Mordechai, A. | bioRxiv Science

bioRxiv2025

Methylation-Based Deconvolution Unveils Glioblastoma Heterogeneity and Cell-Type Composition Linked to Patient Survival

BackgroundIDH-wildtype glioblastoma (GBM) is an aggressive, heterogeneous brain tumor with limited treatment options. DNA methylation profiling allows detailed tumor characterization. This study applies methylation-based deconvolution to define GBMs cellular composition and its association with patient outcomes. MethodsWe 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 brain tumor and tumor microenvironment-relevant cell types, 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 neoplastic component proportions. ResultsDeconvolution uncovered distinct cellular compositions, consistent with single-cell RNA sequencing findings. Tumor purity analysis showed neoplastic fractions averaged 70% of tumor bulk, predominantly oligodendrocyte-like (43%), oligodendrocyte precursor-like (27%), astrocyte-like (19%), and mesenchymal stem cell-like (11%) populations. Non-neoplastic fractions were enriched for macrophages, vascular cells, and immune 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 outcomes (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). ConclusionsMethylation-based deconvolution provides insight into GBM heterogeneity, highlighting the prognostic relevance of the astrocyte-to-oligodendrocyte ratio, which may guide personalized treatment strategies. Graphical Abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=84 SRC="FIGDIR/small/640603v2_ufig1.gif" ALT="Figure 1"> View larger version (22K): org.highwire.dtl.DTLVardef@942af0org.highwire.dtl.DTLVardef@18f2626org.highwire.dtl.DTLVardef@111a756org.highwire.dtl.DTLVardef@1194b_HPS_FORMAT_FIGEXP M_FIG A. We generated oligodendroglial precursor cells (OPs) at various developmental stages from enriched human neural progenitor cultures. A reference atlas of methylation signatures from 14 normal cell types, including the in vitro-generated cells, combined with published profiles of brain tumor and tumor microenvironment-relevant cell types, was applied to deconvolve 263 GBM samples based on their methylation profiles (Heidelberg cohort). Cell type proportions were analyzed with RF_Purify to estimate tumor purity and distinguish neoplastic from non-neoplastic components. B. An independent cohort of 199 GBMs from TCGA and GEO, with available clinical data and all treated with standard-of-care therapy, was similarly deconvolved based on methylation profiles. Kaplan-Meier survival analysis assessed the prognostic impact of cell type proportions, identifying the astrocyte-to-oligodendrocyte ratio as the most significant marker. (Created with BioRender.com) C_FIG Key pointsO_LIMethylation-based deconvolution incorporating in-vitro-derived oligodendrocyte precursor differentiation data reveals GBM heterogeneity C_LIO_LIGBMs neoplastic fraction is dominated by oligodendrocyte-lineage cells C_LIO_LIAstrocyte/oligodendrocyte ratio is a strong prognostic marker for patient survival C_LI Importance of the studyThis study advances neuro-oncology research by using methylation-based deconvolution to uncover the cellular heterogeneity of glioblastoma. By constructing a comprehensive reference atlas and identifying cell-type-specific methylation signatures, it provides prognostic insights into the distinct cellular compositions of glioblastoma. Combining deconvolution results with purity analysis enables the differentiation between neoplastic and non-neoplastic tumor components. This approach complements single-cell RNA sequencing while offering greater clinical applicability, as DNA methylation profiling can be performed on FFPE or fresh frozen samples, unlike the more costly and tissue-sensitive single-cell methods. Importantly, the identification of the astrocyte-to-oligodendrocyte ratio as a strong prognostic marker highlights key cellular determinants of glioblastoma survival. These findings deepen our understanding of glioblastoma biology and offer a practical tool for patient stratification.

cancer biology↗

bioRxiv2022

Detection of A-to-I microRNA editing with miRmedon reveals widespread co-editing of mature microRNA in the human brain

microRNA (miRNA), key regulators of gene expression, are prime targets for adenosine deaminase acting on RNA (ADAR) enzymes. Although ADAR-mediated adenosine-to-inosine (A-to-I) miRNA editing has been shown to be essential for orchestrating complex processes, including neurodevelopment and cancer progression, only few human miRNA editing sites have been reported. Several computational approaches have been developed for the detection of miRNA editing in small RNAseq data, all based on the identification of systematic mismatches of G at primary adenosine sites in known miRNA sequences. However, these methods have several limitations, including their ability to detect only one editing site per sequence (although editing of multiple sites per miRNA has been reproducibly validated), their focus on uniquely mapping reads (even though 20% of human miRNA are transcribed from multiple loci), and their inability to detect editing in miRNA harboring genomic variants (though 73% of human miRNA loci include a reported SNP or indel). To overcome these limitations, we developed miRmedon, that leverages large-scale human variation data, a combination of local and global alignments, and a comparison of the inferred editing and error distributions, for confident detection of miRNA editing in small RNAseq data. We demonstrate the improved performance of miRmedon as compared to currently available methods and describe its advantages. We further use miRmedon to discover editing haplotypes of mature human brain miRNA for the first time. We find that doubly edited mature miRNA are common in the adult human prefrontal cortex, most include a frequently edited site within the miRNA seed region, and are predicted to maintain a stable pre-miRNA structure. These results suggest that co-editing of mature miRNA could enable efficient shifting of gene expression programs.

bioinformatics↗

bioRxiv2019

miRmedon: confident detection of microRNA editing

microRNA (miRNA), key regulators of gene expression, are prime targets for adenosine deaminase acting on RNA (ADAR) enzymes. Although ADAR-mediated A-to-I miRNA editing has been shown to be essential for orchestrating complex processes, including neurodevelopment and cancer progression, only a few human miRNA editing sites have been reported. Several computational approaches have been developed for the detection of miRNA editing in small RNAseq data, all based on the identification of systematic mismatches of G at primary adenosine sites in known miRNA sequences. However, these methods have several limitations, including their ability to detect only one editing site per sequence (although editing of multiple sites per miRNA has been reproducibly validated), their focus on uniquely mapping reads (although 20% of human miRNA are transcribed from multiple loci), and their inability to detect editing in miRNA genes harboring genomic variants (although 73% of human miRNA loci include a reported SNP or indel). To overcome these limitations, we developed miRmedon, that leverages large scale human variation data, a combination of local and global alignments, and a comparison of the inferred editing and error distributions, for a confident detection of miRNA editing in small RNAseq data. We demonstrate its improved performance as compared to currently available methods and describe its advantages.\n\nAvailability and implementationPython source code is available at https://github.com/Amitai88/miRmedon\n\nContactalal@bgu.ac.il

bioinformatics↗