Eur J Pharm Sci. 2026 May 16. pii: S0928-0987(26)00131-4. [Epub ahead of print]
107557
PURPOSE: This study aimed to reveal the mechanism and prognostic significance of migrasome-related genes (MGs) in primary liver cancer (PLC) treated with Sorafenib (Sor) treatment.
METHODS: Datasets were extracted from an online database, and differentially expressed genes (DEGs) were explored. WGCNA was used to detect co-expressed gene modules. Next, MGs and differentially expressed MGs were integrated with DEGs from WGCNA. Prognostic genes were selected using univariate Cox regression and a risk model was developed using 10 types of machine learning methods. The model was validated, followed by immune infiltration, drug sensitivity, and genomic alteration analyses. The correlation between the risk model and the clinical features was assessed using Cox regression and survival analyses. In vitro experiments (qRT‑PCR, Western blot, CCK‑8, and flow cytometry) and in vivo subcutaneous tumor models in nude mice were conducted to validate the biological functions of key genes.
RESULTS: In total, 1,417 DEGs were identified in the TCGA dataset. WGCNA revealed two key modules that strongly correlated with Sor response. Univariate Cox regression analysis identified 83 prognostic genes associated with PLC that were mainly enriched in pathways, such as IL-17 signaling. A prognostic model was built based on seven signature genes (G6PD, KIF20A, IGSF3, BSG, ITGA5, EOGT, and PIGK). The model achieved a C-index of 0.682, demonstrating a strong predictive performance. Survival analysis indicated significantly better survival in the low-risk group than in the high-risk group, with ROC AUC values of 0.708, 0.773, and 0.795 at 1, 3, and 5 years, respectively, indicating good prognostic value. Immune infiltration showed significant associations between the high-risk group and regulatory T cells/M0 macrophages as well as poorer responses to immunotherapy. In vitro experiments showed that the knockdown of G6PD or KIF20A suppressed PLC cell proliferation and promoted apoptosis. In vivo, silencing G6PD or KIF20A significantly inhibited tumor growth in nude mice and suppressed the activation of the IL‑17/AKT/STAT3 signaling pathway.
CONCLUSIONS: This study established a prognostic model based on seven MGs in PLC, correlating them with patient prognosis and Sor treatment response, and provided preliminary evidence that G6PD and KIF20A may promote PLC progression through the IL‑17/AKT/STAT3 pathway.
Keywords: Hepatocellular carcinoma; Machine learning; Migrasomes; Prognostic model; Sorafenib; Validation analysis