Irx4-mediated regulation of Slit1 expression contributes to the definition of early axonal paths inside the retina

During development, axons navigate through 3D space in long, specific paths to establish neuronal connections. In visual system development, one of the early events is intra-retinal axon targeting. Extended from the ganglion cells(RGC) in the ganglion cell layer (GCL), the axons travel in a thin optic fiber layer (OFL) at the vitreal surface. Within OFL, all ganglion axons project towards the optic disc at the center of the retina, where they exit the eye(Fig. 1)(Halfter, 1985;Thanos and Mey, 2001). Within the retina, the newly extended immature axons join and leave fascicles,producing a type of `honeycomb' appearance. The axons soon mature into straight and fasciculated axon bundles.

Several molecules have been reported to be involved in the central projection of retinal axons toward the optic disc. Chondroitin sulfate proteoglycans, a major component of the ECM, are suggested to act as inhibitory molecules to prevent the growth of retinal axons towards the periphery (Brittis et al.,1992). In netrin 1/DCC or the EphB ligand-deficient mouse embryos,RGC axon pathfinding defects have also been observed near the optic disc(Birgbauer et al., 2000;Deiner et al., 1997). L1,receptor tyrosine phosphatase, have also been shown to be involved in axonal outgrowth or projection inside the retina(Brittis et al., 1995;Ledig et al., 1999;Snow et al., 1991). However,the molecular mechanisms involved in guiding the retinal axons specifically within the OFL are not fully understood.

Genetic, biochemical and molecular approaches have led to the identification of an increasing number of molecules as axon guidance cues(reviewed by Brose and Tessier-Lavigne,2000; Mueller,1999; Raper, 2000;Yu and Bargmann, 2001). Slit proteins are large secreted glycoproteins of ∼200 kDa, comprising four leucine rich repeats, seven to nine EGF repeats and a Laminin G-like module. Three distinct Slit genes, Slit1, Slit2 and Slit3, have been cloned in mammals (Brose et al.,1999; Holmes et al.,1998; Itoh et al.,1998; Li et al.,1999). In tissue culture, Slit1 and Slit2 proteins have been shown to function as chemorepellents and collapsing factors for olfactory, motor,hippocampal and retinal axons (Erskine et al., 2000; Li et al.,1999; Nguyen Ba-Charvet et al., 1999; Niclou et al.,2000; Plump et al.,2002; Ringstedt et al.,2000). Slit1 and Slit2 also repel tangentially migrating interneurons in the mouse telencephalon(Hu, 1999;Zhu et al., 1999). However,Slit can also act positively on the axons. Slit2 stimulates the formation of axon collateral branches of the sensory neurons(Wang et al., 1999). InDrosophila, migrating mesodermal cells can switch their response to Slit from repulsion to attraction (Kramer,2001). Robo proteins have been characterized as the receptors for Slits (Brose et al., 1999;Kidd et al., 1999). In mouse,Slit1, Slit2, Robo1 and Robo2 (from the Slit/Robo family)have been reported to be expressed in the ganglion cell layer (GCL) in the retina (Erskine et al., 2000;Niclou et al., 2000;Ringstedt et al., 2000). Slit1, Slit2 and Robo2 control the midline crossing of the axons at the optic chiasm (Hutson and Chien,2002; Plump et al.,2002). However, their roles in intra-retinal axon pathfinding remain unclear.

Interestingly, several members of the Irx homeobox gene family, includingIrx4, are also expressed in GCL in retina(Bruneau et al., 2000;Mummenhoff et al., 2001). Irx genes were identified based on the homology to the Drosophila Iroquois locus genes. The Irx genes belong to a family of evolutionarily conserved homeobox genes. There are three Iro genes in Drosophila,which form the Iroquois complex (Iro-C)(Gomez-Skarmeta et al., 1996;McNeill et al., 1997). Six Irx genes (Irx1-Irx6) have been identified in mouse and human,which are located in two genomic clusters of three genes each(Bosse et al., 2000;Bosse et al., 1997;Bruneau et al., 2000;Christoffels et al., 2000;Cohen et al., 2000;Peters et al., 2000). The Irx genes encode transcriptional controllers that contain a characteristic homeodomain and a sequence motif unique to the family, Iro box. The Irx genes function early in development to specify the identity of diverse territories of the body, such as in the head and mesothorax of Drosophila, the neural plate of Xenopus, vertebrate neural tube and heart (reviewed by Cavodeassi et al.,2001).

To study the molecular mechanisms of retinal axon guidance, we used the chicken system as an in vivo model system because of its experimental accessibility (Nakamoto,1996). Using both gain-of-function and loss-of-function approaches, we found that Irx4 specifically regulates Slit1expression to restrict it to a subset of cells in the ganglion cell layer. We have further analyzed Slit1 function in intra-retinal axon targeting. Our results suggest that Slit1 may act positively to guide retinal axons within the OFL. Irx4 regulation of Slit1 is important for the formation of the `honeycomb' appearance of the immature retinal axons within the retina.

Flat-mount and section in situ hybridization with the digoxigenin-labeled probes were carried out as previously described(Bao et al., 1999). For two-color in situ hybridization, one probe was labeled with fluorescein-12-UTP(Roche), while the other probe was labeled with digoxigenin-11-UTP (Roche) by in vitro transcription. The samples were hybridized with both probes simultaneously. The digoxigenin-labeled probe was detected first using an alkaline phosphatase-coupled anti-digoxigenin antibody and the NBT/BCIP substrate. The subsequent detection of the fluorescein-labeled probe was carried out by using an alkaline phosphatase-coupled anti-fluorescein antibody and developed with the Vector Red kit (Vector Laboratories).

RCAS-Irx4 and RCAS-DN-Irx4 viruses were prepared as previously described(Bao et al., 1999). Viral stocks at 5×10⁸ titer were injected into both optic vesicles at Hamburger-Hamilton (HH) stage10-11 (∼E1.5)(Hamburger and Hamilton,1992). Electroporation was also carried out on HH stage 10-11 chick embryos by using a square wave electroporator CUY-21 (Nepa Gene Company). DNA (0.7-1.3 mg/ml) mixed with the Fast Green dye (0.025%) was injected into the right optic vesicle. The positive electrode was placed next to the outer side of the right optic vesicle while the negative electrode was placed onto the forebrain/midbrain region of the embryo. Three pulses of 50 mseconds duration each at 15 V were applied.

Immunofluorescent staining was carried out on flat mounts of retina. Axons were stained with either monoclonal antibody 270.7 (provided by Dr Virginia Lee, University of Pennsylvania Medical School) or monoclonal antibody 3A10 in the `gene-axon assay'. Retinas were fixed in 4% paraformaldehyde, flat mounted and blocked in 10% calf serum DME with 0.2% Triton X-100. Primary and secondary antibodies were diluted in the block and incubated with the samples for 4 hours at room temperature or at 4°C overnight. Viral infection was confirmed by immunofluorescent staining with either the monoclonal antibody 3C2 (diluted 1:5) or the polyclonal antibody p27 (SPAFAS, Norwich, CT; diluted 1:10,000). Mouse monoclonal antibodies against Islet1 (39.4D5), neurofilament(3A10) and Myc (9E10) were purchased from the Developmental Studies Hybridoma Bank (University of Iowa, Iowa City, IA).

For analysis of the electroporated samples, the retinas were harvested at E7 or E7.5, and fixed in 4% paraformaldehyde. The Slit1-electroporated samples were co-immunostained with an anti-neurofilament antibody 270.7 and a rabbit polyclonal anti-Myc antibody followed by Cy3-conjugated donkey anti-mouse and FITC-conjugated donkey anti-rabbit secondary antibodies (Jackson ImmunoResearch Laboratories, West Grove, PN). Rabbit polyclonal anti-Myc antibody was obtained from Immunology Consultants Laboratory (Sherwood, OR)and used at 1:500 dilutions. The control GFP-electroporated samples were stained with antibody 270.7 followed by Cy3-conjugated anti-mouse secondary antibody. Images of anti-Myc and anti-neurofilament staining were taken from the same field and overlaid using Photoshop 6.0 software. To quantify the results, we used mouse anti-Myc antibody (9E10) instead of rabbit anti-Myc antibody for co-staining with the neurofilament. We could thus examine the transfected cells and axons within the same fluorescent channel. To score the percentage of the transfected cells that superimpose with the axon bundles, we selected transfected cells that were approximately in the same focal plane as the axons when viewed from the flat-mounts, and only in areas that had less than 50% axon coverage. Cells that appeared too out of focus were not included because they might be too deep to influence axon pathfinding.

Using in situ hybridization experiments on cryosections or flat-mounts of retina, we found that the chicken Irx4 gene is expressed in a subset of cells in GCL throughout retinal development(Fig. 2A,B), similar to the patterns reported in mouse retina (Bruneau et al., 2000; Mummenhoff et al., 2001). In addition, the Irx4-expressing cells appeared to be distributed evenly in GCL across the retina, not concentrated in particular sections of the retina (data not shown). Because a number of Slit/Robo family genes have also been reported to be expressed in GCL in mouse(Erskine et al., 2000;Niclou et al., 2000;Ringstedt et al., 2000), we examined whether similar patterns could be observed in the chicken retina. We found that Slit1 is expressed in a subset of cells in GCL, not in any other layers, at E4.5, E6 and E9 (Fig. 2C,D, data not shown). Similarly, Robo1 is also expressed in a subset of the cells in GCL (Fig. 2E). To compare the expression of Irx4 and Slit1in GCL further, we performed two-color in situ hybridization experiments. As shown in Fig. 2F, while some cells express neither gene and a small number of cells appear to have both colors, the majority of the GCL cells express only one of the two genes,either Irx4 or Slit1.

In vertebrates, the underlying mechanism by which the expression of the Slit/Robo family genes is regulated has not yet been reported. Our observation of similar expression patterns of Irx4 and Slit/Robo family genes in GCL gave us a basis to test whether there is a regulatory relationship between Irx4 and the Slit/Robo genes. We used a retroviral vector (RCAS) to express the full-length Irx4 protein,RCAS-Irx4 (Fig. 3A)(Bao et al., 1999). This viral construct can produce replication-competent viral particles that infect proliferating cells and express Irx4 protein in the infected cells. Both optic vesicles were injected at Hamburger-Hamilton (HH) stage 10-11 (∼E1.5)(Hamburger and Hamilton, 1992)with the viral stocks at appropriate titers to achieve incomplete infection(Schulte and Cepko, 2000). The incomplete infection would result in patches of cells infected surrounded by uninfected area, allowing direct comparison of the area with overexpression of Irx4 with the wild-type area. The infected retinas were harvested at embryonic day 8 (E8) and flat-mounted for the ease of analysis and observation(Fig. 1). After in situ hybridization with various RNA probes or immunofluorescent staining by cell type-specific antibodies, the samples were also immunostained with an antibody specific for viral antigen GAG to show the infected area. By comparing gene expression in the infected area immediately adjacent to the uninfected wild-type area, we were able to assess whether any of the genes was subject to the regulation by Irx4.

In the uninfected area, as in the wild-type control, a subset of GCL cells expressed Slit1 at a high level, scattered among cells that did not express Slit1 (Fig. 3B). However, the number of cells expressing Slit1 was significantly reduced in the area infected with RCAS-Irx4 virus(Fig. 3B,C). In the areas that had relatively large patches of infection, very few cells expressedSlit1. The control virus RCAS-GFP expressing a GFP protein did not change the expression of the Slit1 gene (data not shown), indicating that change of Slit1 expression was not merely due to viral infection. To examine whether Irx4 overexpression affected ganglion cell specification or differentiation, we analyzed all the GCL-specific markers available to us, including Brn3a, Brn3b(Liu et al., 2000), RA4(McLoon and Barnes, 1989),nicotinic acetylcholine receptor β 3(Hernandez et al., 1995), and islet 1 (Austin et al., 1995). The expression of these markers was not altered in the samples infected with the RCAS-Irx4 virus (data not shown, Fig. 3F,G). Within the Slit/Robo family genes, the expression ofSlit2, Robo1 and Robo2 was also unaffected by Irx4 overexpression (Fig. 3D,E, data not shown). These results demonstrate that Irx4 specifically downregulates the expression of Slit1 in GCL.

We next examined whether Irx4 function is required for the repression ofSlit1 expression in the retina. We used a dominant-negative Irx4 construct, RCAS-DN-Irx4, which encodes a fusion protein composed of the chicken Irx4 homeodomain and the repressor domain of the DrosophilaEngrailed protein (Fig. 3A)(Bao et al., 1999). Fusion of a DNA-binding domain, such as a homeodomain with the repressor domain of Engrailed, can create a protein that interferes with transcriptional activation by the wild-type protein(Badiani et al., 1994). The retinas were similarly infected by RCAS-DN-Irx4 and analyzed by in situ hybridization at E8. Although Slit1 was expressed in a subset of cells in the uninfected area, the infected area appeared to have substantial upregulation of Slit1 expression. The infected area appeared as round patches within which most if not all cells are expressing Slit1(Fig. 3H,I). Dominant-negative Irx4 did not have any effect on the expression of Slit2, Robo1 orRobo2 (data not shown). Control virus encoding only the Engrailed repressor domain did not have any effect on Slit1 expression(Fig. 3J,K). This indicates that disruption of Irx4 function can relieve the repression of Slit1expression in the cells.

The negative regulatory relationship between Irx4 andSlit1 is consistent with the two-color in situ hybridization result that Irx4 and Slit1 are mostly expressed in distinct cell populations in GCL (Fig. 2F). Although a small number of cells appear to have both colors, it is difficult to determine whether these cells express both genes, or the enzyme detecting the first probe was not completely inactivated. Two-color in situ hybridization is technically difficult especially at the resolution of single-cell level. Recently, we found that Slit1 expression decreases from E7 (data not shown). It is possible that Irx4 may be involved in downregulation of the expression of Slit1 after the axons have reached the optic disc. The overlapping expression in a small number of cells may be due to upregulation of Irx4 and downregulation ofSlit1 expression at later stages.

Although Slit1 is known to function in axon pathfinding(Brose et al., 1999;Kidd et al., 1999;Li et al., 1999;Nguyen Ba-Charvet et al.,1999; Wang et al.,1999), its role in intra-retinal axon guidance has not been reported. We analyzed the intra-retinal axon pathfinding in the samples with reduced Slit1 expression by infection with the RCAS-Irx4 virus. Infected with the RCAS-Irx4 virus or a control RCAS-GFP virus at HH stage 10-11 (E1.5) before the onset of ganglion cell differentiation at E4, the retinas were analyzed at E7, E8 or E12 on flat-mounts. Because retinal axons travel in the OFL, flat-mounts of retina allowed us to examine readily the trajectory of the axons from the vitreal side. Axons were immunostained with an anti-neurofilament antibody (270.7, red) and viral infection was stained using an anti-viral GAG antibody (P27, green).

E7 and E8 were chosen because different stages of axon projection and maturation could be viewed within one retina. There is a center-to-periphery gradient in terms of progression of cell differentiation and axon maturation,the center of the retina being relatively more advanced. As shown inFig. 4A, axons at the periphery of the E8 retina are relatively immature. They join and leave fascicles,producing a type of honeycomb appearance. At the median to center of the retina, however, the axons are more mature, more fasciculated and relatively straight (Fig. 4B). At E7,larger area of the peripheral retina appears immature, whereas at E12, axons in the entire retina appear fasciculated and straight.

In the control virus RCAS-GFP-infected retinas, the axons appeared completely normal in the infected area, suggesting that viral infection alone did not affect intra-retinal axon pathfinding(Fig. 4C,D). However, in the RCAS-Irx4 virus-injected retinas, the axons appeared to distribute unevenly and were overly fasciculated, although the axons still projected towards the optic disc (Fig. 4E-H;Fig. 5). Upon closer inspection, the areas with axonal abnormality were all within the regions that were infected with the RCAS-Irx4 virus. Close to the border of the infected/uninfected area, the axons appeared to avoid the infected area and turned in order to be in the uninfected area(Fig. 4E,F;Fig. 5A-D). This resulted in uneven distribution of the axons, such that the wild-type area had more and denser axon bundles than did the infected area. Because the infected area had a much lower level of Slit1 expression compared with the wild type area (Fig. 3B,C), the preference of the axons for the normal level of Slit1 in the wild-type area suggests that Slit1 is likely to act positively on the axons.

Although the axons at the infected/uninfected border appeared to turn to avoid the infected areas, the axons in the center of the infected area did not turn visibly (Fig. 4E-H;Fig. 5A-D), suggesting that the axons do not respond to Slit1 from a long range. However, many of the axons in the middle of the infected area appeared excessively fasciculated. This is especially evident when the axons projected perpendicularly through a uninfected/infected boundary (Fig. 4G,H). When the axons went from the uninfected area (UI) to an infected area (Ib), the overall direction of axon projection towards the optic disc was not affected. However, an increase in fasciculation occurred at the boundary, which caused an abrupt decrease of the number of axon bundles in the infected area. Because fasciculation is generally believed to be a response to repellents, the uninfected area with normal amount of Slit1 appeared more attractive or permissive, while the infected area with low amount of Slit1 was more repulsive or unpermissive to the axons. However, excessive fasciculation in infected area was also reversible. As the axons went from an infected area(Ia) to an uninfected area (UI), they returned to the appearance of the control with a significant increase in number of axon bundles, presumably by defasciculation (Fig. 4G,H). This suggests that the axonal abnormality is a response to the environment,not a permanent change within the axons.

There is a positive correlation between the infection and axonal abnormality. Most of the RCAS-Irx4-infected patches exhibited axonal phenotypes described above: avoidance of the infected area and excessive fasciculation. We excluded very small infected patches that contained less than 50 cells because we felt that the newly infected patches might not have enough time to downregulate Slit1 and cause a phenotype (indicated by an asterisk * in Fig. 5A-D). Of a total of 177 RCAS-Irx4-infected patches (from 38 retinas) from four independent experiments, 162 of them (91.2%) have the avoidance and fasciculation axonal phenotypes. By contrast, we never observed any axonal phenotype in control RCAS-GFP-infected retinas (0%, n=43 patches). In addition, the axons appeared to respond to the Slit1-low area only when they were passing through it. Once they had passed, they did not move out of the way as the area became Slit1-negative. At later stages (such as E12),the retinas were usually completely infected (data not shown). But the affected areas were still restricted, comparable with earlier stages(Fig. 5E). These results demonstrate that overexpression of Irx4 leads to abnormal retinal axon trajectories inside the retina.

As vertebrate Slits have been reported mostly as repellents in other assay systems (Erskine et al., 2000;Li et al., 1999;Niclou et al., 2000;Plump et al., 2002;Ringstedt et al., 2000) and our results with Irx4 overexpression suggest that Slit1 may act positively on the retinal axons within the retina, we set out to develop an in vivo assay for testing the effect of Slit1 inside the retina. Because Slit1 is too large to be expressed by a retroviral vector, we used electroporation to deliver an expression construct of Slit1 into the optic vesicles. Unlike retroviral infection, the transfected cells appeared as a group of scattered individual cells (Fig. 6D-G).

The rational behind the in vivo assay is described below. As shown inFig. 6A, the axons travel in the very thin optic fiber layer (OFL) above GCL. At E7, there are only two cellular layers in the retina: GCL and the deeper layer containing mainly undifferentiated cells. In flat-mounts, we were able to score the relative positions of the transfected cells in GCL with the axons in OFL. Other transfected cells located deeper would not be in the same focal plane as the axons when viewed from the vitreal side of the flat-mounts. Because they might be too far away to have any influence on the axons, we did not score these cells. If the protein acts positively, the axons would prefer to grow on top of the transfected cells in GCL, superimposed on the transfected cells(Fig. 6B). Conversely, if the protein acts negatively, the axons would avoid the transfected cells, thus would not be superimposed on the transfected cells(Fig. 6C). If the protein has no effect on the axons, such as a GFP protein, the axons would appear to distribute randomly relative to the positions of the transfected cells.

As a control, we electroporated an expression construct encoding GFP into the optic vesicles at HH stage 10-11. The retinas were harvested at E7 and the axons were stained with an anti-neurofilament antibody (270.7) followed by Cy3-conjugated anti-mouse secondary antibody. Some GFP-expressing cells were clearly visible from the vitreal side of the flat-mount retinas, approximately in the same focal plane of the axons, indicating they were the transfected cells in GCL (Fig. 6E,G). Other cells that were localized deeper and difficult to be seen in the focal plane of the axons were not scored. We also scored only in the area with less than 50% of axon coverage (median to periphery of the retina), because the axons were very dense at the vicinity of optic disc. In these control samples, the localization of the control GFP-transfected cells appeared random relative to the axon bundles (Fig. 6E,G). Forty-six percent of the control GFP-transfected cells (n=398, from 12 retinas) have axons growing on top of them.

To test the function of Slit1, we electroporated an expression construct encoding a Myc-tagged Slit1 (pCS2-Slit1myc)(Wu et al., 1999) into the optic vesicles at HH stage 10-11. The retinas were analyzed at E7. TheSlit1-transfected cells were identified by a rabbit anti-Myc antibody, while the axons were stained with a mouse anti-neurofilament antibody, 270.7 (Fig. 6D,F). Interestingly, most of the Slit1-transfected cells appeared to have axons passing on top of them (magenta circles,Fig. 6D,F). To quantify the data, we used a mouse anti-Myc antibody (9E10) to co-stain with the mouse anti-neurofilament antibody (270.7) so that we were able to compare the relative positions within one fluorescence channel. A total of 563 cells from 24 retinas were scored in the area that had less than 50% of axon coverage in three independent experiments. Slit1-transfected cells (91.1%) have axon bundles growing over them. Compared with the control GFP-transfected cells,the axons appeared to travel preferentially above theSlit1-expressing cells. These results suggest that Slit1 may act positively on the growth of the retinal axons.

To confirm that the axonal phenotype of Irx4 overexpression is due to decreased Slit1 expression, and not other molecule(s) regulated by Irx4, we tested whether Slit1 could rescue the phenotype. We co-electroporated RCAS-Irx4 construct with the Slit1 expression construct in ovo at HH stage 10-11. The retinas were analyzed at E8 by co-staining with anti-Myc (9E10,red), anti-GAG (P27, green) and anti-neurofilament (270.7, red) antibodies to show the Slit1-transfected cells, RCAS-Irx4-infected cells and axons,respectively. RCAS-Irx4 virus was made in the cells transfected with the RCAS-Irx4 construct, and rapidly spread inside the retina. However, the expression construct of Slit1 could not produce virus and Slit1 would only be expressed in the cells that were transfected. Therefore, the RCAS-Irx4-infected area was much larger than the area containing theSlit1-transfected cells. Slit1-transfected cells appeared as patches of scattered cells expressing Myc-tag (Fig. 7A-C).

In the area that had both RCAS-Irx4 infection and a large number ofSlit1-transfected cells, the axons appeared largely normal, evenly distributed and dense (Fig. 7A,B). However, in the area with RCAS-Irx4 infection but noSlit1-transfected cells, the axons were excessively fasciculated and unevenly distributed (marked with NR inFig. 7B), similar to the axons in the control experiments in which a GFP expression construct was co-electroporated with the RCAS-Irx4 construct(Fig. 7E). When only a small number of Slit1-transfected cells were present, the axonal phenotype was not fully corrected but did appear improved(Fig. 7C). Interestingly, most of the transfected cells had axons passing on top of them(Fig. 7C). From three independent experiments, 100% of the retinas (n=23) that were co-electroporated with the Slit1 expression construct showed improvement in axonal appearance: axons were more even and not as excessively fasciculated as in the retinas injected with the RCAS-Irx4 virus alone. No improvement in axonal phenotype was observed in the samples that were co-electroporated with the GFP construct (0%, n=11). These results demonstrate that Slit1 can rescue the axonal phenotypes caused by Irx4 overexpression.

To determine how the endogenous Slit1 influences the retinal axon trajectory, we developed another assay (`gene-axon assay') to visualize endogenous Slit1 gene expression simultaneously with axon trajectory. In situ hybridization was carried out on flat-mount retinas with theSlit1 probe, followed by immunofluorescent staining of axons by an anti-neurofilament antibody 3A10. 3A10 was selected because it was able to recognize the denatured neurofilament antigen after in situ hybridization procedure.

Under a fluorescent microscope, we were able to observe the axons that were fluorescently labeled as well as the dark in situ hybridization signals of theSlit1 gene (Fig. 8A-E). The staining of 3A10 appeared somewhat fragmented, possibly because the neurofilament antigen was partially digested by treatment of proteinase K during in situ hybridization procedures. As shown inFig. 8A-D, most of the growth cones of the elongating axons pointed directly toward the cells expressingSlit1 (arrowheads) and the axons traveled on top of theSlit1-expressing cells (Fig. 8A,B). These results support our electroporation data that the endogenous Slit1 appears to act positively on the retinal axons. Because several axons converge on one Slit1-positive cell, and then leave separately to go to the next Slit1-positive cells(Fig. 8A,B), the interaction of the growth cones and Slit1 appears to contribute to the production of the honeycomb appearance of the early retinal axons (see model,Fig. 8F). These results suggest that Slit1 provides intermediate targets for the retinal axons to travel through the optic fiber layer.

However, as the axons mature into relatively straight thick bundles,similar to those close to the center of the retina, they no longer align accurately with the Slit1-expressing cells(Fig. 8E). Some other factors may over-ride the influence of Slit1 by this stage. In addition, the gene-axon assay is also useful for distinguishing the cell type that is expressingSlit1 in GCL. Because of a lack of suitable molecular markers for the two types of cells in GCL: ganglion cells and displaced amacrine cells, it remains unclear which cell type is expressing Slit1. From the result of the gene-axon assay, the axons appear to extend from the cells that are not expressing Slit1 (marked with a `?' inFig. 8D), andSlit1-expressing cells do not have axons (marked with an asterisk inFig. 8C,D). This result suggests that Slit1 is likely to be expressed in the cells without axons, i.e. the displaced amacrine cells, but not in the ganglion cells.

In summary, we have shown that Slit1 plays an important role in intra-retinal axon targeting by providing intermediate targets to the retinal axons in OFL. We further demonstrate that Slit1 expression in GCL is regulated by a homeodomain protein, Irx4. Because multiple guidance cues and many axons are present in vivo, precise spatiotemporal regulation of these guidance cues and receptors ensures correct axon pathfinding in three-dimensional space. In Drosophila, it has been reported that transcription factors including Single-minded, Fish-hook, Drifter and Lola are involved in regulating the expression of the Slit gene in the midline(Crowner et al., 2002;Ma et al., 2000). However, in vertebrates, the mechanisms by which the expression of the Slit/Robo family genes is regulated have not yet been reported. We have shown that Irx4 participates in regulation of Slit1 expression in the retina by using both gain-of-function and loss-of-function approaches. However, our two-color hybridization results also indicate that additional proteins may be involved in repression of Slit1 expression in GCL, because there are cells that express neither Irx4 or Slit1. We now have evidence that other Irx proteins are involved in regulation of Slit1 gene expression in GCL (Z. J. and Z.-Z. B., unpublished). Therefore, the dominant-negative Irx4 construct probably acts as dominant-negative for other Irx proteins as well. Further study is required to understand the significance of the involvement of multiple Irx genes in regulation of Slit1expression.

Irx genes have been shown to function to subdivide territories into smaller domains. In our study, we have shown that Irx4 regulates gene expression in a subset of cells in the GCL instead of specifying domains. In the samples infected with either the RCAS-Irx4 or RCAS-DN-Irx4, no obvious changes in retinal cell differentiation have been detected. As the Engrailed repressor fusion construct acted as a dominant-negative of the full length Irx4, Irx4 is most probably a transcription activator. Regulation of Slit1expression by Irx4 is therefore indirect, mediated through another transcription repressor. It is also interesting to note that the area with decreased Slit1 expression by Irx4 overexpression matched well with the area of viral infection, indicating a relatively rapid response to Irx4 overexpression.

Most guidance cues are now known to be bi-functional: they can act positively as attractants for one navigational event and negatively as repellents for another (Mueller,1999; Song and Poo,1999; Tessier-Lavigne and Goodman, 1996). Recent findings indicate that this bi-functionality can be attributed to differential receptor activation(Hong et al., 1999) or different levels of second messengers in the neuronal cytoplasm, such as the levels of cytoplasmic cAMP/cGMP (Song and Poo, 2001). Many extracellular ligands, including neuromodulators,adhesion molecules and ECM components, can change the level of cyclic nucleotides and are thus capable of modulating axon navigation when they are present concurrently with the guidance cue(Hopker et al., 1999;Nguyen-Ba-Charvet et al.,2001). Interestingly, the same guidance cue can also serve as attractant and repellent for different parts (dendrite versus axon) of the same neuron (Polleux et al.,2000). Thus, the bi-functionality of guidance molecules reflects more the status of the neuron than an intrinsic property of the molecule.

Like most of the guidance cues, Slit can act as a repellent or as an attractant, although mostly as a repellent(Brose et al., 1999;Hu, 1999;Kidd et al., 1999;Kramer, 2001;Li et al., 1999;Nguyen Ba-Charvet et al.,1999; Wang et al.,1999; Wu et al.,1999; Yuan et al.,1999). It can also act both as a short-range and a long-range cue(Kidd et al., 1999;Rajagopalan et al., 2000;Simpson et al., 2000). Several lines of evidence in our study suggest that Slit1 acts positively on chicken retinal axons inside the retina. First, we have shown that the retinal axons`prefer' the area with normal amount of Slit1 to the area with low Slit1 that results from overexpression of Irx4. By rescue experiment, we have shown that Slit1 can correct the axonal phenotype that results from Irx4 overexpression,suggesting that the axonal phenotype is due to decreased Slit1expression. Second, we have shown that retinal axons preferentially travel above the cells that are expressing Slit1 by in vivo electroporation. Third,by gene-axon assay, we have shown that retinal axons project towards the cells that are expressing Slit1 endogenously and that axons align with theSlit1-expressing cells. These results support the idea that Slit1 functions as a positive factor to the retinal axons within the retina.

Because of the complex situation in vivo, we chose to use the word`positive' instead of `attractive'. In vivo, several axon guidance cues co-exist in 3D space. The direction of the axonal projection may be a response to the sum of all the guidance cues present in the environment. Although the distinction is often blurred, `positive' may be more a description of the axonal response than of the internal property of the axon guidance cues.`Positive' also includes both `permissive' and `attractive'(Dodd and Jessell, 1988;Goodman, 1996). Under the current assay systems, we cannot distinguish whether the actions of Slit1 are`permissive' or `attractive'.

Slit1 has previously been shown to act negatively on the retinal axons by gel culture assay (Plump et al.,2002). In addition, Slit1 has also been shown to act positively to increase dendritic growth and branching of the cortical cells(Whitford et al., 2002). Several factors may account for the differences among these results. One possibility is that different extracellular matrix proteins may be present in different assays or different biological systems. The composition of the extracellular matrix is not completely clear in the OFL of the retina. It is possible that these ECM factors may modify the function of Slit1. Extracellular matrix proteins such as laminin have been shown to modulate the function of axon guidance cues including netrin 1 and Slit2(Hopker et al., 1999;Nguyen-Ba-Charvet et al.,2001). In addition, the function of Slit1 may also be modulated by proteolytic cleavage. Slit2 is proteolytically cleaved into N-terminal and C-terminal fragments (Brose et al.,1999). In cultured DRG neurons, the N-terminal fragment (140 kDa)of Slit2 was found to promote axon branching and elongation, whereas the full-length Slit2 was inactive and may actually inhibit the activity of Slit2-N (Wang et al., 1999). Like Slit2, Slit1 is also post-translationally cleaved(Brose et al., 1999;Whitford et al., 2002;Yuan et al., 1999), although the effect of proteolytic cleavage on Slit1 function has not been reported. Interestingly, there is evidence that the Slit proteins may be processed differently in different cells (Brose et al., 1999; Whitford et al.,2002). Hence, this can be another factor contributing to the different effect of Slit1 to the axons under different assay conditions.

Our results also suggest that Slit1 mediates the retinal axon pathfinding in OFL and contributes to the generation of the `honey comb' appearance of the early retinal axons. Slit1 does not appear to direct central projection of the retinal axons towards the optic disc. We did not detect obvious center-to-periphery gradient of Slit1 expression. In addition, in the area that has low Slit1 expression, the overall direction of axon projection towards the optic disc was not affected. Slit1 appears to provide short-range attraction to mediate the navigation of retinal axons across the retina in the optic fiber layer. This may be similar to what has been described as the intermediate target or `stepping stone' phenomenon(Metin and Godement, 1996). The biological significance of the `honey comb' appearance of the early retinal axons is currently unknown. It is possible that the scattered,non-continuous expression of Slit1 is important to keep the attraction at a moderate level. If Slit1 is expressed in all the cells in GCL, instead of a subset of cells, the attraction may be too strong to allow the central projection of the retinal axons.

Using electron microscopy, it has been shown that the growth cones of elongating retinal axons are positioned immediately adjacent to the neuroepithelial endfeet (Halfter and Deiss, 1984). By in situ hybridization, we analyzed Slit1expression during the period of active axon outgrowth, including E4.5, E6 and E9 (Fig. 2 and data not shown). At all stages analyzed, Slit1 is consistently expressed in the ganglion cell layer (GCL), not in the inner nuclear layer (INL), where the cell bodies of the neuroepithelial or Müller glial cells reside. Therefore, Slit1 does not appear to be expressed in neuroepithelial or Müller glial cells, but in displaced amacrine cells, based on the results of the gene-axon assay. Whether displaced amacrine cells also have`endfeet-like' structures that might be observed in the electron microscopy with the growth cones is unclear. In addition, we also found that retinal axons at later stages did not align accurately with theSlit1-expressing cells. It is possible that at later stages, axon bundles are positioned immediately adjacent to neuroepithelial endfeet instead of the Slit1-expressing amacrine cells.

Interestingly, our results also suggest that the retinal axons seem to avoid areas in the retina with low Slit1 expression. Determining whether Slit2 provides the source for repulsion in these areas in the GCL, as Slit2expression is not affected by Irx4 overexpression, requires further study. It is possible that a balance between attractive and repulsive forces is required to ensure that the retinal axons travel through the optic fiber layer at the precise depth, without diving into the deeper layers of the retina or fraying towards the lens.

We thank Virginia Lee, Lin Gan, Marc Ballivet and Amy Chen for antibodies and probes; and Cliff Tabin and Neal Silverman for critical reading of the manuscript. C. L. C. is an Investigator of the Howard Hughes Medical Institute. This work was supported in part by a fellowship to Z.-Z. B. from the American Association of University Women, a grant from Charles H. Hood Foundation and a start-up fund from UMMS.