Gene expression data, clinical survival information, and gene mutation data for patients diagnosed with LUAD were acquired from TCGA database (TCGA-LUAD) and the Gene Expression Omnibus (GEO) database (accession numbers: GSE31210, GSE115002). Specifically, the dataset included 526 LUAD tissue samples and 59 normal lung tissue samples from the TCGA database, 226 LUAD tissue samples and 20 normal lung tissue samples from GSE31210, and 52 LUAD tissue samples and 52 normal lung tissue samples from GSE115002.

Normalization of gene expression data and identification of differentially expressed genes were performed using the R limma package (Version 3.18) (22). The criteria for selecting significant DEGs were set at a p-value of < 0.05 and a logFC of > |1|.

We constructed a PPI network of proteins expressed by DEGs using the STRING database (https://string-db.org/). To identify the most significant modules within the PPI network, we employed the Cytoscape plugin molecular complex detection (MCODE), which is based on graph-theoretic clustering algorithms (23). Selection criteria were established as K-core = 2, degree cutoff = 10, max depth = 100, and node score cutoff = 0.2. Key genes within the PPI network, showing high connectivity and substantial prognostic significance, were determined (24). Mutation data of LUAD patients were obtained from the TCGA database. The Human Protein Atlas (HPA) database (https://www.proteinatlas.org/) was used to validate the protein expression of these key genes in LUAD and normal tissues (25). The images were available from version 23.0.

Building on the collective impact of these ten genes in LUAD progression, we developed risk score signatures to comprehensively examine their role in patient prognosis, immune cell infiltration, and immunotherapeutic response. The risk score signature was formulated using the most robust markers, identified through LASSO Cox regression (26). The optimal penalty parameter lambda was determined through 10-fold cross-validation.

The LUAD tissues and the adjacent paracancerous tissues were obtained from the Department of Pathology, Second Affiliated Hospital, School of Medicine, Zhejiang University. The present study adhered to the ethical principles outlined in The Declaration of Helsinki. The study’s inclusion criteria consisted of the following: i) presence of a primary lung adenocarcinoma; ii) histopathological verification of lung adenocarcinoma diagnosis; and iii) availability of complete clinical records. Conversely, the exclusion criteria comprised: i) presence of autoimmune disorders or other diseases; ii) presence of other severe illnesses; and iii) history of prior immunosuppressive treatments. Study procedures were approved by the Ethics Committee of the Second Affiliated Hospital of Zhejiang University School of Medicine, Hangzhou, China (approval no: 2023-0970). Patient follow-up started in June 2020 and continued until November 2023. Since the present study was retrospective and non-interventional in nature, the requirement to obtain patient informed consent for participation was waived by the Ethics Committee. A total of 93 patients diagnosed with invasive lung adenocarcinoma were selected for this study. These patients had previously undergone surgical resection at the Second Affiliated Hospital of Zhejiang University School of Medicine. The samples of LUAD and their corresponding medical information were collected.

Total RNA was extracted from lung adenocarcinoma tissues of patients following the manufacturer’s instructions using a kit from XiaMen AmoyDx. Reverse transcription was carried out with PrimeScript RT Master Mix at 37°C for 15 minutes followed by 85°C for 5 seconds (Takara). Takara Biotechnology TB Green Premix Ex Taq™ II (cat. no. RR820A) was employed for qPCR, with the following thermocycling conditions: initial denaturation at 95°C for 30 seconds, followed by 40 cycles of 95°C for 5 seconds and 60°C for 30 seconds, and a final extension at 95°C for 15 seconds and 60°C for 60 seconds. The relative expression levels of ten genes were quantified employing the 2−ΔΔCq method (27), with β-actin serving as an internal control (Takara). The primers were designed by Qingke Biotechnology Co. The primers used in this study are shown in the Supplementary Table S1 .

GSEA was performed using the clusterProfiler package in R, where gene sets with normalized enrichment scores (NES) exceeding 1 and false discovery rates (FDR) below 0.05 were considered significant.

We conducted an analysis of immune cell infiltration using the ssGSEA algorithm (28). We compared distinct gene sets characteristic of immune cells and transformed sample gene expression values into enrichment fractions. This approach enabled us to derive the relative abundance of immune cells using the GSVA R package (Version 1.34).

γH2AX expression level in A549 cell were evaluated using Western blot analysis. Initially, cells were lysed with lysis buffer, and protein concentrations in the lysates were measured by the BCA assay (Beyotime Institute of Biotechnology, Shanghai, China). Subsequently, 10 µg of each lysate was separated on 10% SDS-PAGE and transferred to PVDF membranes (EMD Millipore). To reduce non-specific binding, membranes were blocked using 5% non-fat dry milk in TBST. Immunoblotting was performed using an anti-γH2AX antibody (1:1,000; HUABIO), an anti-EXO1 antibody (1:1,000; HUABIO), and an HRP-conjugated anti-rabbit secondary antibody (1:5,000; Santa Cruz Biotechnology, Inc., CA, USA). Blots were developed with ECL reagent (Beyotime Institute of Biotechnology) and imaged with the ChemiDoc MP Imaging System (Bio-Rad Laboratories, Inc.).

Tissue sections (2.5µm) mounted on glass microscope slides were deparaffinized in xylene and rehydrated in graded alcohols. Then the tissue sections were boiled for 15 min in 10 mM citric acid buffer (pH 6.0) for antigen retrieval. After the sections were cooled to room temperature (RT), they were incubated with 3% H₂O₂ for 15 min at 37°C to block endogenous peroxidase activity. Then the sections were blocked with 5% goat serum (Scientific Phygene) for 45 min at room temperature, and incubated with primary antibodies, including EXO1 (1:300; HuaBio), MKI67 (1:1,000; Zhongshan Golden Bridge Biotechnology Co), TTF1 (1:1,000; Zhongshan Golden Bridge Biotechnology Co), and NapsinA (1:1,000; Zhongshan Golden Bridge Biotechnology Co), overnight at 4°C. Subsequently, the sections were incubated with secondary antibodies, including goat anti-rabbit IgG antibody (1:1,000; Abcam), for 30 min. Finally, the substrate color was developed using a diaminobenzidine substrate kit (Abcam), and the sections were counterstained with hematoxylin for 30s. Image acquisition was performed using a light microscope (Nikon Corporation). Immunohistochemistry results were quantified using ImageJ software (version 1.8.0).

The A549 and H322 cell lines were sourced from the American Type Culture Collection (ATCC) in Maryland, USA. A549 and H322 cells were cultured in DMEM/F12 medium (Gibco, California, USA) supplemented with 10% fetal bovine serum (Lonsera, Australia) and 1% penicillin/streptomycin (Beyotime Biotechnology, Jiangsu, China).

EXO1-shRNA1 and shRNA2 were subcloned into EcoRI and AgeI digested pLKO.1vector. All generated plasmids were verified by DNA sequencing (Qingke Biotechnology Co.).

The production of lentiviruses and viral infections of cells were performed previously as described (29). To generate recombinant lentivirus, 293T-17 cells were co-transfected with three plasmids—pLEX-CMV-EXO1, pSPAX2, and pMD2G (SBI System Biosciences, Palo Alto, CA, USA)—using polyethyleneimine (PEI) at a 4:3:1 ratio. Lentiviral supernatants were collected 48 and 72 hours post-transfection, centrifuged at 3,000 rpm for 10 minutes, and filtered through a 0.45 µm membrane before storage at -80°C. Target cells were then infected with EXO1-recombinant lentivirus in the presence of 8 µg/ml polybrene and subsequently selected with 2.5 µg/ml puromycin (Invitrogen, San Diego, USA) for four days.

For the wound-healing assay, cells were seeded in 24-well plates. Cells were scratched with a sterile tip perpendicular to the previously painted line. Scratch wounds were imaged using a light microscope after which cell migration was measured at indicated time points of 0 and 48 h. All experiments were repeated three times.

Cells were collected by centrifugation and resuspended in the medium at a density of 10⁴ cells per milliliter of medium. Then, 200 μl of culture medium was added to each well of a 96-well culture plate. Cell proliferation assays were conducted using CCK-8 (Beyotime Biotechnology, Shanghai) and measured at 0, 24, 48, and 72 hours following the manufacturer’s instructions. The optical density of the cells at 450 nm (OD450) was measured and recorded using GraphPad Prism software to assess their proliferative ability. All experiments were repeated three times.

Statistical and bioinformatics analyses were conducted using R software (version 4.2.0). One-way analysis of variance (ANOVA) was used to compare multiple groups, followed by Tukey’s post hoc test for pairwise comparisons. For comparisons between the two groups, we employed the Wilcoxon rank-sum test, while correlation analyses were based on Spearman and distance correlation methods. The Kaplan-Meier method was used for the survival curve data. The survival period was defined as the time from the start of treatment to the end of the observation.

Figure 1 presents a flow diagram elucidating the study’s procedural framework. In a comprehensive scope, our investigation involved 59 normal samples and 526 tumor samples from TCGA, 20 normal samples and 226 tumor samples from GSE31210, and 52 normal samples and 52 tumor samples from GSE115002. Systematic clustering clearly delineated genomic disparities between normal and LUAD tissues ( Figure 2A ; Supplementary Figures S1A , S2A ). Principal Component Analysis (PCA) underscored a marked separation between the control and tumor groups, with minimal overlap between them ( Figure 2B ; Supplementary Figures S1B , S2B ). The volcano plot of differentially expressed genes highlighted up-regulated and down-regulated genes in LUAD tissues compared to normal tissues ( Figure 2C ; Supplementary Figures S1C , S2C ). The Venn diagram demonstrates the presence of 425 differentially expressed genes across the three datasets ( Figure 2D ).

We established a PPI network to investigate the interrelationships among the protein-encoding DEGs associated with LUAD. Among these DEGs, the top ten genes, namely PBK, ASPM, NCAPG, EXO1, MKI67, RRM2, AURKA, DLGAP5, UBE2C, and CDC6, were identified as key molecules due to their pronounced connectivity ( Figure 3A ). It is noteworthy that the expressions of these key molecules were significantly elevated in tumor samples ( Figures 3B, C ). Principal Component Analysis (PCA) revealed two distinct populations, highlighting a significant difference in key molecule expression between normal samples and LUAD tumors ( Figure 3D ).

Immunohistochemistry staining results for eight key genes were obtained from the Human Protein Atlas (HPA) database. In contrast to the normal group, the protein expression levels of PBK, CDC6, DLGAP5, AURKA, RRM2, MKI67, UBE2C, and NCAPG were substantially increased in the tumor group ( Figure 4A ). We performed a Spearman correlation analysis to examine the mRNA expression levels of 10 key molecules. The analysis identified a significant positive correlation among these molecules in LUAD ( Figure 4B ). Subsequently, we explored the mutational landscape of these genes in LUAD. Out of the 567 samples assessed, approximately 20.28% of patients exhibited mutations in at least one key molecule. ASPM exhibited the highest mutation frequency, closely followed by MKI67 ( Figure 4C ). Furthermore, LUAD patients with lower expressions of PBK, ASPM, NCAPG, EXO1, MKI67, RRM2, AURKA, DLGAP5, UBE2C, and CDC6 enjoyed a significant survival advantage compared to those with elevated expression ( Figure 5A ). These findings underscored the evident disparities in the expression of selected target genes between normal and tumor tissues. Moreover, the expression of these key molecules within LUAD tissues was markedly increased in comparison to normal tissues ( Figure 5B ).

To further explore the clinical significance of the regulatory network orchestrated by the 10 key genes, a risk score signature was expertly fashioned to consolidate the contributions of these 10 pivotal genes, using the LASSO Cox regression model ( Figures 6A, B ). The findings emphasized that the risk score depended on the expression of the four most influential key molecules: DLGAP5, EXO1, RRM2, and PBK. Manifestly, as the risk value increased, a consequential escalation in patient mortality was evident ( Figure 6C ). Revealing the outcome of KM survival analysis, it was distinctly discerned that individuals classified in the low-risk group enjoyed superior survival advantages ( Figure 6D ).

To investigate the variations in the mRNA expression of key moleculars between normal and LUAD tissues, we examined the expression levels of PBK, ASPM, NCAPG, EXO1, MKI67, RRM2, AURKA, DLGAP5, UBE2C, and CDC6 in tissues by using qRT-PCR. The findings indicate that crucial molecules showed significantly higher expression levels in LUAD tissues compared to the adjacent paracancerous tissues ( Figure 7 ).

GSEA demonstrated that the expression of EXO1 is significantly associated with several key molecular mechanisms, which include the essential pathways of PLK1, FOXM1, ATR, E2F, AURORA_B, and FANCONI, in addition to the critical Notch signaling and ESR-mediated signaling pathways ( Figures 8A, B, D ). These compelling findings reveal the intricate potential molecular mechanisms through which EXO1 may promote both cell growth and migration in lung adenocarcinoma (LUAD), suggesting a multifaceted role in the progression of this disease. In addition, the top 10 genes that were positively and negatively linked with EXO1 were exhibited on heat maps ( Figures 8C, E ).

This study further explores the relationship between EXO1 expression and immune infiltration in LUAD patients. We assessed immune-related infiltration to determine the Pearson correlation between the immune microenvironment and EXO1 expression, using ssGSEA algorithms. Our analysis revealed a positive correlation of EXO1 expression with immune infiltration levels from Th2 cells, Tgd, NK CD56dim cells, and T helper cells ( Figure 9A ). Conversely, EXO1 expression showed a negative correlation with Th1 cells, cytotoxic cells, NK CD56bright cells, macrophages, T cells, NK cells, pDC, B cells, CD8 T cells, Th17 cells, DC, TFH cells, eosinophils, iDC, and mast cells ( Figures 9A–M ). Additional analysis highlighted significant differences in EXO1 gene expression levels among various immune cell types among others ( Figures 9N, O ). This research also investigated different functional subsets of T cells, suggesting that EXO1 may play a key role in the immune-inflamed microenvironment of LUAD.

The relationship between EXO1 expression and clinical parameters was evaluated using the Wilcoxon signed-rank tests. Elevated levels of EXO1 expression were associated with higher T stage, N stage, M stage, gender, and smoking status, as well as the occurrence of OS events ( Figure 10A ). Correlations between the clinicopathological characteristics of LUAD patients and their EXO1 protein levels are presented in Table 1 . Patients with high EXO1 expression demonstrated poorer survival outcomes. These findings provide evidence supporting an association between increased EXO1 expression and advanced tumor pathological stage, thus highlighting the crucial role of EXO1 in LUAD prognosis. To advance our investigation, we employed univariate Cox regression model analysis. The subsequent findings confirmed that EXO1 characteristics could be confidently considered as reliable and independent prognostic biomarkers, significantly enhancing the accuracy of prognostic assessment for LUAD patients ( Figure 10B ). We devised a clinical prognostic risk score for LUAD, integrating T stage, N stage, M stage, clinical stage, histological grade, age, and EXO1 expression ( Figure 10C ). Furthermore, we employed a calibration chart to assess the model’s predictive accuracy ( Figure 10D ). EXO1 expression holds promise for enhancing the accuracy of predicting patients’ survival probabilities for 3- and 5-year intervals. Overall, a clear correlation between EXO1 expression and LUAD patient prognosis was evident.

We conducted an investigation into the variations in EXO1 protein expression between cancer-adjacent tissues and LUAD tissues by evaluating the expression levels of EXO1 and its associated proteins through immunohistochemistry ( Figure 11A ). The results revealed significantly elevated expression levels of EXO1, TTF1, MKI67, and NapsinA in LUAD tissues compared to adjacent paracancerous tissues. We evaluated the correlation between EXO1 expression and clinical parameters. Increased EXO1 expression levels were correlated with lymph node metastasis, pleural invasion, poor tumor differentiation, and advanced clinical stage ( Figures 11B–E ).

To assess the roles of EXO1 in LUAD, various in vitro assays were performed. As EXO1 is a promising prognostic biomarker for LUAD, we investigated its role in LUAD progression. qPCR analysis revealed that knockdown of EXO1 led to a decrease in the mRNA levels of EXO1 in A549 and H322 cells ( Figures 12A, B ). Wound healing assays showed that the metastatic ability of A549 and H322 cells was decreased after knocking down EXO1 ( Figures 12C, D, F, G ). The results of Cell Counting Kit-8 (CCK-8) showed that the proliferative ability of A549 and H322 cells was effectively weakened by knocking down EXO1 ( Figures 12E, H ). Knockdown of EXO1 inhibited the growth and migration of LUAD cells, implying that EXO1 promotes the LUAD progression.

To further investigate the role of EXO1 in the DNA damage response and repair processes, our results reveal a significant positive correlation between EXO1 expression and the DNA damage marker H2AX, as well as the repair proteins BRCA1 and BRCA2, in the TCGA database ( Figures 13A–E ). H2AX, an important DNA damage response protein, rapidly accumulates in its phosphorylated form (γH2AX) following DNA double-strand breaks, serving as a key marker of DNA damage. Our experiments show that knockdown of EXO1 leads to a marked decrease in γH2AX expression, suggesting that EXO1 plays a critical role in maintaining DNA damage signaling ( Figure 13F ). Furthermore, BRCA1 and BRCA2, key DNA repair proteins involved in homologous recombination, exhibited significantly reduced mRNA expression levels following EXO1 knockdown, as confirmed by qPCR and consistent with bioinformatic analyses from the TCGA database ( Figure 13G ). These findings indicate that EXO1 not only plays a crucial role in the DNA damage response but may also influence DNA repair mechanisms by regulating the expression of BRCA1 and BRCA2.