Genetically encoded fluorescent sensors of membrane potential
- First Online:
- Cite this article as:
- Baker, B.J., Mutoh, H., Dimitrov, D. et al. Brain Cell Bio (2008) 36: 53. doi:10.1007/s11068-008-9026-7
- 998 Views
Imaging activity of neurons in intact brain tissue was conceived several decades ago and, after many years of development, voltage-sensitive dyes now offer the highest spatial and temporal resolution for imaging neuronal functions in the living brain. Further progress in this field is expected from the emergent development of genetically encoded fluorescent sensors of membrane potential. These fluorescent protein (FP) voltage sensors overcome the drawbacks of organic voltage sensitive dyes such as non-specificity of cell staining and the low accessibility of the dye to some cell types. In a transgenic animal, a genetically encoded sensor could in principle be expressed specifically in any cell type and would have the advantage of staining only the cell population determined by the specificity of the promoter used to drive expression. Here we critically review the current status of these developments.
Optical imaging is a remarkably flexible method for studying various cellular activities. Organic dyes have been developed that can faithfully report many biological variables including calcium concentration, pH, or membrane potential (Davila et al., 1973; Brown et al., 1975; MacDonald and Jobsis, 1976). These optical probes enable simultaneous measurements from many locations and have been used to study the physiology of single neurons as well as large populations of cells (selected work reviewed in Grinvald and Hildesheim, 2004; Baker et al., 2005; and Knöpfel et al., 2006).
Two major drawbacks of these organic dyes are the non-specificity of cell staining and the low accessibility of the dye to some cell types. They either stain all cells in a tissue to which they are exposed, or if delivered in a patch pipette they can only label one or a small number of cells. In the absence of targeting of the organic dyes to specific cell types, optical signals of interest are often drowned out either by background fluorescence from inactive cells or by signals in cells that are not the focus of interest. In addition, with the exception of preparations with sparse cell staining, diffuse labeling limits the spatial resolution so that single cell responses cannot be readily resolved. One general solution to this problem is to make optical reporters from proteins. Since protein-based reporters are encoded in DNA, they can be placed under the control of cell-specific promoters and introduced in vivo using gene transfer techniques. Moreover, protein-based reporters can, in principle, be rationally “tuned” by modification of their functional domains with mutations that adjust their dynamic range of operation. In a transgenic animal, a genetically encoded sensor could in principle be expressed in any cell type and would have the advantage of staining only the cell population determined by the specificity of the promoter used to drive expression.
Protein-based reporters are generally constructed from two parts: a “sensor” protein that undergoes a conformational rearrangement that depends on the parameter measured (e.g., a voltage-gated ion channel that senses membrane potential), and a fluorescent protein (FP) reporter fluorophore whose optical output is modulated by the sensor protein (for a recent review, see Knöpfel et al., 2006). The first FP voltage sensor, denoted FlaSh, was obtained by inserting GFP downstream of the pore region in the Drosophila voltage-gated potassium channel, Shaker (Siegel and Isacoff, 1997). When FlaSh was expressed in Xenopus oocytes, changes in membrane potential were reported by changes of the fluorescence. However, this initial success was not followed by published reports of signals from mammalian brains using FlaSh or the two subsequent first generation voltage sensors, VFSP-1 (Sakai et al., 2001; Knöpfel et al., 2003) and SPARC (Ataka and Pieribone, 2002).
A group of laboratories (Yale, PI Cohen; Berkeley, PI Isacoff; Yale, PI Pieribone; Montana, PI Hughes; and RIKEN BSI, PI Knöpfel) has begun an effort to improve genetically encoded FP voltage sensors for use in mammalian preparations. In order to realize a new, improved generation of sensors, inefficient targeting to mammalian plasma membrane was identified as a major limitation of the first generation FP voltage sensors. Therefore, the group started a concerted effort to find a more suitable signaling protein and FP reporter-combination, whilst also fine-tuning emerging sensor proteins.
Here we review the historical development of the first generation FP voltage sensors, describe the state-of-the-art second generation FP voltage sensors, and outline efforts for the next generation FP voltage sensors. We then compare these traditional fully genetically encodable sensors with new partially genetically encoded probes. Finally, we critically discuss the practical and theoretical limitations of current FP voltage sensor-based imaging techniques.
First generation FP voltage sensors
The very first prototype, FlaSh, was generated in the Isacoff laboratory and obtained by fusing wtGFP to the C-terminus of the Drosophila Shaker potassium channel (Siegel and Isacoff, 1997). When expressed in oocytes, an 80-mV depolarization of the plasma membrane resulted in a 5% decrease in fluorescence (Fig. 1 left panel), this was designated FlaSh, for “fluorescent Shaker”.
To reduce unwanted effects on the cell’s physiology, FlaSh was rendered non-conducting by introducing a W434F mutation, preventing ions from moving through the pore while maintaining voltage-dependent rearrangements (Perozo et al., 1993). In order to resolve action potentials, an ideal sensor would generate a robust signal on a millisecond timescale. The signal strength from FlaSh is comparable to that of the voltage-sensitive organic dye, di4-ANEPPS; however, the on and off rates are rather slow (τ-on ~100 ms; τ-off ~60 ms). In an effort to improve the kinetics of FlaSh, Guerrero et al. (2002) replaced the wtGFP with several different FPs. The change in optical characteristics mediated by the different FPs was substantial. The signal intensity, the direction of the fluorescent change, and the speed of FlaSh were all altered by the chromophore in an unsystematic way. Variants involving wtGFP and uvGFP both have reduced fluorescence in response to depolarization steps, while Ecliptic variants of GFP, YFP, and CFP exhibit increased fluorescence in response to depolarization. Remarkably, upon depolarization, an increase in fluorescence is seen when eGFP is excited at 450 nm, but there is a decrease in fluorescence when excited at 480 nm. The speed of the response is also governed by the chromophore, with the Ecliptic variant of GFP generating the fastest response (τ-on ~5 ms). While the mechanism underlying the fluorescence change that results from an alteration in membrane potential remains poorly understood, it is clear that the fluorescent reporter contributes significantly to the kinetics of the optical response and is an important parameter to vary in an attempt to improve genetically encoded voltage sensors.
The second prototypic design realized in the Knöpfel laboratory and termed VSFP1 exploits the voltage-dependent conformational changes around the fourth transmembrane segment (S4) of the voltage-gated potassium channel Kv2.1 and uses either fluorescence resonance energy transfer (FRET) (Fig. 1, middle panel; Sakai et al., 2001) or a permuted FP (Knöpfel et al., 2003). The third prototype, SPARC (Pieribone laboratory; Fig. 1, right panel), was generated by inserting a FP between domains I and II of the rat skeletal muscle Na+ channel (Ataka and Pieribone, 2002).
The absence of a signal from the first generation probes is in part due to their low membrane expression and in part due to a large, nonresponsive background fluorescence that would mask any voltage-dependent signal. Because Kv1.4 with an N-terminal GFP exhibits excellent membrane expression, several strategies to release Flare or its Kv2.1-based homologues from the ER have been tried. Unfortunately, mutagenesis of potential ER retention signals, addition of ER release motifs, and expression in hippocampal neurons that endogenously express trafficking partners all failed to significantly improve the plasma membrane expression (Ray et al., unpublished observations; Baker and Cohen, unpublished observations).
For the second generation of FP voltage sensors, we hoped to overcome this issue of poor targeting to neuronal membranes.
Second generation FP voltage sensors
Recently, several lines of evidence supported the concept of self-contained voltage sensor domains in naturally occurring proteins such as Kv channels. Thus, a voltage sensor domain may function in the absence of the structures seen in the complete Kv channel complex. The first body of evidence pointing toward this possibility came from crystal structures and structure-function modeling of KvAP and Kv1.2 that indicated that the secondary structure of the S1–S4 portion of Kv channels is relatively independent from neighboring parts of these channels (Jiang et al., 2003a, b; Long et al., 2005a, b). The second line of evidence was based on the identification of voltage sensor domains in non-ion channel proteins. Ci-VSP (Ciona intestinalis Voltage-Sensor-containing Phospatase), for example, is a voltage-controlled enzyme consisting of a transmembrane voltage sensor domain and a cytosolic phosphoinositide phosphatase domain (Murata et al., 2005). The voltage sensor domain from this protein was shown to be functional (i.e., able to generate gating current in response to voltage steps) when the enzyme domain was removed (Murata et al., 2005). More recently, it was also shown by the Isacoff laboratory that the Ci-VSP can exist as a monomer in the plasma membrane (Kohout et al., 2007). A third line of evidence for voltage sensor domains that can function in isolation emerged with the discovery of a new protein family termed Hv1 or voltage sensor domain only proteins (VSOPs). These membrane proteins mediate voltage-dependent proton transport and have a domain homologous to the S1–S4 portion of Kv channels but lack the putative pore forming S5–S6 domain (Ramsey et al., 2006; Sasaki et al., 2006). Hv1 is a dimer, with each of the voltage sensor domains containing its own gated pore; moreover, the pore can function when mutations are made that turn the channel into a monomer consisting of only a single voltage sensor domain (Tombola et al., 2008).
Next generation FP voltage sensors
Further development of VSFP2.1 is progressing in several directions: (i) linker optimization, (ii) alternative FP variants, and (iii) a search for additional ways to fuse or insert FPs into Ci-VSP or alternative sensor proteins.
Linker optimized variants
It is well established for other types of FP sensors that sensitivity of the probe (ΔF/F or ΔR/R values per change in membrane voltage) can be increased by optimizing the length and amino acid composition of the linker between the sensor protein and the FP. To such effect, the performance of VSFP2.1 was improved by removal of five amino acids that initially originated from engineered restriction sites in VSFP2.1 (Dimitrov et al., unpublished observation; Villalba-Galea et al., 2008). Further improvement of sensor sensitivity by optimizing linker is likely to be realized in near future.
Alternative FP colors
Cyan- and yellow-emitting variants of green fluorescent protein (GFP) from Aequorea victoria (CFP and YFP, respectively) are the most widely used components for FRET-based biosensors. However, there are reasons why alternative colors, particularly red shifted versions, are more desirable: (i) for deep tissue imaging, (ii) for reduced interference from tissue autofluorescence and hemoglobin absorption, and (iii) in combination with other fluorescent probes (spectral separation). Several efficient variants of these were very recently generated in the Knöpfel laboratory, the most promising comprised combinations of either a green FP with a red FP or a yellow FP with a far-red FP (Dimitrov et al., unpublished observations).
In addition, for some applications, non-FRET monochromatic probes are more favorable. Single FP sensors have been demonstrated previously including the Flash type of voltage sensors and several types of Ca2+ sensors (Baird et al., 1999; Nakai et al., 2001). Thus, we recently developed analogous single FP voltage sensors based on VSFP2s (Lundby et al., 2008; Mutoh et al., unpublished observation; Baker and Cohen, unpublished observation). The best variant characterized to date exhibited an activation time constant (τ-on) of <3 ms at depolarized membrane potentials (Lundby et al., 2008).
In a canonical voltage sensor domain, it is not only the S4 segment that responds to voltage shifts, but rather the whole domain undergoes conformational changes in a complex manner (Pathak et al., 2007). It follows, therefore, that finding alternative sites for insertions of FPs along the voltage sensor domain that would not disturb the function and folding of either the voltage sensor domain or the FP may lead to new designs of improved functional voltage sensors. To explore this approach, the Hughes laboratory is using its transponson technology to generate a library of key positions in the Ci-VSP that would allow for insertion of a FP without disturbing the folding of the Ci-VSP (critical for voltage sensing), the folding of the FP itself (critical for the fluorescence report), or plasma membrane targeting.
In addition to the above there are many additional, only partially explored approaches for conformational tuning of the parts constituting the sensor. These include addition of prenylation signals or other signals for post-translational modifications that may help improve the plasma membrane targeting and/or conformational states, in a way that would be advantageous for the desired function.
Genetic targeting of neurons
Perhaps the most important feature of genetically encodable sensors is the possibility to achieve indicator loading in specific cell populations. This is the case because cell type-specific protein expression can be achieved by the use of appropriate regulatory gene sequences (promoters). This concept had been proposed many years ago in conjunction with the bioluminescent calcium-sensitive reporter protein aqueorin; however, the light output of bioluminescent probes was rather limited and consumed a co-factor that needed to be applied to the preparation. Over the last years cell type-specific expression of FP variants, or fusion proteins containing them, has been demonstrated in numerous transgenic and gene targeted mouse lines. For some experimental questions it is desirable to label cells sparsely so that individual cells can easily be identified by fluorescence imaging in densely packed tissue. Sparse labeling can be achieved with “shotgun” promoters such as Thy1 (Feng et al., 2000), virus-based transfection (Dittgen et al., 2004), or transgenic strategies for combinatorial expression of FP components (Livet et al., 2007) but is associated with considerable uncertainty about the identity of the labeled cells and/or whether the labeled cells are representative for the given target population.
The specificity of virus-based transfection can be increased by cell-specific promoters, but this technique is limited by the size of recombinant sequence which can be packaged in to the virus. Perhaps more promising is the combination of viruses with mouse lines expressing recombinases, such as Cre and Flp, in specific subsets of neurons. This technology permits targeted expression of FP sensors in any subset neuron, for which one of the now many recombinase-mice lines are available. For that purpose the sensor gene in a lox-stop-lox cassette is inserted into a virus and then expressed under lox recombination as defined by the recombinase mouse.
A more expensive and time-consuming approach is direct transgenic or genetic targeted expression of the FP sensor. The principal advantage being reproducible preparation for experimental convenience and for a greater certainty in establishing what is stained (and—equally important—what is not). With such genetic models the precise targeting allows optical signals to be reliably attributed to specific subsets of cells within a network consisting of many different cell types, whilst also allowing low noise imaging to be performed without optically resolving the labeled cells (Knöpfel et al., 2006; Díez-García et al., 2005, 2007; Qiu and Knöpfel, 2007).
Genetically encoded sensors of membrane potential compared to alternative genetic targeting approaches for fluorescence voltage-sensitive probes
In addition to voltage sensors that are entirely genetically encoded, at least two other approaches to overcoming indiscriminant dye staining use genetics in combination with conventional organic chromophores. The first is a hybrid approach that combines the advantages of conventional voltage-sensitive dyes (relatively large ΔF/F) with genetic targeting (Chanda et al., 2005; DiFranco et al., 2007). The hybrid probe hVOS employs FRET between membrane-anchored GFP and a dye, DPA, that partitions into the plasma membrane in a membrane voltage-dependent manner. While this hybrid probe produces the largest reported ΔF/F values among all genetically encoded voltage-sensitive probes, the high concentration of the lipophilic anion DPA has a significant effect on the activity of excitable membranes (Fernández et al., 1983; Chanda et al., 2005; Zimmermann et al., 2008). Thus reduced DPA concentrations were required to obtain action potential responses (Chanda et al., 2005) resulting in much smaller fractional changes (∆F/F).
Using an optimized DPA-based voltage reporter, Sjulson and Miesenböck (2008) found a failure to detect action potentials in the Drosophila antennal lobe, presumably because spike initiation and/or propagation are inhibited by the capacitive load added even at reduced DPA membrane densities.
The second semi-genetic method uses the genetically targetable expression of an enzyme with an organic dye as a substrate. The idea is that only the enzymatically processed dye stains the plasma membrane where it functions as a conventional voltage-sensitive dye. In a proof of principle study (Hinner et al., 2006), a lipophilic dye was made water-soluble by the addition of phosphate groups to the hydrophobic tail. Cell-specific staining was achieved by genetically expressing an outward-facing membrane-bound phosphatase that cleaves the hydrophilic phosphate group from the engineered dye, resulting in a closely localized population of lipophilic product dye that integrates and stains the membrane of the specified target cells. While this approach has potential and essentially could be used with a variety of established voltage-sensitive lipid dyes, in its current state the applicability in neuronal tissues is unproven. Several obstacles become apparent; first, the usual issues of dye accessibility to the area of study and dye toxicity. Second, a question arises whether the phosphatase would not catalyze intrinsic processes disturbing the physiological state of the tissue, or the dye itself would not become substrate to intrinsic proteins resulting in unwanted staining. Third, in a tissue where the individual cells are closely packed together, the “released” dye is expected to diffuse and bind to plasma membranes of “not-selected-for” cells that are adjacent to the cell expressing the enzyme.
The motivation for the development of FP voltage sensors is to overcome unselective labeling obtained when staining tissue with organic voltage-sensitive dyes. Selective staining of specific subsets of neurons will allow a more stringent interpretation of optically monitored voltage signals, in terms of their source. The second, similarly important benefit is an increase in S/N ratio because undesired staining contributes to noise but not to signal. In this context, it was as a consequence of unspecific labeling (in the form of limited membrane targeting) that the first generation of FP voltage sensors failed. The second generation overcame this problem. Yet, the signal-to-noise ratio (RS/N) is probably the most critical parameter, particularly when a high temporal resolution is required. For in vivo experiments a low RS/N often results from a contamination of the sensor signal by movement-induced changes in the measured light intensity, for instance: changes in blood volume or hemoglobin oxygenation, light scattering or endogenous tissue fluorescence (for a recent review, see Baker et al., 2005). Synchronization with breathing and/or heartbeat along with careful selection of excitation and emission wavelengths can reduce these noise signals significantly (Baker et al., 2005).
There is, however, a physical limit for RS/N in optical recordings due to the statistics of photon detection (expanded in Knöpfel et al. 2006 and Sjulson and Miesenböck, 2007). Thus, in the best case, RS/N is increased with the square root of the number of photons detected per temporal and spatial bin. Accordingly, it can be predicted that the first successful (because least demanding) application of FP voltage sensors in intact tissue will be voltage imaging from populations of subsets of neurons, as previously demonstrated with FP calcium sensors (Díez-García et al., 2005, 2007). Spatial averaging over many neurons that carry a common voltage signal will provide a sufficient emission intensity to detect small modulations (Vranesic et al., 1994). RS/N decreases with the sampling rate (as the number of sampled photon decreases with shorter sampling intervals, but also because of an increased contribution of noise resulting from the sampling process). For these reasons, along with the fact that imaging sampling frequency is limited for some imaging techniques (like laser scanning), an FP voltage sensor with somewhat slow response characteristics may be more suitable for certain applications. Indeed, slow response kinetics act as a signal conditioning mechanisms (as required by the sampling theorem). Yet, timing information may be recovered from slowly responding indicators by deconvolution techniques (Yaksi and Friedrich, 2006).
Capacitive load and other possible caveats
Genetically encoded optical sensors make it possible to record from multiple cells in undamaged tissue. However, as a recombinant gene product, the sensor itself can interfere with intrinsic biochemical and/or physical cell functions. Hence, as in the case of any artificially introduced probe, FP sensor-derived data require verification against controls to demonstrate the absence of probe-induced undesired effects. Unlike synthetic potentiometric probes, e.g. cyanin and oxonal dyes, which are notorious for their phototoxicity (Grinvald et al., 1982; Kalyanaraman et al., 1987), FP-based sensors are not known, and are not expected, to form toxic illumination-induced products. Similar to synthetic probes, however, they add to the nonlinear dielectric polarization of the membrane resulting in increased membrane “capacitive load” (Sjulson and Miesenböck, 2007). The effect is based on a simple physical mechanism: The voltage sensor domain moves positive charges across the membrane in response to changes in membrane potential. This charge movement manifesting as a transmembrane capacitive current (well known as gating currents for ion channels) redistributes charge between the inner and outer membrane surface in a voltage sensor domain state-dependent manner, tantamount to a voltage- and time-dependent nonlinear capacitance. Expression of voltage sensor domain-containing proteins like VSFPs therefore contributes an extra capacitance to the membrane. Depending on its magnitude and activation time course, this extra capacitance may influence the subthreshold and spiking properties of target membranes. The physiological effect of extra mobile membrane charge can be inferred from the behavior of excitable membranes after uptake of lipophilic anions, like dipicrylamine (DPA), causing a dose-dependent amplification of membrane capacitance and, at higher doses, decrement of action potential amplitude, increased action potential width, slowing of spike propagation (Fernández et al., 1983), and reduced field potentials (Chanda et al., 2005). As the biophysical mechanism is fairly basic, simulation models can be used to predict and delimit the effects of extra mobile charges on the electrical response properties of excitable membranes (Fernández et al., 1983; DiFranco et al., 2007).
In this review we have focused on the most advanced approach for FP voltage sensor engineering based on a fusion between a voltage sensor protein and an FP reporter domain. However, an alternative design of an FP voltage sensor might be to use the FP as both the sensor and reporter. Such a protein would have the smallest possible size and minimal interaction with other organism-intrinsic proteins. To achieve this, the FP would need to be localized at the water-lipid interface of the plasma membrane or even largely within the lipid phase. Sensitivity to membrane voltage could be based on either voltage-dependent changes in conformation and lipid–protein interactions or be a direct effect of the transmembrane electric field on the optical properties of the chromophore (Molecular Stark effect). The challenge, however, is to change the amino acid composition of the FP molecule so that it would intrinsically translocate to the hydrophobic phase without interruption to the maturation, constitution, and existence of a functional chromophore.
In conclusion, FP sensors will open paths to yet unexplored territories of functional neuroimaging. In addition to the FP voltage sensors on which this review focuses, there are now many different FP sensors for monitoring biochemical processes and transmitter dynamics (Tsien 2005), which can be applied to neuronal circuits. Reading neuronal circuit activity can also be complemented with emerging optical methods to write activity into neuronal circuits (Miesenbock and Kevrekidis, 2005; Boyden et al., 2005; Knöpfel 2008). In the future these techniques may be combined to yield a bidirectional optical interface to brain function.
The authors thank Dr. Steven Middleton for careful commenting on this article. This article was supported by NIH grant NS057631 (LBC, EYI, VAP, TH, TK) and an intramural grant from RIKEN BSI (TK).
This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.