Abstract PR-14: Meta-analysis of spatial transcriptomics data from pancreatic cancer precursors proposes a common molecular framework for PanIN and IPMN

Abstract

Abstract INTRODUCTION Pancreatic ductal adenocarcinoma (PDAC) arises from either pancreatic intraepithelial neoplasia (PanIN) or intraductal papillary mucinous neoplasm (IPMN). Transcriptomic studies of PanIN and IPMN conducted in isolation both aim to improve early detection, risk stratification, and therapeutic target selection. Here, we curated existing and new whole transcriptome digital spatial profiles (DSP-RNA) from normal to malignant pancreatic epithelium to map the molecular landscape of PDAC progenitors. METHODS New NanoString GeoMx DSP-RNA datasets were generated from 1) surgical IPMN specimens resected at Duke University and 2) deceased donor pancreas specimens collected by the Gift of Life Michigan Donor Care Center that could not be used for transplantation and thus allocated to the University of Michigan for approved research. Published studies were downloaded from GEO (GSE229752, GSE226829, GSE199102) and the NanoString Spatial Organ Atlas. Epithelial areas of interest (AOIs) were processed and analyzed with stgeomx (https://github.com/mkiyer/stgeomx) and R/Bioconductor packages. Quality control (QC) procedures removed low quality AOIs. Batch effects were mitigated by defining highly variable genes within each dataset and merging them into a common set for downstream analysis. Molecular subtype gene signatures were obtained from the pdacR package and quantified using Gene Set Variation Analysis. RESULTS We assembled a pancreatic epithelial DSP-RNA cohort consisting of 768 AOIs from 62 patients. AOIs were annotated as normal duct (n=193), acini (n=53), acinar-ductal metaplasia (ADM) (n=57), PanIN (n=108), IPMN (n=92), and PDAC (n=265). Principal component analysis (PCA) revealed compelling associations between DSP-RNA profiles and the established classical, basal-like, and exocrine molecular subtypes of PDAC. The first principal component (PC1) stratified the exocrine and basal-like PDAC subtypes on opposite ends of its axis (PC1 vs exocrine Pearson r=0.929, p < 0.001, PC1 vs basal-like Pearson r=-0.775, padj < 0.001). PC2 further distinguished classical from basal-like AOIs, and PC3 separated normal (CRISP3+) from neoplastic (MUC5AC+) ductal epithelium. Unsupervised clustering of AOIs proposed four epithelial subtypes: “acinar/ADM” (cAcinar), “normal duct” (cDuct), “low-risk neoplasm” (cLR), and “high-risk/malignant” (cHR). PanIN and IPMN lesions exhibited remarkably similar global gene expression patterns, suggesting that they arise from a common lineage. CONCLUSIONS Meta-analysis of pancreatic epithelial DSP-RNA datasets provided a unifying molecular context for PanIN, IPMN, and PDAC. The established classical, basal-like, and exocrine PDAC subtypes explained global patterns of gene expression variation. Intriguingly, PanIN and IPMN transcriptomes appeared similar, suggesting that early detection and therapeutic selection strategies may be broadly applicable to PDAC precursors. Citation Format: Matthew K Iyer, Ashley Fletcher, Elishama Kanu, Chanjuan Shi, Rosina Carr, Ahmed M. Elhossiny, Daniel P. Nussbaum, Eileen S. Carpenter, Marina Pasca di Magliano, Arul M. Chinnaiyan, Peter J. Allen, Timothy L. Frankel. Meta-analysis of spatial transcriptomics data from pancreatic cancer precursors proposes a common molecular framework for PanIN and IPMN [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Advances in Pancreatic Cancer Research; 2024 Sep 15-18; Boston, MA. Philadelphia (PA): AACR; Cancer Res 2024;84(17 Suppl_2):Abstract nr PR-14.