Epigallocatechin gallate is associated with PDGFRB downregulation and altered PI3K-AKT signaling in gastric cancer

Abstract

Background

Gastric cancer (GC) remains a major cause of cancer-related mortality worldwide. Epigallocatechin gallate (EGCG), a natural polyphenol derived from green tea, exhibits anticancer properties; however, its molecular targets and regulatory mechanisms in GC are not fully elucidated. This study aimed to identify candidate EGCG-associated genes in GC and generate a hypothesis for future mechanistic investigation.

Methods

Differentially expressed genes (DEGs) in GC were identified and intersected with EGCG-associated targets retrieved from The Cancer Genome Atlas (TCGA) and GeneCards public databases. Least absolute shrinkage and selection operator (LASSO) regression and Cox proportional hazards analyses were performed to screen prognostically relevant genes. Diagnostic performance was evaluated using receiver operating characteristic (ROC) curves. Functional enrichment analysis was conducted to explore biological significance. Public single-cell RNA sequencing datasets were analyzed to determine the cellular localization of platelet-derived growth factor receptor beta (PDGFRB), while DepMap transcriptomic data were used to assess its expression across GC cell lines. In vitro assays, 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT), Transwell migration, and Western blotting, were performed to evaluate the biological effects of EGCG on GC-associated signaling pathways.

Results

Thirty-eight EGCG-associated DEGs were identified. Enrichment analysis revealed these genes were involved in cancer-associated pathways. LASSO-Cox modelling identified four candidate genes. Among them, PDGFRB was selected for further investigation based on its prognostic relevance and favorable diagnostic performance. PDGFRB expression was significantly higher in the TCGA genomically stable (GS) subtype than in the other molecular subtypes and was predominantly localized to cancer-associated fibroblasts (CAFs) and pericytes in single-cell RNA sequencing analysis. DepMap data demonstrated heterogeneous PDGFRB expression across GC cell lines. In vitro experiments showed that EGCG inhibited proliferation, migration, and invasion, reduced PDGFRB protein expression, and was associated with apoptosis-related protein changes and altered PI3K-AKT signaling.

Conclusions

Our findings suggest that EGCG treatment was associated with reduced PDGFRB expression and altered PI3K-AKT signaling in GC cells. These findings identify PDGFRB as a candidate EGCG-associated gene and provide a hypothesis for future mechanistic investigation.

Highlight box.

Key findings

• This study integrated bioinformatics analysis and in vitro experiments suggesting that epigallocatechin gallate (EGCG) may suppresses gastric cancer (GC) progression and is associated with platelet-derived growth factor receptor beta (PDGFRB) downregulation and altered PI3K-AKT signaling.

What is known and what is new?

• EGCG has been reported to exert anti-tumor effects in multiple cancers, and PDGFRB is known to participate in tumor progression and oncogenic signaling.

• This study identifies PDGFRB as a potential EGCG-related key gene in gastric cancer and provides preliminary evidence linking EGCG to the PDGFRB/PI3K-AKT axis.

What is the implication, and what should change now?

• These findings suggest that EGCG treatment was associated with reduced PDGFRB expression and altered PI3K-AKT signaling in GC cells. These findings identify PDGFRB as a candidate EGCG-associated gene and provide a hypothesis for future mechanistic investigation.

Introduction

Gastric cancer (GC) remains one of the most common and lethal malignancies of the digestive system worldwide. Despite a gradual decline in global incidence and mortality rates over recent decades, GC still ranks among the leading causes of cancer-related death, reflecting its substantial disease burden (1,2). Due to the absence of specific early symptoms, many patients are diagnosed at advanced stages, when therapeutic options are limited and prognosis is poor. Although systemic chemotherapy, targeted therapy, and immunotherapy have improved survival outcomes to some extent, their clinical efficacy is frequently constrained by toxicity, therapeutic resistance, and inter-patient heterogeneity (3,4).

Large-scale molecular profiling studies have demonstrated that GC is not a single disease entity but comprises multiple biologically distinct subtypes (5,6). The Cancer Genome Atlas (TCGA) classified GC into four molecular subtypes, including Epstein-Barr virus (EBV)-positive, microsatellite instability (MSI), genomically stable (GS), and chromosomal instability (CIN) tumors, each exhibiting unique genomic alterations and therapeutic vulnerabilities (7). These findings highlight that GC is a biologically heterogeneous disease and emphasize the importance of subtype-aware biomarker discovery and therapeutic target prioritization for precision oncology. Therefore, identifying molecular candidates associated with GC progression and treatment responsiveness remains an important strategy for generating hypotheses that may facilitate the development of more personalized therapeutic approaches.

Integrating transcriptomic profiling with drug-target prediction provides an effective strategy for prioritizing candidate compounds and molecular pathways with potential therapeutic relevance. Natural compounds, particularly those with established safety profiles, have attracted increasing interest as potential anticancer agents within this framework.

Epigallocatechin gallate (EGCG), a major bioactive catechin derived from green tea, has been extensively investigated because of its diverse biological activities, including antioxidant, anti-inflammatory, antiproliferative, and pro-apoptotic effects (8-10). Previous studies have shown that EGCG can suppress tumor cell proliferation, migration, and invasion by modulating diverse signaling pathways in ovarian cancer, renal cell carcinoma, and colorectal cancer (11-13). However, owing to its pleiotropic biological effects, the molecular mediators associated with EGCG in GC remain incompletely characterized, and systematic studies integrating bioinformatics analyses with experimental validation are still limited.

Platelet-derived growth factor receptor beta (PDGFRB) is a receptor tyrosine kinase that plays a critical role in regulating cell proliferation, migration, survival, and angiogenesis (14,15). Aberrant activation of PDGFRB has been implicated in tumor progression and unfavorable prognosis in several malignancies (16). Mechanistically, PDGFRB can activate downstream survival pathways, including the PI3K-AKT signaling cascade, thereby promoting tumor cell survival and resistance to apoptosis. Previous studies have shown that EGCG can modulate the PI3K-AKT signaling pathway in various cancer types (17,18). Recent single-cell transcriptomic studies further indicate that PDGFRB is predominantly expressed in stromal cell populations, particularly pericytes and cancer-associated fibroblasts (CAFs), underscoring its important role within the tumor microenvironment (19). Nevertheless, whether PDGFRB represents a molecular candidate associated with EGCG responsiveness in GC has not been systematically investigated.

In the present study, we adopted bioinformatics strategy to investigate molecular candidates associated with EGCG responsiveness in GC. By combining differential gene expression analysis, drug-target prediction, functional enrichment analysis, and prognostic modeling, PDGFRB was prioritized as a candidate gene with potential clinical relevance. To complement the computational findings, limited in vitro experiments were conducted as validation assays to evaluate the effects of EGCG on GC cell proliferation and apoptosis, as well as assessing changes in PDGFRB expression and PI3K-AKT signaling. Through this approach, we present this article to provide a data-driven framework that identifies PDGFRB as a candidate gene associated with EGCG in GC and generates a hypothesis for future mechanistic investigation. We present this article in accordance with the MDAR reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-0722/rc).

Methods

Data acquisition and bioinformatics analysis

The publicly available transcriptomic data and corresponding clinical information of GC patients databases were obtained from TCGA-stomach adenocarcinoma (TCGA-STAD) databases (normal tissue =36, tumor tissue =412) and Gene Expression Omnibus (GEO), GSE79973 and GSE19826 databases. The two GEO datasets included a total of 25 normal gastric tissue samples and 22 GC tissue samples. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.The flowchart is shown in Figure 1.

Figure 1.

The flowchart included TCGA-STAD and GSE79973 and GSE19826 datasets. DEGs, differentially expressed genes; EGCG, epigallocatechin gallate; GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; ROC, receiver operating characteristic; TCGA-STAD, The Cancer Genome Atlas-Stomach Adenocarcinoma.

After data preprocessing and normalization, differential gene expression analysis was performed to identify genes that were significantly dysregulated between GC tissues and normal gastric tissues. Raw count data were analyzed using R software (version 4.3.1) with the limma (3.56.2)-DESeq2 (1.40.2)-edgeR (3.42.4) packages (20-22). Genes with |log₂ fold change (FC)| ≥1 and false discovery rate (FDR) <0.05 were considered differentially expressed genes (DEGs). Volcano plots were generated using the EnhancedVolcano (1.24.0) package.

To enhance the robustness of DEGs identification and minimize method-specific bias arising from different statistical frameworks, only genes consistently identified as differentially expressed by all three algorithms were retained as the final DEGs set for subsequent analyses.

For downstream analyses, including functional enrichment analysis and expression visualization, transcript abundance values were normalized to transcripts per million (TPM). Differential expression analyses were performed exclusively on raw count data, whereas TPM-normalized data were used for visualization and biological interpretation.

To identify potential molecular targets of EGCG, EGCG-associated targets were retrieved from the GeneCards database (23). All targets with relevance scores above the recommended threshold (relevance score >1) were collected as predicted EGCG-related genes. The intersection between DEGs and EGCG-related targets was subsequently determined to identify candidate genes that were both dysregulated in GC and potentially targeted by EGCG. The overlapping genes were visualized using a Venn diagram generated with the VennDiagram package (1.7.3) in R.

Functional enrichment analysis was performed on the intersecting gene set to explore the underlying biological processes (BPs) and signaling pathways. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were conducted using the clusterProfiler package (4.8.3) in R (24). GO enrichment analysis was performed across three categories, including BP, cellular component (CC), and molecular function (MF). Significantly enriched GO terms and KEGG pathways were identified based on adjusted P values (P<0.05). As no custom background gene set was specified, the default background implemented in the GO and KEGG annotation databases was used for enrichment analysis.

To further narrow down key genes with prognostic relevance, least absolute shrinkage and selection operator (LASSO) regression analysis using the glmnet (4.1-8) package in R. This penalized regression approach was employed to reduce model complexity, prevent overfitting, and identify the most robust prognostic features among high-dimensional variables. To assess the robustness of LASSO-based feature selection, a bootstrap resampling strategy was applied. Briefly, the dataset was resampled with replacement for 1,000 iterations, and LASSO Cox regression with 10-fold cross-validation was independently performed in each bootstrap sample. Genes with non-zero coefficients at the optimal penalty parameter (λ_min) were recorded, and the selection frequency was calculated as the proportion of bootstrap iterations in which each gene was retained. Genes with non-zero coefficients were retained as key candidate genes.

Based on the regression coefficients derived from the LASSO-Cox model, a prognostic risk score was constructed for each patient using the following formula:

$$riskscore=exp\left({\displaystyle \sum \left(coefficient\times geneexpressionlevel\right)}\right)$$ [1]

Patients were stratified into high-risk and low-risk groups according to the median risk score. Kaplan-Meier survival analysis and log-rank tests were performed to compare overall survival (OS) between the two groups. The prognostic performance of the risk score was further evaluated using multivariate Cox regression analysis.

The expression levels of the identified key genes (PDGFRB, MAPK10, CAV1, and TIMP1) were extracted from the TCGA-STAD dataset. Gene expression values were log₂ transformed [log₂(TPM + 1)] for visualization. Differences in gene expression between GC tissues and normal tissues were visualized using boxplots and evaluated using the Wilcoxon test. To further explore the clinical relevance of these genes, their expression levels were analyzed in relation to clinicopathological characteristics. The associations between gene expression and clinical parameters were assessed using the Wilcoxon test. All statistical analyses were performed using R software.

To validate the robustness of the identified key genes, two independent GC datasets (GSE79973 and GSE19826) were obtained from the GEO database. Gene expression levels in tumor and normal tissues were compared to evaluate the consistency of expression patterns observed in the TCGA dataset.

In addition, receiver operating characteristic (ROC) curve analysis was performed to assess the diagnostic performance of the candidate genes in the validation datasets. The area under the curve (AUC) was calculated using the pROC (1.18.5) package in R, and ROC curves were generated to evaluate the ability of each gene to distinguish GC tissues from normal tissues.

Kaplan-Meier survival analysis were subsequently performed to evaluate the association between gene expression levels and OS in GC patients. OS was selected as the primary endpoint. Based on prognostic significance and biological relevance, PDGFRB was selected as the core gene for downstream experimental validation.

The single-cell RNA sequencing dataset GSE163558 was downloaded from the GEO database and included three primary GC samples (PT1–PT3). Data preprocessing, quality control, clustering, and cell-type annotation were performed using Seurat (5.0.3, v5). Uniform manifold approximation and projection (UMAP) was used for visualization. PDGFRB expression patterns were analyzed across annotated cell populations to determine its cellular distribution within the GC tumor microenvironment.

To further characterize baseline PDGFRB expression in GC cell lines, transcriptomic data were obtained from the DepMap Portal [Expression (Short-read) Public 26Q1 release]. PDGFRB expression values, reported as log₂(TPM + 1), were retrieved for the AGS and HGC-27 cell lines and used to compare their baseline expression levels.

Reagents and antibodies

Human GC cell lines AGS (Pricella, Wuhan, China) and HGC-27 (Pricella), phosphate-buffered saline (PBS), and complete culture media were obtained from Wuhan Procell Life Science & Technology Co., Ltd. EGCG (≥95%, reagent grade), dimethyl sulfoxide (DMSO), 0.25% trypsin, 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) reagent, Matrix-Gel™ Matrigel (standard type, phenol red–containing), and the bicinchoninic acid (BCA) protein assay kit were purchased from Beyotime Biotechnology (Shanghai, China). Transwell chambers were obtained from Corning Inc. (Corning, NY, USA). Prestained protein markers were purchased from Yeasen Biotechnology (Shanghai, China).

Primary antibodies against PDGFRB (1:1,000, Abclonal, Wuhan, China), BCL-2 (1:1,000, Abclonal, China), BAX (1:1,000, Abclonal, China), AKT (1:1,000, Abclonal, China), phosphorylated AKT (p-AKT, 1:1,000, Abclonal, China), and glyceraldehyde-3-phosphate dehydrogenase (GAPDH, 1:10,000, Abclonal, China) were purchased from Abclonal Technology Co., Ltd. (China). Primary antibodies against PI3K and phosphorylated PI3K (p-PI3K, 1:1,000, Absin, China) were obtained from Absin Bioscience Inc. (Shanghai, China).

MTT cell proliferation assay

Cell proliferation was assessed using the MTT assay (Beyotime, Shanghai, China). Briefly, 100 µL of cell suspension containing 2,000 cells was seeded into each well of a 96-well plate, with three replicate wells per group. After incubation at 37 ℃ for 24 h, cells were treated according to experimental design. Subsequently, 10 µL of MTT solution was added to each well and incubated for 4 h at 37 ℃. Then, 100 µL of formazan dissolution solution was added, followed by an additional 4 h incubation. The absorbance was measured at 570 nm using a microplate reader.

Transwell migration and invasion assays

Cell migration and invasion assays were performed using Transwell chambers with an 8.0 µm pore size (8-µm pore size; Corning). For migration assays, 5×10⁴ cells suspended in 200 µL serum-free medium were seeded into the upper chamber. For invasion assays, the upper chamber was precoated with Matrigel before cell seeding. In both assays, 500 µL of complete medium containing fetal bovine serum (FBS) was added to the lower chamber as a chemoattractant. After incubation for 24 h, non-migrated or non-invaded cells on the upper surface were gently removed with a cotton swab. Cells on the lower surface were fixed with paraformaldehyde and stained with crystal violet. Images were captured under a microscope, and cells were counted in three randomly selected fields.

Western blot analysis

Cells in the logarithmic growth phase were lysed using Racial and Identity Profiling Act (RIPA) (Beyotime, China) buffer supplemented with protease and phosphatase inhibitors. Total protein concentration was determined using the BCA assay Cells in the logarithmic growth phase were lysed using RIPA (Beyotime, China) buffer supplemented with protease and phosphatase inhibitors. Equal amounts of protein were separated by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) (5X, Beyotime, China) and transferred onto polyvinylidene difluoride (PVDF) membranes. Membranes were blocked with 5% non-fat milk at room temperature for 1 h and incubated with primary antibodies at 4 ℃ overnight. After washing with TBST, membranes were incubated with horseradish peroxidase (HRP)-conjugated secondary antibodies (1:2,000) at room temperature for 1 h. Protein bands were visualized using enhanced chemiluminescence (ECL, Beyotime, Shanghai, China) reagents and imaged accordingly. Statistical comparisons between groups were performed using Student’s t-test. Differences were considered statistically significant at P<0.05, as determined by western blot analysis.

Statistics analysis

All statistical analyses were performed using R software (version 4.3.1) unless otherwise specified. Each experiment above was conducted in at least three independent biological replicates. Unless otherwise specified, multiple-testing correction was performed using the Benjamini-Hochberg FDR method. The statistical procedures applied to individual analyses are described before. Random seed set [123].

Results

Identification of EGCG-related genes in GC

Differential expression analysis was performed using three independent analytical approaches. Differential expression analysis between GC tissues and normal gastric tissues identified a total of 2,158 DEGs (Figure 2A-2C). To explore potential molecular targets of EGCG with clinical relevance, these DEGs were intersected with EGCG-related predicted targets, resulting in 38 EGCG-associated DEGs (Figure 2D). The complete list of the 38 overlapping genes has been provided in Table S1.

Figure 2.

EGCG-related genes in gastric cancer. (A-C) Differential expression analysis of gastric cancer using three independent analytical methods. (D) Venn diagram showing the intersection between DEGs and EGCG-related genes. (E,F) GO and KEGG enrichment analyses of the intersecting genes. (G,H) LASSO regression analysis for screening prognostic genes. (G) LASSO coefficient profiles of candidate genes. (H) Cross-validation for selecting the optimal penalty parameter (λ). (I) Heatmap showing the expression patterns of genes included in the risk score model. (J) Kaplan-Meier OS analysis comparing high-risk and low-risk groups based on the median risk score. COVID-19, coronavirus disease 2019; DEGs, differentially expressed genes; EGCG, epigallocatechin gallate; FC, fold change; GO, Gene Ontology; IL, interleukin; KEGG, Kyoto Encyclopedia of Genes and Genomes; LASSO, least absolute shrinkage and selection operator; NS, not significant; OS, overall survival.

Functional enrichment analysis of EGCG-related DEGs in GC

After differential expression analysis of TCGA-STAD data, a total of 2,158 DEGs were identified. Intersection of these DEGs with EGCG-associated targets retrieved from the GeneCards database yielded 38 overlapping genes, which were considered potential EGCG-responsive genes in GC (Figure 2D).

To explore the biological functions and signaling pathways associated with these genes, GO and KEGG enrichment analyses were performed. GO enrichment analysis revealed that the 38 EGCG-associated DEGs were significantly enriched in BPs processes related to cellular stress responses and proliferation regulation, including response to light stimulus, response to ultraviolet (UV), response to radiation, muscle cell proliferation, and regulation of smooth muscle cell proliferation (Figure 2E). These terms suggest that EGCG-related genes may participate in stress-induced cellular responses and modulation of proliferative activity within the tumor microenvironment. KEGG enrichment analysis further demonstrated that these genes were mainly enriched in inflammation- and cancer-related signaling pathways, including hepatitis B, IL-17 signaling pathway, human T-cell leukemia virus 1 infection, coronavirus disease 2019 (COVID-19), and glioma (Figure 2F). Notably, several of these pathways are closely linked to chronic inflammation, aberrant immune responses, and oncogenic signaling, all of which are recognized contributors to gastric carcinogenesis and tumor progression.

However, these annotations likely reflect shared molecular components involved in cell proliferation, survival, inflammation, and extracellular matrix remodeling rather than disease-specific biological mechanisms. Collectively, these enrichment results suggest that EGCG-associated DEGs are participate in inflammation-related signaling and regulation of cell proliferation, providing a functional basis for the subsequent prioritization of candidate genes and supporting the hypothesis that EGCG may exert antitumor effects in GC through modulation of key signaling pathways. Detailed enrichment statistics, including gene counts, GeneRatio, BgRatio, and adjusted P values, are provided in Tables S2,S3.

Identification EGCG-related genes based on LASSO and Cox regression analyses

To further identify EGCG-related genes with potential prognostic significance in GC, the 38 overlapping genes derived from the intersection of DEGs and EGCG-associated targets were subjected to LASSO-COX feature selection. A 10-fold cross-validation procedure was performed to determine the optimal penalty parameter (λ), and the optimal value was selected according to the minimum cross-validation error (λ=0.0618). The coefficient profiles and cross-validation curves are shown in Figure 2G,2H.

As a result, four genes (PDGFRB, MAPK10, CAV1, and TIMP1) were retained in the final model and subsequently incorporated into a multivariable Cox regression model to construct the prognostic signature. The corresponding regression coefficients were 0.0594 (PDGFRB), 0.4623 (MAPK10), 0.0821 (CAV1) and 0.0993 (TIMP1), respectively. To further evaluate the robustness of the feature selection, a bootstrap resampling strategy was applied. Across 1,000 bootstrap iterations, all four candidate genes exhibited moderate to high selection frequencies, indicating that their inclusion in the model was stable and not driven by random sampling variability. These results support the reliability of the LASSO-based screening process.

Based on the expression levels and regression coefficients of these four genes, a risk score model was calculated for each patient. Patients were subsequently stratified into high-risk and low-risk groups according to the median risk score (Figure 2I). Survival analysis demonstrated that patients in the high-risk group exhibited significantly poorer OS compared with those in the low-risk group (logrank P=0.01), suggesting that the risk score model may have prognostic value in GC (Figure 2J).

To determine whether the risk score provided prognostic information independent of conventional clinicopathological characteristics, a multivariate Cox regression model incorporating age, sex, race, pathological grade, tumor-node-metastasis (TNM) stage, and the risk score was constructed. The results demonstrated that the risk score was significantly associated with OS, indicating that it may serve as an independent prognostic indicator for GC patients [hazard ratio (HR) =2.27, 95% confidence interval (CI): 1.329–3.88, P=0.003] (Table S4 and Figure S1). The proportional hazards assumption was satisfied for both the risk score (P=0.19) and the overall model (global P=0.16). The predictive performance of the multivariable prognostic model was further evaluated. The model achieved a concordance index (C-index) of 0.680. Time-dependent ROC analysis yielded AUC values of 0.702, 0.705, and 0.761 for predicting 1-, 3-, and 5-year OS (Figure S2), respectively, indicating good discriminative ability of the integrated prognostic model.

Expression validation of the four candidate genes

To further evaluate the expression patterns of the four candidate genes (PDGFRB, MAPK10, CAV1, and TIMP1), their expression levels were examined in the TCGA GC and adjacent normal tissues using log₂(TPM + 1) normalized RNA-seq data. PDGFRB, MAPK10, and TIMP1 exhibited significant differential expression between tumor and normal tissues (P<0.05), whereas CAV1 did not reach statistical significance (P>0.05) (Figure 3A).

Figure 3.

Validation of key genes, diagnostic performance, clinical relevance, and molecular docking analysis. (A) Boxplots showing the expression levels of four candidate genes (PDGFRB, MAPK10, CAV1, and TIMP1) in gastric cancer and normal tissues from the TCGA dataset. (B,C) Validation of gene expression in independent datasets from the Gene Expression Omnibus (GSE79973 and GSE19826). (D-G) ROC curves evaluating the diagnostic performance of PDGFRB, MAPK10, CAV1, and TIMP1 in the TCGA cohort. (H,I) ROC curves validating the diagnostic performance of the candidate genes in the external datasets (GSE79973 and GSE19826). (J,K) Kaplan-Meier OS curves showing the prognostic significance of PDGFRB and TIMP1 in gastric cancer patients. (L,M) Boxplots illustrating the association between PDGFRB expression and clinicopathological characteristics, including tumor stage and T stage. (N) PDGFRB expression iin molecula subtype. (O-Q) Cell annotation and PDGFRB expression. **, P<0.01; ***, P<0.001; ****, P<0.0001; ns, not significant. AUC, area under the curve; CAF, cancer-associated fibroblast; CI, confidence interval; CIN, chromosomal instability; EBV, Epstein-Barr virus; GS, genomically stable; MSI, microsatellite instability; NK, natural killer; ROC, receiver operating characteristic; OS, overall survival; TCGA, The Cancer Genome Atlas.

To assess the reproducibility of these findings, two independent datasets (GSE79973 and GSE19826) from the GEO were analyzed. In the GSE79973 dataset, the expression levels of PDGFRB and TIMP1 were significantly different between tumor and normal tissues (P<0.05), whereas CAV1 and MAPK10 did not show significant differences (P>0.05) (Figure 3B). Similarly, in the GSE19826 dataset, PDGFRB and TIMP1 remained significantly differentially expressed (P<0.05), while CAV1 and MAPK10 again showed no significant differences (P>0.05) (Figure 3C). These results indicate that PDGFRB and TIMP1 exhibited the most reproducible expression alterations across independent transcriptomic datasets.

Discriminative performance of candidate gene expression

ROC curve analysis was performed to evaluate the ability of candidate genes expression to distinguish tumor tissues from adjacent normal tissues in transcriptomic datasets. In the TCGA dataset, TIMP1 exhibited the highest AUC (0.941), followed by PDGFRB (AUC =0.869), while MAPK10 (AUC =0.657) and CAV1 (AUC =0.578) showed relatively limited discriminative performance (Figure 3D-3G). In GSE79973, the AUC values were 0.97 for TIMP1, 0.89 for PDGFRB, 0.62 for MAPK10, and 0.52 for CAV1 (Figure 3H). Similarly, in GSE19826, the AUC values were 0.917 for TIMP1, 0.944 for PDGFRB, 0.578 for MAPK10, and 0.547 for CAV1 (Figure 3I).

These analyses demonstrate that TIMP1 and PDGFRB exhibit reproducible expression differences between tumor and normal tissues across multiple transcriptomic cohorts. However, these ROC analyses were performed using retrospective tissue expression datasets and therefore should not be interpreted as evidence of clinical diagnostic utility. Rather, they support the robustness of differential expression and provide additional rationale for subsequent prognostic and functional investigation.

Prognostic value of key genes

Kaplan-Meier survival analysis was performed to evaluate the association between candidate gene expression and OS. The results demonstrated that high expression of PDGFRB was significantly associated with poorer OS in GC patients (P=0.03). In contrast, TIMP1 did not show a statistically significant association with OS (P=0.13) (Figure 3J,3K). Considering that prognostic relevance is more closely aligned with the objective of identifying candidate genes associated with GC progression than tumor-versus-normal discrimination alone, PDGFRB was prioritized for subsequent mechanistic analyses.

To further investigate the clinical heterogeneity of PDGFRB, its expression was analyzed in relation to tumor stage and T stage. Stage analysis revealed that PDGFRB expression was significantly higher in stage II and stage III tumors compared with stage I (P<0.05), while no significant difference was observed between stage IV and stage I (P>0.05). In addition, T stage analysis demonstrated that PDGFRB expression was significantly elevated in T2, T3, and T4 tumors compared with T1 tumors (P<0.05), suggesting that higher PDGFRB expression may be associated with tumor advanced local tumor invasion (Figure 3L,3M). However, no significant differences in PDGFRB expression were observed among different groups stratified by tumor grade, lymph node metastasis (N stage), distant metastasis (M stage), or gender (all P>0.05) (Figure S3). We also evaluated its distribution across TCGA molecular subtypes. Kruskal-Wallis analysis revealed significant differences in PDGFRB expression among TCGA molecular subtypes of GC (P=0.001) (Figure 3N). Subsequent Dunn’s post hoc test demonstrated that PDGFRB expression was significantly elevated in the GS subtype compared with the CIN (adjusted P=0.002), EBV (adjusted P=0.03), and MSI (adjusted P=0.002) subtypes. No significant differences were observed among CIN, EBV, and MSI tumors. These findings suggest a subtype-specific enrichment of PDGFRB in the GS subtype (Table S5). These findings indicate that PDGFRB expression is subtype-associated rather than uniformly elevated across GC.

Cellular origin of PDGFRB expression revealed by single-cell RNA sequencing

Given that PDGFRB is expressed by multiple stromal cell populations, including fibroblasts and pericytes, the bulk transcriptomic findings were further investigated using single-cell RNA sequencing to characterize the cellular source of PDGFRB expression within the gastric tumor microenvironment.

To further investigate the cellular source of PDGFRB expression in GC, a publicly available single-cell RNA sequencing dataset (GSE163558) was analyzed. Cell clustering and annotation identified major cell populations within the tumor microenvironment (Figure 3O). The identities of CAFs, pericytes, and epithelial cells were further confirmed by the expression of established marker genes, as shown in Figures S4-S6. PDGFRB expression was predominantly enriched in CAFs and pericytes, whereas little expression was detected in epithelial and malignant epithelial cell populations (Figure 3P,3Q). These findings indicate that the elevated PDGFRB expression observed in bulk transcriptomic datasets is likely influenced, at least in part, by stromal-cell abundance within the tumor microenvironment.

To complement the single-cell findings, PDGFRB expression in GC cell lines was further evaluated using DepMap transcriptomic data. HGC-27 cells exhibited substantially higher PDGFRB expression than AGS cells [log₂(TPM + 1): 2.24 vs. 0.27], indicating that PDGFRB expression varies across GC cell models and is retained in a subset of tumor cells (Table S6, Figure S7). Based on these observations, AGS and HGC-27 cells were selected for subsequent in vitro experiments to further evaluate the biological effects of EGCG and its association with PDGFRB expression and downstream signaling.

EGCG suppresses proliferation of GC cells in a dose-dependent manner

To investigate the effect of EGCG on GC cell proliferation, human gastric adenocarcinoma cell lines AGS and HGC-27 were treated with increasing concentrations of EGCG for 24 hours, followed by assessment using the MTT assay. A series of concentration gradients was applied for each cell line to evaluate dose-response effects.

The results demonstrated that EGCG significantly inhibited the proliferation of both AGS and HGC-27 cells compared with untreated controls. Notably, the inhibitory effect of EGCG increased progressively with rising concentrations, indicating a clear dose-dependent antiproliferative effect (P<0.05). Based on the dose–response curves, the half-maximal inhibitory concentration (IC₅₀) of EGCG was calculated to be 54.6 µmol/L for HGC-27 cells and 181.89 µmol/L for AGS cells (Figure 4A,4B).

Figure 4.

Effects of EGCG on proliferation, migration, invasion, apoptosis, and PI3K-AKT signaling in gastric cancer cells. (A,B) Cell viability analysis of gastric cancer cells treated with EGCG using the MTT assay. (C-F) Transwell assays evaluating the effects of EGCG on the migration and invasion abilities of gastric cancer cells, stained with 0.1% crystal violet, and counted under a microscope at 200× magnification. (G) Representative Western blots showing BAX and BCL-2 protein expression in HGC-27 and AGS cells following EGCG treatment. (H) Quantification of the BAX/BCL-2 ratio in HGC-27 cells. (I) Quantification of the BAX/BCL-2 ratio in AGS cells. (J-Q) Western blot analysis of proteins involved in the PI3K-AKT signaling pathway, including PI3K, phosphorylated PI3K (p-PI3K), AKT, and phosphorylated AKT (p-AKT), after EGCG treatment. Data are presented as mean ± SD from three independent experiments. P<0.05 was considered statistically significant. NC, Vehicle (DMSO). *, P<0.05; **, P<0.01; ns, not significant. DMSO, dimethyl sulfoxide; EGCG, epigallocatechin gallate; GAPDH, glyceraldehyde-3-phosphate dehydrogenase; NC, negative control; SD, standard deviation.

To minimize cytotoxic effects while preserving biological relevance in subsequent functional experiments, EGCG concentrations of 30 µmol/L for HGC-27 cells and 90 µmol/L for AGS cells were selected for further analyses.

EGCG inhibits migration and invasion of GC cells in vitro

Given the observed antiproliferative effects of EGCG, we next examined whether EGCG influences the migratory and invasive capacities of GC cells. After treatment with EGCG at the selected concentrations for 24 hours, Transwell migration and invasion assays were performed.

Compared with the control groups, EGCG-treated HGC-27 and AGS cells exhibited a significant reduction in both migratory and invasive abilities (P<0.05). Quantitative analysis confirmed that EGCG markedly suppressed the number of cells traversing the Transwell membrane, indicating that EGCG effectively impairs the metastatic potential of GC cells in vitro (Figure 4C-4F).

EGCG modulates apoptosis-related protein expression in GC cells

To investigate whether EGCG treatment was associated with alterations in apoptosis-related proteins, Western blot analysis was performed in HGC-27 and AGS cells following 24 h of EGCG treatment. Based on the IC50 results, HGC-27 and AGS cells were treated with 30 and 90 µmol/L EGCG, respectively. Western blot analysis demonstrated that EGCG treatment increased BAX protein expression while decreasing BCL-2 expression in both HGC-27 and AGS cells compared with the vehicle control (Figure 4G). Quantitative analysis showed that the BAX/BCL-2 ratio was significantly increased following EGCG treatment in both HGC-27 (P=0.02) and AGS cells (P=0.002) (Figure 4H,4I). These findings indicate that EGCG treatment was associated with a shift toward a pro-apoptotic protein expression profile.

EGCG modulates PDGFRB expression and PI3K-AKT signaling in GC cells

To further investigate molecular changes associated with EGCG treatment, Western blot analysis was performed to evaluate PDGFRB expression and the PI3K-AKT signaling pathway in HGC-27 and AGS cells.

The results showed that EGCG treatment led to a significant reduction in PDGFRB protein expression in both HGC-27 and AGS cells compared with the vehicle control (Figure 4J,4K). To further investigate molecular changes associated with EGCG treatment, representative Western blot images of phosphorylated and total PI3K and AKT proteins in HGC-27 and AGS cells are shown in Figure 4L,4M. To evaluate pathway activity, the phosphorylation levels of PI3K and AKT were normalized to their corresponding total protein levels. In HGC-27 cells, EGCG treatment significantly reduced both the p-PI3K/PI3K ratio (P=0.03) and the p-AKT/AKT ratio (P=0.02) (Figure 4N,4O), indicating decreased PI3K-AKT pathway activity. In AGS cells, the p-PI3K/PI3K ratio was not significantly altered following EGCG treatment (Figure 4P), whereas the p-AKT/AKT ratio showed a modest but statistically significant reduction (P=0.046) (Figure 4Q). These findings suggest that the effect of EGCG on PI3K-AKT signaling may differ between GC cell lines and appears to be more pronounced in HGC-27 cells.

Taken together, EGCG treatment was associated with reduced PDGFRB protein expression and alterations in PI3K-AKT signaling. These observations are consistent with the bioinformatics analyses identifying PDGFRB as a candidate gene associated with EGCG responsiveness. However, because PDGFRB perturbation, rescue experiments, and direct target-engagement assays were not performed, the present findings do not establish a causal relationship between PDGFRB downregulation and PI3K-AKT pathway modulation.

Discussion

In the present study, we investigated bioinformatics analyses with limited in vitro validation to investigate molecular candidates associated with EGCG responsiveness in GC. Among four candidate genes identified by LASSO-Cox analysis, PDGFRB demonstrated prognostic relevance and was therefore selected for further investigation. EGCG suppresses cell proliferation, migration, and invasion of GC cells, and was associated with reduced PDGFRB expression, increased apoptotic signaling, and alterations in PI3K-AKT signaling. Although both PDGFRB and TIMP1 exhibited relatively high performance in ROC analysis, survival analysis stratified by median expression revealed that only PDGFRB was significantly associated with OS, whereas TIMP1 did not reach statistical significance. Collectively, these findings identify PDGFRB as a candidate gene associated with EGCG responsiveness and generate a hypothesis for future mechanistic investigation rather than establishing a causal signaling axis.

EGCG, the major catechin component of green tea, is a natural polyphenolic compound that has attracted extensive attention due to its diverse biological activities (5). However, the clinical application of EGCG is partly limited by its relatively low oral bioavailability (25). After oral administration, EGCG is primarily metabolized by intestinal microbiota. Previous studies have demonstrated that EGCG can be metabolized by several gut microorganisms, including Raoultella planticola and Klebsiella pneumoniae (26). These metabolic processes may significantly reduce the systemic absorption of EGCG, resulting in limited bioavailability following oral administration. Although nanotechnology-based delivery systems have been explored to enhance the stability and bioavailability of EGCG, the relatively low absorption efficiency remains one of the potential obstacles to its broader therapeutic application (27,28). These findings indicate that EGCG exerts pleiotropic biological activities through multiple molecular pathways rather than a single molecular target.

Despite this limitation, accumulating evidence from numerous preclinical studies—including in vitro experiments, in vivo animal models, and clinical investigations—has highlighted the wide range of biological and pharmacological properties of polyphenolic compounds, particularly EGCG. These activities include antimicrobial, antioxidant, anti-inflammatory and antihypertensive effects (29-31). More importantly, EGCG has been widely reported to exhibit significant anti-tumor activity in various types of cancers (9,32,33). Previous studies have shown that EGCG can suppress tumor invasion and migration through multiple mechanisms. These include the induction of autophagy, lysosomal membrane permeabilization-mediated cell death, as well as both caspase-dependent and caspase-independent apoptotic pathways (34). In addition, EGCG exerts its anti-tumor effects through the regulation of several key signaling pathways involved in cancer progression, including the JAK-STAT signaling pathway, MAPK signaling pathway, PI3K-AKT signaling pathway, Wnt signaling pathway, and Notch signaling pathway (35-37). Furthermore, EGCG has also been reported to modulate the expression of extracellular matrix regulatory proteins such as Tissue inhibitors of metalloproteinases and matrix metalloproteinases, which play critical roles in tumor invasion, metastasis, and tumor microenvironment remodeling. Regulation of these molecules may contribute to the inhibitory effects of EGCG on cancer progression (38).

In vitro experiments demonstrated that EGCG significantly inhibited the proliferation, migration, and invasion of AGS and HGC-27 GC cells in a dose-dependent manner. Consistent with bioinformatics predictions, EGCG treatment was associated with reduced PDGFRB protein expression. Additionally, EGCG exposure was associated with altered expression of apoptosis-related proteins, including increased BAX and decreased BCL-2, suggesting activation of apoptotic associated molecular changes.

PDGFRB is a receptor tyrosine kinase that plays an important role in regulating several critical cellular processes, including cell growth, proliferation, invasion, migration and survival (39,40). Increasing evidence indicates that abnormal activation of PDGFRB contributes to tumor progression and remodeling of the tumor microenvironment. Previous studies have demonstrated that blockade of PDGFRα/β signaling in combination with immune checkpoint inhibition can exert synergistic anti-tumor effects. For instance, combined inhibition of PDGFRα/β and anti-PD-1 therapy has been reported to significantly suppress the growth of fibrotic tumors. These findings highlight the importance of stromal reprogramming in enhancing anti-tumor immune responses and suggest that combination immunotherapy strategies may represent a promising therapeutic approach for patients with fibrosis-associated tumors (41). Furthermore, experimental studies have shown that continuous irradiation of xenograft tumors treated with Imatinib for 14 days resulted in a marked reduction in both PDGFRB and phosphorylated PDGFRB levels, with decreases of approximately 5.3-fold and 4-fold, respectively. Interestingly, the expression of PDGFRB was found to be upregulated in human tumor tissues, suggesting that stromal damage may activate autocrine survival signaling pathways (42). Taken together, these findings suggest that PDGFRB plays a crucial role in tumor progression and stromal remodeling. These findings support the biological plausibility of investigating PDGFRB in GC but do not establish PDGFRB as the principal mediator of EGCG activity.

Our in vitro experiments demonstrated that EGCG treatment suppressed PDGFRB protein expression and attenuated the activation status of the PI3K-AKT pathway, accompanied by increased apoptotic signaling in GC cells. Given that PDGFRB is a known upstream regulator of PI3K-AKT signaling in multiple cellular contexts, it is plausible that EGCG-induced inhibition of PDGFRB may contribute to the observed suppression of survival signaling (43).

An important observation from the present study is that PDGFRB expression was not uniformly distributed across GC molecular subtypes. Additional analysis revealed significantly higher PDGFRB expression in the TCGA GS subtype than in CIN, EBV, and MSI tumors. The GS subtype has been characterized by diffuse histology, epithelial-mesenchymal transition (EMT), stromal enrichment, and extensive tumor-microenvironment interactions. Likewise, the ACRG molecular classification identified the MSS/EMT subtype as a biologically aggressive subgroup associated with extracellular matrix remodeling, fibroblast activation, and poor clinical outcomes (44). These findings suggest that the biological significance of PDGFRB may be particularly relevant in stromal-rich GC subtypes. Recent studies have further emphasized the critical role of stromal heterogeneity in shaping GC progression and immune regulation. For example, Jang et al. (45) demonstrated substantial spatial and microenvironmental heterogeneity across GC ecosystems, highlighting the importance of stromal components in tumor progression and immune regulation.

Recent evidence suggests that the biological interpretation of candidate therapeutic targets should extend beyond transcriptomic associations alone. Although PDGFRB was identified through integrative transcriptomic analyses and subsequently validated in vitro, emerging proteogenomic studies have demonstrated that alterations at the transcript level do not necessarily translate into corresponding changes in protein abundance, phosphorylation status, or therapeutic vulnerability. Chang et al. (46,47) highlighted the importance of integrating genomic, transcriptomic, proteomic, and phosphoproteomic information when prioritizing candidate therapeutic targets across human cancers. Therefore, the present findings should be interpreted as identifying PDGFRB as a candidate EGCG-associated mediator rather than a definitively validated therapeutic target.

Our additional single-cell analysis further demonstrated that PDGFRB expression was predominantly localized to CAFs and pericytes, whereas only minimal expression was observed in malignant epithelial cells. These findings are consistent with previous GC single-cell studies. Sathe et al. (48) identified PDGFRB as a characteristic marker of pericyte populations and demonstrated extensive stromal-cell reprogramming during gastric tumorigenesis. Similarly, Kang et al. (19) reported that PDGFRB was enriched in angiogenesis-associated pericyte subsets and extracellular-matrix-remodeling stromal populations, suggesting a role in vascular remodeling, growth-factor signaling, and tumor invasion. More recently, Jia et al. (49) demonstrated that specific CAF populations contribute to tumor progression and therapeutic resistance in GC, further emphasizing the biological importance of stromal-derived signaling within the tumor microenvironment. Wang et al. (5) demonstrated that stromal cell abundance is closely associated with immune exclusion and activation of multiple immunosuppressive pathways within the GC microenvironment. Given the predominant localization of PDGFRB in CAFs and pericytes observed in our single-cell analysis, it is conceivable that PDGFRB may participate in the regulation of stromal-immune interactions, extracellular matrix remodeling, and angiogenic processes that collectively contribute to tumor progression. Accordingly, PDGFRB may function not only as a tumor-cell-associated signaling molecule but also as a stromal-associated regulator involved in extracellular matrix remodeling, angiogenesis, and tumor-stroma interactions.

Importantly, analysis of DepMap datasets revealed substantial heterogeneity of PDGFRB expression among GC cell lines, with HGC-27 cells exhibiting markedly higher expression than AGS cells. Thus, although PDGFRB expression in clinical tumors appears to be largely stromal-derived, a subset of GC cells may retain biologically relevant PDGFRB expression. Consistent with this observation, EGCG treatment reduced PDGFRB expression and inhibited PI3K/AKT signaling in both HGC-27 and AGS cells. Previous studies have demonstrated that activation of PDGFR signaling promotes GC progression through the PI3K/AKT pathway. Notably, Li et al. (50) reported that GIPC1-mediated activation of PDGFR signaling enhanced GC cell proliferation and migration via PI3K/AKT activation. In agreement with these findings, our data suggest that suppression of PDGFRB-associated PI3K/AKT signaling may contribute to the antitumor effects of EGCG.

It is important to distinguish candidate biomarker discovery from therapeutic-target validation and mechanistic investigation. Although PDGFRB demonstrated differential expression, prognostic relevance, and responsiveness to EGCG treatment in our study, these findings alone do not establish its clinical utility as a diagnostic biomarker nor definitively validate it as a therapeutic target. Contemporary GC research increasingly emphasizes molecular subtyping, multiomic integration, and independent cohort validation as essential requirements for biomarker development (7,44). Large-scale studies, including those from TCGA and subsequent molecular-classification frameworks, have demonstrated substantial biological heterogeneity among GCs and highlighted the importance of integrating genomic, transcriptomic, proteomic, and tumor-microenvironment information when evaluating candidate biomarkers and therapeutic targets (51).

Nevertheless, the present study does not establish direct target engagement or causality between EGCG and PDGFRB. Future studies incorporating genetic perturbation approaches, rescue experiments, phosphoproteomic analyses, proteogenomic integration, and direct target-engagement assays will be necessary to determine whether PDGFRB represents a bona fide therapeutic target mediating EGCG-associated antitumor activity in GC.

Despite these promising biological effects, the clinical translation of EGCG remains challenging due to its limited stability, low oral bioavailability, rapid metabolism, and insufficient tissue exposure. Although nanotechnology-based delivery systems and formulation optimization strategies have been explored to improve EGCG stability and bioavailability, whether these approaches can achieve therapeutically relevant concentrations in GC patients remains to be determined. Therefore, the findings of the present study should be considered as a hypothesis-generating framework for understanding EGCG-associated molecular responses rather than direct evidence supporting EGCG-based therapy for GC.

Several limitations of this study should be acknowledged. First, although an association between EGCG treatment, PDGFRB downregulation, and PI3K-AKT signaling changes was observed, direct mechanistic evidence linking PDGFRB to PI3K-AKT modulation in this context remains lacking. Second, the experimental validation was limited to in vitro assays and no in vivo validation, clinical tissue verification, or direct target-engagement assays were performed. In vivo studies are required to further substantiate the biological relevance of these findings. Finally, EGCG is known to exert pleiotropic effects, and the possibility that additional targets or pathways contribute to its antitumor activity cannot be excluded.

Future studies incorporating PDGFRB-specific knockdown or overexpression, as well as pathway rescue experiments, will be essential to clarify whether PDGFRB acts as a functional mediator of EGCG-induced antitumor effects. Moreover, in vivo validation and clinical correlation analyses may further support the translational potential of PDGFRB as a therapeutic target.

Conclusions

Overall, these results indicate that EGCG treatment is associated with reduced PDGFRB protein expression and alterations in PI3K-AKT signaling in GC cells. Together with the bioinformatics analyses, these findings support PDGFRB as a candidate gene associated with EGCG responsiveness and generate a hypothesis for future validation of its potential role in mediating EGCG-associated antitumor effects in GC.

Supplementary

The article’s supplementary files as

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Data Sharing Statement

Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-0722/dss

tcr-15-08-625-dss.pdf ^{(66.4KB, pdf)}

DOI: 10.21037/tcr-2026-0722