Direct control of neurogenesis by selector factors in the fly eye:regulation of atonal by Ey and So

Transcription factors of the basic helix-loop-helix (bHLH) type play crucial roles during both peripheral and central nervous system development in organisms as diverse as fly, worm and mouse(Lee, 1997; Brunet and Ghysen, 1999; Vetter and Brown, 2001; Ross et al., 2003; Portman and Emmons, 2000). One of the best-studied models for bHLH neurogenic function is the formation of the sensory organs of the Drosophila peripheral nervous system. Here,bHLH proteins function first as `proneural' factors by selecting proneural precursors (PNPs; cells competent, but not committed, to differentiate as neurons) and then as selector factors by establishing sensory organ precursors(SOPs; cells committed to give rise to the sensory neurons and their accessory cells) (Giangrande and Palka,1990; Ghysen and Dambly-Chaudierre, 1993).

In the fly auditory organ (Johnston's organ, JO), the stretch receptors(chordotonals, CH) and the eye, this neurogenic role is fulfilled by the bHLH protein Atonal (Ato) (Jarman et al.,1993; Jarman et al.,1994; Jarman et al.,1995). However, because ato is crucially required in all three organ types, additional factors must contribute to the specification of sensory organ type. The selector factors controlling regional specificity and the establishment of organ primordia are ideal candidates to play this role. Hence, a co-ordination or integration of the genetic networks controlling neuronal specification and organ identity would be necessary to ensure the correct spatial and temporal deployment of the two pathways and the formation of appropriate sensory structures.

In the case of the fly eye, co-expression of several selectors, including the conserved transcription factors Eyeless/Pax6 (Ey), Sine oculis (So), and Eyes absent (Eya), specifies the pool of cells competent to give rise to photoreceptor neurons and other eye cell types (the eye primordium). These and other factors are linked in a complex genetic cascade called the Ey/Pax6 pathway or Retinal Determination (RD) network. Although the onset of gene expression is initially sequential, their transcription soon becomes subject to extensive cross- and feedback regulation(Pappu and Mardon, 2004; Silver and Rebay, 2005). It is these factors that are thought to determine formation of ato-dependent photoreceptor neurons in the eye, rather than mechanoreceptors as in the JO or CH. However, at what level the input from RD factors is integrated with the genetic control of neuronal development is poorly understood and current views vary greatly.

Compelling evidence, collected over the past 12 years, documents the crucial role of the Pax6 pathway in driving eye formation. Among the most dramatic is the observation that misexpression of Ey, or several other RD factors, results in the formation of ectopic eyes(Halder et al., 1995; Bonini et al., 1997; Pignoni et al., 1997; Chen et al., 1997). These and other studies have led to a model of eye specification that places Ey and the RD network at the top, in the role of master controller of eye development(Gehring, 1996; Gehring and Ikeo, 1999). In this model, the RD network would induce expression of ato and directly control neurogenesis in the eye, as well as induce additional factors that confer the appropriate neuronal-type specificity.

More recently, an intriguing alternative, that emphasizes the use of ato in the formation of diverse sensory organs, has been proposed(Niwa et al., 2004). In this model, ato transcription depends on a `Pax6-independent' enhancer that is utilized in all ato-related sensory organs, reflecting the regulation of ato in an ancestral protosensory organ. The modification of this ato-driven neurogenic program by a Pax6-based head-specification pathway would then result in the formation of the eye and its photoreceptor neurons. Thus, Ey/Pax6 would function parallel to or downstream of Ato expression and sensory organ formation, rather than being an upstream master control gene (Niwa et al.,2004).

These two models propose different views of how organ specification and neuronal development are integrated during eye formation. However, our current understanding of ato regulation is still very rudimentary and a detailed dissection of the ato control region is needed to better address this issue.

Expression of the atonal gene during eye morphogenesis is very dynamic (Fig. 1A). Gene expression is controlled at the transcriptional level and begins in rows of eye progenitor cells that form a dorsoventral stripe across the eye epithelium(or eye disc) (blue stripe in Fig. 1A). Soon after, the ato-positive progenitors begin to regulate ato transcription differentially and groups of high-ato-expressing cells (called initial clusters) become visible within the stripe. Next, cells located in between the clusters lose ato expression completely leading to the appearance of separate groups of ato-expressing cells called intermediate clusters. Lastly,the intermediate clusters resolve into single ato-positive cells which become R8 neurons, the founder photoreceptor neuron of each single eye or ommatidium (Jarman et al.,1994; Jarman et al.,1995).

This dynamic pattern of ato expression is achieved through two separate control regions. Genomic DNA flanking the gene on the 3′ side(3′ato) controls the early phase of atoexpression, i.e. the activation of gene expression in eye progenitor cells as well as formation of the initial clusters(Fig. 1A,B)(Sun et al., 1998). Continued expression, first in intermediate clusters and then in the R8 SOPs, depends instead on regulatory elements that lie 5′ of the atotranscription unit and require Ato function(Fig. 1A,B)(Sun et al., 1998). Interestingly, the 3′ato regulatory region also promotes expression in the JO and the femoral CH(Fig. 1C,C′). It has been proposed that expression in all three sensory organs is driven by a common enhancer (Niwa et al., 2004). However, this hypothesis has not been directly tested.

To better understand the relationship between eye specification and ato-driven neuronal development, we chose to investigate the transcriptional activation of ato in eye progenitor cells. Through a detailed analysis of the 3′ato control region, we find that the regulatory elements driving expression in the eye are distinct from those driving expression in JO and CH. Moreover, the RD factors Ey and So directly bind adjacent cis-regulatory sites present within a minimal 3′ato enhancer region. Ey and So bind in vitro and the insertion of a few base pairs between binding sites disrupts reporter gene expression in vivo. Thus, Ey-So protein-protein interactions may play a significant role in the regulation of ato transcription. This study provides insights into the molecular mechanisms that mediate the integration of eye specification and neurogenic pathways, and the developmental transition from organ specification to differentiation during eye morphogenesis.

Fragments of the 3′ato region were generated by PCR. Mutations, deletions and insertions were introduced using the PCR-Based Site-Directed Mutagenesis Kit (Stratagene). Sequences of all primers and details of cloning are available upon request. All constructs were confirmed by sequencing. DNA sequences were aligned using ClustalW. P-element transformation was performed using vectors pCasper-β-gal and pStinger(Rubin and Spradling, 1982; Barolo et al., 2000). Other fly lines used: so¹; eya²; UAS-ey; UAS-so; UAS-eya(Cheyette et al., 1994; Pignoni et al., 1997); and P{GAL4-dpp.blk1}40C.6 (FBti0002123).

Standard antibody (Ab), β-gal and in situ protocols were used(Sullivan et al., 2000). Ab used: mAb anti-Elav (1:1000) (Robinow and White, 1991), mAb anti-Eya (1:100)(Bonini et al., 1993), rabbit anti-β-gal (1:1000) (Cappell), guinea pig anti-Sens (1:1000)(Nolo et al., 2000), anti-GFP(1:1000; Upstate Biotech). Anti-mouse, anti-rabbit and anti-guinea pig Cy2-,Cy3- or Cy5-conjugated (Jackson Immuno Research Laboratories) or hrp-conjugated (Bio-Rad) secondary Ab were used at 1:200 dilution. Confocal microscopy was performed using a Leica TCS microscope. DIG-labeled (Roche) in situ probes were generated from pCRII-TOPO-lacZ.

Ey and Toy were produced using a reticulocyte lysate in vitro transcription-translation system (Promega). Either the full-length So protein produced in S2 cells, or a GSTSo^SixHD-fusion protein purified from Escherichia coli (Six and Homeobox domains)(Kenyon et al., 2005) were used. DNA probes were generated by PCR or by annealing of synthesized oligonucleotides, and were labeled using polynucleotide kinase (NEB) and[γ-³²P]ATP (Amersham). The binding reaction was carried out in 15 μl 10 mM HEPES pH 7.9, 35 mM KCl, 10 mM NaCl, 4 mM MgCl₂,20% glycerol, 1 mM DTT, 1 mM EDTA, 0.06 mM PMSF, 0.1% BSA, 10,000 cpm probe, 1μg poly(dI-dC) (Pharmacia) and 5 out of 50 μl of the in vitro transcription-translation reaction. After incubating 20 minutes at room temperature, samples were analyzed in a 5% non-denaturing polyacrylamide gel followed by autoradiography. For competition experiments, a 40 molar (Ey, Toy)or 100 molar (GSTSo^SixHD) excess of unlabeled DNA was added and incubated for 5 minutes at room temperature before adding the radioactive probe.

GST and GST-So^SIXHD proteins were induced in BL21 cells, and purified with glutathione-agarose beads (Pierce). Full-length[³⁵S]Ey, [³⁵S]luciferase (negative control) or[³⁵S]LexA-Eya^- domain-fusion (positive control) proteins were produced by in vitro transcription-translation in a total volume of 50μl (Promega). For each pull-down experiment, 15 μl of in vitro translated [³⁵S]protein was diluted in 150 μl PBT (PBS, 1%Triton X-100), mixed with either GST or GST-So^SIXHD immobilized on glutathione-agarose beads, and incubated at 4°C for 2 hours. Beads were washed four times with PBT and boiled in 20 μl 2′SDS loading buffer before electrophoresis (10% SDS-PAGE) and autoradiography.

To begin with, we sought to establish whether the activation of 3′ato-dependent transcription in eye, JO (antenna disc) and femoral CH (leg disc) falls under a common cis-regulatory module(CRM), or whether the 3′ region contains separate, disc-specific enhancers. As a first step, we reproduced the findings of Sun and colleagues(Sun et al., 1998) by generating a 3′ato^FL-βgalconstruct that contains the entire 3′ato fragment(Fig. 1B) as well as the endogenous ato promoter region (a 1039 bp fragment that includes 334 bp of the 5′-UTR). As previously described(Sun et al., 1998; Niwa et al., 2004), expression was observed in the eye disc (Fig. 1C), the antenna (Fig. 1C inset) and the leg discs(Fig. 1C′).

Next, we generated three different reporters containing fragments of 3′ato^FL: a 2.9 kb left fragment(3′ato^L- βgal), a 2.1 kb middle fragment (3′ato^M'-β gal) and a 1.8 kb right fragment(3′ato^R- βgal)(Fig. 1B). Multiple independent insertions were recovered and analyzed for each construct (see Fig. S1 in the supplementary material). We found that the 3′ato^M'- βgal could reproduce all aspects of early ato expression within the eye disc(Fig. 1E) and a closely related 3′ato^M” fragment was able to drive transcription in essentially the same pattern(3′ato^M”- βgal and 3′ato^M”-GFP constructs)(Fig. 1B, Fig. 2B and see Fig. S1 in the supplementary material). The M' and M” reporters also showed expression in the ocelli (photoreceptor-containing sensory structures derived from the dorsal region of the eye disc) (Fig. 1E and see Fig. S1 in the supplementary material). No expression was detected in the leg or antennal discs(Fig. 1E,E′).

The expression associated with the JO and CH in antenna and leg discs was observed, instead, in the 3′ato^L-β gal lines (Fig. 1D,D′). The level of expression in the antenna was consistently higher in these lines when compared with 3′ato^FL-βgal. The retina-related stripe and the ocellar expression domain were absent(Fig. 1D). Lastly, lines carrying the 3′ato^R- βgal did not express the reporter in eye, antenna or leg discs(Fig. 2F,F′).

We conclude, therefore, that the 3′ flanking genomic region,similarly to the 5′ regulatory region(Sun et al., 1998), contains multiple CRMs that control ato expression in different sensory organs.

To more precisely identify the region responsible for expression in the eye disc, we further dissected the 3′ato^M” DNA. This analysis resulted in the identification of a 348 bp fragment that is necessary and sufficient to promote reporter gene expression in eye progenitor cells. As summarized in Fig. 2A (and see Fig. S1 in the supplementary material), all reporter constructs that contained this 348 bp region showed expression in a stripe. These include the constructs 3′ato^M”-βgal,3′ato^1.2- βgal and 3′ato³⁴⁸-βgal(Fig. 2B,D,E), 3′ato^FL-βgal and 3′ato^M'- βgal(Fig. 1C,E), 3′ato^M'R- βgal and 3′ato⁴⁸⁸- βgal (see Fig. S1 in the supplementary material). On the other hand, all constructs that lacked this region did not activate the reporter in eye progenitors, including 3′ato^M”-Δ348-βgal(Fig. 2C), 3′ato^L- βgal,3′ato^R- βgal(Fig. 1D,F), and 3′ato^M”-Δ488-βgal(see Fig. S1 in the supplementary material).

Interestingly, the 348 bp region contains two relatively large (A1=99 bp and A2=140 bp) DNA sequences that are highly conserved from D.melanogaster to D. virilis (Fig. 2A, and see Fig. S2 in the supplementary material). Based on this observation, we also generated constructs containing only A1 or A2(3′ato^A1- βgal and 3′ato^A2- βgal)(Fig. 2A). However, we found that neither region alone was sufficient to drive the stripe of reporter gene expression in the eye disc (see Fig. S3 in the supplementary material).

Based on these results, we conclude that the 348 bp region constitutes a`core' or `minimal' enhancer region for the transcriptional activation of ato in eye progenitor cells.

Subtle but significant differences in expression were apparent between reporter lines. Constructs containing the 1.2 kb DNA region closely mimicked early ato expression, including low level expression in a stripe and stronger expression in the initial clusters(3′ato^FL- βgal,3′ato^M'R- βgal,3′ato^M'- βgal,3′ato^M”- βgal,3′ato^1.2- βgal)(Fig. 1C,E, Fig. 2B,D,D′, and see Fig. S1 in the supplementary material). On the other hand, reporters containing the core enhancer region but within fragments smaller than 1.2 kb(3′ato⁴⁸⁸- βgal and 3′ato³⁴⁸- βgal) reproduced some, but not all, aspects of this pattern. First, and perhaps not surprisingly, the latter constructs consistently displayed a lower level of expression than the larger ones (see Fig. S1 in the supplementary material). Second, the differential regulation of ato expression that leads to formation of the initial clusters was not observed in 3′ato⁴⁸⁸- βgal or 3′ato³⁴⁸- βgal. In these lines, mRNA levels remained relatively uniform across the stripe(Fig. 2E,E′, and data not shown). Third, extra rows of anterior progenitor cells expressed ato,thereby generating a broader stripe. This can be clearly seen when comparing the position of the anterior border of the reporter expression domains relative to the domains of two molecular markers: Eya, a protein expressed in all progenitors anterior to the MF, and Senseless (Sens), a protein first expressed just posterior to the early ato domain (i.e. at the level of intermediate ato clusters)(Fig. 1A, Fig. 2D′). In 3′ato³⁴⁸- βgal and 3′ato⁴⁸⁸- βgal, the anterior margin of the β-gal domain was shifted forward, closer to the anterior border of the Eya domain and farther away from Sens-expressing cells(Fig. 2D′,E′, and data not shown).

Phylogenetic comparison of the 1.2 kb DNA among different Drosophila species identified two additional conserved sequences that lie outside the 488 bp region (IC1=88 bp and IC2=133 bp)(Fig. 2A, and see Fig. S2 in the supplementary material). To investigate whether these sequences contained cis-regulatory sites controlling expression level, induction in progenitor cells and/or formation of the initial clusters, we tested the effect of removing a 298 bp region that spans these sequences from the 1.2 kb fragment. Reporter gene expression driven by a 3′ato^1.2-Δ298-βgalconstruct was similar to expression of the 3′ato³⁴⁸-βgal transgenes in that the level of expression was lower and the initial clusters did not form(Fig. 2, compare E′,F′,D′). However, the reporter did not appear to be as precociously activated as 3′ato³⁴⁸- βgal(Fig. 2, compare E′,F′,D′).

These results showed that in addition to the control of gene activation(which resides within the 348 bp region), at least two more enhancers are present within the 1.2 kb fragment, including a clusters formation/expression level regulatory region and an anterior repression module.

To further explore the relationship between the RD network and ato, we focused on the core region responsible for activation. Genetic evidence strongly suggests that the RD factors are required, directly or indirectly, for transcriptional activation of ato. Several strong mutant alleles of the RD genes display `eyeless' phenotypes in the adult and lack expression of ato in the L3 eye disc(Bonini et al., 1993; Cheyette et al., 1994; Jarman et al., 1995). Hence,we chose to investigate how the 3′ato³⁴⁸-β gal construct responded to loss or gain of RD network function and assayed reporter gene expression in: (1) discs lacking the activity of the RD factors So or Eya; (2) discs in which Ey, or So and Eya, are ectopically expressed and induce ectopic eyes; and (3) eya or so mutant discs in which Ey is ectopically expressed but can not induce ectopic eyes.

In order to severely reduce RD network function, we opted to use strong mutant alleles of the eya and so genes: eya², a 100% penetrant eye-specific null allele, and so¹, an eye-specific allele that displays the null eyeless phenotype with ∼95% penetrance. Expression of the 3′ato³⁴⁸- βgal reporter was absent in all eya²/eya² discs and in nearly all so¹/so¹ discs(Fig. 3A-C).

Conversely, in gain of function experiments, we observed ectopic induction of the reporter in the wild-type but not in eya or so mutant backgrounds. To drive Ey or So+Eya expression in tissue other than the eye, we employed the Gal4/UAS system (Brand and Perrimon, 1993). UAS-ey or UAS-So+UAS-eyatransgenes were ectopically induced under the control of the dpp-Gal4^blk driver and reporter gene expression was assayed within the antenna, leg and wing epithelia. When Ey or So+Eya were expressed in a wild-type background, 3′ato³⁴⁸- βgal expression was detected ectopically in all three discs(Fig. 3D-F, and data not shown). However, Gal4-driven expression of Ey in so¹ or eya² mutant discs did not result in activation of the reporter (Fig. 3G, and data not shown).

We conclude that the core eye enhancer is regulated as would be expected for the endogenous ato gene by the RD network.

As mentioned above, deletion of the 348 bp region(3′ato^M'-Δ348-βgal)resulted in loss of reporter gene expression in eye progenitor cells(Fig. 2C). In addition, removal of the A1 conserved region(3′ato^M”-ΔA1-β gal) was sufficient to prevent reporter gene expression (see Fig. S1 in the supplementary material). Interestingly, within the A1 region,we identified sequences matching the consensus binding sites for the transcription factors Pax6/Ey (paired domain) and So(Fig. 4A and see Fig. S2 in the supplementary material) (Niimi et al.,1999; Ostrin et al.,2006; Pauli et al.,2005). The presence of potential binding sites raised the possibility that Ey and/or So directly regulate ato expression.

To test this hypothesis, we carried out electrophorectic mobility shift assays on DNA fragments spanning a ∼500 bp region around these sites(Fig. 4B). The fragment containing the two sites, probe III, could be shifted by either Ey, So or GST-So^SixHD (see Materials and methods)(Fig. 4, lanes 1-7 in C,D, and data not shown). In all cases, competitive binding assays showed that protein binding was specific. Binding was competed by unlabelled probe III DNA or by oligomers containing consensus binding sites, but not by probe III DNA in which either binding site had been mutated(Fig. 4, lanes 8-11 in C and lanes 8-12 in D).

As might be expected, Toy, another Pax6-type factor that functions in eye development, also binds the Pax6 consensus site specifically (see Fig. S4 in the supplementary material). However, it is unlikely that Toy plays a crucial role in directly regulating ato expression in vivo because the 3′ato³⁴⁸- βgal reporter could be induced in the absence of Toy protein. In fact, misexpression of Ey induces ectopic eyes and the 3′ato³⁴⁸-β gal reporter (Fig. 3D-F), but not expression of toy(Czerny et al., 1999). By contrast, ectopic Ey activates transcription of the so gene and So+Eya induces ey (Pignoni et al., 1997; Halder et al.,1998). Hence, Ey and So would both be present to activate the 3′ato³⁴⁸- βgal reporter.

To test whether the Ey- and So-binding sites were required in vivo, we generated reporter constructs containing either a mutated Pax6(3′ato^348Pax6MUT- βgal) or a mutated So (3′ato^348SoMUT-β gal) site, as well as a 1.2 kb construct in which both sites were altered (3′ato^{1.2Pax6-SoMUT}-β gal). The base pair changes introduced had been reported to reduce binding affinity to consensus sites(Ostrin et al., 2006; Pauli et al., 2005; Punzo et al., 2002; Niimi et al., 1999). In all cases, reporter gene expression was severely affected, from strongly reduced to completely absent (Fig. 5 A-D).

In summary, Ey and So can specifically bind sites within the core eye element in vitro and these sites are required for reporter gene expression in vivo.

Strikingly, the So and Pax6 sites are located adjacent to one another(Fig. 4A), being separated by exactly 3 bp in all sequenced Drosophila species (see Fig. S2 in the supplementary material and http://insects.eugenes.org/species/). This arrangement raised the possibility that bound So and Ey may interact with each other and that such interactions are required in vivo. In order to explore these hypotheses, we tested for direct binding in vitro and investigated the significance of the spacing between their cis binding sites in vivo.

For the in vitro assay, Gst-pull-down assays were carried out. In these experiments, a GST-So^SixHD fusion protein containing the evolutionarily conserved domains showed a clear interaction with full-length Ey protein (Fig. 5E). As negative controls, GST alone did not bind Ey, and GST-So^SixHD did not bind luciferase (Fig. 5E). Thus, Ey and So can directly interact at the protein level.

To explore the in vivo significance of this interaction, we reasoned that the conserved 3 bp spacing between sites is likely to be crucial for such interaction. Due to the helical structure of the DNA, differences in spacing between binding sites affect the spatial arrangement of bound factors not only by altering the distance between the proteins, but, perhaps more importantly,by altering the orientation of the bound factors relative to one another. Since ∼11 bp span a complete (360°) rotation of the helix, the insertion of only a few bp in between sites is all that is needed to achieve misorientation. Thus, we generated two constructs: one with a 3 bp insertion between sites (3′ato^348+3A-β gal) and one with a 6 bp insertion(3′ato^348+6A- βgal), thereby introducing a rotation between sites of ∼90° and ∼180°,respectively. We predicted that, if interactions between So and Ey were not crucial, both insertions (3 and 6 bp) would be likely to have little or no effect on reporter gene expression. On the other hand, if Ey-So interactions were essential, the 6 bp insertion, and possibly the 3 bp insertion as well,would disrupt transcriptional regulation. We found that the latter was indeed the case. We detected a modest but consistent drop in expression levels in the 3′ato^348+3A- βgal transgenic lines (Fig. 5F). Moreover, a severe reduction in reporter gene expression was observed with the 3′ato^348+6A- βgal construct(Fig. 5G).

These results show that Ey and So interact directly in vitro. The observation that small insertions between their binding sites affect reporter gene expression is consistent with transcriptional regulation by an Ey-So complex in vivo.

The regulatory elements controlling the early phase of atoexpression in the eye lie within a 1.2 kb region located 3.1 kb downstream of the ato transcription unit. The early phase of atotranscription results from the integration of multiple regulatory inputs through separate cis-regulatory modules present within the 1.2 kb region(Fig. 6A).

Cis-regulatory elements essential for gene activation map to the last 348 bp of the 1.2 kb region and include the So- and Ey-binding sites. Other factors undoubtedly bind to sequences within A1-A2 and regulate gene expression as neither A1 nor A2 alone are sufficient to drive expression in the eye disc. Genetic evidence suggests that signaling by the Bmp4-type factor decapentaplegic (Dpp) also contributes to ato activation and two putative binding sites for Mad (a transcription factor shown to activate Dpp pathway targets) appear to be required for ato expression in all discs (Niwa et al., 2004). However, a Mad consensus site present in the A2 box does not correspond to either one of the two elements identified by Niwa and colleagues. Moreover,both the previously identified sites lie within the L fragment well upstream of the M'-M” interval containing the eye-disc enhancers. A potential source for this discrepancy lies in the use of different promoters. Whereas Niwa and colleagues used the heterologous hsp43 promoter, we relied on the endogenous ato promoter region (∼700 bp of 5′genomic DNA). Future analyses of 3′ enhancer-promoter interactions may resolve this issue.

Separate cis-regulatory elements located within the conserved DNA regions IC1 and IC2 control initial clusters formation. This feature of atoexpression has been shown to require Notch (N) function. Sequence analysis of the IC1-IC2 region does identify a binding site for the effector of N signaling Suppressor of Hairless [Su(H)]. However, contradictory reports have been published on how Notch controls ato expression. Sun and colleagues (Sun et al., 1998)found that transcription of ato is uniformly upregulated upon inactivation of Notch in N^ts1 mutant discs. By contrast,early Ato protein expression is severely reduced in null Notch mutant clones(Baonza and Freeman, 2001; Li and Baker, 2001). Since these experiments made use of different genetic reagents, it is difficult to interpret these results. Notch signaling may independently regulate ato expression at the mRNA and protein levels. Alternatively, the source of the discrepancy may lie in the use of different alleles, one hypomorphic (N^ts1), and the other null(N^54l9).

Lastly, activation of the 3′ato³⁴⁸-βgal reporter (core element) occurs prematurely as compared with endogenous ato. The 3′ato³⁴⁸-βgal mRNA is also found in cells lying just anterior to the proneural domain. Eye progenitors from this region are at a developmental stage referred to as pre-proneural and are characterized by the expression of the transcription factor Hairy (H) in addition to RD proteins (Greenwood and Struhl, 1999). In the absence of Hairy and its partner Extra Macrochaetae, neurogenesis begins precociously within the eye disc(Brown et al., 1995). Thus,Hairy contributes to the downregulation of ato expression and prevents precocious neurogenesis. Activation of the reporters 3′ato³⁴⁸-βgal and 3′ato⁴⁸⁸-βgal (but not 3′ato^1.2-βgal or 3′ato^1.2-Δ298-βgal) in pre-proneural cells suggests that cis-elements mediating anterior repression lie within the 1.2 kb DNA fragment but outside the IC and A boxes. Although a search for canonical Hairy-binding sites does not identify potential regulatory elements, additional short stretches of evolutionarily conserved DNA are present and may contribute to this and/or other aspects of ato regulation.

Over the last few years, Ey and So have been shown to play a crucial role in the deployment and maintenance of the RD network by directly regulating the transcription of several eye-specification genes [ey, so, eya,dachshund (dac) and optix](Niimi et al., 1999; Punzo et al., 2002; Ostrin et al., 2006; Pappu et al., 2005; Pauli et al., 2005). However,little is known about downstream targets of the RD cascade. Although So also activates the post-MF expression of hedgehog and lozenge(Yan et al., 2003; Pauli et al., 2005), this gene regulation is likely to reflect the late, differentiation-related functions of So (Pignoni et al., 1997). Thus, it is unclear how the RD factors induce eye formation and what aspects of the morphogenetic program they control directly.

Our results strongly suggest that the transcription factors Ey and So control activation of ato expression. This is the first example of a gene required during eye morphogenesis that is directly regulated by the RD network. The direct control of ato by Ey and So is a likely reason why ectopic eye induction by Eya+So or Dac depends on the activation of their upstream regulator ey (Pignoni et al., 1997; Chen et al.,1997). Other downstream targets may also be similarly controlled by multiple RD factors.

The in vitro and in vivo evidence presented here also suggest that Ey and So may form a complex when bound to the adjacent cis-regulatory sites in the 3′ato core element(Fig. 6A). Together with the previously reported interactions of Eya-So and Eya-Dac(Pignoni et al., 1997; Chen et al., 1997), this finding raises the possibility that additional multimeric complexes involving several RD factors may also be involved in driving the transcriptional program for eye development. As originally proposed by Curtiss and Mlodzik(Curtiss and Mlodzik, 2000),the observation that normal eye development is severely disrupted when one or another RD factor is over-expressed suggests that the RD proteins must be present at an appropriate level relative to one another. As all four proteins,Ey, Eya, So and Dac, have now been shown to interact in various combinations,the formation of such complexes and the recruitment of additional shared co-factors are likely to be sensitive to the relative concentration of RD factors present in eye progenitor cells.

The model of gene regulation exemplified by the control of atotranscription provides a strong rationale for the feedback regulatory loops that link late and early RD gene expression. This regulation is likely to play a crucial role in ensuring the presence of appropriate levels of all four RD factors to optimize complex formation and co-regulation of downstream targets.

As summarized in the introduction, current models for the co-ordination of organ identity and neurogenesis in the eye place the Pax6 pathway either upstream of, or in parallel to, the control of neurogenesis. The findings presented in this paper favor the former model. We have identified separate regions for the regulation of ato transcription in the eye versus other sensory organs (JO and CH). In addition, the presence of Ey- and So-binding sites that are required in vivo for reporter gene activation strongly suggests that endogenous ato expression is directly regulated by these factors. Thus, the RD network does not merely modify sensory organ development within the eye disc, but does, in fact, directly control it (Fig. 6B). In doing so, it also contributes to the co-ordination of selector and neurogenic inputs required to generate complex sensory structures such as the eye.

Is this regulatory relationship between Ey-So and ato ancestrally derived? That is, was the direct link between ancestral Pax- and Ath-like genes already established in the protosensory organ that gave rise to today's ato-dependent sensory structures? The association of Pax-, Six- and Ath-type factors with sensory perception is not restricted to photic sensation but extends to mechanoreception in diverse organisms including mouse,jellyfish and mollusks (Treisman,2004; Piatigorsky and Kozmik,2004; Fritzsch and Piatigorsky, 2005; Tessmar-Raible et al., 2005). In the jellyfish P. carnea, which lacks eyes but responds to a variety of environmental stimuli including light, expression of a putatively ancestral-like PaxB gene, Six1/2, Six3/6 and atonal-like 1 is associated with neuronal precursors found in the medusa tentacles(Groger et al., 2000; Kozmik et al., 2003; Stierwald et al., 2004; Seipel et al., 2004; Tessmar-Raible et al., 2005). Although the studies carried out in more basal metazoa consist mostly of analyses of gene expression and not function, this evidence does suggest that the association of Pax/Six/Ath-type factors and sensory organ development is ancient and may have been retained over more than 600 million years of evolutionary history.

It is possible that the mechanisms of transcriptional regulation uncovered between Pax and Six genes(Niimi et al., 1999; Pauli et al., 2005; Ostrin et al., 2006) and between Pax/Six and ato (this work) may have arisen early during evolution. Such regulatory interactions may have favored the continued association of Pax/Six/Ath as various modifications of their genetic cascades led to the development of more complex and diverse sensory organs. The investigation of ato/Ath gene regulation in other sensory organs and in basal metazoans is likely to clarify the evolutionary relationship among these pathways and the sensory modalities they control.

We thank Drs J. Malicki and M. Applebury for comments on the manuscript,the Bloomington Stock Center for fly lines, the DSHB for mAbs, Drs Y. Jan, J. Posakony and H. Bellen for fly lines, Abs and DNA clones. Confocal imaging and DNA sequencing were performed at the MEEI Ophthalmology Core Facility(P30EY14104). This work was supported by NEI grant R01EY13167 (to F.P.) and the Massachusetts Lions Eye Research Fund.