Establishment and validation of a prognostic model based on liquid-liquid phase separation-related genes in gastric cancer.

[BACKGROUND] Growing evidence suggests that the imbalance of liquid-liquid phase separation (LLPS) can alter the spatiotemporal coordination ability of biomolecular condensates, thereby playing an important role in carcinogenesis and cachexia. Gastric cancer (GC), ranking as the fifth most prevalent malignancy globally, remains lacking in systematic analysis at the GC-LLPS level within current research. This study aims to identify differentially expressed LLPS-related genes (LLPSGs) in GC and elucidate the role of LLPS in the initiation and progression of GC. Identifying the role of LLPS in carcinogenesis facilitates the development of personalized treatment strategies.

[METHODS] The Cancer Genome Atlas of Stomach Adenocarcinoma (TCGA-STAD) dataset was employed as the training cohort, encompassing RNA sequencing data from 375 GC samples and 32 normal samples, along with comprehensive clinical information from 443 GC patients. Differentially expressed genes associated with GC were identified, and LLPS genes correlated with overall survival (OS) in GC patients were determined using the LLPS database. Univariate Cox analysis and least absolute shrinkage and selection operator (LASSO) Cox regression were applied to construct LLPS-based prognostic models, validated using the GEO15459 dataset containing clinical and gene expression data from 192 GC patients. Model accuracy was assessed via area under the curve (AUC) values. Multiple algorithms were employed to calculate immune cell infiltration scores for high- and low-risk groups. Finally, gene set enrichment analysis (GSEA) enrichment analysis was performed on the selected genes to explore biological processes and pathways.

[RESULTS] Through univariate Cox analysis and the LASSO Cox penalized regression analysis, six genes () associated with the OS of GC patients were found and an LLPSG prognostic model was constructed. In our LLPS-related prognostic model, GC patients in the high-risk group had a poorer OS rate than those in the low-risk group. For 1-, 3-, and 5-year survival rates, the AUC predictive values of the LLPSG nomogram were 0.63, 0.63, and 0.70, respectively. The GSE15459 cohort confirmed the favorable prognostic effect of our model. The predicted survival rates at 1-, 3-, and 5-year are 0.61, 0.64, and 0.66, respectively. There were also significant differences in immune cell infiltration between the high- and low-risk groups of the model. GSEA analysis showed that the six genes were differentially enriched in various cancer-related pathways.

[CONCLUSIONS] Our research establishes and validates an LLPS-associated risk model centered on the genes and . These genes are poised to be considered as potential therapeutic targets in the treatment of GC.