A lactylation-driven prognostic model for lung adenocarcinoma: cellular lactylation heterogeneity and immune insights

Introduction

The lung adenocarcinoma (LUAD), the most prevalent subtype of non-small cell lung cancer (NSCLC), accounts for approximately 45.6% of male and 59.7% of female lung cancer cases worldwide (1). In recent years, the incidence of LUAD has been steadily increasing, with particularly notable trends observed among females and non-smokers (1). Owing to the asymptomatic nature of early-stage disease, diagnosis is frequently established at advanced stages, leading to an unfavorable overall prognosis. The 5-year survival rate of early-stage LUAD remains markedly higher than that of advanced disease (2).

Current therapeutic strategies for LUAD include surgical resection, chemotherapy, radiotherapy, targeted therapy, and immunotherapy based on immune checkpoint inhibitors (ICIs) (3-5). ICIs directed against programmed cell death protein 1 (PD-1)/programmed death-ligand 1 (PD-L1), such as nivolumab and pembrolizumab, have substantially improved survival in a subset of patients without actionable driver mutations (6). Nonetheless, LUAD often displays a “cold tumor” immune phenotype, defined by limited CD8⁺ T-cell infiltration and low PD-L1 expression, which restricts the overall efficacy of ICIs (7). Moreover, durable responses to immunotherapy are observed only in a minority of patients, with resistance arising from factors including an immunosuppressive tumor microenvironment (TME) (e.g., regulatory T-cell infiltration), B-cell dysfunction, and aberrant epigenetic regulation (8). Combination regimens that integrate ICIs with chemotherapy, anti-angiogenic agents, or targeted therapies have been investigated to improve clinical outcomes (8), although enhanced toxicity and suboptimal efficacy remain problematic (6). In addition, conventional biomarkers such as PD-L1 expression and tumor mutational burden (TMB) exhibit limited predictive accuracy, underscoring the urgent need for multi-omics approaches to define molecular subtypes and actionable pathways that can guide personalized LUAD therapy (6,9).

Lactylation, a recently identified lysine post-translational modification, is defined by the covalent attachment of a lactyl group to histone or non-histone proteins through an acylation reaction, thereby modulating gene transcription and protein function (10). In cancer cells, metabolic reprogramming toward aerobic glycolysis, known as the Warburg effect, leads to substantial lactate accumulation, which in turn induces lactylation modifications. A metabolic-epigenetic feedback loop is thereby established, facilitating adaptive tumor growth and malignant progression (11). Mechanistically, lactylation levels have been demonstrated to correlate positively with glycolytic flux, with lactate concentrations directly determining the extent of modification. Within the TME, lactylation functions as an integrator of metabolic signaling and cell fate determination (11). Beyond transcriptional regulation, lactylation has been implicated in immune modulation, DNA repair, and cellular metabolism, collectively driving tumor progression, metastasis, and therapeutic resistance (12,13). In colorectal cancer, enhanced expression of RUBCNL, an autophagy-associated gene, has been attributed to lactylation, thereby promoting resistance to bevacizumab (14). Under nutrient-deprived conditions, lactylation has been shown to regulate metabolic enzyme activity to preserve energy homeostasis (15). Moreover, lactylation serves a pivotal role in remodeling the immune landscape of tumors, including the induction of immunosuppressive factors by macrophages and attenuation of CD8⁺ T-cell effector function (16,17). Consequently, lactylation is now widely recognized as a molecular bridge linking tumor metabolism with immune evasion. Considering its multifaceted involvement in cancer-associated pathways, enzymes governing lactylation have emerged as potential therapeutic targets (16,18), and multiple lactylation-associated genes have demonstrated prognostic relevance (19,20).

Despite the growing attention to lactylation in oncology, its role in LUAD remains insufficiently characterized. In particular, the heterogeneity of lactylation at the cellular level and its immunoregulatory mechanisms within the TME have not yet been systematically delineated. Given the potential of lactylation to serve as a crucial link between tumor metabolism and immune suppression, a comprehensive multi-omics investigation was conducted to clarify its biological significance and clinical implications in LUAD. By integrating single-cell and bulk transcriptomic data with cell-cell communication analysis, machine learning-based modeling, and immune profiling, the objectives were the identification of highly lactylated subpopulations, the development and validation of a lactylation-associated prognostic model, and the exploration of the modification’s impact on the immune microenvironment and therapeutic vulnerabilities. These efforts are expected to yield mechanistic insights and inform actionable strategies for advancing personalized management of LUAD. We present this article in accordance with the TRIPOD and MDAR reporting checklists (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-aw-1170/rc).

Methods

Data collection

Bulk RNA expression matrices were obtained from The Cancer Genome Atlas (TCGA) database (https://portal.gdc.cancer.gov/), including transcriptomic profiles of 568 LUAD patients along with corresponding clinical follow-up information. An external validation cohort was incorporated from the Gene Expression Omnibus (GEO) database, specifically the GSE87340 dataset (21), which contains tumor tissues and matched adjacent normal lung tissues from 27 LUAD patients, together with complete clinical annotations. Single-cell transcriptomic data were retrieved from the GEO dataset GSE131907 (22), comprising 11 untreated LUAD tumor samples and 11 normal lung tissue samples. A lactylation-related gene set was obtained from previously published literature (23,24). The other gene sets used for identifying malignant tumor cells were also derived from previously published literature (25).

Preprocessing of single-cell transcriptomic data

Single-cell transcriptomic data were processed and analyzed using the R packages Seurat (v5.1.0) (26) and Harmony (v1.2.1). Quality control was performed to retain cells expressing >300 genes and exhibiting <20% mitochondrial gene content. The expression matrix was normalized, and 2,000 highly variable genes were selected. Mitochondrial, hemoglobin, and ribosomal genes were excluded to minimize technical artifacts.

Subsequently, data were scaled, and confounding effects from nCount_RNA, nFeature_RNA, and cell cycle phase were regressed out. Principal component analysis was performed using the top 30 principal components, followed by batch effect correction with the Harmony algorithm. Cell clustering was performed using a k-nearest neighbor parameter of 40 and a resolution of 0.1. Uniform Manifold Approximation and Projection (UMAP) was applied for two-dimensional visualization of the cellular landscape. Cell type annotation was conducted using canonical marker genes reported in the literature and further validated with the CellMarker 2.0 database (27).

To investigate the heterogeneity of lactylation across key cellular subsets, epithelial cells, stromal cells, myeloid cells, and lymphocytes were extracted and subjected to refined subcluster analysis.

Copy number variation (CNV) inference in epithelial cells

Genomic instability and malignancy-associated features in epithelial cells were evaluated by inferring CNV from scRNA-seq data using the R package inferCNV (https://github.com/broadinstitute/inferCNV). Epithelial cells were designated as the target population, while B cells and mast cells served as non-malignant reference populations. CNV patterns were estimated by comparing relative gene expression across chromosomal regions between target and reference cells. CNV scores were subsequently compared among epithelial subpopulations to identify clusters potentially exhibiting putative malignant characteristics.

AUCell scoring of feature gene sets

Lactylation activity and other tumor-related gene features in individual cellular subpopulations were assessed using the R package AUCell (28). For each gene set, area under the curve (AUC) scores were calculated based on ranked gene expression profiles, providing a robust, clustering-independent measure of gene set activity, where higher AUC scores indicated greater expression activity. To assess the malignant characteristics of epithelial subpopulations, we scored tumor-related gene sets derived from previously published studies, including cell cycle-G2/M (core mitotic regulators reflecting uncontrolled proliferation), cell cycle high-mobility group (HMG)-rich (chromatin remodeling programs driving aggressive tumor behavior), Unfolded Protein Response (stress-response pathways activated in rapidly proliferating tumor cells), and MYC proto-oncogene (MYC) (MYC-driven programs regulating metabolism, ribosome biogenesis, and cell growth). Epithelial subpopulations exhibiting high AUC scores for lactylation and tumor-related gene sets were selected for downstream analyses, allowing integrated evaluation of both metabolic activity and malignant potential.

Cell-cell communication analysis

To investigate potential interactions between lactylation-high epithelial subpopulations and immune cells, cell-cell communication networks were constructed and analyzed using CellChat (v1.6.1) (29). Ligand-receptor pairs associated with lactylation states were identified, and significantly enriched signaling pathways (P<0.01) were incorporated as input features for downstream machine learning analyses. Bubble plots were generated to visualize the communication probability and strength of ligand-receptor interactions across cellular subsets, elucidating lactylation-driven communication networks.

Prognostic model construction and survival analysis

A lactylation-associated prognostic model was developed using the TCGA-LUAD cohort as the training dataset, and its performance was evaluated in an independent test cohort, GSE87340. Model construction followed the TRIPOD guidelines to ensure rigorous model development, evaluation, and validation (30). It was performed using the R package Mime1 (v0.99.9) (31), which integrates ten machine learning algorithms, including random survival forest (RSF), least absolute shrinkage and selection operator (LASSO), and support vector machine (SVM), to generate a total of 101 combined models. The optimal model was selected based on the AUC and the concordance index (C-index) calculated in both the training and validation cohorts.

Input features included: (I) the lactylation-related gene set; (II) the top 20 characteristic genes from epithelial cells exhibiting high lactylation activity; and (III) ligand-receptor gene pairs significantly enriched between lactylated epithelial cells and immune cells. Genes consistently selected in ≥15 models were defined as key genes and used to reconstruct the 101 models, resulting in the establishment of a lactylation-related prognostic scoring system.

Patients were stratified into high- and low-risk groups based on the optimal cutoff determined using the surv_cutpoint function from the R package survminer (v0.4.9). Kaplan-Meier survival curves were generated using the R packages survival (v3.5.5) and survminer (v0.4.9), and intergroup differences were assessed using the log-rank test.

Immunological correlation analysis

Associations between the lactylation-related risk score and the tumor immune microenvironment, as well as potential immunotherapy responsiveness, were assessed using multiple computational algorithms. The immunophenoscore (IPS) for each sample was calculated with the R package IOBR (v0.99.9), integrating antigen presentation, effector immune function, and immunosuppressive features. Relative infiltration of six major immune cell types (B cells, CD4⁺ T cells, CD8⁺ T cells, neutrophils, macrophages, and dendritic cells) was inferred via the TIMER algorithm implemented in IOBR.

ImmuneScore, StromalScore, and ESTIMATE score were calculated using the R package estimate (v1.0.13) to assess immune cell infiltration, stromal content, and tumor purity. The tumor immune dysfunction and exclusion (TIDE) web platform (http://tide.dfci.harvard.edu/) was applied to evaluate immune dysfunction and exclusion scores and predict responses to immune checkpoint blockade.

Patient sample collection

LUAD tissue specimens were collected from patients undergoing surgical resection at Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology. Written informed consent was obtained from all participants. Tumor and adjacent non-tumorous tissues were collected intraoperatively, immediately snap-frozen in liquid nitrogen, and stored at −80 ℃ until analysis. The study protocol was approved by the Ethics Committee of Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology (approval No. TJ-IRB20220329) and conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Quantitative real-time polymerase chain reaction (qRT-PCR)

Total RNA was extracted using the TRIzol method (TaKaRa RNAiso Reagent, 9180, Shiga, Japan). RNA

quality and integrity were assessed using a NanoDrop 2000 spectrophotometer (Thermo Fisher, Waltham, USA) and 1.5% agarose gel electrophoresis. cDNA synthesis was performed using the PrimeScript™ RT reagent Kit with gDNA Eraser (RR047A, Takara, Shiga, Japan) according to the manufacturer’s instructions. qRT-PCR was conducted with TB Green Premix Ex Taq™ II (RR420A, Takara) on a Bio-Rad CFX96 Real-Time PCR Detection System. Reactions (20 µL) were subjected to the following cycling conditions: 95 ℃ for 30 s, followed by 45 cycles of 95 ℃ for 5 s and 60 ℃ for 30 s, and finalized with melt curve analysis. Relative gene expression levels were calculated using the 2^−ΔΔCt method, with all reactions performed in triplicate. Primer sequences are listed in Table S1.

Cell culture and treatment

The human LUAD cell line A549 was obtained from Xiamen Yimo Biotechnology Co., Ltd. (Xiamen, China; IM-H113). Cells were cultured in Dulbecco’s Modified Eagle Medium (DMEM) (Gibco, Waltham, USA, C11995) supplemented with 10% fetal bovine serum (FBS) (HyClone, Waltham, USA, SH30406.05) and 1% penicillinstreptomycin (Haicheng Yuanhong, Beijing, China, HCCC102) at 37 ℃ with 5% CO₂. Cells were passaged at ~80% confluence using 0.25% trypsin (Invitrogen, Carlsbad, USA, 25200056), and only cultures with >90% viability were used for experiments. For functional assays, cells were seeded at 2×10⁶ cells/mL and subjected to hypoxic treatment for 0, 4, 8, or 12 h at 50–60% confluence. Cells were subsequently collected, lysed with TRIzol, and RNA was extracted for qRT-PCR analysis.

Immunofluorescence staining

Frozen tissue sections (8 μm) were fixed and subjected to antigen retrieval in ethylenediaminetetraacetic acid (EDTA) buffer (pH 8.0, Servicebio, Wuhan, China, G1206) for 30 min in a water bath. Sections were blocked with 3% bovine serum albumin (BSA) or 10% rabbit serum and incubated overnight at 4 ℃ with primary antibodies against ANGPTL4 (1:1,000, rabbit, Abclonal, Wuhan, China, A2011) and cytokeratin 19 (CK19) (1:500, rabbit, Abclonal, A0247). After phosphate-buffered saline (PBS) washes, sections were incubated with horseradish peroxidase-conjugated goat anti-rabbit IgG (1:500, Servicebio, GB23303) for 50 min at room temperature, followed by tyramide signal amplification (TSA, Servicebio). Sequential antibody incubations combined with TSA deposition were performed for multiplex staining. Nuclei were counterstained with 4‘,6-diamidino-2-phenylindole (DAPI) (Servicebio, G1012), autofluorescence was quenched when necessary, and slides were mounted with antifade medium. Images were acquired using a fluorescence microscope (Nikon Eclipse C1, Tokyo, Japan) and a digital slide scanner (Pannoramic MIDI, Budapest, Hungary).

Statistical analysis

All statistical analyses were conducted using R software (v4.2.0). Continuous variables were compared using the Wilcoxon rank-sum test, and categorical variables were analyzed with the Chi-squared test or Fisher’s exact test as appropriate. Kaplan-Meier survival curves were generated, with intergroup differences assessed using the log-rank test. Univariate and multivariate prognostic analyses were conducted using Cox proportional hazards regression models. Two-sided P values <0.05 were considered statistically significant.

Results

Heterogeneity of cell lactylation

After quality control, cells were subjected to dimensionality reduction and clustering, and major cell types were annotated based on known marker genes (Figure S1A), resulting in four primary cell types: epithelial cells, stromal cells, myeloid cells, and lymphocytes. Among these cell types, lactylation activity was higher in epithelial and stromal cells from tumor tissues compared with their counterparts in normal tissues (Figure 1A,1B). Subsequently, major cell populations were further subdivided for refined analysis (Figure S1B).

Figure 1 Lactylation heterogeneity in tumor and stromal cells of lung adenocarcinoma. (A) UMAP plot of all cell types; (B) lactylation AUCell scores across tumor and normal cell populations; (C) UMAP plot of stromal cell subclusters; (D) lactylation AUCell scores of stromal subclusters; (E) UMAP plot of lactylation AUCell scores in stromal subtypes; (F) violin plots comparing lactylation AUCell scores of stromal subclusters between tumor and normal samples. **, P<0.01; ****, P<0.0001; ns, not significant. UMAP, Uniform Manifold Approximation and Projection.

Within stromal cells, subpopulations such as lymphatic endothelial cells, Fibro_3, and myofibroblasts displayed relatively higher lactylation scores. In tumor tissues, lymphatic endothelial cells, myofibroblasts, Fibro_1, Fibro_2, Endo_1, and Endo_2 were characterized by comparatively greater lactylation activity, suggesting that highly lactylated stromal cells may contribute to tumor progression (Figure 1C-1F).

Epithelial cells were analyzed in greater detail, revealing ten subpopulations (Epi_1 to Epi_10). Among these, Epi_10, Epi_5, Epi_4, and Epi_3 displayed relatively higher lactylation scores and comparatively greater CNV levels, suggestive of genomic instability. Notably, Epi_10, Epi_5, and Epi_3 were predominantly derived from tumor tissues, suggesting tumor specificity and a potential association with malignant transformation and tumor progression (Figure 2), a similar trend was observed when feature scoring was performed using other tumor-related gene sets (Figure S2).

Figure 2 Lactylation heterogeneity of epithelial cells in lung adenocarcinoma. (A) UMAP plot of epithelial subclusters; (B) UMAP plot of lactylation AUCell scores across epithelial subclusters; (C) lactylation AUCell scores of individual epithelial subclusters; (D) comparison of lactylation AUCell scores between tumor and normal samples across epithelial subclusters; (E) UMAP plot of CNV scores in epithelial cells; (F) distribution of CNV scores in epithelial cells. Significance is indicated by asterisks: **, P<0.01; ****, P<0.0001; ns, not significant. CNV, copy number variation; UMAP, Uniform Manifold Approximation and Projection.

Construction of 101 machine learning models and identification of key genes

To develop a lactylation-related prognostic model and identify key functional genes, three sources of candidate genes were integrated: signature genes of highly lactylated epithelial cells, ligand-receptor pairs between highly lactylated cells and immune cells, and lactylation-related genes reported in the literature, yielding 770 candidate genes.

Using the R package Mime1, 101 combination models were constructed by integrating ten mainstream machine learning algorithms, including Lasso-Cox, Ridge-Cox, random survival forest, CoxBoost, and GBM-Cox. Gene occurrence frequencies across models were calculated, identifying twelve key genes that appeared in ≥15 models, including nine risk-associated genes (ANGPTL4, ITGA6, SOD1, CCL20, FKBP3, DECR1, SEMA3C, TRIM28, VEGFC) and three protective genes (TNNC2, CRTAC1, HGF). The genes originated from three sources: the lactylation gene set (SOD1, FKBP3, DECR1, TRIM28), ligand-receptor pairs (ANGPTL4, ITGA6, SEMA3C, VEGFC, HGF), and signature genes of highly lactylated cells (CCL20, CRTAC1, TNNC2).

The twelve key genes were subsequently used as input variables to retrain models on the TCGA-LUAD dataset, with independent testing performed on GSE87340. The RSF model achieved the highest AUC and C-index in both datasets (Figure 3), indicating a potential association with patient outcomes. In addition, the importance of these genes was further demonstrated through univariate Cox regression analysis (Figure S3).

Figure 3 Development and evaluation of the machine learning-based signature. (A) AUC values of the lactylation-based prognostic model across 101 machine learning algorithms. (B) Corresponding C-index values. AUC, area under the curve; GBM, gradient boosting machine; HR, hazard ratio; LASSO, least absolute shrinkage and selection operator; RSF, random survival forest; SVM, support vector machine.

Association of risk score with patient prognosis and immune features

A personalized risk score system was established based on the twelve key genes, and patients were stratified into high- and low-risk groups to assess intergroup differences. Survival analyses demonstrated that patients in the low-risk group exhibited longer overall survival (OS) than those in the high-risk group in both the TCGA and GSE87340 cohorts (log-rank P<0.05; Figure 4A,4B). In the TCGA cohort, the median OS was 2,600 days for the low-risk group versus 400 days for the high-risk group. In the GSE87340 cohort, the high-risk group exhibited a median OS of 2,100 days, whereas the low-risk group had not yet reached the median survival time.

Figure 4 Prognostic and immune characteristics of the risk model. (A) OS comparison between high- and low-risk groups in the training cohort; (B) OS comparison in the testing cohort; (C) immune cell infiltration differences between risk groups; (D) ESTIMATE scores of the two groups; (E) TIDE scores of the two groups. Asterisks denote significance: *, P<0.05; **, P<0.01; ****, P<0.0001; ns, not significant. CI, confidence interval; HR, hazard ratio; OS, overall survival; TIDE, tumor immune dysfunction and exclusion.

Associations between the risk score and the tumor immune microenvironment were investigated using TIMER, ESTIMATE, and TIDE algorithms. The high-risk group was observed to have lower infiltration levels of CD4⁺ and CD8⁺ T cells (Figure 4C). ESTIMATE analysis revealed reduced StromalScore, ImmuneScore, and ESTIMATE scores in the high-risk group, reflecting higher tumor purity and impaired immune cell infiltration (Figure 4D).

IPS analyses indicated a higher proportion of immunosuppressive cells in the high-risk group. TIDE analysis demonstrated significantly increased immune exclusion scores in high-risk patients (P<0.001), suggesting a potential association between high-risk status and impaired response to immune checkpoint blockade as well as enhanced immune evasion characteristics (Figure 4E).

Signaling pathways by which highly lactylated cells regulate immune cells

Potential regulatory effects of highly lactylated epithelial cells on the tumor immune microenvironment were investigated using CellChat to infer intercellular communication networks. Frequent interactions were observed between highly lactylated cells and immune cells. Among signaling pathways, fibroblast growth factor (FGF), ephrin type-A receptor (EPHA), and laminin exhibited the highest communication probabilities, with myeloid cells identified as the most active immune cell type in intercellular signaling (Figure 5A-5C).

Figure 5 Interactions between highly lactylated cells and immune cells. (A) Sankey diagram of interactions between highly lactylated cells and immune cells. (B) Number of signaling pathways involving different immune cell types. (C) Top 20 signaling pathways most frequently observed in these interactions. (D) Signaling pathways linked to CellChat-identified genes, with source and target cells on the x-axis and ligand-receptor pairs on the y-axis. Commun. Prob. denotes the communication probability, where higher values suggest a stronger potential interaction through the corresponding signaling pathway between the two cell populations.

The analysis was further focused on five ligand-receptor-related genes identified by the machine learning model. ITGA6 encodes a surface receptor expressed on immune cells, whereas the remaining four genes encode ligands secreted by highly lactylated epithelial cells. Specifically, ANGPTL4 secreted by Epi_10 was predicted to interact with T cells, natural killer cells, and myeloid cells via the integrin subunit alpha 5-integrin subunit beta 1 (ITGA5-ITGB1) complex. SEMA3C secreted by Epi_4 was predicted to interact with myeloid cells through neuropilin 1/neuropilin 2 (NRP1/NRP2) and plexin D1 (PLXND1) (Figure 5D). Both ligand-receptor pairs exhibited strong signal intensities in single-gene cell communication networks, suggesting potential involvement in immune suppression and modulation of anti-tumor immune responses.

Highly lactylated epithelial cells were therefore observed to potentially influence immune cell function through specific ligand-receptor axes, contributing to the regulation of the tumor immune microenvironment.

PCR validation confirmed the reliability of the prognostic model at the transcriptional level

The reliability of the lactylation-related prognostic model at the transcriptional level was assessed using qRT-PCR. Among the twelve selected genes, eleven displayed expression patterns consistent with the model’s risk classification (Figure 6). The first nine genes, previously identified as risk-associated, exhibited higher expression levels in the high-risk group, supporting their potential pro-tumor roles. In contrast, the tenth and eleventh genes, classified as protective, were observed to have higher expression in the low-risk group, in agreement with their predicted tumor-suppressive functions. Only the twelfth gene displayed an expression pattern that did not correspond to the model’s prediction. Collectively, these observations support the potential prognostic value of the model and highlight the biological relevance of most selected genes.

Figure 6 Relative expression levels of lactylation-associated genes in normal lung and tumor lung tissues. (A-L) ANGPTL4, ITGA6, SOD1, CCL20, FKBP3, DECR1, SEMA3C, TRIM28, VEGFC, TNNC2, CRTAC1, and HGF, respectively. Asterisks denote significance: *, P<0.05; **, P<0.01; ***, P<0.001; ns, not significant.

LUAD cell lines were exposed to hypoxic conditions for 4, 8, and 12 hours to simulate a lactate-accumulating microenvironment. With prolonged hypoxia, the expression of glycolytic genes, including HK1, PKM2, G6PD, and LDHA (32-34), progressively increased, indirectly indicating elevated intracellular lactate levels. Moreover, ANGPTL4 expression was significantly upregulated at all time points compared with the control group (P<0.05), showing a correlation with conditions associated with lactate accumulation. These findings provide hypothesis-generating evidence for a potential association between ANGPTL4 expression and lactylation-related activity, but direct mechanistic involvement has not been experimentally established (Figure 7).

Figure 7 mRNA expression of glycolytic genes (A) HK1, (B) PKM2, (C) G6PD, and (D) LDHA in LUAD cell lines under hypoxic conditions (4, 8, and 12 h), (E) ANGPTL4, a potential lactylation-associated gene, was also assessed under the same conditions. For each gene, experiments were performed with five independent biological replicates. Asterisks denote significance: *, P<0.05; **, P<0.01; ***, P<0.001; ****, P<0.0001; ns, not significant. LUAD, lung adenocarcinoma.

Immunofluorescence staining of clinical tissue samples revealed that ANGPTL4 expression was undetectable in para-tumor lung epithelial tissues, whereas tumor epithelial cells in LUAD samples exhibited clear expression (Figure 8). Integration of the in vitro experimental results and clinical tissue analyses suggests a possible link between ANGPTL4 expression and lactylation-related activity, while mechanistic conclusions remain preliminary and require future experimental validation.

Figure 8 Immunofluorescence staining of ANGPTL4 and CK19 in paired para-tumor and tumor tissues. (A) Patient 1 (P1: para-tumor; T1: tumor); (B) Patient 2 (P2: para-tumor; T2: tumor). Nuclei (DAPI, blue), ANGPTL4 (red), and CK19 (green) are shown. Merged and enlarged views (dashed boxes) highlight the colocalization of ANGPTL4 and CK19. Statistical significance is indicated by asterisks: **, P<0.01. DAPI, 4',6-diamidino-2-phenylindole.

Discussion

Lactate, the end product of glycolysis, accumulates extensively in the TME and participates in the regulation of cellular metabolism, gene expression, and the immune landscape, thereby promoting tumor initiation and progression (35,36). In recent years, lactate-induced lysine lactylation has been identified as a novel post-translational modification broadly involved in chromatin remodeling, metabolic reprogramming, and immune evasion (12,36). However, systematic studies of lactylation in LUAD remain limited, and its cellular heterogeneity, immune regulatory mechanisms, and clinical implications have not been fully elucidated. To address this gap, a comprehensive investigation was conducted by integrating bulk and single-cell transcriptomic data with prognostic modeling, immune feature analysis, drug sensitivity prediction, and experimental validation using qRT-PCR and immunofluorescence.

At the cellular level, bioinformatic analysis revealed that epithelial cells and certain stromal cells exhibited high lactylation scores, whereas immune cells generally showed lower levels. This pattern may be attributed to enhanced glycolysis and lactate accumulation in epithelial cells, consistent with the epithelial origin of LUAD (37). Although stromal cells are less glycolytic, their exposure to a lactate-rich TME (38) may contribute to elevated lactylation. In contrast, immune cells displayed relatively lower lactylation, likely due to restricted lactate uptake mediated by the immunosuppressive effects of lactate (39). Subcluster analysis further identified multiple highly lactylated epithelial subpopulations (e.g., Epi_10, Epi_5) with high lactylation activity and CNVs, indicative of a tumor-specific phenotype potentially associated with malignant features. Similarly, certain stromal subtypes (e.g., lymphatic endothelial cells, myofibroblasts) showed high lactylation, underscoring the cellular heterogeneity of lactylation in LUAD.

For prognostic model construction, signature genes from highly lactylated epithelial cells, ligand-receptor interactions, and previously reported lactylation-related genes were integrated. Through 101 combination machine learning models, 12 consistently selected key genes were identified and used to develop a risk score system. This score significantly stratified patient survival in both the TCGA and GSE87340 cohorts, indicating its potential prognostic value. Several key genes with known functional roles in cancer showed expression patterns validated by qRT-PCR that were largely consistent with their predicted classification as risk or protective genes, further confirming the model’s reliability.

Among these genes, ANGPTL4 has been reported as a poor prognostic factor in LUAD in the context of lipid metabolism (40), Studies in other cancers suggest associations with glycolysis and lactate metabolism, including enhanced glucose uptake and lactate production (41-44), indicating that in hypoxic TMEs, ANGPTL4 may be correlated with lactate accumulation and lactylation-related activity. In our study, ANGPTL4 mRNA was upregulated in LUAD tumor tissues compared with adjacent normal tissues and increased progressively in hypoxia-treated LUAD cell lines alongside glycolytic gene upregulation. Immunofluorescence confirmed ANGPTL4 expression in tumor epithelial cells. These observations provide hypothesis-generating evidence for a potential association with lactylation, while direct mechanistic involvement remains to be validated.

Other genes included in the model also exhibited functional associations consistent with their predicted risk effects. ITGA6 has been implicated in promoting lymphangiogenesis and metastasis in LUAD (45), and targeting ITGA6 has been proposed as a potential therapeutic strategy (46). CCL20 has been reported to facilitate LUAD progression through epithelial-mesenchymal transition (47). whereas FKBP3 has been identified as a risk factor in other prognostic models (48). Collectively, these findings highlight the functional relevance of these genes and underscore the necessity for further investigation.

Immune analyses revealed correlations between lactylation-driven risk scores and features of the tumor immune microenvironment. High-risk patients showed reduced infiltration of CD4⁺ and CD8⁺ T cells, and ESTIMATE analysis indicated lower immune and stromal content, consistent with an immune-excluded phenotype. IPS and TIDE analyses further suggested enrichment of immunosuppressive components and potential immune exclusion, despite no significant changes in T cell dysfunction scores. Ligand-receptor interaction analysis indicated that highly lactylated epithelial cells may interact with T and natural killer cells through factors such as ANGPTL4 and SEMA3C. Collectively, these observations are hypothesis-generating and suggest that lactylation-associated immune modulation may involve immunosuppressive cells or stromal barriers, providing a framework for future studies on potential impacts on immunotherapy responsiveness.

Collectively, this study systematically evaluated the potential roles of lactylation in LUAD from multiple perspectives. A lactylation-related prognostic model was constructed and tested, key genes were identified, and their possible involvement in immune regulation and tumor progression was delineated. Notably, the model stratified LUAD patients into high- and low-risk groups, supporting its potential utility as an exploratory prognostic tool and a foundation for further research on clinical implications. Distinct from previous studies, potential interactions between lactylated tumor cells and immune cells were explored at the level of cell-cell communication, supporting the hypothesis that lactylation may contribute to immune escape through factors such as ANGPTL4 and SEMA3C. Several limitations should be noted. First, lactylation activity in this study was inferred indirectly from transcriptomic patterns using AUCell scores derived from lactylation-related gene sets. While this enables large-scale, cell-type resolved analysis, it represents a proxy rather than a direct measurement, as gene expression cannot fully capture post-transcriptional regulation or specific modification sites. Second, because this study is exploratory and based on integrative transcriptomic inference, we did not perform functional perturbation experiments such as ANGPTL4 knockdown; therefore, its proposed mechanistic role remains hypothesis-generating and requires future validation. Third, the lactylation-associated immune exclusion patterns are supported only by computational correlations and need experimental confirmation. Fourth, although the prognostic model performed well in the training cohort, its performance decreased in the small validation cohort (n=27), partly due to platform differences. Finally, despite validation in one external cohort, the model’s clinical utility still requires evaluation in larger prospective datasets, and the findings should be interpreted as exploratory.

Conclusions

This study provides an exploratory analysis of lactylation in LUAD and proposes a 12-gene lactylation-based prognostic model for patient stratification. Integrating single-cell transcriptomic data, immune analysis, and machine learning, we observed predicted high lactylation activity in tumor epithelial cells, which computationally correlates with immune exclusion. ANGPTL4 was identified as a potential risk gene that may be associated with glycolysis and lactate accumulation, suggesting a putative role in lactylation-related processes. Immune analyses indicated that high-risk patients could exhibit immune exclusion and increased immunosuppressive cell components, consistent with a possible contribution of lactylation to immune evasion. Overall, these findings are hypothesis-generating and provide a framework for future experimental studies to investigate lactylation-driven tumor progression and its potential relevance to LUAD prognosis and therapy.

Acknowledgments

We sincerely thank Hangzhou Astrocyte Technology Co., Ltd. for their technical guidance, and we are also grateful to the researchers and participants whose contributions made the public datasets used in this study possible.

Footnote

Reporting Checklist: The authors have completed the TRIPOD and MDAR reporting checklists. Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-aw-1170/rc

Data Sharing Statement: Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-aw-1170/dss

Peer Review File: Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-aw-1170/prf

Funding: This work was supported by the Hubei Chen Xiaoping Science and Technology Development Foundation (grant No. CXPJJH123001-2335).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-aw-1170/coif). All authors report that technical support was provided by Hangzhou Astrocyte Technology Co., Ltd. for this study. The authors have no other conflicts of interest to declare.

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. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The protocol was reviewed and approved by the Ethics Committee of Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology (approval No. TJ-IRB20220329). Written informed consent was obtained from all participants prior to sample collection.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Luo G, Zhang Y, Rumgay H, et al. Estimated worldwide variation and trends in incidence of lung cancer by histological subtype in 2022 and over time: a population-based study. Lancet Respir Med 2025;13:348-63. [Crossref] [PubMed]
2. He S, Li H, Cao M, et al. Survival of 7,311 lung cancer patients by pathological stage and histological classification: a multicenter hospital-based study in China. Transl Lung Cancer Res 2022;11:1591-605. [Crossref] [PubMed]
3. Xing P, Wang S, Cao Y, et al. Treatment strategies and drug resistance mechanisms in adenocarcinoma of different organs. Drug Resist Updat 2023;71:101002. [Crossref] [PubMed]
4. Banerjee A, Vathiotis I, Halder D, et al. Molecular Landscape and Therapeutic Strategies of Lung Cancer Lineage Plasticity. J Thorac Oncol 2025;20:1582-93. [Crossref] [PubMed]
5. Shen Y, Chen JQ, Li XP. Differences between lung adenocarcinoma and lung squamous cell carcinoma: Driver genes, therapeutic targets, and clinical efficacy. Genes Dis 2025;12:101374. [Crossref] [PubMed]
6. Kwok HH, Yang J, Lam DC. Breaking the Invisible Barriers: Unleashing the Full Potential of Immune Checkpoint Inhibitors in Oncogene-Driven Lung Adenocarcinoma. Cancers (Basel) 2023;15:2749. [Crossref] [PubMed]
7. Han M, Wan F, Xiao B, et al. Cell components of tumor microenvironment in lung adenocarcinoma: Promising targets for small-molecule compounds. Chin Med J (Engl) 2025;138:905-15. [Crossref] [PubMed]
8. Spella M, Stathopoulos GT. Immune Resistance in Lung Adenocarcinoma. Cancers (Basel) 2021;13:384. [Crossref] [PubMed]
9. Vokes NI, Pan K, Le X. Efficacy of immunotherapy in oncogene-driven non-small-cell lung cancer. Ther Adv Med Oncol 2023;15:17588359231161409. [Crossref] [PubMed]
10. Hu XT, Wu XF, Xu JY, et al. Lactate-mediated lactylation in human health and diseases: Progress and remaining challenges. J Adv Res 2025;75:229-48. [Crossref] [PubMed]
11. Li F, Si W, Xia L, et al. Positive feedback regulation between glycolysis and histone lactylation drives oncogenesis in pancreatic ductal adenocarcinoma. Mol Cancer 2024;23:90. [Crossref] [PubMed]
12. Chen J, Huang Z, Chen Y, et al. Lactate and lactylation in cancer. Signal Transduct Target Ther 2025;10:38. [Crossref] [PubMed]
13. Hou X, Hong Z, Zeng H, et al. Lactylation in cancer biology: Unlocking new avenues for research and therapy. Cancer Commun (Lond) 2025;45:1367-406. [Crossref] [PubMed]
14. Li W, Zhou C, Yu L, et al. Tumor-derived lactate promotes resistance to bevacizumab treatment by facilitating autophagy enhancer protein RUBCNL expression through histone H3 lysine 18 lactylation (H3K18la) in colorectal cancer. Autophagy 2024;20:114-30. [Crossref] [PubMed]
15. Mao Y, Zhang J, Zhou Q, et al. Hypoxia induces mitochondrial protein lactylation to limit oxidative phosphorylation. Cell Res 2024;34:13-30. [Crossref] [PubMed]
16. De Leo A, Ugolini A, Yu X, et al. Glucose-driven histone lactylation promotes the immunosuppressive activity of monocyte-derived macrophages in glioblastoma. Immunity 2024;57:1105-1123.e8. [Crossref] [PubMed]
17. Raychaudhuri D, Singh P, Chakraborty B, et al. Histone lactylation drives CD8(+) T cell metabolism and function. Nat Immunol 2024;25:2140-51. [Crossref] [PubMed]
18. Xiong J, He J, Zhu J, et al. Lactylation-driven METTL3-mediated RNA m(6)A modification promotes immunosuppression of tumor-infiltrating myeloid cells. Mol Cell 2022;82:1660-1677.e10. [Crossref] [PubMed]
19. Huang H, Chen K, Zhu Y, et al. A multi-dimensional approach to unravel the intricacies of lactylation related signature for prognostic and therapeutic insight in colorectal cancer. J Transl Med 2024;22:211. [Crossref] [PubMed]
20. Liu C, Ni C, Li C, et al. Lactate-related gene signatures as prognostic predictors and comprehensive analysis of immune profiles in nasopharyngeal carcinoma. J Transl Med 2024;22:1116. [Crossref] [PubMed]
21. Sun Z, Wang L, Eckloff BW, et al. Conserved recurrent gene mutations correlate with pathway deregulation and clinical outcomes of lung adenocarcinoma in never-smokers. BMC Med Genomics 2014;7:32. [Crossref] [PubMed]
22. Kim N, Kim HK, Lee K, et al. Single-cell RNA sequencing demonstrates the molecular and cellular reprogramming of metastatic lung adenocarcinoma. Nat Commun 2020;11:2285. [Crossref] [PubMed]
23. Cheng Z, Huang H, Li M, et al. Lactylation-Related Gene Signature Effectively Predicts Prognosis and Treatment Responsiveness in Hepatocellular Carcinoma. Pharmaceuticals (Basel) 2023;16:644. [Crossref] [PubMed]
24. Wan N, Wang N, Yu S, et al. Cyclic immonium ion of lactyllysine reveals widespread lactylation in the human proteome. Nat Methods 2022;19:854-64. [Crossref] [PubMed]
25. Gavish A, Tyler M, Greenwald AC, et al. Hallmarks of transcriptional intratumour heterogeneity across a thousand tumours. Nature 2023;618:598-606. [Crossref] [PubMed]
26. Hao Y, Stuart T, Kowalski MH, et al. Dictionary learning for integrative, multimodal and scalable single-cell analysis. Nat Biotechnol 2024;42:293-304. [Crossref] [PubMed]
27. Hu C, Li T, Xu Y, et al. CellMarker 2.0: an updated database of manually curated cell markers in human/mouse and web tools based on scRNA-seq data. Nucleic Acids Res 2023;51:D870-6. [Crossref] [PubMed]
28. Aibar S, González-Blas CB, Moerman T, et al. SCENIC: single-cell regulatory network inference and clustering. Nat Methods 2017;14:1083-6. [Crossref] [PubMed]
29. Jin S, Guerrero-Juarez CF, Zhang L, et al. Inference and analysis of cell-cell communication using CellChat. Nat Commun 2021;12:1088. [Crossref] [PubMed]
30. Collins GS, Reitsma JB, Altman DG, et al. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD Statement. BMC Med 2015;13:1. [Crossref] [PubMed]
31. Liu H, Zhang W, Zhang Y, et al. Mime: A flexible machine-learning framework to construct and visualize models for clinical characteristics prediction and feature selection. Comput Struct Biotechnol J 2024;23:2798-810. [Crossref] [PubMed]
32. Sharma D, Singh M, Rani R. Role of LDH in tumor glycolysis: Regulation of LDHA by small molecules for cancer therapeutics. Semin Cancer Biol 2022;87:184-95. [Crossref] [PubMed]
33. Jiang P, Du W, Wang X, et al. p53 regulates biosynthesis through direct inactivation of glucose-6-phosphate dehydrogenase. Nat Cell Biol 2011;13:310-6. [Crossref] [PubMed]
34. Liao M, Yao D, Wu L, et al. Targeting the Warburg effect: A revisited perspective from molecular mechanisms to traditional and innovative therapeutic strategies in cancer. Acta Pharm Sin B 2024;14:953-1008. [Crossref] [PubMed]
35. Chen S, Xu Y, Zhuo W, et al. The emerging role of lactate in tumor microenvironment and its clinical relevance. Cancer Lett 2024;590:216837. [Crossref] [PubMed]
36. Li X, Yang Y, Zhang B, et al. Lactate metabolism in human health and disease. Signal Transduct Target Ther 2022;7:305. [Crossref] [PubMed]
37. Gardner EE, Earlie EM, Li K, et al. Lineage-specific intolerance to oncogenic drivers restricts histological transformation. Science 2024;383:eadj1415. [Crossref] [PubMed]
38. Sinjab A, Han G, Treekitkarnmongkol W, et al. Resolving the Spatial and Cellular Architecture of Lung Adenocarcinoma by Multiregion Single-Cell Sequencing. Cancer Discov 2021;11:2506-23. [Crossref] [PubMed]
39. Caslin HL, Abebayehu D, Pinette JA, et al. Lactate Is a Metabolic Mediator That Shapes Immune Cell Fate and Function. Front Physiol 2021;12:688485. [Crossref] [PubMed]
40. Shi X, Li R, Dong X, et al. IRGS: an immune-related gene classifier for lung adenocarcinoma prognosis. J Transl Med 2020;18:55. [Crossref] [PubMed]
41. Zhu C, Teng L, Lai Y, et al. Adipose-derived stem cells promote glycolysis and peritoneal metastasis via TGF-β1/SMAD3/ANGPTL4 axis in colorectal cancer. Cell Mol Life Sci 2024;81:189. [Crossref] [PubMed]
42. Yin D, Fang N, Zhu Y, et al. Adipocytes-induced ANGPTL4/KLF4 axis drives glycolysis and metastasis in triple-negative breast cancer. J Exp Clin Cancer Res 2025;44:192. [Crossref] [PubMed]
43. Zheng X, Liu R, Zhou C, et al. ANGPTL4-Mediated Promotion of Glycolysis Facilitates the Colonization of Fusobacterium nucleatum in Colorectal Cancer. Cancer Res 2021;81:6157-70. [Crossref] [PubMed]
44. Fekir K, Dubois-Pot-Schneider H, Désert R, et al. Retrodifferentiation of Human Tumor Hepatocytes to Stem Cells Leads to Metabolic Reprogramming and Chemoresistance. Cancer Res 2019;79:1869-83. [Crossref] [PubMed]
45. Ren S, Wang J, Xu A, et al. Integrin α6 overexpression promotes lymphangiogenesis and lymphatic metastasis via activating the NF-κB signaling pathway in lung adenocarcinoma. Cell Oncol (Dordr) 2022;45:57-67. [Crossref] [PubMed]
46. Gao W, Wang Q, Li S, et al. Promising therapeutic efficacy and safety of a novel integrin α6-targeting peptide-drug conjugate in lung adenocarcinoma. Mol Cancer 2025;24:190. [Crossref] [PubMed]
47. Fan T, Li S, Xiao C, et al. CCL20 promotes lung adenocarcinoma progression by driving epithelial-mesenchymal transition. Int J Biol Sci 2022;18:4275-88. [Crossref] [PubMed]
48. Zhai Y, Zhao B, Wang Y, et al. Construction of the optimization prognostic model based on differentially expressed immune genes of lung adenocarcinoma. BMC Cancer 2021;21:213. [Crossref] [PubMed]