A histone acetylation score for prognostic stratification and immunogenomic profiling in esophageal cancer

Abstract

Aim

To investigate the prognostic significance of histone acetylation (HAc) regulators in esophageal cancer (EC) and develop a transcriptome-based HAc_score reflecting epigenetic and immunogenomic states.

Methods

Expression and mutation from EC were analyzed to identify prognostic HAc regulators via univariable Cox models. Consensus clustering defined HAc-related expression patterns. Differentially expressed genes (DEGs) among clusters were functionally enriched. A principal component-based HAc_score was constructed from prognostic DEGs and tested for associations with overall survival, tumor mutational burden (TMB), immunophenoscore (IPS), and immune cell infiltration.

Results

Three HAc-related expression patterns showed distinct biological and immune features. From shared DEGs, 19 prognostic genes defined two molecular subtypes and served as the basis for the HAc_score. Higher HAc_score was associated with better overall survival, particularly in early-stage disease. HAc_score correlated inversely with TMB and positively with IPS components, suggesting a transcriptionally active, immunogenic phenotype despite lower mutation burden. Combining HAc_score with TMB improved risk stratification.

Conclusion

HAc_score quantifies HAc–linked transcriptional states in EC and reflects tumor–immune interactions. It stratifies survival risk and complements TMB, supporting its potential use as a prognostic biomarker and integrative epigenetic–immune signature.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12672-025-04258-5.

Introduction

Esophageal cancer (EC) is among the most lethal malignancies worldwide, ranking eighth in incidence and sixth in cancer-related mortality [1]. Despite advances in surgery, radiotherapy, and chemotherapy, outcomes remain poor; the 5-year survival is 20–30% overall [2]. Even in early-stages disease, postoperative recurrence and metastasis are common, the 5-year survival only 30–50% [3]. These observations underscore the need for robust prognostic biomarkers to refine risk stratification and guide therapy in EC.

Histone acetylation (HAc) is a central epigenetic mechanism governing chromatin remodeling and gene regulation [4, 5]. As a reversible post-translational modification, HAc modulates chromatin compaction and the accessibility of the transcriptional machinery, thereby controlling gene expression at specific genomic loci. Proper HAc is critical for normal cellular function, whereas dysregulated HAc can lead to aberrant gene expression and malignant transformation [6–9]. Emerging evidence implicates abnormal HAc in the progression of many cancers, influencing hallmarks of cancer such as uncontrolled proliferation, apoptosis resistance, invasion and metastasis, stem-like phenotypes, and therapeutic resistance [9–12]. These observations highlight the potential importance of HAc-related pathways in EC pathobiology.

Dysregulated HAc also influences tumor–immune crosstalk [13, 14]. Epigenetic modifications such as HAc can facilitate immune evasion by repressing tumor-associated antigens and immune-responsive genes [14, 15]. Dysregulation of HAc writers, erasers, and readers may favor an immunosuppressive tumor microenvironment (TME) [16]. Restoring acetylation balance—for example with histone deacetylase (HDAC) inhibition—can upregulate immunostimulatory molecules on cancer cells and enhance antitumor immune responses [17]. For example, in particular, HDAC inhibitors have been shown to increase MHC expression, costimulatory proteins, and chemokine signals in tumor cells, thereby improving T-cell recognition and function [18]. Thus, HAc perturbation may contribute both to cell-intrinsic aggressiveness and to immune remodeling through reprogramming of immune-related transcription.

Although HAc regulators have been implicated as oncogenic drivers or prognostic markers in multiple malignancies, their roles in EC remain incompletely defined [15, 19, 20]. Prior reports link aberrant HAc to adverse outcomes and immune phenotypes; for example, HDAC overexpression associates with aggressive behavior and poor prognosis [21–23], and HAc-related alterations have been connected to immune evasion pathways within the TME [22, 24]. However, systematic evaluation of HAc-regulating gene programs as risk stratifiers and as correlates of host immunity in EC is limited, representing a clear knowledge gap.

Characterizing the TME in EC is increasingly important, particularly with the adoption of immunotherapy [25]. The composition and functional state of infiltrating immune cells influence progression, metastasis, and treatment response [26]. Immune checkpoint blockade targeting PD-1/PD-L1 has shown clinical activity and improved outcomes in a subset of patients with EC [27]. For example, anti-PD-1 therapy has demonstrated survival benefits in advanced EC, heralding a new therapeutic avenue [27]. Yet durable benefit is limited to a minority—often those with “hot” tumors showing pre-existing T-cell infiltration—while many tumors remain “cold” or resistant, and PD-L1 alone is insufficient as a predictive biomarker [28]. Given this heterogeneity, composite biomarkers that integrate tumor-intrinsic biology with immune context are needed. In particular, coupling epigenetic signatures with immune profiling may better identify patients likely to benefit from immunotherapy and clarify how epigenetic states intersect with antitumor immunity.

Here, we systematically profiled HAc-related transcriptomic and somatic mutational landscapes in EC, derived a patient-level HAc-related score (HAc_score), and assessed its prognostic and immunogenomic relevance. Using The Cancer Genome Atlas esophageal carcinoma (TCGA-ESCA) cohort, we cataloged key HAc regulators, defined three HAc-related expression patterns, and constructed HAc_score from differentially expressed genes across these patterns to quantify HAc-linked transcriptional states at the individual level. We then evaluated the prognostic performance of HAc_score and its associations with immunogenomic features, including tumor mutational burden (TMB), immunophenoscore (IPS), and immune-cell infiltration. Our results indicate that HAc_score captures clinically relevant molecular heterogeneity and immune phenotypes, supporting its potential utility for risk stratification and as a framework for integrating epigenetic and immune profiling in EC.

Materials and methods

Data sources

We analyzed de-identified, publicly accessible datasets; therefore, additional institutional ethics approval was not required. Data use complied with the The Cancer Genome Atlas (TCGA) publication guidelines and the terms of the respective resources. HAc regulators were curated from the Gene Ontology (http://geneontology.org/page/go-database). We downloaded RNA-seq expression profiles, somatic mutation calls, copy number variation (CNV) profiles, and clinicopathologic annotations (overall survival, stage, sex, age) from the Genomic Data Commons (GDC; TCGA-ESCA project) via TCGAbiolinks.

Identification of prognostic HAc regulators

For each regulator, we fitted a univariable Cox proportional-hazards model with OS as the endpoint, using standardized gene expression (z-score) as a continuous predictor; hazard ratios (HRs) and 95% confidence intervals (CIs) were reported. Regulators were screened for downstream analyses using a prespecified nominal threshold (P < 0.20), yielding 23 genes hereafter referred to as the prognostic HAc regulators; their HRs/CIs and P-values are listed in Supplementary Table 1.

Mutational landscape of prognostic HAc regulators in EC

Somatic variant files from the enrolled cohort were processed with maftools to summarize variant classes, per-gene mutation frequencies, and to generate oncoplots. Gene-level copy-number alterations were derived from the prepared CNV matrix and summarized as frequencies of gain/loss across samples; genomic loci of the 23 regulators were mapped and displayed on ideograms using “RCircos” with the provided gene coordinates. To assess coordination among regulators, we computed pairwise Spearman correlations of regulator mRNA expression across tumors and fitted univariate Cox proportional-hazards models for overall survival for each regulator. The resulting correlation matrix and Cox results were integrated to construct a regulator network, which was assembled and visualized with “igraph”.

Consensus clustering analysis

We applied unsupervised consensus clustering with the R package ConsensusClusterPlus to cluster EC tumors based on the expression of the 23 prognostic HAc regulators. Candidate solutions for k = 2–6 were evaluated, and the optimal k was selected by inspecting the cumulative distribution function (CDF) and ΔCDF curves in conjunction with consensus heatmaps and cluster-consensus metrics. The resulting cluster assignments were used for downstream survival, pathway, and immune-infiltration analyses.

Gene set variation analysis (GSVA)

We used GSVA (R/GSVA package) to compute sample-level enrichment scores for KEGG canonical pathways from MSigDB (collection c2.cp.kegg, human gene symbols). Pathway activity scores were then compared across HAc-defined groups (i.e., among HAc_clusters and between high vs. low HAc_score strata) using appropriate statistical tests; multiplicity-adjusted P values were used to determine significance.

Immune infiltration analysis

We estimated immune-cell infiltration using single-sample set enrichment analysis (ssGSEA) as implemented in the GSVA R package. Immune-cell ssGSEA scores were computed using previously published immune-cell gene signatures.

Differential expression and functional enrichment

DEGs were identified across HAc-defined groups using the limma framework. We performed pairwise contrasts among the three HAc_clusters (A vs. B, A vs. C, B vs. C). For downstream analyses, shared DEGs were obtained by intersecting the pairwise results.

DEG lists were subjected to over-representation analysis for Gene Ontology (GO, Biological Process, Molecular Function, Cellular Component) and KEGG pathways using clusterProfiler, with visualization via enrichplot; selected pathways were illustrated using KEGG Mapper.

Construction of the HAc score

To summarize HAc-related expression states at the patient level, we identified prognostic DEGs using univariable Cox proportional-hazards models, yielding a 19-gene set. Gene-wise expression values were scaled, and principal component analysis (PCA) was applied to the 19-gene matrix. For each tumor, the HAc_score was defined as the sum of the sample scores on the first two principal components: HAc_score = PC1_score + PC2_score. The HAc_score was then used in downstream survival and immune-related analyses. This scoring approach was designed to capture the dominant variation in HAc-related gene expression patterns across ECs patients, providing a comprehensive measure of HAc modification.

Correlation between HAc_score and Immunogenomic features in EC

To examine whether the HAc-based signature relates to immunotherapy-relevant biology in EC, we assessed its associations with TMB, mutation spectra, and the IPS.

Somatic mutation MAF files from TCGA-EC were processed with maftools to compute TMB as the number of nonsynonymous single-nucleotide variants and indels per megabase of coding sequence. Associations between HAc_score and TMB were tested by Spearman correlation. Overall survival was compared between TMB groups using Kaplan–Meier curves and log-rank tests. For joint stratification, patients were split into four groups by crossing HAc_score with TMB, and survival was compared by log-rank tests. To describe genomic contexts of the score groups, we generated oncoplots with maftools for the top recurrently mutated genes in high vs. low HAc_score tumors and summarized mutation classes (missense, nonsense, frameshift, splice-site, in-frame indels).

IPS values and component scores (antigen presentation/MHC, effector cells, checkpoints/immunomodulators, suppressor cells) were retrieved from TCIA for TCGA-EC samples. IPS (and component) distributions were compared between high and low HAc_score groups using Wilcoxon rank-sum tests.

Statistical analysis

All analyses were performed in R (v3.6.2; https://www.r-project.org/). Unless otherwise specified, tests were two-sided and P < 0.05 was considered statistically significant. Normality of continuous variables was assessed with the Shapiro–Wilk test. For two-group comparisons we used Student’s t test (parametric) or the Wilcoxon rank-sum test (non-parametric); for ≥ 3 groups, one-way ANOVA or the Kruskal–Wallis test was applied as appropriate. Correlations were evaluated with Pearson (approximately normal) or Spearman (non-parametric) coefficients. Survival was analyzed by Kaplan–Meier estimation with log-rank tests for group differences; effect sizes were obtained from Cox proportional-hazards models.

Results

Genetic variation landscape of the prognostic HAc regulators in EC

We curated 81 HAc–related regulators and, based on univariate screening in TCGA-EC, focused subsequent analyses on a 23-gene prognostic panel (Supplementary Table 1). In tumors with available mutation data (n = 182), 167/182 (91.8%) harbored at least one somatic mutation within this panel (Fig. 1A). This panel-level coverage was largely driven by TP53 (87%), whereas most other regulators were mutated at low individual frequencies (typically ≤ 5%; Fig. 1A; Supplementary Table 1). Stratifying tumors by TP53 status revealed differences in mRNA abundance for several regulators (e.g., lower SET, SIN3A, SNCA, TWIST1, AUTS2, SMARCB1, ING2, NIPBL and higher TADA2B in TP53-mutant tumors; Fig. S1). CNVs were also common across the panel, with gene-specific gains and losses observed at variable frequencies (Fig. 1B) and loci distributed across multiple chromosomes (Fig. 1C). Collectively, these observations indicate that TP53 dominates the mutation landscape within the 23-gene panel while CNVs are widespread among many regulators, which prompted us to interrogate coordinated transcriptomic patterns.

Fig. 1.

Genomic landscape of prognostic HAc regulators in EC. A Somatic mutation profile of 23 prognostic HAc regulators in the enrolled cohort. Each column represents an individual tumor sample, and each row corresponds to a specific HAc regulator. The top bar plot shows tumor mutation burden (TMB) per sample. Mutation types are color-coded as indicated in the legend. The bottom stacked bar plot summarizes the mutation classification for each sample. B Frequencies of copy number variations (CNVs) in the 23 prognostic HAc regulators. Amplification (gain) and deletion (loss) frequencies are shown as red and green dots, respectively. The gray vertical bars indicate total CNV frequency for each gene. C Circular ideogram depicting the chromosomal locations of the 23 prognostic HAc regulators in EC

Consensus clustering of prognostic HAc regulators defines three expression patterns in EC

A correlation network of the 23 prognostic HAc regulators revealed predominantly positive co-expression relationships (20/23), with most genes (17/20) annotated as favorable and a few (3/20) as risk factors by univariate Cox analysis (Fig. 2A). Using consensus clustering of regulator expression, we identified three transcriptomic patterns (HAc_cluster A–C; Fig. 2B). Although the overall difference did not reach statistical significance (log-rank P = 0.084), the Kaplan–Meier curves show a directional separation, with HAc_cluster C tending toward the most favorable survival and HAc_cluster B the least. Heatmaps illustrate cluster-specific expression of the regulators (Fig. 3A). GSVA revealed pathway differences across clusters (Fig. 3B–D). ssGSEA profiling showed between-cluster differences across multiple immune populations (Fig. 3E). HAc_cluster C exhibited higher infiltration in several effector and innate compartments (e.g., activated CD8 T cells, activated dendritic cells, NK subsets, monocytes/neutrophils), this is consistent with its better survival pattern noted in Fig. 2C. The PCA plot shows a trend toward separation among the three clusters, with partial overlap (Fig. 3F). Collectively, these results demonstrate three HAc-related transcriptomic patterns in EC.

Fig. 2.

Consensus clustering analysis identifies three HAc–related transcriptomic subtypes in EC. A Co-expression and prognostic interaction network of the 23 prognostic HAc regulators in EC. Circle size represents statistical significance from univariable Cox proportional hazards analysis. Node color indicates prognostic association: green for favorable and purple for risk factors. Pink and blue lines represent significant Spearman correlations (pink: positive; blue: negative), with line thickness proportional to correlation strength. B Consensus clustering of the 23 prognostic HAc regulators identified three transcriptomic subtypes (HAc_clusters A–C) in the TCGA-EC cohort. The heatmap displays the consensus matrix indicating cluster stability across iterations. C Kaplan–Meier survival curves comparing overall survival among the three HAc_clusters. The number at risk for each cluster is shown below the curves

Fig. 3.

Transcriptomic and immunological characteristics of the three HAc–related transcriptomic subtypes in EC. A Heatmap showing the expression profiles of 23 prognostic HAc regulators across the three identified HAc_clusters. Top annotations indicate clinicopathological features, including sex, T stage, N stage, M stage, survival status, and age. B–D Heatmaps of pathway activity scores derived from GSVA based on KEGG gene sets, comparing functional enrichment between cluster pairs: B cluster A vs. B, C cluster A vs. C, and D cluster B vs. C. E Boxplots illustrating the levels of tumor microenvironment immune cell infiltration across the three clusters. Significance was assessed using Kruskal–Wallis tests; * P < 0.05; ** P < 0.01; *** P < 0.001; ns, not significant. F PCA plot showing the distribution of samples according to transcriptomic profiles. Each dot represents a single sample, color-coded by HAc_cluster subtype (blue: A; orange: B; red: C)

Identifying DEGs across HAc-related transcriptomic patterns

To further probe HAc-related differences in EC, we performed pairwise differential expression among the three HAc_clusters and identified 643 DEGs shared across all three contrasts (Fig. S2A). GO/KEGG analyses indicated enrichment in cell-cycle and genome-maintenance programs—e.g., Cell cycle, DNA replication, homologous recombination, and mismatch/base-excision repair—as well as RNA-processing and proteostasis pathways (Spliceosome, RNA degradation, Proteasome) (Fig. S2B–C). From these DEGs, 19 genes were associated with overall survival in univariate Cox analyses (Supplementary Table 2). Consensus clustering of the 19 DEGs separated the cohort into two gene expression subtypes (geneCluster A and B) (Fig. 4A). Kaplan-Meier survival analysis suggested that geneCluster B showed better overall survival than geneCluster A (P = 0.044; Fig. 4B). The heatmap in Fig. 4C illustrates subtype-specific expression patterns. Expression of the majority (15/23, approximately 65%) of prognostic HAc regulators differs significantly between geneCluster A and geneCluster B (Fig. 4D). Together, these findings define two HAc-related expression subtypes in EC with distinct molecular features and survival patterns.

Fig. 4.

Transcriptomic subtypes based on prognostic DEGs across HAc_clusters. A Consensus clustering analysis of the prognostic DEGs derived from different HAc_clusters identified two robust gene clusters (geneClusters A and B), as shown in the consensus matrix (k = 2). B Kaplan–Meier survival curves comparing overall survival between the two geneClusters. P-value was determined using the log-rank test; the number at risk is displayed below the plot. C Heatmap showing the expression patterns of 19 prognostic DEGs across geneClusters A and B. Annotated clinical features include sex, T stage, N stage, M stage, age at diagnosis, survival status, and original HAc_cluster assignments. D Boxplots comparing the expression levels of 23 HAc regulators between the two geneClusters. * P < 0.05; *** P < 0.001; ns, not significant

The HAc-based signature enables prognostic stratification in EC

To quantify HAc-related patterns at the patient level, we constructed a HAc signature (named HAc_score) from the prognostic DEGs. Patients with higher HAc_scores had better overall survival than those with lower scores (P = 0.028, log-rank; Fig. 5A). The ssGSEA correlation heatmap (Fig. 5B) shows that HAc_score is predominantly negatively associated with immune-cell infiltration signatures, such as activated CD4⁺/CD8⁺ T cells, activated dendritic cells, CD56^dim NK cells, et al. The alluvial plot shows that high HAc_scores are enriched in HAc_cluster C, while low scores are enriched in cluster B, paralleling the survival pattern in Fig. 2C (Fig. 5C). Consistently, HAc_cluster C exhibits the highest HAc_score and cluster B the lowest(Fig. 5D). The high-score group contained a larger fraction of alive cases than the low-score group, and HAc_scores were higher in survivors (P = 0.011; Fig. 6A–B). In stage-stratified analyses, the association was significant in the T0–2, N0, and M0 subsets, whereas trends in T3–4, N1–3, and M1 did not reach significance (Fig. 6C–H). Together, these results support HAc_score as a potential prognostic indicator, particularly in earlier-stage EC.

Fig. 5.

Construction of the HAc score. A Kaplan–Meier survival curves comparing overall survival between high and low HAc_score groups. The red and blue curves represent high and low HAc_score groups, respectively. The log-rank P value is indicated, and the number of patients at risk is shown below. B Correlation matrix displaying associations between the HAc_score and tumor-infiltrating immune cell types in the tumor microenvironment. Red and blue circles indicate positive and negative correlations, respectively. Circle size reflects the strength of the correlation (absolute value of the coefficient). * P < 0.05. C Alluvial diagram visualizing the relationships among HAc_clusters, geneClusters, HAc_score, and survival outcomes. Flow width indicates the relative number of patients transitioning between categories. D Boxplot showing the distribution of HAc_score across the three HAc_clusters (A, B, and C)

Fig. 6.

Prognostic value of the HAc_score in EC. A Stacked bar plot showing the distribution of survival status (Alive vs. Dead) between high and low HAc_score groups. B Boxplot comparing HAc_score between deceased and surviving patients. (C–H) Kaplan–Meier survival curves illustrating overall survival differences between high and low HAc_score groups within clinical subgroups of the TCGA-EC cohort, stratified by T stage (T0–2 [C], T3–4 [D]), N stage (N0 [E], N1–3 [F]), and M stage (M0 [G], M1 [H]). Log-rank P values are indicated in each panel, and numbers at risk are shown below the x-axis

Immunogenomic features associated with HAc_score in EC

We next evaluated immunogenomic correlates of the HAc_score in EC. HAc_score showed a modest inverse correlation with TMB (r = − 0.24, P = 0.002; Fig. 7A), indicating that higher scores tend to occur in tumors with slightly lower mutational load. Consistent with this, patients with low TMB had better overall survival than those with high TMB (P = 0.012; Fig. 7B). Joint stratification by HAc_score and TMB further separated outcomes: the low HAc_score/high-TMB group had the poorest survival, whereas the high HAc_score/low-TMB group had the most favorable survival (Fig. 7C), suggesting that the two metrics provide complementary prognostic information. Mutation landscapes differed between score groups (Fig. 7D–E): although TP53 dominated in both, low-score tumors showed a heavier mutational burden and higher frequencies of large/structural genes (e.g., MUC16, CSMD3, SYNE1), pointing to distinct genomic contexts associated with HAc-related expression states. IPS values (and components related to antigen presentation, effector activity, and checkpoints) were significantly higher in the high-score group (Fig. 7F–J), suggesting a more immunogenic transcriptional program. Collectively, these results indicate that HAc_score reflects a distinct immunogenomic state—characterized by lower TMB, higher IPS, and specific mutation patterns—and, in combination with TMB, enables complementary prognostic stratification in EC.

Fig. 7.

Correlations between HAc_score and immunogenomic characteristics of the tumor microenvironment in EC. A Scatter plot illustrating the negative correlation between HAc_score and tumor mutational burden (TMB). The blue regression line indicates the fitted model with 95% confidence interval (gray area). Data points are color-coded by geneClusters (blue, genCluster A; orange, genCluster B). B Kaplan–Meier curves comparing overall survival between high-TMB (red) and low-TMB (blue) groups. C Kaplan–Meier survival analysis integrating TMB and HAc_score stratification. Log-rank P values are indicated in each panel, and numbers at risk are shown below the x-axis. D Oncoplot of the top 20 most frequently mutated genes in high- and low-HAc_score subgroups of EC. Distinct mutation types are color-coded. E Violin plots comparing immunophenoscore (IPS) between high- (red) and low-HAc_score (blue) groups

Discussion

In current study, we screened prognostic HAc regulators from public datasets and defined three HAc-related expression patterns in EC. Using DEGs across these patterns, we built a patient-level quantitative score, HAc_score, to capture HAc-linked transcriptional states. HAc_score stratified overall survival, with the largest separation in early-stage disease, and it correlated with immunogenomic features of the TME, including lower TMB and higher IPS. This provides a compact readout of HAc-driven programs that helps quantify intertumoral heterogeneity and refine outcome stratification in EC, especially at earlier stages. It also offers a practical way to examine epigenetic–immune crosstalk in EC.

The genomic backdrop we observed included frequent TP53 alterations and broad copy-number changes, which shows the limits of mutation-only stratification for epigenetically mediated biology in EC [29]. TCGA and other large studies have reported high TP53 recurrence and widespread CNVs. These features add to heterogeneity but do not fully explain downstream phenotypes or outcome [30]. In this context, transcriptome-scale signals from HAc regulators can serve as an integrative proxy for chromatin state and its transcriptional effects. The HAc tends to open chromatin and support transcription, whereas deacetylation favors compaction and repression [31, 32]. Prior methods and reviews support the link between HAT/HDAC balance and gene programs, which justifies using HAc-regulator expression as a surrogate of epigenetic state [33].

Next, consensus clustering on prognostic HAc regulators yielded three patterns with coherent pathway differences and distinct immune features. GSVA pointed to shifts in cell-cycle and DNA-maintenance programs and in RNA-processing and proteostasis. Immune deconvolution showed pattern-level variation across multiple immune lineages. GSVA is a reasonable approach to infer pathway activity from bulk RNA profiles, so these clusters likely reflect coordinated biology rather than single-gene effects. Survival curves did not pass a global significance test, but their directions matched the molecular and immune differences. This agrees with pan-cancer observations that regulator-defined transcriptomic states can capture heterogeneity not apparent from genomic calls alone [34, 35].

Then, we identified 643 shared DEGs, of which 19 genes associated with overall survival were used to define two DEG-based subtypes. We then constructed HAc_score, a compact metric quantifying HAc-related states at the patient level. HAc_score stratified survival, and the largest separation was observed in T0–2, N0, and M0 subsets. This suggests particular utility for refining risk in earlier-stage settings, where decisions regarding surveillance intensity and adjuvant therapy are especially sensitive to accurate prognostication. The compact 19-gene basis of HAc_score may facilitate translation across platforms and future integration into multivariable prognostic models. Multi-center studies with large scale are needed in the future to test whether more complex representations add value beyond this simple framework.

In this study, we find that the HAc_score maps onto a distinct immunogenomic state—marked by transcriptionally driven activation of immune programs rather than mutation-derived antigenicity, alongside a decoupling between immune-gene activation and immune-cell infiltration [36]. First, HAc_score showed a modest inverse correlation with TMB. Because TMB is widely used as a surrogate for predicted neoantigenicity, a negative HAc_score–TMB association suggests that the immune features linked to high HAc_score are unlikely to be driven predominantly by mutation-derived antigenicity [37]. Second, IPS was higher in the high-score group—including components reflecting antigen presentation, effector activity, and checkpoints—consistent with activation of immune gene programs on a transcriptional basis. The IPS framework has been validated across cancers as a pathway-level representation of antitumor immunity and has shown associations with checkpoint blockade benefit in multiple settings [38]. Third, at the cellular level, ssGSEA-based estimates of immune-cell infiltration decrease with increasing HAc_score, suggesting that the immune-gene activation captured by IPS is not paralleled by higher immune-cell abundance [39, 40]. Together, these patterns outline an “immune gene–infiltration decoupling” phenotype—gene circuits are activated, yet cellular recruitment appears constrained—which aligns with emerging notions of immune contexture in which transcriptomic “hot” signals can coexist with limited cellular infiltration (for example, immune-excluded states) [39].

TMB-low status was associated with better overall survival. Joint use of HAc_score and TMB further separated outcomes, which implies added prognostic value when considered together. This is in line with the view that single biomarkers rarely capture antitumor immunity. Models that combine tumor-intrinsic features, such as mutations and copy number, with immune readouts, such as IPS and infiltration, explain clinical heterogeneity better. The link between HAc regulators and immune programs is plausible because acetylation controls chromatin access, and shifts in HAT/HDAC balance can rewire networks for antigen presentation, cytokine signaling and checkpoint expression in tumor and immune cells [39]. Overall, HAc_score indexes a relevant immunogenomic state, but validation is still needed.

This study has several limitations. First, analyses were retrospective and limited to TCGA; external validation is still missing. Second, we did not test histology-specific performance (ESCC vs. EAC). Future studies should test the generalizability of HAc_score in independent cohorts, to determine whether the epigenetic states summarized by HAc_score modify therapeutic sensitivity.

In conclusion, HAc_score quantifies HAc-related expression states in EC, stratifies overall survival most clearly in early-stage disease, and associates with specific immunogenomic features. It is a candidate prognostic marker and a practical starting point to study epigenetic–immune interaction in EC.

Author contributions

Fan C., Sun K.: Made contributions to the conception and design of the study; performed data acquisition, analysis, and interpretation; drafted the manuscript and prepared the figures and tables. Feng Z., Zhang P., Xi Y.: Performed data acquisition, statistical analysis, and interpretation; assisted in data processing and result organization. Min X.: Supervised the project; contributed to study design, data acquisition, analysis, and interpretation; and performed critical revision and quality control of the manuscript. All authors have read and approved the final version of the manuscript.

Funding

This research was funded by the National Natural Science Foundation of China [No. 82400892, 82102028]; the Hubei Provincial Natural Science Foundation [No. 2024AFB656]; the Interdiciplinary Research Program of HUST (2025JCYJ062); the Hubei Provincial Health Commission Medical Research Achievement Transformation Project (WJ2023ZH003); the Medical Artificial Intelligence Fund of Tongji Hospital (AI2025A05).

Data availability

The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

Not applicable. This study analyzed publicly available, de-identified data from The Cancer Genome Atlas esophageal carcinoma cohort. No new human participants, identifiable private information, or animal experiments were involved. Therefore, do not require Institutional Review Board review or approval. The study did not involve interaction with or intervention in human participants, and only de-identified public data were analyzed; therefore, consent to participate was not required.

Consent for publication

Not applicable. The manuscript contains no individual person’s data (including images, videos, or identifiable clinical details); therefore, consent for publication was not required.

Competing interests

The authors declare no competing interests.