Machine learning-based model identifies a novel cuproptosis-related mitochondrial gene signature with a key role in the prognosis and treatment of lung adenocarcinoma

Abstract

Lung adenocarcinoma (LUAD) remains one of the most prevalent and lethal cancers globally, making it critical to understand the mechanisms driving its progression and improve prognosis. Moreover, cuproptosis and mitochondrial dysfunction may be involved in lung cancer. Therefore, the present study aimed to identify mitochondrial genes associated with cuproptosis to develop a prognostic model for patients with LUAD, with the potential to predict survival outcomes and responses to treatment. Bulk RNA-sequencing data was utilized from The Cancer Genome Atlas and the Gene Expression Omnibus (GEO), and Pearson correlation analysis was employed to identify mitochondrial genes associated with cuproptosis. A prognostic model was constructed using univariate Cox regression combined with least absolute shrinkage and selection operator analysis, and a nomogram was developed to predict survival with clinical relevance. The accuracy of the model was evaluated using two independent GEO datasets. Additionally, the clinical value of the risk score model was assessed using immune infiltration analysis, tumor mutational burden and drug sensitivity predictions. Furthermore, the effects of superoxide dismutase 2 (SOD2) gene knockdown on tumor metastasis and proliferation were experimentally evaluated. A set of 22 mitochondrial genes associated with cuproptosis were identified: Metabolism of cobalamin associated D, SOD2, human immunodeficiency virus-1 Tat interactive protein 2, cytochrome C somatic, mitochondrial pyruvate carrier 1, adenylate kinase 2, mitochondrial ribosomal protein L44, transforming growth factor β regulator 4, mitochondrial transcription factor A, tetratricopeptide repeat domain 19, coiled-coil-helix-coiled-coil-helix domain containing 4, sideroflexin 1, ATP binding cassette subfamily D member 1, NADH:ubiquinone oxidoreductase complex assembly factor 7, NOP2/Sun RNA methyltransferase 4, NME/NM23 nucleoside diphosphate kinase 6, X-Prolyl aminopeptidase 3, lipoyltransferase 1, mitochondrial methionyl aminopeptidase type 1D, carbonic anhydrase 5B, kynurenine 3-monooxygenase and alcohol dehydrogenase iron containing 1. The model was validated as an independent predictor of overall survival, dividing patients into high- and low-risk groups. Immune infiltration analysis revealed that tumors in the low-risk group displayed more active immune responses and improved immune function. Drug sensitivity analysis suggested that high-risk patients may be more responsive to specific drug treatments. Finally, knockdown of the SOD2 gene suppressed tumor cell metabolism, proliferation and metastasis. In conclusion, the present study successfully established a prognostic model based on cuproptosis-related mitochondrial genes and developed a nomogram to predict LUAD prognosis with high accuracy, thereby providing improved tools for treatment decision-making and enhancing patient outcomes.

Keywords: cuproptosis; immune infiltration; lung adenocarcinoma; machine learning; mitochondrial; prognostic model; superoxide dismutase 2.

Figure 1.

Identification of cuproptosis-related mitochondrial genes in lung adenocarcinoma. (A) Sankey diagram showing the positive correlation (>0.3) between 13 cuproptosis-related genes and mitochondrial-related genes based on Pearson correlation analysis. (B) Volcano plot displaying all differentially expressed genes. (C) Venn diagram showing the intersection of 415 genes obtained from differentially expressed genes and cuproptosis-related mitochondrial genes. (D) Forest plot demonstrating the 67 cuproptosis-related mitochondrial genes identified through univariate Cox regression analysis. (E) Chromosomal localization of these 67 genes. FC, fold change; DEG, differentially expressed gene; Cup-mt, cuproptosis-related mitochondrial genes; HR, hazard ratio; CI, confidence interval.

Figure 2.

Machine learning-based prognostic risk score model construction. (A) C-index for each machine learning prediction model calculated for the training and test sets, with >100 models included. (B) LASSO regression analysis established a model containing cuproptosis-related mitochondrial genes associated with prognosis. (C) Coefficients of the LASSO analysis. (D) Distribution of risk scores, (E) survival status and time distribution in high- and low-risk groups. (F) Kaplan-Meier curves showing the OS of patients in high- and low-risk groups in TCGA training set. (G) Distribution of risk scores, (H) survival status and time distribution in high- and low-risk groups, and (I) Kaplan-Meier curves showing the OS of patients in high- and low-risk groups in the GSE26939 test set. (J) Distribution of risk scores, (K) survival status and time distribution in high- and low-risk groups, and (L) Kaplan-Meier curves showing the overall OS of patients in high- and low-risk groups in the GSE31210 test set. (M) Distribution of risk scores, (N) survival status and time distribution in high- and low-risk groups, and (O) Kaplan-Meier curves showing the overall OS of patients in high- and low-risk groups in the GSE72094 test set. LASSO, least absolute shrinkage and selection operator; TCGA, The Cancer Genome Atlas; OS, overall survival.

Figure 3.

Risk scores and survival curves for different subgroups. (A) Risk score boxplot by survival status. Kaplan-Meier curves for high- and low-risk groups in patients aged (B) ≥65 years and (C) <65 years. (D) Risk score boxplot by sex. Kaplan-Meier curves for high- and low-risk groups in (E) female and (F) male patients. (G) Risk score boxplot by clinical stages M0 and M1. Kaplan-Meier curves for high- and low-risk groups of patients with clinical stage (H) M0 and (I) M1. (J) Risk score boxplot by clinical stages N0 and N1-3. Kaplan-Meier curves for high- and low-risk groups in patients with clinical stage (K) N0 and (L) N1-3. (M) Risk score boxplot by stage T1-2 and T3-4. Kaplan-Meier curves for high- and low-risk groups of patients with clinical stage (N) T1-2 and (O) T3-4. M, metastasis; N, node; T, tumor.

Figure 4.

Nomogram construction to elucidate lung adenocarcinoma prognosis. (A) Univariate and (B) multivariate Cox regression analyses for clinical features and risk scores. (C) Nomogram based on clinical features and risk scores. Calibration curves showing the accuracy of predicted and actual values in (D) TCGA training set, (E) the GSE26939 test set, (F) the GSE31210 test set and (G) the GSE72094 test set. Receiver operating characteristic curves evaluating the performance of the nomogram in (H) TCGA training set, (I) the GSE26939 test sets, (J) the GSE31210 test set and (K) the GSE72094 test set. TCGA, The Cancer Genome Atlas; T, tumor; N, node; OS, overall survival; FPR, false-positive rate; TPR, true-positive rate; AUC, area under the curve.

Figure 5.

Immune cell infiltration in high- and low-risk groups. (A) Expression of 23 immune cell subtypes between high- and low-risk groups. (B) Distribution and expression of immune-related pathways in high- and low-risk groups. (C) Expression of checkpoint-related genes in high- and low-risk groups. (D) Correlation heatmap of risk scores and model genes with immune cells and pathways. *P<0.05; **P<0.01; ***P<0.001. ns, not significant. TNFRSF, tumor necrosis factor receptor superfamily; LAG3, Lymphocyte activation gene 3; PDCD1LG2, Programmed cell death 1 ligand 2; CD200R1, CD200 receptor 1; BTLA, B and T lymphocyte attenuator; IDO2, indoleamine 2,3-dioxygenase 2; BTNL2, Butyrophilin like 2; ADORA2A, Adenosine a2a receptor.

Figure 6.

Immune scores and expression of different immune gene families. (A) ESTIMATE scores. Boxplot of the expression of (B) MHC gene families, (C) inflammatory cytokine gene families and (D) cytotoxic molecule-related gene families between high- and low-risk groups. (E) Difference in TIDE scores between high- and low-risk groups. (F) Proportion of ‘No benefits’ and ‘Responder’ between high- and low-risk groups. *P<0.05; **P<0.01; ***P<0.001. ns, not significant; MHC, major histocompatibility complex; TIDE, Tumor Immune Dysfunction and Exclusion.

Figure 7.

Relationship between TMB, risk scores and gene mutations in different risk groups. (A) Comparison of TMB between high- and low-risk groups. Kaplan-Meier survival curve for overall survival stratified by (B) high and low TMB grouping, and (C) TMB (high or low) and risk score (high or low). Mutation spectrum of common gene mutations in the (D) high-risk group and (E) low-risk group. TMB, tumor mutational burden; H-TMB, high TMB; L-TMB, low TMB; H-RISK, high risk; L-RISK, low risk.

Figure 8.

Correlation between prognostic model and drug sensitivity prediction. (A) Correlation between risk score, model genes and drugs. Blue indicates a negative correlation, whilst orange denotes a positive correlation. (B) Boxplot showing the difference in IC₅₀ of A.770041 between high- and low-risk groups. (C) Scatter plot showing the correlation between risk score and A.770041. (D) Boxplot showing the difference in IC₅₀ of CGP.082996 between high- and low-risk groups. (E) Scatter plot showing the correlation between risk score and CGP.082996. (F) Boxplot showing the difference in IC₅₀ of Obatoclax.Mesylate between high- and low-risk groups. (G) Scatter plot showing the correlation between risk score and Obatoclax.Mesylate. (H) Boxplot showing the difference in IC₅₀ of SL.0101.1 between high- and low-risk groups. (I) Scatter plot showing the correlation between risk score and SL.0101.1. (J) Boxplot showing the difference in IC₅₀ of Thapsigargin between high- and low-risk groups. (K) Scatter plot showing the correlation between risk score and Thapsigargin. (L) Boxplot showing the difference in IC₅₀ of WZ.1.84 between high- and low-risk groups. (M) Scatter plot showing the correlation between risk score and WZ.1.84. *P<0.05; **P<0.01; ***P<0.001.

Figure 9.

Expression levels and biological functions of SOD2 in lung adenocarcinoma cell lines. (A) mRNA levels of SOD2 in the NC and knockdown groups in A549 and PC9 cell lines. Cell Counting Kit-8 analysis showing the effects of SOD2 knockdown on the (B) A549 and (C) PC9 cell lines. Comparison of (D) migration, (E) invasion and (F) proliferation in A549 and PC9 cells between NC and SOD2 knockdown groups. Comparison of mitochondrial fluorescence between the si-NC and si-SOD2 groups in (G) A549 and (H) PC9 cells. ****P<0.0001. SOD2, superoxide dismutase 2; NC, negative control; si, small interfering; ns, not significant OD, optical density.