Deficient FGF signaling causes optic nerve dysgenesis and ocular coloboma

Eye development follows an elaborate sequence of morphogenetic events, which are highly conserved between humans and mice. At mouse embryonic day 9.5 (E9.5) when the proximal optic vesicle connected to the diencephalon forms the optic stalk (OS), the distal optic vesicle contacting the head surface ectoderm folds into two layers: an outer layer of presumptive retinal pigmented epithelium (RPE) and an inner layer of developing neural retina (NR) (Bharti et al., 2006; Fuhrmann, 2010). This remodeling of optic vesicle into a cup-like structure at E10.5 leaves a transient groove opening at the ventral side of the eye, termed the optic fissure (OF), which will be progressively sealed from the proximal to the distal end through fusion of the lip areas of the ventral optic stalk and neural retina until E12.5 (Silver and Robb, 1979; Chang et al., 2006; Morcillo et al., 2006). The interface between optic stalk and neural retina, however, will be specified as optic disc (OD), a unique ocular structure where blood vessels enter the eye. Moreover, although optic disc is immediately adjacent to neuronal progenitor cells in neural retina, it consists exclusively of astrocyte progenitor cells (APCs), which express multiple axon guidance molecules. These extracellular cues direct the fasciculated axons of retinal ganglion cells to exit the eye at the optic disc as the optic nerve (ON), which connects the eye to the brain (Deiner et al., 1997; Dakubo et al., 2003; Morcillo et al., 2006; Bao, 2008). It has been proposed that aberrant closure of the optic fissure and defective formation of the optic disc in human fetuses can cause ocular coloboma and optic nerve dysgenesis, leading to childhood blindness (Gregory-Evans et al., 2004; Chang et al., 2006).

Multiple signaling pathways have been implicated in ocular coloboma and optic nerve hypoplasia. Early midline-derived Shh governs the proximal-distal patterning of optic vesicle by regulating Pax2 and Pax6 expression, which are necessary for the demarcation of the optic stalk, optic disc and neural retina (Macdonald et al., 1995; Chiang et al., 1996). This is followed after the onset of retinal neurogenesis by retinal ganglion cell-derived Shh, which further promotes proliferation of the optic disc and optic stalk (Dakubo et al., 2003). Studies of knockout mice reveal that Bmp4 expressed in the dorsal optic vesicle is crucial for dorsal-ventral patterning, whereas Bmp7 expressed in the ventral optic vesicle and the neighboring periocular mesenchyme is required for initiation of the optic fissure and specification the optic disc (Furuta and Hogan, 1998; Morcillo et al., 2006). Similarly, RA signaling has also been proposed to regulate the closure of the optic fissure by controlling ventral retina morphogenesis and periocular mesenchyme development (Matt et al., 2005; Molotkov et al., 2006; Matt et al., 2008; Lupo et al., 2011). Finally, mutations in Wnt signaling genes Lrp6, frizzled 5 (Fzd5) and β-catenin (Ctnnb1), which are known to regulate the dorsal-ventral patterning of optic vesicle and the specification of the RPE, all cause ocular coloboma (Liu et al., 2007; Liu and Nathans, 2008; Zhou et al., 2008; Fujimura et al., 2009; Westenskow et al., 2009). Many of these coloboma mutants, as well as human patients, exhibit dysregulation of Pax2, a key homeodomain protein required for the closure but not the initiation of the optic fissure (Torres et al., 1996; Alur et al., 2010; Cross et al., 2011). In addition, mutations in ocular transcription factors, e.g. Vax2, Chx10 and Mitf, also result in ocular coloboma (Robb et al., 1978; Silver and Robb, 1979; Scholtz and Chan, 1987; Hero, 1989; Torres et al., 1996; Barbieri et al., 2002; Mui et al., 2005). These animal models provide a framework within which to unravel the etiology of human coloboma and optic nerve defects.

Previous studies in chick and zebrafish models suggest that FGF signaling plays essential roles in eye development, but thus far none of the murine Fgf mutants have shown significant retinal defects. We recently reported that protein tyrosine phosphatase Shp2 (Ptpn11), a downstream mediator of receptor tyrosine kinase signaling, is required for embryonic neural retina specification and adult retinal survival (Cai et al., 2010; Cai et al., 2011). To further explore the signaling pathway upstream of Shp2, we have now generated a conditional knockout of two FGF receptors, Fgfr1 and Fgfr2, specifically in optic vesicle, which resulted in ocular coloboma and optic nerve dysgenesis. Although BMP and midline Shh signaling appear unaffected, Fgfr1/2 deletion disrupts Pax2 expression in the developing optic fissure and optic disc, which instead acquire ectopic expression of RPE marker Mitf. This failure of optic fissure and optic disc specification is also phenocopied by the specific deletion of Frs2a and Shp2, and by mutations that disrupt Fgfr-Frs2α or Frs2α-Shp2 interactions. By contrast, both Fgfr1/2 and Frs2α/Shp2 mutants can be rescued by constitutively active Ras-ERK signaling. Our results thus demonstrate a novel role for FGF-Frs2α-Shp2-Ras-ERK signaling in the development of the optic fissure, optic disc and optic nerve.

Frs2a^flox, Frs2a^2F, Shp2^flox mice have been previously reported (Zhang et al., 2004; Lin et al., 2007; Gotoh, 2008). Fgfr1^ΔFrs mice were derived from the recombinant ES cells kindly provided by Philippe Soriano (Mount Sinai School of Medicine, New York, NY, USA) (Hoch and Soriano, 2006). Fgfr1^flox mice were obtained from the Jackson Laboratory (Bar Harbor, ME, USA) (Hoch and Soriano, 2006). Fgfr2^flox mice were kindly provided by Dr David Ornitz (Washington University Medical School, St Louis, MO, USA) (Yu et al., 2003). Six3-Cre mice were kindly provided by Dr Yasuhide Furuta (MD Anderson Cancer Center, Houston, TX, USA) (Furuta et al., 2000). Rx-Cre mice were kindly provided by Dr Milan Jamrich (Baylor College of Medicine, Houston, TX, USA) (Swindell et al., 2006). LSL-Kras^G12D mice were obtained from the Mouse Models of Human Cancers Consortium (MMHCC) Repository at National Cancer Institute (Tuveson et al., 2004). All mice were maintained in mixed genetic background. Six3-Cre transgene alone or in combination with heterozygous Fgfr1^flox, Fgfr2^flox, Frs2α^flox and Shp2^flox alleles did not cause any phenotype (data not shown) and they were used as wild-type controls. All experiments were performed in accordance with institutional guidelines.

Mouse embryos were staged according to the day the vaginal plug was first observed. The samples were fixed in 4% PFA overnight for frozen or paraffin wax-embedded sections before standard immunohistochemistry was performed as previously described (Pan et al., 2006; Pan et al., 2008). Serial sections (frontal or sagittal sections for ∼E9.5-E11.5 embryos, transverse or sagittal sections for E13.5 embryos) were always stained with anti-Pax2 to identify optic stalk, optic disc and the lip region of the ventral neural retina (optic fissure). We performed regular double fluorescent staining for Pax2/Mitf or Pax2/NF165 and Tyramide Signal Amplification (TSA)-fluorescent staining for Shp2 and phospho-ERK (TSA Plus Fluorescein System, PerkinElmer, Waltham, MA, USA) (Cai et al., 2010). The antibodies used were anti-Pax2 (1:200, PRB-276P) and anti-Pax6 (1:100, PRB-278P) (both from Covance, Berkeley, CA, USA), anti-Mitf (1:100, #MS-771-P0, Thermo Scientific, Fremont, CA, USA), anti-Sox2 (1:100, #Ab5603, Chemicon, Temecula, CA, USA), anti-Shp2 (1:3000, Sc-280, Santa Cruz Biotechnology, Santa Cruz, CA, USA), anti-cleaved-caspase 3 (1:100, #9664), anti-phospho-ERK1/2 (1:1000, #9101) and anti-cyclin D1 (1:100, #2926) (all from Cell Signaling Technology, Beverly, MA, USA), anti-Frs2α (1:1000, ab10425, Abcam, Cambridge, MA, USA), anti-Ki67 (1:10, #550609, BD Pharmingen San Diego, CA, USA) and anti-NF165 (1:100, #2H3, Developmental Studies Hybridoma Bank, University of Iowa, Iowa City, IA, USA). Anti-phospho-Smad1 (PS1) antibody [1:100, kindly provided by Peter ten Dijke (Leiden University Medical Center, Leiden, The Netherlands) and Carl-Henrik Heldin (Ludwig Institute for Cancer Research, Uppsala, Sweden)]. The secondary antibodies used were goat anti-mouse conjugated with Cy2 and goat anti-rabbit conjugated with Cy3 (both from Jackson Laboratory, West Grove, PA, USA). The fluorescent staining was examined under a Leica DM500 fluorescent microscope equipped with Cy2 and Cy3 filters. The cell proliferation and apoptosis rates were calculated as the ratio of Ki67 and TUNEL-positive cells versus DAPI-positive cells, and analyzed by the Student’s t-test. For analysis of each marker, at least three embryos of each genotype were fully sectioned and stained.

RNA in situ hybridization on cryosections was carried out as previously described (Pan et al., 2006; Cai et al., 2010). After hybridization treatment, the sectioned samples were counterstained with the Pax2 antibody to determine the region of the optic stalk, optic disc and ventral neural retina. Probes for the following genes were used: Crx, netrin 1, Shh and Vax2 (from Dr Valerie Wallace, Ottawa Health Research Institute, Ottawa, Ontario, Canada); BF2 (Foxd1 - Mouse Genome Informatics), Bmp7, Bmp4 and Raldh3 (Aldh1a3 - Mouse Genome Informatics) (from Dr Robert Heuckeroth, Washington University Medical School, St Louis, MO, USA); Otx1 and Rx (Rax - Mouse Genome Informatics) (from Dr Naoki Takahashi, Nara Institute of Science and Technology, Nara, Japan); BF1 (Foxg1 - Mouse Genome Informatics) (from Marie Fernandes, Albert Einstein College of Medicine, Bronx, NY, USA); Six3 and Tbx5 (from Dr Guillermo Oliver, St Jude Children’s Research Hospital, Memphis, TN, USA); Vax1 (from Dr Kapil Bharti, National Institutes of Health, Bethesda, MD, USA); Nlz2 (Zfp503 - Mouse Genome Informatics) (from Dr Brian Brooks, National Institutes of Health, Bethesda, MD, USA); Fgf15 (from Dr Minh-Thanh T. Nguyen, National Institutes of Health, Bethesda, MD, USA); Math5 (Atoh7 - Mouse Genome Informatics) (from Dr Tom Glaser, University of Michigan, Ann Arbor, MI, USA); Brn3b (Pou4f2 - Mouse Genome Informatics) (from Dr Lin Gan, University of Rochester, Rochester, NY, USA); and Chx10 (Vsx2 - Mouse Genome Informatics) (from Dr Roderick R. McInnes, Hospital for Sick Children, Toronto, Ontario, Canada). A 1.8 kb fragment of Frs2α (Frs2 - Mouse Genome Informatics) exon 5 was used as an in situ probe to confirm the deletion of Frs2α gene. At least three independent and fully sectioned eye balls of each genotype were analyzed for each RNA probe.

Previous RNA in situ hybridization experiments show that both Fgfr1 and Fgfr2 are expressed in embryonic retina, whereas transcripts for Fgfr3 and Fgfr4 appear undetectable (Visel et al., 2004). We thus used Six3-Cre to ablate Fgfr1 and Fgfr2 in optic vesicle to study FGF signaling in eye development (Furuta et al., 2000). We first carefully investigated the timing and specificity of Six3-Cre using the R26R reporter. As we previously reported, Six3-Cre;R26R embryos exhibited only sparse X-gal staining in the optic vesicle at E9.0 (Cai et al., 2010). At E9.5, however, strong X-gal staining was first observed in the ventral optic vesicle, the precursor region of the optic fissure, optic disc and ventral neural retina (supplementary material Fig. S1A-C). By contrast, the dorsal optic vesicle did not stain strongly for X-gal until E10.5 (supplementary material Fig. S1D-F). After that, transverse sections of the Six3-Cre;R26R embryos indicated that Six3-Cre activity was concentrated in the central retina, optic stalk and optic chiasm (Cai et al., 2010; Cai et al., 2011).

Fgfr1 and Fgfr2 single mutants were viable and fertile without ocular defects (data not shown). Examination of the adult Six3-Cre;Fgfr1^flox/flox;Fgfr2^flox/flox double mutants, however, revealed a ‘tear-drop’-shaped iris, which mimics uveal coloboma found in humans (Fig. 1A,E) (Chang et al., 2006). More strikingly, the optic nerve was either extremely thin or completely missing from the back of the eye (Fig. 1B,F, arrow; Table 1). At E13.5, when the control optic fissure is completely fused, there still remained a significant ventral cleft in the mutant optic cup (Fig. 1C,D,G,H, arrow). These results suggest that FGF signaling is required for the closure of the optic fissure and the formation of the optic nerve.

We next examined the optic fissure and optic nerve defects by preparing sagittal sections of the distal and proximal eye, respectively (see Fig. 1O). Mitf is a crucial regulator of RPE, whereas Pax2 is required for ventral neural retina development and optic fissure closure (Torres et al., 1996; Horsford et al., 2005). In the E13.5 control eye, Pax2 expression marked the fusion of optic fissure in the ventral optic cup, whereas Mitf was restricted to the RPE (Fig. 1I, arrow). In the Fgfr1/2 mutants, however, Mitf was induced in the open optic fissure, where Pax2 expression was significantly downregulated (Fig. 1L, arrow). Further examination of proximal eye sections showed that the mutant optic stalk, marked by Pax2 and netrin 1 expression, was significantly reduced in size (Fig. 1J,M). Although Mitf was no longer ectopically expressed, levels of the axonal marker NF165 were greatly diminished and mislocalized in the proximal eye (Fig. 1K,N; data not shown), consistent with the loss of the optic nerve in the Fgfr1/2 mutants. As shown in Fig. 1V, we also examined transverse sections of the optic disc, which is necessary for the guidance of retinal ganglion cell axons (also known as optic nerve fibers) to exit the eye as the optic nerve. Instead of expressing Pax2 and the axon guidance gene netrin 1, the presumptive optic disc region in the Fgfr1/2 mutants was severely deformed with ectopic Mitf expression (Fig. 1P,Q,S,T, arrows). As a result, the NF165-stained optic nerve fibers were misrouted into the sub-retinal space, failing to form a properly bundled optic nerve (Fig. 1R,U, arrows and arrowheads). Additional transverse sections of the ventral eye consistently revealed gaps in ventral retina (data not shown), confirming a complete penetrance of coloboma in these mutants (Table 1). In 3-week-old animals, bright-field microscopy even revealed ectopic pigmentation in the mutant optic fissure and presumptive optic nerve head (Fig. 1W, arrows). Similar coloboma and optic disc defects were also observed in Rx-Cre;Fgfr1^flox/flox; Fgfr2^flox/flox double mutants, where Fgfr1 and Fgfr2 were ablated using another retina-specific Cre driver, Rx-Cre (supplementary material Fig. S2) (Swindell et al., 2006). Taken together, our histological and molecular analyses suggest that ectopic pigmentation of optic fissure in Fgfr mutants caused ocular coloboma by preventing the fusion of the ventral optic vesicle. Similar conversion of the mutant optic disc region to RPE tissue abolished the proper guidance and assembly of optic nerve, causing optic nerve dysgenesis.

To determine the molecular mechanism of Fgfr1/2 mutant defects, we next investigated the extensive transcriptional network underlying retinal differentiation. Rx and Six3 are transcription factors required for eye field specification and Otx1 marks RPE, optic disc and ciliary margin. Their expressions in the Fgfr1/2 mutants expanded into the presumptive optic stalk (Fig. 2A-F′). In E13.5 control embryos, retinal progenitor marker Sox2 labeled the neural retina and optic disc tissue, whereas Pax6 was present only in the neural retina and RPE. In the Fgfr1/2 mutants, however, the presumptive optic disc region exhibited ectopic Pax6 expression at the expense of Sox2 expression (Fig. 2G-J′, arrows). Finally, Fgf15 and Chx10 were specific to retinal progenitor cells, which express Isl1, Math5, Brn3b and Crx to promote neuronal differentiation. These retinal-specific markers were excluded from the presumptive optic disc and neural retina adjacent to the optic fissure in the Fgfr1/2 mutants (Fig. 2K-T′, arrows; data not shown). Taken together, these results further confirmed that the ocular defects in the Fgfr1/2 mutants were caused by disruption of optic disc and optic fissure specification.

Shh and BMP signaling are known to play important roles in early optic vesicle patterning, which have been implicated in coloboma formation. In the E11.5 Fgfr1/2 mutants, there was no change in expression of Shh and its target gene Gli1 (Fig. 3A-C′, arrows). Similarly, Bmp4 and its downstream effector phospho-Smad1 were maintained in dorsal optic vesicle and Bmp7 was correctly expressed ventrally (Fig. 3D-F′, arrows). Consistent with this, the axial polarity markers Vax1, Tbx5, Vax2, Raldh3, Nlz2, BF1 and BF2 were all expressed normally (Fig. 3G-L′, arrows; data not shown), suggesting that the proximal-distal, dorsal-ventral and temporal-nasal axes were maintained in the Fgfr1/2 mutants. The optic vesicle patterning genes Rx, Six3, Pax6 and Otx1 also appeared unaffected at this stage (Fig. 3M-P′, arrows). By contrast, the neural retinal marker Chx10 was already reduced in the ventral retina adjacent to the optic fissure (Fig. 3Q,Q′, arrows). Importantly, whereas control optic fissure exclusively expresses Pax2 at E11.5, the RPE marker Mitf was ectopically expressed by the Pax2-positive cells in the mutant optic fissure (Fig. 3R,R′, arrows). As shown above in Fig. 1, Fgfr1/2 mutant optic fissure eventually lost all Pax2 expression and acquired ectopic pigmentation. This apparent transition from Pax2/Mitf co-expressing progenitors to purely Mitf-positive RPE in ventral retina cannot be explained by simple infiltration of RPE in the wake of the failed fusion of optic cup, but instead it points to a genuine fate change of neural retina to RPE tissue. Collectively, these results suggested that the lack of Fgfr1/2 signaling disrupted the development of the optic fissure without affecting the general patterning of the optic vesicle.

To further understand the molecular mechanism of FGF signaling in eye development, we next explored its intracellular signaling cascade. As early as E10.5, before the onset of Mitf misexpression, a clear reduction in phospho-ERK staining was observed in the ventral optic vesicle, the presumptive optic disc, the ventral optic stalk and the optic chiasm (Fig. 4A-D,G-J, arrows). At E13.5, when control embryos exhibited strong elevation in ERK phosphorylation in the Pax2-positive optic disc, Fgfr1/2 ablation abolished both phospho-ERK and Pax2 staining in the presumptive optic disc region (Fig. 4E,F,K,L, arrow). ERK signaling is also known to promote cell proliferation and survival. Consistent with this, the presumptive optic disc region in the Fgfr1/2 mutants showed a loss of cell cycle regulator cyclin D1 and cell proliferation marker Ki67, but there was a significant increase in apoptotic marker cleaved caspase 3 and TUNEL staining at E11.5 and E13.5 (Fig. 4M-T′, arrows). These results supported the theory that Fgfr1/2 was required for ERK signaling in the optic fissure and optic disc.

Tyrosine phosphorylation of Frs2 by FGF receptors is known to create docking sites for the adaptor protein Grb2 and protein tyrosine phosphatase Shp2, both of which are important for activating Ras-ERK signaling. Surprisingly, although we confirmed the Six3-Cre-mediated ablation of Frs2a both by RNA in situ hybridization and immunohistochemistry (supplementary material Fig. S3A-D, arrows), ERK phosphorylation was unchanged in the E10.5 Six3-Cre;Frs2a^flox/flox optic vesicle (supplementary material Fig. S3E,I, arrows). Phosphorylated ERK was eventually downregulated in the central retina at E13.5, but the Pax2-positive optic disc and NF165-positive optic nerve were both formed (supplementary material Fig. S3F-H,J-L, arrows). Consistent with this, no ocular coloboma was observed in Six3-Cre;Frs2a^flox/flox animals (data not shown). As we have previously reported (Cai et al., 2010), Six3-Cre;Shp2^flox/flox mutant similarly did not display any defects in embryonic development, including the closure of optic fissure and the formation of optic disc (supplementary material Fig. S3M-P, arrows; data not shown).

To further investigate the functional significance of Frs2a and Shp2 in retinal development, we next generated a combined deletion of these two genes. Interestingly, the Six3-Cre; Frs2α ^flox/flox; Shp2^flox/flox mutants now reproduced the ocular coloboma and optic nerve dysgenesis phenotype observed in the Six3-Cre; Fgfr1^flox/flox;Fgfr2^flox/flox mutants (Fig. 5A-E′, arrows; Table 1). By immunostaining, we confirmed that both Shp2 and phospho-ERK were lost in the ventral optic vesicle at E10.5 (Fig. 5F-G′, arrows). In the E13.5 frontal sections, the Frs2a/Shp2 mutants appeared to maintain the correct dorsal-ventral polarity as indicated by the ventral expression of Vax2, but the Pax2-expressing optic disc structure was absent (Fig. 5H,H′). Similarly, ectopic Mitf expression was again induced in the ventral optic cup and the NF165-labeled optic nerve was severely hypoplastic (Fig. 5I-J′, arrows). Moreover, transverse sections of the Frs2a/Shp2 mutants showed that the presumptive optic fissure and optic disc, and the adjacent ventral retina regions lost the retinal progenitor markers Sox2, Fgf15 and Chx10, the optic disc markers Pax2 and netrin 1, and cell cycle markers Ki67 and cyclin D1 (Fig. 5K-O′, arrows). These results demonstrated that Frs2a and Shp2 were essential for retinal development.

The above results show that ocular coloboma appeared only in the Frs2a/Shp2 double mutants but not in the single mutants, which seemed to suggest that these two genes act redundantly in ocular development. That is, Frs2a can activate downstream ERK signaling even in the absence of Shp2, and vice versa. However, an alternative explanation is that Frs2a and Shp2 are indeed both essential for transmitting FGF signaling, but the conditional knockout of either Frs2a or Shp2 resulted in only a gradual loss of either protein, which by itself may not be rapid enough to abolish downstream ERK signaling before the optic disc and optic fissure are developed. In this model, simultaneous depletion of both the Frs2α and Shp2 proteins in the Frs2a/Shp2 double mutants would accelerate the disruption of ERK signaling, resulting in ocular coloboma. This synergistic effect is thus similar to the classic genetic interaction between two hypomorphic systemic alleles on the same pathway, which produces a far more severe phenotype when combined.

To test this hypothesis, we used the well-characterized Frs2a^2F allele, which contained two point mutations in Frs2a that disrupted the Shp2 binding (Fig. 6A) (Gotoh et al., 2004). We reasoned that if Frs2α could act independently of Shp2, disabling the Frs2α and Shp2 interaction in the Six3-Cre;Frs2a^2F/flox;Shp2^flox/flox mutants would not produce a more severe phenotype than the Six3-Cre; Shp2^flox/flox single mutants. However, if Frs2α must recruit Shp2 to activate downstream signaling, the Six3-Cre;Frs2α^2F/flox;Shp2^flox/flox mutants would phenocopy the same ocular coloboma defects in the Six3-Cre;Frs2α^flox/flox;Shp2^flox/flox mutants. As shown in Fig. 6B-P, the Six3-Cre;Frs2α^2F/flox;Shp2^flox/flox mutants indeed lost both Shp2 and phospho-ERK staining in the central retina. As a result, Mitf was ectopically induced in the presumptive optic disc region, whereas Pax2, netrin 1 and NF165 were downregulated. Although the adult Six3-Cre;Frs2α^2F/flox animals did not display any eye phenotype (data not shown), the Six3-Cre;Frs2α^2F/flox;Shp2^flox/flox mutants presented with ocular coloboma and optic nerve dysgenesis (Table 1). The phenotypic similarities between the Six3-Cre;Frs2α^2F/flox;Shp2^flox/flox and the Six3-Cre;Frs2α^flox/flox; Shp2^flox/flox mutants support the hypothesis that the recruitment of Shp2 by Frs2α is essential in mediating FGF signaling during retinal development.

To further confirm the functional significance of Frs2 in FGF signaling, we next took advantage of a previously characterized Fgfr1^ΔFrs allele (Hoch and Soriano, 2006), in which the Frs2 binding site was deleted in the Fgfr1 cytoplasmic domain (Fig. 7A). By generating the Six3-Cre;Fgfr1^ΔFrs2/flox;Fgfr2^flox/flox mutant, we essentially created a retinal specific Fgfr1^ΔFrs mutant in a Fgfr2-null background. This strategy allows us to probe Fgfr-Frs2 interaction in a tissue-specific manner. The Six3-Cre;Fgfr1^ΔFrs2/flox;Fgfr2^flox/flox mutants completely phenocopied the Six3-Cre;Fgfr1^flox/flox;Fgfr2^flox/flox-null mutants (Table 1). As shown in Fig. 7B-E, phospho-ERK staining was abolished in the presumptive optic disc region in the E13.5 Six3-Cre;Fgfr1^ΔFrs2/flox;Fgfr2^flox/flox mutants. Mitf was ectopically induced at the expense of Pax2 expression, which led to misrouting of NF165-positive retinal ganglion axons and an absence of the optic nerve (Fig. 7F,G,J,K, arrows). In sagittal sections, the mutant optic fissure remained open with clear expression of Mitf but not Pax2, whereas the residual optic stalk appeared hypoplastic without an NF165-positive optic nerve (Fig. 7H,I,L,M, arrows). Therefore, the recruitment of Frs2 to FGF receptor was essential for optic disc and optic fissure development.

Although FGF signaling is capable of activating multiple downstream signals in vitro, our study above showed that both the Fgfr1/2 and the Frs2a/Shp2 coloboma mutants converged upon a Ras-ERK signaling deficiency. To test whether the activation of Ras-ERK signaling alone can account for the role of FGF signaling in vivo, we next took a gain-of-function approach by employing an oncogenic LSL-Kras^G12D allele (Tuveson et al., 2004), which can be induced by Cre recombinase to express a constitutively activated Kras^G12D isoform (Fig. 8A). Induction of Kras^G12D alone in the Six3-Cre; LSL-Kras^G12D animals did not cause any overt phenotype (Cai et al., 2010; Cai et al., 2011). However, in the Six3-Cre;Frs2a^flox/flox;Shp2^flox/flox;LSL-Kras^G12D and the Six3-Cre; Fgfr1^flox/flox;Fgfr2^flox/flox;LSL-Kras^G12D mutants, Kras activation not only led to a recovery of phospho-ERK staining in the central retina, but also restored the formation of the optic disc and optic nerve, as indicated by Pax2, netrin 1 and NF165 staining (Fig. 8B-S, arrows). In adult animals, the occurrence of optic nerve dysgenesis or optic coloboma was significantly reduced in the Six3-Cre;Frs2a^flox/flox;Shp2^flox/flox;LSL-Kras^G12D mutants and completely eliminated in the Six3-Cre;Fgfr1^flox/flox;Fgfr2^flox/flox;LSL-Kras^G12D mutants (Table 1). These genetic rescues thus supported an Fgfr-Frs2α-Shp2-Ras-ERK signaling cascade in ocular development.

In this study, we provided mouse genetic evidence that ocular ablation of Fgfr1 and Fgfr2 caused coloboma and optic nerve dysgenesis, which has not been observed when FGF signaling is disrupted in any other vertebrate organisms. This result is particularly noteworthy, because unlike previous studies of FGF signaling in Xenopus and zebrafish (Take-uchi et al., 2003; Lupo et al., 2005; Picker and Brand, 2005; Picker et al., 2009), our Fgfr1/2 mutants did not display any axial polarity defects, suggesting that the observed ocular coloboma and optic nerve dysgenesis was not secondary to abnormal patterning of early optic cup. As indicated by the misexpression of the ocular astrocyte progenitor cell (APC) marker Pax2 and the RPE gene Mitf, loss of Fgfr1/2 signaling transformed the presumptive optic disc and optic fissure into RPE, suggesting a crucial role for FGF signaling in promoting the astrocyte progenitor cell fate versus the RPE fate. This function of FGF signaling is probably mediated by the reciprocal expression of Pax2 and Mitf, mutation of either one is known to cause ocular coloboma (Scholtz and Chan, 1987; Torres et al., 1996). Interestingly, the potentially antagonistic interaction between Pax2 and Mitf in optic disc and optic fissure formation contrasts with the early role of Pax2 in stimulating Mitf expression in optic vesicle development (Bäumer et al., 2003). Such function reversals are not without precedence, as Pax2 and Pax6 also transition from redundant to antagonistic interactions from early to late retinal development (Schwarz et al., 2000; Bäumer et al., 2003). Similarly, Chx10 and Mitf are co-expressed in early optic vesicle, but later become antagonistic in specifying the neural retina versus the RPE under the influence of FGF signaling, loss of which results in conversion of neural retina into RPE (Nguyen and Arnheiter, 2000; Horsford et al., 2005; Cai et al., 2010). The role of FGF signaling in preventing ectopic RPE formation appears to be unique to mouse and chick, as inhibiting FGF signaling has never produced ectopic pigmentation or coloboma in amphibian or fish. Conversely, none of the axial polarity phenotype observed when FGF signaling is inhibited in amphibian and fish is detected in our FGF signaling mouse mutants. Taken together, our study reveals a distinctive role of FGF signaling in higher organisms in promoting neuronal against glial fate during eye development.

It is notable that our Fgfr1/2 mutants displayed a restricted transformation into the RPE only at the optic disc/fissure and a limited neurogenesis defects only in the lip region of the ventral neural retina adjacent to the optic fissure. One possible explanation is the compensatory signaling by Fgfr3 and Fgfr4, which may be expressed at low levels in retina to mediate FGF signaling. However, we suggest that the developmental timing may also restrict the retinal defects in Fgfr1/2 mutants. From our Cre reporter assay, it was clear that the Six3-Cre deletor in the E9.5 optic vesicle was activated in a ventral-to-dorsal order. Consistent with this, the loss of phospho-ERK staining in the Fgfr1/2 mutants at E10.5 was also limited to the ventral optic vesicle that formed the future optic fissure and optic disc. We propose that the mutant optic fissure and optic disc regions underwent Mitf-mediated transformation into the RPE because they lost FGF signaling by E10.5, whereas the rest of retina that lost FGF signaling after E10.5 was unable to alter their retinal fate. In this model, FGF signaling is required for the initial demarcation of the optic vesicle, but it becomes dispensable for the maintenance of retinal progenitor cell fate after a limited time window. Our results therefore challenge the view that FGF signaling is required for retinal neurogenesis after E12, but instead support our previous study that FGF signaling is necessary only for the establishment of neuronal fate in optic vesicle at E10.5 (Cai et al., 2010).

The transient requirement of FGF signaling in early optic vesicle development is also supported by the lack of an ocular phenotype in the Frs2a and Shp2 single mutants. As we have previously described and have also shown in this study, Six3-Cre;Shp2^flox/flox single mutants did not display any embryonic retinal defects (Cai et al., 2010). By contrast, we have previously demonstrated that the retinal specific ablation of Shp2 by Rx-Cre, a Cre deletor that acts earlier than Six3-Cre, resulted in the transformation of the neural retina into the RPE. Comparing the Six3-Cre;Shp2^flox/flox and the Rx-Cre;Shp2^flox/flox mutants showed that there was a significant perdurance of Shp2 protein after the onset of Cre activities, which resulted in a considerable delay in ERK signaling disruption. As a result, only Rx-Cre was able to abrogate Shp2-ERK signaling early enough to interfere with the determination of the neural retina fate (Cai et al., 2010). In this study, we similarly showed that the Six3-Cre;Frs2a^flox/flox mutants did not significantly downregulate ERK phosphorylation until E13.5. In fact, only by depleting both Frs2α and Shp2 simultaneously in the Six3-Cre;Frs2a^flox/flox;Shp2^flox/flox mutants were we able to disrupt ERK signaling in the ventral retina at E10.5. This established a genetic interaction between Frs2α and Shp2, which form a binding complex in mediating FGF signaling in eye development. Such a binary complex is expected to be relatively resistant to a slow depletion in one of its binding partners, but the complex will disintegrate when the concentrations of both proteins are reduced below a certain threshold. Therefore, only Frs2a/Shp2 double mutants displayed ocular coloboma, albeit still at a slightly lower penetrance than that of Fgfr1/2 mutants (Table 1), which is likely because of the slower recycling rates of intracellular proteins when compared with membrane receptors. Our study thus presents a caution for future investigation of FGF signaling in embryonic development that protein perdurance may be an important confounding factor in interpreting conditional knockout results.

FGF receptors are known to recruit multiple adaptor proteins, including Frs2, Crk and PLCγ, to activate downstream signaling. In particular, recent studies suggest Frs2α, a lipid-anchored protein that is constitutively bound to FGF receptors, mediates some but not all FGF downstream pathways (Eswarakumar et al., 2006; Hoch and Soriano, 2006; Sims-Lucas et al., 2011). It has been previously reported that the Frs2α^2F/2/F mutant, which disrupts the Frs2α-Shp2 interaction, displayed microphthalmia or even anophthalmia (Gotoh et al., 2004). However, as Frs2a^2F/2/F is a systemic mutant, it is not clear whether these severe ocular phenotypes are due to cell autonomous defects or to abnormal tissue-tissue interaction. In this study, we have combined systemic deletion mutants with conditional alleles to disrupt protein-protein interactions in a tissue-specific manner. In the Six3-Cre;Fgfr1^ΔFrs2/flox;Fgfr2^flox/flox mutants, for example, the Fgfr1^ΔFrs mutation was compensated for by the Fgfr1^flox and Fgfr2^flox alleles in every tissue other than the retina, where these conditional alleles were ablated by Six3-Cre. This allowed us to demonstrate that the Fgfr1-Frs2 interaction within retinal progenitor cells was crucial in coloboma formation. Similarly, the coloboma phenotype in the Six3-Cre;Frs2a^2F/flox;Shp2^flox/flox mutants revealed the importance of Frs2α-Shp2 interaction in the retinal-specific FGF signaling. Finally, we showed that constitutive Kras signaling was sufficient to prevent ocular coloboma in these FGF signaling mutants, demonstrating a remarkably specific pathway from Fgfr to Frs2-Shp2 to Ras-ERK signaling in retinal development.

In summary, our present study identifies deficient FGF signaling as a novel cause of ocular coloboma and optic nerve dysgenesis. We show that FGF-activated Ras-ERK signaling is essential for the development of the optic nerve, optic fissure and optic disc, the abnormalities of which have been implicated in not only ocular birth defects but also in adult glaucoma. It is notable that iris and optic disc colobomas have also been observed in some individuals with Noonan syndrome, who have gain-of-functions mutations in Ras-MAPK signaling pathway (Kleanthous et al., 1987; Ascaso et al., 1993; Rudolph et al., 2001; Legius et al., 2002; Carvalho et al., 2003; Tartaglia and Gelb, 2005). Interestingly, we have recently shown that deletion of the genes associated with Noonan syndrome, Shp2 and Nf1, caused lens development defects (Pan et al., 2010; Carbe and Zhang, 2011). These findings suggest that both the gain- and loss-of-function mutations in FGF-Ras-MAPK signaling may have deleterious consequences in human ocular development. Therefore, FGF signaling components should be considered in the diagnosis and therapeutic interventions for ocular anomalies.

The authors thank Drs Philippe Soriano for the Fgfr1^ΔFrs ES cells; Yasuhide Furuta, Milan Jamrich and David Ornitz for mice; Peter ten Dijke and Carl-Henrik Heldin for antibody; and Kapil Bharti, Brian Brooks, Marie Fernandes, Lin Gan, Tom Glaser, Robert Heuckeroth, Roderick R. McInnes, Minh-Thanh T. Nguyen, Guillermo Oliver, David Ornitz, Naoki Takahashi and Valerie Wallace for in situ probes.