Integrative analysis of cancer-associated fibroblast signature in gastric cancer

Fig. 1. Identification of CAFs subtypes in the TME of gastric cancer. (A) t-Stochastic Neighbor Embedding (t-SNE) plot for the 10,741 CAFs in 26 primary gastric cancer tissues. (B) The t-SNE plot of the distribution of 5 subtypes of CAFs after clustering. (C) Cell number and proportions of subtypes of CAFs in 26 primary gastric cancer tissues. (D) Bubble plot showing the expression of the top 5 marker genes in the CAFs subtypes. The proportion of CAFs subgroups with various clinical characteristics, including gender (E), age (F), stage (G), TCGA classification (H), and Lauren classification (I). (J) KEGG enrichment analysis of 5 subtypes of CAFs.

Fig. 2. Exploring cellular communication between 5 subtypes of CAFs and the relationship with gastric cancer prognosis. The circos plots show the top 20 most highly expressed ligand-receptors for CAFs subtypes interactions; the network plots depict the number of ligand-receptor interactions between subtypes of CAFs. Ligand-receptors-related gene pairs include immune checkpoint (A), cytokines/chemokines (B), growth factors (C), and others (D). The ssGSEA combined with Kaplan-Meier analysis was used to analyze the prognostic significance of different CAFs subtypes, including the CAF_0 subtype, CAF_1 subtype, CAF_2 subtype, CAF_3 subtype and CAF_4 subtype in TCGA-STAD (E) and meta-GEO (F) cohorts.

Fig. 3. Correlation analysis of CAF_0 with other clinical characteristics, and exploration of CAF_0 immune infiltration and mutation spectrum in gastric cancer (GC). Kruskal-Wallis test revealed that increasing CAF_0 infiltration correlated with higher T stages, poor pathological stages, progressive tumor grades, and Lauren subtype (A). The difference of immune infiltration in GC patients with different CAF_0 abundance (B). Differences in mutation spectrum among GC patients with varying levels of CAF_0 abundance (C).

Fig. 4. Identification of CAFs hub genes for the building of CAFs-score. (A) The Kaplan-Meier analysis of marker genes for subtypes of CAFs with a poor prognosis, including POSTN, CCL11, APOD, CXCL14, and CFD. (B) Expression of POSTN proteins in human gastric normal and cancer tissues by Immunohistochemical analysis. (C) Expression of POSTN proteins in human gastric normal and cancer tissues by immunofluorescence staining. (D) The expressions of α-SMA and FAP in NF and CAF were detected by immunofluorescence staining. (E) Western blot analysis was conducted to assess the expression of POSTN in gastric cancer tissue, adjacent non-cancerous tissue, as well as in normal fibroblasts and cancer-associated fibroblasts. (F) Volcano plot of differentially expressed genes of normal and cancer tissues in TCGA-STAD. (G) Volcano plot of CAFs prognostic hub genes identified by univariate Cox regression analysis. (H) The lambda trajectory of each independent variable. (I) Plots illustrating the calculated coefficient distributions for the logarithmic (lambda) series used for parameter selection (lambda). (J) Coefficients of multivariate Cox for each gene in the CAFs-score.

Fig. 5. Validation of the prognostic CAFs-related signature based on CAFs-score (KM survival analysis based on the median CAFs-score). The upper left parts are distribution plots for the relationship between CAFs-score and survival status; the lower left parts are heatmaps for the three prognostic CAFs-related genes in the cohorts; the upper right parts are ROC curve for the CAFs-score in the different cohorts; the lower right parts are survival curves between high- and low-CAFs-score groups. Internal cohorts: (A) TCGA-STAD; External cohorts: (B)meta-GEO.

Fig. 6. The establishment of a nomogram for predicting the prognosis of gastric cancer. (A) Univariate and multivariate Cox analyses of CAFs-score and clinicopathological characteristics in TCGA-STAD and meta-GEO. (B) The nomogram model incorporating CAFs-score and stage was established. (C) The time-dependent ROC curves, calibration plots, decision curves, and Comparison curves of the predictive capacity of clinicopathological features of the nomogram for 1-, 2-, and 3-years OS in TCGA-STAD. (D) The time-dependent ROC curves, calibration plots, decision curves, and Comparison curves of the predictive capacity of clinicopathological features of the nomogram for 1-, 2-, and 3-years OS in meta-GEO.

Fig. 7. The TIME of gastric cancer based on the CAFs-score. (A) Evaluation of TIME in different CAFs-score groups based on ESTIMATE algorithm. (B) The relative level of immune infiltration of 22 immune cells between groups with different CAFs-score. (C) The correlation between CAFs-score and different immune cells and stromal cells was evaluated based on multiple algorithms. (D) Correlation of immune cells with the three genes used to calculate the CAFs-score. (E) CAFs-score correlation with other CAFs or stromal scoring systems. (F) Expression of selected immune checkpoint genes in different CAFs-score groups. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

Fig. 8. The mutation landscape in different CAFs-score groups. (A–B) Significantly mutated genes in the TCGA-STAD samples of the subgroups with low and high CAFs-score, respectively. Mutated genes (rows, top 20) are ranked by mutation rate; samples (columns) are grouped to emphasize mutual exclusivity among mutations. The percentage of mutations is presented on the right, while the total number of mutations is shown at the top. The color coding indicates the sort of mutation. (C–D) Distribution of tumor mutational burden (TMB) among different CAFs-score groups and correlation with CAFs-score. (E–F) Distribution of microsatellite instability (MSI) in different CAFs-score groups and CAFs-score in MSS/MSI-L/MSI-H groups. (G) The relationships between the CSC index and CAFs-score.

Fig. 9. The responsiveness of CAFs-score to immunotherapy in different cohorts. (A) Distribution of immune subtypes within different CAFs-score groups in TCGA-STAD. (B–E) TIDE, MSI, T cell dysfunction, and T cell exclusion score in different CAFs-score subgroups within TCGA-STAD, respectively. (F–H) The CAFs-score was used to evaluate the differences in prognosis across the different CAFs-score subgroups in the IMvigor210, GSE176307, and GSE91061 cohorts. (I) Evaluation of differences in immunotherapy response between CAFs-score subgroups within the IMvigor210, GSE176307, and GSE91061 cohorts. ***P < 0.001.

Fig. 10. Drug sensitivity analysis based on CAFs-score. (A) IC50 testing results for drugs with IC50 < 1 in different CAFs-score subgroups. (B–K) Potential drugs with significant treatment differences in the different CAFs-score subgroups.