Background
The recurrence rate of prostate cancer (PCa) remains high. Research have shown that high expression of Golgi apparatus (GA) phosphoprotein 3 is related to occurrence of PCa. Therefore, the purpose of this study was to screen hub genes related to GA in PCa.
Methods
TCGA-PRAD, GSE46602 and 1644 golgi apparatus-related genes (GARGs) were performed. Differentially expressed GARGs (DE-GARGs) were obtained by differential expression analysis and Venn analysis. Next, hub genes were screened through protein-protein interaction networks (PPI), further univariate Cox regression and least absolute shrinkage and selection operator (Lasso) regression were performed to obtain prognostic genes and risk models. Afterwards, Kaplan-Meier (KM) curve and receiver operating characteristic (ROC) curve were utilized to evaluate model. Univariate and multivariate Cox regression analyses were executed to evaluate the independent predictive power of models. Further a nomogram was constructed to assess capability of risk score as an independent prognosis. Meanwhile, the correlation analysis of prognostic genes with clinical features and immune cells and drug sensitivity analysis were also carried out. Finally, the expression level of prognostic gene was analyzed.
Results
Among 13 hub genes were screened, and MUC6, PRAME and VGF were obtained by univariate Cox and Lasso regression, further a risk model was constructed. TCGA-PRAD was divided into high and low risk groups according to the median risk score. Firstly, KM curve showed that there was remarkable difference in biochemical recurrence (BCR) between the two groups, next the AUC value of 1, 3 and 5 years was above 0.65. Eventually, in GSE46602, it was also proved that the risk model had better forecasting ability. Meanwhile risk score could be used as an independent prognostic factor, and it was remarkably different in different clinical features. The better predictive ability of the nomogram was proved by calibration curve and DCA curve. Afterwards, there were remarkable differences in BCR between ESTIMATE score and high-low risk group, likewise, there were significant differences in 14 immune cells, 9 immune checkpoints, and 104 drugs between two risk groups. Lastly, the expression of prognostic genes was consistent with univariate Cox analysis when constructing risk model.
Conclusion
A reliable prognostic model based on MUC6, PRAME and VGF was constructed, which provided valuable information for in-depth exploration of the pathogenesis of PCa.