Data from public databases, including The Cancer Genome Atlas (TCGA; http://portal.gdc.cancer.gov/) and Genotype-Tissue Expression (GTEx; http://gtexportal.org/home/), were used in the present study. TCGA data portal was used to download survival data and tumor staging data for LUAD and LUSC; from a total of 585 cases of LUAD and 504 cases of LUSC. Of these, RNA sequencing data and survival data were available for 516 LUAD cases and 501 LUSC cases. Gene expression profiles were obtained for 59 patients with LUAD and 49 patients with LUSC. Furthermore, data from 578 lung RNA-seq profiles were downloaded from GTEx-lung.

The ferroptosis gene expression profiles of all samples were extracted using R software (version 4.0.3; The R Foundation for Statistical Computing) (10). The ferroptosis genes that were differentially expressed in the tumor group compared with in the non-tumor group were identified using the tidyverse package (www.tidyverse.org) and the Deseq2 package (P<0.05) (https://bioconductor.org/packages/release/bioc/html/DESeq2.html). Subsequently, LUAD and LUSC were divided into three groups according to tumor stage and the differential expression of ferroptosis genes between the groups was analyzed. The Wilcoxon rank-sum test was used for statistical analysis of two groups (11), whereas the Kruskal-Wallis test and Dunn's post hoc test was used for multi-group comparisons.

Survival data were extracted from survminer (https://cran.r-project.org/web/packages/survminer/index.html). Samples were divided into high and low expression groups according to the median value of gene expression. Outcomes included overall survival (OS), disease-free survival (DFS) and progression-free survival (PFS). Kaplan-Meier survival curves (12) were plotted and the log-rank test was performed to analyze the association between the ferroptosis-associated differentially expressed genes and prognostic data. The hazard ratio and 95% CI were used to estimate OS, DFS and PFS.

Sankey diagrams were used to generate graphical representations of the distribution of ferroptosis genes in different tumor stages, and to indicate the relationship between gene expression and patient survival. The R software package ggalluvial was used to construct the Sankey diagram (https://cran.r-project.org/package=ggalluvial). The association between differentially expressed genes and tumor size, lymph node metastasis and distant metastasis was determined using the Kruskal-Wallis test and Dunn's post hoc test.

The Tumor Immune Estimation Resource (TIMER) score, EPIC score and CIBERSORT were calculated using the R immunedeconv package (v4.0.3) (https://omnideconv.org/immunedeconv/). Tumor immunophenotype (TIP) (https://github.com/dengchunyu/TIP) was calculated using the R immunedeconv package (v4.0.3). The TIMER score is used to estimate the number of immune cells present; EPIC provides the relative proportions of cells; CIBERSORT is used for the assessment of T-cell characteristics; and the proportion of tumor-infiltrating immune cells in the cancer immune cycle can be tracked and analyzed using TIP. The immune correlation network graph was used to visualize the correlation between gene expression and immune scores, and the correlation between immune scores themselves.

To clarify the relationship between genes and pathways, the enrichment fraction of each sample in each pathway was calculated using the Single Sample Gene Set Enrichment Analysis (ssGSEA) algorithm (13), and the Spearman correlation between gene expression and the pathway score was calculated.

A total of 99 patients with LUAD who had undergone surgery at the Central Hospital Affiliated to Shenyang Medical College (Shenyang, China) and Shengjing Hospital (Shenyang, China) between January 2015 and December 2018 were included in the present study. The patients ranged in age from 36 to 81 years(average age, 58.97±9.49 years), including 55 men and 44 women. Sectioning and immunohistochemical staining were performed using paraffin-embedded tissue sections from patients with a pathological diagnosis of LUAD, which were obtained from the Pathology Department, Central Hospital Affiliated to Shenyang Medical College. The Central Hospital Affiliated to Shenyang Medical College ethics committee reviewed and approved the present study [approval no. Ke-2024-128(02)]. The use of the samples remaining after clinical diagnosis had no adverse effects on the patients; therefore, the need for consent was waived.

For immunohistochemistry, paraffin-embedded sections were collected under Office for Human Research Protections guidelines. Briefly, the paraffin-embedded sections (3 µm), underwent antigen retrieval with citric acid under high pressure for 3 min and were blocked with 3% hydrogen peroxide for 20 min at 24°C. Subsequently, the sections were incubated with primary antibodies at 4°C for 12 h and with MaxVision™ HRP-Polymer anti-Mouse/Rabbit secondary antibodies (cat. no. kit5030; Fuzhou Maixin Biotechnology Development Co., Ltd.) at 4°C for 0.5 h. The sections then underwent DAB staining for 1 min, after which, they were counterstained with hematoxylin for 1 min, sealed with neutral resin and observed under an Olympus BX53 light microscope (Olympus Corporation). The following primary antibodies were used: CDGSH iron-sulfur domain-containing protein 1 (CISD1; 1:100; cat. no. TA500905; Origene Technologies, Inc.), CD4 (1:100; cat. no. RMA-0620; Fuzhou Maixin Biotechnology Development Co., Ltd.) and CD20 (1:100; cat. no. Kit0001; Fuzhou Maixin Biotechnology Development Co., Ltd.)]. Cytoplasmic brown staining was considered positive for immunohistochemistry. The interpretation of the results was the responsibility of two senior pathologists. The degree of CISD1 positivity was categorized as follows: 0, not positive; 1, weakly positive (light yellow); 2, moderately positive (yellowish-brown); and 3, strongly positive (tan). Regarding the CISD1 staining area score, a score of 0–100 was assigned according to the percentage of positive tumor cells/total tumor cells. The total staining score was equal to the product of the staining area score and the degree of positivity of the staining results. A total score of ≥100 indicated high CISD1 expression, whereas a total score of <100 indicated low CISD1 expression. For CD4 and CD20 staining, only the percentage of positive cells was calculated.

R software (version 4.0.3) was used for both data analysis and visualization. SPSS 17.0 (IBM Corp.) was used for the χ² test. Two groups of specimens were compared using the Wilcoxon rank-sum test, whereas three groups of specimens were compared using the Kruskal-Wallis test and Dunn's post hoc test. Spearman correlation analysis was used to analyze the correlation between immunohistochemical scores of clinicopathological indicators, and Kaplan-Meier analysis and log-rank test was used for survival analysis (SPSS v17.0). P<0.05 was considered to indicate a statistically significant difference.

Data for 516 cases of LUAD and 59 matched samples of paracancerous tissues, and 501 cases of LUSC and 49 matched samples of paracancerous tissues were downloaded from TCGA. A total of 578 lung tissue samples without disease in GTEx were included due to the small amount of data in the control group. Differences between the tumor and non-tumor groups were analyzed by extracting the expression profiles of >20 classical ferroptosis genes. Compared with those in normal tissues, the expression levels of SLC7A1, HSPB1, FANCD2, CISD1, HSPA5, NCOA4, GPX4, DPP4, RPL8 and SLC1A5 were upregulated in LUAD cancer tissues, whereas the expression levels of CDKN1A, EMC2, NFE2L2, MT1G, LPCAT3, GLS2, SAT1, CS, ALOX1, ACSL4, ATL1 and TFRC were downregulated (Figs. 1A and S1A). In LUSC cancer tissues, the expression levels of SLC7A1, NFE2L2, HSPB1, FANCD2, CISD1, FDFT1, SLC1A5, HSPA5, NCOA4, GPX4, RPL8 and TFRC were upregulated, whereas the expression levels of CDKN1A,LPCAT3, GLS2, SAT1, DPP4, CS, ACSL4, ATL1 and EMC2 were downregulated (Figs. 1B and S1B).

Patients with LUAD and LUSC were grouped based on Tumor-Node-Metastasis (TNM) stage. Patients with LUAD or LUSC were divided into three groups, G1 corresponds to stage I, G2 corresponds to stage II, G3 corresponds to stage III and G4 corresponds to stage IV. MT1G, CISD1, RPL8, GLS2, CS, CARS1, ATP5M3 and ACSL4 were associated with the TNM stage of LUAD (Fig. 2A and B). CDKN1A, EMC2 CISD1, TFRC, ALOX15 and ATL1 were associated with the TNM stage of LUSC (Fig. 2C and D). CISD1 was identified in both LUAD and LUSC.

LUAD and LUSC differ in how they are treated and prognosticated. For the prognosis of LUAD and LUSC, CISD1 was further analyzed. The results revealed that in LUAD, CISD1 was associated with prognosis, and high levels were associated with lower OS and PFS (P<0.05; Fig. 3A and C). This suggested that high levels of CISD1 were associated with poor prognosis. In LUSC, CISD1 was not associated with prognosis (Fig. 3D-F). The relationship between CISD1 and LUAD was thus focused on.

There were 187 deaths and 329 survivors among all patients with lung cancer from TCGA. Among the patients who died, there were 69 cases in stage I, 55 cases in stage II, 47 cases in stage III and 16 cases in stage IV, including 111 cases (58.73%) with high CISD1 expression and 76 cases (40.64%) with low CISD1 expression. Among the surviving patients, 147 patients (44.68%) were in the high-level CISD1 group and 182 patients (55.32%) were in the low-level CISD1 group. Although 58.73% of the dead patients had high expression of CISD1, it was mainly concentrated in stage I and II, suggesting that the high expression may delay rather than completely prevent the disease progression. By contrast, patients with low expression were more distributed in stages III and IV, and were associated with 62.5% of stage IV deaths, highlighting its strong association with aggressive phenotypes. Patients with low expression of CISD1 accounted for 55.32% of the survival cohort, which may reflect the effectiveness of early intervention for patients with low expression, or the existence of other compensatory pathways to maintain survival (Fig. 4A). Patients with high expression of CISD1 accounted for 42.0% of patients with T1 cancer, and the survival rate (67.6%) was lower in the high-expression group than that in the low-expression group (79.6%). Notably, the survival rate of patients with high CISD1 expression in T4 stage was low (16.7%), which was in sharp contrast with the low-expression group (71.4%) (Fig. 4B). The survival rate of patients with low CISD1 expression in N0 stage was 76.4% (139/182), which was significantly higher than that of patients with high expression (69.3%), suggesting that CISD1 inhibition may improve the prognosis through immune regulation in the metastasis-free stage. However, the survival rate of patients with high expression in N1-N2 stage decreased sharply (N1: 42.4%; N2: 36.4%), and was significantly lower than that of patients with low expression in the same period, suggesting that CISD1 may drive malignant progression by enhancing tumor cell adaptability (such as antioxidant defense) in lymph node metastasis (Fig. 4C). In M0 stage, high CISD1 expression accounted for 52.4%, and its survival rate (56.6%) was lower than that of the low expression group (67.3%). In M1 stage, high CISD1 expression accounted for 64.0%, and its survival rate (31.3%) was lower than that in the low expression group (55.6%) (Fig. 4D). High CISD1 expression may exacerbate adverse outcomes by promoting metastasis or treatment resistance. However, the χ² test showed that the level of CISD1 was associated with the outcome; a high level of CISD1 could indicate a poor outcome (Table I). In order to exclude intra-group differences, different clinical stages and non-tumor tissues were also compared. CISD1 expression was positively associated with TNM stage (Fig. 4E), and TNM stage was closely associated with tumor prognosis (Fig. 4F). This suggested that, when CISD1 expression was high in patients, the risk of adverse outcomes was also high. It is recommended that consideration be given to the addition of ferroptosis-related drugs in the design of treatment plans for patients.

The present study further analyzed the association between CISD1 and immunity to explore the potential application value of targeting ferroptosis in the immunotherapy of LUAD. The network connection diagram and heat map were used to visualize the association between gene expression and the immune score (Fig. 5). The more red or blue, the stronger the correlation between both, and the larger the ring, the stronger the correlation. The red line represents a positive correlation between the model score or the gene expression and the immune score, and the green line represents a negative correlation between the two. The TIMER score (Fig. 5A), EPIC score (Fig. 5B), CIBERSORT score (Fig. 5C) and TIP score (Fig. 5D) showed that CISD1 was negatively correlated with CD4⁺ T cells and B cells.

The ssGSEA algorithm was used to calculate the enrichment fraction of each sample for a given pathway to investigate the relationship between samples and pathways. These calculated enrichment fractions provided an intuitive way to show how samples and pathways interacted and were associated. The results suggested that CISD1 may be involved in the cellular response to hypoxia, DNA repair, extracellular matrix-related genes, epithelial-mesenchymal transition markers, oxidative phosphorylation, PI3K-AKT-mTOR signaling pathway and metabolic pathways (glycine, serine and threonine metabolism; D arginine and D ornithine metabolism; alanine, aspartate and glutamate metabolism; biotin metabolism) in LUAD (Fig. 6).

CISD1 expression in LUAD cancer tissues (Fig. 7A and B) was significantly higher than that in adjacent tissues (Fig. 7C and D), as determined using Wilcoxon rank-sum test (P<0.001; Z=−7.471; Fig. 7E). In LUAD, the χ² test indicated that CISD1 was positively associated with lymph node metastasis (P=0.010; χ² =6.559), and Ki67 (P=0.005; χ² =7.932) (Table II). By contrast, CISD1 was not associated with sex, tumor differentiation, tumor size and TNM in patients with LUAD. Furthermore, high CISD1 expression indicated poor prognosis and a shorter DFS (P=0.002; Fig. 7F).

A total of 60 cases were randomly selected for CD4 and CD20 staining. The results demonstrated that the expression levels of CISD1 in LUAD were negatively associated with the infiltration of immune cells. The number of CD4 cells was lower in cases with high CISD1 expression (Fig. 8A and B), and the number of CD20 cells was also lower (Fig. 8F and G). The number of CD4 and CD20 immune cells was higher in cases with low CISD1 expression (Fig. 8D, E, I and J). CISD1 was negatively associated with the immune infiltration of CD4 (P<0.001; Z=−6.575) and CD20 (P<0.001; Z=−5.970) cells using Wilcoxon signed-rank test (Fig. 8C and H).