Prognostic and Immunological Analysis of Disulfidoptosis-Related Ferroptosis Genes in Lung Adenocarcinoma

doi:10.21203/rs.3.rs-5324542/v1

Download PDF

Research Article

Prognostic and Immunological Analysis of Disulfidoptosis-Related Ferroptosis Genes in Lung Adenocarcinoma

https://doi.org/10.21203/rs.3.rs-5324542/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Disulfidoptosis, a newly identified form of regulated cell death (RCD), significantly influences the progression of lung adenocarcinoma (LUAD). This study identified 340 ferroptosis-related genes strongly correlated with disulfidoptosis through correlation analysis. By intersecting these genes with those from module genes selected via weighted gene co-expression network analysis (WGCNA), 31 genes were found. In the TCGA-LUAD cohort, Kaplan-Meier(KM) survival analysis initially screened these genes, leading to the selection of 6 Disulfidoptosis-Related Ferroptosis (DRF) genes (IL33, SLC2A1, CDCA3, KIF20A, FANCD2, RRM2) through further screening with Random Forest and SVM-RFE. Based on the expression levels of DRF genes, two distinct groups with differing prognostic and immune characteristics were identified. A machine learning-driven signature (MLS) of 24 DRF-related genes was then constructed using the RSF + SuperPC algorithm and validated in the TCGA-LUAD and GSE31210 datasets. Compared with 77 other signatures, MLS demonstrated superior performance in both datasets. A low MLS score was associated with immune activation, higher tumor mutation burden, and better survival probability. Conversely, a high MLS score correlated with poorer prognosis and reduced potential benefit from immune therapy, although treatments like Doramapimod might still offer benefits. The cell cycle pathway was a key factor distinguishing high from low MLS groups. Overall, MLS shows promise for predicting prognosis in LUAD patients and identifying those who might benefit from immune therapy. Additionally, DRF genes have potential clinical value for diagnosing and treating other cancers, as indicated by pan-cancer analysis. q-PCR experiments targeting select DRF genes confirmed their feasibility as diagnostic markers for LUAD.

disulfidoptosis

ferroptosis

lung adenocarcinoma

machine learning

immune therapy

prognosis

drug sensitivity

Lung cancer is the second most common cancer worldwide and the leading cause of cancer-related deaths. LUAD is the most prevalent histological subtype, accounting for more than 40% of all lung cancer cases^1,2. TNM staging has long been used as an important predictor of lung cancer prognosis³. However, studies have indicated that tumor heterogeneity may lead to prognostic differences among patients with the same stage of lung cancer⁴. Additionally, with the increasing use of immunotherapy for lung cancer, identifying patients who are more likely to benefit from such treatment has become a critical clinical challenge⁵. Anatomical TNM staging alone is not ideal for screening patients who are likely to respond to immunotherapy. Therefore, a comprehensive understanding of the molecular mechanisms underlying LUAD and the identification of new biomarkers for predicting long-term prognosis and response to immunotherapy are crucial for improving preventive strategies and effective interventions^6,7.

Recent studies have shown that cells overexpressing SLC7A11 can undergo rapid cell death due to SLC7A11-mediated depletion of NADPH and intracellular accumulation of toxic cystine, induced by disulfide stress. This process, known as disulfidoptosis, is a newly identified form of regulated cell death (RCD) distinct from known forms such as ferroptosis, necroptosis, and apoptosis. Disulfidoptosis typically occurs under glucose starvation conditions⁸. Regulation of disulfide proteases involves the formation and cleavage of disulfide bonds and the involvement of proteins such as NCKAP1, as well as redox-related signaling pathways. Disulfide proteases have potential as targets for cancer therapy. Inhibiting glucose transporter protein (GLUT) can have therapeutic effects on SLC7A11-overexpressing cancer cells by inducing disulfide proteases. Understanding the regulatory mechanisms and significance of disulfide proteases can provide insights into cellular homeostasis and offer potential strategies for targeted cancer therapy. Ferroptosis is an iron-dependent RCD characterized by excessive lipid peroxidation in the cell membrane, which inhibits antioxidant responses. Key features of ferroptosis include the oxidation of polyunsaturated fatty acids in phospholipids, the presence of redox-active iron, and the loss of lipid peroxidation repair mechanisms. GPX4, a crucial regulator of ferroptosis, inhibits its onset by reducing the formation of phospholipid hydroperoxides. Down-regulation of SLC7A11 triggers ferroptosis by impairing cysteine metabolism pathways^9–11. Ferroptosis plays a significant role in various diseases, including organ damage, degenerative diseases, and tumors^12,13. Moreover, therapeutic resistance in certain cancers is linked to overexpression of GPX4 and reduced expression of SLC7A11¹⁴. Although disulfidoptosis and ferroptosis are distinct forms of cell death, SLC7A11 can act as a common regulator of both, influencing intracellular iron levels and regulating ferroptosis resistance, and plays a vital role in the regulation of disulfidoptosis in lung cancer^10,15,16.

In this study, we identified and validated DRF genes using bioinformatics analysis. Based on different aggregated signatures, we developed a new machine learning-based score, the MLS, for LUAD patients. This score performed well compared to 77 published signatures. We also conducted immunological and drug sensitivity analyses to explore immune differences and identify potentially effective drugs among different patient subgroups. Our study contributes to a better understanding of disulfidoptosis-associated ferroptosis genes in LUAD development and may provide guidance for prognostic prediction and personalized therapy for various cancers, including LUAD.

Data acquisition and processing

RNA-Seq data on gene expression for 33 cancers, including LUAD, along with corresponding clinical data, were downloaded from TCGA (The Cancer Genome Atlas) database (https://portal.gdc.cancer.gov/). Similar data from the validation dataset, GSE31210, were acquired from the GEO database (Gene Expression Omnibus ;https://www.ncbi.nlm.nih.gov/geo/). Somatic mutation data for TCGA-LUAD were acquired from the GDC (Genomic Data Commons; https://portal.gdc.cancer.gov/), and copy number variation (CNV) data were downloaded from UCSC Xena (https://xenabrowser.net). Tumor Immuno-Dysfunction and Rejection (TIDE) data for non-small cell lung cancer were obtained from the TIDE website (http://tide.DRFci.harvard.edu/).

Identifcation of DRF genes

Twenty-three disulfidoptosis-related genes were identified from relevant studies^17,18. The 564 ferroptosis-associated genes were obtained from the FerrDb V2 database (http://www.zhounan.org/). The WGCNA R package was utilized to construct co-expression modules and clarify the relationships between these modules. Pearson correlation analysis was employed to screen for genes highly associated with disulfidoptosis and ferroptosis genes (|cor| > 0.3, p < 0.05), identifying those module genes significantly correlated with LUAD. The intersection of these gene sets was determined. Kaplan-Meier survival analyses were conducted to identify genes more strongly associated with prognosis (p < 0.05). Hyperparameter tuning of ten machine learning algorithms (Classification Tree, Glmnet, KNN, LDA, Logistic, Naive Bayes, NNET, Random Forest, SVM, Xgboost) was performed to select the best screening algorithm. The frequency of CNV of the finalized Disulfidoptosis-Related Ferroptosis (DRF) genes is presented in a bar chart, with their chromosomal positions shown using the "RCircos" package in R. Differential and survival analyses were conducted to validate these genes in the GSE31210 cohort, and the CIBERSORT algorithm was used to analyze the relationship between DRF genes and 20 immune cell types. Genome-wide genomic data for 33 cancers, including DRF gene expression in pan-cancers, survival analyses, and interlinkages, were obtained from TCGA. To better illustrate the intrinsic relationship between DRF and disulfidoptosis genes, the concept of "module" from the WGCNA algorithm was applied, with DRF and disulfidoptosis considered as a module. The association between the expression values of DRF/disulfidoptosis regulators and module signature genes was assessed based on intra-module connectivity. DRF/disulfidoptosis was defined as a regulator with a module membership (MM) greater than 0.75. For each cancer type, the pooled expression levels of the identified DRF/disulfidoptosis regulators were calculated as epigenetic module signature genes (EMEs).

Clinical and immunological characterisation and enrichment analysis of DRF subgroups

Consensus cluster analysis of LUAD samples in the TCGA dataset was performed based on DRF gene expression using the "ConsensusClusterPlus" package. Gene expression and clinical data, such as age, gender, and TNM stage, were visualized between subgroups using the "pheatmap" package. Overall survival (OS) times were then compared between these subgroups.

Subsequently, the "LIMMA" R package was used to identify differentially expressed genes (DEGs) associated with Disulfidoptosis-Related Ferroptosis (DRF) groupings. DEGs were adjusted to a significance level of p < 0.05 with |log2FoldChange (FC)| > 1.2. The "ClusterProfiler" package and "org.Hs.eg.db" were employed to perform GO (Gene Ontology) and KEGG (Kyoto Encyclopedia of Genes and Genomes) enrichment analyses based on the identified DEGs. The ssGSEA package and GSVA (Gene Set Variation Analysis) method were used to compute clustering values for immune-related pathways and immune cells to perform functional and pathway enrichment analyses on the subgroups, with results presented in heatmaps. The package of "estimate" was used to calculate 23 immune-related functional scores.

Quantitative real-time polymerase chain reaction (qRT-PCR)

Normal lung epithelial cells (BEAS-2B) and the human lung adenocarcinoma cell line (A549) were obtained from the Shandong University of Traditional Chinese Medicine. A fluorescence quantitative PCR instrument was acquired from Roche (Basel, Switzerland). Total RNA was extracted using TRNzol Universal reagent (Beijing, China). Complementary DNA (cDNA) was synthesized using the First Strand Synthesis Kit (KR116). The relative expression of genes was analyzed using the 2 − ΔΔCT method with actin as an internal reference gene and normal lung epithelial cells as the control.

MLS prognostic model construction assessment and enrichment analysis

In the TCGA-LUAD training cohort, the KM method was used to identify genes from DEGs significantly associated with OS (p < 0.001). Genes expressed in both TCGA-LUAD and GSE31210 were selected for inclusion in the integrated framework for constructing the MLS. Consistent models were developed based on 101 algorithm combinations in both the TCGA-LUAD training cohort and the GSE31210 validation cohort, and the average C-index of all models in both cohorts was calculated to assess predictive performance (Fig. 3A). MLS scores for each sample in all cohorts were computed using the formula: risk_score = Σ(Expi * Coefi), where Coefi and Expi represent the risk factor and expression of each gene, respectively. Samples were categorized into low risk (score < median) and high risk (score > median) groups based on the median risk score. To illustrate the differences between high and low risk groups, we first depicted the process from DRF cluster grouping to risk grouping to final outcomes using Mulberry diagrams. Subsequently, MLS gene expression between high and low risk groups, as well as MLS scores of DRF subgroups, were analyzed. Finally, the "clusterSUR" package was used for survival analysis between high and low risk groups.

To progressively validate the accuracy of the model, we conducted a systematic literature search that led to the inclusion of 77 published prognostic signatures for LUAD. These signatures were associated with features such as immunity, copper-induced cell death, and iron-dependent cell death. We then compared the MLS to these published signatures using the TCGA-LUAD and GSE31210 datasets. Subsequently, Gene Set Enrichment Analysis (GSEA) was used to analyze the pathways in high and low scoring groups.

Immunoassay

Six immune infiltration algorithms were employed to assess dissimilarities in immune cells between two scoring subgroups. Additionally, the TIDE (Tumor Immune Dysfunction and Exclusion) algorithm predicted the efficacy of anti-tumor immune drugs (anti-PD1, anti-CTLA4) based on TIDE scores, which were positively correlated with drug efficacy¹⁹. TIDE scores were also used to evaluate patient responses to immunotherapy. In the analysis of Tumor Mutational Burden (TMB), the maftools package was utilized to examine the connection between risk scores and TMB. The package of "survival survminer" was used to investigate the impact of TMB on LUAD prognosis, with a p-value < 0.05 considered statistically significant.

Drug screening

To evaluate the sensitivity of high and low scoring MLS groups to different drugs, the oncoPredict²⁰ software package was used to calculate IC50 values for 198 drugs from the GDSC2 database. This analysis aimed to identify potentially effective drugs for patients who are in the high and low scoring groups, with a significance threshold of p < e-10.

Identification and pan-cancer analysis of the DRF gene

Firstly, 340 ferroptosis-related genes associated with disulfidoptosis were identified through correlation studies. Using the "limma" package, TCGA-LUAD samples were categorized into tumor and normal tissues. Co-expression modules were constructed with the WGCNA package. Correlation analyses assessed the relationships between these modules and tumor versus normal tissues, leading to the identification of 13 modules with the optimal soft threshold power of 3 (Fig. 1A, B). Among these modules, 8 were significantly correlated with tumors (P < 0.05), with 5 showing positive correlations and 3 showing negative correlations (Fig. 1C). To identify relevant hub genes, we focused on the MEblue and MEbrown modules, which had the most significant positive and negative correlations (Pearson correlation coefficients of 0.52 and − 0.82, respectively; Fig. 1D, E). From these modules, 1,437 related genes were selected, with 613 genes from the MEblue module. After intersecting these genes with the disulfidoptosis ferroptosis (DRF) genes, 31 genes were obtained.

Kaplan-Meier survival analysis was used to screen for 10 candidate genes with strong prognostic relevance. SVM and Random Forest algorithms were then applied for hub gene screening (Fig. 2A). The intersection of these two algorithms narrowed the 10 candidate genes down to 6 (IL33, SLC2A1, CDCA3, KIF20A, FANCD2, RRM2; Supplementary Table 1, 2). These six DRF genes demonstrated favorable receiver operating characteristic (ROC) values in both the TCGA and GSE31210 cohorts (Supplementary Fig. 1), leading to their identification as DRF genes for LUAD. Subsequent survival and differential analyses for the GSE31210 cohort validated the reliability of these genes (Supplementary Figs. 2, 3).

The interrelationships among DRF genes in the genome-wide data for 33 cancers are illustrated in Fig. 2B. Expression analysis revealed differential expression in 18 cancer types, indicating that IL33 could act as a protective factor for most tumors, while the other five genes appeared as risk factors (Fig. 2C). Survival analysis results suggested that DRF genes can differentiate survival times across multiple cancers (Supplementary Table 3, P < 0.05). Notably, WGCNA was used to identify central regulators involved in the connectivity of DRF and disulfidoptosis regulators across 33 cancer types (Fig. 2D). The analysis showed a strong correlation among central DRF regulators (R = 0.65, P = 4.7e-05; Fig. 2E). In vitro experiments demonstrated significantly higher expression of CDCA3, KIF20A, FANCD2, RRM2, and SLC2A1 in the NSCLC (A549) cell line compared to normal lung epithelial cells (BEAS-2B), with statistically significant differences (Fig. 3), confirming the potential of DRF genes as tumor detection markers.

Mutational landscape and immune profile of the DRF gene

The chromosomal locations of DRF genes are shown in Fig. 4A. The study revealed that CNV deletions were more prevalent in IL33, CDCA3, and KIF20A, while SLC2A1, RRM2, and FANCD2 were more frequently amplified (Fig. 4B). These analyses demonstrated significant genetic and expression heterogeneity of DRF in lung adenocarcinoma, suggesting that imbalanced DRF expression plays a critical role in LUAD development. CIBERSORT analysis indicated that IL33 was positively correlated with various immune-activated cells and negatively correlated with immunosuppressive cells, such as regulatory T cells (Tregs) (Fig. 4C).

Clinical and immunological analysis of DRF subgroups

To further investigate the clinical value and functional biological patterns of these DRFs, we performed clustering analyses on all tumor samples in the TCGA LUAD cohort based on the expression levels of the six DRFs. We compared the expression of DRF-related genes with LUAD subtypes to explore their relationship. Setting the clustering value (K) from 2 to 10, the best aggregation stability was found at K = 2 (Fig. 4D). Consequently, the TCGA LUAD cohort was divided into cluster C1 (n = 347) and cluster C2 (n = 238) based on DRF expression. Heatmaps were used to display gene expression and TNM typing, staging, age (≤ 65 or > 65 years), and gender for the 585 LUAD samples. The IL33 gene expression level was significantly lower in the C1 group compared to the C2 group and was more closely associated with poor prognostic indicators (e.g., higher TNM staging and grading, older age), consistent with previous analyses (Fig. 4E). Kaplan-Meier analysis revealed that the C2 group had a more favorable OS compared to the C1 group, which showed a survival disadvantage consistent with clinical parameters (P = 0.003, Fig. 4F).

Given the importance of tumor immunity in tumor development, we explored the level of cellular infiltration in the tumor microenvironment. ssGSEA scores indicated that the C2 group had richer infiltration of immune cells, such as immature B cells and natural killer(NK) cells. The poorer clinical outcome in the C1 group may be related to immunosuppression induced by a Type 2 T helper cell (Th2) subpopulation in CD4 + T cells²¹ (Fig. 5A). The TME (tumor microenvironment), which includes tumor cells, mesenchymal stromal cells, immune cells, cytokines, and chemokines, was assessed. Figure 5B shows that the C2 group had higher StromalScore, ImmunityScore, and ESTIMATEScore, associated with lower tumor purity and better immune infiltration. This suggests that the C2 group may be linked to an immune microenvironment that promotes tumor death and is likely more sensitive to immunotherapy²². Analysis of the immune environment of the subgroups (Fig. 5C) revealed that Type_II_IFN_Response was more prevalent in C2, indicating characteristics of the immune response. Parainflammation, which is crucial for maintaining homeostasis, was more prominent in C1, potentially driven by p53 mutations, which can promote cancer. Given the sensitivity of parainflammation to NSAIDs, aspirin may offer better therapeutic effects for patients in the C1 group ²³.

To further investigate the pathways dominated by DRFs, we conducted GSVA. Figure 5D illustrates that KEGG analysis revealed a significant enrichment of inflammatory metabolism-related pathways in the C2 group. In contrast, tumorigenic pathways, such as cell cycle and amino acid metabolism, were significantly enriched in the C1 group, which may be related to the previously mentioned parainflammation.

Functional and pathway enrichment analysis of DRF groupings

Using the limma package, 320 DEGs were identified from 37,992 genes, and these DEGs were used for subsequent enrichment analyses. GO analysis indicated that DRF-related genes were primarily associated with microtubule binding, extracellular matrix, and structural constituents in the cellular component (CC). These genes were related to condensed chromosomes, centromeric regions, and chromosomal regions. In terms of biological processes (BP), they were associated with nuclear chromosome segregation and mitotic nuclear division (Fig. 6A-D). KEGG pathway analysis highlighted the cell cycle and p53 signaling pathway as the primary DRF-enriched pathways. The p53 signaling pathway regulates cellular processes such as metabolism, antioxidant defense, and ferroptosis²⁴ (Fig. 6E-G).

Validation of MLS risk model construction

For sample selection, we chose 572 samples with available outcomes and non-zero survival times from TCGA-LUAD as the training group. Using the Kaplan-Meier method, 42 genes significantly associated with OS (p < 0.001) were identified from the 320 DEGs. Among these, 40 genes that were expressed in both TCGA-LUAD and GSE31210 were selected to be included in the integrated framework for constructing the MLS risk model.

In the TCGA-LUAD training cohort, consistent models were constructed using 101 algorithm combinations. The average C-index for each model across all cohorts was calculated to evaluate their predictive ability, with the GSE31210 cohort used for model validation (Fig. 7A). Ultimately, the RSF + SuperPC algorithm, which achieved the highest average C-index, was identified as the most valuable model. This model was developed using 24 hub genes (FAM83A, KRT6A, SLC2A1, RHOV, ANGPTL4, METTL7A, GAPDH, ECT2, LYPD3, GJB2, NAPSA, CYP4B1, LOXL2, CTSH, SFTPB, TMPRSS2, BTG2, ANLN, SLC34A2, CYP2B7P, PFKP, UBE2S, C16orf89, KPNA2). Samples with MLS scores above the mean (n = 286) were classified as the high-risk group, while the remaining samples were categorized as the low-MLS group. All patients in the high-MLS group exhibited a poorer prognosis (Fig. 7B). The expression levels of the 24 genes showed statistical significance in both high and low scoring groups (Fig. 7C). The Mulberry diagram indicated that the C2 group, which comprised a larger portion of the low-MLS group, was associated with better survival outcomes, with a statistically significant difference in scores between the C1 and C2 groups (Fig. 7D, E). Furthermore, to comprehensively assess the predictive power of MLS, we included 77 different features in the study. MLS demonstrated a superior C-index in both TCGA-LUAD and GSE31210 datasets (12/78, 10/78; Fig. 7F), thereby confirming the reliability of the MLS model.

MLS immunological profile

The CIBERSORT algorithm was used to analyze the correlation between the 24 MLS genes, MLS scores, and various immune cells. The results indicated that most MLS genes and MLS scores were negatively correlated with immune-activated cells and positively correlated with immune-suppressed cells. Additional analyses using five other immune infiltration algorithms also showed that immune cell infiltration (including dendritic cells, eosinophils, monocytes, and NK cells) was significantly higher in patients with low MLS scores compared to those with high MLS scores, suggesting an immune activation state (Fig. 8A, B). This implies that LUAD with low MLS levels may be classified as "hot tumors." Conversely, T regulatory cells (Tregs) and tumor-associated fibroblasts (CAFs) were predominantly enriched in patients with high MLS scores, which likely contribute to tumor proliferation, invasion, drug resistance, and an immunosuppressive state, leading to poorer outcomes (Fig. 6B, C). This suggests that LUAD with high MLS levels might be classified as "cold tumors," which are less sensitive to immunotherapy. TIDE scores were negatively correlated with risk scores (Fig. 8C), indicating that low-risk patients may benefit more from immunotherapy. Increased TMB expression is associated with enhanced T cell activation. TMB has potential clinical utility in lung cancer and could serve as a biomarker for immunotherapy²⁵. Our survival analyses showed that patients with lower MLS scores and higher TMB levels typically had significantly prolonged OS, suggesting that MLS could complement TMB in assessing patient prognosis (Fig. 8D, E).

GSEA analysis of MLS

GSEA analysis for subgroups (Fig. 8F) revealed that pathways enriched in low-scoring subgroups were associated with immune cell regulation and GnRH receptor signaling. These pathways, in addition to their role in regulating pituitary gonadotropin secretion, may predict survival and resistance to tumor proliferation, metastasis, and anti-angiogenesis in lung cancer patients²⁶. The Hedgehog signaling pathway, utilized in developing tumor-targeting drugs, and VEGFR inhibitors, known for their effects on vascular smooth muscle and cancer cell proliferation, were also identified²⁷. Interestingly, long-term depression was also enriched, suggesting a potential link between these pathways and tumor biology.

Drug sensitivity analysis of MLS

The drug sensitivity analysis identified nine drugs with varying effectiveness. The low-scoring subgroup samples showed greater sensitivity to drugs such as BMS-754807, Doramapimod, JQ1, OF-1, and SB505124. Conversely, the high-scoring subgroup samples were more sensitive to AZD7762, Dasatinib, Docetaxel, and WIKI4. These findings suggest that MLS may be useful in evaluating the efficacy of targeted therapies and chemotherapy (Fig. 9A-I).

LUAD, the most common type of lung cancer, is often diagnosed at an advanced stage and is highly aggressive. Despite significant advancements in the diagnosis and treatment of lung cancer, new biomarkers and genetic features have improved the ability to predict LUAD prognosis. The development of targeted therapies, immunotherapies, and other novel treatments has benefited patients considerably²⁸. However, due to high resistance to conventional radiotherapy and the heterogeneous nature of LUAD, the prognosis remains poor, with a 5-year survival rate of only 15%²⁹. Comprehensive analysis of patients' multi-omic data enhances our understanding of disease mechanisms, and identifying biomarkers and therapeutic strategies from these data will support clinical precision and personalized medicine.

Recent studies have shown that SLC7A11 can transport cystine in response to glucose starvation, leading to disulfide stress, which disrupts the redox balance and causes disulfide bond death. This process, known as disulfidptosis, is characterized by abnormal formation of disulfide bonds in actin backbone proteins and the breakdown of F-actin^8,30. Similar to ferroptosis, disulfidptosis can induce tumor cell death by altering the structure of cytoskeletal proteins. Elevated SLC7A11 expression has been observed in renal cancer tissues, where it is associated with tumor progression. Disulfidptosis induced by SLC7A11 contributes to cell death in glucose-deficient environments. Notably, SLC7A11 is a critical component of the upstream signaling pathway regulating ferroptosis, a form of cell death driven by oxidative stress and membrane lipid peroxidation, among other factors¹³. Specific genes in LUAD, such as STYK1 and LSH, play significant roles in regulating ferroptosis^31,32. Recent research indicates that cell death occurs through complex and interdependent processes, involving extensive cellular interactions^33,34. Thus, balancing ferroptosis and disulfidptosis could be a novel strategy to enhance outcomes and survival in LUAD patients. A recent study used four disulfidptosis/ferroptosis-related genes to predict prognosis and immune infiltration levels in LUAD patients³⁵. In our study, we established a prognostic signature consisting of 24 DRF-related genes from the TCGA-LUAD cohort based on 6 newly identified DRF diagnostic genes and validated it in the GSE31210 dataset.

In this study, we applied WGCNA along with Pearson correlation analysis, differential expression analysis, and KM survival analysis to selected DRF-related genes. While the LASSO algorithm is valuable for variable selection, it may sometimes favor less relevant predictors when genes are highly correlated with each other³⁶. To address this issue, we compared 10 machine learning methods to identify the most important genes. Evaluation metrics included AUC, classification error rate, accuracy, precision, recall, sensitivity, and specificity. SVM-REF (support vector machine - recursive feature elimination) is a machine learning method that enhances classification accuracy and model robustness by removing less relevant features from the SVM-generated feature vector. Random Forest, another supervised machine learning algorithm, constructs and integrates multiple decision trees to improve prediction performance and stability.

Among the six disulfidptosis-related ferroptosis genes identified, IL-33, a member of the IL-1 family, is closely linked to lung cancer progression. IL-33 enhances the activity of dendritic cells (DCs), CD8 + T cells, and NK cells, thereby inhibiting tumor growth and metastasis³⁷. SLC2A1 expression is significantly elevated in lung cancer and correlates with poorer patient survival³⁸. CDCA3 overexpression may be linked to the activation of various oncogenic signaling pathways and cancer progression, and correlates with sensitivity to platinum-based chemotherapy. Similarly, high KIF20A expression is associated with the activation of oncogenic pathways and poorer OS^39,40. FANCD2 is recognized as a cancer susceptibility gene that coordinates DNA repair pathways and responds to DNA damage through the PI3K-Akt-mTOR signaling pathway, promoting tumor cell survival⁴¹. RRM2 is involved in tumorigenesis and resistance to chemotherapeutic agents^42,43. DRF genes also show potential for aiding in the early diagnosis and treatment of other cancers. This research provides new insights into the mechanisms linking disulfidptosis and ferroptosis. Specifically, KIF20A and SLC2A1, as DRF genes, demonstrate potential for use in the diagnosis and treatment of hepatocellular carcinoma⁴⁴.

To explore the regulatory roles and immunological mechanisms of DRFs in tumor development, our study classified DRFs into two groups using cluster analysis. We examined the differences between DRF subgroups in terms of genomic, transcriptional, and immunological features, and their connections in LUAD. Investigating the potential links between disulfidptosis and ferroptosis can provide deeper insights into mechanisms such as anti-tumor immune activity and inhibition of tumor immune escape in the TME, and help guide the selection of effective immunotherapy regimens. Notably, the two subgroups showed significant differences in OS, TME, and immune infiltration levels. Group C2 was associated with high IL33 expression, immune activation, and better prognosis, indicating an immunoinflammatory phenotype. In contrast, Group C1 exhibited a lower estimate score and higher expression of immunosuppressive cells, which aligns with the immune desert subtype⁴⁵. Additionally, the C1 group had higher TH2/TH1 ratios compared to the C2 group. TH2 cells, a subset of CD4 + T cells, secrete cytokines such as IL-4, IL-5⁴⁶. TH1 cells, also a CD4 subset, primarily produce IFN-γ, IL-2, and TNF, and are involved in cellular immunity and the differentiation of cytotoxic T cells⁴⁷. TH1 and TH2 cells maintain a balance through cytokine secretion, but this balance can be disrupted in the tumor microenvironment, leading to a shift from a Th1 to a Th2 response, known as the "Th1/Th2 shift"⁴⁸. This shift is observed in many tumor patients, including those with lung cancer, and may be linked to tumor immune escape⁴⁹.

Machine learning algorithms are effective tools for analyzing multi-omics data. To understand the molecular differences between prognostic subtypes and enhance clinical utility, we used the TCGA-LUAD dataset as a training set and selected the best MLS through 101 algorithm combinations to address algorithm selection limitations. Overfitting is a common issue in model construction, where a model performs well in the training set but poorly in the validation set ⁽⁶⁴⁾. To mitigate this, we used the C-index of the validation cohort as a ranking criterion. Although the RSF + StepCox[forward] combination showed excellent performance in the training set, it was less effective in the validation set. In contrast, MLS demonstrated better prognostic value across all cohorts compared to other published signatures. Other signatures that performed better in the training set often failed to match this performance in the validation set, likely due to model overfitting.

We analyzed the enrichment of numerous immune-related signatures between high and low score groups. We found that several oncogenic pathways were significantly activated in the high MLS group. This group also exhibited a high expression of Tregs cells, which may contribute to the immunosuppressive phenotype and a cold tumor environment. In contrast, the low MLS group had a higher TMB and a greater variety of immune cell types, suggesting stronger anti-tumor immunity and better prognosis. Survival analyses indicated a more favorable outcome for this group. Additionally, TIDE results showed that the low MLS group responded more effectively to immunotherapy, supporting the notion that MLS could aid in the early identification of immunotherapy-sensitive patients. Interestingly, while macrophages M2 are generally associated with anti-inflammatory responses and are more involved in angiogenesis, immunosuppression, and tumor metastasis⁵⁰, our study observed the opposite. This discrepancy may be due to the heterogeneity among different M2 subpopulations, highlighting the complexity of macrophages in tumor immunity⁵¹. Despite the poor prognosis and limited benefit from immunotherapy in the high MLS group, Dasatinib and Docetaxel may still offer potential therapeutic benefits for this population.

Our study has several limitations. First, the prognostic model was developed using retrospective data from public databases. Although the results indicate the potential of DRF scores in predicting LUAD prognosis, their clinical value needs to be validated with additional prospective real-world data. Second, the mechanisms underlying the relationship between disulfidptosis and ferroptosis require further investigation. Additionally, while bioinformatics analyses provided insights into the potential impact of DRF scores on immune cell infiltration, experimental validation is needed to confirm the correlation between DRF scores and immune activity.

In conclusion, we identified six genes related to disulfidoptosis and ferroptosis, and validated these DRF genes through in vitro experiments. Cluster analysis revealed two molecular subtypes of LUAD, highlighting significant differences in prognosis and immunological profiles between them. We developed a MLS that demonstrated strong performance across multiple cohorts, indicating its robust ability to predict patient prognosis. Comprehensive analyses, including functional, immunological, and pharmacological sensitivity assessments, confirmed the reliability and potential clinical utility of the DRF score for future precision medicine and clinical evaluation.

Ethical Approval

Not applicable

Disclosure

TCGA and GEO belong to public databases. The patients involved in the database have obtained ethical approval. Users can download relevant data for free for research and publish relevant articles. Our study is based on open-source data, so there are no ethical issues and other conflicts of interest.

Competing interests

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest

Author’s contributions

LZC: Data analysis, Writing- Original draft preparation. YRD: Writing- Original draft preparation. ZX and LYX collected, processed and analyzed the data. ZX: Conception and design, Final approval of the version to be published.

Funding

This work was supported by the Chinese Medicine Inheritance and Innovation "Hundred Million" Talent Project (QI Huang Project) QI Huang Scholars Project, Shandong Province Taishan Scholars Specially Appointed Experts Program Project, and the 2021 Shandong Province Natural Science Foundation - General Project [No. ZR202103040386].

Data availability statement

The original contributions presented in the study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding author.

Cappa C, Gregson K, Wardlaw T, et al. Birth registration: a child′s passport to protection[J]. Lancet Glob Health, 2014, 2(2): e67⁃68. DOI: 10.1016/S2214⁃109X(13)70180⁃3.
Barta, J. A., Powell, C. A. & Wisnivesky, J. P. Global Epidemiology of Lung Cancer. Annals of global health 85, doi:10.5334/aogh.2419 (2019).
Sung, H. et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin 71, 209-249, doi:10.3322/caac.21660 (2021).
Woodard, G. A., Jones, K. D. & Jablons, D. M. Lung Cancer Staging and Prognosis. Cancer treatment and research 170, 47-75, doi:10.1007/978-3-319-40389-2_3 (2016).
Hensing, T., Chawla, A., Batra, R. & Salgia, R. A personalized treatment for lung cancer: molecular pathways, targeted therapies, and genomic characterization. Advances in experimental medicine and biology 799, 85-117, doi:10.1007/978-1-4614-8778-4_5 (2014).
Wu, M. et al. Systemic Immune Activation and Responses of Irradiation to Different Metastatic Sites Combined With Immunotherapy in Advanced Non-Small Cell Lung Cancer. Frontiers in immunology 12, 803247, doi:10.3389/fimmu.2021.803247 (2021).
Pan, S. et al. Identification of cuproptosis-related subtypes in lung adenocarcinoma and its potential significance. Frontiers in pharmacology 13, 934722, doi:10.3389/fphar.2022.934722 (2022).
Wang, Y. et al. Molecular subtypes based on cuproptosis-related genes and immune profiles in lung adenocarcinoma. Frontiers in genetics 13, 1006938, doi:10.3389/fgene.2022.1006938 (2022).
Liu, X. et al. Actin cytoskeleton vulnerability to disulfide stress mediates disulfidptosis. Nature cell biology 25, 404-414, doi:10.1038/s41556-023-01091-2 (2023).
Yang, J. et al. Metformin induces Ferroptosis by inhibiting UFMylation of SLC7A11 in breast cancer. Journal of experimental & clinical cancer research : CR 40, 206, doi:10.1186/s13046-021-02012-7 (2021).
Wang, X. et al. Stem Cell Factor SOX2 Confers Ferroptosis Resistance in Lung Cancer via Upregulation of SLC7A11. Cancer research 81, 5217-5229, doi:10.1158/0008-5472.Can-21-0567 (2021).
Wu, S. et al. The role of ferroptosis in lung cancer. Biomarker research 9, 82, doi:10.1186/s40364-021-00338-0 (2021).
Jiang, X., Stockwell, B. R. & Conrad, M. Ferroptosis: mechanisms, biology and role in disease. Nature reviews. Molecular cell biology 22, 266-282, doi:10.1038/s41580-020-00324-8 (2021).
Chen, X., Kang, R., Kroemer, G. & Tang, D. Broadening horizons: the role of ferroptosis in cancer. Nature reviews. Clinical oncology 18, 280-296, doi:10.1038/s41571-020-00462-0 (2021).
Cao, L. et al. Managing ferroptosis-related diseases with indirect dietary modulators of ferroptosis. The Journal of nutritional biochemistry 120, 109427, doi:10.1016/j.jnutbio.2023.109427 (2023).
Koppula, P., Zhuang, L. & Gan, B. Cystine transporter SLC7A11/xCT in cancer: ferroptosis, nutrient dependency, and cancer therapy. Protein & cell 12, 599-620, doi:10.1007/s13238-020-00789-5 (2021).
Zhang, W. et al. RBMS1 regulates lung cancer ferroptosis through translational control of SLC7A11. The Journal of clinical investigation 131, doi:10.1172/jci152067 (2021).
Zhao, S. et al. Crosstalk of disulfidptosis-related subtypes, establishment of a prognostic signature and immune infiltration characteristics in bladder cancer based on a machine learning survival framework. Frontiers in endocrinology 14, 1180404, doi:10.3389/fendo.2023.1180404 (2023).
Wang, T. et al. Disulfidptosis classification of hepatocellular carcinoma reveals correlation with clinical prognosis and immune profile. International immunopharmacology 120, 110368, doi:10.1016/j.intimp.2023.110368 (2023).
Jiang, P. et al. Signatures of T cell dysfunction and exclusion predict cancer immunotherapy response. Nat Med 24, 1550-1558, doi:10.1038/s41591-018-0136-1 (2018).
Maeser, D., Gruener, R. F. & Huang, R. S. oncoPredict: an R package for predicting in vivo or cancer patient drug response and biomarkers from cell line screening data. Briefings in bioinformatics 22, doi:10.1093/bib/bbab260 (2021).
Golubovskaya, V. & Wu, L. Different Subsets of T Cells, Memory, Effector Functions, and CAR-T Immunotherapy. Cancers 8, doi:10.3390/cancers8030036 (2016).
Bejarano, L., Jordāo, M. J. C. & Joyce, J. A. Therapeutic Targeting of the Tumor Microenvironment. Cancer discovery 11, 933-959, doi:10.1158/2159-8290.Cd-20-1808 (2021).
Lasry, A., Aran, D., Butte, A. J. & Ben-Neriah, Y. Cancer Cell-Autonomous Parainflammation Mimics Immune Cell Infiltration. Cancer research 77, 3740-3744, doi:10.1158/0008-5472.Can-16-3383 (2017).
Liu, J., Zhang, C., Wang, J., Hu, W. & Feng, Z. The Regulation of Ferroptosis by Tumor Suppressor p53 and its Pathway. International journal of molecular sciences 21, doi:10.3390/ijms21218387 (2020).
Hu, C. et al. Genomic profiles and their associations with TMB, PD-L1 expression, and immune cell infiltration landscapes in synchronous multiple primary lung cancers. Journal for immunotherapy of cancer 9, doi:10.1136/jitc-2021-003773 (2021).
Wu, Y. et al. Potentially functional variants of HBEGF and ITPR3 in GnRH signaling pathway genes predict survival of non-small cell lung cancer patients. Translational research : the journal of laboratory and clinical medicine 233, 92-103, doi:10.1016/j.trsl.2020.12.009 (2021).
Ran, F. et al. Inhibition of Vascular Smooth Muscle and Cancer Cell Proliferation by New VEGFR Inhibitors and Their Immunomodulator Effect: Design, Synthesis, and Biological Evaluation. Oxidative medicine and cellular longevity 2021, 8321400, doi:10.1155/2021/8321400 (2021).
Spella, M. & Stathopoulos, G. T. Immune Resistance in Lung Adenocarcinoma. Cancers 13, doi:10.3390/cancers13030384 (2021).
Seguin, L., Durandy, M. & Feral, C. C. Lung Adenocarcinoma Tumor Origin: A Guide for Personalized Medicine. Cancers 14, doi:10.3390/cancers14071759 (2022).
Zhong, Z. et al. NFATc1-mediated expression of SLC7A11 drives sensitivity to TXNRD1 inhibitors in osteoclast precursors. Redox biology 63, 102711, doi:10.1016/j.redox.2023.102711 (2023).
Ma, Z. et al. STYK1 promotes tumor growth and metastasis by reducing SPINT2/HAI-2 expression in non-small cell lung cancer. Cell death & disease 10, 435, doi:10.1038/s41419-019-1659-1 (2019).
Yang, R. et al. LSH interacts with and stabilizes GINS4 transcript that promotes tumourigenesis in non-small cell lung cancer. Journal of experimental & clinical cancer research : CR 38, 280, doi:10.1186/s13046-019-1276-y (2019).
Deng, W. et al. The application of nanoparticles-based ferroptosis, pyroptosis and autophagy in cancer immunotherapy. Journal of nanobiotechnology 22, 97, doi:10.1186/s12951-024-02297-8 (2024).
Meng, Q., Ding, B., Ma, P. & Lin, J. Interrelation between Programmed Cell Death and Immunogenic Cell Death: Take Antitumor Nanodrug as an Example. Small methods 7, e2201406, doi:10.1002/smtd.202201406 (2023).
Ma, X. et al. Leveraging a disulfidptosis/ferroptosis-based signature to predict the prognosis of lung adenocarcinoma. Cancer cell international 23, 267, doi:10.1186/s12935-023-03125-z (2023).
Pavlou, M., Ambler, G., Seaman, S., De Iorio, M. & Omar, R. Z. Review and evaluation of penalised regression methods for risk prediction in low-dimensional data with few events. Statistics in medicine 35, 1159-1177, doi:10.1002/sim.6782 (2016).
Yang, K., Tian, C., Zhang, C. & Xiang, M. The Controversial Role of IL-33 in Lung Cancer. Frontiers in immunology 13, 897356, doi:10.3389/fimmu.2022.897356 (2022).
Yi, J. et al. A genome-wide comprehensive analysis of alterations in driver genes in non-small-cell lung cancer. Anti-cancer drugs 29, 10-18, doi:10.1097/cad.0000000000000571 (2018).
Sahin, K. B. et al. Elevating CDCA3 Levels Enhances Tyrosine Kinase Inhibitor Sensitivity in TKI-Resistant EGFR Mutant Non-Small-Cell Lung Cancer. Cancers 13, doi:10.3390/cancers13184651 (2021).
Sun, D., Zhang, H., Zhang, C. & Wang, L. An evaluation of KIF20A as a prognostic factor and therapeutic target for lung adenocarcinoma using integrated bioinformatics analysis. Frontiers in bioengineering and biotechnology 10, 993820, doi:10.3389/fbioe.2022.993820 (2022).
Bhattacharjee, S. & Nandi, S. DNA damage response and cancer therapeutics through the lens of the Fanconi Anemia DNA repair pathway. Cell communication and signaling : CCS 15, 41, doi:10.1186/s12964-017-0195-9 (2017).
Zuo, Z. et al. Ribonucleotide reductase M2 (RRM2): Regulation, function and targeting strategy in human cancer. Genes & diseases 11, 218-233, doi:10.1016/j.gendis.2022.11.022 (2024).
Furuta, E., Okuda, H., Kobayashi, A. & Watabe, K. Metabolic genes in cancer: their roles in tumor progression and clinical implications. Biochimica et biophysica acta 1805, 141-152, doi:10.1016/j.bbcan.2010.01.005 (2010).
Zhang, C. et al. Development and experimental validation of a machine learning-based disulfidptosis-related ferroptosis score for hepatocellular carcinoma. Apoptosis : an international journal on programmed cell death 29, 103-120, doi:10.1007/s10495-023-01900-x (2024).
Hornburg, M. et al. Single-cell dissection of cellular components and interactions shaping the tumor immune phenotypes in ovarian cancer. Cancer cell 39, 928-944.e926, doi:10.1016/j.ccell.2021.04.004 (2021).
Walker, J. A. & McKenzie, A. N. J. T(H)2 cell development and function. Nature reviews. Immunology 18, 121-133, doi:10.1038/nri.2017.118 (2018).
Xu, H. M. Th1 cytokine-based immunotherapy for cancer. Hepatobiliary & pancreatic diseases international : HBPD INT 13, 482-494, doi:10.1016/s1499-3872(14)60305-2 (2014).
Li, J. et al. Immunological modulation of the Th1/Th2 shift by ionizing radiation in tumors (Review). International journal of oncology 59, doi:10.3892/ijo.2021.5230 (2021).
Anichini, A., Perotti, V. E., Sgambelluri, F. & Mortarini, R. Immune Escape Mechanisms in Non Small Cell Lung Cancer. Cancers 12, doi:10.3390/cancers12123605 (2020).
Yunna, C., Mengru, H., Lei, W. & Weidong, C. Macrophage M1/M2 polarization. European journal of pharmacology 877, 173090, doi:10.1016/j.ejphar.2020.173090 (2020).
Li, M. et al. Metabolism, metabolites, and macrophages in cancer. Journal of hematology & oncology 16, 80, doi:10.1186/s13045-023-01478-6 (2023).

No competing interests reported.

Download PDF

Reviewers invited by journal
28 Oct, 2024
Editor assigned by journal
25 Oct, 2024
Submission checks completed at journal
25 Oct, 2024
First submitted to journal
24 Oct, 2024

You are reading this latest preprint version

Prognostic and Immunological Analysis of Disulfidoptosis-Related Ferroptosis Genes in Lung Adenocarcinoma

Status:

Version 1

Abstract

Figures

Introduction

Materials and methods

Results

Discussion

Conclusion

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1