Three-Dimensional Cell and Tissue Reconstruction by Serial Block Face SEM

点击翻译按钮获取中文摘要

We have developed and applied techniques based on serial block-face scanning electron microscopy (SBF-SEM) with a Zeiss SIGMA-VP SEM and a Gatan 3View in situ ultramicrotomy system to determine the 3D ultrastructure of cells and tissues at a lateral (x,y) resolution of 5 to 10 nanometers and a z-resolution of 25 nm to 50 nm, as limited by the minimum thickness that can be removed from the blockface by the in situ microtome. We have applied SBEM extensively to determine the 3-D ultrastructure of human and mouse blood platelets that were rapidly fixed prior to purification to minimize activation (1-3). One objective was to determine the 3D organization of granules, dense granules, mitochondria, and canalicular system in resting human platelets and map their spatial relationships. We found that granule number correlated linearly with platelet size, whereas dense granule and mitochondria number had little correlation with platelet size. 3D data from 30 platelets indicated only limited spatial intermixing of the different organelle classes. Interestingly, almost 70% of granules came within 35 nm of each other, a distance associated in other cell systems with protein mediated contact sites. Size and shape analysis of the 1,500 granules analyzed revealed no more variation than that expected for a Gaussian distribution (1). Although mouse platelets provide an experimental model for hemostasis and thrombosis, important physiological data from this model have received little quantitative, 3D ultrastructural analysis. We have obtained SBEM data from resting mouse platelets. Quantitative analysis revealed that mouse alpha-granules typically had a variable, elongated, rod shape, different from the round/ovoid shape of human granules. This variation in length was confirmed qualitatively by higher-resolution, focused ion beam (FIB) SEM at a nominal 5 nm Z-step size. The unexpected alpha-granule shape raises novel questions regarding alpha-granule biogenesis and dynamics, about whether the variation arise at the level of the megakaryocyte and alpha-granule biogenesis or from differences in alpha-granule dynamics and organelle fusion/fission events within circulating platelets (2). Furthermore, quantitative analysis revealed that the two major organelles in circulating platelets, alpha-granules and mitochondria, displayed a stronger linear relationship between organelle number/volume and platelet size, i.e., a scaling in number and volume to platelet size, than found in human platelets suggestive of a tighter mechanistic regulation of their inclusion during platelet biogenesis. The overall spatial arrangement of organelles within mouse platelets was similar to that of resting human platelets, with mouse alpha-granules clustered closely together with little space for inter-digitation of other organelles. The canalicular system (CS) has been defined as an inward, invaginated membrane connector that supports entry of substances into and out of the platelet, a static structure stable during platelet isolation, and the major source of plasma membrane for surface area expansion during activation. We have used SBEM imaging to reexamine the CS in mouse platelets by generating high-resolution 3D reconstructions from immediately fixed or washed platelets fixed post-washing. We have shown that CS, even in the presence of activation inhibitors, reorganized during platelet isolation to generate a more interconnected network. Furthermore, CS redistribution into the plasma membrane at different times, post-activation, appeared to account for only about half of the PM expansion seen in thrombin-activated platelets, in vitro. This suggests that CS reorganization is not sufficient to serve as a dominant membrane reservoir for activated platelets. In sum, our analysis highlights the need to revisit past assumptions about the platelet CS to better understand how this membrane system contributes to platelet function (3). In another application, we have analyzed the 3D ultrastructure of secretory cells in mouse pancreatic islets of Langerhans, microscopic endocrine organs about 200 to 300 micrometers in size, which secrete insulin, glucagon and other substances for control of blood glucose. We have shown that a combination of 2D and 3D analyses of tissue volume ultrastructure acquired by serial block face scanning electron microscopy (SBF-SEM) can greatly shorten the time required to obtain quantitative information from big data sets that contain many billions of voxels. Thus, to analyze the number of organelles of a specific type, or the total volume enclosed by a population of organelles within a cell, we have shown that it is possible to estimate the number density or volume fraction of that organelle using a stereological approach to analyze randomly selected 2D slices through the cells, and to combine such estimates with precise measurement of 3D cell volumes by delineating the plasma membrane in successive slices. The validity of such an approach can be easily tested since the entire 3D tissue volume is available in the SBF-SEM data set. We have applied this hybrid 3D/2D technique to determine the number of secretory granules in alpha and beta cells of mouse pancreatic islets of Langerhans, and have been able to estimate the total insulin content of beta cells. We have also used the approach to estimate maturation times of secretory granules in beta cells by quantifying the numbers of immature and mature granules and by using data from radioactivity labeling in pulse chase experiments(4). SBF-SEM is capable of producing large 3D images of cellular ultrastructure, but the labor required to manually segment EM images into their semantic components hinders further data analysis. Currently, software pipelines incorporating deep neural networks offer state-of-the-art performance for automated segmentation. However, even state-of-the-art automated segmentation tools require extensive manual correction for many data sets of interest to the structural biology and systems biology communities, and are therefore impractical for image analysis. Our lab is designing novel neural networks and incorporating them into a segmentation software pipeline to improve automated segmentation performance for EM data sets taken from multiple biological systems. We are beginning to develop a design framework and software for constructing segmentation neural networks, and to test these methods on large 3D data sets generated in our laboratory. Whereas previous work in the field of deep learning in SBF-SEM has tended to focus on the identification of cell membranes for mapping neuronal circuits in brain, our approach aims to segment intracellular volumes into multiple classes of organelles for a diverse range of cell types. Preliminary results for blood platelets show considerable promise (M.D. Guay et al., bioRxiv, https://doi.org/10.1101/2020.01.05.895003) Guay MD, Emam ASE, Anderson AB, Aronova MA, Storrie B, Pokrovskaya ID, Leapman RD (2020) Dense cellular segmentation for EM using 2D-3D neural network ensembles. bioRxiv, https://doi.org/10.1101/2020.01.05.895003