RPE specification in the chick is mediated by surface ectoderm-derived BMP and Wnt signalling

During vertebrate eye development, the optic neuroepithelium is specified into three domains: retinal pigment epithelium (RPE), neural retina (NR) and optic stalk. Once these domain specifications have occurred, the surface ectoderm overlying the presumptive NR thickens to form the lens placode. Subsequently, the distal portion of the optic vesicle (ov) invaginates and a double-layered optic cup develops, connected to the brain by the optic stalk. The external layer of the optic cup forms the single-layered, pigmented RPE and the inner layer develops into the multilayered NR (reviewed by Fuhrmann, 2010).

Little is known about the cellular and molecular mechanisms that lead to the subdivision of the ov into an RPE and NR domain (reviewed by Adler and Canto-Soler, 2007; Fuhrmann, 2010). Members of the transforming growth factor β (TGFβ) superfamily, the hedgehog (HH) and fibroblast growth factor (FGF) families appear to be involved in domain specification by modulating the expression of transcription factors (reviewed by Clegg et al., 2008; Fuhrmann, 2010). The boundaries between the RPE and NR domains are established by the Microphthalmia-associated transcription factor (Mitf) and the paired-like homeodomain transcription factor Vsx2, also known as Chx10 (Rowan et al., 2004; Horsford et al., 2005). Disruption of the Mitf gene in mice leads to a hyperproliferating, unpigmented RPE (reviewed by Bharti et al., 2006). Besides MITF, members of the orthodenticle-related family of transcription factors, Otx1/Otx2, the paired domain protein Pax6 and the homeodomain proteins Vax1/Vax2 and Lhx2 are involved in RPE development (reviewed by Bharti et al., 2006; Bharti et al., 2011; Bharti et al., 2012; Fuhrmann, 2010; Nishihara et al., 2012; Ou et al., 2013). In both mouse and chick, Mitf expression is observed in ov cells before NR commitment. In the mouse, the entire ov initially expresses Mitf, whereas in the chick Mitf expression is initially restricted to distal ov cells (Nguyen and Arnheiter, 2000; Müller et al., 2007; Ishii et al., 2009). FGFs released from the overlying surface ectoderm appear to repress RPE specification by inducing Vsx2 expression (Horsford et al., 2005; Nguyen and Arnheiter, 2000). Once NR specification has occurred in the distal ov, Mitf expression becomes restricted to the dorsal ov (Nguyen and Arnheiter, 2000; Nakayama et al., 1998; Müller et al., 2007; Ishii et al., 2009). At this stage, the chick ov is divided into a dorsal RPE and ventral NR domain (Kagiyama et al., 2005; Hirashima et al., 2008; Kobayashi et al., 2010).

Activins and BMPs belong to the TGFβ superfamily and appear to be involved in RPE specification (reviewed by Fuhrmann, 2010; Layer et al., 2010). Activin can substitute for the mesenchyme to induce/maintain RPE-specific gene expression in ov explants (Fuhrmann et al., 2000). During early eye development BMP application inhibits NR and optic stalk development, and instead induces RPE (Ohkubo et al., 2002; Hyer et al., 2003; Müller et al., 2007; Kobayashi et al., 2010). By contrast, inhibition of BMP signalling downregulates RPE-specific genes and results in loss of pigmentation (Adler and Belecky-Adams, 2002; Müller et al., 2007).

In several vertebrates, including humans, Wnt family members and their receptors are expressed during the initial stages of eye development, suggesting that Wnt signalling is involved in ov patterning. Overexpression of Wnt2b induces Bmp7 expression and results in thinning of the NR (Ohta et al., 2011), whereas BMP application at ov stages induces Wnt2b and Wnt8b expression (Müller et al., 2007; Teraoka et al., 2009). By contrast, NR specification requires suppression of Wnt8b in the mouse (Liu et al., 2010). However, at the time the RPE and NR are specified, the Wnt/β-catenin signalling pathway is not active in ov cells (reviewed by Fuhrmann, 2010).

Although several growth factors and transcriptional regulators have been identified, we still do not know which tissues are involved in RPE specification and how the different signalling pathways integrate to regulate cell fate decisions within the ov. We previously proposed that the BMP-expressing surface ectoderm rather than the mesenchyme is involved in RPE specification (reviewed by Clegg et al., 2008; Layer et al., 2010). Mitf expression is induced in the distal ov just underneath the BMP-expressing surface ectoderm at a time when mesenchymal cells are absent (Müller et al., 2007). Here, we show that BMPs and Wnts are released from the surface ectoderm to specify the RPE. Wnt2b is expressed at the right time and place to be involved in directing dorsal ov cells towards an epithelial cell fate. We provide gain-of-function evidence that Wnt-mediated GSK3β inhibition elicits RPE development in vivo. Accordingly, interfering with Wnt signalling prevents MITF expression and hence RPE development. Finally, we show that BMPs and Wnts cooperate at the level of pSmad to induce MITF expression in ov cells. We present a new model in which Wnts, via a GSK3β-dependent but β-catenin-independent pathway, stabilise BMP signalling to specify the RPE in dorsal ov cells.

In the chick, Mitf transcripts were first detected in the distal ov underneath the BMP-expressing surface ectoderm at stage 9 (Müller et al., 2007), whereas MITF protein expression was not yet observed (Fig. 1A). At stage 10, MITF protein expression was restricted to the distal ov adjacent to the overlying surface ectoderm (Fig. 1B).

Activation of the BMP signalling pathway results in receptor-mediated phosphorylation (p) of Smad1, 5 and/or 8 (pSmad), an indicator for BMP signalling. We compared the distribution of pSmad and MITF. At stage 10, pSmad was strongly detected in the surface ectoderm and in the nuclei of the entire distal ov (Fig. 1C-D′). At stage 10/11, pSmad labelling appeared to be stronger in the dorsal ov (Fig. 1E,E′). Thus, at ov stages the pattern of pSmad correlates well with the Mitf/MITF expression pattern.

At stage 10/11, when the distal ov is divided into a dorsal Mitf- and ventral Vsx2-expressing domain, Bmp4 and Bmp7 transcripts are still detected throughout the surface ectoderm (Crossley et al., 2001; Müller et al., 2007). This suggests that another signal is involved in progressively restricting Mitf expression to the dorsal ov. The Wnt signalling pathway directs dorsal cell fate decisions during embryonic development (reviewed by Klaus and Birchmeier, 2008). Therefore, we next studied the expression patterns of various Wnt family members, such as Wnt2b, Wnt3a and Wnt8b (see also Garcia-Lopez et al., 2004; Rossi et al., 2007; Quinlan et al., 2009; Van Raay and Vetter, 2004). Of the analysed Wnt ligands, only Wnt2b transcripts were restricted to the dorsal surface ectoderm and dorsal ov at the time the RPE is specified (Fig. 1G-K). At this time the Wnt/β-catenin signalling is not active in ov cells (Fig. 1F,F′). Co-localisation of MITF and nuclear β-catenin was detected only in a few cells beginning at the late optic vesicle stage (stage 12/13; supplementary material Fig. S1A-D). Taken together, the spatial and temporal expression pattern of Wnt2b and β-catenin suggests that perhaps Wnt signalling, via a β-catenin-independent signalling pathway, is involved in RPE specification.

It has been supposed that the surface ectoderm does not play a role in RPE specification (Fuhrmann et al., 2000; Nguyen and Arnheiter, 2000). Because surface ectoderm removal in these studies was carried out after Mitf expression had been initiated, we removed the surface ectoderm before or at the time Mitf expression is initiated. One day following surface ectoderm removal at stages 7 to 9 (Fig. 2A,B), optic cup and lens formation was not observed (Fig. 2C-H′, Fig. 3C,C′). Despite the presence of Pax7-positive neural-crest-derived mesenchymal cells, MITF protein expression and pSmad labelling was only weak or not detected (12/13 cases; Fig. 2E-H′; supplementary material Fig. S1F-J,Q-R′).

Additional experiments confirmed that the surface ectoderm is required for RPE specification. Optic vesicles from stage 9 and 10 chick embryos were isolated, the surface ectoderm removed and transplanted back to the chicken embryo (Fig. 2I,J). Optic vesicles, where the surface ectoderm had been removed at stages 9 and 10 (ten somites), completely lacked pigmentation 3 days after the operation (3/4 cases, compare Fig. 2K-M). Some pigmentation was observed when the surface ectoderm was removed at late stage 10 (11 and 12 somites; 17/26 cases; data not shown). Reduced pigmentation was also observed in cultured ov explants when surface ectoderm removal was carried out up to the ten-somite stage (36/37 cases; data not shown). Our data demonstrate that the surface ectoderm is required for RPE specification during the initial stages of ov patterning.

If it is correct that BMPs and/or Wnts from the surface ectoderm are responsible for RPE specification, then replacing the removed surface ectoderm with BMP- or Wnt3a-soaked beads should at least in part restore MITF expression. Following surface ectoderm removal at stage 8 to 10, a BMP7-soaked bead was placed either into the lumen of the ov or temporal to it and the embryos were allowed to develop until the late ov or early optic cup stage (stages 13 to 15; Fig. 3D,E,H,I). One day following the operation, we observed strong MITF and pSmad labelling in the distal ov or outer optic cup (Fig. 3F-G,J-K′). Remarkably, in the absence of the surface ectoderm, BMP7 application rescued optic cup and lens formation in 47% (9/19 cases; Fig. 3I-K′).

Next, we removed the surface ectoderm at ov stages and activated the Wnt signalling pathway by implanting Wnt3a-soaked beads (Fig. 3L,O). Like Wnt2b, Wnt3a is a known agonist of both the canonical and noncanonical Wnt signalling pathways (Abdul-Ghani et al., 2011; Gessert and Kühl, 2010; Qiu et al., 2011; Qu et al., 2013), and in contrast to Wnt2b, functional Wnt3a protein is commercially available. In the absence of the surface ectoderm, Wnt3a was sufficient to rescue optic cup and lens formation (8/16 cases) and MITF expression one day after the operation (4/8 cases; Fig. 3N,N′,Q-R′). Alsterpaullone (AP) is a chemical compound that specifically inhibits GSK3β (Broun et al., 2005), thereby mimicking Wnt signalling. When the surface ectoderm was removed and AP-soaked beads implanted at ov stages, only faint MITF expression was observed (3/3 cases; Fig. 3S-U′). In control experiments, where the surface ectoderm was removed and beads soaked in either phosphate-buffered saline (PBS) or dimethyl sulfoxide (DMSO) were implanted, MITF expression and pSmad labelling was not, or only weakly, detected (Fig. 3A-C; supplementary material Fig. S1). Taken together, these results show that BMP or Wnt signalling can replace the surface ectoderm to direct neuroepithelial cells towards an RPE cell fate and to allow optic cup and lens development.

To investigate whether Wnt signalling is required for RPE specification, we interfered with the Wnt signalling pathways at ov stages. Soluble-frizzled-related proteins (sFRP-1) and the Wnt-inhibiting factor-1 (WIF-1) bind to all Wnt ligands, thereby blocking both the canonical and noncanonical Wnt signalling pathways (Kawano and Kypta, 2003). Following treatment with sFRP-1 or WIF-1, some of the embryos were analysed 3 days after the operation to monitor the appearance of pigment granules. The eyes that developed were reduced in size and only faint pigmentation was observed (Fig. 4A,D). A dramatic downregulation or loss of MITF expression in the ov/cup was already observed 1 day after the operation (9/14 cases; Fig. 4B-C′,E-F′). The NR-specific marker Vsx2 was weakly induced in the RPE, which still had RPE-like morphology (supplementary material Fig. S3D,E).

We next carried out experiments that interfered with the Wnt/β-catenin signalling pathway at ov stages. Dickkopf-1 (Dkk1) specifically inhibits the Wnt/β-catenin signalling pathway by binding to low density lipoprotein receptor-related protein 5 or 6 (reviewed by MacDonald et al., 2009). Following implantation of Dkk1-soaked beads at stages 8 to 10, eye development appeared to be normal (Fig. 4G). Pigmentation and MITF expression was not affected compared with the contralateral eye (6/7 cases; Fig. 4G-I′). Thus, we extended previous findings in the mouse (Fujimura et al., 2009; Westenskow et al., 2009) to the ov stage, showing that the Wnt/β-catenin signalling pathway is not required for RPE specification.

To see whether Wnt2b is required for RPE specification, we targeted Wnt2b by electroporating Wnt2b siRNA at ov stages. One day later, MITF expression was not affected in the operated eye compared with the contralateral eye (6/6 cases; supplementary material Fig. S2A-D′). This is in agreement with the observation that Wnt2b mutant mice do not exhibit any eye phenotypes (Tsukiyama and Yamaguchi, 2012). However, these data suggest that Wnt signalling, possibly via a Wnt/β-catenin-independent signalling pathway, is required for RPE specification.

To test the potential role of Wnt signalling in RPE specification, we carried out gain-of-function experiments and implanted Wnt3a beads at ov stages. Compared with controls (Fig. 5A-E), the Wnt3a-treated embryos developed a small pigmented eye with a rudimentary lens 3 days after the operation (Fig. 5F,G). Histological examination showed that Wnt3a treatment prevented optic cup formation (Fig. 5G) and that the entire vesicle had RPE-like morphology, being strongly pigmented (11/15 cases; Fig. 5H,I). Moreover, NR development was completely inhibited and the entire ov, including the optic stalk region, expressed MITF (Fig. 5J; data not shown).

In the chick and frog, both BMP and Wnt signalling cooperate to induce an epidermal cell fate (Wilson et al., 2001; Fuentealba et al., 2007; Patthey et al., 2009). Wnt-mediated GSK3β inhibition stabilises BMP signalling by preventing the degradation of pSmad1, thereby inducing an epidermal cell fate (Fuentealba et al., 2007). To test whether a similar mechanism exists during ov patterning, we inhibited GSK3β at these stages (9-14 somites). Following AP treatment, a small, strongly pigmented, MITF-expressing vesicle developed, phenocopying Wnt3a treatment (34/41 cases; compare Fig. 5F-J with 5K-Y). In those vesicles, pSmad labelling (Fig. 5R,W) was observed, colocalised with the RPE-specific marker RPE65 (Fig. 5X,Y). By contrast, the expression of NR markers, such as Vsx2, was downregulated or lost (supplementary material Fig. S3A,C). We did not detect any nuclear β-catenin staining in the neuroepithelium of AP-treated vesicles 1-3 days after the operation (6/6 cases; Fig. 5S,T). No changes were observed in control experiments (10/10 cases; supplementary material Fig. S1). Thus, our data suggest that Wnt-mediated GSK3β inhibition can, independently of β-catenin nuclear transfer, induce MITF expression in chick ov cells.

BMP application at ov stages induces both ectopic Wnt2b and Mitf expression in the presumptive NR (Müller et al., 2007). Thus, it is possible that BMP-induced Wnt signalling maintains MITF expression and hence RPE development. Alternatively, Wnt-mediated GSK3β inhibition could stabilise BMP signalling to induce an epithelial phenotype in ov cells (see above). Therefore, we next inhibited BMP signalling by implanting Noggin-soaked beads, simultaneously activating the Wnt signalling pathway.

Exposure to Wnt3a or GSK3β inhibition converts the entire ov into a strongly pigmented and MITF-expressing vesicle (Fig. 5). By contrast, in the absence of BMP signalling, activation of Wnt signalling was no longer able to induce MITF expression (4/5 cases; Fig. 6A-E′). In those embryos, we observed only faint, or no, pSmad labelling in the distal ov and surface ectoderm (Fig. 6F,F′).

To investigate whether Wnt signalling is required for BMP-induced MITF expression, we inhibited Wnt signalling by simultaneously activating the BMP signalling pathway. Exposure to BMP alone induces strong MITF/Mitf expression in the ov (Fig. 3) (Müller et al., 2007). Following implantation of two beads, soaked in BMP7 and WIF-1 into the ov (Fig. 6G), activation of the BMP signalling pathway was no longer able to induce or maintain MITF expression and only faint or no pSmad staining was observed (6/6 cases; Fig. 6H-L′). In control experiments, pSmad and MITF expression was observed only in the dorsal ov or optic cup and in the dorsal surface ectoderm (5/5 cases; supplementary material Fig. S1K-P).

Wnts can signal through Smad1 (Fuentealba et al., 2007). Here, GSK3β inhibition prevents degradation of pSmad1, thereby stabilising BMP signalling (reviewed by Eivers et al., 2008). To see whether a similar mechanism exists during ov patterning, we implanted WIF1-soaked beads in the presence of the GSK3β-inhibitor AP (Fig. 6M). One day after the operation, eye development arrested at the ov stage and nuclear β-catenin labelling was not observed (3/4 cases; Fig. 6N-R′). Strikingly, in the absence of Wnt signalling, GSK3β inhibition was sufficient to stabilise pSmad and rescue MITF expression in both the dorsal and ventral ov (3/4 cases; Fig. 6Q-R′), suggesting that pSmad is sufficient to induce MITF expression in the absence of the surface ectoderm.

To confirm that pSmad mediates RPE specification, we removed the surface ectoderm and electroporated a constitutively active form of human Smad1 (CA-Smad1) (Fuentealba et al., 2007) into ov cells (Fig. 7A). In the absence of the surface ectoderm CA-Smad1 was sufficient to rescue MITF expression (6/6 cases; Fig. 7B-C′). In a control experiment, wild-type Smad1 (Fuentealba et al., 2007) did not lead to MITF expression in ov cells in the absence of the surface ectoderm (5/5 cases; Fig. 7D,D′).

In Xenopus, it has been shown that in the absence of Wnt signalling pSmad is degraded, thereby inducing a neuronal cell fate (Fuentealba et al., 2007). Wnt2b expression is restricted to the dorsal surface ectoderm (Fig. 1), so that a similar mechanism could exist in the ov to allow NR development ventrally. To test this, we tried to inhibit pSmad degradation thereby directing presumptive NR cells towards an RPE cell fate. Application of the proteasomal inhibitor MG132 (Lee and Goldberg, 1998) at ov stages (7-10 somites) induced ectopic MITF expression in the ov/cup (4/7 cases; Fig. 7E-G′,J-L′). One to three days following the operation pSmad labelling was increased in the inner optic cup, and downregulation of the NR-specific marker Vsx2 was observed (Fig. 7H,I,M). Taken together, our data are consistent with a model, according to which Wnt-mediated GSK3β inhibition stabilises BMP signalling, thereby restricting RPE development to the dorsal ov.

The classical model of ov patterning holds that the distally located surface ectoderm induces NR development, whereas proximally signals released from the mesenchyme elicit RPE development. In this study, we show that signals released from the surface ectoderm, and not from the mesenchyme, promote RPE specification.

Fate-mapping studies in the chick showed that the distalmost part of the ov at stages 9 and 10 contributes to the eye, whereas the dorsalmost region of the ov consists of telencephalic precursors (Garcia-Lopez et al., 2009; Pombero and Martinez, 2009). Consistent with this, RPE specification initiates distally underneath the surface ectoderm, but soon the MITF domain becomes restricted to the dorsal ov and optic cup (Fig. 1) (Müller et al., 2007; Ishii et al., 2009). Similarly, polarisation of active BMP signalling to dorsal ov cells is observed in both the chick and mouse (this study; Monteiro et al., 2004). However, at this time point BMPs are still expressed in the entire surface ectoderm (Furuta et al., 1997; Müller et al., 2007), suggesting that another signal(s) is involved in restricting MITF expression to the dorsal ov/cup. BMP and Wnt family members are often co-expressed in dorsal tissues and co-regulate vertebrate pattern formation (e.g. Abu-Khalil et al., 2004; Szeto and Kimelman, 2004; Ramel et al., 2005; Zechner et al., 2007; Ohta et al., 2011). Here, we identified Wnt2b as a candidate gene involved in RPE specification. Expression of Wnt2b is detected in the dorsal surface ectoderm, dorsal ov and/or optic cup in several species (this study) (Zakin et al., 1998; Cho and Cepko, 2006; Müller et al., 2007; Veien et al., 2008; Grocott et al., 2011), and Wnt signalling can induce an RPE cell fate in vitro and in vivo (this study; Aoki et al., 2006; Eiraku et al., 2011). Targeting Wnt2b by electroporating Wnt2b siRNA at ov stages did not affect MITF expression (this study) and, although a large number of genes associated with Wnt signalling are expressed during vertebrate eye development, none of the single Wnt mutant mice, including the Wnt2b knockout mice, displays any eye phenotypes (van Amerongen and Berns, 2006; Tsukiyama and Yamaguchi, 2012). However, similar to BMP inhibition (Adler and Belecky-Adams, 2002; Müller et al., 2007) we show that Wnt signalling is required for RPE development. Targeting several Wnt ligands by inhibiting Wnt receptor-mediated signalling results in reduced pigmentation and downregulation of MITF expression (Fig. 4). Moreover, we show that during a narrow time window, the BMP- and Wnt-expressing surface ectoderm is required to mediate RPE specification. The loss of MITF expression following surface ectoderm removal can be rescued by the activation of the BMP or Wnt signalling pathways.

During vertebrate development tissue interactions are required for proper organ formation. We confirm previous findings that the surface ectoderm is required for optic cup formation (Fig. 2) (Hyer et al., 1998; Hyer et al., 2003; Nguyen and Arnheiter, 2000). In the absence of the surface ectoderm, an optic cup fails to form, independently of whether mesenchymal cells remain dorsally or cover the entire ov. Interestingly, in the absence of the surface ectoderm, BMPs or Wnts can rescue optic cup and lens formation (Fig. 3). However, recent in vitro studies suggest that tissue interactions are not required for optic cup formation (reviewed by Sasai et al., 2012; Sasai, 2013). Embryonic stem cell (ESC) reaggregates cultured on matrigel are patterned into NR and RPE in a self-autonomous way (Eiraku et al., 2011; Nakano et al., 2012). It should be noted, that we observed optic cup formation only if optic vesicle explants were cultured on growth factor-containing matrigel, but not in the presence of collagen (Ichie Steinfeld, unpublished observations). It is possible that matrigel releases proteins (Hughes et al., 2010) that allow optic cup formation in the absence of the surface ectoderm. For example, during matrigel-induced hepatocyte differentiation, β-catenin undergoes multifactorial regulation (Monga et al., 2006). Interestingly, patterning into NR and RPE of ESC reaggregates is only observed in the presence of neuroectodermal epithelium or in the presence of both Wnt3a and high concentrations of matrigel (Eiraku et al., 2011). Further work is required to identify the exact cellular and molecular mechanisms that direct optic cup formation in vertebrates.

Wnt signals are context-dependently transduced via the noncanonical or canonical signalling pathways (reviewed by Kikuchi et al., 2007; Katoh and Katoh, 2007). Here, we show that RPE specification is mediated by a Wnt-signalling pathway that is GSK3β-dependent but β-catenin-independent. Firstly, in agreement with previous studies (see Introduction), we did not detect any nuclear β-catenin in the neuroepithelium at the time the RPE is specified (Fig. 1). Secondly, canonical Wnt/β-catenin signalling requires GSK3β inhibition. Following GSK3β inhibition the NR and optic stalk are respecified to develop into a strongly pigmented RPE, independent of nuclear β-catenin accumulation (Fig. 5). Thirdly, following inhibition of Wnt signalling, GSK3β inhibition is sufficient to rescue MITF expression, and this is again independent of nuclear β-catenin accumulation (Fig. 6). Lastly, specific inhibition of the Wnt/β-catenin signalling pathway at optic vesicle stages did not affect MITF expression and pigmentation (Fig. 4). In support of this, eye pigmentation appears normal in mutant mice, where the co-receptors of the Wnt/β-catenin signalling pathway, Lrp5 and 6, are mutated (Fujino et al., 2003; Zhou et al., 2008). Moreover, ectopic activation of the Wnt/β-catenin signalling pathway alone is not sufficient to induce MITF expression (Westenskow et al., 2010) and instead eye development and optic cup formation is disrupted (Smith et al., 2005; Fu et al., 2006; Miller et al., 2006; Fujimura et al., 2009). In particular, β-catenin overexpression within the RPE itself results in a complete loss of pigmentation and Mitf expression (Fujimura et al., 2009). Dkk1, a specific antagonist of the Wnt/β-catenin signalling pathway, is expressed in the chick ov (GEISHA chicken embryo gene expression database: http://geisha.arizona.edu) and in Dkk1 mutant mice, eyes or anterior head structures fail to develop (MacDonald et al., 2004; Mukhopadhyay et al., 2001). This suggests that at ov stages, repression of Wnt/β-catenin signalling is required for proper vertebrate eye development. Accordingly, during the process of stepwise differentiation of human or mouse ESCs towards a retinal cell fate, Dkk1 is initially supplemented to the medium (Osakada et al., 2009).

The status of Wnt signalling regulates neural and epidermal cell fates in chick and frog embryos (Wilson et al., 2001; Fuentealba et al., 2007; Patthey et al., 2009). In Xenopus, the phosphorylation status of Smad1 determines whether a cell acquires an epidermal or neuronal cell fate (Fuentealba et al., 2007; Eivers et al., 2008). A neuronal cell fate is induced in the absence of Wnt signalling. Here, Smad-1 receives inhibitory phosphorylations by GSK3β and by the FGF/MAPK, which leads to proteasomal degradation of pSmad-1 (Fuentealba et al., 2007). By contrast, in the presence of Wnt signalling, GSK3β is inhibited and pSmad-1 can now induce an epidermal cell fate (Fuentealba et al., 2007). A similar mechanism appears to exist during ov patterning to restrict the RPE domain to the dorsal ov. On the one hand, activation of the Wnt signalling pathway at ov stages does not rescue MITF expression lost from BMP inhibition (Fig. 6). On the other hand, in the absence of Wnt signalling, activation of the BMP signalling pathway can no longer induce MITF expression in ov cells. In both cases, we observed no, or only weak, pSmad staining in ov cells. Following inhibition of Wnt signalling, GSK3β inhibition is sufficient to stabilise pSmad in ov cells and to rescue MITF expression and hence RPE development (Fig. 6). Moreover, overexpression of a constitutively active form of Smad1 rescues MITF expression in the absence of the surface ectoderm. Accordingly, inhibition of proteasomal pSmad degradation maintains MITF and results in the loss of Vsx2 expression (Fig. 7).

Based on these findings, we propose a dorsoventral model for the initial stages of ov patterning. Members of the BMP, FGF and Wnt families are released from the surface ectoderm to pattern the ov transiently into a dorsal RPE and ventral NR domain. Initially, the BMP-expressing surface ectoderm initiates Mitf expression through pSmad in the entire distal ov (Fig. 8A). Wnt2b and as yet unidentified Wnt ligand(s) localised to the dorsal surface ectoderm and dorsal ov inhibit GSK3β, thereby stabilising pSmad and maintaining Mitf expression in the dorsal ov and optic cup. In the absence of Wnt signalling, pSmad is degraded through MAPK/ERK and GSK3β phosphorylation, which then leads to NR specification/Vsx2 expression in the ventral ov (Fig. 8B). Once RPE and NR specification occurs, Vsx2 expression is stabilised distally, whereas RPE development will continue in the dorso-proximal region, where mesenchymal cells now become involved in maintaining RPE-specific gene expression.

The molecular interplay between several pathways at the level of pSmad during cell fate specification in the ov may provide the basis for directed differentiation of human ESCs towards functional RPE cells that in the future could be used for cell replacement of RPE-mediated blinding diseases (reviewed by Clegg et al., 2008; Bharti et al., 2011).

Fertilised chicken eggs (Gallus gallus, Linnaeus) were obtained from LSL Rhein-Main (Dieburg, Germany). For all experiments the eggs were incubated at 38°C and 60% humidity, until they reached the desired stages (stage 7 to 11) according to Hamburger and Hamilton (Hamburger and Hamilton, 1951). The eggs were then windowed and the embryonic membranes removed. A small incision was made into the midline of the forebrain, or temporally to the ov using sharpened tungsten needles. One bead (see below) was then transferred into the egg, inserted through the slit and positioned into the ov. The eggs were then sealed and left to develop at 38°C until they reached the desired stage. Owing to the turning of the chick embryo at stage 12/13, some beads placed into the right ov moved into the brain vesicle one day after the operation. The embryos were fixed in 4% paraformaldehyde (PFA) in PBS at 4°C for 2 to 48 hours. Embryos to be used for whole-mount in situ hybridisation (ISH) were dehydrated and stored in 100% methanol. Those intended for ISH on sections were cryoprotected overnight in 10% sucrose in diethylpyrocarbonate-H₂O at 4°C and transferred into 20% and 30% sucrose for at least 1 hour each. Consecutive 10-14 μm thick sections were analysed by ISH or immunohistochemistry. Control experiments were carried out by either implanting PBS- and/or DMSO-soaked beads into the ov of stage 9/10 chick embryos according to the protocol described above.

A 2 μl drop of the designated protein or chemical was placed in a Petri dish and about eight drops (10 μl each) of PBS were placed around it. About 15 beads (see below) were added to the solution and incubated for a minimum of 1 hour. Agarose beads (AffiGel Blue Beads, Bio-Rad) were soaked either in 0.5 or 1 μg/μl BMP5 and BMP7 protein (R&D Systems), 0.5 or 1 μg/μl Wnt3a (R&D Systems), 1 μg/μl DKK1 (R&D Systems), 2 μg/μl sFRP1 (PreProTech) or 1 μg/μl human WIF-1 protein (R&D Systems), 1 μg/μl Noggin (R&D Systems). Alsterpaullone (Sigma-Aldrich) and MG132 (Merck) was dissolved in DMSO (Roth) and used at a concentration of 1 μg/μl using AG1-X2 resin beads (Bio-Rad).

Two per cent Nile Blue A sulfate (Sigma-Aldrich) dissolved in agarose was used to support the removal of the surface ectoderm at stages 7 to 10 using fine glass needles (Geetha-Loganathan et al., 2010).

The right ov was removed from stage 9 (nine somites) or stage 10 chick embryos (10-12 somites), treated with 0.05% collagenase for 30 minutes to loosen the surface ectoderm, which then was removed using sharpened tungsten needles. Using a glass pipette, the remaining ov was then transplanted back to the same chick embryo. Embryos were placed into the incubator and left to develop for another 72 hours, fixed in 4% PFA and processed as described above. In vitro experiments using chick explant cultures were carried out as described before (Kobayashi et al., 2010).

The ov of stage 9-10 chick embryos were usually injected with a green fluorescent protein (GFP) expression construct (1-3 μM) mixed with 1.5-2 μg/μl plasmids or siRNA (see below) containing 0.01% Fast Green. Using a TSS20 Ovodyne Electroporator (Intracel), two pulses (7V, 30-msecond duration) in 970-msecond intervals were delivered, using 0.5 mm diameter, tungsten and platin electrodes. The anode was positioned lateral to the right ov and the cathode was placed into the brain. The DNA/RNA constructs used were Wnt2b siRNA and siRNA-A (both Santa Cruz), pC S2 hSmad1-WT (Addgene plasmid 22989) and pC S2 hSmad1-GM (Addgene plasmid 22992; Edward de Robertis; Pera et al., 2003; Fuentealba et al., 2007). The embryos were fixed 1 day later and assayed for MITF and GFP expression.

RNA ISH on cryostat sections and whole mounts was performed as previously described (Müller et al., 2007). Antisense RNA probes specific for chicken Rx (T. Ogura, Tohoku University, Aoba, Japan), Mitf (M. Mochii, University of Hyogo, Japan), Vsx2 and Wnt3a, Wnt4, Wnt5, Wnt5b, Wnt7b, Wnt8b (D. Schulte, Goethe University of Frankfurt/M., Germany) and Wnt2b (H. Roelink, University of Washington, Seattle, WA), were used.

In explant cultures of ov, MITF expression was carried out as previously described (Kagiyama et al., 2005; Kobayashi et al., 2010). Human MITF antibody (Sigma-Aldrich, HPA003259), PY489-β-catenin (DSHB), pSmad1/5, pSMAD1/5/8 (Cell Signaling, 9516, 9511S), Retinal pigment epithelial-specific protein 65 kDa (Novus Biologicals, 4018BD113D9) and mouse Pax7 (MAB1675) were used on cryostat sections.

We thank Ulrike Hoppe for excellent technical assistance and Anja Heselich, Dorothea Schulte and Hermann Rohrer for critical reading of the manuscript.