Correlative imaging is a powerful synergy of techniques, ideally suited for the analysis of rare or transient cellular structures. For some cellular structures, like phagophores, correlative microscopy may be the only analysis method available. Phagophores are dynamic double-membraned structures created by eukaryotic cells to trap damaged organelles, viruses and bacteria to enable their delivery to the lysosome for degradation. Phagophores are essential for the clearance of many intracellular hazards, but their origin and nature is a topic of considerable debate; they form at locations that cannot be predicted in advance, they only exist for 5 to 15 minutes, and they constantly change shape. To complicate matters further, phagophore sizes can vary by several orders of magnitude; their membranes are only separated by 4nm, yet the structure itself can laterally expand further than 6000nm. These characteristics make phagophores notoriously difficult to study by conventional 2D electron microscopy, and a full 3D structure of a phagophore never been reconstructed before.
Volumetric electron microscopy via FIB-SEM is ideally suited for 3D analyses of cellular membranes. By combining FIB-SEM with correlative imaging, I have reconstructed and morphometrically characterised 12 phagophores, along with all other organelles within 1000nm. The fully segmented and classified 3D data were acquired from four different cells (total volume of 1540µm^3), at a resolution 3.25x3.25x10nm per voxel. Complete segmentation of all organelles within this volume would have traditionally required several years of full-time technician labour. Here, I present a computational workflow in which the laborious task of segmentation is delegated to machine learning based computer vision, enabling 3D reconstruction to be conducted in under 30 days by a single human analyst, using only free and open-source software.