A total of 589 patients who underwent resection of primary GC from January 2000 to December 2006 at Department of Gastroenterological Surgery, Osaka City University (currently, Osaka Metropolitan University; Osaka, Japan) were retrospectively reviewed. The age range of the recruited patients was 21–88 years old, and median age was 67. We generated tissue microarrays (TMA) from these patients and used them for immunohistochemical (IHC) staining. Each TMA core was selected at the invasion front of the cancer. The pathologic diagnoses and classifications were made according to the Japanese Classification of Gastric Carcinoma, fifteenth edition. The study protocol conformed to the ethical guidelines of the Declaration of Helsinki. The Osaka Metropolitan University ethics committee approved this study in 2022 [approval number (approval date): 2022-077 (2022/08/10)]. Informed consent was obtained in the form of opt-out.

IHC staining was performed using 589 GC samples. Slides were deparaffinized and then heated for 10 min at 105°C in an autoclave in Target Retrieval Solution (Dako, Carpinteria, CA). After blocking endogenous peroxidase activity using 3% hydrogen peroxide, the specimens were incubated with CTGF antibody (1:200; Life Technologies) for 1 h at room temperature and were incubated with biotinylated goat antirabbit IgG for 10 min. The slides were treated with streptavidin–peroxidase reagent, followed by counterstaining with Mayer's hematoxylin. Slides were scanned using a Leica Aperio CS2 scanner (Leica Biosystems), and subsequent expression evaluations at each core were performed using the open-source software QuPath (http://qupathe.github.io). Cell detection and IHC quantification were performed using 30 cores (Fig. 1). Furthermore, an objective classifier was trained to classify individual cells as stromal or epithelial cells. After these trainings, CTGF expression in all TMA cores was assessed in every tumor and stromal cell. The percentage of positive cells among total tumor and stromal cells was automatically calculated using the threshold value of 0.15. We diagnosed a case as CTGF positive when the positive percentage of tumor and stromal cells was more than the first quartile of all analyzed cases.

CTGF mRNA expression data and corresponding clinical information of GC samples were collected from The Cancer Genome Atlas (TCGA) database (http://tcga-data.nci.nih.gov/tcga/). We used data from the ‘tcga_pan_cancer_atlas_2018’ study. The third quartile of the mRNA expression level was defined as the cut-off, and we categorized patients into high- and low-expression groups.

Associations between CTGF expression and clinicopathological results were analyzed using the chi-square test. Overall survival (OS) was the time from surgery to death from any cause. The Kaplan-Meier method and the log-rank test were used to estimate and compare the OS, respectively. Disease-free survival was the time from surgery to recurrence or death from any cause. Multivariate analysis was performed using the Cox proportional hazards model. Statistical Package for the Social Sciences statistical software (version 29.0; IBM) was used for all statistical analyses. Two-sided P<0.05 was considered to indicate a statistically significant difference.

CTGF expression was mainly observed in the cytoplasm of cancer and stromal cells (Fig. 1). We investigated the CTGF expressions of cancer and stromal cells in 589 GC tissues using software (Fig. 2). Of 589 patients, 444 and 442 had CTGF-positive cancer and stromal cells, respectively. Table I shows the association between CTGF expression and clinicopathological factors. CTGF expression in stromal cells was significantly associated with CTGF expression in cancer cells (P<0.001). CTGF positivity in cancer cells was significantly associated with sex (female), tumor depth (T1-2), lymph node metastasis (N0-1), and tumor size (<3 cm). CTGF positivity in stromal cells was significantly associated with intestinal type, non-scirrhous type, tumor depth (T1-2), lymph node metastasis (N0-1), lymphatic invasion (negative), and tumor size (<3 cm).

Table II shows the univariate and multivariate analyses using the proportional hazards model. Univariate analysis revealed that the OS of patients was significantly associated with CTGF expression in stromal cells. Multivariate logistic regression analysis revealed that CTGF expression in stromal cells as well as age, tumor depth (T3-4), lymph node metastasis (N2–3), and tumor size (≥3 cm) were independent predictive parameters for OS.

Fig. 3 shows the OS curves of patients by CTGF expression using the Kaplan-Meier method. The 5-year OS rates of patients with high and low CTGF expression in cancer cells were 69.3 and 67.6%, respectively. Regarding the analyses of cancer cells, OS outcomes between patients with high and low CTGF expression were not statistically significant (log-rank; P=0.684). Conversely, the analyses of stromal cells revealed that patients with low CTGF expression demonstrated significantly worse OS than those with high CTGF expression (log-rank; P=0.006). The 5-year OS rates of patients with high and low CTGF expression in stromal cells were 72.1 and 59.8%, respectively.

Fig. 4 shows the analyses for each histological type. The OS rates in terms of the intestinal case were not significantly different depending on CTGF expression in cancer and stromal cells (P=0.183 and P=0.230, respectively). In contrast, among the cases with diffuse type, the OS of patients with low CTGF expression in stromal cells was significantly worse than that of patients with high CTGF expression (P=0.036). The cancer cell analysis in the diffuse type revealed no statistically significant difference, but CTGF-negative cases demonstrated a worse prognosis (P=0.175).

Fig. 5 shows the analyses for each macroscopic type. Among the scirrhous type GC, patients with CTGF-negative cases demonstrated worse survival in both cancer and stromal cell analyses although the difference was not significant.

Fig. 6 shows the survival analyses using TCGA. Among all cases, CTGF positivity did not affect disease-free survival (log-rank; P=0.941). Low CTGF expression cases demonstrated worse prognosis than high CTGF expression cases in genomically stable cases and diffuse type, although the differences were not significant (log-rank; P=0.431 and P=0.169).