Macrophages in tertiary lymphoid structures promote apoptosis of ATF4-positive gastric cancer cells via IL18 Authors: Lu Zhou, Xuelong Li, Junming Wu, Tianqing Zheng, Manru Lin, Yangjian Ou, Hong Zhou, Steven Mo, Xueqiong Han, Jinyu Xiang, Wenjing Gong Published: 2025 ABSTRACT AbstractBackgroundGastric cancer (GC) is a globally prevalent and highly fatal malignancy of the digestive system. Despite its transformative role in gastric cancer, immunotherapy’s efficacy is limited by the complexity and incomplete understanding of the heterogeneous tumor microenvironment, necessitating a deeper analysis to uncover mechanisms of resistance.MethodWe analyzed the immune response status of GC patients and generated a single-cell transcriptome landscape of GC patients under various immunotherapy regimens. Dynamic changes in prognosis, biological functions, developmental trajectories, transcription factors, and intercellular communication among different cell subpopulations were assessed. To complement our computational exploration and provide histological validation, we also collected gastric cancer tissues from two patients who responded to immunotherapy. Hematoxylin and eosin staining and immunofluorescence experiments were performed on these tissues to confirm the pathological features and the spatial distribution of key immune cells identified in our single-cell analysis.ResultsOur study highlighted that PDIA6 in APOA1 + PDIA6 + GC activated the PPAR signaling pathway by increasing APOA1 expression through RNA-binding protein (RBP) action. Given the established role of the PPAR pathway in promoting lipid metabolism and cell proliferation, this activation likely provides a survival and growth advantage to this GC subpopulation. CYP3A5 + GC exhibited a unique metabolic pattern supporting tumor survival and resistance to immunotherapy. In clinical samples with effective immunotherapy, tertiary lymphoid structures showed expression of CD68 and IL18. IL18 + macrophages potentially influenced ATF4 + GC cells via the cytokine signaling axis, promoting immune infiltration. Apoptotic tumor cells were more enriched in the immune-responsive group compared to the immune-unresponsive subgroup. Furthermore, cancer-associated fibroblasts reprogrammed the tumor microenvironment by targeting endothelial cells with CTGF-ITGA5 and immune cells with CTGF-ITGB2. Within the XCL1 + NR4A2 + effector CD8 + T cell subpopulation, NR4A2 transcriptionally regulated the expression of XCL1 and effector genes through negative feedback loop, consequently inhibiting inflammatory pathways and the body’s antitumor immune response.ConclusionOur findings delineate the cellular composition of the tumor microenvironment in GC patients based on GSE183904, significantly advancing the analysis of distinct immune responses in these patients. This study contributes to a comprehensive understanding of immunotherapy efficacy and proposes novel biomarker-driven strategies for patient stratification and targeted therapy in GC.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12967-025-07559-z.Keywords: Gastric cancer, Single-cell RNA sequencing, Tumor microenvironment, Immunotherapy, Intercellular communication Introduction Gastric cancer (GC) is one of the most prevalent and highly fatal malignancies of the digestive system globally [1, 2]. It is characterized by high heterogeneity of tumor cells and the tumor immune microenvironment [3], which brings challenges and opportunities to cancer treatment [4]. Despite advancement in chemotherapy and surgical interventions, the prognosis for advanced GC remains poor, with limited traditional clinical treatment options [5, 6]. In the abovementioned situation, immunotherapy has become a promising approach for GC [5, 6]. However, limited immunotherapy options are available for gastric cancer, and only targeted drugs for human epidermal growth factor receptor 2 (HER-2) and programmed cell death ligand 1 (PD-L1) have been developed, though with limited success to date [7]. Clinical trials have confirmed the limited efficacy of various targeted drugs, whether used alone or in combination with other medications, in patients with advanced GC compared to other tumor types [7]. The limited efficacy of immunotherapy is primarily attributed to the complex immune suppression within the tumor microenvironment (TME) of GC [8]. Consequently, two of the current challenges for this therapy are the low response rates and the emergence of treatment resistance, a clinical reality that is strongly evidenced by a recent comprehensive review of the GC TME [9]. To address these issues, research has focused on combining immune checkpoint blockade antibodies with various drugs to enhance therapeutic efficacy. Understanding the cellular and molecular mechanisms of checkpoint blockade therapy is crucial for the development of combined immunotherapy strategies to overcome resistance mechanisms [9]. For patients with low T-cell levels in tumor tissue, direct injection of interferon-induced molecules or inhibition of immune-suppressive cells, such as regulatory T cells or macrophages, can be employed in addition to antibody combination therapy for immunotherapy [10]. In addition to immune-based approaches, natural product-based anticancer strategies have also attracted growing attention in recent years, such as Potentilla fulgens root extract with cytotoxic effects toward cancer cell lines [11], Elaeagnus angustifolia leaf extract demonstrating anticancer and antioxidant activities [12], and nanocarrier-based delivery systems such as a magnetic nanocomposite derived from okra leaves for targeted delivery of morin hydrate [13]. These studies highlight the expanding diversity of anticancer therapeutic research directions. To investigate the heterogeneity of tumor cells and immune cells in the immune response mechanism of GC patients, we generated comprehensive single-cell transcription landscapes from 36 gastric tissue samples based on GSE183903. Our study overcomes the limitations of conventional population-averaged analyses by employing single-cell RNA sequencing, enabling the precise identification of specific cellular subpopulations and molecular mechanisms that drive immunotherapy resistance in gastric cancer. Our study identify specific resistance mechanisms, including a glutamine-metabolizing CYP3A5 + GC cell subpopulation, immunosuppressive Cancer-Associated Fibroblasts(CAFs) acting through the CTGF axis, and a dysfunctional XCL1 + NR4A2 + CD8 + T-cell state. These findings provide immediately actionable insights: the molecules we identified, such as CYP3A5, RGS5, and NR4A2, serve as novel biomarkers for patient stratification and represent promising therapeutic targets for combination therapies designed to overcome resistance. Methods Human subjects All research was conducted in accordance with the Declaration of Helsinki. Study protocols were approved by the Ethics Committee Medical College of Qingdao University (QDU-HEC-2024107). Written informed consent was obtained for all samples collected and all data included in this study. The sample group GC tissues were collected from patients who responded effectively to immunotherapy. Material source Single-cell RNA sequencing (scRNA-seq) data related to GC were obtained from the Gene Expression Omnibus (GEO) (https://www.ncbi.nlm.nih.gov/geo/). The search covered the period from January 2014 to December 2024, using a combination of keywords and Boolean operators: (‘gastric cancer’ OR ‘stomach neoplasms’) AND (‘tumor microenvironment’ OR ‘TME’) AND (‘single-cell’ OR ‘scRNA-seq’). Ultimately, the dataset with accession number GSE183904 [14] was selected for inclusion in this study. The data are based on the GPL24676 platform and include samples from primary gastric tumors (26 samples), peritoneal metastases of gastric tumors (3 samples), normal gastric tissue (10 samples), and normal peritoneal tissue (1 sample). In this study, peritoneal tissue samples were excluded from the analysis, and gastric tissue samples were selected for single-cell bioinformatics analysis. Additionally, transcriptome data for GC from The Cancer Genome Atlas (TCGA)-Stomach Adenocarcinoma (STAD) cohort were obtained from TCGA database (https://portal.gdc.cancer.gov/projects/TCGA-STAD), comprising 375 STAD samples, 32 adjacent normal controls, and associated clinical information. Estimation of the immune response EaSIeR utilizes multiple RNA-seq datasets to comprehensively describe the immune response in intricate and dynamic TME [15]. This study utilized the easier package to predict the extent of the immune response in GC-related samples. The samples were categorized based on the median of the scoring results: those with scores higher than the median were classified in the immunotherapy response (IR) category, while those with scores lower than the median were classified in the immunotherapy non-response (INR) category. This approach enables a deeper exploration of the heterogeneity in the TME among GC patients with varying levels of immune therapy response. scRNA-seq data preprocessing and clustering Utilizing the Seurat package [16], this study filtered cells exhibiting the highest and lowest gene expression levels (top and bottom 1%) and those with mitochondrial gene expression above 10%. The unique molecular identifier expression data of single cells were normalized through regularized negative binomial regression. To account for potential technical variations between samples, we performed batch effect correction at the sample level using the Seurat integration workflow. The Seurat package [16] was subsequently used to filter and integrate GC scRNA-seq data for cell dimension reduction and clustering. Visualization of the clustering outcomes utilized the uniform manifold approximation and projection (UMAP) algorithm [17], which effectively depicted the single-cell transcription landscape related to differences in immunotherapy efficacy among patients with GC. Identification of cellular clusters This study employed the FindAllMarker function of the Seurat package [16]) to identify genes exhibiting differential expression between different cell clusters and cell subpopulations. Next, we manually annotated cell types by referencing laboratory experiments and prior research articles [14, 18, 19] and further annotated cell subpopulations based on specifically significant representative or functional genes. Estimation of copy number variations in GC cells The R package inferCNV (https://github.com/broadinstitute/inferCNV) was utilized to investigate the scRNA-seq data to identify evidence of large-scale chromosomal copy number alterations, such as the gain or loss of entire chromosomes or large chromosomal segments. To validate the proper annotation of malignant cells, we utilized the InferCNV package to estimate the copy number variations (CNVs) in cells from GC tumor samples, and the findings were visualized using a heatmap. We used the hg19 reference genome and applied default parameters throughout the analytical process, which is consistent with standard practices in the field. Molecular docking To evaluate the binding abilities between different molecules, we obtained the corresponding protein sequences from UniProt (https://www.uniprot.org/) and downloaded the corresponding RNA sequences in FASTA format from NCBI (https://www.ncbi.nlm.nih.gov). The binding regions between the proteins and RNA were predicted using CATRAPID (http://s.tartaglialab.com/page/catrapid). Based on the predicted regions with the highest binding energy, we constructed mRNA models for the corresponding binding sites using RNAComposer (https://rnacomposer.cs.put.poznan.pl/). Subsequently, we performed molecular docking between the mRNA and protein models using Hex 8.0.0 [20]. Finally, we visualized the docking results with PyMOL [21], adjusting the orientation and the chain colors to enhance clarity. Gene set variation analysis Gene set variation analysis (GSVA) transforms the gene expression matrix across various samples into a matrix of gene set expression across samples, enabling the assessment of disparities in distinct gene sets across samples. In this study, the top 20 genes with significant and specific expression in the different cell subpopulations were chosen as the gene set. The single-sample gene set enrichment analysis (ssGSEA) method from the GSVA package [22] was used to compute the scores of different subgroups in the TCGA-STAD cohort, after which the associations between cell subpopulation abundance and clinical prognosis characteristics were analyzed. Based on the algorithm-derived optimal cutoff, the samples were stratified into high-score and low-score groups, and the R packages survival (https://cran.r-project.org/web/packages/survival/index.html) and survminer (https://rdocumentation.org/packages/survminer/versions/0.4.9) were subsequently utilized to determine the overall survival (OS) rate and recurrence-free survival (RFS) rate. The R package pROC was used to analyze the receiver operating characteristic (ROC) curve [23]. Enrichment analysis To study the functional characteristics and potential regulatory mechanisms of the cell subpopulations, the R package clusterProfiler [24] was used for Gene Ontology Biological Process (GO BP) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses. Enriched functions or pathways were considered significant when P-value < 0.05. Gene set enrichment analysis Gene Set Enrichment Analysis (GSEA) [25] was employed to evaluate the gene distribution within an expression profile organized by phenotypic relevance. Reference genomes from MsigDB V7.4 (c5.bp.v7.0.entrez.gmt and c2.cp.kegg.v7.0.symbols.gmt) [26] facilitated our analysis of gene expression and phenotypic profiles in distinct cellular subpopulations. Trajectory analysis Single-cell transcriptional diversity is indicative of developmental potential. To further investigate the heterogeneity of malignant cells, we used the R package CytoTRACE [27] to infer the differentiation state and trajectory of GC cell subpopulations, and further validation was performed using the R package Monocle2 [28]. Cells experience distinct functional changes due to external stimuli or during development, resulting in varying levels of expression. Gene expression patterns can be used to classify cells and infer their developmental trajectories. This study utilized the R packages Monocle2 [28] and Monocle3 [29] to construct developmental trajectories within subpopulations of various cell types, and the UMAP algorithm was used to visualize the results. Gene regulatory network analysis Gene expression is impacted by the levels of transcription factors (TFs), which in turn influence the physiological functions of an organism by influencing the expression levels and patterns of specific genes. To identify the specific TFs regulating each cell subpopulation, an analysis and reconstruction of the TF-centered gene regulatory network (GRN) [30, 31] were conducted utilizing the pySCENIC python module tool. Additionally, transcription factor binding profiles were obtained from the JASPAR database (available at https://jaspar.genereg.net) to establish patterns of transcription factor regulation of each coexpressed gene module. Cell communication analysis The R package iTALK (10.1101/507871) focuses on cell populations as the targets of interactions. The tool identifies genes that exhibit high or differential expression in specific cell subpopulations and subsequently matches these genes with databases of ligands and receptors, thus revealing interactions between various cell types during the developmental process. In this study, we utilized the iTALK package to uncover noteworthy ligand–receptor interactions among subpopulation cells, thus enabling comprehensive identification of intercellular communication events. Hematoxylin and eosin staining Hematoxylin and eosin (H&E) staining was performed on gastric cancer tissue sections to examine general histopathological features(PMC10680339). Fresh GC tissue sections were baked overnight at 62℃ to ensure firm adhesion to the slides. The sections were then sequentially immersed in Deparaffinization Solution I, II, and III (Solarbio, YA0031), each for 10 min, followed by rehydration through a graded ethanol series (100%, 95%, 85%, 75%), with each step lasting 5 min, and finally rinsed in running water for 2 min. Next, the sections were stained with hematoxylin (Solarbio, G1080) for 5 min. After rinsing off the excess stain with running water, the sections were differentiated in hematoxylin differentiation solution for 2 s, rinsed again, and then blued in bluing solution (Solarbio, G1866) for 30 s. After a final rinse with clean water, the nuclear staining was observed under a microscope. Following hematoxylin staining, the sections were stained with eosin (Solarbio, G1100) for 1 min. After staining, the sections were dehydrated sequentially in 75%, 85%, 95%, and absolute ethanol I and II for 1 min each, and in absolute ethanol III for 5 min. The sections were then cleared in Clearing Solution I, II, and III, each for 5 min. Finally, the sections were mounted with neutral balsam (Solarbio, G8590), allowed to air dry, and observed under a microscope. Immunofluorescence Multiplex immunofluorescence staining was performed to identify tertiary lymphoid structures (TLSs) and visualize the spatial distribution of specific immune cells (PMC6918834). Fresh GC tissue sections were prepared and baked overnight at 62℃ to ensure firm adhesion to the slides. The sections were treated with deparaffinization solution (Solarbio, YA0031), followed by immersion in a series of graded ethanol solutions (100%, 95%, 85%, 75%) and rinsed with running water for rehydration. Antigen retrieval was performed using the EDTA method with microwave heating, followed by natural cooling to room temperature. The sections were then incubated in 0.5% Triton X-100 (Servicebio, G1203-250 ml) solution and subsequently washed with PBS (GIBCO, C10010500BT). A barrier was drawn around the tissue with a PAP pen to prevent reagent spread. Next, the sections were blocked with 5% goat serum (Solarbio, SL038) at 37℃ for 1 h. Following the blocking step, primary antibodies CD3 (Abcam, Ab135372), CD20 (Abcam, Ab64088), CD68 (Abcam, ab283654), and IL18 (Abcam, ab243091) were added to cover the tissue, and the sections were incubated overnight at 4℃. The next day, the sections were washed with PBS on a shaker at room temperature. Fluorescent secondary antibodies were then added, and the sections were incubated at 37℃ in the dark for 1 h. After incubation, the sections were washed again with PBS on a shaker at room temperature. The surrounding water was removed, and DAPI-containing antifade mounting medium (Solarbio, S2110) was added under dark conditions. Finally, images were captured using a fluorescence microscope. Data statistics and analysis All the bioinformatics analyses and research in this study were conducted on a bioinformatics cloud platform (http://www.bioinforcloud.org.cn). A significance level of P-value < 0.05 was adopted to determine statistical significance. Results The global landscape of gastric cancer patients with different immunotherapy regimens This study performed bioinformatics analysis on a total of 151,076 cells obtained from 36 samples, comprising 26 primary gastric tumors and 10 normal gastric tissues. Following data preprocessing, 115,845 high-quality single-cell transcriptomes were acquired and subsequently organized into 40 clusters (Fig. 1A). Based on the immune response scores of the 26 GC samples as a basis, we categorized the samples into two groups: IR (39,147 cells) and INR (57,268 cells). In addition, 10 cancer-adjacent samples were selected as the control group (19,429 cells) for comparison and analysis with immunotherapy-scoring samples from GC patients. Subsequently, a UMAP plot was constructed to illustrate the single-cell transcriptional landscape of GC patients with distinct immunotherapy states. The global landscape of GC patients with different immunotherapy regimens. (A). The main workflow of this study. (B). Bubble plots showing specific biomarker genes in different cell types. (C). Single-cell transcriptional landscape of different cell types in GC. (D). CNVs in malignant GC cells. (E). Comparison of cellular ecological components between the control group, IR group, and INR group. GC: gastric cancer; IR: immunotherapy response; CNVs: copy number variations; INR: immunotherapy nonresponse Through correlation analysis and cell markers within each cluster, we identified multiple cell types, encompassing GC cells (CLDN4, CLDN7, CEACAM6, CCND1, MET, and KRT7), epithelial cells (Ep; EPCAM, CD24, ITGB4, CEACAM1, and ITGA6), fibroblasts (Fib; ACTA2, COL1A1, DCN, FN1, and PDGFRA), endothelial cells (En; PLVAP, PECAM1, VWF, ENG, KDR, and CD34), macrophages (Mac; CD68, CD14, and CD163), mast cells (TPSAB1, MS4A2, and KIT), naïve T cells (CD3D, CD3E, CD3G, and CD247), CD8 + T cells (CD8A, CD8B, and IFNG), B cells (MS4A1, CD79A, CD79B, CD19, and CD70), and plasma cells (CD79A, JCHAIN, XBP1, and TNFRSF17) (Fig. 1B, C). Furthermore, the study validated the existence of multiple CNVs in GC cells by employing inferCNV, as illustrated in Fig. 1D. Previous research has demonstrated that the presence of tertiary lymphoid structures (TLSs) within tumors is frequently associated with widespread immune infiltration [32]. Both immune cells and stromal cells play crucial roles in either initiating or sustaining TLSs [32]. The abundances of Mac, naïve T, CD8 + T, and B cells were significantly greater in the IR and INR groups than in the control group (Fig. 1E). This finding suggests enhanced immune cell infiltration in GC tissue, which might contribute to the formation of TLSs. Notably, En and Fib were significantly lower in the INR group than in the control and IR groups. These results further confirm the heterogeneity of the TME in GC patients under different treatment conditions. Dynamics of GC cell subpopulations in immune response States This study utilized dimensionality reduction clustering to analyze GC cells and projected the resulting clusters onto UMAP, identifying 7 distinct GC cell subpopulations (Fig. 2A). The map displayed specific markers associated with each of the GC cell subpopulations (Fig. 2B). By observing the dynamic changes in GC cell subpopulations among different patients, we found that CYP3A5 + GC was a major component of the INR group, while APOA1 + PDIA6 + GC was specifically enriched in the IR group (Fig. 2C; Supplementary Fig. 1A). Furthermore, our analysis of the GC cell atlas revealed a significant positive correlation (Cor = 0.838, p-value < 0.001) between APOA1 and PDIA6, which are co-expressed in APOA1 + PDIA6 + GC (Supplementary Fig. 1B, C). Molecular docking analysis revealed that the PDIA6 protein can target and bind to APOA1-mRNA, resulting in a binding score of -691.57 KJ/mol (Fig. 2D). Furthermore, we observed that elevated levels of APOA1 + PDIA6 + GC are associated with adverse prognosis in GC patients (Supplementary Fig. 1D). The heterogeneity of GC cells under different immunotherapy regimens. (A). Single-cell atlas of GC cell subpopulations. (B). The heatmap shows the expression of each marker gene in the GC cell subpopulations. (C). The dynamic changes in GC cell subpopulations in different groups. (D). Molecular docking of PDIA6-protein with APOA1-mRNA. (E). The signaling pathways associated with APOA1 + PDIA6 + GC or ATF4 + GC. (F). Signaling pathways significantly activated or inhibited by APOA1 + PDIA6 + GC or ATF4 + GC. (G). Boxplots showing CytoTRACE values from the most mature (highest values) to the most immature (low values) GC cell subpopulations. (H). Trajectory analysis of GC cell subpopulations. (Left) Pseudotime analysis mapping of GC cell subpopulations and pseudotime values to display the differentiation trajectory of GC cell subpopulations. (Right) Pseudotime analysis mapping the various GC cell subpopulations. (I). The heatmap of motif modules for GC cell subpopulations. GC: gastric cancer Enrichment analysis revealed that multiple subpopulations of GC cells were significantly enriched in signaling pathways related to cell development, energy metabolism, and cell proliferation, indicating that GC cells adapt to their rapid proliferation and survival requirements by perturbing energy homeostasis (Fig. 2E; Supplementary Fig. 1E). Furthermore, we found that APOA1 + PDIA6 + GC cells significantly activated oxidative phosphorylation and PPAR signaling pathways (Fig. 2F), which are associated with cell growth and development. Interestingly, ATF4 + GC cells significantly activated cytokine-cytokine receptor interaction, cell apoptosis, and chemokine signaling pathways (Fig. 2F), while CYP3A5 + GC significantly inhibited Oxidative phosphorylation and Glycolysis/Gluconeogenesis, instead of activating D-Glutamine and D-glutamate metabolism and ErbB signaling pathway (Fig. 2F, G). These findings suggest the presence of more complex mechanisms within the tumor microenvironment, warranting further investigation. According to the ordering predicted by CytoTRACE, the APOA1 + PDIA6 + GC exhibited the highest degree of cellular differentiation potential, whereas the CYP3A5 + GC subpopulation may represent a functionally mature subpopulation (Fig. 2G). Furthermore, cell stemness scores further support these findings (Supplementary Fig. 2I). The differentiation trajectory of the GC cell subpopulation also indicated that APOA1 + PDIA6 + GC was positioned at the inception of the developmental pathway, while CYP3A5 + GC was situated closer to the later stages of this trajectory (Fig. 2G). These findings suggest that intervention in APOA1 + PDIA6 + GC may influence the differentiation status of GC cells. We subsequently constructed GRNs to investigate the specific TFs that regulate different subpopulations of GC cell (Fig. 2H). Notably, the crucial regulatory TFs RFX5 and HOXB6, which govern CYP3A5 + GC and APOA1 + PDIA6 + GC, respectively, were clustered into the same module, indicating analogous expression profiles within both subpopulations. HOXB6 plays a role in regulating the gene expression pattern of adult esophageal cells [33]. Thus, targeting HOXB6 may inhibit the formation and differentiation of GC cells, thereby enhancing treatment efficacy. Increased CDH13 + TEC and NDUF4L2 + TEC contributed to tumor angiogenesis We performed a subpopulation analysis of En and identified 8 distinct subpopulations (Fig. 3A, B). Patients under different immunotherapy regimens exhibited significant increases in the abundances of ACKR1 + tumor-associated endothelial cell (TEC) and CDH13 + TEC compared to those in the control group (Fig. 3C, D). Notably, the changes in the abundances of both ACKR1 + TEC and CDH13 + TEC subpopulations in the IR and INR patients showed significant discrepancies. Survival analysis revealed that patients with a high abundance of CDH13 + TEC who responded poorly to immune treatment exhibited unfavorable overall survival (OS) and recurrence-free survival (RFS) (Fig. 3E). Additionally, the observed changes in the abundances of ACKR1 + TEC at both the bulk and single-cell levels were consistent (Fig. 3F), indicating that the expression pattern of ACKR1 + TEC was relatively stable in GC. The heterogeneity of endothelial cells under different immunotherapy regimens. (A). Single-cell atlas of the En subpopulations. (B). The density plot shows the expression of each marker gene in the En subpopulations. (C). The single-cell atlas shows En subpopulations across different groups. (D). The dynamic changes in the En subpopulations in the different groups. (E). Survival curve of the CDH13 + TEC subpopulation. (F). Boxplot showing the changes in abundance of the ACKR1 + TEC subpopulation at the bulk level. (G). Hypoxia and angiogenesis gene set scoring of the En subpopulation. (H). Hypoxia and angiogenesis gene set scoring of CDH13 + TEC and NDUFA4L2 + TEC across different groups. (I). The signaling pathways associated with the En subpopulations. (J). Trajectory analysis of the En subpopulations. (K). Activity of TFs that regulate En subpopulations. En: endothelial cell; TEC: tumor-associated endothelial cell; TFs: transcription factors Angiogenesis supplies these vital resources to support the growth of cancer cells [34](. Compared to other cell subpopulations, the hypoxia and angiogenesis scores of CDH13 + TEC and NDUFA4L2 + TEC were markedly different, we infering a potential link to malignant cell proliferation and migration (Fig. 3G). Interestingly, the CDH13 + TEC and NDUFA4L2 + TEC in the IR group had greater hypoxia and angiogenesis scores compared to other groups (Fig. 3H). Enrichment analysis revealed the significant involvement of ACKR1 + TEC, CDH13 + TEC, and NDUFA4L2 + TEC in pathways related to cell proliferation, differentiation, migration, and angiogenesis (Fig. 3I). Pseudotime analysis of the En subpopulations revealed distinct endpoints for ACKR1 + TEC, CDH13 + TEC, and NDUFA4L2 + TEC along the developmental trajectory (Fig. 3J). These findings suggest that these fully differentiated subpopulations might be TEC subpopulations with diverse expression characteristics. GRN analysis revealed that the subpopulations exerting angiogenic effects were predominantly influenced by FOXA1 and NR2F6 (Fig. 3K). Suppressing the expression of these genes in endothelial cells might enhance the effectiveness of immunotherapy. In conclusion, the subpopulations that promote angiogenesis were notably activated in GC patients, thereby fostering tumor growth and complicating immunotherapy. RGS5 + CAF and CFD + CAF promoted reprogramming of the tumor microenvironment The investigation of Fib cell subpopulations identified a total of 6 subpopulations, and distinctive expression markers for each subpopulation were visualized via UMAP (Fig. 4A, B). By assessing the high expression levels of cancer-associated fibroblast (CAF)-related markers (FAP, TAGLN, ACTA2, VIM, S100A4, and CSPG4) and the alterations in subpopulation abundance (Fig. 4C, D), we identified four CAF subpopulations: CA12 + CAF, RGS5 + CAF, NGFR + CAF, and CFD + CAF. Notably, CFD + CAF exhibited high expression of chemokines (CXCL1, CCL2, and CXCL12), which could contribute to the induction of an immunosuppressive microenvironment. Compared to the IR group, the INR group showed a substantial increase in the abundance of RGS5 + CAF and CFD + CAF. Analysis of the survival curves for RFS and OS revealed that an increase in the presence of RGS5 + CAF and CFD + CAF was associated with an unfavorable prognosis in patients with GC (Fig. 4E, F). The heterogeneity of fibroblasts under different immunotherapy regimens. (A). Single-cell atlas of Fib subpopulations. (B). The density plot shows the expression of each marker gene in the Fib subpopulations. (C). Expression of cytokines and CAF-related cell markers in each Fib subpopulation. (D). The dynamic changes in Fib subpopulations in different groups. (E). Survival curve of the RGS5 + CAF subpopulation. (F). Survival curve of the CFD + CAF subpopulation. (G). The signaling pathways associated with the Fib subpopulations. (H). Trajectory analysis of Fib subpopulations. (I). Activity of TFs that regulate Fib subpopulations. Fib: fibroblast; CAF: cancer-associated fibroblast; TFs: transcription factors Enrichment analysis indicated that RGS5 + CAF was significantly enriched in focal adhesion, ECM-receptor interaction, oxidative phosphorylation, and the MAPK signaling pathway, while CFD + CAF was significantly enriched in cytokine-cytokine receptor interaction, the NOD-like receptor signaling pathway, and the MAPK signaling pathway (Fig. 4G). A previous report showed that the TGF-beta signaling pathway can induce the transformation of normal Fib into CAF [35]. Pseudotime analysis suggested that the CXCL14 + Fib subpopulation was at the starting point of the developmental trajectory of the Fib subpopulation (Fig. 4H) and was significantly enriched in the TGF-beta signaling pathway, playing an important role in promoting Fib-to-CAF conversion. CFD + CAF represented an intermediate state of a developmental trajectory, whereas RGS5 + CAF was positioned at one end of the trajectory. Analysis of GRNs revealed that NFIL3 is a key transcriptional regulator of CFD + CAF and CDX2 is a key transcriptional regulator of RGS5 + CAF (Fig. 4I). Targeting these transcription factors might regulate the abundance of tumor-associated fibroblasts, thereby inhibiting tumor growth and invasion. Therefore, the increased abundance of RGS5 + CAF and CFD + CAF promoted reprogramming of the TME. IL18 + Mac was associated with immune infiltration and inflammatory response in TME Dimensionality reduction clustering was performed on the macrophages, resulting in the projection of 7 macrophage subpopulations in the UMAP plot and the identification of specific markers expressed by each subpopulation (Fig. 5A, B). Due to the significant enrichment of FN1 + tumor-associated macrophages (TAM), SLC25A37 + TAM, and IGHG4 + TAM in the two groups of GC patients compared to the control group, these were defined as TAM subpopulations (Fig. 5C). We observed that the IL18 + macrophage subpopulations exhibited significantly greater abundances in the IR group compared to the INR group, which was significantly positively correlated with TLSs (Cor = 0.782, p-value < 0.001) (Fig. 5D). Notably, IL18 + Mac and SLC25A37 + TAM showed relatively high expression levels of various pro-inflammatory factors (Fig. 5F). At the bulk level, the abundance of IL18 + Mac was greater in the IR group than in the INR group, consistent with the changes in macrophage subpopulations observed at the single-cell level (Fig. 5G). The ROC curve indicated the excellent ability of the IL18 + Mac signature (AUC = 0.908) to predict patient response to immunotherapy treatment (Fig. 5H). In this study, most macrophage subpopulations were significantly enriched in immune and inflammation-related signaling pathways (Supplementary Fig. 2A), indicating that modulation of macrophage function could have a significant impact on cancer therapy. GSEA indicated that IL18 + Mac significantly activated the cytokine receptor-ligand interaction signaling pathway (Fig. 5I). Through pseudotime analysis (Supplementary Fig. 2B, C), we constructed a developmental trajectory of macrophages. The IL18 + Mac, situated at the terminus of the developmental trajectory, represents a mature and functionally specialized macrophage subpopulation. The analysis of specific TFs within each subpopulation revealed that LHX9 has the capacity to concomitantly regulate two crucial macrophage subpopulations (Fig. 5K), the IL18 + Mac and SLC25A37 + TAM subpopulations. Overall, we emphasized the dynamic changes in macrophage subpopulations during immune resistance, revealing the potential role of IL18 + macrophages in the immune infiltration and inflammatory response. The heterogeneity of macrophages under different immunotherapy regimens. (A). Single-cell atlas of Mac subpopulations. (B). The density plot shows the expression of each marker gene in the Mac subpopulations. (C). The dynamic changes in Mac subpopulations in different groups. (D). Correlations between IL18 + Mac and TLSs. (E). Expression of cytokines in each Mac subpopulation. (F). Boxplot showing the changes in the abundances of the IL18 + Mac and SLC25A37 + TAM subpopulation at the bulk level. (G). Diagnostic efficacy of the IL18 + Mac subpopulation. (H). Signaling pathways significantly activated or inhibited by IL18 + Mac subpopulation. (I). Pseudotime analysis mapping of macrophage subpopulations and pseudotime values to display the differentiation trajectory of macrophage subpopulations. (J). Activity of TFs that regulate Mac subpopulations. Mac: macrophage; TLSs: tertiary lymphoid structures; TAM: tumor-associated macrophage; TFs: transcription factors IL18 + Mac within TLSs predict immunotherapy response in GC We collected imaging and clinical data from GC patients who underwent four cycles of immunotherapy combined with chemotherapy (Fig. 6A-C). Pathological immunohistochemistry revealed PD-L1 expression levels below 1% in patients, typically indicating a poor response to immunotherapy. However, patients achieved favorable outcomes after four cycles, evidenced by reductions in tumor size and lymph node involvement. This suggests that, in addition to PD-L1, other predictive biomarkers such as TLSs may exist and could more accurately predict immunotherapy efficacy. Subsequently, we employed multiple immunofluorescence staining using markers CD3 and CD20 to identify the organized arrangement of T cells and B cells, respectively, indicative of TLSs. Analysis of clinical tissues from these patients revealed the presence of TLSs, as evidenced by the structured localization of CD3 + and CD20 + cells (Fig. 6D). Furthermore, within these TLSs, we observed positive expression of CD68 and IL18, highlighting their association with the inflammatory microenvironment. These findings not only deepen our understanding of the tumor immune microenvironment in GC but also identify new potential biomarkers for predicting the efficacy of immunotherapy. Imaging studies and immunofluorescence of GC patients responding to immunotherapy. (A). CT imaging of Patient (1) The patient underwent a radical gastrectomy for GC. Postoperative pathology of GC: (gastric antrum) poorly differentiated carcinoma, consistent with poorly differentiated adenocarcinoma. Immunohistochemistry: CEA(+), CK(-), CK7(+), Ki-67 (approximately 70%+), P53(+), HER2 (0), MLH1(+), MSH2(+), MSH6(+), PMS2(+), in situ hybridization for EBER(-). PD-L1 (SP263): TPS less than 1%, CPS less than 1, positive control (+), negative control (-). One month after surgery, the patient was diagnosed with recurrent GC with perigastric and retroperitoneal lymph nodes metastasis, and peritoneal metastasis. Treatment: The patient received 4 cycles of sintilimab combined with oxaliplatin and tegafur as first-line therapy. Efficacy evaluation: After 2 and 4 cycles of treatment, the efficacy was evaluated as reduced stable disease. (B). T2-weighted fat-suppressed MRI images of Patient (2) (C). Enhanced MRI sequence images of Patient 2. Gastroscopy showed GC, and pathology indicated a malignant tumor at the gastric body and antrum junction, considered to be poorly differentiated adenocarcinoma: CK8/18(+), cgA(+), syn(+), HER2(0), PMS2(+), MLH1(+), MSH2(+), MSH6(+), Ki-67 (approximately 80%+), in situ hybridization for EBER(-), PD-L1 (SP263): TPS less than 1%, CPS less than 1, positive control (+), negative control (-). Diagnosis: gastric cancer with perigastric lymph node metastasis and liver metastasis. Treatment: The patient received 4 cycles of first-line therapy with sintilimab, oxaliplatin and tegafu. Efficacy evaluation: After 2 and 4 cycles of treatment, the efficacy was evaluated as partial response. (D). Hematoxylin and eosin staining and immunofluorescence results of tumor tissues from immunotherapy-responsive GC patients. GC: gastric cancer XCL1 + NR4A2 + effector CD8 + T cells play a complex role in antitumor immunity Through clustering analysis of CD8 + T cells, a total of 6 cell subpopulations were identified, and UMAP plots were generated to depict the specific marker expression of each subpopulation (Fig. 7A, B). Compared to the control group, patients in the IR group showed a marked decrease in XCL1 + NR4A2 + effector CD8 + T (Te) cell subpopulation, while the INR group exhibited a significant increase (Fig. 7C). The differentiation and developmental trajectory of CD8 + T-cell subpopulations indicated that GZMK + exhausted T (TEX) cells were located at one endpoint of the developmental trajectory (Fig. 7D). These cells exhibited high levels of exhaustion-related genes (PDCD1 and LAG3) and antigen-presenting genes (HLA-DRB1, HLA-DPB1, HLA-DRB5, and HLA-DPA1) (Fig. 7E, F; Supplementary Table 1). The heterogeneity of CD8 + T cells under different immunotherapy states. (A). Single-cell atlas of CD8 + T cell subpopulations. (B). The density plot shows the expression of each marker gene in the CD8 + T cell subpopulations. (C). Dynamic changes in CD8 + T cell subpopulations in different groups. (D). Trajectory analysis of CD8 + T cell subpopulations. (E). Expression of CD8 + T cell subpopulation biomarkers in each CD8 + T cell subpopulation. (F). GZMK + TEX exhibited distinct patterns of gene expression associated with exhaustion across the different groups. (G). Survival curve of the XCL1 + NR4A2 + Te subpopulation. (H). Boxplot showing the changes in abundance of the GZMK + TEX subpopulation at the bulk level. (I). Diagnostic efficacy of the XCL1 + NR4A2 + Te and GZMK + TEX subpopulations. (J). Co-expression of XCL1 and NR4A2 in the XCL1 + NR4A2 + Te subpopulation. (K). Correlation between XCL1 and NR4A2 in the XCL1 + NR4A2 + Te subpopulation. (L). The signaling pathways associated with CD8 + T cell subpopulations. (M). Activity of TFs that regulate CD8 + T cell subpopulations. TEX: exhausted T cells; Te: effector CD8 + T cells; TFs: transcription factors Notably, a high abundance of XCL1 + NR4A2 + Te was associated with a poor prognosis in patients undergoing immunotherapy (Fig. 7G). At the bulk level, the abundance of CD8 + T-cell subpopulations showed significant consistency, particularly for GZMK + TEX cells (Fig. 7H). Diagnostic analysis demonstrated that the XCL1 + NR4A2 + Te (AUC = 0.857) and GZMK + TEX (AUC = 0.903) subpopulations identified patients’ immunotherapy statuses (Fig. 7I). Detailed analysis of the XCL1 + NR4A2 + Te subpopulation revealed high levels of NR4A1, NR4A2, and NR4A3, along with specific expression of XCL1 and XCL2 (Supplementary Table 1). According to the single-cell atlas of CD8 + T cells, XCL1 and NR4A2 were colocalized and demonstrated a substantial positive correlation in XCL1 + NR4A2 + Te (Cor = 0.751, p-value < 0.001) (Fig. 7J, K). XCL1 + NR4A2 + Te was significantly enriched in various pathways related to cytotoxicity, cell growth, development, and immune responses (Fig. 7L). This highlights the complex role of this subpopulation in regulating immune responses. Furthermore, the CD8 + T cell subpopulations were primarily regulated by KLF16, LTF, FOSL1, FOXP3, and HIVEP1 (Fig. 7M). Therefore, the XCL1 + NR4A2 + Te subpopulation demonstrates intricate immune mechanisms related to antitumor immunity. Cellular communication in the tumor microenvironment To understand the interactions within the TME, we constructed communication networks involving cytokines, immune checkpoints, growth factors, and the extracellular matrix (ECM) (Fig. 8A-D). In cytokine communication, we observed that macrophage subpopulations targeted ATF4 + GC via the IL1B-IL1R2 axis. IL18 + Mac also interacts with ATF4 + GC via CXCL10-SDC4 axis, promoting immune responses with the TME. In immune checkpoint communication, a notable interaction was abserved between the CD24 ligand expressed on ATF4 + GC and the SIGLEC10 receptor on macrophages, indicating a mechanism for immune evasion. During growth factor communication, CAF subpopulations, predominantly expressing CTGF, influence TEC subpopulations via ITGA5 and immune cells through ITGB2, highlighting pathways for tumor angiogenesis and immune modulation. In communication with the ECM, IL18 + Mac and FN1 + TAM expressed SPP1, interacting with CAF and TEC through multiple receptors, which may contribute to the structural support and signaling within the TME. Cellular communication in the tumor microenvironment. (A). The chemokine receptor‒ligand pairs. (B). Immune checkpoint receptor‒ligand pairs. (C). Growth factor receptor‒ligand pairs. (D). Other receptor‒ligand pairs Discussion In this research, using primary GC data obtained from scRNA-seq, we developed a comprehensive understanding of the single-cell transcription profiles of GC patients in various immunotherapy response states to reveal the cell subpopulations that play important roles in the TME of GC and explore the communication between GC cells and TME. This study highlights the heterogeneity of GC cells under different immune states. We found that APOA1 + PDIA6 + GC, specifically enriched in the IR group, was at the developmental starting point of GC cells and exhibited strong cellular stemness. APOA1, linked to the PPAR pathway, is associated with extensive metaplasia and poorer prognosis in intestinal-type GC [36]. PDIA6, an RNA-binding protein (RBP), plays a role in melanoma metastasis [37]. We demonstrated that APOA1 + PDIA6 + GC significantly activated the PPAR signaling pathway, with APOA1 and PDIA6 being co-expressed and positively correlated. Additionally, PDIA6 protein showed strong binding affinity with APOA1 mRNA, suggesting that PDIA6 may regulate APOA1 expression through binding to APOA1 mRNA, thereby activating the PPAR pathway and promoting GC cell proliferation. Conversely, CYP3A5 + GC cells were predominant in the INR group, potentially impacting immunotherapy efficacy. CYP3A5 is crucial for drug metabolism and mediates chemotherapy resistance [38, 39]. While tumor cells typically rely on glycolysis, recent research indicates that immune cells in the tumor microenvironment primarily consume glucose, with cancer cells uptaking glutamine [40–42]. Glutamine plays an indispensable role in maintaining the oxidative-reductive balance and biosynthesis of biomolecules in tumors, actively inhibiting glucose metabolism in the TME [42, 43]. In our study, mature CYP3A5 + GC exhibited a unique metabolic pattern, suppressing oxidative phosphorylation and glycolysis while activating glutamine metabolism and the ErBb signaling pathway. This suggests that mature GC cells may utilize glutamine for energy, inhibit glucose metabolism, and activate the ErBb pathway to promote tumor survival and therapeutic resistance. Targeting CYP3A5 with metabolic inhibitors (e.g., glutamine analogs) represents a potential strategy to overcome this form of resistance. Angiogenesis is one of the hallmarks of cancer, but its association with immunotherapy has not been largely been explored [44]. Our study demonstrated that CDH13 + TEC was associated with both cell proliferation and migration, as well as angiogenesis and hypoxia. When CDH13 is overexpressed in endothelial cells, it is an atypical member of the cadherin family [45] and promotes their proliferation and migration [46]. The induction of CDH13 in microvessels associated with tumors and metastasis promotes tumor angiogenesis [47]. Furthermore, the increase in CDH13 + TEC was significant in GC patients with varying immune response states, consequently supplying increased nutrients and oxygen for tumor development, impacting the efficacy of immunotherapy and contributing to the unfavorable prognosis of GC patients. In contrast to those in the INR group, the angiogenesis and hypoxia scores were greater in the IR group than in the control group, coupled with a lower abundance of CDH13 + TEC and NDUFA4L2 + TEC. These findings further indicate that targeting these cells in the TME may improve the efficacy of immunotherapy in GC patients. CAF constitutes a major TME component, influencing cancer hallmarks and immunotherapy prognosis [48]. We indicated that elevated abundances of RGS5 + CAF and CFD + CAF correspond to unfavorable prognoses in GC patients across the INR. Reports have indicated that RGS5 + CAF perform diverse functions in multiple cancer types, encompassing their impact on the TME, energy metabolism, tumor metastasis, and immune suppression [49–51], which is consistent with our research. Previous studies have shown that the adhesion and signal transduction between cells and the ECM in cancer constitute important factors influencing the resistance of cancer cells to treatment [52]. CAF constitutes the primary regulatory factor that shapes tumor metabolism, contributing to tumor growth, progression, and evasion through metabolic reprogramming [53]. Therefore, enrichment analysis revealed that the involvement of RGS5 + CAF in adhesion, cellular migration and energy metabolism pathways may enhance the barrier effects against drug and immune cell infiltration, promoting tumor cell migration. The participation of CFD + CAF in cellular activities and communication pathways may preserve inhibitory immune cell subpopulations through the secretion of chemokines. Furthermore, RGS5 + CAF and CFD + CAF secreted significant amounts of CTGF, influencing the TME through growth factor interactions. CTGF plays a crucial role in regulating cell processes and promoting inflammation, fibrosis, and cancer [54]. Additionally, ITGA5 promoted tumor angiogenesis and enhanced patient drug resistance, while ITGB2 was essential for immune infiltration in various tumor environments [19, 55–57]. Consequently, RGS5 + CAF and CFD + CAF may facilitate angiogenesis through the CTGF-ITGA5 axis while simultaneously inhibiting immune cell infiltration through the CTGF-ITGB2 axis. These findings suggest that therapeutic inhibition of RGS5 may disrupt these pro-tumorigenic pathways and potentially reverse associated treatment resistance.​. TAM are pivotal innate immune cells that significantly shape the tumor microenvironment [35, 36]. Our study highlights a significant enrichment of IL18 + Mac in the IR group, which are associated with tertiary lymphoid structures (TLSs) and express various proinflammatory factors. IL-18 is a potent cytokine involved in macrophage activation [58, 59]. While current immune therapies predominantly target T cell-mediated responses, T cells are not the only immune cells capable of exerting antitumor effects. In particular, the presence of IL18 and CD68 positive cells within TLSs in tumors of immunotherapy-responsive patients suggests that IL18 + Mac infiltration is indicative of effective immune responses. This capability to differentiate immune therapy responders underscores the potential of IL18 + Mac as biomarkers for therapeutic efficacy. Therefore, strategies to modulate IL-18 signaling, such as supplementation, hold potential to enhance the efficacy of immunotherapy, a premise that warrants future investigation. In conclusion, the elevated abundance of IL18 + Mac may facilitate the formation of TLSs by recruiting inflammatory factors, thereby stimulating immune cell activation and enhancing the effectiveness of immune treatments. Monitoring IL18 + Mac levels and their proinflammatory activity could enable the prediction of treatment outcomes and the adjustment of therapeutic strategies accordingly. Furthermore, previous studies have associated oxidative phosphorylation with anti-inflammatory macrophages and glycolysis with proinflammatory macrophages [60]. Notably, TAM subpopulations characterized by the expression of SLC25A37, which concurrently express various proinflammatory factors, demonstrated significant enrichment in the oxidative phosphorylation pathway. The specific transcription factor LHX9 of SLC25A37 + TAM also regulated the inflammatory phenotype of IL18 + Mac, further supporting the role of SLC25A37 + TAM in promoting inflammation response in the TME. Moreover, GC subpopulations interact with immune cell subpopulations through various ligand–receptor interactions. There was a significant cytokine relationship between the ATF4 + GC tumor cell subpopulation and immune cell subpopulations, where FN1 + TAM, SLC25A37 + TAM, and IL18 + Mac interacted through IL1B-IL1R2. Previous research has demonstrated that tumor cells expressing IL1R2 may inhibit IL1B-mediated pro-inflammatory signaling [61]. These findings further indicate that IL1B-IL1R2 serves as a key ligand in the communication between TAM and GC cells and participates in the regulation of antitumor immunity. Importantly, both ATF4 + GC and IL18 + Mac significantly activated the cytokine receptor-ligand pathway, suggesting a strong communication relationship between them. We found that IL18 + Mac targeted ATF4 + GC through CXCL10-SDC4. CXCL10-producing macrophages initiate a chemokine network that establishes a “hot” TME, correlating with improved responses to immunotherapy [62]. The expression of CXCL10 in GC cell lines contributes to the intratumoral infiltration of T cells [63]. In conclusion, we speculate that IL18 + Mac may influence ATF4 + GC cells through cytokine signaling, potentially activating the chemokine pathway, promoting immune infiltration, with a potential link to tumor cell apoptosis. Furthermore, the abundance of ATF4 + GC was higher in the IR group, further indicating that, compared to the INR group, there was an enrichment of apoptotic tumor cells in the IR group. These findings highlight a potential interaction between IL18 + Mac and ATF4 + GC in modulating the tumor immune microenvironment, suggesting potential therapeutic targets. Future studies are warranted to experimentally validate these interactions and explore their implications in enhancing immunotherapeutic strategies for GC. The efficacy of immune checkpoint blockade therapy may correlate with variances in the status of tumor-infiltrating T cells across cancer types, especially exhaustion [64]. Our research revealed that within the CD8 + T cells in GC, GZMK + TEX was significantly more abundant in the IR group compared to the INR group. These cells exhibited specific expression patterns and demonstrated reliable diagnostic capabilities. Despite expressing exhaustion-associated genes, GZMK + TEX may play a crucial role in overcoming immune suppression and eliminating tumor cells. Notably, most effective cancer immunotherapies involve the effector functions of CD8 + T cells [34]. Therefore, understanding effector T-cell function within cancer cells is vital for improving the immune response to tumors. Compared to those in the IR group, patients in the INR group had a greater abundance of XCL1 + NR4A2 + Te, suggesting their potential role in promoting tumor progression. The NR4A transcription factor family, which comprises NR4A1, NR4A2, and NR4A3, is known to advance tumor growth by repressing cytokine production [65–67]. The chemokine XCL1, serving as a distinctive marker of the XCL1 + NR4A2 + Te subpopulation, is produced by T cells in response to inflammation [68]. Our findings revealed a co-expression and positive correlation between NR4A2 and XCL1 in XCL1 + NR4A2 + Te, with elevated levels correlating with adverse outcomes in GC patients. This finding suggested that NR4A2 potentially modulates the expression of XCL1 and associated genes through a negative feedback loop, consequently inhibiting inflammatory pathways and the body’s antitumor immune response. Furthermore, NR4A family genes are pivotal in regulating CD8 + T cell exhaustion [69, 70], which is triggered and driven by T cell receptor (TCR) signaling, with heightened Nr4a3 levels linked to more robust TCR signals [71, 72]. Notably, excessive TCR signaling may induce T-cell dysfunction and suppress the MAPK signaling pathway [73]. In this study, the XCL1 + NR4A2 + Te subpopulation was significantly enriched in the TCR and MAPK signaling pathways. Hence, we postulated that elevated NR4A levels reflect sustained and intense TCR signaling within the tumor immune microenvironment. This may dampen the MAPK pathway, facilitate CD8 + T cell exhaustion, and thereby undermine the effectiveness of the body’s antitumor immune response. Beyond identifying cellular mechanisms underlying immunotherapy resistance, our study implemented an ML-based prediction strategy using single-cell–defined resistance signatures. By integrating metabolic malignant states (CYP3A5 + GC), stromal remodeling programs (RGS5⁺/CFD⁺ CAFs), and dysfunctional T-cell phenotypes (XCL1 + NR4A2 +Te), the model delineated a resistance spectrum rather than a binary classification. Importantly, IL18 + Mac emerged as resistance-protective immune effectors, in agreement with their localization within TLSs and association with tumor cell apoptosis. These findings demonstrate that biologically grounded ML approaches can predict immune-escape trajectories and identify candidates for combination strategies, such as glutamine metabolism blockade, CAF targeting, and NR4A pathway modulation. This study investigated the heterogeneity of the TME in GC patients receiving different immunotherapy regimens using scRNA-seq data. However, several limitations remain. First, the study exclusively included primary samples of GC patients without incorporating samples from metastatic GC. Second, the absence of detailed immunotherapy regimen data (e.g., dosage, cycles) limits the interpretation of patient-specific responses. Third, the study findings were obtained through bioinformatics analysis, and further validation in additional clinical cohorts is necessary. Finally, the observational nature of our current study limits the establishment of causative relationships. We proposed that IL18 + Mac may interact with ATF4⁺ GC via the cytokine signaling axis to activate apoptotic signaling in ATF4 + GC. This mechanism highlights a targetable interaction that may serve as a viable alternative to immune checkpoint blockade (ICB) therapy in specific patient subsets. Although further functional validation is required, such strategies hold the potential to overcome the limitations of current immunotherapies by addressing tumor microenvironment (TME) heterogeneity in a more comprehensive manner. Further research incorporating metastatic GC samples and additional experimental validation will be crucial to confirm and expand upon these findings. Conclusion Collectively, our study unveiled the intricate TME in GC patients across diverse immune response states, elucidating the roles of distinct cellular subpopulations and their correlations with tumor advancement. Notably, in GC patients who responded effectively to immunotherapy, there was infiltration of IL18 + Mac within TLSs. In summary, while current immunotherapies for gastric cancer are often limited by low response rates, our study provides a high-resolution blueprint of the GC TME that addresses these shortcomings. The novel biomarkers and therapeutic targets we identified, such as the glutamine-metabolizing enzyme CYP3A5 and the PDIA6-APOA1 regulatory axis, represent promising candidates for future development as alternatives. They not only enable better patient stratification beyond conventional markers but also outline novel combination regimens designed to overcome the specific metabolic and stromal mechanisms of resistance, ultimately aiming to improve clinical outcomes for a broader population of GC patients. Supplementary information Below is the link to the electronic supplementary material. Supplementary Material 1: Biological functions enriched in each gastric cancer cell subpopulation SF1A. The single-cell atlas shows GC cells across different groups. SF1B. Correlation between APOA1 and PDIA6 in the APOA1+ PDIA6+ GC subpopulation. SF1C. Co-expression between APOA1 and PDIA6 in the APOA1+ PDIA6+ GC subpopulation. SF1D. Prognostic relevance of APOA1+ PDIA6+ GC in GC patients. SF1E. The signaling pathways associated with the CYP3A5+ GC subpopulation. SF1F. Signaling pathways significantly activated or inhibited by CYP3A5+ GC. SF1G. Regulatory mechanisms of CYP3A5+ GC on glycolysis/gluconeogenesis signaling pathways. SF1H. Regulatory mechanisms of CYP3A5+ GC on oxidative phosphorylation signaling pathways. SF1I. Stemness scores of GC cell subpopulations. GC: gastric cancer Supplementary Material 2: Enrichment analysis and pseudo-temporal analysis of macrophage subpopulations. SF2A. The signaling pathways associated with macrophage subpopulations. SF2B. Pseudotime analysis mapping the various macrophage subpopulations Supplementary Material 3 Acknowledgements Not applicable. Abbreviations Gastric cancer Single-cell RNA sequencing Tumor microenvironment The Cancer Genome Atlas Stomach Adenocarcinoma Immunotherapy response Immunotherapy nonresponse Uniform manifold approximation and projection Copy number variations Gene set variation analysis Single-sample gene set enrichment analysis Overall survival Recurrence-free survival Receiver operating characteristic Gene Ontology Biological Process Kyoto Encyclopedia of Genes and Genomes Gene Set Enrichment Analysis Gene regulatory network Epithelial cells Fibroblasts Endothelial cells Macrophages Tertiary lymphoid structures Tumor-associated endothelial cells Cancer-associated fibroblasts Tumor-associated macrophages Extracellular matrix Author contributions LZ, XLL and WJG designed the study. JMW, and SM performed the data analysis. TQZ, MRL and YJO visualized the results. LZ, XLL, JMW, TQZ, MRL, YJO, HZ and SM drafted the manuscript. XQH, JYX and WJG reviewed and edited the manuscript. All the authors read and approved the final manuscript. Funding This work was supported by Medical and Health Science and Technology Development Program of Shandong Province (Grant No. 202303110610). Data availability The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request. Declarations Ethics approval and consent to participate This study was approved by the Institutional Review Board and Ethical Committee of Medical College of Qingdao University. The studies involving human participants complied with the Declaration of Helsinki and were reviewed and approved by the Ethics Committee Medical College of Qingdao University (QDU-HEC-2024107). Samples were obtained from all patients whose informed consent was obtained from them and/or their legal guardians. Consent for publication Not applicable. Competing interests The authors declare that they have no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Lu Zhou and Xuelong Li co-first authors. Contributor Information Xueqiong Han, Email: hanxueqiong@stu.gxmu.edu.cn. Jinyu Xiang, Email: kiwi2010@163.com. Wenjing Gong, Email: gongwenjing@qdu.edu.cn. References Associated Data Supplementary Materials Supplementary Material 1: Biological functions enriched in each gastric cancer cell subpopulation SF1A. The single-cell atlas shows GC cells across different groups. SF1B. Correlation between APOA1 and PDIA6 in the APOA1+ PDIA6+ GC subpopulation. SF1C. Co-expression between APOA1 and PDIA6 in the APOA1+ PDIA6+ GC subpopulation. SF1D. Prognostic relevance of APOA1+ PDIA6+ GC in GC patients. SF1E. The signaling pathways associated with the CYP3A5+ GC subpopulation. SF1F. Signaling pathways significantly activated or inhibited by CYP3A5+ GC. SF1G. Regulatory mechanisms of CYP3A5+ GC on glycolysis/gluconeogenesis signaling pathways. SF1H. Regulatory mechanisms of CYP3A5+ GC on oxidative phosphorylation signaling pathways. SF1I. Stemness scores of GC cell subpopulations. GC: gastric cancer Supplementary Material 2: Enrichment analysis and pseudo-temporal analysis of macrophage subpopulations. SF2A. The signaling pathways associated with macrophage subpopulations. SF2B. Pseudotime analysis mapping the various macrophage subpopulations Supplementary Material 3 Data Availability Statement The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.