Sei sulla pagina 1di 44

Progress in Biophysics & Molecular Biology 75 (2001) 121–164

Review

Determining the structure of biological macromolecules by


transmission electron microscopy, single particle analysis and
3D reconstruction
Jonathan Ruprechta,*, Jon Nieldb
a
University of Cambridge, Department of Biochemistry, Hopkins Building, Cambridge CB2 1QW, UK
b
Wolfson Laboratories, Department of Biochemistry, Imperial College of Science, Technology and Medicine,
London SW7 2AY, UK

Abstract

Single particle analysis and 3D reconstruction of molecules imaged by transmission electron microscopy
have provided a wealth of medium to low resolution structures of biological molecules and macromolecular
complexes, such as the ribosome, viruses, molecular chaperones and photosystem II. In this review, the
principles of these techniques are introduced in a non-mathematical way, and single particle analysis is
compared to other methods used for structural studies. In particular, the recent X-ray structures of the
ribosome and of ribosomal subunits allow a critical comparison of single particle analysis and X-ray
crystallography. This has emphasised the rapidity with which single particle analysis can produce medium
resolution structures of complexes that are difficult to crystallise. Once crystals are available, X-ray
crystallography can produce structures at a much higher resolution. The great similarities now seen between
the structures obtained by the two techniques reinforce confidence in the use of single particle analysis and
3D reconstruction, and show that for electron cryo-microscopy structure distortion during sample
preparation and imaging has not been a significant problem. The ability to analyse conformational
flexibility and the ease with which time-resolved studies can be performed are significant advantages for
single particle analysis. Future improvements in single particle analysis and electron microscopy should

Abbreviations: 1D, one-dimensional; 2D, two-dimensional; 3D, three-dimensional; CCD, charge-coupled device;
CCF, cross-correlation function; CTF, contrast transfer function; DPR, differential phase residual; EM, electron
microscopy/electron microscope; FEG, field-emission gun; FSC, Fourier shell correlation; LDL, low-density
lipoprotein; LHCII, light-harvesting complex II; MAD, multiwavelength anomalous diffraction; MCF, mutual
correlation function; MIRAS, multiple isomorphous replacement and anomalous scattering; MSA, multivariate
statistical analysis; NMR, nuclear magnetic resonance; OEC, oxygen-evolving complex; PCTF, phase contrast transfer
function; PSII, photosystem II; SCF, sinogram correlation function; SNR, signal-to-noise ratio; TEM, transmission
electron microscopy.
*Corresponding author. Fax: +44-1223-333345.
E-mail address: jjr24@cam.ac.uk (J. Ruprecht).

0079-6107/01/$ - see front matter # 2001 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 7 9 - 6 1 0 7 ( 0 1 ) 0 0 0 0 4 - 9
122 J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164

increase the attainable resolution. Combining single particle analysis of macromolecular complexes and
electron tomography of subcellular structures with high-resolution X-ray structures may enable us to realise
the ultimate dream of structural biology a complete description of the macromolecular complexes of the cell
in their different functional states. # 2001 Elsevier Science Ltd. All rights reserved.

Keywords: Transmission electron microscopy; Single particle analysis; Image processing; 3D reconstruction;
Crystallography

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

2. Specimen preparation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

3. Electron microscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

4. Optical diffraction and densitometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

5. Principles and methodology of single particle analysis and 3D


reconstruction by angular reconstitution . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
5.1. Correcting for the contrast transfer function (CTF) . . . . . . . . . . . . . . . . . . 127
5.2. Particle picking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
5.3. Band-pass filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
5.4. Reference-free alignment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
5.5. Multivariate statistical analysis (MSA) . . . . . . . . . . . . . . . . . . . . . . . . . 131
5.6. Automatic classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
5.7. Multi-reference alignment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
5.8. Angular reconstitution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
5.9. 3D reconstruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
5.10. Iterative refinements to the model . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
5.11. Evaluating the quality of the 3D reconstruction . . . . . . . . . . . . . . . . . . . . 135
5.12. Presenting the model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136

6. Critical review of single particle analysis and 3D reconstruction by


angular reconstitution: a comparison with other methods for
structural studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136

7. The ribosome: comparing single particle analysis with X-ray


crystallography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145

8. Photosystem II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151

9. Electron tomography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

10. Future prospects and conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164 123

1. Introduction

3D structures of molecules can be obtained through the image processing of electron


microscope (EM) images, using the techniques of single particle analysis, angular reconstitution
and 3D reconstruction. First, transmission electron microscope (TEM) images of the molecule(s)
of interest randomly oriented on an EM grid are obtained. These images are 2D projections of the
3D structure, because the depth of focus is much greater than the specimen thickness (DeRosier
and Klug, 1968). Theoretically, if the projection angle of each image could be determined, a 3D
structure of the molecule could be obtained by projecting the 2D images back along their
projection angles (back-projection). However, electron micrographs of biological molecules are
very noisy and show low intrinsic contrast. Single particle analysis was developed to deal with the
problems of noise and low intrinsic contrast by increasing the signal-to-noise ratio (SNR) of the
micrographs, pushing back the limits of resolution and interpretability.
Even in the last three months, as this review was in preparation, the amount of information
about single particle analysis, and the number of examples of its use, have exploded. There have
been several excellent, but concise, reviews of single particle analysis and electron cryo-
microscopy (e.g. Saibil, 2000a, b; Orlova, 2000). This review will focus in depth on single particle
analysis, informing the reader of some of the principles of this technique. Whilst this has been
done in a non-mathematical way, references are provided to more detailed treatments of some of
the topics. By critically reviewing single particle analysis, it is hoped that the reader will become
aware of some of the advantages and also of the problems of the technique, so that they can assess
the quality of the structures they will see. The review first introduces the steps needed
to produce a 3D structure by single particle analysis, from sample preparation to electron
microscopy and then to image processing, using work that has been carried out on photosystem
II as an example. Single particle analysis is then critically compared to X-ray crystallo-
graphy, electron crystallography and NMR spectroscopy. This is further discussed by considering
structural studies of the ribosome, which have been instrumental in driving many of the
technical advances in single particle analysis, and of photosystem II. Finally, related
developments such as electron tomography are considered, and the future of single particle
analysis is discussed.

2. Specimen preparation

Any specimen preparation technique must avoid the collapse of structures during pre-
paration and observation, since the specimens are viewed in a vacuum in the EM (Slayter and
Slayter, 1992). Also, biological specimens are extremely sensitive to bombardment by electrons,
and this is a significant factor in the high noise levels of electron micrographs (discussed in Amos
et al., 1982; Dubochet et al., 1988). The incident electrons lose large amounts of energy by
inelastic collisions, forming highly reactive ions and free radicals, which disrupt bonds and
fragment molecules. Incident electrons can also directly transfer their momentum to atomic nuclei
in the structure, resulting in atom displacement. Furthermore, the atoms typically present
in biological molecules (C, H, O, N, etc.) scatter electrons weakly, producing images with
low intrinsic contrast. Therefore, to increase the attainable resolution, electron-beam damage
124 J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164

to the specimen should be minimised, and image contrast should be maximised (Chen et al.,
1998).
Negative staining (Amos et al., 1982; Chen et al., 1998; Slayter and Slayter, 1992) is technically
simple to perform, and provides high contrast and a low sensitivity to the electron beam.
However, negative staining has several disadvantages (Kiselev et al., 1990). It only shows surface
detail, with the distribution of the heavy metal atoms rather than the density of the specimen
being revealed. Furthermore, negative staining imposes a limit on the resolution of 10–20 A (
(Amos et al., 1982; Chen et al., 1998; Hoenger and Aebi, 1996). This is because of stain movement
during imaging, variable flattening of the 3D structure by dehydration, and stain granularity (the
size of the stain molecules prevents their penetrating the protein surface, so that details such as
bundles of helices will not be revealed).
Electron cryo-microscopy (also called cryo-electron microscopy, cryo-EM; Dubochet et al.,
1982, 1988; Slayter and Slayter, 1992; Baker and Johnson, 1997; Chen et al., 1998) has
revolutionised the analysis of macromolecular structure by electron microscopy, and is
complementing negative staining for 3D reconstruction studies. Often, single particle analysis
and 3D reconstruction of negatively stained specimens will be attempted to gain a first glimpse of
the structure. Then, images will be taken under cryo-conditions, which in principle contain
information to atomic resolution (Chen et al., 1998), and processed to produce a 3D model at
higher resolution. For a detailed comparison of the relative advantages of negative stain and cryo-
EM techniques, see Hoenger and Aebi (1996). In cryo-EM, samples are embedded in vitreous ice
on a ‘‘holey’’ carbon grid and maintained at low temperatures (100–113 K) whilst under the
electron beam. Vitreous ice is essentially a supercooled liquid, produced when water is very
rapidly cooled below 273 K (at about 105 K s1) (Slayter and Slayter, 1992). This avoids damage
to the specimen by ice crystal formation. Vitreous ice forms a structureless medium in which the
molecules are hydrated, despite the requirement for vacuum conditions (the low temperatures lead
to a very slow rate of sublimation of the ice). The specimen is therefore imaged under conditions
more like those in its native environment. The low temperature, and the use of low doses of
electrons (10–20 electrons/A( 2), as described in the next section, minimise beam damage (Stark
et al., 1996).
The procedure for cryo-EM specimen preparation is as follows. The specimen is applied to a
‘‘holey’’ carbon grid, which may have been glow-discharged to improve the hydrophilicity of the
carbon film and therefore increase the amount of specimen adhering to the grid. The grid is then
blotted with filter paper, and plunged into a bath of liquid ethane held at liquid nitrogen
temperature. Liquid ethane rather than liquid nitrogen is used as the cryogen because liquid
ethane has a higher heat capacity. Furthermore, liquid nitrogen will form a gas layer on contact
with the EM grid, which will insulate the sample and slow cooling. The blotting time is an
important factor in controlling the thickness of the ice. If the ice layer is too thick, the electron
beam may not be able to penetrate it.
The frozen-hydrated specimens have to be transferred to the microscope using special
equipment designed to keep them at low temperature under liquid nitrogen (cryo-transfer). This is
essential to avoid specimen warming (even at quite low temperatures, sublimation of ice may be
appreciable, causing variable dehydration of the specimen) and to prevent contamination by
water vapour condensing onto the specimen.
J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164 125

3. Electron microscopy

There are several requirements to produce the high-quality images necessary for high-resolution
work (Slayter and Slayter, 1992). The specimen stage must have a high degree of mechanical
stability in order to reduce specimen drift whilst an image is recorded. Vibrations and stray
magnetic fields can cause a strongly directional loss of resolution. The stage design must also
minimise heating and the accumulation of electric charge. For cryo-EM work a cold stage (cooled
by liquid nitrogen or liquid helium) is necessary (Dubochet et al., 1988; Fujiyoshi et al., 1991;
Fujiyoshi, 1998). This should provide evenness of cooling and mechanical stability. Anti-
contamination devices are required to prevent deposition of materials on the specimen. These are
usually metal surfaces, such as a pair of copper blades, situated near the position of the EM grid in
the microscope, and cooled to liquid nitrogen temperature so that they trap residual gases
(Dubochet et al., 1988; Baker and Johnson, 1997).
Samples may be imaged under conditions in which the electron dose is minimised, thus reducing
beam damage. In this ‘‘low-dose’’ technique, three different ‘‘modes’’ are used: search, focus and
exposure. First, the pin in the microscope is used to correlate the areas imaged in the exposure and
search modes. Search mode is set at a low magnification (e.g. 2650  ), and is used to position the
pin such that a good area of the grid will be imaged in exposure mode. For cryo-EM, a good area
of the grid would be one containing a hole in the carbon film}the sample will hopefully be
embedded in the meniscus of ice spanning the hole. Using low magnification at this stage reduces
beam damage since the irradiation level is kept very low (less than 0.05 electrons/A ( 2/s (Baker and
Johnson, 1997)). Focus mode is then used to focus, at high magnification (e.g. 175 000  ), on an
area of the carbon film near the ice hole. The position of the beam in focus mode is set at an
adjustable distance and angle from its position in exposure mode, such that in focus mode beam
damage is again minimised. Phase contrast is now introduced by defocusing the microscope (e.g.
1–2 mm underfocus, see Section 5.1 for a discussion of phase and amplitude contrast). Finally, in
exposure mode (e.g. at 40 000  ), the image is recorded of the desired area. The magnification at
which the images are taken, the accelerating voltage, the total electron dose, and the defocus can
all be varied depending on the desired resolution of the image (Baker and Johnson, 1997). It can
be seen that in low-dose regimes, the final image is taken ‘‘blind’’, and the quality can only
be assessed after the image has been taken and analysed by optical diffraction (Section 4) or by
on-line image processing which is available with the latest electron microscopes.
Even the most advanced electron lenses are of poor quality in terms of spherical aberration,
astigmatism, lens-current fluctuations and chromatic aberration. Spherical aberration is a
geometrical property of lenses, resulting in rays that travel to the lens margins being brought to a
slightly different focus point compared to those travelling very close to the lens axis, resulting in a
‘‘zone of confusion’’ (Slayter and Slayter, 1992; Reimer, 1997). Astigmatism occurs when the
objective lens field is not perfectly symmetrical, resulting in the formation of line images of point
objects, oriented uniformly across the image (Amos et al., 1982; Slayter and Slayter, 1992; Reimer,
1997). Astigmatism can vary while a sample is being imaged, due to the deposition of
contaminants onto the lens elements. Astigmatism can be corrected by looking at the granularity
of the carbon film at high magnification (e.g. 175 000  ), which should remain rotationally
symmetrical while going in and out of focus (Amos et al., 1982). This can be achieved by adjusting
the field applied by electromagnetic lenses called stigmators. It is essential to compensate for
126 J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164

astigmatism before taking high-resolution TEM images. Chromatic aberration (Amos et al., 1982;
Slayter and Slayter, 1992; Reimer, 1997) is a consequence of the dependence of the focal length of
lenses on the wavelength of incident radiation. In electron microscopy, it results from variations in
electron velocity. Electron velocity is inversely proportional to electron wavelength, as described
by the de Broglie equation. Variations in electron velocity may result from fluctuations in the high
tension supply, differences in the energy of electrons emitted from conventional thermionic
emission guns, and energy losses as a result of inelastic scattering events. The electromagnetic
lenses can no longer bring the electrons of ‘‘incorrect’’ velocity to the correct focus, and this
results in a smearing out of the contrast transfer function (CTF}see Section 5.1) at high
resolutions.
The development of the field-emission gun (FEG) has revolutionised cryo-EM work, and its use
should soon become standard for higher resolution work (Stowell et al., 1998). In FEG
microscopes, very high electric fields are used to release electrons from a very fine tungsten tip
(Kasper, 1982; Reimer, 1997). The increased spatial coherence (due to the higher brightness and
smaller effective size of the electron source) and increased temporal coherence (due to the smaller
energy spread) of this electron source gives great improvements in contrast transfer (see Section
5.1) at high resolution compared to a conventional thermionic emission gun, especially when the
image is strongly defocused to introduce phase contrast during cryo-EM work (Walz and
Grigorieff, 1998). Unfortunately, correcting for astigmatism is difficult with a FEG (Williams and
Barry Carter, 1996). However, developments such as on-line image processing, allowing
observation of the power spectrum (the squared amplitude of the Fourier transform) at the
electron microscope during imaging, are simplifying this (Zemlin et al., 1996).

4. Optical diffraction and densitometry

Optical diffraction is often used to select the images with minimal astigmatism and drift for
subsequent densitometry and image processing (Reimer, 1997). The amorphous carbon film of the
EM grid produces a nearly white spatial frequency spectrum (Reimer, 1997). However, electron
microscopes do not transmit all spatial frequencies equally well, an effect described by the contrast
transfer function (CTF}Section 5.1). The missing frequencies appear as a series of rings in the
Fourier transform or the optical diffraction pattern. These rings are called Thon rings (Thon,
1966). In an image with low astigmatism, the Thon rings show rotational symmetry. In an
astigmatic image, the Thon rings appear stretched in one direction. Blurring of the Thon rings
indicates beam drift during exposure. The spacing of the Thon rings supplies information about
the defocus value when the micrograph was taken, and provides valuable information about the
contrast transfer function (Section 5.1).
Densitometry converts the information in the TEM images into a form that is suitable for
computer analysis. Densitometry is a compromise between quality and time. Whilst it is essential
that the images are digitised accurately, without degrading the image resolution or adding noise, it
is also important that a micrograph can be scanned in a reasonable time (Mitsuoka et al., 1997).
Flat-bed point-illuminating densitometers, such as the Joyce Loebl Mark 4 and the Perkin-Elmer
DS, literally digitise an image point-by-point, which can take hours for a single micrograph. Line-
illuminating densitometers, such as the LeafScan 45 and the Zeiss-SCAI, can scan a line at a time,
J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164 127

and are therefore considerably faster (Mitsuoka et al., 1997). The images are digitised usually with
16-bit grey scale resolution (65,536 possible greys). Shannon’s Theorem specifies the conditions
required for the faithful digital representation of an image (Shannon, 1949; discussed in: Chen
et al., 1998; van Lint, 1982; Mellema, 1980; Radermacher, 1988). In this context, the theorem
shows that you must scan at least at half the desired molecular resolution at the specimen level. As
an example, we might hope to obtain a resolution of 5 A ( in the scanned image. By Shannon’s
(
Theorem, the minimum resolvable distance (5 A}the Nyquist frequency) will be twice the pixel
resolution. Hence, we need to achieve a resolution of 2.5 A( per pixel on the specimen level. If the
images are taken at a magnification of 40 000  , the densitometer step size would have to be
10 mm. However, images are often digitised to a spatial resolution corresponding to about one-
quarter of the desired molecular resolution at the specimen level, this oversampling being essential
to reduce interpolation errors (Chen et al., 1998).

5. Principles and methodology of single particle analysis and 3D reconstruction by angular


reconstitution

Despite attempts made to boost contrast and reduce beam damage during specimen
preparation and imaging, further image processing is essential to obtain meaningful information
from the images, particularly when low-dose techniques are used which result in a low SNR in the
images. A wide range of image processing software is now available (reviewed by Carragher and
Smith, 1996). In this section, the principles of image processing, single particle analysis and 3D
reconstruction will be discussed as implemented in the IMAGIC-5 image processing program (van
Heel et al., 1992b, 1996; Schatz et al., 1995; Orlova, 2000). Fig. 1 summarises the stages of the
procedure. Figs. 3, 6 and 7 illustrate key aspects of image processing and are reproduced from an
investigation into the photosystem II (PSII) supercomplex from higher plants (Nield et al., 2000c).

5.1. Correcting for the contrast transfer function (CTF)

There are two components to contrast in an image: amplitude and phase (discussed in Amos
et al., 1982; Erickson and Klug, 1971; Slayter and Slayter, 1992). Phase contrast is produced by
interference between the elastically scattered waves that pass though the objective aperture of the
microscope and the unscattered, transmitted wave. Amplitude contrast is produced when
electrons are lost by inelastic scattering, or by high angle elastic scattering, which falls outside the
objective aperture. The contrast transfer function (CTF) describes the fidelity with which different
spatial frequencies are transmitted by the electron lenses (Erickson and Klug, 1971; Wade, 1992;
Williams and Barry Carter, 1996; Reimer, 1997). High spatial frequencies represent fine detail
(high-resolution information), and low spatial frequencies represent coarse detail. The CTF
accounts for the effects of spherical aberration and defocusing. The CTF has amplitude and phase
contrast components, which are additive (Amos et al., 1982; Erickson and Klug, 1971).
In cryo-EM work, the phase component of the CTF (the PCTF) predominates because the
objective lens is defocused to introduce phase contrast, and amplitude contrast is low since the
samples are thin and biological specimens contain lighter atoms, which scatter electrons weakly.
The PCTF limits the resolution of the images in the micrographs. Beyond the first zero crossing of
128 J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164

Fig. 1. Single particle analysis, image processing and 3D reconstruction}a flow chart summarising the key steps.
J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164 129

Fig. 2. The phase contrast transfer function (PCTF), and the limits of resolution. The continuous line shows the
variation of the PCTF as a function of a=l (the spatial frequency), where l is the electron wavelength and a is the angle
of scattering in the microscope. The PCTF is given by sinwðaÞ where wðaÞ ¼ ð2p=lÞð14 Cs a4 þ 12 D f 2 Þ. Cs is the
coefficient of spherical aberration and Df is the defocusing. In this example, l ¼ 0:0042 nm, Cs ¼ 1:3 mm and
Df ¼ 300 nm. The dashed line shows the effect of an envelope function.

the PCTF and until the second zero crossing, contrast is reversed (see Fig. 2); adding these
intensities to those that lie before the first zero crossing produces artefacts. Spatial frequencies
corresponding to a PCTF of zero do not appear in the final image. For these reasons, the limit of
resolution of an image (without PCTF correction) is obtained when the first zero crossing of the
PCTF is reached. By correcting for the PCTF, compensating for the contrast reversals, it is
possible to increase the resolution of the images and thus to increase the resolution of the final 3D
model. Correction for the PCTF is achieved as follows. Optical diffraction or the fast fourier
transform (FFT) or the power spectrum of an EM image reveals the Thon rings, which determine
where the maxima and minima of the CTF lie (Amos et al., 1982). Areas where the CTF is positive
or negative appear as bright bands, with the zero crossings appearing as black rings. The PCTF is
given by a simple mathematical equation (given in the legend to Fig. 2) that can be fitted to the
Thon rings. Correction for the PCTF can be carried out in Fourier space by multiplying the data
with the inverse of the PCTF (Erickson and Klug, 1971). The inverse function may be modified to
prevent amplifying noise in the area of the zero crossings (Orlova et al., 1997). PCTF correction
can either be applied to the 2D images, or to the 3D model in 3D Fourier space (e.g. Orlova et al.,
1997; Stark et al., 1997a) provided the micrographs selected for processing are taken at the same,
or similar, defocus and are therefore affected by the same PCTF}the uncorrected 3D model will
then have been modified by a 3D PCTF. Despite PCTF correction, it is impossible to extend the
130 J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164

resolution beyond a point known as the information limit (Slayter and Slayter, 1992). This is
because the PCTF is damped by an envelope function that represents limiting factors such as
chromatic aberration and poor coherence of the electron beam. In the final 3D model, loss of
information at spatial frequencies coinciding with the zeros of the PCTF can be avoided by
combining images taken at different levels of defocus (Slayter and Slayter, 1992; Saibil, 2000b).
Like the phases, image amplitudes also show CTF modulations. It is possible to correct for the
amplitude as well as the phase component of the CTF, and this may be especially important for
low spatial frequencies (Toyoshima and Unwin, 1988; Toyoshima et al., 1993). Details of one
approach for correcting both amplitude and phase components of the CTF may be found in a
paper by Mancini and Fuller (2000).
If the desired resolution of the final model lies within the first zero crossing of the CTF it may
be that no explicit CTF correction need be applied. A low-pass filter can be used to exclude
information beyond the first zero crossing, preventing the creation of artefacts. In a
reconstruction of the 70S ribosome at a resolution within the first zero crossing of the CTF,
CTF correction was thought to enhance the low spatial frequencies while making smaller details
less clear, and figures in the paper are shown without CTF correction (Stark et al., 1997a).
However, Zhu and colleagues have criticised this approach. They believe that CTF correction of
the amplitudes of low-frequency components is important, and they have developed the necessary
computer programs to do this (Zhu et al., 1997).
An additional consideration is that the envelope function decreases the amplitudes of high
spatial frequencies. Efforts are now being made to correct for this, for example by using low-
resolution X-ray data to calculate an inverse temperature factor to boost the high spatial
frequencies (Mancini and Fuller, 2000; also see Section 7). Similar techniques are used in electron
crystallography (e.g. Unger and Schertler, 1995).

5.2. Particle picking

A data set of at least several thousand particle images should be obtained from the micrographs
by picking all discernible particles that are not overlapping or in close contact with other particles
(Harauz et al., 1988). The single particle images are cut into x  x pixel boxes. The size of the pixel
boxes is selected to just enclose the single particle images, removing as much background as
possible. Particle picking can be done directly in IMAGIC (van Heel et al., 1996), or by using the
program XIMDISP (Crowther et al., 1996; Smith, 1999) and then importing the stack of single
particle images into IMAGIC.

5.3. Band-pass filtering

Band-pass filters are applied to suppress the highest and lowest spatial frequencies of the
images. Very low spatial frequencies can represent gradual fluctuations in the average densities
arising from the amount and uniformity of staining; very high spatial frequencies include noise
(Harauz et al., 1988). Band-pass filtering influences the success of the subsequent alignment steps,
which involve the use of correlation functions that can overweigh certain spatial frequencies (van
Heel et al., 1992a). By using Gaussian-based filters, the creation of artefacts at the high and low
cut-off points can be avoided. The spatial frequencies can be restored after an initial 3D model has
J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164 131

been reconstructed. This is especially important for the high spatial frequencies, which include the
desired high-resolution information.
The particles are now surrounded by a circular mask, removing unnecessary background. The
images are normalised to zero average density and an arbitrary standard deviation (van Heel et al.,
1992b).

5.4. Reference-free alignment

Reference-free alignment (Dube et al., 1993) is the first alignment step in the image processing,
and is used to centre the particles. The single particle data set is compared to a rotationally
averaged total sum of the band-pass filtered particles, resulting in a translational alignment that
centres the images in their x  x pixel boxes. Alignment is achieved by using cross-correlation
functions (CCFs) (Frank, 1980; Reimer, 1997). The basis of this approach is that the two images
to be aligned are translationally shifted relative to each other by known vectors, and in each
position, the product of equivalent pixels in the overlapping area are averaged over the area of
overlap. This product is the CCF at the position of the shift vector (Frank, 1980; Frank et al.,
1988). The CCF shows a peak at the place where a motif present in both images overlaps. Good
alignment is therefore achieved when the CCF reaches a maximum. By searching for peaks in the
CCF and shifting molecules by the appropriate vector, translational alignment of the two images
can be achieved. The CCF is actually evaluated in Fourier space, taking advantage of the
convolution theorem (which states that the Fourier transform of the product of two functions is
the product of their individual transforms, this latter being easier to calculate). The rotationally
averaged total sum and the image to be aligned are Fourier transformed, then these transforms
are complex conjugate multiplied together, and the desired CCF is produced by inverse Fourier
transformation (van Heel, 1992b).
The reference-free alignment step is iterated several times until a good alignment is achieved
(the distance, in pixels, in which each image is shifted becomes very small). Initially performing a
reference-free alignment prevents the data set being biased towards a particular reference at an
early stage in the analysis, and uses an averaged image with a higher SNR than a raw image (Dube
et al., 1993).

5.5. Multivariate statistical analysis (MSA)

The multivariate statistical analysis (MSA) technique of correspondence analysis is next used to
identify the principal components of variation in the set of images, as a preliminary to placing the
images into groups of similar molecular images in similar rotational orientations (Frank and van
Heel, 1982; Frank, 1984, 1990; Lebart et al., 1984; van Heel and Frank, 1981).
Each image of x  x pixels can be considered as a point in (x  x)-dimensional space, where
each axis represents the density value at a single pixel. The entire set of images forms a cloud
within this space. The distances between points in this space may be interpreted in terms of the
similarity of the corresponding images. The cloud may be structured into subclouds,
corresponding to different subsets of images (classes). The set of points in (x  x)-dimensional
space are arranged into a large input matrix. Distances between points in the (x  x)-dimensional
space are measured using metrics such as the w2 metric. The w2 distances between any two rows or
132 J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164

columns of the input matrix are calculated, producing a symmetric matrix (van Heel and Frank,
1981). Next, the eigenvalues and eigenvectors (also called eigenimages or factors) of this
symmetric matrix are determined (van Heel and Frank, 1981). The w2 metric cannot deal with
images that have a zero average density, as may be the case with phase contrast EM images which
can be treated as a modulation relative to a constant background (Borland and van Heel, 1990).
For this reason, the modulation metric was developed (Borland and van Heel, 1990). IMAGIC
allows the user to define which metric is used (van Heel et al., 1996). To determine the major
components of variation, a new rotated co-ordinate system is described in which the first axis is
the eigenvector representing the greatest interimage variance (associated with the largest
eigenvalue), the second axis is the eigenvector representing the largest remaining interimage
variance, and so on (van Heel et al., 1992b). By only considering the significant eigenvectors
(typically the first 5–50 (Sherman et al., 1998)), the images can be considered as points in a much
smaller than (xx)-dimensional space. This reduces the total amount of data, vastly speeding up
the image processing, whilst also reducing the effects of noise. Since the eigenvectors show the
variation in the dataset, they can provide useful information about conformational flexibility or
substrate binding, as will be discussed later in the context of the ribosome.
A map plotting different combinations of the significant eigenvectors against each other can
reveal classes of images that have certain features in common. The program WEB can be used to
produce these maps, and it is possible to create an average of images that fall within a selected
area (Frank et al., 1996).

5.6. Automatic classification

The hierarchical ascendant classification algorithm is used to generate a specified number of


.
classes (van Heel and Stoffler-Meilicke, 1985). This algorithm begins with as many classes as there
are images. It then merges the classes that are closest together. At each step, two classes are
merged if the resulting increase in total intra-class variance is minimal. The images in each of these
classes are then averaged to produce ‘‘characteristic views’’ (known as class averages or classums).
In a set of similar aligned images, the noise at any position varies from image to image, but the
information from the molecules is the same. The averaging procedure boosts the common signal
by a factor of n1=2 , where n is the number of images, whilst repressing random noise (Chen et al.,
1998). The class averages therefore have a higher SNR than the raw images, and are much more
easily interpreted.

5.7. Multi-reference alignment

Having produced the first class averages, some are selected and used as references to search
through and align the entire data set, generating improved class averages. In this multi-reference
alignment (Schatz et al., 1995), each image is translationally and now also rotationally aligned, in
turn, to selected class averages, using cross-correlation functions. MSA and automatic
classification are then used to generate new class averages. After a few iterations, good class
averages with greatly improved SNRs can be achieved (see Fig. 3a). The progress of the iterations
can be followed visually by inspecting the class averages, and to a certain extent by looking at
J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164 133

Fig. 3. Single particle analysis of the plant photosystem II supercomplex. (a) A selection of typical class averages used
for the 3D reconstruction. (b) Reprojections of the 3D map in identical orientations with the corresponding class
averages. (c) Surface rendered views of the final 3D map in the same orientation as the class averages. Reprinted with
permission from Nature Structural Biology (Nield et al., 2000c, # 2000, Macmillan Magazines Limited).

changes in the total sum of the aligned images after each iteration. After the first multi-reference
alignment, the total sum of the images should appear less rotationally symmetric.

5.8. Angular reconstitution

Having generated a set of good class averages, it is desirable to construct a 3D model from
them. The first step towards this is to determine projection directions (Euler angles) for the class
averages. This is done using the angular reconstitution technique (van Heel, 1987a; Serysheva
et al., 1995), which is based on the common line projection theorem (van Heel, 1987a). This states
that two 2D projections of the same 3D object have at least one 1D line projection in common.
The common line projection is equivalent to the common tilt axis of the two projections, which is
perpendicular to the projection directions of both input class averages.
To find the common line projection(s) between two 2D class averages, the sinogram of each
class average is calculated. A sinogram is a collection of the 1D line projections of a 2D class
average. The first line projection is obtained by summing all the horizontal lines in the 2D class
average (Serysheva et al., 1995). The second line projection is usually in a direction 18 away from
the first. The sinograms are compared pairwise using sinogram correlation functions (SCFs),
which have maxima at the position corresponding to a pair of shared line projections (Fig. 4). A
least-squares fitting procedure is used to identify the position of the peaks in the SCF, giving the
orientation of the common tilt axis between the two class averages. From the common tilt axis,
the Euler angles of the class averages can be determined (for a mathematical description, see van
134 J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164

Fig. 4. Determining Euler angles using sinograms and sinogram correlation functions (SCF). Left, a self-sinogram
correlation function. The class average is shown in a and again in b. The sinograms of the class average are shown in c
and d. Finally, e shows the sinogram correlation function, with symmetry-related peaks being colour-coded. Right, a
cross sinogram correlation function between two different class averages shown in a and b. The sinograms are in c and
d, and the cross sinogram correlation function in e. The four colour-coded dots indicate the optimal Euler angle
orientations of these two class averages. Reprinted with permission from Nature Structural Biology (Serysheva et al.,
1995, # 1995, Macmillan Magazines Limited).

Heel, 1987a). When the best Euler angles have been assigned to an input class average, the peaks
in the SCF have their highest value, which should be the same for each peak (Serysheva et al.,
1995). The standard deviation of the peak heights can thus be used to exclude poor class averages,
and can provide useful information about the progress of the reconstitution (see Section 6).

5.9. 3D reconstruction

An initial 3D reconstruction is gained by back-projecting the class averages along their assigned
Euler angles, using the exact-filter back-projection algorithm (Harauz and van Heel, 1986;
Radermacher, 1988; Schatz et al., 1995). This algorithm takes into account the heterogeneous
distribution of projection directions that are typically encountered in the class averages, by
downweighting over-populated class averages (Orlova, 2000).
The 3D reconstruction is then reprojected along the Euler angle directions assigned to the class
averages. These resulting reprojections illustrate how well the class averages fit to the 3D model
(compare the reprojections shown in Fig. 3b with the class averages in Fig. 3a). Poor class
averages can be identified, and then removed from the dataset. This identification can either be
done visually or by using a list of the errors between the class averages and their corresponding
reprojections that is generated by IMAGIC. Many class averages will have been discarded by the
J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164 135

end of this procedure. For example, from a pool of 400 initial class averages maybe 200 will be
kept. This is only an approximate figure, and it will depend on the size and quality of the original
data set. The remaining class averages can be back-projected to produce a more reliable 3D
model.

5.10. Iterative refinements to the model

The 3D model is refined as follows (Schatz et al., 1995; Serysheva et al., 1995). The reprojected
images from the latest model are used as references for a multi-reference alignment of the entire
data set of raw, band-pass filtered images. Following MSA and automatic classification, rare
molecular orientations that were missed can now be recognised and grouped into statistically
significant classes.
In the subsequent angular reconstitution, instead of comparing each new input class average
with every other one, they are now compared to the reprojections from the previous model. These
reprojections are known as the ‘‘anchor set’’, and each has assigned Euler angles. This is an
improvement because the reprojections contain less noise than the original input class averages
due to the averaging that occurs when the 2D class averages are merged to produce the 3D
reconstruction (compare Figs. 3b and a). Unlike the input class averages, they also share a
common 3D origin and a common rotational orientation in the plane of the EM grid (Schatz et al.,
1995; Serysheva et al., 1995).
The procedure of alignment, MSA, classification and angular reconstitution is iteratively
applied to refine the 3D results. Poor class averages are discarded before the next round of
iteration. In the later stages of image processing, the CCF may be replaced by the mutual
correlation function (MCF) (van Heel et al., 1992a). The multiplication in Fourier space during
the calculation of the CCF makes strong spatial frequencies (usually low-resolution information)
overwhelmingly stronger than the weak spatial frequencies. The MCF has been developed to
better weigh fine, high-resolution, details in the images (van Heel et al., 1992a). In the MCF, the
Fourier components are divided by the square roots of their amplitudes.
After 3D reconstruction and further iterative refinements, the 3D model shows no further
improvement (no further reduction in the error measures is seen).

5.11. Evaluating the quality of the 3D reconstruction

The resolution can be determined by measuring the Fourier shell correlation (FSC) between
two independent 3D reconstructions (van Heel and Harauz, 1986; Orlova et al., 1997), each based
on half of the available class averages (Fig. 5). FSC measures the normalised cross-correlation
coefficient between two 3D volumes as a function of spatial frequency. The resolution can be
determined as the reciprocal of the spatial frequency at the intersection of the FSC function with a
function that represents three times the standard deviation of random noise (3s) (the 3s threshold
criterion) corrected for the molecular point-group symmetry. Alternatively, resolution can be
determined as the reciprocal of the spatial frequency when the correlation coefficient equals 0.5
(the FSC=0.5 criterion).
136 J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164

Fig. 5. Determining the resolution using the Fourier shell correlation (FSC) technique.

5.12. Presenting the model

Initial 3D surface-rendering can be performed in IMAGIC, with thresholds selected to produce


expected molecular weights (Fig. 3c). The model can be presented as a series of cross-sections,
revealing internal detail, as in Fig. 6. These cross-sections can be printed out, glued onto
Styrofoam sheets and cut at a particular contour level to produce a physical model. The AVS/
Express software package (Sheehan et al., 1996) can be used to produce surface-rendered images,
as shown in Fig. 7. AVS/Express allows easier manipulation of the model than IMAGIC, such as
rotations and magnifications, and the use of colour.

6. Critical review of single particle analysis and 3D reconstruction by angular reconstitution:


a comparison with other methods for structural studies

In this section, single particle analysis will be compared to X-ray crystallography, electron
crystallography and NMR spectroscopy (for reviews of the principles behind electron crystal-
lography see Amos et al., 1982; Slayter and Slayter, 1992; Walz and Grigorieff, 1998; Glaeser,
1999). 3D reconstruction using the angular reconstitution technique that is implemented in
IMAGIC will be compared to the related random-conical tilt method.
J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164 137

Fig. 6. 10 Å thick sections through the 3D map of the plant photosystem II supercomplex at 24 Å. (a) Surface
representation of the 3D map indicating the 10 Å thick sections taken. (b) Projection map of the transmembrane region
towards the stromal surface of the chloroplast thylakoid membrane. (c) Projection map of the region close to the
lumenal surface. (d) Projection map of the region occupied by the extrinsic proteins of the photosystem II oxygen-
evolving complex. Reprinted with permission from Nature Structural Biology (Nield et al., 2000c, # 2000, Macmillan
Magazines Limited).
138 J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164

Fig. 7. Surface-rendered view of the 3D map of the plant photosystem II supercomplex at 24 Å, visualised using AVS
Express. The extrinsic oxygen-evolving complex proteins are labelled, A/A0 (the 33 kDa protein) and B/B0 (23 and
17 kDa proteins). The putative membrane-spanning region is shown. Overall dimensions are also indicated. Reprinted
with permission from Nature Structural Biology (Nield et al., 2000c, # 2000, Macmillan Magazines Limited).

A significant advantage of structural studies by single particle analysis is that there is no need to
grow crystals. Crystallisation has traditionally limited the speed with which a structure can be
determined by X-ray or electron crystallography. The 2D or 3D crystallisation of proteins can be
problematical, given the many variables that can influence crystal formation, and the requirement
for good-quality crystals that diffract to a high resolution (Rhodes, 2000). For electron
crystallography, it is also necessary to achieve uniform flatness across the 2D crystal, otherwise
there is a loss of resolution perpendicular to the crystal plane (Stowell et al., 1998). Despite these
difficulties, X-ray crystallographers are seeking new methods to crystallise increasingly complex
proteins (reviewed in Ostermeier and Michel, 1997), including the use of lipidic cubic phases to
produce 3D crystals of membrane proteins (Rummel et al., 1998). Several new methods have been
developed to grow 2D crystals of soluble and membrane proteins for electron crystallography,
.
exploiting affinity tags on overexpressed molecules (reviewed in Kuhlbrandt and Williams, 1999)
and using streptavidin as an adaptor molecule for crystallisation on a biotinylated lipid monolayer
(see, for example, Darst et al., 1991; Avila-Sakar and Chiu, 1996; Levy et al., 1999; reviewed in
Kornberg and Darst, 1991). Furthermore, image processing programs used in single particle
analysis, such as IMAGIC-5, can now be used to correct for imperfections and short-range
disorder in 2D crystals (Sherman et al., 1998; Walz and Grigorieff, 1998).
Sample preparation for single particle analysis is usually easier and quicker than crystallisation.
The ease of sample preparation, and the fact that for cryo-EM the vitrification process is not very
sensitive to the buffer components, makes it possible to make rapid progress in studying different
functional states of molecules (e.g. bound to different substrates or cofactors). Cryo-EM and
single particle analysis has revealed conformational changes in the GroEL-GroES ATPase cycle
(Chen et al., 1994; Roseman et al., 1996). However, the angular reconstitution technique requires
a random orientation of particles on the EM grid in order to uniformly cover the asymmetric
J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164 139

triangle of the molecular point group symmetry (Orlova et al., 1997), producing a 3D structure
with an isotropic resolution (Fig. 8). An overabundance of one type of view can lead to a 3D
reconstruction artefact, as shown by Boisset in reconstructions using a ‘‘human head phantom’’ as
a test volume (Boisset et al., 1998). The orientations taken up by proteins on an EM grid can be
affected by molecular properties such as shape and hydrophobicity, and grid preparation such as
glow discharging. As yet, there is no reliable way to ensure a random orientation of particles
(Chen et al., 1998). If preferred orientations are observed, a broader range of orientations may be
observed by tilting the specimen by a small amount (10–308) in the EM (van Heel, 1992b; Baker
and Johnson, 1997) or alternatively the random-conical tilt method may be used (described
below).
Another advantage of single particle analysis is the ability to deal with heterogeneous
populations of molecules. MSA and automatic classification allow a heterogeneous population to
be sorted into classes of similar molecules, generating more homogeneous sub-populations that
can be analysed individually. This approach, which has been called ‘‘computational purification’’,
has been exemplified in a recent study of low-density lipoproteins (Orlova et al., 1999). Single
particle images of low-density lipoproteins (LDL) were grouped into 60 classes using MSA and
automatic classification. The class averages showed striking variations in the number of striations
among similarly oriented particles (Fig. 9a). The classes with three striations were prominent, and

Fig. 8. The Euler angle distribution for the class averages used in a reconstruction of the Haliotis tuberculata
hemocyanin type 1 didecamer. This molecule has D5 point group symmetry, and the asymmetric triangle is indicated in
the figure. The red dots show the projection directions. The uniform coverage of the asymmetric triangle is important to
reduce artefacts. Reprinted with permission from the Journal of Molecular Biology (Meissner et al., 2000, # 2000,
Academic Press Limited).
140 J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164

were used to reconstruct a 3D model (Fig. 9b and c). This indicated that the striations resulted
from internal structure in the inner core of the LDL particle. The authors hope to produce 3D
models of the other classes once more data has been obtained. It is, however, essential to ensure
that particles with the same structure are used in the generation of the 3D model, so that the
model is accurate and meaningful. In the study by Orlova and colleagues, the angular
reconstitution technique helped verify that particles with the same structure were used in the
image processing. If this had not been the case, the standard deviation of peaks found in the
search of the Euler angles would have increased as the class averages were added stepwise into
the angular reconstitution. Furthermore, the 3D map would have become blurred, and the
reprojections would have become different to the input class averages and raw images (Orlova
et al., 1999). Computational purification has also been used to reveal different conformations of
an archael chaperonin (Schoehn et al., 2000) and in studies of photosystem II (Section 8). It can be
anticipated that its use will increase in the future. Heterogeneous populations of molecules are
very difficult to study by X-ray or electron crystallography for two main reasons: it is likely to be
extremely difficult to crystallise them, and even if crystals can be produced, the different molecules
will diffract differently, making analysis complicated. Single particle analysis, MSA and automatic
classification are thus at an advantage when studying protein complexes that are unstable and
produce a heterogeneous population of molecules differing in subunit composition, e.g. PSII
(Section 8), or in studying conformational flexibility in proteins (reviewed by Saibil, 2000a; and
see Section 7).
The techniques of cryo-EM have encouraged the use of time-resolved EM studies. Transient
states can be observed by initiating a reaction on the EM grid by flash photolysis of caged
reactants followed by vitrification after a fixed time interval (Chen et al., 1998). Alternatively, the
cryo-EM plunger (Berriman and Unwin, 1994; Chen et al., 1998) can be used. This device can

Fig. 9. Computational purification used in single particle analysis of low-density lipoproteins (LDL). (a) Some class
averages after the first round of alignment and classification. The heterogeneity is striking. (b) 3D model reconstructed
from class averages showing three striations. (c) A cutaway view of the 3D model showing the outer shell and the inner
core structure. Reprinted with permission from Proceedings of the National Academy of Sciences (Orlova et al., 1999,
# 1999, National Academy of Sciences, U.S.A).
J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164 141

rapidly mix a solution of ligand sprayed onto a protein complex supported on an aqueous film on
an EM grid. Rapid freezing in liquid ethane can trap the complex. By altering the time interval
between spraying and freezing, it has been possible to observe conformations associated with
reaction times of 1–100 ms. This technique has been used to study the structure of the open-
channel form of the acetylcholine receptor by electron crystallography (Unwin, 1995), and can
also be applied to single particle work. Computer-controlled cryo-EM plungers have now been
developed, allowing the interactions of three different molecules to be studied (White et al., 1998).
Two solutions can be mixed and held in a delay line, before spraying onto a grid containing the
third molecule. Computer-control allows the precise, reproducible, adjustment of individual
parameters. The ability to sort and classify particles that may be in different conformational states
or bound to different substrates or cofactors using MSA may be a significant advantage for time-
resolved studies using single particle analysis. Time-resolved X-ray crystallography has the
disadvantage that the vast majority of the molecules in a crystal must be in the same conformation
at the same time, in order to produce strong diffracted beams. At the moment, time-resolved X-
ray crystallographic studies are technically considerably more difficult to perform than time-
resolved cryo-EM studies. Furthermore, it is possible that large conformational changes will be
restricted in the context of a 3D crystal, or will result in the crystal fracturing. This is less of a
problem with 2D crystals or single particles, because the molecules are less constrained. This is a
possible reason for differences in the M state of bacteriorhodopsin revealed by X-ray and electron
.
crystallography (Kuhlbrandt, 2000). Despite this, time-resolved X-ray crystallography is
producing impressive results (reviewed in Stoddard, 1998), including structures of intermediates
in the bacteriorhodopsin photocycle (Edman et al., 1999; Royant et al., 2000; Sass et al., 2000).
Recent results from X-ray and electron crystallographic studies of bacteriorhodopsin have led to
an atomic model for proton pumping (Royant et al., 2000; Sass et al., 2000; Subramanian and
.
Henderson, 2000; reviewed in Kuhlbrandt, 2000).
A significant advantage for electron microscopy compared to X-ray crystallography is that we
can directly obtain information about phases from EM. This is because in the electron
microscope, electromagnetic lenses can focus diffracted electrons to form images that contain
information about phase as well as amplitude. The significance of this has been dramatically
demonstrated in the atomic model of the ab tubulin dimer, obtained by electron crystallography
(Nogales et al., 1998). The final resolution of the electron density map is at 3.7 A, ( but shows a
well-defined connectivity that is readily interpretable in terms of secondary structure elements
because of the high-quality phase information obtained directly from the EM images. This
allowed an atomic model to be built, despite the fact that the resolution of the density map is
lower than would be used to produce an atomic model from data derived by X-ray
crystallography (Chiu et al., 1999). This shows the potential advantage of electron microscopy
studies in general over X-ray crystallography. However, multiple wavelength anomalous
diffraction (MAD) phasing is now being increasingly used in X-ray crystallography, producing
high-quality phase information very quickly (Hendrickson, 1991; reviewed in Smith, 1991; Ealick,
2000). The future significance of the ability to directly determine the phases by EM may therefore
be that high-resolution X-ray structures could be successfully modelled into larger structures
produced by single particle analysis that are at much lower resolutions.
Single particle analysis and its derived 3D reconstructions cannot yet compete with electron or
X-ray crystallography in terms of resolution. The best resolutions quoted for single particle
142 J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164

analysis are 7.4 A( for the hepatitis B core protein (Bottcher


. ( for the
et al., 1997; Fig. 10) and 7.5 A
(
70S ribosome (Matadeen et al., 1999), although 15–30 A is more usual. This compares for instance
to 0.83 A( for the 46 amino acid residue protein crambin by X-ray crystallography (Stec et al.,
1995). Significant achievements for electron crystallography include the structures of light-
harvesting complex II at 3.4 A ( (Kuhlbrandt
. et al., 1994), bacteriorhodopsin at 3.5 A ( (Henderson
et al., 1990; Grigorieff et al., 1996), the ab tubulin dimer at 3.7 A ( (Nogales et al., 1998), and
(
aquaporin-1 at 3.8 A (Murata et al., 2000). Although these are different molecules posing different
challenges for structural studies, similar trends have been revealed in molecules that have been
studied by more than one of these techniques, such as the ribosome (see Section 7). Resolution in
the EM is limited to about 50 times the wavelength of the incident electrons (Amos et al., 1982),
due to the small apertures that are used in electron microscopes to reduce aberrations in the
electromagnetic lenses (Saibil, 2000b). Given that the high-energy electrons produced in current
microscopes have a wavelength between 0.015 and 0.040 A ( (Chiu et al., 1999), there is no
theoretical limit to obtaining atomic resolution by single particle analysis. The limit is imposed by
several factors, such as beam-damage. Electrons cause less damage than X-rays (Henderson,
1995) and have a greater scattering power (Chiu et al., 1999), but because of the small sample size
(single particles or 2D crystals versus 3D crystals), beam damage is a more significant problem for
single particle analysis and electron crystallography than for X-ray crystallography. Specimen
movement, beam-induced specimen charging, astigmatism, aberrations, beam drift, and the
inability to correct for the CTF perfectly, further limit the resolution for single particle analysis
and electron crystallography. Specimen charging is thought to lead to several problems, including
specimen movement, film breakage upon irradiation, and blurring of electron diffraction patterns.

Fig. 10. 3D map of the hepatitis B virus core protein shell at 7.4 Å resolution, currently the highest obtained by single
particle analysis. This map revealed a novel fold for viral capsid proteins. Reprinted with permission from Nature
(Böttcher et al., 1997, # 1997, Macmillan Magazines Limited).
J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164 143

A model has been proposed for the effects of specimen charging (Brink et al., 1998). During
imaging, a fraction of the incident electrons are inelastically scattered, and the resulting energy
transfer is proposed to lead to the emission of secondary and Auger electrons from the specimen.
This creates a positive charge on the specimen, which could act as an electromagnetic lens, causing
focus changes and blurring or movement of the specimen. The number of single particle images
that can be handled by the image processing programs is a further factor that can limit the
attainable resolution for single particle analysis. The more single particle images that are used in
the reconstruction, the higher the resolution of the final model. Orlova has described a simple
formula for the number of images needed to obtain a 3D model at a given resolution using
angular reconstitution (Orlova, 2000). Molecular symmetry can decrease the number of images
needed for a given resolution by providing redundant motifs in the specimen and geometric
constraints for the alignment steps (Chiu et al., 1999; DeRosier and Klug, 1968).
There is no upper limit to the size of molecule that can be studied by single particle analysis.
This is a significant advantage compared to NMR spectroscopy, where a maximum size of about
.
100 kDa is imposed with the latest multi-dimensional techniques (Wider and Wuthrich, 1999). X-
ray crystallographic studies of 3D crystals with very large unit cells can be difficult (reviewed in
Murali and Burnett, 1991). These crystals may fail to diffract to high resolution, are often sensitive
to radiation damage (possibly because of a high solvent content), and it may not be possible to
resolve individual reflections in the diffraction pattern. The largest asymmetric structure solved to
date by X-ray crystallography is the 70S ribosome (Cate et al., 1999). A minimum size of 100 kDa
is thought to be imposed on single-particle analysis by the requirements of molecular alignment
(Henderson, 1995). Larger macromolecules are easier to align because, at any given resolution,
there is more scattering of the electrons, and thus the image contains more of the information used
to determine the rotational and translational orientations. The a-latrotoxin dimer (260 kDa)
.
(Orlova et al., 2000; reviewed by Saibil, 2000c) and the haemagglutinin trimer (252 kDa) (Bottcher
et al., 1999) are so far the smallest structures solved by single particle analysis.
X-ray or electron crystallography and NMR spectroscopy are therefore the methods of choice
for studying individual molecules or small macromolecular complexes. Single particle analysis is
very useful for studying large complexes that are difficult to analyse by these higher resolution
techniques. It is also of great value in time-resolved studies, in analysis of different functional
states of macromolecules, and in studying conformational flexibility in large complexes.
The random-conical tilt method is an alternative 3D reconstruction technique that can be used
in single-particle analysis (Frank et al., 1988; Radermacher, 1988; Frank and Radermacher, 1992;
Frank, 1996; Chen et al., 1998) and has been implemented in the programs SPIDER and WEB
(Frank et al., 1996). In contrast to angular reconstitution, it makes use of molecules aligned in
preferred orientations on an EM grid. A high-tilt (45–608) image is recorded under low-dose
conditions, followed by a second untilted image of the same area. The particles from the untilted
images are aligned using rotational CCFs, supplying information about their in-plane
orientations. The particles from the tilted images are translationally aligned, and then a 3D
reconstruction of the tilted images is performed, using the in-plane orientations determined from
the untilted images and the common tilt angle. MSA and classification procedures can be used to
group the untilted images into classes. Class averages can be produced of the corresponding tilted
images, and these class averages can then be used for 3D reconstruction. This technique has
several disadvantages compared to angular reconstitution. First, the untilted image is taken of an
144 J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164

area that has already been exposed to electron damage (van Heel et al., 1992b). Although this
image is not directly used in the 3D reconstruction, it is used to orient the tilted images. Secondly,
specimens cannot be viewed at tilt angles greater than 60–708 because of the fragile nature of EM
specimens (Chen et al., 1998). This results in a missing cone in Fourier space that cannot be
sampled (the ‘‘missing cone problem’’). This produces a non-isotropic resolution in real space,
with the resolution being lower in a direction perpendicular to the specimen plane (Chen et al.,
1998; Radermacher, 1988; Schatz et al., 1995). Furthermore, highly tilted images are difficult to
record and are of poorer quality than untilted images (Chen et al., 1998). Since the defocus level
will change across a tilted image, it may be necessary to correct for the CTF separately for
different areas of the EM image (Schatz et al., 1995). For these reasons, angular reconstitution
may be preferred to the random-conical tilt method. For particles with icosahedral symmetry,
such as some viruses, particle orientations can be determined using a further technique developed
by Crowther (1971). In this technique, particle orientations are determined in reciprocal space by
searching for symmetry-related peaks in the Fourier transforms of individual particles (Crowther,
1971; Fuller et al., 1996; Saibil, 2000b), but this technique will not be discussed in detail here.
Single particle analysis and 3D reconstruction techniques are continually developing to resolve
problems. A key difference between single particle work and crystallographic approaches is that in
single particle analysis there are currently no universally accepted resolution criteria like R-factors
(Rhodes, 2000). Instead, criteria such as the Fourier shell correlation (FSC) are used to obtain an
estimate of the reproducible resolution, and there is currently no consensus about the threshold
that should be used (Stewart et al., 1999). Penczek has criticised the determination of resolution
from the intersection point of the FSC function with the 3s curve (Malhotra et al., 1998). He
asserts that the approach gives higher resolutions than if the resolution limit is set at the point
where the correlation coefficient drops below 0.5. He proposes a method for transferring FSC
results into more objective SNR values. Setting the limit at the point where the correlation
coefficient reaches 0.5 corresponds to an SNR of 1.0. Use of the FSC=0.5 threshold has been
supported by comparisons of cryo-EM and X-ray structures of adenovirus type 2 (Stewart et al.,
1999). On the other hand, van Heel has compared the X-ray crystal structure of the Thermus
thermophilus 70S ribosome at 7.8 A ( (Cate et al., 1999) with his cryo-EM structure of the E. coli
(
50S subunit at 7.5 A (Matadeen et al., 1999). He states that the cryo-EM structure does have the
slightly higher resolution, and that the 3s criterion gives a conservative estimate of the resolution
compared to accepted criteria used in X-ray crystallography (van Heel, 2000). Despite this,
comparison of A-form rRNA helices in the two papers shows that, for the X-ray structure, the
deep major grooves, wide minor grooves and the ridges for the phosphates are as expected for
the resolution, and agree well with the atomic models of the ribosome. This is not apparent in the
cryo-EM structure. It is to be hoped that as more structures are solved at comparable resolutions
by X-ray crystallography and single particle analysis, a resolution criterion for single particle
analysis that is equivalent to those used in crystallographic studies may be developed. A further
problem with the Fourier shell correlation technique is that the two 3D reconstructions are not
statistically independent, each having been derived from an earlier 3D model. For low SNR
images, this may lead to an overestimation of the resolution, because of correlation between noise
in the images (Grigorieff, 2000). This correlation of noise is introduced by alignment of the data
set to common references before it is split in half. Alternative methods of determining the
resolution have been described, such as the differential phase residual (DPR). This uses a sum of
J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164 145

Fourier space amplitudes, rather than a multiplication, to weight the phase differences between
two signals in Fourier space (Frank et al., 1981; Penczek et al., 1994). It has been pointed out that
this is flawed, since the differential phase residual changes when one image is multiplied by a
constant, despite this not changing the information content in the images (van Heel, 1987b;
Orlova et al., 1997). A complete description of the resolution should thus include the criterion
used, and the stated resolution of a model should probably not be implicitly trusted. Currently,
resolution may best be assessed by determining the closest features that can be interpreted in the
3D map, given data from other techniques. The selection of references for multi-reference
alignment still has an unfortunate degree of subjectivity, which will have to be addressed in the
future. These points highlight the fact that at the moment the theory and practice of single particle
analysis and 3D reconstruction are less well developed than those behind the crystallographic
approaches and NMR spectroscopy.

7. The ribosome: comparing single particle analysis with X-ray crystallography

The ribosome is a very attractive target for atomic resolution structural studies because they
will lead to a dramatic increase in our understanding of the mechanism of protein synthesis,
RNA–RNA and protein–RNA interactions. Over the past couple of years, tremendous advances
have been made. The 50S subunit of Haloarcula marismortui (H. marismortui) has recently been
determined at 2.4 A ( resolution by X-ray crystallography, allowing almost the entire length of 23S
rRNA, 5S rRNA and 27 of its 31 proteins to be fitted into the electron density map (Ban et al.,
2000; reviewed by Cech, 2000). This has been followed by the X-ray structure of the 30S subunit
of Thermus thermophilus (T. thermophilus) at 3 A ( resolution, allowing all of the ordered regions of
16S rRNA and 20 associated proteins to be fitted into the electron density (Wimberly et al., 2000;
Carter et al., 2000; reviewed by Williamson, 2000).
For many years, it proved difficult to grow 3D crystals of sufficient order and size for X-ray
crystallography. Yonath and co-workers obtained crystals of the 50S subunit from H. marismortui
in the 1980s (e.g. Shevack et al., 1985). However, because of the large asymmetric unit, accurate
phase determination required the use of very large numbers of heavy atom derivatives to produce
measurable differences in the diffraction patterns, with each heavy atom having to be located very
accurately. The ribosomal crystals grown by Yonath’s group have shown extreme radiation
sensitivity (including deformed diffraction spots and unit cell dimensions that increase during
exposure to the X-ray beam), poor isomorphism and non-isotropic mosaicity (Harms et al., 1999).
These problems dramatically stalled progress, leading to single particle analysis and 3D
reconstruction being seen as an attractive alternative.
Recently, the problems with the X-ray crystallographic approach have been resolved. Heavy
metal clusters of tungsten and tantalum atoms and synchrotron radiation sources have been used
to produce measurable differences in the diffraction patterns, greatly reducing the number of
heavy atom derivatives that have to be made. This has provided low-resolution phase information
(Ban et al., 1998, 1999; Cate et al., 1999; Clemons et al., 1999). Subsequently, most phasing
information has come from anomalous scattering from osmium and iridium compounds (Ban
et al., 1999, 2000; Clemons et al., 1999; Wimberly et al., 2000). The Yale group, working on the
50S subunit, did not encounter the problems of extreme radiation sensitivity and cell dimension
146 J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164

change in their crystals, despite the fact that they were working on the same crystal form, and
collected the entire 2.4 A ( data set using only two crystals. X-ray structures have been pro-
duced which are at much higher resolutions than were obtained by cryo-EM work, and have
revealed a wealth of information about ribosome structure (Ban et al., 1998, 1999, 2000; Cate
et al., 1999; Clemons et al., 1999; Nissen et al., 2000). Rather than being an exhaustive review of
structural work on the ribosome (good reviews can be found in Green and Puglisi, 1999; Davies
and White, 2000; van Heel, 2000; Puglisi et al., 2000; Brimacombe, 2000), this section will
compare the use of single particle analysis and X-ray crystallography as revealed by work on the
ribosome, and will describe some of the resulting technical developments in single particle
analysis.
In 1976, Lake published the first attempt to model the 30S and 50S subunits, and the 70S
ribosome, from E. coli (Lake, 1976), based on the visual interpretation of raw, noisy micrographs
of particles imaged in negative stain. The subjective nature of such interpretations led to the
development of MSA, automatic classification and averaging techniques which could be used to
.
produce characteristic class averages of ribosomes (e.g. Frank et al., 1982; van Heel and Stoffler-
Meilicke, 1985; Verschoor et al., 1984). The more sophisticated image processing techniques
allowed much more detailed analysis of structure. The importance of the MSA technique was
illustrated by eigenvector analysis of the 50S subunit indicating conformational flexibility in the
L7/L12 stalk (Harauz et al., 1988). This has subsequently been confirmed by the fact that the L7/
L12 stalk region consistently appears less elongated in the X-ray structures (Ban et al., 1998, 1999,
2000; Cate et al., 1999) than in cryo-EM work (Stark et al., 1995, 1997b). However, it is interesting
to note that the L7/L12 stalk appears to be bent inwards in some cryo-EM specimens (Frank et al.,
1995; Stark et al., 1995,1997a; Matadeen et al., 1999) compared to negative stained specimens (e.g.
Harauz et al., 1988). It was thought that the extended stalk might be an artefact produced by
negative staining. Nevertheless, extended stalks can be produced in cryo-EM samples under
certain conditions (Agrawal et al., 1996; Malhotra et al., 1998). The presence of the extended stalk
seems to depend on glow-discharging of the EM grids (Matadeen et al., 1999) and the display
threshold (Malhotra et al., 1998). This further supports conformational flexibility, and illustrates
the importance of using a range of sample preparation conditions (stains, cryo-EM, glow-
discharging of grids) before reaching firm conclusions about a structure.
In the 1990s, cryo-EM revolutionised studies of ribosome structure in terms of the attainable
resolution. In 1995, Frank produced a 25 A ( structure of the 70S E. coli ribosome based on image
processing of micrographs from untilted specimen grids (Frank et al., 1995). This revealed a
channel in the 30S subunit, predicted to be a pathway for incoming mRNA, and a bifurcating
tunnel in the 50S subunit, possibly constituting a pathway for the nascent polypeptide chain,
preventing its hydrolysis (Green and Puglisi, 1999). CTF correction was applied to this structure,
and the CTF-corrected model showed fewer tunnels than the uncorrected model. This illustrates
the importance of using CTF correction to restore the correct weight of low spatial frequencies
(Frank et al., 1995). This was followed later in the same month by the structure of the E. coli 70S
ribosome at 23 A ( (Stark et al., 1995). The authors attributed the higher resolution, compared to
earlier work in the 30–55 A ( range, to the use of the angular reconstitution rather than the random
conical tilt method. This study showed an extensive system of channels in the 50S subunit (called
the exit channel complex}ECC). The diameter of the channels was about 20 A. ( Higher resolution
rRNA structures, in particular double-helical regions of 16S rRNA, could be fitted into this
J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164 147

model. With the benefit of the atomic structures of the ribosome, it is now apparent that some of
the extensive channels revealed by these cryo-EM studies are only seen at low resolution. All atom
CPK representations of the ribosome show few gaps inside the molecule, with the exception of the
cleft and channel for mRNA in the 30S subunit, and the exit channel for the nascent polypeptide
in the 50S subunit.
The structure of the 50S subunit from E. coli has recently been published at 7.5 A ( resolution
(Matadeen et al., 1999). This represents a dramatic increase in the resolution obtainable by single
particle analysis, and is significant since it reaches a nominal resolution comparable to that
.
obtained for the hepatitis B virus core protein shell (Bottcher et al., 1997). This resolution was
achieved by processing about 16,000 particles, compared to about 6000 for the hepatitis B
structure in which the icosahedral symmetry of the particle could be exploited. However, it is
important to note that the resolution of the 50S subunit is based on FSC with the 3s threshold
criterion, whereas that for the hepatitis B virus is based on the FSC=0.5 criterion. Indeed, the 50S
structure is not comparable to the hepatitis B structure, where the fold of the coat protein could be
deduced from the structure, or to the 7.8 A ( crystal structure of the 70S ribosome (Cate et al.,
1999). The increase in resolution obtained in this study is attributed to the development of a novel
program for correcting the CTF, implemented in IMAGIC (Matadeen et al., 1999). This program
allows the detection of local defocus and astigmatism variations within a single micrograph. CTF
correction was then applied by reversing the phases beyond each CTF zero crossing, using the
local parameters. The CTF correction was not used to alter the amplitudes. The image processing
was begun by extracting the 50S subunit from an earlier cryo-EM reconstruction of the 70S
ribosome. Reprojections from this earlier model were then used as references for a multi-reference
alignment of the dataset. This approach can vastly speed up the time taken for image processing.
A further interesting point revealed in this study is the use of the eigenvectors produced by MSA
to look for magnification differences within the dataset. The final model shows the classical
features expected of the 50S subunit, such as the central protuberance, the A-site finger, the L7/
L12 stalk and the L1 protuberance. It has also revealed new details, including a collar around the
L1 stalk into which the L9 ribosomal protein could be modelled. Regions of 23S rRNA could also
be modelled into the structure, and a second stalk-like structure has become visible below the L7/
L12 stalk. By comparing this structure with a 13 A ( map of kirromycin stalled 70S particles, the
authors have revealed conformational changes in the 50S subunit between the free and bound (in
the 70S complex) forms.
Shortly after this, the structure of the E. coli 70S ribosome at 11.5 A ( resolution (on the basis of
the FSC=0.5 criterion) was published (Gabashvili et al., 2000). This was based on the processing
of 73,523 particles of the ribosome bound to fMet-tRNAMet f , i.e. in its initiation-like complex. In
this study, the authors used a novel technique for correcting the under-representation of the
Fourier amplitudes at higher spatial frequencies. This under-representation is due to specimen
charging, specimen drift, instrumental instabilities and partial coherence (Gabashvili et al., 2000).
X-ray solution scattering measurements were used to correct the Fourier amplitudes up to the
1/11.5 A( 1 cut-off. The quality of the reconstruction was assessed by comparing the tRNA part
with low-resolution (about 10 A) ( maps calculated from the atomic co-ordinates of yeast initiator
Met
tRNA and E. coli fMet-tRNAf . The shape and the geometry of the tRNA in the cryo-EM map
agrees well with that of E. coli fMet-tRNAMet f , supporting the reliability of other fine details in the
reconstruction. The final reconstruction shows great detail}RNA helices (showing the right-
148 J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164

handed twists and deep grooves characteristic of A-form rRNA), peripheral proteins and
intersubunit bridges can be identified. The resolution actually appears comparable to the 7.5 A (
structure of the 50S subunit (Matadeen et al., 1999). The L1, L6, L14, L11 ribosomal proteins, a
fragment of 23S rRNA, and the a-sarcin-ricin loop can be modelled in, and EF-G can be
positioned to contact the a-sarcin-ricin loop.
Single particle analysis of the unbound 30S subunit (i.e. not attached to 50S) has proved
problematical, and only low resolutions have been achieved. Marin van Heel’s group have
determined the structure of the 30S subunit from T. thermophilus at 25 A ( (based on the 3s
criterion) (Harms et al., 1999). The authors attribute the low resolution to the flexibility and
instability of 30S particles, resulting in the structures produced by single particle analysis
representing a range of conformations, decreasing the resolution (Harms et al., 1999). Joachim
Frank’s group have published the structure of the 30S subunit from E. coli at 23 A ( (based on the
FSC=0.5 criterion) (Gabashvili et al., 1999). The authors noticed that the final 3D model was
blurred compared to the structure of the 30S subunit in a model of the 70S ribosome. They also
attributed the low resolution to conformational flexibility in the 30S subunit, so that the 23 A (
model represents an average of different conformations. Computer simulations of conformational
flexibility in 50S-bound 30S confirmed this. Averaging of the resulting models led to the dis-
appearance of fine structural details, and the average looked more like the 23 A ( model. The high
homogeneity of the 30S sample after sucrose density gradient centrifugation showed that the
structural heterogeneity was not an artefact produced by the isolation or purification steps, but
rather represented real conformational flexibility in the 30S subunit. The authors then decided to
computationally purify the data set, aiming to produce sub-populations of particles, which could
be separately reconstructed to give 3D models of the different conformers. They proceeded by
initially hypothesising that the conformations favoured by the 50S-bound 30S subunit would be
represented in the conformations assumed by unbound 30S. Two previously published
reconstructions showed 30S in different conformations}these earlier models were the 70S
ribosome binding fMet-tRNAMet f (Malhotra et al., 1998) and the 70S ribosome binding EF-G
(Agrawal et al., 1998). The 30S subunit was extracted from each of these models and used as
references for the computational purification. Particles were compared in turn to both models,
and assigned to one or the other on the basis of cross-correlation coefficients. 3D models were
calculated for each sub-population, and then the particles were reassigned to these new models.
This procedure was iterated until stable results were obtained. The resolution of the 3D models of
the conformers was 32 A ( (FSC=0.5 criterion). This is lower than the model representing the
‘‘average’’ structure, due to the smaller number of particles used to produce the models for each
conformer compared to the average (Gabashvili et al., 1999). However, the conformers showed
( By comparing the conformers with
better structural details than the ‘‘average’’ structure at 23 A.
models of the 50S-bound 30S subunit, conformational flexibility in the 30S subunit upon
associating with the 50S subunit has been revealed, in particular, the relative movements of three
main structural domains (the head, platform and main body). As a consequence of this study, the
authors make two important points about the technique of computational purification. First,
there is a compromise between achieving a high degree of homogeneity and a high enough
resolution to allow the reconstructions of the different conformers to be compared. This is because
a high degree of homogeneity requires a large number of sub-populations, thus fewer particles in
each sub-population, resulting in a lower resolution. The authors suggest that particles should be
J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164 149

drawn from two distant regions of the conformational space, and then the sub-populations should
be made large enough to produce reconstructions at a moderate resolution. Secondly, small
conformational changes may not be detected by the procedures used for conformational
purification, and may thus be averaged out in the final maps of the conformers. Conformational
purification therefore requires the processing of many thousands of particles to produce large sub-
populations, and may not reveal fine conformational changes.
The cryo-EM structures have been validated by the recent X-ray structures, indicating that
neither specimen preparation nor single particle analysis and 3D reconstruction are producing
distorted structures. The 5.5 A( (Clemons et al., 1999) and 3 A ( (Wimberly et al., 2000) structures of
the 30S subunit from T. thermophilus confirms the overall shape and dimensions revealed by cryo-
EM structures. The 5 A ( (Ban et al., 1999) and 2.4 A( (Ban et al., 2000) maps of the 50S subunit
from H. marismortui have the same overall shape as that determined from cryo-EM work. In the
5A ( structure, four tungsten clusters line the polypeptide exit channel, suggesting that the diameter
of the channel is about 20 A.( This has been confirmed by higher resolution X-ray studies, showing
that the diameter of the tunnel varies from 10 to 20 A,( and is on average 15 A( (Nissen et al., 2000).
(
Comparing this to the value of 20 A obtained by Stark and colleagues (1995) indicates that
structure flattening due to variable dehydration under the vacuum conditions of the EM has not
been a significant problem in the cryo-EM work. Interestingly, electron density for the L1 protein
( cryo-EM structure (Matadeen et al., 1999) and at a similar position in the
is visible in the 7.5 A
(
9 A X-ray structure (Ban et al., 1998), but is missing in the higher resolution X-ray maps (Ban
et al., 1999, 2000). It has been pointed out that in the 5 A ( map the authors have positioned the
L11-rRNA complex in densities assigned to the L7/L12 stalk in the cryo-EM models (Matadeen
et al., 1999). It may be significant that there is no electron density for the N-terminal domain of
L11 in the 5 A ( map (Ban et al., 1999). Furthermore, in the 7.5 A( cryo-EM map, there is insufficient
density in this region to accommodate the L11-rRNA complex (van Heel, 2000). All of L11 is
present in the 11.5 A ( cryo-EM structure (Gabashvili et al., 2000) but this disagrees with the
position in the 7.5 A ( cryo-EM map (Matadeen et al., 1999). Unfortunately, there is no clear
electron density for L11 in the 2.4 A ( map, so this problem remains unresolved. The 7.8 A ( map of
the 70S subunit from T. thermophilus (Cate et al., 1999) confirms the general shape of the
ribosome revealed by single particle analysis, such as the 3-pointed crown appearance of the 50S
subunit. Comparison of the 11.5 A ( cryo-EM structure of 70S with X-ray crystal structures shows
the high degree of similarity between them (Gabashvili et al., 2000). A few differences have been
noticed, for example in the intersubunit region, and this has been attributed to the constraints of
crystal packing (Gabashvili et al., 2000). Some other small differences between cryo-EM and X-
ray structures have been attributed to the evolutionary separation of the organisms, for example,
H. marismortui does not have the L9 ribosomal protein (discussed in van Heel, 2000).
Compelling evidence for the fact that experimental conditions, single particle analysis and 3D
reconstruction are not producing incorrect or misleading structures comes from the use of cryo-
EM structures to solve the phases for the 9 A ( map of the 50S subunit (Ban et al., 1998) and the
(
7.8 A map of the 70S ribosome (Cate et al., 1999). In both cases, cryo-EM images were positioned
in the crystal lattice and used to provide initial phases for low-resolution reflections. These were
accurate enough to locate the bound heavy metal clusters using difference Fourier maps. For the
50S map, phases were subsequently determined by multiple isomorphous replacement and
anomalous scattering (MIRAS) from the bound heavy metal clusters (Ban et al., 1998). For the
150 J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164

70S map the located heavy atom clusters were subsequently used in MAD phasing experiments
(Cate et al., 1999). The molecular replacement search showed a very good match, implying that
the cryo-EM structure was very superimposable on the X-ray structure. The validity of cryo-EM
structures of the T. thermophilus 70S ribosome is supported by studies in which pure molecular
replacement searches, using a cryo-EM structure, were used to reveal the packing of the molecules
in the 3D crystals (Harms et al., 1999).
Single particle analysis has been especially useful in the study of ribosomes binding their
ligands. This is because cryo-EM allows molecules to be studied in a wide range of solution
conditions without the requirements of crystallisation, allowing ligands and antibiotics (which
can stall the ribosome in particular states) to be added, while single particle analysis allows
computational purification of the resulting sample. In 1996, a 3D cryo-EM structure of poly(U)-
programmed E. coli 70S ribosomes with the A, P and E sites bound by deacylated
tRNAPhe molecules was published (Agrawal et al., 1996). The following year, 20 A ( structures
of the E. coli 70S ribosomes in their pre- and post-translocational states were published
(Stark et al., 1997a). Density difference maps were used to reveal the positions of the tRNA
molecules in the A, P and E sites, and atomic models of the tRNA molecules were fitted to the
density maps. The positions of the tRNA molecules disagree with those proposed by Agarwal,
and the authors attribute this to the fact that the earlier model is ‘‘non-physiological’’, with
three tRNA molecules bound per ribosome. A model of three tRNAs bound to the A, P and E
sites of the ribosome is shown in the 70S X-ray structure (Cate et al., 1999). These binding
sites were located by difference mapping. It has been stated that the positions of tRNAs
in the A and P sites support the work of Stark and disagree with that of Agrawal (van Heel, 2000).
The positions of A, P and E site tRNAs in the 11.5 A ( cryo-EM structure (Gabashvili et al., 2000)
agree well with those in the 70S X-ray structure (Cate et al., 1999) (reviewed in Davies and White,
2000). A 3D reconstruction of kirromycin stalled E. coli ribosomes binding the aminoacyl-
tRNA  EF-Tu  GTP ternary complex at 18 A ( has been published (Stark et al., 1997b). This 3D
reconstruction was performed using an earlier model of a pre-translocational ribosome as an
initial ‘‘anchor set’’. It might be thought that this could bias the model towards the earlier
structure. However, given the iterative refinements that were applied, the effect is likely to be
negligible. EF-G has been visualised bound to the naked E. coli 70S ribosome (lacking mRNA
and tRNA) (Agrawal et al., 1998), and now bound to the ribosome in pre- and post-
translocational complexes (Agrawal et al., 1999; Stark et al., 2000). These cryo-EM studies have
proved useful in the interpretation of the X-ray crystal structures, for example the 50S structure at
5A ( (Ban et al., 1999) used cryo-EM work to position EF-G and EF-Tu complexed with an
aminoacyl-tRNA and GTP onto the ‘‘factor binding domain’’. Comparison of cryo-EM
structures in different functional states obtained by single particle analysis has now led to a low-
resolution picture of translocation, indicating that it is a two-step process (Frank and Agrawal,
2000).
The eukaryotic ribosome has also been studied by single particle analysis, at 25 A ( for the rat
(
liver ribosome (Dube et al., 1998b), and 21 A for the rabbit reticulocyte ribosome (Dube et al.,
1998a). It is much more complex than the prokaryotic ribosome. For example, the prokaryotic
large subunit has 31 proteins and the small subunit 21, compared to the eukaryotic large subunit
with 49 proteins and the small subunit with 33 (Lewin, 2000). As yet the eukaryotic ribosome is
not accessible to X-ray analysis.
J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164 151

Cryo-EM work has thus been of vital importance in obtaining medium resolution structures of
ribosomes, and in studying their functional interactions with ligands, providing a wealth of data
and biochemically testable hypotheses. The potential for structure distortion problems with cryo-
EM, single particle analysis and 3D reconstruction does not appear to have been realised. This
conclusion has been supported by comparisons of virus structures solved by X-ray crystal-
lography and cryo-EM (reviewed by Baker and Johnson, 1996). However, only in the last couple
of years has single particle analysis provided high enough resolutions to enable X-ray crystal
structures of ribosomal components to be modelled into the electron density with real confidence
(Malhotra et al., 1998; Matadeen et al., 1999; Gabashvili et al., 2000). X-ray crystallography has
now revolutionised structural studies of the ribosome, and this work has partly been driven by the
use of cryo-EM structures as phasing models. The 2.4 A ( resolution structure of the 50S subunit
from H. marismortui (Ban et al., 2000) has already revealed that the ribosome is a ribozyme and is
providing fascinating details of RNA tertiary structure interactions (Nissen et al., 2000; reviewed
in Cech, 2000). The 3 A( structure of the 30S subunit from T. thermophilus (Wimberly et al., 2000)
has provided information about protein–RNA and RNA–RNA interactions, with the latter
frequently using the minor groove as an interaction surface. The same group has now revealed
molecular details of the interactions made by A-, P- and E-site codons and tRNA with the 30S
subunit, and have studied the structural basis for the action of the antibiotics paromomycin,
streptomycin and spectinomycin (Carter et al., 2000). Furthermore, the atomic resolution 30S
structure has revealed that the initial step in decoding tRNA may involve the flipping out of bases
A1492/A1493, allowing them to form a decoding surface that can monitor the width of the minor
groove of the codon–anticodon helix, allowing discrimination of cognate and near-cognate
tRNAs (Carter et al., 2000). However, crystals suitable for X-ray diffraction analysis have only
been obtained from halophilic or thermophilic bacteria. E. coli and eukaryotic ribosomes still
remain stubbornly resistant to analysis by X-ray crystallography, and at the moment are only
amenable to cryo-EM and single particle analysis. By using the atomic resolution crystal
structures to interpret models of the different functional states of the ribosome that are being
revealed by cryo-EM and single particle analysis, it should become possible to obtain a complete
knowledge of protein synthesis at the atomic level.

8. Photosystem II

Photosystem II (PSII) is the multisubunit membrane protein found in cyanobacteria, red and
green algae, and higher plants that harnesses light energy in order to split water into molecular
oxygen and reducing equivalents (for reviews of the structure and function of PSII, see Barber
.
et al., 1997; Barber, 1998; Barber and Kuhlbrandt, 1999). PSII catalyses one of the most strongly
oxidising reactions known to occur in biology (Hankamer et al., 1997), ultimately reducing carbon
dioxide to organic molecules, thereby producing almost all global biomass (Boekema et al.,
1998a). It has therefore long been the target of structural studies. These have proved to be
particularly difficult because of the labile nature of this macromolecular complex. As will be
discussed below, it is possible to isolate different complexes of PSII varying in the amount of
peripheral proteins they contain. Furthermore, some of the proteins themselves are intrinsically
unstable, for example the D1 protein has a half-life of approximately 30 minutes in moderate light
152 J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164

(Barber, 1998). The relative ease with which single particle analysis can study membrane proteins
and the ability to computationally purify the data set has allowed this technique to play a crucial
role in recent structural studies of PSII. This section will show how single particle analysis has
advanced our understanding of the structure and function of PSII. First, it is important to briefly
summarise some aspects of PSII structure.
PSII essentially consists of two components: the reaction centre and an inner antenna light-
harvesting system. The reaction centre is responsible for the light-driven water-splitting reaction.
It consists of the D1 and D2 proteins, the a- and b-subunits of cytochrome b559, and the PsbI
protein. The D1 and D2 proteins bind the cofactors necessary for the light-driven charge
separations. The function of the inner antenna, which is composed of the chlorophyll a-binding
proteins CP43 and CP47, is to funnel light-energy towards the reaction centre. CP43 and CP47,
together with D1 and D2, the a- and b-subunits of cytochrome b559 and some other low
molecular weight proteins (less than 10 kDa), form a core responsible for all the electron transfer
reactions. This core has been observed predominantly in a dimeric form. The full PSII complex
capable of oxygen-evolution consists of the reaction centre, the inner antenna, and an extrinsic
oxygen-evolving complex (OEC) which binds the Mn4 cluster. The OEC consists of the proteins
PsbO (33 kDa), PsbP (23 kDa) and PsbQ (16/17 kDa) in higher plants, and PsbO, PsbV (15 kDa)
and PsbU (12 kDa) in cyanobacteria. Outside of these ‘core’ PSII components is a family of
membrane-bound chlorophyll a=b-binding (Cab) proteins, including light-harvesting complex II
(LHCII). The membrane-bound Cabs are not present in cyanobacteria, where they are replaced
by soluble extrinsic phycobiliproteins. When the components of the PSII core are bound to the
Cab proteins CP29, CP26 and LHCII, they constitute the PSII-LHCII supercomplex (Boekema
.
et al., 1995; Barber and Kuhlbrandt, 1999).
Single particle analysis of PSII has so far been dominated by studies of negatively stained
specimens. In 1995, work was published on PSII reaction centre cores from both spinach and the
thermophilic cyanobacterium Synechococcus elongatus (S. elongatus), and also the spinach PSII-
LHCII supercomplex (Boekema et al., 1995). Alignment and MSA techniques, implemented in
IMAGIC, were used to produce class averages for these different PSII complexes. The resulting
top-view class averages showed two-fold symmetry, and the side-views showed protrusions
attributed to the extrinsic proteins of the OEC. Comparison of the top views of the supercomplex
with the dimeric core complexes provided, for the first time, evidence that the core complex was
located in the centre of the supercomplex, with the peripheral regions accommodating the Cab
proteins. Top-view class averages were produced of core dimers following Tris-washing at pH 8.0,
a treatment that removes all of the OEC lumenally bound mass including PsbO, the major 33 kDa
OEC protein. Calculating a difference map of Tris-washed and non-washed core dimers localised
the likely position of at least this 33 kDa subunit. Combined with other biochemical data, this
single particle work led the authors to propose a tentative structural organisation for the
supercomplex. Their conclusions still stand, although more components have been assigned
places. Furthermore, this work provided support for earlier ideas that PSII is a dimer in vivo, due
to the two-fold symmetry and the observation that monomeric core complexes with bound LHCII
are never seen. The supercomplex analysed in this study lacked the 23 kDa extrinsic protein.
Subsequently, the conditions used in the preparation of PSII from spinach were altered,
successfully resulting in the purification of supercomplexes containing the 23 kDa protein, which
could be analysed to produce top- and side-view class averages (Boekema et al., 1998a). MSA was
J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164 153

used to separate a group of single particle images into those containing the 23 kDa protein and
those lacking it. Difference maps were then used to indicate the position of the 23 kDa subunit.
Projection maps produced by single particle analysis have been combined with higher resolution
projection maps of the reaction centre subcore (the PSII core lacking CP43) (Rhee et al., 1998)
.
and the structure of the LHCII trimer (Kuhlbrandt et al., 1994), both obtained by electron
crystallography, to produce improved models of the subunit positioning (Barber et al., 1999). The
resulting model was supported by other biochemical data, such as cross-linking studies.
One of the first examples of the use of single-particle cryo-EM in the PSII field came with the
publication of the 3D structure of the higher plant PSII-LHCII supercomplex from higher plants,
at a resolution of 24 A( (Nield et al., 2000c and see Figs. 3, 6 and 7 in this review). Computational
purification proved to be necessary and of great help in this study. Reference-free alignment of
almost 16,000 particles identified sub-populations differing in size and shape. These sub-
populations represented complexes that had lost some peripheral proteins during sample
preparation. Each sub-population was treated separately to produce a 3D reconstruction. These
reconstructions provided a framework in which higher resolution structures obtained by electron
crystallography (Rhee et al., 1998; Hankamer et al., 1999) could be incorporated. For the first
time, this allowed the binding sites of the OEC proteins to be related to the underlying intrinsic
membrane proteins, information that will be essential for understanding the water-splitting
reaction. This structure further emphasises the dimeric nature of the complex, and the OEC
proteins are revealed as having a tetrameric appearance and being bound to the lumenal surface,
as had been indicated in earlier freeze-etch studies at lower resolution (Seibert et al., 1987). This
model has been refined, giving a better distribution of protein density in the peripheral regions,
allowing the LHCII trimers to be positioned more accurately (Nield et al., 2000a), shown here in
Fig. 11. Recently, single particle analysis of negatively stained PSII supercomplexes, obtained
after treatment to remove the extrinsic proteins, has revealed significant conformational changes
in the peripheral light-harvesting system (Boekema et al., 2000a). This has indicated that the role
of the extrinsic subunits may be to ensure a directed transfer of excitation energy through the
peripheral antenna proteins CP26 and CP29, or to maintain sequestered domains of inorganic
cofactors, a requirement for the water-splitting reaction (Boekema et al., 2000a).
The 3D structure of the higher plant PSII supercomplex was followed by 3D structures of the
PSII supercomplex from the green alga Chlamydomonas reinhardtii and the core complex from S.
elongatus (Nield et al., 2000b). These structures were at the lower resolution of 30 A,( and are based
on image analysis of negatively stained samples. The C. reinhardtii supercomplex is very similar to
the higher plant supercomplex, whilst the S. elongatus core complex is very similar to the higher
plant core complex. By building higher resolution models of the core into these structures, it has
again been possible to relate the positions of their OEC proteins to the underlying intrinsic
membrane proteins. The S. elongatus core complex structure supports earlier work presenting top-
view class averages of the complex (Kuhl et al., 1999). This earlier work also shows class averages
of ‘‘double dimers’’}two core complexes interacting via their tips, hinting at the possible
arrangement of these core complexes in vivo.
Boekema and colleagues have performed several elegant studies attempting to search for
complexes larger than the supercomplex that may exist in vivo (Boekema et al., 1998b, 1999b,
1999a, 2000b). They have successfully taken the approach of subjecting PSII membranes to partial
detergent treatment, analysing the solubilised particles by single particle analysis, and again using
154 J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164

Fig. 11. Superposition of the helical organisation of the known subunits of the PSII-LHCII supercomplex, derived
from electron crystallography (Kühlbrandt et al., 1994; Rhee et al., 1998; Hankamer et al., 1999) on the latest model of
the supercomplex obtained by single particle analysis of cryo-EM images. Adapted from Fig. 7 of Nield et al., 2000a,
with the permission of The Royal Society, London.

MSA to computationally purify the dataset. This work has shown that LHCII can bind in three
types of binding sites, called strong (S), moderate (M) and loose (L) (Boekema et al., 1998b,
1999b), and has identified three different PSII megacomplexes (Boekema et al., 1999b, 1999a).
Later work, using MSA to classify images of crystalline arrays of PSII from the granal membranes
of spinach has indicated that crystalline regions contain predominantly supercomplexes, and that
the megacomplexes may be relegated to non-crystalline areas (Boekema et al., 2000b). This work
has further identified that PSII-LHCII supercomplexes in one membrane are usually adjacent to
domains containing only LHCII in the opposing membrane, hinting at resonance energy transfer
from LHCII domains to PSII-LHCII complexes across the membranes. The main conclusion
from this work is the heterogeneous nature of the associations between PSII and LHCII. It has
been proposed that the formation of these different complexes may allow PSII to react to light and
J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164 155

stress conditions by providing different routes of excitation energy transfer (Boekema et al.,
1999b).
In just five years, single particle analysis has led to rapid progress in our understanding of PSII
structure at the macromolecular level. The ability to rapidly computationally purify and study
individual complexes has led to new insights into the arrangement of PSII, and to new proposals
for its function. By modelling higher resolution structures into the maps provided by single
particle analysis, we are beginning to understand the arrangement of proteins within the
supercomplex. Future work will involve obtaining higher resolution structures of supercomplexes
and megacomplexes, using cryo-EM to model in structures obtained by X-ray and electron
crystallography, thus gaining a near atomic resolution picture of these complexes.

9. Electron tomography

Electron tomography of organelles and whole cells has now grown out of the 3D reconstruction
methods used in single particle analysis (reviewed in Baumeister et al., 1999). In electron
tomography, specimens with a thickness of 0.25–2.0 mm are imaged in TEMs operating at higher
accelerating voltages than are used in single particle analysis or electron crystallography (400–
1200 kV compared to about 120 kV). The higher voltages are essential due to the increased
specimen thickness. The specimen is imaged using tilts of 60–708 at increments of 1–28. The
images have to be precisely aligned, and finally back-projected to produce the 3D model.
Tomographic reconstruction can be implemented in SPIDER (Frank et al., 1996) using alignment
and weighted back-projection algorithms (for example McEwen et al., 1986; Deng et al., 1999) or
in related image processing programs (e.g. Perkins et al., 1997b). Electron tomographic
reconstruction of cilia (McEwen et al., 1986), a whole cell of the archaebacterium Pyrodictium
abyssi (Baumeister et al., 1999) and mitochondria (e.g. Mannella et al., 1997; Perkins et al., 1997a;
Deng et al., 1999) have been attempted. Currently, the resolution lies in the range 50–70 A (
(Baumeister et al., 1999), but improvements in the technique should decrease this towards the
limit of about 28 A( imposed by radiation damage (Glaeser, 1999). Low-dose automated methods
are now available (e.g. Rath et al., 1997), allowing specimens to be imaged whilst embedded in
vitreous ice (after plunging in liquid ethane). This cryo-electron tomography has recently been
applied to Neurospora mitochondria (Nicastro et al., 2000). Tomographic studies of mitochondria
have led to improved models of cristae structure. Furthermore, studies of mitochondria from
patients suffering from mitochondrial myopathies may reveal more information about
mitochondrial function and the effects of disease (Frey and Mannella, 2000). For examples of
electron tomography using the programs SPIDER and Sterecon, see http://www.wadsworth.org/
spider doc/spider/vrml/uncomp/index.html.

10. Future prospects and conclusion

Improvements to single particle analysis and 3D reconstruction are continually being


investigated. In future, it will be important to find ways to reduce specimen drift and beam-
induced charging (Chiu et al., 1999). Microscopes with liquid helium cooled superconducting
156 J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164

objective lenses have been developed, such as SOPHIE (Superconducting Objective lens in a
PHIlips Electron microscope) (Zemlin et al., 1996). These microscopes keep the specimen
temperature near 4 K, and their use has been shown to lead to a two-fold reduction in beam-
damage compared to liquid nitrogen temperatures. This is based on a quantitative evaluation of
the rate at which diffraction spots from 2D crystals of bacteriorhodopsin fade (Stark et al., 1996).
Stages cooled by superfluid helium have been produced, reducing vibration problems that are
associated with conventional cryo-microscopes, and allowing very high instrumental resolutions
to be achieved (better than 2 A)( (Fujiyoshi et al., 1991; Fujiyoshi, 1998). The operation of these
microscopes requires considerably more sophisticated instrumentation and there are practical
problems associated with the poor conductivity of carbon and ice at liquid helium temperatures
(Stowell et al., 1998). Improvements in the resolution of charge-coupled device (CCD) cameras
may eventually see them replacing photographic film for high resolution TEM work, removing the
need for densitometry (for a recent review see Faruqi and Subramanian, 2000). Image processing
programs are continually being developed and updated. These should eventually provide more
reliable methods of compensating for instrumental factors, such as better CTF correction (e.g.
Conway and Steven, 1999; Skoglund et al., 1996). Frank and colleagues have developed a method
of 3D reconstruction that includes CTF correction, eliminating CTF correction as a separate step
in the image processing (Zhu et al., 1997). Grigorieff has developed FREALIGN (Fourier
REconstruction and ALIGNment) which can correct for the CTF in an image (including
astigmatism, and both amplitude and phase contrast components), calculate a 3D reconstruction,
and then refine the angles and translational orientations of particles (Grigorieff, 1998). An exciting
recent software development is EMAN (Electron Micrograph ANalysis) (Ludtke et al., 1999).
This program is designed specifically for high-resolution (beyond 10 A) ( single particle analysis,
and can employ both amplitude and phase CTF correction. As data-handling capacity increases,
it should become possible to successfully align and assign the orientations to larger single particle
data sets, increasing the attainable resolution. Eventually, it should become possible to automate
data collection and structure determination, as is now usual for X-ray crystallography.
The use of medium resolution structures obtained from single particle analysis as phasing
models for X-ray analysis of macromolecular complexes may increase in the future, particularly
when conventional methods of phasing cannot be applied. There is, however, an important
caveat. Flexible particles, such as the 30S ribosomal subunit, may result in a reconstruction that
represents an average of a range of conformations. This is unlikely to be suitable for molecular
replacement searches, since the conformation of the molecule in the crystals is likely to be different
to that in the reconstruction (Harms et al., 1999).
It is now possible to gain atomic resolution details of macromolecular complex complexes by
combining high-resolution X-ray structures of individual components with the medium resolution
structure of the entire assembly, obtained by electron microscopy. With X-ray crystallography
and NMR spectroscopy dominating the high-resolution study of individual protein molecules, the
future of single particle analysis probably lies in producing moderate resolution structures of large
complexes that will serve as the scaffold for hanging the high-resolution structures of components.
It has been suggested that structures determined by EM need to be at less than 7 A ( in order to
decrease the positional uncertainty when modelling X-ray structures within them (Chiu et al.,
1999). This currently lies at the limit of the attainable resolution with single particle analysis.
However, Rossman believes that combining X-ray structures with a cryo-EM model at 22 A ( can
J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164 157

produce structural information accurate to 2.2 A ( (Rossmann, 2000), and domains of GroEL from
a crystal structure have been successfully docked into 12 and 15 A ( cryo-EM maps (Roseman,
2000). Recent examples of the success of this combined approach are the analysis of clathrin
(Musacchio et al., 1999), PSII (Nield et al., 2000a-c), and Semliki Forest Virus (Mancini and
Fuller, 2000). Combining electron tomographic analysis of organelles with molecular structures is
a very exciting prospect.
It will become important to develop a single database, analogous to the Protein Data Bank
(PDB), for storing 3D models obtained by single particle analysis and electron tomography,
allowing users to download and manipulate the models (e.g. to import higher resolution data).
Proposals for such a database have been made (Marabini et al., 1996; Carazo and Stelzer, 1999)
and some prototypes can be seen at http://www.wadsworth.org/spider 3d/home page.html and
http://www.bioimage.org/. Structures derived from single particle analysis are now being
deposited in the Molecular Structures Database (MSD), which is superseding the PDB (Mancini
and Fuller, 2000). There may however be a reluctance to provide models for such databases, since
they provide the ideal references for multi-reference alignment, vastly speeding up the processing
of a competitor’s images!
In 1982, a paper commented on the potential of single particle analysis and MSA to help us
‘‘recognise subtle differences in molecular structure’’ (Frank et al., 1982). By 2001, it can be seen
that this hope has been surpassed. Single particle analysis will help us to realise the ultimate aim of
structural biology}the atomic resolution description of the macromolecular complexes of the cell
in their different functional states.

Acknowledgements

We would like to thank Professor Sir Tom Blundell (Department of Biochemistry, University of
Cambridge), Dr. Venki Ramakrishnan (MRC Laboratory of Molecular Biology, Cambridge) and
Dr. Gebhard Schertler (MRC Laboratory of Molecular Biology, Cambridge) for their critical
comments on the manuscript. Any errors or misinterpretations remain the sole responsibility of
the authors.

References

Agrawal, R.K., Penczek, P., Grassucci, R.A., Li, Y., Leith, A., Nierhaus, K.H., Frank, J., 1996. Direct visualisation of
A-, P- and E-site transfer RNAs in the Escherichia coli ribosome. Science 271, 1000–1002.
Agrawal, R.K., Penczek, P., Grassucci, R.A., Frank, J., 1998. Visualisation of elongation factor G on the Escherichia
coli 70S ribosome: the mechanism of translocation. Proc. Natl. Acad. Sci. U.S.A. 95, 6134–6138.
Agrawal, R.K., Heagle, A.B., Penczek, P., Grassucci, R.A., Frank, J., 1999. EF-G-dependent GTP hydrolysis induces
translocation accompanied by large conformational changes in the 70S ribosome. Nat. Struct. Biol. 6, 643–649.
Amos, L.A., Henderson, R., Unwin, P.N.T., 1982. Three-dimensional structure determination by electron microscopy
of two-dimensional crystals. Prog. Biophys. Molec. Biol. 39, 183–231.
Avila-Sakar, A.J., Chiu, W., 1996. Visualization of b-sheets and side-chain clusters in two-dimensional periodic arrays
of streptavidin on phospholipid monolayers by electron crystallography. Biophys. J. 70, 57–68.
Baker, T.S., Johnson, J.E., 1996. Low resolution meets high: towards a continuum from cells to atoms. Curr. Opin.
Struct. Biol. 6, 585–594.
158 J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164

Baker, T.S., Johnson, J.E., 1997. Principles of virus structure determination. In: Chiu, Burnett, Garcia (Eds.) Structural
Biology of Viruses. Oxford University Press, Oxford, pp. 38–55.
Ban, N., Freeborn, B., Nissen, P., Penczek, P., Grassucci, R.A., Sweet, R., Frank, J., Moore, P.B., Steitz, T.A., 1998. A
9A( resolution X-ray crystallographic map of the large ribosomal subunit. Cell 93, 1105–1115.
Ban, N., Nissen, P., Hansen, J., Capel, M., Moore, P.B., Steitz, T.A., 1999. Placement of protein and RNA structures
(
into a 5 A-resolution map of the 50S ribosomal subunit. Nature 400, 841–847.
Ban, N., Nissen, P., Hansen, J., Moore, P.B., Steitz, T., 2000. The complete atomic structure of the large ribosomal
subunit at 2.4 A( resolution. Science 289, 905–920.
Barber, J., 1998. Photosystem two. Biochim. Biophys. Acta 1365, 269–277.
.
Barber, J., Kuhlbrandt, W., 1999. Photosystem II. Curr. Opin. Struct. Biol. 9, 469–475.
Barber, J., Nield, J., Morris, E.P., Hankamer, B., 1999. Subunit positioning in photosystem II revisited. Trends
Biochem. Sci. 24, 43–45.
Barber, J., Nield, J., Morris, E.P., Zheleva, D., Hankamer, B., 1997. The structure, function and dynamics of
photosystem two. Physiol. Plant. 100, 817–827.
Baumeister, W., Grimm, R., Walz, J., 1999. Electron tomography of molecules and cells. Trends Cell Biol. 9, 81–85.
Berriman, J., Unwin, N., 1994. Analysis of transient structures by cryo-microscopy combined with rapid mixing of
spray droplets. Ultramicroscopy 56, 241.
Boekema, E.J., Hankamer, B., Bald, D., Kruip, J., Nield, J., Boonstra, A.F., Barber, J., Rogner, . M., 1995.
Supramolecular structure of the photosystem II complex from green plants and cyanobacteria. Proc. Natl. Acad.
Sci. U.S.A. 92, 175–179.
Boekema, E.J., van Breeman, J.F.L., van Roon, H., Dekker, J.P., 2000a. Conformational changes in photosystem II
supercomplexes upon removal of extrinsic subunits. Biochemistry 39, 12907–12915.
Boekema, E.J., van Breemen, J.F.L., van Roon, H., Dekker, J.P., 2000b. Arrangement of photosystem II
supercomplexes in crystalline macrodomains within the thylakoid membrane of green plant chloroplasts. J. Mol.
Biol. 301, 1123–1133.
Boekema, E.J., Nield, J., Hankamer, B., Barber, J., 1998a. Localization of the 23-kDa subunit of the oxygen-evolving
complex of photosystem II by electron microscopy. Eur. J. Biochem. 252, 268–276.
Boekema, E.J., van Roon, H., van Breeman, J.F.L., Dekker, J.P., 1999a. Supramolecular organisation of photosystem
II and its light-harvesting antenna in partially solubilized photosystem II membranes. Eur. J. Biochem. 266,
444–452.
Boekema, E.J., van Roon, H., Calkoen, F., Bassi, R., Dekker, J.P., 1999b. Multiple types of association of photosystem
II and its light-harvesting antenna in partially solubilized photosystem II membranes. Biochemistry 38, 2233–2239.
Boekema, E.J., van Roon, H., Dekker, J.P., 1998b. Specific association of photosystem II and light-harvesting complex
II in partially solubilized photosystem II membranes. FEBS Lett. 424, 95–99.
Boisset, N., Penczek, P.A., Taveau, J.-C., You, V., de Haas, F., Lamy, J., 1998. Overabundant single-particle electron
microscope views induce a three-dimensional reconstruction artifact. Ultramicroscopy 74, 201–207.
Borland, L., van Heel, M., 1990. Classification of image data in conjugate representation spaces. J. Opt. Soc. Am. A. 7,
601–610.
.
Bottcher, C., Ludwig, K., Herrmann, A., van Heel, M., Stark, H., 1999. Structure of influenza haemagglutinin at
neutral and at fusogenic pH by electron cryo-microscopy. FEBS Lett. 463, 255–259.
.
Bottcher, B., Wynne, S.A., Crowther, R.A., 1997. Determination of the fold of the core protein of hepatitis B virus by
electron cryomicroscopy. Nature 386, 88–91.
Brimacombe, R., 2000. The bacterial ribosome at atomic resolution. Structure 8, R195–R200.
Brink, J., Sherman, M.B., Berriman, J., Chiu, W., 1998. Evaluation of charging on macromolecules in electron
cryomicroscopy. Ultramicroscopy 72, 41–52.
Carazo, J.M., Stelzer, E.H.K., 1999. The BioImage Database Project: Organizing multidimensional biological images in
an object-relational database. J. Struct. Biol. 125, 97–102.
Carragher, B., Smith, P.R., 1996. Advances in computational image processing for microscopy. J. Struct. Biol. 116, 2–8.
Carter, A.P., Clemons, W.M., Brodersen, D.E., Morgan-Warren, R.J., Wimberly, B.T., Ramakrishnan, V., 2000.
Functional insights from the structure of the 30S ribosomal subunit and its interactions with antibiotics. Nature 407,
340–348.
J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164 159

Cate, J.H., Yusupov, M.M., Yusupova, G.Z., Earnest, T.N., Noller, H.F., 1999. X-ray crystal structures of 70S
ribosome functional complexes. Science 285, 2095–2104.
Cech, T.R., 2000. The ribosome is a ribozyme. Science 289, 878–879.
Chen, S., Roseman, A.M., Hunter, A.S., Wood, S.P., Burston, S.G., Ranson, N.A., Clarke, A.R., Saibil, H.R., 1994.
Location of a folding protein and shape changes in GroEL-GroES complexes imaged by cryo-electron microscopy.
Nature 371, 261–264.
Chen, S., Roseman, A.M., Saibil, H.R., 1998. Electron Microscopy of Chaperonins. In: Lorimer, G.H., Baldwin, T.O.
(Eds.), Methods in Enzymology Vol. 290: Molecular Chaperones. Academic Press, San Diego, pp. 242–253.
Chiu, W., McGough, A., Sherman, M.B., Schmid, M.F., 1999. High-resolution electron cryomicroscopy of
macromolecular assemblies. Trends Cell Biol. 9, 154–159.
Clemons Jr., W.M., May, J.L.C., Wimberly, B.T., McCutcheon, J.P., Capel, M.S., Ramakrishnan, V., 1999. Structure
of a bacterial 30S ribosomal subunit at 5.5 A( resolution. Nature 400, 833–840.
Conway, J.F., Steven, A.C., 1999. Methods for reconstructing density maps of ‘‘single’’ particles from cryoelectron
micrographs to subnanometer resolution. J. Struct. Biol. 128, 106–118.
Crowther, R.A., 1971. Procedures for three-dimensional reconstruction of spherical viruses by Fourier synthesis from
electron micrographs. Phil. Trans. Roy. Soc. Lond. B 261, 221–230.
Crowther, R.A., Henderson, R., Smith, J.M., 1996. MRC image processing programs. J. Struct. Biol. 116, 9–16.
Darst, S.A., Ahlers, M., Meller, P.H., Kubalek, E.W., Blankenburg, R., Ribi, H.O., Ringsdorf, H., Kornberg, R.D.,
1991. Two-dimensional crystals of streptavidin on biotinylated lipid layers and their interactions with biotinylated
macromolecules. Biophys. J. 59, 387–396.
Davies, C., White, S.W., 2000. Electrons and X-rays gang up on the ribosome. Structure 8, R41–R45.
Deng, Y., Marko, M., Buttle, K.F., Leith, A., Mieczkowski, M., Mannella, C.A., 1999. Cubic membrane structure
in amoeba (Chaos carolinensis) mitochondria determined by electron microscopic tomography. J. Struct. Biol. 127,
231–239.
DeRosier, D.J., Klug, A., 1968. Reconstruction of three dimensional structures from electron micrographs. Nature 217,
130–134.
Dube, P., Bacher, G., Stark, H., Mueller, F., Zemlin, F., van Heel, M., Brimacombe, R., 1998a. Correlation of the
expansion segments in mammalian rRNA with the fine structure of the 80S ribosome; a cryo-electron microscopic
reconstruction of the rabbit reticulocyte ribosome at 21 A( resolution. J. Mol. Biol. 279, 403–421.
Dube, P., Tavares, P., Lurz, R., van Heel, M., 1993. The portal protein of bacteriophage SPP1: a DNA pump with 13-
fold symmetry. EMBO J. 12, 1303–1309.
Dube, P., Wieske, M., Stark, H., Schatz, M., Stahl, J., Zemlin, F., Lutsch, G., van Heel, M., 1998b. The 80S rat liver
ribosome at 25 A ( resolution by electron cryomicroscopy and angular reconstitution. Structure 6, 389–399.
Dubochet, J., Adrian, M., Chang, J.J, Homo, J.C., Lepault, J., McDowall, A.W., Schultz, P., 1988. Cryo-electron
microscopy of vitrified specimens. Q. Rev. Biophys. 21, 129–228.
Dubochet, J., Lepault, J., Freeman, R., Berriman, J.A., Homo, J.C., 1982. Electron microscopy of frozen water and
aqueous solutions. J. Microsc. 128, 219–237.
Ealick, S.E., 2000. Advances in multiple wavelength anomalous diffraction crystallography. Curr. Opin. Chem. Biol. 4,
495–499.
Edman, K., Nollert, P., Royant, A., Belrhali, H., Pebay-Peyroula, E., Hajdu, J., Neutze, R., Landau, E.M., 1999. High-
resolution X-ray structure of an early intermediate in the bacteriorhodopsin photocycle. Nature 401, 822–826.
Erickson, H.P., Klug, A., 1971. Measurement and compensation of defocusing and aberrations by Fourier processing
of electron micrographs. Phil. Trans. Roy. Soc. Lond. B 261, 105–118.
Faruqi, A.R., Subramanian, S., 2000. CCD detectors in high-resolution biological electron microscopy. Q. Rev.
Biophys. 33, 1–27.
Frank, J., 1980. The role of correlation techniques in computer image processing. In: Hawkes, P.W. (Ed.), Computer
Processing of Electron Microscope Images. Springer, Berlin, pp. 187–222.
Frank, J., 1984. The role of multivariate image analysis in solving the architecture of the Limulus polyphemus
hemocyanin molecule. Ultramicroscopy 13, 153–164.
Frank, J., 1990. Classification of macromolecular assemblies studied as ‘single particles’. Q. Rev. Biophys. 23, 281–329.
Frank, J., 1996. Three-Dimensional Electron Microscopy of Macromolecular Assemblies. Academic Press, San Diego.
160 J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164

Frank, J., Agrawal, R.K., 2000. A ratchet-like inter-subunit reorganization of the ribosome during translocation.
Nature 406, 318–322.
Frank, J., van Heel, M., 1982. Correspondence analysis of aligned images of biological particles. J. Mol. Biol. 161,
134–137.
Frank, J., Radermacher, M., 1992. Three-dimensional reconstruction of single particles negatively stained or in vitreous
ice. Ultramicroscopy 46, 241–262.
Frank, J., Radermacher, M., Penczek, P., Zhu, J., Li, Y., Ladjadj, M., Leith, A., 1996. SPIDER and WEB: processing
and visualization of images in 3D electron microscopy and related fields. J. Struct. Biol. 116, 190–199.
Frank, J., Verschoor, A., Boublik, M., 1981. Computer averaging of electron micrographs of 40S ribosomal subunits.
Science 214, 1353–1355.
Frank, J., Verschoor, A., Boublik, M., 1982. Multivariate statistical analysis of ribosome electron micrographs. J. Mol.
Biol. 161, 107–133.
Frank, J., Verschoor, A., Wagenknecht, T., Radermacher, M., Carazo, J.-M., 1988. A new non-crystallographic image-
processing technique reveals the architecture of ribosomes. Trends Bioch. Sci. 13, 123–127.
Frank, J., Zhu, J., Penczek, P., Li, Y., Srivastava, S., Verschoor, A., Radermacher, M., Grassucci, R., Lata, R.K.,
Agrawal, R.J., 1995. A model of protein synthesis based on cryo-electron microscopy of the E. coli ribosome. Nature
376, 441–444.
Frey, T.G., Mannella, C.A., 2000. The internal structure of mitochondria. Trends Bioch. Sci. 25, 319–324.
Fujiyoshi, Y., 1998. The structural study of membrane proteins by electron crystallography. Adv. Biophys. 35, 25–80.
Fujiyoshi, Y., Mizusaki, T., Morikawa, K., Yamagishi, H., Aoki, Y., Kihara, H., Harada, Y., 1991. Development of a
superfluid helium stage for high-resolution electron microscopy. Ultramicroscopy 38, 241–251.
Fuller, S.D., Butcher, S.J., Cheng, R.H., Baker, T.S., 1996. Three-dimensional reconstruction of icosahedral
particles}the uncommon line. J. Struct. Biol. 116, 48–55.
Gabashvili, I.S., Agrawal, R.K., Grassucci, R., Frank, J., 1999. Structure and structural variations of the Escherichia
coli 30S ribosomal subunit as revealed by three-dimensional cryo-electron microscopy. J. Mol. Biol. 286, 1285–1291.
Gabashvili, I.S., Agrawal, R.K., Spahn, C.M.T., Grassucci, R.A., Svergun, D.I., Frank, J., Penczek, P., 2000. Solution
structure of the E. coli 70S ribosome at 11.5 A( resolution. Cell 100, 537–549.
Glaeser, R.M., 1999. Review: electron crystallography: present excitement, a nod to the past, anticipating the future.
J. Struct. Biol. 128, 3–14.
Green, R., Puglisi, J.D., 1999. The ribosome revealed. Nat. Struct. Biol. 6, 999–1003.
Grigorieff, N., 1998. Three-dimensional structure of bovine NADH: ubiquinone oxidoreductase (complex I) at 22 A ( in
ice. J. Mol. Biol. 277, 1033–1046.
Grigorieff, N., 2000. Resolution measurement in structures derived from single particles. Acta Cryst. D 56, 1270–1277.
Grigorieff, N., Ceska, T.A., Downing, K.H., Baldwin, J.M., Henderson, R., 1996. Electron-crystallographic refinement
of the structure of bacteriorhodopsin. J. Mol. Biol. 259, 393–421.
Hankamer, B., Morris, E.P., Barber, J., 1999. Revealing the structure of the oxygen-evolving core dimer of photosystem
II by cryoelectron crystallography. Nat. Struct. Biol. 6, 560–564.
Hankamer, B., Nield, J., Zheleva, D., Boekema, E., Jansson, S., Barber, J., 1997. Isolation and biochemical
characterisation of monomeric and dimeric photosystem II complexes from spinach and their relevance to the
organisation of photosystem II in vivo. Eur. J. Biochem. 243, 422–429.
Harauz, G., Boekema, E., van Heel, M., 1988. Statistical image analysis of electron micrographs of ribosomal subunits.
In: Noller, H.F., Moldave, K. (Eds.), Methods in Enzymology Vol. 164: Ribosomes. Academic Press, San Diego,
pp. 35–49.
Harauz, G., van Heel, M., 1986. Exact filters for general geometry three dimensional reconstruction. Optik 73, 146–156.
.
Harms, J., Tocilj, A., Levin, I., Agmon, I., Stark, H., Kolln, .
I., van Heel, M., Cuff, M., Schlunzen, F., Bashan, A.,
Franceschi, F., Yonath, A., 1999. Elucidating the medium-resolution structure of ribosomal particles: an interplay
between electron cryo-microscopy and X-ray crystallography. Structure 7, 931–941.
Henderson, R., 1995. The potential and limitations of neutrons, electrons, and X-rays for atomic resolution microscopy
of unstained biological molecules. Q. Rev. Biophys. 28, 171–193.
Henderson, R., Baldwin, J.M., Ceska, T.A., Zemlin, F., Beckmann, E., Downing, K.H., 1990. Model for the structure
of bacteriorhodopsin based on high-resolution electron cryo-microscopy. J. Mol. Biol. 213, 899–929.
J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164 161

Hendrickson, W.A., 1991. Determination of macromolecular structures from anomalous diffraction of synchrotron
radiation. Science 254, 51–58.
Hoenger, A., Aebi, U., 1996. 3-D reconstructions from ice-embedded and negatively stained biomacromolecular
assemblies: a critical comparison. J. Struct. Biol. 117, 99–116.
Kasper, E., 1982. Field electron emission systems. In: Barer, R., Cosslett, V.E. (Eds.), Advances in Optical and Electron
Microscopy, Vol. 8. Academic, London.
Kiselev, N.A., Sherman, M.B., Tsuprun, V.L., 1990. Negative staining of proteins. Electron Microsc. Rev. 3, 43–72.
Kornberg, R.D., Darst, S.A., 1991. Two-dimensional crystals of proteins on lipid layers. Curr. Opin. Struct. Biol. 1,
642–646.
.
Kuhl, H., Rogner, M., van Breeman, J.F.L., Boekema, E.J., 1999. Localization of cyanobacterial photosystem II
donor-side subunits by electron microscopy and the supramolecular organization of photosystem II in the thylakoid
membrane. Eur. J. Biochem. 266, 453–459.
.
Kuhlbrandt, W., 2000. Bacteriorhodopsin}the movie. Nature 406, 569–570.
.
Kuhlbrandt, W., Wang, D.N., Fujiyoshi, Y., 1994. Atomic model of plant light-harvesting complex by electron
crystallography. Nature 367, 614–621.
.
Kuhlbrandt, W., Williams, K.A., 1999. Analysis of macromolecular structure and dynamics by electron cryo-
microscopy. Curr. Opin. Chem. Biol. 3, 537–543.
Lake, J.A., 1976. Ribosome structure determined by electron microscopy of Escherichia coli small subunits, large
subunits and monomeric ribosomes. J. Mol. Biol. 105, 131–159.
Lebart, L., Morineau, A., Warwick, K.M., 1984. Multivariate Descriptive Statistical Analysis. Wiley, New York,
pp. 30–62.
Levy, D., Mosser, G., Lambert, O., Moeck, G.S., Bald, D., Rigaud, J.-L., 1999. Two-dimensional crystallisation on
lipid layer: a successful approach for membrane proteins. J. Struct. Biol. 127, 44–52.
Lewin, B., 2000. Genes VII. Oxford University Press, Oxford, p. 140.
Ludtke, S.J., Baldwin, P.R., Chiu, W., 1999. EMAN: semiautomated software for high-resolution single-particle
reconstructions. J. Struct. Biol. 128, 82–97.
.
Malhotra, A., Penczek, P., Agrawal, R.K., Gabashvili, I.S., Grassucci, R.A., Junemann, R., Burkhardt, N., Nierhaus,
K.H., Frank, J., 1998. Escherichia coli 70S ribosome at 15 A( resolution by cryo-electron microscopy: localisation of
fMet-tRNAMet f and fitting of L1 protein. J. Mol. Biol. 280, 103–116.
Mancini, E.J., Fuller, S.D., 2000. Supplanting crystallography or supplementing microscopy? A combined approach to
the study of an enveloped virus. Acta Cryst. D 56, 1278–1287.
Mannella, C.A., Marko, M., Buttle, K., 1997. Reconsidering mitochondrial structure: new views of an old organelle.
Trends Bioc. Sci. 22, 37–38.
Marabini, R., Vaquerizo, C., Fern!andez, J.J., Carazo, J.M., Engel, A., Frank, J., 1996. Proposal for a new distributed
database of macromolecular and subcellular structures from different areas of microscopy. J. Struct. Biol. 116,
161–166.
Matadeen, R., Patwardan, A., Gowen, B., Orlova, E.V., Pape, T., Cuff, M., Mueller, F., Brimacombe, R., van Heel,
M., 1999. The Escherichia coli large ribosomal subunit at 7.5 A ( resolution. Structure 7, 1575–1583.
McEwen, B.F., Radermacher, M., Rieder, C.L., Frank, J., 1986. Tomographic three-dimensional reconstruction of cilia
ultrastructure from thick sections. Proc. Natl. Acad. Sci. U.S.A. 83, 9040–9044.
Meissner, U., Dube, P., Harris, J.R., Stark, H., Markl, J., 2000. Structure of a Molluscan Hemocyanin Didecamer
(HtH1 from Haliotis tuberculata) at 12A ( resolution by cryoelectron microscopy. J. Mol. Biol. 298, 21–34.
Mellema, J.E., 1980. Computer reconstruction of regular biological objects. In: Hawkes, P.W. (Ed.), Computer
Processing of Electron Microscope Images. Springer, Berlin, pp. 89–126.
Mitsuoka, K., Murata, K., Kimura, Y., Namba, K., Fujiyoshi, Y., 1997. Examination of the LeafScan 45, a line-
illuminating micro-densitometer, for its use in electron crystallography. Ultramicroscopy 68, 109–121.
Murali, R., Burnett, R.M., 1991. X-ray crystallography of very large unit cells. Curr. Opin. Struct. Biol. 1, 997–1001.
Murata, K., Mitsuoka, K., Hirai, T., Walz, T., Agre, P., Heymann, J.B., Engel, A., Fujiyoshi, Y., 2000. Structural
determinants of water permeation through aquaporin}1. Nature 407, 599–605.
Musacchio, A., Smith, C.J., Roseman, A.M., Harrison, S.C., Kirchhausen, T., Pearse, B.M.F., 1999. Functional
organization of clathrin in coats: combining electron cryomicroscopy and X-ray crystallography. Cell 3, 761–770.
162 J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164

Nicastro, D., Frangakis, A.S., Typke, D., Baumeister, W., 2000. Cryo-electron tomography of Neurospora
mitochondria. J. Struct. Biol. 129, 48–56.
Nield, J., Funk, C., Barber, J., 2000a. Supermolecular structure of photosystem II and location of the PsbS protein.
Phil. Trans. Roy. Soc. Lond. B 355, 1337–1344.
.
Nield, J., Kruse, O., Ruprecht, J., da Fonseca, P., Buchel, C., Barber, J., 2000b. Three-dimensional structure of
Chlamydomonas reinhardtii and Synechococcus elongatus photosystem II complexes allows for comparison of their
oxygen-evolving complex organisation. J. Biol. Chem. 275, 27940–27946.
Nield, J., Orlova, E.V., Morris, E.P., Gowen, B., van Heel, M., Barber, J., 2000c. 3D map of the plant photosystem II
supercomplex obtained by cryoelectron microscopy and single particle analysis. Nat. Struct. Biol. 7, 44–47.
Nissen, P., Hansen, J., Ban, N., Moore, P.B., Steitz, T.A., 2000. The structural basis of ribosome activity in peptide
bond synthesis. Science 289, 920–930.
Nogales, E., Wolf, S.G., Downing, K.H., 1998. Stucture of the ab tubulin dimer by electron crystallography. Nature
391, 199–302 and Nature 393, 191.
Orlova, E.V., 2000. Structural analysis of non-crystalline macromolecules: the ribosome. Acta Cryst. D 56, 1253–1258.
Orlova, E.V., Atiqur Rahman, M., Gowen, B., Volynski, K.E., Ashton, A.C., Manser, C., van Heel, M., Ushkaryov,
Y.A., 2000. Structure of a-latrotoxin oligomers reveals that divalent cation-dependent tetramers form membrane
pores. Nat. Struct. Biol. 7, 48–53.
Orlova, E.V., Dube, P., Robin Harris, J., Beckman, E., Zemlin, F., Markl, J., van Heel, M., 1997. Structure of keyhole
limpet hemocyanin type I (KLH1) at 15 A ( resolution by electron cryomicroscopy and angular reconstitution. J. Mol.
Biol. 271, 417–437.
Orlova, E.V., Sherman, M.B., Chiu, W., Mowri, H., Smith, L.C., Gotto, A.M., 1999. Three-dimensional structure of
low density lipoproteins by electron cryomicroscopy. Proc. Natl. Acad. Sci. U.S.A. 96, 8420–8425.
Ostermeier, C., Michel, H., 1997. Crystallization of membrane proteins. Curr. Opin. Struct. Biol. 7, 697–701.
Penczek, P., Grassucci, R.A., Frank, J., 1994. The ribosome at improved resolution: new techniques for merging and
orientation refinements in 3D cryo-electron microscopy of biological particles. Ultramicroscopy 53, 251–270.
Perkins, G., Renken, C., Martone, M.E., Young, S.J., Ellisman, M., Frey, T., 1997a. Electron tomography of neuronal
mitochondria: three-dimensional structure and organization of cristae and membrane contacts. J. Struct. Biol. 119,
260–272.
Perkins, G.A., Renken, C.W., Song, J.Y., Frey, T.G., Young, S.J., Lamont, S., Martone, M.E., Lindsey, S., Ellisman,
M.H., 1997b. Electron tomography of large, multicomponent biological structures. J. Struct. Biol. 120, 219–227.
Puglisi, J.D., Blanchard, S.C., Green, R., 2000. Approaching translation at atomic resolution. Nat. Struct. Biol. 7,
855–861.
Radermacher, M., 1988. Three-dimensional reconstruction of single particles from random and nonrandom tilt series.
J. Elect. Microsc. Tech. 9, 359–394.
Rath, B.K., Marko, M., Radermacher, M., Frank, J., 1997. Low-dose automated electron tomography: a recent
implementation. J. Struct. Biol. 120, 210–218.
Reimer, L., 1997. Transmission Electron Microscopy, 4th edition. Springer, Berlin.
.
Rhee, K.-H., Morris, E.P., Barber, J., Kuhlbrandt, W., 1998. Three-dimensional structure of the plant photosystem II
reaction centre at 8 A( resolution. Nature 396, 283–286.
Rhodes, G., 2000. Crystallography Made Crystal Clear, 2nd Edition. Academic Press, San Diego, pp. 29–44.
Roseman, A.M., 2000. Docking structures of domains into maps from cryo-electron microscopy using local correlation.
Acta Cryst. D 56, 1332–1340.
Roseman, A.M., Chen, S., White, H., Braig, K., Saibil, H.R., 1996. The chaperonin ATPase cycle: mechanism of
allosteric switching and movements of substrate-binding domains in GroEL. Cell 97, 241–251.
Rossmann, M.G., 2000. Fitting atomic models into electron-microscopy maps. Acta Cryst. D 56, 1341–1349.
Royant, A., Edman, K., Ursby, T., Pebay-Peyroula, E., Landau, E.M., Neutze, R., 2000. Helix deformation is coupled
to vectorial proton transport in the photocycle of bacteriorhodopsin. Nature 406, 645–648.
Rummel, G., Hardmeyer, A., Widmer, C., Chiu, M.L., Nollert, P., Locher, K.P., Pedruzzi, I., Landau, E.M.,
Rosenbusch, J.P., 1998. Lipidic cubic phases: new matrices for the three-dimensional crystallisation of membrane
proteins. J. Struct. Biol. 121, 82–91.
Saibil, H.R., 2000a. Conformational changes studied by cryo-electron microscopy. Nat. Struct. Biol. 7, 711–714.
J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164 163

Saibil, H.R., 2000b. Macromolecular structure determination by cryo-electron microscopy. Acta Cryst. D 56,
1215–1222.
Saibil, H., 2000c. The black widow’s versatile venom. Nat. Struct. Biol. 7, 3–4.
.
Sass, H.J., Buldt, G., Gessenich, R., Hehn, D., Neff, D., Schlesinger, R., Berendzen, J., Ormos, P., 2000. Structural
alterations for proton translocation in the M state of wild-type bacteriorhodopsin. Nature 406, 649–653.
(
Schatz, M., Orlova, E.V., Dube, P., J.ager, J., van Heel, M., 1995. Structure of Lumbricus terrestris hemoglobin at 30 A
resolution determined using angular reconstitution. J. Struct. Biol. 114, 28–40.
Schoehn, G., Quaite-Randall, E., Jim!enez, J.L., Joachimiak, A., Saibil, H.R., 2000. Three conformations of an archael
chaperonin, TF55 from Sulfolobus shibatae. J. Mol. Biol. 296, 813–819.
Seibert, M., DeWit, M., Staehelin, L.A., 1987. Structural localization of the O2-evolving apparatus to multimeric
(tetrameric) particles on the lumenal surface of freeze-etched photosynthetic membranes. J. Cell Biol. 105, 2257–
2265.
Serysheva, I.I., Orlova, E.V., Chiu, W., Sherman, M.B., Hamilton, S.L., van Heel, M., 1995. Electron cryomicroscopy
and angular reconstitution used to visulize the skeletal muscle calcium release channel. Nat. Struct. Biol. 2, 18–24.
Shannon, C.E., 1949. Communication in the presence of noise. Porc. IRE 37, 10–22.
Sheehan, B., Fuller, S.D., Pique, M.E., Yeager, M., 1996. AVS software for visualisation in molecular microscopy.
J. Struct. Biol. 116, 99–106.
Sherman, M.B., Soejima, T., Chiu, W., van Heel, M., 1998. Multivariate analysis of single unit cells in electron
crystallography. Ultramicroscopy 74, 179–199.
Shevack, A., Gewitz, H.S., Hennemann, B., Yonath, A., Wittman, H.G., 1985. Characterisation and crystallisation of
ribosomal particles from Haloarcula marismortui. FEBS Lett. 184, 68–71.
.
Skoglund, U., Ofverstedt, L.-G., Burnett, R.M., Bricogne, G., 1996. Maximum-entropy three-dimensional
reconstruction with deconvolution of the contrast transfer function: a test application with adenovirus. J. Struct.
Biol. 117, 173–188.
Slayter, E.M., Slayter, H.S., 1992. Light and Electron Microscopy. Cambridge University Press, U.K.
Smith, J.L., 1991. Determination of three-dimensional structure by multiwavelength anomalous diffraction. Curr. Opin.
Struct. Biol. 1, 1002–1011.
Smith, J.M., 1999. XIMDISP}a visualization tool to aid structure determination from electron microscope images.
J. Struct. Biol. 125, 223–228.
Stark, H., Mueller, F., Orlova, E.V., Schatz, M., Dube, P., Erdemir, T., Zemlin, F., Brimacombe, R., van Heel, M.,
1995. The 70S Escherichia coli ribosome at 23 A ( resolution: fitting the ribosomal RNA. Struct. 3, 815–821.
.
Stark, H., Orlova, E.V., Rinke-Appel, J., Junke, N., Mueller, F., Rodnina, M., Wintermeyer, W., Brimacombe, R., van
Heel, M., 1997a. Arrangement of tRNAs in pre- and post-translocational ribosomes revealed by electron
cryomicroscopy. Cell 88, 19–28.
Stark, H., Rodnina, M.V., Rinke-Appel, J., Brimacombe, R., Wintermeyer, W., van Heel, M., 1997b. Visualisation of
elongation factor Tu on the Escherichia coli ribosome. Nature 389, 403–406.
Stark, H., Rodnina, M.V., Wieden, H.-J., van Heel, M., Wintermeyer, W., 2000. Large-scale movement of elongaton
factor G and extensive conformational change of the ribosome during translocation. Cell 100, 301–309.
Stark, H., Zemlin, F., Boettcher, C., 1996. Electron radiation damage to protein crystals of bacteriorhodopsin at
different temperatures. Ultramicroscopy 63, 75–79.
Stec, B., Zhou, R., Teeter, M.M., 1995. Full-matrix refinement of the protein crambin at 0.83 A ( and 130 K. Acta Cryst.
D 51, 663–681.
Stewart, P.L., Chiu, C.Y., Haley, D.A., Kong, L.B., Schlessman, J.L., 1999. Review: resolution issues in single-particle
reconstruction. J. Struct. Biol. 128, 58–64.
Stoddard, B.L., 1998. New results using Laue diffraction and time-resolved crystallography. Curr. Opin. Struct. Biol. 8,
612–618.
Stowell, M.H.B., Miyazawa, A., Unwin, N., 1998. Macromolecular structure determination by electron microscopy:
new advances and recent results. Curr. Opin. Struct. Biol. 8, 595–600.
Subramanian, S., Henderson, R., 2000. Molecular mechanism of vectorial proton translocation by bacteriorhodopsin.
Nature 406, 653–657.
164 J. Ruprecht, J. Nield / Progress in Biophysics & Molecular Biology 75 (2001) 121–164

Thon, F., 1966. Zur Defokussierung sabh.angigkeit des Phasenkontrastes bei der elektronenmikroskopischen
Abbildung. Z. Naturforsch. 21a, 476–478.
Toyoshima, C., Unwin, N., 1988. Contrast transfer for frozen-hydrated specimens: determination from pairs of
defocused images. Ultramicroscopy 25, 279–292.
Toyoshima, C., Yonekura, K., Sasabe, H., 1993. Contrast transfer for frozen-hydrated specimens II. Amplitude
contrast at very low frequencies. Ultramicroscopy 48, 165–176.
Unger, V.M., Schertler, G.F.X., 1995. Low resolution structure of bovine rhodopsin determined by electron cryo-
microscopy. Biophys. J. 68, 1776–1786.
Unwin, N., 1995. Acetylcholine receptor channel imaged in the open state. Nature 373, 37–43.
van Heel, M., 1987a. Angular reconstitution: a posteriori assignment of projection directions for 3D reconstruction.
Ultramicroscopy 21, 111–124.
van Heel, M., 1987b. Similarity measures between images. Ultramicroscopy 21, 95–100.
van Heel, M., 2000. Unveiling ribosomal structures: the final phases. Curr. Opin. Struct. Biol. 10, 259–264.
van Heel, M., Frank, J., 1981. Use of multivariate statistics in analysing the images of biological macromolecules.
Ultramicroscopy 6, 187–194.
van Heel, M., Harauz, G., 1986. Resolution criteria for three dimensional reconstruction. Optik 73, 119–122.
van Heel, M., Harauz, G., Orlova, E.V., Schmidt, R., Schatz, M., 1996. A new generation of the IMAGIC image
processing system. J. Struct. Biol. 116, 17–24.
van Heel, M., Schatz, M., Orlova, E., 1992a. Correlation functions revisited. Ultramicroscopy 46, 307–316.
.
van Heel, M., Stoffler-Meilicke, M., 1985. Characteristic views of E. coli and B. stearothermophilus 30S ribosomal
subunits in the electron microscope. EMBO. J. 4, 2389–2395.
van Heel, M., Winkler, H., Orlova, E., Schatz, M., 1992b. Structure analysis of ice-embedded single particles. Scanning
Microsc. (Suppl.) 6, 23–42.
van Lint, J.H., 1982. Introduction to Coding Theory. Springer, New York, pp. 22–29.
Verschoor, A., Frank, J., Radermacher, M., Wagenknecht, T., Boublik, M., 1984. Three-dimensional reconstruction of
the 30s ribosomal subunit from randomly oriented particles. J. Mol. Biol. 178, 677–698.
Wade, R.H., 1992. A brief look at imaging and contrast transfer. Ultramicroscopy 46, 145–156.
Walz, T., Grigorieff, N., 1998. Electron crystallography of two-dimensional crystals of membrane proteins. J. Struct.
Biol. 121, 142–161.
White, H.D., Walker, M.L., Trinick, J., 1998. A computer-controlled spraying-freezing appratus for millisecond time-
resolution electron cryomicroscopy. J. Struct. Biol. 121, 306–313.
.
Wider, G., Wuthrich, K., 1999. NMR spectroscopy of large molecules and multimolecular assemblies in solution. Curr.
Opin. Struct. Biol. 9, 594–601.
Williams, D.B., Barry Carter, C., 1996. Transmission Electron Microscopy: A Textbook for Materials Science. Plenum
Press, New York and London, pp. 459–475.
Williamson, J.R., 2000. Small subunit, big science. Nature 407, 306–307.
Wimberly, B.T., Brodersen, D.E., Clemons, W.M., Morgan-Warren, R.J., Carter, A.P., Vonrhein, C., Hartsch, T.,
Ramakrishnan, V., 2000. Structure of the 30S ribosomal subunit. Nature 407, 327–339.
Zemlin, F., Beckmann, E., van der Mast, K.D., 1996. A 200 kV electron microscope with Schottky field emitter and a
helium-cooled superconducting objective lens. Ultramicroscopy 63, 227–238.
.
Zhu, J., Penczek, P.A., Schroder, R., Frank, J., 1997. Three-dimensional reconstruction with contrast transfer
correction from energy-filtered cryoelectron micrographs: procedure and application to the 70S Escherichia coli
ribosome. J. Struct. Biol. 118, 197–219.

Potrebbero piacerti anche