Identification of functional sine oculis motifs in the autoregulatory element of its own gene, in the eyeless enhancer and in the signalling gene hedgehog

The Drosophila visual system consists of two compound eyes and three ocelli, which are simple eyes located on the adult vertex(Stark et al., 1989). Both types of optical organs develop from a small number of cells that are set aside during development in the early embryo. These cells form the eye part of the eye-antennal imaginal disc and proliferate during the larval stages. The compound eye emerges from the central part of the eye imaginal disc, whereas the ocelli develop from the anterior-medial region. The compound eye in Drosophila consists of a precisely organized array of approximately 750 ommatidia, each containing eight photoreceptor neurons and 12 accessory cells. The ommatidia begin to form in the early third instar larva, when the morphogenetic furrow (MF), a wave of pattern formation marked by an indentation, moves across the eye disc from posterior to anterior (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).

Determination of the eye primordium requires several nuclear proteins that are known to act as transcriptional regulators. The Drosophila Pax6gene eyeless (ey) was the first gene shown to display the capacity to induce ectopic eye morphogenesis upon ectopic expression(Halder et al., 1995). A second Drosophila Pax6 gene, twin of eyeless (toy),like ey encodes a protein with two DNA-binding domains(Czerny et al., 1999). Further genes involved in early eye determination are eyegone (eyg),which also shows some similarity to Pax6(Jun et al., 1998; Chao et al., 2004; Dominguez et al., 2004), eyes absent (eya) and dachshund (dac),both encoding nuclear proteins (Bonini et al., 1993; Mardon et al.,1994), and sine oculis (so)(Cheyette et al., 1994). Analyses of the expression patterns of these genes combined with genetic approaches have revealed a complex genetic regulation network during compound eye development. toy is the first of the mentioned genes to be expressed during embryogenesis and activates ey in the eye primordium(Czerny et al., 1999). so is required later for the development of the entire visual system,including the compound eyes, the ocelli, the optic lobe of the brain and the larval photoreceptors designated as Bolwig's organ(Cheyette et al., 1994; Serikaku and O'Tousa, 1994; Pignoni et al., 1997). eya expression comes up later in the compound eyes and can be found in the ocelli-specifying region in third instar eye imaginal discs. Recently, eya has been shown to have protein phosphatase activity(Li et al., 2003; Tootle et al., 2003). so and eya are both required for compound eye and ocellus formation, as the respective mutants lack both visual systems(Cheyette et al., 1994; Zimmerman et al., 2000). so, eya and dac have been shown to be regulated by ey (Halder et al.,1998; Niimi et al.,1999; Zimmerman et al.,2000). SO and DAC have been proposed to function as co-factors for EYA, and genetic studies in Drosophila have demonstrated synergistic interactions between so, eya and dac during eye development(Chen et al., 1997; Pignoni et al., 1997). The respective protein complexes feed back on ey expression and eya and dac, like ey and toy, are capable of inducing ectopic eye morphogenesis(Bonini et al., 1993; Bonini et al., 1997; Pignoni et al., 1997).

Although much knowledge has been gathered during the last years about the complex genetic network that orchestrates eye development, only a small number of observed regulatory interactions have been analysed down to the level of DNA-protein interactions. Analysis of further components controlling expression patterns of genes involved in early eye development should therefore provide important details on the genetic hierarchy that mediates eye specification and may help to identify direct targets of the known eye specification genes.

Among the already described direct interactions, toy has been shown to induce ey expression by an eye-specific enhancer in embryonic eye precursor cells, but not during larval stages in the later emerging eye imaginal disc (Czerny et al.,1999). However, ey, together with toy, directly regulates so expression by an eye-specific enhancer that is deleted in the so¹ mutant allele(Niimi et al., 1999; Punzo et al., 2002). Furthermore, ey-regulated, eye-specific enhancers have been identified using deletions within the eya gene locus(Zimmerman et al., 2000).

In this study, we address the regulatory potential of a previously described so7 enhancer fragment during ocellar morphogenesis. So7 represents the DNA fragment that is deleted in the so¹mutation and contains the ey- and toy-regulated enhancer element so10 (Punzo et al.,2002). We show that a 27 bp fragment within so7, soAE, is sufficient to expand expression of a reporter gene to the ocellar region when fused to the so10 enhancer and, consequently, this so10-soAE enhancer fragment is sufficient to rescue the eyeless and the ocelliless phenotype of so¹/so¹ flies when used as a driver for so. Furthermore, we show that soAE is a direct target of so in compound eye and ocellar development and that the autoregulatory feedback of so on its own expression is required for the ocellus-specific expression of so.

By analysing the DNA-binding specificity of SO in more detail, we were able to identify those nucleotides that are essential for SO-soAE interaction. Using the emerging cis-regulatory signature for so-dependent regulation, we performed a genome-wide search for additional putative so-target genes. Sequences that fit our selection criteria were identified in the ey and hedgehog (hh) loci. We show that both these genes contain eye-specific enhancers that are directly regulated by so. Our results emphasize the importance of autoregulatory feedback loops in morphogenesis and development.

Flies were reared on standard medium at 25°C. Lines used:UAS-so (Pignoni et al.,1997), UAS-eya (Bonini et al., 1997), so10-lacZ(Niimi et al., 1999),so7-lacZ, so9-lacZ, so10^EY+TOYmt-lacZ(Punzo et al., 2002), dpp^blink-GAL4(Staehling-Hampton and Hoffmann,1994), ey-GAL4(Halder et al., 1998), FRT42D, so³/CyO(Pignoni et al., 1997), eyFLP (Newsome et al.,2000), FRT42D, ubiquitinGFP(Duchek et al., 2001). so²/so² (Bloomington Stock Center). Clones of homozygous so³ mutant cells were generated by the expression of FLP recombinase under the control of an eyenhancer.

Specific genotypes generated: (1) eyFLP; FRT42D,ubiquitinGFP, (2) so²/so²;so10-soAE-lacZ, (3) so²/so²;so7-lacZ, (4)UAS-so/UAS-so;UAS-eya/UAS-eya and (5) so¹/so¹; so10-soAE-so.

lacZ reporter plasmids and rescue constructs were introduced into w¹¹¹⁸ by standard P-element transformation procedures. Three to 10 independent transgenic lines were established for each construct and tested for expression or rescue potential.

Antibody staining on discs was performed according to Halder et al.(Halder et al., 1998). Primary antibodies were anti-EyaMab10H6, 1:10(Bonini et al., 1997), Rabbit anti-β-galactosidase, 1:500 (Promega). Secondary antibodies used were from Jackson ImmunoResearch Laboratories: Cy5 α-rabbit (1:400), Alexa586α-mouse (1:400).

To detect β-galactosidase activity, third instar larval imaginal discs were fixed and subjected to a standard X-gal colour reaction for 2 hours at 37°C.

Inserts of the reporter constructs were obtained by PCR, using so7 as a template, and subcloned into the lacZ pCβ vector(Niimi et al., 1999). For the rescue construct a modified pUAST vector(Brand and Perrimon, 1993) was used. The 5 × UAS sequence was replaced by so10-soAE. so cDNA was placed downstream of hsp70 within the polylinker resulting in so10-soAE–hsp70–so in the pUAST backbone(so10-soAE-so).

For the constructs: ey enhancer, B4M and B4M SOmut, the sequences given in Fig. 5 were used. A BamHI and a KpnI site was added at the 5′ and 3′-end, respectively, and used for subcloning into the lacZpCβ vector.

The hh¹ (bar-3) sequence was obtained by PCR on genomic DNA of wild-type (wt) flies by using the following primer set:5′-CTGTGCGCTCGAGTGGGCCACACAGGGTGGG-3′; rightward orientation,5′-CGGCCCGTCTCAGATCTCGGATCTGAGATC-3′ leftward orientation. Mutations were introduced by PCR. For the deletion construct hh¹ Δ5′,5′-GGGGTACCCAAGACAAGTAATCCCCCACCCTCGC-3′ was used as rightward oriented primer (the SO site is mutated by changing GAG to CCC).

Genomic DNA was amplified by PCR from so²/so² flies and sequenced. The sequences were confirmed on independent amplification events. Genomic DNA isolation was performed according to Bui et al.(Bui et al., 2000a). Primers used for mapping the so² deletion were:5′-GAAGGGCACTGCTTACTGAGAGCTCG-3′,5′-GCCCATCGAATCCGCATCTCCCCCAG-3′ rightward orientation;5′-GCGCACACTCGACAAATTTGCGATCTGGC-3′ leftward orientation. Primers are located at positions 2355, 3116 and 6218, respectively, within the last intron. Nucleotides 3983-5181 are deleted in so² (the first nucleotide of the last intron is set as 1).

Southern blotting was performed according to Sambrook and Russel(Sambrook and Russel, 2001). Genomic DNA was digested using ClaI, EcoRV and XhoI. As probes, DIG-labelled PCR products of so10 and so9 were used. so10 and so9 are described previously(Punzo et al., 2002).

Drosophila S2 cells were maintained in Schneider's insect medium(Invitrogen) supplemented with 10% fetal calf serum and were transfected with the Effectene Transfection Reagent (Qiagen). For reporter gene assays 2×10⁶ cells were transfected with a total of 200 ng plasmid DNA (20 ng reporter plasmid, 5 ng of a plasmid constitutively expressing firefly luciferase, the indicated amounts of expression plasmids and the parental vector pAc5.1B/V5His, to bring the total amount of DNA to 200 ng). Cells were lysated 48 hours after transfection and lysates were assayed forβ-galactosidase and luciferase activity as described previously(Muller et al., 2003).

Radioactively labelled probes were generated by annealing and filling in partially overlapping oligonucleotides in the presence of(α-³²P)ATP. Binding reactions were carried out in 20 μl of 100 mmol/l KCl, 20 mmol/l HEPES pH 7.9, 20% glycerol, 1 mmol/l DTT, 0.3% BSA,0.01% NP40 containing 10,000 cpm probe and 1 μg dIdC. As a protein source,full-length SO protein was synthesized in reticulocyte lysates using the T7 promotor according to the manufacturer's specification (Promega). For the binding reaction, 1 μl of a standard 50 μl reaction was used. After incubation for 30 minutes at 4°C, the reactions were analysed by non-denaturing 6% polyacrylamide gel electrophoresis followed by autoradiography. For the cold competition experiments, the proteins were first incubated with a 100 × molar excess of unlabeled double-strand oligonucleotides for 10 minutes at RT, followed by incubation with the radiolabelled probe at 4°C for 30 minutes.

Putative so target genes were identified by screening the entire Drosophila genomic sequence with the consensus GTAANYNGANAYS using the program FLY ENHANCER [freely available at http://flyenhancer.org(Markstein et al., 2002)].

Alignments of different Drosophila species were obtained from http://hanuman.math.berkeley.edu/genomes/drosophila.html.

A 1.6 kb enhancer fragment (so7, Fig. 1A) spanning the genomic region deleted in so¹is able to recapitulate the expression pattern of so in third instar eye imaginal discs when driving a lacZ reporter gene(Punzo et al., 2002). Furthermore, so7 is able to completely rescue the eyeless, and partially the ocelliless, phenotype of so¹ mutant flies when driving the so gene (Punzo et al.,2002).

So10 (400 bp) and so9 (1.2 kb) (Fig. 1A) are subfragments of so7. so10 mediates expression in the compound eye part of third instar eye-antennal imaginal discs and contains the previously described ey- and toy-specific binding sites(Fig. 1D). These include five binding sites bound by toy. Three of these are also binding sites for ey and are important for compound eye development, whereas the two toy-specific sites are required for ocellar development(Niimi et al., 1999; Punzo et al., 2002). Consistent with its expression pattern, so10 is able to rescue the eyeless phenotype but not the ocelliless phenotype of so¹ mutant flies (Punzo et al.,2002).

so9-mediated expression appears at the posterior margin of the eye disc(similar to Fig. 1C). When combined with so10, so9 provides additional transcriptional input to expand the expression to the ocellar region.

Trans-acting factors that bind the cis-regulatory so9 element and cooperate with toy to confer expression in the ocellar region were unknown when this work was started. In order to locate the binding sites of such additional transcription factors we first aimed at the isolation of the smallest version of so9 that still would be able to drive expression of a lacZ reporter to the ocellar region of eye imaginal discs when combined with so10 (Fig. 1A). Our search resulted in the identification of a fragment as small as 27 bp(Fig. 1A, number 21), which in the following text will be referred to as soAE (sine oculisautoregulatory element).

The expression pattern mediated by a combined so10-soAE-element(Fig. 1A, number 14 and Fig. 1B) was indistinguishable from expression mediated by so7, whereas soAE alone resembled the expression pattern of so9 (Fig. 1A, number 21 and Fig. 1C). In addition,so10-soAE driving so is sufficient to rescue both the eyeless and ocelliless phenotype of so¹ mutant flies(Fig. 1G). Therefore, soAE contains all regulatory elements that are sufficient for so9-mediated expression. Further evidence for the functional relevance of this sequence came from the comparison of D. melanogaster and the genomes of six other Drosophila species in which the soAE sequence shows a high degree of conservation (see Materials and methods).

In soAE three sequence motifs can be found that are reminiscent of well-known transcription-factor-binding sites. These are a motif related to the Pax6-consensus-binding site(Epstein et al., 1994), a TAAT-motif that is a hallmark of most homeodomain recognition sequences and a GATA-motif. We mutated these sites, and tested the respective fragments(so10-mutPAX, so10-mutHD, so10-mutGATA) for the resulting expression patterns.

so10-mutPAX-mediated expression was indistinguishable from the so10-soAE expression pattern (Fig. 2C). Conversely, mutating the putative homeodomain-binding site (so10-mutHD) or the GATA sequence (so10-mutGATA) resulted in loss of reporter gene expression in the ocellar region (Fig. 2A,B).

We then oligomerized soAE four times, to boost its expression. As a result,an expression signal became apparent posterior and slightly in front of the MF(Fig. 2E) as well as in the optic lobe (data not shown). However, 4xsoAE was not able to drive expression in the ocellar region. Additional copies of soAE did not lead to a further strengthened expression. Expression of 10xsoAE, for example, appears blotchy and weaker in the eye disc than expression of 4xsoAE(Fig. 2F).

As the expression pattern of 4xsoAE is reminiscent of so-expression in the eye disc, we hypothesized that soitself might be the soAE regulating factor. Both expression patterns show a signal in the optic lobe as well as posterior to, within and in a few cells in front of the MF. The only difference is the ocellar expression of so,which cannot be seen using the 4xsoAE reporter construct.

The idea that so itself is the soAE-binding factor was further supported by previous work in which Hazbun et al. showed that SO binds in vitro to (C/T)GATA (Hazbun et al.,1997), a motif that is present in soAE(Fig. 4C nt. 7-11).

To determine experimentally whether the expression pattern of the mutated fragments correlates with the ability of these fragments to bind SO in vitro we performed electrophoretic mobility shift assays (EMSAs). SO protein was able to shift radiolabelled mutPAX but failed to bind to mutHD and mutGATA DNA fragments (Fig. 2H; see also Fig. 4A).

These results, in combination with our in-vivo data, strongly suggest that so itself is responsible for the ocellus-specific expression of so10-soAE.

To further test this hypothesis, we moved on to a genetic approach. so² is a hypomorphic allele that originated as a spontaneous partial reversion of so¹(Lindsley and Zimm, 1992). Different from so¹ adult flies, which completely lack compound eyes and ocelli, so² flies develop compound eyes that range from normal appearance to slightly reduced shapes but still lack ocelli completely. In so²/so¹ flies,eyes are of intermediate size (Heitzler et al., 1993). Because of the common origin and the genetic interaction of these two alleles, we tested if there is a mutation in so² flies that affects the genomic so9/so10 sequences. Using PCR on genomic so² DNA, we found a deletion of 1.2 kb that indeed affected so7. We further confirmed this result by Southern blotting (data not shown). The deletions of so¹ and so² partially overlapped(Fig. 1A) and in so², four of the five previously described Pax6-binding sites (Punzo et al.,2002) were missing. In fact, both binding sites exclusively recognized by TOY were deleted. According to Punzo et al.(Punzo et al., 2002), these toy-specific binding sites within the so10 enhancer fragment are required for ocellus development. The sequence representing so9, which contains the soAE fragment, appeared not to be affected by the so² deletion.

Next we took advantage of so² mutant flies to test whether the cis-regulatory potential of soAE depends on SO protein in vivo. Therefore we analysed so7- and so10-soAE-mediated expression in the ocellar region in so² mutant flies. As expected,so7-lacZ and so10-soAE-lacZ expression was lost in the ocellar region of so² mutant flies(Fig. 3D), supporting the idea of so being required for the ocellus-specific expression of so7 and so10-soAE further. The absence of reporter gene expression cannot be explained by a loss of ocellus-specific precursor cells, as eya expression,which represents a marker for this specified cell population, was detectable in so² mutant flies in the prospective ocellar region(Fig. 3E).

Taken together, toy and so binding to so10 and soAE,respectively, seem to cooperatively drive so-expression in the ocellar region of third instar eye discs.

To further examine the hypothesis that so autoregulation is important for ocellus development, we did the reverse experiment by mutating the SO-binding site of so7 (so7mut). So7mut-lacZ expression was hardly detectable in the ocellar region in a wt background(Fig. 1E) and resembled expression of fragments number7 and number11(Fig. 1A).

These data strongly suggest that feedback of so on its own enhancer is needed for ocellus development.

To assess whether soAE is a target of so also in the compound eye part of the eye disc, we tested the expression of the 4xsoAE reporter construct in cells homozygous for so³, a null allele of so (Cheyette et al.,1994). so³ mutant cells, however, tend to overproliferate, fail to differentiate into neurons and subsequently die(Pignoni et al., 1997). Hence,to be able to analyse reporter gene activity in living cells within so³ clones we tested them for eya expression. Eya is a suitable marker for viable cells in so³mutant clones for the following reasons. First, so and eyaare both targets of ey and show the same expression pattern in third instar eye discs (Halder et al.,1998; Niimi et al.,1999; Bui et al.,2000b). Both are expressed in a few cells anterior to the MF,within the MF, and in the differentiating photoreceptors posterior to the MF(Curtiss and Mlodzik, 2000). Second, SO and EYA proteins form a complex that works as a transcriptional activator when the proteins are co-expressed(Pignoni et al., 1997; Silver et al., 2003). Third, so¹ mutant eye discs still express eya, whereas in eya¹ mutants, expression of so is lost(Halder et al., 1998). Finally, so can be induced by eya in third instar eye imaginal discs (Curtiss and Mlodzik,2000). For these reasons we assume eya-positive-cells of third instar eye discs also express so during normal development. Therefore, only eya-expressing cells within so³clones were examined in our assay. In fact, in eya-expressing cells within so³ clones, expression of the 4xsoAE reporter construct was lost (Fig. 3F-I,the clones are negatively marked by the absence of ubiquitin-GFP expression; Fig. 3G). This strongly suggests that SO protein in general is required for activation of the soAE element in the eye field.

To further analyse whether soAE is a general in-vivo target of SO we tested reporter gene activity as a result of ectopic eye induction. so on its own is not able to induce ectopic eyes. By contrast, eya alone,synergistically strengthened by so, is sufficient to induce ectopic eye development on antennae, wings and legs(Pignoni et al., 1997).

We induced ectopic eye development by combining a dpp-GAL4 driver with UAS:so, UAS:eya or both of them and tested whether the reporter construct 4xsoAE-lacZ was induced ectopically. As expected,ectopic so alone did not result in reporter gene activity, whereas eya alone or eya combined with so in a synergistic manner was able to activate the reporter construct in wing discs(Fig. 3A-C).

In another in-vitro approach, we took advantage of Drosophila S2 cells to address whether SO and EYA proteins work cooperatively as a complex on soAE DNA to induce transcription. Consistent with the in-vivo data, our in-vitro results using S2-cells showed that SO, which has DNA-binding properties but lacks a transactivation domain, on its own was not able to activate soAE-mediated lacZ expression(Fig. 3J). Likewise, EYA, which contains a transactivation domain but lacks DNA-binding properties, also failed to induce transcription in S2 cells when expressed alone(Fig. 3J). Only when co-expressed, SO and EYA cooperatively worked as transcriptional activators on soAE (Fig. 3J). Interestingly,both SO and EYA mediated weak transactivation when the oligomerized mutated sites mutHD and mutGATA were used (Fig. 3J), despite the fact that these sites do not mediate transgene expression in vivo in the developing ocellus(Fig. 2A,B).

To date, there is only one direct target of so described in Drosophila, which is the Runx class transcription factor lozenge (lz) (Yan et al., 2003). Consistent with a previous invitro study that addressed the DNA specificity of the SO homeodomain(Hazbun et al., 1997), the authors show that the sequence (C/T)GATA plays a crucial role in SO-DNA interaction. Another study reports that SO together with EYA is able to transactivate by binding to an AREC3/Six4-binding site in cell culture. This motif, however, diverges to some extent from the C/TGATA-motif(Fig. 4C)(Silver et al., 2003).

Our soAE fragment harbours a CGATA motif, which is consistent with the SO-binding consensus of the lz promotor. In our experiments, however,also mutations upstream of this GATA core motif(Fig. 4C nt. 8-11) were able to abolish expression of the reporter construct in vivo and also impaired the capability of SO to shift DNA fragments in the EMSA. This observation suggested that additional sequences upstream of the GATA motif are also necessary for SO binding to its target site.

Therefore we decided to elucidate the sequence specificity of SO-DNA binding by analysing a systematic series of point mutations for their competitive effect on protein-DNA complex formation(Fig. 4A).

These in vitro experiments revealed a stretch of 13 nucleotides to be important for protein-DNA interaction of SO. There are three nucleotides, G,A, A at positions 1, 4, 9, respectively(Fig. 4A lanes 9, 12, 17 and Fig. 4C nt. 1, 4, 9), that appear to be most important for the interaction. These nucleotides, which show the strongest effects upon mutation, are found in the AREC3/Six4-binding site and are also substituted in the constructs so10-mutHD and so10-mutGATA. This provides strong evidence that these nucleotides are also important for soAE-mediated reporter gene expression in vivo (Fig. 2A,B).

Combining our in-vitro data on the autoregulatory element with the known so target sequence of lz and the AREC3/Six4-binding site, we defined the consensus sequence GTAANYNGANAYC/G as necessary for SO binding to DNA. This consensus sequence was taken as a basis for scanning the Drosophila genome for similar sites (see Materials and methods). In total, 1632 putative so targets emerged from this survey. Out of the affected genes several candidates are already known to be involved in eye development. In the following we will describe two of the genes that we picked for further analysis: ey and hh.

The first soAE similar element that caught our attention was located within the previously described eye-specific enhancer of the ey gene(Czerny et al., 1999; Hauck et al., 1999). A positive feedback loop already has been postulated on the basis of the fact that ey is induced in ectopic eye development upon co-expression of so and eya (Pignoni et al., 1997). Furthermore, the ability of so and eya to induce ectopic eyes is lost in ey² mutants(Pignoni et al., 1997). In ey² mutant flies, the previously mentioned eye-specific enhancer of ey is disrupted by insertion of a transposable element(Quiring et al., 1994) (see also Fig. 5A). These experiments genetically show that so and eya are able to feedback on ey and that this feedback loop relies on the eye-specific enhancer of the ey gene. However, a direct interaction between SO,EYA and the ey-enhancer has not been previously demonstrated.

The fact that the potential so target site within the eye-specific enhancer is perfectly conserved between D. melanogaster, D. pseudoobscura and two other Drosophila species (see Materials and methods), encouraged us to perform additional assays to obtain molecular evidence for a direct interaction.

First we showed that oligonucleotides containing this sequence were strong competitors for the binding of SO to soAE in EMSA, whereas this competing potential was lost when the GAT core (Fig. 4C nt. 8-10) of the sequence was mutated(Fig. 4B, eyeless and eyeless mut). We then compared the expression pattern of different mutated versions of a 160 bp fragment, comprising the eye-specific ey-enhancer, driving a lacZ reporter (sequences shown in Fig. 5A). The wt enhancer mediated expression posterior to the MF(Fig. 5B) (see also Hauck et al., 1999)(Fig. 4D). By mutating the Pax6 sites, expression in the eye disc was reduced to the posterior margin (Fig. 5C, observed in all four transgenic lines that were tested) (see also Hauck et al., 1999)(Fig. 4F). Mutating the so site and the Pax6 sites further reduced expression in the eye disc (one transgenic line showed no pattern at all, five independent transgenic lines showed weak activity similar to Fig. 5D). These data indicate that so directly regulates ey expression through the eye-specific enhancer of the ey gene.

hh encodes a secreted signalling protein that plays an important role in patterning the Drosophila eye field. Many lines of evidence suggest that hh signalling is required for the initiation and the propagation of the MF. Accordingly, hh is expressed at the posterior margin of the eye imaginal disc prior to photoreceptor differentiation and in cells posterior to the MF during its progression(Borod and Heberlein, 1998). Loss of hh function blocks initiation of the MF and impedes its progression (Borod and Heberlein,1998). Posterior margin clones of a null allele of smoothened (smo), the cell-autonomous receptor of hh signalling, lack differentiated photoreceptors(Greenwood and Struhl, 1999; Curtiss and Mlodzik, 2000). Conversely, ectopic hh expression anterior to the MF gives rise to a progressing ectopic MF (Heberlein et al.,1995; Pignoni and Zipursky,1997).

so and eya also have been shown to be required for initiation and propagation of the MF(Pignoni et al., 1997), and both are expressed at the posterior margin before initiation and later in front of the MF (Bonini et al.,1993; Serikaku and O'Tousa,1994). Furthermore, ectopic MFs are found in ectopic eyes induced by so together with eya(Pignoni et al., 1997). These data suggest that a feedback loop between hh and so/eyamight influence the proper initiation and propagation of the MF. Consistent with that, hh fulfilled our criteria to be a putative SO target. Both sites found within the hh locus showed almost perfect conservation among seven Drosophila species (see Materials and methods) and were able to compete for SO binding in EMSA(Fig. 4B,C, hh first, hh second). In addition, we found these sites to be located within an area that is deleted in the hh¹ (bar-3) mutant allele, a weak hh allele affecting adult flies. The corresponding deletion can be found in the first intron of the hh gene(Mohler, 1988; Lee et al., 1992). The predominant phenotype of hh¹ is a reduction of eye facets. Therefore, the deletion leading to the hh¹ allele may affect an eye-specific enhancer of hh(Renfranz and Benzer, 1989). This idea is supported by the observation of Kango-Singh et al. that in hh¹ mutant flies targeted expression of ey fails to induce ectopic eyes (Kango-Singh et al., 2003).

We chose to clone 1.4 kb out of the hh¹ deletion encompassing the two so sites and ligated this fragment to the lacZ reporter gene. Expression of the resulting hh¹-lacZ construct was found exclusively in the eye disc in cells posterior to the MF (Fig. 6B), in perfect agreement with the observation of Lee et al. that hh is expressed in differentiating photoreceptor cells(Lee et al., 1992).

Next we mutated the two SO-binding sites within the hh¹-lacZ construct by replacement of GAG by CCC(hh first, nt. 8-10 in Fig. 4C) and GAT by CCC (hh second, nt. 8-10 in Fig. 4C), resulting in the mutated construct hh¹ SOmut-lacZ(Fig. 6A,hh¹ SOmut). In four out of 10 transgenic lines, the resulting construct had lost its capability to induce lacZexpression. In six out of 10 transgenic lines, weak expression in the same pattern as the wt construct was detectable(Fig. 6C). This residual activity is probably due to a weak interaction of SO with the mutated binding sites similar to that seen in our cell culture assays(Fig. 3J). When we tested a construct in which the first SO-binding site was deleted and the second was mutated, this residual expression was lost completely(Fig. 6A,hh¹ Δ5′ and Fig. 6D).

These results show that the two SO-binding sites within the first intron of the hh gene are functional in vivo and sufficient to mediate expression, reflecting the known hh expression pattern in the eye part of late third instar eye imaginal discs. This strongly suggests that hh is directly regulated by so.

so gene activity is crucial for proper development of the entire visual system of Drosophila melanogaster, including the larval visual system (Bolwig's organ), the optic lobe, the compound eye and the ocellus. Previous work from our laboratory identified an eye-specific enhancer of so, so10, that is regulated by ey and toy(Niimi et al., 1999; Punzo et al., 2002). When used as a driver for so, so10 is only sufficient to rescue eye development of so¹ mutant flies but not ocellus development. Here we show that a fragment of 27 bp, soAE, found downstream of so10, was sufficient to rescue the entire mutant phenotype of so¹ mutant flies when combined with so10. We show that the SO protein itself bound to soAE and,in cooperation with EYA, formed an autoregulatory feedback loop that is essential for ocellus development.

As SO binds to its own enhancer and autoregulation cannot initiate expression of a gene, the initiation of so expression in the ocellar region must be triggered by other means. We propose the following model. Initiation of so expression in early third instar eye discs is mediated by ey and toy throughout the eye disc, including the ocellar precursors. Later, after this first induction, socooperatively with eya can maintain its own expression in the ocellar region by a positive autoregulatory feedback. Thus, the initiation of so expression is mediated by so10, whereas for the maintenance of so, soAE is required. This is supported by the observation of Punzo et al. that so10, which is activated by ey and toy mediates expression in early third instar larvae all over the eye disc and only later gets restricted to the compound eye part(Punzo et al., 2002).

In this model the specificity of so expression for ocellar precursor cells is provided by the expression pattern of eya; EYA protein can be found only in the ocellar region itself, where it specifically interacts with SO, and no EYA is present in the proximity of these cells. The importance of eya is further strengthened by the fact that eya⁴ mutants show an eyeless and ocelliless phenotype(Zimmerman et al., 2000). Therefore, to elucidate the mechanisms that control gene expression specifically in ocellar precursor cells, additional studies on eyaare required.

Positioned at the top of the hierarchy of the retinal determination network, ey is a potent inducer of ectopic eyes and is able to directly induce so and eya. Like ey, so and eya are able to induce ectopic eyes but only when co-expressed; so alone fails to do so.

To accomplish this induction, eya and so need to feed back on ey, obviously by binding to the eye-specific enhancer of ey. In an ectopic situation, the feedback of so/eya on ey is strong enough to induce ey for ectopic eye formation.

The function of this feedback loop in normal eye development remains to be elucidated. so and eya are both expressed posterior to the furrow and are important for neuronal development(Pignoni et al., 1997). Nevertheless, ey is tuned down posterior to the MF. The activity of the so-binding site in the ey gene might, therefore, be suppressed by other factors or by so itself during cellular differentiation posterior to the furrow. As co-expression of ey, soand eya is elevated only in a few cells in front of the MF and within the MF, a possible role for this feedback loop might be to boost eyexpression in front of and within the furrow, which leads to a strengthening of so and eya expression in just a few cell rows.

For proper eye development, a well-balanced expression level of the genes belonging to the retinal determination network is crucial. Loss-of-function mutations, as well as overexpression of the eye specification genes ey,eya, so or dac during eye development, impede proper determination of the organ and result in a reduction in eye size(Halder et al., 1998; Curtiss and Mlodzik, 2000). Therefore, we hypothesize that a feedback loop of so on eyis also important for the fine-tuning of ey expression during normal eye development. Due to its previously proposed ability to activate as well as to repress the expression of genes (Silver et al., 2003), so is a potent regulator in this context.

decapentaplegic (dpp) signalling plays an important role in the complex regulatory network of eye development. In dpp mutant eye discs, so, eya and dac are not expressed(Chen et al., 1999), whereas dpp is able to initiate ectopic expression of so and dac when expressed at the anterior margin of the eye disc(Chanut and Heberlein, 1997; Pignoni and Zipursky, 1997). Conversely, dpp expression is patchy in eye discs of eya and so loss-of-function mutants, suggesting that eya and so are required for either initiation or maintenance of dppat the posterior disc margin before MF initiation(Pignoni et al., 1997; Hazelett et al., 1998).

hh is required for dpp expression at the posterior margin before MF initiation (Borod and Heberlein,1998), and dpp expression is induced by hh in the MF (Heberlein et al.,1993), supporting the assumption that dpp is downstream of hh signalling. As dpp alone is not able to rescue posterior margin clones of hh, there have to be more eye-relevant target genes of hh signalling during third instar larval development. dpp in combination with eya can restore photoreceptor differentiation in posterior margin clones lacking smoothened(smo) expression (smo is a cell-autonomous receptor of hh signalling). This shows that dpp, in combination with eya, is able to bypass the requirement of hh during eye development (Pappu et al.,2003). Taken together, it is evident that hh is necessary for proper eya and dpp expression, both of which can induce so, and it contains two so target sites. We therefore hypothesized that the transcriptional complex consisting of EYA and SO, as with ey might also feed back on hh in order to drive the furrow during late eye development. In this model the genetic cascade starts with hh, which induces dpp and eya, moves on to so and through the SO/EYA complex feeds back to hh in order to maintain hh expression as a driving force of the MF.

The impact of these so-binding sites in the hh enhancer on eye development becomes evident from the fact that hh¹(bar-3) mutant flies have smaller eyes. The severity of the hh¹ mutant phenotype is probably diminished by an additional putative SO-binding site that resides outside the area covered by the hh¹ deletion (Fig. 6A, SO-binding motifs). If functional, this region (5′ to the hh¹ deletion) might mediate a residual hh-expression that overcomes the loss of the other sites to some extent. Another possible explanation for the rather weak hh¹ phenotype might be that the feedback of so on hh is not crucial for MF initiation but still might be of importance for the well-balanced expression of hh during MF propagation.

so belongs to the Six gene family. All Six proteins are characterized by a Six domain and a Six-type homeodomain, both of which are essential for specific DNA binding and protein-protein interaction. Based on the amino acid sequence of their homeodomain and Six domain, the Six genes were divided into three subgroups. Each of the three Drosophilahomologues can be assigned to one of these subgroups: so is mostly related to Six1/2, optix to Six3/6 and DSix4 to Six4/5 (reviewed by Kawakami and Kobayashi, 1998).

Promoter analyses of the mouse Six genes (Six1/2, Six4/5) revealed similar target sequence specificities for these mammalian counterparts of so. Six2, Six4/AREC3 and Six5 effectively bind to the same target sequence in a DNA fragment called ARE (Atpla1 regulatory element) that can be found in the Na,K-ATPase α1 subunit gene(Fig. 4C, ARE fragment)(Suzuki-Yagawa et al., 1992; Kawakami et al., 1996a; Kawakami et al., 1996b; Harris et al., 2000). Six1 and Six4 have been shown to bind to MEF3 sites in the myogenin and in the aldolase A muscle-specific (pM) promoters(Fig. 4C, MEF3 site)(Spitz et al., 1998). Recently, mammalian Six4 has been shown to bind additionally to the transcriptional regulatory element X (TreX) within the muscle creatine kinase(MCK) enhancer (Fig. 4C, Trex)(Himeda et al., 2004).

Comparison of all these sites confirmed that the three nucleotides we suggest are the most important for SO-DNA interaction are present and conserved within these motifs (nt. 1, 4 and 9 in Fig. 4C). In the case of the MEF3 site, which comprises seven nucleotides that include only two of the nucleotides important for SO-DNA interaction (nt. 4 and 9 in Fig. 4C), we looked up the original publications to check if the third conserved nucleotide is also present, and in most of the cases were able to verify its conservation(Hidaka et al., 1993; Spitz et al., 1998; Himeda et al., 2004). In fact,there is only one exception published in a study that describes two Six2 target sites (Brodbeck et al., 2004).

Nevertheless, by combining the vast majority of previous studies describing protein-DNA interaction of Six genes and our study of SO-DNA interaction, we infer that SO, Six1, Six2, Six4 and Six5have very similar DNA-binding properties. In the case of so, we propose that the consensus sequence GTAANYNGANAY(C/G) marks a good starting point for the identification of additional targets of SO, thereby helping to unravel the complex genetic interactions that orchestrate the development of the visual systems of Drosophila.

We are grateful to F. Pignoni, B. Dickson and P. Rorth for fly stocks. We would like to thank C. Brink for critical comments on the manuscript. T.P. was supported by an MD/PhD Grant from the Fondation Suisse de Recherche sur les Maladies Musculaires (FSRMM).