CD146+CAFs mediate immunosuppression in gastric cancer via COL4A1: potential therapeutic targets Authors: Jia Chen, Peng Xu, Qingyuan Wang, Juncheng Dai, Shuang Liang, Sheng Yang, Guiping Xie, Xiang Li, Yaohui Wang, Hongjin Hua, Xiaochun Ping, Jiajia Shen, Lizong Shen Published: 2025 ABSTRACT AbstractBackgroundThe immunomodulatory roles of cancer-associated fibroblasts (CAFs) in gastric cancer have been increasingly recognized. This study aimed to identify specific CAF subpopulations and to elucidate the mechanisms underlying their immunosuppressive effects.MethodsWe developed a single-cell RNA sequencing (scRNA-seq) protocol to enrich CAFs and matched normal fibroblasts (NFs) from stomach tissues. Bioinformatics analyses, along with in vitro and in vivo experiments, were performed to functionally validate a distinct CAF subset.ResultsWe profiled 8,322 high-quality fibroblasts from nine patient-matched tumor and adjacent normal tissue pairs. A unique CAF subset, CD146+CAFs, was identified. CD146+CAFs were significantly enriched in the tumor microenvironment and associated with poor survival and M2-like macrophage infiltration. Tumor cells promoted CD146+CAFs expansion via JAG1-NOTCH3 and LGALS3-CD146 signaling. Mechanistically, NOTCH3 acted as a master regulator in CD146+CAFs. Tumor cell-derived JAG1 engaged NOTCH3 on CD146+CAFs, which triggered PI3K/AKT-mediated CD146 transcription and enhanced COL4A1 secretion. Functionally, CD146+CAF-derived COL4A1 bound macrophage CD44/IL7R to drive macrophage M2-like polarization, thereby suppressing CD8+ T-cell function. Therapeutic blockade of CD146 or COL4A1 in vivo impaired M2 polarization, reinvigorated CD8+ T-cell activity, and restrained tumor growth.ConclusionsOur findings underscore the pivotal role of CD146+CAFs in shaping an immunosuppressive microenvironment and highlight CD146+CAFs and COL4A1 as promising candidates for targeted therapy in gastric cancer.Keywords: Gastric Cancer, Immunosuppression WHAT IS ALREADY KNOWN ON THIS TOPIC Immunotherapy for gastric cancer remains of limited efficacy and falls short of clinical expectations. Immunotherapy for gastric cancer remains of limited efficacy and falls short of clinical expectations. The immunomodulatory roles of cancer-associated fibroblasts (CAFs) in gastric cancer have been increasingly recognized. The immunomodulatory roles of cancer-associated fibroblasts (CAFs) in gastric cancer have been increasingly recognized. WHAT THIS STUDY ADDS A unique subset of CAFs, CD146+CAFs, is identified; they are significantly enriched in the gastric cancer microenvironment, correlate with poor survival and M2-like macrophage infiltration. A unique subset of CAFs, CD146+CAFs, is identified; they are significantly enriched in the gastric cancer microenvironment, correlate with poor survival and M2-like macrophage infiltration. CD146+CAF expansion is driven by juxtacrine signaling involving tumor-derived JAG1 and LGALS3. CD146+CAF expansion is driven by juxtacrine signaling involving tumor-derived JAG1 and LGALS3. CD146+CAFs secrete COL4A1 via PI3K/AKT signaling. CD146+CAFs secrete COL4A1 via PI3K/AKT signaling. CD146+CAFs polarize macrophages towards an M2-like phenotype via COL4A1/CD44/IL7R pathway, resulting in immunosuppression. CD146+CAFs polarize macrophages towards an M2-like phenotype via COL4A1/CD44/IL7R pathway, resulting in immunosuppression. Therapeutic blockade of CD146 or COL4A1 in vivo impairs M2 polarization, reinvigorates CD8+ T-cell activity, and restrains tumor growth. Therapeutic blockade of CD146 or COL4A1 in vivo impairs M2 polarization, reinvigorates CD8+ T-cell activity, and restrains tumor growth. HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY These findings advance the understanding of immunosuppression in gastric cancer and nominate CD146+CAFs and COL4A1 as promising targets for future immunotherapy, potentially improving patient outcomes. These findings advance the understanding of immunosuppression in gastric cancer and nominate CD146+CAFs and COL4A1 as promising targets for future immunotherapy, potentially improving patient outcomes. Background Immunotherapy has emerged as a promising therapeutic modality for several solid cancer types, including breast, renal, and lung cancers, as well as melanoma.1 However, its efficacy in gastric cancer remains limited and falls short of clinical expectations.2 3 For instance, phase 3 trials such as KEYNOTE-061, KEYNOTE-062, and JAVELIN Gastric 300 failed to demonstrate superiority of programmed cell death protein-1 (PD-1) or programmed death-ligand 1 (PD-L1) monotherapy over standard chemotherapy in patients with advanced gastric cancer.4 Moreover, maintenance therapy with PD-L1 inhibitors after first-line chemotherapy has not conferred significant survival benefits.5 Although combination regimens of PD-1 blockade and chemotherapy have improved outcomes in selected patients, these benefits are restricted to subgroups defined by stringent eligibility criteria.6 Immunosuppression within the tumor microenvironment (TME) remains a major barrier to effective immunotherapy in gastric cancer,7 underscoring the urgent need to elucidate its underlying mechanisms.8 Cancer-associated fibroblasts (CAFs) are the dominant stromal population in most solid tumors and orchestrate tumor proliferation, invasion, metastasis, angiogenesis, and immune escape.9 10 Over the past decade, their immunomodulatory roles have gained increasing recognition.11 12 CAFs foster an immunosuppressive TME by inhibiting T-cell recruitment and activation, erecting physical barriers, expanding regulatory T cells (Tregs), skewing macrophages (MΦ) and dendritic cells toward a tolerogenic state, and directly driving CD8+ T-cell exhaustion.13,15 In gastric cancer, Li et al identified a niche composed of SPP1+ tumor-associated macrophages (TAMs) and THBS2+ myofibroblastic CAFs (myCAFs) that confers resistance to anti-PD-1 therapy in peritoneal metastasis,16 underscoring the need to decipher CAF-centric intercellular networks. Recently, we profiled 15 paired gastric tumors and adjacent normal mucosae by single-cell RNA sequencing (scRNA-seq) and further implicated that CAFs and TAMs, together with exhausted CD8+ T cells and Tregs, are involved in shaping the immunosuppressive landscape.4 Despite these insights, the precise molecular circuits through which CAFs disable antitumor immunity in gastrointestinal cancers remain incompletely defined. Dissecting these CAF-centered interactions may uncover actionable targets to overcome immunotherapy resistance. We previously demonstrated that gastric cancer patient-derived CAFs drive tumor progression, chemoresistance and immunosuppression via extracellular matrix (ECM) remodeling.17,19 However, the spectrum and dynamics of CAF subpopulations in gastric tumors relative to normal mucosa remained undefined. Implementing the CAF-enrichment workflow of Kieffer et al and Belle et al,20 21 we profiled 8,322 fibroblasts from nine paired gastric tumors and adjacent normal tissues. We identified and validated a specific subpopulation, CD146+CAFs, which was markedly enriched in tumors and was associated with poor survival and M2-like macrophage infiltration. CD146+CAFs were expanded through juxtacrine signaling involving tumor-derived jagged-1 (JAG1) and galectin-3 (LGALS3). Functionally, CD146+CAFs secreted collagen IV α1 (COL4A1) skews TAMs toward an M2-like phenotype, thereby promoting immunosuppression. In vivo, neutralizing antibodies against CD146 or COL4A1 restrained tumor growth and restored CD8+ T-cell cytotoxicity. Our study highlights the pivotal role of this specific subpopulation of CAFs in the immunosuppressive microenvironment of gastric cancer and provides evidence for exploring CD146+CAFs and COL4A1 as new potential therapeutic targets for gastric cancer. Methods Human tissue specimens of patients with gastric cancer Gastric cancer samples for scRNA-seq and multiplex-immunofluorescence (MxIF) assays were obtained from patients who underwent radical gastrectomy at the First Affiliated Hospital of Nanjing Medical University between March 2024 and July 2024. All patients were pathologically diagnosed with primary locally advanced gastric adenocarcinoma and did not receive preoperative chemotherapy or radiotherapy. The samples were obtained after obtaining written consent according to an established protocol. This study was conducted in accordance with the Declaration of Helsinki. Single-cell dissociation, fibroblast enrichment and scRNA-seq for fibroblasts Tumor tissues and corresponding non-cancerous mucosal tissues (at least 5 cm from the outer tumor margin) were collected immediately after resection and preserved in MACS Tissue Storage Solution (Miltenyi Biotec, Bergisch Gladbach, Germany). After washing with phosphate-buffered saline, the tissues were processed into single-cell suspensions using a Tumor Dissociation Kit (Miltenyi Biotec, Bergisch Gladbach, Germany). Referring to the methods of Kieffer et al and Belle et al,20 21 antibodies against CD45 (RRID:AB_2876779), CD31 (RRID:AB_1877152), and EPCAM (RRID:AB_756079) (BioLegend, San Diego, California, USA) were used to deplete immune, endothelial, and epithelial cells, respectively, thereby enriching the fibroblasts. Live DAPI−CD45−CD31−EPCAM− cells were sorted using a BD FACSMelody Cell Sorter (BD Biosciences, San Jose, California, USA; RRID:SCR_023209). Finally, single cells were stained for viability assessment using Calcein-AM (Thermo Fisher Scientific, Waltham, Massachusetts, USA) and Draq7 (BD Biosciences, San Jose, California, USA). scRNA-seq was conducted using a BD Rhapsody Single-Cell Analysis System (BD Biosciences, San Jose, California, USA; RRID:SCR_027096). The “Seurat” R package (V.4.3.1, RRID:SCR_016341) was used for scRNA-seq analysis.22 Before data analysis, each sample underwent quality control with several filtration steps: (1) excluding cells with fewer than 200 unique detected genes; (2) removing genes expressed in fewer than three cells; (3) filtering out cells with mitochondrial transcript levels exceeding 10–50%; (4) retaining cells with 300≤nFeature_RNA<3000 and 500≤nCount_RNA≤sample-specific 90th percentile; and (5) identifying potential doublets with DoubletFinder (V.2.0.3) and removing them separately for each sample. Next, the gene expression data were preprocessed using Seurat functions: “FindVariableFeatures” to identify the top 2,000 variable genes, followed by normalization and scaling using “NormalizeData” and “ScaleData” with default parameters. To eliminate batch effects across samples, the “RunHarmony” function from the “Harmony” R package (V.0.1.1) was applied. Dimensionality reduction was performed using principal component analysis via the “RunPCA” function. The optimal number of principal components (PCs) for further analysis was determined by examining the “elbow” point generated by the “ElbowPlot” function. A graph-based clustering method was employed to cluster the cells, including “FindNeighbors” and “FindClusters” functions. Finally, Uniform Manifold Approximation and Projection (UMAP) was applied to visualize these clusters. After clustering, differentially expressed genes (DEGs) were identified within each cluster using the Wilcoxon rank-sum test, implemented via the “FindAllMarkers” function. DEGs were visualized using UMAP plots generated using the “FeaturePlot” function and dot plots generated using the “DotPlot” function. Differential abundance analysis was performed using miloR (V.1.8.1, RRID:SCR_025512)23 by modeling cell counts within neighborhoods of a k-nearest neighbor (KNN) graph. A KNN graph was constructed from 15 PCs using BuildGraph (k=15, d=15), neighborhoods were defined with makeNhoods, and a negative binomial generalized linear model with trimmed mean of m-values (TMM) normalization was fitted to each neighborhood, with batch included as a covariate in testNhoods. This approach yielded log2-fold changes in cell abundance between the two conditions for each neighborhood. scDist (V.1.1.4)24 was used to identify perturbed cell types across conditions, with results visualized using DistPlot. Additionally, SCISSOR (V.2.0.0, RRID:SCR_003146)25 analysis was used to identify prognostic phenotype-associated cell subpopulations by integrating our scRNA-seq data with bulk transcriptomic profiles and survival data from the Asian Cancer Research Group (ACRG; GSE62254, https://www.ncbi.nlm.nih.gov/geo/; RRID:SCR_004001) and GSE84437 cohorts. Monocle2 (V.2.28.0, RRID:SCR_016339)26 was used to elucidate the pseudotemporal differentiation trajectories of fibroblast subpopulations. First, fibroblast RNA-seq count data were aggregated into a CellDataSet object using the “newCellDataSet” function, retaining genes expressed in more than 10 cells. DEGs for trajectory analysis were identified using “differentialGeneTest”. Dimensionality reduction and cell ordering were performed with “reduceDimension” and “orderCells” functions. Cell lineage and pseudotime were independently inferred using Slingshot (V.2.8, RRID:SCR_017012)27 for validation: the Seurat object was converted to a SingleCellExperiment, and unsupervised trajectory analysis was run on UMAP embeddings with Seurat clustering labels as input. Data acquisition and processing The data of the patients with gastric cancer were obtained from The Cancer Genome Atlas (TCGA; https://tcga-data.nci.nih.gov/tcga/; RRID:SCR_003193) and ACRG databases. The raw scRNA-seq FASTQ files were retrieved from the Genome Sequence Archive (GSA) database (https://bigd.big.ac.cn/gsa; RRID:SCR_025826), with accession numbers CRA002586, HRA000704, and HRA003647. The gene expression levels were aligned to the GRCh38 human reference genome using Cell Ranger (V.3.1.0; RRID:SCR_017344) with default parameters. Consequently, an integrated scRNA-seq database comprising 22 paired gastric cancer tissues and corresponding normal gastric mucosa tissues was established. Cell–cell communication analyses The “CellChat” R package (V.1.6.1; RRID:SCR_021946) was used to explore the interactions between CAFs and other cell types within TME. First, the communication probability was estimated, and the communication network was constructed using the “computeCommunProb” and “filterCommunication” functions. Next, the “computeCommunProbPathway” function was employed to calculate the communication probability at the signaling pathway level by aggregating all relevant ligand-receptor pairs. Lastly, the overall cell–cell communication network was established using the “aggregateNet” function. Gene regulatory network analyses To compute regulon activity scores in CAFs, Single-Cell regulatory Network Inference and Clustering (SCENIC) analyses were performed using the pySCENIC package (V.0.10.0) in Python (V.3.8; RRID:SCR_025802).28 First, GRNBoost2 (RRID:SCR_025614) was employed to deduce co-expression modules linking transcription factors (TFs) with candidate target genes. Next, RcisTarget (RRID:SCR_024860) was applied to scrutinize genes within each module, aiming to uncover enriched motifs. A TF and its potential direct targets were also collectively defined as a regulon. Lastly, AUCell (RRID:SCR_021327) was used to assess the activity levels of each regulon across individual cells. The cell-type specificity of a regulon was assessed by calculating its regulon specificity score, and ggplot2 (RRID:SCR_014601) was used for the corresponding visualization.29 Deconvolution of CAF subtypes Deconvolution of CAF subtypes was performed using BayesPrism (V.2.2.2, RRID:SCR_027499). A reference expression matrix was constructed from our scRNA-seq dataset of gastric fibroblasts, with CAF subtypes defined by cluster-specific markers. This reference was applied to bulk RNA-seq data from the TCGA, ACRG, GSE15459, and GSE84437 cohorts to estimate relative proportions of CAF subtypes per sample. Following developer recommendations, mitochondrial and ribosomal genes were excluded, and the analysis was restricted to protein-coding genes before running deconvolution with the “new.prism” function. Cells and cell culture The gastric cancer cell lines AGS (RRID:CVCL_0139) and MKN45 (RRID:CVCL_0434), and the THP-1 (RRID:CVCL_0006) cell line, were preserved in our laboratory. All cells were cultured in Dulbecco’s Modified Eagle Medium (Invitrogen, Carlsbad, California, USA) supplemented with 10% fetal bovine serum (FBS; Gibco, Grand Island, New York, USA) and 1% penicillin/streptomycin (Thermo Scientific, Waltham, Massachusetts, USA) in a humidified incubator at 37°C with 5% CO2. Isolation and culture of primary normal fibroblasts (NFs) and CAFs were performed according to our previous report.17 Lentivirus infection Lentiviruses carrying EBF2 short hairpin RNA (shEBF2) and TCF3 short hairpin RNA (shTCF3) were constructed by GeneCopoeia (BioPike, Germantown, Maryland, USA). Lentiviral transduction was performed according to the manufacturer’s instructions. The shRNA sequence targeting EBF2 was GAGGUGACAUUAUCUUAU, and the sequence targeting TCF3 was CCGGCCCGGATCACTCAAGCAATAACTCGAGTTATTGCTTGAGTGATCCGGGTTTTTG. Western blotting assay The expression levels of the indicated protein were determined by western blotting. The antibodies used are listed in online supplemental table S1. Relative protein levels were quantified and normalized to glyceraldehyde-3-phosphate dehydrogenase (GAPDH) levels in the same sample using densitometry analysis. Immunofluorescence assay Cells were fixed with 4% paraformaldehyde and permeabilized with 0.1% Triton X-100. Subsequently, cells were incubated with primary antibodies overnight at 4°C, followed by secondary antibodies for 2 hours at room temperature. Cell nuclei were counterstained with 4',6-diamidino-2-phenylindole (DAPI; Sigma-Aldrich, St. Louis, Missouri, USA). Images were acquired using a Leica SP5 II confocal microscope (Leica Microsystems, Germany; RRID:SCR_018714). Multiplex-immunofluorescence MxIF staining was performed according to the manufacturer’s recommended protocols using the Opal 7-Color Automation Detection IHC kit (PerkinElmer, Waltham, Massachusetts, USA). The antibodies used are listed in online supplemental table S1. Flow cytometry Cells were collected and incubated with specific antibodies for 30 min at 4°C. The results were then analyzed using flow cytometry (FCM) (FACS Calibur, BD Biosciences, San Jose, California, USA; RRID:SCR_026372). The antibodies used are listed in online supplemental table S1. Chromatin immunoprecipitation Chromatin immunoprecipitation (ChIP) assays were performed using a ChIP Assay Kit (Beyotime, Shanghai, China) according to the manufacturer’s recommended protocols. Briefly, cell lysates from CAFs were prepared, and chromatin fragments were fragmented to an average size of 200–1,000 bp using microcapsule nuclease. The lysates were then immunoprecipitated with anti-NOTCH3 or IgG antibodies overnight at 4°C. Cross-links between DNA and protein were reversed at 65°C for 4 hours, and the DNA was subsequently analyzed by quantitative PCR (qPCR). The ChIP-qPCR primers for CD146 were as follows: 5’-CCCTCGCCCATTAACTCT-3’ (Forward) and 5’-TTCAAACGCAAGCTCCTG-3’ (Reverse). The ChIP-qPCR primers for COL4A1 were as follows: 5’-CATCAGGGCAGAACGATGTA-3’ (Forward) and 5’-CTGTCCCTCCTCTCCTTACT-3’ (Reverse). These qPCR products were resolved on 2% agarose gels and visualized using a ultraviolet (UV) transilluminator. Luciferase reporter assay Luciferase reporter vectors containing wild-type or mutant melanoma cell adhesion molecules (MCAM, also known as CD146) fragments and COL4A1 fragments were generated by RiboBio (Guangzhou, China). Cells were seeded in 24-well plates and transfected with these luciferase reporter vectors using Lipofectamine 3000 (Invitrogen, Carlsbad, California, USA). After a 48 hours incubation, the cells were lysed, and luciferase activity was measured according to the manufacturer’s protocol using a dual-luciferase reporter assay kit (Promega, Madison, Wisconsin, USA). Each experiment was repeated at least three times. ELISA Cell supernatants were collected at the indicated times and centrifuged. ELISA assays were performed using a human ELISA kit (eBioscience, Shanghai, China) for COL4A1, interleukin (IL)-10, and transforming growth factor (TGF)-β1, respectively. The assays were conducted according to the manufacturer’s instructions. Each experiment was repeated at least three times. 3D co-culture of fibroblasts and tumor cells A 96-well ultra-low attachment plate (Thermo Fisher Scientific, Waltham, Massachusetts, USA) was used for three-dimensional (3D) spheroid culture. A mixture of 10,000 NFs and gastric cancer cells at a 1:1 ratio was added to the wells. Over 24–48 hours, the cells gradually proliferated and formed mixed spheroids with ultra-low attachment properties at the bottom of the U-shaped wells. After 4 days of co-culture, the phenotypes of CD146+CAFs were analyzed by FCM. CD8+ T-cell isolation and stimulation assay 24-well plates were coated with 5 µg/mL anti-CD3 (Invitrogen, Carlsbad, California, USA) overnight at 4°C. Peripheral blood mononuclear cells (PBMCs) were isolated using the Human Peripheral Blood Mononuclear Cell Isolation Kit (Solarbio, Beijing, China). After washing, CD8+ T cells were isolated using CD8 MicroBeads (Miltenyi Biotec, Bergisch Gladbach, Germany; RRID:AB_2889920) using magnetic bead separation. The supernatant was removed, and 2×10⁵ CD8+ T cells per well were seeded. The cells were then cultured for 48 hours in complete Roswell Park Memorial Institute (RPMI) 1640 medium supplemented with 2 µg/mL anti-CD28 antibody (Invitrogen, Carlsbad, California, USA; RRID:AB_2536491) and 0.25 µg/mL IL-2 (MedChemExpress, Monmouth Junction, New Jersey, USA). Animal study The animal studies were approved by the Animal Management and Use Committee of Nanjing Medical University (SYXK-2020–0022). A humanized immune-reconstructed gastric cancer cell-derived xenograft (CDX) model was established using NCG mice (NOD/ShiLtJGpt-Prkdcem26Cd52Il2rgem26Cd22/Gpt, RRID:IMSR_ARC:NOD). This study primarily used male animals to minimize experimental variability. However, we also conducted experiments with female mice, and similar results were observed for both sexes. AGS alone or a mixture of AGS cells and CD146+CAFs (1:1 ratio) was inoculated subcutaneously (1×10⁶ cells per mouse). On day 7 after tumor inoculation, PBMCs from three healthy donors were injected via the tail vein (5×105 cells per mouse). The mice that received the mixture inoculation were randomly divided into three groups: the anti-COL4A1 group, anti-CD146 group, and control group (nine mice per group to meet statistical needs). The anti-COL4A1 and anti-CD146 groups received intraperitoneal injections of neutralizing antibodies against CD146 or COL4A1 (anti-CD146, RRID:AB_3081922; anti-COL4A1, RRID:AB_3206274; Abcam, Cambridge, UK) (5 mg per kilogram of body weight for each injection) on days 14, 21, and 28 after tumor inoculation, respectively. On day 35 after tumor inoculation, the animals were sacrificed and fresh tumors and peripheral blood were collected for further analysis. The tumors were measured and subjected to pathological analysis. The collected tissues and blood samples were used for additional analyses, including MxIF and cytokine measurement. Statistical analysis Statistical analyses were performed using the GraphPad Prism software (V.8.0; La Jolla, California, USA; RRID:SCR_002798). All values are expressed as mean±SD. Differences between groups were compared using a two-tailed unpaired Student’s t-test for comparisons between two groups or one-way analysis of variance for comparisons of multiple groups. Categorical variables were assessed using the χ² test. Overall survival was analyzed using the Kaplan-Meier method and compared using the Gehan-Breslow-Wilcoxon test (log-rank test). All experiments were repeated at least thrice. A p value of <0.05 was considered statistically significant. Results scRNA-seq identifies a unique CAF subpopulation, CD146+CAFs, selectively enriched in the gastric cancer microenvironment Using our established scRNA-seq protocol for enriching CAFs and NFs, we profiled 8,322 fibroblasts from nine paired gastric cancer tissues and matched adjacent normal gastric mucosa (figure 1A). Of these, 3,458 cells were derived from normal tissues and 4,864 cells from tumor tissues (figure 1B,C, online supplemental figure S1A). After integration and stringent quality control, unsupervised clustering resolved eight transcriptionally distinct fibroblast subsets (figure 1B,C, online supplemental figure S1B,C): CD146+ myCAFs, POSTN+ myCAFs, SOCS3+ inflammatory CAFs (iCAFs), CFD+ iCAFs, CFH+ iCAFs, COL12A1+ myCAFs, PDGFA+ myCAFs, and CD74+ antigen-presenting CAFs. Differential distribution analysis revealed that CD146+CAFs (CD146+ myCAFs) were significantly enriched in tumors compared with matched normal mucosa (p=0.026, figure 1C,D, online supplemental figure S1D); the remaining subsets showed no significant difference between normal and tumor tissues (figure 1C, online supplemental figure S1E). Complementary analyses with miloR (figure 1E) and scDist (figure 1F) collectively supported both the selective expansion and altered activation status of CD146+CAFs in tumor tissue. CD146+CAFs were almost undetectable in the normal mucosa, indicating a tumor-specific subpopulation. Differential gene-expression analysis (CAFs vs NFs) demonstrated upregulation of NOTCH3, MCAM (CD146), ACTA2 (α-SMA), COL12A1, PDGFA, and CD74, whereas POSTN, SOCS3, and CFD were downregulated (figure 1G). All eight subsets retained canonical fibroblast markers (COL1A1, COL1A2, COL3A1; online supplemental figure S1F), yet each subset displayed a unique transcriptomic signature (figure 1H–I, online supplemental figure S1G,H). CD146+CAFs were characterized by high expression of NOTCH3, MCAM (CD146), and COL4A1 (figure 1J), underscoring their specialized roles within the gastric cancer microenvironment. scRNA-seq reveals a distinct CD146+CAF subpopulation in gastric cancer microenvironment. (A) Workflow schematic. Human gastric cancer tissues and matched normal tissues were dissociated into single cells. Two fractions were collected by FACS per sample: (1) all viable cells (DAPI−); (2) fibroblast-enriched cells (DAPI−, CD45−, CD31−, EPCAM−). Sorted cells underwent single-cell capture, barcoding, and reverse transcription using the BD Rhapsody platform. (B) UMAP visualization of unsupervised clustering of fibroblasts from nine paired tumor and normal tissues, identifying eight distinct subpopulations. (C, D) Quantification of CD146+CAF enrichment in tumor microenvironment (p=0.026). (E) Milo analysis quantifying selective expansion of CD146+CAF neighborhoods in tumor tissue. (F) scDist analysis demonstrating that CD146+CAFs exhibited the most extensive phenotypic remodeling compared with other CAF subtypes. (G) Violin plots comparing expression of key marker genes between tumor CAFs and NFs. (H) Heatmap of differential gene expression across CAF subtypes. (I) Bubble plot illustrating specific gene markers for each CAF subpopulation. Dot size represents the fraction of expressing cells; color intensity indicates mean expression level. (J) Feature plots showing expression of NOTCH3, MCAM (CD146), and COL4A1 in the CD146+CAF subpopulation. CAF, cancer-associated fibroblast; COL4A1, collagen IV α1; FACS, fluorescence-activated cell sorting; MCAM, melanoma cell adhesion molecule; NFs, normal fibroblasts; scRNA-seq, single-cell RNA sequencing; UMAP, Uniform Manifold Approximation and Projection. CD146+CAFs have previously been described in colorectal cancer and endometrial cancer.30 31 CD146 is also a well-known marker of endothelial cells and pericytes.32 33 To exclude contamination by these lineages, we analyzed the expression of canonical endothelial (PECAM1, VWF, CDH5) and pericyte (RGS5, DES) genes in our CD146+CAF subset. None of these lineage-specific transcripts were detected (online supplemental figure S1I), confirming the fibroblastic identity of the sorted cells and supporting the designation of CD146+CAFs. Pseudotime trajectory inference using Monocle2 (figure 2A–D) positioned SOCS3+iCAFs at the initiation of the inferred differentiation path. Markers of this state (BTG2, ATF3, SOCS3, ATP1B1) decreased along pseudotime, whereas CD146+CAFs and CFD+iCAFs were positioned at the trajectory terminus, characterized by upregulated NOTCH3, MCAM, COL4A1, ACTA2, and ARPC5. Branching analysis with Slingshot (figure 2E) independently supported CD146+CAFs derived from a pre-branch SOCS3+CAFs like progenitor. Finally, Kyoto Encyclopedia of Genes and Genomes and Gene Ontology analyses of CD146+CAFs (figure 2F) showed enrichment for Notch, PI3K-AKT, and ECM-receptor interaction pathways. Single-cell trajectory and functional enrichment analyses of gastric cancer CAFs. (A) Pseudotime plot showing CAF development trajectory in gastric cancer. (B) Trajectory distribution of individual CAF subpopulations. (C) Heatmap displaying scaled expression levels of branch-specific differentially expressed genes, grouped into three major clusters with pseudotime-dependent expression patterns. Representative genes are listed on the left. (D) Expression dynamics of key marker genes along the pseudotime axis. Each point represented a single cell, colored by CAF subpopulation. Solid black lines denoted pseudotime expression kinetics. (E) Branching analysis using Slingshot independently supported CD146+CAFs derived from a pre-branch SOCS3+iCAF like progenitor. (F) KEGG and GO analyses of different CAF subpopulations, revealing enrichment of Notch signaling, ECM-receptor interaction, and PI3K-AKT signaling pathways specifically in CD146+CAFs. apCAF, antigen-presenting CAF; CAF, cancer-associated fibroblast; GO, Gene Ontology; iCAF, inflammatory CAF; KEGG, Kyoto Encyclopedia of Genes and Genomes; myCAF, myofibroblastic CAF; UMAP, Uniform Manifold Approximation and Projection. We next validated CD146+CAFs in primary CAFs and in human gastric cancer tissues. FCM revealed that the prevalence of NOTCH3+CD146+CAFs was 7.31%–22.2% in primary CAFs derived from patients with advanced gastric cancer (figure 3A, online supplemental figure S2A), confirming their substantial presence in the TME. Fluorescence-activated cell sorting (FACS)-isolated CD146+CAFs from primary CAFs (figure 3B) contained only minor fractions of NOTCH3+CD146−CAFs and NOTCH3−CD146+CAFs (<5%, figure 3C, online supplemental figure S2B), underscoring the specificity of CD146+CAF phenotype. Immunofluorescence assay confirmed the co-expression of NOTCH3 and CD146 in isolated CD146+CAFs (figure 3D). MxIF assays demonstrated the significant enrichment of CD146+CAFs within the human gastric cancer microenvironment (figure 3E, online supplemental figure S2C). These cells were phenotypically distinct from CD31+/VEC+ endothelial cells (online supplemental figure S2D). CD146+CAFs are significantly enriched in gastric cancer microenvironment. (A) FCM analysis showing the prevalence of NOTCH3+CD146+CAFs ranged from 7.31% to 22.2% in primary CAF populations. (B) Isolation of CD146+CAFs from primary CAFs. (C) FCM analysis showing that NOTCH3+CD146 or NOTCH3-CD146+ −CD146 constituted <5% of total CAFs. (D) IF confirmed co-expression of NOTCH3 and CD146 in the isolated CD146+CAFs. (E) MxIF demonstrated enrichment of CD146+CAFs in gastric cancer tissue compared with adjacent normal tissue (α-SMA marks CAFs; CEA marks tumor cells). CAF, cancer-associated fibroblasts; FCM, flow cytometry; IF, immunofluorescence; MxIF, multiplex-immunofluorescence. In summary, our scRNA-seq study identified CD146+CAFs, characterized by co-expression of NOTCH3 and CD146, as a distinct subset that is markedly enriched in the gastric cancer microenvironment. This finding refines the molecular taxonomy of CAFs and positions CD146+CAFs as potential drivers of tumor-stromal crosstalk, providing a rationale for potential therapeutic targeting. CD146+CAFs are associated with poor survival and M2-like TAM infiltration in gastric cancer Subsequently, we evaluated the clinical relevance of CD146+CAFs using multiple stomach adenocarcinoma (STAD) cohorts. In TCGA-STAD, NOTCH3 expression strongly correlated with MCAM (CD146) (r=0.55, p=3.88×10−36) and COL4A1 (r=0.70, p=4.58×10−68); CD146 and COL4A1 were even more tightly linked (r=0.76, p=3.62×10−86) (figure 4A). Elevated expression of CD146 (p=0.004), NOTCH3 (p=0.003), or COL4A1 (p<0.001) was associated with shorter overall survival (figure 4B). Patients were stratified into high-abundance and low-abundance CD146+CAF groups by optimal cutoff. The high-abundance group exhibited elevated stromal scores (p<0.05) but reduced immune scores (TCGA-STAD, p<0.001, figure 4C). CD146+CAF abundance was also correlated with tumor T stage (GSE84437, figure 4D) and Lauren classification (GSE15459, figure 4E). High CD146+CAF abundance was significantly associated with worse overall survival across multiple cohorts (TCGA, ACRG, GSE15459, GSE84437; p<0.05, figure 4F), corroborated by SCISSOR analysis (figure 4G). Moreover, high CD146+CAF abundance was associated with enrichment of M2-like TAMs in TCGA (p<0.0001, figure 4H) and ACRG (p=0.011, figure 4I) cohorts. M2-like TAMs are recognized as key contributors to tumor immunosuppression across cancers and portend poorer prognosis in patients with gastric cancer.34 35 Clinicopathological significance of CD146+CAFs in gastric cancer. (A) In TCGA-STAD, NOTCH3 expression was significantly correlated with MCAM (CD146) and COL4A1 expression; CD146 and COL4A1 also showed significant mutual correlation. (B) Higher expression levels of NOTCH3, CD146 and COL4A1 were significantly associated with poorer overall survival in patients with gastric cancer in TCGA-STAD. (C) Higher abundance of CD146+CAFs was positively correlated with stromal scores and negatively correlated with immune scores in TCGA-STAD. (D, E) CD146+CAF abundance correlated with tumor T stage (GSE84437, (D) and Lauren classification (GSE15459, (E) of gastric cancer. (F) High CD146+CAF abundance was significantly associated with worse overall survival in multiple cohorts (TCGA, ACRG, GSE15459, GSE84437; all p<0.05). (G) SCISSOR analysis corroborated this survival association. (H, I) High CD146+CAF abundance was associated with enrichment of M2-like TAMs in TCGA (H) and ACRG (I) cohorts. *p<0.05; **p<0.01; ***p<0.001. ACRG, Asian Cancer Research Group; CAF, cancer-associated fibroblast; COL4A1, collagen IV α1; MCAM, melanoma cell adhesion molecule; STAD, stomach adenocarcinoma; TAM, tumor-associated macrophage; TCGA, The Cancer Genome Atlas. To explore intercellular communication, we analyzed the integrated scRNA-seq dataset (22 paired gastric cancer tissues and adjacent normal mucosa) using CellChat. This revealed that CD146+CAFs communicate predominantly with macrophages (MΦ), T cells, B cells, and natural killer cells through COL4A1–CD44 ECM-receptor interactions, with macrophages being the dominant partners (online supplemental figure S3). Collectively, these bioinformatics analyses indicate that CD146+CAFs are robustly linked to adverse prognosis and immunosuppression in gastric cancer, potentially through their crosstalk with macrophages. Gastric cancer cells foster CD146+CAF enrichment via JAG1-NOTCH3 and LGALS3-CD146 signaling Next, we explored the mechanisms underlying CD146+CAF enrichment in the gastric cancer microenvironment. It is well recognized that both NOTCH3 and CD146 are cell receptors, and they interact with their distinct ligands. We hypothesized that gastric cancer cells supply specific ligands to activate NOTCH3 and CD146 on CD146+CAFs. Using a 3D sphere culture system, we observed that human gastric cancer cell lines AGS and MKN45 significantly increased proliferation of CD146+CAFs derived from primary NFs (p<0.001). This effect was abolished by MRK003, a γ-secretase inhibitor that blocks Notch activation, confirming involvement of the NOTCH pathway (p<0.001, figure 5A,B). Conditioned media (CM) from tumor cells also markedly enhanced the proliferation of isolated CD146+CAFs in colony formation (p<0.001, figure 5C,D), cell counting kit-8 assays (p<0.001, figure 5E), and Ki67 quantification (p<0.001, figure 5F,G), which could be reversed by AA98, an inhibitor of CD146, which blocks CD146 dimerization and inhibits its activation (p<0.001). Analysis of the integrated scRNA-seq dataset comprising 22 patients with gastric cancer identified JAG1 as the most likely NOTCH3 ligand and LGALS3 as the CD146 ligand (online supplemental figure S4A). Consistently, anti-JAG1 antibody significantly attenuated tumor cell-induced proliferation of CD146+CAFs (p<0.001, figure 5A,B), while the LGALS3 inhibitor TD139 reversed CM-induced proliferation (p<0.001, figure 5C–G). Additionally, we quantified Annexin-V staining in CD146+CAFs after exposure to tumor cell CM and found that tumor cell CM significantly decreased the Annexin-V+ apoptotic population (p<0.001, figure 5H,I), an effect that could be reversed by AA98 or TD139 treatment. Gastric cancer cells promote CD146+CAF proliferation and survival. (A, B) Co-culture with tumor cells significantly enhanced CD146+CAF proliferation, which could be reversed by treatment with MRK003 (10 µM) or anti-JAG1 antibody (10 µg/mL). (C-G) Treatment with CMs from tumor cells (24 hours) significantly promoted CD146+CAF proliferation, as assessed by colony formation assay (C, D), CCK-8 assay (E), and Ki67 quantification (F, G). (H, I) Treatment with CMs from tumor cells (24 hours) significantly decreased Annexin-V+ apoptotic population. These effects could be reversed by AA98 (150 µg/mL) and TD139 (10 µM). *p<0.05; **p<0.01; ***p<0.001. CAF, cancer-associated fibroblast; CCK-8, cell counting kit-8; CM, conditioned media. These results indicate that gastric cancer cells promote the proliferation and enrichment of CD146+CAFs through JAG1-mediated NOTCH3 activation and LGALS3-mediated CD146 signaling. CD146+CAFs produce COL4A1 via PI3K/AKT signaling Based on our CellChat analyses predicting that this distinct CAF subset potentially functions through COL4A1, we investigated how CD146+CAFs regulate COL4A1 expression. First, we probed the association between NOTCH3, CD146, and COL4A1 expression. CD146+CAFs (CAF1–3) were isolated from three patients with advanced gastric cancer. Recombinant JAG1 (rJAG1) treatment simultaneously upregulated N3ICD (the intracellular domain of NOTCH3), CD146 and COL4A1, whereas MRK003 abolished these increases (figure 6A,B). Furthermore, treatment with recombinant LGALS3 (rLGALS3) elevated CD146 and COL4A1 levels without affecting N3ICD expression; both effects were reversed by AA98 (figure 6C,D). Conversely, AA98 also blocked rJAG1-induced upregulation of CD146 and COL4A1 but left N3ICD unchanged (figure 6E), indicating that NOTCH3 acts upstream of CD146, which in turn modulates COL4A1. CD146+CAFs produce COL4A1 via PI3K–AKT pathway. CAF1–3 represented three isolates of CD146+CAFs from three primary CAFs derived from three patients with advanced gastric cancer. (A, B) Treatment with rJAG1 (500 ng/mL) upregulated N3ICD, CD146, and COL4A1, which could be reversed by co-treatment with MRK003 (10 µM). (C, D) Treatment with rLGALS3 (5 µg/mL) increased CD146 and COL4A1 levels, which could be reversed by AA98 (150 µg/mL). (E) AA98 (150 µg/mL) inhibited rJAG1-induced CD146 and COL4A1 upregulation without affecting N3ICD levels. (F~H) ChIP and luciferase reporter assays showed that NOTCH3 binds directly to the CD146 promoter region (−477 to −482 bp) via CSL to drive its transcription. (I-K) SCENIC analysis and JASPAR prediction identified EBF2 as a co-TF for CD146 (I), confirmed by ChIP (J) and luciferase reporter assays (K). (L) rLGALS3 activated PI3K/AKT signaling, leading to increased EBF2 and CD146 expression. (M, N) Pharmacological modulation of AKT activity altered EBF2, CD146, and COL4A1 levels accordingly. (O) shRNA-mediated EBF2 knockdown (shEBF2) reduced CD146 expression. (P, Q) ChIP and luciferase assays identified TCF3 as the transcription factor for COL4A1. (R) shRNA-mediated TCF3 knockdown (shTCF3) reduced COL4A1 expression. *p<0.05; **p<0.01; ***p<0.001. CAF, cancer-associated fibroblast; ChIP, chromatin immunoprecipitation; CSL, CBF-1/Suppressor of Hairless/Lag; rJAG1, recombinant jagged-1; rLGALS3, recombinant galectin-3; SCENIC, Single-Cell regulatory Network Inference and Clustering; TF, transcription factor. In canonical Notch signaling, ligand binding liberates N3ICD, which translocates to the nucleus, displaces the CSL (CBF-1/Suppressor of Hairless/Lag-1) co-repressor, recruits co-transcriptional activators (co-TFs), and activates target-gene transcription.36 ChIP assays revealed that N3ICD binds the CD146 promoter at −477 to −482 bp via CSL; luciferase assays confirmed this interaction (figure 6F–H), which indicated that NOTCH3 drives CD146 transcription by binding to its core promoter. SCENIC analysis and JASPAR prediction (figure 6I, online supplemental figure S4B–E) identified EBF2 as the key co-TF in CD146+CAFs, verified by ChIP (figure 6J) and luciferase assays (figure 6K). rLGALS3 activated PI3K/AKT, upregulated CD146 and increased EBF2 (figure 6L). Pharmacologic activation (SC79, an AKT activator) or inhibition (MK-2206, an AKT inhibitor) of AKT correspondingly altered EBF2 and CD146 levels (figure 6M,N), whereas shEBF2 markedly reduced CD146 levels (figure 6O). These findings established a positive feedback loop involving CD146/PI3K/AKT/EBF2/CD146 independently of NOTCH3, which further promoted CD146 transcription. SCENIC and JASPAR further predicted TCF3 as the principal transcription factor for COL4A1 (online supplemental figure S4D, F–H), confirmed by ChIP (figure 6P) and luciferase assays (figure 6Q). As shown in figure 6L–N, PI3K/AKT phosphorylation activated TCF3 and thereby promoted COL4A1 expression, whereas shTCF3 markedly decreased both TCF3 and COL4A1 levels (figure 6R). Collectively, NOTCH3 serves as a “driver” in CD146+CAFs. Tumor cell-derived JAG1 activates NOTCH3, initiating the PI3K/AKT/EBF2 axis to amplify CD146 transcription and the PI3K/AKT/TCF3 axis to enhance COL4A1 production, thereby reinforcing the protumoral phenotype of CD146+CAFs. CD146+CAFs polarize macrophages towards an M2-like phenotype via COL4A1/CD44/IL7R pathway, resulting in immunosuppression The aforementioned bioinformatic analyses indicated that CD146+CAFs may exert immunosuppressive effects by interacting with MΦ through COL4A1-CD44 signaling. To test this hypothesis, we re-clustered TAMs from the integrated scRNA-seq database of 22 patients with gastric cancer and identified three subpopulations. The M2-like subset (MRC1(CD206)+Macro) expressed CD163, C1QB, C1QA, and C1QC (figure 7A,B). THP-1 monocytes were differentiated into macrophages with phorbol myristate acetate and then treated with recombinant COL4A1 (rCOL4A1) or co-culture with CD146+CAFs. FCM showed a marked increase in M2-like macrophages (p<0.001), accompanied by elevated IL-10 and TGF-β1 in the supernatant (p<0.001); both effects were abolished by a monoclonal antibody against CD44 (anti-CD44) (p<0.001, figure 7C,D). However, VEGF, ARG1, and NOS2 remained unchanged in these induced M2-like macrophages (online supplemental figure S5A–C). MxIF assay further revealed co-localization of COL4A1 and CD206 in the tumor stroma (figure 7E). TCGA-STAD data corroborated a positive correlation between COL4A1 and CD206 (r=0.43, p=9.67×10⁻²², figure 7F). These results implicate COL4A1 as a pivotal mediator reshaping the gastric cancer immune microenvironment by driving macrophage polarization. CD146+CAFs polarize macrophages towards an M2-like phenotype via COL4A1/CD44/IL7R pathway. (A) UMAP plots showed different cell types in the integrated scRNA-seq dataset (n=22 patients). Macrophages were subdivided into three subpopulations. (B) Feature plot showed that M2-like MΦ (MRC1(CD206)+Macro) expressed CD163, C1QA, C1QB, and C1QC. (C, D) rCOL4A1 (10 µg/mL) or co-culture with CD146+CAFs increased the proportion of M2-like macrophages (C) and IL-10 and TGF-β1 levels (D) in THP-1-derived macrophages, which could be reversed by anti-CD44. (E) MxIF assay showed co-localization of M2-like macrophages and COL4A1 in the gastric cancer stroma. (F) TCGA-STAD showed a significant positive correlation between COL4A1 and CD206 expression. (G) RNA-seq revealed IL7R as one of the most significantly downregulated genes on anti-CD44 treatment. (H) GSEA indicated that COL4A1-induced CD44 activation activated the JAK–STAT pathway. (I) Anti-CD44 treatment (10 µg/mL) inhibited COL4A1-induced IL7R expression and JAK1–STAT6 activation. (J) Anti-IL7R treatment (10 nM) decreased the proportion of COL4A1-induced M2-like macrophages. (K) COL4A1-induced M2-like macrophages inhibited CD8+ T-cell proliferation and decreased GZMB, TNF-α, and IFN-γ levels; these effects could be reversed by MK-1966 (10 µg/mL, 24 hours) or SB-431542 (10 µM, 24 hours). **p<0.01; ***p<0.001. CAF, cancer-associated fibroblast; GSEA, Gene Set Enrichment Analysis; GZMB, granzyme B; IFN, interferon; IL, interleukin; MxIF, multiplex-immunofluorescence; rCOL4A1, recombinant collagen IV α1; scRNA-seq, single-cell RNA sequencing; STAD, stomach adenocarcinoma; TCGA, The Cancer Genome Atlas; TGF-β1, transforming growth factor-β1; TNF, tumor necrosis factor; UMAP, Uniform Manifold Approximation and Projection. To dissect downstream events, anti-CD44-treated macrophages were subjected to RNA-seq. IL7R was among the most significantly downregulated genes (figure 7G), and Gene Set Enrichment Analysis indicated that COL4A1-CD44 engagement activated the JAK-STAT pathway (figure 7H). Anti-CD44 suppressed COL4A1-induced IL7R expression and JAK1-STAT6 phosphorylation (figure 7I), which is required for M2-like polarization in macrophages.37 Consistently, neutralizing IL7R with a monoclonal antibody against IL7R (anti-IL7R) markedly reduced the fraction of M2-like macrophages (p<0.001, figure 7J), suggesting a potential role of IL7R in modulating the macrophage phenotype. Thus, CD146+CAFs drive M2-like polarization in macrophages through the COL4A1/CD44/IL7R/JAK1/STAT6 cascade. Finally, co-culture assays demonstrated that COL4A1-induced M2-like macrophages significantly inhibited CD8+ T-cell proliferation and decreased granzyme B, tumor necrosis factor-α, and interferon-γ levels in CD8+ T cells. These effects were reversed by MK-1966 (a neutralizing anti-IL-10 antibody) or SB-431542 (a TGF-βR1 inhibitor) (both from MedChemExpress, Monmouth Junction, New Jersey, USA) (p<0.001, figure 7K, online supplemental figure S5D–H). These data establish a CD146+CAFs–COL4A1–CD44+ macrophage–CD8+ T-cell immunosuppressive axis that facilitates gastric cancer immune escape. Neutralizing antibodies against CD146 or COL4A1 boost CD8+ T-cell function and inhibit tumor growth in vivo To evaluate the impact of CD146+CAFs on tumor growth and the therapeutic potential of blocking either CD146 or COL4A1 in vivo, we subcutaneously co-injected isolated CD146+CAFs and AGS cells (1:1) into NCG mice to establish a CDX model. Human PBMCs from healthy donors were then infused via the tail vein to reconstitute the immune function of NCG mice (figure 8A), allowing for the study of human immune responses in this in vivo model. Mice were randomized to receive intraperitoneal anti-CD146 and anti-COL4A1. No significant difference in tumor size was observed among the three groups of tumors treated with PBMCs from different donors (online supplemental figure S6A). Neutralizing antibodies against CD146 or COL4A1 boost CD8+ T-cell function and inhibit tumor growth in vivo. (A) Schematic of the in vivo experimental design. (B) Representative images of subcutaneous tumors from each group. (C) Subcutaneous tumor growth curves. (D) Serum IL-10 and TGF-β1 levels measured by ELISA. (E) MxIF assays showed increased M2-like macrophages (CD206+) in mice receiving AGS+CD146+CAF co-inoculation versus AGS alone; anti-CD146 (5 mg/kg) or anti-COL4A1 (5 mg/kg) treatment significantly reduced CD206+ cell abundance. **p<0.01; ***p<0.001. CAF, cancer-associated fibroblast; COL4A1, collagen IV α1; IL, interleukin; i.v., intravenous; MxIF, multiplex-immunofluorescence; PBMC, peripheral blood mononuclear cells; TGF-β1, transforming growth factor-β1. Both antibody treatments significantly inhibited tumor growth (p<0.001, figure 8B,C). Co-injection elevated serum IL-10 and TGF-β1 (p<0.001), which were markedly reduced by either anti-CD146 or anti-COL4A1 (p<0.001, figure 8D). Mechanistically, MxIF assays revealed increased M2-like macrophages (CD206+) in the co-injection group (p<0.001), whereas both antibody treatments significantly decreased their abundance (p<0.001, figure 8E, online supplemental figure S6B). Conversely, anti-CD146 or anti-COL4A1 significantly enhanced intratumoral CD8+ T-cell infiltration (p<0.001, online supplemental figure S6C,D). Following these treatments, the expression levels of CD146, NOTCH3, COL4A1, and associated key signaling proteins were significantly modulated (online supplemental figure S6D–F). These data demonstrate that targeting CD146 or COL4A1 reprograms the tumor immune microenvironment by reducing M2-like macrophages and augmenting cytotoxic T-cell activity, supporting their potential as novel therapeutic targets in gastric cancer. Discussion Previously, we demonstrated that CAFs modulate the malignant phenotype of gastric cancer, including immune suppression.17,19 The CAFs also exhibited high heterogeneity. However, these mechanisms underlying the immunomodulatory roles of CAFs in gastric cancer as well as the changes in fibroblast subpopulations between tumor tissues and corresponding normal tissues remain unclear. In the present study, our scRNA-seq, focusing on enriched fibroblasts, clearly revealed the subpopulation changes and heterogeneity of CAFs compared with normal fibroblasts in the normal stomach mucosa. Importantly, we identified a distinct subpopulation of CAFs, CD146+CAFs, that were significantly enriched in the gastric cancer microenvironment. Bioinformatic analyses suggested that this CAF subset is associated with poor prognosis and M2-like TAM infiltration in gastric cancer. We further elucidated the mechanisms underlying the enrichment of this subpopulation in gastric cancer as well as the regulatory interactions among proteins such as NOTCH3, CD146, and COL4A1 within the novel CAFs. Our in-depth investigation demonstrated that CD146+CAFs promote M2-like polarization of macrophages via the COL4A1-CD44-IL7R interactions, ultimately suppressing CD8+ T-cell function. Furthermore, in vivo blockade of CD146 or COL4A1 reversed M2 polarization, reinvigorated CD8+ T-cell activity, and inhibited tumor growth. The key finding of this study is the delineation of CD146+CAFs that co-express NOTCH3 and CD146 within the gastric cancer microenvironment. Members of the NOTCH family play important roles in tumorigenesis and tumor progression, with different NOTCH members exhibiting distinct activities and biological functions.36 38 39 The NOTCH3 signaling pathway regulates the differentiation of synovial fibroblasts in inflammatory arthritis, thereby affecting disease progression.40 Recent research has reported that the NOTCH3 pathway can regulate the biological behavior of the stroma in lung cancer, promoting tumor invasion.41 However, the role of NOTCH3 in the gastric cancer stroma and its impact on gastric cancer progression remains unclear. CD146, also known as MCAM, is a member of the immunoglobulin superfamily. It is involved in a variety of biological processes, including cell development, signal transduction, angiogenesis, tumor metastasis, lymphocyte activation, morphogenesis during development, and tissue regeneration.42 Studies have shown that CD146 promotes the invasion and metastasis of various tumor cells and is significantly associated with poor prognosis in cancer.31 43 44 CD146 also participates in the regulation of immune cell functions such as T cells and macrophages.45 46 While CD146+CAFs have been described in colorectal cancer and endometrial cancer,30 31 the role and mechanisms of CD146+CAFs in gastric cancer have not yet been defined. Our comprehensive research has elucidated the origin, functions, and mechanisms of this specific subpopulation. Intriguingly, we uncovered that the expression of CD146 in CD146+CAFs involves a positive feedback loop involving CD146/PI3K/AKT/CD146 in the gastric cancer microenvironment. This loop further promotes CD146 expression and enhances the immunosuppressive function of these CAFs. Collagen is abundant in the ECM and plays an important role in regulating the TME and biological behavior.47 As a type IV collagen, COL4A1 is involved in intercellular interactions. Studies have reported that upregulation of COL4A1 expression can promote migration, invasion, and proliferation of tumor cells in hepatocellular carcinoma and urothelial carcinoma.48 49 In this study, COL4A1 was identified as a key factor in immune suppression mediated by CD146+CAFs in gastric cancer. Specifically, COL4A1 engages CD44 on TAMs to induce IL7R-JAK1-STAT6-mediated M2-like polarization and subsequent CD8+ T-cell dysfunction. Furthermore, the potential value of targeting CD146+CAFs or COL4A1 as new therapeutic targets for immunotherapy in gastric cancer is preliminarily supported by in vivo experiments. This study has several limitations. First, while we identified CD146+CAFs in the gastric cancer microenvironment and validated their clinicopathological significance—particularly their association with immune suppression—using publicly available databases, these findings require confirmation in large prospective cohorts. Second, the therapeutic potential of targeting CD146+CAFs or COL4A1 warrants further preclinical validation in co-engrafted humanized patient-derived xenograft models. In conclusion, we identified a distinct CAF subpopulation, CD146+CAFs, which is significantly enriched in the gastric cancer microenvironment. These CAFs correlate with poor survival, immune suppression, and response to immunotherapy. We systematically delineated the mechanisms driving their accumulation and clarified the roles of both CD146+CAFs and COL4A1 in establishing an immunosuppressive milieu at multiple molecular and cellular levels. Furthermore, we evaluated the therapeutic potential of targeting CD146+CAFs or COL4A1 in preclinical models. Our findings not only advance the understanding of immunosuppression in gastric cancer but also nominate promising targets for future immunotherapy, potentially improving patient outcomes. Subsequent studies should prioritize clinical validation and explore rational combination regimens to maximize therapeutic efficacy. Supplementary material Acknowledgements The authors are grateful to NovelBio (www.novelbio.com) for supporting scRNA-seq. Footnotes Funding: This study was financially supported by the National Natural Science Foundation of China (82373143 and 82103432). Provenance and peer review: Not commissioned; externally peer reviewed. Patient consent for publication: Not applicable. Ethics approval: This study involves human participants and was approved by the Institutional Review Board of Nanjing Medical University (2023-SRFA-111). Participants gave informed consent to participate in the study before taking part. Data availability free text: The datasets used and/or analyzed during the current study are available from the corresponding author upon reasonable request. Data availability statement Data are available upon reasonable request. References Associated Data Supplementary Materials Data Availability Statement Data are available upon reasonable request.