Significance of cuproptosis‐related lncRNA signature in LUAD prognosis and immunotherapy: A machine learning approach

Abstract

Objective

Cuproptosis‐related genes are closely related to lung adenocarcinoma (LUAD), which can be analyzed via the analysis of long noncoding RNA (lncRNA). To date, the clinical significance and function of cuproptosis‐related lncRNAs are still not well elucidated. Further analysis of cuproptosis‐related prognostic lncRNAs is of great significance for the treatment, diagnosis, and prognosis of LUAD.

Methods

In this study, a multiple machine learning (ML)‐based computational approach was proposed for the identification of the cuproptosis‐related lncRNAs signature (CRlncSig) via comprehensive analysis of cuproptosis, lncRNAs, and clinical characteristics. The proposed approach integrated multiple ML algorithms (least absolute shrinkage and selection operator regression analysis, univariate and multivariate Cox regression) to effectively identify the CRlncSig.

Results

Based on the proposed approach, the CRlncSig was identified from the 3450 cuproptosis‐related lncRNAs, which consist of 13 lncRNAs (CDKN2A‐DT, FAM66C, FAM83A‐AS1, AL359232.1, FRMD6‐AS1, AC027237.4, AC023090.1, AL157888.1, AL627443.3, AC026355.2, AC008957.1, AP000346.1, and GLIS2‐AS1).

Conclusions

The CRlncSig could well predict the prognosis of different LUAD patients, which is different from other clinical features. Moreover, the CRlncSig was proved to be an effective indicator of patient survival via functional characterization analysis, which is relevant to cancer progression and immune infiltration. Furthermore, the results of RT‐PCR assay indicated that the expression level of FAM83A‐AS1 and AC026355.2 in A549 and H1975 cells (LUAD) was significantly higher than that in BEAS‐2B cells (normal lung epithelial).

The data preprocessing was performed in the Cancer Genome Atlas (TCGA)‐lung adenocarcinoma (LUAD)to obtain all long noncoding RNAs (lncRNAs) and cuproptosis‐related gene data, and Pearson correlation analysis was performed to obtain cuproptosis‐related lncRNAs. Combining cuproptosis‐lncRNAs with Futime (survival or censoring time) and Fustat (censoring status)for multistage machine learning, the data were subjected to univariate Cox regression, multivariate Cox regression, and least absolute shrinkage and selection operator regression analysis to obtain 13 lncRNAs. The samples were divided into high‐risk and low‐risk groups based on the expression of these 13 lncRNAs, and a series of correlation analysis and verification procedures were conducted.

INTRODUCTION

Lung cancer is one of the most dangerous malignant tumors for human health and life. In recent years, many countries have reported a significant increase in the incidence rate and mortality of lung cancer. ¹ The 85% incidence of lung cancer is non‐small‐cell lung cancer (NSCLC), lung adenocarcinoma (LUAD) accounts for approximately 55% in NSCLC and the 5‐year survival rate is only 15%. ² , ³ As for treatment approaches, surgery can be conducted on early‐stage patients, but survival rates for advanced‐stage patients are extremely low despite receiving combination and targeted treatment. ³ , ⁴ For the treatment and prognosis of LUAD patients, the further study of the behavior characteristics of LUAD cells is significant. Thus, effective and reliable prognostic indicators are urgently needed to improve the diagnosis and prognosis of patients.

Recent studies have illustrated that the mechanism of copper‐induced cell death (cuproptosis) was different from pyroptosis, necrosis, apoptosis, ferroptosis, and other known mechanisms that induce cell death through the regulation of copper ionophores via mitochondrial respiration. ⁵ , ⁶ , ⁷ , ⁸ , ⁹ Specifically, the direct binding of copper to lipidated proteins of the mitochondrial tricarboxylic acid (TCA) cycle leads to fatty acylation‐dependent protein aggregation and destabilization of iron–sulfur clustering, which leads to proteotoxic stress and cell death. ⁵ Mitochondrial respiration is an important factor affecting the dynamic behavior of malignant tumors from different perspectives. ¹⁰ During the process of cancer cell proliferation, the rate of flow of TCA cycles increased significantly. In addition, the growth inhibition of cancer tumor of mouse can be realized via the deletion of Glutaminase (GLS) alleles or the inhibition drugs of GLS. ¹¹ , ¹² NSCLC and many other tumor cells are significantly affected by an abundance of Ferredoxin 1 (FDX1) and lipoylated proteins, which has provided inspiration for the diagnosis and treatment of human tumors. ⁵ However, the role of cuproptosis‐related genes has only recently been proposed and study of the mechanism and clinical effect of cuproptosis‐related genes in LUAD and other cancers is still insufficient, therefore further research is required.

Generally, noncoding RNAs (ncRNAs) with nucleic acid length more than 200 nucleotides and no protein‐coding ability are known as long noncoding RNAs (lncRNAs). ¹³ , ¹⁴ The participation of lncRNAs in the regulation of protein‐coding genes can be realized via transcription, inheritance, and other mechanisms. Since lncRNAs participate in proliferation, migration, and apoptosis via different forms and mechanisms, the biological behavior and state of lncRNAs are considered to be important in various biological processes. ¹⁵ , ¹⁶ Due to the significant influence of the dynamic behavior of lncRNAs in tumor cells, several studies ¹⁴ , ¹⁵ , ¹⁷ have reported that lncRNAs can be a novel indicator of various human malignant tumors. There have been relatively few studies on the regulation mechanism of lncRNAs on cuproptosis‐related genes and this has hindered the effective use of lncRNAs in cancer treatment. Thus, the in‐depth study of the mechanism of the dynamic behavior of cuproptosis‐related lncRNA in cancer cells has important clinical value for the diagnosis and prognosis of LUAD patients.

In this study, an effective approach via machine learning was proposed for the identification of the cuproptosis‐related lncRNA signature (CRlncSig) via the comprehensive analysis of cuproptosis, lncRNA, and clinical characteristics. In addition, the response of CRlncSig to cancer cells and the possibility of using CRlncSig as an indicator of clinical prognosis and immunotherapy response in LUAD patients were comprehensively explored.

METHODS AND ALGORITHMS

Information acquisition for patients with LUAD

Information on the practical clinical, mutation and RNA sequencing transcriptome data of LUAD patients was collected through the Cancer Genome Atlas (TCGA) genomic data commons (GDC).

Selection of cuproptosis‐related lncRNA and genes

Nineteen cuproptosis‐related genes (DBT, NFE2L2, NLRP3, GCSH, ATP7B, ATP7A, SLC31A1, FDLAT, PDHA1, DX1 LIAS, DLST, LIPT1, LIPT2, PDHB, MTF1, DLD, GLS, and CDKN2A) were identified via a literature summary. ⁵ , ¹⁸ , ¹⁹ , ²⁰ , ²¹ , ²²

The Pearson correlation coefficient (PCC) is a statistical measurement unit that is widely used to measure the linear correlation between different variables, and has a value between −1 and 1. According to the screening criteria (|Pearson R| > 0.3 and p value < 0.001), 3450 cuproptosis‐related lncRNAs were screened via the PCC.

Proposal and evaluation of risk signature

The CRlncSig was combined with survival data, and the data were randomized as the testing set and the training set. As shown in Table 1, the results for the clinical features of the training and testing sets showed no significant difference (p value >0.05), indicating that the data were randomly distributed. The least absolute shrinkage and selection operator regression analysis (LASSO) regression analysis constructed a refined model using a penalty function so that the sum of the absolute values of the mandatory coefficients was less than a fixed value.

TABLE 1.

Differences in clinical characteristics between training and testing sets

Covariates	Type	Total	Testing set	Training set	p value
Age	≤65	235 (46.35%)	110 (43.48%)	25 (49.21%)	0.2376
	>65	253 (49.9%)	133 (52.57%)	120 (47.24%)
	Unknown	19 (3.75%)	10 (3.95%)	9 (3.54%)
Gender	Female	273 (53.85%)	138 (54.55%)	135 (53.15%)	0.8211
	Male	234 (46.15%)	115 (45.45%)	119 (46.85%)
Stage	I	271 (53.45%)	143 (56.52%)	128 (50.39%)	0.5366
	II	118 (23.27%)	55 (21.74%)	63 (24.8%)
	III	84 (16.57%)	40 (15.81%)	44 (17.32%)
	IV	26 (5.13%)	11 (4.35%)	15 (5.91%)
	Unknown	8 (1.58%)	4 (1.58%)	4 (1.57%)
Tumor (T)	T1	165 (32.54%)	89 (35.18%)	76 (29.92%)	0.2865
	T2	276 (54.44%)	137 (54.15%)	139 (54.72%)
	T3	44 (8.68%)	17 (6.72%)	27 (10.63%)
	T4	19 (3.75%)	8 (3.16%)	11 (4.33%)
	Unknown	3 (0.59%)	2 (0.79%)	1 (0.39%)
Metastasis (M)	M0	345 (68.05%)	175 (69.17%)	170 (66.93%)	0.6583
	M1	25 (4.93%)	11 (4.35%)	14 (5.51%)
	Unknown	137 (27.02%)	67 (26.48%)	70 (27.56%)
Node (N)	N0	325 (64.1%)	164 (64.82%)	161 (63.39%)	0.4944
	N1	94 (18.54%)	45 (17.79%)	49 (19.29%)
	N2	74 (14.6%)	35 (13.83%)	39 (15.35%)
	N3	2 (0.39%)	2 (0.79%)	0 (0%)
	Unknown	12 (2.37%)	7 (2.77%)	5 (1.97%)

Univariate Cox regression (UCR) analysis was used to analyze the correlation between influencing factors and samples. Thirty‐six significantly cuproptosis‐related lncRNAs were obtained using UCR analysis (p value <0.01). The 20 distinct cuproptosis‐related lncRNAs obtained by LASSO were analyzed using multifactor Cox regression (MCR), and the CRlncSig containing 13 lncRNAs was established. In this study, the CRlncSig scores of each sample were calculated using Equation (1). ²³

$$\mathit{CR}\ln \mathit{\text{cSig scores}}=\sum _{i=1}^n\left[\mathit{\text{Coef}}\left(\mathit{ln}{\mathit{\text{cRNA}}}_i\right)\times \mathit{\text{Expr}}\left(\mathit{ln}{\mathit{\text{cRNA}}}_i\right)\right]$$ (1)

where $\mathit{\text{Coef}}\left(\mathit{ln}{\mathit{\text{cRNA}}}_i\right)$ denoted the correlation coefficient and $\mathit{\text{Expr}}\left(\mathit{ln}{\mathit{\text{cRNA}}}_i\right)$ denoted the lncRNA expression. Based on the CRlncSig scores, the TCGA samples were divided into high‐risk groups (HRG) and low‐risk groups (LRG).

Functional analysis

Gene ontology (GO) enrichment analysis with appropriate criteria was performed to analyze the function enrichment. The criterion for GO analysis was p value < 0.05, which represents significant enrichment.

Immune analysis of the risk model

The CIBERSORT algorithm ²⁴ and ESTIMATE ²⁵ were conducted to analyze tumor immune infiltration. The likelihood of the immunotherapy response was estimated using Tumor Immune Dysfunction and Exclusion (TIDE) data.

Principal component and Kaplan–Meier survival analyses

The principal components of the whole dataset were analyzed using Principal Component Analysis (PCA). Furthermore, the overall survival (OS) between HRG (high risk group) and LRG (low risk group) was analyzed via the survminer and survival packages of the R program, and the Kaplan–Meier (K–M) method was used to obtain the survival curves.

Exploration of potential antitumor drugs

Data from the Cancer Genome Project (CGP) were selected for study to obtain potential antibiotic drugs. In this process, the pRRophetic package was used to predict the drug response to the expression matrix.

Accuracy evaluation of CRlncSig

During the accuracy evaluation of CRlncSig, the timeROC package of the R program was used to obtain the Receiver Operating Characteristic (ROC) curve. In addition, the pec package of the R program was an effective method to obtain the C‐index, and the linear regression plot was drawn by the regplot package of the R program. Finally, the nomogram calculated the predicted value of the individual outcome event by constructing an MCR model.

Independence analysis of CRlncSig

As described above, UCR and MCR analyses were used to explore the relationship and mechanism between different features (age, stage, gender, Tumor (T), Metastasis (M) and Node (N)) of LUAD patients and the prognostic pattern.

Prediction of the correlation between lncRNAs and diseases

In this study, LncRNADisease 2.0 was used to predict the correlation between lncRNA and disease. ²⁶

Details of reverse transcription‐polymerase chain reaction (RT‐PCR) and cell culture

In the experiment, the Trizol method was used for RNA extraction, and the RNA was reversed into complementary DNA (cDNA) using a reverse transcription kit. RT‐PCR was carried out to investigate the expression of lncRNAs. The primer sequences of lncRNAs and Glyceraldehyde‐3‐phosphate dehydrogenase (GAPDH) are shown in Table 2.

TABLE 2.

The primer sequences of long noncoding RNA and GAPDH

lncRNAs and gene	Sequence
Glyceraldehyde‐3‐phosphate dehydrogenase‐F primer	5′‐ACAACTTTGGTATCGTGGAAGG‐3′
Glyceraldehyde‐3‐phosphate dehydrogenase‐R primer	5′‐GCCATCACGCCACAGTTTC‐3′
FAM83A‐AS1‐F primer	5′‐CCCCAGAGCACTTCCTTAGC‐3′
FAM83A‐AS1‐R primer	5′‐CAGGGCCGTCTGTGTTTACT‐3′
AC026355.2‐F primer	5′‐GAGGTTCACTCCCTCAGCTG‐3′
AC026355.2‐R primer	5′‐TTCCTAATGGCAGCCCAAGG‐3′

LUAD (A549 and H1975 cells) and normal lung epithelial (BEAS‐2B cells) cell lines were cultured in the laboratory. All these cells were cultured at 37°C/5% CO₂ unless otherwise stated. During this process, Dulbecco's modified eagle medium (DMEM)‐based and 1640‐based media with 10% fetal bovine serum and 1% penicillin–streptomycin solution were used to culture the cells.

RESULTS

Screening of cuproptosis‐related lncRNAs

Using the proposed approach, 19 genes and 16 876 lncRNAs were selected from the database above. The CRlncs were obtained via coexpression analysis with an appropriate criterion (|Pearson R| > 0.3 and p < 0.001). As presented in Figure 1a 3450 CRlncs were obtained, and the coexpression network between cuproptosis‐related lncRNAs and genes was visualized via a Sankey diagram.

FIGURE 1.

Construction of cuproptosis‐related lncRNA signature (CRlncSig). (a) Sankey diagram of cuproptosis‐related long noncoding RNA (lncRNAs) related to the 19 genes. (b) Univariate Cox regression (UCR) analysis indicated that the 36 lncRNAs were significantly related to cuproptosis‐related genes. (c) log λ values were selected to 10‐fold cross‐validation. (d) The coefficient profile of 36 lncRNAs. Coefficient profiles were drawn based on (log λ) sequences and lanbda.min values were chosen based on 10‐fold cross‐validation, where the optimal λ generated 20 lncRNAs. (e) Multifactor Cox regression (MCR) analysis screened 13 lncRNAs as independent prognosis factors. (f) Correlation heatmap of 19 cuproptosis‐related genes with 13 independent prognosis lncRNAs.

Construction of CRlncSig

The 3450 lncRNAs were combined with the clinical survival data, and the data were randomized as a testing set (253 patients) and a training set (254 patients). The training set was performed using UCR analysis to obtain significant lncRNAs, and 36 CRlncs were correlated with OS (Figure 1b). LASSO regression analysis was used to accurately obtain the cuproptosis‐related lncRNAs (Figure 1c,d). Twenty CRlncs were selected based on lambda.min for the subsequent MCR analysis. As a result, 13 potential independent prognosis lnRNAs were determined via MCR analysis (Figure 1e). Figure 1f shows the correlation between the cuproptosis‐related genes and 13 potential independent prognosis lnRNAs.

Validation of the distinguishing ability of CRlncSig

The prognostic risk of patients with LUAD was evaluated using the CRlncSig, which was constructed through the 13 cuproptosis‐related lncRNAs obtained from the MCR analysis. The accuracy and reliability of the classification results of the HRG and LRG obtained by CRlncSig were comprehensively analyzed. Based on the CRlncSig scores, the samples were divided into HRG and LRG (Figure 2a,b), which demonstrated the survival time and status of the patients in the HRG and LRG. Figure 2c illustrates the expression of cuproptosis‐related lncRNAs in the HRG and LRG. The survival curves of the patients in the HRG and LRG are shown in Figure 2d. The results indicate that patients in the HRG exhibited lower survival rate than those in the LRG, which is consistent with the original intention of the study. Moreover, the prognostic ability of CRlncSig was investigated via the testing set and all the data, respectively. As shown in Figure A1a–h, all the analysis and evaluation results were consistent with those of the training set, indicating that the proposed model had good adaptability to different data and emphasizing the universality of the proposed model.

FIGURE 2.

The cuproptosis‐related lncRNA signature (CRlncSig) scores correlation diagram shows the grouping ability of CRlncSig. (a) Risk curve graph. (b) Survival status diagram. (c) The risk heatmap emphasizes the significant difference of expression level between the high‐risk groups and the low‐risk groups. (d) Kaplan–Meier analysis.

In addition, the survival rates with different ages, genders, stages, and Twere analyzed and predicted using the CRlncSig. Figure A2a–h shows that patients in the HRG exhibited lower survival rates. Furthermore, the prognostic ability of the proposed approach was analyzed using the progression‐free, disease‐free, and disease‐specific survivals of different patients (Figure A2i–k), and all three groups of patients exhibited higher survival rates in the LRG (Figure A2i–k), confirming the effective role of CRlncSig in the prediction of the prognosis of LUAD patients.

Investigation of the grouping ability of the proposed model via PCA

PCA can convert multiple indicators into a few comprehensive indicators through dimensionality reduction, therefore it can be used to judge whether CRlncSig can distinguish the HRG and LRG clearly. The distributions of the profiles of gene expression, cuproptosis‐related lncRNAs, and genes in the HRG and LRG are shown in Figure 3, indicating the effectiveness of using CRlncSig to distinguish the HRG and LRG.

FIGURE 3.

Principal Component Analysis (PCA) in different groups. (a) Profiles of gene expression. (b) Nineteen cuproptosis‐related genes. (c) CRlnc. (d) Profiles of cuproptosis‐related lncRNA signature (CRlncSig).

Effectiveness evaluation of CRlncSig

The effectiveness evaluation of CRlncSig when acting alone was important in the practical clinical use and prognosis analysis. The UCR and MCR analyses were used to evaluate the effect of CRlncSig under the condition of independent use.

The result of UCR analysis is presented in Figure 4a, where the hazard ratio (HR) is 1.061 and the 95% confidence interval (CI) is 1.047–1.075. The results of MCR analysis are presented in Figure 4b, where the HR and 95% CI are 1.052 and 1.037–1.067 (p value < 0.001), respectively. The UCR and MCR analyses indicated that the CRlncSig can act independently of other characteristics, thus serving as an independent factor for prognostic. As shown in Figure 4c–e, the ROC and C‐index were used to evaluate the accuracy and sensitivity of CRlncSig. As a result, the ROC and C‐index values of the CRlncSig were both higher than 0.75 and higher than other clinical characters, emphasizing the good accuracy and reliability of CRlncSig.

FIGURE 4.

Effectiveness evaluation of cuproptosis‐related lncRNA signature (CRlncSig). (a, b) Univariate Cox regression (UCR) and multifactor Cox regression (MCR) analyses illustrate that the CRlncSig could act independently of other characteristics. (c, d) Receiver Operating Characteristic (ROC) curve. Accuracy of CRlncSig outperforms other clinical characteristics. (e) C‐index. Accuracy of CRlncSig outperforms other clinical characteristics.

The immunotherapy response of CRlncSig

The relationship between CRlncSig and immunity was revealed using the CIBERSORT algorithm, ESTIMATE, GO functional enrichment, and immune check genes analyses. As shown in Figure 5a, the GO analysis indicted that CRlncSig is related to humoral immunity. As shown in Figure 5b,c, both the immuneScore and immune infiltrating cells of different groups were different. As shown in Figure 5d, in the HRG, the Infiltrations of activated T cells CD4 memory, macrophages M0, mast cell activated was relatively high, and the infiltration of monocytes, dendritic cells resting, and mast cells resting were relatively low.

FIGURE 5.

Differential evaluation of immune infiltration of cuproptosis‐related lncRNA signature (CRlncSig). (a) Gene ontology analysis. (b) The immuneScore of high‐risk groups and low‐risk groups. (c, d) Analysis of immune infiltration in different groups via CIBERSORT algorithm.

The differences in immunotherapy among different groups were analyzed using immune checkpoint genes, TIDE scores, and Tumor mutational burden (TMB). A previous study ²⁷ demonstrated that TIDE could be used to evaluate the potential therapeutic efficacy of immune checkpoint inhibitor (ICI) therapy in HRG and LRG patients. In addition, the possibility of immune evasion increased with increasing TIDE scores. The expression differences of immune checkpoint genes between the HRG and LRG are shown in Figure 6a,b. Figure 6c shows that patients in the HRG had relatively low TIDE scores, suggesting that the effect of ICI treatment on the HRG may be greater. TMB has been widely used in enzyme immunoassay, enzyme‐linked immunosorbent assay, and response prediction of patients to immunotherapy. The relatively high TMB could enable T cells to recognize more new neoantigens to optimize the effect of immunotherapy. ²⁸ The analysis revealed that the HRG had relatively higher TMB (Figure 6d) and survival rates (Figure 6e), indicating that immunotherapy had better efficacy for high‐risk patients. After grouping the mutation data using CRlncSig, 15 genes with the highest mutation frequency were visualized, as shown in Figure 6f,g. Generally, high TMB resulted in better patient prognosis, ²⁹ therefore the effects of CRlncSig and TMB prognosis were comprehensively compared. Figure 6h shows that the survival rates of patients under Condition 1 (red line represents High‐TMB and high risk) and Condition 3 (purple line represents Low‐TMB and high risk) were significantly lower than for those under Condition 2 (blue line represents High‐TMB and low risk) and Condition 4 (green line represents Low‐TMB and low risk). However, the survival curve of patients under Condition 1 was similar to that for those under Condition 3, thus the effect of TMB diagnosis on distinguishing patients with different risks was relatively poor. This analysis demonstrates that the prognosis significance of CRlncSig is better than that of TMB, proving the effectiveness of the proposed approach.

FIGURE 6.

Response evaluation of cuproptosis‐related lncRNA signature (CRlncSig) to tumor immunotherapy. (a, b) Immune checkpoint genes analysis. (c) The high‐risk groups (HRGs) had relatively low Tumor immune dysfunction and exclusion (TIDE) scores. (d) The HRGs had relatively higher Tumor mutational burden (TMB). (e) Survival analysis. The HRGs had relatively higher survival rates. (f, g) Waterfall plot visualizing the 15 genes in the HRGs and low‐risk groups. (h) Kaplan–Meier curves of patients under different conditions.

Investigation of the prognostic nomogram and potential drug

As shown in Figure 7a, nomograms with CRlncSig scores and clinical features were generated to predict the survival rates (1‐, 3‐, and 5‐year). Additionally, the nomograms show individual clinical scores, total points, and predicted survival rates (78.6% in 1 year, 41.2% in 3 years, and 15.5% in 5 years) of the 20th (randomly selected) patient. The correlation plots demonstrate that all the actual and predicted survival rates of OS were consistent (Figure 7b). Furthermore, 83 potential drugs targeting the CRlncSig based on CRlnc were determined via the pRRophetic of the R program and the CGP algorithm. Figure 7c–h indicated six potential drugs available for the treatment of LUAD patients.

FIGURE 7.

Investigation of prognostic nomograms and potential drugs. (a) The nomograms with cuproptosis‐related lncRNA signature (CRlncSig) scores were generated to estimate the survival rates (1‐year, 3‐year, and 5‐year). (b) The nomogram calibration showed the predicted survival rates of the 1‐, 3‐, and 5‐year overall survival (OS). (c–h) The boxplot describes the sensitivity of IC₅₀ between different groups and the correlation graph reveals the relationship between the CRlncSig score and IC₅₀.

Verification of the expression level of lncRNAs

The results of lncRNA differential analysis of TCGA were verified through RT‐PCR experiments. The existing (FAM83A‐AS1) and newly (AC026355.2) discovered lncRNAs were randomly selected for verification. Figure 8a–f shows that the expression levels of FRAM83A‐AS1 and AC026355.2 in H1975 and A549 cells (LUAD cell lines) are relatively high compared with that of BEAS‐2B cells (lung normal epithelial cell lines), which further verifies the reliability of TCGA differential analysis.

FIGURE 8.

Expression levels of long noncoding RNAs (lncRNAs). (a, b) Expression differences of lncRNAs (FAM83A‐AS1, AC026355.2) between different samples in The Cancer Genome Atlas (TCGA). (c, d) Expression differences of lncRNAs (FAM83A‐AS1, AC026355.2) between tumor and corresponding normal samples in TCGA. (e, f) Expression levels of lncRNAs (FAM83A‐AS1, AC026355.2) in BEAS‐2B cells (lung normal epithelial cell lines) and H1975 and A549 cells (lung adenocarcinoma cell lines).

DISCUSSION

Cuproptosis is a different cell death pattern from apoptosis, pyroptosis, ferroptosis. For LUAD patients that do not respond to chemoradiotherapy or immunotherapy, investigations into cuproptosis may provide an effective way to treat cancer in the future. However, studies on cuproptosis in tumors are relatively few, therefore its regulation in tumors is not clear. Research has increasingly focused on identifying signatures with lncRNAs to estimate the survival and immunotherapy response, therefore the systematic screening of CRlnc is conducive to LUAD treatment.

The lncRNAs related to cuproptosis were determined using PCC analysis, and a prediction approach using cuproptosis‐related lncRNAs was constructed by 13 lncRNAs to estimate the OS in LUAD patients. The patients were divided into HRG and LRG via CRlncSig scores. Among the 13 lncRNAs, four lncRNAs (AC026355.2, AC008957.1, Ap000346.1, GLIS2‐AS1) were downregulated in the HRG and nine lncRNAs (CDKN2A‐DT, FAM66C, FAM83A‐AS1, AL3592.1, FRMD6‐AS1, AC027237.4, AC023090.1, AL157888.1, AL627443.3) were upregulated in the HRG. These lncRNs were distributed on different chromosomes with lengths longer than 200 nt (as shown in Table 3), and among these five lncRNAs (CDKN2A‐DT, FAM83A‐AS1, AC026355.2, FAM66C, FRMD6‐AS1, GLIS2‐AS1) were shown to be related to tumors. CDKN2A‐AS1 promotes the proliferation, migration, and invasion of epithelial ovarian cancer cells by regulating the SOSTDC1 signaling pathway, ³⁰ and patients with LUAD who overexpress CDKN2A‐AS1 had a worse prognosis. ³¹ FAM83A‐AS1 promotes the proliferation, migration, invasion, and autophagy of LUAD cells by regulating microRNA‐141‐3p and FAM83A. ³² , ³³ In addition, FAM83A‐AS1 is also involved in iron death and necrotic apoptosis, and is related to the prognosis of patients with LUAD. ³⁴ , ³⁵ Moreover, FAM66C promotes tumor proliferation in intrahepatic cholangio carcinoma via the regulation of hsa‐miR‐23b‐3p/KKND2, which is involved in the immune infiltration of melanoma. ³⁶ Similarly, FRMD6‐AS1 regulates metastasis in nasopharyngeal carcinoma. ³⁷ In summary, these lncRNAs are potential markers of LUAD, promoting the occurrence and development of LUAD, and indicating that CRlncSig is more reliable for the prognosis of LUAD. Notably, seven lncRNAs (AC008957.1, AP000346.1, AC027237.4, AC023090.1, AL157888.1, AL627443.3, AL359232.1) were first discovered in this study, and the association of lncRNAs with diseases was analyzed using lncRNADisease Version 2.0. ²⁶ The known and predicted associated diseases of 13 lncRNAs are summarized systematically in Table 4. Of these, AC008957.1, AL359232.1, AC023090.1, and AL157888.1 were proved to be closely related to cancers such as lung cancer. These lncRNAs may be potential markers of LUAD, which may provide more target research possibilities for LUAD.

TABLE 3.

Long noncoding RNA information

ncRNA	Transcript	Length (nt)	Chromosome	Gene name
CDKN2A‐DT	NR_024274.1	616	9	CDKN2A divergent transcript
FAM83A‐AS1	NR_024479.1	1175	8	FAM83A antisense RNA 1
FRMD6‐AS1	NR_037676.1	2246	14	FRMD6 antisense RNA 1
GLIS2‐AS1	NR_110901.1	359	16	GLIS2 antisense RNA 1
FAM66C	NR_026788.1	3266	12	Family with sequence similarity 66 member C
AC008957.1	ENST00000512329.2	1924	5	NA
AP000346.1	ENST00000390329.3	1087	22	NA
AC027237.4	ENST00000560870.1	694	15	NA
AC023090.1	ENST00000592906.1	502	18	NA
AL157888.1	ENST00000631930.1	593	10	NA
AL627443.3	ENST00000663046.1	1376	6	NA
AL359232.1	ENST00000556874.1	1206	14	NA

TABLE 4.

Diseases related to long non‐coding RNA (lncRNA)

ncRNA	Disease (experimental and predicted)
CDKN2A‐DT	Lung squamous cell carcinoma, cervical cancer, lymphoma, malignant glioma, non‐small‐cell lung carcinoma, stomach cancer, thyroid cancer, urinary bladder cancer
FAM83A‐AS1	Lymphoma, Alzheimer's disease, breast cancer, cancer, cervical cancer, colorectal cancer, Huntington's disease, lung cancer, malignant glioma melanoma, neuroblastoma, non‐small‐cell lung carcinoma, ovarian cancer prostate cancer
FRMD6‐AS1	Astrocytoma, cervical cancer, lymphoma, malignant glioma, non‐small‐cell lung carcinoma, stomach cancer, thyroid cancer, urinary bladder cancer
GLIS2‐AS1	Lymphoma, acquired immunodeficiency syndrome, Alzheimer's disease, breast cancer, cancer, cervical cancer, colorectal cancer, hepatocellular carcinoma, Huntington's disease, lung cancer malignant glioma, non‐small‐cell lung carcinoma, osteosarcoma, ovarian cancer, prostate cancer
FAM66C	Lymphoma, Alzheimer's disease, breast cancer, cancer, cervical cancer colorectal cancer, hepatocellular carcinoma, Huntington's disease lung cancer, malignant glioma, melanoma, neuroblastoma, non‐small‐cell lung carcinoma, osteosarcoma, ovarian cancer
AC008957.1	Cervical cancer, lymphoma, malignant glioma, non‐small‐cell lung carcinoma, stomach cancer, thyroid cancer, urinary bladder cancer
AP000346.1	None
AC027237.4	None
AC023090.1	Alzheimer's disease, breast cancer, cancer, colorectal cancer, hepatocellular carcinoma, Huntington's disease, lung cancer, lymphoma osteosarcoma, ovarian cancer, prostate cancer
AL157888.1	Cervical cancer, lymphoma, malignant glioma, non‐small‐cell lung carcinoma, stomach cancer, thyroid cancer, urinary bladder cancer
AL627443.3	None
AL359232.1	Cervical cancer, lymphoma, malignant glioma, non‐small‐cell lung, carcinoma, stomach cancer, thyroid cancer, urinary bladder cancer

Generally, clinical characteristics were considered to be an important prognostic factor. In terms of the prognostic accuracy of the CRlncSig, UCR analysis indicated that the CRlncSig and clinical characteristics (T, N, stage) were associated with the prognosis of LUAD. Furthermore, only the CRlncSig was determined to be associated with the prognosis of LUAD using multivariate Cox regression analysis. Additionally, the ROC value of the CRlncSig reached 0.797, which was higher than other clinical features, emphasizing the good prognostic accuracy of the proposed approach. The CRlncSig was superior to clinical characteristics, which could better reflect the prognosis of patients with LUAD. Nomogram is a statistical prediction model for cancer prognosis. All the predicted and actual survival rates were consistent, proving the significance of the proposed model for the LUAD patients. Tumor immunotherapy with ICI was designed to enhance the recognition ability and attack effect of the human immune system against cancer cells, and has a smaller side effect on patients with metastatic cancer and a stronger specificity for tumor recognition. ²⁷ , ³⁸ The Tumor mutational burden (TMB) is regarded as an effective indicator for the prediction of the treatment of patients with ICI. ²⁸ , ³⁹ In this study, the possibility of the proposed approach providing reliable immune biomarkers for tumor treatment was evaluated by TMB and TIDE scores of different risk groups. The analysis results revealed that the CRlncSig could better distinguish the prognostic differences in patients with high TMB for precise immunotherapy. However, the TMB could not distinguish the prognostic differences from the risk model in detail, indicating that the risk model can distinguish potential treatment populations more accurately and an optimal treatment plan can be obtained. The CRlncSig proposed in this study was therefore superior to the TMB prognosis, providing a more accurate identification of patients benefiting from immunotherapy. In clinical practice, patients with LUAD experience immune escape, and the immune evasion mechanism of the tumor can be clarified from two perspectives. Generally, the T cells are in a dysfunctional state, although the infiltration of cytotoxic T cells is relatively high. Additionally, T cells infiltrating tumor tissue are cleared via immunosuppressive factors. Furthermore, the TIDE score is an effective computational framework that integrates these two mechanisms to realize the prediction of treatment with ICI. In this study, patients in the HRG had relatively high TIDE scores, that is, the ICI played a more significant and effective role in high‐risk patients.

This study had some deficiencies and limitations. Although there were 598 LUAD samples, the limited number of samples may lead to some potential biomarkers being missed. Database updates may slightly change the conclusions and results obtained by the proposed model. In addition, although some of the potential prognostic lncRNAs obtained in this study have proved their roles in tumors, further in vitro experiments may be useful to verify the roles and mechanisms of lncRNAs in LUAD. Consequently, the augmentation of sample data and further in vitro experiments to enhance the reliability of the proposed approach will be the focus in future work.

CONCLUSION

As LUAD causes considerable harm to human life and health, a reliable prediction approach for the prognostic of patients with LUAD is important, which was instructive to clarify the correlation and mechanism between cuproptosis and lncRNAs. The prediction approach was able to identify LUAD patients with good responses to immunotherapy, emphasizing the possibility and reliability of CRlncSig as a potential target for the treatment of LUAD.

AUTHOR CONTRIBUTIONS

All authors had full access to the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Conceptualization: Lihong Yang. Formal analysis: Lihong Yang. Writing – original draft preparation: Lihong Yang. Data curation: Yazhou Cui. Funding acquisition: Jianping Lin. Writing – review and editing: Jianping Lin and Lu Liang.

CONFLICT OF INTEREST STATEMENT

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.

APPENDIX A.

FIGURE A1.

Cuproptosis‐related lncRNA signature (CRlncSig) scores correlation diagram. (a, b) Risk curve graph. (c, d) Survival state diagram. (e, f) Risk heatmap. (g, h) Kaplan–Meier analysis.

FIGURE A2.

Kaplan–Meier curves of the differences in survival rates under different risk groups. (a) Patients under 65 years old, (b) patients older than 65 years, (c) patients with stage I–II, (d) patients with stage III–IV, (e) female patients, (f) male patients, (g) patients with T1 and T2, (h) patients with T3 and T4, (i) progression‐free survival, (j) disease‐free survival, (k) disease‐specific survival.

Yang L, Cui Y, Liang L, Lin J. Significance of cuproptosis‐related lncRNA signature in LUAD prognosis and immunotherapy: A machine learning approach. Thorac Cancer. 2023;14(16):1451–1466. 10.1111/1759-7714.14888