RNA sequencing data and corresponding clinical information for patients with gastric cancer were obtained from The Cancer Genome Atlas Stomach adenocarcinoma (TCGA-STAD, portal.gdc.cancer.gov/projects/TCGA-STAD) cohort. A total of 443 patients were initially identified. Subsequently, 5 patients were excluded due to incomplete follow-up information required for survival analysis, including missing survival time or survival status. Therefore, 438 patients were retained for follow-up and prognostic analyses. No newly collected patient data were included in the present study for the first time. All patient-level data were obtained from TCGA-GDC (portal.gdc.cancer.gov/), GEO (https://www.ncbi.nlm.nih.gov/geo/), HPA (proteinatlas.org/), TISIDB (cis.hku.hk/TISIDB/), TIDE (tide.dfci.harvard.edu/) and STRING (https://string-db.org/). GO (http://geneontology.org/) and KEGG (https://www.genome.jp/kegg/) databases were used for enrichment analyses (14–19). The external validation cohort GSE84437 (20) was downloaded from the Gene Expression Omnibus database (https://www.ncbi.nlm.nih.gov/geo/) and included transcriptomic and clinical information from 483 patients with gastric cancer. The GSE84437 SuperSeries contains 483 gastric cancer samples and was generated using the GPL6947 Illumina HumanHT-12 V3.0 expression beadchip platform. Gene expression matrices were annotated using the GPL6947 Illumina HumanHT-12 V3.0 expression beadchip platform annotation file.. When multiple probes corresponded to the same gene symbol, the mean expression value was used for downstream analysis with gastric cancer.

A total of 28 lipo-CAF-associated genes were selected through a structured literature review. Genes were included if previous studies reported their association with lipid metabolism, lipid transport, fatty acid handling, adipocyte-like stromal features, lipid-rich fibroblast phenotypes, CF-associated stromal remodeling or cancer-associated lipid metabolic programs. The selected genes and information regarding their supporting studies are listed in Table I (21–48). This literature-derived gene set was used as an initial biologically informed signature to evaluate lipid-associated CAF-associated features in gastric cancer. To identify gastric cancer phenotype-associated biological processes, differential expression analysis was performed between the lipo-CAF-high and -low groups using the limma package (version 3.52.0) in R software version 4.2.1 (49). Genes with adjusted P-value <0.05 and |log2 fold change| >1 were considered significantly differentially expressed. followed by Gene Ontology (50) and Kyoto Encyclopedia of Genes and Genomes (51) enrichment analyses. A protein-protein interaction (PPI) network of lipo-CAF-associated genes was constructed and visualized using the STRING database (19).

Any association between lipo-CAF gene expression and overall survival (OS) was evaluated using univariate Cox regression analysis. To identify a prognosis-associated signature, Least Absolute Shrinkage and Selection Operator (LASSO) Cox regression analysis was performed. Lipo-CAF genes were selected based on the Akaike information criterion (52), which was used for model selection by balancing model fit and complexity. The risk score for each patient was calculated as the sum of the expression value of each selected gene multiplied by its corresponding regression coefficient. Univariate and multivariate Cox regression analyses were performed to evaluate independent prognostic factors, including clinical features and risk score, in gastric cancer. Kaplan-Meier survival analysis was performed using mRNA expression and overall survival data from TCGA-STAD and GSE84437 gastric cancer cohorts. Patients were stratified into high- and low-risk groups according to the median risk score cutoff of 2.20 and survival differences were compared using the log-rank test.

Stromal, immune and ESTIMATE score and tumor purity were calculated using the ESTIMATE algorithm (53). ESTIMATE is an expression-based method that infers the levels of stromal and immune cell infiltration in tumor tissue based on gene expression signatures. The ESTIMATE algorithm was implemented using the estimate package in R software, and normalized gene expression data from TCGA-STAD cohort were used as input. The stromal score, immune score, ESTIMATE score and tumor purity were compared between the lipo-CAF-high and lipo-CAF-low groups.

Tumor Immune Dysfunction and Exclusion (https://tide.dfci.harvard.edu) was used to evaluate immune escape in patients classified into high- and low-risk groups according to the prognostic signature. The expression levels of human leukocyte antigen (HLA) family genes and immune checkpoint molecules (TIGIT, PDCD1, SIGLEC15, CTLA4, CD274, PDCD1LG2, LAG3 and HAVCR2) were compared between the high- and low-risk groups. Spearman correlation analysis was conducted to assess the association between immune cell infiltration and risk score.

NIH-3T3 mouse fibroblasts were obtained from the American Type Culture Collection (cat. no. CRL-1658) and mouse forestomach carcinoma (MFC) gastric cancer cells were obtained from Procell Life Science & Technology Co., Ltd. (cat. no. CL-0156). NIH-3T3 cells were cultured in high-glucose DMEM supplemented with 10% FBS and 1% penicillin-streptomycin. MFC cells were cultured in RPMI-1640 medium supplemented with 10% FBS and 1% penicillin-streptomycin. Cells were maintained at 37°C in a humidified incubator containing 5% CO₂. Cells were used within 10 passages after thawing and were determined to be free of Mycoplasma contamination. NIH-3T3 cells were used as a technically tractable fibroblast-based model to evaluate whether knockdown of candidate lipo-CAF-associated genes in fibroblasts could affect gastric cancer cell behavior in co-culture. Given that the gastric cancer cells used in the co-culture system were mouse MFC cells, NIH-3T3 cells were selected to maintain species compatibility.

3×xTransduced NIH-3T3 mouse fibroblasts were co-cultured with MFC mouse gastric cancer cells in 96-well plates. Briefly, NIH-3T3 cells were seeded into 96-well plates and allowed to adhere overnight at 37°C. MFC cells were added, and co-cultured in complete medium at 37°C. CCK-8 assays were performed at 0, 24, 48 and 72 h after co-culture. CCK-8 (Beyotime Biotechnology, cat. no. C0038) was added to each well and incubated for 2 h at 37°C. Absorbance was measured at 450 nm using a microplate reader.

xxTransduced NIH-3T3 mouse fibroblasts were co-cultured with MFC mouse gastric cancer cells for 48 h at 37°C. Cells were fixed with 4% paraformaldehyde for 30 min at room temperature. TUNEL assay was performed using a One Step TUNEL Apoptosis Assay kit (Beyotime Biotechnology, cat. no. C1088) according to the manufacturer's instructions. Briefly, cells were incubated with the TUNEL reaction mixture for 60 min at 37°C in the dark. After washing with PBS, nuclei were counterstained with DAPI (1 µg/ml) for 5 min at room temperature. Samples were mounted using anti-fade fluorescence mounting medium and observed under a fluorescence microscope. In total, ≥5 randomly selected fields of view were captured for each sample. Fluorescence signals were analyzed using ImageJ software version 1.53t (National Institutes of Health). The percentage of apoptotic cells was calculated as the number of TUNEL-positive cells divided by the total number of DAPI-positive cells.

HPA (proteinatlas.org) was used to evaluate the protein expression of candidate genes in gastric cancer tissues and normal gastric tissues. For HPA immunohistochemistry images, the corresponding HPA antibody accession numbers were recorded. The antibody accession numbers used for normal and tumor tissues were as follows: Apolipoprotein (APO)-C3 (accession no. HPA073918 for both normal and tumor tissues); APOA2 (accession no. HPA072575 for normal tissue and CAB025885 for tumor tissue); APOC2 (accession no. HPA055877 for both normal and tumor tissue); prostaglandin-endoperoxide synthase 2 (PTGS2; accession no. HPA001335 for both normal and tumor tissues); FABP4 (accession no. CAB024961 for normal tissue and HPA002188 for tumor tissue); CD36 (accession no. CAB025866 for both normal and tumor tissues); and ABCA8 (accession no. HPA044914 for both normal and tumor tissues). The TISIDB database (cis.hku.hk/TISIDB/) was used to assess the associations between candidate gene expression and clinical features in stomach adenocarcinoma. The TISIDB analyses were based on the cohort from TCGA-STAD. The source cohort was recorded as TCGA-STAD, with Genomic Data Commons project ID: TCGA-STAD (Database Of Genotypes And Phenotypes accession no. phs000178).

Transduced NIH-3T3 mouse fibroblasts were lysed using RIPA lysis buffer (Thermo Fisher Scientific, Inc.; cat. no. 89900) supplemented with protease inhibitor cocktail. Total protein concentration was determined using a bicinchoninic acid protein assay kit. Equal amounts of protein (30 µg/lane) were loaded onto 10% SDS-PAGE gels and transferred onto PVDF membranes. The membranes were blocked with 5% skimmed milk in Tris-buffered saline containing 0.1% Tween-20 (TBST) for 1 h at room temperature. The membranes were incubated with primary antibodies against FABP4 (1:1,000; cat. no. #2120S; Cell Signaling Technology, Inc.), CD36 (cat. no. #74002S; Cell Signaling Technology, Inc.), ABCA8 (all 1:1,000; cat. no. ab230896; Abcam) and GAPDH (1:5,000; cat. no. ab8245; Abcam) overnight at 4°C. After washing three times with TBST, the membranes were incubated with HRP-conjugated anti-rabbit (1:5,000; cat. no. #7074S; Cell Signaling Technology, Inc.) and anti-mouse IgG (1:5,000; cat. no. #7076S; Cell Signaling Technology, Inc.), for 1 h at room temperature. Protein bands were detected using Pierce ECL Western Blotting Substrate (Thermo Fisher Scientific, Inc.; cat. no. 32106) and visualized using a chemiluminescence imaging system. Band intensities were quantified using ImageJ software version 1.53t (National Institutes of Health) and normalized to GAPDH.

Lentiviral shRNA plasmids targeting mouse Fabp4, Cd36 or Abca8a were constructed using the pLKO.1-puro lentiviral vector. The pLKO.1-puro vector and psPAX2 packaging and pMD2.G envelope plasmid were obtained from Addgene, Inc. Lentiviral particles were generated using a second-generation lentiviral packaging system. Briefly, 293T cells (American Type Culture Collection; cat. no. CRL-3216) were maintained in high-glucose DMEM supplemented with 10% FBS and 1% penicillin-streptomycin at 37°C. For lentivirus production, 293T cells were co-transfected with 10 µg pLKO.1-shRNA plasmid, 7.5 µg psPAX2 packaging plasmid and 2.5 µg pMD2.G envelope plasmid at a mass ratio of 4:3:1 using Lipofectamine 3000 transfection reagent. Transfection was performed at 37°C for 6 h, after which the medium was replaced with fresh complete medium. Lentiviral supernatant was collected at 48 and 72 h after transfection. NIH-3T3 mouse fibroblasts were infected with the collected lentiviral particles at a multiplicity of infection of 10 in the presence of 8 µg/ml polybrene. After 24 h of transduction at 37°C in a humidified incubator containing 5% CO₂, the medium was replaced with fresh complete medium. Cells were selected with 2 µg/ml puromycin for 72 h and maintained in medium containing 1 µg/ml puromycin. Knockdown efficiency was assessed by western blotting 72 h after transduction, and the sequence with the highest knockdown efficiency for each target gene was used for subsequent functional assays. The shRNA target sequences were as follows: shFABP4-1, 5′-CACCGAGATTTCCTTCAAA-3′; shFABP4-2, 5′-CTGGATGGAAATTTGCATCAA-3′; shFABP4-3, 5′-TGTGTGATGCCTTTGTGGG-3′; shCD36-1, 5′-GGACCATTGGTGATGAGAAGG-3′; shCD36-2, 5′-GGCTGTGTTTGGAGGTATTCT-3′; shCD36-3, 5′-GCTGTGTTTGGAGGTATTCTG-3′; shABCA8A-1, 5′-GCTGCTATGTTCTTCCTGAAA-3′; shABCA8A-2, 5′-GCAGATGATGCTGCTGATGAA-3′; shABCA8A-3, 5′-GCTGATGACCTTCTTCATCAA-3′; and sh-negative control (NC), 5′-TTCTCCGAACGTGTCACGT-3′. For co-culture assays, transduced NIH-3T3 cells were seeded into culture plates and allowed to adhere overnight at 37°C. MFC mouse gastric cancer cells were added and co-cultured with NIH-3T3 cells in complete medium at 37°C. Co-culture was performed for 0, 24, 48 and 72 h for CCK-8 assays and 48 h for TUNEL assays.

Statistical analyses were performed using R software (version 4.2.1; Posit Software, PBC) and GraphPad Prism (version 9.5.1; Dotmatics). Data are presented as the mean ± SD unless otherwise stated. Data normality was assessed using the Shapiro-Wilk test. For two-group comparisons, unpaired two-tailed Student's t-tests or Wilcoxon rank-sum tests were used as appropriate. For comparisons among >2 groups, one-way ANOVAs followed by Tukey's multiple-comparison tests or Kruskal-Wallis tests followed by Dunn's multiple-comparison tests were used as appropriate. For time-course experiments, two-way ANOVA followed by Sidak's multiple-comparison tests was used. Kaplan-Meier survival curves were compared using the log-rank test. Cox regression analysis was used to evaluate prognostic factors and LASSO Cox regression was used to construct the prognostic model. Spearman correlation analysis was used to assess correlations between variables. All tests were two-sided and P<0.05 was considered to indicate a statistically significant difference.

To investigate the role of lipo-CAFs in gastric cancer progression, 28 lipo-CAF-associated genes were collected (Table I). Differential expression analysis revealed that the majority of lipo-CAF-associated genes were expressed at higher levels in gastric cancer compared with in normal tissues (Fig. 1A). protein-protein interaction (PPI) analysis demonstrated that these genes were closely interconnected (Fig. 1B). Based on the expression patterns of these lipo-CAF-associated genes, patients with gastric cancer from TCGA-STAD cohort were classified into two subtypes (Fig. 1C-E). Significant differences in lipo-CAF gene expression were observed between the two subtypes (Fig. 1F). In addition, patients with the C2 subtype (lipo-CAF-high) exhibited significantly shorter OS compared with those with the C1 subtype (Fig. 1G).

Differential gene expression and gene enrichment analyses were performed to compare the two subtypes (Fig. 2A and B). GO enrichment analysis showed that differentially expressed genes between the lipo-CAF-high and -low groups were enriched in terms associated with synaptic signaling, membrane potential regulation and ion-channel activity, including ‘regulation of membrane potential’, ‘modulation of chemical synaptic transmission’, ‘synapse organization’, ‘potassium ion transport’, ‘ion channel complex’ and ‘channel activity’ (Fig. 2C). Pathway enrichment analysis further demonstrated that these genes were enriched in energy metabolism-associated pathways, including the ‘calcium signaling pathway’, ‘cell adhesion molecules’ and the ‘cAMP signaling pathway’ (Fig. 2D). In addition, mutation analysis revealed that patients with gastric cancer in the lipo-CAF-high group exhibited a lower overall mutation rate (Fig. 2E and F). Estimation of Stromal and Immune cells in Malignant Tumours using Expression data-based analyses showed that patients in the lipo-CAF-high group exhibited significantly higher immune and stromal scores, whereas tumor purity was significantly lower, compared with the lipo-CAF-low group (Fig. 3A-D). Immune infiltration analysis (Fig. 3E) further indicated that the lipo-CAF-high group exhibited increased infiltration of naïve B cells, CD8⁺ T cells, monocytes, M2 macrophages, dendritic cells and mast cells (Fig. 3F). Finally, the lipo-CAF-high group demonstrated higher expression levels of HLA family genes (Fig. 4A) and immune checkpoint molecules, including T cell immunoreceptor with Ig and ITIM domains, programmed cell death 1 and cytotoxic T lymphocyte-associated protein 4 (Fig. 4B).

Based on the role of lipo-CAF-associated genes in gastric cancer as aforementioned, a prognostic model was constructed using these genes. Univariate Cox regression analysis identified nine genes with prognostic value (Fig. 4C). Subsequently, a prognostic signature was established using LASSO Cox regression analysis (Fig. 4D and E). The model was internally validated in the TCGA cohort and externally validated in the GSE84437 cohort and was significantly associated with OS in patients with gastric cancer (Fig. 4F and G). Visualization using a heatmap demonstrated that these genes were highly expressed in patients at high-risk and were associated with poor survival outcomes (Fig. 5A-C). Furthermore, univariate and multivariate Cox regression analyses showed that the prognostic model-derived risk score was an independent prognostic factor for patients with gastric cancer (Fig. 5D and E).

To determine whether the prognostic model was associated with immune cell infiltration, correlation analysis was performed between the risk score and immune cell infiltration levels. Results showed that the risk score was positively correlated with M2 macrophages and mast cells and negatively correlated with plasma cells (Fig. 5F-H). Immune therapy response scoring was conducted in TCGA-STAD patients with gastric cancer, with results suggesting that high-risk patients, as defined by the prognostic model, were more likely to exhibit no response to immunotherapy (Fig. 5I).

To determine whether the lipo-CAF genes included in the prognostic model were expressed in CAFs in gastric cancer, single-cell transcriptomic datasets GSE134520 and GSE167297 were analyzed. GSE134520 was originally generated to construct a single-cell transcriptomic atlas of gastric premalignant and early-malignant lesions, including non-atrophic gastritis, chronic atrophic gastritis, intestinal metaplasia and early gastric cancer. GSE167297 was originally generated from diffuse-type gastric cancer samples to characterize spatially distinct tumor microenvironment features The results showed that FABP4, CD36, ABCA8 and PTGS2 were expressed in CAFs (Figs. S1 and S2). Subsequently, clinical validation analyses were performed using publicly available online databases. FABP4, CD36 and ABCA8 were associated with poor prognosis in gastric cancer and showed clinicopathological associations with tumor stage and grade (Fig. S3A, F and G), whereas the remaining genes did not show significant differences (Fig. S3B-E and H). Furthermore, protein expression of lipo-CAF-associated genes was evaluated using the HPA database, which showed that FABP4, CD36 and ABCA8 were highly expressed in CAFs within gastric cancer tissue (Fig. 6A-G). To further investigate their functional roles, these three genes were individually knocked down in mouse fibroblasts using lentiviral vectors (Fig. 6H-J) and subsequently co-cultured with mouse gastric cancer MFC cells (Fig. 6K). Functional assays demonstrated that knockdown of these genes significantly reduced MFC cell proliferation, as measured by CCK-8 assays and was accompanied by increased TUNEL-positive staining in the co-culture system (Fig. 6L-Q). These findings indicate that lipo-CAF-associated genes, particularly FABP4, CD36 and ABCA8, may exert tumor-promoting effects in fibroblast-based co-culture systems.