Abstract Although major advances have been made in the field of mesoscopic imaging and associated tissue clearing protocols, these applications are greatly challenged when applied to imaging of pancreatic ductal adenocarcinoma (PDAC) tissue. Most importantly, penetration of labelling agents, typically antibodies, can be drastically reduced from the characteristically dense PDAC stroma. To circumvent this issue, we present a method by which machine learning assisted segmentation is applied to resolve the 3D PDAC microarchitecture from autofluorescence (AF) based light-sheet fluorescence microscopy (LSFM) scans. Hereby, PDAC tissue features could be studied in 3D space without the need for labelling or sectioning. In this proof of principle study, we applied this imaging pipeline on surgical specimens from five PDAC patients and normal pancreatic tissue, generating mosaics of cm 3 -sized tissue discs at micrometre resolution, each on scanning depths corresponding to thousands of standard pathological 2D tissue sections. Using this method, we generated 3D volumes for quantification of blood vasculature, neoplastic epithelium, islets of Langerhans and stromal components. We further showcase the potential for downstream 2D histochemical and immunohistochemical analysis of scanned specimen. As such, the method may facilitate studies of metastatic routes, vessel microarchitecture, islet phenotypes, and spatial relationships in the PDAC tumour microenvironment.
Lehrstrand et al. (Sun,) studied this question.