Identification of GADD45B, HMGB3, LMNB2, and MFAP5 as lactylation-related prognostic markers for survival prediction in esophageal cancer Authors: Shuo Li, Shaomin Quan, Jun Ma, Yunbo Bai, Jing Sun, Yuan Wang, Zhongjing Wang, Juntao Zhang, Song Zhang, Wanjun Li, Zhigang Fan Published: 2026 ABSTRACT AbstractBackgroundLactylation, a protein modification driven by lactate produced during glycolysis, has been implicated in tumorigenesis and immune suppression, and may influence patient prognosis. This study aimed to identify lactylation-related prognostic genes and explore their potential functional relevance.MethodsThe Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) datasets were used to identify lactylation-related genes (LRGs) in esophageal cancer (ESCA). Prognostic LRGs were determined by univariate and multivariate Cox regression analyses. Correlations of prognostic LRGs with ESCA-associated genes, immune cell infiltration, and five classes of immunomodulatory genes were evaluated, followed by functional enrichment and drug sensitivity analyses. These LRGs were validated using an independent dataset and immunohistochemistry (IHC) assays.ResultsA total of 321 predictive LRGs were identified in ESCA. Four LRGs (GADD45B, HMGB3, LMNB2, and MFAP5) were further recognized as independent prognostic factors (P<0.05). These genes were significantly associated with ESCA-related genes, tumorigenesis, and immune-inflammatory processes. Moreover, they showed significant correlations with immune cells (e.g., activated B cells, natural killer T cells) and multiple immunomodulatory genes (such as CSF1R, CXCL12, CCL23, and CCR2). In addition, the four LRGs had different sensitivities to the drugs 5-fluorouracil, gefitinib, paclitaxel, sorafenib, vincristine, and cisplatin. HMGB3 knockdown significantly inhibited the viability, migration, and invasion of KYSE-140 cells while promoting cell apoptosis.ConclusionsWe identified four genes associated with lactylation activity scores that serve as independent prognostic candidates for ESCA. These LRGs hold promise as potential therapeutic targets and prognostic biomarkers in ESCA.Keywords: Lactylation, esophageal cancer (ESCA), prognosis, immunoregulatory, therapy Highlight box Key findings • GADD45B, HMGB3, LMNB2, and MFAP5 were identified as lactylation-related genes (LRGs), of which their expression levels were significantly associated with the prognosis of esophageal cancer (ESCA). These genes may serve as potential prognostic biomarkers for ESCA. What is known and what is new? • Lactylation has been implicated in tumor initiation, immune suppression, and cancer prognosis. • By integrating bulk and single-cell RNA sequencing data with lactylation activity scores, this study identified four novel LRGs that are differentially expressed in ESCA. These genes were independently associated with patient prognosis and showed significant correlations with multiple immune regulatory genes. Our findings provide new insights into the prognostic relevance of LRGs and their potential involvement in the immune landscape of ESCA. What is the implication, and what should change now? • This study highlights a potential link between LRGs, prognosis, and immune regulation in ESCA. The identified genes may represent promising diagnostic biomarkers and molecular targets, which could contribute to improved prognostic stratification and personalized therapeutic strategies for ESCA patients. Nevertheless, the precise biological functions and mechanistic roles of these genes in ESCA progression and immune modulation require further experimental and clinical validation. Introduction Esophageal cancer (ESCA) ranks as the seventh leading cause of cancer-related mortality worldwide according to GLOBOCAN 2022 statistics (1). More than half of global ESCA cases occur in China (2). ESCA comprises two major histological subtypes: esophageal squamous cell carcinoma (ESCC) and esophageal adenocarcinoma (EAC). Current treatment options include surgery, radiotherapy, chemotherapy, targeted therapy, and immunotherapy (3). However, the overall prognosis of ESCA remains poor, as most patients are diagnosed at an advanced stage, with a 5-year survival rate of only ~20% (4). Therefore, identifying reliable prognostic biomarkers is crucial for predicting clinical outcomes and guiding therapeutic strategies in ESCA. The Warburg effect reveals that, even under sufficient oxygen supply, tumor cells preferentially undergo glycolysis, leading to lactate production and accumulation (5). Lactylation, a lysine modification that depends on intracellular lactate levels, broadly remodels the tumor microenvironment (6). Emerging evidence shows that lactylation participates in multiple critical biological processes and serves as a pivotal bridge between tumor metabolic reprogramming and epigenetic regulation. For instance, tumor-derived lactate-induced lactylation can alter the molecular phenotype of CD4+ T cells or macrophages, thereby promoting tumor immune evasion (7). In ESCA, lactylation plays an indispensable role in regulating tumor initiation, progression, and chemoresistance (8,9). Clinical evidence has shown that pan-lactylation levels are significantly elevated in ESCC tissues from patients who fail to achieve major pathological response following chemoimmunotherapy (10). Mechanistically, H2BC9 lactylation has been reported to promote tumor cell proliferation and growth through activation of the Wnt/β-catenin signaling pathway (11). In addition, Wilms tumor 1-associated protein-mediated m6A modification enhances the stability of PDIA3P1, which promotes glycolysis and increases H4K8 lactylation, thereby driving ESCA progression (9). Beyond its close association with cancer hallmarks and tumor progression, lactylation may also serve as a predictor of prognosis and therapeutic responsiveness (12,13). Nevertheless, studies focusing on lactylation-related genes (LRGs) in the ESCA microenvironment remain limited. In this study, we integrated single-cell RNA sequencing (scRNA-seq) and bulk RNA sequencing datasets to systematically investigate the role of LRGs in ESCA. By comparing differentially expressed genes (DEGs) between tumor and normal tissues, between patients with high vs. low lactylation scores, and between cells with differential lactylation activity, we identified a set of genes highly associated with ESCA progression and lactylation dynamics. Subsequent analyses enabled the identification of prognostic genes and revealed their functional relevance in pathway enrichment, immune regulation and drug sensitivity. These findings provide novel insights into the molecular mechanisms linking lactylation to ESCA and highlight potential targets for early intervention and therapeutic strategies. We present this article in accordance with the TRIPOD and MDAR reporting checklists (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0300/rc). Methods Data acquisition RNA expression profiles and corresponding clinical data for ESCA were obtained from The Cancer Genome Atlas (TCGA) via the UCSC Xena platform (https://xena.ucsc.edu/), comprising 183 tumor tissue samples and 13 normal tissue samples. The validation dataset, GSE53622 (platform: GPL18109), was downloaded from the Gene Expression Omnibus (GEO) database and included 60 tumor and 60 normal esophageal tissue samples. In addition, a scRNA-seq dataset (GSE188900), consisting of seven tumor samples and one distal normal sample, was retrieved from GEO. A total of 332 LRGs were collected based on previous research (13). The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. scRNA-seq data processing The scRNA-seq data were analyzed using the “Read10X” function from the “Seurat” package in R and processed according to the following criteria: (I) ≤10% mitochondrial genes; (II) ≤5% hemoglobin genes; (III) ≤30% ribosomal genes; and (IV) gene count per cell ranging from 200–7,000. All sample data were combined using the merge function and normalized using the “NormalizeData” function. A total of 3,000 highly variable genes were selected. Cell clustering was performed using the “FindNeighbors” and “FindClusters” functions, which construct a K-nearest neighbor (KNN) graph and apply community detection algorithms to identify transcriptionally distinct cell populations. Major cell types were identified based on marker genes as follows: B cells (BANK1, CD79A, MS4A1, CD79B, CD19), endothelial (CLDN5, VWF, FLT1, ICAM1), epithelial (EPCAM, SFN, PERP, KRT18, KRT19, KRT5, CDH1), fibroblasts (DCN, COL1A2, ACTA2, PDGFRA, LUM, COL1A1, THY1), macrophages (LYZ, CD68, FCGR3A), mast (MS4A2, TPSAB1, TPSB2, KIT, GATA2, CPA3), and NK/T cells (CD3D, CD3E, CD3G, CDBA, CCL5). Identification of ESCA-associated LRGs To identify LRGs associated with ESCA, a multi-step analysis was conducted. First, the “AUCell” package in R was used to quantify lactylation activity in scRNA-seq data based on 332 LRGs. Cells were divided into high- and low-lactylation groups according to the median AUCell score. DEGs between the two groups were identified using the “FindMarkers” function with adjusted P value <0.05 and |log2fold change (FC)| ≥0.5. Second, in the TCGA ESCA cohort, the “GSVA” package in R was applied to calculate lactylation scores based on the same 332 genes. Patients were then classified into high- and low-lactylation groups according to the median score. DEGs between the two groups were identified using the “limma” package in R with the same statistical thresholds (adjusted P value <0.05 and |log2FC| ≥0.5). Third, DEGs between tumor and normal tissue samples in the TCGA cohort were also identified using the “limma” package under the same criteria. Finally, the intersecting genes from DEGs between high- and low-lactylation cells in the scRNA-seq dataset, DEGs between high- and low-lactylation patients in the TCGA dataset, and DEGs between tumor and normal tissues in TCGA, were defined as LRGs. Functional enrichment analyses Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were performed using the DAVID database (https://david.ncifcrf.gov/) to assess the potential biological functions and pathways. A P value <0.05 was considered statistically significant. Identification of prognosis-related LRGs All LRGs were included as variables in the univariate Cox regression analysis to evaluate their associations with overall survival (OS) in ESCA patients. Genes with P<0.05 in the univariate analysis were selected as candidate covariates and subsequently included in the multivariable Cox regression model to identify independent prognostic genes. No additional data imputation was performed as the dataset contained no missing values. Kaplan-Meier survival analysis was performed using the “survival” and “survminer” packages in R to evaluate differences in OS between patients with high and low expression levels of prognosis-related LRGs. A P value <0.05 was considered statistically significant. Construction of the nomogram model A nomogram model was developed using the “rms” package in R based on the expression levels of prognosis-related LRGs in the TCGA cohort. Each variable was assigned a specific score, and the total score was used to estimate 1-, 2-, and 3-year OS probabilities for ESCA patients. Calibration curves were generated to assess the agreement between predicted probabilities and actual outcomes. Decision curve analysis (DCA) was performed using the “ggDCA” package in R to evaluate the clinical utility of the nomogram model. Immune cell infiltration analysis The single-sample gene set enrichment analysis (ssGSEA) algorithm was used to determine differences in the infiltration levels of 23 immune cells. Pearson correlation analysis was employed to determine the correlation between prognosis-related LRGs and differential immune cells. Furthermore, various immune gene sets were obtained from the TISIDB database (http://cis.hku.hk/TISIDB/index.php), including chemokine and receptor, major histocompatibility complex (MHC), immunoinhibitory, and immunostimulatory. The correlation between prognosis-related LRGs and immune genes was also calculated through Pearson correlation analysis. The correlation analysis between prognosis-related LRGs and ESCA-associated genes Genes associated with the occurrence of ESCA were retrieved from the GeneCards database (https://www.genecards.org/) using the keyword “Esophageal Carcinoma”. These genes were ranked based on relevance scores, and the top 20 genes were selected for differential expression analysis. Pearson correlation analysis was then performed to assess the relationship between prognosis-related LRGs and differentially expressed ESCA-associated genes. Gene set variation analysis (GSVA) and gene set enrichment analysis (GSEA) A series of Hallmark gene sets (h.all.v2024.1.Hs.symbols.gmt) were downloaded from the MSigDB database (https://www.gsea-msigdb.org/gsea/msigdb/). The “GSVA” package in R was employed to calculate enrichment scores for each gene set in order to evaluate potential biological functional differences across samples. Additionally, GSEA was performed using the package “clusterProfiler” in R. A P value <0.05 was considered statistically significant. Prediction of transcription factors Transcription factor prediction was performed using the “RcisTarget” package in R, and prognosis-related LRGs were input as the gene set. This analysis is based on motif enrichment. We used the precompiled motif ranking file hg19-tss-centered-10kb-7species.mc9nr.feather, which ranks genes genome-wide according to their predicted binding potential for each motif. For each motif, genes were ranked based on their binding score, and a recovery curve was generated for the input gene set. The area under the recovery curve (AUC) is then computed to assess motif enrichment. Next, a normalized enrichment score (NES) is calculated based on the AUC distribution across all motifs. Motifs with high NES values are identified as significantly enriched, and their associated transcription factors are considered as potential regulators. Drug sensitivity analysis The GDSC1 dataset (https://www.cancerrxgene.org/downloads/bulk_download) was downloaded from the Genomics of Drug Sensitivity in Cancer (GDSC) database to obtain drug sensitivity information. The half maximal inhibitory concentration (IC50) values for drugs were calculated using the “oncoPredict” package in R. Then, the correlation analysis was performed between the IC50s of drugs and prognosis-related LRGs. Immunohistochemistry (IHC) We collected formalin-fixed, paraffin-embedded (FFPE) tumor and adjacent normal tissue samples from 15 patients with ESCA. Each patient contributed three tumors and three adjacent normal tissues (total, n=90). Patients with newly diagnosed primary ESCA and no history of preoperative radiotherapy or chemotherapy were enrolled. Cases with recurrent disease or incomplete clinicopathological data were excluded. The clinical characteristics of these patients are summarized in Table S1. IHC was performed to validate the protein expression levels of the hub genes. Briefly, FFPE sections were incubated with the primary antibody (Epredia, TL-125-QPB) at 4 ℃ overnight. The sections were then incubated with the signal-enhancing reagent at 37 ℃ for 10 min, followed by incubation with the HMG4 Rabbit mAb (Nature, A27979), Anti-GADD45B Antibody (Boster, A03508), MAGP2 Rabbit mAb (Nature, A16529), and Lamin B2 Rabbit mAb (Nature, A21845) at 37 ℃ for another 10 min. The staining results were photographed under the microscope (Nikon, ECLIPSE Si RS). The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by the Ethics Committee of 3201 Hospital [20251-005]. Written informed consent was obtained from all patients. Virtual knockout Virtual knockout analysis was performed using the R package scTenifoldKnk based on scRNA-seq data to identify genes perturbed by target gene disruption. These genes (adjusted P<0.05) were subsequently subjected to functional enrichment analysis (P<0.05) to evaluate the potential biological effects of target gene perturbation in ESCA. Cell culture and transfection Human ESCC cell lines KYSE-140 (CL-0953) and KYSE-180 (CL-0760), along with their corresponding specialized culture media (CM-0953, CM-0760), were purchased from Procell (Wuhan, China). Cells were cultured at 37 ℃ in a humidified incubator containing 5% carbon dioxide (CO2) and 95% relative humidity. HMGB3 siRNA and the corresponding negative control sequences (Table S2) were synthesized by Hippo Biotechnology (Huzhou, China). Cell transfection was performed using 125 µL Opti-MEM® Medium (Gibco, 31985-062), 100pmoL siRNA, and 4 µL Lipo8000TM transfection reagent (Beyotime, C0533) for 6 h. After 48 h of culture, the knockdown efficiency was evaluated by real-time quantitative polymerase chain reaction (qPCR). Real-time qPCR Total RNA was extracted from cells using TRIzol reagent (TIANGEN, DP424), and RNA purity was assessed using a NanoVue plus spectrophotometer (Biochrom, 28956057), followed by reverse transcription using the FastQuant cDNA First Strand Synthesis Kit. Subsequently, qPCR was performed according to the manufacturer’s instructions for the SuperReal PreMix Plus Kit (TIANGEN, FP205). The relative expression level of HMGB3 was calculated using the 2−ΔΔCt method, with GAPDH serving as the internal control. Primer sequences (Table S2) were synthesized by Biomed Gene Technology (Beijing, China). Cell viability assay Cell viability was assessed using the Cell Counting Kit-8 (CCK-8) assay. Cells in the logarithmic growth phase were transfected with siRNA-HMGB3 or siRNA-NC. Subsequently, 10 µL CCK-8 solution (Apexbio, K1018) was added to each well and incubated for 4 h. The absorbance was measured at 450 nm using a microplate reader. Terminal deoxynucleotidyl transferase-mediated dUTP nick end labeling (TUNEL) assay Harvested cells were washed with phosphate-buffered saline (PBS) and fixed with 4% paraformaldehyde (Beyotime, P0099) for 30 min. After washing with PBS, cells were permeabilized with 0.3% Triton X-100 in PBS for 5 min. Subsequently, 50 µL of TUNEL reaction mixture prepared according to the manufacturer’s instructions (Beyotime, C1086) was added for immunofluorescence staining, followed by incubation at 37 ℃ for 60 min in the dark. Nuclei were counterstained with DAPI (Solarbio, C0065). Cells were thoroughly washed with PBS between each step. Finally, cell slides mounted with anti-fade mounting medium (Boster, AR0036) were observed under a fluorescence microscope, and green fluorescence was evaluated to assess cell apoptosis. Transwell assay Cell migration and invasion abilities were evaluated using Transwell assays. For the migration assay, Transwell inserts were directly placed into 24-well plates, and 100 µL serum-free medium was added to the upper chamber and incubated at 37 ℃ for 1 h. For the invasion assay, the upper chamber was precoated with Matrigel matrix (Yeasen, 40183ES08), and 500 µL serum-free medium was added to both the upper and lower chambers, followed by incubation at 37 ℃ for 2 h. Subsequently, the medium in the upper chamber was discarded, and cell suspension (1×105 cells/well) was added to the upper chamber, while 500 µL medium containing 10% FBS was added to the lower chamber as a chemoattractant. After incubation at 37 ℃ to allow cell migration or invasion, non-migrated cells on the upper surface were carefully removed. The inserts were washed three times with PBS and fixed with 4% paraformaldehyde for 30 min. The inserts were then air-dried, stained with 500 µL 0.1% crystal violet (Sangon, CB0331) for 30 min, and washed with PBS. Finally, the number of migrated or invaded cells was counted under a microscope in randomly selected fields. Statistical analysis Bioinformatics analyses were performed using R software, while experimental data were analyzed using GraphPad Prism version 10 (Boston, USA). Differences between two groups were analyzed using a two-tailed Student’s t-test. Comparisons among multiple groups were performed using one-way analysis of variance (ANOVA) followed by Tukey’s post hoc test. A P value <0.05 was considered statistically significant. Results Identification of LRGs in ESCA To investigate gene expression patterns between tumor and normal tissues in the TCGA-ESCA dataset, we identified 4,680 DEGs through differential expression analysis (Figure 1A). GO analysis revealed that the upregulated DEGs were significantly enriched in cell cycle regulation and canonical glycolysis, while the downregulated DEGs were enriched in response to hypoxia (Figure 1B). KEGG analysis showed that upregulated DEGs were involved in the cell cycle and pentose phosphate pathway, whereas downregulated DEGs were enriched in metabolic pathways and carbon metabolism (Figure 1C). These findings suggest that significant metabolic reprogramming may occur in ESCA, and lactylation may play a role in its pathogenesis. Identification of LRG signature in ESCA. (A) Volcano plot illustrating DEGs between tumor and normal tissues in the TCGA ESCA dataset. (B) Lollipop plot showing the enriched GO-BP terms of DEGs identified in (A). (C) Lollipop plot showing the enriched KEGG pathways of DEGs identified in (A). (D) Volcano plot illustrating DEGs between high- and low-lactylation groups in the TCGA ESCA dataset. (E) Lollipop plot showing the enriched GO-BP terms of DEGs identified in (D). (F) Lollipop plot showing the enriched KEGG pathways of DEGs identified in (D). (G) UMAP plot illustrating the comprehensive annotation of ESCA samples into seven distinct cell types. (H) UMAP plot displaying lactylation activity in each cell, scored by AUCell; red indicates high lactylation, and blue indicates low lactylation. (I) Volcano plot illustrating DEGs between high- and low-lactylation cells in the scRNA-seq dataset. (J) Venn diagram showing the intersection of DEGs identified from (A), (D), and (I). DEG, differentially expressed gene; ESCA, esophageal cancer; GO-BP, Gene Ontology-Biological Process; KEGG, Kyoto Encyclopedia of Genes and Genomes; LRG, lactylation-related gene; scRNA-seq, single-cell RNA sequencing; TCGA, The Cancer Genome Atlas; UMAP, Uniform Manifold Approximation and Projection. We next calculated lactylation scores for ESCA patients in the TCGA dataset based on the expression of 332 LRGs, and stratified patients into high-lactylation and low-lactylation groups. A total of 1,672 DEGs were identified between the two groups (Figure 1D). To explore the biological processes associated with lactylation, we performed GO and KEGG analyses (Figure 1E,1F). The DEGs were significantly enriched in pathways related to the cell cycle, DNA replication, immune response, and cell adhesion molecules, suggesting that lactylation may affect ESCA progression by promoting tumor-associated pathways and immune evasion. Subsequently, we analyzed scRNA-seq data from ESCA tumor samples and annotated cells into seven major cell types using Uniform Manifold Approximation and Projection (UMAP) (Figure 1G). We computed lactylation scores at the single-cell level using AUCell, and stratified cells into high- and low-lactylation activity groups, identifying 3,908 DEGs between them (Figure 1H,1I). By intersecting the DEGs from three comparisons, tumor vs. normal tissues (4,680 DEGs), high vs. low lactylation score groups in TCGA (1,672 DEGs), and high vs. low lactylation cells in scRNA-seq (3,908 DEGs), we identified 321 LRGs (Figure 1J, Table S3). These LRGs may be closely associated with lactylation activity and ESCA development. Identification of prognosis-related LRGs in ESCA Based on univariate Cox regression analysis, we identified 7 LRGs significantly associated with ESCA prognosis from 321 LRGs (Figure 2A). Multivariate Cox regression analysis further revealed 4 independent prognostic genes for ESCA, including two risk genes (HMGB3 and GADD45B) and two protective genes (LMNB2 and MFAP5) (Figure 2B). Using the optimal cutoff value for each gene’s expression, patients were stratified into high- and low-expression groups to assess the impact of these genes on prognosis (Figure 2C-2F). Kaplan-Meier survival analysis showed that higher expression levels of HMGB3 and GADD45B were significantly associated with poorer OS, whereas higher expression levels of LMNB2 and MFAP5 were associated with better OS. Collectively, these four LRGs may serve as potential prognostic biomarkers for ESCA. Identification of prognosis-related LRG signature in ESCA. (A) Forest plot illustrating 7 out of 321 LRGs significantly associated with ESCA prognosis using univariate Cox regression analysis. (B) Forest plot showing 4 out of the 7 LRGs as independent prognostic factors for ESCA identified by multivariate Cox regression analysis. (C-F) Kaplan-Meier survival curves for each prognostic LRG, plotted according to the optimal cutoff values of gene expression. (G) A nomogram integrating the four prognostic LRGs to predict 1-, 2-, and 3-year OS probabilities in ESCA. **, P<0.01; ***, P<0.001. (H) Calibration curves evaluating the agreement between nomogram-predicted survival and actual observed survival at 1, 2, and 3 years. (I) DCA assessing the clinical utility of the nomogram. DCA, decision curve analysis; ESCA, esophageal cancer; HR, hazard ratio; LRG, lactylation-related gene; OS, overall survival. Based on the four prognosis-related LRGs identified in the TCGA cohort, we constructed a nomogram to predict individual OS at 1, 2 and 3 years (Figure 2G). The nomogram demonstrated that higher total scores were associated with lower predicted survival probabilities. The calibration curves showed that the predicted OS was broadly consistent with the actual observations (Figure 2H). Furthermore, DCA indicated that the nomogram had favorable clinical utility (Figure 2I). Analysis of LRGs and clinical factors in ESCA We analyzed the expression of four LRGs across different pathological stages of ESCA (pathologic T, N, M, and overall pathological stage) using TCGA data (Figure 3A-3D). The results showed that GADD45B and HMGB3 expression levels were significantly upregulated in advanced stages. MFAP5 expression progressively increased with higher T stages (T1–T4), but was reduced in M1 (vs. M0). Analysis of scRNA-seq data revealed distinct cell type-specific expression patterns of the genes (Figure S1). Specifically, LMNB2 was primarily expressed in epithelial cells, MFAP5 was enriched in fibroblasts, and HMGB3 was prominently expressed in both epithelial cells and fibroblasts. GADD45B showed broad expression across multiple cell types. Expression patterns of the four LRGs across different clinical parameters in ESCA. (A) Expression levels in different overall clinical stages (I–IV). (B) Expression levels according to primary tumor size/extent (T stage). (C) Expression levels according to lymph node involvement (N stage). (D) Expression levels according to the presence of distant metastasis (M stage). ESCA, esophageal cancer; LRG, lactylation-related gene; M, metastasis; N, node; T, tumor. Relationship between LRGs and ESCA-related genes/pathways The top 20 genes associated with ESCA were identified from the GeneCards database based on relevance scores. Among them, 12 genes were differentially expressed between tumor and normal tissue samples in the TCGA ESCA dataset (Figure 4A). Correlation analysis revealed that four LRGs were significantly associated with at least nine ESCA-related genes, among which HMGB3 and CHEK2 exhibited the strongest correlation (cor =0.71, P<2.2E-16) (Figure 4B). Subsequently, we investigated the signaling pathways potentially influenced by these four LRGs (Figure 4C-4F). GSVA analysis of high- vs. low-expression groups indicated that inflammatory response-related pathways (e.g., IL2/STAT5 signaling, inflammatory response, IL6/JAK/STAT3 signaling) were activated in the high-expression groups of GADD45B and MFAP5, but suppressed in the high-expression groups of HMGB3 and LMNB2. Carcinogenesis-related pathways (e.g., epithelial-mesenchymal transition, angiogenesis, KRAS signaling) were activated in the high-expression groups of GADD45B, MFAP5, and LMNB2. Mitochondrial metabolism-related pathways (e.g., reactive oxygen species pathway) were enriched in the high-expression groups of HMGB3 and LMNB2, as well as the low-expression groups of GADD45B and MFAP5. Notably, these pathways were all upregulated in tumor tissues compared with normal tissues (Figure S2A,S2B). In addition, glycolysis was enriched in the high-expression groups of HMGB3 and LMNB2, and in the low-expression groups of GADD45B and MFAP5, further suggesting their potential roles in lactylation regulation (Figure 4C-4F). GSEA yielded consistent results (Figure S2C). Together, these findings suggest that LRGs are significantly associated with ESCA-related genes and pathways, potentially contributing to tumorigenesis or modulating tumor immune responses. Association between LRGs and ESCA progression. (A) Box plot of ESCA-related genes obtained from the GeneCards database. **, P<0.01; ***, P<0.001; ****, P<0.0001; ns, not significant. (B) Correlation plots between the four prognostic LRGs and differentially expressed ESCA-related genes. (C-F) GSVA analysis of high- and low-expression subgroups of the four prognostic LRGs. Red indicates enrichment in the high-expression group, whereas blue indicates enrichment in the low-expression group. ESCA, esophageal cancer; GSVA, gene set variation analysis; LRG, lactylation-related gene. Association between LRGs and the immune microenvironment Considering the potential association between LRGs and immune pathways in ESCA, we further investigated the relationship between LRGs and the immune microenvironment. Using the ssGSEA algorithm, we found that the infiltration levels of 11 immune cell types significantly differed between tumor and normal tissues, suggesting immune dysregulation in ESCA (Figure S3). Notably, activated B cells and natural killer T (NKT) cells were significantly correlated with all four LRGs (Figure 5A). Subsequently, we analyzed immune-related gene sets and found significant correlations between the four LRGs and genes from the immunoinhibitor, immunostimulatory, chemokine, receptor, and MHC categories (Figure 5B-5F). Specifically, in the immunoinhibitor category, CSF1R, BTLA, and TGFB1 showed significant correlations with all four LRGs (Figure 5B). In the immunostimulatory category, ENTPD1, CXCL12, CD48, and TNFSF15 were similarly associated (Figure 5C). For chemokines, CCL23, CCL19, CCL16, CCL14, CXCL12, CCL28, and CXCL1 were significantly correlated (Figure 5D). In the receptor group, CCR2 was significantly associated with all four LRGs (Figure 5E). Lastly, in the MHC category, TAP1 exhibited a significant correlation with all four LRGs (Figure 5F). These findings suggest that LRGs may be broadly involved in immune regulation in ESCA, potentially influencing both immune cell infiltration and immune gene expression. Immune relevance of the four prognostic LRGs. (A) Heatmap showing the correlations between the four LRGs and differentially infiltrated immune cells. (B-F) Heatmap showing the correlations between the four LRGs and immunoregulation genes. ×/*, P<0.05; ××/**, P<0.01; ×××/***, P<0.001. LRG, lactylation-related gene. Transcriptional regulation of LRGs in ESCA To explore the transcriptional regulatory mechanisms of LRGs in ESCA, we identified a total of 77 transcription factors that may potentially regulate the four LRGs, among which 42 TFs were significantly differentially expressed between tumor and normal tissues (Figure 6A). Since transcription factors exert their regulatory effects by binding to specific DNA motifs, we conducted motif enrichment analysis to elucidate their potential binding preferences (Figure 6B). The results revealed that the most significantly enriched motif was transfac_pro__M09565 (NES =9.55), followed by cisbp__M1332 (NES =9.46) and transfac_pro__M09570 (NES =8.77). These enriched motifs were mainly associated with the regulation of HMGB3, LMNB2, and GADD45B, suggesting potential upstream regulatory pathways influencing LRG expression in ESCA. Prediction of transcriptional regulators of the four LRGs. (A) Regulatory network of transcription factors and the LRGs. Red triangles represent LRGs; purple rectangles denote transcription factors; transcription factors with black borders indicate those differentially expressed between tumor and normal tissues. (B) Recovery curves of the top three significantly enriched motifs. Green line: observed cumulative number of recovered target genes of the transcription factor; Red line: Expected cumulative number of recovered target genes under random permutation (null background); Blue step curve: statistical significance threshold; the dot and vertical dashed line mark the optimal cutoff rank for leading-edge genes. LRG, lactylation-related gene; SD, standard deviation. Drug sensitive analysis We further analyzed the potential association between the expression levels of the four LRGs and drug sensitivity using IC50 data from six drugs (5-fluorouracil, gefitinib, paclitaxel, sorafenib, vincristine, and cisplatin) obtained from the GDSC database (Figure 7). Patients with low expression of GADD45B and MFAP5 showed increased sensitivity to 5-fluorouracil, gefitinib, paclitaxel, sorafenib, and vincristine, indicating a potentially favorable therapeutic response. Patients with high HMGB3 expression were more likely to be sensitive to Gefitinib and Cisplatin, while high LMNB2 expression was associated with increased sensitivity to gefitinib, paclitaxel, vincristine, and cisplatin. These findings suggest that LRG expression profiles could aid in stratifying ESCA patients and optimizing individualized treatment strategies. Drug sensitivity analysis. Scatter plots showing the correlations between the expression levels of the four LRGs and the IC50 values of commonly used antitumor drugs. IC50, half maximal inhibitory concentration; LRG, lactylation-related gene. Expression verification of LRGs In the TCGA dataset, the expression levels of GADD45B and MFAP5 were significantly decreased, whereas HGMB3 and LMNB2 were significantly upregulated (Figure 8A). Consistently, in the GSE53622 dataset, the expression patterns of GADD45B, HGMB3, and LMNB2 were in agreement with those observed in the TCGA cohort (Figure 8B). We analyzed the protein expression levels of these genes in ESCA tissues and adjacent normal tissues using IHC (Figure 8C). The results were consistent with the RNA-level data, further suggesting a potential association of these genes with ESCA. Validation of LRG expression. (A) Box plots displaying the expression levels of the four LRGs in the TCGA-ESCA dataset. (B) Box plots displaying the expression levels of the four LRGs in the GSE53622 dataset. (C) Representative IHC images of protein expression for four LRGs. Scale bar: 10 μm. *, P<0.05; ****, P<0.0001; ns, no significant. ESCA, esophageal cancer; IHC, immunohistochemistry; LRG, lactylation-related gene; TCGA, The Cancer Genome Atlas. Virtual knockout reveals potential biological functions of LRGs Using scTenifoldKnk, we identified genes potentially perturbed by the four LRGs in scRNA-seq data, followed by GO and KEGG enrichment analyses (Figure 9 and Figure S4). GO biological process analysis showed that MFAP5 and GADD45B were mainly enriched in extracellular matrix (ECM) organization, collagen fibril formation, cell-matrix adhesion, and TGF-β/BMP signaling pathways. In contrast, HMGB3 and LMNB2 were significantly associated with cell cycle-related processes, including mitotic nuclear division, as well as epithelial differentiation processes such as epidermal development. KEGG pathway analysis further indicated that MFAP5 and GADD45B were enriched in ECM-receptor interaction and PI3K-Akt signaling pathways, whereas HMGB3 and LMNB2 were both enriched in pathways related to cornified envelope formation. Collectively, these results suggested that MFAP5 and GADD45B are primarily involved in ECM remodeling and tumor-associated signaling pathways, while HMGB3 and LMNB2 may contribute to epithelial differentiation and cell cycle regulation. Virtual knockout reveals potential biological functions of LRGs. Results of GO-BP analysis of the disrupted genes following sequential knockout of GADD45B, LMNB2, MFAP5, and HMGB3. GO-BP, Gene Ontology-Biological Process; LRG, lactylation-related gene. HMGB3 knockdown suppressed the malignant phenotype of KYSE-140 cell line We further verified the functional role of HMGB3 in ESCA cell lines. Compared with the well-differentiated KYSE-180 cells, HMGB3 expression was markedly higher in the moderately differentiated KYSE-140 cells (Figure 10A). Three HMGB3-targeting siRNAs were constructed, all of which significantly reduced HMGB3 expression (Figure 10B). Among them, siRNA-3 showed the highest silencing efficiency and was selected for subsequent experiments. Compared with the siRNA-NC group, siRNA-HMGB3 significantly suppressed the viability of KYSE-140 cells at both 24 and 48 h (Figure 10C). In addition, HMGB3 knockdown markedly reduced the migratory and invasive capacities of KYSE-140 cells (Figure 10D,10E). TUNEL staining further revealed a higher proportion of apoptotic cells in the siRNA-HMGB3 group (Figure 10F). Overall, these findings showed that inhibition of HMGB3 expression suppressed the malignant phenotype of the KYSE-140 cells. The functional role of HMGB3 in KYSE-140 cells. (A) RT-qPCR was performed to evaluate the mRNA expression levels of HMGB3 in KYSE-140 and KYSE-180 cells. (B) The knockdown efficiency of three HMGB3 siRNAs was validated by RT-qPCR. (C) Cell viability was assessed at 0, 24, and 48 h using the CCK-8 assay to compare the vector control and siRNA-HMGB3 groups. (D) Representative images of cell migration and invasion assessed using the Transwell assay. (E) Quantitative analysis of migrated and invaded cell numbers. (F) TUNEL fluorescence staining was performed to evaluate cell apoptosis. 400× magnification. Scale bar: 10μm. Data are presented as mean ± SD from three biological replicates. *, P<0.05; **, P<0.01; ****, P<0.0001. CCK-8, cell counting kit-8; DAPI, 4,6-diamidino-2-phenylindole; mRNA, messenger RNA; RT-qPCR, reverse transcription quantitative polymerase chain reaction; SD, standard deviation; siRNA, small interfering RNA; TUNEL, terminal deoxynucleotidyl transferase dUTP nick end labeling. Discussion Lactylation is a post-translational modification of histones or other proteins at lysine residues, closely linked to tumor progression and predictive of patient prognosis. In this study, we identified four LRGs (GADD45B, HMGB3, LMNB2, and MFAP5) that were significantly associated with ESCA occurrence and lactylation activity. Both univariate and multivariate Cox regression analyses confirmed these LRGs as independent prognostic factors in ESCA. Previous research showed that GADD45B (growth arrest and DNA damage-inducible beta) participates in DNA damage response and cell cycle arrest (14). It enhances MAP2K3 phosphorylation, thereby activating p38 and p53 signaling to inhibit ESCA tumor growth, suggesting its tumor-suppressive role (15). Intriguingly, although GADD45B expression was significantly downregulated in ESCA, its low expression was paradoxically associated with worse prognosis, a result also observed in gastric cancer (16). Stage-specific analysis in our study further revealed significant upregulation of GADD45B in advanced ESCA. Therefore, we speculate that GADD45B may exert a dual role in ESCA initiation and progression. During tumor evolution, the accumulation of DNA damage can impair the function of GADD45B and create feedback that promotes overexpression, leading to tumor progression (17). LMNB2 (Lamin B2), a nuclear lamins family member, maintains nuclear envelope and nucleolar integrity (18). Aberrant LMNB2 expression has been implicated in tumorigenesis, affecting chromosomal stability, DNA repair, and epithelial-mesenchymal transition (19). MFAP5 (microfibrillar-associated protein 5) is involved in ECM remodeling and is typically enriched in cancer-associated fibroblasts (CAF) (20). MFAP5 has been reported to suppress migration and invasion by blocking the AKT/mTOR pathway in endometrial cancer (21). On the other hand, it may promote cell motility and invasiveness in certain cancers, such as ovarian cancer (22). HMGB3 (high mobility group box 3), an X-linked member of the HMG protein superfamily, has been linked to multiple malignancies, including lung and ovarian cancers (23,24). In ESCA, HMGB3 knockdown inhibits proliferation and metastasis (25). Notably, we observed significant correlations between these LRGs and key ESCA-associated genes, particularly between HMGB3 and CHEK2, a DNA damage sensor gene implicated in ESCA susceptibility (26). Through virtual knockout analysis, we found that MFAP5 and GADD45B are primarily involved in ECM remodeling and tumor-related signaling pathways, whereas HMGB3 and LMNB2 may contribute to ESCA development by coordinating epithelial differentiation and cell cycle progression—processes that are critical for tumor initiation and progression (27). Collectively, these findings highlight indispensable roles of the four LRGs in ESCA occurrence and progression, thereby providing a biological basis for their prognostic relevance. Lactylation includes both histone and non-histone modifications. Histone lactylation is primarily enriched at promoter and enhancer regions, where it remodels chromatin structure and activates downstream gene transcription (28). Non-histone lactylation exerts more diverse regulatory effects, including modulation of protein stability, activity, and subcellular localization, thereby influencing protein function (28). Through these mechanisms, lactylation has been reported to affect key molecules and signaling pathways involved in immunity, metabolism, and DNA damage repair, ultimately shaping the tumor microenvironment (29). Notably, both forms of lactylation have been identified in ESCA (9,30). These lactylation sites identified in ESCA act as glycolytic regulators, influencing tumorigenesis and stemness (30). Consistently, GADD45B downregulation impairs mitochondrial respiration while enhancing glycolysis in bladder cancer cells (31), whereas HMGB3 overexpression promotes glycolysis in nasopharyngeal carcinoma (32). Similarly, MFAP5 suppresses papillary thyroid carcinoma progression by inhibiting aerobic glycolysis and lactate production through the Wnt/β-catenin pathway (33). MFAP5+ CAF have been reported to promote extramural venous invasion in gastric cancer, potentially through the regulation of lactylation (34). Consistent with these findings, our single-cell analysis revealed that MFAP5 was predominantly enriched in fibroblasts. Our analyses demonstrated a significant enrichment of glycolysis and metabolic pathways in groups with high vs. low expression of the four LRGs. Therefore, their potential mechanistic link with lactylation has potential reliability. We further investigated the immune implications of these LRGs in ESCA. Their expression was significantly associated with the infiltration levels of activated B cells and NKT cells. NKT-like cells have been shown to exhibit enhanced glycolysis and lactylation levels in malignant pleural effusions, maintaining immune evasion capacity (35). Additionally, the four LRGs showed significant correlations with various immune modulatory genes, such as ENTPD1 and CCL2/CCR2. Previous research has shown that H3K18 lactylation can upregulate CCR8, thereby promoting the recruitment of regulatory T cells (Tregs) into the tumor microenvironment, as well as ENTPD1, which encodes CD39 and contributes to the generation of immunosuppressive adenosine (7). In addition, non-histone ENSA K63 lactylation has been reported to activate the CCL2/CCR2 signaling pathway, leading to the recruitment of tumor-associated macrophages and suppression of cytotoxic T-cell activity (36). These immune modulatory genes may affect the response to immunotherapy in ESCA (37,38). In addition, these four LRGs exhibit potential roles in immune and inflammatory regulation. For instance, GADD45B can modulate innate immune responses of granulocytes and macrophages through differential regulation of the p38 and JNK signaling pathways (39). Moreover, exosomal HMGB3 has been shown to promote monocyte/macrophage recruitment and inflammatory responses by activating the STAT3/MAPK/NF-κB/CCR2 signaling axis (40). According to the “metabolism-epigenetics-immunity” framework proposed by Wang et al. (41), we hypothesize that these LRGs might partly influence the immunosuppressive tumor microenvironment through lactylation-related mechanisms, potentially promoting immune evasion and progression in ESCA. Nevertheless, further mechanistic studies are warranted. We observed that the expression of these four LRGs was associated with sensitivity to six commonly used anticancer drugs (5-fluorouracil, gefitinib, paclitaxel, sorafenib, vincristine, and cisplatin). Previous studies have demonstrated that targeting lactylation can enhance cancer cell sensitivity to chemotherapeutics such as 5-fluorouracil and cisplatin (42,43). Furthermore, accumulating evidence highlights the intrinsic link between the four LRGs and tumor sensitivity to chemotherapy. HMGB3 upregulation has been associated with cisplatin resistance in tumor cells (44), while LMNB2 deficiency in cisplatin-resistant ESCA cells may partially restore sensitivity to cisplatin (45). In addition, blocking MFAP5 secretion in CAFs has been shown to inhibit fibrosis and enhance the bioavailability of paclitaxel (46). Notably, GADD45B expression is increased in ESCC patients with non-complete response to neoadjuvant chemotherapy (47). Given that chemo-immunotherapy remains the first-line treatment for advanced or metastatic ESCA, our findings highlight the importance of integrating LRG expression, immune modulation, and drug sensitivity into personalized treatment strategies to optimize therapeutic efficacy. Taken together, although GADD45B, HMGB3, LMNB2, and MFAP5 differ in their cellular localization and primary biological functions, their dysregulated expression may converge on common pathological processes. Specifically, these genes may promote glycolysis-associated lactate accumulation and modulate immune-inflammatory responses through distinct molecular pathways. This may collectively contribute to the establishment of an immunosuppressive tumor microenvironment and reduced sensitivity to chemotherapy. Notably, such a “high lactate-immunosuppression-drug resistance” axis has been widely associated with poor prognosis in multiple cancers (48,49). This integrated framework may provide a plausible biological explanation for the prognostic value of these LRGs in ESCA. Conclusions In summary, this study provides a novel perspective by analyzing lactylation activity scores with multi-group cross-analysis, and we identified LRGs with significant differential expression. These genes were found to be closely related to ESCA-associated genes, inflammatory responses, and tumorigenic pathways, suggesting their potential involvement in the pathological process of ESCA and their value as prognostic factors. Among them, HMGB3 knockdown suppressed the malignant phenotype of KYSE-140 cell line. Moreover, the four prognostic LRGs showed correlations with multiple immune cells and immunoregulatory genes, as well as distinct patterns of drug sensitivity, highlighting their potential utility in prognostic stratification and as candidates for personalized therapeutic strategies. However, considering that the study was primarily based on bioinformatics analyses, the conclusions should be regarded as preliminary. Further preclinical studies are required to elucidate the mechanistic roles of these LRGs in ESCA and to clarify their impact on the tumor immune microenvironment. In addition, their prognostic utility should be validated in large-scale clinical cohorts to ensure accuracy and reliability. Supplementary The article’s supplementary files as Acknowledgments None. 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. This study was approved by the Ethics Committee of 3201 Hospital [20251-005]. Written informed consent was obtained from all patients. Footnotes Reporting Checklist: The authors have completed the TRIPOD and MDAR reporting checklists. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0300/rc Funding: This work was supported by the General Medical Research Foundation Program (No. TYYLKYJJ-2022-002). Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0300/coif). The authors have no conflicts of interest to declare. Data Sharing Statement Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0300/dss References Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0300/prf Supplementary Materials The article’s supplementary files as Data Availability Statement Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0300/dss