DNA Methylation-Based Deconvolution Study of Glioblastoma Heterogeneity and Cell Types Associated with Patient Survival

DNA Methylation-Based Deconvolution Study of Glioblastoma Heterogeneity and Cell Types Associated with Patient Survival | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article DNA Methylation-Based Deconvolution Study of Glioblastoma Heterogeneity and Cell Types Associated with Patient Survival Aviel Iluz, Nir Lavi, Hanna Charbit, Mijal Gutreiman, Masha Idelson, and 11 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6743395/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background: IDH-wildtype glioblastoma (GBM) is an aggressive, heterogeneous brain tumor with limited treatment options. This study tries to improve our understanding of GBM by applying DNA methylation-based deconvolution to define its cellular composition and its association with patient outcomes. Methods: We generated oligodendroglial precursor cells at various developmental stages from enriched human neural progenitor cultures and used their DNA methylation signatures, along with published signatures of cells types relevant to brain tumors and the tumor microenvironment, to deconvolve 263 adult GBMs (Heidelberg cohort). Tumor purity was estimated using RF_Purify. An independent cohort of 199 GBMs from TCGA and GEO, all treated with standard-of-care therapy, was similarly deconvolved, followed by Kaplan–Meier survival analysis to assess the prognostic value of the proportions of the neoplastic components. Results: Deconvolution uncovered distinct cellular compositions that differed between the neoplastic and non-neoplastic components of GBM. Tumor purity analysis showed that the neoplastic fractions averaged 70% of the tumor bulk: they were predominantly oligodendrocyte-like (43%), along with oligodendrocyte precursor-like (27%), astrocyte-like (19%), and mesenchymal stem cell-like (11%) populations. The non-neoplastic fractions were enriched for macrophages, vascular cells, and immune cell populations. A higher oligodendrocyte-like signature was linked to poorer survival (median survival 14.3 vs. 15.3 months; p = 0.017), while a higher astrocyte-like signature correlated with improved survival (15.3 vs. 13.4 months; p = 0.044). The astrocyte-to-oligodendrocyte ratio emerged as a strong prognostic marker, with a higher ratio predicting significantly longer survival (15.8 vs. 11.9 months; p < 0.00011). Conclusions: The methylation-based deconvolution data provided insight into GBM heterogeneity, highlighting the prognostic relevance of the astrocyte-to-oligodendrocyte ratio and its potential to guide personalized treatment strategies. Glioblastoma methylation-signature tumor heterogeneity deconvolution tumor microenvironment astrocytes oligodendrocytes glial cells glioblastoma pathology central nervous system tumor Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Introduction IDH-wildtype glioblastoma (GBM) is an aggressive and lethal brain cancer known for its rapid growth and resistance to standard treatments [ 1 , 2 ]. A key feature of GBM is its extensive cellular heterogeneity, which presents challenges for treatment. The tumor contains a complex mix of neoplastic and non-neoplastic cells, including glioma-initiating cells, immune cells, endothelial cells, and glial cells [ 2 – 5 ]. Thus, understanding the cellular drivers of GBM heterogeneity is essential for improving its diagnosis and treatment. Various methods, including proteomics and transcriptomic analyses, have been used to explore the complex cellular dynamics of GBM [ 2 – 4 , 6 ]. Recent breakthroughs in single-cell RNA sequencing (scRNA-seq) have provided new insights into the cellular diversity of GBM, revealing distinct subpopulations of neoplastic cells and their interactions with the tumor microenvironment (TME) [ 2 – 5 ]. A landmark study [ 3 ] identified four distinct cellular states within the neoplastic fraction of GBM: astrocyte-like, oligodendrocyte progenitor cell-like, mesenchymal-like, and neural progenitor cell-like. These findings underscore the tumor’s complex differentiation patterns and plasticity. However, while scRNA-seq has been instrumental in resolving such intratumoral heterogeneity, it is not without limitations. The technique is resource-intensive, involving high cost, extended processing time, and often limited scalability, which constrain its broader application, particularly in clinical or large-cohort studies. Furthermore, scRNA-seq is susceptible to transcriptional noise, dropout effects, and dissociation-induced artifacts, all of which can compromise the accurate representation of in vivo cellular states. DNA methylation signature-based deconvolution has emerged as a promising alternative that can overcome the challenges with scRNA-seq. This approach leverages the relative stability and cell-type specificity of DNA methylation patterns to infer the cellular composition of bulk tumor samples. Unlike scRNA-seq, it is cost-effective and scalable, and can be applied to archival formalin-fixed paraffin-embedded material, making it especially suitable for large-scale or retrospective studies. By bypassing the need for single-cell dissociation and transcript capture, DNA methylation deconvolution provides a robust and clinically viable means of profiling tumor heterogeneity, offering new insights into the cellular architecture of GBM [ 7 ]. Genome-wide DNA methylation profiling has become a transformative tool for classifying central nervous system (CNS) tumors, offering several advantages over traditional diagnostics in terms of patient stratification [ 8 , 9 ]. Each cell type has a unique methylation signature that reflects its tissue identity [ 10 ], and it is well-established that tumors exhibit stable methylation signatures, even after oncogenic events [ 11 ]. Accordingly, the DNA methylation signature of a tumor reflects the cumulative methylation patterns of all its constituent cell types, including both normal and neoplastic cells. This composite profile captures the cellular admixture within the tumor, providing insights into its complex cellular makeup [ 7 ]. These signatures have been validated in thousands of CNS tumors, facilitating accurate sub-classification and diagnoses and better treatment guidance [ 8 , 12 – 14 ], and this method was incorporated into the 2021 WHO Classification of CNS tumors [ 14 ]. In the context of GBM, beyond confirming the diagnosis, DNA methylation profiling has helped identify distinct subclasses, including receptor tyrosine kinase I (RTK1), receptor tyrosine kinase II (RTK2), and mesenchymal[ 8 , 9 , 15 ]. Studies of newly diagnosed GBMs reveal significant spatial diversity within methylation subclasses, underscoring both the complexity of tumor classification and the critical role of intratumoral heterogeneity in driving treatment resistance and disease recurrence [ 16 ]. Methylation-based deconvolution involves the use of a reference atlas of normal cell-type methylation signatures to estimate tissue composition and has been applied to analyze tissue contributions based on circulating cell-free DNA [ 17 , 18 ]. A similar deconvolution methodology has been applied to investigate the composition of non-neoplastic immune cells within the GBM TME [ 7 ]. However, to the best of our knowledge, methylation signatures have not been specifically used to investigate the neoplastic components of GBM. In order to address this gap, in the present study, we have adopted a comprehensive approach that uses methylation-based deconvolution and methylation-based purity analysis to analyze both the neoplastic and non-neoplastic components of GBM tumors. By analyzing methylation signatures from 462 GBM samples across three datasets, we aim to elucidate GBM heterogeneity encompassing all relevant cell types within the TME, with particular focus on the neoplastic fraction, and to investigate associations between the abundance of specific cell types and patient survival. This analysis provides valuable insights into the cell of origin of GBM, a critical factor in understanding early tumorigenic processes and identifying potential targets for early intervention. Since DNA methylation profiles preserve the signatures of the originating cell type, they offer a reliable method for tracing cellular origins. When combined with our uniquely developed in vitro-derived human neural progenitors and oligodendrocyte-lineage precursor methylation dataset, this approach allows for a comprehensive exploration of the cell of origin in GBM [ 19 ]. Finally, to validate and contextualize our findings, we compared our results with data obtained through complementary methodologies, including scRNA-seq and the deconvolution of immune cell profiles from bulk glioma DNA methylation data [ 3 , 7 , 20 – 22 ]. With this integrative approach, we believe that we have provided an enhanced understanding of GBM's cellular complexity that can inform future prognostic and therapeutic strategies. Materials and Methods Generation of Human Neural Progenitors and Oligodendrocyte Lineage Precursor Cells Highly enriched cultures of neural progenitors (NPs) were derived from human embryonic stem cells (hESCs) using an established protocol [ 23 ]. Briefly, hESCs (HAD-C100 [ 24 ]) at passages 22 to 28 were cultured on human recombinant Laminin 521 (Biolamina, Sundbyberg, Sweden) in NutriStem medium supplemented with growth factors (Biological Industries, Beit Haemek, Israel) and 50 units/mL penicillin and 50 µg/mL streptomycin (Pen-Strep; Thermo Fisher Scientific, Waltham, MA, USA). Oligodendroglial precursor cells at various developmental stages were derived from these enriched NPs cultures as previously described [ 23 ]. Briefly, NPs were first induced to differentiate into OLIG2-positive precursor cells (henceforth referred to as OLIG2 cells) and subsequently induced to differentiate into oligodendroglial lineage cells. This process resulted in an enriched population of Olig2+/Nkx2.2 + oligodendrocyte progenitor cells (henceforth referred to as OPCs), which were further differentiated into O4 + pre-oligodendrocytes (henceforth referred to as pre-oligodendrocytes). Immunofluorescent Staining of hESC-Derived Oligodendroglial Lineage Cells to Detect Different Stages of Development To characterize the developmental stages of oligodendroglial lineage cells, markers representing different maturation steps were detected by immunofluorescence. Briefly, cells were plated on glass coverslips pretreated with poly- d -lysine (30–70 kDa) at a concentration of 10 mg/ml and laminin (concentration, 4 mg/ml; both from Sigma, St. Louis, MO) and cultured for 3–5 days. The cells were then fixed with 4% paraformaldehyde and incubated with mouse monoclonal anti-Olig2 (1:150; EMD Millipore, Burlington, MA) and rabbit anti-Nkx2.2 (1:50; Novus Biologicals, Centennial, CO) antibodies. Alexa Fluor 555-conjugated donkey anti-mouse antibody (1:150; Invitrogen/Thermo Fisher Scientific, Waltham, MA) and Alexa Fluor 488-conjugated donkey anti-rabbit antibody (1:150; Jackson ImmunoResearch Laboratories, West Grove, PA) were used for detection of the markers. For detection of the pre-oligodendrocyte stage, the cells were incubated with mouse anti-O4 antibody (1:150; R&D Systems, Minneapolis, MN), fixed with 4% paraformaldehyde, and incubated with Alexa Fluor 488-conjugated donkey anti-mouse antibody (1:150; Jackson ImmunoResearch). Nuclei were counterstained with 4′,6-diamidino-2-phenylindole (Vector Laboratories, Burlingame, CA). Specimens were visualized using an Olympus BX61 fluorescence microscope (Olympus, Hamburg, Germany). Flow Cytometry Analysis Human ESC-derived neural progenitor cells (NPCs) were immunostained with mouse anti-PSA-NCAM-PE (1:50) and mouse anti-A2B5-APC (1:50) antibodies, or the corresponding isotype controls (all from Miltenyi Biotec, Waltham, MA). These markers are used to differentiate between distinct developmental stages and lineage potentials within NPC populations. Specifically, A2B5 is a marker of glial-restricted progenitors and early oligodendrocyte precursor cells, while PSA-NCAM is a marker of neuronal-committed progenitors and early post-mitotic neurons. By analyzing the expression profiles of these markers, the experiment distinguished between A2B5⁺/PSA-NCAM⁻ cells, which represent glial-biased progenitors; PSA-NCAM⁺/A2B5⁻ cells, enriched for neuronal lineage commitment; double-positive or double-negative subsets, which may reflect bipotent progenitors, intermediate states, or uncommitted populations. For each sample, at least 10 4 cells were analyzed on a CytoFLEX flow cytometer (Beckman Coulter, Indianapolis, IN). DNA Methylation Profiling Genome-wide DNA methylation profiling of NPs, OLIG2 cells, OPCs, and pre-oligodendrocytes was performed using the Illumina EPIC array, with data generated at the Genomics and Proteomics Core Facility of the German Cancer Research Center (DKFZ) in Heidelberg, Germany. Data processing, filtering, and normalization were carried out using the minfi R package (v1.34.0). Data Collection We collected 450K or EPIC Illumina cytosine-guanine dinucleotides (CpGs) methylation signature data to assemble the various sample sets analyzed in this study, as detailed below. All data were processed as beta-value CSV files. To ensure consistency, EPIC beta-value probe sets were overlapped with the 450K beta-value probe set, and only common probes were retained for downstream analysis. GBM datasets for deconvolution analysis We selected the DNA methylation profiles of GBM samples classified as "Glioblastoma, IDH-wildtype" according to the WHO 2016 classification of CNS tumors [ 13 ], from the Heidelberg reference cohort (GSE90496). To focus on adult GBM, we included 263 samples further sub-classified into the receptor tyrosine kinase I (RTK1; n = 64), receptor tyrosine kinase II (RTK2; n = 143), or mesenchymal (MES; n = 56) subtypes, as defined by Capper et al [ 8 ].This classification was originally derived using the DNA methylation-based brain tumor classifier (version 11.4), which is known to improve diagnostic accuracy and standardization [ 8 ]. Glioma-initiating cell datasets Methylation data from a glioma-initiating cell (GIC) sample set (n = 20) were obtained from a publicly available dataset referenced in Vinel et al [ 25 ]. (GSE155985). The researchers isolated GICs from GBM samples, which were classified by methylation profiling into the RTK1, RTK2, and MES subtypes. GBM datasets for survival analysis and validation : We obtained methylation data of an independent cohort of 199 GBM cases with available clinical data from multiple sources: (1) a cohort of 84 TCGA samples diagnosed as primary GBM [ 15 ], and (2) a cohort of 115 GBM samples from two GEO datasets (53 from GSE60274 [ 26 ], 62 from GSE195640 [ 27 ]). We selected only patients who were treated according to the Stupp protocol, which consists of radiation therapy with concurrent and adjuvant temozolomide. All the 199 GBM cases were successfully subclassified into the RTK1 (n = 39), RTK2 (n = 97), and MES (n = 63) subclasses using the DKFZ brain tumor classifier (v12.8) on the Heidelberg Epignostix Classifier platform ( https://app.epignostix.com ). Non-glial tumor datasets : Methylation profiles of non-glial tumor sets were obtained from the TCGA database and represent the average methylation profiles of samples from four tumor types: bladder urothelial carcinoma (BLCA) (n = 292), breast carcinoma (n = 721), kidney renal papillary cell carcinoma (KIRP) (n = 210), and prostate adenocarcinoma [ 15 ] (n = 127). Normal cell datasets Methylation data (CpG beta-value CSV files) for various normal cell types, including 25 signatures from Moss et al. [ 18 ], microglia [ 7 ], mesenchymal stem cells (MSCs) (GSM4077441–GSM4077443, GSM4078810–GSM4078818), astrocytes (GSM3938231), oligodendrocytes [ 17 , 18 ], and cortical neurons (GSE98203) were sourced from multiple databases and studies. Construction of GBM Methylation Profile Atlases Constructing the reference atlases We constructed reference atlases by assembling the methylation profiles of the relevant cell types for each atlas. To identify tissue-specific CpG sites within each atlas, we applied the feature selection method described by Moss et al [ 18 ]. Briefly, CpGs (based on hg19 coordinates) with low variance (< 0.1%) across the methylation atlas or missing values were excluded. Methylation values for each CpG across cell types were normalized by their sum, and the top 100 hypermethylated CpGs per cell type were selected based on specificity. A similar procedure was applied to the reversed methylation matrix to identify hypomethylated CpGs. For each cell type, both the top 100 hypermethylated and hypomethylated CpGs, along with neighboring CpGs within 50 bp, were included in the reference matrix. To further refine the feature set, pairwise-specific CpGs were iteratively selected by projecting the atlas onto the current CpG set, calculating Euclidean distances between cell types, and adding CpGs that best distinguished the most similar pairs at each iteration. Refining the reference atlases : We used two reference atlases for GBM deconvolution analysis. The initial reference atlas included 16 cell type signatures: B-cells, CD4 T-cells, CD8 T-cells, NK cells, neutrophils, vascular endothelial cells, monocytes, microglia, MSCs, cortical neurons, astrocytes, oligodendrocytes, NPs, OLIG2 cells, OPCs, and pre-oligodendrocytes, covering a total of 4,712 CpG sites. Based on the results from deconvolving the Heidelberg set of 263 GBM samples, we refined this atlas to include 14 cell type signatures by merging OLIG2 cells, OPCs, and pre-oligodendrocytes into a single oligodendrocyte-lineage precursor component (designated as OPs), thus reducing the total to 4,111 CpG sites. This refined atlas was subsequently used to deconvolve the Heidelberg GBM dataset, the TCGA and GEO GBM datasets, and the GIC datasets. Deconvolution of Methylation Signatures Using Non-Negative Least Squares Linear Regression Non-negative least squares linear regression was employed, as described by Moss et al. [ 18 ], to deconvolve the GBM methylation signature beta-value matrices into cell-type components, using the relevant reference atlas. Validation of the Deconvolution Method To validate the ability of the methylation-based deconvolution method to correctly identify expected cell type signatures, we applied it to methylation data from non-GBM sources using a custom reference atlas we constructed, comprising cell type signatures relevant to the tested samples, as follows: Deconvolution of non-glial tumors : We applied methylation-based deconvolution to non-glial tumors, including BLCA, breast carcinoma, KIRP, and PRAD. The reference atlas used for this analysis comprised 25 cell type signatures, sourced from publicly available data by Moss et al [ 18 ]. and spanning 7,390 CpG sites. The results showed that the predominant cell type in each tumor corresponded to the expected tissue of origin: bladder cells (74%) in BLCA, breast cells (62%) in breast cancer, kidney cells (79%) in KIRP, and prostate cells (90%) in PRAD (Figure S1 ). Deconvolution of normal cell types : We applied methylation-based deconvolution to 11 samples representing various normal cell types. The reference atlas used in this analysis included 11 cell type signatures derived from publicly available data deposited by Moss et al.[ 18 ], using an N-1 approach: for each cell type, one corresponding sample was excluded from the atlas for validation and subsequently deconvolved using the remaining data. The atlas comprised a total of 3,274 CpG sites. The analysis showed that over 90% of the estimated cell type proportions in each sample matched the expected normal cell type. For example, the cortical neuron sample was estimated to contain 100% cortical neurons (Figure S2 ). Assessment of Tumor Purity Tumor purity was calculated using the RF_purify R package (v0.1.2) with the ABSOLUTE method. This method analyzes Illumina methylation data to estimate purity by applying a Random Forest machine learning algorithm trained on copy number variations inferred from methylation patterns [ 28 ]. Survival Analysis Kaplan–Meier analysis and log-rank tests were conducted on 199 GBM samples using the survminer (v0.5.0) and survival (v3.7-0) R packages [ 29 ]. We assessed differences in overall survival based on the median proportion of each cell type across all samples. Patients with proportions greater than the median were classified as "high," while those with proportions less than the median were classified as "low," and their survival outcomes were compared. Astrocytes and oligodendrocytes showed significant overall differences in terms of survival, with opposite survival trends: higher proportions of astrocytes were associated with higher overall survival, while higher proportions of oligodendrocytes were associated with lower overall survival. To assess the prognostic significance of the ratio of astrocyte-like and oligodendrocyte-like signatures within the neoplastic fraction, we determined the optimal cutoff for the astrocyte-to-oligodendrocyte ratio. The optimal cutoff was established by evaluating the ratio values across the entire range, with the threshold identified as the point yielding the most significant log-rank test result. Based on this cutoff, the groups were dichotomized into those with “high” and “low” astrocyte-to-oligodendrocyte ratios, with the groups exhibiting distinct survival probabilities that were consistent with previous reports using different methodologies [ 30 ]. To validate this cutoff and address potential overfitting, we applied a permutation-based approach. That is, the astrocyte-to-oligodendrocyte ratio was randomly shuffled across patients (100,000 iterations) while maintaining survival times and censoring status, and log-rank p-values were recomputed for each permutation. The permutation p-value was calculated as the proportion of permuted p-values that were less than the observed p-value, with statistical significance defined at p < 0.05. A low permutation p-value indicates that the observed p-value is unlikely under randomized conditions, reflecting the robustness of the selected cutoff. This approach confirms the cutoff’s reliability and significance and excludes any effects attributable to chance or dataset-specific overfitting. Statistical Analysis To assess significant differences in cell type proportions across GBM subclasses, pairwise Student’s two-tailed t -tests were conducted for each subclass pair. Two-tailed t -tests were also used to evaluate the statistical significance of Pearson correlation coefficients for correlations between cell type proportions and purity scores. The t -tests were performed using R, Python, or Microsoft Excel with built-in t -test functions. For survival analysis, the log-rank test was applied as previously described. To compare the expected cell-type proportions between the results of survival analysis of the GBM set (n = 199) and the observed results from the initial GBM set (n = 263), we performed a chi-square goodness-of-fit test using the chisq.test function in R (with the following parameters: simulate.p.value = TRUE, B = 2000). For comparisons of subclass distributions between patient groups classified according to survival probability, we used the chi-square test for independence with the chisq.test function in R. A significance threshold of 0.05 was used for all statistical tests. Graph and Plots All graphs and plots were generated using R (v4.0.3+), Python 3.11, Microsoft Excel, or BioeRender.com. Results Generation of NPs and Human Oligodendrocyte Lineage Precursor Cells Through immunostaining studies, we were able to confirm that NPs were successfully derived from hESCs and subsequently differentiated into oligodendrocyte lineage precursor cells through a staged process (Fig. 1 ). These NPs (Fig. 1 A) then differentiated into OLIG2 cells (Fig. 1 B), which further differentiated into cells of the oligodendroglial lineage, which includes enriched populations of OPCs (Fig. 1 C) that subsequently differentiated into pre-oligodendrocytes (Fig. 1 D). These marker studies confirm the progression of the initial NPs along the expected developmental trajectory. Identification of Cell Type Composition and Proportions in GBM and Its Subclasses Reference atlases of the cellular methylation profiles : To investigate the cell type composition in GBM, we constructed a reference atlas of DNA methylation signatures critical for the deconvolution of GBM samples and quantification of cell type proportions. This atlas encompassed methylation data from 16 distinct cell types derived from our previously published data [ 17 , 18 ] and several publicly available datasets: cortical neurons, astrocytes, oligodendrocytes, various immune cells (B cells, CD4⁺ T cells, NK cells, CD8⁺ T cells, and neutrophils), tumor-associated macrophages (TAMs: monocytes and microglia [ 7 ]), and MSCs. To account for the potential differentiation states of cancer cells, we further extended the atlas by incorporating methylation profiles from our in vitro-generated NPs and three defined stages within the oligodendrocyte lineage. These stages represent a continuum of differentiation and include OLIG2-positive progenitors (characterized by expression of the OLIG2 transcription factor essential for oligodendrocyte development), OPCs, and pre-oligodendrocytes identified by O4 expression. This atlas was generated by selecting distinct differentially methylated CpGs for each cell type, following previously described and validated methodologies [ 18 ]. In total, it comprises 4,712 CpG sites across these cell types ( Figure S3 ). Overall cellular composition of GBM samples To estimate the cellular composition of bulk GBM tumor samples, we applied a deconvolution method adapted from cell-free DNA tissue-origin studies [ 18 ]. Using 450K methylation data, we analyzed 263 adult GBM samples, classified according to their methylation profiles as previously described [ 8 ]. In the GBM samples, glial cells formed the majority (62%) of the cellular component, including contributions from oligodendrocyte-like (26.3%), astrocyte-like (13.4%), and pre-oligodendrocyte-like (O4+, 22%) components. The cortical neuron signature was minimally represented (2.6%). Additionally, GBM displayed 7.3% MSC-like cells, 7.8% TAMs comprising microglia (3.6%) and monocytes (4.2%), 13% vascular endothelial cells, and 7.2% immune cells (B-cells, CD4 T-cells, NK cells, CD8 T-cells, and neutrophils). NPs effectively showed no contribution ( Figure S4 ). To further refine the analysis of GBM cell type composition, OLIG2 cells, OPCs, and pre-oligodendrocytes were merged into a single component termed OPs. This adjustment was necessary because OLIG2 cells and OPCs contributed minimally to the GBM cell populations (average, 0.3 ± 1.2%). Thus, their integration ensured representation while preserving the accuracy of cell type contribution. The final reference atlas included 14 cell types (Fig. 2 ). This adjustment had minimal impact on the estimated proportions of astrocytes, cortical neurons, MSCs, TAMs, vascular endothelial cells, and immune cells. However, a slight difference emerged in the oligodendrocyte lineage, with the proportion of oligodendrocytes increasing from 26.3–29.5%, while the proportion of pre-oligodendrocytes, which accounted for 22% in the separated analysis, was reduced to 18.9% in the combined OPs component (Figs. 3 A & S4). Comparison of cellular composition between GBM subclasses : Our findings revealed significant variation in cell type proportions among GBM subclasses. Specifically, the astrocyte proportions were significantly lower in the MES (10.5%) and RTK1 (12.8%) subtypes than in RTK2 subtype (12.6%) (RTK1 vs. RTK2: p < 0.01; MES vs. RTK2: p < 1.60E-07), but were not significantly different between the RTK1 and MES subtypes (Fig. 3 D). Oligodendrocyte proportions also showed significant pairwise differences in all comparisons: RTK1 vs. RTK2: 39% vs. 27% (p < 1.012E-32), RTK1 vs. MES: 39% vs. 24.6% (p < 2.39E-20), and RTK2 vs. MES: 27% vs. 24.6%, p < 1.85E-03) (Fig. 3 B ) . Additionally, significant differences were observed in the proportions of other cell types, including OPs, MSCs, immune cells, and TAMs. All p-values were calculated using a two-tailed t -test ( Figs. 3 B–J ) . Assessment of the Neoplastic and Non-Neoplastic Components of GBM Using RF_Purify The neoplastic fraction, estimated from the 263 GBM samples using methylation-based deconvolution, was 70% ± 11%. Using the cell type proportions determined from the deconvolution analysis, we aimed to distinguish between the neoplastic and non-neoplastic components in GBM tumors. This was done using the RF_Purify method, which applies the ABSOLUTE score to infer tumor purity [ 31 ]. We correlated purity scores with the proportions of seven predominant cell types identified in the deconvolution analysis, excluding cortical neurons (average, 2%) and NPs (average, 0%) due to their minimal contribution. The analyzed components included immune cells, TAMs, vascular endothelial cells, MSCs, astrocytes, OPs, and oligodendrocytes (Fig. 4 ). Significant negative correlations were observed between tumor purity and the proportions of immune cells (R = -0.60, p < 1.E-26), TAMs (R = -0.76, p < 3.E-51), and vascular endothelial cells (R = -0.20, p < 1.E-03) (Figs. 4 A, 4 E). In contrast, tumor purity was correlated positively with three glial lineage cells, that is, astrocytes (R = 0.33, p < 6.E-08), OPs (R = 0.68, p < 1.E-37), and oligodendrocytes (R = 0.41, p < 6.E-12), and with MSCs (R = 0.22, p < 4.E-04) (Figs. 4 B, 4 E). When the proportions of all four positively correlated components (three glial lineage cell types and MSCs) were combined additively as a single component, the positive correlation with tumor purity was stronger (R = 0.79, p < 2.E-56). This implies that these components collectively contribute substantially to the neoplastic portion of the tumor (Figs. 4 D, 4 E). On the other hand, when the proportions of all the negatively correlated components (immune cells, TAMs, and vascular endothelial cells) were combined additively as a single component, the negative correlation with tumor purity was stronger (R = -0.81, p < 3.E-62). Thus, these components collectively contribute substantially to the non-neoplastic normal portion of the tumor (Figs. 4 C, 4 E). All p-values were calculated using a two-tailed correlation t -test. Validation of Purity Analysis: Comparison of Deconvolution Results in Bulk GBM Tumors Versus GICs The three glial cell types and MSCs identified in the neoplastic fraction of GBM were validated by deconvolution analysis of the DNA methylation profiles of CD133 (Prominin-1)-enriched GICs, which represent the neoplastic compartment of the tumor. These profiles were obtained from a previously published study [ 25 ]. and derived from GBM tumors classified as RTK1, RTK2, and MES (see Methods). This comparison allowed us to evaluate whether the cell-type signals associated with the neoplastic fraction in bulk tumors are retained in a highly purified population of tumor cells. As expected, deconvolution of GICs revealed that they comprised a significantly higher tumor fraction composed of the three glial lineage cell types and MSCs compared to GBM bulk tumors, with 88% in GICs versus 70% in GBM (p < 2.2E-20, two-tailed t -test, Figure S5 ). Specifically, GICs exhibited significantly increased proportions of astrocytes (19% vs. 13% in GBM samples, p < 8.7E-04), OPs (26% vs. 19% in GBM, p < 3.1E-05), and MSCs (16% vs. 8% in GBM, p < 5.6E-11), while the oligodendrocyte proportions remained similar between GICs and GBM (28% vs. 30%, p < 0.21). In contrast, the proportions of vascular endothelial cells, immune cells, and TAMs were significantly lower in GICs than in GBM bulk tumors (vascular endothelial cells: 7% vs. 14%, p < 2.4E-09; immune cells: 3% vs. 7%, p < 1.1E-07; TAMs: 0% vs. 8%, p < 6.5E-54) ( Figure S5 ). All p values were calculated using a two-tailed t -test. These findings demonstrate that cell types positively correlated with tumor purity in GBM (Fig. 4 B) are more abundant in GICs, whereas those negatively correlated with tumor purity (Fig. 4 A) are present at lower proportions. Correlation Between Cell Type Proportions and Patient Survival: Prognostic Significance of the Astrocyte-to-Oligodendrocyte Ratio Overall trends in associations between cell types and survival probability To evaluate whether the abundance of specific cell types is associated with patient survival, we analyzed 199 GBM samples with available methylation clinical data from the TCGA and GEO databases. Using our methylation-based deconvolution approach, we estimated cell type proportions in this cohort and confirmed consistency with the original set of 263 samples (chi-square goodness-of-fit test, p = 0.96; Figure S6). We then assessed the association between cell type proportions and overall survival using Kaplan–Meier analysis (Fig. 5 ). We found that oligodendrocyte proportions lower than the median (26.4%) were significantly associated with higher overall survival probability (OSP), whereas higher proportions were linked to lower OSP (median survival time [MST]: 15.3 vs. 14.3 months; median survival time difference [MSTD]: 1 month; p < 0.018, log-rank test; Fig. 5 A). In contrast, higher astrocyte proportions (above the median of 10.6%) were associated with improved OSP, while lower proportions were correlated with shorter OSP (MST: 15.3 vs. 13.4 months; MSTD: 1.9 months; p < 0.044, log-rank test; Fig. 5 B ). A similar trend was observed for microglia, where proportions above the median (3.2%) were correlated with increased OSP compared to lower proportions (MST: 15.7 vs. 14.3 months; MSTD: 1.4 months; p < 0.007, log-rank test; Figure S7B ). No significant correlations were found for other cell types or cell type groups (immune cells, TAMs, vascular endothelial cells, OPs, and MSCs) ( Figure S7A-F ). Prognostic significance of the astrocyte-to-oligodendrocyte ratio Based on the contrasting effects of astrocyte and oligodendrocyte abundance on OSP, and considering the reports of previous studies that highlight dynamic shifts in cellular states within the GBM neoplastic fraction [ 2 – 4 , 32 ], we focused on the impact of the astrocyte-to-oligodendrocyte ratio on survival. The median ratio was used to effectively stratify patients into high- and low-survival groups. A ratio above the median (> 0.41) was significantly associated with increased OSP (MST = 15.7 months), whereas a lower ratio corresponded to reduced OSP (MST = 14.3 months), with an MSTD of 1.44 months (p < 0.03, log-rank test, Fig. 5 C). To evaluate the astrocyte/oligodendrocyte ratio as a prognostic biomarker, we determined the optimal cutoff for classifying the survival groups (see Methods). A cutoff of 0.3, corresponding to a ratio of 1:3.3, was determined. Patients with a ratio above this threshold exhibited significantly improved OSP (MST = 15.8 months), whereas those with a lower ratio had reduced OSP (MST = 11.9 months), with an MSTD of 3.9 months (p < 0.00011, log-rank test, Fig. 5 D). Notably, the average overall survival differed by 6 months between the “high” and “low” ratio groups. The higher ratio group had an average overall survival of 18.1 months, while the lower ratio group had an average overall survival of 11.6 months. To validate this cutoff and mitigate potential overfitting, we applied a permutation-based approach with 100,000 random iterations of the ratio data (see Methods). The permutation p-value was 0.0001, which indicates the proportion of permuted p-values ≤ the observed p-value (0.00011). This validation underscores the robustness of the cutoff and excludes random effects or dataset-specific overfitting. Given the impact of the astro/oligo ratio on overall survival and the differences in astrocyte and oligodendrocyte proportions observed across the three GBM subclasses, RTK1, RTK2, and MES (Fig. 3 ), we examined their association with survival outcomes to determine whether the observed survival differences were driven by a specific subclass. We found that patients with an astrocyte/oligodendrocyte ratio above the defined cutoff, which was associated with long-term survival, were predominantly from the RTK2 subclass (64%), whereas only 7% belonged to the RTK1 subclass. In contrast, patients with a ratio below the cutoff, linked to short-term survival, had a higher proportion of the RTK1 subclass (43%) than the RTK2 subclass (20%). Unlike the RTK1 and RTK2 subclasses, which exhibited opposite distributions between the long-term survival and short-term survival groups, the MES subclass did not follow this pattern ( Figure S8 ). However, a survival analysis based solely on the GBM subclasses did not reveal significant differences in overall prognosis ( Figure S7G ). Discussion Our study establishes methylation-based deconvolution as a powerful tool for unraveling GBM’s cellular composition by shedding light on tumor heterogeneity and linking cell type proportions to patient survival. By leveraging a reference atlas of methylation profiles from GBM microenvironment cell types, particularly through the inclusion of in vitro-differentiated human oligodendrocyte-lineage precursors, we successfully delineated the neoplastic cell composition of GBM using DNA methylation-based deconvolution. To our knowledge, this is the first time the cell composition of the neoplastic component of GBM has been investigated through methylation-based deconvolution, a result previously demonstrated only with techniques such as RNA-seq [ 3 , 6 , 20 – 22 ]. Moreover, the results were comparable to those of scRNA-seq analysis, while offering certain advantages such rapid speed of analysis, lower cost, and better feasibility and applicability, as it is compatible even with formalin-fixed paraffin-embedded and archived samples. Importantly, our approach enabled the identification of key cellular components within the tumor microenvironment and highlighted the astrocyte-to-oligodendrocyte ratio as a significant prognostic indicator in GBM. With regard to the proportions of different cell types, glial cells dominated the GBM composition (62%), which displayed primarily oligodendrocyte-like, oligodendrocyte precursor-like and astrocytes-like signatures, while cortical neurons contributed minimally. Additionally, we identified the presence of MSC-like cells, vascular endothelial cells, TAMs, and immune cells, further underscoring the complexity of the GBM microenvironment. The proportions of vascular endothelial cells were markedly higher in our analysis, averaging 13.9% (± 4.5%) compared to 1.7% (range 0.3–2.8%) reported in prior studies [ 6 , 7 , 20 – 22 ]. It should be noted that a recent report indicated 9.9% vascularization in GBM, which is closer to our observed proportion of 13.9% [ 33 ]. This finding underscores the well-documented angiogenic nature of GBM[ 34 ] and reinforces the reliability of our approach in capturing GBM’s vascular complexity. While most of the cell type proportions were broadly comparable to earlier findings,[ 3 , 6 , 7 , 20 – 22 ] a few notable differences emerged (Table 1 and Fig. 6 ). For instance, we observed a slightly higher proportion of immune cells (7.3% ± 5.4%) relative to the average of 3.3% reported elsewhere. These discrepancies may reflect differences in sample collection, tissue processing, or methodological biases. In particular, certain cell types tend to be underrepresented in single-cell RNA-seq due to their fragility or dissociation inefficiency. Despite these variations, our findings reinforce the robustness of methylation-based deconvolution in capturing the major cellular components of GBM [ 6 , 7 , 20 – 22 ]. Table 1 Comparison of Methylation-Based Deconvolution Results Across Studies Fraction Cell types Neftel et al. Yu et al. Darmanis et al. Yuan et al. Singh et al. Other studies (avg.) Our study (avg.) Non-Neoplastic fraction B cells - - - - 1.1% 1.1% 3.4% CD4T cells - - - - 0.2% 0.2% 0.3% NK-cells 0.0% 1.4% - - 0.4% 0.6% 0.4% CD8T cells 1.2% 3.0% - 0.5% 1.3% 1.5% 1.1% Neutrophils 0.1% 2.8% - - 1.1% 1.3% 2.2% TAMs 5.4% 6.4% 7.7% 0.2% 7.5% 5.4% 7.8% Cortical neurons - - 1.4% 0.0% 0.2% 0.5% 1.5% Normal Oligodendrocytes 2.8% 9.1% 6.9% 1.6% 6.2% 5.3% - Normal Astrocytes - 0.6% 1.4% - - 1.0% - Neoplastic fraction 87% 52% 59% 83% 78% 72% 70% Summary of the proportions of various cell types and neoplastic fraction identified in our study (the right most column) compared to those reported in other studies and their average. The left four data columns represent the scRNA-seq-based studies: Neftel et al.[ 3 ], Yu et al.[ 22 ], Darmanis et al.[ 20 ] and Yuan et al.[ 21 ] The fifth column on the right displays findings from methylation-based analysis by Singh et al.[ 7 ]. The two rightmost columns display the averages from other studies and our study. To distinguish neoplastic from non-neoplastic components, we used the RF_Purify method, which leverages the ABSOLUTE score to infer tumor purity. The neoplastic fraction, estimated from the 263 GBM samples using methylation-based deconvolution, was 70% ± 11%. This value demonstrates strong concordance with prior estimates of 72% (range, 52–87%) derived primarily from scRNA-seq [ 3 , 6 , 20 – 22 ] (Table 1 and Fig. 6 ). Thus, overall, the tumor purity levels are highly consistent across methods. Further, this analysis confirmed that the glial cells OPs, astrocytes, and oligodendrocytes, along with MSCs, constitute the neoplastic fraction. This is line with previous scRNA-seq studies [ 3 , 6 , 20 – 22 ]. Moreover, immune cells, TAMs, and vascular endothelial cells predominantly belong to the tumor microenvironment, as also observed in previous studies [ 3 , 7 , 20 – 22 ]. These tumor purity data were also validated through deconvolution analysis of a GIC dataset that provided an independent validation of the purity analysis in bulk GBM tumors. Cell types positively correlated with tumor purity in the bulk GBM dataset results were enriched in GICs, while those negatively correlated, probably representing non-malignant or microenvironmental cells, were reduced. Thus, the biological validity of the deconvolution approach in terms of effectively distinguishing malignant from non-malignant components was confirmed. We demonstrated that the oligodendrocyte differentiation lineage accounts for approximately 50% of the bulk tumor and a striking 71% of the neoplastic fraction. These findings extend prior work by Liu et al. [ 35 ], Persson et al [ 36 ], and Ligon et al.[ 37 ], which showed that oligodendrocyte precursor cells can give rise to GBM under defined oncogenic conditions. By revealing that oligodendrocyte-like and oligodendrocyte precursor-like cells comprise 71% of the neoplastic fraction, our results strongly reinforce the involvement of the oligodendrocyte lineage in GBM pathogenesis and suggest that pre-oligodendrocytes may serve as cells of origin for GBM. To evaluate the accuracy of our neoplastic cell composition estimates, we applied DNA methylation-based deconvolution to CD133-enriched tumor dissociates from 20 GBM patients [ 38 ]. As expected, the neoplastic fraction was higher than in that in bulk tumors (~ 90% vs. ~70%), consistent with the enrichment results for cancer cells. The remaining ~ 10% consisted largely of endothelial cells (~ 6.8%), which probably reflecting GBM’s prominent vasculature and minor contamination from tumor-associated blood vessels during CD133 + cell enrichment. Significant differences in cell-type proportions were observed among GBM subclasses RTK1, RTK2, and MES, including variations in TAMs and immune cell fractions (e.g., B cells, CD4 + T cells, NK cells, CD8 + T cells, and neutrophils). These findings are consistent with transcriptomic and immunohistochemistry analyses, which report increased tumor-infiltrating lymphocytes and TAMs in the mesenchymal GBM subtype compared to non-mesenchymal subtypes [ 7 , 39 , 40 ]. Our results extend these subclassifications by revealing distinct differences in cellular composition detected through methylation-based deconvolution. We observed distinct astrocyte and MSC methylation signatures in the neoplastic fraction of GBM that correspond to the astrocyte-like and mesenchymal-like states, respectively, identified by scRNA-seq [ 3 – 5 ] as part of the neoplastic fraction of GBM. Moreover, our analysis indicates that early oligodendrocyte lineage stages (OLIG2⁺ or OPC-like cells) and NPs contribute minimally to GBM composition. Instead, most oligodendrocyte-lineage cells display advanced differentiation, as reflected by pre-oligodendrocyte (O4-positive) and mature oligodendrocyte methylation signatures. In alignment with prior research [ 3 , 33 ], our findings indicate that astrocyte-like and oligodendrocyte-like signatures constitute a significant part of the GBM neoplastic fraction. Further, the consistently reported low abundance of normal astrocytes and oligodendrocytes in GBM samples [ 3 , 6 , 7 , 20 – 22 ] supports the conclusion that the elevated proportions we detected primarily reflect the neoplastic component. A key finding of our study is the significant association of GBM cellular composition, particularly the astrocyte and oligodendrocyte proportions, with patient survival. More specifically, higher proportions of astrocytes were linked to improved overall survival, while increased proportions of oligodendrocytes and OPs were associated with poorer outcomes. These results are in alignment with previous studies which showed that the OPC- and NPC-like tumor states are associated with the highest proliferative potential, while the astrocyte-like states are associated with the lowest proliferative potential [ 3 , 4 ]. We identified the astrocyte-to-oligodendrocyte ratio as a key prognostic biomarker. This aligns with the findings that the astrocyte-like state has reduced tumor-initiating potential compared to oligodendrocyte-lineage components, which exhibit greater tumorigenicity in mouse models [ 4 ]. Although the RTK1, RTK2, and MES subtypes exhibited distinct cellular compositions, these differences did not translate into significant overall survival variations across subtypes, as reported in other studies [ 39 , 41 ]. Instead, as suggested above, survival differences were driven by the astrocyte-to-oligodendrocyte ratio. This observation suggests that intertumoral cellular heterogeneity may be a more powerful prognostic indicator than subclassification alone. Nevertheless, patients with lower survival times, as determined by a higher astrocyte-to-oligodendrocyte ratio, had a greater proportion of GBM tumors classified as RTK2 than RTK1, and vice versa. Since RTK2 showed higher proportions of astrocytes and lower proportions of oligodendrocytes, compared to RTK1, this may explain the survival differences between the two subclasses. This study has several limitations. First, the DNA methylation profiles of in vitro-differentiated NPs and oligodendrocyte-lineage precursors may not fully capture the methylation landscapes present in GBM tumors, potentially affecting the accuracy of our findings. Second, assigning the methylation signature of OPs and mature oligodendrocytes to the less differentiated NPC-like and OPC-like transcriptional signatures reported in scRNA-seq [ 3 – 5 ] as part of the neoplastic fraction remains a challenge. The discrepancy between our findings and the previous studies [ 3 – 5 ] probably stems from methodological differences between single-cell transcriptional profiling and bulk methylation analysis. Further investigations, particularly single-cell methylation studies, are needed to validate this interpretation. Despite these methodological differences, both approaches support the notion that these populations represent distinct stages within the oligodendrocyte differentiation lineages. Finally, the survival analysis was conducted on a cohort predominantly receiving standard-of-care therapy, which may limit its applicability to patients undergoing alternative treatments or to underrepresented GBM subtypes. Thus, further in-depth research is required to evaluate the broader applicability of these results. Conclusions DNA methylation-based deconvolution offers a robust framework for resolving the cellular composition of GBM. We identified the astrocyte-to-oligodendrocyte ratio as a significant prognostic biomarker linked to patient survival. In addition, the findings point to an oligodendrocytic cell of origin and shed further light on the pathology of this tumor. Overall, this approach has the potential to provide clinically meaningful insights from initial tumor resections while offering several advantages over current methods like scRNA-seq. It is likely to have high applicability in routine clinical implementation, with the potential to improve patient stratification, guide personalized therapy, and ultimately, impact survival outcomes. Abbreviations BLCA bladder urothelial carcinoma CNS central nervous system CpGs cytosine-guanine dinucleotides GBM IDH-wildtype glioblastoma GIC glioma-initiating cell hESCs human embryonic stem cells KIRP kidney renal papillary cell carcinoma MES mesenchymal MSCs mesenchymal stem cells MST median survival time MSTD median survival time difference NPCs neural progenitor cells NPs neural progenitors OLIG2 cells OLIG2-positive progenitor cells OPs oligodendrocyte-lineage precursor cells OPCs Olig2+/Nkx2.2 + oligodendrocyte progenitor cells OSP overall survival probability PRAD prostate adenocarcinoma RTK1 receptor tyrosine kinase I RTK2 receptor tyrosine kinase II scRNA-seq single-cell RNA sequencing TAMs tumor-associated macrophages TME tumor microenvironment Declarations Ethics approval and consent to participate All data from human patients used in this study were obtained from online databases, including The Cancer Genome Atlas (TCGA) and GEO. Therefore, the need for informed consent and ethics approval was waived by the university. Consent for publication Not applicable Availability of data and materials Reference atlases used for deconvolution analysis and clinical data for the survival analysis cohort are available in the supplementary data. Accession numbers for the datasets supporting the conclusions in this article are provided in the main text. Competing interests The authors declare that they have no competing interests. Funding This research was supported by a generous gift from Leslie and Michael Gaffin. Author Contributions: Conceptualization, A.I., A.E., J.M, I.L.; methodology, A.I., N.L., J.M, BE. R, M.I, D.S, E. BS, I.L. MS; software, A.I., N.L, J.M.; validation, A.I., J.M. M.I, A.Z. I.L.; formal analysis, A.I., I.L.; investigation, A.I, I.L.; resources, A.I., H.C, M.G., M.I, D.S., EB.S, M.S, J.M, BE, R, I.L.; data curation, A.I., N.L., J.M. BE. R, M.I, D.S; writing—original draft preparation, A.I., I.L.; writing-review and editing, A.I., N.L, J.M., I.L., A.L, A.M; Y.F. visualization, A.I., N.L., J.M., A.L, A.M, I.L.; supervision, I.L.; project administration, I.L.; funding acquisition, I.L. All authors have read and agreed to the published version of the manuscript. Acknowledgments The results shown here are in whole or part based on data generated by the TCGA Research Network: https://www.cancer.gov/tcga. We sincerely thank Andreas von Deimling and his laboratory for conducting the EPIC methylation profiling of the in vitro differentiated neural progenitors [42] and oligodendrocyte precursor cells, which greatly contributed to this research. We also extend our heartfelt thanks to Leslie and Michael Gaffin for their generous support. References Ostrom QT, Bauchet L, Davis FG, Deltour I, Fisher JL, Langer CE, Pekmezci M, Schwartzbaum JA, Turner MC, Walsh KMet al et al (2014) The epidemiology of glioma in adults: a state of the science review. Neuro Oncol 16:896–913. 10.1093/neuonc/nou087 Patel AP, Tirosh I, Trombetta JJ, Shalek AK, Gillespie SM, Wakimoto H, Cahill DP, Nahed BV, Curry WT Martuza RL (2014) Single-cell RNA-seq highlights intratumoral heterogeneity in primary glioblastoma. Sci (New York NY) 344: 1396–1401 10.1126/science.1254257 Neftel C, Laffy J, Filbin MG, Hara T, Shore ME, Rahme GJ, Richman AR, Silverbush D, Shaw ML, Hebert CMet al et al (2019) An Integrative Model of Cellular States, Plasticity, and Genetics for Glioblastoma. Cell 178: 835–849 e821 10.1016/j.cell.2019.06.024 Suvà ML, Tirosh I (2020) The Glioma Stem Cell Model in the Era of Single-Cell Genomics. Cancer Cell 37:630–636. 10.1016/j.ccell.2020.04.001 Sharma P, Aaroe A, Liang J, Puduvalli VK (2023) Tumor microenvironment in glioblastoma: Current and emerging concepts. Neurooncol Adv 5:vdad009. 10.1093/noajnl/vdad009 Caruso FP, Garofano L, D'Angelo F, Yu K, Tang F, Yuan J, Zhang J, Cerulo L, Pagnotta SM, Bedognetti D et al (2020) A map of tumor-host interactions in glioma at single-cell resolution. Gigascience 9: 10.1093/gigascience/giaa109 Singh O, Pratt D, Aldape K (2021) Immune cell deconvolution of bulk DNA methylation data reveals an association with methylation class, key somatic alterations, and cell state in glial/glioneuronal tumors. Acta Neuropathol Commun 9:148. 10.1186/s40478-021-01249-9 Capper D, Jones DTW, Sill M, Hovestadt V, Schrimpf D, Sturm D, Koelsche C, Sahm F, Chavez L, Reuss DE et al (2018) DNA methylation-based classification of central nervous system tumours. Nature 555: 469–474 10.1038/nature26000 Capper D, Stichel D, Sahm F, Jones DTW, Schrimpf D, Sill M, Schmid S, Hovestadt V, Reuss DE, Koelsche C et al (2018) Practical implementation of DNA methylation and copy-number-based CNS tumor diagnostics: the Heidelberg experience. Acta Neuropathol 136: 181–210 10.1007/s00401-018-1879-y Dor Y, Cedar H Principles of DNA methylation and their implications for biology and medicine. Lancet 392: 777–786 Doi, Bergman Y, Cedar H (2018) (2013) DNA methylation dynamics in health and disease. Nat Struct Mol Biol 20: 274–281 Doi 10.1038/nsmb.2518 Fernandez AF, Assenov Y, Martin-Subero JI, Balint B, Siebert R, Taniguchi H, Yamamoto H, Hidalgo M, Tan AC, Galm Oet al, Hovestadt V, Jones DT, Picelli S, Wang W, Kool M, Northcott PA, Sultan M, Stachurski K, Ryzhova M et al (2012) Warnatz HJ (2014) Decoding the regulatory landscape of medulloblastoma using DNA methylation sequencing. Nature 510: 537–541 Doi 10.1038/nature13268 Sahm F, Schrimpf D, Stichel D, Jones DTW, Hielscher T, Schefzyk S, Okonechnikov K, Koelsche C, Reuss DE, Capper D al (2017) DNA methylation-based classification and grading system for meningioma: a multicentre, retrospective analysis. Lancet Oncol 18:682–694. 10.1016/S1470-2045(17)30155-9 Louis DN, Perry A, Reifenberger G, von Deimling A, Figarella-Branger D, Cavenee WK, Ohgaki H, Wiestler OD, Kleihues P, Ellison DW (2016) The 2016 World Health Organization Classification of Tumors of the Central Nervous System: a summary. Acta Neuropathol 131:803–820. 10.1007/s00401-016-1545-1 Louis DN, Perry A, Wesseling P, Brat DJ, Cree IA, Figarella-Branger D, Hawkins C, Ng HK, Pfister SM, Reifenberger Get al et al (2021) The 2021 WHO Classification of Tumors of the Central Nervous System: a summary. Neuro Oncol 23:1231–1251. 10.1093/neuonc/noab106 McLendon R, Friedman A, Bigner D, Van Meir EG, Brat DJ, Mastrogianakis M, Olson G, Mikkelsen JJ, Lehman T, Aldape Ket al et al (2008) Comprehensive genomic characterization defines human glioblastoma genes and core pathways. Nature 455:1061–1068. 10.1038/nature07385 Bedard PL, Hansen AR, Ratain MJ, Siu LL (2013) Tumour heterogeneity in the clinic. Nature 501:355–364. 10.1038/nature12627 Loyfer N, Magenheim J, Peretz A, Cann G, Bredno J, Klochendler A, Fox-Fisher I, Shabi-Porat S, Hecht M, Pelet Tet al et al (2023) A DNA methylation atlas of normal human cell types. Nature 613:355–364. 10.1038/s41586-022-05580-6 Moss J, Magenheim J, Neiman D, Zemmour H, Loyfer N, Korach A, Samet Y, Maoz M, Druid H, Arner P et al (2018) Comprehensive human cell-type methylation atlas reveals origins of circulating cell-free DNA in health and disease. Nat Commun 9: 5068 10.1038/s41467-018-07466-6 Alcantara Llaguno SR, Parada LF (2016) Cell of origin of glioma: biological and clinical implications. Br J Cancer 115:1445–1450. 10.1038/bjc.2016.354 Darmanis S, Sloan SA, Croote D, Mignardi M, Chernikova S, Samghababi P, Zhang Y, Neff N, Kowarsky M, Caneda C al (2017) Single-Cell RNA-Seq Analysis of Infiltrating Neoplastic Cells at the Migrating Front of Human Glioblastoma. Cell Rep 21:1399–1410. 10.1016/j.celrep.2017.10.030 Yuan J, Levitin HM, Frattini V, Bush EC, Boyett DM, Samanamud J, Ceccarelli M, Dovas A, Zanazzi G, Canoll P et al (2018) Single-cell transcriptome analysis of lineage diversity in high-grade glioma. Genome Med 10: 57 10.1186/s13073-018-0567-9 Yu K, Hu Y, Wu F, Guo Q, Qian Z, Hu W, Chen J, Wang K, Fan X, Wu X al (2020) Surveying brain tumor heterogeneity by single-cell RNA-sequencing of multi-sector biopsies. Natl Sci Rev 7:1306–1318. 10.1093/nsr/nwaa099 Nishri Y, Hampton D, Ben-Shushan E, Fainstein N, Magnani D, Aharonowiz M, Reubinoff BE, Chandran S, Ben-Hur T (2020) Continuous Immune-Modulatory Effects of Human Olig2 + Precursor Cells Attenuating a Chronic-Active Model of Multiple Sclerosis. Mol Neurobiol 57:1021–1034. 10.1007/s12035-019-01802-7 Tannenbaum SE, Turetsky TT, Singer O, Aizenman E, Kirshberg S, Ilouz N, Gil Y, Berman-Zaken Y, Perlman TS, Geva Net al et al (2012) Derivation of xeno-free and GMP-grade human embryonic stem cells–platforms for future clinical applications. PLoS ONE 7:e35325. 10.1371/journal.pone.0035325 Vinel C, Rosser G, Guglielmi L, Constantinou M, Pomella N, Zhang X, Boot JR, Jones TA, Millner TO Dumas AA (2021) Comparative epigenetic analysis of tumour initiating cells and syngeneic EPSC-derived neural stem cells in glioblastoma. Nat Commun 12: 6130 10.1038/s41467-021-26297-6 De Meyer T, Bady P, Trooskens G, Kurscheid S, Bloch J, Kros JM, Hainfellner JA, Stupp R, Delorenzi M, Hegi ME al (2015) Genome-wide DNA methylation detection by MethylCap-seq and Infinium HumanMethylation450 BeadChips: an independent large-scale comparison. Sci Rep 5:15375. 10.1038/srep15375 Bady P, Marosi C, Weller M, Grønberg BH, Schultz H, Taphoorn MJB, Gijtenbeek JMM, van den Bent MJ, von Deimling A Stupp R (2022) DNA methylation-based age acceleration observed in IDH wild-type glioblastoma is associated with better outcome-including in elderly patients. Acta Neuropathol Commun 10: 39 10.1186/s40478-022-01344-5 Carter SL, Cibulskis K, Helman E, McKenna A, Shen H, Zack T, Laird PW, Onofrio RC, Winckler W, Weir BA et al (2012) Absolute quantification of somatic DNA alterations in human cancer. Nat Biotechnol 30: 413–421 10.1038/nbt.2203 Alboukadel K, Marcin K, Przemyslaw B (2024) survminer: Drawing Survival Curves using 'ggplot2'. R package version 0.5.0. City Shahar T, Granit A, Zrihan D, Canello T, Charbit H, Einstein O, Rozovski U, Elgavish S, Ram Z, Siegal Tet al et al (2016) Expression level of miRNAs on chromosome 14q32.31 region correlates with tumor aggressiveness and survival of glioblastoma patients. J Neurooncol 130:413–422. 10.1007/s11060-016-2248-0 Johann PD, Jäger N, Pfister SM, Sill M (2019) RF_Purify: a novel tool for comprehensive analysis of tumor-purity in methylation array data based on random forest regression. BMC Bioinformatics 20:428. 10.1186/s12859-019-3014-z Chaligne RA-O, Gaiti FA-O, Silverbush D, Schiffman JA-O, Weisman HR, Kluegel L, Gritsch S, Deochand SD Gonzalez Castro LA-O, Richman AR Epigenetic encoding, heritability and plasticity of glioma transcriptional cell states Greenwald AC, Darnell NG, Hoefflin R, Simkin D, Mount CW, Gonzalez Castro LN, Harnik Y, Dumont S, Hirsch D, Nomura Met al et al (2024) Integrative spatial analysis reveals a multi-layered organization of glioblastoma. Cell 187:2485–2501e2426. 10.1016/j.cell.2024.03.029 Jain RK, di Tomaso E, Duda DG, Loeffler JS, Sorensen AG, Batchelor TT (2007) Angiogenesis in brain tumours. Nat Rev Neurosci 8:610–622. 10.1038/nrn2175 Liu C, Sage JC, Miller MR, Verhaak RG, Hippenmeyer S, Vogel H, Foreman O, Bronson RT, Nishiyama A, Luo Let al et al (2011) Mosaic analysis with double markers reveals tumor cell of origin in glioma. Cell 146:209–221. 10.1016/j.cell.2011.06.014 Persson AI, Petritsch C, Swartling FJ, Itsara M, Sim FJ, Auvergne R, Goldenberg DD, Vandenberg SR, Nguyen KN, Yakovenko Set al et al (2010) Non-stem cell origin for oligodendroglioma. Cancer Cell 18:669–682. 10.1016/j.ccr.2010.10.033 Ligon KL, Alberta JA, Kho AT, Weiss J, Kwaan MR, Nutt CL, Louis DN, Stiles CD, Rowitch DH (2004) The oligodendroglial lineage marker OLIG2 is universally expressed in diffuse gliomas. J Neuropathol Exp Neurol 63:499–509. 10.1093/jnen/63.5.499 Singh SK, Clarke ID, Terasaki M, Bonn VE, Hawkins C, Squire J, Dirks PB (2003) Identification of a cancer stem cell in human brain tumors. Cancer Res 63:5821–5828 Wang Q, Hu B, Hu X, Kim H, Squatrito M, Scarpace L, deCarvalho AC, Lyu S, Li P, Li Y al (2017) Tumor Evolution of Glioma-Intrinsic Gene Expression Subtypes Associates with Immunological Changes in the Microenvironment. Cancer Cell 32:42–56e46. 10.1016/j.ccell.2017.06.003 Snuderl M Molecular classification and deconvolution of the immune microenvironment in glioblastoma. Neuro-Oncology 23: 175–176 Doi, Kaffes I, Szulzewsky F, Chen Z, Herting CJ, Gabanic B, Velázquez Vega JE, Shelton J, Switchenko JM, Ross JL (2021) McSwain LF (2019) Human Mesenchymal glioblastomas are characterized by an increased immune cell presence compared to Proneural and Classical tumors. Oncoimmunology 8: e1655360 Doi 10.1080/2162402X.2019.1655360 Drexler R, Schüller U, Eckhardt A, Filipski K, Hartung TI, Harter PN, Divé I, Forster MT, Czabanka M, Jelgersma C al (2023) DNA methylation subclasses predict the benefit from gross total tumor resection in IDH-wildtype glioblastoma patients. Neuro Oncol 25:315–325. 10.1093/neuonc/noac177 Davenport CF, Scheithauer T, Dunst A, Bahr FS, Dorda M, Wiehlmann L, Tran DDH (2021) Genome-Wide Methylation Mapping Using Nanopore Sequencing Technology Identifies Novel Tumor Suppressor Genes in Hepatocellular Carcinoma. Int J Mol Sci 22. 10.3390/ijms22083937 Additional Declarations No competing interests reported. Supplementary Files SupplementaryInformation25.05.2025.xlsx GBMDeconvolutionSM25.05.2025Final.docx floatimage1.png Graphical Abstract Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6743395","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":466156852,"identity":"40c8af3b-f914-4330-a3aa-2ae22060e4b5","order_by":0,"name":"Aviel Iluz","email":"","orcid":"","institution":"Hadassah Hebrew University Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Aviel","middleName":"","lastName":"Iluz","suffix":""},{"id":466156853,"identity":"ed314555-9225-4855-9386-d6a68b382703","order_by":1,"name":"Nir Lavi","email":"","orcid":"","institution":"Hadassah Hebrew University Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Nir","middleName":"","lastName":"Lavi","suffix":""},{"id":466156854,"identity":"512298c5-4b3b-4448-b2c1-44e80396bd2a","order_by":2,"name":"Hanna Charbit","email":"","orcid":"","institution":"Hadassah Hebrew University Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Hanna","middleName":"","lastName":"Charbit","suffix":""},{"id":466156855,"identity":"8b721282-8845-4945-b075-b7f4f0ef863e","order_by":3,"name":"Mijal Gutreiman","email":"","orcid":"","institution":"Hadassah Hebrew University Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Mijal","middleName":"","lastName":"Gutreiman","suffix":""},{"id":466156856,"identity":"51ffc9b5-7c96-4e72-945f-0d475ab4d277","order_by":4,"name":"Masha Idelson","email":"","orcid":"","institution":"Hadassah Hebrew University Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Masha","middleName":"","lastName":"Idelson","suffix":""},{"id":466156857,"identity":"3cef8cb6-a1b3-4027-b884-b2307cf499ee","order_by":5,"name":"Debora Steiner","email":"","orcid":"","institution":"Hadassah Hebrew University Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Debora","middleName":"","lastName":"Steiner","suffix":""},{"id":466156858,"identity":"991a2f96-60db-4214-9e4f-a3cf41422472","order_by":6,"name":"Etti Ben-Shushan","email":"","orcid":"","institution":"Hadassah Hebrew University Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Etti","middleName":"","lastName":"Ben-Shushan","suffix":""},{"id":466156859,"identity":"32b58c8b-3bb6-48e8-91e1-aad3c87345cf","order_by":7,"name":"Aviad Zick","email":"","orcid":"","institution":"Hadassah Hebrew University Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Aviad","middleName":"","lastName":"Zick","suffix":""},{"id":466156860,"identity":"04e7e702-2b83-4ec6-9daf-be3ff4e986f2","order_by":8,"name":"Amir Eden","email":"","orcid":"","institution":"The Hebrew University of Jerusalem","correspondingAuthor":false,"prefix":"","firstName":"Amir","middleName":"","lastName":"Eden","suffix":""},{"id":466156861,"identity":"e8adaec3-aa44-4b2d-9760-f26d0f272533","order_by":9,"name":"Anat Mordechai","email":"","orcid":"","institution":"Hadassah Hebrew University Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Anat","middleName":"","lastName":"Mordechai","suffix":""},{"id":466156862,"identity":"fd67878a-d5f8-45f7-87d7-089edba632cc","order_by":10,"name":"Moscovici Samuel","email":"","orcid":"","institution":"Hadassah Hebrew University Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Moscovici","middleName":"","lastName":"Samuel","suffix":""},{"id":466156863,"identity":"d1d9e914-7447-4076-935d-4a513bafced6","order_by":11,"name":"Yakov Fellig","email":"","orcid":"","institution":"Hadassah Hebrew University Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Yakov","middleName":"","lastName":"Fellig","suffix":""},{"id":466156864,"identity":"cb116775-be93-454c-95b8-1ca050820c52","order_by":12,"name":"Alexander Lossos","email":"","orcid":"","institution":"Hadassah Hebrew University Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Alexander","middleName":"","lastName":"Lossos","suffix":""},{"id":466156865,"identity":"7a9e828f-deb8-421d-8bb8-47f9e020bd19","order_by":13,"name":"Joshua Moss","email":"","orcid":"","institution":"Hadassah Hebrew University Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Joshua","middleName":"","lastName":"Moss","suffix":""},{"id":466156866,"identity":"ae6ba462-6ac9-4381-a74d-806cc085aaf3","order_by":14,"name":"Benjamin E. Reubinoff","email":"","orcid":"","institution":"Hadassah Hebrew University Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Benjamin","middleName":"E.","lastName":"Reubinoff","suffix":""},{"id":466156867,"identity":"f7ed28de-ad7b-4ce2-97cc-73ff41915855","order_by":15,"name":"Iris Lavon","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA5UlEQVRIiWNgGAWjYBAC+QbGBoYHFQcYGICIAcgGgwMP8GhhAylLOIOuJQGvFiBIbEPTwoBXi/Th5g+J8+7I8x0//vjDzx0Mcub9Cxjx28KX2CaRuO2Z4cwzOWaSvWcYjGVuPCDgMB7GNobEbYcZNxzIYWPgBbJnSBwgqAXosDmH7Tecf/74418itTRIJDYcTtxwI8FAGmwLfwNhh0kkHDucPPPGGzNp2TYJYwkJxga8WuR72B9/+FBz2LbvfPrjj2/bbOQk+A8f/vABjxZ0IAFEiQ0kaAAD/gOk6hgFo2AUjIJhDgCqNliLXuofJQAAAABJRU5ErkJggg==","orcid":"","institution":"Hadassah Hebrew University Medical Center","correspondingAuthor":true,"prefix":"","firstName":"Iris","middleName":"","lastName":"Lavon","suffix":""}],"badges":[],"createdAt":"2025-05-25 11:23:35","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6743395/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6743395/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":84216440,"identity":"affc909b-5558-44b2-95f3-2de3b1d327cb","added_by":"auto","created_at":"2025-06-09 10:44:35","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":136351,"visible":true,"origin":"","legend":"\u003cp\u003eImmunostaining results showing the differentiation of hESCs into oligodendroglial lineage cells\u003c/p\u003e\n\u003cp\u003e(A) Dot plot presentation of NPs co-expressing PSA-NCAM and A2B5. (B–D) Immunostaining results showing the expression of specific markers characterizing different stages of the development of oligodendroglial lineage cells: (B) OLIG2-positive progenitors, (C) OPCs co-expressing Olig2 and Nkx2.2 (marked by white arrows), (D) pre-oligodendrocytes expressing O4. Scale bars: (B–C), 100 mm; (D), 200 mm\u003c/p\u003e","description":"","filename":"Picture1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6743395/v1/37239d021d50df76bb79acb2.jpg"},{"id":84215097,"identity":"d9e0d1eb-e5ef-43f4-bfd4-4a067c7d4781","added_by":"auto","created_at":"2025-06-09 10:36:35","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":251392,"visible":true,"origin":"","legend":"\u003cp\u003eReference methylation atlas for GBM deconvolution and cell-type identification\u003c/p\u003e\n\u003cp\u003eThe methylation atlas comprises 14 tissues/cell types (columns) across 4111 CpG sites (rows) that are located in 2610 genomic blocks of 500 bp each. Feature selection was carried out as described in a previous study[18]. Specifically, the top 100 uniquely hypermethylated CpGs and hypomethylated CpGs for each cell type, yielding a total of 2800 tissue-specific CpGs, with neighboring CpGs within 50 bp were included.\u003c/p\u003e","description":"","filename":"Picture2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6743395/v1/a5177171032cc54cb17207e7.jpg"},{"id":84216847,"identity":"52491938-bf66-4c76-9aa6-d4de97570f61","added_by":"auto","created_at":"2025-06-09 10:52:35","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":413815,"visible":true,"origin":"","legend":"\u003cp\u003eCell-type proportions derived from deconvolution analysis of GBM and their differences across GBM subtypes based on the reference atlas comprising 14 cell types\u003c/p\u003e\n\u003cp\u003e(A) Bar plot summarizing the average proportions of 14 cell types (combined into 9 groups) from the deconvolution of the methylation profiles of 263 samples shown in A–I (leftmost bar), subdivided into RTK1 (n = 64), RTK2 (n = 143), and MES (n = 56). The average percentage and standard deviation for each cell type are displayed within its respective bar. (B–J) Boxplots depicting the proportions of 9 cell types/cell-type groups derived from the deconvolution of methylation data across GBM samples (n = 263, blue, leftmost), subdivided into RTK1 (n = 64, orange), RTK2 (n = 143, green), and MES (n = 56, pink). The 9 panels correspond to (B) oligodendrocytes, (c) oligodendrocyte-lineage precursor cells (OPs), (D) astrocytes, (E) mesenchymal stem cells (mesenchymal SCs), (F) neural progenitors, (G) cortical neurons, (H) vascular endothelial cells, (I) tumor-associated macrophages (TAMs, comprising microglia and monocytes), and (J) immune cells (comprising B cells, CD4 T cells, NK cells, CD8 T cells, and neutrophils). The central line represents the median, and the box edges indicate the interquartile range (IQR), with the whiskers extending to 1.5 times the IQR. The black lines denote significant differences between groups, determined by the pairwise two-tailed Student’s \u003cem\u003et\u003c/em\u003e-test. Exact p-values are reported in the main text. (*p \u0026lt; 0.05; **p \u0026lt; 0.01; ****p \u0026lt; 0.0001; ns: not significant).\u003c/p\u003e","description":"","filename":"Picture3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6743395/v1/13ae6073cad28714a11edd32.jpg"},{"id":84216442,"identity":"5ee1a4c9-e4c5-436b-9792-8801b7d89ab4","added_by":"auto","created_at":"2025-06-09 10:44:35","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":280110,"visible":true,"origin":"","legend":"\u003cp\u003eCorrelation between cell-type proportions and tumor purity estimates calculated by RF_Purify_ABSOLUTE\u003c/p\u003e\n\u003cp\u003e(A–D) Correlation between cell type proportions (Y axis) derived from the deconvolution of all 263 GBM samples and sample purity (X axis) estimated by the RF_Purify algorithm using the ABSOLUTE method. The analyzed cell types include (A) total immune cells (purple), tumor-associated macrophages (TAMs, dark blue), and vascular endothelial cells (dark orange); (B) mesenchymal stem cells (MSCs, blue), astrocytes (red), oligodendrocyte-lineage precursor cells (OPs, light green), and oligodendrocytes (dark green); (C) the three cells types with negative correlation combined (immune cells + TAMs + vascular endothelial cells, light purple); (D) the four cell types with positive correlation combined (three glial cell types + MSCs, green). (E) An overview histogram illustrates the Pearson correlation (R) values (Y axis) between the proportions of the nine cellular components shown in panels A, B, C, and D (X-axis) and sample purity. The dashed line divides the graph into two sections, with the far-right bar in each section representing the correlation for the combined cells groups representing cells with positive correlations (glial cells + MSCs, green) or negative correlation (immune cells + TAMs + vascular endothelial cells, pink).\u003c/p\u003e","description":"","filename":"Picture4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6743395/v1/b9a7ea762728f987da68096f.jpg"},{"id":84216849,"identity":"5bc9b03c-952e-4a10-87be-7a91da826caf","added_by":"auto","created_at":"2025-06-09 10:52:35","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":249955,"visible":true,"origin":"","legend":"\u003cp\u003eKaplan-Meier plots based on cell type proportions\u003c/p\u003e\n\u003cp\u003eKaplan-Meier survival analysis of 199 cases showing significant differences in OSP based on cell type proportions. MSTD between the groups is indicated in months. (A–C) Groups are stratified into high and low groups (High \u0026gt; median, Low \u0026lt; median) according to the cell type proportions: (A) oligodendrocyte proportion, (B) astrocyte proportion, (C) astrocyte-to-oligodendrocyte ratio. (D) High and low groups based on an astrocyte-to-oligodendrocyte cutoff value of 0.3 determined based on the threshold yielding the most significant log-rank test result, validated by a permutation-based approach.\u003c/p\u003e","description":"","filename":"Picture5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6743395/v1/9c98b2b59f275ebe3191f2b5.jpg"},{"id":84215100,"identity":"046411a2-7119-4415-860d-ae058689199d","added_by":"auto","created_at":"2025-06-09 10:36:35","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":88299,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of the proportions of different cell types between our study and other studies\u003c/p\u003e\n\u003cp\u003eBoxplots illustrating the distribution of cell type proportions and the neoplastic fraction in GBM samples from other studies (as detailed in Table 1), with each boxplot representing a specific cell type. The boxplot on the far right depicts the neoplastic fraction on a 100% scale. Our findings are indicated by a red circle for comparison. The central line in each boxplot shows the median; the X, the mean; and the box edges, the interquartile range (IQR). The whiskers extend to 1.5 times the IQR.\u003c/p\u003e","description":"","filename":"Picture6.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6743395/v1/3a57b780929b1ae4ebd24501.jpg"},{"id":90776500,"identity":"ee8b5938-b0a1-42ff-a553-273bc743ad29","added_by":"auto","created_at":"2025-09-08 03:16:53","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2869816,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6743395/v1/84bac3fe-199c-4249-bdaa-e2d1d961d64e.pdf"},{"id":84215102,"identity":"dbf7baf1-f1b9-477e-b577-b4a4328d0481","added_by":"auto","created_at":"2025-06-09 10:36:35","extension":"xlsx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":2153677,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryInformation25.05.2025.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-6743395/v1/29754282f6b99773dd6b3d0a.xlsx"},{"id":84216446,"identity":"ec7333a2-de1a-4d24-a194-0820946a916e","added_by":"auto","created_at":"2025-06-09 10:44:35","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":2881394,"visible":true,"origin":"","legend":"","description":"","filename":"GBMDeconvolutionSM25.05.2025Final.docx","url":"https://assets-eu.researchsquare.com/files/rs-6743395/v1/f6069eb8b1c9397b2e5f052c.docx"},{"id":84216444,"identity":"9eb9a2bb-6f8f-4ca2-891a-5cc7a29606d6","added_by":"auto","created_at":"2025-06-09 10:44:35","extension":"png","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":293642,"visible":true,"origin":"","legend":"\u003cp\u003eGraphical Abstract\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-6743395/v1/bc6a58296d2a35027b991069.png"}],"financialInterests":"No competing interests reported.","formattedTitle":"DNA Methylation-Based Deconvolution Study of Glioblastoma Heterogeneity and Cell Types Associated with Patient Survival","fulltext":[{"header":"Introduction","content":"\u003cp\u003eIDH-wildtype glioblastoma (GBM) is an aggressive and lethal brain cancer known for its rapid growth and resistance to standard treatments [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. A key feature of GBM is its extensive cellular heterogeneity, which presents challenges for treatment. The tumor contains a complex mix of neoplastic and non-neoplastic cells, including glioma-initiating cells, immune cells, endothelial cells, and glial cells [\u003cspan additionalcitationids=\"CR3 CR4\" citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. Thus, understanding the cellular drivers of GBM heterogeneity is essential for improving its diagnosis and treatment.\u003c/p\u003e \u003cp\u003eVarious methods, including proteomics and transcriptomic analyses, have been used to explore the complex cellular dynamics of GBM [\u003cspan additionalcitationids=\"CR3\" citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Recent breakthroughs in single-cell RNA sequencing (scRNA-seq) have provided new insights into the cellular diversity of GBM, revealing distinct subpopulations of neoplastic cells and their interactions with the tumor microenvironment (TME) [\u003cspan additionalcitationids=\"CR3 CR4\" citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. A landmark study [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e] identified four distinct cellular states within the neoplastic fraction of GBM: astrocyte-like, oligodendrocyte progenitor cell-like, mesenchymal-like, and neural progenitor cell-like. These findings underscore the tumor\u0026rsquo;s complex differentiation patterns and plasticity. However, while scRNA-seq has been instrumental in resolving such intratumoral heterogeneity, it is not without limitations. The technique is resource-intensive, involving high cost, extended processing time, and often limited scalability, which constrain its broader application, particularly in clinical or large-cohort studies. Furthermore, scRNA-seq is susceptible to transcriptional noise, dropout effects, and dissociation-induced artifacts, all of which can compromise the accurate representation of in vivo cellular states.\u003c/p\u003e \u003cp\u003eDNA methylation signature-based deconvolution has emerged as a promising alternative that can overcome the challenges with scRNA-seq.\u0026nbsp;This approach leverages the relative stability and cell-type specificity of DNA methylation patterns to infer the cellular composition of bulk tumor samples. Unlike scRNA-seq, it is cost-effective and scalable, and can be applied to archival formalin-fixed paraffin-embedded material, making it especially suitable for large-scale or retrospective studies. By bypassing the need for single-cell dissociation and transcript capture, DNA methylation deconvolution provides a robust and clinically viable means of profiling tumor heterogeneity, offering new insights into the cellular architecture of GBM [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Genome-wide DNA methylation profiling has become a transformative tool for classifying central nervous system (CNS) tumors, offering several advantages over traditional diagnostics in terms of patient stratification [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eEach cell type has a unique methylation signature that reflects its tissue identity [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e], and it is well-established that tumors exhibit stable methylation signatures, even after oncogenic events [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. Accordingly, the DNA methylation signature of a tumor reflects the cumulative methylation patterns of all its constituent cell types, including both normal and neoplastic cells. This composite profile captures the cellular admixture within the tumor, providing insights into its complex cellular makeup [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. These signatures have been validated in thousands of CNS tumors, facilitating accurate sub-classification and diagnoses and better treatment guidance [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan additionalcitationids=\"CR13\" citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e], and this method was incorporated into the 2021 WHO Classification of CNS tumors [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. In the context of GBM, beyond confirming the diagnosis, DNA methylation profiling has helped identify distinct subclasses, including receptor tyrosine kinase I (RTK1), receptor tyrosine kinase II (RTK2), and mesenchymal[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. Studies of newly diagnosed GBMs reveal significant spatial diversity within methylation subclasses, underscoring both the complexity of tumor classification and the critical role of intratumoral heterogeneity in driving treatment resistance and disease recurrence [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eMethylation-based deconvolution involves the use of a reference atlas of normal cell-type methylation signatures to estimate tissue composition and has been applied to analyze tissue contributions based on circulating cell-free DNA [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. A similar deconvolution methodology has been applied to investigate the composition of non-neoplastic immune cells within the GBM TME [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. However, to the best of our knowledge, methylation signatures have not been specifically used to investigate the neoplastic components of GBM. In order to address this gap, in the present study, we have adopted a comprehensive approach that uses methylation-based deconvolution and methylation-based purity analysis to analyze both the neoplastic and non-neoplastic components of GBM tumors. By analyzing methylation signatures from 462 GBM samples across three datasets, we aim to elucidate GBM heterogeneity encompassing all relevant cell types within the TME, with particular focus on the neoplastic fraction, and to investigate associations between the abundance of specific cell types and patient survival. This analysis provides valuable insights into the cell of origin of GBM, a critical factor in understanding early tumorigenic processes and identifying potential targets for early intervention. Since DNA methylation profiles preserve the signatures of the originating cell type, they offer a reliable method for tracing cellular origins. When combined with our uniquely developed in vitro-derived human neural progenitors and oligodendrocyte-lineage precursor methylation dataset, this approach allows for a comprehensive exploration of the cell of origin in GBM [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. Finally, to validate and contextualize our findings, we compared our results with data obtained through complementary methodologies, including scRNA-seq and the deconvolution of immune cell profiles from bulk glioma DNA methylation data [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan additionalcitationids=\"CR21\" citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. With this integrative approach, we believe that we have provided an enhanced understanding of GBM's cellular complexity that can inform future prognostic and therapeutic strategies.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eGeneration of Human Neural Progenitors and Oligodendrocyte Lineage Precursor Cells\u003c/h2\u003e \u003cp\u003eHighly enriched cultures of neural progenitors (NPs) were derived from human embryonic stem cells (hESCs) using an established protocol [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. Briefly, hESCs (HAD-C100 [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]) at passages 22 to 28 were cultured on human recombinant Laminin 521 (Biolamina, Sundbyberg, Sweden) in NutriStem medium supplemented with growth factors (Biological Industries, Beit Haemek, Israel) and 50 units/mL penicillin and 50 \u0026micro;g/mL streptomycin (Pen-Strep; Thermo Fisher Scientific, Waltham, MA, USA). Oligodendroglial precursor cells at various developmental stages were derived from these enriched NPs cultures as previously described [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. Briefly, NPs were first induced to differentiate into OLIG2-positive precursor cells (henceforth referred to as OLIG2 cells) and subsequently induced to differentiate into oligodendroglial lineage cells. This process resulted in an enriched population of Olig2+/Nkx2.2\u0026thinsp;+\u0026thinsp;oligodendrocyte progenitor cells (henceforth referred to as OPCs), which were further differentiated into O4\u0026thinsp;+\u0026thinsp;pre-oligodendrocytes (henceforth referred to as pre-oligodendrocytes).\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eImmunofluorescent Staining of hESC-Derived Oligodendroglial Lineage Cells to Detect Different Stages of Development\u003c/h3\u003e\n\u003cp\u003eTo characterize the developmental stages of oligodendroglial lineage cells, markers representing different maturation steps were detected by immunofluorescence. Briefly, cells were plated on glass coverslips pretreated with poly-\u003cspan type=\"SmallCaps\" class=\"SmallCaps\" name=\"Emphasis\"\u003ed\u003c/span\u003e-lysine (30\u0026ndash;70 kDa) at a concentration of 10 mg/ml and laminin (concentration, 4 mg/ml; both from Sigma, St. Louis, MO) and cultured for 3\u0026ndash;5 days. The cells were then fixed with 4% paraformaldehyde and incubated with mouse monoclonal anti-Olig2 (1:150; EMD Millipore, Burlington, MA) and rabbit anti-Nkx2.2 (1:50; Novus Biologicals, Centennial, CO) antibodies. Alexa Fluor 555-conjugated donkey anti-mouse antibody (1:150; Invitrogen/Thermo Fisher Scientific, Waltham, MA) and Alexa Fluor 488-conjugated donkey anti-rabbit antibody (1:150; Jackson ImmunoResearch Laboratories, West Grove, PA) were used for detection of the markers.\u003c/p\u003e \u003cp\u003eFor detection of the pre-oligodendrocyte stage, the cells were incubated with mouse anti-O4 antibody (1:150; R\u0026amp;D Systems, Minneapolis, MN), fixed with 4% paraformaldehyde, and incubated with Alexa Fluor 488-conjugated donkey anti-mouse antibody (1:150; Jackson ImmunoResearch). Nuclei were counterstained with 4\u0026prime;,6-diamidino-2-phenylindole (Vector Laboratories, Burlingame, CA). Specimens were visualized using an Olympus BX61 fluorescence microscope (Olympus, Hamburg, Germany).\u003c/p\u003e\n\u003ch3\u003eFlow Cytometry Analysis\u003c/h3\u003e\n\u003cp\u003eHuman ESC-derived neural progenitor cells (NPCs) were immunostained with mouse anti-PSA-NCAM-PE (1:50) and mouse anti-A2B5-APC (1:50) antibodies, or the corresponding isotype controls (all from Miltenyi Biotec, Waltham, MA). These markers are used to differentiate between distinct developmental stages and lineage potentials within NPC populations. Specifically, A2B5 is a marker of glial-restricted progenitors and early oligodendrocyte precursor cells, while PSA-NCAM is a marker of neuronal-committed progenitors and early post-mitotic neurons. By analyzing the expression profiles of these markers, the experiment distinguished between A2B5⁺/PSA-NCAM⁻ cells, which represent glial-biased progenitors; PSA-NCAM⁺/A2B5⁻ cells, enriched for neuronal lineage commitment; double-positive or double-negative subsets, which may reflect bipotent progenitors, intermediate states, or uncommitted populations. For each sample, at least 10\u003csup\u003e4\u003c/sup\u003e cells were analyzed on a CytoFLEX flow cytometer (Beckman Coulter, Indianapolis, IN).\u003c/p\u003e\n\u003ch3\u003eDNA Methylation Profiling\u003c/h3\u003e\n\u003cp\u003eGenome-wide DNA methylation profiling of NPs, OLIG2 cells, OPCs, and pre-oligodendrocytes was performed using the Illumina EPIC array, with data generated at the Genomics and Proteomics Core Facility of the German Cancer Research Center (DKFZ) in Heidelberg, Germany. Data processing, filtering, and normalization were carried out using the \u003cem\u003eminfi\u003c/em\u003e R package (v1.34.0).\u003c/p\u003e\n\u003ch3\u003eData Collection\u003c/h3\u003e\n\u003cp\u003eWe collected 450K or EPIC Illumina cytosine-guanine dinucleotides (CpGs) methylation signature data to assemble the various sample sets analyzed in this study, as detailed below. All data were processed as beta-value CSV files. To ensure consistency, EPIC beta-value probe sets were overlapped with the 450K beta-value probe set, and only common probes were retained for downstream analysis.\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eGBM datasets for deconvolution analysis\u003c/strong\u003e \u003cp\u003eWe selected the DNA methylation profiles of GBM samples classified as \"Glioblastoma, IDH-wildtype\" according to the WHO 2016 classification of CNS tumors [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e], from the Heidelberg reference cohort (GSE90496). To focus on adult GBM, we included 263 samples further sub-classified into the receptor tyrosine kinase I (RTK1; n\u0026thinsp;=\u0026thinsp;64), receptor tyrosine kinase II (RTK2; n\u0026thinsp;=\u0026thinsp;143), or mesenchymal (MES; n\u0026thinsp;=\u0026thinsp;56) subtypes, as defined by Capper et al [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e].This classification was originally derived using the DNA methylation-based brain tumor classifier (version 11.4), which is known to improve diagnostic accuracy and standardization [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e].\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eGlioma-initiating cell datasets\u003c/strong\u003e \u003cp\u003eMethylation data from a glioma-initiating cell (GIC) sample set (n\u0026thinsp;=\u0026thinsp;20) were obtained from a publicly available dataset referenced in Vinel et al [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. (GSE155985). The researchers isolated GICs from GBM samples, which were classified by methylation profiling into the RTK1, RTK2, and MES subtypes.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eGBM datasets for survival analysis and validation\u003c/b\u003e: We obtained methylation data of an independent cohort of 199 GBM cases with available clinical data from multiple sources: (1) a cohort of 84 TCGA samples diagnosed as primary GBM [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e], and (2) a cohort of 115 GBM samples from two GEO datasets (53 from GSE60274 [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e], 62 from GSE195640 [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]). We selected only patients who were treated according to the Stupp protocol, which consists of radiation therapy with concurrent and adjuvant temozolomide. All the 199 GBM cases were successfully subclassified into the RTK1 (n\u0026thinsp;=\u0026thinsp;39), RTK2 (n\u0026thinsp;=\u0026thinsp;97), and MES (n\u0026thinsp;=\u0026thinsp;63) subclasses using the DKFZ brain tumor classifier (v12.8) on the Heidelberg Epignostix Classifier platform (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://app.epignostix.com\u003c/span\u003e\u003cspan address=\"https://app.epignostix.com\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cb\u003eNon-glial tumor datasets\u003c/b\u003e: Methylation profiles of non-glial tumor sets were obtained from the TCGA database and represent the average methylation profiles of samples from four tumor types: bladder urothelial carcinoma (BLCA) (n\u0026thinsp;=\u0026thinsp;292), breast carcinoma (n\u0026thinsp;=\u0026thinsp;721), kidney renal papillary cell carcinoma (KIRP) (n\u0026thinsp;=\u0026thinsp;210), and prostate adenocarcinoma [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e] (n\u0026thinsp;=\u0026thinsp;127).\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eNormal cell datasets\u003c/strong\u003e \u003cp\u003eMethylation data (CpG beta-value CSV files) for various normal cell types, including 25 signatures from Moss et al. [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e], microglia [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e], mesenchymal stem cells (MSCs) (GSM4077441\u0026ndash;GSM4077443, GSM4078810\u0026ndash;GSM4078818), astrocytes (GSM3938231), oligodendrocytes [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e], and cortical neurons (GSE98203) were sourced from multiple databases and studies.\u003c/p\u003e \u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eConstruction of GBM Methylation Profile Atlases\u003c/h2\u003e \u003cp\u003e \u003cstrong\u003eConstructing the reference atlases\u003c/strong\u003e \u003cp\u003eWe constructed reference atlases by assembling the methylation profiles of the relevant cell types for each atlas. To identify tissue-specific CpG sites within each atlas, we applied the feature selection method described by Moss et al [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. Briefly, CpGs (based on hg19 coordinates) with low variance (\u0026lt;\u0026thinsp;0.1%) across the methylation atlas or missing values were excluded. Methylation values for each CpG across cell types were normalized by their sum, and the top 100 hypermethylated CpGs per cell type were selected based on specificity. A similar procedure was applied to the reversed methylation matrix to identify hypomethylated CpGs. For each cell type, both the top 100 hypermethylated and hypomethylated CpGs, along with neighboring CpGs within 50 bp, were included in the reference matrix. To further refine the feature set, pairwise-specific CpGs were iteratively selected by projecting the atlas onto the current CpG set, calculating Euclidean distances between cell types, and adding CpGs that best distinguished the most similar pairs at each iteration.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eRefining the reference atlases\u003c/b\u003e: We used two reference atlases for GBM deconvolution analysis. The initial reference atlas included 16 cell type signatures: B-cells, CD4 T-cells, CD8 T-cells, NK cells, neutrophils, vascular endothelial cells, monocytes, microglia, MSCs, cortical neurons, astrocytes, oligodendrocytes, NPs, OLIG2 cells, OPCs, and pre-oligodendrocytes, covering a total of 4,712 CpG sites. Based on the results from deconvolving the Heidelberg set of 263 GBM samples, we refined this atlas to include 14 cell type signatures by merging OLIG2 cells, OPCs, and pre-oligodendrocytes into a single oligodendrocyte-lineage precursor component (designated as OPs), thus reducing the total to 4,111 CpG sites. This refined atlas was subsequently used to deconvolve the Heidelberg GBM dataset, the TCGA and GEO GBM datasets, and the GIC datasets.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eDeconvolution of Methylation Signatures Using Non-Negative Least Squares Linear Regression\u003c/h3\u003e\n\u003cp\u003eNon-negative least squares linear regression was employed, as described by Moss et al. [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e], to deconvolve the GBM methylation signature beta-value matrices into cell-type components, using the relevant reference atlas.\u003c/p\u003e\n\u003ch3\u003eValidation of the Deconvolution Method\u003c/h3\u003e\n\u003cp\u003eTo validate the ability of the methylation-based deconvolution method to correctly identify expected cell type signatures, we applied it to methylation data from non-GBM sources using a custom reference atlas we constructed, comprising cell type signatures relevant to the tested samples, as follows:\u003c/p\u003e \u003cp\u003e \u003cb\u003eDeconvolution of non-glial tumors\u003c/b\u003e: We applied methylation-based deconvolution to non-glial tumors, including BLCA, breast carcinoma, KIRP, and PRAD. The reference atlas used for this analysis comprised 25 cell type signatures, sourced from publicly available data by Moss et al [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. and spanning 7,390 CpG sites. The results showed that the predominant cell type in each tumor corresponded to the expected tissue of origin: bladder cells (74%) in BLCA, breast cells (62%) in breast cancer, kidney cells (79%) in KIRP, and prostate cells (90%) in PRAD (Figure \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cb\u003eDeconvolution of normal cell types\u003c/b\u003e: We applied methylation-based deconvolution to 11 samples representing various normal cell types. The reference atlas used in this analysis included 11 cell type signatures derived from publicly available data deposited by Moss et al.[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e], using an N-1 approach: for each cell type, one corresponding sample was excluded from the atlas for validation and subsequently deconvolved using the remaining data. The atlas comprised a total of 3,274 CpG sites. The analysis showed that over 90% of the estimated cell type proportions in each sample matched the expected normal cell type. For example, the cortical neuron sample was estimated to contain 100% cortical neurons (Figure \u003cspan refid=\"MOESM2\" class=\"InternalRef\"\u003eS2\u003c/span\u003e).\u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eAssessment of Tumor Purity\u003c/h2\u003e \u003cp\u003eTumor purity was calculated using the \u003cem\u003eRF_purify\u003c/em\u003e R package (v0.1.2) with the ABSOLUTE method. This method analyzes Illumina methylation data to estimate purity by applying a Random Forest machine learning algorithm trained on copy number variations inferred from methylation patterns [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eSurvival Analysis\u003c/h2\u003e \u003cp\u003eKaplan\u0026ndash;Meier analysis and log-rank tests were conducted on 199 GBM samples using the \u003cem\u003esurvminer\u003c/em\u003e (v0.5.0) and \u003cem\u003esurvival\u003c/em\u003e (v3.7-0) R packages [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. We assessed differences in overall survival based on the median proportion of each cell type across all samples. Patients with proportions greater than the median were classified as \"high,\" while those with proportions less than the median were classified as \"low,\" and their survival outcomes were compared. Astrocytes and oligodendrocytes showed significant overall differences in terms of survival, with opposite survival trends: higher proportions of astrocytes were associated with higher overall survival, while higher proportions of oligodendrocytes were associated with lower overall survival. To assess the prognostic significance of the ratio of astrocyte-like and oligodendrocyte-like signatures within the neoplastic fraction, we determined the optimal cutoff for the astrocyte-to-oligodendrocyte ratio. The optimal cutoff was established by evaluating the ratio values across the entire range, with the threshold identified as the point yielding the most significant log-rank test result. Based on this cutoff, the groups were dichotomized into those with \u0026ldquo;high\u0026rdquo; and \u0026ldquo;low\u0026rdquo; astrocyte-to-oligodendrocyte ratios, with the groups exhibiting distinct survival probabilities that were consistent with previous reports using different methodologies [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eTo validate this cutoff and address potential overfitting, we applied a permutation-based approach. That is, the astrocyte-to-oligodendrocyte ratio was randomly shuffled across patients (100,000 iterations) while maintaining survival times and censoring status, and log-rank p-values were recomputed for each permutation. The permutation p-value was calculated as the proportion of permuted p-values that were less than the observed p-value, with statistical significance defined at p\u0026thinsp;\u0026lt;\u0026thinsp;0.05. A low permutation p-value indicates that the observed p-value is unlikely under randomized conditions, reflecting the robustness of the selected cutoff. This approach confirms the cutoff\u0026rsquo;s reliability and significance and excludes any effects attributable to chance or dataset-specific overfitting.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eStatistical Analysis\u003c/h2\u003e \u003cp\u003eTo assess significant differences in cell type proportions across GBM subclasses, pairwise Student\u0026rsquo;s two-tailed \u003cem\u003et\u003c/em\u003e-tests were conducted for each subclass pair. Two-tailed \u003cem\u003et\u003c/em\u003e-tests were also used to evaluate the statistical significance of Pearson correlation coefficients for correlations between cell type proportions and purity scores. The \u003cem\u003et\u003c/em\u003e-tests were performed using R, Python, or Microsoft Excel with built-in \u003cem\u003et\u003c/em\u003e-test functions. For survival analysis, the log-rank test was applied as previously described.\u003c/p\u003e \u003cp\u003eTo compare the expected cell-type proportions between the results of survival analysis of the GBM set (n\u0026thinsp;=\u0026thinsp;199) and the observed results from the initial GBM set (n\u0026thinsp;=\u0026thinsp;263), we performed a chi-square goodness-of-fit test using the \u003cem\u003echisq.test\u003c/em\u003e function in R (with the following parameters: \u003cem\u003esimulate.p.value\u0026thinsp;=\u0026thinsp;TRUE, B\u0026thinsp;=\u0026thinsp;2000).\u003c/em\u003e\u003c/p\u003e \u003cp\u003eFor comparisons of subclass distributions between patient groups classified according to survival probability, we used the chi-square test for independence with the \u003cem\u003echisq.test\u003c/em\u003e function in R.\u003c/p\u003e \u003cp\u003eA significance threshold of 0.05 was used for all statistical tests.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eGraph and Plots\u003c/h2\u003e \u003cp\u003eAll graphs and plots were generated using R (v4.0.3+), Python 3.11, Microsoft Excel, or BioeRender.com.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eGeneration of NPs and Human Oligodendrocyte Lineage Precursor Cells\u003c/h2\u003e \u003cp\u003eThrough immunostaining studies, we were able to confirm that NPs were successfully derived from hESCs and subsequently differentiated into oligodendrocyte lineage precursor cells through a staged process (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e1\u003c/span\u003e). These NPs (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e1\u003c/span\u003eA) then differentiated into OLIG2 cells (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e1\u003c/span\u003eB), which further differentiated into cells of the oligodendroglial lineage, which includes enriched populations of OPCs (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e1\u003c/span\u003eC) that subsequently differentiated into pre-oligodendrocytes (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e1\u003c/span\u003eD). These marker studies confirm the progression of the initial NPs along the expected developmental trajectory.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eIdentification of Cell Type Composition and Proportions in GBM and Its Subclasses\u003c/h2\u003e \u003cp\u003e \u003cb\u003eReference atlases of the cellular methylation profiles\u003c/b\u003e: To investigate the cell type composition in GBM, we constructed a reference atlas of DNA methylation signatures critical for the deconvolution of GBM samples and quantification of cell type proportions. This atlas encompassed methylation data from 16 distinct cell types derived from our previously published data [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e] and several publicly available datasets: cortical neurons, astrocytes, oligodendrocytes, various immune cells (B cells, CD4⁺ T cells, NK cells, CD8⁺ T cells, and neutrophils), tumor-associated macrophages (TAMs: monocytes and microglia [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]), and MSCs. To account for the potential differentiation states of cancer cells, we further extended the atlas by incorporating methylation profiles from our in vitro-generated NPs and three defined stages within the oligodendrocyte lineage. These stages represent a continuum of differentiation and include OLIG2-positive progenitors (characterized by expression of the OLIG2 transcription factor essential for oligodendrocyte development), OPCs, and pre-oligodendrocytes identified by O4 expression. This atlas was generated by selecting distinct differentially methylated CpGs for each cell type, following previously described and validated methodologies [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. In total, it comprises 4,712 CpG sites across these cell types (\u003cb\u003eFigure S3\u003c/b\u003e).\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eOverall cellular composition of GBM samples\u003c/strong\u003e \u003cp\u003eTo estimate the cellular composition of bulk GBM tumor samples, we applied a deconvolution method adapted from cell-free DNA tissue-origin studies [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. Using 450K methylation data, we analyzed 263 adult GBM samples, classified according to their methylation profiles as previously described [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e].\u003c/p\u003e \u003c/p\u003e \u003cp\u003eIn the GBM samples, glial cells formed the majority (62%) of the cellular component, including contributions from oligodendrocyte-like (26.3%), astrocyte-like (13.4%), and pre-oligodendrocyte-like (O4+, 22%) components. The cortical neuron signature was minimally represented (2.6%). Additionally, GBM displayed 7.3% MSC-like cells, 7.8% TAMs comprising microglia (3.6%) and monocytes (4.2%), 13% vascular endothelial cells, and 7.2% immune cells (B-cells, CD4 T-cells, NK cells, CD8 T-cells, and neutrophils). NPs effectively showed no contribution (\u003cb\u003eFigure S4\u003c/b\u003e).\u003c/p\u003e \u003cp\u003eTo further refine the analysis of GBM cell type composition, OLIG2 cells, OPCs, and pre-oligodendrocytes were merged into a single component termed OPs. This adjustment was necessary because OLIG2 cells and OPCs contributed minimally to the GBM cell populations (average, 0.3\u0026thinsp;\u0026plusmn;\u0026thinsp;1.2%). Thus, their integration ensured representation while preserving the accuracy of cell type contribution. The final reference atlas included 14 cell types (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e2\u003c/span\u003e). This adjustment had minimal impact on the estimated proportions of astrocytes, cortical neurons, MSCs, TAMs, vascular endothelial cells, and immune cells. However, a slight difference emerged in the oligodendrocyte lineage, with the proportion of oligodendrocytes increasing from 26.3\u0026ndash;29.5%, while the proportion of pre-oligodendrocytes, which accounted for 22% in the separated analysis, was reduced to 18.9% in the combined OPs component (Figs.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e3\u003c/span\u003eA \u0026amp; S4).\u003c/p\u003e \u003cp\u003e \u003cb\u003eComparison of cellular composition between GBM subclasses\u003c/b\u003e: Our findings revealed significant variation in cell type proportions among GBM subclasses. Specifically, the astrocyte proportions were significantly lower in the MES (10.5%) and RTK1 (12.8%) subtypes than in RTK2 subtype (12.6%) (RTK1 vs. RTK2: p\u0026thinsp;\u0026lt;\u0026thinsp;0.01; MES vs. RTK2: p\u0026thinsp;\u0026lt;\u0026thinsp;1.60E-07), but were not significantly different between the RTK1 and MES subtypes (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e3\u003c/span\u003eD). Oligodendrocyte proportions also showed significant pairwise differences in all comparisons: RTK1 vs. RTK2: 39% vs. 27% (p\u0026thinsp;\u0026lt;\u0026thinsp;1.012E-32), RTK1 vs. MES: 39% vs. 24.6% (p\u0026thinsp;\u0026lt;\u0026thinsp;2.39E-20), and RTK2 vs. MES: 27% vs. 24.6%, p\u0026thinsp;\u0026lt;\u0026thinsp;1.85E-03) (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e3\u003c/span\u003eB\u003cb\u003e)\u003c/b\u003e. Additionally, significant differences were observed in the proportions of other cell types, including OPs, MSCs, immune cells, and TAMs. All p-values were calculated using a two-tailed \u003cem\u003et\u003c/em\u003e-test \u003cb\u003e(\u003c/b\u003eFigs.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e3\u003c/span\u003eB\u0026ndash;J\u003cb\u003e)\u003c/b\u003e.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003eAssessment of the Neoplastic and Non-Neoplastic Components of GBM Using RF_Purify\u003c/h2\u003e \u003cp\u003eThe neoplastic fraction, estimated from the 263 GBM samples using methylation-based deconvolution, was 70% \u0026plusmn; 11%. Using the cell type proportions determined from the deconvolution analysis, we aimed to distinguish between the neoplastic and non-neoplastic components in GBM tumors. This was done using the RF_Purify method, which applies the ABSOLUTE score to infer tumor purity [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. We correlated purity scores with the proportions of seven predominant cell types identified in the deconvolution analysis, excluding cortical neurons (average, 2%) and NPs (average, 0%) due to their minimal contribution. The analyzed components included immune cells, TAMs, vascular endothelial cells, MSCs, astrocytes, OPs, and oligodendrocytes (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eSignificant negative correlations were observed between tumor purity and the proportions of immune cells (R = -0.60, p\u0026thinsp;\u0026lt;\u0026thinsp;1.E-26), TAMs (R = -0.76, p\u0026thinsp;\u0026lt;\u0026thinsp;3.E-51), and vascular endothelial cells (R = -0.20, p\u0026thinsp;\u0026lt;\u0026thinsp;1.E-03) (Figs.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e4\u003c/span\u003eA, \u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e4\u003c/span\u003eE). In contrast, tumor purity was correlated positively with three glial lineage cells, that is, astrocytes (R\u0026thinsp;=\u0026thinsp;0.33, p\u0026thinsp;\u0026lt;\u0026thinsp;6.E-08), OPs (R\u0026thinsp;=\u0026thinsp;0.68, p\u0026thinsp;\u0026lt;\u0026thinsp;1.E-37), and oligodendrocytes (R\u0026thinsp;=\u0026thinsp;0.41, p\u0026thinsp;\u0026lt;\u0026thinsp;6.E-12), and with MSCs (R\u0026thinsp;=\u0026thinsp;0.22, p\u0026thinsp;\u0026lt;\u0026thinsp;4.E-04) (Figs.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e4\u003c/span\u003eB, \u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e4\u003c/span\u003eE). When the proportions of all four positively correlated components (three glial lineage cell types and MSCs) were combined additively as a single component, the positive correlation with tumor purity was stronger (R\u0026thinsp;=\u0026thinsp;0.79, p\u0026thinsp;\u0026lt;\u0026thinsp;2.E-56). This implies that these components collectively contribute substantially to the neoplastic portion of the tumor (Figs.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e4\u003c/span\u003eD, \u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e4\u003c/span\u003eE). On the other hand, when the proportions of all the negatively correlated components (immune cells, TAMs, and vascular endothelial cells) were combined additively as a single component, the negative correlation with tumor purity was stronger (R = -0.81, p\u0026thinsp;\u0026lt;\u0026thinsp;3.E-62). Thus, these components collectively contribute substantially to the non-neoplastic normal portion of the tumor (Figs.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e4\u003c/span\u003eC, \u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e4\u003c/span\u003eE). All p-values were calculated using a two-tailed correlation \u003cem\u003et\u003c/em\u003e-test.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003eValidation of Purity Analysis: Comparison of Deconvolution Results in Bulk GBM Tumors Versus GICs\u003c/h2\u003e \u003cp\u003eThe three glial cell types and MSCs identified in the neoplastic fraction of GBM were validated by deconvolution analysis of the DNA methylation profiles of CD133 (Prominin-1)-enriched GICs, which represent the neoplastic compartment of the tumor. These profiles were obtained from a previously published study [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. and derived from GBM tumors classified as RTK1, RTK2, and MES (see Methods). This comparison allowed us to evaluate whether the cell-type signals associated with the neoplastic fraction in bulk tumors are retained in a highly purified population of tumor cells.\u003c/p\u003e \u003cp\u003eAs expected, deconvolution of GICs revealed that they comprised a significantly higher tumor fraction composed of the three glial lineage cell types and MSCs compared to GBM bulk tumors, with 88% in GICs versus 70% in GBM (p\u0026thinsp;\u0026lt;\u0026thinsp;2.2E-20, two-tailed \u003cem\u003et\u003c/em\u003e-test, \u003cb\u003eFigure S5\u003c/b\u003e). Specifically, GICs exhibited significantly increased proportions of astrocytes (19% vs. 13% in GBM samples, p\u0026thinsp;\u0026lt;\u0026thinsp;8.7E-04), OPs (26% vs. 19% in GBM, p\u0026thinsp;\u0026lt;\u0026thinsp;3.1E-05), and MSCs (16% vs. 8% in GBM, p\u0026thinsp;\u0026lt;\u0026thinsp;5.6E-11), while the oligodendrocyte proportions remained similar between GICs and GBM (28% vs. 30%, p\u0026thinsp;\u0026lt;\u0026thinsp;0.21). In contrast, the proportions of vascular endothelial cells, immune cells, and TAMs were significantly lower in GICs than in GBM bulk tumors (vascular endothelial cells: 7% vs. 14%, p\u0026thinsp;\u0026lt;\u0026thinsp;2.4E-09; immune cells: 3% vs. 7%, p\u0026thinsp;\u0026lt;\u0026thinsp;1.1E-07; TAMs: 0% vs. 8%, p\u0026thinsp;\u0026lt;\u0026thinsp;6.5E-54) (\u003cb\u003eFigure S5\u003c/b\u003e). All p values were calculated using a two-tailed \u003cem\u003et\u003c/em\u003e-test. These findings demonstrate that cell types positively correlated with tumor purity in GBM (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e4\u003c/span\u003eB) are more abundant in GICs, whereas those negatively correlated with tumor purity (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e4\u003c/span\u003eA) are present at lower proportions.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003eCorrelation Between Cell Type Proportions and Patient Survival: Prognostic Significance of the Astrocyte-to-Oligodendrocyte Ratio\u003c/h2\u003e \u003cp\u003e \u003cstrong\u003eOverall trends in associations between cell types and survival probability\u003c/strong\u003e \u003cp\u003eTo evaluate whether the abundance of specific cell types is associated with patient survival, we analyzed 199 GBM samples with available methylation clinical data from the TCGA and GEO databases. Using our methylation-based deconvolution approach, we estimated cell type proportions in this cohort and confirmed consistency with the original set of 263 samples (chi-square goodness-of-fit test, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.96; Figure S6). We then assessed the association between cell type proportions and overall survival using Kaplan\u0026ndash;Meier analysis (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e5\u003c/span\u003e).\u003c/p\u003e \u003c/p\u003e \u003cp\u003eWe found that oligodendrocyte proportions lower than the median (26.4%) were significantly associated with higher overall survival probability (OSP), whereas higher proportions were linked to lower OSP (median survival time [MST]: 15.3 vs. 14.3 months; median survival time difference [MSTD]: 1 month; p\u0026thinsp;\u0026lt;\u0026thinsp;0.018, log-rank test; Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e5\u003c/span\u003eA). In contrast, higher astrocyte proportions (above the median of 10.6%) were associated with improved OSP, while lower proportions were correlated with shorter OSP (MST: 15.3 vs. 13.4 months; MSTD: 1.9 months; p\u0026thinsp;\u0026lt;\u0026thinsp;0.044, log-rank test; Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e5\u003c/span\u003eB\u003cb\u003e).\u003c/b\u003e A similar trend was observed for microglia, where proportions above the median (3.2%) were correlated with increased OSP compared to lower proportions (MST: 15.7 vs. 14.3 months; MSTD: 1.4 months; p\u0026thinsp;\u0026lt;\u0026thinsp;0.007, log-rank test; \u003cb\u003eFigure S7B\u003c/b\u003e). No significant correlations were found for other cell types or cell type groups (immune cells, TAMs, vascular endothelial cells, OPs, and MSCs) (\u003cb\u003eFigure S7A-F\u003c/b\u003e).\u003c/p\u003e \u003cp\u003e \u003cstrong\u003ePrognostic significance of the astrocyte-to-oligodendrocyte ratio\u003c/strong\u003e \u003cp\u003eBased on the contrasting effects of astrocyte and oligodendrocyte abundance on OSP, and considering the reports of previous studies that highlight dynamic shifts in cellular states within the GBM neoplastic fraction [\u003cspan additionalcitationids=\"CR3\" citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e], we focused on the impact of the astrocyte-to-oligodendrocyte ratio on survival. The median ratio was used to effectively stratify patients into high- and low-survival groups. A ratio above the median (\u0026gt;\u0026thinsp;0.41) was significantly associated with increased OSP (MST\u0026thinsp;=\u0026thinsp;15.7 months), whereas a lower ratio corresponded to reduced OSP (MST\u0026thinsp;=\u0026thinsp;14.3 months), with an MSTD of 1.44 months (p\u0026thinsp;\u0026lt;\u0026thinsp;0.03, log-rank test, Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e5\u003c/span\u003eC).\u003c/p\u003e \u003c/p\u003e \u003cp\u003eTo evaluate the astrocyte/oligodendrocyte ratio as a prognostic biomarker, we determined the optimal cutoff for classifying the survival groups (see Methods). A cutoff of 0.3, corresponding to a ratio of 1:3.3, was determined. Patients with a ratio above this threshold exhibited significantly improved OSP (MST\u0026thinsp;=\u0026thinsp;15.8 months), whereas those with a lower ratio had reduced OSP (MST\u0026thinsp;=\u0026thinsp;11.9 months), with an MSTD of 3.9 months (p\u0026thinsp;\u0026lt;\u0026thinsp;0.00011, log-rank test, Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e5\u003c/span\u003eD). Notably, the average overall survival differed by 6 months between the \u0026ldquo;high\u0026rdquo; and \u0026ldquo;low\u0026rdquo; ratio groups. The higher ratio group had an average overall survival of 18.1 months, while the lower ratio group had an average overall survival of 11.6 months.\u003c/p\u003e \u003cp\u003eTo validate this cutoff and mitigate potential overfitting, we applied a permutation-based approach with 100,000 random iterations of the ratio data (see Methods). The permutation p-value was 0.0001, which indicates the proportion of permuted p-values\u0026thinsp;\u0026le;\u0026thinsp;the observed p-value (0.00011). This validation underscores the robustness of the cutoff and excludes random effects or dataset-specific overfitting.\u003c/p\u003e \u003cp\u003eGiven the impact of the astro/oligo ratio on overall survival and the differences in astrocyte and oligodendrocyte proportions observed across the three GBM subclasses, RTK1, RTK2, and MES (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e3\u003c/span\u003e), we examined their association with survival outcomes to determine whether the observed survival differences were driven by a specific subclass.\u003c/p\u003e \u003cp\u003eWe found that patients with an astrocyte/oligodendrocyte ratio above the defined cutoff, which was associated with long-term survival, were predominantly from the RTK2 subclass (64%), whereas only 7% belonged to the RTK1 subclass. In contrast, patients with a ratio below the cutoff, linked to short-term survival, had a higher proportion of the RTK1 subclass (43%) than the RTK2 subclass (20%). Unlike the RTK1 and RTK2 subclasses, which exhibited opposite distributions between the long-term survival and short-term survival groups, the MES subclass did not follow this pattern (\u003cb\u003eFigure S8\u003c/b\u003e). However, a survival analysis based solely on the GBM subclasses did not reveal significant differences in overall prognosis (\u003cb\u003eFigure S7G\u003c/b\u003e).\u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eOur study establishes methylation-based deconvolution as a powerful tool for unraveling GBM\u0026rsquo;s cellular composition by shedding light on tumor heterogeneity and linking cell type proportions to patient survival. By leveraging a reference atlas of methylation profiles from GBM microenvironment cell types, particularly through the inclusion of in vitro-differentiated human oligodendrocyte-lineage precursors, we successfully delineated the neoplastic cell composition of GBM using DNA methylation-based deconvolution. To our knowledge, this is the first time the cell composition of the neoplastic component of GBM has been investigated through methylation-based deconvolution, a result previously demonstrated only with techniques such as RNA-seq [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan additionalcitationids=\"CR21\" citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Moreover, the results were comparable to those of scRNA-seq analysis, while offering certain advantages such rapid speed of analysis, lower cost, and better feasibility and applicability, as it is compatible even with formalin-fixed paraffin-embedded and archived samples. Importantly, our approach enabled the identification of key cellular components within the tumor microenvironment and highlighted the astrocyte-to-oligodendrocyte ratio as a significant prognostic indicator in GBM.\u003c/p\u003e \u003cp\u003eWith regard to the proportions of different cell types, glial cells dominated the GBM composition (62%), which displayed primarily oligodendrocyte-like, oligodendrocyte precursor-like and astrocytes-like signatures, while cortical neurons contributed minimally. Additionally, we identified the presence of MSC-like cells, vascular endothelial cells, TAMs, and immune cells, further underscoring the complexity of the GBM microenvironment. The proportions of vascular endothelial cells were markedly higher in our analysis, averaging 13.9% (\u0026plusmn;\u0026thinsp;4.5%) compared to 1.7% (range 0.3\u0026ndash;2.8%) reported in prior studies [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan additionalcitationids=\"CR21\" citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. It should be noted that a recent report indicated 9.9% vascularization in GBM, which is closer to our observed proportion of 13.9% [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. This finding underscores the well-documented angiogenic nature of GBM[\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e] and reinforces the reliability of our approach in capturing GBM\u0026rsquo;s vascular complexity. While most of the cell type proportions were broadly comparable to earlier findings,[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan additionalcitationids=\"CR21\" citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e] a few notable differences emerged (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e and Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e). For instance, we observed a slightly higher proportion of immune cells (7.3% \u0026plusmn; 5.4%) relative to the average of 3.3% reported elsewhere. These discrepancies may reflect differences in sample collection, tissue processing, or methodological biases. In particular, certain cell types tend to be underrepresented in single-cell RNA-seq due to their fragility or dissociation inefficiency. Despite these variations, our findings reinforce the robustness of methylation-based deconvolution in capturing the major cellular components of GBM [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan additionalcitationids=\"CR21\" citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e].\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComparison of Methylation-Based Deconvolution Results Across Studies\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"16\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c11\" colnum=\"11\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c12\" colnum=\"12\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c13\" colnum=\"13\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c14\" colnum=\"14\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c15\" colnum=\"15\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c16\" colnum=\"16\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFraction\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eCell types\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003eNeftel et al.\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eYu et al.\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eDarmanis et al.\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c9\" namest=\"c8\"\u003e \u003cp\u003eYuan et al.\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c11\" namest=\"c10\"\u003e \u003cp\u003eSingh et al.\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c13\" namest=\"c12\"\u003e \u003cp\u003eOther studies (avg.)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c15\" namest=\"c14\"\u003e \u003cp\u003eOur study (avg.)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"1\" nameend=\"c16\" namest=\"c16\"\u003e\u0026nbsp;\u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"8\" rowspan=\"9\"\u003e \u003cp\u003eNon-Neoplastic fraction\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eB cells\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c9\" namest=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c11\" namest=\"c10\"\u003e \u003cp\u003e1.1%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c13\" namest=\"c12\"\u003e \u003cp\u003e1.1%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c15\" namest=\"c14\"\u003e \u003cp\u003e3.4%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"1\" nameend=\"c16\" namest=\"c16\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eCD4T cells\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c9\" namest=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c11\" namest=\"c10\"\u003e \u003cp\u003e0.2%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c13\" namest=\"c12\"\u003e \u003cp\u003e0.2%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c15\" namest=\"c14\"\u003e \u003cp\u003e0.3%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"1\" nameend=\"c16\" namest=\"c16\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eNK-cells\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003e0.0%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.4%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c9\" namest=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c11\" namest=\"c10\"\u003e \u003cp\u003e0.4%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c13\" namest=\"c12\"\u003e \u003cp\u003e0.6%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c15\" namest=\"c14\"\u003e \u003cp\u003e0.4%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"1\" nameend=\"c16\" namest=\"c16\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eCD8T cells\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003e1.2%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e3.0%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c9\" namest=\"c8\"\u003e \u003cp\u003e0.5%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c11\" namest=\"c10\"\u003e \u003cp\u003e1.3%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c13\" namest=\"c12\"\u003e \u003cp\u003e1.5%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c15\" namest=\"c14\"\u003e \u003cp\u003e1.1%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"1\" nameend=\"c16\" namest=\"c16\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eNeutrophils\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003e0.1%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2.8%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c9\" namest=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c11\" namest=\"c10\"\u003e \u003cp\u003e1.1%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c13\" namest=\"c12\"\u003e \u003cp\u003e1.3%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c15\" namest=\"c14\"\u003e \u003cp\u003e2.2%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"1\" nameend=\"c16\" namest=\"c16\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eTAMs\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003e5.4%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e6.4%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e7.7%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c9\" namest=\"c8\"\u003e \u003cp\u003e0.2%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c11\" namest=\"c10\"\u003e \u003cp\u003e7.5%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c13\" namest=\"c12\"\u003e \u003cp\u003e5.4%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c15\" namest=\"c14\"\u003e \u003cp\u003e7.8%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"1\" nameend=\"c16\" namest=\"c16\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eCortical neurons\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e1.4%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c9\" namest=\"c8\"\u003e \u003cp\u003e0.0%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c11\" namest=\"c10\"\u003e \u003cp\u003e0.2%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c13\" namest=\"c12\"\u003e \u003cp\u003e0.5%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c15\" namest=\"c14\"\u003e \u003cp\u003e1.5%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"1\" nameend=\"c16\" namest=\"c16\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eNormal Oligodendrocytes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003e2.8%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e9.1%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e6.9%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c9\" namest=\"c8\"\u003e \u003cp\u003e1.6%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c11\" namest=\"c10\"\u003e \u003cp\u003e6.2%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c13\" namest=\"c12\"\u003e \u003cp\u003e5.3%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c15\" namest=\"c14\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"1\" nameend=\"c16\" namest=\"c16\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eNormal Astrocytes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.6%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e1.4%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c9\" namest=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c11\" namest=\"c10\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c13\" namest=\"c12\"\u003e \u003cp\u003e1.0%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c15\" namest=\"c14\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"1\" nameend=\"c16\" namest=\"c16\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eNeoplastic fraction\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003e87%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e \u003cp\u003e52%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c8\" namest=\"c7\"\u003e \u003cp\u003e59%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c10\" namest=\"c9\"\u003e \u003cp\u003e83%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c12\" namest=\"c11\"\u003e \u003cp\u003e78%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c14\" namest=\"c13\"\u003e \u003cp\u003e72%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c16\" namest=\"c15\"\u003e \u003cp\u003e70%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"16\"\u003eSummary of the proportions of various cell types and neoplastic fraction identified in our study (the right most column) compared to those reported in other studies and their average. The left four data columns represent the scRNA-seq-based studies: Neftel et al.[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e], Yu et al.[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e], Darmanis et al.[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e] and Yuan et al.[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e] The fifth column on the right displays findings from methylation-based analysis by Singh et al.[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. The two rightmost columns display the averages from other studies and our study.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eTo distinguish neoplastic from non-neoplastic components, we used the RF_Purify method, which leverages the ABSOLUTE score to infer tumor purity. The neoplastic fraction, estimated from the 263 GBM samples using methylation-based deconvolution, was 70% \u0026plusmn; 11%. This value demonstrates strong concordance with prior estimates of 72% (range, 52\u0026ndash;87%) derived primarily from scRNA-seq [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan additionalcitationids=\"CR21\" citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e] (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e and Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e). Thus, overall, the tumor purity levels are highly consistent across methods. Further, this analysis confirmed that the glial cells OPs, astrocytes, and oligodendrocytes, along with MSCs, constitute the neoplastic fraction. This is line with previous scRNA-seq studies [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan additionalcitationids=\"CR21\" citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Moreover, immune cells, TAMs, and vascular endothelial cells predominantly belong to the tumor microenvironment, as also observed in previous studies [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan additionalcitationids=\"CR21\" citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. These tumor purity data were also validated through deconvolution analysis of a GIC dataset that provided an independent validation of the purity analysis in bulk GBM tumors. Cell types positively correlated with tumor purity in the bulk GBM dataset results were enriched in GICs, while those negatively correlated, probably representing non-malignant or microenvironmental cells, were reduced. Thus, the biological validity of the deconvolution approach in terms of effectively distinguishing malignant from non-malignant components was confirmed.\u003c/p\u003e \u003cp\u003eWe demonstrated that the oligodendrocyte differentiation lineage accounts for approximately 50% of the bulk tumor and a striking 71% of the neoplastic fraction. These findings extend prior work by Liu et al. [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e], Persson et al [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e], and Ligon et al.[\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e], which showed that oligodendrocyte precursor cells can give rise to GBM under defined oncogenic conditions. By revealing that oligodendrocyte-like and oligodendrocyte precursor-like cells comprise 71% of the neoplastic fraction, our results strongly reinforce the involvement of the oligodendrocyte lineage in GBM pathogenesis and suggest that pre-oligodendrocytes may serve as cells of origin for GBM.\u003c/p\u003e \u003cp\u003eTo evaluate the accuracy of our neoplastic cell composition estimates, we applied DNA methylation-based deconvolution to CD133-enriched tumor dissociates from 20 GBM patients [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]. As expected, the neoplastic fraction was higher than in that in bulk tumors (~\u0026thinsp;90% vs. ~70%), consistent with the enrichment results for cancer cells. The remaining\u0026thinsp;~\u0026thinsp;10% consisted largely of endothelial cells (~\u0026thinsp;6.8%), which probably reflecting GBM\u0026rsquo;s prominent vasculature and minor contamination from tumor-associated blood vessels during CD133\u0026thinsp;+\u0026thinsp;cell enrichment. Significant differences in cell-type proportions were observed among GBM subclasses RTK1, RTK2, and MES, including variations in TAMs and immune cell fractions (e.g., B cells, CD4\u0026thinsp;+\u0026thinsp;T cells, NK cells, CD8\u0026thinsp;+\u0026thinsp;T cells, and neutrophils). These findings are consistent with transcriptomic and immunohistochemistry analyses, which report increased tumor-infiltrating lymphocytes and TAMs in the mesenchymal GBM subtype compared to non-mesenchymal subtypes [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]. Our results extend these subclassifications by revealing distinct differences in cellular composition detected through methylation-based deconvolution.\u003c/p\u003e \u003cp\u003eWe observed distinct astrocyte and MSC methylation signatures in the neoplastic fraction of GBM that correspond to the astrocyte-like and mesenchymal-like states, respectively, identified by scRNA-seq [\u003cspan additionalcitationids=\"CR4\" citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e] as part of the neoplastic fraction of GBM. Moreover, our analysis indicates that early oligodendrocyte lineage stages (OLIG2⁺ or OPC-like cells) and NPs contribute minimally to GBM composition. Instead, most oligodendrocyte-lineage cells display advanced differentiation, as reflected by pre-oligodendrocyte (O4-positive) and mature oligodendrocyte methylation signatures.\u003c/p\u003e \u003cp\u003eIn alignment with prior research [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e], our findings indicate that astrocyte-like and oligodendrocyte-like signatures constitute a significant part of the GBM neoplastic fraction. Further, the consistently reported low abundance of normal astrocytes and oligodendrocytes in GBM samples [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan additionalcitationids=\"CR21\" citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e] supports the conclusion that the elevated proportions we detected primarily reflect the neoplastic component. A key finding of our study is the significant association of GBM cellular composition, particularly the astrocyte and oligodendrocyte proportions, with patient survival. More specifically, higher proportions of astrocytes were linked to improved overall survival, while increased proportions of oligodendrocytes and OPs were associated with poorer outcomes. These results are in alignment with previous studies which showed that the OPC- and NPC-like tumor states are associated with the highest proliferative potential, while the astrocyte-like states are associated with the lowest proliferative potential [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eWe identified the astrocyte-to-oligodendrocyte ratio as a key prognostic biomarker. This aligns with the findings that the astrocyte-like state has reduced tumor-initiating potential compared to oligodendrocyte-lineage components, which exhibit greater tumorigenicity in mouse models [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Although the RTK1, RTK2, and MES subtypes exhibited distinct cellular compositions, these differences did not translate into significant overall survival variations across subtypes, as reported in other studies [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e]. Instead, as suggested above, survival differences were driven by the astrocyte-to-oligodendrocyte ratio. This observation suggests that intertumoral cellular heterogeneity may be a more powerful prognostic indicator than subclassification alone. Nevertheless, patients with lower survival times, as determined by a higher astrocyte-to-oligodendrocyte ratio, had a greater proportion of GBM tumors classified as RTK2 than RTK1, and vice versa. Since RTK2 showed higher proportions of astrocytes and lower proportions of oligodendrocytes, compared to RTK1, this may explain the survival differences between the two subclasses.\u003c/p\u003e \u003cp\u003eThis study has several limitations. First, the DNA methylation profiles of in vitro-differentiated NPs and oligodendrocyte-lineage precursors may not fully capture the methylation landscapes present in GBM tumors, potentially affecting the accuracy of our findings. Second, assigning the methylation signature of OPs and mature oligodendrocytes to the less differentiated NPC-like and OPC-like transcriptional signatures reported in scRNA-seq [\u003cspan additionalcitationids=\"CR4\" citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e] as part of the neoplastic fraction remains a challenge. The discrepancy between our findings and the previous studies [\u003cspan additionalcitationids=\"CR4\" citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e] probably stems from methodological differences between single-cell transcriptional profiling and bulk methylation analysis. Further investigations, particularly single-cell methylation studies, are needed to validate this interpretation. Despite these methodological differences, both approaches support the notion that these populations represent distinct stages within the oligodendrocyte differentiation lineages. Finally, the survival analysis was conducted on a cohort predominantly receiving standard-of-care therapy, which may limit its applicability to patients undergoing alternative treatments or to underrepresented GBM subtypes. Thus, further in-depth research is required to evaluate the broader applicability of these results.\u003c/p\u003e"},{"header":"Conclusions","content":"\u003cp\u003eDNA methylation-based deconvolution offers a robust framework for resolving the cellular composition of GBM. We identified the astrocyte-to-oligodendrocyte ratio as a significant prognostic biomarker linked to patient survival. In addition, the findings point to an oligodendrocytic cell of origin and shed further light on the pathology of this tumor. Overall, this approach has the potential to provide clinically meaningful insights from initial tumor resections while offering several advantages over current methods like scRNA-seq.\u0026nbsp;It is likely to have high applicability in routine clinical implementation, with the potential to improve patient stratification, guide personalized therapy, and ultimately, impact survival outcomes.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cdiv class=\"DefinitionList\"\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eBLCA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ebladder urothelial carcinoma\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eCNS\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ecentral nervous system\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eCpGs\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ecytosine-guanine dinucleotides\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eGBM\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eIDH-wildtype glioblastoma\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eGIC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eglioma-initiating cell\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ehESCs\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ehuman embryonic stem cells\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eKIRP\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ekidney renal papillary cell carcinoma\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMES\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003emesenchymal\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMSCs\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003emesenchymal stem cells\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMST\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003emedian survival time\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMSTD\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003emedian survival time difference\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eNPCs\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eneural progenitor cells\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eNPs\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eneural progenitors\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eOLIG2 cells\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eOLIG2-positive progenitor cells\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eOPs\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eoligodendrocyte-lineage precursor cells\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eOPCs\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eOlig2+/Nkx2.2\u0026thinsp;+\u0026thinsp;oligodendrocyte progenitor cells\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eOSP\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eoverall survival probability\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePRAD\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eprostate adenocarcinoma\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eRTK1\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ereceptor tyrosine kinase I\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eRTK2\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ereceptor tyrosine kinase II\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003escRNA-seq\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003esingle-cell RNA sequencing\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eTAMs\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003etumor-associated macrophages\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eTME\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003etumor microenvironment\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003eEthics approval and consent to participate\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAll data from human patients used in this study were obtained from online databases, including The Cancer Genome Atlas (TCGA) and GEO. Therefore, the need for informed consent and ethics approval was waived by the university.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eReference atlases used for deconvolution analysis and clinical data for the survival analysis cohort are available in the supplementary data. Accession numbers for the datasets supporting the conclusions in this article are provided in the main text.\u0026nbsp;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003eFunding\u003c/p\u003e\n\u003cp\u003eThis research was supported by a generous gift from Leslie and Michael Gaffin.\u003c/p\u003e\n\u003cp\u003eAuthor Contributions:\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eConceptualization, A.I., A.E., J.M, I.L.; methodology, A.I., N.L., J.M, BE. R, M.I, D.S, E. BS, I.L. MS; software, A.I., N.L, J.M.; validation, A.I., J.M. M.I, A.Z. I.L.; formal analysis, A.I., I.L.; investigation, A.I, I.L.; resources, A.I., H.C, M.G., M.I, D.S., EB.S, M.S, J.M, BE, R, I.L.; data curation, A.I., N.L., J.M. BE. R, M.I, D.S; writing\u0026mdash;original draft preparation, A.I., I.L.; writing-review and editing, A.I., N.L, J.M., I.L., A.L, A.M; Y.F. visualization, A.I., N.L., J.M., A.L, A.M, I.L.; supervision, I.L.; project administration, I.L.; funding acquisition, I.L. All authors have read and agreed to the published version of the manuscript.\u003c/p\u003e\n\u003cp\u003eAcknowledgments\u003c/p\u003e\n\u003cp\u003eThe results shown here are in whole or part based on data generated by the TCGA Research Network: https://www.cancer.gov/tcga.\u003c/p\u003e\n\u003cp\u003eWe sincerely thank Andreas von Deimling and his laboratory for conducting the EPIC methylation profiling of the in vitro differentiated neural progenitors [42] and oligodendrocyte precursor cells, which greatly contributed to this research.\u003c/p\u003e\n\u003cp\u003eWe also extend our heartfelt thanks to Leslie and Michael Gaffin for their generous support.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eOstrom QT, Bauchet L, Davis FG, Deltour I, Fisher JL, Langer CE, Pekmezci M, Schwartzbaum JA, Turner MC, Walsh KMet al et al (2014) The epidemiology of glioma in adults: a state of the science review. Neuro Oncol 16:896\u0026ndash;913. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/neuonc/nou087\u003c/span\u003e\u003cspan address=\"10.1093/neuonc/nou087\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePatel AP, Tirosh I, Trombetta JJ, Shalek AK, Gillespie SM, Wakimoto H, Cahill DP, Nahed BV, Curry WT Martuza RL (2014) Single-cell RNA-seq highlights intratumoral heterogeneity in primary glioblastoma. Sci (New York NY) 344: 1396\u0026ndash;1401 \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1126/science.1254257\u003c/span\u003e\u003cspan address=\"10.1126/science.1254257\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNeftel C, Laffy J, Filbin MG, Hara T, Shore ME, Rahme GJ, Richman AR, Silverbush D, Shaw ML, Hebert CMet al et al (2019) An Integrative Model of Cellular States, Plasticity, and Genetics for Glioblastoma. Cell 178: 835\u0026ndash;849 e821 \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.cell.2019.06.024\u003c/span\u003e\u003cspan address=\"10.1016/j.cell.2019.06.024\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSuv\u0026agrave; ML, Tirosh I (2020) The Glioma Stem Cell Model in the Era of Single-Cell Genomics. Cancer Cell 37:630\u0026ndash;636. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ccell.2020.04.001\u003c/span\u003e\u003cspan address=\"10.1016/j.ccell.2020.04.001\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSharma P, Aaroe A, Liang J, Puduvalli VK (2023) Tumor microenvironment in glioblastoma: Current and emerging concepts. Neurooncol Adv 5:vdad009. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/noajnl/vdad009\u003c/span\u003e\u003cspan address=\"10.1093/noajnl/vdad009\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCaruso FP, Garofano L, D'Angelo F, Yu K, Tang F, Yuan J, Zhang J, Cerulo L, Pagnotta SM, Bedognetti D et al (2020) A map of tumor-host interactions in glioma at single-cell resolution. Gigascience 9: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/gigascience/giaa109\u003c/span\u003e\u003cspan address=\"10.1093/gigascience/giaa109\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSingh O, Pratt D, Aldape K (2021) Immune cell deconvolution of bulk DNA methylation data reveals an association with methylation class, key somatic alterations, and cell state in glial/glioneuronal tumors. Acta Neuropathol Commun 9:148. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s40478-021-01249-9\u003c/span\u003e\u003cspan address=\"10.1186/s40478-021-01249-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCapper D, Jones DTW, Sill M, Hovestadt V, Schrimpf D, Sturm D, Koelsche C, Sahm F, Chavez L, Reuss DE et al (2018) DNA methylation-based classification of central nervous system tumours. Nature 555: 469\u0026ndash;474 \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/nature26000\u003c/span\u003e\u003cspan address=\"10.1038/nature26000\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCapper D, Stichel D, Sahm F, Jones DTW, Schrimpf D, Sill M, Schmid S, Hovestadt V, Reuss DE, Koelsche C et al (2018) Practical implementation of DNA methylation and copy-number-based CNS tumor diagnostics: the Heidelberg experience. Acta Neuropathol 136: 181\u0026ndash;210 \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00401-018-1879-y\u003c/span\u003e\u003cspan address=\"10.1007/s00401-018-1879-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDor Y, Cedar H Principles of DNA methylation and their implications for biology and medicine. Lancet 392: 777\u0026ndash;786 Doi, Bergman Y, Cedar H (2018) (2013) DNA methylation dynamics in health and disease. Nat Struct Mol Biol 20: 274\u0026ndash;281 Doi 10.1038/nsmb.2518\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFernandez AF, Assenov Y, Martin-Subero JI, Balint B, Siebert R, Taniguchi H, Yamamoto H, Hidalgo M, Tan AC, Galm Oet al, Hovestadt V, Jones DT, Picelli S, Wang W, Kool M, Northcott PA, Sultan M, Stachurski K, Ryzhova M et al (2012) Warnatz HJ (2014) Decoding the regulatory landscape of medulloblastoma using DNA methylation sequencing. Nature 510: 537\u0026ndash;541 Doi 10.1038/nature13268\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSahm F, Schrimpf D, Stichel D, Jones DTW, Hielscher T, Schefzyk S, Okonechnikov K, Koelsche C, Reuss DE, Capper D al (2017) DNA methylation-based classification and grading system for meningioma: a multicentre, retrospective analysis. Lancet Oncol 18:682\u0026ndash;694. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/S1470-2045(17)30155-9\u003c/span\u003e\u003cspan address=\"10.1016/S1470-2045(17)30155-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLouis DN, Perry A, Reifenberger G, von Deimling A, Figarella-Branger D, Cavenee WK, Ohgaki H, Wiestler OD, Kleihues P, Ellison DW (2016) The 2016 World Health Organization Classification of Tumors of the Central Nervous System: a summary. Acta Neuropathol 131:803\u0026ndash;820. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00401-016-1545-1\u003c/span\u003e\u003cspan address=\"10.1007/s00401-016-1545-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLouis DN, Perry A, Wesseling P, Brat DJ, Cree IA, Figarella-Branger D, Hawkins C, Ng HK, Pfister SM, Reifenberger Get al et al (2021) The 2021 WHO Classification of Tumors of the Central Nervous System: a summary. Neuro Oncol 23:1231\u0026ndash;1251. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/neuonc/noab106\u003c/span\u003e\u003cspan address=\"10.1093/neuonc/noab106\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMcLendon R, Friedman A, Bigner D, Van Meir EG, Brat DJ, Mastrogianakis M, Olson G, Mikkelsen JJ, Lehman T, Aldape Ket al et al (2008) Comprehensive genomic characterization defines human glioblastoma genes and core pathways. Nature 455:1061\u0026ndash;1068. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/nature07385\u003c/span\u003e\u003cspan address=\"10.1038/nature07385\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBedard PL, Hansen AR, Ratain MJ, Siu LL (2013) Tumour heterogeneity in the clinic. Nature 501:355\u0026ndash;364. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/nature12627\u003c/span\u003e\u003cspan address=\"10.1038/nature12627\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLoyfer N, Magenheim J, Peretz A, Cann G, Bredno J, Klochendler A, Fox-Fisher I, Shabi-Porat S, Hecht M, Pelet Tet al et al (2023) A DNA methylation atlas of normal human cell types. Nature 613:355\u0026ndash;364. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41586-022-05580-6\u003c/span\u003e\u003cspan address=\"10.1038/s41586-022-05580-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMoss J, Magenheim J, Neiman D, Zemmour H, Loyfer N, Korach A, Samet Y, Maoz M, Druid H, Arner P et al (2018) Comprehensive human cell-type methylation atlas reveals origins of circulating cell-free DNA in health and disease. Nat Commun 9: 5068 \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41467-018-07466-6\u003c/span\u003e\u003cspan address=\"10.1038/s41467-018-07466-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlcantara Llaguno SR, Parada LF (2016) Cell of origin of glioma: biological and clinical implications. Br J Cancer 115:1445\u0026ndash;1450. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/bjc.2016.354\u003c/span\u003e\u003cspan address=\"10.1038/bjc.2016.354\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDarmanis S, Sloan SA, Croote D, Mignardi M, Chernikova S, Samghababi P, Zhang Y, Neff N, Kowarsky M, Caneda C al (2017) Single-Cell RNA-Seq Analysis of Infiltrating Neoplastic Cells at the Migrating Front of Human Glioblastoma. Cell Rep 21:1399\u0026ndash;1410. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.celrep.2017.10.030\u003c/span\u003e\u003cspan address=\"10.1016/j.celrep.2017.10.030\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYuan J, Levitin HM, Frattini V, Bush EC, Boyett DM, Samanamud J, Ceccarelli M, Dovas A, Zanazzi G, Canoll P et al (2018) Single-cell transcriptome analysis of lineage diversity in high-grade glioma. Genome Med 10: 57 \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s13073-018-0567-9\u003c/span\u003e\u003cspan address=\"10.1186/s13073-018-0567-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYu K, Hu Y, Wu F, Guo Q, Qian Z, Hu W, Chen J, Wang K, Fan X, Wu X al (2020) Surveying brain tumor heterogeneity by single-cell RNA-sequencing of multi-sector biopsies. Natl Sci Rev 7:1306\u0026ndash;1318. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/nsr/nwaa099\u003c/span\u003e\u003cspan address=\"10.1093/nsr/nwaa099\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNishri Y, Hampton D, Ben-Shushan E, Fainstein N, Magnani D, Aharonowiz M, Reubinoff BE, Chandran S, Ben-Hur T (2020) Continuous Immune-Modulatory Effects of Human Olig2\u0026thinsp;+\u0026thinsp;Precursor Cells Attenuating a Chronic-Active Model of Multiple Sclerosis. Mol Neurobiol 57:1021\u0026ndash;1034. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s12035-019-01802-7\u003c/span\u003e\u003cspan address=\"10.1007/s12035-019-01802-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTannenbaum SE, Turetsky TT, Singer O, Aizenman E, Kirshberg S, Ilouz N, Gil Y, Berman-Zaken Y, Perlman TS, Geva Net al et al (2012) Derivation of xeno-free and GMP-grade human embryonic stem cells\u0026ndash;platforms for future clinical applications. PLoS ONE 7:e35325. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1371/journal.pone.0035325\u003c/span\u003e\u003cspan address=\"10.1371/journal.pone.0035325\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVinel C, Rosser G, Guglielmi L, Constantinou M, Pomella N, Zhang X, Boot JR, Jones TA, Millner TO Dumas AA (2021) Comparative epigenetic analysis of tumour initiating cells and syngeneic EPSC-derived neural stem cells in glioblastoma. Nat Commun 12: 6130 \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41467-021-26297-6\u003c/span\u003e\u003cspan address=\"10.1038/s41467-021-26297-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDe Meyer T, Bady P, Trooskens G, Kurscheid S, Bloch J, Kros JM, Hainfellner JA, Stupp R, Delorenzi M, Hegi ME al (2015) Genome-wide DNA methylation detection by MethylCap-seq and Infinium HumanMethylation450 BeadChips: an independent large-scale comparison. Sci Rep 5:15375. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/srep15375\u003c/span\u003e\u003cspan address=\"10.1038/srep15375\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBady P, Marosi C, Weller M, Gr\u0026oslash;nberg BH, Schultz H, Taphoorn MJB, Gijtenbeek JMM, van den Bent MJ, von Deimling A Stupp R (2022) DNA methylation-based age acceleration observed in IDH wild-type glioblastoma is associated with better outcome-including in elderly patients. Acta Neuropathol Commun 10: 39 \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s40478-022-01344-5\u003c/span\u003e\u003cspan address=\"10.1186/s40478-022-01344-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCarter SL, Cibulskis K, Helman E, McKenna A, Shen H, Zack T, Laird PW, Onofrio RC, Winckler W, Weir BA et al (2012) Absolute quantification of somatic DNA alterations in human cancer. Nat Biotechnol 30: 413\u0026ndash;421 \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/nbt.2203\u003c/span\u003e\u003cspan address=\"10.1038/nbt.2203\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlboukadel K, Marcin K, Przemyslaw B (2024) \u003cem\u003esurvminer: Drawing Survival Curves using 'ggplot2'.\u003c/em\u003e R package version 0.5.0. City\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShahar T, Granit A, Zrihan D, Canello T, Charbit H, Einstein O, Rozovski U, Elgavish S, Ram Z, Siegal Tet al et al (2016) Expression level of miRNAs on chromosome 14q32.31 region correlates with tumor aggressiveness and survival of glioblastoma patients. J Neurooncol 130:413\u0026ndash;422. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s11060-016-2248-0\u003c/span\u003e\u003cspan address=\"10.1007/s11060-016-2248-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJohann PD, J\u0026auml;ger N, Pfister SM, Sill M (2019) RF_Purify: a novel tool for comprehensive analysis of tumor-purity in methylation array data based on random forest regression. BMC Bioinformatics 20:428. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12859-019-3014-z\u003c/span\u003e\u003cspan address=\"10.1186/s12859-019-3014-z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChaligne RA-O, Gaiti FA-O, Silverbush D, Schiffman JA-O, Weisman HR, Kluegel L, Gritsch S, Deochand SD Gonzalez Castro LA-O, Richman AR Epigenetic encoding, heritability and plasticity of glioma transcriptional cell states\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGreenwald AC, Darnell NG, Hoefflin R, Simkin D, Mount CW, Gonzalez Castro LN, Harnik Y, Dumont S, Hirsch D, Nomura Met al et al (2024) Integrative spatial analysis reveals a multi-layered organization of glioblastoma. Cell 187:2485\u0026ndash;2501e2426. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.cell.2024.03.029\u003c/span\u003e\u003cspan address=\"10.1016/j.cell.2024.03.029\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJain RK, di Tomaso E, Duda DG, Loeffler JS, Sorensen AG, Batchelor TT (2007) Angiogenesis in brain tumours. Nat Rev Neurosci 8:610\u0026ndash;622. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/nrn2175\u003c/span\u003e\u003cspan address=\"10.1038/nrn2175\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu C, Sage JC, Miller MR, Verhaak RG, Hippenmeyer S, Vogel H, Foreman O, Bronson RT, Nishiyama A, Luo Let al et al (2011) Mosaic analysis with double markers reveals tumor cell of origin in glioma. Cell 146:209\u0026ndash;221. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.cell.2011.06.014\u003c/span\u003e\u003cspan address=\"10.1016/j.cell.2011.06.014\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePersson AI, Petritsch C, Swartling FJ, Itsara M, Sim FJ, Auvergne R, Goldenberg DD, Vandenberg SR, Nguyen KN, Yakovenko Set al et al (2010) Non-stem cell origin for oligodendroglioma. Cancer Cell 18:669\u0026ndash;682. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ccr.2010.10.033\u003c/span\u003e\u003cspan address=\"10.1016/j.ccr.2010.10.033\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLigon KL, Alberta JA, Kho AT, Weiss J, Kwaan MR, Nutt CL, Louis DN, Stiles CD, Rowitch DH (2004) The oligodendroglial lineage marker OLIG2 is universally expressed in diffuse gliomas. J Neuropathol Exp Neurol 63:499\u0026ndash;509. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/jnen/63.5.499\u003c/span\u003e\u003cspan address=\"10.1093/jnen/63.5.499\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSingh SK, Clarke ID, Terasaki M, Bonn VE, Hawkins C, Squire J, Dirks PB (2003) Identification of a cancer stem cell in human brain tumors. Cancer Res 63:5821\u0026ndash;5828\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang Q, Hu B, Hu X, Kim H, Squatrito M, Scarpace L, deCarvalho AC, Lyu S, Li P, Li Y al (2017) Tumor Evolution of Glioma-Intrinsic Gene Expression Subtypes Associates with Immunological Changes in the Microenvironment. Cancer Cell 32:42\u0026ndash;56e46. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ccell.2017.06.003\u003c/span\u003e\u003cspan address=\"10.1016/j.ccell.2017.06.003\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSnuderl M Molecular classification and deconvolution of the immune microenvironment in glioblastoma. Neuro-Oncology 23: 175\u0026ndash;176 Doi, Kaffes I, Szulzewsky F, Chen Z, Herting CJ, Gabanic B, Vel\u0026aacute;zquez Vega JE, Shelton J, Switchenko JM, Ross JL (2021) McSwain LF (2019) Human Mesenchymal glioblastomas are characterized by an increased immune cell presence compared to Proneural and Classical tumors. Oncoimmunology 8: e1655360 Doi 10.1080/2162402X.2019.1655360\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDrexler R, Sch\u0026uuml;ller U, Eckhardt A, Filipski K, Hartung TI, Harter PN, Div\u0026eacute; I, Forster MT, Czabanka M, Jelgersma C al (2023) DNA methylation subclasses predict the benefit from gross total tumor resection in IDH-wildtype glioblastoma patients. Neuro Oncol 25:315\u0026ndash;325. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/neuonc/noac177\u003c/span\u003e\u003cspan address=\"10.1093/neuonc/noac177\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDavenport CF, Scheithauer T, Dunst A, Bahr FS, Dorda M, Wiehlmann L, Tran DDH (2021) Genome-Wide Methylation Mapping Using Nanopore Sequencing Technology Identifies Novel Tumor Suppressor Genes in Hepatocellular Carcinoma. Int J Mol Sci 22. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/ijms22083937\u003c/span\u003e\u003cspan address=\"10.3390/ijms22083937\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Glioblastoma, methylation-signature, tumor heterogeneity, deconvolution, tumor microenvironment, astrocytes, oligodendrocytes, glial cells, glioblastoma pathology, central nervous system tumor","lastPublishedDoi":"10.21203/rs.3.rs-6743395/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6743395/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground:\u003c/h2\u003e \u003cp\u003eIDH-wildtype glioblastoma (GBM) is an aggressive, heterogeneous brain tumor with limited treatment options. This study tries to improve our understanding of GBM by applying DNA methylation-based deconvolution to define its cellular composition and its association with patient outcomes.\u003c/p\u003e\u003ch2\u003eMethods:\u003c/h2\u003e \u003cp\u003eWe generated oligodendroglial precursor cells at various developmental stages from enriched human neural progenitor cultures and used their DNA methylation signatures, along with published signatures of cells types relevant to brain tumors and the tumor microenvironment, to deconvolve 263 adult GBMs (Heidelberg cohort). Tumor purity was estimated using RF_Purify. An independent cohort of 199 GBMs from TCGA and GEO, all treated with standard-of-care therapy, was similarly deconvolved, followed by Kaplan\u0026ndash;Meier survival analysis to assess the prognostic value of the proportions of the neoplastic components.\u003c/p\u003e\u003ch2\u003eResults:\u003c/h2\u003e \u003cp\u003eDeconvolution uncovered distinct cellular compositions that differed between the neoplastic and non-neoplastic components of GBM. Tumor purity analysis showed that the neoplastic fractions averaged 70% of the tumor bulk: they were predominantly oligodendrocyte-like (43%), along with oligodendrocyte precursor-like (27%), astrocyte-like (19%), and mesenchymal stem cell-like (11%) populations. The non-neoplastic fractions were enriched for macrophages, vascular cells, and immune cell populations. A higher oligodendrocyte-like signature was linked to poorer survival (median survival 14.3 vs. 15.3 months; p\u0026thinsp;=\u0026thinsp;0.017), while a higher astrocyte-like signature correlated with improved survival (15.3 vs. 13.4 months; p\u0026thinsp;=\u0026thinsp;0.044). The astrocyte-to-oligodendrocyte ratio emerged as a strong prognostic marker, with a higher ratio predicting significantly longer survival (15.8 vs. 11.9 months; p\u0026thinsp;\u0026lt;\u0026thinsp;0.00011).\u003c/p\u003e\u003ch2\u003eConclusions:\u003c/h2\u003e \u003cp\u003eThe methylation-based deconvolution data provided insight into GBM heterogeneity, highlighting the prognostic relevance of the astrocyte-to-oligodendrocyte ratio and its potential to guide personalized treatment strategies.\u003c/p\u003e","manuscriptTitle":"DNA Methylation-Based Deconvolution Study of Glioblastoma Heterogeneity and Cell Types Associated with Patient Survival","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-06-09 10:36:30","doi":"10.21203/rs.3.rs-6743395/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"1d162d47-dcf4-47a6-ba1b-e0c454afff68","owner":[],"postedDate":"June 9th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-09-08T03:08:45+00:00","versionOfRecord":[],"versionCreatedAt":"2025-06-09 10:36:30","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6743395","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6743395","identity":"rs-6743395","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}