high-resolution. Higher throughput, faster FIB milling together with streamlined
workflows will facilitate high-resolution subtomogram averaging in situ, as will
development of additional genetic techniques to acquire thinner cells.
Improved algorithms will be needed that capitalize upon higher resultant
signal-to-noise ratios to provide higher quality reconstructions. One approach will
be to develop tomogram reconstruction algorithms that compensate for specimen
movement or warping over the tilt series, and mosaic reconstructions based on
detection of ‘tectonic plates’ of structure within the specimen that move synchronously; combinations of multiple individual localized reconstructions of such
areas will avoid the ‘one size fits all’ global fiducial model approach. Algorithms to
correct for CTF and phase plate CTF will also be enabled by better quality images.
CTF correction of phase plate images requires fitting to two unknowns, defocus and
phase plate phase shift, as opposed to solely defocus.
Given higher quality images from hardware and software advances it will be
important to increase the throughput of cryo-tomographic imaging. Higher
throughput imaging will be enabled by a number of developments. Camera
frame-rate and stage speed and stability would enable continuous data collection
from a continuously moving stage on timescales of seconds as opposed to minutes,
facilitating collection of datasets as large as used in single particle analysis projects,
and cutting dataset acquisition times from weeks to hours [35, 54]. Continued
algorithm development to use the full tilt series for initial alignment but only the
relatively undamaged initial frames for final reconstruction will capitalize on the
wealth of data. Increased data acquisition speeds will make specimen availability
and targeting the new bottlenecks. Greater availability of sample will avoid having
insufficient sample to image; this could be facilitated by automated switching of
grids in automated cartridge-based systems, ability to load larger numbers of grids,
or more reliable production of grids with optimal specimen density such as the
SpotItOn [130] or microfluidics approaches. Fully automated tomography targeting, reconstruction, and annotation will also need to be developed in light of
increased throughput, incorporating automated sophisticated target recognition
(e.g., automated recognition of cells or holes in carbon), followed by fully automated tomogram reconstruction and automated segmentation of features. Higher
quality data will spur advances in global template matching and automated
segmentation.
It will be important to be able to interpret higher quality tomograms. Higher
signal-to-noise ratio tomograms will lend themselves to easier template matching
strategies. It will also be necessary to further develop tagging strategies to identify
structures in tomograms. Integrated streamlined workflows incorporating CLEM
and cryo-FIB milling in which all metadata is tracked on behalf of the user will
remove the overhead of tracking targets by the user are essential. Development of
streamlined workflows for super-resolution CLEM is needed to enable fluorescent
tags to be used as standard tools for identifying components of tomograms, but it
will also be critical to develop a general-purpose tag to be used solely for ECT
visualization of protein localization. The recent proof-of-principle study employing
ferritin suggests the way forwards [93]. Tagging advances are also needed for
86
J. L. Ferreira et al.
workflows will facilitate high-resolution subtomogram averaging in situ, as will
development of additional genetic techniques to acquire thinner cells.
Improved algorithms will be needed that capitalize upon higher resultant
signal-to-noise ratios to provide higher quality reconstructions. One approach will
be to develop tomogram reconstruction algorithms that compensate for specimen
movement or warping over the tilt series, and mosaic reconstructions based on
detection of ‘tectonic plates’ of structure within the specimen that move synchronously; combinations of multiple individual localized reconstructions of such
areas will avoid the ‘one size fits all’ global fiducial model approach. Algorithms to
correct for CTF and phase plate CTF will also be enabled by better quality images.
CTF correction of phase plate images requires fitting to two unknowns, defocus and
phase plate phase shift, as opposed to solely defocus.
Given higher quality images from hardware and software advances it will be
important to increase the throughput of cryo-tomographic imaging. Higher
throughput imaging will be enabled by a number of developments. Camera
frame-rate and stage speed and stability would enable continuous data collection
from a continuously moving stage on timescales of seconds as opposed to minutes,
facilitating collection of datasets as large as used in single particle analysis projects,
and cutting dataset acquisition times from weeks to hours [35, 54]. Continued
algorithm development to use the full tilt series for initial alignment but only the
relatively undamaged initial frames for final reconstruction will capitalize on the
wealth of data. Increased data acquisition speeds will make specimen availability
and targeting the new bottlenecks. Greater availability of sample will avoid having
insufficient sample to image; this could be facilitated by automated switching of
grids in automated cartridge-based systems, ability to load larger numbers of grids,
or more reliable production of grids with optimal specimen density such as the
SpotItOn [130] or microfluidics approaches. Fully automated tomography targeting, reconstruction, and annotation will also need to be developed in light of
increased throughput, incorporating automated sophisticated target recognition
(e.g., automated recognition of cells or holes in carbon), followed by fully automated tomogram reconstruction and automated segmentation of features. Higher
quality data will spur advances in global template matching and automated
segmentation.
It will be important to be able to interpret higher quality tomograms. Higher
signal-to-noise ratio tomograms will lend themselves to easier template matching
strategies. It will also be necessary to further develop tagging strategies to identify
structures in tomograms. Integrated streamlined workflows incorporating CLEM
and cryo-FIB milling in which all metadata is tracked on behalf of the user will
remove the overhead of tracking targets by the user are essential. Development of
streamlined workflows for super-resolution CLEM is needed to enable fluorescent
tags to be used as standard tools for identifying components of tomograms, but it
will also be critical to develop a general-purpose tag to be used solely for ECT
visualization of protein localization. The recent proof-of-principle study employing
ferritin suggests the way forwards [93]. Tagging advances are also needed for
86
J. L. Ferreira et al.