Integrative single-cell and bulk transcriptomics define cell death patterns and ZDHHC22 in gastric cancer progression

Abstract

Introduction‌

Gastric cancer remains a major global health burden with high incidence and mortality. Despite therapeutic advances, long-term survival remains unsatisfactory. Reliable biomarkers to predict therapeutic efficacy and clinical outcomes are urgently needed. This study aimed to elucidate the heterogeneity of gastric cancer using an integrative bioinformatics approach that combines single-cell RNA-sequencing (scRNA-seq) and bulk RNA-seq data.

Methods‌

scRNA-seq datasets were obtained from the GEO database, and bulk RNA-seq data were obtained from the TCGA. Cell communication, pseudotime analysis, and cell death scoring (cuproptosis, ferroptosis, autophagy, and pyroptosis) were performed on the basis of the scRNA-seq datasets. Stemness, survival, drug sensitivity and posttranslational modification (PTM) analyses were conducted using TCGA data.

Results‌

Cancer cells were clustered into three molecular subtypes with distinct molecular characteristics, pathway activation profiles and cell death patterns. Cell–cell communication analysis revealed subtype-specific regulatory interactions, whereas pseudotime and cell death scoring demonstrated dynamic cellular states. Bulk RNA-seq revealed five cell death patterns, among which “cell death 2” and high cuproptosis scores emerged as independent risk factors for poor prognosis. Genes such as LMF1-AS1, CAMKV, ERVW-1, ACLY, GAS2L2, RGS8, and RASSF8-AS1 were differentially expressed in the high-risk group. Samples with low stemness and elevated pyroptosis showed enhanced inferred drug resistance, particularly to LBH589. Additionally, ZDHHC22 was identified as a potential protective prognostic factor, and in vitro assays suggested its association with cell death modulation in gastric cancer.

Conclusion‌

This study highlights the molecular heterogeneity of gastric cancer, demonstrating how distinct cell death patterns and PTM features shape prognosis and drug sensitivity. The identified biomarkers and resistance profiles may provide valuable guidance for personalized therapeutic strategies.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12920-026-02405-7.

Introduction

Gastric cancer is among the most prevalent malignant tumors worldwide, with particularly high incidence and fatality rates [1].China has a disproportionate burden, accounting for approximately 40% of global cases and deaths [2]. Despite ongoing advancements in multimodal therapies, long-term survival outcomes remain unsatisfactory [3]. Certain therapeutic agents, such as histone deacetylase (HDAC) inhibitors, have shown limited clinical efficacy in gastric cancer because of various limitations [4]. Thus, identifying and validating reliable biomarkers is an urgent clinical priority for predicting therapeutic efficacy and patient outcomes in patients with gastric cancer, ultimately facilitating personalized treatment and improving survival.

Recent studies have further underscored the complexity of gastric cancer biology and the need for multidimensional analytical approaches. Single-cell transcriptomic analyses have shown that crosstalk between cancer-associated fibroblasts and myeloid cells shapes the heterogeneous tumor microenvironment and contributes to immune evasion [5]. At the molecular level, the TP53 signaling cascade has been implicated in apoptosis induced by anticancer bioactive peptides [6], whereas the lncRNA WT1-AS has been reported to promote autophagy and suppress tumor growth by inhibiting the PI3K/Akt/mTOR pathway [7]. In addition, mutational profiling has identified FAT4 mutation as a favorable prognostic factor associated with increased tumor mutation burden, suggesting its potential value as a biomarker for immunotherapy response [8]. Despite these advances, an integrated understanding of how diverse cell death modalities and post-translational modifications, such as palmitoylation, collectively influence gastric cancer progression remains limited.

Cuproptosis is a characterized form of cell death triggered by copper-mediated metabolic disruption through direct engagement with lipoylated components in the tricarboxylic acid (TCA) cycle [9]. Elevated expression of cuproptosis-associated immune checkpoints, such as CD209 and HAVCR2, is correlated with significantly poor survival in patients with gastric cancer [10]. Ferroptosis is an iron-catalyzed, nonapoptotic cell death pathway driven by the excessive accumulation of lipid hydroperoxides [11, 12]. Notably, inhibition of ferroptosis has been linked to chemotherapy resistance in gastric cancer [13]. Autophagy, a physiological process in which cellular components are sequestered into autophagosomes and degraded in lysosomes [14], plays a dual role in gastric cancer, as it functions as both a tumor suppressor and a tumor promoter depending on context [15]. Pyroptosis, mediated by gasdermin (GSDM)-derived N-terminal fragments that form membrane pores, disrupts cellular integrity and promotes the release of intracellular contents and inflammatory factors, accompanied by chromatin coagulation and DNA damage. In gastric cancer, pyroptosis also has context-dependent effects, with evidence supporting both tumor-promoting and tumor-suppressive functions [16–18].

Recent studies have further expanded our understanding of gastric cancer pathogenesis. Epigenetic and transcriptional regulators such as PHOX1 [19] and Fra-1 [20] have been shown to drive tumor progression, while metabolic vulnerabilities involving glycolysis [21], one-carbon metabolism [22], and lipid homeostasis [23] have emerged as potential therapeutic targets. In parallel, immunomodulation and cell death pathways, including ferroptosis [24, 25], necroptosis [26], and immunogenic cell death [27], are increasingly recognized as critical determinants of the tumor microenvironment [28, 29]. Additionally, novel prognostic biomarkers, such as GXYLT2 [30], have been identified. Despite these advances, the interplay between diverse cell death modalities and post-translational modifications, such as palmitoylation, in gastric cancer remains poorly characterized.

Posttranslational modification (PTM) has been shown to control tumorigenesis and progression in human cancer [31]. Palmitoylation involves the reversible covalent attachment of palmitic acid to cysteine residues, regulating protein localization, stability, and signaling activity [32]. The functional role of protein palmitoylation in gastric carcinogenesis is now well established. For example, inhibition of the Wnt palmitoyltransferase porcupine suppresses cell proliferation and attenuates activation of the Wnt/β-catenin signaling pathway in gastric cancer [33]. One study identified four palmitoylation-related genes, including ASPA, RBM20, COL4A1, and MAL, using machine learning to predict gastric cancer with high accuracy [34]. Another study revealed that methionine restriction regulates Th1 cells by promoting ZDHHC23-mediated T-bet palmitoylation and degradation, thereby affecting CD8 + T-cell immunity and enhancing anti-PD-1 therapy efficacy in gastric cancer [35]. Moreover, elevated PD-L1 palmitoylation has been shown to stabilize the protein, suppress T-cell-mediated immunity, and facilitate immune evasion in gastric carcinoma [36]. Previous studies have reported that either pharmacological inhibition of Nrf2 palmitoylation or genetic silencing of the palmitoyltransferase zinc finger DHHC2 enhances antitumor immune responses in both cellular systems and murine models [37]. However, beyond these findings, the role of other palmitoyltransferases in regulating gastric cancer remains largely unexplored.

Single-cell RNA sequencing (scRNA-seq) enables cell clustering to dissect tissue heterogeneity, thereby advancing precision medicine [38, 39]. This technology also facilitates precise evaluation of gene function within distinct cellular subtypes, significantly strengthening mechanistic insights into malignant transformation and tumor progression [40, 41]. Despite these advances, integrated analyses of palmitoylation and multiple programmed cell death modalities using scRNA-seq remain largely unexplored in gastric cancer. To address this gap, we integrated scRNA-seq and bulk RNA-seq data to develop a prognostic signature aimed at improving outcome prediction and guiding treatment decisions.

In this study, we used scRNA-seq data to cluster gastric cancer cells and calculated cell death scores on the basis of the expression of genes associated with cuproptosis, ferroptosis, autophagy, and pyroptosis. We systematically analyzed differences in cell death patterns, tumor stemness, and drug sensitivity across clusters. These findings were subsequently validated using bulk RNA-seq data to identify prognosis-related palmitoylases. The cell death scoring system enabled comparative assessment of distinct cell death patterns, followed by evaluation of stemness indices, drug sensitivity, and survival outcomes. Finally, we identified key prognostic markers and potential therapeutic candidates.

Methods

Data collection

The scRNA-seq datasets were retrieved from the Gene Expression Omnibus (GEO) database under accession numbers GSE183904, comprising primary gastric cancer tissues from 27 patients; GSE234129, comprising a primary tumor sample from one patient; and GSE239676, comprising primary tumor samples from 12 patients [42–44]. For GSE234129 and GSE239676, gene expression matrices were generated from 10X Genomics data using the Read10X() function in R. For GSE183904, processed count matrices were provided in CSV format and read using fread(). Only primary tumor samples were retained for analysis. In GSE239676, samples with identifiers ending in “P” were selected as primary tumor samples; in GSE234129, samples ending in “Ca” were selected as cancer samples; and all GSE183904 samples represented primary tumors. After this filtering step, 31 samples were merged for downstream analysis.

Bulk RNA-seq data for 443 stomach adenocarcinoma (STAD) samples were obtained from The Cancer Genome Atlas (TCGA) database (https://portal.gdc.cancer.gov/). After excluding cases with incomplete clinical annotation, 327 samples were retained for downstream analyses. Hallmark cancer-related gene sets were downloaded from the Molecular Signature Database (MSigDB) [45].

Quality control of scRNA-seq data

We performed quality control using the Seurat package in R [46]. Cells were excluded if they expressed fewer than 300 or more than 7,000 genes, had more than 100,000 unique molecular identifiers (UMIs), contained more than 10% mitochondrial reads from genes matching ^MT-, or showed more than 1% erythrocyte gene expression based on HBA1, HBA2, HBB, HBD, HBE1, HBG1, HBG2, HBM, HBQ1, and HBZ. Samples with fewer than 1,000 cells after filtering were removed. The merged dataset was then normalized using the NormalizeData() function with the LogNormalize method, and highly variable genes were identified using FindVariableFeatures(). The data were scaled with ScaleData(), followed by principal component analysis using RunPCA(). No explicit batch correction was applied because visual inspection of the UMAP embeddings did not reveal strong sample-specific clustering. After quality control, 130,433 cells were retained for downstream analyses.

Cell type annotation and extraction of cancer cells

Marker genes for each cluster were identified using the FindAllMarkers function in Seurat. Dimensionality reduction was achieved through both Uniform Manifold Approximation and Projection (UMAP) and t-distributed stochastic neighbor embedding (t-SNE) approaches. Cell type annotation was performed using the SingleR package [47] in combination with canonical marker genes (Table 1) [48]. InferCNV (https://github.com/broadinstitute/inferCNV) analysis was utilized to calculate copy number variation (CNV) scores in epithelial cells, with T cells serving as references. Clusters with high CNV scores were defined as cancer cells. Samples with fewer than 100 cancer cells were excluded. Hierarchical clustering of cancer cells was then carried out using the top 50 highly variable genes. GSEA algorithm [49] based on hallmark cancer gene set (https://www.gsea-msigdb.org/gsea/index.jsp) was utilized for enrichment analysis of marker genes across cancer cell clusters.

Table 1.

Reported marker genes of each cell type

Cell type	Marker genes
Epithelial_cells	EPCAM, KRT18, MUC1
Endothelial_cells	PECAM1, VWF, ENG, MCAM
Fibroblasts	FAP, DCN, PDPN, COL1A2, COL3A1
SMCs	ACTA2, ACTN2, MYL2, MYH2
Monocytes	ITGAX, ITGAM, MS4A7
T_cells	CD2, CD3D, CD3E, CD3G
Treg_cells	FOXP3, CTLA4, CD4
B_cells	CD79A, CD19
Plsama_cells	IGLL5, JSRP1
Mast_cells	TPSAB1, TPSB2
NK_cells	FGFBP2, SPON2, FCGR3A
Macrophages	CD14, CD163, CD68, CSF1R
myeloid_cells	FCER1G, SPI1, CPA3

Cell–cell communication and pseudotime analyses

Cell–cell communication networks were investigated using the CellChat package [50], which models intercellular networks by ligand–receptor interactions. This analysis systematically characterized cellular crosstalk and downstream signaling pathways. Pseudotime trajectories were performed using the monocle package to reconstruct the developmental progression of cancer cell populations and to reveal dynamic expression patterns of cluster-specific marker genes along inferred trajectories [51].

Cell death scoring and transcription factor analysis

Four types of cell death, namely, cuproptosis, autophagy, ferroptosis, and pyroptosis, were included in the analysis. The corresponding gene sets were constructed on the basis of the GeneCards database (https://www.genecards.org). To ensure a robust association between candidate genes and their respective cell death pathways, we ranked genes in descending order based on their GeneCards score values. After rounding these scores to the nearest integer, we selected the top 50 genes (including ties) for inclusion in our analysis. The AverageExpression function was applied to calculate the average gene expression of each cancer cell subtype and sample type. We used the gsva function from the GSVA package with the following parameters: gsva(exp1, gmt, mx.diff = FALSE, kcdf = “Poisson”), where exp1 represents the normalized gene expression matrix. The parameter kcdf = “Poisson” was selected because the expression data showed a count-like distribution, which is generally appropriate for untransformed or unlogged RNA-seq count data. GSVA was subsequently performed using the four cell death-related gene sets to derive cell death scores [52], and the expression profiles of critical cell death-related genes were visualized along pseudotemporal trajectories. Moreover, the AUCell package was used to quantify cell death scores at single-cell resolution [53]. AUCell analysis was performed using the AUCell R package with the following exact call: AUCell_calcAUC(gmt, cells_rankings, nCores = 1, aucMaxRank = nrow(cells_rankings) * 0.1). Here, gmt represents the gene-set list, cells_rankings denotes the per-cell gene expression rankings, and aucMaxRank was set to the top 10% of all ranked genes, calculated as nrow(cells_rankings) * 0.1. Transcription factor annotation for critical cell death-related genes was performed using RcisTarget [54]. Functional enrichment analyses (GO and KEGG) were subsequently conducted for both these genes and their predicted regulatory factors [53].

Stemness analysis and bulk RNA-seq analysis

The CytoTRACE2 package (https://github.com/digitalcytometry/cytotrace2) was used to assess cellular stemness across distinct tumor subtypes and sample types in the scRNA-seq datasets. For the bulk RNA-seq data, cancer samples from TCGA-STAD dataset were included. Four-cell death scores for each sample were calculated using the GSVA algorithm [52]. Stemness indices across TCGA cohorts were further quantified using a one-class predictive model developed with the gelnet package and validated against the Progenitor Cell Biology Consortium (PCBC) dataset (https://www.synapse.org).

Survival, differential expression, and drug sensitivity analyses

TCGA samples were hierarchically clustered based on cell death scores. Age, sex, AJCC tumor stage, cell death cluster, cell death scores, and RNA stemness index (RNAsi) were included as covariates in univariate and multivariate Cox proportional hazards models. Grade and treatment data were not consistently available in the TCGA-STAD dataset and were therefore not included as adjustment variables. The DESeq2 package was used for differential expression analysis of each group, with thresholds of log2-fold change (logFC) > 1 and a p value < 0.05.

Drug sensitivity was predicted using the calcPhenotype function from the oncoPredict R package. The training dataset was obtained from the Cancer Therapeutics Response Portal v2 (CTRP2) and included gene expression profiles expressed as TPM values without log transformation, together with corresponding drug sensitivity measurements, with area under the dose–response curve values converted to IC50 values. The test expression matrix consisted of the averaged log-normalized gene expression values for each sample and was restricted to genes shared by both the training and test datasets. Prediction was performed using the following parameters: batchCorrect = “eb” for empirical Bayes batch correction, powerTransformPhenotype = FALSE, removeLowVaryingGenes = 0.2, and minNumSamples = 20. The output comprised predicted IC50 values for 545 drugs in each sample. Group differences in predicted IC50 values were evaluated using the Wilcoxon rank-sum test. Drugs with an absolute log2 fold change greater than 0 and an FDR less than 0.05 were considered to show significantly different predicted sensitivity between groups.

Posttranslational modification analysis

Subtype-specific differences in palmitoylation were examined across the TCGA samples. The expression profiles of all 23 members of the palmitoyl S-acyltransferase (PAT) zinc finger DHHC (ZDHHC) gene family were systematically quantified [55]. Differentially expressed genes were subjected to Pearson correlation analysis with other subtype-specific DEGs. Genes showing significant correlations were further analyzed via KEGG pathway enrichment and protein–protein interaction (PPI) network analyses.

Cell culture

The gastric cancer cell lines AGS and HGC-27 were sourced from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China). Cells were grown in RPMI-1640 medium containing 10% fetal bovine serum (FBS) and 1% penicillin–streptomycin, and maintained under standard culture conditions at 37 °C in a humidified atmosphere of 5% CO₂. The authenticity of all cell lines was verified through short tandem repeat (STR) analysis, and they were routinely screened to exclude mycoplasma contamination.

Transfection

The full-length ZDHHC22 coding sequence was cloned and inserted into the pcDNA3.1 vector to construct an overexpression plasmid. The recombinant plasmid was verified by sequencing. AGS and HGC-27 cells were subsequently transfected with the pcDNA3.1-ZDHHC22 construct using Lipofectamine 3000 (Invitrogen, CA, USA) following the manufacturer’s instructions. After 72 h of incubation, the cells were subjected to G418 selection to establish stable cell lines, and successful ZDHHC22 overexpression was validated through reverse transcription polymerase chain reaction (RT‒PCR) and western blot analysis.

RT-PCR

Total RNA was isolated from gastric cancer cells using TRIzol reagent (Invitrogen, USA). Complementary DNA (cDNA) was synthesized with the TransGen cDNA Synthesis SuperMix Kit (Beijing, China) according to the manufacturer’s guidelines. Quantitative real-time PCR was subsequently conducted with the SYBR Green PCR Kit (Qiagen, USA). Gene expression was analyzed using the 2^−ΔΔCt method, with β-actin serving as the internal control. The sequences of the primers were as follows: ZDHHC22 forward 5’-GTG ACC TTC GTG CTG CAG CT-3’; reverse 5’- AGG TCG TCT GGG GAG TTC TG TG-3’; β-actin forward 5’-GGA GAT TAC TGC CCT GGC TCC TA-3’; reverse 5’-GAC TCA TCG TAC TCC TGC TTG CTG-3’.

Western blotting

Total protein was extracted from gastric cancer cells using RIPA buffer (Beyotime, China) and quantified with a BCA protein assay kit. Equal amounts of protein were separated by SDS‒PAGE, transferred to PVDF membranes, blocked with 5% nonfat milk, and incubated overnight at 4 °C with anti-ZDHHC22 (PA5-71288, Thermo Fisher Scientific), anti-FDX1 (PA5-102955, Thermo Fisher Scientific), and anti-LIAS (MA5-56418, Invitrogen) antibodies. Protein signals were visualized using an enhanced chemiluminescence (ECL) detection system [56].

CCK-8 assays and EdU assays

Cell viability was measured using the Cell Counting Kit-8 (CCK-8) following the manufacturer’s protocol. Transfected gastric cancer cells were seeded into 96-well plates and cultured overnight. The cells were then treated with rapamycin (AY-22989; MCE, Shanghai, China) or Cu(II)-elesclomol (HY-156376; MCE, Shanghai, China) for 72 h. After treatment, 10 µL of CCK-8 solution was added to each well, and the cells were incubated at 37 °C for 2 h. Absorbance was then measured at 450 nm using a microplate reader [57]. Cell proliferation was examined with a 5-ethynyl-2-deoxyuridine (EdU) assay, in which cells were incubated with EdU for 2 h, fixed in 4% formaldehyde for 30 min, stained with Hoechst 33,342, and observed under a fluorescence microscope to calculate the percentage of EdU-positive cells [58].

Apoptosis assays

Cell apoptosis was examined with a TUNEL detection kit following the supplier’s protocol. Logarithmic-phase gastric cancer cells were plated in 24-well plates (1 × 10⁵ cells per well), fixed with 4% paraformaldehyde after 24 h, stained according to the kit guidelines, and visualized under a fluorescence microscope.

Wound healing assays

A wound healing assay was performed to assess cell migration. Transfected gastric cancer cells were grown in 6-well plates until they reached over 90% confluence, after which a scratch was made in the center of the monolayer using a 100 µL pipette tip. The cells were rinsed with PBS to remove debris, and images of the wound area were taken at different time points. The extent of wound closure was quantified using ImageJ software.

Cell invasion assays

Cell invasion was evaluated using Matrigel-coated Transwell chambers (Corning, USA). Transfected cells were seeded in the upper chamber with 200 µL serum-free medium, while the lower chamber contained medium supplemented with 10% FBS as a chemoattractant. After 24 h of incubation at 37 °C, the cells that remained on the upper surface were gently removed with a cotton swab, and the invading cells on the underside were fixed with 4% paraformaldehyde, stained with crystal violet, and visualized under a microscope for counting.

Statistical analysis

All the statistical computations were performed with R software (version 4.3.1). A significance threshold of p < 0.05 (two-tailed) was applied throughout the study.

Results

Cell type annotation and cancer cell extraction

We collected 40 scRNA-seq samples and, after quality control and dimensionality reduction, retained 130,433 cells for downstream analyses. These cells were then clustered into 18 distinct groups (Figure S1A-G). Heatmap visualization revealed marker gene expression patterns across clusters (Fig. 1A), while bubble plots illustrated the distribution of pivotal cell type-defining transcripts (Fig. 1B). All the cells were subsequently annotated into 12 major cell types, namely, epithelial cells, endothelial cells, fibroblasts, smooth muscle cells (SMCs), monocytes, T cells, regulatory T cells (Tregs), B cells, plasma cells, mast cells, NK cells, and macrophages (Fig. 1C, Figure S1H). From this dataset, 13,595 epithelial cells were extracted and subjected to InferCNV analysis (Fig. 1D–E, Figure S1I–J). On the basis of the hierarchical clustering, a total of 12,153 epithelial cells with high CNV scores were identified as cancer cells (Fig. 1F–H, Figure S2). After screening based on the sample cancer cell content, cancer cells from 31 samples were retained for subsequent analyses.

Fig. 1.

scRNA-seq data processing and cancer cell extraction. A Heatmap depicting the spatial distribution of marker genes across 18 distinct clusters. B Bubble chart displaying cell-type-specific marker gene expression across 18 identified clusters. C UMAP visualization of the annotated cell type distribution following dimensionality reduction. D UMAP projection illustrating epithelial cell distribution across individual samples. E UMAP projection illustrating epithelial cell distribution across 13 distinct clusters. F InferCNV analysis of copy number variation (CNV) scores in epithelial cells, with T cells used as reference controls. G UMAP projection illustrating cancer cell distribution across individual samples. H UMAP projection illustrating the cancer cell distribution across 13 distinct clusters

Cancer cell clustering

To identify novel gastric cancer cell subtypes and uncover the emerging mechanisms underlying tumor microenvironment (TME)-driven cancer progression, we selected the top 50 highly variable genes and performed hierarchical clustering on cancer cells (Fig. 2A). This analysis revealed three distinct subtypes (Fig. 2B). A heatmap of the top 50 variable genes clearly revealed subtype-specific expression profiles (Fig. 2C). GSEA revealed distinct pathway activation patterns across the three subtypes. E2F_TARGETS, KRAS_SIGNALING-DN, and ALLOGRAFT-REJECTION were maximally enriched in subtypes 1, 2 and 3, respectively. Conversely, suppression of PANCREAS-BETA_CELLS and INTERFERON_GAMMA_REPONSE was observed in subtypes 1 and 2, respectively (Fig. 2D–F). Hierarchical clustering of the 31 tumor specimens, based on the compositional ratios of these three subtypes, revealed three sample types. Each sample type was predominantly enriched for one specific cancer cell subtype, with sample types 1–3 corresponding to subtypes 1–3, respectively (Fig. 2G–H).

Fig. 2.

Clustering of cancer cells. A Heatmap of highly variable genes across 13 cancer cell clusters. B UMAP visualization of cancer cell distribution across three distinct subtypes. C Heatmap illustrating the mean expression profiles of the top 50 highly variable genes across the three subtypes. D–F Hallmark cancer gene set–based GSEA enrichment profiles for subtypes 1–3. G–H Distribution of three cancer cell subtypes across individual tumor samples

Cell–cell communication analysis

After defining the three cancer cell subtypes, we next investigated how these subtypes communicate with other cells in the tumor microenvironment. Following quality-control filtering based on tumor purity, 111,339 cells from 31 samples were retained for dimensionality reduction analysis (Figure S3A–C). UMAP and t-SNE projections were employed to visualize the distribution of samples, cellular subtypes, and sample types across the 31 specimens (Fig. 3A–C, Figure S3D–G). Circular plots illustrate the overall interaction counts and weights among cell types (Figure S4A–B), while bubble plots highlight the intensity of specific ligand–receptor pairs (Figure S4C–D).

Fig. 3.

Cell–cell communication analysis. A UMAP projection illustrating the cellular distribution across 31 scRNA-seq samples. B UMAP visualization of annotated cell types across 31 samples. C UMAP projection showing the distribution of the three sample types across 31 samples. D–G Interaction strength of ligand–receptor pairs within the COLLAGEN, EGF, CD96, and CD99 signaling pathways across cancer cell subtypes and other cell populations

In addition, the interaction counts and weights between each cellular population and all others are displayed in circular diagrams (Figure S5–S6). Heatmaps were constructed to summarize the incoming and outgoing signaling strengths between cell populations (Figure S7A). When cancer cells served as the receptor population, subtype-specific differences emerged in several pathways. Compared with those of subtypes 2 and 3, the signaling strength of the COLLAGEN, EGF, CD96, and CD99 pathways was markedly weaker or even absent in subtype 1. These differences were most pronounced in terms of stromal cell interactions (Figure S7B). Similar trends were evident across the four pathways when overall signaling strength was examined (Fig. 3D–G).

Pseudotime analysis and cell death scoring

To investigate the dynamic progression among cancer cell subtypes, we performed pseudotime trajectory analysis, which revealed distinct developmental relationships across the three cancer cell subtypes. Subtype 1 spanned all developmental phases and was enriched in both the early and late stages, whereas subtypes 2 and 3 were largely confined to intermediate differentiation phases (Fig. 4A–C). Genes related to cuproptosis, ferroptosis, autophagy, and pyroptosis exhibited three temporal expression patterns, with expression peaking in early, middle, or late pseudotime stages (Fig. 4D–G). Heatmap visualization of GSVA-derived cell death scores demonstrated subtype-specific associations. Subtype 1 exhibited elevated pyroptosis activity, which matched that of sample type 1, whereas subtype 2 demonstrated concurrent cuproptosis and autophagy activation, which corresponded to sample type 2. Even when scaled across all cells, the cuproptosis and autophagy scores remained prominent in sample type 2. In contrast, subtype 3 (sample type 3) was not strongly associated with any specific cell death pathway (Fig. 4H–J).

Fig. 4.

Pseudotime and cell death scoring analyses. A Developmental progression of cancer cells visualized by pseudotime coloring. B Developmental progression of cancer cells visualized by subtype coloring. C Distribution of three cancer cell subtypes along the pseudotemporal axis. D–G Expression profiles of key cell death–related genes across pseudotime. H Heatmap of cell death scores across three cancer cell subtypes. I Heatmap of cell death scores for cancer cells across three sample types. J Heatmap of cell death scores across all cells in the three sample types. K Heatmap comparing cancer cell death scores across sample types and subtypes

Comparative analysis further revealed subtype-specific cell death patterns: subtype 1 in sample type 1 showed elevated pyroptosis, and subtype 2 in sample type 2 displayed concurrent cuproptosis–-autophagy coactivation, whereas subtype 2 in sample type 3 demonstrated predominant ferroptosis (Fig. 4K). We examined the dynamics of gene expression across pseudotime using heatmaps (Fig. 5A–D). We then performed transcription factor enrichment analysis on marker genes from subtype 1 (highly expressed during early or late stages) and from subtypes 2 and 3 (enriched at intermediate stages), respectively (Figure S8). Interestingly, AUCell-based single-cell quantification of the four cell death pathways revealed patterns that closely matched the GSVA-derived results (Fig. 5E–H), supporting the robustness of these findings.

Fig. 5.

Key cell death markers based on pseudotime and AUCell analyses. A–D Expression profiles of representative cell death–related genes across pseudotemporal progression in cancer cells. E–H UMAP projections illustrating AUCell-based scores for four major types of cell death across cancer cells

Cell stemness, survival, and drug sensitivity analyses

Given the distinct cell death patterns and differentiation states observed, we next evaluated the stemness properties of each subtype using CytoTRACE2 to infer the differentiation status of individual cancer cells (Fig. 6A–C). The results revealed similar differentiation trends across the three cancer cell subtypes and sample types. Notably, both subtype 1 and sample type 1 exhibited a relatively low degree of stemness (Fig. 6D–E). To validate our findings and to systematically explore the roles of gastric cancer cell subtypes and various cell death modes in tumor progression, we leveraged bulk RNA-seq data from 412 TCGA-STAD samples. After patients whose demographic or clinical data were incomplete were excluded, 327 samples were included in the downstream analyses.

Fig. 6.

Stemness, survival, and drug sensitivity analyses. A–C UMAP projections showing the CytoTRACE-based stemness distribution in cancer cells. D CytoTRACE scores across three cancer cell subtypes. E CytoTRACE scores across three sample types. F Heatmap of GSVA-derived cell death scores in the TCGA-STAD cohorts. G Heatmap of stemness scores and clinical characteristics in TCGA samples. H Forest plot of multivariate Cox proportional hazards regression. I Comparison of stemness scores between cell death Group 4 and non–cell death group 4 samples. J–K Differential drug sensitivity profiles in cell death Group 4 and sample type 1 relative to their controls. L–M Differences in sensitivity to LBH589 (panobinostat) in cell death Group 4 and sample type 1 compared with controls

On the basis of the GSVA-derived scores of the four programmed cell death pathways, hierarchical clustering stratified the TCGA samples into five distinct cell death patterns (Fig. 6F). PCBC-based RNAsi values were visualized across demographic features, tumor stage, cell death patterns, and cell death score variables using a hierarchical clustering heatmap (Fig. 6G). Multivariate Cox regression analysis incorporating these variables identified that cell death 2 (hazard ratio (HR) = 2.484; p = 0.047) and high cuproptosis score (HR = 3.333; p = 0.014) were independent risk factors for poor prognosis. Conversely, cell death 3 (HR = 0.142; p = 0.015) and an elevated autophagy score (HR = 0.127; p = 0.009) emerged as independent protective factors (Fig. 6H).

Differential expression analysis, adjusted for covariates in the Cox model, highlighted several key molecular features: suppression of lipase maturation Factor 1 antisense RNA 1 (LMF1-AS1) with upregulation of CaM kinase like vesicle associated (CAMKV) in high cuproptosis cases (Figure S9A–B); reduction of endogenous retrovirus Group W member 1, envelope (ERVW-1), family with sequence similarity 228 member A (FAM228A), alanine and arginine rich domain-containing protein (AARD), and ATP citrate lyase (ACLY) in autophagy-high samples (Figure S9C-D); opposing regulation of growth arrest specific 2 like 2 (GAS2L2) (increased) and regulator of G protein signaling 8 (RGS8) (decreased) in cell death Group 2 (Figure S9E–F); and downregulation of Ras association domain family 8 antisense RNA 1 (RASSF8-AS1) and ACLY levels in autophagy-enriched specimens (Figure S9G–H).

To bridge the scRNA-seq and bulk RNA-seq datasets, we used cell death scores derived from the four programmed cell death pathways. Notably, scRNA-seq sample type 1 and TCGA cell death Group 4 exhibited highly similar patterns, featuring low overall cell death scores but elevated pyroptosis (Figs. 4J and 6F), along with reduced differentiation potential according to stemness analysis (Fig. 6E and I). The Wilcoxon test revealed no significant differences in drug sensitivity between sample type 3 and the other scRNA-seq groups. In contrast, the remaining scRNA-seq and TCGA-defined subgroups showed distinct differences in predicted drug sensitivity, as illustrated in Figure S10. Interestingly, the predicted drug sensitivity profiles of cell death Group 4 and sample type 1 were highly consistent. Compared with their respective controls, both groups shared 27 overlapping drugs with differential responses, 23 of which demonstrated reduced computationally inferred sensitivity (Fig. 6J–K). Among these, LBH589 showed the most pronounced difference in sensitivity across both groups (Fig. 6L–M).

Posttranslational modification analysis

ZDHHC22 expression was significantly upregulated in cell death Group 3 samples compared with that in noncell death Group 3 samples (Fig. 7A). Correlation analysis further revealed that 24 marker genes specific to this group were significantly associated with ZDHHC22 expression (Fig. 7B). KEGG pathway enrichment analysis revealed distinct functional pathways for these correlated genes. Specifically, IL20RA was enriched in the viral protein interaction with cytokine and cytokine receptor pathway, whereas ST8SIA1 participated in glycosphingolipid biosynthesis. Additionally, KCNJ10 was implicated in gastric acid secretion pathways (Fig. 7C–D). Moreover, PPI network analysis uncovered an indirect but functionally relevant regulatory association between ZDHHC22 and KCNJ10, suggesting potential crosstalk between their biological pathways (Fig. 7E). These findings highlight ZDHHC22 as a potential protective prognostic factor in gastric cancer and suggest that it functions through interconnected regulatory networks.

Fig. 7.

Posttranslational modification analysis. A Volcano plot of differentially expressed genes between cell death Group 3 and non–cell death Group 3 samples. B Correlation heatmap of ZDHHC22 and its significantly coexpressed genes in cell death Group 3 samples. C KEGG enrichment dot plot of ZDHHC22 coexpressed genes. D KEGG enrichment circular plot of ZDHHC22 coexpressed genes. E Protein–protein interaction (PPI) network of ZDHHC22 and its significantly coexpressed genes in cell death Group 3 samples

ZDHHC22 inhibits cell viability and proliferation

Our bioinformatics analysis revealed that ZDHHC22 upregulation was significantly associated with favorable prognosis in gastric cancer patients. To functionally validate its tumor-suppressive role, we transfected AGS and HGC-27 cells with a ZDHHC22 expression plasmid. RT–PCR analysis confirmed a significant increase in ZDHHC22 mRNA levels in both cell lines following transfection (Fig. 8A). Consistently, western blotting demonstrated a marked increase in ZDHHC22 protein expression (Fig. 8B). Functional assays revealed that enforced expression of ZDHHC22 markedly suppressed cell viability as measured by a CCK-8 assay (Fig. 8C), and inhibited cell proliferation, as shown by an EdU incorporation assay (Fig. 8D–E). Collectively, these findings indicate that ZDHHC22 overexpression reduces the growth and proliferation of gastric cancer cells.

Fig. 8.

ZDHHC22 inhibits cell viability and proliferation. A RT–PCR was performed to measure the mRNA levels of ZDHHC22 in gastric cancer cells after ZDHHC22 plasmid transfection. B Western blotting was conducted to measure the protein expression levels of ZDHHC22 expression in gastric cancer cells after ZDHHC22 plasmid transfection. C CCK-8 assays were performed to measure the viability of gastric cancer cells. D EdU incorporation assays were performed to measure the proliferation of gastric cancer cells. E Quantitative results for Panel D. **p < 0.01, ***p < 0.001

ZDHHC22 induces apoptosis and inhibits invasion

We used a TUNEL assay to evaluate apoptosis in AGS and HGC-27 cells after ZDHHC22 plasmid transfection and found that ZDHHC22 overexpression markedly promoted apoptotic cell death (Fig. 9A–B). Wound-healing assays demonstrated that elevated ZDHHC22 expression impaired the migratory capacity of both cell lines (Fig. 9C–D). Additionally, Matrigel-coated Transwell assays showed that ZDHHC22 overexpression significantly reduced invasive potential (Fig. 9E–F). Collectively, these findings indicate that ZDHHC22 upregulation suppresses the migration and invasion of gastric cancer cells. Next, we examined the expression of FDX1 and LIAS, two key regulators of the cuproptosis pathway, in ZDHHC22-overexpressing gastric cancer cells. We found that ZDHHC22 overexpression increased the levels of both FDX1 and LIAS in gastric cancer cells (Fig. 9G).

Fig. 9.

ZDHHC22 induces apoptosis and inhibits invasion. A A TUNEL assay was performed to evaluate apoptosis in AGS and HGC-27 cells. B Quantitative results for Panel A. C Wound healing assays were performed to measure the migratory capacity of gastric cancer cells. D Quantitative results for Panel C. E Matrigel-coated Transwell assays were performed to measure the invasive potential of gastric cancer cells. F Quantification of the data in Panel E. G Western blotting was performed to measure the expression of FDX1 and LIAS in gastric cancer cells. H CCK-8 assays were performed to measure the viability of gastric cancer cells after they were treated with 30 nM LBH589 plus ZDHHC22 plasmid transfection for 72 h. **p <0.01, ***p <0.001

However, the precise molecular mechanism by which ZDHHC22 upregulation increases FDX1 and LIAS levels remains to be fully elucidated. Moreover, a CCK-8 assay revealed that 30 nM LBH589 inhibited gastric cancer cell viability, and that ZDHHC22 overexpression further enhanced the LBH589-mediated suppressive effect (Fig. 9H). These findings suggest that ZDHHC22 upregulation is associated with increased sensitivity to LBH589 in gastric cancer cells, warranting further investigation into its role in drug response.

ZDHHC22 is associated with cuproptosis and autophagy modulation

To investigate whether ZDHHC22 participates in the regulation of cuproptosis in gastric cancer cells, ZDHHC22-overexpressing cells were treated with the cuproptosis inducer Cu(II)-elesclomol, and cell viability was subsequently assessed. Cu(II)-elesclomol treatment reduced cell viability in both AGS and HGC-27 cells (Supplementary Figure S11). Notably, ZDHHC22 overexpression combined with Cu(II)-elesclomol treatment resulted in a greater reduction in cell viability than either treatment alone (Supplementary Figure S11). To determine whether ZDHHC22 is also involved in the regulation of autophagy, ZDHHC22-overexpressing gastric cancer cells were treated with the autophagy activator rapamycin, followed by assessment of cell viability. Rapamycin treatment likewise reduced cell viability in both AGS and HGC-27 cells (Supplementary Figure S11). Furthermore, the combination of ZDHHC22 overexpression and rapamycin treatment led to a more pronounced decrease in cell viability compared with either condition alone (Supplementary Figure S11). Collectively, these findings suggest that ZDHHC22 overexpression is associated with enhanced cuproptosis and altered autophagy-related features in gastric cancer cells. However, whether ZDHHC22 directly regulates these pathways or acts indirectly through other mediators remains to be determined. Further mechanistic studies are warranted to clarify the nature of this association.

Discussion

In this study, we integrated scRNA-seq data with bulk transcriptomic profiles from TCGA datasets to characterize heterogeneity within the gastric cancer TME. Through this approach, we delineated distinct cell death patterns, evaluated their associations with patient prognosis, and assessed their implications for drug sensitivity prediction. Hierarchical clustering of cancer cells revealed three subtypes, and GSEA of marker genes highlighted distinct functional enrichments. Subtype 1 showed significant upregulation of E2F target, G2M checkpoint, and mitotic spindle pathway genes (Fig. 2D).

Increased E2F activity is well known to drive uncontrolled proliferation [59], while enrichment of G2M checkpoint and mitotic spindle signaling pathways underscores their role in the cell cycle and mitosis [45]. Taken together, these findings suggest that subtype 1 represents a proliferative cluster closely associated with cell proliferation. Subtype 2 was enriched for the PI3K/AKT/mTOR pathway, a key regulator of cell survival, antiapoptotic processes, and metastatic dissemination ‌ [60–62], indicating that tumor progression occurred in this cluster. Subtype 3 was enriched in the apical junction pathway, which has been linked to enhanced metastatic potential and unfavorable prognosis in gastric cancer [63]‌.

Analysis of cell subtype proportions across samples revealed that subtype 1 was the most abundant, whereas subtypes 2 and 3 may play a more critical role in advanced disease progression (Fig. 2G–H). This distribution pattern suggests that although subtype 1 may constitute the majority of tumor cells at earlier stages, subtypes 2 and 3 may play more critical roles during disease progression. Consistently, pseudotime and stemness analyses demonstrated that subtype 1 has the lowest degree of stemness, supporting its role as an early stage tumor population [64, 65]. In contrast, subtypes 2 and 3 were predominantly confined to intermediate differentiation phases and exhibited higher stemness, further supporting their potential association with tumor progression and cellular plasticity.

Cell–cell communication analysis revealed stronger interaction intensities for subtypes 2 and 3 than for subtype 1 across the COLLAGEN, EGF, CD96, and CD99 pathways. Activation of the tumor-associated collagen pathway in gastric cancer has been implicated in immunosuppression and adverse clinical outcomes [66]‌. EGF upregulation strongly correlated with enhanced metastatic potential [67, 68] and unfavorable clinical outcomes in gastric cancer [69]. CD96 has been identified as an immunosuppressive checkpoint associated with poor prognosis [70]. However, findings on CD99 in gastric cancer have shown discordant results, with some studies reporting that its downregulation is associated with worse prognosis [71]‌, which requires further mechanistic investigation.

Survival analyses using TCGA data confirmed the prognostic impact of programmed cell death signatures. Multivariate Cox regression revealed that cell death Group 2 and a high cuproptosis score were independent risk factors, whereas cell death Group 3 and an elevated autophagy score conferred protective effects. These results underscore the prognostic relevance of cell death–associated transcriptional programs in gastric cancer and suggest their potential for patient risk stratification.

Cuproptosis, triggered by copper overload, is strongly positively correlated with lymph node metastasis and inversely associated with both overall survival (OS) and disease-free survival (DFS) in gastric cancer patients [9, 72], which is in agreement with our findings. Similarly, cytoprotective autophagy has been shown to suppress tumorigenesis by maintaining genomic stability and cellular homeostasis during malignant transformation [73], which also supports our findings. Given the complexity of gastric cancer progression, relying solely on a single cell death score may be inadequate for accurate prognostic assessment.

In contrast, clustering on the basis of four cell death modalities—particularly cell death Groups 2 and 3—offers a more promising approach to effectively stratify patient survival risk. Differential expression analyses based on cell death Group 2, the cuproptosis score, cell death Group 3, and the autophagy score as stratification variables revealed several clinically relevant candidate genes. LMF1-AS1 expression was significantly lower in the high cuproptosis score group, which is consistent with its positive association with survival outcomes. Similarly, in esophageal adenocarcinoma (EAC), high LMF1 expression has been linked to improved prognosis, supporting the favorable prognostic association observed in this study [74].

Conversely, we observed significantly higher CAMKV expression in the high cuproptosis group. Although CAMKV-IgG has been reported to be a diagnostic marker for immunotherapy-responsive paraneoplastic encephalitis and uterine malignancies [75], its role in gastric cancer and cuproptosis remains largely unexplored and warrants further study.

ERVW-1, FAM228A, AARD, and ACLY expression decreased in the high autophagy score group, suggesting potential links to adverse prognosis. Notably, both the upregulation of ERVW-1 expression and the increase in autophagy have both been implicated in endosomal hyperplasia, which is a precursor of endosomal cancer [76, 77]. Although the role of FAM228A and AARD in cancer biology remains largely unexplored, their association with autophagy identified here highlights novel directions for future research. Additionally, the expression of RGS8 was markedly downregulated in the cell death Group 2, which is consistent with reports in thyroid cancer showing its reduced expression in tumor tissues [78], suggesting its potential tumor-suppressive role. In contrast, we detected increased GAS2L2 expression in the cell death Group 2, which has previously been implicated as a potential therapeutic target in esophageal cancer radiotherapy [79].

Together, these findings highlight the prognostic and therapeutic potential of genes associated with specific cell death clusters in gastric cancer. As depicted in Fig. 6H, cell death Group 3 emerged as a protective factor, with RASSF8-AS1 downregulated in this group. These findings are consistent with those of prior reports demonstrating the oncogenic role of RASSF8-AS1 in esophageal squamous cell carcinoma [80]. ACLY inhibition suppresses gastric cancer progression by blocking fatty acid synthesis and autophagic flux [62]. ACLY expression decreased in both the high-autophagy group and the cell death type 3 group. Together with the protective role of these prognostic signatures these findings further support a protumorigenic role for ACLY.

Comparative analyses revealed that sample type 1 and cell death Group 4 exhibited similar cell death patterns, and we subsequently examined their relationships with tumor stemness and inferred drug resistance. Both groups demonstrated low stemness indices and increased resistance to multiple chemotherapeutic agents. LBH589, also known as panobinostat, a histone deacetylase (HDAC) inhibitor [81], showed the most prominent increase in resistance among the above 2 groups. Conversely, in the scRNA-seq datasets, compared with sample type 1, sample types 2 and 3 exhibited greater sensitivity to LBH589. The marked resistance of sample type 1/cell death Group 4, characterized by low stemness and high pyroptosis, to LBH589 is noteworthy. This resistance may stem from its high pyroptotic state or may instead reflect parallel outcomes driven by a common upstream regulator. In addition, potential molecular bridges, such as specific inflammasome components or gasdermin proteins, may influence sensitivity to HDAC inhibitors. Together with previous reports highlighting the limited clinical efficacy of LBH589 in gastric cancer treatment [4], these findings suggest that our molecular classification could help guide the more selective use of HDAC inhibitors in clinical practice.

PTMs play a key role in oncogenesis in various cancer types [82, 83]. Protein palmitoylation, primarily catalyzed by the Asp-His-His-Cys (DHHC) family of palmitoyltransferases, has recently attracted attention because of its roles in tumor biology [84, 85]. ZDHHC22, a member of the DHHC family, has been identified as a favorable prognostic biomarker in breast cancer, with its upregulation correlated with improved relapse-free survival [86]. Although its function in gastric cancer has not been previously described, the upregulation of ZDHHC22 in cell death Group 3, together with the protective prognostic effect observed in this subgroup, suggests a potential tumor-suppressive function for ZDHHC22-mediated palmitoylation in gastric cancer. Despite these findings, we acknowledge that our experimental validation does not yet fully capture the heterogeneity of the three cancer cell subtypes and five cell death patterns identified by our integrative analysis. The current in vitro experiments focused primarily on the effects of ZDHHC22 on apoptosis, autophagy, and cuproptosis. Further studies using mouse models and patient-derived organoids will be necessary to more comprehensively elucidate the underlying biology.

Our study integrates multimodal cell death analysis with palmitoylation profiling to characterize molecular features associated with gastric cancer progression. Using in vitro assays, we identified ZDHHC22 as a candidate protective factor whose upregulation is associated with altered cell death phenotypes. Nevertheless, several limitations warrant acknowledgment. First, our bioinformatics analysis relied only on publicly available datasets. Second, functional validation was performed using a limited number of cell lines, which may not fully capture the biological variability of gastric cancer. Third, independent cohort validation of the cell death patterns and prognostic signatures is necessary to further confirm their reliability. Fourth, the functional role of ZDHHC22 remains hypothesis-generating in gastric cancer. Additional validation using independent cohorts, quantitative PCR (qPCR), immunohistochemistry (IHC), or public protein-level datasets would further strengthen the evidence supporting the role of ZDHHC22 in gastric cancer. Fifth, whether ZDHHC22 directly palmitoylates key components of the autophagy or cuproptosis pathways, and whether these in vitro observations translate to clinical cohorts, require independent validation. In addition, the drug sensitivity analyses are purely computational predictions based on CTRP2‑trained models. Although these analyses generate testable hypotheses, direct experimental validation in gastric cancer cell lines or patient‑derived models is required to confirm the predicted sensitivity patterns, particularly for agents such as LBH589.

Conclusion

In conclusion, this integrative single-cell and bulk transcriptomic study characterized the molecular heterogeneity of gastric cancer and identified three cancer cell subtypes with distinct functional features, cell death profiles, and stemness properties. Among the five cell death-based clusters identified from bulk RNA-seq data, specific patterns were independently associated with prognosis. Notably, sample type 1 and cell death Group 4 exhibited low stemness, elevated pyroptosis, and reduced computationally inferred sensitivity to multiple drugs, most prominently LBH589. We further identified ZDHHC22 as a potential protective factor whose upregulation was associated with enhanced cuproptosis, altered autophagy-related features, and increased drug sensitivity in vitro. These findings provide new insights into gastric cancer heterogeneity and suggest that cell death-based subtyping may help inform prognostic evaluation and guide the selective use of therapeutic agents, such as HDAC inhibitors. External cohort validation and further mechanistic investigation remain necessary.

Abbreviations

TCA

Tricarboxylic acid

GEO

Gene Expression Omnibus

scRNA-seq

Single-cell RNA sequencing

TCGA

The Cancer Genome Atlas

MSigDB

Molecular Signature Database

UMAP

Uniform Manifold Approximation and Projection

t-SNE

T-distributed stochastic neighbor embedding

CNV

Copy number variation

RNAsi

RNA stemness indices

logFC

Log2 fold change

PPI

Protein-protein interaction

SMCs

Smooth muscle cells

PCBC

Progenitor Cell Biology Consortium

HR

Hazard ratio

OS

Overall survival

DFS

Disease-free survival

LMF1-AS1

LMF1 antisense RNA 1

LMF1-AS1

Lipase maturation factor 1 antisense RNA 1

CAMKV

CaM kinase like vesicle associated

ERVW-1

Endogenous retrovirus group W member 1,envelope

FAM228A

Family with sequence similarity 228 member A

AARD

Alanine and arginine rich domain containing protein

ACLY

ATP citrate lyase

GAS2L2

Growth arrest specific 2 like 2

RGS8

Regulator of G protein signaling 8

RASSF8-AS1

Ras association domain family member 8 antisense RNA 1

HDAC

Histone deacetylase

Authors’ contributions

JW and JL designed the study. JW, CC, GD, LH, YZ, ZC, and QM performed the study and analyzed the data. JW, PW and JL writing and/or revising the manuscript. All authors read and approved of the final manuscript.

Funding

This study was supported by Zhejiang Basic Public Welfare Research Program (LBY24H180005), Medical and Health Science and Technology Program of Zhejiang Provincial (2024KY431) and and National Oncology Clinical Key Speciality (2023-GJZK-001).

Data availability

All datasets utilized in this investigation were obtained from publicly accessible repositories, with detailed accession numbers and database identifiers provided in the Methods section.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.