This retrospective, single-institution study utilized surgically resected gastric tumor specimens collected at Gifu University Hospital (Gifu, Japan) between June 1, 2000, and December 31, 2008. Clinical records were reviewed to collect information on patient background, treatment, and postoperative course. Cases with histological diagnoses other than adenocarcinoma (e.g., small cell carcinoma, undifferentiated carcinoma, or neuroendocrine tumor) were excluded from the analysis. Patients were also excluded if they were under 20 years old, had incomplete clinical data, exhibited inadequate immunohistochemical staining, or were deemed unsuitable for the study by the attending physician. Histological classification was primarily performed according to the 15th edition of the Japanese Classification of Gastric Carcinoma (15). In addition, tumors were further categorized into intestinal/indeterminate and diffuse types based on the WHO Classification of Digestive System Tumours, 5th edition (36), to facilitate international comparison. Tumor staging was based on the 8th edition of the American Joint Committee on Cancer/Union for International Cancer Control Tumor-Node-Metastasis classification of malignant tumors (37). This study was approved by the Institutional Review Board of Gifu University (approval no. 2024-253; approved on December 4, 2024) and conducted in accordance with the Declaration of Helsinki (2013 version). The study was publicly disclosed on the institution's website, and patients were given the opportunity to opt out of participation. The study design, data collection, analysis, and reporting were performed in accordance with the Reporting Recommendations for Tumor Marker Prognostic Studies guidelines (38).

All paraffin-embedded tissues were cut into 4-µm-thick serial sections, deparaffinized, and stained with hematoxylin and eosin or used for IHC. For IHC, the sections were placed in citrate buffer (10 mmol/l, pH 6.0) and autoclaved in a pressure chamber at 121°C for 20 min to retrieve the antigens, followed by rinsing and blocking in 3% hydrogen peroxide in methanol for 10 min at room temperature to remove endogenous peroxidase. Non-specific binding sites were blocked in 0.01 M phosphate-buffered saline (pH 7.4) containing 2% bovine serum albumin (Wako Pure Chemical, Osaka, Japan) for 40 min at room temperature. The sections were then incubated overnight (16 h) at 4°C in a humidified chamber with the following primary antibodies: FGF10 (rabbit polyclonal IgG, 1:500; Abcam, Cambridge, UK, #ab80064), FGFR2 (rabbit polyclonal IgG, 1:500; Abcam #ab10648), HER2 (rabbit monoclonal IgG, 1:400, Clone 4B5; Roche Diagnostics, Basel, Switzerland, #790-2991), epidermal growth factor receptor (EGFR) (rabbit monoclonal IgG, 1:200, Clone D38B1; Cell Signaling Technology, Danvers, MA, USA; #4267), CD44 (mouse monoclonal IgG, 1:50, Clone G44-26; BD Pharmingen, San Diego, CA, USA; #550392), and Ki67 (mouse monoclonal IgG, 1:100, Clone MIB-1; DAKO, Agilent Technologies, Santa Clara, CA, USA; #M7240). The next day, the sections were exposed to biotinylated goat anti-rabbit IgG (1:200; Vectastain Elite ABC kit; Vector Laboratories, Newark, CA, USA) for 60 min at room temperature, followed by incubation with avidin-biotin peroxidase complex (Vectastain Elite ABC kit; Vector Laboratories) to activate immunoreactivity. The bound complex was visualized using 3,3′-diaminobenzidine (Sigma, St. Louis, MO, USA) for 3–5 min under a microscope, to prevent overstaining. Finally, the sections were counterstained with hematoxylin for 2 min and dehydrated through graded alcohols. The specificity of the antibodies for IHC was validated using positive and negative controls. FGF10 expression was confirmed in plasma cells and FGFR2 expression in the glandular epithelium in normal intestinal tissue (positive control), and negative controls showed no notable immunoreactivity (Fig. S1). The normal intestinal tissue used for positive and negative controls was obtained from the archival tissue bank of Gifu University Hospital. These samples were non-neoplastic tissues sourced from surgical resections and were processed and paraffin-embedded in the same manner as the study specimens.

The distinction between tumor and stromal cells was performed by two board-certified pathologists (K.K. and H.T.) based on standard histopathological assessment of cellular morphology in hematoxylin and eosin-stained sections. Tumor cells were identified as malignant epithelial cells characterized by features such as nuclear pleomorphism and the formation of abnormal glandular structures. Stromal cells were identified as the surrounding non-neoplastic connective tissue components, including fibroblasts, inflammatory cells, and vascular elements.

The four-tier score was determined based on the visual assessment of staining intensity by two independent pathologists. The categories were defined as follows: 0 (negative) for no discernible immunoreactivity; 1+ (weakly positive) for faint, hard-to-see staining; 2+ (moderately positive) for clear, intermediate-intensity staining; and 3+ (strongly positive) for intense staining. Representative images illustrating these four tiers of staining intensity for FGF10 and FGFR2 are provided in Fig. 1.

For example, expression levels of FGF10 and FGFR2 in tumor cells and FGF10 in the stroma were scored using a four-tier scale: 0 (negative), 1+ (weakly positive), 2+ (moderately positive), and 3+ (strongly positive) (Fig. 1). HER2 expression was classified as negative (0 or 1+) or positive (2+ or 3+), EGFR was classified as negative (0) or positive (1+ to 3+), and CD44 was classified as negative (0), weakly positive (1+), or strongly positive (2+). Immunohistochemical evaluations were independently performed by two board-certified pathologists (K.K. and H.T.) who were blinded to clinical outcomes and patient characteristics. Inter-observer agreement was assessed using weighted Cohen's κ coefficients, demonstrating substantial agreement for FGF10 tumor expression (κ=0.78), FGF10 stromal expression (κ=0.74), and FGFR2 tumor expression (κ=0.82). Discrepancies (6.8%) were resolved through collaborative microscopic review and consensus.

Patient characteristics were summarized using medians and ranges for continuous variables and frequencies and percentages for categorical variables. Baseline characteristics were compared between groups defined by FGF10 expression using the Mann-Whitney U test for continuous variables and Fisher's exact test for categorical variables.

The association between FGF10 expression (in tumor cells and stroma separately) or FGFR2 expression and differentiation status was evaluated using multivariable ordinal logistic regression models adjusting for patient age, HER2 status, and EGFR status, with differentiation status as an ordinal outcome (pap, tub1, tub2, por1, por2, and sig). Age is a fundamental demographic factor known to influence tumor biology and clinical outcomes, and HER2 and EGFR were included owing to their known roles as molecular markers in gastric cancer and their potential associations with histological differentiation (5,39). These variables were selected a priori based on biological plausibility and their relevance in previous gastric cancer studies. For the binary classification of differentiation (pap, tub1, tub2, and por1 as intestinal/indeterminate type; and por2 and sig as diffuse type), multivariable binary logistic regression models were employed with the same adjustment factors, and predicted probabilities from the multivariable models were plotted.

Overall survival (OS) was defined as the time from the date of surgery to the date of death from any cause, with data for surviving patients censored at the date of their last follow-up. OS curves were generated using the Kaplan-Meier method, and differences between groups were compared using the log-rank test. Multivariable Cox proportional hazards regression models were used to assess the prognostic impacts of FGF10 and FGFR2 expression on OS, adjusting for patient age, sex, and tumor stage. Covariates included in the multivariable Cox proportional hazards regression models were selected based on their established clinical relevance and biological plausibility in relation to survival outcomes. FGF10 and FGFR2 were treated as continuous variables in all analyses.

Multicollinearity among covariates in all multivariable models was assessed using variance inflation factors (VIF). All statistical analyses were performed using R software version 4.3.2. A two-sided P-value <0.05 was considered to indicate a statistically significant difference for all analyses. A post hoc power analysis indicated that the sample size in this study provided 73% power to detect a medium effect size (Cohen's d=0.5) with a significance level of α=0.05.

In the predicted probability plots, solid lines represent the point estimate of the predicted probability and the gray shaded areas indicate the 95% confidence interval (CI), both derived from a logistic regression model. The estimates were calculated using maximum likelihood estimation. For ordinal logistic regression models, maximum likelihood estimation was used to estimate the variance of the regression coefficients under the proportional odds assumption, and the CIs for the adjusted ORs were constructed using the estimated variance. In the Cox proportional hazards model, the variance of the regression coefficients was estimated using the partial likelihood method, and the CIs for the hazard ratios were similarly constructed using the estimated variance. The CIs for Kaplan-Meier survival probabilities were calculated using Greenwood's formula.

A total of 120 surgically resected gastric tumor specimens were initially collected, of which 117 gastric adenocarcinoma specimens were included in the final analysis, after excluding three non-adenocarcinoma tumors (1 small cell carcinoma, 1 undifferentiated carcinoma, and 1 neuroendocrine tumor). FGF10 and FGFR2 expression levels in these 117 gastric adenocarcinoma tissues were evaluated by IHC (Fig. S2). The clinicopathological characteristics of the 117 patients are shown in Table I. The median age was 68 years (range: 31–92 years), with a male predominance (65.0%). Regarding tumor progression, most cases were T4 (66.7%) and N3 (41.0%). There were no cases of Stage I disease and Stage III was the most common (63.2%), followed by Stage II (23.1%) and Stage IV (13.7%). Notably, lymphatic invasion was present in 95.7% of cases and vascular invasion was observed in 82.1% of cases.

The expression patterns of FGF10 and FGFR2 in gastric adenocarcinoma tissues are shown in Fig. 1. FGF10 expression was observed in tumor cells in 72 of 117 cases (61.5%) and in the stroma in 75 cases (64.1%). Among the FGF10-positive cases in tumor cells, 44 cases (37.6%) showed weak expression (1+) and 28 cases (23.9%) showed high expression (≥2+; 13 cases 2+, 15 cases 3+). In the stromal component, 43 cases (36.8%) showed weak expression (1+) and 32 cases (27.4%) demonstrated high expression (≥2+; 25 cases 2+, 7 cases with 3+). FGF10 expression in the stromal compartment was comparable to its expression in tumor cells, suggesting the importance of paracrine signaling within the tumor microenvironment. FGFR2 expression in tumor cells was observed in 114 of 117 cases (97.4%), with 29 cases (24.8%) showing weak expression (1+) and 85 cases showing high expression (≥2+) (72.6%; 49 cases 2+, 36 cases 3+). This extremely high expression frequency suggested an important role of FGFR2 in gastric adenocarcinoma.

When patients were divided into two groups based on the presence or absence of FGF10 expression in tumor cells (negative: 0, positive: 1+ to 3+), significant differences were observed in histological subtype (P=0.041, Fisher's exact test) and MIB-1 index (P=0.012, Mann-Whitney U test). A detailed analysis of histological subtypes revealed that moderately differentiated tubular adenocarcinoma (tub2) had a higher tendency for FGF10-positivity (30/37 cases, 81.1%), whereas poorly differentiated adenocarcinoma (por2) had a more balanced distribution (22/43 cases, 51.2%). The MIB-1 index was significantly higher in the FGF10-positive group compared with the FGF10-negative group (median: 45.7 vs. 37.1, respectively), indicating that FGF10 expression was associated with increased tumor cell proliferation.

We clarified the relationship between FGF10 and FGFR2 expression levels and the histopathological gland-forming differentiation pattern in gastric adenocarcinoma using a multivariable ordinal logistic regression model. Expression levels of FGF10 showed significant positive associations with histological classification in tumor cells and stromal tissue (Fig. 2). Multivariable ordinal logistic regression analysis, adjusted for age, HER2 status, and EGFR expression, revealed adjusted ORs of 1.749 (95% CI 1.231–2.487, P=0.002) for FGF10 expression in tumor cells (Fig. 2A) and 2.418 (95% CI 1.123–5.206, P=0.024) for stromal FGF10 expression (Fig. 2B). These findings indicate that increased FGF10 expression was associated with a higher probability of well-differentiated histological subtypes. This association was consistently observed in tumor cells and stromal tissue, with stromal FGF10 expression demonstrating a stronger association (higher adjusted OR). Subsequent analysis of FGFR2 found no significant association between FGFR2 expression levels and differentiation (P=0.788) (Fig. 3A).

A predicted probability plot is shown for the classification of pap, tub1, tub2, and por1 as intestinal/indeterminate type and por2 and sig as diffuse type. FGF10 expression in tumor cells had an adjusted OR in multivariable binary logistic regression analysis of 1.672 (95% CI 1.093–2.556, P=0.018), demonstrating a significant association with histological classification (Fig. 2C); however, FGF10 expression in the stroma (adjusted OR=1.782, 95% CI 0.724–4.386, P=0.209) (Fig. 2D) and FGFR2 expression in tumor cells (adjusted OR=0.831, 95% CI 0.31–2.227, P=0.713) showed no significant associations with histological classification (Fig. 3B). These results suggest that higher FGF10 expression levels were more strongly associated with the intestinal/indeterminate type. Furthermore, the association between FGF10 expression and histological classification was stronger in tumor cells than in the stroma.

Survival analysis revealed no significant associations between OS and the expression levels of FGF10 (in tumor cells and stroma) and FGFR2. In addition, Kaplan-Meier survival curve analysis (Fig. 4) showed no significant differences in survival rates in relation to the expression levels of each marker (log-rank test: FGF10 in tumor cells, P=0.596; FGF10 in stroma, P=0.806; FGFR2 in tumor cells, P=0.594). Multivariable Cox proportional hazards regression analysis, adjusted for age, sex, and disease stage, demonstrated no significant associations for FGF10 expression in tumor cells [hazard ratio (HR) 0.823, 95% CI 0.640–1.057, P=0.127], stromal FGF10 expression (HR 0.675, 95% CI 0.385–1.183, P=0.170), or FGFR2 expression in tumor cells (HR 1.080, 95% CI 0.559–2.085, P=0.819).

Assessment of multicollinearity indicated that all VIF values were close to 1.0 across all multivariable regression models, indicating no evidence of multicollinearity among the included covariates.