Two-step selection of a single R8 photoreceptor: a bistable loop between senseless and rough locks in R8 fate

The stereotyped pattern of adult Drosophila eyes results from exact specification of R8 precursors within the field of undifferentiated cells forming the larval eye imaginal disc (reviewed by Frankfort and Mardon, 2002; Hsiung and Moses, 2002). Once the R8 precursor is selected, it initiates ommatidial assembly by recruiting undifferentiated cells to the photoreceptor fate(Freeman, 1996; Tio et al., 1994). As there is no migration of cells in the eye, the establishment of the R8 cell array sets the pattern for the rest of eye development(White and Jarman, 2000). The proneural gene atonal (ato) is required for Drosophila eye development and resolution of its expression within the morphogenetic furrow (MF) determines the arrangement of R8 cells(Jarman et al., 1993b; Jarman et al., 1994). The MF is a physical marker of the wave of differentiation that progresses across the eye disc during the third larval instar and leaves a developing array of photoreceptors in its wake (Ready et al.,1976; Tomlinson and Ready,1987). Ato is initially expressed in a dorsal-to-ventral stripe just anterior to and within the MF (Fig. 1A) (Jarman et al.,1994; Jarman et al.,1995). This stripe of Ato resolves to evenly spaced clusters of 10-15 cells known as intermediate groups (IGs)(Fig. 1B). This column of IGs is defined as column 0. From these IGs, a single cell is selected to continue to express Ato and begin R8 differentiation. The first column of selected R8s lies immediately posterior to the IGs and is identified as column 1. A new column of selected R8s emerges from the MF every 2-3 hours and is staggered out of phase with previous and subsequent columns, producing the characteristic hexagonal pattern of the adult eye(Wolff and Ready, 1991). R8 development also depends on the zinc finger transcription factor Sens(Frankfort et al., 2001; Nolo et al., 2000). Without sens, an R8 precursor is selected from the IG but fails to differentiate as an R8. Ato activates sens expression in a subset of IG cells, and then resolves together with Sens to a single R8 precursor in column 1. Ato and Sens are then co-expressed in the selected R8 until Ato is downregulated after column 3 (Frankfort et al., 2001; Jarman et al.,1994). Sens continues to be expressed in R8 through adult stages and is required for terminal R8 differentiation during pupation(Domingos et al., 2004; Sprecher and Desplan, 2008; Xie et al., 2007).

Two models have been proposed to explain how a single R8 is selected from an IG. The distinction between these models is important because they define different populations of cells with equal potential to become R8 precursors. One model is based on parallels between retinal differentiation and the early development of other peripheral nervous system (PNS) organs (reviewed by Bray, 2000; Ghysen and Dambly-Chaudiere,1989). In this model, IGs are considered roughly equivalent to proneural clusters in PNS differentiation, R8 cells are considered analogous to sensory organ precursors (SOPs), and Notch-mediated lateral inhibition is necessary to select a single SOP or R8 precursor (reviewed by Lai, 2004; Voas and Rebay, 2004; Baker et al., 1996; Baker and Yu, 1997; Baker and Zitron, 1995; Cagan and Ready, 1989; Lee et al., 1996). Thus, if lateral inhibition is disrupted, clusters of R8 precursors are predicted to develop (Fig. 1C). However,mutations in genes outside the lateral inhibition pathway, such as rough (ro), which encodes a homeodomain transcription factor, are also capable of generating additional R8 photoreceptors,suggesting other mechanisms may be required(Cagan, 1993; Rawlins et al., 2003; Spencer and Cagan, 2003). Therefore, a second model was introduced, the `R8 equivalence group' model(Dokucu et al., 1996). In this model, nuclei of three cells at the posterior edge of an IG migrate apically and continue to express Ato while in neighboring IG cells Ato expression is lost (Dokucu et al., 1996; Sun et al., 1998). Then, in two of the three R8 equivalence group cells, Ro represses Ato, leading to the selection of a single R8. Thus, in ro-null mutants up to three R8 precursors per ommatidium are predicted to develop(Fig. 1D)(Dokucu et al., 1996; Heberlein et al., 1991). Consistent with this model, ectopic Ro is capable of repressing Ato anterior to the MF and Ro is expressed in a pattern complementary with the posterior border of Ato expression in the MF (Chanut et al., 2000; Dokucu et al.,1996). Despite the differences between the two models, both are based on the premise that ectopic R8s are formed when the initial pattern of Ato expression is not refined to a single cell. In other words, an undifferentiated cell can only develop as an R8 in the context of ongoing Ato expression. Both models were proposed before sens was identified and its pivotal role in R8 development explored.

Sens is required in R8 to repress Ro expression and thereby allow R8 differentiation (Frankfort et al.,2001). In the absence of sens, Ro expression expands into the previously selected R8 precursor and this extra Ro-expressing cell switches fate to an R2,5 cell type. How Ro functions in this process is unknown as the molecular mechanisms controlling recruitment of the R2,5 precursors are poorly understood. In addition, ectopic Sens is capable of repressing endogenous Ro expression and ectopically inducing the R8 cell fate. Moreover, ro is probably a crucial early target of Sens repression as R8 differentiation is often restored in sens, ro double mutant ommatidia (Frankfort et al.,2001). The presence of three Ro-expressing cells in sensmutants also indirectly supports the presence of an R8 equivalence group,defined as three cells with equal potential to differentiate as R8 precursors. However, the issue of how a single R8 is selected from this potential equivalence group has not been specifically addressed or answered.

In this work, we show that ro is not required for the initial selection of a single R8 precursor. We show that in ro-null mutants,the single R8 fate persists for several hours and that ectopic Sens-expressing R8s develop in the absence of ectopic Ato 6 hours after normal R8 specification. Furthermore, we show that Ro directly represses senstranscription in the R2,5 precursors. Together with our previous report that Sens repression of Ro is required for R8 differentiation, our current findings suggest that the R8 and R2,5 cell precursors comprise the R8 equivalence group and that a negative regulatory loop between sens and ro is required to select the R8 vs. R2,5 cell fate from among these cells. We also report the identification of an enhancer that is necessary and sufficient for R8-specific sens expression and characterize distinct elements within the enhancer responding to inductive and repressive signals during R8 selection. Our data suggest that a two-step process is required for R8 selection. Initially, Ato directly activates sens expression in the IG and lateral inhibition transiently selects a single R8 from the IG cells. Then, after the correct pattern of R8 precursors is established by lateral inhibition, Ro is required to maintain this pattern by direct repression of sens in the R2,5 precursors.

Twelve PCR products spanning the sens genomic locus were generated by PCR with 5′ EcoRI and 3′ BamHI tails and cloned into pH-Pelican or pH-Stinger(Barolo et al., 2000). Mutations in predicted homeodomain binding sites and E-boxes were made by using mutagenic primers; sequences can be obtained upon request. The sens-L genomic rescue was prepared as described previously(Venken et al., 2006). TheΔ F2 genomic rescue construct was created by inducing recombination between the sens-L genomic rescue and a mutagenic PCR fragment that lacks the F2 region (see Fig. S1 in the supplementary material). F2-sens was generated in the pCaSpeR-4 vector(D'Avino and Thummel, 1999). The F2 enhancer sequence and hsp70 promoter were excised from F2-GFP and cloned upstream of the 2558 bp sens cDNA and an SV40 poly A sequence.

For transgenic fly generation, embryo progeny of yw virgins crossed to yw; KiΔ2-3 males were injected with pH-Stinger-based constructs. Third instar larvae heterozygous for reporter constructs were dissected and stained (see Pepple et al., 2007).

The following stocks were used: sens^E1: ro^X63: RM104:Dl^6B: Dl^RF: sens^E2 FRT80B/TM6B: FRT82D ato¹: FRT82D ro^X63: Fragment E1/CyO, hs-hid;sens^E2 FRT80B/TM6B: yw,hsflp; M(3),arm-lacZ,FRT80B/TM6B: yw,hsflp;FRT82D arm-lacZ/TM6B. Clones were generated by standard protocols (see Pepple et al., 2007) with minute heat shocks at 37°C for 1 hour 40-42 hours after egg laying. Dl^ts is Dl^6B/RF(Baker and Zitron, 1995). Dl^ts animals were raised at 18°C until third instar then vials were submersed in a 31°C water bath for 6 hours. Larvae were immediately dissected and stained (see Pepple et al.,2007). Primary antibodies used were: rabbit anti-GFP (1:1000,Molecular Probes); guinea-pig anti-Ato (1:1000, a gift from Hugo Bellen) for DAB stains; rabbit anti-Ato (1:3000, a gift from Kwang Choi) for fluorescent stains; guinea pig anti-Sens (1:2000, a gift from Hugo Bellen); and mouse anti-β galactosidase (1:1000, Promega). Goat anti-rabbit Cy3 and goat anti-guinea pig Cy5 secondary antibodies were obtained from Jackson laboratories (West Grove, Pennsylvania, USA). Goat anti-mouse Alexa, goat anti-guinea pig Alexa and goat anti-rabbit Alexa secondary antibodies were obtained from Molecular Probes (Eugene, Oregon, USA). All secondary antibodies used at a 1:500 dilution in PAXDG (PBS with 1% BSA, 0.3% Triton X-100, 0.3%sodium deoxycholate and 5% NGS). Scanning electron microscopy was performed(see Pepple et al., 2007).

In Fig. 3I-J, Sens and Ato expression were counted only in ommatidia that could be unambiguously assigned a column designation. For each column, percentages of ommatidia containing single positive cells and multiple positive cells were determined. Ato data were generated from nine wild-type discs and 17 ro^X63discs. Sens data were generated from 13 wild-type discs and 17 ro^X63 discs. An average percentage for each column was determined using the normalized percentage from each disc. Error bars represent the standard error of the mean. Student's t-tests were performed to determine P values.

Ato/Da EMSAs were performed (see Jarman et al., 1993b). Wild-type probes for Ato/Da EMSAs are as follows:E1, TTAGTACCGGACCGACATATGGTCAAAAAGCCGA; E2, TTAAGCCGACGAAGACAGTTGCCAGAGTCCTTTG; E3, TTAGTCACTGTTCTTCAGCTGTTTATGTATAAAA; and E4, TTAATTCGTGCTTTACATCTGTTCACCATTGGAG. Italicized thymidines were added to probe sequence for radiolabeling withαP³²-dATP, core E-boxes sequence underlined. For all mutant probes, CANNTG was mutated to AANNTT. Ro EMSAs were performed (see Heberlein et al., 1994). Probes for Ro EMSAs are as follows: wt,ATTTATGTACAAATTACAATCATAATAATTT;H1^*,ATTTATGTACAAGGGGCAATCATAATAATTT;H2^*,ATTTATGTACAAATTACAATCATGGGGATTT;H1,2^*,ATTTATGTACAAGGGGCAATCATGGGGATTT. Gels were dried before autoradiography.

The two models for R8 selection have been tested primarily using Ato expression as the R8 marker (Baker et al.,1996; Baker and Yu,1997; Baker and Zitron,1995; Dokucu et al.,1996; Heberlein et al.,1991; Lee et al.,1996). Sens is an additional, consistent and more specific marker of the R8 cell fate. We therefore re-evaluated the role of lateral inhibition and rough in R8 selection using Sens to mark R8s. Lateral inhibition was disrupted using a combination of Delta (Dl) alleles that generate Dl^ts animals (Dl^6B/RF)(Baker and Zitron, 1995). Previous work has demonstrated that, after disruption of lateral inhibition in Notch^ts and Dl^ts animals, Ato is expressed in groups of cells in column 1 rather than in single R8s (Model 1)(Lee et al., 1996). In agreement with this data, we found that Sens expression also fails to resolve to a single cell in column 1 when lateral inhibition is disrupted(Fig. 2C,G). This supports the previous model that lateral inhibition is necessary for selection a single Ato- and Sens-expressing R8 precursor in column 1.

In contrast to the Dl^ts phenotype, in ro^x63 null mutants only single Sens-expressing R8 cells are found in columns 1 and 2 (Fig. 2D,H,I). This is not consistent with a previous report that in ro mutants Ato expression is found in two or three cells in column 1,owing to failure of resolution of the R8 equivalence group (Model 2)(Dokucu et al., 1996). To evaluate this discrepancy, we closely re-examined Ato expression in ro^x63 null mutants and found that 6±3% of ommatidia in column 1 do have two Ato-staining cells (data not shown). However, this is not significantly different (P=0.8) from wild-type discs, where two Ato-positive cells are found in 6±2% of column 1 ommatidia (data not shown). The occasional second Ato-expressing cell does not persist in either genotype, and by column 2, only a single Ato staining cell is seen in all ommatidia (data not shown). These data indicate that, although Ato expression is not limited to a single cell in column 1, by column 2 in both wild-type and ro^x63 mutant discs, expression of Ato and Sens is always restricted to a single cell. Thus, ro is not required for selection of a single Ato- or Sens-expressing R8 photoreceptor.

In ro^x63 null mutant discs, single Sens-expressing cells are selected and persist for two columns, demonstrating that rois not required for selection of a single R8 precursor. However, in column 3,Sens expression is occasionally found in multiple cells per ommatidium and by column 5 many ommatidia have three Sens-positive cells(Fig. 2I, asterisk). This suggests that the ro phenotype may be due to a later effect on R8 differentiation than previously reported. To better characterize the ro^x63 phenotype, we closely examined ro^x63 discs using Ato and Sens expression as R8 cell markers. Initially, expression patterns of Ato and Sens are the same in both wild-type and ro^x63 discs(Fig. 3A-H). In a subset of IG cells, Ato and Sens are co-expressed (circled in Fig. 3B-D,F-H) and in column 1 and 2 Ato and Sens colocalize to a single R8 precursor (open arrowhead in Fig. 3B-D,F-H). The first difference is found in column 3 where, in wild-type discs, Ato and Sens are expressed in only one cell per ommatidium. By contrast, in ro^x63 discs, 9±3% of ommatidia have multiple Ato-positive cells and 21±5% have multiple Sens-positive cells(Fig. 3I). The difference between wild-type and ro^x63 mutants further increases after column 4, where Sens is expressed in up to three cells per ommatidium in more than 50% of ommatidia (Fig. 3F, white arrowhead; Fig. 3J). Although Sens is a specific and consistent R8 marker, it is possible that not all extra Sens-positive cells in ro^x63mutants are equivalent to wild-type R8s. Therefore, in ro^x63 mutants, supernumerary cells expressing Sens are considered putative R8s (`R8s'). To determine when extra `R8s' form, we calculated the average percentage of ommatidia with extra Sens-expressing`R8s' in the first seven columns of ro<sup>X63</sup> mutant discs(Fig. 3J). We find that the ro<sup>X63</sup> phenotype evolves gradually starting in column 3,where 21±5% of ommatidia contain extra `R8' cells. By column 6, 60-70%of ommatidia have multiple `R8s'. By contrast, in wild-type discs, a single R8 cell is found in every ommatidium. Therefore, in ro<sup>X63</sup>mutants, additional `R8s' develop from cells that begin to express Sens starting in column 3.

Ro is a homeodomain-containing protein and has been shown to bind DNA at two sites in its own enhancer containing an ATTA core sequence(Heberlein et al., 1994). To explore the possibility that Ro directly represses sens, we identified the R8 specific sens enhancer and characterized the mechanisms regulating sens expression. A 645 bp fragment within the second intron of the sens genomic locus named F2 was identified that is sufficient to drive reporter expression specifically in photoreceptors of the developing eye-antennal imaginal disc(Fig. 4). To test whether the F2 region is necessary for R8-specific sens expression, the 645 bp region was specifically deleted from the sens-L genomic rescue construct generating ΔF2 (see Fig. S1 in the supplementary material). In sens-null mutants, one copy of ΔF2rescues the null phenotype in all tissues except the eye (see Fig. S2 in the supplementary material). Thus, F2 is the sens eye enhancer and is necessary and sufficient for R8-specific sens expression.

F2 contains two potential Ro-binding sites known as H1 and H2, for homeodomain 1 and 2 (Fig. 4B,highlighted in red). To test for a direct interaction, electrophoretic mobility shift assays (EMSAs) were performed. A probe containing H1 and H2 is bound specifically by Ro protein in vitro(Fig. 5A). Complete loss of binding occurs with mutation of H2. Mutation of H1 does not prevent Ro binding, but there may be a mild decrease in binding compared with the wild-type probe. To test the in vivo significance of these interactions, each site was mutated in a reporter generated with the minimal R8-specific enhancer, B-short-GFP, and the effect on GFP was evaluated(Fig. 5D-L). Although H1 is not required for Ro binding in vitro, mutation of H1 in B-short (termed H1^*) leads to consistent expression of GFP in two extra cells per ommatidium (arrowheads in Fig. 5H). These two cells were identified as the R2,5 photoreceptor pair by co-localization of GFP with β-galactosidase from the R2,5-specific enhancer trap RM104 (Fig. 5M-O). GFP expression is also expanded into the R2,5 pair with the H2 mutation (H2^*) (Fig. 5K). Mutation of both H1 and H2 (H1,2^*) results in a GFP expression pattern indistinguishable from H2^* (see Fig. S3 in the supplementary material). To test whether the loss of ro function has the same effect on B-short-GFP expression as does mutation of the Ro-binding sites, ro^X63 clones were generated. In the absence of ro function, both Sens and B-short-GFP expression are detected in two to three cells per ommatidium(Fig. 5Q-S, arrowheads). Together with the in vitro binding data, these in vivo results suggest that Ro directly represses sens expression in R2,5 photoreceptors.

In order to identify mechanisms activating sens expression in R8,we performed binding site analysis and functional assays with subfragments of F2 (Fig. 6). Sens expression in the eye is dependent on the proneural bHLH protein Ato(Frankfort et al., 2001). To activate target gene expression, Ato heterodimerizes with another bHLH protein, Daughterless (Da), and binds to the minimal E-box consensus sequence CANNTG (core nucleotides underlined)(Brown et al., 1996; Jarman et al., 1993a; Murre et al., 1989a; Murre et al., 1989b). Four potential E-boxes are present in F2 (E1-E4, identified by green boxes in Fig. 4B and green vertical lines in Fig. 6A), suggesting that Ato directly regulates sens. To test whether these potential binding sites are required for reporter expression, we generated three subfragments of F2 (fragments A, B and C). Fragment A lacks all E-box sequences and does not express GFP in the eye(Fig. 6F). Fragment B contains two E-boxes, E1 and E2, and drives GFP expression strongly starting in column 1, but lacks significant IG expression(Fig. 6L, IGs bounded by white vertical lines). The lack of IG GFP expression with fragment B suggests that E3 and E4 may also be required. Therefore, we predicted that fragment C, which contains all four E-boxes, would recapitulate the complete F2-GFP expression pattern. Owing to the deletion of the Ro-binding sites, additional expression in the R2,5 was also anticipated. As predicted, C-GFP is expressed at high levels in the IGs, suggesting that multiple E-boxes are required for the earliest expression of sens (Fig. 6R). However, GFP expression also expands to nearly every cell posterior to the IGs, suggesting that additional negative regulatory elements other than the Ro-binding sites are missing from fragment C(Fig. 6A, blue bracket).

In order to identify a minimal enhancer containing all positive and negative regulatory regions, fragments B-short(Fig. 6I) and B-long(Fig. 6O) were generated. Fragment B-short contains only E1 and is sufficient for expression in the selected R8s, but lacks IG expression. Fragment B-long contains all four E-boxes and is sufficient for both IG and selected R8 GFP expression. Thus,fragment B-long is the minimal enhancer containing all necessary negative and positive regulatory elements and recapitulates the complete eye specific sens expression pattern.

To test for a direct interaction between Ato/Da heterodimers and the four E-boxes present in the R8 enhancer, EMSAs were performed(Fig. 7A). Ato/Da heterodimers bind strongly to E1 with weaker binding detectable for E4. Binding is lost with mutation of the core E-box sequence. No binding is detected to probes containing E2 or E3. Da homodimers also bind to E1 and E4 (indicated by arrowhead). This binding is lost with mutation of the E-box core sequence. Interaction of Da homodimers with E-box sites has been described previously,but the significance of this interaction in vivo is unknown(Jarman et al., 1993b; Jarman et al., 1994). These data suggest that Ato/Da heterodimers directly regulate sensexpression by binding E1 and E4 in the R8 enhancer. This supports the subfragment analysis that shows that both E1 and E4 are required in fragment B-long for IG reporter expression.

Fragment B-short contains a single E-box, E1, and is sufficient for R8-specific expression starting in column 1. To determine whether E1 is necessary for expression in vivo, two base pairs in the E-box core sequence were mutated in B-short, generating E1^*(Fig. 7G). Mutation of the E-box does not abolish all R8 reporter expression in vivo but delays the onset of GFP expression by three or four columns(Fig. 7G, bracket). Deletion of the entire E-box has the same effect on GFP expression (data not shown) and suggests an Ato-independent enhancer that is sufficient to maintain sens expression is present in fragment E1^*. This is not an unexpected finding as Ato expression ends after column 3, whereas Sens continues to be expressed until early adult stages. No additional transcription factors with the ability to directly activate sensexpression in an Ato-independent manner have been identified.

In this work, we show that Ro directly represses sens in developing R2,5 cells and that de-repression of Sens is sufficient to initiate R8 cell fate in the absence of ectopic Ato(Fig. 8B). Although there are a small number of ectopic Ato-expressing cells in column 3 in ro^x63 mutants, it is not likely that the additional `R8'cells are due to misregulation of Ato as the great majority of ectopic `R8s'never express detectable Ato protein after the intermediate group stage. It is more likely that the extra Ato-positive cells are due to secondary Sens activation of proneural gene expression, a previously reported phenomenon(Acar et al., 2006; Nolo et al., 2000).

In a previously published report, we have shown that sens is required for R8 differentiation to occur through repression of Ro in R8, and that ectopic Sens is sufficient to repress endogenous Ro expression(Frankfort et al., 2001). Thus, in the absence of sens, three R2,5 cells develop and in the absence of ro up to three R8 cells form per ommatidium. This reciprocal phenotype supports the existence of the three cell R8 equivalence group and a mechanism of mutual repression between sens and ro that specifies opposite cell types(Fig. 8C). Although we have shown that one mechanism regulating this mutual repression is the direct repression of sens by Ro, other roles for Ro may exist. We observe that the Ro-binding site mutations do not produce the same level of GFP reporter protein expression elevation in R2,5 precursors that would be predicted from the level of GFP expressed in ro mutants. This suggests that Ro may also regulate sens by repressing an activator of sens expression in R2,5 precursors.

Regardless of the mechanism, the negative-feedback loop between sens and ro is secondary to the initial force driving R8 selection in which Ato and Sens are transiently repressed by lateral inhibition in all but one cell within an IG. Thus, lateral inhibition transiently represses neural differentiation in the eye, establishing the patterned array of precisely spaced ommatidia while retaining the potential for later recruitment of undifferentiated cells to the photoreceptor cell fate. If the effects of lateral inhibition were to repress permanently the potential for neuronal differentiation, further retinal development would be blocked. Therefore, the effects of lateral inhibition must be limited and our data indicate that column 3 is the boundary of its influence. As the effects of lateral inhibition diminish, the negative-feedback loop between sens and ro reinforces the pattern of selected R8s and ensures that only one Sens-expressing cell from the R8 equivalence group develops as an R8. This simple bistable loop translates the transient developmental signal of lateral inhibition into a committed irreversible fate(Ferrell, 2002).

In later R8 differentiation, another bistable loop is used to specify the`pale' or `yellow' subtypes of R8 photoreceptors(Mikeladze-Dvali et al.,2005). During this late developmental step, the bias for the`pale' R8 fate is provided by a signal from a `pale' R7. We propose that the bias signal that tips the fate decision in the sens-ro loop is provided by resolution of Ato to a single cell by lateral inhibition. Ato then directly activates Sens expression and biases that cell to the R8 cell fate. It is not yet known what activates Ro expression and thereby establishes the R2,5 cell fates. However, it has been suggested that epidermal growth factor receptor (EGFR) or Hedgehog signaling may be required for Ro expression(Dominguez, 1999; Dominguez et al., 1998). Ligands for both of these signaling pathways are expressed in developing R8s(Freeman, 1994; Heberlein et al., 1993; Ma et al., 1993; Tio et al., 1994; Tio and Moses, 1997). As a result, after the R8 bias is established, a signal such as the EGFR ligand Spitz could be sent from R8 to the two neighboring cells that bias their sens-ro loop towards Ro expression and the R2,5 fate. Once Ro expression is initiated in the R2,5 pair, the pattern of a single Sens-expressing R8 per ommatidium becomes irreversible.

Proper patterning of the Drosophila eye requires precise selection of R8 precursors in a highly ordered array. Previously, the potential to assume the R8 fate was generally believed to reside in the single cell that achieved the highest balance of proneural induction by ato and escaped repression by lateral inhibition. This concept has influenced the interpretation of mutants that exhibit multiple R8 phenotypes, such as ro, by linking the extra R8s that form to cells that inappropriately maintain Ato expression. However, our data show that the expression pattern of Ato and Sens in a ro-null mutant is not altered in a manner consistent with this model. Our re-evaluation of the ro phenotype suggests the intriguing possibility that undifferentiated cells posterior to the furrow retain the developmental plasticity to develop as R8s even in the absence of ongoing Ato expression.

The ro phenotype demonstrates that, despite initial repression of the R8 cell fate by lateral inhibition, at least two additional cells have the potential to develop as R8s starting in column 3 if Sens expression is de-repressed. One of the subfragments of the sens eye enhancer,fragment C-GFP, is expressed in nearly all cells posterior to the MF,suggesting that sens could be de-repressed in cells other than the R2,5 cell precursors and initiate R8 development. The widespread expression of fragment C-GFP suggests that it lacks an important negative regulatory region distinct from Ro repression. One potential mechanism that may explain the fragment C-GFP expression pattern is that the stripe of Ato expression in the MF confers R8 potential to all cells and that this potential is only transiently repressed by lateral inhibition during patterning. Then, as the effects of lateral inhibition fade, secondary mechanisms repress sensexpression and R8 differentiation in cells posterior to the MF. This model,demonstrated by the function of Ro and suggested by fragment C-GFP expression,is distinct from the previous concept that R8 cell fate is limited to cells of the IG.

The minimal eye specific enhancer of sens, fragment B-long,contains at least four potentially discreet regulatory elements that balance the positive and negative inputs required to specify a single R8 precursor per ommatidium. The first positively acting element is under the direct control of Ato/Da heterodimers and contains E-boxes 1 and 4. This element is required for Ato-dependent sens expression in the IGs and in columns 1-3. Although ato is at the top of the genetic cascade required for eye differentiation, sens is only the third direct target identified in the eye after bearded (brd) and dacapo(dap) (Powell et al.,2004; Sukhanova et al.,2007). We find that Ato/Da heterodimers bind to two E-boxes (E1 and E4) to drive early sens expression in R8. This is in contrast to the previously described direct regulation of sens in SOPs of the embryonic and developing adult PNS by Ato and Scute at a single E-box in their common enhancer (Jafar-Nejad et al.,2003).

The second positively acting regulatory element resides within the boundaries of fragment E1^*, although we did not specifically identify the minimal necessary sequence. This element responds to an Ato-independent mechanism that is sufficient to maintain Sens expression in selected R8 cells after column 3. Sens is known to respond to Ato-independent inductive cues much later in R8 development (48 hours after pupation) when Sens expression requires the spalt genes(Domingos et al., 2004). However, larval expression of Sens is not disrupted in spalt mutants,suggesting the existence of yet another unidentified positive regulator.

In addition to these two positively acting elements, there are also at least two negative regulatory elements. We specifically identified the Ro-binding element H2 that is responsible for repressing Sens expression in R2,5 cells. The second element was not specifically identified, but its presence is suggested by the nearly ubiquitous expression of fragment C-GFP. Together these positive and negative regulatory elements outline an elegant strategy for the multi-staged selection of a single R8 per ommatidium and highlights a model where blocking R8 cell fate potential with sequential,independent, repressive mechanisms is an important strategy for patterning and cell fate development in the Drosophila eye.

We thank Kenneth Dunner, Jr, at the University of Texas High-Resolution Electron Microscopy Facility for assistance with SEM. We also thank Richard Atkinson for expert guidance on confocal microscopy, and Hugo Bellen, Kwang Choi, the Developmental Studies Hybridoma Bank and the Bloomington Stock Center (Indiana University, Bloomington, IN) for flies and antibodies. We especially thank the members of the Mardon Laboratory, Georg Halder, Hamed Jafar-Nejad, Paul Overbeek and Kartik Pappu for their support and comments on this manuscript. This work was supported by funds awarded to G.M. by the Retina Research Foundation, the National Eye Institute (R01 EY011232), an Ophthalmology Core Grant (5P30EY002520) and the National Eye Institute (T32 EY007102). Additional support for M.A. provided by NSF Grant(DGE-0440525).