A novel natural killer cell-related signatures to predict prognosis and chemotherapy response of pancreatic cancer patients

Abstract

Background: Natural killer (NK) cells are involved in monitoring and eliminating cancers. The purpose of this study was to develop a NK cell-related genes (NKGs) in pancreatic cancer (PC) and establish a novel prognostic signature for PC patients.Methods: Omic data were downloaded from The Cancer Genome Atlas Program (TCGA), Gene Expression Omnibus (GEO), International Cancer Genome Consortium (ICGC), and used to generate NKG-based molecular subtypes and construct a prognostic signature of PC. NKGs were downloaded from the ImmPort database. The differences in prognosis, immunotherapy response, and drug sensitivity among subtypes were compared. 12 programmed cell death (PCD) patterns were acquired from previous study. A decision tree and nomogram model were constructed for the prognostic prediction of PC.Results: Thirty-two prognostic NKGs were identified in PC patients, and were used to generate three clusters with distinct characteristics. PCD patterns were more likely to occur at C1 or C3. Four prognostic DEGs, including MET, EMP1, MYEOV, and NGFR, were found among the clusters and applied to construct a risk signature in TCGA dataset, which was successfully validated in PACA-CA and GSE57495 cohorts. The four gene expressions were negatively correlated with methylation level. PC patients were divided into high and low risk groups, which exerts significantly different prognosis, clinicopathological features, immune infiltration, immunotherapy response and drug sensitivity. Age, N stage, and the risk signature were identified as independent factors of PC prognosis. Low group was more easily to happened on PCD. A decision tree and nomogram model were successfully built for the prognosis prediction of PC patients. ROC curves and DCA curves demonstrated the favorable and robust predictive capability of the nomogram model.Conclusion: We characterized NKGs-derived molecular subtypes of PC patients, and established favorable prognostic models for the prediction of PC prognosis, which may serve as a potential tool for prognosis prediction and making personalized treatment in PC.