Therapeutic and prognostic effect of disulfidptosis-related genes in lung adenocarcinoma

Abstract

Disulfidptosis, a new form of cell death, may be induced by disulfide stress associated with cystine disulfide buildup, which can promote cell toxicity, leading to cell death. Nevertheless, the role of direct prognosis and the mechanism underlying the regulation of disulfidptosis-related genes (DRGs) in lung adenocarcinoma (LUAD) are still unknown. This study aimed to investigate the role of DRGs in LUAD prognosis and diagnosis through multiomics analysis. First, copy number variations (CNVs) and mutations in the 10 genes were assessed. Considering that five differentially expressed genes (DEGs) were associated with disulfidptosis, a novel DRG score that can be utilized to anticipate LUAD prognosis was developed. Next, the generated receiver operating characteristic (ROC) and survival curves demonstrated that the model had an excellent predictive quality in LUAD in both the training and validation cohorts. Meanwhile, substantial functional disparities between the high DRG group and the low DRG group were observed, and the second gap mitosis (G2M) checkpoint, E2 promoter-binding factor (E2F) targets, and myelocytomatosis (MYC) target activities were consistently higher in the high DRG group than in the low DRG group. Additionally, the T-cell dysfunction score and tumor inflammation signature (Merck18) were negatively correlated with DRGs, whereas myeloid-derived suppressor cells (MDSCs) were positively correlated with DRGs. Moreover, DRGs were negatively linked to most of the immunological checkpoints. Meanwhile, samples of low DRGs benefited more from immune checkpoint blockade (ICB). The correlation analysis between DRGs and clinical characteristics revealed increasing malignancy with increasing DRG scores. Drug sensitization experiment results indicated that sensitivity to cisplatin, vincristine, docetaxel, and gemcitabine was higher in the high DRG group than in the low DRG group. The function of model genes in LUAD was also verified using immunohistochemistry, quantitative reverse transcription-polymerase chain reaction (qRT-PCR), western blotting, 5-ethynyl-2′-deoxyuridine (EDU), and clonogenic formation.

1. Introduction

Lung cancer is the major reason for cancer-related deaths worldwide [1,2]. Most patients with lung cancer present with advanced stages at the time of diagnosis. Indeed, even in patients undergoing final treatment with chemotherapy or radiotherapy for locally advanced or early-stage resectable disease, up to 90 % of patients will eventually experience relapse of their disease, with a <60 % survival rate at 5 years [2,3]. Non–small cell lung cancer (NSCLC) accounts for 80%–85 % of all lung cancers, which comprise three main types: large cell carcinoma, squamous cell carcinoma, and adenocarcinoma [4]. Of these subtypes, LUAD which is the most heterogeneous and aggressive type, has a very high tumor mutation burden associated with epidermal growth factor receptor (EGFR), Kirsten rat sarcoma viral oncogene homolog (KRAS), B-Raf proto-oncogene, serine/threonine kinase gene (BRAF), tyrosine kinase receptor 2 (ERBB2), tumor protein p53 gene (TP53), anaplastic lymphoma kinase (ALK), serine/threonine kinase 11 (STK11), and thyroid transcription factor 1 (TTF1) mutations [[5], [6], [7], [8]]. Although progress has been made in chemotherapy, radiotherapy, targeted therapy, and immunotherapy in the past 10 years, owing to the limited understanding of the molecular mechanisms underlying early lung carcinogenesis, and because most cases are diagnosed very late, effective prevention and treatment of lung cancer remain challenging [9]. Therefore, it is of significant value to explore novel molecular biological mechanisms for the treatment and prognosis of lung cancer.

As one of the common molecular biological mechanisms related to tumor treatment and prognosis, programmed cell death is a mechanism through which cell death occurs and unnecessary cells are removed. Cell death through different forms, including apoptosis, ferroptosis, and cuproptosis and is related to tumorigenesis, progression, and drug resistance [10,11]. Apoptosis is precisely modulated at the level of genes, leading to the orderly and effective clearance of injured cells, and numerous antineoplastic drugs utilize this well-established signaling pathway to initiate cancer cell death [12]. Ferroptosis, another mode of cell death, involves iron accumulation that manifests primarily as cytological changes, including a decrease in or absence of mitochondrial cristae, breakdown of the outer mitochondrial membrane, and merger of mitochondrial membranes, thereby increasing the vulnerability to cisplatin in lung cancer [13]. Cuproptosis, a copper-induced cell death that regulates the tricarboxylic acid cycle, can cause cytotoxicity and affect tumor multiplication in different cancers [14]. Disulfidptosis is a new kind of cell death unlike the traditional modes of cell death such as apoptosis, resulting from excessive intracellular cystine accumulation under glucose scarcity conditions. It is an adenosine triphosphate (ATP) level–independent mode of rapid cell death associated with disulfide stress [15]. Under glucose scarcity conditions in solute carrier family 7 member 11 (SLC7A11)^high cells, the substantial acceleration of disulfide bond accumulation results in the aberrant cross-linking of disulfide among actin cytoskeletal proteins and cytoskeletal contractions, disturbing the tissue and, eventually, leading to cell death because of the collapse of the actin network [15,16]. Disulfidptosis correlated closely with several types of cancers, which lead to disulfide metabolism disorders that affect cancer cell survival and proliferation [17,18]. Additionally, disulfide metabolism in cancer cells is also associated with biological behaviors such as drug resistance and metastasis [19,20]. Furthermore, disulfidptosis may have a certain correlation with tumor immune response. The cell death signal produced by disulfidptosis may be recognized by tumor immune cells, thereby activating the immune response of tumor-specific T cells and enhancing the effectiveness of humoral and cellular immunity, ultimately improving cancer treatment efficacy. Zhao et al. reported that disulfidptosis-related gene (DRG) signature is an excellent tool to identify the immune landscape of patients with bladder cancer and predict their prognosis [21]. Additionally, prognostic signature based on disulfidptosis–DEGs could better predict survival outcomes in patients with breast cancer [22]. Five long non-coding RNAs (lncRNAs) associated with disulfidptosis could predict the survival of patients with cervical cancer and facilitate clinical management of patients in a stratified manner combined with clinical characteristics [23]. However, the exact predictive role of DRGs in LUAD and their innate modulatory mechanisms are unclear. Thus, it is worthwhile to investigate how disulfidptosis functions in LUAD.

Because of the role of disulfidptosis and its related genes in the progression of multiple tumors and their value in diagnosis and prognosis, it can be inferred that DRGs may also play a role in the occurrence and development of LUAD and may serve as effective predictors and therapeutic targets for patients with LUAD. Therefore, this study conducted multi-directional analysis on the transcriptomic data of LUAD from The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) databases to reveal the correlation between DRGs and prognosis prediction, tumor immune evasion, and drug resistance in LUAD, providing certain theoretical support for the treatment and pathogenesis of LUAD. Firstly, DRGs were analyzed for messenger RNA (mRNA) expression, copy number variations, mutations, and predictive capacity. Furthermore, their biofunction was identified utilizing both enrichment analysis and unsupervised clustering analysis. Moreover, differential gene analysis was performed according to the outcomes of the unsupervised analyses. Subsequently, DEGs were identified using Cox regression and LASSO algorithms to establish a model based on the DEGs. The ROC and decision curve analysis (DCA) curves were plotted to demonstrate a high predictive value of the model for LUAD. Next, the DRGs were correlated with clinical characteristics, tumor immune dysfunction and exclusion (TIDE), immunotherapy, targeted therapy, and chemotherapy using relevant bioinformatics algorithms. Finally, the function of model genes in LUAD was validated using immunohistochemistry, qRT-PCR, western blotting, EDU, and clonogenic formation. The results of these analyses may demonstrate that disulfidptosis possibly engages in LUAD progression, thereby helping guide the treatment and predict LUAD prognosis.

2. Materials and methods

2.1. Data and material acquisition

Overall, 310 cases of TCGA-LUAD were retrieved from the TCGA, which was set as the training cohort. The validation cohort (GSE31210) was downloaded from the GEO, which consisted of 226 patients (105 men and 121 women), whose mean age was 60 years, ranging from 30 to 76 years. This cohort included 115 patients who had never smoked and 111 who had ever smoked. The pathological stages included 114 stage IA cases, 54 stage IB cases, and 58 stage II cases [24]. Fourteen pairs of fresh LUAD tissue (cancer and adjacent normal tissues) and paraffin-embedded sections from 28 cases of patients with LUAD were collected for immunohistochemistry (IHC). All experimental procedures were approved by the Ethics Committee of The Third People's Hospital of Luoyang.

Moreover, 10 genes related to disulfidptosis, including glycogen synthase 1 (GYS1), leucine-rich pentatricopeptide repeat–containing (LRPPRC), NCK-associated protein 1 (NCKAP1), NADH:ubiquinone oxidoreductase subunit A11 (NDUFA11), NADH:ubiquinone oxidoreductase core subunit S1 (NDUFS1), nucleotide-binding protein–like (NUBPL), mitochondrial 3-oxoacyl-ACP synthase (OXSM), ribophorin I (RPN1), solute carrier family 3 member 2 (SLC3A2), and solute carrier family 7 member 1 (SLC7A1), were assessed [15].

2.2. Cluster analysis of 10 disulfidptosis-related genes

All 310 tumor samples from the training cohort were subjected to unsupervised hierarchical cluster analysis using the R package ConsensusClusterPlus (version: 4.22) [25] based on the 10 DRGs.

2.3. DEG analysis of 10 disulfidptosis-related genes

The R statistical package limma (version: 4.22) was utilized to evaluate DEGs between the two groups [26]. The bubble chart, waterfall diagram, and pie chart demonstrated the CNV and mutation information of the 10 DRGs, which were derived from gene set variation analysis (GSVA; http://bioinfo.life.hust.edu.cn/GSVA).

2.4. Enrichment analysis

A relevant threshold of p < 0.05 was selected for the gene ontology (GO) terms of molecular function (MF), cell components (CC), and biological process (BP) as well as for the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment, both of which were performed using the database for annotation visualization and integrated discovery (DAVID) database.

2.5. Establishment and validation of the disulfidptosis-related model

First, DEGs in the training cohort were screened using the univariate Cox analysis. Lasso algorithm was subsequently applied using the “glmnet” statistical software package (version 4.22) to identify prognostic DEGs. Subsequently, based on the outcomes of the lasso algorithm, the model was structured utilizing multivariate Cox regression. Finally, weighted estimates according to the expression levels of each gene were used to evaluate the DRG of each sample. DRG = β₁X₁ + β₂X₂ + … + β₅ X₅. Two training and validation groups were established, which were categorized into low DRG and high DRG groups according to median DRG scores. Prognostic survival probability analysis in the training and validation groups was applied using the Kaplan–Meier survival and ROC curves.

2.6. Nomogram and univariate and multivariate Cox regression analyses of risk

The prognostic impact of factors associated with LUAD was investigated using univariate and multivariate Cox risk regression analyses. A nomogram was established utilizing the “rms” and “survivalRoc” statistical packages in R (version 4.22) to anticipate the probability of survival. Using the nomogram, the predictor score of each patient can be computed and the aggregate of the predictor scores can be used to forecast the likelihood of survival [27]. The nomogram was also evaluated using a calibration curve. Eventually, the nomogram, with a greater net advantage than other factors, was verified using DCA.

2.7. Gene set variation analysis

KEGG (https://www.kegg.jp/) and 50 Hallmarks-of-Cancer [28] pathways were utilized as background gene sets for GSVA [29] to identify differences in functioning with respect to high and low DRGs.

2.8. Correlation of DRGs with immune parameters and immunotherapy biomarkers

The TIDE algorithm, using gene expression matrixes, is a method that can evaluate tumor immune escape potency [30]. To obtain various immunological parameters, the gene expression matrices of training cohorts were uploaded to the TIDE website (http://tide.dfci).

2.9. Differences of model and clinical characteristics associated with disulfidptosis

Differences among DRGs, clinical characteristics, and immunotherapeutic biomarkers were evaluated. Differences were verified using t-test, with p < 0.05 denoting statistical significance.

2.10. Analysis of drug sensitivity using OncoPredict

Drug sensitivity can be predicted by using the R statistical package OncoPredict [31]. In the present study, sensitivity differences between drugs with high and low DRGs were assessed using OncoPredict, which can compute the IC₅₀ of 335 drugs.

2.11. Human protein atlas

The HPA (https://www.proteinatlas.org/) is billed as the biggest and most comprehensive database in the world devoted to identifying the distribution of proteins spatially expressed in human tissues and cells. It maps protein expression across tissues, cells, and organs, including both normal and tumor tissues, using data from systems biology, transcriptomics, and proteomics. It uses transcriptomics and proteomics techniques to assess how proteins are expressed at the RNA and protein levels in various human tissues and organs [32].

3. Immunohistochemistry

Immunohistochemistry was performed using a standard technique (streptavidin–biotin–peroxidase technique). Tissue slices were removed from wax, hydrated, heated with 10 mmol/L sodium ethylenediamine tetraacetic acid (EDTA; pH 9.0) in a steamer for 20 min to remove the antigen, and treated with 3 % H₂O₂-methanol solution for 10 min to decrease the endogenous peroxidase activity. After washing with phosphate-buffered saline (PBS), the slices were treated with the initial antibody surfactant protein B (SFTPB; Proteintech, 13664-1-AP, 1:30) overnight at 4 °C. In proper negative controls, Tris-buffered saline substituted the primary antibody. For all antibodies, immunoreactivity was recognized as positive in cases where >10 % alveolar epithelial cells were stained. In alveolar epithelial cells, SFTPB stains membrane and cytoplasmic expression.

3.1. Quantitative reverse transcription-polymerase chain reaction

The SYBR Green Master Mix kit was used to extract the total RNA of cells, and cDNA was prepared from the extracted RNA. Amplification was performed in 40 cycles of 5 s at 95 °C and 15 s at 60 °C after the first denaturation step of 1 min at 95 °C. The mRNA expression of SFTPB was measured utilizing the 2^−ΔΔCt relative quantification, with glyceraldehyde-3-phosphate dehydrogenase (GAPDH) as the internal reference.

3.2. Western blot

Cell culture flasks were first cleaned with PBS before adding lysis buffer (Solarbio, China) to cells. Cell lysate was collected at 4 °C and 12,500 rpm and boiled for 10 min in loading buffer. Proteins were transferred onto polyvinylidene fluoride (PVDF; 0.2 μm) membranes after solubilization on 12.5 % sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) gels and blocking with 5 % skim milk. The following primary antibodies were used: SQSTM1/p62 (1:1000; #8025; CST), GAPDH (1:5000; #5174; CST), and LC3B (1:1000; #43566; CST). The primary antibodies for LC3B and p62 were diluted as per the manufacturer's recommendations and then incubated overnight at 4 °C on a shaker at low speed. Next, secondary antibodies were added and incubated at room temperature on a slow speed shaker for 2 h. After three PBS washes, the PVDF membranes were examined using the electrochemiluminescence Plus kit to observe the immunoreactivity process. The protein levels were normalized relative to GAPDH concentration.

3.3. Knockdown SFTPB in LUAD cell lines

MISSION siRNA (Sigma Chemical Co., Ltd, St. Louis, MO) was used to prepare the target-specific small interference RNA (siRNA) of SFTPB. A negative control siRNA was obtained from the same company. The A549 cell line was derived from the Model Culture Collection of the Chinese Academy of Sciences (Shanghai, China). To silence SFTPB expression, LUAD cell lines were transfected either with the siRNA targeting SFTPB (sh-SFTPB) or negative control (NC) siRNA.

4. EDU assay

EDU experiments were performed to evaluate cell proliferation utilizing the EDU Apollo 567 kit (Solarbio, Beijing, China). After planting in 24-well plates, LUAD cells (5 × 10⁴/well) were treated with EDU and sealed with 4 % paraformaldehyde. Microscopy was used to count LUAD cells.

4.1. Clonogenic formation assay

A549 cells were transfected, cultured for 2 weeks under ideal conditions, stained, and fixed in 6-well plates for the sh-SFTPB and NC groups.

4.2. Statistics and software

Data were processed and analyzed using R software (version 4.22). The groups were compared using t-test, whereas parametric variables were correlated using Pearson's correlation. Associations between DRGs and immunological parameters and immunotherapeutic biomarkers were analyzed using Pearson's correlation analysis, and p < 0.05 was considered statistically significant.

5. Results

5.1. CNVs and 10 mutation genes related to disulfidptosis in LUAD

NCKAP1 mutation was the predominant mutation type in the 10 genes related to disulfidptosis in the training cohort, followed by LRPPRC mutation, and missense mutation was the major mutation type [Supplementary Figs. 1A and C]. NDUFA11 and SLC7A1 had more CNVs [Supplementary Fig. 1D]. The primary CNVs of NDUFA11, SLC7A1, OXSM, and GYS1 was deletion and that of NDUFS1, NCKAP1, NUBPL, and LRPPRC was amplification [Supplementary Fig. 1D]. The CNVs of 10 genes related to disulfidptosis were positively correlated with mRNA expression. Additionally, the CNVs of NUBPL exhibited the greatest association with the mRNA expression of NUBPL than other genes [Supplementary Fig. 1B].

5.2. Cluster analysis and identification of DEGs in the training cohort

The training cohort was initially categorized into two subgroups in accordance with the expression of 10 genes related to disulfidptosis [Supplementary Figs. 3A and B]. Furthermore, the prognostic analyses of 10 genes revealed that 2 genes (OXSM and SLC7A1) had a good predictive value [Supplementary Figs. 2C and D]. Then, differential gene expression analysis between the two subclusters was performed [Supplementary Table 1], identifying 83 DEGs (illustrated with volcano [Fig. 1A] and heat maps [Fig. 1B]).

Fig. 1.

Identification of differentially expressed genes in the training cohort. (A). Volcano plot was used to visualize DEGs. (B). Heatmap plot was used visualize DEGs. (Up, upregulated genes; Down, downregulated genes).

DEG enrichment pathways were determined via enrichment analysis using GO and KEGG [Supplementary Figs. 3A–D]. DEGs were prominent in pathways linked to metabolism, including glutathione metabolism, fatty acid metabolism, and lipid metabolism [Supplementary Figs. 3A–D]. This suggested clear distinctions in the material metabolism of the two subclusters.

5.3. Development and validation of a gene model related to disulfidptosis

To determine the prognostic effect of 83 genes, univariate Cox analysis was performed, identifying 42 genes [Supplementary Table 2]. The LASSO algorithm was used with minimum lambda to further assess the nine best candidate genes [Fig. 2A and B]. Five genes anillin actin-binding protein (ANLN), human leukocyte antigen class II (HLA-DPB1), alkaline phosphatase (ALPL), SFTPB, and carbamoyl phosphate synthase 1 (CPS1) were selected to be modeled with multivariate Cox analysis based on the outcomes of the LASSO algorithm [Fig. 2C] [Supplementary Table 3]. The expression level of five genes for which the model was built was then used to separate the samples into high- and low-expression groups, followed by prognosis analysis. Four of the five genes showed a significant influence on patients’ overall survival [Fig. 2E–H], whereas ALPL exhibited a poor prognostic value [Fig. 2D]. According to these findings, the survival probability of ANLN and CPS1 within the high-expression group [Fig. 2E and F] was significantly lower than that of the low-expression group, whereas the survival probability of HLA-DPB1 and SFTPB in the low-expression group [Fig. 2G and H] was markedly lower than that of the high-expression group.

Fig. 2.

Development of a disulfidptosis-related gene model. (A–B). LASSO analysis for other important genes associated with prognosis. (C). Multivariate Cox regression analysis for nine genes associated with disulfidptosis. (D–H). Survival curves of ALPL, ANLN, CPS1, HLA-DPB1, and SFTPB between the high and low expression groups.

5.4. Model demonstrated strong prognostic value with both the training and validation cohorts

LUAD specimens were categorized into high and low DRG groups based on median DRGs, which confirmed this model's predictive prognostic value [Fig. 3A]. A scatter plot revealed that the high DRG group had more deaths than the low DRG group [Fig. 3B]. Furthermore, the expression of ANLN, ALPL, and CPS1 genes was higher in the high DRG group, whereas that of SFTPB and HLA-DPB1 genes was higher in the low DRG group [Fig. 3C]. Meanwhile, a better prognostic model value was demonstrated by the survival and ROC curves [Fig. 5D and E]. Survival curves showed that the low DRG group had a substantially better rate of survival than the high DRG group [Fig. 3D]. The area under the curve (AUC) values were 0.754, 0.742, and 0.706 at 1, 2, and 3 years, respectively [Fig. 3E].

Fig. 3.

Novel model exhibited high prognostic value in the training cohort. (A). Based on DRGs, the training cohort was classified into the high and low DRG groups. (B). Scatter plot exhibiting the live and death cases in the high and low DRG groups. (C). Expression of ALPL, ANLN, CPS1, HLA-DPB1, and SFTPB between the high and low DRG groups in the training cohort. (D). Kaplan–Meier survival analysis indicated the difference in the high and low DRG groups. (E). Time-dependent ROC analysis of the model for 1, 2, and 3 years in the training cohort.

Fig. 5.

GSVA and correlation between DRGs and immune parameters. (A). Rectome pathway (B). Hallmark pathway (C). KEGG pathway was scored using the R package GSVA. (D). Correlation between DRGs and immune parameters.

GSE31210 data from the GEO database were used to validate the model's predictive accuracy [Supplementary Fig. 4]. The results of the validation cohort were identical to those of the training cohort, with AUC values of 0.717, 0.671, and 0.701 for 1-, 2-, and 3-year survival in the validation cohort, respectively. This valuable and stable predictive model for LUAD prognosis prediction was subsequently validated using these verification cohort results.

5.5. Nomogram using DRG and clinical characteristics from the training cohort

The association between DRGs and prognosis of patients with LUAD was assessed using univariate and multivariate regression analyses. Cox regression analysis revealed that DRGs were an individual influencing factor in patients with LUAD [Supplementary Fig. 5)][Supplementary Tables 4 and 5]. A nomogram was established in accordance with the DRGs and clinicopathological characteristics (including pathological staging and pathological tumor node metastasis (TNM)) to anticipate the survival possibility of 1, 2, and 3 years [Fig. 4A]. The calibration curve demonstrated a significant predictive accuracy of the nomogram at 1, 2, and 3 years [Fig. 4B]. The model's prognostic value was investigated by plotting a DCA curve. The DCA curve exhibited that the nomogram had a better net benefit at 1 [Fig. 4C], 2 [Fig. 4D], and 3 years [Fig. 4E] in patients with LUAD when compared with the other clinical characteristics.

Fig. 4.

Development of a nomogram by integrating DRGs and clinicopathological characteristics in the training cohort. (A). DRGs, pathological stage, and pathological TMN were used to construct the nomogram. (B). Calibration plot for the training cohort of the nomogram. (C–E). DCA curve exhibiting that the model had a good net benefit at 1, 2, and 3 years.

5.6. Correlation among DRGs and tumor microenvironment and immune checkpoints

GSVA enrichment revealed large functional differences in high and low DRGs [Fig. 5A–C]. The enrichment scores for the hallmark MYC targets, hallmark G2M checkpoints, and DNA replication pathways were higher in the high DRGs, whereas enrichment scores for the “reactome disease association with surfactant metabolism” and “reactome defective colony-stimulating factor 2 receptor beta subunit (CSF2RB) causes pulmonary surfactant metabolism dysfunction 5 (SMDP5)” were higher in the low DRGs group [Fig. 5A and B]. To explore the relationship with DRGs and tumor microenvironment (TME), TIDE was utilized to detect the immunological characteristics of LUAD [Fig. 5D]. As MDSCs are important mediators that promote tumor evasion from immune surveillance and foster tumor proliferation, patients with high MDSC infiltration have a worse prognostic outcome [33]. The abundance of MDSC infiltration in every sample was computed using the TIDE algorithm, which revealed a robust positive association of DRGs with MDSC infiltration abundance [Fig. 5D]. Meanwhile, the results revealed that the low DRG group benefited more from immune checkpoint blockade (ICB). Publicly available gene expression profiles, which contain the transcriptome of worn-out cluster of differentiation 8 (CD8) T cells, were used to investigate T-cell dysfunction [34], regulated T-cell activation [35], and acquired anti-ICB tumors [36]. Recent research on mouse tumor models has unveiled two phases of T-cell dysfunction [37,38]. In this study, DRG was negatively correlated with the dysfunction score [Fig. 5D], which suggested that LUAD with high DRGs had a greater capacity for immune escape. Meanwhile, Merck18 was remarkably correlated with clinical response and total survival, which supported its paralleling assessment in conventional practice and future clinical trials with other biomarkers. A negative correlation was observed between DRGs and Merck18 [Fig. 5D], suggesting that low DRGs benefited more from ICBs.

The five model gene expression and DRGs were strongly correlated with 20 immune checkpoint expression [Fig. 6A]. The results showed that most of the immunological checkpoints, including cytotoxic T lymphocyte-associated antigen-4 (CTLA4), hepatitis A virus cellular receptor 2 (HAVCR2), and T cell immunoreceptor with Ig and ITIM domain (TIGIT) [Fig. 6A], were negatively correlated with DRGs, indicating that the low DRG group benefited more from ICB treatment than the high DRG group. Furthermore, the proportion of samples with an immunotherapy response was higher in low DRGs of the TIDE algorithm than in high DRGs [Fig. 6B].

Fig. 6.

Low DRG group was strongly sensitive to ICB. (A). Correlation among DRGs and model genes and immune checkpoints. (B). Prediction of nonresponder and responder sample number in the high and low DRG groups.

5.7. Differences between DRGs and clinical characteristics

The model was correlated with clinical features to discuss the connection between the LUAD model and prognosis. The violin plot revealed a remarkable difference between DRGs in the status [Fig. 7A], M stage [Fig. 7B], N stage [Fig. 7C], and pathological stage [Fig. 7D]. The study results demonstrated that DRGs increased in patients with higher levels of pathological staging, pathological M, and pathological N according to the hierarchical analysis of clinical characteristics. Overall, the predictive model demonstrated excellent predictive value.

Fig. 7.

Correlation analysis of DRGs and different clinical characteristics. (A–D). Violin diagrams showing the difference between DRGs and status (A), Pathological-M (B), Pathological-N (C), and Pathological stage (D).

5.8. Model was relevant to both chemotherapy and targeted therapy for LUAD

The analysis of the association between the LUAD model and chemotherapy identified some effective chemotherapeutic agents. Based on the LUAD expression matrix, the drug sensitivity of 181 drugs, 13 of which were used in lung cancer, was assessed [Supplementary Table 6]. The difference in drug sensitivity between the high and low DRG groups was calculated and the consequence was visualized [Fig. 8A]. The results showed that 123 medications, including 11 drugs used to treat lung cancer, differed in both the low and high DRG groups. Of these, 11 drugs [Fig. 8B-L], including cisplatin [Fig. 8C], vinorelbine [Fig. 8L], docetaxel [Fig. 8E], and gemcitabine [Fig. 8H], were more sensitive in the high DRG group. Thus, these results may be beneficial in the individualized management of patients with LUAD.

Fig. 8.

High DRGs were strongly sensitive to chemotherapy and targeted therapy. (A). Volcano plot exhibited the difference between the two groups in drug sensitivity. (B–L). Box plot exhibited the difference in IC₅₀ of afatinib (B), cisplatin (C), dabrafenib (D), docetaxel (E), erlotinib (F), gefitinib (G), gemcitabine (H), osimertinib (I), paclitaxel (J), temozolomide (K), and vinorelbine (L) between the high and low DRG groups.

5.9. Five model-associated genes for immunohistochemistry

To determine model-related gene expression in LUAD, the expression of ANLN, ALPL, CPS1, SFTPB, and HLA-DPB1 in normal, tumor, and metastatic lung tissues was evaluated using immunohistochemistry data from the lung tissue database of HPA [Supplementary Figs. 6A–E]. The results of immunohistochemical staining revealed that ANLN was less expressed in the normal tissue and diffusely and moderately expressed in the nucleus of tumor tissue [Supplementary Fig. 6E]. Of note, the percentage of ANLN-positive cells was higher in metastatic tissues than in nonmetastatic tissues [Supplementary Fig. 6E]. The expression level of ALPL was low in the normal tissue; however, it was high within the cytoplasm and cell membrane of cancer cells [Supplementary Fig. 6A]. CPS1, located in both the cytoplasm and cell membrane, was weakly expressed in the tumor tissue and not expressed in the normal tissue [Supplementary Fig. 6B]. High SFTPB expression was noted in the cytoplasm and cell membrane of the tumor tissue, with no expression in the normal tissue [Supplementary Fig. 6D]. HLA-DPB1 expression was high in the cytoplasm and cell membrane of the tumor tissue and low in the normal tissue [Supplementary Fig. 6C].

5.10. SFTPB enhanced LUAD cell proliferation and corresponded with LUAD pathogenic characteristics

To verify SFTPB protein expression and investigate its association with the pathological characteristics of LUAD, immunohistochemical staining was performed on the collected LUAD samples. The results indicated high SFTPB expression in the tumor tissue, particularly in lymph node metastases [Fig. 9A]. Western blotting showed that SFTPB expression was elevated in LUAD and that it was greater in lymphatic metastases than in tumors [Fig. 9B]. To explore the role played by SFTPB in LUAD, SFTPB expression in A549 and type II alveolar epithelial cells (AEC II) was examined, followed by the construction of a cellular model of SFTPB deletion in LUAD cells. After transfecting LUAD cells with the plasmid, qRT-PCR and western blotting were performed to confirm that siRNA had a knockout effect on SFTPB in LUAD cells. SFTPB expression in the sh-SFTPB group was decreased compared with that in the NC-control group [Fig. 9C–E]. This result suggested that following plasmid transfection, SFTPB expression in LUAD cells was downregulated. Moreover, A549 cells treated with sh-SFTPB had a decreased ability for proliferation than A549 cells treated with NC [Fig. 9F]. Likewise, A549 cells treated with sh-SFTPB had a smaller clone size and number than those treated with NC [Fig. 9G]. This indicated that SFTPB knockdown resulted in decreased cell proliferation and that SFTPB possibly plays an important role in LUAD.

Fig. 9.

SFTPB enhanced LUAD cell proliferation and corresponded to LUAD pathogenic characteristics. (A). Immunohistochemistry of SFTPB in normal, paracancerous tissues, tumor, and lymphatic metastases of the LUAD tissue. (B) Western blotting demonstrated difference in the expression of SFTPB in normal, tumor, and lymphatic metastases. (C). SFTPB expression was verified in A549 and AEC II cells using qRT-PCR. (D). Expression of SFTPB knockdown in A549 cells. (E). Expression of SFTPB was verified through qRT-PCR. (F). Representative EDU staining (red) of SFTPB in A549 cells treated with sh-SFTPB or NC. Nucleus marked with DAPI (blue). (G). Cell proliferation capacity of A549 cells treated with NC-control and sh-SFTPB was measured using the clonogenic formation assay. Note bene: an uncropped version of the original Fig. 9B–D is shown in Supplementary Fig. 7A.

6. Discussion

LUAD tumors are highly heterogeneous, as demonstrated by thousands of different mutations in each cancer genome concerning single nucleotide variants, CNVs, gene fusions, and main chromosomal events [39,40]. Targeted therapies against these genetic aberrations have been developed, which makes the molecular characterization of a tumor a requisite to achieve success in designing a targeted therapy with the most effective strategy. Although there is hope for LUAD treatment owing to the exploration of novel drugs that target key disease-driving mutations, only a minority of patients with LUAD have benefited from these therapies. This is because of resistance to primary and secondary drugs and because the clinically effective targeting of all other mutated genes has not yet been achieved [41]. Identification of mechanisms of immune resistance to cancer may contribute to the development of individualized immunotherapies. This requires the search for new treatment strategies. In this context, disulfidptosis may become a hot topic in future research. Numerous studies have indicated that targeted therapy for disulfidptosis is possibly a promising new treatment for LUAD. Disulfide metabolism is closely associated with the modulation of the redox state, an essential mechanism for tumorigenesis and development [42]. For instance, certain tumor cells are known to extend their survival and tolerance by modifying redox environments within the cells [43]. Additionally, disulfides are potentially useful in cancer treatment. Some antitumor drugs such as cisplatin and paclitaxel exhibit their antitumor effects through reaction with intracellular disulfides [44,45]. In the present study, the biological role, prognostic potential, and relationship of 10 DRGs with immunotherapy, TIDE, chemotherapy, and resistance to targeted therapy in LUAD were assessed from multiple viewpoints. First, mRNA expression, CNV, and mutation of the 10 genes associated with disulfidptosis in LUAD were analyzed. Multiomics data analysis revealed that the 10 DRGs may upregulate the mRNA expression of related genes through higher CNV, thereby contributing to LUAD development. High CNV was also exhibited by the genes in LUAD, and all CNVs were positively associated with mRNA expression. The 10 DRGs were then included in a consistent cluster analysis, which enabled the separation of data into cluster A and cluster B. The DEGs of the two clusters were then analyzed, producing 83 DEGs. Next, GO and KEEG enrichment studies based on the 83 DEGs were performed. The genes were selected using univariate Cox regression, LASSO, and multivariate Cox analysis, yielding five DEGs. Based on these five DEGs, a model was established to assess LUAD prognosis and guide treatment. The model exhibited good prognostic ability in the training and validation cohorts.

The protein-coding gene ANLN is present on chromosome 7q14.2. It codes for an 1125-amino-acid protein, which plays an essential role in many biological processes, including cell growth, division, mitosis, and migration, as well as in human disorders [46]. ANLN participates in the regulation of tumor cell growth, invasion, and metastasis through reciprocal associations with cytoskeletal components [47]. In various tumor types, including breast cancer, lung cancer [48,49], pancreatic cancer [50,51], liver cancer [52], bladder cancer [53], and colorectal cancer [54], ANLN is overexpressed [55]. In addition, it deteriorates outcomes by facilitating the progress of malignant tumors. One of the crucial proteins for cell division, ANLN associates with the RHO Family GTPases (GTP-Rho), actin, and cytokinin and participates in cell multiplication, differentiation, adhesion, migration, apoptosis, and cycling [56]. The dysregulation of cell division because of aberrant ANLN expression is crucial for cancer development, growth, and metastasis. ANLN overexpression in bladder cancer was confirmed to facilitate tumor cell proliferation, migration, and infiltration. The modulation of ANLN protein expression by siRNA suppresses tumor cell growth and migration that causes tumor cell apoptosis [57]. Furthermore, the fact that ANLN is a crucial regulatory factor for stimulating PI3K/Akt signaling indicates the indirect involvement of ANLN in apoptosis induction [58]. HLA consists of a cell surface αβ-heterodimer of two immunoglobulin chains, of which only one carries an allelic variant. The polymorphic A chain (heavy chain) of HLA class I is encoded by the A, B, and C loci on the short arm of chromosome 6 on the major histocompatibility complex (MHC), whereas the nonpolymorphic B chain (light chain) is encoded on the long arm of chromosome 15 (i.e., outside the MHC). The broad variety of polymorphisms in the HLA region is generally acknowledged to be the outcome of selective pressure induced by the functional significance of HLA molecules in the immune response. ALPL is indispensable in the regulation of intracellular signaling and protein activity [[59], [60], [61]]. Rao et al. reported that ALPL knockdown can reduce prostate cancer cell migration [62]. Another research showed that ALPL overexpression suppressed high-grade serous ovarian cancer (HGSOC) cell migration and invasion [63]. These studies indicated that the effect of ALPL and its mechanism in cancer are controversial. ALPL regulates the mechanism underlying the molecular biology of the invasion and metastasis of LUAD by influencing the c-Myc/RhoA cascade. This discovery offers a rationale for the target-based treatment of clinical LUAD [64]. Pathogenic deviations in genes that encode the vital elements of lung surfactants such as surfactant protein B (SP–B, SFTPB gene), surfactant protein C (SP–C, SFTPC gene), and ATP-binding cassette transporter A3 are the main root cause of both infant respiratory distress syndrome and childhood interstitial lung disease [65]. It is known that surfactant proteins B and C have a role in lowering surficial stress and that surfactant proteins A and D have an important role in innate immunity [66]. In multiple tumor types, low SFTPB expression is correlated with unfavorable prognosis [67,68]. N-acetylglutamate allosterically activates CPS1, which produces carbamoylphosphate from bicarbonate, ammonia, and ATP. Biallelic mutations in CPS1 that manifest in neonates cause hyperammonemia and decreased citrulline [69]. Meanwhile, a significant decrease in metabolic CPS1 expression occurs in weakly differentiated hepatic cancer cell lines [70]. Additionally, CPS1 overexpression was correlated with two adverse treatment responses in patients with colorectal cancer treated with neoadjuvant synchronized radiotherapy [71]. Owing to its potential stimulation of the Janus kinase (JAK)-signal transducer and activator of transcription (STAT) signaling pathway, CPS1 may be carcinogenic in terms of LUAD biology [72]. Generally, the activation of the JAK-STAT signaling system promotes the growth of tumors by controlling inflammation, invasion, and immunosuppression [73], which offered an idea on how to explore CPS1 as a potential LUAD biomarker. The present study results revealed that the survival rate was significantly lower in the HLA-DPB1 and SFTPB low-expression groups but significantly higher in the ANLN and CPS1 low-expression groups. Moreover, DRGs were an independent predictor of prognosis in patients with LUAD. Additionally, the total survival was lower in the high DRG group than in the low DRG group. The prognostic value of the model was subsequently assessed using ROC and DCA curves. In the training cohort, the model predicted 1-, 2-, and 3-year overall survival with sensitivities of 0.754, 0.742, and 0.706, respectively. Moreover, a stable prognostic value was obtained in the validation cohort. DRGs offered a higher net benefit to patients than other factors. In summary, these findings suggest that the model of five genes related to disulfidptosis is of a strong prognostic value for patients with LUAD.

The TME plays a role in the promotion or inhibition of tumor development. Various immune cells are known to infiltrate TME by regulating tumor immunity. Thus, TME influences tumor cell growth [[74], [75], [76]]. TME, which comprises lymphocytes, cancer cells, mesenchymal stromal cells, and cell-secreted cytokines, has an important role in tumor immunosuppression. Inhibitory receptor expression mediates the intrinsic control of tumor immunosuppression, whereas intercellular interaction and/or inhibitory factor secretion mediates extrinsic control [77]. In certain malignant tumors, the existence of T cells is correlated with a better outcome [78]. There is flexibility for macrophages to progress to distinct phenotypes (such as M1-like and M2-like macrophages) in response to various environmental stimulus signals. Certain tumor-associated macrophages (TAMs) have antitumor properties and display an M1-like macrophage phenotype, particularly in the initial phases of tumorigenesis. Nevertheless, most TAMs manifest an M2-like macrophage phenotype mediated by tumor reprogramming of TAMs, which is crucial in facilitating tumor progression [79,80]. The infiltration of M2 within the TME is related to resistance and adverse outcomes. M2 macrophages are engaged in lung cancer invasion and metastasis by promoting epithelial–mesenchymal transition [81]. A reduction in polarized M2 macrophages restricts lung cancer growth and proliferation and enhances responsiveness to immunotherapy and chemotherapy [82]. Regulatory T cells (Tregs) serve a complex “dual-edged sword” function in tumors. Tregs, a subset of CD4⁺ T cells, regulate excessive inflammation and are crucial for maintaining immunological homeostasis [[83], [84], [85]]. However, the immunomodulatory qualities of Tregs can accelerate cancer development. Moreover, during the tumorigenesis process, Tregs can suppress the body's antitumor response, which can facilitate tumor cells' immunological escape [86]. Patients with bladder cancer who have a higher level of Treg infiltration have a better prognosis. High T-cell exclusion and a high TIDE score indicate a high degree of immunological escape [87]. The current study results revealed that patients with minimal MDSC infiltration benefited more from ICB, in line with the findings of other research, when DRGs were correlated with clinical features, immunological microenvironment, immunotherapy, and medication sensitivity. The immunological features of the high and low DRG groups were then estimated, which revealed that LUAD dysfunction scores were higher in the low DRG group than in the high DRG group. This meant that the higher the DRGs, the greater the immune escape capacity. In terms of biological function, high DRGs exhibited increased hallmark MYC targets, hallmark G2M checkpoints, and DNA replication activity. Moreover, based on the hierarchical analysis of clinical features, DRGs were higher in patients at a higher pathological stage and pathological TNM stage. Higher tumor malignancy was indicated in patients with LUAD in the high DRG group. These findings indicated that ICB may be more beneficial for patients with LUAD in the low DRG group. A linkage analysis revealed that DRGs were negatively correlated with the expression of eight immune checkpoints and positively correlated with the expression of two immune checkpoints. These findings indicated that patients with LUAD who have low DRGs might have a better chance of treatment success with ICB. These results are in line with the finding that a higher proportion of patients with low DRGs responded to ICB treatment. Additionally, the high DRG group was more responsive to a few frequently prescribed chemotherapy drugs such as cisplatin, gemcitabine, vincristine, docetaxel, and other targeted therapeutic agents, when compared with the low DRG group. These findings demonstrated that in cases with poor immunotherapy response, cisplatin and other chemotherapy and targeted therapy drugs may be considered for patients with LUAD having high DRGs. By contrast, in patients with LUAD having low DRGs, immune therapy or immunotherapy, in combination with chemotherapy, can be an alternative, potentially leading to better prognosis. However, it is important to know that this needs to be validated and confirmed in clinical practice.

Model-related genes strongly correlated with prognosis in patients with LUAD, suggesting their possible involvement in LUAD development. The immunohistochemistry results of five genes were obtained from the HPA database to validate the model. Immunohistochemical data of the lung tissue from the HPA database were utilized to validate the expression of ANLN, ALPL, CPS1, SFTPB, and HLA-DPB1 in normal, malignant, and metastatic states to determine the expression of model-related genes in LUAD. As demonstrated by immunohistochemical staining, ANLN expression was diffusely moderate in tumor tissue nuclei but low in normal tissues. The percentage of ANLN cells was significantly higher in metastatic tissues than in nonmetastatic tissues. ALPL was weakly expressed in normal tissues but moderately expressed in the cytoplasm and cell membrane of cancer cells, whereas CPS1 expression was weak in the cytoplasm and cell membrane of tumor tissues but negative in normal tissues. High SFTPB expression in the tumor tissues was observed in the cytoplasm and cell membrane, whereas its expression was negative in the normal tissue. Finally, HLA-DPB1 expression was high in tumor tissues, situated in both the cytoplasm and cell membrane but low in normal tissues. The function and role of SFTPB in LUAD were subsequently validated. SFTPB expression was higher in lymphatic metastases than in tumors. SFTPB knockdown revealed a decreased proliferative ability in the sh-SFTPB group. The exact mechanism, however, is unknown and requires further research.

Despite the prominent strength of disulfidptosis-associated prognostic markers for recognizing immune landscapes and anticipating prognosis, some limitations need to be recognized and should be addressed with subsequent studies. In the current study, data from public databases were analyzed, which may have introduced bias in the prediction results from the actual situation. While some approaches were implemented to minimize this limitation, more data from patients with LUAD are needed to validate the utility of the model and accuracy of the immunotherapy predictions. Furthermore, significant basic and clinical research is needed to validate these findings.

Taken together, the mechanisms regulating the tumor immunity, TME, and prognosis of LUAD were identified to establish a new model that accurately predicted the prognosis of patients with LUAD through a comprehensive analysis of disulfidptosis-associated genes. The model's therapeutic capability with respect to chemotherapy, targeted therapy, and immunotherapy was also identified, which offered novel concepts to guide individualized therapeutic tactics in patients with LUAD.

Ethics statement

Review and/or approval by an ethics committee was not needed for this study because de-identified data analysis from TCGA and GEO databases does not require institutional review board approval and informed consent.

Funding

This research was supported by Henan Provincial Medical Science and Technology Program (No. LHGJ20190567) and (No. LHGJ20200582).

Consent for publication

The manuscript has been read in its final version by all authors and agreed upon.

Data and material availability

The original data of this study will be shared by authors with no reservation.

CRediT authorship contribution statement

Changshuan Li: Writing – review & editing, Writing – original draft. Chao Sun: Investigation. Yakun Li: Investigation. Lin Dong: Investigation. Xian Wang: Investigation. Ruixin Li: Visualization, Software, Formal analysis. Junjie Su: Visualization. Qiong Cao: Validation, Methodology. Shiyong Xin: Supervision, Funding acquisition, Conceptualization.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

List of abbreviations

DRGs

disulfidptosis-related genes

LUAD

lung adenocarcinoma

CNVs

copy number variations

DEGs

differentially expressed genes

ROC

receiver operating characteristic

G2M

second gap mitosis

E2F

E2 promoter binding factor

MYC

myelocytomatosis

MDSCs

myeloid-derived suppressor cells

ICB

immune checkpoint blockade

qRT-PCR

quantitative real-time polymerase chain reaction

EDU

5-ethynyl-2′-deoxyuridine

NSCLC

non-small cell lung cancer

EGFR

epidermal growth factor receptor

KRAS

Kirsten rat sarcoma viral oncogene homolog

BRAF

B-Raf proto-oncogene, serine/threonine kinase gene

ERBB2

tyrosine kinase receptor 2

TP53

tumor protein p53 gene

ALK

anaplastic lymphoma kinase

STK11

serine/threonine kinase 11

TTF1

thyroid transcription factor 1

ATP

adenosine triphosphate

SLC7A11

solute carrier family 7 member 11

lncRNAs

long non-coding RNAs

TCGA

the cancer genome atlas

GEO

gene expression omnibus

mRNA

messenger RNA

DCA

decision curve analysis

TIDE

tumor immune dysfunction and exclusion

IHC

immunohistochemistry

GYS1

glycogen synthase 1

LRPPRC

leucine rich pentatricopeptide repeat containing

NCKAP1

NCK associated protein 1

NDUFA11

NADH:ubiquinone oxidoreductase subunit A11

NDUFS1

NADH:ubiquinone oxidoreductase core subunit S1

NUBPL

nucleotide binding protein-like

OXSM

mitochondrial 3-oxoacyl-ACP synthase

RPN1

ribophorin I

SLC3A2

solute carrier family 3 member 2

SLC7A1

solute carrier family 7 member 1

GSVA

gene set variation analysis

GO

gene ontology

MF

molecular function

CC

cell components

BP

biological process

KEGG

Kyoto Encyclopedia of Genes and Genomes

DAVID

database for annotation visualization and integrated discovery

HPA

Human Protein Atlas

EDTA

ethylenediamine tetraacetic acid

PBS

phosphate buffered saline

SFTPB

surfactant protein B

GAPDH

glyceraldehyde-3-phosphate dehydrogenase

PVDF

polyvinylidene fluoride

SDS-PAGE

sodium dodecyl sulfate-polyacrylamide gel electrophoresis

siRNA

small interfering RNA

NC

negative control

ANLN

anillin actin-binding protein

HLA-DPB1

human leukocyte antigen class II

ALPL

alkaline phosphatase

CPS1

carbamoyl phosphate synthase 1

AUC

area under the curve

TNM

tumor node metastasis

CSF2RB

colony-stimulating factor 2 receptor beta subunit

SMDP5

pulmonary surfactant metabolism dysfunction 5

TME

tumor microenvironment

CD8

cluster of differentiation

CTLA4

cytotoxic T lymphocyte-associated antigen-4

HAVCR2

hepatitis A virus cellular receptor 2

TIGIT

T cell immunoreceptor with Ig and ITIM domain

AEC II

type II alveolar epithelial cells

GTP-Rho

the RHO Family GTPases

MHC

major histocompatibility complex

HGSOC

high-grade serous ovarian cancer

JAK

Janus kinase

STAT

signal transducer and activator of transcription

TAMs

tumour-associated macrophages

Appendix A. Supplementary data

The following are the Supplementary data to this article.

figs1.

figs2.

figs3.

figs4.

figs5.

figs6.

figs7.