The winged helix transcription factor Foxg1 facilitates retinal ganglion cell axon crossing of the ventral midline in the mouse

During brain development, neurons project axons tipped with navigating growth cones. Ultimately, these axons reach and form synaptic connections with their target cells. Axon tracts connecting brain regions can be divided into two classes depending on whether they cross the midline. Ipsilateral (or longitudinal) tracts, such as the thalamocortical tract, connect regions on the same side of the brain. Contralateral (or commissural) tracts, such as the corpus callosum, cross the midline of the brain and connect regions in different halves. The visual system exhibits both types of tract. Each optic nerve carries RGC axons from the retina towards the optic chiasm on the ventral surface of the hypothalamus. At the optic chiasm some RGC axons progress contralaterally and others progress ipsilaterally to join the optic tract and grow towards their targets in the forebrain and midbrain. This way of organising retinal output results in parts of the brain receiving binocular input, which is useful for judging positions of objects in space.

The behaviour of a growth cone at a specific point along its route is defined by its interaction with the cells and molecules it encounters. In the mouse, RGC axons exit the retina and start to navigate along the optic stalk(forming the optic nerve) at embryonic day 11 (E11)(Colello and Guillery, 1990). By E15.5, the chiasm has acquired its mature configuration although progressively more axons navigate this route during subsequent development(Marcus et al., 1995). At the optic chiasm RGC growth cones respond to cues generated by a specialised population of midline cells (Wizenmann et al., 1993; Sretavan and Reichardt, 1993; Marcus et al., 1995). Their importance is shown by the failure of optic tract formation in mice with an immunoablated chiasm(Sretavan et al., 1995) and the development of only ipsilateral projections in Pax2^-/-or Vax1^-/- mice with agenesis of the chiasm(Torres et al., 1996; Hallonet et al., 1999). Contralaterally projecting RGCs (the majority in mice) are distributed all over the retinal surface. Ipsilaterally projecting RGCs are concentrated in ventrotemporal retina and their axons are repelled at the optic chiasm(Sretavan and Reichardt, 1993; Herrera et al., 2003; Williams et al., 2003).

Several transcription factors are regionally expressed in the developing visual system and are poised to regulate the development of its structures and axonal connections (Torres et al.,1996; Halonet et al., 1999; Barbieri et al., 2002; Mui et al., 2002; Wang et al., 2002; de Melo et al., 2003; Herrera et al., 2003). The winged helix transcription factor Foxg1 is strongly expressed in the developing nasal retina and optic stalk, optic chiasm, telencephalon and superior colliculus (Xuan et al.,1995; Huh et al.,1999; Marcus et al.,1999). The most obvious anatomical defects in embryos homozygous for a targeted deletion of Foxg1 are abnormally shaped eyes and a hypoplastic telencephalon (Xuan et al.,1995; Huh et al.,1999). We have shown by injecting tract-tracers into the eyes of these mutants that the retina projects axons to an optic tract growing over the surface of the thalamus (Pratt et al.,2002).

To test the potential for Foxg1 to influence axon navigation, we used carbocyanine dyes to trace retinal axons in mice carrying a targeted deletion of Foxg1 (Xuan et al.,1995; Hebert and McConnell,2000). We found that mice lacking Foxg1 have an increased ipsilateral projection arising from both nasal and temporal retina. To characterise this defect, we (1) used lacZ reporter transgenes to identify potential sites at which Foxg1 expression might influence the navigation of retinofugal axons; and (2) examined several molecular markers expressed in the normal retina and optic chiasm, namely the transcription factors Foxg1 and Nkx2.2, the receptor tyrosine kinase Ephb2 and its ligand ephrin B2, and the stage-specific embryonic antigen 1 (SSEA-1; Fut4 - Mouse Genome Informatics) (Marcus et al.,1999; Barbieri et al.,2002; Williams et al.,2003) in Foxg1^-/- embryos.

To identify cells in which the Foxg1 locus is transcriptionally active, we exploited the Foxg1^tm1M (or Foxg1^lacZ) allele in which coding sequences are replaced by a lacZ cassette (Xuan et al.,1995). For tract tracing, we used the Foxg1^tm1(cre)Skm (or Foxg1^Cre) allele in which coding sequences are replaced by a Cre recombinase cassette(Hebert and McConnell, 2000). Foxg1^Cre and Foxg1^lacZ are predicted null alleles (designated Foxg1^-)(Xuan et al., 1995; Hebert and McConnell, 2000). For lineage tracing, we used Gtrosa26^tm1Sho(R26RS) reporter mice (Mao et al., 1999). Cre-mediated recombination in Foxg1^Cre;R26RS embryos irreversibly activates lacZ expression from the ROSA26 locus(Mao et al., 1999; Hebert and McConnell, 2000). Tract tracing and lineage tracing experiments used the Foxg1^Cre allele on an albino Swiss Webster background (as described by Hebert and McConnell,2000). Other experiments used Foxg1^lacZ and/or Foxg1^Cre alleles on a mixed pigmented CBA/C57Bl6/Swiss Webster background. Foxg1 homozygous mutants(Foxg1^Cre/Cre) and compound heterozygotes(Foxg1^Cre/lacZ) were identified by their hypoplastic telencephalon and distorted eyes. Heterozygous embryos (i.e. Foxg1^+/-) were identified by PCR genotyping: they were indistinguishable morphologically from wild type(Xuan et al., 1995; Huh et al., 1999) and we detected no RGC axon projection defects at the optic chiasm (data not shown).

Foxg1⁺ was detected with primers Foxg1ORFFor 5′-CTG ACG CTC AAT GGC ATC TA-3′ and Foxg1ORFRev 5′-TTT GAG TCA ACA CGG AGC TG-3′ which give a product of 438 bp; Foxg1^lacZ allele was detected with primers Foxg1_5′UTRF 5′-GCT GGA CAT GGG AGA TAG GA-3′ and Foxg1lacZ_ORFR 5′-GAC AGT ATC GGC CTC AGG AA-3′ which give a 550 bp product; Foxg1^Cre allele was detected with primers NLSCreF 5′-CAT TTG GGC CAG CTA AAC AT-3′; NLSCreR 5′-ATT CTC CCA CCG TCA GTAC G-3′, which give a 307 bp product. For all primers, cycling conditions were 96°C for 2 minutes followed by[96°C for 30 seconds, 58.5°C for 30 seconds and 72°C for 30 seconds] for 35 cycles.

E14.5 Foxg1^lacZ/+ and Foxg1^lacZ/Creembryonic heads were dissected and fixed for 1 hour at 4°C in 4%paraformaldehyde, 0.02% NP40, 0.01% sodium deoxycholate, 5 mM EGTA, 2 mM MgCl₂ in phosphate-buffered saline (PBS). In some cases, heads were equilibrated in 30% sucrose/PBS and sectioned (10 μm) on a cryostat. Tissues were rinsed several times in wash buffer (2 mM MgCl₂, 0.02%NP40, 0.01% sodium deoxycholate in PBS), transferred to staining solution(wash buffer supplemented with 5 mM potassium ferricyanide, 5 mM potassium ferrocyanide and 1 mg/ml X-gal), and stained overnight (cryostat sections on slides) or for 2 days with agitation (wholemounts) at 37°C in darkness. Staining was stopped with 20 mM EDTA in PBS. Some wholemounts were sectioned with a vibratome (200 μm) or processed to wax, sectioned (10 μm) and mounted. Cryostat sections were counterstained with Nuclear Fast Red.

Embryos (E13.5) or heads (E15.5) were fixed at 4°C in 4%paraformaldehyde in PBS overnight. For retrograde labelling, the back of the head and tissues overlying the thalamus were removed, and small DiI or DiA crystals (Molecular Probes) were placed in a line over the dorsal thalamus on one side to ensure unilateral labelling of the optic tract. For anterograde labelling, the lens was removed and the optic cup packed with larger clumps of crystals. Embryos were returned to 4% paraformaldehyde in PBS in the dark at room temperature for about one month to allow tracers to diffuse along axons. Vibratome sections (200 μm) were cleared by sinking in 1:1 glycerol: PBS containing the nuclear counterstain TOPRO3 (0.2 μM, Molecular Probes) and then 9:1 glycerol:PBS. Sections were stored at 4°C. Images were acquired using an epifluorescence microscope and digital camera (Leica Microsystems,Germany) or a TCS NT confocal microscope (Leica Microsystems, Germany). For quantification DiI-labelled RGCs were assigned to either nasal or temporal retina (Fig. 1B,C,J,K) in serial horizontal sections (for examples of sections used, see Fig. 4). In epifluorescence(TRITC filter) images (Fig. 3A-J; Fig. 4D-F,K,L) DiI appears orange and TOPRO3 appears red and in confocal images (Fig. 3K,L; Fig. 4A-C,G-J,M-O) DiI appears red, DiA is green and TOPRO3 appears blue.

Ephb2 and ephrin B2 immunohistochemistry was performed as described(Batlle et al., 2002). Embryos(E13.5) or heads (E16.5) were fixed overnight in 4% paraformaldehyde in phosphate buffer high salt (PBHS) [0.04 M phosphate, 0.03 M NaCl (pH7.0)] at 4°C, processed to wax and sectioned (10 μm). After rehydration,endogenous peroxidase activity was blocked with 90% methanol:3% hydrogen peroxide. Antigens were unmasked by microwaving in 10 mM sodium citrate buffer(pH 6.0) before reacting overnight with primary antibody followed by a bridge rabbit anti-goat antibody and detection using the Envision⁺(Rabbit) Kit (Dako K4010). Antibody incubation and pre-incubation blocking were carried out in 1% bovine serum albumin in PBS at room temperature. SSEA-1 and Nkx2.2 immunohistochemistry was performed as described [SSEA-1(Marcus and Mason, 1995),Nkx2.2 (Pratt et al., 2000)]except that detection of the primary antibody was carried out using the Envision⁺ (Mouse) Kit (Dako K4006). Primary antibodies were goat anti-mouse ephrin B2 (R&D Systems AF496) used at 1:500; goat anti-mouse Ephb2 (R&D Systems AF467) used at 1:1000; rabbit anti-goat (Dako Z0228)used at 1:400; mouse monoclonal anti-Nkx2.2 antibody (DSHB, USA 74.5A5) used at 1:30; and mouse monoclonal anti-SSEA1 antibody (DSHB, USA MC-480) was used at 1:50. Following a diaminobenzidine colour reaction, sections were dehydrated and mounted.

To identify cells that express Foxg1, we took advantage of a`knock-in' to the Foxg1 locus in which a lacZ cassette replaces Foxg1-coding sequences and is expressed in cells transcriptionally activating the Foxg1 locus. These cells are stained blue after incubation with X-gal solution(Fig. 1). If these cells possess a functional Foxg1 allele (i.e. in Foxg1^lacZ/+ embryos) they can make Foxg1 protein.

We examined Foxg1 expression in Foxg1^lacZ/+embryos at E14.5, because this is the time at which RGC axons are navigating the midline. Foxg1 expression was strongest in the nasal retina(Fig. 1B,C); 51% of nasal cells were lacZ⁺ in counts across several sections. By contrast,only a few temporal cells were lacZ⁺ (2%) (visible at higher magnifications in Fig. 1D). These findings are consistent with previous results of in situ hybridisation(Hatini et al., 1994). In the centronasal retina, staining was particularly prominent in the inner layer occupied by the RGCs (Fig. 1E). Foxg1 was expressed in the anterior hypothalamus(Fig. 1F) with a population of expressing cells at the ventral midline overlying bundled axons at the optic chiasm (Fig. 1G,H). Foxg1 was also expressed by other structures encountered (or avoided) by RGC axons,including the telencephalon, dorsal midline and superior colliculus(Fig. 1A).

We examined lacZ expression in Foxg1^lacZ/Crecompound heterozygotes, which can produce no functional Foxg1 protein, so as to identify cells in which the mutant gene was transcriptionally activated at E14.5. As Foxg1^lacZ/Cre cells had the same lacZgene dose as Foxg1^lacZ/+ cells (one copy), lacZstaining intensity could be directly compared between the two genotypes. The Foxg1^-/- eye is distorted(Fig. 1J,K) with an abnormally small lens and a medially elongated retina that extends towards the optic chiasm. The nasal bias of Foxg1 expression in Foxg1^lacZ/+ embryos described above was also apparent in Foxg1^lacZ/Cre embryos (see whole mount in Fig. 1N and sectioned material in Fig. 1J-M). We counted 21%lacZ⁺ cells in nasal retina compared with only 3%lacZ⁺ cells in temporal retina. The reduced count nasally reflects the absence of lacZ⁺ cells in outer layers of the Foxg1^lacZ/Cre lateral retina (compare Fig. 1B,C with J,K). Previous work at a younger age, E12.5, also showed lacZ expression in the eyes of Foxg1^lacZ/lacZ embryos confined mainly to the nasal region of the mutant retina (Huh et al.,1999). As in Foxg1^lacZ/+ embryos, lacZ staining was prominent in the inner layer of the nasal retina(compare Fig. 1J-M with B-E). These data show that the nasotemporal polarity of the retina is maintained in Foxg1^-/- mutants until at least E14.5; this conclusion is further supported by data presented later.

In E14.5 Foxg1^lacZ/Cre compound heterozygotes, the anterior hypothalamus continued to express Foxg1 around the ventral midline at the point contacted by the expanded mutant retina(Fig. 1N). As in Foxg1^lacZ/+ embryos, stained cells were detected at the ventral midline (Fig. 1O,P)indicating that Foxg1 is not required for the maintenance of this population of cells at the optic chiasm. lacZ staining was also retained in the telencephalon, along the dorsal midline and in the superior colliculus(Fig. 1I).

The experiments described above show that Foxg1 is expressed by many nasal RGCs at the time their growth cones are navigating the chiasm. The few expressing cells we detected in the temporal retina were near to the lateral lip (or ciliary margin) of the temporal retina where progenitor cells proliferate (Perron et al.,1998; Zuber et al.,2003). It was therefore possible that Foxg1-expressing progenitors might populate the temporal retina with RGCs, the navigational responses of which were influenced by transient expression of Foxg1 earlier in their lineage. Crossing Foxg1^Cre/+ mice with R26RSreporter mice generates embryos in which a Cre recombinase-mediated recombination event irreversibly enables lacZ expression from the Rosa26 locus in cells that express Foxg1(Mao et al., 1999; Hebert and McConnell, 2000). These cells and their descendants express lacZ protein, regardless of whether the Foxg1 locus remains active. The lacZ expression pattern provides a convenient means of visualising cells whose fate may have been influenced by autonomous exposure to Foxg1.

As expected, the nasal retina of Foxg1^Cre/+; R26RSembryos exhibited uniform lacZ staining(Fig. 2A-D). Strong staining was also seen in lens, optic nerve and at the optic chiasm. In the temporal retina of Foxg1^Cre/+; R26RS embryos, lacZ stained cells were found in radial stripes most of which were one or two cells wide while others were wider (Fig. 2B-D; arrows indicate radial stripes in temporal retina). A similar stripy distribution has been described before in chimeras and arises because the retina is constructed from radially oriented clones of cells(Reese et al., 1999). Examination of serial sections though the eyes of six embryos did not reveal any clear pattern in the size or density of these radial stripes, implying that the exposure of their ancestors to Foxg1 did not predispose them to occupy any particular part of the temporal retina. It was also apparent that most temporal RGCs were located outside the lacZ-expressing clones. These lacZ-expressing clones included cells outside the RGC layer showing that ancestral Foxg1 expression was not restricted to RGCs. We also detected mosaic activation of lacZ expression in the temporal retinal pigment epithelium (a derivative of the optic cup, as is the retina)suggesting that a subpopulation of these cells are derived from Foxg1-expressing cells.

In conclusion, nasal RGCs express Foxg1 at the time their axons are negotiating the optic chiasm. By contrast, although some temporal RGCs are descended from Foxg1-expressing precursors, the vast majority do not express Foxg1 at this time and have never expressed Foxg1 at any time in their lineage.

The polarisation of Foxg1 expression in the retina suggests that it could play an autonomous role in the navigation of nasal RGC growth cones but is unlikely to play an autonomous role in the navigation of most temporal RGC growth cones. Foxg1 may, however, influence the navigation of both nasal and temporal RGC growth cones by regulating the properties of the optic chiasm. To test these possibilities, we used DiI and DiA to trace the RGC projections in embryos lacking Foxg1.

We examined RGC projections in wild type at E13.5(Fig. 3A-C) and E15.5(Fig. 3F-H), over which period the optic chiasm develops its mature configuration. Progressing through the brain from rostral to caudal, sections show the injection in the eye(Fig. 3A,F), the optic nerve approaching the optic chiasm along the ventral surface of the brain(Fig. 3B,G) and the optic tract growing dorsally over the thalamus (Fig. 3C,H). At both ages, the contralateral optic tract is more strongly labelled, indicating that the majority of RGC axons cross the midline at the optic chiasm (Fig. 3C,H). Double labelling with DiI injected into one eye and DiA into the other illustrates the X-shape formed by RGC axons at the chiasm(Fig. 3K). Labelling of the optic tract is predominantly from the contralateral eye, as shown by cross-sections of the optic tract as it grows dorsally over the thalamus(Fig. 3K′).

In E13.5 and E15.5 Foxg1^-/- embryos many of the general features of RGC axon navigation were similar to those in wild type. RGC axons from both eyes converged on the ventral surface of the hypothalamus where they were sorted into ipsilateral and contralateral optic tracts growing dorsally over the thalamus (Fig. 3D,E,I,J). In contrast to the wild-type, the ipsilateral tract in mutants appeared greatly increased [compare panels for wild type(Fig. 3C,H) with mutants(Fig. 3E,J); arrows indicate the ipsilateral tract]. Double labelling with DiI injected into one eye and DiA into the other indicated that, although RGC axons converged at the hypothalamic midline, the balance of midline crossing was shifted in the mutants. Although many RGC axons still crossed the midline(Fig. 3L), a large proportion of axons in the optic tracts was labelled from the ipsilateral eye(Fig. 3L′).

To quantify the shift in the proportion of RGC axons projecting ipsilaterally and to identify the sites of origin in the retina of the aberrant ipsilateral projections, we retrogradely labelled RGCs with DiI or DiA from the thalamus. The numbers of nasal and temporal RGCs projecting ipsilaterally and contralaterally were counted in wild-type and mutant embryos at E15.5. As gene expression studies (Huh et al., 1999) (present study) indicate that, despite its medial elongation, the mutant retina does not exhibit greatly disturbed polarity,mutant eyes were divided into nasal and temporal halves using the same anatomical criteria as for wild-type eyes.

Wild-type embryos showed the expected patterns of projection. Retrograde label filled the optic tracts, optic chiasm and optic nerves, and showed most axons originating contralaterally (Fig. 4A-H). Contralaterally projecting RGCs were distributed across nasal and temporal retina (Fig. 4F,G). The much smaller number of ipsilaterally projecting RGCs were more frequently found temporally (Fig. 4G; Table 1). The density of labelled RGCs was highest medially, towards the optic nerve head,and decreased sharply in more lateral retina(Fig. 4F,G). RGC cell bodies were orientated radially with their axons projecting towards the optic nerve head along the inner surface of the retina(Fig. 4H).

In the mutant embryos, retrograde label filled the optic tracts, optic chiasm and optic nerves as in wild type(Fig. 4I-O) but many more axons originated ipsilaterally than in wild type [compare mutant(Fig. 4J) with wild type(Fig. 4C)]. This increased ipsilateral projection arose from both nasal and temporal retina(Fig. 4K,K′,M). Counts of ipsilateral RGCs in serial sections from four mutant and four wild-type E15.5 embryos (Table 1) showed an approximately eightfold increase in ipsilaterally projecting RGCs from both nasal and temporal retina in the mutants. There were no significant changes in the mutants in the numbers of RGCs projecting contralaterally from the nasal or temporal retina (Table 1). Total numbers of RGCs counted in the retina were not significantly altered in mutants. Ipsilaterally and contralaterally projecting RGCs were intermingled in the nasal and temporal retina (Fig. 4M, with higher magnification of RGCs in N) showing that all parts of the mutant retina contain RGCs capable of projecting to either side of the brain. We did not detect double-labelled cells, indicating that the increased ipsilateral projection did not result from large numbers of individual RGCs projecting to both sides of the brain.

As in the wild type, the majority of mutant RGCs are located in the inner layer of the retina and are radially orientated with their axons projecting along the inner surface of the retina (compare Fig. 4O with Fig. 4H), although the mutant RGCs do exhibit a broader radial spread. Again, reminiscent of the wild-type situation, mutant RGC axons fasciculate to form the mutant correlate of the optic nerve, which exits the elongated retina at its point of attachment to the ventral brain (seen in Fig. 4K) and approaches the optic chiasm (seen in a ventral adjacent section in Fig. 4L). As in the wild type, mutant retinal axons form an X-shaped chiasm (compare wild types in Fig. 3K, Fig. 4B,C,E with mutants in Fig. 3L, Fig. 4I,J,L), although their shapes are not identical. Given the distortions to the mutant eye it is remarkable that its retinal projection retains so many features in common with the wild type.

Our results in Foxg1^-/- mutants indicate an increase in the proportion of RGCs whose axons are repelled by the midline and join the ipsilateral optic tract. As in wild type, significantly more ipsilateral projections originated in temporal retina (χ² test; Table 1). This finding provides further evidence for the retention of nasotemporal polarity, as was suggested by the expression pattern of Foxg1 in mutants (see above).

As well as being expressed in the developing retina itself, Foxg1 is expressed in cells that surround and may influence the patterns of expression of other genes in the retina as it develops(Dou et al., 1999). It was possible that in embryos lacking Foxg1 the patterning of the retina might have been altered fundamentally and this might have had consequences for RGC axon navigation. Results above suggested that the nasotemporal polarity of the Foxg1^-/- retina is similar to wild type. Here, we addressed the nature of the other major axis of the mutant retina, i.e. dorsoventral. This was important as most ipsilateral RGC projections normally originate ventrally in the temporal retina(Drager, 1985) and so a repatterning of Foxg1^-/- retina, such that it adopts a predominantly ventral character might have explained an increased ipsilateral projection.

The receptor tyrosine kinase Ephb2 and its ligand ephrin B2 are expressed in complementary dorsoventral gradients in the developing retina and their distributions in the retina and chiasm have been related to RGC axon midline crossing behaviour (Barbieri et al.,2002; Williams et al.,2003). In wild-type embryos at E13.5 and E16.5, ephrin B2 exhibits a dorsal^[high] to ventral^[low] gradient of expression in the retina with the strongest expression dorsolaterally(Fig. 5A,B). At E13.5, Ephb2 is expressed throughout the retina in a ventral^[high] to dorsal^[low] gradient (Fig. 5E). By E16.5 Ephb2 is more evenly distributed across the retina(Fig. 5F), although levels still appear to be slightly higher ventrolaterally. Ephb2 is also strongly expressed in the optic nerve, consistent with expression of Ephb2 protein on the RGC axons (Fig. 5E-F,I-J). RGC axons continue to express Ephb2 protein as they approach the optic chiasm and exit into the optic tract (Fig. 5I,J). By contrast, the optic nerve stains weakly for ephrin B2 protein (Fig. 5A,B).

The dorsoventral gradients of Ephb2 and ephrin B2 proteins are retained in the mutant. In E13.5 Foxg1^-/- embryos, the retina expresses ephrin B2 protein in a dorsal^[high] to ventral^[low] gradient (Fig. 5C) and Ephb2 protein in a complementary ventral^[high]to dorsal^[low] gradient (Fig. 5G). At E16.5, ephrin B2 expression is strongest in the dorsolateral retina (Fig. 5D)and Ephb2 protein is more evenly distributed across the retina with slightly higher levels ventrolaterally (Fig. 5H). As in wild types Ephb2 protein is expressed by mutant RGC axons as they cross the optic chiasm and enter the optic tracts(Fig. 5K,L).

In the wild type the eye is roughly spherical and is attached to the ventral surface of the diencephalon by the optic stalk, whereas the mutant eye lacks an optic stalk and is attached directly to brain. Perhaps as a mechanical consequence of this the mutant eye is elongated medially and squashed dorsoventrally relative to the wild-type eye. To distinguish whether this is only a change in shape or also involves a change in retinal volume, we measured retinal area (mm²) in every 20th 10 μm horizontal serial section through the eyes of wild-type and mutant embryos. As each section was 0.01 mm thick, retinal volumes approximated to ∑(retinal area× 0.2) mm³. Average volumes obtained for wild-type (0.86 mm³) and mutant (0.88 mm³) eyes were very similar. The proportion of the Foxg1^-/- retina that is elongated medially (marked `er' in Fig. 1K) probably corresponds to the central third of the wild-type retina. Consistent with this idea, the medial expansion of the mutant retina expressed levels of ephrin B2 and Ephb2 intermediate between the extremes of the expression gradients shown in Fig. 5D,H (data not shown), as did the wild-type central retina.

These data indicate that, as in the nasotemporal axis, patterning of the retina along the dorsoventral axis is similar in Foxg1^-/-and in wild-type embryos.

The presence of cells at the chiasm that turn on Foxg1 in both Foxg1^+/+ and Foxg1^-/- embryos(Fig. 1) prompted us to examine the mutant chiasm in more detail. The expression patterns of the transcription factor Nkx2.2 and the cell-surface molecule stage-specific embryonic antigen 1(SSEA-1) define stages of normal chiasm development(Marcus and Mason, 1995; Marcus et al., 1999). SSEA-1 is expressed by a population of early differentiating neurons that provides an anatomical template to guide RGC axons at the developing chiasm(Sretavan et al., 1994; Marcus and Mason, 1995; Sretavan et al., 1995). We speculated that changes in the distribution of neurons expressing SSEA-1 in the mutant might perturb RGC axon midline crossing behaviour. However, we were unable to detect any clear shift in the expression domains of SSEA-1 between the wild-type (Fig. 7A,C,E) and mutant (Fig. 7B,D,F) chiasm at E13.5. The transcription factor Nkx2.2 defines a domain through which RGC axons grow at the midline (Marcus et al.,1999), but we did not find evidence of differences between Nkx2.2 expression in the wild-type and mutant (compare Fig. 7G with 7H). Ephrin B2 is expressed by radial glial cells that form a palisade at the chiasm and control RGC axon divergence (Marcus and Mason,1995; Marcus et al.,1995; Nakagawa et al.,2000; Williams et al.,2003). Although the shape of the tissues surrounding the optic chiasm was slightly abnormal in Foxg1 mutants, expression of ephrin B2 was observed along the ventral midline of the diencephalon in both wild-type and mutant embryos at E13.5 and E16.5(Fig. 6). These data show that the mutant chiasm does not show major defects in the expression of at least three molecules known to be important for its interaction with RGC axons.

During normal development, the eye field in the anterior neural plate splits into a right and a left half separated by the developing floor plate(reviewed by Rubenstein and Beachy,1998; Rubenstein et al.,1998). On each side, the optic cup (from which retina and retinal pigment epithelium are derived) forms laterally, the optic stalk forms medially and the retina projects axons along the optic stalk to the chiasm. Foxg1 is first expressed in the surface ectoderm immediately anterior to the neural plate at around E8 in mice. Over the following day, its expression appears in the anterior neural plate and, as this region becomes the anterior neural tube, demarcates the telencephalic neuroepithelium, optic chiasm and nasal retina (Hatini et al.,1994; Xuan et al.,1995; Dou et al.,1999; Huh et al.,1999). The domain of Foxg1 expression in nasal retina includes the locations of projecting RGCs. Only a few cells in the temporal retina, which are located towards the ciliary margin, express Foxg1. Our data show that most RGCs in the temporal retina are descended from lineages in which Foxg1 has never been expressed. These expression patterns indicate that Foxg1 is unlikely to have a significant cell autonomous effect on the development of cells in temporal retina, but may have such an effect on cells in nasal retina.

Our results show that Foxg1 is required to ensure a normal ipsilateral projection from RGCs. The Foxg1^-/- retina produces an increased ipsilateral projection, arising from RGCs located both nasally and temporally. Given the expression patterns discussed above, abnormal development of RGC axons from both nasal and temporal territories argues against a completely cell-autonomous role for Foxg1 in the control of RGC projections. Foxg1 may autonomously regulate the behaviour of nasal RGC growth cones at the optic chiasm, e.g. by influencing their expression of cell surface molecules, but this can not be the case for most temporal RGCs.

The eyes of mice lacking Foxg1 have an abnormal shape(Xuan et al., 1995; Huh et al., 1999) (present study). A striking feature of the Foxg1^-/- phenotype is the lack of an optic stalk connecting the retina to the brain. This could result in the retina becoming stretched medially as the head grows. Consistent with this, the overall volume and patterning of gene expression along its nasotemporal and dorsoventral axes are similar to wild-type retina. This argues against a fundamental repatterning of the retina in the absence of Foxg1. Changes in Foxg1^-/- retinal morphology might also be secondary to defective development of the lens, which expresses Foxg1 early in its formation and is smaller than normal in mutants (Figs 2, 5)(Hatini et al., 1994; Xuan et al., 1995; Dou et al., 1999; Huh et al., 1999). Previous studies have shown that, although lens is not necessary to instruct the differentiation of neuroretina, it does influence the morphology of the retina either mechanically or molecularly(Ashery-Padan et al.,2000).

Foxg1 can be added to the growing list of proteins with a dual role in tissue morphogenesis and axon growth and morphology. Others on this list include transcription factors such as Pax6(Ericson et al., 1997; Mastick et al., 1997; Pratt et al., 2000; Pratt et al., 2002) and secreted proteins of the Wnt (Hall et al.,2000; Yoshikawa et al.,2003), bone morphogenetic protein (BMP)(Augsburger et al., 1999) and hedgehog (Charron et al., 2003)families. Enzymes involved in regulating the sulphation status of extracellular heparan sulphate proteoglycans are implicated in Shh, Wnt and BMP signalling, tissue morphogenesis(Merry and Wilson, 2002; McLaughlin et al., 2003a) and RGC axon guidance (Walz et al.,1997; Irie et al.,2002).

The simplest explanation of our finding an increased ipsilateral projection in embryos lacking Foxg1 is that Foxg1 normally regulates the proportions of crossed and uncrossed projections by a direct action on both temporal and nasal axons at the chiasm itself. In the chick, Foxg1 has been shown to autonomously regulate the retinotectal mapping of RGC axons(Yuasa et al., 1996) and it is possible that in the mouse Foxg1 expression by RGCs is involved in retinotectal mapping rather than in regulating midline crossing at the optic chiasm.

In the wild-type the diencephalon is surrounded on each side by a large telencephalic vesicle. This physical support is missing in the mutant, which has a hypoplastic telencephalon and, as a consequence, the diencephalon adopts a more `open' conformation. Nevertheless, we found that many morphological and molecular features of the wild-type chiasm were retained in the mutant. The normal developing chiasm has been characterised in terms of the expression of transcription factors and regulatory molecules whose expression domains coincide with navigating RGC axons (Marcus et al., 1999; Nakagawa et al.,2000; Williams et al.,2003). We did not detect any gross abnormalities in the distribution of cells expressing the transcription factors Foxg1 or Nkx2.2 or of cells expressing cell surface molecules SSEA-1 or ephrin B2 at the developing chiasm. Despite its elongated shape, the mutant eye projects a correlate of the optic nerve, which converges to form the optic chiasm. It is extremely unlikely that anatomical alterations of the mutant eye or brain are solely responsible for the alterations in the balance of ipsilateral and contralateral projections. Taken together, these findings point to a more specific defect causing the change in RGC axon midline crossing in mutants.

The mechanisms that regulate RGC growth cone navigation at the normal optic chiasm are not well understood, although it is known that Ephb and ephrin B family members are likely to participate in regulating midline crossing of RGC growth cones (Nakagawa et al.,2000; Barbieri et al.,2002; Williams et al.,2003). We did not detect obvious alterations in the distributions of Ephb2 and ephrin B2 protein in the mutant retina and optic chiasm but this does not necessarily preclude a role for Foxg1 in the control of Ephb/ephrin B signalling. There are several precedents for a particular receptor/ligand interaction leading to either attraction or repulsion depending on other factors. The presence of the extracellular matrix molecule laminin has been shown to convert the attraction of RGC growth cones towards a source of netrin to a repulsion of RGC growth cones away from a source of netrin(Hopker et al., 1999). Levels of an intracellular guanylate cyclase control the response of neocortical growth cones to a Sema3a gradient (Polleux et al., 2000). Although an analagous mechanism has not been demonstrated for Ephb/ephrin B signalling it appears that Ephb-expressing axons can be either attracted or repelled by ephrin B in different situations(McLaughlin et al., 2003b; Hindges et al., 2002).

Foxg1 may, therefore, regulate expression at the optic chiasm of one or more of a potentially very large number of modulators of signalling pathways activated by, for example, Ephb/ephrin B. The identification of transcriptional targets of Foxg1 will allow a systematic survey of these possibilities. The final choice made by each RGC growth cone whether to go ipsilateral or contralateral at the optic chiasm probably depends on the balance between attractive and repulsive signals at the chiasm and the nature of the response of each growth cone to those signals. Most probably loss of Foxg1 promotes ipsilateral projections at the optic chiasm by perturbing these balances.

We thank Katy Gillies, Ben Martynoga, James O'Connor, Duncan Mcneil and Linda Wilson (Biomedical Sciences Confocal Facility) for their contribution to this work. We thank Lijian Shen and Eseng Lai for providing the Foxg1^lacZ mice, and Jean Hebert and Susan M^cConnell for providing the Foxg1^Cre mice. The Nkx2.2 (74.5A5) and Ssea1 (MC-480) antibodies developed by Thomas Jessell and Davor Solter, respectively, were obtained from the Developmental Studies Hybridoma Bank developed under the auspices of the NICHD and maintained by The University of Iowa, Department of Biological Sciences, Iowa City, IA 52242. This work was supported by the Wellcome Trust, MRC and BBSRC.