Unraveling the Molecular Interplay Between Pancreatic Cancer and Diabetes Mellitus: A Multilevel Model Approach Applied to Spatial Transcriptomics

Abstract

Pancreatic ductal adenocarcinoma (PDAC) is a highly malignancy with increasing incidence, where type 2 diabetes mellitus (T2DM) emerges as a risk factor and a possible consequence. The molecular and spatial mechanisms behind this relationship remain poorly understood. Spatial transcriptomics offers a powerful tool in this context. This study analyzed in situ sequencing (ISS) data from 18 PDAC patients stratified by diabetes status (no diabetes, new-onset diabetes mellitus (NODM), and long-standing diabetes mellitus (LSDM)). Cell types were identified using overrepresentation analysis. Differential expression was tested using the iDESC R package. Spatial co-expression of glucagon and insulin was quantified using different metrics, and visualized using the SEAGAL Python package. Ectopic expression of glucagon and insulin was assessed using generalized linear mixed-effects models and associated gene signatures in acinar cells were derived through penalized mixed-effects models. Cell-cell distances were analyzed using linear mixed-effects models. 30 cell types were identified, with antigen-presenting cancer-associated fi broblasts enriched in LSDM patients. Differential expression analysis revealed diabetes-associated gene signatures, particularly in LSDM. The spatial co expression of glucagon and insulin was weak but consistent across samples. Ectopic expression of both hormones was primarily observed in acinar cells, and two gene signatures related to it were constructed. The distance between T and T helper cells was increased in patients with LSDM compared to nondiabetic patients. This Master’s Thesis presents a multilevel statistical framework for analysing ISS data. The methods presented here can be adapted to larger cohorts and emerging technologies to refine our understanding of the PDAC-T2DM interface.