L E T T E R S Many genetically encoded biosensors use F��rster resonance energy transfer (FRET) between fluorescent proteins to report biochemical phenomena in living cells. Most commonly, the enhanced cyan fluorescent protein (ECFP) is used as the donor fluorophore, coupled with one of several yellow fluorescent protein (YFP) variants as the acceptor. ECFP is used despite several spectroscopic disadvantages, namely a low quantum yield, a low extinction coefficient and a fluorescence lifetime that is best fit by a double exponential. To improve the characteristics of ECFP for FRET measurements, we used a site-directed mutagenesis approach to overcome these disadvantages. The resulting variant, which we named Cerulean (ECFP/S72A/Y145A/H148D), has a greatly improved quantum yield, a higher extinction coefficient and a fluorescence lifetime that is best fit by a single exponential. Cerulean is 2.5-fold brighter than ECFP and replacement of ECFP with Cerulean substantially improves the signal-to-noise ratio of a FRET-based sensor for glucokinase activation. Two fluorescent proteins derived from Aequorea victoria, ECFP1 and YFP2, have appropriate fluorescence excitation and emission proper- ties for the measurement of close molecular distances. When these two molecules are positioned within ���5 nm of each other3, energy can transfer from the excited state of the donor fluorophore to the unoc- cupied excited state of the acceptor fluorophore. This strategy has been used to detect molecular interactions in living cells and is the basis of a wide variety of molecular biosensors4. In principle, FRET is easily applied to a broad range of biological problems. In practice, however, these assays are often quite problematic because limitations in the fluorophore properties result in a small dynamic range. The change in the overall YFP-to-ECFP ratio is typically only 10���30% for most of the FRET-based indicators5���8, with few notable excep- tions9,10. This level of contrast pushes the limit of modern digital microscopy, because noise is often over 10% of the signal at low inten- sity levels11. One of the major disadvantages of using ECFP in FRET experi- ments is ECFP���s dim fluorescence, which results in a low signal-to- noise ratio. Although current varieties of YFP, such as Citrine12 and Venus13, are quite bright, the total brightness (proportional to the product of the extinction coefficient and quantum yield) of ECFP is ~33% of that of the most popular variant of green fluorescent protein (GFP), enhanced GFP (EGFP) (Table 1), and ~20% of that of YFP14. Figure 1a shows the 458-nm-excited emission spectra taken from cells expressing both ECFP and the monomeric variant15 of the Citrine YFP (mCit) at similar levels. Even though excitation parameters favor ECFP excitation, the peak intensity of mCit is 50% greater than the peak intensity of ECFP. One solution to the problem of poor signal-to-noise ratio with intensity-based FRET measurements is the adaptation of intensity- independent approaches to the measurement of FRET in biological specimens, such as fluorescence lifetime microscopy16. Because the fluorescence lifetime for the FRET donor (that is, ECFP) decreases when FRET occurs with the acceptor (that is, YFP), monitoring changes in the fluorescence lifetime of the donor fluorophore is a rig- orous method for detecting FRET. To examine the suitability of using ECFP as a probe for lifetime-FRET measurements, we used time- correlated, single-photon counting spectroscopy to measure the fluo- rescence lifetimes of ECFP and a mCit:ECFP fusion protein (Table 2). Unfortunately, the fluorescence lifetime of ECFP is not best fit by a single exponential. This finding is in agreement with fluorescence lifetime measurements taken in live cells expressing ECFP17. We also found that a fusion between two ECFP molecules reduced the fluorescence lifetime. This suggests that ECFP molecules have multiple excited states, which interconvert by homotransfer. To prove the presence of energy transfer in ECFP:ECFP, we compared the time- resolved fluorescence anisotropy decay of ECFP:ECFP with mono- meric ECFP (Fig. 1b). The anisotropy decay for ECFP:ECFP included a fast component not present in the anisotropy decay for ECFP alone, which indicates the presence of energy transfer. In addition, the steady-state anisotropy for ECFP:ECFP (r = 0.268) was lower than with ECFP alone (r = 0.292). We also found a ���10% reduction in the total fluorescence of ECFP:ECFP compared to ECFP alone (Fig. 1c). This proportion is consistent with the efficiency of energy transfer detected by our fluorescence lifetime measurements (FRET efficiency = 9.3%). Taken together, these results show the presence of energy transfer within ECFP:ECFP, and are consistent with the presence of multiple fluorescence states. The existence of multiple states in differ- ent ECFP molecules complicates the interpretation of the fluores- cence lifetime results for ECFP, and this limits the applicability of a lifetime assay for ECFP:YFP FRET. Department of Molecular Physiology and Biophysics, Vanderbilt University Medical Center, 735 Light Hall, Nashville, Tennessee 37232, USA. Correspondence should be addressed to D.W.P. (dave.piston@vanderbilt.edu). Published online 29 February 2004 doi:10.1038/nbt945 An improved cyan fluorescent protein variant useful for FRET Mark A Rizzo, Gerald H Springer, Butch Granada & David W Piston NATURE BIOTECHNOLOGY VOLUME 22 NUMBER 4 APRIL 2004 445 �� 2004 Nature Publishing Group http://www.nature.com/naturebiotechnology

L E T T E R S The presence of multiple fluorescent states in ECFP is likely due to the two different conformations found in the crystal structure of ECFP18. The two conformations result from alternate displacement of two hydrophobic residues (Tyr145 and His148) to the surface of the protein, which is exposed to solvent. Thus, we hypothesized that mutation of one of these residues to an amino acid with a hydrophilic side chain would stabilize a single conformation, and thus produce a CFP with a single-exponential fluorescence lifetime. Mutation of His148 to aspartate resulted in a fluorescence lifetime that was best fit by a single exponential (Table 2). In addition, the fluorescence life- time of ECFP/H148D:ECFP/H148D was comparable to that of the monomer. Therefore, the ECFP/H148D variant is much more suit- able as a FRET donor for fluorescence lifetime studies. We next looked to increase the overall brightness and usefulness of ECFP/H148D. We incorporated a mutation known to improve the folding of fluorescent proteins at 37 ��C (S72A)19, and did ran- dom mutagenesis studies on Tyr145. The brightest variant isolated (clone D10) was found to have a greater extinction coefficient than ECFP/H148D, although the quantum yield was less than that of ECFP/H148D (Table 1). Because DNA sequencing revealed substi- tution of Tyr145 with glycine in the D10 variant, we reasoned that a Y145A mutation would provide additional stability. The resulting variant (ECFP/S72A/Y145A/H148D), which we have named Cerulean, is 2.5-fold brighter than ECFP (Table 1) and its fluores- cence lifetime is best fit by a single exponential (Table 2). Cerulean is more resistant to photobleaching than ECFP and has equivalent pH stability (Table 1). Unexpectedly, the absorption and fluores- cence spectrum of Cerulean are similar to those of ECFP, with only a slight reduction in the 505-nm emission peak observed (Fig. 2a). Although the presence of two spectral peaks suggests the existence of multiple fluorescent states, these states are not resolved in the fluorescence lifetime measurements. In addition, no energy loss 446 VOLUME 22 NUMBER 4 APRIL 2004 NATURE BIOTECHNOLOGY Figure 1 Limitations of using ECFP in FRET experiments. (a) The mean relative fluorescence emission spectra of COS-7 cells electroporated with equal amounts of plasmid DNA encoding ECFP and mCit is shown in the bottom panel as detected by spectral imaging (458-nm excitation n = 7 error bars indicate s.e.m.). Spectra were normalized to total fluorescence. Photobleaching of mCit with a 514-nm laser did not result in increased ECFP intensity, indicating the absence of FRET between mCit and ECFP. The top panel shows reference spectra for ECFP ( ) and mCit ( ). (b) The time-resolved anisotropy decay of 1 ��M ECFP ( ) and ECFP:ECFP ( ) was measured using time-resolved spectroscopy. The fast decay in ECFP:ECFP anisotropy indicates the presence of energy transfer. (c) Steady-state fluorescence emission spectra were collected (434-nm excitation) for equivalent absorbances of ECFP:ECFP (dotted line) and ECFP (solid line). Data were obtained using magic angle conditions to eliminate the effects of polarization. Figure 2 Comparison of Cerulean and ECFP. (a) Absorbance (dotted lines) and fluorescence emission spectra (solid lines) of purified ECFP (black) and Cerulean (blue). Measurements were taken in TE buffer (pH 8.0) and normalized to peak intensity. (b,c) Cyan and yellow fluorescence was resolved by spectral imaging (458-nm excitation) and linear unmixing in COS-7 cells expressing either ECFP:mCit (b) or Cerulean:mCit (c). A histogram of the intensity distribution of pixels inside the cell for the cyan (cyan) and mCit (yellow) channels is shown on the right. (d) Cyan and mCit channels are shown from a ��TC3 cell expressing the Cerulean glucokinase biosensor before and after insulin stimulation (100 nM, 2 min). (e,f) Spectral imaging (800-nm excitation) was used to collect the emission spectra from cells expressing either the glucokinase biosensor containing ECFP and mCit (e) or the glucokinase biosensor containing Cerulean and mCit (f). Intensities were normalized to peak cyan fluorescence for both pre- (solid line) and post- (broken line) insulin stimulation of ��TC3 cells (n = 4). Error bars indicate s.e.m. a b c d e f a b c �� 2004 Nature Publishing Group http://www.nature.com/naturebiotechnology