Beam-induced motion of vitrified specimen on holey carbon film.

Abstract

The contrast observed in images of frozen-hydrated biological specimens prepared for electron cryo-microscopy falls significantly short of theoretical predictions. In addition to limits imposed by the current instrumentation, it is widely acknowledged that motion of the specimen during its exposure to the electron beam leads to significant blurring in the recorded images. We have studied the amount and direction of motion of virus particles suspended in thin vitrified ice layers across holes in perforated carbon films using exposure series. Our data show that the particle motion is correlated within patches of 0.3-0.5 μm, indicating that the whole ice layer is moving in a drum-like motion, with accompanying particle rotations of up to a few degrees. Support films with smaller holes, as well as lower electron dose rates tend to reduce beam-induced specimen motion, consistent with a mechanical effect. Finally, analysis of movies showing changes in the specimen during beam exposure show that the specimen moves significantly more at the start of an exposure than towards its end. We show how alignment and averaging of movie frames can be used to restore high-resolution detail in images affected by beam-induced motion.

Full text

Page 1

Beam-induced motion of vitrified specimen on holey carbon film
Axel F. Brilota, James Z. Chena,b,1, Anchi Chengc, Junhua Pand, Stephen C. Harrisond,f, Clinton S. Potterc,
Bridget Carragherc, Richard Hendersone, Nikolaus Grigorieffa,b,⇑
aDepartment of Biochemistry, Rosenstiel Basic Medical Sciences Research Center, Brandeis University, MS029, 415 South Street, Waltham, MA 02454, USA
bHoward Hughes Medical Institute, Brandeis University, MS029, 415 South Street, Waltham, MA 02454, USA
cDepartment of Cell Biology, The Scripps Research Institute, La Jolla, CA 92037, USA
dHarvard Medical School and Children’s Hospital, Boston, MA 02115, USA
eMRC Laboratory of Molecular Biology, Hills Road, Cambridge CB2 0QH, UK
fDepartment of Biological Chemistry and Molecular Pharmacology and Howard Hughes Medical Institute, Harvard Medical School, Boston, MA, USA
a r t i c l ei n f o
Article history:
Received 9 December 2011
Received in revised form 6 February 2012
Accepted 7 February 2012
Available online 16 February 2012
Keywords:
Single-particle electron cryo-microscopy
Beam-induced motion
Radiation damage
Dose rate
High-resolution EM
a b s t r a c t
The contrast observed in images of frozen-hydrated biological specimens prepared for electron cryo-
microscopy falls significantly short of theoretical predictions. In addition to limits imposed by the current
instrumentation, it is widely acknowledged that motion of the specimen during its exposure to the elec-
tron beam leads to significant blurring in the recorded images. We have studied the amount and direction
of motion of virus particles suspended in thin vitrified ice layers across holes in perforated carbon films
using exposure series. Our data show that the particle motion is correlated within patches of 0.3–0.5 lm,
indicating that the whole ice layer is moving in a drum-like motion, with accompanying particle rotations
of up to a few degrees. Support films with smaller holes, as well as lower electron dose rates tend to
reduce beam-induced specimen motion, consistent with a mechanical effect. Finally, analysis of movies
showing changes in the specimen during beam exposure show that the specimen moves significantly
more at the start of an exposure than towards its end. We show how alignment and averaging of movie
frames can be used to restore high-resolution detail in images affected by beam-induced motion.
? 2012 Elsevier Inc. All rights reserved.
1. Introduction
Electron cryo-microscopy (cryo-EM) can be used to visualize the
three-dimensional (3D) structure of a broad variety of specimens,
including two-dimensional (2D) crystals (e.g., Gonen et al., 2004;
Henderson et al., 1986), helical specimens (for example, Ge and
Zhou, 2011; Miyazawa et al., 1999; Sachse et al., 2007; Yonekura
et al., 2003) and isolated (single) particles. In recent years the
application of the single particle approach has led to 3D reconstruc-
tions of a number of highly symmetrical virus particles at near-
atomic resolution (4 Å or better, see Grigorieff and Harrison,
2011) for a recent review). Despite this success, it is commonly
acknowledged that contrast in images of vitrified specimens falls
significantly short of predicted physical limits (Glaeser, 1999;
Henderson, 1995). Physical limits are imposed by the radiolysis of
biological molecules caused by the high-energy electron beam
which limits the electron dose to 5–10 electrons/Å2(Baker et al.,
2010; Henderson, 1992, 1995) if high-resolution features are to
be preserved. Under idealized conditions, particle images are
predicted to contain sufficient signal to obtain a 3D reconstruction
at 3 Å resolution by averaging of a few thousand molecular images
(Glaeser, 1999; Henderson, 1995). In practice however, the recent
reconstructionsof particlesatnear-atomicresolutionhave required
averaging signal from several 100,000 to over 10 million images of
subunits or asymmetric units (Grigorieff and Harrison, 2011). The
contrast transfer function of the electron microscope, image
detector noise and motion in the specimen induced by the incident
electron beam all contribute to the loss of contrast in cryo-EM
images (for a recent review, see Glaeser and Hall, 2011). The first
two issues concern limitations of current instrumentation and are
being addressed by technological improvements (Cambie et al.,
2007; Danev and Nagayama, 2001; Majorovits et al., 2007; McMu-
llan et al., 2009; Milazzo et al., 2011, 2005; Muller et al., 2010).
Beam-induced specimen motion is thought to be caused by the
reaction of the specimen to the high-energy electron beam, result-
ing in a build-up of positive charge on the specimen (Brink et al.,
1998) and radiolysis of the sample and vitrified embedding med-
ium (Glaeser, 2008; Glaeser and Taylor, 1978). Charge build-up
leads to a weak deflection of the electron beam that can blur the
image, especially of tilted samples in which the component of the
deflection perpendicular to the beam is not zero (Glaeser and
1047-8477/$ - see front matter ? 2012 Elsevier Inc. All rights reserved.
doi:10.1016/j.jsb.2012.02.003
⇑Corresponding author at: Department of Biochemistry, Rosenstiel Basic Medical
Sciences Research Center, Brandeis University, MS029, 415 South Street, Waltham,
MA 02454, USA. Fax: +1 781 736 2419.
E-mail address: niko@brandeis.edu (N. Grigorieff).
1Present address: Department of Biology, Massachusetts Institute of Technology,
77 Massachusetts Avenue, Cambridge, MA 02139, USA.
Journal of Structural Biology 177 (2012) 630–637
Contents lists available at SciVerse ScienceDirect
Journal of Structural Biology
journal homepage: www.elsevier.com/locate/yjsbi

Page 2

Downing, 2004; Henderson, 1992). Radiolysis of the specimen is
thought to lead to a build-up of internal pressure as the radiolysis
products take up more space than the original molecules (Glaeser,
2008). The mechanical stress is sufficiently high to cause specimen
deformations (and ultimately breakdown of the entire fabric – so-
called bubbling), again blurring the final image. In a recent study,
Glaeser and Henderson (Glaeser et al., 2011) studied the beam-in-
ducedmotionof paraffin2Dcrystalssupportedbya continuouscar-
bon film and showed that film thicknesses greater than 35 nm
significantly reduced the observed motion, thereby improving the
fraction of images with strong high-resolution signal. These exper-
iments further corroborate mechanical instability as one of the
leading factors allowing beam-induced motion. Unfortunately, the
use of a continuous carbon film is often not ideal for non-crystalline
single particles as it adds background to an image and can induce
preferred particle orientation.
We have recently investigated an imaging protocol in which the
electron dose rate was varied, to image single particles embedded
in ice over holes in a carbon support film (Chen et al., 2008). These
experiments suggested that a lower dose rate allows for a higher
total dose before bubbling occurs, but did not clearly demonstrate
that beam-induced motion prior to bubbling was reduced. In the
present study, we have investigated beam-induced motion by
monitoring positions and orientations of rotavirus double-layer
particles (DLPs). These particles are very regular and have a molec-
ular mass of 70 MDa, allowing alignment with a reference struc-
ture with accuracies of about 0.2 Å and 0.2?, for translational and
orientational alignments, respectively (Zhang et al., 2008). We col-
lected exposure series from holes inside perforated carbon films
containing rotavirus embedded in ice, varying dose rate and hole
size. Changes in the particle orientations between exposures were
then taken as an indication for specimen motion. In a second set of
experiments, we investigated the timing of the beam-induced mo-
tion during exposures by recording movies using a new type of
camera, a direct electron detector. Particles visible in individual
frames or frame averages were analyzed in terms of their orienta-
tional and translational changes during the exposure.
2. Materials and methods
2.1. Sample preparation
Rotavirus DLPs were prepared as described (Street et al., 1982).
Three microliters of sample with a concentration of 2.5–4 mg/ml
was applied to Quantifoil?or C-flat™ grids and plunge-frozen using
either a Gatan CP3 plunger (for all exposure series experiments) or
anFEIVitrobotMark2(forallmovieexperiments),witha4or6 sblot
time and at relative humidity between 65% and 80%. The following
grid types were used: Quantifoil?1.2/1.3 Cu 400 mesh (measured
hole size = 1.6 lm; this difference between nominal and measured
hole size was observed for all grids in this batch), C-flat™ 0.6/
2.0 Cu 400 mesh (measured hole size = 0.6 lm), C-flat™ 1.0/1.0 Cu
400 mesh (measured hole size = 1.0 lm), and C-flat™ 1.2/1.3 Cu
400 mesh(measuredholesize = 1.2 lm).TheQuantifoil?gridswere
cleaned prior to their use by immersion in a small amount of ethyl
acetateontopoffilterpaperinaglassPetridish.Immediatelybefore
plunging, all grids were subject to glow discharge for 45 s at 20 mA.
2.2. Electron microscopy – exposure series
Images were collected on a Gatan US4000 4 k ? 4 k CCD camera
mounted on an FEI TF30 electron microscope operating at 300 kV,
and using a side-entry Gatan 626 cryo holder. The calibrated mag-
nification was 49,053, giving a pixel size on the specimen of 3.06 Å.
Magnification calibration was performed using a DLP reference
structure (Zhang et al., 2008) and maximizing correlation coeffi-
cients found by varying the pixel size during Frealign (Grigorieff,
2007) runs (see below). We used an underfocus of 2.5–3.5 lm,
and an electron dose per exposure of 8 electrons/Å2. Images in each
series were taken about 60 s apart, and occasionally 600 s apart to
test if charging was a factor in the observed particle motions (see
Section 3). The exposed area on the grid was centered on each hole
and held approximately constant at 2.0 lm to ensure that the elec-
tron beam made contact with the carbon support film everywhere
around holes. Furthermore, an objective aperture was used for all
exposures. The ice thickness was measured using holes in the ice
layer produced by a focused electron beam with subsequent tilting
to 30? (Wright et al., 2006). The ice next to virus particles was
determined to be between 800 and 1000 Å thick while it was thin-
ner in the center of holes, with the thinnest ice (500 Å) measured in
the center of 1.6 lm holes on Quantifoil?grids.
2.3. Electron microscopy – movies
Images were collected on a Direct Electron DE-12 4 k ? 3 k di-
rect electron detector mounted on an FEI TF20 electron microscope
operating at 200 kV, and using a side-entry Gatan 626 cryo holder.
The calibrated magnification was 17,858, giving a pixel size on the
specimen of 3.36 Å. The underfocus was set between 2.5 and
3.5 lm, and the electron dose per frame was 0.5 electrons/Å2. Mov-
ies were recorded at a rate of 40 frames/s. The exposed area was
adjusted as above and ice thickness measurements were essen-
tially identical to those measured above.
2.4. Image processing – exposure series
Virus particles were semi-automatically selected from the first
image in an exposure series using e2boxer from the EMAN2 process-
ing package (Tang et al., 2007). Micrographs were cross-correlated
to each other using the Spider processing package (Frank et al.,
1996)todeterminetranslationaloffsets,andparticleswereselected
in subsequent exposures by coordinates updated with the offsets.
The defocus for each micrograph was determined using CTFFIND3
(Mindell and Grigorieff, 2003). Particle images were decimated
using 2 ? 2 pixel averaging and aligned in an exhaustive search
with Frealign (Grigorieff, 2007) (Mode = 4 with DANG = 200 and IT-
MAX = 200 using data between 18 and 300 Å resolution), followed
by 10 rounds of refinement (Mode = 1), using a DLP reference struc-
ture (Zhang et al., 2008). The rotation angles and axes of reorienta-
tions experiencedby particles between exposures were determined
using the program Tiltdiff (Henderson et al., 2011). The rotation axis
ofmostparticleswaswithin0.5? inthe imageplane.About6%ofthe
particles were measured to have tilt axes with larger out-of-plane
angles. These were excluded from the analysis to eliminate cases
were the exhaustive parameter search failed. Histograms of the
magnitude of the measured rotation angles were generated with a
bin size of 0.1?. The histograms show a clear positive skewness,
indicating that a simple average and standard deviation might not
be appropriate to characterize the distribution of rotation angles.
We used the computer program distfit from the Theseus package
(Theobald and Wuttke, 2006) to test 20 different distributions and
selected the Rayleigh distribution as the most appropriate accord-
ing to the Bayesian information criterion. The Rayleigh distribution
(Lalitha and Mishra, 1996) is characterized by a single parameter, k
which indicates the maximum (mode) of the distribution:
Rðx;kÞ ¼x
The maximum likelihood estimate of k2is:
1
2N
i¼1
k2e?x2=2k2:
ð1Þ
k2¼
X
N
x2
i
ð2Þ
A.F. Brilot et al./Journal of Structural Biology 177 (2012) 630–637
631

Page 3

and standard error in k2is:
SDðk2Þ ¼
k2
ffiffiffiffiffiffiffiffiffiffiffiffiffi
N ? 1
p
:
ð3Þ
Therefore, standard error in k2is:
SDðkÞ ?SDðk2Þ
2k
¼
k
2
ffiffiffiffiffiffiffiffiffiffiffiffiffi
N ? 1
p
:
ð4Þ
This error is reported together with k in Table 1 for the fitted
distributions.
2.5. Image processing – movies
Raw movie frames were corrected for dark current and gain-
normalized using dark frames and flat fields recorded directly be-
fore each movie and at the beginning of a session, respectively.
Frame averages were calculated using the computer program label
from the MRC image processing package (Crowther et al., 1996).
Particle rotations and translations were analyzed using Frealign
as described above, except using data between 25 and 800 Å.
3. Results
3.1. Exposure series
In a first set of experiments we investigated the influence of
dose rate on beam-induced motion using exposure series. Fig. 1A
shows a field of rotavirus particles prepared using C-flat™ 1.2/
1.3 Cu 400 mesh grids (hole size = 1.2 lm). The image was re-
corded using a dose rate of 20 electrons/Å2/s and the exposure
lasted 0.4 s, giving a total dose of 8 electrons/Å2. Subsequent expo-
sures are shown in panels B–D, indicating no bubbling in the sam-
ple even in the final exposure, after which the total dose was
32 electrons/Å2. Virus particles are clearly visible in all exposures
and could readily be aligned using a reference structure (Zhang
et al., 2008). Using projection matching implemented in the com-
puter program Frealign (Grigorieff, 2007), the orientation of most
of the virus particles in all four exposures were determined suc-
cessfully, as judged by their small angular differences and small
out-of-plane errors (see Materials and methods) determined by
the program Tiltdiff (Henderson et al., 2011). The angular differ-
ences between exposures are plotted as vectors in panels E–G to
show both the magnitude and direction of the rotations. A striking
feature of these plots is the local correlation in the rotation vectors,
extending over a distance of 300–500 nm. This correlation also
indicates that most particle orientations were determined cor-
rectly within a degree or better. Furthermore, it shows that the ori-
entation changes between exposures is not due to the motion of
individual virus particles in the ice, but that it must be due to a
deformation in the ice layer itself, moving the particles with it.
The angular changes of up to 2? seen in panel E were also observed
in many other exposure series. The direction of the rotation re-
verses from one side of the hole to the other, indicating that the
ice undergoes a drum-like motion with a curvature radius of
25 lm and translation of the central region perpendicular to the
specimen plane of about 150 Å (see Section 4). Subsequent expo-
sures produce further particle rotations approximately in the same
Table 1
Rotation angles measured between the first and second exposures in exposure series of rotavirus particles in vitrified ice over holes in perforated carbon support films. The dose
per exposure was 8 electrons/Å2. The dose rate was varied from 2 electrons/Å2/s – 160 electrons/Å2/s and support films with different hole sizes, varying from 0.6 to 1.6 lm were
used.
Dose RateHole = 0.6 lm Hole = 1.0 lmHole = 1.2 lmHole = 1.6 lm
2 e?/Å2/s
20 e?/Å2/s
80 e?/Å2/s
160 e?/Å2/s
0.412 ± 0.015 (N = 187)
0.460 ± 0.013 (N = 307)
0.453 ± 0.013 (N = 302)
0.480 ± 0.017 (N = 210)
0.413 ± 0.014 (N = 211)
0.403 ± 0.013 (N = 257)
0.415 ± 0.013 (N = 261)
0.433 ± 0.014 (N = 255)
0.394 ± 0.008 (N = 599)
0.384 ± 0.006 (N = 879)
0.443 ± 0.007 (N = 889)
0.461 ± 0.010 (N = 538)
0.470 ± 0.008 (N = 783)
0.608 ± 0.010 (N = 919)
0.481 ± 0.009 (N = 694)
0.558 ± 0.011 (N = 678)
Fig.1. Exposure series of rotavirus DLPs in ice in a C-flat™ 1.2/1.3 grid (hole size = 1.2 lm). Four successive exposures are shown in panels A, B, C and D. The total dose applied
to the sample is noted above each exposure. Panels E, F and G show vector plots corresponding to the changes in particle orientations between exposures (exposure numbers
are indicated in the plots). Panel H plots the summed vectors from panels E, F and G.
632
A.F. Brilot et al./Journal of Structural Biology 177 (2012) 630–637

Page 4

direction but their magnitude decreases, suggesting that beam-in-
duced motion is largest during the initial exposure, and that it is
dose-dependent. The particle rotations observed in the ice also ap-
ply to the few virus particles found on the carbon next to the hole
in Fig. 1, indicating that the motion in the ice extends into the
neighboring carbon support film.
We observed the same pattern of correlated particle rotations in
many other exposure series, with vectors pointing either away
from or towards the center of the hole, indicating that the direction
of the drum motion is random. Furthermore, the magnitude of the
observed rotations varied considerably from series to series. To
gain a more quantitative understanding of the orientation changes,
and to investigate the influence of dose rate and hole size, we re-
corded exposure series from 438 holes on 11 grids, yielding a total
of 8089 DLPs for rotation measurements. The dose rates used were
2, 20, 80 and 160 electrons/Å2/s. To vary the hole size, we used
Quantifoil?grids (1.6 lm holes), as well as C-flat™ grids (0.6, 1.0
and 1.2 lm holes). Fig. 2 shows the histograms of measured rota-
tions between the first and second exposures for all tested dose
rates and a hole size of 1.2 lm. To interpret the histograms, we cal-
culated maximum-likelihood fits of the Rayleigh distribution (see
Materials and methods). The fitted Rayleigh distributions are
superimposed on each histogram in Fig. 2, and their distribution
maxima are given in Table 1 (for all conditions tested), together
with their standard errors and the number of particles analyzed.
The results in Fig. 2 and Table 1 show that the magnitude of parti-
cle motion is quite variable and that alterations in dose rate and
hole size do not produce predictable changes in the particle mo-
tions. Clear trends are discernible, however. Particles suspended
over larger holes show larger rotations than particles in smaller
holes. Thus, the largest average rotations for all dose rates were ob-
served with 1.6 lm holes (Quantifoil?grids) while the rotation
averages determined for 1.0 lm holes (C-flat™ grids) were equal
(within error estimates) or smaller than those seen with 1.2 lm
holes (C-flat™ grids). The results for 0.6 lm holes do not appear
to follow this trend which may reflect irregular hole boundaries
and/or a larger contribution of the carbon support to the overall
motion (see Section 4). There is also a clear trend with dose rate,
showing larger rotations as the rate increases, with the highest rate
(160 electrons/Å2/s) producing the largest average rotations in
most cases. We also varied the time between exposures between
60 and 600 s but did not detect noticeable changes in the magni-
tude of beam-induced motion. Analysis of rotations in subsequent
exposures generated similar histograms with rotations of some-
what reduced but still significant magnitude, in agreement with
the observations shown in Fig. 1.
3.2. Movies
Particle rotations of a few degrees during beam exposure result
in movement of peripheral density features of the 700 Å virus par-
ticles by 10 Å or more, thereby eliminating any signal at sub-nano-
meter resolution in the direction of rotational movement. Since
cryo-EM data recorded under similar conditions led to a recon-
struction of the peripheral rotavirus coat proteins at 3.8 Å resolu-
tion (Chen et al., 2009; Settembre et al., 2011; Zhang et al., 2008)
the observed particle rotations must occur in such a way that sig-
nificant high-resolution signal is preserved. The rotation vectors
suggest that the rotation for any given particle occurs around an
approximately fixed axis. Subunits closer to the rotation axis will
move less than those further away from the axis (see Discussion
and conclusions), thereby maintaining stronger high-resolution
contrast in the images. Other particles may not move much at
all, as indicated by the histograms in Fig. 2, and will exhibit even
stronger high-resolution contrast. It is also possible that the parti-
cle rotations do not occur at a constant rate during the exposure.
For example, if particles rotate more rapidly initially and then slow
down during an exposure, blurring in the final image would be less
severe compared to blurring produced by particles that rotate at a
constant rate.
To investigate the timing of the motion, we recorded movies at
40 frames/s of virus specimens prepared using Quantifoil?grids as
these showed the largest beam-induced motions (see Supplemen-
tary material). At a dose rate of 20 electrons/Å2/s, the dose per
frame is only 0.5 electrons/Å2, too low to produce sufficient con-
trast for particle alignment (Movie S1). We therefore calculated
10-frame averages to improve the contrast while still allowing a
temporal resolution of 0.25 s/averaged frame (Movies S2 and S3).
We also made sure that we captured the onset of the exposure
by starting recording before the beam shutter (above the speci-
men) was opened. Therefore, the first few frames of a movie con-
tain dark frames. Fig. 3 shows results for a 1.5 s exposure with
Fig.2. Histograms for particle rotations measured between the first and second exposures of rotavirus DLPs in ice on C-flat™ 1.2/1.3 Cu 400 mesh grids (hole size = 1.2 lm).
The dose applied in each exposure was 8 electrons/Å2. The dose rate varied between 2 and 160 electrons/Å2/s, giving exposure times between 0.05 and 4 s. The histograms
were fitted with a Rayleigh distribution and maximum (k) is indicated in each plot together with the total number of measurements.
A.F. Brilot et al./Journal of Structural Biology 177 (2012) 630–637
633

Page 5

63 raw frames (Movie S1) of which the first three frames were
dark. The vector plots in Fig. 3A–E indicate the incremental rota-
tions between 10-frame averages (see corresponding Movie S2).
The plots exhibit the same overall pattern as those produced from
the exposure series, with local correlations and opposite directions
on opposite sides of the hole. As the exposure progresses, rotation
increments become smaller until they are frequently within the
detection limit of our method (rotations smaller than about 0.2?).
The rotation vectors suggest smaller rotations for particles closer
to the center of the hole, consistent with the above-mentioned
drum-like motion, which predicts a larger rotational motion near
the edge. The largest overall rotational motion during the 1.5 s
exposure of the movie is about 4?, suggesting that the drum motion
in the center of the hole could be up to 250 Å perpendicular to the
specimen plane, producing a curvature of the ice layer with a
12 lm radius. The rotations seen in the movies are larger than
those we observed with most of the exposure series. However, in
the exposure series we were not able to observe the full extent
of the rotations at the beginning of an exposure, only the average
orientation over its entire duration. The larger rotations seen in
the movies therefore do not contradict the results from the expo-
sure series.
We also analyzed the incremental shifts of each particle, shown
in Fig. 3G–K (see corresponding Movie S3). The largest overall
translational motion is about 70 Å (Fig. 3L), part of which may be
attributable to the drum motion (see Discussion and conclusions).
Unlike the rotations, the particle translations are not distributed
roughly symmetrically around the center of the hole. Instead,
translations in the upper right corner are larger than elsewhere
in the hole. Inspection of the movies (Supplementary material,
Movies S2 and S3) shows that the edge of the carbon moves in
the same direction as the adjacent particles. The larger translations
are directed towards the center of the hole, suggesting that the
hole contracts in the direction of the translations (along the diago-
nal from the bottom left to the top right of the image) and that the
distances between particles in the image decrease. We found sim-
ilar patterns of translation in many other movies. A reduction in
distance was also observed previously in images of gold particles
in ice (Chen et al., 2008) and is most consistent with a drum mo-
tion of the ice layer that is induced by a shrinking dimension of
the hole, rather than an expansion of the ice. Hence, deformation
of the carbon support film contributes to the overall particle mo-
tion, consistent with the observed rotation of the virus particles
found on the carbon next to the hole in Fig. 1.
4. Discussion and conclusions
4.1. Beam-induced motion is caused by changes in the ice layer and
carbon support film
Beam-induced specimen motion has long been recognized as
one of the main factors attenuating high-resolution signal in
cryo-EM (Bottcher, 1995; Bullough and Henderson, 1987; Glaeser
et al., 2011; Henderson, 1992). The main cause of this motion is
beam damage occurring to the specimen as it is exposed to the
high-energy electron beam (Glaeser, 2008; Glaeser and Taylor,
1978; Glaeser et al., 2011) although charging may also play a role
(Glaeser and Downing,2004;
mechanical stability of the continuous support film used with 2D
crystals reduces or entirely avoids beam-induced motion, but an
equivalent solution has not yet been found for single particles sus-
pended in ice layers over holes in the perforated carbon film. In the
present study we investigated the nature of beam-induced motion
of particles in ice over holes. Our analysis of images obtained dur-
ing an exposure series indicates that the ice inside the holes under-
goes drum-like motion with a curvature radius of 12–25 lm and
translation of the central region perpendicular to the specimen
plane of 150–250 Å. In earlier studies, it was observed that the
ice layer spanning the holes in perforated carbon films assumes a
dome shape (Wright et al., 2006). This is consistent with the pres-
ent study, which implies that the observed dome shape was at
least partially induced by a drum motion. It is likely that this drum
motion is caused to some extent by an increase in internal pressure
in the ice due to the molecular radicals generated by radiolysis,
leading to an expansion of the ice layer as previously suggested
(Glaeser, 2008). However, a significant amount of additional mo-
tion is produced by changes in the carbon support film. This is
clearly indicated by the particle translations during beam exposure
shown in Fig. 3. The changes appear to be anisotropic, reducing the
hole dimension predominantly in one direction. The hole analyzed
Henderson,1992).Increased
Fig.3. Movie frame averages of rotavirus DLPs in ice on a Quantifoil?grid (hole
size = 1.6 lm). The movie was recorded for 1.5 s at 40 frames/s using the DE-12
direct electron detector (Direct Electron). 10-frame averages of the resulting 60
frames were subjected to the same analysis as the exposure series. Vector plots in
panels A–E indicate particle rotations measured between frame averages. The total
average of the movie is shown in the background in A and the summed rotations
from panels A–E are shown in F. Panels G–K show vector plots indicating the
particle translations measured between frame averages. Panel L shows the summed
translations from panels G–K. The square highlights the area shown as a close-up in
Fig. 5. The movie is further analyzed in Supplementary material (Movies S1–S3).
634
A.F. Brilot et al./Journal of Structural Biology 177 (2012) 630–637

Page 6

in Fig. 3 changes its dimension by about 50 Å (about 0.3%) in the
direction of the largest translations. In the case shown in Fig. 3,
the main cause of the observed beam-induced motion must there-
fore be due to changes in the carbon support. It is unlikely that
these changes are the result of specimen heating by the electron
beam which are estimated for a dose rate of 20 electrons/Å2/s to
be between 0.2 K (Egerton and Rauf, 1999) and 0.5 K (Dietrich
et al., 1978), leading to motions between 0.03 and 0.05 Å (assum-
ing a thermal expansion coefficient of amorphous carbon of
5 ? 10?6/K, Booy and Pawley, 1993; Vonck, 2000) over a length
of the beam diameter (2 lm). Instead, the changes may affect the
atomic structure of the support film and explain why pre-irradia-
tion of the support film (before application of the sample) with
an electron beam may reduce beam-induced motion (Ge and Zhou,
2011; Miyazawa et al., 1999) (see below). Additional experiments
are required to quantify how much beam-induced motion can be
reduced by pre-irradiation of the carbon support, and to determine
if a support film with higher mechanical stability (Glaeser et al.,
2011) could also reduce the motion of the particles in the holes.
4.2. Deformations in the ice and carbon layers are plastic
The total dose used in the movie in Fig. 3 was 30 electrons/Å2.
Our experiments with exposure series indicate that subsequent
exposures further increase the particle rotations, and with it the
drum motion. The amount of motion therefore depends on the to-
tal dose applied to the sample. On the other hand, variation of the
time between exposures did not significantly alter the amount of
motion. This is inconsistent with a mechanism of beam-induced
motion that depends on the build-up of charge on the specimen.
Although we expect charging to occur as with any other cryo spec-
imen, the charge drains after a few minutes (Brink et al., 1998). It is
therefore likely that the deformation of the ice layer and carbon
support is predominantly plastic, as one might expect for amor-
phous media such as vitrified ice and amorphous carbon. Since
dose rate affects the magnitude of the motion, it appears that a
lower dose rate reduces the strain induced in the specimen some-
what. The origin of the strain is not entirely clear. As previously
discussed (Glaeser, 2008), radiolysis of the water and macromole-
cules likely leads to increased pressure inside the ice layer. A lower
dose rate may reduce the pressure by giving molecular radicals
more time to diffuse to the ice surface where they are released into
the microscope vacuum. However, the carbon support is much
more resistant to radiolysis and the induced strain in the carbon
must have a different origin. The effectiveness of pre-irradiation
in reducing beam-induced motion suggests that the electron beam
has an ‘annealing effect’ on the carbon layer, allowing it to relax
into a more stable configuration. The annealing will also depend
on the energy input by the beam and hence the dose rate. Both pro-
cesses (pressure increase due to radiolysis, relaxation by anneal-
ing) are consistent with the initial larger motion at the beginning
of the exposure, followed by gradual relaxation into a new state
that is stable under the beam. In this new state, the ice layer will
have undergone a drum-like motion, producing the observed par-
ticle rotations and translations. When the exposure ends, the con-
centration of radicals will drop again, presumably leading to some
relaxation of the ice layer. However, as the results from the expo-
sure series show, this does not restore the latter to its original state
as the annealing of the carbon and the straining of the ice lead to
plastic deformation. Renewed exposure to the electron beam re-
starts the process and produces motion almost as large as the first
exposure. As the specimen is repeatedly exposed (or exposed for a
long time) to the electron beam, annealing of the carbon will pro-
gress to a more stable state. The events occurring during beam
exposure are summarized in the schematic in Fig. 4.
4.3. Loss of high-resolution contrast is caused by particle rotation and
translation
There is considerable variability in the amount of motion ob-
served between different experiments (see Fig. 2), on average
about 0.3?. The high-resolution detail will therefore be better pre-
served for some particles compared with others. The variability is
presumably due to small random differences between holes and
the ice in them. Further variability is due to the drum motion,
which will lead to little or no particle motion in the center of the
hole if it occurs perpendicular to the image plane, while rotations
and some translations will be noticeable for particles closer to the
edge of the hole. If the specimen is tilted with respect to the image
plane, the drum motion will produce additional translations and
hence blurring. For example, if the specimen is tilted by 30?, the
translation in the image plane will be 75–130 Å for a drum motion
of 150–250 Å, sufficient to obliterate all high-resolution contrast.
Even at a small tilt angle of 3? which can occur unintentionally
(Vonck, 2000), the translation in the image plane is still 7–13 Å,
which will produce blurring corresponding to a temperature factor
(B-factor) of about 4000–13,000 Å2(Jensen, 2001) (B-factors are gi-
ven in X-ray crystallographic notation). Additional blurring arises
due to particle rotations. If we assume an average rotation angle
of 2? we expect the signal of coat molecules within a radius of
about 100 Å from the axis to be attenuated by a B-factor of about
450 Å2(Jensen, 2001), which is the average B-factor determined
for the reference rotavirus reconstruction (Zhang et al., 2008).
The surface area near the rotation axis included within this radius
is about 1/50 of the total area. Therefore, if rotation alone were
responsible for the signal loss, we would expect to obtain a resolu-
Fig.4. Schematic of beam-induced motion. (A) Particles are suspended in vitrified
ice over a hole. (B) When the electron beam illuminates the sample, the carbon
support film deforms such that the size of the hole shrinks by a small amount. At
the same time, radiolysis of the solvent and macromolecules produces radicals that
increase the pressure inside the ice layer. Both changes cause a drum motion of the
ice layer that lead to particle rotations and translations. The direction of the drum
motion is random. (C) After the beam is turned off, the changes induced in the
specimen remain (plastic deformation). This model does not consider specimen
charging which could add to the blurring of images.
A.F. Brilot et al./Journal of Structural Biology 177 (2012) 630–637
635

Page 7

tion of 3 Å from images of 50,000 to 500,000 molecules, still less
than what has been observed in the recent virus reconstructions
at near-atomic resolution. The blurring factor due to translation
can easily account for the additional loss of signal to account for
the observed requirement of several 100,000 to over 10 million
molecules (Grigorieff and Harrison, 2011).
4.4. Beam-induced motion depends on hole size and dose rate
We have previously shown that a reduced dose rate increases
the total dose needed to observe bubbling in the specimen (Chen
et al., 2008), prompting us to examine the effect that dose rate
has on beam-induced motion. Furthermore, the importance of
mechanical stability observed with 2D crystals led us to include
in our study specimens with different hole sizes in the support
film. Our measurements suggest that neither dose rate nor hole
size can be adjusted to avoid beam-induced motion entirely. How-
ever, lower dose rates and smaller holes appear to reduce motion
on average while the largest rotations are observed with 1.6 lm
holes. Apart from their larger size, the ice layers in the 1.6 lm holes
also exhibit the greatest variation in thickness, 1000 Å near the
edge and 500 Å in the middle. The thinner ice regions may be more
susceptible to deformation than the thicker regions, allowing lar-
ger deformations in larger holes (Yoshioka et al., 2010).
Measurements made with the smallest hole size of 0.6 lm rep-
resent an exception of the observed trend as the average rotations
are larger than those observed with 1.0 and 1.2 lm in most cases. A
possible reason might be the frequently observed irregular edges of
the 0.6 lm holes (Figure S1) which may contribute to the mechan-
ical instability of the specimen. Furthermore, as the size of the illu-
minated area for all hole sizes was kept approximately constant
(2 lm diameter), there was more carbon within the area. It is
therefore possible that the instabilities of the carbon support film
discussed above and observed previously (Glaeser et al., 2011) con-
tributed to the larger particle rotations inside the 0.6 lm holes.
4.5. Movie frame alignment and averaging can reduce blurring in the
images
There has been a common belief that beam-induced motion is
worst at the beginning of the exposure. The movie in Fig. 3 shows
that this is indeed the case and that the first 10 electrons/Å2cause
about twice as much motion (about 2.5?) as the following 20 elec-
trons/Å2(about 1?). The analysis of other movies yielded essen-
tially identical results in terms of the temporal behavior of
beam-induced motion. If beam-induced motion cannot be avoided,
the high-resolution signal might be boosted somewhat if 5–
10 electrons/Å2dose are accumulated on the specimen before the
camera shutter is opened. However, movies of the particle motions
may make it possible to avoid image blurring entirely. If the direc-
tion, magnitude and timing of the particle translations and rota-
tions during the exposure can be determined, particle images
from individual frames could be averaged with their correct orien-
tations and positions, thus preserving the high-resolution signal.
This will require low-noise image recording, a feature that is
becoming available with the latest generation of direct electron
detectors (McMullan et al., 2009; Milazzo et al., 2011, 2005). A fur-
ther requirement will be sufficient signal in frame averages of the
movies to determine the translations and rotations taking place
during the exposure. For the 70 MDa virus particle used in our
studies, images with a combined dose of 5 electrons/Å2at 200 kV
(10-frame averages) were sufficient. Using the translations deter-
mined in Fig. 3, we performed frame alignment for a small area
of the hole with three virus particles (Fig. 5). Compared to the aver-
age in Fig. 5A calculated without alignment, the average with
alignment in Fig. 5B shows a substantial reduction of blurring.
Blurring due to the particle rotations cannot be removed by simple
2D alignment but could be corrected for by appropriate adjustment
of the Euler angles in a 3D reconstruction from individual movie
frames.
For particles with smaller molecular mass, such as the 70S ribo-
some (2.6 MDa), the signal in the images will be weaker and it
might be necessary to add larger particles, such as the rotavirus
particles used here, to enable tracking of the ice layer. However,
since their radius is also smaller, which reduces the blurring of
peripheral regions compared to the virus particles examined here,
the effect of particle rotation on the final 3D map will be less. For
example, the 70S ribosome has a diameter of about 250 Å, suggest-
ing that an angular accuracy of 0.6? would be sufficient to obtain a
reconstruction at 4 Å resolution. Therefore, although tracking of
the ice deformation would still be necessary to reduce blurring
by the expected beam-induced particle rotations, the required
accuracy is not as high as with the virus particles.
4.6. Other factors may play a role in beam-induced motion
We have investigated here only two factors with a possible
influence on beam-induced motion, namely dose rate and hole size
(within a limited range of sizes from 0.6 to 1.6 lm). Much smaller
holes (with more regular edges and illuminated with a beam with
smaller diameter) and dose rates may be required to achieve a
more significant reduction of beam-induced motion than observed
Fig.5. Translational alignment of movie frames to reduce blurring in images affected by beam-induced motion. (A) Average of 60 frames of an area of Movie S1 that
experienced translations of about 60 Å (see Fig. 3L) Particles are significantly blurred and high-resolution information is lost. (B) Average of the same 60 frames as in panel A
after translational alignment of individual frames. The translations for the individual frames were calculated from the translations determined in Fig. 3 for the 10-frame
averages by linear interpolation. The frame average in panel B exhibits substantially improved contrast with details at higher resolution that were not visible in panel A.
636
A.F. Brilot et al./Journal of Structural Biology 177 (2012) 630–637

Page 8

in our experiments. Other factors that may be important include
carbon thickness, pre-irradiation of the carbon support, size of
the illuminated area, ice thickness, ice uniformity, protein density,
buffer composition, as well as additives that may stabilize the ice
layer or act as sinks for the molecular radicals produced by radiol-
ysis. Although the processing of movies tracking the ice layer
deformation may be one way to reduce the deteriorating effect of
beam-induced motion of the high-resolution data, it would be
more desirable for practical reasons to reduce or eliminate beam-
induced motion altogether. Our results show that large virus parti-
cles such as rotavirus used here provide effective probes to moni-
tor and quantify beam-induced motion. These particles therefore
represent a valuable new tool to investigate beam-induced motion
further and identify procedures that help preserve high-resolution
detail in cryo-EM images.
Acknowledgments
The authors would like to thank Jim Pulokas and John Crum
(National Resource for Automated Molecular Microscopy at the
Scripps Research Institute) for technical support, Douglas Theobald
for advice on the use of Rayleigh distributions, and Chen Xu for
maintaining the Brandeis EM facility in meticulous condition. The
work was supported by National Institutes of Health Grant P01
GM62580 (awarded to NG) and an NSERC Fellowship (awarded
to AB), and RR017573 (BC, CP, AC). Some of the work presented
here was conducted at the National Resource for Automated
Molecular Microscopy which is supported by the National Insti-
tutes of Health though the National Center for Research Resources’
P41 program (RR017573).
Appendix A. Supplementary data
Supplementary data associated with this article can be found, in
the online version, at doi:10.1016/j.jsb.2012.02.003.
References
Baker, L.A., Smith, E.A., Bueler, S.A., Rubinstein, J.L., 2010. The resolution dependence
of optimal exposures in liquid nitrogen temperature electron cryomicroscopy of
catalase crystals. Journal of Structural Biology 169, 431–437.
Booy, F.P., Pawley, J.B., 1993. Cryo-crinkling: what happens to carbon films on
copper grids at low temperature. Ultramicroscopy 48, 273–280.
Bottcher, B., 1995. Electron cryomicroscopy of graphite in amorphous Ice.
Ultramicroscopy 58, 417–424.
Brink, J., Sherman, M.B., Berriman, J., Chiu, W., 1998. Evaluation of charging on
macromolecules in electron cryomicroscopy. Ultramicroscopy 72, 41–52.
Bullough, P., Henderson, R., 1987. Use of spot-scan procedure for recording low-
dose micrographs of beam-sensitive specimens. Ultramicroscopy 21, 223–229.
Cambie, R., Downing, K.H., Typke, D., Glaeser, R.M., Jin, J., 2007. Design of a
microfabricated, two-electrode phase-contrast element suitable for electron
microscopy. Ultramicroscopy 107, 329–339.
Chen, J.Z., Sachse, C., Xu, C., Mielke, T., Spahn, C.M., et al., 2008. A dose-rate effect in
single-particle electron microscopy. Journal of Structural Biology 161, 92–100.
Chen, J.Z., Settembre, E.C., Aoki, S.T., Zhang, X., Bellamy, A.R., et al., 2009. Molecular
interactions in rotavirus assembly and uncoating seen by high-resolution cryo-
EM. Proceedings of the National Academy ofSciences of the United States of
America 106, 10644–10648.
Crowther, R.A., Henderson, R., Smith, J.M., 1996. MRC image processing programs.
Journal of Structural Biology 116, 9–16.
Danev, R., Nagayama, K., 2001. Transmission electron microscopy with Zernike
phase plate. Ultramicroscopy 88, 243–252.
Dietrich, I., Fox, F., Heide, H.G., Knapek, E., Weyl, R., 1978. Radiation damage due to
knock-on processes on carbon foils cooled to liquid helium temperature.
Ultramicroscopy 3, 185–189.
Egerton, R.F., Rauf, I., 1999. Dose-rate dependence of electron-induced mass loss
from organic specimens. Ultramicroscopy 80, 247–254.
Frank, J., Radermacher, M., Penczek, P., Zhu, J., Li, Y., et al., 1996. SPIDER and WEB:
processing and visualization of images in 3D electron microscopy and related
fields. Journal of Structural Biology 116, 190–199.
Ge, P., Zhou, Z.H., 2011. Hydrogen-bonding networks and RNA bases revealed by
cryo electron microscopy suggest a triggering mechanism for calcium switches.
Proceedings of the National Academy ofSciences of the United States of America
108, 9637–9642.
Glaeser, R.M., 1999. Review: electron crystallography: present excitement, a nod to
the past, anticipating the future. Journal of Structural Biology 128, 3–14.
Glaeser,R.M.,2008. Retrospective:radiation
‘‘information limitations’’. Journal of Structural Biology 163, 271–276.
Glaeser, R.M., Taylor, K.A., 1978. Radiation damage relative to transmission electron
microscopy of biological specimens at low temperature: a review. Journal of
Microscopy 112, 127–138.
Glaeser, R.M., Downing, K.H., 2004. Specimen charging on thin films with one
conductinglayer:discussionof
Microanalysis 10, 790–796.
Glaeser, R.M., Hall, R.J., 2011. Reaching the information limit in cryo-EM of
biological macromolecules: experimental aspects. Biophysical Journal 100,
2331–2337.
Glaeser, R.M., McMullan, G., Faruqi, A.R., Henderson, R., 2011. Images of paraffin
monolayer crystals with perfect contrast: minimization of beam-induced
specimen motion. Ultramicroscopy 111, 90–100.
Gonen, T., Sliz, P., Kistler, J., Cheng, Y., Walz, T., 2004. Aquaporin-0 membrane
junctions reveal the structure of a closed water pore. Nature 429, 193–197.
Grigorieff, N., 2007. FREALIGN: high-resolution refinement of single particle
structures. Journal of Structural Biology 157, 117–125.
Grigorieff, N., Harrison, S.C., 2011. Near-atomic resolution reconstructions of
icosahedral viruses from electron cryo-microscopy. Current Opinion in
Structural Biology 21, 265–273.
Henderson, R., 1992. Image contrast in high-resolution electron microscopy of
biological macromolecules: TMV in ice. Ultramicroscopy 46, 1–18.
Henderson, R., 1995. The potential and limitations of neutrons, electrons and X-rays
for atomic resolution microscopy of unstained biological molecules. Quarterly
Reviews of Biophysics 28, 171–193.
Henderson, R., Baldwin, J.M., Downing, K.H., Lepault, J., Zemlin, F., 1986. Structure of
purple membrane from halobacterium halobium: recording, measurement and
evaluation of electron micrographs at 3.5 Å resolution. Ultramicroscopy 19,
147–178.
Henderson, R., Chen, S., Chen, J.Z., Grigorieff, N., Passmore, L.A., et al., 2011. Tilt-pair
analysis of images from a range of different specimens in single-particle
electron cryomicroscopy. Journal of Molecular Biology 413, 1028–1046.
Jensen, G.J., 2001. Alignment error envelopes for single particle analysis. Journal of
Structural Biology 133, 143–155.
Lalitha, S., Mishra, A., 1996. Modified maximum likelihood estimation for Rayleigh
distribution. Commun Stat Theory 25, 389–401.
Majorovits, E., Barton, B., Schultheiss, K., Perez-Willard, F., Gerthsen, D., et al., 2007.
Optimizing phase contrast in transmission electron microscopy with an
electrostatic (Boersch) phase plate. Ultramicroscopy 107, 213–226.
McMullan, G., Chen, S., Henderson, R., Faruqi, A.R., 2009. Detective quantum
efficiency of electron area detectors in electron microscopy. Ultramicroscopy
109, 1126–1143.
Milazzo, A.C., Cheng, A., Moeller, A., Lyumkis, D., Jacovetty, E., et al., 2011. Initial
evaluation of a direct detection device detector for single particle cryo-electron
microscopy. Journal of Structural Biology 176, 404–408.
Milazzo, A.C., Leblanc, P., Duttweiler, F., Jin, L., Bouwer, J.C., et al., 2005. Active pixel
sensor array as a detector for electron microscopy. Ultramicroscopy 104, 152–
159.
Mindell, J.A., Grigorieff, N., 2003. Accurate determination of local defocus and
specimen tilt in electron microscopy. Journal of Structural Biology 142, 334–
347.
Miyazawa, A., Fujiyoshi, Y., Stowell, M., Unwin, N., 1999. Nicotinic acetylcholine
receptor at 4.6 A resolution: transverse tunnels in the channel wall. Journal of
Molecular Biology 288, 765–786.
Muller, H., Jin, J., Danev, R., Spence, J., Padmore, H., et al., 2010. Design of an electron
microscope phase plate using a focused continuous-wave laser. New Journal of
Physics, 12.
Sachse, C., Chen, J.Z., Coureux, P.D., Stroupe, M.E., Fandrich, M., et al., 2007. High-
resolution electron microscopy of helical specimens: a fresh look at tobacco
mosaic virus. Journal of Molecular Biology 371, 812–835.
Settembre, E.C., Chen, J.Z., Dormitzer, P.R., Grigorieff, N., Harrison, S.C., 2011. Atomic
model of an infectious rotavirus particle. The EMBO Journal 30, 408–416.
Street, J.E., Croxson, M.C., Chadderton, W.F., Bellamy, A.R., 1982. Sequence diversity
of human rotavirus strains investigated by northern blot hybridization analysis.
Journal of Virological Methods 43, 369–378.
Tang, G., Peng, L., Baldwin, P.R., Mann, D.S., Jiang, W., et al., 2007. EMAN2: an
extensible image processing suite for electron microscopy. Journal of Structural
Biology 157, 38–46.
Theobald, D.L., Wuttke, D.S., 2006. THESEUS: maximum likelihood superpositioning
and analysis of macromolecular structures. Bioinformatics 22, 2171–2172.
Vonck, J., 2000. Parameters affecting specimen flatness of two-dimensional crystals
for electron crystallography. Ultramicroscopy 85, 123–129.
Wright, E.R., Iancu, C.V., Tivol, W.F., Jensen, G.J., 2006. Observations on the behavior
of vitreous ice at approximately 82 and approximately 12 K. Journal of
Structural Biology 153, 241–252.
Yonekura, K., Maki-Yonekura, S., Namba, K., 2003. Complete atomic model of the
bacterial flagellar filament by electron cryomicroscopy. Nature 424, 643–650.
Yoshioka, C., Carragher, B., Potter, C.S., 2010. Cryomesh: a new substrate for cryo-
electron microscopy. Microscopy and Microanalysis 16, 43–53.
Zhang, X., Settembre, E., Xu, C., Dormitzer, P.R., Bellamy, R., et al., 2008. Near-atomic
resolution using electron cryomicroscopy and single-particle reconstruction.
Proceedings of the National Academy of Sciences of the United States of
America 105, 1867–1872.
damage anditsassociated
physical principles.Microscopy and
A.F. Brilot et al./Journal of Structural Biology 177 (2012) 630–637
637