Logic of Wg and Dpp induction of distal and medial fates in the Drosophila leg

Animal body plans require the specification of unique cell fates along two primary axes: anteroposterior (AP) and dorsoventral (DV). In addition, for those animals bearing appendages, a third, proximodistal (PD), axis must be generated orthogonal to the primary body axes. Studies in Drosophilahave revealed many of the signals and pathways that control the formation of these three axes (reviewed by Morata,2001). However, as many of the same pathways are used for forming all three of these axes, it remains unclear how they are uniquely specified. Moreover, once established, it is not understood how positional information is specified along an the PD axis of an appendage.

In Drosophila, the appendages are derived from imaginal discs,sheets of epithelial cells that are patterned during larval development. Imaginal discs are divided into anterior (A) and posterior (P) compartments,groups of cells that are segregated from each other early in development(Lawrence and Morata, 1977). Compartment boundaries are sources of signaling molecules, morphogens, that provide positional information to the cells in the developing discs(Tabata and Takei, 2004). In the leg disc, Hedgehog (Hh) is expressed and secreted by cells of the P compartment, and induces the expression of two long-range signaling molecules,Decapentaplegic (Dpp) and Wingless (Wg), in A compartment cells that are adjacent to the AP compartment boundary. Hh activates dpp in the dorsal half of the leg disc and wg in the ventral half(Basler and Struhl, 1994). Once activated, the wg and dpp expression domains are maintained by a mutually antagonistic repression between them(Brook and Cohen, 1996; Jiang and Struhl, 1996; Johnston and Schubiger, 1996; Morimura et al., 1996; Penton and Hoffmann, 1996; Theisen et al., 1996). dpp is required to pattern the dorsal half of the leg(Morimura et al., 1996; Theisen et al., 1996), whereas wg is required to specify ventral leg fates(Couso et al., 1993; Johnston and Schubiger, 1996; Struhl and Basler, 1993; Wilder and Perrimon, 1995)(Fig. 1A). Thus, the DV axis of the leg is specified by these two opposing morphogens, probably by regulating unique sets of target genes in a concentration-dependent manner(Abu-Shaar and Mann, 1998; Brook and Cohen, 1996; Hays et al., 1999).

In contrast to the DV axis, Dpp and Wg act combinatorially to generate the proximodistal (PD) axis of the leg by inducing a different set of target genes, including Distalless (Dll) and dachshund(dac) (Campbell et al.,1993; Diaz-Benjumea et al.,1994; Lecuit and Cohen,1997; Mardon et al.,1994) (Fig. 1A). Unlike Wg and Dpp, which are expressed in ventral and dorsal sectors of the leg disc, respectively, Dll and dac are expressed in approximately circular domains whose centers are located where the Wg and Dpp sectors meet in the middle of the disc(Fig. 1A). Dll is expressed in a large central domain of the leg disc that gives rise to the distalmost positions of the adult leg (tarsus and distal tibia), whereas dac is expressed in more medial PD positions. It has been proposed that the PD identity of a cell and, consequently, whether it activates Dll or dac, is determined by integrating the levels of both Wg and Dpp: high levels of Wg and Dpp activate Dll and repress dac, whereas intermediate levels of Wg and Dpp activate dacbut are insufficient to activate Dll(Lecuit and Cohen, 1997). However, an important and unresolved question stemming from this model is how do the sector-like expression domains of wg and dpp, which give rise to ventral and dorsal gradients of Wg and Dpp, respectively,generate nearly circular and concentric Dll and dacexpression domains?

After the expression of Dll and dac is initiated, these genes become independent of Wg and Dpp, and maintain their expression by an unknown mechanism (Galindo et al.,2002; Lecuit and Cohen,1997). By the end of larval development, the PD axis is divided into at least four domains based on the expression of these two genes: a distal Dll-only domain, a domain that expresses both genes (Dll+Dac), a Dac-only domain, which is expanded dorsally, and a proximal domain that expresses neither gene (Fig. 1A). In addition, the distalmost (tarsal) segments of the leg are more finely patterned by the graded activity of the Epidermal Growth Factor Receptor (EGFR) pathway, that controls another set of target genes, including aristaless (al), Drosophila Lim1 (Lim1), Bar and apterous (ap)(Campbell, 2002; Galindo et al., 2002). The activation of this pathway also depends on Wg and Dpp, which trigger the expression and secretion of EGFR ligands from the center of the leg disc.

In this study, we investigate how Wg and Dpp specify the DV and PD axes of the leg. How do these same morphogens regulate different target genes when specifying these two axes? How are Wg and Dpp inputs integrated during the activation of Dll and dac? How do Wg and Dpp create an asymmetric DV axis such that dorsal fates are expanded relative to ventral fates (Fig. 1A)? We provide answers to these questions in part by analyzing the role of brinker(brk) in leg development. brk encodes a transcriptional repressor of Dpp target genes (Campbell and Tomlinson, 1999; Jazwinska et al., 1999; Minami et al.,1999). In the wing, where it has been best studied, brkis itself repressed by Dpp signaling and is therefore expressed as a reciprocal gradient to the Dpp activity gradient. The expression domains of two Dpp target genes in the wing, optimotor blind (omb) and spalt (sal), are established by different sensitivities to Brk repression (Moser and Campbell,2005; Muller et al.,2003). By extension, different levels of Brk are thought to pattern other aspects of the AP axis of the wing. Dpp signaling represses brk by triggering the assembly of a transcriptional repressor complex that is composed of at least three proteins: Medea, Schnurri and Mothers against Dpp (Mad), the nuclear localization of which is induced by phosphorylation upon Dpp signaling (Marty et al., 2000; Pyrowolakis et al., 2004). In addition to repressing brk, Dpp signaling is likely to control the transcription of other target genes via Mad, Medea,and, in some cases, Schnurri. Thus, in principle, Dpp signaling has the potential to regulate gene expression in at least two ways: (1) via Mad/Medea and (2) by silencing Brk (reviewed by Affolter et al., 2001).

In the leg, we show, surprisingly, that brk is not required for DV axis specification but that it plays an essential role in PD axis formation. Thus, Dpp uses distinct mechanisms for specifying these two axes in the leg. In PD axis formation, we show that Dll and dac are activated by Wg signaling and repressed by Brk. Moreover, Dll and dachave different sensitivities to both Brk levels and Wg signaling. The different sensitivities to these two inputs establish the Dll and dac expression domains at different PD positions. By demonstrating that Brk is a repressor and Wg is an activator of PD axis genes, our data explain the requirement for both Wg and Dpp inputs for their activation. When considered together with the expression pattern of brk, our findings also suggest a mechanism for the generation of asymmetry in the DV axis of the leg.

brk^XA is a P (lacZ) insertion and is larva lethal (Campbell and Tomlinson,1999). Mad^1-2 is a strong hypomorph(Wiersdorff et al., 1996),whereas the rest are considered nulls: tkv^a12(Nellen et al., 1994), brk^M68 (Jazwinska et al., 1999), gro^E48 and CtBP^l(3)87De-10. dpp^discs mutant larvae are heterozygous for In(2L)dpp^d12 and Df(2L)dpp^d14 (St Johnston et al., 1990), and wg^CX3/CX4 are hypomorph (Brook and Cohen,1996). Other stocks used were: omb-Z(Grimm and Pflugfelder, 1996), dpp-Z (Blackman et al.,1991), H15-Z (Brook and Cohen, 1996), wg-Z(Ingham, 1991),UAS-brk (Jazwinska et al.,1999; Muller et al.,2003), UAS-axin(Willert et al., 1999),UAS-gro (Hasson et al.,2005), UAS-tkv^QD(Nellen et al., 1996) and UAS-TCF-RNAi (gift by Barry Dickson and Daniel McKay).

Clonal analysis was performed using the FRT/Flp technique(Xu and Rubin, 1993) using the following stocks: gain-of-function clones, y w hs FLP122;tub>y+>Gal4 UAS-GFP. Flip-out clones were induced by heat-shocking the larvae for 10 minutes at 37°C. Loss-of-function clones were generated using the following genotypes: y f^36a brk^M68FRT 18A/FRT18A; hsFlp, w hsFlp GFP FRT 19A/y brk^M68 FRT 19A, gro^E48 or CtBP^l(3)87De-10 82B/FRT82B ubiGFP,Mad^1-2 or tkv^a12 FRT 40A/arm-Z or ubiGFP FRT 40A. Double mutant clones for brk and tkv or Mad were generated as described previously(Campbell and Tomlinson, 1999; Jazwinska et al., 1999). Clones were induced by heat-shocking the larvae for 1 hour at 37°C.

More than 10 clones were analyzed for each genotype, except for the double brk; tkv or brk; Mad clones, where, owing to their low frequency, more than five clones were analyzed for each genotype. Except where indicated, all clones were generated 48-72 hours AEL.

We used the MARCM technique (Lee and Luo, 1999) to express brk+ or axin+ in brk^- cells using: yw tubGal80FRT19A/brk^M68 FRT 19A; tubGal4-UAS lacZ,hsFlp (from G. Struhl)/UAS-axin or UAS-brk. To vary the amount of Axin or Brk, the larvae were grown at different temperatures,17°C versus 29°C (Brand et al.,1994).

More than 10 clones were analyzed for each genotype. Except when indicated,all clones were generated 48-72 hours AEL.

Imaginal discs were fixed and stained using standard procedures. The primary antibodies used were: rabbit and mouse anti-β-Gal (Capell and Promega), mouse anti-Dachsund, mouse anti-Engrailed, mouse anti-Wingless(Hybridoma bank), guinea pig anti-P-Mad (Ed Laufer), guinea pig anti-Distal-less, rabbit anti-Homothorax, rat anti-Aristaless (Andrew Tomlinson) and rat anti-Lim (Gerard Campbell).

In the leg imaginal disc, the expression of dpp and wg is generally restricted to the dorsal and ventral halves, respectively, owing to a mutually antagonistic repression that exists between these two pathways(Brook and Cohen, 1996; Jiang and Struhl, 1996; Johnston and Schubiger, 1996; Morimura et al., 1996; Penton and Hoffmann, 1996; Theisen et al., 1996). As in wing discs, throughout development, brk expression was reciprocal to Dpp signaling in leg discs, as visualized with an antibody directed against P-Mad (Fig. 1B; see Fig. S1A,B in the supplementary material). In addition, at all stages, brk was expressed laterally and ventrally, where it overlapped with wgexpression (Fig. 1B; see Fig. S1B in the supplementary material). In late third instar discs, brkexpression was partially downregulated in ventral cells, which also have weak P-Mad staining at this stage (Fig. 1B). Such a pattern suggests that Dpp signaling represses brk in the leg. To test this, we generated clones expressing a constitutively activated form of the Dpp receptor thickveins(tkv^QD) and also examined Mad loss-of-function clones, in which the Dpp pathway was inactive. Expression of Tkv^QDrepressed brk cell autonomously, whereas brk was derepressed in Mad^- clones (see Fig. S2A,B in the supplementary material). Thus, as in the wing, brk is negatively regulated by Dpp signaling in the leg.

To specify the DV axis of the leg, Dpp and Wg regulate distinct sets of target genes that pattern the dorsal and ventral halves of the disc,respectively. For example, Dpp signaling represses wg and activates omb, whereas Wg signaling represses dpp and activates H15 (Fig. 1A)(Brook and Cohen, 1996; Couso et al., 1993; Hays et al., 1999; Johnston and Schubiger, 1996; Morimura et al., 1996; Penton and Hoffmann, 1996; Theisen et al., 1996; Wilder and Perrimon, 1995). As described above, Dpp also negatively regulates brk in dorsal cells. As Brk is a transcriptional repressor in the Dpp pathway, we tested the possibility that brk contributes to ventral fate specification by repressing dorsal genes such as dpp and omb. To address this issue, we generated mitotic clones of a null allele of brk and examined the expression of dpp, wg, omb and H15. Surprisingly, all four of these genes were expressed normally in brk^- clones, no matter where they arose in the disc(Fig. 1C-F). Most importantly,ventral brk^- clones still expressed wg and H15, and did not derepress dpp or omb. By contrast, Tkv^QD-expressing clones, in which the Dpp pathway was fully active, repressed both H15 and wg (see Fig. S4A,B in the supplementary material). Leg discs entirely mutant for a hypomorphic allele of brk also did not show any ventral expansion of P-Mad staining or wg repression (see Fig. S3C,D in the supplementary material). These results strongly suggest that brk is not playing a role in DV axis specification in the leg. Notably, these findings contrast with those of Theisen et al. (Theisen et al., 2007) who suggested that brk is a repressor of dpp in the ventral leg disc. However, our genetic tests of this model, which were not carried out by Theisen et al., suggest that this model cannot be correct, at least in its most simple form.

The conclusion that brk does not play a role in DV axis specification in the leg is further supported by the phenotype of brk^- clones in the adult appendage. Although brk^- clones can be recovered anywhere in the adult leg,when they arose in the distal leg (tarsus and distal tibia) they either had no phenotype (see Fig. S4E in the supplementary material) or only subtly changed the bristle pattern (proximal tibia and distal femur, see below). By contrast,when they arose in the ventral or lateral proximal leg (proximal femur,trochanter or coxa) these clones generated leg-like outgrowths(Fig. 2A). These outgrowths(n=15) were strictly cell-autonomous and had no leg joints or distal leg structures, such as tarsi or claws. An analysis of the bristle pattern in these outgrowths suggests that they were composed of a single DV identity that is most similar to the ventral or lateral tibia. Consistent with the expression pattern of brk, clones in the dorsal proximal leg were wild type. Moreover, consistent with the marker analysis in leg discs (see above), brk^- clones behave very differently from clones expressing Tkv^QD, which cause a clear transformation to a dorsal identity in both proximal and distal regions of the leg (see Fig. S4C,D in the supplementary material). Thus, based on these molecular and morphological readouts, we conclude that brk does not play a role in DV axis specification. However, as explored below, these results suggest that brk is playing an important role in PD axis specification.

The PD axis duplications that result from removing brk suggest that Brk normally acts to repress Dll, dac and perhaps other PD genes. To address this issue we analyzed the expression of Dll and dac in brk^- clones. brk^-clones generated anywhere in the Dll or Dll+Dac domains did not affect the expression of these genes (Fig. 2B), suggesting that brk does not regulate these genes once their expression domains are established. This conclusion is consistent with the fact that by the third instar, both the Dll and Dac domains partially overlap with brk expression in the ventral disc.

In contrast to distal clones, all ventral or lateral clones generated outside (i.e. proximal to; n=25) the Dll and Dac domains activated both genes cell autonomously (Fig. 2B). As expected from its expression pattern, dorsal brk^- clones failed to derepress Dll or dac. In brk^- clones that derepressed these genes,the pattern of Dll and dac expression showed significant organization. Cells that had high levels of Dll generally had low levels of Dac and vice versa. However, despite this relationship, 100% of the clones(n=25) had cells that co-expressed both genes. By contrast, 52% of these clones had cells that expressed Dac, but not Dll, whereas only 4% had cells that expressed Dll, but not Dac. This pattern of Dll and Dac derepression is consistent with the observation that brk^-clones give rise to tibia-like outgrowths in the adult(Fig. 2A), as the tibia is derived from cells in the leg disc that express both Dll and Dac(Abu-Shaar and Mann, 1998; Lecuit and Cohen, 1997). brk^- clones in the Dac-only expression domain derepress Dll, accounting for the subtle change in bristle pattern observed in distal tibia and femur clones (see above).

Previous results suggest that repression by Brk uses two co-repressors,C-terminal binding protein (CtBP) and Groucho (Gro)(Hasson et al., 2001). Consistent with a role for these co-repressors in Dll and dac repression in the leg, clones mutant for gro or CtBP derepressed both genes in the same region of the leg disc(proximal ventral and lateral) as brk^- clones(Fig. 2C,D). In addition, like brk^- clones, gro^- or CtBP^- clones had no affect on Dll and dac expression in their normal domains and did not derepress these genes in dorsal clones. These results suggest that, as elsewhere in development, Brk uses these two co-repressors to repress Dll and dac in the leg disc. Interestingly, they also illustrate an inherent asymmetry to the way in which Dll and dac are regulated during leg development: while Dll and dac are repressed ventrally and laterally by Brk/Gro/CtBP, our data suggest that there is no Gro- or CtBP-dependent repressor that functions to keep these genes switched off in the dorsal leg disc. We will present additional evidence to support this idea.

Our finding that brk loss-of-function clones derepress Dll and dac in the ventral half of the leg disc(Fig. 2B) suggest two models for the role of Dpp signaling in PD gene activation. The first model suggests that Dpp signaling is only required to eliminate Brk repression. Alternatively, Dpp signal transduction may still be required to activate Dll and dac even in the absence of brk. This model follows from the observation that Brk can repress some Dpp target genes by competing with Mad for DNA binding(Kirkpatrick et al., 2001),raising the possibility that in the absence of Brk, Dll and dac may respond to lower levels of Dpp signaling. We carried out two experiments that both support the first model, namely that, in the absence of Brk, Dpp signaling is not essential for Dll and dacexpression.

In the first approach, we examined leg discs mutant for hypomorphic alleles of dpp and brk. In dpp^d12/d14(dpp^discs) leg discs, very little Dll and dac expression remained and the discs were very small(Diaz-Benjumea et al., 1994)(Fig. 3A). The remaining Dll and dac expression was derived from the trochanter region of the leg that is formed without Dpp input(Abu-Shaar and Mann, 1998; Diaz-Benjumea et al., 1994). Conversely, brk^XA hemizygotes had overgrown leg discs(Campbell and Tomlinson,1999). Notably, the expression of Dll and dacwas expanded ventrally and laterally in brk^XA discs compared with wild type (Fig. 3B). In addition, and consistent with the brk^-clonal analysis, wg expression was still limited to the ventral leg disc in brk^XA discs, as it is in wild-type discs (see Fig. S3B,D in the supplementary material). In brk^XA;dpp^discs double mutants, the shape and size of the leg discs resembled brk^XA leg discs. In this double mutant, no P-Mad staining was detected (see Fig. S3E in the supplementary material) and,consistently, wg was expressed along the entire AP compartment boundary, creating a leg disc with ventral-ventral symmetry(Fig. 3D). Surprisingly,although these discs exhibited no evidence of Dpp signaling, Dll and dac were still expressed, and their relative spatial domains were similar to wild type: Dll was expressed in the center of the disc, dac in a medial domain and there was a region in between where both genes were co-expressed (Fig. 3C, compare with Fig. 1A′). However, despite the overall integrity of the Dll and Dac domains, brk^XA; dpp^discs leg discs no longer had the normally large Dac-only domain in the dorsal region of the disc, and dac expression was expanded laterally(Fig. 3C). The Dlldomain was also expanded dorsally compared with wild type. These phenotypes are consistent with wg derepression in the dorsal leg disc, creating a ventral-ventral symmetric appendage.

Other genes required for PD axis formation are the EGFR pathway target genes al and Lim, which are both expressed, among other places, in the center of the leg disc (the future pretarsus, Fig. 3F)(Campbell, 2002; Galindo et al., 2002). In brk^XA; dpp^discs leg discs, the distal expression of al and Lim was absent whereas the other, more proximal, expression domains remained (Fig. 3E and data not shown). Thus, despite the expression of Dll and dac, the PD axis formed in brk^XA;dpp^discs leg discs was incomplete.

A potential caveat in the brk^XA; dpp^discsexperiments is that both of these alleles are hypomorphs, raising the possibility that sufficient Dpp activity exists to activate Dll and dac when Brk levels are compromised. To address this possibility, we generated clones doubly mutant for a null allele of brk and a null allele of tkv (Fig. 4A-D) or a strong hypomorphic allele of Mad (data not shown). Dll and dac were derepressed in both brk^-; tkv^--null mutant clones and brk^- Mad^- clones(Fig. 4A-D and data not shown). As brk^-; tkv^- clones cannot transduce the Dpp signal, these findings demonstrate that in the absence of Brk, Dpp signal transduction is not essential for Dll and dac expression. As with brk^- clones, when brk^-;tkv^- clones have had enough time to grow there is significant organization to Dll and dac expression. This organization cannot be dependent on Dpp signaling but, instead, as shown below, is probably due to Wg input.

We have shown that in the absence of Brk, Dll and dacbecome derepressed, even in the absence of Dpp signaling. These findings raise the issue of what activates Dll and dac in the brk^-; tkv^- clones, a situation where Dpp signal transduction cannot occur. One likely candidate is Wg, which is required for the initial activation of Dll and dac(Diaz-Benjumea et al., 1994). Consistent with this idea, only a subset of brk^- (or brk^-; tkv^-) clones derepress Dll and dac; dorsal or dorsolateral clones, which are far form the Wg source,fail to activate Dll or dac.

We carried out two experiments that provide more direct evidence that Wg signaling is required to activate these two genes. In the first approach, we examined the consequences of reducing both Wg signaling and Brk levels. As reported above, brk^XA larvae had overgrown leg discs with large Dll and Dac domains (Fig. 3B). If in this background we also reduced Wg signaling (i.e. in brk^XA; wg^CX3/CX4 larvae), the Dll and Dac domains were reduced in size (compared with brk^XA discs),suggesting that Wg signaling is playing a crucial role in the activation of these two genes (Fig. 5C;compare with Fig. 3B). Remarkably, brk^XA;wg^CX3/CX4 leg discs, as well as their Dll and Dac domains, were nearly circular; the Dac domain no longer had a dorsal bias that normally exists in wild-type leg discs (compare with Fig. 5A). In wg^CX3/CX4 hypomorphic larvae, in which Wg signaling was compromised but Brk levels were normal, the leg discs had a dorsal-dorsal symmetry, as evidenced both by the dac and brk expression patterns (Fig. 5B). This dorsal-dorsal symmetry probably occurs because dpp is derepressed ventrally in wg^CX3/CX4 leg discs(Brook and Cohen, 1996; Hays et al., 1999). In addition to providing evidence that Wg signaling is an activator of Dll and dac, these results suggest that the ratio of Wg signaling to Brk levels is important for controlling the shape and size of the Dll and Dac domains and, ultimately, the leg disc (see Discussion).

In a second experiment, we blocked Wg signaling in brk^-clones by expressing the Wg pathway inhibitor, Axin, using the MARCM method(Lee and Luo, 1999; Willert et al., 1999). In all brk^-; tub>axin+ clones, neither Dllor dac were activated, even in ventral positions of the leg disc(Fig. 5D). Taken together,these results strongly support the idea that Wg signaling is an essential activator of Dll and dac.

Because Wg is a diffusible morphogen, we also tested the idea that high levels of Wg signaling would be required to activate Dll, whereas lower levels would be sufficient for dac activation. Our previous experiments demonstrate that both genes are activated in ventral or ventrolateral brk^- clones(Fig. 2B) and that neither is activated in brk^-; tub>axin+ clones(Fig. 5D). We tested whether intermediate levels of Wg signaling were sufficient for dac (but not Dll) activation by expressing lower levels of Axin in brk^- clones, by growing the larvae at 17°C instead of 25°C (see Materials and methods). Strikingly, brk^-;tub>axin+ clones generated in larvae grown at 17°C often derepressed dac, but not Dll(Fig. 5E), a phenotype that was never observed in brk^- clones. These results suggest that during normal leg development, different levels of Wg signaling help to pattern the PD axis by activating different target genes; high Wg levels activate Dll, moderate levels activate dac but are insufficient for Dll activation, and low or no Wg signaling fails to activate either target.

Like Wg, Dpp is a secreted morphogen that can produce different effects depending on its concentration. Although our previous results show that Dpp signaling is not required for Dll and dac activation in the absence of Brk, they do not address the possibility that different levels of Dpp signaling contribute to target gene choice during wild-type development. According to this idea, Dll and dac may be differentially sensitive to different levels of Dpp and, consequently, Brk. To test this idea, we resupplied different levels of Brk in brk^- null clones and examined Dll and dac expression (see Materials and methods). As shown previously, ventral brk^- clones derepressed both Dll and dac(Fig. 2B). When grown at 25°C, brk^-; tub>brk+ clones failed to derepress either gene (Fig. 6A). However,when grown at 17°C, ventral brk^-; tub>brk+ clones derepressed dac but not Dll(Fig. 6B). These results suggest that, as with Wg, Brk levels, which are governed by Dpp levels, help to establish the Dll and dac expression domains along the PD axis of the leg.

The above genetic manipulations demonstrate that the levels of both Brk and Wg signaling determine whether dac or Dll are activated in the leg disc. To place these observations in the context of wild-type leg development, we examined the temporal sequence of Dll and dac activation during larval development, and compared it with the Wg, Brk and P-Mad patterns. In the mid-second instar (∼60 hours AEL),P-Mad, a readout of Dpp signaling, was broadly observed throughout the dorsal half of leg discs (see Fig. S1A in the supplementary material). At this same stage, brk was expressed ventrally, where there is no detectable P-Mad, and Wg staining was also limited to ventral cells(Fig. 7A; see Fig. S1A in the supplementary material). At this stage, Dll was expressed in the center of the disc, in cells that (1) had no detectable Brk, (2) stained for P-Mad and (3) had detectable Wg staining. dac was not expressed at this stage (Abu-Shaar and Mann,1998). Slightly later in development, in early third instar discs(∼72h AEL), the brk expression domain was larger and P-Mad staining was more restricted to a subset of dorsal cells (see Fig. S1B in the supplementary material). At about this time, dac was first observed in dorsal cells where there was no detectable Brk and in a small number of lateral and ventral cells that had low Brk levels(Fig. 7B). These early Dac+cells also had low levels of Dll. In older leg discs (∼108 hours AEL) the expression domains of Dll and dac overlapped with brk (Fig. 7C),consistent with our results showing that Brk does not play a role in Dll or dac expression once their domains have been established.

Although the evolutionary history of chordate and arthropod appendages is distinct, numerous similarities exist in the underlying genetic pathways that control appendage formation in these two phyla, raising the possibility that a common ancestor had appendage-like outgrowths that used many of the same mechanisms that currently operate in both phyla (reviewed by Shubin et al., 1997). As all appendages share the property of having PD axis, an understanding of the mechanisms governing the formation of this ancient axis may reveal additional common themes that operate in all animal appendages. Moreover, as dorsal appendages, such as wings, are thought to have been derived from ventral appendages (i.e. legs), the analysis of leg development is more likely to reveal such commonalities (reviewed by Shubin et al., 1997).

In Drosophila, central to the development of this appendage are two signaling pathways, Wg and Dpp, that are used repeatedly in animal development. In the leg, these signaling pathways serve at least two purposes:to specify positional values along the DV axis and to establish the PD axis. Our experiments reveal that the logic by which Dpp input is used to generate these two axes is distinct: for Dll and dac activation, the primary targets of the PD axis of the leg, Dpp input is mediated via Brk. By contrast, Dpp uses Brk-independent mechanisms to specify the DV axis. Also unclear from previous studies is how Wg and Dpp induce Dll and dac at distinct positions along the PD axis. In particular, how do Wg and Dpp, which are present in ventral and dorsal gradients, respectively,create concentric rings of PD fates? Finally, how are Wg and Dpp inputs integrated by Dll and dac? As discussed below, our results provide answers to these questions and allow us to present a revised model for PD axis formation in the Drosophila leg.

In the leg, as in other tissues, brk is expressed in a pattern that is complementary to the pattern of Dpp pathway activation, as revealed by P-Mad levels. Brk, a transcriptional repressor, could in principle be responsible for repressing Dpp targets relevant to both the DV and PD axes. Surprisingly, our experiments do not support a role for brk in DV axis specification. For all four DV readouts examined (dpp, wg, omband H15), expression was normal in brk^- clones. By contrast, two of the primary PD markers, dac and Dll, are repressed by Brk early in leg development. Thus, Dpp signaling controls gene expression in the DV and PD pathways in different ways: Dpp signaling both activates (e.g. omb) and represses (e.g. wg) DV genes independently of brk, perhaps via the Mad-Med and Schnurri transcription factors (Muller et al.,2003). By contrast, Dpp regulates Dll and dac in a brk-dependent manner. We suggest that these different modes of gene regulation by Dpp signaling are necessary for Dpp to execute these two distinct functions during leg development.

In contrast to the Dpp pathway, our evidence suggests that Wg signaling does not go through a transcriptional repressor to control Dll and dac expression. From previous results, the most likely candidate for such a Wg pathway repressor is Tcf, the downstream transcription factor in the Wg pathway. In the absence of Wg, Tcf has been shown to repress Wg target genes in a gro- and CtBP-dependent manner(Brannon et al., 1999; Brantjes et al., 2001; Cavallo et al., 1998; Lawrence et al., 2000; Roose et al., 1998; Valenta et al., 2003). In the leg, such a Wg pathway repressor would be predicted to be active in the dorsal leg disc, away from the source of Wg. An argument against this scenario,however, is our observation that gro^- or CtBP^- clones fail to derepress Dll or dac in the dorsal leg disc, but derepress these genes in the ventral leg disc. We also examined the consequences of knocking down Tcffunction in clones by RNAi. Although Tcf-RNAi-expressing clones were able to block Wg repression of dpp in the ventral leg disc, no derepression of Dll or dac was observed in dorsal regions of the disc (data not shown). Taken together, these results suggest that whereas Dpp signaling controls Dll and dac by repressing the Brk repressor, Wg signaling controls Dll and dac positively,without going through an intermediate repressor.

During leg development, Dll is activated in the center of the disc whereas dac is activated in medial regions of the disc. Previous genetic studies suggested that the activation of these two outputs depends on different levels of Wg and Dpp(Diaz-Benjumea et al., 1994; Lecuit and Cohen, 1997). Our results support this view, and provide additional insights into how Dll and dac are activated by these two secreted signals during leg development. By independently manipulating the levels of Brk or Wg signaling, we found that both of these regulators impact whether Dllor dac is activated. Specifically, we created situations in which both genes have the potential to be expressed (i.e. in ventral brk^- clones), but at the same time varied the amount of Wg signal transduction or Brk. We found that Dll can be activated only when the Wg pathway is fully active. By contrast, dac can be activated when the Wg pathway is only partially active. Analogously, when we expressed different levels of Brk in brk^- clones, we found that Dll could only be activated in the absence of Brk, but that dac could be activated in the presence of intermediate Brk levels. As Brk levels normally vary in response to Dpp activity, these data demonstrate that the levels of both Wg and Dpp inputs play a role in choosing which PD target gene is activated during wild-type leg development(Fig. 7D).

Although, the requirement for multiple inputs into gene regulation is typical in transcriptional regulation, for most cases where multiple positive inputs are required, the underlying mechanism is not well understood (reviewed by Arnosti, 2003; Barolo and Posakony, 2002; Guss et al., 2001; Mann and Carroll, 2002; Merika and Thanos, 2001). By contrast, there are many examples of genes that integrate both positive and negative inputs to be activated in a spatially or temporally restricted manner(Arnosti, 2003). The enhancers mediating this control can be thought of as `logic integrators' that are active only in the presence of the correct activators and in the absence of repressors (Istrail and Davidson,2005). We suggest that such a mechanism can account for why Dpp and Wg are both required for the activation of Dll and dacduring leg development. Although both pathways are required for activating these genes, our finding that Dpp functions by repressing a repressor creates a situation in which positive and negative inputs ultimately determine where Dll and dac are activated. For Dll, we suggest that activation occurs only in cells that satisfy two conditions: (1) they have little or no brk expression owing to high Dpp signaling and (2) they are experiencing a strong Wg input. These conditions are satisfied in the center of the young leg disc, where Wg and Dpp signals coincide, precisely where Dll is expressed (Fig. 7A,D). Importantly, Brk plays a crucial role in this model because it keeps Dll switched off in a large part of the disc, thus allowing Wg-mediated activation of Dll to occur in only a small subset of the disc. Consistently, a Dll enhancer that activates expression in the young leg disc directly integrates Wg and Dpp inputs by binding Tcf, Mad and Brk (Estella et al., 2008). In addition to these inputs, there must be additional factors that limit Dll activation to ventral, but not dorsal (e.g. wing), imaginal discs. Ventral specificity may be mediated by the expression of Dll,itself, via an early embryonic enhancer or by the activity of other ventral-specific factors such as buttonhead (btd), which is required for Dll expression(Castelli-Gair and Akam, 1995; Estella et al., 2003; Vachon et al., 1992).

The activation of dac appears to be more complex, but we speculate works using similar logic as that used for Dll activation. When dac is first activated in the wild-type leg disc, it is turned on in a relatively large number of cells dorsal to the Dll domain, but also in a smaller number cells lateral and ventral to the Dll domain. Based on their position in the leg disc, it is likely that cells in each of these regions have different levels of Brk and Wg signaling at the time when dac is first activated. Cells in the dorsal dac domain have no (or very low levels of) Brk and intermediate levels of Wg signaling,whereas cells ventral to the Dll domain have intermediate levels of Brk and high levels of Wg signaling (Fig. 7B,D). We suggest that both of these ratios of Brk to Wg (Brk:Wg)inputs are competent to activate dac. Furthermore, we speculate that these Brk:Wg inputs are not sufficient to activate Dll. We suggest that an analogy can be made between how Dll and dac are activated in the leg disc, to how even-skipped (eve) is activated during embryogenesis. As we suggest for Dll and dac, the activation of eve stripe 2 integrates both repressors and activators (Small et al.,1992). Moreover, for eve stripe 2 activation, the ratio of activation:repression is more important than the absolute amount of activation (Arnosti et al.,1996). By analogy, dac may be able to be activated by a wide range of Wg levels, as long as there is a compensatory change in the levels of Brk.

Another previously unexplained aspect to leg development is that dac expression, both when it is first activated and throughout all of development, is asymmetric: there are many more cells expressing dacdorsal to the Dll domain than there are lateral or ventral to the Dll domain. If dac were simply activated by lower levels of Dpp and Wg, it is difficult to account for this asymmetry. We suggest that the activation of dac by specific ratios of Brk:Wg, together with the asymmetric expression of brk in the leg, accounts for this asymmetry. The absence of brk expression in the dorsal leg disc allows those cells to be more sensitive to activation by Wg signaling. By contrast, lateral or ventral cells express brk, making them less responsive to Wg input. This idea is supported by our experiments in which we varied the Brk:Wg ratio. Most informative are the brk^XA;wg^CX3/CX4 leg discs, which have nearly symmetrical and circular dac and Dll expression domains(Fig. 5C). We suggest that these phenotypes result from a reduction in both Wg signaling and Brk levels. Less Wg results in less dac activation in dorsal cells, but the Wg that remains is more effective at activating dac ventrally and laterally owing to lower Brk levels. Thus, we suggest that the shape of the dac expression domain, and ultimately the shape of the adult appendage, is governed by the ratio of Brk:Wg and the responses to these inputs by key genes such as Dll and dac.

Prior to this study, the model for PD axis formation in the leg failed to explain how the activities of Wg and Dpp, present in ventral and dorsal gradients, respectively, are integrated to create the PD axis. In light of our data, we suggest the following steps in PD axis formation(Fig. 7D). First, we suggest that Dll is activated only in cells that are receiving high Dpp and Wg signals, because it is kept off in most of the disc due to repression by brk. Based on previous results(Diaz-Benjumea et al., 1994; Lecuit and Cohen, 1997), we also posit that high levels of both signals repress dac, keeping this gene switched off in cells that activate Dll. As the disc grows, we suggest that dac is activated in cells that have the appropriate ratios of Brk:Wg inputs, as outlined above, and that Dll cannot be activated by these input ratios.

Once the initial dac and Dll domains are initially defined, we further suggest that they are locked into place and maintained by a mechanism that no longer requires Wg and Dpp inputs. This idea is consistent with previous genetic studies showing that the requirement of Dll and dac for Wg and Dpp inputs is transient(Galindo et al., 2002; Lecuit and Cohen, 1997). Once activated, we suggest that an autoregulatory and/or Polycomb-dependent mechanism can create the three main domains that exist along the mature PD axis: a distal Dll-only domain, a medial Dll+Dac domain and more proximal Dac-only domain. The Dll-only and Dac-only domains are simplest to understand,as they are derived from cells that were only able to activate one of these two genes because of the Brk:Wg input they received. We suggest that the domain that co-expresses Dll and dac is derived from cells present at the interface between the initial dac and Dllactivation domains, which have, by virtue of their position in the leg disc,the potential to activate both genes. Once both Dll and dacare activated, even in the same cell, we imagine that they, too, become locked into an expressed state by an autoregulatory and/or Polycomb-dependent mechanism. A recent analysis of the cis-regulatory elements controlling Dll expression in the Drosophila leg is consistent with this model (Estella et al., 2008). In particular, Dll expression in the leg disc uses two cis-regulatory elements. One directly receives input from Wg and Dpp and, consequently, is only active in the center of the leg disc where both signals meet. The second element maintains this initial expression at least in part by via an autoregulatory mechanism, by directly binding Dll(Estella et al., 2008). In the future, it will be important to further test the model proposed here by analyzing the regulatory elements controlling dac during leg development.

We thank Gerarld Campbell, Gary Struhl, Konrad Basler, Miriam Zecca, Ze'ev Paroush, Daniel McKay and Andrew Tomlinson for reagents, and Gary Struhl,Andrew Tomlinson, Laura Johnston, Gines Morata, Daniel McKay, Matt Giorganni,Francisco A. Martin and Hector Herranz for comments on the manuscript. This work was supported by NIH and March of Dimes grants to R.S.M., and by funding from the Spanish Ministry to C.E.