The formalin-fixed paraffin-embedded (FFPE) samples were prepared per the standard protocol. Next, slides were mounted with 5-µm thick tissue sections and stained with hematoxylin and eosin. Multiple stainings were performed on the adjacent section with antibodies against IDH1 R132H (Dianova GmbH, H09 clone), GFAP (Dako, 6F2 clone), p53 (Santa Cruz Biotechnology, DO-1 clone), ATRX (Sigma-Aldrich, polyclonal), and Ki-67 (Dako, Mib-1 clone). All specimens were evaluated by two pathologists (S.M. and K.K.).

Genomic DNA (gDNA) was extracted from the buffy coat and frozen tumor specimens using a QIAamp DNA extraction kit (Qiagen, Hilden, Germany) or from unstained FFPE samples using a QIAamp DNA FFPE Kit (Qiagen, Hilden, Germany). The concentration of gDNA was determined using Qubit (Thermo Fisher Scientific, Waltham, MA, USA).

For IDH1 sequencing, we amplified a 129-bp fragment including codon 132. Polymerase chain reaction (PCR) was performed as follows: denaturation with 35 cycles at 95 ℃ for 30 seconds, annealing at 56 ℃ for 40 seconds, and extension at 72 ℃ for 50 seconds, with a final extension step at 72 ℃ for 7 minutes. The forward primer used was 5’-CGGTCTTCAGAGAAGCCATT-3’, and the reverse primer was 5’-GCAAAATCACATTATTGCCAAC-3’. Sequence analysis was performed using ApE v2.0.60.

DNA libraries were prepared using the Agilent Sureselect V6 58 M according to the manufacturer’s instructions. The NovaSeq6000 platform (Illumina, San Diego, CA, USA) was used for sequencing, and it generated 150-bp paired-end reads. FASTQ files were applied to the GENOMON pipeline and mutation calls were performed (https://genomon-project.github.io/GenomonPagesR/). Called mutations were sorted by the following criteria: Func.refGene = “exonic” or “splicing” or “exonic; splicing”, misRate_tumor ≥ 0.05, misRate_normal < 0.02, P-value fisher_realignment > 1.1, variantPairNum_tumor ≥ 3, variantPairNum_normal < 3, strandRatio_tumor ≠ 1 & strandRatio_tumor ≠ 0. The sorted mutations are listed in Supplementary Table 1. The copy number analyses were performed using the Sequenza package [8]. The optimal phylogenetic tree was estimated using maximum parsimony and drawn using the bootstrap method in MEGA11 [9].

The Illumina Infinium Human Methylation EPIC BeadChip array (Illumina, San Diego, CA, USA) was utilized for a comprehensive genome-wide methylation analysis. gDNA, amounting to 600, was extracted from the FFPE samples. The preprocessing for the analysis of the EPIC array data and the computation of the beta score were performed with the Minfi package using R software, version 3.4.1 [10]. Following the application of the ChAMP package for filtering, the remaining probes for analysis amounted to 384,906 [11]. The beta scores were normalized using the BMIQ method in the ChAMP package. The top 5000 probes exhibiting the highest median absolute deviation on the CpG islands were selected for further analysis. The methylation profiles were analyzed with two-dimensional t-distributed stochastic neighbor embedding (t-SNE) in 2 dimensions using the Rtsne package. Reference methylation data for gliomas (GSE90496) were obtained from the Gene Expression Omnibus database (http://www.ncbi.nlm.nih.gov/geo/). A molecular classification algorithm and copy number analysis from the German Cancer Center (DKFZ classifier, https://www.molecularneuropathology.org/mnp) was performed [12]. In the copy number analysis by DKFZ classifier, a log2 ratio ± 0.4 was used as the cutoff for amplification/loss and a log2 ratio [13].

For the analysis of public datasets, the cBioPortal for cancer genomics was utilized (https://www.cbioportal.org). Primary astrocytoma, IDH-mutant with CDKN2A/B homozygous deletion cases were collected (n = 15) [1, 14, 15], and the published clinical outcomes were analyzed, and the published clinical outcomes were analyzed. Kaplan–Meier curves were drawn using Prism (version 9.3.1), and statistical differences were verified using a log-rank test.