Project Summary/Abstract
Volume electron microscopy is the only technique to-date that provides both sufficient resolution (<20 nm) and sufficient
field of view (>100 μm) for the dense reconstruction of neuronal wiring diagrams. Currently, there exist two systems that
have already delivered mm3-sized synaptic resolution electron microscopy stacks: Multi-beam scanning electron
microscopy(Eberle et al. 2015; Ren and Kruit 2016) (mSEM) and Gridtape-based automated transmission electron
microscopy(Yin et al. 2020; Maniates-Selvin et al. 2020) (Gridtape-TEM). In mSEM, the sample is scanned with up to 91
parallel beams and an image is formed by low energy secondary electrons that are generated during scanning.
Gridtape-TEM detects transmitted electrons with one or multiple fast cameras simultaneously. Both techniques currently
rely on collecting and imaging thousands of ultrathin serial sections (30 - 40 nm) that are being cut with a diamond knife
on an ultramicrotome. For mSEM, the sections are either collected using an automated tape collecting ultramicrotome or
directly onto silicon wafers. For Gridtape-TEM, the sections are collected onto an electron-transparent film in
millimeter-sized apertures on Gridtape. However, serial collection of ultrathin sections is delicate and inherently prone to
failures and artifacts such as section loss, folds and cracks or knife marks. More than 50% of the errors of today’s
state-of-the-art automated neuron segmentation algorithms can be attributed to missing information due to
serial-sectioning. As a consequence, more than 40 hours of manual segmentation proofreading by human experts are
currently required to accurately reconstruct a single cortical pyramidal cell. Some of the remaining automated
segmentation issues can certainly be addressed by improving the underlying algorithms. But in order to scale dense
automated neuronal circuit reconstructions to whole mouse brains with about 70 million neurons, it is necessary to
significantly reduce the experimental artifacts. The collection of semi-thin sections with a thickness around 100 - 500 nm
has been proposed as a much more robust alternative to ultrathin sectioning. In order to maintain or even increase the
resolution in Z, these semi-thin sections could be iteratively milled and scanned in the case of mSEM or a series of images
at different tilt angles could be acquired in the case of Gridtape-TEM. Here we propose to combine the commercially
available multi-beam scanning transmission electron microscope FASTEM from Delmic with iterative broad ion beam
milling of semi-thin sections (BIB-mSTEM). First, hundreds of semi-thin sections will be collected directly onto
scintillator plates using the commercially available MagC magnetic collection system. Subsequently, these sections will be
iteratively thinned and imaged by going back and forth between broad ion beam milling and imaging with FASTEM. For
each section, this will produce a series of iteratively milled TEM projection images that can be used to reconstruct a
high-resolution 3d stack of each section. BIB-mSTEM will be substantially more robust and reliable than mSEM and
Gridtape-TEM based workflows: In contrast to Gridtape-TEM, the sections are collected onto a solid substrate and not on
a fragile support film. In contrast to mSEM, BIB-mSTEM forms the image from high energy transmitted electrons that are
much less sensitive to local electromagnetic fields and milling-induced irregular surface topography than low energy
secondary electrons.
