Molecular, histologic, and clinical characterization of methylation class pleomorphic xanthoastrocytoma: An analysis of 469 tumors

Abstract

Background

Methylation class pleomorphic xanthoastrocytoma (mcPXA) comprises tumors with the DNA methylation signature of classical PXA but with a wider histologic spectrum, including overlap with glioblastoma (GBM).

Methods

To clarify the histologic and molecular scope of mcPXA and characterize its clinical behavior, a cohort of 469 tumor samples from 458 patients matching to mcPXA by the DKFZ classifier (v12.6 score ≥0.85) was interrogated.

Results

Patient median age was 23 years (range 1-73 years) with a female predominance (259 female/199 male). CDKN2A/B homozygous deletion was observed in 406 of 469 (87%) samples. In samples tested for BRAF p.V600E mutations (n = 279), 240 (86%) harbored the mutation. A chr7+/chr10− pattern was observed in 103 of 469 (22%) samples. Among samples tested for TERT promoter mutations (n = 143), 32 (22%) harbored the mutation. Progression-free and overall survival of patients with mcPXA were comparable to patients with methylation class IDH-mutant astrocytoma, low grade, but a GBM-like subset (ie, cases with a pre-methylation working diagnosis of GBM) showed shorter survival. Histologic features of high grade, including palisading necrosis and microvascular proliferation, were prognostic in mcPXA. Compared to patients with BRAF p.V600E-altered GBM, patients with mcPXA were younger and had a lower frequency of TERT promoter mutations.

Conclusion

Tumors in mcPXA share molecular characteristics with histologically defined PXA, and high-grade histologic features can help predict their clinical behavior. The use of an epigenetic classification of PXA reveals that this group of tumors is more common than previously appreciated and warrants in-depth study to identify efficacious therapeutic options.

Graphical Abstract

Graphical Abstract.

Key points.

• mcPXA tumors have stereotypic molecular profiles across a spectrum of histologic features.

• GBM-like histology is a prognostic factor in mcPXA.

• mcPXA tumors can be clinically and molecularly distinguished from BRAF-mutant GBM.

Importance of the Study.

This study represents the most comprehensive evaluation of mcPXA tumors to date and provides an integrated perspective on the epidemiologic, histologic, and molecular spectrum of mcPXA. The results provide evidence of histologic features of prognostic significance within mcPXA and may be used to inform future grading criteria for tumors of this methylation class. The results also underscore the importance of establishing a reliable reference set for DNA methylation-based tumor classification.

Methylation class pleomorphic xanthoastrocytoma (mcPXA) is an epigenetically distinct group of central nervous system (CNS) tumors with a characteristic DNA methylation pattern^1,2 based on the methylation profile of tumors with the histologic features originally described for PXA by Kepes et al.³ Histologically defined PXAs are usually circumscribed and superficially located cerebral gliomas composed of epithelioid and spindle-shaped astrocytic cells with frequent multinucleation, xanthomatous changes, and intranuclear vacuoles/pseudoinclusions, often intermixed with eosinophilic granular bodies and lymphocytes.^3,4 Although first observed in histologically defined PXA, the PXA methylation signature has since been observed in tumors with diverse histologic patterns not typical of classical PXA. Of particular interest are tumors that meet histologic criteria for a diagnosis of epithelioid glioblastoma (GBM), IDH-wildtype but harbor a methylation profile corresponding to mcPXA.⁵ Prior studies of PXA have been predominantly focused on histologically defined tumors.^6–8 Given that many tumors in mcPXA may not show classical histologic features of PXA, further study of this epigenetically characterized group of tumors is warranted.

Molecular profiling of histologically defined PXAs has identified the presence of mitogen-activated protein kinase (MAPK) pathway gene alterations (usually BRAF p.V600E) and homozygous deletion of CDKN2A and/or CDKN2B in nearly all cases.^7–10 These 2 molecular features have also been observed in nearly all cases of mcPXA.¹¹ The additional presence of TERT promoter alteration or +7/−10 copy number change has been identified in a subset of mcPXA, and TERT promoter alterations were recently found to have prognostic significance independent of histologic grading (ie, mitotic rate quantification).¹¹ The biologic significance of GBM-like molecular findings in tumors with the mcPXA epigenetic signature requires clarification before such tumors receive definitive classification.

To better characterize the histologic, molecular, and clinical spectrum of mcPXA, tumors evaluated with DNA methylation profiling at 4 methylation referral centers (US National Cancer Institute: NCI; German Cancer Research Center: DKFZ; University College London: UCL; NYU Langone Health: NYU) as well as tumors with previously published methylation profiles were considered for inclusion in a mcPXA cohort. A total of 469 tumors with a calibrated confidence score of at least 0.85 by version 12.6 (v12.6) of the DKFZ CNS tumor DNA methylation classifier were selected for inclusion in the study. These tumors came from 458 unique patients. Histologic features, genetic alterations, and clinical parameters were interrogated for all samples with available data, and copy number changes and select, locus-specific amplifications and deletions were inferred from DNA methylation signals. Histologic and molecular features were tested for possible associations with progression-free survival (PFS) and overall survival (OS).

This multi-institutional study represents a comprehensive characterization of mcPXA tumors and highlights key features of concordance between histologically defined and DNA methylation-based classifications of PXA. The results provide evidence of prognostically significant histologic features within mcPXA and help clarify the relationship between mcPXA and GBM.

Materials and Methods

Sample Selection

To establish the mcPXA study cohort, methylation classifier results of tumor samples evaluated by DNA methylation profiling on the Illumina 450K or EPICv1 beadchips at NCI, DKFZ, UCL,¹² and NYU through February 2024, as well as previously published tumor samples, were queried.^1,13–26 Samples with a classification result of PXA and a calibrated confidence score ≥0.85 by v12.6 of the DKFZ CNS tumor classifier were selected for inclusion unless they were classified as high-grade glioma with pleomorphic and pseudopapillary features (HPAP)²⁷ with a mean confidence score ≥0.85 by the NCI-Bethesda classifier (https://methylscape.ccr.cancer.gov/). Raw DNA methylation array data in IDAT format (Illumina 450K or EPICv1 beadchips) from previously published samples were acquired from the following sources: PMID 28966033 (E-MTAB-5528), PMID 29539639 (GSE109381), PMID 29763623 (E-MTAB-5552), PMID 30876455 (GSE125450), PMID 31346129 (E-MTAB-7490), PMID 31554817 (GSE135017), PMID 32238360 (E-MTAB-7802; E-MTAB-7804), PMID 32991787 (GSE143843), PMID 34538273 (GSE152561), PMID 34545083 (GSE131482), PMID 35191072 (GSE196490), PMID 36734226 (GSE224218), PMID 36928815 (GSE215240), PMID 36973520 (GSE218542), E-MTAB-9297.

To establish an astrocytic glioma comparison cohort for survival analyses, methylation classifier results of tumor samples evaluated by DNA methylation profiling as part of The Cancer Genome Atlas (TCGA) GBM²⁸ and lower grade glioma²⁹ cohorts were queried. Samples with a classification result of A_IDH_LG, GBM_MES_ATYP, GBM_MES_TYP, GBM_RTK1, or GBM_RTK2 and a calibrated confidence score ≥0.85 by v12.6 of the DKFZ CNS tumor classifier were selected for inclusion.

To establish a BRAF-altered GBM comparison cohort for molecular and histologic feature analyses, methylation classifier results of tumor samples evaluated by DNA methylation profiling on Illumina EPIC beadchips at NCI through February 2024 were queried. Samples with a classification result of GBM_MES_ATYP, GBM_MES_TYP, GBM_RTK1, or GBM_RTK2 and a calibrated confidence score ≥0.85 by v12.6 of the DKFZ CNS tumor classifier were then investigated for the presence of BRAF p.V600E mutations. All tumors with a BRAF p.V600E mutation were selected for inclusion in the BRAF-altered GBM cohort.

The research received ethical clearance from the NCI Central Institutional Review Board, as well as from the UK Brain Archive Information Network and the Institutional Review Board of the Medical Faculty at Heinrich Heine University Duesseldorf, allowing the use of biospecimens with associated anonymized clinical data.

Histologic Evaluation

To evaluate histologic features, a single digital whole slide image (WSI) of an H&E-stained section from the block selected for DNA methylation profiling was examined. In many cases, the single WSI did not represent the entire case originally submitted for histologic evaluation. Surgical pathology reports were also consulted when available. Histologic features were evaluated by WSI for 197 mcPXA tumors and 19 BRAF-altered GBM tumors. Digital slides were generated at NCI and UCL using a NanoZoomer s60, s210, or s360 Digital Slide Scanner. Slides were anonymized using a Python script (https://github.com/bgilbert/anonymize-slide) and uploaded to HALO Link (Indica Labs) for viewing and annotation.

Digital slides were scored for the presence of the following 11 histologic features selected for their prevalence in histologically defined PXA and high-grade gliomas (HGGs) that may have a similar appearance to PXA: (1) epithelioid cells, (2) spindle-shaped cells, (3) multinucleated cells, (4) xanthomatous cells, (5) nuclear pleomorphism, (6) intranuclear vacuoles/pseudoinclusions, (7) eosinophilic granular bodies, (8) lymphocytic infiltrates, (9) microvascular proliferation, (10) necrosis (ie, geographic), and (11) palisading necrosis. As noted, surgical pathology reports were consulted when available. These reports were given priority when discrepant with the observed histology. Reported findings from the entire case were usually observed in the block selected for nucleic acid extraction; discrepancies were recorded for only 15 cases (8%). Histologic infiltration was not formally scored due to uncertainty regarding an adequate method of scoring this feature, which may be seen at the periphery of histologically defined PXA without carrying the same clinical significance as seen in diffuse gliomas.

Grading criteria for mcPXA have not been codified. To test the applicability of histologic grading by mitotic activity to tumors matching to mcPXA, tumors were assigned a low or high mitotic activity level according to the CNS WHO 5th Edition (2021) grading criteria for histologically defined PXA.³⁰ A mitotic count of <5 mitoses per 10 high power fields or <2.5 mitoses per mm² denoted a low mitotic activity level. A higher mitotic count denoted a high mitotic activity level. Again, only a single WSI was reviewed for mitotic count.

Primary scoring and mitotic activity level assignment by digital WSI analysis were performed by 2 observers (C.H.D. and N.S.), and discrepancies were resolved by discussion. Guidance and oversight of scoring were provided by 2 experts (K.D. and M.Q.). Primary scoring and mitotic activity level assessment of an additional 31 tumor samples for which digital WSI were not available was provided by report of the contributing neuropathologist.

Immunophenotypic Evaluation

To determine immunophenotypic features in mcPXA, immunostain results for ATRX, BRAF (p.V600E), CD34, GFAP, and H3K27me3 were extracted from submitted surgical pathology reports when available. No additional immunostaining or interpretation was performed for this study.

DNA Methylation Profiling

To measure DNA methylation signals in tumor samples, H&E-stained sections were evaluated to identify regions of higher tumor content, and these regions were marked for collection. Tumor-enriched formalin-fixed, paraffin-embedded (FFPE) material was collected from 5 to 10 unstained slides of 5-10 µm sections, and genomic DNA was extracted using the AllPrep DNA/RNA FFPE Kit (Qiagen). Extracted DNA (target amount > 250 ng) was bisulfite-converted (EZ DNA Methylation Kit, Zymo Research D5001) and processed with the Infinium FFPE DNA Restore Kit (Illumina). When standard quality controls confirmed adequate DNA quality and bisulfite conversion, DNA methylation was assayed on the Infinium Methylation450K or MethylationEPICv1 beadchips (Illumina), according to the Infinium HD FFPE Methylation Assay automated protocol (Illumina).

DNA Methylation-Based Tumor Classification

To classify samples based on DNA methylation signals, all samples were independently classified at NCI using v12.6 of the DKFZ CNS tumor methylation classifier, which employs a random forest algorithm to classify query samples based on the methylation status of 10 000 select probes shown to have the best ability to discriminate among CNS tumor entities.¹ All samples were also classified using version 2 (v2) of the NCI-Bethesda classifier because of known similarity in epigenetic profile between mcPXA and HPAP and the absence of the HPAP class in v12.6 of the DKFZ CNS tumor classifier. Normalization, probe filtering, and adjustment for the possible effect of FFPE or frozen tissue were performed as implemented in the mnp.v12b6 R package.

Copy Number Analysis

To algorithmically assess for somatic copy number variation across the genome of each sample, the conumee R package³¹ was used to calculate probe intensity ratios with baseline correction. Aneuploidy for select regions was assessed by comparing the combined raw intensity values of methylated and unmethylated probes on either strand in each region from each tumor sample to the combined intensities of the same probes in a reference cohort of diploid genomes. A set of about 100 000 probes with potentially misleading information³² was excluded from analysis. The reference cohort was constructed by selecting TCGA samples with diploid genomes as assessed by multiple genomics assays. Separate reference cohorts were constructed for males and females for the 450K and EPICv1 methylation arrays from a set of 166 diploid genomes.

Homozygous loss of the CDKN2A/B locus was called using a query to reference log2 ratio threshold of −0.415, as previously published.³³ Amplifications of the EGFR, MDM2, and PDGFRA loci were called using a query to reference log2 ratio threshold of 0.6, as previously published.³⁴ Gains and losses of broad structural variants (ie, whole chromosome arms) were called using GISTIC³⁵ based on conumee segmentation files with query to reference log2 ratio thresholds of 0.1 and −0.1, respectively (ie, GISTIC defaults). A genome-wide relative aneuploidy score was calculated for each sample by taking the sum of 2 parameters. First, samples were sorted in ascending order by the standard deviation (SD) of all probes (ie, 1 = lowest SD). Next, samples were sorted in descending order by the conumee noise estimate (ie, 1 = highest noise). The highest-ranked samples by both parameters are those with the most variance but least noise across probes. The sum of the relative ranks of the SD and noise parameters was used as the relative aneuploidy score.

MGMT Promoter Methylation Analysis

To predict the methylation status of the MGMT gene promoter, the MGMT-STP27 model was used with DNA methylation array measurements as model inputs.³⁶

Gene Variant Analysis

To determine the mutation status of ATRX, BRAF, MAP2K1, NF1, PTEN, TERT promoter, and TP53, and to identify MAPK pathway-related fusions, previously reported results were extracted from the literature or medical record, when available, or targeted next-generation sequencing (NGS) was performed at NCI. For samples sequenced at NCI, tumor-enriched FFPE material was collected from 5 to 10 unstained slides of 5-10 µm sections, and genomic DNA and RNA were extracted using the AllPrep DNA/RNA FFPE Kit (Qiagen). NGS was performed using either a custom amplicon-based brain tumor-specific panel (PBTP)³⁷ or the commercial TruSight Oncology 500 panel (TSO500; Illumina).

For the PBTP assay, libraries were prepared using AmpliSeq technology and sequenced using the Ion S5™ XL Sequencing System (Thermo Fisher Scientific). Signal processing, base calling, and alignment to the GRCh37/hg19 human genome assembly were performed using Torrent Suite™ software packages (Thermo Fisher Scientific). Variant annotation and interpretation were performed with Ion Reporter Software v.5.10 (Thermo Fisher Scientific).

For the TSO500 assay, libraries were prepared using the TSO500 Kit according to manufacturer instructions. Amplified pre-enriched libraries were hybridized to probes specific to the 523 genes targeted by the TSO500 panel. Enriched libraries were amplified, quantified, and normalized to 2 nM, then sequenced as paired-end reads on a high-output NextSeq 500/550 flow cell. The TSO500 local application was used for alignment and variant calling. Final interpretation of variants was based on integration of data from multiple bioinformatics databases and experimental and clinical data reported in biomedical literature.

Survival Analysis

To characterize the clinical behavior of tumors in this study, PFS and OS of patients were extracted from the medical record or published reports, as available. Prognostically relevant clinical factors (eg, extent of resection, presence and type of adjuvant therapy) were also extracted, but data availability and quality were inconsistent, so these factors were not included in formal testing. Prior to testing for associations between survival outcomes and patient and tumor characteristics, samples were filtered to ensure all patients included in survival analyses were independent. When multiple samples from a single patient were present, the tumor sample from the earliest resection was selected. Single variable survival curves for categorical variables were estimated using the Kaplan-Meier method with differences in survival tested with the log-rank test. Differences in survival for single categorical variables were also tested with Cox proportional hazards models using a likelihood ratio test. Cox proportional hazards multiple regression models with likelihood ratio tests were used to evaluate the effects of continuous and multiple predictor variables. A P-value <.05 was considered statistically significant in both log-rank and likelihood ratio tests.

Feature Comparisons Between Groups

To test for enrichment of categorical variables across groups, contingency tables were built, and Pearson’s chi-squared tests with Yates’ continuity correction as implemented in base R were employed. To test for differences in continuous variables across groups, the Wilcoxon rank sum test as implemented in base R was employed. Given the multiplicity of hypotheses being tested in enrichment analyses, the Benjamini Hochberg method was used to control the false discovery rate (FDR). An FDR < 0.05 was considered statistically significant in chi-squared tests.

Results

mcPXA Tumors Are Most Frequent in the Temporal Lobe of Adolescents and Young Adults and Are Often Histologically Considered as HGG or GBM Prior to DNA Methylation-Based Classification

The study cohort included 469 tumor samples classified as mcPXA by the DKFZ CNS tumor classifier, v12.6. These samples came from 458 unique patients. Patient median age was 23 years (range 1-73 years, interquartile range 13-36 years), and females (F) were more frequently affected than males (M) (n = 259 F, n = 199 M, F:M ratio 1.3) (Supplementary Figure 1). Primary versus recurrent status of tumor was available for 172 samples (37%), and of these, most tissue samples (n = 131, 76%) were from primary tumors. Information on tumor location was available for 265 samples (57%), and of these, the most frequent was temporal lobe (n = 108, 41%), followed by frontal lobe (n = 42, 16%) and parietal lobe (n = 41, 15%) (Supplementary Figure 1). A pre-methylation working diagnosis was available for 440 samples (94%), and the most frequent was PXA (n = 132, 30%), followed by HGG (n = 81, 18%) and GBM (n = 80, 18%) (Figure 1A). There were 8 samples submitted with the differential diagnosis of PXA versus GBM prior to methylation classification.

Figure 1.

Pre-methylation diagnosis and molecular features of the mcPXA cohort. (A) Distribution of common pre-methylation working diagnoses ordered by frequency, with GBM-like working diagnoses grouped together. PXA = pleomorphic xanthoastrocytoma; GBM = glioblastoma; HGG = high-grade glioma; NOS = not otherwise specified; GG = ganglioglioma; EPN = ependymoma; MNG = meningioma. (B) Summary of molecular features including copy number alterations and point mutations. Samples are ordered by patient age (when available) and grouped by GBM-like pre-methylation diagnosis. Pre-Me Dx = pre-DNA methylation diagnosis; CDKN2A/B DEL = CDKN2A/B homozygous deletion; MGMT METH = MGMT promoter methylation status; AMP = amplification; TERT MUT = TERT promoter mutation status.

mcPXA Tumors Harbor Key Genomic Changes of Classical PXA and a Subset Show Genomic Changes That Overlap with GBM

Of the 469 tumor samples in the study cohort, BRAF p.V600E status was known for 279 (59%), and most (n = 239 of 279, 86%) harbored the mutation (Figure 1B). Of the 40 samples tested in which this mutation was not detected, alternative MAPK pathway-activating alterations were identified in 12 samples, and one other had a variant of uncertain significance in BRAF (p.G503_V504delinsVIHKS). Of these 12 samples, 3 had pathogenic variants in NF1, one of which also had a non-canonical pathogenic BRAF variant (p.D594N). Two samples had pathogenic variants in MAP2K1. Two samples had ATG7::RAF1 fusions, and two had NTRK2 rearrangements (one NACC2::NTRK2 fusion, one detected by fluorescence in situ hybridization without a known fusion partner). One sample had an AGK::BRAF fusion, one sample had a CUL1::BRAF fusion, and one sample had a PAPD7::RAF1 fusion. In the subset of samples without available BRAF p.V600E testing (n = 190, 41%), rearrangements involving BRAF (GTF2I::BRAF) or NTRK2 (KCTD8::NTRK2, BEND5::NTRK2, SPECC1L::NTRK2) were identified in six, and 2 additional samples were associated with prior resections with NTRK2 or BRAF fusions. One sample occurred in a patient with an NF1 germline mutation. Overall, results suggested a high frequency of BRAF p.V600E alterations, with additional MAPK alterations in the remaining tumors, consistent with prior studies.^7,9

Samples were examined for the presence of CDKN2A/B homozygous loss based on DNA methylation-based copy number profiles, and most samples (n = 406 of 469, 87%) demonstrated evidence of homozygous loss (Figure 1B). Of the 63 samples without quantitative evidence of homozygous deletion of CDKN2A/B, 51 (81%) harbored changes in copy number profile suggestive of at least hemizygous deletion, with approximate log2 ratio values between −0.135 and −0.414 and partial loss of CDKN2A/B on qualitative visual inspection of copy number plots (data not shown). Of the 12 remaining cases without any degree of loss of CDKN2A/B, 4 had possible gains of CDK4 and/or CDK6 on qualitative visual inspection of copy number plots (data not shown).

About one-fifth of samples (n = 103 of 469, 22%) demonstrated co-occurrence of whole chromosome 7 gain with whole chromosome 10 loss (+7/−10). MGMT promoter methylation was uncommon (n = 72 of 469, 15%), and amplifications of EGFR, MDM2, and PDGFRA were very rare (EGFR n = 1, MDM2 n = 2, PDGFRA n = 1, all < 1%). Of the 143 samples (30% of full cohort) with known TERT promoter status, about one-fifth (n = 32 of 143, 22%) had TERT promoter mutations, most with concomitant BRAF p.V600E mutation (26 of 28 when the status of both was known). The 2 samples with TERT promoter mutation without BRAF p.V600E mutation harbored MAPK pathway-activating fusions (ATG7::RAF1, CUL1::BRAF). Eight TERT promoter-altered tumors (25%) had a concomitant +7/−10 chromosomal copy number pattern, and none of the TERT promoter-mutated mcPXA tumors had amplifications of EGFR, MDM2, or PDGFRA. Alterations in select genes with more frequently available data are summarized in Supplementary Table 1.

mcPXA Tumors Demonstrate Histologic Features of Classical PXA but Also of Other Glioma Types, Including GBM

The presence of specific histologic features was evaluated in 228 tumor samples (49% of full cohort), and one additional tumor sample with reported mitotic count was included in comparisons of mitotic activity level. Of the 228 samples included in histologic feature counts, 197 (86%) were available for central review by WSI, and 31 were available by report without WSI (Supplementary Table 2). Not all features were reported or could be assessed in all samples; percentages were calculated based on total samples with available data for a given feature and not on total samples overall. Among the 228 samples with histologic evaluations, the most common pre-methylation working diagnosis was HGG (n = 60, 26%), followed by PXA (n = 52, 23%), glioma NOS (n = 33, 14%), and GBM (n = 29, 13%).

Epithelioid histology, at least focally evident, was observed in nearly all tumors (n = 183 of 194, 94%), and a spindled component was observed in over half (n = 120 of 195, 62%). Nuclear pleomorphism was very common (n = 176 of 201, 88%), and xanthomatous changes, multinucleation, and intranuclear vacuoles/pseudoinclusions were each seen in almost three-fourths of samples (n = 146 of 200, 73%; n = 144 of 198, 73%; n = 143 of 197, 73%, respectively). A prominent, frequently perivascular, lymphocytic infiltrate was seen in over half of the samples (n = 111 of 201, 55%), and eosinophilic granular bodies were seen in nearly half (n = 94 of 197, 48%) (Figure 2A). Geographic necrosis was present in almost half of the samples (n = 102 of 227, 45%), whereas palisading necrosis was seen in one-fifth of samples (n = 45 of 226, 20%) (Figure 2B). Microvascular proliferation was identified in about two-fifths of samples (n = 98 of 227, 43%) (Figure 2C). Of samples for which mitotic counts were determined (n = 229, 49% of full cohort), there were nearly equal numbers of low- and high-mitotic activity tumors (n = 113 low, n = 116 high) using a threshold of 5 mitoses/10 high power fields (HPF) or 2.5 mitoses/mm², the threshold recommended to distinguish CNS WHO grade 2 from grade 3 for histologically defined PXA. A summary of histologic features in the 197 samples evaluated by WSI-based central review is included in Figure 2D, and a summary of features for all 228 samples with histologic feature counts is included in Supplementary Table 2.

Figure 2.

Histologic features of tumors in the mcPXA cohort. (A) Classical features including epithelioid and spindle-shaped cells intermixed with lymphocytes and frequent eosinophilic granular bodies. Scale bar = 100 micrometers. (B, C) GBM-like features including palisading necrosis (B) and microvascular proliferation (C). Scale bars = 100 micrometers. (D) Summary of histologic features with samples ordered by patient age (when available) and grouped by GBM-like pre-methylation diagnosis. MVP = microvascular proliferation, EGB = eosinophilic granular bodies.

A limited analysis of select immunophenotypic features available in submitted surgical pathology reports showed GFAP immunoreactivity in 129 of 137 (94%) tumors. Loss of nuclear ATRX staining was observed in 15 of 112 (13%) tumors. Mutant BRAF (p.V600E) protein was positive in 85 of 103 (83%) tumors. CD34 staining was variable (positive in 35 of 75 tumors, 47%). Loss of nuclear H3K27me3 staining was observed once in 25 cases (4%). A summary of immunophenotypic features is included in Supplementary Table 3.

Palisading Necrosis, Microvascular Proliferation, and Pre-Methylation Diagnosis of GBM Are Associated with Shorter Survival in mcPXA

PFS data were available for 124 patients (27%), and OS data were available for 173 patients (38%). The age distribution of patients with available clinical outcomes data was not statistically different from the age distribution observed in the full cohort. Survival data were available for male patients more frequently than for female patients. Select demographic, molecular, and histologic features were examined for their relationship to patient outcomes. Neither patient age (n = 121 for PFS, n = 166 for OS) nor sex (n = 59 F, n = 65 M for PFS; n = 82 F, n = 91 M for OS) were significantly associated with PFS or OS (Supplementary Tables 4 and 5). The +7/−10 copy number signature (n = 95 absent, n = 29 present for PFS; n = 135 absent, n = 38 present for OS), a molecular feature common to GBM, was not associated with PFS or OS. TERT promoter mutations (n = 45 absent, n = 11 present for PFS; n = 70 absent, n = 17 present for OS), another feature common to GBM, were not associated with PFS or OS in this cohort. Presence of BRAF p.V600E mutation (n = 8 absent, n = 88 present for PFS; n = 13 absent, n = 120 present for OS) was not associated with PFS or OS. Relative aneuploidy (n = 124 for PFS, n = 173 for OS) showed a clinically insignificant association with PFS and was not associated with OS.

For patients with available data on the presence of high-grade histologic features and outcome, palisading necrosis (n = 63 absent, n = 19 present for PFS; n = 81 absent, n = 21 present for OS) was associated with shorter PFS (P = .04) and shorter OS (P = .03) (Figure 3 and Supplementary Tables 4 and 5). Microvascular proliferation (n = 50 absent, n = 33 present for PFS; n = 55 absent, n = 48 present for OS) was associated with shorter OS (P = .04) but was not associated with PFS. A composite histologic assessment based on the presence of a pre-methylation working diagnosis of GBM (ie, GBM-like; n = 105 absent, n = 19 present for PFS; n = 132 absent, n = 41 present for OS) was also associated with shorter OS (P = .02) but was not associated with PFS. Tumor mitotic activity level (n = 36 low, n = 48 high for PFS; n = 49 low, n = 56 high for OS) was not associated with PFS or OS.

Figure 3.

Histologic correlates of patient survival in the mcPXA cohort. (A) Kaplan-Meier (KM) plot demonstrating OS stratified by presence of palisading necrosis. (B) KM plot demonstrating OS stratified by presence of microvascular proliferation. (C) KM plot demonstrating PFS stratified by presence of palisading necrosis. (D) KM plot demonstrating OS stratified by methylation class with mcPXA substratified by GBM-like pre-methylation diagnosis. (E) KM plot demonstrating OS stratified by mitotic activity level. (F) KM plot demonstrating PFS stratified by mitotic activity level. A-IDH = IDH-mutant astrocytoma, low grade; GBM = glioblastoma classes including mesenchymal and RTK1/2, MC = methylation class.

To provide clinical context for survival characteristics of patients with mcPXA tumors, OS curves for patients with tumors matching to mcPXA, methylation class IDH-mutant astrocytoma, low grade (A-IDH), and methylation class mesenchymal or RTK1/2 GBM were assessed (Figure 3D). Kaplan-Meier analysis demonstrated that OS for most patients with mcPXA tumors was approximately as long as that of patients with A-IDH and substantially longer than that of those with GBM. However, patients with mcPXA tumors with a pre-methylation diagnosis of GBM (ie, GBM-like) demonstrated OS intermediate between patients with non-GBM-like tumors of mcPXA and patients with GBM. Cox proportional hazards models simultaneously testing the effects of age and methylation class (with mcPXA limited to GBM-like samples) showed age but not methylation class was associated with OS (Supplementary Table 6).

mcPXA Tumors Considered PXA Prior to DNA Methylation-Based Classification Are Similar to Other Tumors in mcPXA

Given the variety of pre-methylation diagnoses of tumors in mcPXA, histologically classical PXAs within mcPXA (ie, tumors with a pre-methylation working diagnosis of PXA) were compared to the others. Patients with histologically classical PXAs were of approximately the same ages as those with non-classical tumors; this result was consistent whether GBM-like tumors were excluded from or included in the non-classical cohort. Histologically classical PXAs shared approximately equal proportions of +7/−10 copy number change, TERT promoter mutation, CDKN2A/B homozygous deletion, MGMT promoter methylation, and BRAF p.V600E mutation with non-classical tumors; this result was consistent whether GBM-like tumors were excluded from the non-classical cohort or not (Supplementary Tables 7 and 8). Most histologic features were observed in approximately equal proportions in histologically classical PXAs and other tumors of mcPXA, but multinucleation and lymphoid infiltrates were relatively over-represented in the histologically classical PXA subset (FDR = 0.04) when GBM-like tumors were excluded from the non-classical subset. This enrichment was present but not statistically significant when GBM-like tumors were included in the non-classical cohort. When GBM-like tumors were included in the non-classical cohort, this cohort was enriched for necrosis (FDR = 0.04). Regarding patient outcomes, those with classical PXA and those with non-classical PXA had comparable OS (Supplementary Table 9).

Age, TERT Promoter Status, MGMT Promoter Methylation Status, and Necrosis May Help Distinguish BRAF p.V600E Mutant GBM and mcPXA

Given molecular and histologic similarity between mcPXA and BRAF-altered GBM tumors, characteristics of the 2 groups were compared. A total of 19 samples classified as GBM (DKFZ CNS v12.6, ≥0.85) were identified with a BRAF p.V600E mutation. The predominant subtype was “GBM-mesenchymal atypical” (n = 13, 68%), but “GBM-mesenchymal typical” samples constituted almost a third (n = 6, 32%). Within the BRAF-mutant GBM cohort, median age was 34 (range 11-79 years), and males were more frequently affected than females (n = 7 F, n = 12 M, F:M ratio 0.6). Temporal lobe was the most frequent location (n = 5 of 16 with location information), followed by frontal lobe (n = 4). The most frequent pre-methylation diagnosis was HGG (n = 9 of 18 with such diagnoses), followed by neoplasm NOS (n = 4). One sample received a pre-methylation diagnosis of PXA. The +7/−10 copy number signature was present in nearly a third (n = 6, 32%), CDKN2A/B homozygous loss was found in two-thirds (n = 13, 68%), and MGMT promoter methylation was present in nearly half (n = 9, 47%) (Figure 4A). Of the 15 GBM samples in this group with known TERT promoter status, almost all (n = 14, 93%) had TERT promoter mutations. Differences between GBM and mcPXA cohorts were statistically significant for age (GBM older, FDR = 0.01), proportion with TERT promoter mutation (GBM higher, FDR < 0.01), and proportion with MGMT promoter methylation (GBM higher, FDR < 0.01, Supplementary Table 10).

Figure 4.

Molecular and histologic features of GBM with BRAF p.V600E mutation. (A) Summary of molecular features in BRAF-altered GBM cohort, including copy number alterations and point mutations. Samples are ordered by patient age. CDKN2A/B DEL = CDKN2A/B homozygous deletion; MGMT METH = MGMT promoter methylation status; AMP = amplification; TERT MUT = TERT promoter mutation status. (B) Summary of histologic features in BRAF-altered GBM cohort, with samples order by patient age. MVP = microvascular proliferation, EGB = eosinophilic granular bodies.

Epithelioid histology was observed in all BRAF-altered GBM samples (n = 19, 100%), and a spindled component was observed in over half (n = 13, 68%) (Figure 4B). Nuclear pleomorphism was a consistent feature (n = 19, 100%), as was necrosis (n = 19, 100%). Xanthomatous changes, multinucleation, and intranuclear vacuoles/pseudoinclusions were all common (n = 17, 89%; n = 17, 89%; n = 16, 84%, respectively). A prominent lymphocytic infiltrate and eosinophilic granular bodies were each seen in less than a third of samples (n = 6, 32% for both features). Palisading necrosis was seen in over half of the samples (n = 12, 63%). Microvascular proliferation was identified in about three-fourths of samples (n = 14, 74%). Differences between GBM and mcPXA cohorts were statistically significant for proportion with any necrosis and with palisading necrosis (GBM higher, FDR < 0.01 for both features, Supplementary Table 10).

Discussion

The classification of CNS tumors by DNA methylation signature has revealed new relationships between histologic and molecular features of brain tumors. PXA was originally defined by a characteristic histology and growth pattern.³ The PXA methylation class includes tumors with classic PXA features but also includes glial neoplasms that do not fit the original histologic description. These epigenetically defined PXA tumors not only match to the PXA methylation class, but they also show, at high frequencies, the genomic changes typical of PXA, including BRAF p.V600E mutation and CDKN2A/B homozygous loss. The presence of these additional molecular characteristics provides further support for the molecular similarity of mcPXA to histologically defined PXA. Of particular interest is a subset of tumors matching to mcPXA with GBM-like histologic features. The aim of the present study was to provide a comprehensive histologic, molecular, and clinical assessment of mcPXA and to clarify the extent to which epigenetic variation can be relied upon to distinguish this entity from GBM.

Molecular assessment confirmed a high prevalence of BRAF p.V600E mutations and homozygous loss of CDKN2A/B, both well-established features of classical PXA. Molecular findings more typical of GBM, such as +7/−10 and TERT promoter mutations, were uncommon, and amplifications of EGFR, MDM2, and PDGFRA were exceptionally rare. These results confirm prior findings on the molecular landscape of mcPXA^7,11,38 (Supplementary Table 11). In our cohort, we could not identify a significant association between TERT promoter mutation and shorter OS in mcPXA, in contradistinction to a prior study.¹¹TERT promoter mutations in mcPXA are relatively uncommon, and further study to evaluate their prognostic value is warranted.

Histologic assessment, although limited to less than half of the full cohort and to a single slide from each tumor, confirmed the presence of several classic histologic features in mcPXA. Histologic evaluation also demonstrated an unexpectedly high rate of microvascular proliferation and palisading necrosis, neither of which is typical of histologically defined PXA and both of which were associated with shorter OS. Similarly, a pre-methylation diagnosis of GBM, relatively common in mcPXA and frequently coinciding with the presence of microvascular proliferation and/or palisading necrosis, was also associated with shorter OS. The prevalence of high-grade features in mcPXA and histologic overlap with GBM has been reported previously,^2,5,6,11,39 as has the association of pre-methylation diagnosis of GBM with shorter OS¹¹ (Supplementary Table 11). The present study therefore confirms results of prior studies and demonstrates the prognostic value of palisading necrosis and microvascular proliferation in mcPXA. The present study also clarifies an important relationship between tumors matching to GBM methylation classes and those mcPXA tumors with a pre-methylation diagnosis of GBM: although the GBM-like tumors of mcPXA are associated with longer OS relative to GBM, the survival advantage may in part be explained by younger patient age, which is a favorable prognostic factor in GBM.

Grading of tumors matching to mcPXA remains an area of active investigation, and there is an urgent clinical need to clarify grading criteria for tumors of this methylation class. mcPXA appears to include glial neoplasms that may not fit the histologic definition of PXA, so grading criteria established for histologic PXA may not be appropriate for tumors of mcPXA. Given the available evidence, mitotic activity level should continue to determine grading for histologically defined PXA and should be considered when grading tumors of mcPXA, although its prognostic value is uncertain in this context. We and others are actively investigating factors that may affect survival in mcPXA, and grading of these tumors will be a focus of our future work.

This work highlights that the incidence of mcPXA tumors is likely to be higher than histologically defined PXAs. Of the 440 tumors in our cohort with working diagnoses given prior to DNA methylation-based classification, only 132 (30%) were considered PXA, while 300 others were given alternative designations, supporting the hypothesis that the total incidence of mcPXA tumors is likely to be more common than histologically defined PXA. Interrogation of BRAF status is not mandatory for a diagnosis of GBM, so there may be a subset of tumors diagnosed as GBM for which mcPXA is not suspected. Given the overlap of histologic and molecular features between GBM and mcPXA, detection of a BRAF alteration in a tumor otherwise considered GBM may warrant methylation profiling. Despite the variety of working diagnoses given prior to DNA methylation-based classification, the finding that those tumors of mcPXA with pre-methylation working diagnoses of PXA showed molecular and, to a lesser degree, histologic similarity to other tumors of mcPXA suggests that an epigenetic-based classification of PXA identifies a molecularly coherent group of neoplasms that may share important histological features as well. Of note, DNA methylation-based classification for mcPXA was reliable across different classifiers. The NCI/Bethesda (v2) classifier returned high-score matches to mcPXA in 431 (92%) of the 469 tumors in this cohort, which was defined using the DKFZ v12.6 classifier (Supplementary Table 12).

While methylation profiling is the only reliable method for unequivocal recognition of mcPXA tumors, particularly those that do not demonstrate classic histologic features, surrogate approaches for identifying tumors with potential to match to mcPXA may be considered. Immunohistochemical testing to demonstrate expression of BRAF p.V600E mutant protein is helpful. Other immunophenotypic characteristics typical of mcPXA tumors are non-specific but include expression of GFAP and retained nuclear expression of ATRX and H3K27me3. Copy number variation testing to demonstrate CDKN2A/B homozygous loss is helpful, and amplifications of EGFR, PDGFRA, and MDM2 should be absent. DNA sequencing to demonstrate a canonical BRAF alteration as well as CDKN2A/B deletion is helpful. Pathogenic variants in ATRX, PTEN, and TP53 are unusual but not completely absent.

Finally, we present a comparison between mcPXA and a smaller number of GBM with BRAF p.V600E mutations. BRAF-altered GBM tumors affected a slightly older population and had higher proportions of TERT promoter mutation, MGMT promoter methylation, and necrosis compared with mcPXA tumors. The findings of this study underscore the importance of establishing accurate, meticulously curated reference sets for classifier training to reliably discriminate between epigenetically similar entities. The creation of a multi-institutional consensus reference set of methylation patterns for specific CNS tumor entities analogous to Genome Reference Consortium genome builds may be considered a pre-requisite to widespread deployment of machine learning-based DNA methylation classifiers for routine diagnostic purposes.

This study has several limitations. First, the cohort was collected retrospectively and includes tumor samples with a variety of treatment exposures. Second, only a single WSI from each case was reviewed. Although the associated surgical pathology reports were also reviewed when available, information was heterogeneous and limited. This limitation may have affected mitotic counts, which can be spatially variable. Third, a formal analysis of growth patterns seen in mcPXA was not possible due to inconsistent availability of neuroimaging for most cases. Of 36 cases with at least some neuroimaging results available for review, a single case was described as infiltrative, and 13 were described as cystic, but this analysis was not robust enough to warrant inclusion in this study. Analysis of growth patterns in mcPXA is a priority of future research, along with identification of robust grading criteria. Fourth, because DNA methylation-based tumor classification is not yet standard of care, contemporary DNA methylation-based cohorts such as this one may be enriched for higher-grade, more diagnostically challenging cases. DNA methylation-based cohorts will become more representative of the true population as more samples are assayed. Fifth, even with a relatively large sample size, some of the key features under study were of low frequency, in part due to data availability (eg, patient outcomes, TERT promoter mutation status). Incomplete data limited the statistical power to detect relationships between some features and clinical outcomes and negated the possibility of testing for survival differences in patients with BRAF (p.V600E) mutant GBM tumors and GBM-like mcPXA tumors. Moreover, information on treatment-related prognostic factors was too sparse to permit inclusion in the analysis. Acknowledging these limitations, this study further elucidates mcPXA and identifies hypotheses that could be tested in future studies.

In summary, this study highlights, as expected, that mcPXA tumors affect adolescents and young adults, share histologic and molecular features of classical PXA, and occasionally manifest aggressive clinical behavior that can be predicted by histologic evaluation. Our results, together with prior studies, suggest that a diagnosis of PXA should be strongly considered in the setting of a high-confidence methylation match to the mcPXA class, especially in the setting of corroborating genomic alterations (CDKN2A/B homozygous loss and BRAF p.V600E mutation). The use of an epigenetic definition of PXA reveals this entity is more common than previously appreciated and warrants in-depth study for therapeutic options.

Conflict of Interest

The authors declare no conflict of interest.

Funding

This work was supported by the intramural research program of the National Cancer Institute, National Institutes of Health.

Author Contributions

C.H.D. and K.A. conceived of the study and wrote the manuscript. C.H.D. performed analyses and wrote the initial draft of the manuscript. C.H.D. and N.S. performed formal histologic scoring of features by WSI for NCI and UCL cases. C.G. performed a detailed histologic review of a subset of WSI. K.A., P.J.C., and M.Q. integrated histological and molecular findings for NCI cases. C.H.D., N.S., S.R., Z.A., H.-J.C., K.D., and C.K. extracted molecular and clinical data from the medical record for NCI cases. H.-J.C., C.K.F., I.L., and M.Raffeld analyzed sequencing and PCR data generated for NCI cases. Z.A., O.S., and R.T. generated DNA methylation array data for NCI cases. K.G. and M.S. reviewed histologic findings, developed working diagnoses, generated DNA methylation array data, and extracted molecular and clinical data from the medical record for NYU cases. A.E., N.E., S.E.F., W.P., M.P., M.Ratliff, G.R., J.S., D.S., C.T., and A.D. reviewed histologic findings, developed working diagnoses, generated DNA methylation array data, and extracted molecular and clinical data from the medical record for DKFZ cases. S.B. reviewed histologic findings, developed working diagnoses, generated DNA methylation array data, and extracted molecular and clinical data from the medical record for UCL cases. O.L.A.N., S.A., F.A., T.B., S.C., K.S.C., J.C., F.D.A.C., I.F., A.G., J.G., H.S.L., M.B.S.L., Q.M., M.S.M., M.M., S.G.N., J.J.N., P.P., L.R., D.T., J.T.O., A.T., R.V., N.W., and C.G. reviewed histologic findings, developed working diagnoses, and extracted molecular and clinical data from the medical record for cases submitted to NCI. T.A., M.G., B.H.O., M.P.-P., and J.W. extracted clinical data from the medical record for cases evaluated at NCI. All authors read, revised, and approved of a draft of the manuscript.

Data Availability

The methylation data generated for this study will be made available upon reasonable request.