Two CAF cell lines were isolated from clinical GC samples and routinely maintained in advanced DMEM/F12 medium (Thermo Fisher Scientific, Inc.) enriched with 10% fetal bovine serum (FBS; Thermo Fisher Scientific, Inc.) in a humidified incubator at 37˚C in 5% CO₂. Human GC cell line AGS (cat. no. CRL-1739) was purchased from American Type Culture Collection and routinely maintained in a humidified incubator at 37˚C in 5% CO₂. The primary GC samples on tissue microarrays were employed, containing 278 patients collected between January 1998 and December 2006 at Prince of Wales Hospital (Hong Kong, China). A total of 184 male and 96 female patients with GC (age range, 32-88 years; median age, 65 years) was included in the cohort. The clinical information of each case, together with survival status and survival time, were collected. Human primary sample usage was approved (approval no. 2024.219) by the Joint Chinese University of Hong Kong-New Territories East Cluster Clinical Research Ethics Committee (Hong Kong, China), which is responsible for performing ethics and scientific review and oversight of clinical studies undertaken by Chinese University of Hong Kong and/or NTEC.

After dewaxing with xylene and rehydration with descending ethanol solutions (100, 95, 70%), tumor microarray sections (4 µm) were put into Tris-EDTA (pH 8.0) and boiled in a pressure cooker for 3 min for antigen retrieval. Then the sections were covered by 3% H₂O₂ solution and incubated at 25˚C for 15 min, followed by blocking at 25˚C with 5% bovine serum albumin (BSA, cat. no. 10033; MP Biomedicals) for 1 h. The prepared sections were incubated with POSTN (1:2,000; cat. no. ab219057; Abcam) as the primary antibody for 3 h, then incubated with anti-rabbit IgG-HRP (1:2,000; cat. no. P0448; Dako; Agilent Technologies, Inc.) for 1 h and stained with DAB (Dako; Agilent Technologies, Inc.). IHC staining results were evaluated using brightfield microscopy. Samples with more than 5% POSTN-positive fibroblast cells were defined as POSTN-positive cases. Kaplan-Meier survival analysis followed by the log-rank test was conducted by R package ‘survival’ (v3.3.1) (https://cran.r-project.org/web/packages/survival/).

Cell transfections were carried out using Lipofectamine 3000 Transfection Reagent (Thermo Fisher Scientific, Inc.). Subsequent experimentations were conducted after a 1-day incubation at 37˚C following transfection. Sequences of siYAP1-1, siYAP1-2, siTEAD1-1, siTEAD1-2, siPOSTN-1 and siPOSTN-2 are listed in Table SI and adopted at the dosage of 10 nM. Cell lysates were prepared by incubating cells on ice for 30 min in RIPA lysis buffer (150 mM NaCl, 1.0% Triton X-100, 0.5% sodium deoxycholate, 0.1% SDS, 50 mM Tris, adjusted pH to 8.0). The total protein concentration was then quantified using the BCA Protein Assay Kit (cat. no. 23225; Thermo Fisher Scientific, Inc.). Protein samples (50 µg per lane) were separated by 10% (w/v) sodium dodecyl sulfate-polyacrylamide gel electrophoresis and transferred onto polyvinylidene fluoride membranes, which were blocked for 2 h at 25˚C in Tris-buffered saline with Tween (TBST) containing 5% BSA. The membranes were incubated overnight at 4˚C with primary antibodies, washed twice with TBST, and then incubated for 1 h with HRP-conjugated secondary antibodies. Protein bands were visualized using Clarity Western ECL substrate (Bio-Rad Laboratories, Inc.). The primary YAP1 (1:5,000; cat. no. ab52771) and POSTN (1:1,000; cat. no. ab219057) antibodies were purchased from Abcam. TEAD1 (1:1,000; cat. no. LS-C334933) antibody was purchased from LS Bio (https://www.lsbio.com/). pRb (Ser807/811; 1:1,000; cat. no. 9308), PARP (1:1,000; cat. no. 9542), cleaved PARP (1:1,000; cat. no. 9541), p21 (1:1,000; cat. no. 2947) and Cyclin D1 (1:1,000; cat. no. 55506) were obtained from Cell Signaling Technology, Inc. β-actin (1:5,000; cat. no. YM3028) was obtained from ImmunoWay Biotechnology Company. Anti-mouse IgG-HRP (1:2,000; cat. no. P0260) and anti-rabbit IgG-HRP (1:2,000; cat. no. P0448; both from Dako; Agilent Technologies, Inc.) were used as secondary antibodies.

Cell viability was determined by the Cell Counting kit-8 (CCK-8; MedChemExpress). A total of 1:10 of CCK-8 reagent was added to the cell culture medium for 1.5-2 h of incubation, then determined by the spectrophotometer (SpectraMax iD3; Molecular Devices, LLC) at an absorbance of 450 nm (OD450). Cell invasion assays were performed in Matrigel Invasion Chambers (Corning, Inc.). CAFs suspended in serum-free medium were added to the upper chamber, and the medium containing 10% FBS was added to the lower chamber. After incubation for 16-20 h, invasive cells to the lower membrane surface were stained with 0.4% (w/v) crystal violet (MilliporeSigma) and then quantified.

The Cancer Genome Atlas (TCGA)-stomach adenocarcinoma (STAD) dataset was used in the present study (https://xenabrowser.net/). The clinical and survival information was collected. Samples with POSTN expression levels higher than the median expression level were defined as POSTN-high-expression cases. Kaplan-Meier survival analysis was conducted by R package ‘survival’. For the differentially expressed gene (DEG) identification and functional enrichment analysis with regard to the POSTN expression level, the whole genomic expression level alteration was first evaluated by comparing the 10% samples with the lowest POSTN expression (POSTN^-, n=37) and the 10% samples with the highest POSTN expression (POSTN⁺, n=37). DEGs between two groups were identified using the R package ‘DESeq2’. The Gene Ontology (GO) enrichment analysis and gene set enrichment analysis (GSEA) was performed based on the gene expression alterations using the R package ‘clusterProfiler.’ The expression heatmap was visualized using the R package ‘pheatmap’ (v1.0.12) (https://cran.r-project.org/web/packages/pheatmap/). Spearman correlations between the expression level of POSTN and ECM remodeling gene set were calculated using the R package ‘stats’ (v4.1.3). Two-tailed P<0.05 was defined as indicating a statistically significant difference. The binding motifs of YAP1/TEAD1 on the promoter region of POSTN were predicted by the Eukaryotic Promoter Database (https://epd.epfl.ch//index.php) and JASPAR 2022 database (https://jaspar.genereg.net).

The single-cell RNA sequencing (scRNA-seq) analysis was carried out based on a public dataset (https://dna-discovery.stanford.edu/research/datasets/) in the sub-dataset ‘Gastric scRNAseq’ (24) using the R package ‘Seurat’ (v4.0.2) (https://satijalab.org/seurat/). After routine data normalization and cluster labeling, gene expression in individual cells was visualized using ‘FeaturePlot’ and ‘Nebulosa’. Fibroblasts were extracted from the whole dataset for further subgrouping into four subtypes based on the 12 pan-cancer CAF markers identified in our previous study (23).

For subcutaneous xenograft formation assay, CAF-1 with shCtrl or shPOSTN (Vector backbone: pLKO.1; target sequence: 5'-CACUUGUAAGAACUGGUAUAATT-3') treatment were mixed with AGS cells (10⁶ cells/mouse) and then subcutaneously injected into 4-week-old NOD scid gamma (NSG) mice (n=5/group). NSD mice (weighted 18-22 g) were routinely housed under controlled conditions at 22-24˚C with 40-60% humidity, maintained on a 12/12-h light/dark cycle, and provided ad libitum access to standard laboratory chow and autoclaved water. Anesthesia with intraperitoneal (i.p.) injection of a mix of Ketamine (75 mg/kg) and Xylazine (10 mg/kg) will be conducted when sacrificing the mice. The tail reflexes and breathing patterns were monitored until anesthesia was achieved, after which cervical dislocation was performed for euthanasia. The mice were directly sacrificed following anesthesia. The xenografts were harvested at the 16th day after injection. All mice experiments were authorized by the Animal Ethics Experimentation Committee from CUHK (approval no. 24-036-NSF; Hong Kong, China).

Unpaired student's t-test was used to compare the variations between assay groups and control. Spearman's correlation was adopted for correlation analyses. Statistical analyses were performed by GraphPad Prism 8.0 (GraphPad; Dotmatics) and SPSS software (version 22.0; IBM Corp.). Data were expressed as the mean ± standard error of the mean (SEM) of triplicate independent experiments. Two-tailed P<0.05 was defined as indicating a statistically significant difference.

The distribution of five ECM remodeling-related CAF markers was re-plotted in the previously established scRNA-seq atlas reflecting the GC TME, which comprises eight cell types: T cells, B cells, cancer cells, myeloid cells, fibroblasts, plasmacytoid dendritic cells, mast cells and endothelial cells (Fig. 1A). The expression of POSTN was predominantly enriched in fibroblasts, with ~30% of fibroblasts exhibiting high levels of POSTN. Additionally, POSTN expression was also observed in endothelial cells, where a significant percentage of cells were POSTN positive, albeit at lower expression levels. POSTN expression was rarely detected in other cell types (Fig. 1B). The other four CAF markers, C7, IGF1, COL10A1 and CTHRC1, also showed high enrichment in fibroblasts (Fig. S1A).

Further analysis of the subgroups of GC fibroblasts identified four cancer-associated subtypes: Progenitor-like CAF (proCAF), matrix-remodeling CAF (matCAF), myeloid CAF (myCAF) and inflammatory CAF (iCAF) (Fig. 2A). POSTN expression was notably enriched in matCAFs, along with COL10A1 and CTHRC1 (Figs. 2B and S1B). Meanwhile, C7 and IGF1 were significantly upregulated in proCAFs (Fig. S1B).

In the TCGA-STAD cohort, patients with GC with high levels of POSTN demonstrated poorer overall survival rates (P=0.0012) (Fig. 3A). When stratified by Lauren classification, patients with GC with relatively higher POSTN expression exhibited a trend toward poorer prognosis, although this was not statistically significant (P=0.11 and 0.07, respectively) (Fig. 3B and C). Furthermore, the proCAF markers C7 (P=0.085) and IGF1 (P=0.084) did not show significant associations with prognosis prediction, while the other two matCAF markers COL10A1 (P=0.039) and CTHRC1 (P=0.016) were associated with poorer prognosis (Fig. S2).

Further analyses were performed to assess the association between clinicopathological features and disease-specific survival in patients with GC (Table I). Sex distribution showed no significant difference, with females comprising 35.8% and 37.2% in the low and high POSTN groups, respectively (P=0.879). Similarly, age distribution (≥60 years: 64.7 vs. 70.3%, P=0.319) and tumor stage (Stage I: 16.2 vs. 9.9%, Stage IV: 11.0 vs. 9.3%, P=0.237) did not differ significantly between groups. However, tumor depth (T category) exhibited a highly significant association with POSTN expression (P<0.001), as no T1 tumors were observed in the high POSTN group, while T4 tumors were more prevalent in the high POSTN group (33.1 vs. 22.0%). Lymph node involvement (N category) and distant metastasis (M category) showed no significant differences (P=0.827 and P=0.676, respectively).

In the Cox proportional hazards model (Table II), age <60 years was significantly associated with improved survival (HR=0.628, P=0.016), a finding confirmed in multivariate analysis (HR=0.571, P=0.005). Advanced tumor stage (Stage III: HR=2.464, P=0.008; Stage IV: HR=4.411, P<0.001) and deeper tumor invasion (T3: HR=8.653, P=0.032; T4: HR=9.220, P=0.028) were significantly associated with poorer survival in univariate analysis, though these associations were not significant in multivariate analysis. Lymph node involvement (N1: HR=1.666, P=0.038; N2: HR=1.707, P=0.044; N3: HR=2.721, P<0.001) and distant metastasis (M1: HR=2.045, P=0.014) also predicted worse survival in univariate analysis, with N3 remaining significant in multivariate analysis (HR=2.367, P=0.038). High POSTN expression was significantly associated with poorer survival in both univariate (HR=1.665, P=0.003) and multivariate analyses (HR=1.635, P=0.007), highlighting its potential as an independent prognostic factor in GC.

The potential of POSTN as a prognosis predictor was further supported by analyses of in-house primary GC samples. Images from IHC staining of eight representative cases, showcasing both POSTN-low (Fig. 4A) and POSTN-high (Fig. 4B) groups, were presented. The results indicated that POSTN was differentially overexpressed in the cytoplasm of CAFs in patients with GC.

Similar to the results generated from TCGA-STAD dataset, elevated POSTN protein levels were demonstrated to be associated with poor prognosis (P<0.001) (Fig. 5A). When classified by Lauren classification, patients with intestinal-type GC and high POSTN levels exhibited a noticeable decrease in survival probability after 50 months of follow-up (P<0.001) (Fig. 5B). Similarly, patients with diffuse-type GC demonstrated a stronger association between POSTN levels and prognosis predictions (P<0.001) (Fig. 5C).

Further analysis of clinicopathologic characteristics associated with POSTN protein levels in a Hong Kong cohort revealed critical insights into the correlation between POSTN expression and GC progression (Table III) A slight male predominance was indicated in both groups, with 63.4% in the low and 69.6% in the high POSTN group; however, this difference was not statistically significant (P=0.337). Notably, age emerged as a significant factor (P=0.033), with a higher percentage of patients aged 60 years or older in the high POSTN group (69.6%) compared with 56.2% in the low group. A statistically significant association was observed between elevated POSTN expression levels and advanced tumor staging across T, N and M classifications. Notably, the high POSTN cohort demonstrated substantially higher rates of advanced T-stage disease (T3/T4: 74.4 vs. 45.1% in the low POSTN group; P<0.001). This progression-dependent pattern extended to nodal involvement, with significantly greater N-stage positivity in patients with high POSTN expression (84.0 vs. 73.2%; P=0.044). Similarly, distant metastasis (M1 stage) occurred more frequently in the high-POSTN subgroup (20.8 vs. 11.1%; P=0.04), completing a consistent correlation between POSTN overexpression and metastatic progression.

The prognostic analyses for disease-specific survival using univariate and multivariate Cox regression models are presented in Table IV. Univariate analysis indicated that age ≥60 years, diffuse type, advanced tumor staging (T, N and M) and high POSTN levels were significant predictors of poor survival (all P<0.05). Subsequent multivariate analysis further confirmed the independent prognostic value of age (HR=1.982, P<0.001), advanced T classification (HR=1.949, P=0.002), lymph node involvement (HR=3.449, P<0.001), metastasis (HR=2.808, P<0.001), and high POSTN (HR=1.443, P<0.05), with POSTN emerging as a novel independent prognostic biomarker beyond traditional TNM parameters. These findings collectively highlight the multifactorial nature of GC progression.

To further demonstrate the oncogenic function of POSTN, DEGs between POSTN-highly expressed (POSTN⁺) samples and POSTN-lowly expressed (POSTN^-) samples were identified in the TCGA cohort. The volcano plot highlighted several ECM remodeling-related genes that were significantly upregulated in POSTN⁺ samples, including COL5A2, COL11A1, FNDC1, FN1, COL10A1, MMP13 and MUC2 (Fig. 6A). Pearson's correlation analysis revealed that the expression levels of multiple matrix metalloproteinase (MMP) family proteins, collagen family proteins, fibroblast growth factors and tumor migration markers were significantly correlated with POSTN (Fig. 6B). Detailed examination of each sample showed that markers associated with fibroblast proliferation and migration were generally upregulated in POSTN⁺ samples (Fig. 6C). Further GO enrichment analysis demonstrated significant enrichment of ECM remodeling-related pathways in POSTN⁺ samples, including fibroblast proliferation and migration, ECM organization, collagen biosynthetic and metabolic processes, cell-substrate and cell-matrix adhesion and cell junction assembly (Fig. 6D). GSEA results also highlighted several upregulated biological processes in POSTN⁺ samples, such as cell adhesion molecules, ECM-receptor interactions and focal adhesion. Notably, the hippo signaling pathway was also upregulated in POSTN⁺ samples, underscoring its importance in the regulation of POSTN expression (Fig. 6E). Further validations demonstrated that deletion of POSTN significantly inhibited the cell viability and invasion ability of CAFs (Fig. 6F-H). Western blot assay demonstrated cell cycling arrest features in POSTN-deleted CAFs, which was indicated by the downregulation of pRb and cyclin D1, and the upregulation of p21 and cleaved-PARP (Fig. 6I). Besides, the xenograft formation assay revealed that cancer cells mixed with POSTN-deleted CAFs generated smaller xenografts (Fig. 6J and K).

In the TCGA-STAD-based correlation analysis, strong correlations were observed between POSTN mRNA levels and YAP1/TEAD1 (R=0.26 and 0.31, respectively), as well as the downstream effectors of YAP1 signaling, CTGF and CYR61 (R=0.59 and 0.45, respectively) (Fig. 7A). In the scRNA-seq dataset, the expression levels of POSTN, YAP1 and TEAD1 were enriched in similar clusters (Fig. 7B). Pseudotime analysis revealed that the expression levels of POSTN, YAP1, TEAD1, CTGF and CYR61 were enriched during the early stages of CAF cell cycling and were chronologically associated with CAF proliferation markers (Fig. 7C). Furthermore, the upregulation of POSTN, YAP1, CTGF and CYR61 was particularly prominent in patients with early-stage GC (Fig. 7D). Binding motif prediction identified a putative YAP1/TEAD1 binding site in the promoter sequence of POSTN, located 322 base pairs upstream of the transcription start site (Fig. 7E). The western blot assay was subsequently employed to verify the regulatory function of YAP1/TEAD1 on POSTN expression. The results demonstrated that the knockdown of both YAP1 and TEAD1 downregulated POSTN expression in two GC CAF cell lines, as well as the cell cycling marker pRb (Fig. 7F).