SBM department letter #21

On December 13, 2021, the multidisciplinary consortium coordinated by Christpohe Grosset (MIRCADE team, BMGIC – Inserm U1035) published an original article entitled « Deciphering tumour tissue organization by 3D electron microscopy and machine learning » in the Communications Biology journal.

Since 2015, the MIRCADE team (Methods and Innovations for Research on Childhood Cancers) has been studying the biology of several solid tumors in children, including hepatoblastoma, a primary liver cancer. Following observations made in laboratory, Christophe Grosset wondered about the parameters controlling the internal organisation of tumor tissue, knowing that they are very poorly known. To answer this question, he surrounded himself with researchers and engineers from the Departments of Biological and Medical Sciences, Bordeaux Neurocampus and Engineering and Digital Sciences of the University of Bordeaux, including Baudouin Denis De Senneville (Institute of Mathematics), Fatma Zohra Khoubai (MIRCADE team), Jean Ripoche (Inserm, University of Bordeaux), Marc Bevilacqua and Etienne Gontier (Bordeaux Imaging Center, Electron Microscopy Pole). He also approached Alexandre Labedade, independent graphic designer, Christophe Chardot and Sophie Branchereau, both pediatric surgeons at AP-HP, as well as Kathleen Flosseau and Stefano Cairo from XenTech society. After analyses of the biological samples by the « Serial block-face scanning electron microscopy » technique at the BIC and the 3D reconstruction of the tissue by stacking 250 electron microscopy images, the consortium developed a methodology based on the semi-automatic segmentation of the main elements structures of tumor tissue such as blood capillaries, cells, nuclei and mitochondria (https://github.com/bsenneville/Onconanotomy). As manual segmentation is too laborious to segment thousands of elements, the consortium traines an artificial neural network of the U-net type for the automatic segmentation of tissue elements, thus generating several training databases (protected by patent). Using this approach, 182 cells, 113 nuclei and 3 portions of a blood capillary were segmented and then analysed using mathematical approaches. The consortium then discovered (i) that blood capillaries and structures near the bile canaliculi influence the planar alignment, polarity and size of tumor cells and their mitochondrial network, and (ii) that the size of hepatoblastoma cells is correlated with that of their nucleus, their cytoplasm and their mitochondrial network. Through this work, the consortium provides some information about the bioarchitectural parameters controlling the 3D organisation of a tumor tissue, thus paving the way for future investigations in an emerging field called « Onconanotomy ».