The ethical committee of The First Hospital of China Medical University approved the biological experiments conducted in this study (Ethics number: AF-SOP-07-1.2-01).

LncRNA expression data for HGG patients was obtained from The Atlas of Noncoding RNAs in Cancer (TANRIC) (15), incorporating data from the TCGA glioma RNA-seq database (http://cancergenome.nih.gov/) and independent RNA-seq studies in the CGGA database (16). Coding mRNA expression profiles were sourced from the TCGA and CGGA-325 datasets (http://www.cgga.org.cn). Sample annotations were performed using barcode IDs based on clinical information from the UCSC Xena (https://xenabrowser.net/datapages/) and CGGA databases. Detailed clinical and molecular information is provided in Supplementary Table S1 .

Fifteen clinically relevant hypoxia-related metagenes (HRMG, Supplementary Table S2 ) were curated from literature reviews (17). To identify hypoxia subtypes within the TCGA-HGG cohort, the expression matrix of these metagenes was used with the Consensus Cluster Plus R package, utilizing Euclidean distance and K-means clustering. Cluster-consensus, delta area from unsupervised consensus clustering, and average silhouette width from the Silhouette R package were employed to validate clustering stability.

To develop the hypoxia-related lncRNA signature, we utilized lncRNA expression profiles from TCGA RNAseq as a training set. Differentially expressed lncRNAs (|logFC| > 1, FDR < 0.001) were identified in comparisons of hypoxia clusters (cluster 1 vs. cluster 3, cluster 1 vs. cluster 4, cluster 2 vs. cluster 3, cluster 2 vs. cluster 4) using the Limma package as descibed in the previous study (18). Prognosis-related lncRNAs (p-value < 0.05) were identified via univariate Cox regression analyses of HGG, WHO grade 3, and WHO grade 4 glioma cohorts from TCGA RNAseq. Pearson correlation analysis (|Pearson r| > 0.5, FDR < 0.001) was used to correlate lncRNAs with hypoxia metagenes. Eight lncRNAs were identified by overlapping the above five correlated gene lists to construct a hypoxia-related lncRNA signature. Risk scores were computed by linearly aggregating the expression values of these lncRNAs weighted by coefficients from univariate Cox regression analysis in TCGA. These coefficients were also applied to calculate risk scores in the validation dataset (CGGA).

Pearson’s correlation analysis was used to determine the relationship between coding mRNA expression and risk scores in HGG. Genes with Pearson r > 0.7 and p-value < 0.05 were selected for Gene Ontology (GO) analysis using the ClueGO plug-in in Cytoscape.

Scores for gene sets, including HRMG, immunosuppressive gene sets, immune-cell-related gene sets, and gene programs and pathway signatures (19–21), were calculated using ssGSEA. The stromal score, immune score, and tumor purity were calculated using the ESTIMATE R package (22).

Human glioma cell lines LN229 and U87 were procured from the American Type Culture Collection (ATCC, Manassas, VA, USA). Initially, cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM, Gibco) supplemented with 10% fetal bovine serum (Gibco). Upon reaching logarithmic growth phase, cells were enzymatically dissociated into single cells using 0.25% trypsin (Gibco). All cell lines underwent fewer than 20 passages and passed mycoplasma and short tandem repeat (STR) DNA profiling tests.

Total RNA was extracted using TRIzol (TaKaRa, Japan) as per the manufacturer’s instructions. Subsequently, RNA was reverse-transcribed into cDNA using PrimeScript™ RT Master Mix (RR036A, Takara). Quantitative PCR was conducted with TB Green^® Premix Ex Taq™ (RR420A, Takara) using a LightCyclerR480 (Roche Diagnostics Ltd., Basel, Switzerland) under identical amplification conditions. Each reaction was performed in triplicate. Gene expression levels were normalized to 18S and quantified using the 2^-△△Ct method. The primer sequences were as follows: TP73-AS1 forward, 5’-CTCCGGACACTGTGTTTTCTC-3’; TP73-AS1 reverse, 5’-TCTTTTAAGGCGGCCATATC-3’; 18S forward, 5’-GCAGAATCCACGCCAGTACAAGAT-3’; 18S reverse, 5’-TCTTCTTCAGTCGCTCCAGGTCTT-3’.

Proteins from glioma cell lines were extracted using RIPA lysis buffer with PMSF. Protein concentration was determined using the BCA method, and 15-30 µg of protein was loaded per well and separated by SDS-PAGE. Proteins were transferred onto PVDF membranes (0.22 μm pore size) and incubated with primary antibodies against GAPDH (1:5000; 10494-1-AP; ProteinTech) and HIF-1A (1:2000; 20960-1-AP; ProteinTech). After incubation with secondary antibody (ProteinTech), signals were detected using the chemiluminescence Femto-sig ECL kit (Tanon, Shanghai, China).

Consent was obtained from all glioma patients, and the use of human samples for IHC was approved by the Institutional Review Board of The First Affiliated Hospital of China Medical University. Brain tumor tissues were fixed in 4% paraformaldehyde, paraffin-embedded, and sectioned into 4 µm slices. Sections were incubated with primary antibodies against IBA-1 (1:500; 10904-1-AP, Proteintech), HIF-1A (1:500; 20960-1-AP, Proteintech), and TGF-β (1:500; 21898-1-AP, Proteintech). Imaging was performed using an optical microscope (Olympus, Tokyo, Japan), and German immunohistochemical staining intensity was applied.

THP-1 cells were differentiated into macrophages by treatment with 100 nM phorbol 12-myristate 13-acetate (PMA; Sigma) for 24 hours. After differentiation, 2×10⁵ THP-1 cells were seeded into the lower chamber of a 24-well plate in RPMI-1640 medium supplemented with 10% FBS. LN229 cells were seeded at a density of 1×10⁵ cells per well in the upper chamber of a co-culture insert with a 3 µm pore polycarbonate membrane (Corning, Corning, NY, USA) in medium under different conditions (si-NC + DMSO (control), si-NC + CoCl₂, si-TP73-AS1 + CoCl₂). Co-culture was maintained for 48 hours to allow interaction between LN229 and THP-1 cells. Following co-culture, transwell inserts were removed, THP-1 cells from the lower chamber were collected, and PE-CD206 (PE-65155, Proteintech), FITC-MHC II (FITC-65218, Proteintech), and APC-CD11b (101211, BioLegend) flow cytometry antibodies were then added and incubated at 4°C for 30 min while in the dark. The unbound antibodies were cleaned with pre-cooled PBS after centrifugation, and the supernatant was then discarded. The fluorescence values of the macrophage differentiation antigen on cell surface were measured using a flow cytometer LS Fortessa (BD Biosciences).

FISH was performed to visualize the location of TP73-AS1 in glioma using Cy3-labeled probes synthesized by Servicebio (Wuhan, China). Hybridization was conducted overnight following the manufacturer’s instructions, and images were captured using a laser scanning confocal microscope (Olympus, Tokyo, Japan).

Clinical glioma tissue samples were fixed in 4% paraformaldehyde, embedded in paraffin, and sectioned at 4 µm. Sections were processed by Wuhan Servicebio Technology Co., Ltd. for multicolor immunofluorescence staining. After antigen retrieval and blocking, primary antibodies against HIF-1α (green; 20960-1-AP, Proteintech) and IBA-1 (yellow; 10904-1-AP, Proteintech) were applied overnight at 4°C. Following washing, sections were incubated with Alexa Fluor 488-conjugated anti-rabbit IgG and Alexa Fluor 594-conjugated anti-mouse IgG secondary antibodies. FISH staining for TP73-AS1 (red) was conducted using a commercial kit, and nuclei were stained with DAPI (blue). Images were acquired using a microscope (Olympus).

In FISH and IHC staining results, three fields of view from the same specimen were selected for positive cell counting, and mean values were calculated. Correlation analysis was performed using GraphPad, and regression curves were plotted.

To induce hypoxia, CoCl₂ was used as a hypoxia-mimetic agent (23). LN229 and U87 cells were seeded at a density of 5×10⁵ cells per well in 6-well plates and allowed to adhere overnight. The next day, medium was replaced with DMEM containing 100 µM CoCl₂ (Macklin, Shanghai). Cells were exposed to CoCl₂ for varying duration (6, 12, and 36 hours) to assess hypoxia induction. Post-treatment, cells were washed with PBS, and HIF-1α levels were evaluated by Western blotting to confirm hypoxia induction.

siRNAs targeting TP73-AS1 were synthesized by Ribobio (Guangzhou, China) with the following sequences: si-TP73-AS1-1 sense, 5’-GTTTCCTGCTTTCCAAGTAAA-3’; si-TP73-AS1-2 sense, 5’-CGGTGTAAATTGACTCAGAAA-3’; and si-TP73-AS1-3 sense, 5’-CTTTCCTTTAACTCAAGTAAA-3’. Suppression of TP73-AS1 was validated using qRT-PCR.

LN229 glioma cells were transfected with siRNA negative control (si-NC) or siRNAs targeting TP73-AS1 (si-TP73-AS1-1 and si-TP73-AS1-2). After 48 hours, transfected cells were trypsinized, counted, and seeded in 12-well plates at a density of 500 cells per well. Cells were cultured in DMEM supplemented with 10% fetal bovine serum for 14 days to allow colony formation. Colonies were fixed with 4% paraformaldehyde, stained with 0.1% crystal violet, and counted. The experiment was performed in triplicate to assess the effect of TP73-AS1 knockdown on LN229 cell clonogenic potential.

Cell viability of glioma cells was assessed using a Cell Counting Kit-8 (Beyotime) according to the manufacturer’s protocol. Glioma cells were seeded in triplicate in 96-well plates at a density of 1×10³ cells/well and incubated for various duration (0, 24, 48, 72, 96, and 120 hours). Cell viability was measured using a microplate reader (BioTek).

THP-1 cells were differentiated into macrophages by treatment with 100 nM PMA (Sigma) for 24 hours. Following differentiation, 1×10⁵ THP-1 cells were seeded in the upper chamber of a transwell insert with an 8 µm pore membrane (Corning) in RPMI-1640 medium supplemented with 10% FBS. LN229 cells were seeded at a density of 2×10⁵ cells per well in the lower chamber of a 24-well plate under different conditions (si-NC + DMSO (control), si-NC + CoCl₂, si-TP73-AS1 + CoCl₂). Co-culture was maintained for 48 hours to allow interaction between LN229 and THP-1 cells.

Statistical analyses were primarily conducted using GraphPad Prism 8.0.2 and R v3.5.0 (http://www.R-project.org) software. Group differences were assessed using Wilcoxon rank-sum test, Welch t-test, one-way ANOVA, or chi-square test. Kaplan-Meier survival curves were estimated using the log-rank test. Univariate and multivariate Cox regression analyses identified prognostic factors. Pearson correlation analysis determined associations between continuous variables. All experiments underwent at least three independent replications and were presented using mean ± standard deviation. A two-sided p-value < 0.05 was considered statistically significant unless otherwise stated. Benjamini and Hochberg (BH or FDR) correction was applied to reduce false positive rates.

To explore the prognostic value of hypoxia status in HGG (Grade 3/4), we conducted an unsupervised consensus cluster analysis of HGG samples from TCGA based on the gene expression profiling of a common HRMG ( Supplementary Figure S1 ). According to the cluster consensus, delta area, and average silhouette width, four robust hypoxia clusters were established for HGG samples ( Supplementary Figures S2A–D ). We observed that different hypoxia clusters yielded distinct survival prognosis ( Supplementary Figure S2E ). Tumors stratified into Cluster 1 and 2 were closely correlated with worse prognosis. Taken together, the status of hypoxia status may significantly impact clinical prognosis.

To further explore the underlying molecules involved in HGG hypoxia status, we conducted univariate Cox analyses on lncRNAs in HGG, WHO grade 3 (G3) glioma, and WHO grade 4 (G4) glioma cohorts from TCGA set. Meanwhile, we performed the Pearson correlation analysis between lncRNAs and HRMG, and conducted differential lncRNA expression analysis across hypoxia clusters. We overlapped the above 5 correlated lncRNA lists (Cox analysis: p-value < 0.05; Pearson correlation analysis: |Pearson r| >0.5, FDR < 0.001; differential expression analysis: |logFC| >1, FDR < 0.001) and obtained eight hypoxia-related lncRNAs ( Figures 1A, B ): AP000695.4, OSMR-AS1, AC078883.3, RP11-545E17.3, LINC01057, LINC01503, TP73-AS1 and LINC00672.

To evaluate the prognosis of glioma patients by hypoxia-related lncRNAs, we calculated the risk score based on coefficients from Cox analysis and lncRNA expression for each of the TCGA and CGGA glioma samples ( Figure 1C ). To analyze the prognosis of the hypoxia-related 8-lncRNA risk signature, we dichotomized TCGA and CGGA glioma samples based on the median risk score. The overall survival (OS) time differed significantly between the low- and high-risk groups ( Figures 1D, E ). Low risk score resulted in a significantly favorable prognosis than high risk score. Taken together, we developed a hypoxia-related lncRNA signature with robust prognostic value in HGG.

To analyze the relation between hypoxia-related lncRNA risk score and HGG clinical and molecular features, we drew heatmaps ( Figure 2A ; Supplementary Figure S3A ) to show the spectrums of clinical and molecular features based on risk score. There were robust differences in the clinical and molecular characteristics of the two risk groups ( Figure 2A , Supplementary Figure S3A , Supplementary Table S1 ). High-risk groups were closely correlated with older age at diagnosis, higher Karnofsky performance score (KPS), higher WHO grade, glioblastoma (GBM) phenotype, classical or mesenchymal subtype, Chr 7 gain/Chr 10 loss, non-codeleted 1p/19q, wild type IDH, unmethylated MGMT promoter, mutant TERT promoter, mutant PTEN, wild type ATRX, wild type TP53, and mutant EGFR. We also explored the association between risk score and various clinical, pathological, and molecular features. Higher risk score was observed in HGG with risky clinical factors and subtypes ( Figures 2B, C ; Supplementary Figure S3B ). HGG with wild-type IDH, unmethylated MGMT promoter, and non-codeleted 1p/19q presented higher risk score ( Figures 2D, E ; Supplementary Figure S3B ). In summary, these results indicated that the hypoxia-related lncRNA risk signature was positively associated with risky behaviors in HGG.

To precisely predict the prognosis of HGG, we examined the prognostic significance of the hypoxia-related 8-lncRNA risk signature across various stratified cohorts. Within the TCGA and CGGA dataset, our analysis demonstrated that the risk score had substantial predictive value in groups categorized by age, gender and KPS ( Figure 3A ; Supplementary Figure S4 ). Additionally, high-risk groups had a shorter survival time than low-risk groups, irrespective of whether patients underwent radiotherapy or chemotherapy ( Figure 3A ; Supplementary Figure S4 ). Then, we evaluated the prognostic significance of the risk score across various molecularly stratified cohorts. In cases with different statuses of MGMT promoter and TP53, those with high risk score exhibited shorter survival times compared to those with low risk score ( Figure 3B ). High risk score was associated with reduced OS time in cases without Chr 7 gain/Chr 10 loss and with mutation of IDH and TERT promoter and wild-type of PTEN, ATRX, EGFR and EZH2 in the TCGA dataset ( Figure 3B ), which was similar in the CGGA dataset ( Supplementary Figure S4 ). Furthermore, multivariate Cox analysis indicated that the risk signature was an independent prognostic factor in HGG ( Figure 4A ). Finally, the nomogram plots and calibration curves based on IDH status, WHO grade, and the lncRNA risk score suggested that the risk signature could increase prediction accuracy of prognosis in HGG ( Figure 4B ; Supplementary Figure S5 ). Overall, the risk signature proved to be a significant predictor of prognosis in HGG.

To explore the biological phenotype of HGG cases between the high- and low-risk groups, we first selected the coding genes that positively correlated with the risk score (Pearson r > 0.7, p-value < 0.05) in HGG from TCGA dataset. These coding genes was functionally annotated by the ClueGO plugin in Cytoscape ( Figure 5A ). High-risk group were mainly associated with increased expression of regulation of I-kappaB kinase/NF-kappaB signaling, glycosyl compound catabolic process, Various types of N-glycan biosynthesis, Salmonella and Human immunodeficiency virus 1 infection ( Figure 5B ). Specifically, we analyzed the distribution of gene score calculated from nonredundant gene programs, pathway signatures for drug targets, and canonical pathways across various risk groups. We found that Hypoxia/glycolosis gene set and immune-related gene sets, such as GP2_Immune-T cell/B cell, GP11_Immune-IFN, PD1_signaling, and CTLA4_pathway ( Figure 5C ; Supplementary Figure S6 ), were consistently more enriched in high-risk group compared to low-risk group. We then analyzed the anti-tumor immunity cycle using the Tumor Immunophenotype (TIP) algorithm (19). Effective anti-tumor immunity was reduced at steps 6 (tumor cell recognition by T cells) and 7 (killing of cancer cells) in high-risk group ( Figure 5C ; Supplementary Figure S6 ). Additionally, high-risk group also had higher immunosuppressive score as determined by a previously described method (24) ( Figure 6A lower panel and Supplementary Figure S7A lower panel). Collectively, these findings indicate that HGG with high risk score possess an immunosuppressive phenotype.

According to immunoediting theory, tumor immune microenvironment was remodeled by external and internal factors (25). We first explored potential external factors by investigating the microenvironmental composition. Our findings showed that high-risk group had elevated immune score and reduced tumor purity compared to low-risk group ( Figure 6A ; Supplementary Figure S7A ). Furthermore, macrophages, Tregs, Neutrophils, Nkdim, iDC, Tgd, Esoinophils, aDC and CD8⁺ T cells were more prevalent in high-risk group. Based on the CIBERSORT algorithm, we found that macrophages were the predominant cells in HGG ( Supplementary Figure S7B ). High-risk group had higher macrophage score generated by single-sample ssGSEA algorithm based on three established gene sets ( Figure 6B ). Additionally, higher proportion of M2 macrophages was observed in high-risk group ( Supplementary Figure S7B ). We then concentrated on the expression of immunomodulators, including both costimulatory and coinhibitory molecules, as a significant factor in intrinsic immune escape regulation. In high-risk group, these molecules were notably upregulated ( Figure 6C ; Supplementary Figure S7C ), indicating cytotoxic T-cell dysfunction. Subsequently, prediction analysis of glioma immunotherapy responsiveness based on the TIDE algorithm revealed that HGG with high risk score exhibited a poorer response to immunotherapy ( Figure 6D ). Therefore, enrichment of M2 macrophages and aberrant expression patterns of immunomodulators might serve as external and internal factors in maintaining immunosuppression in HGG with high risk score, resulting in immunotherapy resistance.

To further validate the link between hypoxia-related lncRNAs and tumor immunosuppression, TP73-AS1 with highest expressional level in TCGA and CGGA compared to other hypoxia-related lncRNAs was selected as a representative lncRNA for this purpose ( Supplementary Figure S8 ). In in-house glioma samples, we detected TP73-AS1, HIF1α, IBA-1, and TGF-β, which revealed that patients with elevated TP73-AS1 expression showed increased TAM (IBA-1⁺) infiltration and higher levels of HIF1a and TGF-β ( Figures 7A, B ), indicating a positive relationship between hypoxia-related TP73-AS1 and tumor immunosuppression. TP73-AS1 expressional levels were higher in HGG cell lines than in normal glial cells ( Figure 7C ). Under hypoxic conditions, HGG cells expressed more HIF1α and TP73-AS1, suggesting that hypoxia induces TP73-AS1 expression ( Figures 7D, E ). Moreover, TP73-AS1 knockdown ( Figure 7F ) inhibited HGG cell hypoxia status, proliferation, and colony formation ( Figures 7G–I ). To elucidate the role of hypoxia-related TP73-AS1 in hypoxia-induced TAM’s biological phenotype, we co-cultured HGG cells with suppressed TP73-AS1 with TAM under hypoxic conditions. The results showed that TP73-AS1 suppression in HGG cells significantly reduced hypoxia-induced TAM infiltration and M2 polarization ( Figures 7J, K ). These findings highlight the critical role of hypoxia-related lncRNA in regulating HGG’s malignant behaviors and TAM infiltration.