Ratiometric genetically encoded voltage indicators (GEVIs) are needed to remove unwanted artifacts of membrane motion so that changes in membrane potential can be discriminated from motion-induced changes in fluorescence. Membrane motion occurs when cells migrate or change shape, for example when cultured cardiomyocytes contract. Membrane motion is also an inherent feature of contractile tissues such as skeletal muscles, the gastrointestinal tract, or the heart, and sample movement relative to the camera is an omnipresent artifact during microscopy in live animals due to breathing, hemodynamic motions, or animal movement1.
A properly designed ratiometric GEVI would be able to provide a readout corrected for changes in indicator abundance caused by motion if two criteria are fulfilled. First, a reference fluorophore’s brightness must be either independent or inversely correlated to that of a voltage-modulated fluorophore of different wavelengths in the same molecule, so that the ratio of emissions at the two wavelengths at any point in space can be related to voltage independently of indicator abundance at that point. Second, both the fluorophores should be excitable by the same wavelength of light, as this would enable simultaneous recording of the two fluorophores in separate channels, thereby increasing speed, simplicity, and precision. GEVIs that utilize a change in FRET efficiency between two fluorescent proteins meet these requirements, but engineering large responses in FRET-based GEVIs has proven to be difficult2.
Another solution may be to use a voltage-independent fluorophore with a large Stokes shift that can be excited at the same wavelength as the voltage sensitive-fluorophore while keeping their emissions easily separable. Many recently developed GEVIs with high voltage responsivity function via brightness modulation of a single fluorophore. For example, ASAP-family GEVIs utilize a single circularly permuted green fluorescent protein (GFP) domain3,4,5,6. We sought to create a high-performance ratiometric GEVI by fusing a voltage-independent large-Stokes-shift (LSS) red fluorescent protein (RFP) to ASAP3.
While many LSS RFPs have been developed, they suffer from residual oligomerization7,8 or poor brightness9. To create a ratiometric GEVI, we initially fused ASAP-family GEVIs at their C-terminus to the cyan-excitable red fluorescent protein CyRFP1 or its more monomeric variant mCyRFP1. The cyan excitation of CyRFP1 and mCyRFP1 allows them to be co-excited efficiently with ASAP1 by 488-nm light (or 950-nm light in two-photon mode). We also tested mRuby3, FusionRed, and mCherry, which have been fused to a variety of proteins without causing toxicity10,11. Disappointingly, when expressed in neurons, all five fusions accumulated in the cell body and dendrites in structures too large to represent single vesicles, suggesting protein aggregation (Supplementary Fig. 1). These accumulations were more visible in the red channel than the green channel, likely because the RFPs are located in the pH-neutral cytosol while the GFP moiety of ASAP is located in the acidic lumen of the secretory pathway. The resulting differences in green/red ratios in the cell prevents using the red channel to correct for movement artifacts. In addition, GEVI molecules trapped in intracellular accumulations will not respond to voltage changes at the membrane. As the CyRFP1 fusion appeared to be better localized to the membrane compared to other RFP fusions, we chose CyRFP1 as a template for creating a more monomeric RFP.
mCyRFP1 had shown increased monomericity relative to CyRFP1 in vitro due to a surface A161K mutation, which breaks a hydrophobic contact with Tyr-148 across the cross-dimer interface8 (Fig. 1a). We hypothesized that A161K may not be the optimal mutation for breaking this hydrophobic contact, e.g. if retaining tyrosine at position 148 is not ideal for folding or hydration in the monomeric state. Indeed, performing double saturation mutagenesis and screening for brightness yielded a Y148T A161V mutant that was more monomeric in vitro (Fig. 1b). We designated this protein mCyRFP3, since the name mCyRFP2 was already assigned to a different mCyRFP1 variant12. mCyRFP3 exhibited absorbance and excitation centered near 500 nm and peak emission of 586 nm, slightly red-shifted from mCyRFP1 (Fig. 1c). mCyRFP3 showed improved fluorescence quantum yield of 0.72 and an extinction coefficient of 38 mM−1 cm−1 (Table 1), making its molar brightness more than 50% brighter than its predecessor mCyRFP1. mCyRFP3 fluorescence demonstrated a pKa of 4.1 and was extremely stable in the physiological pH range (Fig. 1d). mCyRFP3 photobleached with a normalized half-time of 75 s (Fig. 1e), and exhibited a maturation half-time of 12.5 min (Fig. 1f).
Engineering and characterization of mCyRFP3. Engineering of mCyRFP3. (a) Crystal structure of dimeric predecessor CyOFP1 displaying interacting residues Y148 and A161. Image was generated in Schrödinger MacPyMol 1.8. (b) Size-exclusion chromatography of EGFP, tdTomato, CyRFP1, and mCyRFP3. EGFP was used as a monomeric standard, while tdTomato was used as a dimeric standard. (c) Excitation and emission spectra of mCyRFP3 compared to its parent mCyRFP1. (d) pH dependence of mCyRFP3 fluorescence, demonstrating a pKa of 4.1. Error bars are s.em.m of triplicate measurements. (e) Photobleaching kinetics of purified cyan- excitable red fluorescent proteins under illumination by a 120-W metal-halide arc lamp through a 490/20-nm excitation filter. The time-axis was adjusted for each fluorophore to simulate excitation conditions producing 1000 photons per s per molecule. Lighter shading on CyRFP1, mCyRFP1, and mCyRFP3 lines represents standard deviation of five measurements. (f) Maturation kinetics of mCyRFP3 demonstrating a half-life of t = 12.5 min. Graphs in (a) through (f) were generated in Microsoft Excel for Mac 16.
We next characterized mCyRFP3 performance in mammalian cells. Fusions to various subcellular proteins were properly localized (Fig. 2a) and a histone H2B fusion enabled observation of mitosis (Fig. 2b). In the organized smooth endoplasmic reticulum (OSER) assay, mCyRFP3 showed a similar score (87% whorl-free cells) as mCyRFP1 (Fig. 2c). While purified proteins in vitro provide extinction coefficients and quantum yield parameters as an objective measure of a mature FP’s per-molecule brightness, an FP’s effective brightness in cells is influenced by its maturation efficiency and stability. We compared the brightness of cells expressing mCyRFP3, mCyRFP1, or CyRFP1 using a bicistronic construct co-expressing EGFP, where the EGFP fluorescence was used to normalize for differences in transfection and mRNA levels. In HEK293A and HeLa cells, mCyRFP3 was significantly brighter than mCyRFP1 (Fig. 2d).
Performance of mCyRFP3 in mammalian cells. (a) HeLa cells expressing mCyRFP3 fused various subcellular proteins. For each fusion, the original of the fusion partner and its normal subcellular location are indicated in parentheses. (i) mCyRFP3-2aa-tubulin (human, microtubules), (ii) mCyRFP3-7aa-actin (human, actin cytoskeleton), (iii) Calnexin-14aa-mCyRFP3 (human, endoplasmic reticulum), (iv) mannosidaseII-10aa-mCyRFP3 (mouse, Golgi complex), (v) mCyRFP3-10aa-lamin B1 (human, nuclear envelope) (vi) PDHA-10aa-mCyRFP3 (human, mitochondrial pyruvate dehydrogenase), (vii) connexin43-7aa-mCyRFP3 (rat, cell–cell adhesion junctions), (viii) paxillin-22aa-mCyRFP3 (chicken, focal adhesions), (ix). mCyRFP3-2aa-CAAX. Scale bar, 10 µm. (b) mCyRFP3-10aa-H2B (human, nucleosomes) in (i) interphase, (ii) prophase, (iii) metaphase, (iv) anaphase. Spinning-disk confocal image stacks were acquired using Improvision Volocity 6.0 and flattened and scaled in NIH Fiji 2.1. (c) Performance of mCyRFP3 compared to mCyRFP1, mEGFP and mCardinal on CytERM monomericity assay. Error bars represent standard deviation of measurements from > 150 cells in each of three separate experiments. (d) mCyRFP3 characterization in mammalian cells. Brightness comparison of mCyRFP3 in HEK293A and HeLa cells expressing a bicistronic construct EGFP-P2A-RFP where RFP = CyRFP1, mCyRFP1, and mCyRFP3. The red fluorescence generated from each of the RFPs were normalized to the fluorescence of EGFP to normalize for expression and were charted relative to the value of CyRFP1. Excitation was performed with a 480/10-nm filter and emission was collected from 580 to 800 nm. Integrated red emission relative to the integrated GFP emission is shown as mean ± s.e.m. of 6–8 biological replicates. Graphing and two-tailed t tests with Bonferroni correction were performed in Microsoft Excel for Mac 16.
We were curious if optimization of amino acids 148 and 161 could also improve the performance of another RFP derived from eqFP578, the natural tetrameric predecessor to mCyRFP1. Thus, we also performed combinatorial saturation mutagenesis at these positions in the mMaroon1 far-red fluorescent protein, which is descended from eqFP578 via mNeptune113. Indeed simultaneous Y148T A161G mutations led to the development of a more monomeric and brighter variant, mMaroon2 (Supplementary Note, Supplementary Figs. 2, 3, Supplementary Table 1).
We recently developed ASAP3, a GEVI with improved responsivity and brightness compared to previous ASAP-family GEVIs5. ASAP3 consists of a 4-helix transmembrane voltage-sensing domain (VSD) bearing a circularly permuted GFP domain inserted into an extracellular loop. Voltage-dependent motions of the VSD are transduced into changes in GFP fluorescence under blue light excitation. An independent comparison found that ASAP3 exhibits the largest power-normalized signal-to-noise ratios for both subthreshold voltage dynamics and action potentials (APs) among GEVIs developed to date14. We thus explored whether mCyRFP3 could serve as a non-perturbing voltage-independent label for ASAP3.
We tested fusions of ASAP3 to mCyRFP3, mCyRFP1, mRuby3, which we had previously tested with ASAP1, and the recently developed mScarlet, which is non-perturbing in fusions with a variety of cellular proteins15. ASAP3-mCyRFP3 (ASAP3-R3) showed the clearest membrane localization in dendrites and the smallest amount of cell body accumulations, followed by mCyRFP1, then mRuby3, and finally mScarlet, which completely lacked membrane expression and whose cell body accumulations were especially large (Fig. 3a). We also compared ASAP3-R3 to ASAP3 fused to mCyRFP2, which was developed during the course of this work. ASAP3-mCyRFP2 showed more accumulation of red fluorescence in the cell body and less ASAP3 signal at the membrane, consistent with trapping in secretory vesicles (Supplementary Fig. 4). In neurons expressing ASAP3-R3, high-magnification confocal microscopy confirmed tight correspondence between the green signal of ASAP3 and the red signal of mCyRFP3 (Fig. 3b).
Characterization of ASAP3-mCyRFP3 (ASAP3-R3). (a) Representative images of ASAP3 fusions to various RFPs expressed in primary rat hippocampal neurons acquired on a spinning-disk confocal microscope using Improvision Volocity 6.0 and processed in NIH Fiji 2.1. (b) Laser-scanning confocal images of ASAP3-R3 in a rat hippocampal neuron. Images were acquired with Zeiss Zen Blue software and processed in Fiji 2.1. (c) Above, normalized excitation spectra of mCyRFP3 and mEGFP, showing a single excitation wavelength can be used. Below, normalized emission spectra of mCyRFP3 and mEGFP, showing their emissions are separable (d) Steady-state voltage responsivity of ASAP3-R3. Error bars represent standard error of the mean (n = 8 HEK293 cells). Graphs were made in Microsoft Excel for Mac 16.
Conveniently, both fluorophores of ASAP3-R3 can be excited by the same wavelength, while their emissions are easily separable (Fig. 3c). The steady-state voltage responsivity of ASAP3-R3 in the green channel upon depolarization from − 70 mV to 30 mV, measured as the change in fluorescence intensity relative to that at the–70 mV baseline condition (∆F/F0), was − 41 ± 1% (mean ± standard error of the mean in eight HEK293 cells) (Fig. 3d). The slightly lower response amplitude compared to ASAP3 (− 51 ± 1%)5 may be due to minor contamination by voltage-insensitive mCyRFP3 fluorescence in the green emission channel (Fig. 3c).
Cardiomyocytes (CMs) derived from induced pluripotent stem cells (iPSCs) enable the study of patient-specific mechanisms in cardiovascular diseases16. A key functional readout of iPSC-CMs is the pattern of spontaneous cardiac APs, which drive contraction and whose appearance indicates the maturation of ion homeostasis and channel properties. Recording APs is also useful for studying signaling pathways leading to lineage specification of CM subtypes17. The gold standard for characterizing APs in iPSC-CMs is patch-clamp electrophysiology, but this method is low-throughput and destructive. An alternative method for AP measurements is to use fluorescent GEVIs, but GEVIs preferentially would be ratiometric so that changes in membrane potential are self-normalized and can be discriminated from motion-induced changes in fluorescence.
We tested the utility of ASAP3-R3 in reporting APs of iPSC-CMs. ASAP3-R3 fluorescence from beating iPSC-CMs was excited using blue light, then green and red channels were recorded by a single CMOS camera using an emission beam splitter. We found some apparently non-motile iPSC-CMs were firing APs, as ASAP3-R3 showed rhythmic decreases in fluorescence in the green channel and no change in the red channel (Fig. 4a). At the peak of the action potential (most depolarized timepoint), ASAP3-R3 showed a change in the green/red ratio relative to its maximum (most polarized) value (∆R/Rmax) of − 46.6 ± 4.8%. This was similar to the change in green fluorescence relative to maximum fluorescence (∆F/Fmax) of − 47.8 ± 5.21% for ASAP3 in spontaneously beating iPSC-CMs (Fig. 4b), demonstrating the ASAP3-R3 GEVI is fully functional. Photobleaching was not observed in the recording period (Fig. 4b).
ASAP3-R3 performance in cardiomyocytes. (a) Representative induced pluripotent stem cell derived cardiomyocytes (iPSC-CMs) expressing ASAP3 or ASAP3-R3. Epifluorescence images were acquired using Hamamatsu HCImage software and color channels assigned in NIH Fiji 2.1. (b) Representative single-trial ∆R/R or ∆F/F responses to spontaneous cardiac APs of cells expressing ASAP3, expressing ASAP3-R3, or expressing ASAP3-R3 and treated with isoproterenol (ISO). Graphs were generated in Microsoft Excel for Mac 16. (c) Characterization of traces obtained from ASAP3 (n = 6), ASAP3-R3 (n = 10), or ASAP3-R3 + ISO (n = 7). Graphing and statistical analyses were performed in GraphPad Prism 7.
We also tested if ASAP3-R3 can faithfully report drug induced changes of APs in iPSC-CMs. Isoproterenol (ISO) activates β-adrenergic receptor signaling in CMs to increase heart rate. ASAP3-R3 showed that ISO treatment of iPSC-CMs also increased the spontaneous contraction rate, from ~ 30 beats per minute (bpm) to > 60 bpm (Fig. 4b,c). The rise rate, which should correlate with the up-stroke velocity in patch clamp measurements, was also significantly increased (Fig. 4c). As a result, the AP duration to 90% repolarization (APD90) was significantly shorter in the ISO treated group (Fig. 4c).
We next demonstrated how ASAP3-R3 can be used to correct for motion artifacts in moving cells. Different edges of a contracting iPSC-CM can exhibited disparate responses in green intensity (Fig. 5a,b). On one edge, green signal fluctuations were larger than those in non-contracting CMs (Fig. 5a). On another edge, the green signal was inverted from the expected response and did not resemble AP waveforms (Fig. 5b). The voltage-insensitive red channel showed different rhythmic fluctuations at the two edges, indicating different patterns of membrane movement. Thus, the disparate green fluorescence responses from the same iPSC-CM were due to different contributions of membrane movement to green fluorescence. Indeed, when the green/red ratio was tracked instead, different portions of the membrane responded in the same direction and magnitude as in non-contracting iPSC-CMs (Fig. 5a,b). The ratio changes in different membrane regions also resembled each other in waveform shape even though the individual channels did not (Fig. 5a,b).
ASAP3-R3 accurately reports voltage in motile cells and detects voltage fluctuations during the cell cycle. (a) Top, one segment of membrane in a beating CM was analyzed. Middle, raw traces of green and red signals in the selected region. Red trace indicates significant cell movement in and out of selected region detected by the voltage-independent mCyRFP. Bottom, relative change in green fluorescence alone (− ∆F/F) and the relative change in the green/red ratio (− ∆R/R). While green intensity changes are larger than those observed previously in non-contracting CMs due to these motion artifacts, ratio changes are similar in magnitude to those observed previously. (b) Similar analysis on a different membrane segment of the same cell. Here, the green fluorescence changes are opposite in direction from that expected due to movement, which is detected in the red channel. While green intensity traces do not resemble AP waveforms, ratio traces are indistinguishable from AP waveforms in non-motile CMs and similar in shape to those in (a). Epifluorescence images were acquired using Hamamatsu HCImage software and color channels assigned and overlaid in NIH Fiji 2.1. Graphs were generated in Microsoft Excel for Mac 16. (c) Top, images of the ASAP3-R3 green/red ratio at different time points relative to cytokinesis in a HeLa cell. Middle, individual green/red ratio time courses of three HeLa cells during cell division. The cell in the images is represented by the blue trace. Cytokinesis occurs immediately before the 0-h time point. Below, mean green/red ratios of the three cells. Error bars represent standard deviation. Epifluorescence images were acquired using Hamamatsu HCImage software and pseudocoloring with a ratiometric lookup table was performed in NIH Fiji 2.1. Graphs were generated in Microsoft Excel for Mac 16.
Finally, using ASAP3-R3, we demonstrated regulation of the membrane potential during the cell cycle. Potassium and chloride channel activities are regulated by cell cycle phase18, and electrode measurements have found fluctations in the transmembrane potential during cell cycle progression19,20. Interestingly, manipulations of channel function or extracellular ion concentrations suggest a role for transmembrane voltage in the timing of cell cycle transitions18. However, direct electrical measurements across the cell cycle have only been reported as single-timepoint measurements in non-motile cell types. We asked whether we could observe transmembrane voltage in a motile cell type throughout an entire cell cycle. HeLa cells demonstrate pronounced cellular motility, which would make repeated membrane voltage measurements of a single cell throughout the cell cycle difficult. The morphological dynamism of HeLa cells would also prevent accurate voltage readings from a non-ratiometric GEVI.
We thus expressed ASAP3-R3 in HeLa cells and visualized the green/red ratio over time. We observed a ~ 35% change in green/red ratios throughout the cell cycle, with the lowest ratio value (representing depolarization) in mitosis prior to cytokinesis (Fig. 5c). The highest green/red ratios (representing hyperpolarization) occurs 6–9 h following cytokinesis, which corresponds to S-phase in HeLa cells13. The timing of these voltage changes are consistent with previous reports using intracellular electrodes on non-motile cell types19,20. Thus, using ASAP3-R3, we were able to detect membrane voltage changes throughout the cell cycle in moving cells, finding that HeLa cells are most depolarized during mitosis and most polarized during S-phase.
Our results demonstrate the utility of structure-guided multi-site saturation mutagenesis for fluorescent protein improvement. Obtaining mCyRFP3 would have been exceedingly difficult by random mutagenesis. mCyRFP3 differs by mCyRFP1 by three nucleotides within a 700 bp gene, so in a library bearing 3 random mutations per clone without duplicates, this combination would have occurred once in 700 choose 3 × 43, or 3.6 × 109 mutants. In reality, the library would need to be at least an order of magnitude larger to account for Poisson distributions of mutation combinations. If the ideal combination of amino acids had required six nucleotide changes, then it would have arisen in a perfectly distributed library of six-nucleotide mutants once per 6.6 × 1017 clones. By identifying two sites on the crystal structure, we only needed to screen for 1024 possible combinations using an NNS codon at each site (32 × 32). Thus, hypothesis-driven structure guided mutagenesis can identify beneficial multi-site mutants that would have been unobtainable using random mutagenesis. Similar methods have been used to produce functional improvements in a variety of RFPs and in ASAP3 itself5,7,10,13,21,22.
Indeed, our observation that coordinated mutations at positions 148 and 161 produced more monomeric and brighter variants of multiple RFPs suggests a general strategy of monomerization in which positions interacting across a dimer interface are mutated to all possible combinations. While this may seem obvious in retrospect, this does not appear to be a standard step in the monomerization workflow for fluorescent proteins. mCyRFP1 is derived from TagRFP, which in turn is a weakly dimeric mutant of the natural tetrameric RFP eqFP578. To our knowledge, positions 148 and 161, which are distant from each other on the surface of a single monomer, have not been previously mutated in combination and to saturation in any RFPs derived from TagRFP.
As mCyRFP3 is more monomeric in vitro than mCyRFP1, it is not surprising that it would be less perturbing to ASAP3 membrane localization. However, in the OSER assay, fusions of mCyRFP1 and mCyRFP3 to CytERM showed similar propensities to form ER whorls. In addition, although FusionRed and mScarlet are less perturbing in the OSER assay than mCyRFP3, they performed worse when fused to ASAP3. These observations demonstrate that the OSER assay only tests the performance of fluorescent proteins fused to one cellular protein, and may not necessarily predict performance of fusions with any other protein. One possible complicating factor in the OSER assay is protein expression level; fusions with less stable fluorescent proteins may accumulate to lower concentrations and thereby create whorls in a smaller percentage of cells.
In summary, we have created a cyan-excitable red fluorescent protein, mCyRFP3, with improved brightness and monomericity. Among RFPs, mCyRFP3 performs especially well as a fusion tag for the ASAP3 voltage indicator. The resulting ASAP3-R3 enables voltage imaging with an internal reference channel for motion correction, as we demonstrate by imaging transmembrane voltage changes during cardiomyocyte contraction and during the cell cycle. Compared to voltage indicator dyes, ASAP3-R3 has the advantage of being genetically encoded, allowing for stable expression and targeting to specific cell types. We expect that ASAP3-R3 will also be useful for imaging voltage in the brains of living animals, where the mCyRFP3 reference channel can allow for correction of motion artifacts in the voltage-modulated ASAP3 signal.
