Comprehensive assessment of base excision repair (BER)-related lncRNAs as prognostic and functional biomarkers in lung adenocarcinoma: implications for personalized therapeutics and immunomodulation

Abstract

Background

Lung adenocarcinoma (LUAD) is the most prevalent subtype of lung cancer, and comprehending its molecular mechanisms is pivotal for advancing treatment efficacy. This study aims to explore the prognostic and functional significance of base excision repair (BER)-related long non-coding RNAs (BERLncs) in LUAD.

Methods

A risk score model for BERLncs was developed using the least absolute shrinkage and selection operator regression and Cox regression analysis. Model validation and prognostic evaluation were performed using Kaplan–Meier and receiver-operating characteristic curve analyses. Gene Ontology and Kyoto Encyclopedia of Genes and Genomes analyses were conducted to elucidate the potential biological functions of BERLncs. Comparative analyses were carried out to investigate disparities in tumor mutation burden (TMB), immune infiltration, tumor immune dysfunction and exclusion (TIDE) score, chemosensitivity, and immune checkpoint gene expression between the two risk groups.

Results

A predictive risk score model comprising 19 BERLncs was successfully developed. Patients were divided into high-risk and low-risk groups according to the median risk score. The high-risk subgroup exhibited significantly inferior overall survival. Functional enrichment analysis revealed pathways associated with lung cancer development, notably the neuroactive ligand–receptor interaction pathway. High-risk patients demonstrated elevated TMB, diminished TIDE scores, and an immunosuppressive tumor microenvironment, while low-risk patients displayed potential benefits from immunotherapy. Additionally, the risk model identified potential anticancer agents.

Conclusion

The risk score model based on BERLncs shows promise as a prognostic biomarker for LUAD patients, providing valuable insights for clinical decision-making, therapeutic strategies, and understanding of underlying biological mechanisms.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00432-023-05435-1.

Introduction

Lung cancer is the second most prevalent cancer globally and the leading cause of cancer-related mortality, posing a significant threat to the physical and mental well-being of the global population (Siegel et al. 2023). It can be classified into small-cell lung cancer and non-small-cell lung cancer based on histological types, with lung adenocarcinoma (LUAD) being the predominant histological subtype of non-small-cell lung cancer (Tan et al. 2021; Zhong et al. 2023). Due to the lack of evident clinical symptoms in the early stages, most cases of lung cancer are diagnosed at an advanced stage, resulting in a 5-year relative survival rate of only 5% for stage IV patients (Xie et al. 2021; Ye et al. 2021). Consequently, the available treatment options are limited and ineffective. Current treatment modalities for lung cancer mainly include surgical resection, radiotherapy, chemotherapy, and immunotherapy, with the latter experiencing significant breakthroughs in recent years. Immunotherapy, encompassing therapies targeting cytotoxic T-lymphocyte-associated antigen 4 (anti-CTLA-4), programmed cell death-1 (PD-1), and PD-L1, has shown remarkable efficacy in certain patient populations (Tan et al. 2020; Wu et al. 2021). However, its effectiveness remains limited (Brueckl et al. 2020; Li et al. 2020). Despite advancements in the understanding of tumor progression mechanisms and the promotion of early diagnosis, the survival rate of non-small-cell lung cancer patients remains low. Thus, there is an urgent need to comprehend the heterogeneity of LUAD patients and establish a more accurate and comprehensive risk model to assess patient risk, personalize treatment approaches concerning prognosis prediction, immunotherapy response prediction, drug selection, and ultimately improve the survival rate.

Long non-coding RNAs (lncRNAs) are characterized as nucleotide sequences longer than 200nt, lacking protein-coding function, and are detectable in various tissues and body fluids, such as serum, plasma, urine, and saliva (Chandra Gupta and Nandan Tripathi 2017). Once considered “junk DNA,” recent study has demonstrated that lncRNAs can regulate gene expression at the epigenetic, transcriptional, and post-transcriptional levels, playing a critical role in tumorigenesis and tumor progression (Gao et al. 2020). Therefore, lncRNAs have emerged as promising biomarkers for early cancer diagnosis and targeted therapy.

Base excision repair (BER), a prevalent DNA repair mechanism in eukaryotes, plays a crucial role in tumor progression. BER is primarily responsible for repairing small base damage caused by oxidative and alkylation damage (Almeida and Sobol 2007). It typically involves four stages: recognition, excision of damaged bases, filling of defective nucleotides, and gap sealing, all of which contribute to maintaining genome stability and suppressing tumor growth (Sweasy et al. 2006; Wallace et al. 2012). Key BER genes have been implicated in the development and progression of certain cancers, such as gastric and breast cancers (Jian et al. 2022; Shinmura et al. 2011). However, the comprehensive involvement of BER-related lncRNAs in tumor development remains understudied. Given the prognostic and immunotherapeutic response predictive potential of N7-methylguanosine-associated lncRNAs (Wei et al. 2022), it is reasonable to speculate that BER-associated lncRNAs possess a comparable potential. These lncRNAs are postulated to function as regulatory factors, influencing pivotal processes including immune response, angiogenesis, cell proliferation, and metastasis within the tumor microenvironment of LUAD. Through their interactions with tumor-specific signaling pathways and cytokines, these lncRNAs have the potential to emerge as critical prognostic factors for evaluating the effectiveness of immunotherapy and predicting outcomes in LUAD. Hence, this study aims to construct and validate a prognostic risk model based on BER-related lncRNAs, exploring their role in evaluating tumor mutation burden, tumor immune microenvironment, immunotherapy response, and drug sensitivity. The detailed workflow of the study is illustrated in Supplementary Fig. 1.

Materials and methods

Data sources

The RNA-seq data from LUAD patients with complete clinical information [including overall survival (OS), survival status, age, sex, stage, and tumor node metastasis (TNM) stage] and mutation data were obtained from the TCGA database (https://portal.gdc.cancer.gov/). The RNA-seq data were processed using annotated human GTF files from the Ensemble database (http://asia.ensembl.org), resulting in the identification of 19,509 mRNAs and 13,482 lncRNAs. A total of 59 normal samples and 526 tumor samples were included in the study, and patients with missing OS values were excluded to minimize statistical bias. Transcriptome data and clinical information from the IMvigor210 cohort were obtained using the R package “IMvigor210CoreBiology”.

To ensure the quality and reliability of the research data, we implemented various measures. Initially, we employed the robust multi-array average (RMA) method for normalizing the raw gene expression data. This ensured comparability of gene expression features across different samples while reducing the impact of batch effects and technical variations. Additionally, we applied the ComBat algorithm to correct gene expression data from diverse batches, platforms, or centers, thereby enhancing the accuracy of subsequent analyses. During gene selection, only genes surpassing a predefined threshold expression level were considered, and low-expressed genes were filtered out to mitigate any insignificant interference with the analysis. Moreover, to satisfy the assumption of normality, we logarithmically transformed (log2) the gene expression values. In terms of data exclusion, we adhered to agreed-upon criteria. Primarily, we discarded samples exhibiting extensive missing data to ensure the reliability of the analysis results. Besides, we employed robust statistical methods to detect and exclude outliers, preventing their unfounded influence on the outcomes. Finally, any duplicate samples were eliminated to prevent excessive representation of specific cases within the analysis.

Selection of BER genes and BER-related lncRNAs

The study incorporated 400 BER-related genes (BRGs) obtained from various databases and functional pathways. Among these, 222 BRGs were identified from a previous study (Wood et al. 2001), and additional 68, 147, and 287 BRGs were extracted from the DNArepairK database (Babukov et al. 2021), REPAIRtoire database (Milanowska et al. 2011), and Gene Cards database (Safran et al. 2010), respectively. Furthermore, 221 genes were obtained from 26 BER-related functional pathways using the GSEA database, including GO_BASE_EXCISION_REPAIR, GO_BASE_EXCISION_REPAIR_AP_SITE_FORMATION, GO_BASE_EXCISION_REPAIR_GAP_FILLING, GO_GLOBAL_GENOME_NUCLEOTIDE_EXCISION_REPAIR, GO_NUCLEOTIDE_EXCISION_REPAIR, GO_NUCLEOTIDE_EXCISION_REPAIR_COMPLEX, GO_NUCLEOTIDE_EXCISION_REPAIR_DNA_DAMAGE_RECOGNITION, GO_NUCLEOTIDE_EXCISION_REPAIR_DNA_DUPLEX_UNWINDING, GO_NUCLEOTIDE_EXCISION_REPAIR_DNA_GAP_FILLING, GO_NUCLEOTIDE_EXCISION_REPAIR_DNA_INCISION, GO_NUCLEOTIDE_EXCISION_REPAIR_PREINCISION_COMPLEX_ASSEMBLY, GO_NUCLEOTIDE_EXCISION_REPAIR_PREINCISION_COMPLEX_STABILIZATION, GO_TRANSCRIPTION_COUPLED_NUCLEOTIDE_EXCISION_REPAIR, GO_UV_DAMAGE_EXCISION_REPAIR, KEGG_BASE_EXCISION_REPAIR, KEGG_NUCLEOTIDE_EXCISION_REPAIR, REACTOME_BASE_EXCISION_REPAIR, REACTOME_BASE_EXCISION_REPAIR_AP_SITE_FORMATION, REACTOME_DISEASES_OF_BASE_EXCISION_REPAIR, REACTOME_GLOBAL_GENOME_NUCLEOTIDE_EXCISION_REPAIR_GG_NER, REACTOME_NUCLEOTIDE_EXCISION_REPAIR, REACTOME_PCNA_DEPENDENT_LONG_PATCH_BASE_EXCISION_REPAIR, REACTOME_POLB_DEPENDENT_LONG_PATCH_BASE_EXCISION_REPAIR, REACTOME_TRANSCRIPTION_COUPLED_NUCLEOTIDE_EXCISION_REPAIR_TC_NER_, WP_BASE_EXCISION_REPAIR and WP_NUCLEOTIDE_EXCISION_REPAIR. After removing duplicated genes, non-coding genes, and pseudogenes, a set of 400 BRGs was retained for correlation analysis with lncRNAs in the TCGA-LUAD dataset. Pearson correlation analysis was conducted, and lncRNAs with an absolute correlation coefficient greater than 0.4 (|R|> 0.4) and a p value less than 10 to the power of − 60 (p value < 10e−60) were selected for subsequent analysis. The lncRNAs filtered out based on the aforementioned thresholds are defined as BER-related lncRNAs (BERLncs).

Construction of a risk score model for BER-related lncRNAs

A random division was performed on 494 LUAD patients, resulting in a training set (n = 330) and a validation set (n = 164). Using the training set, a risk model was constructed and then validated using the entire cohort and the validation set. Univariate Cox regression analysis was conducted to identify BERLncs significantly associated with overall survival (OS) in the training set (p value < 0.05). To avoid overfitting the data with multiple variables, LASSO regression analysis was performed on these BERLncs using the “glmnet” package with 1000 iterations. Furthermore, multivariate Cox regression analysis was carried out to select 19 BERLncs for constructing the risk scoring model. The risk score (Lin et al. 2023; Xiong et al. 2023; Xu et al. 2022; Ye et al. 2022), representing the prognostic feature of patients, was calculated as follows:

$$\mathrm{BERLncs}={\sum }_{i=1}^n(\mathrm{Coef}(i)\times x(i)),$$

where $\mathrm{Coef}(i)$ represents the regression coefficient and $x(i)$ represents the expression of each BERLnc. Additionally, the association between BER genes and the selected 19 BERLncs was visualized using heat maps.

Evaluation and validation of the BERLncs risk model

A risk score was computed for each patient, and subsequently, the patient cohort was stratified into low-risk and high-risk subgroups based on the median value. To assess the differences in OS between these subgroups, Kaplan–Meier survival curves were employed. The statistical software packages “SurvMiner” and “Survival” were utilized as powerful tools to facilitate this analysis. Additionally, the prognostic effect of the model was evaluated using ROC curves and C-index curves. To gain a comprehensive understanding of the distribution of samples across different risk subgroups, three-dimensional principal component analysis (3d-PCA) was applied to visualize the high-dimensional expression data of all genes, BER-related genes, 2012 BERLncs, and the 19 selected BERLncs within the model. This visualization process was performed using the “limma” and “scatterplot3d” packages, enabling a comprehensive examination of the spatial arrangement of samples.

Relationship between risk score and clinicopathologic variables

In this study, we performed a comprehensive stratification of clinical and pathological characteristics in LUAD patients, considering variables such as age, sex, race, stage, primary tumor (T), regional lymph nodes (N), and distant metastasis (M). The aim was to investigate the relationship between the risk score and these clinical variables. Furthermore, we analyzed the disparity in overall survival (OS) between the two risk subgroups. To ascertain the potential of the risk score as an independent prognostic factor for LUAD patients, we conducted both univariate and multivariate Cox regression analyses, incorporating the clinical variables and risk score. This analysis aimed to determine if the risk score could serve as a reliable indicator in prognostic evaluation. Additionally, we developed a nomogram based on the results obtained from the multivariate Cox regression analysis. This nomogram was designed to predict the 1-, 3-, and 5-year survival rates of patients. To assess the accuracy of the prediction, we drew a calibration curve, which provides an intuitive method to evaluate whether the prognostic model accurately reflects the actual status of the sample. A calibration curve in which the points align closely to the 45-degree line indicates a high level of accuracy in the model.

Functional enrichment analysis of differentially expressed lncRNA genes

Differential expression analysis of lncRNA genes between high-risk and low-risk subgroups was conducted using the R package “limma.” The screening threshold of FDR < 0.05 and ∣log FC∣ ≥ 1 was applied to identify significantly differentially expressed genes. Subsequently, to unravel the potential biological functions and pathways associated with these genes, GO and KEGG enrichment analyses were performed. The R package “clusterProfiler” was utilized for these analyses. The GO analysis was categorized into three domains: biological process (BP), cellular component (CC), and molecular function (MF). Enriched terms with a p value < 0.05 were considered significant. To execute the GO and KEGG analyses, we employed the online bioinformatics database known as the Database for Annotation, Visualization, and Integrated Discovery (DAVID). This resource (https://david.ncifcrf.gov/summary.jsp) facilitated the exploration of functional annotations and pathway information linked to the differentially expressed lncRNA genes.

Association of risk score models with 13 types of immune function

To quantify the infiltration levels of immune-related functions, we employed the single-sample gene set enrichment analysis (ssGSEA) algorithm, which is widely utilized in medical research. The ssGSEA algorithm allowed us to assess the level of immune-related function infiltration using RNA sequence data obtained from TCGA-LUAD (Finotello and Trajanoski 2018). To perform the analysis, we assembled a specific gene set related to immune functions, as documented in Table S1 from a previous study (Ye et al. 2022).

Potential compounds of targeted BERLncs risk model in clinical therapy

To identify potential compounds for targeted therapy in the context of the BERLncs risk model, we utilized the R package “pRRophetic” and its associated pRRophetic algorithm. This algorithm allowed us to predict the maximum half-inhibitory concentration (IC50) of chemotherapy drugs specifically for LUAD patients using the cancer drug sensitivity genomics (GDSC) dataset (Geeleher et al. 2014). It is worth noting that a lower IC50 value indicates a higher sensitivity to the drug in question. This information can guide clinicians in determining the most suitable compounds for targeted therapy within the context of the BERLncs risk model.

Immunotherapy research based on the risk model

The R software package “maftools” was employed to calculate the tumor mutation burden (TMB) in patients belonging to the high-risk and low-risk subgroups. Subsequently, KM survival analysis was conducted to examine the differences in overall survival (OS) among patients with varying TMB states. To predict the immune response of LUAD patients to checkpoint blockade (ICB) therapy after treatment, the tumor immune dysfunction and rejection (TIDE) score were calculated using the TIDE tool (http://tide.dfci.harvard.edu/). Additionally, the expression levels of other immune checkpoint molecules, such as CD8, CD274, IFNG, Merck18, among others, were evaluated in LUAD patients.

Evaluation and validation of the model using the IMvigor210 clinical cohort

The IMvigor210 clinical cohort study focused on assessing the effectiveness of Atezolizumab, a PD-L1 targeting antibody, in platinum-treated patients with locally advanced or metastatic urothelial carcinoma (Powles et al. 2014). In this study, the R software package “timeROC” was utilized to determine the accuracy of the relevant BERLncs in predicting the probability of survival in patients with bladder cancer within the LASSO model. Furthermore, box plots were generated to identify variations in risk scores among patients receiving PD-L1 inhibitors (specifically Atezolizumab).

Calculation of mRNAsi in LUAD patients

Using patient gene expression data, we employed the OCLR algorithm to calculate the mRNAsi in the context of lung adenocarcinoma. The mRNAsi serves as an indicator of tumor dedifferentiation characteristics and aids in predicting patient outcomes (Malta et al. 2018). Based on the median mRNAsi score, patients were stratified into high and low mRNAsi subgroups. The differences in OS between these subgroups were evaluated using K–M survival analysis. Additionally, mRNAsi levels were compared across different subgroups based on clinical features, such as age, sex, stage, and TNM stage.

Statistical analysis

Statistical analysis was performed to compare groups and assess significance. The Kruskal–Wallis test compared differences among multiple groups, while the Wilcoxon test compared differences between two groups. Survival analysis utilized the Kaplan–Meier method and log-rank test. Univariate Cox regression models calculated hazard ratios for candidate genes. All p values were bidirectional, with p < 0.05 considered significant. Analysis was conducted using R software version 4.2.2.

Result

Identification of BER-related lncRNAs in LUAD patients

Using Pearson correlation analysis, we identified a total of 1741 BERLncs based on the following criteria: |R|> 0.4 and p value < 10e−60. To visualize the co-expression relationship between BER-related genes (BRGs) and BERLncs, we employed the R package “ggalluvial” to generate a Sankey diagram (Supplementary Fig. 2).

Construction and validation of a BER-related lncRNAs risk model for LUAD patients

A total of 494 LUAD patients were randomly divided into a training cohort (n = 330) and a test cohort (n = 164) in a 3:2 ratio. Supplementary Table 1 presents the clinical parameters, demonstrating no significant differences between the training and testing cohorts (p value > 0.05). The random division ensured a well-balanced grouping and minimized statistical bias. To construct a reliable model for predicting the prognostic risk of LUAD patients, we initially identified 82 BERLncs significantly associated with OS in LUAD patients through univariate Cox regression analysis in the training cohort (p value < 0.05) (Supplementary Table 2). Subsequently, LASSO regression analysis (Fig. 1A, B) and multivariate Cox regression analysis were performed on these BERLncs to effectively reduce the features in high-dimensional data and optimize the indicators for predicting clinical outcomes. Finally, 19 BERLncs were selected (Supplementary Table 3), and based on their expression levels and risk score coefficients, the BERLncs risk score model was constructed to assess the prognostic risk of LUAD patients. Additionally, we validated the grouping ability of the risk model using 3dPCA, which revealed that the 19 risk-related BERLncs exhibited the best grouping ability (Fig. 1C), while the grouping ability of entire genes, 400 BRGs, and 2012 BERLncs was comparatively poor (Fig. 2D–F). This indicates that our risk model effectively distinguishes between high- and low-risk patients.

Fig. 1.

Construction and validation of the BERLncs model for LUAD. A Lasso regression coefficients were obtained for lambda values. B The optimal Lambda value corresponding to the minimum Partial Likelihood Deviation was selected, and 38 lncRNAs were screened for subsequent analysis. C Three-dimensional Principal Component Analysis (3dPCA) based on 19 BERLncs in the entire cohort. D 3dPCA analysis based on 400 BRGs in the entire cohort. E 3dPCA analysis based on 2012 BER-associated lncRNAs in the entire cohort. F 3dPCA analysis based on all genes in the entire cohort

Fig. 2.

Prognostic value of 19 BERLncs in the training group. A Kaplan–Meier survival curve analysis of the training cohort. A p value < 0.05 indicates a statistically significant difference. B Distribution of risk scores in the training cohort (the x-axis represents LUAD patients arranged according to the risk score, and the y-axis represents the risk score value for each patient). C Scatter plot of survival status and risk score in the training cohort (the x-axis represents LUAD patients arranged by risk score, and the y-axis represents the survival time of each patient). D Heatmap of the expression profiles of 19 BERLncs. E Receiver-Operating Characteristic (ROC) curve of the BERLncs model showing the Area Under the Curve (AUC) values for 1, 3, and 5 years in the entire cohort

Based on the median risk score, LUAD patients were divided into high- and low-risk subgroups. The Kaplan–Meier curve analysis demonstrated a statistically significant difference in OS between the high- and low-risk subgroups (p value < 0.001), with the low-risk subgroup exhibiting superior OS compared to the high-risk subgroup (Fig. 2A). The risk score curve and scatter plot further illustrated that higher risk scores correlated with higher mortality rates (Fig. 2B, C). Moreover, the heat map displayed up-regulation of AC009970.1, AL031600.2, AL590822.2, AL078645.1, and NAV2-AS2 in the low-risk group and down-regulation in the high-risk subgroup (Fig. 2D), suggesting that high expression of these lncRNAs is associated with improved patient survival. The findings in the testing and entire cohorts were consistent with those in the training cohort (Supplementary Fig. 3A–H). ROC curve analysis demonstrated that the BERLncs risk model accurately predicted patients’ OS, with AUC values of 0.789 at 1 year, 0.769 at 3 years, and 0.826 at 5 years (Fig. 2E).

Association between risk score models and clinical features

The association between risk scores and clinicopathological characteristics was analyzed. Patients who died had significantly higher risk scores compared to those who survived (p value < 0.01) (Supplementary Fig. 4 A). Moreover, an increasing trend in overall risk scores was observed with increasing T, N, and stages (p value < 0.05) (Supplementary Fig. 4 B–D). These findings suggest that low-risk scores are associated with early tumor stages, while high-risk scores are associated with advanced tumor stages. No correlation was found between risk scores and age, sex, race, or M stage (Supplementary Fig. 4 E–H).

When analyzing the differences in OS between the two risk subgroups under different clinicopathological stratifications, no statistical difference in OS was observed between the two risk subgroups in patients at Stage IV, Stage III, N2, and M1 (p value > 0.05) (Supplementary Fig. 5 A–D). However, statistically significant differences in OS between the two risk subgroups were observed in the remaining clinicopathological stratifications (p value < 0.05), with the high-risk group exhibiting significantly lower survival rates compared to the low-risk group (Supplementary Fig. 5 D–R). Due to the small number of patients in the T4 and Asian stratifications, these results were not considered (Supplementary Fig. 5 S–T).

Univariate and multivariate Cox regression analyses were performed to evaluate the association between BERLncs risk score and clinicopathological features, including age, TNM stage, gender, race, and stage (Fig. 3A, B). The results indicated that stage, T stage, N stage, and risk score were all associated with prognosis. However, only the risk score could be identified as an independent prognostic factor in LUAD patients (HR = 1.008, p = 0.006). To assess the predictive effect of BERLncs risk score and other clinicopathological features on OS in LUAD patients, ROC curve and C-index curve analyses were conducted. The results demonstrated that the BERLncs risk score had higher accuracy in predicting 1-year, 3-year, and 5-year OS in LUAD patients compared to the traditional TNM stage and other clinical variables (Fig. 3C–F). Based on the results of the multivariate Cox regression analysis, a nomogram was constructed to predict 1-year, 3-year, and 5-year survival by integrating risk scores and other clinicopathological variables (Fig. 3G). Calibration curves showed good agreement between actual and predicted survival in LUAD patients (Fig. 3H).

Fig. 3.

Influence of BERLncs risk score and clinicopathological features on the prognosis of patients with LUAD. A Results of univariate Cox regression analysis. B Results of multivariate Cox regression analysis. C C-index curves for risk scores and clinicopathological features; D–F 1-year, 3-year, and 5-year ROC curves and respective Area Under the Curve (AUC) values for risk score and other clinicopathological features. G Construction of a nomogram based on risk scores and multiple clinicopathological features. H Calibration curves depicting the relationship between the observed and nomogram-predicted OS rates

Functional analysis of differentially expressed lncRNA genes

Functional annotation and classification of the differentially expressed lncRNA genes were performed using Gene Ontology (GO) analysis. The results revealed that the differentially expressed genes were enriched in various biological processes (BP), cell components (CC), and molecular functions (MF). The BP analysis showed enrichment in humoral immune response, while the CC analysis indicated involvement in apical structures and intermediate fibers. The MF analysis highlighted peptidase inhibitor activity as a significant function (Fig. 4A). Figure 4B displays representative differential genes corresponding to the most valuable biological processes.

Fig. 4.

Functional analysis of differential BERLncs genes. A, B GO enrichment analysis. C, D KEGG enrichment analysis

In addition to GO analysis, Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis was conducted to identify the significant signaling pathways associated with the differentially expressed lncRNA genes. The top four statistically significant pathways for differential gene enrichment were “Staphylococcus aureus infection” (hsa05150), “Neuroactive ligand–receptor interaction” (hsa04080), “Retinol metabolism” (hsa00830), and “Drug metabolism–cytochrome P450” (hsa00982) (Fig. 4C). Figure 4D illustrates some representative differential genes and their corresponding KEGG pathways.

Analysis of TMB, TIME, and immunotherapy response based on the BERLncs model

The TMB of the two risk subgroups was calculated using the “maftools” package. The results demonstrated that the high-risk subgroup had a significantly higher TMB than the low-risk subgroup in the entire cohort (Fig. 5A), which was consistent with findings in the training and testing cohorts (Supplementary Fig. 6A-B). The optimal TMB cut-off value was determined using the “survminer” package, and patients were categorized into high-TMB and low-TMB groups accordingly. Survival analysis showed no statistically significant difference in OS between the two subgroups in the entire cohort (Fig. 5B). However, when combining the BERLncs risk score and TMB, the survival analysis revealed that high-TMB patients in the low-risk subgroup had a more prolonged survival compared to low-TMB patients in the high-risk subgroup, indicating the superior prognostic significance of the BERLncs risk model over TMB in LUAD (Fig. 5C). Similar results were observed in the training and testing cohorts (Supplementary Fig. 6C–F).

Fig. 5.

Analysis of TMB, TIME, and response to immunotherapy based on the BERLncs model in the entire cohort. A Comparison of TMB between high- and low-risk subgroups. B Kaplan–Meier curve analysis of OS between high-mutation and low-mutation subgroups. C Kaplan–Meier curve analysis of OS based on TMB and BERLncs risk score. D Differences in 13 immune events between high- and low-risk subgroups were assessed using single-sample gene set enrichment analysis (ssGSEA). E Differences in TIDE scores between high- and low-risk subgroups were assessed using the TIDE algorithm. F, G Exclusion and dysfunction scores for high- and low-risk subgroups. H–N Expression levels of seven immune checkpoint genes between high- and low-risk groups

To analyze the TIME of different risk subgroups in LUAD patients, single-sample gene set enrichment analysis (ssGSEA) was performed. The results indicated significant statistical differences in multiple immune functions between the two risk subgroups in the entire cohort. Several immune functions, such as T-cell co-inhibition, checkpoint, and T-cell co-stimulation, were significantly suppressed in the high-risk group (Fig. 5D). Consistent findings were observed in the training and testing cohorts (Supplementary Fig. 7A-B).

The immunotherapy response of LUAD patients was evaluated using the TIDE score. Surprisingly, the high-risk subgroup had a significantly higher TIDE score compared to the low-risk subgroup in the entire cohort (Fig. 5E). Additionally, the high-risk group exhibited a higher Exclusion score (Fig. 5F), while the low-risk group showed a higher Dysfunction score (Fig. 5G). Furthermore, the expression of seven immune checkpoint genes was analyzed between the high and low-risk groups. The results revealed that CD8, Merck18, and IFNG were upregulated in the low-risk group (Fig. 5H–J), whereas MDSC and CAF were upregulated in the high-risk group (Fig. 5K–L). TAMM2 exhibited higher expression in the low-risk group (Fig. 5M), while there was no statistically significant difference in CD274 (PD-L1) expression between the high and low-risk groups (Fig. 5N). Similar results were observed in the training and testing cohorts (Supplementary Fig. 7C–V).

Validation of the impact of BERLncs on immune therapy efficacy in IMvigor210

In the IMvigor210 dataset, the expression levels of target genes (VAC14-AS1, ABCA9-AS1, FRMD6-AS1, and NPSR1-AS1) identified by the LASSO model were used to conduct K-M survival analysis on high- and low-risk groups. However, no statistically significant difference in OS was observed between the two subgroups (Supplementary Fig. 8A). The ROC curve analysis showed that the model could still predict the 1-, 3-, and 5-year survival rates of bladder cancer with an AUC greater than 0.5 (Supplementary Fig. 8B). Unfortunately, the risk scores of these target genes did not show a significant difference between patients with complete response (CR)/partial response (PR) and those with stable disease (SD)/disease progression (PD) (Supplementary Fig. 8C). This suggests that additional factors beyond these target genes may be required to predict the response of patients to immune therapy accurately.

Relationship between BERLncs risk score and chemosensitivity

Using the pRRophetic algorithm to calculate the IC50 values of chemotherapy drugs in the GDSC database, significant differences in IC50 values were observed between high- and low-risk subgroups for 76 chemotherapy drugs. This indicates that the BERLncs model has the potential to predict chemotherapy sensitivity. The top 15 chemotherapy drugs with significant differences are shown in Supplementary Fig. 9 A-O, which could be further analyzed in the context of LUAD patients. Therefore, incorporating these drugs into clinical practice may provide more benefits.

Correlation between mRNAsi and clinical features in LUAD

The correlation between mRNAsi and clinical features in LUAD was assessed. There was no statistically significant difference in overall survival between the high and low mRNAsi groups (p = 0.182) (Fig. 6A). However, the mRNAsi in cancer samples was significantly higher than in standard samples (Fig. 6B). In terms of clinical features, male patients exhibited significantly higher mRNAsi than female patients (Fig. 6C), and mRNAsi showed an overall increasing trend with tumor progression (stages I–IV, T1–T4, M0–M1) (Fig. 6D–F). This may be attributed to the fact that most lung cancer patients are diagnosed in the advanced stages.

Fig. 6.

The correlation of mRNAsi with LUAD. A Kaplan–Meier survival curve showed no significant difference in survival rate between the high and low mRNAsi groups. B Differences in mRNAsi between normal (57 samples) and LUAD (512 samples) tissues. C–F Distribution of mRNAsi according to sex, tumor stage, T stage, and M stage

Discussion

In this study, we developed a risk score model called BERLncs, which incorporates 19 long non-coding RNAs (lncRNAs) associated with BER and clinical-pathological information from patients with LUAD in the TCGA database. Notably, we employed a repeated LASSO regression method with 1000 iterations to identify a robust set of lncRNA features, surpassing the performance of traditional LASSO regression. The 19 lncRNAs included in the model were AC009970.1, AC011611.2, AL078645.1, AC018529.1, SNHG6, AL590822.2, AC025741.1, AC019186.1, AC008957.1, AC026355.2, NHS-AS1, AL117335.1, FAM66C, AC008883.1, NAV2-AS2, AL031600.2, AL162632.3, AC103591.3, and AL442125.1. Previous studies have reported the functional significance of some of these lncRNAs. For instance, SNHG6 promotes proliferation and inhibits apoptosis in non-small-cell lung cancer (Dong et al. 2020), AC026355.2 serves as a prognostic biomarker for LUAD (Liu et al. 2022), and NAV2-AS2 plays a significant role in LUAD prognosis due to its aberrant expression (Hu et al. 2019). Our study confirmed the prognostic value of these lncRNAs. However, other lncRNAs identified in our model require further investigation and validation through subsequent studies.

We stratified LUAD patients into high-risk and low-risk subgroups based on the median risk score derived from the BERLncs model. Subsequently, we evaluated the prognostic performance of the BERLncs risk score model using K-M and ROC curves. Our findings revealed a positive correlation between the BERLncs risk score and patient mortality across all datasets. Moreover, SNHG6, AC011611.2, and AC019186.1 exhibited high expression levels in the high-risk subgroup, suggesting their potential as tumor antigens for developing therapeutic vaccines against LUAD.

Analyzing the relationship between the BERLncs risk score and clinical-pathological features of LUAD patients, we observed that a low-risk score was indicative of early-stage tumors, while a high-risk score correlated with advanced-stage tumors. This observation may be attributed to the fact that most patients are diagnosed in the middle and late stages of the disease. Univariate and multivariate Cox regression analyses demonstrated that the BERLncs risk score served as an independent prognostic factor for LUAD, outperforming other clinical–pathological features, such as age, gender, race, stage, and TNM stage in predicting OS. Based on these analyses, we conclude that the BERLncs model exhibits high accuracy and provides valuable insights for future studies.

GO analysis of the differentially expressed genes identified in our study revealed their involvement primarily in the humoral immune response within the biological process category. Previous research has highlighted the importance of the BER mechanism in maintaining genomic stability and immune function in B cells (Bahjat and Guikema 2017; Guo et al. 2023; Li et al. 2023), while immune dysfunction has been shown to promote tumor development. Molecular function analysis revealed that the differentially expressed genes were primarily associated with serine-type peptidase inhibitor activity. Regarding cellular composition, these genes were predominantly distributed in the cell membrane and intermediate cytoskeleton regions. KEGG enrichment analysis identified pathways, such as “Staphylococcus aureus infection” (hsa05150) and “Neuroactive ligand-receptor interaction” (hsa04080) as being enriched by the differentially expressed genes. These findings suggest a potential association between these pathways and the risk of LUAD, which aligns with the previous research. Staphylococcus aureus infection is an inflammatory pathway known to promote cancer development, particularly chronic inflammation (Coussens and Werb 2002). A recent study has shown that Staphylococcus aureus-induced LTA encourages proliferation in NSCLC cell lines derived from adenocarcinoma (Hattar et al. 2017). This study found that the neuroactive ligand–receptor interaction pathway is one of the most important pathways in LUAD. The research shows that the expression levels of multiple neurotransmitters are significantly increased under chronic stress conditions in the tumor microenvironment. Specifically, the high expression levels of certain neuroactive ligands such as acetylcholine have been shown to promote invasion and metastasis of LUAD through the α5-nAChR/FHIT pathway (Jiao et al. 2023). This suggests that neuroactive ligands and their corresponding receptors may be involved in regulating the proliferation and metastatic ability of lung adenocarcinoma cells. In addition, the study observed that D2 dopamine receptors (D2 DA receptors) are overexpressed in lung adenocarcinoma. Stimulation of D2 DA receptors can inhibit the ERK1/2 and AKT signaling pathways, thereby suppressing the proliferation of tumor cells (Roy et al. 2017). This finding indicates the important role of neurotransmitter receptors in lung cancer. Drugs targeting neurotransmitter receptors are also widely used in the treatment of lung cancer, such as EGFR inhibitors. Therefore, a deeper understanding of the pathway will help provide new insights into the development and progression of lung adenocarcinoma and provide a theoretical basis for potential treatment targets.

In recent years, immune checkpoint blockade (ICB) therapy has emerged as a crucial treatment option for lung cancer (Endris et al. 2019). TMB and TIDE scores are vital biomarkers for predicting the efficacy of ICB therapy. High TMB is associated with better response to ICB therapy due to the increased production of neoantigens (Maleki Vareki 2018), which facilitates immune recognition of the tumor (Zhang et al. 2023). However, it is important to note that only a small fraction of neoantigens generated by mutated tumor genes are immunogenic and capable of eliciting T-cell immune responses (Yadav et al. 2014). Therefore, TMB may not always correlate with the effectiveness of immune therapy. The TIDE algorithm integrates gene expression features related to T-cell dysfunction and T-cell exclusion to simulate tumor immune evasion. The TIDE score, derived from pre-treatment tumor characteristics, predicts the response to ICB therapy. A higher TIDE score indicates a greater likelihood of immune evasion, implying poorer therapeutic efficacy of ICB treatment in patients (Jiang et al. 2018). Based on our research results, we have found that high-risk patients exhibit higher Exclusion scores and CAF gene expression levels. CAF cells have been confirmed to hinder the migration of immune cells into the tumor microenvironment (Denton et al. 2018). This indicates that high-risk patients might employ T-cell exclusion as a mechanism for immune evasion. Furthermore, study has also shown that TAMM2 cells can produce cytokines, such as IL-10 and TGF-β, which inhibit tumor immune response and facilitate tumor progression (Meng et al. 2022). Our findings reveal that low-risk patients have higher Dysfunction scores and increased TAMM2 gene expression levels, suggesting that T-cell dysfunction could be a potential mechanism contributing to immune escape. Nevertheless, relying solely on TIDE scores for predicting patient prognosis after receiving ICB therapy has limitations. The authors suggest the incorporation of additional data types and methods to improve the TIDE algorithm and a comprehensive combination of diverse biomarkers to achieve a more accurate prediction of ICB therapy effectiveness (Jiang et al. 2018). In contrast to TMB and TIDE scores, we propose that evaluating the TIME and the performance of publicly available immune checkpoint molecules in two risk groups can provide superior predictions of ICB therapy efficacy, as they directly reflect the patient’s immune status.

In our study, we observed that the high-risk group exhibited higher TMB and lower TIDE scores, suggesting a potentially favorable prognosis for ICB treatment. However, ssGSEA analysis indicated that immune functions were largely suppressed in the high-risk group, making it more prone to immune evasion. Moreover, the expression levels of most immune checkpoint-positive biomarkers, including CD8, Merck18, and IFNG, were significantly lower in the high-risk group compared to the low-risk group. Conversely, most immune checkpoint-negative biomarkers displayed higher expression levels in the high-risk group. Taking these findings into consideration, we can make a more accurate hypothesis that the low-risk group may have a higher likelihood of benefiting from immunotherapy. This is because BERLncs may serve as critical regulatory factors in modulating immune cell infiltration, activity, and functionality within the tumor microenvironment. They also affect the regulation mechanisms of immune modulation pathways and immune checkpoint regulators. Through their interactions with immune-related genes, BERLncs may exert a direct influence on the efficacy of immunotherapy. Further research is required to investigate the interactions between BERLncs and specific immune cell types, as well as their regulatory mechanisms within the tumor microenvironment. The IMvigor210 dataset initially focused on bladder cancer and provides valuable insights into immune regulation across various cancer types. By leveraging this dataset for external validation, we evaluated the effectiveness of immune-related features in lung cancer and explored potential similarities with other cancers. While no significant differences were observed in the efficacy of Atezolizumab immune therapy between high-risk and low-risk groups, our analysis demonstrated that BERLncs can predict 1-, 3-, and 5-year OS in bladder cancer patients (AUC > 0.5). This suggests a potential association between BERLncs and bladder cancer. Therefore, further confirmation with additional real-world clinical cohort data is necessary. Additionally, the combination of risk score, TIME, TMB, TIDE scores, and multiple immune therapy biomarkers may enhance the accuracy and predictive performance of ICB therapy outcomes.

Drug resistance poses a significant challenge in cancer treatment, as most lung cancer patients eventually develop resistance to anticancer drugs (Zhang et al. 2022). To provide more precise clinical guidance, we utilized the GDSC database to evaluate the relationship between the BERLncs risk model and the sensitivity of 76 anticancer drugs. Substantial differences in IC50 values were observed between the two risk subgroups for the majority of these drugs, indicating that the BERLncs score can serve as a potential biomarker for predicting drug sensitivity in LUAD patients and guiding personalized treatment.

The mRNA stemness index (mRNAsi) is a measure of dedifferentiation in cancer. During cancer progression, tumor cells gradually lose their differentiated phenotype and acquire stem cell-like characteristics. This dedifferentiation process is particularly pronounced in metastatic tumors (Malta et al. 2018). Cancer stem cells (CSCs) play critical roles in tumor differentiation, metastasis, drug resistance, and epigenetic changes, making them valuable targets in cancer research (Hou et al. 2022). In our study, we conducted a preliminary analysis of the relationship between clinical features and mRNAsi scores in LUAD. Consistent with previous research (Wan et al. 2022), we found significantly higher mRNAsi scores in LUAD patients compared to normal individuals. Denton et al. confirmed that cancer-associated fibroblasts (CAFs) can inhibit immune cell function and impede their migration into the tumor microenvironment, thereby maintaining tumor stemness and promoting cancer development (Denton et al. 2018). Our findings revealed that the expression of CAF genes was significantly higher in the high-risk subgroup compared to the low-risk subgroup. Furthermore, ssGSEA analysis indicated an association between our high BERLncs risk score and an immunosuppressive tumor microenvironment in LUAD. These results suggest that BERLncs may directly modulate tumor stemness and immune suppression to drive cancer development.

However, our study has certain limitations that warrant acknowledgement. It is an inherently biased retrospective analysis, and further prospective investigations are necessary to confirm our findings. Moreover, the sample size primarily relied on the TCGA database, which limits the generalizability of our model. Validation using LUAD samples from other databases is crucial. Additionally, some BER-related lncRNAs identified in our study have not been previously reported in the context of LUAD. Further in vivo and in vitro research is required to elucidate their mechanisms of action in LUAD. Nonetheless, our study provides a valuable reference for the development of prognostic markers in LUAD.

Overall, our risk scoring model offers individualized treatment strategies based on molecular biomarkers for patients with LUAD. By conducting an analysis of gene expression and clinical features, the model calculates a risk score that enables physicians to more accurately predict the patient’s response to immunotherapy and select chemotherapy drugs suitable for treating LUAD patients. Consequently, this optimizes the treatment plan by aiding doctors in making well-informed decisions regarding treatment options. As a clinical decision support tool, this model demonstrates its capacity to assist medical professionals. To ensure the dependability and consistency of the model, we will undertake multicenter clinical research to validate its efficacy and uphold ethical guidelines and privacy regulations. Furthermore, we are committed to regular model updates and enhancements to sustain precision and practicality.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

XZ and XZ conceived the work. JZ studied and drafted the manuscript. LS, AL, and XZ discussed and edited the manuscript. GL, XC, and YH assisted the analysis. XZ checked the statistical and bioinformatic accuracy as an expert in statistics and bioinformatics. All authors read and approved the final version of the manuscript.

Funding

This study was partially supported by the National Natural Science Foundation of China (32170915 and 82172931).

Data availability

All data generated or analyzed during this study are included in this published article and its supplementary materials.

Declarations

Conflict of interest

The authors declare that they have no competing interests.

Ethical approval and consent to participate

The work was approved by the Guangdong Medical University committee (YS2021159). Informed consent forms are not required for patient data extracted from public databases.

Consent for publication

Not applicable.