Comorbidity of diabetes mellitus (DM) and gastric cancer (GC) represents a clinical challenge because metabolic dysregulation and tumor biology can interact to worsen outcomes. Epimedium, a traditional botanical with reported anti-tumor and metabolic regulatory properties, has been proposed as a candidate therapeutic for this comorbid state. The precise molecular mechanisms linking Epimedium activity to pathways common to DM and GC, however, remained unclear. This study aimed to identify shared molecular targets between DM and GC and to determine whether Epimedium can modulate those targets, using an integrated computational and experimental pipeline.
The investigators first integrated bulk transcriptomic datasets for DM and GC to identify common differentially expressed genes (DEGs) between the two conditions. From that set of shared DEGs, machine learning algorithms were applied to prioritize core genes implicated across both disease contexts. The machine learning analysis was complemented by SHapley Additive exPlanations (SHAP) analysis to improve interpretability and to isolate the most influential candidate genes within the models. The resulting prioritized genes were then considered for overlap with predicted molecular targets of Epimedium.
Predicted molecular targets of Epimedium were intersected with the machine learning–derived hub gene candidates to generate a focused list of putative Epimedium-responsive genes that also link DM and GC. From the integrated analysis, two genes—GABRA1 and IGHG1—emerged as primary hub genes for further evaluation based on the combined transcriptomic and machine learning evidence.
To refine target selection at cellular resolution, single-cell RNA sequencing (scRNA-seq) analysis was performed on gastric cancer tissue and matched normal tissue. The scRNA-seq data revealed differential cell-type expression patterns among candidate genes. IGHG1 showed significant overexpression in GC tissue compared with normal tissue and was notably enriched within the B-cell population. In contrast, GABRA1 demonstrated insufficient expression in the single-cell datasets for meaningful further analysis, and thus was deprioritized in the single-cell context.
Cellular experiments were conducted using the human HGC-27 gastric cancer cell line to test Epimedium's functional effects and to validate target modulation. Cell viability assays (CCK-8) demonstrated that Epimedium inhibited HGC-27 cell viability in a dose-dependent manner. The reported half-maximal inhibitory concentration (IC50) for Epimedium in this cell line was 632.3 μg/ml.
Gene and protein expression assays were used to assess target engagement. Quantitative PCR (qPCR) showed downregulation of IGHG1 transcript levels following Epimedium treatment. Western blot analysis confirmed a corresponding decrease in IGHG1 protein expression. These in vitro findings indicate that Epimedium reduces gastric cancer cell viability while decreasing IGHG1 expression at both mRNA and protein levels.
The combined computational and experimental evidence supports IGHG1 as a candidate molecular bridge linking DM and GC pathobiology that is responsive to Epimedium. The study's integration of bulk transcriptomics, machine learning with interpretable SHAP outputs, predicted herbal targets, and cell-type resolution via scRNA-seq strengthened the biological plausibility of IGHG1 as a key node. In vitro experiments in HGC-27 cells demonstrated that Epimedium not only reduces tumor cell viability but also suppresses IGHG1 expression, consistent with the hypothesis that Epimedium's anti-tumor effects in this comorbidity operate, at least in part, through modulation of IGHG1.
Within the scope of the reported work, IGHG1 was identified as a central shared gene in DM and GC and as a putative actionable target of Epimedium. The study combined bioinformatics integration, interpretable machine learning, and single-cell resolution with cellular assays to show that Epimedium decreases HGC-27 viability and downregulates IGHG1 expression. These results support continued investigation of IGHG1 as a therapeutic target in patients with concurrent diabetes and gastric cancer and provide a rationale for additional mechanistic and in vivo studies. The abstract does not provide exhaustive methodological parameters or broader preclinical/clinical safety data; readers should consult the full text for complete experimental details.