Telomerase-related gene EHHADH drives lung cancer progression and shapes the immunosuppressive tumor microenvironment Authors: Wei Wei, Yuheng Wang, Diliyaer Abudukeremu, Jinhua Luo, Mingjun Du Published: 2026 ABSTRACT Highlights•Identified 8 prognostic telomerase-related genes in lung adenocarcinoma and built a robust risk model.•The risk model links to immune microenvironment; low-risk patients show immune-hot phenotype.•EHHADH acts as an oncogene in lung adenocarcinoma; its high expression predicts poor prognosis.•EHHADH is enriched in epithelial cells and shapes an immune-desert tumor microenvironment.Keywords: Lung adenocarcinoma, Telomerase, Tumor microenvironment, Prognostic model, EHHADH Introduction Lung cancer remains one of the most prevalent and lethal malignancies worldwide, imposing an enormous burden on human health and socioeconomic development. According to GLOBOCAN 2022, approximately 2.48 million new lung cancer cases and 1.82 million lung cancer-related deaths are recorded annually on a global scale, underscoring the staggering magnitude of this disease burden [1]. Despite remarkable advances in diagnostic technologies and therapeutic modalities—including surgery, radiotherapy, chemotherapy, targeted therapy, and the emerging field of immunotherapy [2]—the overall five-year survival rate for lung cancer patients remains unsatisfactory. In particular, the prognosis for patients with advanced-stage disease continues to be dismal, with the five-year survival rate for stage IV patients estimated at only approximately 8–11% [3]. This grim reality stems not only from the pronounced heterogeneity and aggressive biological behavior of lung cancer, but also from the substantial inter-individual variability in treatment response [4]. Consequently, elucidating the molecular mechanisms underlying lung cancer initiation and progression, identifying biomarkers with prognostic predictive value and therapeutic guidance potential, and developing precise prognostic assessment models on this basis are of paramount clinical significance and scientific importance for realizing individualized diagnosis and treatment, as well as improving the quality of life and long-term outcomes of lung cancer patients [5]. Telomerase is a specialized ribonucleoprotein enzyme responsible for maintaining the length and structural integrity of telomeres at chromosomal ends. Under physiological conditions, telomerase activity is typically suppressed or maintained at minimal levels in normal somatic cells. In stark contrast, aberrant upregulation of telomerase activity is detected in the vast majority of malignant tumor cells, representing a pivotal mechanism by which cancer cells acquire unlimited proliferative capacity and evade senescence and apoptosis [6]. Activation of the telomerase catalytic subunit TERT is not only implicated in tumor initiation and progression, but also exhibits a mutation frequency of approximately 11% in non-small cell lung cancer (NSCLC), with its mutational status closely associated with clinical features including smoking history, sex, and distant metastasis [7]. Furthermore, TERC, the RNA component of the telomerase complex, has been demonstrated to participate in NSCLC progression by facilitating the assembly and nuclear translocation of the telomerase holoenzyme [8]. In recent years, with the widespread application of genomic sequencing technologies and bioinformatic analytical tools, the prognostic value of telomere-related genes (TRGs) has attracted considerable attention. Accumulating evidence indicates that TRG-based prognostic signatures can effectively predict survival outcomes in lung adenocarcinoma patients and are closely correlated with tumor immune infiltration status and treatment sensitivity [9,10]. Nevertheless, the precise mutational landscape of these genes in lung cancer, their systematic association with patient clinical prognosis, and the mechanisms by which they cooperate with other molecular events to influence disease course remain major focuses and challenges in the field [10,11]. In addition, the tumor microenvironment (TME) serves as the indispensable "ecosystem" upon which tumor cells depend for their survival and progression, playing a critical role in tumor development, metastasis, and therapeutic resistance. The TME is composed of a diverse array of cellular and non-cellular constituents—including tumor cells, immune cells, mesenchymal cells, vascular endothelial cells, and extracellular matrix components—that interact with one another through intricate signaling networks to collectively orchestrate tumor biological behavior [12]. In particular, the composition and functional state of the tumor immune microenvironment (TIME) exert a decisive influence on the efficacy of immunotherapy [13]. Studies have demonstrated that "immune-hot" tumors enriched with effector T cells generally respond favorably to immune checkpoint inhibitors, whereas "immune-cold" or "immune-excluded" microenvironments frequently display treatment resistance, attributable to multifaceted mechanisms including metabolic reprogramming, immune cell dysfunction, and impaired T cell infiltration [14]. Therefore, delineating the intrinsic relationships between prognosis-associated genes and the characteristics of the tumor immune microenvironment in lung cancer not only deepens our mechanistic understanding of disease progression, but also provides potential targets and conceptual frameworks for identifying patients most likely to benefit from immunotherapy and for developing novel combinatorial therapeutic strategies [15]. In light of the foregoing, the present study aimed to integrate large-scale multi-omics data with advanced bioinformatic analytical approaches to comprehensively investigate the mutational landscape and prognostic significance of telomere-related genes in lung cancer. On this foundation, we sought to construct a robust risk model capable of accurately predicting patient prognosis, systematically characterize its associations with the tumor immune microenvironment, and functionally validate and mechanistically explore key prognostic genes through in vitro cellular functional assays and single-cell RNA sequencing. The findings of this study are expected to provide novel molecular tools for precise risk stratification in lung cancer, offer evidence-based decision support for clinicians in formulating individualized treatment regimens, and uncover new targets and research directions for the diagnosis, treatment, and prognostic improvement of lung cancer, ultimately contributing to enhanced long-term survival and quality of life for lung cancer patients. Method Data acquisition and preprocessing Gene expression data and corresponding clinical information for lung adenocarcinoma (LUAD) patients were downloaded from The Cancer Genome Atlas (TCGA) database and used as the model training cohort. Six independent GEO public datasets (GSE13213 [16], GSE26939 [17], GSE29016 [18], GSE30219 [19], GSE31210 [20], and GSE42127 [21]) were collected as external validation cohorts. Single-cell RNA sequencing (scRNA-seq) data from the HRA001130 dataset were obtained for tumor microenvironment (TME) characterization. Prior to analysis, batch effects across all datasets were corrected, and data quality was verified by principal component analysis (PCA). Mutational landscape analysis of telomerase-related genes The mutational characteristics of telomerase-related genes were systematically analyzed in the TCGA database, encompassing variant type distribution, single nucleotide variant (SNV) base substitution spectra, and gene mutation frequencies. Waterfall plots were employed to visualize the mutation landscape across individual samples, and frequently mutated genes were prioritized for downstream analysis. Identification of prognostic-related genes Univariate Cox proportional hazards regression analysis was independently performed across seven public datasets to identify telomerase-related genes significantly associated with overall survival (OS). Candidate genes demonstrating consistent prognostic relevance across multiple datasets were retained, ultimately yielding eight key prognostic genes: INMT, ALDH3A2, OGDHL, TDO2, EFNA3, STAT1, EHHADH, and AANAT. Differential expression patterns of these eight genes between normal and tumor tissues were visualized by heatmap analysis. Construction and validation of the prognostic risk model The eight prognostic genes were integrated into multiple machine learning algorithms for model optimization, with the mean C-index across six GEO validation cohorts serving as the primary selection criterion. A risk score was computed for each patient based on the expression levels of model genes weighted by their corresponding regression coefficients. Patients were dichotomized into high-risk and low-risk groups using the optimal cutoff value as the threshold. Kaplan–Meier survival analysis with log-rank testing was used to compare OS between the two groups. Time-dependent receiver operating characteristic (ROC) curve analysis was performed to assess predictive accuracy at 1-, 3-, and 5-year time points. Furthermore, PCA based on model gene expression was applied to validate the patient stratification capability of the model, and systematic comparisons with previously published models were conducted to evaluate its performance superiority. Tumor microenvironment analysis The Immunophenoscore (IPS) for TCGA-LUAD samples was calculated using The Cancer Immunome Atlas (TCIA), and IPS differences between high- and low-risk groups were compared across four immune checkpoint profiles (CTLA4-pos/PD1-pos, CTLA4-neg/PD1-pos, CTLA4-pos/PD1-neg, and CTLA4-neg/PD1-neg). The ESTIMATE R package was applied to calculate StromalScore, ImmuneScore, ESTIMATEScore, and TumorPurity for each sample, and Spearman correlation analysis was used to evaluate the associations between these scores and the risk score. Immune cell infiltration levels were systematically assessed in the TIMER 2.0 database by integrating seven quantification algorithms: QUANTISEQ, CIBERSORT, CIBERSORT-ABS, XCELL, MCPCOUNTER, EPIC, and TIMER. Additionally, differential expression of immune-related genes, including co-stimulatory molecules, co-inhibitory molecules, and MHC-related genes, was compared between high- and low-risk groups. Functional validation of the key prognostic gene EHHADH Spearman correlation coefficients between the expression of each model gene and the overall risk score were calculated, with particular focus on EHHADH. Kaplan–Meier survival analysis was performed in four independent cohorts to compare OS between EHHADH high- and low-expression groups. Immunohistochemistry images of EHHADH protein expression in normal lung tissue and lung adenocarcinoma tissue were retrieved from the Human Protein Atlas (HPA) database (antibody: HPA036401) to assess protein-level expression differences. EHHADH was stably knocked down in A549 and H1299 lung cancer cell lines using short hairpin RNA (shRNA; sh-EHHADH), with a non-targeting sequence serving as the negative control (NC). Knockdown efficiency was confirmed by quantitative real-time PCR (qRT-PCR). Cell adhesion assays were performed to evaluate the effect of EHHADH knockdown on cell attachment capacity, and wound healing assays (24-h) were conducted to assess changes in cell migration. All experiments were performed in biological replicates. Data are presented as mean ± standard deviation (SD), and between-group differences were evaluated using Student's t-test. Single-cell RNA sequencing analysis Raw scRNA-seq data from the HRA001130 dataset [22] were processed using the Seurat package. Quality control was performed to exclude low-quality cells, retaining cells with at least 200 detected genes and genes expressed in at least 3 cells. Cells with a high proportion of mitochondrial gene expression (>10%) were excluded. Data were log-normalized using the “LogNormalize” method with a scaling factor of 10,000, followed by identification of highly variable genes using the “FindVariableFeatures” function. Batch effects were corrected using Seurat’s integration workflow. Dimensionality reduction was conducted using principal component analysis (PCA), and the top principal components were selected based on standard criteria (ElbowPlot). Cell clustering was performed using the “FindNeighbors” and “FindClusters” functions with default resolution parameters, and visualization was achieved using t-distributed stochastic neighbor embedding (t-SNE). Cell type annotation was conducted based on canonical marker genes. Marker genes for each cluster were identified using the “FindAllMarkers” function with default thresholds (logfc.threshold = 0.25, min.pct = 0.1) and the Wilcoxon rank-sum test. A total of 14 distinct cell types were identified. Samples were stratified into High_EHHADH and Low_EHHADH groups based on the proportion of EHHADH-positive cells. The average expression level of EHHADH across different cell types was calculated, and differences in cell type proportions between groups were assessed using the Wilcoxon rank-sum test. A p-value < 0.05 was considered statistically significant. Result Mutational landscape and identification of prognostic-related genes Fig. 1 is the schematic overview of the research workflow of this study. To systematically analyze the role of telomerase-related genes in lung cancer, we first examined the mutation characteristics of these genes in the TCGA database (Fig. 2A). Results revealed that missense mutations were the most common variant type, followed by frame shift and nonsense mutations. Among single nucleotide variants (SNVs), C→A transitions were most frequent (184 cases), suggesting specific mutational drivers. Waterfall plot analysis showed that 197 samples (87.95%) harbored at least one mutation in telomerase-related genes, with ATP7A, OGDHL, and AOX1 being the most frequently mutated genes, highlighting their importance in lung cancer (Fig. 2B). To identify clinically significant genes from numerous candidates, we performed integrated analysis across seven independent public datasets. Through Cox proportional hazards regression analysis for each cohort and considering consistency across multiple datasets, we identified eight key genes: INMT, ALDH3A2, OGDHL, TDO2, EFNA3, STAT1, EHHADH, and AANAT. Heatmap analysis revealed distinct expression patterns of these eight genes between normal and tumor tissues (Fig. 2C). To validate the biological significance of these eight candidate genes, we performed comprehensive multi-cohort analysis. Through batch-effect corrected PCA analysis, samples from seven cohorts (gse613213, gse26939, gse29016, gse30219, gse31210, gse42127, and TCGA) showed good clustering and separation (Fig. 2D, center scatter plot). These eight genes were distributed across different chromosomal locations in the human genome, with red dots representing genes upregulated in tumors and black dots representing genes upregulated in normal tissue, demonstrating distinct associations between individual genes and tumor phenotypes (Fig. 2D, peripheral circular plot). Univariate and multivariate Cox regression analysis further revealed the prognostic significance of these eight genes for overall survival in lung cancer patients (Fig. 2E). INMT, ALDH3A2, and OGDHL emerged as protective factors (HR<1, blue dots), suggesting that elevated expression of these genes correlates with improved prognosis. Conversely, EFNA3, STAT1, EHHADH, TDO2, and AANAT served as hazardous factors (HR>1, orange dots), with higher expression associated with poorer prognosis. The risk coefficients of these genes can be utilized to construct prognostic prediction models. To elucidate the biological processes involving these eight genes, we performed KEGG and GO enrichment analysis. KEGG pathway analysis revealed that these genes were primarily enriched in multiple metabolic-related pathways including 2-oxocarboxylic acid metabolism, propanoate metabolism, aspartate metabolism, and citric acid cycle, while also being associated with histone metabolism, fatty acid degradation, and valine/leucine/isoleucine degradation (Fig. 2F). GO enrichment analysis indicated that these genes were functionally associated with microbody protein degradation, peroxisomes, and tricarboxylic acid cycle enzyme complexes at the molecular level, while also involving intramolecular oxidoreductase activity and peptidyl-CoA hydrolase activity (Fig. 2G). These functional annotations suggest that telomerase-related genes participate in lung cancer development and progression by modulating cellular metabolism and energy metabolism processes. Schematic overview of the study design and analytical workflow. This study systematically illustrates the role of telomerase-related genes (TRGs) in lung adenocarcinoma (LUAD). Public datasets were integrated to identify prognostic TRGs, followed by the construction of a robust machine learning–based prognostic model for risk stratification. The model was further validated across independent cohorts. Tumor microenvironment (TME) characteristics and immune infiltration were comprehensively analyzed. Finally, key target genes, particularly EHHADH, were functionally validated through in vitro experiments and single-cell RNA sequencing to elucidate their roles in tumor progression and immune modulation. Identification and Characterization of Prognostic Telomerase-Related Genes in Lung Cancer. (A) Mutational landscape of telomerase-related genes in TCGA lung cancer samples. Bar plots show the distribution of variant classifications (left), variant types (middle), and SNV classes (right). C→A transitions represent the most frequent mutational event (144 cases). (B) Waterfall plot displaying the mutation profile across 224 lung cancer samples. 197 samples (87.95%) harbored at least one mutation in telomerase-related genes. ATP7A, OGDHL, and AOX1 are the most frequently mutated genes, with mutation frequencies shown in the right panel. (C) Heatmap showing expression profiles of the eight prognostic genes across normal and tumor tissues from multiple cohorts. INMT shows elevated expression in normal tissues, while ALDH3A2, OGDHL, TDO2, EFNA3, STAT1, EHHADH, and AANAT display higher expression in tumor tissues, indicating tissue-specific expression patterns. (D) Multi-cohort integrated analysis of the eight candidate genes. Center scatter plot shows PCA analysis with batch-effect correction across seven datasets (gse613213, gse26939, gse29016, gse30219, gse31210, gse42127, and TCGA), demonstrating good sample clustering and separation. Peripheral circular plot shows chromosomal distribution of the eight genes, with red dots indicating tumor-upregulated genes and black dots indicating genes upregulated in normal tissues. (E) Univariate Cox regression analysis showing hazard ratios (HR) and 95% confidence intervals for each gene. Blue dots represent protective factors (HR<1), while orange dots represent hazardous factors (HR>1), significantly correlating with overall survival in lung cancer patients. (F) KEGG pathway enrichment analysis showing top enriched pathways associated with the eight genes. (G) Gene Ontology (GO) enrichment analysis displaying biological processes (BP), cellular components (CC), and molecular functions (MF) associated with the eight genes. Construction and validation prognostic risk model To establish an optimal prognostic prediction model, we integrated eight prognostic-related genes into multiple machine learning algorithms for model selection. Using the average C-index values from six GEO validation cohorts as the primary criterion, we identified the best-performing risk model (Fig. 3A). The model demonstrated strong predictive capability in the TCGA training cohort, effectively discriminating between high-risk and low-risk patients. Kaplan-Meier survival analysis revealed significant differences in overall survival between the two groups (log-rank p < 0.0001). Time-dependent ROC curve analysis at 1-, 3-, and 5-year time points showed AUC values of 0.64, 0.62, and 0.62, respectively, indicating excellent predictive accuracy of the model (Supplementary Figure 1). We further validated the generalizability of the risk model across six independent GEO validation cohorts (Fig. 3B). Kaplan-Meier survival analysis demonstrated that the model effectively stratified patients into high-risk and low-risk groups across all validation cohorts, with significant survival differences between the two groups. Specifically, in the GSE13213 cohort, high-risk patients exhibited substantially reduced overall survival compared with low-risk patients (log-rank p = 0.0012), with time-dependent ROC curve analysis showing AUC values of 0.8, 0.73, and 0.7 at 1-, 3-, and 5-year time points, respectively. In the GSE26939 cohort, high-risk patients demonstrated significantly worse survival outcomes than low-risk patients (log-rank p = 0.011), with AUC values of 0.63, 0.59, and 0.57 at 1-, 3-, and 5-year time points, respectively. In the GSE29016 cohort, survival curves were clearly separated between the two groups (p = 0.00011), with AUC values of 0.54, 0.7, and 0.71. In the GSE30219 cohort, high-risk patients exhibited significantly worse prognosis (p < 0.0001), with AUC values of 0.58, 0.7, and 0.74. In the GSE31210 cohort, survival differences were similarly significant (p = 0.00015), with AUC values of 0.72 and 0.77. In the GSE42127 cohort, significant survival differences were observed between the two groups (p = 0.0073), with AUC values of 0.62, 0.64, and 0.66. These findings comprehensively demonstrated that the established risk model possesses robust discriminatory ability and excellent predictive accuracy across multiple independent validation cohorts, indicating important clinical utility and potential for translational application of this prognostic risk model. Construction and multi-cohort validation of the prognostic risk model. (A) Comparison of model performance across multiple machine learning algorithms in the TCGA training cohort and six independent GEO validation cohorts (GSE13213, GSE26939, GSE29016, GSE30219, GSE31210, and GSE42127), evaluated by the average C-index. (B) Kaplan–Meier survival curves and time-dependent ROC curves for the prognostic risk model across six independent GEO validation cohorts. Clustering analysis and performance superiority verification of the prognostic model To further validate the accuracy and discriminatory ability of the established risk model, we performed principal component analysis (PCA) based on the expression of model genes in the TCGA training cohort and six GEO validation cohorts. Results demonstrated that patients in all seven cohorts were clearly segregated into two distinct clusters (Fig. 4A). Specifically, the red cluster represented high-risk patients, while the green cluster represented low-risk patients, with good separation between the two clusters and clear spatial distribution. This distinct clustering pattern was most prominent in the TCGA cohort and was consistently verified across the six validation cohorts (GSE13213, GSE26939, GSE29016, GSE30219, GSE31210, and GSE42127), fully demonstrating the model's excellent patient stratification capability and high accuracy. Validation of the Five-Gene Risk Model (A) Principal component analysis (PCA) of patients stratified by the five-gene risk model. High-risk patients are shown in red, low-risk patients in green. Clear separation between groups across TCGA and six GEO validation cohorts demonstrates the model's discriminatory power. (B) Performance comparison of the five-gene model with previously published prognostic models. The five-gene model ranks highest based on concordance index (c-index) values, indicating superior predictive accuracy. To evaluate the superiority of our established model relative to previously reported models, we systematically compared the risk scores calculated based on model gene expression and corresponding coefficients with models documented in previous literature. Through ranking and performance evaluation of risk scores for all patients in the TCGA training cohort, results revealed that our constructed model (MY MODEL) ranked first in performance metrics and substantially outperformed previously published models (Fig. 4B). These findings fully confirmed the superiority of the eight prognostic-related genes selected and optimized through machine learning algorithms, demonstrating that our model possesses stronger discriminatory ability and higher predictive accuracy compared to existing models. Exploring the association between prognostic risk model and tumor microenvironment To investigate the potential reasons for the differential prognosis between high-risk and low-risk patients, we focused on the immune infiltration within the tumor microenvironment (TME). Firstly, we utilized The Cancer Immunome Atlas (TCIA) to calculate the Immunophenoscore (IPS) for TCGA-LUAD samples. Our analysis revealed that the low-risk group exhibited significantly higher IPS scores across different immune checkpoint profiles (CTLA4-pos PD1-pos, CTLA4-neg PD1-pos, CTLA4-pos PD1-neg, and CTLA4-neg PD1-neg) compared to the high-risk group (Fig. 5A, all p < 0.01 or p < 0.001). These findings suggest that patients in the low-risk group may be more likely to benefit from immunotherapy. Next, we assessed the correlation between the prognostic risk score and four key scores derived from the ESTIMATE R package (StromalScore, ImmuneScore, ESTIMATEScore, and TumorPurity) (Fig. 5B). Our analysis indicated that TumorPurity was positively correlated with the risk score (r = 0.35, q = 4.4e-16), meaning higher tumor purity was associated with higher risk. Conversely, StromalScore (r = −0.29, q = 3e-11), ImmuneScore (r = −0.18, q = 2.8e-05), and ESTIMATEScore (r = −0.23, q = 1.4e-07) were all negatively correlated with the risk score, implying that a higher risk score corresponds to a lower stromal and immune component within the TME. To further characterize the immune landscape, we extracted quantitative data from seven immune assessment methods within the TIMER 2.0 database for TCGA data. By comparing high-risk and low-risk groups, we observed that the low-risk group consistently exhibited higher levels of immune cell infiltration (Fig. 5C). This comprehensive analysis, encompassing various immune cell types quantified by different algorithms (e.g., QUANTISEQ, CIBERSORT, CIBERSORT-ABS, XCELL, MCPCOUNTER, EPIC, TIMER), highlights a more robust immune presence in the low-risk patient cohort. Finally, we examined the expression levels of immune-related genes between the high-risk and low-risk groups (Fig. 5D). Our results indicated that a majority of immune-related genes, including co-stimulatory molecules (e.g., CD27, ICOS, CD86), co-inhibitory molecules (e.g., PDCD1, CD274, CTLA4), other MHC molecules, MHC class II molecules, and MHC class I molecules, were generally highly expressed in the low-risk group. This widespread upregulation of immune-related genes in the low-risk group further supports the notion of an "immune-hot" phenotype associated with better prognosis, contrasting with the "immune-cold" phenotype potentially characterizing the high-risk group. Association between the prognostic risk model and the tumor immune microenvironment in TCGA-LUAD. (A) Violin plots comparing the Immunophenoscore (IPS) between low-risk and high-risk groups across four immune checkpoint combination profiles (CTLA4-pos PD1-pos, CTLA4-neg PD1-pos, CTLA4-pos PD1-neg, and CTLA4-neg PD1-neg). The low-risk group exhibited significantly higher IPS scores in all subgroups (** p < 0.01; *** p < 0.001). (B) Scatter plots showing the Spearman correlations between the risk score and ESTIMATE-derived scores, including StromalScore, ImmuneScore, ESTIMATEScore, and TumorPurity. Green and red dots represent low-risk and high-risk patients, respectively. (C) Heatmap displaying immune cell infiltration levels across high-risk and low-risk groups, as quantified by seven deconvolution algorithms (QUANTISEQ, CIBERSORT, CIBERSORT-ABS, XCELL, MCPCOUNTER, EPIC, and TIMER). The low-risk group demonstrated consistently higher immune cell infiltration across all algorithms. (D) Heatmap illustrating differential expression of immune-related genes between risk groups, categorized into co-stimulatory molecules, co-inhibitory molecules, other MHC molecules, MHC class II molecules, and MHC class I molecules. The majority of these genes were significantly upregulated in the low-risk group. Significance levels: ns, not significant; * p < 0.05; ** p < 0.01; *** p < 0.001. Functional validation of key prognostic genes To further investigate the individual contributions of the identified prognostic genes, we first examined the correlation between each model gene and the overall risk score. Among them, EHHADH showed a significant positive correlation with the risk score (r = 0.4, q = 0) (Fig. 6A), indicating that higher EHHADH expression is associated with an increased risk. To validate the prognostic significance of EHHADH, we performed survival analysis across multiple independent cohorts. Consistent with its positive correlation with the risk score, higher expression of EHHADH was significantly associated with poorer overall survival in four distinct cohorts (Log-rank p < 0.0001, p = 0.00015, p = 0.00018, and p = 1e-04, respectively) (Fig. 6B). This robustly confirms EHHADH as a hazardous factor in lung cancer prognosis. Furthermore, we explored the expression pattern of EHHADH in human tissues using the Human Protein Atlas (HPA) database. The immunohistochemistry images revealed that EHHADH expression was markedly higher in lung cancer tumor tissues compared to normal lung tissues (Fig. 6C). Specifically, in normal alveolar cells (HPA036401), EHHADH staining was low (intensity: moderate, quantity: <25%), while in lung cancer tumor cells (HPA036401, adenocarcinoma, NOS), EHHADH staining was medium (intensity: moderate, quantity: >75%) and predominantly localized in the cytoplasmic and membranous/nuclear regions (Fig. 6D), further supporting its role in tumorigenesis. To experimentally elucidate the functional role of EHHADH in lung cancer, we performed knockdown experiments in lung cancer cell lines. We successfully knocked down EHHADH in both A549 and H1299 cell lines using shRNA (sh) compared to non-targeting control (NC) (Supplementary Figure 2A and 2B show relative mRNA expression levels, with significant reduction of EHHADH mRNA in sh-EHHADH groups, ****p < 0.0001 and ***p < 0.001 respectively). Functional assays demonstrated that EHHADH knockdown led to a significant increase in cell adhesion in both A549 and H1299 cell lines (Fig. 6E, Supplementary Figure 2C and 2D). Concurrently, wound healing assays showed a notable decrease in cell migration rate in both A549 and H1299 cells upon EHHADH knockdown (Fig. 6F, Supplementary Figure 2E and 2F, with reduced migration rates in sh-EHHADH groups compared to NC). Collectively, these in vitro results indicate that knocking down EHHADH significantly reduces the migratory and invasive capabilities of lung cancer cells, suggesting a crucial role for EHHADH in promoting tumor metastasis. Functional validation of EHHADH as a key prognostic gene in lung cancer. (A) Spearman correlation between the expression of five prognostic model genes (STAT1, INMT, EHHADH, EFNA3, and ALDH3A2) and the risk score. Orange and blue dots represent high- and low-risk patients, respectively. (B) Kaplan–Meier survival curves of EHHADH high- and low-expression groups across four independent cohorts. High EHHADH expression was significantly associated with worse overall survival (all log-rank p < 0.001). (C–D) Representative immunohistochemistry images from the Human Protein Atlas (HPA) database showing EHHADH protein expression in normal lung tissue (C) and lung adenocarcinoma tissue (D). EHHADH expression was markedly elevated in tumor cells (staining quantity >75%) compared to normal alveolar cells (staining quantity <25%). (E) Cell adhesion assays in A549 (left) and H1299 (right) cells following EHHADH knockdown (sh) or negative control (NC) treatment. (F) Wound healing assays in A549 (left) and H1299 (right) cells. EHHADH knockdown markedly decreased cell migration compared to NC groups. Single-cell RNA sequencing reveals EHHADH's role in shaping the tumor microenvironment To further dissect the role of EHHADH within the tumor microenvironment (TME) at single-cell resolution, we analyzed single-cell RNA sequencing data from the HRA001130 dataset. Samples were stratified into high-EHHADH and low-EHHADH groups based on the percentage of EHHADH-positive cells (Fig. 7A), reflecting varying levels of EHHADH expression across the samples. Following preliminary dimensionality reduction and clustering (Supplementary Figure 3A for marker gene annotation), we identified 14 distinct cell types within the TME (Fig. 7B). Subsequent analysis of EHHADH mean expression across these different cell types revealed that EHHADH was predominantly expressed in epithelial cells (Fig. 7C), suggesting its primary role within the tumor cells themselves. Further investigation into cell composition analysis between the high-EHHADH and low-EHHADH groups (Fig. 7D and Supplementary Figure 3B) demonstrated a significant shift in the TME landscape. Specifically, the low-EHHADH group exhibited a higher proportion of immune cell infiltration, including T cells (p = 0.0011), NK cells (p = 0.007), and proliferating cells (p = 0.05) compared to the high-EHHADH group. While epithelial cell populations were higher in the high-EHHADH group (p = 0.0019), other immune cell types such as B cells, macrophages, monocytes, and neutrophils showed trends, though not always statistically significant, toward higher proportions in the low-EHHADH group. These findings suggest that high EHHADH expression may be associated with an "immune-desert" or "immune-excluded" phenotype within the TME, whereas lower EHHADH expression correlates with a more immune-infiltrated microenvironment. Single-cell RNA sequencing analysis of EHHADH expression and its impact on the tumor microenvironment. (A) Bar chart showing the percentage of EHHADH-positive cells across individual samples from the HRA001130 dataset. Samples are stratified into High_EHHADH (yellow) and Low_EHHADH (red) groups based on EHHADH-positive cell proportion. (B) t-SNE plots illustrating the cellular composition of the tumor microenvironment in High_EHHADH (n = 43,597 cells) and Low_EHHADH (n = 47,400 cells) groups. A total of 14 distinct cell types were identified and color-coded accordingly. (C) Line plot depicting the mean expression level of EHHADH across 14 cell types. EHHADH was predominantly expressed in epithelial cells, indicating its primary activity within tumor epithelial compartments. (D) Box plots comparing the proportions of 14 cell types between High_EHHADH and Low_EHHADH groups across individual samples. The Low_EHHADH group exhibited significantly higher proportions of T cells (p = 0.0011), NK cells (p = 0.007), and proliferating cells (p = 0.05), while epithelial cells were enriched in the High_EHHADH group (p = 0.0019). p-values were calculated using the Wilcoxon rank-sum test. Discussion The complexity and heterogeneity of lung cancer represent the principal determinants of its unfavorable prognosis, and the development of precise prognostic assessment tools and individualized therapeutic strategies remains a central focus of contemporary lung cancer research [23]. In the present study, through the integration of multi-omics data, advanced bioinformatic analytical approaches, and experimental validation, we comprehensively investigated the roles of telomere-related genes (TRGs) in lung cancer, characterized their associations with the tumor immune microenvironment, and successfully constructed a prognostic risk model with high predictive performance. We first conducted a comprehensive dissection of the mutational landscape of TRGs in the TCGA lung adenocarcinoma cohort. Although previous studies have sporadically reported correlations between mutations or aberrant expression of specific telomerase component genes—such as TERT—and clinical prognosis in lung cancer, the present study systematically surveyed a broader spectrum of TRGs, revealing that as many as 87.95% of samples harbored at least one mutation, with missense mutations constituting the predominant variant type [24]. Of particular note, circulating TERT has recently been validated as a novel prognostic and diagnostic biomarker in resectable non-small cell lung cancer (NSCLC), with its perioperative dynamic changes closely correlated with patient survival [25]. Collectively, these findings suggest that genetic alterations in TRGs are a pervasive feature of lung cancer and may drive tumorigenesis and disease progression by impairing protein function, thereby providing a more comprehensive perspective for understanding the oncogenic mechanisms of this gene family. Through multi-cohort integrative analysis, we rigorously identified eight core genes with independent prognostic value: INMT, ALDH3A2, OGDHL, TDO2, EFNA3, STAT1, EHHADH, and AANAT. The expression patterns of these genes differed significantly between tumor and normal tissues, and they were further stratified into protective and risk factors. Among them, TDO2 has been demonstrated to orchestrate an immune-excluded, ferroptosis-resistant tumor microenvironment in solid-predominant lung adenocarcinoma through the cooperative interplay between IL4I1+ tumor-associated macrophages and TDO2+ myofibroblasts, which drives aryl hydrocarbon receptor (AhR) signaling activation and is closely associated with poor prognosis [26]. STAT1, a pivotal node in IFN-γ signal transduction, has been shown to participate in the regulation of anti-tumor immune responses through reciprocal inhibition with TP63, while the MED12–STAT1–TAP2 axis has been identified as a key regulator of CD8+T cell cytotoxic activity that influences immunotherapy efficacy in NSCLC [27,28]. The prognostic risk model constructed in this study demonstrated superior predictive capability and robustness across multiple independent validation cohorts, outperforming the majority of previously reported models built upon a limited number of genes or a single cohort. Many existing lung cancer prognostic models have focused on specific driver mutations—such as those in EGFR or KRAS—or a restricted set of immune-related genes, and their applicability and predictive accuracy may be constrained by the inherent heterogeneity of patient populations [29]. By contrast, our model integrates genes implicated in diverse biological processes—including metabolism, immunity, and cellular proliferation—and its broader applicability and predictive stability were ensured through rigorous machine learning optimization and multi-cohort cross-validation [30]. This multi-dimensional, biologically informed gene integration strategy for prognostic model construction, combined with the advantages of large-scale data mining and extensive independent external validation, substantially enhances the model's potential for clinical translation [31]. Of particular importance, the present study conducted an in-depth investigation of the complex associations between the prognostic risk model and the tumor microenvironment. We observed that high-risk patients typically exhibited an "immune-cold" microenvironmental phenotype characterized by low levels of immune cell infiltration, whereas low-risk patients displayed an "immune-hot" phenotype with markedly enhanced immune cell infiltration and upregulated expression of immune-related genes. It has been well established that "cold" and "hot" tumors differ fundamentally in the density and functional states of infiltrating immune cells, with these differences attributable to multifaceted mechanisms including metabolic reprogramming, immune exclusion, and T cell functional exhaustion, which collectively determine the therapeutic response to immune checkpoint inhibitors [14]. In the field of lung cancer research, single-cell RNA sequencing has been extensively employed to elucidate differences in the cellular composition of the tumor microenvironment among patient populations with divergent immunotherapy responses [32], while spatial multi-omics approaches have further confirmed that the spatial distribution patterns of intratumoral immune cells serve as effective tools for predicting immunotherapy prognosis in advanced NSCLC [33]. To our knowledge, the present study is among the first to deeply integrate a TRG-based prognostic risk model with tumor immune microenvironmental characteristics, unveiling the intimate bidirectional interplay between these two dimensions and providing novel mechanistic insights into lung cancer progression and immunotherapy resistance, as well as a potential biological rationale for identifying patients most likely to benefit from immunotherapy. Furthermore, we performed cytological functional validation and single-cell RNA sequencing analysis of the key prognostic gene EHHADH. Identified as a risk factor, elevated EHHADH expression was associated with adverse clinical outcomes. Recent investigations have established that EHHADH functions as a critical regulator of lipid metabolism in lung squamous cell carcinoma, with lipidomic analyses confirming its involvement in the regulation of lung cancer initiation, progression, and treatment response, underscoring its potential value as both a diagnostic biomarker and a target for individualized therapy [34]. Our experimental findings confirmed that EHHADH knockdown significantly suppressed the migratory and invasive capacities of lung cancer cells, a result that is consistent with single-cell analyses demonstrating predominant EHHADH expression in epithelial cell populations, which may attenuate immune cell infiltration through remodeling of the tumor microenvironment. Single-cell transcriptomic analyses further revealed that pronounced heterogeneity among AT2-type epithelial cells in lung adenocarcinoma is closely associated with patient prognosis, and that specific epithelial cell subpopulations may serve as candidate prognostic biomarkers [35]. These findings provide novel mechanistic insights into the roles of EHHADH in lung cancer cell biology and tumor microenvironmental remodeling, identify it as a promising candidate target for future therapeutic development, and further underscore the innovation and translational significance of the present study. Nevertheless, several limitations of the present study warrant acknowledgment. Although large-scale public databases were utilized and multi-cohort validation was performed, all bioinformatic analyses were conducted on retrospective data. Future studies incorporating prospective clinical cohorts are needed to further substantiate the validity of the model. Moreover, the present study focused primarily on mRNA expression levels and mutational characteristics; future investigations integrating deeper layers of omics data—including proteomics and metabolomics—would facilitate a more comprehensive understanding of the regulatory networks and molecular mechanisms governing these genes in lung cancer. The specific signaling pathways and upstream/downstream regulatory mechanisms through which each prognostic gene exerts its effects in lung cancer also warrant further elucidation through in-depth molecular biological experimentation. In conclusion, the present study not only constructed a highly accurate and extensively validated prognostic risk model for lung cancer, but also systematically delineated the mutational landscape of TRGs in lung cancer, their prognostic significance, and their interactions with the tumor immune microenvironment. These innovative findings provide a robust theoretical foundation and potential biomarkers for the precise diagnosis of lung cancer, the formulation of individualized treatment regimens, and the optimization of immunotherapeutic strategies, and are anticipated to exert a positive impact on the clinical management and long-term prognosis of lung cancer patients. Ethics statement No new patient samples, animal experiments, or other procedures requiring additional ethical approval were involved in the present study. Therefore, no additional ethics statement is required. Data availability statement All data used in this study were obtained from public databases or previously published literature. Gene expression, somatic SNV, and CNV data were downloaded from The Cancer Genome Atlas (TCGA, https://portal.gdc.cancer.gov/), and normal lung tissue expression data were obtained from the Genotype-Tissue Expression (GTEx, https://gtexportal.org/) database. LUAD-related GEO cohorts (GSE13213, GSE26939, GSE29016, GSE30219, GSE31210, GSE42127) were sourced from the Gene Expression Omnibus (https://www.ncbi.nlm.nih.gov/geo/). Single-cell transcriptomic data were obtained from the Genome Sequence Archive for Human (GSA for Human) of the China National Center for Bioinformation (CNCB, https://ngdc.cncb.ac.cn/gsa-human/, accession number HRA001130). Detailed information on data access and literature references can be found in the Materials and Methods section. Processed data and analysis code are available from the corresponding author upon reasonable request. Funding No funding. CRediT authorship contribution statement Wei Wei: Writing – review & editing, Writing – original draft, Methodology, Investigation. Yuheng Wang: Supervision, Software. Diliyaer Abudukeremu: Visualization, Validation, Software, Project administration. Jinhua Luo: Formal analysis, Data curation. Mingjun Du: Formal analysis, Data curation. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this manuscript. No commercial or financial entities have provided funding or other forms of support for this study, and none of the authors hold any relevant patents, stock, or other financial assets related to the content of the research. All authors confirm that there are no other potential conflicts of interest that need to be disclosed. Footnotes Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.tranon.2026.102852. Contributor Information Jinhua Luo, Email: Ljhua1966@126.com. Mingjun Du, Email: dumingjun1989@163.com. Supplementary materials References Associated Data Supplementary Materials Data Availability Statement All data used in this study were obtained from public databases or previously published literature. Gene expression, somatic SNV, and CNV data were downloaded from The Cancer Genome Atlas (TCGA, https://portal.gdc.cancer.gov/), and normal lung tissue expression data were obtained from the Genotype-Tissue Expression (GTEx, https://gtexportal.org/) database. LUAD-related GEO cohorts (GSE13213, GSE26939, GSE29016, GSE30219, GSE31210, GSE42127) were sourced from the Gene Expression Omnibus (https://www.ncbi.nlm.nih.gov/geo/). Single-cell transcriptomic data were obtained from the Genome Sequence Archive for Human (GSA for Human) of the China National Center for Bioinformation (CNCB, https://ngdc.cncb.ac.cn/gsa-human/, accession number HRA001130). Detailed information on data access and literature references can be found in the Materials and Methods section. Processed data and analysis code are available from the corresponding author upon reasonable request.