not much thicker so as not to lower the contrast; (ii) particles that form a dense population, but no or not
too many aggregates, and are well distributed across the field of view (crowding toward the edges of
the film is a frequent problem); and (iii) in general, particles that adopt random orientations, so that
each projection angle can be thoroughly sampled.
In all cases, the grid holder can be tilted, so that the same specimen can be imaged at various
angles with respect to the electron beam. This feature can be used, in particular, to reconstruct 3D
models from pairs of projection images collected at a known angle to each other (random conical tilt
reconstruction method; see Radermacher et al. 1987; Frank 2006).
Images of NS specimens have enough contrast so that they can be easily identified and studied
individually. They provide very useful information about the purity of the sample and its homogeneity:
presence of aggregates or, on the contrary, fragmentation of the complexes, overall shape, internal
symmetry, conformational variability, etc. (reviewed in Ohi et al. 2004; Vahedi-Faridi et al. 2013).
Studies in NS can stop here or proceed to a limited analysis such as the search for rotational symmetry.
Most of the time, however, the noise is averaged out by combining numerous images of projections at a
similar angle, and tomography is applied to reconstructing a 3D structure: computer programs class
images according to the direction of the projection, orient them in-plane with respect to one another,
average them, and reconstruct a 3D volume by combining all orientations (see e.g. Cheng et al. 2015).
The resolution of the resulting 3D structures is limited to the 15–20 Å range, and no information is
obtained on the internal variations of electron density of the object, its shape only being preserved. In
addition, NS introduces artifacts, such as specimen flattening. 3D reconstructions of negatively stained
specimens are, these limitations notwithstanding, an extremely useful first step in the EM analysis of a
macromolecule.
In cryo-EM, the contrast is very limited (the electron density of the ice film is typically ~80% that
of the particle; Henderson and McMullan 2013), and individual images are not interpretable in details
(Fig. 12.3B). Numerous factors contribute to degrading their quality, among which artifacts arising
from aberrations in the electron optics, the need to defocus to generate contrast, the high levels of noise
arising from the finite number of electrons used (typically 20 electrons per Å
2 , most of which go
through unperturbed), movement of the particles during exposure, and the limited efficiency of the
photographic film or charge-coupled device (CCD) detectors used until very recently. Nevertheless,
image correction, computerized identification, and analysis of hundreds of thousands if not millions
of images (Fig. 12.4) permit, provided the objects are large enough to be aligned – theoretically
!50–100 kDa (Henderson 1995; Henderson and McMullan 2013), in practice, !150 kDa – to
reconstruct 3D structures to a resolution that, until recently, could reach, in the best cases, 6–7 Å
(for recent reviews, see e.g. Bai et al. 2015a; Carazo et al. 2015; Cheng 2015; Cheng et al. 2015;
Nogales and Scheres 2015; De Zorzi et al. 2016; Vinothkumar and Henderson 2016). Recent technical
progress (including the use of a Volta phase plate) shows that this size limit can be lowered (see
Khoshouei et al. 2017), whereas the resolution has now reached the critical level at which large amino
acid side-chains can be identified (see below, § 12.2.3).
Critical assessment of the resolution achieved in cryo-EM reconstructions is a complicated issue.
It seems to be best addressed by comparing quantitatively the Fourier transforms of two models
extracted independently from two sets of images (the so-called “gold standard” procedure), taking as a
criterion the resolution at which Fourier shell correlation (FSC) falls below 0.143 (Rosenthal and
Henderson 2003; Scheres and Chen 2012; see Box 12.1), a procedure that has been agreed upon by
microscopists (Henderson et al. 2012) and is now in common use. Note that it is not rare that, either as a
consequence of the alignment procedure or because they are more or less disordered, some regions of
the maps are better resolved than others.
536
12 The Use of Amphipols for Electron Microscopy
too many aggregates, and are well distributed across the field of view (crowding toward the edges of
the film is a frequent problem); and (iii) in general, particles that adopt random orientations, so that
each projection angle can be thoroughly sampled.
In all cases, the grid holder can be tilted, so that the same specimen can be imaged at various
angles with respect to the electron beam. This feature can be used, in particular, to reconstruct 3D
models from pairs of projection images collected at a known angle to each other (random conical tilt
reconstruction method; see Radermacher et al. 1987; Frank 2006).
Images of NS specimens have enough contrast so that they can be easily identified and studied
individually. They provide very useful information about the purity of the sample and its homogeneity:
presence of aggregates or, on the contrary, fragmentation of the complexes, overall shape, internal
symmetry, conformational variability, etc. (reviewed in Ohi et al. 2004; Vahedi-Faridi et al. 2013).
Studies in NS can stop here or proceed to a limited analysis such as the search for rotational symmetry.
Most of the time, however, the noise is averaged out by combining numerous images of projections at a
similar angle, and tomography is applied to reconstructing a 3D structure: computer programs class
images according to the direction of the projection, orient them in-plane with respect to one another,
average them, and reconstruct a 3D volume by combining all orientations (see e.g. Cheng et al. 2015).
The resolution of the resulting 3D structures is limited to the 15–20 Å range, and no information is
obtained on the internal variations of electron density of the object, its shape only being preserved. In
addition, NS introduces artifacts, such as specimen flattening. 3D reconstructions of negatively stained
specimens are, these limitations notwithstanding, an extremely useful first step in the EM analysis of a
macromolecule.
In cryo-EM, the contrast is very limited (the electron density of the ice film is typically ~80% that
of the particle; Henderson and McMullan 2013), and individual images are not interpretable in details
(Fig. 12.3B). Numerous factors contribute to degrading their quality, among which artifacts arising
from aberrations in the electron optics, the need to defocus to generate contrast, the high levels of noise
arising from the finite number of electrons used (typically 20 electrons per Å
2 , most of which go
through unperturbed), movement of the particles during exposure, and the limited efficiency of the
photographic film or charge-coupled device (CCD) detectors used until very recently. Nevertheless,
image correction, computerized identification, and analysis of hundreds of thousands if not millions
of images (Fig. 12.4) permit, provided the objects are large enough to be aligned – theoretically
!50–100 kDa (Henderson 1995; Henderson and McMullan 2013), in practice, !150 kDa – to
reconstruct 3D structures to a resolution that, until recently, could reach, in the best cases, 6–7 Å
(for recent reviews, see e.g. Bai et al. 2015a; Carazo et al. 2015; Cheng 2015; Cheng et al. 2015;
Nogales and Scheres 2015; De Zorzi et al. 2016; Vinothkumar and Henderson 2016). Recent technical
progress (including the use of a Volta phase plate) shows that this size limit can be lowered (see
Khoshouei et al. 2017), whereas the resolution has now reached the critical level at which large amino
acid side-chains can be identified (see below, § 12.2.3).
Critical assessment of the resolution achieved in cryo-EM reconstructions is a complicated issue.
It seems to be best addressed by comparing quantitatively the Fourier transforms of two models
extracted independently from two sets of images (the so-called “gold standard” procedure), taking as a
criterion the resolution at which Fourier shell correlation (FSC) falls below 0.143 (Rosenthal and
Henderson 2003; Scheres and Chen 2012; see Box 12.1), a procedure that has been agreed upon by
microscopists (Henderson et al. 2012) and is now in common use. Note that it is not rare that, either as a
consequence of the alignment procedure or because they are more or less disordered, some regions of
the maps are better resolved than others.
536
12 The Use of Amphipols for Electron Microscopy