mCerulean3-Based Cameleon Sensor to Explore Mitochondrial Ca²⁺ Dynamics In Vivo

Summary

Genetically Encoded Ca²⁺ Indicators (GECIs) are extensively used to study organelle Ca²⁺ homeostasis, although some available probes are still plagued by a number of problems, e.g., low fluorescence intensity, partial mistargeting, and pH sensitivity. Furthermore, in the most commonly used mitochondrial Förster Resonance Energy Transfer based-GECIs, the donor protein ECFP is characterized by a double exponential lifetime that complicates the fluorescence lifetime analysis. We have modified the cytosolic and mitochondria-targeted Cameleon GECIs by (1) substituting the donor ECFP with mCerulean3, a brighter and more stable fluorescent protein with a single exponential lifetime; (2) extensively modifying the constructs to improve targeting efficiency and fluorescence changes caused by Ca²⁺ binding; and (3) inserting the cDNAs into adeno-associated viral vectors for in vivo expression. The probes have been thoroughly characterized in situ by fluorescence microscopy and Fluorescence Lifetime Imaging Microscopy, and examples of their ex vivo and in vivo applications are described.

Graphical Abstract

Highlights

• •Donor substitution in a mitochondrial Ca²⁺ sensor improves photo-physical properties
• •Mitochondria-targeting sequence amelioration enhances the sensor localization
• •Donor substitution allows FLIM-FRET analysis, with a compensation for pH bias
• •The performance of the sensor is improved in situ, ex vivo, and in vivo

Biological Sciences Tools; Cell Biology; Optical Imaging

Introduction

The development of Genetically Encoded Indicators (GEIs) based on green fluorescent protein (GFP) mutants and their expression in cell subcompartments has revolutionized the study of intracellular signaling. Fluorescent proteins (FPs)-based GEIs can be divided into two major classes: single FP-based sensors and FRET-based indicators (see Pendin et al., 2017 for a recent review). Single FP-based GEIs are often characterized by high dynamic range (R_max/R_min), but they are generally “non-ratiometric,” i.e., they respond to changes in their specific ligand concentration with changes in fluorescent intensities but not of their excitation or emission spectra. This characteristic makes the calibration of the changes in Ca²⁺ concentration ([Ca²⁺]) in situ quite difficult for single fluorophore-based GECIs, and, more importantly, movement artifacts, occurring in some specimens, cannot be easily corrected. The FRET signal, on the contrary, is intrinsically ratiometric; changes in [Ca²⁺] are most commonly evaluated by recording the fluorescence intensity of both acceptor and donor FPs (exciting only the donor) and calculating their ratio (R). The vast majority of FRET-based sensors uses as the donor chromophore the cyan FP variant (CFP) and a yellow variant (YFP) as the acceptor. Among FRET-based sensors, the Ca²⁺-sensitive probes Cameleons are the most used. The basic Cameleon structure is represented by CFP and YFP linked together by the Ca²⁺-binding protein calmodulin and a calmodulin-binding domain from myosin light-chain kinase (M13). Many variants on this backbone have been produced over the last decade (Nagai et al., 2004, Palmer and Tsien, 2006, Horikawa et al., 2010). Surprisingly, the main stumbling blocks preventing the extensive use of Cameleons (and other FRET sensors based on CFP and YFP), i.e., the relatively low fluorescence quantum yield of the donor ECFP (Enhanced CFP), its complex kinetics, and strong tendency to photo-switch, have not been satisfactorily addressed. Recently, two CFP variants (mCerulean3 [Markwardt et al., 2011] and mTurquoise2 [Goedhart et al., 2010]) have been generated. These FPs are endowed with higher quantum yield, improved photostability, and decreased photo-switching behavior. Both these FPs are characterized by a single exponential decay rate, compared with the multiple components decay typical of CFP. The last feature makes these FPs particularly attractive for FLIM (Fluorescence Lifetime Imaging Microscopy) experiments.

As a model construct, we focused on the D3 Cameleon variant (D3cpv) because of its single and suitable affinity for Ca²⁺ (K_d) and its good R_max/R_min in response to Ca²⁺ changes. We thus substituted the ECFP with mCerulean3, and we extensively engineered the Cameleon cDNA developing a few approaches to modulate the features of CFP-YFP FRET-based sensors.

We thoroughly characterized the cytosolic and the mitochondria-targeted variant of mCerulean3-based Cameleon in situ using both intensity- and lifetime-based approaches. Finally, we inserted the mCerulean3-based probes into adeno-associated viral vectors (AAVs) to monitor cytoplasmic and mitochondrial Ca²⁺ dynamics ex vivo, in isolated neurons, in mature isolated cardiomyocytes, in acute slices from mouse brain, and in vivo in neurons of the somatosensory cortex.

Results

Mitochondria-Targeted and Cytosolic mCerulean3-Based Cameleon Probes Optimization

We first focused on the modification of the mitochondria-targeted Cameleon, 4mtD3cpv, substituting the ECFP with mCerulean3 (Palmer et al., 2006) and generating the probe 4mtD3mCerulean3 (hereafter called 4mtD3mC3, Figure 1A) that was eventually transiently expressed in HeLa cells (Figure 1B). Filippin et al. previously showed a significant mislocalization of 4mtD3cpv to the cytosol (Filippin et al., 2005). We confirmed this result: indeed, analyzing the number of cells that display a mitochondrial localization of the sensors 24 h after transfection, we found that 4mtD3cpv was properly localized in the mitochondrial matrix in only 38.6% of the cells (Figure 1C). We thus redesigned the whole N terminus of the 4mtD3mC3 probe (see Transparent Methods for further details). This targeting sequence enabled proper mitochondrial cleavage (Figure 1D) and localization (87.5% of cells after 24 h and ≅100% after 72 h, Figure 1C). To check whether mCerulean3 brightness improvement, reported for the cytosolic version of the protein (Markwardt et al., 2011), was maintained in the complex contest of a Cameleon probe expressed in a different environment, the mitochondrial matrix, the fluorescent intensity at 480 nm (the peak emission wavelength of the donor) in HeLa cells transfected with 4mtD3cpv or 4mtD3mC3 was measured. Since FRET efficiency is very similar in the two probes (see Figure 2B), the fluorescence intensity of cpV emission (excited at 512 nm, emission 535 nm), present in both probes, was used to normalize for the protein expression level (see Transparent Methods). The fluorescence mean intensity in 4mtD3mC3 expressing cells was about 42% higher than that of cells expressing 4mtD3cpv (Figure 1E).

Figure 1.

Optimization of Mitochondrial and Cytosolic Cameleons

(A–D) Cloning and localization of the mitochondrial probe. (A) Schematic representation of the cloning strategy used to modify the mitochondria targeting sequence (4mt) with the elongated version (4mt*) and to substitute ECFP with mCerulean3 in 4mtD3cpv Cameleon probes. (B) Confocal images of 4mtD3mC3. The mitochondrial localization of the probe was evaluated as the co-localization with a mitochondrial marker, TOM20. Yellow color indicates co-localization of the mitochondrial maker and 4mtD3mC3. Scale bar, 10 μm. (C) The bar chart represents the mean ± SEM of the number of cells (N) showing a proper mitochondrial localization, normalized to the number of transfected cells for each field. N ≥ 17 cells for each condition. (D) Mitochondria-targeting sequence cleavage. Twenty-four hours after transfection, total proteins were extracted and subjected to Western blot analysis with antibodies anti-GFP.

(E–H) The newly generated mitochondrial mCerulean3-based Cameleon is a functional and brighter probe compared with the original 4mtD3cpv. (E) The bar chart represents the mean ± SEM of ECFP and mCerulean3 fluorescence normalized to cpV fluorescence, in HeLa cells expressing mitochondrial Cameleon probes. N ≥ 39 cells for each condition. (F) The bar chart shows the R_max/R_min of the mitochondrial probes as mean ± SEM (normalized to 4mtD3cpv) of N ≥ 24 cells for each condition. (G) Representative kinetics of mitochondrial Ca²⁺ uptake in HeLa cells expressing 4mtD3cpv (black), 4mtD3mC3 (gray) and 4mtD3mC3+16 (light gray). HeLa cells transiently transfected with 4mtD3cpv, 4mtD3mC3, or 4mtD3mC3+16 were treated with 100 μM histamine (Hist) and perfused with 600 μM EGTA where indicated. (H) Left. Image of HeLa cells expressing 4mtD3mC3+16 along with the analyzed ROIs. Right. Representative kinetics of single mitochondrial Ca²⁺ uptake analysis in HeLa cells expressing 4mtD3mC3+16, using the protocol described for panel G. Data are plotted as ΔR/R₀ as defined in the Transparent Methods section. (I–M) Cloning and localization of the mitochondrial probe.

(I) Schematic representation of the cloning strategy used to substitute ECFP with mCerulean3 in D3cpv Cameleon probes.

(J) Fluorescence microscope image of the donor (cyan) and the acceptor (yellow) cytosolic D3mC3. Scale bar, 10 μm.

(K–M) The newly generated mCerulean3-based Cameleon is a functional and brighter probe compared with the original D3cpv. (K) Representative kinetics of cytosolic Ca²⁺ uptake in HeLa cells expressing D3cpv or D3mC3+16 employing the same protocol as in panel A. Data are presented ΔR/R₀. (L) The bar chart represents the mean ± SEM of ECFP and mCerulean3 fluorescence, normalized to cpV fluorescence, in HeLa cells expressing cytosolic Cameleon probes. N ≥ 27 cells for each condition. (M) The bar chart shows the R_max/R_min of the cytosolic probes as mean ± SEM (normalized to D3cpv) of N ≥ 18 cells for each condition. Statistical significance (*p < 0.05) was detected by Wilcoxon test for comparison between two groups and by one-way ANOVA and Bonferroni post hoc for comparison among three different groups.

See also Figure S1 and Table S7.

Figure 2.

Ca²⁺ Affinity and Mitochondrial Ca²⁺ Uptake upon Stimulation

(A–C) Titration protocol. (A) Representative kinetics of R% in permeabilized HeLa cells transiently expressing the 4mtD3cpv. Where indicated, digitonin-permeabilized cells were perfused with an intracellular-like medium without energy sources and containing different [Ca²⁺] together with 5 μM of the mitochondrial uncoupler, FCCP. In the representative trace 10 μM CaCl₂ was perfused; 5 mM CaCl₂ was finally added to reach the maximal FRET. (B and C) In situ Ca²⁺ titration mCerulean3-based Cameleon (red trace) or original Cameleon (black trace). The graph represents the data as mean ± SEM of N ≥ 5 cells for each [Ca²⁺], at pH 8.0 for mitochondrial (B) and at pH 7 for cytosolic probes (C).

(D–G) Stimulated mitochondrial Ca²⁺uptake. Average kinetics of mitochondrial Ca²⁺ uptake in HeLa (D) or MEFs (F) cells expressing 4mtD3mC3+16 (gray trace), 4mtD3cpv (black trace), or mt-Aequorin (mt-aeq, blue trace) employing the same protocol as in Figure 1G, with the only exception to MEFs, stimulated with ATP, 100 μM. Data are presented as [Ca²⁺], mean ± SEM. The bar chart shows the average of the mitochondrial Ca²⁺ peaks elicited by histamine application in HeLa cells (E) and ATP application in MEF cells (G) as mean ± SEM of N ≥ 18 and 7 cells, respectively, for each condition. Statistical significance (*p < 0.05) was detected by one-way ANOVA and Bonferroni post hoc.

In ratiometric GECIs, a key parameter for their practical use is the value of R_max/R_min in situ. To this end, HeLa cells have been permeabilized to obtain accurate minimal and maximal Ca²⁺ levels thanks to the equilibration between the perfused saline and the mitochondrial matrix (see Transparent Methods). The substitution of the ECFP with mCerulean3 caused a reduction (about 20%) in R_max/R_min of 4mtD3mC3 (R_max/R_min = 1.94, Table 1) compared with 4mtD3cpv (R_max/R_min = 2.44, Table 1). We first attempted to substitute the acceptor FP (Nagai et al., 2004), replacing cpV with other YFP variants, Citrine or YPet, but none of them allowed a recovery of R_max/R_min (data not shown). Finally, we increased the linker flexibility between the two Ca²⁺ responsive elements, by including poly-Gly linkers of different lengths, from 2 to 16 (Chichili et al., 2013). The addition of a 16-amino acid poly-Gly allowed the complete recovery of the R_max/R_min, increasing it to 2.37 (Figure 1F and Table 1). The ability of the newly generated variant (called 4mtD3mC3+16) to report changes in mitochondrial [Ca²⁺], compared with the other two probes, was finally tested and confirmed in intact cells (Figure 1G). Furthermore, the increased signal to noise ratio has been exploited to monitor the dynamics of [Ca²⁺] in single mitochondria upon stimulation with histamine (Figure 1H and Video S1).

Table 1.

In Situ Properties of Cameleon Probes

		pH 7.0	pH 8.0
4mtD3mC3+16	Kd (μM)	6.48 ± 0.81	6.18 ± 0.99
	N	1.07 ± 0.13	0.72 ± 0.07
	Rmax/Rmin	1.92 ± 0.03	2.37 ± 0.03
4mtD3cpv	Kd (μM)	2.49 ± 0.23	3.22 ± 0.86
	N	0.99 ± 0.09	0.68 ± 0.11
	Rmax/Rmin	2.48 ± 0.13	2.44 ± 0.03
D3mC3+16	Kd (μM)	4.15 ± 1.12
	N	0.77 ± 0.12
	Rmax/Rmin	3.48 ± 0.07
D3cpv	Kd (μM)	1.83 ± 0.55
	N	0.68 ± 0.14
	Rmax/Rmin	5.36 ± 0.13

K_d, dissociation constant; n, Hill constant; R_max/R_min, dynamic range. The K_d and R_max/R_min values have been compared for pH values (7.0 and 8.0) that can be observed in the mitochondrial matrix in different experimental conditions (pH 7.0 in Figure S1I, related to Figure 1). The estimated Ca²⁺ affinities vary by about 5% at pH 7.0 compared with those at pH 8.0 for the 4mtD3mC3+16 probe and by about 20% for the 4mtD3cpv, indicating that the substitution of the donor decreases significantly the pH sensitivity of the mitochondrial probe. The acidification from pH 8.0 to 7.0 results also in a significant change of R_max/R_min in both 4mtD3cpv and 4mtD3mC3+16. Data are presented as mean ± SEM of N ≥ 5 (number of cells).

Video S1. Histamine Stimulated Mitochondrial Ca²⁺ Uptake in HeLa Cells, Related to Figure 1

The video shows the increase in the mitochondrial [Ca²⁺] induced by histamine perfusion of HeLa cells expressing 4mtD3mC3+16. The mitochondrial Ca²⁺ uptake is visualized employing a pseudo-colored scale that starts from blue-green, low [Ca²⁺], and turns to yellow-red, high [Ca²⁺].

These promising results were exploited to generate a mCerulean3-based cytosolic Cameleon. ECFP was thus substituted with mCerulean3 in D3cpv sensor, and the glycine-linkers were added between the two Ca²⁺ responsive elements, obtaining the probe D3mC3+16 (Figure 1I). The cytosolic localization of the probe was confirmed using fluorescence microscopy (Figure 1J). The ability of the probes to detect cytosolic Ca²⁺ ([Ca²⁺]_c) changes upon histamine stimulation (Figure 1K) and the improvement in mCerulean3 fluorescence compared with ECFP (Figure 1L) were confirmed.

Since the classical digitonin treatment (as used earlier) causes a rapid probe leakage into the medium, to measure the R_max/R_min of the cytosolic probes, we employed a very low dose of digitonin (Hofer et al., 1998, see Transparent Methods), a protocol that prevents probe leakage. Despite the addition of 16 Gly, the D3mC3+16 shows a significant reduction of the R_max/R_min ratio (Figure 1M) compared with the original probe.

Photostability and pH Effect

Photostability is an important factor in experiments requiring prolonged imaging of living cells. Photobleaching of mCerulean3 has been shown to be significantly less than that of ECFP (Markwardt et al., 2011), but the effect of this property in an FRET sensor such as Cameleon is difficult to predict (see Transparent Methods for further details). HeLa cells expressing the ECFP-based or mCerulean3-based Cameleons were thus imaged for 20 min with the same settings for both illumination and acquisition. R was monitored in conditions of basal [Ca²⁺] or in the absence of Ca²⁺ in the medium and cytosolic [Ca²⁺] clamped at very low levels (see Transparent Methods). Figure S1A shows time-dependent R changes at basal [Ca²⁺] in representative cells (data quantification is provided in Figure S1B). The decrease of R is faster (about twofold) in D3cpv-expressing cells compared with cells expressing either D3mC3 or D3mC3+16. A similar difference was observed in cells in the absence of external Ca²⁺ (i.e., employing a Ca²⁺ clamp method, see Transparent Methods; Figures S1C and S1D) and for the mitochondria-targeted Cameleon probes (Figures S1E and S1F for basal FRET photostability and Figures S1G and S1H for minimal FRET photostability). Given the much better performance of mCerulean3-based Cameleons in the presence of 16 glycines, most of the following data were performed using only these constructs.

Last, since variations in pH can affect several GECI properties, depending on single FPs pK_a (Koldenkova and Nagai, 2013), we also investigated this issue in HeLa cells transiently expressing the cytosolic or the mitochondrial Cameleons, by imposing the different pH that can be experienced by these two compartments in live cells. We found that, as predicted by the mCerulean3 pK_a, the sensors based on this protein are much less sensitive to pH changes (see Table 1 and Figures S1I–S1P for further details).

Ca²⁺ Affinity

The profound structural changes introduced in the probes may also affect their affinity for Ca²⁺ (Horikawa et al., 2010). The in situ calibration of the Ca²⁺ affinity was thus carried out for the ECFP- and mCerulean3-based mitochondria-targeted probes, using the passive Ca²⁺ loading procedure in digitonin-permeabilized cells (Figures 2A, Filippin et al., 2005). The experimentally calculated K_d in situ (at the physiological matrix pH 8.0) is lower for 4mtD3cpv (K_d = 3.2 ± 0.9 μM) (Table 1, Figure 2B, black trace) compared with that of 4mtD3mC3+16 (K_d = 6.2 ± 0.99 μM) (Table 1, Figure 2B, red trace). The in situ calibration has been performed also for the cytosolic probes. The estimated K_d in situ was 1.8 ± 0.5 μM for D3cpv (Table 1, Figure 2C, black trace) and 4.1 ± 1.1 μM for D3mC3+16 (Table 1, Figure 2C, red trace).

We then exploited the calculated K_d to measure the peak of [Ca²⁺]_m, reached in the mitochondrial matrix, upon ER Ca²⁺ release induced by IP₃R (Inositol trisphosphate receptor) stimulation. [Ca²⁺]_m peaks, elicited by histamine, were measured in a selected low responding HeLa cell batch (Figures 2D and 2E) and by ATP in a strongly responding MEF (Mouse Embryonic Fibroblast) cell line (Figures 2F and 2G) and compared with the values measured with targeted aequorin. In HeLa cells, the mean [Ca²⁺] peak in mitochondria was 18 μM, as measured with aequorin, while those measured with mitochondrial targeted Cameleons were on average 11 μM and 14 μM for 4mtD3mC3+16 and 4mtD3cpv, respectively (Figure 2E). In some cells (9% and 29% of cells expressing 4mtD3mC3+16 and 4mtD3cpv, respectively), the R values of the peaks were so close to R_max that the calculation of the [Ca²⁺] in the matrix was prone to major errors. For this reason, those cells have been excluded from the analysis.

In MEF cells, stimulated with ATP, the mean Ca²⁺ peaks measured with aequorin were about 131 μM, i.e., much higher than those reported by the two Cameleons (44 and 45 μM for 4mtD3mC3+16 and 4mtD3cpv, respectively). Also in this latter case, as explained before, 20% and 40% of cells expressing 4mtD3mC3+16 and 4mtD3cpv, respectively, were very close to R_max, and thus they were not included in the means.

FRET Efficiency Evaluation and pH Sensitivity with FLIM

To evaluate FRET efficiency, two methods are usually employed: the intensity-based (mostly using the acceptor photobleaching approach) or fluorescence lifetime-based measurements. FLIM measurements, compared with intensity-based images, have the advantage of being mostly independent of fluorophore concentration, excitation intensity fluctuation, sample thickness, or photobleaching (Marcu et al., 2014). Thus, analysis based on a time-resolved technique has often been employed to precisely estimate FRET efficiency. Usually, in FLIM-FRET measurements, energy transfer is observed as a decrease of the donor fluorescence lifetime. To calculate FRET efficiency under different conditions (see Transparent Methods), it is first necessary to determine the lifetime of the donor only. To this end, HeLa cells were transfected with ECFP or mCerulean3 targeted to the mitochondrial matrix. As previously reported, the fluorescence decay of ECFP can be fitted by a two-exponential terms model (Millington et al., 2007), whereas the mCerulean3 decay is mono-exponential (Markwardt et al., 2011) (Table S1 and Figures S2A and S2B). Next, applying the same Ca²⁺ titration protocol described for the intensity-based approach, we evaluated FRET efficiency in HeLa cells expressing 4mtD3cpv or 4mtD3mC3+16. The lifetime of 4mtD3cpv can be fitted by either a double or a triple exponential model, whereas in the other probe, the lifetime can be fitted with a double exponential model, where only the second shorter donor lifetime is associated to the FRET process. Given the low accuracy of a triple exponential fitting, a double exponential fitting model has been used also in the case of 4mtD3cpv (Geiger et al., 2012). For both sensors, the shorter donor lifetime component is clearly dependent on [Ca²⁺], whereas the longer donor lifetime is less affected by the Ca²⁺ levels (Table S2 for 4mtD3mC3+16 and Table S3 for 4mtD3cpv). At maximum [Ca²⁺], FRET efficiency was 43% in the case of 4mtD3cpv (Table 2 and Figure 3A) and 36% for 4mtD3mC3+16 (Table 2 and Figure 3B), whereas in the absence of Ca²⁺ the FRET efficiencies decrease to 31% (Table 2 and Figure 3A) and 18% (Table 2 and Figure 3B) respectively. Thus, the dynamic range, estimated as the ratio between maximal and minimal FRET efficiencies at pH 8.0, is 2 for 4mtD3mC3+16 and 1.4 for 4mtD3cpv (Figures 3A and 3B and Table 2).

Table 2.

Decay Fitting Parameters of Cells Expressing 4mtD3cpv and 4mtD3mC3+16 at pH Equal to 8.0 or 7.5

Sample	τ1 (ns)	A1	τ2 (ns)	A2	<τ>	EFRET
4MTD3CPV PH = 8, CAMIN	2.89 ± 0.02	49 ± 1%	0.88 ± 0.02	51 ± 1%	1.86 ± 0.01	31%
4MTD3CPV PH = 8, CASAT	2.66 ± 0.02	41 ± 1%	0.73 ± 0.01	59 ± 1%	1.53 ± 0.02	43%
4MTD3CPV PH = 7.5, CAMIN	2.84 ± 0.01	44 ± 1%	0.75 ± 0.02	56 ± 1%	1.68 ± 0.02	32%
4MTD3CPV PH = 7.5, CASAT	2.75 ± 0.02	43 ± 1%	0.71 ± 0.02	57 ± 1%	1.58 ± 0.02	36%
4MTD3MC3+16 PH = 8, CAMIN	3.58 ± 0.03	72 ± 1%	1.24 ± 0.05	28 ± 1%	2.94 ± 0.01	18%
4MTD3MC3+16 PH = 8, CASAT	3.21 ± 0.05	65 ± 1%	0.63 ± 0.05	35 ± 1%	2.31 ± 0.06	36%
4MTD3MC3+16 PH = 7.5, CAMIN	3.55 ± 0.03	61 ± 1%	0.94 ± 0.03	39 ± 1%	2.53 ± 0.03	25%
4MTD3MC3+16 PH = 7.5, CASAT	3.39 ± 0.03	60 ± 2%	0.75 ± 0.03	40 ± 2%	2.34 ± 0.08	30%

Ca²⁺_min and Ca²⁺_sat are, respectively, the minimum and the saturation Ca²⁺ concentrations (500 μM). τ_i are the lifetimes, A_i are the relative amplitudes, and average lifetimes (<τ>) are the amplitude-weighted lifetimes. Data are reported as mean ± SEM of N ≥ 5 cells for each [Ca²⁺] and pH condition. See also Tables S1–S5.

Figure 3.

FLIM Analysis of FRET Efficiency and Alterations in MCUC Components Protein Level Effect on Mitochondria [Ca²⁺] in Resting Conditions

(A) Normalized fluorescence decay of 4mtD3cpv at minimum FRET (black squares) and maximum FRET (blue dots). The full lines are the fitting curves with a two-exponential model. For comparison, the decays of 4mt-CFP expressed in mitochondria are reported (red lines). The insets are the false-color FLIM images (amplitude-weighted lifetime) at minimum and maximum FRET, in the 1- to 4-ns range. Data are from N ≥ 15 cells for each condition.

(B) Normalized fluorescence decay of 4mtD3mC3+16 at minimum FRET (black squares) and maximum FRET (blue dots). The full lines are the fitting curves with a two-exponential model. For comparison, the decay of mCerulean3 expressed in mitochondria is reported (red line, one-exponential fitting model). The insets are the false-color FLIM images (amplitude-weighted lifetime) at minimum and maximum FRET, in the 2- to 5-ns range. Data are from N ≥ 15 cells for each condition.

(C) FLIM images of 4mtD3mC3+16 in intact HeLa cells in resting conditions. FLIM images report the amplitude-weighted lifetime in the 2- to 3-ns range. Scale bar, 10 μm.

(D) The bar chart represents the <τ> mean ± SEM of N ≥ 27 cells expressing 4mtD3mC3+16 and overexpressing MCU, MICU1, or MCU and MICU1.

(E) The bar chart represents the mean ± SEM of the R (cpV/mCerulean3), normalized to control, of N ≥ 25 cells expressing 4mtD3mC3+16 and overexpressing MCU, MICU1, or both.

(F) The bar chart represents the mean ± SEM of R (% of max, where maximum is the R at pH 9.0) of N ≥ 13 cells transfected with mt-SypHer in the presence of void vector, MCU, MICU1, or MCU and MICU1.

(G) The bar chart represents the <τ> mean ± SEM of N ≥ 15 cells expressing 4mtD3mC3+16 and overexpressing MCU, MICU1, or MCU and MICU1, including only data that display the negative rise term in the acceptor channel.

See also Figure S2 and Tables S1–S6.

Since we found that R is affected by pH (see Figures S1I–S1P and Table 1) and previous studies have demonstrated that pH has a significant impact on ECFP lifetime (Villoing et al., 2008), we carried out FLIM-based analysis of 4mtD3cpv and 4mtD3mC3+16 at different [Ca²⁺] and different pH. We found a decrease in the dynamic range of both probes upon acidification (see Figure S2). Furthermore, we found that lifetime analysis allows the discrimination between pH and Ca²⁺ effect, employing the rise term analysis (Borst et al., 2008). The rise term reflects the delay in acceptor emission decay due to indirect excitation from donor emission, and it is represented by a negative amplitude contribution to the acceptor emission decay. Upon mitochondrial matrix acidification, the disappearance of this rise term in the acceptor channel occurs independently of changes in [Ca²⁺] (see Figures S2E and S2F and Tables S4 and S5 for further details).

FLIM Analysis Reveals Mitochondrial [Ca²⁺] Heterogeneity in Resting Condition

The FLIM approach has been finally applied (using only the 4mtD3mC3+16) to estimate the resting Ca²⁺ level in the mitochondrial matrix in different conditions: naive cells, cells overexpressing the mitochondrial Ca²⁺ uniporter (MCU), the MCU modulator MICU1, or both proteins. The first observation is that, under resting conditions, the FLIM maps reveal a substantial heterogeneity of mitochondria in terms of [Ca²⁺] (Figure 3C), possibly because of constitutive transfer of Ca²⁺ from the ER to some mitochondria (those located near the mouth of IP₃R) (Cárdenas et al., 2010). As expected (Csordas et al., 2013, Mallilankaraman et al., 2012, Patron et al., 2014), overexpression of MCU alone or MCU and MICU1 decreased the donor lifetime (Figure 3D), indicating the predicted increase in [Ca²⁺]_m. Surprisingly, also the overexpression of only MICU1 caused a small reduction of donor lifetime. A small increase in the [Ca²⁺]_m upon MICU1 overexpression has been previously reported employing GCaMP family Ca²⁺ sensors (Csordas et al., 2013, Mallilankaraman et al., 2012, Patron et al., 2014). We did not observe such an increase employing 4mtD3mC3+16 using the intensity-based approach (Figure 3E), possibly due to the lower K_d for Ca²⁺ of the Cameleon compared with GCaMP sensors. Given the similar effect of acidification and Ca²⁺ increase on donor lifetime, we investigated whether a reduction in matrix pH could contribute to the shorter lifetime observed in MICU1-overexpressing cells. Indeed, the rise term analysis suggested a decrease in matrix pH in cells overexpressing MICU1 (Table S6). We thus directly verified whether overexpression of MICU1 induces changes in the matrix pH by using the GEI mt-SypHer (Poburko and Demaurex, 2012). A significant decrease in matrix pH from 7.7 in control cells to 7.5 in cells overexpressing MICU1 and/or MCU was observed (Figure 3F). We thus repeated FLIM analysis including only cells that presented an acceptor rise (Figures 3G and S2C, Table S6). In these cells, MICU1 overexpression causes a marginal, non-significant increase in matrix [Ca²⁺], whereas MCU (overexpressed alone or together with MICU1) induces a much larger [Ca²⁺] rise.

In Vivo Delivery of the Generated Probes in Mice

To express mCerulean3-based Cameleons in vivo, adeno-associated viruses (AAVs) were generated with capsid serotype 9 (Bish et al., 2008, Pacak et al., 2006, Foust et al., 2009). AAV9-CMV-D3mC3+16 or AAV9-CMV-4mtD3mC3+16 was first tested in cultured neonatal rat cardiomyocytes. Ninety-six hours after transfection, spontaneous cytosolic and mitochondrial Ca²⁺ oscillations were observed with both probes (Figures 4A and 4B, respectively). Of note, the cytosolic peaks displayed a larger change in ΔR/R₀ compared with the mitochondrial ones. Figure S3 shows one of the advantages of a ratiometric probe such as 4mtD3mC3+16 compared with one of the brightest “non-ratiometric” sensors available, mtGCaMP6f (Mammucari et al., 2015, Tosatto et al., 2016). The response of the ratiometric probe nicely corrects for the substantial movement artifact due to contraction in all transfected cells. Although in many cells the mtGCaMP6f appears to nicely report the mitochondrial Ca²⁺ increases during spontaneous beating (Figure S3A), in others (Figure S3B) the changes in fluorescence are minimal and hardly distinguishable from the changes in focus due to contraction observed in cells solely transfected with a mitochondrial GFP (Figure S3C).

Figure 4.

Monitoring [Ca²⁺] Changes in Cardiomyocytes

Representative traces of spontaneous Ca²⁺ oscillations in neonatal rat cardiomyocytes expressing AAV9-CMV-D3mC3+16 (A) or AAV9-CMV-4mtD3mC3+16 (B). Representative traces of spontaneous Ca²⁺ oscillations in adult cardiomyocytes isolated from mice injected with AAV9-CMV-D3mC3+16 (C) or AAV9-CMV-4mtD3mC3+16 (D). Representative traces of [Ca²⁺] changes in response to caffeine in adult cardiomyocytes from mice injected with AAV9-CMV-D3mC3+16 (E) or AAV9-CMV-4mtD3mC3+16 (F). Representative traces of spontaneous Ca²⁺ oscillations in human induced pluripotent stem cell-derived cardiomyocytes expressing AAV9-CMV-D3mC3+16 (G) or AAV9-CMV-4mtD3mC3+16 (H). Data are plotted as ΔR/R₀. See also Figure S3.

AAV9-CMV-D3mC3+16 or AAV9-CMV-4mtD3mC3+16 was then used to infect adult cardiac cells in mice via tail vein injection. Six months after the infection, cardiomyocytes were isolated. Figure 4 shows the typical kinetics of cytosolic and mitochondrial Ca²⁺ changes due to spontaneous Ca²⁺ oscillations (Figures 4C and 4D, respectively) or induced by the application of caffeine (Figures 4E and 4F, respectively). Noteworthy, the amplitude of mitochondrial Ca²⁺ rise (in terms of ΔR/R₀) was larger in adult cardiomyocytes compared with that in neonatal cells.

The different heart-beating rate among different animals, which inversely correlate with body weight (Ostergaard et al., 2010), raises the question of the suitability of these probes in models that can contribute to the study of human cardiac pathophysiology. To address this issue, we exploit the transduction of AAV9-CMV-D3mC3+16 or AAV9-CMV-4mtD3mC3+16 in human induced pluripotent stem cell-derived cardiomyocytes. In Figures 4G and 4H, we show that both cytosolic and mitochondrial mCerulean3-based Cameleons were able to report spontaneous and synchronous (among cells of the same coverslips) Ca²⁺ oscillations at a beating frequency similar to that of human beings.

To specifically express the probes in neurons of mouse brain, the CMV promoter was substituted with the neuronal-specific promoter synapsin 1 (Kugler et al., 2001). The AAVs were first tested in cultures of primary neonatal cortical neuron (typically contaminated by non-neuronal cells) where the exclusive neuronal expression and the proper localization of both probes have been tested (see Figure S4). Furthermore, we exploit cultured neurons to test the ability of both probes to detect spontaneous activity (see Figures S5A and S5B), Ca²⁺ changes in response to a depolarizing stimulation (see Figures S5C and S5D), or the fast Ca²⁺ spikes elicited by the GABA-A antagonist picrotoxin (see Figures S5E–S5H).

AAV9-syn-D3mC3+16 or AAV9-syn-4mtD3mC3+16 was then injected intracranially in 1-month old mice, and the cytosolic and mitochondrial Ca²⁺ levels were tested in acute hippocampal slices 2 weeks post infection. The neuronal specificity and the correct subcellular localization were verified ex vivo (see Figure S6). In the hippocampal slices, a larger increase in mitochondrial Ca²⁺ (recorded at the two-photon microscope; see Figures S7A and S7B for spectra characterization), compared with that of the cytosol, was observed upon stimulation with group I metabotropic glutamate receptor agonist ((S)-3,5-Dihydroxyphenylglycine, DHPG, 50 μM) (Figures 5A and 5B). DHPG response can be detected as a single Ca²⁺ rise (as shown in Figure 5A, cytosol, and 5B, mitochondria),or it can generate an oscillatory pattern (Figures 5C, cytosol, and 5D, mitochondria). However, in the mitochondria the oscillations of [Ca²⁺] were slower and more prolonged than the fast oscillations of the cytosolic compartment. Additional experiments were carried out in hippocampal slices from adult mice, using a muscarinic agonist (Carbachol, CCH, Figure 5E) or the ionotropic glutamatergic agonist (N-methyl-D-aspartate, NMDA, Figure 5F), and in somatosensory slices from 2-week old animals upon local application of the glutamatergic agonist NMDA (Figures 5G and S7C–S7F) and depolarization with KCl (Figure 5H). Spontaneous Ca²⁺ oscillations were rarely and randomly observed in the cytosol of neurons of hippocampal slices expressing AAV9-syn-D3mC3+16 (Figure S7G), whereas no spontaneous Ca²⁺ spikes were detected in the mitochondrial compartment.

Figure 5.

Monitoring [Ca²⁺] Changes in the Brain: Hippocampus Slices from Mouse, Somatosensory Cortex Slices from Mouse, and In Vivo Mouse Cortex

(A and B) Average traces of Ca²⁺ rise induced by DHPG (50 μM) in neurons of hippocampal slices from mice injected with AAV9-syn-D3mC3+16 (A) or AAV9-syn-4mtD3mC3+16 (B) of N ≥ 4 cells.

(C and D) Representative traces of Ca²⁺ rise induced by DHPG in neurons of hippocampal slices from mice injected with AAV9-syn-D3mC3+16 (C) or AAV9-syn-4mtD3mC3+16 (D).

(E) Average traces of Ca²⁺ rise induced by carbachol (CCH, 500 μM) in neurons of hippocampal slices from mice injected with AAV9-syn-D3mC3+16 (black) or AAV9-syn-4mtD3mC3+16 (gray). N ≥ 5.

(F) Average traces of Ca²⁺ rise induced by perfusion of NMDA 50 μM in neurons of hippocampal slices from mice injected with AAV9-syn-D3mC3+16 (black) or AAV9-syn-4mtD3mC3+16 (gray). N ≥ 10.

(G) Average traces of Ca²⁺ rise induced by neuronal depolarization induced by NMDA receptor activation with 1 mM NMDA (puff administration) in neurons of cortical slices from mice slices from mice injected with AAV9-syn-D3mC3+16 (black) or AAV9-syn-4mtD3mC3+16 (gray). N ≥ 22.

(H) Average traces of Ca²⁺ rise induced by neuronal depolarization with 30 mM KCl in neurons of cortical slices from mice slices from mice infected with AAV9-syn-D3mC3+16 (black) or AAV9-syn-4mtD3mC3+16 (gray). N ≥ 5.

(I and J) In vivo experiment showing representative traces of Ca²⁺ rise induced by NMDA receptor activation with 1 mM NMDA (puff administration) in neurons from mice injected with AAV9-syn-D3mC3+16 (I) or AAV9-syn-4mtD3mC3+16 (J). [Ca²⁺] changes have been measured in neuronal cell bodies. Data are plotted as ΔR/R₀.

See also Figures S4–S7.

We finally tested the ability of both probes to detect Ca²⁺ changes in vivo. We performed two-photon Ca²⁺ imaging experiments in the cortex of anesthetized mice and locally applied NMDA (1 mM). As expected, we recorded large rise in [Ca²⁺] in both cytosolic and mitochondrial compartments (Figures 5I and 5J). It is worth noting that, despite the substantial movements due to blood flow and respiration, we were able to record a stable basal mitochondrial Ca²⁺ level, confirming the efficacy of the ratiometric measurement to correct such characteristic confounding artifacts of in vivo imaging (Figures 5I and 5J).

Discussion

The central role of Ca²⁺ signaling in the regulation of physiological and pathological processes, as well as the importance of mitochondria in this process, is widely accepted (Rizzuto et al., 2012). Thanks to the development of GECIs, the role of subcellular compartments in Ca²⁺ regulation has been intensively explored; however, their role in vivo remains still largely based on indirect information (Pendin et al., 2015). The main reason for this difficulty in monitoring organelle Ca²⁺ handling directly and in vivo largely depends on the limitations of the available targeted Ca²⁺ indicators. Movement of the samples (due to breathing, heart- beat, and blood flow) or of the organelle themselves, relatively low fluorescence of ratiometric indicators (ideal for correcting movement artifacts), mislocalization of the expressed probes, pH sensitivity, and photobleaching are particularly relevant problems for investigating mitochondrial Ca²⁺ handling in vivo. To overcome these drawbacks, we decided to improve the properties of a widely used FRET-based Ca²⁺ indicator, i.e., the Cameleons, by extensively modifying the primary structure. In particular, we have (1) modified the N-terminus of the 4mtD3cpv, by eliminating a potential additional methionine initiation site (that could be in part responsible for the mistargeting of the original probe to the cytosol) and increasing the number of amphiphilic amino acids in each of the four targeting sequences (i.e., four repeats of the human cytochrome c oxidase subunit VIII mitochondria-targeting peptide); (2) substituted ECFP with the more brilliant and less pH sensitive isoform mCerulean3; (3) introduced a flexible linker between the two Ca²⁺ sensor domains (calmodulin and M13). The cytosolic and mitochondrial targeted probes, D3mC3+16 and 4mtD3mC3+16, have been extensively characterized in situ, and they show a clear improvement in brightness, photostability, selective localization (for the mitochondrial version), and reduced pH sensitivity. Although most of the improvements were somehow predictable, it is still unclear why the simple substitution of ECFP with another very similar FP isoform results in a drastic modification of R_max/R_min in response to Ca²⁺. However, similar examples exist, such as that observed by Thestrup et al. (2014). The strategy of introducing an artificial poly-Gly linker was successful in the mitochondrial target probe and can, in principle, be generalized also to other probes, at least those that rely on two distinct sensor elements for their functionality. However, the same approach failed in the complete recovery of this drawback in the cytosolic probe. Despite that we do not have a conclusive explanation, we can hypothesize that the different environments between the cytosol and the mitochondria, in terms of pH, temperature (Chrétien et al., 2018), and protein composition, can affect many photophysical properties of both FPs and Ca²⁺-binding domains.

A debated issue in the mitochondrial field is the amplitude of Ca²⁺ transients that can be experienced by mitochondrial matrix upon stimulation of IP₃R, i.e., up to 3 μM (Kristián et al., 2002, Chalmers and Nicholls, 2003) or up to tenths or hundreds of micromoles (Montero et al., 2000, Montero et al., 2002, Arnaudeau et al., 2001, Pinton et al., 2004). The values reported here, based on an accurate in situ calibration of the Cameleons, are much closer to those obtained with mitochondrial low-affinity aequorin than those previously reported with many other GECIs and chemical dyes. The peak [Ca²⁺]_m here monitored with the Cameleons (that have an affinity for Ca²⁺ much lower than chemical dyes and of most other mitochondrial GECIs) are similar to those measured previously using ratiometric pericams (Filippin et al., 2003) or using a low-affinity Cameleon (Palmer et al., 2006). The average Ca²⁺ peaks reported here, however, are still lower than those measured with mutated aequorin. However, it should be stressed that (1) the average values of the [Ca²⁺] here reported using 4mtD3mC3 and 4mtD3mC3+16 are substantially underestimated given our choice to selectively eliminate the more strongly responding cells/organelles (see Figures 2F and 2G) and (2) the very high Ca²⁺ values reported by aequorin (averaged over all cells) depend also on the intrinsic characteristics of the aequorin signal calibration that is dominated by the most responsive cells/organelles (for review see Rizzuto and Pozzan, 2006).

Given the recent interest in the FLIM technique to study cellular signaling, we then exploit TCSPC (time-correlated single-photon counting) FLIM-FRET analysis. The use of FLIM microscopy, employing both time and frequency domain approaches, is becoming quite popular not only in classical biophysical studies in vitro but also for studying intracellular dynamics in cells, either in situ (Klarenbeek et al., 2011, Laine et al., 2012, Zhao et al., 2015) or in vivo (Omer et al., 2014, Raspe et al., 2016, Zhao et al., 2015). Here we show that the single exponential lifetime of mCerulean3 allows a precise evaluation of FRET efficiency and increases the dynamic range compared with ECFP-based Cameleon. One unexpected result was the pronounced pH sensitivity of lifetime, in particular in the case of the mitochondrial probe. At low [Ca²⁺] and at pH 7.0, there is an important and heterogeneous reduction of the acceptor rise term in many cells. This effect is not observed at pH 7.0 in the cytosolic probe or for the mitochondrial sensor in the physiological pH range 7.5–8.0. Although we do not have a clear explanation for this phenomenon, we can hypothesize that it could depend on some specific event occurring in the mitochondrial matrix at this non-physiological pH value. Indeed, changes in the mitochondrial environment have been reported to impact on FPs lifetime (Rieger et al., 2017, Sohnel et al., 2016). However, combining acceptor rise term and lifetime analysis, we revealed an unpredicted consequence of MCU and/or MICU1 overexpression, i.e., a significant acidification of matrix pH at least in a subpopulation of cells (then confirmed by direct measurement with a mitochondria-targeted pH sensitive probe).

After this accurate evaluation of the improved performance of the probes in living cells, we generated AAV9 to express them in primary cultures (neurons and cardiomyocytes), brain slices, and in vivo in mouse brain cortex. The data obtained in these models confirm the excellent signal to noise ratio of the mCerulean3-based probes and provide the proof of principle for an in situ and in vivo evaluation of cytosolic and mitochondrial Ca²⁺ dynamics upon different stimulation protocols, allowing the dissection of different patterns of neurotransmission. In particular, spontaneous synchronized oscillatory activity has been detected in different cardiomyocyte models with both cytosolic and mitochondrial probes. Spontaneous activity has been recorded in both neuronal primary cultures and brain slices; cytosolic and mitochondrial Ca²⁺ changes have been easily monitored in cultured neurons, acute brain slices, and in vivo using different pharmacological stimuli.

Generally, the main obstacle preventing the extensive use of Cameleons (and other FRET sensors based on CFP/YFP couple) is the relatively low fluorescence quantum yield of the donor. The substantial increase in the signal to noise ratio we obtained in mCerulean-based sensors could allow experiments in cells with low expression levels that otherwise will be troublesome. In addition, they will also allow quantitative measurements by FLIM, which would be impossible with CFP-based probes because of the double exponential lifetime of this variant, thus expanding the available tools for studying Ca²⁺ signaling in vivo.

In conclusion, the mCerulean3-based Cameleons we generated are updated versions of the original probes with a number of improvements in their photo-physical properties that allow their usage in both intensity- and lifetime-based approaches and in different biological contexts. Although the dynamic range of Cameleons in response to [Ca²⁺] changes is lower compared with the best intensity-based probes, the FRET-based indicators appear superior to correct movement artifacts, photobleaching, and pH sensitivity, and most relevant, they allow one to quantitatively estimate the values of [Ca²⁺] at rest and during stimulation.

Limitations of the Study

Beyond the reported advantages, mCeruelean3-based Cameleon probes retain some of the general limitations of FRET-based biosensors, i.e., lower dynamic range and lower fluorescence intensity compared with chemical indicators or to some of the single FP-based sensors. Furthermore, although pH sensitivity has been ameliorated, pH can still affect some properties of mCerulean3-based Cameleons. Lifetime measurements, and in particular the rise term analysis of 4mtD3mCerulean3+16, can help in distinguishing Ca²⁺ and pH effects on the FRET signal. Finally, since Cameleons are high-molecular-weight sensors, their fusion with other cDNAs in viral vector with small expression cassettes is hampered.

Methods

All methods can be found in the accompanying Transparent Methods supplemental file.

Published: June 28, 2019