Hedgehog regulated Slit expression determines commissure and glial cell position in the zebrafish forebrain

In bilateral organisms, some axons cross the midline of the CNS to form commissures that functionally connect the two sides of the nervous system. Axonal growth cones navigate towards, across, and then away from the midline in response to precise signals present along the growth substrate, including intermediate cellular and molecular guidance cues (reviewed by Goodman, 1996; Grunwald and Klein, 2002; Kaprielian et al., 2001; Steward, 2002). A variety of secreted molecules, including Netrins and Sonic Hedgehog, function to attract axonal growth cones towards the midline of the spinal cord, whereas other guidance cues, including Slits, Semaphorins and Ephrins, typically function to repel growth cones. In the Drosophila nerve cord, the midline repellent Slit plays a major role in establishing which commissural axons cross the midline and how far from the midline non-crossing axons are positioned (Battye et al.,1999; Harris and Holt,1999; Kidd et al.,1999; Simpson et al.,2000a; Simpson et al.,2000b). Slit repulsion is mediated by Roundabout (Robo) receptors expressed in growth cones (Keleman et al.,2002).

In the vertebrate forebrain, Slit-Robo-mediated repulsion is crucial for the appropriate dorsoventral position of the optic chiasm(Rasband et al., 2003). In mice lacking both Slit1 and Slit2 function, retinal axons cross the midline in multiple locations around the true chiasm(Plump et al., 2002). Similarly, zebrafish lacking astray (robo2) function show aberrant retinal axon growth across the midline(Hutson and Chien, 2002; Karlstrom et al., 1996). Analyses of these two phenotypes led to the proposal that gradients of Slit expression descending toward the optic chiasm channel Robo-expressing axons within the appropriate pathway(Hutson and Chien, 2002; Plump et al., 2002; Rasband et al., 2003; Richards, 2002b). Slit1 and Slit2 have also been shown to channel callosal axons across the cortical midline in the mouse forebrain (Bagri et al., 2002; Shu et al.,2003d). It is not known whether the Robo/Slit system also influences formation of the first forebrain commissures in vertebrates, the postoptic commissure (POC) and the anterior commissure (AC). However, the fact that commissures form in astray (robo2) mutants indicates that Robo2 alone is not necessary for POC and AC midline crossing(Karlstrom et al., 1996).

The cellular growth substrate for retinal axons as they cross the midline consists of neurons, neuroepithelial cells and glia(Marcus and Easter, 1995; Mason and Sretavan, 1997). CD44/SSEA1-positive neurons border the optic nerve in and across the mammalian midline, and may provide a barrier that directs RGC axons to cross the midline in the correct position (Jeffery,2001; Marcus et al.,1995; Marcus and Mason,1995; Mason and Sretavan,1997). Other identifiable midline cells include a radial glial`palisade' (Misson et al.,1988) that helps to guide ipsilaterally projecting retinal axons(Mason and Sretavan, 1997). More dorsally, midline glial populations that make up the `glial sling',`glial wedge', and indusium griseum have all been implicated in midline crossing in the corpus callosum (Richards,2002a; Shu et al.,2003a; Shu et al.,2003b; Shu et al.,2003c; Shu and Richards,2001; Silver et al.,1982; Silver and Ogawa,1983). Little is known about how midline cells influence the formation of the forebrain commissures.

The zebrafish forebrain provides an easily accessible system for the study of chiasm and commissure formation. Prior to chiasm formation, the POC forms in the diencephalon, while the AC forms more anterodorsally in the telencephalon (Bak and Fraser,2003; Wilson et al.,1990). RGC axons grow across the ventral midline in close proximity to the POC (Wilson et al.,1990). Pioneering growth cones of the POC and optic nerve grow superficially (Wilson and Easter,1991; Wilson et al.,1990), and their growth substrate primarily consists of poorly-characterized neuroepithelial cells and basal lamina(Burrill and Easter, 1995; Wilson and Easter, 1991; Wilson et al., 1990). It is currently unknown whether these neuroepithelial cells are neural and/or glial precursor cells, neurons or glia (Burrill and Easter, 1995), although some express Glial Fibrillary Acidic Protein (Gfap) and have a radial glial morphology, and so appear to be astroglia (Marcus and Easter,1995). These Gfap+ cells come in direct contact with POC and optic axons at the midline at 48 hours post-fertilization(Marcus and Easter, 1995).

Several achiasmatic zebrafish mutants have been identified in which forebrain axons fail to cross the midline and most RGC axons project ipsilaterally (Karlstrom et al.,1997; Karlstrom et al.,1996). Many of these ipsilateral/midline mutants affect Hedgehog(Hh)-mediated midline patterning, with detour (dtr) and you-too (yot) encoding the Hh-responsive transcription factors Gli1 and Gli2, respectively(Karlstrom et al., 1999; Karlstrom et al., 2003). Complex regulation and processing of the vertebrate gli genes results in either repression or activation of Hh target gene expression(Koebernick and Pieler, 2002),and both dtr(gli1) and yot(gli2) mutations block the ability of cells to activate Hh signaling(Karlstrom et al., 2003). The cellular and molecular basis for the axon guidance defects in the Hh/midline mutants is currently unknown. Besides playing an important role in directing cell differentiation in the forebrain(Nybakken and Perrimon, 2002),Hh proteins directly attract commissural axons to the midline in the mouse spinal cord (Charron et al.,2003; Ogden et al.,2004) and may repel RGC axons in the chick forebrain(Trousse et al., 2001). It is thus possible that the midline guidance errors in Hh pathway mutants are due to a loss of direct Hh-mediated axon guidance. Alternatively, they may be indirectly caused by cell specification defects in the forebrain(Karlstrom et al., 2003; Sbrogna et al., 2003).

Although much has been learned about midline guidance in the forebrain,many fundamental questions remain. First, what is the molecular and cellular architecture of the midline growth substrate that leads to proper commissure and then chiasm formation? Second, how do Slits influence the first midline-crossing axons in the forebrain, those of the POC and AC? Finally,given the loss of midline crossing in Hh pathway mutants, does Hh directly or indirectly affect axon guidance in the forebrain? We show that Gfap+ cells express slit1a and form a `bridge' across the midline prior to commissure and chiasm formation, and that Slit molecules help to establish the POC. We show that Hh is not needed directly to guide commissural and retinal axons near the midline, but instead is necessary to establish the proper patterning of midline glia and the patterned expression of slitguidance cues. We propose a model in which slit1a expressed by a glial bridge provides a substrate for axon crossing at the midline, whereas slit2 and slit3 repulsion helps to channel growth cones into commissures.

Fish lines were maintained in the University of Massachusetts-Amherst Zebrafish Facility. Wild-type embryos were from the TL line; the you-too (yot^ty119) and detour(dtr^ts269) lines were maintained in TL and Tübackgrounds. The you-too (yot^ty119) allele encodes a truncated Gli2 protein that acts as a dominant repressor(Gli2DR) to block Gli-mediated activation, while dtr^ts269is a gli1 loss-of-function mutation(Karlstrom et al., 2003). Embryos were grown at 28.5°C and staged in hours post-fertilization (hpf)(Kimmel et al., 1995),dechorionated with 0.2 mg/ml pronase (Sigma). 0.003% 1-phenyl-2-thiourea (PTU)was added starting at 23 hpf to reduce pigmentation(Westerfield, 1993). No differences were seen between PTU-treated and untreated embryos in overall growth or assayed phenotypes.

Anti-acetylated tubulin (IgG2b monoclonal, Sigma) was used at a dilution of 1:800 to label axons (Wilson et al.,1990). A polyclonal anti-goldfish-Gfap antibody (generous gift of Sam Nona, University of Manchester, UK)(Nona et al., 1989) was used at 1:400 to label astroglial cells (Marcus and Easter, 1995). F59 (IgG1 monoclonal, Developmental Studies Hybridoma Bank) is specific for slow muscle myosins in chicken(Crow and Stockdale, 1986), and diluted at 1:10 preferentially labels slow muscle fibers in zebrafish(Devoto et al., 1996). Secondary antibodies from Jackson ImmunoResearch Laboratories were diluted as follows: FITC goat anti-mouse IgG, 1:200; Cy5 goat anti-mouse IgG, 1:200;TRITC goat anti-Rabbit, 1:500; FITC goat anti-Rabbit, 1:200.

Embryos were fixed in 4% formaldehyde (Ted Pella) for 2 hours at room temperature and antibody-labeled, as previously described(Karlstrom et al., 1999) with several modifications. Briefly, embryos were dehydrated, incubated in 100%acetone for 7 minutes at –20°C, rehydrated in PBS+0.2% Triton X-100(PBS-Tx), digested with 10 μg/ml Proteinase K for 3-5 minutes depending on age, washed for 3×5 minutes in PBS-Tx, and incubated in blocking solution (PBS-Tx-2%BSA-5%NGS-1%DMSO) for 1 hour at room temperature. All primary and secondary antibodies were diluted in blocking solution and applied for 2 hours at room temperature.

Digoxigenin-labeled anti-sense mRNA probes were synthesized using SP6/T7/T3 Dig RNA Labeling Kits (Roche). The probes used were pax2a(Krauss et al., 1991), shh (Krauss et al.,1993), ptc1 (Concordet et al., 1996), sema3d(Halloran et al., 1998), slit1a (Hutson et al.,2003), slit2 and slit3(Yeo et al., 2001), robo1,robo2 and robo3 (Lee et al.,2001), and robo4(Park et al., 2003). A standard in situ hybridization protocol(Jowett, 1997) was modified to enable multi-antibody labeling following mRNA probe hybridization. Briefly,embryos were fixed with 4% formaldehyde (Ted Pella) for 2 hours at room temperature. Anti-acetylated tubulin and anti-Gfap primary antibodies were added with anti-DIG antibodies. Following a Fast Red (Roche) color reaction,embryos were washed for 2×15 minutes with PBS-0.1% Tween20, and then for 3×15 minutes with PBS-Tx, blocked, and incubated in secondary antibodies. Embryos were never post-fixed.

Embryos were cleared in 75% glycerol and examined using a Zeiss LSM510 laser-scanning confocal microscope at a magnification of 40× or 63×. For lateral views of the forebrain, eyes were removed. Three-dimensional stacks were acquired using multitracking for double- and triple-labeled embryos to ensure sequential acquisition and preclude signal crossover. Stacks ranged from 20 μm to 50 μm in Z-distance with constant pinhole settings of 105.7 μm (∼1.0 Airy unit) for all channels and an(optimal) interval of 0.4 μm. Images are maximum intensity projections unless otherwise noted. For consistency, throughout this paper we use`anterior' to mean closer to the telencephalon, and `posterior' for closer to the diencephalon, despite changes in orientation due to flexure of the nervous system.

At several different time points, wild-type embryos were treated with 100μM or 200 μM Cyclopamine (Toronto Chemical) in embryo medium + 0.5% EtOH(Incardona et al., 1998) at 28.5°C. No effect on the assayed phenotypes was seen in control embryos incubated in embryo medium + 0.5% EtOH. Thirty embryos were treated in 2 ml volumes in 12-well plates.

Translation-blocking (Nasevicius and Ekker, 2000) zebrafish slit1a (GACAA CATCC TCCTC TCGCA GGCAT), slit2 (CATCA CCGCT GTTTC CTCAA GTTCT) and slit3(TATAT CCTCT GAGGC TGATA GCAGC) morpholinos (MOs, GeneTools) were kept as 10 mg/ml stocks in Danieau's solution (58 mM NaCl, 0.7 mM KCl, 0.4 mM MgSO₄, 0.6 mM Ca(NO₃)₂, 5 mM Hepes, Phenol Red) and diluted in Danieau's solution. Injections of 400-500 pl were made into the yolk of one- to four-cell stage embryos obtained from wild-type crosses or yot^ty119 (gli2DR) heterozygous intercrosses. slit1a MOs were injected at 1 ng, slit2 or slit3 at 2-4 ng, with no phenotypic differences seen between these doses of slit2 or slit3 MOs. Co-injections of slit2 and slit3 MOs used 2.5 ng of each. Injected embryos were fixed at 28 hpf for triple antibody labeling of axons, glial cells and slow muscle fibers. Overexpression of Slit genes can cause convergent extension defects(Yeo et al., 2001), as can high doses of slit1a, slit2 or slit3 MOs (L.D.H. and C.-B.C., unpublished). To rule out the possibility that defects in axon guidance were due to indirect morphogenetic effects, each injected embryo was stained with F59, and only those that displayed normal body shape, somite formation and patterning of the superficial slow-muscle layer were used to score axon and glial defects. Homozygous yot (gli2DR) mutant embryos were identified by the loss of embryonic slow muscle fibers at 28 hpf(Stickney et al., 2000). Zebrafish shh was transcribed from the T7TSshh plasmid(Ekker et al., 1995) and microinjected as previously described(Barresi et al., 2000).

To better characterize the cellular growth substrate for commissural axons,we examined the relationship between Gfap+ cells, gene expression borders, and commissural and retinal axons in the forebrain (Figs 1 and 2). In the regions of the AC and POC, Gfap+ cells span the midline superficial to neuroepithelial cells and send radial fibers into the forebrain (Fig. 1B, see also Movies 1, 2 and 3 in the supplementary material) with AC and POC axons arranged perpendicular to these fibers. To orient these cells in relation to the optic stalk and diencephalic midline, we examined Gfap+cells in relation to pax2a(Macdonald et al., 1997) and shh (Ekker et al.,1995; Krauss et al.,1993) expression. The anteriormost Gfap+ cells in the diencephalon are adjacent to, but do not overlap with, pax2a-expressing cells of the optic stalk (Fig. 1C,D). These Gfap+ cells lie superficial to neuroepithelial cells that express sonic hedgehog (shh; Fig. 1E,F). The astroglial fibers that stretch into the brain appear to overlap with shh-expressing cells at the midline from 15 hpf to 27 hpf of development; however, by 36 hpf the Gfap+ fibers no longer overlap with the zone of shh expression (Fig. 1E,F, data not shown).

Previous studies reported that Gfap+ cells appear in the midline region only after forebrain commissures have formed, suggesting that astroglia probably do not guide axons across the midline(Marcus and Easter, 1995). However, our confocal analysis indicates that Gfap+ cells span the midline of the forebrain at 15 hpf, well before the appearance of any axons in the brain(Fig. 2)(Wilson et al., 1990). Gfap+cells span the diencephalic midline in the future POC region(Fig. 2A), and by 18 hpf Gfap+cells span the midline where both the POC and AC will form(Fig. 2B). As development proceeds, Gfap+ cells become increasingly restricted to the locations of the AC and POC (Fig. 2C, bracket; Fig. 2D-F). The first POC axons begin crossing the midline at approximately 23-24 hpf(Bak and Fraser, 2003) in association with Gfap+ cells (Fig. 2C, arrow). RGC axons (Fig. 2E,F, arrowheads) then grow towards the midline close to, but not in direct contact with, POC axons (Fig. 2E,F, bottom arrow) (Burrill and Easter, 1995). A small cluster of Gfap+ cells extends anteriorly from the glial bridge directly in the position where RGC axons cross the midline (Fig. 2F,arrowhead).

The expression of shh adjacent to the POC and optic chiasm(Fig. 1D), combined with the loss of midline axon crossing in Hh mutants (reviewed by Russell, 2003) and the recent findings that Shh can directly guide midline crossing axons in mouse(Charron et al., 2003) and chick (Trousse et al., 2001),suggest that Shh may directly guide POC and/or RGC axons near the midline of the forebrain. To test this, we used cyclopamine (CyA) to pharmacologically block Smoothened-mediated Hedgehog signaling starting just prior to the times when POC (22 hpf) or RGC (27 hpf) axons grow toward the midline(Fig. 3A). In embryos treated with 100 μM CyA from 27 hpf to 36 hpf, RGC axons crossed the midline normally, forming a wild-type chiasm (Fig. 3E,F). Similarly, the POC and optic chiasm were completely formed in embryos treated with CyA from 22 hpf to 36 hpf(Fig. 3D,F). To verify that CyA was completely blocking Hh signaling at these ages, we examined late-onset expression of the Hh receptor and the Hh target gene patched1(ptc1) in the fin bud, which normally begins at 32 hours of development (Concordet et al.,1996; Lewis et al.,1999). Treatment with 100 μM CyA from 27 hpf to 36 hpf led to a complete loss or major reduction of ptc1 expression in the fin bud(data not shown), whereas 200 μM CyA treatment eliminated all ptc1expression in the fin bud, indicating a complete block of Hh signaling(Fig. 3B-F; upper insets)(Neumann et al., 1999). Higher doses of CyA (200 μM) applied between 22-36 hpf also led to a reduction in the number of RGCs, consistent with a role for Hh in RGC specification(Neumann and Nuesslein-Volhard,2000). However, both POC and RGC axons correctly crossed the midline in these treated embryos (Fig. 3D,E; lower inset).

Because axons fail to cross the midline of the forebrain in Hh signaling pathway mutants (Culverwell and Karlstrom,2002), we next treated embryos with 100 μM CyA at earlier time points to determine when Hh plays a role in commissure formation. CyA treatment from 10 to 36 hpf resulted in major axon pathfinding errors of all commissural and the retinal axons (Fig. 3B, green) phenocopying the more severe Hedgehog pathway mutants(Culverwell and Karlstrom,2002; Russell,2003; Varga et al.,2001). In addition, Gfap+ cells were reduced and highly disorganized (Fig. 3B, red)following these early CyA treatments. Similar to the Hh mutants, these embryos exhibited a loss of midline forebrain tissues and partial cyclopia(Fig. 3B), suggesting that axon and glial cell disruptions could be due to the loss of midline cells. However,embryos treated with CyA from 15 h to 36 h of development showed defects in axon guidance despite the presence of midline tissue(Fig. 3C). In these embryos,POC axons were disorganized and often formed several separate bundles(Fig. 3C, arrow). In some embryos, the POC was extremely reduced and many POC axons wandered into the preoptic area (Fig. 3C; lower left inset). In addition, RGC axons completely failed to cross the midline and grew ipsilaterally (Fig. 3C,green, arrowheads). Importantly, CyA treatment at these same stages disrupted the glial bridge that spans the midline in the region of the POC(Fig. 3C, red).

To better understand the relationship between Hh signaling and formation of the midline glial bridge, we examined yot (gli2DR) mutants that have a repressed transcriptional response to Hh without a major loss of midline tissue (Karlstrom et al.,1999). In yot(gli2) mutants, Gfap+ cells were reduced at the midline and many cells were aberrantly positioned in the region between the two commissures (Fig. 4). This disorganization of Gfap+ cells was apparent before axons are present(Fig. 4A,D). As seen in the 15-36 hpf CyA-treated embryos, misguided POC axons in yot mutants were predominantly associated with mis-positioned Gfap+ cells(Fig. 3C, Fig. 4F). To assess whether ectopic expression of Hh would disrupt axon and glial guidance, we injected Shh-encoding mRNA at the two-cell stage. The POC was reduced and defasciculated following global Shh overexpression(Fig. 3G-I, arrows) and Gfap+cells were similarly expanded along the anteroposterior axis(Fig. 3G-I, brackets). Together, these results suggest that Hh signaling is required for the proper patterning of glial bridges in the forebrain, and that axon errors seen in Hh pathway mutants may be the indirect effect of disruptions in this axon substrate.

To better understand how Hh signaling is involved in establishing the commissural axon growth substrate, we next examined axon guidance molecule expression in yot (gli2DR) mutants. Given the role for Robo/Slit-mediated growth cone repulsion in RGC axon guidance(Hutson and Chien, 2002; Hutson et al., 2003), we first examined the expression of slit1a and slit2/3 during commissure formation (Fig. 5). Consistent with the surround repulsion guidance mechanism seen for retinal axons (Rasband et al., 2003), slit2 and slit3 were expressed adjacent to the commissures,with expression being absent from regions where commissural axons and glial bridges cross the midline (Fig. 5A,C). In yot(gli2) mutants, slit2 and slit3 were expanded, with expression filling the commissural regions normally devoid of slit2/slit3 expression(Fig. 5B,D). In contrast to slit2/slit3, slit1a was expressed in the preoptic area and AC region, with expression overlapping both POC and AC axons(Fig. 5E). A subset of Gfap+cells in the glial bridge and the majority of the Gfap+ cells in the AC region expressed slit1a (Fig. 5G,H, data not shown), although none of these cells expressed either slit2 or slit3. slit1a expression was reduced in yot (gli2) mutants in the preoptic area(Fig. 5F), unlike the expansion of expression seen for slit2/slit3.

The secreted guidance molecule sema3d(Halloran et al., 1998) is expressed at the midline directly underneath the POC and is thus in a position to guide axons at the midline (Fig. 5G). Similar to slit1a, sema3d was severely reduced in the diencephalon of yot (gli2DR) mutants(Fig. 5H). We also examined expression of ephrinB2a and ephrinB3, secreted molecules that are expressed near the chiasm and POC, and that are known to play a role in RGC axon guidance (Chan et al.,2001). Surprisingly, despite other midline patterning defects in the region (Karlstrom et al.,1997), expression of ephrinB2a and ephrinB3 was largely unaffected in yot (gli2DR) mutant embryos (data not shown).

To test whether and how Slit proteins regulate glial bridge and commissure formation, we next used morpholino antisense oligonucleotide (MO) injections to reduce Slit function in wild-type and mutant embryos. Inhibition of slit2, slit3, and slit2/slit3 together in wild-type embryos resulted in defasciculation of the POC, with many distinct axon bundles crossing throughout the pre- and post-optic areas(Fig. 6A-D,K). To semi-quantify these POC defects, we generated a crossing index of POC formation based on a scale of 1 to 7 (with 7 being an ideal wild-type commissure and 1 being no commissure) for individual embryos injected with different MOs(Fig. 6L). Inhibition of slit2, slit3, and slit2/slit3 together in wild type embryos significantly disrupted POC formation(Fig. 6L, purple bars). Besides leading to defasciculation, slit2 or slit2/slit3 MO injections caused some POC axons to wander into inappropriate regions of the forebrain in most embryos (Table 1). slit3 MO and slit1a MO injections did not significantly increase axon wandering (Fig. 6A-D,K; Table 1). In some slit MO-injected embryos the AC was variably reduced, whereas in others it appeared to be unaffected, despite clear POC defects(Fig. 6A-E,K).

Inhibition of Slit1a function in wild-type embryos disrupted POC formation more severely than inhibition of Slit2 and/or Slit3, and produced distinct axon guidance errors with a marked reduction in crossing fibers(Fig. 6A,D,E,K). slit1a MO injection reduced POC formation in wild-type embryos by approximately 50% (Fig. 6E,K,L), with only a small increase in the number of wandering POC axons (Table 1). Reduction in POC formation by slit1a MOs suggests that Slit1a promotes midline crossing, whereas defasciculation and increased wandering following Slit2/Slit3 inhibition suggests that Slit2/Slit3 restrict where POC axons can cross the midline.

Because injection of slit2/slit3 MOs led to axon defects consistent with the known roles for Slit proteins as axon repellents, we wondered whether the expansion of slit2/slit3 expression in yot (gli2DR) mutants could be responsible for the inability of POC and RGC axons to grow across the midline. To answer this question, we assayed POC formation in yot (gli2DR) mutants injected with slit MOs(Fig. 6F-I,K). slit2and slit3 single MO injections partially rescued POC formation in yot (gli2DR) mutants (Fig. 6L), whereas injection of both slit2 and slit3MOs together rescued POC formation to a level that was not significantly different from that seen in slit2/slit3 MO-injected wild-type embryos. yot (gli2DR) mutations led to a significant increase in POC axon wandering, which was further increased after injection of slit2 and/or slit2/slit3 MOs(Fig. 6; Table 1). Together, these results support the idea that Slit2 and Slit3 act as repellent guidance cues for POC axons, and that their misexpression in yot (gli2DR) mutants is a major cause of POC defects seen in this mutant. By contrast, slit1a MO injection did not rescue POC formation in yotmutants, and instead caused a further reduction in axon crossing(Fig. 6J-L). These knockdown results support the idea that Slit1a functions distinctly from Slit2/Slit3,with Slit1a being necessary for POC axons to grow across the midline and Slit2/Slit3 acting as repellent guidance cues.

Given the slit2/slit3 expression on either side of the forebrain glial bridges (Fig. 5), and the correlation between Gfap+ cell position and axonal defects at the midline in yot (gli2DR) mutants (Figs 1, 2, 4), we next wondered whether Slit molecules might play a role in positioning glial cells in the forebrain. Indeed, loss of Slit1a, Slit2, Slit3, or both Slit2 and Slit3 function in wild-type embryos led to an anteroposterior spreading of Gfap+ cells in the pre- and post optic areas (Fig. 6A-E). This spreading was apparent at 22 hpf, prior to midline axon growth in the forebrain (Fig. 6A,D, insets).

Glial cells responded more drastically to Slit knockdown when we injected slit MOs into yot (gli2DR) mutants. In this case, injection of slit2, slit3 and slit2/slit3 MOs often led to a complete restoration of the glial bridge across the midline in the POC region(Fig. 6A,F-I). The extent of glial bridge rescue correlated with rescued POC formation. In contrast to Slit2 and Slit3, knocking down Slit1a function did not rescue glial bridge formation in yot (gli2DR) mutants(Fig. 6F,J). Again, in all slit MO-injected yot (gli2DR) embryos, aberrantly positioned axons appeared to be associated with Gfap+ cells.

To determine which cell populations are potentially capable of responding directly to Slit cues, we determined whether commissural neurons and midline glial cells express one or more of the Roundabout (Robo) family of Slit receptors. Double labeling using different robo in situ probes in combination with the anti-tyrosine hydroxylase (TH) antibody that labels commissural neurons of the nucleus of the tract of the POC (ntPOC)(Holzschuh et al., 2001)showed that robo1, robo2 and robo3 are expressed in the majority of ntPOC cells, but that robo4 is not expressed in cell bodies in the ntPOC region (data not shown).

To determine whether Robo receptors are expressed in cells of the diencephalic glial bridge, we triple labeled 28 hpf embryos with the anti-AT antibodies, anti-Gfap antibodies, and robo1, robo2, robo3 or robo4 in situ probes (Fig. 7). This triple labeling revealed that Gfap+ cells of the glial bridge appear to express robo1 and robo3, but not robo2 (Fig. 7A-L). Interestingly, robo4 expression overlaps with only a subset of Gfap+cells in the anterior portion of the glial bridge; those Gfap+ cells that come in direct contact with POC axons and RGC axons(Fig. 7M-P). robo4expression was not seen in Gfap+ cells ventral to the commissure(Fig. 7M). Thus, both glial cells and commissural neurons express different combinations of Robo receptors, and may therefore be capable of responding directly to Slit guidance signals.

We have characterized a population of midline-spanning glial cells (a glial bridge) that precedes the formation of the major commissures in the zebrafish forebrain, and have shown that Slit molecules play a central role in positioning both these glial cells and the commissures themselves. Our data suggest that Gfap and slit1a co-expressing cells form a midline bridge that provides a favorable substrate for POC and RGC axon growth(Fig. 8A). Based on Slit and Robo expression analysis and knockdown experiments, we propose that Slit2- and Slit3-mediated repulsion excludes Robo-expressing glia from Slit2/Slit3-expressing regions, restricting these cells to bands that span the midline and prefigure the commissures (Fig. 8A). Similarly, Slit2- and Slit3-mediated repulsion constrains growth of Robo-expressing commissural and RGC axons to the same Slit2/Slit3-negative region of the forebrain where this glial bridge is positioned. In contrast to Slit2 and Slit3, our data show that Slit1a is required for commissural axons to cross the midline, but does not play a major role in patterning the glial bridge. Sema3d does not appear to be necessary for commissure formation, as the POC forms in slit2/slit3MO-injected yot (gli2DR) mutants that lack sema3dexpression. We show that Hh signaling is needed to establish the proper expression of Slit genes (Fig. 8D,E), and for formation of the glial bridge(Fig. 8B). Finally, we show that cell-type-specific Robo receptor `codes' could differentially mediate the response of RGC axons, POC axons and glial cells to Slit signaling in the POC/chiasm region (Fig. 8C).

slit2 and slit3 expression surrounds the POC and AC (see also Hutson and Chien, 2002),and loss of Slit2 and/or Slit3 function results in the abnormal spreading of POC axons into anterior and posterior regions of the diencephalon (Figs 5, 6). Thus both Slit2 and Slit3 appear to constrain POC axon growth. The finding that Slit2 but not Slit3 knockdown also led to increased axon wandering(Table 1) suggests that Slit2 plays a more significant role in preventing growth outside the POC region. These Slit2/Slit3 loss-of-function data are consistent with the surround-repulsion model proposed for Robo/Slit retinal axon guidance(Hutson and Chien, 2002; Plump et al., 2002; Rasband et al., 2003; Richards, 2002b). Interestingly, although these POC `morphotypes' are relatively subtle in wild-type embryos, slit MO knockdowns in the context of Hh pathway mutants (which have altered Slit expression; Fig. 5, Fig. 8B) uncovered both overlapping and distinct functions for different Slits in the formation of the POC and the diencephalic glial bridge. The dramatic rescue of yot(gli2DR) commissure and glial bridge defects following slit2/slit3 knockdown was surprising given the general forebrain patterning defects associated with the loss of Hh signaling,including the reduction of the guidance molecule Sema3d at the midline(Fig. 5). The fact that slit2/slit3 MO injection can rescue commissure formation in yot (gli2DR) mutants to the same level seen in slit2/slit3 MO-injected wild-type embryos strongly suggests that the expansion of slit2/slit3 expression in yot(gli2DR) mutants is the major cause of POC (and most likely RGC) midline crossing errors (Fig. 8), and that Hh signaling and Sema3d are not directly required for axons to cross the midline.

Several lines of evidence show that Slit1a(Hutson et al., 2003) may function distinctly from Slit2/Slit3 during commissure formation in the zebrafish forebrain. First, slit1a expression is complementary to slit2/slit3 expression and unlike slit2/slit3, commissural axons grow in slit1a-expressing regions (Fig. 5). Second, expression of slit1a is distinctly affected by the loss of Hh signaling, with slit1a expression being reduced rather than expanded in yot (gli2DR) mutants(Fig. 5F). Third, blocking Slit1a function in wild-type embryos led to a marked reduction in POC crossing rather than the defasciculation seen in slit2/slit3morphants (Fig. 6). Fourth,reducing Slit1a function in yot (gli2DR) mutants failed to rescue axon crossing in yot (gli2DR) mutants and instead led to an even further decrease in POC axon crossing (Fig. 6). Interestingly, retinal axons in the optic tract also navigate across domains of slit1a expression, and knocking down Slit1a function causes retinal axon guidance errors that are not easily explained by removal of a repulsive signal (L.D.H. and C.-B.C., unpublished).

How might Slit1a influence commissural and retinal axon guidance? Based on our loss of function experiments, Slit1a does not appear to be a major repellent for commissural axons. One possibility is that slit1aexpression on a subset of Gfap+ cells in the glial bridge contributes to the permissive/attractive growth environment provided by these cells(Fig. 8A). If true, our slit1a MO experiments indicate that Slit1a is not the only guidance molecule needed for commissural axons to recognize these glial cells, as axons continue to follow aberrantly placed glial cells in slit1aMO-injected embryos (Fig. 6). An alternative model is that Slit1a acts as a repellent molecule for commissural axons, but is less repellent than Slit2/Slit3(Fig. 8A). Thus, when confronted with a choice between Slit1a-, Slit2- and Slit3-expressing domains,commissural axons choose to travel on the `path of least repulsion', growing on slit1a-expressing cells that border slit2/slit3-expressing cells(Fig. 8).

Our studies have uncovered a novel role for Slit molecules in the positioning of astroglia in the vertebrate forebrain(Fig. 8). We show that Gfap+cells are positioned in slit1a-expressing regions that do not express slit2 or slit3 (Fig. 5), and that loss of Slit2/Slit3 (but not Slit1a) function results in an expansion of the glial bridge in the POC region prior to axon growth(Fig. 6). In addition, glial cell defects in yot (gli2DR) mutants were largely rescued following slit2/slit3 MO knockdown, suggesting that the mis-positioning of glial cells in yot (gli2DR) mutants is due to the expanded zones of slit2/slit3 expression in these embryos(Figs 6, 8). Such a role for Slits in cell positioning has been shown for a variety of cell types, including glial cells and mesodermal cells (for reviews, see de Castro, 2003; Piper and Little, 2003). In the Drosophila CNS, Slit/Robo signaling is required to position glial cells that later serves to guide longitudinal and commissural axons (reviewed by Hidalgo, 2003). A role of Slits in cell migration has also been shown in mouse, where Slit1 is required cell autonomously for subventricular neuroblasts to migrate in vitro(Nguyen-Ba-Charvet et al.,2004). The fact that a subset of Robo receptor genes are expressed in zebrafish forebrain glial cells (Fig. 7, Fig. 8C)indicates that Robo-mediated Slit repulsion might directly prevent glial cells from occupying slit2- and slit3-expressing domains, perhaps by influencing cell migration in a way that is analogous to growth cone repulsion.

Our confocal analysis of Gfap expression during forebrain development revealed that two glial bridges span the midline prior to AC, POC and RGC formation, and that these Gfap-expressing cells are present at the right position to provide cues for midline axon growth. These astroglial cells send short fibers perpendicular to the pial surface into the forebrain, and the developing commissure forms as a ribbon along these fibers (see Movies 1, 2, 3 in the supplementary material) (Marcus and Easter, 1995). A previous immuno-EM study used the same Anti-Gfap antibodies to show that retinal and commissural axons directly contact Gfap+radial glia-like fibers (Marcus and Easter, 1995). These anatomical data strongly suggest that the glial bridge provides the growth substrate for the first commissural axons to cross the midline (Fig. 8A). Disruption of the glial bridge correlates with midline axon crossing errors,and glial cell defects precede axon defects both in yot (gli2DR)mutants (Figs 3, 4) and in slit2/slit3 morphants(Fig. 6D). These functional data suggest that these glial cells provide a positive or at least permissive cue for midline axon crossing (Fig. 8A,B). A positive role for glial cells is also supported by the observation that aberrantly growing axons appear to preferentially grow along mis-positioned Gfap+ cells rather than following their normal pathway or an alternative non-glial route. It will now be important to directly test whether glial cells are both necessary and sufficient for midline guidance of commissural and retinal axons.

shh is expressed at the right place and time to influence both cell fate and axon guidance at the midline in the diencephalon(Fig. 1E,F, Fig. 8D,E). Secreted Shh was recently shown to play a direct role in attracting axons to the ventral midline of the mouse neural tube (Charron et al., 2003). In addition, in vitro and in vivo experiments in chick suggest that Shh may inhibit RGC axon growth(Trousse et al., 2001). However, by using CyA to block Smoothened-mediated Hh signaling at different times, we show that Hh signaling does not appear to directly guide commissural or retinal axons in the zebrafish forebrain(Fig. 3). The rescue of POC formation in yot (gli2DR) mutants by injection of slit2/slit3 MOs (Fig. 6) also indicates that Hh signaling does not play a major role in the midline guidance of POC axons. Instead, Hh signaling appears to affect axon guidance indirectly through its role in patterning of the midline(Dale et al., 1997; Karlstrom et al., 2003; Mason and Sretavan, 1997),including the formation of the glial bridge and the regulation of axon guidance-molecule expression (Figs 3, 4, 5).

How can we reconcile our CyA results with the direct inhibitory affect seen for Shh in chick retinal explants? One possibility is that this particular guidance function for Shh has not been evolutionarily conserved between fish and chick. However, this seems unlikely given the high degree of conservation in other Hh-mediated developmental processes(Ingham and McMahon, 2001),and the similar expression of Hh in the chiasm region in chick and fish(Trousse et al., 2001)(Fig. 1E,F). Alternatively,loss-of-function approaches may be more informative than ectopic expression studies regarding the direct requirement for Hh in vivo. Thus, although retinal axons may be able to directly respond to ectopic Hh signals in vitro,direct Hh-mediated growth-cone guidance may not play a significant role in the context of other guidance cues present in the chiasm region. The finding that overexpression of shh in zebrafish not only disrupted AC, POC and RGC axon crossing but also disrupted glial bridge formation(Fig. 3G-I) is also consistent with a primary role for Hh signaling in patterning the axon growth substrate.

Rescue of POC formation in yot (gli2DR) mutants following slit2/slit3 MO injections, combined with the expansion of slit2/slit3 expression in Hh pathway mutants, suggests that the indirect effect of Hh on glial and axon guidance in the diencephalon is largely mediated by Slit proteins. Hh may regulate Slit gene expression directly, or through its regulation of cell fates. By contrast, loss of Hh or Slit function had little effect on AC formation, indicating that other patterning and guidance molecules are sufficient to pattern the ventral telencephalon and establish the AC. These data are consistent with the known role for Hh in ventral forebrain patterning, and begin to shed light on the downstream molecular targets that result in proper glial position and neural connectivity in this region.

Special thanks to Samuel Nona for the Gfap antibodies, Hitoshi Okamoto for morpholinos to slit2 and slit3, and the zebrafish community for in situ probes. We thank Jeanne Thomas for technical assistance and Brendan Delbos for fish care. Many thanks to Carol Mason, Stephen Devoto,Abbie Jensen and the members of the Karlstrom Laboratory for thoughtful critique. This work was supported in part by NS043872 (M.J.F.B.) and NIH NS39994 (R.O.K.).