Development of the TP53 mutation associated hypopharyngeal squamous cell carcinoma prognostic model through bulk multi-omics sequencing and single-cell sequencing

The aim of this study was to construct a prognostic model based on the TP53 mutation to calculate prognostic risk scores of patients with HPSCC. TP53 mutation and transcriptome data were downloaded from the TCGA databases. Gene expression data from GSE65858, GSE41613, GSE3292, GSE31056, GSE39366, and GSE227156 datasets were downloaded from the GEO database. GSEA, univariate, multivariate Cox analysis and LASSO analysis were employed to identify key genes and construct the prognostic model. ROC curves were utilized to validate the OS and RFS results obtained from the model. The associations between risk scores with various clinicopathological characteristics and immune scores were analyzed via ggplot2, corrplot package, and GSVA, respectively. Single-cell sequencing data was analyzed via unbiased clustering and SingleR cell annotations. Initially, two key genes, POLD2 and POLR2G, were identified and utilized to construct the prognostic model. Samples were divided into different risk groups via the risk scores obtained from the model, with high-risk group samples exhibiting poorer prognosis. Furthermore, the risk score exhibited a positive correlation with lymphatic metastasis in patients and the immune scores of CD4+ T, CD8+ T, dendritic cell, macrophage, and neutrophil. The immune responses also exhibited notable disparities between the high- and low-risk groups. The results of single-cell sequencing analysis demonstrated that epithelial cells and macrophages were relatively abundant in HPSCC samples. POLD2 and POLR2G exhibited higher expressions in epithelial cells, with most of the identified pathways also enriched in epithelial cells. The prognostic model exhibited a significant capacity for predicting the prognosis of HSPCC samples based on the TP53 mutation conditions and may also predict the cancer characteristics and immune infiltration scores of samples via different risk scores obtained from the model. Level 5.