Comprehensive analysis of gene signature linking hypoxia and lactylation for predicting prognosis and immunotherapy response in patients with gastric cancer

Introduction

Gastric cancer (GC) ranks as the fifth most commonly diagnosed malignancy worldwide and represents the third leading cause of cancer-related mortality globally (1). In 2022, there were over 968,000 new cases of GC, with nearly 660,000 deaths attributed to the disease (2). The uncertainty regarding the etiology of tumors and the late diagnosis of malignancies contribute to elevated mortality rates (3). Historically, surgical resection followed by adjuvant chemotherapy was a cornerstone of treatment. However, the paradigm has now evolved, with perioperative chemotherapy emerging as the current standard of care for resectable locally advanced disease (4). Nevertheless, there is an urgent need to explore innovative treatment strategies for GC from a novel perspective.

The tumor microenvironment (TME) is the internal environment in which tumor cells are generated and survive. The interaction of components in the TME promotes or inhibits tumor progression (5). Hypoxia, a hallmark of cancer, serves as a primary source of the metabolic byproduct lactate. Previous studies have established a significant association between hypoxia and poor prognosis in GC (6). Hypoxia activates hypoxia-inducible factor-1 (HIF-1) and upregulates pro-angiogenic factors, such as vascular endothelial growth factor (VEGF), thereby facilitating nutrient supply to the tumor and fostering a proliferative microenvironment (7). Under hypoxic conditions, tumor cells undergo a metabolic shift from oxidative phosphorylation to aerobic glycolysis, resulting in increased adenosine triphosphate (ATP) production and lactate accumulation. Notably, lactate functions as a key substrate for protein lactylation, a recently identified post-translational modification. Moreover, hypoxia-driven lactate accumulation remodels the extracellular matrix (ECM) by modulating enzymes, such as matrix metalloproteinases (MMPs), thereby promoting tumor invasiveness and metastatic potential (8,9). Substantial evidence indicates that lactate accumulation in the TME can promote protein lactylation levels (10-12). The elevated levels of lactylation modification activate the transcriptional programs, ultimately increasing the malignant behavior of tumor cells (13). However, most studies focus solely on hypoxia or lactylation, and research analyzing the prognosis and treatment of GC patients by combining the characteristics of hypoxia and lactylation genes is limited.

In this study, we integrated single-cell RNA sequencing (scRNA-seq) and spatial transcriptome (ST) data to reveal the histological structure of GC. A novel hypoxia- and lactylation-related gene (HALRG) -based risk scoring model was developed using The Cancer Genome Atlas (TCGA) database and subsequently validated with Gene Expression Omnibus (GEO) datasets. Furthermore, we also analyzed the differences in the TME and treatment responses between the high- and low-risk groups. We present this article in accordance with the TRIPOD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1453/rc).

Methods

Data download

We downloaded scRNA-seq data for 26 patients from GSE183904 and ST data from GSE203612, both obtained from the GEO database (http://www.ncbi.nlm.nih.gov/geo/). The RNA expression profiles and corresponding clinical information for the TCGA-stomach adenocarcinoma (STAD) project were retrieved from the TCGA database (https://portal.gdc.cancer.gov/), comprising 36 normal tissues and 412 GC tissues. The validation dataset consisted of 200 GC samples from the GSE15459 project within the GEO database. Besides, gene sets associated with hypoxia and lactylation were obtained from the Gene Set Enrichment Analysis (GSEA) database, as detailed in table available at https://cdn.amegroups.cn/static/public/tcr-2025-1453-1.xlsx. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Single-cell transcriptome data processing

To ensure data quality, all analyses in our study were conducted using R software (version 4.4.2) (14). We used the “CreateSeuratObject” function to convert the GC samples into a Seurat object. Low-quality cells with fewer than 300 or more than 5,000 expressed genes, or those in which more than 20% of the mitochondrial gene count was removed. The “NormalizeData” function was used for data normalization, and the “FindVariableFeatures” function was employed to select 2,000 highly variable genes for subsequent analysis. Then, the data were scaled and analyzed using principal component analysis (PCA). Cell clustering was performed using the “FindNeighbors” and “FindClusters” functions, with the resolution parameter set to 0.4. The Uniform Manifold Approximation and Projection (UMAP) method was used to visualize the data.

Epithelial cells analysis in GC

In order to distinguish malignant cells from normal epithelial cells, we employed the “inferCNV” R package (version 1.22.0) to calculate the initial copy number variation (CNV) levels of each region. The inferCNV package compared the gene expression of each tumor cell to reference gene expression from T cells and endothelial cells. Then, epithelial cell subclusters exhibiting a significant CNV spectrum were classified as malignant cells. Malignant epithelial cells were visualized using t-distributed stochastic neighbor embedding (t-SNE). Additionally, normal gastric epithelial cells were clustered and annotated into distinct subtypes by screening the literature (15-17).

Cell-cell communication analysis

The CellChat R software package combines gene expression data with established knowledge of signaling ligand-receptor interactions and their cofactors to model potential communication between cells (18). To investigate potential cell-cell communication between epithelial cells and other cell types, ligand-receptor interactions across different cells were analyzed using the “CellChat” package. The “aggregateNet” function was utilized to construct cell communication networks.

Pseudotime trajectories analysis

We employed the Monocle package to analyze the pseudo-time trajectories of malignant epithelial cells (19). We constructed a Monocle object using the “newCellDataSet” function, with the parameter expressionFamily = negbinomial.size. Prior to trajectory inference, genes were filtered to retain only those with a mean expression ≥0.1. Dimensionality reduction was performed through the “reduceDimension” function, using the parameters method = “DDRTree” and max_components =2. Finally, cell developmental trajectories were visualized using the “plot_cell_trajectory” function.

ST analysis

We utilized the “Load10X_Spatial” function to import ST data and normalized the data using the SCTransform algorithm. For deconvolution analysis, we applied the Conditional Auto-Regressive Deconvolution (CARD) algorithm, which integrates single-cell sequencing data to predict the cell types at each spatial location (20). The CARD package (version 1.1) was subsequently used to visualize these cell type distributions within the spatial dataset.

Development and validation of a prognostic signature based on HALRGs

First, differentially expressed genes (DEGs) were identified in the TCGA training cohort using the “DESeq2” package, applying a significance threshold of P value <0.05 and |log₂fold change (log₂FC)| >1. Subsequently, univariate Cox regression analysis was conducted on these DEGs to identify prognostic genes. We conducted the least absolute shrinkage and selection operator (LASSO) regression algorithm to select the hub genes and develop a prognostic signature. These HALRGs were then analyzed through multivariate Cox regression. The HALRG prognostic model was constructed using the formula: risk score = (coef1 × gene1 exp) + (coef2 × gene2 exp) + ... + (coefN × geneN exp), where exp denotes the expression levels of the genes, and coef represents the coefficients of the genes calculated by multivariate Cox regression. The TCGA cohort was stratified into high-risk and low-risk groups using the median risk score. Besides, we utilized the GSE15459 dataset to validate the reliability of the risk model. The “timeROC” package was used to generate Kaplan-Meier survival curves and receiver operating characteristic (ROC) curves. The corresponding area under the curve (AUC) was employed to evaluate the predictive ability of the prognostic signatures. Univariate and multivariate analyses were performed to assess the independent prognostic significance of the risk model.

Construction and evaluation of a nomogram

We utilized the “survival” package (version 3.8.3) and “rms” package (version 7.0.0) in R to develop a prognostic nomogram incorporating age, gender, stage, and risk score for predicting survival outcomes. Calibration curves were plotted at 1-, 3-, and 5-year intervals to evaluate the predictive accuracy of the nomogram model.

Assessment of treatment response

CIBERSORT is a computational algorithm for estimating the abundance of immune cells based on gene expression profiles (21). In this study, we used the CIBERSORT R package to analyze the distribution of 22 immune cell types in the TCGA-STAD cohort. Besides, we utilized the “ESTIMATE” software package to estimate stromal and immune cells in malignant tumors, calculating the immune score, stromal score, ESTIMATE score, and tumor purity (22). Meanwhile, the Tumor Immune Dysfunction and Exclusion (TIDE) score was determined using the TIDE algorithm to assess the possibility of immune escape. To evaluate chemotherapy sensitivity, the “oncopredict” software was employed to compare the differential sensitivity to chemotherapy agents between high- and low-risk patient cohorts.

Statistical analysis

LASSO and Cox regression analyses were employed to develop a HALRG prognostic model for GC. The Wilcoxon rank-sum test was used to assess the relationship between the two groups. AUC of ROC was utilized to evaluate diagnostic performance and predictive capability. Statistical analyses were conducted using R software version 4.4.2, with statistical significance set at P<0.05.

Results

Annotation of cell types

A total of 26 GC samples were screened from GSE183904 in this study (Figure 1A). After filtering out low-quality cells and normalizing the data, we used the PCA method for dimensionality reduction. Then, 104,150 cells were divided into 19 clusters using the UMAP algorithm (Figure 1B). Based on the annotations in the literature, these clusters were further divided into seven cell types, including T cells (clusters 0, 1, 8, 15), B cells (clusters 2, 6, 16), endothelial cells (clusters 7, 18), epithelial cells (clusters 3, 4, 17), mast cells (cluster 11), macrophages (clusters 5, 10, 12), and fibroblasts (clusters 9, 13, 14) (Figure 1C). The bubble plot illustrated the marker genes unique to each cell type (Figure 1D). Figure 1E depicted the expression patterns of these marker genes across different cell types.

Figure 1 Integration and clustering of GC scRNA-seq data. (A) The UMAP algorithm visualized 26 GC samples. (B) UMAP of the 19 cell clusters. (C) UMAP of the seven different cell types. (D) Dot plot showing the expression of marker genes for each cell subtype. (E) Feature plots showing the expression of marker genes across the nine cell types. GC, gastric cancer; scRNA-seq, single-cell RNA sequencing; UMAP, Uniform Manifold Approximation and Projection.

Inference of CNVs (InferCNV) analysis of epithelial cells

To explore the epithelial characteristics of the tumor, we identified and extracted transcriptional profiles of 17,670 epithelial cells from scRNA-seq data. The epithelial cells were re-clustered and divided into 12 clusters (Figure 2A). We employed the “inferCNV” package to distinguish malignant tumor cells from normal cells through genome-wide CNV analysis, with normal endothelial cells and T cells serving as reference cells (Figure 2B). The results revealed that subpopulations of epithelial cells, specifically clusters 0, 2, 3, 6, 7, 8, and 11, displayed notable amplifications in CNV, indicating their malignant characteristics (Figure 2C). A total of 10,994 epithelial cells were classified as tumor epithelial cells. The remaining non-malignant epithelial cells were subjected to subclustering, displaying seven distinct subclusters of normal epithelial cells (Figure 2D). Figure 2E presented the subclusters of normal epithelial cells, including pit mucous cells (PMCs; marked with TFF1 and MUC5AC), PMC-like cells (marked with SOX4), gland mucous cells (GMCs; marked with PGC and MUC6), enteroendocrine cells (marked with CHGA), and intestinal cells (marked with REG4).

Figure 2 Overview of epithelial cell lineage. (A) UMAP of the 12 cell clusters. (B) CNV heatmap of epithelial cell subtypes. (C) Box plot displaying CNV scores for epithelial cell subtypes. (D) The t-SNE algorithm visualized five cell clusters of malignant epithelial cells. (E) t-SNE of the nonmalignant epithelial cells. C, cluster; CNV, copy number variation; GMC, gland mucous cell; InferCNV, inference of copy number variations; PMC, pit mucous cell; T, tumor; t-SNE, t-distributed stochastic neighbor embedding; UMAP, Uniform Manifold Approximation and Projection.

Pseudotime analysis and cell-cell communication analysis

We conducted pseudotime analysis and cell-cell communication analysis to further investigate the characteristics of malignant cells. Four distinct subpopulations of malignant epithelial cells were identified. Trajectory analysis using the Monocle 2 package revealed that these cells could be categorized into five differentiation states (Figure 3A). Figure 3B showed that state 1 malignant cells represented the initiation of differentiation, while states 4 and 5 corresponded to the terminal stages of differentiation. Furthermore, CellChat analysis demonstrated that epithelial cells exhibited robust communication with T cells and displayed a high interaction frequency across diverse cell populations (Figure 3C). The results revealed that ligand-receptor-mediated cellular interactions were predominantly present in the MIF signaling pathway (Figure 3D,3E). Notably, the MIF-CD74 + CXCR4 ligand-receptor pair contributed most significantly to this pathway, underscoring its pivotal role in mediating intercellular communication (Figure 3F).

Figure 3 Pseudotime analysis and cell-cell communication analysis. (A) Distribution of subtypes and states in malignant epithelial cells in the trajectory curve. (B) Display of developmental time in malignant epithelial cells. (C) The number and weights of cell-cell communication between cell types. (D) Cell-cell communication network of MIF signaling pathway. (E) Bubble diagram illustrating ligand-receptor interactions between epithelial cells and other cell types. (F) Cell-cell communication analysis identified that MIF-CD74 + CXCR4 play essential roles. Commun prob., communication probability.

ST analysis

Spatial information plays a pivotal role in understanding tumor biology. In this study, ST data obtained from GEO were processed using the “Seurat” package, followed by dimensionality reduction with the UMAP algorithm (Figure 4A). We performed systematic analysis and integration of ST data with scRNA-seq datasets, enabling the spatial mapping of seven distinct cell types identified from scRNA-seq onto GC tissue sections. Spatial pie chart visualization revealed the proportional distribution of these seven cell types, with epithelial cells emerging as the predominant type (Figure 4B). Additionally, the “CARD” package was utilized to generate high-resolution spatial maps, providing detailed insights into the tissue distribution of these cell types (Figure 4C).

Figure 4 ST analysis in GC. (A) UMAP of dimensionality reduction in ST data and spatial distribution map for the clusters. (B) Pie plot displaying the proportions of the cells in GC tissue. (C) Spatial distribution of seven cell types. GC, gastric cancer; ST, spatial transcriptome; UMAP, Uniform Manifold Approximation and Projection.

Construction of HALRG risk model

To investigate the characteristics of HALRGs in GC, we conducted messenger RNA (mRNA) expression analysis on a training cohort consisting of TCGA-STAD samples. Using the R package DESeq2, we identified 6,432 upregulated and 3,343 downregulated DEGs (Figure 5A). The DEGs between the tumor and normal groups were visualized in a heatmap (Figure 5B). A total of 522 HALRGs were extracted from the GSEA database. Venn diagram analysis revealed 81 differentially expressed HALRGs at the intersection of the DEGs and HALRGs (Figure 5C). Subsequently, univariate Cox regression analysis identified 31 potential prognostic genes (Figure 5D). LASSO and multivariable Cox regression analyses were further employed to identify three prognostic genes: SERPINE1, IGFBP1, and DTNA (Figure 5E). The risk score was calculated using the regression coefficients as follows: risk score = (0.15888533 × SERPINE1 expression) + (0.06842331 × IGFBP1 expression) + (0.10684378 × DTNA expression). Based on the median risk score, the training cohort was divided into high- and low-risk groups. In addition, Kaplan-Meier survival analysis revealed that the high-risk group exhibited a significantly lower survival rate compared to the low-risk group (Figure 5F). The AUC values for the risk model at 1, 3, and 5 years were 0.668, 0.653, and 0.687, respectively, indicating the potential predictive efficacy of the risk model (Figure 5G). However, the AUC values are modest, limiting their use for clinical applicability.

Figure 5 Construction of the HALRG risk model. (A) The volcano plot displaying DEGs in the TCGA-STAD dataset. Upregulated genes are shown as red dots, while downregulated genes are depicted as blue dots. (B) The heatmap showing the expression levels of the DEGs. (C) The Venn diagram showing the quantities of the DEGs and HALRGs. (D) Univariate Cox analysis of HALRGs-related DEGs. (E) LASSO-Cox analysis in the training cohort. (F) Kaplan-Meier survival curves in the training cohort. (G) ROC curves showing the predictive efficacy of the HALRG risk model. AUC, area under the curve; DEG, differentially expressed gene; HALRG, hypoxia- and lactylation-related gene; HR, hazard ratio; N, normal; ROC, receiver operating characteristic; STAD, stomach adenocarcinoma; T, tumor; TCGA, The Cancer Genome Atlas.

Development of the nomogram and validation of the prognostic model

Through integrating risk score and clinical information, we developed a nomogram to predict the overall survival (OS) at 1, 3, and 5 years for GC patients (Figure 6A). Calibration curves demonstrated excellent agreement between nomogram-predicted survival probabilities and the actual observed survival rates (Figure 6B). Additionally, the external validation cohort GSE87211 was conducted to determine the strong predictive ability of the risk model. Kaplan-Meier survival analysis revealed significantly reduced survival rates in the high-risk group, and the ROC curve further evaluated the predictive accuracy of the model (Figure 6C,6D). Figure 6E illustrated the calibration curves for 1, 3, and 5 years in the validation cohort. Both univariate and multivariate Cox regression analyses confirmed that the risk score maintained significant independent prognostic predictive capability (P<0.001) compared to age, gender, and stage (Figure 6F).

Figure 6 The prognostic value and validation of the HALRG risk model. (A) The nomogram integrating the risk score with clinical information. (B) Calibration curves for 1, 3, and 5 years in the training cohort. (C) Kaplan-Meier survival curves in the validation cohort. (D) ROC curves in the validation cohort. (E) Calibration curves for 1, 3, and 5 years in the validation cohort. (F) Univariate and multivariate Cox regression analyses of clinical information and risk scores. AUC, area under the curve; CI, confidence interval; HALRG, hypoxia- and lactylation-related gene; M, metastasis; N, node; ROC, receiver operating characteristic; T, tumor.

Tumor immune infiltration and drug sensitivity analysis

Based on the TCGA-STAD dataset, we employed the CIBERSORT algorithm to evaluate the infiltration landscape of 22 immune cell types in GC and constructed immune cell profiles (Figure 7A). Figure 7B illustrated the differences in the immune microenvironment between high-risk and low-risk groups. Subsequently, we used the ESTIMATE algorithm to analyze the TME. The results indicated that the high-risk group exhibited significantly elevated ESTIMATE scores and stromal scores compared to the low-risk group, along with reduced tumor purity. However, there was no significant difference in immune scores between the two risk groups (Figure 7C). Notably, immune scores did not differ significantly between the two risk groups, indicating comparable levels of infiltrating immune cells. This suggests that the distinct TME in the high-risk group is defined more by reactive stroma than immune infiltration. In terms of TME scores, the high-risk group demonstrated significantly higher TIDE scores, dysfunction scores, and exclusion scores, suggesting that immune evasion is more common (Figure 7D). Furthermore, we analyzed the drug sensitivity in GC. The results showed that high-risk patients had higher sensitivity to chemotherapy drugs oxaliplatin and docetaxel, whereas low-risk patients showed higher sensitivity to vinblastine and irinotecan (Figure 7E).

Figure 7 Tumor immune infiltration and drug sensitivity analysis in different risk groups. (A) Twenty-two immune cell profiles analyzed by the CIBERSORT method. (B) Comparison of immune infiltration between high- and low-risk groups. (C) Box plots showing the proportion of ESTIMATEScore, StromalScore, ImmuneScore, and TumorPurity in GC tissues. (D) Box plots comparing the TIDE, dysfunction, exclusion, and MDSC between high- and low-risk groups. (E) Predicted responsiveness to drugs in GC. *, P<0.05; **, P<0.01; ***, P<0.001; ****, P<0.0001; ns, not significant (P>0.05). GC, gastric cancer; MDSC, myeloid-derived suppressor cell; NK, natural killer; TIDE, Tumor Immune Dysfunction and Exclusion; TME, tumor microenvironment.

Discussion

Cancer is widely recognized as a metabolic disorder. In hypoxic environments, malignant cells enhance the glycolytic process (Warburg effect) to accelerate lactate production, which provides essential energy for cancer cells (8). There is evidence to suggest that circulatory lactate not only functions as an energy substrate but also serves as a metabolic precursor, facilitating glucose liberation to support critical tumorigenic processes (23). A groundbreaking discovery reveals lactylation, a novel post-translational modification mechanism, that plays a pivotal role in regulating tumor metabolism and malignant behaviors under hypoxia. Sun et al. demonstrated that lactylation-mediated modulation of METTL16 protein enhances m6A modification of FDX1 mRNA, thereby inducing copper death to inhibit tumor progression (24). Concurrently, AARS1-driven lactylation of p53 promotes the onset of GC (25). The hypoxic TME in GC orchestrates multifaceted oncogenic processes, including therapeutic resistance and immunosuppressive TME remodeling, through enhanced infiltration of regulatory T cells (Tregs) and tumor-associated macrophages (TAMs), which collectively undermine antitumor immunity (26). Therefore, hypoxia and lactylation provide valuable insights into underlying tumorigenic mechanisms and serve as promising therapeutic targets for GC treatment.

This study used transcriptome sequencing data and clinical prognostic information from TCGA and GEO databases to explore the regulatory roles of hypoxia and lactate in the initiation and progression of GC, as well as their potential value as prognostic biomarkers. We developed a prognostic model using LASSO Cox analysis with three HALRGs, including SERPINE1, IGFBP1, and DTNA. Mechanistically, SERPINE1 is a member of the serine protease inhibitor superfamily. Clinical studies have demonstrated that downregulating SERPINE1 expression significantly diminishes the proliferation, invasion, and metastatic properties of GC cells, while simultaneously promoting apoptosis (27). Another study showed that SERPINE1 activates the JAK2/STAT3 signaling pathway, stimulating let-7g-5p-enriched exosome secretion from cancer cells to induce M2 macrophage polarization, reshaping the immunosuppressive TME and accelerating tumor progression and metastasis (28). IGFBP1 plays a critical role in various physiological and pathological processes, including embryonic development, metabolic regulation, and tumor growth and metastasis. Furthermore, IGFBP1 promotes cellular survival and facilitates metastasis in GC by inhibiting AKT1-mediated phosphorylation of SOD2 and enhancing the activity of antioxidant enzymes, thereby attenuating the accumulation of mitochondrial reactive oxygen species (ROS) (29). DTNA, a scaffold protein of the dystrophin-associated protein complex (DAPC), has been shown to be associated with connecting the ECM and the muscle cell cytoskeleton, and with maintaining the structural integrity of muscle cell membranes (30). Besides, the interaction between DTNA and STAT3 induces STAT3 phosphorylation, upregulating pro-fibrogenic factors like TGF-β1 and suppressing the tumor-suppressive function of p53, thereby accelerating the progression from hepatic fibrosis to hepatocellular carcinoma (31). Zhang et al. identified DTNA in the diffuse subtype of GC, where it serves as a marker of poor prognosis and may contribute to pathogenesis, particularly in this subtype (32). In conclusion, the three target genes identified exhibit hypoxia and lactylation modification sites and play crucial regulatory roles in pathological processes associated with GC, including proliferation and invasion.

This study investigated the biological significance of the TME through an integrated analysis of single-cell transcriptomic and ST data. The application of scRNA-seq enabled us to cluster 104,150 cells into seven major cell types. Notably, epithelial cells—the primary drivers of malignancy—were further characterized using InferCNV analysis of chromosomal CNVs, which enabled discrimination between malignant cells and normal epithelial subtypes, revealing potential origins of intratumoral heterogeneity. ST complemented these findings by mapping the spatial distribution of cell types within tumor tissues. To analyze the evolutionary trajectory of malignant epithelial cells, we employed the Monocle 2 algorithm to construct pseudotemporal developmental trajectories, demonstrating the dynamic changes of epithelial cells in different states. Cell-cell communication network analysis using the CellChat algorithm identified epithelial cell crosstalk as a pivotal axis in GC progression. Based on these discoveries, we innovatively established a HALRG risk model that stratified the TCGA-STAD cohort into high- and low-risk groups. Subsequent survival analysis showed that the high-risk group was associated with poor survival rates. It has been reported that hypoxia and lactate-related genes are associated with poor prognosis in several cancers, such as hepatocellular carcinoma (33), melanoma (34), and triple-negative breast cancer (35). ROC curve analysis validated the prognostic efficacy of the HALRG model in GC patients, with consistent validation in GEO cohorts. We further developed a multivariate nomogram that integrates HALRG scores with tumor-node-metastasis (TNM) staging and age parameters. Calibration curves demonstrated a high concordance between predicted and observed survival probabilities, reinforcing the prognostic utility of the HALRG model. Both univariate and multivariate Cox analyses confirmed that the risk score is an independent prognostic factor for GC.

The TME comprises a complex network of non-cancerous cells, stromal components, and metabolic byproducts. It significantly influences the initiation and progression of GC by mediating immunosuppressive microenvironments (36). In terms of immune cell infiltration, the high-risk group exhibited significant reductions in CD8⁺ T cells, T helper cells, Tregs, and M1 macrophages. The analysis of TME revealed elevated ESTIMATE scores and stromal scores in the high-risk group, while lower tumor purity. The TIDE score demonstrated higher TIDE, dysfunction, and exclusion scores in the high-risk group. These findings suggest that patients with high HALRG scores benefit less from immunotherapy. We hypothesize that hypoxia and lactylation may drive immune cell depletion, thereby promoting immune evasion. Previous studies have indicated that hypoxia induces tumor cells to secrete immunosuppressive factors such as IL-6 and TGF-β while recruiting Tregs to establish an immunosuppressive microenvironment (37). Additionally, hypoxia disrupts mitochondrial function, such as by transferring mutated mitochondria to T cells, impairing their energy metabolism and inducing functional deficits, which exacerbates immune escape (38). Lactate suppresses antitumor immunity by upregulating transporters like MCT1 to enhance lactate uptake in exhausted T cells, inhibiting their cytotoxic function (39). Hypoxia is closely related to the immune microenvironment and can promote the progression of GC. Drug sensitivity analyses revealed that high-risk patients exhibited higher sensitivity to chemotherapeutic agents such as oxaliplatin and docetaxel, but poorer outcomes with vinblastine and irinotecan. However, it should be noted that these drug sensitivity findings remain predictive in nature and require further validation in GC cell lines or patient-derived models. By focusing on hypoxia- and lactylation-mediated mechanisms in GC, this study provides novel therapeutic insights and directions for patient stratification.

Limitations

Although our study provides a comprehensive multi-omics landscape of hypoxia and lactylation in GC, several limitations warrant consideration. First, the HALRG model necessitates prospective validation in clinical trials to evaluate its real-world clinical utility. Second, the mechanisms underlying hypoxia and lactylation in immune evasion require experimental validation using both in vitro and in vivo models. Third, the ST data are limited to a small subset of cohorts, necessitating larger-scale studies to confirm regional TME patterns.

Conclusions

In conclusion, we have elucidated the heterogeneity of epithelial cell differentiation trajectories and intercellular communication in GC by integrating extensive RNA-seq data and ST profiles from public databases. We identified three HALRGs—SERPINE1, IGFBP1, and DTNA—that reveal their multifaceted roles in GC progression and have innovatively constructed a HALRG prognostic model that demonstrates satisfactory prognostic value. The clinical utility was further enhanced by incorporating the HALRG risk score, TNM staging, and age into a nomogram, enabling individualized risk stratification. Furthermore, the HALRG signature showed significant correlations with immune microenvironment characteristics and drug responsiveness, making it a reference for predicting immune therapy responses in GC patients.

Acknowledgments

We express our sincere gratitude to the researchers who provided the original open single-cell data. Additionally, we appreciate the research teams that played a significant role in generating data from GC samples from the TCGA and GEO databases.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1453/rc

Peer Review File: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1453/prf

Funding: This work was supported by the Natural Science Foundation of Sichuan Province (No. 2022NSFSC1378).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1453/coif). The authors have no 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. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

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. Smyth EC, Nilsson M, Grabsch HI, et al. Gastric cancer. Lancet 2020;396:635-48. [Crossref] [PubMed]
2. Bray F, Laversanne M, Sung H, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 2024;74:229-63. [Crossref] [PubMed]
3. Machlowska J, Baj J, Sitarz M, et al. Gastric Cancer: Epidemiology, Risk Factors, Classification, Genomic Characteristics and Treatment Strategies. Int J Mol Sci 2020;21:4012. [Crossref] [PubMed]
4. Kang YK, Kim HD, Yook JH, et al. Neoadjuvant Docetaxel, Oxaliplatin, and S-1 Plus Surgery and Adjuvant S-1 for Resectable Advanced Gastric Cancer: Updated Overall Survival Outcomes From Phase III PRODIGY. J Clin Oncol 2024;42:2961-5. [Crossref] [PubMed]
5. Zubair H, Khan MA, Anand S, et al. Modulation of the tumor microenvironment by natural agents: implications for cancer prevention and therapy. Semin Cancer Biol 2022;80:237-55. [Crossref] [PubMed]
6. Piao HY, Liu Y, Kang Y, et al. Hypoxia associated lncRNA HYPAL promotes proliferation of gastric cancer as ceRNA by sponging miR-431-5p to upregulate CDK14. Gastric Cancer 2022;25:44-63. [Crossref] [PubMed]
7. Liu L, Yu J, Liu Y, et al. Hypoxia-driven angiogenesis and metabolic reprogramming in vascular tumors. Front Cell Dev Biol 2025;13:1572909. [Crossref] [PubMed]
8. 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]
9. Zhang D, Tang Z, Huang H, et al. Metabolic regulation of gene expression by histone lactylation. Nature 2019;574:575-80. [Crossref] [PubMed]
10. Chen J, Huang Z, Chen Y, et al. Lactate and lactylation in cancer. Signal Transduct Target Ther 2025;10:38. [Crossref] [PubMed]
11. Hu Y, He Z, Li Z, et al. Lactylation: the novel histone modification influence on gene expression, protein function, and disease. Clin Epigenetics 2024;16:72. [Crossref] [PubMed]
12. 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]
13. 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]
14. Hao Y, Hao S, Andersen-Nissen E, et al. Integrated analysis of multimodal single-cell data. Cell 2021;184:3573-3587.e29. [Crossref] [PubMed]
15. Jeong HY, Ham IH, Lee SH, et al. Spatially Distinct Reprogramming of the Tumor Microenvironment Based On Tumor Invasion in Diffuse-Type Gastric Cancers. Clin Cancer Res 2021;27:6529-42. [Crossref] [PubMed]
16. Zhang P, Yang M, Zhang Y, et al. Dissecting the Single-Cell Transcriptome Network Underlying Gastric Premalignant Lesions and Early Gastric Cancer. Cell Rep 2019;27:1934-1947.e5. [Crossref] [PubMed]
17. Kumar V, Ramnarayanan K, Sundar R, et al. Single-Cell Atlas of Lineage States, Tumor Microenvironment, and Subtype-Specific Expression Programs in Gastric Cancer. Cancer Discov 2022;12:670-91. [Crossref] [PubMed]
18. 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]
19. Qiu X, Mao Q, Tang Y, et al. Reversed graph embedding resolves complex single-cell trajectories. Nat Methods 2017;14:979-82. [Crossref] [PubMed]
20. Ma Y, Zhou X. Spatially informed cell-type deconvolution for spatial transcriptomics. Nat Biotechnol 2022;40:1349-59. [Crossref] [PubMed]
21. Newman AM, Liu CL, Green MR, et al. Robust enumeration of cell subsets from tissue expression profiles. Nat Methods 2015;12:453-7. [Crossref] [PubMed]
22. Yoshihara K, Shahmoradgoli M, Martínez E, et al. Inferring tumour purity and stromal and immune cell admixture from expression data. Nat Commun 2013;4:2612. [Crossref] [PubMed]
23. Hui S, Ghergurovich JM, Morscher RJ, et al. Glucose feeds the TCA cycle via circulating lactate. Nature 2017;551:115-8. [Crossref] [PubMed]
24. Sun L, Zhang Y, Yang B, et al. Lactylation of METTL16 promotes cuproptosis via m(6)A-modification on FDX1 mRNA in gastric cancer. Nat Commun 2023;14:6523. [Crossref] [PubMed]
25. Zong Z, Xie F, Wang S, et al. Alanyl-tRNA synthetase, AARS1, is a lactate sensor and lactyltransferase that lactylates p53 and contributes to tumorigenesis. Cell 2024;187:2375-2392.e33. [Crossref] [PubMed]
26. Bigos KJ, Quiles CG, Lunj S, et al. Tumour response to hypoxia: understanding the hypoxic tumour microenvironment to improve treatment outcome in solid tumours. Front Oncol 2024;14:1331355. [Crossref] [PubMed]
27. Xu X, Zhang L, Qian Y, et al. A SERPINE1-Based Immune Gene Signature Predicts Prognosis and Immunotherapy Response in Gastric Cancer. Pharmaceuticals (Basel) 2022;15:1401. [Crossref] [PubMed]
28. Ye Z, Yi J, Jiang X, et al. Gastric cancer-derived exosomal let-7 g-5p mediated by SERPINE1 promotes macrophage M2 polarization and gastric cancer progression. J Exp Clin Cancer Res 2025;44:2. [Crossref] [PubMed]
29. Cai G, Qi Y, Wei P, et al. IGFBP1 Sustains Cell Survival during Spatially-Confined Migration and Promotes Tumor Metastasis. Adv Sci (Weinh) 2023;10:e2206540. [Crossref] [PubMed]
30. Wan L, Ge X, Xu Q, et al. Structure and assembly of the dystrophin glycoprotein complex. Nature 2025;637:1252-60. [Crossref] [PubMed]
31. Hu ZG, Zhang S, Chen YB, et al. DTNA promotes HBV-induced hepatocellular carcinoma progression by activating STAT3 and regulating TGFβ1 and P53 signaling. Life Sci 2020;258:118029. [Crossref] [PubMed]
32. Zhang C, Min L, Liu J, et al. Integrated analysis identified an intestinal-like and a diffuse-like gene sets that predict gastric cancer outcome. Tumour Biol 2016;37:16317-35. [Crossref] [PubMed]
33. Hong H, Han H, Wang L, et al. ABCF1-K430-Lactylation promotes HCC malignant progression via transcriptional activation of HIF1 signaling pathway. Cell Death Differ 2025;32:613-31. [Crossref] [PubMed]
34. Mallardo D, Fordellone M, White A, et al. CD39 and LDHA affects the prognostic role of NLR in metastatic melanoma patients treated with immunotherapy. J Transl Med 2023;21:610. [Crossref] [PubMed]
35. Jing X, Liang H, Hao C, et al. Overexpression of MUC1 predicts poor prognosis in patients with breast cancer. Oncol Rep 2019;41:801-10. [PubMed]
36. Liu Y, Li C, Lu Y, et al. Tumor microenvironment-mediated immune tolerance in development and treatment of gastric cancer. Front Immunol 2022;13:1016817. [Crossref] [PubMed]
37. Wang H, Franco F, Ho PC. Metabolic Regulation of Tregs in Cancer: Opportunities for Immunotherapy. Trends Cancer 2017;3:583-92. [Crossref] [PubMed]
38. Ikeda H, Kawase K, Nishi T, et al. Immune evasion through mitochondrial transfer in the tumour microenvironment. Nature 2025;638:225-36. [Crossref] [PubMed]
39. Peralta RM, Xie B, Lontos K, et al. Dysfunction of exhausted T cells is enforced by MCT11-mediated lactate metabolism. Nat Immunol 2024;25:2297-307. [Crossref] [PubMed]