6.5 Array Tomography: Computer Assisted Imaging
of Serial Sections on Rigid Substrates
The immunofluorescent and electron microscopic images obtained from the AT
sample are at very different scales; these images have to be computationally registered to identify corresponding points (Fig. 6.5). Image registration is required at
two different steps of the imaging process. First, low magnification IF and SEM
images from the entire ribbon are registered in order to identify the positions for
high magnification SEM imaging. And second, the high magnification IF and SEM
images are registered to assign immunofluorescence from different channels to the
underlying ultrastructure. Image registration is done by identifying corresponding
points from the sample in the two modes of imaging, IF and SEM. For example,
DAPI fluorescence from nuclei corresponds precisely to the darkly stained nuclear
heterochromatin in the EM images; MBP immunofluorescence corresponds to the
dark concentric circles of myelin in EM. Registration of light and EM images can
be achieved in a variety of ways; below is a brief description of the workflow used
in our lab which is described in more detail in [18].
To identify positions for high magnification SEM imaging, first a low-magnification
SEM mosaic from the entire ribbon is obtained, and this mosaic is stitched together
using the stitching algorithms available through FIJI [50]. MosaicPlanner, a custom
Python-based graphical user interface, (available at http://code.google.com/p/
smithlabsoftware), is then used to automatically find corresponding locations across
sections and generate a position list which is used to acquire SEM images at an intermediate magnification. The series of acquired SEM images are aligned across the
different sections using a translational transformation and an updated more precise
position list is calculated, correcting for inaccuracies in the original position list. Finally,
each serial section is reimaged at high resolution (2.23–3.7 nm/pixel) using this
updated position list.
Image registration between the IF and SEM modalities is accomplished using the
TrakEM2 plugin [51] within FIJI [50]. The registration procedure begins with
electron micrographs taken at intermediate magnification, and then registration is
refined using high magnification electron micrographs. DAPI images are histogramnormalized to make the spatial structure in both the dim autofluorescence and brighter
DAPI fluorescence equally apparent. This is useful because variations in the dim
autofluorescence correspond to ultrastructural features visible in the electron
microscope, such as large dendrites with weak autofluorescence, and mitochondria
with strong autofluorescence. Several corresponding features in the DAPI images and
the medium magnification EM images are identified by eye, and used to fit a similarity
transformation (rigid rotation plus uniform scaling). This transformation is automatically applied to the other light microscopy images. Once registered, identical
alignment transformations that bring the images from section to section into correspondence can be calculated and applied to images from a set of sections. Alignment
is based upon the EM images, using the elastic alignment algorithm in TrakEM2 [52],
and these alignment transforms are then applied to IF data. These image reconstruction tools, including tools to assist in deconvolution, stitching, registration, and
alignment, are available at http://code.google.com/p/smithlabsoftware.
6 Conjugate Immunofluorescence—SEM Array Tomography …
159
of Serial Sections on Rigid Substrates
The immunofluorescent and electron microscopic images obtained from the AT
sample are at very different scales; these images have to be computationally registered to identify corresponding points (Fig. 6.5). Image registration is required at
two different steps of the imaging process. First, low magnification IF and SEM
images from the entire ribbon are registered in order to identify the positions for
high magnification SEM imaging. And second, the high magnification IF and SEM
images are registered to assign immunofluorescence from different channels to the
underlying ultrastructure. Image registration is done by identifying corresponding
points from the sample in the two modes of imaging, IF and SEM. For example,
DAPI fluorescence from nuclei corresponds precisely to the darkly stained nuclear
heterochromatin in the EM images; MBP immunofluorescence corresponds to the
dark concentric circles of myelin in EM. Registration of light and EM images can
be achieved in a variety of ways; below is a brief description of the workflow used
in our lab which is described in more detail in [18].
To identify positions for high magnification SEM imaging, first a low-magnification
SEM mosaic from the entire ribbon is obtained, and this mosaic is stitched together
using the stitching algorithms available through FIJI [50]. MosaicPlanner, a custom
Python-based graphical user interface, (available at http://code.google.com/p/
smithlabsoftware), is then used to automatically find corresponding locations across
sections and generate a position list which is used to acquire SEM images at an intermediate magnification. The series of acquired SEM images are aligned across the
different sections using a translational transformation and an updated more precise
position list is calculated, correcting for inaccuracies in the original position list. Finally,
each serial section is reimaged at high resolution (2.23–3.7 nm/pixel) using this
updated position list.
Image registration between the IF and SEM modalities is accomplished using the
TrakEM2 plugin [51] within FIJI [50]. The registration procedure begins with
electron micrographs taken at intermediate magnification, and then registration is
refined using high magnification electron micrographs. DAPI images are histogramnormalized to make the spatial structure in both the dim autofluorescence and brighter
DAPI fluorescence equally apparent. This is useful because variations in the dim
autofluorescence correspond to ultrastructural features visible in the electron
microscope, such as large dendrites with weak autofluorescence, and mitochondria
with strong autofluorescence. Several corresponding features in the DAPI images and
the medium magnification EM images are identified by eye, and used to fit a similarity
transformation (rigid rotation plus uniform scaling). This transformation is automatically applied to the other light microscopy images. Once registered, identical
alignment transformations that bring the images from section to section into correspondence can be calculated and applied to images from a set of sections. Alignment
is based upon the EM images, using the elastic alignment algorithm in TrakEM2 [52],
and these alignment transforms are then applied to IF data. These image reconstruction tools, including tools to assist in deconvolution, stitching, registration, and
alignment, are available at http://code.google.com/p/smithlabsoftware.
6 Conjugate Immunofluorescence—SEM Array Tomography …
159