To study age and survival-associated molecular features of oligodendroglioma, we consulted 550 oligodendroglioma tumors from seven publicly available cohorts, including the TCGA, Chinese Glioma Genome Atlas (CGGA), Capper et al., Glioma Longitudinal AnalySiS Consortium (GLASS), Jonsson et al., Oligo Nation, and the EORTC 26951 Phase III clinical trial [7, 12, 13, 37, 41, 68, 77]. Patient age (N = 481), overall survival (N = 342), and bulk RNA-sequencing data (N = 232) from the TCGA, CGGA, and GLASS were used to nominate a molecular marker associated with an older, more aggressive subset of oligodendroglioma. The DNA methylation profile of this marker was subsequently analyzed in a group of 364 tumors from the TCGA, Capper et al., and GLASS cohorts and, along with RNA-sequencing data, subjected to association tests with features in 156 TCGA pathology reports and 27 pre-operative multimodal magnetic resonance imaging volumes. To isolate our analyses to neoplastic cells and interrogate proliferating phenotypes, we used single-nucleus RNA sequencing from 7817 nuclei and single-nucleus ATAC-sequencing data from 18,549 nuclei published by Blanco-Carmona et al. and Wang et al. [10, 82]. All data analyzed in this study are summarized in Supplemental Table 1.

HOXD12 expression is age-associated and prognostic in oligodendroglioma

Contrary to most cancer types, the age distribution of oligodendroglioma patients has been reported to appear bimodal, possibly indicating that certain age ranges are associated with different tumor biology [17, 64]. Although we were unable to statistically judge whether oligodendroglioma patient age distributions were multimodal in our data (Hartigans' dip test), we did find many cohorts exhibited a non-Gaussian distribution. Unlike IDH-wildtype glioblastoma and IDH-mutant astrocytoma adult-type diffuse gliomas in the TCGA, patient age in TCGA oligodendrogliomas was not normally distributed (p < 0.04, Shapiro–Wilk) (Supplemental Fig. 5). In addition to the TCGA, oligodendroglioma age was significantly or marginally significantly non-Gaussian distributed among primary tumors in cohorts published by Jonsson et al. (N = 84, p = 0.01, Shapiro–Wilk), Capper et al. (N = 111, p = 0.048), the CGGA (N = 48, p = 0.06), the GLASS Consortium (N = 31, p = 0.08), and Oligo Nation (N = 34, p = 0.09) (Fig. 1a) [7, 13, 19, 37, 80, 91]. In addition, all non-TCGA datasets were enriched for younger, longer surviving patients compared to the TCGA, suggesting that older patients may be underrepresented and obscure a clearer bimodal distribution in these datasets (Fig. 1b). Given that age is prognostic and non-Gaussian distributed in oligodendroglioma, we hypothesized that a distinctive, aggressive subtype of oligodendroglioma that occurs at older ages may exist.

Fig. 1

Investigation of the non-Gaussian age distribution in oligodendroglioma identifies HOXD12 expression as age and survival associated. a The distribution of oligodendroglioma patient age was significantly and marginally significantly non-Gaussian in six independent oligodendroglioma cohorts. b Older age was associated with worse outcomes in cohorts with survival data. c Unbiased differential gene expression analysis between older and younger patients returned 16 significantly differentially expressed genes, including several genes in the HOX, GATA, and keratin gene families. d, e Of the four differentially expressed genes that also stratified TCGA oligodendroglioma survival, HOXD12 had the strongest association with poor outcomes. f In addition to the TCGA, older oligodendroglioma patients were also enriched for HOXD12-positive expression status in the CGGA. g HOXD12-positive expression status was prognostic in the CGGA, validating the results from the TCGA. h HOXD12-positive expression status was prognostic independent of patient age and CNS WHO grade in the TCGA and independent of CNS WHO grade in the CGGA. i Age distributions of TCGA and CGGA oligodendrogliomas when partitioned by HOXD12-positive and HOXD12-negative expression status

To test genes for age and survival association, we first performed an unbiased differential gene expression analysis comparing older (≥ 45 years) and younger TCGA oligodendrogliomas and identified 16 differentially expressed genes. These genes included developmental transcription factors in the HOX and GATA gene families, which were upregulated in older patients, and keratin genes, which were downregulated in older patients (Fig. 1c). Gene ontology analyses of the results from this differential gene expression analysis showed that the most significantly activated pathways in older TCGA oligodendrogliomas were linked to developmental transcription factors and included DNA-binding transcription activator activity (padj < 1e−6) and the development of the embryonic skeletal system (padj < 1e−7), appendages (padj < 1e−6), and limbs (padj < 1e−6) (Supplemental Fig. 6a,b). Suppressed pathways included cornification (padj < 1e−13) and keratinization (padj < 1e−10) and other pathways that involve structural proteins that control cell shape. Our results were supported by a gene set enrichment analysis, whose most significant results were the suppression of cornification (padj < 1e−13) and the activation of embryonic skeletal system development (padj < 1e−7) (Supplemental Fig. 6c).

We followed these age-related analyses with survival analyses focused on the 16 differentially expressed genes identified above. To compare high and low gene expression levels, we compared patients with gene expression above a gene’s median expression level (elevated) to patients whose gene expression is below or equal to that gene’s median expression level (low). Of these 16 genes, 4 genes conferred significantly worse outcomes when elevated, of which HOXD12 was the most significant and severe (HR = 9.3, FDR < 1e−5, log-rank) (Fig. 1d,e). Given that the median HOXD12 expression in the TCGA was zero, we refer to elevated HOXD12 expression as HOXD12-positive. HOXD12-positive expression status’ statistical associations with age and survival were validated in the CGGA (Fig. 1f,g). Older (≥ 45 years old) CGGA oligodendroglioma patients were significantly enriched for HOXD12-positive patients (p = 0.03, Fisher’s), and HOXD12-positive status was prognostic in the CGGA (p < 0.001, log-rank), incurring a similar hazard ratio (HR) as observed in the TCGA (TCGA HR = 9.3, CGGA HR = 8.4). In multivariate survival analyses, HOXD12 was the only one of the four genes identified as prognostic whose elevated expression was also prognostic independent of patient age and WHO grade in the TCGA (FDR = 0.02, CPH) (Fig. 1h, Supplemental Fig. 7). HOXD12-positive status was independently prognostic of WHO grade in the CGGA but lost its significance when patient age was accounted for, possibly because the sample size (N = 44) was relatively small for a multivariate analysis. For both TCGA and CGGA cohorts, when HOXD12-positive and HOXD12-negative patients were considered separately, their age distributions more closely resembled Gaussian distributions (Fig. 1i). Similar trends toward age and survival association were found in the GLASS dataset; however, low patient count limited these analyses (N = 9) (Supplemental Fig. 8).

HOXD12 gene body hypermethylation is age-associated and predictive of poor survival independent of HOXD12 expression

Given that DNA methylation and gene transcription are related, we sought to evaluate the relationship between HOXD12 methylation levels and patient age and survival. We found that every Illumina 450 k array DNA methylation probe associated with HOXD12 (N = 14) was positively correlated with HOXD12 expression in the TCGA, suggesting that HOXD12 hypermethylation was associated with higher transcriptional activity (Fig. 2a). Accordingly, we hypothesized that HOXD12 hypermethylation would be more prevalent in older oligodendroglioma patients and associated with poor survival. To test this hypothesis, we defined a HOXD12 methylation signature and set a threshold above which we deemed patients HOXD12 hypermethylated. This signature was developed by testing each probe for age and survival association in both univariate and multivariate analysis (Methods) (Fig. 2b, Supplemental Fig. 9). Interestingly, the only probes that reached significance (p < 0.05) in all tests were HOXD12’s three gene body probes, which is consistent with the notion that gene body methylation is often positively correlated with gene expression and a report that HOXD12 hypermethylation is linked to higher HOXD12 expression levels [69, 87]. To formally establish a HOXD12 gene body hypermethylation threshold, we considered the mean of the beta values for the three HOXD12 gene body methylation probes for patients with HOXD12-positive expression status and for patients with HOXD12-negative expression status. We chose the mean of these two values (0.3577) as the HOXD12 gene body hypermethylation threshold.

Fig. 2

HOXD12 Hypermethylation is Associated with Age and Predictive of Poor Survival Independent of HOXD12 Expression. a All HOXD12-associated methylation probes (N = 14) were significantly positively correlated with HOXD12 expression in the TCGA. b HOXD12-associated probes were tested for age, univariate survival, and multivariate survival associations in the TCGA using specified thresholds. Only the three probes located within HOXD12’s gene body passed all such tests. c HOXD12 hypermethylation was associated with age in the TCGA and the Capper et al. cohort. d HOXD12 hypermethylated oligodendrogliomas from Capper et al. and the TCGA were enriched for higher CNS WHO grade. e HOXD12 hypermethylation was prognostic in the TCGA. f Nomogram for predicting 5- and 10-year overall survival using patient age, HOXD12 methylation level, and CNS WHO grade. g Multivariate survival analyses showed HOXD12 hypermethylation was prognostic independent of HOXD12-positive expression status, age, and CNS WHO grade. the TCGA. h Oligodendroglioma TCGA patients harboring HOXD12-positive expression status and HOXD12 gene body hypermethylation had the worst outcomes (overall survival = 2.6 years)

Using this gene body methylation signature and threshold, we found that HOXD12 gene body hypermethylation was associated with age in the TCGA (p < 1e-6, Mann–Whitney U) and Capper et al. cohorts (p = 0.002, Mann–Whitney U) (Fig. 2c, Supplemental Fig. 10). While this age association was not as strong in the Capper et al. cohort, this is likely because the Capper et al. cohort was disproportionally enriched for younger, WHO grade 3 tumors compared to all other datasets we analyzed (Supplemental Fig. 11). Furthermore, HOXD12 gene body hypermethylation was prognostic in the TCGA in a univariate analysis (HR = 4.2, p < 0.001, log-rank) and two multivariate analyses, the first of which accounted for HOXD12 expression status (p < 0.003, CPH) and the second accounted for HOXD12 expression status, patient age, and WHO grade (p < 0.007, CPH) (Fig. 2d,e). Patients harboring HOXD12-positive expression status and HOXD12 gene body hypermethylation in the TCGA formed the class of oligodendrogliomas with the worst outcomes, with a median overall survival of only 2.6 years (Fig. 2f). Although outcome data was not available for the Capper et al. cohort, oligodendrogliomas from this cohort were substantially enriched for central nervous system (CNS) WHO grade 3 tumors (p = 0.01, Fisher’s), as were their TCGA counterparts (p < 0.001, Fisher’s), indicating that HOXD12 gene body hypermethylation was likely prognostic in the Capper et al. cohort (Fig. 2g). HOXD12 gene body hypermethylation in the GLASS dataset showed similar patterns but suffered from a low sample count (N = 10) (Supplemental Fig. 12). To explore the clinical applicability of HOXD12 gene body methylation, we developed a nomogram predicting 5- and 10-year overall survival using patient age, HOXD12 gene body methylation level, and CNS WHO grade using TCGA data (Fig. 2h). Unfortunately, immunohistochemistry for HOXD12 was not a sufficiently sensitive surrogate marker to detect a difference between HOXD12 gene body methylation levels in a small set of oligodendrogliomas (N = 10) (Supplemental Fig. 13).

HOXD12-positive expression status and gene body hypermethylation are prognostic independent of key histopathologic, genomic, and radiographic features

Having shown that HOXD12-positive expression status and HOXD12 gene body hypermethylation were markers for poor survival in oligodendroglioma, we interrogated their relationship with clinically relevant histopathologic, genomic, and radiographic features in the TCGA (Fig. 3a). Among histopathologic features, HOXD12 gene body hypermethylation was linked to a high Ki-67 proliferative index (≥ 10%) (p < 0.0001, Fisher’s) and the presence of mitotic figures (p = 0.015, Fisher’s) but not to the presence of necrosis or microvascular proliferation (MVP) (Fig. 3b). Among proposed prognostic genomic alterations, we found that HOXD12-positive expression status and HOXD12 gene body hypermethylation were associated with the loss of chromosome arm 15q (p < 0.001, p = 0.049, Fisher’s) and MYC activation (p = 0.04, p < 0.03, Fisher’s) but not with NOTCH1 mutations or PIK3CA mutations. CDKN2A homozygous deletions were not tested because fewer than 1% (1/171) of TCGA oligodendrogliomas harbored these alterations. Radiographically, HOXD12 gene body hypermethylation was associated with the presence of T1-post contrast enhancement in a set of 27 multimodal pre-operative MRIs from the TCGA (p = 0.03, Fisher’s), while HOXD12-positive expression status was not significantly associated with any tested radiographic feature. These associations suggest that tumors expressing HOXD12 and harboring HOXD12 gene body hypermethylation are not only prognostic, but they also appear more aggressive histopathologically, molecularly, and radiographically. To highlight these features, we identified a representative tumor harboring both HOXD12-positive expression status and HOXD12 gene body hypermethylation that displays mitotic figures, necrosis, and MVP (Fig. 3c), as well as contrast enhancement and substantial peritumoral edema (Fig. 3d).

Fig. 3

HOXD12-Positive Expression and Gene Body Hypermethylation are Independently Prognostic of Key Histopathologic, Genomic, and Radiographic Features. a List of key tested histopathologic, genomic, and radiographic features. b HOXD12 gene body hypermethylation was linked to high Ki-67 proliferative index and the presence of mitotic figures. HOXD12-positive expression status and hypermethylation were associated with the loss of chromosome arm 15q and MYC activation. HOXD12 hypermethylation was associated with the presence of T1-post-contrast enhancement. c Representative H&E-stained sections of a HOXD12-positive and gene body hypermethylated TCGA oligodendroglioma (TCGA-DU-7018) showed increased mitotic activity, necrosis, and MVP. d MRI of the same patient showed strong contrast enhancement (blue) and extensive peritumoral edema (red) on T1-post contrast and FLAIR MRI sequences, respectively. e, f HOXD12-positive expression status and hypermethylation were independently prognostic of all tested histopathologic and genomic variables. g HOXD12-positive expression status was independently prognostic of all tested radiographic features

Remarkably, HOXD12-positive expression status and HOXD12 gene body hypermethylation were independently prognostic of all tested histopathologic and genomic variables (Fig. 3e, f). These multivariate analyses indicated that the statuses of HOXD12 expression and gene body methylation may be better indicators of outcome than histopathologic features used to determine CNS WHO grade (mitotic features, necrosis, MVP) and genomic markers nominated by other studies [2, 3, 30, 38, 74]. Similarly, HOXD12-positive expression status was independently prognostic of all testable radiographic features (total tumor volume omitted due to collinearity). However, HOXD12 gene body hypermethylation was not (p = 0.12), possibly because the MRI dataset was small (Fig. 3g, Supplemental Fig. 14). Regrettably, it was not possible to validate our findings in the CGGA or Capper et al. cohorts because of insufficient imaging or histopathology data.

Importantly, the TCGA survival associations observed here were likely not due to treatment differences because the use of adjuvant chemotherapy (N = 152), predominately temozolomide, and/or adjuvant radiation therapy (RT) (N = 151) were not statistically associated with HOXD12 gene body hypermethylation or HOXD12-positive expression status (Supplemental Fig. 15a, b, Supplemental Table 2). Of note, TCGA WHO grade 2 oligodendrogliomas were treated with postoperative chemoradiation dramatically less frequently than TCGA WHO grade 3 oligodendrogliomas (9% vs. 55%, p < 0.0001, Fisher’s) (Supplemental Fig. 15c). It is conceivable that if HOXD12 gene body methylation status had been incorporated as a molecular risk factor, patients with HOXD12 gene body hypermethylated tumors may have been treated more commonly with chemoradiation, as the case with WHO grade 3 tumors, and the survival differences we observed may have been slimmer.

To further control for confounding treatment differences and to test HOXD12 gene body hypermethylation in a prospective cohort, we also analyzed DNA methylation data from the prospective EORTC 26951 Phase III trial] [41, 77]. EORTC 26951 Phase III compared RT to RT and six cycles of chemotherapy agents procarbazine, CCNU, and vincristine (PCV) in WHO grade 3 diffuse gliomas. For our analysis, we included EORTC patients who met the following three criteria: (1) their tumor met the current WHO CNS classification system definition of oligodendroglioma (including 1p/19q-codeletion), (2) they were treated with the trial’s uniform chemoradiation regimen (RT + PCV), and (3) their tumor had methylation data sufficient to determine HOXD12 gene body methylation status. Unfortunately, most of the 115 tumors with profiled DNA methylation data were excluded because, unlike Illumina 450 k arrays, they were profiled with Illumina 27 k arrays, which lack two of the three HOXD12 gene body probes in our HOXD12 gene body methylation signature. Only 10 tumors met these three selection criteria, of which four were HOXD12 gene body hypomethylated and six were HOXD12 gene body hypermethylated. Our analysis of these data, although severely underpowered, did show a weak trend toward worse survival among HOXD12 gene body hypermethylated patients in both overall survival (HR = 2.8, p = 0.36) and progression-free survival (HR = 2.8, p = 0.36) (Supplemental Fig. 16). Overall, more extensive prospective follow-up data and more feasible clinical testing are necessary to definitively establish any prognostic utility of HOXD12-positive expression status or HOXD12 gene body methylation status.

Increased HOXD12 activity in neoplastic cells is associated with a proliferative phenotype

To determine HOXD12 activity in specific cell compartments and associated cell states, we analyzed high-quality paired oligodendroglioma single-nucleus ATAC sequencing (snATAC-seq) and single-nucleus RNA sequencing (snRNA-seq) data published by Blanco-Carmona et al. (N = 7729) and snATAC-seq data published by Wang et al. (N = 10,281), from which we estimated snRNA-seq reads using Signac’s GeneActivity function [10, 73, 82]. To identify neoplastic nuclei, we used inferCNV to compute the somatic copy number alteration status of chromosome arms 1p and 19q to determine whether they were codeleted, a unique feature of oligodendroglioma cells; we used published gene markers to assign cell types to non-neoplastic nuclei (Fig. 4a, Supplemental Figs. 17, 18) [48, 53, 75, 81, 84]. Neoplastic cell states were assigned using lineage scores for oligodendrocyte progenitor cell (OPC)-like, Astro-like, Cycling, and ribosomal-enriched (RE) cell states for each nucleus, which we computed using published gene sets [10, 81]. Using a previously described permutation test, we calculated and corrected p-values which we used for final cell state assignment [10]. All RE cells were excluded due to a low sample count. Stemness scores and lineage scores for OC-like and AC-like cell states were also computed as previously described [76].

Fig. 4

HOXD12 activity is associated with neoplastic nuclei, proliferating cells, and an increased stem-like phenotype. a UMAP embeddings of single-nuclei RNA-seq and single-nuclei ATAC-derived single-nuclei RNA-seq from the Blanco-Carmona et al. and Wang et al. cohorts, respectively, labeled by neoplastic cell state or non-neoplastic cell type. Neoplastic nuclei are labeled as Gradient, Astro-like, OPC-like, or Cycling. b In both the Blanco-Carmona et al. and Wang et al. cohorts, single-nucleus HOXD12 RNA reads and single-nucleus ATAC HOXD12 reads were more prevalent in neoplastic tissue compared to non-neoplastic tissue. c, d Among neoplastic nuclei, HOXD12 RNA and ATAC reads were most common in OPC-like and cycling cells, which are associated with stem-like and proliferative phenotypes. e Lineage and stemness scores for oligodendroglioma tumors nuclei in the Blanco-Carmona et al. and Wang et al. cohorts. f Nucleus stemness scores are higher in tumor nuclei that have HOXD12 RNA or ATAC reads than those without in both the Blanco-Carmona et al. and Wang et al.

We found that HOXD12 expression was significantly more common in cycling/proliferative neoplastic cells than in non-proliferative neoplastic cells, which, in turn, had a higher rate of HOXD12 expression than non-neoplastic brain cells. In general, HOXD12 expression and chromatin accessibility were significantly more prevalent in neoplastic nuclei compared to non-neoplastic nuclei in the Blanco-Carmona et al. and Wang et al. datasets (Fig. 4b). After cycling/proliferative neoplastic cells, HOXD12 snRNA-seq expression was most prevalent in OPC-like neoplastic cells in both datasets, followed by gradient and astro-like neoplastic cells (Fig. 4c). Likewise, snATAC-seq confirmed that neoplastic cells classified as cycling or OPC-like had the highest prevalence of HOXD12 chromatin accessibility followed by cells classified as gradient or astro-like (Fig. 4d). Given that cycling cells are in the process of dividing and that OPC-like cells are known for their proliferative potential and stem-like properties, we computed stemness scores for all neoplastic cells. For both snRNA-seq and snATAC-seq data, and in both the Blanco-Carmona et al. and Wang et al. datasets, nuclei with HOXD12 activity had significantly higher stemness scores (Student’s t-test) (Fig. 4e,f). Altogether, these single-nuclei data suggested that HOXD12 activity is associated with a proliferative phenotype in neoplastic cells and indicated that our previous analyses of bulk sequencing data reflect biological processes within the neoplastic cell compartment, rather than the non-neoplastic cells of the tumor microenvironment.

A Pan-HOX methylation signature, driven by HOXD12, is associated with increased proliferation and poor outcomes

To place our HOXD12-centric findings in the context of all HOX genes, we analyzed the HOX gene body DNA methylation probes present in all 365 oligodendroglioma tumors from the TCGA (N = 171), Capper et al. (N = 170), and GLASS (N = 24) datasets. Two distinct, consistent clusters, not related to dataset membership, appeared across 1000 UMAP projections of these data generated using different initializations (Fig. 5a, Supplemental Fig. 19). We restricted our analysis to samples that remained in the same cluster in at least 90% of the UMAP projections (N = 345) (Supplemental Fig. 20). These two clusters were characterized by global HOX gene body DNA methylation levels: nearly all HOX-associated gene body probes had higher beta values in one cluster (HOX-high) compared to the other (HOX-low) (Fig. 5b, c). Patients in the HOX-high group were significantly older in the TCGA (p = 3e−15, Mann–Whitney U) and Capper et al. (p = 1e−8, Mann–Whitney U) datasets, and they disproportionally harbored higher CNS WHO grade (p = 2e−5, p = 0.0004, Fisher’s) (Fig. 5d, Supplemental Fig. 21). TCGA patients in the HOX-high group harbored more aggressive tumors, measured both by clinical outcomes and histological features. HOX-high TCGA patients suffered worse outcomes than those in the HOX-low group (HR = 7.6, p = 0.01). HOX-high TCGA tumors also more commonly had increased mitotic figures (p = 0.01, Fisher’s) and higher Ki-67 proliferative index (p < 0.01, Mann–Whitney U) (Fig. 5e, f). The presence of palisading necrosis and microvascular proliferation (MVP) were also elevated in HOX-high TCGA patients; however, their difference was less significant. Although the sample size of stable samples in the GLASS dataset was small (N = 9), we observed more movement from the HOX-low cluster to the HOX-high cluster as tumors evolved than vice versa (Supplemental Fig. 22a–c). The scarcity of these data made conclusions impossible to draw with confidence; however, GLASS samples in the HOX-high cluster tended to be recurrent, higher WHO grade, and shorter-lived (Supplemental Fig. 22d–f). As was the case with HOXD12 gene body hypermethylation and HOXD12-positive expression status, HOX-high survival associations were not linked to lower rates of adjuvant therapy in the TCGA (Supplemental Fig. 23). In fact, the only significant associations we found were higher rates of adjuvant chemotherapy (p = 0.01, Fisher’s) and chemoradiation (p < 0.05) in HOX-high patients.

Fig. 5

A Pan-HOX signature led by HOXD12 is associated with poor outcomes and proliferation. a A UMAP embedding of all HOX gene body associated DNA methylation probes for oligodendroglioma patients in the TCGA, Capper et al., and GLASS datasets (N = 365) formed two clear clusters termed HOX-high and HOX-low. b The mean value of HOX gene body probes is significantly higher in the HOX-high cluster than in the HOX-low cluster. c The mean value of 98.4% of HOX gene body probes is higher in the HOX-high cluster than in the HOX-low cluster. d The HOX-high cluster is associated with patient age and tumor WHO grade in the cohorts of the TCGA and Capper et al. Differences in TCGA and Capper et al. age distributions are likely attributable to the abundance of young, WHO grade 3 oligodendrogliomas in the Capper et al. cohort. e The TCGA patients in the HOX-high cluster harbor significantly worse outcomes than those in the HOX-low cluster. f TCGA tumors in the HOX-high cluster more commonly had mitotic figures and high Ki67 index; the presence of palisading necrosis and MVP were also elevated. g In a differential gene expression analysis between the two methylation classes, HOXD12 had the highest RNA expression log fold change. h TCGA patients in the HOX-high cluster were associated with the mitotic cell cycle pathway in a gene ontology analysis. i In a differential methylation analysis, gene body probes on the HOXD locus had higher log fold change on average compared to gene body probes from other HOX loci. j In this differential methylation analysis, gene body probes associated with HOXD12 had the highest average log fold change compared to other HOX genes

Next, we performed unbiased differential gene expression and differential DNA methylation analyses by comparing patients in the HOX-high group to patients in the HOX-low group. Remarkably, HOXD12 was the gene with the highest RNA expression log-fold change between the HOX-high and HOX-low TCGA patients (4.8, padj < 7e−5) (Fig. 5g). Gene ontology analyses of the results from our differential gene expression analysis on TCGA oligodendroglioma tumors showed that the most significantly activated pathways in HOX-high patients were related to mitotic activity and cell cycle (padj < 1e−6) (Fig. 5h). In our differential DNA methylation analysis, gene body probes from the HOXD locus had more significant, higher log-fold change on average compared to gene body probes from other HOX loci (Fig. 5i). Among HOX genes, gene body probes associated with HOXD12 had the highest average log-fold change compared to gene body probes associated with other HOX genes (Fig. 5j, Supplemental Fig. 24). In general, gene body DNA methylation of HOX genes partitioned oligodendroglioma into two groups with lower and higher HOX gene body DNA methylation levels. The group of oligodendroglioma tumors with high HOX gene body DNA methylation levels was associated with poor survival, older patients, and more aggressive histology and was best differentiated from the HOX-low group by genes in the HOXD locus, especially HOXD12.