Genetically encoded calcium indicators for multi-color neural activity imaging and combination with optogenetics

Structure-guided engineering of color-shifted GECIs

We initially mutated GCaMP3 (Tian et al., 2009) to incorporate sets of the core mutations of EBFP, ECFP, and EYFP: in GCaMP3 numbering Phe221Leu/Thr222Ser/Tyr223His (“BCaMP1a”); Tyr223Trp (“CyCaMP1a”); and Val116Tyr/Lys119Val (GCaMP3 incorporates the Thr116Val mutation; “YCaMP1a”). CyCaMP1a and YCaMP1a were fluorescent and responded to in vitro calcium changes [(ΔF/F)_max = 2.6 ± 0.1, s.d., n = 3; and (ΔF/F)_max = 3.0 ± 0.1, s.d., n = 3, respectively] (Table 1), whereas BCaMP1a was fluorescent but not a sensor. The Cerulean (Rizzo et al., 2004) mutation Ser229Ala (GFP numbering Ser72Ala; Cerulean also has mutations at GFP positions 145 and 148, but these positions are absent in GCaMP) resulted in higher apo brightness and lower (ΔF/F)_max (1.9 ± 0.3, s.d., n = 3) in CyCaMP1b (Table 1, Figure 1B). YCaMP1a was improved by the incorporation of the Citrine (Griesbeck et al., 2001) mutations Thr65/222Gly, Val68/225Leu, Gln69/226Met, and Ser72/229Ala, resulting in YCaMP1b [(ΔF/F)_max = 9.2 ± 0.4, s.d., n = 3] (Table 1, Figure 1C). We incorporated the EBFP2 (Sato et al., 2007) and Azurite (Mena et al., 2006) mutations into BCaMP1a, but these did not result in improved variants. Screening variants of the M13pep-cpFP linker (“linker1”) produced a variant BCaMP1b, with amino acids Leu-Glu replaced by Met-Pro, with (ΔF/F)_max of 0.9 ± 0.2 (s.d., n = 3; Table 1). A cpFP-CaM linker (“linker2”) variant BCaMP1c with Phe-Pro instead of Thr-Arg, had a (ΔF/F)_max of 2.0 ± 0.1 (s.d., n = 3; Table 1, Figure 1A). The combination of the two linker variants resulted in a poor sensor (data not shown).

Initial engineering of the RCaMP scaffold

A red-colored (albeit quite dim) variant of Aequorea victoria GFP has been published (Mishin et al., 2008); we made the corresponding mutations to GCaMP3, but the resulting protein displayed no fluorescence (data not shown). Subsequently, we circularly permuted the red FP mCherry (Shaner et al., 2004) and replaced cpGFP in GCaMP3 with cp-mCherry; this protein was not fluorescent either (data not shown), in agreement with the absence of chromophore formation in cp-mCherry described previously (Carlson et al., 2010). We therefore selected another protein, mRuby (Kredel et al., 2009), as a template for a red calcium indicator. We reasoned mRuby would be more amenable to circular permutation because of its robust thermodynamic stability. Its parental protein, eqFP611, has been shown to contain a trans-conformation of its chromophore (Petersen et al., 2003), shifting the tyrosyl moiety of the chromophore in between β-strands 7 and 8 of the FP barrel. Interestingly, mutagenesis of eqFP611 to RFP639 shifts the chromophore to the cis-conformation (Nienhaus et al., 2008), bringing the tyrosyl moiety in close proximity to strand 10 as well. We therefore tried two different circular permutations, at residues 159 (strand 7) and 196 (strand 10), and additionally tried swapping the M13pep and CaM domains in each circular permutation (see Figure 2A for a schematic of the RCaMP family). The RCaMP scaffold with the circular permutation at the same region as GCaMP (after amino acid 159), sandwiched between N-terminal M13pep and C-terminal CaM in the same positions, was very dimly fluorescent in Escherichia coli colonies after prolonged (4 days) incubation at 4°C; the other scaffolds were not fluorescent (not shown) and abandoned during further optimization.

Random mutagenesis of linker1 (M13pep-to-cp-mRuby; originally the same Leu59-Glu60 as in GCaMP, from the translated XhoI site) resulted in four variants (Pro-Arg-Ile, Cys-Ile, Arg-Ile, Ala-Ile) with faster onset of fluorescence in E. coli (~1 day). The variants expressed well in E. coli and eluted as monomers during size exclusion chromatography (SEC) (Figures 2B,C). Denaturing SDS-PAGE gels showed that the RCaMPs exhibit a significant population of backbone-cleaved species (Figure 2D). Of the four variants, the Ala-Ile linker variant (“RCaMP-AI”) showed the largest (ΔF/F)_max (2.6 ± 0.1, s.d., n = 3) (Table 1, Figure 2E) and was selected for further optimization. A spontaneous mutation that arose during gene assembly (Asp114Ser) was also incorporated into RCaMP (Table 1). Fluorescence spectra of purified RCaMP-AI with 450 nm excitation indicated the presence of a substantial green fluorescent component (not shown), potentially from incomplete maturation of the red species by stalling in a green intermediate state, as proposed for the red FP zFP574 (Ivashkin et al., 2011). To optimize maturation, we performed error-prone PCR of the RCaMP-AI template and screened E. coli colonies for variants with faster and more complete red chromophore maturation. This produced RCaMP1a, with the additional mutations Ala112Tyr and Thr364Ile (Table 1).

Crystal structures of mRuby, RCaMP, and R-GECO1

To aid our protein engineering efforts, we sought to solve the crystal structures of mRuby, RCaMP, and R-GECO1. Structural information on the parent protein of mRuby, eqFP611 (86% sequence identity with mRuby), has been published (Petersen et al., 2003), however, no protein structure of mRuby was available at the time of initial RCaMP engineering, hampering optimization efforts. It has previously been shown that pH can have a large effect on chromophore conformation in red FPs (Battad et al., 2007; Pletnev et al., 2008; Pletneva et al., 2011); therefore we determined the crystal structure of mRuby at two different pHs: 4.5 and 8.5 (Table 2). The large difference in pH had no notable effect on overall mRuby protein structure (RMSD = 0.13 Å for 208 C_α atoms). The structure of mRuby at pH 8.5 shows a trans-isomer of the red 2-(iminomethyl)-4-(4-hydroxybenzylidene)-imidazol-5-one chromophore, while the structure at pH 4.5 shows a mixture of ~60% trans and ~40% cis. Similar pH-dependent cis-trans conformational changes also occur in the related FP eqFP578 (Pletneva et al., 2011) (70% sequence identity with mRuby). For other RFPs [mKate (Pletnev et al., 2008), Katushka (Pletneva et al., 2011) and RtmS5-H148S (Battad et al., 2007); 68, 67, and 51% sequence identity with mRuby, respectively], an opposite pH-induced isomerization of the chromophore has been described. The RMSD between mRuby and eqFP611 is 1.11 Å for 195 common C_α atoms, with the largest differences in surface loops resulting from different crystal packing. It was previously shown that the equilibrium of cis-trans isomers of the chromophore in eqFP611 can be influenced by altering the hydrogen-bonding network surrounding the chromophore (Nienhaus et al., 2008), further complicating the predictability of RCaMP chromophore behavior. To directly observe the chromophore orientation and the structure and extent of cp-mRuby/CaM packing interactions, both of which assisted GCaMP engineering (Akerboom et al., 2009, 2012a; Tian et al., 2009), we solved the crystal structure of the early RCaMP variant RCaMP1a.

RCaMP1a was crystallized in the Ca²⁺-bound form and its structure was determined to 2.1 Å resolution by molecular replacement (Table 2). RCaMP crystallized as a monomer, consistent with solution measurements (Figures 3A, 2B,C).

Although the mRuby circular permutation was based on GCaMP, large differences between RCaMP and GCaMP in both the orientation of, and interface between, the CaM/M13 complex and the cpFP domain can be seen (Figures 3A,B). The CaM/M13 complex of Ca²⁺-RCaMP is generally similar to that of Ca²⁺-GCaMP (RMSD of 1.9 Å for 152 common C_α atoms), but adopts a slightly more closed and compact conformation. The cp-mRuby domain of RCaMP is structurally very similar to mRuby, with an RMSD of 0.5 Å for 209 C_α atoms. Strikingly, unlike GCaMP, there is no large opening in the side of the FP β-barrel created by the circular permutation of the mRuby domain in Ca²⁺-bound RCaMP (Figure 3B); the termini of cp-mRuby cross close to one another and produce minimal disruption to the β-barrel. In contrast to the construction of the original GCaMP protein (Nagai et al., 2001), no amino acids of mRuby were replaced by linker amino acids. As a consequence of the circular permutation, though, linker amino acids partake in β-barrel formation in the cp-mRuby domain of RCaMP1a: residue Asn143 of mRuby, which directly hydrogen bonds the phenolate oxygen of the chromophore, is spatially substituted by Thr294 of the second linker of RCaMP (Figure 3C); the crystal structure shows that the asparagine (Asn61 in RCaMP) is co-opted into the first linker. Mutagenesis of this threonine to any other amino acid, or deletion of amino acids to generate an opening similar to GCaMP, greatly diminishes fluorescence of RCaMP (data not shown).

In contrast with the chromophore in the structure of mRuby crystallized at pH 8.5, the RCaMP structure (crystallized at pH 7.5) shows the (4-hydroxybenzylidene)-imidazol-5-one moiety of the chromophore to be a cis-isomer (Figures 3C,D). Superimposition of both mRuby structures and the cp-mRuby domain of RCaMP demonstrates a slight (~1 Å) shift of the RCaMP chromophore (red structure in Figure 3C) relative to both isomers of the mRuby chromophore (cyan in Figure 3C). Closer inspection of the electron density map for the RCaMP chromophore revealed a discontinuity between Phe213 and Met214, which forms part of the chromophore (Met214-Tyr215-Gly216) (Figure 3D). The electron density map is consistent with imine hydrolysis of the RCaMP chromophore, resulting in a ketone on Met214, maintaining the extended π-conjugation of the red chromophore (thus 2-acyl-4-(4-hydroxybenzylidene)-imidazol-5-one), and an amide on Phe213 (Figures 3D,E). This imine hydrolysis results in a discontinuity in the polypeptide backbone of RCaMP, also observed by SDS-PAGE of purified protein samples (Figure 2D). R-GECO1 showed a comparable level of backbone cleavage, with two smaller protein bands corresponding to imine hydrolysis of the chromophore (Figure 2D). The electron density map for mRuby shows no significant imine hydrolysis (Figure 3D). However, after boiling in the presence of the reducing agent 2-mercaptoethanol during sample preparation for SDS-PAGE, cleavage products can clearly be detected (Figure 2D), in agreement with structural and SDS-PAGE analysis of the red FP DsRed (Gross et al., 2000). Equivalent amide hydrolysis is reported for the photoactivatable Kindling Fluorescent Protein (asFP595) (Tretyakova et al., 2007). Interestingly, cp-mRuby does not show significant levels of imine hydrolysis (Figure 2D), and also lacks chromophore formation (not shown). The red-shifting (~15 nm excitation, ~4 nm emission) of the RCaMP chromophore relative to mRuby may result from rearrangements around the chromophore, as seen for eqFP611, RFP611, RFP618, RFP630, and RFP639 (Kredel et al., 2008), or from planarization following chromophore hydrolysis.

In addition to mRuby and RCaMP, we also crystallized R-GECO1 (Zhao et al., 2011a) in the presence of Ca²⁺ and solved the crystal structure to 2.2 Å by molecular replacement (Table 2). Strikingly, the position of the CaM domain relative to cp-mApple is distinct from that seen in RCaMP and GCaMP (Figure 3A)—the homology model used during R-GECO1 design (Zhao et al., 2011a)—underlining the importance of structure determination in protein engineering. Many of the mutations incorporated into the R-GECO1 sensor during initial screening and selection (Zhao et al., 2011a) are concentrated at the proto-interface between mApple and CaM (Figure 3G), as is the case for RCaMP (Figure 3F). Three residues of the M13pep-cp-mApple linker (Val60, Val61, and Ser62) make hydrophobic and hydrogen-bonding contacts with CaM, cp-mApple, and the rest of the linker. Similar to GCaMP (Wang et al., 2008; Akerboom et al., 2009), the cp-mApple-CaM linker is not in close proximity to the chromophore, and no cp-mApple-CaM linker mutations were selected in R-GECO1 (Zhao et al., 2011a). In addition to the linkers, R-GECO1 mutations Gly79, Arg81, Cys129, Phe366, and Asn380 are in the vicinity of the proto-interface (Figure 3G). No backbone cleavage was observed in the vicinity of the R-GECO1 chromophore, as was seen in the RCaMP crystal structure. R-GECO-1-Lys78, from strand eight of cp-mApple, adjacent to the circular permutation site, forms an ionic interaction with the phenolate oxygen of the chromophore and is stabilized in place via a hydrogen bond from Ser62, immediately following the M13pep-cp-mApple linker (Figure 3H).

Optimization of RCaMP

The structure of RCaMP explains the initial selection of mutations resulting in RCaMP1a, and allowed us to prioritize sites of mutagenesis for additional rounds of sensor engineering. As in GCaMP (Wang et al., 2008; Akerboom et al., 2009), the circular permutation of mRuby and fusion of M13pep and CaM resulted in a tightly packed adventitious interface between CaM and cp-mRuby, excluding solvent from the chromophore environment. The first linker, connecting M13pep and cp-mRuby, comes in close proximity to the chromophore, and the strongly selected isoleucine side-chain packs tightly against the inter-domain linker of CaM and cp-mRuby. Thr294, originally from the translated MluI restriction site in the second linker, hydrogen-bonds the RCaMP chromophore, stabilizing it in the deprotonated state (Figure 3C). The spontaneous (PCR-derived) Asp114Ser mutation improves packing against the second linker, especially Gln297 (Figure 3F).

Both mutations from error-prone PCR mutagenesis, Ala112Tyr and Thr364Ile, improve packing at the mRuby-CaM interface (Figure 3F). The improved packing of the interface by these mutations simultaneously increases brightness, red state maturation completeness and kinetics, (ΔF/F)_max, affinity and cooperativity. A second round of mutagenesis selected Gly109Ala, further improving packing at the mRuby-CaM interface (Figure 3F) in RCaMP1b, with increased affinity at a slight loss in (ΔF/F)_max (Table 1). The GCaMP5 mutation CaM-Asp380Tyr (Akerboom et al., 2009) is positioned at the GFP-CaM interface near the GFP chromophore and led to an increase in the calcium-bound fluorescence of GCaMP (Akerboom et al., 2012a). The corresponding mutation to RCaMP1b, Asp372Tyr, gave rise to RCaMP1c, with (ΔF/F)_max = 7.5 (Table 1). RCaMP1d was discovered as a spontaneous, fast-maturing mutant; faster chromophore maturation comes at the cost of affinity (1.6 ± 0.06 μM, s.d., n = 3) and (ΔF/F)_max (6.5 ± 0.2, s.d., n = 3) (Table 1). RCaMP1d was found to contain the mutation Ala270Val; the corresponding mRuby mutation, Ala120Val, was independently observed recently during the generation of mRuby2 (Lam et al., 2012), which exhibits greater brightness and photostability. Deletion of the RSET peptide resulted in RCaMP1e, which was significantly dimmer in cells, perhaps indicating that RCaMP is stabilized in situ by this N-terminal fusion peptide, as were early GCaMP variants (Tallini et al., 2006). We attempted to increase affinity by introducing the CaM mutation Arg90Gly (Sorensen and Shea, 1996; Akerboom et al., 2012a) (RCaMP numbering Arg384Gly) and screening libraries around the four EF-hands of CaM, which produced RCaMP1f. Affinity surprisingly decreased slightly to 1.9 ± 0.08 μM (s.d., n = 3), but (ΔF/F)_max rose to 12.3 ± 0.2 (s.d., n = 3; Table 1, Figures 2A,E). Lastly, targeted mutagenesis of the second linker (cp-mRuby-to-CaM; encompassing the Thr294-Arg295 linker, as in GCaMP, from the translated MluI site, as well as the subsequent residue Asp296) resulted in RCaMP1h, with 1.3 ± 0.06 μ M affinity, and (ΔF/F)_max of 10.5 ± 0.1 (s.d., n = 3; Table 1, Figure 2A). The crystal structure of RCaMP1a suggests that the charge-reversal mutation Arg295Asp likely improves interactions with Arg116 from cp-mRuby (Figure 3F). (RCaMP1g did not perform well in initial characterization and was not pursued further).

Fluorescence response to Ca²⁺ titrations of purified RCaMP and R-GECO1 sensor proteins are shown in Figure 2E-left; a Hill plot of the data shows linear ranges (Figure 2E-right). All RCaMP variants after RCaMP-AI show near complete maturation to the red state and similar extent of backbone cleavage (Figure 2D). All RCaMPs behave as monomers in solution (Figures 2B,C) and are fluorescent indicators of Ca²⁺ under both 1- and 2- photon excitation (RCaMP1f is shown in Figure 2F).

Photophysical characterization of chromatic variants

Next we purified several RCaMP variants, as well as R-GECO1 and the parent fluorescence proteins mRuby and mApple, and subjected them to systematic photophysical characterization, under both 1- and 2-photon illumination. We have previously analyzed the fluorescence mechanism of the GCaMP calcium indicator (Akerboom et al., 2009; Mütze et al., 2012), which involves a Ca²⁺-dependent deprotonation of the p-hydroxybenzylideneimidazolinone chromophore, leading to increased fluorescence. In GCaMP, the quantum yield (QY; and fluorescence lifetime) of the Ca²⁺-bound (bright) and Ca²⁺-free (dim) states are identical; the fluorescence increase upon Ca²⁺ binding is manifested entirely as an increase in extinction coefficient (ε), reflecting a greater population of deprotonated, fluorescent chromophore (Mütze et al., 2012). Strikingly, in the RCaMP indicators we found that more of the Ca²⁺-dependent increase in brightness was attributable to changes in quantum yield and fluorescence lifetime, than to changes in the extinction coefficient (Table 3). The peak of the RCaMP Ca²⁺-bound state is slightly spectrally shifted relative to the unbound state (from ~8 nm blue-shifted to ~1 nm red-shifted, for the different RCaMP sensors). R-GECO1 shows a larger peak shift: 12 nm blue-shifting of absorption, and 10 nm blue-shifting of emission. Under 1-photon illumination, the parent protein mRuby is slightly brighter (brightness = ε × QY) than EGFP; the brightness of the Ca²⁺-bound state of the different RCaMP indicators is 66–86% that of mRuby (Table 3).

pH titrations of the new sensors show some complex effects; some blue and red GECIs exhibit multiple titratable groups affecting fluorescence (Figures 1, 2, Table 1). Intriguingly, for some sensors, the pK_a of the Ca²⁺-bound state is higher than that of the Ca²⁺-free state (unlike GCaMP), implying that calcium binding decreases the acidity of the chromophore proton, increasing the protonated, and potentially non-fluorescent state of the chromophore (Tables 1, 3). The fact that the sensors nonetheless exhibit such large increases in fluorescence implies that the photophysical transitions upon calcium binding are more complicated than in GCaMP, consistent with the Ca²⁺-dependent increase in quantum yield, and the pH-dependent cis-trans conformational changes seen in RCaMP and other RFPs (Battad et al., 2007; Pletnev et al., 2008; Pletneva et al., 2011). R-GECO1, by contrast, shows a GCaMP-like Ca²⁺-dependent decrease of pK_a from 8.9 to 6.6 (Zhao et al., 2011a), consistent with a fluorescence increase driven largely by extinction coefficient effects (Table 3).

Importantly, RCaMP shows large (ΔF/F)_max signals under 2-photon illumination, on the order of the 1-photon signal change (Figure 2F). Under 2-photon illumination, the RCaMP sensors are red-shifted and surprisingly brighter than mRuby (Table 3). For the red indicators, excitation with either a Ti:Sapphire laser or an optical parametric oscillator (OPO) coupled to a primary exciplex laser produced large fluorescence changes upon calcium addition (Figure 2F).

Red FPs frequently exhibit poor photostability (Drobizhev et al., 2011); this is a serious drawback that limits their utility in experiments. Other photophysical phenomena also complicate usage; for instance, mApple shows significant photoswitching (Shaner et al., 2008). We examined the photoswitching and photobleaching properties of the red GECIs and their parent FPs under 1-photon (either lamp excitation or wide-field laser illumination) and 2-photon (laser) regimes. As published, mApple showed an immediate, ~50% drop in fluorescence upon illumination, which recovered slowly during darkness (data not shown). R-GECO1 showed similar rapid-photoswitching properties (Figure 4A), which may confound functional imaging. RCaMP1h displayed a more conventional, irreversible form of photobleaching, and appeared to bleach more slowly for the first several minutes of bright mercury lamp illumination (2.7 W/cm²). Two-photon illumination of HEK293 cells expressing RCaMP1h or R-GECO1 showed a somewhat faster decay of fluorescence for RCaMP1h compared to R-GECO1 (Figure 4B). However, 2-photon brightness of several RCaMPs was more than double that of R-GECO1 (Table 3, Figure 4C). Stopped-flow fluorescence showed that decay kinetics of R-GECO1 are about twice as fast as GCaMP5G; RCaMP1h is slightly slower than GCaMP5G (t_1/2(decay): R-GECO1, 120 ms; GCaMP5G, 280 ms; RCaMP1h, 410 ms) (Figure 4D).

Strikingly, R-GECO1 displayed significant photoactivation after illumination (~2-fold) with 405 nm, 488 nm and 561 nm light for both calcium-free and calcium-loaded states of the protein, although the activation is different in nature between the two states. This effect was strongest for 488 nm, followed by 405 nm and approximately 10-fold less for 561 nm. Applying 1 s pulses of 488 nm light resulted in a red-shift in absorbance of calcium-loaded R-GECO1 (Figure 4E-left), resulting in a temporary decrease (~40%) in fluorescence when illuminated with 561 nm (Figure 4E-right). For calcium-free R-GECO1, 1 s pulses of 488 nm light resulted in an apparent decrease in absorbance of protonated chromophore and a simultaneous increase in absorbance in deprotonated chromophore (Figure 4F-left), mimicking the effect on fluorescence of calcium binding to R-GECO1 (Figure 4F-right) when illuminated with 561 nm light. This effect resulted in an apparent increase in fluorescence (~50%, Figure 4F-right). Photo-activation could be observed using a range of illumination intensities (0.056–0.66 mW/cm², 488 nm, not shown). Transient decay of the photoactivated state for both calcium-free and calcium-loaded R-GECO1 followed one-phase decay kinetics, with a half-life of 560 ms and 940 ms, respectively (Figures 4H,I). There is also a wavelength shift in the transient fluorescence, a small red-shift of 2 nm for Ca²⁺-bound and 7.5 nm for Ca²⁺-free.

These photophysical effects are not observed for RCaMP (Figure 4G), suggesting that RCaMP may be a better sensor to be used in combination with optogenetic tools.

Characterization in HEK293 cells and neurons

The new sensors, as well as R-GECO1, were next tested in cultured HEK293 cells and neurons, according to protocols that we have established for optimizing GCaMP (Tian et al., 2009; Akerboom et al., 2012a,b). The red GECIs were first tested in transfected HEK293 cells following acetylcholine (ACh)-evoked Ca²⁺ mobilization (Figures 5A,B). Subsequently, cultured rat hippocampal neurons were infected with lentivirus driving GECI expression and imaged following AP elicitation with a bath electrode (1 AP per field stimulation) (Akerboom et al., 2012a) (Figures 5C,D). In both HEK293 cells and neurons, the RCaMP and R-GECO1 sensors were expressed in both the cytoplasm and nucleus, in contrast to GCaMP, which appears to have a cryptic nuclear-exclusion sequence (Tian et al., 2009). RCaMP- and R-GECO1- expressing cells appeared healthy; thus the nuclear expression does not seem to correlate with the “cytomorbid” state seen with long-term GCaMP over-expression (Tian et al., 2009). The expression in the nucleus may give rise to a slower, weaker Ca²⁺ response (Bootman et al., 2009) than the cytoplasmic response, with the conflated epifluorescence signal thus appearing slower and less sensitive to APs.

The four RCaMP sensors tested (1d, 1e, 1f, and 1h) performed similarly well in neurons, showing observable responses following a minimum of 5 field stimuli, and reaching a maximum ΔF/F of 1.0–1.5 after 160 field stimulations. RCaMP1f and RCaMP1h were approximately twice as bright in cultured neurons as the other RCaMPs, and displayed almost identical ΔF/F and SNR. R-GECO1 was more sensitive, responding to 2 APs. SNR for R-GECO1 was also higher than for the RCaMPs, although the RCaMPs are brighter (Figure 5D). YCaMP1b showed twice the response of the RCaMPs at higher numbers of field stimuli, but required a minimum of 20 field stimuli to observe any response. However, SNR for both RCaMP1f and RCaMP1h was higher than YCaMP1b at high field stimulations, as YCaMP1b is dimmer.

The RCaMPs were also somewhat slower than R-GECO1 in the rise and decay from electrical stimulation. For the 10 AP stimulation, the time-to-half-decay t_1/2(decay) and time-to-half-rise t_1/2(rise) values for the GECIs were (R-GECO1: 0.78 ± 0.13/0.09 ± 0.02 s, n = 15; RCaMP1d: 1.61 ± 0.85/0.29 ± 0.03 s, n = 16; RCaMP1e: 1.57 ± 0.20/0.29 ± 0.02 s, n = 13; RCaMP1f: 1.80 ± 0.33/0.27 ± 0.02 s, n = 16; RCaMP1g: 1.69 ± 0.63/0.28 ± 0.05 s, n = 16; RCaMP1h: 1.83 ± 0.30/0.32 ± 0.04 s, n = 15; all measurements s.d.).

Two-color sub-cellular imaging in neurons

Having established the baseline performance level of the RCaMP indicators in neurons, we sought to demonstrate the utility of RCaMP in two-color imaging, beginning by labeling two compartments of one cell. First, RCaMP1e was fused with an N-terminal cytochrome C oxidase subunit VIII tag for targeting to the mitochondrial matrix, expressed by AAV6 virus, and co-infected along with cytoplasmic GCaMP3.5 (Tian et al., 2009) (GCaMP3 without the CaM-Asn60Asp mutation; affinity ~3-fold lower than GCaMP3) into primary rat cortical neurons. Upon stimulation of Ca²⁺ release from the endoplasmic reticulum (ER) by addition of 10 mM caffeine, RCaMP fluorescence initially rose together with GCaMP fluorescence (Figure 5E), consistent with Ca²⁺ release from the ER through the mitochondrial-associated membrane (MAM) (Hayashi et al., 2009), while the bulk of Ca²⁺ released from the ER entered the cytoplasm before being buffered in mitochondria. Likewise, depolarization with 64 mM KCl demonstrated entry of extracellular Ca²⁺ into the cytoplasm and subsequently into mitochondria (Figure 5F). These experiments demonstrate the simultaneous visualization and tracking of sub-cellular Ca²⁺ fluctuations using RCaMP together with GCaMP. This paradigm may be expanded to tracking of sub-cellular Ca²⁺ mobilizations under various physiological or pathophysiological conditions in various cell types.

Two-color imaging in mixed culture of neurons and astrocytes

Communication between astrocytes and neurons has been implicated in information processing in the brain. We therefore sought to test the utility of dual-color neuron/astrocyte imaging using RCaMP1h and GCaMP5G (Akerboom et al., 2012a) (Figure 6). RCaMP1h is 80 nm red-shifted relative to GCaMP5G; therefore signals from each population of cells are expected to be easily deconvoluted. First, RCaMP was expressed in cultured rat hippocampal neurons via AAV2/1-synapsin1 virus infection; GCaMP5G was expressed in astrocytes via AAV2/1-GFAP virus (Figure 6A). The two infected cell populations were then co-cultured. One week post-infection, we observed robust labeling of neurons only with RCaMP and astrocytes only with GCaMP. Simultaneous dual-color imaging of neuronal and astrocytic responses to field stimulations was then performed. The number and magnitude of evoked calcium events in both neurons and astrocytes increased with the number of evoked APs, and most importantly, dual-color imaging revealed the spatiotemporal relationship between field stimulus-evoked neuronal and astrocytic calcium responses (Figures 6B–D). At lower number of field potential stimuli, calcium events could be detected only in astrocytes in close proximity to responding neurons (Figures 6B,C). Following larger stimuli, more neuronal activity was detected, which triggered broad astrocytic calcium events in the imaged field (Figure 6B). The peak responses of astrocytes typically lagged neuronal responses by ~1 s (Figure 6D). These results clearly show that dual-color imaging with RCaMP and GCaMP (and other indicators) holds great promise to dissect neuron-astrocyte, or other inter-cellular, communication.

Imaging RCaMP activity in Drosophila larval motor neuron terminals

We next sought to establish RCaMP performance in Drosophila larvae using the GAL4-UAS system (Brand and Perrimon, 1993). We drove expression of RCaMP1f in larval motor neurons using a GAL4 driver line specific for glutamatergic neurons [OK371-GAL4 (Mahr and Aberle, 2006)]. Muscle 13 neuromuscular junctions (NMJs) were then imaged in dissected fillet preparations. Motor neuron terminals showed bright fluorescence; RCaMP transients in the terminals were imaged under 1-photon illumination in response to motor nerve stimulation (Figure 7A). Terminals showed little or no response to long (2 s) bursts of APs at 1–10 Hz (Figure 7B). Above 20 Hz, motor terminals showed modest peak amplitudes, peaking at ~30% ΔF/F in response to 160 Hz stimulation (Figure 7C). RCaMP responded slowly to all stimulation frequencies; times to half-peak and half-decay were ~1 s at 40, 80, and 160 Hz (rise and decay time constants could only be accurately measured at the highest stimulation frequencies).

Imaging sensory-evoked Ca²⁺ transients in worms

We subsequently imaged RCaMP activity in the AWC neurons of C. elegans in response to odor addition and removal sequences, as previously described (Akerboom et al., 2012a). RCaMP1a and RCaMP1e were selected for testing. All RCaMP constructs were expressed in the AWC^on sensory neuron under the control of the str-2 promoter. After 10 s of observation, odor (IAA @ 10⁻⁴ v/v) was added for 5 min. At the last 10 s of odor addition, imaging was commenced for another minute to observe the neuron's response to odor removal (the AWC^on neuron activates in response to odor removal). Interestingly, RCaMP1a outperformed RCaMP1e during IAA addition; ΔF/F and SNR of RCaMP1a were higher compared to RCaMP1e for IAA addition (Figure 7D), although RCaMP1e showed almost identical ΔF/F and SNR upon odor removal.

In addition, we demonstrated the utility of RCaMP1h in visualizing contractions of the pharyngeal muscles (strain ZX1423; zxEx795[pmyo2::RCaMP1h; pmyo-3::eCFP]). Worm heads were cut off (Avery et al., 1995), placed in an imaging chamber in buffer Em D50, and illuminated with a 590 nm LED (80 mW/cm²). As a control, worm heads of another transgenic strain (expressing mCherry in pharynx muscle) were imaged separately. RCaMP1h signal tracked pharyngeal pumping events, with large, reproducible increases up to ~40% ΔF/F (Figure 7E). In control animals, mCherry signal did not vary during the same preparation, indicating that motion artifacts play no role in the observed RCaMP signal (Figure 7E, bottom).

In vivo imaging in zebrafish trigeminal neurons

We next imaged activity in trigeminal sensory neurons of 48 h post-fertilization zebrafish using RCaMP1a and RCaMP1b (Figure 7F). Trigeminal neurons are usually silent, and fire one or a few spikes in response to light touch (Douglass et al., 2008). We recorded fluorescence changes in response to trains of brief (1 ms) pulses of electrical field stimulation (Materials and Methods). Both indicators performed similarly, giving steadily increasing responses for trains of 20–320 pulses, reaching a maximum ΔF/F of about 100% (Figure 7F-left) and SNR of ~25 (Figure 7F-right).

Integrated optogenetics/imaging with RCaMP and channelrhodopsin-2 (ChR2)

As discussed above, simultaneous light-mediated activation of neural activity (optogenetics) and fluorescent imaging of activity (either in the same or a different cell) is currently quite difficult, as the action spectra of opsins are very broad (Zhang et al., 2011), overlapping the excitation spectra of commonly used indicators such as GCaMP, Yellow Chameleon, and TN-XXL. In such regimes it is impossible to image the indicator without inadvertently driving the actuators, precluding baseline measurements. Furthermore, activators are typically driven with strong light intensities to achieve fast, reliable cell depolarization, and this can bleach indicators if they overlap. Integrated, non-overlapping imaging/optogenetics will enable qualitatively new sorts of neuroscience experiments, such as mapping projections of numerous single neurons through brain tissue, akin to an “all-optical” channelrhodopsin-assisted circuit mapping (CRACM) experiment (Petreanu et al., 2007). To facilitate such experiments, GECIs must either be red-shifted past effector action spectra, or effectors must be shifted away from GECI excitation. Significant progress has been made at red-shifting optogenetic activators (the C1V1 variant (Yizhar et al., 2011b) being the most red-shifted variant available, with maximum absorption at 550 nm); however, the action spectrum of this effector still substantially overlaps GECI excitation. Therefore, blue light activation combined with red Ca²⁺ indicators appears to be the best option for integrated optogenetics. The use of R-GECO1 and its mutants in combination with ChR2 in pyramidal neurons has been recently reported (Chang et al., 2012; Ohkura et al., 2012). However, the blue light-induced photoactivation of RGECO-1, as described above in protein and below in vivo, calls into question the nature of the fluorescent transients in these papers. Both papers employed mammalian neurons, which have sufficient levels of all-trans retinal (ATR) to support ChR2 function. In both cases, no ChR2-free controls to exclude the possibility of R-GECO1 blue-light photoactivation being the principal contributor to the observed signals were performed. Our results with R-GECO1 suggest that some or all of the observed signals previously reported on may be artifactual.

First, we tested RCaMP1e together with ChR2-Thr159Cys [ChR2(TC)] (Berndt et al., 2011), as well as GCaMP3 (Tian et al., 2009) with the red-shifted channelrhodopsin C1V1 (Figure 8A). Initial experiments were performed in cultured cells, using GECIs directly fused to effector proteins (Figure 8A), in order to measure Ca²⁺ flux into the depolarized cells close to the cell membrane (Figure 8B). This arrangement creates equimolar amounts of effector and sensor, largely controlling for expression level differences in this preliminary intra-cellular assay. To measure Ca²⁺ flux we employed a stable HEK cell line expressing the leaky mTrek potassium channel and a voltage-gated calcium channel (Ca_V3.2). In darkness, membrane voltage is controlled by extracellular potassium. Upon light-activation of ChR, cells depolarize and subsequent Ca_V3.2 opening leads to an increase in intracellular [Ca²⁺] (Figure 8B). Expression of the two fusion constructs was comparable (Figure 8C). The excitation spectrum of GCaMP3 and the action spectrum of C1V1 show significant overlap compared to RCaMP1e and ChR2(TC) (Figure 8D). To characterize the level of cross-activation we first tested photocurrent amplitudes induced by light of 400, 450, and 560 nm, for both constructs (Figure 8E). ChR2(TC) exhibits virtually no activation at 560 nm (near the peak of RCaMP1e excitation), but can be activated with 400 and 450 nm (Figure 8F). C1V1 shows activation of more than 40% at all tested wavelengths, including GCaMP3 excitation at 450 nm (Figures 8E,F).

The separation of ChR2(TC) activation and RCaMP excitation allowed intense, simultaneous stimulation and imaging, at 450 nm (56 mW/cm²) and 560 nm (20 mW/cm²), respectively (Figure 8G-top). With 2 mM extracellular [Ca²⁺], 300 ms pulses of 450 nm light were sufficient to open calcium channels, giving rise to robust RCaMP transients (7.5 ± 0.9%, s.d. 5.43, n = 36), although variability between single-trial traces was relatively high (Figure 8H). Although it was also possible to image GCaMP3 fluorescence in conjunction with C1V1 activation, the spectral overlap required the excitation beam to be significantly attenuated (0.7 mW/cm², 440 nm light), which produced very little fluorescence (Figure 8G-bottom). With 5 mM extracellular [Ca²⁺], 300 ms pulses of 560 nm light yielded only small increases in GCaMP3 fluorescence (3.7 ± 0.1%, s.d. 0.43, n = 7) (Figure 8H). Increasing extracellular [Ca²⁺] improved fluorescence signals in both preparations (Figure 8H), with the RCaMP increases reaching 35% (ΔF/F)_max, and GCaMP3 increases reaching 6% (ΔF/F)_max. Addition of the calcium channel blocker mibefradil abolished signals in both cases, indicating that very little Ca²⁺ is flowing directly through the opsins (Figure 8H) (Prigge et al., 2010).

Having established integrated optogenetics with RCaMP and ChR2 in cell culture, we then deployed them in vivo in C. elegans, expressing RCaMP in muscles, and ChR2 either in muscles as well, or in the upstream motor neurons. RCaMP expression was bright and restricted to muscles (Figure 9A). First, we tested the performance of RCaMP1e in visualizing spontaneous muscle contractions, in semi-restrained animals (Figure 9B). Fluorescence traces in ROIs from the dorsal and ventral sides showed clear antiphasic fluorescence increases (Figure 9C), consistent with expected muscle contractions during crawling. Fluorescence increases approached 50–60% during spontaneous muscle contraction.

Next ChR2 was co-expressed in muscles alongside RCaMP1e or RCaMP1h, to compare the responses of light-evoked contractions to spontaneously arising ones. Worms lack ATR, thus to facilitate ChR2 activation, worms were cultivated in the presence of ATR. Negative control experiments were performed in the absence of added ATR. Light pulses from a blue LED (100 ms, 470 nm, 50 mW/cm²) were used to activate ChR2; little photobleaching of RCaMP was seen during experiments. RCaMP was excited with a yellow-orange LED (590 nm, 79.3 mW/cm²), which elicits no activation of ChR2 (Figures 8D,E, 9D–H). The blue light stimulus caused a small fluorescence increase in the RCaMP channel, due to autofluorescence from the tissue (Figures 9D–F,I). To account for this, background signals were subtracted from images during the photostimulation, when prolonged stimuli were applied (Figures 9G–I). Blue light pulses evoked large, slow RCaMP transients in muscles; signals were comparable to those from naturalistic muscle contractions, although they decayed more slowly (Figure 9D). In the absence of retinal, no RCaMP signal increases were observed. Thus, the observed RCaMP signals are due exclusively to ChR2-evoked activity, and not to any potential artifacts.

Subsequently, we expressed ChR2 in cholinergic motor neurons [transgene zxIs6, using the punc-17 promoter (Liewald et al., 2008)], pre-synaptic to muscle cells, and RCaMP1h in muscles. Again, we presented 100 ms blue light pulses and could detect a subsequent Ca²⁺ increase on the order of that seen from spontaneous muscle contraction and ChR2-evoked muscle depolarization (Figure 9E). The cholinergic neuron-evoked signal reached its peak after ~3 s and returned to baseline after ~12 s (Figure 9E). Again, this signal increase was only observed if retinal was added during cultivation. In muscles that were not transgenic for RCaMP, but nevertheless were innervated by ChR2-positive neurons, no fluorescence increase could be observed (Figure 9F). Similarly small amounts of autofluorescence were observed between RCaMP-positive and RCaMP-negative worms (data not shown).

We further characterized the motor neuron-induced Ca²⁺ transients in muscles expressing RCaMP1e. We presented prolonged photostimuli (1.2 s, 470 nm) to the ChR2-expressing cholinergic neurons; robust responses were seen only in RCaMP1e- positive muscles (Figure 9G). With increasing intensity of ChR2 activation, Ca²⁺ transients in muscles rose accordingly, from ~2% to ~25% (ΔF/F)_max, for 1 and 50 mW/cm² stimulus intensity, respectively (Figure 9H).

Above we showed that R-GECO1 shows dramatic reversible photoactivation by blue and green light, both in the presence and absence of Ca²⁺. We subsequently extended these results in vivo, imaging R-GECO1 with ChR2 expressed in cholinergic neurons. Although red fluorescence transients were observed upon blue light stimulation, identical responses were observed in worms not supplemented with ATR (Figure 9I), suggesting the signals result not from ChR2 activation of the muscles, but from photoactivation of R-GECO1. Interestingly, the half-life of fluorescence decay for R-GECO1 in C. elegans was similar to the half-life of photoactivation (Figures 4H,I, 9I). This illustrates that not only does R-GECO1 photoactivation produce a confounding signal to neural activity imaging, but also that the time course of the artifact is sufficiently similar to responses expected from Ca²⁺ transients following trains of APs that it may be mistakenly assigned as such (Chang et al., 2012; Ohkura et al., 2012; Wu et al., 2013).

Initial RCaMP construction

For the red protein in RCaMP we selected mRuby (Kredel et al., 2009). DNA encoding mRuby was generated by assembly PCR (Stemmer et al., 1995), and cloned in pRSETa (Invitrogen, USA). We decided to test two different circular permutation sites in mRuby; (1) between Pro142 and Thr143 in beta strand seven [homologous to cpEGFP in GCaMP (Nakai et al., 2001)], and (2) between His196 and Arg197 in beta strand 10 [in close proximity to a potential planar trans conformation of the chromophore in mRuby, as described for its parental protein eqFP611 (Petersen et al., 2003)]. M13pep and CaM DNA, including linkers, were identical to their counterparts present in GCaMP3 (Tian et al., 2009). We also swapped the M13pep and CaM domains for each circular permutation site. Assembly of the four resulting RCaMP constructs was performed as follows; the four different parts (the N-terminal (from Met1 to Pro142 and from Met1 to His196) and C-terminal part (N143 to G227 and Arg197 to G227) of mRuby, the DNA encoding the M13pep including (M1 to E60) and DNA encoding CaM (T302 to K450) (mRuby and GCaMP3 numbering, respectively) were PCR amplified from GCaMP3 and mRuby DNA. Seven codon overlaps were encoded in the primers. PCR products were purified by gel extraction, pooled in equimolar amounts for the corresponding constructs, and assembled in a PCR assembly reaction. One to five microliter of this reaction was then used as a template in a following PCR reaction using primers annealing to the N-terminus and C-terminus of the final PCR product. Resulting DNA was gel purified, digested with NdeI and HindIII and ligated into pre-digested pRSETa and sequenced for verification.

Directed evolution and rational optimization of RCaMP

Construction of YCaMP, CyCaMP, and BCaMP

YCaMP, CyCaMP, and BCaMP were constructed by Quikchange mutagenesis (Stratagene, USA) using GCaMP3 (Tian et al., 2009) in pRSETa as template. Reactions were performed according to the manufacturer's instructions and constructs were sequenced for verification. We used combinations of published mutations to generate the different excitation and emission characteristics. For YCaMP, we used V115Y and K118V to shift the excitation/emission to yellow. For cyan, we used Y223W and Y223W in combination with S229A to generate CyCaMP1a and CyCaMP1b, respectively. For BCaMP, we initially mutated Thr222Ser and Tyr223His. This resulted in a very dim non-responsive BCaMP1a, so we screened variants of both the M13pep-cpEBFP and cpEBFP-CaM linker as described above. We found two linker variants (LeuGlu to MetPro for M13pep-cpEBFP linker and ThrArg to PhePro for cpEBFP-CaM linker) with improved brightness and ΔF/F₀.

Protein expression and purification

Proteins were expressed and purified as described previously (Akerboom et al., 2009), with the following adaptations: proteins were expressed at 30°C for 36 h for RCaMP and at 25°C for the other color variants.

Size exclusion chromatography

To examine the oligomeric state of RCaMP, an aliquot of purified protein in 20 mM Tris, pH 8.0, 100 mM NaCl, 2 mM CaCl₂ was injected onto a Superdex 200 SEC column with 20 mM Tris, pH 8.0, 100 mM NaCl, 2 mM CaCl₂ as the running buffer.

Crystal structure determination

mRuby, RCaMP, and R-GECO1 proteins were expressed in E. coli BL21(DE3) from the pRSETa plasmid (Invitrogen). Proteins were purified by immobilized metal ion affinity and SEC. The N-terminal tag encoded by the pRSET vector was cleaved from mRuby proteins using enterokinase prior to SEC. Proteins were concentrated to ~10 mg/mL for crystallization using centrifugal concentrators (Millipore) with a 10 kDa molecular weight cut-off (MWCO). Crystallization using commercially available sparse matrix screens (Hampton Research) was done by mixing 1.2 μL of protein solution with 1.2 μL of precipitant in 96-well plates. X-ray diffraction data were collected at 100 K at the LRL-CAT beamline (31-ID) of the Advanced Photon Source. X-ray diffraction data were processed using Mosflm (Leslie, 1992; Battye et al., 2011)/Scala (Winn et al., 2011) (Table 2). Structures were solved by molecular replacement using Phaser (McCoy et al., 2007) with previously published structures of EqFP611 (PDB 3E5W) (Petersen et al., 2003), mCherry (PDB 2H5Q) (Shu et al., 2006) and CaM-M13 (from GCaMP, PDB 3EK7) (Akerboom et al., 2009). Iterative cycles of model building in Coot (Emsley et al., 2010) and refinement in Refmac/CCP4 (Winn et al., 2011) led to the final models described in Table 2.

Spectrophotometric analysis of purified protein

Fluorescence spectra were obtained as described previously (Akerboom et al., 2009). In short, pure proteins were dialyzed into 20 mM TRIS.HCl, pH 8.0, 100 mM NaCl and diluted 10-fold into zero-free calcium buffer (Invitrogen) for calcium-free spectra, and in 39 μM free-calcium buffer (Invitrogen) for calcium-loaded spectra. For absorbance spectra, undiluted samples were taken and EGTA or CaCl₂ was added to a final concentration of 10 mM. Affinity measurements were done in calcium standards prepared by mixing different amounts of 39 μM free calcium buffer and zero calcium free buffer together according to http://maxchelator.stanford.edu (Patton et al., 2004) and the manual provided with the calcium kit. Measurements were taken in a Tecan Safire² (Tecan Group Ltd., Switzerland), and data was analyzed using the GraphPad Prism software package (http://www.graphpad.com/prism/Prism.htm).

Stopped-flow kinetic analysis

The kinetics of calcium release by the indicators was determined from a single exponential fit to the fluorescence decay following rapid mixing of 1 μM protein samples in 10 μM calcium with a solution of 10 mM EGTA at room temperature, both buffered with 50 mM MOPS, 100 mM KCl at pH 7.2, using a stopped-flow device (Applied Photophysics) coupled to a fluorometer (Varian).

pH titrations

The pKa of both calcium-bound and calcium-free fluorescent indicators was determined by measuring the fluorescence at different pH values. Purified protein was diluted 20-fold into buffer A (20 mM Glycine, 20 mM Citrate, 20 mM TRIS.HCl) at different pH values, containing 2mM CaCl₂ for calcium-loaded spectra and 2 mM BAPTA for calcium-free spectra.

SDS-PAGE analysis

For SDS-PAGE, small (15 μL) protein samples were mixed with 5 μL LDS Sample buffer (Life Technologies, USA) supplemented with 2-Mercaptoethanol, incubated at 100°C for 5 min, cooled on ice, and loaded onto pre-cast NuPAGE Bis-Tris gels (Invitrogen), and run at 50 V in MES buffer provided, until satisfactory migration of protein bands was reached. Protein mass was determined by running 5–10 μL SeeBlue Plus2 protein marker (Life Technologies, USA) in parallel. Gels were stained with staining buffer [50% methanol, 10% glacial acid, 40% distilled water, 1 g/L coomassie brilliant blue R-250 (FisherBioscience)], and destained with destaining buffer (similar to staining buffer, with coomassie brilliant blue omitted). Gels were imaged using an AlphaImager HP (ProteinSimple, USA), and sizes were calculated using the AlphaEaseFC software package (Genetic Technologies, Inc., USA).

Photophysical measurements

Photophysical measurements were taken on purified protein in buffer (30 mM MOPS, 100 mM KCl) in the absence (10 mM EGTA) or presence (10 mM Ca-EGTA) of free Ca²⁺ (calcium calibration buffer kits, Invitrogen). All solutions had a measured pH of 7.28 ± 0.08. Absorption measurements were performed on a UV/VIS spectrophotometer (Lambda 35, Perkin Elmer) with 2-nm slit settings. Emission spectra were taken on a fluorescence spectrometer (LS55, Perkin Elmer) with 5 nm slit settings, using a quartz microcuvet to obtain excitation spectra in the UV in addition to the visible range. Quantum yield measurements were performed with an integrating-sphere spectrometer (Quantaurus, Hamamatsu), using protein solutions with an absorbance of about 0.1 OD at 570 nm. Two-photon spectroscopy measurements were performed on an inverted microscope (IX-81, Olympus) outfitted with a 60× 1.2 NA objective, using an 80 MHz Ti:Sapphire laser (Chameleon, Coherent) spanning 700–1080 nm, as described previously (Akerboom et al., 2012a; Mütze et al., 2012). An extended wavelength range (1000–1400 nm) was accessed by the addition of an OPO (Chameleon OPO, Coherent). Fluorescence collected by the objective, after passing through shortpass and bandpass filters (720SP, 625/90, Semrock), was detected using a fiber-coupled avalanche photodiode (APD) (SPCM_AQRH-14, Perkin Elmer), which operates in single-photon counting mode. Output pulses from the APD were fed to an autocorrelator (Flex03LQ, Correlator.com). This system was run under computer control with automated data acquisition, and allowed us to determine the two-photon-excited fluorescence excitation spectra. To perform time-resolved fluorescence lifetime measurements, we employed a fast-timing APD (PDF-CCTB, Micro Photon Devices) fed into a single-photon counting board (TimeHarp200, PicoQuant). We used two-photon-excited fluorescence correlation spectroscopy (FCS) to quantify chromophore concentration and two-photon specific brightness using EGFP at known concentration as a reference to calibrate the excitation volume. All FCS autocorrelation fits were performed using a custom fitting program in MatLab (Vijay Iyer, unpublished). We also used FCS to determine the fluorescence rate per molecule, or specific brightness, given by <F>/N = <F>G(0), as the laser power is varied. The peak brightness, the maximum value of the specific brightness, provides a metric to compare the photostability under two-photon illumination (Mütze et al., 2012). One-photon photobleaching measurements were made in an upright microscope (Imager.Z2, Zeiss), using a 20× 0.85NA objective, with illumination from either a Hg-lamp (excitation filter 550/25, Semrock), or a 561-nm laser (Sapphire 561 LP, Coherent). The intensity at the sample was 2.7 W/cm² for lamp illumination, and 6.0 W/cm² for 561-nm laser illumination. Samples consisted of proteins in aqueous buffer droplets (50 mM MOPS, 100 mM KCl, 1 mM CaCl₂, 0.1% BSA, pH 7.3) suspended in a nonpolar solvent (1-octanol) between two coverslips, following the method of Kremers and Piston (2010). Fluorescence from individual droplets contained within the area of uniform illumination was collected by the objective, passed through a bandpass filter (605/70, Semrock), and sent to either a CCD camera (CoolSnapEZ, Roper) or an APD to record the emission with high temporal resolution. In order to investigate photoswitching in the protein samples, additional illumination by a 405 laser (Sapphire 405 LP, Coherent) was combined along the path of the 561-nm laser. All light sources were outfitted with shutters, and exposure sequences were set by a custom software program.

Transient absorption and emission of R-GECO1 following 488-nm excitation

Absorption and emission properties of R-GECO1 and RCaMP1h were measured using a sub-microcuvet (16.40F-Q-10/Z15, Starna Cells) of 0.036 cm² area and 1 cm path length, in a setup providing 488-nm laser light delivered to the cuvet under shutter control, and a counter-propagating probe beam of lamp-light transiting through the same area of the cuvet. The transmitted probe light was sent to a fiber-coupled CCD-based spectrometer (QE65000, Ocean Optics), which captured and stored full absorption spectra of 100-ms duration at a frame rate of 10 s⁻¹. Optical density was calculated from -Log[S(1)/S_o(1)], where S_o(1) is the lamp-light signal recorded at each wavelength of the spectrometer in the absence of R-GECO1 and S(1) the signal measured with R-GECO1 present. Due to baseline drifting for this setup, we found it useful over the course of long experiments to determine just the transient absorption, ΔOD, from −Log[S_post-488(1, t)/S_pre-488(1)].

Transient fluorescence was recorded in this setup from side-window measurements of the cuvet. Emission light was focused onto a fiber-coupled florescence spectrometer (Pixis256E CCD + SB-2358 Acton spectrograph, Princeton Instruments), capable of recording 100 ms per spectra at a frame rate of 7.9 s⁻¹. The transient fluorescence was excited by weak 561 nm light (0.025 W/cm²). Measurements at 10-fold higher intensity yielded equivalent results, indicating the weak 561 nm light was not perturbing the photophysics.

Excitation laser light overfilled the 0.036 mm² open area of the cuvet, and incident intensities for 488, 405, and 561 nm were determined from measurements of laser power transmitted through the cuvet containing buffer only, divided by the open area of the cuvet (0.036 cm²). Laser exposure time was controlled with a shutter (Uniblitz) under software control. The spectrometers were operated in free-running data-acquisition mode, i.e., not synchronized to the laser shutter. However, we could determine accurately the timing from the end of the 488-pulse to the beginning of the first and subsequent data frames from laser fiducials appearing in each frame.

Samples of R-GECO-1 or RCaMP1h were diluted in the microcuvet containing buffer (MOPS/KCl/10mM-EGTA, pH 7.25), and data taken for this “no-calcium” case. Following these measurements, CaCl₂ was added to provide 10 mM final concentration in the buffer, the sample allowed to equilibrate, and measurements performed for the “+ Ca” case. Absorption experiments were performed for 3 concentrations (5, 10, 20 μM) and fluorescence experiments performed at 10 μM and 3 μM, in order to assess possible concentration effects (none observed).

HEK293 cell acetylcholine Ca²⁺ mobilization assay

RCaMPs were cloned into pCAGGS by PCR amplification, subsequent digestion with AgeII and NotI and ligation into predigested pCAGGS (Niwa et al., 1991). Constructs were sequenced for verification. Transfections were done in low passage number HEK293 cells using the Amaxa 96 nucleofector kit (Lonza Group Ltd., Switzerland). Cells were then incubated for 48 h in DMEM.HG (Corning, USA) supplemented with 10% Fetal Bovine Serum (FBS) at 37°C, 10% CO₂. Media was renewed after 24 h. After 48 h, growth media was aspirated and 100 μl 1× TBS buffer supplemented with 2 mM CaCl₂ was gently pipetted on top of the cells. Imaging was performed immediately afterwards in an FDSS plate reader (Hamamatsu, Japan). Dilutions of acetylcholine in 1× TBS supplemented with 2mM CaCl₂ were added 10 s after read initiation. Data was analyzed using the GraphPad Prism software package (http://www.graphpad.com/prism/Prism.htm).

Hippocampal and cortical neuronal culture imaging

Procedures were as described previously (Akerboom et al., 2012a). Briefly, dissociated hippocampal or cortical neurons obtained from P0 rat pups were plated in 24-well plates and transduced with GECI variants expressed from the human synapsin-1 promoter using lentivirus or adeno-associated virus (AAV). Fluorescent signals were imaged concomitant with trains of APs at 30 Hz, induced by field electrode stimulation. Imaging data were analyzed using a customized algorithm in MATLAB (MathWorks).

Channelrhodopsin—HEK293 cell measurements

All fluorescence experiments were performed on an IX70 microscope (Olympus, Tokyo, Japan) with a 40× and 1.35 N.A. oil objective (Fluar Zeiss, Germany). For dual-wavelength excitation of ChR and RCaMP1e or GCaMP3, two independent light sources were applied. For RCaMP1e and GCaMP3 excitation we used a Polychrome V unit (TILL Photonics, Planegg, Germany) set to 560 nm or 440 nm (FWHM 7 nm), respectively. ChRs were activated with a 75W-Xenon lamp (Leistungselektronik Jena, Jena, Germany) bandpass filtered either with 450 nm (FWHM 15 nm) or 560 nm (FWHM 15 nm) (AHF Analysentechnik, Tübingen, Germany). Both light sources were controlled by the Tillvision Software (TILL Photonics, Planegg, Germany). The xenon lamp was modulated via a fast shutter (Uniblitz, Rochester, NY). Fluorescence was selected via a dichroic mirror (FF01-468/553-25, Semrock, Rochester, NY) and projected onto a CCD camera (Imago, TILL Photonics, Planegg, Germany). Fluorescence was sampled every 3 s at an exposure time of 500 ms and a resolution of 320 × 20 pixels (2 × 2 binning). ΔF/F was calculated by subtracting the baseline fluorescence level (average of 20 data points before ChR activation) and subtracted from the actual fluorescence response and normalized to F₀. All fluorescence data was background corrected.

Electrophysiology was performed as previously described (Prigge et al., 2010). Briefly, patch pipettes were pulled from micro-haematocrit-tubes (Hecht-Assistant, Sondheim, Germany). Pipettes were then filled with (in mM): 110 NaCl, 10 EGTA, 2 MgCl₂, 2 CaCl₂, 2 KCl, and 10 HEPES buffer (pH 7.5). External solution contained (in mM): 140 NaCl, 2 KCl, 2 MgCl₂, 1, 2, 5, or 10 CaCl₂ and 10 HEPES (pH 7.2). pH and osmolarity were adjusted with NMG/HCl to 7.2 and Glucose to 280 or 320 mOsm, respectively. Voltage clamp measurements were performed with a HEKA EPC 7 amplifier (Heka Electronics, Darmstadt, Germany) and visualized with pClamp Software (Molecular Devices, Foster City, CA).

A HEK293 cell line constitutively expressing Ca_V3.2 and mTrek (Yao et al., 2006) was transfected one day after seeding using Fugene HD (Roche, Indianapolis, IN). Fluorescence or electrophysiology was performed one day after transfection. ATR was added into cell growth media to a final concentration of 1 μM.

C1V1 or ChR2(TC) were fused C-terminally to GCaMP3 or RCaMP1e by overlap-extention PCR. The product was cloned into pECFP-N1 (Clontech, Mountainview, CA) via NheI and XbaI digestion and subsequent ligation.

Imaging GECIs in C. elegans with channelrhodopsin photostimulation

For Ca²⁺ imaging of body wall muscles, worms were fully or partially immobilized by gluing them with surgical glue (Octyl Cyanoacrylate) onto 2% agar pads in M9 buffer. RCaMP1e and RCaMP1h fluorescence was measured on an inverted fluorescence microscope (Zeiss, Axio Observer) equipped with two high power light-emitting diodes (LEDs; 590 and 470 nm, KSL70 Rapp Optoelektronik), coupled with a beam splitter that allows simultaneous illumination at two wavelengths, for excitation of RCaMP, as well as for photostimulation of ChR2. The filter set for imaging consisted of a double band pass excitation filter 479 and 585 nm combined with a 605 nm beam splitter and a 647/57 band pass filter for emission (AHF Analysentechnik: F74-423, F38-605, F37-647). Illumination protocols were run using the Lambda SC Smart Shutter controller (Sutter Instruments). Videos were obtained with a CCD camera (Photometrics Coolsnap HQ², Roper Scientific). Calcium traces were taken by a 40× oil immersion objective (Zeiss EC Plan Neofluar 40× 1.3 NA) and analyzed using ImageJ (http://rsbweb.nih.gov/ij/index.html), generating ROIs for single or few muscle cells that did not show major movement. As a control, muscles in animals raised in the absence of ATR were analyzed, or non-transgenic muscles next to the measured transgenic muscle (animals contained extra-chromosomal arrays of RCaMP1e, and expression was thus subject to minor mosaicism). Data were calculated to obtain ΔF/F. For some experiments involving prolonged photostimulation, the fluorescence increase due to the autofluorescence was deduced by taking the fluorescence value of the first video frame of the blue stimulus, and subtracting this value from each measured fluorescence value during the photostimulus. To obtain normal fluorescence images, the same microscope was used with a HBO 100 lamp and a filter set for red fluorescent probes, e.g., Cy3. Some animals were only partially immobilized by gluing, to allow motility, and the ROIs were then traced manually for analysis, using non-ambiguous anatomical landmark features that could be followed even under fluorescence imaging. For Ca²⁺ imaging of the pharynx, heads were cut off and placed in an imaging chamber (ALA Scientific Instruments) in buffer Em D50 (140 mM NaCl, 3 mM KCl, 3 mM CaCl₂, 1 mM MgCl₂, 50 mM D-Mannitol, 10 mM HEPES, pH 7.3). Worm heads were illuminated with yellow light (80 mW/cm²) using a 590 nm LED and filterset as described above. Calcium traces were taken by a 25× Oil immersion objective (Zeiss LCI Plan-Neofluar 25 × 0.8 Imm Corr DIC) and analyzed by ImageJ, generating ROIs for terminal bulb and background. As a control, worms expressing mCherry in the pharynx (pmyo2::mCherry) were used. Data were calculated to obtain ΔF/F. Pump events were either scored by eye, judging the slight movements visible in the image stream (for RCaMP), or by analyzing kymographs of line scans through the terminal bulb (mCherry), where the lumen opening could be easily visualized.

Imaging GECIs in C. elegans AWC^on sensory neuron

The same experimental protocol was used as in previous GCaMP experiments (Akerboom et al., 2012a). All recordings are from the AWC^on sensory neuron. After 10 s of observation, odor (IAA @ 10⁻⁴ dilution) was added for 5 min. At the last 10 s of odor addition, imaging was commenced for another minute to observe the neuron's response to odor removal (the AWC^on neuron activates in response to odor removal).

Imaging RCaMP1f activity in Drosophila motor neuron terminals

RCaMP activity in zebrafish trigeminal neurons

mitfa^−/− (nacre) zebrafish were maintained under standard conditions at 28°C and a 14:10 h light:dark cycle. Larvae were raised in E3 embryo medium (5 mM NaCl, 0.17 mM KCl, 0.33 mM CaCl₂ and 0.33 mM MgSO₄). Embryos were injected at the 1 cell stage with 20 ng/μl plasmid DNA encoding RCaMP under the control the pan-neuronal elavl3/HuC promoter (elavl3:RCaMP), and 40 ng/μL Tol2 transposase mRNA diluted in E3 medium with 0.025% Phenol Red. Forty-eight hour post-fertilization embryos showing expression in isolated trigeminal neurons were treated with 1 mg/mL bath-applied alpha-cobratoxin for 15 min to block tail movements. They were mounted in 1.5% low melting temperature agarose, and imaged using an microscope with a 40×/0.8NA objective lens (Olympus) and a CCD camera (Hamamatsu Orca). 1 ms field stimuli were applied using two mesh electrodes in the bath and a stimulator (Grass SD9). Stimulation voltage was calibrated to the threshold to elicit a behavioral response to one pulse in the set up. One spike in a trigeminal neuron has previously been shown to be sufficient to elicit tail movement (Douglass et al., 2008). Software written in LabView (National Instruments) was used for image acquisition and stimulus control. ROI were selected manually, and data was analyzed using MATLAB (MathWorks).

Conflict of interest statement

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.