The GC cell lines AGS and HGC-27 were purchased from Procell Life Science & Technology Co., Ltd. Both cell lines were cultured in RMPI-1640 medium (cat. no. SH30809.01; HyClone; Cytiva) supplemented with 10% fetal bovine serum (cat. no. 10099141; Gibco; Thermo Fisher Scientific, Inc.) and 1% penicillin-streptomycin (cat. no. C0222; Beyotime Institute of Biotechnology) at 37°C in an atmosphere containing 5% carbon dioxide. Mycoplasma contamination was assessed every 2 weeks using a Mycoplasma staining detection kit (cat. no. C0296; Beyotime Institute of Biotechnology).

The gene expression data and clinical follow-up data of 370 patients with GC were obtained from TCGA database (TCGA-STAD dataset). The GSE66229 (22), GSE84433 (23) and GSE163558 (24) datasets were downloaded from the Gene Expression Omnibus (GEO) database. The data in these datasets were analyzed using the GPL570 HG-U133_Plus_2 Affymetrix Human Genome U133 Plus 2.0 Array (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GPL570) and the GPL6947 Illumina HumanHT-12 V3.0 expression beadchip (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GPL6947), respectively. The gene expression data and clinical follow-up data of 300 and 357 patients with GC were retained from the two datasets, respectively. Bioinformatics analyses were conducted using the following public databases: Tumor Immune Dysfunction and Exclusion (TIDE; http://tide.dfci.harvard.edu/) for predicting immune checkpoint inhibitor response; Tumor IMmune Estimation Resource (TIMER2.0; http://cistrome.shinyapps.io/timer/) for evaluating immune cell infiltration; TCGA (https://portal.gdc.cancer.gov/) for acquiring multi-omics GC data; GEO (https://www.ncbi.nlm.nih.gov/geo/) for validating differential gene expression; Kyoto Encyclopedia of Genes and Genomes (KEGG; http://www.kegg.jp/) for pathway enrichment analysis of differentially expressed genes; and Gene Ontology (GO; http://geneontology.org/) for functional annotation of gene sets.

Single-cell and transcriptome-wide sequencing were performed using the R software (version 4.5.1; http://www.r-project.org/). For single-cell RNA sequencing (scRNA seq), cell-level QC thresholds were applied as follows: cells with nCount-RNA between 1,000 and 60,000 and those with 500–7,000 genes (nFeature-RNA) were retained, whereas cells with a mitochondrial gene percentage of >20% were excluded. For gene-level QC thresholds, genes expressed in ≥5 cells were retained, and batch effects were corrected using the Harmony package (version 1.2.3; http://cran.r-project.org/web/packages/harmony/). Batch effects in expression data were corrected through quantile normalization using the normalizeBetweenArrays function in the limma package (version 3.46.0; http://bioconductor.org/packages/release/bioc/html/limma.html). Data reliability was validated through principal component analysis (PCA) implemented using the FactoMineR package (version 2.12; http://cran.r-project.org/web/packages/FactoMineR/), and PCA plots were generated using the ggplot2 package (version 3.5.2; http://cran.r-project.org/web/packages/ggplot2/). Differential expression analysis was performed using the limma or DESeq2 package (version 1.49.4; http://bioconductor.org/packages/release/bioc/html/DESeq2.html); prognostic analysis was performed using the survival (version 3.8–3; http://cran.r-project.org/web/packages/survival/) and survminer (version 0.5.0; http://cran.r-project.org/web/packages/survminer/) packages; weighted gene co-expression network analysis (WGCNA) was performed using the WGCNA package (version 1.73; http://cran.r-project.org/web/packages/WGCNA/); cell-cell communication was analyzed using CellChat (version 2.1.2; http://github.com/jinworks/CellChat); pseudotime series analysis was performed using Monocle2 (version 2.36.0; bioconductor.org); and copy number variations (CNVs) were analyzed using the inferCNV package (version 1.24.0; http://bioconductor.org/packages/release/bioc/html/infercnv.html). Data were visualized using the ggplot2, ComplexHeatmap (version 2.25.2; bioconductor.org), and Seurat packages (version 5.3.1; cran.r-project.org). A flowchart demonstrating the detailed data processing protocol is shown in Fig. 1. scRNA QC parameters included nCount-RNA, nFeature-RNA, mitochondrial gene percentage and the number of genes expressed per cell.

AGS and HGC-27 cells in the logarithmic growth phase were seeded in 24-well plates and transfected when the cell density reached 80%. A total of 1.25 µl of 20 µM RRM2 siRNA (a negative control siRNA was included in the same kit; cat. no. HY-RS12286; MedChemExpress) was diluted with serum-free medium to a final volume of 50 µl, gently mixed, and incubated at room temperature for 5 min (solution A). Subsequently, 2.5 µl of siRNA/miRNA transfection reagent (cat. no. HY-K2017; MedChemExpress) was diluted with serum-free medium to a final volume of 50 µl, mixed gently and incubated at room temperature for 5 min (solution B). Solution B was gradually added dropwise to solution A, mixed gently and incubated at room temperature for 20 min to form a transfection complex (this mixture was used within 60 min). After the cell culture medium was removed, 400 µl of serum-free medium and 100 µl of the transfection complex were sequentially added and gently mixed. After 6 h, the medium was replaced with a complete medium containing serum, and incubation was continued. Protein extraction and western blotting were performed to assess RRM2 knockdown efficiency.

Cells in the logarithmic growth phase were seeded in a 6-well plate at a density of 500–1,000 cells/well, mixed evenly, and cultured for 14 days at 37°C. After colonies were visible, the medium was removed, and the plates were washed twice with phosphate-buffered saline (PBS). Thereafter, the cells were fixed with 4% paraformaldehyde (cat. no. P0099-3L; Beyotime Insitute of Biotechnology) for 20 min at room temperature and stained with 0.1% crystal violet (cat. no. Y268091; Beyotime Insitute of Biotechnology) for 20 min at room temperature. After rinsing and air-drying, the cells were imaged and counted using the ImageJ software (version 2.14.0; National Institutes of Health).

Cells (3×10⁵) in the logarithmic growth phase were harvested, and the culture supernatant and digested cells were centrifuged separately at 300 × g for 5 min at 4°C. The supernatants were discarded, and the pellets were combined, resuspended in binding buffer, and incubated with Annexin V-FITC and PI (cat. no. C1062M; Beyotime Insitute of Biotechnology) at 4°C for 15 min in the dark. Flow cytometry (BD FACSVerse; BD Biosciences) was used to calculate the proportion of early apoptotic cells in the lower right quadrant (Annexin V-FITC⁺ and PI⁻), and the data were analyzed with FlowJo 10.4 (BD Biosciences).

Transwell inserts (cat. no. 3422; Corning, Inc.) were purchased in advance. Logarithmically growing cells were digested and resuspended in serum-free medium. A total of 2,000 cells were counted and seeded into the upper chamber. The lower chamber was filled with complete medium containing 10% FBS. After 24 h of culture at 37°C, the inserts were removed, washed twice with FBS, fixed with 4% paraformaldehyde for 20 min at room temperature and stained with crystal violet for 20 min at room temperature. Non-migrated cells were removed with a cotton swab, allowed to dry, and images were captured under a fluorescence microscope (Olympus IX73; Olympus Corporation). Cell counting and analysis were performed using ImageJ (version 2.14.0) software.

Logarithmically growing cells were seeded in a 6-well plate. When the cell confluence reached above 95%, a straight scratch was made in the cell layer using a sterile pipette tip. The scratched surface was gently rinsed with PBS. After imaging, 4% FBS-reduced serum medium was added and cultured for another 24 h (25). The supernatant was discarded, the cells were gently rinsed with PBS and images were obtained again. All images were captured under a fluorescence microscope (Olympus IX73; Olympus Corporation). The width of the scratch was measured and analyzed using ImageJ software (version 2.14.0).

Using the Comet Electrophoresis Kit (cat. no. C2041S; Beyotime Institute of Biotechnology), approximately 1×10⁴ to 5×10⁴ cells in the logarithmic growth phase were mixed with 0.7% low-melting-point agarose and dropped onto pretreated slides. The mixture was incubated at 4°C for 10 min. After the agarose solidified, the slides were placed in lysis buffer at 4°C in the dark for at least 1 h. The slides were then denatured in alkaline electrophoresis buffer (300 mM NaOH, 1 mM EDTA, pH 13) for 60 min. Subsequently, electrophoresis was performed at 25 V for 30 min in the dark at 4°C. Following electrophoresis, the cells were neutralized, stained with PI at 2 µg/ml for 10 min in the dark at room temperature, and observed under a fluorescence microscope (Olympus IX73; Olympus Corporation) and images were obtained.

Using an immunofluorescence kit (cat. no. C2037S; Beyotime Institute of Biotechnology), cells in the logarithmically growing phase were seeded onto slides, removed from the slides, gently rinsed twice with PBS, fixed with 4% paraformaldehyde for 15 min at room temperature, permeabilized with 0.1% Triton X-100 (cat. no. NG1031A; Beyotime Insitute of Biotechnology) for 10 min, blocked with 5% BSA (cat. no. A8806; Sigma-Aldrich; Merck KGaA) for 1 h at 37°C and incubated overnight at 4°C with a dropwise diluted (1:500) primary antibody (γH2AX; cat. no. 9718S; Cell Signaling Technology, Inc.). The cells were rinsed twice with PBS, and diluted (1:100) fluorescently labeled (Alexa Fluor 488) donkey anti-rabbit IgG secondary antibodies (cat. no. ab150061; Abcam) were added for 1 h at room temperature. Cell nuclei were counterstained with DAPI for 10 min at room temperature, then mounted and observed and photographed under a fluorescence microscope.

Cells in the logarithmically growing phase were centrifuged at 300 × g for 5 min at room temperature, the supernatant discarded, and the pellet added to lysis buffer (cat. no. P0013B; Beyotime Insitute of Biotechnology). Total cellular protein was extracted and quantified using the BCA assay (cat. no. P0010; Beyotime Biotechnology). A total of 20 µg extracted proteins were subjected to 10% SDS-PAGE electrophoresis (60 V for 30 min and then transferred to 120 V for 60 min). After cutting the gel, the membrane was transferred to a PVDF membrane (300 mA for 100 min) and blocked with 5% skim milk for 2 h at room temperature. The membrane was then incubated with the primary antibody at 4°C overnight, washed three times with TBST (including 0.1% Tween-20), and incubated with the diluted (1:1,000) secondary antibody (anti-rabbit IgG, HRP-linked antibody; cat. no. 7074; and anti-mouse IgG, HRP-linked antibody; cat. no. 7076; both from Cell Signaling Technology, Inc.) at room temperature for 2 h. After washing the membrane three times with TBST, the membrane was developed and photographed using ECL luminescent solution (NCM Biotech; Suzhou Xinsaimei Biotechnology Co., Ltd.), and the grayscale value of the bands was analyzed using ImageJ software (version 2.14.0). The primary antibodies used were as follows: Poly(ADP-ribose) polymerase 1 (PARP-1; cat. no. 9532T; Cell Signaling Technology, Inc.; diluted 1:1,000), X-ray repair cross-complementing protein 1 (XRCC1; cat. no. ET1704-01; Huabio; diluted 1:1,000), phosphorylated tumor protein p53 (p-p53; cat. no. HA722761; Huabio; diluted 1:2,000), RRM2 (cat. no. ET1705-62; Huabio; diluted 1:1,000), RAD51 recombinase (RAD51; cat. no. ET1705-96; Huabio; diluted 1:5,000), phosphorylated histone H2AX (γH2AX; cat. no. 9718S; Cell Signaling Technology, Inc.; diluted 1:1,000), GAPDH (cat. no. 2118S; Cell Signaling Technology, Inc.; diluted 1:1,000).

Each experiment was performed in triplicate or more. Data are presented as the mean ± standard deviation (SD). Statistical analyses and figure creation were conducted using GraphPad Prism 10 (Dotmatics). Comparisons between two groups (Figs. 5 and 6) were conducted using Student's t-test. One-way ANOVA was used for comparisons among mutliple groups (Fig. 4C and H). When a significant main effect (P<0.05) was found, pairwise comparisons between groups were further assessed using Tukey's post-hoc test. A P<0.05 was considered to indicate a statistically significant difference. The following significance levels were applied: P<0.05, P<0.01, P<0.001.

The scRNA-seq data of patients with GC were subjected to quality-control analysis and highly variable genes were selected (Fig. S1A-C). Cell populations were compared between tumor and normal tissues, and each cell cluster was annotated (Fig. 2A-C). The inferCNV algorithm was used to analyze CNVs across different cell populations. The results revealed that D-epithelial cells and general epithelial cell clusters exhibited a significantly high number of CNVs (Fig. 2D and E).

These two epithelial cell populations were subdivided into six clusters for further analysis. Among them, clusters 0 (C0) and 1 (C1) accounted for the largest proportions and were present only in tumor tissues (Fig. 2F and G). C0 (including genes such as KRT19, GABRP, GPRC5A, TMC5, CYP3A5, CD55, MYH14, S100P, KRT8 and ANXA2) exhibited an epithelial cell subtype with secretory/metabolic function in tumors. C1 (including genes such as TOP2A, MKI67, BIRC5, PBK, RRM2, CDC20, ASPM, CEP55, TPX2 and KIF20B) contained highly proliferating epithelial cells. To identify potential clinical molecular targets, analysis focused on the more abundant tumor epithelial cell subpopulation. Pseudotime trajectory analysis revealed that C1 cells were more primitive than C0 cells (Fig. 2H). Gene set enrichment analysis (GSEA) indicated that C1 cells were primarily enriched in DNA repair and MYC target-related pathways (Fig. 2I). Intercellular communication analysis revealed that both C0 and C1 cells strongly interacted with other epithelial cell populations (Fig. 2J). Except for C0 and C1, C2 represents a differentiated cell group with some characteristics of stem/progenitor cells from epithelial origin (including genes such as PGC, ALDH1A1, FAM3B, CXCL17, OLFM4, MT1M, REG1A, TCEAL9, CDKN2A and EID1). The C3 group corresponds to activated/effector T cells associated with epithelial cells (including genes such as CD2, FYB1, PTPRC, SAMSN1, WIPF1, SRGN, LCP1, CD52, LSP1 and SLA), while C4 comprises neuroendocrine-like epithelial cells (including genes such as RIMBP2, ST18, SCG5, HEPACAM2, SCG3, INSM1, RFX6, ARX, FGF14 and PCSK1). Finally, the C5 group represents an inflammatory/stress-responsive epithelial subpopulation (including genes such as NPPB, TNF, CYP2E1, CXCL5, HLA-DQA2, LY6K, CDKN2A, MYL9, TAGLN, PNMA5) (Table SI).

As shown in Fig. 2K, C1 cells were involved in extensive ligand-receptor interactions with other clusters. In addition, pathway enrichment analysis revealed activated EGFR, WNT, and BMP signaling in C1 (Figs. 2L and S1D).

GO and KEGG analyses revealed that C1 cells were predominantly enriched in the ‘chromosomal region’, ‘ribonucleoprotein complex biogenesis’, and ‘ATP hydrolysis activity’ and were closely associated with the ‘cell cycle’ and ‘DNA replication’-related processes (Fig. 3A and B). Furthermore, WGCNA was performed on C1 cells, selecting a soft-threshold power of 8, and divided C1 into 28 subgroups (Fig. 3C and D). The associations among these subgroups are presented in Fig. 3E and F. Among the 28 modules identified by WGCNA, their associatons with the C0-C5 cell subpopulations (Fig. 3E) were systematically evaluated. In the association study between module 28 and subpopulation C1, four subclusters (C11, C127, C112, and C114) exhibited significantly higher average expression levels compared with other subclusters. Based on this observation, these four subclusters were selected for subsequent functional validation experiments (Fig. 3G). The key genes expressed in these subpopulations are shown in Fig. 3H.

GC-specific transcriptomic datasets GSE66229 and GSE84433 were obtained from GEO, and batch effects were removed. Differentially expressed genes (DEGs) associated with GC prognosis were identified based on survival analysis (Fig. S2A-D). Core genes from the four key malignant epithelial cell subclusters (C11, C112, C114, and C127) were intersected with the DEGs, resulting in the identification of two overlapping genes, RRM2 and PRNP. GO and KEGG enrichment analyses showed that RRM2 was strongly associated with ‘DNA replication’; therefore, it was selected as the primary candidate molecular target for GC. Pan-cancer analysis revealed that RRM2 expression was significantly higher in tumor tissues than in adjacent normal tissues in various cancer types (Fig. 4A). In both GSE66229 and GSE84433 datasets, patients with GC were stratified into high- and low-RRM2-expression groups. Expression trend analysis showed that RRM2 expression gradually decreased with disease progression (Fig. 4B, C, G and H).

RRM2 expression was positively associated with the abundances of activated memory CD4⁺ T cells, M0 and M1 macrophages, and resting NK cells (Fig. 4D and I). Using the TIMER database, it was found that high RRM2 expression in GC was associated with increased infiltration of CD8⁺ T cells, CD8⁺ naive T cells, cytotoxic lymphocytes (CTLs) and cytotoxic cells (Fig. S1E). The TIDE scores between two GEO GC datasets (GSE66229 and GSE84433) were compared. A high TIDE score indicates an increased risk of immune evasion by the tumor and is assoicated with a poorer prognosis for patients. Both datasets showed that the TIDE scores in the high-RRM2-expression group were lower than those in the low-RRM2-expression group, with statistically significant differences observed in GSE84433 dataset. These findings suggest that patients with GC with higher RRM2 expression may exhibit reduced immune evasion and potentially an improved response to immune checkpoint blockade therapy (Fig. S2E). The associations between RRM2 expression and prognosis across three datasets (GSE66229, GSE84433 and TCGA) are shown in Fig. 4E, J, and O. Notably, statistically significant differences are shown in Fig. 4E and O. In Fig. 4J, although P=0.088, a consistent trend is observed in which higher RRM2 expression is associated with longer survival time. It is hypothesized that if the GSE84433 dataset in Fig. 4J were expanded with additional data, it would likely achieve a P<0.05. KEGG analysis showed that RRM2 was enriched in ‘DNA replication’ and ‘cell cycle’ pathways (Fig. 4F and K). To validate these findings, patients with GC in the TCGA-STAD dataset were divided into RRM2-high (G1) and RRM2-low (G2) groups (Fig. 4L). G1 exhibited higher abundances of activated memory CD4⁺ T cells, M0 and M1 macrophages and resting NK cells, as well as significantly higher expression levels of immune checkpoint genes such as LAG3, PDCD1, and CTLA4 (Fig. 4M and N). The results of GO and KEGG enrichment analyses in the TCGA-STAD dataset were consistent with those observed in the GSE66229 and GSE84433 datasets (Fig. S2G-I).

To assess the role of RRM2 in GC progression, it was knocked down in two GC cell lines, AGS and HGC-27. Colony formation assay showed that RRM2 knockdown significantly suppressed colony formation in both cell lines (Fig. 5A and C), suggesting that RRM2 contributes to GC cell proliferation. In addition, RRM2 knockdown led to an increase in the number of apoptotic cells (Fig. 5B and C).

Transwell and wound healing assays showed that RRM2 knockdown reduced the number of GC cells migrating through the Transwell membrane and impaired the wound closure ability of the cells. These results suggest that RRM2 regulates the migratory abilities of GC cells (Fig. 5D-F).

Analysis of GC datasets from the GEO and TCGA databases indicated a strong association between RRM2 expression and DNA replication in GC. To validate this association, a comet assay was performed using GC cells transfected with control siRNA (CTRL group) or RRM2-targeting siRNA (siRRM2 group). The siRRM2 group exhibited pronounced comet tails and an increased tail DNA content, suggesting a significantly high proportion of DNA strand breaks (Fig. 6A). Consistently, γH2AX expression was markedly higher in the siRRM2 group than in the CTRL group, indicating substantial DNA damage upon RRM2 knockdown (Fig. 6B).

To examine the underlying mechanisms, the expression levels of proteins associated with DNA damage and repair were assessed. Western blotting revealed that RRM2 knockdown increased the protein expression levels of the DDR markers γH2AX and p-p53 but significantly downregulated the HR repair marker RAD51. Conversely, the expression levels of PARP-1, a key protein in single-strand break (SSB) repair, and XRCC1, a key protein in BER, remained unchanged. These findings suggest that RRM2 depletion induces DNA damage and is associated with HR repair in GC cells (Fig. 6C and D).