Optogenetic Tools for Subcellular Applications in Neuroscience

Summary

The ability to study cellular physiology using photosensitive, genetically encoded molecules has profoundly transformed neuroscience. The modern optogenetic toolbox includes fluorescent sensors to visualize signaling events in living cells and optogenetic actuators enabling manipulation of numerous cellular activities. Most optogenetic tools are not targeted to specific subcellular compartments but are localized with limited discrimination throughout the cell. Therefore, optogenetic activation often does not reflect context-dependent effects of highly localized intracellular signaling events. Subcellular targeting is required to achieve more specific optogenetic readouts and photomanipulation. Here we first provide a detailed overview of the available optogenetic tools with a focus on optogenetic actuators. Second, we review established strategies for targeting these tools to specific subcellular compartments. Finally, we discuss useful tools and targeting strategies that are currently missing from the optogenetics repertoire and provide suggestions for novel subcellular optogenetic applications.

Main Text

The use of light-sensitive proteins to observe and control cellular activities is considered one of the most groundbreaking innovations in the field of neuroscience in recent years and is starting to replace conventional methods for probing neuronal functions (Scanziani and Häusser, 2009). Deisseroth et al. (2006) first coined the term “optogenetics” to describe genetically targeted photoreceptor expression in neurons for their selective activation or inhibition with light. The term was later extended to other genetically encoded photosensitive proteins, including both actuators (which control the activity of the neuron in a light-dependent manner) and sensors (which monitor neuronal activity) (Dugué et al., 2012, Miesenböck, 2009). Over the past decade, optogenetics thrived in the neurosciences for several reasons. First, optogenetic manipulations can match the speed of electrical neuronal activity, offering, in principle, full control over neuronal activity using closed-loop optogenetics, wherein the light stimulation pattern is automatically updated based on the difference between desired and measured output functions (Grosenick et al., 2015). Second, optogenetics enables an experimental intervention to be limited to specific time windows but also allows for repetitive experiments to be performed over long timescales with defined stimulation intensities. Third, genetically defined, cell-type-specific expression of optogenetic probes has been improved to a level where neuronal circuits can be more thoroughly untangled and the activity of dispersed cell populations controlled or monitored in situ.

Besides enabling the manipulation of neuronal excitability and network activity in vitro and in vivo, various optogenetic tools also allow one to target intracellular processes that are confined to specific compartments or organelles (Figure 1). Given the morphological complexity of neurons and their functional compartmentalization and polarization, the targeting of optogenetic tools with this level of specificity provides a powerful approach for investigating subcellular physiology, providing detailed insights into functional and computational processes performed by the cell. Numerous combinations of fluorescent sensors and compartment-specific protein tags have been employed, enabling more precise labeling of structures of interest compared with chemical reporter dyes. Neuroscientists have only just begun to transfer these targeting strategies to optogenetic actuators to actively control subcellular processes. Recent studies have demonstrated the promise and benefits of compartment-specific optogenetic interrogations. These include re-localization of organelles (van Bergeijk et al., 2015) and neurotransmitter receptors (Sinnen et al., 2017), control of spine size (Hayashi-Takagi et al., 2015), and light-activated filling of synaptic vesicles (Rost et al., 2015).

In this review, we provide a detailed overview of the optogenetic tools that are currently available. We then summarize established subcellular targeting strategies for both optogenetic actuators and sensors that have been applied in neuroscience. Finally, we provide suggestions for future developments and applications of subcellular optogenetic tools and offer a perspective on the future of this rapidly evolving field.

Optogenetic Tools: Actuators

Sensory Photoreceptors

Figure 1
Figure 2
Figure 3
Figure 4

Light-sensitive proteins, known as sensory photoreceptors, adapt cellular activity to ambient light conditions. They are present in all kingdoms of life and consist of seven main protein classes: rhodopsins, xanthopsins or photoactive yellow proteins, light-oxygen-voltage (LOV) sensors, blue light-sensors that use flavine adenine dinucleotide (BLUF), cryptochromes, phytochromes, and UV-B receptors (UVR8) (Figure 2). UVR8 receptors absorb UV-B light via intrinsic tryptophan clusters. All other identified photoreceptors bind organic compounds called chromophores that exhibit delocalized electrons distributed across conjugated π-electron systems, which allow for light absorption in the range between 300 and 800 nm. Photon absorption triggers primary photochemical reactions usually leading to conformational changes in the protein that are eventually propagated to the respective effector domains. Depending on the photoreceptor, signal transduction occurs within a sensor domain that unifies sensory and effector functions (e.g., microbial ion pumps, channelrhodopsins) or involves signal propagation from the light sensor to a secondary domain of the same protein (e.g., as in LOV, BLUF proteins, and phytochromes) or signaling to distinct interacting protein partners (as with visual rhodopsins). The palette of natural light-sensing proteins is complemented by synthetically engineered photoreceptors that are obtained via modification and recombination of existing receptors to produce modified or completely novel light-activated outputs. In the following section, we introduce the photoreceptors used for optogenetic experiments.

Membrane Voltage Modulation Using Microbial Rhodopsins

Type I rhodopsins, also referred to as microbial rhodopsins (Figure 2A), originate from archaea, bacteria, algae, and fungi, where they generate membrane ion gradients for energy production or have photosensory functions, such as mediating phototactic and photophobic responses. These heptahelical (seven transmembrane) proteins are characterized by light-induced isomerization of their all-trans retinal chromophore, followed by reorientation of the Schiff-base lysine, deprotonation of the Schiff base, and/or distinct conformational changes within the opsin, which ultimately induce protein activation. In optogenetics, ion-transporting rhodopsins, such as light-driven ion pumps and light-gated ion channels, are used to modulate the membrane potential of target cells.

In microbial ion pumps, photon absorption triggers active transport of H⁺ or Na⁺ from the cytosol to the extracellular space or transport of Cl⁻ in the opposite direction. Proton and chloride pumps, as well as recently identified Na⁺ pumps, are applied to hyperpolarize the neural plasma membrane in response to light stimulation, increasing the electrical threshold necessary to trigger action potentials (APs) (Chow et al., 2010, Han and Boyden, 2007, Inoue et al., 2013, Kato et al., 2015, Zhang et al., 2007) and suppressing neurotransmitter release (El-Gaby et al., 2016, Mahn et al., 2016). Because only one ion per absorbed photon is transported within the photocycle, effective hyperpolarization demands high light intensities together with high protein expression levels in the membranes. Optogenetic neuronal silencing has been most commonly achieved using the proton pump archaerhodopsin (Arch) from Halorubrum sodomense or the related archaerhodopsin from the Halorubrumstrain TP009 (ArchT) (Chow et al., 2010, Han et al., 2011, Ihara et al., 1999). Chloride-pumping halorhodopsins also achieve neuronal AP inhibition by membrane hyperpolarization with an efficiency that is similar to that of proton pumps but without intra- or extracellular pH changes (Han and Boyden, 2007, Zhang et al., 2007). Because of its superior expression level in neurons, Natronomonas pharaonishalorhodopsin (NpHR) has been applied for neuronal inhibition in vitro and in vivo, including in zebrafish, rodents, and non-human primates (Arrenberg et al., 2009, Diester et al., 2011, Gradinaru et al., 2008). The recently identified green light-driven Na⁺ pump KR2 from Krokinobacter eikastus (Inoue et al., 2013) has been applied to hyperpolarize cultured rat cortical neurons and inhibited C. elegans locomotion during pan-neuronal activation (Kato et al., 2015). Interestingly, mutations in its selectivity filter converted this pure Na⁺ pump into a mixed Na⁺/K⁺ pump. KR2 is an interesting inhibitory tool because it can hyperpolarize neuronal plasma membranes without changing H⁺ and Cl⁻ concentrations. However, the potency of this new inhibitory rhodopsin still requires further testing in different neuronal populations.

The optogenetic counterparts of microbial ion pumps are Channelrhodopsins (ChRs), light-gated ion channels that passively conduct ions along electrochemical membrane gradients (Nagel et al., 2002, Nagel et al., 2003). Of these, Channelrhodopsin-2 (ChR2) from Chlamydomonas reinhardtii is the prototype for “depolarizing” ChRs, which are used to elicit neural firing following defined photostimulation patterns (Boyden et al., 2005, Ishizuka et al., 2006). In fact, ChR2 and related ChRs are highly selective for H⁺ but conduct approximately equal amounts of H⁺ and Na⁺ under physiological conditions. In pioneering studies, ChR2 was used to induce synaptic plasticity (Zhang et al., 2007) and to map neuronal circuits (Petreanu et al., 2007), and it has since been applied to tackle numerous neuroscientific questions in different model organisms (Arenkiel et al., 2007, Li et al., 2005, Nagel et al., 2005, Schroll et al., 2006).

Several ChR variants have been identified, both by engineering approaches and genome mining, that broaden the range of optogenetic applications (for a review, see Schneider et al., 2015). In short, ChR2-H134R and ChR2-T159C exhibit increased photocurrents because of improved folding and retinal binding (Berndt et al., 2011, Nagel et al., 2005). Mutating ChR2 C128 and/or D156 of the retinal-binding pocket decelerates photocycle kinetics, generating variants with long-lived open states (Bamann et al., 2010, Berndt et al., 2009, Dawydow et al., 2014, Yizhar et al., 2011). These so-called step-function rhodopsins (SFRs) have the unique advantage that they are activated with a short light pulse of one color (e.g., blue light), remain open for a prolonged time, and can be closed with light of another color (e.g., UV or red light). SFRs are complemented by fast-cycling ChRs, such as ChR2-E123T (ChETA), ChR2-E123T/T159C (ChETA_TC), and the ChR from Stigeoclonium helveticum (Chronos), which allow one to trigger APs at frequencies of up to 200 Hz because of their fast and mostly voltage-independent photocycle kinetics (Berndt et al., 2011, Gunaydin et al., 2010, Klapoetke et al., 2014). A selection of ChRs with distinct spectral properties also allows two neural populations to be independently activated by light of different wavelengths. Examples include the blue-shifted ChR from Platymonias subcordiformus (PsChR) and Tetraselmis striata(TsChR) and the green to orange light-activated variants C1V1, ReaChR, bReaChES, and Chrimson (Govorunova et al., 2013, Klapoetke et al., 2014, Lin et al., 2013a, Prigge et al., 2012, Rajasethupathy et al., 2015, Yizhar et al., 2011). Other remarkable ChR variants include the H⁺-selective ChR of Dunaliella salina(DChR1) as well as ChR2 L132C (CatCh), a ChR2 mutant with weak inactivation and increased Ca²⁺ and Mg²⁺ conductance (Kleinlogel et al., 2011, Schneider et al., 2013). In contrast to cation-selective ChRs, the recently engineered Cl⁻-conducting ChRs (ChloC and iC1C2) effectively inhibit APs by clamping the membrane potential to the reversal potential of Cl⁻ (Berndt et al., 2014, Wietek et al., 2014). Structure-based protein engineering provided second-generation Cl⁻-conducting ChRs with increased protein expression levels and Cl⁻ conductance, such as improved ChloC (iChloC) and the chimeric light-activated chloride channels iC++ and SwiChR++ (Berndt et al., 2016, Wietek et al., 2015). Furthermore, engineered Cl⁻ channels are complemented by natural anion-selective ChRs (ACRs) from Guillardia theta and Proteomonas sulcata, which show large conductance upon light activation (Govorunova et al., 2015, Govorunova et al., 2016, Wietek et al., 2016).

In summary, the ensemble of light-activated pumps and channels allows the neuronal membrane potential to be modulated in diverse ways. Because of the small size of their protein-encoding genes, the availability of all-trans retinal in mammalian neurons, as well as their activation and deactivation kinetics on the millisecond timescale, microbial rhodopsins represent by far the most commonly used optogenetic actuators, and their application to cell-type-specific optical interrogations of neuronal activity has revolutionized neuroscience research.

Rhodopsins and Optical G Protein Activation

Type II rhodopsins are light-activated G protein-coupled receptors (GPCRs) that initiate cellular signaling cascades, mediating visual perception in animals (Figure 2A).

Khorana et al. (1988) showed that light activation of bovine rhodopsin expressed in Xenopus oocytes evoked inward-directed photocurrents, but the underlying mechanism at the time was not known. Zemelman et al. (2002) later established the first optogenetic system (ChARGe) to depolarize cultured hippocampal neurons by coexpressing Drosophila rhodopsin NinaE, arrestin-2, and the alpha subunit of the cognate heterotrimeric G protein (Zemelman et al., 2002). Because of its complex nature and slow kinetics, ChARGe was soon thereafter replaced by the single-gene ChR2 as an excitatory tool. In contrast, heterologously expressed vertebrate rhodopsin 4 (Ro4) reduces neuronal excitability by activating postsynaptic K⁺currents and inhibiting presynaptic Ca²⁺ currents via the intrinsic G_i/o protein pathway. The combined use of ChR2 and Ro4 to synchronize neuronal activity in the spinal cord of living chicken embryos formed the first reported in vivo optogenetic experiment (Li et al., 2005).

While Ro-4 responses decline during sustained or repeated light stimulation, short- and long-wavelength rhodopsins from visual cones enable repetitive photoactivation of G_i/o protein signaling (Karunarathne et al., 2013a, Karunarathne et al., 2013b, Masseck et al., 2014). Alternatively, invertebrate rhodopsins, such as the box jellyfish opsin (JellyOp), can be used to repetitively activate G_s signaling in mammalian cells (Bailes et al., 2012). Belonging to the group of melanopsins, Opn4 is a bistable photoreceptor found in specialized, intrinsically photosensitive retinal ganglion cells. Opn4 variants from mouse and human can be turned on and off with blue and yellow light, respectively, but show distinct light sensitivity and inactivation kinetics of their photoresponses (Spoida et al., 2016). Pioneering applications of Opn4 in mice include vision restoration in non-intrinsically photosensitive retinal ganglion cells and the activation of hypothalamic orexin neurons that control sleep/wakefulness cycles (Lin et al., 2008, Tsunematsu et al., 2013).

OptoXRs are chimeric GPCRs consisting of a rhodopsin, the intracellular loops of which are replaced by the corresponding loops of a ligand-activated GPCR, creating a light-activated GPCR with new G protein-coupling specificity (Figure 2A). Examples of OptoXRs include the β2 adrenergic receptor-like opsin-GPCR and the α1a adrenergic receptor-like opsin-GPCR, which selectively control G_s and G_qsignaling, respectively (Airan et al., 2009, Kim et al., 2005). Following a similar design, the chimeric rhodopsin Crblue was created by replacing the intracellular loops of the blue cone opsin with the corresponding regions from JellyOp, resulting in a blue light-activated G_s-coupling receptor (Karunarathne et al., 2013a). In a further advance, Siuda et al. (2015) created photosensitive μ-opioid-like receptors (opto-MORs) that resemble canonical μ-opioid peptide receptors in their subcellular localization, internalization, and signaling properties and applied these to alter preference behavior in mice in vivo. The optoXR strategy has also been extended to the vertebrate melanopsin OPN4, which was combined with intracellular domains of the metabotropic glutamate receptor type 6 (mGluR6). Activation of the resulting opto-mGluR6 in retinal ON-bipolar cells of blind mice enabled vision restoration at ambient light intensities (van Wyk et al., 2015). Notably, the optoXR concept has been successfully applied to a variety of rhodopsins and GPCRs, thus representing a general strategy, and we await new developments in this direction.

Following a slightly different approach, Oh et al. (2010) added the C-terminal domain of the serotonin receptor 5-HT_1A to vertebrate rhodopsin to create a light-activated serotonin (5-hydroxytryptamine)-like receptor (Rh-CT_5-HT1A). Analogous to intrinsic 5-HT_1A receptors, Rh-CT_5-HT1A was selectively targeted to somatodendritic sites, where it induced membrane hyperpolarization via the activation of G protein-coupled inwardly rectifying K⁺ (GIRK) channels, thus functionally compensating for loss of 5-HT_1A (Oh et al., 2010). To elucidate the function of endogenous 5-HT_2C receptors, Spoida et al. (2014) expressed vertebrate melanopsin carrying the C terminus of 5-HT_2C (vMo-CT_5-HT2c) in GABAergic neurons of the dorsal raphe nuclei in mice. When activated by blue light, vMo-CT_5-HT2c triggered G_q-mediated Ca²⁺ influx, activating GABAergic cells and inhibiting the firing of postsynaptic serotonergic neurons, which reduced anxiety-related behavior in mice in vivo.

Thus, the OptoXR concept for G protein activation opens new experimental routes to selectively trigger intracellular signaling cascades. We envision further applications of this approach in the CNS to both manipulate cellular activity and to unravel the role of GPCR signaling in different brain states, behavior, and disease.

Modulating Second Messenger Concentrations with Light-Activated Enzymes

Complementary to GPCRs, several light-activated enzymes allow the concentration of cyclic nucleotide second messengers to be controlled (Figure 2B). Photoactivated adenylyl cyclases (PACs), such as euPAC from Euglena gracilis, bPAC from Beggiatoa, and OaPAC from Oscillatoria acuminate, are modular photoreceptors that consist of blue light-sensing BLUF domains coupled to catalytic domains that produce cyclic AMP (cAMP). The PACα subunit of euPAC has been used to modulate cAMP levels in sensory neurons of Aplysia, in different neuronal populations of Drosophila melanogaster, in cholinergic neurons of Caenorhabditis elegans, and in rat dentate gyrus granule cells (Bellmann et al., 2010, Nagahama et al., 2007, Schröder-Lang et al., 2007, Weissenberger et al., 2011, Zhou et al., 2016). bPAC, which is 60% smaller and more soluble relative to PACα, was first applied to activate cyclic nucleotide-gated (CNG) channels in hippocampal slice cultures and to alter grooming activity in bPAC-transgenic fruit flies in vivo (Stierl et al., 2011). It has also been expressed specifically in the pituitary cells of freely behaving zebrafish larvae to enhance the rise of endogenous cortisol levels triggered by stress (De Marco et al., 2013) and in mouse sperm to complement adenylyl cyclase deficiency and recover motility in a light-dependent manner (Jansen et al., 2015). However, two important caveats for nearly all photoreceptors with flavin-based chromophores are their significant dark activity, which might modulate cell physiology or animal behavior prior to the actual experimental interventions, and the inability to shift the absorption spectra by engineering.

The rhodopsin from Blastocladiella emersonii (RhGC, also referred to as BeCyclOp; Figure 2A) belongs to the recently discovered group of membrane-spanning, light-activated guanylyl cyclases (Avelar et al., 2014, Gao et al., 2015, Scheib et al., 2015). Upon green light activation, this unusual rhodopsin catalyzes cyclic guanosine monophosphate (cGMP) synthesis with millisecond precision but does not show detectable dark activity. Thus, it represents a powerful tool for modulating cGMP levels and has been used to activate cGMP-gated ion channels in rat hippocampal neurons and C. elegans myocytes (Gao et al., 2015, Scheib et al., 2015). The selection of light-activated enzymes is complemented by an engineered light-activatable cAMP/cGMP-phosphodiesterase (LAPD; Figure 2B), in which a light-sensing domain from a bacterial phytochrome is hybridized to the catalytic domain of the human phosphodiesterase 2A (Gasser et al., 2014). So far, LAPD has been employed to reduce cyclic nucleotides in Chinese hamster ovary (CHO) cells and in zebrafish embryos; however, a neuroscientific application has yet to be reported. Together, light-activated cyclases and phosphodiesterases are single-component tools that can be used to photoadjust intracellular second messenger concentrations. Their use requires thorough controls to exclude effects from the enzyme’s dark activity, which can be minimized by the use of rhodopsin-coupled enzymes instead of flavin or phytochome photoreceptor modules.

LOV Protein-Based Tools for Protein Activation and Inactivation

LOV domains are photosensory modules found in photoreceptors from plants, bacteria, fungi, and algae (Crosson et al., 2003, Huala et al., 1997, Salomon et al., 2000). Upon light absorption, the flavin mononucleotide (FMN) forms a transient thioadduct of a cysteine with the C4a position of the flavin. The resulting conformational change in the LOV domain is transmitted to the catalytic domains via the connecting linkers, resulting in structural alterations and activation of the catalytic domains. In optogenetics, LOV domains coupled to non-endogenous effector proteins enable blue light-mediated modulation of protein accessibility and activity (Figure 2B).

In an early approach to engineer a blue light-activated enzyme, Lee et al. (2008)created chimeras in which the LOV2 domain from Avena sativa phototropin was inserted via its N- and C-terminal helical extensions into E. coli dihydrofolate reductase. However, the resulting light-mediated control of the protein’s enzymatic activity was low in these early variants (Lee et al., 2008). Shortly afterward, Wu et al. (2009) developed photoactivatable Rac1 (PaRac1) by fusing the LOV2 domain and the helical linker (Jα) from phototropin as an allosteric activator for guanosine triphosphate hydrolase (GTPase) Rac1 mutants. In the dark, the LOV domain sterically inhibits PaRac1, preventing it from interacting with effector proteins. Upon blue light exposure, the GTPase activity of PaRac1 is restored, enabling investigations into the role of Rac1 in the regulation of actin cytoskeletal dynamics. Notably, Wu et al. (2009) highlighted linker length to be of critical importance because the addition or removal of a single residue abolished photocontrol. So far, the PaRac1 design has not been used to photocontrol other enzymes. In follow-up studies, PaRac1 has been successfully applied to analyze cocaine-induced structural plasticity of nucleus accumbens neurons (Dietz et al., 2012), to dissect cytoskeletal modulation of postsynaptic potentiation in vitro (Schwechter et al., 2013), and to selectively shrink potentiated spines in vivo (Hayashi-Takagi et al., 2015). Other chimeric proteins consisting of different LOV domains, linkers, and effector proteins have been constructed to create a light-controlled histidine kinase (Möglich et al., 2009), light-activated molecular motors (Nakamura et al., 2014), and a light-regulated tetracycline (Tet) repressor (Moon et al., 2014). However, to our knowledge, to date, none of these tools has been applied in neurons. A photoactivatable Ca²⁺-releasing protein, created by the insertion of LOV2 into the Ca²⁺-binding calmodulin-M13 fusion protein (photoactivatable Ca²⁺ releaser/PACR), might nevertheless prove to be a useful candidate for altering Ca²⁺ concentrations in neurons (Fukuda et al., 2014). Also acting on cellular ion concentrations, but based on a different design strategy, Schmidt et al. (2014) coupled a membrane-anchored LOV/Jα motif to peptide toxins (so-called lumitoxins), enabling photomodulation of voltage-gated K⁺ channel conductances. Moreover, a direct fusion of LOV/Jα with the miniature K⁺ channel pore K_cv (BLINK1) from a Chlorella virus enabled light-enhanced K⁺ currents in cultured cells and zebrafish embryos (Cosentino et al., 2015). However, using BLINK1 as a reliable tool to suppress APs in excitable cells will require an improved variant that possesses enhanced plasma membrane targeting combined with reduced dark activity and faster photocycle kinetics.

An interesting LOV-based optogenetic tool enables light-induced production of singlet oxygen (miniSOG) (Lin et al., 2013b, Shu et al., 2011). MiniSOG is useful for both local polymerization of diaminobenzidine that can be visualized by electron microscopy as well as for specific protein inactivation (see below for a further discussion of miniSOG applied to inactivate synaptic transmission).

Optogenetic Tools to Modulate Protein-Protein Interactions

Light-activated changes to the oligomeric states of homo- and heteromeric protein complexes are common biological phenomena and have been applied in various optogenetic experiments to study protein-protein interactions and protein recruitment to specific subcellular compartments (Figures 1 and 2C and 2D). Phytochromes, LOV proteins, cryptochromes, UVR8, and fluorescent protein-derived interaction domains have been generated to enable photocontrol of protein oligomerization, each offering specific advantages and limitations with respect to the experimental design. Important criteria include activation wavelength and chromophore availability, low dark activity, reversibility, and suitable binding affinities.

Phytochromes. Phytochromes are plant photoreceptors that use covalently bound, linear tetrapyrroles as chromophores. Photoisomerization of the chromophore triggers an allosteric transition between two conformational states (red-absorbing and far red-absorbing in classic phytochromes). Red light-induced binding of phytochrome B (PhyB, Arabidopsis thaliana) to its natural interaction partner, phytochrome interaction factor 3 (PIF3), was the first system to enable light-controlled protein translocation. This approach has also been applied to recruit proteins to the plasma membrane, nucleus, peroxisomes, and endosomes as well as to mitotic cellular structures (Levskaya et al., 2009, Toettcher et al., 2013, Yang et al., 2013).

Despite the spectral window for phytochrome activation and deactivation, together with their bistable nature, phytochrome-based optogenetic systems have rarely been applied in multicellular organisms. In animal cells, the obligate bilin chromophores require either synthesis via one or two modification steps from heme or supplementation, hampering in vivo application (Müller et al., 2013c). Recently, two publications demonstrated the applicability of PhyB-PIF-6 in zebrafish embryos. Beyer et al. (2015) used a PhyB-PIF3-based system to induce nuclear protein import in zebrafish embryos in vivo. They applied phycocyanobilin (PCB) to the culture medium, which enabled its diffusion into dechorinated embryos at levels sufficient to trigger nuclear protein import using whole-embryo illumination (Beyer et al., 2015). Buckley et al. (2016) injected PCB into 16-cell stage zebrafish embryos, where it complemented a membrane-bound PhyB, allowing for red light-induced recruitment of PIF-6-tagged target proteins. Spatially restricted illumination then enabled asymmetric protein inheritance in mitotic cells of the neuroepithelium.

LOV Proteins. Yazawa et al. (2009) established the LOV protein FKF1 and the interacting GIGANTEA (Arabidopsis thaliana) (Figure 2C) as protein tags for blue light-activated dimerization and demonstrated light-induced membrane recruitment of Rac1 triggering lamellipodia formation. FKF1 and GIGANTEA were also applied to photostimulate the oligomerization of FKF1-tagged L-type Ca_v1.2 channels to increase Ca²⁺ influx into cardiomyocytes, enhancing excitation-contraction coupling (Dixon et al., 2012), an interesting approach not applied to neurons so far. Although the FKF1-GIGANTEA system uses the ubiquitously available FMN chromophore, its application is hampered by the large size of the two interacting proteins as well as by its slow on and off responses in the minute and hour range.

A different strategy was pursued by Lungu et al. (2012), who used the AsLOV2 domain from Avena sativa phototropin to photomodulate the affinity of peptides for specific binding partners. Similarly, Niopek et al. (2014) established light-inducible nuclear localization signals (LINuSs). In their design, LOV2-Jα-caged nuclear localization sequences become accessible upon blue light exposure, thus binding the corresponding endogenous importins to trigger nuclear protein import (Niopek et al., 2014). A complementary light-inducible nuclear export system (LEXY) consists of AsLOV2 coupled to a mutated J_α helix, which acts as an artificial nuclear export signal that is exposed upon illumination, inducing nuclear export of tagged proteins. Finally, a constitutively nuclear chromatin-anchored LEXY variant enables photoinhibition of endogenous nuclear protein export (Niopek et al., 2016).

Strickland et al. (2012) developed tunable light-induced dimerization tags (TULIPs) based on AsLOV2 fused to a peptide epitope (LOVpep) and an engineered PDZ domain (Figure 2C). Different mutants of the TULIP-interacting pair provide a range of binding affinities and different kinetic constants. TULIPs have been used to enable light-induced recruitment of proteins to various cellular compartments to regulate the activity of nucleotide exchange factors and scaffold proteins as well as kinases (Strickland et al., 2012). Notably, TULIP-like tools have been shown to effectively relocate entire organelles, including peroxisomes, mitochondria, and recycling endosomes (van Bergeijk et al., 2015; see also Optogenetic Control of Organelle Positioning). The repertoire of AsLOV2-based peptide interaction systems has recently been extended by the improved light-induced dimer (iLID) system, which displays a high dynamic range of activation, and variants with nanomolar to micromolar binding affinities (Guntas et al., 2015).

Yet another LOV domain-based dimerization system with tunable kinetics, called Magnets, has been developed by Kawano et al. (2015), who created two light-induced heterodimerizing isoforms of the small Vivid photoreceptor. Pilot applications of this system include Magnet-based membrane recruitment and phosphatidylinositol 3-kinase (PI3K) activation (Kawano et al., 2015).

By contrast, Wang et al. (2016) have created an optogenetic system for the photoinduction of protein dissociation (called LOV2 trap and release of protein [LOVTRAP]). This system is based on a modified version of the Z domain of the immunoglobulin-binding staphylococcal protein A (ZdK), which tightly binds to AsLOV2 in the dark. Interestingly, the LOVTRAP system showed more than a 150-fold change in the dissociation constant upon blue light activation and allowed for the reversible release of proteins trapped at the outer mitochondrial membrane, with time constants of return kinetics that were tunable between 2 s and 500 s (Wang et al., 2016).

As of today, many LOV-based tools to induce protein-protein interactions and guide protein localization have been developed. However, only a few of these have been applied in neurons, and their full potential for subcellular optogenetics has yet to be exploited.

Cryptochromes. Cryptochromes (Crys) are photoreceptors with a flavin adenine dinucleotide (FAD) chromophore that is characterized by blue light-induced reduction. Crys are the most widely distributed photoreceptor class found in animals, plants, algae, and bacteria. Light-triggered homo- and heterodimerization of plant Cry2 (Arabidopsis thaliana) has been applied in a number of optogenetic experiments.

Cry2-CIB Dimerization Systems. Kennedy et al. (2010) developed a subcellular protein recruitment system based on Cry2 and its interaction partner, the Arabidopsis cytochromes-interacting basic helix-loop-helix protein 1 (CIB1), or the shorter N-terminal variant (CIBN). They also conceived the so-called split protein approach, wherein the protein of interest is expressed as two inactive fragments that reconstitute a functional protein upon light absorption.

In follow-up studies, Cry2 and CIBN enabled recruitment of different enzymes to the plasma membrane. For example, the blue light-mediated membrane targeting of an inositol 5-phosphatase led to dephosphorylation of phosphatidylinositol 4,5-diphosphate or phosphatidylinositol 3,4,5-trisphosphate (PIP₂ or PIP₃, respectively), which resulted in the disappearance of clathrin-coated pits as well as in inhibition of voltage-gated K⁺ channels in non-neuronal cells (Idevall-Hagren et al., 2012). In contrast, membrane recruitment of a PI3K triggered local PIP₃ production in cultured hippocampal neurons, inducing F-actin polymerization and growth cone expansion but not neurite elongation (Kakumoto and Nakata, 2013). In another study, the Cry2-tagged Raf1 kinase was activated by membrane relocation (Figure 3A), activating the Raf/MEK/ERK signaling pathway, leading to significant neurite outgrowth in a rat PC12 pheochromocytoma cell line (Zhang et al., 2014). Furthermore, localized membrane recruitment of the GTPase-accelerating protein RGS4 induced rapid inactivation of G protein α subunit signaling, whereas recruitment of the C-terminal domain of G protein-coupled receptor kinase 2 (GRK2ct) inhibited Gβγ signaling via sequestration, allowing investigators to study the influence of G protein gradients on cell migration in a macrophage cell line (O’Neill and Gautam, 2014).

Notably, photo-induced association of organellar membrane-anchored Cry2 with motor protein-coupled CIBN enabled directional transport of entire organelles, including mitochondria, peroxisomes, and lysosomes (Duan et al., 2015). Dimerization of Cry2-CIB1 also empowered a system called exosomes for protein loading via optically reversible protein-protein interactions (EXPLORs), which enables light-controlled loading of exosomes with any Cry2-conjugated cargo protein (Yim et al., 2016).

Recently, Taslimi et al. (2016) identified truncated versions of Cry2 and CIB1 with decreased dark activity as well as slow- and fast-cycling Cry2 mutants (Cry2 L348F and Cry2 W349F, respectively) that may be useful for future Cry2-CIB1 interaction experiments.

Homooligomerization of CRY2 and Combinational Systems

Blue light-mediated Cry2 homodimerization can be used to reversibly dimerize coupled target proteins (Figure 1D). Accordingly, Chang et al. (2014) coupled the photolyase homology region (PHR) of Cry2 to the brain-derived neurotrophic factor (BDNF) receptor, TrkB, to mimic ligand-induced receptor dimerization with light. The resulting optoTrk could still be activated by its natural ligand BDNF and was used to photoactivate PI3K signaling, regulating neurite outgrowth and filopodium formation in rat hippocampal neurons. A PHR dimerization approach was also used to construct a light-controlled fibroblast growth factor receptor (optoFGFR1) (Kim et al., 2014).

Similarly, several groups have applied light-induced oligomerization of Cry2 to reversibly control target protein clustering. Light-triggered clustering of low-density lipoprotein receptor-related protein 6 (LRP6) activated the Wnt/β-catenin signaling pathway, whereas oligomerization of Cry2-coupled GTPase Rac1 increased enzymatic activity (Bugaj et al., 2013). In a follow-up study, Cry2 fused to receptor-targeting binding domains allowed clustering and activation of endogenous receptors of the receptor tyrosine kinase class to be stimulated by light, as shown for a fibroblast growth factor receptor, a platelet-derived growth factor receptor, and a β-integrin receptor (Bugaj et al., 2015). Similarly, Cry2-mediated oligomerization of STIM1 (optoSTIM1) activated endogenous Ca²⁺ release-activated Ca²⁺ (CRAC) channels, enhancing local Ca²⁺ concentrations in zebrafish embryos and in human embryonic stem cells (Kyung et al., 2015). Notably, optoSTIM1-induced Ca²⁺ entry in mouse hippocampal neurons was shown to selectively reinforce contextual memory formation in vivo (Kyung et al., 2015).

Taslimi et al. (2014) reported a Cry2 mutant (Cry2 E490G, Cry2olig) with enhanced clustering capability that was effective for localized induction of actin polymerization and for inhibition of clathrin-mediated endocytosis by light.

Combining Cry2 homo-oligomerization with Cry2-CIB1 heterodimerization can trigger the formation of large protein clusters via photoactivation of co-expressed Cry2 and a CIB1-tagged multimeric protein. This system has been used for light-induced protein inhibition, termed light-activated reversible inhibition by assembled trap (LARIAT) (Lee et al., 2014). Very recently, the light-induced binding of Cry2 to CIB1-Rab GTPase fusion proteins was used to reversibly induce intracellular membrane aggregations (IM-LARIAT), inhibiting specific membrane trafficking pathways dependent on the employed Rab GTPase (Nguyen et al., 2016).

Taken together, Cry2- and CIB1-based systems are versatile tools for blue light-mediated oligomerization and clustering of target proteins, but again, a generalization of these concepts would greatly facilitate their application by neuroscientists. A side-by-side comparison of the existing tools and their interaction properties in neurons would be a first step toward this goal.

UVR8. The UV-B receptor UVR8 is a photosensor that is involved in stress signaling in plants. Upon absorption of UV-B (280–315 nm) light, homodimeric UVR8 receptors dissociate, enabling their binding to their interaction partner COP1, an E3 ubiquitin ligase, to control transcription in the nucleus. In 2013, several teams reported optogenetic experiments based on UVB-induced UVR8 dissociation or UVR8-Cop1 interaction (Figures 2C and 2D; Chen et al., 2013a, Crefcoeur et al., 2013, Müller et al., 2013b). Dimerization of UVR8 coupled to target proteins was enhanced by using UVR8 tandems with two UVR8 moieties joined by a flexible peptide linker, as shown for a fusion construct of UVR8 and histone H2B (Crefcoeur et al., 2013). A tandem approach, using a target glycoprotein coupled to several copies of UVR8 in series, has also been developed to create a UV light-driven protein secretion system. In the dark, UVR8 oligomerization retains complexes of coupled proteins in the endoplasmic reticulum (ER). Upon exposure to UV light, UVR8 dissociation is triggered, and protein secretion occurs. This system enabled light control of local secretory trafficking in the dendrites of primary hippocampal neurons (Chen et al., 2013a).

Fluorescent Protein-Based Interaction Tools. Photochromic fluorescent proteins can be switched between a fluorescent and a non-fluorescent state using light of two different wavelengths. They were originally optimized for super-resolution microscopy techniques that rely on stochastic activation of single fluorophores. Zhou et al. (2012) described the K145N mutant of the GFP Dronpa, which forms fluorescent tetramers upon exposure to violet light, with cyan light switching off fluorescence and causing monomerization. Light-induced switching between oligomeric states was used to recruit the red fluorescent mNeptune protein to the plasma membrane. Furthermore, C-and N-terminal flanking of target proteins by the Dronpa mutant enabled light-induced protein activation (creating so called fluorescent light-inducible proteins [FLIPs]) (Zhou et al., 2012).

Researchers can now choose from a variety of photo-controlled dimerization or multimerization systems that are characterized by different activation wavelengths, association and dissociation time constants, affinities, dark activities, and cofactor requirements, all of which need to be considered when planning an experiment (Table 1). Several instructive reviews now provide comprehensive overviews of dimerization tools (e.g., Zhang and Cui, 2015). Moreover, recent studies in yeast (Pathak et al., 2014) and mouse fibroblasts (Hallett et al., 2016) provide direct comparisons of optical dimerizers within a single system; however, a comparable systematic evaluation is still missing for neurons.

Table 1Optogenetic Tools for Controlling Protein-Protein Interactions and Protein Oligomerization

Optogenetic Interaction System, Chromophore and Color of Activation	Advantages (+) and Disadvantages (−) of the Tools
Phytochrome
→ PhyB-PIF3/PIF6	+ bimodal switchable
Bilin chromophore	+ deep tissue penetration of red/far-red light
Activation by red light (660 nm), far-red inactivation (730 nm)	+ color tuning possible using different bilin variants
	− chromophore not ubiquitously available
LOV domain
→ FKF1 and GIGANTEA	+ ubiquitous chromophore availability
→ AsLOV2-peptides	+ tuned variants with different time constants and affinities
→ TULIPs	+ high dynamic range of improved variants
→ Magnets	+ small size of LOV domain
FMN chromophore	− no color tuning
Activation by blue light (470 nm)
Cryptochrome
→ Cry2 and CIB(N)	+ ubiquitous chromophore availability
FAD chromophore	+ tuned variants with different time constants and affinities
Activation by blue light (470 nm)	+ high dynamic range of improved variants
	− large protein size
	− no color tuning
UVR8
→ homodimerization or heterodimerization with Cop1	+ no additional chromophore
Intrinsic tryptophan cluster as chromophore	+ selective activation when combined with phytochromes
Activation by UV-B (280 nm)	+ color-tuned variants absorbing UV-C
	− UV-light induced photodamage
	− irreversible
	+ increased homodimerization affinity when using UVR8 tandems
Fluorescent proteins
→ Dronpa K145N	+ bimodal switchable
Cys-Trp-Gly as chromophore	+ GFP-based: small protein, tunable
UV/cyan (variable)	− UV light for activation
	− low dynamic range
	− only homodimerization

Light-Induced DNA Modification, Transcription, and Protein Expression

In parallel to the development of diverse optogenetic tools for photocontrolled protein-protein interactions, a variety of approaches have emerged that enable light-controlled modification and editing of DNA as well as light-activated transcriptional control and light-induced post-translational modifications. Although these systems have the potential to control the onset of transcription and uncover the effects of initiating protein expression, most have not been used by other laboratories after their initial description. This is partly because they require co-expression of multiple components and because only a few support the activation of endogenous genes with light, for example by using zinc-finger proteins, transcription activator-like effectors (TALEs), or modified Cas9.

Phytochrome-Based Systems. The first optogenetic approaches used to modulate transcription were based on phytochromes. Shimizu-Sato et al. (2002) used the phytochrome-PIF3 pair to recruit the Gal4 activation domain to the Gal4 DNA-binding domain, allowing reversible transcriptional activation in yeast. Using a different strategy, Levskaya et al. (2005) built a chimeric photoreceptor consisting of a cyanobacterial phytochrome (Synechocystis phytochrome, Cph1) and a bacterial histidine kinase (E. coli, EnvZ) that enabled light-induced activation of the corresponding response regulator (OmpR) in E. coli. Phytochrome applications also included post-translational protein modification, as shown for a light-activated protein splicing system in yeast (Tyszkiewicz and Muir, 2008). It was not until 2013 that Müller et al. (2013a) coexpressed PhyB fused to the transactivation domain VP16 and PIF6 coupled to the Tet repressor TetR to photoactivate gene expression in mammalian cells.

LOV Domain-Based Tools. By fusing AsLOV2 to the trp repressor from E. coli, Strickland et al. (2008) developed a light-activated DNA-binding module that selectively protects target DNA from nuclease digestion. Shortly afterward, Yazawa et al. (2009) used their light-activated dimerization system to heterodimerize the DNA-binding domain of Gal4 with the transactivation domain VP16, activating transcription in mammalian cell culture. Polstein and Gersbach (2012) advanced this system by replacing Gal4 with engineered zinc-finger proteins, which allow this tool to be targeted to almost any DNA sequence of interest (an approach called light-inducible transcription using engineered zinc-finger proteins [LITEZ]). This approach offers a significant advantage over other light-activated transcriptional systems that use viral, bacterial, or yeast transcription factors, which require the gene of interest to be expressed under the control of a matching promoter. Wang et al. (2012) fused a truncated Gal4 DNA-binding domain to a mutated Vivid photoreceptor and to the p65 transactivation domain to enable light-induced DNA binding of this tool and subsequent transcriptional activation of a reporter gene in rodents in vivo (LightOn system). Adapting the intrinsically light-sensitive bacterial transcription factor EL222 for mammalian transcriptional activation, Motta-Mena et al. (2014) reported a gene expression system with fast activation and deactivation kinetics in response to light that was applied in zebrafish embryos.

Indirect transcriptional activation has also been achieved by LOV2-regulated nuclear import or export of transcriptional activators or repressors (see also the above section on LOV2-based dimerization systems). Proof-of-principle experiments include light control of an engineered transcription factor that consisted of a bacterial DNA-binding domain fused to the VP64 transcriptional activator sequence, a nuclear export sequence, and a LINuS domain and optical control of p53 transcriptional activity using the LEXY approach (Niopek et al., 2014, Niopek et al., 2016). In addition, blue-light mediated activation of the transcription factor LIN-1 has been used to specifically manipulate cell fate during C. elegans development in vivo (Yumerefendi et al., 2015).

Cryptochrome-Based Tools. Cryptochrome dimerization systems offer an alternative to LOV proteins for blue light-mediated control of gene expression and genome targeting. Based on the CRY2-CIB1 interaction, split Gal4 transcription factors were created and applied to activate transcription in yeast and in zebrafish embryos (Kennedy et al., 2010, Liu et al., 2012, Taslimi et al., 2016). The split protein approach has also been applied to Cre recombinases, enabling light-dependent control of DNA recombination (Kennedy et al., 2010, Taslimi et al., 2016).

Second-generation techniques use customizable DNA-binding domains, such as TALEs or inactive CRISPR-Cas9, which enable most endogenous genes of interest to be targeted. Konermann et al. (2013) engineered a TALE-CRY2 and CIB1-effector (VP64)-interacting pair to activate the transcription of diverse endogenous genes in cultured neurons and in the mouse prefrontal cortex in vivo (so-called light-inducible transcriptional effectors [LITEs]). Furthermore, by using histone-modifying enzymes, including deacetylases, methyltransferases, and acetyltransferase inhibitors, as CIB1-coupled effectors, the authors could photoinduce epigenetic modifications, indirectly controlling target gene expression. Konermann et al. (2013) also suggested the replacement of TALEs with mutants of Cas9 that are nucleolytically inactive, which was independently followed up by two other groups (Nihongaki et al., 2015, Polstein and Gersbach, 2015). Interestingly, both of these groups reported an inversely constructed interacting pair, with CRY2 fused to the respective transcriptional activator (e.g., VP64, p65AD) and CIB1 coupled to either the N or C terminus of dCas9, the catalytically inactive form of Cas9. Coexpression of the interacting pair with several single guide RNAs (sgRNAs) enabled multiple endogenous genes to be simultaneously activated by light-induced transcription factor recruitment. Of all systems described so far, LITEs are the only transcriptional control system to have been applied in neurons to date. Nevertheless, Cas9-based systems also allow endogenous mammalian genes to be targeted, and we hope to see follow-up studies using this approach in neurons in the near future.

UVR8-Based Transcriptional Activation. UVB-induced heterodimerization of UVR8 and COP1 represents an alternative means by which to recruit transcriptional activators (e.g., nuclear factor κB [NF-κB], VP-16) to specific DNA-binding domains (Gal4, macrolide-responsive repressor E), as shown in proof-of-principle experiments in different mammalian cell culture systems (Crefcoeur et al., 2013, Müller et al., 2013b). Notably, independent transcriptional activation of UVR8- and phytochrome-controlled genes was achieved by applying either pulsed UVB illumination or continuous far-red light. UVR8 was also combined with the blue light-activated Vivid system to either activate the expression of a Vivid-controlled gene (with blue light) or both the UVR8- and the Vivid-controlled genes (with UVB), but this combination was not suitable to independently control gene expression (Müller et al., 2013b).

Optogenetic Tools: Sensors

Genetically encoded optical sensor proteins allow a wide range of cellular parameters to be monitored, including ion and metabolite concentrations, enzyme activities, and membrane voltage. Available sensors comprise various environmentally sensitive fluorescent proteins as well as bioluminescent indicators. Although of equal importance as optogenetic actuators, we provide here only a brief introduction to sensors and refer the interested reader to several excellent reviews for further information (Badr and Tannous, 2011, Germond et al., 2016, Lin and Schnitzer, 2016, Tantama et al., 2012).

Fluorescent indicators encompass intrinsically sensitive proteins, such as the original pH-sensitive GFP, as well as engineered sensor proteins. New indicators are created by mutation, circular permutation, and combination of fluorescent proteins with protein domains that change their conformation upon interaction with a molecule of interest (Tantama et al., 2012). These conformational changes are then transmitted to the fluorescent output domains, altering fluorescence spectra, fluorescence intensity, or fluorescence resonance energy transfer (FRET) efficiency between the donor-acceptor pair. Measurements can thus be either intensiometric, following changes in single-wavelength fluorescence, or ratiometric, in the case of FRET. Ratiometric sensors enable the determination of absolute values independent of their expression level and are mostly insensitive to focus drifts that might occur in in vivo experiments but are more difficult to implement in multi-color experiments.

In neuroscience, the most commonly applied fluorescent sensors are genetically encoded calcium indicators (GECIs), genetically encoded voltage indicators (GEVIs), and pH sensors (pHluorins). GECIs are employed to report neuronal AP firing represented by somatic Ca²⁺ transients (Knöpfel, 2012) and Ca²⁺ fluxes in the presynaptic terminal (Dreosti et al., 2009), in postsynaptic compartments (Mao et al., 2008), or in dendrites (Xu et al., 2012). Many GECIS are based on calmodulin (CaM) and the Ca²⁺-CaM-binding peptide, M13, as in the intensiometric green fluorescent GCaMPs (Chen et al., 2013b, Nakai et al., 2001) and genetically encoded calcium indicators for optial imaging (GECOs) (Zhao et al., 2011b), and the ratiometric yellow cameleon-nano sensors (Horikawa et al., 2010, Nagai et al., 2004). Troponin-based FRET sensors represent alternative ratiometric Ca²⁺ indicators (Mank et al., 2008, Thestrup et al., 2014). The latest GECI developments have yielded green fluorescent indicators with sensitivity and kinetics that are comparable with inorganic dyes; further developments aim to generate spectrally distinct variants with similar optical performance (Rose et al., 2014).

Complementary to the GECIs, there are two distinct classes of protein-based voltage reporters. The first class consists of the voltage-sensitive fluorescent proteins (VSFPs), which employ the voltage-sensitive domain of the Ciona intestinalisvoltage-sensing phosphatase (Ci-VSP). Examples include the prototypic VSFP1.2 as well as VSFP-Butterfly, Arclight, and ASAP1 (Akemann et al., 2012, Dimitrov et al., 2007, Jin et al., 2012, St-Pierre et al., 2014). The second class of GEVIs uses the intrinsic voltage sensitivity of pumping-deficient mutants derived from microbial rhodopsins like Arch, Mac from Leptosphaeria maculans, or Ace from Acetabularia acetabulum (Gong et al., 2014, Gong et al., 2015, Hochbaum et al., 2014, Kralj et al., 2011). Both VSFP and opsin-derived sensors are available as intensiometric or ratiometric variants, and the latest versions allow single AP firing to be detected in acute slices or in vivo (Gong et al., 2015).

pHluorins are widely used to analyze the exocytosis of synaptic vesicles and their subsequent re-acidification following endocytosis (Miesenböck et al., 1998, Sankaranarayanan et al., 2000), but they can also be employed to monitor the surface removal of neurotransmitter receptors (Ashby et al., 2004). Ratiometric pHluorin enables quantitative measurement of intravesicular pH using 410 and 470 nm excitation, whereas the fluorescence of ecliptic pHluorin decreases with pH and is extinguished (“eclipses”) at pH < 6. Refined pH sensors include the superecliptic (SE)-pHluorin, which has improved fluorescence properties (Sankaranarayanan et al., 2000), the red-shifted pH indicators pH-tomato (Li and Tsien, 2012) and mOrange2 (Egashira et al., 2015, Li et al., 2011, Shaner et al., 2008), as well as “pseudoratiometric” constructs combining intravesicular SE-pHluorin with cytosolic red fluorophores (Kim and Ryan, 2010, Rose et al., 2013). A vast body of literature exists on readouts of organellar pH using genetically encoded pH sensors, and we refer the interested reader to several excellent reviews that provide in-depth coverage of this topic (Benčina, 2013, Bizzarri et al., 2009, Miesenböck, 2012, Royle et al., 2008).

Numerous other fluorescent indicators have been applied to study brain cell activity; these include sensors for inorganic ions and molecules (chloride, zinc, hydrogen peroxide) and for organic signaling molecules and metabolites (glutamate, acetylcholine, ATP, NADH, cyclic nucleotides, glucose, pyruvate, phospholipids) (Tantama et al., 2012). Several sensors that directly monitor enzyme activity have also been developed. These allow the activity of small GTPases (Kraynov et al., 2000), kinases (e.g., protein kinase A) (Nagai et al., 2000, Zhang et al., 2001), and proteases (Xu et al., 1998) to be detected. Moreover, GPCR activation can be followed by intra- or intermolecular FRET, either within the heterodimeric G protein or between the GPCR and a binding partner (G protein or arrestin). This approach has been applied to analyze the activation kinetics of metabotropic glutamate receptors, GABA_B receptors, and M1 muscarinic acetylcholine receptors (Janetopoulos et al., 2001, Jensen et al., 2009, Matsushita et al., 2010, Tateyama et al., 2004, Vilardaga et al., 2003).

Although existing fluorescent sensors enable simultaneous imaging of multiple cellular processes, specific targeting of sensor proteins to subcellular structures of interest provides a potent means by which to study functional compartmentalization in neurons. In the following section, we review targeting strategies for key neuronal compartments.

Subcellular Targeting Strategies

Figure 1
Figure 2
Figure 3
Figure 4

Optogenetics is especially well suited to uncovering compartment-specific processes that cannot easily be resolved using classical electrophysiological, pharmacological, biochemical, or genetic methods. The subcellular enrichment of optogenetic tools has been achieved using short peptides that encode intracellular retention signals, specific targeting or anchor signals, or fusions with full-length organellar marker proteins (Figure 3; Table 2). Below, we summarize strategies for targeting optogenetic tools to different subcellular compartments and provide examples of their applications in neurons and non-neuronal brain cells.

Table 2 Subcellular Targeting Strategies for Optogenetic Tools in Neuroscience
Targeting Motifsa	Tool	cDNA/Sequenceb	Reference
Membrane Tethering or Enhanced Surface Trafficking
C-terminal K-Ras CAAX domain (KEKMSKDGKKKKKKSKTKCVIM)	CIB1-GFP-CAAX + Cry2-mCherry-Raf1c	A 79574	(Zhang et al., 2014)
	Phy-mCherry-CAAX + PI3K-binding domain (iSH)-YFP-PIFd	A 50839, A 50841	(Toettcher et al., 2011)
	CIBN-CAAX + Cry2- PI3K-binding domain (iSH)	A 79574, A 66839	(Idevall-Hagren et al., 2012, Kakumoto and Nakata, 2013)
	CIB1-GFP-CAAX + mCherry-Cry2-5ptaseOCRLe	A 79574, A 66836	(Idevall-Hagren et al., 2012)
	YC3.60-CAAX (YC3.60pm)	GB AB178714f	(Nagai et al., 2004)
N-terminal CD4	hCD4-GCaMP2	A 18932	(Mao et al., 2008)
N-terminal MARCKS (first 41 aa)	MARCKS-GCaMP2g		(Mao et al., 2008)
N-terminal MARCKS (first 8 aa: MGCCFSKT)	MARCKS-GCAMP3	A 50022	(Vrontou et al., 2013)
N-terminal Lck	Lck-GCaMP2	A 24794	(Shigetomi et al., 2010b)
	Lck-GCaMP3	A 26974	(Shigetomi et al., 2010a)
	Lck-GCaMP5G	A 34924h	(Akerboom et al., 2012)
N-terminal signal peptide of NAchR β-subunit (MRGTPLLLVVSLFSLLQD), C-terminal ER export signal of Kir2.1 (FCYENEV)	eNpHR/eNphR2.0	A 20949	(Gradinaru et al., 2008)
C-terminal Kir2.1 Golgi trafficking signal (KSRITSEGEYIPLDQIDINV) and C-terminal ER export signal of Kir2.1 (FCYENEV)i	eNpHR 3.0	A 26972	(Gradinaru et al., 2010)
	TS-VChR1, TS-C1V1	A 35498	(Yizhar et al., 2011)
	eArch3.0 (eA3), eArchT3.0 (eAT3), eMac3.0 (eM3)	A 35514, A 35513, A 35515	(Mattis et al., 2011)
Axon and AIS
C-terminal myosin VI binding domain	ChR2-MVIBD-GFP		(Lewis et al., 2011)
C-terminal ankyrin G-binding domain of the intracellular loop II-III of Nav1.2 (221 aa)	ChR2-YFP-NavII-III	A 26057	(Grubb and Burrone, 2010)
C-terminal ankyrin G-binding site of Nav1.6 (TVRVPIAVGESDFENLNTEDVSSESDP)	ChR2-GFP-NavII-III		(Wu et al., 2011a)
N-terminal ankyrin G polypeptide (aa 1–837)	ankyrin-hChR2, ankyrin-NpHR		(Greenberg et al., 2011)
Postsynaptic Density and Spines
C-terminal PSD clustering motif ETQV from NMDA receptors	ChR2-EYFP-ETQV		(Gradinaru et al., 2007)
C-terminal chicken β-actin	GCaMP2-actin	A 18928	(Mao et al., 2008)
N-terminal PSD-95	PSD95-GCaMP2	A 18931	(Mao et al., 2008)
	PSD-95-GCaMP5K	A 18931	(Leitz and Kavalali, 2014)
N-terminal PSD-95ΔPDZ1.2	AS-PaRac1 (PSD-95ΔPDZ1.2-Venus-PaRac1)j		(Hayashi-Takagi et al., 2015)
N-terminal PSD-95 (aa 1–745)	PSD-95-ChR2, PSD95-NpHR		(Greenberg et al., 2011)
C-terminal homer1c	CRY2-GFP-homer1c	A 89442	(Sinnen et al., 2017)
C-terminal PSD95_FingR (intrabody against PSD95)	CRY2PHR-PSD95FingR-GFP	A 89443	(Sinnen et al., 2017)
N-terminal mCherry-GluA1	mCherry-GluA1-CIB	A 89444	(Sinnen et al., 2017)
Somatodendritic Region (Excluding Exons)
C-terminal MBD of melanophilin (aa 176–201: RDQPLNSKKKKRLLSFRDVDFEEDSD)	ChR2-MBD	A 21484	(Lewis et al., 2009)
			(Wu et al., 2013)
C-terminal restriction and clustering signal from the cytoplasmic C terminus of the voltage-gated K+ channel 2.1 (QSQPILNTKEMAPQSKPPEELEMSSMPSPVAPLPAR TEGVIDMRSMSSIDSFISCATDFPEATRF)	ChR2-Kv2.1	A 89256	(Baker et al., 2016, Wu et al., 2013)
C-terminal dendritic targeting motif of neuroligin-1 (VVLRTACPPDYTLAMRRSPDDVPLMTPNTITM)	ChR2-NLG1		(Wu et al., 2013)
Lumen of Secretory Vesicles
N-terminal synaptobrevin (VAMP2)	synapto-pHluorink	GB AF058695.1, GB AF058694.2	(Miesenböck et al., 1998)
	superecliptic synapto-pHluorinl	GB AY533296.1	(Sankaranarayanan et al., 2000)
	VAMP2-pHTomato		(Li and Tsien, 2012)
Synaptophysinm	SypHy/SypHluorin	A 24478, A 37003, A 37004, A 37005	(Granseth et al., 2006, Zhu et al., 2009)
	synaptophysin-mOrange2 (syp-mOr)		(Egashira et al., 2015)
	SypHTomato		(Li and Tsien, 2012, Pech et al., 2015)
	synaptophysin-pHluorin-Arch3-mKate2-βHK (pHoenix)	A 70111	(Rost et al., 2015)
	ratio1XsypHy	A 44268	(Rose et al., 2013)
C-terminal synaptotagmin 1–7, 9–12, 17	pHluorin-Syt-1 etc.		(Dean et al., 2012)
VGLUT1n	VGLUT1-pHluorin		(Balaji and Ryan, 2007, Voglmaier et al., 2006, Zhu et al., 2009)
	VGLUT1-mOr2		(Li et al., 2011)
N-terminal VGAT	VGAT-pHluorin (VGAT-pH)	A 78578	(Santos et al., 2013)
Cytosol of Presynaptic Terminal
N-terminal synaptophysin	synaptophysin-mCherry-luciferase: synATP	A 51819	(Rangaraju et al., 2014)
	SyGCaMP2/synaptoGCamp2	A 26124, A 26125	(Dreosti et al., 2009, Zhao et al., 2011a)
	SyGCaMP3		(Li et al., 2011)
	SyGCaMP5G		(Akerboom et al., 2012)
	SyGCaMP6		(Mahn et al., 2016)
	InSynC (SYP1-miniSOG)	A 50971, A 50972	(Lin et al., 2013b)
C-terminal synaptobrevin (VAMP2)	InSynC (miniSOG-VAMP2)	A 50969, A 50970	(Lin et al., 2013b)
N-terminal synaptotagmin-1 (C. elegans)	InSynC (snt-1-miniSOG)		(Lin et al., 2013b)
C-terminal microtubule-associated protein 1 light chain 3 beta (LC3)	mRFP-GFP tandem fluorescence-tagged LC3 (tfLC3)	A 21074	(Kimura et al., 2007)
CD63	Lyso-pHoenix (CD63-pHluorin-Arch3-mKate2-βHK)	A 70112	(Rost et al., 2015)
Endoplasmic Reticulum
N-terminal: signal peptide from calreticulin (MLLSVPLLLGLLGLAVA); C-terminal: KDEL retention motif, e.g., last seven amino acids of the ER chaperone GRP78/BiP (TAEKDEL)	GCaMPer: CRTsigpep-GCaMP3 (D324G,D360G,397G,D435G)-KDEL	A 63885	(Henderson et al., 2015)
	D1ER (second-generation cameleon GECI targeted to the ER)	A 36325	(Palmer et al., 2004)
	erGAP (pH-insensitive aequorin-based low-affinity GECI targeted to the ER)	A 78118, A 78120	(Navas-Navarro et al., 2016, Rodriguez-Garcia et al., 2014)
	LAR-GECO (low-affinity red fluorescent genetically encoded Ca2+ indicators for optical imaging)	A 61244, A 61245	(Wu et al., 2014)
	CEPIA1er: calcium-measuring organelle-entrapped protein indicator (GCamp2 -E31D/F92W/E104D/D133E)	A 58215, A 58216, A 58217o	(Suzuki et al., 2014)
C-terminal 2x or 3x vesicular stomatitis virus glycoprotein (VSV-G) C terminus	UVR8-VSVG-YFP: light-triggered secretion from the ER	A 49800	(Chen et al., 2013a)
Mitochondria
N-terminal COX 8 localization sequence (MSVLTPLLLRGLTGSARRLPVPRAKIHSLGDP). Tandem repeats with up to eight copies of targeting signal increase mitochondrial enrichment.	CEPIAmt (calcium-measuring organelle-entrapped protein indicators: GCaMP2-derived low-affinity GECI)	A 58218 A 58219 A 58220p	(Suzuki et al., 2014)
	Mito-GEM-GECO1 (blue-green emission ratiometric genetically encoded Ca2+ indicators for optical imaging)	A 32461	(Zhao et al., 2011b)
	RCaMP1e (red-shifted genetically encoded calcium indicator constructed from calmodulin and cp-mRuby	A 42874q	(Akerboom et al., 2013)
	LAR-GECO (low-affinity red fluorescent genetically encoded Ca2+ indicators for optical imaging)	A 61245	(Wu et al., 2014)
	COX8-GCaMP5G	A 58509	(Venkatachalam and Cohen, 2014)
	HyPer-Mr inserted into the H2O2-sensitive domain of the E. coli protein OxyR)		(Belousov et al., 2006)
	Grx1-roGFP2: glutathione redox potential sensors	A 64977	(Breckwoldt et al., 2014, Gutscher et al., 2008)
	Mito-SypHer/SypHer-mt/mito-pHt: pH sensor	A 48251	(Azarias et al., 2011, Marland et al., 2016, Poburko et al., 2011)
	Mito-ZapCY1 zinc sensoru	A 58996	(Park et al., 2012)
N-terminal localization sequence (MLSLRQSIRFFK) of cytochrome c oxidase subunit IV	Mito-Gcamp2		(Chen et al., 2011, Marland et al., 2016)
	Mitochondrial matrix-roGFP2	A 49437	(Guzman et al., 2010)

aN- or C-terminal denotes localization of the trafficking motif relative to the actuator and/or sensor

bGB, GenBank sequence; A, Addgene clone

cRaf1 membrane recruitment activates the Raf/MEK/ERK pathway

dMembrane binding of PI3K induces PIP₃ production

eMembrane recruitment of 5ptase triggers PIP₂ and PIP₃ dephosphorylation

fAvailable from the lab of Atsushi Miyawaki

gThird and fourth amino acid of MARCKS mutated to Cys to generate palmitoylation sites

hAlso available from Addgene: LCK-GCAMP6 (52924)

iTS placed between NpHR and YFP; additional ER export sequence at the C terminus of YFP

jAvailable from the lab of Haruo Kasai

kGenBank ID for ecliptic and ratiometric phLuorin without synaptobrevin

lF64L and S65T mutations in the original pHluorin enhanced fluorescence; GenBank ID for SE phLuorin without synaptobrevin

mInsertions of fluorophores between the 3^rd and 4^th transmembrane domain of synaptophysin

nInsertion of fluorophores in the first luminal loop (between Val-103 and Val-104) of rat VGLUT1

oA 58216, intensiometric red-fluorescent (R-CEPIA1er); A 58217, ratiometric blue/green (GEM-CEPIA1er) variant

pA 58218, CEPIA2mt (K_D = 0.16 μM); A 58219, CEPIA3mt (K_D = 11 μM); A 58220, CEPIA4mt (K_D = 59 μM). Intensiometric red fluorescent (R-CEPIA1er) and ratiometric blue/green (GEM-CEPIA1er) variants are also available

qUntargeted variant RCaMP1h

rHyPer, circular permutated YFP (cpYFP). Fluorescence is also pH sensitive

sImproved version of the ratiometric and pH-insensitive redox sensor roGFP

tHyPer-derived pH sensor (C199S), insensitive to H₂O₂

uCombines the truncated cyan fluorescent protein (CFP)- and citrine-flanking Zn²⁺-binding domain from the first 2 zinc fingers of Zap 1 from S. cerevisiae

Plasma Membrane

The first attempts to direct the localization of optogenetic actuators in neurons aimed to increase their functional expression at the plasma membrane to yield more efficient control of membrane voltage while reducing intracellular aggregation and toxicity (Figure 3A). Gradinaru et al. (2008) described enhanced halorhodopsin (eNpHR), later termed eNphR2.0, which incorporated an N-terminal signal peptide from the β-subunit of the nicotinic acetylcholine receptor (β-NAchR) (Isenberg and Meyer, 1989) and a C-terminal ER export sequence from the potassium channel Kir2.1 (Stockklausner and Klocker, 2003). These modifications abolished intracellular aggregation and increased peak photocurrents (Gradinaru et al., 2008). Remarkably, the β-NAchR export sequence alone or in combination with other ER export sequences yielded only minor improvements of surface expression, highlighting the fact that targeting motifs may exert additive and context-dependent effects. In a further refinement, the β-NAchR export sequence was replaced by a Golgi trafficking signal (TS) from Kir2.1 (Hofherr et al., 2005), inserted between the opsin and the fluorophore, generating eNphR3.0, which had improved membrane expression and 3-fold increased photocurrents compared with eNpHR2.0 (Gradinaru et al., 2010). The Golgi TS and ER export sequence have since been adopted to optimize the membrane expression of many actuators and sensors; for example, the green-absorbing Volvox channelrhodopsin 1 (VChR1) (Yizhar et al., 2011), the proton pumps Mac and Arch (Mattis et al., 2011), the sodium pump KR2 (Kato et al., 2015), and Arch-derived voltage sensors (Flytzanis et al., 2014, Gong et al., 2013, Hochbaum et al., 2014).

Ca²⁺ transients exhibit faster kinetics and higher amplitudes in the vicinity of Ca²⁺ channels than in the overall cytosol (Augustine et al., 2003). Soluble GECIs have been localized to the sub-membrane cytosolic compartment to visualize these Ca²⁺ microdomains near the plasma membrane. The attachment of GECIs to the plasma membrane was achieved by fusing the GECI to the transmembrane domain of CD4 (Mao et al., 2008) or to membrane-tethered peptides, such as variants of the N-termini of MARCKS (myristoylated alanine-rich C-kinase substrate) or Lck (lymphocyte-specific protein tyrosine kinase; Figure 3A). In neurons, fusion of CD4 or MARCKS to GCaMP2 did not improve visualization of Ca²⁺ transients for the detection of APs relative to cytosolic GCaMP2 (Mao et al., 2008). However, Lck-GCaMP2 enabled better detection of Ca²⁺ signals in finger-like astrocytic processes (Shigetomi et al., 2010b). Ca²⁺ imaging in astrocytic processes was further improved by Lck-GCaMP3 (Shigetomi et al., 2010a) and Lck-GCaMP5 (Akerboom et al., 2012), revealing the presence of “spotty” Ca²⁺ microdomains in astrocytes (Shigetomi et al., 2011, Shigetomi et al., 2013). The FRET-based GECI YC3.60 was also targeted to the plasma membrane using the polyisoprenylated C-terminal CAAX box of K-Ras (Hancock et al., 1989); however, this construct performed poorly in a transgenic mouse model because of the compressed dynamic range of fluorescence changes (Kotlikoff, 2007, Nagai et al., 2004).

Neuronal Plasma Membrane Compartments

The uniform distribution of optogenetic actuators in the plasma membrane enables effective manipulation of the overall activity state of neurons; however, wide-field illumination can cause excessive and unphysiological ion fluxes (Zhang and Oertner, 2007). This issue can be addressed by expression of depolarizing or hyperpolarizing actuators in specific neuronal membrane compartments for more precise spatial confinement of light-induced membrane potential fluctuations. Several strategies have been used to restrict opsin expression to the axon’s plasma membrane and to the axon initial segment (AIS), the somatodendritic region, and the postsynaptic density (PSD) (Figures 3B and 3D).

Axonal Compartment

Optogenetic excitation or inhibition of the axon allows neuronal output to be directly controlled and neurotransmitter release from long-range projection neurons to be locally manipulated within their target areas. To date, most studies involving direct light stimulation of long-range projection fibers have relied on passive accumulation of ChR2 in axons, a process that takes up to several weeks (Britt et al., 2012) but can be accelerated by targeted axonal trafficking. For example, fusion of ChR2 to an artificial binding domain for myosin VI (ChR2-MVIBD-GFP) enriches surface ChR2 in axons of cortical neurons to a level that abolishes super-threshold depolarization by dendrite illumination in some cells (Lewis et al., 2011). However, it is unclear whether investigations of long-range projections would benefit from using this modification because the authors did not evaluate whether coupling to myosin VI expedited the expression of ChR2 in axons and only analyzed local axon projections.

The AIS is of specific interest for voltage manipulation by optogenetic actuators because it represents the final integration site of dendritic inputs and promotes the generation of APs because of its high density of voltage-gated sodium (Na_v) channels. Na_v channel clustering in the AIS (and in the nodes of Ranvier) is mediated by the interaction of ankyrin G with nine amino acids in the II-III intracellular loop of Na_v channels (Zhou et al., 1998). Grubb and Burrone (2010) devised a way to specifically cluster ChR2-YFP (yellow fluorescent protein) at the AIS by adding the 221 amino acid-long Na_v1.2 II-III intracellular loop, including the ankyrin G-binding motif, to its C terminus. This construct did not impair basic neuronal electrical properties but failed to elicit APs, most likely because of its sparse expression at the AIS together with the small single-channel conductance of ChR2. A similar construct (ChR2-GFP attached to the 27-amino acid ankyrin G-binding domain from Na_v1.6) was enriched at a dendritic site with high ankyrin density in axonless but spiking retinal Aii amacrine cells, identifying this region as an AIS-like structure in these cells (Wu et al., 2011a). Strikingly, overexpression of ChR2 fused to the ankyrin G-binding domain disrupted the endogenous Na_v clusters in the AIS-like segments, abolishing AP generation in AII amacrine cells, a finding that was also reported for retinal ganglion cells (Zhang et al., 2015b).

Somatodendritic Compartment

For optogenetic experiments aiming to specifically excite a neuronal subpopulation, exclusion of microbial opsins from the axon can greatly enhance the precision of optical activation by avoiding light stimulation of fibers that cross the illuminated brain area (Zalocusky et al., 2013). The best attempt at this has exploited the polarized transport of membrane proteins into dendrites by myosin motor proteins. Attachment of the myosin Va-binding domain (MBD) of melanophilin to ChR2 (Geething and Spudich, 2007) enhanced its expression in the somatodendritic compartment (Lewis et al., 2009). Spiking induced by illumination of the axonal compartment was reduced by 80% in mouse cortical L2/3 pyramidal neurons that expressed ChR2-MBD, and the dendritic enrichment of ChR enabled investigators to specifically trigger APs by dendritic photocurrents under wide-field illumination. Myosin-dependent transport was also employed to confine ChR2 expression to the soma and proximal dendrites by addition of a clustering motif from the K_v2.1 voltage-gated potassium channel (Lim et al., 2000). In combination with temporal focusing of 2-photon excitation, this allowed for single-cell resolution circuit mapping in acute brain slices (Baker et al., 2016).

The polarized subcellular localization of depolarizing and hyperpolarizing optogenetic actuators provides an attractive approach for creating a center-surround antagonism in retinal ganglion cells (RGCs). RGCs are prime candidates for optogenetic therapeutic strategies that aim to restore vision because they might still survive after photoreceptor degeneration. To achieve contrast enhancement, on and off center-surround receptive fields have to be restored. Two studies have explored the potential of using somatic ChR2 targeting and dendritic NpHR enrichment to create an excitatory “center” and an antagonistic “surround” field in RGCs. Greenberg et al. (2011) employed N-terminal fragments of ankyrin G (1–2512) and PSD-95 (1–2235) fused to the N terminus of ChR2 and eNpHR to achieve somatic and dendritic targeting, respectively, and reported morphologically and functionally complementary localization of these two constructs in RGCs. By contrast, Wu et al. (2013)systematically assessed different targeting motifs fused to the C terminus of ChR2-EGFP to achieve complementary expression in the somata (center) and distal dendrites (surround) of RGCs after adeno-associated virus (AAV)-mediated gene transfer. The C-terminal attachment of a targeting sequence from K_v2.1 enriched ChR2 at the soma and reduced its expression in distal dendrites and axons. K_v2.1 tagging was superior to fusion with the ankyrin G-binding domain of Na_v1.6, which decreased somatic expression levels of ChR2. Dendritic trafficking motifs derived from the GluR1 cytoplasmic C terminus (Ruberti and Dotti, 2000) and from K_v4.2 (Rivera et al., 2003), nAchR α7 (Xu et al., 2006), and telencephalin (Mitsui et al., 2005) did not affect the subcellular localization of ChR2 compared with its unmodified form, whereas the melanophilin MBD and the dendritic targeting motif of neuroligin-1 (Rosales et al., 2005) resulted in its axonal exclusion and somatodendritic enrichment. The neuroligin-1 motif currently presents the most promising candidate for surround targeting in RGCs because the melanophilin-derived tag caused intracellular inclusions (Wu et al., 2013). Together, these reports describe exciting optogenetic applications involving subcellular targeting strategies in the retina, but they also highlight the challenges and limitations of these approaches and the need for careful controls.

PSD and Spines

Spines are thought to represent the smallest computational units in neurons, and they are prime targets for optogenetic readouts and manipulation (Figure 3D). Targeting of ChR2 to PSDs was first achieved by C-terminal addition of the short PDZ domain binding motif ETQV to ChR2-YFP (Gradinaru et al., 2007). GECIs have been targeted to spines by fusing either chicken β-actin (Mao et al., 2008) or PSD-95 (Leitz and Kavalali, 2014) to the C terminus of GCaMPs. Interestingly, Mao et al. (2008)reported that immobilization of GCaMP2-actin at the spine cytoskeleton resulted in increased photobleaching of the GECI because of its impaired replacement by freely diffusible cytosolic GCaMP2. However, recently developed ultrasensitive GCaMP6 versions produce >20-fold larger signals relative to GCaMP2 and enable high-resolution imaging of spine Ca²⁺ signals in vivo using sparse neuronal labeling with non-targeted, cytosolic indicators (Chen et al., 2013b).

Optogenetic manipulation of spines advanced significantly with the development and application of activated synapse (AS)-PaRac1, a photoactivatable version of the small GTPase Rac1 that is targeted to ASs (see the discussions above of LOV protein-based tools for protein activation and inactivation, and Hayashi-Takagi et al., 2015and Wu et al., 2009). Hayashi-Takagi et al. (2015) fused a PaRac1 mutant (PaRac1 L514K/L531E, which has reduced dark activity compared with the original PaRac1) to the PSD95 mutant PSDΔ1.2, which clusters at postsynaptic sites but does not interact with PDZ binding proteins. They then introduced an mRNA dendritic targeting element (DTE) downstream of the PSDΔ1.2-Venus-PaRAC1 coding sequence to induce transportation of the construct’s mRNA to dendritic segments and, thus, its localized translation following NMDA receptor (NMDAR) activation (Kobayashi et al., 2005). The resulting AS-targeting cassette (AS-PaRac1) caused sparse labeling of spine heads in vitro under baseline conditions. The expression of this construct was strongly enhanced by protein synthesis-dependent, but not by protein synthesis-independent, potentiation of synapses. Optical activation of AS-PaRac1 induced spine shrinkage, accompanied by a reduction in postsynaptic calcium entry, but did not reduce spine numbers. AS-PaRac1 expressed under the control of an immediate-early gene promoter in vivo preferentially localized to newly formed or enlarged spines following motor training. Its photoactivation within 24 hr post-training disrupted acquired learning without disturbing an unrelated motor skill. This light-induced depotentiation of spines revealed the presence of task-specific synapses (“synaptic ensembles”) because many of the optically shrunk spines were re-potentiated by task retraining, providing evidence for causal links between spine morphology and memory formation. In the future, combination of AS-PaRac1 with cell-type- or connectivity-specific expression systems might enable the light-driven elimination of protein synthesis-dependent long-term potentiation (late LTP), promising exciting new insights into the role of synaptic plasticity at specific synaptic contacts in shaping neuronal network events and behavior. Such AS-targeting cassettes might also enable the specific expression of other optogenetic tools at activated spines. Recently, Sinnen et al. (2017) presented a method for reversible optogenetic relocalization of both soluble and membrane proteins to the PSD. Fusions of CRY2 with homer1c, PSD95, or FingR (an intrabody that binds PSD95; Gross et al., 2013) anchored CRY2 to the PSD and enabled recruitment of CIB1-tagged mCherry, Ca²⁺/calmodulin-dependent protein kinase II (CamKII), or GluA1 following blue light exposure. Remarkably, Sinnen et al. (2017) discovered that optogenetically controlled incooperation of additional GluA1 into the PSD enhanced excitatory input not by increasing synaptic strength but, rather, by unmasking previously silent postsynaptic sites, an observation only possible by the high spatiotemporal control of optogenetic GluA1 positioning. Whereas Hayashi-Takagi et al. (2015) adopted the AS-PaRac1 system for in vivo applications, the method presented by Sinnen et al. (2017) cannot be easily applied in living animals because this system is too large for packaging into AAVs. Nevertheless, these two publications outline a new experimental strategy for the optogenetic control of synaptic inputs, not only by manipulating the firing of the presynaptic neuron but, rather, by altering the strength of individual synapses. So far, only glutamatergic synapses have been manipulated, but we expect similar tools to be developed for other transmitter systems, such as for GABAergic synapses.

Synaptic Vesicles and Presynaptic Terminals

Soluble fluorescent sensors can be readily targeted to axonal terminals by fusing indicators to synaptic vesicle (SV) proteins, which are highly enriched at presynaptic boutons (Figure 3C). Both the presynaptic cytosol and the lumen of SVs are accessible to sensors, providing compartment-specific optogenetic readout strategies. Miesenböck et al. (1998) created the prototypic vesicular pH sensor synapto-pHluorin by fusing a pH-sensitive GFP derivate (pHluorin) to the intravesicular C terminus of synaptobrevin/VAMP-2. Although widely used to visualize exocytosis and reacidification of endocytosed vesicles, the initial synapto-pHluorin suffered from two main limitations: a relatively high surface fraction (about 15%) (Sankaranarayanan et al., 2000) and its significant lateral diffusion following exocytosis (Granseth et al., 2006). To overcome these limitations, other SV proteins have been tested for their ability to anchor fluorophores in vesicles. Insertion of SE-pHluorin into the second intravesicular loop of synaptophysin yielded “sypHy,” which shows reduced surface expression and a superior signal-to-noise ratio compared with synapto-pHluorin (Granseth et al., 2006). The lowest surface expression (2%) was found with SE-pHluorin inserted into the first luminal loop of vesicular glutamate transporter 1 (Voglmaier et al., 2006), which even allowed optical detection of single-vesicle fusion events (Balaji and Ryan, 2007). Localization of pHluorin to exocytic vesicles has also been achieved by fusing it to the N termini of different synaptotagmin isoforms (Dean et al., 2012, Diril et al., 2006, Fernández-Alfonso et al., 2006) and to the C terminus of the vesicular GABA transporter (Santos et al., 2013) as well as by inserting it into the first luminal loop of SV protein 2A (Zhang et al., 2015a). This intraluminal tagging of different components of the SV proteome allows researchers to study their individual trafficking and recycling mechanisms. However, it should be noted that overexpression of synaptobrevin-pHluorin and synaptophysin-pHluorin resulted in larger protein clusters than those formed by their endogenous counterparts, a difference that could not solely be attributed to overexpression of the targeting protein (Opazo et al., 2010).

When sensors are fused to the cytosolic regions of vesicular proteins, their localization in the presynaptic cytosol is largely unaffected by exo- and endocytosis (in contrast to sensors targeted to the SV lumen). Synaptophysin has been proven to be a particularly powerful tool for targeting optical sensors to the presynaptic cytosol. Dreosti et al. (2009) fused GCaMP2 to the cytoplasmic C terminus of synaptophysin, creating the first genetically encoded presynaptic Ca²⁺ indicator. This tool was later refined by development of synaptophysin-GCaMP3, 5, and 6 (Akerboom et al., 2012, Li et al., 2011, Mahn et al., 2016). A synaptophysin-based targeting strategy was also employed to develop a quantitative genetically encoded presynaptic ATP sensor (Syn-ATP) that consists of mCherry in tandem with optimized firefly luciferase tagged to the cytosolic C terminus of synaptophysin (Rangaraju et al., 2014). Syn-ATP enables ratiometric luminescence-to-fluorescence imaging and reporting of presynaptic ATP level changes elicited by neuronal firing. With multiple targeting systems at hand, it is now feasible to perform multiparametric imaging on synaptic contacts using genetically encoded, spectrally distinct sensors; for example, pre- or postsynaptically targeted GCaMPs combined with red-shifted vesicular pH sensors indicating exocytosis (Leitz and Kavalali, 2014, Li et al., 2011, Pech et al., 2015).

Similar to sensors, optogenetic actuators have also been targeted to presynaptic terminals by fusion to SV proteins. Light-driven acidification of SVs, in combination with pHluorin-based readout of vesicular pH, was achieved by embedding Arch3 and a luminal pHluorin between the 3^rd and 4^th transmembrane domain of synaptophysin (Rost et al., 2015). The resulting fusion construct, “pHoenix,” was enriched at presynaptic terminals and localized as a bona fide SV protein. Activation of pHoenix resulted in accumulation of protons in synaptic vesicles, providing an additional driving force for vesicular transmitter loading and acutely increased transmitter release. Following pharmacological inactivation of vesicular H⁺-ATPases, pHoenix could functionally substitute for endogenous H⁺ pumps, enabling precise light-controlled vesicular acidification and transmitter uptake. This pHoenix-driven control of the vesicular fill state was then utilized to elucidate that incomplete SV filling reduces vesicular fusion competence.

Light-controlled inactivation of transmitter release is another type of optogenetic interference with presynaptic function that would represent an excellent tool for studying the role of individual synapses or synaptic circuits in neuronal computation and behavior. Two strategies have been reported to achieve synaptic inhibition. Based on chromophore-assisted light inactivation (CALI), Shu et al. (2011) developed inhibition of synapses with CALI (InSynC). This tool consists of the LOV2-derived flavoprotein miniSOG (Shu et al., 2011) fused to the cytosolic ends of synaptobrevin or synaptophysin to enable light-induced ablation of synaptic transmission via local generation of reactive oxygen species (ROS). ROS then inactivate the surrounding proteins via oxidation of tryptophan, tyrosine, histidine, cysteine, and methionine residues (Lin et al., 2013b). Using this method, synaptic inactivation was achieved in dissociated neuronal cultures, slice cultures, and C. elegans neurons in vivo and was partially reversible within 24 hr. An alternative approach to synaptic inhibition involved replacing endogenous synaptotagmin in C. elegans with synaptotagmin fused to a photosensitive degron (Hermann et al., 2015). When worms expressing this construct were exposed to blue light, their mobility was impaired within 15 min, and synaptotagmin was completely depleted within 1 hr of illumination. Unfortunately, both InSynC and the photosensitive degron-based approaches are limited by slow kinetics and irreversibility because they require long-term illumination to bring about protein inactivation/degradation and rely on protein synthesis for recovery; thus, alternative methods for optogenetic synapse inactivation are still needed.

Other Organelles

Subcellular optogenetic applications have not only been developed for neuron-specific compartments, such as dendrites, axons, and synapses, but also for organelles common to all eukaryotic cells. In fact, soon after cloning of the GFP coding sequence (Prasher et al., 1992), researchers realized its potential as a fluorescent sensor for organellar pH. Kneen et al. (1998) combined GFP with trafficking motifs for the ER, Golgi, and mitochondria, creating a first set of genetically encoded subcellular pH indicators. Similar subcellular targeting strategies were later developed for other genetically encoded fluorescent sensors and, to a much lesser extent, optogenetic actuators. Below, we summarize work that has reported applications in neurons and astrocytes.

ER and Golgi Apparatus. Export of optogenetic actuators from the ER and Golgi apparatus is essential for their efficient expression on the cell surface, especially because their intracellular retention can cause toxic protein accumulations that impair neuronal viability (Rein and Deussing, 2012). As outlined above, several studies have addressed these issues, and ER export sequences and Golgi TSs are now routinely added to membrane-spanning optogenetic tools. However, the ER is not just the site of transmembrane and secreted protein synthesis, processing, and trafficking but is also crucially involved in cellular metabolism and buffering of H⁺ and Ca²⁺. Thus, strategies have been developed to target genetically encoded fluorescent sensors to the ER to enable direct optical monitoring of ER physiology in neurons and astrocytes. Most commonly, a hydrophobic peptide—for example, the 30-amino acid signal peptide of bovine preprolactin (Sasavage et al., 1982)—is linked to the N terminus of the protein of interest. Signal peptides then facilitate this protein entering the secretory pathway and are post-translationally removed. Addition of C-terminal ER retention signals containing the amino acid sequence KDEL for luminal or KKXX for membrane proteins prevents further protein export to the Golgi (Jackson et al., 1990, Munro and Pelham, 1987). Importantly, no additional tags or fluorophores should be added to the C-terminal side of the KDEL motif because even short peptides can shield the ER retention signal (Costantini and Snapp, 2013). Another efficient ER targeting signal is the signal peptide of calreticulin (Fliegel et al., 1989), which, in combination with KDEL motifs, has been widely used to target both ratiometric and single-wavelength GECIs to the neuronal ER (de Juan-Sanz et al., 2017, Henderson et al., 2015, Navas-Navarro et al., 2016, Rodriguez-Garcia et al., 2014, Suzuki et al., 2014, Wu et al., 2014).

In contrast to ER localization, precise targeting strategies for Golgi structures are challenging because many Golgi proteins cycle between the Golgi and the ER or the Golgi and endosomes. In non-neuronal cells, pHluorins linked to β-1,4-galactosyltransferase, trans-Golgi network protein 38 (TGN38), or α-2,6-sialyltransferase have been used to visualize pH in the trans-Golgi, whereas pHluorin fused with mannosidase 1A has been used to label the cis-Golgi (Benčina, 2013).

Mitochondria. Besides providing energy in the form of ATP, mitochondria serve as intracellular Ca²⁺ sources and sinks and are interesting candidates for the targeted expression of GECIs (Akerboom et al., 2013, Suzuki et al., 2014, Wu et al., 2014) and other optogenetic indicators, including sensors for pH, Zn²⁺, H₂O₂, oxidant stress, and glutathione redox potential (Azarias et al., 2011, Belousov et al., 2006, Breckwoldt et al., 2014, Guzman et al., 2010, Marland et al., 2016, Park et al., 2012). Sensors can be localized in the mitochondrial matrix via addition of the N-terminal targeting sequence of cytochrome c oxidase (COX) subunit VIII (Rizzuto et al., 1995). The use of tandem repeats of COX VIII targeting sequences greatly improves the mitochondrial enrichment of optogenetic indicators (Filippin et al., 2005). Importantly, the high Ca²⁺ transients in the ER and mitochondria can saturate classical high-affinity GECIs (dissociation constant [K_D] values in the range of 100–700 nM). Accordingly, both green and red low-affinity sensors with K_D values > 1 μM have been developed, which are better suited for non-saturating quantification of intra-organellar Ca²⁺concentrations (Akerboom et al., 2013, de Juan-Sanz et al., 2017, Henderson et al., 2015, Palmer et al., 2004, Suzuki et al., 2014, Wu et al., 2014). As an additional caveat, pH changes interfere with mitochondrial Ca²⁺ measurements (Marland et al., 2016, Whitaker, 2010), demanding careful calibrations or the use of pH-insensitive GECIs (Rodriguez-Garcia et al., 2014). MitoSypHer, a mitochondrially targeted, pH-sensitive YFP with a pKa of 8.7, was derived from the H₂O₂ sensor HyPer by mutating one of the two H₂O₂-sensing cysteines (Poburko et al., 2011). Live-cell imaging of mitoSypHer-expressing astrocytes revealed that the uptake of synaptically released glutamate by astrocytes acidifies the mitochondrial matrix, which decreases the cytosol-to-matrix H⁺ gradient. As a consequence, glutamate uptake impairs mitochondrial respiration and transiently reduces local O₂ demand by astrocytes (Azarias et al., 2011, Perreten Lambert et al., 2014). Using the same probe, Marland et al. (2016)detected matrix acidification of presynaptic mitochondria during sustained transmitter release.

Although selective expression of genetically encoded sensors in mitochondria is well established, it seems surprising how little has been achieved in the area of mitochondrial manipulation using optogenetic actuators. Over two decades ago, successful expression of the archaeal bacteriorhodopsin in yeast mitochondria was reported (Hoffmann et al., 1994). In this construct, N-terminal addition of the signal sequence from COX IV directed bacteriorhodopsin to the inner mitochondrial membrane, with the N terminus localized in the intermembrane space and the C terminus in the matrix. In this orientation, bacteriorhodopsin provided an additional proton-driving force over the inner membrane upon illumination. Given the importance of mitochondrial energy supply for brain physiology and its diverse roles in aging and pathophysiological conditions (Gribkoff et al., 2015), it would be exciting to see similar tools developed for application in mitochondria of mammalian cells. In C. elegans, miniSOG has been successfully localized to the outer mitochondrial membrane by fusing it to TOMM-20, enabling efficient ablation of neurons via light-induced generation of ROS (Xu and Chisholm, 2016).

Autophagosomes and Lysosomes. Autophagy is an essential cellular clearing mechanism that conveys cytosolic components, such as damaged proteins or organelles, to lysosomes for degradation. Neurons in particular rely on autophagy because they do not dilute cellular waste by cell division. Impaired autophagy is linked to various neurodegenerative diseases, whereas excessive autophagy induces neuronal cell death (Damme et al., 2015). Various marker proteins have been identified for lysosomes, including lysosome-associated membrane proteins (LAMP1 and LAMP2), lysosome integral membrane protein 2 (LIMP2), and CD63 (Saftig et al., 2010). Interestingly, autophagic flux can be followed by monitoring the state and localization of microtubule-associated protein 1 light chain 3 (LC3), which is converted into the membrane-bound form, LC3-II, and subsequently enriched at the inner membrane during formation of the double-membrane autophagosome. Thus, when tandems of RFP-EGFP are attached to the intraluminal N terminus of LC3, they enable real-time imaging of autophagosome maturation (Kimura et al., 2007). Initially, autophagosomes have a neutral pH, resulting in orange fluorescence of both RFP and EGFP. However, when they fuse with lysosomes, autophagosomes become acidified, which quenches EGFP (pKa 5.9), whereas RFP (pKa 4.5) remains fluorescent. RFP-EGFP-LC3 and related constructs have been widely applied to investigate autophagic flux in healthy neurons (Wang et al., 2015) as well as in the contexts of Alzheimer’s disease (Lee et al., 2010), Parkinson’s disease (Ginet et al., 2014), and Niemann-Pick disease type C1 (Elrick et al., 2012, Ordonez et al., 2012).

Subcellular optogenetic approaches that aim to manipulate the physiology of autophagosomes and lysosomes are still at an early stage of development. A light-driven proton pump for lysosomes, termed lyso-pHoenix, was recently developed by insertion of an optogenetic actuator-sensor cassette consisting of a luminal pHluorin and Arch3 into the first intraluminal loop of CD63 (Rost et al., 2015). Lyso-pHoenix strongly co-localizes with LAMP-2 and represents the first optogenetic tool that enables light-controlled acidification of lysosomes; however, it remains to be applied in neurons.

Optogenetic Control of Organelle Positioning. Cytoskeleton-interacting motor proteins position organelles within the cell, but researchers are only beginning to understand how organellar localization and function affect cellular physiology. Based on the dimerization system TULIP, van Bergeijk et al. (2015) developed a method that enabled photo-controlled repositioning of peroxisomes, recycling endosomes, and mitochondria. Repositioning was elicited by blue light-triggered heterodimerization of organellar anchor proteins, fused to LOVpep, with ePDZb1, which was fused to the cytoskeletal motor proteins kinesin, dynein, or myosin (van Bergeijk et al., 2015). These authors showed that, in hippocampal neurons, light-induced, myosin-Vb-mediated recruitment of either peroxisomes or recycling endosomes induced their entry into dendritic spines. They also showed that axonal mitochondria could be mobilized by photocontrolled kinesin coupling, whereas light-induced synaptophilin binding stalled mitochondrial movement. Finally, van Bergeijk et al. (2015) also demonstrated that kinesin-induced accumulation of recycling endosomes in growth cones enhanced outgrowth, whereas endosome removal by dynein interaction reduced axonal outgrowth. These photocontrolled organelle positioning systems were later applied in C. elegans, which required redesigning the constructs by codon optimization and use of C.elegans homologs of the targeting sequences (Harterink et al., 2016).

Developing and Applying Subcellular Optogenetic Tools

The extensive variety of optogenetic tools currently available provide many opportunities to investigate specific subcellular compartments, and readers will no doubt come up with many new ideas for how to target their subcellular structure of interest. However, only a few of the available optogenetic tools are easy to use, and a number of general principles have evolved for subcellular targeting of photosensitive proteins. In this section, we provide advice for future development and application of subcellular optogenetic tools.

Choosing the Most Appropriate Optogenetic Tool

The choice of which optogenetic tool to use to evoke a specific cellular output is not always a straightforward one. Even in the case of ChR—the obvious tool of choice for generating fast membrane depolarization—there are many varieties from which to choose that have different biophysical properties, including photocycle kinetics, activation spectra, light and voltage sensitivities, inactivation, and relative ion conductance (Schneider et al., 2015).

To begin with, every photoreceptor is characterized by its specific dynamic range of light activation. While rhodopsins do not show any detectable activity in the dark, modular-built photoreceptors show an intrinsic equilibrium between the inactive and the active state, resulting in considerable dark activity. If the basal activity of a photoreceptor is not carefully controlled, then it can disturb the intracellular milieu, produce artificial cellular responses, and affect reference readouts of “control” conditions. Whereas high expression levels of microbial ion pumps are preferred for hyperpolarizing the plasma membrane, low expression of actuators with dark activity is advised. For example, in the case of PaRac1, expressing it under control of a TetCMV promoter enables the effects of its dark activity to be minimized by titration of doxycycline, allowing its expression level to be optimized for potent light activation (Wu et al., 2011b).

Photoreceptor on and off kinetics are also important criteria for identifying the best-suited tool for mimicking a cellular process. Microbial rhodopsins are ideal for modulating neural excitability with millisecond precision. Analysis of protein expression during brain development, on the contrary, requires tools with long-lived active states, such as LOV- or cryptochrome-based transcriptional control systems. Bistable photo-interconvertible systems are powerful tools for analyzing the effects of the duration-output relationship of cellular activities because their activation can be maintained without continuous illumination and can be terminated by application of a second light pulse of a different color. Examples of bistable optogenetic systems applied in neuroscience include engineered SFRs, natural bistable rhodopsins from invertebrates, and melanopsins (Bailes et al., 2012, Bamann et al., 2010, Berndt et al., 2009, Spoida et al., 2016, Yizhar et al., 2011). In comparison with photoactivation, the effective off-switching of an actuator requires higher light intensities to inactivate all molecules, and wavelengths have to be chosen that do not excite its dark state. Bistable rhodopsins with active states that have red-shifted absorption relative to the respective dark state are preferable because all rhodopsins exhibit significant dark-state absorption at wavelengths shorter than their main absorption band (β-band absorption). Accordingly, SFR channel closure is most effective when triggered by a longer wavelength (e.g., with orange or red light). In this regard, melanopsins, neuropsins, and parapinopins are well qualified for optogenetic applications because their two photo-interconvertible states are spectrally well separated, and their active states are red-shifted compared with their respective dark states, allowing them to be effectively switched off (Figure 4; Koyanagi et al., 2004, Yamashita et al., 2014). Off-switching of these proteins is also facilitated by the increased quantum efficiencies of the off reactions. By contrast, JellyOP forms a blue-shifted active state upon green illumination, and the dark state recovers by thermal isomerization within minutes but cannot be retrieved in a light-dependent manner (Koyanagi et al., 2008).

In general, action spectra and light sensitivity are crucially important parameters, especially when designing multi-wavelength experiments that aim to independently activate several actuators or a combination of actuators and sensors. For example, because of the abovementioned β-band absorption in the blue range, rhodopsins work best in combination with fluorescent sensors with red-shifted excitation (Ziegler and Möglich, 2015). Because the brain is most permeable to red light between 600 nm and 700 nm (Jacques, 2013), red light-activated systems or 2-photon excitation enable deeper tissue penetration at moderate light intensities, the latter requiring photoreceptors with adequate 2-photon cross-sections (Prakash et al., 2012). In contrast, activation of UVR8 by UVB light should be minimized to prevent UVB-induced DNA damage (Müller et al., 2013b).

Similar biophysical criteria should also be considered when using or designing sensor proteins. Most importantly, the dynamic range in which fluorescence is sensitive to the cellular parameter of interest should match the range of expected fluctuations that give rise to high signal-to-noise ratios (Tantama et al., 2012). Moreover, sensors should be fast enough to track dynamic changes and should be resistant to photobleaching and pH-insensitive, especially when used in small compartments.

Tool Expression

Even a perfect optogenetic tool would be useless if it cannot be functionally expressed in the target cell of interest. Targeted expression requires efficient gene delivery by transfection, electroporation, microinjection, or viral delivery using replication-deficient shuttle viruses, such as a lentivirus or AAV. Depending on the presented coat proteins, the chosen virus might confer increased affinity for a specific neuronal cell type or a preferred route to enter neurons; for example, via axon terminals (Tervo et al., 2016). Commonly, cell-type-specific gene activation is achieved by transcriptional control using specific promoter sequences or by site-specific recombination technologies (e.g., Cre/LoxP technology), as often applied for specific expression of optogenetic tools in transgenic rodents. Activity-dependent expression using immediate-early gene promoters and retrograde or anterograde neuronal labeling are examples of more sophisticated targeting approaches (Packer et al., 2013).

Codon optimization of genes can also help to improve protein translation, especially when adapting proteins across large evolutionary distances. Unfortunately, correct protein folding depends on many factors, including cell-type-specific interacting proteins (e.g., chaperones) and lipids, and the effectiveness of an optogenetic system is generally difficult to predict when applying it in a different cell type or species. Independent of protein expression, output functions can also differ depending on the respective expression system. For example, Opn4 has been reported to exclusively activate the G_q pathway in retinal ganglion cells and preferentially couple to G_q in neurons but activate both G_i/o and G_q/11 signaling in HEK293 cells (Bailes and Lucas, 2013, Spoida et al., 2016).

Whereas UVR8 and GFP-related fluorescent proteins use amino acid-based light absorption, organic chromophores are required for the activation of most other photoreceptors. Accordingly, sufficient chromophore availability in target cells is essential for their activation. This is especially important in the case of vertebrate rhodopsins, which cannot recycle 11-cis retinal. However, cone photoreceptors and bistable rhodopsins from invertebrates can offer alternative tools for light activation of GPCR signaling. In addition, although sufficient levels of endogenous retinal are available in mammalian cells for its functional incorporation into rhodopsin-based optogenetic tools, retinal has to be externally provided (for example, in food supplements) to enable its activation in worms and flies (Nagel et al., 2005, Schroll et al., 2006). In the case of phytochrome-based systems that bind bilins, bilins either need to be added to the medium or to animal food, or they need to be synthesized in animal cells when coexpressed with genes encoding the respective enzymes from plants and cyanobacteria (Müller et al., 2013c). In general, high chromophore binding affinities are required for tissues with low cofactor concentrations. Accordingly, the affinity of the apoprotein for the respective chromophore severely limits the application of chromophore analogs that carry extra side groups to shift their absorption bathochromically because chromophore modifications usually reduce the K_D value for apoprotein binding (AzimiHashemi et al., 2014).

Subcellular targeting is a major advantage of optogenetic tools; however, the introduction of light-sensitive proteins into cells and into subcellular compartments comes with caveats. Their expression might affect cellular health per se, which has to be carefully controlled both at a functional and morphological level, including passive and active membrane properties, synaptic transmission, trafficking of endogenous proteins, spine density, neurite outgrowth, and cell number. For example, fusing ChR2 to red fluorescent proteins, such as mCherry and tdTomato, can lead to intracellular protein inclusions, reduced membrane expression, and smaller photocurrents compared with ChR2-eYFP (enhanced yellow fluorescent protein) fusions and may ultimately affect neuronal health (Asrican et al., 2013, Madisen et al., 2012). Long-term, high-level expression of ChR2 in the rat somatosensory cortex by in utero DNA electroporation was reported to induce aberrant morphology of L2/3 pyramidal cells in the absence of any light activation, an effect that was diminished when ChR2 was expressed by viral transduction (Miyashita et al., 2013). Moreover, activation of optogenetic actuators might exert paradoxical effects. Sustained activation of the chloride pump eNpHR3.0 causes depolarizing shifts in the reversal potential of chloride-conducting GABA_A receptors, adversely increasing the probability of spiking after photoactivation (Raimondo et al., 2012). Although the same authors did not observe a change in spiking probability after activation of the proton-pumping Arch, Mahn et al. (2016) showed that repolarization after a step-like ending of Arch3 activity triggers APs and unintended transmitter release. Interestingly, Arch-mediated cytosolic alkalization of nerve terminals suppresses evoked transmitter release but increases spontaneous transmitter release (El-Gaby et al., 2016, Mahn et al., 2016) and might activate acid-sensing ion channels (Ferenczi et al., 2016). Finally, as another note of caution when interpreting results from experiments involving optogenetic activation and inhibition, on the network level, transient optogenetic manipulation can affect connected but functionally independent brain regions. These acute off-target effects might alter the behavior of an animal and lead to conflicting results compared with experiments that involve chronic neuronal ablation or inactivation (Otchy et al., 2015).

Combining Target Sequences, Actuators, and Sensor Proteins

Finding appropriate and efficient trafficking signals for subcellular targeting strategies is not a straightforward undertaking. Several databases on subcellular trafficking signals exist (King and Guda, 2007, Negi et al., 2015), but unfortunately, none of these cover neuron-specific targeting motifs. Common targeting motifs for neuron-specific structures are synaptophysin and vesicular glutamate transporter 1 (VGLUT1) for synaptic vesicles, PSD95 for the PSD, and ankyrin G-interacting domains for the axon initial segment. However, these proteins also exemplify the need for careful controls when establishing subcellular targeting strategies. Although we are not aware of any reported synaptophysin overexpression artifacts in mammalian neurons (but see Alder et al., 1995, for Xenopus), Royle et al. (2008) reported that a truncated synaptophysin-pHluorin (lacking the 4^th transmembrane domain) showed incorrect targeting. This is in line with observations on other tetraspanins, in which deletion of one or more helices caused ER retention (Pols and Klumperman, 2009). Overexpression of VGLUT1 in neurons increases miniature excitatory postsynaptic current (mEPSC) amplitudes (Weston et al., 2011, Wojcik et al., 2004) and converts GABAergic neurons into GABA and glutamate co-releasing neurons (Takamori et al., 2000), indicating that introduction of VGLUT1-pHluorin for imaging purposes may dramatically alter synaptic transmission. Similarly, overexpression of PSD95 strongly increases synaptic strength and saturates long-term potentiation (Béïque and Andrade, 2003, Ehrlich and Malinow, 2004), which prompted researchers to establish alternative PSD-targeting molecules, including a truncated version of PSD95 (Hayashi-Takagi et al., 2015), homer1c, or a PSD-binding intrabody (Sinnen et al., 2017). AIS-targeting strategies involving parts of the Na_v II-III intracellular loop were reported to perturb trafficking of endogenous Na⁺ and K⁺ channels, which disrupted channel clustering at the AIS and impaired neuronal firing (Garrido et al., 2003, Grubb and Burrone, 2010, Wu et al., 2011a). Finally, modifications intended to bring about subcellular targeting might also impair the function of optogenetic tools. We experienced that the same targeting motifs that route functional Arch3 to SVs or to lysosomes render ChR2 non-functional (Rost et al., 2015). The reason for this is unclear, but the different multimerization properties of trimeric microbial pumps (Yoshimura and Kouyama, 2008) and dimeric ChR variants (Kato et al., 2012, Müller et al., 2011) might contribute to this problem.

These examples highlight the fact that trafficking motifs, tags, or fluorophores can impair the performance of optogenetic tools, whereas overexpression of optogenetic tools and trafficking motifs can interfere with the proper function of endogenous proteins. When developing a new targeting approach, it is advisable to start with the full-length protein for targeting. Next, the mutation of functional sites is recommended to prevent overexpression artifacts. Ultimately, reducing the size of the expression construct to a minimum might improve expression efficiency, as in the case of viral gene delivery, where gene size is limited by the capacity of the viral vector system (approximately 4.9 kb for AAVs and 9–10 kb for lentiviruses).

Outlook

Despite the tremendous progress achieved to date in the development of optogenetic tools and subcellular targeting strategies, a number of essential tools and efficient targeting strategies for various neuronal compartments remain to be developed.

Tools

First, the neuroscience community needs potent tools to silence neuronal activity with light (see also Wiegert et al., 2017). Existing inhibitory tools include microbial proton, chloride, and sodium pumps, which require high light intensities and show considerable side effects, as well as anion-selective ChRs, which can dramatically alter the chloride reversal potential and become ineffective when the membrane resting potential is lower than the reversal potential for chloride (Mahn et al., 2016, Raimondo et al., 2012). The ideal candidate to overcome these challenges would be a selective light-gated K⁺ channel that mimics endogenous K⁺ conductances. In addition to the K⁺ channel, efficient optogenetic tools to control second-messenger signaling cascades are still lacking. Foremost, a light-activated, selective Ca²⁺ channel would be most useful here to directly trigger intracellular Ca²⁺-dependent signaling events, ranging from activation of Ca²⁺-sensitive enzymes and transcription factors to stimulation of structural plasticity and neurotransmitter release. Such a channel remains to be created, but, in the meantime, alternative approaches are emerging (Ma et al., 2017), including optogenetic Ca²⁺ uncaging (Fukuda et al., 2014), photocontrolled activation of Ca²⁺-selective Orai1 ion channels (Ishii et al., 2015), and light-gated Ca²⁺ release from the ER (Feldbauer et al., 2016, Kyung et al., 2015). Similar to Ca²⁺ signals, cyclic nucleotides are important second messengers that have defined output functions in different neuronal compartments, depending on the local activity of downstream protein kinases, phosphodiesterases, nucleotide exchange factors, and CNG and HCN (hyperpolarization-activated, cyclic nucleotide-modulated) channels. The palette of available light-activated cAMP and cGMP cyclases (bPAC, PACα, oaPAC, RhGC), the light-activated phosphodiesterase LAPD, and numerous genetically encoded cAMP and cGMP sensors (Gorshkov and Zhang, 2014, Rich et al., 2014) now offer potent tools with which to investigate the regulatory role of these molecules in neuronal differentiation, gene expression, excitability, and synaptic transmission. Membrane lipids also shape local signaling cascades and influence neuronal outgrowth, intracellular trafficking, excitability, and exo- and endocytosis (van Meer et al., 2008). Although tools have emerged to enable light-mediated control of PIP₂ and PIP₃ levels (Idevall-Hagren et al., 2012, Kakumoto and Nakata, 2013, Toettcher et al., 2011), optogenetic manipulation of membrane lipids remains in its infancy and will hopefully be advanced by tools that control other membrane components, such as cholesterol and sphingolipids, and also lipid-derived hydrophobic messengers, such as endocannabinoids or diacyglycerol. When combined with subcellular targeting strategies, these systems will enable cell-type-specific, spatially restricted, and temporally precise fine-tuning of the plasma membrane composition and promise exciting applications for neuroscientists and cell biologists. We further expect light-induced dimerization systems to replace chemically induced protein dimerization and translocation (DeRose et al., 2013) because most chemical dimerizers suffer from irreversibility. Subcellular application of optogenetic dimerizers would be facilitated by generation of transgenic animals that constitutively express one part of the dimer targeted to a subcellular compartment (e.g., CRY2 targeted to the plasma membrane, mitochondria, or PSD), whereas the interaction partner (in this case CIB1) coupled to diverse target proteins could be provided by viral gene delivery.

Subcellular Targets

The exquisitely refined cellular machinery of neurons is governed by a variety of highly localized signaling events (e.g., Ca²⁺ sparks, GPCR signals, local translation). Optogenetic approaches will only match these processes when they can be appropriately targeted. Clearly, efficient targeting strategies for several subcellular compartments remain to be developed. For instance, no trafficking motif has been described that specifically targets ChRs and hyperpolarizing pumps into the plasma membrane of presynaptic terminals, despite the fact that several membrane proteins (e.g., GPCRs and voltage-gated Ca²⁺ channels [VGCCs]) show almost exclusive expression in the presynapse. A presynaptic Ca²⁺-selective ChR would be a useful tool to specifically elevate Ca²⁺ in axonal boutons and to probe the strength of individual synapses. In the future, optogenetic Ca²⁺ cages (Fukuda et al., 2014) could be targeted to the presynaptic terminal; however, the consequences of long-term expression of presynaptic Ca²⁺ buffers remain to be determined. Finally, specific subcellular targeting strategies are not only relevant for studying neurons but also for investigating astrocytes, oligodendrocytes, microglia, and brain-resident stem cells.

The past decade has been characterized by the rapid development and application of optogenetic methods, many of which have become routine applications in neuroscience labs. Recently, identification of a growing number of subcellular trafficking motifs has enabled the precise intracellular localization of these tools. We envision that this development will culminate in all-optical interrogation of any subcellular compartment, facilitating the development of completely new approaches to studying the physiology and pathophysiology not only of the brain but also of other organs.