Enhanced genetically encoded voltage indicators advance their applications in neuroscience

Abstract

Genetically encoded voltage indicators report membrane voltage with high spatiotemporal resolution. Extensive recent efforts to improve the GEVIs’ brightness, sensitivity, and kinetics have greatly increased the GEVIs’ signal-to-noise performance over ten-fold and lowered their response time to the sub-millisecond regime. Such capabilities have broadened the GEVIs’ ability to measure membrane voltage of neural populations at cellular resolution in vitro and in vivo, all at high speeds. The GEVIs’ high voltage fidelity and fast response have revealed novel physiological phenomena in multiple neuroscientific applications. Such applications portend future targeted studies of voltage activity that take advantage of the GEVIs’ ability to report rapid dynamics from genetically-targeted neural populations.

1. Introduction

The neuroscience community has long pursued the ability to analyze neural circuits with cell-type specificity and millisecond precision. These demands are partially met by staple neural recording techniques such as multi-electrode recordings and optical recording of genetically encoded calcium indicators. Electrical methods can record neural activity with high temporal resolution, but lack genetic specificity. Genetically encoded calcium indicators impart genetic specificity, but lack the kinetics required to record neural activity with high temporal resolution. Genetically encoded voltage indicators (GEVIs) are a promising technology that combines genetic specificity, voltage accuracy, and speed. When compared to electrical recordings or optical recordings of calcium activity, voltage imaging of GEVIs would be able to uniquely answer a range of neuroscientific questions that specifically examine the millisecond subthreshold or super-threshold voltage dynamics from genetically identifiable neurons or neural populations. Recent development of GEVIs has led to unprecedented recording fidelity in vitro and greatly enhanced neuroscience studies in vivo. The goal of this review is to discuss three outcomes of this development: (1) the upward trajectory of performance from multiple state-of-the-art classes of sensors, (2) the screening and assessment strategies of sensor development, and (3) the capabilities of these sensors to address neuroscientific questions in various systems and at various scales.

2. A broad diversity of GEVIs sense neuronal voltage

We outline the recent development of multiple categories of GEVIs, including fluorescent protein (FP)-based GEVIs, circularly permuted FP-based GEVIs, rhodopsin-based GEVIs, FRET-opsin GEVIs, and chemogenetic FRET GEVIs (Fig. 1).

Figure 1 -. Multiple GEVI configurations can probe membrane potential.

(a) FP-based GEVI (VSFP-butterfly). (b) FP-based GEVI (Arclight). (c) Circularly-permuted FP-based GEVIs (ASAP). (d) Rhodopsin-based GEVI, (Arch). (e) FRET-opsin GEVIs (Ace2N-mNeon). (f) Chemogenetic FRET GEVIs (Voltron).

2.1. FP-based GEVIs

FP-based GEVIs combine the bright readout of engineered fluorescent proteins with conformational or electronic changes in voltage sensing domains (VSDs). Because FP-based GEVIs employ intact FPs, these sensors retain the high brightness and photostability of their FPs components; these sensors often perform with low relative shot-noise even during millisecond timescale imaging. Initial FP-based GEVIs in the VSFP series combined the Ciona intestinalis voltage-sensing phosphatase (Ci-VSP), which had a voltage-sensitive conformation, and a CFP and YFP fluorescence resonance energy transfer (FRET) pair [1,2] (Fig. 1a). More recent VSFP variations further tuned the FRET configurations and VSD. First, variants such as VSFP2.4, Mermaid, VSFP-CR, and NIR-GEVI employed additional FPs in the visible spectrum [3-5] or near-infrared spectrum (Monakhov, et al., bioRxiv doi:10.1101/536359) as elements in the FRET pair. The high FRET efficiencies of these sensors improved the voltage sensitivity of the VSFP sensors to ΔR/R ~ 5-10% per 100 mV [4]. Next, mutating key residues in the VSD shifted the voltage response of Ci-VSP toward the physiological resting membrane potential; these sensors responded to voltage depolarizations with ~5 millisecond kinetics [6]. VSFP-Butterfly combined VSD tuning and red-shifted FRET pairs by flanking the FP donor and acceptor on different sides of the VSD; it reported up to 200 Hz voltage fluctuations [7,8]. Interactions between Ci-VSP and a single fused FP also produced a voltage response [5]. The first series of VSFP3 sensors fused a palette of FPs to Ci-VSP to create sensors with millisecond-timescale sensitivity but moderate response [9,10]. New pairings of FPs and Ci-VSP produced ArcLight [11] and Bongwoori [12] (Fig. 1b). Both Arclight and Bongwoori responded to voltage with ~30% ΔF/F per 100 mV and had on-kinetics of ~10 milliseconds. This fast response resolved spikes within 60 Hz spike trains in vitro.

2.2. Circularly-permuted FP-based GEVIs

Circularly-permuted FPs (cpFPs) have historically coupled to sensing domains to report neural activity [13,14]. Conformational changes in the sensing domain pull on the FP’s β-barrel cage, thereby modulating the local environment of the FP chromophore and the FP’s fluorescence intensity. cpFPs have high allosteric coupling to VSDs that change conformation; GEVIs based on cpFPs thus typically report the modulation of their VSDs with high voltage sensitivity and fast response. GEVIs initially employed the same concept by fusing one terminal of a circularly permuted eGFP or mKate to Ci-VSP [15,16]. These sensors had moderate responses of ~1% ΔF/F per 100 mV. ASAP1 advanced the cpFP GEVI design by inserting the cpFP into the middle of a homolog of Ci-VSP to increase the coupling between the FP and the VSD [17] (Fig. 1c). ASAP1 produced ~20% ΔF/F per 100 mV, and on-kinetics of ~2 milliseconds. Additional mutations on the linker between the cpFP and VSD generated a second generation of ASAP sensors that had either higher voltage sensitivity (ASAP2s) or faster kinetics (ASAP2f); these sensors reported voltage transients from live flies or mouse brain slices [18,19]. FlicR1, a fusion of cp-mApple and Ci-VSP, further extended the cpFP sensors into the red spectrum [20]. FlicR1 produced ~6% ΔF/F per 100 mV and reported spikes in vitro.

2.3. Rhodopsin-based GEVIs

A variety of microbial rhodopsins re-purposed from the optogenetic toolbox [21] have also served as GEVIs. Membrane potentials modulate the local electrochemical potential of protons on the chromophore of these rhodopsins; in turn, the chromophore increase the protein’s absorption and fluorescence (Fig. 1d). Electrochemical modulation of the rhodopsin VSD supports some of the fastest and largest voltage responses among all GEVIs. However, because the native function of rhodopsins is converting light into photocurrent, the fluorescence quantum yield of rhodopsin sensors lags far behind that of FP-based sensors. Initial engineering of rhodopsin-based GEVIs focused on Archaerhodopsin (Arch); these efforts eliminated photocurrent and took advantage of strategies that localized the protein to the membrane of neurons [21-23]; the Arch-D95N variant produced 30% ΔF/F per 100 mV [24]. The subsequent generations of Arch-based GEVIs (Arch-EEQ/EEN [25], Archer [26], and QuasAr1/2 [27], QuasAr3 [28], and Archon1 [29]), incrementally engineered superior voltage sensitivity, faster kinetics, and higher brightness. The series culminated in Archon1, which had high voltage sensitivity (~80% ΔF/F per 100 mV) and rapid response kinetics (<1 millisecond). This sensor successfully tracked high-speed spiking and subthreshold activity of multiple neurons simultaneously in vivo. Still, Archon1 had low quantum yield compared to that of FP-based sensors.

2.4. FRET-opsin GEVIs

One promising strategy coupled bright FP readouts with the fast voltage sensitivity of rhodopsin VSDs via electronic FRET (eFRET): the FP served as the FRET donor and fluorescent readout while the rhodopsin VSD served as the FRET acceptor. Voltage depolarization increased rhodopsin absorption; depolarization thus increased FRET between the rhodopsin and FP, and reduced the emission of the FP in FRET-opsin sensors. The rhodopsin-FP fusion trades off some voltage response for brightness when compared to rhodopsin GEVIs; the combination of moderate voltage response but high brightness supports a higher spike detection fidelity than that of rhodopsin GEVIs at equal excitation powers. The first generation of FRET-opsin GEVIs fused a variety of green, yellow, orange, and red FPs to either the rhodopsin from Leptosphaeria maculans (Mac) [30] or an existing Arch sensor (QuasAr2) [31]. These sensors had comparable voltage sensitivity to Ci-VSP-based sensors but had significantly faster kinetics; the Mac- and Arch-based sensors had on-kinetics of ~3-5 milliseconds. The second generation of FRET-opsin sensors further improved kinetics by using the rhodopsin from Acetabularia acetabulum (Ace) as the VSD, which had a sub-millisecond voltage response (Fig 1e). Fusions of the Ace VSD with either green or red FPs produced a broad spectrum of fast FRET-opsin GEVIs such as Ace2N-mNeon [32] and VARNAM (AcemRuby3) [33]. These sensors enabled recordings of fast spike trains from neurons in awake mice and flies.

2.5. Chemogenetic FRET GEVIs

Much like the FRET-opsin fusions described above, fusions of protein VSDs and chemical dyes can also sense voltage. These chemogenetic sensors employ chemical elements as either VSDs or fluorescent readouts. When the chemical component serves as the VSD, it could supply voltage sensitivity comparable to protein VSDs. When the chemical component serves as the fluorescent readout, it often supplies a brighter and more photostable source of emission when compared to FPs. Chemical VSDs have a long history of reporting voltage. For example, the hybrid Voltage Sensors (hVoS) coupled the dipicrylamine (DPA) VSD to a membrane-localized FP via FRET [34-36]. The voltage-dependent motion of DPA within a neuron’s membrane modulated the distance between the DPA and FP, thus producing voltage-dependent FRET efficiency and fluorescence. More recently developed chemogenetic sensors, such as Voltron [37] and FlareFRET [38], employ the eFRET configuration; instead of a FP donor, chemogenetic eFRET sensors employ a bright dye as the donor fluorescent readout (Fig. 1f). Compared to their fully genetically encoded counterparts, the dye-based eFRET sensors had a ~3-4 times higher brightness and had voltage sensitivity of ~20-30% ΔF/F per 100 mV. Using its high brightness, Voltron imaged action potentials from large populations of neurons in live mice and zebrafish.

3. Large-scale screens have iteratively improved GEVIs

Two complementary methods driving advances in the GEVIs’ performance are site-directed mutagenesis and directed evolution. Site-directed mutagenesis focuses on critical residues in a protein’s structure to increase the speed and amplitude of the voltage response or enhance interactions between VSDs and FPs. Common targets for site-directed mutagenesis include key residues of voltage-sensing domains and linkers between the components of fusion proteins. For example, site-directed mutagenesis of the proton-conducting residues in Arch improved the sensor’s sensitivity, brightness, and kinetics [24,25,27]. In addition, site-directed mutagenesis identified the type and number of amino acids between the rhodopsin VSD and fluorescent donor in the FRET-opsin configuration [30-32].

Going beyond screening key individual residues, directed evolution stochastically mutagenized the complete sensor gene and selected variants with superior brightness or localization to the cell membrane. For example, the initial evolution of QuasAr1/2 and FlicR1 employed 3-5 rounds of directed evolution based on a manual selection of >10⁴ colonies for brightness [20,27]. More recently, the development of Archon1 [29] expanded automated screening to test >10⁶ variants in HEK cells by using robotic selection of cells. Screening in HEK cells also enabled the immediate identification of variants with superior membrane localization in mammalian cells.

4. SNR metrics properly assess GEVI performance

Large-scale development of GEVIs requires an accurate assessment of GEVI performance. While GEVI performance varies between preparations, key metrics such as signal-to-noise ratio (SNR) and d’ can predict the theoretical limits of sensor performance in each preparation. SNR is defined as the ratio between a GEVI’s peak change in fluorescence (ΔF) and the standard deviation of its baseline fluorescence ($\sqrt{F_0}$, for shot-noise limited measurements). $\mathrm{SNR}=\Delta F∕F\times \sqrt{F_0}$ thus depends both on voltage sensitivity and brightness. The spike discriminability metric, d’, considers not only the sensor’s sensitivity and brightness, but also the sensor’s kinetics and the expected waveform in response to an action potential [39]. Intuitively, sensors with slow kinetics can accumulate more emissions over their long response transients and enable high fidelity discrimination, but at the cost of temporal resolution. Overall, d’ provides a comprehensive measure of spike detection fidelity; linear increases in d’ translates to orders of magnitude lower false positive and false negative spike detection error rates. Table 1 summarizes the properties of recently developed GEVIs and their corresponding SNR or d’. Most sensors function at high SNR or d’ (>10), which corresponds to less than one spike detection error per hour of recording; this suggests that robust voltage recordings are feasible when discounting photobleaching. Because linear increases in SNR metrics lead to exponential decreases in spike detection errors, additional incremental increases in sensor performance will likely further improve recordings at fine spatial resolutions or large scales in practical neuroscience settings. These sensors’ high SNR and fast kinetics suggest that robust voltage recordings are feasible when discounting photobleaching over seconds of recording. Given their combination of moderate photobleaching rates (Table 1) and high SNR, GEVIs can accurately detect spikes within optical recordings for experiments lasting tens of seconds to several minutes.

Table 1 –

Recently developed GEVIs with high SNR are readily usable for in vivo imaging.

	ΔF/F (%)		Excitation intensity (mW/mm2)	Excitation wavelength (nm)	On-kinetics (ms)		Off-kinetics (ms)		SNR	d’	Photobleaching time constant (s)
	per 100 mV (−70 to +30 mV)	per action potential			τfast	τslow	τfast	τslow
ASAP2s [19]	−39	−12	24	460	5.2	63	24	106	37	54	100
FlicR1 [20]	6.6	2.6	100	561	3.0	41	2.8	18	6	9.5**	150
QuasAr2 [27]	90	48	8000	640	1.2	11.8	1.0	15.9	70	100**	1020
Archon1 [29]	81	30	800	637	0.61	8.1	1.1	13	36	57**	10,000 ***
ArcLight [11]	−35	−3.2	30	473	14	78	27	72	9	3**	90
QuasAr2-mCitrine [31]	−13	−9.7	30	532	4.8	21	3.1	21	8.8	12.4	No data available
Ace2N-mNeon [32]	−18	−12	15	505	0.36	4.2	0.42	5.2	57	100	143 ***
VARNAM [33]	−12	−8.4	15	565	0.88	5.2	0.8	4.7	36	53**	256
Voltron525 [37]	−23	−12	20	510	0.64	4.1	0.78	3.9	99*	140**	329

^*estimated assuming shot-noise limited recordings;

^**estimated from kinetics;

^***estimated from reported photobleaching rate.

5. GEVIs enable in vivo neuroscientific investigations at various scales

The recent advances in GEVIs have enabled researchers to probe broad classes of novel questions in neuroscience. The GEVIs’ superior kinetics, brightness, and sensitivity can capture fast voltage dynamics with single-spike resolution between distinct subsets of neurons or in subcellular compartments in vivo. In recent years, multiple groups have taken advantage of these benefits to make new discoveries in neuroscience at the sub-cellular, cellular, and mesoscopic scales.

5.1. Sub-cellular recordings

The rapid response and high voltage-response fidelity of GEVIs have enabled novel studies of voltage propagation with micrometer spatial resolution. For example, Ace2N-mNeon captured the dynamics of voltage propagation throughout different portions of a single neuron’s dendritic and axonal arbor in awake flies with sub-millisecond precision [32]. Such a study could reveal the function of regions within the arbor inaccessible to traditional electrophysiology. Targeted voltage imaging could thus identify whether specific sub-cellular regions serve as efferent or afferent signaling pathways, and how the rapid voltage dynamics of these signals differ in these pathways. Similarly, the photoactivatable NovArch sensor tracked the back-propagation of action potentials in L5 pyramidal neuron dendrites (Chien, et al., bioRxiv doi:10.1101/211946). Such a study could help identify the specific voltage dynamics underlying signal transmission and integration throughout the structure of a neuron.

5.2. Cellular-resolution recordings

Recently developed GEVIs can image individual action potentials in small or transparent organisms such as zebrafish, worms, and fruit flies, as well as optically accessible regions of the mouse brain such as the cortex and hippocampus [28,32,33]. These fast, precise recordings of voltage demonstrated novel phenomena such as correlations between firing rate and subthreshold potential fluctuations in mice. For example, targeted expression of hVoS in several classes of interneurons or mossy cells [35] resolved the action potential duration of these cell types with millisecond precision. In addition, GEVIs have helped probe state-dependent changes in defined regions of the hippocampus in behaving mice. The observed correlation between the subthreshold fluctuations of different neuron cell types and the local field potential (LFP) established state-dependent changes in excitability [28].

The single cell recordings have naturally scaled to simultaneous multi-neuron recordings within a genetically or anatomically defined subset of neurons. These simultaneous recordings examine millisecond-scale differences in spike timing between subsets of neurons as well as coordinated subthreshold fluctuations on a scale larger than that of electrophysiology. For example, Voltron recorded single spikes from individual neurons in the tegmental area of the zebrafish brain [37]. Voltron’s superior temporal resolution helped identify distinct neurons with highly separable responses at milliseconds timescales before or after visual stimulus-driven swims. This difference in timing suggested that one set of neurons encodes the motor command, while another set encoded an efferent copy.

Voltage imaging can work in parallel with optogenetic actuators to observe the causal effects of external stimulation. One combination of red-fluorescent GEVI and blue-light-activated rhodopsin, called Optopatch, probed the mechanisms of lateral inhibition of L1 interneurons in the mouse cortex (Fan, et al., bioRxiv doi:10.1101/614172). The varying effect of different amplitudes of optogenetic stimulation on interneurons’ firing rates and spike timing suggested that thalamic and top-down neuromodulatory inputs converge in L1 with different timing to produce precise responses to various sensory inputs.

Specific expression strategies that target GEVI expression can significantly improve the fidelity of multi-neuron recordings by drastically reducing background fluorescence. For example, soma-targeting of Archon doubled the number of detectable neurons per field of view in vivo (Piatkevich, et al. bioRxiv doi:10.1101/616094). The restricted expression of soma-targeted sensors enabled simultaneous recordings of the spikes and subthreshold oscillations from neurons in the hippocampus and striatum. These recordings found divergent spiking patterns between individual neurons despite highly correlated subthreshold fluctuations, practically demonstrating the power of voltage imaging to measure fast voltage dynamics from genetically targeted neural populations. An alternate strategy of sparse GEVI expression also improved SNR through the same principles. Here, a destabilized Cre line sparsely expressed VSFP in live mice, allowing for high SNR, single-cell imaging of cortical voltage dynamics with cellular resolution [40].

5.3. Ensemble-scale recordings

GEVIs can label populations of neurons via Cre-dependent expression to record voltage patterns from distinct classes of cells or between multiple populations with minimal cross-population contamination. Mesoscopic recording of voltage has inspired widefield imaging [41] or fiber-optic-based recording [42] experiments. In wide-field, GEVIs helped report the propagation of voltage throughout the rodent cortex. In fiber-photometry experiments, GEVIs identified stereotypical, cell-type-specific, subthreshold hyperpolarization voltage transients in the medium spiny neurons of the striatum absent in LFP recordings [42]. Such novel cell-type-specific physiology could provide details about the Parkinsonian disease state not accessible by other technologies. Moving forward, this method, in conjunction with the expanding spectral diversity of GEVIs [20,33], could enable relatively non-invasive, simultaneous recordings from distinct neural populations.

6. Conclusion

Recent neuroscientific studies have demonstrated the advantages of using GEVIs to probe neural activity in many models. Due to their enhanced brightness, expression, and voltage sensitivity, GEVIs can track fluctuations in membrane potential and fast-firing action potentials with high spatial resolution in vitro and in vivo. Further developments in screening technology will lead to improvements in GEVIs’ metrics, possibly attaining large-scale voltage recordings with single-cell resolution. Still, GEVIs are now already a promising technology to investigate neural signaling between genetically defined populations at the mesoscopic scale. These modern applications employ the speed and genetic specificity of GEVIs to extract detailed neural activity absent in other forms of recording.

Highlights.

• The performance of genetically encoded voltage indicators has steadily improved

• The sensors have attained the requirements for high-speed, in vivo imaging

• The sensors have revealed previously unseen neural phenomena in vivo