Zhentao Hou, Chen Gao
Pancreatic cancer has a poor prognosis. This study investigated the prognostic role of the keratin gene family and aimed to develop a keratin-based biomarker. Bioinformatics analysis was performed using RNA-seq data from gemcitabine-treated patients in the TCGA-PAAD cohort (discovery set, n = 62). Differential expression and protein interaction analyses identified resistance-associated keratin genes. A Keratin Switch Score (KSS) was constructed via LASSO-Cox regression based on the KRT13/KRT4 expression ratio. Its prognostic value was validated in the full TCGA cohort (n = 183) and an independent GEO dataset (GSE205154, n = 289) using Kaplan–Meier and multivariate Cox analyses. Associations with pathways, tumor mutational burden, and immune microenvironment were assessed using GSEA, GSVA, and immune deconvolution algorithms. Multiple keratin genes were upregulated in chemotherapy-resistant tumors. The KRT13/KRT4-based KSS was an independent prognostic factor for worse overall survival in both validation cohorts. The high-KSS subtype was enriched for E2F targets, hypoxia, epithelial-mesenchymal transition, and p53 pathway activity, and exhibited higher tumor mutational burden. Its tumor immune microenvironment was characterized by functional suppression (e.g., downregulated T-cell exhaustion genes). KSS did not stably correlate with in vitro sensitivity to gemcitabine or other chemotherapeutics. The KRT13/KRT4 ratio defines a novel, high-risk pancreatic cancer subtype with integrated aggressive tumor cell phenotypes and an immunosuppressive microenvironment. KSS is a robust, independent prognostic biomarker that may guide risk stratification and combination therapy strategies.