Robo1 regulates the development of major axon tracts and interneuron migration in the forebrain

The CNS midline plays an important role in guiding commissural projections. The chemorepulsive ligand Slit and its receptors of the Robo family are expressed in the developing and adult brain(Yuan et al., 1999; Marillat et al., 2002) and are crucially involved in the formation of midline commissures(Kidd et al., 1998a; Kidd et al., 1998b; Brose et al., 1999; Plump et al., 2002; Bagri et al., 2002; Long et al., 2004; Sabatier et al., 2004).

In the Drosophila nerve cord and in the mammalian spinal cord,Robo (Roundabout in Drosophila) protein expression is upregulated after midline crossing, when commissural growth cones become highly responsive to Slit, preventing them from re-crossing the midline(Kidd et al., 1998a; Kidd et al., 1998b; Zou et al., 2000; Long et al., 2004; Sabatier et al., 2004). However, brain commissural and decussating axons, including the corpus callosum, optic chiasm and the corticospinal tract express Robo protein(Sundaresan et al., 2004), and respond to Slit2 both before and after they cross the midline(Plump et al., 2002; Bagri et al., 2002; Shu et al., 2003a). Thus,Slit2 may serve a different role than that observed in flies or at the midline of the spinal cord, probably because brain commissural axons grow away from the midline after they cross it, rather than remaining in close proximity to the midline as spinal commissural axons do.

Slit2 and Slit1/2 double knockout animals display defects in corticothalamic and thalamocortical targeting, callosal and hippocampal commissure projections (Bagri et al., 2002), and defects in the formation of the optic chiasm(Plump et al., 2002). In these mice, large ectopic commissures are formed at the midline from corticothalamic axons that would not normally cross the midline. These data suggest that the Slits normally prevent these axons from crossing the midline and instead guide them to their respective targets in the thalamus.

In addition to regulating commissural axon guidance and axonal branching(Wang et al., 1999; Ozdinler and Erzurumlu, 2002; Sang et al., 2002), Slit/Robo signalling also regulates cellular migration(Hu, 1999; Wu et al., 1999; Zhu et al., 1999). Slit secreted from the ventricular zone of the lateral ganglionic eminence (LGE)repels cortical interneurons from the subventricular zone of LGE explants and inhibits tangential migration when added locally at the corticostriatal boundary of brain slices (Zhu et al.,1999). However, tangential migration was reported to take place normally in Slit1/Slit2 double knockout mice, while the ganglionic eminence (GE) retained its repulsive activity(Marín et al.,2003).

In order to examine Slit signalling and its involvement in axonal guidance and neuronal migration, we have generated Robo1 knockout mice by targeted deletion. Here, we analyse the migration of interneurons into the neocortex and the formation of the corpus callosum, hippocampal commissure,corticothalamic and thalamocortical projections. The results reveal striking differences between the phenotypes of Robo1 and Slit knockouts, and suggest that additional mechanisms are involved in Slit/Robo signalling in these systems.

A description of the generation of the Robo1 knockout mice can be found in the Results section and the legend to Fig. 1. Mice were genotyped by polymerase chain reaction (PCR) using the following primers: M1510F(5′-CGAGGARGAAARSTSATGATC-3′) and 216R(5′-CCACAAGACTTGTGACAATACC-3′). For reverse transcriptase(RT)-PCR, primers 1510F and 200R (5′-CCTCTCTTCCAAAAGATAGCTGG-3′)were employed. Total RNA was extracted from tissue using Trizol (Invitrogen,UK) and the RT reaction was performed with either oligo dT or random primers to specifically amplify between exons 4 and 7.

Whole mount in situ hybridization for Robo1 was performed on embryonic day(E)12.5 intact mouse embryos using a modified protocol(Wilkinson, 1992; Henrique et al., 1995) as described by Camurri et al. (Camurri et al., 2004). To assess protein expression in the Robo1 knockouts,gel-electrophoresis and immunoblotting were performed as previously described(Hivert et al., 2002). Robo1 was detected using a rabbit polyclonal antibody (205) raised against the C-terminal peptide of DUTT1/ROBO1 (Xian et al., 2001). To demonstrate equivocal loading, blots were re-probed using a mouse monoclonal antibody raised against β-actin (Jackson Laboratories, MN).

Imaging was performed using a General Electric Omega 400 (9.4 Tesla) NMR spectrometer. A custom-made solenoid volume coil was used as the radio frequency signal transmitter and receiver. Brains were placed in home-built,MR-compatible tubes filled with fomblin (Fomblin Profludropolyether, Ausimont,NJ) to prevent dehydration. Diffusion-weighted images were acquired with a 3D diffusion weighted multiple echo sequence(Mori and van Zijl, 1998). The imaging field of view was 11 × 7.5 × 7.5 mm³. The imaging matrix had a dimension of 128 × 70 × 72. The spectral data were apodized by a 10% trapezoidal function and then zero-filled to 256× 140 × 144. The pixel size after the zero-filling was 43 ×53.5 × 52.1 μm³. Eight diffusion-weighted images with different diffusion gradient direction and magnitude were acquired for each sample. A repetition time of 900 ms, an echo time of 37 ms and two signal averages were used for a total imaging time of ∼28-30 hours.

The diffusion tensor was calculated using a multivariate linear fitting method, and three pairs of eigenvalues and eigenvectors were calculated for each pixel. The eigenvector associated with the largest eigenvalue was referred to as the primary eigenvector. For the quantification of anisotropy,a linear measure (CL) was used (Westin et al., 2002). Using the primary eigenvector and CL, colour maps were calculated and the red, green and blue values of each pixel were defined by the orientation of its primary eigenvector with the intensity proportional to the CL. Red was assigned to the fibre orientation along the anteroposterior axis, green to the right-left axis and blue to the dorsoventral axis.

Brains were collected between E12.5 and E18.5. Embryos were either fixed by immersion in 4% paraformaldehyde (PFA) or transcardially perfused with saline,followed by 4% PFA, and then postfixed in the same fixative solution overnight. Brains were blocked in 3% agar and cut at 40 μm on a Vibratome(Leica) or 30 μm on a cryostat (Bright). Sections were washed in 1×phosphate buffered saline (PBS), blocked in a solution of 2% serum (v/v) and 0.2% Triton X-100 (v/v) (Sigma) in PBS for 2 hours. Normal goat serum (S-1000,Vector Laboratories, Burlingame, CA) or normal donkey serum (017-000-121,Jackson ImmunoResearch Laboratory, West Grove, PA) was used for primary antibodies made in rabbit or rat, respectively. Sections were incubated in either rabbit anti-GFAP (1:30,000; Z0334, Dako, Glostrup, Denmark); rabbit anti-neurofilament M C-terminal (1:75,000; AB1987, Chemicon, Temecula, CA);rat anti-L1 (1:5000; MAB5272, Chemicon); rabbit anti-calbindin (1:10000;D-28K, Swant, Bellinzona, Switzerland); mouse anti-GAD65 (1:200 Affinity Research Products, Exeter, UK); or rabbit anti-Robo1 or -Robo2 (1:10,000 and 1:5,000, respectively; antibodies prepared by Dr Murakami) overnight. Sections were washed in PBS and incubated in biotinylated goat anti-rabbit (1:500;Vector Laboratories) or biotinylated donkey anti-rat (1:500; Jackson ImmunoResearch Laboratory) for 2 hours, then incubated in avidin-biotin solution (1:500; Vector Laboratories) and processed as previously described(Shu et al., 2000).

Calbindin-positive cells were counted in 200 μm coronal strips of dorsomedial neocortex at different levels along the rostrocaudal extent of the cortex at E18.5 (eight sections at each level from each of three animals for each condition). In all counts, the experimenter did not know the condition of the animal. Strips were longitudinally divided into six equal bins/sectors,from bin 1 (ventricular zone) to bin 6 (marginal zone). Interneuron migration was assessed at E12.5 by counting the total number of calbindin-positive cells that had crossed the corticostriatal notch and entered the cortex.

Injections were made using fine-tipped glass pipettes (1-5 μl, Dummond Scientific Company, Broomall, PA) attached to a pressure injector(Picospritzer, Parker Instrumentation, NJ). Pipettes were filled with solutions of either 10% DiI or DiA (D-282 and D3883, Molecular Probes) in dimethylformamide (data shown in Fig. 5). Other brains (data shown in Fig. 7) were labelled by placing a single crystal of either DiI or DiA in the brain as previously described (Métin and Godement,1996; Molnár et al.,1998). DiA crystals were placed in the presumptive somatosensory cortex to label corticofugal axons, and DiI crystals were placed in the dorsal thalamus to label thalamocortical axons. To label the corpus callosum and the hippocampal commissure, injections were first made into the cingulate cortex(DiI) and then the brains were cut coronally at the level of the hippocampus to allow the injection of DiA directly into the dentate gyrus. Labelled brains were stored at 37°C in darkness for 2-6 weeks and then blocked in 4%agarose and cut at either 40 μm or 100 μm using a Vibratome (40 μm sections were cut on Leica Vibratome and 100 μm sections were cut on a Vibroslice, Campden Instruments). Injection sites were verified after sectioning by the presence of a fluorescent bolus and a pipette track. Sections were washed and incubated overnight with 4′-6-Diamidino-2-Phenyllindole (DAPI, 1:20,000; D-9542, Sigma) in PBS or bisbenzimide (10 minutes in 2.5 μg/ml solution in PBS, Sigma). Images were collected using a confocal microscope (Fluoview FV5000 Olympus, NY or Leica,Microsystems, UK). Sequential images collected with the Leica microscope were subsequently reconstructed using Metamorph imaging software (Universal Imaging Corporation).

Dissociated cell cultures were derived from E15 mouse telencephalons according to the method of Cavanagh et al.(Cavanagh et al., 1997). Briefly, GEs were dissected out from embryonic forebrains in Hanks' solution under a stereo microscope, and isolated tissue was dissociated enzymatically in Neurobasal media with trypsin (0.1%) and DNase I (0.001%) for 15 minutes at 37°C. Trypsin was inactivated by 10% foetal calf serum (FCS) in Neurobasal media for 5 minutes and cells dissociated by delicate tritiation with a sterile pipette tip. The resulting suspension was centrifuged at 1000 g for 3 minutes, the supernatant discarded and cells resuspended in Neurobasal media containing B27 Supplement, 100 μg/ml penicillin/streptomycin and 2 mM L-glutamine. They were then plated at a density of 2×10⁵ cells on poly-L-lysine (10 μg/ml) and laminin (5 μg/ml) coated 13 mm coverslips in 24-well plates. Cultures plated were kept in a humidified incubator (95% Air/5%CO₂) at 37°C and cells were allowed to attach to the coverslips for 30 minutes. Fresh medium was then added, and again on the following morning.

To evaluate the role of Robo1 in forebrain development, we first generated Robo1 knockout mice. In this procedure, we floxed one of a pair of exons coding for the immunoglobulin domain Ig 3a to generate a frame shift mutation. We expected a stop codon downstream and, consequently, nonsense mRNA message decay leading to a `null' phenotype (Li and Wilkinson, 1998). To achieve this, a targeting vector was generated using a 9.9 kb genomic fragment identified from a mouse bacterial artificial chromosome which encompasses exons 4 to 6 of Robo1(Fig. 1A). Exon 5 was floxed with a neomycin resistance gene (positive selection) cassette at the 3′end of exon 5, and a thymidine kinase cassette (for negative selection) was cloned into the vector arm. 129sv-derived ES were targeted and clones selected and analysed for homologous recombination by Southern blotting(Fig. 1B). Three independent clones demonstrating homologous recombination were propagated and then injected into C57Bl/6 derived blastocysts and chimeras generated. Two independently derived chimeras that had transmitted the recombinant allele were mated with Actin Cre mice (obtained from Dr D. Acampora, Kings College London) to generate founder mice where the exon5/neo cassette had been excised. These animals were bred onto a C57Bl/6 background and the resulting heterozygotes were crossed to generate homozygous deficient mice.

Southern blotting and PCR analysis of E14.5 pup DNA/RNA demonstrated that exon 5 and the neo cassette had been deleted(Fig. 1B-D), and sequence information showed that a frame shift had occurred in mice carrying the deleted allele (Fig. 1E). We predicted that the altered RNA species would undergo nonsense message decay(Li and Wilkinson, 1998), and homozygous mice would be negative for Robo1 message and, hence, protein. In situ hybridization studies on wild-type and homozygous mutant E12.5 embryos revealed that Robo1 mRNA expression was completely absent in the Robo1 knockout, whereas Robo2 and Rig1 expression remained high in the spinal cord(Fig. 1F). Similarly, western blot analysis on whole embryo (E14.5) extracts, which were probed with an antiserum against the C terminus of Robo1 (205)(Xian et al., 2001), showed that Robo1 expression was high in the wild-type specimen, reduced in the heterozygote and completely absent in the knockout(Fig. 1G). As a control,western blots were stripped and re-probed with a β-actin antibody to demonstrate equal loading. These results indicate that our Robo1knockout mice produce no Robo1 mRNA or protein and thus should be considered as complete null mutants.

All phenotypic analyses described here were performed on Robo1mutant mice backcrossed onto the C57Bl/6 background for 6 to 10 generations. Complete mortality of homozygous deficient mice was seen at birth in the C57Bl/6 line.

We have previously shown, using an antibody directed against both Robo1 and Robo2, that Robo receptors are expressed on callosal axons(Shu et al., 2003a). However,the generation of antibodies by F. Murakami that recognize Robo1 and Robo2 independently (Long et al.,2004; Sabatier et al.,2004) has demonstrated that Robo1 is expressed at high levels on callosal axons at E17 (Fig. 2A,C,E), whereas Robo2 is only faintly expressed in this region(Fig. 2B,D). In the same brain,Robo2 is expressed at high levels on other axonal tracts within the brain such as the nigrostriatal pathway, the optic tract and, to a lesser extent, on axons within the internal capsule (Fig. 2F).

Robo1 knockout brains from E17 and E18 embryos were first examined by DTMRI to identify large fibre tracts. This revealed gross abnormalities in the development of the corpus callosum and hippocampal commissure(Fig. 3–shown at E17). DTMRI colour maps demonstrate the diffusion anisotropy of water molecules in ordered structures (Zhang et al.,2003) and can be `colour-coded' to demonstrate axonal fibre tracts running in different orientations within the brain. In the horizontal plane,the presence of the corpus callosum (Fig. 3A, arrow in A′) and hippocampal commissure(Fig. 3A, arrowhead in A′) are shown in wild-type (n=3) and heterozygote(n=3) littermates. However, disruption in both of these tracts was evident in the knockout (n=2; Fig. 3G-I). A large reduction in the size of the corpus callosum and hippocampal commissure was observed in the mid-sagittal plane (compare green structures with a large arrow in Fig. 3C′ with Fig. 3I′). However, the anterior commissure was present in mice of each genotype (small arrow in Fig. 3C′,F′,I′). Coronal sections revealed that callosal axons in Robo1 knockout mice were blue at the midline(Fig. 3H, arrow in H′)rather than green (Fig. 3B,arrow in B′), indicating that axons were coursing in the dorsoventral plane rather than mediolaterally. This could indicate that when callosal axons reach the midline, they turn to grow ventrally rather crossing the midline.

To further analyse the phenotype, sections (control mice, n=5;knockouts, n=2) were stained with an antibody against L1-CAM that labels all axons of the corpus callosum and projections from the hippocampus including the fimbria, fornix and hippocampal commissure(Fig. 4). Callosal axons and fibres in the fornix formed large fasciculated bundles at the midline in the knockout (Fig. 4I-L, arrows). Although some axons still crossed the midline(Fig. 4L,L′, asterisks),others were clearly mis-routed into these large fascicles. Mis-projection of callosal axons in the Robo1 knockout brain occurred throughout the tract at both rostral (Fig. 4K,K′)and caudal (Fig. 4L,L′)levels. Thus, rather than forming classic Probst bundles, callosal axons and axons of the hippocampal commissure formed tight fascicles that projected ventrally. This suggests that loss of Robo1 results in the formation of segregated axonal fascicles rather than the normal single bundle of axons that make up the corpus callosum. Our immunohistochemical analysis could not discern whether these fascicles were derived primarily from the cortex or the hippocampus as L1-CAM labels both projections. To examine these projections independently, we injected DiI into the cortex (n=6 knockouts) to label the callosal axons and DiA into the hippocampus (n=2 knockouts)to label the fornix and hippocampal commissure(Fig. 5). We found that in both the callosal and hippocampal commissure projections, the axons bundled together in large clusters that ectopically projected into the septum (rather than forming a single large bundle that crossed the midline; Fig. 5). The axons of the corpus callosum and hippocampal commissure normally remain segregated at all rostrocaudal levels (n=7 controls; overlays in Fig. 5E,J,O) and do not mix together. However, in Robo1 knockout brains, these large clusters of axons from the callosal and hippocampal projections were mixed (red and green bundles in the overlay Fig. 5I,N). An alternative phenotype would have been for single axons to mix between the two projections, but this was not observed. Thus, Robo1 appears to be involved in maintaining the complete segregation of these two commissures. Finally, as seen by immunohistochemical analysis above, some axons still crossed the midline in the rostral region(Fig. 5B, arrow).

Previous work has described midline glial structures that guide callosal axons at the midline (Silver,1993; Shu and Richards,2001). In order to examine these structures, we labelled brain sections with glial fibrillary acidic protein. All three midline glial populations were present–the glial wedge, indusium griseum glia and the midline zipper glia (Shu et al.,2003b)–in both wild-type (n=5; Fig. 6A,B) and Robo1knockout brains (n=2; Fig. 6C,D). Although some disruption of the midline zipper glia was evident (compare Fig. 6B,D), it appeared that this was secondary to the formation of the large axonal fascicles described above.

In Slit2 knockout mice, corticothalamic and thalamocortical projections deviate within the internal capsule resulting in an ectopic commissure(Bagri et al., 2002). To investigate these projections in Robo1 knockout mice, DiI and DiA crystals were placed in the dorsal thalamus and cortex, respectively, of wild-type and knockout littermate brains (E12.5-18.5; Fig. 7). At E12.5, similar to what has been described in the rat(Molnár and Cordery,1999), we observed thalamocortical and corticothalamic fibres,bearing growth cones at their tips, growing out of their sites of origin and directed towards the region of the internal capsule. There was no apparent difference in the pattern or extent of labelling between mutant and Robo1 wild-type mice (n=3 for each condition, data not shown). However, at E14.5, although both thalamocortical and corticothalamic projections in knockout brains (n=4) followed paths comparable with those in wild type (n=4), they were further advanced. Thus, although axons of both systems in wild-type brains had not advanced past the lateral cortex(Fig. 7D,E), thalamocortical projections (DiI, red) in the knockouts were observed well into the cortex(compare Fig. 7A with 7D) where a few cortical plate cells had been back-labelled(Fig. 7B, arrows). Similarly,the corticothalamic projections in these animals (DiA, green) were already present in the thalamus (Fig. 7A). Furthermore, DiI-labelled axons crossing transversely over the thalamocortical fibre bundle were observed at the level of the internal capsule. This `knot' structure (Fig. 7C, arrow), observed in all knockout brains examined at this age,could be the result of a misplaced subgroup of thalamic axons or other misrouted axons such as retrogradely labelled optic tract axons. These data indicate that early in development, Robo1 plays a role in the timely projection of thalamocortical and corticofugal axons. Examination of E18.5 brains (n=4 for each condition) showed the advance of thalamocortical axons persisted in the Robo1 knockouts where the axons projected further medially into the cortex (Fig. 7G, arrow) compared with controls(Fig. 7K, arrow). In addition,back-labelled cells appeared in greater numbers in the thalamus of mutants following placement of dye in the cortex (compare Fig. 7I with 7M, arrows),further indicating that thalamocortical axons had arrived earlier in these brains. Furthermore, similar to DTMRI analysis, we observed corpus callosum axons deviating ventrally at the midline(Fig. 7F, arrow). In addition,similar to E14.5, DiI-labelled axons coursing transversely at the level of the internal capsule formed a `knot' structure in all but one of the Robo1 knockout brains examined (Fig. 7H, arrow), while they were absent in the controls(Fig. 7L, arrow).

Robo1 mRNA is localized in the developing neocortex and the proliferative zone of the GE (Marillat et al.,2002), suggesting that this receptor may regulate the movement of cortical interneurons from the ventral telencephalon. We examined Robo1 expression in coronal sections of E13.5 and E15.5 mouse brains, ages when interneuron migration is at its peak. Staining was identified in the mantle zone of the GE and along the marginal zone (MZ) and lower intermediate zone(IZ) of the cortex (Fig. 8A),well-defined routes for tangentially migrating interneurons(Fig. 8B)(Anderson et al., 1997; Parnavelas, 2000). The staining in these zones was rather dense and it was difficult to discern individual cell bodies from processes oriented parallel to the ventricular surface. At high magnification, the Robo1 receptor was clearly visualized in some individual cells, especially in the GE near the start of the migratory route (Fig. 8A′). These cells often displayed elongated somata and leading processes, features typical of migrating interneurons. Further analysis of dissociated GE cell cultures prepared from E15 animals showed Robo1 expression in GABAergic cells as identified by co-labelling with GAD65 (Fig. 8C-E).

A number of studies have reported that interneurons use corticofugal projections to migrate from the ventral telencephalon to the cortex(Parnavelas, 2000; Denaxa et al., 2001; Morante-Oria et al., 2003; McManus et al., 2004),although such an association has not been supported by the work of others(Tanaka et al., 2003). Thus,given the defects observed in the corticofugal projections, we wanted to elucidate the possible role of Robo1 in interneuron migration and distribution in the cortex. We examined brain sections (E12.5-E18.5) stained for calbindin,a marker of cortical interneurons. Although not all interneurons express calbindin (López-Bendito et al.,2004), this marker has been used routinely to label a large subpopulation of GABAergic interneurons migrating tangentially to the cortex(Anderson et al., 1997; Sussel et al., 1999). Our results clearly showed that the pattern of calbindin staining closely resembled that of Robo1 immunoreactivity(Fig. 8A,B). Analysis of wild-type E12.5 mice showed relatively few calbindin-positive cells predominantly in the lateral cortex (Fig. 9D) in agreement with earlier observations(Anderson et al., 1999; Anderson et al., 2001). However, in Robo1 knockout littermates, significantly more calbindin-positive cells were observed in the cortex (Fig. 9E). Cell counts at E12.5 (n=4 for each condition)revealed that almost twice as many calbindin-positive neurons had migrated into the cortices of Robo1 knockouts (220±12, P<0.001) compared with controls (98±10)(Fig. 9A,D,E). A comparable increase in the number of calbindin-positive cells was noted in the cortex of Robo1 knockouts at E15.5 (n=4 for each condition).

Analysis at E18.5 (n=3 for each condition) demonstrated an∼35% increase in the total number of calbindin-positive cells counted in a 200 μm wide strip of knockout dorsomedial cortex (DMC) compared with counts in a similar area in wild-type brains (Fig. 9B). Counts at different levels along the rostrocaudal axis showed that the increased number of calbindin-positive cells was evident in rostral(Fig. 9C,D,E) and middle(parietal) cortical areas (Fig. 9C,F,G), but not in the caudal (occipital) cortex(Fig. 9C). Furthermore, there was an abundance of calbindin-positive cells in the striatal region(Fig. 9G,G′, arrows), an area that is normally repulsive to migrating interneurons(Marín et al., 2001). The presence of calbindin labelled cells in the striatum was also evident in E15.5 knockout brains (data not shown). These observations indicate that Robo1 influences both the number of interneurons entering the cortex early and their migratory route through the ventral telencephalon.

Evidence presented here and elsewhere suggests that although the cellular processes involved in cell migration and axon guidance are fundamentally different, similar molecules may be involved. Here, we have found that Robo1 is indeed involved in both cell migration, and axon growth and guidance events. Furthermore, we found significant differences between the phenotype of Robo1 knockout mice and that of Slit2 or Slit1/2 double knockout mice. The largest differences are in the formation of the corticothalamic and thalamocortical pathfinding between the two types of mutants and in the migration of interneurons to the neocortex from the ventral forebrain. Slit2 or Slit1/2 double mutants display large ectopic commissures in the diencephalon that are made up of axons from the cortex which would normally project to the thalamus(Bagri et al., 2002). Thalamocortical axons in these mice display ectopic projections ventrally and very few enter the cortical plate. However, in Robo1 knockouts,thalamocortical and corticothalamic axons reach their targets at least one day earlier than in controls and no ectopic commissures are present in the diencephalon. Furthermore, previous data show no defect in interneuron migration into the cortex of Slit or Slit/Netrin (Ntn) mutants(Marín et al., 2003),whereas we see an increase, compared with controls, in the number of interneurons entering the cortex and reaching their targets early. These differences are significant and imply one of three possibilities for Slit signalling in this system: (1) both Robo1 and Robo2 are required for Slit signalling; (2) Robo1, together with an unknown receptor is required for Slit signalling, or (3) a completely novel Slit receptor is involved in these processes. An additional possibility is that Robo1 acts alone (possibly as a homodimer) in one of more of these systems.

The role of Slit2 in callosal formation has been demonstrated in vivo using both gene mutation and antisense knockdown of the protein(Bagri et al., 2002; Shu et al., 2003a). However,the Robo receptor involved has been unclear. Our data show that Robo1 is required for callosal formation, but some subtle differences also exist between the Robo1 and Slit2 callosal phenotypes. For example, when Slit2 is removed, axons tend to defasciculate (Bagri et al., 2002; Plump et al.,2002; Shu et al.,2003a), whereas when Robo1 is absent, the axons form tight axon clusters or bundles that project ectopically en masse (data presented here). Our data also suggest that Robo1 might be involved in maintaining a crucial distance between callosal and hippocampal commissure axons. Our finding that axon bundles from each commissure were mixed together at the midline, rather than remaining segregated, suggests the involvement of Robo1 in normally maintaining the separation of these commissures into two distinct fibre tracts, perhaps either through Robo/Slit signalling or through Robo1 homophilic interactions.

Our results demonstrate that corticothalamic and thalamocortical projections reach their targets prematurely when compared with their wild-type littermates early in development. This may be interpreted in one of two ways:either Robo1 normally acts as a growth retardant for these axons in a Slit-independent manner; or the interactions between Robo1 and either Slit1 expressed in the cortical plate (Whitford et al., 2002) or Slit1 and Slit2 expressed in the thalamus(Marillat et al., 2002; Bagri et al., 2002), prevent these axons from entering their respective targets until later. Slit2and Slit1/2 double mutants display major pathfinding defects in the formation of these projections(Bagri et al., 2002). However,the pathfinding defects were not observed in the Robo1 knockouts (other than overshooting their targets at each age as described). Hence, either these effects are mediated by a different receptor or by a different combination of receptors. The finding that Robo2 is not highly expressed on callosal axons makes it unlikely that elimination of both Robo1 and Robo2 simultaneously might be required to observe similar defects as those described in Slit mutants. Regardless, we and others (Long et al., 2004) have been unable to generate Robo1 and Robo2 double knockout mice by breeding owing to the proximity of these two genes on the same chromosome. We found that Robo1 is expressed in axons within the IZ at a time when both the corticofugal and thalamocortical axonal systems develop. By E18.5, when these axons have reached their final targets, Robo1 labelling is downregulated in these axons. Thus, the phenotype observed in the mutants might be directly mediated by Robo1 on the developing axons in the cortex. In addition, Robo1 is expressed within the GE and could,therefore, mediate guidance events on both thalamic and cortical projections within this region.

Our results show a significant increase in the number of interneurons that enter the cerebral cortex from the ventral forebrain in Robo1-null mice throughout the period of corticogenesis. Furthermore, we found that these neurons migrated through the striatum and were not repelled by it. Previous studies have shown that the striatum expresses Sema3a and Sema3f(Marín et al., 2001),and GABAergic interneurons expressing neruopilin 1 (Npn1) are repelled away from the striatum into the cortex(Marín et al., 2001; Morante-Oria et al., 2003). Slits have also been shown to repel GABAergic interneurons in vitro(Zhu et al., 1999). However,in vivo, Slit1/2 double knockout mice and Slit1/2 and Ntn1 triple knockout mice do not display defects in cortical interneuron migration(Marín et al., 2003). It is difficult to reconcile these data with our findings and explain why in the Robo1 knockout animals, interneurons migrate through the repulsive striatal region and enter the cortex earlier and in larger numbers than normal.

The data suggest that there may be two different developmental events occurring. First, that Robo1 is required to repel the cells around the striatum, which may be a Slit-dependent or independent event. Neurons still avoid the striatal area in Slit1/2 double mutant mice(Marín et al., 2003),indicating that this might be a Slit-independent event. However, presumably these neurons still express Npn1 and the striatal region would still express Sema3a and Sema3f. Thus, perhaps, both Npn1/Sema and Robo1 signalling are required to steer the cells around the striatum. The second abnormal developmental event we observed is that more interneurons migrate into the cortex. The most plausible explanation for this is that these cells follow the cortical and thalamic axons within the internal capsule that enter their respective targets earlier than normal in Robo1 knockout mice. A close association with the TAG-1 expressing axonal bundles of the developing corticofugal fibre system has been observed, prompting speculation that interneurons use this system as a scaffold for their migration into the neocortex (Denaxa et al., 2001; Morante-Oria et al., 2003; McManus et al., 2004). Considering that Robo1 is expressed in the GE, where interneurons originate from, and in the IZ, a zone that contains the majority of migrating interneurons to the cortex, it is possible that their trajectory is both directly delineated by Robo1 and indirectly regulated through fasciculation and migration along cortical and thalamic axons.

Why was the increase in interneuron number restricted to the rostral and middle cortical areas of the Robo1 knockout mice? Recent studies by Yozu et al. (Yozu et al.,2005) have clearly shown that the sources and mechanisms of migration of interneurons that populate these areas are different from those destined for caudal cortical regions. Specifically, these investigators have demonstrated that interneurons destined for the caudal cortex and hippocampus arise in the caudal GE and use a novel migratory path, the so-called caudal migratory stream, whereas those that populate rostral and middle areas arise predominantly from the MGE/LGE. Finally, we cannot exclude the possibility that the increase in interneuron numbers in the rostral and middle cortical areas is due to a differential increase in proliferation in different parts of the GE.

In summary, we found that Robo1 is required for the correct formation of the corpus callosum and the hippocampal commissure, as well as the timely projection of thalamocortical and corticofugal axons. In addition, the absence of Robo1 led to premature migration of interneurons to the cortex. The differences between the Robo1 knockout phenotype and the phenotypes of Slit-deficient mice suggest that additional components contribute to Robo/Slit signal transduction mechanism.

The authors gratefully acknowledge the contribution of Melissa Barber for Robo1 labelling. V.S. was a recipient of an MRC (UK) Career Establishment Award. The work in J.P.'s laboratory was supported by a Wellcome Trust Grant(074549). Work in L.J.R.'s laboratory was supported by a grant (FY05-785) from The March of Dimes Foundation for Birth Defects. Work in S.M.'s laboratory was supported by NIH grant EB003543-01A.