A Breg-associated lncRNA signature predicts prognosis and immune landscape in esophageal carcinoma Authors: Wen-Tao Xiao, Si-Qi Wu, Ya-Rui Han, Jia-Hui Song, Jun-Yan He Published: 2026 ABSTRACT AbstractBackgroundEsophageal carcinoma (ESCA) is a highly aggressive malignancy with poor prognosis, and growing evidence indicates that the tumor immune microenvironment, particularly regulatory B cells (Bregs), has been increasingly recognized as an essential contributor to tumor progression. However, the mechanisms linking Breg-associated long non-coding RNAs (lncRNAs) to ESCA prognosis have not been elucidated. This study aimed to develop a Breg-associated lncRNA prognostic signature for ESCA and to characterize its associations with the tumor immune microenvironment, mutational landscape, and therapeutic drug sensitivity.MethodsTranscriptome and clinical data for ESCA patients were retrieved from The Cancer Genome Atlas (TCGA). Consensus clustering based on Breg marker gene expression was conducted to identify immune-related molecular subtypes. Differentially expressed lncRNAs were identified, and prognostic candidates were screened based on the results of univariate Cox regression. A prognostic signature was subsequently established by integrating least absolute shrinkage and selection operator (LASSO) selection with multivariate Cox analysis, and was further assessed in both training and validation cohorts. Kaplan-Meier survival curves, time-dependent receiver operating characteristic (ROC) analysis, and Cox regression were applied to comprehensively assess the predictive capacity of the signature. To facilitate clinical application, a nomogram was generated by combining the risk score with key clinicopathological variables. Multiple bioinformatics approaches were employed to characterize tumor mutational burden (TMB), immune infiltration, and functional pathways. Drug response profiles were predicted via the OncoPredict algorithm based on pharmacogenomic data from the GDSC2 database.ResultsA four-lncRNA prognostic signature consisting of LINC00298, AC003077.1, GPC6-AS2, and COPDA1 was developed and effectively stratified ESCA patients into high- and low-risk groups. The model showed strong predictive performance, achieving area under the curve (AUC) values of 0.728, 0.755, and 0.751 for predicting 1-, 2-, and 3-year overall survival, respectively. The risk score remained prognostically relevant across clinical subgroups, and integration into a nomogram improved predictive accuracy. The signature was constructed from lncRNAs differentially expressed across Breg-defined molecular subtypes, and the resulting risk score reflects the Breg-associated transcriptional state of the tumor. Consistent with a Breg-mediated immunosuppressive phenotype, high-risk patients exhibited elevated TMB, increased mutational frequency in several driver genes, and significantly reduced infiltration of multiple immune cell subsets—including activated T cells, B cells, macrophages, natural killer (NK) cells, and regulatory T cells—within the tumor microenvironment, indicating that the Breg-associated lncRNA program contributes to immune evasion in ESCA. Drug response prediction also indicated that the high-risk group showed differential sensitivity to several candidate compounds, implying potential therapeutic relevance.ConclusionsIn this study, we developed a Breg-associated lncRNA signature that enables prognostic prediction and provides insights into the immune microenvironment of ESCA. Furthermore, the nomogram integrating the risk score and clinical factors could provide a quantitative reference for individualized survival assessment.Keywords: Esophageal carcinoma (ESCA), long non-coding RNA (lncRNA), regulatory B cells (Bregs), prognostic signature, immune infiltration Highlight box Key findings • A prognostic signature composed of four regulatory B cells (Bregs)-associated long non-coding RNAs (lncRNAs) (LINC00298, AC003077.1, GPC6-AS2, and COPDA1) was constructed for esophageal carcinoma (ESCA). • The signature effectively stratified patients into high- and low-risk groups and showed stable predictive performance for overall survival. • The high-risk group showed lower immune cell infiltration and reduced ESTIMATE, immune, and stromal scores. What is known and what is new? • Bregs are important regulators of the tumor immune microenvironment. • This study identified a Breg-associated lncRNA signature and revealed its association with immune infiltration patterns in ESCA. What is the implication, and what should change now? • The Breg-associated lncRNA signature may provide a potential biomarker for prognostic evaluation in ESCA. • Further experimental and clinical studies are needed to validate the biological functions of these lncRNAs and their potential clinical relevance. Introduction Esophageal carcinoma (ESCA) remains a major global public health challenge. Recent epidemiological evidence indicates that ESCA ranks among the ten most common malignancies worldwide and accounts for a substantial proportion of cancer-related mortality (1). Epidemiological surveys have consistently ranked ESCA among the major causes of cancer mortality. Although recent decades have witnessed significant strides in multimodal diagnosis and treatment, patient prognosis remains far from satisfactory (2). Registry data reveal that despite these therapeutic improvements, the overall 5-year survival rate is still poor, and long-term outcomes have not significantly improved (3). This dismal prognosis is further complicated by substantial epidemiological and histological heterogeneity: squamous cell carcinoma remains predominant in East Asia and other high-incidence regions, whereas adenocarcinoma has increased markedly in Western populations, reflecting distinct etiological patterns and clinical characteristics (4). Taken together, current evidence indicates that prognostic evaluation in ESCA remains challenging. Therefore, developing clinically relevant biomarkers is essential for improving patient stratification. In recent years, immune checkpoint blockade targeting the programmed cell death protein 1/programmed death-ligand 1 (PD-1/PD-L1) axis has substantially reshaped the landscape of ESCA immunotherapy. In the perioperative setting, adjuvant nivolumab significantly improved disease-free survival in patients with resected esophageal or gastroesophageal junction cancer following neoadjuvant chemoradiotherapy (CheckMate 577) (5). In advanced disease, phase III trials such as ATTRACTION-3 and CheckMate 648 have demonstrated survival benefits of PD-1 based regimens in esophageal squamous cell carcinoma (6,7). Moreover, pembrolizumab-based therapy has also shown clinical efficacy in biomarker-selected ESCA populations (KEYNOTE-181) (8). These findings underscore the pivotal role of immune regulation in determining therapeutic response and long-term outcomes in ESCA. With the rapid development of cancer immunology, immune infiltration and the tumor microenvironment (TME) have emerged as key determinants of tumor progression, metastasis, and therapeutic response (9). Increasing evidence indicates that the immune composition of the TME significantly influences therapeutic response in esophageal cancer (10). Among immune components, regulatory B cells (Bregs) have attracted growing attention as a specialized B-cell subset capable of suppressing antitumor immunity by dampening cytotoxic T-cell and natural killer (NK) cell activity (11,12). Although tumor-infiltrating B cells may exert antitumor functions in certain contexts, Bregs are increasingly implicated in promoting immune escape and tumor progression depending on their phenotype and surrounding microenvironment (13,14). Historically, cancer research has focused primarily on protein-coding genes; however, protein-coding sequences account for less than 2% of the human genome, and a substantial proportion of the transcriptome consists of non-coding RNAs (ncRNAs) (15). Among them, long non-coding RNAs (lncRNAs) have emerged as critical regulators of gene expression and are frequently dysregulated in multiple cancers, including esophageal cancer (16-18). Beyond their oncogenic roles, lncRNAs have also been implicated in immune regulation by modulating immune cell differentiation and cytokine signaling pathways (19,20). Nevertheless, the interaction between lncRNAs and immunosuppressive B-cell subsets, particularly Bregs, remains poorly understood in esophageal cancer. Given the immunosuppressive role of Bregs in facilitating tumor immune evasion, investigating Breg-related lncRNAs may uncover novel immunoregulatory mechanisms and provide potential biomarkers for prognostic assessment and therapeutic targeting in ESCA (21). Here, we explored Breg-associated lncRNAs and evaluated their potential prognostic relevance in ESCA using transcriptomic datasets combined with immune landscape analyses. We further evaluated its predictive performance, explored its association with immune infiltration patterns and functional pathways, and developed a nomogram for individualized survival prediction. Additionally, we assessed potential therapeutic implications through drug-sensitivity analyses. Together, this work proposes a Breg-related lncRNA signature with potential value for prognostic assessment and personalized treatment guidance in ESCA. We present this article in accordance with the TRIPOD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0409/rc). Methods RNA-seq data processing and sample filtering The bulk RNA-seq data and corresponding clinical profiles for ESCA were retrieved from The Cancer Genome Atlas (TCGA) repository (https://www.cancer.gov/ccg/research/genome-sequencing/tcga). Samples with incomplete clinical variables or inadequate follow-up information were removed, resulting in a final cohort of 160 eligible patients. Fragments per kilobase per million mapped reads (FPKM) values were converted to transcripts per million (TPM) and transformed as log2(TPM + 1). Prior to normalization and variance-stabilizing transformation, genes with a median raw count of 0 across samples were excluded. To ensure data comparability across samples, additional normalization and procedures were performed using the DESeq2 R package (https://bioconductor.org/packages/DESeq2/). Figure 1 illustrates the comprehensive workflow followed in this study. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The flowchart of the whole study. AUC, area under the curve; Breg, regulatory B cell; ESCA, esophageal carcinoma; GSEA, gene set enrichment analysis; LASSO, least absolute shrinkage and selection operator; lncRNA, long non-coding RNA; TCGA, The Cancer Genome Atlas; WGCNA, weighted gene co-expression network analysis. Breg-associated subtype stratification and lncRNA profiling We performed unsupervised consensus clustering on ESCA patients using the expression levels of 27 previously reported Breg-related genes identified by Zhou et al. to delineate Breg-associated molecular subtypes (22). The ConsensusClusterPlus package in R was applied with 1,000 resampling iterations to ensure clustering robustness, resulting in two stable subtypes (C1 and C2). To characterize molecular differences between Breg-related subtypes, we performed differential expression analysis of lncRNAs across the two clusters. Differential expression analysis was conducted using the Wilcoxon rank-sum test, with significance defined as |fold change| >1.5 and P<0.05. In total, 1,406 lncRNAs were identified as significantly differentially expressed and carried forward for downstream analyses. Univariate Cox regression analysis indicated that 33 prognostic lncRNAs, from a total of 1,406, were significantly linked to patient survival outcomes. These prognostic lncRNAs were then incorporated into feature-selection procedures for subsequent model development. Derivation and validation of the Breg-derived prognostic signature Using the 33 prognostic lncRNAs identified by univariate Cox regression, a Breg-related prognostic signature was developed using least absolute shrinkage and selection operator (LASSO) regression followed by multivariate Cox regression. The TCGA-ESCA cohort was randomly divided into training and validation cohorts at 1:1 ratio to ensure balanced sample sizes for model development and internal validation. LASSO regression was implemented with the “glmnet” R package (https://CRAN.R-project.org/package=glmnet) to reduce overfitting and select the most informative lncRNAs. The lncRNAs retained after LASSO regularization were then incorporated into a multivariate Cox model to obtain their corresponding regression coefficients. For each patient, a prognostic score was generated by integrating the expression levels of selected lncRNAs with their corresponding coefficients estimated from multivariate Cox regression, as described by the following equation: where “exp” denotes the expression level of each lncRNA and “coef” represents the coefficient derived from multivariate Cox regression. Risk stratification in the training cohort was based on the median risk score, resulting in high- and low-risk groups, and the same cutoff was applied to the validation cohort. Group survival distributions were analyzed using the Kaplan-Meier analysis and log-rank test. We assessed the model’s discriminatory capability through time-dependent receiver operating characteristic (ROC) curve analysis using the pROC R package (23), evaluating area under the curve (AUC) values at 1, 2, and 3 years for 1-, 2-, and 3-year survival in both the training and validation sets. Stratified and independent prognostic analyses Stratified overall survival (OS) analyses were performed to examine whether the prognostic signature remained effective across different clinicopathological subgroups. Tumor grade was classified into low-grade (G1–G2) and high-grade (G3). The lymph node ratio (LNR)—defined as the ratio of metastatic to examined lymph nodes—has previously been demonstrated as an independent prognostic factor in esophageal cancer (24,25). In this study, four LNR categories were applied: pN0 (LNR = 0); low (0< LNR ≤0.2); medium (0.2< LNR ≤0.5); and high (LNR >0.5). For downstream survival analyses, pN0 and low groups were combined as low-LNR, while medium and high were combined as high-LNR. Kaplan-Meier curves were generated using the survival (https://CRAN.R-project.org/package=survival) and survminer (https://CRAN.R-project.org/package=survminer) R packages, with differences assessed via the log-rank test. Univariate and multivariate Cox proportional hazards regression analyses were performed using the “survival” package to determine whether the risk score served as an independent prognostic indicator after adjusting for clinical variables. Genomic alterations and tumor mutational burden (TMB) analysis TMB was quantified and somatic mutation profiles were computed using the “maftools” R package (26). Somatic mutation landscapes, including oncoplots and mutation frequency summaries, were generated using maftools and visualized using “ggplot2” and “forestplot” (https://CRAN.R-project.org/package=forestplot). Nomogram construction The nomogram was developed using the “rms” package (https://CRAN.R-project.org/package=rms) within the R environment to visualize the integrated prognostic model. The nomogram integrated several prognostic variables, including gender, tumor status, pathologic stage, LNR grouping, and the constructed risk score. The accuracy of predicted survival was evaluated using calibration plots comparing predicted probabilities with real outcomes. The discriminative ability of the nomogram was quantified by the concordance index (C-index) and time-dependent AUC, computed using the riskRegression (available at https://CRAN.R-project.org/package=riskRegression) and timeROC (27) R packages. Weighted gene co-expression network analysis (WGCNA) WGCNA was performed using the “WGCNA” R package (28) to identify gene modules associated with the Breg-related lncRNA signature. To establish a scale-free network topology within the TCGA-ESCA dataset, a soft-thresholding power of 12 was selected. Subsequently, network construction and module detection were executed using the blockwiseModules function, with a minimum module size of 100 and a merge cut height of 0.25. Genes with highly similar expression patterns were clustered into distinct modules through dynamic tree cutting. Correlation analysis identified the module most strongly linked to the Breg-associated lncRNA signature, which was selected for functional enrichment. Functional enrichment analyses To understand the biological relevance of the selected module, we performed multiple enrichment analyses including Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG), and Hallmark pathway enrichment analyses. Functional annotations were derived using “clusterProfiler” and “org.Hs.eg.db”, while pathway visualization was generated using the “enrichplot” package. Significance for enrichment was defined as an adjusted P value <0.05. Gene set enrichment analysis (GSEA) GO-based (GSEA was additionally performed between immune subtypes (C1 vs. C2) using ranked differential mRNAs to characterize subtype-associated biological programs. To characterize pathway-level differences, we performed GSEA between high- and low-risk groups. Genes were ordered by their differential expression statistics prior to enrichment analysis and enrichment scores were conducted using the ‘clusterProfiler’ package in R (29). Pathways with adjusted P values below 0.05 were regarded as significantly enriched. Immune infiltration analysis The tumor immune microenvironment was evaluated using multiple computational approaches. ImmuneScore, StromalScore, and ESTIMATEScore were calculated using the “ESTIMATE” R package (30) to assess the proportions of immune and stromal components in each tumor sample. Immune infiltration levels for 28 immune cell subsets were quantified using single-sample GSEA (ssGSEA) implemented in the “GSVA” R package (31). In addition, immune cell infiltration was assessed using seven algorithms—TIMER, CIBERSORT, CIBERSORT-ABS, QUANTISEQ, EPIC, MCPcounter, and xCell—were integrated to provide a comprehensive evaluation. Correlation analyses between the risk score and immune cell infiltration levels were conducted using Spearman or Pearson methods as appropriate. Drug sensitivity Drug sensitivity was inferred using the oncoPredict package (32), which estimates chemotherapeutic susceptibility across patient samples based on large-scale pharmacogenomic reference datasets. The GDSC2 training expression matrix (GDSC2_Expr) and drug response data (GDSC2_Res, 198 compounds) were obtained from the CancerRxGene portal (https://www.cancerrxgene.org/). The calcPhenotype function was applied to the TCGA-ESCA expression matrix with the following parameters: batch correction using the empirical Bayes method (batchCorrect = “eb”), power transformation of drug response values enabled (powerTransformPhenotype = TRUE), removal of the lowest 20% of genes by variance computed from raw data (removeLowVaryingGenes = 0.2, removeLowVaryingGenesFrom = “rawData”), and a minimum sample threshold of 10 (minNumSamples = 10). The predicted output for each sample was a drug sensitivity score positively correlated with IC50, and scores were log2-transformed prior to visualization. Drugs with statistically significant differences between high- and low-risk groups were identified using the Wilcoxon rank-sum test (P<0.005). These analyses were based on cell-line pharmacogenomic data and did not incorporate patient-derived drug response data. Statistical analysis All statistical analyses were carried out using R software (version 4.2.0), including differential expression testing, survival modeling, and correlation assessments. To identify differentially expressed genes or lncRNAs across subgroups, Wilcoxon tests or “limma”-based linear modeling approaches were employed, with Benjamini-Hochberg FDR correction applied to adjust for multiple comparisons. The prognostic significance of clinical variables and model-derived risk scores was determined through Cox proportional hazards analyses, supplemented by Kaplan-Meier survival comparisons using the log-rank test. Associations between immune cell abundance and risk scores were quantified via Spearman or Pearson correlation analyses, chosen according to the distribution of the variables. Two-tailed testing was applied throughout the study, and statistical significance was determined using nominal P<0.05 or FDR-adjusted P<0.05. Results Overview of immune cell infiltration in ESCA Consensus clustering analysis leveraging Breg-associated marker genes identified two immune-related subtypes within the TCGA-ESCA cohort. Cumulative distribution function analysis was first performed to assess clustering stability. Delta area curves further suggested that two clusters provided the optimal solution (Figure 2A-2C). Accordingly, the TCGA-ESCA cohort was divided into Cluster 1 (C1, n=83) and Cluster 2 (C2, n=77) (Figure 2D). The enriched fractions of different immune cell sets obtained by ssGSEA between the two subtypes were significantly different. In Cluster 1, several immune cells exhibited higher enrichment scores for multiple immune cell subsets compared with Cluster 2. Consistently, ESTIMATE analysis showed that ESTIMATE score, Immune score, and Stromal score were all significantly higher in C1 compared with C2 (Figure 2E-2G), supporting an immune-infiltrated phenotype in C1. A heatmap integrating immune-cell signatures and clinical annotations further highlighted the molecular and clinical divergence between these two subtypes (Figure 2H). Identification and immune landscape of Breg-related molecular subtypes. Consensus matrix heatmap showing sample clustering stability based on Breg marker gene expression (A). Consensus CDF curves for k=2–7 (B). Delta area plot of the relative change in the area under the CDF curve, supporting k=2 as the optimal clustering solution (C). ssGSEA-based comparison of immune cell infiltration scores between Cluster 1 (C1, n=83) and Cluster 2 (C2, n=77) (D). Comparison of ESTIMATEScore (E), ImmuneScore (F), and StromalScore (G) between the two clusters, indicating significantly higher immune and stromal components in C1 (E-G). Heatmap illustrating immune cell enrichment patterns and corresponding clinical annotations between C1 and C2 (H). ns, P≥0.05; **, P<0.01; ***, P<0.001; ****, P<0.0001. Breg, regulatory B cell; C1, cluster 1; C2, cluster 2; CDF, cumulative distribution function; MDSC, myeloid-derived suppressor cell; ssGSEA, single-sample gene set enrichment analysis. Functional characterization of immune infiltration subtypes and identification of prognostic Breg-related lncRNAs To explore the functional implications of the two identified immune infiltration clusters, GSEA was performed using GO gene sets based on the ranked list of differentially expressed mRNAs between C1 and C2 clusters. The results revealed that Cluster C1 was significantly enriched in immune-related biological processes, which were mainly associated with adaptive immune functions, antigen receptor-mediated signaling cascades, B cell receptor activity, and leukocyte proliferation, while Cluster C2 was predominantly associated with epithelial differentiation, keratinization, and ribosome-related pathways (Figure 3A-3C). Subsequently, KEGG enrichment analysis of differential mRNAs between the two clusters revealed that immune-associated pathways, including IgA-related intestinal immune networks and chemokine signaling, were prominently represented (Figure 3D). Functional characterization and screening of Breg-associated lncRNAs. GSEA was performed using a ranked list of differentially expressed mRNAs between C1 and C2 (A). Representative GSEA enrichment plots showing immune-related biological processes enriched in C1 (B) and epithelial differentiation-, keratinization-, and ribosome-related pathways enriched in C2 (C). KEGG pathway enrichment analysis of differentially expressed mRNAs between C1 and C2 (D). Volcano plot of differentially expressed lncRNAs between C1 and C2 (E). The 33 prognostic lncRNAs identified by univariate Cox regression analysis represent a subset of the 1,406 Breg-related lncRNAs (F). Breg, regulatory B cell; GO, Gene Ontology; GSEA, gene set enrichment analysis; KEGG, Kyoto Encyclopedia of Genes and Genomes; lncRNA, long non-coding RNA. Next, we analyzed the differentially expressed lncRNAs between the two clusters by setting thresholds of |fold change| >1.5 and P<0.05. A total of 1,406 lncRNAs were identified as characteristic lncRNAs of distinct immune subtypes (Figure 3E). Univariate Cox regression analysis was then conducted for these differential lncRNAs, and 33 lncRNAs demonstrated statistically significant relationships with OS in the cohort (P<0.05) (Figure 3F). These prognostic lncRNAs were subsequently retained for further model construction (Table S1). Construction and validation of the Breg-related signature We retrieved transcriptomic and clinical data for 160 ESCA patients from the TCGA database for subsequent analysis. These patients were randomly partitioned into a training cohort (n=80) and a validation cohort (n=80) to construct and internally validate the prognostic model. To identify key prognostic markers among the 33 Breg-associated lncRNAs, we performed univariate Cox proportional hazards regression followed by LASSO regression. Using cross-validation and analysis of penalized regression coefficients, eight lncRNAs were shortlisted as potential predictors for model development (Figure 4A,4B). Subsequent multivariate Cox regression refinement yielded a final prognostic signature consisting of four lncRNAs: LINC00298, AC003077.1, GPC6-AS2, and COPDA1 (Figure 4C,4D). Based on these four lncRNAs, a risk scoring system was established. The risk score for each patient was computed as the sum of the expression levels of each lncRNA multiplied by its corresponding regression coefficient derived from the multivariate Cox model. The median risk score was adopted as the threshold to categorize patients into high- and low-risk groups. The overall distribution of risk scores, survival outcomes, and group assignments is depicted in Figure 4E, providing an overview of stratification. A heatmap of the four prognostic lncRNAs is shown in Figure 4F, highlighting their differential expression across risk groups. Survival analysis based on Kaplan-Meier curves showed a clear survival disadvantage in the high-risk group (Figure 4G), underscoring the prognostic significance of the Breg-related lncRNA signature. Time-dependent ROC analysis demonstrated good discriminative performance, reaching AUCs of 0.744, 0.840, and 0.871 for 1-, 2-, and 3-year OS, respectively (Figure 4H). The genomic locations of the four prognostic lncRNAs were annotated and visualized in Figure 4I. To further validate the biological relevance of these four lncRNAs within the Breg-associated immune context, their expression levels were compared between the C1 and C2 subgroups defined by Breg marker gene clustering. All four lncRNAs were significantly upregulated in the immunosuppressive C2 subgroup relative to the immune-infiltrated C1 subgroup (all P<0.05; Figure S1), confirming their preferential activity within the Breg-shaped tumor immune microenvironment. Development of a prognostic signature based on Breg-associated lncRNAs. LASSO regression analysis for screening prognostic Breg-associated lncRNAs (A,B). Four prognostic lncRNAs (LINC00298, AC003077.1, GPC6-AS2, and COPDA1) were identified by multivariate Cox regression analysis (C). Risk coefficients of the four lncRNAs in the prognostic signature (D). Distribution of risk scores (top) and survival status (bottom) of ESCA patients (E). Heatmap showing expression patterns of the four lncRNAs across the high- and low-risk groups (F). Kaplan-Meier survival analysis showing significantly worse overall survival in the high-risk group compared with the low-risk group (G). ROC curves evaluating the predictive performance of the prognostic signature for 1-, 2-, and 3-year overall survival (H). Chromosomal locations of the four lncRNAs included in the prognostic signature (I). *, P<0.05; ***, P<0.001. AUC, area under the ROC curve; Breg, regulatory B cell; ESCA, esophageal carcinoma; LASSO, least absolute shrinkage and selection operator; lncRNA, long non-coding RNA; OS, overall survival; ROC, receiver operating characteristic. To evaluate the predictive power and robustness of the Breg-related lncRNA signature, survival analysis was performed using two internal datasets: the TCGA validation cohort and the TCGA entire cohort. In the validation cohort, the distribution of patient risk scores and corresponding survival status is displayed in Figure 5A, and the expression profiles of the four lncRNAs are shown in Figure 5B. Kaplan-Meier analysis demonstrated that patients in the high-risk group had significantly poorer OS compared to those in the low-risk group (Figure 5C). Further evaluation using time-dependent ROC curves showed AUC values of 0.734, 0.692, and 0.731 at 1, 2, and 3 years, respectively, supporting the model’s predictive performance (Figure 5D). Prognostic performance and internal validation of the signature. Distribution of risk scores (top) and survival status (bottom) in the validation cohort (A) and the entire cohort (E). Heatmaps showing the expression patterns of the four signature lncRNAs across high- and low-risk groups in the validation cohort (B) and the entire cohort (F). Kaplan-Meier survival curves comparing overall survival between high- and low-risk groups in the validation cohort (C) and the entire cohort (G). ROC curves evaluating the predictive performance of the prognostic signature for 1-, 2-, and 3-year overall survival in the validation cohort (D) and the entire cohort (H). AUC, area under the ROC curve; Breg, regulatory B cell; lncRNA, long non-coding RNA; OS, overall survival; ROC, receiver operating characteristic. These results were further confirmed in the full TCGA cohort. The distribution of risk scores together with survival outcomes is shown in Figure 5E, and the expression profiles of the signature lncRNAs are visualized in Figure 5F. Survival curves in Figure 5G again exhibited clearly divergent survival outcomes between the two risk groups, reinforcing the prognostic relevance of the signature. The ROC curves in Figure 5H demonstrated AUCs of 0.728, 0.755, and 0.751 at 1, 2, and 3 years, indicating high predictive accuracy in the overall cohort. These results support the stability and accuracy of the Breg-associated lncRNA signature in predicting OS in ESCA patients. Prognostic stratification of the Breg-related signature across clinical subgroups The stratified analysis further revealed that across different subgroups (age, sex, TNM stage, grade, and LNR), the risk score effectively distinguished survival outcomes in most subgroups, as shown in Figures 6,7. Patients with high-risk scores consistently exhibited poorer OS. Notably, the signature maintained prognostically significant in both low- and high-grade populations, as well as among patients with varying pathological stages. Even within the low-LNR subgroup, where lymph node involvement was minimal, the risk signature effectively separated patients into distinct prognostic groups (Figure 7E). Overall, the prognostic value of this signature was consistently stable across diverse clinical subgroups, supporting its robustness. Stratified survival analysis of the Breg-associated lncRNA signature across different clinical subgroups. Kaplan-Meier survival curves comparing overall survival between high- and low-risk groups stratified by age ≤60 years and age >60 years (A,B), female and male (C,D), tumor status (tumor-free and with tumor) (E,F), clinical T stage (T1T2 and T3T4) (G,H), clinical N stage (N0 and N+) (I,J), clinical M stage (M0 and M1) (K,L), clinical stage (stage I+II and stage III+IV) (M,N), pathologic T stage (T1–T2 and T3–T4) (O,P), pathologic N stage (N0 and N+) (Q,R), and pathologic M stage (M0 and M1) (S,T). P values were calculated using the log-rank test. Breg, regulatory B cell; lncRNA, long non-coding RNA; M, metastasis; N, node; OS, overall survival; T, tumor. Stratified survival analysis by pathologic stage, grade, and lymph node ratio. Kaplan-Meier survival curves comparing overall survival between high- and low-risk groups stratified by pathologic stage I+II and stage III+IV (A,B), tumor grade (low-grade and high-grade) (C,D), and LNR group (low-LNR and high-LNR) (E,F). P values were calculated using the log-rank test. LNR, lymph node ratio; OS, overall survival. Comparative genomic alterations and TMB analysis As shown in Figure 8A,8B, mutation prevalence was substantially higher in the high-risk population compared to the low-risk population (98.73% vs. 28.75%). Given this discrepancy, we further examined whether TMB also varied between the two risk groups. Differential mutation analysis identified several genes with remarkably different mutation frequencies between the two groups (Figure 8C). Mutations in DNAH11, TTN, LRP1B, HMCN1, CSMD1, SYNE1, and VCL were predominantly enriched in the high-risk cohort, whereas FLG mutations were more frequently observed in the low-risk group. We next compared TMB levels between the two groups. In line with the mutation landscape, the high-risk group showed significantly higher TMB values than the low-risk group (Figure 8D). Survival analysis revealed that patients with high TMB had poorer OS compared with those with low TMB (Figure 8E). Moreover, integrating both TMB status and the risk score generated four distinct subgroups with markedly different survival outcomes (Figure 8F), indicating that combining TMB with the Breg-related signature may improve prognostic stratification. Genomic alterations and TMB analysis in high- and low-risk ESCA patients. Mutation landscape of the low-risk (A) and high-risk (B) groups, showing the frequency and distribution of somatic mutations in ESCA patients. Comparison of significantly mutated genes between the low-risk and high-risk groups (C). Violin plot comparing TMB levels between the two risk groups (D). Kaplan-Meier survival analysis of ESCA patients stratified by TMB status (E). Kaplan-Meier survival analysis of ESCA patients stratified by both TMB status and risk group (F). *, P<0.05; **, P<0.01. CI, confidence interval; ESCA, esophageal carcinoma; H-Risk, high-risk; H-TMB, high tumor mutational burden; L-Risk, low-risk; L-TMB, low tumor mutational burden; OR, odds ratio; OS, overall survival; TMB, tumor mutational burden. Nomogram construction To evaluate whether the risk score provides prognostic information beyond conventional clinical factors, we conducted univariate and multivariate Cox regression analyses. As shown in Figure 9A,9B, multiple variables were associated with OS, and tumor status and pathologic stage emerged as independent prognostic factors in ESCA. A nomogram integrating gender, tumor status, pathologic stage, LNR group, and the risk score was subsequently established to improve clinical applicability (Figure 9C). This nomogram serves as a clinically practical tool for estimating personalized 1-, 2-, and 3-year survival risks in ESCA patients. Calibration plots showed good concordance between predicted and observed survival across all examined time points (Figure 9D). The time-dependent C-index further confirmed that the nomogram offered improved discrimination over individual clinical features (Figure 9E). Kaplan-Meier curves depicted inferior survival in the high-score group (Figure 9F). As shown in Figure 9G, the AUC values for 1-, 2-, and 3-year OS prediction were 0.638, 0.776, and 0.823, indicating satisfactory predictive accuracy. Collectively, these results confirm the nomogram’s effectiveness in providing individualized prognostic evaluation for ESCA patients. Nomogram construction and evaluation. Univariate Cox regression analysis of clinical factors and risk score (A). Multivariate Cox regression analysis identifying independent prognostic factors (B). Nomogram integrating significant clinical variables and risk score for predicting 1-, 2-, and 3-year overall survival (C). Calibration curves evaluating nomogram predictive accuracy (D). Time-dependent C-index comparison between the nomogram and individual prognostic factors (E). Kaplan-Meier survival analysis based on nomogram-derived stratification (F). Time-dependent ROC curves assessing nomogram performance at 1-, 2-, and 3-year OS (G). AUC, area under the ROC curve; C-index, concordance index; LNR, lymph node ratio; OS, overall survival; ROC, receiver operating characteristic. WGCNA and GSEA We constructed a WGCNA network to identify co-expression modules associated with the risk groups. Of the multiple gene modules revealed by hierarchical clustering, the brown module exhibited the strongest association with the high-risk group (Figure 10A,10B). Module eigengene-trait analysis further demonstrated a significant association between module membership and gene significance within this module (cor =−0.67, P<0.001), indicating robust internal consistency (Figure 10C). Functional enrichment analysis revealed that genes in the brown module were mainly enriched in muscle- and structure-related pathways, including muscle contraction, regulation of membrane potential, synaptic membrane components, and sarcolemma (Figure 10D,10E). KEGG analysis further highlighted calcium signaling, dilated cardiomyopathy, vascular smooth muscle contraction, neuroactive ligand-receptor interaction, and focal adhesion pathways (Figure 10F). Although these pathways are not classically immune-related, many of them, such as calcium signaling, extracellular matrix (ECM) organization, and focal adhesion, play essential roles in immune cell activation, adhesion, and tumor-immune interactions. We next carried out GSEA to characterize functional discrepancies between the high- and low-risk groups. GSEA revealed that the signature was linked to pathways related to ECM remodeling, synaptic structure, integrin-mediated interactions, and mitochondrial metabolic processes such as oxidative phosphorylation (Figure 10G,10H). As the signature was generated from Breg-associated lncRNAs, the observed structural and metabolic disturbances may represent downstream manifestations of immune dysregulation, providing a rationale for subsequent immune profiling. WGCNA and functional enrichment analyses of the Breg-related lncRNA signature. Cluster dendrogram of genes based on WGCNA, with different colors representing distinct gene modules (A). Module-trait relationship heatmap showing correlations between module eigengenes and risk groups, highlighting the brown module as the most relevant module (B). Correlation between module membership and gene significance within the brown module (C). GSEA dot plot of enriched GO terms in the brown module (D). GO enrichment bar plot showing the top enriched biological processes and cellular components in the brown module (E). KEGG pathway enrichment analysis of brown module genes (F). GSEA plots of representative GO pathways enriched in the activated group (G) and suppressed group (H). Breg, regulatory B cell; GO, Gene Ontology; GSEA, gene set enrichment analysis; KEGG, Kyoto Encyclopedia of Genes and Genomes; lncRNA, long non-coding RNA; ME, module eigengene; WGCNA, weighted gene co-expression network analysis. Immune infiltration analysis To investigate immune infiltration and the tumor immune microenvironment in ESCA, we first compared the ESTIMATE score, Immune score, and Stromal score between the two risk groups. As presented in Figure 11A, all three scores were significantly lower in the high-risk group, indicating reduced immune and stromal components within the TME. We next applied ssGSEA to evaluate the relationship between the risk score and tumor-infiltrating immune cells. The heatmap revealed overall lower immune infiltration levels in the high-risk group across multiple immune cell types (Figure 11B). IConsistent with this trend, violin-plot comparisons of individual immune cell subsets showed that the ssGSEA enrichment scores of multiple immune populations differed significantly between the high- and low-risk groups (Figure 11C). We next employed a panel of multi-algorithm immune cell deconvolution approaches—TIMER, CIBERSORT, CIBERSORT-ABS, QUANTISEQ, MCPCOUNTER, XCELL, and EPIC—to comprehensively compare immune infiltration patterns between high- and low-risk groups. Results from all algorithms uniformly indicated reduced immune cell infiltration in the high-risk subgroup (Figure 12), reflecting a broadly immunosuppressed TME. To further characterize specific immunosuppression-relevant subsets, pairwise comparisons were conducted for CD8+ T cells, Tregs, M2 macrophages, and Bregs across algorithms (Figure S2). Statistically significant reductions were observed for CD8+ T cells (CIBERSORT-ABS and MCPCOUNTER), Tregs (CIBERSORT), and M2 macrophages (CIBERSORT-ABS and QUANTISEQ), while Breg activity assessed by ssGSEA was also significantly lower in high-risk patients (P<0.001). Taken together, these findings consistently support the presence of an immune-depleted microenvironment in high-risk ESCA patients, characterized by deficiency of both antitumor effector and immunosuppressive populations. Immune infiltration analysis between high- and low-risk groups. Comparison of ESTIMATEScore, ImmuneScore, and StromalScore between high- and low-risk groups based on ESTIMATE analysis (A). Heatmap illustrating ssGSEA-derived immune cell infiltration profiles across high- and low-risk patients (B). Violin plots comparing the enrichment scores of immune cell subsets between the two risk groups (C). ns, P≥0.05; *, P<0.05; **, P<0.01; ***, P<0.001; ****, P<0.0001. MDSC, myeloid-derived suppressor cell; ssGSEA, single-sample gene set enrichment analysis. Multi-algorithm validation of immune cell infiltration differences between risk groups. Correlation analysis of immune cell infiltration estimates and risk score based on multiple immune deconvolution algorithms, including xCell, TIMER, quanTIseq, MCP-counter, EPIC, CIBERSORT, and CIBERSORT-ABS (A). Heatmap showing immune infiltration profiles across high- and low-risk groups inferred by different immune deconvolution methods (B). EPIC, Estimating the Proportions of Immune and Cancer cells; MCP, microenvironment cell populations; TIMER, Tumor Immune Estimation Resource. Breg-related lncRNAs correlate with therapeutic drugs To further investigate the potential therapeutic implications of the risk signature, we employed the OncoPredict algorithm combined with drug response data from the GDSC2 database to estimate drug sensitivity. As shown in Figure 13, the predicted IC50 values of ulixertinib, VX-11E, trametinib, and SCH772984 were significantly lower in the high-risk group (Wilcoxon test, P<0.05), indicating enhanced sensitivity of high-risk ESCA patients to these agents. The observed drug sensitivity pattern implies a potential therapeutic relevance of MAPK/ERK-targeted agents in the high-risk subgroup, although further experimental validation is required. Overall, the results indicate that the proposed risk signature may aid in therapeutic stratification and the selection of potential drug candidates in ESCA. Drug sensitivity analysis between high- and low-risk groups. Predicted drug sensitivity scores were compared between low- and high-risk ESCA patients for ulixertinib (A), VX-11E (B), trametinib (C), and SCH772984 (D). ESCA, esophageal carcinoma. Discussion ESCA remains a lethal malignancy with persistently poor long-term outcomes and a substantial global burden. Despite multimodal treatment strategies—ranging from endoscopic therapy for early disease to trimodality approaches and systemic therapy for advanced ESCA—OS is still unsatisfactory (33). These observations suggest that conventional clinicopathological staging may be insufficient to capture the biological heterogeneity underlying disease progression and therapeutic vulnerability, highlighting the need for molecular and immune-informed prognostic frameworks (34,35). In addition, although immune checkpoint inhibitors have reshaped the therapeutic landscape of ESCA, substantial variability in clinical response persists, underscoring the need for more precise biomarkers to guide patient selection and optimize immunotherapeutic benefit (36-38). These clinical advances underscore the importance of dissecting immune heterogeneity to improve patient stratification and optimize therapeutic strategies. Immune-related stratification has increasingly been explored to refine prognostic assessment in ESCA; however, most existing models emphasize T cell-dominant activity or rely on composite immune infiltration scores. Emerging evidence indicates that tumor-infiltrating B cells exert context-dependent functions—supporting antitumor immunity through antigen presentation and tertiary lymphoid structure formation, while also contributing to immunosuppressive niches depending on subset composition and spatial organization within the TME (39). Bregs, in particular, have been identified as key immunomodulatory subsets capable of suppressing antitumor T-cell responses via cytokine-mediated and contact-dependent mechanisms, thereby promoting immune escape and tumor progression (40,41). In parallel, lncRNAs have emerged as important regulators of immune differentiation and checkpoint-related signaling, offering potential immune-informed biomarkers (42). Nevertheless, Breg-oriented lncRNA frameworks integrating immune phenotype with prognostic stratification remain limited in ESCA, warranting systematic investigation. Despite accumulating evidence implicating Bregs and lncRNAs in shaping an immunosuppressive TME, few studies have systematically integrated Breg-associated transcriptional features with lncRNA-based modeling to achieve robust prognostic stratification and therapeutic insight in ESCA. To address this gap, we leveraged Breg marker gene expression to delineate two stable Breg-associated molecular subtypes (C1 and C2) within the TCGA-ESCA cohort, which exhibited markedly distinct immune landscapes as confirmed by complementary immune deconvolution approaches. Based on these subtypes, subtype-associated lncRNAs were systematically screened and refined to establish a prognostic model. Subsequent multivariate Cox regression yielded a four-lncRNA signature comprising LINC00298, AC003077.1, GPC6-AS2, and COPDA1. Importantly, the signature showed consistent prognostic stratification across the training, testing, and overall cohorts, prompting us to further evaluate its association with clinicopathological characteristics and its potential utility for individualized risk assessment in ESCA. To facilitate individualized risk assessment, a nomogram integrating gender, tumor status, pathological stage, LNR group, and the lncRNA-based risk score was constructed. Importantly, despite the growing number of prognostic models proposed for esophageal cancer based on transcriptomic features, immune-related gene sets, or pathway-oriented signatures, many existing models are primarily constructed from differential expression patterns and emphasize global immune activity or tumor-intrinsic processes (43-45). While these approaches have improved prognostic stratification to some extent, they often do not explicitly incorporate immunosuppressive immune subsets or stratification strategies that explicitly account for immune contexture (46-48). In contrast, the present model was developed by anchoring lncRNA selection to Breg marker-defined molecular subtypes, thereby integrating an immunologically relevant stratification step into the modeling framework. By linking Breg-associated transcriptional programs with lncRNA-based risk modeling, our approach connects prognostic stratification with immune contexture rather than relying solely on expression-derived correlations, providing a complementary perspective for risk assessment in ESCA. Our data also revealed an apparent dissociation between TMB and immune cell infiltration: high-risk patients exhibited elevated TMB yet markedly reduced infiltration of CD8+ T cells, Tregs, M2 macrophages, and Breg activity. We propose that this reflects Breg-mediated immune suppression overriding neoantigen-driven immune activation—a decoupling consistent with an immune-excluded phenotype despite high mutational burden. Notably, while high TMB alone was associated with improved survival (P=0.028), this benefit was entirely abrogated when combined with high-risk score (P<0.001), suggesting that the Breg-associated lncRNA program captures an immunosuppressive state that functionally neutralizes the prognostic advantage of TMB. These findings highlight the potential value of integrating both biomarkers for patient stratification, particularly in the context of immunotherapy selection. To gain insight into the biological basis of the Breg-associated lncRNA signature, WGCNA was performed. Genes within the identified risk module showed prominent enrichment in pathways related to mitochondrial translation, oxidative phosphorylation, and ECM organization. These findings suggest a tumor phenotype characterized by metabolic hyperactivity and extensive structural remodeling. Previous studies have consistently linked metabolic reprogramming—particularly mitochondrial respiration—to aggressive tumor behaviors and adverse clinical outcomes across various solid malignancies (49,50). In parallel, ECM remodeling is a well-established driver of tumor progression and therapeutic resistance, acting as a physical barrier to treatment (51,52). Importantly, GSEA results showed a high degree of concordance with the WGCNA findings, confirming that upregulation of mitochondrial metabolism and ECM-associated pathways are stable and robust biological features of the high-risk group. Biologically, this enhanced metabolic capacity supports tumor cell survival under stress, while ECM reorganization facilitates invasion and stromal interactions (50,52). To further explore the functional relevance of the four prognostic lncRNAs, GSEA was performed by stratifying samples into high- and low-expression groups for each lncRNA and ranking genes by differential expression statistics. LINC00298, COPDA1, and AC003077.1 were enriched in immune regulatory pathways including lymphocyte differentiation, T cell activation, and cytokine-cytokine receptor interaction, while GPC6-AS2 was associated with ECM organization and integrin signaling, suggesting distinct but complementary roles in shaping the immune microenvironment (Figure S3). Beyond driving intrinsic tumor growth, these biological alterations likely reshape the immunological landscape of the TME. Compared with the high-risk group, the low-risk group showed increased ImmuneScore and ESTIMATEScore values and a wider spectrum of infiltrating immune cells, while the high-risk group presented an immune-suppressed phenotype. While direct evidence in ESCA is evolving, mechanistic studies in other solid tumors suggest that excessive ECM deposition and matrix stiffness can physically restrict immune cell infiltration, leading to an “immune-excluded” phenotype (53). Concurrently, hyperactive tumor metabolism can impose nutrient competition and metabolic stress on infiltrating immune cells, thereby dampening their effector functions (54). The robustness of this immune-depleted status in high-risk tumors was validated across multiple deconvolution algorithms (including TIMER, CIBERSORT, and others). Consequently, the Breg-associated lncRNA signature appears to identify a specific immunosuppressive niche where metabolic aggression and stromal barriers coexist with reduced immune infiltration. Given the established role of Bregs in mediating immune tolerance (55), this signature may reflect a broader regulatory network that coordinates these immunosuppressive processes. Additionally, drug sensitivity profiling further implied that the high-risk group might be more susceptible to pharmacological inhibition of the MAPK/ERK signaling pathway. These predicted drugs include trametinib (MEK inhibitor), ulixertinib and SCH772984 (ERK inhibitors), and VX-11E (ERK2 inhibitor). The MAPK/ERK signaling pathway has been widely implicated in ESCA progression, with established roles in enhancing cellular proliferation and metastatic potential (56). Also, evidence suggests that MAPK/ERK signaling can help tumors escape the immune system by affecting the TME (57). Trametinib is a well-known MEK inhibitor that works by blocking downstream ERK activation (58). Similarly, ulixertinib and SCH772984 were developed to target ERK directly and are often used to overcome resistance to other treatments (59,60). Although the present study is primarily based on bioinformatics analyses and lacks experimental validation, the observed sensitivity to MAPK/ERK-targeting agents suggests that the proposed signature may be functionally linked to MAPK/ERK signaling. This implies that the Breg-associated lncRNA signature could be useful for finding potential treatment targets in ESCA. Although this study provides a robust, biologically informed prognostic model for ESCA, the results should be interpreted in light of several limitations. First, the prognostic signature was developed and internally validated mainly based on the TCGA-ESCA cohort, and the relatively limited sample size and retrospective nature of the dataset may restrict the generalizability of our findings. Additional independent cohorts are needed to further confirm the stability and clinical applicability of this Breg-associated lncRNA model. Second, the identification of Breg-related subtypes and immune infiltration patterns was inferred from bulk transcriptomic data and computational deconvolution algorithms, which cannot fully reflect the actual abundance or functional status of immune cell subsets within the TME. Third, although four lncRNAs were included in the final signature, their biological functions and mechanistic roles in ESCA progression and immune regulation remain unclear and require further experimental investigation. Finally, drug sensitivity predictions were performed using the OncoPredict algorithm based on pharmacogenomic data from the GDSC2 database, which may not fully recapitulate the complex interplay between the tumor and the immune microenvironment in patient tissues. Despite these limitations, this study highlights the potential relevance of Breg-associated lncRNAs in ESCA prognosis and immune microenvironment heterogeneity, warranting further experimental and clinical validation. Conclusions In conclusion, this study systematically characterized Breg-associated immune heterogeneity in ESCA by stratifying patients into two molecular subtypes (C1 and C2) based on Breg marker gene expression. Building upon subtype-related transcriptional differences, we developed a novel four-lncRNA prognostic signature (LINC00298, AC003077.1, GPC6-AS2, and COPDA1), which effectively stratified patients into distinct risk groups with significantly different survival outcomes. Furthermore, a nomogram integrating the risk score with key clinicopathological variables was constructed to facilitate individualized survival estimation. Functional analyses based on WGCNA and GSEA suggested that the high-risk phenotype was associated with metabolic reprogramming and ECM remodeling, whereas the low-risk group exhibited a more immune-infiltrated TME. In addition, exploratory drug sensitivity prediction indicated potential differences in response to MAPK/ERK pathway inhibitors between risk groups. Unlike existing ESCA prognostic models that are primarily constructed from mRNA-based immune-related gene sets or pathway-oriented signatures without explicitly incorporating immunosuppressive immune subsets, our signature is uniquely anchored in Breg-associated transcriptional heterogeneity and leverages lncRNAs as prognostic mediators—a regulatory layer that remains underexplored in ESCA prognosis. Collectively, these findings provide an integrative framework linking Breg-associated transcriptional programs with lncRNA-based prognostic stratification in ESCA, offering a basis for further validation and mechanistic investigation. Supplementary The article’s supplementary files as Acknowledgments We acknowledge the TCGA database for providing the platform and its contributors for uploading meaningful datasets. Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. Footnotes Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0409/rc Funding: This work was supported by National Natural Science Foundation of China (No. 82304083 to J.Y.H.) and Natural Science Foundation of Hunan Province (No. 2023JJ40584 to J.Y.H.). Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0409/coif). The authors have no conflicts of interest to declare. References Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0409/prf Supplementary Materials The article’s supplementary files as