Time-lapse imaging of neural development: Zebrafish lead the way into the fourth dimension

Abstract

Time-lapse imaging is often the only way to appreciate fully the many dynamic cell movements critical to neural development. Zebrafish possess many advantages that make them the best vertebrate model organism for live imaging of dynamic development events. This review will discuss technical considerations of time-lapse imaging experiments in zebrafish, describe selected examples of imaging studies in zebrafish that revealed new features or principles of neural development, and consider the promise and challenges of future time-lapse studies of neural development in zebrafish embryos and adults.

During development cells move and change their shape as they integrate into functional tissues and organs; they divide in precise orientations, grow in a polarized fashion, gain or lose adhesive contacts, migrate, project appendages, form specialized connections with other cells, trim off pieces of themselves, die in an orderly manner, or alter their structure in response to physiological activity. Nowhere are these dynamic processes more complicated, nor is the need for their precision more crucial, than in the developing nervous system, where cells must connect with one another to form neural circuits. It is difficult to describe these behaviors or understand how they contribute to neural development solely by examining static developmental snapshots. Trying to reconstruct complicated cellular behaviors from still images of different animals would be like trying to decipher the rules of baseball from still photographs of several different games, or attempting to appreciate the grace of a dance from images of several different performances. Repeatedly imaging the same cells over time to create a time-lapse movie is a powerful approach for ordering events into a comprehensive, coherent sequence.

Zebrafish have emerged as the premier model for directly imaging dynamic cellular processes. Time-lapse imaging of cellular development during embryonic stages is virtually impossible in mammals. To study embryonic development in mice, a pregnant mother must be sacrificed and each embryo removed from the uterus and dissected, providing a single snapshot of a particular developmental stage. As a proxy for investigating dynamic developmental behaviors in mammals, cultured cells, tissue slices, and organ explants have been used for time-lapse imaging. However, since cellular behaviors are often influenced by the embryonic environment, these preparations do not always recapitulate in vivo developmental processes with complete accuracy. The zebrafish model provides an opportunity for imaging development in live, intact vertebrate animals. Zebrafish embryos are optically clear and fertilized externally. Because they develop in water, the entire process of development can be observed over many hours without molesting the embryo. These advantages, combined with the availability of many mutants and an always-improving molecular toolkit, make zebrafish embryos a powerful setting in which to study dynamic developmental events by time-lapse. Time-lapse imaging is a complement, rather than a substitute, for traditional analyses in fixed animals, which make it possible to collect larger sample sizes and to stain for endogenous messenger RNAs and proteins. Together time-lapse and fixed tissue experiments make possible a more complete view of neural development. This review will discuss technical considerations of time-lapse imaging experiments in zebrafish, describe selected examples of imaging studies in zebrafish that revealed new features or principles of neural development, and consider the promise and challenges of future time-lapse studies of neural development in zebrafish embryos and adults.

Making movies: tools and methods for time-lapse imaging in zebrafish

Making time-lapse movies of zebrafish development is conceptually simple: a labeled embryo is placed under a microscope and images are captured repeatedly at varying intervals. To position a cell for imaging, embryos are anaesthetized or paralyzed, and typically embedded in agarose. Zebrafish embryos are surprisingly resilient and can develop normally in agarose for hours. Laser scanning confocal microscopy is the most common method for imaging. However, mulitiphoton microscopy is better for imaging deep tissues, and if scanning speed is the major concern, spinning disk confocal microscopy, which can capture millisecond dynamic events, may be the best choice. Several technical parameters must be considered when designing each experiment, including how long the imaging session should be, what magnification will be used, what mounting method will be used (which depends on the developmental stage and whether an upright or inverted microscope is used), how to balance imaging speed (which affects image quality) with the number of time-lapse intervals, and how to quantify time-lapse data. Although this review does not focus on these technical aspects of microscopy, optimizing imaging parameters for each particular neuronal population and developmental stage is a major challenge that must often be accomplished through trial-and-error.

Labeling specific cells

The first step in designing zebrafish imaging experiments is choosing a method for labeling specific cells and, if desired, manipulating gene expression within those cells. The optimization of molecular tools, many of them borrowed from the Drosophila system, has made these tasks easier and provided a wealth of options for creating sophisticated transgenes tailored for specific experimental questions.

There are two general methods for visualizing cells in zebrafish with transgenic reporters: expressing transgenes transiently (sometimes called somatic transgenesis) or making stable transgenic lines (germline transgenics). Because extrachromosomal DNA is unstably inherited during cell division, simply injecting a DNA transgene into fertilized eggs results in its mosaic inheritance (Figure 1A). This makes it possible to perform analyses at a single-cell level, even when a transgene drives reporter expression broadly (Downes et al., 2002). If, on the other hand, labeling an entire population of cells is the goal, stable transgenic lines can be generated (Figure 1B). Making stable transgenic fish used to be a somewhat burdensome undertaking, but the development of transposon-based methods, which increase the frequency at which exogenous DNA integrates into chromosomes, has greatly facilitated the process (Kikuta and Kawakami, 2009). As an alternative to genetic labeling, cells can be filled with fluorescent dye by injection or electroporation (Figure 1C) (Cox and Fetcho, 1996; Hutson and Chien, 2002; Tawk et al., 2009). Although this is an older method, it is still in frequent use, since it eliminates the necessity for creating a transgene and makes it possible to target specific cells.

Figure 1.

Three approaches to cell labeling. (A) DNA transgenes injected into early stage embryos are mosaically inherited, resulting in stochastic expression of the transgene in a subset of cells. Embryos expressing the reporter in a single cell can thus be identified for imaging. (B) An entire population of cells can be imaged in embryos harboring a stable, integrated transgene. (C) Electroporation or pressure injection of fluorescent dyes makes it possible to image a selected cell.

There are several approaches to finding enhancers/promoters for creating a cell type-specific reporter transgene. If a gene expressed in the cells of interest is known, it can be relatively easy to clone regulatory elements from the region surrounding that gene (Figure 2A). Most commonly regions upstream of the gene are chosen, but cross-species comparative analysis can aid in the identification of conserved regulatory regions anywhere near the gene (e.g. Allende et al., 2006). This approach, however, does not guarantee a fully accurate reporter, since regulatory elements can be located quite far from the start of a gene. A more reliable method for obtaining faithful reporter expression is to recombine a fluorescent reporter gene into a bacterial artificial chromosome (BAC) containing the target gene (Figure 2B) (Yang et al., 2009). The larger sequence context of a BAC reporter improves the fidelity of reporter gene expression. Recently, transposon (Tol2)-mediated BAC transgenesis has been developed in zebrafish and mice to precisely deliver single-copy BAC transgenes (Suster et al., 2009). As an alternative to creating reporters based on specifically selected genes, many labs have recently performed enhancer- and gene-trap screens (Asakawa et al., 2008; Balciunas et al., 2004; Davison et al., 2007; Kawakami et al., 2004; Kikuta et al., 2007; Laplante et al., 2006; Lobo et al., 2006; Parinov et al., 2004; Scott et al., 2007). In these screens, a reporter gene is randomly inserted into the genome, where it may land near specific regulatory elements (Figure 2C). These screens have together created a large collection of reporter transgenes with novel expression patterns.

Figure 2.

Three methods for obtaining cell type-specific reporters. (A) A region upstream of a selected gene can be subcloned upstream of a fluorescent reporter, and often drives expression in a pattern approximating the endogenous gene. (B) To improve the chances of faithfully recapitulating an endogenous expression pattern, a fluorescent reporter gene can be recombined into a targeted location in a BAC. (C) Tol2 transposition can be used to randomly integrate a reporter throughout the genome. The reporter will occasionally land near regulatory elements that drive expression in a useful pattern.

Reporters and effectors

Many fluorescent proteins can be used as reporters. In addition to the common ones, such as GFP and RFP, a few photoconvertible fluorescent reporters have proven valuable for tracking cell fate and analyzing the morphology of single cells (Aramaki and Hatta, 2006; Arrenberg et al., 2009; Caron et al., 2008; Davison et al., 2007; Hatta et al., 2006; Marriott et al., 2008; McLean and Fetcho, 2009; Sato et al., 2006; Scott et al., 2007). Kaede (Ando et al., 2002), KikGR (Tsutsui et al., 2005) and Dendra (Gurskaya et al., 2006) are all photoconvertible fluorescent proteins with emission spectra that can be shifted from green to red upon exposure to UV light. Dronpa is also a GFP-like protein, but 405- and 488-nm light can reversibly switch the fluorescent “on” -state and nonfluorescent “off” -state of Dronpa (Ando et al., 2004). Such reversible photoswitching could be repeated more than 100 times with a millisecond response time, making it possible to monitor dynamic subcellular events (Habuchi et al., 2005).

The goal of many experiments is not only to image a cell, but also to simultaneously express an effector gene that alters its function. Several strategies have been developed for tandem co-expression of two genes. For examples, a fluorescent reporter and effector gene can both be placed under control of the Gal4 upstream activation sequence (UAS) and activated with the same Gal4 driver (Koster and Fraser, 2001). Alternatively, two separate proteins can be translated from one messenger RNA with an internal ribosome entry site (IRES) (Fahrenkrug et al., 1999), or, to ensure equimolar co-expression, a single fusion protein can be split into two autonomous proteins with viral 2A peptides, which, though often called “self-cleaving”, create two peptides by a “ribosome-skipping” mechanism (Provost et al., 2007). These techniques can be used to overexpress wildtype, dominant negative, or constitutively active proteins, along with a fluorescent reporter, making it possible to compare the behavior of molecularly manipulated cells to wildtype cells.

Gal4/UAS transgenes

The yeast Gal4/UAS transcriptional activation system has been an extraordinarily successful tool for manipulating gene expression in Drosophila. With modifications, this system has also been widely incorporated into zebrafish transgenes (reviewed in Halpern et al., 2008). The beauty of this binary approach is that Gal4 drivers and UAS reporters can be produced separately and put together in different combinations, providing powerful versatility. Fusion of Gal4 to the VP16 transcriptional activation domain provides substantial amplification, allowing abundant expression even from weak enhancers/promoters (Koster and Fraser, 2001). A fusion of Gal4 to the ligand-binding domain of the ecdysone receptor allows hormone-mediated temporal control of gene expression (Esengil et al., 2007). Silencing and variegation has been a significant problem with the Gal4/UAS system in zebrafish, but the creation of non-repetitive UAS elements alleviates this problem (Akitake et al., 2011). In addition to using known tissue-specific regulatory elements to direct Gal4 expression, several labs have also used Gal4 for enhancer trap screens, creating an enormously versatile set of tools for imaging cells and misexpressing genes in tissues throughout the animal (Asakawa and Kawakami, 2009; Asakawa et al., 2008; Davison et al., 2007; Scott et al., 2007).

What have we learned from watching movies?

Time-lapse imaging provides at least three kinds of insights into mechanisms of neural development. First, it can reveal a link between a cell’s behavior at one moment and a previous experience, suggesting a mechanistic connection between two discrete events. For example, the ultimate fate of newborn cells can sometimes be predicted by the orientation of a cell division, a relationship that can only be definitively deduced by watching division occur and tracking the daughter cells over time. Second, time-lapse imaging is the most direct way to identify a migratory pathway, and may be the only way of deducing these routes if migratory cells are intermingled with a heterogeneous group of cells exhibiting different behaviors. Third, time-lapse imaging can reveal transient cellular behaviors that are characteristic features of developmental processes. For example, developing axons often project branches that are later pruned away; if degeneration is rapid or stochastic it may be impossible to know that pruning occurred without time-lapse imaging. Identifying linked sequences of events, migratory pathways and transient phenomena is fundamental to understanding how the nervous system is constructed.

Watching cells divide and differentiate

During central nervous system development, highly organized multi-layered structures, such as the cortex and retina, emerge from single-layered neuroepithelia. Recent time-lapse imaging studies of the developing zebrafish retina and hindbrain have challenged several aspects of widely accepted models describing how these laminar brain regions develop (Figure 3). These studies exemplify the powerful ability of live imaging to order discrete events into a progressive sequence.

Figure 3.

Studies in the retina and hindbrain have revised models for the formation of laminar structures from simple neuroepithelia. In all images, “A” indicates the apical surface; for simplicity, basal processes are not shown in dividing cells. (A) Interkinetic nuclear migration was thought to be a smooth movement of nuclei in concert with the cell cycle, but a recent study in the zebrafish retina showed that nuclear movements are stochastic (Norden et al., 2009). (B) It was believed that during asymmetric cell division in simple neuroepithelia, apical daughters become progenitors that continue to divide and basal daughters differentiate into neurons, but a recent study found the opposite to be true in the developing zebrafish hindbrain (Alexandre et al., 2010). (C) A recent study in the zebrafish retina found that daughters of cells that divided at the apical surface often migrated to the inner nuclear layer to divide again and generate horizontal cells (Godinho et al., 2007).

Before the onset of neurogenesis in the central nervous system, elongated progenitor cells span the width of the neuroepithelium. Their nuclei are positioned all along the apical-basal axis, giving the neuroepithelium a pseudostratified appearance. According to a prevalent model, the nuclei of progenitor cells were thought to migrate from the apical to the basal surface during G1 and S-phase, return towards the apical pole during G2, and divide at the apical surface, a sequence termed interkinetic nuclear migration (IKNM) (Miyata, 2008). Though the model relating nuclear position to cell cycle stage was widely accepted, IKNM was never visualized directly in living animals. To image IKNM, Norden and colleagues tracked progenitor nuclei in the developing zebrafish retina labeled with the nuclear marker H2B:RFP, using confocal microscopy (Norden et al., 2009). Rather than observing smooth, directed nuclear movements in concert with the cell cycle, the authors found that nuclear movements were mostly stochastic, only becoming rapidly directed towards the apical surface just before mitosis (Figure 3A). Thus, IKNM movements, at least in the zebrafish retina, are not as predictable or organized as was once thought.

Once at the apical surface, cells can divide parallel or obliquely to the apical surface. Symmetric divisions generating two proliferative progenitor cells occur when the cleavage plane is parallel to the apical-basal axis (in other words, the long axis of dividing cells is parallel to the apical surface), whereas asymmetric divisions generating one progenitor and one neuronal daughter cell occur when the cleavage plane is perpendicular to the apical-basal axis (Chenn and McConnell, 1995; Miyata, 2008). The prevalent model relating cell division with asymmetric cell fates proposed that inheritance by the apical daughter cell of an apical protein complex caused it to retain the progenitor fate, whereas the basal cell that did not inherit the complex adopted a neural fate (Huttner and Brand, 1997). A recent study tested this model directly by imaging asymmetrically dividing cells and their progeny in the developing posterior hindbrain (Alexandre et al., 2010). Expressing an apical complex marker (Par3-GFP), a marker for the basolateral compartment (Numb-GFP), and a reporter for neuronal cell fate (HuC-GFP), in neuroepethelial cells allowed the authors to monitor cell division, inheritance of the apical complex, and the fate of daughter cells. Consistent with previous models, progenitors that cleaved parallel to the apical-basal axis mostly generated two proliferative progenitor daughter cells, whereas cleavages more than 15° off this axis generated a progenitor and a neuron. As expected, the “apical footprint” (Par3-GFP) was inherited by both proliferative daughters during symmetric divisions, but was asymmetrically inherited by one daughter during asymmetric division. The surprising finding was that during asymmetric division the cell fated to become a neuron, as judged by the fact that it turned on the HuC:GFP reporter, was apically positioned and inherited the apical Par3 complex, while the basal daughter cell became the progenitor and re-established Par3-GFP expression after several minutes (Figure 3B). These compelling results challenge the prevalent model, which posits that the apical cell becomes the progenitor, and provide a striking example of time-lapse analysis clarifying an issue that was difficult to resolve in analyses of static images.

Following cell division, developing neurons migrate from the apical surface to the appropriate laminar position. In the retina, the frequent appearance of mitotic cells in the inner nuclear layer raised the interesting possibility that neurons could also be born away from the apical surface, either close to or within their respective cell layers (Smirnov and Puchkov, 2004). Godinho and colleagues addressed this idea with time-lapse multiphoton imaging of fluorescently labeled horizontal cells from early stages, when only a few horizontal cell precursors were present, to stages when the horizontal cell layer was completely formed (Godinho et al., 2007). Strikingly, rather than directly generating horizontal cells, many asymmetric cell divisions at the apical surface yielded unipotent proliferative cells that migrated to the inner nuclear layer and divided again to generate horizontal cells (Figure 3C). In fact, the majority of horizontal cells (~85%) were formed by symmetric division of progenitor cells in the layer where mature horizontal cells reside, rather than by migration of terminally fated daughter cells from an apical division. Together, these studies of IKNM, cell division, and differentiation in the retina and hindbrain, using cell type-specific transgenic reporters, revised long-standing models of laminar development (Figure 3).

Watching neurons migrate

Many neurons migrate long distances from their birth site to their final residence. Live imaging reveals migratory pathways, since a neuron’s trajectory from its birthplace to its destination can be traced directly. A study of cell migration in the developing zebrafish cerebellum illustrated how the migratory paths of individual cells can be distinguished, even within a heterogeneous population of cells, by time-lapse imaging (Volkmann et al., 2008). In the cerebellum, granule cells migrate from the upper rhombic lip, a proliferative neuroepithelium, deep into the cerebellar cortex and ventral hindbrain, where they form neuronal clusters that connect with various parts of the brain to mediate motor control and other functions (Wingate, 2001). Although diverse cells emerge from the rhombic lip, the neuroepithelium displays no anatomically obvious regional divisions. Imaging granule cell precursor migration and tracking cells trajectories with software-assisted tracing revealed that cells emerge from distinct spatial subdomains along the mediolateral axis of the rhombic lip (Volkmann et al., 2008); cells born in a lateral subdomain of the rhombic lip migrated along a different route than cells born in a dorsomedial subdomain and formed distinct anatomical structures. Thus, time-lapse imaging made it possible to distinguish the origin and migratory paths of different subpopulations of cells without requiring distinct molecular markers for each cell type.

Comparing migratory pathways in mutant and wildtype cells can illuminate the molecular mechanisms controlling cell motility. For example, another study of cerebellar cell migration revealed an unexpected function for Cadherin 2 (Cdh2), a major constituent of adherens junctions in neuronal migration (Rieger et al., 2009). Previous work in chick and zebrafish had shown that loss of Cdh2 function resulted in mispositioned neurons and neural crest cells close to their place of origin in the brain. This observation suggested that the motility of migratory cells was compromised in the absence of Cdh2 (Kasemeier-Kulesa et al., 2006; Lele et al., 2002). To re-examine this idea, Rieger and colleagues traced individual granule cells over time in Cdh2 mutants. Unexpectedly, cell velocities were similar in mutant and wildtype animals, but in Cdh2 mutants the direction of migration was randomized, resulting in circling behaviors of granule cells (Figure 4). The lack of directionality of migrating mutant granule cells was confirmed with the expression of a red-fluorescent centrosome marker. In migrating wildtype cells, the centrosome was located predominantly anteriorly, in the direction of migration, but its location within the cell was random in Cdh2-deficient granule cells. Moreover, the position of an adherens junctions marker (a fusion of the Cdh2 protein to mCherry) within migrating neurons correlated with the distinct migratory steps of granule cells and was predominantly located at the anterior cell cortex during forward movement, close to the centrosome. Thus, rather than controlling the rate of migration, adherens junctions control the direction of cell migration, an insight that would have been difficult to glean from still images of migration.

Figure 4.

Granule cell precursors in wildtype and cadherin 2 (cdh2) mutants start out in the same area of the upper rhombic lip (start), but although wildtype cells migrate a substantial distance, mutant cells stay close to their starting point (end). Time-lapse imaging revealed that cdh2 mutant cells are defective in the directionality of migration, rather than the speed of movement (in between) (Rieger et al., 2009).

Watching growth cones navigate

Once a neuron has chosen its fate and its cell body has reached the appropriate location, it projects neurites to establish synaptic connections. Watching dendrites and axons grow in real time can reveal aspects of their navigational strategies that are not possible to deduce simply by determining whether or not they reached their destination. For example, transient growth cone behaviors, including changes in their speed, shape or size are most clearly detected by time-lapse imaging and often indicate axon guidance choice points. Comparing these behaviors in wildtype and mutant growth cones can reveal how guidance molecules regulate these choices (e.g. Hutson and Chien, 2002; Jing et al., 2009; Sato-Maeda et al., 2006). Time-lapse imaging can also reveal whether guidance decisions result from attraction to a particular path or the avoidance of alternative paths. For example, the peripheral axons of somatosensory neurons are complex branched structures that are segregated from those of their neighbors in a “tiled” arrangement. In principle, this segregated arbor arrangement could result from competition between adjacent axons for a limiting positive factor or from the active avoidance of neighboring cells. Time-lapse imaging of the developing peripheral arbors of trigeminal neurons, which innervate the head, and Rohon-Beard neurons, which innervate the trunk and tail, clearly demonstrated that when growth cones contacted a neighboring axon they collapsed and turned away (Liu and Halloran, 2005; Sagasti et al., 2005), providing a mechanistic explanation for the formation of a tiled pattern.

Perhaps the most dramatic transient events are the degeneration or rerouting of selective neurite branches. These events leave no trace, making them impossible to detect by examining mature arbor structures. For example, time-lapse imaging of trigeminal sensory neuron arborization in zebrafish revealed that sporadic pruning of axonal arbors and spontaneous neuronal death are common during the course of peripheral sensory axon development (Sagasti et al., 2005). These sporadic pruning events often occurred when axon branches crossed over one another, suggesting that pruning may be a backup mechanism for ensuring a tiled arrangement when repulsion between neighboring arbors fails. Quantitative analyses of these local degeneration events demonstrated that axon branch detachment, fragmentation, and clearance were all complete in less than three hours, which makes this process difficult to detect in still images (Martin et al., 2010). Imaging the regeneration of trigeminal peripheral axons that were precisely severed with intense laser power on a multiphoton microscope revealed that related behaviors regulate the ability of sensory axons to successfully reinnervate their former territory (O’Brien et al., 2009a). Following axotomy, injured axons could reinitiate growth but were often repelled by their former territories, in much the same manner that neighboring axon branches repel one another during tiling. Strikingly, time-lapse movies also showed that when regenerating axons did reenter their former territories they invariably retracted back out of the region, degenerated locally, or the parent neuron died (Figure 5A). Together these observations suggested that local inhibitory factors persist in denervated skin after a severed axon branch has degenerated.

Figure 5.

Time-lapse imaging can reveal transient events in growth cone development and regeneration. (A) Following axotomy of the peripheral axon arbor of a trigeminal neuron (start), regenerating axons never reinnervate former territory (blue patch), despite the fact that severed portions of the axons are quickly cleared (end). Examining regeneration by time-lapse imaging revealed that sometimes axons do grow into former territory, but always degenerate or retract out of it (in between) (O’Brien et al., 2009a). (B) Wildtype retinal ganglion cell axons projecting into the brain (dotted line represents the midline) almost never misroute (end). However, examining intermediate stages of migration by time-lapse revealed that some axons (green axon) leave the common path (orange axons), but these errors are almost always corrected (in between) (Hutson and Chien, 2002).

As with studies of migration, comparing transient events in axon development between mutant and wildtype neurons can uncover unexpected molecular mechanisms regulating axon guidance. A study of a mutant in a receptor of the Robo family illustrates this point. Robo proteins are receptors for secreted guidance signals of the Slit family and their activation is thought primarily to induce repulsive growth cone behaviors. Hutson and Chien investigated the phenotype of the astray mutant, which has a mutation in the robo2 gene, using time-lapse imaging of dye-filled retinal ganglion cells to study its role in the establishment of retinotectal axon pathways (Hutson and Chien, 2002). These time-lapse analyses revealed that in wildtype zebrafish, retinal ganglion cell (RGC) axons growing across the optic chiasm occasionally made pathfinding errors, which were always eventually corrected (Figure 5B). In the astray/robo2 mutant, RGC axons made more errors, and those errors were rarely corrected. Thus, in addition to its guidance role, robo2 is unexpectedly also required for correcting erroneous projections—both functions that may depend on sensing repulsive gradients. These misprojections can be detected in fixed specimens, but their correction in wild-type animals would have been difficult to detect without time-lapse imaging, and defects in the astray mutant may have been interpreted solely as guidance defects.

Into the future: more to watch

The use of zebrafish time-lapse imaging studies for identifying links between developmental events, tracing migratory routes, and revealing transient cellular behaviors, perhaps constitutes the model’s most valuable contribution to the field of neural development. These studies are continually becoming more sophisticated as microscopy techniques and molecular tools improve. Although the studies we have described here demonstrate that zebrafish imaging experiments are already making significant contributions, several aspects of neural development have only just started being explored with this powerful approach. These new frontiers include studies of subcellular events, the relationship between physiological states and development, and the regulation of plasticity in the mature nervous system.

Watching events within cells

Some of the studies mentioned in this review employed genetically encoded fluorescent reporter fusions to image subcellular structures (Alexandre et al., 2010; Norden et al., 2009; Rieger et al., 2009). As more of these fusion proteins are created, it becomes possible to track an increasing variety of organelles, subcellular compartments, and protein complexes. Imaging techniques that minimize exposure of the sample, such as multiphoton or spinning disk microscopy, can help reduce bleaching of these sometimes-vulnerable fusion proteins. One caution with these studies is that overexpression of fusion genes could drive localization to ectopic locations within a cell or interfere with normal development, but appropriate controls such as co-staining with an antibody for an endogenous protein, or using the fusion protein to rescue a mutant phenotype, can alleviate these concerns. Live imaging studies of the development of synapses, perhaps the most interesting subcellular specializations in neurons, using both pre- and post-synaptic reporter fusions, have already yielded insights into the assembly of synapses and the establishment of connections (Jontes et al., 2004; Meyer and Smith, 2006; Niell et al., 2004). Synapse formation is likely to become a growing focus of zebrafish research, and these live imaging studies are poised to make substantial contributions in new areas, such as how organelles and vesicles are trafficked in axons and dendrites, or how cytoskeletal dynamics are regulated during cell and growth cone migration.

Linking physiology and development

Many developmental events are influenced by a neuron’s physiological state. For example, neuronal activity can influence processes as diverse as neurite arborization and the initiation of cell death (Ben Fredj et al., 2010; Hua et al., 2005; Paulus et al., 2009; Svoboda et al., 2001). Calcium sensors are faithful indicators of calcium transients and, indirectly, of action potentials. For several years now, imaging calcium in live zebrafish larvae with calcium-sensitive dyes or genetic indicators, such as GCaMP and cameleon, has helped decipher neural circuit organization (reviewed in McLean and Fetcho, 2008). Combining these indicators with long-term imaging of developmental processes could help detect correlations between activity and specific cellular behaviors. The causality of these relationships could then be tested with drugs that alter cellular activity, or more precisely with genetically encoded proteins that either stimulate or block cellular activity, such as the light-activated channels channelrhodopsin, halorhodopsin, and LiGluR (reviewed in Baier and Scott, 2009). The development of indicators for other physiological phenomena, such as synaptic transmission (Dreosti et al., 2009), the presence of reactive oxygen species (Niethammer et al., 2009), pH (Miesenbock et al., 1998), voltage (Baker et al., 2008), or the activity of specific channels (Richler et al., 2008), will make it possible for zebrafish researchers to ask a broader array of questions about the influence of physiological states on development.

Imaging cellular plasticity in mature animals

Dynamic cellular behaviors do not cease when neurons attain their mature structures. Cells grow and change as animals age, they undergo regulated responses to cellular damage, and they establish new neural connections as animals learn. A dramatic example of extreme cellular plasticity was provided by a recent zebrafish time-lapse study that documented the transdifferentiation of cells from dorsal root ganglia neurons into sympathetic neurons (Wright et al., 2010). It is likely that many other surprising transformations occur in the nervous system as animals grow and transition from embryonic to larval, and larval to adult forms.

At any age, structural plasticity is particularly dramatic after cellular damage. Molecules that regulate the ability of axons to regenerate are conserved between fish and mammals (Bhatt et al., 2004; O’Brien et al., 2009a), giving these studies particular medical relevance. The ability to create precise damage to cells, dendrites or axons with lasers makes it possible to image regeneration in live animals, but this method is only just beginning to be exploited for zebrafish studies (Martin et al., 2010; O’Brien et al., 2009b).

Characterization of the adult zebrafish brain is just beginning, with studies of adult neurogenesis leading the way (e.g. Adolf et al., 2006; Chapouton et al., 2006; Grandel et al., 2006). One of the next challenges is to create fluorescent reporters expressed in adult neurons, of which there are currently few. This could be accomplished by screening enhancer trap lines for adult expression or creating BAC transgenic reporters of genes expressed in the adult. Even enhancers that drive early transient expression can be used to permanently mark neurons, either in combination with the Cre-lox recombination system or with a self-maintaining Gal4-based positive feedback system (dubbed “Kaloop”) (Collins et al., 2010; Distel et al., 2009; Thummel et al., 2005). The development of the tol2 transposon system in fish, which greatly increases the integration of transgenes into the genome of somatic cells, has even made it possible to perform imaging experiments in transient transgenic adults (e.g. Tu and Johnson, 2010). It should be possible to do studies in zebrafish analogous to recent tour-de-force multiphoton imaging studies of the adult mouse cortex (Pan and Gan, 2008). Combining adult imaging studies with new reporters for subcellular structures and physiological indicators would provide a powerful opportunity to gain deeper insights into the structural plasticity of adult brains.