Dynamic Coupling of Pattern Formation and Morphogenesis in the Developing Vertebrate Retina

In this Research Article, Picker et al. show how cells in the retina get their spatial coordinates.


Introduction
Map-like representation of sensory information is an evolutionary conserved principle of brain organization and function [1]. The point-to-point mapping of retinal ganglion cell (RGC) axons onto the midbrain tectum/superior colliculus of the vertebrate, is a hallmark example for the requirement of precise pattern formation during embryonic development, since mapping occurs according to the position of RGCs along the nasal-temporal (anteriorposterior) and dorsal-ventral axes of the retina. The topographic projections of RGC axons accurately preserve information on cell positions and neighborhood relationships in the retina as a continuous map of terminals in the tectum [2].
Cell-surface axon guidance molecules expressed in gradients across the retina and tectum control the formation of retinotopic connections [3][4][5][6][7][8]. Guidance molecule expression along the nasaltemporal retina axis is regulated by the nasal-and temporalspecific transcription factors Foxg1, Foxd1, SOHo, and GH6 [9][10][11][12]. However, expression of these factors in the retina is asymmetrical from the onset, indicating that they act downstream of nasal-temporal axis specification. Retinotopic mapping consequently occurs as a function of RGC position along molecular gradients within a coordinate system set by the major retinal axes. This suggests that axis formation and mapping are intimately connected developmental processes, but the nature and timing of the signals that establish cell positional identities in this coordinate system are largely unknown.
Resolving the mechanisms underlying the allocation of positional identity to retinal cells is confounded by the complex morphogenetic rearrangements of forebrain tissues that occur during eye formation [13][14][15][16]. Morphogenesis of the retina begins with the lateral displacement of cells in the eye field to the site of future optic vesicle evagination [17,18]. Subsequently, cells continuously evaginate from the forebrain, steadily increasing the size of the optic vesicle. Next, the optic vesicle invaginates to form the two-layered optic cup, with the outer layer, which faces the surface ectoderm and lens, fated to become neural retina and the inner layer the retinal pigment epithelium. Cell movements from the presumptive pigmented epithelium into the neural retina may occur during this phase [19]. Later optic cup development depends on integrin-mediated focal adhesion at the basal side of the retinal epithelium [20]. However, although the anatomy of the optic cup is well described, dynamic in vivo analysis of its formation is lacking, and axial patterning of the prospective retina has not been studied prior to the completion of optic cup formation [21,22]. It thus remains to be determined how and when axial cell positions in the retina are specified and how the orientation of the retinal axes is related to the axes of the neural tube during eye morphogenesis.
Fgfs have conserved functions in axial patterning of the neural tube. Originating from local organizers, they can control cell positional identities in adjacent regions, with several Fgfs often acting in a combinatorial manner [23,24]. We have previously shown that Fgf8 signaling contributes to nasal-temporal patterning of the retina, and proposed a combinatorial Fgf signal to exert full control over the specification of this retinal axis [25], but the nature of the Fgfs involved, their dynamic requirement, and sites of action during eye morphogenesis are not resolved.
We show here that a combined Fgf8/3/24 signal specifies the positional identities of cells along the nasal-temporal retinal axis in zebrafish embryos. Absence of all three factors leads to completely temporalized, and ectopic activation of Fgf signaling to completely nasalized retinae. Axis specification occurs very early, at the onset of optic vesicle evagination, when fgf8/3 are expressed in the dorsal forebrain and fgf24 in the dorsally located olfactory placode. During formation of the optic vesicle, Fgf signaling from these dorsal sources is required to confine expression of foxg1 (and other future nasal genes) to the dorsal half, and foxd1 (and other temporal genes) to the ventral half of the evaginating optic vesicle. Thus, at the moment of specification, nasal-temporal cell positional identities align parallel to the dorsal-ventral axis of the neural tube and asymmetrically relative to the dorsal sources of Fgf.
By in vivo tracking of GFP-labeled nasal and temporal retina progenitor cells in transgenic lines, we further show how tightly synchronized morphogenetic cell movements and cell shape changes during optic cup formation lead to axis reorientation and the final nasal-temporal subdivision of the neural retina. This occurs as the result of two temporally concordant morphogenetic processes: (1) compaction of nasal retina progenitors already residing in the future neural retina domain-possibly by shortening along the lateral and elongation along the apical-basal cell axis-and (2) directed movement of temporal retina progenitors into that domain. In this process, Foxg1 promotes cohesion of nasal progenitor cells in an Fgf-dependent manner, thereby probably allowing the gradual addition of temporal progenitors to the growing neural retina. Thus, the dynamic coordination of pattern formation with neuroepithelial morphogenesis through Fgf8/3/24 signaling controls the final arrangement of axial cell positions in the retina.

Combined Fgf8/3/24 Signaling Controls Nasal-Temporal Patterning of the Retina
Fgf8 is involved in nasal-temporal patterning of the retina, but loss of fgf8 results only in subtle patterning defects, demonstrating the presence of other unknown factors controlling this process [25]. To test whether Fgf8 acts in combination with other Fgfs, we studied the expression of nasal and temporal marker genes in embryos that lack two or more Fgfs.
We find that fgf3 and fgf24 strongly interact with fgf8 in nasaltemporal patterning ( Figure 1A and 1B). In wild-type (wt) control embryos at 28 h, the Eph receptor epha4b is expressed in a temporalto-nasal decreasing gradient, and the ephrin ligand efna5a in a complementary nasal-to-temporal decreasing gradient. In fgf8 2/2 mutants, epha4b expression expands into the dorsonasal retina, and efna5a expression is reduced in that region. This phenotype is enhanced in fgf8/3 2/2 double mutants; and in fgf8/24 2/2 double mutants, the changes in epha4b and efna5a expression are even stronger, now also affecting the ventronasal retina. Upon inactivation of all three fgfs, by fgf24-morpholino injection in fgf8/3 2/2 double mutants (referred to as fgf8/3/24 2/2 ), all retinal cells express epha4b and none efna5a. The same result is obtained upon blocking of all Fgf receptor signaling with a pharmacological inhibitor (FgfRinh.) between the 1-and 5-somite stages (ss). Fgf3 2/2 and fgf24 2/2 single-mutants have no detectable retinal patterning or eye defects (unpublished data).

Nasal-Temporal Pattern Is First Evident along the Dorsal-Ventral Axis of the Optic Vesicle
The eye undergoes highly complex morphogenetic movements during its evagination, and the locations of cells contributing to the future nasal and temporal retina have not been followed during this process. To investigate the initial orientation of the retinal axes, we studied the early expression of genes that are later restricted along the nasal-temporal retina axis. In wt 10ss embryos, the future nasal marker, foxg1, is expressed in the dorsal leaflet of the evaginating optic vesicle ( Figure 1E, left), whereas the future temporal marker, foxd1, (see Materials and Methods and Figure  S11 and S12 for foxd1 gene nomenclature) is expressed in the

Author Summary
The vertebrate brain contains a point-to-point representation of sensory input from the eye. This visual map forms during embryonic development, by neuronal cells of the retina sending targeted axon projections to the brain. Since the projection needs to wire up neighboring cell positions in the retina to neighboring target areas in the brain, all retinal cells must harbor a defined spatial coordinate as prerequisite for map formation. How such a retinal coordinate system is established and maintained in the dynamically evolving embryo is a fundamental, but unresolved, question. By combining genetic analysis and in vivo imaging in zebrafish embryos, we have tracked the developmental origin of cell coordinates in the retina. We find that three related Fgf signals emanating from outside the eye define relative cell positions in the retina very early, already at the onset of its formation. But the absolute position of retinal cells relative to the body axes is greatly rearranged during subsequent development. In this phase, surprisingly, the same Fgf signals that at first defined retinal cell positions now balance asymmetric cell movements and cell shape changes, which are required for harmonic retinal growth and the final alignment of cell coordinates in the eye.
ventral optic vesicle leaflet ( Figure 1F, left). Surprisingly, nasaltemporal markers, therefore, initially align with the dorsal-ventral axis of the neural tube. By 20ss, the future nasal markers, foxg1 and efna5a, are restricted to the anterior-medial optic cup ( Figure 1G and 1I, left), whereas the future temporal markers foxd1 and epha4b are expressed in a complementary posterior-lateral domain ( Figure 1H and 1J, left). From 24 h, these markers are expressed in the anterior nasal half and posterior temporal half of the retina ( Figure 1A and 1B and unpublished data). Time-lapse imaging of transgene-labeled retinal cells during optic cup morphogenesis supports the conclusion from these gene expression analyses that the dorsal-ventral axis of the optic vesicle corresponds to the later nasal-temporal axis of the retina (see below).
We next addressed the source of the Fgfs that pattern the optic vesicle along its dorsal-ventral axis, the future nasal-temporal axis of the retina. Early, spatially restricted fgf8/3/24 expression relative to the domains of nasally/temporally expressed genes explains the patterning activity of Fgfs along the dorsal-ventral axis of the optic vesicle ( Figure 1K). At 5ss, towards the end of the requirement phase for Fgf signaling in nasal-temporal patterning, fgf8 and -3 are expressed in the dorsal forebrain, which is contiguous with, and close to, the foxg1-expressing, dorsal optic vesicle leaflet and distant to the foxd1-expressing, ventral leaflet. fgf24 is expressed in cells of the nascent olfactory placode [26], at the hinge between the dorsal forebrain and the dorsal optic vesicle leaflet. Fgf24-expressing cells remain in close contact with the developing optic vesicle and cup during later morphogenesis ( Figure S1). Thus, expression of future nasal markers occurs close to the dorsal source of Fgfs and expression of future temporal markers distant to it ( Figure 1L). Local and graded Fgf signaling in the dorsal optic vesicle leaflet is further supported by nested, Fgfdependent expression of the Fgf pathway target genes erm, pea3, spry2, and spry4 ( Figure S2).

Fgf Signaling Enhances Proliferation of Nasal Retina Progenitors
Since Fgfs are known mitogens and Foxg1 promotes neural progenitor proliferation [27][28][29][30], we assessed whether the nasaltemporal asymmetry in Fgf signaling affects cell proliferation in the optic vesicle.
Control embryos, stained for the mitosis marker Phospho-Histone H3 (PH3), show slightly more PH3-positive cells in the dorsal than the ventral optic vesicle leaflet at 5ss (Figure 2A, top). This asymmetry becomes clearer at 10ss ( Figure 2B, top), when the apical side of the dorsal leaflet is often densely populated by PH3positive cells, a pattern never observed in the ventral leaflet. In FgfR-inh.-treated embryos, this asymmetric proliferation pattern is lost (Figure 2A and 2B, bottom). Counting and plotting of the mean ratios of dorsal/ventral leaflet PH3-positive cells shows the increasing asymmetry in proliferation in control embryos and its loss upon FgfR-inhibition: at 10ss, 2-fold more dividing cells are found in the dorsal than in the ventral leaflet in control embryos, whereas the ratio is near 1:1 after FgfR inhibition ( Figure 2C). Analysis of the mean PH3-positive cell number per optic vesicle leaflet, shows that progenitor proliferation is selectively affected in the nasal retina primordium/dorsal optic vesicle leaflet after FgfR inhibition, whereas proliferation of temporal retina progenitors in the ventral optic vesicle leaflet is unchanged ( Figure 2D). Similarly, BrdU incorporation at 10ss is severely reduced in nasal progenitors of the dorsal optic vesicle leaflet upon FgfR inhibition. Treatment has no obvious effect on temporal progenitors in the ventral leaflet ( Figure 2E). Thus, Fgf signaling is selectively required for enhanced proliferation of nasal retinal progenitors during optic vesicle evagination, and this requirement coincides with the Fgfdependent regulation of foxg1 in the dorsal optic vesicle leaflet.

Directed Movements of Temporal Retina Progenitor Cells into the Optic Cup
The unexpected initial alignment of future nasal-temporal markers along the dorsal-ventral axis raised the question how the nasal-temporal axis reaches its final anterior-posterior orientation. We thus analyzed the dynamic development of the nasal-temporal axis by in vivo imaging of eye formation in transgenic Tg (-8.0cldnb:lynGFP)zf106 embryos [31], which we find express GFP in the nasal retina throughout development ( Figure 3A and S3, see Materials and Methods). From 10-to 15ss, cldnb:GFP is expressed throughout the dorsal optic vesicle leaflet, but between 18-and 25ss, cldnb:GFP expression becomes progressively restricted to the dorsal half of the outer layer of the optic cup. The portion of the optic vesicle that contacts the lens ectoderm will form the neural retina, and henceforth, we use the term outer layer to describe this in the nasal retina of wild-type (wt) control, fgf8 mutant (fgf8 2/2 ), fgf8/3 double mutant (fgf8/3 2/2 ), fgf8/ 24 double mutant (fgf8/24 2/2 ), fgf24-morpholino injected fgf8/3 double mutant (fgf8/3/24 2/2 ), and wt embryos after Fgf receptor inhibitor treatment (wt+FgfR-inh.) at 28 h (nasal is to the left and dorsal is up). Dashed-line circle indicates position of the lens. (C) Nasal-temporal profile of epha4b and efna5a gene expression levels (mean grey value for n = 8 images analyzed for each genotype, y-axis, in defined nasal-temporal (NT) regions of the retina, x-axis, see inset, one representative standard deviation is plotted). (D) Schematic illustrating the graded, nasal expansion of the temporal marker epha4b in the phenotypic series of fgf mutants. (E and F) Expression of the future nasal marker foxg1 (E) and the future temporal marker foxd1 (F) at 10ss in the optic vesicle (ov) of wt (left) and fgf8/3/24 2 part of the forming optic cup. At 28 h, after completion of optic cup morphogenesis, cldnb:GFP is restricted to the nasal half of the neural retina.
To investigate the mechanism that gradually restricts cldnb:GFP expression to the dorsal half of the outer layer of the optic cup, we performed time-lapse imaging between 18-and 24ss ( Figure 3B and S4). cldnb:GFP expression initially reaches the distal limit of the optic cup, which we term the ridge, but within about 2.5 h, its distal limit is approximately nine cell diameters from the ridge. Time-lapse movies show pronounced outward cell movement in the inner layer of the optic cup towards the distal ridge (Video S1). This suggests a gradual displacement of cldnb:GFP-positive, nasal retina progenitor cells in the outer optic cup layer-the future neural retina domain-by a late movement of cldnb:GFP-negative cells, presumably from the inner layer, around the optic cup ridge.
To determine whether outer layer cells are indeed displaced in this way, we followed the movement of DsRed2-expressing outer layer cell clones in transgenic membrane-GFP Tg(Bactin:HRAS-EGFP)vu119 embryos [32]. A representative cell-tracking experiment shows how a cell in the ridge region (blue) of the outer optic cup (distance to ridge at 0 min: one cell diameter) is gradually displaced dorsally and proximally (distance to ridge at 1 h 25 min: six cell diameters), and as this happens, it elongates along the apicalbasal axis (apical-basal axis at 0 min: 22 mm, at 1 h 25 min: 41 mm) ( Figure 3C and Video S2). Notably, this lateral displacement within the optic epithelium occurs with the same kinetics as the displacement of the distal limit of cldnb:GFP expression (compare Figure 3B to 3C). A representative cell positioned further dorsally (white) is barely displaced laterally and elongates only slightly (apical-basal axis at 0 min: 34 mm, at 1 h 25 min: 45 mm) ( Figure 3C). At 36 h, the clone in Figure 3C, which initially covered the complete extent of the outer optic cup layer (see insets in Figure 3C), is restricted to the nasal hemiretina ( Figure S5; n = 5/5 analyzed outer layer clones) confirming that outer layer cells are all initially destined for nasal retina. This suggested that cell movements from the inner optic cup layer around the distal ridge region gradually add nonnasal retina progenitors to the outer layer of the optic cup. This addition of cells occurs coincident with the elongation of the apical-basal axis of nasal progenitors already residing in the outer optic cup layer, suggesting a gradual compaction of the future neural retina epithelium.
The gradual encroachment of GFP expression into the outer layer in the HGn42A enhancer trap line [33] is complementary to the restriction in cldnb:GFP expression ( Figure 3D). The insertion in HGn42A maps to a site 52-kbp downstream of the foxd1 locus, and GFP expression in this line recapitulates endogenous foxd1 expression in the prospective temporal retina ( Figure S6). These results support the conclusion that HGn42A:GFP-positive, prospective temporal, retinal cells move around the distal optic cup

Nasal Retina Progenitors Delaminate from the Optic Vesicle Upon Loss of FgfR Signaling
Fgf8/3/24 2/2 embryos form smaller, but otherwise morphologically normal, retinae ( Figure S7) despite the complete loss of nasal-temporal polarity and reduced proliferation, suggesting global eye morphogenesis is not compromised in the absence of nasal-temporal patterning. However, we do find that Fgf signaling specifically and regionally affects epithelial cell morphology and behavior at the onset of optic cup formation.
Increasingly reduced levels of cldnb:GFP expression in the dorsal optic vesicle leaflet after FgfR-inh.-treatment indicates that reporter expression depends on Fgf signaling ( Figure 4A and 4B; n = 7/7). At 15ss, cells in the dorsal leaflet appear disorganized, whereas cells in the lower leaflet appear normal ( Figure 4B), the tight apical membrane apposition of the optic vesicle leaflets is lost, and the ventricle contains delaminated, weakly cldnb:GFP-positive (unpublished data) cells ( Figure 4C and 4D, n = 5/5). The delaminated cells eventually undergo apoptosis, but viability of the disorganized cells in the optic vesicle neuroepithelium is not compromised ( Figure 4E). To test whether an early defect in apical-basal cell polarity causes later delamination, we studied the expression of the two apical markers aPKC and ZO1. At 10ss, prior to cell delamination, expression of both markers is normal in FgfR-inh.-treated embryos compared to controls ( Figure S8A and S8B). At 15ss, after the onset of delamination, expression reflects the loss of apical membrane apposition and accumulation of cells To directly assess the requirement for FgfR-signaling in nasal retina progenitors, we transplanted Tg(hsp70l:dnfgfr1-EGFP)pd1positive cells [34] that express a dominant-negative, GFP-tagged version of Fgfr1 under the control of a heat-shock promoter, into wt host embryos and heat-shocked the chimeras at the onset of optic vesicle evagination. At 12ss, embryos with dnfgfr1-EGFP clones in the dorsal optic vesicle leaflet show reduced foxg1 (n = 10/ 12) and ectopic expression of foxd1 (n = 8/12) ( Figure S9), reminiscent of fgf8/3/24 2/2 embryos ( Figure 1E and 1F Although the behavior of prospective nasal cells is disrupted upon abrogation of Fgf signaling, an optic cup still forms. To explore how eye morphogenesis occurs in such circumstances, we followed the movements of HGn42A:GFP-labeled temporal progenitors by in vivo imaging in FgfR-inh.-treated embryos. GFP mRNA in HGn42A embryos is ectopically found in the dorsal optic vesicle leaflet after FgfR-inh. treatment, similar to the effect on foxd1, but GFP protein maturation appears to lag behind, therefore allowing the tracking of ventral optic leaflet cells ( Figure 4J). At 20ss, more HGn42A:GFP-positive temporal progenitors have moved into the future neural retina domain of the outer optic cup layer in FgfR-inh.-treated embryos compared to controls ( Figure 4I; n = 7/7). This suggests that the loss of misspecified foxg1-negative, nasal progenitors in the optic cup after FgfR inhibition causes an enhanced movement of temporal progenitors from the ventral leaflet into the future neural retina domain.

Temporal Progenitor Cell Movement Can Occur Independent of Fgf-Mediated Patterning
To complement the analysis of morphogenetic movements in eyes lacking nasal identity, we created eyes that lack temporal identity by implanting Fgf8 beads adjacent to the nascent temporal retina, below the ventral optic vesicle leaflet ( Figure 5A and 5B). The movement of HGn42A:GFP-positive cells from the ventral optic vesicle leaflet into the outer optic cup layer is delayed after Fgf8 bead implantation compared to the control side of the same embryo ( Figure 5C; n = 8/9). However, perdurance of HGn42A:GFP expression shows that these cells eventually reach their normal axial position in the retina at 36 h ( Figure 5D; n = 12/12). Thus, the morphogenetic movement of temporal retina progenitor cells from the ventral optic vesicle leaflet around the distal ridge into the prospective neural retina can occur independent of correctly restricted foxg1 and foxd1 expression.

Foxg1 Is Required for Epithelial Cell Cohesion in the Optic Vesicle
The dependence of foxg1 expression upon Fgf signaling raised the possibility that Foxg1 may be a transcriptional mediator of some or all of the effects of Fgf signaling upon presumptive nasal progenitors. To test this idea, we studied foxg1 in loss-and gain-offunction assays.
Live imaging of optic vesicles in memGFP-labeled embryos, shows that abrogation of Foxg1 using a translation-blocking morpholino (foxg1MO; Figure S10) results in the delamination and accumulation of cells in the ventricle of the optic vesicle at 15ss (n = 7/8), when compared to controls injected with a 5-bp mismatch control morpholino (foxg1-5MM-MO, n = 8/8) ( Figure 6H, left two panels). This phenotype is highly similar to the delamination observed in fgf8/3/24 2/2 embryos and after FgfR inhibition ( Figure 6H, right two panels). Except for this phenotype and the previously reported changes during telencephalic development [35], this foxg1 morpholino does not create any morphological defects (unpublished data), consistent with its specificity and the restriction of foxg1 expression to the developing forebrain at this stage. Live imaging reveals that only foxg1MO-injected DsRed2 donor cells delaminate when transplanted into noninjected HRAS-GFPexpressing hosts (n = 9/10); neither host cells nor DsRed2-positive control transplanted cells (n = 10/10), nor foxg1-5MM-MOinjected cells show this phenotype ( Figure 6A-6C). Next, we performed clonal overexpression by transplanting foxg1cherryinjected DsRed2 donor cells into noninjected HRAS-EGFP hosts. Compared to clones of noninjected cells, which scatter by mixing with host cells, foxg1cherry-overexpressing cells form highly coherent clusters, with hardly any host cells intermingling at 10ss ( Figure 6D and 6E; n = 15/15). Together, these results suggested that Foxg1 may act downstream of Fgf signaling to promote epithelial cohesion of cells in the nascent nasal retina. Supporting this hypothesis, we found that Foxg1 could rescue the delamination of nasal retinal cells that occurs upon abrogation of Fgf signaling. In chimeras in which Fgf signaling is blocked, foxg1cherry-overexpressing cells are rescued from delamination (n = 12/12), when compared to host cells and noninjected transplanted cells (Figure 6F and 6G; n = 9/11). This strongly suggests that Foxg1 promotes cell cohesion in the dorsal optic vesicle leaflet, and that the delamination and death of misspecified nasal progenitors observed in the absence of Fgf signaling is due to the lack of Foxg1.

Foxg1 Promotes Cell Clustering at Sites of High Fgf-Signaling Activity
In chimeras carrying foxg1-overexpressing cells, we observed that the large majority of coherent clones were found in the dorsal forebrain, dorsal optic vesicle leaflet, head mesenchyme surrounding the optic vesicle, and olfactory primordium (unpublished data), all sites of high Fgf pathway activity. To explore this phenomenon, we tracked the lateral spreading/clustering of foxg1-overexpressing cell clones in response to exogenous Fgf provided from a bead.
For this purpose, first, 20-25 cells from memGFP-labeled donors, either overexpressing foxg1cherry or cherry (control) protein, were transplanted into nonlabeled hosts at sphere stage. Directly afterwards, beads, either coated with recombinant Fgf8 protein or PBS (control), were implanted next to the transplanted cell clone. Individual bead-implanted chimeras were then separately analyzed by live imaging at sphere (directly after implantation), bud (6 h after implantation), and 15ss (12 h after implantation) ( Figure 7A).
Directly after implantation (sphere), spreading of foxg1cherryoverexpressing clones in the presence of a PBS bead ( Figure 7B, top), foxg1cherry-overexpressing cells in the presence of an Fgf8 bead ( Figure 7C, top) and cherry-overexpressing cells in the presence of an Fgf8 bead ( Figure 7D, top) is very similar, when comparing the average of the measured maximal (d max ), measured minimal (d min ), and calculated median (d med ) distance of cells to the bead surface from four representative experiments for each condition ( Figure 7E, first radial plot). At bud, live imaging and measuring d max ( Figure 7E, second radial plot) of foxg1cherryoverexpressing cells in the presence of an Fgf8 bead ( Figure 7C, middle) show slightly decreased spreading, when compared to foxg1 overexpression with a PBS bead ( Figure 7B, middle) or cherry overexpression with an Fgf8 bead ( Figure 7D, middle). At 15ss, all analyzed foxg1cherry-overexpressing clones in the presence of an Fgf8 bead formed a tightly aggregated, single cluster at the site of bead implantation ( Figure 7C, bottom). Foxg1cherry-overexpressing clones in the presence of a PBS bead form several scattered clusters that do not coincide with the bead implantation site ( Figure 7B, bottom), and Fgf8 beads do not induce clustering of cherry-overexpressing clones ( Figure 7D, bottom). Analysis of d max at 15ss revealed a significant reduction for foxg1cherry-overexpressing clones in the presence of an Fgf8 bead compared to foxg1cherry-overexpressing clones in the presence of a PBS bead (p = 0.03305) or cherry-overexpressing clones in the presence of an Fgf8 bead (p = 0.00005) ( Figure 7E, third radial plot). Plotting d max over time shows that the spreading behavior of foxg1cherry-overexpressing clones in the presence of an Fgf8 bead diverges from bud stage onwards when compared to cherry-overexpressing cells in the presence of an Fgf8 bead. At 15ss, d max of foxg1cherry-overexpressing cells in the presence of a PBS bead shows a high degree of variability between individual clones, probably depending on their relative location to endogenous Fgf sources ( Figure 7F).
These results show that foxg1-expressing cells preferentially cluster around Fgf sources, suggesting a positive feedback between Fgf-dependent regulation of foxg1 gene expression and sustained cohesion of foxg1-expressing cells close to Fgf sources.

Stages and Dynamics of Nasal-Temporal Axis Development
Experiments in chick embryos have long suggested specification of the nasal-temporal axis of the retina at early stages of eye morphogenesis [36], and our previous work indicated that nasaltemporal axis formation requires early Fgf8 signaling [25]. We now show that a combined Fgf8/3/24 signal along the dorsalventral axis of the neural tube fully controls the nasal-temporal subdivision of the future neural retina already at the onset of optic vesicle evagination.
This unexpected, early orientation of the future nasal-temporal axis raises the question of how its final alignment with the anteriorposterior body axis is achieved. After the onset of evagination, the optic vesicle enters a phase of growth and lateral extension [37]. We show that during this process, the dorsal leaflet of the optic vesicle, which is facing the ectoderm, consists only of nasal retinal progenitors. Subsequently, the optic vesicle is transformed into the two-layered optic cup. At this stage, nasal retina progenitors in the outer layer gradually regress into their final axial position by epithelial compaction, concordant with late movement of temporal retina progenitors from the inner layer, around the distal ridge of the optic cup. The synchronization of these morphogenetic processes is crucial for the final nasal-temporal subdivision of the neural retina and could explain why fate-mapping experiments previously revealed an alignment of nasal-temporal cell positions along the medial-lateral axis of the optic cup [19].
Thus, formation of the nasal-temporal axis occurs as the results of three processes: (1) axis specification in the optic vesicle along the dorsal-ventral axis of neural tube, (2) morphogenetic axis reorientation by cell morphology changes and directed cell movements, and (3) 90u anterior rotation of the optic cup [37], leading to the alignment with the anterior-posterior body axis ( Figure 8A).

Fgf Spreading and Competence during Pattern Formation in the Optic Vesicle
All three Fgfs involved in nasal-temporal pattern formation originate asymmetrically relative to the future axis, consistent with their role in nasal fate specification in the dorsal optic vesicle leaflet. Fgf8 and -3 are both expressed in the dorsal forebrain [38,39] and fgf24 in cells of the olfactory placode [26,40].
The graded transformation of nasal into temporal retina fates upon stepwise elimination of fgf8/3/24 and the nested expression of Fgf target genes in the optic vesicle, suggest a morphogen-like mechanism with continuous determination of cell positional identities along an Fgf gradient. This would be similar to the mechanism of action of Fgf signaling during mesoderm development [41]. Such a gradient could arise by propagation of Fgf8 and Fgf3 within the neuroepithelium, since early, the optic vesicle is contiguous with the dorsal forebrain. In contrast, the source of Fgf24 lies outside the neural tube, and it thus must signal vertically through the basal side of the optic vesicle neuroepithelium. Bead implantations into the tissue surrounding the optic vesicle indicate that Fgf8 can also signal effectively via the basal side of the neuroepithelium. Interestingly, Fgf signaling occurs via the basal side of the epithelium during otic vesicle invagination [42]. We thus favor the model of an extra-neuroepithelial gradient of Fgf8/ 3/24 during optic vesicle patterning.
Bead implants show that the ventral optic vesicle leaflet-the future temporal retina-is also competent to respond to Fgfs. The spatially restricted response to Fgfs in the dorsal optic vesicle leaflet could, therefore, occur by limiting signal spreading. One possibility is that the tight apposition between cells of the prelens ectoderm and the distal optic vesicle ridge prevents Fgf spreading to ventral cells, thereby leading to a sharp signaling threshold that underlies the precise dorsal-ventral restriction of foxg1 and foxd1 expression in the optic vesicle.
Recently, Fgf19 has been suggested to act during lens and retina development, possibly downstream of Fgf8 and Fgf3 [43]. Fgf19 loss-and gain-of-function experiments can alter retinal gene expression levels, but fail to produce any consistent nasal-temporal patterning shift independent of severe abnormalities in optic cup morphology. Together with its unrestricted and late expression, this makes a specific contribution of Fgf19 to nasal-temporal patterning very unlikely. The previously reported cooperation of Fgf8 and Fgf3 during retinal neurogenesis [44] appears to be unrelated to the function of Fgf8/3/24 during early pattern formation and eye morphogenesis (unpublished data).

Roles of Foxg1 in Pattern Formation and Neural Progenitor Proliferation
Experiments in mouse and chick embryos have shown a mutually repressive interaction between foxg1 and foxd1 [10,45]. This is supported by the sharp foxg1/-d1 expression boundary at the distal ridge of the optic vesicle. Additionally, we find that foxg1 and foxd1 expression always occurs in a mutually exclusive pattern in both Fgf loss-and gain-of-function experiments. Similar to the mouse telencephalon [46,47], foxg1 in the optic vesicle may be directly activated by Fgf signaling. Since Fgf signaling only regulates the development of nasal fates, temporal fate specification and foxd1 expression in the ventral optic vesicle leaflet could either occur as default, in the absence of Fgf signals, and/or as response to a different signal, potentially of ventral origin.
Foxg1 and -d1 are likely to act as direct regulators of efn/eph guidance cue expression [10,11]. Although Efna5a and Epha4b are guidance cues for retinal ganglion cell axons, we show that their spatially restricted expression is already established in the optic vesicle, shortly after that of foxg1 and foxd1 and long before neuronal differentiation and retinotectal map formation. Given that Efns/Ephs are known effectors of cell sorting at lineagerestricted compartment boundaries [48], a potential, early role for Efn/Eph signaling could be maintenance of a nasal-temporal lineage boundary between cells in the dorsal and ventral leaflets of the optic vesicle during growth and morphogenesis. Indeed, the presence of such lineage-restricted compartments along the dorsalventral axis of the retina has previously been suggested by fate mapping experiments in chick embryos [49].
Foxg1 is best studied for its function in maintaining neural progenitor proliferation [27][28][29]. We find that the optic vesicle contains 2-fold more mitotic, foxg1-positive, nasal retina progenitors than mitotic, foxd1-positive, temporal retina progenitors. This asymmetry depends on Fgf signaling and coincides with the strict Fgf requirement for foxg1 expression. One role for foxg1 could thus be maintenance of a high rate of neural progenitor proliferation, close to the source of Fgfs, similar to the role proposed for Fgfs in the vertebrate spinal cord stem zone [50].

Epithelial Cohesion and Cell Movements during Optic Cup Morphogenesis
Changes in motility and/or shape of individual cells must be tightly balanced with cell-cell adhesion to assure tissue integrity during epithelial morphogenesis, but how positional identities defined during pattern formation contribute to regional differences in morphogenetic cell behavior is poorly understood [51,52]. We find that local cell behavior and epithelial integrity during optic cup morphogenesis directly depend on correct prior patterning of the optic vesicle by Fgfs.
When foxd1-expressing, temporal progenitors move into the optic cup, foxg1-expressing, nasal progenitors increasingly regress dorsoproximally, leading to the definitive alignment of nasaltemporal fates in the primordium of the neural retina ( Figure 8B, left). In the course of this morphogenetic movement, nasal progenitors appear increasingly immotile and elongated, whereas temporal progenitors in the inner optic cup layer appear motile and more cuboidal. There are two plausible, but not exclusive, explanations for the driving force behind the morphogenetic cell movement and cell shape changes in the optic cup: on one hand, an active movement of temporal cells could exert a force within the plane of the optic cup neuroepithelium, leading to the gradual compaction of nasal progenitors already residing in the future neural retina domain (pushing force). Fgf8 bead implants indicate that the movement of cells from the ventral optic leaflet into the neural retina can occur independent of their Fgf-dependent nasaltemporal fate. Thus, if this movement is active, it appears to be under the control of a yet unidentified factor.
Alternatively, since the optic cup epithelium is overall highly coherent, the compaction of future nasal tissue could tow temporal progenitors into the neural retina, independent of active, lateral cell movements (pulling force). The higher rate of temporal progenitor movement observed in the absence of Fgf signaling suggests that intact cohesion of nasal progenitors at least restricts morphogenetic movements of temporal cells. Evidently, compaction by cell elongation can only occur if the nasal progenitors, already residing in the future neural retina domain, have a high degree of lateral cohesion. In support of this, we find that in the absence of Fgf signaling and Foxg1 activity, nasal progenitors start to delaminate exactly at the onset of temporal progenitor movement into the optic cup ( Figure 8B, right). Together with the ability of Foxg1 to rescue delamination in the absence of Fgf signaling and the strong effect of Foxg1 overexpression on lateral cell spreading/clustering, this suggests a novel role for Foxg1 as a positive regulator of neuroepithelial cell cohesion.
When Foxg1 is ectopically expressed, cell cohesion is more pronounced in regions of high Fgf signaling activity. There could be several explanations for this. First, Fgf signaling might enhance Foxg1-dependent foxg1 expression, translation or posttranslational efficacy, such that Foxg1 function is more effective in an Fgfsignaling environment. Indeed, it has been recently shown that Fgf signaling posttranslationally regulates the subcellular localization of Foxg1 [53], indicating a role for persistent Fgf signaling on Foxg1 function. Furthermore, there could be a feed-forward mechanism whereby Foxg1 enhances an independent role for Fgf signaling in mediating cell cohesion ( Figure 8C). In support of this, recent results suggest a requirement for Fgf signaling in epithelial organization during collective cell migration in the lateral line primordium [54,55] and perhaps to a lesser extent in the parapineal nucleus [56]. In the lateral line primordium, Fgf signaling may promote epithelialization and the formation of apical junctional complexes between the polarized epithelial cells [54]. A failure in maintaining neuroepithelial integrity and junctional complexes could certainly contribute to the observed extrusion of prospective nasal cells in the absence of Fgf signaling. Additionally, chemotactic effects of Fgfs [57] could limit nasal progenitor spreading and thereby indirectly lead to epithelial compaction. The future identification of Fgf-and Foxg1-regulated effectors of cell adhesion will advance the understanding of this mechanism. The phenotype of fgf8/3/24 2/2 embryos is reminiscent of the general loss of neuroepithelial integrity in N-cadherin mutants [58,59]. It is thus possible that Fgf-dependent foxg1 expression is locally required for cadherin-mediated cell adhesion. However, a disruption of apical-basal cell polarity does not appear to be the cause of cell delamination in fgf8/3/24 2/2 .
In summary, by in vivo tracking, the development of one retinal axis from its specification, through morphogenetic rearrangement until its final orientation, we show that retinal pattern formation and morphogenesis are tightly coordinated processes. Considering the differences of optic vesicle morphology between vertebrate species [60,61], it will be important to assess the conservation of morphogenetic axis reorientation. In first support of this, a recent fate-mapping study suggests that nasal-temporal cell positions initially align along the dorsal-ventral neural tube axis also in chick embryos [62]. Interestingly, in frog embryos, late cell movement from the optic stalk contributes to formation of the ventral neural retina [63], suggesting comparable cell movements may shape the retina along both of its major axes.

Fish Lines and Maintenance
Fish were maintained and bred according to standard procedures [64]. AB or tupl wild-type and ace ti282a (acerebellar/fgf8), ika t22030 (ikarus/fgf24), and fgf3 t24149 mutant fish were used for intercrossing and to breed [38,65,66]. Adult carrier fish and mutant embryos were identified by direct sequencing after PCR on genomic DNA, using the following primers:  [31][32][33][34]67], and a line expressing DsRed2 under the control of the Xenopus ef1a promoter [68]. The cldnb gene is not expressed in the retina, and nasal retina expression is only present in the single Tg(-8.0cldnb:lynGFP)zf106 line (D. Gilmour, personal communication), thus representing a positional effect. The nature of the trapped enhancer is currently unknown.

Cell Transplantation and Heat-Shock Treatment
All cell transplantations were carried out between the 40% and 50% epiboly stages at the animal pole. Dnfgfr1-expressing cell clones were created by transplantation of cells from Tg(hsp70l:dnfgfr1-EGFP)pd1 embryos into the animal pole of wt embryos at late blastula stages. Chimeras were heat-shock induced by transfer into 37uC E3 medium at the 3ss stage and subsequently incubated at 28.5uC. Controls chimeras carrying clones from a hsp70l:eGFP transgenic line (S. Hans, unpublished data) were treated identically.

Qualitative Profiling of Gene Expression
Sets of 8-bit grayscale images of dissected, flat-mounted eyes were captured after in situ hybridization and imported into ImageJ 1.32j (http://rsb.info.nih.gov/ij/). Using the ''Analyze.Mean'' option, mean intensities in eight axial, 30-mm 2 regions were measured and further analyzed with Microsoft Excel. Analysis was done on identically processed embryos.

BrdU Incorporation and Detection
BrdU treatments were done by incubating the embryos in 5 mg/ml BrdU and 15% DMSO, for 20 min on ice followed by 10 min at 28.5uC just prior to fixation. BrdU detection was done as previously described [70].

Microscopy
Live embryos were imaged with an upright Leica TCS SP5 confocal microscope using a 636dipping lens after immobilization in 1.2% LMP agarose in embryo medium. A 406 UV-corrected lens was used for imaging fluorescently stained embryos. Image analysis and assembly was done with ImageJ, Metamorph, Volocity, and Leica LAS software. Injection of 25-50 pg in vitro-transcribed palmitoylated mRFP [71] or lynGFP RNA/ embryo was used as in vivo membrane counterstain.

Foxd1 Nomenclature
At the time of this study, the zebrafish genome contained one gene annotated as foxd1 (GenBank accession number: NM_131271.1) and one gene annotated as foxd1 like (GenBank accession number: NM_212913.1). Phylogenetic sequence analysis revealed that zebrafish foxd1 like is orthologous to other vertebrate foxd1 genes, whereas zebrafish foxd1 is orthologous to foxd2 (Figures S11 and S12). Therefore, foxd1 like is referred to as foxd1 throughout this study. Figure S11 Phylogenetic tree analysis of vertebrate Foxd1/2/3. Maximum likelihood phylogeny of Foxd1/2/3 sequences from Homo sapiens, Mus musculus, Gallus gallus, Xenopus laevis, and Danio rerio as determined by PHYML (Guindon and Gascuel, 2003 [72]) shows that zebrafish Foxd1-like (light red shading) is orthologous to other vertebrate Foxd1 genes, whereas zebrafish Foxd1 (light blue shading) is orthologous to other vertebrate Foxd2 genes (PHYML parameters if not default: bootstrapping = 1,000 pseudo datasets; transition ratio and proportion of invariable sites = estimated; number of substitution categories = 8; gamma distribution parameter = estimated; only bootstrap values.900 are shown, based on alignment in Figure S12). Found at: doi:10.1371/journal.pbio.1000214.s011 (0.14 MB TIF) Figure S12 Trimmed multiple sequence alignment of vertebrate Foxd1/2/3 proteins. A multiple alignment of known zebrafish, human, mouse, chick, and frog Foxd1/2/3 protein sequences was calculated with MAFFT (Katoh et al., 2005 [73]) (623 aligned amino acids [aa]) and trimmed with GBlocks (Talavera and Castresana, 2007 [74]) (232 aligned aa) before phylogenetic tree calculation (see Figure S11). The following GenBank sequences were used: D. rerio Foxd1-like: NP_998078, D. Video S1 Dynamic restriction of cldnb:GFP-expressing cells in the optic cup. A Tg (-8.0cldnb:lynGFP)zf106 transgenic embryo (green), colabeled with membrane-targeted RFP (red) was imaged between 18-and 24ss. Cells expressing cldnb:GFP first cover the complete outer optic cup layer and then get gradually localized to the dorsal half -the future nasal retina. Cross-section through one half of the forebrain, dorsal is up and lateral to the left. Frame interval: 10 min. Number of frames: 17. Each frame is a single confocal section.