Fgf signals from a novel signaling center determine axial patterning of the prospective neural retina

The retinotopic organization of the midbrain optic tectum in fish,amphibians and birds, or superior colliculus in mammals is a highly precise representation of positional cell identities in the retina, because, according to the position of their somata in the eye, retinal ganglion cell (RGC) axons project topographically to the tectum. As a result, the nasal-temporal (NT)and dorsal-ventral (DV) axes of the retina map inversely to the anterior-posterior (AP) and DV axes of the tectum(McLaughlin et al., 2003). Molecularly, topographic mapping of RGC axons is controlled by axon guidance cues of the Efn (Ephrin) class, which mediate axon repulsion by binding Eph tyrosine kinase receptors. Complementary, axial gradients of Eph-receptors on RGCs in the retina and Efn-ligands on the tectum direct this process(Drescher et al., 1995; Cheng et al., 1995; Flanagan and Vanderhaeghen,1998). The topology of the retinotectal map requires precedent axial patterning to define positional cell identities and to restrict the expression of guidance cues in the retina and tectum. For the tectal target field, the midbrain-hindbrain boundary (MHB), located posterior to the tectum,controls patterning along the anterior-posterior (AP) axis during embryonic development. The secreted growth factor Fgf8 is expressed at the MHB,asymmetric to the tectum, and has a `posteriorizing' influence on early midbrain polarity and later retinotopic organization of the tectum(Crossley et al., 1996; Itasaki and Nakamura, 1996; Friedman and O'Leary, 1996; Logan et al., 1996; Reifers et al., 1998; Martinez et al., 1999; Picker et al., 1999; Nakamura and Sugiyama,2004).

The anatomical stages of vertebrate eye development are well described(Chow and Lang, 2001; Chuang and Raymond, 2002). In the zebrafish, eye morphogenesis commences with the lateral evagination of the optic vesicle from the diencephalic neural keel at 5- to 6-somite stage(Schmitt and Dowling, 1994). But it is not well understood how early eye morphogenesis relates to axial patterning of the prospective neural retina and how this influences topography of the retinotectal map (Peters,2002). DV eye patterning involves bone morphogenetic protein 4(BMP4), which is expressed in the dorsal eye and leads to upregulation of the dorsal T-box transcription factor Tbx5 and to downregulation of the ventral homeodomain transcription factors Vax and Pax2 upon misexpression(Koshiba-Takeuchi et al.,2000; Sasagawa et al.,2002). BMP loss/inhibition results in ventralization of the retina(Sakuta et al., 2001; Sasagawa et al., 2002; Murali et al., 2005). Vax genes are required for ventral retina development and to delineate optic stalk from retinal territories; the latter through a midline-derived Hedgehog (Hh)signal (Schulte et al., 1999; Barbieri et al., 1999; Mui et al., 2002; Take-uchi et al., 2003). Asymmetric distribution of retinoic acid (RA) has also been implicated in DV eye patterning (Hyatt et al.,1992; Kastner et al.,1994; Marsh-Armstrong et al.,1994; Hyatt et al.,1996). A recent study suggested that RA and Fgfs can ventralize the eye by collaborating with Hh, but the exact temporal requirement for these factors remains largely unclear (Lupo et al., 2005). In the chick, retinal DV polarity is gradually determined between the 8- and 14-somite stage(Uemonsa et al., 2002).

By contrast, retinal patterning along the NT axis is only poorly understood. The forkhead transcription factors Foxg1 (BF1) and Foxd1 (BF2)display restricted and complementary expression along the retinal NT axis and cause RGC axon misprojections upon ectopic expression(Hatini et al., 1994; Yuasa et al., 1996). Two transcription factors, sensory organ homeobox (SOHo) and Gallus gallus homeobox 6 (GH6) are expressed in the nasal half of the eye, and cause RGC axon misprojections upon ectopic expression and downregulation of temporal expression of Epha3a (Deitcher et al., 1994; Schulte and Cepko,2000; Stadler and Solursh,1994). Foxg1 can induce ectopic nasal gene expression (SOHo, GH6,Efna5, Efna2) and suppress temporal gene expression (Epha3, Foxd1) upon misexpression in the temporal retina(Takahashi et al., 2003). Despite increasing knowledge about late DV and NT axial restriction of gene expression in the retina, it is largely unknown when and through which factors these patterns are established (McLaughlin et al., 2003). Importantly, the currently known factors that regulate NT eye patterning are themselves already expressed asymmetrically. Fgf signals have been implicated in various aspects of eye development,including the segregation of neural from pigmented retina, lens induction and differentiation, retinal cell fate specification, photoreceptor survival, RGC axon outgrowth and axon guidance, and are thus good candidates to play a role in axial eye patterning (Russell,2003; Yang, 2004). Furthermore, Fgf-dependent differentiation of the retina is mediated by a combined Fgf8/Fgf3 signal (McCabe et al.,1999; Martinez-Morales et al.,2005).

We had previously shown that RGCs from fgf8-mutant eyes form topographically incorrect axon projections on wild-type tecta, suggesting a role for Fgf8 in axial patterning of the retina(Picker et al., 1999). Here,we analyze the exact temporal and spatial requirement for Fgf signaling during axial patterning of the retina. Gene expression along the NT and DV axis of the retina in the fgf8-mutant acerebellar (ace),upon pharmacological inhibition of Fgf-receptor (Fgfr) signaling and upon ectopic Fgf8 expression shows that Fgf8 together with other Fgf proteins determines the NT axis. The temporal requirement for Fgfs is tightly restricted to the onset of eye morphogenesis between the 5- and 10-somite stage. Heterotopic eye transplantations show that this early patterning phase is sufficient to drive subsequent autonomous NT eye patterning. We also show that the early, Fgf-dependent positional code of cells along the NT axis of the eye determines later topographic mapping of RGC axons along the anterior-posterior axis of the midbrain tectum. The developmental timing of(1) fgf8 expression and its patterning activity, (2) of the requirement for Fgfr signaling and (3) of activation of the Fgf-target gene spry4 in the diencephalon and eye suggest that the telencephalon acts as extrinsic Fgf-source and thus is a novel signaling center for retinal patterning.

Fish were maintained at 28°C on a 14 hour light/10 hour dark cycle(Westerfield, 2000). Embryos were staged according to (Kimmel et al.,1995) or in hours post-fertilization at 28°C for 24 hours old or older embryos. The data we present in this study were obtained from analysis of AB or TL wild-type fish and of homozygous acerebellar^ti282a (ace) embryos(Brand et al., 1996). Molecular analysis indicates that ace^ti282a is a complete loss-of-function allele (Reifers et al.,1998) (M.B., unpublished).

Whole-mount mRNA in situ hybridization were carried out as described(Reifers et al., 1998). Digoxigenin-labeled probes were prepared from linearized templates using an RNA labeling and detection kit (Roche). Receptor alkaline phosphatase staining with the chick for Epha3-AP fusion protein was carried out as previously described (Cheng and Flanagan,1994; Brennan et al.,1997). Apoptotic cells were detected by an in-situ nick-end labeling procedure using a commercial kit (Roche). Acridine Orange (AO,Molecular Probes) was used at a concentration of 2 μg/ml to stain live embryos (Brand et al.,1996).

For intensity profiling of retinal gene expression, sets of 8-bit grayscale images of dissected, flat mounted eyes where captured after in situ hybridization as described above. Images where subsequently imported into the ImageJ 1.32j(http://rsb.info.nih.gov/ij/)image analysis software. Using the `Analyze>Plot Profile' option intensity profiles along radial trajectories were captured and further analyzed with Microsoft Excel. Comparisons of gene expression in wild-type and mutant where performed on embryos from the same egg lay and were stained, imaged and analyzed under identical conditions.

The Fgf-receptor signaling inhibitor SU5402 (Calbiochem)(Mohammadi et al., 1997) was diluted into 10 mM stocks in DMSO and added to E3 medium at a working concentration of 5 μM, followed by repeated washing in excess medium. Control embryos were treated in 0.05% DMSO(Reifers et al., 2000).

Optic vesicle transplantation and anterograde RGC labeling was carried out as described before on the same side of donor and host embryo(Picker et al., 1999). Fgf8-protein beads were prepared by overnight incubation of heparin-coated acrylic beads (Sigma) or 45 μm polystyrene beads (Polysciences) in a 250μg/ml solution of recombinant zebrafish Fgf8 protein in PBS. Before implantation, beads were rinsed three times for 10 minutes in PBS. Control beads were loaded with 250 μg/ml BSA.

Various Fgfs are expressed in and around the forming eye from neural plate stages of embryogenesis onwards. We therefore designed experiments to determine if Fgf signaling plays a role in patterning of the prospective retina along the nasal-temporal (NT) and dorsal-ventral (DV) axis. Our initial approach was to alter levels of Fgf signaling by (1) implantation of Fgf8-protein-coated beads to increase signaling; and (2) treatment of embryos with the pharmacological Fgf receptor (Fgfr) inhibitor SU5402 to block signaling during early eye development and to study the response of genes with restricted expression along the retinal axes at the 28 hour stage. Of the genes that are restrictively expressed along the NT axis we analyzed the forkhead-transcription factor foxg1 (BF1), which is expressed in the nasal half of the prospective retina, and two axon guidance molecules of the Eph/Efna class – the efna5a ligand, which is expressed nasally, and the epha4b receptor, which is expressed temporally. Of the genes that are discretely expressed along the DV axis, we analyzed the T-box transcription factor tbx5 as dorsal marker and the homeobox gene vax2 as ventral marker.

In wild-type embryos at 28 hours, expression of efna5a is confined to cells in the nasal half and epha4b to cells in the temporal half of the prospective retina (Fig. 1B,E). Fgf8-bead implantation between the 5- and 7-somite somite stage close to the retina (Fig. 1A,D) leads to a strong `temporal shift' in retinal gene expression: temporal expansion of the nasally expressed gene efna5a(n=7/7) and complementary repression of the temporally expressed epha4b (n=8/8) (Fig. 1C,F). Similar to Fgf8 beads, transplantation of cells from the telencephalic fgf8 expression domain at the 5-somite stage (see Fig. 5A,B) into the optic vesicle can repress temporal epha4b expression (n=3/4)showing that they can act as source of the patterning activity(Fig. 1G-I). Implantation of Fgf8-soaked beads after the 5- to 7-somite stage does not alter NT patterning(n=17/17). This Fgf8-misexpression phenotype strikingly resembles the phenotype of the aussicht (aus) mutant, where Fgf8 expression is generally increased(Heisenberg et al., 1999).

To determine the exact timing of the requirement for Fgf8 in NT eye patterning, we applied the pharmacological Fgfr inhibitor SU5402 to wild-type embryos at discrete developmental intervals between the 1- and 15-somite stage, and studied retinal gene expression at the 28 hour stage(Mohammadi et al., 1997; Reifers et al., 2000; Raible and Brand, 2001). Blocking Fgfr signaling induces a complete `nasal shift' of retinal NT polarity: in wild-type control embryos, the expression of epha4b is strictly confined to the temporal half of the retina at the 28 hour stage(Fig. 2A). After treatment between the 1- and 5-somite stage, the expression of epha4b partially expands into the nasal retina (Fig. 2B). Treatment between the 3- and 8-somite stage and the 5- and 10-somite stage leads to an apparently complete expansion of epha4bthroughout the retina (Fig. 2C,D). Complementary to this expansion, the nasal expression domains of efna5a (Fig. 2G, compare with Fig. 3C for wild-type expression) and foxg1(Fig. 2F, compare with Fig. 3A for wild-type expression) are completely lost. The dorsal expression domain of tbx5is not affected in its DV extent but shifts position nasally(Fig. 2H, compare with Fig. 3G). The ventral domain of vax2 expression is unaffected(Fig. 2I, compare to Fig. 3I). Treatment beyond the 10-somite stage has no influence on epha4b expression(Fig. 2E) and retinal patterning in general.

To qualitatively analyze even subtle shifts in gene expression, we quantified RNA expression levels in the eye in flat-mount preparations by image analysis after in situ hybridization(Fig. 2J,K). For this purpose,linear intensity profiles were captured along a defined trajectory through the eye on digital image sets. As it has been noticed that regulatory genes in the eye are differentially expressed not only along a single axis(McLaughlin et al., 2003)`straight-line intensity profiling' along the canonical NT and DV axes would not capture all changes in gene expression. We therefore applied a `radial profiling scheme' along a clockwise 360° trajectory through the eye(Fig. 2J). Intensity profiling of epha4b gene expression (mean values, n=3 eyes) reveals that treatment between the 1- and 5-somite stage leads to partial nasal shift of the epha4b expression gradient into the dorsal retina,complementary to the partial temporal shift upon ectopic Fgf8 expression(compare with Fig. 1C,F). After treatment between the 3- and 8-somite stage, a remnant gradient is still detectable, whereas treatment between the 5- and 10-somite stage eliminates the gradient completely (Fig. 2K). Apart from a small size reduction along the NT axis(n=9) and of the lens, eyes of inhibitor treated embryos have no obvious morphological defects (Fig. 2L).

Thus, increased Fgf signaling through local misexpression of Fgf8 between the 5- and 7-somite stage can induce a `temporal shift' in retinal patterning,suggesting that Fgf8 itself is the patterning activity. Blocking Fgf signaling through Fgfr-inhibitor treatment conversely induces a `nasal shift'. In contrast to the partial shift in NT patterning after Fgf8-bead implantation,the temporalization after Fgfr-inhibitor treatment is complete. The requirement for Fgf signaling peaks between the 5- and 10-somite stage, the same period when ectopic Fgf8 protein can induce NT patterning shifts. Telencephalic, Fgf8-expressing cells can repress temporal cell fates upon ectopic transplantation into the eye in the same developmental interval, which suggests that the telencephalon – the only anterior expression domain of fgf8 at that time (see Fig. 5A,B) – is the source of Fgf8.

To determine if fgf8 is required for retinal patterning, we analyzed gene expression in fgf8-mutant acerebellar(ace) embryos at 28 hours of development by in situ hybridization(Fig. 3A-J).

In wild-type embryos, foxg1 is expressed in the nasal half of the eye, including the complete ventronasal but only the ventral part of the dorsonasal retinal quadrant. In ace mutant embryos, foxg1expression in the ventronasal quadrant is unaffected but appears reduced in the dorsonasal retina (Fig. 3A,B). Intensity profiling, along a clockwise 180° trajectory through the dorsal eye (mean values, n=8 eyes/gene)(Fig. 3L), reveals that foxg1 is expressed in a nasal-to-temporal decreasing gradient, which is flattened in the dorsonasal retinal segment in the ace mutant, but reaches the same maximum and minimum expression levels(Fig. 3K). Similar to foxg1, efna5a is expressed in the nasal half of the wild-type retina;however, efna5a is also expressed in the complete dorsonasal quadrant. Thus, foxg1 expression is nested within the efna5aexpression domain. In ace mutant embryos, efna5a expression is overall strongly reduced (Fig. 3C,D). This leads to a global flattening of the nasal-to-temporal decreasing efna5a expression gradient in the mutant(Fig. 3K). In the wild type, epha4b is expressed throughout the temporal retina, complementary to efna5a. In ace, the epha4b expression domain expands into the dorsonasal retinal quadrant(Fig. 3E,F). As a result, the nasal-to-temporal increasing epha4b gradient is shifted in the ace mutant, indicating a global upregulation of expression(Fig. 3K). In the wild type, tbx5 is expressed in the dorsomedial retina. In ace mutant embryos, the NT extent of tbx5 appears reduced and shifted nasally(Fig. 3G,H). Tb5x is expressed in a bell-shaped gradient, which is centered on the dorsalmost point of the retina and bilaterally fades out into the dorsonasal and dorsotemporal quadrants. In the ace mutant, this bell-shaped gradient is flattened,indicating a global reduction in gene expression(Fig. 3K). In the wild type, vax2 is expressed in the ventral retina. The expression domain covers two spatially distinct sub-domains: one in the ventronasal quadrant and one in the ventrotemporal quadrant, with the latter extending further dorsally than the other. The medial part of the ventral retina only weakly expresses vax2. In the ace mutant, vax2 expression is unaffected (Fig. 3I,J). In sections along the canonical medial DV axis through the retina, we could not detect any differences between wild type and mutant (data not shown).

We used the intensity profiles to define the radial positions, where gene expression reaches half-maximal intensity values, E_max/2(Fig. 3K). We defined this position as `gene expression boundary'. Comparison between wild type and ace mutant eyes reveals that the position of the thus defined dorsal NT gene expression boundary undergoes a `nasal shift' for foxg1(radial offset=16°), efna5a (radial offset=13°) and epha4b (radial offset=11°). The two lateral gene expression boundaries of tbx5 both shift towards the dorsomedial eye (radial offset, nasal=12°, radial offset, temporal=14°), but importantly the central peak of tbx5 expression, E_max value, is shifted nasally in the mutant (radial offset=12°). Schematic plotting on the basis of these boundary positions faithfully reproduces the gene expression patterns determined by microscopic imaging (compare schematics in Fig. 3M with Fig. 3A-J).

As described above, we found that ectopic Fgf8 protein cannot induce NT patterning shifts beyond the 10-somite stage. Nevertheless, late implanted beads are capable of rescuing the ace mutant eye phenotype: an Epha3-alkaline-phosphatase fusion protein detects Efna-ligand protein in the nasal half of the retina of wild-type embryos(Fig. 4A). The expression of Efna ligands is strongly reduced in ace mutant eyes, and rescued after bead implantation into the forebrain at the 10-somite stage(n=10/10) (Fig. 4C,D). Bead implantations into both wild type and ace mutants at this stage lead to ectopic expression of Efna ligands in the ventralmost region of the nasal retina, but do not induce nasal gene expression in the temporal retina(Fig. 4B,D).

One scenario that could explain the observed NT patterning shifts is the selective loss of the nasal and concomitant expansion of temporal retinal fates in ace and Fgfr-inhibitor-treated embryos. We therefore studied cell death during eye development in these embryos. Although we find a general increase in cell death in the eyes of ace and Fgfr-inhibitor treated embryos, this effect is not significant before the 20-somite stage. Likewise we do not observe a bias of cell death in ace and Fgfr-inhibitor treated embryos to the proximal, future nasal half of the eye (see Fig. S1 in the supplementary material). We therefore exclude the possibility that local cell death is the cause of the alterations in retinal NT patterning upon loss of Fgf signaling.

We find that the lack of fgf8 in the ace mutant causes a partial nasal shift of positional identities in the retina: expansion of temporal gene expression (epha4b) at the expense of nasal gene expression (foxg1, efna5a). Patterning along the DV axis (tbx5,vax2) appears unaffected. Fgf8 is able to rescue the ace mutant eye phenotype beyond the 5- to 7-somite stage, but without shifting NT patterning beyond its range in wild-type embryos. The observations in the genetic loss-of-function situation are consistent with the complete nasal shift after Fgfr-inhibitor treatment, and the temporal shift upon Fgf8 misexpression. Since the nasal shift in the ace mutant is partial compared with the phenotype after Fgfr-inhibitor treatment, we conclude that at least one other Fgf signal together with Fgf8 determines NT retinal patterning.

Fgfs bind to high-affinity tyrosine kinase receptors that activate the intracellular Ras/Mapk signal transduction pathway, and in response the Fgf-feedback inhibitors Sprouty 2 (Spry2), Sprouty 4 (Spry4), Spred and Sef,and the ETS domain factors Erm and Pea3(Furthauer et al., 2001; Raible and Brand, 2001; Lin et al., 2002; Furthauer et al., 2002; Raible and Brand, 2004; Sivak et al., 2005). To provide further evidence for a reception of Fgf signals by optic vesicle cells, we examined the activation of spry4 as Fgf-target gene in domains overlapping and surrounding sites of Fgf gene expression. If the proposed Fgf8 signal from the telencephalon has a direct effect on NT eye patterning, Fgf-target genes should be expressed around the telencephalon,including the optic vesicle during the crucial patterning phase between the 5-and 10-somite stage.

We find that at the 5-somite stage spry4 is co-expressed with fgf8 in the telencephalon of wild-type embryos(Fig. 5A-D). Importantly, we find that spry4 is expressed broader than fgf8 and extends along the DV axis of the forebrain into the dorsal diencephalon and the proximal part of the evaginating optic vesicle. Spry4 is also expressed in nasal placode precursors that surround the telencephalon at this stage (Fig. 5C,D). In ace embryos, the telencephalic and diencephalic expression of spry4, including the proximal optic vesicle, is strongly reduced,whereas the nasal placode precursors express spry4 at normal levels(Fig. 5E,F). Thus, spry4 is expressed and selectively dependent on fgf8 in the diencephalon and proximal optic vesicle in the crucial phase for NT eye patterning, supporting the idea of an Fgf signal emanating from the telencephalon and reaching into the optic vesicle. Low-level spry4expression in the optic vesicle could explain the partial phenotype of ace mutants and argues for at least one other Fgf being involved in retinal NT patterning.

To test (1) whether the effect of the Fgf signal on retinal patterning is irreversible during later development and (2) what the functional consequences of altered Efna/Eph expression in ace and Fgfr-inhibitor-treated embryos are, we studied the topographic projections of RGC axons to the midbrain optic tectum. For this purpose we grafted wild-type optic vesicles with or without prior Fgfr-inhibitor-treatment and optic vesicles from ace mutant embryos to wild-type hosts at the 10-somite stage(Fig. 6A). Anterograde co-labeling of the dorsotemporal (DT) and dorsonasal (DN) retinal quadrants with the lipophilic tracers DiI and DiO in wild-type control transplantations shows that DT and DN RGC axons form two spatially distinct termination zones(TZs) along the anterior-posterior axis of the ventral tectum, according to the NT position of their somata in the retina(Fig. 6B-D). Wild-type larvae with an optic vesicle graft from an Fgfr-inhibitor-treated embryo show a pronounced topographic shift in the TZs of RGC axons: although the projection of DT RGC axons is unaltered and forms an anterior TZ, DN RGC axons display a completely `temporalized projection behavior', terminating in an ectopic anterior TZ like the DT RGC axons (Fig. 6E-G). Wild-type larvae that received an optic vesicle graft from an ace embryo show an `intermediate' RGC axon projection phenotype:the projections of DT RGC axons are unaltered and form an anterior TZ, but DN RGC axons form an ectopic anterior TZ in addition to the topographically correct posterior TZ (Fig. 6H-J).

These results demonstrate that the early defects in retinal NT patterning observed in the ace mutant and Fgfr-inhibitor-treated embryos produce corresponding topographic shifts in the later projection of RGC axons: partial temporalization of the nasal retina in ace and full temporalization upon Fgfr-inhibitor treatment. Furthermore, this shows that the alterations in eye patterning are irreversible after optic vesicle transplantation to wild-type host at the 10-somite stage and do not impair RGC differentiation. Consistent with the correct early DV patterning of the eye in the absence of fgf8 and Fgfr signaling, we do not observe RGC axon misprojections along the DV axis of the tectum.

Finally, we asked whether any other or later signal derived from the forebrain or head of the embryo, apart from the Fgf signal, is required for retinal NT patterning. To achieve this, we grafted optic vesicles between the 5- and 7-somite stage ectopically onto the yolk cell, by slipping the graft into a pocket between the host epidermis and yolk cell membrane(Fig. 7A-C). Earlier transplants were not possible, because optic vesicle morphogenesis has just started at this stage. The transplantation spatially isolates the optic vesicle from the embryonic body proper and the above-described Fgf sources in the head region. Observation of live embryos shows that these heterotopic grafts integrate, grow and develop autonomously and ultimately form an ectopic, morphologically normal eye (Fig. 7D,E). At the 28 hour stage (16 hours post-transplantation) we tested NT gene expression, and find that both nasal expression of efna5a (n=5/7) and temporal epha4b expression(n=7/8) are correctly established and maintained in such grafts(Fig. 6F-I). This shows that the retina can undergo correct NT patterning in the absence of all signals from the forebrain or head after the 5- to 7-somite stage. Together with the results of the Fgfr inhibition experiments, this argues for an early Fgf signaling phase around the 5-somite stage that is sufficient to trigger autonomous NT eye patterning. The missing retinal pigment epithelium in the grafts at later stages of development (inset in Fig. 7E) is probably due to a lack of induction from the head mesenchyme in the ectopic situs(Fuhrmann et al., 2000).

In this study, we have investigated the role of Fgf signaling in axial patterning of the prospective neural retina. We find that Fgf signaling determines positional cell identities along the NT axis of retina already at the time of optic vesicle formation (Fig. 8A). Temporally restricted inhibition of all Fgf receptor (Fgfr)signaling in embryos causes a complete `nasal shift' of cell fates(Fig. 8C), while the `nasal shift' observed in fgf8-mutant ace eyes is partial(Fig. 8B). The peak of sensitivity towards Fgfr-inhibition is between the 5- and 10-somite stage. Conversely, ectopic expression of Fgf8-protein induces a `temporal shift' in NT patterning during this interval (Fig. 8D). Normal retinal DV patterning in the absence of Fgf signaling shows that this axis is determined independently of the NT axis. We therefore suggest that the crucial period for Fgf-dependent retinal patterning is between the 5- and 10-somite stage. During this period, fgf8expression is restricted to the telencephalon and neighboring the optic vesicle field; ectopically transplanted cells from this region can alter retinal NT patterning similar to Fgf8. Thus, Fgf8, emanating from the telencephalon probably acts as an extrinsic signal on eye patterning. Fgf8-dependent expression of the Fgf-target gene spry4 around the telencephalon, in the dorsal diencephalon and in the optic vesicles further supports this hypothesis. Based on the local activation of spry4, it is most likely that Fgfs directly promote nasal cell fates.(Fig. 8E). Topographic misprojections of RGCs with Fgf-dependent shifts in NT patterning axons along the AP axis of the midbrain tectum underscore the irreversibility and functional relevance of these observations for later eye development and retinotectal mapping.

We conclude (1) that the period of optic vesicle evagination between 5- and 10-somite stage is the crucial phase for terminal, retinal NT patterning; (2)that Fgf8 acts together with at least one other, as yet unknown Fgf to determine this axis; and (3) that the source of the Fgf8 signal for NT patterning is the telencephalon. We thus suggest that the telencephalic precursors at this stage function as novel signaling center for axial patterning of the retina by signaling bilaterally into the evaginating optic vesicles. Interestingly, we find that in most experiments that the dorsonasal retina seems to be more sensitive to the Fgf signal than the ventronasal retina, which suggests a differential competence of primordium.

We show that NT gene expression patterns in retinae compromised for Fgf signaling display a `nasal shift' compared with the wild type(Fig. 8A-C). The complete`nasal shift' in retinal polarity upon blocking of all Fgfr signaling,compared with the partial effect in the ace mutant and remnant spry4 expression in ace mutant early optic vesicles,demonstrates the combinatorial nature of the Fgf signal(Fig. 8C). fgf3 and fgf24 have been previously described to function combinatorially with fgf8 and are thus good candidates to be part of such a combinatorial signal (Leger and Brand, 2002; Walshe and Mason, 2003; Draper et al., 2003; Martinez-Morales et al.,2005). fgf3 is strictly co-expressed with fgf8in the telencephalon at the 5-somite stage and there it regulates spry4 expression together with fgf8(Fürthauer et al., 2001; Raible and Brand, 2001; Herzog et al., 2004). It thus could signal from the same source as fgf8 during retinal patterning. fgf24 is not expressed in the telencephalon, but in the mesenchymal precursor cells of the nasal placode, which are surrounding the anterior neural keel at the 5-somite stage, where spry4 is also expressed but not dependent on Fgf8 (Whitlock and Westerfield, 2000). Previous studies have suggested the extraocular mesenchyme could act as source of extrinsic factors for optic vesicle differentiation (Fuhrmann et al.,2000). It is therefore possible that the fgf24-expressing precursors cells of the nasal placode act as an alternative source of Fgf during retinal NT patterning. Another candidate is the recently published fgf17b, which may be co-expressed with fgf8 and fgf3 in the telencephalon (Cao et al., 2004). Preliminary results with Fgf3 morpholinos(Leger and Brand, 2002)injected into ace mutant embryos suggest that Fgf8+Fgf3 cannot account for the signal (A.P. and M.B., unpublished). Further loss-of-function experiments will clarify the molecular nature of the Fgfs that function with Fgf8 in retinal NT patterning.

Gene expression along the retinal DV axis is not affected upon Fgf-dependent alterations in NT patterning. By experimentally interfering with Fgfs through gastrulation stages, two studies had suggested Fgfs to signal in parallel with Nodal, Hedgehog (Hh) and retinoic acid (RA) in DV eye patterning(Take-uchi et al., 2003; Lupo et al., 2005). In addition, interactions between the DV and NT patterning mechanisms have been proposed (Huh et al., 1999; Mui et al., 2002; Sakuta et al., 2001; Koshiba-Takeuchi et al.,2000). Our results show that the crucial phase for Fgf-dependent retinal NT patterning is after gastrulation; interference with Fgf signaling during this time does not change DV patterning. The reduction in the dorsal expression domain of tbx5, that we observe in the ace mutant is restricted along the NT axis and does not affect the DV extent of the domain. This argues for the independence of the NT and DV eye patterning mechanisms.

The exact developmental timing of the events that lead to axial patterning of the retina has not been clearly resolved. Grafting studies in chick embryos suggest that the retinal NT axis is determined before HH stage 10-11 (10-13 somite stage) (Dutting and Meyer,1995). Through combining loss- and gain-of-function analysis, we show that the crucial phase for Fgf-dependent NT patterning is between the 5-and 10-somite stage, when fgf8 expression is clearly confined to the telencephalon and the optic vesicle has just started to evaginate from the diencephalon and is still connected along its full AP extent to the forebrain(Schmitt and Dowling, 1994). Fgfr inhibitor application during discrete developmental intervals was crucial for this observation, as a complete nasal shift in gene expression along the NT axis of the eye can only be induced by treatment between the 5- and 10-somite stage. Partial NT shifts induced by treatments directly before that stage, are most readily explained by persistence of the inhibitor. The importance of the phase between the 5- and 10-somite stage for eye patterning is independently supported by the reduced expression of the Fgf target spry4 in the diencephalon and optic vesicle in ace mutant by the 5-somite stage, which indicates a non-autonomous requirement for Fgf8 in eye development at this early stage.

Interestingly, the ace mutant eye shows an extended responsiveness to Fgf8, beyond the early NT patterning phase, in rescue experiments. This is reminiscent of the observed plasticity in the midbrain-hindbrain region(Jaszai et al., 2003). But the observed, retained competence to respond to Fgf8 is limited as Fgfr-inhibitor treated and ace mutant eyes do not restore normal NT patterning, when transplanted into wild-type host brains beyond the 10-somite stage. Instead,patterning is terminally altered in the grafts, which is reflected by topographic RGC axon misprojections 5 days after transplantation. Thus, beyond the 10-somite stage, the endogenous patterning activity of Fgfs is not effective.

Previous studies have taken advantage of culturing optic vesicles or retinal explants to determine the influence of extrinsic signals on eye development (Pittack et al.,1997; Fuhrmann et al.,2000). Unexpectedly, our heterotopic transplantations of optic vesicles to the yolk cell at the 5- to 7-somite stage demonstrate that the inductive events prior to this stage are sufficient to drive correct NT patterning, autonomously in the absence of extrinsic cues from the forebrain/head. Taken together with our findings on the requirement for Fgfs,in particular Fgf8, during that period, we propose that a short pulse of Fgfs around the 5-somite stage triggers NT eye patterning, and the eye becomes independent of exogenous signal supply shortly thereafter.

Because epha4b expression is confined to the temporal retina in normal zebrafish, its expansion could directly explain the misprojection of RGCs from the nasal retina of ace mutant and Fgfr-inhibitor treated eyes to an ectopic, anterior termination zone (TZ) by increasing the sensitivity of axons for repellent Efna-ligands expressed in gradients on the tectum (Brennan et al., 1997; Picker et al., 1999). Remarkably, nasal RGC axons from an ace mutant eye form an ectopic anterior TZ in addition to a normal posterior TZ, indicating a partial temporalization, in contrast to a full temporalization upon blocking of all Fgfr signaling. These duplicated TZs are similar to the observations made in mice with targeted deletions of Efna5, Efna2 and Efna5/Efna2(Frisen et al., 1998; Feldheim et al., 2000). Consistent with normal gene expression along the DV axis of the retina, we do not find DV mapping defects of RGC axons from ace or Fgfr-inhibitor treated eyes. It has recently been shown that nasally expressed foxg1(CBF1) represses temporal and induces nasal gene expression, similar to the results that we obtain for Fgfs(Takahashi et al., 2003). It is thus possible that Fgf signaling acts upstream of the transcriptional regulator foxg1.

We propose that the telencephalon is the source of the Fgf8 signal that influences NT eye patterning. At the onset of optic vesicle evagination,between the 5- and 10-somite stage, the telencephalon is the only fgf8 expression domain in the head region. Expression there coincides with the timing of the observed effects of Fgfs on NT eye patterning.

(1) Fgfr inhibitor treatments show that Fgf signaling is crucial between the 5- and 10-somite stage, and not required for NT patterning beyond the 10-somite stage, when Fgf8 starts to be expressed in the optic stalk and retina.

(2) Ectopic Fgf8 protein expression shows a restricted competence to induce NT patterning shifts between the 5- and 10-somite stage.

(3) Reduction of the graded expression of the Fgf target spry4 in the dorsal diencephalon, including the proximal optic vesicle in the ace mutant at the 5-somite stage, shows active Fgf signaling in the region.

Although our loss- and gain-of-function experiments consistently show that Fgf8 influences retinal NT patterning, this effect could still be indirect and reflect a previously reported, autonomous requirement for Fgf8 during telencephalon development rather than a direct signal(Shanmugalingam et al., 2000; Walshe and Mason, 2003). However, the fact that (1) the Fgf target gene spry4 is dependent on fgf8 in the diencephalon and optic vesicle and (2) altered NT patterning after Fgf8-bead implantation support that Fgf8 itself is the signaling entity. As the simplest model, we thus favor a`trans-neuroepithelial' propagation of the Fgf8 signal from the dorsal telencephalic source along the DV axis of the neural tube into the ventrally located diencephalon and the contiguous proximal-distal axis of the evaginating optic vesicle (Fig. 8E). Fate-mapping of cells in the optic vesicle at the 10-somite stage shows that the proximal-distal and posterior-anterior axis of the vesicle match the NT and DV axis of the retina at later stages(Li et al., 2000). Thus, in the crucial patterning phase between the 5- and 10-somite stage, the NT axis of the prospective retina is perpendicular to the AP axis of the telencephalon, which therefore could function as an asymmetric Fgf source. Such a `trans-neuroepithelial' Fgf8 distribution could well explain (1) the graded activation of spyr4 in the diencephalon and optic vesicle, and(2) the Fgf-dependent induction of proximal/future nasal (foxg1- and efna5a-expressing) and suppression of distal/future temporal(epha4b-expressing) retinal cell fates. The similarity between the ace phenotype and the effects caused by partial Fgfr-inhibition suggests that the other Fgf(s) act via a similar mechanism, but analysis of further candidate Fgfs will be needed to determine whether the other molecular components of the combinatorial signal also emanate from the telencephalon.

Considering the role of Fgf expression at the midbrain-hindbrain boundary for tectal AP patterning (Lee et al.,1997; Picker et al.,1999), it is intriguing to find that Fgfs also determine eye patterning along the NT axis. Because both axes are matched in the topography of the retinotectal projection, it is tempting to speculate that this dual requirement reflects a common control mechanism for the positional code of cells in the source and target area of retinotectal map. If this were true,graded interference with Fgfs, common to source and target field, would create parallel shifts in axial patterning of eye and tectum. In evolutionary terms,this might confer `robustness' to the map, because global fluctuations in Fgf signaling would result in coordinate changes of axial patterning of eye and tectum, which are compensated for each other in the retinotectal topography.

We thank Caroline Brennan and Uwe Drescher for reagents, and Carl-Philipp Heisenberg and Andrew Oates for comments on the manuscript. This work was supported by funds from the EU (Zf models), DFG SFB 655 and the Human Frontiers Science Program, and by the Max-Planck-Society to M.B.