Breaking epigenetic shackles: targeting ARID1A methylation and the PI3K/AKT/mTOR-PD-L1 axis to overcome immune escape in gastric cancer

Methylation analysis of ARID1A

The methylation status of ARID1A in GC tissues and adjacent normal tissues was evaluated using the UALCAN database (http://ualcan.path.uab.edu/analysis.html, accessed 15 September 2024). The correlation between ARID1A methylation density and mRNA expression was evaluated using Spearman correlation analysis. The data for this analysis were downloaded from cBioPortal (http://www.cbioportal.org, Stomach Adenocarcinoma dataset, accessed 20 September 2024). Additional validation of site-specific methylation effects was performed using MEXPRESS (https://mexpress.ugent.be/, accessed 20 September 2024). To validate site-specific methylation patterns identified in the MEXPRESS analysis (FDR-adjusted p-value < 0.05), RNA expression profiles, methylation data, and clinical information for GC were retrieved from the UCSC Xena (accessed September 2024). Publicly available datasets included RNA-seq expression profiles from 375 GC and 33 normal tissues and methylation data from 396 tumors and two normal tissues (Illumina Human Methylation 450K BeadChip). Gene identifiers were converted to gene symbols using Ensembl (Homo sapiens, http://asia.ensembl.org/index.html, accessed 20 September 2024). All data were obtained from publicly accessible, de-identified databases (UALCAN, cBioPortal, MEXPRESS, UCSC Xena). According to NCI Guidelines (2015) and TCGA policies (https://www.cancer.gov/ccg/research/genome-sequencing/tcga/history/ethics-policies), this secondary analysis of pre-deidentified public data is exempt from ethics committee approval.

Identification of differentially expressed genes

Gastric cancer patients were stratified by ARID1A promoter β-values (cg05445839) into hypermethylated (Hyper-group, ≥75th percentile) and hypomethylated (Hypo-group, ≤25th percentile) subgroups. Methylation-regulated differentially expressed genes (MeDEGs) were identified using the DESeq2 package (version 1.38.3) in R software, with significant thresholds set at |log₂FoldChange| ≥1.5 and false discovery rate (FDR) adjusted p-value < 0.05 for subsequent analyses.

Functional enrichment analysis and protein interaction network analysis

MeDEGs were annotated using Gene Ontology (GO) functional enrichment and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses. GO enrichment included biological processes (BP), molecular functions (MF), and cellular components (CC) (Ashburner et al., 2000). KEGG, a database resource, was used to elucidate gene functions at the molecular and higher-order levels, including biochemical pathways (Kanehisa et al., 2017). Enrichment analysis was conducted using a hypergeometric test. For gene set functional enrichment, the clusterProfiler R package (v3.14.3) (Yu et al., 2012) was used with GO annotations serving as the background reference to map genes to predefined functional categories. Statistically significant terms were defined by a p-value < 0.05 and a false discovery rate (FDR) <0.05. Protein-protein interaction (PPI) networks were constructed using the STRING (v11.5, confidence score >0.7, active sources: Experiments & Databases) (Szklarczyk et al., 2019) and visualized with Cytoscap (v3.10.0) (Shannon et al., 2003). Hub genes were identified using the cytohubba plugin (v0.1) in Cytoscape (v3.10.0) by selecting the top 10 nodes ranked by Closeness centrality. Functional modules were detected with MCODE (v2.0.2) using default parameters (degree cutoff = 2, node score cutoff = 0.2, k-core = 2, max depth = 100).

Gene set enrichment analysis

Gene set enrichment analysis (GSEA) was conducted using the clusterProfiler (v3.14.3) in R to identify significant differences in predefined gene sets between the methylation-defined groups, applying Benjamini–Hochberg correction (FDR-adjusted p-value < 0.05) and size filtering (10–500 genes). Statistically significant enrichment was defined by the following cutoff criteria: |Normalized Enrichment Score (NES)| > 1, nominal p- value < 0.05, and FDR-adjusted p-value < 0.05.

TISIDB database analysis

The “Chemokine” module of the TISIDB database (http://cis.hku.hk/TISIDB/, accessed 1 October 2024) was used to investigate the association between ARID1A and chemokine/chemokine receptor expression levels in tumor-infiltrating immune cells. To further elucidate the immunological relevance of ARID1A in cancer, the “Immunomodulator” module was used to evaluate correlations between ARID1A expression and immune checkpoint gene levels.

Comprehensive immune landscape profiling in methylation-defined subgroups

Immunological heterogeneity between the hypermethylated and hypomethylated subgroups was comprehensively profiled using seven computational algorithms and visualized via hierarchical clustering heatmaps (Euclidean distance; ComplexHeatmap v2.14.0). Immune cell composition was first deconvoluted using CIBERSORT to calculate the relative enrichment scores for 22 functionally distinct immune cell subtypes (Newman et al., 2015). TME characteristics, including the immune/stromal cell ratios and neoplastic purity, were quantified using the ESTIMATE algorithm (Yoshihara et al., 2013). Stromal-immune interplay was further resolved using an MCP-counter, which provides absolute quantification of eight stromal and two immune cell populations (Becht et al., 2016). The EPIC framework was implemented to differentiate cancer epithelial cells from seven non-malignant microenvironment components, enhancing the resolution of tumor-immune interactions (Racle et al., 2017). Bulk transcriptome data were analyzed via TIMER 2.0 to estimate the abundance of six functionally key immune infiltrates (CD8⁺ T cells, B cells, macrophages, etc.) (Li et al., 2020). The quanTIseq pipeline enabled absolute quantification of 10 immune cell types through signature matrix deconvolution of RNA-seq data (Finotello et al., 2019). Finally, the Immunophenoscore (IPS) algorithm integrated MHC class I/II presentation, immunostimulatory/checkpoint molecules, and effector cell signatures to generate composite immunogenicity indices (Charoentong et al., 2017).

Cell culture and reagent preparation

Human GC cell lines HGC-27 and AGS were cultured in RPMI 1640/F-12 medium (Gibco, Thermo Fisher Scientific, Waltham, MA, USA) supplemented with 10% fetal bovine serum (FBS; Gibco, Waltham, MA, USA) and 1% penicillin/streptomycin (Beyotime, Jiangsu, China). All cells were maintained in a humidified incubator at 37 °C with 5% CO₂. Cell line authenticity was routinely verified by short tandem repeat DNA profiling, and mycoplasma contamination was excluded using the MycoBlue Mycoplasma Detector (Vazyme Biotech, Nanjing, China) before experimental use. 5-Aza-2’-deoxycytidine (5-aza-CdR; Selleck Chemicals, Houston, TX, USA) was reconstituted in dimethyl sulfoxide (DMSO; Gibco, Waltham, MA, USA) to a stock concentration of 50 mM, aliquoted, and stored at −80 °C until use. SC79 (MedChemExpress, Monmouth Junction, NJ, USA) was similarly prepared in DMSO at a stock concentration of 20 mM and stored under identical conditions.

RNA isolation and quantitative real-time PCR analysis

Total RNA was isolated using the RNA-easy Isolation Reagent Kit (Vazyme Biotech, Nanjing, China) according to the manufacturer’s protocol. RNA concentration and purity were measured spectrophotometrically using a NanoDrop 2000 system (Thermo Fisher Scientific, Waltham, MA, USA). For cDNA synthesis, 1 µg of total RNA was reverse-transcribed using SuperScript™ IV Reverse Transcriptase (200 U/μL; Invitrogen, Waltham, MA, USA) under conditions specified by the supplier.

Quantitative real-time PCR (qRT-PCR) was performed on a Bio-Rad CFX96 Touch Real-Time PCR Detection System using FASTStart SYBR Green Master Mix (PeqLab Biotechnologies GmbH). Each 20 μL reaction contained 2 μL of cDNA template. Primer sequences were as follows: ARID1A: Forward: 5’-CAGTACCTGCCTCGCACATA-3’; Reverse: 5’-GCCAGGAGACCAGACTTGAG-3’.

β-actin (ACTB; endogenous control): Forward: 5’-TCCTGTGGCATCCACGAAACT-3’; Reverse: 5’-GAAGCATTTGCGGTGGACGAT-3.’ Thermal cycling conditions included an initial denaturation at 95 °C for 30 s, followed by 40 cycles at 95 °C for 10 s and 60 °C for 30 s. Relative gene expression was calculated using the 2ˆ(−ΔΔCt) method, with β-actin as the normalization control.

DNA isolation and quality assessment

Genomic DNA was isolated from GC cell lines using the TIANamp Genomic DNA Kit according to the manufacturer’s protocol. DNA concentration and purity were assessed spectrophotometrically using a NanoDrop™ 2000 system (Thermo Fisher Scientific, Waltham, MA, USA). Samples meeting the following quality criteria were retained for downstream analyses: DNA concentration ≥100 ng/μL and A260/280 absorbance ratios approximating 1.8 (±0.2).

Bisulfite treatment

Genomic DNA (1 µg per sample) was subjected to bisulfite conversion using a DNA Bisulfite Conversion Kit (Beyotime Biotechnology, Jiangsu, China) according to the manufacturer’s protocol. The conversion reaction was performed in a thermal cycler under the following conditions: initial denaturation at 95 °C for 3 min, followed by 12 cycles at 95 °C for 30 s and 70 °C for 10 min, with a final hold at 4 °C. The converted DNA was then transferred to purification columns for desulfonation, washing, and elution. The bisulfite-converted DNA samples were ultimately stored at −20 °C for further molecular analyses.

Methylation-specific PCR analysis of ARID1A

The bisulfite-converted DNA was subjected to methylation-specific PCR (MSP). The MSP primers were designed using the MethPrimer program (MethPrimer-Design MSP/BSP primers and CpG islands prediction; Li Lab, PUMCH; available at  http://www.urogene.org, accessed 5 December 2024). The following primer sets were used for the MSP analysis:

ARID1A-M-F: 5’-GTAATAATTTGGCGTTTTAGCGA-3’;

ARID1A-M-R: 5’-ACCCAATCCTTATAAAAAAAATCGT-3’ (methylated-specific),

ARID1A-U-F: 5’-GGGGTAATAATTTGGTGTTTTAGTG-3’,

ARID1A-U-R: 5’-CCCAATCCTTATAAAAAAAATCATC-3’ (unmethylated-specific).

The 50 uL MSP-PCR reaction mixture contained 5 μL of 10 × PCR Buffer (Mg²⁺ Plus, 25 mM; R007A, TaKaRa Bio Inc., Shiga, Japan), 4 μL of dNTP Mixture (2.5 mM, R007A, TaKaRa, Shiga, Japan), 1 μL each of forward and reverse ARID1A primers (100 μM, Sangon Biotech, Shanghai, China), 0.25 μL of TaKaRa LA Taq HS DNA polymerase (5U/uL, R007A, TaKaRa, Shiga, Japan), 36.75 μL of nuclease-free H₂O, and 2 μL of bisulfite-converted DNA template (50 ng/μL). Thermal cycling conditions for MSP amplification were as follows: Initial denaturation at 95 °C for 5 min, followed by 40 cycles of denaturation at 95 °C for 30 s, primer-specific annealing (59 °C for methylated [M] primers or 55 °C for unmethylated [U] primers) for 45 s, and extension at 72 °C for 45 s, with a final extension at 72 °C for 10 min. Methylation-specific PCR products were resolved on a 2% agarose gel pre-stained with YeaRed Nucleic Acid Gel Stain (YEASEN Biotechnology, Cat# 10202ES76, Shanghai, China) at 140 V for 40 min. DNA bands were visualized under UV light, and the 106 bp target fragments were identified by comparison with DL2000 Plus DNA Marker (MD101-01-AA; 100–5000 bp; Vazyme, Beijing, China).

Lentiviral transduction for ARID1A knockdown and cell infection

Lentiviral vectors targeting ARID1A (shARID1A:5’-TTCTCCGAACGTGTCACGT-3’) and non-targeting control shRNA (shCtrl: 5’-GTTGATGAACTCATTGGTT-3’) were obtained from Genepharma (China). For lentiviral transduction, GC cells at 40% confluence were co-transduced with two lentiviral constructs: one carrying a ARID1A-specific shRNA and the other containing scrambled control shRNA. Transduction was performed at a multiplicity of infection of 10 in serum-free basal medium to enhance viral entry efficiency. After 8–12 h of incubation, the medium was replaced with a complete growth medium containing 10% FBS and maintained for 72 h post-transduction.

GFP expression in transduced cells was quantitatively assessed using an A1 confocal laser-scanning microscope (Nikon, Japan). Successful transduction was confirmed by the detection of green fluorescence indicative of GFP-positive cells. Subsequently, transduced cells underwent rigorous selection with 2 μg/mL puromycin for 7 d to isolate puromycin-resistant clones. Validation of stable ARID1A knockdown was confirmed by western blotting and qRT-PCR before downstream functional assays.

Western blot

Cells were washed with phosphate-buffered saline (PBS) and lysed in RIPA lysis buffer (P0013B, Beyotime) supplemented with Protease Inhibitor Cocktail (HY-K0010, MedChemExpress) and phosphatase inhibitors (HY-K0021, MedChemExpress). Lysates were incubated on ice for 30 min, centrifuged at 12,000 rpm for 10 min at 4 °C, and supernatants were collected. Protein concentrations were quantified using the BCA Protein Assay Kit (P0010S, Beyotime). Equal amounts of protein from each sample were separated by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and transferred onto nitrocellulose membranes (EMD Millipore, Billerica, MA, USA). Membranes were blocked with 5% non-fat milk at room temperature and then incubated overnight at 4 °C with primary antibodies, including: Anti-ARID1A (1:1000, ab182560, Abcam), Anti-PD-L1 (1:1000, PB0166, BOSRER), Anti-PI3K (1:2000, 60225-1-AP, Proteintech), Anti-p-PI3K (1:1000, #4228, Cell Signaling Technology [CST]), Anti-AKT (1:1,000, D290056, Sangon), Anti-p-AKT (Ser473) (1:1,000, T40067, Abmart), Anti-mTOR (1:1000, #CBD-13, BOSRER), Anti-p-mTOR (1:1000, #IFF-13, BOSRER), and Anti-β-actin (1:10,000, A3854, Sigma-Aldrich), Tubulin (1:2000,66200-1-AP, Proteintech). Membranes were subsequently incubated with horseradish peroxidase-conjugated goat anti-mouse/rabbit IgG secondary antibodies (LEF22060, Life-iLab) at room temperature. Protein bands were visualized via chemiluminescence, and signal intensities were quantified by densitometry using the ImageJ software.

Migration and invasion assays

For migration analysis, GC cells (3 × 10⁴ cells/well), either treated or untreated, were seeded into the upper chamber of Transwell inserts (8-µm pore size, Corning Inc., Corning, NY, USA). The lower chamber was filled with 600 μL of complete medium alone or medium containing a specified concentration of 5-aza-CdR. After 24 h of incubation, cells in the upper chamber were fixed with 4% paraformaldehyde, stained with 0.1% crystal violet, and imaged under an optical microscope. The number of migrating cells was quantified by counting cells in six randomly selected microscopic fields per insert. For the invasion assay, the protocol was identical, except that Transwell inserts were pre-coated with 4 μL of Matrigel (Corning Inc., Corning, NY, USA) before cell seeding.

Wound-healing assays

A wound healing assay was performed to evaluate its effect on the cellular migratory capacity. Following successful transfection, well-grown cells were seeded into six-well plates and cultured to form confluent monolayers. A sterile 200 µl pipette tip was used to create a linear mechanical wound by vertically scraping the cellular monolayer. After three consecutive washes with PBS to eliminate cellular debris, a serum-free medium was added. Wound closure progression was systematically monitored at baseline 0 h and 24 h intervals, with photomicrographic documentation performed under standardized conditions. High-resolution images were captured using an Olympus IX83 inverted light microscope equipped with phase-contrast optics. Quantitative analysis of cellular migration was conducted by measuring the interwound distance using the ImageJ software (National Institutes of Health, Bethesda, MD, USA) with calibrated digital image-processing protocols.

Cell counting kit-8 cell proliferation assays

Cell viability was systematically evaluated using the Cell Counting Kit-8 (CCK-8; Sangon Biotech, Shanghai, China) following the manufacturer’s recommended protocols. GC cells were seeded into 96-well microplates at a density of 1,000 cells per well and allowed to stabilize overnight. After designated incubation intervals, 10 µl of CCK-8 reagent was aseptically dispensed into each experimental well. The plates were then transferred to a light-protected incubator and maintained at 37 °C for 2 h to facilitate chromogenic substrate conversion. Absorbance was quantified at 450 nm using an Accuris SmartReader 96 Microplate Reader (Benchmark Scientific, Sayreville, NJ, USA). Proliferation curves were plotted using GraphPad Prism software (v9.5.1; San Diego, CA, USA) based on the acquired optical density (OD) values, enabling quantitative analysis of temporal cellular proliferation dynamics.

Plate colony formation assays

For clonogenic assays, GC cells were seeded at 500 cells/well in six-well plates (Corning, Corning, NY, USA) and cultured for 14 d (37 °C, 5% CO₂). Colonies were fixed with 4% paraformaldehyde (15 min), stained with 0.1% crystal violet (10 min), washed twice with PBS to remove unbound dye, and vertically air-dried before quantification.

Flow cytometry

Apoptosis was analyzed using the Annexin V-647/Propidium iodide (PI) Dual Staining Kit (Cat# AC12L043, Life-iLab, China) following the manufacturer’s optimized protocols. The experimental procedures were as follows: HGC-27 and AGS cells were gently washed with ice-cold PBS, then dissociated using Trypsin solution (0.25%) without EDTA (T1350, Solarbio, Beijing, China) at 37 °C through gradient digestion. After digestion, the cell suspension was centrifuged at 300×g for 5 min using a Beckman Coulter Allegra X-15R centrifuge. Cell pellets were resuspended in a pre-cooled Annexin V Binding Buffer and adjusted to a density of 2 × 10⁵ cells/mL. Each experimental group received 5 μL Annexin V-647 fluorochrome and 5 μL PI staining solution (working concentration: 50 μg/mL), with parallel preparation of unstained controls and single-color compensation controls. After gentle mixing, samples underwent light-protected incubation at 4 °C for 15 min, with strict avoidance of vortex agitation to preserve membrane integrity. The fluorescence was detected within 60 min post-staining using a CytoFLEX LX flow cytometer (Beckman Coulter, Brea, CA, USA). Apoptotic indices, defined as the cumulative percentage of early apoptotic (Annexin V⁺ PI⁻) and late apoptotic (Annexin V⁺ PI⁺) populations within the viable cell gate, were calculated using CytExpert 2.4 software with optimized doublet discrimination parameters.

Dose-response curve

GC cells were seeded in 96-well plates at a density of 7 × 10 cells/well and incubated for 24 h until they reached 70% confluence. The cells were then treated with graded concentrations of 5-aza-CdR (0–40 μM) for 24–96 h. After treatment, cell viability was assessed using the CCK-8 assay (Dojindo) by adding the reagent to the culture medium at a 1:10 dilution ratio, followed by incubation at 37 °C for 1 h. Absorbance was measured at 450 nm using a microplate reader. Dose–response curves (logarithmic concentration vs. normalized response) were generated by fitting the data to a sigmoidal dose–response model (variable slope) using nonlinear regression analysis.

Statistical analysis

Data processing, statistical analyses, and visualization were performed using the R software (v4.2.1). Normal distributions were analyzed using unpaired Student’s t-tests, while non-normal distributions were analyzed using the Wilcoxon rank-sum test. Pearson’s correlation coefficient was used to evaluate the association between two continuous variables. Given the potential influence of skewed data, Spearman’s correlation analysis was conducted to ensure a comprehensive assessment of variable relationships. A two-tailed p-value < 0.05 was defined as the threshold for statistical significance.

ARID1A methylation and expression in GC

Aberrant DNA methylation patterns are a recognized epigenetic mechanism that drives oncogenic transcriptional dysregulation (Torres et al., 2016). To investigate the causal relationship between ARID1A aberrant expression and methylation dynamics, we performed a multi-platform bioinformatics analysis. Although ARID1A promoter methylation levels through UALCAN (Chandrashekar et al., 2022) showed a trend toward reduction in normal gastric mucosa, this difference did not reach statistical significance (p > 0.05; Fig. 1A). Genomic correlation analysis via cBioPortal demonstrated a robust inverse association between ARID1A methylation density and mRNA expression levels (Spearman’s ρ =  − 0.29, p = 2.06 × 10⁻⁸; Fig. 1B). Complementary evidence from MEXPRESS (Koch et al., 2019) further confirmed this anti-correlation pattern, particularly emphasizing that hypermethylation of CpG islands within the ARID1A promoter region was inversely associated with gene expression in gastric adenocarcinoma specimens (Fig. 1C). These convergent findings suggest that promoter hypermethylation serves as a potential regulatory mechanism underlying ARID1A suppression in gastric carcinogenesis.

To refine the macro-level dysregulation of methylation into site-specific microvariations, we performed additional validation using the UCSC Xena database. Stratification by ARID1A expression levels revealed significant inverse correlations between transcriptional activity and methylation at promoter-associated probes cg04081153 (p-value = 0.0002, FDR-adjusted p-value = 0.0008, Fig. 2H) and cg05445839 (p-value < 0.0001, FDR-adjusted p-value ≤ 0.0008, Fig. 2I). Consistent with this, samples grouped by methylation levels at these loci demonstrated lower ARID1A expression in the hypermethylated subgroups (p-value = 0.0001, FDR-adjusted p-value = 0.0001, Fig. 2J; p-value <0.0001, FDR-adjusted p-value < 0.0001, Fig. 2K). Complementary analysis via MEXPRESS identified 12 CpG sites exhibiting significant negative correlations with ARID1A expression (Pearson’s R: −0.269 to −0.122), including cg25601319 (R = −0.194, p < 0.001), cg14851949 (R = −0.122, p < 0.05), cg24532126 (R = −0.268, p < 0.001), cg05445839 (R = −0.123, p < 0.05), cg11856093 (R = −0.177, p < 0.01), and cg00371107 (R = −0.265, p < 0.001) (Fig. 2A). These findings highlight cg05445839, a recurrently hypermethylated probe within the ARID1A promoter, as a potential therapeutic target for restoring physiological ARID1A expression in gastric adenocarcinoma through epigenetic modulation.

Comprehensive profiling and functional annotation of methylation-driven gene networks in GC

Differential expression analysis between the ARID1A promoter methylation-defined subgroups (Hyper-group vs. Hypo-group) identified 191 methylation-regulated differentially expressed genes (MeDEGs), comprising 37 upregulated and 154 downregulated genes meeting the established significance thresholds (FDR-adjusted p-value < 0.05, |log₂FC| ≥ 1.5) (Table S1). Transcriptional alterations were visualized using a volcano plot highlighting the MeDEG distribution (Fig. 3A), while hierarchical clustering of the top 50 statistically significant MeDEGs demonstrated robust subgroup segregation (Fig. 3B). Furthermore, a correlation matrix delineated the co-expression patterns between the 10 most significantly upregulated and downregulated MeDEGs, revealing potential regulatory networks influenced by ARID1A promoter hypermethylation (Fig. S1A). To elucidate the biological significance of the MeDEGs in gastric carcinogenesis, we performed integrative GO and KEGG pathway enrichment analyses on the 191 identified candidates. GO analysis revealed nine enriched BP, eight CC, and nine MF, with cell adhesion, apoptotic regulation, and signaling receptor activity identified as the most significantly enriched terms, along with immune-related pathways (Fig. 3C). KEGG pathway mapping further demonstrated the predominant enrichment of oncogenic signaling cascades, including cell-matrix adhesion, PI3K-AKT activation, and mTOR-mediated proliferation (Fig. 3D). PPI network reconstruction using STRING and Cytoscape identified key regulatory hubs (Fig. S1B), with the top 10 centrality-ranked nodes (closeness metrics) highlighting potential master regulators (Fig. 3E). Modular decomposition using the MCODE algorithm resolved two functionally cohesive subnetworks implicated in distinct carcinogenic processes (Figs. 3F–3G). Comparative analysis of simple nucleotide variation data from TCGA revealed distinct mutational architectures between ARID1A hypermethylated and hypomethylated GC cohorts. In the hypermethylated subgroup, the five most frequently mutated genes were TTN (54.9%), TP53 (45.1%), MUC16 (34.1%), LRP1B (26.8%), and ARID1A (25.6%) (Fig. S1C). Conversely, the hypomethylated cohort exhibited predominant alterations in TTN (53.3%), TP53 (50.7%), LRP1B (32.0%), MUC16 (30.7%), and CSMD3 (28.0%) (Fig. S1D).

GSEA of MeDEGs

Stratification based on ARID1A promoter methylation β-values identified 16,030 MeDEGs. GSEA revealed significant associations with hallmark oncogenic pathways, including apoptotic regulation, immune response, NF-kappa B signaling, and T-cell receptor signaling, with NES and p-value annotated for each pathway (Figs. 4A–4L).

ARID1A methylation and tumor immune microenvironment modulation

Integrated analysis of the TISIDB database revealed significant correlations between ARID1A expression and chemokine/chemokine receptor dynamics in GC (Figs. 5A–5B). We found that ARID1A expression positively correlated with several chemokine/chemokine receptor genes, as illustrated in Figs. 5C–5G: CXCL14 (r = 0.102, p = 0.0374), CCR7 (r = 0.134, p = 0.00616), CCR9 (r = 0.105, p = 0.032), CXCR5 (r = 0.131, p = 0.0075), and XCR1 (r = 0.103, p = 0.0363). Conversely, ARID1A levels were inversely associated with a broad spectrum of chemokine ligands and their receptors (Figs. S2A–S2T).

ARID1A expression demonstrated a complex immunoregulatory signature in GC, as overviewed in the correlation heatmap (Figs. 5H–5I). Specifically, ARID1A levels showed significant positive correlations with several key immunomodulatory molecules, including the immunostimulators CD40LG (r = 0.102, p = 0.037; Fig. 5J), TNFRSF13B (r = 0.11, p = 0.0249; Fig. 5K), TNFRSF13C (r = 0.244, p = 5.11e-7; Fig. 5L) and TNFRSF25 (r = 0.18, p = 0.000228; Fig. 5M), as well as the immunoinhibitors ADORA2A (r = 0.146, p = 0.00294; Fig. 5N) and CD160 (r = 0.154, p = 0.00168; Fig. 5O). Conversely, ARID1A was inversely associated with a broader set of immunomodulatory molecules (e.g., HAVCR2, IL-10, LGALS9, CD276; complete details of all significant negative correlations are provided in Figs. S3A–S3N). This bidirectional immunoregulatory signature establishes ARID1A as a key modulator of tumor-immune interactions in gastric cancer.

To investigate the influence of ARID1A promoter hypermethylation on immunomodulation, we systematically investigated immune infiltration patterns across methylation-defined subgroups using seven computational algorithms. CIBERSORT deconvolution revealed notable differences in lymphoid populations; the hypomethylated subgroup exhibited elevated infiltration of resting memory CD4⁺ T cells (p < 0.0001) and resting dendritic cells (p < 0.05), along with reduced M0 macrophage abundance (p < 0.0001). Consensus across the MCP-counter, TIMER, and Quantiseq algorithms identified significant B-cell infiltration divergence between subgroups, suggesting methylation-dependent regulation of humoral immunity. Stratification using the ESTIMATE algorithm revealed higher immune scores (p < 0.05) and ESTIMATE scores (p < 0.05), along with lower tumor purity (p < 0.05) in the hypermethylated cohort. Immunophenoscore (IPS) analysis further revealed subgroup-specific disparities in CP and AZ compartments (Fig. 6A).

Given the significant divergence in immune infiltration between the two clusters, we further evaluated their association with canonical immune checkpoints. Immune checkpoint analysis revealed distinct expression patterns of CTLA-4 (p < 0.05) and BTLA (p < 0.05) across methylation-defined subgroups (Fig. 6B). Emerging evidence has highlighted the interaction between tumor-associated sialoglycans and sialic acid-binding immunoglobulin-like lectins (Siglecs) in infiltrating immune cells as a novel immune checkpoint axis, offering therapeutic potential in cancer immunotherapy (Stanczak & Läubli, 2023). Supporting this paradigm, Haas et al. (2019) demonstrated that Siglec-7 and Siglec-9 suppress T cell activation upon TCR(T-cell receptor) stimulation (Stanczak et al., 2018). Notably, our Siglec profiling identified elevated expression of Siglec-7 (p < 0.01) and Siglec-9 (p < 0.001) in the hypermethylated subgroup compared to their hypomethylated counterparts, suggesting methylation-dependent regulation of this immunosuppressive axis (Fig. 6B).

Promoter hypermethylation inhibits ARID1A expression in gastric cancer cells

Epigenetic modifications constitute one of the primary mechanisms underlying gene silencing (Long et al., 2023). To determine whether ARID1A can be regulated by DNA methylation, we examined the ARID1A promoter sequence using the NCBI database and predicted it with the MethPrimer database. Bioinformatics prediction using the MethPrimer identified three hypermethylated CpG islands in the ARID1A promoter region (Fig. 7A). Among these, we selected a CpG island adjacent to the transcription start site (TSS: chr1:26692500) for primer design and subsequent validation using MSP. MSP detected exclusive amplification of methylated ARID1A promoter fragments in HGC-27 cells, while AGS cells exhibited both methylated and unmethylated amplification products (Fig. 7B). Dose–response profiling across a 0–40 μM 5-aza-CdR gradient demonstrated concentration-dependent suppression of HGC-27 proliferation (IC50 = 18.57 μM; Fig. 7E). Conversely, AGS cells showed no significant proliferation suppression even at the maximum tested concentration (40 μM) of 5-aza-CdR (Fig. S4A). Given the biallelic methylation pattern (partial promoter methylation) (Fig. 7B) and low baseline ARID1A expression in AGS cells (Figs. 7C–7D), pharmacological demethylation with 5-aza-CdR failed to significantly restore ARID1A expression (Fig. S4B). This resistance likely stems from incomplete epigenetic silencing, as evidenced by residual unmethylated alleles that permit constitutive low-level transcription. Conversely, HGC-27 cells exhibited complete ARID1A promoter hypermethylation coupled with reversible transcriptional silencing (post-5-aza-CdR), making them an optimal model for investigating methylation-dependent oncogenic mechanisms.

CCK-8 proliferation assays revealed significant suppression of HGC-27 cell proliferation at 10 μM 5-aza-CdR (p < 0.01, Fig. 7F). This observation suggests that maximal epigenetic efficacy is achieved at 10 μM, beyond which cytotoxicity-driven effects dominate. Consequently, 10 μM was chosen for subsequent experiments to avoid nonspecific toxicity while maintaining effective target modulation. Time- and concentration-resolved analyses identified 10 μM 5-aza-CdR with 48 h exposure as the optimal regimen, achieving maximal ARID1A transcriptional reactivation (Fig. 7H) and promoter demethylation (Fig. 7G). Prolonged treatment (>48 h) paradoxically increased methylation (Fig. 7G) and reduced ARID1A protein levels (Figs. 7I–7J), indicating a potential escape from the initial epigenetic modulation.

ARID1A depletion promotes GC progression

To evaluate the tumor-suppressive role of ARID1A in GC, we performed stable ARID1A knockdown in HGC-27 and AGS cell lines. Western blot and qPCR analyses confirmed significant reductions in ARID1A protein (Figs. 8A–8B) and mRNA levels (Fig. 8C) in the HGC-27 cell line, with similar trends observed in AGS cells (Figs. S5A–S5C). Transwell assays demonstrated that ARID1A silencing enhanced the migratory (p < 0.0001) and invasive capacities significantly (p < 0.0001) in HGC-27 cells (Figs. 8D–8F). Consistent findings were observed in AGS cells with a significant increase in migration (p < 0.0001) and invasion (p < 0.001) post-knockdown (Figs. S5D–S5F). Wound healing assays further corroborated the accelerated migration rates in both ARID1A-depleted cell lines (HGC-27: p < 0.05; AGS: p < 0.0001) (Figs. 8G–8H, Figs. S5G–S5H). CCK-8 proliferation assays revealed significant growth potentiation in ARID1A-deficient GC cells (Fig. 8L; Fig. S5L). Long-term clonogenic survival assays confirmed a sustained proliferative advantage (HGC-27: p < 0.01; AGS: p < 0.0001) (Figs. 8I–8J, Figs. S5I–S5J). Flow cytometric analysis using Annexin V-647/PI staining showed significant suppression of apoptosis in ARID1A-silenced cells (HGC-27: p < 0.01; AGS: p < 0.01) (Figs. 8K, 8M; Figs. S5K, S5M). These findings mechanistically link ARID1A loss to proliferative enhancement and apoptotic resistance in GC progression.

Functional impact of ARID1A demethylation in gastric cancer

Pharmacological demethylation with 10 µM 5-aza-CdR in HGC-27 cells reversed ARID1A promoter hypermethylation, significantly impairing metastatic potential. Transwell assays indicated a marked reduction in both migration and invasion compared to controls (Figs. 9A–9C). Clonogenic survival assays showed greatly reduced colony formation (Figs. 9D–9E). This anti-proliferative effect was consistent with the substantial growth suppression revealed by CCK-8 assays (Fig. 9H). Flow cytometry analysis demonstrated a significant increase in apoptosis relative to control cells (Figs. 9F–9G). Together, these results identify ARID1A promoter hypermethylation as a critical driver of gastric cancer progression and highlight the therapeutic potential of 5-aza-CdR in HGC-27 cells via promoting apoptosis and suppressing proliferation and invasion.

ARID1A methylation drives immune evasion via PD-L1 upregulation and PI3K/AKT/mTOR activation in GC

PD-L1, a key immunosuppressive checkpoint molecule, is elevated in GC and is correlated with adverse clinical outcomes (Wu et al., 2006; Hou et al., 2014). Previous studies have demonstrated elevated PD-L1 expression in ARID1A-mutated tumors (Kim et al., 2019). In our experimental models, ARID1A knockdown in HGC-27 cells induced an increase in PD-L1 expression (p < 0.01) (Figs. 10A–10C), with similar PD-L1 upregulation observed in AGS cells (p < 0.0001) (Figs. S6A–S6C). Notably, pharmacological demethylation of HGC-27 cells with 5-aza-CdR restored ARID1A expression while suppressing PD-L1 (Figs. 10D–10F).These findings suggest an inverse regulatory relationship between ARID1A and PD-L1 in the studied models, implying that ARID1A loss-of-function may reprogram immune checkpoint signaling in gastric cancer.

Based on our earlier profiling data that identified PI3K–AKT–mTOR signaling as enriched in ARID1A-hypermethylated tumors (Fig. 3D), we further investigated its role as a mechanistic link. Transcriptomic analysis of TCGA cohorts confirmed PI3K/AKT pathway activation in ARID1A-hypermethylated tumors, marked by elevated expression of PIK3R1, PIK3R2, PIK3CA, and AKT3 compared to hypomethylated controls (Figs. S6E, S6F, S6G, and S6J, respectively). Accordingly, the restoration of ARID1A expression by 5-aza-CdR treatment in HGC-27 cells suppressed PI3K/AKT/mTOR pathway activation, as evidenced by reduced phosphorylation of PI3K, AKT, and mTOR (Figs. 10D, 10G–10L).

To functionally validate this mechanism, we performed a rescue assay in which SC79, an AKT agonist, was applied to 5-aza-CdR-demethylated HGC-27 cells, resulting in reactivated AKT/mTOR phosphorylation and restored PD-L1 expression (Figs. 10M–10S). A comprehensive rescue experiment involving four experimental groups further established the functional hierarchy of the ARID1A–PI3K/AKT–PD-L1 axis (Figs. 10T–10V): the control group showed baseline ARID1A and PD-L1 levels, SC79 alone enhanced PD-L1 expression confirming pathway sufficiency, 5-aza-CdR alone restored ARID1A and suppressed PD-L1, and most crucially, the addition of SC79 reversed 5-aza-CdR-induced PD-L1 suppression, confirming that PI3K/AKT activation is necessary for PD-L1 upregulation following ARID1A methylation.