Citations

Abstract Diffuse hemispheric glioma, H3 G34‐mutant (DHG‐H3 G34) has been primarily molecularly characterized by methylation profiling and sequencing studies. We describe an integrated histomolecular evaluation including high‐resolution copy number profiling of a series of 60 DHG‐H3 G34 to further our understanding of the spectrum of genetic changes associated with this tumor type. Cases were clinically tested using an 187‐gene mutation and fusion targeted neuro‐oncology next‐generation sequencing panel ( n = 60) and Oncoscan chromosomal microarray ( n = 26) by a single laboratory (2018–2022). A subset of cases had immunohistochemical results for OLIG2 ( n = 42), p53 ( n = 48), and ATRX ( n = 46), and methylation array data ( n = 8). Median age at testing was 21 years (range, 12–50). No significant difference was noted in clinical, histopathological, and mutational profile between pediatric and adult patients. H3‐3A G34 mutations included G34R ( n = 56; 94%), G34V ( n = 3), and a non‐canonical G34E ( n = 1). Concurrent mutations most often involved TP53 ( n = 55; 92%), ATRX ( n = 50; 83%), and PDGFRA ( n = 34; 57%). A reportedly primary tumor was confirmed to be hypermutant and had a PMS2 mutation. A single case also showed an FGFR3::FAM184B fusion. All cases with available chromosomal microarray data had unbalanced genomes, which were often complex (14/26; 54%). The most frequent recurrent copy number abnormalities were losses involving 3q, 4q, 10q, 13q, and 18q, and 17p copy‐neutral loss of heterozygosity (cnLOH) encompassing TP53 . This copy number profile was reminiscent of that seen in Grade 4 IDH‐mutant astrocytomas. Collectively, a TP53 abnormality at copy number (12/26, all cnLOH), sequence (55/60) and protein expression (46/48) level was detected in all 60 cases. In conclusion, integrated high‐resolution copy number and histomolecular analysis expanded the spectrum of genetic changes associated with DHG‐H3 G34, including the presence of universal TP53 abnormalities with frequent cnLOH—a copy number abnormality that has been largely unrecognized—for this new 2021 World Health Organization central nervous system tumor type.

Abstract Computer vision now enables deep learning models to link histopathologic features to molecular features in a quantitative manner not previously achievable by human pathologists. However, human interpretability is often limited. We trained a deep neural network to predict gene expression from digital whole slide images (WSIs) using 848 central nervous system (CNS) tumors with paired WSI and RNA-sequencing data. We used inferred RNA expression levels as a surrogate for protein expression of the CNS tumor lineage markers GFAP, OLIG2, and SSTR2, as well as the proliferation marker MKI67 (Ki67). We established thresholds for categorizing the inferred expression levels as positive or negative based on levels observed in cross-validation testing in select tumor types to approximate routine clinical interpretation of immunohistochemical (IHC) staining. We tested the sensitivity and specificity of our ‘virtual’ IHC on an independent, multi-institutional cohort of over 2,000 CNS tumors with objective diagnostic labels derived from DNA methylation-based tumor classification. As a dichotomous variable, inferred GFAP expression showed a sensitivity of 74% and a specificity of 99% as a glial marker in a cohort of gliomas and meningiomas. In the same cohort, OLIG2 showed a sensitivity of 75% and a specificity of 97% as a glial marker, while SSTR2 showed a sensitivity of 83% and a specificity of 97% as a marker of meningioma. In a cohort of ependymomas and gliomas, virtual IHC for OLIG2 showed a sensitivity of 75% and a specificity of 82% in distinguishing ependymomas from gliomas. Finally, in a cohort of gliomas, inferred expression of MKI67 demonstrated a trend consistent with what would be expected by actual IHC, with increasing MKI67 expression with increasing tumor grade. Our model provides the basis for a human interpretable and clinically applicable deep neural network to aid human pathologists in the diagnosis and grading of CNS tumors.

Abstract Molecular testing has become critical components in the diagnostic classification of the central nervous system (CNS) tumors. However, these methods require substantial resources, limiting accessibility for many patients. Recent advances in artificial intelligence (AI) and computer vision empower deep learning models to infer molecular features from histopathology images, which may often be more readily available than molecular testing, to classify CNS tumors. We trained pan-CNS tumor models to predict DNA methylation and gene expression from whole-slide images (WSIs). These molecular predictions were then used in a hierarchical machine learning framework, termed Neuropath-AI, to predict nine broad tumor families and 52 tumor types using a large diverse cohort of 5,715 histopathology samples, of which 2,988 had paired DNA methylation profiling and 848 had paired RNA-sequencing. We evaluated Neuropath-AI on another large independent multi-institutional cohort of 5,516 CNS tumors, for which Neuropath-AI predicted tumor class with an associated confidence score. Neuropath-AI designated 46% of test samples as predictable with high-confidence, which it then accuracy classified in 97% of samples. It made moderate-confidence (and above) predictions for 87% of test samples, for which it achieved a diagnostic accuracy of 80% for the top-1 prediction and 86% accuracy for a top-2 predictions. A comparison with human neuropathologists showed that Neuropath-AI is comparable at classifying tumors. A second test showed that human neuropathologists were more accurate when the Neuropath-AI classifier results were provided to them, testifying to the potential translational benefit of integrating AI classification tools into the pathologist’s workflow. Our model provides the basis for a clinically applicable deep learning assistant to improve human efficiency and accuracy of CNS tumor diagnosis. The model will be made publicly available and can be readily implemented to assist human pathologists in clinical workflows, in further expanded prospective studies.

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.

Abstract Background Adult onset medulloblastoma (aMB) is a rare tumor with limited available evidence. We present a large multi-institutional retrospective cohort of aMB patients treated in the modern era, with an emphasis on understanding the role of chemotherapy at initial diagnosis. Methods We included 267 consecutive patients with aMB treated at 7 different institutions from 2000-present, controlling for chemotherapy regimen and cycles received. Results Treatment factors were highly intercorrelated with one another and with treating institution. Concurrent chemotherapy was not associated with overall survival (OS). Adjuvant chemotherapy was associated with OS on univariable analyses (HR = 0.55, P = .029) and on multivariable analysis when adjusting for risk status (HR 0.55, P = .026) but not when also adjusting for treating institution. Proton craniospinal irradiation was associated with improved survival on univariable (HR = 0.50, P = .019) and multivariable analysis adjusting for risk status (HR = 0.51, P = .024) but not when treating institution was also considered. On subgroup analysis, adjuvant chemotherapy was associated with improved survival in M0 (HR = 0.55, P = .043) but not M1 disease, in patients with subtotal resection (HR = 0.43, P = .048) but not those with GTR. Similarly, progression-free survival was improved with chemotherapy in patients with M0 (HR = 0.57, P = .032) but not M1 disease, and in patients with subtotal (HR = 0.50, P = .054) but not gross total resection. Conclusions There was no benefit of concurrent chemotherapy. Adjuvant chemotherapy was associated with improved overall survival and this effect was driven by select subgroups, specifically those with M0 disease and those with residual tumor. We could not confirm that these associations are independent of the treating institution.

Abstract Most TERT promoter (TERTp) clinically relevant sequence variants in CNS tumors are the canonical C228T and C250T mutations. Non-canonical TERTp mutations have been reported mainly in non-CNS tumors. We describe 12 CNS tumors with non-canonical TERTp variants interpreted as clinically relevant. Using clinical targeted NGS panels (2020-2024), we detected c.-100_-79dup (n=2), c.-92_-91insGGCGGCCCCGCCCCTTCCTTTC (n=1), c.-118_-117insTCCCCGGCCCAGCCCCTTCCGGG (n=1), c.-123_-76dup (n=1), c.-110_-89dup (n=2), and c.-57A>C (n=5). Most cases had chromosomal microarray (n=11) and/or methylation array data analyzed using the NCI/Bethesda v.2 classifier (n=9). All non-canonical TERTp variants occurred at a variant allele frequency suggestive of somatic clonal origin (mean, 34%; range, 18-46), and in adults (mean age, 60 years; range, 36-88). Except for the c.-57A>C variant, all non-canonical TERTp variants occurred in tumors with histological and/or molecular diagnostic features of glioblastoma, IDH-wildtype (GBM), often epigenetically aligning to this tumor type. All five tested cases matched to superfamily glioblastoma (score ≥ 0.9), with three to class GBM_RTK_I (>0.9 in 2, 0.86 in 1) and two to GBM_RTK_II (>0.9). The size and sequence of such variants suggest that these may be functionally equivalent to canonical mutations or generate favorable TERTp activation motif conformations like previously reported activating duplications. The c.-57A>C variant is a well-established oncogenic mutation observed in other cancer types but has not been reported in CNS tumors. We describe five cases with the c.-57A>C variant: one GBM matching to methylation class GBM_MES_TYP (0.99); two oligodendroglioma, IDH-mutant and 1p/19q-codeleted, one of which had methylation data matching to class O_IDH (0.99); one diffuse glioma, IDH-mutant, which was partially 1p/19q-codeleted and matched to class O_IDH (0.99); and one anaplastic meningioma, aligning to superfamily meningioma (0.99)/MNG_MAL (0.79). Our findings show that non-canonical TERTp variants typically occur in tumor types that often harbor TERTp canonical mutations, expanding the spectrum of TERTp clinically relevant variants that are molecular biomarkers in CNS tumors.