Keywords: Dehydrogenase, Glioma, Isocitrate, Oligodendrocyte, Transcription Factor 2

GBMLGG Dataset

Previously harmonized clinical (including histologic diagnoses and survival data), mutational, DNA methylation (by Illumina HumanMethylation450 BeadChip), and gene expression (by RNA-sequencing) data from GBM and LGG cohorts (GBMLGG) (21) were obtained from Xena Platform hosted by University of California, Santa Cruz (http://xena.ucsc.edu) (22). The IDH mutant group (MT) was defined as cases with eitherIDH1 p.R132 orIDH2 p.R172 mutations. The IDH-wild-type group (WT) was defined as cases without any somaticIDH1 orIDH2 mutations. Three cases were excluded due toIDH2 somatic alterations in sites other thanp.R172.

Detection of Chromosomal Codeletion of 1p/19q andCDKN2A/CDKN2B Deletion

To molecularly identify oligodendrogliomas, we utilized DNA copy number analysis (CNA) data (with germline variations removed) as log2 (tumor/normal) scores (17,23) to computationally detect chromosomal deletions of the short arm of chromosome 1 (1p) and the long arm of chromosome 19 (19q). For convenience, CNA segments are mapped and averaged (with weights of segment lengths) to chromosome bands. A deletion of chromosome band is defined as ≤−0.5 in the CNA score. Although a loss of the chromosome is defined as the loss of at least half of their bands (totally 33 bands in 1p and 12 in 19q), the final cases with 1p/19q codeletion have lost at least 29 bands in 1p and 9 in 19q.

CDKN2A andCDKN2B are 2 closely adjacent genes located in chromosome 9p21.3. We utilized DNA CNA GISTIC2-Threshold data (23) provided by UCSC Xena repository, which produced gene-level copy number estimates that were further thresholded to −2, −1, 0, 1, and 2. Only cases with the same copy number betweenCDKN2A andCDKN2B and nongain copy numbers are included (only 8 out of 431 cases in the IDH-mutant group or 24 out of 809 total cases in the combined GBMLGG dataset were excluded). All cases with 1p/19q codeletions were removed from this analysis due to lack of clinical significance from previous studies (19,20).

Differences in Genomic DNA-Methylation and Expression Between IDH Mutant and Wild-Type Groups

The DNA methylation level for an individual CpG site (with an Illumina cg identifier) was presented as the percentage of alleles methylated (ranging from 0 to 1), which was termed beta-value. Because the beta-values commonly had bimodal distribution in the 2 ends (either near-100% unmethylated or near-100% methylated for a site), they were converted to M-values (24) for normalization with a formula M-value = log2 ([beta + psc]/[1−beta + psc]), where pseudocount (psc, 0.0001) was introduced to limit the skewing from outliers. For each CpG site, we compared the M-values between the IDH mutated and wild-type groups with 2-tailed Studentt-tests for independent groups. In the discussion of individual CpG sites (such as in theOLIG2 gene), we reported the difference of median beta-values (percentages of methylation) between the 2 groups to facilitate understanding.

Expression levels of each gene were presented as RSEM scores (25), a normalized and base 2 logarithm-transformed score from RNA-sequencing counts. For each gene, theRSEM scores were compared between the IDH mutant and wild-type groups with 2-tailed Studentt-tests for independent groups.

To study the correlation of theOLIG2 expression with the transcriptome and DNA methylome (in M-value), we used the Empirical Bayes statistic models (26,27), adjusted with IDH mutation status.

Statistical significance was defined byBonferroni corrected cutoffs (28) with the numbers of tests (ɑ = 0.05/numbers of tests). All the Studentt-tests were performed as equal variances not assumed to avoid false positivity.

Independent Gene Expression Profile Data for Corroboration

As a corroboration, we used gene expression profiles (GEPs) from an independent glioma cohort (GSE4290) (29) from Gene Expression Omnibus (GEO), which was procured with Affymetrix Human Genome U133 Plus 2.0 microarray platform. Expression levels are log 2 transformed and annotated withGPL570 file from GEO (accessed on September 27, 2021, last update on December 14, 2020).OLIG2 expression level for each specimen was defined as the average of the log 2 readings of probes “213824_at” and “213825_at.” Due to the unavailability of IDH data, the grouping was defined by reported histology and clustering shown by Uniform Manifold Approximation and Projection (UMAP) dimension reduction.

OLIG2-Protein Binding Sites

To determine OLIG2 target genes, we used OLIG2 chromatin immunoprecipitation sequencing (ChIP-Seq) data previously reported by Jolma et al (European Nucleotide Archive [ENA] accession numbers ERP001824 and ERP001826) (30). The alignment and peak-calling were performed as described in the original paper (116 688 peaks), with GRCh38/hg38 as the reference genome. After comparison of output and peak-calling quality, we selected the ChIP-Seq dataset performed against the full OLIG2 protein (ERR193878, ERR193879, ERR193880, and ERR193881), rather than the DNA-binding domain. The genes corresponding to the peaks were defined by neighboring genes within a distance of 5000 bp from peaks as in previous mouse studies described by Satoh et al (31). A total of 13 694 genes were assigned as OLIG2 target genes (compared with ∼6000 genes in mouse reported by Satoh et al). The CpG sites corresponding to the ChIP-Seq peaks were defined as being within the peaks and a total of 25 893 CpG sites were identified.

Pathway Analysis

We used Illumina Beadchip annotation file (GPL16304) to associate the genes and the neighboring CpG sites. Significant differentially expressed genes (between IDH mutant and wild-type groups) were separated into 3 lists: increased expression in IDH mutant group with associated differentially methylated CpG site(s), decreased expression in IDH mutant group with associated differentially methylated CpG site(s), and increased or decreased expression without associated differentially methylated CpG sites. Pathway enrichment analysis was performed onhttps://reactome.org/AnalysisService/, with all default settings.

Statistics

Data transformation and Student’st-tests are performed onKNIME v4.4 platform (www.knime.com). The Empirical Bayes statistic models were performed with the “limma” (27) package on R v3.6 (www.r-project.org). Survival analysis and diagram plotting were performed with Kaplan-Meier survival curves with the “survminer” (32) and “survival” (33) packages on R. All the UMAP models for GEP were constructed with the “umap” package (34,35) on R with “n_neighbors = 15, random_state = 123” setting, although the plotting was performed on Python 3 for esthetic considerations.

Glioma Cohorts and Datasets

There were 516 patients in the LGG cohort and 595 patients in the GBM cohort. There were 822 patients whose IDH statuses were ascertained through whole exome sequencing data. There were 661 patients whose 1p/19q statuses were ascertained through CNA. There were 702 patients with gene expression data and 685 patients with DNA methylation data, although the total patient numbers for individual genes or CpG sites may vary due to the nature of high-throughput platforms (Table 1).

The IDH mutation pattern comprised 375 cases withp.R132H, 17 cases withp.R132C, 12 cases withp.R132G, and 9 cases withp.R132S IDH1 mutations. In addition, there were 20 cases withIDH2 p.R172 mutations (12 of which [or 60%] had 1p/19q codeletion).

Similar to previous observations, patients with LGG and IDH wild-type status (median OS = 758 days) had significantly worse overall survival than those with LGG and IDH mutant status (median OS = 2988 days), although better than patients with IDH wild-type GBM (median OS = 406 days) (Fig. 2A). The histologic features of IDH mutant and IDH wild-type tumors were similar within each cohort (LGG and GBM) from reviewing the Digital Pathology Images in TCGA (Fig. 2E–H). The overall mRNA sequencing data showed distinct separation between IDH mutant and wild-type gliomas by UMAP dimension reduction (Fig. 3A, B).

OLIG2 Expression

OLIG2 expression was significantly higher in IDH mutant tumors compared to wild-type tumors overall (1.93 [1.73–2.12] fold, p = 2.08 × 10⁻⁶⁷), as well as both LGG (1.54 [1.22–1.87] fold, p = 2.64 × 10⁻¹⁶) and GBM cohorts (2.00 [1.24–2.77] fold, p = 0.0002) (Fig. 4A).

Within the IDH mutant tumors,OLIG2 expression was significantly but only slightly higher in 1p/19q codeleted tumors (oligodendrogliomas) compared to those without the codeletion (astrocytomas) (0.20 [0.05–0.35] fold, p = 0.011). After excluding oligodendrogliomas (with 1p/19q codeletion),OLIG2 expression was significantly higher in IDH mutant gliomas compared to IDH wild-type tumors (1.87 [1.62–2.12] folds, p = 2.76 × 10⁻³⁹) (Fig. 4B).

Within IDH mutant, 1p/19q noncodeleted astrocytoma cases in both LGG and GBM cohorts, 179 cases wereCDKN2A/CDKN2B copy number neutral (denoted as GISTIC2-Threshold score 0), 69 cases were −1 (possibly single allelic deletion or subclonal homozygotic deletion), and 26 cases were −2 (likely homozygotic deletion). For each GISTIC2-Threshold score deletion (0, −1, and −2), OLIG2 expression was decreased by 0.29 (0.12–0.46) folds (p = 0.00086). The scale of this change is much smaller than the comparison between IDH-mutant and wild-type groups but similar to that between IDH mutant astrocytomas and oligodendrogliomas (Fig. 4C).

A receiver operating characteristic (ROC) curve for usingOLIG2 expression level to predict IDH mutation status yielded an area under the curve (AUC) of 0.899 (Fig. 4D). Within the IDH mutant tumors, an ROC curve for usingOLIG2 expression level to predict 1p/19q codeletion yielded an AUC area of 0.546 (Fig. 4E).

DifferentialOLIG2 Expression Corroboration by Independent Dataset

To corroborate the finding of differential expression ofOLIG2 between IDH mutant and wild-type gliomas, we used GEP dataset of glioma samples (GSE4290) (29). This dataset did not provide IDH mutation status, so we used histologic grades as surrogate for IDH status (i.e. Grade II and III to imply IDH mutant, while Grade IV to imply IDH wild-type). As a sanity check (36), we performed a global UMAP clustering with all the genes and observed that most of the Grade IV gliomas were clustered together (66/76, 87%) and most of the Grade II and III gliomas were clustered together (54/76, 71%) (Fig. 3C, compared withFig. 3A, B from TCGA cohorts), suggesting grouping by histologic grades is imperfect but informative. In this dataset,OLIG2 expression is 0.83 (0.44–1.22)-fold higher in Grade II or III gliomas compared to Grade IV gliomas (p = 4.29 × 10⁻⁵).

Alternatively, we also attempted to reclassify the cases by GEP clusters suggested by UMAP (Fig. 3D). The cluster with enriched Grade IV gliomas was inferred as “IDH-wild-type-like” cases (n = 87); the cluster with enriched Grade II, III, and oligodendrogliomas was inferred as “IDH-mutant-like” cases (n = 49); the cluster admixed with nontumor brain tissue (from brain tissue resected for epilepsy) was inferred as “normal-like” cases (n = 16), which were removed from the analysis due to a concern about tumor purity. Compared with IDH-wild-type-like cases, IDH-mutant-like cases have 1.41 (1.06–1.76) folds higher expression ofOLIG2 (p = 6.2 × 10⁻¹³).

DNA Methylation ofOLIG2

Methylation analysis revealed that, in comparison with IDH wild-type tumors, IDH mutant tumors had a significantly hypermethylated region (−6124 to −5397 nucleotides upstream fromOLIG2 transcription start site [TSS], up to 0.47 in difference of median beta-value [e.g. cg06412358]) and a significantly hypomethylated region (−1647 to −1018 nucleotides upstream fromOLIG2 TSS, up to −0.48 in difference [e.g. cg07601542]). This finding was similar in both LGG and GBM cohorts (Fig. 4F).

Outcome Analysis forOLIG2 Expression

For each analysis, cases were divided into 3 groups based onOLIG2 expression levels: Group 1 was the lowest third inOLIG2 expression levels while Group 3 was the highest third. In the overall GBMLGG cohort, the median overall survivals are 548 (485–675) days for Group 1, 2282 (1762–4695) days for Group 2, and 2660 (1933–4229) days for group 3. Group 3 shows significantly longer overall survival than Group 1 with a hazard ratio of 0.418 (0.351–0.499, p < 0.0001) (Fig. 2B). However, there were no overall survival differences between groups within either IDH mutant (Group 3 vs Group 1 HR = 1.236 [0.9182–1.663], p = 0.2) or wild-type (0.8635 [0.7144–1.044], p = 0.1) cohorts (Fig. 2C, D).

Global DNA Methylation and Gene Expression Differences Between IDH Groups

In our gene expression (transcriptome) study, 20 242 genes were successfully compared between the IDH mutant and wild-type tumors, with theBonferroni correction cutoff for the p value of 2.47 × 10⁻⁶. There were 11 605 genes significantly differentially expressed between the 2 groups, with decreased expression of 6398 (31.6%) and increased expression of 5207 (25.7%) genes in the IDH mutant group (Fig. 5A).

In our genome-wide methylation (methylome) study with combined cohort GBMLGG, 393 153 CpG sites were successfully compared between IDH mutant and wild-type groups. Therefore, theBonferroni correction cutoff for the p value of the methylome was 1.27 × 10⁻⁷. There were 210 193 sites (53.5%) significantly differentially methylated CpG sites between the 2 groups, with 192 407 (48.9%) hypermethylated and 17 786 (4.5%) hypomethylated sites in the IDH mutant group (Fig. 5B). As a visual comparison, we also performed a similar comparison between IDH mutant and wild-type acute myeloid leukemia with the TCGA LAML cohort (37) (Fig. 5C).

Among the interrogated genes, 19 926 genes had at least 1 CpG site interrogated in the methylome study. Among the significantly differentially expressed genes, 11 044 genes also had at least 1 differentially methylated CpG site, while 476 genes did not. Among the nonsignificantly differentially expressed genes, 7951 genes had significantly differentially methylated CpG sites, while 455 genes did not. The p value for chi-square test was 2.3 × 10⁻⁵, suggesting the differential methylation pattern was significantly associated with the differential gene expression pattern.

Genes and CpG Sites Associated WithOLIG2 Expression

Since IDH mutation status has been shown to be highly associated withOLIG2 expression, as well as with the transcriptome and methylome, to assess correlation of gene expression withOLIG2 (instead of IDH), we utilized a linear model between the expressions ofOLIG2, and genes of interest adjusted by IDH status. Out of 20 529 genes tested, 5504 genes (26.8%) were significantly associated withOLIG2 expression. Although there were similar numbers of positively associated genes (n = 2869, 14.0%) as the negatively associated genes (n = 2635, 12.8%) that passed theBonferroni correction line, the top genes with top p values and absolute values of coefficients were exclusively positively associated withOLIG2 expression (Fig. 5D andSupplementary Data S1). Out of 8956 genes with neighboring OLIG2-protein binding peaks, 2573 genes (28.7%) were significantly associated withOLIG2 expression after a correction with the overallBonferroni cutoff, with 1391 genes (15.5%) positively and 1182 genes (13.2%) negatively associated.

Similarly, we utilized a linear model to highlight the expression ofOLIG2 and DNA methylome adjusted by IDH status (Fig. 5E andSupplementary Data S2). Out of 396 065 tested CpG sites, 30 180 sites (7.6%) were significantly associated withOLIG2 expression afterBonferroni correction, with 22 881 sites (5.8%) positively and 7299 sites (1.8%) negatively associated. Out of 23 002 CpG sites that locate within the OLIG2 ChIP peaks, 1144 sites (5.0%) were positively and 301 sites (1.3%) negatively associated withOLIG2 expression after a correction with the overallBonferroni cutoff (1445 sites [6.3%] in total).

Pathway Analysis

The top enriched pathways for genes (n = 4195) with increased expression in the IDH mutant group with differentially methylated CpG sites (likely driving-genes) in IDH mutant tumors, were “protein-protein interactions at synapses” (p = 1.6 × 10⁻⁶), “neurexins and neuroligins” (p = 1.8 × 10⁻⁵), and “neuronal system” (p = 1.3 × 10⁻³). On the other hand, genes (n = 5333) with decreased expression in the IDH mutant group with differentially methylated CpG sites were enriched in “interferon gamma signaling” (p = 2.4 × 10⁻¹⁴), “interferon alpha/beta signaling” (p = 8.3 × 10⁻¹²), and “interferon signaling” (p = 1.7 × 10⁻¹¹) pathways.

Differentially expressed genes without differentially methylated CpG sites (n = 381) were enriched in “processing of capped intron-containing pre-mRNA” (p = 1.3 × 10⁻⁵), “mRNA splicing—major pathway” (p = 2.1 × 10⁻⁵), and “mRNA splicing” (p = 4.3 × 10⁻⁵). These genes were likely related to IDH-independent mechanisms in IDH wild-type tumors.

To evaluate the expressions of the genes potentially regulated by OLIG2, we also performed a gene set enrichment analysis on OLIG2 target genes that were significantly underexpressed in the IDH mutant tumors (n = 3030). Pathways for “G1/S Transition” (p = 8.7 × 10⁻⁶) and “DNA Replication” (p = 1.1 × 10⁻⁵) were enriched, while the enrichment of genes involving in “interferon gamma signaling,” seen in the underexpressed genes associated with differential methylation, was absent. There were 2489 OLIG2 target genes overexpressed in the IDH mutant tumors, and pathways “neuronal system” (p = 2.9 × 10⁻⁷), “protein-protein interactions at synapses” (p = 1.4 × 10⁻⁶), and “neurexins and neuroligins” (p = 8.8 × 10⁻⁶) were enriched, similar to the overexpressed genes associated with differential methylation (Table 2).

• 1.Koh HJ, Lee SM, Son BG, et al. Cytosolic NADP+-dependent isocitrate dehydrogenase plays a key role in lipid metabolism. J Biol Chem 2004;279:39968–74 [DOI] [PubMed] [Google Scholar]
• 2.Hurley JH, Dean AM, Koshland DE, et al. ; Stroud RM. Catalytic mechanism of NADP(+)-dependent isocitrate dehydrogenase: Implications from the structures of magnesium-isocitrate and NADP+ complexes. Biochemistry 1991;30:8671–8. [DOI] [PubMed] [Google Scholar]
• 3.Xu X, Zhao J, Xu Z, et al. Structures of human cytosolic NADP-dependent isocitrate dehydrogenase reveal a novel self-regulatory mechanism of activity. J Biol Chem 2004;279:33946–57 [DOI] [PubMed] [Google Scholar]
• 4.Waitkus MS, Diplas BH, Yan H.. Isocitrate dehydrogenase mutations in gliomas. Neuro Oncol 2016;18:16–26 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 5.Yan H, Parsons DW, Jin G, et al. IDH1 and IDH2 mutations in gliomas. N Engl J Med 2009;360:765–73 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 6.Dang L, White DW, Gross S, et al. Cancer-associated IDH1 mutations produce 2-hydroxyglutarate. Nature 2009;462:739–44 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 7.Christensen BC, Smith AA, Zheng S, et al. DNA methylation, isocitrate dehydrogenase mutation, and survival in glioma. J Natl Cancer Inst 2011;103:143–53 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 8.Noushmehr H, Weisenberger DJ, Diefes K, et al. ; Cancer Genome Atlas Research Network. Identification of a CpG island methylator phenotype that defines a distinct subgroup of glioma. Cancer Cell 2010;17:510–22 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 9.Louis DN, Perry A, Reifenberger G, et al. The 2016 World Health Organization classification of tumors of the central nervous system: A summary. Acta Neuropathol 2016;131:803–20 [DOI] [PubMed] [Google Scholar]
• 10.Brat DJ, Verhaak RGW, Aldape KD, et al. ; Cancer Genome Atlas Research Network. Comprehensive, integrative genomic analysis of diffuse lower-grade gliomas. New Engl J Med 2015;372:2481–98 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 11.Brat DJ, Aldape K, Colman H, et al. cIMPACT-NOW update 3: Recommended diagnostic criteria for “Diffuse astrocytic glioma, IDH-wildtype, with molecular features of glioblastoma, WHO grade IV.” Acta Neuropathol 2018;136:805–10 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 12.Yokoo H, Nobusawa S, Takebayashi H, et al. Anti-human Olig2 antibody as a useful immunohistochemical marker of normal oligodendrocytes and gliomas. Am J Pathol 2004;164:1717–25 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 13.Popova SN, Bergqvist M, Dimberg A, et al. Subtyping of gliomas of various WHO grades by the application of immunohistochemistry. Histopathology 2014;64:365–79 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 14.Suzuki A, Nobusawa S, Natsume A, et al. Olig2 labeling index is correlated with histological and molecular classifications in low-grade diffuse gliomas. J Neurooncol 2014;120:283–91 [DOI] [PubMed] [Google Scholar]
• 15.Tsigelny IF, Kouznetsova VL, Lian N, Kesari S.. Molecular mechanisms of OLIG2 transcription factor in brain cancer. Oncotarget 2016;7:53074–101 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 16.Lu F, Chen Y, Zhao C, et al. Olig2-dependent reciprocal shift in PDGF and EGF receptor signaling regulates tumor phenotype and mitotic growth in malignant glioma. Cancer Cell 2016;29:669–83 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 17.Cancer Genome Atlas Research Network. Comprehensive genomic characterization defines human glioblastoma genes and core pathways. Nature 2008;455:1061–8 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 18.Louis DN, Perry A, Wesseling P, et al. The 2021 WHO classification of tumors of the central nervous system: A summary. Neuro Oncol 2021;23:1231–51 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 19.Shirahata M, Ono T, Stichel D, et al. Novel, improved grading system(s) for IDH-mutant astrocytic gliomas. Acta Neuropathol 2018;136:153–66 [DOI] [PubMed] [Google Scholar]
• 20.Reis GF, Pekmezci M, Hansen HM, et al. CDKN2A loss is associated with shortened overall survival in lower-grade (World Health Organization Grades II–III) astrocytomas. J Neuropathol Exp Neurol 2015;74:442–52 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 21.Ceccarelli M, Barthel FP, Malta TM, et al. ; TCGA Research Network. Molecular profiling reveals biologically discrete subsets and pathways of progression in diffuse glioma. Cell 2016;164:550–63 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 22.Goldman MJ, Craft B, Hastie M, et al. Visualizing and interpreting cancer genomics data via the Xena platform. Nat Biotechnol 2020;38:675–8 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 23.Mermel CH, Schumacher SE, Hill B, et al. GISTIC2.0 facilitates sensitive and confident localization of the targets of focal somatic copy-number alteration in human cancers. Genome Biol 2011;12:R41. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 24.Du P, Zhang X, Huang CC, et al. Comparison of Beta-value and M-value methods for quantifying methylation levels by microarray analysis. BMC Bioinformatics 2010;11:587. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 25.Li B, Dewey CN.. RSEM: Accurate transcript quantification from RNA-Seq data with or without a reference genome. BMC Bioinformatics 2011;12:323. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 26.Ritchie ME, Phipson B, Wu D, et al. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res 2015;43:e47. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 27.R: Empirical Bayes Statistics for Differential Expression [Internet]. Available at:http://web.mit.edu/~r/current/arch/i386_linux26/lib/R/library/limma/html/ebayes.html. Accessed May 25, 2022
• 28.Noble WS.How does multiple testing correction work?Nat Biotechnol2009;27:1135–7 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 29.Sun L, Hui AM, Su Q, et al. Neuronal and glioma-derived stem cell factor induces angiogenesis within the brain. Cancer Cell 2006;9:287–300 [DOI] [PubMed] [Google Scholar]
• 30.Jolma A, Yan J, Whitington T, et al. DNA-binding specificities of human transcription factors. Cell 2013;152:327–39 [DOI] [PubMed] [Google Scholar]
• 31.Satoh JI, Asahina N, Kitano S, et al. A comprehensive profile of ChIP-Seq-based Olig2 target genes in motor neuron progenitor cells suggests the possible involvement of Olig2 in the pathogenesis of amyotrophic lateral sclerosis. J Cent Nerv Syst Dis 2015;7:1. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 32.Kassambara A, Kosinski M, Biecek P, et al. Survminer: Drawing Survival Curves using “ggplot2” [Internet]. 2021. Available at:https://CRAN.R-project.org/package=survminer. Accessed May 25, 2022
• 33.Therneau TM, Elizabeth A, Cynthia C..survival: Survival Analysis [Internet]. 2022. Available at:https://CRAN.R-project.org/package=survival. Accessed May 25, 2022
• 34.Konopka T. umap: Uniform Manifold Approximation and Projection [Internet]. 2022. Available at:https://CRAN.R-project.org/package=umap. Accessed May 26, 2022
• 35.McInnes L, Healy J, Melville J,UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction [Internet]. arXiv. 2020. Report No.: arXiv:1802.03426. Available at:http://arxiv.org/abs/1802.03426. Accessed May 26, 2022
• 36.Tran PMH, Tran LKH, Nechtman J, et al. Comparative analysis of transcriptomic profile, histology, and IDH mutation for classification of gliomas. Sci Rep 2020;10:20651. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 37.The Cancer Genome Atlas Research Network. Genomic and epigenomic landscapes of adult de novo acute myeloid leukemia. N Engl J Med 2013;368:2059–74 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 38.Malta TM, de Souza CF, Sabedot TS, et al. Glioma CpG island methylator phenotype (G-CIMP): Biological and clinical implications. Neuro Oncol 2018;20:608–20 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 39.Meijer DH, Kane MF, Mehta S, et al. Separated at birth? The functional and molecular divergence of OLIG1 and OLIG2. Nat Rev Neurosci 2012;13:819–31 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 40.Novitch BG, Chen AI, Jessell TM.. Coordinate regulation of motor neuron subtype identity and pan-neuronal properties by the bHLH repressor Olig2. Neuron 2001;31:773–89 [DOI] [PubMed] [Google Scholar]
• 41.Marshall CAG, Novitch BG, Goldman JE.. Olig2 directs astrocyte and oligodendrocyte formation in postnatal subventricular zone cells. J Neurosci 2005. Aug 10;25:7289–98 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 42.Chen Y, Miles DK, Hoang T, et al. The basic helix-loop-helix transcription factor olig2 is critical for reactive astrocyte proliferation after cortical injury. J Neurosci 2008;28:10983–9 [DOI] [PMC free article] [PubMed] [Google Scholar]

Supplementary Materials