8.8 Subtomogram Reconstruction
When a tomogram contains particles thought to be identical in structure, they can be
averaged to decrease the noise. This can be done in multiple ways, each with its
own complexities and limitations. Although it is covered in Chap. 9 in more detail,
here it is treated as an extension of the original reconstruction approaches.
8.8.1 3D Alignment and Averaging
The most common approach is to locate and pick 3D subvolumes from tomograms.
These are then aligned with respect to a reference or template and averaged [48, 50,
70–82]. With the introduction of direct detectors, it is now possible to easily correct
for the CTF (see Sect. 8.5) and achieve sub-nanometer resolutions [83].
Bharat et al. [84] developed a correction method that includes compensating for
the accumulated dose, effectively low-pass filtering images taken later in the tilt
series to avoid degradation of high resolution information in the earlier images. Lin
et al. [85] classified individual subtomograms to reveal the dynamics of dynein in
sea urchin sperm flagella.
8.8.2 2D Alignment, Reconstruction and Averaging
The large fields-of-view covered in tomography means that parts of the specimen
might shift relative to other parts. One solution may be to do local alignments and
do a distortion correction using a vector field (e.g., as is done in IMOD [27]). If the
target is to extract subtomograms, the task becomes somewhat easier. The global
alignment can be used to locate the 2D submicrographs, refine their individual
alignments, and then do a subtomogram reconstruction (implemented in the current
version of Bsoft [12]). This is effectively the same as single particle analysis, with
the advantage of taking the orientational constraints of the tilt series into
consideration.
8.8.3 Model-Based Averaging
Macromolecules are often arranged in defined locations and orientations. For
example, the glycoproteins on the surface of an enveloped virus typically adopt a
specific orientation with respect to the membrane. This can be used to select the
glycoprotein densities with considerable orientation constraints to aid in alignment
and averaging [74, 76]. The clathrin networks of coated vesicles can be modeled as
230
J. Bernard Heymann
When a tomogram contains particles thought to be identical in structure, they can be
averaged to decrease the noise. This can be done in multiple ways, each with its
own complexities and limitations. Although it is covered in Chap. 9 in more detail,
here it is treated as an extension of the original reconstruction approaches.
8.8.1 3D Alignment and Averaging
The most common approach is to locate and pick 3D subvolumes from tomograms.
These are then aligned with respect to a reference or template and averaged [48, 50,
70–82]. With the introduction of direct detectors, it is now possible to easily correct
for the CTF (see Sect. 8.5) and achieve sub-nanometer resolutions [83].
Bharat et al. [84] developed a correction method that includes compensating for
the accumulated dose, effectively low-pass filtering images taken later in the tilt
series to avoid degradation of high resolution information in the earlier images. Lin
et al. [85] classified individual subtomograms to reveal the dynamics of dynein in
sea urchin sperm flagella.
8.8.2 2D Alignment, Reconstruction and Averaging
The large fields-of-view covered in tomography means that parts of the specimen
might shift relative to other parts. One solution may be to do local alignments and
do a distortion correction using a vector field (e.g., as is done in IMOD [27]). If the
target is to extract subtomograms, the task becomes somewhat easier. The global
alignment can be used to locate the 2D submicrographs, refine their individual
alignments, and then do a subtomogram reconstruction (implemented in the current
version of Bsoft [12]). This is effectively the same as single particle analysis, with
the advantage of taking the orientational constraints of the tilt series into
consideration.
8.8.3 Model-Based Averaging
Macromolecules are often arranged in defined locations and orientations. For
example, the glycoproteins on the surface of an enveloped virus typically adopt a
specific orientation with respect to the membrane. This can be used to select the
glycoprotein densities with considerable orientation constraints to aid in alignment
and averaging [74, 76]. The clathrin networks of coated vesicles can be modeled as
230
J. Bernard Heymann