PanIN and CAF Transitions in Pancreatic Carcinogenesis Revealed with Spatial Data Integration

Abstract

AbstractSpatial transcriptomics (ST) is a powerful new approach to characterize the cellular and molecular architecture of the tumor microenvironment. Previous single-cell RNA-sequencing (scRNA-seq) studies of pancreatic ductal adenocarcinoma (PDAC) have revealed a complex immunosuppressive environment characterized by numerous cancer associated fibroblasts (CAFs) subtypes that contributes to poor outcomes. Nonetheless, the evolutionary processes yielding that microenvironment remain unknown. Pancreatic intraepithelial neoplasia (PanIN) is a premalignant lesion with potential to develop into PDAC, but the formalin-fixed and paraffin-embedded (FFPE) specimens required for PanIN diagnosis preclude scRNA-seq profiling. We developed a new experimental pipeline for FFPE ST analysis of PanINs that preserves clinical specimens for diagnosis. We further developed novel multi-omics analysis methods for threefold integration of imaging, ST, and scRNA-seq data to analyze the premalignant microenvironment. The integration of ST and imaging enables automated cell type annotation of ST spots at a single-cell resolution, enabling spot selection and deconvolution for unique cellular components of the tumor microenvironment (TME). Overall, this approach demonstrates that PanINs are surrounded by the same subtypes of CAFs present in invasive PDACs, and that the PanIN lesions are predominantly of the classical PDAC subtype. Moreover, this new experimental and computational protocol for ST analysis suggests a biological model in which CAF-PanIN interactions promote inflammatory signaling in neoplastic cells which transitions to proliferative signaling as PanINs progress to PDAC.SummaryPancreatic intraepithelial neoplasia (PanINs) are pre-malignant lesions that progress into pancreatic ductal adenocarcinoma (PDAC). Recent advances in single-cell technologies have allowed for detailed insights into the molecular and cellular processes of PDAC. However, human PanINs are stored as formalin-fixed and paraffin-embedded (FFPE) specimens limiting similar profiling of human carcinogenesis. Here, we describe a new analysis protocol that enables spatial transcriptomics (ST) analysis of PanINs while preserving the FFPE blocks required for clinical assessment. The matched H&E imaging for the ST data enables novel machine learning approaches to automate cell type annotations at a single-cell resolution and isolate neoplastic regions on the tissue. Transcriptional profiles of these annotated cells enable further refinement of imaging-based cellular annotations, showing that PanINs are predominatly of the classical subtype and surrounded by PDAC cancer associated fibroblast (CAF) subtypes. Applying transfer learning to integrate ST PanIN data with PDAC scRNA-seq data enables the analysis of cellular and molecular progression from PanINs to PDAC. This analysis identified a transition between inflammatory signaling induced by CAFs and proliferative signaling in PanIN cells as they become invasive cancers. Altogether, this integration of imaging, ST, and scRNA-seq data provides an experimental and computational approach for the analysis of cancer development and progression.