Performance of a Slow-Scan CCD Camera for Macromolecular Imaging in a 400 kV Electron Cryomicroscope

Michael B. Sherman, Jacob 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

We have evaluated the feasibility and limitations of a 1024 x 1024 slow-scan charge-coupled device (CCD) camera for imaging in a 400 kV electron cryomicroscope. We used catalase crystals and amorphous carbon film as test specimens. Using catalase crystals, we found that the finite (24 um) pixel size of the slow-scan CCD camera governs the ultimate resolution in the acquired images. For instance, spot-scan images of ice-embedded catalase crystals showed resolutions of 8 Å and 4 Å at effective magnifications of 67,000x and 132,000x, respectively. Using an amorphous carbon film, we evaluated the damping effect of the modulation transfer function (MTF) of the slow-scan CCD camera on the specimen's Fourier spectrum relative to that of the photographic film. The MTF of the slow-scan CCD camera fell off more rapidly compared to that of the photographic film and reached the value of 0.2 at the Nyquist frequency. Despite this attenuation, the signal-to-noise ratio of the CCD data, as determined from reflections of negatively stained catalase crystals, was found to decrease to ~ 50% of that of photographic film data. The phases computed from images of the same negatively stained catalase crystals recorded consecutively on both the slow-scan CCD camera and photographic film were found to be comparable to each other within 12o. Ways of minimizing the effect of the MTF of the slow-scan CCD camera on the acquired images are also presented.

INTRODUCTION

In protein electron crystallography, both electron diffraction patterns and images are used to provide amplitudes and phases for a 3-dimensional reconstruction. In principle these data can be recorded on a slow-scan charge-coupled device (CCD) camera. Indeed, such data has been obtained from a variety of samples at various accelerating voltages up to 400 kV (Dierksen et al., 1992; Koster et al., 1992; Dierksen et al., 1993; Pan and Crozier, 1993; Brink and Chiu, 1994; Dierksen et al., 1995). For recording electron diffraction patterns the slow-scan CCD camera is an ideal detector because of its broad dynamic range, high linearity and low noise (Krivanek and Mooney, 1993; Brink and Chiu, 1994; Faruqi et al., 1994; Faruqi et al., 1995; de Ruijter, 1995). It has been demonstrated that the quality of electron diffraction patterns of glucose-embedded crotoxin complex crystals is high as reflected in a 4.5% R-factor for Friedel symmetry-related reflections up to 2.7 Å resolution (Brink et al., 1995). Moreover, more than 100 diffraction patterns could be recorded in a single tilt series of crotoxin complex crystals to 3.5 Å resolution with an accumulative dose of less than 3 electrons/Å2. This approach to collect data is more efficient than the conventional method of recording one pattern per crystal, particularly for crystalline specimens with a variable thickness or when large and sufficiently flat crystals are rare. As part of this approach, the slow-scan CCD camera is useful to provide an on-line evaluation of the quality of a crystal in terms of crystallinity, flatness and thickness before the tilt data is recorded (Brink and Chiu, 1994).

In image acquisition the slow-scan CCD camera has been used to enable an on-line evaluation of the defocus and astigmatism (Koster and de Ruijter, 1992; Krivanek and Mooney, 1993), to facilitate automatic data collection in electron tomographic applications (Dierksen et al., 1992; Koster et al., 1992; Dierksen et al., 1993; Dierksen et al., 1995) and to record high resolution images in material science (Pan and Crozier, 1993). The performance of slow-scan CCD cameras is known to be dependent on several factors including the accelerating voltage, and in many cases better at lower voltages (Daberkow et al., 1991). The image resolution is ultimately limited by the finite pixel size of the slow-scan CCD camera as formulated in the Whittaker-Shannon sampling theorem (Goodman, 1968). Therefore, the electron microscope magnification has to be set to a minimum value so as to achieve the intended image resolution (Table 1). The table shows the minimum magnifications that are required for different resolutions for a Gatan model 679 slow-scan CCD camera, which has 24x24-um2 pixels. In general, biological electron microscopists record images at the lowest possible magnification. In this way, the illumination dose can be kept low enough to minimize radiation damage while maintaining sufficient optical density above the fog level of the photographic emulsion (Kuo and Glaeser, 1975; Unwin and Henderson, 1975; Chiu and Glaeser, 1980). For instance, 3 Å images of crotoxin complex crystals have been recorded at 40,000x magnification (Brink et al., 1992). For the same image resolution, the magnification on the slow-scan CCD camera must be at least four times higher than that used on photographic film (Table 1).

Table 1. Minimum magnifications required at different target
resolution using a slow-scan CCD camera with 24-micron pixel.

Target resolution (Å)

Minimal effective magnification
15
32,000x
7
68,500x
3
160,000x

This implies a higher electron dose which may exceed the radiation tolerance of the specimen. However, there is a possibility that due to the practically fog-free nature of the images acquired on a slow-scan CCD camera the required electron dose at the high magnification could be lower than generally used for photographic film recording.

In this paper, we used ice-embedded and negatively stained catalase crystals and thin amorphous carbon film to evaluate the performance of a 1024 x 1024 slow-scan CCD camera for imaging purposes at 400 kV in terms of resolution, modulation transfer function (MTF), signal-to-noise ratio (SNR) and reliability of retrieved phases. The carbon film and negatively stained catalase crystals have allowed direct comparison between consecutive images recorded from the same specimen area on both the slow-scan CCD camera and photographic film because of their relative insensitivity to electron radiation.

MATERIALS AND METHODS

Specimen preparation

Beef liver catalase crystals were purchased from Boehringer Mannheim. The crystal suspension was dissolved in 10% (w/v) NaCl solution in water and was recrystallized by dialysis against 50 mM phosphate buffer at pH 6.3 for 1 to 3 days (Unwin, 1975). For low-dose spot-scan imaging the catalase crystals were embedded in vitreous ice on thin carbon films which spanned holes in a holey carbon support net on 400 mesh copper grids. Fast freezing was done in liquid nitrogen-cooled ethane using a guillotine-like device (Jeng et al., 1988). The frozen, hydrated specimen was held in a Gatan 626 cryoholder at -167oC in a JEOL 4000EX electron microscope outfitted with a side-entry goniometer stage and a LaB6 electron gun. Negative staining was done using a 1% aqueous solution of uranyl acetate.

Electron optical conditions

A 50 um diameter condenser aperture was used together with a relatively high excitation (spot size 5) of the first condenser lens resulting in a measured angular source size of 0.07 mrad (Spence, 1988). A 70 micron objective lens aperture was used. We used the user function memory slots of the microscope's microprocessor to store and recall settings of lenses and deflection coils for several operational modes including search, focus and imaging using either spot-scan or flood-beam illumination (Brink et al., 1992).

The microscope has a Gatan (Model 679) 1024 x 1024 slow-scan CCD camera attached directly below the projection chamber. The camera was equipped with an extra fiber optics coupling plate and was shifted off the microscope axis by about 2.5 cm to reduce the number of x-ray photons hitting the CCD chip. To maintain sensitivity the camera had a P43 phosphor instead of a single-crystal YAG scintillator (Brink and Chiu, 1994; Mooney and Krivanek, 1994).

Low dose imaging on the slow-scan CCD camera and Kodak SO-163 photographic film was done basically using the standard technique described elsewhere (Jeng and Chiu, 1983). Thin crystals at least 2 um on edge were searched for with a TV-rate CCD camera (Gatan, model 673) at a dose-rate of 5 x 10-3 electrons/Å2/s in a defocused diffraction mode. Search mode images of crystals and their electron diffraction patterns were recorded on the slow-scan CCD camera to evaluate crystal thickness and quality. Focusing was done off-axis on an area adjacent to a crystal at the magnification used for image recording. Because the photographic film and the slow-scan CCD camera were located at different planes, images on the slow-scan CCD camera were recorded at an effective magnification 1.67x of that used on photographic film. To test the slow-scan CCD camera's performance at various resolutions, two different microscope magnifications were used, namely 40,000x and 80,000x. Kodak SO-163 photographic films used for imaging were processed for maximum speed and contrast according to the manufacturer's instructions in undiluted Kodak D-19 developer for 12 minutes at 20oC.

Spot-scan imaging on the slow-scan CCD camera

Fig.1
Fig. 1. Diagram representing the spot-scan procedure implemented for imaging on the slow-scan CCD camera. The electron beam is deflected above the specimen using the beam shift coils to scan an area of interest. It is shifted back onto the optical axis using the image shift coils below the specimen. The final image is centered on the CCD chip using a projector shift, which remains constant during spot-scanning.

Spot-scan imaging on the slow-scan CCD camera was done using a computer program written in C, which was incorporated as a custom function in the Gatan DigitalMicrograph program (version 2.5) running on a Macintosh IIfx (Mooney et al., 1990). The program controlled, besides the slow-scan CCD camera, the microscope's deflection coils above and below the specimen to obtain a synchronous shift of the beam so that the final image would remain stationary on the CCD chip during spot-scan imaging (Fig. 1). The program consisted of two separate modules. The first module was used to set up and store on a hard disk the experimental parameters including exposure time per frame, amount of beam shift between adjacent spots, and adjustment of projector deflector coils to center the beam on the CCD chip. The second module recalled these parameters and acquired the actual spot-scan images. The acquisition time per spot was ~30 s, which included serial port communication between the computer and the microscope (~0.5 s), exposure of the CCD chip (typically 1 s), readout of the CCD chip and subsequent transfer of the raw data to the Macintosh memory (3.5 s). The remaining time was used for subtraction of the dark current reference, gain-normalization of the image (de Ruijter and Weiss, 1992; Kujawa and Krahl, 1992), display of the image on the computer screen in a reduced size and its storage on a local magnetic disk. Spot-scan images were acquired using the full CCD chip with 1024 x 1024 pixels. It is possible to reduce the acquisition time per spot considerably by adopting 2 modifications. Raw images can be stored without the dark current subtraction and gain normalization, which can in principle be done later. If the images are not displayed at all the acquisition time would be reduced from 30 s to ~ 6 s. We chose to gain-normalize and display the acquired images so as to have an immediate visual check on the spot-scan procedure and image quality. The specimen dose for imaging catalase crystals was 6 - 10 electrons/Å2 and 10 - 14 electrons/Å2 at 40,000x and 80,000x microscope magnification, respectively.

Quantitative image analysis

The image data recorded on photographic film was digitized using a Perkin Elmer (PDS) microdensitometer 1010M at 3.5 Å/pixel (carbon film), or 1.25 Å/pixel (ice-embedded catalase crystals). In the case of negatively stained catalase crystals the step size was 6.5 Å/pixel. Images recorded with the slow-scan CCD camera were transferred to a Silicon Graphics R4400 workstation for subsequent analysis. Image processing was done using the MRC program package (Henderson et al., 1986; Crowther et al., 1996) in conjunction with Spectra (Schmid et al., 1993). The processing included reciprocal lattice refinement and unbending. The computed amplitude for each reflection and its local background obtained from the program mmbox were used to calculate the signal-to-noise ratio (SNR). The IQ value which has been commonly used in protein electron crystallography, is derived from the SNR by taking its reciprocal value (1/SNR) followed by truncation of the result to integer and ranking it from 1 to 9. An IQ of 1 denotes the highest peak-to-background amplitude ratios (Henderson et al., 1986). To compare the CCD performance versus photographic film, we used a 'decay plot' in which the logarithm of reflection amplitudes was plotted against the square of the spatial frequency.

For images of carbon film obtained using flood-beam illumination at 40,000x microscope magnification, the computed amplitudes of the Fourier transforms were averaged azimuthally and used to obtain radial profiles of the transforms (Zhou et al., 1996). These radial profiles were used to assess the fall-off of the data. Comparison of carbon film image data recorded on both the CCD camera and photographic film was done by plotting the ratio as a function of the spatial frequency.

Estimate of CCD's modulation transfer function

For the CCD camera, we have determined the MTF by analyzing the Fourier amplitude fall-off in a difference image obtained with the slow-scan CCD camera without any specimen in the beam. This method is similar to that described by Ishizuka (1993). In addition, we introduced an indirect procedure to measure the slow-scan CCD camera's MTF relative to that of photographic film. The same area of a thin carbon film was recorded on both the slow-scan CCD camera and photographic film. The Fourier amplitude spectrum of an image registered by a detector is attenuated by the detector's MTF (de Ruijter and Weiss, 1992; Mooney et al., 1993) according to eq. 1a:

(1a)

where A(g)in and A(g)out are the Fourier amplitude spectra of the input and output images, respectively, is the MTF of the detector, and g is the spatial frequency. For the CCD camera and photographic film the ratio of the MTFs can be expressed as follows:

(1b)

This equation can be simplified to eq. 1c if the input signal is the same for both detectors. If one of the MTFs is known, for instance , then the remaining unknown MTF can be derived:

(1c)

In our experiment with the same carbon film area on both detectors, the electron flux through the object, the exposure time and the electron optical parameters except the magnification at the detector levels were identical. We compensated for this magnification difference by using a smaller step size in digitizing the photographic film image, so that the number of electrons per pixel was identical in both sets of digitized images. Table 2 summarizes the experimental and computational conditions for the carbon film experiment.

Table 2. Experimental settings used for amorphous carbon film imaging.

Recording medium

Effective magnification
Exposure time (s)
Pixel size (um)
Image area(pixels)
Gatan 679 slow-scan CCD camera
67,000x
1
24
1024 x 1024
Kodak SO-163 photographic film
40,000x
1
14
1024 x 1024

Images of negatively stained catalase crystals

Images from the same negatively stained catalase crystal area were taken at a microscope magnification of 20,000x on both slow-scan CCD camera and Kodak SO-163 photographic film. Negatively stained crystals were used because there was no significant structural change at low resolution after multiple exposures with an accumulated dose of 30 electrons/Å2. We acquired 3 to 5 images by recording them alternatively on the CCD camera and the photographic film under the same electron-optical conditions. Images acquired on photographic films were digitized using the PDS with the step size of 14 um, matching the 24 um pixel size of the CCD. The ratio of the corresponding SNR values for all reflections was plotted against the spatial frequency. The phase consistency among the crystal images was evaluated from the merging phase residuals using the origtilt program modified after Thomas and Schmid (1995).

RESULTS

Attainable resolution in the spot-scan images of ice-embedded catalase crystals recorded with CCD camera

Fig.2 Fig.3
Fig. 2. (a) 400 kV spot-scan image of an ice-embedded catalase crystal acquired with a 1024 x 1024 Gatan model 679 slow-scan CCD camera with an illumination dose of 6 electrons/Å2. The microscope's magnification was set at 40,000x and the effective magnification on the camera was 67,000x. Bar is 1000 Å. The unit cell dimensions of the catalase crystal are a = 173.5 Å and b = 69 Å. (b) IQ plot of the computed diffraction pattern of the image in (a). Only reflections with IQ <= 6 are plotted. For smaller IQ values the corresponding spot has a larger radius. The edge resolution was 7.2 Å. A dashed circle is drawn indicating 8 Å resolution. The first zero of the microscope's CTF can be observed at 10 Å resolution leading to weaker reflections around this spatial frequency. It can be seen that 85% of the reflections up to this resolution have IQ <= 4 (SNR better than 2). Fig. 3. (a) Similar to Figure 2a, but image was recorded at an effective magnification of 132,000x on the slow-scan CCD camera with an illumination dose of 12 electrons/ Å2. The Nyquist frequency corresponds to 1/3.6 Å-1. (b) IQ plot similar as in Figure 2b. Dashed circles are drawn to indicate 8, 5, and 4 Å resolution. The `gap' in the resolution zone between 8 and 6 Å reflects a near-zero region of the microscope's contrast transfer function (CTF).

Figure 2a is an example of a 400 kV spot-scan image of an ice-embedded catalase crystal recorded with a specimen dose of 6 electrons/Å2 on the slow-scan CCD camera at an effective magnification of 67,000x. The IQ plot of the computed amplitudes of this image (Fig. 2b) shows that ~ 85% of all reflections out to 8 Å had an IQ value of 4 or better. Even close to the edge of the image's Fourier transform, i.e. the Nyquist frequency, which corresponds to 7.2 Å resolution, rather strong reflections of IQs 2 and 3 could be found. A zero of the microscope's contrast transfer function (CTF) can be discerned at 10 Å resolution which is consistent with a defocus of approximately 0.5 um. All 11 processed spot-scan micrographs of the ice-embedded catalase crystals yielded results similar to those shown in figure 2. The thickness of the crystals, as measured from search-mode images (Brink and Chiu, 1994), ranged from 1 to 2.5 unit cells with an assumed c-spacing of 180 Å (Unwin, 1975). From these images one could conclude that the image resolution was limited by the pixel size of the slow-scan CCD camera.

Images with higher resolution should then be acquired using the slow-scan CCD camera at an increased magnification. Figure 3a shows a spot-scan image of a catalase crystal recorded at an effective magnification twice as high as the previous image, namely 132,000x. The total dose for this image was 12 electrons/Å2. The IQ plot in figure 3b shows that at least 4 Å resolution is attainable using the slow-scan CCD camera at this magnification. The lack of strong reflections at 6 - 8 Å resolution reflects the microscope CTF with a defocus determined to be 0.3 um. A plot of the logarithm of the reflections' amplitudes against the square of the spatial frequency shows a slope of ~45 Å2 (Fig. 4a). For comparison we analyzed 8 sets of spot-scan images of ice-embedded catalase crystals recorded close to Scherzer focus on photographic film.

Fig.4
Fig. 4. The logarithm of the Fourier amplitudes of reflections with IQ <=7 plotted against the square of the spatial frequency computed from 400 kV spot-scan images of ice-embedded catalase crystals acquired at a microscope magnification of 80,000x on (a) the slow-scan CCD camera, and at a microscope magnification of 40,000x (b) on the photographic film. The graphs represent the largest estimate of the data decay, because the unscaled amplitudes were used. The slope of the least-squares fitted line through the data (= decay factor), which represents the fall-off in the amplitudes, was measured to be ~45 Å2 in (a) and ~20 Å2 in (b). This decay factor is analogous to the temperature or B-factor used in X-ray crystallography.

An example of a plot similar to Fig. 4a is shown in Fig. 4b which has a slope of the fitted line of ~20 Å2. The fall-off observed in both plots might simply reflect the molecular transform of catalase. However, the larger fall-off that we observe in the CCD data could be caused by additional factors such as the damping effect of the slow-scan CCD camera's MTF (de Ruijter and Weiss, 1992; Mooney et al., 1993) and the microscope's envelope functions. The latter factor, however, can be ignored, because of the relatively small effects of the envelope functions due to spatial and temporal coherence for the electron-optical conditions used in the experiment. In fact, they would only account for a reduction of 3 to 5 % at 4 Å resolution at a defocus value of 0.3 um in our JEOL4000EX assuming an energy spread around 2 eV (Chiu, 1978) .

Estimate of the slow-scan CCD camera's MTF

Fig.5
Fig. 5. Modulation transfer function (MTF) of the Gatan (Model 679) 1024 x 1024 slow-scan CCD camera in our JEOL 4000EX microscope measured at 400 kV using an analysis of the autocorrelation function of a difference image similar as described by (Ishizuka, 1993). Our camera is equipped with a P43 phosphor scintillator. In addition, it has an extra fiber optic plate to filter out most of the X-ray photons generated inside the microscope's column or the camera system.

The MTF of our slow-scan CCD camera was measured in two ways. Using the analysis of the Fourier amplitude fall-off the MTF was found to decrease to 0.2 at the Nyquist frequency of the detector, namely 1/48 um-1 (Fig. 5). In an indirect way, we measured the MTF by comparing the transforms of carbon film images recorded on both the slow-scan CCD camera and photographic film. Figs. 6a-b show the computed Fourier transforms of such images. Their defocus was determined as 1.4 um underfocus from the clearly visible CTF rings. The general appearance of the Fourier transform of the CCD image showed a faster fall-off of the amplitudes towards high resolution. This was clearly evident from the radial profiles (Fig. 6c and d). All the microscope-dependent envelope functions affecting these consecutive images can be ruled out since the microscope settings did not change during the experiment. Therefore, the difference in fall-off observed here can only be attributed to a difference in MTFs between both recording media. This difference is illustrated in Fig. 7 showing the ratio of the radial profiles, i.e. the ratio of the MTFs, of data from the slow-scan CCD camera and the photographic film. The ratio fell to a value of 0.25 at the Nyquist frequency. This fall-of was found to be independent of defocus (data not shown).
Fig.6 Fig.7
Fig. 6. Computed diffraction patterns from 400 kV images of carbon film taken at 40,000x microscope magnification and ~1.4 um underfocus on (a) slow-scan CCD camera and (b) photographic film. Exposure times were identical in both cases. Both images were digitized using equivalent pixel size (24 um for CCD and 14 um for photographic film). (c - d) represent the corresponding radial profiles of computed amplitudes. The profiles were calculated using a running average along the radius and additional rotational averaging. For the image recorded on photographic film, the higher order CTF maxima were found to be larger compared to those recorded on the CCD camera. Fig. 7. Plot of the ratio of the radial profiles using the amplitudes from the images of carbon film shown in Fig. 6. As explained in the Materials and Methods, this plot represents the ratio of the MTFs of the slow-scan CCD camera and the photographic film (). The ratio has reached 0.25 at the Nyquist frequency.

Phases and SNRs from the images of negatively stained catalase crystals

Consecutive images of the same negatively stained catalase crystal were evaluated for their similarity in the computed phases. Merging phase residuals were computed which were obtained from reflections of IQs 4 and better up to a resolution of 15 Å, which was close to the Nyquist frequency (1/14.4 Å-1) of the CCD camera. The overall phase residuals ranged from 7o to 13o for all crystals imaged on either detector. Average residuals of ~ 12o were found for merging images recorded on both detectors. Typical examples of the phase residuals as a function of resolution are shown in Table 3.

Table 3. Phase residuals of images for reflections of IQ <= 4 merged from the slow-scan CCD camera and Kodak SO-163 photographic film. The residuals shown here were obtained by merging 2 images recorded on the slow-scan CCD camera (CCD data), 2 images on photographic film (film data), and 5 recorded on both CCD camera and the photographic film (CCD versus film). The number of reflections refers to the common reflections found in each resolution zone.


CCD data


film data

CCD versus film

Resolution zone
(Å)

Phase residual (deg)
No. of reflections
Phase residual (deg)
No. of reflections
Phase residual (deg)
No. of reflections
180.0 - 41.1

12.4
13
3.4
13
4.6
42
40.3 - 29.1

8.1
13
6.4
13
8.7
48
28.7 - 23.7

9.4
6
4.5
6
9.0
24
23.6 - 20.5

9.0
13
12.5
13
15.7
37
20.5 - 18.4

16.4
10
8.9
11
14.7
20
18.3 - 16.9

12.3
11
6.7
13
20.5
22
16.7 - 15.5

7.2
5
6.2
6
14.9
7
15.5 - 14.5

18.2
4
4.2
4
16.5
6

To evaluate the amount of noise that the slow-scan CCD camera adds to the image data due to the data read-out, digitization and normalization, we computed the SNR of each reflection for images of the same crystal area recorded on both detectors. If no extra noise is present, the ratio of the SNR for each reflection common to both data sets would be equal to 1. In all eleven negatively stained catalase crystal images analyzed, 30 - 40% of the reflections had an IQ value of 1 whether recorded on photographic film or on the slow-scan CCD camera. A plot of the ratio of SNRs from 4 image pairs as a function of spatial frequency is shown in figure 8.
Fig.8 Fig. 8. Plot reflecting the noise behavior of the slow-scan CCD camera relative to photographic film as function of spatial frequency. Identical areas of negatively stained catalase crystals were imaged on both the slow-scan CCD camera and photographic film. The signal-to-noise ratio (SNR) for each reflection was calculated from the computed diffraction patterns as the ratio of the reflection's amplitude and its local background. Using the reflections common to both types of images, the ratios of their SNR-values were computed and plotted. Data from 4 image pairs recorded on the slow-scan CCD camera and photographic film was used to generate this plot. All the reflections in the spectra of the image pairs were included in the graph regardless of their IQ values. The data points from reflections present in image spectra recorded on film, which were missing in those recorded using the slow-scan CCD camera are plotted on the abscissa. They constituted 5 - 10% of the total number of reflections in individual diffraction patterns. In fact, we always found that the reflections were missing from image spectra obtained using the slow-scan CCD camera. The least-squares fitted line reveals that towards the Nyquist frequency the ratio of the SNRs falls to 0.5. This means that the SNR of the data obtained using the slow-scan CCD camera is 50% of that of the photographic film data at this particular frequency.

It exhibits a linear and shallow fall-off to 0.5 at the Nyquist frequency, indicating that the CCD camera indeed increases the noise level at higher spatial frequencies. In fact, we found that occasionally reflections were absent, which occurred only in those images that were recorded on the slow-scan CCD camera. This parallels the observed decrease in the SNR in the images recorded on the slow-scan CCD camera as compared to the images recorded on photographic film. The graph displays quite a spread in the data, which reflects the sensitivity of the ratio to small errors in the measurements of the reflections' amplitude and background. If IQ ratios instead of the SNR ratios were used, this spread would not be visible (data not shown).

DISCUSSION

The choice of recording medium is important to assure an optimal and accurate signal acquisition of electron images. This is particularly true for electron crystallography of biological macromolecules which impose a stringent constraint on the electron dose for recording high resolution information. Photographic emulsions have been used as a primary recording medium. Different emulsions have been investigated in terms of their effectiveness for either diffraction or imaging purposes (Chiu and Glaeser, 1980; Glaeser, 1985) . The most thorough analysis has been carried out to characterize the MTF and noise level of different photographic films for electron imaging of beam-sensitive materials at 100 kV (Downing and Grano, 1982). With the development of electronic detectors such as slow-scan CCD cameras, which are characterized by a high linearity and sensitivity, as well as a broad dynamic range, it is logical to consider them as an alternative recording medium. In this paper we studied the characteristics of a Gatan (Model 679) 1024 x 1024 slow-scan CCD camera relevant to imaging of biological materials. Instead of determining typical parameters, such as MTF and the signal-transfer-efficiency, to describe a recording medium in a conventional way, we have taken a practical approach of using both protein crystals and an amorphous carbon film as test specimens to judge the recorder's performance in terms of resolution, MTF, SNR and phase reliability.

Attainable image resolution

In each experiment the choice of the image magnification for a certain resolution was based on the camera pixel size (Table 1). Catalase was chosen as a resolution test sample because it is readily available and crystallizes easily; its crystalline nature allows for easy assessment of resolution, and its unit cell parameters are sufficiently large to give a reasonable number of reflections even at a medium resolution. For example at 14 Å resolution ~110 unique reflections can be found. Using the computer controlled operation as outlined in Fig. 1, we demonstrated the feasibility of spot-scan imaging on the slow-scan CCD camera using catalase crystals. Figs. 2-3 showed that the image resolution of ice-embedded catalase crystals was indeed limited by the pixel size of the slow-scan CCD camera (i.e. 8 Å resolution at 67,000x and 4 Å resolution at 132,000x). A practical problem encountered in recording high magnification, spot-scan images of these catalase crystals on the slow-scan CCD camera was related to magnetic hysteresis at 400 kV. It was often difficult to keep the beam centered on the CCD chip during spot-scan imaging. This problem was overcome by a careful alignment of the microscope and by setting up a hysteresis loop for searching and imaging the crystals.

The required magnification to achieve 4 Å resolution in the images was about three times higher using the slow-scan CCD camera than used so far with Kodak SO-163 film (Brink and Chiu, 1991; Soejima et al., 1993). Normally, the electron dose would increase proportionally to the square of the magnification to maintain the same optical density in the image. However, we were able to use only 2 instead of 9 times higher dose to acquire the images at a higher magnification. The relatively smaller electron dose suggests the benefit of imaging on a slow-scan CCD camera compared to photographic film due to the virtually fog- or background-free nature of images acquired with the camera.

The 8 and 4-Å resolution data obtained with the slow-scan CCD camera on the ice-embedded catalase crystals confirmed, that image resolution was limited by the discrete data sampling by the camera. In addition it is known that this sampling combined with the scintillator's response to electrons results in a MTF which damps the amplitudes towards higher spatial frequencies (Daberkow, et al., 1991; de Ruijter and Weiss, 1992; Kujawa and Krahl, 1992; de Ruijter et al., 1993). For instance, although we have seen a decay factor of 20 Å2 for ice-embedded catalase crystals recorded on photographic film (Fig. 4b), similar images now recorded on the slow-scan CCD camera showed a larger decay factor, i.e. 45 Å2 (Figs. 4 a). This suggests a difference in the fall-off due to the recording medium itself. In addition, the discrete image sampling by the CCD camera could affect the reliability of the phases of the reflections especially at high resolution. The data obtained using both detectors on the negatively stained crystals revealed small phase residuals upon merging (Table 3). This suggests that the phase retrieval for reflections has the same reliability for both detectors. Nevertheless, it should be pointed out that the data set that we analyzed had a large percentage of high-SNR reflections. Therefore, the phase residuals for reflections with a smaller SNR might become worse.

Modulation Transfer Function

The decay plots obtained from the ice-embedded catalase crystals were the first indication of the influence of the slow-scan CCD camera's MTF on the object amplitude spectrum (Figs. 4 a-b). In order to measure more rigorously a relative effect of the MTF between the slow-scan CCD camera and photographic film, we prefer to use the same specimen area to eliminate possible factors of difference such as specimen thickness and structural motif. We have chosen amorphous carbon film as a test specimen because it is sufficiently resistant to the electron irradiation so as to allow multiple images of the same specimen area to be recorded on both detectors without changing its structure. In addition, the carbon film also mimics a non-crystalline biological specimen whose Fourier transform is continuous. In our experiment, we imaged the same area of carbon film under identical electron-optical conditions. After normalizing the electron statistics in both sets of data, the comparison of computed amplitudes revealed a pronounced difference in the data fall-off (Figs. 6 c-d). An assessment of the MTF could be made from a ratio plot of CCD and film data, since this ratio would eliminate the influence of the specimen transform, the effect of the CTF, as well as the effects of the microscope's envelope functions. This ratio of the image data is the ratio of MTFs of the slow-scan CCD camera and the photographic film () (Fig. 7).

From this plot, the actual MTF of the CCD camera could be deduced if the MTF of the photographic film was known. At 100 kV the MTF of Kodak SO-163 film falls to ~0.6 at 1/28 um-1 (Downing and Grano, 1982) , which corresponds to the Nyquist frequency for the sampling of our images recorded on photographic film. At higher voltage the MTF of photographic film is likely to improve due to a reduction in the scattering cross-section of the emulsion, as was pointed out initially by Valentine (1966) and confirmed experimentally by Frieser and Klein (1958). If, however, we assume that the photographic film's MTF at 400 kV is simply similar to that at 100 kV, then the CCD camera's MTF would be ~ 0.15 at the Nyquist frequency. The direct measurement of the MTF from the fall-off of the Fourier amplitudes in the difference image yielded a very similar value, i.e. 0.2 (Fig. 5). This confirms not only the MTF of the photographic film at the Nyquist frequency, but also corroborates the indirect measurement of the slow-scan CCD camera's MTF.

Ways of reducing MTF effects of the CCD camera on images

The obvious effect of a detector's MTF is the contrast reduction at high spatial frequencies. One way to reduce this effect would be to use even higher magnifications than those suggested in Table 1. However, the necessarily higher dose may exceed the radiation tolerance of the specimen. Another practical way to minimize the MTF effects is to use pixel binning together with a higher magnification. Binning combines adjacent pixels into a single, larger pixel (de Ruijter, 1995). However, the extent of the MTF improvement would depend on the MTF's shape. In our case the MTF was found to improve by a factor of 2 at the Nyquist frequency for 2 x 2 binning (data not shown). In this experiment the magnification had to be increased two-fold to maintain the image resolution. Since the effective pixel size on the detector increases proportionally with magnification, the dose per effective image pixel remains the same. However, at a higher magnification the specimen area imaged on the chip becomes smaller, and therefore a larger CCD chip (4096 x 4096 or even larger) would be desirable for many biological applications.

The modification of the image spectrum by the CCD camera's MTF can in principle be compensated for by proper weighting of the amplitudes by (Mooney et al., 1993; de Ruijter, 1995). Based on our SNR ratio measurement (Fig. 8), the SNRs are not significantly different between the two detectors. This means that our CCD camera does not introduce a significant amount of noise to the image. This observation suggests that for strong signals the MTF correction can safely be applied. However, for very weak signals, where the intrinsic CCD camera noise dominates the MTF-attenuated signal in the output image, it might be difficult to extract the signal reliably. To predict the lowest signal level at which the MTF correction is still reliable, one would need to know the absolute value of the inherent camera noise level. Furthermore, since the data close to the Nyquist frequency could suffer from aliasing, it is not advisable to use data beyond approximately 2/3rd of the Nyquist frequency for image reconstruction (P. Mooney, personal communication). For instance, one would use 50,000x magnification for a 15 Å targeted resolution instead of 32,000x (cf. Table 1).

CONCLUSIONS

Near-atomic resolution images of ice-embedded catalase crystals were obtained at 400 kV using a Gatan 1024 x 1024 slow-scan CCD camera. The resolution was found to be dictated by the electron microscope magnification because of the CCD camera's finite pixel size. The sampling of the data by the slow-scan CCD camera together with the scintillator's electron response at 400 kV gives rise to a decrease of the modulation transfer function (MTF) towards high spatial frequencies. This causes damping of the amplitude spectrum, which could be corrected for by applying an appropriate restoration scheme. This scheme should be governed by the signal present in the image relative to the noise generated by the slow-scan CCD camera. We found that for strong signals the extra noise from the CCD camera decreased the signal-to-noise ratio to 50% when compared to film data. In light of our observations, the current generation of the 1024 x 1024 Gatan slow-scan CCD camera is adequate for recording images of highly ordered specimen up to 4 Å resolution with our 400 kV electron cryomicroscope. The phase reliability, however, for weak signals might be slightly worse, since our experiments dealt mostly with strong reflections. For non-crystalline objects the attenuation of the amplitude spectrum by the camera's MTF could pose a problem if one aims to study these objects at high resolution. The use of binning to reduce the MTF effect would result in a considerable reduction of the specimen area imaged on the camera (i.e. 2-3 large objects per image frame). Therefore, the present slow-scan CCD camera with an array size of 1024 x 1024 pixels may not be practical for recording images of large non-crystalline objects beyond 15 Å resolution. A higher resolution structure determination would require a combination of computational data correction such as MTF restoration and hardware improvement including the electron response of the scintillator, the characteristics of the fiber optics and a larger CCD array size. It is clear that such a development is essential for the eventual replacement of the photographic film by the slow-scan CCD camera for data collection in structural biology. Finally, we would like to point out that the performance characteristics of a given slow-scan CCD camera depend on the electron voltage, as well as on the type of scintillator and the fiber optics assembly. The methods that we have described here could be readily adopted for testing new camera in any electron microscope intended for molecular biological applications.

Acknowledgments

This research has been supported by grants from the NCRR of NIH (RR02250), NSF (BR9202199) and the W. M. Keck Foundation. We thank Paul Mooney and Michael Lieber for helpful discussions throughout the investigation; Paul Mooney, Kenneth Downing, Robert M. Glaeser and Amy McGough for comments on the manuscript.

REFERENCES

Brink, J. and 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. and Chiu, W. 1994. Applications of a slow-scan CCD camera in protein electron crystallography. J. Struct. Biol., 113, 23-34.

Brink, J., Chiu, W. and 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., Sherman, M. B. and Chiu, W., 1995. Application of a slow-scan CCD camera in protein electron crystallography at 400 kV. In: Proc. 52nd Ann. Mtg. Microsc. Soc. Amer., Kansas City, MI, Bailey, G. W. and Rider, C. L., (eds.), San Francisco Press, pp. 8-9.

Chiu, W. 1978. Factors in high resolution biological structure analysis by conventional transmission electron microscopy. Scanning Electron Microscopy, 1, 569-580.

Chiu, W. and Glaeser, R. M., 1980. Evaluation of photographic emulsions for low-exposure imaging. In: Electron microscopy at molecular dimensions. State of the art and strategies for the future., Baumeister, W. and Vogell, W., (eds.), Springer-Verlag, Berlin, pp. 194-199.

Crowther, R. A., Henderson, R. and Smith, J. M. 1996. MRC Image Processing Programs. J. Struct. Biol., 116, 9-17.

Daberkow, I., Herrmann, K.-H., Liu, L. and Rau, W. D. 1991. Performance of electron image converters with YAG single-crystal screen and CCD sensor. Ultramicroscopy, 38, 215-223.

de Ruijter, W. J. 1995. Imaging properties and applications of slow-scan charge-coupled device cameras suitable for electron microscopy. Micron, 26, 247-276.

de Ruijter, W. J., Mooney, P. E. and Krivanek, O. L., 1993. Signal transfer efficiency of slow-scan CCD cameras. In: Proc. 51st Ann. Mtg. Micros. Soc. Amer., Cincinnati, OH, Bailey, G. W. and Rieder, C. L., (eds.), San Francisco Press, pp. 1062-1063.

de Ruijter, W. J. and Weiss, J. K. 1992. Methods to measure properties of slow-scan CCD cameras for electron detection. Rev. Sci. Instrum., 63, 4314-4321.

Dierksen, K., Typke, D., Hegerl, R. and Baumeister, W. 1993. Towards automatic electron tomography. II. Implementation of autofocus and low-dose procedures. Ultramicroscopy, 49, 109-120.

Dierksen, K., Typke, D., Hegerl, R., Koster, A. J. and Baumeister, W. 1992. Towards automatic electron tomography. Ultramicroscopy, 40, 71-87.

Dierksen, K., Typke, D., Hegerl, R., Walz, J., Sackmann, E. and Baumeister, W. 1995. Three-dimensional structure of lipid vesicles embedded in vitreous ice and investigated by automated electron tomography. Biophys. J., 68, 1416-1422.

Downing, K. H. and Grano, D. 1982. Analysis of photographic emulsions for electron microscopy of two-dimensional crystalline specimens. Ultramicroscopy, 7, 384-404.

Faruqi, A. R., Andrews, H. N. and Henderson, R. 1995. A high sensitivity imaging detector for electron microscopy. Nucl. Instrum. Meth. Phys. Res., A367, 408-412.

Faruqi, A. R., Andrews, H. N. and Raeburn, C. 1994. A large area cooled-CCD detector for electron microscopy. Nucl. Instrum. Meth. Phys. Res., A 348, 659-663.

Frieser, H. and Klein, E. 1958. Properties of emulsions with respect to electron irradiation. Z. Angew. Phys., 10, 337-346.

Glaeser, R. M. 1985. Electron crystallography of biological macromolecules. Ann. Rev. Phys. Chem., 36, 243-275.

Goodman, J. W., 1968. Introduction to Fourier optics. Heffner, H. and Siegman, A. E., (eds.), McGraw-Hill Book Co., San Francisco.

Henderson, R., Baldwin, J. M., Downing, K. H., Lepault, J. and 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.

Ishizuka, K. 1993. Analysis of electron image detection efficiency of slow-scan CCD camera. Ultramicroscopy, 52, 7-20.

Jeng, T. W. and Chiu, W. 1983. Low dose electron microscopy of the crotoxin complex thin crystal. J. Mol. Biol., 164, 329-346.

Jeng, T. W., Talmon, Y. and Chiu, W. 1988. Containment system for the preparation of vitrified-hydrated virus specimens. J. Electron Microsc .Tech., 8, 343-8.

Koster, A. J., Chen, H., Sedat, J. W. and Agard, D. A. 1992. Automated microscopy for electron tomography. Ultramicroscopy, 46, 207-228.

Koster, A. J. and de Ruijter, W. J. 1992. Practical autoalignment of transmission electron microscopes. Ultramicroscopy, 46, 189-197.

Krivanek, O. L. and Mooney, P. E. 1993. Applications of slow-scan CCD cameras in transmission electron microscopy. Ultramicroscopy, 49, 95-108.

Kujawa, S. and Krahl, D. 1992. Performance of a low-noise CCD camera adapted to a transmission electron microscope. Ultramicroscopy, 46, 395-403.

Kuo, I. A. M. and Glaeser, R. M. 1975. Development of methodology for low exposure, high resolution electron microscopy of biological specimens. Ultramicroscopy, 1, 53-66.

Mooney, P. E., de Ruijter, W. J. and Krivanek, O. L., 1993. MTF restoration with slow-scan CCD camera. In: Proc. 51st Ann. Mtg. Microsc. Soc. Am., Cincinnati, OH, Bailey, G. W. and Rieder, C. L., (eds.), San Francisco Press, pp. 262-263.

Mooney, P. E., Fan, G. Y., Truong, K. V., Bui, D. B. and Krivanek, O. L., 1990. Slow-scan CCD camera for transmission electron microscopy. In: Proc. XIIth Int. Cong. Electron Microsc., Seattle, WA, Peachey, L. D. and Williams, D. B., (eds.), San Francisco Press, Inc., pp. 164-165.

Mooney, P. E. and Krivanek, O. L., 1994. Image-coupling methods in CCD cameras for electron microscopy. In: Proc. 52nd Ann. Mtg. Microsc. Soc. Am., New Orleans, LA, Bailey, G. W. and Garratt-Reed, A. J., (eds.), San Francisco Press, pp. 406-407.

Pan, M. and Crozier, P. A. 1993. Quantitative imaging and diffraction of zeolites using a slow-scan CCD camera. Ulramicroscopy, 52, 487-498.

Schmid, M. F., Dargahi, R. and Tam, M. W. 1993. SPECTRA: a system for processing electron images of crystals. Ultramicroscopy, 48, 251-264.

Soejima, T., Sherman, M. B., Schmid, M. F. and Chiu, W. 1993. 4 Å projection map of bacteriophage T4 DNA helix-destabilizing protein (gp32*I) crystal by 400-kV electron cryomicroscopy. J. Struct. Biol., 111, 9-16.

Spence, J. C. H., 1988. Experimental high-resolution electron microscopy. Oxford University Press, NY.

Thomas, I. M. and Schmid, M. F. 1995. A cross-correlation method for merging electron crystallographic image data. J. Microsc. Soc. Amer., 1, 167-173.

Unwin, P. N. T. 1975. Beef liver catalase structure: Interpretation of electron micrographs. J. Mol. Biol., 98, 235-242.

Unwin, P. N. T. and Henderson, R. 1975. Molecular structure determination by electron microscopy of unstained crystalline specimens. J. Mol. Biol., 94, 425-440.

Valentine, R. C., 1966. The response of photographic emulsions to electrons. In: Advances in Optical and Electron Microscopy, Barer, R. and Cosslett, V. E., (eds.), Academic Press, New York, pp. 180-203.

Zhou, Z. H., Hardt, S., Wang, B., Sherman, M. B., Jakana, J. and Chiu, W. 1996. CTF determination of images of ice-embedded single particles using a graphics interface. J. Struct. Biol., 116, 216-223.