The RNA sequencing data, mutation data, and corresponding clinical information of 551 LUAD patients were downloaded from the TCGA database(https://portal.gdc.cancer.gov/), which consisted of 497 LUAD tumor specimens and 54 normal specimens. Then, each gene featuring zero values was removed, and the average excess gene expression was calculated. Samples featuring no survival information and those with OS below 30 days were removed from the clinical information. The gene expression information of 397 LUAD tumor specimens and the relevant clinical information were obtained from the GEO (https://www.ncbi.nlm.nih.gov/geo/) GSE31210/GSE30219/GSE37745 datasets. For better dataset comparability, all of the expression data were converted to FPKM format. Batch effects were then eliminated using the combat function of the “sva” package. All data were transformed using Log2 before analysis. Ten cuproptosis-related genes (CRGs) were then retrieved from one published work and the researchers identified seven positively regulated and three negatively regulated CRGs (10).

The correlations between the 10 CRGs and lncRNA expression were analyzed using Pearson correlation analysis. Each CRlncRNA must be in accordance with the following standard of correlation coefficients (|Pearson R|): >0.3 and P <0.01. Using difference analysis, we ultimately obtained 76 differentially expressed CRlncRNAs (log2 fold alteration >1, false discovery rate (FDR) <0.05).

Using the “caret” R package, the 497 TCGA LUAD samples were randomly divided into a training group and a testing group at a ratio of 1: 1. The training group was used for constructing the CRlncRNAs signature, while the testing group was used for internal validation. In combination with the LUAD clinical information in TCGA, the differentially expressed CRlncRNAs were imported into the Cox and LASSO regression to create the predictive signature. Finally, the risk score was determined as follows: risk score = ∑ k = 1 n Coef ( k )   ×  Expr ( k ) . Coef (k) were the short name for the regression coefficients and Expr (k) was the expression of prognostic CRlncRNAs. Patients were divided into low- and high-risk groups based on the mean risk score. The Kaplan–Meier estimator and the log-rank test were used to analyze whether a difference existed between the high and low-risk groups in terms of the OS of LUAD patients and the “survival” R package was employed for this. Then, external validation was performed for this signature, with the risk scores of 397 LUAD samples from the GEO database calculated by the above formula for survival analysis. K-M survival curves with log-rank tests were used to assess the prognostic efficacy of the risk model across all cohorts. The accuracy of this signature in foretelling OS of LUAD patients was evaluated using receiver operating characteristic (ROC) curve analysis. Both univariate and multivariate Cox regression analyses were used to estimate the independent prognostic value.

Using the “rms” R package, the risk scores and the clinical variables of gender, age, and tumor phase were combined to construct a nomogram for the one-, three- and five-year OS of LUAD patients and correction plots based on the Hosmer-Lemeshow test for illustrating the consistency between actual outcomes and predicted outcomes. Then, we separated the clinicopathological characteristics into subgroups and ran a survival analysis to determine whether they had any effect on the model.

Utilizing the “GSVA” package, Gene Set Variation Analysis (GSVA) was carried out to investigate the heterogeneity of diverse biological processes. Hallmark gene sets “h.all.v7.5.1.symbols.gmt” from MSigDB (https://www.gseamsigdb.org/gsea/msigdb/index.jsp) were utilized for the GSVA. Using the R package “clusterProfiler” and “org.Hs.eg.db”, gene set enrichment analysis (GSEA) was applied to identify clearly enriched signaling paths in various groups on the basis of the following criterion: FDR < 0.25 and P < 0.05.

In order to assess the correlation between the prognostic signature and the TIME, the immune infiltrating values of the TCGA-LUAD specimens were calculated according to 7 algorithms: CIBERSORT, CIBERSORT-ABS, EPIC, MCPCOUNTER, QUANTISEQ, TIMER, and XCELL (26–32). Spearman correlation analysis was used to evaluate the correlation between immune cell sub-populations and risk scores and the outcomes were presented in a bubble chart. Then, the abundance of immune cells and stromal cells between various groups was explored. The immune scores, stromal scores, and estimate scores (stromal scores + immune scores) of patients were computed to assess the TME differences by the “estimate” R package. Machine learning can accurately assess and quantify immunogenicity. The Cancer Immunome Atlas (TCIA) database was used to retrieve the Immunophenoscores (IPS) for LUAD (33). To forecast immunotherapy sensitivity, the IPS between the high-risk and low-risk groups was examined. Tumor Immune Dysfunction and Exclusion (TIDE) algorithm was used online (http://tide.dfci.harvard.edu/) (34). Patients were more likely to respond to ICI treatment with greater effects if their TIDE scores were lower. Then, single-sample GSEA (ssGSEA) was applied to explore the difference between various groups in terms of immune cells and immune-related function. Subsequently, comparisons were also made between high-and low-risk groups in terms of immune checkpoints.

The IC50 of commonly used chemotherapeutic medications in the TCGA-LUAD dataset was computed using the R package “pRRophetic” to analyze the risk model in the clinic for LUAD treatment (35). By building statistical models using gene expression and drug sensitivity data from cell lines in the Cancer Genome Project, this program enables users to predict the clinical chemotherapeutic response using just baseline tumor gene expression data. Violin plots depict the Wilcoxon signed-rank test findings comparing the IC50s of commonly used antitumor medicines between the high-and low-risk groups.

To explore potential molecular subtypes, all patients with LUAD were classified into different clusters based on the expression of prognostic CRlncRNAs via the R package “ConsensusClusterPlus”. Differences in survival, TIME, immune checkpoints, and response to immunotherapy between subgroups were assessed in the same manner as before. At the same time, we used the TCGA-LUAD mutation data to explore the mutation differences between high-and low-risk groups and different clusters, which were presented in the form of heat maps.

Normal human lung epithelial BEAS-2B cells and human LUAD cell lines (A549, H1299) were purchased from the Cell Resource Center of Shanghai Life Sciences Institute. These cells were grown in F12K or RPMI-1640 (Gibco BRL, USA) with 10% fetal bovine serum (FBS), 1% streptomycin, and penicillin (Gibco, Invitrogen, Waltham, MA, USA). 5% CO2, 95% humidity, and 37°C were used to cultivate the cells.

LINC00592 knockdown was generated using small interfering RNAs (siRNAs) constructs (GenePharma, Suzhou, China). In addition, LINC00592 siRNA sequences were listed in Supplementary Table S1 . Briefly, cells were seeded at 50% confluence in a 6-well plate and infected with negative control (NC), and knockdown (siLINC00592). All transfections were carried out with Lipofectamine 3000 (Invitrogen, USA).

Total RNA from cell lines was extracted by the manufacturer’s instructions using TRIzol (15596018, Thermo). After that, cDNA was created using the PrimeScriptTMRT kit (R232-01, Vazyme). SYBR Green Master Mix (Q111-02, Vazyme) was used to perform the Real-time polymerase chain reaction (RT-PCR), and the expression levels of each mRNA were normalized to the level of mRNA GAPDH. The 2^−ΔΔCt method was used to count the expression levels. Tsingke Biotech (Beijing, China) provided all primers, and Supplementary Table S1 has full primer sequences.

In 96-well plates, we seeded the cell suspension with 3×10³ cells per well. The plate was then incubated for 2 hours at 37°C in the dark with 10 mL of CCK-8 labeling agent (A311-01, Vazyme) each well. The enzyme-labeled meter (A33978, Thermo) was used to measure the absorbance of the cells at 450 nm for 0, 24, 48, 72, and 96 hours in order to determine the vitality of the cells.

We transfected 1×10³ cells into each well of a 6-well plate and kept the cells alive for 14 days. Before Crystal violet (Solarbio, China) staining, the cells were washed twice with PBS and fixed for 15 minutes in 4% paraformaldehyde.

A 96-well plate with 2×10⁴ treated cells in each well was used for the experiment, which was conducted after the cells had attached to the wall. The manufacturer’s recommended 5-Ethynyl-2’-deoxyuridine (EdU) assay was then carried out (Ribobio, China). An inverted microscope was used to count the number of proliferating cells.

In 6-well plates, transfected cells were plated and cultured in a cell incubator until they were 95% confluent. In each cultured well, a single straight line was scraped using a sterile 20-L plastic pipette tip, and unattached cells and debris were gently washed away twice with PBS. Finally, we used the Image J software to measure the width of the scratches after taking photos of the scratch wounds at 0h and 48h.

Cell invasion and migration studies were performed in transwells. The top chamber of 24-wells was filled with treated A549 and H1299 cells (2×10⁵), which were then incubated for 48 hours. To assess the cells’ ability to invade and migrate, the top section of the plate was either precoated with matrigel solution (BD Biosciences, USA) or left untreated. The remaining cells on the bottom layer were then fixed with 4% paraformaldehyde and stained with 0.1% crystal violet after the cells on the top surface were removed (Solarbio, China).

All of our data and statistics were processed using R (version 4.1.3). Experiment data were analyzed using the applications Graphpad and Image J. Kaplan-Meier curves with a log-rank test were used to compare the OS rates of the two groups. A “Survminer” R package was used to generate all survival curves. To assess prognostic factors, univariate and multivariate Cox regression analysis (CRA) was employed. Lasso regression was used to discover characteristics that had a greater impact on the outcomes. The “ggplot2” R program was used to visualize the data, and the “survival” R package was used to calculate the OS and risk ratings. The heatmap is created using the “Pheatmap” R package. For normally distributed data, a two-tailed t-test or a one-way ANOVA was employed to discover significant quantitative differences. To determine if there were significant differences for nonnormally distributed data, a Wilcoxon test or a Kruskal-Wallis test was utilized. Every statistical analysis was completed using R software. P < 0.05 is a statistically significant value.

The flow chart of this study was shown in Figure 1 . A total of 76 differentially expressed CRlncRNAs were obtained by correlation analysis (|Pearson R| > 0.3 and P < 0.01) and variance analysis (|Log2FC| > 1 and P < 0.05), and the correlation network diagram ( Supplementary Figure S1A ) and corresponding heatmap and volcano charts were constructed ( Figures 2A, B ).

Through univariate CRA, 8 prognostic CRlncRNAs related to OS (P < 0.05) were obtained, and each of the CRlncRNAs was modulated positively through CRGs as shown in the Sankey diagram ( Figure 2C ). The circle diagram displays the prognostic HR value of 8 CRlncRNAs, and it is clear that TPMO-AS1, LINC00205, OGFRP1, MIR31HG, LINC00592, OGFRP1, and TFAP2A-AS1 are high-risk genes. In contrast, AQP4-AS1 and CADM3-AS1 were shown to be low-risk genes ( Figure 2D ). Subsequently, the prognostic signature consisting of 7 CRlncRNAs (LINC00592, OGFRP1, TFAP2A-AS1, AQP4-AS1, CADM3-AS1, MIR31HG, and LINC00205) was constructed by LASSO regression ( Supplementary Figures S1B, C ) and multivariate CRA. Some of the prognostic CRlncRNAs (LINC00205, OGFRP1, MIR31HG) have previously been proven to be closely related to NSCLC (36–38). Then, the expression levels of these genes were examined in the normal and tumor groups ( Figure 2E ), and the findings revealed that AQP4-AS1 and CADM3-AS1 were strongly expressed in the normal group, while the other genes were substantially expressed in the tumor group. Meanwhile, the clinical correlation heatmap was constructed based on clinical information and the expression levels of the 7 CRlncRNAs ( Figure 2F ). The risk scores of all LUAD patients were then calculated according to regression coefficients, and the patients were divided into high-and low-risk groups based on the mean risk score. The risk scores were calculated using the following formula: risk score = (0.6418 × LINC00592 expression) + (0.6236 × OGFRP1 expression) + (0.5246 × TFAP2A-AS1 expression) + (-2.4203 × AQP4-AS expression) + (-1.47395306825907 × CADM3-AS1 expression) + (0.3009 × MIR31HG expression) + (0.5019 × LINC00205 expression). Similarly, we calculated the risk value for 397 LUAD samples from the GEO database using the above formula and subsequently divided them into high-and low-risk groups. In order to determine the prognostic capability of the risk signature, we compared Kaplan-Meier survival plots for OS ( Figures 2G-J ), the distribution of risk scores, the survival time and survival status, and the related expression of 7 CRlncRNAs between the high-and low-risk groups in the TCGA cohort (training group, the testing group, all groups) and the GEO cohort (GSE31210, GSE3021,9 and GSE37745) ( Supplementary Figures S1D-G ).

Univariate and multivariate CRA confirmed that our risk model was an independent prognostic indicator in TCGA and GEO cohorts (P <0.001, Supplementary Figures S1H-K ). After this, it can be demonstrated using principal component analysis (PCA) analysis that the risk score may separate the TCGA and GEO cohorts into two different clusters ( Figures 3A-D ). Then, the ROC curves were used for validation of the specificity and sensitivity of the signatures in train groups, and the AUCs for one-, three- and five-year survival reached 0.703, 0.710, and 0.757, respectively, implying a good predictive value ( Figure 3E ). Additionally, the ROC curves showed high predictive abilities in the other validation groups ( Figures 3F-H ). After that, we evaluated the ability of the model genes to predict the high-and low-risk of patients, and the results are shown in Supplementary Figures S2A-G . The AUC values were all greater than 0.6, among which the AUC values of OGFRP1 and LINC00205 were both greater than 0.7, showing a strong predictive ability. Clinical data and risk classification were merged to construct a nomogram ( Figure 3I ), which was then utilized to evaluate the prognosis of TCGA-LUAD patients. The nomogram may aid in more precisely determining patient risk and directing future treatment options. We also created calibration curves ( Figure 3J ) and observed that this nomogram could reliably predict the prognosis of LUAD patients after one-, three-, and five years. Prognostic ROC analysis was used to further evaluate the accuracy of this nomogram, and the findings significantly beat those of other clinical shapes and risk scores. The AUC in 1, 3, and 5 years was 0.724, 0.734, and 0.708, respectively, according to the results ( Figures 3K-M ). These outcomes further suggested that this predictive signature constitutes a promising biomarker to predict the prognosis of LUAD. To determine whether the constructed prognostic model could forecast OS based on various clinical features, we conducted survival analysis for several clinical subgroups. The following analytical details are provided: Age (≤65 or >65), Gender (Male or Female), Clinical Stage (Stage I or Stage II-IV), T Status (T1-2 or T3-4), M Status (M0 or M1), and N Status (N0 or N1-3). Figures 4A-L showed that independent of age, gender, clinical stage, T status, M status, or N status, the OS rates of high-risk patients were lower than those of low-risk patients.

Analysis of hallmark pathway gene signatures highlighted that the high-and low-risk groups showed some differences. A direct comparison of Risk-High versus Risk-Low revealed that the top 5 enriched signatures in high-risk group included E2F targets, G2M checkpoint, MYC targets v1, MYC targets v2 and unfolded protein response and the low-risk group was mainly active in some metabolic related pathways ( Figure 5A ). According to clinical research, E2F family members are directly linked to the incidence, growth, proliferating, and apoptosis of cancerous tumors such as gastric, pulmonary, liver, esophagus, prostrate, bladder, and ovarian cancer (36, 37). In order to control cell proliferation, the G2M checkpoint also functions as a cell cycle regulatory route. High G2M checkpoint pathway activation has been linked in studies to considerably worse survival in people with pancreatic cancer (38). Furthermore, c-Myc is necessary for tumorigenesis (39), Almost often, Myc may increase transcription (40), which showed that LUAD cells could be susceptible to Myc inhibition. Extremely crucial nuclear transcription factors involved in controlling the cell cycle are encoded by the E2F family (36, 41). Unfolded protein response (UPRmt) is known to be preserved in cancer and is able to become active in response to mitochondrial stress (42). This helps to preserve mitochondrial integrity and promotes the development of tumors. As a result, these pathways, which were more prevalent in the high-risk group, may play a crucial role in controlling tumor development in LUAD. Then, GSEA was further applied to determine the obvious pathways between different risk groups ( Figure 5B ). It was found that the T/B cell receptor signaling pathway (SP), the JAK/STAT SP, and the cytokine receptor interaction were clearly enriched in the low-risk groups, implying a close association of low-risk patients with tumor- and immune-related pathways. On the other hand, high-risk groups were obviously enriched in cell cycle, DNA replication and proteasome pathway.

The correlation between risk scores and tumor-infiltrating immune cells was explored ( Figure 5C ). Most of the levels of immune cell infiltration, such as memory activated CD4+T cells, B cells and macrophages, were negatively correlated with the risk score, which may imply that patients in the low-risk group had higher levels of immune cell infiltration in the TME, as we confirmed in the following validation. In order to further assess TIME patterns and immunotherapeutic responses across different risk groups, we conducted an attempt. The components of LUAD’s TME score were calculated using the ESTIMATE method. Correlation analysis showed that the risk score was negatively correlated with immune scores, stromal cores and estimate scores, and positively correlated with tumor purity ( Figure 5D ). The violin diagrams confirmed that the low-risk group had higher immune scores, stromal scores and estimate scores, and the high-risk group had higher tumor purity ( Figure 5E ). The IPS may also aid in identifying those who might benefit from immunotherapy. It was predicted that LUAD patients will respond well to either PD-1/PD-L1 or CTLA4 inhibitors, or perhaps to both ( Figures 6A-D ). The results showed that the low-risk group showed higher IPS scores when patients were not treated with PD1/PDL1 and CTLA-4, and no significant differences were seen in the other subgroups. Surprisingly, Dysfunction scores were much lower in the high-risk group and Exclusion scores were significantly greater than in the low-risk group (P < 0.001; Figures 6E-G ), which suggests that high-risk patients are more likely to benefit from immunotherapy. Subsequently, the ssGSEA enrichment scores of different immune cells and immune related functions were quantified ( Figures 6H, I ). Compared to the high-risk group, B cells, DCs, immature dendritic cells (iDCs), mast cells, neutrophils, plasmacytoid dendritic cells (pDCs) and T helper cells had higher enrichment scores in the low-risk group. In the low-risk group, CCR, HLA, MHC class I and Type II TFN response were enriched to a higher degree, which may indicate that the TME of low-risk patients has a more active immune status to fight against tumor progression and thus has a better prognosis. Meanwhile, a comparison of various risk groups in terms of immune checkpoint activation revealed that most immune checkpoints expressed more activity in low-risk groups, such as HHLA2, CD40LG, CD48 and CD244 ( Figure 6J ). Some of these immune checkpoint expression levels were found to be highly expressed in the high-risk group, such as LAG3, IDO1, CD276 and PDCD1 (also called PD1), which may provide therapeutic guidance for high-risk groups.

Different chemotherapeutic medicines’ effectiveness was compared between different groups. For patients with higher risk score, we found that the IC50 for the following chemotherapeutic medications was lower: Vinblastine, Cisplatin, Tipifarnib, Axitinib, Cytarabine, Foretinib, Tivozanib, Masitinib, Bexarotene (P < 0.001, Figures 7A-I ). Our results support the use of this risk score to forecast chemotherapeutic drug sensitivity and immunotherapy response in LUAD patients, which makes it easier to construct a personalized medication treatment.

In order to identify distinctive molecular subtypes according to the expression of prognostic CRlncRNAs, unsupervised consensus clustering was applied. LUAD patients were separated into three clusters, with k=3 identified as the optimal clustering stability ( Figure 8A ). The PCA plot showed the distribution of Cluster 1(C1), Cluster 2 (C2) and Cluster 3 (C3) samples ( Figure 8B ), and a Sankey diagram was constructed to display the connection between clusters and high-and low-risk groups ( Figure 8C ). Samples C2 and C3 mainly belong to the high-risk group, while samples C1 are mainly distributed in the low-risk group. Subsequent survival analysis showed that patients in the C3 group had the worst prognosis, while patients in the C1 group had the best prognosis ( Figure 8D ). The heat map of immune cell infiltration including 7 algorithms was drawn ( Figure 8E ), and the results showed that there was higher immune cell infiltration in the TME of C1 group, such as M2 macrophages, T cells, etc., which may mean that more immune cells in the TME of C1 were recruited and activated adaptive immune response to form a hot tumor state. The TME scores of various clusters were then analyzed, and the findings revealed that C1 had a higher immune score, stromal score and estimate score, as well as lower tumor purity, than C2.

TME scores, immune checkpoint and immune cell infiltration results calculated by CIBERSORT were integrated to draw a heat map ( Figure 9A ), which showed that there were different degrees of highly expressed immune checkpoints in the C3 group, such as LAG3, PDL1, CD70, IDO1, PDL2, etc. This may indicate that C3 group is more likely to benefit from immune checkpoint blockade treatment, whereas no significant difference was seen in the heat map of immune cell infiltration. Subsequently, the expression of immune checkpoints in different clusters is shown by boxplot ( Figure 9B ). Patients with LUAD mutational landscape were presented in Figure 9C , which showed the highest 20 mutation frequency of gene mutation, it can be seen in the high-risk group, C2/C3 gene mutation frequency increases. Figure 9D confirmed that higher tumor mutation burden in high-risk groups, so we further according to the median risk scores and the median TMB, divided the patients into four groups (H-TMB+high-risk, H-TMB+low-risk, L-TMB+high-risk, and L-TMB+low-risk) and the survival differences between them were further observed ( Figure 9E ). Survival analysis illustrated that H-TMB+low-risk group had the best prognosis and L-TMB+high-risk group had the worst prognosis.

In the above studies, we found that LINC00592 exhibited the highest HR value and was substantially elevated in tumor samples compared to normal samples. Using the GEPIA database, we further determined the impact of LINC00592 on survival, and patients with high expression of LINC00592 had a worse prognosis ( Supplementary Figure S2H ). Further research was done on the LINC00592 in the signature. We conducted in vitro research to better comprehend LINC00592’s role in LUAD. Compared to BEAS-2B cell lines from healthy human lung epithelial tissue, Figure 10A showed that LINC00592 was significantly elevated in two LUAD (A549, H1299) cell lines. LINC00592 in LUAD cell lines was knocked down for subsequent experiments. First, we employed the qRT-PCR technique to measure the level of LINC00592 expression 5 days after transfection in order to gauge the effectiveness of siRNA knockdown of LINC00592 in A549 and H1299 cell lines ( Figure 10B ). We discovered that all siRNA sequences could significantly reduce LINC00592 expression (P < 0.001). Following LINC00592 knockdown, there was also a significant reduction in the cells’ vitality as evaluated by the CCK8 assay ( Figure 11A ). The findings of the experiment indicate that LINC00592 may have a key role in LUAD cell surviva. Cell proliferation was evaluated using the colony formation assay. Comparing cells with decreased LINC00592 expression to the siRNA NC group, the reduced LINC00592 expression cells showed a significantly lower number of colonies ( Figure 11B ). Therefore, a slower rate of colony formation was seen in LINC00592 knockdown cells, suggesting that LINC00592 may be essential for the proliferation of the LUAD cell line. EdU assays revealed that lower expression of LINC00592 decreased the proliferative activity of A549 and H1299 cells in comparison to the NC group ( Figure 11C ). The outcome of the scratch-wound healing experiment was comparable. Cells with reduced LINC00592 expression showed a significantly delayed rate of wound healing ( Figure 11D ). Transwell experiments revealed that LINC00592 downregulation severely restricted the migration and invasion of A549 and H1299 cells ( Figure 11E ). To demonstrate the accuracy and reproducibility of the results, all experiments were repeated in two LUAD (A549, H1299) cell lines and all data were presented as the means ± SD of three independent experiments. *P < 0.05, **P < 0.01, ***P < 0.001.