Deciphering synergistic and redundant roles of Hedgehog, Decapentaplegic and Delta that drive the wave of differentiation in Drosophila eye development

The adult eye of Drosophila is remarkable for a wave of differentiation that spreads from posterior to anterior across the retina field of the eye imaginal disc. Eight hundred or so ommatidia begin differentiation in about 30 columns. Adjacent columns begin differentiation approximately 90 minutes apart, so that it takes about two days before the entire retinal field is differentiating. The advancing anterior boundary of the differentiating region is morphologically recognizable from an indentation known as the morphogenetic furrow, and is associated with changes in cell cycle, adhesion and gene expression (Wolff and Ready, 1993; Heberlein and Moses, 1995; Lee and Treisman,2002). A potentially similar wave of differentiation also occurs in zebrafish retina (Neumann and Nusslein-Volhard, 2000; Stenkamp et al., 2000).

Progression of the morphogenetic furrow depends primarily on Hedgehog,which is secreted by differentiating photoreceptor neurons. Hedgehog is necessary for furrow progression (Ma et al., 1993). Ectopic Hh expression is sufficient to initiate ectopic furrows in the anterior undifferentiated region(Heberlein et al., 1995). Despite the primary role of Hh, cells unable to receive Hh signals are still able to differentiate because Hh triggers the secretion of secondary signals,and cells unable to respond to Hh still respond to these other signals(Strutt and Mlodzik, 1997). Dpp is one important secondary signal, and acts redundantly with Hh. Cells must be able to respond to either Hh or Dpp to begin differentiating(Heberlein et al., 1993; Greenwood and Struhl, 1999; Curtiss and Mlodzik, 2000). It has been proposed that Notch signaling also contributes(Baker and Yu, 1997; Li and Baker, 2001). Ectopic Notch and Dpp together can initiate differentiation as effectively as ectopic Hh does (Baonza and Freeman,2001).

In this paper we sought to define the individual roles and interactions of each signal through the study of loss-of-function mutations affecting response to Hh, Dpp or N signals, both alone and in combination. We also sought to determine the basis of the redundancy between Hh and Dpp, and how the two signals replace each other, and we investigated whether there is redundancy between Hh and N, and for what events, if any, each signal is individually sufficient in the absence of the others.

Although redundant functions can be studied through ectopic expression experiments, a loss-of-function approach that removes components from the redundant pathways has the advantage of addressing gene function at the normal time and place, and at normal expression levels. As each of Hh, Dpp and N signals are important many times during Drosophila development from embryogenesis onwards, it was necessary to use a conditional genetic approach. We employed mosaic analysis using the FLP/FRT system to obtain clones of retinal cells lacking function of Hh, Dpp and N pathway components. Each of these genes also plays roles in the initiation of the morphogenetic furrow at the posterior margin of the eye field, so that furrow initiation fails when eye margin cells are mutated, leaving the region to the anterior undifferentiated (Lee and Treisman,2002). The present study is therefore restricted to the progressive onset of differentiation by cells within the retinal field.

Initially, we found that removing Smo did not prevent accumulation of the Ci activator Ci155. However, we found that differentiation of Smo-mutant cells depended not on Ci, but on Dpp and N signaling. Perhaps because Dpp and Hh would otherwise act over different ranges, the pace of furrow progression is constrained by inhibitors, such as Hairy, which are themselves regulated by Dpp, Hh and N.

Clones of cells mutant for relevant genes were obtained by FLP-mediated mitotic recombination (Golic,1991; Xu and Rubin,1993). Homozygous mutant cells were identified through lack of Ci155 antibody staining, or the absence of the transgene-encoded markers armβ-gal or hsGFP (Vincent et al.,1994; Motzny and Holmgren,1995; Methot and Basler,1999).

Clones were obtained in: eyF; smo³FRT40/[armlacZ] FRT40, hsF; smo³FRT40/[armlacZ] FRT40, hsF; smo^D16FRT40/[armlacZ] FRT40, and y hsF; smo¹ FRT42 en/smo³ FRT42 [smo⁺ hs:Gfp][ci⁺]; ci⁹⁴/Dp(1;4)y⁺spa larvae with equivalent results. smo³ and smo^D16 are both null alleles (Chen and Struhl,1998).

Clones were obtained in: hsF; smo³tkv^strII FRT40/M [armlacZ] FRT40 larvae. tkv^strII is a null allele(Greenwood and Struhl,1999).

Clones were obtained in: hsF; tkv^a12 FRT40/M[armlacZ] FRT40 larvae as described(Burke and Basler, 1996). tkv^a12 is a null allele(Penton et al., 1994).

Clones were obtained in: hsF; smo³Mad^1-2 FRT40/[armlacZ] FRT40 larvae as described (Curtiss and Mlodzik,2000). Mad^1-2 is an insertion in Mad regulatory sequences that prevents most Dpp signaling with little effect on growth (Wiersdorff et al.,1996).

Clones were obtained in: y hsF; FRT42[ci⁺]/FRT42; ci⁹⁴ larvae as described (Methot and Basler,1999). ci⁹⁴ is a null allele from which the promoter region and first exon have been deleted, including start sites for transcription and translation(Slusarski et al., 1995; Methot and Basler, 1999).

Clones were obtained in: y hsF; smo¹FRT42 en /smo³ FRT42[smo⁺hs:Gfp] [ci⁺]; ci⁹⁴ larvae as described(Methot and Basler, 1999). Control smo clones were obtained in parallel from phenotypically y⁺ larvae: y hsF; smo¹ FRT42 en/smo³ FRT42 [smo⁺hs:Gfp] [ci⁺]; ci⁹⁴/Dp(1;4)y⁺spa.

Clones were obtained in: y hsF; Mad^1-2 FRT40/[ci⁺] FRT40; ci⁹⁴ larvae. This [ci⁺]transgene was provided by R. Holmgren.

Clones were obtained in: y hsF; tkv^a12 FRT40/[ci⁺] FRT40; ci⁹⁴, and y hsF; tkv^a12 FRT40/M21 [ci⁺] FRT40; ci⁹⁴ larvae.

Clones were obtained in: y hsF; Su(H)^Δ47 FRT40 [w⁺l(2)35Bg⁺]/[ci⁺] FRT40; ci⁹⁴ larvae. Su(H)^Δ47 is a 1.9 kb deletion that removes the Su(H) l(2)35Bg intergenic region, including the transcriptional start site and ATG codon of both genes, and is a null allele(Morel and Schweisguth,2000).

Clones were obtained in: y hsF; Su(H)^Δ47 FRT40 [w⁺l(2)35Bg⁺]/[armlacZ] FRT40 larvae.

Clones were obtained in: y hsF; Mad^1-2Su(H)^Δ47 FRT40 [w⁺l(2)35Bg⁺]/[ci⁺] FRT40; ci⁹⁴ larvae.

Fng:Gal4 (gift of M. Mlodzik) was used to misexpress ci derivatives from UAS transgenes. We also used the Gal4 line hairy^H10 to misexpress UAS:ci^Cellanterior to the morphogenetic furrow(Ellis et al., 1994).

Eye discs were labelled for Atonal as described(Lee et al., 1996). Other labelling was performed either in 0.1 M sodium phosphate (pH 7.2), 1% normal goat serum, 0.1% saponin, following a 45 minute fixation in cold PLP(Tomlinson and Ready, 1987),or in 0.1 M sodium phosphate (pH 7.2), 0.3% sodium deoxycholate, 0.3% Triton X-100, after fixation at room temperature in 3.7% formaldehyde, 100 mM PIPES(pH 6.95), 1 mM EGTA, 2 mM MgSO₄. Preparations were examined using BioRad MRC600 and Radiance 2000 confocal microscopes, and digital images were manipulated using Adobe PhotoShop 4.0 and NIH Image 1.62 software. Antibodies against β-galactosidase were obtained from Cappel and from the Developmental Studies Hybridoma Bank (mAb40-1), polyclonal rabbit anti-GFP was obtained from Molecular Probes, and other antibodies were obtained from their developers: anti-ato (Jarman et al.,1994); anti-Ci155 (mAb2A1)(Motzny and Holmgren, 1995);anti-senseless (Nolo et al.,2000); and anti-hairy antibodies(Brown et al., 1995). Preparations were obtained over several years. Consequently, variations in procedures, antibody batches, and confocal hardware and software make comparing signal intensity between preparations unreliable, except where specifically noted in the text.

The crucial step initiating retinal differentiation is specification of a founder R8 photoreceptor cell within each ommatidium. Once specified, R8 cells initiate the recruitment of other retinal cells in response to the receptor tyrosine kinase Egf receptor and Sevenless, which are activated by signals emanating from the R8 cells (Tomlinson and Ready, 1987; Freeman,1997; Nagaraj et al.,2002). R8 cells are specified by the proneural gene atonal (ato) (Jarman et al., 1994). Cell specification after R8 does not depend directly on Atonal, Hh or Dpp function, and can occur in cells genetically null for atonal, ahead of the morphogenetic furrow, or in cells unable to respond to Hh and Dpp, so long as R8 cells are present nearby(Jarman et al., 1994; Dominguez et al., 1998; Curtiss and Mlodzik,2000).

R8 specification in the morphogenetic furrow is illustrated in Fig. 1. senseless and other target genes are expressed in response to rising Atonal activity(Baker, 2002). Atonal expression is transient, whereas Senseless is maintained in differentiating R8 cells throughout the eye disc (Nolo et al., 2000) (Fig. 1). hairy encodes a negative regulator of Atonal function, and is expressed ahead of the morphogenetic furrow and downregulated as differentiation begins (Fig. 1) (Brown et al.,1995). Notch activation downregulates Hairy(Baonza and Freeman, 2001). These events coincide with peak Hh signal at the anterior of the morphogenetic furrow, as revealed by accumulation of the Ci155 protein(Motzny and Holmgren, 1995; Strutt and Mlodzik, 1996)(Fig. 1).

Ectopic expression has identified roles of Dpp and N downstream and parallel to Hh (Pignoni and Zipursky,1997; Baonza and Freeman,2001). Mutations of Dpp receptors or N pathway genes do not affect the progression of differentiation (Burke and Basler, 1996; Baker and Yu,1997). To ascertain the necessary and sufficient roles of Hh, Dpp and N signals at their normal time and place of action, and at endogenous expression levels, we have determined the effects on differentiation of mutations of ci, Mad and Su(H), the essential transcription factor targets of Hh, Dpp and N, respectively, in all single-, double- and triple-mutant combinations. We also assessed the affects of some other genotypes. The complete results are summarized in Table 1. Overlap and redundancy between signals makes parallel comparison of all these results necessary in order to obtain a full picture (Table 1).

Ato expression at first fails in clones mutant for the Hh transduction protein Smo. Ato is absent cell autonomously, consistent with Hh being the cell-cell signal that normally regulates the onset of Ato expression at this time. Several hours later, smo-mutant cells differentiate following the appearance of weak, delayed Ato expression(Strutt and Mlodzik, 1997; Dominguez, 1999)(Fig. 2A).

To test whether the smo mutations completely abolished Hh signaling, Ci155 accumulation was examined. As well as stimulating Ci155 transcriptional activator activity, Hh causes Ci155 accumulation by preventing SCF^Slimb from processing Ci155 to the repressor form Ci75(Ingham and McMahon, 2001). Ci155 is specifically detected by monoclonal antibody 2A1, which does not recognize Ci75 (Motzny and Holmgren,1995; Aza-Blanc et al.,1997). In the smo-mutant cells, Ci155 stabilization was not abolished but only delayed (Dominguez,1999) (Fig. 2B). As the smo mutations have since been shown to be null(Chen and Struhl, 1998), a slower-acting, smo-independent mechanism of Ci155 stabilization was implied. Ci155 was later lost from the most posterior margins of smo-mutant clones (Fig. 2B).

Since smo-mutant cells differentiate in response to Dpp(Greenwood and Struhl, 1999; Curtiss and Mlodzik, 2000), we tested whether Ci155 was being stabilized by Dpp signaling. Ci155 levels were examined in clones of cells mutant for both smo and tkv, or for both smo and Mad. tkv encodes the Type I Dpp receptor, Mad encodes the essential transcription factor target. Ci155 did not accumulate when both Hh and Dpp signal reception was inactivated, indicating that Dpp signal reception was required to accumulate Ci155 proteins in the absence of Hh signal reception (Fig. 2C,D). A modest but reproducible reduction in Ci155 levels was seen in cells mutant for tkv alone, indicating that Dpp signaling through Tkv contributes to the level of Ci155, even in the presence of the Hh pathway (Fig. 2E).

Dpp signaling might promote furrow progression and differentiation by regulating Ci155 processing to Ci75, as Hh does. If so, ci-mutant cells should resemble cells that are unable to respond to either Hh or Dpp,and thus would be unable to differentiate. An alternative model was that Dpp acted independently of its effects on Ci. If so, ci-mutant clones should exhibit delayed differentiation as do clones mutant for smo. Cells homozygous for a deletion of the ci gene were examined to distinguish between these two models.

Unexpectedly, clones of ci-mutant cells differentiated normally in all respects (Fig. 3A-C). The temporal and spatial pattern of ato expression was completely normal in ci-mutant cells, unlike in cells unable to respond to Hh or in cells unable to respond to Hh or Dpp (Fig. 3A). Normal neural differentiation occurred in ci-mutant cells without any delay (Fig. 3B and data not shown). Cells lacking ci were morphologically normal in adults, both externally and on sectioning to reveal differentiated cellular morphology (Fig. 3C). There were no defects in ommatidial chirality or planar polarity. The genotype of the ci-mutant cells was unequivocally confirmed using antibodies against Ci to detect ci-mutant cells directly (Fig. 3A). While this paper was under review, Pappu et al. also reported normal eye development by large clones lacking ci (Pappu et al., 2003).

One possibility suggested by these results is a novel pathway of Hh signal transduction that is dependent on smo but not on ci. A second possibility is that Hh signaling is essential only to prevent formation of the repressor protein Ci75. If Ci155 was unimportant, deletion of the entire ci gene could mimic Hh signaling, by eliminating Ci75.

If smo signals independently of ci in the eye, smo ci clones should show delayed differentiation like smo clones(e.g. Fig. 3D). If smois only essential to eliminate Ci75, smo ci clones should develop normally without any delay, like ci clones. Fig. 3E shows that smo ci clones developed normally like ci, and were not delayed like smo clones of the same size (Fig. 3D). Therefore, smo appeared to be essential only to prevent processing of Ci to the Ci75 repressor protein.

We targeted ectopic expression of different forms of Ci protein to the developing eye. The Gal4 line hairy^H10 was used to target UAS:ci^Cell expression anterior to the morphogenetic furrow (Ellis et al., 1994). ci^Cell encodes a Ci protein truncated at amino acid 975 and mimics Ci75 (Methot and Basler,1999). Atonal expression was reduced, although eye patterning occurred normally (data not shown). Fng:Gal4 (a gift of M. Mlodzik) was used to misexpress UAS:ci^Cell ventrally(Fig. 3F). When expressed under the control of Fng:Gal4, ci^Cell prevented furrow progression and differentiation in ventral cells, but permitted furrow progression across the dorsal region of the eye disc (Fig. 3G). Fng:G4 and h^H10 were also used to drive expression of UAS:ci^U, which encodes a Ci protein with a deletion of amino acids 611-760, which is defective in processing to Ci75 and which behaves as a Hh-dependent activator protein(Methot and Basler, 1999), and UAS:ZnAD and UAS:ZnRD transgenes, in which the DNA binding domain of Ci was coupled to transcriptional activator and repressor domains, respectively(Hepker et al., 1997). These were without detectable effect (data not shown). Taken together, these findings suggested that differentiation might depend on blocking production of Ci75 in response to Hh or Dpp.

There also had to be a ci-independent signal required to induce differentiation even in the absence of Ci. If differentiation depended entirely on eliminating Ci75, deletion of the ci gene would remove the barrier to differentiation anywhere in the eye field, but instead ci-mutant cells initiated differentiation in the same temporal progression as wild-type cells (Fig. 3A). ci-independent differentiation could not depend on Hh, because smo ci cells also initiated differentiation with precisely normal timing (Fig. 3E).

The second signal might be Dpp, acting independently of Ci. If this was correct, we would expect that Mad ci mutant cells would fail to differentiate. Mad ci mutant cells were examined and were found to differentiate, but such differentiation was delayed(Fig. 4A,B). This result confirmed that Dpp signaling through Mad contributed to differentiation in the absence of ci, but that some signal from posterior cells was still able to progress slowly through Mad cimutant cells.

One possibility is that mutation of Mad did not completely abolish Dpp signaling. The Mad^1-2 allele, although similar to null alleles of tkv in effects on patterning, permits more normal growth than do tkv-null alleles and so retains a minimal response to Dpp. Perhaps this limited activity becomes significant in the absence of ci. Another possibility is that N signaling might contribute to differentiation in Mad ci cells. To distinguish whether N or residual Dpp signaling was responsible for the differentiation of Mad ci-mutant cells, we examined tkv ci-mutant cells, as well as cells triply mutant for Su(H), Mad and ci.

Differentiation was not observed in tkv ci-mutant clones. Clones of cells mutant for both tkv and ci grew poorly, and only small clones were recovered (Fig. 4C,D). Such clones were rarely recovered in the posterior region of the eye disc, where they seemed to sort out. Large clones of tkv ci-mutant cells were obtained using the Minute Technique. These large clones always had round shapes, indicating that there was reduced mixing with wild-type cells throughout the eye disc (data not shown). No retinal differentiation was detected. These findings indicate that some Dpp signaling occurred in Mad ci-mutant cells, which contributed to the slow differentiation of Mad ci cells. Clones of Mad Su(H)ci-mutant cells also failed to differentiate, indicating a role for N signaling in the delayed differentiation of Mad ci-mutant cells(Fig. 4E).

Two interpretations of the role of N could be considered. One was that N signaling and Dpp signaling were each required for differentiation in the absence of ci. The alternative was that N activity enhanced sensitivity to Dpp so that the limited Dpp signaling that occurs in Mad-mutant cells became sufficient for delayed differentiation.

If N signaling was required for differentiation in the absence of ci, we would predict that Su(H) ci-mutant cells would be unable to differentiate. However, Su(H) ci-mutant cells did initiate differentiation, but such differentiation was delayed(Fig. 4F). Thus N signaling was not absolutely required for differentiation. Delayed differentiation in the absence of Su(H) and ci must depend on Dpp signaling, as it did not occur in Mad Su(H) ci-mutant cells. Su(H) mutations must reduce the effectiveness of Dpp, as the timing of differentiation was normal in ci-mutant cells but delayed in Su(H) ci-mutant cells (differentiation in ci-mutant clones must be due to Dpp signaling, because tkv ci-mutant cells did not differentiate). These results indicate that N signaling made Dpp signaling more effective, at least in the absence of Ci.

The differentiation of Su(H) ci-mutant cells differed from that of Su(H)-mutant cells. Su(H)-mutant cells show a profound neurogenic phenotype in which the majority of mutant cells take an R8 fate(Ligoxygakis et al., 1998)(Fig. 4G). Morphogenetic furrow progression can accelerate so that differentiation begins earlier at the anterior of large Su(H)-null clones than it does in nearby wild-type regions (Li and Baker, 2001)(Fig. 4G). By contrast,differentiation of Su(H) ci-mutant cells was delayed, and R8 differentiation was increased only moderately compared with wild type(Fig. 4F). These data show that differentiation without Su(H) was more effective in the presence of ci. This was the first data to indicate a role of Ci155 that could not be replaced by deleting the ci gene to eliminate Ci75. More evidence is reported below.

One model for the importance of Ci75 was that Ci75 antagonized Dpp and N function. In this view the only normal role of Hh would be to help downregulate Ci75 at the furrow so that Dpp and N could act more promptly. Delayed differentiation in smo clones would be due to ci75 antagonizing the Dpp/N function that we have found is sufficient for normal furrow progression in the absence of ci. This model does not predict delayed differentiation in Su(H) ci-mutant clones(Fig. 4F; see above).

To isolate the role of Hh we examined cells mutant for Mad and Su(H). If the only role of Hh was to derepress Dpp and N signaling,then Mad Su(H)-mutant cells would not differentiate because of missing Dpp and N signals. By contrast, Mad Su(H) cells differentiated at the normal rate (Fig. 4H). A neurogenic phenotype reflected dependence of lateral inhibition on Su(H) (Ligoxygakis et al., 1998). Initiation of Atonal expression has also been reported in Dl Medea-mutant cells, which should resemble Mad Su(H) cells in lacking N and Dpp signaling(Baonza and Freeman, 2001). These results show that Dpp and N were dispensable for differentiation if Hh signaling was intact. As Mad Su(H) ci cells did not differentiate,differentiation of Mad Su(H) depended on a positive role of Ci that was not mimicked by deleting the ci gene to remove Ci75. Thus Hh,through Ci155, could drive differentiation in the absence of Dpp and N signaling.

Notch signaling contributes to differentiation by downregulating Hairy and Extramacrochaetae expression at the morphogenetic furrow(Baonza and Freeman, 2001). Hairy is a transcriptional repressor protein that antagonizes Atonal(Ohsako et al., 1994; Brown et al., 1995). We monitored Hairy expression to evaluate the contributions of Hh, Dpp and N signaling to turning off Hairy, and to correlate this with differentiation. The results are summarized in Table 1. Hh, Dpp and N signals do not appear essential to turn Hairy on,although Dpp does contribute because Hairy levels appear lower in clones mutant for tkv, Mad and their combinations with other mutations(Greenwood and Struhl, 1999)(Fig. 4, and data not shown).

One, or more, of Dpp, Hh or N signaling is required to turn Hairy off at the morphogenetic furrow, because Hairy expression was maintained cell autonomously in Mad Su(H) ci clones(Fig. 4E). Hairy was turned off in Mad ci clones and tkv ci clones, although downregulation was delayed in the center of the clone(Fig. 4B,D). This implies that N signaling is sufficient to downregulate Hairy in response to a signal from more posterior cells outside the clone.

N signaling was required, and was sufficient, for Hairy downregulation,because Hairy was not shut off in Su(H) ci clones(Fig. 4F). If N signaling was essential to shut off Hairy under all circumstances then we would expect Hairy expression to be maintained autonomously in Su(H) clones. By contrast, there was only a brief delay to shutting off Hairy in Su(H)clones (Fig. 4G). Hairy was also shut off in Mad Su(H) clones, although after a delay(Fig. 4H). These results show that either N or Hh signals from posterior cells is sufficient to shut off Hairy expression, but that Hairy expression is maintained indefinitely in Mad Su(H) ci and Su(H) ci cells unable to respond to either pathway. Downregulation of Hairy in response to Hh as well as N explains why N is not required for differentiation in response to Hh, even though it is required for differentiation in response to Dpp.

Downregulating Hairy was neither necessary nor sufficient for differentiation. Whereas Mad Su(H) ci-mutant cells both failed to differentiate and maintained Hairy expression(Fig. 4E), tkv cicells did not differentiate even though Hairy expression was shut off(Fig. 4C,D). By contrast, Hairy expression was maintained in Su(H) ci-mutant clones, even though Su(H) ci cells could differentiate. However, Hairy downregulation may contribute to prompt differentiation because differentiation is delayed in Su(H) ci compared with ci, and is less neurogenic in Su(H) ci than in Su(H)(Fig. 4F,G).

As Hairy was shut off by either Hh or N signaling, Hairy maintenance away from the boundaries of Mad ci, tkv ci and Mad Su(H) clones shows that these genotypes were defective for Hh and Dl secretion. Hh is normally secreted by differentiating photoreceptor cells(Ma et al., 1993). Atonal and then Egf receptor activity promotes expression of Dl, the activating ligand for Notch (Baker and Yu, 1998; Tsuda et al., 2002). Both Hh and Dl expression should be absent or delayed in Mad ci, tkv ci and Mad Su(H) clones.

Table 1 summarizes all of our data. Initially our results suggested that Dpp and Hh might be redundant through common regulation of Ci. When Ci was found to be dispensable, the simplest interpretation was that the repressor Ci75 was more important than Ci155. Others recently arrived at a similar conclusion(Pappu et al., 2003). However,further work revealed roles for Ci155, and redundancies between Hh, Dpp and N,as the explanation for normal progression of differentiation in the absence of any one of the important components ci, Mad or Su(H). There might be other signals awaiting discovery, as Hh, Dpp and N cannot explain the expression of hairy or of genes such as eyes absent during furrow progression (Curtiss and Mlodzik,2000) (Fig. 4E and N.E.B., unpublished).

The pathways implied by our results shown in Fig. 5. Fig. 5A shows how Hh, Dpp and N signaling pathways act within each cell. Fig. 5B illustrates the spatial and temporal relationships of the extracellular signals during morphogenetic furrow progression.

The development of Mad Su(H) ci-mutant cells is a helpful starting point, as they may reflect a `ground state' of eye development that requires extracellular signals to differentiate. Mad Su(H) ci cells fail to express the atonal or senseless genes that initiate R8 differentiation, and, consequently, fail to support retinal differentiation. This shows that the absence of Ci75 is not sufficient for differentiation. Dpp alone can induce Ato (e.g. in Su(H) ci clones), but N and Dpp signaling together are required to activate Atonal with normal kinetics, as occurs in ci-mutant cells. N signaling alone (in tkv ciclones) is insufficient. In the presence of Ci, prompt differentiation required Hh to downregulate Ci75, and differentiation was delayed in Smo clones that lacked this input. The normal role of Hh is not just to remove Ci75 thus permitting Dpp and N to work, because Atonal is turned on normally in Mad Su(H) clones that do not respond to Dpp or N signals. Such differentiation depends exclusively on Hh yet progresses normally, except that a neurogenic phenotype reflects dependence of lateral inhibition on Su(H). Hh depends positively on ci to drive differentiation in Mad Su(H) cells and, therefore, requires Ci155. The positive role of ci can also be inferred from the delayed differentiation of Su(H) ci clones in comparison with Su(H) clones.

We find that Hairy is downregulated redundantly by Hh and N signaling. Prolonged Hairy expression is not sufficient to block differentiation completely but it does antagonize it (e.g. in Su(H) ci clones). Downregulation of Hairy in response to Hh as well as N explains why both ci and Su(H) mutant clones can differentiate promptly, and why N enhances differentiation in response to Dpp but is not required for differentiation in response to Hh.

Comparison between Mad Su(H) ci cells that do not differentiate and Mad Su(H) cells that do shows that Hh signaling is sufficient to initiate eye differentiation. This is consistent with previous studies of ectopic Hh activation (Heberlein and Moses, 1995; Li et al.,1995; Ma and Moses,1995; Pan and Rubin,1995; Strutt et al.,1995). Our experiments confirm this conclusion at the normal time and place of Hh signaling at the anterior of the morphogenetic furrow, and confirm directly that Dpp and N signaling are not necessary for Hh signaling to be sufficient.

Comparison between Mad Su(H) ci cells and Su(H) ci cells shows that Dpp signaling is sufficient to initiate eye differentiation in its normal location in the absence of Hh or N signals, but such differentiation is delayed. The normal timing of differentiation is restored by combined Dpp and N signals (in ci clones). This is the basis for the ectopic differentiation on co-expression of Dpp and Dl ahead of the furrow(Baonza and Freeman, 2001).

Superficially, our results differ from previous ectopic expression studies that concluded that Dpp signaling alone was not sufficient to induce ectopic differentiation in all locations (Pignoni and Zipursky, 1997; Greenwood and Struhl, 1999; Baonza and Freeman, 2001). This discrepancy is probably explained by the baseline repressor activity of Su(H) protein(Hsieh and Hayward, 1995; Morel and Schweisguth, 2000). Our previous work shows that without N signaling, repressor activity of Su(H)protein retards differentiation (Li and Baker, 2001). Dpp signaling is sufficient for differentiation in our experiments where the Su(H) gene has been deleted. In the presence of the Su(H) gene, Dpp may be most effective at locations where there is little Su(H) repressor activity, such as close to the morphogenetic furrow where N signaling is active.

Comparison between Mad Su(H) ci cells, which do not differentiate,and Mad ci or tkv ci cells, which differentiate slowly or not at all, shows that Notch signaling alone is insufficient for differentiation. Premature differentiation reported when N is activated ectopically ahead of the furrow must reflect endogenous Dpp signaling at such locations (Baonza and Freeman,2001; Li and Baker,2001).

Our experiments reveal an outline of the mechanisms of Hh, Dpp and N redundancy (Fig. 5A). First,our results show that Mad and Ci independently reinforce differentiation,presumably through the transcription of target genes because Mad is sufficient for differentiation in the absence of Ci, and vice versa. Our results show unequivocally that the transcriptional activator Ci155 activates differentiation in addition to Ci75 antagonizing differentiation.

It was surprising to find that Dpp stabilizes Ci155 in the absence of Smo,which suggested Dpp input into Hh signal transduction. Although the requirement for smo-dependent input through fused makes it unlikely that Ci155 is functional in smo clones(Ohlmeyer and Kalderon, 1997; Methot and Basler, 1999),Ci155 accumulation might be associated with reduced Ci75 levels. Ci75 is shown to repress differentiation in smo clones because smo ciclones differentiation normally. Ci155 stabilization cannot be due to an indirect effect of Dpp signaling on Hh, Ptc or Smo expression levels because the effect is detected in the absence of smo, and, therefore,reflects an effect on Hh signal transduction components downstream of Smo. One idea is that Dpp signaling (or Dpp-induced differentiation) may replace SCF^Slimb processing of Ci (which cleaves Ci155 to Ci75) with Cullin3-mediated Ci degradation, just as normally occurs posterior to the morphogenetic furrow (Ou et al.,2002). In a smo clone, Ci155 would accumulate because Smo is required for Cullin3 to degrade Ci (Ou et al., 2002) (N.E.B., unpublished). However, the SCF^Slimb-to-Cullin3 switch may not be the only effect of Dpp on Ci processing, because Tkv slightly enhanced Ci155 accumulation even when smo is present (Fig. 2E).

Finally, downregulation of Hairy by N requires the Su(H) gene. N also overcomes baseline repressor activity of Su(H) protein to promote progression of differentiation (Li and Baker, 2001). This role of N must be independent of Hairy.

Dl, Hh and Dpp are generally thought to signal over very different distances. How can signals of such different range substitute for one another to permit normal eye development? Fig. 5B shows signal sources and targets in the eye disc. Dpp is transcribed in response to Hh signaling and is produced where Ci155 levels are highest (Heberlein et al.,1993; Strutt and Mlodzik,1996). Dl is regulated by Hh indirectly through Ato and Ato-dependent Egfr activity in differentiating cells(Baker and Yu, 1998; Tsuda et al., 2002). Hh is expressed most posteriorly of the three, in differentiating photoreceptors(Ma et al., 1993).

Eye differentiation uses Hh to progress through cells unable to respond to Dpp (tkv, Mad) or N (Su(H)). The range of Hh diffusion depends in part on the shape of the morphogenetic furrow cells(Benlali et al., 2000). The Dpp that drives differentiation through ci-mutant cells unable to respond to Hh must diffuse from outside the ci clones because Dpp synthesis is Hh dependent (Heberlein et al.,1993; Methot and Basler,1999). Large ci clones develop normally so Dpp diffusion cannot be limiting (dpp-mutant clones offer no information about the range of Dpp because they express and differentiate in response to Hh). Instead the rate of progression in response to Dpp is controlled by Dl. Dl signals over, at most, one or two cell diameters at the morphogenetic furrow(Baker and Yu, 1997).

The previous view of eye patterning was influenced by the morphogen function of Hh and Dpp in other discs(Nellen et al., 1996; Strigini and Cohen, 1997). It was thought that domains of Ato and Hairy expression reflected increasing concentrations of Hh and Dpp (Greenwood and Struhl, 1999; Lee and Treisman, 2002). Our data shows that, in the eye, the combination of signals is important. Differentiation is triggered where Dl and/or Hh synergize with Dpp, regardless of where the source of Dpp is. The additional requirements limit Dpp to initiating differentiation at the same locations that Hh does.

We thank Y. Chen, J. Curtiss, L. Firth, D. Kalderon, A. Koyama-Koganeya, G. Mardon, M. Mlodzik, K. Moses, J. Treisman and D. Tyler for comments on the manuscript; and S. Carroll, Y.-N. Jan, H. Bellen, F. Schweisguth, G. Struhl,A. Tomlinson, J. Treisman, M. Mlodzik, the Developmental Studies Hybridoma Bank at the University of Iowa, and, especially, K. Basler and R. Holmgren for strains and antibodies. We also thank Sung-Yun Yu, Cynthia Zapata and Heather Schultz for technical assistance. Confocal microscopy was performed at the AECOM Analytical Imaging Facility. This work was supported by grants from the NIH (GM47892 and GM61230). N.E.B. is a scholar of the Irma T. Hirschl Trust.