Académie royale de Médecine de Belgique


Résumé Denis Wirtz, membre étranger


par Denis WIRTZ (Johns Hopkins University - USA), membre étranger.

As pancreatic ductal adenocarcinoma (PDAC) is predicted to become the second leading cause of cancer death in the US, a better understanding of the spatial complexity of the human tumor microenvironment is necessary for an improved design of in vitro and in vivo model systems.  Pancreatic intraepithelial neoplasia (PanIN) is a known precursor to PDAC.  Our group has developed a pipeline for building large (cm-scale) models of PanIN and PDAC tumors to inform understanding of how tumors progress in 3D.  We reconstructed serially sectioned histologically stained human pancreatic tissue volumes of up to 3x3x0.7cm3

We derived an elastic image registration method that creates aligned volumes of pancreatic tissue from scanned whole-slide images. To quantify the tumor microenvironment, we developed a DeepLab semantic segmentation model that identifies nine tissue types from the H&E images at 93% validation accuracy: lipid, connective tissue, blood vessels, normal ductal epithelium, islets of Langerhans, acini, PanIN, PDAC, and lymphocytes.  Tumor classification was performed on all sample images, creating 3D cm scale tumor maps at single cell resolution. 

Using our 3D models, we calculated tumor connectivity and branching, immune cell infiltration, cancer cell intravasation, angiogenesis, and radial densities of tissue components surrounding normal and diseased epithelium.  We have modelled pancreatic cancer cells that grow within a muscular vessel for over 3mm.  This is significant as vascular invasion can lead to metastasis to the liver.  In PanIN, we have noted great heterogeneity in tumor inflammation: single lesions can have >500% variability in radial lymphocyte density in different tissue sections.