Since its discovery1 in the jellyfish Aequorea victoria and the cloning of the gene,2, 3, 4 green fluorescent protein (GFP) has proven to be a valuable tool for labeling proteins: instead of labeling the proteins directly, the genetic code of GFP or of one of its variants is fused by simple cloning techniques to the sequence of the protein of interest. Upon expression of the protein, the fluorescent tag is also expressed and develops fluorescence without additional cofactors.5 As opposed to classical labeling techniques involving fluorescent dyes, proteins of interest can be labeled selectively in this way. Because of this and their ease of application, visible fluorescent proteins (VFPs) have been widely used to study the subcellular localization and trafficking of proteins. For the same reason, VFPs are also employed extensively in fluorescence resonance energy transfer (FRET) experiments to detect and trace protein-protein interactions in living cells. When the proteins of interest interact, donor and acceptor fluorophores are brought into close vicinity so that the donor can transmit its excitation energy to the acceptor. As a result, both the intensity and the lifetime of the donor fluorescence decrease, whereas the intensity of the acceptor emission increases.6 This offers several approaches to determine FRET efficiency on the stage of a fluorescence microscope: One is to detect changes in the intensity of donor and acceptor emission,7, 8 the other is to measure changes in the lifetime of the donor molecule.9, 10 One important advantage of the latter approach is that fluorescence lifetimes are independent of the actual protein concentrations. Therefore, control measurements can be obtained from different cells. Moreover the fluorescence decay offers additional information that can be used to calculate the relative fractions of free and associated donor molecules.11
As opposed to the well-established biochemical methods, microscopic techniques that exploit FRET between VFPs are able not only to detect but also to track protein-protein interactions in living cells. When and where proteins associate with each other in living cells are key questions in many biological research areas, and the preceding techniques have the power to provide answers to many of these questions. By visualizing protein-protein interactions within a cell, it might be possible to disentangle the complex network of interacting proteins and to gain an understanding of how the activity of proteins in the complex machinery of a cell is regulated by interacting proteins.
To conduct reliable measurements of the FRET efficiency between interacting proteins, however, not only a sophisticated setup but also bright and stable fluorescent labels are essential. The latter factor becomes increasingly important in experiments, where protein-protein interactions are traced over longer observation periods. Nowadays, a remarkable set of fluorescent proteins covering the entire visible spectrum is available.12 From this collection, donor and acceptor fluorophores have to be chosen such that the emission spectrum of the donor and the absorption spectrum of the acceptor overlap to a certain extent, which is a prerequisite for FRET to occur. At the same time, the overlap between the donor and acceptor emission spectra should be as low as possible to minimize the cross-talk between donor and acceptor detection channels. These requirements narrow the choice of suitable fluorescent donor-acceptor pairs. Among them, the enhanced cyan fluorescent protein (ECFP) and the enhanced yellow fluorescent protein (EYFP) are very popular.13, 14 For intensity-based FRET measurements, ECFP and EYFP can be easily excited with a mercury lamp and appropriate filter sets or with the and line of an Argon laser. With a titanium:sapphire laser, ECFP can be excited by two-photon excitation, or the second harmonic of the laser beam can be used for one-photon excitation. The high repetition rate and the short pulse width (100 to ) of this laser makes it an excellent excitation source for lifetime-based FRET measurements as well.
In this paper, we show that FRET measurements involving fluorescent proteins need to be evaluated carefully, especially when ECFP is used as a donor fluorophore, since photobleaching and/or photoconversion can bias the results considerably. We studied these effects with a streak camera system that was connected to one of the detection channels of our confocal microscope. This system allowed us to record the fluorescence decays in a spectrally resolved manner.15 To follow up changes in the emission properties of the fluorescent proteins during the course of the experiment, data were recorded in the so-called dynamic photon counting (dpc) mode. These data can be used for an offline analysis of the photon counting streak images as a function of irradiation time. Changes in both the spectra and the fluorescence decay of ECFP, which might occur upon irradiation, can be detected in this way. The results obtained for ECFP are compared with Cerulean, an ECFP variant that has been recently introduced.16
DNA Vector Preparation, Transfection, and Cell Culture
The plasmid encoding for ECFP (pECFP-C1) was obtained from Clontech (Clontech, Mountain View, California). pCerulean-C1 was designed by mutating pECFP-C1 at the following residues: S72A, Y145A, and H148D, according to Rizzo 16 The plasmid encoding the EYFP-Cerulean hybrid protein was constructed as described earlier,11 and the plasmid encoding the EYFP-ECFP hybrid protein was created in the same way. Restriction and sequencing analysis confirmed the correctness of both fusion constructs.
HEK293 cells were transfected using the conventional calcium phosphate precipitation technique and grown on coverslips. For confocal imaging and lifetime measurements, the coverslips were transferred to a microscope chamber that was superfused with phosphate buffered saline solution (PBS) containing (in ): , , , , .
Isolation of Fluorescent Proteins
Fluorescent proteins were expressed as soluble N-terminal hexa-His-tag fusions under control of an isopropyl- -D-1-thiogalactopyranoside (IPTG) inducible promoter (pET16b vector) in E. coli BL21 (DE3). After cell lysis by ultrasonication, proteins were purified by metal chelate affinity chromatography (His-trap HP, Amersham), eluted with imidazole, and dialyzed against a sodium phosphate buffer ( , , pH 7.4) and diluted to a final concentration of .
To investigate the effects of irradiation on the isolated proteins, the diluted protein solution was filled into capillaries with an inner diameter of (Hilgenberg, Malsfeld, Germany). The droplets were small enough so that they could be scanned completely by our laser scanning microscope (LSM). To prevent evaporation during the course of the experiment, the capillaries were sealed with modeling clay.
A scheme of the setup used in this study is shown in Fig. 1a . It consists of a laser scanning microscope (LSM-510, Carl Zeiss , Jena, Germany), a polychromator (250is, Chromex Inc., Albuquerque, New Mexico), and a streak camera (C5680 with a M5677 sweep unit, Hamamatsu Photonics Deutschland , Herrsching, Germany). The sample was excited by ultrashort light pulses that were generated by a mode-locked Titanium:Sapphire laser (Mira 900, Coherent GmbH, Dieburg, Germany) pumped by a , continuous-wave, frequency-doubled laser ( Verdi, Coherent). For measurements in cells, a pulse picker (Model 9200, Coherent) was used to decrease the frequency of the laser pulses to , the maximum repetition rate of the single-sweep unit of our streak camera. Isolated fluorescent proteins in capillaries were, however, exposed to the full repetition rate of the Titanium:Sapphire laser , while streak images were still acquired with a repetition rate of . In this way, we could follow up the effects of irradiation in a reasonable period of time despite the relatively large volume of the solution in the capillaries.
The second harmonic of the pulse-picked or unmodified laser beam was generated by a -barium borate (BBO) crystal. The beam was directed to the scan head of the confocal microscope. Cells were imaged with a C-Apochromat water immersion objective (NA 1.20, Zeiss); capillaries were imaged with a C-Apochromat water immersion objective (NA 0.45, Zeiss).
To measure the laser power in the focal plane, the diffuser and the sensor cell of a commercial power meter head (FieldMaster with LM2 head, Coherent) were removed and attached to a coverslip, which could be mounted in the same type of chamber we used for the confocal measurements of living cells. The active area of the sensor was large enough to collect all the light exiting the pupil of the objective. Comparison with a calibrated power meter showed that the power measurement with the modified power head was accurate.
For experiments in cells, the intensity of the laser beam was attenuated by a neutral density filter so that the power in the focal plane was . This beam was used to scan the sample continuously. The pixel dwell time was . These settings correspond to settings that are used for fluorescence lifetime imaging experiments. For experiments in capillaries, the intensity of the laser beam was adjusted with the neutral density filter so that the power in the focal plane was . The region of interest (ROI) that was chosen for bleaching had a height of 10 pixels and a length that was equal to the full width of an image (512 pixel, ). The ROI was kept constant in all experiments and always positioned in the middle of the capillary so that all experiments are comparable to each other. The droplets were small enough so that the full length of the droplet was always scanned and exposed to the Titanium:Sapphire laser beam, which excludes the possibility that the measurement was biased by axial diffusion of ECFP or Cerulean from unexposed parts of the capillaries [Fig. 3a]. The scanning speed was adjusted such that the pixel dwell time was .
In both in vivo and in vitro experiments, the fluorescence emitted by the sample was collected by the water immersion objective, fed back through the scan head, and guided to one of the confocal channels. Here, the emitted light was coupled into an optical fiber and directed to a polychromator. The polychromator spread the incident light along the horizontal axis and focused it on the entrance slit of the streak camera. The operation principle of the streak camera is explained in more detail in Fig. 1b.15, 17 In brief, incoming photons hit the photocathode, where they are converted to photoelectrons. The photoelectrons are accelerated by the accelerating electrode, pass through a pair of deflection plates, are multiplied in a multichannel plate (MCP) and hit a phosphor screen, where they are reconverted to an optical image, the so-called streak image. At the instant the photoelectrons pass through the deflection electrodes, a voltage ramp is applied so that the electrons are “swept” from top to bottom. Electrons leaving the photocathode at earlier times arrive at the phosphor screen in a position close to the top, whereas those electrons that leave the photocathode at a later time arrive at a position closer to the bottom. Thus, the vertical position of a spot depends on the time at which the photoelectron has left the photocathode; the horizontal position depends on the wavelength of the incident photon, because a spectrograph was used to focus the spectrum onto the streak-camera entrance slit. The streak images are read out by a CCD camera and transferred in intervals to a computer, where they are accumulated to yield the final photon counting streak images. In parallel, detected photons were recorded in dynamic photon counting (dpc) files, in which photons are saved with a time tag allowing for a detailed time-dependent analysis of both the spectrum and the fluorescence decays during the measurement period.
To visualize and evaluate changes, which occur during irradiation, streak images were reconstructed in time intervals from the dpc files. From these streak images, fluorescence spectra and fluorescence decays were extracted, as shown in Fig. 1b. Fluorescence decays were evaluated using the “iterative reconvolution method” and a modified Gaussian-Newtonian algorithm to approximate the parameters of a mono- or biexponential model. All steps of the analysis were performed using our own software, which was developed using MATLAB (MathWorks, Natick, Massachusetts). The goodness of the fit was judged by the value of the reduced .
Since, in most cases, a biexponential model had to be chosen to fit the data adequately, we decided to fit all data to a biexponential model to better quantify the acceleration of the fluorescence decay upon irradiation. To facilitate the direct comparison of the data, amplitude-weighted mean lifetimes were calculated from the parameters of the biexponential fit according to:and are the relative amplitudes of the fast and slow lifetime component, respectively.18 is equal to the area under the (normalized) fluorescence decay curve. It can also be regarded as a measure for the intensity one would have observed in an intensity-based FRET experiment.
The FRET efficiency is defined as the fraction of energy absorbed by the donor that is transferred to the acceptor. It is given by the ratio between the energy transfer rate and the total fluorescence decay rate of the donor. It can be calculated by comparing the fluorescence intensity or fluorescence lifetime of the donor in presence of an acceptor with the fluorescence intensity or fluorescence lifetime of the donor in absence of an acceptor:
In an ideal scenario, the donor fluorophore employed in FRET experiments exhibits a monoexponential fluorescence decay. Then, in the presence of an interaction partner labeled with an appropriate acceptor fluorophore, one would observe a biexponential fluorescence decay, in which the fast lifetime component could be attributed to the fraction of interacting donor molecules whose lifetime is quenched by FRET. The slow lifetime component would be caused by free, noninteracting donor molecules and should be close to the control values obtained in absence of the acceptor.11, 19 In this case, the FRET efficiency of a donor-acceptor pair would have been calculated by relating the fast lifetime component of the biexponential fit to the donor fluorescence lifetime obtained in control measurements.
In this study, however, we show that fluorescence lifetimes of ECFP and Cerulean decrease upon prolonged irradiation and exhibit a pronounced biexponential fluorescence decay. To better describe the fluorescence decay and quantify the area under the decay curve, we used the amplitude weighted mean lifetimes for calculating the “apparent” FRET efficiency. This approach is justified in so far as we do not investigate a mixed system of interacting and noninteracting donor molecules: Cells express either the donor molecules alone or hybrid proteins, in which donor and acceptor are covalently linked to each other. Thus, only one population of donor molecules exists. The one and only aim of this approach is to demonstrate the errors one can make when the observed decrease in the fluorescence lifetime is erroneously attributed to FRET.
Streak Camera Measurements of ECFP and Cerulean
To investigate the changes in the photophysical properties of ECFP and Cerulean induced by irradiation, we expressed ECFP and Cerulean in HEK293 cells and scanned the cells continuously on the stage of a confocal microscope with the pulse-picked and frequency-doubled laser beam . The intensity was attenuated by a neutral density filter so that the power in the focal plane was . These settings correspond to settings that are used for fluorescence lifetime imaging experiments. Figure 2b shows streak images of ECFP (upper panel) and Cerulean (lower panel) that have been obtained during an acquisition period of . In addition to the streak images, so-called dynamic photon counting (dpc) files were recorded, in which each detected photon is saved with a time tag along with its wavelength and time coordinates. This information can be used to follow up in an offline analysis the total number of emitted photons as a function of time, or to recover streak images, fluorescence spectra, or fluorescence decays in any time intervals of interest. Figures 2c and 2e show an overlay of spectra and fluorescence decays, respectively, that have been reconstructed from the dpc data in intervals. The overlay of the fluorescence spectra of ECFP [Fig. 2c, upper panel] shows that the amplitude of the fluorescence spectra decreased during the course of the experiment. Superposition of the normalized fluorescence decays [Fig. 2e, upper panel] shows that the fluorescence decay also underwent changes during the course of the experiment. It was accelerated upon prolonged irradiation. In the example shown, the mean lifetime decreased by 19% from to and further down to . In a series of eight experiments, the mean fluorescence lifetime decreased from to .
Cerulean (Fig. 2, lower row) showed a similar behavior, although the changes in the fluorescence spectra [Fig. 2c, lower row] were not so pronounced as in the ECFP spectra. Also, the fluorescence decay curves were accelerated upon prolonged irradiation. In the example shown, the mean lifetime decreased by 22% from to and further down to . In a series of nine experiments, the mean fluorescence lifetime decreased from to .
These results could be confirmed for ECFP and Cerulean that were expressed as -terminal His-tag fusions in E. coli and purified by nickel affinity chromatography. The purified proteins were diluted with PBS to a final concentration of . A small aliquot of this solution was filled into a capillary with an inner diameter of and imaged on the stage of our confocal microscope. To better observe the effects of irradiation despite the bigger volume, the sample was exposed to the full repetition rate of the Titanium:Sapphire laser. The intensity was adjusted so that the power in the focal plane was . The pixel dwell time was set to . Again, ECFP exhibited a decrease in both fluorescence intensity and fluorescence lifetime upon irradiation (Fig. 3 , upper row). Here, the mean fluorescence lifetime decreased by 23% from to after exposure. In a series of eight experiments, the mean fluorescence lifetime decreased from to .
Under the same conditions, Cerulean (Fig. 3, lower row) exhibits similar changes. In the example shown, the mean fluorescence lifetime changed by 23% from at the beginning to at the end of the experiment. In a series of nine experiments, the mean fluorescence lifetime decreased from to .
Streak Camera Measurements of Hybrid Proteins
The fluorescence decays measured in cells or in the capillary show that ECFP and Cerulean are not ideal donor molecules in experiments, where FRET efficiency is monitored over a longer observation period: In intensity-based estimates of the FRET efficiency, the decrease in fluorescence intensity would be erroneously attributed to FRET. Likewise, the decrease in the ECFP lifetime might invalidate measurements of the donor fluorescence lifetime. To demonstrate how an estimate of FRET can be biased by prolonged irradiation, we constructed an EYFP–Cerulean hybrid protein, in which the donor fluorophore (e.g., ECFP or Cerulean) and the acceptor fluorophore (e.g., EYFP) were closely linked by a short sequence of 10 amino acids (see Sec. 2.1). Figure 4b shows the streak images obtained from Hek293 cells expressing these hybrid proteins. Comparison with the respective streak images from cells expressing only the donor molecules [Fig. 2b] shows that the donor (ECFP or Cerulean) fluorescence signal in the wavelength range between 465 and is decreased, whereas the acceptor (EYFP) fluorescence between 510 and is increased. This is also demonstrated by the spectrum [Fig. 4c] and the fluorescence decays [Fig. 4e] that were reconstructed from the respective dpc files. Compared to the spectrum of ECFP or Cerulean, the emission peak of the fluorescence spectrum of the EYFP–ECFP and the EYFP–Cerulean hybrid protein was shifted toward acceptor emission wavelengths, and the fluorescence decay of the ECFP and Cerulean moiety in the hybrid protein was accelerated. Compared to ECFP, whose lifetime was determined to be in living cells, the fluorescence lifetime of the EYFP–CFP hybrid protein was accelerated to . Upon irradiation, the donor fluorescence lifetime decreased further to . In a series of seven experiments, the mean fluorescence lifetime decreased from to . When it is assumed that donor and acceptor fluorophores are separated by a single distance, a FRET efficiency of 37% can be calculated with Eq. 3 from the initial value. During the course of the experiment, however, the lifetime decreased to . Based on the latter value, a FRET efficiency of 58% would be calculated, which is beyond the maximal FRET efficiency that should be expected for an ECFP–EYFP FRET pair: GFP and its variants consist of an 11-stranded -barrel, which has the shape of a cylinder with a length of and a diameter of .14, 20 Since the fluorophore is located almost in the geometric center of the cylinder,14, 20 the minimal distance between the fluorophores that is physically possible is about . This is close to the Förster distance of for the ECFP and EYFP pair.21 Thus, with a stoichiometry of 1:1 between donor and acceptor molecules, FRET efficiencies above 50% should not be observed.
Similar results are observed for the analog hybrid-protein containing the Cerulean fluorophore: Compared to Cerulean alone, whose lifetime was determined to be in living cells, the fluorescence lifetime of the EYFP–Cerulean hybrid protein was decreased to and . In a series of eight experiments, fluorescence lifetimes decreased from to . The first value would correspond to a FRET efficiency of 35%, which is almost identical to the FRET efficiency calculated in the case of the EYFP–ECFP hybrid protein. After irradiation, a FRET efficiency of 62% would have been calculated.
When and where proteins associate with each other in living cells are key questions in many biological research projects. One way to address these questions is to measure the extent of FRET between proteins that have been labeled with appropriate donor and acceptor fluorophores. Many methods have been published to estimate the extent of FRET. Among them, fluorescence lifetime imaging (FLIM) microscopy is considered to be one of the most reliable methods for FRET measurements in living cells, since fluorescence lifetimes are insensitive to variations in the concentration of the fluorophores, fluctuations in the power of the illumination source, and alterations in the optical path.22 But the experiments presented in this paper demonstrate that special care has to be taken when ECFP and Cerulean are used as donor fluorophores. It has been already shown that fluorescence intensities of ECFP, Cerulean, and other VFPs23, 24 decrease upon one- or two-photon excitation. Here, we show also that the fluorescence lifetimes of ECFP and Cerulean decrease upon irradiation, which might seriously impede the interpretation of lifetime-based FRET experiments using ECFP or Cerulean as donor molecules. If these effects are not taken into account, the decrease in donor fluorescence lifetime might be erroneously attributed to FRET, resulting in overestimated FRET efficiencies: During the course of our experiment, ECFP bleached to an extent of more than 50%; during this period, the lifetime decreased from to . The lifetime after irradiation was close to a value we observed initially for the hybrid protein. Being unaware of this effect, one would have misinterpreted the decrease in donor fluorescence lifetime as FRET. The same applies to Cerulean, which also bleached to an extent of more than 50% during the course of the experiment and whose lifetime decreased from to . These results could be confirmed for the isolated proteins: Upon irradiation, the fluorescence lifetimes of both ECFP and Cerulean decreased.
In principle, several processes can contribute to such a decrease in fluorescence lifetime:
• Photobleaching: If ECFP and Cerulean were a mixture of two isomers of different lifetimes and if the component with the longer fluorescence lifetime photobleached more rapidly, then the overall fluorescence lifetime of the mixture of both isomers would decrease upon irradiation.
• Photoconversion: Irradiation of ECFP and Cerulean could induce a photochemical reaction of the chromophore that yielded an isomer with a shorter fluorescence lifetime.
In Cerulean, the mutation H148D was introduced to stabilize the solvent exposure of this residue (by substituting the imidazole with a carboxyl group), but surprisingly, the structural arrangement of Cerulean in this region follows closely the structure of the ECFP structure.27 The hydrophilic carboxyl group of H148D points toward the interior, whereas the hydrophobic methyl group of Y145A is exposed to the solvent. Contrary to ECFP, the indole group of Cerulean adopts a cis configuration with respect to the imidazoline ring chromophore itself. The x-ray structure of Cerulean shows only a single conformation of the chromophore and the surrounding residues. Accordingly, only one fluorescence lifetime should be observed, which is (almost) the case for the unexposed sample. However, our results show that upon irradiation, a second isomer predominates. Relating these structural details to the fluorescence properties is, however, difficult. It appears that the protein main chain plays an important role in stabilizing the chromophore structure in both the trans- or cis-configuration and might be an important factor influencing the fluorescence lifetime of ECFP and Cerulean, respectively. It might be that subtle structural changes, which cannot be resolved by the x-ray structure, have a dramatic impact on the fluorescence lifetime. On the other hand, the fact that the fluorescence lifetimes are changed irreversibly (at least on the time scale of our experiments) indicates that the energy barrier between the isomers must be at least high enough that the isomers do not interconvert at room temperature. In this light, a photoinduced reaction (e.g., photoconversion) might appear as a good explanation of the decrease of the fluorescence lifetime. But most likely both mechanisms, photobleaching and photoconversion, occur simultaneously. No matter what mechanism is involved, the fact that the shape of the fluorescence emission spectra does not change noticeably during exposure indicates that the isoforms involved have similar emission spectra.
In our experiments, ECFP and Cerulean were exposed to the same laser power. Under these conditions, ECFP and Cerulean show a considerable decrease in both the fluorescence intensity and the fluorescence lifetime. But when comparing the data, one has to keep in mind that compared to ECFP, Cerulean has a higher extinction coefficient ( versus ) and a higher quantum yield (0.62 versus 0.37). Thus, with respect to the number of photons emitted by the sample, Cerulean makes up a much better tag for both intensity- and lifetime-based FRET measurements. This, however, does not mean that it is not possible to get reasonable results with ECFP as a donor fluorophore. The comparison of the FRET efficiencies obtained from the EYFP–ECFP hybrid protein after of irradiation with the homolog Cerulean protein shows also that experiments based on ECFP lifetime measurements can yield good results. One has only to make sure that excitation time and excitation power are kept low. In the setting of a confocal microscope, this applies not only for the laser, but also for the mercury lamp. Tramier 28 have nicely shown also that illumination with a mercury lamp for less than a minute can decrease the ECFP lifetime considerably. Thus, even during the screening for transfected cells, fluorescence lifetimes can be artificially decreased. This can explain the variability observed in ECFP lifetime measurements and emphasizes the need for additional controls to verify that observed decreases in donor fluorescence lifetime are due only to FRET and not to photoconversion. Such an additional control can be achieved by spectrally resolved fluorescence lifetime measurements in a streak camera or a multiwavelength time-correlated single photon counting (TCSPC) setup, in which both the donor and the acceptor fluorescence decay are recorded simultaneously. With these techniques, it is possible to verify that the observed decrease in donor fluorescence lifetime is matched by a corresponding increase of acceptor fluorescence, which at the same time excludes that the decrease in donor lifetime is caused by other quenching processes, photobleaching, and/or photoconversion. Another useful control, which can even be performed in a conventional TCSPC setup, would be to photobleach the acceptor after the experiment, as it is shown in the example presented in Fig. 5 . If a decrease in donor fluorescence lifetime is exclusively due to FRET, then the donor lifetime must recover to control values after photobleaching of the acceptor.
We thank Karin Schoknecht for excellent technical assistance and Gerhard Rousseau for helpful hints in modifying our streak camera setup.