An Ion-insensitive cAMP Biosensor for Long Term Quantitative Ratiometric Fluorescence Resonance Energy Transfer (FRET) Measurements under Variable Physiological Conditions*

Ratiometric measurements with FRET-based biosensors in living cells using a single fluorescence excitation wavelength are often affected by a significant ion sensitivity and the aggregation behavior of the FRET pair. This is an important problem for quantitative approaches. Here we report on the influence of physiological ion concentration changes on quantitative ratiometric measurements by comparing different FRET pairs for a cAMP-detecting biosensor. We exchanged the enhanced CFP/enhanced YFP FRET pair of an established Epac1-based biosensor by the fluorophores mCerulean/mCitrine. In the case of enhanced CFP/enhanced YFP, we showed that changes in proton, and (to a lesser extent) chloride ion concentrations result in incorrect ratiometric FRET signals, which may exceed the dynamic range of the biosensor. Calcium ions have no direct, but an indirect pH-driven effect by mobilizing protons. These ion dependences were greatly eliminated when mCerulean/mCitrine fluorophores were used. For such advanced FRET pairs the biosensor is less sensitive to changes in ion concentration and allows consistent cAMP concentration measurements under different physiological conditions, as occur in metabolically active cells. In addition, we verified that the described FRET pair exchange increased the dynamic range of the FRET efficiency response. The time window for stable experimental conditions was also prolonged by a faster biosensor expression rate in transfected cells and a greatly reduced tendency to aggregate, which reduces cytotoxicity. These properties were verified in functional tests in single cells co-expressing the biosensor and the 5-HT1A receptor.

Several constructs of fluorescent proteins have been developed to measure cyclic adenosine monophosphate concentration ([cAMP]) in living cells (1,2). In these biosensors, Epac1 or Epac2 (exchange proteins directly activated by cAMP) (3,4) or their cAMP binding domains are used as backbones (5)(6)(7) to which fluorescent proteins are tethered. The binding of cAMP to the biosensors leads to a conformational change, which alters the relative distance and/or orientation between the FRET pair, thus changing the efficiency of energy transfer and donor and acceptor fluorescence emission.
For such biosensors with fixed donor-acceptor stoichiometry, common ratiometric fluorescence measurements can be performed to identify changes in FRET. These changes can be calibrated with reference ligand solutions using a single fluorescence excitation wavelength (e.g. see Ref. 8). In contrast to multi-excitation wavelength approaches like the Ef DA /␥ analysis (1,9) or lux-FRET (10), which require more complex algorithms and calibrations, single excitation wavelength measurements profit from a higher time resolution and reduced bleaching of the fluorophores. Ratiometric analysis also gives a better signal-to-noise ratio than the more complex approaches (11).
cAMP biosensors using the conventional FRET pair (enhanced cyan fluorescence protein (eCFP) 2 and enhanced yellow fluorescence protein (eYFP)) (12) might be critical when ratiometric cAMP measurements are performed with singlechannel fluorescence excitation, as demonstrated in a previous work (1). Such ratiometric FRET analysis of the intensity ratio between eCFP and eYFP emission of the Epac1 biosensor CFP-Epac(␦DEP-CD)-YFP (Ponsioen et al. (7)) is susceptible to eYFP fluorescence intensity changes whenever ion concentrations fluctuate (e.g. [Cl Ϫ ] (13)). Considering that ion concentrations change significantly during physiologically cellular activities (e.g. [H ϩ ] and [Cl Ϫ ] change with normal synaptic interactions of neurons) (e.g. see Refs. 14 and 15), FRET signals can be affected by secondary effects and may not reliably monitor the real [cAMP]. The condition becomes worse under pathologic conditions, when ionic homeostasis is unbalanced, further falsifying [cAMP] readouts. During hypoxia, for example, intracellular pH decreases, and chloride ions accumulate within neurons, whereas calcium influx is increased or calcium is released from intracellular stores (16,17). Exchange of the FRET pair had already been used to gain a higher FRET effi-ciency of the traditional cAMP biosensor (18); no special focus, however, was directed on ion sensitivity. Particularly for quantitative ratiometric FRET analysis in vivo, it seemed necessary, therefore, to exchange the conventional FRET pair eCFP/eYFP for other fluorophores that are less sensitive to changes in these ion concentrations. From a variety of improved CFP and YFP derivatives, we have chosen mCerulean and mCitrine, which are less sensitive to pH changes and show only insignificant halide sensitivity. mCitrine also has a much higher photostability as compared with eYFP and has an improved folding efficiency (19). mCerulean is a better FRET donor than eCFP when combined with mCitrine (20,21). Förster distance R 0 of the mCerulean/mCitrine FRET pair is almost 10% larger than for eCFP/mCitrine as calculated by a 68% larger quantum yield of mCerulean. Another important advantage also is that the monomeric versions (22) of Cerulean and Citrine show a much lower tendency to oligomerization as compared with the eCFP/ eYFP pair, which may cause additional artifacts.
Here, we report a reduced ion sensitivity (to pH and Cl Ϫ and indirectly also Ca 2ϩ ) of quantitative radiometric measurements by exchanging the eCFP/eYFP FRET pair of the biosensor for mCerulean/mCitrine. In addition, we report an improvement of the biosensor aggregation behavior.

Cell Culture
Mouse N1E-115 neuroblastoma cells from the American Type Culture collection (LGC Promochem, Wesel, Germany) were grown at 37°C and 5% CO 2 in Dulbecco's modified Eagle's medium (Sigma-Aldrich) containing 10% fetal calf serum (FCS) and 1% penicillin/streptomycin. Neuroblastoma cells were seeded at low density (1 ϫ 10 6 cells) either in 60-mm dishes (for fluorescence spectroscopy measurements) or in 10-mm dishes, including glass coverslips on the bottom (for microscopic measurements). After 24 h, cells were transfected with appropriate vectors using Lipofectamine2000 reagent (Invitrogen) according to the manufacturer's instructions. 3 h after transfection, cells were serum-starved overnight and then used in the experiment. We had to avoid longer incubation periods, because EPAC* proteins aggregated, which impedes direct comparison with the biosensor CEPAC*.

Changing of Intracellular pH
To change the intracellular pH (pH i ) of N1E cells, we used a modified protocol described previously (25), using extracellular buffer exchange with different pH values to affect intracellular conditions. Extracellular buffer (150 mM NaCl, 5 mM KCl, 2 mM CaCl 2 , 1 mM MgCl 2 , 10 mM HEPES, 10 mM sodium-D-glucose) was prepared with pH values ranging from pH 6.5 to pH 8.0 and given to the N1E cells, immediately producing a pH i shift from 7.05 to 7.4, respectively. Under normal culture conditions at an extracellular pH of 7.4, N1E neuroblastoma cells exhibit a pH i of 7.35. The pH i shift was verified by pH indicator dye SNARF-5F (Invitrogen), which was also used to calibrate the pH range. The experiments with cAMP biosensor transfected cells were performed in an upright fluorescent microscope, whereas the experiments with SNARF-5F were performed in a laser-scanning microscope with a spectral resolving emission unit in order to measure the full emission spectrum of SNARF-5F.

Receptor Activation in N1E Cells
N1E-115 cells co-transfected with 5-HT 1A R and either cAMP biosensor were kept in an extracellular solution of 150 mM NaCl, 5 mM KCl, 2 mM CaCl 2 , 1 mM MgCl 2 , 10 mM HEPES, 10 mM sodium-D-glucose (pH 7.4), and 100 nM forskolin. 5-HT 1A R agonist and antagonist were applied by exchanging the standard bath solution for a solution containing or 1 M serotonin or WAY 100635 (Sigma-Aldrich), respectively. Experiments were performed in an upright fluorescence microscope.

Cell Viability Test
To test the viability of cells transfected with either of the two constructs, the MTS cell proliferation assay (Promega) was used. Approximately 5 ϫ 10 3 N1E cells were seeded into 96-well plates and allowed to attach overnight. Transfection with a mock control, CEPAC*, or EPAC* was carried out with Lipofectamine2000 according to the manufacturer's instructions. After 4,8,12,16,and 20 h, absorbance at 485 nm of triplicates of each transfection was measured in a SkanIt plate reader (Thermo Scientific). MTS is reduced by cells into a formazan product, and the quantity of formazan product as measured by the amount of absorbance is directly proportional to the number of living cells in culture. Normalized absorbance was plotted against time after transfection.

Capturing of Dendritic Spines
Primary neurons transfected with the cAMP biosensors were kept in an extracellular solution of 128 mM NaCl, 2 mM KCl, 2 mM CaCl 2 , 1 mM MgCl 2 , 10 mM HEPES, 10 mM sodium-D-glucose (pH 7.4). Dendritic spines could be captured using an inverse laser-scanning microscope.

Fluorescence Spectroscopy
The sample preparation was obtained as described previously (1), and only modifications are mentioned here. For [cAMP] measurements, transfected N1E-115 cells were suspended in a buffer and lysed by ultrasonic treatment (DIGITAL Sonifier S-450D, Branson (Danbury, CT)). The homogenate was centrifuged for 1.5 h at 16,000 ϫ g and 4°C to extract the cytosol. For [cAMP] measurements, the lysis buffer contained 140 mM KCl, 5 mM NaCl, 1 mM MgCl 2 , and 10 mM HEPES at pH 7.2. For measuring the pH dependence, the solution contained 140 mM KCl, 5 mM NaCl, 1 mM MgCl 2 , and 10 mM HEPES at pH 7.2. For measuring the Ca 2ϩ dependence, a calcium-free lysis buffer was used containing 140 mM KCl, 5 mM NaCl, 1 mM MgCl 2 , 1 mM EGTA, and 10 mM HEPES at pH 7.2. Such lysate contained a free [Ca 2ϩ ] below 10 nM as a starting condition for the calcium titration as proven by calcium measurements with Fluo-5 (Invitrogen) (data not shown). The nominal free [Ca 2ϩ ] was changed by titrating a defined amount of calcium to the cuvette and was estimated by the Max Chelator Ca-Mg-ATP-EGTA Calculator version 1 (26,27) (available on the World Wide Web). Addition of high calcium concentrations led to a pH change due to proton release from EGTA. Therefore, the pH in the cuvette was adjusted during each test. For measuring the Cl Ϫ dependence, a chloride-free lysis buffer was used, containing 140 mM K ϩ , 5 mM Na ϩ , 1 mM Mg 2ϩ , 147 MeSO 3 Ϫ , and 10 mM HEPES at pH 7. Spectroscopic measurements were performed as described previously (1) with a Fluorolog-322 (Horiba Jovbin Yvon, Munich, Germany). A donor excitation wavelength of 420/2 nm was chosen. A second excitation wavelength of 500/2 nm was necessary to apply the formalism of Equation 2. Reference emission spectra (F D ref () and F A ref ()) of the donor and acceptor fluorophores used for the unmixing procedure (see Equation 1) were obtained at 420/2 nm excitation from cells transfected with donor or acceptor only. To estimate the spectral contributions of light scattering and autofluorescence of cells as an additional background component, reference spectra of pcDNA-transfected cells were used.

Microscopy
Preparation-A glass coverslip carrying the biosensor transfected or loaded N1E-115 cells or primary neurons was positioned in a perfused bath chamber of a microscope. Cells were perfused with an appropriate bath solution at room temperature. For sequential measurements with different solutions, total solution exchanges in the chamber were performed in a time range of 1 min.
Wide Field Fluorescence Microscopy-We used an upright epifluorescence microscope equipped with a water immersion objective (XLUMPlanFI, ϫ20, numerical aperture 0.95, Olympus, Germany). A 100-watt xenon lamp attached to a monochromator (Optoscan, Kinetic Imaging) served as an excitation light source and was coupled to the microscope via fiber optics. Emission intensities were measured at two excitation wavelengths using 420/10 nm ( 1 ) and 500/10 nm ( 2 ), frequently called donor and acceptor excitation, respectively. Fluorescence emission was separated from the excitation light by a dichroic mirror (505 nm). Using a DualView (Optical Insights, Tucson, AZ), the fluorescence emission signal was split by a dichroic mirror (515 nm) for the donor channel at 470/30 nm and the acceptor channel at 535/30 nm. This allowed us to acquire three images with an iXon camera DV887DCS (Andor Technology, South Windsor, CT): (a) the "donor image" at donor excitation and donor emission wavelength; (b) the "FRET image" at donor excitation and acceptor emission wavelength; (c) the "acceptor image" at acceptor excitation and acceptor emission wavelength. Due to the DualView, the donor image and the "FRET image" were obtained simultaneously, which required only two exposures. Exposure times (of about 3 s) were chosen according to the fluorescence intensity of the cells and were equal for all image series of an individual experiment. Special care was taken that bleaching can be ignored.
Laser-scanning Microscope-For capturing dendritic spines and for recording intracellular pH in single cells with SNARF-5F, we used a confocal laser-scanning microscope LSM 510 META from Zeiss (Göttingen, Germany). CEPAC* in spines was observed at 458 nm excitation, and a bright emission bandwidth was observed from 464 to 603 nm using a C-Apochromat ϫ 63/1.2 W corrected objective. For the pH i recording, the emission spectrum (562-646 nm) of SNARF-5F was detected by using the META DETECTOR with a resolution of 8 points (10 nm bandwidth) at an excitation of 543 nm and a Plan-Neofluar ϫ 40/1.3 oil objective.

Data Analysis
Formation of Fluorescent Granules-Single images were taken with a wide field microscope at various times after trans-

Ion-insensitive cAMP Biosensor
JULY 1, 2011 • VOLUME 286 • NUMBER 26 fection to measure protein aggregation seen as "granules" with the ImageJ software version 1.41 (National Institutes of Health, Bethesda, MD). Protein aggregation was determined in each cell using a line scan through the entire cells, excluding obvious organelle structures, such as the nucleus. This line scan represents the fluorescence intensity at any given pixel. Calculating the S.D. value for each given fluorescence intensity plot gave a measure of the granularity of the cell because the line profile did not vary greatly in cells showing a uniform fluorophore expression, whereas in cells showing aggregates, the line profile changed between regions with high and low fluorescence. For the different time points, at minimum, six cells were analyzed, and the mean granularity (as the sum of S.D. values from different scans) was plotted over time after transfection.
Length of Neurites-To image a single primary cell, multiple images were taken and stitched together using the ImageJ plugin MosaicJ (28). Neurite length was then measured on the reconstructed image using the ImageJ plugin NeuronJ (29).
Displaying Dendritic Spines-Captured z-stacks of dendritic spine pictures were projected on a single picture by using the maximal z-projection of the ImageJ software version 1.41.
CEPAC* and EPAC* Calibration-We calibrated the [cAMP] sensitivity of CEPAC* and EPAC* by the "sensitized emission" FRET signal using a fluorescence spectrometer (1) as described previously. Spectra were obtained at two excitation wavelengths ( i ): at donor excitation wavelength ( 1 ), where mainly the donor is excited, and at the acceptor excitation wavelength ( 2 ), where the donor must not be excited (9). For each excitation wavelength, the CEPAC* as well as the EPAC* fluorescence signals F i () were fitted with a linear combination of the donor D and acceptor A reference spectra.
The apparent donor concentration, [D i ], and the apparent acceptor concentration, [A i ], were used as fitting factors. The donor F D Ref and the acceptor F A Ref reference spectra for both excitation wavelengths i were obtained in a reference measurement of cells containing only donor and acceptor, respectively.
The apparent acceptor concentrations [A i ] at both excitation wavelengths were used to calculate the following term, where ␣ represents the relative acceptor emission intensity ratio used for the two excitations obtained in a separate experiment using an "acceptor only" sample, ␥ is the relative acceptor/donor extinction ratio, and f DA is the fraction of biosensor proteins in FRET state. For calibration of the FRET biosensors CEPAC* and EPAC*, the cAMP dependence of Ef DA /␥ was fitted by an adapted Hill equation, where n H is the Hill coefficient, indicating the amount of cAMP-binding places; p o and p max are offset and maximum amplitude parameters, respectively; and EC 50 is [cAMP] when 50% of the cAMP binding sites are occupied. Also the cAMP dependence of the acceptor/donor ratio as well as the ion dependences were fitted by a similar Hill equation. The apparent FRET efficiency (Ef DA ) was calculated from the donor fluorescence signal in quenched (F D,FRET ) and nonquenched (F D,NON-FRET ) state (30).

Performance and Calibration of CEPAC*-[cAMP]
calibration curves of both biosensors were measured to ensure similar binding properties and to compare the dynamic range of the FRET signals. We exposed cell lysates from CEPAC* or EPAC*transfected cells to various [cAMP] values, ranging from nominal 0 to 1 mM, and measured the biosensor FRET signal acceptor/donor ratio or Ef DA /␥ as a function of [cAMP]. Fluorescence spectra were recorded, as described previously (1), and the contributions of the donor and acceptor reference spectra were calculated as shown in Fig. 1, A and B. Both biosensors revealed a similar behavior regarding the contribution of reference spectra; the donor emission signal was decreased, and the acceptor emission signal increased with a rise of [cAMP], as expected for FRET changes. During donor excitation, photon counts for the donor and acceptor component of CEPAC* reached similar values in contrast to EPAC*, which is beneficial for the signal/noise ratio of the ratiometric FRET analysis. For both biosensors, we compared the FRET signal acceptor/donor channel ratio as the ratio between the acceptor and donor intensities (here at 420 nm excitation; Fig. 1, C and D). The intensities of the acceptor and donor channels were corrected for spectral bleed-through and background, whereas the channel bandwidth was adapted to that used in the microscope. Due to the higher energy transfer efficiency of the FRET pair mCerulean/mCitrine compared with eCFP/eYFP, the [cAMP] calibration curves of the acceptor/donor channel ratio of CEPAC* started at a higher value (0.94 Ϯ 0.03) than for EPAC* (0. 43  FRET Efficiency-Besides verifying comparable functions of both biosensors, we also determined the FRET efficiency (E) for both biosensors by using the multi-excitation wavelength approach to calculate Ef DA /␥. Fig. 1, E and F, shows the [cAMP] calibration curves for both biosensors, using Ef DA /␥ as the signal. Fitting of these curves with a Hill equation (Equation 3) gave Hill coefficients and EC 50 values that were similar to those obtained in the ratio analysis of both biosensors. We estimated the absolute value of E from the donor quenching (Fig. 1, A and  B) and Ef DA /␥. Considering that the donor is partially quenched for high [cAMP] (low FRET), CEPAC* reached a value for E of 26.9 Ϯ 0.4% in cAMP-free solution, when all biosensor molecules are in the highest FRET state (assuming f DA ϭ 1). In the low FRET state, when all biosensor molecules are bound to cAMP, E remained at 13.3 Ϯ 0.7%. Thus, the dynamic range resulted in ⌬E ϭ 13.6%, which corresponds to 51% of the maximum E value. For EPAC*, E changed only from 29.4 Ϯ 4.9% to 22.9 Ϯ 2.9%, with ⌬E ϭ 6.5%, revealing less accuracy due to the lower donor intensity than in the case of CEPAC*. In summary, CEPAC* provides a larger dynamic range in FRET efficiency than EPAC*. Thus, CEPAC* offers a better signal/noise ratio.
Sensitivity to Changes of Ion Concentration-As mentioned before, the fluorescence of GFP-based fluorophores are affected by ion concentration changes. In order to assess the degree of such influences on FRET measurements with cAMP biosensors, fluorescence spectra of CEPAC* and EPAC* were measured at various ionic concentrations by a fluorescence spectrometer. Fig. 2 illustrates pH, Ca 2ϩ , and Cl Ϫ dependences of the unmixed donor and acceptor components. Fig. 3 shows the acceptor/donor ratio, as used for ratiometric FRET analysis, and the FRET value Ef DA /␥. All ion dependence curves were described by a Hill kinetic as shown in Figs. 2 and 3. The data of the Hill fits (EC 50 and n H ) are presented in Table 1. Among all three ion dependences, the pH dependence exhibited the strongest effect on the acceptor/donor ratio of both biosensors

Ion-insensitive cAMP Biosensor
JULY 1, 2011 • VOLUME 286 • NUMBER 26 ( Fig. 3A). Although CEPAC* revealed a relatively weak pH sensitivity of only 14% change within a range from pH 6.8 to pH 7.5, a much stronger pH dependence of about 71% change was measured for EPAC*. The high pH dependence seems to be predominantly caused by the acceptor because the donor components of both biosensors revealed a much lower sensitivity (Fig. 2, A and B). As shown in Fig. 3B, the Ef DA /␥ values exhibited similar pH stability for both cAMP biosensors within a range from pH 5.5 to 8.5.
The donor and acceptor components (Fig. 2, C and D, black/ gray lines) exhibited no significant dependence directly to calcium when pH was constantly regulated to 7.2 during calcium titration (8% Ϯ 2% and 9% Ϯ 3% max. change for mCerulean and eCFP, as well as 5% Ϯ 3% and 8% Ϯ 4% max. change for mCitrine and eYFP). Consequently, the ratio of CEPAC* as well as EPAC* (Fig. 3C, black/gray lines) remained almost constant (2% Ϯ 1% and 3% Ϯ 4% max. change). However, as Ca 2ϩ is rapidly bound to intracellular buffers in vivo, there arise indirect effects through the exchanges of H ϩ from the calcium-buffers. We mimicked this proton-carried secondary calcium effect in vitro by leaving the pH to change freely after a first adjustment to pH 7.2 before any addition of calcium. EGTA was acting as a replacement for intracellular calcium buffer-proteins. Under these conditions, the calcium dependence of the donor and acceptor components (Fig. 2, C and D, blue lines) exhibited a quenching of the fluorescence at increasing [Ca 2ϩ ] levels mainly for the acceptor. At very low [Ca 2ϩ ] levels of 1-30 nM, there were no significant changes in the fluorescence intensity of donor and acceptor components. At physiological [Ca 2ϩ ] fluctuations, ranging from 30 nM up to several hundred M, a strong calcium dependence of 51 Ϯ 1% fluorescence reduction was seen for the eYFP component, whereas it was significantly lower (10 Ϯ 2%) for mCitrine. The donor components did not show any significant calcium dependence (4 Ϯ 1% and 9 Ϯ 4% reduction for mCerulean and eCFP). The acceptor/ donor ratio of EPAC* changed within the same [Ca 2ϩ ] range (Fig. 3C). Consequently, the ratio of CEPAC* remained almost constant (6 Ϯ 2%), whereas the ratio of EPAC* was reduced by  Table 1. Fig. 3D exhibited a [Ca 2ϩ ] insensitivity in the range from 1 nM to 1 mM for both biosensors.

Ion-insensitive cAMP Biosensor given in
The chloride dependence of the unmixed donor and acceptor component is shown in Fig. 2 (E and F). Within a [Cl Ϫ ] fluctuation range from 400 M to 80 mM, the donor components remained stable (changes of only 3 Ϯ 1% and 8 Ϯ 8% for mCerulean and eCFP). The strongest dependence was observed for the eYFP component of EPAC*, which decreased by 28 Ϯ 4%, whereas mCerulean varied only by 3 Ϯ 2%. A similar [Cl Ϫ ] characteristic was found in the FRET ratio (Fig.  3E). Although the ratio remained constant for CEPAC* (Ϯ1%), it dropped by about 34 Ϯ 2% for EPAC*. Consequently, the FRET value, Ef DA /␥, shown in Fig. 3B, exhibited a high [Cl Ϫ ] stability in the range of 400 M to 80 mM for both cAMP biosensor versions.
In summary, the results show that there is an inaccuracy in ratiometric FRET measurements using the cAMP biosensors that may be caused by pH fluctuations and to a smaller extent during physiological fluctuations in chloride concentrations. Such imprecision in [cAMP] estimates measured with EPAC* are significantly reduced when CEPAC* is used. The multi-excitation wavelength approach, Ef DA /␥, however, showed no significant ion sensitivity.
Formation of Fluorescent Granules-In our studies, we observed a progressive formation of fluorescent granules in cells expressing EPAC*. The fluorescent granules seem to consist of protein aggregates that severely disturb the analysis of FRET signals by depleting the available biosensor from surrounding regions. The aggregates also seem to exert some degree of cytotoxicity. To analyze this phenomenon, we compared the protein expression rate and cell viability of both biosensor constructs.
The expression of CEPAC* occurred much faster than for EPAC* (Fig. 4A). In the case of CEPAC* expression, fluorescent cells appeared already at 6 h after transfection, and there was a sufficient number of cells 10 h after transfection. As for the expression of EPAC*, there was a delay of 10 h  Table 1.

Ion-insensitive cAMP Biosensor
until the first fluorescent cells were detectable, and a 14 -16 h delay was necessary to allow measurements in a sufficient number of cells.
In cell cultures expressing CEPAC*, there were no signs of protein aggregation until 20 h post-transfection time, which normally was the end of the analysis time frame (Fig. 4, B and C).  Cells even revealed significant expression up to 2 days after transfection (not shown). In contrast, EPAC* showed an early onset of protein aggregation already at 16 h after transfection, and after 20 h, all cells showed protein aggregation. Cytotoxicity of the aggregated EPAC* became obvious when transfected cells started to get smaller showing signs of apoptosis and cell density within the culture decreased. In a cell viability assay of cells transfected with either CEPAC* or EPAC* using mocktransfected cells as control (Fig. 4D), we observed that CEPAC*-transfected cells showed a similar viability as mocktransfected cells, whereas EPAC*-transfected cells revealed a decreasing viability already 10 -12 h after transfection, which is the time window when EPAC* expression started (Fig. 4A).

Operation of CEPAC* and Dependence of the Biosensor Signal on Intracellular pH Fluctuations in Single Cell Experiments-
The function of CEPAC* was confirmed by FRET measurements in single N1E-115 neuroblastoma cells co-transfected with plasmids encoding for CEPAC* and for 5-HT 1A R. Temporal changes in [cAMP] were measured as described previously (1). 5-HT 1A R is a G i -protein-coupled seven-helix receptor inducing inhibition of adenylyl cyclases and consequently a decline of intracellular [cAMP] (31). In order to effectively provoke a decrease of [cAMP] levels by receptor agonist application, the cAMP level was first elevated by application of the adenylyl cyclase activator forskolin (100 nM) 10 min before the receptor agonist was applied. The applied forskolin concentration corresponded with its EC 50 value (98 Ϯ 45 nM) for cAMP formation, as determined in separate experiments (data not shown). The 5-HT 1A R agonist serotonin applied in a 1 M concentration induced an increase in the acceptor/donor channel ratio (implying a [cAMP] decrease), whereas the receptor antagonist WAY 100635, applied at 1 M, returned the ratio values to control levels. Because the absolute values of the acceptor/donor channel ratio depend on the microscopic equipment, they are not necessarily equal to the acceptor/donor channel ratio presented in the cAMP calibration curve of Fig. 1, C and D. Therefore, we performed a separate cAMP calibration curve for the microscope (data not shown) in order to assign quantitative [cAMP] values. This revealed a 5-HT 1A Rinduced change of [cAMP] from ϳ70 to 7 M and a recovery to 50 M after application of the 5-HT 1A R antagonist (mean values from 26 cells; Fig. 5A). Fig. 5B demonstrates the variability between cells of the quantitative [cAMP] values, including two measurements of each cell before and during serotonin stimulation. The analysis shows that several cells reached [cAMP] levels above 1 mM, which is outside the dynamic range of CEPAC* where the ratio to [cAMP] assignment is defective. The time constant of about 100 s of the increase and decrease seems to be a consequence of the slow exchange of the solution in the bath chamber.
Although in standardized tests, CEPAC* are clearly superior to EPAC*, we studied whether the biosensor readout of both biosensors is influenced by pH i fluctuations within (patho)physiological range. To simulate hypoxia (32, 33), we exposed the cells to a series of extracellular solutions buffered at different pH (pH 7.4, 6.5, and 8.0). Flögel et al. (25) showed that a change of extracellular pH results in a significant change of pH i in N1E-115 cells. In calibration experiments with a pH indicator, SNARF-5F (data not shown), we verified that such manipulations result in pH i values of 7.35, 7.05, and 7.4 under our experimental conditions. Fig. 5C shows the result of the normalized biosensor readout (here a record of one cell in each case) by changing the intracellular buffer of CEPAC*-and EPAC*-transfected cells. Application of 1 M forskolin after the series of pH change resulted in an acceptor/donor channel ratio decrease (implying a [cAMP] increase), which proved that cells were still functional and intact. Fig. 5D shows an average of normalized acceptor/donor channel ratio of 12-13 cells at the maximum pH difference and the maximum forskolin difference for both biosensors.
Although the pH i change did not affect CEPAC*, EPAC* reacted with an apparent false positive cAMP signal. After changing the pH i , both biosensors still reacted normally to forskolin, indicating a rise in [cAMP] i . The false positive cAMP signal (Ϫ0.09 Ϯ 0.01) of EPAC* was nearly twice as large as the forskolin response (Ϫ0.05 Ϯ 0.01).
Impact on Neuronal Dendrite Growth and Measurable cAMP Activity-The impact of biosensors on the dendrite growth of neurons was measured in primary hippocampal cells that were dissociated and transfected with CEPAC* or EPAC* after 4 days of explantation on a coverslip. After an additional 4 days allowed for biosensor protein expression and dendrite growth, cells were fixed and analyzed. We found more neurons transfected with CEPAC*, revealing a more homogenous distribution of CEPAC* even in spines of dendrites Fig. 6C, whereas EPAC* was clustered in many cases (Fig. 6B), and other neurons were dead. As a parameter for neuronal growth, the total dendrite length of still intact neurons was analyzed. Fig. 6A shows an exemplary primary cell transfected with CEPAC*, including the lines for length measurements. Fig. 6D compares the total dendrite length of 6 neurons/coverslip for each biosensor. EPAC* shows a clear tendency (p ϭ 0.07 for t test) to have a reduced dendrite growth compared with CEPAC*.
Measurements of cAMP levels after receptor activation in CEPAC* ϩ 5-HT 1A R-and EPAC* ϩ 5-HT 1A R-co-transfected cells revealed that not all of these cells reacted to receptor activation. Therefore, we selected a region of interest on every coverslip (e.g. see Fig. 5B) and analyzed a comparable number of cells in their average response rate. The response rates of 7-8 different preparations are presented in Fig. 6D. Cells expressing CEPAC* revealed a much higher response rate of about 89 Ϯ 9% as compared with only 33 Ϯ 7% of cells expressing EPAC*.

DISCUSSION
Most tandem fluorophore FRET biosensors have been designed for ratiometric measurements, which use single fluorescence excitation rather than multi-excitation protocols that require more complex formalisms and calibration steps. It has been shown, however, that a cAMP biosensor with a conventional eCFP/eYFP FRET pair is not optimal for studying physiological processes quantitatively because changes in the ionic environment lead to changes of the acceptor fluorescence intensity (1) that are unrelated to cAMP. Improvements of this biosensor with respect to its FRET efficiency were achieved by exchanging the FRET pair with advanced fluorophores (18). Here, we investigated the consequences of such an improvement on the ion dependence (within the physiological range) and the aggregation behavior of the biosensor. We replaced the eCFP/eYFP FRET pair (EPAC*) with mCerulean and mCitrine (CEPAC*), respectively, in order to allow reproducible and reliable quantitative ratiometric [cAMP] analyses. In the following, we discuss the function, FRET efficiency, ion dependence, and aggregation behavior of the new biosensor in comparison with EPAC*. CEPAC* has a single cAMP-binding domain (Hill coefficient of 1) and therefore shows the same biosensor function as EPAC*. Binding of cAMP leads to a decrease of acceptor and an increase of donor fluorescence (Fig. 1, A and B). A similar binding affinity toward cAMP can be assumed, because the [cAMP] calibration curves reveal an EC 50 value for CEPAC* similar to that for EPAC* and correspond to the value presented by Ponsioen et al. (7) for EPAC*.
The FRET efficiency difference (⌬E) revealed a higher dynamic range for CEPAC* (CEPAC* ⌬E ϭ 13.6% versus EPAC* ⌬E ϭ 6.5%). Also, other groups have shown that such exchange of the FRET pair by advanced fluorophores results in an increase of the ⌬E (18) and improves the signal/noise ratio. The signal/noise ratio of CEPAC* was further improved due to the equally strong brightness of donor and acceptor under experimental conditions (mainly at low [cAMP]; Fig. 1, A  and B).
We confirmed that CEPAC* was fully functional by measuring quantitative [cAMP] response curves (Fig. 5B) in single cell experiments after activation and blockade of a G i -coupled receptor. As expected, CEPAC* detected a decrease of [cAMP] induced by 5-HT 1A R activation that was blocked by an antagonist application. Preincubation of cells in 100 nM forskolin increased the [cAMP] on average to 100 M and in some cells even over 1 mM. Such a large concentration might not happen in all of a cell cytosol under physiological conditions, but this helped us to observe a clear [cAMP] decrease.
With regard to the sensitivity of the FRET signal to changes of H ϩ , chloride, and calcium concentration (Fig. 3), we found different characteristics between the multi-excitation approach Ef DA /␥ and donor/acceptor ratio using a single excitation. For Ef DA /␥, both biosensors exhibit a similar stability at varying ion concentrations, which can be expected because Ef DA /␥ is insensitive against non-FRET-related intensity changes of the acceptor (1). In contrast, the acceptor/donor ratio of the EPAC* biosensor is heavily disturbed by ionic concentration changes within physiological levels; no such disturbances were found for CEPAC*. These changes were mainly attributed to changes in acceptor fluorescence intensity (Fig. 2). In the case of H ϩ and Cl Ϫ , the Hill values of the present measurements for eYFP (EC 50 : pH 7.2, and 130 mM [Cl Ϫ ]) correlate well with published data for the single fluorophore (EC 50 : pH 6.9, and 110 mM [Cl Ϫ ]) (34). Such data were not available for the Ca 2ϩ dependence of the used fluorophores. The acceptor/donor ratio of both biosensors is insensitive to direct calcium changes measured in vitro (Fig. 3C, black/gray lines). It is, however, important to note that binding of Ca 2ϩ to calcium buffering proteins results in a proton release (39 -41). We mimicked this in vitro by using EGTA as a replacement for a calcium-buffer protein.
Under these conditions, CEPAC* exhibits only a slight decrease in acceptor/donor ratio of about 6% by varying [Ca 2ϩ ] within the functionally important range of 30 nM to 1 M. In contrast to EPAC*, such alterations in the calcium level lead to major changes of the acceptor/donor ratio (42%; Fig. 3C, blue lines), which cover almost the complete dynamic range of the biosensor (46%; Fig. 1D). In vivo, such changes may not be negligible when calcium is released at high M concentrations within the microdomains near Ca 2ϩ -channels or at release sites from intracellular stores (42)(43)(44)(45). In this case, the proton release may change pH i and, therefore, contribute to the EPAC* signal. We ascribe this as a secondary calcium effect on EPAC* in vivo. Such secondary calcium effects can be neglected when CEPAC* is used. In the case of pH, the acceptor/donor ratio changes were found to be about 4 times larger than the complete dynamic range of the biosensor. Within a normal physiological pH range (pH 6.8 -7.5), however, the ratio changes were negligible (14%) for CEPAC*, whereas they were significantly higher (71%) for EPAC*. Finally, changes in chloride concentration also influenced the acceptor/donor ratio of EPAC* but to a lesser extent than other ions. However, in the case of CEPAC*, there was no chloride dependence on the acceptor/donor ratio, making it an improved and more suitable biosensor for quantitative ratiometric [cAMP] measurements in physiology.
The disturbance of [cAMP] measurements caused by fluctuations of ion concentrations is an important factor, particularly for neurons, which normally are continuously active and experience permanent influence by neuromodulators often targeting not only the cAMP pathway but directly or indirectly also the calcium signaling cascades (e.g. see Ref. 35). The influence of changes of the pH and chloride concentration on FRET measurements is essential under pathophysiological condi-tions, such as hypoxia and ischemia. Under such pathophysiological conditions, all three variables, such as H ϩ , Cl Ϫ , and Ca 2ϩ , accumulate to disturb reliable cAMP measurements.
The pH i in healthy neurons can rapidly change upon activity. For example, in respiratory neurons, the pH i of 7.15 falls up to 0.2 pH unit during synaptic inhibition (14). Glutamate-induced excitotoxicity in hippocampal neurons was found to provoke an acidification of up to 0.5 unit of pH i due to displacement of H ϩ from internal binding sites by glutamate-induced increase in cytosolic Ca 2ϩ (36). In pathological situations, hypoxia and ischemia are accompanied by a dramatic drop of pH i (decreasing by 0.5-1 unit). This decrease probably results from an immediate blockade of oxidative phosphorylation and activation of glycolysis (37). To simulate such conditions, we evoked mild changes in pH i homoeostasis and analyzed how the two different biosensors reacted. Although CEPAC* was not influenced by pH i fluctuations, a moderate pH i drop from 7.25 to 7.05 resulted in a significant EPAC* biosensor signal equivalent to a change of [cAMP] induced by 1 M forskolin. Due to the described ion-dependent characteristic of the biosensors, CEPAC* is preferred to EPAC* for the usage of the ratiometric FRET methods.
The ideal behavior of a genetically encoded FRET biosensor is a fast protein expression and a uniform distribution, which was achieved with CEPAC* (Fig. 4). In contrast to EPAC*, CEPAC* showed virtually no sign of protein aggregation and, because of reduced cytotoxicity, allows measurements in small and narrow cellular compartments like dendrites. We suppose that the tendency for aggregation restricts the function of cAMP sensing because our cell experiments exhibited twice the number of responsive cells transfected with CEPAC* as compared with those with EPAC* (Fig. 5C). The aggregation might be caused by the dimerization motif present in eCFP and eYFP. This motif was removed in mCerulean and mCitrine, whereas the biosensor protein backbone was left unmodified in CEPAC*. The slow expression of EPAC* might also hint at problems associated with a failure of chaperone function resulting in incorrect protein folding and therefore aggregation. Such cytotoxicity became obvious in primary hippocampal neuron cultures transfected with either biosensor. The total dendrite length and the maximal length of dendrites in CEPAC*-transfected neurons were nearly twice as large as the values for EPAC*-transfected cells. Although the difference was not highly significant, one must remember the fact that we could only analyze living EPAC* cells. Although the overall appearance of the EPAC* neurons was poorer, many cells showed protein aggregations similar to those seen in N1E cells. Therefore, we assume that many neurons underwent apoptosis before being fluorescently labeled. Such aggregation was virtually absent in CEPAC*-transfected neurons.
In conclusion, we show that the use of ion-dependent FRET pairs like eCFP/eYFP results in significant artifacts in particular for ratiometric FRET measurements (using a single wavelength excitation), assuming physiological fluctuations of intracellular ion concentrations (38). FRET pairs, such as mCerulean/ mCitrine, with more stable fluorescent properties with respect to ion concentration and with an enhanced dynamic range of the FRET efficiency response, are superior. This is especially the case when measuring specific physiological processes or in certain samples in which background effects, such as fluctuations in ion concentration, are expected. Additional advantages include faster expression, longer stability, and a reduced tendency to aggregate, which allow measurements in microdomains. The improvements made to the FRET-based [cAMP] biosensors can be implemented for any other FRET-based biosensor.