Transcriptome profiling data with clinical information were obtained from the TCGA-LUAD project by R (version 4.4.1) with the R package TCGAbiolinks. The inclusion of 513 primary solid tumor cases included intact clinical information (age, sex, T stage, N stage, M stage, and prognostic information) and 58 normal samples.

Read counts per gene were generated using HTSeq-count and used as input for pairwise differential expression analysis with DESeq2; the threshold values were |log₂FoldChange| > 1 and adjusted P-value < 0.05. TPMs were transformed to log₂(TPM+1) for further analyses. The R package maftools was used to perform the mutation spectrum analysis. Patient simple nucleotide variation data (MuTect2) retrieved through the R package maftools. Waterfall plots were created using the ComplexHeatmap package. Mutational burden is calculated by multiplying the number of mutations by the number of bases covered in a sample and is reported as mutations per megabase.

Estimate is a method used in tumor samples to determine the proportions of stromal and immune cells by analyzing gene expression patterns. We applied it to assess the TME of each LUAD patient, along with stromal score (stromal content), immune score (extent of immune cell infiltration), ESTIMATE score (synthetic mark of stroma and immune), and tumor purity using the R package estimate (17).

CIBERSORT was applied to calculate immune cell composition from gene expression profiles. This deconvolution algorithm was used to estimate the abundance of different immune cell populations from expression data (18).

The level of infiltration of immune cell types in the R package gsva was analyzed by using ssGSEA. The expression levels of genes in 28 published gene sets for immune cells were used to calculate the extent of infiltration of 28 immune cell types (19).

Survival analysis was performed using R (version 4.4.1) with the packages survival and survminer. Survival was plotted using a Kaplan–Meier survival curve, and statistical significance was determined by the log-rank (Mantel-Cox) test. p < 0.05 was considered significant.

Expressions of FLNA and NDUFS1 were extracted and clustered coherently using the R package ConsensusClusterPlus. The samples were divided into two clusters. Heatmaps were produced by the R language with the package pheatmap.

The C5 gene set was used to download gene sets for canonical pathways and Gene Ontology from the Molecular Signatures Database (MSigDB). GSEA was conducted with the R package clusterProfiler to identify the significant functional difference between the two clusters. Significant pathway enrichment was identified by the normalized enrichment score (|log₂FoldChange| >1), p-value < 0.05, and FDR q-value < 0.05.

Weighted Gene Co-Expression Network Analysis (WGCNA) was performed using the R package WGCNA. To ensure that the constructed co-expression networks are more in line with the characteristics of scale-free networks, we chose 7 as the soft power. We got 12 modules and determined the correlations between them and cluster, stromal score, immune score, ESTIMATE score, and tumor purity.

GO enrichment analyses were performed with R packages clusterProfiler, enrichplot, and ggplot2. Only terms with both p- and q-values of <0.05 were considered significantly enriched.

The PPI network was constructed using the STRING database with a minimum interaction score of medium confidence (0.400), followed by reconstruction using Cytoscape (version 3.10.2).

Human LUAD epithelial cells (A549 and NCI-H1299) were obtained from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China) and cultured in DMEM (low glucose) (Gibco, Shanghai, China), supplemented with 10% fetal bovine serum (FBS) (Gibco). A humidified atmosphere with 5% CO₂ was used to maintain human cells at 37°C and test for mycoplasma contamination. Cell lines were regularly examined to confirm morphological and growth characteristics that indicate that cell line. In this study, the following antibodies and reagents were employed: β-actin (Abcom), FGA (Sigma-Aldrich, Abcom), xCT (Proteintech), goat anti-rabbit secondary antibody (Abcom), Ki-67 (Abcom), and Alexa 488–labeled secondary antibody (Abcam).

Overexpression plasmids were constructed by Genechem. The following siRNA targeting FGA was used: siFGA: 5′-UUUGUAUUUGUGAAGAUGCtt-3′; Lipofectamine 3000 (Invitrogen) was used to transfect cells with the indicated plasmids and siRNAs in accordance with the manufacturer’s instructions.

Total RNA was extracted by using the SPARKeasy Cell RNA Kit (Sparkjade). Reverse transcription was performed using a cDNA synthesis kit (Thermo Fisher Scientific). Primer sequences were as follows: FGA, forward:5′- GTCTTCTTTGCTAGAGAAGTGGAGA-3′, reverse: 5′- AAAGAATGTTTCTCTTGCCTTCCTG-3′. GAPDH, forward:5′-GAAGGTGAAGGTCGGAGTC-3′, reverse: 5′-GAAGATGGTGATGGGATTTC-3′. Relative expression levels were determined by normalizing the expression level of each target to GAPDH, and relative mRNA fold changes were determined using the 2^−ΔΔCt method.

Immunofluorescence staining for Ki-67 to evaluate cell proliferation, following the manufacturer’s instructions. Cells were incubated with a primary Ki-67 antibody (Abcam, USA; 1:300), followed by a secondary Alexa 488–labeled antibody (Abcam, USA; 1:1000) for 1h at room temperature. Subsequently, cells were counterstained with 10 mg/ml DAPI.

A cell proliferation assay was performed using the Cell Counting Kit-8 (CCK-8; beyotime). Transfected with siRNA or plasmid, cells were seeded in 96-well plates and incubated at 37°C for 24h; each well was incubated with 100 μl DMEM (low glucose) (Gibco) supplemented with 10 μl CCK-8 reagent at 37°C for 2h. The absorbance of every well was measured at 450 nm using a microplate reader (Thermo Fisher Scientific, Waltham, MA, USA).

For cell invasion and migration assays, cells were serum-starved for 24h pre-plating, and 2.5 × 104 cells were seeded in with or without Matrigel-coated transwell inserts (8-μm pore size, Corning). Briefly, cells in 100 μl serum-free media were added into the upper chamber, and 500 μl culture media with 10% FBS was in the lower well. After incubation for 24h, cells that remained on the upper surface of the membrane were removed with a cotton swab, and those on the underside of the membrane were fixed with 4% paraformaldehyde and stained with 0.5% crystal violet and microscopically counted from three random fields of each membrane. Carefully wounding confluent cell layers in six-well plates with a sterile 200 µl tip, washing them twice with medium, and culturing them for 48h. At a specific time, the area between the wound edges was measured before and after the recovery and quantified using light microscopy and ImageJ (version 1.54).

Intracellular ROS levels were measured using the CM-H2DCFDA assay kit (Beyotime Biotechnology, Shanghai, China) as per the manufacturer’s instructions. Treated or control adherent cells (incubated in low glucose medium) were washed with PBS. Cells were then incubated in situ with 5 µM CM-H2DCFDA diluted in the assay diluent provided with the kit for 30 min at 37°C in the dark. After washing the cells with PBS, they were analyzed using a flow cytometer (Agilent NovoCyte). The resulting data were processed using FlowJo software.

All animal experiments were conducted in accordance with accepted standards of animal care and approved by the Institutional Animal Care and Use Committee of Affiliated Hospital of Jining Medical University (Jining, China). Seven-week-old female BALB/c nude mice were obtained from Shandong Pengyue (Shandong, China) and randomly divided into two groups. The mice were subcutaneously injected into the right flank with logarithmic-phase A549 cells (8 × 10⁶ cells per mouse) transfected with shFGA or Ctrl-GFP to establish a subcutaneous lung cancer xenograft model. Tumor size was measured every other day using calipers, and volume (V) was calculated as V = (L × W²)/2, where L is the longest diameter and W is the perpendicular shorter diameter.

Co-IP assays were performed using Classic Magnetic Protein A/G IP Kits (Epizyme, Shanghai, China). Briefly, protein lysates were incubated with anti-FGA or anti-xCT (mouse monoclonal) antibodies for 1h, followed by incubation with pre-washed magnetic beads for an additional hour. After magnetic separation for 1 min, the bead-antibody-protein complexes were washed four times with lysis buffer to remove nonspecific interactions. Finally, the immunoprecipitated proteins were separated by SDS-PAGE for downstream analysis.

A non-contact co-culture system was established using 0.4-μm pore Transwell inserts (Corning). A549 cells and CD8⁺ T cells were co-cultured at a 1:1 ratio. Transfected A549 cells (2 × 10⁵ cells/well) were seeded in six-well plates. Pre-activated CD8⁺ T cells (2 × 10⁵ cells/insert) were added to the Transwell inserts. Cells were maintained in RPMI-1640 medium supplemented with 10% FBS and 100 IU/ml IL-2. After 48h, CD8⁺ T cells from the inserts were collected for PCR analysis.

Statistical analysis was performed using GraphPad Prism 9.0.0 (GraphPad Software, Inc.) and is shown as mean ± SD as indicated. Replicates are indicated in figure legends. The equality of variance was determined through an F-test. In samples that were normally distributed, a two-tailed t-test was used to compare the means of the variables between two groups. In samples with nonnormal distributions, the medians of the variable between two groups were compared by a Mann–Whitney U test. For analysis of more than two groups, we used the analysis of variance (ANOVA) to determine the equality of variance. Comparisons between groups were performed with Tukey–Krammer post-hoc analysis. For all tests, P < 0.05 was considered statistically significant.

To determine the relationship between the estimated proportions of immune and stromal cells with the survival rate, we performed Kaplan–Meier survival analysis with 95% confidence for ImmuneScore, StromalScore, and ESTIMATEScore, respectively. The ESTIMATE algorithm by the R package ESTIMATE (20, 21) was used to assess the immune and stromal components in the TME for each sample. This assessment yielded three scores: ImmuneScore, StromalScore, and ESTIMATEScore. These scores were positively correlated with the proportions of immune, stromal, and the sum of both, respectively. Therefore, higher scores indicated a larger proportion of the corresponding components in the TME. As illustrated in Figures 1A–C, elevated ImmuneScore, StromalScore, and ESTIMATEScore demonstrated significant associations with improved overall survival rate (p < 0.05). According to these findings, the immune and stromal components in TME were effective predictors of prognosis in patients with LUAD.

We analyzed the clinicopathological data from TCGA-LUAD cases to investigate associations between the proportions of immune and stromal components with their clinicopathological characteristics. ESTIMATE score demonstrated progressive decline with advancing TNM stage (Figure 2A). Immune Score showed inverse correlations with T classification, N classification, and overall stage (Figure 2B). StromalScore exhibited negative associations with T, M classification, and overall stage, but not N classification (Figure 2C). These findings indicate that altered immune-stromal compositional ratios correlate with advanced LUAD progression, particularly in mediating invasive and metastatic processes.

To characterize the immune landscape, we applied the CIBERSORT algorithm to quantify tumor-infiltrating subsets, generating 22 kinds of immune cell profiles in LUAD samples (Figure 2D). Comparative transcriptomic analysis of LUAD versus normal tissues in TCGA identified 5,404 differentially expressed genes (DEGs, |log₂FC|>1, FDR < 0.05), and results were visualized using a volcano plot (Figure 2E). The genes were then considered for the WGCNA (Figures 2F, G). To identify a module related to immunity, we performed a correlation between modules and traits (Figure 2H). Notably, the brown module tended to be significantly associated with immunity (R = 0.84, P = 3e-140) (Figure 2H). We conducted Gene Ontology (GO) enrichment analysis on the brown module genes, revealing significant enrichment (FDR < 0.05) in biological processes including regulation of cell–cell adhesion and positive regulation of cytokine production, cellular components such as secretory granule membrane and the external side of the plasma membrane, as well as molecular functions encompassing immune receptor activity and carbohydrate binding (Figure 2I).

To delineate the immunomodulatory role of disulfidptosis in LUAD, 24 disulfidptosis-related genes were identified, and correlations between these genes and ESTIMATE scores were evaluated (Figure 3A). FLNA and NDUFS1 were included in the next analysis based on the highest absolute value of the correlation with the immune score. RNA-seq data from the TCGA-LUAD cohort were analyzed using the Wilcoxon rank-sum test, demonstrating significantly lower FLNA expression levels (p < 0.001) and elevated NDUFS1 expression (p < 0.001) in tumor tissues compared to normal counterparts (Figures 3B, C). Consensus clustering of 513 TCGA-LUAD samples based on FLNA and NDUFS1 expression stratified two clusters. The heatmap shows Cluster 1 (n = 268) has high expression of FLNA and low expression of NDUFS1, while Cluster 2 (n = 245) had low expression of FLNA and high expression of NDUFS1 (Figure 3D).

ESTIMATE and GSEA were performed to understand the differences in immunological function better. In ESTIMATE analysis, Cluster 1 demonstrated higher stromal, immune, and ESTIMATE scores along with lower tumor purity than Cluster 2 (Figures 3E–G). To functionally annotate inter-cluster expression differences, we conducted Gene Set Enrichment Analysis (GSEA) using the c5.all.v7.0.Symbols.gmt (version 7.0 of the Molecular Signatures Database) reference set from the Molecular Signatures Database (v7.0). Our analysis incorporated all differentially expressed genes between Cluster 2 and Cluster 1. Multiple immune-related pathways showed significant enrichment, including leukocyte cell–cell adhesion, positive regulation of leukocyte cell–cell adhesion, regulation of leukocyte-mediated immunity, regulation of T-cell activation, and T-cell activation (Figure 3H). Clinical and tumor pathologic features of patients are summarized in Figure 3I.

We conducted CIBERSORT and ssGSEA analyses to characterize immune infiltration heterogeneity between the two clusters. CIBERSORT analysis demonstrated that Cluster 1 exhibited significantly higher proportions of B-cell memory, T-cell CD4 memory resting, T-cell regulatory (Tregs), monocytes, macrophages M0, macrophages M2, dendritic cells resting, and mast cells resting compared to Cluster 2 (Figure 4A). The ssGSEA further identified 28 immune cell subtypes, including activated B cells, activated CD8 T cells, activated dendritic cells, mast cells, macrophages, natural killer cells, and natural killer T cells, highly expressed in cluster 1 (Figure 4B). Next, we compared the expression levels of immunomodulatory targets in the two clusters. Most targets, including PD1, PDL1, PDL2, CTLA4, CD80, CD86, LAG3, TIM3, TIGHT, OX40, GITR, 4-1BB, ICOS, CD27, and CD70, were significantly higher in Cluster 1 (Figures 4C–F). Landscapes of mutation profiles between the two clusters are depicted (Figures 4G, H). Cluster 1 had higher TMB than Cluster 2 (Figures 4I). These results suggest that Cluster 1 had a stronger immune infiltration and a stronger response to immunotherapy than Cluster 2.

To identify a disulfidptosis and immunity-associated module, we analyzed the intersection between WGCNA-derived brown module genes and inter-cluster differentially expressed genes, yielding 164 consensus genes (Figure 4J). The protein-protein interaction (PPI) network was constructed using STRING (Search Tool for the Retrieval of Interacting Genes/Proteins) and visualized through Cytoscape v3.10.2, with nodes representing proteins and edges indicating functional associations (Figure 4K). Based on network topology analysis, genes were prioritized by node connectivity, with the 15 most interconnected candidates selected for further investigation (Figure 4L). These genes were subsequently advanced for functional validation studies.

Based on our literature mining and analysis of relevant studies, we identified FGA as a potential target gene and have decided to focus our research efforts on exploring the functional significance role of FGA in LUAD. A549 cells were cultured in DMEM (low glucose) and transfected with either control small interfering RNA (siCtrl) or FGA-targeting small interfering RNA (siFGA). To verify the effectiveness of these knockdowns, we analyzed RNA prepared from cells 24h after transfection by qPCR. mRNA expression of RNAi targets was efficiently reduced (Figure 5A).

Cell proliferation was first assessed through Ki-67 immunofluorescence labeling, revealing significantly elevated proliferative activity in siFGA-treated cells versus controls (Figures 5B, C). Subsequently, we performed an assessment of the effect of transfection on cell proliferation at 48 (Figure 5D) and 72h (Figure 5E) using a CCK8 assay. The results indicated that the knockdown of FGA significantly stimulated the proliferation of A549 cells in comparison to the siCtrl cells. Transwell migration and invasion assays demonstrated that siFGA significantly increased the number of migrating and invading cells (Figures 5F–H). To investigate whether siFGA reduces intracellular reactive oxygen species (ROS) levels, thereby decreasing glutathione (GSH) consumption and inhibiting disulfide bond formation and subsequent disulfidptosis, we measured intracellular ROS levels 24h post-transfection using the ROS Assay Kit (CM-H2DCFDA probe and Diluent). The results demonstrated that siFGA significantly reduced ROS levels (Figure 5I).

To assess the impact of FGA on tumor growth in vivo, tumor growth kinetics were evaluated in 7-week-old female immunodeficient BALB/c nude mice. Mice were subcutaneously inoculated with shFGA-expressing A549 cells or vector control A549 cells. Results showed that compared to tumors derived from vector control cells, xenograft tumors from shFGA A549 cells exhibited significantly accelerated growth (Figure 5J). These findings demonstrate that FGA suppresses LUAD progression in vivo. Collectively, these results suggest that FGA potentially functions as a tumor suppressor in LUAD via disulfidptosis.

The association between FGA and disulfidptosis was explored through Spearman’s rank correlation analysis across disulfidptosis-related gene sets. This revealed a statistically significant positive correlation between FGA and xCT (R = 0.35, P = 3.63e−16) (Figure 6A). Survival analysis revealed a negative correlation between xCT expression and overall survival (P = 0.011) (Figure 1D). Western blotting demonstrated that FGA-knockdown cells showed reduced xCT protein expression compared to siCtrl-transfected cells at 48h post-transfection (Figure 6B). Densitometric quantification of protein bands using ImageJ software confirmed significant downregulation of both FGA and xCT in the experimental group (Figures 6C, D).

To delineate whether FGA exerts tumor suppressive effects via xCT-mediated pathways, we conducted rescue experiments by transfection of xCT overexpression plasmid (OExCT). LUAD cells were co-transfected with either siCtrl or siFGA, along with either a control vector or OExCT. There were four experimental groups: the siCtrl and vector co-transfection group, the siFGA and vector co-transfection group, the siCtrl and OExCT co-transfection group, and the siFGA and OExCT co-transfection group. We found that transfection of the xCT overexpression plasmid significantly restored xCT protein expression in siFGA-transfected cells (Figure 6E).

To mechanistically define the tumor suppressor function of FGA dependent on xCT, functional phenotypic assays were employed. Functional characterization in both A549 and NCI-H1299 cells demonstrated that xCT overexpression attenuated the phenotypes induced by FGA knockdown; xCT overexpression partially diminished the enhanced effects of FGA knockdown on cell migration and invasion (Figure 7A). Similarly, xCT overexpression partially counteracted the enhanced effects of FGA knockdown on migration of A549 and NCI-H1299 cells, as demonstrated in a wound-healing assay (Figure 7B). Additionally, CCK8 assay results showed that in A549 and NCI-H1299 (Figure 7C) cells, the stimulatory effect of FGA knockdown on cell proliferation was partially reversed by restoring xCT expression. Taken together, these results suggest that FGA through xCT inhibited tumor progression by suppressing cell viability, formation, migration, invasiveness, and tumor growth of lung cancer cells.

To explore the undefined molecular mechanism by which FGA regulates xCT in LUAD cells, we performed co-immunoprecipitation (co-IP) assays to determine physical interaction between these proteins. For all co-IP experiments, whole-cell lysates were used as positive controls in immunoblotting. Normal mouse IgG served as negative controls matched for isotype in immunoprecipitation. Reciprocal co-immunoprecipitation assays in A549 cells confirmed a specific physical interaction between FGA and xCT. Endogenous FGA immunoprecipitates pulled down xCT from whole-cell lysates (Figure 7D), while reverse IP with anti-xCT antibodies co-precipitated FGA (Figure 7E). Collectively, reciprocal co-immunoprecipitation analyses demonstrate that endogenous FGA and xCT form a protein complex in A549 cells. At the subcellular level, FGA resides in the endoplasmic reticulum as a secretory protein, while xCT resides in vesicles as a membrane transporter protein (Figure 7F). Source: The Human Protein Atlas. This physical interaction mechanistically underlies FGA-driven disulfidptosis regulation in LUAD.

To investigate the impact of tumor cell FGA on CD8⁺ T-cell activation, we employed a non-contact Transwell co-culture system. Following 48h of co-culture in a non-contact Transwell system, where pre-activated CD8⁺ T cells (upper chamber) were exposed to soluble factors from either control or siFGA A549 cells (lower chamber), qPCR analysis of the CD8⁺ T cells co-cultured with siFGA A549 cells significantly enhanced the expression of activation markers CD69 (p < 0.01) and CD25 (p < 0.05) (Figure 7G) in CD8⁺ T cells compared to the control group. This augmented expression of T-cell activation markers suggests an enhanced anti-tumor immune response in the context of tumor cell FGA deficiency.

The Human Protein Atlas contains extensive transcriptomics and proteomics data for specific human tissues and includes the Tissue Atlas, Cell Atlas, and Pathology Atlas. In this study, we examined the protein expression of the FGA gene in normal lung tissues and lung cancer tissues using this database, and the results suggest that FGA is decreased in normal tissue and increased in lung cancer tissues (Figure 7H). This spatial compartmentalization pattern, integrated with phenotypic rescue evidence, delineates a mechanistic framework wherein FGA within the TME suppresses oncogenesis and influences immune infiltration through xCT-mediated regulation of disulfidptosis.