Single-cell sequencing data of paired gastric cancer tissue and normal stomach tissue were obtained from the GEO database (https://www.ncbi.nlm.nih.gov/geo; ID: GSE184198), including data from 1 gastric cancer sample and 1 normal stomach sample (18). To externally validate the results of the single-cell analysis, we obtained GSE268238 from the GEO database. The retrieved expression matrix was used to create a Seurat object, which was then matched with the acquired metadata. The data uploader had cleaned and filtered the data, so there is no need for further data cleaning (filtering thresholds: nCount_RNA > 1000 & nFeature_RNA < 5000 & percent.mt < 30 & nFeature_RNA > 600). The CellMarker database (http://xteam.xbio.top/CellMarker/) was used for manual annotation of cells (19). The harmony R package was used to remove batch effects between samples.

The subset function was used to extract a subset Seurat object from the Seurat object. The NormalizeData function was used to normalize the Seurat object. The FindVariableFeatures function was used to calculate highly variable genes. All cells were divided into different Seurat clusters at a resolution of 0.4. The RunUMAP function was applied to perform dimensionality reduction on the Seurat object. The genes related to oxidative phosphorylation were obtained from MsigDB (https://www.gsea-msigdb.org/gsea/msigdb), and the AddModuleScore function was utilized to evaluate the oxidative phosphorylation score of each cell. The R package ‘Seurat’ was used to perform the above analysis (20).

We conducted an analysis using the STAD data from the TCGA database (https://portal.gdc.cancer.gov/) and gastric cancer data (GSE62254) from GEO database (https://www.ncbi.nlm.nih.gov/geo/). After data cleaning and filtering, we obtained an expression matrix (TCGA) composed of 435 gastric cancer patients and their corresponding clinical data, and expression matrix (GEO) composed of 300 gastric cancer patients and their corresponding clinical data. Patients were divided into groups with high and low expression based on the median value of gene expression. The R package ‘survival’ was used to perform the survival analysis. The Kaplan-Meier (KM) analysis was utilized to determine survival differences, and a P-value of less than 0.05 was considered statistically significant.

We used high-OXPHOS-macrophage-related genes screened by single-cell sequencing to construct a prognosis-related signature. Therefore, the final selection of genes used to construct the prognosis-related signature was associated with macrophages and high OXPHOS. The R package ‘survival’ was utilized for univariate Cox analysis and multivariate Cox analysis. KM analysis was employed to determine survival differences, with a P-value less than 0.05 considered statistically significant. After univariate and multivariate Cox analyses, the selected genes comprised the prognosis-related signature. The following formula was used to calculate the risk score:

The expression level of each RNA was denoted as exp (RNAi), and the multivariate Cox regression coefficient for each RNA was denoted as coef (RNAi). The R package ‘survivalROC’ was used to assess the accuracy of prognosis -related signature. The R package ‘ggplot2’ was used for data visualization.

Based on the expression level of NPC2, LY96 and TPP1, we classified all macrophages into gene-positive macrophages and gene-negative macrophages. A Seurat object was used to construct a CellChat object. After loading the human receptor-ligand pair database into the CellChat object, an intercellular interaction network was established based on the existing receptor-ligand pair information. The R package CellChat was used to carry out cell communication analysis, and the analysis was carried out according to the standard protocol document (21).

We acquired the tissue microarray from Origin Biotechnology Inc., which contained 80 gastric cancer tissues and 80 normal stomach tissues. We followed the standardized procedure provided by the company to perform immunofluorescence staining on the tissue microarrays. We used anti-NPC2 (Invitrogen, CAT#PA5-143858, 1:500), anti-LY96 (Invitrogen, CAT#MA5-15766, 1:200), anti-TPP1 (Invitrogen, CAT#PA5-22274, 1:200) and anti-CD68 (Invitrogen, CAT#14-0688-82, 1:500) as primary antibodies, with goat anti-mouse-488 (abcam, CAT#ab150113, 1:1000) and goat anti-rabbit-594 (abcam, CAT#ab150080, 1:1000) as secondary antibodies. After the antibody incubation was completed, we stained the cell nuclei with DAPI. Once all staining procedures were finished, we processed the tissue microarrays with a confocal microscope (ZEISS, LSM 980).

THP1 cells were seeded into plate/dish (P3-P6, 1×10⁴ cells/ml) and cultured in RPMI 1640 medium (Servicebio, CAT #G4531) supplemented with 10% fetal bovine serum (FBS) (Servicebio, CAT #G8003), 0.05mM β-mercaptoethanol (MCE, CAT#HY-Y0326), and 1% penicillin/streptomycin (Servicebio, CAT#G4003). IACS-010759 (MCE, CAT#HY-112037) was used to inhibit OXPHOS in THP1 cells. The cells were treated with IACS-010759 (1.4 nM) for 48 hours and then fixed with paraformaldehyde. After fixation, the cells underwent immunofluorescence staining, following the same procedure as the aforementioned tissue immunofluorescence staining.

The R software (version 4.2.2) was used for all analyses. Prism 9 was used for data analysis. Kolmogorov-Smirnov test was used to confirm normality of data. Student’s t-test was used to confirm statistical significance if data were normally distributed. Mann-Whitney U test was used to confirm statistical significance if data were not normally distributed. P< 0.05 was considered statistically significant.

A total of 17,985 cells were classified into 8 cell types, including non-immune cells such as epithelial cells, endothelial cells, and fibroblasts, as well as immune cells such as T or NK cells, mast cells, B cells, plasma cells, and macrophages ( Figure 1A ). Epithelial cells originating from tumor tissue are considered to be malignant tumor cells, while those from normal tissue are normal functioning gastric epithelial cells. Similarly, macrophages from tumor tissue are considered to be tumor-associated macrophages, while those from normal tissue are deemed to have normal functions. After removing the batch effect between samples, the cells from tumor tissues and normal tissues exhibit a uniform distribution ( Figure 1B ). The proportions of the 8 different cell types also show significant differences between tumor and normal tissues ( Figure 1C ). The proportions of T or NK cells are nearly the same in both types of tissues, while the proportion of macrophages is significantly higher in tumor tissue compared to normal tissue. Therefore, it can be acknowledged to some extent that macrophages play a key role in the initiation and progression of tumors. Figure 1D displays the marker genes for each cell type.

A total of 956 macrophages were divided into 6 subgroups ( Figure 2A ). Among these, the proportion of TAMs originating from tumor tissues was significantly higher than that of macrophages derived from normal tissues ( Figure 2B ). Studies have indicated that TAMs are primarily M2-like macrophages, which have tumor-promoting functions (22–24). Furthermore, the metabolic characteristics of TAMs are marked by active oxidative phosphorylation that produces a large number of ligands that promote tumor invasion and growth (25, 26). Therefore, we calculated oxidative phosphorylation scores based on the expression levels of genes related to macrophage oxidative phosphorylation ( Supplementary Tables S1 and S2 ). We found that C3 macrophages had the highest oxidative phosphorylation scores ( Figure 2C ), suggesting that C3 macrophages may possess the capability to promote tumor invasion and growth. Subsequently, we extracted the signature genes of C3 macrophages and conducted Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses. In the KEGG analysis, signature genes were significantly enriched in pathways related to the lysosome ( Figure 2D ). Research has confirmed that TAM lysosomes can release enzymes that dissolve the extracellular matrix, thereby enhancing the invasive ability of cancer cells (27, 28). This further substantiates the role of C3 macrophages in tumor promotion. In the GO analysis, the signature genes were also significantly enriched in lysosome-related functions ( Figure 2E ), further supporting the important role of C3 macrophages, consistent with the results above. C3 macrophages may affect the TME of GC.

C3 macrophages had the highest OXPHOS score, thus marker genes of C3 macrophages would be analyzed as high-OXPHOS-macrophage related genes in the next step, with a total of 510 C3 macrophage marker genes. A total of 510 high-OXPHOS-macrophage related genes were screened out for further analysis. Univariate Cox regression analysis was used to filter the genes related to prognosis of gastric cancer, obtaining 50 prognosis-related genes ( Table 1 ). Then, to identify more prognostically valuable genes, multivariate Cox regression analysis was conducted on the 50 prognostic-related genes, eventually yielding 19 prognostic genes ( Table 2 ). Based on these 19 prognostic genes, a risk score was calculated for each patient. Patients were divided into high-risk and low-risk groups according to the median risk score. Figure 3A displays the distribution range of the risk scores. Patients with higher risk scores also had a higher risk of death ( Figure 3B ). Patients in the low-risk group had a higher survival rate compared to those in the high-risk group ( Figure 3C ). Moreover, compared to other clinical characteristics, the risk score achieved a higher AUC ( Figure 3D ), indicating that the high-OXPHOS-macrophage related prognostic signature has better predictive capability for prognosis. Similarly, the risk score demonstrated significant prognostic prediction abilities in both univariate and multivariate Cox regression analyses, further indicating that the risk score is an independent risk factor for gastric cancer ( Figures 3E, F ). Finally, we validated the risk model using an external gastric cancer dataset, further demonstrating the prognostic value of the risk model ( Figures 3G, H ). The high-OXPHOS-macrophage-related prognostic signature can effectively predict the prognosis of GC patients.

The 19 genes that make up the high-OXPHOS-macrophage related prognostic signature are all expressed in macrophages, however, NPC2, LY96, and TPP1 are specifically expressed in macrophages ( Figure 4A ). Among these, patients with high expression of NPC2, LY96, and TPP1 have a poorer prognosis ( Figures 4B–E ). In macrophages, NPC2, LY96, and TPP1 are mainly expressed in type C3 macrophages, and the expression level of NPC2 is higher than that of LY96 and TPP1 ( Figure 4F ). Moreover, NPC2, LY96, and TPP1 are primarily highly expressed in TAMs ( Figure 4G ), and the expression level of NPC2 is significantly higher than LY96 and TPP1 ( Figure 4H ). This demonstrates that NPC2 plays an important role in TAMs. At the single-cell level, NPC2, LY96, and TPP1 are significantly elevated in TAMs. Data from TCGA were used to explore the expression differences of NPC2, LY96, and TPP1 at the overall tumor level. Ultimately, we found that NPC2, LY96, and TPP1 are highly expressed in tumor tissues ( Figures 4I–K ). NPC2, LY96, and TPP1 were most likely related to the function of TAM.

We classified macrophages into NPC2 positive macrophages and NPC2 negative macrophages based on the expression level of NPC2. When observing the overall communication strength of cells, cell communication in tumor tissues was significantly stronger than in normal tissues ( Figure 5A ). The communication strength between different cell types in tumor tissues was clearly stronger than that between cell types in normal tissues ( Figure 5B ). NPC2 positive macrophages secrete TNF and TGFB1 that act on tumor cells, thereby enhancing the proliferation of tumor cells ( Figure 5C ). Furthermore, research has confirmed that MIF acts on T cells, causing the suppression of T cell activity and thereby mediating tumor immune escape. Similarly, NPC2 positive macrophages secrete MIF acting on T cells, inhibiting T cell activity and mediating tumor immune escape ( Figure 5C ). The above results indicate that NPC2 positive macrophages can affect the tumor microenvironment, ultimately promoting tumor growth and invasion. Lastly, we confirmed the high expression of NPC2 in tumor tissue macrophages using immunofluorescence ( Figures 5D, E ). Through external validation using GSE268238, we confirmed that NPC2 was highly expressed in macrophages from tumor tissues ( Figure 5F ). After using the OXPHOS inhibitor, we found that the expression of NPC2 in macrophages was significantly reduced ( Figures 5G, H ). This further illustrates the close relationship between NPC2 and OXPHOS.

We classified macrophages into LY96 positive macrophages and LY96 negative macrophages based on the expression level of LY96. The overall cell communication strength is significantly stronger in tumor tissues compared to normal tissues ( Figure 6A ). The communication strength among different cell types within tumor tissues is significantly stronger than that within normal tissues ( Figure 6B ). Similar to before, LY96 positive macrophages secrete TNF and TGFB1 that act on tumor cells, thereby enhancing tumor cell proliferation ( Figure 6C ). Like NPC2, LY96 positive macrophages secrete MIF affecting T cells, inhibiting T cell activity, and mediating tumor immune escape ( Figure 6C ). These results demonstrate that LY96 positive macrophages can influence the tumor microenvironment, ultimately promoting tumor growth and invasion. Finally, we confirmed the high expression of LY96 in tumor tissue macrophages using immunofluorescence ( Figures 6D, E ). Through external validation using GSE268238, we confirmed that LY96 was highly expressed in macrophages from tumor tissues ( Figure 6F ). After using the OXPHOS inhibitor, we found that the expression of LY96 in macrophages was significantly reduced ( Figures 6G, H ). This further illustrates the close relationship between LY96 and OXPHOS.

We classified macrophages into TPP1 positive macrophages and TPP1 negative macrophages based on the expression level of TPP1. The overall cell communication strength in tumor tissues was notably stronger than in normal tissues ( Figure 7A ). Communication strength among different cell types within tumor tissues was clearly stronger than that within normal tissues ( Figure 7B ). Similar to the previous findings, TPP1 positive macrophages secrete TNF and TGFB1 acting on tumor cells, thereby enhancing the proliferation of tumor cells ( Figure 7C ). In a resembling manner, TPP1 positive macrophages secrete MIF acting on T cells, thus inhibiting T cell activity and mediating tumor immune escape ( Figure 7C ). The aforementioned results suggest that TPP1 positive macrophages can affect the tumor microenvironment, leading to the promotion of tumor growth and invasion. Lastly, using immunofluorescence, we verified the high expression of TPP1 in the macrophages of tumor tissue ( Figures 7D, E ). Through external validation using GSE268238, we confirmed that TPP1 was highly expressed in macrophages from tumor tissues ( Figure 7F ). After using the OXPHOS inhibitor, we found that the expression of TPP1 in macrophages was significantly reduced ( Figures 7G, H ). This further illustrates the close relationship between TPP1 and OXPHOS.