Address correspondence and reprint requests to Ethan K. Scott, The University of Queensland, School of Biomedical Sciences and The Queensland Brain Institute, Brisbane, 4072 QLD, Australia. E-mail: email@example.com
Over recent years, several groundbreaking techniques have been developed that allow for the anatomical description of neurons, and the observation and manipulation of their activity. Combined, these approaches should provide a great leap forward in our understanding of the structure and connectivity of the nervous system and how, as a network of individual neurons, it generates behavior. Zebrafish, given their external development and optical transparency, are an appealing system in which to employ these methods. These traits allow for direct observation of fluorescence in describing anatomy and observing neural activity, and for the manipulation of neurons using a host of light-triggered proteins. Gal4/Upstream Activating Sequence techniques, as they are based on a binary system, allow for the flexible deployment of a range of transgenes in expression patterns of interest. As such, they provide a promising approach for viewing neurons in a variety of ways, each of which can reveal something different about their structure, connectivity, or function. In this study, the author will review recent progress in the development of the Gal4/Upstream Activating Sequence system in zebrafish, feature examples of promising studies to date, and examine how various new technologies can be used in the future to untangle the complex mechanisms by which behavior is generated.
One of the overarching goals of neuroscience is to explain how animals sense, interpret, and respond to the world around them. As such, identifying the brain regions, connected networks, and specific types of neurons driving a range of behaviors has been a major goal of the neuroscience community. The approach is conceptually simple. First, the nervous system must be described in terms of the neurons composing it, and then these neurons must be characterized with regard to their connectivity, activity, and roles in behavior. Once enough circuits and cell types are described, their connectivity will converge into a coherent whole. The types of stimuli or behaviours that coincide with activity in each will inform us as to their probable role in the perception of stimuli or generation of the behavior. The distinct and sometimes overlapping behavioral effects of losing different circuits will help to confirm this functional role, and to explain how different groups of neurons interact.
In reality, the complexity of the human brain and its function far outstrip our current ability to describe it in this way. With its hundreds of distinct anatomical features and thousands of cell types, we have only begun to describe this network and the flow of information through it. The relative simplicity and experimental accessibility of model systems’ brains offer ways around these obstacles. For example, the anatomical description of the Caenorhabditis elegans nervous system, with its 302 neurons and roughly 5000 synapses, is complete at the cellular level (White et al. 1986). Many of these neurons have also had behavioral functions ascribed to them (reviewed by Chatterjee and Sinha 2007). Of course, as model systems become more complex, so do the challenges of describing them.
Over the past two decades, zebrafish has emerged as a powerful model for studying the nervous system and its function. One reason for this is the simple feature that zebrafish are transparent at early life stages. This, combined with external development, means that the brain can be observed directly in intact, behaving animals. Additionally, many of the forward genetic tools that have been staples of research in C. elegans and Drosophila are now available in zebrafish. At the same time, zebrafish share portions of their neural architecture with other vertebrates, including humans. This means that zebrafish combine certain useful traits from invertebrate model systems (simplicity, small size, easy imaging, and tractable genetics) with those of mammalian systems (complex behaviors and similarity to major portions of the human nervous system). At the same time, some techniques are still missing from the zebrafish genetic toolbox (homologous recombination, facile reverse genetics, and clonal genetic screens), and certain regions of the mammalian nervous system (notably the cerebral cortex) lack close counterparts in zebrafish. As with any model system, the effective use of zebrafish requires combining its technical and inherent biological strengths.
An important step in the development of the zebrafish model system has been the introduction and refinement of transgenes using the Gal4/Upstream Activating Sequence (UAS) system. As I will describe in this review, these tools are poised to deliver faster and more direct methods for describing the structure and function of the nervous system and how behaviors are generated. A few recent reviews have addressed aspects of this topic. Specifically, two groups have recently reviewed Gal4/UAS tools in zebrafish, with especially clear and detailed descriptions of the history of Gal4/UAS in zebrafish, the technical details of various Gal4 constructs, and the methods for generating transgenic lines (Asakawa and Kawakami 2008), and the past uses of Gal4/UAS in a variety of model systems, UAS lines that are available or in development, and limitations and challenges associated with these tools in zebrafish (Halpern et al. 2008). Okamoto et al. (2008) and Higashijima (2008) have recently highlighted some of the more elegant uses of transgenic lines to describe neural anatomy and activity in zebrafish, and McLean and Fetcho (2008) have focused on imaging techniques in the zebrafish spinal cord. Accordingly, I will cover these topics briefly if at all. Here, I will focus on the utility and future potential of the Gal4/UAS system for describing the anatomy, physiology, activity, and behavioral relevance of different brain regions and neuronal populations in zebrafish.
The Gal4/UAS system as a tool for neuroscience
The Gal4/UAS system is drawn from yeast, where it regulates genes involved in the metabolism of galactose. It comprises two parts: the Gal4 protein and its UAS (Giniger et al. 1985; Giniger and Ptashne 1987). In cells where the Gal4 gene is expressed, the Gal4 protein targets the UAS sequence (Guarente et al. 1982; West et al. 1984), thus driving expression of any open reading frame immediately downstream of the UAS. In a typical experiment, Gal4 is driven in an expression pattern of interest by linking the Gal4 gene to a known enhancer for that pattern. As a second step, a UAS-linked transgene is introduced, which is then expressed in the same pattern as the Gal4 gene. In principle, this allows for any gene (linked to UAS) to be expressed in any pattern, so long as there is a known enhancer with which to drive Gal4.
Over the past 15 years, the Gal4/UAS system has facilitated hundreds of studies into the structure and function of various nervous systems, most notably Drosophila (Brand and Perrimon 1993, and reviewed in Elliott and Brand 2008). The key to the system’s efficacy is that a given Gal4 line can be used repeatedly with a variety of different UAS-linked transgenes. Conversely, a UAS-linked transgene can be expressed in any cell type for which a Gal4 driver is available. This means that a study can begin with an anatomical description of the neurons composing a Gal4 expression pattern of interest, and then use different transgenes to probe these neurons’ activity. Finally, other transgenes can be used to manipulate the function of the neurons in order to gauge their role in generating behavior. In Drosophila, this has allowed for transsynaptic tracing (Yoshihara 2002), calcium imaging (Wang et al. 2003, 2004; Suh et al. 2004), observation of neurotransmitter release (Ng et al. 2002), and the silencing (Sweeney et al. 1995; Kitamoto 2001) and activation (Schroll et al. 2006, Zhang et al. 2007) of neural circuits. These studies and others have contributed greatly to our understanding of activity in the Drosophila nervous system, and how the functions of particular neurons contribute to behavior.
While the Gal4/UAS system was first reported in zebrafish a decade ago (Scheer and Campos-Ortega 1999), appeal was initially hampered by concerns over Gal4 toxicity (reviewed by Asakawa and Kawakami 2008; Halpern et al. 2008) and the inefficiency of transgenesis in the zebrafish model system. As two transgenes are generally needed for expression of the target protein, and as the system is most powerful when a range of UAS-linked transgenes can be combined with an array of Gal4 expression patterns, a large number of transgenic lines are needed for the system to be fully effective. For this reason, the recent expansion of the Gal4/UAS toolbox in zebrafish awaited the development of techniques for creating a large number of transgenic lines quickly and easily.
Targeting circuits: methods for driving Gal4 expression
In early studies using Gal4/UAS in zebrafish, Gal4 expression was driven by known enhancers that were included in the transgenesis construct. Such directed expression included neural patterns in the retina (Scheer et al. 2001; Campbell et al. 2007; Xiao and Baier 2007), tectum (Sato et al. 2007), forebrain (Jeong et al. 2007), various parts of the developing embryonic nervous system (Scheer and Campos-Ortega 1999; Koster and Fraser 2001b; Hans et al. 2004; Inbal et al. 2006), and in non-neural tissues such as the pancreas (Zecchin et al. 2007) and circulatory system (Beis et al. 2005). Another approach for expressing transgenes is through enhancer trapping (ET) or gene trapping (GT). ET and GT screens involve making a very large number of independent lines for a transgene that is linked to either a basal promoter in the case of ET, or a splice acceptor sequence in the case of GT. In either case, the transgene is not driven unless its insertion site in the genome provides an endogenous enhancer (for ET) or an endogenous gene with a splice donor (for GT). In the cases of successful insertions, the marker gene are expressed in a pattern similar to that of the endogenous gene. Initially, the genes for fluorophores were used in ET (Balciunas et al. 2004; Parinov et al. 2004; Ellingsen et al. 2005) and GT (Kawakami 2004) screens in zebrafish, and these screens resulted in dozens of lines with distinct and specific expression of fluorophores in a variety of tissues. The success of these strategies, and the advent of reliable UAS-marker lines led others to perform ET (Scott et al. 2007; Asakawa et al. 2008; Ogura et al. 2009) and GT (Davison et al. 2007; Asakawa et al. 2008) screens, this time inserting the Gal4 gene rather than a fluorophore. As they can be paired with existing and future UAS lines, the hundreds of resulting Gal4 ET and GT lines offer genetic access to a wide range of tissues and cell types within the zebrafish nervous system.
Enhancer trap/GT strategies and the use of selected promoters each provide advantages, and are in some ways analogous to forward and reverse genetic strategies, respectively. In ET or GT (as in forward genetics), the experimenter queries the genome in an unbiased fashion, allowing expression patterns (or phenotypes) to arise, and selecting those of greatest interest for further study. Advantages of this approach are that it is easy to generate a large number of different expression patterns, and unknown enhancers may be discovered in the process. ET and GT, however, are not practical for targeting chosen expression patterns of interest, and generally result in broad expression patterns, often including non-neural tissues (Davison et al. 2007; Scott et al. 2007; Asakawa et al. 2008; Ogura et al. 2009). The breadth of most ET patterns currently restricts their use for studying discrete cell types in dense and complex regions of the brain. This may explain why behavioural studies using these lines thus far have focused on spinal circuits and relatively simple escape behaviour (Szobota et al. 2007; Asakawa et al. 2008). The use of known enhancers to drive Gal4 requires more work for the generation of each transgenic line, but this approach makes more sense when the target is known in advance, and when a tightly restricted expression pattern is needed. The mutually exclusive advantages of each approach ensure that both ET/GT lines and lines generated with chosen promoters will continue to be used into the future.
As Gal4 and UAS work in trans, the Gal4/UAS system alone does not allow multiple UAS-linked transgenes to be expressed in distinct cell types. For example, a fish bearing a retinal-specific Gal4 transgene, a tectum-specific Gal4 transgene, UAS:GFP, and UAS:RFP will have red and green fluorescence in both retinal and tectal cells. This lack of flexibility precludes studies that aim to observe or manipulate specific parts of a circuit differentially. In the future, this could be overcome by combining Gal4/UAS transgenes with those for expression systems that would not interfere, such as the LexPR/LexA (Emelyanov and Parinov 2008) or cre/LoxP (Liu et al. 2008) systems.
Seeing circuits: Gal4/UAS tools for neuroanatomy in zebrafish
As mentioned above, the primary advantage of using Gal4/UAS in the nervous system is that a given circuit or cell type can be targeted with a variety of different UAS-linked transgenes. This provides invaluable flexibility for describing the structure, connectivity, and cellular composition of brain regions of interest. In studies to date, a majority of zebrafish Gal4 expression patterns have been reported using UAS-linked fluorophores that reside in the cytoplasm, although UAS lines have now been developed that target fluorophores to subcellular compartments such as nuclei and mitochondria (Halpern et al. 2008). Such labeling is excellent for revealing the positions of cell bodies, and in some cases can be used to image major tracts. Connectivity, however, depends on axons and dendrites that, as fine structures, are typically membrane-rich and cytoplasm-poor. Accordingly, these are better observed using membrane-targeted UAS-linked fluorophores (Halpern et al. 2008).
Even clear staining of an expression pattern’s membrane structure does not necessarily reveal the neurons’ connectivity. This is because, although the termination points of the neurites should be clear, their identities as axons or dendrites may not be. Additionally, membrane staining will not reveal en passant synapses that may be important to the connectivity of the circuit in question. Both of these issues could be addressed in the future by using markers targeted to pre- and post-synaptic terminals. Although they have only been used thus far in transient transfection experiments, UAS:synaptophysin-GFP (Meyer and Smith 2006) and UAS:PSD95-GFP (Niell et al. 2004) have clearly labeled the axonal termini of retinal ganglion cells (RGCs) and the dendritic arbors of tectal neurons, respectively. It is likely that, in the context of a stable UAS line, they and other such markers will aid in the description of Gal4-expressing neurons’ connectivity.
Certain UAS-linked markers provide further flexibility in describing an expression pattern’s structure. One example is Kaede (Ando et al. 2002), a photoconvertible fluorophore that initially fluoresces in a green wavelength. With exposure to violet or UV light, it irreversibly converts into a red-emitting fluorophore. We have used this feature to describe the coarse connectivity of Gal4 ET lines (Scott et al. 2007). By focusing violet light on a restricted region of cell bodies in an expression pattern, we can convert a majority of the Kaede in those cells to its red form. As this red Kaede diffuses down the axonal shafts of these neurons, it reveals the axons belonging specifically to the irradiated neurons. In this way, we can determine which parts of an arborization field are innervated by neurons in the restricted area of irradiation. Examples of this are shown in Fig. 1.
This sort of mapping could be taken further by bringing the Brainbow system into zebrafish. Brainbow involves stochastic Cre-mediated recombination at Lox sites in a transgenic construct bearing the genes for several fluorophores (Livet et al. 2007). As these recombinations occur differently in different cells, each neuron expressing the construct has one of numerous possible combinations of fluorophores. The morphologies of many neurons in a dense mass can then be mapped, as all of the structures belonging to a given cell are marked with a unique (or rare) fluorophore profile. This has resulted in spectacular images and the spatial untangling of neurons and their axons in the mouse brain (Livet et al. 2007). It should have the same potential in zebrafish, but without the need for dissection and fixation of the brain. All that should be necessary is the generation of a UAS:Brainbow line, which would be combined with a Gal4 line of choice and a transgene capable of delivering Cre.
Genetic Golgi staining: observing individual neurons using Gal4/UAS
Complex nervous systems comprise hundreds of distinct nuclei, ganglia, tracts, and larger structural regions. While descriptions of these structures have been critical to our understanding of the brain and its function, this is far from the end of the story. This is because these structures themselves are composed of dozens or hundreds of different types of neuron, each with its own distinct anatomy, connectivity, physiology, and behavioral relevance. In order fully to describe the workings of the nervous system or one of its parts, we must start with a description of the cells composing it.
The description of individual neurons goes back more than a century to the time of Golgi and Ramón y Cajal. While Golgi staining has been used in zebrafish (e.g. Fuller and Byrd 2005), it has been supplanted by techniques that make better use of one of the model’s strengths: the ability to observe internal structures in intact, and in many cases live, animals. Some of these involve the backfilling of cell bodies from an axonal target region using dyes (e.g. Bernhardt et al. 1990; Eisen et al. 1986; Hale et al. 2001) or horse radish peroxidase (e.g. Myers et al. 1986), which allows for histochemical labelling. Alternatively, dyes can be introduced using biolistics (Connaughton et al. 2004). These studies have made important contributions to our understanding of the anatomy and development of neural circuits, especially in the spinal cord, but as they are not genetically based techniques, they cannot be targeted to particular cell types, and are not compatible with Gal4/UAS tools.
Another way to express Gal4 stochastically is by combining the Gal4/UAS and Cre/LoxP (Dong and Stuart 2004) systems. Sato et al. (2007) have done this by injecting a plasmid containing a promoter for tectal neurons linked to the Gal4 gene (see Fig. 2b). Intervening, however, is the DsRed2 gene, which itself is flanked by LoxP sites. The result is that DsRed2, and not Gal4, is expressed in tectal neurons. Stochastic expression of Cre, provided from a second plasmid, leads to the excision of the DsRed2 gene in a small number of these neurons, resulting in the expression of the Gal4 protein. A third plasmid, bearing UAS:GFP, allows these Gal4-positive cells to be labeled in green, and described anatomically. They were also able to over-express an axon guidance gene, ephrinB2a, in green neurons by using a UAS:ephrinB2a:UAS:GFP plasmid, and thus demonstrate its cell-autonomous role in axon targeting.
Approaches for simultaneously observing and manipulating small numbers for neurons rely on expression off of multiple UAS transgenes, or on internal ribosomal entry sites (IRES). Co-expression of such transgenes, however, does not occur in all Gal4-positive cells (Koster and Fraser 2001b; Hua et al. 2005; Sato et al. 2007), and this can result in unwanted noise in the data, as some cells are scored as manipulated when they are not actually expressing the manipulating transgene, or in the case of IRES, are expressing it at variable levels (Mizuguchi et al. 2000). Recently, however, the use of viral 2A peptides has been transferred into the zebrafish model system (Provost et al. 2007). By separating distinct open reading frames with short spacers that are subject to ribosomal skips during translation, equal amounts of multiple open reading frames can be targeted to the same cells (Szymczak et al. 2004). This approach is likely to supplant the use of IRES, and should be easily deployed in UAS transgenes.
The above labeling techniques have in common that they lead to Gal4 expression in a small number of cells, which are then visualized. They can also be used to mis- or over-express transgenes in these sparse cells to generate mosaics, which is important for studying cell-autonomous effects (Sato et al. 2006) or interactions between genetically distinct cells (Hua et al. 2005). However, because of the sparse expression of Gal4, they are not good approaches for genetically manipulating whole populations of neurons. Another set of techniques exists for sparse labeling within large populations of Gal4-expressing neurons.
One of these is the focused photoconversion of proteins such as Kaede or Dronpa. By irradiating individual Kaede-positive trigeminal sensory neurons, Sato et al. (2006) were able to convert them from green to red, and thus distinguish them from the rest of the unconverted neurons. Dronpa is a reversible green fluorophore that can be darkened by blue light and restored with violet light (Habuchi et al. 2005). Aramaki and Hatta (2006) expressed Dronpa in numerous hindbrain neurons, erased fluorescence throughout the pattern, and then serially brightened and darkened individual neurons, each time noting their morphology. This allowed them to describe these neurons both as individuals and as a group (Fig. 2c). The only drawback of this approach is the spatial limit of the targeted irradiation. Thus far, it has proven effective in neurons that are large and superficial (Sato et al. 2006), or that are already expressing fluorophore stochastically (Aramaki and Hatta 2006). It remains to be seen whether it will be possible to photoconvert individual neurons that are small, densely packed, or deep in the brain.
Another approach involves the highly variegated UAS:GFP in the Brn3c:Gal4, UAS:GFP (BGUG) transgene. For unknown reasons presumably relating to interference from the adjacent Brn3c:Gal4, expression of GFP is inefficient, appearing in only a small subset of Gal4-positive neurons. When combined with a Gal4 line, this allows for the sparse labeling of the targeted expression pattern with membrane-targeted GFP (Fig. 2d). BGUG has proven effective for labeling individual RGCs (Xiao and Baier 2007), tectal neurons (Scott et al. 2007), and assorted other types of neurons throughout the brain (Ethan K. Scott, unpublished experiments). An advantage of this system is that it works regardless of an expression pattern’s density or depth. However, as the mechanism of BGUGs variegation is unknown, it is difficult to prove that the labeling of neurons is genetically unbiased. If some cell types are more likely to be labeled than others, then a catalog of the neurons described in repeated BGUG observations would not accurately reflect their true abundance.
The anatomical techniques presented above each has its own strengths and shortcomings. Combined, however, they offer the possibility of fully describing Gal4 expression patterns in terms of their overall structure, connectivity, and cellular composition. For the rest of this review, I will focus on the tools with which neurons can be analyzed functionally, and how the Gal4/UAS system can best be used to deploy them in zebrafish.
Transgenic tools for observing activity in neurons
The functions of the neurons that form a circuit can be studied in either of two ways: by characterizing their patterns of activity under a range of conditions, or by identifying the behaviors that they help to generate. For decades, the gold standard for observing activity in neurons has been electrophysiology. Fluorescent techniques for monitoring neural activity have been developed more recently that serve as indicators of membrane potential, Ca2+ concentration, or neurotransmitter release. Although the cellular details and certain interpretations differ, each of these approaches can indicate when a neuron is active. The fact that their readout is visual makes them ideal for the zebrafish model system, because neurons or groups of neurons can be observed in intact animals as they behave or receive stimuli.
Many of the most trusted fluorescent indicators of activity are synthetic voltage- or Ca2+-sensitive dyes. While these have certain advantages in performance over genetically encoded indicators (Pologruto et al. 2004; Knopfel et al. 2006), they must be loaded broadly or by injection, and therefore cannot be targeted to specific neurons based on their genetic profiles. Protein indicators, on the other hand, can be driven transiently or in stable transgenic lines, and are compatible with the Gal4/UAS system. A recent proliferation in the genetically encoded indicators (reviewed by Miyawaki 2005), some of which have already been demonstrated in zebrafish, should allow for the flexible analysis of activity throughout the zebrafish nervous system.
Voltage reporting proteins allow the electrical properties of a neuron to be observed by varying their fluorescence based on the voltage across the membrane (reviewed by Baker et al. 2008). Thus, they can report on the occurrence of action potentials. To form these proteins, parts of naturally occurring voltage-gated cation channels are fused to one or more fluorophores. Depending on the construct, voltage-based conformational changes lead either to an increase in fluorescence from a single fluorophore [as in FlaSh (Guerrero et al. 2002; Siegel and Isacoff 1997) and SPARC (Ataka and Pieribone 2002)], or to increased Förster resonance energy transfer (FRET) between two fluorophores [e.g. in voltage-sensitive fluorescent proteins (Lundby et al. 2008; Sakai et al. 2001) and mermaid (Tsutsui et al. 2008)]. With the same goal, sensor proteins have been designed for Ca2+ concentration, which increases as voltage-gated Ca2+ channels open in response to an action potential. These involve fusing calmodulin (as in the cases of cameleon (Miyawaki et al. 1997), inverse pericam (Nagai et al. 2001), G-CaMPs (Nakai et al. 2001), and camgaroo (Baird et al. 1999) or more recently troponin (for TN-L15 Heim and Griesbeck 2004), TN-XL (Mank et al. 2006), and TN-XXL (Mank et al. 2008) to fluorophores. Conformational changes in the Ca2+-sensitive proteins lead to changes in the fluorescence or in FRET, depending on the construct.
Another useful readout of neural activity is the release of neurotransmitter, and pHluorin is a genetically encoded probe designed to detect it (reviewed by Yuste et al. 2000). Targeted to the acidic lumen of a secretory vesicle, this pH-sensitive GFP mutant is quenched, but it increases in fluorescence upon vesicle fusion and neutralization of the internal membrane of the vesicle (Miesenbock et al. 1998). Thus, it reports vesicle fusion as an indicator of neurotransmitter release.
The tools described so far are indicators of excitatory input to a neuron. The reciprocal, inhibitory input can be detected using the Cl−-sensitive protein Clomeleon (reviewed by Berglund et al. 2008). Clomeleon is a fusion of yellow fluorescent protein (YFP) and cyan fluorescent protein (CFP), where the CFP is a FRET donor to YFP. YFP is quenched by Cl−, and CFP is not, so at higher Cl− concentrations, the ratio of CFP/YFP fluorescence increases (Kuner and Augustine 2000; Markova et al. 2008). This has allowed, for example, the study of Cl− gradients in bipolar cells of the mouse retina (Duebel et al. 2006), and the visualization of inhibition in mouse hippocampal neurons (Berglund et al. 2006).
With all currently described activity indicators in zebrafish, some form of fluorescence microscopy is required, with specified excitation and emission of an activity-dependent fluorophore. Accordingly, the fish needs to be partially or completely immobilized as activity is observed. This, in turn, limits the types of behavior that can be linked to activity in targeted neurons. This limitation has recently been overcome through the use of Aequorin, a jellyfish enzyme that cleaves its cofactor, coelenterazine, selectively at high physiological Ca2+ concentrations (Shimomura et al. 1993; Shimomura 1995). This results in the generation of a blue photon, which drives excitation of GFP in a GFP-aequorin fusion, and emission of a green photon (Baubet et al. 2000; Hastings and Johnson 2003). The result is a probe that requires no optical excitation, and that luminesces only at high Ca2+ concentrations. This allows for activity of genetically targeted neurons to be monitored in a larva freely swimming in a dark box. Locomotor or other behaviors can be simultaneously observed in infrared wavelengths. This has already been used to correlate activity in a small number (< 20) of hypocretin-positive neurons to circadian behavior of free-swimming larvae (Florian Engert, personal communication). The great advantage of the technique, the freedom from magnified observation of the neurons in the swimming animal, brings with it a need for stringent genetic targeting. Since signals from different types of neurons cannot be separated spatially as they can on a microscope, the best results will come when GFP-aequorin is expressed in sets of neurons that are coordinated and tightly temporally linked with observed behaviors.
Most of these tools for visualizing neural activity have been used extensively since their creation, but have caught on slowly in the zebrafish model system. A few highlights in zebrafish include the identification of spinal neurons involved in larval escape response using cameleon (Higashijima et al. 2003) and the use of gCaMP to characterize the dynamics of cardiac conduction during early heart development (Chi et al. 2008). The further expansion of these tools in zebrafish, especially through the generation of UAS transgenic lines, will open the door to varied and flexible analyses of activity in any cell types currently targetable by Gal4 lines.
Patterns of activity within a brain region or cell type provide important information about the probable roles of neurons in sensing the world or generating an animal’s responses. A more conclusive demonstration of function comes from manipulations of targeted neurons that result in corresponding changes in behaviour. Silencing and activation of circuits can be used to demonstrate their necessity or sufficiency for a particular behavior, respectively. Furthermore, genetically encoded constructs allow for the targeted manipulation of distinct sets of neurons in structurally and genetically complex brain regions. The toolbox for genetic manipulations of neurons has recently been reviewed comprehensively (Luo et al. 2008), so here I will focus on relevant past work and the newly developed techniques that are likely to be most effective in zebrafish.
Genetic targeting assures that the experimenter can drive an activating or silencing transgene in a particular cell type, but many experiments require further temporal or spatial control over the manipulation. Temporal control is important because chronically silenced neurons may carry out homeostatic responses to resume activity even in the presence of the silencing mechanism. In cases where the targeted neurons remain silenced, the broader circuit may change connectivity or activity in a way that permits a detour around the silenced neurons. Either scenario would lead to a false negative result with regard to the targeted neurons’ normal necessity for the tested behavior. Spatial control is important when there is a broad expression pattern, and an experiment needs to address the role of the genetically targeted neurons in a particular part of the brain. Fortunately, many of the approaches for manipulating activity offer temporal control, spatial restriction, or both.
A few groups have already begun silencing zebrafish neurons with genetically targeted proteins. As part of their Gal4 GT screen, Asakawa and Kawakami (2008) used UAS:TeTxLC in tandem with Gal4 expression patterns to prevent synaptic transmission in Gal4-positive neurons. This allowed them to describe sets of spinal cord neurons necessary for different aspects of the larval escape response. Also with the goal of preventing neurotransmitter release, Hua et al. (2005) expressed a dominant-negative vesicle-associated membrane protein (dnVAMP) in RGCs, and studied the effects of this silencing on the development of the RGC axon termini. Finally, membrane-tethered toxins have been shown to be effective in blocking post-synaptic receptors, and thus preventing synaptic transmission (Ibanez-Tallon et al. 2004). As various toxins can be tethered this way, they allow for the interruption of specific types of receptors, thus providing for studies of specific neurotransmitter systems. These approaches lack inherent conditionality, so will not be effective for some types of studies. The most promising possibility for using them conditionally would be to pair their UAS constructs with lines for a ligand-dependent form of Gal4 that has recently become available in zebrafish (Esengil et al. 2007). This would allow for temporal control over their expression, based on ligand application, even if the proteins themselves are not conditional.
Among the approaches for removing a neuron’s activity from a circuit, none is more decisive than simply killing it. One appeal of this approach is the assurance that the neuron has been completely removed, eliminating concern over leaky activity or cell-autonomous homeostatic adjustments. It also opens the door to studying cell non-autonomous rewiring of circuits and regeneration of lost circuits. The Escherichia coli enzyme nitroreductase (NTR, encoded by the nfsB gene) converts a harmless prodrug into a potent cytotoxin, leading to the death of cells expressing NTR when the prodrug is presented (Curado et al. 2007, 2008; Pisharath 2007). Importantly, Davison et al. (2007) have already used UAS:nfsB-mCherry to kill targeted cells in combination with their Gal4 GT lines. An advantage of this system is the ease with which large numbers of animals can be treated, and it is therefore a good approach for large-scale behavioral or drug screens. While NTR can be genetically targeted, ablation cannot be further restricted spatially, so it cannot be used for subtle manipulations in combination with broad Gal4 expression lines. As a complementary approach, the adaptation of light-mediated apoptosis techniques to zebrafish would be desirable, as these would permit regional control through the use of targeted application of light. One such approach could use the KillerRed protein, a fluorophore that has been engineered to release cytotoxic reactive oxygen species as it goes through its excitation/emission cycle (Bulina et al. 2006). Illumination with green light serves to induce apoptosis in KillerRed-expressing cells, and this characteristic should allow for spatially restricted killing of neurons in embryonic and larval zebrafish.
Another technique for conditionally silencing neurons is to express ligand-dependent ion channels. Through the application of the ligand, neurons in a particular expression pattern can be silenced by the opening of the channel and the resulting hyperpolarization. Several approaches along this line have been demonstrated in other model systems, and a few of these are potentially applicable to zebrafish. These include the modified C. elegans glutamate receptor GluCl, which is gated by the drug ivermectin (Slimko et al. 2002; Lerchner et al. 2007), and the Drosophila allatostatin receptor, which opens in response to the allatostatin peptide (Lechner et al. 2002; Gosgnach et al. 2006; Tan et al. 2006). While these techniques offer temporally controlled induction, they lack spatial control, and therefore, like other approaches noted here, will have limited utility in patterns that are broad or that have significant background in non-neural tissues.
The most effective conditional manipulation of zebrafish neurons thus far has used LiGluR, a light activated glutamate receptor. This receptor can be covalently linked to a synthetic tethered agonist, Maleimide/Azobenzene/Glutamate (Volgraf et al. 2006). The tether is shifted into a cis alignment by 380 nm light, bringing the agonist into contact with the channel’s binding site. This opens the channel, allowing for Na+ influx, and leading to activation of the affected neurons. 488 nm light reverses this alignment and leads to closure of the channel. Szobota used a UAS:LiGluR line in combination with a Gal4 ET line to drive LiGluR in various neurons, including the Rohon-Beard touch-sensitive cells of the dorsal spinal cord. Activation of the channel with 380 nm light prevented the normal escape response from a touch stimulus, and this response was returned to normal by irradiation with 488 nm light (Szobota et al. 2007).
Other light-activated channels that have a great deal of potential for zebrafish include Channelrhodopsin 2 (ChR2) from the green algae Chlamydomonas reinhardtii and halorhodopsin [Natronomas pharaonis Halorhodopsin (NpHR)], drawn from the bacterium N. pharaonis (reviewed by Evanko 2007; Zhang et al. 2007). Both can be targeted genetically, and can activate (ChR2) or silence (NpHR) neurons in response to short pulses of intense light (Boyden et al. 2005; Gradinaru et al. 2008; Han and Boyden 2007; Li et al. 2005; Nagel et al. 2005; Zhang et al. 2006). The activation or silencing provided by these transgenes can be restricted extremely tightly both temporally and spatially, and this has allowed for the linkage of discrete sets of neurons to specific behaviours in C. elegans (Nagel et al. 2005), Drosophila larvae (Schroll et al. 2006), and zebrafish (Douglass et al. 2008). The need to apply light to neurons in an intact animal presents a challenge to using these technologies for studying mammalian behavior. While this has been proven possible through the delivery of light from an embedded fiber-optic cable (Adamantidis et al. 2007), the activation or silencing of targeted mammalian neurons in behavioral studies will remain much more technically challenging and labor-intense than in invertebrates and zebrafish. While they offer extremely sensitive control over target neurons, ChR2 and NpHR have the disadvantage that they require intense light pulses which will be best delivered to immobilized larvae through a microscope (Douglass et al. 2008). Therefore, for the time being, the use of ChR2 and NpHR may be restricted to behaviors that do not require mobility.
Most reports of Gal4/UAS use in zebrafish have noted some degree of variegation in the expression of UAS-linked markers, as a likely consequence of methylation of the UAS repeats (reviewed by Halpern et al. 2008). The impact of variegation varies depending on the study being performed. Mild variegation has little effect on descriptions of expression patterns or analyses of activity, as a majority of the cells being studied are still labeled. However, in experiments where circuits are being silenced and behavior assayed, variegation could allow leaky function in the targeted circuit. This could result in the animal’s successfully performing a behavior that actually does rely on the targeted cells. As a consequence, interpretations of negative results from such experiments will have to be very conservative until the issue of Gal4/UAS variegation can be resolved.
Behavior in zebrafish
As outlined above, the Gal4/UAS system in zebrafish has great utility for the anatomical description of the nervous system and the experimental control of neurons once they have been described. The final step is to decipher how these interconnected neurons drive behavior. This will involve the pairing of activation or silencing with behavioral assays to see whether given circuits mediate those behaviors. Screening for such links could take either of two forms: the silencing/activation of a library of expression patterns while testing for a specific behavior of interest, or the use of a range of behavioral assays to identify the function of a particular circuit or brain region. In either case, it will be important to have simple, high-throughput, and easily interpreted behavioral assays with which to test the importance of a group of manipulated neurons.
In larval zebrafish, there are two categories of behavior that have been studied effectively and in detail: visual behaviors and the motor control of swimming and escape. As zebrafish larvae need to blend into different backgrounds, capture prey, and avoid predators at early stages, a functional visual system is established by 5 dpf (reviewed by Baier 2000; Fleisch and Neuhauss 2006; Li 2001; Neuhauss 2003). The simplest assay is for visual background adaptation, where larvae on dark surfaces disperse melanin granules so as to appear dark while they constrict these granules to blend into a light background. This involves endocrine signaling mediated by a circuit linking the retina, hypothalamus, and pituitary (reviewed by Kawauchi 2006). The response has been shown to depend on RGCs (Kay et al. 2001), but the cellular details in other brain regions are not fully elucidated. Other visual behaviors include the optokinetic and optomotor responses (OKR and OMR). In the OKR, immobilized larvae are exposed to a horizontally moving visual stimulus, and their eyes sweep to follow the stimulus, occasionally saccading back to a starting location for another sweep (reviewed by Huang and Neuhauss 2008; Rinner et al. 2005). For the OMR, larvae are placed on top of an upward-facing monitor that provides a stimulus of laterally moving lines (Orger et al. 2000). Larvae swim with these lines, a response that would keep them stable in a natural setting where the visual world is stationary and the water is flowing. While mutations and resulting anatomical phenotypes have been tied to OKR and OMR deficits (Huang et al. 2006; Neuhauss et al. 1999; Rick et al. 2000; Roeser & Baier 2003), the underlying circuitry remains poorly understood. Prey capture is a more demanding visual assay, as it requires larvae to track and pursue small prey items (generally paramecia). This requires both motion detection and high-acuity vision, and vision must be properly linked to motor outputs (Budick and O’Malley 2000; Borla et al. 2002; McElligott and O’Malley 2005). Roles for the optic tectum and certain pre-motor areas have been established for efficient prey capture (Gahtan et al. 2005; Smear et al. 2007), but the details and other necessary circuits have not been explored fully. Given the richness of assays available for visual behavior and basic questions that remain unanswered about the neural mechanisms underlying them, this seems like a front on which the Gal4/UAS system offers particular potential for progress.
The last decade has seen extraordinary advances in the ways that neuroscientists can observe, characterize, and manipulate neurons and the networks that they form. These new techniques have already begun to change our view of how the brain assembles and functions to produce behavior. Each approach has its advantages and disadvantages, just as every model system has strengths and weaknesses. It is useful, therefore, to consider which technologies and what model systems are best suited to addressing specific neurobiological questions. As a relatively simple vertebrate, zebrafish combine some of the advantages (and disadvantages) of both invertebrates and mammals.
Historically, transgenic techniques have been a weakness of the zebrafish model system, but this has changed with the introduction of efficient techniques for integrating transgenes. This has allowed the Gal4/UAS system to take hold. As this system allows for a set of Gal4-expressing neurons to be targeted with a range of UAS-linked transgenes, circuits can be studied from a range of different anatomical, functional, and behavioral perspectives. In future years, this is likely to permit analyses of circuits in zebrafish that are begun with detailed anatomical analyses, followed by characterizations of activity, and finished with conclusive behavioral roles for the neurons of interest.
Such flexibility requires tools for all of these manipulations. Having a large assortment of Gal4 expression patterns is important so that a wide range of different cell types and circuits can be targeted. While hundreds of Gal4 lines are available from recent ET and GT screens, many of these have broad expression in the brain and other tissues. Therefore, future expansion of the Gal4 collection should focus on creating lines with tightly restricted expression either through ET and GT with improved minimal promoters (reviewed by Asakawa and Kawakami 2008), or through the use of defined enhancers.
The flexibility of Gal4/UAS in zebrafish rests equally on having UAS-linked transgenes that can perform a range of functions in Gal4-positive cells. In contrast to the numerous Gal4 lines, roughly a dozen stable transgenic zebrafish lines have been reported for UAS-linked constructs. Most of these deliver fluorophores to Gal4-positive cells. As a result, while many different patterns can be targeted, relatively few observations and manipulations can be performed in them. Therefore, it will be important in the coming years to expand this part of the toolbox to include UAS-linked Ca2+ indicators, voltage indicators, activating transgenes, and silencing transgenes, along with UAS lines from unbiased over-expression screens (Maddison et al. 2009). The existing bank of Gal4 lines means that, as these UAS tools become available, they can be deployed to numerous and diverse different patterns within the nervous system. These additions will be important to the future utility of the Gal4/UAS system in zebrafish, and to the eventual dissection of its behavioral circuits.
I would like to thank Herwig Baier for comments on this manuscript and Kohei Hatta, Stephen Smith, Hitoshi Okamoto, and Florian Engert for sharing images or preliminary data. My work is supported by University of Queensland New Staff Startup and Early Career Researcher Awards, a NARSAD Young Investigator Award, and the Australian Zebrafish Phenomics Facility (NHMRC Enabling Grant #455871).