Gene expression data, associated mutations, and clinical data of GC patients were downloaded from TCGA (https://portal.gdc.cancer.gov/). Series matrix file of GSE13195, GSE66229, GSE15459, GSE26253 and GSE84433 were downloaded from the GEO dataset (https://www.ncbi.nlm.nih.gov/geo/). The differential HMEs were determined using limma R package with adjusted P-value less than 0.05. A tissue microarray comprising 86 paired GC tumor tissues and adjacent normal specimens, supplemented with 8 additional GC cases, was purchased from Shanghai Qutdo Biotech Company (Shanghai, China). The research related to human use has been complied with all the relevant national regulations, institutional policies and in accordance with the tenets of the Helsinki Declaration, and has been approved by the Ethics Committee of the Shandong Cancer Hospital and Institute (Jinan, China).

AGS cells were obtained from the Cell Bank of the Chinese Academy of Science (Shanghai, P.R. China). MKN-45 cells were obtained from the National Infrastructure of Cell Line Resource (Beijing, P.R. China). AGS cells were cultured in Ham’s F12-medium (HyClone, Logan, UT, USA) supplemented with 12% fetal bovine serum (Gibco, Carlsbad, CA, USA). MKN-45 cells were cultured in RPMI-1640 medium (Gibco) containing 10% fetal bovine serum. All cell lines were incubated under standardized conditions in a humidified atmosphere (37°C, 5% CO₂) using a Thermo Fisher Scientific HERAcell 150i incubator (Waltham, MA, USA). All cells were authenticated by short tandem repeat profiling, and tested free from Mycoplasma.

A total of 74 HMEs, listed in Supplementary Table 1 , were obtained from previous publications (10). 34 shared HMEs between TCGA and GEO GC expression datasets were selected for clustering analysis using the “ConsensusClusterPlus” R package. Three novel GC subtypes were identified based on the cumulative distribution function curve parameters, which exhibited a gradual and smooth increase.

Functional enrichment analysis for KEGG and GO terms was performed using the clusterProfiler R package.

Stromal score, Immune score, and ESTIMATE score for each GC patient were calculated from gene expression profiles using the ESTIMATE R package (31). Immune cell composition in GC tissues was quantified using the CIBERSORT algorithm (32).

The TIDE scores and the Microsatellite Instability status of GC samples were assessed utilizing the TIDE online platform (http://tide.dfci.harvard.edu/login/) (33, 34).

Lipofectamine 2000 (Invitrogen, Carlsbad, CA, USA) was used to transfect siRNAs into GC cells. Cells were collected 72 hours post-transfection. RT-PCR was used to verify the transfection efficiency. The siRNA sequences were listed as follows: si-SMYD2-1 5’-CUCGGAGACUGUAAGACUATT-3’ and si-SMYD2-2 CCCAACGGAAGAUAGAAAUTT-3’.

H. pylori strains 26695 and 11637 were cultured in Brucella broth supplemented with 5% fetal bovine serum at 37°C under microaerophilic conditions. AGS and MKN-45 cells were infected with H. pylori at varying concentrations and collected at specified time points.

Trizol reagent (Invitrogen, Carlsbad, CA, USA) was used to extract RNAs following the manufacturer’s instructions. RT reagent Kit gDNA Eraser (Takara, Shiga, Japan) was employed to reverse mRNAs to cDNA. Finally, SYBR Green (Takara) was used for qRT-PCR of cDNAs. Primer sequences were listed as follows: SMYD2 forward 5’-TGTGTTTGAGGACAGTAACGTG-3’ and reverse 5’-GAGGGAGTACAAAGGATAGTGC-3’; β-actin forward 5’ - AGTTGCGTTACACCCTTTCTTG - 3’ and reverse 5’ - CACCTTCACCGTTCCAGTTTT - 3’.

500 cells per well with different treatments were plated in 6-well plates and cultured for 1–2 weeks. Colonies were fixed with methanol and stained with Giemsa solution. Colonies containing > 50 cells were counted for analysis.

1200 cells per well with different treatments were plated in 96-well plates and cultured for 24, 48, or 72 hours for CCK8 assay. CCK8 reagent (Med Chem Express, Monmouth Junction, NJ, USA) was added to the conditioned medium and incubated for 3 hours. Absorbance was measured at 450 nm using a spectrophotometer. Both assays were performed in triplicate.

For wound healing assays, cells (5 × 10⁵) with different treatments were seeded into 12-well plates and grown to confluence. A sterile pipette tip was used to create a linear scratch. Wound width was measured at 0 and 48 hours. Wound closure rate was calculated as: (wound width at 0 hours – wound width at 48 hours)/wound width at 0 hours × 100%. An inverted phase-contrast microscope was used to view the cells and photograph.

For migration assays, cells (5 × 10⁴) were seeded into the upper chamber of a Matrigel-coated transwell insert. The medium containing 20% FBS was added to the lower chamber. After 48 hours, methanol and crystal violet staining solution were used to fix and stain the cells respectively.

Tissue microarray on glass slides was subjected to deparaffination and dehydration. Then, the array was subjected to epitope retrieval. Next, the array was treated with H₂O₂ and blocked in goat serum for 30 minutes. Primary antibody of SMYD2 was used to incubate the array at 4°C overnight. The array was incubated with the corresponding secondary antibody on the second day. Finally, a DAB staining kit (Vector Laboratories, Burlingame, CA, USA) was used to stain. Images were used to assess the IHC score. The intensity of positive staining was scored as follows: 0 (negative), 1 (weak), 2 (moderate) and 3 (strong). A scale from 0 to 3 was used to score the proportion of positively stained cells: 0 (0%), 1 (<25%), 2 (25-75%) and 3 (>75%). Final IHC score = Intensity score × Proportion score.

Lysis buffer supplemented with protease inhibitors was used to extract proteins from cells. The lysates were resolved by SDS-PAGE. Protein was then transferred to PVDF membranes, which were blocked with 5% non-fat milk powder and subsequently incubated with primary antibodies overnight at 4°C. Following this, PVDF membranes were incubated with appropriate secondary antibodies. Signals were detected using an ECL detection reagent (Millipore, St. Louis, MO, USA). The primary antibody used was SMYD2 (D14H7; Cell Signaling Technology, Danvers, MA, USA).

Data were presented as mean ± SEM from three independent experiments. Mann-Whitney U-test or Student’s t-test was used to compare variables between experimental group and control group by GraphPad PRISM version 9 (San Diego, CA, USA). SPSS version 23.0 (IBM, Armonk, NY, USA) was used to analyze clinical data. P < 0.05 was considered statistically significant.

In order to investigate the association between HMEs and GC, we performed a comprehensive analysis of somatic mutation frequencies for 74 HMEs using TCGA-STAD data. The results indicated a relatively high mutation frequency in GC samples. In detail, 171 GC samples (90.96%) had HME mutations. Among these, KMT2D exhibited the highest mutation frequency (36%), followed by KMT2C, KMT2B, KMT2A, JARID2, JMJD1C, KMT2E, PHF2, NSD3, SETDB1, NSD2, PRDM16 and HR ( Figure 1A ). Next, we compared expression of HMEs between GC tissues and adjacent normal samples from the TCGA cohort. 24 HMEs were differentially expressed, with 15 up-regulated and 9 down-regulated genes in GC ( Figure 1B ). Similar trends were observed in the GSE13195 and GSE66229 datasets ( Supplementary Figure 1A ). Correlation analysis revealed moderately correlated HMEs expression patterns between TCGA and GEO datasets ( Figure 1C ; Supplementary Figure 1B ). We categorized GC patients into three subtypes based on the expression profiles of HMEs using a consensus clustering algorithm ( Figure 1D ; Supplementary Figures 2A, B ). Survival analysis indicated that patients with GC in cluster 2 (C2) had a higher survival probability than those in cluster 1 (C1) and cluster 3 (C3) ( Figure 1E ; Supplementary Figure 2C ). To identify subtype-associated pathways, gene set variation analysis was performed comparing C1 + C3 with C2. The results showed that C1+C3 was significantly enriched in cell cycle and p53 pathways ( Figure 1F ). These findings collectively highlighted the critical regulatory role of HMEs in GC tumorigenesis and progression.

We have found that the expression of some HMEs was different in GC tissues compared to para-carcinoma tissues. To evaluate their diagnostic potential, we systematically analyzed HMEs expression profiles in TCGA, GSE13195 and GSE66229 datasets, respectively. Consistent intersection analysis revealed 9 consistently up-regulated and 1 down-regulated HME across all datasets ( Figure 2A ; Supplementary Figure 3A ). We used 9 up-regulated HMEs to train the diagnostic model with ten-fold cross-validation using Random Forest, Logistic Regression, Support Vector Machine, Naive Bayes, Linear Discriminant Analysis, Bagging, and Gradient Boosting Decision Tree ( Supplementary Figure 3B ). Notably, all models demonstrated high diagnostic accuracy with AUC values ranging from 0.95 to 0.96 ( Figure 2B ). Feature selection via Random Forest reduced the biomarker set to six HMEs using an importance threshold of 0.075 ( Figure 2C ). The simplified 6-HME model maintained comparable performance with AUC 0.96 ( Figure 2D ). Validation in TCGA and GSE13195 cohorts confirmed robust diagnostic capability, achieving AUC values of 0.88 and 0.96 respectively ( Figure 2E ). Collectively, these findings demonstrated that HME-based biomarkers exhibit strong potential for gastric cancer diagnosis.

To identify pivotal regulators of GC development, we focused on the six HMEs including SETDB1, PRMT1, PRMT5, PRMT3, HSPBAP1 and SMYD2. There are fewer studies on SMYD2 than the other genes. Notably, elevated SMYD2 expression correlated significantly with poorer patient prognosis ( Figure 3A ). SMYD2 was also up-regulated across multiple cancer types ( Figure 3B ). Comparative analysis of paired tumor and para-carcinoma tissues showed SMYD2 overexpression across most cancer types, except for lung adenocarcinoma ( Figure 3C ). Furthermore, SMYD2 expression was positively related to tumor mutation burden in GC ( Supplementary Figure 4A ). The high-SMYD2 group exhibited higher mutation counts ( Figure 3D ) and increased mutational co-occurrence compared to the low-SMYD2 group ( Supplementary Figure 4B ).

To identify SMYD2-associated genes and pathways, we performed comparative transcriptomic analysis of TCGA and GEO datasets, analyzing differentially expressed genes (adjusted p-value < 0.05) between SMYD2-high and SMYD2-low expression groups (stratified by median expression cutoff) ( Figure 4A ). Venn diagram analysis identified 252 consistently up-regulated genes ( Figure 4B ) and 277 down-regulated genes ( Supplementary Figure 5A ). GO enrichment analysis revealed that up-regulated genes were predominantly associated with mitochondrial function, mitotic processes and cell cycle checkpoint pathways ( Figure 4C ). Concurrently, KEGG pathway analysis demonstrated significant enrichment of up-regulated genes in proteasome activity, cell cycle regulation and DNA replication pathways ( Figure 4D ). For down-regulated genes, both GO and KEGG analyses showed primary enrichment in biological processes related to cell-substrate adhesion, cytoskeletal organization, and focal adhesion pathways ( Supplementary Figures 5B, C ). These collective findings suggest that SMYD2 may modulate signaling networks governing tumor cell proliferation and metastatic processes.

To explore the association between SMYD2 expression and tumor immune infiltration in GC patients, we performed CIBERSORT analysis in TCGA and GEO datasets. SMYD2 expression positively correlated with CD8⁺ T cells, activated CD4⁺ T cells, and M0/M1 macrophage infiltration, but negatively correlated with M2 macrophages and regulatory T cells (Tregs) ( Figure 5A ). Subsequent analysis revealed SMYD2 positively correlated with tumor purity while inversely correlating with immune score, stromal score, and ESTIMATE score ( Figure 5B ). Using the TIDE algorithm (34), we evaluated immunotherapy response potential, where lower TIDE scores indicate greater predicted benefit. High SMYD2 expression associated with reduced TIDE, dysfunction, and exclusion scores ( Figure 5C ), suggesting enhanced immunotherapy sensitivity in SMYD2-high GC patients. These findings implicate SMYD2 as a predictive biomarker for immunotherapy efficacy in GC.

We measured SMYD2 expression in human GC samples and adjacent normal samples using a tissue microarray. Results showed that SMYD2 expression was significantly elevated in GC samples ( Figures 6A, B ). ROC curve analysis revealed SMYD2’s discriminative capacity for GC diagnosis, yielding an AUC of 0.877 ( Figure 6C ). Association analysis between SMYD2 expression and clinicopathological characteristics revealed significant correlation with Borrmann subtype (P = 0.005), while showing non-significant associations with other parameters ( Supplementary Table 2 ).

We next explored the functions of SMYD2, and we performed siRNA-mediated knockdown in GC cells ( Figures 7A, B ). SMYD2 suppression significantly attenuated cellular proliferation, as evidenced by colony formation and CCK-8 assays ( Figures 7C, D ). Furthermore, SMYD2 silencing markedly impaired invasion and migration capacities in Transwell and wound healing assays, respectively ( Figures 7E, F ). These results reiterated the positive role of SMYD2 in tumorigenesis and progression of GC in vitro.

Infection of H. pylori is a well-known risk factor for GC (35, 36). Therefore, we explored whether SMYD2 expression was regulated by H. pylori. We detected SMYD2 expression after H. pylori strains (26695 and 11637) infection in AGS and MKN-45 cells. The results showed that SMYD2 expression was significantly up-regulated after infection with H. pylori at different MOIs. Meanwhile, SMYD2 expression was also increased at different time points after infection with H. pylori ( Figures 8A, B ). This finding was corroborated by bioinformatics analysis of GEO datasets, demonstrating significantly elevated SMYD2 levels in H. pylori-positive clinical specimens compared to uninfected controls ( Supplementary Figure 6A ).