Alignment of neurites between light and electron microscopy

For tissue where fine-scale features (such as neurites) are connected to cell bodies within the high-resolution EM volume, the initial link between cell body identification and matching of smaller features is straightforward. In the example retina dataset, neurites are reconstructed by tracing the primary neurites from the cell body located in medium resolution EM volumes into high-resolution EM volumes (Figure 8A). Cell matching and tracing accuracy is confirmed by superimposing 3D renderings of optical images and EM segmentations using Amira (ThermoFisher Scientific) (Figure 8B).

Matching neurites between optical images and EM. (A) The proximal neurites (red) of cell bodies identified in a large field, medium resolution image volume (40 × 40 × 40 nm voxel size, Figure 5H) are traced into the high-resolution image volume (blue, 4 × 4 × 40 nm voxel size). (B) Matching the morphology of EM reconstructed neurites (blue) to images from fixed (green) and live (red) optical imaging confirms the initial matching of cell bodies. (C) Single slice from EM volume (viewed in VAST) where live two-photon (red) and fixed confocal (green) fluorescence images of the neurons of interest are aligned with the EM volume to determine neurite-to-neurite matches with EM traced (blue) neurons. (D) Optical image (red) affine transformed to better fit the EM traced neurites (blue). (E) Correspondence points (red targets) where positions in the optical image (gray) have been mapped onto positions of EM traced neurites.

Once the target cells are reconstructed, the ease of matching subcellular features within an arbor depends on the structural details of the neurons and the quality and sparseness of the optical maps. For the live retinal imaging example, optical labeling of the neuropil was too dense for most neurites to be matched using a manual side-by-side comparison of the light and EM images. To align optical data and EM segmentations at the micrometer scale, we use matched fiducial points from cell nuclei and large neurites to calculate an affine transformation of the optical data into the EM volume space (ImageJ). The matching is further refined using a thin plate spline transform with additional fiducial points. We then are able to view the optical images superimposed on the raw EM data and EM segmentations in VAST (Figure 8C) where we can identify additional fine correspondence between optical images and traced neurites. By iteratively adding more tracing, more correspondence points, and re-transforming the optical data, we generate a dense mapping of correspondence between the optical and EM images (Figures 8D,E).

The fine-scale projection of the optical data into the EM volume also allows us to identify optically characterized neurites that were not previously traced in the EM volume. Within a plexus of labeled neurites (Figures 1A,B), we could iteratively (1) identify a correlated neurite, (2) determine where the next closest optically imaged neurite should be, and (3) perform dense neurite segmentation in the projected region to find the neurite with morphology matching the optical image. Leapfrogging through the optically imaged plexus is significantly more difficult than matching neurites from labeled cell bodies. The general approach to matching light and EM neurites is described in detail by Drawitsch et al. (2018), and the efficiency of the approach is aided by starting with a saturated segmentation of all neurites in the region of interest.

For tissue where the cell bodies of the neurites of interest are not included in the sectioned volume, it is possible to match features between scales using other dense correlation features. For example, to identify contralaterally projecting retinal ganglion cell boutons in the lateral geniculate nucleus, reflected light imaging of fiber tracts (low and medium-resolution features), DAPI staining (medium and high-resolution markers), and fluorescently tagged Cholera Toxin B (CtB) labeling of the axons of interest was sufficient (Figure 9). This application benefits from the distinctive ultrastructural profile of retinal ganglion cell boutons.

Application of multiresolution feature matching in a brain slice. (A) The initial alignment of light and EM uses myelinated fiber tracts. Top panel shows confocal image of aldehyde fixed dLGN coronal slice. Red arrows indicate three myelinated tracts that are visible in both reflected light and EM. Green = reflected light, Blue = Nissel stain, Red = axon terminals of CtB injected retinal ganglion cells. Bottom panel shows EM image of the surface of the same brain slice. The white outline in the optical image indicates the position of the EM section. Rectangles indicate the position of images in (B). (B) Secondary alignment uses blood vessels. Top and bottom panels are higher resolution image acquisitions from positions shown in (A). Corresponding blood vessels are indicated by asterisks. Rectangles indicate the position of (C). (C) Cropped images from (B) showing correspondences of synaptic bouton and chromatin signals. In the EM image, retinal ganglion cell boutons (ultrastructurally identified by light mitochondria) are highlighted in red and match with the CtB labeled boutons in the optical image. The nucleus and chromatin pattern in EM are highlighted in blue and corresponds to the DAPI labeling in the optical image.