mode remove contributions from inelastically scattered electrons [76] that cause
image blur. The combination of moderate accelerating voltage (300 kV) and an
energy filter therefore compensate for the poor SNR characteristics inherent in frozen
specimens. They also facilitate tomography on slightly thicker specimens but with a
concomitant trade-off in resolution. Note that, for unstained (cryo) specimens, SNR is
discussed in preference to contrast, the latter of which can be enhanced artificially.
Contrast enhancement does not suppress noise.
In cryo- electron tomography, the radiation sensitivity of the frozen-hydrated
specimen means that tomograms comprise only about half as many projections as
for plastic sections, collected at increments of 1.5 or 2 degrees. Further complications arise from the increase in apparent thickness of the slab-shaped sample
during tilting and ultimately, the limited range of the tilt holder/sample grid combination. This gives rise to the so-called “missing (information) wedge”, which
describes the unsampled region in Fourier space. The information deficit results in a
real-space artefact whereby features parallel to the electron beam appear to be
smeared, while features orthogonal to the beam are unresolved. A ± 70° tilt range
provides 78% of the available information but with an anisotropic, i.e. biased,
representation of information. A dual-axis acquisition scheme [77, 78] addresses the
fundamental geometric bias. In this acquisition scheme, a complementary series of
projections is acquired from the orthogonal axis and merged with the initial tilt
series. If the second tilt series covers a similar angular range, dual-axis tomography
improves sampling to 93% of the corresponding Fourier space [79], and the resolution is more isotropic. Long, filamentous structures such as microtubules are
resolved equally well, regardless of whether they are parallel to the tilt axis or
perpendicular.
Much of the preceding discussion is applicable to plastic sections, although the
beam damage experienced is more subtle in the sense that it is deceptively gradual
and to the novice, might go unnoticed. It is not less important. For
plastic-embedded sections, a tomographic tilt-series may comprise more than 150
separate projections collected at 1° increments. If we assume that the sample
thickness remains constant, 150 similar exposures corresponds to 150 times the
electron dose for a single projection. In practise, it is much higher due to the
increased beam cross-section at progressively higher tilt, where thickness is related
to the cosine of the tilt angle. Thus, although the specimen might not appear to be
distorting, it is in fact undergoing continuous mass loss and shrinkage. Comparison
of projections 1 and 150 will show marked differences attributed to non-uniform
shrinkage alone. Apart from inadequately representing the specimen, alignment of
the projections will be difficult unless the projections are modified after acquisition
by applying stretching factors, and by accounting for non-linear movements of gold
particles. Prior to acquisition, the beam is spread for 10–15 min to expose the
region of interest and the surrounding area to a higher flux of electrons. This allows
the region of interest to undergo gross changes prior to recording the tilt-series.
Even so, further changes cannot be halted entirely—only the rate of change slows.
To minimise further changes during data acquisition, the low-dose acquisition
1 Electron Tomography: A Primer
11
image blur. The combination of moderate accelerating voltage (300 kV) and an
energy filter therefore compensate for the poor SNR characteristics inherent in frozen
specimens. They also facilitate tomography on slightly thicker specimens but with a
concomitant trade-off in resolution. Note that, for unstained (cryo) specimens, SNR is
discussed in preference to contrast, the latter of which can be enhanced artificially.
Contrast enhancement does not suppress noise.
In cryo- electron tomography, the radiation sensitivity of the frozen-hydrated
specimen means that tomograms comprise only about half as many projections as
for plastic sections, collected at increments of 1.5 or 2 degrees. Further complications arise from the increase in apparent thickness of the slab-shaped sample
during tilting and ultimately, the limited range of the tilt holder/sample grid combination. This gives rise to the so-called “missing (information) wedge”, which
describes the unsampled region in Fourier space. The information deficit results in a
real-space artefact whereby features parallel to the electron beam appear to be
smeared, while features orthogonal to the beam are unresolved. A ± 70° tilt range
provides 78% of the available information but with an anisotropic, i.e. biased,
representation of information. A dual-axis acquisition scheme [77, 78] addresses the
fundamental geometric bias. In this acquisition scheme, a complementary series of
projections is acquired from the orthogonal axis and merged with the initial tilt
series. If the second tilt series covers a similar angular range, dual-axis tomography
improves sampling to 93% of the corresponding Fourier space [79], and the resolution is more isotropic. Long, filamentous structures such as microtubules are
resolved equally well, regardless of whether they are parallel to the tilt axis or
perpendicular.
Much of the preceding discussion is applicable to plastic sections, although the
beam damage experienced is more subtle in the sense that it is deceptively gradual
and to the novice, might go unnoticed. It is not less important. For
plastic-embedded sections, a tomographic tilt-series may comprise more than 150
separate projections collected at 1° increments. If we assume that the sample
thickness remains constant, 150 similar exposures corresponds to 150 times the
electron dose for a single projection. In practise, it is much higher due to the
increased beam cross-section at progressively higher tilt, where thickness is related
to the cosine of the tilt angle. Thus, although the specimen might not appear to be
distorting, it is in fact undergoing continuous mass loss and shrinkage. Comparison
of projections 1 and 150 will show marked differences attributed to non-uniform
shrinkage alone. Apart from inadequately representing the specimen, alignment of
the projections will be difficult unless the projections are modified after acquisition
by applying stretching factors, and by accounting for non-linear movements of gold
particles. Prior to acquisition, the beam is spread for 10–15 min to expose the
region of interest and the surrounding area to a higher flux of electrons. This allows
the region of interest to undergo gross changes prior to recording the tilt-series.
Even so, further changes cannot be halted entirely—only the rate of change slows.
To minimise further changes during data acquisition, the low-dose acquisition
1 Electron Tomography: A Primer
11