Convergence of signaling pathways underlying habenular formation and axonal outgrowth in zebrafish

The habenulo-interpeduncular (Hb-IPN) conduction system is a conserved feature of the vertebrate brain that modulates a variety of processes such as sleep, fear/anxiety, pain, learning, feeding, reproduction and reward (Sutherland, 1982; Benarroch, 2015; Namboodiri et al., 2016), and has been associated with mood disorders (Ranft et al., 2010; Savitz et al., 2011; Lawson et al., 2016) and addiction (Fowler et al., 2011; Baldwin et al., 2011). Recent attention has focused on the role of the Hb-IPN circuitry in nicotine dependence and withdrawal symptoms (Antolin-Fontes et al., 2015; Salas et al., 2009; Zhao-Shea et al., 2013; Pang et al., 2016) and the finding that, in zebrafish and other non-mammalian species, the bilaterally paired dorsal habenulae [dHb; equivalent to the medial habenulae (mHb) of mammals] (Aizawa et al., 2011) show prominent left-right differences in size, organization, molecular properties and connectivity (Braitenberg and Kemali, 1970; Concha and Wilson, 2001; Güntürkün and Ocklenburg, 2017; Duboué and Halpern, 2017). Although accumulating evidence demonstrates the importance and diverse functions of the habenular nuclei, our understanding of their development is incomplete.

The Hedgehog (Hh), Wnt and fibroblast growth factor (Fgf) signaling pathways have all been implicated in early habenular development. However, there are conflicting reports on whether Hh signaling antagonizes or promotes formation of the dHb (Chatterjee et al., 2014; Halluin et al., 2016). Mutations in the zebrafish wntless (wls) gene, which encodes a protein essential for Wnt secretion, or in the fibroblast growth factor 8a (fgf8a) gene, result in small dHb (Regan et al., 2009; Kuan et al., 2015). The small size of the dHb in wls and fgf8a mutant zebrafish is likely to be caused by a reduction in dHb progenitors (Kuan et al., 2015; Dean et al., 2014), which are characterized by their expression of the developing brain homeobox 1b (dbx1b) gene (Dean et al., 2014). The Dbx1 gene was first identified as a marker of habenular progenitors in mice (Vue et al., 2007). In zebrafish, there are two homologs, dbx1a and dbx1b, but only dbx1b is expressed in dHb progenitors (Dean et al., 2014). Transcripts for the chemokine (C-X-C) motif, receptor 4b (cxcr4b) gene are also presumed to localize to habenular progenitors or newly born neurons (Roussigné et al., 2009; Halluin et al., 2016). However, the relationship between cells that express dbx1b and those that produce components of the Cxcr4-chemokine pathway is unresolved.

Through long-term lineage experiments, we demonstrate that dbx1b⁺ progenitors give rise to all neurons in the dHb of adult zebrafish. During larval development, these progenitors undergo a transition during which they express both dbx1b and cxcr4b, then downregulate dbx1b and differentiate into cxcr4b-expressing neural precursors and, ultimately, dHb neurons. Hh, Wnt and Fgf signaling pathways are required for formation of dbx1b⁺ progenitors. Analyses of zebrafish mutants indicate that Hh signaling is upstream or independent of Wnt signaling, and that Wnt activity restricts the domain of Fgf signaling. Wnt and Fgf signals are also necessary to delimit the expression domains of genes encoding Cxcr4-chemokine pathway components. We discovered that disruption of chemokine signaling perturbs the posteriorly directed outgrowth of dHb efferents, resulting in ectopic axonal projections. Thus, the action of multiple signaling pathways, through a complex regulatory network, promotes both the generation and connectivity of habenular neurons.

Previous studies suggest that the dbx1b and cxcr4b genes are expressed by dHb progenitors (Dean et al., 2014; Roussigné et al., 2009; Halluin et al., 2016). A detailed analysis indicates that expression of the two genes is only partially overlapping at 35 h postfertilization (hpf) and 3 days postfertilization (dpf) (Fig. S1A-D). Thus, the precise relationship between the dbx1b and cxcr4b cell populations is unclear.

The identity of dbx1b⁺ cells as progenitors of the dHb is well established on the basis of evidence from experiments in zebrafish (Dean et al., 2014) and the mouse homolog, Dbx1 (Vue et al., 2007). In larval zebrafish, dbx1b⁺ progenitors are located in a ventromedial location and give rise to neurons throughout the dHb (Dean et al., 2014). We confirmed these observations (Fig. S2A,B), and extended lineage tracing experiments to adults (Fig. S2C). Descendants of dbx1b⁺ progenitors, distinguished by their mCherry-labeled nuclei, are found throughout the dHb of larvae and adults bearing Tg(gng8:CAAX-GFP), which labels the membranes of dHb neurons and their efferent axons with GFP (deCarvalho et al., 2013).

Owing to the known role of Cxcr4b in cell migration (Lewellis and Knaut, 2012), we performed time-lapse imaging to determine whether cxcr4b⁺ cells migrate to the developing dHb or whether cxcr4b expression is initiated de novo in dbx1b⁺ progenitors. Images of TgBAC(dbx1b:GFP) and TgBAC(cxcr4b:nls-tdTomato) double-labeled embryos were captured every 15 min between 27 and 48 hpf (Fig. 1, Movie 1). At 27 hpf, GFP-labeled dHb progenitors are detected in wild-type (WT) embryos and, by 32 hpf, dTomato-positive nuclei appear within these GFP-labeled cells, demonstrating that cxcr4b expression initiates within pre-existing dbx1b⁺ progenitors. We found no evidence for migration of dTomato-labeled cells into the presumptive habenular region.

In wls mutants, which produce fewer dbx1b⁺ cells and have smaller dHb owing to reduced Wnt signaling (Kuan et al., 2015), there is a delay in the appearance of GFP-labeled and dTomato-labeled cells (35 and 40 hpf, respectively) (Fig. 1), and there are far fewer cells, consistent with a reduction in dHb progenitors. However, despite these defects, dTomato labeling is also only detected within the dbx1b⁺ cells in wls mutants.

We next examined how dbx1b and cxcr4b expression changes during the course of habenular development (i.e. from 2 to 30 dpf) (Fig. 2). By 2 dpf, dTomato labeling is found in nuclei within the anterodorsal region of the dHb. By 3 dpf, GFP labeling becomes confined to a medial and ventral region of the developing dHb. The population of cells with dTomato-labeled nuclei increases in the dorsal and lateral regions, where GFP labeling is absent. However, a small number of colabeled cells are found at the border between the GFP and dTomato uniquely labeled domains. Restriction of GFP to an increasingly smaller ventromedial region, and the dorsolateral expansion of the dTomato-labeled cell population, continues over the course of several weeks.

Owing to persistence of dTomato labeling (Fig. S3A,B), we also examined the distribution of endogenous cxcr4b transcripts by fluorescent RNA in situ hybridization. Double labeling for ELAV like neuron-specific RNA binding protein 3 (elavl3; formerly known as HuC) (Fig. 3A) or anoctamin 2 (ano2) (Fig. 3B), which are expressed in mature dHb neurons (Kim et al., 1996; deCarvalho et al., 2014), indicates that cxcr4b transcription is rapidly terminated as dHb neurons differentiate. The results suggest that dbx1b⁺ progenitors undergo a transition to cxcr4b⁺ neural precursors, which then give rise to differentiated neurons.

Previously, Sonic hedgehog (Shh) signaling from the zona limitans interthalamica (ZLI) was found to be essential for formation of the dHb, as mutants for the Shh receptor Smoothened (Smo) lack cxcr4b expression in this region of the brain (Halluin et al., 2016) (Fig. 4A,A′). We find that expression of dbx1b is also absent in smo homozygous mutants (Fig. 4B-C′), indicative of the loss of dHb progenitors. Previously, mutations in the wls gene, which encodes a protein essential for secretion of Wnt proteins, were also found to reduce the dHb progenitor population (Kuan et al., 2015). To examine the relationship between Shh and Wnt signaling in dHb development, we assessed expression of sonic hedgehog a (shha) and the Hh-responsive gene patched 1 (ptch1) in the ZLI of wls mutants. Expression of both genes appeared indistinguishable from that of WT siblings (Fig. 4D-G′), suggesting that Shh acts upstream or independently of Wnt signaling in the specification of dHb progenitors.

Reduction of Fgf signaling using chemical inhibitors, or in fgf8a homozygous mutants, also results in fewer dbx1b-expressing progenitors (Dean et al., 2014) and small habenulae (Regan et al., 2009). Therefore, the reduced progenitor population observed in wls mutants could be caused by a lack of Fgf signaling. In WT embryos, expression of fgf8a is limited to cells just anterior of the developing dHb (Fig. 5A-C), within a region 28.9±1.8 µM long and 95.0 µM±5.3 µM wide (n=7). However, rather than being lost or smaller, the fgf8a domain is expanded posteriorly (2.2-fold) and medially (1.4-fold) in wls mutants (n=7) (Fig. 5A′-C′, Table 1, Fig. S4A). A similar enlargement of the fgf8a domain is seen after heat shock-induced overexpression of the Wnt signaling inhibitor Dickkopf 1 (Dkk1) in 24 hpf Tg(hsp70:dkk1-GFP) embryos (Fig. 5D,D′). The result of expanded fgf8a expression in wls mutants is increased Fgf signaling, as evidenced by a corresponding expansion in the expression domains of Fgf-responsive genes, such as dusp6 and etv5b (Fig. 5, Table 1, Fig. S4B,C). These findings indicate that, surprisingly, the reduction of dHb progenitors in wls mutants is not the result of decreased Fgf signaling. Instead, Wnt signaling normally acts to restrict the spatial extent of Fgf signaling in the dorsal diencephalon.

Given that Wnt and Fgf signaling are both important for the formation of dbx1b⁺ dHb progenitors, we explored the epistatic relationship between these pathways. Transcripts for dbx1b are reduced in the dorsal diencephalon of wls and fgf8a single mutants compared to WT siblings. However, mutants doubly homozygous for wls and fgf8a completely lack dbx1b expression in this region and fail to form dHb (Fig. 5I-L and data not shown). Treatment of wls homozygous mutants with the Fgf receptor inhibitor SU5042 produces a similar phenotype, reducing dbx1b-expressing progenitors in a dose-dependent manner (Fig. S5). Thus, Wnt and Fgf signaling act in an additive manner to generate dHb progenitors.

In addition to their role in the formation of dHb progenitors, Wnt and Fgf signaling are necessary for proper expression of Cxcr4-chemokine pathway components. This pathway includes Chemokine (C-X-C) motif ligand 12 (Cxcl12; formerly known as Stromal cell-derived factor-1 or Sdf1), which binds to the Cxcr4 receptor and the Atypical chemokine receptor 3 (Ackr3; formerly known as Chemokine receptor 7 or Cxcr7). Cxcr4 is the signaling receptor, whereas Ackr3 (the Ackr3b homolog) sequesters excess Cxcl12 to establish a local gradient (Dambly-Chaudiere et al., 2007; Lewellis and Knaut, 2012; Miyasaka et al., 2007; Boldajipour et al., 2008). Zebrafish have two Ackr3 and Cxcl12 homologs (a and b). At 35 hpf, cxcl12a transcripts are found in cells posterior to the dbx1b⁺ developing habenulae (Fig. 6A-C), whereas cxcl12b is expressed in cells surrounding the habenular region, at higher levels anterior to the region and at lower levels posterior to it (Fig. 6D-F, Fig. S6A). Cells expressing ackr3b lie anterior to the cxcl12b⁺ (Fig. S6B) and dbx1b⁺ (Fig. S6C) populations, as well as in the midline between the developing dHb.

Disruption of Wnt or Fgf signaling significantly expands the expression patterns of cxcl12b and ackr3b. In wls mutants, cxcl12b is ectopically transcribed within the developing habenulae and at high levels both anteriorly and posteriorly [3-fold increase along the anterior-posterior (A-P) axis, Fig. 6G-H′, Table 1, Fig. S4D]. By contrast, cxcl12a transcripts appear unaffected (Fig. S3D). In wls mutants, midline expression of ackr3b also extends more laterally (2-fold wider) and posteriorly (1.9-fold longer) (Fig. 6I-J′, Table 1, Fig. S4E).

The enlarged cxcl12b and ackr3b expression domains (Fig. 6G-J′) resemble the expanded fgf8a expression observed in wls mutants (Fig. 5A-C′), suggesting that Fgf8a positively regulates expression of these chemokine pathway members. Consistent with this, the cxcl12b domain in the dorsal diencephalon in fgf8a homozygous mutants is almost half the size of the WT domain at 35 hpf (Fig. 6K-L′, Table 1, Fig. S4). Expression of ackr3b is also affected but to a lesser degree (Fig. 6M-N′, Table 1, Fig. S4). In wls;fgf8a double mutants (Fig. 6O-R), or in wls mutants treated with SU5042 (Fig. 6S-U′), the domain of cxcl12b expression is reduced relative to wls homozygous mutants and becomes more similar in size to the WT pattern (Table 1, Fig. S4D). Taken together, the results demonstrate that Fgf and Wnt signaling act antagonistically to define the expression domains of these members of the chemokine signaling pathway.

The function of the Cxcr4b-chemokine pathway in the developing dHb is unknown. Chemokine signaling is known to mediate cell cohesion and migration (Miyasaka et al., 2007; Palevitch et al., 2010; Aman and Piotrowski, 2008); however, the size and morphology of the dHb appear normal in cxcr4b homozygous mutants (Fig. 7A,B and data not shown). Because Cxcr4-related receptors are also involved in axonal outgrowth and pathfinding by diverse neuronal types (Li et al., 2005; Miyasaka et al., 2007; Chalasani et al., 2003a,b; Lieberam et al., 2005), we examined the projections of dHb neurons.

As visualized by Tg(gng8:CAAX-GFP) labeling (Fig. 7A), axons of dHb neurons typically emerge at a lateral posterior position and fasciculate in the prominent bilateral fiber bundles of the fasciculus retroflexus (FR), which extend posteriorly to the midbrain IPN. We found that the majority of cxcr4b mutants have defective dHb axonal projections (Fig. 7B,E). Axonal phenotypes fall into three classes: in Class I larvae, the dHb-IPN projection is similar to WT; Class II larvae exhibit correctly directed axons that extend posteriorly, as well as ectopic anteriorly projecting axons; in Class III larvae, axons from the left and right dHb merge to form a thick bundle that extends anteriorly along the midline. In contrast to their WT or heterozygous siblings that predominantly showed the Class I phenotype (97.6%; n=41), most homozygous cxcr4b mutants displayed either Class II (48.6%) or Class III (45.7%) phenotypes (n=35). Similar defects were observed upon loss of the chemokine ligand Cxcl12a (Class II 66.6% and Class III 13.3%; n=30) (Fig. 7C,E). Consistent with the misregulation of cxcl12b and ackr3b expression observed in wls mutants (Fig. 6G-J′), axons that emerge from their smaller dHb are also misdirected (Class II 58.1%, Class III 14%; n=43) (Fig. 7D,E). Thus, although the Cxcr4b-chemokine signaling pathway is not required for the generation of dHb neurons, it is necessary for the correct outgrowth of their efferent axons.

Axon outgrowth from the dHb begins just prior to 2 dpf, with some axon terminals reaching the IPN between 2 and 3 dpf (Beretta et al., 2017; Kuan et al., 2007; Gamse et al., 2005). Cxcr4b receptors are present on the membrane of dHb neurons during this timeframe (Fig. 8A,B), as assayed using the transgenic reporter TgBAC(cxcr4b:cxcr4b-mKate2-IRES-GFP-CAAX) (Venkiteswaran et al., 2013). Internalization of the Cxcr4b-mKate2 fusion protein from the GFP-tagged cell membrane is indicative of active chemokine signaling (Venkiteswaran et al., 2013). Cxcr4b-mKate2 labeling is detected in the medioventral region of the dHb, where newly born neurons arise, as well as on their emerging axons. Although GFP labeling of axonal membranes extends from the dHb toward the IPN (Fig. 8A,B, arrows), the mKate2 signal is found on the proximal portion of these efferent projections and is absent more distally (Fig. 8A,B, arrowheads). This distribution of Cxcr4b receptors further supports a role for chemokine signaling in the initial outgrowth of axons from dHb neurons.

We investigated the formation of the zebrafish dHb nuclei and examined how the convergence of multiple signaling pathways influences their development and axonal projections. As summarized schematically in Fig. 9, Shh acts upstream or parallel to Wnt signaling, which functions additively with Fgf signaling to generate dbx1b-expressing dHb progenitors. The progenitors then give rise to cxcr4b-expressing neural precursors that differentiate into the mature neurons of the dHb. A second role for Wnt signaling is to delimit the spatial extent of Fgf activity in the dorsal diencephalon, which, in turn, regulates the expression of cxcl12b and ackr3b genes in the Cxcr4b chemokine signaling pathway. In the absence of Wnt signaling or upon loss of the Cxcr4b chemokine receptor or Cxcl12a ligand, dHb neurons exhibit aberrant axonal outgrowth.

In prior studies, two populations of cells were presumed to correspond to habenular progenitors in zebrafish: cells expressing cxcr4b and those expressing dbx1b (Roussigné et al., 2009; Halluin et al., 2016; Dean et al., 2014). Because only a small subset of cells shows overlapping expression, we sought to determine the relationship between the cell populations that express these genes.

The earliest known marker of the developing dHb is dbx1b, with transcripts detected at 24 hpf (Dean et al., 2014). The dbx1b-expressing cells are found in a ventromedial domain relative to mature habenular neurons. Consecutive imaging of the dHb demonstrated that labeling from a dbx1b transgenic driver decreases over time, becomes restricted to the most medioventral region, and persists until at least 4 weeks postfertilization. Lineage tracing experiments demonstrate that the dbx1b-expressing cells are progenitors, since they give rise to neurons throughout the larval (Dean et al., 2014) as well as the adult dHb. These findings are in agreement with expression analyses of the murine homolog Dbx1 and lineage studies in the mouse brain (Vue et al., 2007).

Expression of cxcr4b in the habenulae is first detected at 28 hpf, 4 h later than dbx1b (Roussigné et al., 2009). Time-lapse imaging reveals that cxcr4b-expressing cells do not migrate to the developing habenulae from more caudal regions, as was previously proposed (Kuan et al., 2015). Rather, labeling from a cxcr4b driver is initiated in a subset of cells that express GFP under the control of a dbx1b promoter. This indicates that cxcr4b⁺ cells are derived from dbx1b⁺ progenitors. Over time, dbx1b expression is downregulated in cxcr4b⁺ cells, which are found dorsal and lateral to the dbx1b⁺ progenitors. Expression of cxcr4b is also gradually lost, as cells in more dorsolateral positions begin to express markers of differentiated dHb neurons. The temporal pattern of cxcr4b expression parallels the previously described medial to lateral progression of dHb neurogenesis (Aizawa et al., 2007). Persistence of dTomato labeling from TgBAC(cxcr4b:nls-dTomato) in mature dHb neurons provides further support that cxcr4b expression demarcates a neural precursor population. Thus, development of the dHb is characterized by the transition from dbx1b⁺ progenitors to cxcr4b⁺ neural precursors and finally to elavl3⁺ neurons.

Disruption of Fgf or Wnt signaling pathways results in small dHb due to a reduction in progenitor cells (Regan et al., 2009; Dean et al., 2014; Kuan et al., 2015). It was previously reported that embryos homozygous for a mutation in fgf8a fail to form dbx1b⁺ progenitors (Dean et al., 2014). However, we find that dbx1b-expressing cells are present in fgf8a homozygous mutants at a later stage, although there are considerably fewer than in WT siblings. Fgf signaling could be responsible for generating the appropriate number of progenitors and/or for regulating dbx1b transcription. In support of the latter hypothesis, dbx1b expression is lost upon treatment of zebrafish embryos with SU5402, an inhibitor of Fgf receptor 1, but restored after its removal (Dean et al., 2014).

Through the analysis of wls mutants, we discovered an unsuspected role for the Wnt pathway in spatially restricting expression of fgf8a, and thereby limiting the extent of Fgf signaling in the developing dorsal diencephalon. In wls mutants, however, increased Fgf signaling does not compensate for the loss of Wnt signaling, as there are still fewer dbx1b-expressing dHb progenitors. Furthermore, wls;fgf8a double mutants or wls mutants treated with SU5402 fail to produce any dbx1b⁺ progenitors, indicating that the two signaling pathways work in an additive manner to generate the dHb progenitor population.

One hypothesis is that Wnt establishes a ‘pro-habenular’ territory, in which cells express dbx1b only after activation by Fgf signaling. Loss of Wnt signaling, as in wls mutants, results in smaller pro-habenulae and, therefore, fewer cells that could express dbx1b, irrespective of the level of Fgf signaling. This is a challenging hypothesis to test because dbx1b expression is the earliest known indicator of dHb formation.

Recent studies have also implicated Hh signaling as an early regulator of habenular development (Chatterjee et al., 2014; Halluin et al., 2016). In zebrafish embryos homozygous for a mutation in the smo gene, which encodes a G protein-coupled receptor essential for Hh signaling, expression of cxcr4b was reported as absent in the epithalamic region in one study (Halluin et al., 2016) and expanded in another (Chatterjee et al., 2014). In our experiments, neither cxcr4b nor dbx1b transcripts were detected in this region of the smo mutant brain. Thus, Hh signaling is required to produce dHb progenitors and/or to induce dbx1b expression. Based on these results, and the finding that expression of shha is unperturbed in the ZLI of wls homozygotes, we presume that the Hh pathway acts upstream of or parallel to Wnt signaling in defining the pro-habenular territory.

Homozygous fgf8a mutants show reduced expression of genes encoding the chemokine signaling components Cxcl12b and Ackr3b. However, in wls mutants, fgf8a expression and Fgf signaling domains are dramatically enlarged, and, as a consequence, so are the cxcl12b and ackr3b expression domains. Reduction of Fgf signaling in wls mutants by mutation of fgf8a or chemical inhibition restores cxcl12b expression to a WT pattern. These results demonstrate that Fgf8a normally functions downstream of Wnt signaling to define the spatial expression of chemokine pathway components in regions surrounding the habenulae.

A regulatory relationship between Wnt signaling, Fgf signaling and ackr3b expression was previously reported in the migrating lateral line primordia of zebrafish (Aman and Piotrowski, 2008). In this tissue, Fgf induces the expression of Wnt inhibitors, whereas Wnt induces expression of both Fgf ligands and Fgf signaling inhibitors (Aman and Piotrowski, 2008). This creates mutually exclusive regions of Wnt and Fgf signaling, where Wnt signaling is active at the leading tip of the migratory primordium and Fgf signaling occurs in the trailing zone. Expression of ackr3b is spatially restricted to the trailing cells, but it is unclear whether this is caused by inhibition by Wnt in the leading zone, or by activation by Fgf and/or relief of Wnt inhibition in the trailing cells. Although the precise genetic interactions in the lateral line may differ from those in the dorsal diencephalon, there are parallels between the two, as both tissues rely on the opposing forces of Wnt and Fgf signaling to regulate the spatial expression pattern of a chemokine receptor.

We have uncovered a new role for the Cxcr4-chemokine pathway in the establishment of dHb-IPN connectivity. Although cxcr4b transcription is downregulated in dHb neurons, the protein likely persists, as evidenced by labeling of a Cxcr4b fluorescent fusion protein. Loss of the chemokine receptor results in atypical rostral projections of habenular efferents. Similar axonal phenotypes are observed in mutants lacking the Cxcl12a chemokine ligand and in wls mutants, in which expression of genes encoding the Cxcl12b chemokine and the chemokine receptor Ackr3b are both misregulated.

The function of Cxcl12 chemokines in axonal pathfinding is complex and depends on the neuronal type being assayed. For example, Cxcl12 is a repulsive cue for efferents of rat cerebellar neurons (Xiang et al., 2002), but an attractant for axons of retinal ganglion cells (Li et al., 2005), and creates a permissive environment for a number of cell types including olfactory neurons (Miyasaka et al., 2007), retinal ganglion cells (Chalasani et al., 2003a, 2007), dorsal root ganglia neurons (Chalasani et al., 2003b) and ventral motor neurons (Lieberam et al., 2005).

The projection defects of dHb neurons that result from loss of Cxcr4b could, thus, be caused by the inability of their axons to respond to a posterior attractive cue and/or an anterior repulsive cue. The cxcl12a gene is expressed posterior to the developing habenulae. The defects in axonal outgrowth observed in cxcl12a mutants indicate that the chemokine might serve as an attractant that guides axons caudally, or, alternatively, creates a permissive environment for dHb axon outgrowth. Conversely, cxcl12b transcripts are abundant anterior to the dHb, suggesting that Cxcl12b repels axons from extending rostrally. An examination of cxcl12b mutants is required to test this hypothesis. The defective axonal outgrowth phenotype is less penetrant in cxcl12a mutants than in cxcr4b mutants, suggesting that the Cxcl12a and Cxcl12b ligands might act in conjunction to regulate the direction of dHb axonal outgrowth.

In wls mutants, expression of cxcl12a is unaltered and cxcr4b is expressed in the developing dHb, yet axons still project aberrantly. The gene encoding the Cxcl12b ligand is inappropriately transcribed throughout the dHb. High levels of chemokines have been reported to function as dominant negatives for chemokine signaling by binding to and driving internalization of the receptor (Miyasaka et al., 2007). It is possible that the expanded cxcl12b expression in wls mutants also has a dominant-negative effect, resulting in aberrant, anteriorly directed axonal outgrowth. In cxcr4b, cxcl12a and wls mutants, the many rostrally directed axons from both the left and right dHb extend toward the midline, where they fasciculate and project anteriorly as a single bundle (Class III phenotype). This suggests that, in the absence of normal chemokine signaling, these axons can now respond inappropriately to an anterior midline cue.

These results indicate that chemokine signaling influences the direction of dHb axon outgrowth, and is reminiscent of the action of ventral motor neuron axons, which upon loss of chemokine signaling choose the exit point of dorsal motor neurons (Lieberam et al., 2005). Axons that do emerge caudally from mutant dHb successfully join the FR and innervate the IPN. Successful pathfinding by this subset of axons indicates that other signaling pathways mediate axon fasciculation, extension of the FR through the midbrain, and target recognition. Several axon guidance cues and receptors have been implicated in aspects of this process including Netrin/DCC (Schmidt et al., 2014; Funato et al., 2000), Sema3F/Nrp2 (Giger et al., 2000; Chen et al., 2000; Sahay et al., 2003), Sema5A (Kantor et al., 2004) and Sema3D/Nrp1a (Kuan et al., 2007). Disruption of most components leads to defasciculation of the FR or to incorrect innervation of the IPN. Only the loss of DCC or Netrin results in ectopic projections outside the FR (Schmidt et al., 2014).

If the chemokine signaling pathway is not required for guidance of dHb neurons to the target, but rather influences the initial outgrowth of their axons, newly born dHb neurons and their emerging axons, but not older neurons or axons approaching or innervating the IPN, would be expected to respond to chemokine signals. Indeed, labeling by the Cxcr4b fluorescent fusion protein is detected in medially positioned neurons found closest to the ventricular zone. Receptors are also enriched in axons newly emerging from the dHb, but are absent distally as they extend closer to the IPN. This distribution of Cxcr4b receptors suggests that they are rapidly internalized upon chemokine binding and degraded as axons elongate.

Understanding the processes underlying development of the habenular nuclei and the generation of their neuronal diversity and connectivity is key to unraveling their roles in behavior and mood disorders. Integration of the multiple genes and signaling pathways implicated in dHb development is an important step toward determining how their neurons arise and form appropriate synaptic connections.

Zebrafish were kept on a 14 h-light/10 h-dark cycle at 27°C. Embryos were raised in system water or, for experiments, maintained in system water with 0.003% phenylthiourea (1× PTU) to reduce pigmentation. The AB WT strain (Walker, 1998) and lines bearing the wls^c186 (Kuan et al., 2015), fgf8^ti282a (Brand et al., 1996), cxcr4b^t26035 (Knaut et al., 2003) and smo^hi1640Tg (Chen et al., 2001) mutations were used. Homozygous wls^c186 mutants were genotyped as embryos or identified as larvae according to morphological criteria (Kuan et al., 2015). Embryos and larvae homozygous for cxcr4b^t26035 were genotyped using published primers (Miyasaka et al., 2007). The transgenic lines used include TgBAC(gng8:Eco.NfsB-2A-CAAX-GFP) [referred to as Tg(gng8:CAAX-GFP)] (deCarvalho et al., 2013), TgBAC(dbx1b:Cre-mCherry) (Koyama et al., 2011), TgBAC(cxcr4b:nls-dTomato) (Donà et al., 2013), TgBAC(dbx1b:GFP) (Kinkhabwala et al., 2011), Tg(hsp70:dkk1-GFP) (Stoick-Cooper et al., 2007), TgBAC(dbx1b:tdTomato-nls) (Satou et al., 2012), Tg(HuC:H2B-GCamP6S) (Vladimirov et al., 2014), TgBAC(cxcr4b:cxcr4b-mKate2-IRES-GFP-CAAX) (Venkiteswaran et al., 2013) and Tg(β-actin:loxP-hmgb1-eCFP-loxP-H2B-mCherry) (M. Parsons, Johns Hopkins University, Baltimore, MD, USA, unpublished). Zebrafish maintenance and experimental procedures were performed according to protocols approved by the Carnegie Institution for Science Animal Care and Use Committee.

Colorimetric and fluorescent in situ hybridization were performed according to previously described methods (deCarvalho et al., 2013). RNA probes were synthesized for dbx1b (Gribble et al., 2007), etv5b, ackr3b, cxcr4b, wls (Thisse et al., 2001), fgf8 (Fürthauer et al., 1997), ano2 (deCarvalho et al., 2014), dusp6 (Tsang et al., 2004), cxcl12a (Doitsidou et al., 2002), cxcl12b (Thisse and Thisse, 2005) and elavl3 (Kim et al., 1996). GFP was detected using rabbit antisera (TP401, Torrey Pines Biolabs) and a Cy3-conjugated AffiniPure anti-rabbit IgG secondary antibody (1:500, 111-165-144, Jackson ImmunoResearch). Immunolabeling was performed as described (deCarvalho et al., 2013).

To inhibit Wnt signaling, Tg(hsp70:dkk1-GFP) embryos were heat shocked at 24 hpf according to published methods (Stoick-Cooper et al., 2007). For inhibition of Fgf signaling, embryos were maintained in 1× PTU and, between 24 and 48 hpf, treated with the FGFR inhibitor SU5402 (12-24 µM, SML0043, Sigma-Aldrich) in 0.3% dimethyl sulfoxide (DMSO) or exposed to vehicle alone. Embryos were washed for 5 min before fixation in 4% paraformaldehyde (PFA).

A Zeiss AxioCam HRc camera mounted on a Zeiss Axioskop was used to obtain bright-field images. Fluorescent images were captured on an upright or inverted Leica SP5 confocal microscope fitted with a 25×/0.95 NA water immersion objective. Fixed and live embryos were mounted in 1.2% low-melt agarose (SeaPlaque, Lonza) in 1× phosphate buffered saline (PBS) or fish system water with 200 mg/l tricaine (Sigma-Aldrich), respectively. Sectioning and imaging of adult brains was performed as previously described (deCarvalho et al., 2014), and sections were cut every 60 µm.

Confocal images were analyzed using Imaris software (Bitplane). Measurements obtained from colormetric in situ hybridization images were collected using the Fiji imaging processing package (https://fiji.sc/) (Schindelin et al., 2012; Schindelin et al., 2015) and provided as the mean±s.e.m. Statistical analyses were performed using the Student’s two-tailed t-test (Microsoft Excel).

Embryos were sedated in 165 mg/l tricaine in 1× PTU and mounted in 1% low-melt agarose in a glass-bottomed dish (WillCo Wells). These tricaine and agarose concentrations were shown to reduce developmental defects in long-term imaging experiments (Kaufmann et al., 2012). Time-lapse imaging was performed on a Leica SP5 inverted confocal microscope with a 25×/NA 0.95 water immersion lens. Z-stacks of the dorsal embryonic brain (∼100-300 µm, step size 1 µm) were acquired at 15 min intervals. Imaging was performed for 24 h, during which deionized water was continually pumped between the objective and glass-bottomed dish. A metal collar (Leica Water Immersion Micro Dispenser cap) was mounted over the objective and attached to a 10 ml syringe using flexible polymer tubing (Tygon). The syringe was placed in a syringe pump (NE-300, New Era Pump Systems), which was set to dispense at a rate of 40 µl/h.

We thank Stephen Devoto, Richard Dorsky, Joshua Gamse, Darren Gilmour, Holger Knaut, Michael Parsons and Michael Tsang for providing reagents and mutant and transgenic zebrafish; Holger Knaut for his valuable advice on experimental design; Michelle Macurak for technical assistance; and Andrew Rock and Carmen Tull for care of the aquatics facility.