Introduction

Lung adenocarcinoma (LUAD) accounts for approximately 40% of all lung cancer (LC) cases, the most common cause of cancer-related death worldwide1,2. The tumor microenvironment (TME) in LUAD is highly infiltrative and destructive, allowing cancer cells to invade blood vessels and lymphatics, leading to metastasis via the circulatory as well as lymphatic systems. The introduction of immunotherapies targeting immune checkpoints has significantly improved therapeutic outcomes and revolutionized the treatment of LUAD3,4. Despite notable advancements in treatment, the five-year overall survival (OS) for LUAD is still below 20%5,6. Additionally, tumor heterogeneity, low response rates, and therapeutic resistance significantly hinder the efficacy of existing targeted therapies7,8. Therefore, to enhance the therapy of LUAD, it is imperative to identify certain molecular markers and establish precise therapeutic techniques.

The stimulator of interferon genes (STING) serves as a critical adaptor molecule triggering innate immune response following the recognition of intracellular pathogenic DNA, playing a critical role in controlling antiviral responses and detecting tumor formation9,10,11,12. Research has identified two primary mechanisms through which STING influences tumor development. On the one hand, STING activation promotes Type I interferon (IFN) synthesis and induces the maturation of dendritic cells (DCs) and macrophages, thereby enhancing the adaptive immune response and inhibiting tumor growth11,13,14. On the other hand, STING may promote tumor progression in the context of chronic inflammation by contributing to tumorigenesis or metastasis, a process linked to endoplasmic reticulum (ER) stress and T-cell survival, particularly CD8 + T cells. This mechanism is critical and independent of IFN signaling15,16. For example, Ze Lin et al. reported that low STING expression in LUAD tissues promotes cancer progression through the cyclic GMP-AMP synthase (cGAS)-STING pathway17,18. Jianjun Wu et al. demonstrated that STING mutations result in lung inflammation and immune deficiency, characterized by reduced T-cell levels in mice, suggesting a link between this mechanism and ER stress15. Activation of STING signaling in antigen-presenting cells (APCs) within the tumor microenvironment (TME) drives the production of Type I interferons, thereby enhancing the cross-priming of effector CD8⁺ T cells19,20. Previous studies have shown that reduced expression of STING/cGAS in lung cancer is associated with poor prognosis21. However, few comprehensive studies exist, and no definitive conclusions have been drawn regarding the role of STING-related genes (STING-RGs) in LUAD progression.

Consequently, we aim to identify STING-related prognostic genes in LUAD. We explored the potential mechanisms by which STING regulates LUAD progression. Additionally, we constructed a new STING-related prognostic model and performed drug sensitivity analysis, offering a theoretical basis for clinical prognostication and LUAD treatment prediction.

Materials and methods

Preparation of relevant data

The inclusion criteria of patients in our research were as follows: (1) Powerful evidence for the diagnosis of LC, (2) Confirming LUAD at the histopathological level, (3) Available survival time, status and clinical information. The Cancer Genome Atlas (TCGA) was queried for LUAD-related datasets, including gene expression and clinical information matrices for 526 LUAD samples and 59 control samples (training dataset). The data types included HTSeq-FPKM and HTSeq-Counts. For differential expression analysis, the HTSeq-Counts expression matrix was used, whereas the HTSeq-FPKM transformed TPM expression matrix (exp(log(fpkm) - log(sum(fpkm)) + log(1e6))) was employed for all other analyses. Gene symbols were employed to convert the ensemble IDs in the gene expression matrix, in conjunction with the corresponding annotation files for the respective species. Furthermore, the Gene Expression Omnibus provided the GSE13213 dataset, which is a validation dataset based on the GPL6480 platform and includes gene expression as well as clinical data matrices for 117 LUAD samples22. A query of the Molecular Signatures Database (MSigDB) with the term “STING” identified 47 STING-related genes (STING-RGs).

Analysis of the differential expression (DE) between the LUAD and control groups in the training dataset (TD)

Differential expression (DE) analysis between LUAD and control groups in training dataset was performed using DESeq2 (version 1.38.0) to identify differentially expressed genes (DEGs)23. The thresholds for differential expression analysis were set at |log2FC| > 0.5 and p < 0.05.

Acquisition of key modules and genes by gene co-expression network analysis

The GSVA program (version 1.42.0)’s single-sample gene set enrichment analysis (ssGSEA) was utilized to determine the 47 STING-RG scores for every LUAD sample in the TD24. LUAD samples were then classified into HSG and LSG based on the median score. Kaplan-Meier (KM) analysis, using Wilcox test, was carried out on the two score groups to evaluate the impact of STING-RGs on LUAD patient survival.

To identify co-expressed gene modules as well as key modular genes associated with STING, co-expression networks were built via the weighted gene co-expression network analysis package, with STING-RG scores as the trait25. First, outlier samples were removed, followed by hierarchical cluster analysis. The optimal soft-threshold power (β) was then determined via a scale-free topology fit index of 0.9. Dissimilarity coefficients between genes were derived by calculating neighborhood and similarity measures, resulting in a systematic clustering tree with a minimum module size of 200. The association between STING-RG scores and module performance was then examined via Pearson correlation analysis. Modules exhibiting the strongest correlation with STING-RG scores, as depicted in a correlation heatmap, were designated as key modules. To illustrate the relationship between genes, modules, and traits, scatter plots of Gene Significance and Module Membership were generated. Genes with |MM| > 0.8 as well as |GS| > 0.2 were chosen as critical module genes.

Acquisition of differentially expressed STING-RGs (DE-STING-RGs) with prognostic significance

STING-RGs with prognostic significance were identified by performing the proportional hazards (PH) assumption test and univariate CR analysis on key module genes related with STING in disease samples from the training dataset (p < 0.05). Differentially expressed STING-RGs (DE-STING-RGs) with prognostic significance were then identified by intersecting DEGs and STING-RGs with prognostic significance. Additionally, the chromosomal distribution of DE-STING-RGs with prognostic significance was analyzed and visualized using Circos (http://circos.ca/)26.

Functional annotations for DE-STING-RGs with prognostic significance

To investigate the biological processes responsible for DE-STING-RGs with prognostic significance in the progression of LUAD, functional enrichment analysis of DE-STING-RGs with prognostic significance was performed using Metascape (https://www.metascape.org). A minimum overlap of 3, a minimum enrichment factor of 1.5, as well as a p-value threshold of less than 0.05 were the requirements for the screening process. Additionally, the Molecular Complex Detection (MCODE) algorithm was used to identify highly interconnected biological processes and annotate them in the network generated through functional enrichment analysis, using the PHYSICAL-CORE database. To build a PPI network, a range of 500 was selected for the maximum network size, with a minimum node size of 3.

The development, assessment, and verification of a risk model (RM)

The least absolute shrinkage and selection operator (LASSO) algorithm in the glmnet package (edition 4.1-4.1) was initially employed to screen prognostic genes and develop a RM based on DE-STING-RGs with prognostic significance27. The formula was: Risk Score = sum(Xi * coefi), in which Xi represented the relative expression of prognostic gene i, as well as coefi was the LASSO Cox coefficient for prognostic gene i. Subsequently, LUAD individuals in the TD were categorized into HRG and LRG based on the RS produced by the RM, via the optimal cut-off value. The validity of RM was assessed by conducting KM analysis using the survminer package (edition 0.4.9) for the two risk groups, with log-rank tests used to determine survival rate differences28. Additionally, receiver operating characteristic (ROC) curves for LUAD patients at 1, 3, and 5 years were generated via survival ROC package (edition 1.0.3.1)29. The Area Under the Curve (AUC) values were computed to evaluate RM’s effectiveness, with values exceeding 0.6 indicating commendable diagnostic efficacy. A heatmap was used to illustrate the expression patterns of prognostic genes in the two risk groups. Finally, the RM was validated using the validation dataset.

The associations with clinical features and RS

The clinical features of LUAD, including age (> 60 or ≤ 60 years), gender (female or male), tumor (T) stage (T1, T2, T3, T4), node (N) stage, as well as metastasis (M) (M0, M1), were analyzed using the Wilcoxon test to evaluate their association with the RS22. The association between prognostic genes and clinical features of patients was also explored. The survival ROC function was used to generate ROC curves at 1, 3, as well as 5 years to evaluate the efficacy of RS and clinical features in predicting LUAD prognosis.

Evaluation of independent prognostic factors (IPFs) and nomogram construction

The clinical features and RS were incorporated into independent prognostic analyses22. Univariate CR analysis, the PH assumption test, as well as multivariate CR analysis were sequentially employed to find IPFs (p < 0.05). A nomogram for 1-, 3-, and 5-year survival in LUAD patients was then constructed on the basis of the independent prognostic factors. In addition, calibration curves were produced in order to evaluate the nomogram’s validity for LUAD.

Enrichment analysis (EA) between two risk group

Differential expression (DE) analysis between the two risk groups in the TD was carried out using DESeq2, as well as log2FC values were produced. The log2FC values were ranked from largest to smallest, followed by gene set EA using clusterProfiler package (edition 4.7.1.3)30,31. The reference gene set used was “cc2.cp.kegg.v7.4.symbols.gmt” from the MSigDB database. Furthermore, GSEA was also employed to identify biological processes or pathways associated with prognostic genes. To further investigate variations in biological functions between the two risk groups, gene set variation analysis (GSVA) scores were generated for each pathway via the GSVA package, with the same reference gene set as in the gene set EA. Additionally, associations between the RS and SPs were explored.

Immune landscape analysis of LUAD

Immune cell infiltration levels in the two risk groups from TD were quantified using ssGSEA. The ssGSEA algorithm was then used to calculate enrichment scores (ESs) for 13 immunological functions in the two risk groups28. Differences in immune function between the two groups were analyzed via Wilcoxon test (p < 0.05). Pearson correlation analysis was employed to evaluate the correlation between prognostic genes and various differential immune cells. Immunotherapies targeting immune checkpoints eliminate tumor cells by modulating T-cell activity through co-suppressing or co-stimulating signaling pathways. An analysis of differences in immune checkpoints, using data from 48 published immune checkpoints, was applied to the two risk groups32. To study the proportions of relevant immune cells, stromal cells, as well as tumor cells in the LUAD microenvironment, the StromalScore, ImmuneScore, and ESTIMATEScore of the biological samples were computed via ESTIMATE algorithm32. Additionally, Spearman correlation analysis was used to evaluate associations between RS, StromalScore, ImmuneScore, and ESTIMATEScore.

Exploration of the correlation with prognostic genes and the microenvironment of LUAD

To study the association between prognostic genes and single cells in a TME 16 single-cell RNA sequencing (scRNA-seq) datasets associated with LUAD were identified from the TME Single Cell Database (TISCH). The distribution and expression of prognostic genes in single cells were analyzed using the Seurat package (version 4.1.0)33.

Analysis of immunotherapy responses between two risk group

The tumor immune response is a critical process in tumor development, and immune escape (IE) facilitates tumor proliferation, invasion, and metastasis. The TIDE method was utilized to determine the TIDE score, Dysfunction score, as well as Exclusion score for the two groups in order to evaluate the possibility of tumor IE in various risk categories of LUAD patients. An increased TIDE score indicated a higher probability of IE in LUAD patients during treatment. To further analyze the distinct impacts of immune checkpoints (IC), specifically PD-L1 as well as CTLA4, on cytotoxic T lymphocytes across different groups, Immunophenotype Score (IPS) data for IC inhibitors were obtained from the Cancer Immunity Database. Finally, ssGSEA scores for predicted immunotherapy pathways in the training dataset of LUAD samples were calculated using the GSVA package. Subsequently, Spearman analysis was employed to evaluate the relevance of RS with immunotherapy pathways as well as the immune cycle.

Analysis of mutated landscapes

To assess the relationship between tumor mutational burden (TMB) and survival, LUAD samples were grouped into high-TMB as well as low-TMB groups on the basis of their TMB levels. OS rates in the two risk groups as well as TMB groups was then compared using KM survival curves. The relationship between RS and TMB was assessed via Spearman correlation analysis, incorporating all samples in the training dataset. Somatic mutations in LUAD patients within the two risk groups were further visualized and analyzed via maftools package to display tumor cell (TC) mutations34.

Sensitivity analysis of LUAD to chemotherapeutic drugs

To assess the susceptibility of LUAD patients to chemotherapy drugs, drug sensitivity data were obtained from the GDSC database. The pRRophetic function was then used to compute the IC50 of each drug for the samples in the training dataset (version 0.5)35. Finally, the Wilcoxon test was employed to analyze variations in IC50 values between the two groups, identifying differential responses to chemotherapy drugs. The association between prognostic genes and drug efficacy was determined by Spearman correlation analysis, using the NCI-60 cell line as the reference.

Expression analysis of prognostic genes

To confirm prognostic gene expression, an analysis was conducted on the mRNA expression patterns of these genes in both tissues and cells. Specifically, the analysis focused on comparing gene expression in LUAD tissues with control tissues in the training dataset. Additionally, prognostic gene expression levels in 570 LUAD cell lines obtained from the CCLE online database were analyzed. Protein expression levels of GAB3 were validated in three paired LUAD tumor and adjacent normal tissues obtained from surgically resected patients, using the Western blotting protocol described in Sect. “Western blotting”.

Cell culture and reagents

Human LUAD cell lines A549 and NCI-H23 were purchased from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China). Both cell lines were cultured in Dulbecco’s Modified Eagle Medium (DMEM) supplemented with 10% Fetal Bovine Serum (FBS) and 1% penicillin-streptomycin solution (100 U/mL penicillin, 100 µg/mL streptomycin). All cells were maintained in a humidified incubator at 37 °C with 5% CO2.

Plasmid and SiRNA transfection

The GAB3 overexpression plasmid (GAB3-OV) and the corresponding empty vector (Ctrl) were constructed by Genechem, Shanghai, China. Small interfering RNA targeting GAB3 (siGAB3) and a negative control siRNA (si-NC) were synthesized by GenePharma, Shanghai, China. For transfection, cells were seeded into 6-well plates and grown to 70–80% confluence. Plasmids 2.5 µg were transfected into cells using Lipofectamine 3000 according to the manufacturer’s protocol. Cells were collected for subsequent experiments at 48 h post-transfection.

Western blotting

Protein expression levels of prognostic genes were confirmed using Western blot. Samples were treated with RIPA lysis buffer, as well as protein content was measured via a BCA test kit (Thermofisher, USA). 20 µg of protein was loaded into each lane. Proteins were transferred to a PVDF membrane (Millipore, USA) after separation by polyacrylamide gel electrophoresis. The membrane was blocked with 5% milk as well as incubated overnight at 4 °C with primary antibodies diluted 1:1000: GAPDH (ab181602, Abcam, Cambridge, MA, USA), and anti-GAB3 (ab233411, Abcam, Cambridge, MA, USA). An improved chemiluminescence kit (Millipore, USA) was utilized to detect target proteins following a one-hour room temperature incubation with the matching horseradish peroxidase-linked secondary antibody.

Wound healing assay

GAB3 siRNA, overexpression A549 or NCI-H23 cells, as well as constructed empty vector (Ctrl) were all seeded at a density of 1 × 10^5 cells/well on a 24-well culture plate. The cells were then grown for 24 h at 37 °C with 5% CO2. After removing the culture media, a 10 µl pipette tip was used to gently scrape the surface of the seeded cells. The cells were gently washed twice with PBS before being given 1 milliliter of DMEM media containing 10% FBS. At 0, 24, and 48 h, pictures of the scratches were captured. All experiments were performed in triplicate.

CCK-8 assay

Constructed empty vector (Ctrl), GAB3 siRNA, as well as overexpression A549 or NCI-H23 cells were seeded in 96-well culture plates at a density of 1 × 10^4 cells/well as well as grown overnight in a 37 °C, 5% CO2 incubator. Following a pair of PBS washes, 100 µl of DMEM media containing 10% FBS was introduced to the cells. In each well of the 96-well plate, 90 µl of medium containing 10 µl of CCK-8 solution was added after the medium was changed at 0, 24, 48, and 72 h. After that, the plates were incubated for two hours at 37 °C. Using a microplate reader (BioTek Instruments, VT, USA), absorbance at 450 nm was determined for each well. Then, the ability of each group to proliferate was plotted.

Transwell migration assay

Transwell migration assays were performed using 8-µm pore size inserts (Corning, NY, USA). Forty-eight hours post-transfection, A549 and NCI-H23 cells were resuspended in serum-free medium and seeded into the upper chamber ($5 \times 10^4$ cells/well), while medium containing 10% FBS was added to the lower chamber. After incubation for 24 h, non-migrated cells were removed. The cells on the lower surface were fixed with 4% paraformaldehyde, stained with 0.1% crystal violet, and counted in five randomly selected fields under an inverted microscope.

Statistical analysis (SA)

For statistical analysis, R programming language (edition 4.2.3) as well as GraphPad 9.5 was used. The data was compared between different categories via the Wilcoxon test. Unless otherwise noted, p < 0.05 as well as p.adj < 0.05 was deemed statistically significant. Statistical significance was defined as follows: P < 0.05 (*), P < 0.01 (**), and P < 0.001 (***).

Results

Identification of 133 key module genes

The training dataset (LUAD vs. control) identified 9,585 DEGs, encompassing 6,119 upregulated as well as 3,466 downregulated DEGs (Fig. 1A-B). The LUAD samples in the training dataset (TD) were categorized into high-score group (HSG) and low-score group (LSG) based on the STING-RG enrichment scores. Significant variations in survival between the two groups were found using KM survival analysis (p = 0.029). Consequently, STING-RG enrichment scores were used as a trait to construct co-expression networks (Fig. 1C). Subsequently, clustering analysis was carried out on the TD samples, as well as no outliers were excluded (Fig. 1D). To ensure biologically meaningful scale-free topology, an optimal β value of 5 was selected, considering a scale independence criterion of > 0.9 (Fig. 1E). 13 co-expression modules were found with a minimum module size of 200, among which the MEyellow module (Cor = 0.72, p < 0.05) exhibited the strongest positive correlation with STING-RG enrichment scores (Fig. 1F-G). A total of 133 key module genes were extracted from the MEyellow module on the basis of the standard |MM| > 0.8 and |GS| > 0.2 (Fig. 1H).

Fig. 1
figure 1

Identification of key modular genes. (A).Genes with no discernible difference or a small fold change are shown by grey dots, whereas red dots are upregulated genes as well as blue dots denote downregulated genes. (B) Heatmap of differential gene expression analysis. The top half displays expression density, while the bottom half shows the expression heatmap. (C) KM survival analysis comparing survival rates between the two groups. (D) Hierarchical clustering analysis of gene expression. (E) The choice the power value for the soft threshold. The scale-free model fit index is displayed in the left panel, whereas the mean connectivity of these values is shown in the right side. (F) Gene cluster dendrogram using module assignment as well as dissimilarity metrics. Gene modules are shown in the bottom half, whereas a hierarchical clustering dendrogram of genes is displayed in the top half. (G) Heatmap showing the association between module genes and clinical characteristics. Groups are represented by the horizontal axis, association strength is shown by the numbers in the heatmap, as well as p-values are displayed by the values in parenthesis, where red denotes positive correlation and blue signifies negative correlation. (H) Selection of core genes from filtered modules.

Full size image

Screening of 38 prognostically significant DE-STING-RGs

From the 133 key module genes, those that met the PH assumption test (p > 0.05) were encompassed in univariate CR analysis (p < 0.05), yielding 48 STING-RGs with prognostic significance (Fig. 2A). Subsequently, genes present in both the 9,585 DEGs and the 48 prognostically significant STING-RGs were identified, resulting in a total of 38 DE-STING-RGs with prognostic relevance (Fig. 2B). The majority of these genes were localized on human chromosome 7 and exhibited significant downregulation in LUAD samples (Fig. 2C). Functional enrichment analyses as well as protein-protein interaction (PPI) networks indicated ACADVL, ACADM, ACADL, ACAA1, ACADSB, ACADS, and ACAT1 interact and cooperate in the fatty acid metabolism process (Fig. 2D-E).

Fig. 2
figure 2

Co-screening of DE-STING-RGs with prognostic significance. (A) Filtering of STING-RGs with prognostic significance using univariate Cox regression analysis. (B) Venn diagram showing the intersection of DEGs as well as STING-RGs with prognostic significance. (C) Chromosomal map of gene localization. (D-E) Functional enrichment analysis and PPI network maps.

Full size image

GAB3 and IL16 were confirmed as prognostic genes, and RM based on prognostic genes had promising predictive capability for survival in LUAD

Prognostication is critically important for patients with LUAD. Two key genes, GAB3 and IL16, were identified from DE-STING-RGs with prognostic significance using LASSO regression analysis the optimal log(λ) was − 3.507, and a RM was developed (Fig. 3A-B). To evaluate the reliability of the RM, the RS was computed using the equation: RS = (expression of GAB3 * −0.0599294226493178) + (expression of IL16 * −0.201038468078862). Based on the optimal RS cutoff (−1.335668), the LUAD samples were divided into LRG and HRG (Fig. 3C). As RS increased, a higher mortality rate was observed in LUAD samples, indicating a positive association between RS and mortality (Fig. 3D). Additionally, KM survival analysis showed that individuals in the low-risk LUAD group exhibited a remarkablly higher probability of OS (p < 0.05) (Fig. 3E). Moreover, AUC values for 1-, 3-, and 5-year survival exceeded 0.6, suggesting the RM effectively predicts survival outcomes in LUAD patients (Fig. 3F). Our analysis further revealed that high-risk LUAD samples were associated with low GAB3 and IL16 expression (Fig. 3G). Similar findings were replicated in the validation dataset, confirming the reliability of these results (Fig. 4A-E).

Fig. 3
figure 3

Predictive power of risk models of prognostic genes for LUAD survival. (A-B) LASSO regression coefficient plots and cross-validation curves. (C) LUAD samples were categorized into LRG and HRG based on the optimal RS threshold. Blue denotes the LRG, and red signifies HRG. (D) Correlation between RS and survival status. Blue is alive patients, and red denotes deceased patients. (E) Overall probability of survival in the low risk as well as high risk LUAD groups. Blue is the LRG, while red is the HRG. (F) 1-, 3-, and 5-year survival rates (SRs) in LUAD patients. Red indicates 1-year, green indicates 3-year, and blue indicates 5-year SRs. (G) Expression of GAB3 as well as IL16 in LUAD samples from HRG as well as LRG. Values closer to −1 mean lower expression, while those closer to 1 signify higher one.

Full size image
Fig. 4
figure 4

Validation data sets confirm the reliability of the analysis in Fig. 3. (A) LUAD samples were categorized into LRG as well as HRG based on the optimal RS threshold. Blue is the LRG, while red is the HRG. (B) Correlation between RS and survival status. Blue represents alive individuals, and red represents deceased patients. (C) Overall probability of survival in the low risk as well as high risk LUAD groups. Blue is the LRG, while red is the HRG. (D) 1-, 3-, and 5 survival rates (SRs) in LUAD patients. Red is 1-year, green denotes 3-year, as well as blue signifies 5-year SRs. (E) Expression of GAB3 and IL16 in LUAD samples from HRG as well as LRG. Values closer to −1 mean lower expression, while those closer to 1 signify higher one.

Full size image

Significant association between clinical features and RS

The Wilcoxon test of risk score (RS) by sex, age, T stage (TS), N stage (NS), M stage (MS), and overall stage revealed significant differences for gender, stage (i-iii), TS (T1-T3), and NS (N2/N3-N0) (Fig. 5A-F). The violin plots of IL-16 and T, N, M, and M stages showed significant differences among the subgroups of T, N, and total stages (Fig. 5G). Moreover, AUC values of the ROC curves combining six clinical characteristics and RS were approximately 0.6 for 1, 2, and 3 years. Stage demonstrated superior predictive capability, with AUC values exceeding 0.6 for 1-, 2-, and 3-year survival (Fig. 5H).

Fig. 5
figure 5

Association of Clinical Characteristics with Risk Scores. (A-F) Wilcoxon plots of risk scores versus sex, age, TS, NS, MS, as well as overall staging. (G) Difference analysis between clinical characteristics and IL-16. (H) AUC values at 1, 2, as well as 3 years for ROC curves representing six combinations of clinical characteristics as well as RS.

Full size image

A nomogram integrating RS, TS, NS, and MS effectively predicted survival in patients with LUAD

RS, TS, NS, and MS were shown to be independent predictive factors (p < 0.05) (Fig. 6A-B). Therefore, a nomogram model was generated to forecast the likelihood of LUAD patient survival at 1, 3, as well as 5 years, incorporating RS, TS, NS, MS, and total points,.with a higher total points corresponding to a decreased likelihood of patient survival (Fig. 6C). The high precision of the nomogram in calculating LUAD patient survival was demonstrated through calibration curves (Fig. 6D).

Fig. 6
figure 6

Nomogram integrating risk scores, TS, NS, as well as MS effectively predicted survival in LUAD patients. (A-B) Column plot showing the predictive power of RS, TS, NS, as well as MS on survival in LUAD patients. (C) RS, TS, NS, MS, as well as total score predict the probability of survival at 1, 3, and 5 years in LUAD patients. (D) Calibration curves confirming the high accuracy of the column line plots.

Full size image

Transcriptional and immune-related functional items were activated in the two risk groups, respectively

GSEA analysis revealed that the HRG showed ribosome activation, while the LRG showed activation of the intestinal immune network (IIN) for IgA production, graft-versus-host disease, autoimmune thyroid disease, as well as allograft rejection (Fig. 7A). GSEA was also employed to explore the potential pathways in which prognostic genes might be involved. The results demonstrated that GAB3 and IL16 were enriched in pathways such as ribosome, oxidative phosphorylation, spliceosome, and hematopoietic cell lineage (Fig. 7B, C). These pathways provide essential protein molecules and cellular carriers for STING-mediated immune responses. Moreover, GSVA results indicated remarkable variations in 20 pathways between the two groups. In the HRG, aminoacyl-tRNA biosynthesis, RNA polymerase, base excision repair, and pyrimidine metabolism were activated. Conversely, the LRG demonstrated activation of the IIN for IgA production and hematopoietic cell lineage (Fig. 7D). In summary, transcriptional and immune-related pathways were activated in the respective risk groups. Notably, the two risk groups exhibited significant positive and negative correlations with IIN for IgA production (Fig. 7E).

Fig. 7
figure 7

Transcriptional and immune-related functional programs in in LUAD. (A) Gene Set Enrichment Analysis (GSEA) results. (B, C) GSEA showed that GAB3 and IL16 were enriched in pathways such as ribosome, oxidative phosphorylation, spliceosome, and hematopoietic cell lineage. (D) Pathways showing significant differences between the two groups. (E) Relationship between immune-related and transcriptional processes in the two risk categories.

Full size image

Analysis of the immune landscape revealed a worse prognosis for high-risk LUAD

The ssGSEA algorithm identified 16 differentially infiltrating immune cells (ICs) between the two risk groups. All 16 cell types were heavily infiltrated in the LRG, including CD8 + T cells, activated DCs, as well as B cells (Fig. 8A). Additionally, 13 immune function enrichment scores were remarkably higher in LRG, revealing that high risk LUAD had diminished immune functions (Fig. 8B). Correlation analysis between prognostic genes and immune cells revealed that IL-16 was positively correlated with CD4 + T cells and CD8 + T cells, indicating its critical role in anti-tumor immunity (Fig. 8C). In the training dataset, 39 immunological checkpoints were expressed, with 36 showing differential expression between the two groups. TMIGD2 expression was higher in the HRG, while the remaining 35 checkpoints were overexpressed in the LRG, suggesting that blocking TMIGD2 expression may benefit high-risk LUAD patients (Fig. 8D). Additionally, our findings revealed an inverse correlation between immunity score, stromal score, ESTIMATE score, and RS (Fig. 8E-H).

Fig. 8
figure 8

Analysis of immunization status. (A) Expression levels of 16 cell types in HRG as well as LRG. (B) Expression of 13 immune function enrichment scores in HRG as well as LRG. (C) The correlation between IL-16 and immune cells. (D) Expression levels of 36 immune checkpoints in HRG as well as LRG. (E) Immunity Score, Stromal Score, and ESTIMATE Score in HRG as well as LRG. (F-H) Association between Immunity Score, Stromal Score, ESTIMATE Score, and RS. “ns” indicates non-significant results.

Full size image

The correlation between prognostic genes and the microenvironment of the LUAD

Given the prognostic significance of the LUAD microenvironment in tumor development and progression, we analyzed the expression of GAB3 and IL16 in TME-associated cells using 16 datasets from the TISCH database. GAB3 expression was found to be high in CD4Tconv, CD8T, CD8Tex, as well as Tprolif cells in the NSCLC-GSE99254 and NSCLC-GSE151537 datasets. IL16 was highly expressed in CD4Tconv, CD8T, CD8Tex, Tprolif, as well as Treg cells in NSCLC-GSE99254 and NSCLC-GSE151537 datasets (Fig. 9A-B). We further analyzed the GSE99254 dataset, which was classified into six cell types. The number of each cell type, ranked in order, was CD4Tconv, CD8T, CD8Tex, Treg, Tprolif, and Mono/Macro (Fig. 9C). IL16 expression was higher in TME-associated cells compared to GAB3. These findings support the conclusion that the prognostic genes are closely related to LUAD (Fig. 9D).

Fig. 9
figure 9

The association between prognostic genes and the microenvironment of the LUAD. (A-B) Expression levels of 18 cell types in 16 datasets. (C) Proportions of six different cell types. (D) Correlation between GAB3, IL16, and the tumor microenvironment (TME).

Full size image

Immunotherapy response that performed effectively in low-risk LUAD

The TIDE score, Exclusion score, as well as Dysfunction score was obtained via TIDE algorithm to assess immunotherapy performance in each LUAD patient across different groups. The results showed that the Dysfunction score was remarkably elevated in LRG, while the Exclusion score and TIDE score were substantially higher in HRG, suggesting high risk LUAD patients were more likely to exhibit immune escape. The response to immunotherapy in the two risk groups was 153:128 and 173:61, respectively, further demonstrating that immunotherapy was more effective for low-risk LUAD patients (Fig. 10A-B).

Fig. 10
figure 10

Immunotherapeutic response in LUAD. (A-B) TIDE Score, Exclusion Score, and Dysfunction Score used to evaluate immunotherapy outcomes in LUAD patients from HRG as well as LRG. (C) Immune checkpoint blockade (ICB) analysis plots for both risk groups. (D-E) Association between immunotherapy pathways, RS, and immune cycles.

Full size image

Additionally, analysis of immune checkpoint blockade (ICB) indicated remarkable variations between the two risk groups in their immunotherapeutic response to CTLA-4 as well as PD-1 in cytotoxic T lymphocytes. Scores for IPS, IPS-CTLA4 blocker, IPS-CTLA4 as well as PD1/PDL1/PDL2 blocker, along with IPS-PD1/PDL1/PDL2 blocker were all higher in LRG. This finding supports the efficacy of ICB targeting CTLA-4 and PD-1 in preventing cancer cells from evading the immune system in low-risk LUAD patients, thus contributing to their treatment (Fig. 10C).

Moreover, understanding TME heterogeneity could improve the precision of tumor immunotherapy. Hence, this study examined the correlations between immunotherapy pathways, RS, and the immune cycle. Most signaling pathways and RS showed significant positive correlations, with the IFN-γ signature exhibiting the highest correlation with the RS (r = 0.330, p = 0.001). A strong positive association was observed between the RS and three specific stages in the seven-step cancer immune cycle: cancer antigen presentation (step 2), induction and activation (step 3), as well as migration of ICs to the tumor (step 4). The correlation coefficients for these three stages were 0.450, 0.528, and 0.432, respectively (p = 0.001) (Fig. 10D-E).

Through the findings, we speculate low-risk LUAD patients have an activated immune response, making ICB more effective, while high-risk LUAD patients have a suppressed immune response, increasing the likelihood of cancer cell escape. However, as the RS increases, most immunotherapeutic pathways and immune cycles become activated, enabling sustained immunotherapy in high-risk patients (HRPs).

LUAD patients in the HRG as well as low-TMB group suffered from lower survival rates

Based on the combination of RS and TMB, LUAD patients were classified into four groups: High-TMB + High-risk, High-TMB + Low-risk, Low-TMB + High-risk, as well as Low-TMB + Low-risk. Patients with somatic hypermutations showed improved survival rates (Fig. 11A). Correlation analysis further supported a positive association between RS and TMB (r = 0.19, p < 0.001) (Fig. 11B). A waterfall plot illustrated somatic mutations in patients across both risk groups, showing that somatic mutations were concentrated in TTN, TP53, CSMD3, MUC16, and PYR2. TTN showed a higher mutation frequency in HRPs, whereas TP53 showed a more consistent mutation pattern in low-risk LUAD patients. Overall, high-risk LUAD patients showed a higher likelihood of harboring mutations (Fig. 11C-D).

Fig. 11
figure 11

Survival of LUAD patients in the HRG and the high/low TMB groups. (A) Survival rates for high TMB + high risk, high TMB + low risk, low TMB + high risk, as well as low TMB + low risk groups. (B) Correlation between RS and TMB levels. (C-D) Somatic mutation waterfall plots for patients in the two LUAD risk groups.

Full size image

The high-risk LUAD patients displayed sensitivity to most chemotherapy treatments

Chemotherapy and targeted therapy are essential treatments for LUAD. The results suggested low-risk LUAD patients were more susceptible to several drugs, including Axitinib, Gefitinib, as well as Lenalidomide. Docetaxel, Paclitaxel, Vinorelbine, Erlotinib, and Lapatinib can be more effective in treating high-risk LUAD patients (Fig. 12A). Correlation analyses revealed that most anticancer drugs (e.g., Docetaxel and Lapatinib) were positively correlated with IL16 and GAB3, whereas Gefitinib and Lenalidomide exhibited a negative correlation with IL16 and GAB3 (Fig. 12B-C). These findings suggest that our prognostic model could serve as a valuable tool for predicting the responsiveness of chemotherapy drugs in LUAD patients.

Fig. 12
figure 12

Sensitivity to chemotherapeutic agents in the two risk groups. (A) Comparison of drug IC50 between the high- and low-risk groups. (B) Correlation between IL16 and drugs. (C) Correlation between GAB3 and drugs.

Full size image

Expression of prognostic genes

In LUAD samples from three surgically resected individuals, GAB3 protein expression was lower than in nearby normal tissues (Fig. 13A). Furthermore, GAB3 overexpression (GAB3-ov) appeared to inhibit the proliferation and migration of A549 as well as NCI-H23 cells. Conversely, GAB3 siRNA interference (siGAB3) promoted the proliferation and migration of A549 and NCI-H23 cells (Fig. 13B-D). The validation results have been further enhanced from bioinformatics analysis and cellular function experiments, suggesting that GAB3 may function as a tumor suppressor gene.

Fig. 13
figure 13

Validation of GAB3 as a tumor suppressor in LUAD. (A) Western blot (WB) of GAB3 expression in paired LUAD tumor (T) and adjacent normal (N) tissues. Bar chart shows relative quantification. (B) WB validation of GAB3 overexpression (GAB3-OV) and knockdown (siGAB3) efficiency in A549 and NCI-H23 cells. GAPDH served as the loading control. (C) CCK-8 proliferation assays (0, 24, 48 h, 72 h) in A549 and NCI-H23 cells. GAB3-OV inhibited proliferation, while siGAB3 promoted it. (D) Transwell migration assays in A549 and NCI-H23 cells. GAB3-OV reduced migration, while siGAB3 increased it. Representative images (left) and quantification (right) are shown. (E) Wound healing assays in A549 cells (0, 24, 48 h). GAB3-OV delayed wound closure, while siGAB3 accelerated it. Representative images (left) and migration area quantification (right) are shown. *Data represent mean ± SD (n = 3).

Full size image

Discussion

STING affects cancer cell proliferation and invasion, and is closely linked to LUAD tumor progression through the cGAS-STING pathway, primarily by modulating the tumor immune microenvironment11,13,14,15. However, there are limited comprehensive studies on investigating the role of STING-related genes (STING-RGs) in LUAD progression, and no definitive conclusions have been reached. In this study, we identified 38 DE-STING-RGs with prognostic significance in LUAD. These genes were associated with LUAD development and were all linked to poor prognostic outcomes. Functional enrichment analysis revealed that these 38 DE-STING-RGs were involved in the biosynthesis and metabolism of various amino acids, including tryptophan and branched-chain amino acids (like alanine, leucine, as well as isoleucine), as well as fatty acid metabolism, specifically oxidation and degradation. It is well established that there are distinct differences in carbohydrate, amino acid, as well as lipid metabolism between tumor cells (TCs) and normal cells (NCs)30. Amino acid metabolism and its derivatives have been reported as key factors in the progression of LUAD2. Zheng Wang et al. demonstrated that tryptophan metabolism is associated with tumor immunosuppression and immune escape, and that abnormal tryptophan metabolism can worsen the prognosis of LUAD patients36. Similarly, Kayo Ikeda et al. suggested that branched-chain amino acids promote tumor proliferation and stimulate tumor growth, which is associated with alterations in the tumor immune microenvironment37,38. Furthermore, synthetic lipids are utilized by cancer cells as an energy source to drive angiogenesis, invasion, growth, and survival39. According to Suyu Wang et al., fatty acid metabolism is essential for LUAD carcinogenesis, progression, and immune modulation40,41. Our results agreed with those of these earlier investigations.

Based on the 38 DE-STING-RGs, we identified two prognostic genes in LUAD: GAB3 and IL16. Consistent with our findings, expression levels of both genes were significantly lower in high-risk samples compared to low-risk samples. GAB3, or Grb2-associated binding protein 3, is an adaptor protein belonging to the Grb2-associated binders’ family of multi-substrate docking proteins. These proteins resemble insulin receptor substrate 1 and can bind phosphatidylinositol lipids in biological membranes42,43. Grb2-associated binders facilitate immune cell activation by forming signalosomes and interacting with other signaling partners, acting as both positive and negative regulators44. Specifically, GAB3 is a critical mediator of cytokine-driven NK-cell expansion and antitumor immune responses45. In addition, PI3Kδ rapidly cooperates with multiple adaptor proteins, including GAB3, during CD4⁺ T-cell activation, thereby playing an essential role in adaptive immune signaling46,47. It may also promote tumor growth in cancers such as ovarian cancer and gliomas48,49. Lei Zhang et al. demonstrated that the LINC00324/miR-9-5p (miR-33b-5p)/GAB3 regulatory axis is critical in controlling tumor-associated macrophage risk as well as prognosis in LUAD patients50,51. In the present study, By analysing samples from three surgically resected LUAD patients, GAB3 protein expression was found to be lower in tumour tissues than in adjacent normal tissues. Further in vitro experiments showed that GAB3 overexpression significantly inhibited the proliferation and migration of lung cancer cells, while GAB3 siRNA interference promoted the proliferation and migration of these cells. These results suggest that GAB3 may function as a tumor suppressor gene in LUAD. The in vitro functional validation not only further supports our conclusions drawn from the bioinformatics analysis, but also highlights the significance of GAB3 as a potential therapeutic target.

Interleukin-16 (IL16) is a pro-inflammatory cytokine and chemoattractant for various immune cells52,53. Lifeng Li et al. identified IL16 as a prognostic marker for LUAD and reported that IL16 expression levels in low-risk samples were remarkably higher than in high-risk samples, which agrees with our findings54,55. IL16 in LUAD has been implicated in immune tolerance in non–small cell lung cancer (NSCLC)56,57.IL16 also modulates T cell function and induces T cell immune responses, while IL16 deficiency can promote dendritic cell (DC) maturation58. Interestingly, Jianjun Wu et al. showed that STING mutations regulate T cell function by inducing ER stress, thereby alleviating lung disease15. Notably, STING activation promotes the synthesis of IFNs, which induces dendritic cell maturation, enhances the adaptive immune response, and suppresses tumor growth12. Evidence suggests that the STING pathway may be associated with IL16 in inflammation-related pathological processes59. Therefore, we speculate that GAB3 and IL16 may influence LUAD progression by regulating T cell activation and immune response. Specifically, GAB3 was linked to macrophage activity, while IL16 was associated with dendritic cell maturation.

STING is widely expressed in non-small cell lung cancer (NSCLC) and is associated with T-cell functional gene signatures, contributing to improved overall survival. T-cell responses and the composition of the immune microenvironment are key determinants of therapeutic efficacy21. Therefore, a RS model according to the two prognostic genes was constructed, as well as we found that LUAD risk was associated with multiple immune-related pathways, like B cell receptor, T cell receptor, chemokine, as well as JAK-STAT signaling pathways (SPs). The results agreed with Sijin Sun et al.60,61. The JAK-STAT pathway, which is controlled by cytokines, is critical in the innate immunity initiation, inflammation suppression, as well as regulation of immune responses, and it is also important in LUAD62,63. Anqi Lin et al. identified the B cell receptor signaling pathway (SP) as a possible biomarker for predicting the effectiveness of immune checkpoint inhibitor therapy in LUAD patients64. Notably, B cells activated through the B cell receptor SP can lead to the activation of tumor-specific CD8 + T cells, elevating anti-tumor immunity65. Similarly, chemokines and the T cell receptor signaling pathway regulate immune cells within tumors, thereby influencing the therapeutic response and prognosis in LUAD patients66,67,68. Additionally, we found that activated DCs, B cells, as well as CD8 + T cells were linked to LUAD risk. As previously mentioned, GAB3 and IL16 link to T cell activation and immune response, with IL16 being linked to DC maturation. Single-cell localization results also showed that GAB3 and IL16 were primarily distributed in CD8 + and CD4 + T cells. Studies have shown that the expression of STING in small-cell lung cancer (SCLC) carries prognostic significance in immune-infiltrated tumors. Increased STING expression is associated with higher densities of CD45⁺ and CD8⁺ immune cells in both tumor compartments69. These findings further support our hypothesis. Notably, IFN responses differed significantly between the risk groups, further confirming that IL16 promotes IFN synthesis and subsequently induces DC maturation. ESTIMATE analysis suggested that tumor purity was higher in the HRG, implying a greater inclination for tumor metastasis, consistent with the recurrent nature and poor prognosis of LUAD6. A prospective cohort study could be designed to include patients with lung adenocarcinoma (LUAD) who are scheduled to receive immune checkpoint inhibitor (ICI) therapy. The expression levels of GAB3 and IL16 in their tumor tissues would be assessed, followed by long-term monitoring of objective response rate, progression-free survival, and overall survival.

Additionally, we found that Docetaxel, Paclitaxel, Vinorelbine, Erlotinib, and Lapatinib were potentially more effective in treating high-risk LUAD patients, while Axitinib, Gefitinib, and Lenalidomide showed greater sensitivity in low-risk patients. Furthermore, correlation analyses revealed that Docetaxel and Lapatinib were positively correlated with IL16 and GAB3, whereas Gefitinib and Lenalidomide exhibited a negative correlation with IL16 and GAB3. These drugs have been widely reported in the treatment response of LUAD. The downregulated expression of GAB3 protein in clinical tumor samples, as well as its functional validation through in vitro experiments, further elucidates the tumor-suppressive role of GAB3 in LUAD and underscores its potential as a therapeutic target. Future early-phase clinical trials of candidate therapeutics could be conducted by first validating, in vitro, the effects of the drugs on LUAD cell proliferation and STING pathway activation, followed by clinical evaluations to assess their safety and preliminary efficacy.

In summary, this study, through bioinformatics analysis and experimental validation, identified two STING-related prognostic genes, GAB3 and IL16, which may influence LUAD progression by regulating T cell activation and immune responses. Moreover, we found that LUAD risk is associated with multiple immune-related pathways, with extensive roles played by T cells and dendritic cells. However, the precise influence of these prognostic genes on the mechanisms underlying LUAD progression remains to be confirmed, necessitating further in vivo and in vitro studies to provide more robust scientific evidence and support clinical interventions.

Future studies will expand the sample size to further validate the protein expression patterns of GAB3 and IL16, thereby improving the generalizability of our conclusions. We also plan to collect larger independent clinical cohorts to further validate and refine the proposed risk model, enhancing its predictive stability and clinical applicability across diverse populations. To address the current limitation in the number of LUAD cell lines, future work will incorporate additional LUAD models (such as PC9 and H1975) to cover a broader spectrum of LUAD driver mutation subtypes, enabling analyses of the functional differences of GAB3 under various oncogenic backgrounds. Moreover, the current evidence supporting the immune regulatory mechanisms is primarily indirect; direct indicators such as T-cell cytotoxic activity and dendritic cell maturation markers (e.g., CD80 and CD86) have not yet been examined. Future experiments will focus on these directions to more comprehensively validate the roles of GAB3 and IL16 in modulating immune responses in LUAD. In addition, high- and low-risk LUAD cellular models will be constructed to conduct in-vitro drug sensitivity assays for the predicted therapeutic agents, providing empirical confirmation of differential drug responses between the two groups and supporting the translational value of our findings. Finally, the present study only infers a potential association between STING-related genes and fatty acid metabolism based on existing datasets, and has not yet validated this relationship through direct measurements such as lipid content or metabolic enzyme activity in LUAD cells or tissues.