Comprehensive analysis of the tumor immune microenvironment in gastric cancer and peritoneal metastasis based on single-cell RNA sequencing analysis

Abstract

The development and metastasis of gastric cancer are complex processes involving the TME (Tumor microenvironment) and the interactions between various cell types. Here, we investigated the molecular mechanisms and biological processes underlying gastric cancer and its metastasis using single cell RNA-sequencing (scRNA-seq) with the aim of identifying new targets and approaches for clinical treatment. R version 4.4.1 and the SeuratV5 package were used to process 20 scRNA-seq samples sourced from the GEO database. Highly variable genes were selected for GO (Gene Ontology), KEGG (Kyoto Encyclopedia of Genes and Genomes) and GSAV (Gene set variation analysis) enrichment analyses. CytoTRACE, CellChat, and Monocle3 analyses were conducted to investigate cell communication and pseudotemporal dynamics in the PM (Peritoneal Metastasis) and GC (Gastric Cancer) groups. Additionally, the prognostic significance of key genes was assessed by integrating data from the TCGA clinical database. A total of 2,626,594 peritoneal metastasis cells and 17,894 gastric cancer cells were identified, revealing 13 distinct cell clusters. Gene enrichment analyses identified high expression in several pathways (P53, Wnt and JAK-STAT3) in TAMs and mast cells. Cell communication was more robust in the GC group than the PM group, with TAMs (Tumor-associated Macrophages) and mast cells showing elevated expression of the CCL5-CCR1 ligand-receptor signaling axis in both groups. Pseudotemporal analysis demonstrated the differentiation potential of TAMs into mast cells, with APOC1, C1QB, FCN1, FTL, S100A9, CD1C, CD1E and FCER1A identified as the top eight genes driving this process. High expression levels of these genes, along with CCL5 and CCR1, were associated with poor long-term survival in cancer patients. scRNA-seq identified the intricate tumor immune microenvironment, highlighting the pivotal roles of TAMs and mast cells in gastric cancer peritoneal metastasis. The CCL5-CCR1 pathway emerged as a potential immune checkpoint, offering novel insights for future immunotherapeutic and targeted therapeutic strategies in the treatment of gastric cancer peritoneal metastasis.

Introduction

Gastric cancer (GC) poses a substantial global health challenge and represents the third leading cause of cancer-related mortality worldwide¹. Furthermore, the incidence of GC is disproportionately high in East Asian populations². According to the latest data published by the Chinese National Cancer Center³, the age-standardized incidence rate (ASIR) of gastric cancer in China is 13.72 per 100,000, and the age-standardized mortality rate (ASMR) is 9.39 per 100,000. The primary treatment for early GC is surgical resection⁴. Although radiotherapy and chemotherapy are standard treatments for over half of all patients eligible for GC surgery, the immune heterogeneity of tumors remains a significant obstacle for prognosis and recovery⁵. Clinically, almost 60% of all patients with GC experience peritoneal metastasis (PM), a common manifestation in advanced stages. PM represents a severe threat to the lives of patients with GC, with a 5-year overall survival (OS) rate of only 2%⁶. Even standard treatment approaches for recurrent or metastatic GC are ineffective in managing GC-PM, with patients receiving palliative chemotherapy often surviving for less than six months.

Therefore, exploring and analysing the physiological and pathological mechanisms underlying GC with PM, utilizing genomics and bioinformatics to investigate immune evasion pathways and tumor implantation migration at the microscopic level⁷, is of significant clinical benefit.

Single-cell RNA sequencing (scRNA-seq), particularly 10 × sequencing technology^8,9, has shown remarkable potential for uncovering the transcriptional heterogeneity and extensive reprogramming of the tumor microenvironment¹⁰, making comprehensive analysis of the human immune system more accessible. Many scholars have not only gained deeper insights into the pathophysiological microenvironment of gastric cancer¹¹ through single-cell studies, but also provided numerous novel perspectives for further clinical prediction and treatment¹². Thus, in-depth exploration of single-cell data from GC and PM samples could contribute to further understanding the similarities and differences in their immune mechanisms, as well as reveal specific pathways for treating PM.

In PM, immune evasion and tumor migration/implantation rely on the complex interactions and signalling pathways between various cellular subpopulations¹³. In this study, we utilized single-cell data to characterize and analysed the cellular subpopulations in samples of GC and PM. By performing bioinformatics analysis, we aimed to gain a comprehensive understanding of the roles of different cells in tumor progression and the impact of key genes on survival prognosis. We hope that our work will contribute to the clinical treatment of tumors and provide new insights for further mechanistic studies.

Analysis methods

Data source

The original 10 × RNA-seq data were obtained from GEO (Gene Expression Omnibus) (https://www.ncbi.nlm.nih.gov/geo/). A total of 20 samples were extracted from GSE163558, GSE228598 and GSE112302, including 10 gastric cancer samples and 10 peritoneal metastasis samples. The complete dataset can be found in the supplementary file named ‘rawdata’.

scRNA-seq data set quality control

After batching read 20 single cell samples through R, looping through the single cell data list scRNAlist, we used the ‘Merge’ function to integrate and read the data. The ‘lapply’ function was used to batch filter the single cell data list; the min.cells parameter was set to < 3%, the min. features gene number was set to 300, and the number of genes expressed per cell was set to > 300 and < 7000. For each cell, the UMI (Unique Multiplex Index) count content was > 1000, and the largest top 3% of cells were eliminated.

scRNA-seq data normalization processing and dimensionality reduction clustering

We used the ‘NormalizeData’ function to normalize the data, scale the dataset, and regress out mitochondrial gene effects. Then we used > ‘FindVariableFeatures’ function to identify highly variable genes and applied Principal Component Analysis (PCA) for dimensionality reduction. Next, we used the ‘Harmony’ package to correct for batch effects. Then, we applied t-distributed Stochastic Neighbor Embedding (t-SNE) and Uniform Manifold Approximation and Projection (UMAP) for unsupervised cell clustering for the visualization of cell subpopulations in a two-dimensional plot.

Cell type annotation

The ‘FindAllMarkers’ function was used to analyze differences between all cell clusters to identify marker genes. We applied the Wilcoxon Rank Sum test, with a logFC threshold set to 0.25, and the top 20 marker genes with an eligible logFC value in each cell cluster were screened and retained. The top 20 marker genes were manually annotated by cellmarker (http://xteam.xbio.top/CellMarker/) and Enrichr (https://maayanlab.cloud/Enrichr/) websites; and automatically annotated by the ‘singleR’ package. According to the confidence of singleR annotation and manual annotation results, the final annotation was performed by comparing the biological information background.

Differential gene analysis

The ‘FindAllMarkers’ function was then used to analyze the differentiation of all cell subsets, and the Wilcoxon non-parametric test was used with a log2FC differential multiplication threshold of 0.25 and a gene expression ratio threshold of 0.25. The top 100 genes were screened for enrichment analysis based on the magnitude of the avg_log2FC value. The ‘compareCluster’ function was then used to perform GO enrichment analysis for the top 100 genes, and the ‘ggupset’ package was used to draw the GSVA (Version 1.48) single-cell score results for each cell subset. Volcano plots were visualized using the ‘JJVolcano’ package in R for the five most significantly upregulated and downregulated genes of all cell types of differential genes.

Cellchat analysis

The ‘cellchat’ package in R was used to search for hyper-variable genes (ligand-receptor) and their mutual pathways in 10 samples from the GC group and 10 samples of the PM group. Hyper-variable genes were then projected to a protein–protein interaction (PPI). The ‘networkfilterCommunication’ function was used to filter the cell communication network with < 10 cells, summarize all relevant ligands/receptors, calculate the communication probability and aggregate network at the signaling pathway level, and visual analysis was performed according to the results.

CytoTRACE cell differentiation score

CytoTRACE was installed using the ‘Biomanager’ package in R. RNA information in annotated Seurat objects was extracted, genes expressed in < 5 cells were filtered out, analyzed using CytoTRACE, and the analysis results, cell type annotation, and UMAP embedding were used to draw visual graphs. The mean gene expression values in the cell annotation results were obtained, converted to Matrix, and the standard deviation for each gene row was calculated. The ‘msigdbr’ package in R was installed (https://www.gsea-msigdb.org/gsea/msigdb/human/collections.jsp#H), the gene sets were converted, and the ‘Amex GSVA’ package was used for standardized enrichment analysis and visualization.

Pseudotime analysis of cell progression

Pseudotime analysis is a method used to simulate time. Monocle¹⁴ utilizes advanced machine learning techniques to learn explicit master graphs from single-cell data to rank cells, which can robustly and accurately resolve complex biological processes. The pseudotime analysis of TAMs, cDC1s and mast cells by monocle3 was combined with Seurat to evaluate gene changes. The R package ‘monocle3.22’, based on R 4.4.1, was downloaded from the functional net (https://cole-trapnell-lab.github.io/monocle3).

Gene survival analysis

GEPIA2 (http://gepia2.cancer-pku.cn/) integrates data from the Cancer Genome Atlas (TCGA) and GEO database, including information relating to various disease subtypes. Survival analysis of highly variable genes was performed using the GEPIA2 online analysis platform.

Results

Cell annotation

After data filtering, quality control, dimensionality reduction, clustering, and batch effect removal, we obtained 26,594 PM cells and 17,894 GC cells. Figure 1A shows dimensionality reduction clustering for 20 samples prior to Harmony integration. The group annotation (GC and PM) is presented in Fig. 1B. Figure 1C, D respectively show the integrated sample distribution and group annotations following Harmony batch correction. The dimensionality reduction results for each individual sample post-Harmony integration are shown in Fig. 1E. During annotation and based on a cell cluster tree resolution of 0.2 (Fig. 2A), we defined 13 cell clusters, numbered from 0 to 12. Comparative dimensionality reduction (UMAP: Fig. 2B; t-SNE: Fig. 2C) revealed GC/PM group distributions (Fig. 2D). Initial automatic annotation was performed using SingleR, and according to the confidence scores in Fig. 2E and UMAP clustering (Clustering via UMAP demonstrates enhanced topological preservation and structural fidelity), we observed that SingleR annotated cluster 10 represented NK cells (Fig. 2F). However, based on spatial visualization from dimensionality reduction clustering and further annotation with marker genes, we found that the annotation of cluster 10 as NK cells lacked confidence. Further in-depth annotation was performed for clusters 0 and 1, with more detailed annotations.

Fig. 1.

(A) Distribution of 20 independent samples before Harmony integration; (B) Sample distribution of GC group and PM group before Harmony integration; (C) Distribution of 20 independent samples after harmony integration; (D) Sample distribution of GC group and PM group after Harmony integration; (E) Distribution of independent samples after dimensionality reduction and clustering).

Fig. 2.

(A) Resolution cell cluster tree; (B) UMAP dimensionality reduction clustering plot of 13 cell clusters; (C) t-SNE dimensionality reduction clustering plot of 13 cell clusters; (D) UAMP distribution of cell clusters in the two groups; (E) Confidence score of SingleR annotation; (F) Final annotation results according to SingleR; (G) Cell annotation marker genes; (H) The violin plot of Top20 gene expression in each cell cluster).

We prepared a heatmap for the top 20 marker genes (Fig. 2G) and a violin plot (Fig. 2H) of each cell cluster. Combining with biological significance from cell annotation websites, we obtained the final annotation results. The final annotation identified 13 cell subpopulations: TAMs, T cells, neutrophils, conventional dendritic cells (cDC1s), B cells, gastric mucosal cells, proliferating cells, fibroblasts, plasma cells, mast cells, endothelial cells, and plasmacytoid dendritic cells (pDCs) (Fig. 3A).

Fig. 3.

(A) Final annotation result dimensionality reduction clustering diagram; (B) Volcano plot of Top5 hypervariable genes; (C) Enrichment results of GO; (D) GO enrichment results for specific signaling pathways; (E) Results of enrichment of the GSVA; (F) Comparison of GSVA enrichment intensities).

Highly variable gene analysis

Volcano plot visualization of differential gene expression for each cell subpopulation revealed the top five upregulated and downregulated genes (Fig. 3B). In T cells, the most upregulated genes were FGFBP2, ACO22126.1, KLRF1, KLRC3, and KIR2DL4, while the most downregulated were RAB3IL1, ASGR2, CSTA, TLR8, and LYPD2. In B cells, the top five upregulated genes were SEC14L3, IL17A, H2AFB1, KRT73-AS1, and IL2, while the most downregulated genes were VSIG4, CLEC5A, FCN1, CSF1R, and MS4A4A. In TAMs, FOLR3, ACOX2, FCN1, APOBEC3A, and VSTM1 were the most upregulated, while LIPF, TPSAB1, ACO44849.1, TPSB2, and CPA3 were the most downregulated. In mast cells, the top five upregulated genes were CD207, CLEC4F, CD1E, FCER1A, and ASCL4, while the most downregulated were LIPF, IGHG3, TPSB2, CPA3, and TPSAB1. These results suggest that T cells exhibit dual characteristics: while upregulating genes associated with cytotoxicity and growth factors (such as FGFBP2, KLRF1) to maintain immune activity, they are also subjected to immunosuppressive effects from the tumor microenvironment (e.g., upregulation of KIR2DL4 and downregulation of TLR8), thus reflecting dynamic and complex immune regulation during tumor progression. The role of B cells in the tumor environment may involve a balance between pro-inflammatory and anti-inflammatory functions. The upregulation of IL17A and IL2 may reflect B cell immune activation, while the downregulation of CLEC5A and CSF1R suggests that a tumor might inhibit its interactions with other immune cells. TAMs may enhance tumor growth through metabolic activity (e.g., FOLR3, ACOX2) and inflammation via complement and DNA modification (e.g., FCN1, APOBEC3A), while the downregulation of TPSAB1 and CPA3 suggests a reduction in their ability to break down the tumor matrix.

Gene enrichment results

GO analysis

GO enrichment results (Fig. 3C) for the top 100 genes associated with each cell subpopulation revealed diseases and signaling pathways linked to these genes. For TAMs, the diseases involved complement and coagulation cascades, malaria, ‘Legionnaires’ disease and tuberculosis. Further pathway-specific GO analysis (Fig. 3D) identified highly significant pathways, including the Wnt signaling pathway, methylation pathways, CCR chemokine receptor binding, the CXCL12-activated CXCR4 signaling pathway, regulation of immune response and the MAPK cascade. T cells, B cells, and plasma cells showed higher enrichment in pathways such as interleukin-10 production, the positive regulation of NF-kappaB transcription factor activity, myeloid leukocyte activation, and lymphocyte activation compared to TAMs.

GSVA analysis

As shown in Fig. 3E, TAMs had high GSVA scores in signaling pathways related to IL-10, IL-6, TGF-β, IL-2, PI3K-AKT-MTOR, JAK-STAT3, Wnt, and P53, with the highest score in the TGF-β pathway. Mast cells, cDC1s and gastric mucosal cells showed high enrichment in anti-inflammatory pathways, consistent with TAMs. T cells, B cells, and plasma cells had lower enrichment intensity in these pathways (Fig. 3F).Combined with specific GO gene enrichment results (Fig. 3D), TAMs may promote immune suppression by secreting IL-10 and TGF-β, inhibiting effector T cells and activating Tregs, facilitating tumor cell transformation and epithelial-mesenchymal transition (EMT), and enhancing the immunosuppressive state of immune cells, leading to stronger tumor invasiveness and metastatic potential.

Cell communication

Signal pathway interactions were abundant between the GC and PM groups (Fig. 4A). Overall, the GC group showed higher interaction quantity and intensity than the PM group. In both groups, the major interactions between cell subpopulations involved lymphocytes and myeloid cells. The most intense interactions occurred between TAMs, mast cells, T cells, B cells, and cDC1s. The pattern of interaction (Fig. 4B) between mast cells, TAMs, and cDC1s was primarily Pattern 1, involving pathways such as CXCL, MK, IL-1, and NOTCH. T cells and B cells primarily followed Pattern 2, involving pathways such as CCL, TNF, and IFN-II. In the GC group, the most intense signaling flows were found in pathways such as VISFATN, TGF-β, JAM, EGF, and VEGF, with VISFATN showing the highest signaling flow. The PM group featured specific pathways such as CHEMERM and LT, but with lower signaling flow intensity. Both groups showed high signaling flow in FN1, MHC-I, MHC-II, CD45 and MF pathways, indicating active immune escape, tumor cell migration and proliferation during tumor progression.

Fig. 4.

(A) The intensity and quantity of cell communication interactions between the GC group and PM group; (B) Cell communication patterns of GC group and PM Group; (C) Lymphocyte interaction pathways with myeloid cells; (D) Outgoing communication pattern of secreting cells; (E) Display of the interaction intensity between all signal initiators and receivers).

It is noteworthy that the proliferating cells in the PM group exhibited significantly more active interactions with gastric mucosal cells and lymphocytes compared to the GC group (Fig. 4A). The numbers and intensities of outgoing communication patterns of secreting cells in GC group were more robust than in the PM group (Fig. 4B). In contrast, lymphocytes and other myeloid cells in the GC group had more active interactions than those in the PM group (Fig. 4C), suggesting that tumor cells in peritoneal metastasis may exhibit higher proliferation and metabolic activity. Combining the results from Fig. 4B, we conclude that the PM group influences surrounding cells, including gastric mucosal and lymphocytes, through the release of growth factors (e.g., EGF, TGF-β), chemokines (e.g., CXCL1, CXCL8), and extracellular matrix (ECM) remodeling signals (e.g., FN1, MMPs); these findings were also supported by GO enrichment analysis (Fig. 3).

The analysis of outgoing communication patterns of secreting cells (Fig. 4D) revealed that lymphocytes in the GC group featured more signaling pathways than in the PM group, especially in endothelial cells, mast cells, cDC1s, and TAMs. 2D spatial visualization showed that TAMs and T cells played similar roles as signaling transmitters and receivers in both groups. The strongest signal transmitters were gastric mucosal cells and cDC1s in both groups (Fig. 4E), but in the GC group, cDC1s were the strongest signal receivers, while in the PM group, T cells were the strongest receivers.

To further analyze the differences in immune signal transduction between the two groups, we focused on the CCL and CXCL chemokine family signaling pathways. The interactions between lymphocytes and myeloid cells in the CCL pathway were more complex in the GC group (Fig. 5A), with the highest number of interactions, while the PM group exhibited the strongest interaction intensity (Fig. 5B). The highest interaction intensity between cDC1s and T cells in both groups was followed by interactions with gastric mucosal cells and T cells. Mast cells and neutrophils exhibited higher interactions with other cells in the PM group for the CCL pathway, but minimal interactions in the GC group. 2D spatial visualization (Fig. 5C) also showed that the PM group had lower outgoing strength in the CCL pathway, but a higher receiving intensity. TAMs in both groups mainly acted as influencers, while cDC1s functioned as receivers, influencers and mediators. The analysis of CCL pathway roles in different cell clusters (Fig. 5D) showed that mast cells in the PM group exhibited a more prominent influencer role compared to a weaker influencer in the GC group. CCL5-CCR1 was the most specific CCL pathway in both GC and PM groups (Fig. 5E), with more complex interactions in the GC group, mainly involving TAMs, cDC1s, fibroblasts and gastric mucosal cells (Fig. 5F). However, there was no evidence to suggest that mast cells played a co-influential role. In contrast, mast cells interacted with TAMs and cDC1s in the PM group by signaling with myeloid cells, such as T and B cells, demonstrating a distinct role compared to the GC group.

Fig. 5.

(A) Results of CCL signal pairing interactions between lymphocytes and myeloid cells; (B) CCL pathway intensity between all cells; (C) Display of the interaction intensity of CXCL/CCL pathway between all signal pathway initiators and receivers; (D) The role positioning of each cell in the CCL pathway; (E) Contributions of each Ligand-Receptor pair of CCL pathway; (F) Display of CCL pathway Ligand-Receptor pairing with the highest specificity).

These findings suggest that in the peritoneal metastatic environment, the interactions between immune cells are intensified, especially in terms of immune regulation in the local tumor microenvironment. This could be a key factor in immune evasion and the promotion of metastasis. Tumor cells in peritoneal metastasis may more effectively utilize chemotactic signals from immune cells to migrate, colonize and grow. The CCL5-CCR1 pathway might be a potential immune target for further treatment in gastric cancer and peritoneal metastasis.

CytoTRACE analysis

Visualization analysis of the CytoTRACE Predicted order (Fig. 6A) shows cell types including mast cells, cDC1s, gastric mucosal cells, TAMs, monocytes, and fibroblasts in lighter hues, indicating relatively reduced differentiation potential. In contrast, lymphocytes such as plasma cells, B cells, T cells, neutrophils, and pDCs exhibited enhanced differentiation capacity. This observation aligns with immune evasion mechanisms and functional specialization of effector cells during tumor progression and metastasis-consistent with the substantial clonal expansion potential of lymphocytes.

Fig. 6.

(A) Cell differentiation capability score of cytoTRACE; (B) The pseudo-temporal differentiation starting points of cDC1 cells, mast cells, and tumor-associated macrophages; (C) The pseudo-temporal differentiation starting point of the CCR1;D.The pseudo-temporal differentiation starting point of the CCL5; (E) Pseudo-temporal trajectories of Top8 gene; (F) Spatiotemporal changes of top8 gene in Mast Cells, TAMs, cDC1s).

Pseudo-temporal analysis

Combined with the spatial distribution of dimensionality reduction clustering, cDC1s gradually differentiate into TAMs and mast cells under PM and GC conditions (Fig. 6B), while TAMs differentiated from cDC1s will differentiate into mast cells. Expression of ligand receptors of CCL5-CCR1 specific pathway was investigated in a simulated time sequence; analysis showed that CCR1 (Fig. 6C) was more abundant than CCL5 (Fig. 6D) and was balanced among cDC1s, TAMs and mast cells. These results suggest that the differentiation trajectory of cDC1s and the enhanced function of TAMs/mast cells jointly promote immunosuppression in the tumor microenvironment and help tumor cells evade immune surveillance. APOC1, C1QB, FCN1, FTL and S100A9 showed highly differentiated expression during the time sequence change (Fig. 6E). Of these, the expression levels of S100A9 and FCN1 (Fig. 6F) increased gradually during the differentiation of cDC1s into TAMs and decreased gradually during the differentiation into mast cells. APOC1, C1QB and FTL showed a gradual downwards trend during the entire differentiation process. However, the expression levels of CD1C, CD1E and FCER1A genes were gradually increased during the differentiation of mast cells, and their expression in mast cells was much higher than that in other cells. The high expression levels of S100A9 and FCN1 in TAMs suggest that they promote tumor growth by regulating inflammation and tumor-related immune escape, while mast cells may stimulate tumor-related angiogenesis by high expression molecules such as FCER1A and promote metastasis by secreting factors such as histamine and trypsin.

Key gene prognostic survival analysis

Based on the gene trajectory changes in the pseudo-temporal analysis results, the top eight genes with the most significant changes were integrated with the TCGA tumor clinical database in the GEPIA2 online analysis platform. Analysis indicated that within the first 200 months, the downregulation of CD1C (Fig. 7C), CD1E (Fig. 7D), and FCER1A (Fig. 7E) compared to their upregulation leads to a significant decrease in the clinical survival rate of patients. However, the upregulation and downregulation of APOC1 (Fig. 7A) and FCN1 (Fig. 7F) did not differ significantly in terms of tumor survival rates within 200 months. The downregulation of C1QB (Fig. 7B), S100A9 (Fig. 7H), and FTL (Fig. 7G), on the other hand, increased the survival rate of patients within 200 months.

Fig. 7.

Prognostic survival curves of Top8 gene in TCGA tumor clinical sample cohort (A) Overall survival analysis of gene APOC1; (B) Overall survival analysis of gene C1QB; (C) Overall survival analysis of gene CD1C; (D) Overall survival analysis of gene CD1E; (E) Overall survival analysis of gene FCER1A; (F) Overall survival analysis of gene FCN1; (G) Overall survival analysis of gene FTL; (H) Overall survival analysis of gene S100A9).

Conclusion

The exact mechanisms underlying GC and PM are not fully understood. The ‘seed and soil’ hypothesis is currently one of the most widely accepted theories of metastasis¹⁵. The occurrence of PM depends on the interaction between cancer cells (seeds) and the peritoneal microenvironment (soil)¹⁶. Tumor cells secrete transforming growth factor beta 1 (TGF-β1), which is involved in the generation and maturation of the extracellular matrix (ECM). The ECM provides binding sites for β1-integrin and CD44H, ultimately facilitating the invasion and adhesion of cancer cells^17,18. Specific structures in the peritoneum, such as lymphatic pores and lymphatic lacunae, make this structure conducive for the ‘seeding’ of cancer cells. Peritoneal lymphatic lacunae are unique immune structures surrounded by macrophages and lack a continuous mesothelial layer, making them prone to the colonization of free cancer cells¹⁹.

In addition, free cancer cells tend to accumulate at lymphatic pores that connect the lymphatic lacunae with sub-peritoneal lymphatic vessels. These pores are specifically distributed on the serosal surfaces of the diaphragm, mesentery, omentum, and pelvic peritoneum, making them susceptible to early involvement in PM^20,21. Therefore, neglect of the TME, the ‘soil’ formed by immune cells and tumor cells, has often been implicated as a major reason for suboptimal treatment outcomes²². TAMs represent the major immune cell population in the TME, predominantly exhibiting an M2 phenotype (promoting tumor growth)²³. TAMs participate in tumor progression by secreting cytokines, promoting angiogenesis, and suppressing immune responses. In GC, the degree of TAM infiltration is closely associated with poor prognosis. M1-type TAMs enhance tumor immune clearance by increasing TNF-α secretion and improving the function of CD8 + T cells, while M2-type TAMs inhibit T cell function by promoting lipid accumulation in the TME, resulting in the immune evasion of GC cells²⁴. Our single-cell sequencing of TAMs, based on highly variable gene expression and dimensionality reduction clustering, showed a mixed M1/M2 macrophage phenotype. The high enrichment of the PI3K/AKT signaling pathway in gene enrichment analysis further suggests that M2 TAM polarization is active, supported by high expression levels of chemokines such as CXCL5, CXCL12 and CCL5 in the TME²⁵.

Our research indicates that TAMs in the primary tumor secrete large amounts of immunosuppressive molecules (IL-10, TGF-β) and chemokines (CCL2, CXCL8)²⁶, recruiting more suppressive cells and inhibiting T cell activity through cell–cell contact. TAMs exhibit dual functions: (1) suppressing lymphocyte-mediated anti-tumor responses, and (2) promoting pro-inflammatory signaling via crosstalk with myeloid cells. This establishes a ‘chronic inflammation-tumor’ axis that facilitates tumor progression through enhanced proliferation, invasion, and angiogenesis. During the process of PM, tumor cell proliferation is closely linked to microenvironmental remodeling²⁷. Proliferating cells promote the ‘activation’ of peritoneal mesothelial cells and gastric mucosal cells, thus facilitating tumor cell adhesion, diffusion, and invasion. Gastric mucosal cells may be induced by tumor cell signals to transition into a cancer-associated fibroblast (CAF)-like or activated mesothelial cell state²⁸, further promoting ECM secretion and tumor cell migration²⁹. TAMs induce peritoneal mesothelial cells to transform into CAFs, forming a pre-metastatic niche. During this process, TAMs and proliferating cells in the peritoneal metastasis group may influence lymphocytes through the expression of PD-L1³⁰ or by secreting immunosuppressive factors such as IL-10 and TGF-β, leading to functional exhaustion or conversion to an immunosuppressive phenotype (such as Tregs)³¹. Elevated levels of pro-inflammatory factors, such as IL-6 and TNF-α, activate the STAT3 pathway, further promoting the implantation of cancer cells into the peritoneum³², a process consistent with the M2 polarization of TAMs mentioned earlier. We summarize these findings as follows. First, the TME in the primary tumor relies heavily on immune regulation mediated by TAMs: TAM-driven immune suppression in primary gastric cancer is likely a key factor in resisting host immune clearance and promoting early tumor progression. Second, the activity of TAMs may serve as preparatory work for the pre-metastatic niche: the interactions of TAMs with myeloid cells and lymphocytes may lay the groundwork for tumor cell subsequent metastasis and systemic spread.

Mast cells, another important immune cell type, have been controversial in their contribution to the TME in solid malignancies. In GC and PM, mast cells play a role in promoting disease progression, migration and reduced survival³³. Pseudotime analysis also revealed a biological differentiation relationship from cDC1s to TAMs and then to mast cells. Among the highly variable genes (Fig. 3C), CD1E and FCER1A in mast cells, along with FCN1 in TAMs, were the most upregulated and key genes in the pseudotime trajectory. FCN1 was also one of the top five most downregulated genes in B cells. The upregulation of these genes significantly improved patient survival.

During the progression of GC, mast cells promote the spread of tumor cells by releasing chemotactic ligand CCL2, vascular endothelial growth factor (VEGF)-A and TNF, which increase epithelial permeability and vasodilation³⁴. TAMs also use these factors and signaling pathways to promote tumor progression and metastasis. The CCL/CXCL pathway plays an extraordinary role in both mast cells and TAMs within the TME. Our research further confirms that the CCL5-CCR1 receptor-ligand pathway is the most specific in both GC and PM, with the CCL5-CCR1 signal activating NF-κB and STAT3 pathways, upregulating EMT transcription factors such as Snail and Twist, thereby enhancing cancer cell invasiveness. CCL5 recruits TAMs, Treg cells, and myeloid-derived suppressor cells (MDSCs) to the tumor site, forming an immunosuppressive microenvironment. TAMs then further secrete IL-10 and TGF-β³⁵, suppressing CD8 + T cell function. Moreover, CCR1 induces the expression of PD-L1 on gastric cancer cells, promoting T cell exhaustion. Following CCR1 activation, the PI3K/AKT and MAPK/ERK pathways promote GC cell proliferation and inhibit apoptosis (downregulating the Bax/Bcl-2 ratio). This provides a new targeted treatment strategy for GC and PM, such as using small molecule inhibitors: CCR1 antagonists (e.g., CCX354, BX471)³⁶, which can inhibit gastric cancer peritoneal metastasis in preclinical models. Additionally, monoclonal antibodies targeting CCL5 (e.g., Met-RANTES) could block the ligand-receptor binding, reduce immune cell infiltration, and, combined with CCR1 inhibitors and PD-1 antibodies, reverse the immunosuppressive microenvironment and target the CCL5-CCR1 pathway alongside chemotherapy to overcome resistance in peritoneal metastasis.

Author contributions

SUN WH collected the data and authored this paper.  DENG ZM provided guidance throughout the process. JIANG ZW offered financial support for publication and direction on the research scope.

Funding

This study was supported by the following funding sources: Natural Science Foundation of Nanjing University of Chinese Medicine (Project No. XZR2020017). National Natural Science Foundation of China (Grant No. 82074432). Key Discipline Construction Project of Jiangsu Provincial Health Commission: Traditional Chinese Surgery (Project Code: ZDXK202251). Open Research Fund of China National Clinical Research Base for Traditional Chinese Medicine (Jiangsu Province Hospital of Chinese Medicine; Project No. JD2023SZ11).

Data availability

Single-cell RNA-seq data from 10 gastric cancer samples (GSM3067368, GSM3067369, GSM3067370, GSM3067373,GSM3067374, GSM3067375, GSM5004180, GSM5004181, GSM5004182, GSM5004187, GSM5580154) and 10 peritoneal metastasis samples (GSM7133741-GSM7133749&GSM5004187) are available in the NCBI GEO(https://www.ncbi.nlm.nih.gov/geo/) database under super-series accession number GSE163558, GSE228598 and GSE112302.Related gene prognosis survival data is derived from the TCGA clinical database (https://www.cancer.gov/ccg/research/genome-sequencing/tcga). All of the raw data can be found in the related files. Raw data can be obtained by contacting the corresponding author directly.

Competing interests

The authors declare no competing interests.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-13892-6.