Illuminating Brain Activities with Fluorescent Protein-Based Biosensors

Fluorescent protein-based biosensors are indispensable molecular tools for life science research. The invention and development of high-fidelity biosensors for a particular molecule or molecular event often catalyze important scientific breakthroughs. Understanding the structural and functional organization of brain activities remain a subject for which optical sensors are in desperate need and of growing interest. Here, we review genetically encoded fluorescent sensors for imaging neuronal activities with a focus on the design principles and optimizations of various sensors. New bioluminescent sensors useful for deep-tissue imaging are also discussed. By highlighting the protein engineering efforts and experimental applications of these sensors, we can consequently analyze factors influencing their performance. Finally, we remark on how future developments can fill technological gaps and lead to new discoveries.


Introduction
The human brain has an average of 86 billion neurons, forming complex neuronal networks that are essential for behavior, intelligence, learning, and memory [1]. A fundamental goal in neuroscience has been to dissect neuronal circuits in vivo and to elucidate how identified populations of neurons contribute to behavioral outputs [2]. In fact, understanding brain function is not only a challenging scientific quest aiming to disentangle the functional relationship between the electrical, chemical, and hemodynamics (the dynamics of blood flow) in the brain with the cognitive, emotional, and behavioral outputs [3], but it's also a clinically crucial endeavor toward the study, diagnostics, treatment, and intervention of neurological diseases, such as Alzheimer's [4], stroke [5], and epilepsy [6]. As brains are structurally compact and functionally susceptible to invasive manipulations, optical imaging is highly compelling for interrogating their function [7].
The past 30 years have seen rapid development of brain imaging techniques [8]. Information obtained through brain imaging facilitates both functional interpretation and medical advancements toward addressing neurological diseases. While each method provides unique merits in studying brain activities, it also accompanies certain pitfalls that prevent one technique to dominate. For example, imaging modalities, such as functional magnetic resonance imaging (fMRI) [9], positron emission tomography [10], and X-ray computed tomography (X-ray CT) [11] have established themselves as clinically important tools for neurosurgical planning and diagnosing [12]. However, these techniques can only provide limited spatiotemporal information about brain function. Complementary to these techniques are electrophysiology and optical neuroimaging, through which electric and neurochemical signals in the brain can be detected and further related to neuronal and cortical functions [13]. While electrophysiology, the 'gold standard' for investigating neuronal functions [14,15], can directly measure electrical activity of cells with high sensitivity, the invasive requirement of physical contact with tissues and the accompanied poor spatial resolution lend its dominant position in neuroscience to be challenged by the emerging optical neuroimaging techniques [16]. The use of light to study neuronal signaling has several advantages, including limited invasiveness, tunable wavelength, high spatial resolution, and sensitive detection [17].
Optical neuroimaging measures neuronal activities by converting neuronal signals, such as voltage, calcium ions (Ca 2+ ), and neurotransmitters into light signal outputs ( Figure 1). It is thus an indirect approach towards studying neuronal signaling that necessitates the use of reporters (or probes, sensors, indicators, etc.) [16]. As such, the invention of optical reporters for brain imaging has been a highly intensive area of research [18]. Historically, calcium- [19] and voltage-sensitive dyes [20] have played pivotal roles in optically recording brain signals in vitro and in vivo. The use of dyes, however, is invasive and may result in nonspecific staining. Moreover, the spatial and temporal localization of dyes cannot be easily controlled, making targeted or chronic imaging difficult. As a complementary and competing technology, genetically encoded fluorescent indicators are becoming increasingly popular because they (1) can be introduced into cells through transgenic DNA expression; (2) enable targeted expression with precise spatial and temporal control; (3) facilitate largescale recording of neuronal activities in vitro and in vivo; (4) are amenable to protein engineering and optimizations; (5) are relatively noninvasive; and (6) are suitable for chronic imaging. However, since genetically encoded fluorescent indicators often require genetic delivery, they are less appropriate for direct, in vivo applications on human subjects. The past two decades has witnessed the fast development of a growing list of genetically encoded fluorescence probes, which have found broad applications within the neuroscience acceptor FPs with enhanced photophysical properties as they became available, reorganization of the CaM-M13 sensing module, evolution of the CaM-Ca 2+ binding site, and modulation of the linker length and/or composition between CaM-M13 and the flanking FPs [35]. For example, substitution of EYFP in YC2.1 with Citrine (YC2.3 and YC3.3) [45] or Venus (YC2.12) [72] FP produced GECIs with improved protein folding and maturation at 37 °C, that are indifferent to chloride ions, and that are more resistant to pH fluctuations at physiological conditions. Insertion of CaM binding peptide of CaM-dependent kinase kinase (CKKp) [73] into the linker region that connects the two EF-hands of CaM resulted in YC6.1, a GECI that increased the dynamic range of YC2.1 to reach 200% [34]. Despite the aforementioned improvements, these GECIs still suffer from a poor signal-to-noise ratio (SNR), especially when targeted to organelles or submicroscopic environments. To engineer high SNR, FRET-based GECIs, Nagai and co-workers employed a circularly permuted YFP (cpYFP) as the FRET acceptor [33]. Since cpYFPs have new N-and C-termini in close proximity to the chromophore [44], their use as FRET acceptors would result in different FRET performances, compared to their wild-type counterparts, due to an alteration of the relative orientation and dipole of the donor and acceptor chromophores. Accordingly, after sampling circularly permutated Venus (cpVenus) at different permutation sites and substituting them in YCs, they found that cp173Venus absorbs a greater amount of energy from excited CFP donor, thereby producing YC3.6 with Ca 2+ -dependent FRET ratio change of nearly 600%. Its large dynamic range enabled imaging of stimulated, fast-Ca 2+ dynamics of hippocampal brain slices from YC3.6-genetically encoded transgenic mice [33]. YC3.6 has since become one of the most frequently used GECIs.
Of equal importance to an indicator's dynamic range is its Ca 2+ binding affinity, which dictates GECI sensitivity in a given biological context [74]. To this end, Palmer et al. engineered improved cameleons (D2cpV, D3cpV, and D4cpV) based on computationally redesigned calmodulin-peptide pairs with Ca 2+ sensitivities tuned to more than a 100-fold range (0.6-160 μM) [35]. In 2010, Horikawa and colleagues developed ultrasensitive GECIs, termed yellow cameleon-Nano (YC-Nano, K d = 15-140 nM), by adjusting the length of the linker peptide between CaM and M13 ( Figure 2A) [36]. Together with its large signal change (14.5-fold) in the presence of Ca 2+ , YC-Nano enabled imaging of spontaneous motor activities in living zebrafish embryos [36]. To date, YC-Nano remains popular for detecting subtle Ca 2+ transients and basal-level neuronal activity in multicellular networks. For example, transgenic mice expressing YC-Nano15 or YC-Nano50 allowed ultrasensitive imaging of exocytotic events associated intracellular Ca 2+ dynamics of pancreatic acinar cells [75], and subtle, transient, and local activity in the fine processes of astrocytes (Ca 2+ twinkle) [76], respectively.
An apparent drawback to these YC-type GECIs resides is their use of CaM and M13 peptide as the Ca 2+ sensing module, which can potentially interfere with cellular biochemical machineries [77]. CaM is a ubiquitous and tightly regulated signal protein that interacts with a multitude of CaM-binding proteins [78] to activate and modulate kinases, phosphatases, and ion channels [79]. Therefore, expression of CaM-containing GECIs may inadvertently perturb normal cellular functions to an unpredictable extent. Moreover, unwanted binding of M13 to endogenous CaM can compromise sensor functions. Indeed, earlier attempts of using cameleons in transgenic invertebrates have encountered various problems, such as attenuated indicator performance in vivo and suboptimal targeting efficiencies [80][81][82]. To create minimally perturbing, and more "bio-orthogonal" GECIs, Griesbeck and colleagues explored the use of troponin C (TnC) [83], a skeletal and cardiac muscle-specific Ca 2+binding protein [84,85], to construct new types of GECIs. Inserting truncated or full-length TnC variants between CFP and Citrine leads to several TN-type GECIs, among which TN-L15 shows a 140% dynamic response and was targeted to the plasma membrane of HEK293 cells and primary hippocampal neurons [37]. Further engineering of the Mg 2+ -and Ca 2+binding sites within the C-terminal lobe of TnC, and use of alternative donor/acceptor pair (ECFP/Citrine cp174), improved the ion selectivity, dynamic range, and response kinetics of TN-L15, resulting in TN-XL with a 400% ratiometric change upon Ca 2+ binding. With a fast off-rate and large dynamic range, TN-XL enables fast, stable, and reproducible imaging of presynaptic motor neuron terminals of transgenic fruit flies [38]. Subsequently, domain rearrangement (i.e., deleting the lower-affinity N-terminal and concatenating the remaining C-terminal lobe of TnC) and mutagenesis inside EF hands III and IV, gave rise to a more sensitive GECI, designated TN-XXL [39], which can be used for chronic, in vivo, twophoton imaging of mouse visual cortex. By replacing the ECFP/cpCitrine pair with a 3xCFP/cpVenus pair, Liu et al. further improved the dynamic range of TN-XXL by nearly 11-fold [40]. More recently, the Griesbeck lab reported another TnC-based GECI, called "Twitch," which was developed by using the minimized C-terminal domain of Opsanus tau troponin C (tsTnC), which has fewer Ca 2+ -binding sites, higher Ca 2+ -binding affinity, and lower potential buffering of endogenously Ca 2+ [41]. Interestingly, the authors optimized the FRET responses of the Twitch indicators with a two-step, large functional screen: first in bacterial colonies and then in rat hippocampal neuronal cultures. The performance of the resulting indicators, such as Twitch-3 and Twitch-2B, rivals that of the synthetic dye, Oregon Green BAPTA1-AM [86,87]. They demonstrated the usefulness of the Twitch indicators by recording visually evoked calcium signals in mouse cortical layer II/III neurons and imaging T lymphocyte activation [41].

GECIs Based on BRET
While FRET-based GECIs have well distinguished themselves as powerful tools for neuronal imaging, the idea of deploying BRET (bioluminescent resonance energy transfer) for GECI design was a rare but viable one. In 2010, Saito et al. reported an auto-luminescent Ca 2+ sensor, BRAC [42], which is comprised of a Ca 2+ -sensing module (CaM-M13) sandwiched between a fluorescent protein acceptor (Venus) and a luciferase donor (RLuc8) [88,89]. Ca 2+ binding induced a conformational change within CaM-M13, followed by a 60% change in BRET efficiency [42]. They subsequently improved the BRET efficiency of the indicator and developed Nano-lantern (Ca 2+ indicator) by inserting CaM-M13 in a Nterminally truncated R-Luc8 of an enhanced Nano-lantern BRET pair. Nano-lantern has a 300% BRET dynamic range and was successfully used in concert with channelrhodopsin2 (ChR2) [90][91][92] in dissociated rat hippocampal neurons to conduct optogenetic activation and functional imaging experiments. By swapping the donor in existing FRET-or BRETbased biosensors with a recently developed, highly bright luciferase, NanoLuc [93], new chemiluminescent BRET biosensors were derived. CalfluxVTN is a genetically encoded NanoLuc based-chemiluminescent Ca 2+ biosensor modified from Twitch3 that can be used to couple optogenetic stimulation with Ca 2+ sensing [94]. Moreover, enhanced Nano-lanterns (eNLs) and an eNL-based Ca 2+ sensor with a 500% dynamic range have been recently reported [95]. BRET-based GECIs have the advantage of excitation-free, lower phototoxicity, and deeper tissue imaging capacity, but suffer from low photon output and the need of delivering bioluminescent substrates to points of interest. The recent development of a brighter and red-shifted teLuc luciferase [96] and the progress in engineering high efficiency BRET pairs in general is expected to partially alleviate this problem [43,[94][95][96]].

Single FP-Based GECIs
Historically, GECIs based on single-FPs flourish, having more popularity than FRET-based GECIs. Although different in many aspects of their working principles and photophysical properties, GECIs based on single-FPs and FRET share common grounds in their developmental history, from proof-of-principle ideas, to reiterative optimizations, to universal adoptions. In 1999, Tsien and colleagues reported an interesting finding, showing that major structural rearrangements of GFP are tolerable, including circular permutation and domain insertion [44]. This finding paved the foundation for designing new biosensors using circularly permutated fluorescent proteins (cpFPs), affording the first single-FP based GECI, termed "Camgaroo1", by insertion of CaM at Y145 of EYFP [44]. EYFP substitution for Citrine in camgaroo1 with Citrine lead to camgaroo2, which has improved expression at 37 °C, higher photostability, and reduced sensitivity to pH and chloride ions [45]. Despite proof-of-concept demonstrations, camgaroo-type GECIs failed to prove practical for Ca 2+ imaging in live cells, primarily due to their unfavorable Ca 2+ binding affinities (apparent K d for Ca 2+ is ~7 μM). However, the subsequent adoption of cpFPs [44,97] in GECI designs led to indicators with significantly better performances. In 2001, two Japanese groups independently reported single-FP based GECIs, designated G-CaMP [47] and pericam [46], which are comprised of CaM and M13 fused to the N-and C-termini of a cpFP, respectively. These GECIs not only exhibit large dynamic ranges, but also have Ca 2+ affinities (0.2-2 μM) within physiological ranges (50-5000 nM) [26], and are capable of imaging Ca 2+ dynamics in subcellular compartments and mouse myotubes [46,47]. Incorporation of known beneficial mutations for GFP gave rise to several G-CaMPs, including G-CaMP1.6 [48] and G-CaMP2 [49], with improved folding and maturation but marginal change to the dynamic range. Transgenic organisms expressing G-CaMP1.6 or G-CaMP2 were used to monitor neuronal activities of presynaptic boutons of the Drosophila larval neuromuscular junction [48], and mice cerebellar parallel fibers [49], respectively. In both cases, however, significant attenuation of SNR in vivo was observed [48,49], highlighting the need for improved G-CaMPs. In a study of cell polarization during Xenopus gastrulation, Shindo et al. briefly mentioned GCaMP4.1 but did not detail how it was developed [51]. Muto et al. tested substitution of "superfolder GFP" [98] mutations into cpEGFP of GCaMP2 and developed GCaMP-HS (GCaMP-hyper sensitive) [52], which is capable of imaging spatiotemporal activation of the spinal motor neurons in zebrafish. Attempts to further optimize these single-FP based GECIs, not surprisingly, proved difficult, owing to the intricacies and unpredictability of CaM-M13 triggered environment changes around the cpFP chromophore. To aid rational design of improved G-CaMPs, two research groups independently determined the X-ray crystal structure of G-CaMP2 and elucidated its Ca 2+dependent fluorescence transformation mechanism: Ca 2+ binding to CaM stabilizes a deprotonated chromophore in its otherwise protonated unbound state [50,99]. Soon afterwards, Tian and coworkers used a combination of G-CaMP2 structure-guided mutagenesis and semi-rational library screening to develop G-CaMP3 [50], a new G-CaMP with increased baseline fluorescence, increased dynamic range, and higher affinity for Ca 2+ . GCaMP3 outperforms all previous GECIs, including GCaMP2, D3cpVenus, and TN-XXL in pyramidal cell dendrites, and was successfully used to image neuronal activities in worms, flies, and mice [50]. As a state-of-art GECI at that time, GCaMP3 soon become the default indicator for various imaging applications, including imaging neuronal activity in the motor cortex [50], barrel cortex [100], and hippocampus [101] of mice; chronic imaging of learning-related circuit changes in mice in vivo [102]; imaging light-evoked responses in neuronal populations of zebrafish tectum [103]; Drosophila [104] and mouse retina [105]; and probing dendritic excitation in mouse cortical layer V dendrites in vivo [106]. Meanwhile, further improvements of GCaMP3 continued. Through structure-guided, targeted mutagenesis, linker modulation, and high-throughput screening, Looger's laboratory created a family of "GCaMP5" sensors with a wide range of Ca 2+ affinities, dynamic range, baseline fluorescence, and on-off kinetics [53]. GCaMP5s have been systematically characterized in cultured neurons and astrocytes, and in animal models, including Caenorhabditis elegans, Drosophila, zebrafish, and mice, where single action potential (AP)-evoked signals are clearly detectable [53].
While the brightness, Ca 2+ affinity, and dynamic ranges of GCaMPs are approaching those of synthetic Ca 2+ indicator dyes, such as OGB1-AM (Oregon Green Bapta-1-AM) [106,107], their sensitivities and kinetics are still a far cry. GCaMPs typically have half time for rise (τ on ) and decay (τ off , the time from peak fluorescence to 50% of the peak fluorescence was defined as the half time decay) of τ on = 20 ms-1.4 s and τ off = 0.4-5 s, respectively, which is much slower than physiological Ca 2+ flux signaling (rise within 1 ms and fall within 10-100 ms) [108], disfavoring their use to resolve spike times and firing rate variations over synthetic dyes (τ on <1 ms and τ off = 7 ms for OGB-1) [54]. Almost concomitantly, two groups reported their attempts, wherein they focused mutagenesis on CaM and the cpGFP-CaM interface to address this issue. Sun et al. developed Fast-GCaMPs, which shows up to a 20-fold accelerated off-responses and K d values spanning from 0.16 to 6 μM, and can track natural song in Drosophila auditory neurons as well as the rapid responses in mammalian neurons [54]. Chen et al. reported a family of ultrasensitive protein calcium sensors (GCaMP6) and showed that GCaMP6 is the first GECI to outperform OGB-1 in vivo [55]. Notably, these GCaMP6 sensors (GCaMP6s, 6m, and 6f for slow, medium, and fast kinetics, respectively) were directly screened from GCaMP variants expressed in dissociated rat hippocampal neurons, underscoring the importance of direct sensor screening in physiologically relevant settings. To date, GCaMP6 remains one of the most advanced GECIs for use in vivo.
Just as GCaMP indicators had been gaining satisfactory acceptance, Campbell's research group enthusiastically stepped into another untouched territory, engineering GCaMP spectral variants. Despite a decade of availability, the hue of single FP-based GECIs has, until 2011, been limited to monochromatic green. A red version of GCaMP indicator would be of particular interest in that it enables deeper tissue imaging, reduced phototoxicity, multicolor Ca 2+ imaging with other GCaMP indicators, and multiparameter imaging with existing indicators derived from other species. In a landmark study in 2011, Zhao et al. developed a blue, improved green, and red ratiometric excitation and emission Ca 2+ indicators with large dynamic ranges ( Figure 2B) [56]. These indicators were developed from a high-throughput, colony-based screening system in which indicator variants are secreted to the periplasmic space of E. coli. Manipulation of Ca 2+ concentrations in the periplasm by spraying Ca 2+ or EGTA solutions to LB agar plates altered the Ca 2+ on-off state, and thus the fluorescence of the indicators, which was then quantified by an imaging system [56]. To distinguish this new palette of indicators from the prior GCaMP series, a new term was coined for them, GECO (genetically encoded calcium indicators for optimal imaging). Multicolor imaging of Ca 2+ was successfully achieved in HeLa cells transfected with nucleus-localized R-GECO1, cytoplasmic G-GECO1, and mitochondria-localized GEM-GECO1 [56]. In addition, they performed multiparameter imaging of Ca 2+ and ATP by using R-GECO1 and ATeam1.03 [109]. Subsequently improved GECOs, such as R-CaMP1.07 [58] and R-GECO1.2 [57], GECOs with Ca 2+ affinities suitable for mitochondria and endoplasmic reticulum (i.e., LAR-GECO1 [62]), and GECOs of other spectral variants, such as O-GECO1 [57], CAR-GECO1 [57], and Y-GECO1 [59], have since followed. More recently, a GECO with a large Stokes shift with suitability for two-photon imaging [110], namely REX-GECO1, was developed and used in organotypic hippocampal slice cultures and the visual system of albino tadpoles [60]. GR-GECOs are green-to-red photo-convertible GECOs that can be used as an optical "highlighter" for Ca 2+ within a population of specified cells [61]. This diverse lineup of GECOs should find broad applications in neuronal imaging [22,111]. Independently, Akerboom et al. rationally designed and engineered G-CaMP color variants, of which the blue (BCaMP1c), cyan (CyCaMP1a), and yellow (YCaMP1h) indicators were developed by grafting GFP mutations that produced blue [112], cyan [113] and yellow [65] FPs, whereas the red (RCaMP1h) indicator was developed by using a cp version of mRuby [114] within the GCaMP scaffold [63]. Interestingly, RCaMP1h, when used in combination with optogenetic actuators [90][91][92][115][116][117][118], such as blue light-activatable channelrhodopsin-2 (ChR2) [90][91][92], was found to be superior than R-GECO1, which undergoes reversible photoactivation upon blue and green light illumination [63]. Coupling optogenetic manipulation with functional Ca 2+ imaging using the ChR2/RCaMP1h pair enables the investigation of a signal input-output relationship within defined neuronal populations, both intra-and inter-cellularly (such as with neurons and astrocytes) [63].
Advancements in FP technology itself have also inspired new GECI designs. A recent example of a green-red ratiometric Ca 2+ indicator based on FP exchange (FPX) technology underlies such new possibilities [64]. The tripartite, single-polypeptide Ca 2+ biosensor, with the structure RA-CaM-B-M13-GA, employs dimerization-dependent green [119] and red [119] FPs (GA, RA), which share a common dimerization partner (B). Ca 2+ binding shifts the equilibrium between RA-B and GA-B, resulting in a ratiometric red-to-green signal change [64]. The apparent advantage of FPX technology lies in its simple configuration from which new sensors can be quickly derived before extensive optimizations. The shortcomings, however, are that sensors derived from FPX technology are not suitable for quantitative imaging [64].

Genetically Encoded Voltage Indicators (GEVIs)
While tracking Ca 2+ dynamics represents a robust way for detecting downstream signaling events following membrane depolarization, transmembrane voltage measurements provide more direct information on neural activity [120]. During the course of an action potential, the membrane potential undergoes a rapid change that results in a single signal spike or subthreshold oscillation [121]. Transmembrane voltage fluctuations can lead to subsequent downstream Ca 2+ signaling or synaptic transmission within a single neuron, and the underlying information can also be relayed across long distances. Moreover, subthreshold membrane potential fluctuations cannot be detected by GECIs because they do not lead to any change in Ca 2+ concentrations. Therefore, voltage indicators are intrinsically attractive tools for observing neuronal activities [122]. However, the lack of well-established voltagesensing platforms together with the relatively faster temporal voltage fluctuations makes the design and application of voltage indicators far more daunting compared to Ca 2+ indicators. Nevertheless, the progress in the development of engineering genetically encoded voltage indicators (GEVIs) has been exciting ( Table 2) [123], and as we shall see, many voltage indicators of various classes have already found practical utility in vivo [124,125]. Several important aspects of GEVIs including sensor development, sensing mechanism, and in vivo applications are also reviewed elsewhere [24, 123, 126-129].

GEVIs Based on Voltage Sensitive Domains (VSDs)
Naturally found in some ion channels or phosphatases, voltage sensitive domains (VSDs) are membrane-bound components that bear voltage-sensing capabilities [158,159]. Typically, they contain four structural transmembrane helices (S1-S4) that modulate the gate of ion channels or control phosphatase activities. A defining feature that has been repeatedly harnessed by GEVI engineers is that the positively charged fourth helix (S4), or certain loop regions of VSDs, undergo substantial conformational changes upon membrane hyperpolarization or depolarization [123,160]. Not surprisingly, coupling the conformational change of VSDs to fluorescence perturbations, either in the form of intensity change of a single FP or FRET efficiency change between two FPs, has been the principle guiding strategy in early GEVI designs. In 1997, Siegel et al. reported the first GEVI, FlaSh, which is composed of GFP inserted into a non-conducting mutant of Shaker K + channel, and demonstrated that FlaSh GFP fluorescence can faithfully report membrane potential changes in Xenopus laevis oocytes [130]. Several improved GEVIs have since followed [136]. GFP replacement with other FP variants and/or rational incorporation of mutations known to alter gating kinetics and voltage dependence of the Shaker K + channel, yielded FlaSh variants with improved folding and distinct spectra, kinetics, and voltage dependence [161]. Use of a reversibly non-conducting form of the rat μI skeletal muscle voltage-gated sodium channel led to SPARC, which sports a faster response kinetics [131]. Tandem fusion of CFP-YFP to the C-terminal of S4 of a K V potassium channel produced a FRET-type GEVI, VSFP1 [132]. Yet, none of these first generation GEVIs gained widespread utility, predominately due to their poor mammalian expression and membrane trafficking [162].
Several attempts to fuse a single FP directly to the C-terminus of CiVSD only yielded GEVIs [138] with small dynamic ranges (<3%) such as VSFP3.1 [134] and VSFP3.1_mOrange2 [138]). A rare and exceptional example is ArcLight ( Figure 3C) [143], a GEVI with a surprisingly large dynamic response (18.1%) over a physiological voltage range (−70 mV to +30 mV). The large response amplitude of ArcLight and of its linker variants, such as ArcLight A242 and ArcLight Q239 (~35%) [143], are believed to have originated from an unintended point mutation A227D within the super ecliptic pHluorin [168], although the detailed mechanism is still insufficiently understood [169]. In cultured hippocampal neurons, ArcLight variants allow reliable detection of single APs and subthreshold electrical events [143]. More recently, Cao et al. demonstrated that ArcLight enables precise optical measurements of membrane potentials from intact neuronal circuits of whole Drosophila brain [170], presaging a bright future of in vivo optical electrophysiology with GEVIs. Combinational mutagenesis of CiVSD, or swapping of CiVSD with VSDs from other species in ArcLight, gave rise to GECIs with faster kinetics but attenuated response amplitudes, such as in Bongwoori [145] and chicken ArcLight [144], respectively. Direct evolution of key residues in the FP domain of ArcLight produced Marina, a GEVI that exhibits a ΔF/ΔV with a positive slope relationship [147].
One shared caveat to the abovementioned GEVIs is their relatively slow response kinetics (typically >10 ms) when considering the time-scale of action potentials (~1 ms [171]), thereby compromising the ability of these GEVIs to effectively detect subthreshold potentials and rapid trains of APs. To address this issue, the Lin group recently developed a fast and highly responsive probe (2 ms on-off kinetics, 18-29% ΔF/F), ASAP1 ( Figure 3B) [148], whereby they inserted a cpGFP into the S3-S4 loop of chicken VSD (GgVSD). Efficient transduction of VSD conformational changes to the chromophore environment of cpGFP is thought to be responsible for the accelerated kinetics, as was also similarly observed in VSD-cpGFP tandem fusion constructs [172,173]. Owing to its fast kinetics and large dynamic range, ASAP1 faithfully detected single APs and subthreshold potential changes at kHz frame rates [148]. Recently, improved variants of ASAP1 with higher sensitivity and more rapid kinetics including ASAP2s [149] and ASAP2f [150] has been developed and applied with two-photon imaging, revealing unique insights into neural processing in subcellular domains and neuronal tissues. Using a similar strategy, Abdelfattah and co-workers developed a red shifted GEVI (FlicR1) that can be used in conjunction with a blue-shifted channelrhodopsin for all-optical electrophysiology [146].

GEVIs Based on Microbial Rhodopsins
Another emerging class of GEVIs, whose development was pioneered by Cohen and coworkers, leverages the use of microbial rhodopsins as both the voltage sensing and reporting element [174]. Rhodopsins are typical membrane-bound G-protein coupled receptors (GPCRs) that function as channels, ion pumps, or light sensors [91,175]. Mechanistic studies of microbial-and bacterio-rhodopsins identified retinal and a retinallysine Schiff-base linkage in the protein core as key components in the photocycle that confers light sensitivity [175,176]. In 2011, Cohen and colleagues found that changes in membrane potential could induce a detectable absorbance shift in the retinal chromophore of green-absorbing proteorhodopsin [151], opening up the possibility of using rhodopsins as voltage sensors and detectors. In their initial report, they used a proteorhodopsin mutant (D97N) lacking its proton-pumping capability, designated PROPS, to detect electrical spiking in Escherichia coli [151]. However, the probe failed to localize to the plasma membrane of mammalian cells [152]. Fortunately, other rhodopsins, prominently Archaerhodopsin 3 (Arch), a light-driven proton-pump from Halorubrum sodomense [117], do not suffer similar localization issues. In cultured hippocampal neurons, Arch robustly detected single electrically evoked AP with an optical SNR >10 and shows a submillisecond response time [152]. A non-pumping mutant of Arch (Arch-D95N) can significantly reduce the photocurrent generated by light irradiation and has a high degree of sensitivity (50% greater than Arch) to resolve single APs, albeit with a slower response (30-36 ms) [152].
A serious drawback that soon became apparent with rhodoposin-based GEVIs was their intrinsic low brightness (quantum yield of Arch is only 0.0009) [152]. Rational mutagenesis and directed evolution methods targeting residues that modulate the Schiff base charge have identified Arch mutants showing higher brightness and sensitivity, such as Arch-EEQ, Arch-EEN [153], and Arch-7 [177], although these Arch mutants are still two to three orders of magnitude dimmer than commonly used FP fluorophores (QY of EGFP is 0.6). Nevertheless, mechanistic insights into the voltage-sensitive fluorescence of Arch and other rhodopsins [175,176,178], combined with novel screening platforms [154], still hold much promise for the discovery of improved rhodopsin-based GEVIs. The recent engineering of QuarsArs [154], which are Arch variants with improved brightness (QuarsAr1 is 15-fold brighter than Arch), sensitivity (90% ΔF/F per 100 mV for QuarsAr2), and kinetics (0.05 ms and 1.2 ms for QuarsAr1 and QuarsAr2, respectively), exemplifies such endeavors. Interestingly, combined use of QuarsAr with a spectrally compatible optogenetic actuator, CheRiff, in a coexpression vector enables all-optical electrophysiology (Optopatch) in mammalian neurons-totally abolishing the need for conventional electrodes [154]. More recently, Flytzanis et al. engineered two Arch variants with enhanced radiance (Archers) and demonstrated their use in probing voltage dynamics in behaving C. elegans [156]. Notably, Archer1 has wavelength-specific dual functionality, as a voltage sensor under red light and as an inhibitory actuator under green light [156]. The development of Optopatch and Archer1 presages an exciting era of all-optical neurophysiology [154].
Alternative strategies to enhance the brightness of rhodopsin-derived GEVIs focused on FPopsin fusions, whereby voltage induced absorbance changes of the opsin quench the emission of its brighter FP partner via electrochromic FRET (eFRET) [154]. As fluorescence change is usually detected in the FP channel, much less laser intensity is needed. Although FP-opsin fusions generally have slower response kinetics compared to standalone opsins, as observed in QuarsAr2-Citrine [154] and MacQ-mCitrine ( Figure 3D) [155], the combined benefits of high voltage sensitivity of opsins and high brightness of FPs still enable these GEVIs to reliably detect single APs and sub-threshold voltage dynamics [155], thereby outperforming VSFPs, such as ArcLight. Furthermore, swapping FP spectral variants and optimizing the intervening FP-opsin linker are promising strategies toward further enhancing the performance of FP-opsin eFRET sensors [154]. Combined use of a fast response rhodopsin (Acetabularia acetabulum rhodopsin, Ace) and a bright FP (mNeonGreen) gave rise to Ace-mNeon, an ultra-fast GEVI that enable high-fidelity imaging of fast spike trains in live mice and flies [157].

Genetically Encoded Fluorescent Indicators of Synaptic Activity
Synaptic transmission is a signature event for neuronal information processing downstream of neuronal firing as it plays essential roles in information processing and memory formation [179]. During this process, APs approaching the synaptic bouton prompt presynaptic vesicles to exocytose and release neurotransmitters into the synaptic cleft, allowing neurons to communicate with each other via electrochemical signaling [180]. The development of indicators for synaptic activity lags far behind indicators for calcium and of membrane voltage [181]. Indicators of this class include those purposed for detecting neurotransmitter concentrations (such as glutamate) and pH changes during synaptic vesicle recycling (Table  3).

Genetically Encoded Glutamate Indicators
Glutamate is a major excitatory neurotransmitter in the brain [190]. Postsynaptic glutamate release and extrasynaptic glutamate signaling ('spillover') give rise to many important neuronal processes, including synaptic crosstalk, learning, and memory [191,192]. In 2005, Okumuto et al. reported the first fluorescent indicator protein for glutamate (FLIPE) by bracketing the bacterial periplasmic glutamate binding protein ybeJ (also known as "GltI") [193] with CFP and Venus [182]. Glutamate binding triggers a conformational change within GltI, leading to a detectable FRET efficiency change. Tsien and colleagues later adopted a similar design and created a glutamate-sensing fluorescent reporter (GluSnFR) [183]. Both indicators showed glutamate-dependent FRET ratio changes in vitro, but their limited response amplitudes was insufficient for quantitation applications in neurons. Linker and binding affinity optimizations of GluSnFR gave rise to an enhanced probe, Super GluSnFR (44% ΔF/F), that was successfully used to quantitatively image synaptic glutamate spillover and reuptake in cultured hippocampal neurons with centisecond temporal and dendritic spine-sized spatial resolution [184]. In addition, an intensity-based, glutamatesensing, fluorescent reporter (iGluSnFR) derived from cpEGFP insertion into GltI, was engineered with SNR and kinetics appropriate for in vivo imaging ( Figure 4A) [185]. With a large dynamic range (4.5 ΔF/F) and fast response kinetics, iGluSnFR enables visualization of glutamate neurotransmission in intact neurological systems, including somata, dendrites, and dendritic spines in mouse retina, worms, zebrafish, and mice [185].

Genetically Encoded pH Indicators for Synaptic Vesicle Recycling
During the neurotransmitter-containing vesicles fusion to the presynaptic membrane upon AP propagation, the lumen of the vesicle, which is normally maintained at an acidic pH ~5.6, is exposed to the extracellular space (pH ~ 7.4). Subsequent vesicle recycling resets the acidity of the vesicle pH [194,195]. Thus, tracking the vesicle pH changes associated with vesicle exocytosis and recycling events represent a plausible way to report on synaptic transmission. pH-sensitive variants of GFP are well suited to such applications. The first genetically encoded pH indicator tailored for this application is SynaptopHluorin, in which a pH sensitive GFP (pHluorin) was fused to the C-terminus of synaptobrevin/VAMP2 (vesicular associated membrane protein-2) [168]. The indicator was successfully localized to the inner surface of synaptic vesicles and faithfully reported local pH changes associated with transmissions at individual synaptic boutons. To explore the in vivo application of synaptopHluorin as a neurotransmitter indicator, Ng et al. targeted the probe to all three classes of neurons in the antennal lobe of Drosophila and performed functional imaging of olfactory circuits in response to natural odors [196]; Bozza et al. expressed the probe in mice mature olfactory sensory neurons, allowing them to monitor odorant-evoked activation of the sensory neurons in glomeruli of the olfactory bulb, and to reveal the spatial patterns of odorant-activated glomeruli [197]. Alternative targeting domains, such as synaptophysin [186] and vesicular glutamate transporter (vGlt1) [187], were used to better target pHluorin to the synaptic vesicle. Red-shifted variants of synaptopHluorin, including VGLUT1-mOrange2 [188] and sypHTomato ( Figure 4B) [189], were also developed by switching pHluorin to pH-sensitive, red-shifted FP mutants. These color variants are spectrally compatible with GCaMP and permit concurrent measurements of calcium dynamics and synaptic vesicle recycling [188,189].

Future Perspectives
As with electrophysiology, genetically encoded fluorescent probes offer a powerful approach to interrogate neuronal activities. While a plethora of fluorescent probes have been successfully developed and deployed to analyze neuronal-specific parameters, such as calcium, membrane voltage, neurotransmitter, and synaptic vesicle recycling, the currently available sensor toolbox, experimental methodologies, and optical modalities fall short of achieving the ultimate goal of proficiently investigating systems neuroscience, i.e., understanding how populations of nerve and glial cells form circuits underlying behaviors in awaken animals. The progress made in the past two decades with fluorescent probe development provides a strong foundation for future innovations aiming to realize the full potential of these optical probes for which research opportunities and challenges coexist and await further exploration.
First of all, the toolbox of genetically encoded fluorescent probes need to be further expanded to supplement currently widely used calcium and voltage indicators. Neuronal firing involves a cascade of signaling activities from which rich information are encoded and transmitted. While calcium and voltage play pivotal role in this process, other species, such as neurotransmitters, neuropeptides, and neuromodulators are also of paramount importance, but for which very few probes are available. In particular, indicators for many neurotransmitters, such as glutamate, γ-aminobutyric acid (GABA), histamine, serotonin, zinc ion [198] and dopamine remain attractive but hard-to-make [199]. Since voltage, calcium, and neurotransmitter changes represent functionally relevant, yet nonlinear, aspects of synaptic transmission [200], reporters for each class are of distinct value to study neuronal circuits. The development and in vivo application of the glutamate indicator iGluSnFR [185] have already showcased its indisputable potential in this area. In addition to creating new indicators for distinct species, further optimization of existing indicators will facilitate the recording of neuronal actives with higher SNRs and higher fidelity over the spatiotemporal scales most relevant to neurophysiological conditions. Importantly, both single-FP and FRET-based indicators have their own merits and limitations, and thus, should be selected with caution for specific applications. Single-FP based intensiometric biosensors typically have larger SNR than FRET-based biosensors, but are often sensitive to probe concentration and pH changes; ratiometric probes often suffer from high background signals due to the intrinsic inefficiency of the FRET process. Moreover, turn-off responses are less favorable than turn-on responses as photobleaching may complicate data interpretation. Furthermore, exploration of new sensing mechanisms such as FPX may provide ample opportunities for further innovation. In addition, indicators with red-shifted spectra (ideally near-infrared) are desirable for in vivo applications because they can potentially alleviate the light scattering and tissue penetration issues [201] encountered by commonly used green fluorescence indicators. Conceivably, red-shifted indicators can be used simultaneously with the reiteratively optimized GCaMP calcium indicators or blue light-activated channelrhodopsins [202,203] to record and manipulate neuronal signals in multiplexing imaging experiments, as these indicators are spectrally orthogonal to each other. Parallel to the development of new indicators, the very nature of these fast and interconnected neuronal activities naturally calls for continuous optimizations of the brightness, photostability, response kinetics, and dynamic ranges of existing indicators, some of which have already found proof-of-concept demonstrations in vivo. In particular, the dynamic range and SNR of current GEVIs are far from optimal and necessitate major engineering efforts before their wide spread adoption.
The continuing advancement in the bioluminescence technology is also sparking a revolution in the development of bioluminescent sensors for brain imaging. On the one hand, caged luciferins are powerful tools to detect small and low abundant species such as ATP [204], copper-(I) [205], nitric oxide [206], and highly reactive oxygen species (hROS) [207]. In response to specific analytes, these caged luciferins were converted to better substrates for luciferases, resulting in enhanced bioluminescence signals. On the other hand, by swapping the donor FPs in existing FRET-based biosensors with luciferases, new chemiluminescent BRET biosensors were developed. In particular, the replacement of CFP in FRET sensors with NanoLuc resulted in BRET sensors with good sensitivity and response magnitude. For example, BTeam [208] modified from ATeam, LOTUS-V [209] modified from Mermaid2, and BLZinCh [210] modified from eZinCh-2 are NanoLuc-based BRET biosensors for ATP, membrane voltage, and Zn 2+ , respectively. We expect that the recently developed teLuc-cyOFP1 BRET pair may be utilized to further enhance these BRET-based biosensors [96]. As most FRET-based biosensors are modular, this approach could in principle be applicable to the development of BRET biosensors for various biomolecules or functions. Moreover, one may directly insert sensory domains into luciferases, such as NanoLuc and teLuc, to derive intensiometric, bioluminescent biosensors. Together, these efforts will create a versatile optical toolbox for deciphering the logics of neural activity in live, non-sedated animals.
Of similar importance to probe development is the advancement in optical instrumentation and imaging data processing. Recent progresses in multiphoton imaging [211][212][213], digital light sheet microscopy [214], aberration-corrected multifocus microscopy [215], and spatial light modulator microscopy [216] hold much potential for evolutionary modalities for brain imaging [217]. Accordingly, imaging acquisition and data processing methods [8,22] need to be updated to accommodate larger and faster scales of information extraction from experiments.
Finally, optimization of the transduction methods to deliver these probes to specific cell or tissue types will be crucial to achieve spatial resolution in brain regions of interest. These include careful choice of promoters and vectors in viral packaging, and on-demand optimization of other transgenic techniques, such as in utero electroporation and stable transgenesis. Novel genome editing techniques, such as clustered regularly interspaced short palindromic repeats (CRISPR)-Cas9 [218] may also play pivotal roles in standardizing and streamlining the generation of transgenic rodent lines expressing various probes.        [156] Ace-mNeon Acetabularia acetabulum rhodopsinl (Ace), Neurons, awake mice and flies [157] a Dynamic range (%) here refers to the absolute value of fluorescence change (ΔF) normalized by the initial fluorescence (F 0 ), expressed as % ΔF/F 0 , following 100 mV depolarization or repolarization steps. Values are steady-state or peak changes, as determined in the original reports. One has to be cautious when comparing these numbers, because experimental conditions could be different; b,c The value here refers to one of the representative values measured in one cell type under a certain experimental condition specified in the original work, and may differ in other publications repeating the original work; d Unlike other similar GEVIs (e.g., ArcLight, ASAP1, ASAP2s, ASAP2f, and Ace-mNeon) that have a "negative" response (i.e., decrease in fluorescence upon neuronal activation), Marina was engineered for an inverted response (i.e., increase in fluorescence upon neuronal activation).
Chen et al. Page 36 Table 3 A List of genetically encoded indicators for synaptic activity.  [189]