Identification of Novel Metabolism-Associated Subtypes for Pancreatic Cancer to Establish an Eighteen-Gene Risk Prediction Model

Abstract

Pancreatic cancer (PanC) is an intractable malignancy with a high mortality. Metabolic processes contribute to cancer progression and therapeutic responses, and histopathological subtypes are insufficient for determining prognosis and treatment strategies. In this study, PanC subtypes based on metabolism-related genes were identified and further utilized to construct a prognostic model. Using a cohort of 171 patients from The Cancer Genome Atlas (TCGA) database, transcriptome data, simple nucleotide variants (SNV), and clinical information were analyzed. We divided patients with PanC into metabolic gene-enriched and metabolic gene-desert subtypes. The metabolic gene-enriched subgroup is a high-risk subtype with worse outcomes and a higher frequency of SNVs, especially in KRAS. After further characterizing the subtypes, we constructed a risk score algorithm involving multiple genes (i.e., NEU2, GMPS, PRIM2, PNPT1, LDHA, INPP4B, DPYD, PYGL, CA12, DHRS9, SULT1E1, ENPP2, PDE1C, TPH1, CHST12, POLR3GL, DNMT3A, and PGS1). We verified the reproducibility and reliability of the risk score using three validation cohorts (i.e., independent datasets from TCGA, Gene Expression Omnibus, and Ensemble databases). Finally, drug prediction was completed using a ridge regression model, yielding nine candidate drugs for high-risk patients. These findings support the classification of PanC into two metabolic subtypes and further suggest that the metabolic gene-enriched subgroup is associated with worse outcomes. The newly established risk model for prognosis and therapeutic responses may improve outcomes in patients with PanC.

Keywords: metabolic genes; pancreatic cancer; risk model; subtype; transcriptome.