Dr J. G. McNally, Laboratory of Receptor Biology and Gene Expression, 41 Library, National Cancer Institute, Bethesda, MD 20892, USA. Tel.: +1 301 402 0209; e-mail: firstname.lastname@example.org
One manifestation of fluorescence resonance energy transfer (FRET) is an increase in donor fluorescence after photobleaching the acceptor. Published acceptor-photobleaching methods for FRET have mainly used wide-field microscopy. A laser scanning confocal microscope enables faster and targeted bleaching within the field of view, thereby improving speed and accuracy. Here we demonstrate the approach with CFP and YFP, the most versatile fluorescent markers now available for FRET. CFP/YFP FRET imaging has been accomplished with a single laser (argon) available on virtually all laser-scanning confocal microscopes. Accordingly, we also describe the conditions that we developed for dual imaging of CFP and YFP with the 458 and 514 argon lines. We detect FRET in a CFP/YFP fusion and also between signalling molecules (TNF-Receptor-Associated-Factors or TRAFs) that are known to homo- and heterotrimerize. Importantly, we demonstrate that appropriate controls are essential to avoid false positives in FRET by acceptor photobleaching. We use two types of negative control: (a) an internal negative control (non-bleached areas of the cell) and (b) cells with donor in the absence of the acceptor (CFP only). We find that both types of negative control can yield false FRET. Given this false FRET background, we describe a method for distinguishing true positive signals. In summary, we extensively characterize a simple approach to FRET that should be adaptable to most laser-scanning confocal microscopes, and demonstrate its feasibility for detecting FRET between several CFP/YFP partners.
There is now widespread interest in the ability to detect protein–protein interactions in fixed or living cells by using fluorescence resonance energy transfer (FRET) microscopy. Although FRET has been used in both spectrophotometry and microscopy for a number of years (Herman, 1989; Tsien et al., 1993; Clegg, 1996), it has yet to become a routine technique in microscopy studies. Part of this relates to the difficulty in labelling cells with donor–acceptor fluorophores.
Reports of CFP–YFP FRET detected by microscopy are rather limited in number considering the intense, widespread interest in the approach. This reflects the fact that at present FRET is far from a routine technique. One reason for this may be the difficulty in achieving FRET in the first place, given the requirements for proximity and orientation of the fluorophores (Clegg, 1996). An additional contributing factor, however, is the difficulty of analysing FRET reliably. The most widely used approach requires excitation of the donor and then detection of acceptor emission. This approach suffers from complications caused by bleed-through of CFP fluorescence into the YFP channel as well as inadvertent excitation of YFP during CFP excitation. These effects can be mitigated by different algebraic correction schemes, which vary in their sophistication and complexity (Gordon et al., 1998; Xia & Liu, 2001). Corrections for spectral overlap must be performed first, and therefore FRET cannot be immediately analysed.
An alternative and apparently appealing approach to detect FRET is acceptor photobleaching (Bastiaens & Jovin, 1996; Bastiaens et al., 1996; Wouters et al., 1998; Kenworthy & Edidin, 1999; Kenworthy, 2001). Its principle is that energy transfer is reduced or eliminated when the acceptor is bleached, thereby yielding an increase in donor fluorescence. Such an increase in fluorescence following bleaching is particularly diagnostic of FRET, because in most circumstances fluorescence normally decreases following a bleach. Another advantage of acceptor photobleaching is that correction issues are mitigated. Increases in donor fluorescence cannot be related to acceptor bleed-through, because the acceptor was bleached.
The acceptor photobleaching method has found application in several recent studies (Wouters et al., 1998; Kenworthy et al., 2000), including some using CFP and YFP fusion proteins (Llopis et al., 2000; Siegel et al., 2000; Day et al., 2001). These studies with YFP acceptor photobleaching have been carried out using wide-field microscopes. In this approach, acceptor bleaching has taken from 1–5 min, during which time the specimen may have moved, or in living cells the proteins of interest may have exchanged with unbleached counterparts or the cells themselves incurred damage due to prolonged bleaching. Thus, the speed of bleaching is critical, especially for live cell imaging.
Bleaching speed can be increased with lasers available on all confocal microscopes. Here we describe an approach for acceptor photobleaching using a confocal microscope to detect FRET between CFP and YFP fusion proteins. We provide a thorough characterization of the technique, and test its feasibility using three different positive controls with CFP and YFP. Importantly, we developed rigorous controls and demonstrated that they are essential for discriminating between true FRET and false-positive signals. We describe a simple approach to detect true FRET above the background of false-positive signals.
Tissue culture and transfection
HeLa cells were cultured at 37 °C under 10% CO2 in DMEM supplemented with 10% FBS, 100 µg mL−1 streptomycin sulphate and 100 U mL−1 penicillin (Invitrogen Life Technologies, Inc.). Transfections of 500 ng of the appropriate plasmids into log-phase growth Hela cells were carried out using Lipofectamine Plus reagents according to the manufacturer's protocol (Invitrogen Life Technologies, Inc). Double transfections were done at a 1 : 1 ratio. Overnight cell cultures were then fixed in 4% paraformaldehyde for 20 min and mounted on slides. ProLong antifade kit without antifade reagent was used as a mounting solution (Molecular Probes, Inc., Engene, OR, USA).
CFP and YFP were expressed from pEYFP-C1 and pECFP-C1 vectors (C-terminal enhanced fluorescent protein vectors, BD Biosciences Clontech, Palo Alto, CA, USA). The CFP-YFP fusion plasmid was constructed by inserting the BspEI-BglII YFP-containing fragment of pEYFP-C1 into the appropriate sites of the pECFP-C1 plasmid. The YFP-containing fragment was obtained by PCR using the following primers: 5′-AAG TCC GGA ATG GTG AGC AAG GGC GAG GA-3′ and 5′-TCG AGA TCT CTT GTA CAG CTC GTC CAT GAC-3′. In this fusion, CFP and YFP were separated by two amino acid residues. Human TRAF genes were amplified directly from a cDNA library and cloned into pECFP-C1 and pEYFP-C1.
We used a Zeiss LSM510 confocal microscope (Carl Zeiss. Inc, Thornwood, NY, USA) operating with a 40 mW argon laser. The laser was tuned to lines at 458, 488 and 514 nm. Cells were examined with a 100× 1.3 NA Zeiss oil immersion objective and 2× zoom. FRET was measured using the acceptor photobleaching method (Kenworthy, 2001). According to this procedure, if FRET is occurring, then photobleaching of the acceptor (YFP) should yield a significant increase in fluorescence of the donor (CFP). In our acceptor photobleaching protocol, we bleached cells in the YFP channel by scanning a region of interest (ROI) 20 times using the 514 argon laser line at 75% intensity (95 µW laser power at the specimen). Depending on the experiment, we selected anything from one to three ROIs for bleaching in each cell. The time of bleach ranged from 3 to 10 s, depending on the number and localization of bleach ROIs. Before and after this bleach, CFP images were collected to assess changes in donor fluorescence. Any increase in CFP fluorescence caused by bleaching of the YFP acceptor could be masked, at least in part, by bleaching of CFP related simply to the imaging process itself. To minimize the effect of photobleaching due to imaging, images were collected at 0.15% of the laser intensity, 500 times less than bleach intensity. To ensure that bleaching due to imaging was minimal, we monitored the level of bleaching in each experiment by collecting five CFP/YFP image pairs before the bleach and five after the bleach. Each image was collected first in the CFP channel, and then in the YFP channel. The gain of the PMTs was adjusted so as to eliminate cross-talk and to achieve the best possible dynamic range (see Results).
To calculate the FRET energy transfer efficiency as a percentage (EF) we used the formula EF = (I6–I5) × 100/I6, where In is the CFP intensity at the nth timepoint. As the bleach occurred between timepoints 5 and 6, this formula yields the increase in CFP fluorescence following a YFP bleach normalized by CFP fluorescence after the bleach. As a control, we always performed a similar calculation in non-bleached regions of the specimen: CF = (I6 − I5) × 100/I6. To evaluate measurement noise we performed a similar calculation on timepoints 4 and 5 preceding the bleaching: NF = (I5–I4) × 100/I5. Throughout the paper, we report EF, CF and NF with their standard errors of the mean. In all cases background (pixel values outside the cells) was relatively low (< 10% of the signal) (see Results). Therefore, to minimize data manipulation, background subtraction was not performed.
Dual imaging of CFP and YFP with a single laser (argon)
To optimize the imaging of CFP and YFP in our system and to eliminate cross-talk between the channels, we modified the existing filter set on the LSM510 Zeiss confocal microscope and fine-tuned the conditions of imaging. Based on the excitation spectra for CFP and YFP (Tsien, 1998), the 458 argon laser line should primarily excite CFP, the 514 line should primarily excite YFP and the 488 line should excite neither fluorophore efficiently (Fig. 1A). Accordingly, we used the 458 line to excite CFP and the 514 line to excite YFP. To permit this dual excitation, the microscope was configured with a 458/514 nm double dichroic in the excitation path (Fig. 1C). In the emission path, we inserted a 505 nm beam-splitter to help separate fluorescence from CFP and YFP. In our confocal system, the beam-splitter transmits longer wavelengths to PMT1. To improve specific detection of YFP emission at PMT1, we used a 530 nm long-pass filter (LP530, Fig. 1B and C). The 505 nm beam-splitter reflects shorter wavelengths corresponding to CFP emission to either PMT2 or PMT3. Because the light path to PMT3 is more direct, we used it for detection of CFP fluorescence. To improve specific detection of CFP emission at PMT3, we inserted a 470–500 nm band pass filter into the light path (BP470-500, Fig. 1B and C). Henceforth, we use the term ‘CFP channel’ to refer to the combination of the 458 laser-line excitation with the BP470-500 emission filter and PMT3. We use the term ‘YFP channel’ to refer to the combination of the 514 laser-line excitation with the LP530 emission filter and PMT1.
There are several reasons why the preceding configuration should significantly restrict cross-talk between CFP and YFP. First, in the CFP channel, YFP should be excited by the 458 line at < 10% of its peak excitation efficiency (Fig. 1A), and < 5% of the emitted light resulting from this marginal excitation will be transmitted by the CFP emission filter (Fig. 1B). Conversely, in the YFP channel, although as much as 15% of total CFP emission could be passed by the YFP emission filter (Fig. 1B), the YFP channel's 514 line should excite CFP at < 1% peak efficiency (Fig. 1A), and so there will be very little CFP emission at the outset. In addition to minimal cross-talk, the preceding CFP and YFP channels are also reasonably well tuned to the excitation and emission of their respective fluorophores, and so true signal should significantly exceed cross-talk signal. In particular, the YFP channel excites YFP at > 80% of its peak efficiency, and collects over 60% of the total YFP emission (Figs 1A and B). The CFP channel also excites CFP at ∼80% peak efficiency, but the narrower CFP emission filter passes only ∼25% of the total CFP emission (Fig. 1A and B).
To test these considerations for distinguishing CFP from YFP, we transfected cells with constructs encoding either CFP or YFP, or a CFP–YFP fusion containing equal amounts of CFP and YFP. For these tests, we used both the fluorescent proteins themselves, and also fusions to other proteins of interest. As expected, we found in all cases that we could detect some signal in either the CFP or YFP channel, if we set the PMT gains high enough. This bleed-through signal depended on the transfection efficiencies and also on the particular protein fusions studied. Consequently, on each day when an experiment was performed, we first established PMT gain settings that would eliminate cross-talk while preserving as much dynamic range as possible. This was done by examining the brightest cells among a population containing both CFP and YFP, and then imaging these cells with a cross-talk configuration. The CFP cross-talk channel was defined as the 514 laser line, but with CFP detection, namely the CFP emission filter and PMT3, and conversely the YFP cross-talk channel was defined as the 458 laser line, but with YFP detection, namely the YFP emission filter and PMT1. In each cross-talk channel, the appropriate PMT gain was lowered until there was no detectable signal in that channel. This ensured, for example, in the CFP cross-talk channel, that the 514 line did not lead to detectable excitation of CFP. It also ensured that any YFP fluorescence excited by the 514 line would not be detected by PMT3. A comparable analysis applies to the YFP cross-talk channel.
PMT settings established by the preceding procedure consistently yielded no cross-talk when CFP-only or YFP-only cells were imaged (Fig. 2A–D). To confirm the absence of cross-talk, we compared the mean intensity values within the cell and outside the cell. For both CFP-only cells in the YFP channel and for YFP-only cells in the CFP channel there was no difference between the mean intensity values within and outside the cell, indicating that any bleed-through fluorescence was undetectable. With the optimized PMT settings, we also achieved a good dynamic range for both fluorophores, as demonstrated by histograms of pixel intensity values (Fig. 2G and H). In addition, low signal-to-noise was never a problem. For instance, in the cell of Fig. 2(E) and (F), mean intensity was 9.3 times higher than background in the CFP channel, and 17.7 times higher than background in the YFP channel. Once these optimal gain settings were determined, they were not exceeded on that day for any of the subsequent FRET experiments.
Although we achieved a high signal-to-noise ratio for both CFP and YFP, the optimized gain settings consistently required higher gain values in the CFP channel (PMT3) compared to the YFP channel (PMT1). This resulted in noisier images for CFP than for YFP (Fig. 2E–H), even though in this control experiment and in all subsequent FRET experiments equal amounts of CFP and YFP constructs were transfected into cells. This disparity in apparent sensitivity between the CFP and YFP channels is due to several factors. First, as noted above, the CFP channel collects ∼25% of the total CFP emission, whereas the YFP channel collects ∼60% of the total YFP emission. Second, the 458 laser line used in the CFP channel is about three times weaker than the 514 laser line used in the YFP channel. Finally, even with equivalent excitation and detection, CFP has a lower quantum yield than YFP (Day et al., 2001). Despite these limitations, we have achieved reasonable signal-to-noise levels for CFP (as well as for YFP) using a variety of different fusion constructs, and conclude that detection of CFP using our current configuration would only be problematic under conditions of very low expression.
FRET by acceptor photobleaching
Negative control: CFP only Cells were transfected with a construct containing only CFP (Fig. 3). In this case, FRET should be impossible, because the acceptor fluorophore, YFP, is absent. To assess inadvertent bleaching related merely to imaging of the specimen, we always collected five pre-bleach and five post-bleach images (Fig. 3C). We consistently found some reduction in CFP intensity from one time point to the next, suggesting that some imaging bleaching was in fact occurring. To minimize the impact of bleaching due to imaging we imaged with both laser lines at 0.15% of full laser power. For the intentional bleach, cells were illuminated at selected sites by the 514 laser line at 75% of full laser power and then the change in CFP fluorescence was measured. FRET efficiency, EF, was calculated as the difference between CFP fluorescence before and after the bleach (time points 5 and 6), and then normalized by the fluorescence after the bleach (EF = (I6 − I5) × 100/I6). We found that EF varied from position to position within a cell, and also from one cell to another. This variability nonetheless yielded a smooth histogram distribution for EF values (blue bars, Fig. 3D, EF measured from three different locations in each of 10 different cells). For comparison, we selected from the same cells a total of 30 other regions that had not been bleached. For each of these unbleached regions, we performed an analogous computation to calculate a control: CF = (I6 − I5) × 100/I6. Distributions of CF (unbleached) and EF (bleached) values were similar (Fig. 3D), but the mean for CF (0.17 ± 0.19) was somewhat lower than the mean for EF (0.81 ± 0.18) (n = 30).
Given the spread of EF and CF values seen in the preceding experiment, we expanded the sample size of the negative control to determine more completely the level of fluctuation in our system. We repeated the experiments with CFP alone on three separate days, and examined several other negative controls also containing only CFP, but this time fused to a protein of interest (Fig. 4). We found that the negative controls exhibited considerable variability in the apparent FRET efficiency (Fig. 4A). In bleached regions (blue bars), the efficiencies EF ranged from −0.06 to +0.05. In unbleached regions (red bars), the control values CF ranged from –0.01 to +0.07.
As before when these data were pooled, they yielded smooth histograms for the distribution of FRET efficiencies EF or control values CF (Fig. 4B). For bleached regions (blue bars), the mean EF was −1.43 ± 0.47 (n = 290), whereas for unbleached regions the mean CF was 1.78 ± 0.23 (n = 289). Because these two values were significantly different from each other as determined by a Student's t-test (P ≤ 0.01), and because the bleached regions yielded lower values than the non-bleached regions, bleaching by the 514 laser line may lead to some bleaching of CFP. Data from the complete set of experiments with CFP (Fig. 4A) demonstrate that this effect varied from day to day and from sample to sample (compare the blue bars for example in CFP/expt2 vs. CFP/expt4). The source of this bleaching is unclear, because theoretically the 514 line should not excite and bleach CFP (Fig. 1A). As this sporadic bleaching could prevent detection of FRET, we monitored subsequent samples to detect such bleaching and when it occurred we adjusted the 514 laser line to reduce the effect.
Detection of pseudo FRET in negative controls In addition to some apparent bleaching of CFP by the 514 line, another surprise was that in non-bleached regions the mean control FRET values CF varied erratically and were on average positive. Several explanations for this effect were tested. First, we investigated whether it might arise from noise in the imaging process itself. In the simplest scenario, mean imaging noise should be zero or even slightly negative (if there is some imaging bleaching). However, other factors such as drift of the stage could introduce a positive mean, as regions of the same size measured out of focus will appear to be less bleached than the same regions in focus. For best comparison to our FRET procedure, we measured imaging noise by comparing CFP intensities in the same region at two consecutive time points right before the intentional bleach (namely time points 4 and 5). The noise was computed as NF = (I5 − I4) × 100/I5. The mean noise (NF) was close to zero, namely −0.13 ± 0.11 (n = 494), and significantly different from the mean control for FRET CF (1.78 ± 0.23) (Fig. 4C). Both of these measurements are made from unbleached regions, with the only difference being that a bleach occurred somewhere else in the specimen between time points 5 and 6 but not between time points 4 and 5.
Therefore, the positive values for the controls CF are somehow induced by the bleaching procedure itself, even though the bleach occurred elsewhere in the specimen. We also investigated whether this increase was due to imperfect acousto optical tunable filter (AOTF) performance and so tested for fluctuations in post-bleach AOTF function. However, none of these tests detected any post-bleach increase or instability in laser power.
Thus, the variability in calculated pseudo-FRET efficiencies arises from an unknown source or sources. Until these sources are identified, caution must be exercised in determining whether FRET occurs. We used the following two steps to assess FRET. First, we compared the histogram of FRET efficiencies (EF) from the actual experiment to the histogram obtained from unbleached regions (CF) of all previous negative controls (Fig. 4B, red bars). We used the unbleached regions (CF) as a baseline because their mean was positive and therefore provided a more stringent test for true FRET as compared to bleached regions (EF) of the negative controls, where the mean was negative (Fig. 4B, blue bars). This choice of baseline reduces the risk of false positives, but increases the chance that weak FRET could be overlooked.
If the preceding step showed a difference between measured FRET efficiencies (EF) and all negative controls (CF), then we asked if the negative control for the current experiment was unusual. We compared bleached (EF) and unbleached (CF) regions for the current experiment only. If FRET efficiencies for bleached regions (EF) were not significantly higher than the control calculation for unbleached regions (CF), then we concluded that FRET had not occurred, regardless of how high the FRET efficiency was for the bleached regions. This controls for the unlikely possibility that in a particular experiment an even larger pseudo-FRET occurred exceeding all ∼300 previous measurements from negative controls.
Random FRET control: CFP and YFP To evaluate the preceding strategy for FRET detection, we performed a second control. Cells were co-transfected with two separate constructs, one encoding CFP (or in one case a CFP fusion protein) and the other encoding YFP (Fig. 5). Recent studies suggest that at high enough concentrations, individual CFP and YFP molecules may dimerize (Zacharias et al., 2002). Thus, in this control we expected either no FRET or low levels of FRET due to chance dimerization. We measured FRET efficiencies (EF) from 65 bleached regions of, in this case, 14 different cells. The histogram for FRET efficiencies was not clearly different from the control CF (Fig. 5D). However, the mean EF was significantly different from the mean CF (3.27 ± 0.44 vs. 1.78 ± 0.23, P = 0.003), indicating that some FRET occurs in the CFP and YFP mix. This result suggests that our method is sensitive enough to detect FRET caused by random interactions and/or dimerization. This FRET, although true, constitutes another source of background. As a result, in future experiments we also compare the measured FRET values to that obtained for this CFP/YFP control.
Positive control no. 1: CFP-YFP fusion As the simplest positive control to assess whether our acceptor photobleaching protocol was feasible for detecting genuine FRET, we examined cells transfected with a construct containing CFP fused to YFP (CFP-YFP) (Fig. 6). In this chimeric molecule, the two fluorescent proteins are separated by a two amino acid linker, and therefore FRET should occur. For example, an analogous CFP-YFP fusion separated by a short linker (19 amino acids) has been reported to produce FRET in studies of calpain activity (Vanderklish et al., 2000).
As before, we measured FRET efficiencies (EF) from 38 bleached regions of, in this case, 14 different cells containing the CFP-YFP fusion. This distribution of EF values was compared to that for unbleached regions (CF) from all negative controls. The two distributions showed a clear difference, with the EF values shifted positively compared to the negative control (Fig. 6D), thereby providing evidence for FRET in the CFP-YFP fusion (7.96 ± 0.38 vs. 1.78 ± 0.23). As our second step to confirm FRET, we also compared the FRET efficiencies (EF) to the control values for 38 unbleached regions (CF) obtained specifically from this experiment. These two distributions also showed a clear difference with again a rightward shift of the experimental (EF) values (7.96 ± 0.38 vs. 2.21 ± 0.26). Finally, FRET for the CFP-YFP fusion was significantly higher than that measured for a mixture of CFP and YFP (7.96 ± 0.38 vs. 3.27 ± 0.44). We conclude that FRET has been detected with the CFP-YFP fusion.
Positive control no. 2: FRET in TRAF2 homotrimers The preceding results prove that bona fide FRET can be detected using acceptor photobleaching on the confocal microscope operating with different lines from a single argon laser. To determine if this method was sensitive enough to detect FRET between more distantly separated fluorophores, we tested for FRET using TRAF2, a molecule known to homotrimerize. TRAFs (TNF-Receptor-Associated Factors) are involved in signal transduction, where they function as scaffolds for cytokine receptors and kinases (review in Wajant et al., 2001). TRAFs have been purified, and homotrimerization of TRAF2 has been shown by laser light scattering, analytical ultracentrifugation and crystallography studies (McWhirter et al., 1999; Park et al., 1999; Pullen et al., 1999). TRAF2 homotrimers provide an alternative and presumably more sensitive test for FRET, because the fluorophores are not on the same molecule, as in the case of the CFP/YFP fusion.
To evaluate TRAF2 FRET, HeLa cells were transfected with two separate constructs, one encoding CFP-TRAF2 and the other encoding YFP-TRAF2 (Fig. 7). Following transfection, TRAF2 accumulated in cytoplasmic clusters exhibiting perfect colocalization of the CFP and YFP signals. In these colocalized areas, we measured FRET efficiencies (EF) from two or three bleached regions in six different cells. When the distribution of EF values was compared to the control distribution CF, we observed a significant rightward shift in the experimental values (8.48 ± 0.79 vs. 1.78 ± 0.23) (Fig. 7D). As our second step to confirm FRET, we also compared the FRET efficiencies (EF) to the control values for unbleached regions (CF) obtained specifically from this experiment (8.48 ± 0.79 vs. −0.04 ± 0.61). These two distributions also showed a clear difference, again with a rightward shift of the experimental (EF) values. Finally, FRET for CFP-TRAF2/YFP-TRAF2 was significantly higher than that measured for a mixture of CFP and YFP (8.48 ± 0.79 vs. 3.27 ± 0.44). We conclude that FRET was detected between the TRAF2 homotrimers.
Positive control no. 3: FRET between TRAF1 and TRAF2 As a further test for FRET in a physiological situation, we examined the interaction between TRAF1 and another TRAF protein, namely TRAF2. These molecules are thought to interact based on coimmunoprecipitation assays (Rothe et al., 1994; Takeuchi et al., 1996) and thereby provide a test for FRET in which the fluorophores are on different types of molecule.
To detect FRET between TRAF1-TRAF2 heterotrimers, HeLa cells were transfected with CFP-TRAF1 and YFP-TRAF2 (Fig. 8). Once again, these molecules accumulated in cytoplasmic clusters. In a fraction of cells that contained both TRAF2- and TRAF1-labelled proteins, CFP and YFP signals colocalized perfectly. In these colocalized areas, we measured FRET efficiencies (EF) from two or three bleached regions in 10 different cells. When the distribution of EF values was compared to the control distribution CF we once again observed a significant rightward shift in the experimental values (23.1 ± 0.68 vs. 1.78 ± 0.23) (Fig. 8D). As our second step to confirm FRET, we also compared the FRET efficiencies (EF) to the control values for unbleached regions (CF) obtained specifically from this experiment (23.1 ± 0.68 vs. 1.58 ± 0.84). These two distributions also showed a clear difference, again with a rightward shift of the experimental (EF) values. Finally, FRET for CFP-TRAF1/YFP-TRAF2 was significantly higher than that of a mixture of CFP and YFP (23.1 ± 0.68 vs. 3.27 ± 0.44). We conclude that FRET was detected between the TRAF1/TRAF2 heterotrimers.
FRET by acceptor photobleaching is an increasingly popular approach. However, in most studies wide-field microscopes with mercury arc lamps are used. Bleaching with an arc lamp is slow compared to a laser. Therefore, we optimized methods for FRET by acceptor photobleaching utilizing a laser confocal microscope and CFP and YFP fluorophores. This pair of fluorophores is not only well suited for FRET and widely used in protein tagging, but also may be conveniently excited by two laser lines of a single, commonly used confocal laser, namely the argon laser with laser lines 458 and 514. As a prerequisite, we first established a technique for dual imaging of CFP and YFP. We distinguished these fluorophores using the 458 nm line to excite CFP and the 514 nm line to excite YFP. Appropriate emission filters (a 470–500 nm band pass for CFP and a 530 nm long pass for YFP) and adjustment of the PMT gains to eliminate residual cross-talk were essential to discrimination. This strategy provided a simple and affordable approach for dual imaging of CFP and YFP. Other approaches, such as a separate laser to excite CFP, are considerably more expensive, but should in principle yield even less cross-talk and so permit more sensitivity in detecting low levels of CFP or YFP. Nevertheless, with our approach we have been able to obtain CFP and YFP images from a variety of cells and fusion proteins with excellent signal-to-noise ratios.
Having established a method to distinguish CFP and YFP, we then developed a procedure to detect FRET between interacting proteins tagged with these fluorophores. Using the acceptor photobleaching method on the confocal we observed FRET in three different positive controls, namely in cells containing (1) CFP fused to YFP, (2) proteins known to form a homotrimer (TRAF2-CFP and TRAF2-YFP) (McWhirter et al., 1999; Park et al., 1999; Pullen et al., 1999), and (3) proteins known to form a heterotrimer (TRAF2-YFP and TRAF1-CFP) (Rothe et al., 1994; Takeuchi et al., 1996). The ability of the confocal microscope to detect FRET in these cases demonstrates the feasibility and effectiveness of our acceptor photobleaching approach. In addition to detecting FRET in these positive controls, we were also able to detect low levels of FRET in cells transfected with both CFP alone and YFP alone. This combination yielded a FRET efficiency, which is higher than the negative control, but lower than that of positive controls. This FRET probably reflects CFP-YFP dimerization (Zacharias et al., 2002), and therefore suggests that our method is sensitive enough to detect this phenomenon.
To detect FRET reliably, we developed a more stringent set of control standards than previously applied in comparable experiments. First, we introduced an internal control for each FRET experiment. We routinely compared bleached regions to non-bleached regions in each cell. Second, we compared cells carrying both fluorophores (and thus potentially capable of FRET) to cells carrying CFP alone (and thus incapable of FRET). Our rigorous study demonstrates that, in fact, those controls are essential. By applying them we were able to discover a potential source of error –‘false FRET’. We performed over 300 negative controls using cells that contained only CFP or a CFP fusion protein, but no YFP. We found that in the absence of YFP, CFP fluorescence often increased after a bleach, even in regions of the cell that were not directly bleached. These observations undermine the widespread impression that an increase in donor fluorescence following acceptor bleaching is unlikely to occur by any route other than FRET. Particularly disturbing was the finding that this effect varied from day to day and from sample to sample. Thus, without extensive negative controls, false positives are a distinct possibility. This is especially critical if FRET is the only reliable means to establish protein interactions, and so this possibility must be assayed each time that FRET by acceptor photobleaching is performed, otherwise false FRET may be mistaken for real FRET. Therefore, we developed a system of criteria for detecting a baseline for FRET.
We suggest that the histogram of FRET efficiencies from the actual experiment should be compared to the histogram obtained from unbleached regions of the negative control (CFP alone). The unbleached regions may be used as a baseline because they potentially may exhibit pseudo FRET and therefore provide a more stringent test for true FRET. Of course, these precautions may obscure weak FRET, but this is outweighed by the benefit of discarding a pseudo FRET. We suggest that data from several negative experiments should be pooled to obtain a representative sample – the more the better, because pseudo FRET levels fluctuate from day to day. FRET should be accepted as true FRET only (a) if measured FRET efficiency in the actual experiment is significantly higher than the combined data for pseudo FRET of the negative control and (b) if measured FRET efficiency is also higher than the negative control data obtained on the same day.
The cause of pseudo FRET remains unknown. We determined that it was not due to either random noise in the imaging process or fluctuations in the AOTF decay. One clue to its origin comes from the fact that false FRET appeared not only at the site of bleaching, but also elsewhere in unbleached regions of the same cell or even in other cells. This ‘action at a distance’ rules out many possible explanations. For example, it could be that photobleaching by some as yet unknown mechanism converts YFP into a CFP-like form. However, because the cells are fixed, this photoconversion would induce only a local false FRET, not the global phenomenon observed. ‘Action-at-a-distance’ could arise via diffusion of components within the medium. Photobleaching might alter a chemical in the medium that would then diffuse and change CFP or YFP properties elsewhere, leading to an increase in fluorescence from the CFP channel. This could arise by either a chemically induced photoconversion of YFP into CFP or a dequenching of CFP. Further experiments are necessary to prove or disprove this hypothesis. Whatever the cause of this pseudo FRET, our findings underscore the need to perform extensive negative controls to characterize the contribution of the effect in FRET measurements by acceptor photobleaching.
Although we have quantified all the FRET data, we believe this quantification at present is valid only for establishing the presence or absence of FRET. Without calibration experiments, caution must be exercised in assigning significance to the absolute magnitude of FRET values. Several problems may affect the validity of quantification. First, the magnitude of false FRET varies on a day-to-day basis, and this presumably affects the magnitude of true FRET. Second, quantitative differences may be due to CFP and YFP orientation in different constructs. Third, the ratio between CFP and YFP fusions in a protein complex may affect FRET efficiency. Fourth, even if cells are always transfected with equal amounts of DNA for both constructs, fluctuations may occur in CFP/YFP ratio for individual cells and even for individual protein complexes.
Despite the need to perform a large number of negative controls to establish a baseline, the acceptor photobleaching approach on the confocal has a number of advantages. Although we developed this approach for the LSM510 Zeiss confocal microscope, the procedure should be readily adaptable to any standard confocal operating with a single argon laser, provided that appropriate filter sets are used. With this argon laser, FRET can be assessed using CFP and YFP, the green fluorescent protein variants that are currently the most practical available for FRET with fluorescent protein fusions. On a confocal, bleaching of the acceptor is rapid. We achieved complete bleaches in only a few seconds by using 20 scans of a 40 mW laser. This is much faster than the time required to bleach acceptor molecules on a wide-field microscope, where several minutes are necessary (Kenworthy et al., 2000; Llopis et al., 2000; Day et al., 2001). During this prolonged period, drift of the specimen or stage can change the region imaged, and so result in changes in donor fluorescence unrelated to the bleach. A second advantage of bleaching on the confocal is the ability to perform targeted bleaching. Regions of interest can be bleached and then compared to unbleached regions in the same field as a first negative control.
Both the speed of bleaching and the targeted bleaching are advantageous for FRET on living cells. In this study we examined fixed specimens, but the only limitation in applying the approach to live cells is that the mobility of the tagged proteins is slower than the time required to do the bleach. If the tagged protein exchanges rapidly, then there will be significant fluorescence recovery following the bleach and so any increase in donor fluorescence due to FRET will quickly disappear after the bleach. When such a scenario arises, it is of course still possible to use the acceptor photobleaching method, but the entire cell must be bleached to abolish any fluorescent recovery of the acceptor. In this latter scenario, time-lapse measurements of FRET would be impossible.
Even when time-lapse measurements are possible with acceptor photobleaching, temporal resolution will be limited by the time required to perform the bleach plus the time required for fluorescence recovery before a new bleach can be undertaken. In these situations, where multiple time-lapse measurements are required, alternative approaches such as FRET detection by direct measurement of intensity changes (Periasamy & Day, 1999; Sorkin et al., 2000; Vanderklish et al., 2000; Janetopoulos et al., 2001; Luo et al., 2001) or fluorescence lifetime changes (Gadella & Jovin, 1995; Ng et al., 1999; Elangovan et al., 2002) are more suitable. Besides very high temporal resolution, additional advantages of the lifetime imaging approach to FRET include no dependence on probe concentration, spectral cross-talk or photobleaching. These advantages are offset at present by the limited availability of lifetime systems. On the other hand, acceptor photobleaching using a confocal should be achievable on any confocal microscope equipped with an argon laser and appropriate excitation and emission filters. This approach should therefore find broad application to the measurement of FRET in fixed cells or in live cells at a single time point.
We thank Michael Stanley of Chroma Techonology Corporation for assistance in selecting the proper filters and providing filter samples for testing dual imaging of CFP and YFP.