2270P Integrative analyses of bulk and single-cell RNA-seq identified diabetes mellitus-related signature as a prognostic factor in pancreatic adenocarcinoma

Pancreatic Adenocarcinoma (PAAD) is a highly fatal disease and often associated with a poor prognosis. The development of diabetes mellitus (DM) is an important risk factor. There have been studies exploring markers or prognostic models that affect the prognosis of patients with pancreatic cancer, but there are few studies on pancreatic cancer-associated diabetes mellitus (PCDM) patients. The differentially expressed genes (DEGs) between DM and non-DM were detected and selected to build model. PCDM patients were randomly divided into training (70%) and test cohort (30%). Then the univariate COX was performed to screen OS associated genes (p < 0.05) in the training cohort. The LASSO-penalized COX regression and ten-fold cross validation were used to train and select models. The risk score was calculated and verified in both training and test cohort. GSVA, ESTIMATEScore, CIBERSORT and ssGSEA were used to compare the tumor microenvironment (TME). Moreover, the value of genes used in the model was detected on single cell level. Totally 197 patients (including 60 with DM) were enrolled, 23 genes from DEGs were related to the OS of PCDM patients in the training cohort, and 10 genes were selected to construct the risk model. The risk score for each subject was calculated as follow: risk score=0.434*ACACA+0.119*ATG7+0.373*DEFB123+0.135*FSTL3-0.168*NIPSNAP3B-0.069*RASSF1+0.225*RBPJ+0.364*SLC35F2-0.168*SLC37A1-0.145*ZC3H12D. The C-index was 0.83 in the training cohort (hazard ratio [HR] = 7.61, p < 0.001) and 0.76 in the test cohort (HR = 3.22, p = 0.015). According to the median value -0.054, PCDM patients were split into different risk groups, and patients in the low-risk group showed more enriched T cell related pathways (60 vs 1), higher ImmuneScore (p < 0.001), and more CD8 positive T cell infiltration (CIBERSORT: p < 0.001; ssGSEA: p < 0.001). We calculated the score of genes used in the model in single cells. Malignant cells with the highest score (top 10%) had enhanced activities of angiogenesis. The prognostic model showed potential value in predicting the OS of patients with PCDM. Patients with low-risk score had a hotter TME and might benefit from immunotherapy.