Mutations in H3F3A, which encodes histone H3.3, commonly occur in pediatric glioblastoma. Additionally, H3F3A K27M substitutions occur in gliomas that arise at midline locations (eg, pons, thalamus, spine); moreover, this substitution occurs mainly in tumors in children and adolescents. Here, we sought to determine the association between H3F3A mutations and adult thalamic glioma. Genomic H3F3A was sequenced from 20 separate thalamic gliomas. Additionally, for 14 of the 20 gliomas, 639 genes—including cancer-related genes and chromatin-modifier genes—were sequenced, and the Infinium HumanMethylation450K BeadChip was used to examine DNA methylation across the genome. Of the 20 tumors, 18 were high-grade thalamic gliomas, and of these 18, 11 were from patients under 50 years of age (median age, 38 y; range, 17–46), and 7 were from patients over 50 years of age. The H3F3A K27M mutation was present in 10 of the 11 (91%) younger patients and absent from all 7 older patients. Additionally, H3F3A K27M was not detected in the 2 diffuse astrocytomas. Further sequencing revealed recurrent mutations in TP53, ATRX, NF1, and EGFR. Gliomas with H3F3A K27M from pediatric or young adult patients had similar, characteristic DNA methylation profiles. In contrast, thalamic gliomas with wild-type H3F3A had DNA methylation profiles similar to those of hemispheric glioblastomas. We found that high-grade thalamic gliomas from young adults, like those from children and adolescents, frequently had H3F3A K27M.

Immune-based therapies have shown limited efficacy in glioma thus far. This might be at least in part due to insufficient numbers of neoantigens, thought to be targets of immune attack. In addition, we hypothesized that dynamic genetic and epigenetic tumor evolution in gliomas might also affect the mutation/neoantigen landscape and contribute to treatment resistance through immune evasion. Here, we investigated changes in the neoantigen landscape and immunologic features during glioma progression using exome and RNA-seq of paired primary and recurrent tumor samples obtained from 25 WHO grade II–IV glioma patients (glioblastoma, IDH-wild-type, n = 8; grade II–III astrocytoma, IDH-mutant, n = 9; and grade II–III oligodendroglioma, IDH-mutant, 1p/19q-codeleted, n = 8). The number of missense mutations, predicted neoantigens, or expressed neoantigens was not significantly different between primary and recurrent tumors. However, we found that in individual patients the ratio of expressed neoantigens to predicted neoantigens, designated the “neoantigen expression ratio,” decreased significantly at recurrence (P = 0.003). This phenomenon was particularly pronounced for “high-affinity,” “clonal,” and “passenger gene–derived” neoantigens. Gene expression and IHC analyses suggested that the decreased neoantigen expression ratio was associated with intact antigen presentation machinery, increased tumor-infiltrating immune cells, and ongoing immune responses. Our findings imply that decreased expression of highly immunogenic neoantigens, possibly due to persistent immune selection pressure, might be one of the immune evasion mechanisms along with tumor clonal evolution in some gliomas.

Machine learning models for automated magnetic resonance image segmentation may be useful in aiding glioma detection. However, the image differences among facilities cause performance degradation and impede detection. This study proposes a method to solve this issue. We used the data from the Multimodal Brain Tumor Image Segmentation Benchmark (BraTS) and the Japanese cohort (JC) datasets. Three models for tumor segmentation are developed. In our methodology, the BraTS and JC models are trained on the BraTS and JC datasets, respectively, whereas the fine-tuning models are developed from the BraTS model and fine-tuned using the JC dataset. Our results show that the Dice coefficient score of the JC model for the test portion of the JC dataset was 0.779 ± 0.137, whereas that of the BraTS model was lower (0.717 ± 0.207). The mean Dice coefficient score of the fine-tuning model was 0.769 ± 0.138. There was a significant difference between the BraTS and JC models (p < 0.0001) and the BraTS and fine-tuning models (p = 0.002); however, no significant difference between the JC and fine-tuning models (p = 0.673). As our fine-tuning method requires fewer than 20 cases, this method is useful even in a facility where the number of glioma cases is small.

Recent studies have demonstrated that tumor-driving alterations are often different among gliomas that originated from different brain regions and have underscored the importance of analyzing molecular characteristics of gliomas stratified by brain region. Therefore, to elucidate molecular characteristics of diffuse cerebellar gliomas (DCGs), 27 adult, mostly glioblastoma cases were analyzed. Comprehensive analysis using whole-exome sequencing, RNA sequencing, and Infinium methylation array (n = 17) demonstrated their distinct molecular profile compared to gliomas in other brain regions. Frequent mutations in chromatin-modifier genes were identified including, noticeably, a truncating mutation in SETD2 (n = 4), which resulted in loss of H3K36 trimethylation and was mutually exclusive with H3F3A K27M mutation (n = 3), suggesting that epigenetic dysregulation may lead to DCG tumorigenesis. Alterations that cause loss of p53 function including TP53 mutation (n = 9), PPM1D mutation (n = 2), and a novel type of PPM1D fusion (n = 1), were also frequent. On the other hand, mutations and copy number changes commonly observed in cerebral gliomas were infrequent. DNA methylation profile analysis demonstrated that all DCGs except for those with H3F3A mutations were categorized in the “RTK I (PDGFRA)” group, and those DCGs had a gene expression signature that was highly associated with PDGFRA. Furthermore, compared with the data of 315 gliomas derived from different brain regions, promoter methylation of transcription factors genes associated with glial development showed a characteristic pattern presumably reflecting their tumor origin. Notably, SOX10, a key transcription factor associated with oligodendroglial differentiation and PDGFRA regulation, was up-regulated in both DCG and H3 K27M-mutant diffuse midline glioma, suggesting their developmental and biological commonality. In contrast, SOX10 was silenced by promoter methylation in most cerebral gliomas. These findings may suggest potential tailored targeted therapy for gliomas according to their brain region, in addition to providing molecular clues to identify the region-related cellular origin of DCGs.