Differential interactions of eyeless and twin of eyeless with the sine oculis enhancer

The Drosophila visual system consists of the two compound eyes and the ocelli, which are located on the adult vertex and consist of three simple eyes (Stark et al., 1989). Both types of eyes develop from a small number of cells that are set aside in the embryo. These cells form the eye part of the eye-antenna imaginal disc and proliferate during the larval stages. The compound eye develops from the central part whereas the ocellus develop from the anterior-medial region of the eye imaginal disc. The compound eye in Drosophila consists of a precisely organized array of approximately 750 ommatidia, each containing eight photoreceptor neurons and twelve accessory cells. The ommatidia form in the early third instar larva, when a wave of pattern formation, marked by an indentation called the morphogenetic furrow, moves across the eye disc in a posterior to anterior direction (reviewed by Wolff and Ready, 1993). Although committed to retinal fate, cells anterior to the furrow are still undifferentiated, whereas cells posterior to it are sequentially recruited into ommatidial clusters undergoing retinal differentiation (reviewed by Treisman and Heberlein, 1998). Our understanding of the molecular events that occur in and posterior to the furrow, such as pattern formation, ommatidial assembly and cell differentiation, has advanced dramatically in recent years. Early studies focused largely on late events of pattern formation. More recently genes involved in eye specification functioning early in eye morphogenesis have also been discovered.

Determination of the eye primordium requires several nuclear proteins that are known to act as transcriptional regulators. The Drosophila Pax6 gene ey was the first gene shown to display the capacity to induce ectopic eye morphogenesis (Halder et al., 1995). Like ey, toy encodes a Pax6 gene containing two DNA-binding domains (Czerny et al., 1999). eye gone (eyg) encodes a Pax-like protein (Jun et al., 1998), sine oculis (so) is a homeobox gene (Cheyette et al., 1994) while eyes absent (eya) and dachshund (dac) both encode different nuclear proteins (Bonini et al., 1993; Mardon et al., 1994). Analysis of the expression patterns of these genes combined with genetic approaches, have revealed a sequential and hierarchical cascade during compound eye development. toy is the first to be expressed during embryogenesis and activates ey in the eye primordium (Czerny et al., 1999). so is required for the development of the entire visual system, including the compound eyes, the ocelli, the optic lobe of the brain and the larval photoreceptor designated as Bolwig’s organ (Cheyette et al., 1994; Pignoni et al., 1997; Serikaku and O’Tousa, 1994). eya is expressed later in the compound eyes and the ocelli specifying region in third instar eye imaginal discs. Like so, it is also required for compound eyes and ocelli formation since eya mutants lack both visual systems (Zimmerman et al., 2000). so, eya and dac have been shown to be downstream of ey and regulated by it (Halder et al., 1998; Niimi et al., 1999; Zimmerman et al., 2000). These proteins form complexes that feed back on ey expression and they are also capable of inducing ectopic eye morphogenesis (Bonini et al., 1993; Bonini et al., 1997; Pignoni et al., 1997). Despite a high sequence homology in their DNA binding domains, EY and TOY appear to exhibit different biological roles during development (Czerny et al., 1999). First, these genes are expressed differentially during embryonic development. Second, ey and toy are co-expressed in the eye imaginal disc but only toy is expressed in the ocellar region. Therefore, the lack of ey in ey² mutants impairs compound eye formation but not ocellar development. This suggests that so and eya are not under the control of ey in the ocelli but require other regulators to ensure their proper expression in these cells. The precise role of toy during Drosophila development is still not understood.

Analysis of elements that control the expression pattern of genes involved in early eye development should provide additional details on the genetic hierarchy during eye specification. It has been shown that toy induces the expression of ey through the eye-specific enhancer of the ey gene in the eye precursor cells of the embryo, but not during the larval stages in the eye disc (Czerny et al., 1999). The so10 enhancer, which is part of an eye-specific regulatory sequence deleted in the so¹ mutant, has been shown to be a direct ey response element (Niimi et al., 1999). Recently, genomic deletions proximal to the promoter found in eya¹ and eya² mutants were identified as eye-specific enhancers. These elements have been shown to be inducible by EY in the antennal disc (Zimmerman et al., 2000). All these data provide additional information leading to a better understanding of the complexity of the interacting network during early eye development.

In this study, we addressed the regulation of the so10 enhancer during eye morphogenesis. We show that the so10 enhancer is bound and regulated by EY and TOY through their paired domain (PD). Despite extensive sequence homology, the PD of EY and TOY bind different sites in this enhancer. Targeted mutagenesis experiments allowed us to establish that these different binding sites are functional and required to ensure proper activity of the so10 enhancer in the eye disc. Finally, rescue experiments of the so¹ mutant using the different mutated versions of the so10 enhancer demonstrated that TOY and EY have different functions in the formation of the compound eyes and the ocelli through the same enhancer. Thus the TOY binding sites are absolutely required for ocelli development and the EY binding sites are required for compound eye formation.

Flies were reared on standard medium at 25°C. Lines used: so10-lacZ (Niimi et al., 1999), dpp^blink-Gal4 (Staehling-Hampton and Hoffmann, 1994), UAS-ey (Halder et al., 1995), UAS-toy (Czerny et al., 1999), ey² (Quiring et al., 1994), so¹ (Cheyette et al., 1994), eya¹ (Bonini et al., 1993), spa^pol (Fu and Noll, 1997), UAS-eyΔPD and UAS-eyΔHD (Punzo et al., 2001), ey^J5.71 and toy^G7.39 (S, F., U. Kloter, and W. J. G., unpublished).

Specific genotypes were generated: (1) so10-lacZ/so10-lacZ; dpp^blink-Gal4/TM6B,Tb,Hu; spa^pol/spa^pol, (2) so10-lacZ/so10-lacZ; ey²/ey², (3)so¹/so^1;so10-lacZ/so10-lacZ, (4) eya¹/eya^1;so10-lacZ/so10-lacZ, UAS-toy/UAS-toy; ey^J5.71/ci^D, UAS-toy/UAS-toy; ey²/ey². Transgenic lines were generated by P-element-mediated germline transformation in yw¹¹¹⁸. New lines were created for this study: so10^EY/TOYmt, so10^TOYmt, so10^EY+TOYmt-Gal4 and so10^EY+TOYmt-lacZ lines; so9-lacZ, so7-lacZ, so7-Gal4, so7^EY/TOYmt-Gal4, so7^TOYmt-Gal4, so7^EY+TOYmt-Gal4; UAS-toyΔPD and UAS-toyΔHD. For each construct at least two lines with P-element insertions onto different chromosomes were analysed to ensure correct expression patterns. Each Gal4 line was verified for correct expression patterns by crossing to a UAS-lacZ followed by X-gal staining on discs. lacZ expression was detected by X-gal staining on discs, or by means of immunohistochemestry with a monoclonal anti-β-gal antibody (Promega) according to Halder et al. (Halder et al., 1998).

Gel shift assays were performed with the 128 bp so fragment and full-length EY and TOY proteins as described previously (Niimi et al., 1999). Full-length EY and TOY proteins were synthesized in reticulocyte lysates from the T7 and T3 promoter, respectively, according to the manufacturer specification (Promega) by using the pBSK-ey and toy plasmids (Czerny et al., 1999; Quiring et al., 1994). For the DNaseI footprinting assay, 6xHis tagged paired domains (pQE30-EY PD and TOY PD) were produced and purified in native condition onto Ni-NTA columns according to the manufacturer’s specifications (Qiagen). The proteins were diluted in the binding buffer to obtain a final concentration of 10 mM Tris pH7.5, 75 mM NaCl, 1 mM EDTA, 1 mM DTT, 0.5 mM PMSF, 10% glycerol, 25 mM imidazole, and 50 μg/ml poly(dI-dC). Reactions were carried out in 100 μl by incubating 100 ng of His-PD purified proteins with 10 ng of the so10 fragment end-labelled with [α-³²P]dATP at Asp718 for 15 minutes at room temperature followed by digestion with 0.001 U of DNaseI (Boehringer Manheim) for 1 minutes at 20°C. After phenol-chloroform extraction, analysis was performed on a denaturing sequencing gel followed by autoradiography.

Western blot experiments were done with a rabbit anti-EY antibody or with a rabbit anti-TOY antibody at a dilution of 1:200. The antibody was preabsorbed with larval tissue. Each lane was loaded with extracts from 10 leg discs. All extracts for western blotting were boiled for 6 minutes. Transfer was verified by Ponceau Red staining. The secondary antibody for detection of the signal was used at a dilution of 1:2000 (HRP-coupled swine anti-rabbit antibody from DAKO A/S) and the signal was revealed using a chemoluminescence kit according to the manufacture’s specifications (Amersham).

The EY and TOY binding sites within the so10 enhancer were sequentially mutated using standard PCR amplification procedures. The resulting 428 bp Asp718-EcoRI mutated fragments were subcloned in pBluescript SK, sequenced, further excised using BamHI-Asp718 and subcloned into the lacZ pCβ vector (Niimi et al., 1999) at Asp718-BamHI. To generate so10 and so7-GAL4 expression vectors, we first modified the pUAST vector (Brand and Perrimon, 1993) by deleting the XbaI-SphI fragment, removing the polylinker, UAS sequences and the hsp70 promoter. A new polylinker was inserted through the same sites to create the new AscI-NotI-SpeI-BglII-KpnI-NgoMIV-EagI-EcoRI-AvrII-NheI-SphI multiple cloning sites. The Gal4 encoding sequence was amplified by PCR in order to create AscI-NotI cloning sites at both ends and subcloned into the modified pUAST vector to generate the pP-Gal4 vector. A 300 bp NotI minimal hsp70 promoter fragment was further inserted in front of the Gal4 at NotI. This resulted in the pPhsp70-Gal4 vector. The 428 bp so10 enhancer and derivatives were inserted into this pPhsp70-Gal4 vector at EcoRI-Asp718 to generate the resulting so10-Gal4 constructs. To generate the so7 constructs, the 1.6 kb EcoRI so7 fragment (Niimi et al., 1999) was first subcloned into Bluescript SK at EcoRI. Then, the Asp718 1.2 kb subfragment so9 was excised and subcloned at Asp718 in the correct orientation into the different derivatives of the so10-hsp70-Gal4 constructs resulting in the so7-Gal4 constructs. The so9-lacZ construct was generated by cloning the 1.2 kb Asp718 so9 fragment into pCβ (Niimi et al., 1999) at Asp718. The 128 amino acids EY- and TOY-PD were amplified by PCR in order to create BglII and Asp718 cloning sites at both ends to ensure cloning in the correct ORF into the pQE30 vector at BamHI-Asp718 sites (Qiagen). After PCR amplification, each construct was verified by sequencing. The toy cDNA was deleted between P₁₃-Q₁₆₂ for toyΔPD using the NsiI-BlpI sites to generate the toyΔPD. The toyΔHD was deleted between L₂₁₉-R₂₉₁ using BbsI-EagI sites. The deleted cDNAs were further excised from Bluescript and inserted as an Asp718-XbaI fragment in pUAST. Detailed description of the primers used will be given upon request.

Recent genetic analysis demonstrated that the induction of so mediated by EY occurs during second instar larvae in the eye progenitor cells (Halder et al., 1998). A genomic fragment deleted in the so¹ mutant has been shown to interact specifically with ey in a yeast one hybrid assay and to be ectopically inducible by EY in vivo (Niimi et al., 1999). Therefore, we were interested in the regulation of this enhancer element (so10) during normal eye development. To address this question, we first studied the expression of the so10 enhancer in an ey mutant background. When compared to a wild-type eye disc (Fig. 1A), the expression of the transgene (so10-lacZ) is dramatically reduced in the anterior to posterior central region of an ey² mutant eye disc (Fig. 1B) and also of an ey null mutant eye disc (Fig. 1C) (ey null: ey^J5.71; S. F., U. Kloter and W. J. G., unpublished) (Punzo et al., 2001). The residual lateral expression of the enhancer detected in the eye disc indicates that additional transcription factors are involved in so10 regulation. To investigate whether the loss of expression is due to a loss of EY rather than to a loss of so10-expressing cells by apoptosis or a loss of the expression of ey downstream genes, we studied the expression of the so10 enhancer in so¹ and eya¹ mutants. Like ey mutants these two mutants also exhibit an increased amount of apoptosis (reviewed by Treisman and Heberlein, 1998). As shown in Fig. 1D,E the expression in the anterior region of the eye disc is not affected in so¹ and eya¹ mutants, respectively. These results strongly support the idea that the loss of expression detected in the anterior part of an ey² or an ey^J5.71 mutant eye disc is directly due to a loss of EY and not to the loss of expression of ey downstream genes. The loss of expression in the posterior region shows that this enhancer is regulated differently in the anterior than in the posterior part of the eye disc. It suggests a concerted involvement of EY/SO/EYA in the expression of this element in differentiating cells, which has not been further investigated in this study.

The sparkling (spa) gene, which is involved in lens formation, belongs to the Pax2/5/8 gene family and is expressed posterior to the morphogenetic furrow (Fu and Noll, 1997). We also investigated a possible involvement of this Pax gene in so10 enhancer regulation posterior to the morphogenetic furrow. As shown in Fig. 1F the expression of so10 seems not to be affected in this mutant. toy, the second Pax6 gene expressed in the developing eye disc, reveals a similar expression pattern to ey (Czerny et al., 1999). Since the so10 enhancer exhibits residual expression in ey mutants, we asked whether TOY could also be involved in its regulation. We addressed this question by using the recently isolated toy mutant toy^G7.39 (S. F., U. Kloter and W. J. G., unpublished). Owing to the hypomorphic character of this mutant the size of the eye disc can vary a lot within the same larva from a small cluster of cells to almost wild-type size (Fig. 1I). According to this, toy transcript is still detected in eye discs of normal size whereas it is not detected in strongly reduced eye discs (data not shown). As shown in Fig. 1G-I so10-lacZ expression is reduced in the central region of a toy^G7.39 mutant eye disc. Whereas in Fig. 1G the expression is restricted to the posterior margin of the eye disc, in Fig. 1H a clear territory is not definable owing to the shape of the eye disc although both discs derive from the same larva. These results suggested to us that toy is involved in so10 regulation but because of a high variability of the toy transcript (data not shown), and therefore a high variability in eye disc size and shape (Fig. 1G-1I), it was not possible to asses a clear toy-dependent territory. To further corroborate an involvement of TOY in so10 regulation we switched to ectopic expression experiments using the UAS-GAL4 system (Brand and Perrimon, 1993). We performed experiments in an ey² mutant where ectopic expression of toy is not able to induce ey (Czerny et al., 1999). As shown in Fig. 1J, ectopic expression of TOY in ey² induces expression of the so10 enhancer. This led us to the conclusion that so10 enhancer might be regulated by both, EY and TOY in Drosophila, and that the residual staining found in ey² or ey^J5.71 eye discs could be in part due to the activation by toy.

To determine whether TOY directly binds to the so10 fragment in vitro, we performed bandshift experiments using full-length EY and TOY on a 128 bp so10 sub-fragment previously shown to be bound by EY (Niimi et al., 1999). As shown in Fig. 2A TOY is able to bind specifically to this fragment since a 10-fold molar excess of cold competitor inhibits binding. This result further enforced our finding that toy is also able to regulate so10 and encouraged us to study this regulation.

Pax6 proteins contain two DNA binding domains, the paired domain (PD) and the homeodomain (HD). It has recently been shown that the EY-PD is required to induce endogenous so expression and to direct eye development, whereas the EY-HD is dispensable for this process (Punzo et al., 2001). Therefore, we investigated whether the same mechanism applies to so10 regulation. We used EY- and TOY-deleted proteins in which the PD or the HD was missing. To ensure that the TOY constructs would not activate endogenous ey that in turn would activate so10, we performed the experiment in an ey² mutant background, as previously described for full-length toy. Ectopic expression of these proteins in the appendages by dpp^blink-Gal4 revealed that the HD deletion has no effect on so10 inducibility (Fig. 2E,H) when compared to the full-length proteins (Fig. 2D,G), whereas the removal of the PD domain abolished the ability of both EY and TOY to induce so10 expression (Fig. 2F,I). To ensure that the deleted proteins were expressed at comparable levels, we performed western blotting experiments probed with an anti-TOY or an anti-EY antibody, respectively. As shown in Fig. 2B,C the deleted proteins (asterisk) are expressed at comparable or lower levels than the full-length TOY or EY proteins. Despite a lower level for the HD deleted constructs these proteins still induce so10-lacZ whereas the PD-deleted protein does not. Altogether, these results demonstrate that EY and TOY activate the so10-lacZ transgene through their paired domain.

These findings prompted us to identify the PD binding sites of EY and TOY within this element. We performed in vitro DNase I footprinting experiments using His-tagged EY and TOY PD. As shown in Fig. 3B, three protected regions were found by using the EY-PD whereas five protected regions were found by using the TOY-PD. Whereas sites 3 and 4 are only protected by TOY, sites 1,2 and 5 are equally bound by EY and TOY in vitro. Therefore, we were not able to find specific binding sites protected exclusively by EY, in contrast to TOY. As shown in Fig. 3C, these protected sites reveal sequence similarity to the in vitro selected Pax6 consensus binding sequence (Czerny and Busslinger, 1995; Epstein et al., 1994). To address the question of the functional relevance of these binding sites, we mutated the different binding sites and studied the effect of the mutations on the enhancer activity. We mutated the bases fitting with the Pax6 consensus in order to lose the core homology (Fig. 3A). Since the Pax6 consensus is bipartite and highly degenerated it was necessary to introduce several mutations within each binding site to ensure the complete loss of EY-PD and TOY-PD binding, respectively (Fig. 3A: mutation of binding sites 1 to 5 referred to as so10^EY+TOYmt). In addition, after generating transgenic lines carrying the enhancer with all five binding sites mutated, we verified that this element was not inducible anymore by EY and TOY in wing (Fig. 3D) and leg discs (data not shown). These results strongly suggest a direct role of these proteins in so10 activation through the five identified binding sites.

To correlate the different binding properties of EY and TOY on the so10 enhancer to their function during eye development, we generated two additional so10-lacZ transgenic lines. One line carries mutations in TOY binding sites 3 and 4 referred to as so10^TOYmt. The other carries mutations in the EY and TOY binding sites 1, 2 and 5 referred to as so10^EY/TOYmt.

The expression of these different so10-lacZ constructs was further analyzed with regard to eye development during the larval stages. As shown in Fig. 4C and 4G, when the two TOY-specific binding sites 3 and 4 are disrupted (so10^{TOY mt}), the expression is maintained in the entire eye disc in early third instar larvae, and only later becomes restricted to the posterior side. The expression is completely lost in ey² mutants (data not shown) suggesting that the residual staining is due to EY activity. Disrupting the EY/TOY binding sites (so10^EY/TOYmt) consequently targets the expression of the transgene in the lateral edge of the eye disc (Fig. 4D,H). Interestingly, the expression patterns of so10^{TOY mt} and so10^EY/TOYmt mimic the expression patterns obtained with the wild-type so10 enhancer in a toy^G7.39 and in an ey² or ey^J5.71 mutant background (Fig. 1G, 1B and 1C, respectively). Thus during larval stages, completely removing EY gives the same result as disrupting the EY/TOY binding sites whereas partially removing TOY (hypomorph) leads only in some cases (Fig. 1G) to the same result as disrupting the TOY binding sites. Mutation of the five binding sites (so10^EY+TOYmt) abolishes any residual expression in the eye disc (Fig. 4E,I) suggesting a role for both EY and TOY in specifying expression of so10 in the eye disc.

To define functional sites within the so10 enhancer, we tried to correlate the different expression patterns observed to their requirements. We therefore asked whether the different mutated enhancers had the ability to rescue the so¹ mutant phenotype, which lacks both compound eyes and ocelli, and to what extent this phenotype can be rescued. We used the UAS-Gal4 system to generate transgenic flies expressing the Gal4 transcription factor under the control of the different mutated enhancers. Those lines were then crossed to a UAS-so line in a so¹ mutant background (Fig. 5A). First we confirmed that the different so10-Gal4 drivers directed expression of a UAS-lacZ reporter transgene in similar manner to that obtained with the so-lacZ constructs previously examined (data not shown). Then we tested if the wild-type so10 element is sufficient for rescuing the so¹ mutant eye phenotype. As shown in Fig. 5B, expression of the so cDNA under the control of the so10 fragment in a so¹ mutant rescued the compound eyes effectively but failed to rescue the missing ocelli. This result is consistent with the expression pattern of this element showing expression in the eye-specifying region but lacking expression in the ocellus region (Fig. 1A arrowhead). Thus, the transgene contains appropriate DNA regulatory sequences in order to direct and restore so expression in compound eye formation in a so¹ mutant.

We next asked how far the mutated Gal4 drivers were also able to restore proper so expression in a so¹ mutant background and therefore rescue eye development. As shown in Fig. 5C, mutations introduced in the TOY binding sites (so10^TOYmt) have only a minor effect since the eye is rescued to approximately the same size as the rescue obtained with the wild-type so10 enhancer. However, disruption of the EY/TOY binding sites (so10^EY/TOYmt), which largely inactivated the enhancer (Fig. 4D and H), gave hardly any rescue (Fig. 5D). Indeed, we observed in almost 100% of the offspring a partial rescue only, characterized by a strongly reduced eye on one side and no eye on the other side of the fly head. This phenotype is rarely seen in so¹ mutants. Interestingly, it is reminiscent and strongly resembles the eye phenotype observed in the ey² mutant (Halder et al., 1998). It has recently been shown by Punzo et al. (Punzo et al., 2001) that the variable eye phenotype is not due to a residual expression of ey. The expression of EY is not detectable in the eye disc of ey² mutants and the same phenotype is observed in the ey null mutant ey^J5.71. Thus, the presence of the TOY binding sites within the so10^EY/TOYmt enhancer allows a partial rescue, suggesting that TOY can overcome the lack of EY function to some extent. Therefore, this finding might explain why ey mutant flies have a strong variation in eye size. Finally, the so10^EY+TOYmt enhancer did not rescue the so¹ mutant phenotype (Fig. 5E). These results suggest a direct role for the EY/TOY binding sites in so10 regulation during compound eye development, whereas the TOY binding sites seem to contribute only little to the formation of the compound eye.

To further investigate whether TOY can overcome the lack of EY, we tried to rescue the eye phenotype of an ey mutant by TOY. Flies carrying either the UAS-ey or the UAS-toy cDNA were crossed to ey-enhancer-Gal4 flies in an ey^J5.71 mutant background and the offspring was examined. As shown in Fig. 5H, toy was also able to rescue an ey^J5.71 mutant eye phenotype if expressed under the control of the ey enhancer. This demonstrates that in the absence of ey, toy can partially complement ey, and suggests that the two Pax6 homologues have partially redundant functions.

Since the rescue obtained with the so10 enhancer was incomplete and resulted in a reduced compound eye (Fig. 5B), irregular ommatidia arrangement, lack of inter-ommatidial bristles and of ocelli (data not shown), we investigated whether additional regulatory sequences on the adjacent 1.2 kb fragment, encompassing the deletion in so¹, were required for proper so expression. Therefore, we generated a new so-Gal4 driver, which contained the 1.2 kb fragment adjacent to the 400 bp so10 element, in the same orientation as found in the endogenous gene. This construct was designated as so7 driver (the 1.2 kb fragment corresponds to so9, whereas so7 includes fragments so9 + so10) (Niimi et al., 1999). The so9 fragment was not activated by EY in a yeast one hybrid assay, whereas so7 containing the so10 was (Niimi et al., 1999). The mutated so7-Gal4 drivers were generated using the same mutations as previously described for the various mutated so10 enhancers, and are referred to as so7^EY/TOYmt, so7^TOYmt and so7^EY+TOYmt, respectively. As a control, we generated a lacZ line carrying only the so9 fragment to test whether this element is also expressed in the eye disc. As revealed by lacZ staining, a weak expression was detectable at the posterior margin of the eye disc (data not shown). When crossed to UAS-lacZ, the so7 driver extended the expression further into the ocellar region of the eye disc (Fig. 6B) than the so10 driver (Fig. 4F). The so7^EY/TOYmt line produced expression in the ocellar region but not in the eye-specifying region except at the posterior margin of the eye disc, which was caused by to the so9 fragment (Fig. 6C). This result indicates that the expression in the eye-specifying region depends on the presence of the EY/TOY binding sites 1, 2 and 5. Compared with the so10^EY/TOYmt (Fig. 4H), the so7^EY/TOYmt is not expressed in the lateral edges of the eye disc, showing that so9 restricts the activity of so10 through the TOY binding sites. The so7^TOYmt enhancer (Fig. 6D) shows a similar expression pattern as the so10^TOYmt enhancer (Fig. 4G) in the posterior part of the eye disc. In addition, the so7-enhancer-specific expression in the ocellar region is missing in so7^TOYmt indicating that it is dependent on the TOY binding sites. In summary, in the context of the so7 enhancer, disrupting the EY/TOY binding sites affects expression mainly in the eye field whereas disrupting the TOY binding sites affects the expression in the ocelli. Thus, so9 cooperates with the so10 enhancer to specify expression in the ocelli through the TOY binding sites. Finally, when all five binding sites were mutated (so7^EY+TOYmt) only staining reflecting the pattern of so9 was detected (Fig. 6E). Overall, the expression patterns of the so7-Gal4 lines confirmed our previous findings on the so10 element and further attested a different requirements for the EY and TOY binding sites.

To correlate the function of the so7 enhancers with different expression patterns, we performed a rescue experiment by restoring so expression using the UAS-Gal4 system in a so¹ mutant background. We crossed the different so7-Gal4 enhancers to UAS-so in a so¹ mutant background and tested their ability to rescue the so¹ mutant phenotype. As shown in Fig. 6F, so7 was able to fully rescue the compound eyes and partially rescue the ocelli (Fig. 6J). so7^EY/TOYmt was not able to restore compound eye development (Fig. 6G) but was still able to direct ocelli formation (Fig. 6K). In contrast, so7^TOYmt rescued compound eye formation (Fig. 6H) at the same efficiency as the wild-type enhancer, but did not rescue ocelli development (Fig. 6L). so7^EY+TOYmt was unable to rescue either the compound eye or the ocelli phenotype (Fig. 6I and M). This demonstrates that the EY/TOY binding sites are absolutely required for compound eye development whereas the TOY binding sites are required for the development of the ocellus. Thus, these results suggest a distinct functions for the two Drosophila Pax6 proteins in regulating the same target gene during compound eye and ocellus development.

Genes involved in eye development are highly conserved between vertebrates and invertebrates. Given the complex genetic network during early eye development, studies of regulatory sequences controlling eye-specific expression provide insight into eye specification. Analyzing the expression of the eye-specific so10 enhancer in different mutant background gives information about the regulatory circuit. The loss of so10-lacZ expression in the ey², ey^J5.71 or toy^G7.39 mutants indicates, that EY and TOY are required for activation of this element. In the case of so and eya, the analysis in the null mutant was not possible due to embryonic lethality of these mutants. Since the so¹ and eya¹ mutations eliminate expression of so and eya transcript and protein in the eye disc, but not elsewhere in the organism, these mutants were defined as eye-specific nulls (Bonini et al., 1993; Cheyette et al., 1994). These findings render them appropriate for the analysis in the eye, especially since so¹ specifically deletes those intronic sequences that contain the so10 enhancer. The results in the so¹ and eya¹ mutants clearly show that the maintenance of activity of this element is dependent on the presence of both proteins posterior to the morphogenetic furrow. These data complement the findings obtained in ectopic eyes and fit with the current model previously published on the regulation of retinal differentiation (Bonini et al., 1997; Chen et al., 1997; Czerny et al., 1999; Pignoni et al., 1997). Whereas EY and TOY seem to be required for activation of the so10 eye-specific enhancer, an interacting network between SO, EYA, EY and TOY seems to be responsible for maintenance and proper expression in differentiating cells. In ectopic eye development, the genetic pathways governing these events have been shown to involve feedback loops so that all these genes regulate the expression of each other. Our findings indicate that it is true in the normal eye as well.

We have studied in more detail the so10 eye-specific enhancer during eye morphogenesis. These studies demonstrate that this element, which is deleted in the so¹ mutant is regulated by both Pax6 proteins in Drosophila. Using gel shift and DNase footprinting experiments we demonstrated that TOY and EY are able to bind specifically to this enhancer in vitro. Furthermore, we showed that so10 is also activated ectopically by toy in an ey² mutant background suggesting a direct activation by toy. Our rescue assays performed with the different so10 mutant drivers, which are exclusively expressed in the compound eye field but not in the ocelli, reflect the requirement of the EY/TOY binding sites and to a lesser extend those of TOY for compound eye development. The phenotype obtained with the so10^EY/TOYmt driver nicely parallels the variability in eye size observed in ey mutants (Halder et al., 1998). Moreover, expressing toy in the ey domain of the eye disc by using the ey enhancer Gal4 driver, established that toy could partly complement the lack of ey activity. These findings lead us to propose that the variable eye phenotype seen in ey² mutant is due to the presence of toy rather than a residual expression of ey. In this respect, we did not succeed in detecting any ey transcript or protein in the ey² mutant eye disc (Callaerts et al., 2001; Punzo et al., 2001; Quiring et al., 1994). Furthermore, the ey null mutant ey^J5.71 also has a variable eye phenotype (Callaerts et al., 2001). Thus, the two Pax6 proteins share similar biological properties to direct eye development.

The incomplete rescue obtained with the so10 driver led us to search for additional cis regulatory sequences. We found that the so9 fragment adjacent to the so10 enhancer, which has been shown not to be inducible by EY (Niimi et al., 1999), contains cis regulatory sequences acting in combination with the so10 eye-specific enhancer to ensure correct expression in the eye. This combined fragment (so7=so10 + so9), which encompasses the so¹ deletion, is able to ensure complete rescue of the compound eye and partial rescue of the ocellus when used as a Gal4 driver. The mutated so7 drivers confirmed our previous findings on the EY/TOY binding sites of the so10 enhancer with regard to compound eye formation and further indicated a functional difference for the TOY binding sites since the latter are absolutely required for ocellar development. This further demonstrates that TOY requires additional transcription factors present in the eye disc for its proper function. Thus, the different biological properties of EY and TOY previously described are not limited to their DNA binding specificities (Czerny et al., 1999) but extend to their ability to cross talk with different sets of proteins.

Beside its role in ocelli specification, the so9 fragment also exerts a finely tuned effect on the so10 activity in the compound eye, since so7^EY/TOYmt is more down-regulated in the eye field than so10 (Fig. 6C). These expression patterns are in line with the rescue experiments; there is no rescue of the compound eye with so7^EY/TOYmt (Fig. 6G), whereas so10^EY/TOYmt gives at least a partial rescue (Fig. 5D). The compound eye is fully rescued with the wild-type so7 driver and partly rescued with the so10 driver.

Using our in vitro approach, we could not define any EY-specific sites since the sites 1, 2 and 5 are equally bound by EY and TOY. These sites might be more EY-specific since the residual expression of the so10^TOYmt is lost in ey² eye discs. In addition, so10 expression in ey mutant eye discs is similar to the so10^EY/TOYmt expression in wild-type eye discs. Nevertheless, we cannot exclude the possibility that both proteins bind in a complex combinatorial arrangement to these sites in a wild-type eye context. Alternatively, it is also possible that these sites will be occupied by either EY or TOY depending on which protein is present in the nucleus. In this respect, the efficiency of the ocellar rescue is relevant. The rescue obtained with the so7^EY/TOYmt lacking binding sites 1, 2 and 5 is reduced qualitatively (ocelli smaller) and quantitatively (40% of the flies show a rescue) as compared to the rescue efficiency obtained with the wild-type enhancer (80% of rescue efficiency). This strongly suggests a requirement of all five binding sites for TOY to ensure full enhancer activity in the ocelli.

The footprinting experiment did not reveal any binding of the EY-PD onto the binding sites 3 and 4. Since Czerny et al. (Czerny et al., 1999) demonstrated a reduced binding activity for the EY-PD when compared to the TOY-PD on the TOY binding sites located in the ey enhancer, we cannot exclude that the EY-PD might also bind weakly to the TOY-BS 3 and 4 of the so enhancer in vivo.

All animals, ranging from flatworms to mammals, analyzed so far, have a Pax6 gene which from our current state of knowledge, is universally required for eye specification. In contrast to vertebrates, where generally a single Pax6 gene gives rise to several differentially spliced transcripts, Drosophila and other holometabolous insects have two Pax6 genes, raising the question of functional redundancy. Gene duplication and subsequent divergence of developmental control genes is a major driving force in evolution increasing the diversity and complexity of the organisms. A second mechanism for recruiting additional genes into a developmental pathway is enhancer fusion (Gehring and Ikeo, 1999). The acquisition of new cis-regulatory elements represent an important mechanism for functional diversification (Bouchard et al., 2000; Greer et al., 2000; Hanks et al., 1995; Suda et al., 1999; Xue and Noll, 1996). Our findings strongly support both of these hypotheses since toy is able to rescue the eye development in an ey mutant when expressed in the ey domain. The finding that ey and toy exhibit different expression patterns during embryogenesis might account in part for their functional biological diversity (Czerny et al., 1999; Kammermeier et al., 2001). In the eye, both genes are co-expressed, except for the ocellar territory where only toy is expressed (Czerny et al., 1999). In addition, it has been proposed that TOY and EY diverged to regulate different sets of target genes because of a N14G mutation that changes the DNA binding specificity of the PD domain of ey (Czerny et al., 1999). Indeed, using the so10 regulatory element we found that TOY does not bind to the same sequences as EY, but interestingly, TOY and EY regulate the same target enhancer in different cells. The phenotypes obtained in our rescue experiments using either the EY/TOY or TOY binding site mutated enhancers, nicely parallel the phenotypes observed in those mutants. The ey null mutant still has ocelli but lacks compound eyes. Interestingly escapers from the recently isolated toy mutant in our laboratory (toy^G7.39) exhibit no eye reduction whereas the ocelli are partially missing (S. F., U. Kloter and W. J. G., unpublished). Therefore, removal of the common target gene of both Pax6 proteins in the eye (e.g. so¹ mutant) consequently leads to a loss of both compound eyes and ocelli. Therefore, we propose that one of the developmental programs of toy is in part to specify ocellar development in addition to head formation, since toy mutants generated are characterized by pupal lethality, pharate adults lacking half of the head or the entire head capsule (Kammermeier et al., 2001). Thus, we can propose that the so gene is regulated by toy to specify the ocelli and by ey to specify the compound eyes during larval development.

Our analysis of ey and toy allows us to dissect the evolutionary changes after the gene duplication event that has happened during insect evolution. First, the cis-regulatory regions of the two genes have diverged, leading to both temporal and spatial changes of expression; toy is expressed much earlier than ey during embryogenesis, whereas ey is not expressed in the ocellar region of the larval eye disc (Czerny et al., 1999). Second, the protein coding regions of the two genes have diverged, most importantly in the paired domain where asparagine 14 which is present in most Pax6 homologs, has been mutated in ey to glycine, which changes the DNA binding properties of the protein significantly (Czerny et al., 1999). Third, the positive autocatalytic feedback loop found in vertebrates for their single Pax6 gene (Chow et al., 1999; Grindley et al., 1995; Okladnova et al., 1998; Plaza et al., 1995), has evolved into a heterocatalytic control loop in which toy transcriptionally activates ey by binding to the eye-specific enhancer of ey (Czerny et al., 1999; Hauck et al., 1999). Fourth, both toy and ey cooperate in differentially regulating the so target gene, reflecting the fact that earlier in evolution so was regulated by a single Pax6 gene. These findings strongly support the hypothesis of intercalary evolution (Gehring and Ikeo, 1999) showing that the ey gene has been intercalated into the eye developmental pathway between toy and so. The observation that toy activates ey in the eye progenitor cells of the embryo, where neither so and eya are expressed (Kumar and Moses, 2001), indicate that toy and ey are acting high up in the genetic hierarchy leading to eye development.

Our current model of eye development (Fig. 7) shows that the eye specifying genes are expressed and sequentially regulated during different stages of development. Our study demonstrate a distinct role for TOY and EY in so regulation to specify ocelli and compound eye development, respectively.

We acknowledge U. Walldorf for the rabbit anti-EY and the rabbit anti-TOY antibodies. We also would like to thank C. Dozier for critical reading of the manuscript and L. Michaut for helpful discussion. This work was supported by the Kantons of Basel-Stadt and Basel-Land and a grant from the Swiss National Science Foundation. S. P. was supported by the Centre National de la Recherche Scientifique and by an EMBO fellowship.