Abstract Objective Glioblastoma multiforme (GBM) presents significant therapeutic challenges due to its heterogeneous tumorigenicity, drug resistance, and immunosuppression. Although several molecular markers have been developed, there still lack of sensitive molecular for accurately detection. Studying the mechanisms underlying the development of GBM and finding relevant prognostic biomarkers remains crucial. Methods Single-cell RNA sequencing, bulk RNA-seq, and cancer immune cycle activities of GBM were used to assess the expression of different molecular related to GBM. Bioinformatics analyses were carried to evaluate the functional of the high mobility group protein B3 (HMGB3) in GBM. Results HMGB3 was highly expressed in GBM tissues and influenced the interpatient and intratumoral transcriptomic heterogeneity as well as immunosuppression in GBM. HMGB3 also contributes to a no inflamed tumor microenvironment (TME) and has an inhibitory effect on tumor-associated immune cell infiltration. Besides, HMGB3 participated GBM chemotherapeutic sensitivity and negative correlation with 140 medicines. Conclusion HMGB3 as a heterogeneous and immunosuppressive molecule in the GBM TME, making it a potential target for precision therapy for GBM. Supplementary Information The online version contains supplementary material available at 10.1007/s12672-025-02235-6. Keywords: Glioblastoma multiforme, scRNA-seq, Bulk RNA-seq, HMGB3, Tumor microenvironment Introduction Glioblastoma multiforme (GBM) is one of the most prevalent and aggressive malignancies of the central nervous system, characterized by rapid proliferation, extensive invasiveness, and high mortality rates [[30]1]. Patients diagnosed with GBM have an overall survival (OS) of approximately 14 months [[31]2, [32]3]. The standard clinical therapy for GBM patients involves surgical resection followed by radiotherapy and chemotherapy [[33]4]. However, the effectiveness of these treatments is often limited by the presence of the blood-brain barrier and the tumor’s poorly defined boundaries with surrounding tissues. GBM exhibits a heterogeneous population of genetically unstable peripheral immune cells in GBMs, including tumor-associated microglia, macrophages and regulatory T cells [[34]5–[35]8]. Although GBM is typically characterized by a low infiltration of tumor-infiltrating lymphocytes, it also demonstrates resistant to immunotherapy [[36]9]. Evidence of immunoreactivity in the lymphatic vessels and dural sinuses suggests that immunotherapy could serve as a potential adjunctive treatment for GBM patients [[37]10]. Evidence indicates that immunotherapeutic approaches, such as immune checkpoint blockade (ICB), show promising results in reducing the tumor burden of GBM Emerging [[38]11]. However, sustained therapeutic responses are rare, largely due to the highly immunosuppressive tumor microenvironment and heterogeneous and plasticity observed at the single-cell levels [[39]12]. To overcome these challenges, a deeper understanding of the factors influencing cellular transcriptomic heterogeneity and the identification of immunosuppressive targets are essential. The tumor microenvironment (TME) constitutes a complex and dynamic ecosystem characterized by intricate interactions between neoplastic and non-neoplastic components. These interactions yet synergistic foster a chronic inflammatory, immunosuppressive, and tumor-promoting milieu that supports neoplastic progression [[40]13–[41]16]. Consequently, the identification of key molecular targets capable of reversing tumor-mediated immunosuppression represents a critical step in developing effective combination therapies [[42]17]. Single-cell RNA sequencing (scRNA-seq) is a powerful tool for dissecting the lineage identity and heterogeneity of cancers. This technology offers significant potential for discover immune system heterogeneity and develop personalized therapeutic strategies [[43]18, [44]19]. Based on the scRNA-seq analysis, Jiang et al. [[45]20] had discovered transcriptional heterogeneity in organ-specific metastasis of gastric cancer (GC). Smalley et al. [[46]21] identified that dendritic cells could positively regulate the immune environment through modulation of activated T cells and major histocompatibility complex (MHC) expression and increase OS. Bulk RNA-seq is an important tool for understanding genome-wide transcriptome variation. Integrating scRNA-seq with bulk RNA-seq has proven effective in revealing dynamic changes within the TME, particularly in immunology [[47]22]. Joanito et al. used a combined analysis to characterize the status of two epithelial tumor cell populations and refine the molecular classification of colorectal cancer [[48]23]. Kang et al. employed scRNA-seq combined with bulk RNA-seq to identify key characteristics of gastric cancer’s TME using scRNA-seq combining with bulk RNA-seq [[49]24]. Gong et al. further applied this integrated approach to elucidate the dynamics of stromal and tumor-specific characteristics of nasopharyngeal carcinoma microenvironments [[50]25]. Despite these advancements, the analysis of TME in GBM is still insufficient, particularly in understanding anti-tumor immunity. Recent studies have highlighted that the efficacy of anti-tumor immunity in cancer cells is closely linked to the activity of the cancer immune cycle [[51]26]. Wang et al. [[52]27] combined the cancer-immunity cycle with bulk RNA-seq to identify ZNF207 as a key promoter of HCC immunosuppression. High-mobility group box 3 (HMGB3) is an important proinflammatory cytokine that plays a pivotal role in DNA repair, transcription, tumorigenesis, and chemotherapy resistance [[53]28, [54]29]. Previous studies have a pivotal role that HMGB3 is high expressed in non-small cell lung cancer (NSCLC) and breast carcinoma (BC), and its elevated expression is associated with poor clinical outcomes [[55]30, [56]31]. Nevertheless, the specific mechanisms by which HMGB3 accelerates GBM progression remain poorly understood. The role of HMGB3 in modulating tumor immunity has not been thoroughly explored. In this study, we integrated scRNA-seq with bulk RNA-seq to systematically identify potential immunosuppressive targets in GBM. Our findings suggest that HMGB3 may serve as a key molecule contributing to interpatient and intratumoral transcriptome heterogeneity, as well as a potential immune suppressive target for GBM. Furthermore, we reveal that HMGB3 contributes to the formation of a noninflamed TME and may serve as a predictive biomarker for clinical response to immune checkpoint blockade (ICB) therapy. Methods Data collection The single-cell transcriptomic characterization of the GBM scRNA dataset was derived from the Gene Expression Omnibus (GEO) database [[57]32]. The accession number was [58]GSE224090, and the platform utilized was [59]GPL24676. The bulk RNA-seq data for the normal brain (NB) tissue samples (n = 5) and the GBM samples (n = 170) were retrieved from The Cancer Genome Atlas (TCGA) [[60]33]. The corresponding survival information was obtained from the UCSC Xena data portal [[61]34]. The cancer-immunity cycle scores for GBM samples were acquired from the Tracking Tumor Immunophenotype (TIP) database [[62]35]. The expression of HMGB3 protein were evaluated using the Human Protein Atlas (HPA) database [[63]36]. In instances where multiple probes corresponded to the same gene symbol, the highest value was retained as the final expression measurement. Additionally, the data were transformed using the log 2^(x + 1) method. The pathways associated with tumor progression were sourced from the Molecular Signature Database (MSigDB) [[64]37]. All data were procured from publicly available databases; therefore, ethics approval and informed consent were not necessary. ScRNA analysis The scRNA-seq data collected from the GBM samples represented by [65]GSM7011674 and [66]GSM7011676 were merged using the “merge” function in the R “Seurat” package [[67]38]. Cells of suboptimal quality were excluded based on the following criteria: (1) fewer than 5 cells, (2) fewer than 500 features, (3) mitochondrial gene expression less than 5%, (4) ribosome gene expression less than 40%, and (5) erythroid gene expression less than 0.1%. Gene expression normalization was conducted using the Log Normalize method. The top 2000 genes exhibiting the highest intercellular coefficient of variation were identified utilizing the Find Variable Features function [[68]39, [69]40]. The software package “harmony” was employed to integrate data from multiple samples [[70]41]. The Uniform Manifold Approximation and Projection (UMAP) algorithm was utilized to reduce dimensionality and visualize clustered cells [[71]42]. Cell subpopulations were identified using the R “SingleR” package [[72]43]. Marker genes for various cell types were determined utilizing the “FindAllMarkers” function with the parameters min. pct = 0.25, only. Pos = TRUE, and logfc.threshold = 1.0 and the Wilcoxon rank sum test. WGCNA and hub gene screening The activities of the cancer-immunity cycle and mRNA data obtained from GBM patients underwent comprehensive analysis using the R “WGCNA” package [[73]44]. mRNAs exhibiting high variability, defined as a median absolute deviation (MAD) greater than 1 in the normalized expression profile were retained. All outlier samples identified in the clustering dendrograms were excluded prior to applying WGCNA. A fit value of 9 was chosen to ensure that the co-expression network maintained a scale-free structure. The similarity matrix was subsequently transformed into an adjacency matrix, which was further transformed into a topological overlap matrix (TOM). To avoid the formation of excessively small or large modules, the blockwisemodules function was applied to cluster genes with similar expression profiles or function into different modules using the hybrid dynamic tree-cutting algorithm. Module eigengenes (MEs) were used to assess the association between the modules and each phenotype using Pearson’s correlation test. The module exhibiting the strongest association coefficient was recognized as the central module. Simultaneously, analyzing the correlation between the all genes and central module, genes with a correlation >0.8 were selected as candidate genes. Finally, genes contained in the marker genes for different cell types and candidate genes from WGCNA were defined as hub genes. Association of HMGB3 with immunological characteristics of TME and ICB response Following the methodology of a previous study [[74]45], we conducted a multi-dimensional evaluation of the relationship between HMGB3 and immunologic characteristics of TME in GBM. To prevent calculation errors stemming from different in algorithms, we used eight independent algorithms, including CIBERSORT-ABS, MCP-counter, quantIseq, TIMER, xCell, EPIC, TISIDB, and TIP, to comprehensively calculate the infiltration level of TIICs in GBM. The infiltration level of TIICs for all algorithms, excluding TISIDB and TIP, were obtained from the TIMER 2.0 web tool [[75]46]. Additionally, we collected effector genes of TIICs and investigated their correlation with HMGB3 in GBM. Furthermore, markers of ICB and the pan-cancer T-cell inflamed score were utilized to evaluate host susceptibility to ICB [[76]47]. Generating immune-related DEGs (IRDEGs) The immune score and stromal score within the TME were calculated using the ESTIMATE R package. Based on the median immune score, stromal score, and HMGB3 mRNA expression, GBM patients were categorized into two groups. The R “Limma” package [[77]48] was applied to designate DEGs associated with positive or negative HMGB3, immune, and stromal factors from the RNA-seq data. The common DEGs across the different groups were defined as IRDEGs. Subsequently, GO and KEGG analyses were applied using the R “ClusterProfiler” [[78]49] package to investigate the potential biological functions associated with the HMGB3‐related DEGs in GBM. Analysis of the sensitivity of chemotherapy To further examine the potential role of HMGB3 in guiding chemotherapy, the R “oncoPredict” package was used to extract the IC50/AUC values of the chemotherapy drugs [[79]50]. Simultaneously, the relationship between HMGB3 mRNA expression and IC50/AUC values of all drugs was analyzed, and drugs with an absolute correlation of |correlation| > 0.5 were selected. Differences in IC50/AUC values of the drugs between high and low HMGB3 mRNA expression samples were compared. Results Identification of GBM cell subtypes Based on ScRNA-seq data analysis, all cells were grouped into 21 categories. Further annotation was carried and the cells were divided into six cell types: macrophages, natural killer (NK) cells, B cells, CD8^+ T cells, monocytes, and astrocytes (Fig. [80]1). Additionally, 1150 specific marker genes were defined (Supplementary Table [81]1). Fig. 1. [82]Fig. 1 [83]Open in a new tab ScRNA-seq data from two human samples of GBM. A–C The UMAP plot shows the aggregated data from the two specimens, which were classified into six cell clusters. The plot is colored by individual samples. D The dot plot displays the 5 marker genes with the highest expression levels for each cell cluster HMGB3 is a potential immunosuppressive target After excluding mRNAs and samples of suboptimal quality, 19,434 mRNA and 167 GBM samples were selected for WGCNA network analysis. The genes were divided into 15 modules in cluster analysis (Fig. [84]2A–D). Further analysis revealed that the brown module was positively associated with macrophage recruitment (R = 0.62, p = 8e−19), while the green module was negatively associated with infiltration of immune cells (R = −0.65, p = 2e−21). Thus, the two modules were deemed core for further analysis. GO and KEGG pathway analysis showed that genes within the brown module exhibited enrichment in immune-related pathways. Meanwhile, genes within the green module were mainly enriched in cell cycle (Fig. [85]2E, F). For immunosuppressive targets in GBM, we examined the green module. Based on the screening criteria, HMGB3, ILF2, CENPU, LMNB1, MCM7, and CENPF were found as the core genes (Fig. [86]2G). Fig. 2. [87]Fig. 2 [88]Open in a new tab Selection of an immunosuppressive module related to GBM. A–C Highly variable genes were categorized into 15 modules. D A heatmap was created to display the correlation and p value of each module with the scores of the cancer immunity cycle. E,F Two key modules were analyzed for GO and KEGG enrichment analysis. G Venn diagram of common genes identified among the MMhub genes and marker genes of cell subtypes. H Correlation between HMGB3 and the immune escape signature. I,J Correlation between HMGB3 and complement immune responses The correlation analysis proved that HMGB3 had a positive correlation with immune evasion signature [[89]51] in GBM tissue (Fig. [90]2H). Therefore, we focused mainly on HMGB3. Correlation analysis revealed that HMGB3 has strong negative correlations with C2 and C3, which were associated with complement immune responses (F[91]ig. [92]2I, [93]J). Based on the bioinformatics analysis, HMGB3 was identified as a potential target for cellular transcriptomic heterogeneity and immunosuppression in GBM. HMGB3 was highly expressed and may serve as an indicator of TME modulation in GBM The mRNA expression of HMGB3 in matched GBM tissues was higher than that in NB tissues (Fig. [94]3A). Data from immunohistochemistry (IHC) in the HPA database revealed a consistent trend and HMGB3 was mainly overexpressed in the nucleus (Fig. [95]3B). The univariate Cox regression, Kaplan–Meier curves, and log-rank test results indicated that HMGB3 did not affect the prognosis of GBM concerning OS and disease-specific survival (DSS). However, higher expression of HMGB3 was correlated with longer progression-free survival (PFS) in GBM (Fig. [96]3C). We then analyzed how HMGB3 contributed to the advancement of GBM through gene set enrichment analysis (GSEA). The results demonstrated that genes in the HMGB3^high group are significantly involved in negative regulation of epidermis development, mitotic DNA replication, noradrenergic neuron differentiation, and double-strand break repair, etc. On the other hand, genes in the HMGB3^low group are associated with processes such as killing by a host of symbiont cells, defense response to fungus, complement activation, response to fungus, and monocyte chemotaxis (Fig. [97]3D). Furthermore, KEGG pathway analysis revealed enrichment of pathways like the notch signaling pathway, mismatch repair, cell cycle, and DNA replication in the HMGB3^high group, whereas the HMGB3^low group was mainly enriched in pathways related to asthma, graft versus host disease, and allograft rejection (Fig. [98]3E). The genes in the HMGB3^low and HMGB3^high groups was underwent GSVA analysis. The results revealed that the HMGB3^high group was activated in markers associated with G[2]M checkpoint, E[2]F targets, mitotic spindle, etc. In contrast, the HMGB3^low group was primarily activated in markers associated with coagulation, xenobiotic metabolism, inflammatory response, and other markers (Fig. [99]3F). These findings revealed that HMGB3 could serve as a promising indicator of the TME status. Fig. 3. [100]Fig. 3 [101]Open in a new tab HMGB3 is aberrantly expressed in GBM and is associated with TME remodeling. A HMGB3 is highly expressed in the TCGA GBM cohort. B Immunohistochemistry was used to measure the HMGB3 protein level in cerebral cortex and high-grade glioma samples. C Prognostic relationship between HMGB3 and patients with GBM. D,E GSEA of GO and KEGG in HMGB3^high and HMGB3^low groups. F GSVA analysis of all genes in the HMGB3^high and HMGB3^low groups Correlations of HMGB3 with anticancer immunity of the TME in GBM Cluster analysis revealed that HMGB3 have a negative relevance with the majority of immunomodulators (Fig. [102]4A). In the HMGB3^high group, a significant decrease in the expression of numerous MHC-related molecules was observed. Which indicating a compromised antigen presentation and processing capacity in the group. Pivotal monocyte or macrophage chemokines were also found to be downregulated in the HMGB3^low group, leading to reduced inflammatory response and monocyte or macrophage phagocytosis in GBM. The results also indicated that the HMGB3^low group exhibited a diminished anti-cancer immune profile across various steps of the immune cycle (Fig. [103]4B). It should be noted that the HMGB3^low group has a more robust immune status, which likely contributes to decrease TIIC infiltration in the GBM-TME. Our observation shows a positive correlation between HMGB3 mRNA expression and T cell recognition of cancer cells. The HMGB3^high group suggesting an improved ability of T cell receptors to recognize cancer cells. Additionally, we found that priming and activation and cancer cell killing activities were increased in the HMGB3^high group. Subsequent analysis of the infiltration of TIICs revealed a negative correlation between the infiltration level of TIICs (neutrophils, macrophages, monocytes, CD8^+ T cells, NK cells, and dendritic (DC) cells) and HMGB3 in various algorithms (Fig. [104]4C). Besides, the effector genes of these TIICs were decreased in the HMGB3^high group (Fig. [105]4D). Consistent with previous findings, HMGB3 showed a robust negative correlation with macrophage marker genes (CD11b, CD45, CD68, and EMR1) (Fig. [106]4E). Fig. 4. [107]Fig. 4 [108]Open in a new tab HMGB3 has a negative association with anti-tumor immunity in the TME of GBM. A Variances in the 122 immunomodulators were observed in both the high- and low-HMGB3 expression groups. B Variances in the cancer immunity cycle scores were observed between the high- and low-HMGB3 expression groups. C The associations between HMGB3 and the infiltration level of neutrophils, macrophages, monocytes, CD8^+ T cells, NK cells, and DC cells, which were calculated using eight independent algorithms. D Variances in the effector genes of the above TIICs between the high and low HMGB3 expression groups. E The relationship between HMGB3 and the effector genes of macrophages. F The correlation between HMGB3 and 20 immune checkpoint inhibitors Our analysis also identified a strong inverse relationship between HMGB3 and inhibitory immune checkpoints (PD-L1, CD274, CD86, CD200R1, etc.) (Fig. [109]4F), pan-cancer T cell inflamed score, and T-cell inflammatory genes (Fig. [110]5A–C). These findings emphasize the potential significance of HMGB3 in shaping a noninflamed TME and its influence on the effectiveness of immunotherapy in GBM. At last, we confirmed the association between HMGB3 and TME in the NU GBM cohort. Notably, HMGB3 expression was consistently associated with immunomodulator expression, cancer immune cycle activities, TIIC infiltration levels, immune checkpoint expression, pan-cancer T-cell inflamed score, and T-cell inflammatory genes (Fig. [111]5D–J). In conclusion, we have demonstrated that increased of HMGB3 created a non-inflamed TME in GBM, whereas decreased of HMGB3 correlated with an inflamed TME in GBM. Fig. 5. [112]Fig. 5 [113]Open in a new tab HMGB3 can be used to predict the response to ICB in GBM and validate the function of HMGB3 by the NU GBM cohort. A,B The correlations between HMGB3 mRNA expression and T-cell inflamed score and T-cell inflammatory genes. C Differences in T-cell inflammatory genes between high- and low-HMGB3 groups. D Differences in the 122 immunomodulators between the high- and low-HMGB3 groups in the NU GBM cohort. E Correlation between HMGB3 and cancer immunity cycle score. F Correlation between HMGB3 and 28 TIICs were calculated by the ssGSEA algorithm. F–J Associations between HMGB3 and marker genes of macrophages, inhibitory immune checkpoints, T-cell inflamed score, and T-cell inflammatory genes, respectively The mechanism of HMGB3 in forming the non-inflamed TME A total of 268 genes were identified as HMGB3-related DEGs from the TCGA-GBM cohort (Fig. [114]6A). The immune score and stromal score were calculated using the ESTIMATE algorithm. Subsequently, DEGs were determined for different score groups (Fig. [115]6B, C). Interestingly, there was no overlap between HMGB3-positive or -negative related DEGs and immune or stromal scores. These finding suggest that HMGB3 may negatively regulate immune-related components of the TME in GBM (Fig. [116]6D). GO and KEGG analysis revealed that HMGB3-negative-related DEGs were significantly enriched in pathways associated with cell chemotaxis and migration, cytokine-cytokine receptor interaction, and IL-17 signaling (Fig. [117]6E, F). These findings indicate that HMGB3 might play a role in establishing a non-inflamed TME in GBM by suppressing immune-related pathways. Fig. 6. [118]Fig. 6 [119]Open in a new tab The biological functions of HMGB3 in TCGA-GBM. A The volcano plot of HMGB3-related DEGs. B The volcano plot of stromal score-related DEGs. C The volcano plot of immune score-related DEGs. D The analysis focused on DEGs related to HMGB3, immune score and stromal score interactions. E,F Bubble plots of the GO and KEGG pathways of HMGB3-related DEGs The drug sensitivity analysis related to HMGB3 The GDSC database revealed a negative correlation between 28 chemotherapeutic drugs and HMGB3 (R < −0.5 and p < 0.05). Figure [120]7A displayed the top six drugs. The names, targets and pathways of 22 of these drugs were showed in Table [121]1. The sensitivity of 22 chemotherapeutic drugs was significantly downregulated in the HMGB3^high group (Fig. [122]7B). Evaluation of the CTRP database revealed that 10 drugs showed the strongest negative association with HMGB3 (R < −0.65 and p < 0.05) (Fig. [123]7C). The AUC values of these drugs differed significantly between the HMGB3^high and HMGB3^hlow groups (Fig. [124]7D). Taken together, these results suggest that these chemotherapeutic agents are promising for the treatment of GBM. Fig. 7. [125]Fig. 7 [126]Open in a new tab Drugs and compounds for the treatment of GBM are evaluated by assessing HMGB3 expression. A The list of 28 potential chemotherapeutic drugs was generated using the GDSC database. B The sensitivity of these 28 drugs was compared in the high- and low-HMGB3 expression groups. C The top ten compounds with the strongest negative correlation to HMGB3 were identified using the CTRP database. D Differences in AUC value of these ten compounds between high and low HMGB3 groups Table 1. Names, IDs, targets and pathways of action of 22 drug compounds Drug name Drug ID Drug target Target pathway Screening set Tozasertib 1096 AURKA, AURKB, AURKC, others Mitosis GDSC2 UMI-77 1939 MCL1 Apoptosis regulation GDSC2 WEHI-539 1997 BCL-XL Apoptosis regulation GDSC2 [127]P22077 1933 USP7, USP47 Protein stability and degradation GDSC2 Pyridostatin 2044 G-quadruplex stabiliser DNA replication GDSC2 Telomerase Inhibitor IX 1930 Telomerase Genome integrity GDSC2 Wee1 Inhibitor 1046 WEE1, CHEK1 Cell cycle GDSC2 MK-1775 1179 WEE1, PLK1 Cell cycle GDSC2 Navitoclax 1011 BCL2, BCL-XL, BCL-W Apoptosis regulation GDSC2 I-BRD9 1928 BRD9 Chromatin other GDSC2 MIRA-1 1931 TP53 p53 pathway GDSC2 Cediranib 1922 VEGFR, FLT1, FLT2, FLT3, FLT4, KIT, PDGFRB RTK signaling GDSC2 ML323 1629 USP1, UAF1 Protein stability and degradation GDSC2 BMS-345541 1249 IKK-1, IKK-2 Other GDSC2 Fulvestrant 1200 ESR Hormone-related GDSC2 GSK1904529A 1093 IGF1R, IR IGF1R signaling GDSC2 Ipatasertib 1924 AKT1, AKT, AKT3 PI3K/mTOR signaling GDSC2 MIM1 1996 MCL1 Apoptosis regulation GDSC2 BIBR-1532 2043 TERT Genome integrity GDSC2 Axitinib 1021 PDGFR, KIT, VEGFR RTK signaling GDSC2 Sepantronium bromide 1941 BIRC5 Apoptosis regulation GDSC2 Uprosertib 2106 AKT1, AKT2, AKT3 PI3K/mTOR signaling GDSC2 [128]Open in a new tab Discussion Despite numerous efforts in the treatment of GBM, its intratumoral heterogeneity and immunosuppressive properties pose significant limitations for therapeutic evaluation [[129]52]. There is a pressing need for innovative approaches to identify immunosuppressive targets that can aid in the development of effective therapeutic strategies. In contrast to bulk RNA-seq, which primarily focuses on average gene expression levels, scRNA-seq technology is critical for characterizing cellular subpopulations, discovering novel biomarkers, and enhancing the understanding of the heterogeneity of different cell types across various cancers [[130]53, [131]54]. In the present study, we performed a combined analysis of bulk RNA-seq and scRNA-seq to identify molecules exhibiting both tumor heterogeneity and immunosuppressive functions. Meanwhile, 1150 different marker genes were identified. Through WGCNA analysis, we identified 886 genes as critical for the immunosuppression model. GO and KEGG analysis revealed significant enrichment of these gene in processes of related to the cell cycle, DNA replication, and homologous recombination. Additionally, we identified 173 MMhub genes as candidate immunosuppressive targets in GBM. Four candidate genes were identified by the overlap between the two sets of genes, including HMGB3, ILF2, CENPU and MCM7. Given that immune-related functions influence clinical outcome in patients, and HMGB3 has been previously associated with immunity and inflammation [[132]55], we selected HMGB3 as a potential immunosuppressive target in GBM for further research. HMGB3, also known as HMG2A or HMG4, is involved in processes such as DNA repair, replication, transcription and recombination. Its involvement has been implicated in the development, onset, and maintenance of dedifferentiation and reduced survival in various cancers, including cervical carcinoma [[133]28], ovarian carcinoma [[134]29], NSCLC [[135]30], BC [[136]31], colorectal carcinoma [[137]56], and leukemia [[138]57]. Additionally, HMGB3 has been shown to play a vital role in therapy resistance across multiple cancers [[139]28, [140]29]. Previous reports have established that a molecule’s suitability as an immunotherapy target is largely dependent on its specific overexpression and immunosuppressive function in the TME [[141]58]. It has been observed that HMGB3 mRNA expression is significantly elevated in GBM, suggesting that anti-HMGB3 treatment may have fewer side effects. The development of the cancer-immunity cycle reflects the overall effect of interactions between cancer cells and the immune system. An interesting finding was that the negative correlation between HMGB3 and various stages of the cancer-immunity cycle. The findings suggested a potential immunosuppressive role for HMGB3 in GBM. In addition, there was a significant decrease in the activity of T cell recruitment in the HMGB3^high group, leading to reduced infiltration levels of several key immune cells. Moreover, HMGB3 showed a negative correlation with several inhibitory immune checkpoints, including PD-L1, CD86, CD200R1, HAVCR2, LAIR1, IDO1, CEACAM1, and LGALS3. Additionally, the immune checkpoints gene were significantly downregulated in the HMGB3^high group. The results indicate that GBM with high HMGB3 expression may not respond well to ICB therapy. Furthermore, HMGB3 was inversely associated with the pan-cancer T cell-inflamed score and with effector genes relevant to ICB. These findings were verified in external validation cohorts and support the conclusion that HMGB3 defines a non-inflamed TME. Therefore, it is critical to explore alternative treatment approaches for GBM, especially in cases with high levels of HMGB3 expression. Theoretically, a potential treatment approach for GBM with high levels of HMGB3 would be to convert a non-inflamed TME into an inflamed TME to stimulate an anti-cancer immune response. However, the presence of immune checkpoints could increase due to negative feedback regulatory mechanisms. Consequently, subsequent treatment with ICB may help revive suppressed anti-cancer immunity and improve the efficacy of anti-HMGB3 therapy. Several preliminary phase II and III clinical trials have shown that combining of anti-PD1/PD-L1 with anti-CTLA4 or anti-angiogenic agents is more effective than monotherapy [[142]59, [143]60]. The negative correlation between HMGB3 and specific ICBs suggests that combining anti-HMGB3 therapy with current ICB therapies may offer complementary benefits in the treatment of GBM. Our results indicated that HMGB3 contributes to a non-inflamed TME by suppressing the activities of immune-related cytokine/chemokine signaling pathways in GBM. We also found that these druggable targets and their corresponding compounds were negatively associated with HMGB3. The drug targets included those related to the cell cycle (Wee1 inhibitor and MK-1775), apoptosis regulation (UMI-77, WEHI-539, Navitoclax, MIM1, Sepantronium bromide, and BRD-K35604418), RTK signaling (Cediranib and Axitinib), protein stability and degradation ([144]P22077 and ML323), PI3K/MTOR signaling (Ipatasertib and Uprosertib), genome integrity (Telomerase Inhibitor IX and BIBR-1532), and p53 signaling (MIRA-1 and SCH-529074). The results are consistent with the pathway enrichment analysis. The genes in the immunosuppression module and DEGs positively associated with HMGB3 indicate potential as effective antitumor drugs. The limitations of this study include the relatively small sample size of the scRNA data and the use of median HMGB3 mRNA expression as a cutoff. Further experiments are needed to elucidate the role of HMGB3 in promoting GBM progression through tumor immunity. Conclusion This study systematically identified HMGB3 plays a role in interpatient and intratumoral transcriptomic heterogeneity, as well as cellular and immunosuppressive functions in GBM. HMGB3 represents a promising immunotherapy target for GBM in the future. Supplementary Information [145]12672_2025_2235_MOESM1_ESM.xlsx^ (82.6KB, xlsx) Additional file 1: Table S1 1150 significantly different marker genes. (XLSX 83 KB) Acknowledgements