Integrated clinical and molecular landscape of disseminated pediatric low-grade glioma

Abstract

Background

Pediatric-type low-grade gliomas (PLGG) are the most common central nervous system (CNS) tumor in children. Many are indolent and have excellent outcomes; however, some inexplicably spread throughout the CNS leading to increased morbidity and mortality.

Methods

To better understand this rare and difficult-to-treat entity, as well as the features associated with dissemination in CNS tumors, we assembled a large international cohort (n = 269) of patients with disseminated PLGG with detailed clinical and molecular characterization, including DNA sequencing and methylome profiling.

Results

We identified three subgroups of patients based on the temporal and spatial distribution of dissemination. Tumors with diffuse leptomeningeal spread without a primary tumor mass and those occurring in infants had the worst clinical outcomes. The genetics overlapped substantially with that of non-disseminated PLGG, suggesting that non-genetic mechanisms are an important contributor to dissemination. Therapeutically, targeted RAS/MAPK-pathway inhibition was more effective than conventional chemotherapy as first or second-line treatment.

Conclusions

In sum, this cohort increases our clinical and biological understanding of this rare disease, provides insights for improving patient care, and directs future clinical trials and basic science research.

Key Points.

• Disseminated pediatric low-grade gliomas form 3 temporo-spatial subgroups.

• Patients with diffuse tumors and infants are high-risk, with equivalent survival to standard-risk medulloblastoma.

• Use of targeted therapies improves outcomes as first or second-line therapy.

Importance of the Study

Low-grade gliomas are the most common brain tumors in children; among these, some of the most devastating are those that disseminate throughout the CNS (DLGG). DLGG are highly heterogeneous and the presentation may overlap with non-disseminated LGG, making it hard to predict who will disseminate, as well as how to best treat them. Previous studies have been limited by small numbers, inconsistent clinical data, and limited molecular characterization, which highlights the need for a comprehensive study. Here, we present an international cohort of 269 DLGG with detailed clinical and molecular annotation, representing by far the largest study every published. Our findings reveal that DLGG from 3 temporo-spatially defined subgroups, enable risk stratification and identification of the highest-risk patients, and provide strong evidence that targeted therapies improve outcomes compared to chemotherapy. This yields insights that are immediately clinically actionable as well as a foundation for future clinical trials and basic science research.

Low-grade gliomas/glioneuronal tumors (LGG) are the most common central nervous system (CNS) tumors in children.^1,2 Most pediatric-type LGG (PLGG) have excellent long-term survival, although they can incur significant morbidity related to multiple recurrences and lines of therapy.³ Some of most devastating PLGG, with substantially higher mortality, are those that disseminate in the CNS (disseminated PLGG; DLGG), which is estimated to occur in 5%-6%.^4,5 This is despite usually remaining histologically low-grade.⁶ DLGG are highly heterogeneous and their presentation and early clinical course overlaps with non-disseminated LGG. It is therefore currently impossible to determine which PLGG are at risk for aggressive behavior, how therapies can potentially mitigate outcomes, and what is the best overall management approach for these complex tumors.

PLGG are typically driven by genetic alterations in the RAS/MAPK signaling pathway, most frequently involving BRAF, FGFR1, and NF1.⁶ Importantly, many of these alterations can be specifically targeted for therapy.^7-10 In the past, treatments for unresectable PLGG consisted of cytotoxic chemotherapy and, in some cases, radiation. With improved knowledge of the biology of these tumors and increasing evidence that their use improves outcomes,^9,10 current therapeutic strategies are increasingly switching towards upfront use of targeted therapies such as BRAF, MEK, and/or pan-RAF inhibitors.

One recognized subset of DLGG is diffuse leptomeningeal glioneuronal tumor (DLGNT), which classically presents with widespread leptomeningeal coating without a primary tumor mass. Most have a RAS/MAPK pathway alteration (usually a BRAF fusion) with chromosome 1p deletion and in some cases additional 1q gain.^11–14 However, BRAF fusions are common in PLGG and 1p deletion occurs in wide range of cancers,¹⁵ so the biologic mechanism of dissemination is unclear and is not explained by genetics. Furthermore, the DNA methylation cluster “DLGNT” does not correspond precisely to the clinical-morphologic group of diffuse leptomeningeal glioneuronal tumor. Specifically, (1) some PLGG classify as DLGNT but present with a single localized mass¹⁶ and (2) classic clinical-pathologic cases fitting the above definition may not cluster/classify with methylation class DLGNT. It is unclear what the significance of DLGNT is in this context.

Previous studies on PLGG dissemination have been limited by small numbers, inconsistent clinical data collection and follow up information, and limited molecular characterization (summarized in our systematic review¹⁷). Only one study has included more than 40 patients⁵ but lacks genetic testing, while smaller studies focusing on DLGNT specifically have included more detailed molecular and survival data, but do not represent the full breadth of this disease.^12,13 Furthermore, most studies precede the use of MAPK pathway targeted therapies, and DLGG may have challenges in meeting eligibility criteria for trials of targeted therapies (due to inability to accurately measure disease or leptomeningeal disease not meeting minimum size criteria for target lesion^18,19), meaning there is limited trial data showing the effect of targeted therapy in DLGG. Therefore, while the literature has identified some important clinical and molecular features, there remains no comprehensive study that captures the breadth of this disease with sufficient data to guide the next phase of clinical studies on these patients.

There are numerous outstanding and clinically important questions regarding dissemination in PLGG,¹⁷ including (1) which tumors are at highest risk for dissemination, (2) what are the genetic and non-genetic mechanisms that contribute, (3) how are DLGG best managed and is this different than treatment paradigms for less aggressive PLGG, and (4) what is the significance of pathologic and molecular tumor subtyping, particularly when it is discordant from the clinical behavior? To answer these knowledge gaps and better understand this challenging group of patients, we compiled an international cohort of DLGG patients (n = 269) with detailed clinical, radiographic, pathologic, and molecular annotations. With this dataset, we identify clinically relevant subgroups, risk stratification, and treatment responses, which inform on current management of DLGG and provide the foundation for future study.

Methods

Clinical Data Collection

This study was approved by the Research Ethics Board (# 1000079566) at the Hospital for Sick Children. The inclusion criteria were: (1) a pediatric-type low-grade glioma and (2) dissemination, which was defined as ≥2 non-contiguous lesions or the presence of significant leptomeningeal tumor spread with or without a discrete primary mass. Clinical data (including patient demographics, dissemination pattern and timing, and treatment courses), imaging, pathology, and molecular testing were entered by collaborators in an online REDCap database (see Figure 1A for study workflow). Where available, samples were transferred to the Hospital for Sick Children for central pathology review and molecular testing. There were 269 patients who met study inclusion criteria, of which 139 samples had tissue available for additional molecular testing run at SickKids. Overall, 219/269 (81%) patients had some form of molecular testing, including 187 (70%) with testing performed by the referring site, 84 (31%) with NGS at SickKids, and 99 (37%) with methylome profiling.

Figure 1.

Disseminated PLGG form 3 spatiotemporally defined subgroups. (A) Schematic of study workflow. Created using BioRender.com. (B) Diagram of 3 spatiotemporal dissemination groups of DLGG. (C) Pie chart of breakdown of 3 patient groups in cohort. A small number of additional non-disseminated tumors that classified as DLGNT by methylation with high confidence were also included in our cohort. (D) Intraoperative endoscopic image of a primary disseminated PLGG. The temporal horn of the left lateral ventricle was accessed via a parieto-occipital approach with the endoscope directed anteriorly (screen left is lateral, top is superior). The arrow indicates a dominant lesion (T) in the anterior aspect of the temporal horn, being partially obstructed from view by the choroid plexus (CP); the asterisk indicates disseminated tumor along the lateral ependymal surface. The pathology from this biopsy demonstrated pilocytic astrocytoma with a BRAF p. V600E mutation. (E) Violin plot of age for three patient groups. Boxplots show median and interquartile range. P values by two-tailed T-test. (F) Barcharts of sex breakdown by dissemination group. P-values for diffuse versus primary = .006, diffuse versus secondary = .004, and primary versus secondary = .7 (Chi-square test). (G) Kaplan-Meier survival curves for overall survival (OS) and progression-free survival (PFS) comparing DLGG and reference cohort of non-disseminated PLGG from Ryall et al. P = 2 × 10⁻⁹ for OS and 1 × 10⁻⁹⁹ for PFS (log-rank test). (H) Dotplot showing overall survival for cohort, separated by whether patients were alive at the most recent follow-up. (I) Stacked barchart of initial therapy for DLGG cohort separated by dissemination group. P = .0001, Chi-square test.

To confirm tumors were PLGG, pathology reports were uploaded and (where available) representative slides were shipped for central review by a neuropathologist (A.L.). To confirm that tumors qualified as disseminated, radiology reports and representative images were uploaded and reviewed by study investigators (A.L., J.B., M.D.). For analysis of medical therapy, regimens were grouped according to the main agent used, primarily as “vinblastine-based” and “carboplatin-based.” For comparison of with non-disseminated PLGG, we used our previously published population-based cohort of 1000 PLGG.⁶ Please see Supplemental Methods for further details regarding the data collection and review.

Molecular Testing

NGS panel sequencing.—DNA was extracted from FFPE blocks using DNAstorm (CellData) and libraries generated using custom hybrid capture probes (Twist Bioscience) targeting commonly altered glioma driver genes. Samples were sequenced on NextSeq instruments (Illumina) and fastq files aligned to hg19 using bwa (v2.2.1). SNVs and INDELS were detected using Mutect2²⁰ and structural variants were identified using Arriba²¹ (v2.4.0), and Manta.²²

Methylation analysis.—Genome-wide methylation profiles were obtained using Infinium MethylationEPIC Array (mainly v1 chips, some v2 for more recent samples). FFPE tissue had a minimum of 200 ng of genomic DNA bisulfite-converted and assayed on Infinium MethylationEPIC kit, and scanned on the NextSeq-550, all per manufacturer protocols. Raw methylation files were imported, pre-processed, and normalized (ssNoob method) using minfi²³ (v1.44.0). CNV profiles were inferred using conumee²⁴ (v1.34.0) and heatmaps were created using ComplexHeatmap (v2.15.1). Samples were classified using the DKFZ CNS tumor methylation classifier (version 12.5 for v1 chips and 12.8 for v2 chips) primarily,²⁵ as well as the NIH Bethesda v2 classifier.

Circulating tumor DNA (ctDNA) analysis.—NGS from ctDNA in CSF was performed as previously described.²⁶ In brief, CSF samples were collected in plain sterile collection tubes or Streck tubes, cfDNA was extracted using the QIAamp Circulating Nucleic Acid Kit, and up to 30 ng used as input for library preparation. Libraries were prepared using Illumina TruSight Oncology 500 ctDNA v2 and Twist library preparation kits. Hybrid capture was performed using 500 ng library DNA with customized 21-gene targeted panel. Sequencing was run on a NextSeq 550 and raw reads were aligned to GRCh37 using bwa-mem v0.7.12. CNV analysis was performed using ichorCNA²⁷ and SNV and fusion calling was run in BaseSpace with DRAGEN v.3.10.4.

Single-Nucleus RNA Sequencing

Snap frozen tissue samples (5-15 mg) were thawed in HB buffer and mechanically homogenized. A discontinuous iodixanol density gradient was prepared in a separate tube, the tissue homogenate was layered, and nuclei isolated by centrifugation at 10 000 × g for 18 min, with the fraction at the 30%-40% iodixanol interface collected. Using an estimated 10 000 nuclei, single-nucleus encapsulation was performed using the 10× Genomics Chromium platform, libraries generated using 10× Genomics multiomics kit, and sequencing run on Illumina NovaSeq 6000 S4 platform. Raw sequencing reads were processed using the CellRanger v7.0.1 pipeline, with alignment to GRCh38. The count matrix was processed using Seurat²⁸ (v4.3.0.1) for quality control, clustering, and dimensionality reduction. Cell type annotation was performed using SingleR²⁹ (v2.2.0) labelling in comparison to reference datasets.^30,31 Oncogenic pathway activation scores were calculated using the PROGENy³² Bioconductor package (v1.20.0).

Statistical Analysis

No statistical method was used to predetermine sample size. Analysis was performed using R, with statistical tests reported in the figure legends. Univariate survival analysis of known prognostic factors in PLGG and the dissemination group was performed using the Kaplan-Meier method; ­multivariate analysis was performed on the most significant univariate predictors using Cox proportional hazards ­models. Progression-free survival for the first treatment course defined as the time from initial surgery to first progression, and for additional treatment courses as the time between progression events. Progression events were defined as clinical and/or radiographic changes leading to alteration in management, based on the discretion of study investigators.

For further details regarding the clinical data collection and review, molecular testing protocols, data analysis, and availability, please see the online Supplemental Methods.

Results

Overall Cohort Clinical Presentation, Treatment, and Outcomes

We assembled a diverse cohort of patients from 39 sites in 13 countries across all continents (see FigureS1A for map of sites and S1b for patient recruitment). Patients were diagnosed between 1988 and 2023, although the year of diagnosis did not impact survival (Figure S1C and D). The median age at diagnosis was 4.5 years (range 1 month to 44 years) and 52% (n = 140) of patients were male. Germline cancer predisposition was relatively rare—10 patients (4%) had neurofibromatosis type 1 and 3 (1%) had Noonan syndrome. The median follow-up period was 5.9 years (range 0-35 years), with 41 patients (15%) who died during the follow-up period. Most patients had at least one episode of progression (n = 229/259 patients, 88%). CSF diversion was required in 54.6% (n = 147/269) of patients at some point during the disease course, with 31.2% (n = 84/269) within one month or less from diagnosis. As has been previously described in DLGG,⁴ there were elevated CSF protein levels in many patients, with a median of 91 mg/dL (SD = 465 mg/dL; n = 73 patients with data available; Figure S2A), recognizing that reference ranges vary slightly between different labs.

Based on initial imaging and clinical course we identified 3 groups of spatiotemporal growth in DLGG (Figure 1B and C): (1) primary DLGG with a clear dominant/primary lesion and concurrent lesions elsewhere in the CNS (n = 139/269, 52%; Figure 1D), (2) secondary DLGG, who presented with a single localized mass and had dissemination at a later date (n = 84, 31%), and (3) diffuse DLGG with primary leptomeningeal or other dissemination and without a dominant lesion (n = 42, 16%). Importantly for consideration of patient surveillance, nearly a third of patients were in the secondary dissemination category. Diffuse DLGG were older at diagnosis (Figure 1E) with a median age of 7.0 years, compared with 3.8 years for primary (P = .02 vs diffuse) and 4.1 years for secondary DLGG (P = .002). Diffuse DLGG had a male predominance with only 26% female patients (Figure 1F), compared to 52% for primary (P = .006) and 55% for secondary (P = .004). CSF diversion rates were not significantly different between groups, with 51% (n = 71) of primary, 60% (n = 50) of secondary, and 60% (n = 25) of diffuse requiring (P = .4, Chi-square test).

For comparison of clinical, pathologic, and genetic features with non-disseminated PLGG we used the population-based control cohort of >1000 PLGG previously published by Ryall et al.⁶ DLGG had dramatically worse progression-free (PFS; 18% at 5 years; P < .0001) and overall survival (OS; 89% at 5 years, 81% at 10 years; P < .0001) than non-disseminated PLGG (PFS 73% at 5 years; OS 96% at 5 years, 94% at 10 years; Figure 1G). Out of 41 patients who died, 39 (95%) were within 10 years of diagnosis (Figure 1H). While this observation is subject to follow-up biases (69 patients [26%] had >10-year follow-up), it suggests that beyond this time point, this disease may stabilize.

Most patients (58%; n = 155/265) underwent surgery followed by adjuvant therapy (either conventional chemotherapy or targeted agents) as first-line treatment (Figure 1I). Less frequently, patients received only surgery (30%; n = 79) or only chemotherapy (5%; n = 14). Secondary DLGG had lower rates of upfront chemotherapy use (46% total), compared to primary (73%) and diffuse (70%). Surgically, most patients had either a biopsy (46%; 118/258) or partial resection (30%; n = 78) of the dominant lesion (Figure S2b). Fewer patients had a near total (NTR; 13%, n = 33) or gross total resection of the dominant lesion (GTR; 5%, n = 12), which was associated with improved OS (P = .03) compared to biopsy or partial resections, even when diffuse tumors were excluded (P = .06; Figure S2C). Upfront radiation (with surgery ± chemotherapy) was uncommon (7%; n = 19), and only 24% of patients ever received radiation in their treatment course, with a slightly greater proportion of craniospinal (13%) compared to focal fields (7%; Figure S2D-F). Recognizing that this is highly biased by patient selection, there was worse survival for patients receiving craniospinal vs focal radiation (P = .05; Figure S2F), as well as a trend toward worse survival for patients receiving upfront radiation (Figure S2G; P = .09 vs no upfront radiation).

Molecular and Clinical Features of Three DLGG Subgroups

Characteristics of Primary DLGG.—To better understand the biology and features of DLGG classification, we performed integrated clinical, morphological, and genomic analysis of DLGG according to their clinical subgroups. Over half of the patients in our cohort (n = 139; 52%) had a clear dominant mass (the presumed originating lesion) and additional metastatic lesions, termed primary dissemination (primary DLGG). Primary DLGG had a striking predilection for the dominant mass occurring in the suprasellar/optic pathway region, which comprised 55% of primary DLGG compared to 18% of overall PLGG (Figure 2A, Figure S3A; p < 2.2*10⁻¹⁶, Chi-square test).

Figure 2.

Clinicopathologic features of primary disseminated PLGG show enrichment for suprasellar dominant location. (A) Barchart comparing locations for primary DLGG with non-disseminated PLGG from Ryall cohort. P < 2.2 × 10⁻⁶, Chi-square test. (B) Violin plot comparing age at diagnosis for the 4 most common locations of primary DLGG. Boxplots show median and interquartile range. P-values by Wilcoxon test with Holm adjustment. (n = 14 hemispheric, 21 posterior fossa, 14 spinal, and 75 suprasellar). (C) Violin plots comparing age at diagnosis for DLGG versus non-disseminated PLGG (from Ryall et al.) for suprasellar and posterior fossa tumors. n = 74 disseminated and n = 171 non-disseminated for suprasellar; n = 21 disseminated and n = 265 non-disseminated for posterior fossa. P-value by Wilcoxon test. (D) Stacked barchart comparing pathologic phenotype by dominant location. “Astrocytic” includes tumors with diagnosis of pilocytic/pilomyxoid astrocytoma, optic glioma, PXA, and diffuse/low-grade astrocytoma. “Glioneuronal” includes diagnosis of DLGNT, ganglioglioma, DIA/DIG, DNT, RGNT. “Non-specific” refers to tumors with only a diagnosis of LGG/GNT given. P =1 .6 × 10⁻⁷ by Chi-square test, comparing the four locations. (n = 14 hemispheric, 20 posterior fossa, 14 spinal, and 75 suprasellar). (E) Oncoplot of driver alterations stratified by dominant location. (F) Pie chart of driver alterations in primary DLGG (top) and non-disseminated PLGG from Ryall cohort (bottom). (G) Alluvial plot comparing pathologic diagnosis and methylation classification by DKFZ algorithm. y-axis is number of samples. (H) Kaplan-Meier curve comparing survival for diffuse DLGG, and non-diffuse DLGG (both primary + secondary) in patients <1 year age and >1 year age. P-value by log-rank test. CNV, copy number variant; DIA/DIG, desmoplastic infantile astrocytoma/ganglioglioma; DLGNT, diffuse leptomeningeal glioneuronal tumor; DNT, dysembryoplastic neuroepithelial tumor; LFS, Li Fraumeni syndrome; LGG/GNT, low-grade glioma/glioneuronal tumor; NF-1, neurofibromatosis type 1; PA, pilocytic astrocytoma; PXA, pleomorphic xanthoastrocytoma; RGNT, rosette forming glioneuronal tumor; SNV, single nucleotide variant; SV, structural variant.

Suprasellar tumors occurred more commonly in younger patients (median age = 1.8 years; Figure 2B) compared to tumors originating in the next three most common locations—posterior fossa (16%; median age = 9.7 years), cerebral hemispheres (12%; median age = 10.1 years), and spinal cord (10%; median age = 4.4 years). Suprasellar tumors that disseminated occurred in younger patients compared with non-disseminated suprasellar PLGG (median age 1.8 versus 6.8 years, respectively; P = 9*10⁻¹², Wilcoxon; Figure 2C), while posterior fossa PLGG (the most common overall location) did not show a significant difference in age between disseminated (median age = 9.7 years) and non-disseminated tumors (median age = 7.9 years; P = .12, Wilcoxon). Spinal tumors had worse OS than those in other locations (P = .04, log-rank test; Figure S3B), although all locations had similar PFS (P = .26). The most common pathologic diagnosis (Figure 2D,E, Figure S3C) was pilocytic/pilomyxoid astrocytoma (n = 91; 66%). In general, tumors located in the suprasellar region and posterior fossa had predominantly an astrocytic phenotype, while those in the cerebral hemispheres and spine had a greater proportion with glioneuronal histology (Figure 2D, P = 2 × 10⁻⁸, Chi-square test).

As expected, most primary DLGG harbored RAS/MAPK pathway activating alterations (Figure 2E andF). The most frequent molecular alteration was BRAF fusion which was present in 53% of patients with molecular testing available (n = 61), compared with 30% of non-disseminated PLGG. BRAF p. V600E occurred in similar proportions (n = 19, 17% of DLGG; 19% of non-disseminated PLGG), while germline NF1 alterations were rare in primary DLGG (n = 6, 5%; compared to 21% of non-disseminated PLGG). There were no recurrent CNVs, and 1p deletion, 1q gain, and CDKN2A deletion were all rare. Methylation classification was generally highly concordant with histologic diagnoses (Figure 2G), with the most common classification being location-specific subsets of pilocytic astrocytoma (n = 29; 62%). As has previously been reported in PLGG,³³ even with permissive score cutoffs there were several unclassified samples (n = 4; 9%) and misclassifications as control tissue (n = 3, 7%) using the methylation classifier.

Younger age at presentation was a significant marker of worse outcome. Infants (age <1 year) with primary or secondary DLGG had dramatically worse overall survival than patients >1 year of age at presentation (10 y OS 68% vs 87%, P = .003, log-rank test; Figure 2H), identifying this as one group of high-risk DLGG. Infants with DLGG were virtually all suprasellar in location (n = 42/47; 89%). They were mainly primary disseminated (n = 34/47; 72%: Figure S3e), with no survival difference between those that were primary versus secondary disseminated (Figure S3f).

Characteristics of PLGG with Secondary Dissemination.—Nearly a third of patients (n = 84, 31%) had secondary dissemination, that is, they presented with a single lesion and developed disseminated disease at progression (secondary DLGG). Most secondary DLGG had significant residual disease (partial resection in 42%, biopsy in 25%, no surgery or CSF diversion only in 10%; Figure 3A, S4a). However, occasional cases had GTR (6%, n = 5; see Figure S4B and C for example cases) or NTR (17%, n = 14), most of whom did not receive adjuvant therapy (n = 2/14 NTR patients received chemotherapy and 2 received radiation). In comparison, 40% of non-disseminated LGG (Ryall et. al⁶ cohort) had a GTR (Figure S4A). Therefore, in our cohort, secondary dissemination is rare in patients with local disease control and adjuvant therapy—although among all PLGG most tumors with subtotal resections do not disseminate.⁶

Figure 3.

Clinicopathologic features of PLGG with secondary dissemination indicate peak risk within 5 years of diagnosis. (A) Nested pie chart of initial treatment and extent of surgical resection for secondary DLGG. Note that 2 patients in the “surgery, radiation” group also received chemotherapy but are grouped here for clarity. (B) Histogram of time from presentation to dissemination. (C) Representative MRI images (T2 FLAIR, T1 with contrast) of brain and spine showing a DNT with intraventricular and subsequently spinal dissemination (DLGG40). (D) Stacked barcharts comparing dominant location for DLGG with primary versus secondary dissemination. P = .04 by Chi-square test, comparing suprasellar/spine/posterior fossa/hemispheric/other locations. (E) Pie chart of pathologic diagnosis. (F) Oncoprint of genetic and methylation results for patients with available testing. SNV: single nucleotide variant; CNV: copy number variant; SV, structural variant. (G) Pie chart of driver alterations. (H) Pie chart of methylation classifier results. (I) Kaplan Meier curves of OS and PFS for secondary DLGG patients with germline-NF1 versus without cancer predisposition syndrome. P = .57 for OS and P = .2 for PFS, log-rank test. DIG/A, desmoplastic infantile ganglioglioma/astrocytoma; DLGNT, diffuse leptomeningeal glioneuronal tumor; DNT, dysembryoplastic neuroepithelial tumor; LGG/GNT, low-grade glioma/glioneuronal tumor; PA, pilocytic astrocytoma; PXA, pleomorphic xanthoastrocytoma; RGNT, rosette forming glioneuronal tumor.

Among 81 patients with adequate documentation to confirm timing of dissemination, the median time between initial radiographic diagnosis and clear disseminated disease was 2.7 years (Figure 3B and C). Overall, 68% (n = 55/81) of patients had dissemination occur within 5 years of diagnosis, with 26% (n = 21) patients disseminating between 5-10 years and only 6% (n = 5) more than 10 years after the initial diagnosis.

In addition to the clear distinction in clinical behavior and timeline between primary and secondary DLGG, we observed differences in the locations and genetics. Compared to primary DLGG, secondary DLGG had a more diverse range of originating locations; specifically, a greater proportion of tumors arising in the brainstem, ventricles, diencephalon, and tectum (Figure 3D; P = .04 comparing suprasellar/hemispheric/post fossa/spine/other locations, Chi-square test). Secondary DLGG had lower rates of leptomeningeal spread and higher rates of intraventricular dissemination (Figure S5A and B). There was shorter time to first progression for patients who developed extensive leptomeningeal spread (P = .04 vs focal or no spread; Figure S5C) and a trend towards inferior overall survival (P = .15).

Pathologic diagnosis was similar, with 64% (n = 52) of patients having pilocytic/pilomyxoid astrocytoma (Figure 3E and F), as were the methylation classifier results, which were again concordant with histology (Figure 3H). In terms of genetics, BRAF p. V600E was rare in these patients (8%, compared to 17% of primary DLGG). The second most common driver (after BRAF fusion) was germline NF1 loss in 13% (n = 8; Figure 3F and G). While the small number of patients limits statistical robustness, there appeared to be a trend toward improved outcomes for NF1-driven compared to other secondary DLGG (5-year PFS 38% for NF1 vs 16% for non-NF1, P = .2; Figure 3I).

Characteristics of Diffuse DLGG.—Only 16% (n = 42) of patients had diffuse DLGG, which did not have a clear dominant lesion. Patients with diffuse DLGG had significantly worse OS than primary/secondary DLGG (5-year OS 71% vs 92%, P = .02, log-rank test; Figure 4A), identifying this as the second high-risk subgroup of DLGG (along with infants with primary/secondary DLGG; see Figure S10 for multivariate survival analysis). Interestingly, although numbers were small, diffuse DLGG patients had few deaths beyond 5 years from diagnosis, while there was no plateau in the survival of primary/secondary DLGG, suggesting ongoing tumor changes. There was no difference in PFS between the three subgroups (P = .44, Figure 4A), as essentially all patients progress within 5 years.

Figure 4.

PLGG with diffuse growth have enrichment for oligodendroglial histology and chromosome 1p deletion. (A) Kaplan Meier curves of OS and PFS comparing diffuse, primary, and secondary DLGG. P = .02 for OS of diffuse vs. primary/secondary DLGG, log-rank test. (B) Alluvial plot comparing histologic phenotype with pathologic diagnosis. (C) Representative brain and spine MRI images of a DLGNT (DLGG036). From top left clockwise, sequences are T2 FLAIR, T2 FLAIR, T2, and T2. (D) Representative histology and immunohistochemistry images of a DLGNT showing cells are Olig2+/synaptophysin+/GFAP-(DLGG291). (E) Oncoprint of genetic alterations and methylation classifications. (F) Pie chart of driver genetic alterations. (G) Pie chart of 1p deletion rates. (H) Pie chart of DKFZ methylation classifier results. CNV, copy number variant; DLGNT, diffuse leptomeningeal glioneuronal tumor; HGAP, high-grade astrocytoma with piloid features; LGG/GNT, low-grade glioma/glioneuronal tumor; PA, pilocytic astrocytoma; PHGG, pediatric-type high-grade glioma; SNV, single nucleotide variant; SV, structural variant.

Histologically, 80% (n = 32) of diffuse DLGG had oligodendroglial morphology, denoted by cells with uniform circular nuclei, as opposed to astrocytic morphology (18%, n = 7), which has cells with more elongated and variable nuclear morphology (Figure 4B). Most patients (n = 31, 74%) had an integrated pathologic diagnosis (incorporating histology and molecular) of diffuse leptomeningeal glioneuronal tumor (DLGNT; Figure 4B,E). This entity classically has diffuse leptomeningeal sugar-coating of the brain and spinal cord (although growth patterns can be variable) and histologically has oligodendroglial cells that are Olig2+/synaptophysin +/GFAP- by IHC (Figure 4C and D).

As shown by previous work,¹² there were high rates of BRAF fusion (19/36 tested; 53%; Figure 4E and F) and 1p deletion (20/30 tested; 67%; Figure 4G, Figure S6A). That said, 1p deletion was by no means exclusive to diffuse DLGG, as 1/3 of cases with this alteration (11/33) were in other subgroups (Figure S6B). No survival difference was observed in patients with/without 1p deletion (Figure S6C). Chromosome 1q gain was present in 11/24 tested (46%). While this has been reported as a high-risk marker in DLGNT,³⁴ we also did not identify a significant difference in OS or PFS between diffuse DLGG with/without this alteration (Figure S6C).

Diffuse DLGG most frequently were classified as DLGNT by methylation arrays (28%, 5/17; Figure 4E andH) although 3/17 (18%) were classified as control tissue. In contrast, of all tumors classified as DLGNT by methylation (n = 13), 6 (46%) were diffuse (Figure S7A), 3 were DLGG with primary/secondary dissemination, and 4 tumors were localized disease at presentation and during reported follow-up (3 in spinal cord and 1 in the frontal lobes; Figure S7B-E). Overall, fewer than a half of DLGG classified as DLGNT by methylation belonged to the diffuse group.

DLGG Undergoing Malignant Transformation Incur Additional Genetic and Epigenetic Alterations

Overall, 16 (6%) DLGG underwent high-grade transformation, denoted by increased cellularity, cytologic atypia, and proliferative rate (Figure 5A). Of these, half (n = 8) had high-grade histologic features noted at presentation (but in a tumor with low-grade precursor, see Supplemental Methods under “Pathology review” for further explanation), while the other half showed these features on repeat surgeries at progression, ranging from 1 to 23 years from diagnosis (Figure 5B). While the survival of these cases from diagnosis was marginally different from our overall cohort (P = .06, log-rank test; Figure 5C), from the time of high-grade transformation there was increased risk, with the majority of patients dead by 10 years (OS 52% at 5 years and 26% at 10 years; P = 1.4 × 10⁻⁵, log-rank test comparing OS from transformation vs overall cohort survival).

Figure 5.

High-grade transformation of disseminated LGG is associated with additional genetic alterations. (A) Representative histology images of DLGG with high-grade histologic features, including high cellularity, cytologic pleomorphism/atypia, and elevated mitotic rate. Cases are DLGG181, DLGG233, DLGG264, DLGG267 clockwise from top right. Arrows point to mitotic figures. (B) Dotplot showing time from diagnosis to transformation. (C) Kaplan Meier curves comparing OS from diagnosis and from timepoint of high-grade transformation, compared to overall cohort. P = .06 comparing OS from diagnosis for transformed patients vs non-transformed patients. P = .003 comparing OS from time of transformation versus OS for non-transformed patients (from diagnosis). P-values by long-rank test. (D) Oncoprint of genetic and epigenetic alterations in transformed DLGG, including longitudinal sampling where available. H3K27me3 loss is on IHC as assessed by referring neuropathologist. CNV, copy number variant; DLGNT, diffuse leptomeningeal glioneuronal tumor; GBM, glioblastoma; LGG/GNT, Low-grade glioma/glioneuronal tumor; SNV, single nucleotide variant; SV, structural variant. (E) Barcharts showing rate of additional high-risk genetic/epigenetic alterations in transformed versus non-transformed DLGG (specifically CDKN2A deletion, ATRX mutation, EGFR mutation, chromosome 1q gain, TERT promoter mutation, and H3 K27me3 loss). (F) T-distributed stochastic neighbor embedding dimensionality reductions of single-nucleus RNAseq on two DLGNT samples, one of which remained low-grade (left) and the other that developed anaplastic features (right). Cell type annotations based on match to developmental CNS reference atlases (see “Methods” section). Tumor cells outlined with dash lines, comprising OPC-like and proliferating OPC-like cells. OPC, oligodendrocyte precursor cell. (G) Violin plots showing PROGENy oncogenic pathway scores for MAPK, P53, and PI3K pathways in different cell types for snRNAseq data from transformed sample in (F). Boxplots show median and interquartile range. P value by two-tailed T-test.

Second-hit alterations in tumor suppressor genes were identified in this group of tumors including CDKN2A deletion (n = 4), ATRX mutation (n = 1), and TERT promoter mutation (n = 1), as well as EGFR activating mutations (n = 1) and chromosome 1q gain (n = 8) (Figure 5D). In particular, 1q gain was highly associated with transformation, occurring in 8/13 (62%) transformed tumors with testing, compared to 6/103 (6%) non-transformed (P = 8 × 10⁻⁸, Chi-square test). The epigenetics also changed in some cases—2 patients had methylation classifications of “glioblastoma, IDH-wildtype,” one of whom had progressed from an initial classification of DLGNT. Three patients demonstrated loss of H3K27me3 (in the absence of an H3 K27M mutation), expanding the spectrum of tumors in which this epigenetic alteration is observed. Overall, 69% of high-grade patients had at least 1 high-risk genetic/epigenetic feature, compared to 4% of non-transformed patients (Figure 5E).

To better understand how intra-tumoral heterogeneity contributes to high-grade progression, we performed single-nucleus RNA sequencing (snRNAseq) on one sample of anaplastic DLGNT from a patient who progressed on trametinib (MEK inhibitor) and ultimately died from disease, comparing it to a sample from a non-transformed DLGNT, both with KIAA1549::BRAF fusions (Figure 5F, S8a-f). In both cases, most tumor cells matched to an oligodendrocyte precursor cell (OPC)-like phenotype, in keeping with the oligodendroglial morphology observed histologically. However, in the high-grade case, there was a subset of cells with a more proliferative phenotype (“proliferating OPC”) with elevated cell-cycle scores, which were not present in the low-grade sample (Figure S8B and E). While we hypothesized that these proliferating cells would have activation of alternate oncogenic pathways such as PI3K, they instead had even greater MAPK-pathway activity relative to other tumor cells (Figure 5G; P < 2.2 × 10⁻¹⁶). Furthermore, the proliferating OPC cluster had depressed p53 signaling in the absence of a known TP53 mutation (Figure 5G, Figure S8F andG), implying a non-genetic mechanism of pathway suppression.

To investigate epigenetic regulation mechanisms, we compared genome-wide DNA methylation profiles on DLGG with (n = 7) versus without (n = 77) high-grade transformation. While overall methylation levels were not different between the 2 groups (Figure S8H), transformed DLGG had hypomethylation at CpG sites associated with both TP53 (P = 1 × 10⁻⁴) and its primary negative regulator MDM2 (P = .0018; Figure S8I). As p53 activity is heavily regulated at the protein-level (rather than transcriptional level),³⁵ we hypothesize that increased MDM2 expression due to hypomethylation is a possible mechanism of reducing p53 activity, and as a result avoiding oncogene-induced senescence in the cells with highest MAPK pathway activity.

MAPK-Pathway Targeted Therapy Improves Outcomes at Diagnosis and Progression

A key question underlying this study was what is the optimal therapeutic strategy for patients who present with DLGG. While the approach among neuro-oncologists for treatment of incompletely resected or growing PLGG has shifted to increasing use of MAPK-pathway targeted therapy (BRAF, pan-RAF, and/or MEK inhibition), rather than conventional chemotherapy,³⁶ it remained unclear whether this would be as effective in patients with disseminated PLGG.

While many patients had multiple courses of chemotherapy (Figure 6A) and several different agents (median = 2), the most common initial regimens were carboplatin-based (usually carboplatin + vincristine) in 43% (n = 114/218 patients; Figure 6B), and vinblastine-based regimens (21%; n = 54/218). The use of temozolomide, TPCV, targeted agents, and other regimens was rare as first-line treatment. There was a trend suggesting better PFS in those treated with carboplatin versus vinblastine-based regimens (P = .07; Figure S9A). As this cohort compiled patients from over two decades, only 13 patients were treated with upfront MAPK pathway inhibitor therapy (MEK, BRAF, and/or pan-RAF inhibition). These patients had dramatically better PFS than the 180 patients treated with upfront chemotherapy (median PFS 5.9 vs 1.8 years, P = .056, log-rank test; Figure 6C,D). While numbers were small, there were better outcomes for patients treated with BRAF ± MEK inhibitors (n = 6) versus MEKi only (n = 6; P = .05, Figure S9B).

Figure 6.

RAS/MAPK pathway targeted therapies improve outcomes compared to conventional chemotherapy. (A) Dotplot of number of lines of medical therapy. Each dot represents one patient. Median = 2 lines of therapy. (B) Pie chart of first-line chemotherapy regimen used in patients who received upfront chemotherapy. Chemotherapy regimens described further in “Methods” section. TPCV, thioguanine, procarbazine, lomustine, and vincristine. (C) Kaplan-Meier curve comparing PFS for patients treated with first-line chemotherapy versus first-line MAPK-pathway targeted therapy. P = .056, log-rank test. Inset pie-chart shows dissemination category for 13 patients on upfront targeted therapy. (D) Pie chart of targeted agents used in c. MEK includes 5 patients on trametinib, 1 on selumetinib. BRAF includes 3 on dabrafenib, 1 on vemurafenib, and 2 on dabrafenib + trametinib. Pan-RAF is 1 patient on tovorafenib. (E) Kaplan-Meier curve of PFS for patients treated sequentially with chemotherapy then targeted therapy, as well as non-disseminated LGG at SickKids treated with targeted therapy (mainly second line). Inset shows proportion of patients who progressed on versus off treatment, out of 32 patients who progressed on targeted therapy. PFS is calculated in reference to the initiation of each respective treatment. P = .0003 comparing DLGG on TT versus chemotherapy; P = .02 comparing DLGG versus non-disseminated LGG on targeted therapies (by log-rank test). (F) Pie chart of target agents used in e. MEK includes 49 patients on trametinib, 10 on selumetinib, 4 on binimetinib. BRAF includes 2 on dabrafenib, 1 on vemurafenib, and 2 on dabrafenib + trametinib. Pan-RAF is 4 patients on tovorafenib. mTOR is in 2 patients on everolimus. Other includes 1 patient on the ALK inhibitor LDK-378, 1 on alpelisib (PI3K inhibitor), and 1 on permagitinib (FGFR inhibitor). (G) Oncoprint of NGS results on CSF ctDNA from disseminated LGG patients. 7 patients are included, one with sampling at 2 timepoints. BRAF fusions are all KIAA1549-BRAF. CSF, cerebrospinal fluid; DLGNT, diffuse leptomeningeal glioneuronal tumor; EVD, external ventricular drain; LP, lumbar puncture; SNV, single nucleotide variant; TKD, tyrosine kinase domain duplication. (H) Stacked barcharts showing CSF positivity (as determined by driver alteration identified on NGS) for disseminated and non-disseminated LGG, separated by method of CSF sampling. (I) Kaplan Meier curve comparing survival outcomes for 2 high-risk DLGG subtypes (primary/secondary DLGG with age <1 year and diffuse DLGG), with standard risk DLGG (remainder of cohort), and other pediatric brain tumor types. P-value by log-rank test. SickKids reference cohorts of standard-risk medulloblastoma and posterior fossa ependymoma were used for comparison (see “Reference cohorts” in “Methods” section for further information). Also see Figure S10 for multivariate analysis showing these 2 subtypes are the only significant prognostic factors in our patient cohort.

We identified 78 patients who were treated sequentially with chemotherapy followed by targeted therapy upon progression, comparing PFS on each agent (Figure 6E and F; also see Figure S9H for clinical/molecular features). These patients had substantially longer PFS on targeted therapy, with median PFS of 2.7 years on targeted therapy vs 1.7 years on chemotherapy (P = .0003, log-rank test)—although still worse than non-disseminated PLGG on second-line targeted therapy at our institute (median PFS = 5.4 years, P = .02). Progression on targeted therapy occurred at similar rates during and after completion of treatment (Figure 6E inset). This benefit was also observed when comparing all patients on first-line chemotherapy and second-line chemotherapy to second-line targeted therapy (Figure S9C and D). Indeed, even some patients who progressed rapidly on chemotherapy had durable long-term responses to targeted therapy, demonstrating the potential for salvage treatment for patients who may be considered medically intractable otherwise.

Importantly, not all DLGG groups benefited equally from targeted therapy. The 10 diffuse DLGG patients who received sequential chemotherapy and targeted therapy (out of 78 in Figure 6E) did not have better survival on targeted therapy (P = .3; Figure S9E and F). While the numbers are small, this highlights this patient group as a priority for developing alternate treatment strategies. Conversely, there remained a strong survival benefit for the high-risk infant patients on targeted therapy (P = .002, Figure S9G).

Non-Invasive Identification of Targetable Genetic Alterations

Given the benefits of targeted therapy it is critical to identify the specific genetic alteration (as effective targeted therapy differs for BRAF fusions, BRAF p. V600E, FGFR alterations, etc). We hypothesized that this could be done non-invasively through circulating tumor DNA (ctDNA) analysis of CSF, as has been previously reported in CNS tumors with leptomeningeal disease.³⁷ We analyzed 8 CSF samples by NGS (see Methods for further details) from 7 DLGG patients ­(Figure 6G), including 5 DLGNTs and 2 pilocytic/optic gliomas, which were compared to 30 CSF samples from non-disseminated PLGG.

In 6/7 patients with DLGG a driver alteration was identified in their CSF sample, including both SNVs and structural variants in BRAF and FGFR, as well as copy number alterations (Figure 6H, Figure S11). Strikingly, 75% (n = 3/4) lumbar puncture samples from DLGG were positive for a targetable alteration, while LP sampling was negative in all 7 non-disseminated PLGG samples (Figure 6H). Additionally, all 3 samples from intraventricular collection (shunt/EVD) in DLGG were positive. This indicates that for DLGG, unlike for PLGG in general, CSF sampling either from LP or during CSF diversion procedures is a viable method for (1) identifying therapeutically targetable alterations in cases that are not amenable to surgical biopsy or resection and (2) longitudinal sampling to identify mechanisms of treatment resistance (ie secondary alterations).

Discussion

This study highlights the diverse range of PLGG that can disseminate through the CNS. This occurs in 3 patterns: (1) primary DLGG with a clear dominant/primary lesion, (2) secondary DLGG with presentation as a localized mass with later dissemination, and (3) diffuse DLGG with primary leptomeningeal or other dissemination and lacking a dominant lesion. Primary and secondary DLGG had high rates of dominant lesions in the suprasellar region and most had astrocytic histology. Both had high rates of BRAF fusion, while BRAF p. V600E was largely limited to primary DLGG. Diffuse DLGG occurred in older children, who were predominantly male. These had oligodendroglial histology and high rates of chromosome 1p deletion. While some findings here have been previously reported (such as the potential for primary vs secondary dissemination and high rates of BRAF fusion),¹⁷ this study is the first to comprehensively profile DLGG with sufficient clinical and molecular detail to understand the full spectrum of this disease and provide the rationale for future biologically-informed clinical studies.

Previous studies have identified genetic risk factors for malignant transformation in PLGG, particularly BRAF p. V600E with concomitant CDKN2A deletion.^38,39 Other than a small proportion of tumors that underwent high-grade transformation, most of our cohort did not have additional alterations in tumor suppressor genes. Further, other than the subset of diffuse DLGG with chromosome 1p deletion, most cases do not have any genetic alterations or methylation profiles to distinguish from non-disseminated PLGG. This is critical for physicians treating and studying PLGG patients to recognize—histology, genetics, and methylation do not predict dissemination, and factors outside these need to be further investigated. Our analysis also illustrates the potential discrepancies that can occur between methylation-based classifications and clinical behavior, which is relevant both for this patient cohort and more broadly for the use of methylation as a diagnostic tool in CNS tumors.⁴⁰

There are several immediately actionable points that can alter the clinical approach to these patients. First is the recognition that approximately 1/3 of our cohort presented with localized disease and only disseminated at progression. In the majority, this occurred within 5 years from initial diagnosis, but rare patients disseminated over 10 years from the time of initial diagnosis. This is in keeping with previous work from our group indicating that beyond the first decade from diagnosis, in the absence of second-hit genetic alterations, tumors often stabilize and have increased senescence.⁴¹

Second is the identification of two high-risk groups of DLGG—specifically tumors with diffuse spread (diffuse DLGG) and primary/secondary DLGG occurring in infants, which have equivalent survival to standard-risk medulloblastoma (Figure 6I). In multivariate survival analysis these 2 factors (specifically diffuse category and age < 1) were the only significant prognostic variables, while factors such as genetic driver, dominant location, pathology, extent of resection, and sex were not significant. Given the striking enrichment for suprasellar tumors among DLGG (55% of primary DLGG vs 18% of our populated-based cohort of 1000 PLGG6), with most occurring in patients under 3 years of age (median = 1.8 years), when possible we recommend upfront craniospinal imaging and increased surveillance in this patient group.

Third is the evidence that targeted therapies against the RAS/MAPK pathway are more effective than chemotherapy, either when used upfront, or as second line in patients that progressed on chemotherapy—and can produce long-term responses in the latter. Notably, few patients were treated with radiation and those who received CSI had poor outcomes, in contrast to some prior reports.⁴² While this may reflect the more aggressive nature of those tumors receiving CSI rather than the effect of the treatment itself, treatment with targeted therapy should be strongly considered in this scenario. Given that identification of a driver alteration is necessary before initiating targeted therapy, this indicates a critical role for surgical biopsy or potentially CSF-based analysis for guiding therapy. Further long-term study is needed to determine if targeted therapy also improves overall survival, as PFS was used for our analysis. As such, our large, multinational study team suggest the specific inclusion of DLGG in future PLGG trials for subgroup analysis, using response metrics that are suitable for leptomeningeal disease.⁴³ Specifically, we recommend the incorporation of ctDNA positivity into the assessment of residual disease. Our data also suggest that, when feasible and safe, there may be value in aiming for resection of a dominant mass, even in the presence of metastatic lesions.

From a scientific perspective, this study provides insights into the cellular origins of PLGG and the mechanisms contributing to cancer metastasis. PLGG can occur throughout the CNS, with some correlations between location and histology/genetics, but nonetheless with significant overlap between tumors in different regions.⁶ The striking enrichment of suprasellar tumors in our cohort compared to overall PLGG implies a predilection for tumors in this location to disseminate. Prior methylation-based analysis^44–47 and animal modeling studies^48,49 have implicated different cells of origin for hemispheric, suprasellar/third ventricular, and posterior fossa PLGG. The fact that suprasellar DLGG presented at a younger age than both DLGG in other locations and non-disseminated suprasellar PLGG further suggests a different temporal window at which this originating cell is vulnerable. Therefore, further studies to pinpoint the specific cellular origins of PLGG in regions and developmental stages will be essential to understand lineage-specific dissemination risk and therapeutic vulnerabilities.⁵⁰

More broadly, PLGG is a valuable model system to study cancer metastasis, as it is (1) a genetically quiescent disease and (2) is able to widely disseminate while remaining low-grade indefinitely. While other CNS tumors often present as high-grade and metastatic⁵¹ or become increasingly malignant and genetically unstable over time,^52,53 the unusually stable nature of most PLGG allows for the disentangling of cancer grade and genetics with tumor dissemination. Our analysis indicates that profoundly different biological mechanisms contribute to malignant transformation versus dissemination, with the former more explicable by DNA-level alterations. This is in keeping with the broader body of research in metastasis in CNS and systemic cancers that has failed to identify recurrent genetic mechanisms contributing to metastasis across cancer types.^54–56

To summarize, this study expands our understanding of the clinical, pathologic, and molecular features of this challenging disease as well as supporting the use of non-invasive CSF analysis and upfront targeted therapy for treatment. As genetics does not distinguish between PLGG that remain localized versus disseminate, nor explain the mechanisms underlying dissemination, future work will be oriented toward non-genetic factors, including immune microenvironment, tumor metabolism, and epigenetics to explain and predict this behavior. Further, most DLGG patients do eventually progress on targeted therapies, therefore, it will also be important to further investigate the mechanisms of therapy resistance and identify novel options for combination therapies.

Limitations of this study

The rarity of this disease required us to recruit patients from many sites and over a broad time period, resulting in heterogeneity with respect to clinical management, data collection, and molecular testing. The retrospective analysis of treatment response to chemotherapy versus targeted therapy requires confirmation in a prospective study with uniform treatment protocols, radiographic assessment (ideally volumetric analysis), and criteria for progression and response. The single-cell and ctDNA analysis have small sample numbers and should be regarded as hypothesis generating and requiring further exploration in future studies. As many patients have limited tissue sampling, there was a lack of patients with sufficient amount and quality of material for this testing—a greater emphasis on prospective tissue banking is recommended to enable future scientific studies.

Supplementary Material

Supplementary material is available online at Neuro-Oncology (https://academic.oup.com/neuro-oncology).

Author Contributions

Conceptualization: A.B.L., J.B., P.Pati., M.C.D., U.T., C.E.H. Data Curation: A.B.L., C.L., A.S., A.T.L. Formal Analysis: A.B.L. Funding acquisition: A.B.L., M.C.D., C.E.H., U.T. Investigation: A.B.L., I.B., R.S., M.R., R.Y., Y.N., P.PPate., I.H., M.K. Project Administration: C.L., J.H.C. Resources: P.Pati., J.E.V.V., M.J.S., C.E., C.C., M.S., E.J.N., S.Ch, C.D., B.W., K.B., F.v.L., D.D.E., A.S.G.S., A.W., V.L., P.G., A.F., L.W., I.L.F., A.S.P.F., A.F., S.Ca., N.M., S.I.B., K.M., R.T.G., S.R., F.J.C., L.L., M.V.M., J.M.M.L., S.D., K.F., K.Y., T.S., F.O., S.H., Y.A., T.I., F.Y., J.R.H., A.Lu., M.P.N., H.T., R.M.M., A.La., L.B.S., J.P.K., A.T., C.D., J.J., A.T.H., P.B.D., J.T.R. Supervision: C.E.H., U.T., M.C.D. Visualization: A.B.L., A.T.L. Writing—Original Draft: A.B.L., J.B., P.Pati., M.C.D., C.E.H., U.T. All authors were involved in drafting of the manuscript and final approval.

Conflict of Interest Statement

The authors report the following relevant disclosures:

C.H., honorarium from Servier Canada.

Y.A., honoraria from Nippon Kayaku, Novocure, UCB Japan, Ono Pharmaceutical, Brainlab, Merck, Chugai, Eisai, Daiichi Sankyo, Carl Zeiss, Nihon Medi-Physics, Stryker; grants to institution from Philips, Otsuka, Chugai, Nihon Medi-Physics, Daiichi Sankyo, Stryker, Esai, Japan Blood Products Organization, Ono Pharmaceutical, Taiho Pharma, Sumitomo Dainippon Pharma, Astellas Pharma, Incyte Biosciences, Servier.

J.B., honoraria from Servia Canada, Alexion Canada.

I.B., funding support for data collection and analysis from Alexion Pharmaceuticals, Day One Pharmaceuticals.

S.C., honorarium from Alexion PharmaceuticalsCE: consulting fees from Alexion Pharmaceuticals; payment for expert testimony from the Canadian Drug Agency; DSMB member for Collaborative Network for Neuro-oncology Clinical Trials.

A.F., honorarium from DayOne Biopharmaceuticals.

A.G.S., consulting fees from Novartis and Alexion; payment to institution from Novartis.

J.H., honoraria from Alexion Pharmaceuticals; ­consulting fees from Servier International and ECOR1; participation on scientific advisory board of Servier International; director and lead CNS group for ANZCHOG.

T.I., research funding from Nihon Servier Co, Ohara Pharmaceutical Co.

J.P.K., UK Tovorafenib advisory board memberALasalletta: consulting fees from Servier, Day101, Ipsen; support for attending meetings from Lilly, Servier.

A.T., consulting fees from MedisonFvL: honorarium from Servier Canada.

Funding

Rally for the Cure Foundation; Pediatric Brain Tumor Foundation; Brain Tumor Foundation of Canada (to A.B.L.); SickKids Clinician Scientist Training Program (to A.B.L.); Canadian Institutes of Health Research (399868 to C.H., 480606 to C.H.); Terry Fox Research Institute Program Projects Grant at the Hospital for Sick Children (TFRI Project # 1156-03 to A.B.L., U.T., C.H., P.B.D., J.T.R., J.B.); SickKids Foundation donors David Nowak and Allyson Whyte-Nowak.

Ethics Statements

This study was approved by the Research Ethics Board (# 1000079566) at the Hospital for Sick Children and complied with all relevant ethical regulations and Declaration of Helsinki principles. All study investigators at other sites complied with their specific institutions’ ethical requirements.

Data Availability

Newly generated methylation data for DLGG are available at GEO under GSE285849. Newly generated single-nucleus RNA seq data are available at GEO under GSE288107. Newly generated panel sequencing data for DLGG are available at EGA under EGAD50000001230. Newly generated ctDNA low-pass WGS and panel sequencing data for PLGG will be made available on EGA as part of a larger study that includes other types of pediatric brain tumors. The data are available under controlled access to comply with data protection regulations and can be accessed by application to the data access committee via C.H. (cynthia.hawkins@sickkids.ca). Previously published RNA and targeted DNA sequencing data for PLGG6 are available on EGA (EGAD00001005987), and can be accessed by application to the data access committee via C.H. (cynthia.hawkins@sickkids.ca). Additional clinical data and molecular characterization (using IHC, FISH, Nano-String gene fusion panels, SNP array) are available as source data at the manuscript website https://doi.org/10.1016/j.ccell.2020.03.011.