Deciphering the glioblastoma microenvironmental landscape with multi-modal radiogenomics to guide prognosis and personalized therapy.

Deciphering the glioblastoma microenvironmental landscape with multi-modal radiogenomics to guide prognosis and personalized therapy - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Search Log in Primary site navigation Search Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice J Transl Med . 2026 May 5;24:718. doi: 10.1186/s12967-026-08226-7 Deciphering the glioblastoma microenvironmental landscape with multi-modal radiogenomics to guide prognosis and personalized therapy Hongying Zhao 1, ✉, # , Kailai Liu 1, 2, # , Marui Guan 1 , Xun Tang 1 , Yangxinyue Zheng 1 , Hongzheng Yu 1 , Shangwei Ning 1, ✉ , Li Wang 1, ✉ Author information Article notes Copyright and License information 1 College of Bioinformatics Science and Technology, Harbin Medical University, Harbin, 150081 China 2 Basic Medical Science College, Harbin Medical University, Harbin, 150081 China ✉ Corresponding author. # Contributed equally. Received 2025 Oct 23; Accepted 2026 Apr 27; Collection date 2026. © The Author(s) 2026 Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/ . PMC Copyright notice PMCID: PMC13221763  PMID: 42087207 Abstract Background Glioblastoma (GBM) is a highly heterogeneous and treatment-refractory tumor, where the tumor microenvironment (TME) plays a central role in shaping progression and therapeutic response. However, the molecular and cellular basis of TME-linked heterogeneity, and how it can be captured through noninvasive imaging, remains poorly understood. Objective This study aims to establish a noninvasive framework for characterizing GBM heterogeneity by bridging radiomic features (RFs) with TME architecture, and to identify novel druggable vulnerabilities for personalized treatment. Methods We analyzed magnetic resonance imaging (MRI)-derived RFs to identify prognostic RFs characterizing tumor heterogeneity. These RFs were correlated with TME composition through integrated analysis of single-cell transcriptomic data and functional enrichment. We utilized a ranking-based computational approach to evaluate gene set activity at single-cell resolution, assessing the enrichment of critical gene subsets within individual cells’ expressed genes. Drug sensitivity was assessed by matching RF-associated gene signatures with pharmacogenomic perturbation profiles. Results Leveraging noninvasive MRI, we identified 31 prognostic RFs that effectively stratified patients into distinct risk groups (C-index = 0.84; HR = 2.16, p < 0.001). These RFs showed significant associations with key dimensions of TME heterogeneity, showing significant associations with specific cellular states—including neural progenitor cell-like (NPC-like)/oligodendrocyte progenitor cell-like (OPC-like) tumor subclasses, macrophages, and myeloid-derived suppressor cells (MDSCs)—as revealed by single-cell RNA-sequencing (scRNA-seq) analysis. Computational drug screening based on these associations identified targeted agents capable of reversing high-risk expression patterns linked to specific RFs, thereby suggesting potential therapeutic strategies aligned with individual TME profiles. Conclusion Our findings indicate that combining imaging-derived RFs with transcriptomic profiling of the TME offers a promising approach to decode GBM heterogeneity and uncover therapeutic opportunities. This multimodal strategy enables noninvasive stratification and may aid in the design of personalized treatment approaches in GBM. Supplementary information The online version contains supplementary material available at 10.1186/s12967-026-08226-7. Keywords: Glioblastoma, Tumor microenvironment, Single-cell RNA sequencing, Radiogenomics, Personalized therapy Highlights A preoperative MRI-based radiomic model was developed to predict survival in glioblastoma patients with high accuracy. Radiogenomic analysis linked imaging features to underlying gene expression patterns, suggesting their biological basis. Radiomic features were associated with tumor microenvironment components, including neoplastic and immune cell subpopulations. Specific RFs correlated with key biological processes such as DNA repair, cell cycle, stemness, and inflammation. Targetable cellular states linked to RFs suggest potential therapeutic strategies, including nocodazole and parthenolide. Supplementary information The online version contains supplementary material available at 10.1186/s12967-026-08226-7. Introduction The tumor microenvironment (TME) plays a pivotal role in cancer progression, immune evasion, and therapeutic resistance, making it a key focus in oncology research [ 1 ]. Composed of tumor cells, immune and stromal cells, extracellular matrix, and signaling molecules, the TME dynamically influences tumor behavior through complex cellular and molecular interactions [ 2 , 3 ]. In glioblastoma (GBM), a highly aggressive brain malignancy, the TME contributes significantly to tumor heterogeneity and treatment failure, particularly through immune cell infiltration and stromal remodeling [ 4 , 5 ]. However, conventional TME analysis relies heavily on invasive biopsies, which are prone to sampling bias due to spatial tumor heterogeneity [ 6 ]. Recent advances highlight the potential of non-invasive magnetic resonance imaging (MRI) in characterizing tumor heterogeneity and inferring TME characteristics [ 7 – 9 ]. Radiomic features (RFs) extracted from MRI have been shown to correlate with underlying biological processes, offering insights into tumor phenotypes without the need for tissue sampling. Yet, the associations between RFs, TME components, and drug responses remain largely unexplored. In this study, we present a radiogenomic framework integrating multi-center MRI and RNA-seq data from The Cancer Genome Atlas (TCGA) and the Clinical Proteomic Tumor Analysis Consortium (CPTAC) to identify prognostic RFs and their corresponding gene expression signatures. We further link these RFs to specific cellular states using single-cell RNA-seq and explore their associations with drug sensitivity. Our work connects imaging phenotypes with tumor biology and provides novel insights into RF-based stratification and candidate therapeutic targeting in GBM (Fig. 1 ). Fig. 1. Open in a new tab Overview of the study. ( A ) Identification of survival-related RFs. ( B ) Screening of key radiogenomic genes associated with survival RFs. ( C ) Analysis of tumor TME using MRI. ( D ) An atlas of radiomic feature-gene-tumor microenvironment (RF-gene-TME) associations Methods Experimental design Our retrospective multicenter study was performed in four steps (Fig. 1 ). First, we extracted and selected RFs from the preoperative MRI data of GBM patients to construct a radiomic prognostic model. Second, by leveraging multimodal data, we identified key radiogenomic genes associated with each prognostic RF and developed a transcriptomic prognostic model to elucidate potential underlying differences in prognosis. Third, by employing single-cell RNA-sequencing (scRNA-seq) data, we characterized the cellular types and potential therapeutic drugs associated with each prognostic RF. Finally, on the basis of these results, we constructed an atlas illustrating the relationships among RFs, genes, and TME. Data collection MRI data were obtained from TCGA and CPTAC, with tumor subregion annotations provided through the Brain Tumor Segmentation 2021 (BraTS2021) challenge [ 10 , 11 ] and from the REMBRANDT dataset [ 12 ], for which segmentations were generated using an automated pipeline and validated by a board-certified neuroradiologist. All cases include 4 multi-modal MRI sequences (T1, T1ce, T2, FLAIR) and annotations for enhancing tumor (ET), necrotic core (NCR), and peritumoral edema (ED) consistent with the BraTS labeling protocol. Bulk transcriptomics data from TCGA, CPTAC, REMBRANDT, Chinese Glioma Genome Atlas (CGGA) and Gene Expression Omnibus (GEO) were used for gene-level validation. Single-cell RNA-seq data from GEO ( GSE182109 , GSE131928 , GSE125587 ) (Fig. S1 A). Preprocessing of transcriptomic data Bulk transcriptomic data from TCGA-GBM, CPTAC, REMBRANDT, CGGA (CGGA-325 and CGGA-301), and GEO datasets ( GSE4412 and GSE7696 ) were processed using a uniform preprocessing pipeline. Expression values were log₂-transformed with the addition of a pseudocount of 1. For microarray data, probe-to-gene mapping was performed using platform-specific annotation files, and expression values of multiple probes mapping to the same gene were collapsed by averaging. Single-cell RNA-seq data underwent Seurat-based preprocessing and Harmony batch correction (using dataset origin as the batch variable; max.iter.harmony = 20). The first 15 Harmony-corrected dimensions were used for Uniform Manifold Approximation and Projection (UMAP)/t-Distributed Stochastic Neighbor Embedding (t-SNE) visualization, graph construction (FindNeighbors), and clustering (FindClusters, resolution = 0.5). Cell types were annotated by manual curation: glioma cells were classified into mesenchymal-like (MES-like; DDIT3, ENO2, HILPDA), astrocyte-like (AC-like; CST3, SLC1A3), oligodendrocyte progenitor-like (OPC-like; ALCAM, OLIG1, OMG, PLLP, PLP1), and neural progenitor-like (NPC-like; SOX4, DLL3, STMN4) states [ 13 ], non-malignant populations (e.g., myeloid cells, T cells) used markers from CellMarker 2.0 [ 14 ]. All annotations were validated by dot plots of key markers across clusters. This pipeline yielded 211,182 high-quality cells for in-depth tumor microenvironment characterization. Extraction of RFs The tumor subregion segmentations included in TCGA, CPTAC and REMBRANDT datasets were used for RF extraction. Five volumes of interest (VOIs), including ET, NCR, ED, whole tumor (WT) and tumor core (TC) were defined in accordance with the BraTS annotation protocol across four MRI modalities (T1, T1ce, T2, and FLAIR). Prior to feature extraction, voxel intensities within each VOI and modality were z-score normalized to reduce inter-scanner variability and improve feature comparability. RFs were extracted using PyRadiomics with fixed parameters. For each VOI, features were computed across 18 image types (including Original, Wavelet, Square, SquareRoot, Logarithm, Exponential, Gradient, and LBP filters), resulting in a total of 33,760 features per patient. Construction of the radiomic prognostic model To identify robust RFs significantly associated with overall survival (OS), univariate Cox regression was first performed in the TCGA cohort, and features with P < 0.05 were selected. Subsequently, highly correlated features ( R > 0.9) were removed to reduce redundancy. Five machine learning algorithms—random survival forest (RSF), Lasso, the elastic network (Enet), and ridge and stepwise Cox regression— were applied to refine feature selection. To prevent overfitting, each algorithm incorporates distinct strategies: Lasso implements L1 regularization [ 15 ]; Enet combines both L1 and L2 regularization [ 16 ]; ridge regression shrinks coefficients through L2 regularization [ 17 ]; stepwise Cox regression determines the optimal feature subset through iterative stepwise selection [ 18 ]; and RSF averages predictions across numerous survival trees to reduce individual tree overfitting. Furthermore, the model was validated and evaluated in the multiple external validation sets mentioned previously. Finally, a radiomic prognostic risk model was constructed using multivariate Cox regression on 101 patients from the TCGA cohort (serving as the training set). To validate the performance of the radiomic prognostic model, we defined two external validation sets, which included 19 patients from CPTAC and 16 patients from REMBRANDT (Fig. S1 A). Identification of key radiogenomic genes for each RF To identify genes associated with RFs, we integrated radiomic and transcriptomic data from the TCGA and CPTAC cohorts. Batch effects between cohorts were corrected using the ComBat algorithm applied separately to the radiomic and gene expression matrices, with dataset origin (TCGA vs. CPTAC) as the batch variable and no additional covariates included to preserve biological heterogeneity; data were then normalized by min–max scaling. Pearson correlation analysis was performed between RFs and gene expression profiles, with thresholds of R > 0.3 and P < 0.05 being widely adopted standards in previous radiogenomic association studies [ 19 ]. Partial least squares (PLS) regression was applied to capture the dominant gene expression patterns linked to each RF. Permutation tests and bootstrapping were used to assess significance and estimate variability, respectively. Genes with Z-scores > 3 were considered significantly associated with RFs, a cutoff established in prior imaging–transcriptomic work and widely used to identify highly contributory features [ 20 ]. Key radiogenomic genes were defined as the intersection between PLS-derived signatures and correlated gene sets. Construction of the radiogenomic prognostic signature For the development of the radiogenomic prognostic signature, first, univariate Cox regression was performed to identify survival-associated radiogenomic genes ( P < 0.05). Next, LASSO-penalized Cox regression was applied to reduce model complexity and select the most predictive features. Finally, stepwise multivariate Cox regression was used to further refine variable selection and construct the final prognostic model. Specifically, we constructed multiple independent datasets: the training set utilized 142 patient samples from TCGA, while six independent external validation sets were established, including 97 patient samples from CPTAC, 210 from REMBRANDT, 107 from CGGA-301, 85 from CGGA-325, 80 cases from GSE7696 , 59 cases from GSE4412 (Fig. S1 A). Analysis of radiogenomic prognostic differences We used the limma R package to perform differential expression analysis and identify survival-associated differentially expressed genes (DEGs), with significance defined as P < 0.05 and fold change (FC) > 1. Subsequently, gene set enrichment analysis (GSEA) was conducted on the DEGs, applying Benjamini–Hochberg correction, to explore functional distinctions between GBM patients with differing prognostic outcomes. Functional characterization of RFs To explore the potential biological correlates of radiogenomic genes from 20 prognostic RFs, KEGG pathway enrichment analysis was performed using ClusterProfiler [ 21 ], with significance set at FDR-corrected P < 0.05. Associations between RFs and functional state activities were assessed by applying single-sample gene set enrichment analysis (ssGSEA) to bulk RNA-seq data, and AUCell to scRNA-seq data. Associations were considered significant if |R| > 0.3 and P < 0.05. To further elucidate the mediating role of radiogenomic genes in the causal effects of RFs on biological pathways as well as on cellular states, we performed causal mediation analysis using mediation package [ 22 ], the results with p < 0.05 were considered statistically significant. Identification of cell subpopulations In scRNA-seq analysis, the AUCell R package was employed to assess the activity of key radiogenomic genes for each RF. AUCell ranks genes by expression, computes cumulative enrichment curves, and calculates AUC scores to infer gene set activity per cell. Cells were classified as active based on default AUC thresholds. Fisher’s exact test was used to detect enrichment of active cells in subpopulations containing more than 200 cells, with significance set at FDR-corrected P < 0.05 and odds ratio (OR) > 1. For bulk RNA-seq, cell type deconvolution was performed using BayesPrism. A reference matrix derived from scRNA-seq data modeled gene expression profiles of distinct cell types via a multinomial distribution. Bayesian inference estimated cell fractions and expression levels in bulk samples, enabling decomposition into constituent cell populations. Prediction of RF-linked therapeutic drugs To identify potential drugs associated with RFs, an anti-correlation-based strategy was implemented. For each significant RF-linked cell subpopulation identified above, a subpopulation-specific transcriptional signature was constructed: an ‘across-cell subpopulation’ signature, derived from genes that are differentially expressed in the target subpopulation relative to all other RF-linked subpopulations, identified using the Wilcoxon rank-sum test implemented in the FindAllMarkers function of Seurat, with the following parameters: min.pct = 0.25, |log₂ fold-change| ≥ 0.25, and only.pos = FALSE; a ‘within-cell subpopulation’ signature, defined by genes exhibiting the highest and lowest average expression levels within the subpopulation. Drug-specific signatures were obtained from the Connectivity Map (CMap). Anti-correlation was used to match downregulated drug target genes with upregulated cell genes (and vice versa), with associations evaluated using chi-square tests. We further applied the Benjamini–Hochberg false discovery rate (FDR) correction to the derived P-values, and associations with an FDR-adjusted P < 0.05 were defined as statistically significant. To validate scRNA-seq–derived RF–drug associations in bulk tumors, drug IC50 values were predicted using pRRophetic [ 23 ] from GBM RNA-seq data. Spearman correlation and Wilcoxon rank-sum tests were used to assess associations between RF radiogenomic scores or inferred cell-type fractions and IC50, testing whether single-cell–identified drug sensitivities are recapitulated at the bulk level. For additional information, see Supplemental Materials. Results A total of 721 patients were enrolled in this study, comprising 101 patients in the radiomics analysis set (mean age, 58 years ± 14; 63 men), 142 patients in the transcriptomics training set (mean age, 60 years ± 14; 94 men), 40 patients in the Meta set (mean age, 58 years ± 13; 28 men), and 6 test sets, as shown in Fig. S1 A and Table S1 . Radiomic features reveal prognosis-associated groups To identify prognostic RFs in GBM, we analyzed three publicly available cohorts: TCGA ( n = 101, training cohort), CPTAC ( n = 19), and REMBRANDT ( n = 16). A total of 33,760 RFs were extracted from tumor regions. Univariate analysis followed by the intersection of five machine learning algorithms identified 31 key RFs (RF1–RF31; Fig. 2 A and B; Table 1 ), including first order statistics ( n = 10), gray level co-occurrence matrix (GLCM) ( n = 3), gray level dependence matrix (GLDM) ( n = 2), gray level run length matrix (GLRLM) ( n = 1), gray level size zone matrix (GLSZM) ( n = 11), neighbouring gray tone difference matrix (NGTDM) ( n = 2), and shape features ( n = 2). These RFs originated from different tumor subregions: NCR ( n = 9), ED ( n = 1), ET ( n = 6), TC ( n = 7), and WT ( n = 8). It has been shown that the mean of WT region is associated with worse prognosis in GBM, as is the case for RF10 [ 24 , 25 ]. Fig. 2. Open in a new tab RFs used to predict the outcomes of patients with GBM. (A) A total of 31 prognostic RFs were identified by 5 machine learning algorithms. (B) The radiomic prognostic risk model consisting of 31 RFs. (C) Independence of radiomic risk scores was tested by multivariate cox regression. (D) The radar chart shows the proportion of eight clinical factors in the two radiomics prognostic groups of patients. Chr: chromosome. Kaplan-Meier curves of OS for radiomics risk score and ROC to evaluate the predictive ability of radiomic risk score on 1-year, 2-year and 3-year survival rates in (E) training cohort, and (F , G) two independent validation cohorts. (H) Representative MRI scans of a low-risk patient (top; RF31 = 0.479) and a high-risk patient (bottom; RF31 = 0.348). RF31: Sphericity_Original_Shape_WT_Flair. (I) Radiomic feature values of 31 RFs across the high- and low-risk groups (Wilcoxon rank-sum test). (J) The RF31 feature values are significantly different between the high-risk and low-risk groups. For each RF, we classified them as risk or protective factors based on the results from the first section of the Cox regression model Table 1. 31 Prognostic Radiomic features (RFs) Feature ID Feature name Image type Feature class Subregion Modality RF1 Skewness Wavelet-HLH First order NCR Flair RF2 Energy Squareroot First order NCR T2 RF3 10 percentile Original First order ET T1gd RF4 Maximum Wavelet-LLH First order ET T1gd RF5 10 percentile Wavelet-LLL First order TC T1gd RF6 Robust Mean Absolute Deviation Squareroot First order WT Flair RF7 Median Wavelet-LHH First order WT Flair RF8 90 Percentile Original First order WT T1gd RF9 Variance Squareroot First order WT T2 RF10 Mean Wavelet-LHL First order WT T2 RF11 Imc1 Wavelet-LLL GLCM ET T1gd RF12 Imc2 Wavelet-LLL GLCM ET T1gd RF13 Correlation Lbp_3D GLCM TC Flair RF14 Large Dependence High Gray Level Emphasis Wavelet-LLH GLDM NCR Flair RF15 High Gray Level Emphasis Wavelet-LLL GLDM TC T1gd RF16 Short Run Low Gray Level Emphasis Wavelet-HLL GLRLM NCR T1gd RF17 Gray Level Non-Uniformity Original GLSZM NCR Flair RF18 Gray Level Non-Uniformity Original GLSZM NCR T1 RF19 Small Area Low Gray Level Emphasis Wavelet-LLL GLSZM NCR T1 RF20 Small Area High Gray Level Emphasis Lbp_3D GLSZM NCR T2 RF21 High Gray Level Zone Emphasis Wavelet-HHH GLSZM NCR T2 RF22 Gray Level Non-Uniformity Exponential GLSZM ED Flair RF23 Large Area Emphasis Logarithm GLSZM ET T1gd RF24 Large Area Low Gray Level Emphasis Wavelet-HHL GLSZM TC T1gd RF25 Gray Level Non-Uniformity Wavelet-LLL GLSZM TC T1gd RF26 Size Zone Non-Uniformity Normalized Wavelet-LLL GLSZM WT Flair RF27 Small Area Low Gray Level Emphasis Wavelet-LLH GLSZM WT T2 RF28 Strength Original NGTDM ET Flair RF29 Busyness Wavelet-LLL NGTDM TC T1gd RF30 Elongation Original Shape TC Flair RF31 Sphericity Original Shape WT Flair Open in a new tab A multivariate Cox regression model was built using these 31 RFs to generate a radiomic risk score for each patient, associated with individual disease progression hazards (Fig. 2 B). The radiomic risk score was significantly associated with overall OS in GBM patients ( P < 0.001, Fig. 2 C), independent of clinical variables. Patients were stratified into high-risk and low-risk groups based on the mean risk score from the TCGA cohort. Kaplan–Meier analysis confirmed that high-risk patients had significantly poorer survival in the training cohort ( P = 1.0e-04, HR = 2.16, 95% CI: 1.84–2.53; Fig. 2 E), which was validated in two independent cohorts (log-rank P = 8.6e-04 and P = 0.02; Fig. 2 F–G). High-risk classification also correlated with chromosome 7 gain/10 loss and reduced MGMT promoter methylation [ 26 ] (Fig. 2 D). Numerous reports have indicated that chromosome 7 gain/ chromosome 10 loss and MGMT promoter methylation are associated with higher drug resistance and worse prognosis [ 27 , 28 ]. ROC analysis demonstrated strong predictive performance of the radiomic risk score across all cohorts, with AUC values of 0.98, 0.94, and 0.96 at 1-, 2-, and 3-year time points in the TCGA cohort, respectively (Fig. 2 E–G). Of the selected features, RF17, RF18, and RF19 demonstrated the most significant prognostic impact according to the feature importance analysis based on Wald chi-square values adjusted for degrees of freedom (Fig. S1 B). Among the 31 RFs, 20 showed significant differences between prognostic groups, including RF31 (Wilcoxon rank sum test, P = 0.049; Fig. 2 I–J), which also varied by Karnofsky performance status (KPS) and age ( P = 0.02 and P < 2.22e-16, respectively; Fig. S1 C–D). RF31 was higher in low-risk patients and may associated with tumor sphericity, where lower values indicate more irregular and aggressive tumors. It has been reported that higher sphericity in glioblastoma is associated with a stemness-enriched, proliferative subtype [ 29 ]. Correlation analysis revealed radiogenomic signatures associated with radiomic prognostic classification To explore the relationship between RFs and transcriptomic profiles, we analyzed a meta-cohort combining TCGA-GBM and CPTAC data. Using Pearson correlation ( R > 0.3, P < 0.05) and PLS algorithms (Z > 3), we identified 1061 key radiogenomic genes linked to 31 RFs (average of 34 per RF; Fig. 3 A, Table S2 ). Overall, the selected RFs showed strong associations with specific gene expressions profiles. RF5 correlated with EGF and BSCL2, both elevated in high-risk patients. A wide variety of evidence has pointed to a critical role of EGF and BSCL2 in tumor metastasis [ 30 , 31 ]. RF31, which may be associated with tumor sphericity, inversely correlated with CCL18, TFF3, and TF ( R =-0.36 to -0.41, all P < 0.05). These genes have been reported to facilitate glioma progression via tumor-promoting signaling mechanisms [ 32 – 34 ]. RF12, measured as the GLCM imc2 of the ET region on T1gd, quantifies texture complexity and is a protective RF. It is positively correlated with SLC22A2 ( R = 0.37, P = 0.019), a gene downregulated in aggressive gliomas [ 35 ]. Previous studies have shown that reduced texture complexity may indicate extensive vascular remodeling, which is associated with a poorer prognosis [ 36 ]. RF1, RF8, and RF10 were associated with poor OS and linked to prognostic genes such as ETV2, POLE2, EZH2, KLF11, and GPRASP1, which have been implicated in adverse outcomes [ 31 , 37 – 41 ]. Fig. 3. Open in a new tab Radiogenomic signatures linked to radiomic prognostic stratification. (A) Number of key radiogenomic genes associated with each prognostic RF (upper panel) and their pearson correlation strength with the corresponding RF (bottom panel). The colors represent the classification of RFs. Label colors denote prognostic risk factors (pink) and protective factors (blue) for each RF, respectively. (B) Lollipop plot showing the coefficients in the radiogenomic risk score model. (C) Multivariate cox analysis revealed that risk score was an independent prognostic factor for GBM patients. (D) Kaplan-Meier survival curves for patients in the high-risk and low-risk groups of the TCGA-GBM cohort. (E) Differentially expressed genes between the high-risk and low-risk groups. Red/blue: upregulated/downregulated genes in the high-risk group relative to the low-risk group. (F) GSEA comparing the high- and low-risk groups, highlighting acitvated (left) and suppressed (right) pathways in the high-risk groups To identify radiogenomic signatures as a surrogate for the radiomics prognostic groups, we first screened the 1061 radiogenomic genes using univariate Cox regression, followed by LASSO-penalized Cox regression. The resulting gene set was then subjected to stepwise multivariate Cox regression, which yielded a final model of 14 significant predictors: ZFR2, TRIM3, SYNDIG1L, SWT1, SULT1E1, STOX1, SCYGR1, PYCR3, OSMR, NRG1, MSTN, GOLGA6B, CPNE6, and CLEC4C. A risk score derived from gene expression and regression coefficients was significantly associated with overall survival ( P < 1.0e-03; Fig. 3 B and C). Kaplan–Meier analysis confirmed worse outcomes for high-risk groups ( P < 0.001, HR = 2.72, 95% CI: 2.15–3.43; Fig. 3 D, S2 A- S2 G). We further compared gene expression between high- and low-risk groups and identified 3931 differentially expressed genes (Fig. 3 E). Among the 14 model genes, eight (OSMR, CLEC4C, GOLGA6B, NRG1, SYNDIG1L, TRIM3, CPNE6, and ZFR2) were upregulated and three (STOX1, SCYGR1, and SWT1) were downregulated in the high-risk group. For instance, TRIM3 has been shown to regulate tumor growth through the modulation of c-Myc transcriptional activity in GBM [ 42 ]. OSMR forms a co-receptor complex with EGFRvIII to amplify receptor tyrosine kinase signaling and drive GBM tumorigenesis [ 43 ]. The prognostic biomarker NRG1 promotes the malignant transformation of GBM cells by inhibiting autophagy via the ERBB2/AKT/mTOR pathway [ 44 ]. STOX1 functions as a novel ferroptosis-associated tumor suppressor that exerts its anti-tumor effects by modulating the ferroptosis signaling pathway in GBM [ 45 ]. Furthermore, genome-wide CRISPR-Cas9 screening data from the Cancer Dependency Map (DepMap) database were used to assess the functional effects of prognostic gene knockout. CRISPR-based gene effect analysis (CERES score) indicated that GOLGA6B and CPNE6 depletion markedly suppressed the proliferation in the vast majority of GBM cell lines (Fig. S2 H). Additionally, MGMT and POLD4 (RF21) were upregulated, while CTBP2 and ZNF22 (RF14 and RF20) were downregulated. Consistent with our findings, MGMT and POLD4 expression have been reported to predict poor survival and recurrence [ 27 , 46 – 49 ], CTBP2 overexpression is linked to adverse GBM outcomes [ 50 ], whereas ZNF22 and STOX1 are associated with favorable prognosis [ 51 , 52 ]. DEG-based GSEA revealed activation of pathways related to cytokine signaling and NF-κB in high-risk patients (Fig. 3 F). Among these, CXCL8 (RF18) was positively correlated with risk score ( R = 0.42, P = 7.4e-03) and has been implicated in therapeutic resistance, poor survival, and maintenance of glioma stem cells via PI3K/AKT and NF-κB signaling [ 53 , 54 ]. Collectively, the dysregulation of oncogenic signaling, cytokine signaling, and cell death pathways may provide a mechanistic basis for the poor prognosis associated with the high-risk group. Analyzing the biological function of prognostic RFs via bulk RNA-seq To explore the biological functions of 20 RFs significantly differing between high-risk and low-risk GBM patients (Fig. 2 I), we performed KEGG pathway enrichment on the key radiogenomic genes for each RF. Significant enrichments were observed in human disease-related pathways, with 7 RFs mapping to five major biological categories (Fig. 4 A). RF1 was enriched in DNA replication pathways, and its key radiogenomic gene POLE2 showed strong positive correlation with DNA repair-related processes ( R = 0.53, P = 4.5e-4; Wilcoxon P = 1.4e-3). These findings suggest a potential association between RF1 and DNA replication and repair, processes that are critical for GBM progression and therapeutic response [ 39 , 55 , 56 ]. RF17 and RF18 were enriched in immune-related pathways (IL-17, chemokine, cytokine signaling) [ 57 ]. Uniquely, RF17 also involved TNF signaling, which may contribute to its prognostic distinction [ 58 ]. RF10 was enriched in GABAergic synapse pathways, with its key radiogenomic gene TUG1 supporting glioma stem cell maintenance [ 59 , 60 ]. Given that cellular states critically shape tumor heterogeneity [ 61 ], we next evaluated the functional states of tumor cells using key radiogenomic gene sets across all 20 RFs (Fig. 4 B). Notably, the RF31 feature value was negtively correlated with tumor stemness, while its corresponding gene CASQ2 showed a significant negative correlation ( R = -0.37, P = 0.02; Wilcoxon P = 0.055; Fig. 4 C), consistent with reports linking high CASQ2 expression to enhanced cancer stemness [ 62 ]. In line with KEGG enrichment results, RF1, which is associated with skewness of the NCR region on FLAIR after Wavelet-HLH transformation, was strongly associated with DNA damage, cell cycle, and DNA repair processes ( R = 0.76, P = 2.2e-16; Wilcoxon P = 2.2e-16; Fig. 4 D–E), and its key gene POLE2 has been implicated in DNA repair and mitotic regulation [ 55 , 56 , 63 ]. RF17 and RF18 were positively correlated with inflammation, with texture features from FLAIR (RF17) and T1 (RF18) closely associated with CXCL5 expression ( R = 0.45, P = 3.3e-3 and 3.9e-3; Wilcoxon P = 0.06 and 0.049) and overall inflammatory signatures ( R = 0.81 and 0.85, P = 2.2e-16 for both; Wilcoxon P = 2.2e-16 for both). CXCL5 has been widely reported to be involved in inflammatory processes [ 64 ]. RF17, derived from GLSZM-based gray-level nonuniformity of the NCR region in Flair, was strongly correlated with CSF3 expression ( R = 0.77, P = 2.2e-16; Wilcoxon P = 2.5e-13) and overall inflammation ( R = 0.86, P = 2.2e-16; Wilcoxon P = 2.2e-16), as well as with CD34 ( R = 0.37, P = 0.02), suggesting links to angiogenesis [ 65 – 67 ]. These results highlight the close relationship between RF17/18 texture features, key radiogenomic genes, and inflammatory and vascular processes in GBM (Fig. 4 C–E). These findings were consistent in the TCGA-GBM cohort (Fig. S3 A, S3 D- S3 F). We further explored the mediation role of radiogenomic genes in the association between RFs and cellular states as well as biological pathways using causal mediation analysis. Consistent with our findings, RF1 demonstrated significant effects on DNA repair processes and replication pathway, while RF17 and RF18 showed significant effects on inflammatory processes and immune-related pathways, validating the biological relevance of these RFs (Fig. S3 B, S3 C). Fig. 4. Open in a new tab Biological functional characterization of prognostic RFs based on bulk RNA-seq analysis. (A) The KEGG enrichment of the key radiogenomic genes of RFs. (B) Pearson correlation between RFs and cell functional states. (C ) RF31 feature value versus CASQ2 expression. (D) Anatomical region captured by RF1. (E) RF1 feature value versus its key radiogenomic gene POLE2. (F) ssGSEA score of RF1 key radiogenomic genes versus DNA repair signature activity. For ( C , E , F ), scatter plots show Pearson correlations; box plots compare gene or signature scores between high- and low-RF groups (Wilcoxon rank-sum test). RF31: Sphericity_Original_Shape_WT_Flair. RF1: Skewness_Wavelet-HLH_First order_NCR_Flair Radiomics predicts important components of the tumor microenvironment via key radiogenomic gene sets of RFs and single-cell RNA-seq To explore the association between RFs and the GBM TME, we integrated GBM data with scRNA-seq profiles from three datasets ( GSE182109 , GSE131928 , GSE125587 ), yielding 211,182 cells. Unsupervised clustering identified 12 major cell populations, including four neoplastic subtypes—MES-like ( n = 17,647), AC-like ( n = 16,687), OPC-like ( n = 25,239), and NPC-like ( n = 11,074)—and eight non-neoplastic types, such as T cells, myeloid cells, astrocytes, and oligodendrocytes (Fig. 5 A, S4 A– S4 D). Batch effects were corrected prior to analysis. Activity of key radiogenomic gene sets was evaluated across these cell subpopulations, revealing significant associations between RFs and specific TME components (Fig. 5 B), which were further corroborated in TCGA using BayesPrism-estimated cell fractions (Fig. S5 A). RF21, representing the High Gray Level Zone Emphasis of the GLSZM feature on T2 images (wavelet_HHH, NCR region), was significantly associated with macrophages and CD4 + T cells. RF31, whose key radiogenomic gene is TFF3, showed a strong association with oligodendrocytes, which promote glioblastoma stemness [ 68 , 69 ]. We observed distinct patterns among RF types: features associated with neoplastic cells (e.g., RF1, RF5, RF7) were predominantly first-order statistical features, whereas those linked to immune responses (e.g., RF3, RF8, RF15, RF16, RF17, RF18, RF21, RF23, RF26, RF28, RF30) were mostly texture-based (Fig. 5 C). Correlation analysis between RF-linked populations and functional cancer cell states revealed that different RFs corresponded to distinct functional states, with some cell subpopulation involved in multiple states (e.g., DNA damage and proliferation). For instance, RF1-linked MES-like cells correlated positively with DNA damage, repair, and cell cycle pathways, whereas OPC-like cells were enriched in proliferation signatures (Fig. 5 D). These findings were confirmed using TCGA bulk RNA-seq data (Fig. S3 A, S5 A). A hierarchical model was built to map the RF–gene–TME–function network in GBM (Fig. 5 E and Fig. S4 E, Table S3 ). Several RFs showed consistent correlations with specific TME components across both scRNA-seq and TCGA-GBM datasets. RF1 was associated with NPC-like infiltration ( R = 0.19, P < 2.2e-16) and BIRC5 expression ( R = 0.71, P < 2.2e-16), both elevated in high-RF1 patients (Wilcoxon P < 2.2e-16; Fig. 5 A–E, S6 A–B). NPC-like cells linked to RF1 also correlated with cell cycle activity ( R = 0.50, P < 2.2e-16). Consistently, BIRC5 shows GBM-specific overexpression in Human Protein Atlas (HPA) immunohistochemistry, and decreased BIRC5 expression has been reported to induce cell cycle arrest [ 70 ] (Fig. S5 B). RF10, measuring the first-order mean of the WT region on T2 (wavelet-LHL), was positively correlated with OPC-like neoplastic cell infiltration ( R = 0.43, P < 2.2e-16) and its key radiogenomic gene TNR expression ( R = 0.30, P < 2.2e-16), both elevated in high-RF10 patients (Wilcoxon P < 2.2e-16; Fig. 5 A–C, S6 C–D). OPC-like cells associated with RF10 were strongly linked to stemness ( R = 0.77, P = 2.2e-16) and TGF-β activation, consistent with TNR’s detection in GBM via HPA immunostaining (Fig. S5 C) and its known role in maintaining glioma stem cell properties [ 13 , 59 , 71 – 73 ]. Similar patterns were observed in the TCGA-GBM cohort (Fig. S5 D). RF20 correlated with MES-like infiltration ( R = 0.056, P < 2.2e-16) and PARP2 expression ( R = 0.13, P < 2.2e-16), both elevated in the high-RF20 group (Wilcoxon P < 2.2e-16). MES-like cells showed enrichment in DNA repair pathways ( R = 0.61, P < 2.2e-16), consistent with PARP2’s known role in recruiting repair proteins and assembling base excision repair complexes [ 74 ], validated in TCGA-GBM data. RF18 was positively correlated with macrophage infiltration and its key radiogenomic gene CXCL3 ( R = 0.52, P < 2.2e-16), with both elevated in high-RF18 patients (Wilcoxon P < 2.2e-16; Fig. 5 A–B). Macrophages associated with RF18 were enriched in quiescence pathways ( R = 0.31, P = 2.2e-16), consistent with its link to poor prognosis, and similar patterns were observed in TCGA-GBM data [ 53 , 75 – 77 ] (Fig. S5 E). RF21 was correlated with M2 macrophage infiltration and LILRA2 expression ( R = 0.76 and 0.29, P < 2.2e-16), whereas RF17 was associated with CXCL1-mediated MDSC recruitment ( R = 0.64, P < 2.2e-16) and hypoxia responses ( R = 0.48, P < 2.2e-16), consistent with reported mechanisms involving CXCL1 secretion and HIF-1α regulation [ 78 – 80 ]. RF13 captured GLCM correlations in Flair-derived TC regions, consistent with its key radiogenomic genes (NEU4, CDHR1, VIPR2, PCDH15) are highly expressed in oligodendrocytes [ 81 ]. These results suggest that RFs are associated with diverse aspects of the TME and reveal links between imaging phenotypes and molecular and cellular tumor characteristics. Fig. 5. Open in a new tab Associations between RFs and the tumor microenvironment cells in GBM. (A) From left to right, t-SNE and UMAP maps of 211,182 cells, respectively. (B) t-SNE maps of three independent GBM scRNA-seq cohorts ( GSE182109 , GSE131928 , GSE125587 ). (C ) Enrichment of 20 RFs key radiogenomic gene sets across 12 cell subpopulations (Fisher’s exact test). (D) Pearson correlation between RF-linked cell subpopulations and 14 cancer functional states of CancerSEA. (E) Schematic summary of a hierarchical model for systematic characterization of the top-ranked RF–gene–TME–function association network in GBM Radiomics predicts the cellular drug response via key radiogenomic gene sets of RFs and single-cell RNA-seq To explore the potential therapeutic implications of radiogenomic features, we performed anti-correlation screening between RF-linked cell subpopulation gene signatures and drug-induced gene expression profiles from CMap. Differentially expressed genes for each RF-linked cell population were used as subpopulation-specific signatures. For each drug, a signature was defined based on its ranked gene expression changes. Using these signatures, we screened CMap drugs across all RF-linked cell subpopulation and identified the top five drugs with the strongest inverse correlations for each RF–cell subpopulation pair. An RF–cell subpopulation–drug association network was constructed (Fig. 6 A and Fig. S7 A, Table S4 ). Among the identified drugs, several showed strong associations with specific RFs. As the second top-ranked compound targeting RF31-linked oligodendrocytes, ipratropium bromide was predicted to induce significant changes in subpopulation-specific gene expression following treatment (Wilcoxon P < 2.2e-16; Fig. 6 B). Consistent with previous reports, ipratropium bromide has been identified as a potential inhibitor of oncogenic pathways in glioma and is known to promote oligodendrocyte differentiation, suggesting a plausible role in modulating the glioblastoma microenvironment [ 82 , 83 ]. Similarly, Trichostatin A (TSA) was also predicted to target RF31-linked oligodendrocytes, with treatment-associated gene expression changes (Wilcoxon P < 2.2e-16; Fig. S7 B). This aligns with prior studies showing that TSA suppresses glioblastoma cell proliferation, migration, and invasion [ 84 , 85 ]. Nocodazole was identified as a candidate drug for MES-like neoplastic cells associated with RF1, RF7, RF12, RF14, and RF20. Treatment significantly altered gene expression in RF1-linked MES-like cells (Wilcoxon P = 2.2e-16; Fig. 6 C), including key radiogenomic genes such as TUBA1B, VEPH1, PEA15, and DTX3, consistent with nocodazole’s inhibitory effects on tumor cell proliferation [ 86 , 87 ], and its G2/M phase arrest activity [ 88 ], and interacting with the RF1 key radiogenomic genes TUBA1B and UBA2 [ 89 , 90 ]. Notably, the RF1, RF14, and RF20-linked MES_like neoplastic cells (Fig. 5 D, S4 E), and nocodazole’s GI50 is negatively correlated with RF13 gene expression via CellMiner (VEPH1, PEA15, DTX3; Figs. S7 C– S7 E) [ 91 ], further supporting its selective efficacy in these proliferative subpopulations. Parthenolide was identified as a potential therapeutic agent for RF15-linked MDSCs, significantly altering their cell-type-specific gene expression (Wilcoxon P = 2.2e-16; Fig. 6 D). In the TCGA-GBM cohort, IC50 values were negatively correlated with RF15 radiogenomic gene expression (Rho = -0.24, P = 3.4e-3; Fig. S7 F) and with MDSC infiltration (Rho = -0.34, P = 2.6e-5; Wilcoxon P = 4.3e-4; Fig. S7 G), indicating higher sensitivity in high-RF15 patients. These results align with previous reports that parthenolide targets myeloid cells and exhibits antitumor activity in GBM [ 92 , 93 ]. Some drugs appeared across multiple RF-linked populations; for instance, clonidine was associated with RF16. While its role in GBM requires further study, clonidine has been used as an adjunct therapy for cancer pain management [ 94 ]. These findings indicate that RFs can be associated with underlying tumor biology and may serve as predictive biomarkers for drug response. Fig. 6. Open in a new tab Candidate drug identification for RF-linked cell subpopulations in GBM. (A) The top 2 ranked potential drugs for RF-linked cell subpopulations. Distinct shapes denote different categorical groups (GBM, RFs, cell subpopulations, potential drugs), and distinct colors represent specific RFs linked to cell subpopulations and potential drugs. Rank distributions of the subpopulation-specific transcriptional signatures (left) and drug-specific signatures (right) used in wilcoxon rank-sum test: (B) RF31-linked oligodendrocytes after trichostatin A treatment, (C) RF1-linked MES-like neoplastic cells after nocodazole treatment, and (D) RF15-linked MDSCs after parthenolide treatment. Box plots show the median and interquartile range. RF1: Skewness_Wavelet-HLH_First order_NCR_Flair; RF31: Sphericity_Original_Shape_WT_Flair; RF15: High Gray Level Emphasis_Wavelet-LLL_GLDM_TC_T1gd Discussion The TME plays a crucial role in therapeutic resistance and clinical outcomes in GBM [ 95 ]. Integrating preoperative imaging with molecular profiling offers a non-invasive strategy to characterize the TME, supporting clinical decision-making and personalized treatment strategies [ 96 ]. We identified 31 prognostic RFs from preoperative MRIs of TCGA-GBM patients and constructed a radiomic risk model with strong prognostic performance across two independent validation cohorts. These RFs captured tumor heterogeneity through shape, intensity, and texture features, especially those derived from multiregional segmentation. Texture features were particularly informative, consistent with their reported value in predicting prognosis in GBM [ 97 ]. Multiregional radiomics has been associated with survival and molecular subtypes [ 98 ], and our results confirm that the radiomic risk score serves as an independent predictor of overall survival [ 24 , 25 ]. Notably, RF31, which measures tumor sphericity, was associated with better outcomes and higher expression in low-risk patients, suggesting that more regular tumor shapes are associated with less aggressive disease [ 62 ]. To provide functional interpretation, we developed a radiogenomic model using gene expression profiles linked to each RF. This model recapitulated the risk stratification observed in the radiomic model and extended its applicability to larger patient cohorts independent of imaging data. Unlike earlier studies focusing on differentially expressed genes [ 8 , 99 ], our approach links RFs to biological pathways, offering deeper insights into how imaging phenotypes map onto tumor biology. We further characterized the biological functions of 20 RFs showing significant prognostic differences. Functional enrichment analysis revealed that RF1 was enriched in DNA replication and repair pathways, aligning with the role of genomic instability in tumor aggressiveness [ 39 , 55 , 56 ]. In contrast, RF17 and RF18 were associated with immune-related pathways such as IL-17 signaling, chemokine signaling, and cytokine–cytokine receptor interactions, suggesting their involvement in modulating inflammation and immune infiltration in the TME [ 7 ]. These findings are consistent with evidence linking inflammatory signatures to poor prognosis and therapy resistance in GBM [ 46 , 58 , 100 ]. By integrating scRNA-seq and bulk RNA-seq data, we mapped RFs to specific cellular states. High RF1 scores were associated with NPC-like neoplastic cells exhibiting enhanced cell cycle activity, indicating that RF1 is strongly correlated with transcriptional signatures of tumor proliferation and genomic instability [ 39 , 55 , 56 , 70 ]. RF10 was linked to OPC-like cells with elevated stemness gene expression, suggesting it may capture imaging features related to tumor cell plasticity. Importantly, TNR, a key radiogenomic gene of RF10, was associated with stemness pathways and TGF-β activation, consistent with its potential role in maintaining glioma stem cell properties [ 13 , 71 , 72 ]. We also explored the therapeutic implications of key radiogenomic genes of RFs by predicting drug responses. Nocodazole emerged as a candidate for targeting RF1-, RF13-, and RF14-linked MES-like neoplastic cells, consistent with its known effects on G2/M phase arrest and tumor growth inhibition [ 87 – 89 ]. It interacts with key radiogenomic genes such as TUBA1B and UBA2 [ 90 ], and its GI50 value negatively correlated with RF13 gene expression [ 91 ] Parthenolide showed sensitivity in RF15-linked MDSCs based on IC50 analysis, supporting its use in TME-targeted immunomodulation [ 92 , 93 ]. Compared to previous studies focusing on the link between RFs and molecular changes, our work extends this by associating RFs with specific cell populations and functional states within the tumor ecosystem. Meanwhile, while integrative analysis of scRNA-seq and bulk RNA-seq has been increasingly applied in cancer research [ 101 , 102 ], studies combining RFs with such multi-layered transcriptomic data for functional interpretation remain relatively limited. Our integrative approach provides a more comprehensive view of how imaging phenotypes are associated with tumor heterogeneity and TME interactions. Conclusion In summary, our findings suggest that RFs derived from preoperative MRI can be associated with diverse aspects of tumor biology, including genetic programs, cellular composition, and functional states. These results highlight the translational potential of radiomics in guiding personalized treatment strategies for GBM patients. Electronic supplementary material Below is the link to the electronic supplementary material. Supplementary Material 1 (6.2MB, docx) Supplementary Material 2 (15.1KB, xlsx) Supplementary Material 3 (19KB, xlsx) Supplementary Material 4 (11.9KB, xlsx) Abbreviations GBM Glioblastoma TME Tumor microenvironment RF Radiomic feature MRI Magnetic resonance imaging TCGA The Cancer Genome Atlas CPTAC Clinical Proteomic Tumor Analysis Consortium BraTS Brain Tumor Segmentation ET Enhancing tumor NCR Necrotic core ED Peritumoral edema WT Whole tumor TC Tumor core CGGA Chinese Glioma Genome Atlas GEO Gene Expression Omnibus OPC-like Oligodendrocyte progenitor cell-like NPC-like Neural progenitor cell-like MDSCs Myeloid-derived suppressor cells FDR False discovery rate GLCM Gray-level co-occurrence matrix GLDM Gray-level dependence matrix GLRLM Gray-level run length matrix GLSZM Gray level size zone matrix NGTDM Neighbouring gray tone difference matrix GSEA Gene set enrichment analysis ssGSEA Single-sample gene set enrichment analysis scRNA-seq Single-cell RNA sequencing DEG Differentially expressed gene FC Fold change Chr Chromosome HPA Human Protein Atlas UMAP Uniform Manifold Approximation and Projection t-SNE t-Distributed Stochastic Neighbor Embedding Author contributions Hongying Zhao: Study conception and design, methodological design, data processing and analysis, manuscript drafting and revision. Kailai Liu: Data processing and analysis, figure and table visualization, manuscript drafting and revision. Marui Guan: Data processing. Xun Tang: Data processing. Yangxinyue Zheng: Data collection and collation, figure and table visualization. Hongzheng Yu: Data collection and collation. Shangwei Ning: Study conception and design. Li Wang: Study conception and design, methodological design, resource provision and support, supervision and guidance. All authors read, reviewed, and approved the final manuscript. Funding This work was supported by National Natural Science Foundation of China (62372144, 62573169, 62572155) and Outstanding Youth Foundation of Heilongjiang Province (YQ2023F004). Data availability All data and codes in this study are available under proper request, please contact wangli@hrbmu.edu.cn or ningsw@ems.hrbmu.edu.cn. All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Competing interests The authors declare that they have no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Hongying Zhao and Kailai Liu contributed equally to this work. Contributor Information Hongying Zhao, Email: zhaohongying@hrbmu.edu.cn. Shangwei Ning, Email: ningsw@ems.hrbmu.edu.cn. Li Wang, Email: wangli@hrbmu.edu.cn. References 1. Kharouf N, Flanagan TW, Hassan S-Y, Shalaby H, Khabaz M, Hassan S-L, et al. Tumor microenvironment as a therapeutic target in melanoma treatment. Switzerland: Cancers; 2023. p. 15. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 2. El-Tanani M, Rabbani SA, Babiker R, Rangraze I, Kapre S, Palakurthi SS, et al. Unraveling the tumor microenvironment: Insights into cancer metastasis and therapeutic strategies. Cancer Lett Irel. 2024;591:216894. [ DOI ] [ PubMed ] [ Google Scholar ] 3. Zhao H, Yin X, Wang S, Geng Z, Yu W, Yu H, et al. So3D: a comprehensive three-dimensional spatial omics resource for decoding tissue architecture in physiology and disease. Nucleic Acids Res. 2026;54:D1281–90. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 4. Ren J, Xu B, Ren J, Liu Z, Cai L, Zhang X et al. The importance of M1-and M2-polarized macrophages in glioma and as potential treatment targets. Brain Sci Switz. 2023;13. [ DOI ] [ PMC free article ] [ PubMed ] 5. Pang L, Zhou F, Chen P. Lipid-Laden macrophages recycle myelin to feed glioblastoma. Cancer Res United States. 2024;84:3712–4. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 6. Onubogu U, Gatenbee CD, Prabhakaran S, Wolfe KL, Oakes B, Salatino R, et al. Spatial analysis of recurrent glioblastoma reveals perivascular niche organization. JCI Insight. United States; 2024. p. 9. [ DOI ] [ PMC free article ] [ PubMed ] 7. Sun Q, Chen Y, Liang C, Zhao Y, Lv X, Zou Y, et al. Biologic pathways underlying prognostic radiomics phenotypes from paired MRI and RNA sequencing in glioblastoma. Radiology. 2021;301:654–63. [ DOI ] [ PubMed ] [ Google Scholar ] 8. Beig N, Singh S, Bera K, Prasanna P, Singh G, Chen J, et al. Sexually dimorphic radiogenomic models identify distinct imaging and biological pathways that are prognostic of overall survival in glioblastoma. Neuro-Oncol. 2021;23:251–63. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 9. Choi Y, Nam Y, Jang J, Shin N-Y, Lee YS, Ahn K-J, et al. Radiomics may increase the prognostic value for survival in glioblastoma patients when combined with conventional clinical and genetic prognostic models. Eur Radiol. 2021;31:2084–93. [ DOI ] [ PubMed ] [ Google Scholar ] 10. Baid U, Ghodasara S, Mohan S, Bilello M, Calabrese E, Colak E et al. The RSNA-ASNR-MICCAI BraTS 2021 Benchmark on Brain Tumor Segmentation and Radiogenomic Classification [Internet]. arXiv; 2021 [cited 2026 Jan 5]. 11. Bakas S, Akbari H, Sotiras A, Bilello M, Rozycki M, Kirby JS, et al. Advancing The Cancer Genome Atlas glioma MRI collections with expert segmentation labels and radiomic features. Sci Data. 2017;4:170117. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 12. Sayah A, Bencheqroun C, Bhuvaneshwar K, Belouali A, Bakas S, Sako C, et al. Enhancing the REMBRANDT MRI collection with expert segmentation labels and quantitative radiomic features. Sci Data. 2022;9:338. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 13. Neftel C, Laffy J, Filbin MG, Hara T, Shore ME, Rahme GJ, et al. An integrative model of cellular states, plasticity, and genetics for glioblastoma. Cell. 2019;178:835–e84921. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 14. Hu C, Li T, Xu Y, Zhang X, Li F, Bai J, et al. CellMarker 2.0: an updated database of manually curated cell markers in human/mouse and web tools based on scRNA-seq data. Nucleic Acids Res. 2023;51:D870–6. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 15. Bloniarz A, Liu H, Zhang C-H, Sekhon JS, Yu B. Lasso adjustments of treatment effect estimates in randomized experiments. Proc Natl Acad Sci U S A. 2016;113:7383–90. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 16. Pokhriyal N, Jacques DC. Combining disparate data sources for improved poverty prediction and mapping. Proc Natl Acad Sci U S A. 2017;114:E9783–92. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 17. Xu G, Hu B-G, Principe JC. Robust C-Loss Kernel Classifiers. IEEE Trans Neural Netw Learn Syst. 2018;29:510–22. [ DOI ] [ PubMed ] [ Google Scholar ] 18. van Lammeren GW, Catanzariti LM, Peelen LM, de Vries J-PPM, de Kleijn DPV, Moll FL, et al. Clinical prediction rule to estimate the absolute 3-year risk of major cardiovascular events after carotid endarterectomy. Stroke. 2012;43:1273–8. [ DOI ] [ PubMed ] [ Google Scholar ] 19. Cui H, Sun Y, Zhao D, Zhang X, Kong H, Hu N, et al. Radiogenomic analysis of prediction HER2 status in breast cancer by linking ultrasound radiomic feature module with biological functions. J Transl Med. 2023;21:44. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 20. Bo T, Li J, Hu G, Zhang G, Wang W, Lv Q, et al. Brain-wide and cell-specific transcriptomic insights into MRI-derived cortical morphology in macaque monkeys. Nat Commun. 2023;14:1499. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 21. Wu T, Hu E, Xu S, Chen M, Guo P, Dai Z, et al. clusterProfiler 4.0: A universal enrichment tool for interpreting omics data. Innov Camb Mass. 2021;2:100141. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 22. Jones J, Ertefaie A, Strawderman RL. Causal mediation analysis: selection with asymptotically valid inference. J R Stat Soc Ser B Stat Methodol. 2025;87:678–700. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 23. Geeleher P, Cox N, Huang RS. pRRophetic: an R package for prediction of clinical chemotherapeutic response from tumor gene expression levels. PLoS ONE. 2014;9:e107468. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 24. Verduin M, Primakov S, Compter I, Woodruff HC, van Kuijk SMJ, Ramaekers BLT, et al. Prognostic and predictive value of integrated qualitative and quantitative magnetic resonance imaging analysis in glioblastoma. Cancers. 2021;13:722. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 25. Zhang X, Lu H, Tian Q, Feng N, Yin L, Xu X, et al. A radiomics nomogram based on multiparametric MRI might stratify glioblastoma patients according to survival. Eur Radiol. 2019;29:5528–38. [ DOI ] [ PubMed ] [ Google Scholar ] 26. Ceccarelli M, Barthel FP, Malta TM, Sabedot TS, Salama SR, Murray BA, et al. Molecular profiling reveals biologically discrete subsets and pathways of progression in diffuse glioma. Cell. 2016;164:550–63. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 27. Mansouri A, Hachem LD, Mansouri S, Nassiri F, Laperriere NJ, Xia D, et al. MGMT promoter methylation status testing to guide therapy for glioblastoma: refining the approach based on emerging evidence and current challenges. Neuro-Oncol. 2019;21:167–78. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 28. Wu S, Li X, Gao F, de Groot JF, Koul D, Yung WKA. PARP-mediated PARylation of MGMT is critical to promote repair of temozolomide-induced O6-methylguanine DNA damage in glioblastoma. Neuro-Oncol. 2021;23:920–31. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 29. Zhang Z, Liu Y, Zhang Z, Gui T, Chen Y, Chen Q, et al. MRI-based radiomic clustering identifies a glioblastoma subtype enriched for neural stemness and proliferative programs. Front Oncol. 2025;15:1662401. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 30. Bartolomé RA, Martín-Regalado Á, Jaén M, Zannikou M, Zhang P, de Los Ríos V, et al. Protein Tyrosine Phosphatase-1B Inhibition Disrupts IL13Rα2-Promoted Invasion and Metastasis in Cancer Cells. Cancers. 2020;12:500. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 31. Fan F, Zhang H, Dai Z, Zhang Y, Xia Z, Cao H, et al. A comprehensive prognostic signature for glioblastoma patients based on transcriptomics and single cell sequencing. Cell Oncol Dordr Neth. 2021;44:917–35. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 32. Huang Y, Motta E, Nanvuma C, Kuhrt LD, Yuan Y, Xia P, et al. Microglia/macrophage-derived human CCL18 promotes glioma progression via CCR8-ACP5 axis analyzed in humanized slice model. Cell Rep. 2022;39:110670. [ DOI ] [ PubMed ] [ Google Scholar ] 33. Shukla A, Gupta P, Singh R, Mishra DP. Glycolytic inhibitor 2-Deoxy-d-Glucose activates migration and invasion in glioblastoma cells through modulation of the miR-7-5p/TFF3 signaling pathway. Biochem Biophys Res Commun. 2018;499:829–35. [ DOI ] [ PubMed ] [ Google Scholar ] 34. Han M, Xu R, Wang S, Yang N, Ni S, Zhang Q, et al. Six-transmembrane epithelial antigen of prostate 3 predicts poor prognosis and promotes glioblastoma growth and invasion. Neoplasia N Y N. 2018;20:543–54. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 35. Vassileva V, Braga M, Barnes C, Przystal J, Ashek A, Allott L, et al. Effective Detection and monitoring of glioma using [18F]FPIA PET imaging. Biomedicines. 2021;9:811. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 36. Ruan B, Chen M, Zhuang Q, Guan L, Xie W, Wang L, et al. Chest Computed Tomography-Based Radiomics for the Diagnosis and Prognosis of Pulmonary Hypertension. J Am Heart Assoc. 2025;14:e043221. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 37. Zhao J, Yang S, Cui X, Wang Q, Yang E, Tong F, et al. A novel compound EPIC-0412 reverses temozolomide resistance via inhibiting DNA repair/MGMT in glioblastoma. Neuro-Oncol. 2023;25:857–70. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 38. Zhao C, Gomez GA, Zhao Y, Yang Y, Cao D, Lu J, et al. ETV2 mediates endothelial transdifferentiation of glioblastoma. Signal Transduct Target Ther. 2018;3:4. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 39. Zhang P, Chen X, Zhang L, Cao D, Chen Y, Guo Z, et al. POLE2 facilitates the malignant phenotypes of glioblastoma through promoting AURKA-mediated stabilization of FOXM1. Cell Death Dis. 2022;13:61. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 40. Xin L, Tan Y, Zhu Y, Cui X, Wang Q, Zhao J, et al. EPIC-0307-mediated selective disruption of PRADX-EZH2 interaction and enhancement of temozolomide sensitivity to glioblastoma via inhibiting DNA repair and MGMT. Neuro-Oncol. 2023;25:1976–88. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 41. Li J, Wang C, Cheng R, Su H, Wang L, Ji L, et al. KLF11 promotes the progression of glioma via regulating Holliday junction recognition protein. Cell Biol Int. 2022;46:1144–55. [ DOI ] [ PubMed ] [ Google Scholar ] 42. Teng W, Ling Y, Liu Z, Jiang L, Fu G, Zhou X, et al. Advances in the antitumor mechanisms of tripartite motif-containing protein 3. J Cancer Res Clin Oncol. 2024;150:105. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 43. Jahani-Asl A, Yin H, Soleimani VD, Haque T, Luchman HA, Chang NC, et al. Control of glioblastoma tumorigenesis by feed-forward cytokine signaling. Nat Neurosci. 2016;19:798–806. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 44. Lin J-Z, Lin N, Zhao W-J. A prognostic biomarker NRG1 promotes U-87 MG glioblastoma cell malignancy by inhibiting autophagy via ERBB2/AKT/mTOR pathway. J Cell Biochem. 2023;124:1273–88. [ DOI ] [ PubMed ] [ Google Scholar ] 45. Wang S, Xing S, Li K, Liu S, Yang Y, Zhu X et al. Development and validation of a novel glioblastoma prognostic model: identifying STOX1 and ZNF248 as ferroptosis-related biomarkers. Cancer Cell Int. 2025. [ DOI ] [ PMC free article ] [ PubMed ] 46. Luce A, Abate M, Scognamiglio G, Montella M, Iervolino D, Campione S, et al. Immune cell infiltration and inflammatory landscape in primary brain tumours. J Transl Med. 2024;22:521. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 47. Chen L, Wen A. Unveiling the role of O(6)-methylguanine-DNA methyltransferase in cancer therapy: insights into alkylators, pharmacogenomics, and others. Front Oncol. 2024;14:1424797. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 48. Fan L, You H, Jiang X, Niu Y, Chen Z, Wang H, et al. UCHL3 induces radiation resistance and acquisition of mesenchymal phenotypes by deubiquitinating POLD4 in glioma stem cells. Cell Mol Life Sci CMLS. 2024;81:247. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 49. Knoll M, Waltenberger M, Furkel J, Wirkner U, Gahlawat AW, Dokic I, et al. Whole blood transcriptional fingerprints of high-grade glioma and longitudinal tumor evolution under carbon ion radiotherapy. Cancers. 2022;14:684. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 50. Chen L, Wang L, Qin J, Wei D-S. CtBP2 interacts with ZBTB18 to promote malignancy of glioblastoma. Life Sci. 2020;262:118477. [ DOI ] [ PubMed ] [ Google Scholar ] 51. Cheng Q, Huang C, Cao H, Lin J, Gong X, Li J, et al. A novel prognostic signature of transcription factors for the prediction in patients with GBM. Front Genet. 2019;10:906. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 52. Tong S, Xia M, Xu Y, Sun Q, Ye L, Yuan F, et al. Identification and validation of a novel prognostic signature based on mitochondria and oxidative stress related genes for glioblastoma. J Transl Med. 2023;21:136. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 53. Yuan W, Zhang Q, Gu D, Lu C, Dixit D, Gimple RC, et al. Dual role of CXCL8 in maintaining the mesenchymal state of glioblastoma stem cells and M2-Like tumor-associated macrophages. Clin Cancer Res Off J Am Assoc Cancer Res. 2023;29:3779–92. [ DOI ] [ PubMed ] [ Google Scholar ] 54. Hasan T, Caragher SP, Shireman JM, Park CH, Atashi F, Baisiwala S, et al. Interleukin-8/CXCR2 signaling regulates therapy-induced plasticity and enhances tumorigenicity in glioblastoma. Cell Death Dis. 2019;10:292. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 55. Guo H, Ren H, Han K, Li J, Dong Y, Zhao X, et al. Knockdown of HDAC10 inhibits POLE2-mediated DNA damage repair in NSCLC cells by increasing SP1 acetylation levels. Pulm Pharmacol Ther. 2023;83:102250. [ DOI ] [ PubMed ] [ Google Scholar ] 56. Jian W, Zhang L. POLE2 silencing inhibits the progression of colorectal carcinoma cells via wnt signaling axis. Cancer Biol Ther. 2024;25:2392339. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 57. Alghamri MS, Banerjee K, Mujeeb AA, Mauser A, Taher A, Thalla R, et al. Systemic delivery of an adjuvant CXCR4-CXCL12 signaling inhibitor encapsulated in synthetic protein nanoparticles for glioma immunotherapy. ACS Nano. 2022;16:8729–50. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 58. Zhang Y, Kang T, Wang Y, Song C, Li H, Mi H, et al. A low level of tumor necrosis factor α in tumor microenvironment maintains the self-renewal of glioma stem cells by Vasorin-mediated glycolysis. Neuro-Oncol. 2024;26:2256–71. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 59. Fan X, Gong M, Yu H, Yang H, Wang S, Wang R. Propofol enhances stem-like properties of glioma via GABAAR-dependent Src modulation of ZDHHC5-EZH2 palmitoylation mechanism. Stem Cell Res Ther. 2022;13:398. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 60. Katsushima K, Natsume A, Ohka F, Shinjo K, Hatanaka A, Ichimura N, et al. Targeting the Notch-regulated non-coding RNA TUG1 for glioma treatment. Nat Commun. 2016;7:13616. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 61. Yuan H, Yan M, Zhang G, Liu W, Deng C, Liao G, et al. CancerSEA: a cancer single-cell state atlas. Nucleic Acids Res. 2019;47:D900–8. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 62. Kim JH, Lee E-S, Yun J, Ryu HS, Kim HK, Ju YW, et al. Calsequestrin 2 overexpression in breast cancer increases tumorigenesis and metastasis by modulating the tumor microenvironment. Mol Oncol. 2022;16:466–84. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 63. Krähling H, Musigmann M, Akkurt BH, Sartoretti T, Sartoretti E, Henssen DJHA, et al. A magnetic resonance imaging based radiomics model to predict mitosis cycles in intracranial meningioma. Sci Rep. 2023;13:969. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 64. Yaghi OK, Hanna BS, Langston PK, Michelson DA, Jayewickreme T, Marin-Rodero M, et al. A discrete early-responder stromal-cell subtype orchestrates immunocyte recruitment to injured tissue. Nat Immunol. 2023;24:2053–67. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 65. Gao Y, Zhang E, Liu B, Zhou K, He S, Feng L, et al. Integrated analysis identified core signal pathways and hypoxic characteristics of human glioblastoma. J Cell Mol Med. 2019;23:6228–37. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 66. Brown KH, Ghita-Pettigrew M, Kerr BN, Mohamed-Smith L, Walls GM, McGarry CK, et al. Characterisation of quantitative imaging biomarkers for inflammatory and fibrotic radiation-induced lung injuries using preclinical radiomics. Radiother Oncol J Eur Soc Ther Radiol Oncol. 2024;192:110106. [ DOI ] [ PubMed ] [ Google Scholar ] 67. Liu M-Y, Xie F, Zhang Y, Wang T-T, Ma S-N, Zhao P-X, et al. Molecular hydrogen suppresses glioblastoma growth via inducing the glioma stem-like cell differentiation. Stem Cell Res Ther. 2019;10:145. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 68. Kawashima T, Yashiro M, Kasashima H, Terakawa Y, Uda T, Nakajo K, et al. Oligodendrocytes Up-regulate the Invasive Activity of Glioblastoma Cells via the Angiopoietin-2 Signaling Pathway. Anticancer Res. 2019;39:577–84. [ DOI ] [ PubMed ] [ Google Scholar ] 69. Mikolajewicz N, Zhai K, Puri A, Miletic P, Tatari N, Wei J, et al. Reactive oligodendrocytes promote glioblastoma progression through CCL5/CCR5-mediated glioma stem cell maintenance. Neuron. 2026;114:237–e24910. [ DOI ] [ PubMed ] [ Google Scholar ] 70. Hattori EY, Masuda T, Mineharu Y, Mikami M, Terada Y, Matsui Y, et al. A RUNX-targeted gene switch-off approach modulates the BIRC5/PIF1-p21 pathway and reduces glioblastoma growth in mice. Commun Biol. 2022;5:939. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 71. Aubert A, Mercier-Gouy P, Aguero S, Berthier L, Liot S, Prigent L, et al. Latent TGF-β activation is a hallmark of the tenascin family. Front Immunol. 2021;12:613438. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 72. Esaki S, Nigim F, Moon E, Luk S, Kiyokawa J, Curry W, et al. Blockade of transforming growth factor-β signaling enhances oncolytic herpes simplex virus efficacy in patient-derived recurrent glioblastoma models. Int J Cancer. 2017;141:2348–58. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 73. Li N, Rodriguez JL, Yin Y, Logun MT, Zhang L, Yu S, et al. Armored bicistronic CAR T cells with dominant-negative TGF-β receptor II to overcome resistance in glioblastoma. Mol Ther J Am Soc Gene Ther. 2024;32:3522–38. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 74. Saville KM, Al-Rahahleh RQ, Siddiqui AH, Andrews ME, Roos WP, Koczor CA, et al. Oncometabolite 2-hydroxyglutarate suppresses basal protein levels of DNA polymerase beta that enhances alkylating agent and PARG inhibition induced cytotoxicity. DNA Repair. 2024;140:103700. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 75. Korbecki J, Bosiacki M, Chlubek D, Baranowska-Bosiacka I. Bioinformatic analysis of the CXCR2 ligands in cancer processes. Int J Mol Sci. 2023;24:13287. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 76. O’Reilly E, Zeinabad HA, Szegezdi E. Hematopoietic versus leukemic stem cell quiescence: challenges and therapeutic opportunities. Blood Rev. 2021;50:100850. [ DOI ] [ PubMed ] [ Google Scholar ] 77. Zhang Y, Zou J, Chen R. An M0 macrophage-related prognostic model for hepatocellular carcinoma. BMC Cancer. 2022;22:791. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 78. Peng C, Li L, Luo G, Tan S, Xia R, Zeng L. Integrated analysis of the M2 macrophage-related signature associated with prognosis in ovarian cancer. Front Oncol. 2022;12:986885. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 79. Wang M, Wang W, You S, Hou Z, Ji M, Xue N, et al. ACAT1 deficiency in myeloid cells promotes glioblastoma progression by enhancing the accumulation of myeloid-derived suppressor cells. Acta Pharm Sin B. 2023;13:4733–47. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 80. Wang J, Yu F, Jia X, Iwanowycz S, Wang Y, Huang S, et al. MicroRNA-155 deficiency enhances the recruitment and functions of myeloid-derived suppressor cells in tumor microenvironment and promotes solid tumor growth. Int J Cancer. 2015;136:E602–13. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 81. Kokkosis AG, Madeira MM, Mullahy MR, Tsirka SE. Chronic stress disrupts the homeostasis and progeny progression of oligodendroglial lineage cells, associating immune oligodendrocytes with prefrontal cortex hypomyelination. Mol Psychiatry. 2022;27:2833–48. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 82. Wang B, Ma Q, Wang X, Guo K, Liu Z, Li G. TGIF1 overexpression promotes glioma progression and worsens patient prognosis. Cancer Med. 2022;11:5113–28. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 83. Ren J, Ascencio M, Raimondi T, Rainville EC, Valenzuela RM, Asche CV. Association between exposure of ipratropium and salmeterol and diagnosis of multiple sclerosis: a matched case-control study. Clin Ther. 2019;41:1477–85. [ DOI ] [ PubMed ] [ Google Scholar ] 84. Rampazzo E, Manfreda L, Bresolin S, Cani A, Mariotto E, Bortolozzi R, et al. Histone deacetylase inhibitors impair glioblastoma cell motility and proliferation. Cancers. 2022;14:1897. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 85. Li KK-W, Yang L, Pang JC-S, Chan AK-Y, Zhou L, Mao Y, et al. MIR-137 suppresses growth and invasion, is downregulated in oligodendroglial tumors and targets CSE1L. Brain Pathol Zurich Switz. 2013;23:426–39. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 86. Abbassi RH, Recasens A, Indurthi DC, Johns TG, Stringer BW, Day BW, et al. Lower tubulin expression in glioblastoma stem cells attenuates efficacy of microtubule-targeting agents. ACS Pharmacol Transl Sci. 2019;2:402–13. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 87. Antonucci L, Di Magno L, D’Amico D, Manni S, Serrao SM, Di Pastena F, et al. Mitogen-activated kinase kinase kinase 1 inhibits hedgehog signaling and medulloblastoma growth through GLI1 phosphorylation. Int J Oncol. 2019;54:505–14. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 88. Booy EP, McRae EK, Koul A, Lin F, McKenna SA. The long non-coding RNA BC200 (BCYRN1) is critical for cancer cell survival and proliferation. Mol Cancer. 2017;16:109. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 89. Li H-P, Peng C-C, Wu C-C, Chen C-H, Shih M-J, Huang M-Y, et al. Inactivation of the tight junction gene CLDN11 by aberrant hypermethylation modulates tubulins polymerization and promotes cell migration in nasopharyngeal carcinoma. J Exp Clin Cancer Res CR. 2018;37:102. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 90. Cannon M, Stevenson J, Stahl K, Basu R, Coffman A, Kiwala S, et al. DGIdb 5.0: rebuilding the drug-gene interaction database for precision medicine and drug discovery platforms. Nucleic Acids Res. 2024;52:D1227–35. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 91. Reinhold WC, Sunshine M, Liu H, Varma S, Kohn KW, Morris J, et al. CellMiner: a web-based suite of genomic and pharmacologic tools to explore transcript and drug patterns in the NCI-60 cell line set. Cancer Res. 2012;72:3499–511. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 92. Liu J, Wu R, Yuan S, Kelleher R, Chen S, Chen R, et al. Pharmacogenomic analysis of combined therapies against glioblastoma based on cell markers from single-cell sequencing. Pharm Basel Switz. 2023;16:1533. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 93. Hou Y, Sun B, Liu W, Yu B, Shi Q, Luo F, et al. Targeting of glioma stem-like cells with a parthenolide derivative ACT001 through inhibition of AEBP1/PI3K/AKT signaling. Theranostics. 2021;11:555–66. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 94. Amna S, Øhlenschlaeger T, Saedder EA, Sigaard JV, Bergmann TK. Review of clinical pharmacokinetics and pharmacodynamics of clonidine as an adjunct to opioids in palliative care. Basic Clin Pharmacol Toxicol. 2024;134:485–97. [ DOI ] [ PubMed ] [ Google Scholar ] 95. White J, White MPJ, Wickremesekera A, Peng L, Gray C. The tumour microenvironment, treatment resistance and recurrence in glioblastoma. J Transl Med. 2024;22:540. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 96. Wu J, Mayer AT, Li R. Integrated imaging and molecular analysis to decipher tumor microenvironment in the era of immunotherapy. Semin Cancer Biol. 2022;84:310–28. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 97. Guan F, Wang Z, Qiu Y, Guo Y, Pei D, Wang M, et al. Biological underpinnings of radiomic magnetic resonance imaging phenotypes for risk stratification in IDH wild-type glioblastoma. J Transl Med. 2023;21:841. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 98. Zhou W, Wen J, Huang Q, Zeng Y, Zhou Z, Zhu Y, et al. Development and validation of clinical-radiomics analysis for preoperative prediction of IDH mutation status and WHO grade in diffuse gliomas: a consecutive L-[methyl-11 C] methionine cohort study with two PET scanners. Eur J Nucl Med Mol Imaging. 2024;51:1423–35. [ DOI ] [ PubMed ] [ Google Scholar ] 99. Beig N, Bera K, Prasanna P, Antunes J, Correa R, Singh S, et al. Radiogenomic-based survival risk stratification of tumor habitat on Gd-T1w MRI is associated with biological processes in glioblastoma. Clin Cancer Res Off J Am Assoc Cancer Res. 2020;26:1866–76. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 100. Sa P, Sahoo SK, Dilnawaz F. Responsive role of nanomedicine in the tumor microenvironment and cancer drug resistance. Curr Med Chem United Arab Emirates. 2023;30:3335–55. [ DOI ] [ PubMed ] [ Google Scholar ] 101. Zhang Z, Wang Z-X, Chen Y-X, Wu H-X, Yin L, Zhao Q, et al. Integrated analysis of single-cell and bulk RNA sequencing data reveals a pan-cancer stemness signature predicting immunotherapy response. Genome Med. 2022;14:45. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 102. Tan Z, Chen X, Zuo J, Fu S, Wang H, Wang J. Comprehensive analysis of scRNA-Seq and bulk RNA-Seq reveals dynamic changes in the tumor immune microenvironment of bladder cancer and establishes a prognostic model. J Transl Med. 2023;21:223. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials Supplementary Material 1 (6.2MB, docx) Supplementary Material 2 (15.1KB, xlsx) Supplementary Material 3 (19KB, xlsx) Supplementary Material 4 (11.9KB, xlsx) Data Availability Statement All data and codes in this study are available under proper request, please contact wangli@hrbmu.edu.cn or ningsw@ems.hrbmu.edu.cn. All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Articles from Journal of Translational Medicine are provided here courtesy of BMC ACTIONS View on publisher site PDF (3.2 MB) Cite Collections Permalink RESOURCES Similar articles Cited by other articles Links to NCBI Databases Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top

Related documents

• Radiogenomic and Prognosis Analysis in Glioma#831106
• Radiogenomics in Glioblastoma: Correlation Between Multiparametric Imaging Biomarkers and Genetic Biomarkers#832190
• Molecular Genetic, Host-derived and Clinical Determinants of Long-term Survival in Glioblastoma#991836
• Glioma Patients Registry Based on Radiological, Histopathological and Genetic Analysis#1085589
• Image 1_HIST1H2BJ is associated with poor prognosis and an immune-altered tumor microenvironment in glioma.jpeg#922330