RNA transcriptome data (FPKM-normalized), clinical data (age, sex, AJCC TNM stage, survival status), and somatic mutation data for patients with LUAD (n=515) were obtained from the TCGA database (https://cancergenome.nih.gov/).

Gene expression files of lncRNAs and cuproptosis-related genes were downloaded from the TCGA database. In total, 19 cuproptosis-related genes, namely, NFE2L2, NLRP3, ATP7B, ATP7A, SLC31A1, FDX1, LIAS, LIPT1, LIPT2, DLD, DLAT, PDHA1, PDHB, MTF1, GLS, CDKN2A, DBT, GCSH and DLST (8,12-16), were obtained.

Pearson correlation coefficients (R) were calculated between the expression levels of each of the 19 curated cuproptosis-related genes and all the annotated lncRNAs (n=16,876) within the TCGA-LUAD cohort.

Selection thresholds. Correlation Strength (|R|>0.3): This threshold was chosen to identify lncRNAs with moderate-to-strong co-expression relationships with cuproptosis-related genes. A value >0.3 is a commonly accepted benchmark in transcriptomic studies for identifying biologically relevant co-expression partners, balancing specificity against overly stringent exclusion of potentially important regulators.

Statistical significance. This stringent P-value threshold minimizes false positives, ensuring that only lncRNAs with a highly statistically significant association with at least one cuproptosis gene were retained.

Scope. lncRNAs meeting the above |R| and p criteria for any one or more of the 19 cuproptosis-related genes were included. This broad approach captures lncRNAs potentially regulating specific facets of cuproptosis without requiring universal association with all the cuproptosis-related genes.

Output. The application of these criteria resulted in the identification of 3,385 cuproptosis-related lncRNAs for subsequent prognostic model construction.

Construction and validation of the lncRNA signature. TCGA-LUAD samples (n=515) were stratified by AJCC stage (I/II vs. III/IV) and survival status (alive/dead) and then randomly partitioned into training (70%, n=360) and test (30%, n=155) sets (random seed=2023) to preserve prognostic factor distributions. This 70:30 split adheres to the TRIPOD guidelines for prognostic model development. Using the training cohort, a cuproptosis-related lncRNA signature was developed, which was subsequently validated in the test cohort. Prognostic lncRNAs were initially screened from 3,385 cuproptosis-associated candidates based on OS in the TCGA-LUAD cohort (P<0.05). These were subsequently refined via LASSO-Cox regression analysis (R package ‘glmnet’), yielding a final signature of 52 lncRNAs (17). Further multivariate Cox regression identified 4 lncRNAs, which were used to construct a prognostic lncRNA signature. A risk score was accordingly established and formulated as follows: Risk score=[coef (lncRNA1) x expr (lncRNA1)] + coef(lncRNA2) x expr (lncRNA2)] +......+ [coef(lncRNAn) x expr (lncRNAn)]NA2)] +......+ [coef(lncRNAn) x expr (lncRNAn)]; where coef (lncRNAn) represents the Cox regression coefficient and expr (lncRNAn) represents the expression of lncRNAn. Samples from the training and test sets were grouped based on their risk scores (18).

Somatic mutations were processed using maftools (v2.12.0; https://bioconductor.org/packages/maftools/). TMB equals to (total mutations)/[exome size (38 Mb)].

Immunotherapy response prediction. The TIDE algorithm (http://tide.dfci.harvard.edu/) was applied to predict the anti-PD1/CTLA4 response.

Immune cell infiltration. The data were estimated using CIBERSORTx and ssGSEA (R package GSVA).

Drug sensitivity screening. IC₅₀ values for 73 GDSC compounds were predicted using pRRophetic (v0.5; https://bioconductor.org/packages/pRRophetic/) based on the TCGA transcriptomes.

Functional analysis. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were performed using the R package ‘clusterProfiler’ to characterize the biological functions of the signature lncRNAs (P<0.05).

Nomogram construction and validation. A nomogram incorporating the risk score, age, sex, and TNM stage classification was developed to predict 1-, 3-, and 5-year OS in patients with LUAD. Calibration performance was assessed using the Hosmer-Lemeshow test to evaluate the agreement between the predicted and observed outcomes.

Survival analysis. Kaplan-Meier survival analysis was performed to compare OS and other survival endpoints [including progression-free interval (PFI), disease-specific survival (DSS), and disease-free interval (DFI)] between the high-risk and low-risk patient groups stratified by the median risk score. The statistical significance of the differences in survival curves was assessed using the two-sided log-rank test. P<0.05 was considered to indicate a statistically significant difference.

Construction of an ARHGEF26-AS1-associated competing endogenous RNA (ceRNA) Network. The ceRNA network focused on ARHGEF26-AS1 was constructed by first predicting its interacting miRNAs using miRanda (http://www.microrna.org/microrna/home.do) and DIANA-LncBase (https://diana.imis.athena-innovation.gr/DianaTools/index.php). The downstream target genes of these miRNAs were then retrieved from the StarBase database (https://starbase.sysu.edu.cn/), focusing on interactions supported by experimental evidence. The integrated network, depicting ARHGEF26-AS1, miRNAs, and target genes, was finally assembled and visualized using Cytoscape.

Cell culture. Normal human bronchial epithelial cells (BEAS-2B) and human LUAD cells (A549) (Fuheng Biotechnology Co., Ltd.) were maintained in DMEM (HyClone; Cytiva) and RPMI 1640-medium (HyClone; Cytiva), respectively. Both media were supplemented with 10% fetal bovine serum (Gibco; Thermo Fisher Scientific, Inc.) and 1% penicillin/streptomycin. The cells were incubated at 37˚C in a humidified atmosphere of 95% air and 5% CO_2.

Reverse transcription-quantitative PCR (RT-qPCR). Total RNA was extracted from BEAS-2B and A549 cells using TRIzol^® (Invitrogen; Thermo Fisher Scientific, Inc.). RNA purity/concentration was measured by a NanoDrop 2000 (A260/A280 >1.8). For lncRNA analysis, 1 µg of total RNA was treated with DNase I (Thermo Fisher Scientific, Inc.) to remove genomic DNA contamination, followed by cDNA synthesis using M-MLV Reverse Transcriptase (Promega Corporation) with random hexamers (50 ng/µl) and dNTPs (0.5 mM) in a 20 µl reaction. The thermal profile for reverse transcription was as follows: 25˚C for 10 min (priming), 42˚C for 50 min (synthesis), and 70˚C for 15 min (inactivation). qPCR was performed using SYBR Premix Ex Taq™ (TaKaRa Bio, Inc.) on a QuantStudio 6 Flex system (Applied Biosystems; Thermo Fisher Scientific, Inc.) with the following thermal cycling conditions: An initial denaturation at 95˚C for 30 sec, followed by 40 cycles of 95˚C for 5 sec and 60˚C for 30 sec. A melt curve analysis was generated at the end of each run. lncRNA-specific primers (Table SI) were designed with the following criteria: amplicon length: 80-150 bp; GC content: 40-60%; no secondary structures (validated by OligoAnalyzer); and spanning exon-exon junctions (where applicable). β-actin was used as a reference gene. The 2^-ΔΔCq method (19) was utilized to assess fold changes. Raw Ct values and amplification curves are provided in Table SII.

Statistical analyses were performed using R software (version 4.1.2; https://cran.r-project.org). A two-sided P<0.05 was considered to indicate statistical significance. For the differential expression analysis of the 4 lncRNAs in LUAD (A549) versus normal bronchial (BEAS-2B) cells (Fig. S3), an unpaired Student's t-test was applied to compare the means between the two groups, with results presented as the mean ± SEM from three technical replicates; P<0.05 was considered to indicate a statistically significant difference.

The present study included 515 patients with LUAD from the TCGA database. The baseline clinicopathological characteristics of the entire cohort are summarized in Table SIII. The expression matrices of 19 cuproptosis-related genes and 16,876 lncRNAs were extracted from the TCGA-LUAD dataset. To identify biologically relevant lncRNAs, Pearson correlation analysis (|R|>0.3, P<0.001) was performed between each lncRNA and the 19 cuproptosis-related genes. This stringent threshold captured lncRNAs with moderate-to-strong co-expression, yielding 3,385 preliminary candidates (Fig. 1A). Prognostic lncRNA selection followed a multistep regression framework; univariate Cox analysis (P<0.05) revealed 52 lncRNAs significantly associated with OS. LASSO-penalized Cox regression (10-fold cross-validation, λ=0.032) reduced dimensionality, eliminating multicollinear lncRNAs (Fig. 2A and B). Multivariate Cox regression further refined the model, retaining only 4 lncRNAs (AC026355.2, AP000695.1, ARHGEF26-AS1, and AP005137.2) that independently predicted survival (P<0.01, Wald test). The correlation between the 19 cuproptosis-related genes and 4 cuproptosis-related lncRNAs is demonstrated in Fig. 1B.

Using the TCGA-LUAD cohort (n=515), a prognostic signature comprising four cuproptosis-related lncRNAs (AC026355.2, AP000695.1, ARHGEF26-AS1, and AP005137.2) identified by LASSO-penalized Cox regression were established (Fig. 2). Risk scores were computed for each patient, and the cohorts were stratified into high- and low-risk subgroups using cohort-specific median risk cut-offs. Visualization of risk score distributions (Fig. 3A and E), survival status (Fig. 3B and F) and signature lncRNA expression (Fig. 3C and G) confirmed distinct stratification. Critically, compared with their low-risk counterparts, high-risk patients had significantly shorter OS in both the training and test cohorts (P<0.001; Fig. 3D and H). The prognostic utility of the signature was further validated across additional endpoints, with significant disparities observed in the PFI, DSS and DFI between risk subgroups (Fig. S1A-C).

Patient cohorts were further stratified by age, sex and TNM stage. Across all the subgroups, compared with their high-risk counterparts, low-risk patients consistently demonstrated superior OS (Fig. 4A-F). To assess clinical relevance, the associations of the signature with key clinicopathological parameters were examined. High-risk patients had significantly more advanced TNM stages (Stage III-IV) than low-risk patients did (P=0.002; Fig. 4D). Elevated risk scores were also observed in patients with lymph node metastasis (N1-N3) compared with N0 patients (P<0.001; Fig. S2). Collectively, these findings highlight the utility of the signature in stratifying LUAD disease severity.

Principal component analysis (PCA) was conducted using four distinct feature sets in the TCGA-LUAD samples: i) the 4 signature cuproptosis-related lncRNAs, ii) 19 established cuproptosis-associated genes, iii) whole-genome expression profiles, and iv) risk stratification based on the lncRNA signature (Fig. 5). The first three principal components collectively accounted for 58.7% of the total variance (PC1: 32.4%; PC2: 16.1%; PC3: 10.2%), indicating that substantial biological heterogeneity was captured by the signature (Fig. 5A-C). Notably, high- and low-risk patients presented distinct spatial clustering in PCA space based on the signature alone (Fig. 5D). This clear separation confirms the effectiveness of the signature in stratifying patient risk groups.

High-risk patients with LUAD demonstrated an immunosuppressive tumour microenvironment characterized by elevated infiltration of regulatory T cells, M2 macrophages and myeloid-derived suppressor cells. Conversely, low-risk tumours contained predominantly cytotoxic CD8⁺ T cells, natural killer cells and activated dendritic cells (Fig. 6A). Pathway analysis revealed key immune dysregulation; i) high-risk enrichment: TGF-β signalling, PD-1 checkpoint and IL-10 pathways; and ii) low-risk activation: T-cell receptor signalling, NK cytotoxicity and IFN-γ response (Fig. 6B and C). This ‘immunologically active yet functionally impaired’ phenotype explains the following paradoxical prediction: Ηigh-risk tumours present T-cell exhaustion, with dominant immunosuppression precisely where checkpoint inhibitors show maximal efficacy.

Driver gene analysis was used to identify 15 variants with the greatest difference in frequency between risk groups (Fig. 6D and E). Critically, high-risk patients demonstrated a superior immunotherapy response (P<0.01), which was consistent with elevated TIDE scores (Fig. 6F). TMB in high-risk patients (Fig. 7A) correlated with poorer OS when patients were stratified by TMB status (P<0.001). Notably, within the matched TMB strata, high-risk patients consistently had lower survival rates than their low-risk counterparts did (Fig. 7B and C). This integrative analysis confirms the dual prognostic and immunotherapeutic predictive value of the signature.

Univariate and multivariate Cox regression analyses confirmed that the risk score was an independent predictor of OS in patients with LUAD, irrespective of clinicopathological covariates [univariate: Hazard ratio (HR)=1.835; 95% confidence interval (CI): 1.565-2.150, P<0.001; multivariate: HR=1.737, 95% CI 1.468-2.054, P<0.001; Fig. 8A and B]. To evaluate prognostic performance, time-dependent C-indices and Receiver Operating Characteristic (ROC)-Area under the curve (AUC) values were calculated. Compared with conventional clinical features, the risk score maintained superior discriminative accuracy over time (Fig. 8C). predictive performance for 1-, 3-, and 5-year OS significantly exceeded that of the other prognosticators (Fig. 8D), and it showed competitive power against key clinical variables (Fig. 8E). These findings establish the cuproptosis-related lncRNA signature as a robust independent prognostic factor in patients with LUAD.

To elucidate the regulatory mechanism of ARHGEF26-AS1, a ceRNA network was constructed based on its predicted microRNA (miRNA or miR) interactions. As illustrated in Fig. S3, the network reveals that ARHGEF26-AS1 potentially sponges multiple miRNAs, thereby modulating the expression of downstream target genes involved in key cellular processes such as cell proliferation and migration. Notable interactions include miRNAs such as hsa-let-7a-5p and hsa-miR-124-3p, which are linked to genes including ACTB, THBS1, VAMP3 and BEX3. This ceRNA network underscores the potential role of ARHGEF26-AS1 as a key regulatory lncRNA in cancer progression.

A prognostic nomogram integrating the risk score, age, sex and TNM stage was developed to predict 1-, 3-, and 5-year OS in patients with LUAD. ROC curve analysis demonstrated that the discriminative ability of the nomogram was superior to that of individual clinical factors (Fig. 9A). Calibration plots revealed close alignment between the predicted probabilities and observed outcomes across all the time points (Fig. 9B).

Drug sensitivity profiles were evaluated by estimating the half-maximal inhibitory concentration (IC₅₀) for 73 compounds from the GDSC database. Significant differential sensitivity emerged between risk groups, with high-risk patients presenting an enhanced response to all compounds (Fig. 10). The top 9 prioritized agents showing the greatest differential efficacies are presented in Fig. 10.

Differential expression of the four signature lncRNAs was quantified in LUAD (A549) and BEAS-2B cell lines using RT-qPCR. The expression of AC026355.2, AP000695.1 and AP005137.2 was significantly greater in A549 cells than in BEAS-2B cells (P<0.05), whereas ARHGEF26-AS1 expression was significantly lower (P=0.035) (Fig. S4). This experimental validation demonstrates directional concordance with prior bioinformatic analyses.

The present study has several limitations. First, the retrospective design of the TCGA analysis may introduce selection bias. Although internally validated, the present signature requires confirmation in a fully independent cohort, a challenge compounded by scarce annotation of these lncRNAs in public LUAD datasets. Second, the in vitro validation of the signature was limited to a single LUAD cell line (A549). Although these experiments provided preliminary support for the differential expression of the identified cuproptosis-related lncRNAs, the reliance on one cellular model restricts the generalizability of the currnet findings. To address this, future work will incorporate additional LUAD cell lines with diverse genetic backgrounds as well as primary patient-derived tissues to comprehensively validate the prognostic and therapeutic relevance of the signature. Third, the unknown subcellular localization of these lncRNAs hinders mechanistic insight. To address these issues, the authors plan to validate the signature in a prospective cohort of 200 immunotherapy-treated patients with LUAD; map lncRNA spatial expression via RNA fluorescence in situ hybridization; interrogate functions using CRISPRi under copper stress (with DLAT aggregation monitoring) and FRET-based copper flux assays; and evaluate ASO-based targeting in PDX models, including synergy testing with copper ionophores. Preliminary functional studies are underway, including CRISPRi knockdowns in copper-stressed A549 cells, copper flux assays and RNA pulldown/mass spectrometry to identify interactors (for example, FDX1 and DLAT). These will be fully reported in a subsequent mechanistic manuscript. Additionally, while internal validation showed robust performance, a slight AUC decrease in a preliminary Gene Expression Omnibus cohort (0.712 vs. 0.761) highlights the need for external validation. Further validation in prospective immunotherapy cohorts (for example, IMpower150) are strongly endorsed and these efforts are pursued.

To conclude, the present study establishes a computationally validated 4-lncRNA signature that stratifies patients with LUAD into distinct risk groups with different survival outcomes (OS, DSS and DFI) and predicts immunotherapy response. While the signature shows promise as a prognostic biomarker, its clinical utility requires prospective validation in immunotherapy-treated cohorts.