have a much lower SNR than the crystal averages, and their alignment parameters are expected to not deviate too much from the
crystal average. Therefore, besides the auto-refinement parameters
(Subheading 3.4.2), it is now recommended to activate the alignment restraints available in frealign-2dx (see Note 8):
– Restraint for Euler angles, in degrees (sigma_angles): standard deviation for a Gaussian restraint on the Euler angles.
– Restraint for x,y shifts, in pixels (sigma_shifts): standard
deviation for a Gaussian restraint on the x, y shifts.
More information about these restraints can be found in the
Supplementary Note 1 of [38]. Upon convergence, a higher quality
map should have been obtained, and a more diverse set of views
should be observed, as shown in Fig. 3. This evidences that particles
from the same 2D crystal are not exactly always in the same orientation. This map, commonly referred to as a “consensus map”
because no 3D classification has been performed (yet), can now
be post-processed as described in Subheading 3.7. The user is also
encouraged to experiment with more advanced features, such as
defocus refinement [32, 39].
3.6 3D Classification
Perhaps the most interesting feature of the single particle method is
its ability to classify heterogeneous data into conformationally
homogeneous classes [35, 52]. Structural variability is also a source
of disorder in 2D crystals, thus limiting the achievable resolution by
conventional crystallographic methods. We have used 3D classification to detect distinct conformations of the MloK1 CNBD in
relation to the transmembrane domain (TMD) [38].
Differently from conventional SPA, however, in 2D crystal
particles the protein of interest is always surrounded by neighboring proteins. This means that, in tilted views, the projection of the
neighbors overlaps with the projection of the classification target.
This problem is illustrated in Fig. 4.
Therefore, 3D classification will only work properly if the signal
from the neighbor proteins is subtracted from the particle images.
For more accurate results in signal subtraction of 2D crystal data,
the user should calculate a completely unmasked reconstruction in
frealign-2dx (see Fig. 4). This can be done by setting a reconstruction radius larger than the particle box in the mparameters file and
then running the command frealign_calc_reconstructions from
the refinement directory (see Note 9). Based on this reconstruction,
the user should then define a soft mask focused on the protein of
interest (e.g., the central MloK1 tetramer), and invert this mask to
select everything else that should be subtracted from the experimental projections (i.e., the particles). The focus.postprocess tool
can be used to create such masks (see Subheading 3.7). The signal
278
Ricardo Righetto and Henning Stahlberg
crystal average. Therefore, besides the auto-refinement parameters
(Subheading 3.4.2), it is now recommended to activate the alignment restraints available in frealign-2dx (see Note 8):
– Restraint for Euler angles, in degrees (sigma_angles): standard deviation for a Gaussian restraint on the Euler angles.
– Restraint for x,y shifts, in pixels (sigma_shifts): standard
deviation for a Gaussian restraint on the x, y shifts.
More information about these restraints can be found in the
Supplementary Note 1 of [38]. Upon convergence, a higher quality
map should have been obtained, and a more diverse set of views
should be observed, as shown in Fig. 3. This evidences that particles
from the same 2D crystal are not exactly always in the same orientation. This map, commonly referred to as a “consensus map”
because no 3D classification has been performed (yet), can now
be post-processed as described in Subheading 3.7. The user is also
encouraged to experiment with more advanced features, such as
defocus refinement [32, 39].
3.6 3D Classification
Perhaps the most interesting feature of the single particle method is
its ability to classify heterogeneous data into conformationally
homogeneous classes [35, 52]. Structural variability is also a source
of disorder in 2D crystals, thus limiting the achievable resolution by
conventional crystallographic methods. We have used 3D classification to detect distinct conformations of the MloK1 CNBD in
relation to the transmembrane domain (TMD) [38].
Differently from conventional SPA, however, in 2D crystal
particles the protein of interest is always surrounded by neighboring proteins. This means that, in tilted views, the projection of the
neighbors overlaps with the projection of the classification target.
This problem is illustrated in Fig. 4.
Therefore, 3D classification will only work properly if the signal
from the neighbor proteins is subtracted from the particle images.
For more accurate results in signal subtraction of 2D crystal data,
the user should calculate a completely unmasked reconstruction in
frealign-2dx (see Fig. 4). This can be done by setting a reconstruction radius larger than the particle box in the mparameters file and
then running the command frealign_calc_reconstructions from
the refinement directory (see Note 9). Based on this reconstruction,
the user should then define a soft mask focused on the protein of
interest (e.g., the central MloK1 tetramer), and invert this mask to
select everything else that should be subtracted from the experimental projections (i.e., the particles). The focus.postprocess tool
can be used to create such masks (see Subheading 3.7). The signal
278
Ricardo Righetto and Henning Stahlberg