To ensure the reliability and representativeness of bioinformatics analyses of GPNMB’s role in GC, this study systematically extracted and screened GC-related data from three authoritative public databases while strictly adhering to predefined inclusion criteria.

First, data sources were selected to cover multi-dimensional validation needs: TCGA database provided the core Stomach Adenocarcinoma (STAD) dataset, encompassing GPNMB mRNA expression profiles and matched clinical follow-up information from 408 primary GC tissue samples and 211 adjacent normal tissue samples. This large sample size laid the foundation for initial differential expression analysis and prognostic modeling. To verify the consistency of findings across independent cohorts, three datasets (GSE26253, GSE62254, GSE54129) from the GEO database were included, focusing on cross-validation of GPNMB’s differential expression pattern between GC and normal tissues, as well as its correlation with patient survival trends. Additionally, the GEPIA database was utilized to confirm GPNMB’s status as a key differential gene in GC, leveraging its integrated pan-cancer expression data to help rule out database-specific biases.

To minimize confounding factors, strict inclusion criteria were applied to all bioinformatics datasets: Only samples from patients with pathologically confirmed primary GC were included; cases with incomplete clinical information or a history of preoperative chemotherapy/ radiotherapy were excluded. this ensured that the observed associations between GPNMB expression and GC characteristics were not confounded by prior treatment or data ambiguity, thereby enhancing the credibility of subsequent analytical results.

Differential expression analysis: R software was used to compare GPNMB mRNA expression between GC and normal tissues, with significance thresholds set at |log₂(fold change)|>1 and p < 0.05.

Prognostic analysis: Kaplan-Meier survival curves and univariate Cox regression were used to analyze the association between GPNMB expression and overall survival in TCGA and GEO datasets; median expression was used as the cutoff for high/low expression groups.

Pan-cancer expression analysis: GEPIA was used to generate pan-cancer GPNMB expression profiles, and Interactive Bodymap was used to visualize tissue-specific expression differences between tumor and normal samples.

Shanghai Outdo Biotech Company (Shanghai, China) provided the GC tissue microarray (HStmA180Su32). The time of specimen collection was 2015-01 to 2016-08, Final follow-up time was 2021-09. Initially, 80 pairs of GC and adjacent paracancerous tissue samples were collected; 5 cases were excluded due to incomplete data, and 2 due to tissue detachment, resulting in 73 valid pairs.

All procedures involving human participants complied with the Declaration of Helsinki. Written informed consent was obtained from each participant (covering tissue and clinical data usage), and the study was approved by the Ethics Committee of Shanghai Biotechnology Co., Ltd (Approval No. YB M-05-01).

Apparatus: Constant-temperature drying oven (PH-070A, Shanghai Yiheng), automatic dewaxing machine (LEICA ST5020, LEICA), antigen retrieval instrument (PT Link, Dako), automatic IHC staining system (Autostainer Link 48, Dako), centrifuge (1-14, Sartorius), Aperio scanner (Aperio XT, LEICA).

Reagents: EnVision™ FLEX+ Mouse Secondary Antibody Kit (K8002, Dako), Antibody Diluent with Background Reducing Components (S3022, Dako), Gene IHC Kit (GK800711, Gene Technology), GPNMB primary antibody (dilution at 1:25).

Paraffin-embedded gastric cancer tissue sections were baked at 63°C for 1 hour to melt the wax. Slides were dewaxed in xylene (2 × 15 minutes) and rehydrated through graded alcohols (2 × 100% alcohol for 7 minutes each, then 90%, 80%, 70% alcohol for 5 minutes each). Antigen retrieval was done using PT Link, followed by cooling in distilled water for more than 10 minutes. Slides were incubated with diluted GPNMB primary antibody at 4°C overnight, warmed to room temperature for 45 minutes, and washed with PBS. Then, they were processed in Autostainer Link 48 for blocking, secondary antibody binding, and DAB staining. Counterstaining was performed with hematoxylin (1 minute), differentiated with 0.25% hydrochloric acid alcohol (10 seconds), and rinsed with tap water (5 minutes). Finally, the slides were dried and mounted with neutral balsam.

Two pathologists, blinded to sample origin and patient outcomes, evaluated all images. Staining was scored using a semiquantitative system, considering both intensity (nuclear and membranous staining: 0 = blank, 1 = light yellow, 2 = yellow, 3 = brown) and extent of stained cells (0 = 0%, 1 = 1–24%, 2 = 25–49%, 3 = 50–74%, 4 = 75–100%).

Categorical variables: Gender (male/female), surgical approach (total gastrectomy/partial gastrectomy), tumor location (cardia/fundus vs. body/antrum/angle), pathological grade (I/II vs. III), HER2 status (negative/ positive), HP status (negative/positive), T stage (T1+T2 vs. T3+T4), N stage (N0 vs. N1+N2+N3), M stage (M0 vs. M1), TNM stage (I+II vs. III+IV). Continuous variables: Age, tumor size, CEA level.

(Kaplan-Meier OS curves (TCGA/GSE26253/GSE62254/ GSE54129); high GPNMB correlates with shorter OS (p < 0.05, HR = 1.499–1.7)).

Kaplan-Meier survival analysis across multiple datasets (TCGA, GSE26253, GSE62254) showed that high GPNMB expression correlated with shorter overall survival (Fig. 3A-D). For example, in TCGA-STAD, high GPNMB expression was associated with HR (1.499-1.7 and p=0.0011-0.015), confirming its potential as a prognostic indicator.

This heatmap uses TCGA data to show GPNMB’s expression correlations with other genes in gastric cancer (Fig. 4). Red/blue gradients reflect correlation strength: red indicates positive co-expression, and blue indicates negative correlation. It links GPNMB to immune checkpoints, T-cell markers, etc. These ties suggest that GPNMB may shape the tumor immune microenvironment-like connecting to immunosuppressive genes or anti-tumor T-cell signatures. Overall, this supports the potential role of GPNMB in driving immune evasion and affecting immunotherapy response in gastric cancer, pending further validation.

This set of figures contains 6 sub-figures. Figure 5A-D are violin box plots that group samples based on GPNMB expression and show differences in the distribution of different immune-related indicators. Figure 5E and F are scatter plot, displaying the linear associations between GPNMB expression and specific immune indicators. The P-values and correlation coefficients (r) are used to quantify the strength of these associations.

From the perspective of how immunotherapy works: If Figure 5A and C relate to PD-L1, the PD-L1 levels in the group with high GPNMB expression are usually much higher. This shows that GPNMB may boost PD-L1 production, which then prevents T cells from functioning properly and helps tumors evade the immune system. If Figure 5B and D involve Tumor Mutation Burden (TMB), there is no significant difference in TMB between the groups with high and low GPNMB expression. This means GPNMB does not affect how well immunotherapy works by changing the number of tumor mutations; it acts by regulating pathways related to immune checkpoints. Scatter plots in Figure 5E and F further confirm the linear relationships between GPNMB and immune indicators. Altogether, these findings support the idea that “high GPNMB expression can create an environment in the tumor that suppresses the immune system, making immunotherapy less effective”. They also provide clues for future research on using GPNMB as a marker to predict whether immunotherapy will work and to develop targeted treatments.

This set of Figure 6A-D explores the relationship between GPNMB expression and the tumor immune microenvironment. Figure 6A uses violin-box plots to show differences in three immune-related scores (StromalScore, ImmuneScore, ESTIMATEScore) between groups with low and high GPNMB expression. Figure 6B is a grouped scatter plot, displaying the infiltration levels of 28 immune cell subsets in groups with low and high GPNMB expression. Figure 6C is a correlation bar chart, ranking the correlation strength between GPNMB and 28 immunotherapy targets. Figure 6D is a partial correlation scatter plot, analyzing the linear association between GPNMB expression and the infiltration levels of 7 immune cell types while controlling for tumor purity.

GPNMB expression is closely associated with the composition and function of immune cells in the tumor microenvironment. From Figure 6B, high GPNMB expression selectively regulates the infiltration of immune cell subsets. For example, it may inhibit the infiltration of effector T cells and recruit immunosuppressive cell subsets such as regulatory T cells, thereby facilitating tumor immune evasion and immune surveillance. Figure 6D further confirms this. After adjusting for tumor purity, GPNMB shows different correlation patterns with various immune cells. It may promote the infiltration of immunosuppressive cells, like macrophages, and inhibit the infiltration of anti-tumor immune cells such as CD8 + T cells. These changes indicate that GPNMB can reshape the tumor immune microenvironment. By altering the composition of immune cell subsets, it creates an immunosuppressive environment, which may reduce the efficacy of immunotherapy. In summary, GPNMB affects the anti-tumor immune response by regulating the infiltration and function of immune cells, and it has potential as a marker for predicting immunotherapy outcomes.

(1) Staining intensity scoring: 0 (negative), 1 (1+), 2 (2+), 3 (3+).

(2) Staining positivity rate scoring: Raw percentage (0–100%).

(3) Total score: Calculated as the product of the staining intensity score and the staining positivity rate score.

(4) Survival analysis grouping: Patients were stratified into the low-expression group (GPNMB cytoplasmic total score ≤ 50) and high-expression group (GPNMB cytoplasmic total score > 50); multiple scoring methods and cutoff values were tested but showed no prognostic significance, so the median value was selected as the cutoff.

IHC staining (Figure 7A-D) showed that GPNMB was mainly expressed in the cytoplasm of GC cells: non-tumor tissues mostly had a score of 0, while GC tissues had scores of 1-3. Paired analysis (Figure 7E and Table 1) confirmed GPNMB IHC scores were significantly higher in GC tissues (57 ± 53) than adjacent non-cancerous tissues (38 ± 39, p = 0.018), consistent with bioinformatics results. The results show that the IHC scores for GPNMB in gastric cancer tissues are significantly higher than those in adjacent tissues, and the difference is statistically significant (P < 0.05), which further quantifies the trend of high expression of GPNMB in gastric cancer.

From Table 2, this study concludes that:

a) Demographic characteristics (gender, age), surgical and tumor features (surgical approach, tumor location, size, pathological grade), tumor size and lymph node metastasis (T stage, N stage), as well as molecular and infection indicators (HER2 expression, HP infection, CEA level) showed no statistically significant differences in distribution between the GPNMB low-expression and high-expression groups (p > 0.05).

b) Although statistical significance was not reached for certain features, numerical trends were observed: in terms of gender, the proportion of females in the GPNMB high-expression group (36.4%) was higher than that in the low-expression group (17.5%), approaching statistical significance (p = 0.068); regarding pathological grade, the proportion of poorly differentiated (Grade III) tumors in the GPNMB high-expression group (53.1%) was slightly higher than that in the low-expression group (38.9%).

c) The proportion of patients with stage M1 in the GPNMB high-expression group (24.2%) was significantly higher than that in the low-expression group (5.0%), with a statistically significant difference (χ²=5.663, p = 0.017 < 0.05), suggesting that GPNMB may be associated with the process of distant metastasis in gastric cancer.

Multivariate Cox regression analysis (adjusted for age, gender, TNM stage, etc.) showed that GPNMB expression was an independent prognostic factor for GC (HR = 1.52, 95% CI: 1.03-2.24, p = 0.035), confirming its prognostic value beyond clinical parameters.