A Retrospective Genomic Landscape of 661 Young Adult Glioblastomas Diagnosed Using 2016 WHO Guidelines for Central Nervous System Tumors

Abstract

Introduction

Glioblastomas are primary malignant brain tumors that develop across all age groups, including children and adults. Standard of care treatment includes maximal surgical resection followed by chemotherapy (eg, temozolomide) and radiation therapy; however, median overall survival (OS) remains poor at 14.6 months^1,2

Integration of next-generation sequencing (NGS) into clinical workflows enables clinicians to identify genomic alterations with targeted therapies and/or prognostic value for many tumor types.³ While brain tumors have traditionally been classified on a histopathological basis, updates in the 2021 World Health Organization Classification of Tumors of the Central Nervous System have integrated genomic biomarkers into their diagnostic indications.^4,5 Among adult WHO grade 2/3 brain tumors: tumors of astrocytic lineage are associated with IDH1/2, TP53, and ATRX mutations, while oligodendroglial lineage tumors are associated with mutations in CIC, FUBP1, and 1p/19q codeletion.⁴ These updates affect the definition of disease entities based on these features, and a comparison between prior and current disease classifications is desirable.

Recent advances in comprehensive genomic profiling (CGP) have made accessible the ability for clinicians to probe an array of actionable, disease-defining alterations. The inclusion of CGP into existing histopathological classifications provides an “integrated diagnosis,” adding depth to disease ontologies and further optimizing clinical management.^6-8 Genomic markers of glioblastomas appear to be stratified by age, with notable differences between pediatric and adult populations in thematic oncodrivers.^9,10 One study exploring somatic copy-number alterations (SCNAs) found significant differences between young glioblastomas (≤40 yo, CDKN1C/CEND1 deletion) and older glioblastomas (>40 yo, EGFR amplification and CDKN2A/B deletion) patients.¹¹ Studies prior to the 2021 WHO guidelines indicated young adults (adults ≤ 40 yo) show a unique molecular landscape to both pediatric and classical adult populations; enriched for alterations to IDH1/2, ATRX, BRAF, and TP53; motivating the inclusion of molecular features in the diagnostic criteria for several disease entities.¹² Most notably, IDH1/2 mutant glioblastomas were reclassified to astrocytoma, IDH-mutant; in part due to their positive prognosis and overall survival.^4,13 Given the varied genomic landscape, identifying driving alterations in patients is crucial to identifying promising new therapeutic modalities.^8,14,15

In this study, we present a cohort of 661 young adult glioblastomas diagnosed using the 2016 WHO World Health Organization Classification of Tumors of the Central Nervous System, utilizing CGP to explore their genomic landscape and assess their relationship to currently defined disease entities. Our analysis explored variants with evidence of pathogenic function, common copy number variants (CNVs), and several novel fusion events not described in the literature. We further explored tumor mutational burden (TMB) mutational signatures, anatomic location, and tumor recurrence. Using data collected from CGP, we leveraged unsupervised machine learning techniques to identify 10 genomic classes in previously assigned young adult glioblastomas. We relate these molecular classes to current World Health Organization guidelines; and reference current literature to give therapeutic and prognostic descriptions where possible.

Methods

Comprehensive Genomic Profiling

Comprehensive genomic profiling (CGP) was performed in a College of American Pathologists (CAP)-accredited, Clinical Laboratory Improvement Amendments (CLIA)-certified, New York State regulated reference laboratory (Foundation Medicine). All samples underwent central histopathologic review by a board-certified neuropathologist (S.H.R.) using 2016 World Health Organization diagnostic criteria and were determined to be brain glioblastomas. This study was approved by the Western Institutional Review Board (IRB #20152817) and includes a waiver of informed consent and a HIPAA waiver of authorization.

Genomic Sequencing of Tissue Specimens

Specimen underwent sequential sequencing using≥50 ng of DNA, extracted from 661 formalin-fixed, paraffin-embedded (FFPE) clinical specimens (August 2014—August 2020). NGS was performed using hybridization-captured, adaptor-ligation libraries to a high, uniform coverage (Mean: 789.9X), covering 324 common oncogenes, and the intronic regions for 28 genes involved in rearrangement events (FoundationOne and FoundationOneCDx).¹⁶ NGS data were analyzed for pathogenic genomic alterations, as determined by an internal caller; including base pair substitutions, insertions/deletions, copy number alterations, and rearrangements/fusions. Sample inclusion was contingent on testing negative for 1p/19q codeletion. Specimen calls for mutational signatures were made using the methodology as described by Zehir et al¹⁷ TMB was calculated on ≤1.2 megabase (Mb), defined as the number of somatic, coding point mutations, and indels per megabase of genomic material. Microsatellite instability (MSI) status was determined for each tumor and reported as MSI stable (MSS), ambiguous (MSA), and MSI High (MSI-H). Our investigation was limited to known/likely pathogenic variants. Variants identified in the previous literature as being pathogenic and/or variants occurring twice in COSMIC were considered known pathogenic. Truncating or frameshift events in tumor suppressor genes were considered likely pathogenic. Manual review of pathology reports submitted was used to abstract tumor location and primary/recurrent disease status as available. A histological review was also done to determine specimens matching gliosarcoma, primitive neuroectodermal, or oligodendroglial histology as defined using the 2016 WHO CNS guidelines (S.H.R.).

Comparisons of Mutation Frequency

To compare our cohort of 2016 WHO young adult glioblastomas (age 18–40 yo.; n = 661) to 2016 WHO pediatric (age < 18 yo.; n = 182) and 2016 WHO classic adult (Age > 40 yo.; n = 661) glioblastomas, specimens matching pediatric and classic adult age ranges that had undergone tissue-based sequencing during the same period (August 2014—August 2020) were selected. All available data from pediatric specimens were collected, and 661 classic adult glioblastomas were selected at random for inclusion. Comparisons between IDH^mut and IDH^wt specimens and patients, primary and recurrent patients were also performed. Statistical comparisons of alteration frequency were performed for the 25 most frequently altered genes in young adult glioblastomas, using a series of Fisher Exact tests for the 25 most frequently altered genes in young adult glioblastomas, using Holm-Bonferroni Correction for multiple comparisons.

Latent Class Analysis

To identify molecular subtypes of young adult glioblastomas, we performed a latent class analysis on a subset of these patients (n = 572), imposing the FDA TMB cutoff of≥10 mut/Mb to reduce noise from hypermutated cases and excluding cases primarily arising from MMR deficiency.¹⁸ A series of Gaussian Mixture Models were used, varying the number of Gaussian mixtures between.^5-20 Model choice was performed using each mixture’s Calinski–Harabasz score, and the model with the first local maxima was selected. Defining genes for each latent molecular class were determined using Gaussian mixture model weights (Supplementary Material).

Results

Patient demographics

CGP was performed on a cohort of 661 young adult glioblastoma specimens, as determined using 2016 WHO guidelines (≥18–≤40 yo.). The cohort median (mean) age was 33 yo (32 yo); distributed among the following age ranges: ≥18 to ≤19 (n = 16; 2.4%), ≥20 to ≤24 (n = 82; 12.4%), ≥25 to ≤29 (n = 121; 18.3%), ≥30 to ≤34 (n = 157; 23.8%), and ≥ 35 to ≤40 years (n = 285; 43.1%). Of these young adult glioblastoma patients, 62.0% (410/661) were male and 38.0% (251/661) were female (Supplementary Table S1, top).

Clinicopathological features were abstracted from pathology reports, when available. Patient recurrence status (de novo vs recurrent) was available in 89.5% (592/661) of samples; of which 58.1% (344/592) were identified as de novo and 41.9% (248/592) were identified as recurrent. MGMT promoter methylation status was available for 32.2% (213/661) of samples, of which 41.7% (89/213) were reported as methylated. Biopsy site data was available for 89.3% (590/661) of cases (Supplementary Table S1, bottom).

Mutational signatures were determined in 79.6% (526/661) 2016 WHO young adult glioblastomas, of which 15.9% (84/526) presented with a dominant mutational signature. Alkylating signatures were used as a proxy for recurrence after standard-of-care treatment with temozolomide and/or nitrosoureas. Alkylating mutational signatures were detected in 6.8% (36/526) of cases, whereas MMR, POLE, and BRCA mutational signatures were detected in 5.9% (31/526), 0.6% (3/526), and 0.6% (3/526) of cases, respectively.

Genomic Landscape of Young Adult Glioblastomas

Genomic alterations were identified in almost all (660/661) specimens, with an average of 7.1 GAs per specimen. Commonly altered genes included TP53 (69.3%, 458/661), IDH1 (45.1%, 298/661), and ATRX (43.3%, 286/661); a finding consistent with previous reports demonstrating frequent mutations in these genes in adult lower grade gliomas.¹⁹ Other frequently altered genes included CDKN2A (41.9%, 277/661), CDKN2B (36.2%, 239/661), TERT promoter (20.4%, 135/661), PTEN (20.1%, 133/661), PIK3CA (15.6%, 103/661), NF1 (15.1%, 100/661), and CDK4 (13.9%, 92/661) (Fig. 1A; Supplementary Fig. S1A). Mutations in IDH2 were detected in rare cases (0.6%, 4/661).

Figure 1.

(A) Genomic landscape of 2016 WHO young adult glioblastoma (n = 661). (B) Genomic landscape of yaGBM-PNET (n = 36). (C) Genomic landscape of yaGBM-O (n = 18).

Open in new tabDownload slide

A subset of specimens showed histologic features including primitive neuroectodermal tumor features (GBM-PNET, n = 36), oligodendroglial features (GBM-O, n = 18), and gliosarcoma features (GS, n = 12). While IDH1, TP53, and ATRX mutations were enriched for across all glioblastoma subtypes, H3F3A and MYCN alterations were detected in 19.4% (7/36) and 19.4% (7/36) of glioblastoma-PNETs, respectively (Fig. 1B). Among GBM-Os, 15/18 (83.3%) were positive for mutations involving IDH1 (Fig. 1C) in contrast to GS which showed only 2/12 (16.7%) cases were IDH1 mutant.

Young Adult Glioblastomas Show Distinct Genomic Profiles Compared to Pediatric and Classic Adult Glioblastomas

Although the diagnostic classification of gliomas, including glioblastomas, was based on the absence or presence of microscopic features established by 2016 WHO guidelines, we sought to determine whether the genomic profile of 2016 WHO young adult glioblastomas is significantly different compared to 2016 WHO classic adult glioblastomas (age > 40 years) and pediatric glioblastomas (age < 18 years). We compared the frequencies of genomic alterations in our cohort of young adult glioblastomas to a cohort of 661 randomly selected classic adult glioblastoma samples interrogated by the same genomic assay. Similarly, we compared the genomic landscape of our young adult glioblastoma cohort to a cohort of 182 consecutively tested pediatric glioblastomas using the same assay (Fig. 2).

Figure 2.

(A) Comparison of the Genomic landscape of 2016 WHO young adult glioblastoma (n = 661) to 2016 WHO classic adult glioblastoma (n = 661) (B) Comparison of the Genomic landscape of 2016 WHO young adult glioblastoma (n = 661) to 2016 WHO pediatric glioblastoma (n = 182) (C) Comparison of the Genomic landscape of IDH1/2 Mutant 2016 WHO young adult glioblastoma (n = 302) to IDH1/2 Wildtype 2016 WHO young adult glioblastoma (n = 359).

Open in new tabDownload slide

Young adult glioblastomas, diagnosed using 2016 WHO CNS guidelines, showed distinct genomic profiles compared to 2016 WHO classic adult glioblastoma samples, with 15/25 genes significantly enriched in young adult glioblastomas including TP53, IDH1, and other transcription factor, chromatin modifiers, cell-cycle regulators and fibroblast growth factor pathway activators (Fig. 2A; Supplementary Table S2A). Conversely, classic adult glioblastomas demonstrated significant enrichment for alterations in CDKN2A, CDKN2B, TERT promoter, PTEN, and EGFR. We also note significant differences in the rate of EGFR intragenic deletions (eg, EGFRvIII/vII) (Fig. 2A; Supplementary Table S2A). All significant differences reported were similarly found when performing this comparison on specimens with TMB < 10 (Supplementary Table S2B).

In comparing 2016 WHO young adult glioblastomas and 2016 WHO pediatric glioblastomas, 9/20 genes demonstrated significantly different mutation rates with the young adult glioblastomas enriching for GAs in similar gene signatures described above, while pediatric glioblastomas showed significant enrichment for genomic alterations in H3F3A (33.5% vs 10.1%) and PDGFRA (20.9% vs 11.6%) (Fig. 2B; Supplementary Table S3A). Comparison to patients with TMB < 10 produced similar findings, though we additionally found enrichment for TP53 and PTEN alterations in young adult glioblastomas and did not find significant enrichment in pediatric glioblastomas for PDGFRA (Supplemental Table S3B).

IDH^mut Young Adult Glioblastomas Show Distinct Genomic Profiles Compared to IDH^wt Young Adult Glioblastomas

Prior to the 2016 WHO CNS Guidelines, young adult glioblastomas could be broadly divided into IDH1/2 mutant and wild-type groups; however, whether other genes show significantly different mutation rates remains unclear. To address this question, we compared the mutation rates of the 25 most frequently altered genes in 2016 WHO IDH1/2^mut young adult glioblastomas (n = 302) to 2016 WHO IDH1/2^wt young adult glioblastomas (n = 359) (Fig. 3). Alteration frequencies were significantly different in 12 (12/25) genes between the IDH1/2^mut and IDH1/2^wt groups. Although TP53 was the most frequently altered gene in both cohorts, it showed significant enrichment in IDH1/2^mut young adult glioblastomas (95.7% vs 47.1%). Genes typical to the molecular presentation of glioblastomas, including TERT promoter, PTEN, and EGFR were more frequently altered in IDH1/2^wt young adult glioblastomas, whereas GAs in ATRX, MET, FGF, CCND2, and MYCN were enriched in IDH1/2^mut cases; highlighting the difference in genomic landscapes between histologically indistinct tumors (Fig. 2C; Supplementary Fig. 1B and 1C; Supplementary Table S4A). This comparison among patients with TMB < 10 showed similar findings, except for MET no longer being significantly enriched in IDH^mut specimens (Supplementary Table S4B).

Figure 3.

(A) Genomic landscape of IDH wild-type specimens (n = 359). (B). Genomic landscape of IDH mutant specimens (n = 302).

Open in new tabDownload slide

Among the 298 IDH1^mut 2016 WHO young adult glioblastomas, R132H accounted for 85.9% (256/298) of variants; others included R132G (15/298, 5.0%), R132C (13/298, 4.4%), R132S (12/298, 4.0%), and R132L (2/298, 0.7%). Four (4/661) young adult glioblastomas with IDH2 mutations were identified (R140Q, R172K, R172S, R172M alterations). These findings highlight that approximately 14% of 2016 WHO IDH1/2^mut young adult glioblastomas would not be detected by routine IDH1 R132H immunohistochemical methods.

Recurrent H3F3A and BRAF Mutations in IDH1/2^wt Young Adult Glioblastomas

A sub-analysis of tumors diagnosed as young adult glioblastomas found with H3F3A mutations (n = 67), indicative of pediatric high-grade gliomas. The majority (65.7%, 44/67) of cases harbored the K28M variant, followed by the G35R variant (29.9%, 20/67), and less frequently the G35V variant (4.5%, 3/67). All H3F3A mutant young adult glioblastomas were IDH^wt and had frequent co-occurring alterations in TP53 (44/67), ATRX (32/67), NF1 (19/67), and FGFR1 (17/67) (Fig. 4A; Supplementary Table S5).

Figure 4.

(A) Genomic landscape of H3F3A mutant specimens (n = 67). (B) Genomic landscape of BRAF mutant specimens (n = 41).

Open in new tabDownload slide

Tumor anatomic location information was available for a subset (n = 60) of H3F3A mutant samples and categorized into 3 groups: hemispheric (n = 29), midline (n = 23), and spinal cord (n = 8). Of the 41 H3F3A^K28M mutant tumors with anatomic site data, 75.6% (31/41) presented in the midline or spinal cord, and 24.4% (10/41) presented as hemispheric. Of the 17 H3F3A^G35R and 2 G35V mutant samples with anatomic site data, all (19/19) were hemispheric (Supplementary Table S6).

BRAF alterations were identified in 6.2% (41/661) of 2016 WHO young adult glioblastomas, with V600E identified as the most common variant detected (82.9%, 34/41). Seven additional BRAF variants detected in individual cases included SNVs (D594G, K601E, D587G, G466E), insertions (R506_K507insVLR, V600_W604 > DG), and a single case of amplification. Frequently co-altered genes among the BRAF^mut cohort are CDKN2A (35/41), CDKN2B (34/41), TERT promoter (8/41), and ATRX (7/41) (Fig. 4B; Supplementary Table S7)

Recurrent Gene Fusions are Rare in Young Adult Glioblastomas

While driver gene fusions are reported with increased frequency in pediatric gliomas (eg, KIAA1549-BRAF), it has not been determined whether fusions are significant drivers in young adult glioblastomas.⁹ In our cohort, we identified gene fusions in 2.9% (19/661) of cases including FGFR3-TACC3 (n = 3), FGFR1-TACC1 (n = 2), KIAA1549-BRAF (n = 2), ROS1-GOPC (n = 2), EGFR-SEPT4 (n = 4), RET-CCDC6 (n = 2), FGFR2-GKAP1 (n = 1), ATR-CPNE4 (n = 1), ASTN2-ERG (n = 1), and ARHGEF2-NTRK1 (n = 1) (Supplementary Fig. 4).

A Subpopulation of Hypermutated Young Adult Glioblastomas Harbor Mutations in DNA Mismatch-Repair Genes

Tumor mutational burden (TMB, mutations/Mb) was determined on up to 1.2 Mb of sequenced DNA, and TMB ≥ 10 mutations/Mb was considered high TMB (TMB-high) per the US FDA CDx approval of pembrolizumab in solid tumors. Of the young adult glioblastoma specimens with evaluable TMB scores (643/661), specimens typically showed TMB-low character. Only 11.0% (71/643) reported≥10 mutations/Mb (Fig. 5A; Supplementary Table S8). Among TMB-high patients, we found that 33.8% (24/71) were ultra-mutated, which we defined as having a TMB ≥ 100 mutations/Mb. Ultra-mutated tumors have been frequently associated with mismatch repair (MMR) or POLE mutations. Among patients with MMR/POLE mutational signatures (5.3%, 35/661), ultra-mutation was detected in 14.3% (5/35) patients (Fig. 5B; Supplementary Fig. S5A and Table S9). Similarly, patients with alkylating mutational signatures (36/661) also showed enrichment for ultra-mutator phenotype (33%, 12/36) (Supplementary Fig. 5B; Supplementary Table S10).

Figure 5.

(A) TMB Distribution in 2016 WHO Young Adult Glioblastoma (n = 643). (B) TMB Score by mutational signature.

Open in new tabDownload slide

Tumors classified as TMB-high (n = 71) frequently harbored mutations in TP53 (68/71), ATRX (52/71), IDH1 (40/71), and CDKN2A (34/71). Among TMB-high young adult glioblastomas, 63.9% (45/71) harbored functional mutations in MMR or proofreading genes (MSH6, MLH1, MSH2, POLE1, and PMS2), while MMR genes were not present in any TMB-low specimens. Among the 40 IDH1/2^mut TMB-H tumors, 70% (28/40) harbored mutations in MMR genes.

Latent Class Analysis Reveals 10 Distinct Genomic Subtypes of Young Adult Glioblastomas

Subtyping young adult glioblastomas based on histology features was well established with distinct disease entities (per 2016 WHO Classification of CNS tumors), including young adult glioblastomas with oligodendroglial features (GBM-O), gliosarcoma features (GS), or primitive neuroectodermal tumor features (GBM-PNET), among others. Most tumors, however, show a more typical glial appearance that cannot be further subclassified based on tumor cell morphology. Determining how histologically similar tumors differ genetically can improve the overall understanding of gliomagenesis and potentially guide novel treatment strategies. Using the methodology previously described, we identified 10 distinct molecular classes of young adult glioblastomas (Fig. 6; Supplementary Table S11).

Figure 6.

(A) Latent molecular classes of 2016 WHO young adult glioblastoma (n = 572). (B) Features of latent molecular classes in 2016 WHO young adult glioblastoma (n = 572).

Open in new tabDownload slide

2016 WHO IDH^mut young adult glioblastomas, now classified as IDH^mut Astrocytomas,⁴ were distributed among classes 2, 4, 8, and 10 (Fig. 6; Supplementary Table S11). Class 4 (n = 98) specimens showed enrichment for TP53, IDH1, and ATRX. Class 2 (n = 56) specimens were similarly enriched for alterations in TP53, IDH1, and ATRX; with additional CDKN2A/B codeletion. Class 8 (n = 30) specimens showed enrichment for alterations in TP53, IDH1, and CDKN2A/B codeletion. Class 10 (n = 30) specimens were primarily characterized by TP53, ATRX, and IDH1 alterations; with co-occurring amplifications of CCND2, FGFR23, FGFR6, and CDK4. In summary, IDH^mut tumors were stratified by ATRX truncations, CDKN2A/B codeletion, and CCND2/FGFR23/FGFR6/CDK4 amplifications.

IDH^wt tumors were distributed among classes 1, 5, 6, and 9. Class 1 (n = 44) specimens showed enrichment for EGFR amplifications and internal deletion events (EGFR vII/vIII), often co-occurring. Further investigation found that 4.4% (29/661) of specimens had co-occurring EGFR amplification and internal deletion (vII/vIII) events, and specimens typically were from the primary biopsy (66.7%, 18/27). Class 1 specimens showed accompanying CDKN2A/B codeletion and TERT promoter alterations. Class 5 (n = 77) specimens were enriched for H3F3A mutant tumors, corresponding to a pediatric-type diffuse high-grade glioma diagnosis. Class 6 (n = 81) specimens were characterized by alterations in BRAF with CDKN2A/B codeletion. Class 9 (n = 21) specimens showed co-amplification to CDK4, FGFR1, and GLI1. Both classes 6 and 9 can be categorized into various high-grade 2021 WHO entities including high-grade astrocytoma with piloid features, pleomorphic xanthoastrocytoma, and gangliogliomas with anaplastic features.

Classes 3 and 7 specimens were composed of a mixture of IDH^mut and IDH^wt tumors. Class 3 (n = 99) specimens showed enrichment for CDK4 amplifications as well as TERT promoter, PTEN, and RB1 alterations. Class 7 (n = 30) specimens were characterized by co-amplification to PDGFRA, KIT, and KDR. CDK4 amplification and CDKN2A/B codeletion were also present. Alterations to IDH1 and ATRX were common as well.

Discussion

Glioblastomas occur at all ages but can broadly be divided into 3 age groups encompassing pediatric (<18 years), young adult (18 to ≤40 years), and classic adult glioblastomas (>40 years).¹¹ In this study, we performed a retrospective analysis of 661 young adult glioblastomas (diagnosed using 2016 WHO classification criteria) that underwent CGP to characterize the genomic landscape of this unique and understudied patient population. To exclude the possibility of misclassified oligodendrogliomas confounding our analysis, all tumors included in the young adult glioblastoma study cohort were confirmed to be computationally negative for 1p/19q codeletion, the biomarker that defines oligodendroglial lineage.²⁰

Our study revealed that 45.1% of young adult glioblastomas were IDH1 mutants, with rare (0.6%) cases harboring mutations in IDH2. Under the 2021 WHO CNS Guidelines, these tumors are now classified as IDH^mut astrocytomas. These specimens were enriched for mutations in TP53 and ATRX, consistent with prior studies highlighting this recurrent genomic signature in adult lower-grade astrocytomas.^6,8CDKN2A/B deletions were detected in 37.4% and 29.1% of IDH^mut and IDH^wt specimens, respectively. Comparisons of frequently altered genes in our young adult glioblastoma cohort with pediatric glioblastomas revealed striking differences including enrichment in young adult glioblastoma for alterations in IDH1, ATRX, CDKN2A/B, TERT promoter, and EGFR; whereas pediatric glioblastomas were enriched for H3F3A and PDGFRA alterations. Similarly, when comparing 2016 WHO young adult glioblastomas to 2016 WHO classic adult glioblastomas, 2016 WHO young adult glioblastomas were enriched for alterations in IDH1, TP53, ATRX, H3F3A, CCND2, FGF23, GLI1, FGF6, and FGFR1 while classic adult glioblastomas enriched for EGFR, PTEN, CDKN2A/B, and TERT promoter alterations.

To establish a genomic framework for subclassifying 2016 WHO young adult glioblastomas, we utilized a Gaussian mixture model as an objective way to identify molecular subtypes. LCA revealed 10 latent classes of 2016 WHO young adult glioblastomas based on pathogenic variants. We note first that IDH1/2^mut status stratified molecular classification; with 3 showing computed means > 80% (classes 2, 4, 8, and 10) and 4 showing computed means<20% (classes 1, 5, 6, and 9). Our analysis tiers histopathological diagnosis with molecular alterations to subclassify young adult glioblastomas beyond 2016 WHO diagnostic entities and gives provides additional detail for stratifying molecular features. Our analysis has limited meaningfulness to current young adult glioblastomas since patients were assigned as young adult glioblastomas prior to the 2021 WHO CNS guidelines.

Among 2016 WHO IDH^mut classes (classes 2, 4, 8, and 10), enrichment for pathogenic mutations in both TP53 and ATRX were frequent, a combination that has been shown to lead gliomagenesis by suppression of SOX2.²¹ Classes 2, 4, and 8 showed similar features, stratifying primarily based on CDKN2A/B and ATRX alteration status. Functionally, loss of CDKN2A has been linked to faster tumor growth and confers a poorer overall prognosis for glioma patients.^22-24 Class 10 showed CDK4 amplification and coamplifications of CCND2, FGFR23, and FGF6, which have shown susceptibility to alternative therapies employing CDK4/6 inhibitors, such as abemaciclib.^25,26 Ongoing clinical trials are exploring the use of CDK4/6 inhibitors in glioblastomas, either as monotherapy or in combination with immunotherapeutic strategies (NCT02981940, NCT04118036). These findings indicate that, for young adult glioblastomas diagnosed using the 2016 WHO CNS guidelines, IDH^mut tumors associate with a distinct molecular landscape from IDH^wt wildtype tumors.

Mutations to PTEN and TERT promoters showed significant enrichment among several classes (classes 1, 3, and 6). PTEN mutations, often co-occurring with TERT promoter mutations, interact through the PTEN-AKT pathway to produce heightened cell proliferation.²⁷ In glioblastomas, TERT promoter mutations carry a significantly worse overall survival (81.7 vs. 152.6 weeks; P = .026).²⁸ Targeted therapies for TERT promoter are in early stages, but Eribulin shows promise in early trials of glioblastoma mouse models.²⁹

Class 5 showed significant enrichment of H3F3A point mutations, primarily occurring as K28 and G35 alterations; sometimes accompanied by FGFR1 alterations. These have been previously described as showing epigenetically distinct populations, and whose features reflect pediatric-type, diffuse high-grade gliomas.³⁰

Class 6 was defined primarily by BRAF V600E alterations, along with CDKN2A/B codeletion and PTEN/TERT promoter mutants. Initial clinical data have shown some efficacy of combination of BRAF (dabrafenib) and MEK (trametinib) inhibitors to slow disease progression.

Classes 7 and 10 appeared to be defined by the existence of CNAs, implicating large-scale chromosomal amplification events. These have been described previously in Zhang et al,³¹ which divided patients into RMPA^high (RAS-RAF-MEK-MAPK pathway) and RMPA^low (PIK3-AKT pathway) subgroups. From our results, PDGFRA-KIT-KDR CNAs (class 7) were associated with the RMPA^high group and FGF6 & FGF23 CNAs (class 10) were associated with the RMPA^low group, of which, gene-dosage dependent expression was confirmed in PDGFRA/KIT and FGF23, respectively. This distinction has clinical relevance, as RMPA^low patients consistently show far better survival than their RMPA^high counterparts (P < .0001).³¹ Given that both groups show relatively high copy numbers (22.9 copies in FGF6/23, 24.0 in PDGFRA/KIT/KDR), our data indicate these 2 amplicons may enhance gliomagenesis and define a unique genomic profile of young adult glioblastomas.

Checkpoint inhibitor-based (PD-L1 and PD-1 inhibitors) therapies for glioblastoma patients have advanced into clinical trials and initial studies in adult and pediatric glioblastoma patients have shown profound anti-tumor responses in glioblastomas that harbor constitutional mismatch repair (MMR) gene mutations^32,33 or germline POLE mutations.³² Elevated TMB is a biomarker associated with increased response to checkpoint inhibitors in peripheral solid tumors. Hypermutation in gliomas was recently explored in gliomas; with evidence showing this phenotype is likely driven by the emergence of sub-clonal MMR-mutant, TMZ-resistant tumor populations during therapy.³⁴ Increased TMB often leads to a pattern of MMR mutations and a mutational signature unique to alkylating agents (TMZ) upon recurrence. We examined MMR mutations, TMB, and mutational signature data within our cohort to understand the landscape of hypermutation in young adult glioblastomas. We identified a bimodal distribution, with 87.7% (572/661) presenting with a TMB < 10 and 12.3% (89/661) presenting as TMB^high (TMB ≥ 10). Our findings confirm that TMZ-associated hypermutation is a recurrent feature in a subset of young adult glioblastomas and is frequently associated with underlying MMR gene mutations.

Taken together, these findings highlight the importance of molecular features in the diagnosis of young adult glioblastomas and high-grade gliomas. We offer insight into driver mutations in these latent molecular classes, which vary significantly, and feel these findings should be considered when designing and/or planning clinical trials.

Funding

The authors indicated are employees of Foundation Medicine, which provided the data and funding for this manuscript. This research was performed as part of their duties while employed at Foundation Medicine Inc.

Conflict of Interest

James F. Haberberger reported employment with Foundation Medicine Inc. (FMI), Pathology Research Coordinator, intellectual property: FMI (a patent submission is pending for portions of the methods involving classifier methods for patient tumor molecular data; however, much of the methodology is based on publicly available machine-learning techniques and are described in the methodology; patent filing was after internal review at FMI, and the paper was (barring minor edits) in its current form; while the first author is listed as the inventor, Foundation Medicine Inc. owns the rights to the patent). Worthy Pegram is an employee of Foundation Medicine Inc. Nicholas Britt reported employment with Foundation Medicine Inc., and ownership interests with F. Hoffmann-La Roche AG (Roche). Kelsie Schiavone is an employee of Foundation Medicine Inc. Eric Severson was previously affiliated with Foundation Medicine Inc., and is an employee with ownership interest in Labcorp. Radwa Sharaf is an employee with Foundation Medicine Inc., and owns stock in Roche. Lee A. Albacker is an employee of Foundation Medicine Inc., and has ownership interests in Roche Holdings AG. Erik Williams reported employee/consultant of Foundation Medicine Inc., a wholly owned subsidiary of Roche Holdings, Inc. and Roche Finance Ltd, and equity interest in an affiliate of these Roche entities. Mirna Lechpammer is an employee of Foundation Medicine Inc. Amanda Hemmerich is an employee of Foundation Medicine Inc. Douglas Lin is a full-time employee of Foundation Medicine Inc., a whole subsidiary of Roche, and has Roche stocks. Richard S.P. Huang is an employee of Foundation Medicine Inc., a wholly owned subsidiary of Roche and has equity interest in Roche. Matthew Hiemenz is an employee of Foundation Medicine Inc. Julia Elvin is an employee of Foundation Medicine Inc. Ryon Graf is an employee of Foundation Medicine Inc. James Creeden is an employee of Foundation Medicine Inc. Jeffrey S. Ross is an employee of Foundation Medicine Inc. Brian M. Alexander reported employment at Foundation Medicine Inc., and equity in Roche as part of compensation package. Shakti H. Ramkissoon is an employee of Foundation Medicine Inc. The other authors indicated no financial relationships.

Author Contributions

Conception/design: J.H., S.H.R. Provision of study material or patients: W.P., N.B., K.S., R.S. Collection and/or assembly of data: W.P., N.B., K.S., R.S. Data analysis and interpretation: R.S., J.H., S.H.R. Manuscript writing: J.H., E.S., G.L., W.L.B., L.A.R., M.B.C., P.R., S.H.R. Final approval of manuscript: All authors.

Data Availability

The data underlying this article were provided by Foundation Medicine Inc under licence/by permission. Data will be shared on request to the corresponding author with permission of Foundation Medicine Inc.

References