Histone delactylation shapes epigenetic landscapes and therapeutic stratification in hepatocellular carcinoma

Abstract

Background and aims

Histone lysine lactylation has emerged as a key epigenetic link between metabolism and gene regulation. Its homeostasis depends on both lactylation and delactylation. The biological and clinical relevance of histone lysine delactylation in hepatocellular carcinoma (HCC) remains unclear. This study aimed to investigate the clinical relevance of histone lysine delactylation (HLDL) in hepatocellular carcinoma and to establish an HLDL-based molecular classification for precision therapy.

Methods

We systematically profiled HLDL across multi-omics HCC cohorts (n > 1700) to establish an HLDL-based molecular classification. Distinct subtypes were characterized by genomic, transcriptomic, and immune features. Therapeutic susceptibilities were predicted through pharmacogenomic modeling and validated by in vitro modulation of histone lysine lactylation.

Results

The HLDL classification delineated two major subtypes with distinct biological and therapeutic profiles. Low-HLDL tumors displayed higher genomic instability, activated oncogenic signaling, and an immune-infiltrated yet suppressive microenvironment, corresponding to poorer outcomes but higher sensitivity to immune checkpoint inhibitors, atezolizumab plus bevacizumab therapy, and sorafenib. In contrast, high-HLDL tumors exhibited a metabolically stable and delactylation-active phenotype, responding more favorably to lenvatinib and transarterial chemoembolization. Functional experiments confirmed that reduced HLDL activity led to elevated histone lysine lactylation, PD-L1 upregulation, and lenvatinib resistance, whereas restoring delactylation reversed these effects.

Conclusions

Histone lysine delactylation constitutes a critical epigenetic determinant of HCC heterogeneity and therapeutic response. The HLDL-based classification developed in this study integrates metabolic, immune, and pharmacologic features into a concise framework, offering new avenues for precision therapy and potential targets for modulating the lactylation-delactylation balance in HCC.

1. Introduction

Hepatocellular carcinoma (HCC) ranks among the most prevalent malignant tumors worldwide. HCC is characterized by complex pathophysiology, pronounced heterogeneity, and highly variable clinical outcomes, posing substantial challenges for precision oncology.¹ Emerging research has indicated that the development and progression of HCC involve not only genetic mutations and transcriptional dysregulation but also epigenetic modifications that fundamentally contribute to tumor heterogeneity.²

Histone lysine lactylation is an epigenetic modification driven by aberrant lactate accumulation within the tumor microenvironment. First documented in 2019, histone lactylation is a lactate-dependent epigenetic modification that modulates gene expression and has been increasingly linked to oncogenesis. Histone lactylation is a reversible modification that is dynamically regulated by “writers” (e.g., p300) and “erasers” (HDAC1-3).³ Accumulating evidence suggests that histone lysine lactylation contributes to transcriptional regulation and remodeling of the tumor microenvironment in HCC. Zhang et al. reported that loss of HBO1, which possesses lactate transferase activity, resulted in markedly decreased H3K9 lactylation levels and suppressed HCC cell proliferation, migration, and invasion.⁴ The histone deacetylases HDAC1-3 have been established as key regulators of histone lysine delactylation, and the SIRT2 deacetylase was also identified as a histone delactylase. Several studies have shown these deacetylases play a critical role in HCC.^5,6 However, the role and clinical implications of histone delactylation in HCC remain poorly understood, particularly regarding its potential value in molecular subtyping and personalized therapy.

In this study, we hypothesized that HDAC1-3-associated histone delactylation serves as a key driver of biological heterogeneity and therapeutic outcomes in HCC. We aimed to construct a prognostic model based on histone lysine delactylation (HLDL)-related genes by integrating multiple transcriptomic datasets. Our specific objectives were to stratify patients into distinct molecular subgroups and systematically investigate the association of HLDL levels with the tumor immune microenvironment, oncogenic signaling, and drug sensitivity. This study seeks to provide a theoretical rationale for epigenetically-guided precision subtyping in HCC.

2. Materials and methods

2.1. Data sources

Transcriptomic profiles of patients with HCC were obtained from GEO (Gene Expression Omnibus, GSE14520, GSE202069, GSE104580, GSE109211, and GSE97098), TCGA (The Cancer Genome Atlas), ICGC (International Cancer Genome Consortium), and EGAS00001005503.⁷ For RNA-seq-derived data, uniform processing was performed to convert the data to transcripts per million. The examined datasets included TCGA, ICGC, GSE202069, EGAS00001005503, and GSE97098. For non-standardized microarray-derived datasets, Log₂ normalization was applied. Proteomic and clinical data were retrieved from CPTAC (Clinical Proteomic Tumor Analysis Consortium) and CNHPP (Chinese Human Proteome Project).^8,9 Single-cell RNA sequencing (scRNA-seq) data (GSE149614, GSE151530, GSE282701) were also downloaded from GEO. Details of all datasets are summarized in Supplementary Table 1.

2.2. Differentially expressed gene (DEG) analysis

The ‘limma’ package was used to identify DEGs between the tumor group and non-tumor group in the GSE14520 dataset. An adjusted p < 0.05 and |log₂FC| > 1.5 were determined as the liminal value. The top 20 genes (top 10 down-regulated and top 10 up-regulated) were displayed by heat map.

2.3. Weighted gene co-expression network analysis (WGCNA)

The expression of HDAC1, HDAC2, and HDAC3 was considered as a clinical trait for WGCNA (version 1.70-3) package. First, we clustered all samples and removed outliers. Next, a sample dendrogram combined with trait heat map was constructed. The soft-thresholding power was determined as softpower = 4, which was selected using the scale-free topology fit index R² > 0.85 to meet the standard of a scale-free network (see Supplementary Fig. 1). Gene similarity was calculated using the weighted adjacency matrix, and the gene phylogenetic tree was further constructed. Gene co-expression modules were divided using the dynamic tree cutting algorithm, with the parameter of minModuleSize = 50 set to control the minimum module size. Finally, the modules showing the highest correlation with HDAC1, HDAC2, and HDAC3 expression were screened as key modules for subsequent downstream analysis. Hub genes within the key modules were identified under the threshold criteria of module membership > 0.4 and gene significance > 0.2.

2.4. Prognostic model construction

Using HLDL-DEGs, univariate and LASSO regression analysis were used to obtain prognostic genes in the training dataset. The LASSO model was trained with 10-fold cross-validation to avoid model overfitting, and the optimal λ values were calculated as lambda.min = 0.02309288 and lambda.1se = 0.17879976. Lambda.1se was selected based on the 1-standard error criterion to construct a more parsimonious and robust gene signature. Genes with non-zero coefficient values screened under the optimal λ value were defined as core signature genes for subsequent HLDL score calculation. The risk model calculating formula (1):

$$riskscore=\sum _{i=1}^n\left(\text{coefficient}\left(\text{genei}\right)\ast \text{expr}\left(\text{genei}\right)\right)$$ (1)

Using the median value of the risk score from each sample, patients with HCC with survival information were categorized into two risk subgroups (low-risk and high-risk groups). The prognostic reliability of the model was assessed by Kaplan-Meier and receiver operating characteristic (ROC) curves (1-, 3-, and 5-year). TCGA-LIHC, ICGC, and GSE202069 datasets were used as an external verification set for the risk model.

2.5. Evaluation of immune score, homologous recombination deficiency (HRD), and intratumor heterogeneity (ITH)

Immune infiltration was estimated by the ESTIMATE algorithm.¹⁰ HRD scores were obtained from published resources,¹¹ and ITH was quantified using the DITHER algorithm.¹²

2.6. ScRNA-seq data preprocessing

For unique molecular identifier data, gene expression values were normalized with the "NormalizeData()" function in Seurat (version 4.3.0, developed and maintained by the Satija Lab at the New York Genome Center, New York, NY, USA) using default parameters. Two quality metrics were calculated per cell: the percentage of mitochondrial genes and the total number of genes detected. We excluded cells exhibiting mitochondrial gene expression exceeding 20% or those with fewer than 200 or more than 6000 genes. Cell annotation was performed with canonical markers from the CellMarker database.¹³ T-cell subsets were classified as cytotoxic or exhausted on the basis of marker expression.¹⁴

2.7. Human cell lines and compounds

The human HCC cell lines PLC/PRF/5 and Huh7 were cultured in DMEM supplemented with 10% fetal bovine serum, 100 μg/mL penicillin, and 100 μg/mL streptomycin at 37°C in a humidified atmosphere of 5% CO₂. Lenvatinib (L125518), sodium lactate (S108838), and entinostat (E125068) were purchased from Aladdin Bio-Chem Technolog (Shanghai, China).

2.8. CCK-8 assays

PLC/PRF/5 and Huh7 cells seeded in 96-well plates (3000 cells/well) were incubated with drugs for 48 h at 37 °C, 5% CO₂. Viable cells were quantified using a detection kit (Cell Counting Kit-8) purchased from MedChemExpress (Cat. HY-K0301, USA). The absorbance at 450 nm was measured on a spectrophotometer (Molecular Devices, San Jose, CA, USA). Four parallel replicates were set for each group.

2.9. IC₅₀ values

IC₅₀ values were calculated using CCK-8 assays. Cells were seeded in 96-well plates overnight. The next day, cells were cultured with increasing concentrations of drugs for 48 h at 37 °C, 5% CO₂. IC₅₀ values were calculated using nonlinear dose-response regression curve fits using GraphPad Prism 9.5 (GraphPad Software, CA, USA). Results were obtained from three independent experiments.

2.10. Quantitative reverse transcriptase-PCR

Total RNA was extracted from cells using FreeZol Reagent (Cat. R711, Vazyme, Nanjing, China) following the manufacturer's instructions. HiScript III RT SuperMix for qPCR (Cat. R323, Vazyme) was used to generate cDNA following the manufacturer's instructions. Real-time reverse transcriptase-PCR was performed using Taq Pro Universal SYBR qPCR Master Mix (Cat. Q712, Vazyme) following the manufacturer's instructions. The relative expression of the target gene was calculated by the 2^−ΔΔCt method; β-actin mRNA was used as an internal control. Primer sequences are listed in Supplementary material.

2.11. Western blot

Cells were lysed in RIPA (Radio Immunoprecipitation Assay) Lysis buffer (Cat. P0013B, Beyotime Biotechnology, China) containing 1 mM PMSF (Phenylmethylsulfonyl fluoride, Cat. ST506, Beyotime Biotechnology). Protein concentration was measured using the Pierce BCA protein assay kit (Cat. 23,227, Thermo Scientific, USA). Protein samples were mixed with 5 × loading buffer containing 2-hydroxy-1-ethanethiol. Equal amounts of protein (40 μg) were separated on a 12% SDS-PAGE (sodium dodecyl sulfate - polyacrylamide gel electrophoresis) and transferred to a polyvinylidene fluoride membrane (Cat. IPVH00010, Merck Millipore, Germany). The membrane was blocked with 5% non-fat milk in 1 × TBST (Tris-buffered saline with Tween 20) for 2 h at room temperature. The membranes were then incubated with anti-Histone H3 Rabbit antibody (Cat. PTM-1001RM, 1:1000, PTM Bio Inc., China), pan anti-Kla Rabbit antibody (Cat. PTM-1401, 1:1000, PTM Bio Inc.), and Vinculin Monoclonal antibody (Cat. ET1705-94, 1:20,000, Huabio, China). The membrane was washed with 1 × TBST three times and then incubated with HRP Conjugated Goat anti-Rabbit IgG polyclonal Antibody (Cat. HA1001, 1:50,000, Huabio) as the secondary antibody for 2 h at room temperature. Bands were visualized using ECL chemiluminescence solution (Cat. WBKLS0500, Merck Millipore, Germany).

2.12. Statistical analysis

For comparisons of continuous variables, the Student's t-test was used for normally distributed data (Shapiro-Wilk test) with homogeneous variance (Levene's test), otherwise the Mann-Whitney U test was applied. Categorical variables were compared using the Chi-square test or Fisher's exact test as appropriate. Correlations were assessed via Spearman's rank correlation. To control for false discovery rate, the Benjamini-Hochberg method was applied to all relevant statistical tests. The associations between HLDL (high/low) grouping and clinicopathological categorical variables were assessed for statistical significance using the Chi-square test or Fisher's exact test. Data were analyzed using GraphPad Prism 9.5 (GraphPad Software, CA, USA) and are expressed as mean ± SD. A two-tailed p < 0.05 indicated statistical significance.

3. Results

3.1. Development and validation of the HLDL prognostic model

To elucidate transcriptomic alterations linked to HLDL in HCC, we integrated differential expression analysis and WGCNA in the GSE14520 cohort. By intersecting DEGs with HDAC1-3-correlated modules, we identified 188 HLDL-related genes (Fig. S1A-F). Univariate Cox and LASSO regression analysis identified nine hub genes—ADH1B, AQP9, CD14, CDC37L1, CYP3A43, NR1I2, NR1I3, SPP2, and TTR—comprising the HLDL prognostic signature (Fig. 1A and B). The model effectively stratified patients into high- and low-risk groups, with higher expression of these genes in the low-risk subgroup (Fig. 1C). Time-dependent ROC curves in the training cohort yielded AUCs of 0.74, 0.72, and 0.68 for 1-, 3-, and 5-year survival, respectively (Fig. 1D). External validation in TCGA-LIHC, ICGC-LIRI-JP, and GSE202069 cohorts confirmed consistent predictive performance (Fig. 1E-G), underscoring the robustness and generalizability of the HLDL model across independent datasets. We applied this model for subsequent analysis of clinical relevance, molecular mechanisms, and therapeutic implications.

Fig. 1.

Construction and validation of an HLDL gene signature for patient prognosis. (A-B) Identification of nine HLDL genes by univariate Cox and LASSO analysis. (C-D) Expression patterns, survival status, and ROC curves of patients in GSE14520. (E-G) Validation of predictive performance of the model in TCGA-LIHC (E), ICGC-LIRI-JP (F), and GSE202069 (G) cohorts. Abbreviations: HLDL, histone lysine delactylation; ROC, receiver operating characteristic; HR: hazard ratio.

3.2. Clinical and prognostic relevance of the HLDL score

Having established a robust HLDL-based prognostic model, we next evaluated its clinical and prognostic relevance in patients with HCC. Kaplan-Meier analysis revealed that patients with high HLDL scores (HLDL-H) had significantly better overall survival across the cohorts (Fig. 2A-F, Fig. S2A and B). To specifically highlight the unique advantages of the HLDL classification by excluding interference from the complex tumor immune microenvironment, we performed prognostic analysis on patients stratified into high and low groups on the basis of ESTIMATE immune scores. Notably, while patients stratified by the ESTIMATE immune score showed no significant difference in OS (Fig. S2C), a significant difference in OS was observed between the two patient groups stratified by HLDL subtype classification (Fig. 2A, p < 0.0001). Univariate Cox analysis identified the HLDL score as an independent prognostic factor (Fig. S2D-H).

Fig. 2.

Association of HLDL score with survival and clinicopathological features in HCC cohorts. (A-F) Kaplan-Meier curves across multiple HCC cohorts. (G-I) Correlations of HLDL subtypes with clinicopathological parameters and existing HCC classifications. Abbreviations: HLDL, histone lysine delactylation; K-M: Kaplan-Meier.

We next analyzed the correlation between the HLDL score and various clinical features in multiple HCC datasets. Lower HLDL scores were associated with clinical parameters indicative of HCC malignancy, such as advanced clinical stage, larger tumor size, and higher AFP levels (Fig. 2G-H, Fig. S3A-D). Notably, we also found a significant correlation between hepatitis B virus infection and low HLDL scores. Biologically, HLDL subtypes demonstrated significant correlations with alterations at the genomic, mRNA, miRNA, and protein levels (Fig. S3E). Comparison with established HCC molecular classifications revealed strong concordance while requiring fewer genes (Figs. 2I),^8,9,15, 16, 17, 18 highlighting the efficiency and clinical utility of the HLDL model.

3.3. Multi-omics characterization of HLDL subtypes

To further elucidate the biological basis underlying the prognostic differences between HLDL subtypes, we performed multi-omics analysis, which revealed distinct molecular features between the subtypes. HLDL-L tumors displayed higher genomic instability, as reflected by increased HRD scores, the enrichment of DNA repair-related pathways (Fig. 3A and B), ITH scores (Fig. 3C), and elevated TP53 mutation frequency (Fig. 3D). Gene set enrichment analysis demonstrated activation of cell cycle, MAPK, PI3K/AKT, VEGF, and Wnt signaling pathways (Fig. 3E). Expression of proliferation markers (MKI67, PCNA, TOP2A) was inversely correlated with HLDL scores (Fig. 3F; Fig. S6A-B). Furthermore, multi-omics-based pathway enrichment results revealed that the HLDL-L subtype was significantly enriched in the E2F pathway (Fig. S5A-F). We speculate that the high lactate microenvironment may drive uncontrolled cell cycle progression and malignant proliferation of hepatoma cells through sustained activation of the E2F transcriptional program. A pan-cancer analysis confirmed the negative correlation across 32 tumor types (Fig. S6D), and HLDL scores were generally lower in tumor tissues (Fig. 3G). Collectively, these findings reveal that low HLDL scores indicate high proliferative activity and genomic instability, linking delactylation status to tumor aggressiveness.

Fig. 3.

Genomic instability and pathway differences between HLDL subtypes. (A-C) HRD scores, DNA repair-related pathway enrichment, and ITH scores across subtypes. (D-E) Mutation profiles and TP53 alterations. (F) Enriched oncogenic pathways. (G) Correlations of HLDL score with MKI67 expression. (H) Pan-cancer analysis of tumor versus normal tissues. Abbreviations: HLDL, histone lysine delactylation; HRD: homologous recombination deficiency; ITH: intratumor heterogeneity.

3.4. Immune landscape of HLDL subtypes

Given the close interplay between tumor proliferation, genomic instability, and immune modulation, we next characterized the immune landscape associated with HLDL subtypes. Immune profiling revealed that HLDL-L tumors exhibited a more inflamed yet immunosuppressed microenvironment. The ESTIMATE algorithm indicated significantly higher immune and stromal scores in HLDL-L tumors (Fig. 4A). Single sample GSEA showed enhanced immune-related pathway activity in HLDL-L tumors (Fig. 4B), with upregulated immunoregulatory genes involved in stimulation, inhibition, and antigen presentation (Fig. 4C). The HLDL score negatively correlated with immune cell infiltration (Fig. 4D), markers of exhausted immune microenvironment (Fig. 4E), and regulatory T-cell abundance (Fig. 4F), and HLDL-L tumors showed an elevated M1/M2 macrophage ratio (Fig. 4G).

Fig. 4.

Immune landscape differences between HLDL-L and HLDL-H tumors. (A-B) Immune and stromal scores and pathway enrichment. (C-D) Expression of immune modulator genes and immune cell composition. (E-G) Correlations of HLDL score with markers of immune exhaustion, regulatory T-cell abundance, and M1/M2 ratio. (H-N) Single-cell transcriptomic profiling and intercellular communication analysis showing immune activation with checkpoint suppression in HLDL-L tumors.

ScRNA-seq analysis across three distinct datasets further validated these findings, revealing a higher proportion of T-cell infiltration along with an increased proportion of exhausted T cells in HLDL-L tumors (Fig. 4H-K; Fig. S7A-C, F-H). Cell-cell interaction analysis demonstrated enhanced communication between malignant cells and T-cell subsets in HLDL-L tumors (Fig. 4L-M; Fig. S7D and I), accompanied by elevated activity of the PD-1/PD-L1 signaling pathway (Fig. 4N; Fig. S7E and J). Together, these results indicate that HLDL-L tumors possess an immune-rich yet checkpoint-suppressed microenvironment.

3.5. Therapeutic implications of HLDL classification

Considering the distinct immune and molecular features of HLDL subtypes, we further explored their potential implications for reflecting therapeutic response. In the EGAS00001005503 cohort, low HLDL score was potentially associated with a better response to atezolizumab monotherapy and combined anti-PD-L1/anti-VEGF therapy (Fig. 5A and B), consistent with PD-1/PD-L1 activation in HLDL-L tumors. Notably, combination therapy increased the response rate in HLDL-H patients from 4% to 29%. Conversely, HLDL-H tumors exhibited potentially higher sensitivity to TACE, while HLDL-L tumors showed potentially higher sensitivity to sorafenib (Fig. 5C-E). Moreover, across three independent datasets, the HLDL score showed a negative correlation with response to both sorafenib and lenvatinib, suggesting that HLDL-H patients may be more sensitive to these therapies (Fig. 5F and G). These data indicate that HLDL classification may serve as a predictive biomarker for therapy selection, linking the epigenetic state of tumors to treatment response.

Fig. 5.

Therapeutic responses associated with HLDL classification. (A-B) Responses to atezolizumab or atezolizumab plus bevacizumab therapy. (C-D) TACE outcomes. (E) Response to sorafenib. (F-G) Correlation of HLDL score with sorafenib and lenvatinib sensitivity. Abbreviations: HLDL, histone lysine delactylation; T + A, atezolizumab plus bevacizumab; RECIST, response evaluation criteria in solid tumors; TACE, transcatheter arterial chemoembolization.

3.6. Experimental validation of the HLDL signature

Finally, to experimentally validate the biological relevance of the HLDL classification, we examined HCC cell lines with contrasting HLDL scores. PLC/PRF/5 cells (low HLDL score) showed markedly higher histone lysine lactylation levels than Huh7 cells (high HLDL score) (Fig. 6A and B). Moreover, Huh7 cells demonstrated greater sensitivity to sorafenib and exhibited lower PD-L1 expression (Fig. 6C and D). Sodium lactate treatment increased histone pan-lactylation, suppressed expression of HLDL signature genes, and upregulated MKI67, PCNA, PD-L1, and HLA-A (Fig. 6E-H). Furthermore, lactate exposure induced lenvatinib resistance by upregulating JUNexprssion, a critical driver of lenvatinib resistance. (Fig. 6I and J).

Fig. 6.

Experimental validation of HLDL subtypes and effects of histone lactylation in HCC cells. (A-D) Subtype identification, histone lactylation, and key phenotypes in HCC cell lines. (E-J) Effects of sodium lactate on histone pan-lactylation, gene expression, proliferation, immune modulation, and lenvatinib resistance in HCC cells in vitro. (K-N) Effects of entinostat on gene expression and proliferation in Huh7 cells in vitro. Statistical analysis was performed by unpaired Student's t-test (D, G-I, N) or two-way ANOVA followed by Tukey's multiple comparison test (F, J, L-M). Data are presented as means ± SD (n = 3, ∗p < 0.05,∗∗∗p < 0.001,∗∗∗∗p < 0.0001, ns: No significance).

To validate the delactylation activity of HDACs and its association with HLDL signature genes, we conducted experiments on Huh7 cells using the class I HDACs inhibitor entinostat. In Huh7 cells treated with entinostat, Western blot analysis revealed increased global lactylation, indicative of inhibited delactylation activity of class I HDACs, and qPCR confirmed the downregulation of HLDL signature genes. (Fig. 6K-L). CCK-8 assays further confirmed that entinostat-induced delactylation inhibition, as reflected by decreased HLDL levels, was correlated with enhanced cell proliferation. (Fig. 6M). Additionally, treatment of Huh7 cells with entinostat significantly upregulated the expression of MKI67, PD-L1, and JUN, which was consistent with the trend observed in the sodium lactate experiments (Fig. 6N). These experiments confirmed that histone lactylation promotes tumor cell proliferation, immune modulation, and drug resistance, validating the computationally-derived HLDL classification.

4. Discussion

Most existing molecular classifications of HCC emphasize genomic or clinical features but provide limited therapeutic guidance. In this study, we identified a novel HLDL subtype of HCC using multi-omics data and explored the clinical, genomic, signaling pathway and immunological characteristics. Our results indicate that HLDL subtyping may facilitate personalized medicine for the clinical management of HCC, in which the HLDL-L subtype may represent responders to sorafenib and immunotherapy, whereas the HLDL-H subtype may correspond to responders to TACE. We also found that the addition of anti-C antibodies enhanced the response of the HLDL-H subtype to immunotherapy. Furthermore, the association between HLDL subtyping and molecular features of HCC was validated through in vitro experiments. Compared with other systems, ^8,9,15, 16, 17, 18 the HLDL subtyping framework established in this study provides a precise and actionable basis for clinical treatment decision-making in patients with HCC (Fig. 7).

Fig. 7.

Summary of the HLDL subtypes of HCC and the HLDL subtype-based personalized treatment strategies. Abbreviations: HLDL, histone lysine delactylation; Kla, histone lysine lactylation. T + A, atezolizumab plus bevacizumab; RECIST, Response Evaluation Criteria in Solid Tumors; TACE, transcatheter arterial chemoembolization.

The balance of histone lactylation is disrupted by the aberrant overexpression of HDAC1-3.⁵ Our analysis showed that dysregulated delactylation alters a set of tumor-associated metabolic and signaling pathways, consistent with previous reports.19, 20, 21, 22 Notably, while HDAC1-3 function as both deacetylases and delactylases, our Western blot analysis (Fig. 6B) confirmed that the HLDL signature acts as a specific phenotypic indicator of lactylation status, validating the HLDL score as a robust surrogate for histone delactylation activity. The HLDL model, focused on histone delactylation-related genes, connects epigenetic regulation with metabolic and immune heterogeneity, offering a unified framework that can inform individualized therapeutic strategies in HCC based on the tumor's epigenetic-metabolic profile.

Lactate accumulation, a hallmark of metabolic reprogramming, drives immune exhaustion by impairing T-cell metabolism.^23,24 HLDL-L tumors contain abundant cytotoxic yet exhausted T cells with enhanced PD-1/PD-L1 signaling. This phenotype likely explains their improved immune checkpoint inhibitor response, as checkpoint blockade may partially reverse lactate-induced suppression. In contrast, the more metabolically stable, delactylation-active HLDL-H tumors responded better to TACE. Mechanistically, this heightened sensitivity is likely attributable to the dependence of HLDL-H tumors on oxidative phosphorylation and angiogenic signaling, rendering them particularly vulnerable to ischemic stress induced by TACE. Conversely, reduced TACE efficacy in HLDL-L tumors may result from hypoxia-induced resistance under glycolytic stress.²⁵ Of notable clinical value, we found a significant negative correlation between HLDL scores and patient resistance to lenvatinib and sorafenib, which aligns with previously reported studies showing that excessive lactylation induces lenvatinib resistance in HCC.²⁶

The HLDL signature holds promise for translation into a simplified clinical tool for patients with HCC, potentially through the development of standardized 9-gene qRT-PCR or NanoString assays compatible with formalin-fixed paraffin-embedded specimens. Such a stratification method, if validated, could facilitate decision-making by identifying HLDL-L patients as potential candidates for immune checkpoint blockade and HLDL-H patients for tyrosine kinase inhibitors (e.g., lenvatinib) or TACE regimens. Collectively, our findings underscore the potential of the HLDL model to bridge the gap between epigenetic remodeling and clinical practice, providing a robust framework for optimizing personalized therapeutic strategies in HCC.

Despite these promising findings, several limitations should be acknowledged. First, our model was constructed using retrospective data from public databases, which may introduce inherent selection bias. Second, while we provided ex vivo validation, large-scale prospective clinical cohorts and further in vivo experiments are required to fully confirm the robustness and therapeutic predictive value of the HLDL signature. Third, while we validated global lactylation changes, high-resolution mapping is required to elucidate the specific chromatin-level targets of HLDL-associated delactylation. Fourth, the functional role of JUN in the proposed “lactate-JUN-lenvatinib resistance” regulatory axis remains to be further verified by in-depth in vitro and in vivo functional experiments, and the detailed molecular regulatory mechanism of this axis needs to be further explored and clarified.

5. Conclusion

Histone delactylation represents a key epigenetic determinant of tumor aggressiveness and therapy response. The HLDL score integrates metabolic, genomic, immune, and pharmacologic features into a compact and clinically practical index. Future prospective studies are warranted to validate its predictive utility and explore therapeutic strategies targeting the lactylation-delactylation balance in HCC.

CRediT authorship contribution statement

Zijie Wu: Writing – review & editing, Writing – original draft, Validation, Software, Investigation, Formal analysis, Data curation. Ximing Li: Investigation. Zhijun He: Data curation. Binghua Li: Resources. Yanchao Xu: Resources. Weiwei Hu: Software, Investigation. Jintong Li: Validation. Wenbing Yao: Supervision. Qiang Liu: Resources. Xiangdong Gao: Supervision, Resources. Tao Jiang: Supervision, Resources, Investigation. Hong Tian: Writing – review & editing, Visualization, Supervision, Resources, Project administration, Methodology, Investigation, Conceptualization.

Informed consent

It is not applicable.

Organ donation

This study did not involve the use of donated human organs or tissues.

Ethics statement

Ethical statements were waived since we used only publicly available data and materials in this study.

Data availability statement

All data supporting the findings of this study are available within the paper and its Supplementary material.

Animal treatment

No animal experiments were performed in this study.

Declaration of generative AI and AI-assisted technologies in the writing process

During the preparation of this work, the authors used Kimi (Moonshot AI) in order to language polishing, grammatical refinement, and formatting consistency checks of the manuscript. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.

Funding

This work was supported by the National Natural Science Foundation of China (No. 82273840, No. 82073754), and the Key R&D Program of Xinjiang Uygur Autonomous Region (2020B03003).

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article: