Selecting the right fluorophores and flow cytometer for fluorescence resonance energy transfer measurements

Abstract

Background

Fluorescence resonance energy transfer applied in flow cytometry (FCET) is an excellent tool for determining supramolecular organization of biomolecules at the cell surface or inside the cell. Availability of new fluorophores and cytometers requires the establishment of fluorophore dye pairs most suitable for FCET measurements.

Methods

A gastric tumor cell line (N87) was labeled for major histocompatibility complex class I heavy chain and β2-microglobulin with antibodies conjugated with fluorescein- and indocarbocyanine-like fluorophores and analyzed in FCET measurements on a cell-by-cell basis using three flow cytometers: FACSCalibur, FACSDiVa, and FACSArray.

Results

Normalized fluorescence intensity values were measured and normalized energy transfer efficiencies, spectral overlap integrals, and crucial dye- and instrument-dependent parameters were calculated for all matching pairs of seven fluorophores on the three commercial cytometers. The most crucial parameter in determining the applicability of the donor-acceptor pairs was the normalized fluorescence intensity and the least important one was the spectral overlap.

Conclusions

On the basis of available laser lines, the optimal dye pair for all three cytometers is the Alexa546-Alexa647 pair, which produces high energy transfer efficiency values and has the best spectral characteristics with regard to laser excitation, detection of emission, and spectral overlap. © 2005 Wiley-Liss, Inc.

Investigation of protein-protein associations is important in understanding the structure-function relations in living cells. Fluorescence resonance energy transfer (FRET) based methods are excellent tools for determining association patterns of biomolecules at the cell surface or inside the cell. With the help of FRET, molecular dimensions can be measured and determined in functioning, live cells, thus providing information that would be impossible to obtain with other classic approaches, e.g., electron microscopic methods.

The theory of FRET was introduced by Förster in the late 1940s (1); however, it did not become widespread in biomedical studies until Stryer's popularizing article (2). The FRET process is a long-range dipole-dipole interaction in which an excited donor fluorophore transfers its energy to a neighboring acceptor molecule in a nonradiative way. The main application of FRET as a spectroscopic ruler is based on the fact that the rate of energy transfer depends on the inverse sixth power of the separation of the two interacting molecules. The most apparent manifestation of the FRET process is the decrease of donor fluorescence in the presence of an acceptor in close vicinity (1–10 nm), which is accompanied by an increase of fluorescence of the acceptor (if the acceptor is a fluorescent molecule). The FRET phenomenon has been adapted for microscopy (3–6) and flow cytometry (7–9) and several biological structures have been successfully investigated: receptors involved in immune response (10–12), growth factor receptors on tumor cells (13–15), determination of protein conformation (16, 17), and lipid microdomain structures (9, 18, 19). Several reviews are also available for biomedical and clinical applications of FRET (8, 18, 20–24).

One of the first FRET approaches carried out in flow cytometry was the flow cytometric energy transfer (FCET) method with two-color excitation and proper handling of spectral crosstalk between fluorescent dyes (7, 25). This method was developed for a classic cytometer equipped with 488- and 514-nm lasers suitable for the frequently applied fluorescein and rhodamine dye pair. During FCET measurements, these excitation wavelengths are not optimal for the fluorescein and rhodamine pair because of the substantial crosstalk between the dyes in various detection channels. In new versions of flow cytometric systems, which usually incorporate 488- and 635-nm laser excitations, the fluorescein and rhodamine pair cannot be applied because the 635-nm excitation is outside the spectral range of rhodamine and the spectral overlap between fluorescein and a typical dye excitable at 635 nm is minimal, practically prohibiting the FRET process. However, recently developed dye pairs (Cy3-Cy5) were successfully adopted to a newer, bench-top cytometer (FACSCalibur), and the analysis procedures were improved to consider autofluorescence on a cell-by-cell basis by means of an otherwise unused detection channel (26).

In recent years, the availability of cheaper diode and solid-state lasers with higher power output and matching fluorophore pairs, which span the entire visible spectrum reaching the far-red region, raised the question of selecting the ideal dye pair for the instrument in FRET studies. In our study, we investigated the performance of fluorescent dyes: Cy3, Alexa546, Alexa555, and Alexa568 (FRET donors) and Cy5, Alexa633, and Alexa647 (FRET acceptors) in studying molecular associations with the help of various flow cytometers: FACSCalibur, FACSDiVa, and FACSArray. Some other types of fluorescent molecules were not included in this investigation, but their general roles and usefulness were also considered. In this study, we have summarized the results obtained for all matching pairs of the seven fluorophores on three flow cytometers and established parameters crucial for FCET measurements. Taking into consideration all the significant parameters that influence the magnitude and accuracy of the FRET efficiency value measured, we determined the most suitable donor-acceptor pair for FCET measurements in all cytometers tested.

MATERIALS AND METHODS

Cell Lines

Human gastric cancer cell line N87 with high major histocompatibility complex (MHC) class I expression level was obtained from the American Type Culture Collection (Rockville, MD, USA) and grown according to the manufacturer's specification (in RPMI containing 10% fetal calf serum, 2 mM l-glutamine, and 0.25% gentamicin in 5% CO₂ atmosphere) to confluency. For flow cytometry, cells were harvested by treatment with 0.05% trypsin and 0.02% ethylenediaminetetraacetic acid.

Conjugation of Antibodies With Fluorescent Dyes

The W6/32 (against MHC class I heavy chain) and L368 (against β2-microglobulin) antibodies were a kind gift from Francis Brodsky. Aliquots of antibodies (∼1 mg/ml concentration) were conjugated with the succinimide derivative of fluorescein-like (Alexa546, Alexa555, Alexa568, Alexa633, and Alexa647; Molecular Probes, Eugene, OR, USA) and of indocarbocyanine-like (Cy3 and Cy5; Amersham Biosciences, Freiburg, Germany) dyes, as described previously (27). In the experiments, the W6/32 antibody was labeled with donor fluorophores (Cy3, Alexa546, Alexa555, and Alexa568), and the L368 antibody with acceptor fluorophores (Cy5, Alexa633, and Alexa647). The labeling ratios (L) were in the range of 1 to 5, where concentration quenching does not yet have a significant effect on the fluorescence quantum yield of the donor and acceptor dyes.

Labeling Cells with Fluorescent Antibodies

Freshly harvested cells were washed twice in ice-cold phosphate buffered saline (PBS; pH 7.4). The cell pellet was suspended in PBS at a final concentration of 4 × 10⁷ cells/ml, and then 25 μl of conjugated antibodies was added to 25 μl of cell suspension and cells were incubated for 30 min on ice. The excess of antibody was at least fivefold above saturating concentrations during the incubation. The same procedure was used for FRET samples, in which a mixture of donor- and acceptor-labeled antibodies was added to the cell suspension. The labeled cells were washed twice with excess cold PBS and fixed with 1% formaldehyde and PBS.

Flow Cytometers

The FCET measurements were carried out on three commercial cytometers: FACSCalibur, FACSVantage SE with DiVa option, and FACSArray (Becton Dickinson, Franklin Lakes, NJ, USA). The FACSCalibur is equipped with a 488- and a 635-nm laser and three detectors were used: FL2 (585/42 band pass), FL3 (670 long pass), and FL4 (661/16 band pass filter). The FACSDiVa features a 532-nm diode-pumped solid-state laser, a 633-nm air-cooled HeNe laser, and a 488-nm Ar laser for forward scatter triggering. The emission filters were a 585/42 band pass filter and two 650 long pass filters. The FACSArray is shipped with a 532-nm solid-state laser and a 635-nm diode laser, and for FRET measurements the detectors with 585/42 band pass (yellow), 685 long pass (far red), and 661/16 band pass (red) filters were used.

Theory of Fluorescence Resonance Energy Transfer

The FRET phenomenon is a dipole-dipole interaction in which an excited donor fluorophore transfers its energy to a neighboring acceptor molecule in a nonradiative way (28). This interaction manifests in the decrease of lifetime of the donor molecule. The measure of FRET is the efficiency (E) of energy transfer:

• (1)

i.e., the fraction of absorbed photons that are transferred without radiation to the acceptor, where k_T is the rate of energy transfer, k_F is the rate of fluorescence of the donor without an acceptor, and k_n is the sum of all nonradiative decay rates of the donor except of energy transfer. The rate of interaction can be described as follows:

• (2)

The combination of these two equations provides the distance dependence of FRET efficiency:

• (3)

where R₀ is the so-called Förster distance, i.e., the distance at which FRET efficiency is 50%. The following term was derived for R₀ (nanometers):

• (4)

where Q_D is the donor quantum yield in the absence of acceptor dyes, n is the index of refraction of the conveying medium (usually 1.4 is used for cell surface antibody labeling), κ² is the orientation factor (its value is 0 to 4 and equals ⅔ for dynamic averaging of isotropic transition moments) (29, 30), and J_DA is the overlap integral (28). The J_DA overlap integral (moles per cubic centimeter) can be calculated in the knowledge of the donor emission and acceptor excitation spectra:

• (5)

where f_D(λ) is the normalized fluorescence spectrum of the donor dye:

• (6)

The overlap integrals of the dye pairs applied in this study were calculated from spectroscopic data available at the Web site of the manufacturer (http://www.probes. com), except for Cy3 and Cy5, which were measured on a FluoroLog-3 spectrofluorometer (Jobin Yvon, Edison, NJ, USA). The excitation spectra of the fluorophores can be seen in Figure 1.

Figure 1. Normalized excitation spectra of donor and acceptor fluorophores. The three laser lines used in our flow cytometric setups are indicated by vertical lines as wavelength of excitation.

Flow Cytometric Energy Transfer Measurements

The easiest way to measure FRET in a flow cytometer is to use the background corrected fluorescence signal of the donor from a donor-only labeled sample (I) and a donor- and acceptor-labeled sample (I) and to calculate FRET efficiency (E) as:

• (7)

The disadvantage of this “donor quenching” method is that the mean fluorescence intensity of two separate samples is used and these intensities are not corrected for spectral crosstalk from the acceptor, and corrections for antibody competition also have to be considered.

To make corrections for spectral crosstalk and be able to measure FRET efficiency on a cell-by-cell basis, one has to detect three independent signals (7):

• (8)

• (9)

• (10)

The equation set contains the direct donor excitation in I₁, the acceptor sensitization in I₂, the direct acceptor excitation in I₃, the unquenched donor fluorescence (I_D), and the pure acceptor signal (I_A). All intensities are background corrected. To solve the equation set, one has to determine the S_1–4 spectral overlap factors in separate measurements with donor- and acceptor-only labeled samples. Thereby the following equation can be derived for FRET efficiency:

• (11)

The α-factor is determined empirically according to the following formula:

• (12)

where I is the I₂ signal of acceptor-only labeled sample and I is the I₁ signal of donor-only labeled sample, L is the labeling ratio of the antibodies, B is the mean number of receptors labeled, and ϵ is the excitation coefficient of the dyes at the donor excitation wavelength (the D and A indexes refer to donor and acceptor, respectively). The ratio of molar extinction coefficients of the fluorophores at the donor excitation wavelength (ϵ_D/ϵ_A) is the so-called efr coefficient.

The E transfer efficiency values are calculated on a cell-by-cell basis with AFlex software (31) and are presented as mean values from approximately normally distributed, unimodal energy transfer histograms of 20,000 cells.

RESULTS

Establishing Factors That Influence FCET Measurements

We found four parameters that are relevant for determining the practicality of FRET dye pairs, namely normalized fluorescence intensity (I_norm), overlap integral (J_DA), normalized energy transfer efficiency (E_norm), and the efr coefficient.

The fluorescence intensity is very important in FCET measurements because one measure of FRET efficiency is the decrease of donor signal in the presence of an acceptor. Thus, it is preferable that the donor signals are at least twice as high as the background intensities to detect donor quenching. In these measurements, the fluorescence intensities were at least three times, but mostly fifty times, above background intensity. The background subtracted intensities were corrected for labeling ratios to obtain the fluorescence intensities for the same number of fluorescent dyes (Table 1). The I_norm values were in the ranges of 5 to 200 and 60 to 1,100 for FACSCalibur and FACSDiVa, respectively.

Table 1. Normalized Fluorescence Intensities and Molar Extinction Coefficients of Donor and Acceptor Fluorophores*

	Fluorophore	Imeas	L	Inorm	ϵmax (1/M* cm)	ϵexc (1/M* cm)
* Imeas, measured fluorescence intensity (in arbitrary units); L, labeling ratio of antibody (W6/32 in the case of donor and L368 in the case of acceptor labeling); Inorm, normalized fluorescence intensity; ϵmax, molar extinction coefficient at maximum intensity; ϵexc, molar extinction coefficient at excitation wavelength. a Autofluorescence.
FACSCalibur
Donor (488 nm)	—	8a
	Cy3	124	3.56	33	150,000	28,000
	Alexa546	100	1.85	50	112,000	8,000
	Alexa555	147	3.46	40	158,000	23,000
	Alexa568	29	4.19	5	88,000	5,000
Acceptor (635 nm)	—	1a
	Cy5	411	2.91	158	250,000	187,000
	Alexa633	80	1.06	75	159,000	154,000
	Alexa647	1054	5.2	202	250,000	141,000
FACSDiVa
Donor (532 nm)	—	27a
	Cy3	1737	3.56	481	150,000	86,000
	Alexa546	2094	1.85	1118	112,000	37,000
	Alexa555	2215	3.46	632	158,000	78,000
	Alexa568	280	4.19	61	88,000	30,000
Acceptor (633 nm)	—	8a
	Cy5	1130	2.91	386	250,000	174,000
	Alexa633	392	1.06	362	159,000	159,000
	Alexa647	2954	5.2	567	250,000	129,000

Because the rate of energy transfer is influenced by the Förster distance of the dye pair, which contains physical constants distinctive for the dyes (see equation 4), this parameter should be a good tool for characterizing the FRET dye pairs. Unfortunately, the quantum yield highly depends on the environment, so it is difficult to tell the value of Q_D for a given donor dye under measurement conditions. However, it should be noted that R₀ is mostly sensitive to Q_D in the range of 0.0 to 0.3 (28), which is the characteristic range for cyanine dyes, whereas the Alexa dyes have quantum yield values about or higher than 0.3 in our experience (data not shown); for further information, please see Shapiro (32), pp. 336–338. In the present study, the actual distances were not important, so we used the overlap integral rather than the Förster distance. The J_DA overlap integrals for the 12 combinations were calculated as described in Materials and Methods and are presented in Table 2. The J_DA values were in the range of 5–15 × 10⁻¹³ M/cm³ and were the highest for dye pairs containing Alexa568 as a donor and lowest for pairs containing Cy3 as a donor.

Table 2. Overlap Integrals of Donor and Acceptor Dye Pairs*

Donors	Acceptors
	Cy5	Alexa633	Alexa647
* Overlap integrals (JDA; 10^13 M/cm^3) are calculated from emission and excitation spectra with equation 5 as explained in Materials and Methods.
Cy3	5.5	5.5	5.2
Alexa546	8.1	7.2	7.8
Alexa555	8.7	7.4	8.6
Alexa568	14.6	10.5	14.9

The actual measured FRET efficiency (E) also has to be taken into account to find a good dynamic range for the E value. Low E values compromise the statistical accuracy of the measurements and hinder monitoring changes in the association pattern. High E values are also error prone, especially above 90%, because of the way E is calculated from the experimentally determined A parameter (E = A/[1 + A]). The E values were obtained by detecting all three intensities in equations 8–10, solving for equation 11 on a cell-by-cell basis, and tabulating the means of unimodal efficiency distribution histograms. To compare the efficiency values between dye pairs, they have to be normalized for their excitation spectrum, quantum yield, and the donor-acceptor ratios of the dyes. The excitation spectra are already considered in the α-factor, although the quantum yields are not known, but the correction for donor-acceptor ratios can be easily implemented. In our biological system, the light and heavy chains of MHC class I molecules are expressed constitutively on the cell surface, and only a small portion (<10%) of the heavy chains stands alone. So it can be surmised that almost all donor-labeled antibodies are in close proximity of 1 and only one acceptor-labeled antibody. In such cases, only the ratio of the labeling ratios (L_D/L_A = L_R) has to be considered because the ratio of the number of antibodies is approximately 1. Because the FRET efficiency is not linearly proportional to the labeling ratios (or the α-factor) but the A value is, the A = E/(1 − E) values have to be used for normalization to L_D:L_A = 1:1. The normalized efficiencies (E_norm) listed in Table 3 were calculated with the formulas derived from equations 11 and 12:

• (13)

where L_Rnorm = 1. Then, the obtained A_norm values were converted to normalized transfer efficiencies:

• (14)

Dye pairs containing Alexa568 as a donor showed the highest FRET efficiency values; on one occasion, for the Alexa568 and Alexa633 pair, the normalized FRET efficiency value was even close to 0.80 on the FACSDiVa instrument. The lowest normalized FRET efficiency values (about 0.22) were obtained for dye pairs containing Alexa546 dyes as a donor on the FACSCalibur instrument. FRET efficiency histograms for the dye pairs containing Alexa647 as an acceptor are shown in Figure 2. The highest mean value of the FRET efficiency distribution histogram was obtained for the Alexa568-Alexa647 pair, but the width of the distribution curve was also the largest for this pair.

Figure 2. FRET efficiency distribution histograms of donor fluorophores with Alexa647 as an acceptor measured on a FACSDiVa flow cytometer. The FRET efficiency values were calculated on a cell-by-cell basis by equation 11 as described in Material and Methods.

Table 3. Normalized Energy Transfer Efficiencies (E_norm) of Dye Pairs*

Donors	Acceptors
	FACSCalibur			FACSDiVa
	Cy5	Alexa633	Alexa647	Cy5	Alexa633	Alexa647
* The E values were normalized to labeling ratios of donor and acceptor dyes.
Cy3	0.341	0.327	0.298	0.345	0.326	0.244
Alexa546	0.236	0.213	0.202	0.311	0.412	0.258
Alexa555	0.366	0.382	0.393	0.388	0.537	0.451
Alexa568	0.574	0.641	0.585	0.655	0.796	0.657

Another important factor is the α-factor or the efr coefficient. Because the α-factor contains parameters that may vary in each experiment, we found that the efr coefficient provides a more reliable means to characterize FRET dye pairs. The efr coefficient is given by the ratio of molar extinction coefficients of the fluorophores at the donor excitation wavelength (ϵ_D/ϵ_A). Because this physical parameter cannot be reliably obtained from observing the absorption spectrum of the dyes, another possibility is to determine the FRET efficiency from equation 11 and change the efr factor (or the related α-factor) until the efficiency value equals the one obtained by the donor quenching calculation (corrected for antibody competition) and/or the unquenched donor fluorescence intensity (I_D) equals the fluorescence intensity of the donor-only labeled sample. The efr coefficients determined this way are summarized in Table 4 for two cytometers (FACSCalibur and FACSDiVa). The efr values were in the range of 0.2 to 17 and were the highest for dye pairs containing Cy3 as a donor and lowest for pairs containing Alexa568 as a donor.

Table 4. The efr Coefficients of FRET Dye Pairs*

Donors	Acceptors
	FACSCalibur			FACSDiVa
	Cy5	Alexa633	Alexa647	Cy5	Alexa633	Alexa647
* The efr coefficient is the ratio of molar extinction coefficients (ϵD/ϵA) of the fluorophores at the donor excitation wavelength (488 nm on FACSCalibur and 532 nm on FACSDiVa). The values were determined empirically by comparing the FRET efficiency values obtained on a cell-by-cell basis with that obtained by the donor quenching method.
Cy3	17.0	10.4	14.8	8.5	3.7	7.1
Alexa546	9.0	7.1	8.1	4.2	1.3	3.3
Alexa555	12.7	6.8	9.2	6.0	1.3	3.1
Alexa568	1.1	0.7	0.9	0.7	0.2	0.4

Selecting the Most Suitable FRET Dye Pair

Because of the large number of combinations between the four donor and three acceptor dyes, we wanted to narrow the scope of our search to obtain the most suitable dye pair. The relevant approach seemed to be to focus first on the overlap integrals and FRET efficiency values to sieve out the less favorable dyes. As can be seen in Table 2, Cy3 has the worst J_DA characteristics and is known to have the lowest fluorescence quantum yield (see product description), although the E_norm and efr values are quite high and the intensity is average. Another striking feature is the very high J_DA and E values for the Alexa568 dye. However, these values should be contrasted with the extremely low intensity and efr values, which result in less reliable FRET efficiency histograms (broad distribution) and render FCET measurements practically useless with this dye. For a comparison of FRET efficiency distribution histograms of donor dyes with the acceptor Alexa647, see Figure 2. Thus, two donor dyes, Cy3 and Alexa568, should be excluded from further investigations. Looking at the data of acceptor dyes, we found no significant differences. Although Alexa647 had slightly smaller E_norm values, it produced the highest intensity, whereas Alexa633 had lower J_DA and efr values, but its E_norm was remarkable. So none of the acceptors should be excluded.

For a more thorough investigation, Alexa546 and Alexa555, as donors, and Cy5, Alexa633, and Alexa647, as acceptors, giving six dye pairs, were used.

First, we analyzed the intensities (Table 1) and found that Alexa546 had a fluorescence intensity that was two times as high as that for Alexa555. Among the acceptors, we found no great differences and the intensities fluctuated a bit from one cytometer to another. Nonetheless, it should be noted that Alexa647 has the highest intensity and Alexa633 the lowest intensity. So the Alexa546-Alexa647 pair is the best choice when considering fluorescence intensity.

Second, we examined the overlap integrals (Table 2) because they determine the range of sensitivity of energy transfer via the Förster distance. In all cases, Alexa633 as an acceptor produced the smallest overlap integral and Alexa555 as a donor had the highest overlap integral. The other two acceptors, Cy5 and Alexa647, produced similar values, although the value for Cy5 was a bit higher. Thus, in this respect, the Alexa555-Cy5 pair was the best dye pair, but the other dyes should not be excluded because of the small differences between dye pairs (J_DA values were in the range of 7.2 to 8.7 × 10⁻¹³ M/cm³).

Then we investigated the normalized FRET efficiency values (Table 3). For the two donor dyes, we obtained fairly similar values on both cytometers, so we do not believe this is a decisive factor in either case. However, there were larger differences between the acceptor dyes, and Alexa633 had the highest FRET values, whereas Alexa647 had the lowest. Thus the Alexa555-Alexa633 pair is the most effective when viewing the normalized FRET efficiency values. It should be noted that the E_norm values measured on FACSDiVa (532- and 633-nm excitation) were higher in all cases than those measured on FACSCalibur (488- and 635-nm excitation), although they should be the same independent of the instrument. This difference is most likely due to the better signal-to-noise ratio of the donor fluorescence because the 532-nm excitation is more suited for the donors than is the 488-nm excitation.

The last factor we investigated was the efr coefficient (Table 4), and the largest differences occurred in this dataset. The efr coefficients of Alexa555-containing dye pairs were higher than those of the Alexa546-containing pairs, which should have been the other way if only the excitation spectra had been considered (Fig. 1). However, the values with Alexa633 were (almost) the same on both cytometers, and on FACSDiVa the Alexa546-Alexa647 pair had a higher efr value than did the Alexa555-Alexa647 pair. The efr values for the acceptors tested increased in the order Alexa633, Alexa647, and Cy5 independent of the donor on both cytometers. Therefore, when considering the efr values, the Alexa555-Cy5 pair seems to be the best choice for FCET measurements.

The three instrument-dependent parameters were also determined for the Alexa546-Alexa647 dye pair on the FACSArray. The normalized fluorescence intensities were 5,356 and 20,852 (with background intensities 83 and 11) for Alexa546 (L_D = 4.05) and Alexa647 (L_A = 4.54), respectively. These values were even larger than those obtained with FACSDiVa, probably due to the cuvette system. The normalized FRET efficiency value was 0.352, which was larger than those for FACSCalibur and FACSDiVa. The efr coefficient, determined as described in Materials and Methods, was the same (3.3) as that for the FACSDiVa, which can be explained by the similar laser lines and detection optics.

DISCUSSION

In our study, we selected seven recently developed fluorescent dyes, Cy3, Alexa546, Alexa555, and Alexa568 (FRET donors) and Cy5, Alexa633, and Alexa647 (FRET acceptors), for flow cytometric energy transfer measurements. The measurements were carried out on two cytometers, FACSCalibur and FACSDiVa, and the Alexa546-Alexa647 dye pair was also tested on a FACSArray, with excitation and detection characteristics similar to those used for the FACSDiVa. To elucidate the comparison of the fluorophores and cytometers, we characterized and measured four parameters that influence FCET measurements: normalized fluorescence intensity, overlap integral, normalized FRET efficiency, and the efr coefficient.

If we compare all datasets obtained on the three cytometers, the most significant feature is the pronounced fluorescence intensity (and signal-to-noise ratio) with the 532-nm donor excitation, which is not surprising in view of their excitation spectrum. However, the normalized fluorescence intensities measured with different flow cytometers cannot really be compared even if the detector and laser energy profiles are known because of the different excitation and light harvesting conditions. One might also note large differences between intensities due to the different digitization schemes of the instruments, e.g., the FACSCalibur uses 10-bit analog-to-digital conversion, whereas the FACSDiVa, for pulse peak area, uses 18-bit analog-to-digital conversion. Therefore, to increase the sensitivity of FCET measurements, we can state that it is not enough to use fluorophores that emit in the red region of the visible spectrum (where the amount of autofluorescence is smaller) as was previously described (26), but the flow cytometer has to be also equipped with lasers that provide optimal excitation of these fluorophores.

The difference in signal-to-noise ratio also had a great effect on the FRET efficiency values. Signal-to-noise ratio and FRET efficiency were higher on the FACSDiVa than on the FACSCalibur for all tested dye pairs, although in principle the FRET efficiency values should be independent of the instrument. This discrepancy challenges the comparability of FRET results measured at the different research centers and/or on diverse equipment. In general, FRET efficiencies (even the distribution histograms) should be quite similar if measured on the same instrument or the same type of instrument, and the differences that arise can be attributed to the biological variance of the samples. This argument is probably also valid for the instruments with similar excitation laser lines, optical filters, and detectors. However, the change in the excitation wavelength of the donor can seriously influence its quantum yield (this physical parameter is usually not independent of the wavelength, except in special cases), which in turn has a large effect, via the Förster distance, on the value of FRET efficiency. A more ideal excitation wavelength also increases the donor signals and highly narrows the instrument-dependent variance of the FRET efficiency distribution histograms, which can manifest in altered mean FRET efficiencies. Thus, a feasible strategy to match data from different laboratories should at least include the comparison of fluorescence intensities and normalized FRET efficiencies and, if possible, quantum yield data of the fluorophores in use.

The J_DA and efr coefficient values presented in this report may be useful in other experiments. Because the overlap integrals were calculated from known excitation and emission spectra of the dyes, these values can be used to calculate the Förster distance in the knowledge of the donor quantum yield. The only problem may be the insertion of κ², which in special cases can be estimated if enough evidence is supplied. The most important feature of the efr coefficient is that these values are transitive between the same types of cytometers and can be used for FCET measurements on any cytometer with the same optical setup (this mainly stands for the commercial configurations of FACSCalibur and FACSArray). These coefficients can also be derived from the excitation spectrum of the fluorophores, but they are not so reliable because of the small signal-to-noise ratio of the fluorescence of acceptor dyes at the donor excitation wavelength. It should be noted that the efr coefficients obtained in this work and those derived from the spectra are different, and the ones presented in this work are more valid for measurements on biological systems.

During the analysis of the results, two donor dyes (Cy3 and Alexa568) were left out from thorough investigation, but they certainly can be considered with different cytometers and in other applications. Despite its large popularity in genomic research, we recommend the use of Cy3 as a FRET donor for protein studies only if the expression level of the molecule in question is high. However, the Cy3-Cy5 dye pair is an ideal candidate for microscopic acceptor photobleaching FRET experiments (33, 34) because Cy5 is highly susceptible to photobleaching, and with the available detection filters and mercury arc lamps or 543-nm laser excitation Cy3 is very practical. The strong argument for entirely excluding the Alexa568 dye was its insufficient excitability with the 488- and 532-nm lasers of our cytometers. However, if a cytometer is equipped with a laser light source that has a characteristic emission line closer to the maximum of the excitation spectrum of the dye (e.g., the 543-nm line of a HeNe laser or the 568-nm line of an Ar/Kr laser), then Alexa568 would be a far superior FRET donor fluorophore. Similar lasers are already used in confocal microscopes, so the Alexa568-Alexa647 pair (see J_DA values listed in Table 2) might be the most suitable FRET pair for intensity-based FRET microscopy (4).

The two remaining donor fluorophores, Alexa546 and Alexa555, were quite similar in every aspect of our investigation, but Alexa555 was superior in most cases. In our model system of highly expressed surface antigens, differences in signal-to-noise ratios of donor fluorescence were not prominent with these two dyes, but the low fluorescence signal of the donor dye on a low-expressed antigen can unfavorably influence the variance of FRET distribution histograms and seriously limit the sensitivity of FRET measurements. Thus the most important parameter in the case of a donor dye is its fluorescence intensity and, hence, signal-to-noise ratio in the donor fluorescence channel, which makes Alexa546 a better partner for FCET experiments. (Here the higher normalized fluorescence intensity can only be attributed to the better quantum yield of Alexa546.) In the case of an acceptor dye, the degree of spectral crosstalk to the donor fluorescence channel and the overlap integral is more important. In this respect, Cy5 seemed to be the best choice and Alexa633 the worst choice as a FRET acceptor and Alexa647 is very similar to Cy5. The only reason we think Cy5 should be considered as a less favorable fluorophore is its low quantum yield and weak photostability, which have to be taken into account if we want to expand our observations over microscopic and fluorometric studies.

After analyzing the relevant four parameters (normalized fluorescence intensity, I_norm; overlap integral, J_DA; normalized energy transfer efficiency, E_norm; and efr coefficient), we found that the Alexa555-Cy5 and Alexa546-Alexa647 pairs have the best characteristics. The Alexa555-Cy5 pair exhibits the highest overlap integral and FRET efficiency, whereas the Alexa546-Alexa647 pair has the highest fluorescence intensity. Because the most important parameter in fluorescence measurements is the excitability and detectability of the fluorescent dye, the Alexa546-Alexa647 pair is the most suitable fluorophore pair for FCET experiments. In contrast, in some microscopic techniques the photodestructibility of the fluorophores is also important, so the Alexa555-Cy5 pair (from the set of dyes we tested in this study) is the best choice for acceptor photobleaching FRET measurements (9).

Unfortunately, we could not test flow cytometers other than those from Becton Dickinson, but we have assessed other cytometers from different manufacturers available on the market; for a thorough list, see chapter 8 in Shapiro (32). We found that the FC500 (Beckman Coulter, Fullerton, CA, USA) and the CyAn (DakoCytomation, Fort Collins, Colorado, USA) have characteristics similar to the FACSCalibur; they share the same light sources, detection filters, and cuvette system, so the results presented in this report are presumably valid for these two instruments and for two other instruments from Becton Dickinson, the LSR II and FACSAria. There is also the CyFlow instrument (Partec GmbH, Münster, Germany), which can be equipped with a whole range of diode lasers, so it is possible to build a system that is even more suitable (regarding optics) for FCET studies than the commercially available setups we investigated.

FACSArray, a relatively cheap and incredibly fast instrument, seems to be a great asset for large-scale FCET measurements because it combines flow cytometric analysis with FRET to yield a powerful, high-throughput assay for the detection of molecular associations in intact cells.

Some other sets of fluorophores were also discarded from this report, namely the phycobilin proteins, variants of the green fluorescent protein (GFP), and quantum dots.

The phycobilin proteins, R-phycoerythrin and allophycocyanin, and their tandem conjugate derivatives are very popular in immunofluorescence studies because their exceptional excitability (∼2 million 1/M* cm for R-phycoerythrin) allows for the detection of hundreds of antigens (35). However, the size of these proteins is comparable to that of antibodies and thus renders them inaccurate or even impractical for detecting small molecular distances. Another consequence of their size is that they move relatively slowly and cannot rotate as freely as small fluorescent molecules, which has a tremendous effect on their orientational freedom. This inability manifests in such κ² values, where the range of FRET efficiency is impractical (36).

When fluorophores conjugated to antibodies are used as donor-acceptor pairs in FRET studies on live cells, these approaches are often restricted to the extracellular part of molecules and depend on the availability of appropriate antibodies. The recent development of GFP variants suitable for FRET has expanded the utility of this methodology by permitting the study of intracellular and extracellular processes (37–39). However, despite their known advantages and recently gained importance, there are several real disadvantages that render them difficult to use in FRET studies. Combining GFP variants to form good FRET pairs takes some amount of prudence because the excitation and detection ranges are usually quite close so that spectral crosstalk may hinder the accurate determination of FRET efficiencies. These difficulties caused by spectral overlap can be overcome by the detection of the entire spectrum and mathematically unmixing the components (40, 41). Another limitation is that most available laser lines are not optimal for excitation, and the GFP variants mostly cover the lower part of the visible spectrum, where autofluorescence greatly limits the sensitivity. Nonetheless, some new variants emitting in the red region (DsRed) may be successfully applied for FRET studies (42). A great disadvantage is the need for cotransfection of two proteins. This difficulty can be circumvented by studying the interactions of GFP variants with homotransfer, which occurs between fluorescent molecules of the same kind and manifests in the decrease of fluorescence anisotropy. This method has recently been applied to microscopy (43) and with little investment and few alterations can be implemented for flow cytometry.

The use of quantum dots in biomedical studies is an emerging methodology that should be considered in relation to FRET measurements. The main advantages of quantum dots are the wide excitation spectra, narrow and symmetrical emission spectra, good quantum yield, and photostability (44, 45). However, some of these parameters may manifest as disadvantages in FRET studies. Due to the extremely broad excitation spectrum, quantum dots cannot be used as FRET acceptors because the enhancement on the acceptor side would be small and thus compromise the accuracy of FRET measurements. When quantum dots are used as donors, the narrow emission spectra would decrease the overlap integral with the acceptor dye, thus decreasing the R₀ value for the donor-acceptor pair. Another disadvantage is their size: although the inner core is relatively small (2–5 nm), the coating for protection and biological linking increases the size to 15 to 20 nm, when FRET measurements become impractical.

In summary, to study associations of membrane proteins in cells in suspension, we have established the parameters of normalized fluorescence intensity, overlap integral, normalized FRET efficiency, and efr coefficient as crucial for FCET measurements and found that the optimal donor-acceptor pairs are Alexa546-Alexa647 and Alexa555-Cy5 in the three flow cytometers tested, with the latter also practical in microscopic acceptor photobleaching experiments. The use of high-throughput FRET measurements is greatly facilitated by the FACSArray flow cytometer. If the optical parameters of the flow cytometer change (e.g., a new laser line is introduced), a similar evaluation procedure as described in this report can be used to determine the optimal donor-acceptor pair for the given instrument setup.