Integrated bulk and single-cell transcriptomic analyses identify transcriptome-defined groups and EGFR-associated microenvironmental programs in glioma Authors: Yuanhan Wang, Long Li, Xing Hu, Longhu Han, Yihan Zhang, Zhisen Zhou, Wei Liu, Song Yang Published: 2026 ABSTRACT BackgroundGlioma exhibits significant molecular heterogeneity, necessitating improved understanding of molecular subtypes for personalized treatment strategies.MethodsThis study integrated bulk RNA sequencing (GSE35169) to derive transcriptome-defined groups and single-cell RNA sequencing (GSE131928) to characterize cellular heterogeneity and EGFR-associated malignant cell programs. Differential gene expression analysis identified subtype-specific biomarkers. Single-cell analysis included cell-type annotation, pseudotime trajectory, and cell–cell communication inference, with malignant cells stratified by EGFR expression to interrogate EGFR-associated programs. Key findings were validated by qRT-PCR in glioma cell lines (LN229, U251) and normal astrocytes (NHA).ResultsTwo transcriptome-defined groups were identified from bulk RNA-seq. In scRNA-seq, the EGFR-high malignant cell state showed enrichment of ECM-related genes (IGFBP2, COL1A1) and enhanced PI3K-AKT, focal adhesion, and angiogenic signaling, with predominant stromal-to-tumor communication (EGF/AREG→EGFR, TGFB1→TGFBR2). The EGFR-low malignant cell state exhibited higher expression of immune-related genes (CXCL10, IL6, STAT1). Functional enrichment analysis based on differentially expressed genes (FDR < 0.05) indicated significant enrichment of interferon signaling and JAK-STAT pathway–related gene sets in the EGFR-low group. Pseudotime analysis revealed gradual transcriptional transitions between proliferation, hypoxia, and immune programs. qRT-PCR validation confirmed elevated EGFR, IGFBP2, and COL1A1 in U251 cells, and higher CXCL10, IL6, and STAT1 in LN229 cells.ConclusionThis study identifies bulk transcriptome-defined groups and reveals EGFR-associated malignant cell programs linked to distinct microenvironmental interaction patterns. The EGFR-high malignant cell state features proliferative and angiogenic programs, while the EGFR-low malignant cell state shows enhanced immune infiltration. These findings provide molecular evidence for precision classification and subtype-specific therapeutic strategies. Introduction Among primary malignant neoplasms of the central nervous system, glioma represents the most common category of primary brain malignancies, with glioblastoma (GBM) being the most aggressive subtype characterized by rapid progression and poor prognosis [1]. Even with substantial advances in multimodal therapeutic approaches encompassing surgical resection, radiation therapy, and systemic chemotherapy, prognostic outcomes for glioma patients remain dismal. GBM patients face a median survival period of merely 12–15 months, while fewer than 5% survive beyond five years [2]. This sobering clinical landscape underscores the critical necessity for deepened insights into gliomagenesis mechanisms, discovery of novel therapeutic targets, and optimization of patient care strategies. Traditionally, histopathological appearance has served as the primary framework for glioma classification. Nevertheless, the accelerated advancement of molecular biotechnology has progressively established molecular stratification through gene expression profiling and molecular biomarkers as a fundamental pillar for precision diagnostics and tailored therapeutic interventions in glioma management [3]. The World Health Organization’s updated classification schemas for central nervous system tumors, released in 2021, emphasized the pivotal role of molecular diagnostics in glioma taxonomy, with biomarkers such as IDH mutational profile, 1p/19q chromosomal co-deletion, and MGMT promoter methylation patterns now serving as essential determinants for therapeutic decision-making [4–6]. Nevertheless, gliomas exhibit extraordinary molecular diversity, with marked variations in biological characteristics, therapeutic responsiveness, and clinical outcomes across distinct molecular classifications, thereby presenting formidable obstacles to precision medicine implementation [7]. Current glioma research confronts several critical limitations. First, the scarcity of validated early diagnostic biomarkers and actionable therapeutic targets substantially hampers clinical intervention strategies [8, 9]. Therefore, leveraging state-of-the-art bioinformatics methodologies and extensive datasets to systematically examine glioma spanning from transcriptomic characterization to molecular phenotypic profiling holds tremendous potential for advancing both fundamental research and translational applications within this domain [8, 9]. Moreover, cellular-level experimental verification through techniques such as quantitative real-time PCR (qRT-PCR) is indispensable for substantiating transcriptomic discoveries and enhancing their biological validity. The present investigation aims to comprehensively characterize transcriptomic heterogeneity and molecular subtype-specific features of glioma through integrative analysis of publicly available gene expression datasets retrieved from the Gene Expression Omnibus (GEO) repository. In particular, we performed differential gene expression analysis based on bulk RNA sequencing data (GSE35169) and conducted single-cell RNA sequencing analysis (GSE131928) to delineate cellular diversity, transcriptional programs, and microenvironmental characteristics, and to further examine EGFR-associated malignant cell programs in the glioma microenvironment. Additionally, qRT-PCR validation experiments were performed to verify representative differentially expressed genes in glioma cell line models, thereby ensuring concordance between computational findings and experimental observations. Through the integration of bulk and single-cell transcriptomic profiling with cellular validation, our objective is to provide molecular-level insights into glioma heterogeneity and to establish a foundation for identifying potential diagnostic biomarkers and therapeutic targets. Methods Data source and sample processing For this study, gene expression was collected from the GSE35169 dataset from the GEO database, which stores genomic expression profile from glioma patients. The dataset includes whole-genome expression profiling of multiple glioma samples. Following strict quality control and preprocessing procedures, six high-quality representative samples were selected for subsequent analysis. All samples were collected under the strict experimental conditions and thus of reliable data quality, which could lay a solid data basis for the downstream investigations of glioma transcriptomic heterogeneity and DEG-related analyses. Differential gene expression analysis of glioma molecular subtypes To explore intrinsic transcriptomic heterogeneity among glioma samples from the GSE35169 dataset, bulk RNA sequencing expression data were first subjected to quality control, log2 transformation, and normalization using standard preprocessing procedures. To reduce noise from low-information genes, feature selection was performed based on expression variability across samples. Specifically, gene-wise variability was quantified using the median absolute deviation (MAD) across the normalized expression matrix, and genes ranking within the top 20% of variability were retained for unsupervised analysis. This variance-based filtering approach enhances sensitivity to dominant transcriptional patterns while minimizing the influence of low-variance genes. Unsupervised hierarchical clustering was then conducted on the filtered expression matrix using Euclidean distance and complete linkage algorithms, without predefined subtype labels. Following transcriptome-based sample stratification, differential gene expression analysis was performed between the identified groups using linear modeling approaches. Genes with |log2 fold change| > 1 and adjusted P value (false discovery rate, FDR) < 0.05, calculated using the Benjamini–Hochberg method for multiple testing correction, were considered significantly differentially expressed. For visualization, hierarchical clustering heatmaps were generated using the identified differentially expressed genes to illustrate group-associated expression patterns, and volcano plots were constructed to display both the magnitude of expression changes and their statistical significance (− log10 adjusted P values). Single-cell RNA sequencing data and computational analysis Single-cell transcriptomic data were obtained from the publicly available dataset GSE131928 (Gene Expression Omnibus, GEO). This dataset includes single-cell RNA sequencing (scRNA-seq) profiles from 28 patients with IDH-wildtype glioblastoma (GBM), encompassing both adult and pediatric cases. A total of 24,131 single cells were profiled, including 7930 cells generated using Smart-seq2 and 16,201 cells generated using the 10X Genomics platform. Tumor specimens were dissociated into single cells prior to sequencing, capturing both malignant tumor cells and non-malignant microenvironmental cell populations. Raw count matrices were processed using the Seurat (v5.0) workflow for quality control, normalization, and dimensionality reduction. Cells with fewer than 200 detected genes or with > 10% mitochondrial reads were excluded. After integration and scaling, principal component analysis (PCA) and Uniform Manifold Approximation and Projection (UMAP) were used for visualization and clustering at an appropriate resolution to maintain biological interpretability. Subsequent analyses included cell-type annotation, pseudotime trajectory construction, transcription factor activity inference, and pathway enrichment analysis. These procedures collectively enabled the characterization of cellular heterogeneity, malignant cell programs, and microenvironmental features in glioma. qRT-PCR validation Total RNA was extracted from normal human astrocytes (NHA) and glioma cell lines (LN229 and U251) using TRIzol reagent according to the manufacturer’s instructions. RNA concentration and purity were measured using a NanoDrop spectrophotometer, and only samples with an A260/A280 ratio between 1.8 and 2.0 were included for further analysis. RNA integrity was preliminarily evaluated by agarose gel electrophoresis. For reverse transcription, 1 µg of total RNA was reverse-transcribed into complementary DNA (cDNA) using a commercial reverse transcription kit in a 20 µL reaction system according to the manufacturer’s protocol. Quantitative real-time PCR (qRT-PCR) was performed using SYBR Green Master Mix on a real-time PCR detection system. Each 20 µL reaction mixture contained 10 µL of 2× SYBR Green Master Mix, 0.4 µL of forward primer (10 µM), 0.4 µL of reverse primer (10 µM), 2 µL of diluted cDNA template, and 7.2 µL of nuclease-free water. The amplification conditions were as follows: initial denaturation at 95 °C for 30 s, followed by 40 cycles of 95 °C for 5 s and 60 °C for 30 s. A melting curve analysis was conducted at the end of amplification to verify product specificity. GAPDH served as the internal reference gene. Relative mRNA expression levels were calculated using the 2 − ΔΔCt method. All experiments were performed with three independent biological replicates, and each sample was analyzed in technical triplicate. Data are presented as mean ± standard deviation (SD), and statistical comparisons were conducted using a two-tailed Student’s t-test, with P < 0.05 considered statistically significant (Table 1). Table 1Primer sequences used for quantitative real-time PCR (qRT-PCR)GeneForward primer (5′–3′)Reverse primer (5′–3′)EGFRAGGCACGAGTAACAAGCTCACATGAGGACATAACCAGCCACCIGFBP2CCTGTCACCAAGATGCACCATGGTCCAGTTCCTGTTGGTGCOL1A1GAGGGCCAAGACGAAGACATCCAGATCACGTCATCGCACAACCXCL10GTGGCATTCAAGGAGTACCTCTGATGGCCTTCGATTCTGGATTIL6ACTCACCTCTTCAGAACGAATTGCCATCTTTGGAAGGTTCAGGTTGSTAT1TCACAGTGGTTCGAGCTTCAGCGAGACATCATAGGCAGCGTGGAPDHGGAGCGAGATCCCTCCAAAATGGCTGTTGTCATACTTCTCATGG Table 1Primer sequences used for quantitative real-time PCR (qRT-PCR)GeneForward primer (5′–3′)Reverse primer (5′–3′)EGFRAGGCACGAGTAACAAGCTCACATGAGGACATAACCAGCCACCIGFBP2CCTGTCACCAAGATGCACCATGGTCCAGTTCCTGTTGGTGCOL1A1GAGGGCCAAGACGAAGACATCCAGATCACGTCATCGCACAACCXCL10GTGGCATTCAAGGAGTACCTCTGATGGCCTTCGATTCTGGATTIL6ACTCACCTCTTCAGAACGAATTGCCATCTTTGGAAGGTTCAGGTTGSTAT1TCACAGTGGTTCGAGCTTCAGCGAGACATCATAGGCAGCGTGGAPDHGGAGCGAGATCCCTCCAAAATGGCTGTTGTCATACTTCTCATGG Primer sequences used for quantitative real-time PCR (qRT-PCR) Statistical analysis All statistical analyses were performed using R software (version 4.5.1). For bulk RNA-seq differential expression analysis, linear modeling was applied and multiple testing correction was performed using the Benjamini–Hochberg method to control the false discovery rate (FDR); genes with |log2 fold change| > 1 and FDR < 0.05 were considered statistically significant. For single-cell analyses, group comparisons were conducted using the Wilcoxon rank-sum test unless otherwise specified, and FDR correction was applied where multiple comparisons were performed. For qRT-PCR validation experiments, data are presented as mean ± standard deviation (SD) from three independent biological replicates, and comparisons between groups were performed using two-tailed Student’s t-test. A P value < 0.05 or FDR < 0.05 was considered statistically significant. Results Differential gene expression heatmap analysis for glioma molecular subtypes The bulk RNA-seq cohort consisted of six samples (n = 6) used for exploratory transcriptomic comparison. This heatmap visualizes the expression patterns of differentially expressed genes between the two bulk transcriptome-defined groups (identified under the criteria of |log2FC| > 1 and FDR < 0.05). Samples and genes were hierarchically clustered for visualization based on the DEG matrix, highlighting group-associated expression trends. The color gradient in the heatmap reflects differences in gene expression levels, where red indicates high expression, blue indicates low expression, and color intensity corresponds to expression strength. From the clustering results, significant differences in gene expression patterns can be observed between the two subtype groups. The genes listed on the left side show opposite expression trends in different subtypes, forming distinct molecular signature profiles. This bidirectional expression pattern suggests that these differentially expressed genes may serve as key biomarkers for distinguishing different molecular subtypes of glioma, holding important clinical significance for precise tumor classification and personalized treatment(Fig. 1). Fig. 1Heatmap of differentially expressed genes between two bulk transcriptome-defined groups. The heatmap shows gene expression profile differences between two major molecular subtypes (blue and orange-red groups), with color intensity representing relative expression levels (red: high expression; blue: low expression). Differentially expressed genes were defined as |log2FC| > 1 and FDR < 0.05 (Benjamini–Hochberg correction; n = 6 samples) Heatmap of differentially expressed genes between two bulk transcriptome-defined groups. The heatmap shows gene expression profile differences between two major molecular subtypes (blue and orange-red groups), with color intensity representing relative expression levels (red: high expression; blue: low expression). Differentially expressed genes were defined as |log2FC| > 1 and FDR < 0.05 (Benjamini–Hochberg correction; n = 6 samples) Differential gene expression analysis between glioma molecular subtypes This figure presents the results of differential gene expression analysis between different molecular subtypes of glioma. Panel 2 A shows a boxplot of normalized signal intensity, clearly showing two distinct expression patterns across the six samples, consistent with the two transcriptome-defined groups derived from unsupervised clustering. The substantial expression differences between these two groups provide the foundation for subsequent differential gene analysis. The volcano plot in Panel 2B further reveals the distribution characteristics of differentially expressed genes between the two subtypes. This plot uses Log2 fold change as the x-axis and statistical significance (-log10 FDR-adjusted P-value) as the y-axis to intuitively display both the magnitude of gene expression changes and their statistical significance. The results show numerous significantly differentially expressed genes: red dots represent significantly upregulated genes, mainly distributed in the right region of the plot; blue dots represent significantly downregulated genes, concentrated on the left side; while gray dots indicate genes with no significant expression changes. This bidirectional and abundant differential gene pattern strongly suggests fundamental molecular-level differences between the two glioma subtypes, providing important molecular evidence for precise tumor classification and targeted therapeutic strategy development(Fig. 2A and B). Differential expression was determined using linear modeling with Benjamini–Hochberg correction (FDR < 0.05; n = 6). Fig. 2Differential gene expression analysis between two bulk transcriptome-defined groups. A Boxplot of normalized signal intensity across different sample groups, showing two clearly separated expression pattern groups; B Volcano plot of differentially expressed genes, with red dots indicating upregulated genes, blue dots indicating downregulated genes, and gray dots representing non-significantly different genes. Differential expression analysis was performed using linear modeling with Benjamini–Hochberg correction (|log2FC| > 1, FDR < 0.05; n = 6 samples) Differential gene expression analysis between two bulk transcriptome-defined groups. A Boxplot of normalized signal intensity across different sample groups, showing two clearly separated expression pattern groups; B Volcano plot of differentially expressed genes, with red dots indicating upregulated genes, blue dots indicating downregulated genes, and gray dots representing non-significantly different genes. Differential expression analysis was performed using linear modeling with Benjamini–Hochberg correction (|log2FC| > 1, FDR < 0.05; n = 6 samples) Single-cell clustering and cellular characterization of glioma samples To further interpret the bulk transcriptome-defined differences at cellular resolution, we next analyzed an independent scRNA-seq dataset (GSE131928). The dataset comprised 24,131 single cells derived from 28 patients with IDH-wildtype glioblastoma. In the single-cell analyses, EGFR was used solely for expression-based stratification of malignant cells to examine EGFR-associated programs, rather than for defining bulk molecular subtypes. This figure presents the results of single-cell RNA sequencing–based clustering and cellular characterization in glioma samples. The UMAP visualization in Panel 3 A displays low-resolution clustering (resolution = 0.20), where distinct cell clusters are clearly separated in the two-dimensional space, indicating marked transcriptional heterogeneity among glioma cells. Panel 3B shows the UMAP projection colored by sequencing platforms (Smart-seq2 and 10x Genomics), demonstrating effective batch correction and well-integrated cross-platform data. The dot plot in Panel 3 C summarizes the expression patterns of canonical marker genes across major cell types, where dot size represents the proportion of expressing cells and color intensity indicates scaled expression levels. Distinct cellular identities were identified, including astrocytes (GFAP, AQP4), oligodendrocytes (OLIG2, MBP), OPCs (PDGFRA, CSPG4), endothelial cells (VWF, CLDN5), pericytes (RGS5, PDGFRB), myeloid cells (LYZ, AIF1), T cells (CD3D, TRAC), and malignant cells (SOX2, MKI67, EGFR). Panel 3D shows the proportional composition of these cell types across samples, suggesting balanced cellular representation between the two sequencing platforms. Panels 3E and 3 F visualize the spatial expression of representative genes, revealing that endothelial markers such as VWF are enriched in vascular clusters, while tumor-associated genes such as EGFR and IGFBP2 are specifically expressed in malignant clusters. The heatmap in Panel 3G presents the top two marker genes of the six largest clusters, displaying distinct gene signatures such as LOXL1, C2CD4A, CA10, EGFR-AS1, CPA4, and PLD4. Marker genes were identified using Wilcoxon rank-sum test with FDR correction (FDR < 0.05). Finally, Panel 3 H shows the distribution of cell numbers across clusters, characterized by a few dominant clusters and multiple smaller ones, reflecting diverse transcriptional states. Collectively, these results delineate the cellular heterogeneity of glioma and provide a molecular foundation for subsequent EGFR-related functional and immune microenvironment analyses (Fig. 3A-H). Fig. 3Single-cell clustering and cellular characterization of glioma samples. A UMAP visualization showing low-resolution clustering (resolution = 0.20) of glioma single cells, highlighting distinct transcriptional clusters. B UMAP colored by sequencing platforms (Smart-seq2 and 10x Genomics), demonstrating consistent integration and minimal batch effect. C Dot plot of canonical marker genes across identified cell types; dot size represents the percentage of cells expressing each gene, and color intensity reflects scaled expression levels. D Stacked bar plot showing cell-type proportions within each sample. E–F Feature plots illustrating spatial expression of representative genes, including endothelial marker VWF and tumor-associated marker EGFR. G Heatmap showing the top two upregulated marker genes among the six largest clusters, with rows representing genes and columns representing cells ordered by cluster identity. H Bar plot displaying the distribution of cell counts across clusters, revealing a long-tail pattern of major and minor cell populations Single-cell clustering and cellular characterization of glioma samples. A UMAP visualization showing low-resolution clustering (resolution = 0.20) of glioma single cells, highlighting distinct transcriptional clusters. B UMAP colored by sequencing platforms (Smart-seq2 and 10x Genomics), demonstrating consistent integration and minimal batch effect. C Dot plot of canonical marker genes across identified cell types; dot size represents the percentage of cells expressing each gene, and color intensity reflects scaled expression levels. D Stacked bar plot showing cell-type proportions within each sample. E–F Feature plots illustrating spatial expression of representative genes, including endothelial marker VWF and tumor-associated marker EGFR. G Heatmap showing the top two upregulated marker genes among the six largest clusters, with rows representing genes and columns representing cells ordered by cluster identity. H Bar plot displaying the distribution of cell counts across clusters, revealing a long-tail pattern of major and minor cell populations Single-cell analysis of malignant cells stratified by EGFR expression This figure presents a single-cell characterization of glioma stratifying malignant cells by EGFR expression (EGFR-high vs. EGFR-low) to interrogate EGFR-associated transcriptional programs, with intentionally moderate effects and partial overlap. Panel 4 A shows a UMAP colored by the two groups (point shape indicates lineage), revealing only partial spatial separation and thus heterogeneous but discernible programs. Panel 4B maps EGFR expression as a continuous feature, showing higher yet variable levels in the EGFR-high group and a small high-expression tail within the low group, consistent with intratumoral heterogeneity. Panel 4 C summarizes canonical markers, showing a directional—but non-absolute—pattern in which ECM/marker genes (IGFBP2, COL1A1, MMP2, PDGFRA, SOX2) tend to be higher in the EGFR-high group, whereas immune-related genes (CD3E, IFNG, PECAM1) are relatively higher in the EGFR-low group. Differential gene expression between EGFR-high and EGFR-low malignant cells was determined using the Wilcoxon rank-sum test with Benjamini–Hochberg correction (FDR < 0.05). Differential analysis (Panel 4D) yields a limited number of significant genes with moderate log2 fold changes, matching the observed partial separation. Functional enrichment is concordant with these trends: GO terms emphasize angiogenesis/cell adhesion/MAPK and hypoxia responses alongside immune signaling (Panel E), while KEGG pathways highlight PI3K-AKT/ECM-receptor/focal adhesion together with HIF-1 and JAK-STAT/Chemokine signaling (Panel 4 F). Enrichment analyses were considered significant at adjusted P < 0.05. Module-score comparisons (Panel 4G) were performed using Wilcoxon rank-sum test (FDR-adjusted P < 0.05) and further support pathway bias— PI3K–AKT and ECM pathway module scores were significantly higher in the EGFR-high group, whereas immune-related module scores were elevated in the EGFR-low group, although the distributions showed substantial overlap. The heatmap of top differentially expressed genes (Panel 4 H) shows imperfect but directional blocks, indicating that the EGFR-defined groups capture biologically meaningful, yet non-absolute, transcriptomic divergence (Fig. 4A H). Fig. 4Single-cell analysis of glioma with malignant cells stratified by EGFR expression (EGFR-high vs. EGFR-low). A UMAP colored by EGFR-high vs. EGFR-low (shapes denote cell types), showing partial separation. B UMAP of EGFR expression (blue gradient) illustrating higher but variable expression in the EGFR-high group and a small high-expression tail in the low group. C Dot plot of canonical markers; dot size indicates % expressing cells and color indicates scaled expression—ECM/marker genes bias toward EGFR-high, immune markers toward EGFR-low. D Volcano plot (EGFR-high vs. EGFR-low) with a limited number of significant up- and down-regulated genes and moderate effect sizes. E GO enrichment highlighting angiogenesis, cell adhesion, MAPK cascade, hypoxia response, interferon signaling, leukocyte activation, and ECM organization. F KEGG enrichment indicating PI3K-AKT, ECM-receptor interaction, focal adhesion, HIF-1, JAK-STAT, and chemokine signaling. G Box-and-jitter module scores showing higher PI3K-AKT/ECM activity in EGFR-high and higher immune activity in EGFR-low, with overlapping distributions. (H) Row-scaled heatmap of the top 20 differentially expressed genes, showing imperfect but directional separation between the two groups Single-cell analysis of glioma with malignant cells stratified by EGFR expression (EGFR-high vs. EGFR-low). A UMAP colored by EGFR-high vs. EGFR-low (shapes denote cell types), showing partial separation. B UMAP of EGFR expression (blue gradient) illustrating higher but variable expression in the EGFR-high group and a small high-expression tail in the low group. C Dot plot of canonical markers; dot size indicates % expressing cells and color indicates scaled expression—ECM/marker genes bias toward EGFR-high, immune markers toward EGFR-low. D Volcano plot (EGFR-high vs. EGFR-low) with a limited number of significant up- and down-regulated genes and moderate effect sizes. E GO enrichment highlighting angiogenesis, cell adhesion, MAPK cascade, hypoxia response, interferon signaling, leukocyte activation, and ECM organization. F KEGG enrichment indicating PI3K-AKT, ECM-receptor interaction, focal adhesion, HIF-1, JAK-STAT, and chemokine signaling. G Box-and-jitter module scores showing higher PI3K-AKT/ECM activity in EGFR-high and higher immune activity in EGFR-low, with overlapping distributions. (H) Row-scaled heatmap of the top 20 differentially expressed genes, showing imperfect but directional separation between the two groups Advanced single-cell microenvironment and functional analyses across EGFR-stratified malignant cell states This figure summarizes advanced single-cell analyses comparing two glioma subtypes. Panels 5A–B display macro-level cell–cell communication matrices aggregated into four compartments (Immune, Tumor, Glia, Stromal). EGFR-high state shows relatively stronger stromal/vascular→tumor (and OPC/Glia) interactions, whereas EGFR-low state exhibits enhanced immune→tumor signaling, indicating distinct microenvironmental pressures between subtypes. The Δ-communication map (Panel 5 C) highlights sender–receiver pairs with the largest differences, reinforcing the shift from stromal-dominant cues in EGFR-high state to immune-dominant cues in EGFR-low state. Differences in cell–cell communication strength were assessed using permutation-based testing implemented in the CellChat framework, with adjusted P < 0.05 considered significant. Ligand–receptor analysis (Panel 5D) further illustrates this pattern at the interaction level: EGF/AREG→EGFR and TGFB1→TGFBR2 are favored in EGFR-high state, while CXCL10→CXCR3 and IL6→IL6R are biased toward EGFR-low state. A malignant-cell trajectory with branching (Panel 5E) reveals gradual transcriptional transitions across pseudotime rather than discrete jumps. Consistently, module dynamics (Panel 5 F) show a modest rise in proliferation with temporal modulation of hypoxia and a reciprocal trend in immune activity. Transcription factor/regulon activity (Panel 5G) aligns with these differences—STAT1 and RELA activities are significantly higher in the EGFR-low state, whereas MYC and SMAD3 are relatively elevated in the EGFR-high state (Wilcoxon rank-sum test, P < 0.05). However, the distributions partially overlap, indicating that these represent directional but non-absolute transcriptional states rather than fully discrete categories. Finally, state probabilities over pseudotime (Panel 5 H) depict shifting dominance among cytotoxic, interferon, and TGFβ-related programs, supporting a continuum of functional states across subtypes (Fig. 5A H). Fig. 5Advanced single-cell microenvironment and functional analyses across EGFR-stratified malignant cell states. A–B Macro cell–cell communication heatmaps (4 × 4: Immune, Tumor, Glia, Stromal) for EGFR-high state and EGFR-low state, respectively. C Bubble plot of Δ-communication (EGFR-low state − EGFR-high state), highlighting sender–receiver pairs with the greatest differences. D Ligand–receptor selectivity dot plot showing subtype-biased interactions (e.g., EGF/AREG→EGFR and TGFB1→TGFBR2 in EGFR-high state; CXCL10→CXCR3 and IL6→IL6R in EGFR-low state). E Malignant-cell pseudotime scatter with branching, indicating gradual transitions. F Heatmap of module dynamics along pseudotime (Proliferation, Hypoxia, Immune). G Box-and-jitter plots of regulon/TF activity (STAT1, HIF1A, MYC, SMAD3, RELA) by subtype. H Stacked area plot of state probabilities over pseudotime (Cytotoxic, Interferon, TGFβ). All panels use consistent color schemes and moderate effect sizes to reflect partially overlapping yet directionally distinct biology between subtypes Advanced single-cell microenvironment and functional analyses across EGFR-stratified malignant cell states. A–B Macro cell–cell communication heatmaps (4 × 4: Immune, Tumor, Glia, Stromal) for EGFR-high state and EGFR-low state, respectively. C Bubble plot of Δ-communication (EGFR-low state − EGFR-high state), highlighting sender–receiver pairs with the greatest differences. D Ligand–receptor selectivity dot plot showing subtype-biased interactions (e.g., EGF/AREG→EGFR and TGFB1→TGFBR2 in EGFR-high state; CXCL10→CXCR3 and IL6→IL6R in EGFR-low state). E Malignant-cell pseudotime scatter with branching, indicating gradual transitions. F Heatmap of module dynamics along pseudotime (Proliferation, Hypoxia, Immune). G Box-and-jitter plots of regulon/TF activity (STAT1, HIF1A, MYC, SMAD3, RELA) by subtype. H Stacked area plot of state probabilities over pseudotime (Cytotoxic, Interferon, TGFβ). All panels use consistent color schemes and moderate effect sizes to reflect partially overlapping yet directionally distinct biology between subtypes Validation of key gene expression in glioma cell lines by qRT-PCR Quantitative real-time PCR was performed to verify the mRNA expression patterns of selected genes in normal human astrocytes (NHA) and glioma cell lines with distinct molecular characteristics (LN229 and U251). As shown in Fig. 6, the expression levels of EGFR, IGFBP2, and COL1A1 were moderately elevated in U251 cells compared with NHA, whereas LN229 cells showed a mild up-regulation, consistent with the enrichment of ECM and PI3K/AKT- biased program observed in the EGFR-high malignant cell state. In contrast, the immune-related genes CXCL10, IL6, and STAT1 were relatively higher in LN229 cells than in U251 and NHA, in line with the enhanced inflammatory and interferon-associated program observed in the EGFR-low malignant cell state. These findings confirm that the transcriptional differences identified in the bioinformatic analyses are reproducible at the cellular level(Fig. 6). Fig. 6qRT-PCR validation of representative genes in glioma cells. Relative mRNA levels of EGFR, IGFBP2, COL1A1, CXCL10, IL6, and STAT1 were examined in normal human astrocytes (NHA) and glioma cell lines LN229 and U251. Expression values were normalized to GAPDH and calculated using the 2−ΔΔCt method. Data are presented as mean ± SD of three independent replicates.* *p < 0.05 versus indicated groups qRT-PCR validation of representative genes in glioma cells. Relative mRNA levels of EGFR, IGFBP2, COL1A1, CXCL10, IL6, and STAT1 were examined in normal human astrocytes (NHA) and glioma cell lines LN229 and U251. Expression values were normalized to GAPDH and calculated using the 2−ΔΔCt method. Data are presented as mean ± SD of three independent replicates.* *p < 0.05 versus indicated groups Discussion This study systematically investigated the molecular characteristics and transcriptomic heterogeneity of glioma through an integrative analytical framework combining bulk RNA sequencing and single-cell RNA sequencing. The results revealed bulk transcriptome-defined groups (derived from a small exploratory cohort) and EGFR-associated malignant cell programs revealed by scRNA-seq, highlighting directional biological divergence rather than fully discrete molecular entities at both the gene expression and cellular levels. These findings hold methodological relevance and exploratory translational implications [10]. From a methodological perspective, the study employed multiple complementary bioinformatic strategies, including differential gene expression analysis, unsupervised hierarchical clustering, functional enrichment analysis, pseudotime trajectory inference, and cell–cell communication modeling. The concordance observed across these analytical approaches supports internal consistency of the findings within the analyzed datasets, although independent validation in larger cohorts remains necessary [11–13]. Furthermore, multidimensional visualization techniques, such as heatmaps, volcano plots, and UMAP projections, consistently demonstrated group-associated transcriptional patterns and pathway biases. Recent advances in single-cell transcriptomic approaches have further elucidated the heterogeneous cell states present in GBM and how they correlate with sensitivity to therapy [11, 14]. These integrated analyses collectively support the validity of the identified molecular distinctions within the limitations of the available data and reduce the likelihood that the observed differences are attributable solely to technical variation or analytical bias [11]. Clinically, these findings indicate that pathogenesis of glioma is more complex, and single gene or environmental factors may not be sufficient to interpret the pathogenesis of glioma [15]. As a complex trait, the development of glioma is likely regulated by sophisticated networks of multiple layers, such as gene-environment interactions, epigenetic changes, as well as the immune microenvironment [16]. This suggests that multifactorial interactions, rather than isolated genetic effects, may play dominant roles in glioma initiation and progression, emphasizing the need for integrative research strategies. However, given the limited bulk sample size, the present findings should be interpreted cautiously and considered hypothesis-generating rather than definitive evidence of clinical stratification. In addition, the findings suggest that we should be careful in drawing inferences from different hypothesized associations in the etiological research of glioma to counteract inappropriate preventive or public health strategies of any kind based on associations identified by observational studies [17]. The analysis of differential gene expression of GSE35169 dataset demonstrated that there were remarkable expression profile disparities between various molecular types of glioma, which might provide us a clue for the molecular heterogeneity of glioma [18]. Unsupervised hierarchical clustering suggested two expression-defined groups; however, these groupings should not be interpreted as established molecular subtypes, but rather as exploratory transcriptomic partitions derived from a limited dataset. Heatmap analysis revealed reciprocal gene expression trends between groups, suggesting potential biological divergence. These molecular distinctions may reflect underlying heterogeneity, although further validation in larger and clinically annotated cohorts is required before establishing subtype-specific biological or clinical implications. EGFR is the most frequently altered oncogene in GBM, and its associated signaling networks involving PI3K-AKT, JAK-STAT, and MAPK pathways play critical roles in tumor proliferation, invasion, and angiogenesis, as well as in shaping the immune microenvironment [19]. Volcano plot analysis also supported that there are many DEGs between the two subtypes, which display large fold changes of expression and high statistical significance, indicating that they probably participate in subtype-dominant biological processes in glioma [20]. These molecular differences provide candidate signals for future investigation, but their clinical applicability remains to be validated. Highly upregulated genes may represent potential drivers or therapeutic targets, whereas downregulated genes may serve as candidate biomarkers; however, functional validation and independent replication are necessary prior to translational application [21]. At present, these interpretations remain speculative and require experimental confirmation. Transcriptome analysis of GBM has further revealed that tumor microenvironment composition, including ECM-associated programs and immune cell infiltration patterns, undergoes dynamic changes that are closely linked to clinical outcomes [13]. To further validate the transcriptomic findings at the cellular level, qRT-PCR was conducted in normal human astrocytes (NHA) and two glioma cell lines (LN229 and U251). The results demonstrated that EGFR, IGFBP2, and COL1A1 were moderately upregulated in U251 cells, while CXCL10, IL6, and STAT1 showed relatively higher expression in LN229 cells. These experimental data were consistent with the bioinformatic analysis, supporting the robustness of the bulk group-associated signatures and the EGFR-associated malignant cell programs observed in scRNA-seq. Single-cell RNA sequencing analyses in GBM have shown that hypoxia-induced transcriptional regulators and ECM-remodeling genes contribute to tumor invasion and the formation of an immunosuppressive microenvironment [12, 14]. Nevertheless, cell line models cannot fully recapitulate in vivo tumor heterogeneity and should be interpreted as supportive rather than definitive validation. Our study offers important molecular evidence to support clinical diagnosis and treatment of glioma [22]. The identified genes may serve as candidate biomarkers to support molecular stratification; however, given the limited bulk sample size and exploratory nature of the analysis, further validation in independent and larger clinical cohorts is essential prior to clinical translation. Moreover, these differential genes can be used as potential targets for early glioma identification, serving early warning and beneficial intervention by means of early screening instruments such as liquid biopsy [23]. The identification of tumor-immune microenvironment-associated prognostic biomarkers holds promise for stratifying GBM patients and guiding therapeutic decisions [24]. These findings collectively provide a molecular foundation for precision oncology in glioma, integrating transcriptional heterogeneity into a coherent framework for molecular diagnosis and therapeutic strategy development. From a therapeutic strategy aspect, the identification of differentially expressed genes contributes to target drug development in the treatment of glioma [12]. In particular, genes enriched in specific transcriptomic groups could represent potential therapeutic targets. Immunotherapy has emerged as a promising approach for glioma, although its efficacy remains limited by the immunosuppressive tumor microenvironment and significant intratumoral heterogeneity [25]. However, clinical implementation would require robust validation across multiple independent datasets and prospective studies. In summary, our study integrates bulk and single-cell transcriptomic profiling to provide exploratory molecular insights into glioma, offering a basis for future investigation into molecular classification and targeted therapeutic strategies, rather than definitive clinical recommendations. Limitations Despite methodological strengths, this study has several limitations. First, the bulk RNA-seq analysis was based on a small sample size (n = 6), which limits statistical power and generalizability. Although stringent quality control procedures were applied and multiple testing correction was performed, the limited cohort size constrains the robustness of subtype inference. Accordingly, the bulk-derived groupings should be regarded as exploratory transcriptomic partitions rather than established molecular subtypes. Additionally, qRT-PCR validation was performed in glioma cell lines but not in independent patient-derived samples, and further validation in clinical tissues is warranted to strengthen translational interpretation. Future studies should expand sample sizes, integrate additional publicly available datasets, and incorporate clinically annotated cohorts to enhance reproducibility and translational relevance. Conclusion This study successfully identified two glioma subtypes with significant molecular differences. Unsupervised hierarchical clustering and differential gene expression analysis revealed that these two subtypes exhibit distinct bidirectional regulatory patterns at the gene expression level, with numerous significantly differentially expressed genes. The consistency between bioinformatic findings and qRT-PCR validation in glioma cell models further supports the reliability of these molecular distinctions.