Spectroscopic approach for monitoring two-photon excited ﬂuorescence resonance energy transfer from homodimers at the subcellular level

. We have employed a spectroscopic approach for monitoring ﬂuorescence resonance energy transfer (FRET) in living cells. This method provides excellent spectral separation of green ﬂuorescent protein (GFP) mutant signals within a subcellular imaging volume using two-photon excited ﬂuorescence imaging and spectroscopy (TPIS-FRET). In contrast to current FRET-based methodologies, TPIS-FRET does not rely on the selection of optical ﬁlters, ratiometric image analysis, or bleedthrough correction algorithms. Utilizing the intrinsic optical sectioning capabilities of TPIS-FRET, we have identiﬁed protein–protein interactions within discrete subcellular domains. To illustrate the applicability of this technique to the detection of ho-modimer formation, we demonstrated the in vivo association of promyleocyte (PML) homodimers within their corresponding nuclear body. nm and 501/476 nm, re- spectively. The ratio values for ten cells expressing the donor and the acceptor (PMLCFP/PMLYFP) were 0.86 6 0.02 and 0.95 6 0.01, respec- tively. The ratios are independent of expression levels from cell to cell. An unpaired Student’s t-test analysis for the 501/476-nm ratios was not statistically signiﬁcant (p 5 0.773), reﬂecting the consistency of this ratio for donor (CFP) ﬂuorescence. An unpaired Student’s t-test analysis of the 527/476-nm ratios was highly signiﬁcant (p , 0.0001), indicating a statistical difference between donor and donor–acceptor cells, revealing the occurrence of FRET.


Introduction
A myriad of proteins found to interact in vitro has raised the issue of demonstrating the occurrence of these direct interactions in vivo and more specifically, their spatial localization within discrete cellular domains. The advent of green fluorescent protein ͑GFP͒ fluorophores has prompted the utilization of fluorescence resonance energy transfer ͑FRET͒ to monitor protein-protein interaction. [1][2][3] In FRET, direct excitation of a donor molecule results in nonradiative energy transfer to an acceptor molecule, provided the molecules are within approximately 10 to 100 Å of each other. 4,5 For efficient FRET, the emission spectrum of the donor must sufficiently overlap with the excitation spectrum of the acceptor molecule. While traditionally FRET has been used to calculate intra-and intermolecular distances between two sites, 3,6 determination of the occurrence of a protein-protein interaction relies mainly on whether FRET is observed, since this process occurs within the Förster distance. 7 Recent studies employing spectrally overlapping GFP mutants and FRET methods to identify protein-protein interactions have relied extensively on utilization of optical filters to separate the donor's fluorescence from that of the acceptor's fluorescence followed by intensity-based ratiometric imaging. 2,8 Major obstacles have been encountered with these approaches and are not limited to distinguishing the FRET signal from simple donor and acceptor fluorescence spectral overlap and accounting for differences in expression levels of the donor and acceptor molecules. This has introduced ambiguities in the interpretation of a FRET signal 2,3 and has largely limited the utilization of FRET to monitoring inducible cellular protein interactions. [9][10][11] Here, the donor intensity image in the absence of FRET is obtained from the sample to perform intensity-based ratiometric image analysis. A ratiometric measurement requires approximately equal levels of donor and acceptor expression from cell to cell, typically relying on the creation of multiple, stably transfected cell lines, careful selection of optical filters, and the development of advanced algorithms to minimize spectral bleedthrough. Alternative approaches to quantitatively monitoring FRET have been based on fluorescence lifetime imaging and donor photobleaching. 3,12,13 Two-photon excited fluorescence microscopy relies on the simultaneous absorption of two near-infrared photons to cooperatively excite an electronic transition equivalent to that obtained by single-photon, ultraviolet excitation followed by fluorescence emission. 14 Two-photon excited fluorescence imaging is advantageous over its single-photon counterpart for monitoring protein dynamics, such as FRET, in living cells owing to its intrinsic optical sectioning capability, which minimizes photobleaching and photodamage. 15,16 Identification of the FRET signal still relies on filters and ratiometric image analysis. In contrast to imaging approaches, spectral acquisition negates the need for filters and ratiometric mea-surements, and facilitates quantification of the individual components of the FRET signal.
Detection of noninducible homodimerization versus heterodimerization is inherently more difficult since at most only 50% of the dimers will contribute to the FRET signal. Furthermore, dimers formed exclusively by a donor or acceptor obscure the FRET signal. The promyleocyte ͑PML͒ has been shown to localize to discrete nuclear bodies 17 whose function remains unclear. Its dissociation from these bodies and participation in a chromosomal translocation with the retinoic acid receptor ␣ ͑RAR␣͒ are hallmarks for acute promyelocytic leukemia ͑APL͒. Dimerization of PML has been demonstrated in vitro to be mediated by the coiled-coil domain of the protein, which is also responsible for the heterodimerization between PML and the fusion protein PMLRAR␣ in APL. 18 To this end, we chose to investigate the formation and subcellular spatial distribution of PML homodimers in vivo. We demonstrate that integration of two-photon excited fluorescence microscopy, spectral acquisition, and second-generation GFP mutants enable the detection of PML homodimers in subcellular domains in the living cell.

Plasmid Construction
CFP ͑Clontech, Palo Alto, California͒ and YFP ͑Topaz͒ ͑Packard BioSciences, Meriden, Connecticut͒ mutants were subcloned C-terminally into a CMV-driven expression plasmid containing the PML sequence as previously described. 19 Specifically, both GFP mutants were generated by a polymerase chain reaction ͑PCR͒ using the respective plasmids as a template and oligos containing an Mlu1 site at the 5Ј end and an Xba1 site at the 3Ј end. The fragments were inserted into the CMX-PML vector that was digested with Mlu1 and Nhe1 at the C-terminal. For control experiments, pCDNA3-NLS-ECFP, containing a nuclear localization signal ͑a gift from Drs. B. Forman and C. Solier͒ was subcloned from the pCDNA3-NLS-ECFP-RAR␣ expression plasmid, 9 ͑a gift from Dr. C. Glass͒.

Cell Culture and Microinjection
HEp-2 cells ͑ATCC, Manassas, Virginia͒ were cultured as previously described 19 in drilled 35-mm dishes with an alphanumerically gridded coverslip ͑Bellco Glass, Vineland, New Jer-sey͒ affixed to the bottom. Approximately 10 Ϫ15 liters/cell of DNA plasmid ͑50 ng/l͒ were introduced into the nucleus of the cell using a semiautomated microinjection system ͑Eppendorf Scientific, Westbury, New York͒ and a Zeiss Axiovert microscope ͑Zeiss Inc., Oberkochen, Germany͒. The DNA plasmid concentration was kept equal for CFP and YFP to ensure equivalent expression within a cell. The cells were incubated overnight before FRET analysis to permit expression.

Two-Photon Imaging and Spectroscopy System
The two-photon imaging microscope used for this study has been described 20 and has been modified to include a SpectraPro-150 spectrograph. Briefly, it consists of a modelocked Ti:sapphire laser ͑170-fs pulse width, 76-MHz repetition rate; Mira 900F, Coherent, Santa Clara, California͒ pumped by a 5-W Verdi laser ͑Coherent͒. The beam exiting the laser is deflected into the back port of an inverted Axiovert 100 microscope ͑Zeiss Inc.͒ and scanned across the sample via a personal computer ͑PC͒ controlled galvanometer-driven x-y scanner ͑Series 603X, Cambridge Technology, Watertown, Massachusetts͒. The beam is reflected by a short-pass 675-nm dichroic beamsplitter ͑Chroma Technology Corp., Brattleboro, Vermont͒ and focused onto the sample with a 63ϫ, c-apochromat, NAϭ1.2, water immersion microscope objective ͑Zeiss Inc.͒. For NLS-CFP control experiments, a 100ϫ, NAϭ1.45 oil immersion microscope objective ͑Zeiss Inc.͒ was used. The average power entering the microscope is approximately 30 mW ͑2.5 mW at the sample site͒. Twophoton excited fluorescence from the sample is epicollected, discriminated with the dichroic beamsplitter, filtered by a short-pass 600-nm filter ͑CVI Laser Corp., Livermore, Cali-fornia͒ and detected. Two-dimensional (x-y plane͒ images (256ϫ256 pixels) are acquired at a rate of 1 frame/s ͑pixel dwell time of 16 s/pixel͒ with a single-photon counting photomultiplier tube ͑PMT͒ ͑Hamamatsu Corp., Bridgewater, New Jersey͒ and cover an area of 35ϫ35 m for the 63ϫ microscope objective. An SBG39 wide-pass ͑322 to 654-nm͒ blue-green emission filter ͑CVI Laser Corp., Albuquerque, New Mexico͒ is placed in front of the PMT. Spectra are obtained with a SpectraPro-150 spectrograph equipped with a 300 grooves/mm grating blazed at 500 nm ͑Acton Research Corp., Acton, Massachusetts͒, and a high dynamic range MicroMax:512BFT CCD camera ͑Princeton Instruments, Trenton, New Jersey͒, which is controlled by an ST-133 Controller ͑Princeton Instruments͒. The spectrograph and camera settings are PC-controlled through commercially available software ͑WinSpec/32 v. 2.4.6.6, Roper Scientific Inc., Trenton, New Jersey͒. The CCD temperature is maintained at the minimum possible temperature ͑Ϫ45°C͒ for all the experiments to ensure low dark noise. The entrance slit of the spectrograph is set to a width of 0.5 mm. The spectral acquisition time was 60 s. For the NLS-CFP control, the spectral acquisition time was 10 s. Switching between imaging and spectral acquisition is achieved by changing the position of a built-in microscope mirror. When both two-photon images and spectra are acquired from the sample, the two-photon images are acquired and stored; this is immediately followed by acquisition of emission spectra from the same depth ͑z͒ into the sample. Autofluorescence of the cell is negligible. Dark noise spectra are subtracted from the acquired sample spectra.   t-test analysis for the 501/476-nm ratios was not statistically significant (p ϭ0.1036), reflecting the consistency of this ratio for donor (CFP) fluorescence. An unpaired Student's t-test analysis of the 527/476-nm ratios was not significant (pϭ0.6076), which is indicative of the absence of FRET. Scale bar equals 10 m.

Data Analysis
Acquired 256ϫ256 pixel images were converted to bitmap format with no postprocessing modifications. Spectral data were imported into Microsoft Excel 2000 ͑Microsoft Corp.͒ and plotted with intensity ͑arbitrary units͒ versus the emission wavelength ͑nanometers͒. Because of the low signal-to-noise ratio in the acceptor alone case, a moving average ͑period: 20͒ function was used to plot the spectra. Peak intensity ratios were statistically analyzed ͑Instat, San Diego, California͒ by an unpaired Student's t-test, and a probability value was calculated.

Results and Discussion
Plasmids encoding fusion proteins of PML with either a donor, CFP ͑Clontech͒, or an acceptor, YFP or Topaz ͑Packard BioSciences͒, were introduced alone or in combination into HEp-2 cells ͑ATCC͒. Two-photon images of cells expressing the donor ͑PMLCFP͒ alone, the acceptor ͑PMLYFP͒ alone, and both ͑PMLCFP/PMLYFP͒ were acquired with a twophoton imaging and spectral detection system ͑Fig. 1͒ at an excitation wavelength of 810 nm. This wavelength was selected for its minimal excitation of the acceptor, which is a crucial requirement for the detection of FRET. 8 Other wavelengths tested throughout the tuning range of the Ti:sapphire laser ͑730 to 880 nm͒ exhibited efficient, direct excitation of the acceptor ͑YFP͒. Subsequent to single-scan image acquisition, the emission spectrum from the same optical section was collected at a 60-s acquisition time. Images of one representative cell out of ten examined cells for a single experiment are presented for each case. To ensure reproducibility of the results, experiments were performed in triplicate.
Two-photon excited fluorescence images ͓Figs. 2͑a͒, 2͑b͒, and 2͑c͒ ͑see color plate͔͒ depict the subcellular localization of the PML fusion proteins to nuclear bodies as previously reported. 19 As evident in Fig. 2͑b͒, the fluorescence signal of the acceptor protein ͑PMLYFP͒ is minimal. In contrast, the donor ͑PMLCFP͒ alone and the combination ͑PMLCFP/ PMLYFP͒ are efficiently excited ͓Figs. 2͑a͒ and 2͑c͔͒. The standard approach of ratioing the FRET to donor images for determination of FRET is not reliable here since the expression levels of the proteins may vary between different optical sections within the bodies and from cell to cell. Furthermore, since a decrease in the donor fluorescence intensity is observed in the presence of FRET compared with the donor alone, image ratioing may overcorrect the FRET response. The two-photon emission spectrum of CFP ͓Fig. 2͑d͒, blue͔ covers a region from 450 to 580 nm, with maxima at 476 and 501 nm. When YFP is efficiently excited at 866 nm, its emission spans a region from 510 to 580 nm, with a maximum at 527 nm ͓Fig. 2͑e͔͒. Maximal two-photon excitation of YFP is not feasible owing to the limitations imposed by the tuning range of the laser ͑730 to 880 nm͒. The in vivo spectra ͑Fig. 2͒ are consistent with the in vitro spectra obtained from purified GFPs. 2 Control studies were performed and demonstrated that prolonged irradiation ͑Ͼ9 min͒ of the sample did not alter the recorded image or spectral intensities as expected owing to the photostability of CFP. 21 As expected in FRET, a decrease in CFP intensity was observed in the donor-acceptor spectrum ͓Fig. 2͑d͒, red͔, with a concomitant increase in YFP intensity compared with the individual spectra for the donor ͓Fig. 2͑d͒, blue͔ and the acceptor ͓Fig. 2͑d͒, green͔ alone.
To further illustrate the occurrence of FRET between CFP and YFP, the ratio of intensity of the 527-nm peak to that of the 476-nm peak was calculated for cells expressing the donor alone and for those expressing the FRET pair. In the presence of CFP alone, this ratio is consistently 0.52. For the 810-nm two-photon excitation, YFP intensity values at 527 nm are negligible for the acceptor alone. Thus, an increase in the 527/476-nm ratio in cells expressing both CFP and YFP would serve as a measure of FRET occurrence, which would be indicative of a protein-protein interaction. The 501/ 476-nm ratio corresponds to CFP expression only and serves as a spectral fingerprint for CFP whether expressed alone or in combination with YFP. Inherent to the spectra of CFP, this ratio is preserved at 0.9 irrespective of changes in expression levels between cells.
As described earlier, ratios were calculated for individual cells expressing PMLCFP and those expressing PMLCFP/ PMLYFP, and the average ratio values are presented in Fig. 3. The average of the 501/476-nm ratio was 0.94Ϯ0.002 and 0.95Ϯ0.01, for PMLCFP and PMLCFP/PMLYFP, respectively. The average of the 527/476-nm ratios was 0.52Ϯ0.01 and 0.86Ϯ0.02 for PMLCFP and PMLCFP/PMLYFP, respectively. An unpaired Student's t-test analysis for the 501/ 476-nm ratios was not statistically significant (pϭ0.773) reflecting the consistency of this ratio for donor ͑CFP͒ fluorescence. An unpaired Student's t-test analysis of the 527/ 476-nm ratios was highly significant (pϽ0.0001) indicating a statistical difference between donor and donor-acceptor cells. The observed increase in the 527/476-nm ratio is a direct result of the energy transfer between CFP and YFP and reveals a protein-protein interaction in PML homodimers. The ratio values for ten cells expressing the donor and the acceptor (PMLCFP/PMLYFP) were 0.86Ϯ0.02 and 0.95Ϯ0.01, respectively. The ratios are independent of expression levels from cell to cell. An unpaired Student's t-test analysis for the 501/476-nm ratios was not statistically significant (pϭ0.773), reflecting the consistency of this ratio for donor (CFP) fluorescence. An unpaired Student's t-test analysis of the 527/476-nm ratios was highly significant (p Ͻ0.0001), indicating a statistical difference between donor and donor-acceptor cells, revealing the occurrence of FRET.
As a negative control, NLSCFP, which contains a nuclear localization signal and directs CFP to the nucleus, was introduced into cells as the donor. PMLYFP was maintained as the acceptor. Figure 4͑a͒ ͑see color plate͒ depicts the two-photon fluorescence image of a cell expressing NLSCFP/PMLYFP at 810-nm excitation with fluorescence distributed throughout the nucleus. The corresponding spectrum ͓Fig. 4͑b͔͒ has two peaks at 476 and 501 nm. The 527/476-nm ratio for NLSCFP/ PMLYFP is 0.52Ϯ0.01, which is identical to that measured for the donor alone (0.52Ϯ0.01), indicating the absence of FRET. The 501/476-nm ratio for NLSCFP/PMLYFP was 0.95Ϯ0.003, which is consistent with donor-alone values.
Monitoring FRET by two-photon imaging and spectroscopy uniquely enables the examination of protein-protein interactions at the subcellular level. It reveals for the first time the localization and formation of PML homodimers within the corresponding nuclear bodies in vivo. These findings open new avenues of exploration into the dynamics of PML and the proteins that reside within these bodies, which in turn will provide new insight into the function of the PML body. The TPIS-FRET method can be extended to other cellular questions that focus on protein interactions that may occur within discrete domains, such as transcription. Temporal and spatial resolution of these interactions can be achieved by this approach, which inherently provides structural and functional information from subfemptoliter volumes.
TPIS-FRET is advantageous over ratiometric analysis for the qualitative determination of FRET because it negates the need for postprocessing of the images and the utilization of optical filters. It is independent of protein expression levels from cell to cell, and transiently transfected or microinjected DNA expression plasmids can be readily evaluated. This spectroscopic technique should be applicable to other mutant GFP pairs. It invites future analysis of engineered protein-protein interactions in tissue by exploiting the ability of near-infrared light to penetrate deeply into living tissue.
Understanding the function of a protein by following its dynamic interplay with other proteins in a living cell can contribute fundamentally to understanding the overall cellular process or disease in which it participates. Future studies will examine spatial and temporal variations of PML interactions with other proteins, including PMLRAR␣, within living cells and tissues.