A computational analysis of programmed cell death-associated LncRNA signatures in glioma and drug prediction Authors: Xingyu Liu, Dongliang Luo, Peng Zhang, Di Tian, Wei Liu, Liyan Jia Published: 2025 ABSTRACT AbstractBackgroundDespite advancements in multimodal therapy, glioma remains a lethal brain tumor with limited prognostic biomarkers. Programmed cell death (PCD) pathways and long non-coding RNAs (lncRNAs) are emerging as therapeutic targets, but their combined prognostic potential in glioma remains underexplored.MethodsRNA sequencing data from The Cancer Genome Atlas (TCGA) and Chinese Glioma Genome Atlas (CGGA) cohorts were analyzed to identify PCD-related lncRNAs. Least Absolute Shrinkage and Selection Operator (LASSO) Cox regression and multivariate Cox proportional hazards regression analyses were applied to construct a prognostic signature, with external validation performed in an independent cohort. Tumor immune microenvironment characterization was subsequently conducted through the Tumor IMmune Estimation Resource (TIMER) database using the inverse fold product algorithm, which quantified six tumor-infiltrating immune cells (TIIC) subtypes across 10,897 pan-cancer samples. Immune checkpoint differential expression between risk strata was assessed via single-sample gene set enrichment analysis (ssGSEA). Drug sensitivity prediction was performed using the Cancer Genome Project (GCP) databases.ResultsA 4-lncRNA signature effectively stratified patients into high-risk and low-risk groups. The prognosis of the high-risk group is worse. In multivariate analysis, a risk score can independently predict the prognosis and has a robust prediction efficiency in the CGGA external verification set. The infiltration of immunosuppressive cells, such as M2 macrophages, increased the expression of immune checkpoints, such as PD-L1, increased, and the infiltration of CD8 + T cells decreased in high-risk patients. Computational screening identified elesclomol as a potential therapeutic agent showing subtype-specific efficacy.ConclusionsThis study highlights the prognostic significance of lncRNAs associated with PCD in glioma and provides computational evidence for their potential as therapeutic targets. While the results suggest novel avenues for treatment development, they must be interpreted cautiously due to the lack of experimental validation. Future studies should aim to validate these results through controlled clinical trials and explore underlying molecular mechanisms.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12883-025-04372-5.Keywords: Glioma, Long non-coding RNA, Programmed cell death, Prognosis model, Immune microenvironment, Targeted therapy, Necroptosis, Pyroptosis, Ferroptosis, Autophagy Background Glioma is the most common primary tumor in the central nervous system, and its clinical heterogeneity and diversity of molecular characteristics have long restricted the development of accurate diagnosis and treatment. Traditional histological types provide a basic framework for treatment [1]. However, it can’t explain the difference in the prognosis of tumors with the same grade. This dilemma has directly promoted the innovation of the diagnosis system. In 2021, the WHO CNS5 classification established an integrated diagnosis model with molecular markers such as IDH mutation and 1p/19q co-deletion as the core, which improved the accuracy of prognosis prediction by 37% [2, 3]. However, the therapeutic targets revealed by molecular typing still face multiple challenges: IDH wild-type GBM has limited sensitivity to the Stupp scheme [4]. The Stupp regimen remains the standard therapy, providing survival benefits for patients [5]. However, tumor heterogeneity frequently induces drug resistance, undermining treatment efficacy [6]. Furthermore, systemic drug delivery faces substantial challenges due to the blood-brain barrier’s restrictive nature [7]. Concurrently, the immunosuppressive tumor microenvironment compromises immunotherapy responses [8]. Moreover, targeted therapies are hindered by compensatory activation of alternative molecular pathways [9]. These treatment bottlenecks not only highlight the urgency of developing new combination therapies but also reveal a deeper need- only by discovering new biomarkers and intervention targets can the prognosis of patients be fundamentally improved [10]. Programmed cell death (PCD) is a genetically regulated process crucial for maintaining cellular homeostasis and organismal development [11]. Several distinct forms of PCD, including necroptosis, pyroptosis, ferroptosis, and autophagic apoptosis, have been identified, each with unique characteristics and roles in disease pathology [12–14]. Necroptosis serves as a protective inflammatory response to viral infections, allowing cells to undergo caspase-independent cell suicide [15]. Pyroptosis is an inflammatory cell death triggered by intracellular pathogen infections, releasing cytokines that attract immune cells to the infection site [16]. Ferroptosis, distinguished by its iron dependency, is implicated in various diseases, including cancer [17]. Autophagic apoptosis helps maintain cellular homeostasis and viability in eukaryotic cells [18, 19]. Emerging evidence indicates that PCD significantly impacts cancer treatment by affecting tumor growth, metastasis, and therapy response [20–23]. Targeting PD-1 and PD-L1 with immune checkpoint inhibitors has notably enhanced survival rates in advanced non-small-cell lung cancer (NSCLC) patients [24]. In gliomas, available studies have shown that apoptosis resistance can lead to treatment failure through Bcl-2 family protein overexpression and p53 mutations [25]. Autophagy also exhibits a double-edged role in temozolomide resistance, either removing damaged organelles to maintain survival [25] or inducing type II programmed death [26]. Ferroptosis can also be associated with glioma stemness maintenance through GPX4-regulated lipid peroxidation, and the inhibition of iron death enhances radiotherapy resistance [27]. In cancer, necrotizing apoptosis can lead to tumor progression and metastasis because the inflammatory environment caused by the death of necrotic cells may promote tumor growth and immune escape [28]. The inflammatory milieu created by pyroptotic cell death can synergize with immune checkpoint blockade, leading to improved therapeutic outcomes [29]. The interaction of these PCD mechanisms constitutes an important molecular basis for therapeutic resistance in glioma. However, the specific lncRNAs associated with programmed cell death (PCD) in gliomas remain incompletely characterized. This study integrates multidimensional data from TCGA and CGGA, aiming to systematically identify key lncRNAs in the PCD regulatory pathway in gliomas. By constructing a prognostic risk assessment model based on the expression profiles of these lncRNAs, we will quantitatively evaluate their predictive ability for patient survival outcomes. Meanwhile, we will further analyze the sensitivity of tumor microenvironment characteristics and chemotherapeutic drug response. Ultimately, we will provide potential novel biomarkers for precise prognostic stratification and optimization of targeted-immune combination therapy strategies for gliomas. Methods Data source The gene expression and clinical data used in this study were obtained from the TCGA and CGGA public databases, and all raw data were subjected to ethical review, informed patient consent, and de-identification, following the relevant data use protocols and citation specifications. Clinical samples from 159 TCGA and 657 CGGA glioma patients were included, with inclusion criteria involving the removal of duplicates and the exclusion of patients with a survival time of less than 10 days. The clinical data included all disease subtypes. We utilized the Ensembl Human Genome Browser GRCh38.p13 to differentiate between mRNAs and lncRNAs and to extract mRNA expression data related to PCD. Necroptosis-related mRNAs were identified through relevant literature [30], while pyroptosis-associated mRNAs were found via a literature search [31]. Ferroptosis-related mRNAs were sourced from the FerrDB database (http://www.zhounan.org/ferrdb/legacy/). Autophagy-related mRNAs were obtained from the Human Autophagy Database (HADb, http://www.autophagy.lu/). Clinical trial number: Not applicable. Identification of differential lncRNAs Associations between lncRNAs involved in necroptosis, pyroptosis, ferroptosis, and autophagy and their corresponding mRNAs were evaluated using Pearson correlation analysis. Significantly correlated lncRNAs were identified using a threshold of |R| >0.4 and P < 0.001(multiple testing correction was not applied). To avoid the zero value problem and ensure the symmetry of the data, before the difference analysis, we carried out log2(exprSet + 1) transformation on the expression data, which ensured that all the expression values were positive and made the data more suitable for subsequent statistical analysis. Differential expression of PCD-related mRNAs and lncRNAs in TCGA was analyzed using the limma package in R. Differentially expressed mRNA (DEmRNAs) and lncRNA (DElncRNAs) were identified by the screening conditions of absolute logarithmic multiple change (|logFC| >1) and false discovery rate (FDR ≤ 0.05). Enrichment analysis ClusterProfile, an R software package, is used to enrich and analyze Gene Ontology (GO) and Kyoto Genome Encyclopedia (KEGG) to study the potential interaction between these genes. The R package ggplot2 is used to visualize the enrichment results of the GO function and KEGG signaling pathway. We use the significance threshold P < 0.05 to define GO function and KEGG signaling pathway enriched by differentially expressed genes. Prognostic risk model for programmed cell death-associated lncRNA In the cross-database analysis, lncRNAs simultaneously associated with all four programmed cell death (PCD) pathways were first screened based on the TCGA dataset using the Venn online tool (https://bioinfogp.cnb.csic.es/tools/venny/); and further, in the CGGA dataset, significant covariations with PCD pathways were identified through the Pearson correlation analysis (|R| >0.4, P < 0.001) to identify lncRNAs significantly co-variant with PCD pathways, and ultimately retained the intersecting lncRNAs of TCGA and CGGA for subsequent analysis. Subsequently, for these candidate lncRNAs, univariate Cox proportional risk regression (R package survival) was used to screen molecules significantly (P < 0.05) associated with overall survival (OS), and the feature set was optimized by lasso regression (LASSO, R package glmnet) to reduce the overfitting, and ultimately, the individual risk score formula was constructed based on the multivariate Cox regression coefficients: Risk score = (lncRNA 1 expression × coefficient) + (lncRNA 2 expression × coefficient) +.. + (lncRNA n expression × coefficient). In the model validation stage, patients were divided into high-risk and low-risk groups based on the median risk score, and the survival differences between groups were assessed using the Kaplan-Meier survival curves (R package survival) and the log-rank test. To validate the independent prognostic value of the risk score, univariate and multivariate Cox regression (adjusting for variables such as age, sex, glioma grade, IDH status, and 1p19q co-deletion) was used to analyze its independence. Model predictive performance was quantified by time-dependent subject operating characteristic curves (R package timeROC), with accuracy assessed at different time points by the area under the curve (AUC). All analyses were done in the R language (version 4.2.1) environment, and data visualization was implemented by the ggplot2 package. GSEA functional enrichment analysis We performed pre-ranked GSEA (Gene Set Enrichment Analysis) to investigate the GO functions and KEGG pathways of PCD-associated lncRNAs in the TCGA database. In this analysis, we used KEGG gene sets from the MSigDB C2 collection, which contains curated pathway gene sets, and GO gene sets from the GO database. For the pre-ranked GSEA, we ranked genes based on log2 fold changes derived from differential expression analysis of PCD-associated lncRNAs. The analysis was conducted with the following parameters: a p-value cutoff of 0.05, minimum gene set size of 10, and maximum gene set size of 500. All analyses were performed using the clusterProfiler R package. Tumour immune microenvironment (TIME) analysis The abundance of TIICs in TCGA datasets was quantified using seven algorithms embedded in the TIMER2.0 web tool(http://timer.cistrome.org/)(32), including XCELL, TIMER2.0, QUANTISEQ, MCPCOUNTER, EPIC, CIBERSORT-ABS, and CIBERSORT. Single-sample Gene Set Enrichment Analysis (ssGSEA) was performed using the `GSVA` R package to validate immune cell infiltration levels across samples. The immune microenvironment scores (ESTIMATE, Immune, and Stromal scores) were calculated using the `ESTIMATE` R package to compare high- and low-risk groups. Differential expression of immune checkpoints and immunomodulators between differential groups was analyzed using the `limma` package, with results visualized via `ggplot2`. Drug analysis The pRRophetic R package was employed to predict potential targeted chemotherapeutics and calculate half-maximal inhibitory concentrations (IC50) for high- and low-risk groups, with higher IC50 values indicating reduced drug sensitivity. Group comparisons were performed using Wilcoxon tests with a stringent significance threshold (P ≤ 0.001). Subsets identification of glioma samples Based on the key lncRNAs in the risk group, we performed a clustering analysis on glioma samples using Euclidean distance and k-means clustering to determine their optimal number of clusters. Additionally, we conducted a survival analysis on glioma sample subtypes to evaluate their impact on the survival of glioma patients. Statistical treatment Descriptive statistical analysis was initially conducted on the GBM patient cohort from TCGA. Continuous variables were displayed as the average standard deviation, whereas categorical variables were described in terms of frequencies and proportions. The Wilcoxon rank-sum test was used to assess the relationship between risk scores and clinicopathological characteristics. To evaluate the overall survival (OS) differences among different subgroups, a Kaplan-Meier analysis accompanied by a log-rank test was performed. Statistical analyses were conducted using R version 4.2.2 with a significance level set at P < 0.05. Results Identification of lncRNAs and mRNAs associated with PCD PCD plays a critical role in tumor progression. We extracted 67 necroptosis-, 52 pyroptosis-, and 382 ferroptosis-related mRNAs from FerrDB, alongside 232 autophagic apoptosis-associated mRNAs from HADb. Using thresholds of |R| >0.4 and P < 0.001, we identified 1,110 necroptosis-, 1,822 pyroptosis-, 2,615 ferroptosis-, and 2,521 autophagic apoptosis-linked lncRNAs (Supplementary Fig. 1A-D). A total of 865 lncRNAs associated with programmed cell death (PCD) were identified from the CGGA database. Differential expression analysis (|logFC| ≤ 1, FDR ≤ 0.05) revealed the following regulated mRNAs: 26 up-regulated and 6 down-regulated in necroptosis (Supplementary Fig. 2 A), 22 up-regulated and 2 down-regulated in pyroptosis (Supplementary Fig. 2B), 67 up-regulated and 35 down-regulated in ferroptosis (Supplementary Fig. 2 C), and 53 up-regulated and 19 down-regulated in autophagy (Supplementary Fig. 2D). Further differential analysis of lncRNAs showed 533 up-regulated and 316 down-regulated in necroptosis (Supplementary Fig. 2E), 388 up-regulated and 133 down-regulated in pyroptosis (Supplementary Fig. 2 F), 588 up-regulated and 273 down-regulated in ferroptosis (Supplementary Fig. 2G), and 566 up-regulated and 267 down-regulated in autophagy (Supplementary Fig. 2H). Enrichment analyses on PCD-related differential mRNAs. A GO enrichment analyses on PCD-associated differential mRNAs, including BP, CC, and MF. B KEGG enrichment analyses on PCD-associated differential mRNAs. Development and validation of the lncRNA prognostic model. A Venn diagram illustrating PCD-related lncRNAs in TCGA and CGGA databases. B Heat map illustrating PCD-associated lncRNA expression in the TCGA database. C Forest plot for univariate Cox regression of lncRNA’s expression and glioma prognosis. D Lasso Cox regression analysis suggests 5 PCD-related differential lncRNAs as good candidates for constructing the prognostic signature. E, F Survival curves for high- and low-risk glioma patient groups from the TCGA and CGGA databases. G, I Forest plots for univariate Cox regression analysis of risk scores and glioma prognosis in TCGA and CGGA cohorts. H, J Forest plots for multivariate Cox regression analysis of risk scores and glioma prognosis in TCGA and CGGA cohort, respectively. K, M ROC model to validate the prognostic impact of value-at-risk in TCGA and CGGA glioma patients, respectively. L ROC model to validate the prognostic impact of risk value, age, and gender in TCGA patients with glioma. N ROC model to validate the prognostic impact of risk value, grade, gender, age, radio, TMZ, IDH, and 1p19q in TCGA patients with glioma. GO enrichment analysis indicated that differential mRNAs are involved in biological processes, including the regulation of peptidase activity, cellular responses to tumor necrosis factor, and cysteine-type endopeptidase activity. For cellular components (CC), they were enriched in membrane rafts, microdomains, and regions. Molecular functions (MF) included ubiquitin-protein ligase binding and cysteine-type endopeptidase activity in apoptosis (Fig. 1 A). KEGG analysis showed that PCD mRNAs were mainly involved in the NOD-like receptor signaling pathway, human papillomavirus infection, and lipid metabolism/atherosclerosis (Fig. 1B). Development and validation of the lncRNA prognostic model To identify valuable lncRNAs linked to PCD, the Venn tool revealed 356 lncRNAs involved in necroptosis, pyroptosis, ferroptosis, and autophagy (Supplementary Fig. 3 A). Using Pearson correlation (|R| >0.4, P < 0.001), 865 lncRNAs were identified from the CGGA dataset. Venn analysis identified 69 PCD-linked lncRNAs common to both TCGA and CGGA, shown in a heat map of normal and glioma tissues (Fig. 2A and B). A one-way Cox proportional risk regression (P < 0.05) based on the TCGA cohort screened 10 PCD-related lncRNAs that were significantly associated with overall survival (Fig. 2C). To optimize model efficacy, LASSO regression (Fig. 2D) and multivariate Cox analysis were ultimately employed to establish a 4-lncRNA prognostic signature comprising ZNF197-AS1, AC079210.1, RFPL3S, and PAXIP1-AS2. The risk score formula was defined as: Risk Score = (0.345 × PAXIP1-AS2) + (0.702 × AC079210.1) - (0.725 × ZNF197-AS1) - (0.981 × RFPL3S). The results of the survival analysis showed that the median overall survival (OS) of glioma patients in the high-risk group was significantly shorter than that in the low-risk group (P = 0.002, Fig. 2E), suggesting the clinical characteristics of poor prognosis. This discovery was further confirmed in the CGGA independent verification queue (P < 0.001, Fig. 2F). The data of the two cohorts jointly reveal the robust predictive value of the risk scoring model in the prognosis stratification of glioma. In the TCGA cohort, the univariate Cox proportional hazard model showed that the risk score was significantly correlated with the patient’s survival (HR = 1.347, 95%CI 1.094–1.658, P = 0.005, Fig. 2G), and it still maintained independent prognostic value after multivariate adjustment (HR = 1.311, 95%CI 1.058–1.625, P = 0.013, Fig. 2H). The analysis of the CGGA validation cohort further confirmed that the risk ratio of the univariate model was increased to 1.418(95%CI 1.307–1.539, P < 0.001, Fig. 2I), and the risk score still had significant predictive power (HR = 1.181, 95%CI 1.059–1.318, P = 0.001, Fig. 2J). The discriminant efficiency of the prognosis model was evaluated by time-dependent ROC curve analysis. In the TCGA cohort, the area under the prediction curve (AUC) of risk score for 3-year survival rate was 0.677 (Fig. 2K). When integrating clinical parameters, the risk score (AUC = 0.616) is significantly superior to the predictive efficacy of age (AUC = 0.571) and gender (AUC = 0.494) (Fig. 2L). In the CGGA validation queue, the 3-year AUC of the risk score is 0.676 (Fig. 2M). When combined with molecular indicators such as IDH (AUC = 0.723) and 1p19q (AUC = 0.619), the risk score still maintains its independent predictive value (AUC = 0.622) (Fig. 2N). These results show that the prognosis model has stable cross-cohort prediction ability, but the prediction efficiency of age and gender is relatively limited. LncRNAs GSEA functional enrichment analysis We performed Gene Set Enrichment Analysis (GSEA) to explore the GO and KEGG enrichment of PCD-related lncRNAs in the TCGA database. GO terms in the high-risk group were significantly associated with acute inflammatory response, membrane invasion, regulation of phagocytosis, nitrogenophil granules, and phagocytic vesicles (Fig. 3 A). KEGG pathways in the high-risk group were mainly enriched in apoptosis, cell adhesion, complement and coagulation cascades, leukocyte transendothelial migration, and lysosome pathways (Fig.3B). GSEA functional enrichment analyses of LncRNAs. A GO and (B) KEGG enrichment analyses in the high-risk group. PCD-associated lncRNAs with tumor immune infiltration and microenvironment in TCGA database. (A) Air bubble map multiple software analyses on PCD-associated lncRNAs for immune infiltration in TCGA database samples. (B-D) Boxplots illustrate ssGSEA analysis of PCD-associated lncRNA expression across immune functions, immune cells, and checkpoints in high- and low-risk groups. (E-G) Boxplots illustrating ESTIMATE, immune, and stromal scores for high and low-risk groups within the tumor microenvironment (P < 0.001). (H, I) The boxplot illustrated the sensitivity to Docetaxel and CGP.60,474 in high and low risk groups within the TCGA database (P < 0.001). (J-K) Chemical structure formula of Docetaxel and CGP.60,474 PCD-associated lncRNA and tumour immune infiltration microenvironment and drug analysis TIICs in the TCGA dataset were assessed using TIMER2.0, XCELL, QUANTISEQ, MCPCOUNTER, EPIC, CIBERSORT-ABS, and CIBERSORT. PCD-related lncRNAs demonstrated significant correlations with bone marrow dendritic cells, cancer-associated fibroblasts, CD4 + T cells, and M1/M2 macrophages (Fig. 4A). SsGSEA validated and assessed immune cell infiltration levels associated with PCD lncRNAs. The high-risk group demonstrated increased immune functions, such as APC co-inhibition and co-stimulation, CCR, checkpoint activity, cytolytic activity, HLA expression, inflammation promotion, parainflammation, T cell co-inhibition and co-stimulation, and Type II IFN response (Fig. 4B), compared to the low-risk group. The high-risk group exhibited increased levels of activated dendritic cells (aDCs), B cells, dendritic cells (DCs), immature DCs (iDCs), macrophages, mast cells, neutrophils, T helper cells, Tfh cells, Th1/Th2 cells, TILs, and Tregs (Fig. 4C). Significant differences in immune checkpoints were identified between the high- and low-risk groups (Fig. 4D). The high-risk group demonstrated increased ESTIMATE (P < 0.001, Fig. 4E), immune (P < 0.001, Fig. 4F), and stromal scores (P < 0.001, Fig. 4G). Using the “pRRophetic” R package, 35 potential chemotherapeutic agents were identified, including CGP.60,474 (P < 0.001, Fig. 4H) and Elesclomol (P < 0.001, Fig. 4I), both with low IC50 values. The molecular structures of CGP.60,474 (Fig. 4J) and Elesclomol (Fig. 4K) were retrieved from PubChem and validated against immune-infiltration data in the CGGA dataset. PCD-associated lncRNAs with tumor immune infiltration and microenvironment in TCGA database. A Air bubble map multiple software analyses on PCD-associated lncRNAs for immune infiltration in TCGA database samples. B-D Boxplots illustrate ssGSEA analysis of PCD-associated lncRNA expression across immune functions, immune cells, and checkpoints in high- and low-risk groups. E-G Boxplots illustrating ESTIMATE, immune, and stromal scores for high and low-risk groups within the tumor microenvironment (P<0.001). (H, I) The boxplot illustrated the sensitivity to Docetaxel and CGP.60474 in high and low risk groups within the TCGA database (P<0.001). J-K Chemical structure formula of Docetaxel and CGP.60474. In the CGGA dataset, the high-risk group showed increased immune functions, such as APC co-inhibition and co-stimulation, CCR, checkpoint activity, cytolytic activity, HLA expression, inflammation promotion, parainflammation, T cell co-inhibition and co-stimulation, and Type II IFN response (Fig. 5A). Immune infiltration analysis revealed elevated levels of aDCs, B cells, DCs, iDCs, macrophages, mast cells, neutrophils, T helper cells, Tfh cells, Th1/Th2 cells, TILs, and Tregs in the high-risk group (Fig. 5B). Differential expression of immune markers was observed in both risk groups (Fig. 5C), with comparable differences in ESTIMATE (P < 0.001, Fig. 5D), immune (P < 0.001, Fig. 5E), and stromal scores (P < 0.001, Fig. 55). Bortezomib (P < 0.001, Fig. 5G) and Docetaxel (P < 0.001, Fig. 5H) were identified as chemotherapeutic agents with the lowest predicted IC50 in the CGGA dataset, with their chemical structures shown (Fig. 5I-J) from PubChem. PCD-associated lncRNAs with tumor immune infiltration and microenvironment in CGGA database. A-C Boxplots of ssGSEA analyses on PCD-associated lncRNAs expression in immune function, immune cells, and immune checkpoints for high- and low-risk groups in the CGGA database. D-F Boxplots depicting ESTIMATE, immune, and stromal scores for high and low-risk groups in the tumor microenvironment (P<0.001). G-H Boxplot illustrating Bortezomib and Docetaxel sensitivity in high and low risk groups from the CGGA database (P<0.001). I-J Chemical structure formula of Bortezomib and Docetaxel. Subsets identification of glioma samples The comprehensive correlation coefficient in the TCGA database indicated that the optimal number of clusters (k) is 3. Glioma samples were categorized into three subclasses: Cluster 1, Cluster 2, and Cluster 3 (Fig. 6A). Similarly, CGGA analysis confirmed k = 3, resulting in the same subclass divisions (Fig. 6B). The consensus matrix for k = 3 showed a clear boundary, indicating stable clustering. T-Sensitivity PCA of TCGA and CGGA subtypes revealed a three-dimensional distribution, further validating the clustering (Fig. 6C, D). Survival analysis of glioma subgroups in TCGA showed that Clusters 2 and 3 had significantly shorter survival than Cluster 1 (Fig. 6E). A similar trend was observed in CGGA, reinforcing these findings (Fig. 6F). Subsets of glioma samples.A-B) Consensus matrix method clustering using 4 PCD-associated lncRNAs in glioma subtypes from the TCGA and CGGA databases, respectively. Clustering was performed using the k-means method with Euclidean distance. C-D PCA analyses of the TCGA and CGGA datasets illustrate the distribution of three glioma subclasses: TCGA: PC1: 0.321, PC2: 0.263, indicating that the first two principal components explain 32.1% and 26.3% of the variance in the TCGA dataset, respectively. CGGA: PC1: 0.746, PC2: 0.158, indicating that the first two principal components explain 74.6% and 15.8% of the variance in the CGGA dataset, respectively. E-F Survival curves for glioma patient subtypes C1, C2, and C3 in the TCGA and CGGA databases. Association between glioma sample subsets and tumor immune infiltration microenvironment To examine the infiltration of immune cells in glioma subgroups, we similarly calculated the infiltration by TIMER online database multiple different software. The C2 and C3 groups showed higher cancer-associated fibroblast, macrophage, and T-cell CD4 + expression compared with C1 (Fig. 7 A). In particular, the C1, C2 and C3 subgroups in the TCGA database glioma subtypes showed significant differences in the expression of immune checkpoints, especially CD44, and LAIR1 was significantly upregulated in the C2 and C3 subgroups (Fig. 7B). Similar findings were obtained in the CGGA database glioma subtypes (Fig. 7C). We scored the immune microenvironment of a subset of glioma samples and found higher expression scores for ESTIMATE (Fig.7D) immune (Fig. 7E) and stromal (Fig. 7F) in the C2 and C3 groups. Of particular note, targeted chemotherapeutic agents continue to play an important role in the treatment of glioma patients. Potential targeted chemotherapeutic agents for glioma subgroups C1, C2, and C3 were analysed, yielding a total of 57 results. Among them, the IC50 of CGP.60,474 (Fig. 7G) was lower in the C2 and C3 groups, and the IC50 of Elesclomol was slightly higher (Fig. 7H). Glioma sample subsets with tumor immune infiltration microenvironment. A Air bubble map multiple software analyses on glioma sample subsets for immune infiltration in TCGA database samples. B-C Boxplots illustrate the expression of immune checkpoints across glioma sample subtypes in the TCGA and CGGA databases. D-F ESTIMATE, Immune, and stromal scoring of glioma sample subsets in the tumor microenvironment by boxplots, respectively. G-H CGP.60474 and Elesclomol sensitivity in glioma sample subtypes by boxplots, respectively. Discussion In this study, the prognostic value of PCD-related lncRNAs was systematically analysed based on bioinformatics methods. We systematically constructed a prognostic model for glioma based on lncRNAs and comprehensively analysed the impact of risk scores on glioma at the levels of glioma genome, functional pathways, and immune microenvironment. In addition, we found that patients with different risk scores had very different sensitivities to different chemotherapeutic agents. And subtype analysis revealed that the prognosis, immune infiltration, and prediction of response to chemotherapeutic agents varied among different subtypes, which provides ideas for the treatment of glioma patients with cell death-associated pathways. In this study, data from multiple sources such as FerrDB, HADb, CGGA and TCGA were analysed to identify lncRNAs associated with key PCD pathways (necroptosis, pyroptosis, ferroptosis, and autophagy). through correlation analysis, we selected candidate lncRNAs that are closely related to these pathways. further studies of their expression in glioma samples found that there were significant differences in their expression levels. Notably, lncRNAs associated with ferroptosis and autophagy were significantly upregulated, suggesting that these pathways play a key role in glioma progression. These lncRNAs regulate the expression of their associated mRNAs either through transcriptional regulation or through mechanisms such as post-transcriptional modifications. To explore whether they affect PCD-related pathways by influencing mRNA expression and thus, we made PCD-differential mRNA enrichment analysis. The enrichment results were largely consistent with our speculation, mainly focusing on the PCD-related pathway. To assess the clinical utility of these lncRNAs, we constructed a prognostic model based on their expression profiles. Through statistical analyses such as LASSO and Cox regression, we identified four key genes lncRNA-ZNF197-AS1, AC079210.1, RFPL3S, and PAXIP1-AS2-that were closely associated with patient survival. Bioinformatic analysis suggests that ZNF197-AS1 may participate in competitive endogenous RNA (ceRNA) regulatory networks. This lncRNA potentially modulates programmed cell death (PCD)-associated pathways through its ceRNA activity. Mechanistic studies indicate that ZNF197-AS1 could influence tumor cell malignancy via regulation of specific miRNA-mRNA axes [33]. In our study, the clinical correlation analysis showed that the increased expression of ZNF197-AS1 was independently related to the improvement of survival outcome of glioma patients. Currently, there is no clear report on the cancer-promoting mechanism of AC079210.1. In this study, we found that AC079210.1 is a risk factor for glioma prognosis, and patients tended to have a better prognosis when its expression level was low. RFPL3S exhibits tumor-suppressive functions across multiple cancer types. In gliomas, its elevated expression independently correlates with favorable prognosis, serving as a protective prognostic factor for patient survival. Functional characterization in non-small cell lung cancer demonstrates RFPL3S-mediated enhancement of telomerase activity through transcriptional upregulation of human telomerase reverse transcriptase (hTERT) [34]. Notably, clinical data from testicular germ cell tumors (TGCT) corroborate its tumor-suppressive role, where reduced RFPL3S expression significantly predicts poorer clinical outcomes [35]. This is consistent with our study. This study also identified PAXIP1-AS2 as a prognostic risk factor for gliomas. PAXIP1-AS2 has been identified as a potential biomarker for various cancers, including glioblastoma and uveal melanoma. In glioblastoma, it is part of a prognostic signature associated with patient prognosis and the tumour immune microenvironment [36]. In uveal melanoma, PAXIP1-AS2 promotes cell migration and invasion through upregulation of JAK2, suggesting a role in metastasis [37]. These lncRNAs influence the prognostic progression of gliomas either through transcriptional regulation or by regulating the expression of their associated mRNAs through mechanisms such as post-transcriptional modifications. Kaplan-Meier analysis shows that the survival rate of high-risk patients is obviously lower than that of low-risk group, and the results are verified by independent data sets from CGGA, which proves the wide applicability of the results. In addition, the time-dependent ROC analysis showed that the risk score combined with glioma stage, age, IDH, 1p19q and other clinical features also had good prediction results. Comparing with some of the current models for glioma [38], our predictive performance is relatively excellent, showing the good predictive ability of the model. In addition to the prognostic value, we explored the relationship between PCD-associated lncRNAs and the glioma immune microenvironment. The elevated immune function-related indicators exhibited by patients in the high-risk group may reflect a dynamic imbalance between immune activation and suppression in the tumour microenvironment. Specifically, the upregulation of APC co-inhibition and T cell co-inhibition may be associated with low expression of ZNF197-AS1.ZNF197-AS1 reduces infiltration of effector immune cells through the miR-425/GABARAPL1 axis. It then induces compensatory high expression of suppressive immune checkpoints such as PD-L1 and LAIR-1 in dendritic cells and T cells [33]. Mechanistically, AC079210.1 upregulation drives aberrant TGF-β signaling activation in the tumor microenvironment. This oncogenic pathway induces dual regulatory effects: [1] enhanced antigen-presenting cell (APC) co-stimulation through increased expression of CD80/CD86 molecules on cancer-associated fibroblasts (CAFs), and [2] paradoxical potentiation of T-cell co-stimulation. The sustained TGF-β activation simultaneously promotes pro-inflammatory cytokine secretion, particularly IL-6, while establishing an immunosuppressive niche. This dichotomy results in concurrent inflammatory microenvironment activation and immune escape mechanisms [39–41]. The high-risk cohort demonstrates concurrent upregulation of cytolytic effectors, HLA class I/II complexes, and type II interferon signaling pathways, indicative of tumor microenvironment-specific immune activation. This immunophenotypic pattern suggests compensatory anti-tumor immunity despite progressive disease status. However, this process may be counteracted by PAXIP1-AS2-mediated immunosuppressive mechanisms. It may do so by inducing macrophage M2 polarisation and secretion of IL-10 and TGF-β. In this way PAXIP1-AS2 not only promotes Treg cell expansion, but also leads to effector T cell depletion through upregulation of the PD-1/PD-L1 axis, thus explaining why high immunoreactivity fails to translate into an improvement in clinical prognosis [42, 43]. In terms of immune cell infiltration characteristics, the significant increase of aDCs, iDCs and DCs in the high-risk group may reflect compensatory activation of the antigen-presenting system, but TGF-β signaling induced by the high expression of AC079210.1 may impede the maturation of DCs, leading to the accumulation of immature DCs and weakening their antigen-presenting function [44]. The synchronous increase in macrophage M1/M2 dual subtypes suggests a sustained tug-of-war between pro-inflammatory and anti-inflammatory signaling in the microenvironment, in which PAXIP1-AS2 promotes M2 polarisation by regulating PD-L1 expression [45–47], whereas infiltration of neutrophils and Th1 cells may be driven by deregulation of Th17 inhibition triggered by low expression of RFPL3S, and this dysregulation of the Th1/Th17 balance may exacerbate tissue inflammation but not effectively kill tumours [48]. Treg and Th2 cell dominant infiltration further highlights the formation of an immunosuppressive microenvironment, and the synergistic effect of PAXIP1-AS2 and ZNF197-AS1 may maintain regulatory T cell dominance through the IL-10-TGF-β axis and inhibit CD8 + T cell function [49, 50]. Elevated ESTIMATE, immune and stromal scores in the high-risk group reveal the high complexity of the tumour microenvironment. The elevated immune scores mainly resulted from the extensive infiltration of B cells, T cell subsets, and myeloid cells, which were numerically superior despite the fact that a significant proportion of these cells were functionally inhibitory subsets (e.g., M2 macrophages, Treg) [51]. Elevated stromal scores are closely associated with AC079210.1-driven activation of CAFs, and TGF-β signaling forms a physical barrier and limits the efficiency of immune cell infiltration by inducing collagen deposition and extracellular matrix (ECM) remodeling [52]. This enhanced immune-stromal interaction may give rise to a “rejection-type” immune microenvironment, in which a large number of immune cells are blocked at the edge of the tumour without effective infiltration of the core area. The intergroup differences in immune checkpoints may be driven by a multifaceted regulatory network: low expression of ZNF197-AS1 deregulates CTLA-4 transcriptional repression through GABARAPL1 down-regulation, while high expression of PAXIP1-AS2 directly up-regulates PD-L1 expression through the NF-kB signaling pathway [51, 53]. In addition, IL-10 secreted by M2 macrophages induces TIM-3 expression through the STAT3 pathway, whereas low expression of RFPL3S may impair its negative regulatory effect on LAG-3 [54, 55]. This mechanism of synergistic upregulation of multiple checkpoint molecules allows tumour cells in the high-risk group to establish multiple immune escape defences, ultimately leading to immunotherapy resistance and poor prognosis. These findings suggest that therapeutic strategies for high-risk patients need to balance immune checkpoint blockade with stromal microenvironment modulation, e.g., PD-1 inhibitors in combination with TGF-β receptor antagonists may reverse the immunosuppressive state [56]. In addition, enrichment analysis showed that lncRNAs from high-risk patients were significantly involved in immune-related processes, which further confirmed the idea that these molecules play a central role in the immune microenvironment, and the specific regulatory mechanisms need to be confirmed by subsequent experimental studies. In this study, we systematically revealed the significant heterogeneity of glioma molecular subtypes (C1, C2 and C3) in terms of immune microenvironmental characteristics and therapeutic susceptibility by integrating the TIMER database with the multi-omics analysis of TCGA and CGGA cohorts. It was found that compared to the C1 subtype, the C2 and C3 subtypes exhibited a more active pattern of immune cell infiltration, including significant enrichment of cancer-associated fibroblasts (CAFs), macrophages, and CD4 + T cells. This finding is consistent with previous studies on the mechanism of formation of the immunosuppressive microenvironment in gliomas - CAFs can remodel the tumour mesenchyme through secretion of extracellular matrix and promote immune escape [57];In contrast, the infiltration of M2-type macrophages may further exacerbate the local immunosuppressive state by releasing inhibitory cytokines such as IL-10 and TGF-β [57]. Notably, the high infiltration of CD4 + T cells may be associated with the expansion of regulatory T cells (Tregs), suggesting that C2/C3 subtypes may have stronger immune tolerance properties, which may partly explain the molecular basis of their poorer clinical prognosis [58]. At the level of immune checkpoint molecules, the significant upregulation of CD44 and LAIR1 in the C2 and C3 subtypes is clinically important.CD44, as a hyaluronan receptor, has been shown to promote glioma stem cell self-renewal through activation of the PI3K/AKT pathway [59], whereas LAIR1 inhibits T-cell function by mediating collagen signaling [60]. Co-expression of these two molecules may together constitute a unique immune escape mechanism in the C2/C3 subtype. This not only helps to explain the poor prognosis, but also suggests that immunotherapy against these targets may be a potential intervention strategy for this subgroup of patients. In addition, the ESTIMATE scoring system showed that the C2/C3 group had a higher immune/stroma score, suggesting that stromal remodeling and immune cell interactions were more active in their tumour microenvironment [61]. However, the C2/C3 isoforms in the present study showed significantly elevated sensitivity to the cell cycle inhibitor CGP-60,474, which may be correlated with their high proliferative activity and cell cycle pathway-dependent [62]. In contrast, the resistance of the C2/C3 isoforms to the ROS-inducing agent Elesclomol may stem from the oxidative stress adaptive mechanisms of their immunosuppressive TMEs, for instance CAFs or M2 macrophages by scavenging ROS [63] or secretion of protective factors (e.g. IL-10) to enhance tumour survival [64]. However, these findings need to merit further experiments to validate them! Therefore, C1 subtype may benefit from immunotherapy for sparse immune infiltration. C2/C3 subtype needs drugs for proliferation (CGP-60474) and TME-mediated resistance (such as LAIR1/CD44 blockade). These insights emphasize the necessity of combined therapy for molecular subtypes and immune environment. In this study, a relatively robust prediction model was constructed by integrating the resources of a multi-centre database. However, there are still the following limitations that need to be carefully considered. Firstly, the design of retrospective studies based on public databases inherently has selection bias and potential confounders that are difficult to fully control. Particularly in the context of insufficient standardisation of clinical treatment protocols, there may be systematic interference with the quantitative assessment of tumour immune microenvironment characteristics. Secondly, it is worth pointing out in particular that the model’s prediction of the immune microenvironment and chemotherapeutic drug sensitivity mainly relies on bioinformatics inference, and its specific molecular mechanism of regulating the immune microenvironment has not yet been mechanistically corroborated by functional experiments. Therefore, the future research needs to be deepened in the following aspects: on the one hand, it is suggested to carry out prospective cohort study and establish standardized collection norms of treatment parameters to eliminate the interference of clinical intervention heterogeneity on biomarker analysis; On the other hand, it is necessary to analyze the molecular mechanism of the key lncRNA predicted in this study by combining the cutting-edge technologies such as spatial transcriptomics and single cell sequencing with the experimental system of functional genomics, especially to verify its potential value as a predictive marker of immunotherapy response at the level of clinical transformation. Conclusions In summary, our study highlights the prognostic significance of programmed cell death-associated lncRNAs in glioma and suggests their possible candidacy as therapeutic targets worthy of experimental validation. The integration of lncRNA-based risk models with chemotherapeutic agents like Elesclomol represents a promising approach to improving glioma treatment outcomes. Although further validation and mechanistic studies are required, these findings provide a solid foundation for the development of more effective and personalized therapies for glioma patients. Supplementary Information Supplementary Material 1. Acknowledgements The author thanks the TCGA and CGGA databases for their contributions. Abbreviations Programmed cell death Non-small-cell lung cancer Long non-coding RNAs The Cancer Genome Atlas Chinese Glioma Genome Atlas Tumor IMmune Estimation Resource Human Autophagy Database Differentially expressed mRNAs Differentially expressed lncRNAs Gene Ontology Kyoto Encyclopedia of Genes and Genomes Cellular component Molecular function Biological process Least Absolute Shrinkage and Selection Operator Receiver Operating Characteristic Area Under Curve Tumour immune microenvironment Gene Set Enrichment Analysis Tumor-infiltrating immune cells Single-sample Gene Set Enrichment Analysis Cancer Genome Project Activated Dendritic Cells Dendritic Cells Immature DCs Inhibitory concentration Authors’ contributions LDL, LXY, designed the study. LDL, LXY ZP and TD conducted the bioinformatics analysis and data interpretation. LDL and LXY co-authored the manuscript. LW and JLY revised the manuscript and gave final approval of the version to be published. All authors read and approved the final manuscript. Funding This work was supported by Shandong Medical and Health Science and Technology Development Project (202204040712) and Shandong Second Medical University Graduate Research Innovation Fund Project (2022-25). Data availability The datasets utilized and examined in this study are accessible via the TCGA (https://xenabrowser.net/datapages/) and CGGA (http://www.cgga.org.cn/index.jsp) databases. Declarations Ethics approval and consent to participate Not applicable. This study follows the Helsinki Declaration and the usage specification of public databases, and all data are from anonymous data sets of TCGA and CGGA. TCGA and CGGA belong to public databases. The patients involved in the database have obtained ethical approval. Users can download relevant data for research and publish relevant articles for free. Our research is based on open source data, so there are no ethical issues and other conflicts of interest. Such secondary data analysis does not require additional ethical approval. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Xingyu Liu and Dongliang Luo have contributed equally to this work and share first authorship. Contributor Information Wei Liu, Email: liuwdandy@sdsmu.edu.cn. Liyan Jia, Email: honeyjly@126.com. References Associated Data Supplementary Materials Supplementary Material 1. Data Availability Statement The datasets utilized and examined in this study are accessible via the TCGA (https://xenabrowser.net/datapages/) and CGGA (http://www.cgga.org.cn/index.jsp) databases.