Jaap Brink and Wah Chiu
Verna and Marrs McLean Department of Biochemistry and
The W. M. Keck Center for Computational Biology, Baylor College of Medicine,
One Baylor Plaza, Houston, TX 77030, U.S.A.
ABSTRACT
A Gatan 1024 x 1024 slow-scan charge-coupled-device (CCD) camera has been interfaced to a JEOL4000EX electron cryomicroscope and explored for its usefulness in the electron crystallographic analysis of thin, glucose-embedded crystals of crotoxin complex kept at -125o C. We show that the camera allows for an on-line assessment of the crystals' crystallinity, flatness and thickness. Intensities obtained from electron diffraction patterns acquired with the camera have been statistically analyzed and were found to be consistent with theoretically expected values. A quantitative analysis of the diffraction intensity as function of the accumulated electron dose suggests the possibility of recording up to 250 diffraction patterns with 3.5 Å resolution from a single crotoxin complex crystal 128 Å thick. Tilt series of 125 electron diffraction patterns with 3.5 Å data acquired from a single crystal are shown to be practically feasible. The current study demonstrates for the first time the effectiveness of using a slow-scan CCD camera for electron diffraction data collection from thin protein crystals at near atomic resolution.
INTRODUCTION
Electron crystallography has become a realistic technique for solving the atomic resolution structure of biological macromolecules that are arranged in two-dimensional periodic arrays (Henderson et al., 1990; Jap et al., 1991; Kühlbrandt et al., 1994). Despite these successes technical improvements can be made particularly to the recording process of the structural data to enhance the data collection efficiency. The slow-scan charge-coupled device (CCD) camera as well as the imaging plate are known for their excellent recording characteristics in terms of dynamic range, linearity and subtractable background noise (Mori et al., 1988; Isoda et al., 1991; de Ruijter & Weiss, 1992; Krivanek & Mooney, 1993). The slow-scan CCD camera offers a smaller field of view (typically 1024 x 1024 pixels) than the imaging plate (3000 x 4000 pixels) (Ogura et al., 1994). But the slow-scan CCD camera has a practical advantage to provide instant access to the acquired data, which has enabled its use in automated tuning of an electron microscope as well as automated electron tomography (Downing et al., 1992; Koster et al., 1992; Koster & De Ruijter, 1992; Dierksen et al., 1993). This paper describes the application of a Gatan 1024 x 1024 slow-scan CCD camera in an electron crystallographic study of glucose-embedded crotoxin complex crystals using an intermediate voltage electron microscope. The camera is tested whether it can provide the operator with an on-line evaluation of the specimen in terms of the degree of crystallinity and flatness as well as its thickness prior to the data collection. In addition, we evaluate the practical limit of the amount of high resolution diffraction data that can be obtained with this camera from a single protein crystal.
MATERIALS AND METHODS
2.1. Installation of the CCD camera
A Gatan 1024 x 1024 slow-scan CCD camera (model 679) was attached below the camera chamber of a JEOL4000EX intermediate voltage electron cryomicroscope. The flange on the slow-scan CCD camera puts the camera off-axis with respect to the column's optical axis by 2 to 3 cm. This was done to prevent detection by the camera of a large number of X-ray photons, that originated from within the electron gun of the microscope. Additionally, our slow-scan CCD camera has two fiber optics bundles to increase the filtering efficiency of X-ray photons. To maintain sufficient sensitivity under these conditions a P43 phosphor scintillator of 20 to 30 um thickness was installed. Because the camera is mounted off-axis, images as well as electron diffraction patterns may suffer from small distortions caused by the projector lens, like spiral and pin-cushion. These distortions, however, are easily detected and corrected for in both electron diffraction patterns and images of crystalline specimens with currently established processing software (Baldwin & Henderson, 1984; Henderson et al., 1986).
2.2. Data acquisition with the CCD camera
The slow-scan CCD camera is controlled from a Macintosh IIfx computer using the DigitalMicrograph program from Gatan (Mooney et al., 1990). The microscope gun shift was used as a shutter rather than the projector shift deflector, so that the beam is blanked away from the CCD chip as well as the radiation-sensitive specimen during the CCD readout. The readout process, which occurs at 350,000 pixels/second, includes the charge transfer through the CCD chip coupled to the analog-to-digital conversion of each pixel's charge to a numeric value (in ADU). The sensitivity of the slow-scan CCD camera was determined as 1.5 ADU per primary electron using a method described by Ishizuka (1993), where a focused electron probe was used with a diameter smaller than the CCD chip under non-saturating illumination conditions. Frames acquired with the slow-scan CCD camera are temporarily put in RAM (24 Mbytes). They are stored in Gatan format on the Macintosh's local optical or magnetic disk after applying a gain-normalization, which corrects for the dark current and the pixel-to-pixel gain variation (de Ruijter & Weiss, 1992; Krivanek & Mooney, 1993). The system comes standard with a Peltier cooler, which cools the CCD chip to about -35o C, to minimize the dark current (Krivanek & Mooney, 1993). By monitoring this dark current upon cooling down, we established that thermal equilibrium was reached in about 8 hours. We measured the dark current rate at room temperature as ~50 counts/pixel/s, whereas at low temperature it was ~0.2 counts/pixel/s. The cool-down time was substantially longer than that of the standard Gatan 679 slow-scan CCD camera, because our camera has the additional fiber optics. In the event thermal equilibrium is not established, changes in the dark current as well as the orientation between the fiber optic bundles and the CCD chip will occur. Thermal equilibrium of the slow-scan CCD camera is therefore essential for attaining the highest possible accuracy of the gain normalization. The basic characteristics of our CCD camera were also determined in terms of the modulation transfer function (MTF) and the detection quantum efficiency (DQE) for the 400 keV electrons as described in Ishizuka (1993).
2.3. Specimen preparation
Crotoxin complex was purified and crystallized as described elsewhere (Jeng & Chiu, 1983). The crystals were prepared for electron microscopy by embedding them in 2% glucose on 200 mesh copper or molybdenum grids. Molybdenum grids were preferred because of their similar shrinkage characteristics compared to carbon film and protein upon cooling to low temperature (Booy & Pawley, 1993). The grid with the glucose-embedded crotoxin complex crystals was mounted in a Gatan cryo-holder (model 626) at room temperature, inserted into our JEOL4000EX electron microscope, and cooled down to a temperature between -120o C and -130o C to minimize radiation damage effects (Jeng & Chiu, 1984).
2.4. Electron microscopy and processing
The JEOL4000EX intermediate voltage electron cryomicroscope was operated at 400 kV. The various lens and deflectors settings, that were used to enable the low dose procedure, were recalled from the microscope's User Functions (Brink & Chiu, 1991). Serial port communication to facilitate other aspects of the data collection, such as stage positioning and setting of the tilt angle, was done as described in Brink et al. (1992). A 50 um condenser aperture was used together with a strongly excited first condenser lens (spot size 6). The electron dose rate was calibrated with a pico-ammeter (Brink et al., 1992). Suitable crystals were searched for in the defocused diffraction mode with a 7-8 um diameter beam at a dose rate of 0.005 e/Å2/s with the goniometer tilted to an angle between 55o and 60o . A Gatan TV-rate CCD camera (model 673 Mk. 3) was used to capture the search-mode image and display it on a TV-monitor. This wide-angle CCD camera was installed at a side port of the microscope's projection chamber. To improve the visibility of the crystals either two or four frames were averaged using the Gatan TV-rate image processing unit (model 663).
Once a suitable crystal was identified, the beam was deflected away from the specimen grid, and the appropriate lens and deflector settings for recording the electron diffraction pattern were recalled. Electron diffraction patterns were acquired from crystal areas 2-3 um in diameter at an electron flux of 0.04 e/Å2/s per pattern using the full area of the CCD chip. Exposures ranged from 0.1 to 2 seconds; typical exposures lasted 0.25 second. A camera length of 1220 mm was chosen resulting in an edge resolution of 2.6 Å for an electron diffraction pattern acquired using the entire CCD chip. Acquisition of a single pattern would take ~15 seconds from opening the camera shutter to the display of the gain-normalized pattern on the computer screen. Acquired data, that was used in further processing, was converted to and stored in the MRC format (Cambridge, UK) on a remote magnetic disk of a Silicon Graphics power series 4D/320 file server. Electron diffraction patterns were processed as described previously (Brink et al., 1992) using modified programs from others (Baldwin & Henderson, 1984; Prasad et al., 1990).
2.5. Crystal thickness estimate
The thickness of crotoxin complex crystals was determined according to the mass thickness theory (Hall, 1951; Zeitler & Bahr, 1962). It was computed from search-mode images acquired with the slow-scan CCD camera using the intensity difference between the crystal and its adjacent carbon film. Intensities were averaged from areas measuring 32 x 32 pixels from both the crystal and the carbon film. The thickness was calibrated using multi-layered n-paraffin crystals of 115 Å thickness per layer (Dorset, 1980) and 45 Å thick purple membrane (Henderson & Unwin, 1975). The thickness estimate allowed us to select only those crystals whose thickness was measured as close as possible to 128 Å.
2.6. Quantitative assessment of measured electron diffraction intensities
The statistical definition of electron diffraction patterns of glucose-embedded crotoxin complex crystals acquired with the slow-scan CCD camera was assessed by examining the intensity distributions in these patterns. The reflections were grouped according to their integrated intensities. We computed for each group, the mean intensity and the standard deviation for the differences between the Friedel-related reflections. These values were subsequently evaluated for their inter-dependency.
The Friedel R-factor, Rf , is most commonly used to evaluate electron diffraction patterns by comparing the integrated intensities between Friedel symmetry-related reflections in a single pattern (Baldwin & Henderson, 1984). Rf depends on a number of experimental parameters including the diffracting power of the crystal and the number of atoms in the crystal. It is approximately related to the number of electrons per reflection (n), and the number of unit cells (u) according to:
We obtained a theoretical estimate for Rf of our crystal by using the values for Rf , n and u known from the light-harvesting complex 2-dimensional crystal (Wang & Kühlbrandt, 1992). This Rf would be an approximately expected value for the electron diffraction patterns acquired from our crystal using the slow-scan CCD camera.
In addition, Rf 's were computed from electron diffraction patterns acquired of several crystals each 128 Å thick obtained at exposures ranging from ~0.001 to 0.1 e/Å2 per pattern. For statistically defined electron diffraction patterns and a constant recording efficiency of the slow-scan CCD camera, Rf should follow eq. (1) as function of the dose. This analysis in conjunction with the evaluation of the statistical definition would indicate whether the detection process of the electron diffraction patterns with the CCD camera was governed by counting statistics (Blundell & Johnson, 1976) .
2.7. Radiation damage assessment
To determine the extent of radiation damage on glucose-embedded crotoxin complex crystals, series of electron diffraction patterns from a single crystal were recorded using the slow-scan CCD camera at increasing accumulated electron dose, but with a constant exposure per acquisition (0.01 e/Å2). Several series were obtained from crystals at different tilt angles. The damage was evaluated quantitatively according to Jeng and Chiu (1984) by analyzing the relative crystalline disordering factor (Bn ) and the similarity factor (Rn ) as a function of accumulated electron dose. This quantitative analysis would set an upper limit on the number of electron diffraction patterns potentially recordable from a single crystal in a tilt series.
2.8. Tilt data collection
Tilt series of electron diffraction patterns were recorded from
glucose-embedded crotoxin complex crystals kept at -125o C starting
at the highest tilt angle that was used in the search-mode, and by decreasing
the tilt angle for each subsequent pattern. This angular decrement was
0.5o from +60o to +40o as well as from
-40o to -60o , whereas 1o was used
at tilt angles between +40o and -40o . These angular
intervals ensured sampling of the lattice rods close to 1/T
Å-1 over the entire tilt range, with T the thickness of
the protein crystal, i.e. 128 Å (Prasad et al., 1990).
The tilt angle as well as the stage position were controlled from a 286 IBM-PC
compatible computer, which runs JEOL's STAGE software capable of four axis
control, namely x, y, z, and 
.
A specimen
feature, located off the optical axis but on the mechanical tilt axis, was used
to monitor and correct for crystal movement during tilt angle adjustment caused
by the backlash and/or non-eucentricity of the stage. This correction was done
manually.
RESULTS
1. Electron diffraction data acquired on the CCD camera
 |
 |
| Fig. 1. Search-mode image of a glucose-embedded crotoxin complex crystal in the untilted position acquired with the slow-scan CCD camera. The frame measures 512 x 512 pixels. Part of the crystal covers the holey carbon net, which itself is covered with a thin (~150 Å) carbon film. The crystal thickness was measured as 128 Å. The scale bar indicates 1 um. | Fig. 2. Radial background-subtracted electron diffraction patterns of a 128 Å-thick glucose-embedded crystal of crotoxin complex acquired with the slow-scan CCD camera. These patterns were acquired using the full area of the slow-scan CCD camera, i.e. 1024 x 1024 pixels. The crystal was in the untilted position (A), and tilted to 53.8o (B). The electron dose for each pattern was 0.02 e/Å2. The basic reciprocal lattice vectors (1/38.8 Å-1) are indicated in (A). Reflections in both patterns extend to 3.2 Å resolution. A vertical streak, which is caused by the undiffracted beam, can be seen running through each pattern's center. |
Figure 1 shows an image of a typical crystal of crotoxin complex
embedded in glucose and acquired in the search-mode using the slow-scan CCD
camera. This crystal was determined to be 128 Å or half a unit cell
thick. The full unit cell has been characterized as a = b = 38.8
Å, c = 256.8 Å and
= 90o (Jeng & Chiu,
1983).
Figure 2 shows examples of background-subtracted electron diffraction patterns
acquired from a crotoxin complex crystal in both untilted and tilted positions
(at 53.8o ). All of our CCD-acquired patterns display one streak right down
the center of each pattern. This streak is the result of overflow of charge
built up on a few pixels of the CCD chip around the central, undiffracted beam,
an effect known as blooming. It contains pixels with intensities above 4,000
ADU compared to pixels in a typical reflection that range from 10 to 100 ADU.
For our 12-bit camera the upper limit in detection would be 4,095.
Occasionally, the spike overlaps with one or more reflections of the crystal
lattice depending on the crystal's orientation and the electron exposure.
2. Quantitative evaluation of measured diffraction intensities
 |
| Fig. 3. Profiles of three reflections selected from an unprocessed electron diffraction pattern of glucose-embedded crotoxin complex acquired in the untilted position using the slow-scan CCD camera. The electron dose was 0.01 e/Å2. Shown are 13 x 13 pixel areas at the position of the strongest reflection (A), a medium strong reflection (B), and a weak reflection (C). All three reflections are at resolutions around 4.1 Å. Integrated intensities for these reflections were determined as 820, 210 and 110 ADU respectively. The mean intensities of the perimeter, its standard deviation, and peak-to-perimeter ratios for these reflections were 6.5, 1.2 and 10.3 for (A), 4.6, 0.9 and 3.9 for (B), 4.0, 0.9 and 3.0 for (C). |
Figure 3 shows intensity profiles of three reflections at ~4 Å resolution with different diffraction strength from an electron diffraction pattern of an untilted crystal acquired at 0.01 e/Å2. The reflections have an apparent diameter ranging from 70 to 200 um (3-8 pixels) depending on their intensity, and are separated by an average of 850 um (35 pixels). Despite the very small electron dose all reflections stand clearly above the background, which itself shows only minor intensity variations. Each reflection's integrated intensity was computed by summing the intensities over a 9 x 9-pixel area around its computed center of gravity and by subtracting the average local background. This background was derived from 4 areas each measuring 20 x 20 pixels that were chosen at the corners of the 9x9-box, which itself was centered on each reflection. The integrated intensities of the three reflections shown here are 820, 210 and 110 ADU, or 550, 150, and 75 electrons, respectively, using the sensitivity of the CCD camera of 1.5 ADU per primary electron. The strongest reflections found in electron diffraction patterns of glucose-embedded crotoxin complex crystals were observed predominantly at 4-5 Å resolution at a variety of tilt angles. After measuring the diffraction strength of the undiffracted beam as ~ 1.8 x 107 ADU, we calculated that the intensity of these strongest reflections relative to that of the undiffracted beam ranged from 1.5-5.5 x 10-5.
|
Table 1. Integrated intensity distribution observed in a 400 kV electron diffraction pattern of a 128 Å-thick glucose-embedded crotoxin complex crystal kept at -125o C and tilted to 51.5o acquired using the slow-scan CCD camera.  r refers to
the standard deviation in the difference in integrated intensities observed
for the Friedel-related reflections; n.d. is not determined. |
Range of integrated intensities | Mean integrated intensity | No. of electrons/reflection (ADU) | No. of reflections | r |
| 18 - 73 | 49 | 33 | 50 | 30.6 |
| 73 - 128 | 97 | 65 | 54 | 45.7 |
| 128 - 184 | 154 | 103 | 40 | 37.7 |
| 184 - 239 | 205 | 137 | 20 | 25.7 |
| 239 - 295 | 269 | 179 | 8 | 61.5 |
| 295 - 350 | 308 | 205 | 4 | 5.0 |
| 350 - 406 | 372 | 248 | 4 | 46.0 |
| 406 - 461 | 0 | 0 | 0 | n.d. |
| 461 - 517 | 498 | 332 | 2 | n.d. |
| 517 - 572 | 0 | 0 | 0 | n.d. |
| 572 - 628 | 628 | 419 | 2 | n.d. |
Table 1 shows a typical distribution of integrated intensities extracted from an electron diffraction pattern of a crotoxin complex crystal tilted to 51.5o . Shown are the mean integrated intensities for different intensity ranges, the equivalent number of electrons, the number of reflections in each range, and the standard deviation of the difference in the integrated intensities for the Friedel-pairs. About 75% of its reflections have an integrated intensity ranging from 20 to ~200 ADU, i.e. 13 and ~130 electrons, respectively. We found that the (mean intensity)1/2 exhibits a linear dependence with the standard deviation over all intensity ranges (Fig. 4A).
| Fig. 4. Statistics of electron diffraction patterns of glucose-embedded
crotoxin complex crystals acquired using the slow-scan CCD camera. (A) Plot of
the standard deviations of the intensity difference measured for
Friedel-related reflections, r ,
as function of the (mean
intensity)1/2. Intensities from 12 different electron diffraction
patterns were used. Each pattern was obtained at 0.01 e/Å2.
The solid line was obtained by least-squares fitting. Final accumulated
electron dose for the crystals was 4 e/Å2. (B) Dependence of
the Friedel R-factor, Rf , with the electron exposure per pattern, as
(electron dose)1/2. Data is shown that was obtained from 4 series
of electron diffraction patterns acquired on the slow-scan CCD camera at
electron exposures ranging from ~0.001 to 0.1 e/Å2.
|
|
Both the MTF and the DQE could affect the statistical definition of the patterns. The MTF was determined to fall to 0.2 at the Nyquist frequency, and the DQE was found to be essentially constant above 2 electrons per pixel. These values are similar to those obtained with a similar camera (Ishizuka, 1993). The Full-Width at Half-Maximum of the scintillator's point-spread function (which is the Fourier transform of the MTF) was measured as 30 um, which indicates that the signal of each pixel would be mixed with those of its neighboring pixels. However, the electron diffraction intensity is measured by integrating the counts accumulated in several pixels for each reflection. Consequently, its accuracy would not be significantly effected by the MTF. Most of the pixels in each of the electron diffraction patterns will receive more than 2 electrons, which is in the range where the DQE is relatively constant at 0.6-0.7. Therefore, the DQE will also not be important in determining the statistical definition of acquired electron diffraction intensities.
An estimate of Rf was made using data obtained on the light-harvesting crystals by Wang and Kühlbrandt (1992). They obtained an Rf of 22 % for their crystals, which typically contain ~120,000 unit cells (u) and where the strongest reflection contains 1,000 electrons (n). For the crotoxin complex, the crystal area used in the diffraction experiment contains ~330,000 unit cells (u) with the stronger reflections composed of 250 to 530 electrons (n) (Table 1 ). Using eq. 1 Rf for crotoxin complex crystal was estimated to range from 18 to 26 %. The intensity equivalence of Friedel symmetry-related reflections, Rf, was typically between 12 to 25 % for electron diffraction patterns acquired with the slow-scan CCD camera. When Rf was plotted as a function of (electron dose)1/2, a non-linear behavior was observed (Fig. 4B). Below a dose of 0.01 e/Å2 Rf increased significantly to values of 40 to 60 %. At higher dose, however, Rf decreased rapidly until it leveled off at 10-15 % at 0.04 e/Å2.
3. Radiation damage study of crotoxin complex crystals
The highest possible number of electron diffraction patterns, that
could be acquired with the slow-scan CCD camera from a single crotoxin complex
crystal under the constraints of radiation damage, was assessed by recording
successive patterns from the same crystal. Three of such damage series were
recorded up to an accumulated electron dose of ~5 e/Å2. We
analyzed quantitatively two series obtained from untilted crystals and one
series obtained from a tilted (45o ) crystal. Rf was found to be
less than 25 % for each pattern in these series up to an accumulated electron
dose of 2.5 e/Å2. In fact, Rf remained below 30 % up
to a final dose of 5 e/Å2.
Fig. 5. Relative crystalline disordering factor (Bn )
as function of
accumulated electron dose determined from electron diffraction patterns
acquired using the slow-scan CCD camera. The data was obtained in a radiation
damage experiment from three 128 Å-thick glucose-embedded crotoxin
complex crystals kept at -125o C in the untilted position
( ) and ( ) and tilted
by 45o (*).
|
Fig. 6. Plot of structural similarity R-factor, Rn , as function of the
accumulated electron dose determined from electron diffraction patterns
acquired using the slow-scan CCD camera. The data was obtained in a radiation
damage experiment from three 128 Å-thick glucose-embedded crystals of
crotoxin complex kept at -125o C at the untilted position
() and () and tilted
by 45o ( ).
|
The quantitative assessment of the radiation damage of the crotoxin complex crystal was made using the relative crystalline disordering factor, Bn, and the structure similarity factor, Rn , between the first and nth pattern from each series (Jeng & Chiu, 1984). Bn, which represents the increase of the B-factor of the diffraction intensity, ranged from 2.5 to about 15 Å2 (Fig. 5). Rn , which measures the changes in the diffraction amplitudes, remained below 30 % for all three series up to an accumulated electron dose of 2.5 e/Å2. Above this dose most of the structural similarity disappeared, as suggested by the rise of Rn to 40 or 60 % at a final dose of 5 e/Å2 (Fig. 6).
4. Acquisition of tilt series using the CCD camera
| Fig. 7. Plot of Friedel R-factor,
Rf , as function of tilt angle
determined from electron diffraction patterns acquired using
the slow-scan CCD camera. The patterns were obtained in three separate tilting
experiments each on individual glucose-embedded crotoxin complex crystals kept
at -125oC. The average Rf and its standard deviation (in
parenthesis) for the three sets are
14.5% (1.8%, ),
18.9% (1.9%, ), and 17.4% (1.9%, *). |
We collected several tilt series of glucose-embedded crotoxin complex crystals with the slow-scan CCD camera ranging from 50 to 125 electron diffraction patterns per series. The longest tilt series covered an angular range from +55o to -47o . An example is shown in Fig. 2B of an electron diffraction pattern acquired at high tilt that displayed sharp reflections in all directions indicating the absence of any significant crystal curvature. Because this highly tilted pattern was acquired first, using the slow-scan CCD camera, the flatness of the crystal was evaluated before time was spent on collecting an entire data set. All the patterns acquired in all the tilt series so far showed reflections out to better than 4 Å resolution. For three series shown here the average Rf ranged from 15 to 19 % with a standard deviation of 1.9 % per tilt series (Fig. 7). In fact, for more than 500 electron diffraction patterns acquired so far with the slow-scan CCD camera Rf ranged from 12 to about 25 %.
DISCUSSION
1. CCD installation in a JEOL4000 electron cryomicroscope
For an electron crystallographic analysis of macromolecules, one needs to record both diffraction and image data. A 400 kV electron microscope has been shown to be advantageous for obtaining high resolution (2.7 Å) images, primarily because of the smaller chromatic aberration effects (Brink & Chiu, 1991; Brink et al., 1992). The slow-scan CCD camera has been shown to be a valuable accessory to an electron microscope for direct data acquisition as used in on-line electron optical adjustments and electron tomographic applications (Downing et al., 1992; Koster et al., 1992; Koster & De Ruijter, 1992; Dierksen et al., 1993). Use of a slow-scan CCD camera for the acquisition of diffraction data in an electron crystallographic application would allow for a fast evaluation and an immediate, subsequent numerical analysis of the data, contrary to the imaging plate, which too has been used for the acquisition of electron diffraction intensities (Burmester & Schröder, 1994). Furthermore, the quality of the acquired data should be higher, since this camera performs better than the photographic film in terms of linearity, background noise and dynamic range (de Ruijter & Weiss, 1992; Krivanek & Mooney, 1993). Its applicability to protein electron crystallography at 400 kV has resulted from a number of engineering changes, that were made to the slow-scan CCD camera, such as minimization of the spurious X-ray signals picked up by the camera.
2. Assessment of crystal flatness and thickness
In order to solve the high resolution structure of glucose-embedded crotoxin complex, we must address technical issues such as crystal flatness, sorting out data from different crystal thicknesses, and finally, recording and merging of the 3-dimensional diffraction and image data. We have demonstrated previously that individual electron diffraction patterns and images of glucose-embedded crotoxin complex crystals can be obtained showing structural data beyond 2.7 Å resolution (Chiu & Jeng, 1980; Jeng & Chiu, 1983; Brink et al., 1992). However, we have been hampered in our attempts so far to merge the data coherently in three dimensions partly because of an uncertainty in the crystal thickness for most of the data, and partly because of the often observed lack of crystal flatness. Therefore, our primary concern has been to develop an experimental strategy which allows us to record the maximum amount of data from a crystal whose thickness and flatness can be ascertained before the data collection is started. The availability of the slow-scan CCD camera has opened up a new opportunity to resolve these issues at a point when the microscope operator selects suitable crystals for data collection, which was impractical to achieve when data was recorded on photographic film.
Our strategy for the intensity data collection would be to first evaluate the crystal thickness from search-mode images as shown in Figure 1. This would ensure that data necessary for the 3-dimensional structure determination would be obtained from crystals close to the desired thickness, namely 128 Å. Crystals of this thickness were measured the thinnest of those displaying diffractive power (Jeng & Chiu, 1983; Leapman et al., 1993). By comparing the mass thickness measurement with those performed on specimens of known thickness, like n-paraffin and bacteriorhodopsin, we are able to obtain an estimate of the protein crystal thickness. The reliability of this approach was confirmed by comparing our thickness data with that obtained using the parallel EELS (Leapman et al., 1993). The exact same crystals used for the thickness measurement in our JEOL4000 microscope were now used for the thickness measurement in a VG Microscopes HB501 STEM with a parallel EELS attachment. Preliminary data have indicated an accuracy of our method using the CCD camera of approximately 70 Å (Leapman et al., 1993; Brink et al., 1994). This means that we can confidently exclude crystals of 256 Å thickness or more from our data collection procedure. A crystal thinner than 256 Å or a full unit cell for the 3D reconstruction is preferred to ensure the validity of the weak-phase approximation without the complications of dynamical scattering and effects of Ewald sphere curvature (Cohen et al., 1984; Ho et al., 1988; Glaeser & Ceska, 1989).
Once after a crystal of suitable thickness has been observed, its flatness is assayed by evaluating the sharpness of diffraction spots in an electron diffraction pattern from a highly tilted crystal acquired with the slow-scan CCD camera. For crotoxin complex crystal, the diffraction pattern is usually quite strong (Fig. 2B). Hence, the sharpness of the reflections as well as the extent of the pattern, which relates to the crystallinity, can be readily judged from an unprocessed pattern. However, for crystals of lesser diffracting power, it may be necessary to subtract the local background resulting in patterns as shown in Figure 2.
3. Characteristics of electron diffraction patterns acquired on the CCD camera
The overall appearance of an electron diffraction pattern of a protein crystal acquired by the slow-scan CCD camera is rather similar to that recorded on photographic film except for the streak through the pattern's center (Fig. 2). Because the streak contains pixels around 4,000 ADU, reflections that are too close to this streak will have integrated intensities that are several orders of magnitude higher than their actual values. The large amount of charge that is generated by the undiffracted beam could be prevented from overflowing into adjacent pixels by means of a hardware alternative, called anti-blooming electrodes (Krivanek & Mooney, 1993). Since these are not implemented in our system, a post-acquisition correction must be made to exclude these reflections from symmetry calculations and subsequent merging operations to prevent erroneous results.
Although blooming due to the high intensity, undiffracted beam is relatively pronounced in the electron diffraction pattern (Fig. 2), search-mode images recorded after an electron diffraction pattern was recorded with the CCD camera never showed an after-image of this high intensity beam. This effect was not seen even after a large number (> 100) of electron diffraction patterns were recorded. This after-image does become visible when a very high electron dose (> 50-100 e/Å2) is used in the focussed diffraction mode. The undiffracted beam will then generate charge that accumulates not just in the top layer of the CCD chip, but also in the bulk silicon of the chip. Of the entire chip only the charge in the top layer is cleared before each exposure. Erasure of the remaining charge requires heating it back to room temperature.
Despite additional fiber optics, X-ray pixels are observed in the acquired frames. To obtain accurate integrated intensities, these X-ray pixels, whose intensities range from 100 to 4,095 ADU, must be excluded as well. Cosmic rays that hit the CCD chip, or "hard X-ray pixels", typically result in clusters of 5 to 10 pixels each at 4,095 ADU. They are seen on average once every 200 to 300 acquisitions in full-sized frames and are easily removed by applying an intensity threshold. X-ray photons on the other hand, that originate from within either the microscope column or the slow-scan CCD camera will typically yield a single pixel at 100 ADU. Usually every acquired frame will contain 3 to 5 of these pixels. Identification of these X-ray pixels is done by statistically analyzing a single, background-subtracted electron diffraction pattern. In this analysis, the highest intensity pixel is compared with the standard deviation computed from the remaining pixels in a 3 x 3-pixel box centered on this suspect X-ray pixel. Typically, X-ray pixels will show large intensities and small standard deviations as opposed to genuine data pixels. An identified X-ray pixel is effectively removed by replacing that pixel's intensity with the mean intensity computed over the remaining pixels in a 5 x 5-box centered on the X-ray pixel. Alternatively, removal can be done through a method known as "double take", which has been used with specimens that are fairly beam-resistant (Koster et al., 1992). Our method, however, has the advantage that it does not require 2 frames for X-ray filtering. It may occasionally fail to remove all X-ray pixels, although for electron diffraction data this will have a negligible effect. First, our patterns contain only a small number of X-ray pixels. Second, since they appear at random places in the diffraction pattern as opposed to the Bragg reflections, the choice of an appropriately small window for integrating the pixel intensities (9 x 9 pixels) will ensure exclusion of practically all remaining X-ray pixels from the intensity integration. Nevertheless, a robust computational method to remove X-ray pixels based upon single frames would be helpful.
4. Statistical definition of diffraction patterns acquired with the CCD camera
Three methods were used to assess the statistical definition of the acquired electron diffraction patterns of glucose-embedded crotoxin complex crystals. First, we analyzed a set of patterns in terms of their intensity distributions (Table 1). The linear graph that we obtained from the (mean intensity)1/2 versus standard deviation (Fig. 4A) implies that the registration of an incoming electron by the slow-scan CCD camera is governed by counting statistics (Blundell & Johnson, 1976). The progressively larger spread in the standard deviations at higher intensity reflects the fact that fewer reflections are found in the higher intensity ranges (Table 1). Second, we estimated Rf for the crotoxin complex from eq. 1 using the results obtained on the light harvesting complex by Wang and Kühlbrandt (1992), and by making the approximation that Rf is dominated by the stronger reflections. Our estimate of 18 to 26 % is confirmed by over 500 experimentally observed values of Rf that mostly range from 12 to 25 %. Third, at increasing dose Rf approached a linear relationship with (electron dose)1/2 as is expected for a process governed by counting statistics (Fig. 4B). The deviation from linearity at very low exposures coincides with a significant drop in the camera's DQE. We determined the DQE for our CCD camera to range from 0 to 0.5 below 1 electron/pixel and to remain constant at 0.6-0.7 above 2 electrons/pixel, which is similar compared to results on a similar camera obtained by others (Ishizuka,1993; Krivanek & Mooney, 1993). At the smallest dose used in the experiment, the majority of the pixels received zero or one electron strongly suggesting that the small DQE at low electron dose is responsible for the unproportionally high Rf. This analysis indicates that the electron diffraction data recorded with the slow-scan CCD camera at a dose equal to or more than 0.01 e/Å2 follows statistical principles and therefore can be analyzed with statistical confidence.
The strongest reflection in a typical electron diffraction pattern of glucose-embedded crotoxin complex acquired with the slow-scan CCD camera measure approximately 400 to 800 ADU, or 250 to 530 electrons, respectively (Fig. 3 and Table 1). The normalized intensity for these reflections has been measured relative to the undiffracted beam as 1.5-5x10-5, which is ~1.5 times smaller than that obtained previously on crotoxin complex crystal at 100 kV (Henderson & Glaeser, 1985). This difference in diffraction contrast is expected based upon the relative changes in elastic and inelastic scattering cross-sections at higher electron energy (Glaeser, 1985). An identical change in the diffraction contrast of n-paraffin was observed using data collected on photographic film at 100 and 400 kV (Brink & Chiu, 1991). Since the same change in the diffraction contrast was observed using data collected using film and the slow-scan CCD camera, the comparison made here would reinforce the capability of the camera for measuring electron diffraction intensities of protein crystals.
The impact on the statistical definition of 14- or even 16-bit digitization would be negligible. The strongest reflections are adequately sampled with the current 12-bit system. Merging several frames at different exposures is therefore not only necessary, but in fact would lead to a loss of definition since the total noise in the merged diffraction pattern would increase due to the read-out noise's additive character (de Ruijter & Weiss, 1992). Therefore, the choice of electron exposure and quantization would result in patterns with the smallest amount of read-out noise and the highest possible statistical definition.
4.5. Radiation damage series of crotoxin complex crystals
In a tilting experiment the limit on the number of usable views is ultimately set by radiation damage. It has been common to characterize radiation damage by the critical dose, which is defined as the dose at which the diffraction intensity has fallen to 1/e of its initial value (Unwin & Henderson, 1975). Invariably, however, one finds certain reflections exhibiting anomalous behavior such as an increase in diffraction intensity at higher accumulated dose (Unwin & Henderson, 1975; Jeng & Chiu, 1984). Therefore, we used the parameters Bn and Rn , which are commonly used in X-ray crystallography to assess the radiation damage on a global scale (Hendrickson, 1976). In particular, Rn measures the similarity of the amplitudes of the structure factor of the crystal.
We found that the fall-off for Bn is comparable whether the crystal is tilted or untilted (Fig. 5). Compared to a similar study done on crotoxin complex at 100 kV using data collected on photographic film, the fall-off observed at 400 kV is lower by a factor of 1.5 to 2 (Jeng & Chiu, 1984). This reduction in the amount of damage is similar to that found previously in a radiation damage study on l-valine crystals at different electron voltages (Howitt et al., 1976), which is in accordance with the Bethe stopping power theory (Glaeser, 1975). Rn increases above an accumulated dose of 2.5 e/Å2 in a way which is similar for both tilted and untilted crystals (Fig. 6). The lowest value of Rn for a particular series is apparently determined by its Rf. Because the diffraction data in a tilt series is obtained at varying levels of radiation damage, a correction will have to be made on the intensities before they could be used in a structural determination. Based upon the data shown here, we conclude that such a correction depends only on the electron exposure and not on the tilt angle. According to the behavior of Rn , the glucose-embedded crotoxin complex crystals kept at -125o C can tolerate a total electron dose of approximately 2.5 e/Å2 at 400 kV. Since electron diffraction patterns acquired at a dose of 0.01 e/Å2 have been shown to exhibit enough statistical definition, the similarity data suggests the feasibility of recording up to 250 electron diffraction patterns with minimal damage.
4.6. Tilt series acquisition of crotoxin complex crystals
Our strategy for the three-dimensional diffraction data collection is to obtain as much intensity data as possible from single protein crystals in a tilting experiment (Jeng & Chiu, 1980; Prasad et al., 1986). The advantages here will be 3-fold: first, a much more accurate 3-dimensional merging of the diffraction data will be possible as more sample points along each of the lattice rods are obtained. Second, a better verification will be possible on the crystal thickness by evaluating the symmetry in each of the tilt series, which essentially form partial 3-dimensional intensity data sets. Crystals of the same thickness must display exactly the same kinds of symmetry, e.g. a particular combination of in-plane 2-folds and 21-axes at particular angles with possibly out-of-plane 2-folds. More intensity data basically allows for a more accurate determination of all the R-factors for each of the symmetries analyzed. The quality of the high tilt data is extremely dependent on crystal flatness, which incidentally is more crucial for the crotoxin complex than for bacteriorhodopsin due to the larger crystal thickness of the former. Despite the use of the molybdenum grids, searching for a flat crystal on the grid is still a labor-intensive task. Therefore, the third advantage is that if more intensity data can be retrieved from a single crystal, fewer crystals will be needed to obtain a complete 3-dimensional data set.
Several tilt series were recorded from single glucose-embedded crystals of crotoxin complex to illustrate the feasibility of recording very large numbers of electron diffraction patterns with minimal damage to the crystal. The success of a tilt series acquisition, i.e. the number of patterns per crystal, depends on the size of the crystal, its crystallinity, the quality of the embedding by the glucose, and the extent of the unproductive use of electrons due to repositioning the crystal in the beam. Variability in these factors explains why certain tilt series are relatively short and also why we have not succeeded in collecting a series of 250 patterns. For example, some tilt series contained only 50 electron diffraction patterns while other series were composed of 125 patterns. Essentially, Rf for each pattern throughout each tilt series would remain the same (Fig. 7). The spread in Rf per tilt series of only ~2 % indicated that Rf was essentially independent of the tilt angle. For example, for one tilt series that covered more than 100o in tilt, Rf changed by less than 5 %.
As mentioned above, one of the factors limiting the number of obtainable diffraction patterns from a crystal, has been the electron dose required only for re-positioning the crystals in the beam. This is related to stage movement caused first by backlash, which is particularly large when the tilt angle is changed in the opposite direction from either one of the extreme angles, e.g. +60o or -60o , and second by non-eucentricity of the stage. For example, shifts caused by backlash were measured to range from 3 to 10 um at various positions on the grid. Shifts due to non-eucentricity were typically around 1 um. Unfortunately, off-crystal markers used to monitor crystal movement may not lie exactly on the mechanical tilt axis, or sometimes are not present at all. Under these circumstances when the stage must be re-positioned the crystal itself must be viewed. Consequently, the crystal is irradiated resulting in damage and a loss of patterns that could potentially have been obtained from that particular crystal. We estimated that this is the main factor responsible for the loss of potential patterns in all tilt series recorded so far. In principle, one can improve this situation by implementing the computational procedure for automatic specimen realignment as now used in electron tomography on cellular materials and single molecules to minimize the nonproductive use of electrons during a tilting experiment (Koster et al., 1992; Dierksen et al., 1993). Alternatively, stage designs utilizing sapphire-free specimen rods might help to minimize stage backlash by allowing a smaller surface area being subject to temperature gradients as well as friction (Kaneyama et al., 1989; Lücken et al., 1994).
4.7. General applicability of the CCD camera in protein electron diffraction
The crotoxin complex crystal is well suited for the acquisition of high resolution electron diffraction intensities with the slow-scan CCD camera. For one, it has a small two-dimensional unit cell, so that adjacent reflections will not overlap despite the limited size of the CCD chip. Furthermore, its high resolution reflections are relatively strong. However, one may encounter problems with protein crystals that have large unit cells and that are weaker diffractors, like for instance the tropomyosin crystal with a = 799 Å and b = 55 Å (Avila-Sakar et al., 1993). A larger unit cell results in more closely spaced Bragg reflections at the same edge resolution of the CCD chip. In order to get accurate intensities, one would need to develop alternative ways to correct for the background when the reflections are separated by only a few pixels. In the extreme case, one can use a computational method to deconvolute overlapping reflections as has been done in X-ray crystallography (Rossmann, 1979). Alternatively, one can use a combination of a larger chip (up to 4096 x 4096 pixels) and a larger camera length so as to achieve a separation between the reflections which is adequate for the current algorithms. For weaker diffraction intensity, one can develop more sophisticated computational procedures to average multiple diffraction patterns in order to obtain the necessary statistics. However, the read-out noise that will accumulate upon combining several frames might then present a problem (de Ruijter & Weiss, 1992). Alternatively, a higher electron dose can be used, which necessitates the use of anti-blooming electrodes (Krivanek & Mooney, 1993). Further development should be done on the use of alternative computer platforms such as UNIX workstations with the appropriate software tools to speed up the data acquisition and the on-line analysis. In addition, development of CCD cameras with multiple read-out speeds to allow for video-rate frame transfers as well as faster 12- or 14-bit digitization would be very helpful. It would obviate a separate TV-camera for specimen search and reduce the acquisition time. The latter factor is of practical importance because only small grid areas from a single frozen, hydrated specimen would be examined and recorded before it becomes contaminated with condensed ice after a period of 3 hours of observation.
5. CONCLUSIONS
We have demonstrated the usefulness of a 1024 x 1024 slow-scan CCD camera in the electron crystallographic study of glucose-embedded crotoxin complex crystals. Use of the slow-scan CCD camera allows for an on-line assessment of crystallinity of the specimen, crystal thickness and flatness in a way that was not possible before. The subtractable background noise in a frame acquired using the slow-scan CCD camera allows for a relatively low electron exposure (0.01-0.02 e/Å2) per acquisition whilst still accurately measuring the relatively weak reflections. The quantitative analysis of the diffraction intensities revealed that the R-factor for the Friedel-related reflections, Rf, ranged from 12 to 25 %, which was expected based upon electron statistics and crystal sizes. The quantitative analysis of the diffraction intensity in the radiation damage series suggested that we can record up to 250 patterns per crystal with 3.5 Å data. So far, the longest of many recorded tilt series contained 125 electron diffraction patterns from a single, glucose-embedded crotoxin complex crystal with tilt angles ranging from +55o to -47o . Overall, the data collection has been made more efficient by using a slow-scan CCD camera to record electron diffraction intensities from crotoxin complex crystals. This indicates that with the wider application of the CCD camera a more rapid development of electron crystallography for the structural analysis of protein crystals is possible.
6. ACKNOWLEDGMENTS
We would like to acknowledge support from the W. M. Keck Foundation, the National Center for Research Resources of NIH (RR02250) and the National Science Foundation (BR9202199). We express our gratitude to Mr. M. Dougherty for his help during the initial stage of setting up the computer interface between the JEOL4000EX microscope and the Macintosh computer; Dr. K. Downing for the bacteriorhodopsin specimens. We thank Drs. H. de Ruijter, M. Sherman and M. F. Schmid for the discussions and their comments on the manuscript.
REFERENCES
Avila-Sakar, A. J., Schmid, M. F., Li, L. S., Whitby, F., Phillips, G. N., Jr. & Chiu, W. (1993). Preliminary electron crystallographic analysis of ice-embedded tropomyosin crystals. J. Struct. Biol. 110, 67-74.
Baldwin, J. & Henderson, R. (1984). Measurement and evaluation of electron diffraction patterns from two-dimensional crystals. Ultramicroscopy 14, 319-336.
Blundell, T. L. & Johnson, L. N. (1976). Protein Crystallography. New York, Academic Press.
Booy, F. P. & Pawley, J. B. (1993). Cryo-crinkling: what happens to carbon films on copper grids at low temperature. Ultramicroscopy 48, 273-280.
Brink, J. & Chiu, W. (1991). Contrast analysis of cryo-images of n-paraffin recorded at 400 kV out to 2.1 Å resolution. J. Microsc. 161, 279-295.
Brink, J., Chiu, W. & Dougherty, M. (1992). Computer-controlled spot-scan imaging of crotoxin complex crystals with 400 keV electrons at near atomic resolution. Ultramicroscopy 46, 229-240.
Brink, J., Leapman, R. D. & Chiu, W. (1994). Manuscript in preparation.
Burmester, C. & Schröder, R. R. (1994). Qauntitative Electron Diffraction on Frozen Hydrated Protein 3D Microcrystals. In: Proc. XIIIth Int. Congr. Elec. Microsc., Paris (France), Vol. 1, pp. 593-594, Les Editions de Physique Les Ulis.
Chiu, W. & Jeng, T.-W. (1980). Electron diffraction study of crotoxin complex at 1.6 Å. In: Electron Microscopy at Molecular Dimensions. State of the Art and Strategies for the Future. Berlin, Springer-Verlag. 137-142.
Cohen, H., Schmid, M. F. & Chiu, W. (1984). Estimates of validity of projection approximation for three-dimensional reconstructions at high resolution. Ultramicroscopy 14, 219-226.
de Ruijter, W. J. & Weiss, J. K. (1992). Methods to measure properties of slow-scan CCD cameras for electron detection. Rev. Sci. Instru. 63, 4314-4321.
Dierksen, K., Typke, D., Hegerl, R. & Baumeister, W. (1993). Towards automatic electron tomography. II. Implementation of autofocus and low-dose procedures. Ultramicroscopy 49, 109-120.
Dorset, D. L. (1980). Electron diffraction intensities from bent molecular organic crystals. Acta Cryst. A36, 592-600.
Downing, K. H., Koster, A. J. & Typke, D. (1992). Overview of computer-aided electron microscopy. Ultramicroscopy 46, 189-198.
Glaeser, R. M. (1975). Radiation damage and biological electron microscopy. In: Physical aspects of electron microscopy and microbeam analysis. N. Y., John Wiley & Sons. 205-229.
Glaeser, R. M. (1985). Electron crystallography of biological macromolecules. Ann. Rev. Phys. Chem. 36, 243-275.
Glaeser, R. M. & Ceska, T. A. (1989). High-voltage electron diffraction from bacteriorhodopsin (purple membrane) is measurably dynamical. Acta Cryst. A45, 620-628.
Hall, C. E. (1951). Scattering Phenomena in Electron Microscope Image Formation. J. Appl. Phys. 22, 655-662.
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.
Henderson, R., Baldwin, J. M., Downing, K. H., Lepault, J. & Zemlin, F. (1986). Structure of purple membrane from Halobacterium halobium: Recording, measurement and evaluation of electron micrographs at 3.5 Å resolution. Ultramicroscopy 19, 147-178.
Henderson, R. & Glaeser, R. M. (1985). Quantitative analysis of image contrast in electron micrographs of beam-sensitive crystals. Ultramicroscopy 16, 139-150.
Henderson, R. & Unwin, P. N. T. (1975). Three-dimensional model of purple membrane obtained by electron microscopy. Nature 257, 28-32.
Hendrickson, W. (1976). Radiation damage in protein crystallography. J. Mol. Biol. 106, 889-893.
Ho, M.-H., Jap, B. K. & Glaeser, R. M. (1988). Validity domain of the weak-phase-object approximation for electron diffractionof thin protein crystals. Acta Cryst. A44, 878-884.
Howitt, D. G., Glaeser, R. M. & Thomas, G. (1976). The energy dependence of electron radiation damage in l-valine. J. Ultrastruct. Struct. 55, 457-461.
Ishizuka, K. (1993). Analysis of electron image detection efficiency of slow-scan CCD camera. Ultramicroscopy 52, 7-20.
Isoda, S., Saitoh, K., Moriguchi, S. & Kobayashi, T. (1991). Utility test of imaging plate as a high-resolution image-recording material for radiation-sensitive specimens. Ultramicroscopy 35, 329-338.
Jap, B. K., Walian, P. J. & Gehring, K. (1991). Structural architecture of an outer membrane channel as determined by electron crystallography. Nature 350, 167-170.
Jeng, T.-W. & Chiu, W. (1980). Experimental strategy in three-dimensional structure determination of crotoxin complex thin crystal. Ultramicroscopy 13, 27-34.
Jeng, T. W. & Chiu, W. (1983). Low dose electron microscopy of the crotoxin complex thin crystal. J. Mol. Biol. 164, 329-346.
Jeng, T. W. & Chiu, W. (1984). Quantitative assessment of radiation damage in a thin protein crystal. J. Micros. 136, 35-44.
Kaneyama, T., Naruse, M., Ishida, Y. & Kersker, M. (1989). Development of 200kV ultrahigh-resolution analytical electron micrscope. In: Proc. 47th Ann. Meeting EMSA, San Antonio TX, Vol. pp. 108-109, San Francisco Press, Inc.
Koster, A. J., Chen, H., Sedat, J. W. & Agard, D. A. (1992). Automated microscopy for electron tomography. Ultramicroscopy 46, 207-228.
Koster, A. J. & De Ruijter, W. J. (1992). Practical autoalignment of transmission electron microscopes. Ultramicroscopy 46, 189-197.
Krivanek, O. L. & Mooney, P. E. (1993). Applications of slow-scan CCD cameras in transmission electron microscopy. Ultramicroscopy 49, 95-108.
Kühlbrandt, W., Wang, D. N. & Fujiyoshi, Y. (1994). Atomic model of plant light-harvesting complex. Nature 367, 614-621.
Leapman, R., Brink, J. & Chiu, W. (1993). Low-dose thickness measurement of glucose-embedded crotoxin complex crystals by electron energy loss spectroscopy and STEM dark-field imaging. Ultramicroscopy 52, 157-166.
Lücken, U., Busing, W. M., Felsmann, M. & de Jong, F. (1994). Developments in TEM for Life Science Research. In: Proc. XIIIth Int. Congr. Elec. Microsc., Paris (France), Vol. 1, pp. 277-278, Les Editions de Physique Les Ulis.
Mooney, P. E., Fan, G. Y., Truong, K. V., Bui, D. B. & Krivanek, O. L. (1990). Slow-scan CCD camera for transmission electron microscopy. In: Proc. XIIth Int. Cong. Electron Microsc., Seattle, WA, Vol. 1, pp. 164-165, San Francisco Press, Inc.
Mori, N., Oikawa, T., Katoh, T., Miyahara, J. & Harada, Y. (1988). Application of the "imaging plate" to TEM image recording. Ultramicroscopy 25, 195-201.
Ogura, N., Yoshida, K., Kojima, Y. & Saito, H. (1994). Development of the 25 micron Pixel Imaging Plate System for TEM. In: Proc. XIIIth Int. Congr. Elec. Microsc., Paris (France), Vol. 1, pp. 219-220, Les Editions de Physique Les Ulis.
Prasad, B. V. V., Degn, L.-L., Jeng, T.-W. & Chiu, W. (1986). 3-D diffraction intensities of crotoxin complex crystal. In: Proc. 44th Ann. Meeting EMSA, Albuquerque, NM, Vol. 1, pp. 162-163, San Francisco Press, Inc.
Prasad, B. V. V., Degn, L. L., Jeng, T. & Chiu, W. (1990). Estimation of allowable errors for tilt parameter determination in protein electron crystallography. Ultramicroscopy 33, 281-285.
Rossmann, M. G. (1979). Processing oscillation diffraction data for very large unit cells with an automatic convolution technique and profile fitting. J. Appl. Cryst. 12, 225-238.
Unwin, P. N. T. & Henderson, R. (1975). Molecular structure determination by electron microscopy of unstained crystalline specimens. J. Mol. Biol. 94, 425-440.
Wang, D. N. & Kühlbrandt, W. (1992). Three-dimensional electron diffraction of plant light-harvesting complex. Biophys. J. 61, 287-297.
Zeitler, E. & Bahr, G. F. (1962). A photometric procedure for weight determination of submicroscopic particles quantitative electron microscopy. J. Appl. Phys. 33, 847-853.