Elucidating the molecular pathogenesis of glioma: integrated germline and somatic profiling of a familial glioma case series

Abstract

Background

The genomic characterization of sporadically arising gliomas has delineated molecularly and clinically distinct subclasses of disease. However, less is known about the molecular nature of gliomas that are familial in origin. We performed molecular subtyping of 163 tumor specimens from individuals with a family history of glioma and integrated germline and somatic genomic data to characterize the pathogenesis of 20 tumors in additional detail.

Methods

Immunohistochemical analyses were performed on formalin-fixed, paraffin-embedded tumor sections to determine molecular subtypes of glioma. For 20 cases, tumor DNA was exome sequenced on an Illumina HiSeq 2000 platform and copy number profiling was performed on the Illumina HumanOmniExpress BeadChip. Genotypes at glioma risk polymorphisms were determined from germline DNA profiled on the Illumina Infinium OncoArray and deleterious germline mutations were identified from germline sequencing data.

Results

All 3 molecular subtypes of sporadic glioma were represented in the overall case series, including molecular glioblastoma (n = 102), oligodendroglioma (n = 21), and astrocytoma (n = 20). Detailed profiling of 20 of these cases showed characteristic subtype-specific alterations at frequencies comparable to sporadic glioma cases. All 20 cases had alterations in genes regulating telomere length. Frequencies of common glioma risk alleles were similar to those among sporadic cases, and correlations between risk alleles and same-gene somatic mutations were not observed.

Conclusions

This study illustrates that the molecular characteristics of familial tumors profiled largely recapitulate what is known about sporadic glioma and that both germline and somatic molecular features target common core pathways involved in gliomagenesis.

Key Points

1. Familial and sporadic gliomas display highly comparable molecular landscapes.

2. Germline and somatic molecular events target common core pathways involved in gliomagenesis.

3. Carriage of germline glioma risk variants is not associated with somatic events in the same gene.

Importance of the study

While approximately 5%–10% of malignant gliomas occur in individuals with a family history of glioma, little is known about the molecular nature of tumors occurring in such individuals. In this study, we determine the molecular subtypes of glioma for 163 tumors arising in families with a history of glioma and perform multiplatform germline and somatic genomic characterization of a subset of 20 cases. Taken together, our analyses indicate that the molecular landscape of familial glioma tumors in this study largely recapitulates what is known about sporadic glioma. Furthermore, the study demonstrates that both germline and somatic molecular features target common core pathways involved in gliomagenesis in a complementary manner.

Approximately 20000 cases of diffuse glioma, the most common primary malignant brain tumor in adults, are diagnosed annually in the United States and cause significant morbidity and mortality due to aggressive behavior and limited treatment efficacy.¹ While gliomas have been traditionally classified according to their histopathological appearance, efforts to molecularly characterize this disease in recent years have revealed highly recurrent somatic alterations. These alterations, which include isocitrate dehydrogenase 1 (IDH1) and IDH2 mutations and codeletion of chromosome arms 1p and 19q, define 3 major subtypes of diffuse gliomas.^2–5 IDH-mutant tumors occur in 2 broad groups, including those with 1p/19q codeletion and typically oligodendroglial morphology, and those without 1p/19q codeletion—predominantly lower-grade astrocytomas enriched for mutations in both alpha thalassemia/mental retardation syndrome X-linked protein (ATRX) and tumor protein 53 (TP53). A third group without IDH mutation comprises primarily aggressively behaving glioblastomas (GBMs). Together, these features delineate robust, objectively defined subclasses of glioma that correlate strongly with clinical behavior.

While these efforts have elucidated the molecular pathogenesis of sporadically arising glioma, far less is known about the molecular nature of gliomas that are familial in origin. It is estimated that 5%–10% of gliomas occur in families with 2 or more affected relatives, and studies have indicated that individuals with a family history of the disease carry approximately twice the risk of glioma development as compared with the general population.^6–10 A small fraction of gliomas are caused by single-gene hereditary cancer syndromes, including Li–Fraumeni syndrome, neurofibromatosis type I/II, Lynch syndrome, and melanoma-neural system tumor syndrome.^11–13 Additionally, several families with glioma aggregation have been found to harbor mutations in protection of telomeres protein 1 (POT1),¹⁴ a shelterin complex member involved in telomere maintenance and DNA damage response.^15–17

Although these observations have increased our understanding of the causes of familial glioma, little is known about the molecular profiles of tumors arising in such families. Here, we performed molecular subtyping of a series of 163 familial glioma cases derived from participants in the Gliogene International Consortium, and conducted comprehensive germline and molecular profiling on a subset of 20 cases. Together, these analyses highlight the similarities in the overall molecular landscape of familial and sporadic glioma tumors while exhibiting the interplay of germline and somatic events involved in the molecular pathogenesis of familial glioma.

Methods

Sample Collection and Processing

We collected formalin-fixed, paraffin-embedded tumor samples from participants in the Gliogene Consortium, a large study of familial glioma that recruited 435 families from 14 international sites from 2007 to 2011.¹⁸ At the time of the study, tumor specimens from 219 cases had been collected. A board-certified neuropathologist (J.T.H.) reviewed all slides and tissue blocks for tumor content before submitting suitable samples for downstream genetic analysis. We extracted DNA using Qiagen GeneRead DNA FFPE Kits from tissue scrolls or from macrodissected regions using marked hematoxylin and eosin‒stained slides as guides for blocks with partial tumor content. We measured the DNA sample quality and concentration to determine suitability for whole exome sequencing.

Immunohistochemistry for Molecular Subtyping

Tumor specimens from 163 cases were suitable for immunohistochemistry (IHC) analyses. We conducted IHC profiling for IDH1 R132H mutations and ATRX deficiency on all samples using two 5-micron tissue sections with automated IHC processing equipment. Slides were reviewed by a neuropathologist (J.T.H.). As ATRX deficiency and 1p/19q codeletion occur with almost complete mutual exclusivity, these paired analyses were used to efficiently establish molecular subclass for 88% of the cohort.

Common Germline Variant Genotyping

We used Illumina’s Infinium OncoArray-500K BeadChip to genotype germline DNA. We extracted the genotypes at the 25 known glioma risk SNPs¹⁹ from existing databases for the directly genotyped SNPs, otherwise we imputed genotypes using the Michigan Imputation Server²⁰ according to 1000 Genomes Phase 3 reference haplotypes. For the analysis of the association between same-gene germline variants and somatic mutations, we categorized each case with respect to the presence or absence of corresponding alterations, and summed events across genes to increase power. We evaluated the associations by Fisher’s exact test, with P < 0.05 considered to be statistically significant.

Germline Sequencing and Analysis

We identified putative glioma-associated germline mutations using whole exome¹⁴ or whole genome sequencing data. Briefly, whole exome sequencing was performed at the Baylor College of Medicine Human Genome Sequencing Center (BCM-HGSC) on the Illumina HiSeq 2000 platform after exome capture using NimbleGen VCRome 2.1 capture libraries.²¹ Whole genome sequencing was performed on the Illumina HiSeq X platform and reads were preprocessed using the HGSC Mercury pipeline. Reads were mapped to the GRCh37 Human reference genome with the Burrows-Wheeler aligner²² and recalibrated using the Genome Analysis Toolkit (GATK).²³ Germline SNPs and indels were subsequently called with respect to the human reference genome using GATK HaplotypeCaller software,²⁴ and variants were annotated with ANNOVAR²⁵ for protein impact, allele frequency from the Exome Aggregation Consortium database (ExAC, non–Cancer Genome Atlas), and deleteriousness prediction scores for non-synonymous variants. We filtered variants and examined them using multiple criteria to identify candidate glioma-associated events, including entry in the ClinVar database as a pathogenic variant, prior evidence for a gene in glioma or other familial cancer predisposition, rarity (<1/5000), functional impact (truncating, splice site), and high Combined Annotation Dependent Depletion²⁶ score (>20, denoting top 1% of most deleterious variants).

Tumor Exome Sequencing and Mutation Calling

Tumor exome sequencing was performed for a subset of 20 cases for which (i) sufficient quality and quantity of tumor DNA could be extracted and (ii) germline DNA sequencing data were previously collected to permit identification of somatic mutations via tumor-normal comparison. We conducted whole exome capture of paired-end tumor DNA libraries at the BCM-HGSC with supplementary probe sets to enhance coverage of clinically relevant genes and the telomerase reverse transcriptase (TERT) gene promoter region. Sequencing was performed on the Illumina HiSeq 2000 platform, and we processed the samples as described above. We used MuTect2 to call single-nucleotide variants and indels with respect to germline DNA sequencing data.²⁷ We used Oncotator to filter and annotate variants for those with a minimum tumor variant allele fraction of 10% with at least 5 variant allele reads.²⁸ We used the Broad Integrative Genomics Viewer to manually examine recurrent TERT promoter (TERTp) mutations C228T and C250T.

Tumor Copy Number Profiling

We conducted copy number profiling for the 20-case set using the Illumina HumanOmniExpress BeadChip and processed the data using GenomeStudio v2.0. Subsequently, we used ASCAT v2.4²⁹ software to generate genome-wide allele-specific copy number profiles and we manually called copy number alterations in conjunction with review of Log R and B Allele Frequency plots.

Ethical Approval and Informed Consent

Written informed consent was obtained from each subject or from his or her guardian. Approval from local institutional review boards was received at participating institutions. The study sponsors had no role in the design of the study, the collection, analysis, or interpretation of the data, the writing of the manuscript, or the decision to submit the manuscript for publication.

Results

Molecular Subtypes of Familial Glioma

In order to ascertain the molecular subtypes of familial glioma tumors in our study, we performed IHC staining for IDH1 R132H, the most frequent IDH mutation in glioma, and ATRX, loss of which tends to occur mutually exclusively with 1p/19q codeletion. In a previous analysis by our group, these 2 assays stratified a set of 137 sporadic glioma cases into the 3 molecular subtypes with high sensitivity (0.92) and specificity (0.95) relative to definitive molecular classification including targeted sequencing of IDH1, IDH2, and the TERT promoter and global methylation profiling (unpublished data).

We conducted IHC analyses on 163 samples and were able to successfully subclassify 143 of these (87.7% of cohort). The distribution of molecular subtypes and representative results are presented in Fig. 1. Forty-three samples stained positive for IDH1 R132H mutations, while 20 samples were shown to have loss of ATRX expression. Taken together, 102 samples were designated as molecular glioblastoma, 21 samples as molecular oligodendroglioma, and 20 samples as molecular astrocytoma. We were unable to reliably subclassify 20 tumors by this approach because of uninterpretable ATRX staining; 18 of these did not show IDH mutation and are likely glioblastomas, while the remaining 2 IDH-mutant cases are likely molecular oligodendroglioma or astrocytoma.

Fig. 1.

Distribution of molecular subtypes of 163 familial glioma cases. Molecular subtypes of familial glioma cases were determined by immunohistochemistry to determine presence of IDH1 R132H mutations and loss of ATRX. Subtypes were determined for 143 cases; for the remaining 20, ATRX status was indeterminate but IDH1 R132H mutation status is known. wt, wildtype; mut, mutant.

Integrated Germline and Somatic Profiling

In order to better characterize the molecular pathogenesis of familial glioma, we selected a subset of 20 tumors from unrelated familial glioma patients for multiplatform germline and somatic molecular profiling. An overview of key molecular alterations identified in these cases is presented in Fig. 2, where we stratify cases into the 3 aforementioned molecular subtypes. Three cases harbored IDH1 mutations and 1p/19q codeletion (Subtype 1) and were all previously histopathologically classified as oligodendrogliomas. Seven cases with IDH1 mutation and without 1p/19q codeletion (Subtype 2) displayed a mix of histopathological classes, including 4 lower-grade astrocytomas, 2 GBMs, and 1 oligodendroglioma. The remaining 10 tumors were classified as IDH wildtype (Subtype 3), comprising 8 tumors histopathologically designated as GBM and 2 as anaplastic astrocytomas. The median age at diagnosis for IDH-mutant cases was 42.5 years, while the median age for IDH wildtype cases was 55 years. Additional subject characteristics, including information on glioma cases reported in each family, are provided in Supplementary Table 1.

Fig. 2.

Germline and somatic molecular profiles of 20 familial glioma cases. Cases are separated into molecular subtype according to IDH mutation and 1p/19q codeletion status and are annotated with histopathological designation and age. Genotypes for major glioma risk polymorphisms are presented along with known deleterious germline mutations, somatic mutations, and somatic copy number aberrations of interest. Along the right side, risk allele frequencies (RAFs) and frequency of somatic alterations are presented for this study and the Glioma International Case Control Study (GICC) and the LGG+GBM TCGA cohort for germline and somatic events, respectively.

Twenty-five single nucleotide polymorphisms (SNPs) have been associated with glioma risk in the largest genome-wide association study to date.¹⁹ Genotypes at these loci are presented for each of the 20 cases in Supplementary Table 2, and genotypes for major glioma risk polymorphisms are illustrated in Fig. 2A for each profiled case. Additionally, risk allele frequencies (RAFs) among the 20-case set and among glioma cases genotyped in the Glioma International Case-Control Consortium³⁰ are presented to permit comparison to a large set of sporadically arising glioma cases. Frequencies of top glioma risk alleles among the familial cases do not systematically differ from those in the sporadic case set.

To identify potential high-risk glioma-predisposing mutations, germline sequencing data generated for each of the 20 cases were examined for highly rare alleles with predicted deleterious protein impact, emphasizing genes previously associated with glioma, familial cancer syndromes, or cancer biology. We observed 11 mutations of interest in 10 cases (Fig. 2B), including 6 cases with mutations in genes involved in DNA damage recognition or repair, including TP53, ATR, MLH3, and POLD1. Three cases displayed germline mutations in shelterin complex members POT1 (as previously reported¹⁴) and TERF2, and 2 cases harbored mutations in brain-tumor associated gene DMBT1. Additional data on these mutations, including genomic position, allele frequency, and protein impact, are provided in Supplementary Table 3.

We identified nonsilent somatic mutations in corresponding patient tumor specimens by exome sequencing and comparison to germline DNA. Tumor exome sequencing was completed at an average target read depth of 116X and the analysis yielded 1137 single nucleotide variants and indels across the 20 samples (median, 58.5 nonsilent mutations per sample; range, 32–101; Supplementary Table 4). Recurrent mutations in sporadic glioma observed at 2 TERTp loci³¹ were determined via targeted DNA capture and manual review (Supplementary Fig. 1). Additionally, we determined copy number alterations by array-based copy number profiling, analyzed data using the ASCAT algorithm, and manually reviewed to identify amplification, deletion, and copy-neutral loss of heterozygosity (cnLOH) events (Supplementary Fig. 2). Overall, the occurrence and frequency of alterations observed in this set were highly concordant with recurrent alterations seen in the combined GBM and lower-grade glioma cohort analyzed by The Cancer Genome Atlas² (Fig. 2C, D). Furthermore, subtype-characteristic alterations were observed, including (i) TERTp and CIC mutations in Subtype 1 tumors; (ii) ATRX mutations plus TP53 mutations and co-incident cnLOH in Subtype 2 tumors; and (iii) TERTp and assorted receptor tyrosine kinase (RTK)/phosphatidylinositol-3 kinase (PI3K)/mitogen-activated protein kinase mutations along with characteristic chromosome 7 gains and chromosome 9/10 losses in Subtype 3 tumors.^2,4,5,32

Alterations to Telomere Regulatory Genes

We inspected telomere regulation-related events in greater detail for each sequenced familial glioma case and results are shown in Fig. 3. Notably, we observed an inherited and/or acquired mutation to a gene involved in telomere maintenance in all of the 20 cases. Characteristic acquired alterations were identified in most cases, including TERTp mutations in 2 of 3 Subtype 1 tumors and 8 of 10 Subtype 3 tumors. As has been previously reported,³³ C228T mutations were more common than C250T mutations (90% vs 10% of TERTp mutations, respectively). We observed protein-altering mutations of ATRX in 6 of 7 Subtype 2 cases, as is typical in this subtype of sporadic glioma, as well as 1 ATRX mutation in an IDH wildtype case, which is uncommon but has been previously observed.⁴ Interestingly, all 3 cases without a characteristic acquired telomere-related mutation harbored germline alterations to telomere shelterin complex genes POT1 and TERF2. A compound heterozygous mutation was observed in one case in which a POT1 E450X mutation was inherited and a POT1 R117H mutation was acquired.

Fig. 3.

Mutations in telomere regulatory genes are observed in all familial glioma cases studied. Inherited and/or acquired deleterious alterations to telomere regulation are presented for each case studied according to molecular subtype along with the specific amino acid or nucleotide change. All alterations involved either shelterin complex gene POT1 or TERF2, an activating mutation in the TERT gene promoter (TERTp), or loss of function mutations in ATRX.

Correlation of Common Germline Variants and Somatic Mutations

Several glioma-associated SNPs have been observed to occur in or near genes that are frequently somatically altered in glioma. It is plausible that these predisposing events may act as a “first hit” to a given gene, such that a positive association between same-gene events would be expected; alternately, it is conceivable that inherited variants may be sufficient for alteration of a given gene and that a negative association would be observed. To explore these hypotheses, we catalogued all inherited and acquired events in 5 gene regions (EGFR, TP53, TERT, CDKN2A/B, and IDH1) and present these results in Fig. 4. We characterized each case with respect to the presence or absence of germline and somatic alterations in each gene, and summed events across genes to increase power, yielding 100 total events (Supplementary Table 5). Overall, there was no statistically significant association observed for the occurrence of a somatic alteration in a gene for which an individual harbors a risk allele (or alleles) (odds ratio = 0.79; 95% CI = 0.31–2.01, P = 0.65).

Fig. 4.

Carriage of germline risk variants is not associated with occurrence of somatic events in the same gene. Cases were annotated for germline and somatic alterations for 5 genes displaying both glioma-associated risk polymorphisms and somatic alterations. Each column represents one case, and columns are ordered according to germline and somatic features independently for each gene. Acquisition of somatic alterations was found to be statistically independent of carriage of germline risk alleles in the same gene when assessed in aggregate across the 5 genes (P = 0.65).

Discussion

While approximately 5%–10% of malignant gliomas occur in individuals with a family history of glioma,^6,8,34 little is known about the molecular nature of tumors occurring in such individuals. Increasing knowledge of familial glioma etiology can potentially yield benefits for individual families in which a key predisposing factor is identified and can inform understanding of glioma etiology in general through analyses of cases that are likely enriched in germline genetic predisposition factors. Here, we determine the molecular subtypes of glioma for 163 tumors arising in families with a history of glioma and perform multiplatform germline and somatic genomic characterization of a subset of 20 cases.

Taken together, our analyses indicate that the molecular landscape of familial glioma tumors analyzed in this study largely recapitulates what is known about sporadic glioma. With respect to the molecular subtypes of the 163 tumors in our study, our analysis indicated that the tumor subtypes occur among familial glioma cases with comparable frequencies as among sporadic cases. This finding is consistent with a previous study of 75 cases from the Gliogene cohort (55 of which overlap with cases in this analysis) that used genomic copy number profiles to classify cases into glioma subtypes based on characteristic subtype-specific alterations.³⁵ From the molecular profiling of a subset of 20 tumors we were able to further demonstrate characteristic subtype-specific alterations such as CIC mutations in IDH-mutant 1p/19q codeleted tumors, co-occurring mutations in ATRX and TP53 in IDH-mutant non-codeleted tumors, and co-occurring chromosome 7 gains and chromosome 10 losses in IDH-wildtype tumors. These observations suggest that both familial and sporadic tumors can develop from alterations to a core set of glioma-associated genes.^2,4,5,32

However, differences in etiology are apparent in multiple cases, most clearly among those with germline mutations in components of the telomere shelterin complex, which is involved in the protection of telomeric DNA and maintenance of its structure.³⁶ Specifically, case 2 (as labeled in Fig. 2) carries a truncating germline mutation in POT1 (p.E450X), while a non-synonymous mutation (POT1 p.G95C) predicted to affect DNA binding was identified in case 9 (both mutations were previously reported in Bainbridge et al¹⁴). In case 18, a non-synonymous mutation was identified in TERF2 (p.D481G) that localizes to the DNA binding domain (MYB) of the protein, suggesting a similar functional mechanism of shelterin complex disruption as predicted in case 9. Notably, subtype-characteristic acquired alterations at TERTp and ATRX are not observed in these 3 cases, suggesting that select loss-of-function mutations to shelterin complex proteins may be sufficient to enable maintenance of the telomere, which appears to be a requisite event for gliomagenesis (Brennan et al³²; Fig. 3). However, if telomere shelterin complex mutations are sufficient to enable telomere maintenance and contribute to gliomagenesis, it might be expected that somatic mutations in these genes would be observed in glioma tumors, yet they are not; future investigations will be required to understand this distinction.

Germline mutations that may have increased predisposition to glioma development were also observed in 7 subjects in genes including TP53, ATR, MLH3, POLD1, and DMBT1. Germline mutations in TP53 have previously been linked with glioma development.^10,37,38 While germline mutations in ATR, MLH3, and POLD1 have not previously been linked with glioma, all 3 are involved in DNA damage recognition and/or repair and have been linked to hereditary cancer risk.^39–41 Deleterious mutations in DMBT1, which may play a role in immune defense,⁴² were observed in 2 cases; DMBT1 has been suggested as a potential tumor suppressor in multiple cancers and is deleted somatically in approximately 2% of glioma cases.² However, these findings must be cautiously interpreted in the absence of functional data implicating these specific variants in gliomagenesis, and future investigations will be required to determine whether mutations in genes not previously linked with familial glioma do indeed contribute to gliomagenesis in these individuals and their relatives. Notably, no candidate germline mutations were identified in 10 of 20 sequenced cases, potentially suggesting a polygenic mode of inheritance or potential interactions between genetic and shared environmental risk factors between related cases. It is also possible, albeit statistically unlikely, that multiple cases arose in some families due to chance alone.

In the 20-case analysis, we were also able to explore the potential contribution to gliomagenesis of common polymorphisms previously shown to relate to glioma risk.¹⁹ Although the sample was small, several observations can be made from these data. First, it could be hypothesized that a higher frequency of common risk alleles may account for glioma heritability in families; however, a systematic difference in frequency compared with sporadically arising cases was not observed. Secondly, we did not observe a statistically significant association when comparing the inheritance of common risk alleles with same-gene somatic alterations, suggesting that inherited alleles are not acting as “first hits” to a gene, nor are they sufficient alone to alter a given gene’s function as required for gliomagenesis among cases in this study. Thus, glioma-associated polymorphisms are unlikely driving tumorigenic events, and do not appear to play any greater role in the development of familial glioma compared with sporadic glioma, but they likely modify the expression of genes on key pathways related to gliomagenesis.

Finally, this analysis reveals that both germline and somatic molecular features target common core pathways involved in gliomagenesis. An illustration of this pattern is presented in Fig. 5, which highlights 4 molecular pathways (RTK/PI3K, telomere maintenance, retinoblastoma/cell cycle, and p53) that are impacted by both inherited and acquired glioma-associated molecular events. For example, genes related to telomere maintenance are impacted by rare germline mutations (eg, POT1), common germline risk variants (eg, RTEL1, TERT), and somatic mutations (eg, TERTp, ATRX). Glioma-associated events impacted several specific genes at both germline and somatic levels, including EGFR, TERT, CDKN2A/B, and TP53. Overall, this study demonstrates that the same pathways understood to be impacted by somatic alterations are associated with germline predisposition events in glioma.

Fig. 5.

Germline and somatic glioma-associated molecular features target common core pathways. Cells colored in red denote that the corresponding alteration type (rare germline mutation, common germline variant, somatic mutation, or somatic copy number alteration) in a given gene was observed. Asterisks denote rare glioma-associated germline mutations identified previously.^38,43,44 CNA, copy number alteration.

While this study was limited by its small sample size, it offers a unique perspective on the molecular pathogenesis of glioma by integrating multiple levels of germline and somatic molecular data on a series of cases. This study sheds light on the molecular similarities between the familial cases studied and sporadic glioma, and on the interplay of germline and somatic events on glioma etiology. Furthermore, it offers potential avenues for future work to delineate causes of glioma heritability, including studies of polygenic effects of common risk variants or interactions of rare and common variants. Additionally, given the enrichment of molecular events within core pathways, future sequencing efforts may focus on genes in these pathways to identify germline variants of intermediate frequency that are associated with gliomagenesis. Through these and other efforts to reach a greater understanding of the factors that contribute to glioma heritability, new opportunities for screening and targeted therapy may arise.

Funding

D.I.J. is supported by the CPRIT Post-Graduate Training Program in Integrative Cancer Epidemiology (award ID RP160097). Work at The University of Texas MD Anderson Cancer Center and Baylor College of Medicine was supported by the National Institutes of Health grants R01 CA119215 to M.L.B, R01 CA070917 to M.L.B, R01 CA139020 to M.L.B; American Brain Tumor Association to M.L.B, and the National Brain Tumor Society to M.L.B.

Authorship statement

Study conception and design: DIJ, JTH, MLB

Acquisition of study specimens and data: DIJ, KF, MNB, GNA, JTH, MLB

Analysis and interpretation of data: DIJ, MNB, ST, XG, HVD, JH, JTH, MLB

Writing, review, and/or revision of manuscript: DIJ, GNA, JTH, MLB

Administrative, technical, or material support: JCJ, DMM

Study supervision: JTH, MLB