3.3 Exposure
and Dose
In electron microscopy, dose and exposure are often used interchangeably, whereas strictly, they refer to different quantities
[4]. Exposure measures the number of electrons delivered to the
sample, usually counted in electrons per unit area, while dose refers
to the actual energy absorbed by the sample per unit mass. In
electron microscopy, the exposure can be estimated using a Faraday
cup, the microscope’s phosphor screen, or an evenly exposed image
recorded on a well-calibrated camera. Dose is much more difficult
to quantify due to the complex interactions of chemical composition, crystal size, and beam shape. In X-ray crystallography, the
dose can be calculated using, e.g., RADDOSE [27, 28], but there
are no equivalent programs for electron diffraction yet. If the
elemental composition and solvent content of the sample is
known, the dose in MicroED can instead be approximated from
the exposure at a given acceleration voltage.
Ignoring the effects of the crystalline structure of the sample
and the vitrified solvent around it, the collision stopping power of a
typical protein sample [12] for a 200 keV electron is
2.76 MeV cm
2 g
À1 [29]. In a typical protein sample with density
ρ ¼ 1.17 g cm
À3 , a 200 keV electron loses 323 eV μm
À1 to the
sample. The relative energy loss of the electron during its traversal
of the sample is small, even for the thickest protein crystals, and the
change in collision stopping power is negligible. Because both total
deposited energy and mass increase linearly with the thickness of
the sample, the dose does not dependent on crystal thickness, and a
single 200 keV electron hitting a 1 A ˚ 2 surface of a crystal will deliver
a dose of 4.41 Â 10
6 J kg
À1 (4.41 MGy). For 300 keV electrons, an
exposure of 1 e
À A ˚ À2 corresponds to 3.72 MGy; this reflects the
smaller number of scattering events at higher acceleration voltages,
such that a higher exposure is required to maintain the same signal
when the energy of the incident electrons is increased [5].
Several estimates of the dose tolerances of biomolecules are
available in the literature [30, 31]. These numbers often constitute
limits which are rarely attained in practice. The actual tolerance will
vary significantly from one sample to another [32, 33], depend on
precisely how damage is quantified [34], and additionally change
with the resolution interval under consideration. Finding the ideal
exposure can be posed as an optimization problem with respect to
information content: if the dose is too small, the signal may not be
strong enough to be integrated accurately; if it is too large, the
sample may succumb to radiation damage before the desired
amount of information has been recovered. The optimum dose
will lie somewhere between these extrema. For MicroED, the
proper exposure will often have to be established empirically.
3.4 Multicrystal Data
Collection
Even within a dataset recorded from a fully illuminated, single
crystal, the intensities on frames recorded at different timepoints
during data collection are not on a common scale. Effects due to
314
Johan Hattne
and Dose
In electron microscopy, dose and exposure are often used interchangeably, whereas strictly, they refer to different quantities
[4]. Exposure measures the number of electrons delivered to the
sample, usually counted in electrons per unit area, while dose refers
to the actual energy absorbed by the sample per unit mass. In
electron microscopy, the exposure can be estimated using a Faraday
cup, the microscope’s phosphor screen, or an evenly exposed image
recorded on a well-calibrated camera. Dose is much more difficult
to quantify due to the complex interactions of chemical composition, crystal size, and beam shape. In X-ray crystallography, the
dose can be calculated using, e.g., RADDOSE [27, 28], but there
are no equivalent programs for electron diffraction yet. If the
elemental composition and solvent content of the sample is
known, the dose in MicroED can instead be approximated from
the exposure at a given acceleration voltage.
Ignoring the effects of the crystalline structure of the sample
and the vitrified solvent around it, the collision stopping power of a
typical protein sample [12] for a 200 keV electron is
2.76 MeV cm
2 g
À1 [29]. In a typical protein sample with density
ρ ¼ 1.17 g cm
À3 , a 200 keV electron loses 323 eV μm
À1 to the
sample. The relative energy loss of the electron during its traversal
of the sample is small, even for the thickest protein crystals, and the
change in collision stopping power is negligible. Because both total
deposited energy and mass increase linearly with the thickness of
the sample, the dose does not dependent on crystal thickness, and a
single 200 keV electron hitting a 1 A ˚ 2 surface of a crystal will deliver
a dose of 4.41 Â 10
6 J kg
À1 (4.41 MGy). For 300 keV electrons, an
exposure of 1 e
À A ˚ À2 corresponds to 3.72 MGy; this reflects the
smaller number of scattering events at higher acceleration voltages,
such that a higher exposure is required to maintain the same signal
when the energy of the incident electrons is increased [5].
Several estimates of the dose tolerances of biomolecules are
available in the literature [30, 31]. These numbers often constitute
limits which are rarely attained in practice. The actual tolerance will
vary significantly from one sample to another [32, 33], depend on
precisely how damage is quantified [34], and additionally change
with the resolution interval under consideration. Finding the ideal
exposure can be posed as an optimization problem with respect to
information content: if the dose is too small, the signal may not be
strong enough to be integrated accurately; if it is too large, the
sample may succumb to radiation damage before the desired
amount of information has been recovered. The optimum dose
will lie somewhere between these extrema. For MicroED, the
proper exposure will often have to be established empirically.
3.4 Multicrystal Data
Collection
Even within a dataset recorded from a fully illuminated, single
crystal, the intensities on frames recorded at different timepoints
during data collection are not on a common scale. Effects due to
314
Johan Hattne