Photoswitching FRET to monitor protein–protein interactions

Significance Since protein–protein interactions are extremely important in understanding normal and abnormal cell behavior, cell biologists have often relied on fluorescence techniques, such FRET, as a way to monitor interactions. FRET occurs when donor fluorescent molecules tagged to a protein of interest transfer their excited-state energy to acceptor molecules tagged to another protein of interest. The need to accurately and easily measure FRET led to our development of photoswitching FRET (psFRET). It provides advantages normally associated with advanced methods, like fluorescence lifetime imaging microscopy, with the ease and accessibility of other widely used methods, such as sensitized emission or acceptor photobleaching. The psFRET approach will greatly enhance the ability of cell biologists to utilize FRET in their studies.


Fluorescence lifetime imaging to measure FRET
To provide context for our subsequent derivations of photoswitching kinetics use in monitoring energy transfer, here we present a short derivation of the use of fluorescence lifetime imaging as an approach to monitor FRET. To assist readers and reviewers checking and/or correcting our derivations throughout this SI Appendix, we have included almost every step (i.e. we have shown our work) and request feedback if errors or invalid assumptions are found. Assume a pulse of light is used to excite a population of donor molecules. This will be dependent upon the intensity of the excitation light and the excitation rate constant $ . This ignores any instrument response and considers only a population of molecules transitioning from the ground state ( ) to the excited state ( * ). The molecules can return to the ground state through several pathways. The non---radiative pathways including photobleaching are lumped together as '( , the fluorescence emission pathway is denoted by ) and the energy transfer pathway is denoted by *(+, . [ * ] = −( ) + '( + *(+, )[ * ] (1) Rearrange to combine [ * ] on the left side.
= 1 − t GH t G (16) Donor photoswitching to measure FRET Now consider a population of photoswitchable donor molecules under constant illumination. This follows the theory developed previously for photobleaching FRET 1---3 .
Here the various states are denoted by K' for switched "ON" donors in the ground state, by K' * for switched "ON" donors in the excited state, by L for photobleached donors, and by K** for switched "OFF" donors. The illumination intensity is indicated by I, the excitation rate by $ , the fluorescence emission rate by ) , the rate constants for the non--radiative pathways are combined as '( , the energy transfer rate constant is *(+, , the photobleaching rate constant is L , and the photoswitching "OFF" rate constant is K** . * also has two other depopulation pathways.
O⎯Q K** It can also be photoswitched back to the "on" state by irradiation.
For Dronpa, the donor in our experiments, photoswitching "on" requires light of a different wavelength (~400nm) than imaging and photoswitching "off" (~488nm). Therefore, this pathway is considered separate from the processes discussed below.
If we assume this system under constant illumination I, it can be described by the following two coupled differential equations.
To keep the number of terms to a manageable level, let XYY = ) + '( + *(+, + L + K** (23) Expand and substitute XYY into equation 22  =M> N ( > T @> VCC > ? @> AB @> CBDE @> T @> VCC ), ) + '( + *(+, + L + K** (45) In the absence of an acceptor, *(+, = 0 and does not contribute to depopulating the excited state, but in the presence of an acceptor *(+, is nonzero and will result in increased depopulation of the excited state. If observed on the time scale of photobleaching or photoswitching off, this will be observed as a more slowly decaying fluorescence. In Young et. al. 3 , the authors simplified this equation by noting that L is orders of magnitude slower than the other terms in the denominator. And although K** is faster than L , it is also many orders of magnitude slower than ) , '( , or *(+, . For our purposes, we note that the coefficient of t in equation 45, − $ ( > T @> VCC > ? @> AB @> CBDE @> T @> VCC ), represents a combination of the apparent photobleaching and photoswitching rate constants. In our studies, photoswitching occurs on a much faster time scale than photobleaching, so we are effectively monitoring K** . Monitoring ( ) in the presence or absence of an acceptor provides the apparent bcK** and bK** values, respectively.

Sensitized emission analysis (Donor EfD)
Here, we consider another analysis to determine the FRET efficiency from a photoswitching FRET experiment. Although measuring the photoswitching kinetics requires imaging only in the donor channel, in practice we perform what is essentially a four---cube experiment using a Dual---View image splitter. Three of these images are the normal imaging channels associated with three---cube FRET imaging, IDD (donor excitation and donor emission), IDA (donor excitation and acceptor emission), and IAA (acceptor excitation and acceptor emission) with the fourth being the rarely collected IAD (acceptor excitation and donor emission). In general, IAD is unnecessary with a proper choice of filter sets to avoid acceptor emission bleed---through into the donor emission channel, nevertheless we collect it simply due to our use of the Dual---View.
This analysis follows closely the sensitized emission approaches of Hoppe et. al. 4 and Chen et. al. 5 . We try to maintain most of the formalism of Chen et. al. 5 and refer readers to that work as a guide. Before we discuss our deviations from a normal three---cube experiment, we consider a typical sensitized emission FRET experiment in which IDD, IDA, and IAA images of a sample containing a donor and acceptor are collected. In keeping with the approach of Chen et. al. 5 , we consider all signals and crosstalk in each of these images. bb = f + g + X (52) cc = X + g + f (53) bc = g + f + X (54) In these equations, Id and Ia represent the directly excited donor and directly excited acceptor fluorescence signals, respectively. The crosstalk factors cannot generally be deduced from this one set of three images and require additional imaging of donor alone and acceptor alone samples. For example, the IDD image (equation 52) contains the donor signal, but also may contain crosstalk from the sensitized emission FRET signal (Fc) and the directly excited acceptor (Ia). A three---cube imaging experiment on the acceptor alone sample will provide quantitative information necessary for determining these crosstalks. For each of these factors, the concentration of the acceptor or the donor will impact the level of crosstalk. Therefore, the parameters are normalized based on expression level using images IDD and IAA.
The crosstalk parameter, a, describing the acceptor emission channel signal is derived from direct donor wavelength excitation of the acceptor can be determined from (IDA) and (IAA) collected on acceptor alone samples.
= bc / cc (55) The crosstalk parameter, b, describes the donor emission channel signal derived from direct excitation of the acceptor and can be determined from the images IDD and IAA of the acceptor alone. = bb / cc (56) The IAA image (equation 53) contains the acceptor excited signal (Ia), but also may contain crosstalk emission from the donor in the form of directly excited donor (Id) and sensitized emission (Fc). The crosstalk parameter, c, describes donor emission channel signal derived from direct excitation by the acceptor channel and can be determined from the images IDD and IAA of the donor alone. Similarly, the donor concentration will impact the level of crosstalk, so the parameters are normalized to (IDD), which provides the readout for the donor level. The crosstalk parameter, d, describes the FRET channel signal derived from direct donor wavelength excitation of the donor and bleed---through into the acceptor emission channel. It can be determined from (IDA) and (IDD) collected on donor alone samples.
The image IDA (equation 54) contains the sensitized emission FRET signal (Fc) in which we are interested but also has artifact signals from the donor excited donor fluorescence (Id) and the donor excited acceptor fluorescence (Ia). While crosstalk in IDD and IAA are usually small, the crosstalk in the IDA image can often be substantial depending on the fluorophore pairing. With these parameters, equations 52---54 can be modified to quantitatively reflect the relative crosstalks. bb = f + g n o + X (59) cc = X + g p q + f (60) bc = g + f + X (61) The next step in determining a FRET efficiency requires extraction of Fc from the IDA image by removing the crosstalk signals. By rearranging equations 59---61, the signals for Id, Ia, Fc are isolated.
Equations 62 and 63 can be further derived to express Id and Ia as functions of IDD, IAA, IDA and the crosstalk parameters, but these simplify considerably if the parameters b and c are approximated as 0, which is valid for properly chosen excitation and emission filters. For example, b and c in Chen et. al 5 were found to be 0.0004 and 0.0013, respectively and in our system, we found them to be 0.00447±0.00255 and ---0.00008±0.00011, respectively (mean±sem). Assuming b = 0 and c = 0, then equations 56 and 57 simplify to f = bb (65) X = cc (66) Substituting into equation 64 provides an equation for using a normal three---cube set of images and the predetermined crosstalk parameters a and d to extract the sensitized emission signal from the IDA image. g = bc − bb − cc (67) Next, we discuss our deviations from a normal three---cube experiment for extracting Fc from the image IDA in our photoswitching FRET experiments. Since the photoswitching FRET experiment requires imaging of the donor (Dronpa in our case) as it switches off, we also must measure the photoswitching kinetics of the donor in the absence of the acceptor in another sample. Therefore, we will have the sample control necessary to determine donor crosstalk into the FRET channel, which in the conventional three---cube experiment discussed earlier is the crosstalk parameter d. Bear in mind that our use of the image splitter makes convenient the imaging of the donor channel (IDD) and FRET (IDA) simultaneously, but this is not a strict requirement. For the Dronpa alone samples measured on our system, d = 0.04253±0.00044 (mean±sem).
Here we deviate from a normal 3---cube data collection. Although we do collect image IAA in our experiments and have determined a from mCherry expressing cells for some experiments, we generally do not for every experiment and therefore do not use the crosstalk parameter, a, to calculate the acceptor direct excitation crosstalk. Instead, we rely on photoswitching "off" of the donor during the experiment which decreases both the donor bleed---through and sensitized emission from the IDA image. The signal in the IDA image at the end of a photoswitching cycle provides a good approximation of the Ia crosstalk and acceptor levels.
Consider again equation 61, which describes the components of the FRET image, but this time we consider IDA at the beginning of the experiment before photoswitching off (IDA on) and after the photoswitching cycle (IDA off).
bc K' = g K' + f K' + X K' (68) bc K** = g K** + f K** + X K** (69) Before photoswitching, IDA on has the same crosstalk as a conventional three---cube experiment, so we include the crosstalk parameter d for Id off. Since the donor fluorescence is decreased, it will also decrease its probability to energy transfer and Fc off will also be decreased. Here we make the assumption that Fc off and Fc on will be proportional to donor fluorescence in the respective "off" and "on" states.
o − f K** (73) If we assume the direct excitation of the acceptor is stable.
X K** = X K' (74) Then equation 73 can be substituted into equation 68. bc If a small underestimation of Fc on and thus underestimation of the energy transfer is acceptable, we can further simplify equation 82 by assuming IDD off = 0. As mentioned earlier, the "off" state fluorescence of Dronpa is approximately 0.01371±0.00074 of the "on" so the underestimation in Fc on will be <2%. Setting IDD off = 0 gives a straightforward equation to separate the sensitized emission from the crosstalk artifacts. g K' = bc K' − bc K** − bb K' (83) In this equation, we are subtracting the direct excitation of the acceptor determined when the donor is switched off ( bc K** ) and the donor bleed---through ( bb K' ) scaled by a predetermined crosstalk parameter ( ) from the FRET channel image before photoswitching ( bc K' ).
However, just as in the three---cube experiment, the g K' value is simply the sensitized emission signal and not a FRET efficiency. As has been recounted in numerous papers 4---7 and reviews 8---10 , the best option is to convert this into a FRET efficiency which can be compared across instruments and laboratories. To do so, one must determine a G or γ factor which is specific for each imaging system and fluorophore pairing. The G factor is a ratio relating the sensitized emission signal ( g ) to the loss in donor fluorescence due to energy transfer. Again, numerous approaches have been devised to determine G, but our sensitized emission analysis uses a similar approach as Hoppe et. al. 4 and the equation and terminology from Chen et. al. 5 .
Chen et. al. 5 based their use of the G factor on work from Zal and Gascoigne 7 and relied on a determination through donor---acceptor chimeras with different length linker peptides.
= g / f + g / (84) Hoppe et. al. 4 relied on determining energy transfer of a single donor---acceptor chimera using FLIM---FRET. Thus, with an independent measurement of energy transfer, the G factor (γ in 4 ) could be determined after removal of the crosstalk signals in the FRET channel and measuring the donor channel fluorescence of the same donor---acceptor chimera. Since we can measure FRET efficiency of one of our chimeras by fitting the photoswitching kinetics and we can isolate the sensitized emission FRET signal in a straightforward manner, we rearranged equation 84 and back calculated a G factor for our instrument and donor--acceptor pairing.
Once the G factor (0.15603±0.00197) was determined for one of our donor---acceptor chimeras (D5Ch), we used it in equation 84 to determine the FRET efficiency from our measurements of the sensitized emission signals from our other chimeras and experimental samples.

Sensitized emission analysis (Acceptor EfA)
Next, we consider energy transfer from the perspective of the acceptor. Although energy transfer is generally considered and expressed as the loss from the donor, it can also be expressed as a measure of energy gained by the acceptor and thus provide a read--out of the fraction of acceptors in complex with the donor molecules. The sensitized emission signal (Fc on) described above will have the same spectrum as that of the direct acceptor excitation (Ia on) with the relationship dependent on the relative absorption of the donor and acceptor at the donor excitation wavelengths. Consider the following equation taken from Lakowicz 11 and discussed elsewhere 4,10,12 in which the energy transfer is described for sensitized emission of the acceptor. Lakowicz 11 , fD was used to express fractional labeling of the acceptor, whereas subsequent discussions 4,10,12 used fA. Equation 89 is usually discussed in limited terms since it requires that the FRET channel fluorescence be determined in the absence of the donor. Experiments in cells with these capabilities are usually rare since the absence of a donor precludes energy transfer from occurring. However, the signals measured at the end of a photoswitching cycle of a psFRET experiment closely mimic the condition of imaging the acceptor in the absence of the donor. Thus, the photoswitchable donor gives us an opportunity to directly apply such an equation. Rearranging equation 89 and substituting fA for fD, gives the following. c = [ c ( c +) ) \ (90) Note that cb ( c +) ) − c ( c +) ) represents direct acceptor excitation subtracted from the FRET channel signal in the absence of the donor and this is the same as Fc on which we determine as described previously (equation 83). As we also discussed above, we take advantage of the high contrast between the "on" and "off" states of Dronpa to approximate the direct acceptor excitation ( c ( c +) )) in the absence of the donor using the FRET channel signal after it has been switched off ( bc K** ). The relative absorptions of the donor and acceptor are given by b ( b +$ ) and c ( b +$ ), respectively. These can be determined from the spectra and extinction coefficients derived from the literature. We determined for excitation at 488 nm that b ( b +$ ) = 62600 mol ---1 cm ---1 and c ( b +$ ) = 7700 mol ---1 cm ---1 for Dronpa and mCherry, respectively. By substituting these known or measured values in equation 90, we can determine the energy transfer and fractional labeling of the acceptor. . Data represent mean ± sd (n ≥ 27). For the multi--exponential fits, weighted average rate constants were determined and substituted for bK** or bcK** as appropriate. The weighted average rate constants for double exponential fits were calculated using z * > { @g * > | z@g and triple exponential fits were calculated using z * > { @g * > | @b * > } z@g@b . Although better fits could be obtained by increasing the number of terms, the multi---exponential fits resulted in FRET efficiencies similar to the single exponential fits. We have developed ImageJ plugins for analyses of psFRET data. The plugins use Bio---Formats to extract the timestamp information automatically, otherwise user supplied time intervals are requested. These screenshots show these in use with example data and fit plots along with residuals plots. a. The psFRET_T_Profiler allows analyses of the mean pixel values in specified regions of interest over time. The fluorescence decays can be to single, double, or triple exponential equations (although we suggest limiting the fits to single or double). The data, fit, and the fit parameters are displayed in one plot and the residuals are displayed in a separate plot window. The data and fit parameters can be written to the results table for each ROI. The results table can be saved and opened in other software, such as Excel, for further analysis and calculation of FRET efficiencies. b. The Photoswitching_Pixel_Fitter plugin extracts intensity values at each pixel over a photoswitching experiment and fits that data to a single exponential with offset equation. The final output are images containing the initial signal (Azero), the rate constant, the offset, and the Chi---square at each pixel. The plugin also offers the capability to examine the pixel fits by displaying the data for a selected pixel, the fitted function for selected pixel, and the fit parameters in one plot window. A second window shows the residuals of the fit. COS7 cells expressing Dronpa or D5Ch were subjected to Dronpa photoswitching and analysis using the Photoswitching_Pixel_Fitter plugin. The Chi---square value for the fit at each pixel was used to create new images of Dronpa (a) and D5Ch (b). c. The rate constants (sec ---1 ) for Dronpa (green) and D5Ch (magenta) are displayed as histograms to show the distributions. d. The reduced Chi---square values for Dronpa (green) and D5Ch (magenta) are displayed as histograms to show the distributions. The psFRET images were processed using a 1 pixel mean filter and then analyzed using the plugin. Reduced Chi--square values for the fits at each pixel were used to create new images of Dronpa (e) and D5Ch (f). The rate constants (sec ---1 ) (g) and reduced Chi---square (h) for Dronpa (green) and D5Ch (magenta) are displayed as histograms to show the distributions. Scale bars = 10 µm. (magenta squares and blue diamonds) imaged for figure 6c (main text) and SI Appendix figure 6 were used to assess the extent of photobleaching in these experiments. The green fluorescence signals (Dronpa green fluorescence and D5Ch green fluorescence) measured in the first image immediately after photoswitching "on" with 405 nm light were averaged (n = 8 and n = 10 respectively) and normalized to the initial intensity. In our psFRET protocol, we collect a 568nm excited red fluorescence image at each cycle prior to photoswitching "on" with 405 nm light. The mean intensities of the cells in these images were averaged (n = 10) and normalized to the initial value (D5Ch red fluorescence).
SI Appendix figure S12. Dronpa photoswitching rate constant fusion protein independence. SI Appendix figure S12. Dronpa photoswitching rate constant fusion protein independence. a. COS 7 cells expressing Dronpa (green) or Ch5D (magenta) were imaged using our psFRET protocol and the photoswitching rate constants determined for each cycle. Linear fits of these data show similar slopes. Data represent mean ± sd (n = 13). b. COS 7 cells expressing Dronpa, H2B---Dronpa, Mito---Dronpa, or Vimentin---Dronpa were imaged using our psFRET protocol and the photoswitching rate constants determined for each cycle. Linear fits of these data indicate similar slopes. Data represent the means. Error bars are omitted for clarity. The standard deviations ranged from 0.015 ---0.032. (n ≥ 19 for each). a b SI Appendix figure S13. Photoswitching rate constant cycle dependence in Dronpa and other photoswitchable fluorescent proteins. SI Appendix figure S13. Photoswitching rate constant cycle dependence in Dronpa and other photoswitchable fluorescent proteins. a. COS 7 cells expressing Dronpa were imaged at three power levels (indicated on the plot) using our psFRET protocol and the photoswitching rate constants determined for each cycle. Linear fits of these data indicate that all three have similar slopes. Data represent mean ± sd (1.5mW for n = 17, n = 19 for 5.85mW, and n = 16 for 7.91mW). b. COS 7 cells expressing rsEGFP2, rsEGFP, SkylanNS, SkylanS, or Dronpa were imaged at the same illumination power using our psFRET protocol and the photoswitching rate constants determined for each cycle. Linear fits of these data show slightly positive slopes ranging from 0.001 to 0.0035 per cycle. Data represent mean ± sd (n = 18 for each protein). c. COS 7 cells expressing Dronpa, SkylanNS, rsEGFP, or rsEGFP2 were imaged using our psFRET protocol using illumination powers to closely match their photoswitching rate constants. The photoswitching rate constants were determined for each cycle followed by linear fits of these data. Data represent the means. Error bars are omitted for clarity. The standard deviations ranged from 0.012 ---0.026. (n ≥ 16 for each). The nucleosome is shown from the side (orthogonal to the view in a). The distance between the fluorescent protein chromophores was estimated using the distance tool of the Swiss PDB Viewer. Images were produced using the Swiss PDB Viewer using the nucleosome structure pdb file 1eqz and the Dronpa structure pdb file 2IE2.