A procedure for a time series of three-dimensional Förster resonance energy transfer observation of a cell has been established. It employs quantitative deconvolution and three-dimensional views of intensity-modulated displays. The development was done with Raichu, a synthetic Ras protein capable of producing Förster resonance energy transfer upon its activation, for which two-dimensional imaging has been established. This method gave much clearer images of Förster resonance energy transfer than does the usual method without deconvolution, even though the signals were relatively weak. The results suggest that this procedure is compatible with weak fluorescent light, which is prone to photobleaching.
For in situ analysis of a function of protein in living cells, synthetic fluorescent proteins are widely used. A Förster resonance energy transfer (FRET, also referred to as fluorescence resonance energy transfer) system constructed with fluorescent proteins has become popular for visualizing biological events such as activation of proteins or interaction of proteins (Miyawaki et al., 1997). The technique requires two fluorescence molecules, donor and acceptor. They are specifically designed so that they can represent reactions occurring in living cells. Upon excitation, a donor molecule of the FRET system transfers its excitation energy directly to the acceptor molecule, which absorbs the energy and emits fluorescent light (Zhang et al., 2002). The energy transfer takes place only when the donor and the acceptor molecules are in close proximity (1–10 nm).
The time sequence of the three-dimensional (3D) view of the fluorescent microscopic image is useful for elucidating biological reactions that are not clear in series of conventional 2D images. FRET observations have been used mostly in studies of a series of 2D images, because the fluorescent proteins are usually sensitive to photobleaching or photochromism and do not allow researchers to obtain a sufficient number of images for a series of 3D displays with a laser confocal microscope. Even for the visualization of a single plane without haze, researchers have to attenuate the output power of the laser and have to obtain images with larger confocal pinholes, which do not give well-resolved images. Sometimes, the fluorescent protein used cannot be excited with the laser provided as a standard option for the confocal microscope.
Deconvolution of conventional fluorescent microscopic images can give a series of 3D images because it does not require intense excitation, unlike confocal microscopy. The reason for this difference is that the deconvolution system provided with the conventional wide-field microscope utilizes 100% of the signal coming through the objective lens, whereas the laser confocal microscope with pinholes usually employs less than 1% of the signal coming through the objective lens. However, some algorithms are not quantitative and so cannot be used for FRET analysis. In addition, the background signal makes a great contribution to the mathematical correctness of the deconvolution process.
We found a background subtraction method which eliminates the contribution of the background signal. We here report on a new method for visualizing a time sequence of 3D images of FRET in a cell.
Materials and methods
Cells of a COS7 subclone, which exhibit efficient membrane ruffling after epidermal growth factor treatment, were cultured in Dulbecco's modified minimal essential medium supplemented with 5% fetal bovine serum (Sawano et al., 2002).
Raichu-Ras of Raichu (Ras and Interacting Protein Chimeric Unit) was used for FRET analysis (Mochizuki et al., 2001). The protein contains two fluorescent proteins, cyan fluorescence protein (CFP) and yellow fluorescence protein (YFP). The protein is constructed so that the CFP and the YFP are placed at opposite ends of a linear molecule containing Ras and its counterpart. Upon activation of Ras, the Ras binds to Raf, and the resulting molecule is bent in the middle. This causes the CFP and the YFP to come closer to each other, and thus the molecule exhibits the FRET phenomenon. In other words, the fluorescence protein exhibits FRET when Ras is activated.
An expression vector for Raichu-Ras was introduced into Cos7 cells by electroporation, and the cells were cultured on a glass-bottomed plate (30 mm diameter). After 24 h, the cells were used for the FRET analysis. Cells were stimulated with 20 µg mL−1 epidermal growth factor.
System for FRET observation
The optical setup is summarized in Fig. 1. The system contains a Zeiss Axiovert 100 microscope with a ZEISS Plan Neofluar 40 × 1.3 Oil and a ZEISS Plan Apochromat 63 × 1.4 Oil DIC objective lens. The temperature was controlled by use of a custom-made enclosure and a thermostatic controller with heater. The temperature was set to 37 °C.
A XF88-2 FRET filter (Omega Optical Inc., Brattleboro, VT) set was used. Specifically, 440A21 for CFP (donor) excitation, 480AF30 for CFP (donor) emission, 535AF26 for YFP (acceptor) and FRET emission and a dichroic mirror 455DRLP were the filters used in this study. Neutral density filters with 3–12% transparency were used for attenuating the excitation light.
The focusing device was a P-721.10 piezoelectric focusing device (Physik Instrumente, Palmbach, Germany) with an E-662 controller. The device enables acquisition of 3D stack images moving the lens.
Two wheels (Ludl Electronic Products, Williston, VT), each accommodating up to six filters, were used: one for excitation and another for emission. The emission filter wheel was placed adjacent to the CCD camera (Quantix; Photometrics, Tucson, AZ). The camera was used with 4 × 4 binning.
Scanalytics’iplab for Windows was used as software, which is run with Microsoft Windows 98 and Microsoft Windows 2000 and used for acquisition, calculation, and analysis of the FRET images. This software was used to control the focusing device, the filter wheels, and the CCD camera. Although the time course experiment was set manually for the present study, the software has the capability of acquiring time-lapse 3D stack image series automatically (see Discussion). The epr system (Scanalytics Inc., Fairfax, VA) was used for deconvolution of the images.
Fluorescent beads for point spread function
The deconvolution program epr requires an optical reference image, the point spread function (PSF). The PSF shows how a point light source is blurred by the optical system used for imaging. Fluorescent beads, FC102/112 (Bangs Laboratories, Inc, Fishers, IN), were used to obtain the PSF. The diameters of the beads were about 0.2 µm, which is smaller than the limit of optical resolution. They have a relatively wide range of emission, and images can be taken at both 480 nm and 535 nm. The excitation maximum of the beads was at 424 nm and the emission maximum at 480 nm.
Acquisition of cell images. Acquisition of optical slices covering the entire cell thickness with a Z interval of 0.5 µm was performed. The excitation wavelength was 440 nm. Two sets of images were taken by switching of the filters by use of an emission filter wheel (wavelengths at 480 nm and 535 nm). The exposure times for the two sets were adjusted so that roughly equal levels of the signals could be detected. The Z interval, number of optical slices, XYZ coordinates of the image, and emission and excitation wavelengths, were kept the same throughout the process of imaging of a cell. Damage to the cells from exposure to the excitation beam was minimized by closing of the shutter.
After selection of appropriate cells visualized with the excitation wavelength of CFP (440 nm), under the microscope the cells were kept away from the excitation light with a shutter for 5 min to allow the YFP to recover from possible photochromism. After the first set of images, cells were stimulated with epidermal growth factor (100 ng mL−1). After 4, 10 and 15 min, images were taken. Typically, cells showed membrane ruffling at from 3 to 10 min after stimulation. Sometimes the membrane ruffling lasted for 30 min.
Acquisition of PSF. The PSF was obtained under the same conditions as those used for acquisition of cell images.
Background subtraction. The imaging process creates a series of image stacks obtained at different times after stimulation. Each image stack contained multiple Z optical slices. A small area that may represent the lowest background level in the plane was selected as background area. The background level was subtracted from the stacked image with a custom-made script of iplab for windows. The script enabled us to subtract an average of the background of the designated area in each plane (optical slice) from the corresponding plane. The background was subtracted from the PSF in the same way.
After subtraction of the backgrounds, the image was deconvolved with epr software. The parameters used were: convergence 10E-4; smoothing 13; maximum iteration 100; scaling factor ‘User-defined’ 1. The deconvolution software, epr, is accepted as giving quantitative results. The epr algorithm estimates the unknown dye density, f, as the non-negative density that minimizes the expression
where blur(f)(i,j,k) is the value at voxel(i,j,k) of the function (f)(i,j,k) computationally blurred by the PSF (Kenneth et al., 1993). The parameter α determines the smoothness setting of the restored image and is set at a level that avoids noise and graininess in the image (Kenneth et al., 1993). The algorithm is for a constrained iterative deconvolution with non-negativity restriction that was proved to be quantitative (Fay et al., 1989; Isenberg et al., 1996). The name epr was given for the commercial release of the software by Scanalytics.
After epr calculation, a series of time-resolved images was converted to stacks in one 4D file. The image contained dimensions of X, Y, Z, and time. Two image stacks were produced: one for the image taken at 480 nm, and one for that at 535 nm. The former corresponds to the CFP (donor) emission channel and the latter to the YFP-FRET (acceptor) emission channel. Using the time point and Z-plane in which the cell shows the most apparent membrane ruffling, we adjusted the contrast of the image. The minimum intensity of the fluorescence in the selected plane was set to pure black and the maximum to pure white. The YFP-FRET channel image was divided by the CFP channel image. In the FRET phenomenon, energy is transferred from CFP to YFP. This means that the fluorescence of CFP is decreased and the fluorescence of YFP is increased. The image of the division is a way of clearly indicating the reduction of the fluorescence energy in the CFP channel and the fluorescence energy increase in the YFP-FRET channel. Therefore, the ratio image (the image of the division) shows the FRET efficiency (Mochizuki et al., 2001). The ratio of intensity values of the peak in the YFP-FRET channel image to that in the CFP channel was calculated; this represents the exposure factor (Fig. 2). The ratio image was multiplied by the exposure factor to adjust the intensity difference between the two channels. The corrected ratio image was pseudocoloured so that red represented the highest ratio and blue represented the lowest ratio. With the pseudocoloured image and the YFP-FRET channel image, an intensity-modulated display (IMD) image was created. A 3D time series image was created using the IMD image.
To achieve 3D FRET, one must consider the chromatic aberration between the CFP channel and the YFP-FRET channel. Because images are taken every 0.5 µm, the chromatic aberration in the Z-axis of 0.2 µm may affect the result for the ratio between the fluorescence of CFP and that of YFP. We screened for the lenses that avoid this problem. An image of fluorescent beads, FC102/112, was obtained at 37 °C. After subtraction of the backgrounds, the intensity in each plane was measured. Among various objective lenses, the ZEISS Plan Apochromat 63 × 1.4 Oil DIC gave the most satisfactory results (Fig. 3a). The peaks of the CFP and YFP fluorescence conformed well to each other. After deconvolution with another bead image (PSF), the two curves also corresponded well (Fig. 3b). This lens was used for FRET imaging.
Cos7 cells expressing Raichu-Ras were stimulated with epidermal growth factor. After 5 min, most of the cells exhibited membrane ruffling. According to a previous study, FRET should be observed at the membrane ruffles (Mochizuki et al., 2001). By use of the data without deconvolution, FRET image processing was performed. As shown in Fig. 4(a), FRET was observed at the membrane ruffles in the IMD of the 2D image. Before the stimulation, FRET was not seen in any part of the cell (data not shown for the image before deconvolution). The IMD shows efficiency of the FRET in colour and intensity. In short, it displays the amount and efficiency of Ras activation. However, when 3D construction was done, the resolution of the Z-axis was not sufficient for analysis of the cells (Fig. 4b).
To clarify, we performed deconvolution on the same image. After deconvolution, a much clearer IMD was obtained (Fig. 5a). The 3D configuration successfully gave fine resolution on the Z-axis (Fig. 5b). Figure 5(c) shows the time-resolved 3D FRET image display rotated in the X-direction. The display shows that the FRET is observed at the membrane ruffles at 4 min after the stimulation and declines as the membrane ruffling comes to an end. A side view (bottom) indicates that FRET and the ruffling occurred not entirely, but almost, in the same position.
Because data processing of the FRET image includes calculation of image ratio, the images should be quantitative by means of representing the correct signal intensity of the fluorescence. To achieve this, we selected the following equipment. A cooled CCD camera, Quantix, is designed to represent the linear response to the intensity of the imaged object. The linearity extends up to the full 12-bit data scale of the camera. Thus, the acquired image is expected to exhibit a linear relationship to the intensity of the fluorescent light. The imaging and analysis software, iplab for windows, handles the image so that the intensity of each pixel is preserved even if the image is contrast-adjusted or pseudocoloured.
The backgrounds of the image stacks were found to be slightly different in the different planes. Thus, correct background subtraction before epr treatment is an important step for preserving the integrity of the data. The integrity of the data is important for the FRET experiment, because the experiment involves the step of taking an image ratio. We first tried to use a minimum of each plane's designated ROI (region of interest) and subtracted it from all the Z-planes. Although this method preserved the detail of the image best, it did not eliminate the differences in the background among planes. It gave an erroneous convergence in the deconvolution process (Fig. 3c). The Z-axis intensity plots of a fluorescent bead taken at the CFP and YFP-FRET channels showed differences after the deconvolution. The preferred method is subtracting the average of each plane's ROI from the corresponding plane. This method eliminates the difference in the backgrounds among planes and was found empirically to be effective. The method preserved most of the detail of the image and gave correct convergence in the deconvolution process (Fig. 3b).
The IMD image (Fig. 5a) gives information on both the intensity of CFP and the ratio of YFP to CFP. The intensity reflects the amount of Raichu protein existing in specific regions of a cell. The ratio reflects the degree of FRET that occurred at the specific regions. In other words, in this study, the IMD showed both the amount and the activity of Ras protein in a cell. The display is also compatible with a 3D view, because it contains intensity information that can be used for creating the view. Figure 5(c) shows that FRET caused by Ras activation was most intense at 4 min after the stimulation and gradually faded out over the next 46 min. The maximum stimulation occurred in the area where membrane ruffles were seen, and the area was not limited to the cell border attached to the glass.
The detailed time-lapse 3D image of FRET (Fig. 5c) enables one to obtain information that was not available before. Compared to the method described here, an alternative method, such as using a spinning disc confocal microscope, requires intense laser excitation even with a high-sensitivity camera, which might result in photobleaching or photochromism of the system. An image intensifier improves the sensitivity, but the image quality is compromised, and it is questionable whether it should be used with the FRET observation system, which requires ratio calculation between high-resolution images. In addition to this, use of the Coolsnap-HQ camera (Roper Scientific, Trenton, NJ) instead of Quantix raised the sensitivity, which allowed us to use a shorter exposure time. Now FRET images can be obtained every 2 min. The maximum intensity can be as low as 300 grey levels of the 12-bit camera, and the epr program gives images sufficient for ratio analysis because the noise level is very low.
We thank Mr Fernando Delaville (Scanalytics Inc., Fairfax, VA) and Dr Michiyuki Matsuda (University of Osaka, Osaka) for critical reading of the manuscript.