Evaluation of intensity-based ratiometric FRET in image cytometry—Approaches and a software solution

Abstract

The intensity-based ratiometric FRET (fluorescence resonance energy transfer) method is a powerful technique for following molecular interactions in living cells. Since it is not based on irreversibly destroying the donor or the acceptor fluorophores, the time course of changes in FRET efficiency values can be monitored by this method. ImageJ, a sophisticated software tool for many types of image processing allows users to extend it with programs for various purposes. Implementing intensity-based ratiometric FRET with ImageJ vastly enhances the applicability of the FRET method. We developed an efficient ImageJ plugin, RiFRET, which calculates FRET efficiency on a pixel-by-pixel basis from ratiometric FRET images. It allows the user to correct for channel cross-talk (bleed-through) and to calculate FRET from image stacks, i.e., from 3D data sets. Semiautomatic processing for larger datasets is also included in the program. Furthermore, several options for calibrating FRET efficiency calculations were tested and their applicability to various expression systems is discussed. Although the ratiometric FRET method is widely applied, our plugin is the first freely available software for evaluating such FRET data. The program is user friendly and provides reliable, standardized results. © 2009 International Society for Advancement of Cytometry

In biological systems, fluorescence resonance energy transfer (FRET) is a widely used method for studying molecular interactions and processes (1). FRET is a nonradiative transfer of energy from an excited fluorophore, the donor, to another fluorophore, the acceptor, that are at a distance of 1–10 nm (2). The efficiency of FRET (E) depends on the inverse sixth power of the distance between the donor and the acceptor, so monitoring E can serve as a spectroscopic ruler (3). Several spectroscopic parameters (e.g., intensity, lifetime, anisotropy) can be used to determine E (1, 4–8). In flow cytometry, the intensity-based ratiometric FCET (flow cytometric FRET) method provides statistics for large cell populations. On the other hand, when subcellular details are needed, various microscopic FRET methods can be used (9–16). The ratiometric method (10), which was adapted from flow cytometry to microscopy (17), has the advantage of preserving the donor and acceptor dyes, which is necessary for following the time course of molecular interactions in living cells. Although the ratiometric approach is similar to the 3-cube FRET method (14), it uses less complicated calculations.

The achievable signal to noise ratio is usually limited by the autofluorescence of the cells. Since autofluorescence is less pronounced toward the red spectral region, the preferred donor acceptor dye pair should be red-shifted as well, such as the AlexaFluor546 - AlexaFluor647 pair with 543 nm and 633 nm laser excitations, respectively. In addition to their spectral advantages, these dyes are more photostable than their commonly used spectral analogues (18). For flow cytometric FRET, this dye pair was found to be the optimal choice (19) and it can be extrapolated that this FRET pair is also favorable in image cytometry. In many cases, AlexaFluor488 and AlexaFluor546 dyes with 488 nm and 543 nm laser excitations can be applied as well, when the expression levels of proteins labeled with antibodies conjugated to these dyes is high enough to surmount cellular autofluorecence. In the case of fluorescent proteins, such as the CFP - YFP pair, the ratiometric FRET method is also a good alternative (20). In this case, 458 nm and 514 nm laser lines can be used for excitation. Simultaneous optical separation of fluorescent proteins is discussed in (21).

Calibrating ratiometric FRET efficiency calculations is a substantial point in the evaluation of both flow and image cytometric data (22, 23). In this article, we summarize the available calculation methods of various correction factors related to spectral cross-talk between channels and expression levels. In general, donor, transfer and acceptor channels can be defined, which denote excitation/emission wavelength (band) pairs optimal for donor/donor, donor/acceptor and acceptor/acceptor, respectively. To estimate FRET efficiency, the sensitized emission intensity (corrected signal in the transfer channel) is ratioed to the quenched donor fluorescence intensity (corrected signal in the donor channel). To convert this special intensity ratio into a precise FRET efficiency value, one has to know the ratio of molar absorption coefficients of the donor and the acceptor at the donor excitation wavelength and a special factor called α which is the ratio of the fluorescence intensity of a given number of excited acceptor molecules measured in the transfer channel and fluorescence intensity of the same number of excited donor molecules detected in the donor channel [see also in more detail at Eq. (6)]. Different methods to determine these specific factors are compared here and their advantages and disadvantages are discussed.

Image analysis in medical and biological research is often performed with the ImageJ software, which is freely available, and allows the users to extend it with programs for various purposes (24). We have already successfully used ImageJ for evaluating acceptor photobleaching FRET data (25), which is a simple, self-controlled method for the calculation of FRET. However, in many instances there is a need for following the time course of molecular interactions in living cells. In response to these expanding needs, we present here an ImageJ plugin for the analysis of intensity-based ratiometric FRET images.

MATERIALS AND METHODS

Cell Culture

SK-BR-3 cells (ATCC No. HTB-30) were cultured according to specifications: DMEM with 10% FCS, 2 mM L-glutamine and 100 μg/ml gentamicin. HeLa cells (ATCC No. CCL-2) were cultured in RPMI (10% FCS, 2 mM l-glutamine and 100 μg/ml gentamicin).

Labeling Cells with Antibodies

In one set of experiments, SK-BR-3 breast cancer cells expressing high level of ErbB2 were used. Cell surface ErbB2 molecules were labeled specifically by noncompeting fluorescently tagged antibodies 2C4 (donor, AlexaFluor546) and rHu4D5 (acceptor, AlexaFluor647) binding to the same molecule but to different epitopes. Before labeling, cells cultured to semiconfluence in 8 well Lab-Tek II chambers (Fisher Scientific) were washed three times with Hepes buffer. Antibodies were used at 10 μg/ml final concentrations. After 10 min of incubation with the antibodies, cells were washed three times with Hepes buffer, and fixed with 4% formaldehyde.

Cerulean-Venus Standard Constructs

In another set of experiments, Hela cells transfected with fluorescent proteins were applied. We used constructs C5V, C17V, and C32V in which CFP (Cerulean version, donor) and YFP (Venus version, acceptor) are separated by 5, 17 or 32 amino acid long linkers, respectively. The expression plasmids were a kind gift of Steven S. Vogel (NIH, Rockville MD, USA) (23).

Transfection with Standard Constructs

HeLa cells were seeded in 8 well Lab-Tek II chambers (Fisher Scientific) and grown to 90% confluence prior to transfection. Transient transfection was performed using Lipofectamine2000 (Invitrogen) according to the manufacturer's directions. After 24 h, growth medium was changed to Hepes buffer and measurements were performed on live cells.

Confocal Microscopy

Confocal imaging was carried out with a Zeiss (Göttingen, Germany) LSM 510 confocal laser scanning microscope (CLSM) using a 40× 1.2 NA water immersion objective. In experiments using fluorescently labeled antibodies, AlexaFluor546 was excited with a 543 nm HeNe laser and detected through a 560–615 nm emission filter. AlexaFluor647 was excited with a 633 nm HeNe laser and detected through a 650 nm longpass filter. In the experiments using fluorescent proteins, Cerulean and Venus were excited with 458 and 514 nm Argon ion laser lines and detected trough 470–500 nm bandpass and 530 nm longpass filters, respectively.

Spectrophotometry

Absorption measurements of dye samples with known concentration were performed using a NanoDrop ND-1000 Spectrophotometer (Thermo Scientific) using 2 μl volumes.

The Intensity-Based Ratiometric FRET Method

To calculate the FRET efficiency E, three images are recorded for four distinctly labeled samples: nonlabeled, donor only, acceptor only, and double (donor plus acceptor) labeled. These images are 1) the donor image—excitation: donor absorption wavelength, emission: donor emission maximum; 2) the transfer image—excitation: donor absorption wavelength, emission: acceptor emission maximum; and 3) the acceptor image—excitation: acceptor absorption wavelength, emission: acceptor emission maximum. The images should be acquired line-by-line to avoid registration problems owed to movement of the sample in time. If, in spite of this, registration is necessary, it can be done by selecting the “Register to donor channel” item in the “Image” menu.

Fluorescence intensities of these images from a double labeled sample can be derived as follows.

• (1)

• (2)

• (3)

I_1(i,j), I_2(i,j), and I_3(i,j) are the detected intensities of pixel (i, j) in the donor, transfer and acceptor channels, respectively. All I values are background and autofluorescence corrected throughout. Background correction means that an average of intensities in a given area without fluorophores or cellular materials is subtracted from the image. Autofluorescence can be corrected by subtracting average background corrected intensities of an unlabeled sample in the corresponding channel. If the autofluorescence of cells is low, this subtraction can be omitted, especially when the majority of the signal originates from the cell membrane. I_D(i,j) and I_A(i,j) in the equations are the pure donor and acceptor signals. The calculation of correction factors S₁ through S₄ requires the same set of images for donor only and for acceptor only labeled samples. To use the appropriate correction factors, the same microscope setup (including laser intensities and detector gains) has to be used with the double labeled sample as with the single labeled ones.

and are correction factors calculated from donor only labeled samples, where , and are average intensities in the donor, transfer and acceptor channels, respectively.

and are correction factors calculated from acceptor only labeled samples, where , and are average intensities in the donor, transfer and acceptor channels, respectively.

FRET efficiency E_(i,j) is obtained pixel-by-pixel according to

• (4)

and

• (5)

For determining the factor α, we discuss three methods. In the simplest case (Method 1), the ratio of expression levels of donor and acceptor molecules should be the same on all cells and in each pixel of the cell used for calibration. In this event, α can be calculated from and measured on acceptor only and donor only labeled samples and averaged for at least of 5–10 images, the mean number of dye molecules attached to the donor antibody (L_D), the mean number of dye molecules attached to the acceptor antibody (L_A), the mean number of receptors per cell labeled by the donor antibody (B_D), the mean number of receptors per cell labeled by the acceptor antibody (B_A), and the molar absorption coefficients of the donor and acceptor dyes for the wavelength of donor excitation (ε_D and ε_A):

• (6)

If the expression level of the examined proteins varies from cell to cell, an antibody against a protein with uniform expression level similar to that of the examined ones, conjugated with the same dyes that are used for the FRET measurement provides the simplest approach to determine α.

When the ratio of the expression levels is not the same in all measurements, which is usually the case for proteins with independent regulation of their expression, and the fluorescent labels cannot be bound to alternative targets that are expressed at fixed ratios (such as in the case of labeling with fluorescent proteins), α can be determined by iterative approximation (26) (Method 2), using a donor-acceptor fusion protein which ensures the same expression level of the donor and the acceptor, i.e., B_D/B_A = 1. Since in such a system the visible fluorescent fusion proteins cannot be separated, the possibility of FRET between the two labels needs to be considered. In this case, factor α can be obtained with the following equation.

• (7)

In the first approximation, E = 0 is assumed, and the resulting α can be used for calculating E, which can be applied in the second iteration for the calculation of α. The optimized E and α values can be achieved in a few (3–5) iterations.

In both Method 1 and Method 2, the ε_D/ε_A ratio at the donor excitation wavelength can be determined either by spectrophotometry, or by the method described in (27). In the latter case, which might be preferred since the spectral characteristics of the dyes may change upon conjugation to an antibody, the following equation needs to be solved:

• (8)

The FRET efficiency E_bl should be determined by donor dequenching after photobleaching the acceptor using equation

• (9)

where I_1,before, I_1,after and I₂ are average (background and if necessary, autofluorescence subtracted) intensities of the donor before photobleaching the acceptor, of the donor after photobleaching the acceptor, and of the transfer images, respectively.

The third approach (Method 3) tackles the problem of different expression levels as proposed for flow cytometry based ratiometric FRET in a manner obviating the need for determining the ε_D/ε_A ratio (22). To calculate α with this approach in microscopic measurements, a set of three or more FRET pair chimeras containing the donor and acceptor at a 1:1 ratio are needed with known distinct E values.

First R_F and R_I, as described in (22), have to be calculated for each of the chimeras.

• (10)

and

• (11)

where I₁, I₂ and I₃ are average fluorescence intensities for one cell.

For determining α, R_F and R_I values should be averaged for at least 5–10 randomly chosen cells and not only for a single cell.

The aforementioned factors relate in the following manner:

• (12)

Linearization of the previous equation yields:

• (13)

where c_D and c_A are the molar concentrations for donor and acceptor.

Therefore, if is plotted against , a straight line should be the result with a slope of , and intercept . Consequently, α can be determined from dividing the slope by the intersect:

• (14)

If none of the aforementioned methods is feasible, it still should be possible to measure a reliable FRET efficiency on a control sample using the high statistical power of flow cytometry (but of course, without regard to subcellular localization). This FRET efficiency can than be used as a standard to set α of the ratiometric microscopic approach so that FRET efficiency obtained for the control matches that obtained in flow cytometry.

RESULTS

The RiFRET plugin that we developed for ImageJ can calculate intensity-based ratiometric FRET efficiencies on a pixel-by-pixel basis, providing the possibility for examining molecular interactions in cellular compartments. The use of the program requires that the method of calibration be determined and appropriate samples are prepared before acquiring the images for evaluation (see Fig. 1 for a schematic guide). In addition to single images of the donor, transfer and acceptor channels, image stacks can also be analyzed, thus FRET can be calculated easily for three dimensional structures, or for time series of two dimensional images. The program is written in Java v1.6, and tested with ImageJ version 1.42k. It can be downloaded from its homepage (28), along with some example images. It requires only a computer which can run ImageJ, and screen resolution high enough for the main window to fit in.

Figure 1. Procedure to follow for a ratiometric FRET experiment. First, the antibodies have to be chosen with the corresponding dyes. AlexaFluor546–AlexaFluor647 and AlexaFluor488– AlexaFluor546 are the commonly used dye alternatives. In the case of fluorescent proteins, CFP–YFP or GFP–RFP and their variants are most popular. Then, based on the calibration method selected, the necessary nontreated calibration samples are prepared for determining the S1–S4 and α factors. FRET is measured on double-labeled samples, usually before and after various treatments. Filters, laser intensities, detector gains, scan parameters must be the same for calibration samples and the FRET specimen. Zoom settings are best when subcellular details can be well resolved; signal intensities should be set to the center of the digitization range to allow for biological variation and similar mean values for each channel. For determining the S factors, the threshold should preferably be determined only on the high intensity channel to restrict evaluation to the meaningful pixels. As an auxiliary measure, the histogram of the so gated low intensity channel should be checked against the histogram of the background pixels.

The program leads the user through the evaluation process (see Fig. 2), which consists of setting the source images, subtracting their respective background values (and optionally the autofluorescence as well, if necessary), Gaussian blurring, thresholding for lower and upper limits, and creating the FRET image. RiFRET allows the user to manually enter the S1-S4 correction factors (see Materials and Methods) required in the FRET efficiency calculations, but also provides the option of calculating these factors from donor only and acceptor only labeled images as well. Since calculating the α factor is a crucial point of the evaluation, we addressed the possibilities in more detail, and implemented and tested three methods derived from flow and image cytometry.

Figure 2. Main window of the RiFRET plugin. At the top, the values of the correction factors are shown, followed by the steps of evaluation. Error and information messages appear in a text box at the bottom. [Color figure can be viewed in the online issue, which is available at www.inerscience.wiley.com]

We have compared the two methods for determining the ε_D/ε_A ratios on ErbB2 expressing cells using direct immunofluorescent labeling with the AlexaFluor546-AlexaFluor647 dye pair, conjugated to 2C4 and rHu4D5 antibodies, respectively. From spectrophotometry, this ε_D/ε_A value was 4.52 ± 0.41, while the acceptor photobleaching method described in (27) and evaluated using our new program yielded 4.44 ± 0.58 (averages of three independent measurements ± SD). The difference is not high, but the relatively high standard deviations indicate that it is prudent to perform at least 5–10 measurements. For obtaining the factor α, Method 1 was implemented in our program, providing options for either entering the ε_D/ε_A ratio after spectrofluorimetric determination, or calculating it from calibration images using the acceptor photobleaching method.

We have also compared the utility of Method 2 and Method 3 in determining α. The donor, transfer and acceptor channel images of fused Cerulean and Venus fluorescent proteins expressed in HeLa cells were evaluated using our program to obtain the necessary S factors and average pixel intensities. Using the iterative Method 2, α = 0.75 ± 0.03 was obtained, while fitting the data with Method 3 yielded α = 0.67 ± 0.18.

Having obtained the S and α factors, the consecutive evaluation steps will lead to creating the FRET image. In this, various types of ROIs (region of interest) can be selected and site-specific evaluation can be performed. In Figure 3, the source and FRET images with the corresponding histograms are shown with two types of fluorophores: fluorescent dyes and fluorescent proteins. For larger scale experiments, RiFRET offers a semiautomatic mode (tested with Zeiss LSM 510, AIM Version 4.0), which sequentially opens images in a given directory, and performs the steps of analysis except for thresholding, and the results are displayed as FRET images. In semiautomatic mode, the upper left corner (1/6 × 1/6 of the image) is considered as background.

Figure 3. Donor, transfer and acceptor channel pictures, and FRET map together with the corresponding FRET histograms are depicted in the case of two epitopes of ErbB2 on SK-BR-3 cells labeled with AlexaFluor546-2C4 and AlexaFluor647-rHu4D5 (row A), and HeLa cells transfected with the C5V construct (row B). The C5V construct yields very accurate and homogeneous FRET efficiency, as expected. However, the dispersion of FRET values is also quite low in the case of ErbB2 molecules on SK-BR-3 cells.

We have evaluated our FRET sample with AlexaFluor546-2C4 and AlexaFluor647-rHu4D5 antibodies both using the ratiometric method implemented in this new program and using the acceptor photobleaching method exploiting the program AccPbFRET (25). Mean FRET values ± SD (five samples) were 0.24 ± 0.01 and 0.25 ± 0.01 for the two approaches, respectively. In the case of the Cerulean–Venus C5V, C17V, and C32V constructs, we obtained 0.45 ± 0.02, 0.37 ± 0.02 and 0.29 ± 0.02 with the ratiometric method, and 0.43 ± 0.05, 0.34 ± 0.04 and 0.30 ± 0.01 with acceptor photobleaching, respectively. Data are the average ±SD of the results of 10 samples. These data correlate rather well, with a slope 0.96 of the fitted linear function and R = 0.96. Furthermore, the data also fit very well to the values published originally for these FRET reference standards (0.43 ± 0.02, 0.38 ± 0.03 and 0.31 ± 0.02) (23).

DISCUSSION

FRET can be measured in several ways both in flow cytometry and in microscopic imaging. Flow cytometric FRET can present large scale statistics in a short time, while image cytometric methods carry the advantage of providing information on association of molecules in subcellular compartments. When monitoring the time course of FRET changes in living cells, we need a method that can quickly acquire the images of each examined state while preserving the intensity of the fluorescent labels. For this type of measurement, the donor photobleaching or the acceptor photobleaching FRET methods, which are perhaps the easiest ones with relatively low equipment requirements, cannot be applied. The intensity-based ratiometric method is quite adequate for these experiments, but requires the determination of a set of bleed-through and other correction factors, and some more complicated calculations, which make this approach less feasible. However, these drawbacks can be reduced by decent software tools.

In spite of the wide range of application possibilities of ratiometric FRET imaging, our plugin RiFRET is the first freely available software for the evaluation of intensity-based ratiometric FRET images. It is capable of handling image stacks, making possible the calculation and visualization of three dimensional formations, like cells, or creating two-dimensional FRET maps as a function of time. All bleed-through corrections are included in the calculations, together with the calculation of these correction constants.

In the precise determination of FRET efficiencies, the computation of factor α is a highly important aspect. We have collected three methods for determining this factor, allowing the use of the ratiometric FRET approach in various types of systems. Method 1 can be applied when the ratio of donor and acceptor proteins is roughly the same in all measurements. Another solution here can be the use of an antibody against a protein with uniform expression level similar to that of the transfected ones, conjugated with the same dyes that are used for the FRET measurement. In the event of considerably diverse expression levels, Method 2 can approximate α using a fusion of the two fluorescent proteins of the FRET pair, which ensures the constant donor/acceptor ratio for the calculations. Method 3 is especially useful in the case of fluorescent proteins. It applies FRET pair chimeras, and contrary to the other methods, does not require the knowledge of the ratio of molar absorption coefficients, which can be a remarkable advantage.

We calculated ε_D/ε_A ratios using spectrophotometry and the acceptor photobleaching method described in (27). Our recommendation is to use acceptor photobleaching in the case of dye-labeled antibodies to obtain this ratio, and feed the result into Method 1 to estimate, α as this yields the most unambiguous results. When fluorescent proteins are used, the expression level is so widely dispersed that only Method 2 and Method 3 are applicable to the determination of α. Of these two, and if the ε_D/ε_A ratio is known, then Method 2 is preferred for its lower uncertainty.

Although intensity-based ratiometric FRET is a well established method, we compared its results obtained from the same samples with acceptor photobleaching using the AccPbFRET program (25). Based on the results, we conclude that the two methods provide coherent estimates of FRET, as opposed to the case of the donor photobleaching FRET which tends to overestimate E in comparison to ratiometric FRET (17).

Besides the easy and efficient image processing, RiFRET can speed up the evaluation even more in the semiautomatic mode. For automatic background subtraction, the upper left corner (1/6 × 1/6 of the image) cannot contain cells. At the expense of this condition, the evaluation process can be about two times faster.

Acknowledgements

The authors would like to thank Steven S. Vogel for having kindly provided the C5V, C17V, and C32V plasmids.