cartography, or toponomics, enables the imaging of a hundred of antibodies with
subcellular resolution [6–8]. Array tomography [9, 10] allows the imaging of
dozens of different antibodies with even higher resolution (100 Â 100 Â 70 nm
using deconvolution, [11]). Super-resolution fluorescence imaging that can bring
the resolution down to about 10–20 nm laterally is also becoming compatible with
multiplexing (e.g. [12–14]).
Ultimately, the combination of light-level immunolabeling and electron microscopy can provide both the multiplexing capabilities needed to study the molecular
architecture of biological tissues and the nanometer resolution to explore their
ultrastructure. Efforts to achieve this have been hampered for many years by
conflicting requirements of tissue preparation; procedures that preserve antigenicity
are generally detrimental to ultrastructure, and vice versa. Furthermore, the computing power and tools required to register light-level and electron microscopic
images were not available. Recently, several powerful approaches, such as automated TEM [15, 16], conjugate array tomography [17–19], and serial multiplex
immunogold labeling [20], have succeeded in combining large-scale light level
immunolabeling with electron microscopy.
6.2 Approaches to Large-Scale Molecular
and Ultrastructural Imaging
An effective approach to obtain molecular information from ultrastructurally
resolved tissue is automated TEM (ATEM) with computational molecular phenotyping [15, 16]. Designed and implemented by the Marc lab at the University of
Utah, this approach uses serial ultrathin sections (50–70 nm) of plastic-embedded
chemically fixed tissue to study the organization of the mammalian retina. A small
subset of individual sections are taken from the series at regular intervals, processed
for immunohistochemistry, and imaged at the light level. All the other sections are
viewed by automated TEM with nanometer resolution. More than 1000 image tiles
are obtained from each section to cover a large area for neuronal circuit reconstruction. The immunostained sections are then registered with the EM sections,
allowing the molecular composition of cells to be established; this procedure is
termed computational molecular phenotyping. Neuronal processes are traced and
synapses ultrastructurally identified on the TEM sections. This method takes
advantage of the fact that immunoreactivity for small molecules, such as GABA
and glutamate, can be well preserved in tissue prepared with conventional methods
for electron microscopy [21, 22]. Up to 11 different antibody labels have been used
for computational molecular phenotyping, including the excitation marker AGB
(1-amino-4-guanidobutane), a channel-permeant organic cation, used to probe prior
in vivo activity. Typical ATEM image datasets range from several terabytes to more
than a petabyte, and require exquisite new image processing, assembly, navigation
and analysis algorithms, as well as new interpretive frameworks. This method has
150
K. D. Micheva and K. D. Phend
subcellular resolution [6–8]. Array tomography [9, 10] allows the imaging of
dozens of different antibodies with even higher resolution (100 Â 100 Â 70 nm
using deconvolution, [11]). Super-resolution fluorescence imaging that can bring
the resolution down to about 10–20 nm laterally is also becoming compatible with
multiplexing (e.g. [12–14]).
Ultimately, the combination of light-level immunolabeling and electron microscopy can provide both the multiplexing capabilities needed to study the molecular
architecture of biological tissues and the nanometer resolution to explore their
ultrastructure. Efforts to achieve this have been hampered for many years by
conflicting requirements of tissue preparation; procedures that preserve antigenicity
are generally detrimental to ultrastructure, and vice versa. Furthermore, the computing power and tools required to register light-level and electron microscopic
images were not available. Recently, several powerful approaches, such as automated TEM [15, 16], conjugate array tomography [17–19], and serial multiplex
immunogold labeling [20], have succeeded in combining large-scale light level
immunolabeling with electron microscopy.
6.2 Approaches to Large-Scale Molecular
and Ultrastructural Imaging
An effective approach to obtain molecular information from ultrastructurally
resolved tissue is automated TEM (ATEM) with computational molecular phenotyping [15, 16]. Designed and implemented by the Marc lab at the University of
Utah, this approach uses serial ultrathin sections (50–70 nm) of plastic-embedded
chemically fixed tissue to study the organization of the mammalian retina. A small
subset of individual sections are taken from the series at regular intervals, processed
for immunohistochemistry, and imaged at the light level. All the other sections are
viewed by automated TEM with nanometer resolution. More than 1000 image tiles
are obtained from each section to cover a large area for neuronal circuit reconstruction. The immunostained sections are then registered with the EM sections,
allowing the molecular composition of cells to be established; this procedure is
termed computational molecular phenotyping. Neuronal processes are traced and
synapses ultrastructurally identified on the TEM sections. This method takes
advantage of the fact that immunoreactivity for small molecules, such as GABA
and glutamate, can be well preserved in tissue prepared with conventional methods
for electron microscopy [21, 22]. Up to 11 different antibody labels have been used
for computational molecular phenotyping, including the excitation marker AGB
(1-amino-4-guanidobutane), a channel-permeant organic cation, used to probe prior
in vivo activity. Typical ATEM image datasets range from several terabytes to more
than a petabyte, and require exquisite new image processing, assembly, navigation
and analysis algorithms, as well as new interpretive frameworks. This method has
150
K. D. Micheva and K. D. Phend