Single-cell transcriptomics unveils pyroptosis-related immune microenvironment dynamics and prognostic modeling in esophageal squamous cell carcinoma Authors: Hui Zou, Qinglin Ren, Jun Wu, Jiansheng He Published: 2026 ABSTRACT AbstractBackgroundEsophageal cancer remains a leading cause of global cancer mortality with limited therapeutic efficacy. This study aims to characterize pyroptosis-related immune mechanisms in esophageal cancer and develop a clinical prediction model to improve prognostic evaluation.MethodsSingle-cell RNA sequencing data from seven esophageal cancer tissues and one normal control were integrated to construct a cellular atlas. Immune subpopulations were isolated, and pyroptosis-associated differentially expressed genes were identified. We intersected these genes with pyroptosis-related genes(PRG), identifying 13 key pyroptosis genes. Computational analyses included pyroptosis scoring, subcluster re-analysis, pathway enrichment, and intercellular communication mapping. Pseudotime trajectory analysis was applied to high-pyroptosis immune subsets. A clinical prognostic model incorporating PRG was validated using bulk sequencing data, followed by immune infiltration quantification and chemotherapy response assessment.ResultsBy integrating single-cell and bulk transcriptomic datasets, we found that pyroptosis-related genes were enriched in immune-inflammatory pathways in ESCC immune cells. B cells showed extensive interactions within the ESCC immune microenvironment. The identified pyroptosis-related genes appeared to be involved in immune-cell state transitions and intercellular communication. A two-gene pyroptosis-related prognostic model based on GSDMB and CYCS showed moderate predictive performance in the TCGA training cohort. However, its performance in the external GSE53625 validation cohort was limited, indicating that the model should be interpreted as an exploratory prognostic signature rather than a clinically established prediction tool. A nomogram integrating the risk score with pathological stage was further constructed to explore its potential clinical applicability.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s13019-026-04474-2.Keywords: Pyroptosis-related genes, Esophageal squamous cell carcinoma, Single-cell transcriptomics, Prognosis Introduction Esophageal cancer (EC) ranks as the 11th most common cancer globally and is the seventh leading cause of cancer-related deaths [1]. The two primary histological subtypes are esophageal squamous cell carcinoma (ESCC) and esophageal adenocarcinoma (EAC) [2]. The most common histological subtype of EC in China is ESCC, and China accounts for 53% of global esophageal cancer cases [3]. Despite advances in diagnosis and treatment, most ESCC patients are diagnosed at an advanced stage, resulting in a poor 5-year overall survival rate of less than 30% [4]. Identifying novel prognostic biomarkers or developing an integrated prediction model is crucial for improving clinical outcomes in ESCC, as current tools for routine practice remain inadequate [5–7]. Pyroptosis, an inflammasome-mediated form of programmed cell death, involves rapid cell expansion leading to membrane rupture and release of cellular contents, which triggers a strong inflammatory response [8, 9]. Emerging evidence highlights the critical role of pyroptosis in cancer prognosis, influencing tumor proliferation, invasion, and metastasis. ESCC is increasingly recognized for its complex tumor microenvironment (TME), which contains diverse, highly specialized immune cells with dynamic transcriptional profiles [10]. These immune cells, with heterogeneous phenotypes and functions, play key roles in tumor development and are associated with patient prognosis. The Pyroptosis-Related Genes (PRGs) signature has recently been identified as a prognostic marker in ovarian [11], gastric [12], and lung adenocarcinoma [13]. A newly constructed model based on PRGs was found to be associated with the prognosis and immune microenvironment of ESCC, providing new insights for constructing prognostic models [14]. Another pioneering prognostic signature for ESCC, developed using four carefully selected PRGs, accurately predicts clinical outcomes and aids in the tailored selection of therapeutic strategies [15]. Single-cell transcriptome sequencing (scRNA-Seq) provides a detailed view of gene expression at the single-cell level, allowing for high-resolution analysis of cellular heterogeneity, development, and differentiation across various systems, offer a powerful means to explore the molecular landscape of individual cells [16]. This technology is particularly valuable for identifying distinct cell subtypes and revealing the complexity of tumor subpopulations, revealing new therapeutic targets for future therapies [17]. This study constructs a clinical prediction model for ESCC based on PRGs, which is associated with prognosis and the immune microenvironment in ESCC. The model may provide guidance for clinical treatment and offers a new perspective on predicting ESCC prognosis. Materials and methods Data download ESCC scRNA-seq data GSE188900 was downloaded from the GEO database [18]. This dataset comprises 8 samples, including 7 tumor samples from ESCC patients and 1 normal control. Importantly, all patients were treatment-naïve, meaning they had not received any prior chemotherapy, radiotherapy, or immunotherapy before sample collection [18]. Clinical and gene expression data for 130 ESCC patients and 34 normal controls were sourced from TCGA database. Additionally, bulk RNA sequencing data for ESCC (GSE53625), comprising 179 ESCC patients and 179 normal controls, were downloaded from the GEO database [19]. Quality control and clustering dimension reduction The Seurat R package [20] (version 4.2.0) was utilized to merge and normalize the scRNA-seq data. Cells exhibiting over 20% mitochondrial gene expression were filtered out. Additionally, cells with fewer than 200 counts were excluded to eliminate low-quality cells and empty droplets, while those with over 2,000 genes—indicative of doublet cells—were also removed for subsequent analyses. We normalized the sequencing depth using the “Normalize Data” function, applying the “Log Normalize” method. Variable features were identified using the ‘vest’ method within the ‘FindVariableFeatures’ function, yielding 4,000 features. Data scaling was performed with “ScaleData” to mitigate the effects of sequencing depth. Principal Component Analysis (PCA) was employed to identify significant principal components, from which we selected 20 PCs for t-distributed stochastic neighbor embedding (t-SNE) for dimensionality reduction. The “FindNeighbors” function was executed with default parameters, using the 20 PCs to construct a K-nearest neighbor graph based on Euclidean distance in PCA space. Subsequently, the “FindClusters” function clustered the cells into 26 distinct groups. Dimensionality reduction was finalized using the RunTSNE function. Cell clusters were annotated via the singleR package (version 3.8.1) [21], resulting in the identification of seven cell types: T cells, B cells, epithelial cells, monocytes, iPS cells, smooth muscle cells, and endothelial cells. Differentially expressed genes were identified using the FindAllMarkers function (adjusted P-value < 0.05), and the most significant differentially expressed genes were visualized in a heatmap generated by the DoHeatmap function. Identification of differentially expressed genes of PRGs A total of 66 candidate PRGs were initially collated, including 6 genes from the GeneCards database (https://www.genecards.org/) with a relevance score > 7 for the keyword ‘pyroptosis’, 27 genes from the Molecular Signature Database (MsigDB, version 7.4) for Reactome pyroptosis, and 33 genes from previous literature reports. After removing 15 duplicate genes found across these datasets, a final set of 51 unique PRGs was included in this study. We intersected the differentially expressed genes from TCGA-ESCA data with the marker genes of immune cell subpopulations exhibiting high AUCell scores in the single-cell dataset. AUCell analysis We used the AUCell package (version 1.5.5) in R to analyze the enrichment of pyroptosis-related genes at the single-cell level. We classified cells into high and low Area Under the Curve (AUC) score groups based on an activation threshold (AUC Score = 0.18) calculated using the AUCell_exploreThresholds function. We then used the “FindMarkers” function to identify differentially expressed genes between the high and low pyroptosis-score immune cells, with the filtering criteria set to P < 0.05 and |avg_log2FC|>0.5. Sub-cluster reclustering Based on AUCell scores, we extracted the high-scoring cell subsets and performed reclustering using the same method. We selected 20 principal components for analysis, resulting in 14 sub-clusters. Similarly, we used the “FindAllMarkers” function to identify differentially expressed genes between cell subtypes, with the filtering criteria set to P < 0.05. We visualized the expression of marker genes in high-scoring sub-clusters using violin plots. Cell–cell communication analysis R package ‘CellChat’ (version 1.1.3) was used to perform cell-cell communication analysis. We visualized the number and strength of interactions using circular plots, and the key ligand-receptor pairs involved in signaling between endothelial cells and immune cells were visualized using bubble plots. Pseudotime analysis Pseudotime cell developmental trajectory analysis was analyzed by Monocle 2 package to explore the developmental sequence of the cell clusters from a dynamic perspective. Based on the pseudo-time trajectory, explore the branch points in the development trajectory. The “plot_genes_branched_pseudotime” function was used to the differences in expression trends of PRGs at branch points. TCGA-ESCA differential expression analysis The DESeq2 R package was used to perform analysis of differential expression, with an adjusted p-value threshold of 0.05. Genes with a logFC > 0.5 were considered up-regulated and down-regulated if the logFC were < 0.5. The results of the differential expression analysis were visualized with a heatmap using the R package Pheatmap. The differentially expresssed genes were intersected with differentially expressed genes in TCGA ESCA, and the marker genes of high AUCell-score immune cell subpopulations in TCGA and single-cell data databases were obtained. Analysis of functional enrichment Using the clusterProfiler R package, we performed Gene Ontology(GO) annotation (see Tables 1 and 2) and Kyoto Encyclopedia of Genes and Genomes(KEGG)pathway (see Table 2) enrichment analysis on The clusterProfiler package of R software was utilized to perform GO function and KEGG pathway enrichment analysis of differentially expressed genes between the high/low AUCell score groups, as well as on the intersecting differentially expressed genes between the TCGA-ESCA data and the marker genes of highAUCell score group, an FDR p-value 0.05 was considered statistically significant. Table 1Characteristics of patients in TCGA-ESCA grouped by pyroptosis-related riskscoreCharacteristicsLow_riskHigh_risk P N6564Age, n (%)0.1875< 6026 (20.2%)33 (25.6%)>=6039 (30.2%)31 (24%)Gender, n (%)0.5867FEMALE8 (6.2%)10 (7.8%)MALE57 (44.2%)54 (41.9%)Pathologic_Stage, n (%)0.2897Stage I8 (6.8%)8 (6.8%)Stage II24 (20.5%)27 (23.1%)Stage III23 (19.7%)20 (17.1%)Stage IV1 (0.9%)6 (5.1%)OS.status, n (%)0.2346Alive47 (36.4%)40 (31%)Dead18 (14%)24 (18.6%)OS.time, median (IQR)0.72603 (0.26301, 1.9315)0.61781 (0.19452, 1.3815)0.3670TCGA-The cancer genome atlas program; ESCA-esophageal cancer Characteristics of patients in TCGA-ESCA grouped by pyroptosis-related riskscore TCGA-The cancer genome atlas program; ESCA-esophageal cancer Table 2GO analysis of differentially expressed genes between subclusters in high AUCell score and low AUCell scoreONTOLOGYIDDescription P Q valueBPGO:0032611Interleukin-1 beta production4.68788E-052.7033E-05BPGO:0032612Interleukin-1 production5.89119E-053.3972E-05BPGO:0019730Antimicrobial humoral response5.89119E-053.3972E-05BPGO:0050715Positive regulation of cytokine secretion6.92354E-053.9925E-05BPGO:1,990,266Neutrophil migration0.0004925060.000284BPGO:2,000,403Positive regulation of lymphocyte migration0.0004925060.000284BPGO:0071621Granulocyte chemotaxis0.0005409570.00031194BPGO:0072678T cell migration0.001923460.00110916BPGO:0050792Regulation of viral process0.0028112860.00162113BPGO:0030595Leukocyte chemotaxis0.0031264570.00180287CCGO:0061702Inflammasome complex3.10292E-052.2126E-05CCGO:0000815ESCRT III complex0.00180550.00128745CCGO:0036452ESCRT complex0.0072789570.00519043CCGO:0060205Cytoplasmic vesicle lumen0.0072789570.00519043CCGO:0031983Vesicle lumen0.0072789570.00519043CCGO:0035580Specific granule lumen0.0243747540.01738098CCGO:0044445Cytosolic part0.025310850.01804848CCGO:0034774Secretory granule lumen0.0414082570.02952711CCGO:0070469Respiratory chain0.0414082570.02952711MFGO:0005126Cytokine receptor binding0.0010118420.00043143MFGO:0008009Chemokine activity0.0010118420.00043143MFGO:0042379Chemokine receptor binding0.0016538270.00070516MFGO:0061134Peptidase regulator activity0.0017981760.00076671MFGO:0005125Cytokine activity0.0017981760.00076671MFGO:0097153Cysteine-type endopeptidase activity involved in apoptotic process0.0017981760.00076671MFGO:0008656Cysteine-type endopeptidase activator activity involved in apoptotic process0.0017981760.00076671MFGO:0048018Receptor ligand activity0.0019887050.00084795MFGO:0016505Peptidase activator activity involved in apoptotic process0.0019887050.00084795MFGO:0070851Growth factor receptor binding0.0040362480.00172099GO-Gene Ontology; BP-biological process; CC-cellular component; MF-molecular function GO analysis of differentially expressed genes between subclusters in high AUCell score and low AUCell score GO-Gene Ontology; BP-biological process; CC-cellular component; MF-molecular function Construction of prognostic model To investigate the role of differentially expressed PRGs in esophageal cancer, we conducted univariate Cox regression on 13 PRGs and identified two genes significantly associated with survival (P < 0.05). Using the TCGA-ESCA dataset as a training set and the GSE53625 dataset as a validation set, we applied least absolute shrinkage and selection operator (LASSO) regression to construct a risk model based on gene expression levels. The risk score, calculated using LASSO-Cox regression coefficients, was used to divide patients into high- and low-risk groups. Kaplan-Meier survival analysis and log-rank tests were performed to compare overall survival (OS) between the groups, while time-dependent receiver operating characteristic (ROC) curves were used to assess the model’s predictive accuracy, with the AUC calculated for further validation. The model’s performance was confirmed using the validation dataset, incorporating clinical data to ensure robust survival prediction. Construction of nomogram model To determine whether the risk score’s predictive power for OS is independent of other clinical characteristics (such as age, sex, and pathological stage), Univariate and multivariate Cox regression analyses of clinical variables were performed using 129 esophageal cancer samples with available clinical data from the TCGA database(see Table 1). Subsequently, the nomogram was constructed, and the clinical pathological staging was incorporated into the nomogram for risk prediction. The Harrell concordance index (C-index) was measured to quantify the discrimination performance of the nomograms. Additionally, A calibration curve was generated by comparing the predicted value from the nomogram with the actual survival to assess the performance of the nomogram. Immune infiltration analysis of the prognostic model The TCGA-ESCA gene expression matrix (TPM, Transcripts Per Kilobase Million) was uploaded to the CIBERSORT by employing the LM22 signature. Filtered out samples with P < 0.05, and developed the immune cell infiltration matrix. The ggplot2 package was used to draw bar graphs to show the distribution of 22 immune cell infiltrations in each sample. Drug sensitivity analysis of the prognostic model The Genomics of Drug Sensitivity in Cancer (GDSC) database (www.cancerrxgene.org) can be used to search for tumor drug response data and genome sensitive markers. Using the pRRophetic algorithm, A ridge regression model based on the CCLE cell line expression profile and the TCGA-ESCA gene expression profile was constructed to predict the sensitivity of high- and low-risk groups to common anticancer drugs using IC50 values. Correlation between prognostic pyroptosis-related gene expression and immune checkpoints Using the TCGA esophageal cancer samples, we analyzed the correlation between the expression of prognostic pyroptosis-related genes and common immune checkpoints (CD27, TIGHT, LAIR1, CD96, BTLA, CTLA4, PDCD1, ICOS, CD44, CD276, CD274, TNFSF18, LGALS9, TNFSF15, HHLA2, TNFRSF14, etc.). Scatter plots were generated for the most significant correlations. P < 0.05 was considered statistically significant. Statistical analysis All statistical analyses were performed using R software (version 4.2.0). For comparisons between two groups, statistical significance was estimated using the unpaired Student t-tests and Wilcoxon tests for normally distributed variables and non-normally distributed variables, respectively. Chi-square or Fisher’s exact tests were used for categorical variable comparisons. For multiple group comparisons, Statistical significance was analyzed Kruskal–Wallis test. Result Single-cell sequencing data reveal the heterogeneity among esophageal cancer cells The technical roadmap of this study is shown in Fig. 1. First, we performed dimensionality reduction, clustering, and annotation on the single-cell data of esophageal cancer patients obtained from GSE188900, resulting in 38,122 cells and 20 clusters (Fig. 2A), which were annotated into 7 cell types (Fig. 2B). Clusters 2, 3, and 14 were annotated as T cells (6,747, 18.24%); cluster 15 was annotated as iPS cells (1,119, 3.03%); clusters 0, 7, 11, 18, and 20 were annotated as Monocytes (8,830, 23.87%); cluster 4 was annotated as B cells (2,871, 7.76%); clusters 5, 8, 12, 16, and 19 were annotated as Epithelial cells (7,045, 19.05%); clusters 10 and 16 were annotated as Smooth muscle cells (7,008, 18.95%); and clusters 6 and 10 were annotated as Endothelial cells (3,371, 9.11%).We extracted three types of immune cells and used their marker genes (T cells: CCL5, IL32; B cells: CD79a, CD83; Monocytes: C1QB, LYZ) to create bubble plots (Fig. 2C), violin plots (Fig. 2D), and heatmaps based on the top 5 marker genes for each cell type (Fig. 2E). Finally, we presented the proportion of the three types of immune cells in seven esophageal cancer samples (Fig. 2F). Flow chart (PCA: Principal component analysis, tSNE: T-Distributed random neighbor embedded, GO: Gene Ontology, KEGG: Kyoto encyclopedia of genes and genomes, TCGA: The Cancer genome atlas program, ESCA: Esophageal cancer, DEGs: differentially expressed genes) Single-cell RNA-seq data characterization. (A) TSNE plot of cell distribution in the GSE188900 dataset. (B) Cells are annotated into seven types using SingleR: Monocytes, Smooth Muscle Cells, T Cells, B Cells, Epithelial Cells, Endothelial Cells, and iPS Cells. (C) Bubble plot showing marker gene expression levels for three immune cells (Monocytes, T Cells, B Cells), where bubble size represents the proportion of gene expression in each cluster and darker color indicates higher average expression. (D) Violin plot illustrating marker gene expression within immune cell clusters. (E) Heatmap displaying the top 5 differentially expressed genes for each of the three immune cell types (Monocytes, T Cells, and B Cells), with color ranging from purple (low expression) to yellow (high expression). (F) Bar chart showing the proportion of each cell type across the seven samples. (TSNE: T-distributed Stochastic Neighbor Embedding) Differentially expressed pyroptosis-related genes in esophageal cancer immune cells We extracted the Differmarkers in Immune Cells and intersected them with PRGs, resulting in 13 differentially expressed key pyroptosis genes: AIM2, IRF1, CASP8, CASP1, GZMB, GSDMB, PYCARD, IL18, IL1B, GPX4, CHMP3, CHMP4B, and CYCS (Fig. 3A). We conducted a correlation analysis of the 13 key pyroptosis genes, and the results indicated that these genes are consistently positively correlated (Fig. 3B). We then used t-SNE plots to illustrate the expression levels of the 13 key pyroptosis genes in esophageal cancer tissue cells: AIM2 (Fig. 3C), IRF1 (Fig. 3D), CASP8 (Fig. 3E), CASP1 (Fig. 3F), GZMB (Fig. 3G), GSDMB (Fig. 3H), PYCARD (Fig. 3I), IL18 (Fig. 3J), IL1B (Fig. 3K), GPX4 (Fig. 3L), CHMP3 (Fig. 3M), CHMP4B (Fig. 3N), and CYCS (Fig. 3O). Expression levels and correlation analysis of key prgs in single-cell datasets. (A) Venn diagram showing the selection of key PRGs: 13 key genes were identified by intersecting differentially expressed genes with a set of 51 PRGs. (B) Correlation heatmap of the 13 key pyroptosis genes. Blue indicates positive correlations, while red indicates negative correlations.(C-O) TSNE plots displaying the expression levels of the key pyroptosis genes, including AIM2 (C), IRF1 (D), CASP8 (E), CASP1 (F), GZMB (G), GSDMB (H), PYCARD (I), IL18 (J), IL1B (K), GPX4 (L), CHMP3 (M), CHMP4B (N), and CYCS (O). The intensity of purple, from light to dark, represents increasing gene expression levels. (TSNE: T-distributed Stochastic Neighbor Embedding) Scoring of single-cell pyroptosis We employed the AUCell algorithm to score cells based on the expression of key PRGs in a single-cell dataset, using AUCell explore Thresholds to calculate the activation threshold (AUC = 0.18). Cells were classified into high and low scoring groups according to this threshold (Fig. 4A). First, we visualized the AUC scores of various immune cells using a t-SNE plot (Fig. 4B). Subsequently, we identified differentially expressed genes between the high and low scoring cell groups using the “FindMarkers” function, with selection criteria of P < 0.05 and abs(avg_log2FC) > 0.5. We intersected the differentially expressed genes with 51 pyroptosis-related genes, identifying 24 genes associated with pyroptosis. GO analysis of these 24 genes revealed significant enrichment in biological processes (BP) such as leukocyte chemotaxis, T cell migration, granulocyte chemotaxis, neutrophil migration, and antimicrobial humoral response. Cellular components (CC) were primarily enriched in cytosolic part, specific granule lumen, vesicle lumen, cytoplasmic vesicle lumen, and inflammasome complex, while molecular functions (MF) were predominantly enriched in peptidase activator activity, growth factor receptor binding, receptor ligand activity, cytokine activity, and chemokine activity (Fig. 4C; Tables 3 and 4). KEGG enrichment analysis highlighted pathways related to immune inflammation, including Legionellosis, Necroptosis, Pertussis, NOD-like receptor signaling pathway, NF-kappa B signaling pathway, Toll-like receptor signaling pathway, and TNF signaling pathway (Fig. 4D; Table 5). Single-cell pyroptosis scoring and GO/KEGG enrichment analysis based on the GSE188900 dataset. (A) Pyroptosis scores and activation thresholds were calculated using AUCell, based on the expression of key PRGs; (B) TSNE plot showing AUC scores for individual cells; (C) GO enrichment analysis for differentially expressed genes between the high and low pyroptosis score groups, defined by the AUC activation threshold; (D) KEGG enrichment analysis for differentially expressed genes between the high and low pyroptosis score groups, defined by the AUC activation threshold. (TSNE: T-Distributed Stochastic Neighbor Embedding, GO: Gene Ontology, KEGG: Kyoto Encyclopedia of Genes and Genomes, AUC: Area Under the ROC Curve) Table 3KEGG analysis of differentially expressed genes between patients in high pyroptosis-related riskscore and low pyroptosis-related riskscore based on TCGA-ESCAONTOLOGYIDDescription P Q valueKEGGhsa04145Phagosome2.0905E-061.674E-06KEGGhsa04064NF-kappa B signaling pathway6.8449E-065.481E-06KEGGhsa04621NOD-like receptor signaling pathway2.01E-051.6095E-05KEGGhsa04657IL-17 signaling pathway2.01E-051.6095E-05KEGGhsa04650Natural killer cell mediated cytotoxicity2.1277E-051.7037E-05KEGGhsa04062Chemokine signaling pathway0.00227320.00182027KEGGhsa04668TNF signaling pathway0.004844630.00387935KEGGhsa04217Necroptosis0.005435490.00435248KEGGhsa04620Toll-like receptor signaling pathway0.007817430.00625982KEGGhsa04210Apoptosis0.008639370.00691799KEGGhsa04060Cytokine-cytokine receptor interaction0.012109520.00969672KEGGhsa04662B cell receptor signaling pathway0.012109520.00969672TCGA: The cancer genome atlas program; ESCA༚Esophageal cancer; KEGG-Kyoto encyclopedia of genes and genomes KEGG analysis of differentially expressed genes between patients in high pyroptosis-related riskscore and low pyroptosis-related riskscore based on TCGA-ESCA TCGA: The cancer genome atlas program; ESCA༚Esophageal cancer; KEGG-Kyoto encyclopedia of genes and genomes Table 4GO analysis of differentially expressed genes between patients in high pyroptosis-related riskscore and low pyroptosis-related riskscore based on TCGA-ESCAONTOLOGYIDDescription P Q valueBPGO:0002446Neutrophil mediated immunity6.703E-155.0751E-15BPGO:0045088Regulation of innate immune response6.703E-155.0751E-15BPGO:0042110T cell activation1.7115E-131.2959E-13BPGO:0009615Response to virus7.5865E-165.744E-16BPGO:0001819Positive regulation of cytokine production4.7109E-113.5668E-11BPGO:0051249Regulation of lymphocyte activation2.1601E-101.6355E-10BPGO:0050900Leukocyte migration3.2432E-082.4555E-08BPGO:1,903,706Regulation of hemopoiesis2.7247E-082.063E-08BPGO:0045785Positive regulation of cell adhesion3.3616E-092.5452E-09BPGO:2,001,233Regulation of apoptotic signaling pathway1.4035E-091.0626E-09CCGO:0060205Cytoplasmic vesicle lumen7.0426E-105.5636E-10CCGO:0101002Ficolin-1-rich granule3.6618E-082.8928E-08CCGO:0034774Secretory granule lumen1.2686E-071.0022E-07CCGO:0005925Focal adhesion1.6606E-071.3119E-07CCGO:0005924Cell-substrate adherens junction1.6805E-071.3276E-07CCGO:0070820Tertiary granule5.3172E-064.2006E-06CCGO:0030135Coated vesicle5.3141E-054.1982E-05CCGO:0030139Endocytic vesicle0.000112518.8883E-05CCGO:0042470Melanosome0.00019290.00015239CCGO:0071682Endocytic vesicle lumen0.000314610.00024854CCGO:0098589Membrane region0.000929250.00073411MFGO:0050839Cell adhesion molecule binding1.8953E-051.6192E-05MFGO:0003725Double-stranded RNA binding8.4784E-057.2434E-05MFGO:0042379Chemokine receptor binding0.0007160.0006117MFGO:0044389Ubiquitin-like protein ligase binding0.000801820.00068502MFGO:0005178Integrin binding0.000968680.00082758MFGO:0050786RAGE receptor binding0.000968680.00082758MFGO:0008009Chemokine activity0.000968680.00082758MFGO:0003924GTPase activity0.000968680.00082758MFGO:0005126Cytokine receptor binding0.000993740.00084898MFGO:0005125Cytokine activity0.001896180.00161997TCGA The cancer genome atlas program; ESCA-Esophageal cancer; GO-Gene Ontology, BP-biological process; CC-Cellular component; MF: Molecular function GO analysis of differentially expressed genes between patients in high pyroptosis-related riskscore and low pyroptosis-related riskscore based on TCGA-ESCA TCGA The cancer genome atlas program; ESCA-Esophageal cancer; GO-Gene Ontology, BP-biological process; CC-Cellular component; MF: Molecular function Table 5KEGG analysis of differentially expressed genes between subclusters in high AUCell score and low AUCell scoreONTOLOGYIDDescription P Q valueKEGGhsa05134Legionellosis3.04197E-061.8862E-06KEGGhsa04217Necroptosis0.0002223940.00013789KEGGhsa05133Pertussis0.0002223940.00013789KEGGhsa04621NOD-like receptor signaling pathway0.0002443890.00015153KEGGhsa05323Rheumatoid arthritis0.0002980640.00018481KEGGhsa04625C-type lectin receptor signaling pathway0.000386480.00023963KEGGhsa04668TNF signaling pathway0.0004435440.00027502KEGGhsa05132Salmonella infection0.0005692530.00035296KEGGhsa04623Cytosolic DNA-sensing pathway0.0015618510.00096842KEGGhsa04064NF-kappa B signaling pathway0.0451475310.02799342KEGGhsa04620Toll-like receptor signaling pathway0.0451475310.02799342KEGGhsa04062Chemokine signaling pathway0.0187279560.01161214KEGG kyoto encyclopedia of genes and genomes KEGG analysis of differentially expressed genes between subclusters in high AUCell score and low AUCell score KEGG kyoto encyclopedia of genes and genomes Analysis of pyroptosis-related subgroups Based on the single-cell dataset, we re-clustered immune cell subgroups with high pyroptosis scores into 14 clusters (Fig. 5A). The violin plot illustrates the marker genes for each cluster (Fig. 5B). Additionally, we used t-SNE plots to show the expression of key PRGs AIM2 (Fig. 5C), IRF1 (Fig. 5D), CASP8 (Fig. 5E), CASP1 (Fig. 5F), GZMB (Fig. 5G), GSDMB (Fig. 5H), PYCARD (Fig. 5I), IL18 (Fig. 5J), IL1B (Fig. 5K), GPX4 (Fig. 5L), CHMP3 (Fig. 5M), CHMP4B (Fig. 5N), and CYCS (Fig. 5O) in the cell subgroups. Re-clustering of pyroptosis-related subpopulations based on the GSE188900 dataset. (A) TSNE plot displaying the re-clustering of pyroptosis-related subpopulations into 14 clusters. (B) Violin plot showing the expression levels of marker genes across the 14 subpopulations. (C-O) TSNE plots illustrating the expression of key pyroptosis-related genes: AIM2 (C), IRF1 (D), CASP8 (E), CASP1 (F), GZMB (G), GSDMB (H), PYCARD (I), IL18 (J), IL1B (K), GPX4 (L), CHMP3 (M), CHMP4B (N), and CYCS (O) within the subpopulations. (TSNE: T-Distributed Stochastic Neighbor Embedding) Analysis of cell -cell communication in pyroptosis-related subpopulations We used CellChat to infer and quantify the communication between immune cells (Monocytes, B cells, T cells) with high and low pyroptosis scores. The number (Fig. 6A) and strength (Fig. 6B) of cellular communications were visualized through circular plots. It can be observed that immune cells with both high and low pyroptosis scores exhibit a greater number of interactions, with higher interaction strength, among which B cells show the most diverse and complex interactions with other cells. We also compared all significant receptor-ligand pairs during signal emission/reception in high- and low-scoring immune cell groups, using chord diagrams (Fig. 6C) and dotplots (Fig. 6D). The results demonstrated that the MIF-(CD74 + CD44) signaling pathway plays a critical role in the interactions between high-scoring cell groups and immune cells. In addition, we analyzed the secretory cell communication patterns of endothelial and immune cells with high and low pyroptosis scores. Through a heatmap (Fig. 6E), we visualized how endothelial and immune cells with high and low pyroptosis scores coordinate with each other, as well as the specific signaling pathways driving these communications. The results revealed that for high-scoring B cells, the outgoing pathway is MIF, and the incoming pathways are MIF, GALECTIN, and BAFF. For high-scoring monocytes, the outgoing pathways are GALECTIN, BAFF, and CXCL, while the incoming pathways are MIF and GALECTIN. For high-scoring T cells, the outgoing pathway is MIF, and the incoming pathways are MIF, GALECTIN, and CXCL. For low-scoring B cells, the outgoing pathway is MIF, and the incoming pathways are MIF, GALECTIN, and BAFF. For low-scoring monocytes, the outgoing pathway is BAFF, and the incoming pathway is GALECTIN. For low-scoring T cells, the outgoing pathway is MIF, and the incoming pathways are MIF and GALECTIN. Moreover, the heatmap indicates that the interaction pathways among immune cells with high pyroptosis scores are more frequent and exhibit stronger interactions compared to those with low pyroptosis scores. Illustrates the analysis of cell communication related to cellular pyroptosis. Circles (A, B) and the chord diagram (C) display the quantity and intensity of communication between immune cells (monocytes, B cells, T cells) with high and low pyroptosis scores. Nodes represent different cell types, arrows indicate the direction of signaling, and the thickness of lines reflects the number and strength of communication—thicker lines indicate more frequent or stronger interactions. In the chord diagram, the outer colors represent the cell types involved in communication, while the internal connections show receptor-ligand pairs. (D) depicts the receptor-ligand pairs involved in cell communication, with the vertical axis showing cell types and the horizontal axis showing receptor-ligand pairs; the size of circles indicates significance, and colors shift from blue to red as communication probability increases. (E) features a heatmap that reveals the signaling patterns of immune cells with high and low pyroptosis scores, where the depth of green squares indicates correlation strength Analysis of immune cell pseudotime based on high pyroptosis scoring Single-cell sequencing technology provides a powerful tool for further understanding the mechanisms of cancer due to its high resolution; however, cancer development is a dynamic process. In the pseudotime analysis, high-pyroptosis immune cells were further annotated based on canonical immune-cell markers. Cluster 0 was annotated as cytotoxic T cells, cluster 1 as CD4 + cytotoxic T cells, clusters 2, 11, 12, and 13 as B cells, clusters 3, 5, and 14 as regulatory T cells, cluster 7 as CD8 + T cells, cluster 8 as monocytes, and cluster 10 as mast cells. Cluster 9 could not be confidently assigned to a known immune-cell subtype based on the marker genes available in the current analysis and was therefore labeled as “Unknown.” This cluster was retained in the pseudotime analysis to avoid arbitrary exclusion of potentially informative immune cells. Nevertheless, its biological identity remains uncertain, and further annotation using additional reference atlases or experimental validation will be required. The t-SNE plot displays the annotation results for each cell population (Fig. 7A). Subsequently, to analyze the developmental processes of the cell clusters from a dynamic perspective, we performed pseudotime analysis on these cells using the R package Monocle. Our results indicate that the developmental processes between cell clusters do not follow a single pathway; instead, they exhibit a relatively complex landscape, with two branching points dividing the clusters into a near-dumbbell development pattern (Fig. 7B) and allowing for the distinction of five different developmental stages (Fig. 7C). We further illustrated the changes in each cell type during the developmental process (Fig. 7D). Finally, we conducted further analysis on the genes that drive the processes during these five developmental stages (Fig. 7E), revealing that the expression of 42 pyroptosis-related genes changed, thereby extending the cell developmental trajectory into different stages. This further validates the significant impact and importance of pyroptosis on the development of immune cells in esophageal cancer from a dynamic perspective. We illustrated six representative pyroptosis-related genes (Fig. 7F). Presents the analysis of developmental trajectories of immune cells with high pyroptosis scores. (A)TSNE categorizes these cells into eight types based on classical immune cell markers: cytotoxic T cells, CD4 + cytotoxic T cells, B cells, regulatory T (Treg) cells, monocytes, CD8 + T cells, mast cells, and unclassified cells. (B) shows pseudotime values reflecting the direction of trajectory development, with darker colors indicating earlier stages and lighter colors indicating later stages. (C) illustrates the developmental trajectory of the cells, using distinct colors to represent different stages, aligning with previous clustering results. (D) presents the developmental processes of various cell types, distinguished by color and corresponding to prior clusters. (E) features a heatmap of pyroptosis gene expression during cell development, with each cluster representing different developmental stages. (F) depicts the gene expression trends across stages, where the x-axis represents pseudotime values and the y-axis represents expression levels, with points indicating cells, differentiated by color for each developmental stage Analysis of esophageal cancer patients and pyrotosis We divided the TCGA-ESCA samples into two groups (ESCA and Normal) and initially used the DESeq2 package for differential analysis of sequencing data to identify differentially expressed genes. Genes with logFC > 0.5 and P value < 0.05 were classified as up-regulated (up regulated genes), while genes with logFC < -0.5 and P value < 0.05 were classified as down-regulated (down regulated genes), resulting in 2898 up-regulated genes and 2399 down-regulated genes (Table S1). The heatmap illustrates the expression of 13 key PRGs identified from single-cell data in TCGA tumor samples compared to normal samples (Fig. 8A), while the volcano plot displays the gene expression differences between the two groups (Fig. 8B). Subsequently, we performed differential analysis on the expression of the 13 key pyroptosis-related genes between tumor and normal samples (Fig. 8C), revealing that these genes were significantly overexpressed in tumor samples (AIM2: P = 0.01; GSDMD: P < 0.001; IL1B: P = 0.002; GZMB: P < 0.001; PYCARD: P < 0.001; IRF1: P < 0.001; CASP8: P = 0.001; CHMP4B: P < 0.001; GPX4: P < 0.001; CYCS: P < 0.001; CHMP3: P = 0.033), suggesting that pyroptosis plays a significant role in the pathogenesis of esophageal cancer. Additionally, we intersected the differentially expressed genes obtained from TCGA data with marker genes selected from single-cell data based on pyroptosis scoring, resulting in 551 differentially expressed genes associated with pyroptosis (Fig. 8D). GO analysis of these 551 genes revealed that they are primarily enriched in biological processes (BP) such as neutrophil-mediated immunity, T cell activation, positive regulation of cytokine production, regulation of lymphocyte activation, regulation of apoptotic signaling pathway, and leukocyte migration. In terms of MF, the genes are mainly enriched in chemokine receptor binding, integrin binding, RAGE receptor binding, chemokine activity, and cytokine activity. For CC, they are predominantly associated with cytoplasmic vesicle lumen, focal adhesion, tertiary granules, coated vesicles, and endocytic vesicles (Fig. 8F; Table 2). KEGG analysis results indicate that these genes are significantly enriched in pathways such as Cytokine-cytokine receptor interaction, Apoptosis, Chemokine signaling pathway, TNF signaling pathway, B cell receptor signaling pathway, Toll-like receptor signaling pathway, NOD-like receptor signaling pathway, IL-17 signaling pathway, Necroptosis, and Natural killer cell-mediated cytotoxicity (Fig. 8F; Table 3). These pathways and functions are all related to immunity and cell death. Correlation Analysis Between Esophageal Cancer and Pyroptosis Based on the TCGA-ESCA Dataset. (A) Gene expression levels in tumor (ESCA) and normal samples from the TCGA-ESCA dataset. (B) Volcano plot showing the differential gene expression between tumor and normal groups. (C) Expression differences of key PRGs between ESCA tumor tissues and normal tissues. (D) Intersection analysis of differentially expressed genes (DEGs) from TCGA data with cell marker genes, resulting in 551 intersecting DEGs. (E, F) GO enrichment analysis of intersecting DEGs.Significance is indicated as p < 0.05 (*, p < 0.05; **, p < 0.01; ***, p < 0.001). Abbreviations: TCGA – The Cancer Genome Atlas Program; ESCA – Esophageal Cancer; GO Gene Ontology; KEGG – Kyoto Encyclopedia of Genes and Genomes; DEGs – Differentially Expressed Genes Construction and validation of a pyroptosis gene scoring model We established a pyroptosis-related gene scoring model using TCGA-ESCA as the training cohort and GSE53625 as the external validation cohort. Univariate Cox regression identified two survival-associated genes, GSDMB and CYCS, which were subsequently incorporated into a LASSO-Cox prognostic model. Patients were stratified into high- and low-risk groups according to the median risk score. In the TCGA training cohort, the time-dependent ROC analysis yielded AUC values of 0.701, 0.607, and 0.645 for 2-, 3-, and 4-year overall survival, respectively, suggesting moderate predictive performance. Kaplan–Meier analysis showed that patients in the low-risk group had significantly better overall survival than those in the high-risk group (P = 0.019). In the external GSE53625 validation cohort, the corresponding AUC values were 0.560, 0.561, and 0.545 for 2-, 3-, and 4-year overall survival, respectively. Although the Kaplan–Meier curve showed a trend toward better survival in the low-risk group, the difference did not reach statistical significance (P = 0.076). Therefore, the external validation results suggest that the prognostic model has limited generalizability across independent datasets. These findings indicate that the two-gene model may capture pyroptosis-related prognostic signals in ESCC, but its clinical predictive utility requires further validation in larger, harmonized, and prospectively collected cohorts. Construction of pyroptosis associated clinical prognostic model We integrated pyroptosis related risk scores with different clinical pathological features based on the TCGA-ECSA dataset, starting with univariate and multivariate Cox analyses. The results of the univariate forest plot indicated that both necroptosis risk scores and tumor pathological stage are risk factors (Figs. 9 and 10A). We then constructed a predictive nomogram to estimate OS for esophageal cancer patients (Fig. 11B). We utilized the C-Index to calculate the discriminatory ability of the nomogram, which demonstrated a high concordance (concordant = 0.686). Calibration curves indicated a good agreement between the nomogram’s estimated OS values for 1, 2, and 3 years and the actual observed values of the patients (Fig. 9C). The results of the decision curve analysis (DCA) showed that this prognostic model outperformed the extreme line within a certain range, and that the prognostic model incorporating necroptosis risk scores had better efficacy than one using only pathological staging (Fig. 10D). The risk score was constructed based on the LASSO coefficients. Specifically, CYCS functions as a risk factor, where its upregulation contributes to a higher risk score and is associated with an immunosuppressive microenvironment. In contrast, GSDMB functions as a protective factor, where its upregulation is associated with increased immune infiltration and a lower risk score. Construction of a pyroptosis-related scoring model based on the TCGA-ESCA dataset. (A) LASSO regression coefficient trajectory plot; (B) Ten-fold cross-validation for LASSO regression; (C) timeROC curve for the TCGA-ESCA training set;(D) Kaplan-Meier (K-M) survival curve for the high- and low-risk groups in the TCGA-ESCA training set; (E) Risk factor plot for the TCGA-ESCA training set; (F) timeROC curve for the GSE53625 validation set; (G) K-M survival curve for the high- and low-risk groups in the GSE53625 validation set. (P < 0.05 is considered statistically significant; LASSO: Least Absolute Shrinkage and Selection Operator; TCGA: The Cancer Genome Atlas; ESCA: Esophageal Cancer; K-M: Kaplan–Meier; ROC: Receiver Operating Characteristic Curve) Construction of a pyroptosis-based clinical prognostic model using the TCGA-ESCA dataset. (A) The hazard ratios (HR) and p-values from univariate and multivariate Cox regression analyses, combining the risk score with clinicopathological features. (B) A nomogram integrating the risk score and tumor pathological stage into the pyroptosis prognostic model for prediction. (C) Calibration curve for the pyroptosis prognostic model, where the closer the curve is to the diagonal dashed line, the better the predictive performance. The model demonstrates an ideal predictive capability (C-Index=0.686). (D) Decision curve analysis (DCA) for the pyroptosis clinical prognostic model, with the yellow line representing the constructed model and the blue curve representing a model with only pathological staging included. (P < 0.05 is considered statistically significant) (TCGA: The Cancer Genome Atlas Program; ESCA: Esophageal Cancer; DCA: Decision Curve Analysis) Analysis of Immune Infiltration Levels in the Pyroptosis-Related Clinical Prognostic Model. (A) A comprehensive landscape of immune cell infiltration in TCGA-ESCA samples, analyzed using the CIBERSORTx algorithm. (B) A heatmap showing the correlation of immune cell infiltration in TCGA-ESCA samples; blue indicates a negative correlation, red indicates a positive correlation, and the deeper the color, the stronger the correlation. (C) Correlation between the pyroptosis-related gene GSDMB and the ImmuneScore. (D-F) Correlation of CYCS with ImmuneScore (D), StromalScore (E), and ESTIMATE Score (F). (G-H) Lollipop charts illustrating the correlation between the expression of pyroptosis-related genes GSDMB (G) and CYCS (H) and immune cell infiltration levels. (I) Analysis of immune cell infiltration differences between the high- and low-risk groups in the pyroptosis-related prognostic model. Statistical significance: p < 0.05 is considered significant (*, p < 0.05; **, p < 0.01; ***, p < 0.001). Abbreviations: TCGA (The Cancer Genome Atlas Program); ESCA (Esophageal Cancer) Analysis of immune infiltration levels in pyroptosis-related clinical prognostic models To analyze the differences in immune infiltration levels associated with the pyroptosis scoring model in esophageal cancer, we utilized the CIBERSORTx algorithm to calculate the infiltration levels of 22 types of immune cells in the TCGA-ESCA dataset (Fig. 11A). We then analyzed the correlation between two genes and immune scores, finding that GSDMB was significantly positively correlated with the Immune Score (P = 0.005, R = 0.220) (Fig. 11C), while CYCS showed significant negative correlations with the Immune Score (P < 0.001, R=-0.276) (Fig. 11D), Stromal Score (P < 0.001, R=-0.302) (Fig. 11E), and ESTIMATE Score (P < 0.001, R=-0.304) (Fig. 11F). We also analyzed the correlation between two prognosis-related pyroptosis genes and immune cell infiltration in esophageal cancer. We found that CYCS was significantly negatively correlated with the infiltration levels of NK cells (P < 0.001, R=-0.464), Mast cells (P < 0.001, R=-0.375), iDC (P < 0.001, R=-0.347), Cytotoxic cells (P < 0.001, R=-0.327), TFH (P < 0.001, R=-0.313), DC (P < 0.001, R=-0.306), Tcm (P < 0.001, R=-0.296), NK CD56dim cells (P < 0.001, R=-0.296), TReg (P = 0.001, R=-0.254), T cells (P = 0.001, R=-0.253), Macrophages (P = 0.003, R=-0.229), B cells (P = 0.005, R=-0.221), CD8 T cells (P = 0.020, R=-0.183), and Th1 cells (P = 0.034, R=-0.167) (Fig. 11G), suggesting that high expression of CYCS may be associated with the suppression of various immune cells. The expression of GSDMB was positively correlated with the infiltration levels of Th17 cells (P < 0.001, R = 0.581), Eosinophils (P < 0.001, R = 0.553), T helper cells (P < 0.001, R = 0.320), NK CD56dim cells (P < 0.001, R = 0.314), Tem (P < 0.001, R = 0.313), CD8 T cells (P < 0.001, R = 0.303), pDC (P < 0.001, R = 0.298), T cells (P < 0.001, R = 0.284), B cells (P < 0.001, R = 0.265), and Mast cells (P = 0.026, R = 0.175), while it was negatively correlated with Tcm (P < 0.001, R=-0.327), NK cells (P < 0.001, R=-0.306), Tgd (P = 0.009, R=-0.204), NK CD56dim cells (P = 0.016, R=-0.190), and iDC (P = 0.042, R = 0.160) (Fig. 11H). We then compared the immune cell infiltration levels between high-risk and low-risk groups in the cell death-related prognostic model. We found that in the high-risk group, the levels of B cells memory (P = 0.015) and Mast cells resting (P = 0.018) were lower than those in the low-risk group, while the level of Dendritic cells activated (P = 0.018) was significantly higher in the high-risk group (Fig. 11I), suggesting that this prognostic model may be associated with the activation or suppression of these three types of immune cells. Analysis of pyroptosis-related scoring and chemotherapeutic drug sensitivity We analyzed the relationship between the pyroptosis-related scoring and chemotherapeutic drug sensitivity. We found that patients with ESCA in the TCGA-ESCA high-risk and low-risk groups exhibited significant differences in sensitivity to various chemotherapeutic drugs, including CCT007093 (P < 0.001) (Fig. 12A), LFM.A13 (P < 0.001) (Fig. 12B), Lenalidomide (P < 0.001) (Fig. 12C), JNJ.26,854,165 (P = 0.022) (Fig. 12D), VX.702 (P = 0.001) (Fig. 12E), AZD6482 (P = 0.004) (Fig. 12F), KIN001.135 (P = 0.008) (Fig. 12G), AP.24,534 (P = 0.014) (Fig. 12H), and IPA.3 (P = 0.020) (Fig. 12I), with the high-risk group showing significantly higher sensitivity than the low-risk group. This suggests that pyroptosis-related scoring system constructed in this study may have certain implications for the clinical use of chemotherapeutic drugs. Chemotherapy drug sensitivity analysis based on the pyroptosis-related clinical prognostic model. (A-L) Sensitivity analysis of common chemotherapy drugs, comparing high-risk and low-risk groups classified by the median pyroptosis score in TCGA-ESCA patients. The drugs analyzed include CCT007093 (A), LFM.A13 (B), Lenalidomide (C), JNJ.26,854,165 (D), VX.702 (E), AZD6482 (F), KIN001.135 (G), AP.24,534 (H), and IPA.3 (I).Statistical significance: p < 0.05 is considered significant (*, p < 0.05; **, p < 0.01; ***, p < 0.001).Abbreviations: TCGA (The Cancer Genome Atlas Program); ESCA (Esophageal Cancer) Analysis of prognosis-related pyroptosis genes on immune checkpoints Finally, we conducted a correlation analysis between the expression of prognosis-related key pyroptosis genes from the TCGA-ESCA dataset and the expression of commonly known immune checkpoint genes. The results showed that the expression of CYCS in esophageal cancer was significantly negatively correlated with the expression of common immune checkpoint genes, including CD27 (P < 0.001, R=-0.39) (Fig. 13A), TIGHT (P < 0.001, R=-0.36) (Fig. 13B), LAIR1 (P < 0.001, R=-0.36) (Fig. 13C), CD96 (P < 0.001, R=-0.35) (Fig. 13D), BTLA (P < 0.001, R=-0.33) (Fig. 13E), CTLA4 (P < 0.001, R=-0.33) (Fig. 13F), PDCD1 (P < 0.001, R=-0.30) (Fig. 13G), and ICOS (P = 0.0013, R=-0.28) (Fig. 13H). The expression of GSDMB in esophageal cancer showed a significant positive correlation with the expression levels of CD44 (P < 0.001, R=-0.42) (Fig. 13I), CD276 (P < 0.001, R=-0.31) (Fig. 13J), CD274 (P = 0.008, R=-0.23) (Fig. 13K), and TNFSF18 (P = 0.025, R=-0.20) (Fig. 13L). Additionally, it was significantly positively correlated with the expression levels of LGALS9 (P < 0.001, R = 0.61) (Fig. 13M), TNFSF15 (P < 0.001, R = 0.54) (Fig. 13N), HHLA2 (P < 0.001, R = 0.51) (Fig. 13O), and TNFRSF14 (P < 0.001, R = 0.53) (Fig. 13P). Correlation analysis between prognosis-related pyroptosis genes and immune checkpoints in the TCGA-ESCA dataset. (A-H) Scatter plots showing the correlation between CYCS expression and the following immune checkpoints: CD27 (A), TIGIT (B), LAIR1 (C), CD96 (D), BTLA (E), CTLA4 (F), PDCD1 (G), and ICOS (H). (I-P) Scatter plots illustrating the correlation between GSDMB expression and the following immune checkpoints: CD44 (I), CD276 (J), CD274 (K), TNFSF18 (L), LGALS9 (M), TNFSF15 (N), HHLA2 (O), and TNFRSF14 (P). Statistical significance is indicated by p < 0.05.Abbreviations: TCGA (The Cancer Genome Atlas Program); ESCA (Esophageal Cancer). (TCGA: The Cancer Genome Atlas Program; ESCA: Esophageal Cancer) Discussion Many patients with ESCC are diagnosed at an advanced or metastatic stage, making prognosis challenging. For locally advanced cases, surgery alone has not produced satisfactory outcomes [22]., Traditional histopathological prognostic factors, such as tumor size, stage, and grade, often fail to fully address diagnostic and prognostic needs in ESCC [23], To improve prognostic stratification in ESCC, we developed a pyroptosis-related risk score based on GSDMB and CYCS. In the TCGA training cohort, this model showed moderate predictive performance and was associated with overall survival. However, the external validation performance in GSE53625 was limited, with relatively low AUC values for 2-, 3-, and 4-year survival and a Kaplan–Meier log-rank P-value that did not reach statistical significance. Therefore, the prognostic model should not be interpreted as a fully validated clinical tool at this stage. Instead, it should be regarded as an exploratory pyroptosis-related signature that requires further optimization and validation. Several factors may explain the reduced performance in the external cohort, including differences in sequencing platforms, sample composition, clinical annotation, treatment background, batch effects, and cohort-specific biological heterogeneity. In addition, the use of only two genes may restrict model complexity and reduce its ability to capture the full prognostic heterogeneity of ESCC. Future studies should validate this signature in additional ESCC-specific cohorts and may consider integrating clinical variables, immune features, and additional pyroptosis-related genes to improve model robustness and generalizability.When integrated with pathological staging into a nomogram, the combined model demonstrated strong clinical efficacy, aligning with findings from previous studies [14, 24]. Models with high pyroptosis-related risk scores are strongly associated with poor prognosis in esophageal cancer. This is supported by single-cell immune cell analyses, which reveal elevated expression of 13 key PRGs in tumor cells. The significance of this elevated expression lies in its role in promoting an inflammatory tumor microenvironment. Pyroptosis, as a form of proinflammatory cell death, fosters conditions favorable to tumor growth, thereby exacerbating disease progression and leading to a worse prognosis. This explanation is consistent with findings from previous studies, which have confirmed the tumor-promoting effects of pyroptosis [25, 26]. In our study, AUCell scores based on 13 pyroptosis-related genes were used to classify immune cells into high and low expression groups. Immune cells with higher pyroptosis gene expression exhibited more frequent and stronger interactions. The MIF-(CD74 + CD44) signaling pathway played a crucial role in communication between high-scoring immune cells, with B cells showing the most complex interactions. Tumor-infiltrating B lymphocytes (TIL-Bs) were significantly elevated in responders to immune checkpoint blockade (ICB) therapy for ESCC [27]and were associated with improved outcomes in patients receiving anti-PD-1/PD-L1 treatment [28]. Additionally, cancer-derived HMGB1 promoted angiogenesis by driving proliferating B cells toward proangiogenic signals, further contributing to disease progression [29]. We found that patients with high pyroptosis-related gene risk scores exhibited greater sensitivity to drugs such as CCT007093, LFM-A13, Lenalidomide, JNJ-26,854,165, VX-702, AZD6482, KIN001-135, AP-24,534, and IPA-3 compared to the low-score group. This discovery offers new insights for the development of chemotherapy drugs for esophageal cancer. Future experiments based on these findings could potentially lead to breakthroughs in esophageal cancer treatment. Emerging evidence suggests that pyroptosis, a newly recognized form of programmed cell death, plays a crucial role in both tumor development and response to cancer therapies [30, 31]. Our findings align with these observations, showing that PRGs in ESCC immune cells are significantly enriched in key immune and inflammatory pathways, including cytokine-cytokine receptor interaction, apoptosis, chemokine signaling, TNF signaling, B cell receptor signaling, Toll-like receptor signaling, NOD-like receptor signaling, IL-17 signaling, necroptosis, and natural killer cell-mediated cytotoxicity. This indicates that PRGs are closely intertwined with the immune microenvironment in ESCC. Our findings provide potential guidance for clinical immunotherapy strategies in ESCC. Specifically, the correlation analysis revealed that GSDMB (a protective factor associated with the low-risk group) was significantly positively correlated with critical immune checkpoints, including PD-L1 (CD274), CTLA4, and LGALS9. High expression of PD-L1 is a widely accepted biomarker for response to Immune Checkpoint Blockade (ICB). Therefore, patients in the low-risk group, characterized by high GSDMB expression and a ‘hot’ immune microenvironment enriched with CD8 + T cells, may derive greater clinical benefit from immunotherapy. Conversely, the high-risk group (associated with high CYCS) exhibited a negative correlation with immune checkpoints and lower immune infiltration, indicative of an immunosuppressive ‘cold’ tumor phenotype. For these high-risk patients, immunotherapy alone might be insufficient, and the specific chemotherapeutic agents identified in our drug sensitivity analysis (e.g., inhibitors targeting specific pathways shown in Fig. 12) might offer a more effective therapeutic alternative. In addition, cluster 9 in the pseudotime analysis remained unannotated. Although alternative marker-based annotation was attempted using canonical immune-cell markers, this cluster did not show a sufficiently specific marker profile for confident assignment to a recognized immune-cell subtype. This may reflect a transitional immune-cell state, low-quality or mixed transcriptional features, or limitations of reference-based annotation in the ESCC immune microenvironment. Because cluster 9 participated in the inferred pseudotime trajectory, we retained it in the analysis but interpreted it cautiously. Future studies using larger single-cell datasets, multi-reference annotation strategies, and protein-level validation will be needed to clarify the identity and functional relevance of this cluster. This study has several limitations. First, the conclusions are primarily based on bioinformatic analyses and require validation through in vivo and in vitro experiments. Another limitation of this study is the imbalance between tumor and normal samples in the single-cell RNA-seq dataset. The scRNA-seq analysis included seven ESCC tumor samples but only one normal control sample. Although this dataset allowed us to characterize tumor-associated immune heterogeneity and pyroptosis-related immune-cell states at single-cell resolution, the limited number of normal controls may introduce bias when comparing tumor and non-tumor cell populations. Inter-individual heterogeneity in the single normal sample may also influence the identification of differentially expressed genes and marker genes. Therefore, the single-cell findings should be interpreted with caution, and validation using larger scRNA-seq datasets with balanced tumor and normal samples is needed. Third, although the pyroptosis-related prognostic model showed moderate performance in the TCGA training cohort, its performance in the external GSE53625 validation cohort was limited. Therefore, the model should be considered exploratory and requires further validation in larger, independent, and clinically harmonized ESCC cohorts. Finally, one immune cluster in the pseudotime analysis could not be confidently annotated, and its biological identity requires further investigation. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1. Acknowledgements None. Author contributions Conception and design:HZ, QLRAdministrative suppor: WJ, JSHProvision of study materials or patients: HZ, JSHCollection and assembly of data: HZ, JSHData analysis and interpretation: HZManuscript writing: All authors. Final approval manuscript: All authors. Funding This study was supported by the Jiangsu Provincial Health Commission Research Project on Elderly Health (LKZ2022019). Data availability The scRNA-seq datasets analyzed in this study are publicly available in the Gene Expression Omnibus (GEO) repository under accession number GSE188900. No new datasets were generated in the current work. Declarations Ethics approval and consent to participate The single-cell RNA sequencing data analyzed in this study were obtained from the publicly available dataset GSE188900 on the Gene Expression Omnibus (GEO) repository. The data were fully anonymized, and no additional ethics approval was required for this secondary analysis. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Reference Associated Data Supplementary Materials Supplementary Material 1. Data Availability Statement The scRNA-seq datasets analyzed in this study are publicly available in the Gene Expression Omnibus (GEO) repository under accession number GSE188900. No new datasets were generated in the current work.