Peak levels of BMP in the Drosophila embryo control target genes by a feed-forward mechanism

A primary mechanism for establishing cell fates during animal development is to use morphogen gradients. A single gradient formed across a broad field of cells can determine a number of different cell types by generating thresholds for downstream target gene responses. For example, it is thought that the peak level of a gradient regulates a specific set of target genes,whereas lower levels regulate an additional set of genes.

In Drosophila, patterning of the embryonic dorsoventral axis depends on the combined action of two morphogens: Dorsal (Dl) and Decapentaplegic (Dpp). Dl is a maternally loaded transcription factor that is responsible for setting up the overall dorsoventral axis of the embryo(reviewed by Stathopoulos and Levine,2002). A gradient of Dl protein is formed during early embryogenesis with peak levels in the ventral nuclei. The gradient acts to subdivide the axis into three main regions – ventral (mesoderm), lateral(neuroectoderm) and dorsal ectoderm – by eliciting several threshold responses from batteries of zygotic patterning genes. For example,transcriptional activation of twist (twi) and snail(sna) require high levels of Dl, while short gastrulation(sog), brinker (brk) and rhomboid(rho) can be activated by lower levels of Dl. Genes such as zerknüllt (zen), decapentaplegic(dpp) and tolloid (tld) are repressed by Dl and thus come to be expressed only in the dorsal region. Differential target gene responses are mediated largely by the affinity of Dl-binding sites in the target enhancers (Jiang and Levine,1993).

Dpp acts to further subdivide the dorsal domain into amnioserosa (the dorsal most region) and dorsal ectoderm, while also inhibiting neuroectoderm formation (Ferguson and Anderson,1992a; Wharton et. al.,1993; Biehs et al.,1996). Dpp is a member of the TGFβ superfamily of ligands,which are most closely related to the BMPs (bone morphogenetic proteins), and signals through a pathway comprising the type I and type II serine-threonine kinase transmembrane receptors and the intracellular Smad proteins, Mother against Dpp (Mad) and Medea (Med) (reviewed by Raftery and Sutherland, 2003). Upon ligand binding and receptor activation, Mad is phosphorylated (PMad),thereby allowing translocation into the nucleus along with the co-Smad Medea. In the nucleus, Smads function as DNA-binding transcription factors (reviewed by Shi and Massagué,2003).

Although dpp RNAs are evenly distributed across the dorsal region of the precellular embryo, the Dpp activity gradient takes shape during stage 5, as the embryo is undergoing cellularization (reviewed by Raftery and Sutherland, 2003). The gradient, which also includes a second BMP ligand Screw (Scw)(Arora et al., 1994), is formed through a dynamic process involving the secreted protein Sog, which emanates from the adjacent ventral region(Srinivasan et al., 2002). As Sog diffuses dorsally, it sequesters BMPs with the help of another secreted protein Twisted Gastrulation (Tsg) (Mason et al., 1994; Ross et al.,2001). This tripartite complex is thought to serve two purposes that have opposite effects on BMP signaling. First, it antagonizes BMP signaling by preventing BMP ligands from interacting with their receptors; and second, it promotes signaling by allowing the redistribution of BMPs to more dorsal regions (Holley et al.,1995; Shimmi and O'Connor,2003). Tolloid (Tld), a metalloprotease localized in the dorsal region (Shimell et al., 1991),cleaves Sog, thereby releasing BMPs as active ligands. The net effect is a BMP gradient that can be visualized by detecting the output of the BMP pathway,the nuclear Smad proteins (reviewed by Raftery and Sutherland, 2003). Initially PMad is in a broad gradient containing relatively low levels of protein. This develops into a steep step-wise gradient with increasingly high levels in a five- to six-cell-wide stripe along the dorsal midline, the presumptive amnioserosa, and lower levels in the three or four cells adjacent to either side of the stripe. In more lateral regions of the dorsal ectoderm,PMad protein is not detectable by antibody staining.

There are several candidate BMP target genes whose expression domains correlate with the stepwise PMad gradient. For example, hindsight(hnt; peb – FlyBase)(Yip et al., 1997) and Race (related to angiotensin-converting enzyme; Ance –FlyBase) (Tatei et al., 1995)are expressed specifically in the peak level PMad domain. u-shaped(ush) (Cubadda et al.,1997; Jazwinska et al.,1999b), tail-up (tup)(Thor and Thomas, 1997; Ashe et al., 2000) and rhomboid (rho) (Bier et al., 1990; Ross et al.,2001) are expressed more broadly in 12-14 cells encompassing the adjacent lower level PMad domain. pannier (pnr)(Ramain et al., 1993; Winick et al., 1993) is expressed in a broad domain covering about 36 cells (or 25% of the dorsal-ventral circumference), the border of which does not correlate with a clear domain of PMad activity. The mechanism underlying threshold responses to the BMP/Smad gradient is not fully understood.

A possible mechanism emerged from studies on Brk, a transcriptional repressor (Campbell and Tomlinson,1999; Jazwinska et al.,1999a; Jazwinska et al.,1999b; Minami et al.,1999). As Dpp signaling represses brk expression, brk domains are largely complementary to dpp domains, which allows target genes to be transcribed in the dpp domains. In areas where dpp and brk overlap slightly, Smads and Brk may compete for DNA binding on target enhancers as Brk sites often overlap Smad sites (Kirkpatrick et al.,2001; Rushlow et al.,2001; Saller and Bienz,2001). It has therefore been suggested that a Brk gradient,inverse to the Smad gradient, acts to spatially restrict target gene activation and consequently sets borders of expression(Jazwinska et al., 1999a; Ashe et al., 2000; Müller et al., 2003). However, this mechanism does not explain the threshold responses of those genes that are not Brk targets, such as Race. Alternatively, a mechanism involving differential binding site affinity for Smads may play a role in establishing their borders of expression. Indeed Wharton et al.(Wharton et al., 2004) used modified Race enhancers to demonstrate that it is possible to broaden expression domains by increasing the affinity of Smad-binding sites.

We demonstrate that Race activation requires the combined input of Smads and Zen. Zen-binding sites lie adjacent to Smad sites in the Race enhancer, and we show that Smads facilitate the DNA binding of Zen and that this requires protein interaction between Smads and Zen. As zen is regulated by peak levels of the BMP gradient and thus becomes expressed only in the dorsalmost region(Rushlow et al., 2001), the regulation of Race resembles a feed-forward loop whereby one regulator, BMP/Smad, controls a second regulator, Zen, and then both bind and activate a common target gene, Race. In addition, we tested the respective roles of Zen and Smads in setting the expression borders of Race. When zen was expressed ectopically, Raceexpression broadened to encompass both the high and low level regions of the BMP gradient. Thus, as long as Zen is present, low levels of BMP activity are sufficient to activate Race, indicating that the purpose of the peak of the BMP gradient is to set the Zen domain.

dpp^hr4 is a weak hypomorphic allele and dpp^H46 is a null allele(Wharton et al., 1993). dpp^hr4 was balanced over SM6 eve-lacZ, while dpp^H46 was balanced over CyO23,P[dpp⁺], a chromosome that contains two copies of dpp(Wharton et al., 1993). 4X dpp embryos are of the genotype: CyO23,P[dpp⁺]/CyO23, P[dpp⁺] and were derived from the dpp^H46/CyO23, P[dpp⁺] stock. sog^SY06 is a strong hypomorphic sog allele(Ferguson and Anderson, 1992b)balanced over FM7c, ftz-lacZ. zen^w36 is a null allele(Rushlow et al., 1987a)balanced over TM3, ftz-lacZ or TM3, hb-lacZ. 4X dpp; zen^- embryos were derived from the stock CyO23, P[dpp⁺]/+; zen^w36/TM3,hb-lacZ (1/16 of the embryos). The double heterozygous embryos, dpp^hr4/+; zen^w36/+,were identified by the lack of lacZ staining in embryos derived from a cross of the dpp^hr4 and zen^w36stocks (1/4 of the embryos).

The zen cDNA (+10 to +1234 from the transcription starting site)(Rushlow et al., 1987a) was cloned into pUAST (Brand and Perrimon,1993) via EcoRI and XbaI sites on the 5′and 3′ ends, respectively. The zen-Del cDNA was made by PCR mutagenesis (Expand High Fidelity PCR system, Roche Applied Science) using oligos spanning the deletion region (amino acids 152-198) and cloned into pUAST via the EcoRI and XbaI sites. Transgenic flies were generated by the standard transformation protocol(Spradling and Rubin, 1982). Flies carrying UAS-zen and UAS-zen-Del were crossed to stripe-2 eve-Gal4 drivers (gift from S. Small) and the expression of ectopic zen or zen-Del proteins was confirmed by staining with anti-Zen antibodies. Guinea pig or rabbit anti-Zen antibodies were generated (Covance) as described by Rushlow et al.(Rushlow et al., 1987b). To obtain uniform early embryonic expression of the UAS-zen and UAS-zen-Del transgenes, a maternal Gal4 driver was used in which the GAL4-VP16 fusion protein is expressed maternally under the control of the α–tubulin 67C promoter. These were further crossed into a zen^w36/TM3, hb-lacZ background for the zen mutant rescue experiments.

The Race 533 bp enhancer DNA was kindly provided by M. Levine(Rusch and Levine, 1997). Deletions and point mutations were created by PCR mutagenesis. An internal deletion of 66 bp (nucleotides 432-497 of the Race enhancer) deletes most of the Mad-binding region. The two proximal Zen-binding sites were mutated as follows: ATATTAAT was changed to ATCTAGAT, and ATTAAAAATAAATAAT was changed to TAGAAAAATAACTGCA. Race-lacZ constructs were prepared by subcloning the wild-type and mutated versions of the Race enhancer into a modified Casper transformation vector that contains the minimal promoter sequence from the even-skipped gene fused to the lacZ reporter gene (eve-lacZ Casper)(Small et al., 1992). At least three transformant lines for each construct were tested.

Wild-type, mutant and transgenic embryos were fixed, hybridized with zen, Race, hnt or lacZ antisense RNA probes, and stained(Roche Molecular Biochemicals), dehydrated and mounted in araldite(Polysciences) as described by Rushlow et al.(Rushlow et al., 2001). Anti-PMad polyclonal antibodies were kindly provided by P. ten Dijke(Persson et. al., 1998) and used at a final dilution of 1:1000 in PBS. Secondary anti-rabbit antibody staining was performed using the Vectastain ABC kit (Vector Labs). Embryo preparations were photographed using DIC optics on a Nikon FX-A microscope.

The GST-Zen and GST-Zen-Del fusion constructs were cloned by introduction of EcoRI sites at the initial ATG and at the 15th nucleotide downstream of the stop TAA codon in the zen cDNA by PCR mutagenesis,followed by excision of the EcoRI fragment from the PCR product and ligation into the EcoRI site of the pGEX-4T-2 vector (Pharmacia). Expression plasmids encoding GST-Mad^N and GST-Medea fusion protein containing the N-terminal MH1 domains were obtained from A. Laughon(Kim et al., 1997) and M. Frasch (Xu et al., 1998),respectively. The expression and the purification of the recombinant proteins were carried out as described before(Kirov et al., 1993). The concentration of the isolated proteins was determined by SDS-PAGE after staining with Coomassie R-250, together with defined amounts of bovine serum albumin.

DNAse I footprint analyses were carried out as previously described(Kirov et al., 1993). The 533 bp Race enhancer was originally cloned into the pBluescript KS+vector (Stratagene) NotI site(Rusch and Levine, 1997). Three fragments were generated for footprint analysis using the BssHI and XhoI sites on each side of the vector polylinker: BssHI-ApoI (–50 to 163), HinfII-TthIII (107 to 349) and TthIII-XhoI(349-603). The electrophoretic mobility shift assays were performed as described before (Kirov et al.,1993) except that the electrophoresis was run at room temperature. The sequence of the 42 bp oligonucleotide (Race sequences 486-527)spanning the wild-type Smad-binding sites and the proximal Zen-binding site is(consensus core sites are underlined):5′-TCAGACGCGACTAAGCCGATCTCGCATTAAAAATAAATAATG-3′. The Smad-binding site mutations are (mutated core sites are underlined):5′-TTAGATGCGAGTAAGATGATCTCGCATTAAAAATAAATAATG-3′. The Zen-binding site mutations are (mutated core sites are underlined):5′-TCAGACGCGACTAAGCCGATCTCTCTAGAAAAATAACTCGAG-3′. The sequence of the 128 bp DNA fragment is (consensus core sites are underlined; vector sequences are in lower case):5′gaattcgcccttAACGTCGGCTTATCTTCGCGCCTACCTGGCCGAGAACCCCAGACGGATTGGAAACATCAGACGCGACTAAGCCGATCTCGCATTAAAAATAAATAATGCTCGAGaagggcgaattc-3′. The end-labeled oligonucleotides or fragment, which was isolated from a subclone of the Race enhancer by EcoRI digestion, were added to the reactions containing Zen, Mad, Medea or combinations of different proteins and incubated on ice for 30 minutes before loading on the gel.

GST pull-down assays were carried out as previously described(Kirov et al., 1996). Wild-type (and mutant forms) of zen were cloned into the pAR vector(NdeI and EcoRI sites)(Rosenberg et al., 1987) by introduction of an NdeI site at the initial (or internal sites to make truncated proteins) and an EcoRI site four nucleotides downstream of the stop codon (or at internal sites to make truncated proteins). These constructs were expressed in vitro with the TNT Coupled Reticulocyte lysate system (Promega) using the T7 promoter in the pAR vector. Expression of these proteins was confirmed by electrophoresis on a 12% SDS polyacrylamide gel.

zen encodes a homeodomain protein of the Antennapedia class, and is initially expressed broadly like dpp but becomes restricted to the dorsal most region, the presumptive amnioserosa(Doyle et al., 1986). This refinement of the zen pattern is dependent on Dpp, and proceeds simultaneously with the refining PMad gradient(Rushlow et al., 2001). Thus,by mid-late stage 5, zen transcripts are present only where there are peak levels of PMad, in the dorsal most cells. At this time, Racetranscripts begin to accumulate in the same domain(Fig. 1A)(Tatei et al., 1995). Race activation is known to be dependent on dpp and zen as Race is not transcribed in either mutant(Tatei et al., 1995; Rusch and Levine, 1997). To better understand the mechanism underlying the requirement for both genes, we first examined Race expression in different genetic backgrounds that alter levels of dpp and zen, and then performed molecular analyses with Smad and Zen proteins to characterize protein-protein and protein-DNA interactions with the well-defined Race enhancer(Rusch and Levine, 1997; Wharton et al., 2004). We are particularly interested in how Dpp morphogenetic activity is interpreted by high level target genes so they come to be expressed only in the peak level domain: the presumptive amnioserosa.

dpp^hr4 is a weak dpp allele(Wharton et al., 1993) and sog^SY06 is a strong sog allele(Ferguson and Anderson,1992b). Mutant embryos of both genotypes fail to develop amnioserosa (Wharton et al.,1993; Zusman et al.,1988) presumably because of insufficient Smad signaling. In dpp^hr4 and sog^SY06 mutants, PMad does not accumulate in the dorsalmost five or six cells, but instead fades quickly in dpp^hr4embryos and is present in a relatively strong broad domain in sog^SY06 embryos(Rushlow et al., 2001). Correspondingly, the broad early zen pattern does not refine in either mutant, but follows closely PMad activity. The observed changes in the zen pattern reflects the failure of these embryos to generate a steep BMP gradient with the peak levels of PMad necessary for refined zenexpression (Rushlow et al.,2001). Race transcripts are absent in dpp^hr4 mutants (Fig. 1B). In sog^SY06 mutants, Raceexpression is observed in a broad spotty pattern in the middle body region that fades quickly by gastrulation (Fig. 1C). These results taken together indicate that high levels of both PMad and Zen, above the levels present in each mutant, are required for proper Race activation.

dpp^hr4/+ heterozygous embryos have normal Race expression (Fig. 1D). Similarly, zen^w36/+heterozygotes also have normal Race expression (data not shown). However, in double heterozygous dpp^hr4/+; zen^w36/+ embryos, Race expression is absent in the middle body region (Fig. 1E). zen expression appears normal in these embryos(Fig. 1F) as does PMad staining(data not shown). Thus, although the level of Dpp activity in these embryos is sufficient to drive refined zen expression, it is not able to compensate for the reduction in zen dose in order to properly activate Race. Likewise, there is not enough Zen to compensate for the reduction in Dpp activity. Again, we conclude that Race responds to high levels of combined Zen and PMad activities.

To further investigate the combinatorial requirement for dpp and zen, we performed additional dosage studies. We first examined Race in embryos with four copies of dpp (4X dpp). The expression domains of peak-level PMad, zen, and Race are broader in these embryos covering about 12-14 cells(Fig. 2D-F). This demonstrates a clear correlation between peak levels of PMad/Zen and high-level Dpp target gene expression. Next, we examined Race expression in embryos with 4X dpp that lack zen activity. Race transcripts are present, but only in the dorsalmost five or six cells(Fig. 2I), comparable with that in 2X dpp embryos (Fig. 2C), even though peak levels of PMad and refined zencover 12-14 cells in these embryos (Fig. 2G,H). Therefore, Dpp can activate Race in a zen-independent manner but only if expressed above wild-type levels,as was also concluded by Rusch and Levine(Rusch and Levine, 1997). However, Race is not activated across the entire domain of PMad, but only in the dorsal most region, suggesting that Dpp signaling is higher along the dorsal midline in a 4X dpp embryo than in a wild type 2X dpp embryo, though not obvious in our antibody staining experiment.

Is the role of Zen to potentiate Dpp activity by reducing the threshold of Dpp required for Race activation, or are Zen and Dpp activities interchangeable, one being able to replace the other if the total amount of activity reaches the threshold? To discern these two possibilities, we examined whether zen can activate Race in the absence of PMad by overexpressing zen ubiquitously throughout the embryo using a maternal-GAL4 driver and UASzen(Fig. 2K; see Materials and methods). The PMad staining pattern is normal in these embryos(Fig. 2J). However, the Race stripe is broader than normal, including both the high-level and low-level PMad domains (Fig. 2L), and is also observed in the posterior region (see arrow). Race is not detectable in more lateral regions where there is no detectable PMad. Therefore, zen cannot activate Race in the absence of PMad, even when overexpressed, suggesting that Raceresponds to a combination of activities provided by Zen and Dpp, but these activities are not exchangeable. Only when Zen was fused to a strong activation domain, VP16, did it become independent of Dpp in activating Race (Rusch and Levine,1997).

These results reveal a dual role for Dpp in Race activation. Peak levels of BMP/Smads define the domain of zen in the dorsalmost cells. Next, BMP/Smads, together with Zen, activate downstream targets such as Race; however, this function does not require peak-level Smads as Race was activated in regions of low-level Smads when Zen was ectopically expressed (Fig. 2L). Thus, we propose that the role of the peak of the BMP gradient is to set the zen domain. Hence, Zen defines the expression borders of the downstream high level targets (see Discussion). We sought to investigate the molecular mechanism underlying the combinatorial requirement of Zen and Smads for the activation of Race transcription.

We assayed for Mad and Medea binding to the Race enhancer, a 533 bp DNA fragment that lies 1.5 kb upstream of the Race transcription start site (Rusch and Levine,1997). When fused with a lacZ reporter gene, this DNA fragment drives a similar expression pattern to that of wild-type Race mRNA (Fig. 3B)(Rusch and Levine, 1997). We performed DNA-binding assays with recombinant GST fused Mad protein and GST fused Medea protein, both of which contain the DNA-binding MH1 domain and the linker region (Kim et al.,1997; Xu et al.,1998).

DNase I footprinting assays revealed a Mad-binding region in the Race enhancer (Fig. 3A, lanes 3-5). It spans 82 bp(Fig. 3A, nucleotides 421-502 in blue and purple below the footprint) and is highly GC rich, containing several GNCN motifs (boxes labeled a-g on sequence), a configuration that has been shown to mediate strong inducible reporter gene responses upon binding of Smad complexes in cell culture (Johnson et al., 1999). The Mad-binding region contains a strongly protected area (nucleotides 464-502) (see also Wharton et al., 2004) and a weakly protected area (nucleotide 421-463) demarcated by solid and hatched rectangles, respectively (see Fig. 3 legend).

A 66 bp deletion of this region (nucleotides 432-497) abolishes the in vitro footprint binding of Mad (data not shown), and also greatly reduces in vivo expression of the Race-enhancer reporter gene, Mdel-lacZ (Fig. 3C),suggesting that Mad can directly activate Race via binding to this region. The observed residual activity of this reporter might be due to the remaining Smad-binding site (box g). We also detected four Zen-binding sites in the Race enhancer, three of which are shown in the footprint in Fig. 3A (lanes 6-9; nucleotides in red). Point mutations in all four of the ATTA core sites abolished lacZ reporter expression (Rusch and Levine, 1997), as did mutations in the three most proximal sites (Fig. 3D; ZBS-lacZ).

The above data favor the existence of a relatively short regulatory module containing a cluster of Smad-binding sites bordered by Zen-binding sites that regulates the essential aspects of early Race expression. Similarly,short sequences containing multiple Smad sites and other transcription factor binding sites have been found to regulate the Dpp targets tin(Xu et al., 1998) and Ubx (Saller and Bienz,2001) in Drosophila.

The closely apposed Smad- and Zen-binding sites in the Raceregulatory module hinted at the possibility that their combined activity,which is essential for Race activation, might partially depend on their direct interaction. The most immediate result from such an interaction could be to facilitate the binding of one or the other protein to DNA. To test this, we tried to detect cooperative binding of Mad and Zen proteins to DNA. In preliminary experiments, we found the range of concentrations of Zen and Smad proteins that produce complexes with a 42 bp oligonucleotide spanning a cluster of three Smad-binding sites and the two most proximal Zen sites. Then by incubating one component with suboptimal amounts of the other, which by itself is not sufficient to produce complexes, we expected that the DNA binding of the protein of suboptimal concentration would increase if there is cooperative DNA binding, and/or possibly form supershift complexes containing both proteins.

Mad and Medea form similar complexes with the 42 bp oligonucleotide(Fig. 4B, lanes 2-3), and do not bind oligonucleotides in which the three Smad sites (CAGAC, GACT, GCCG)were mutated (TAGAT, GAGT, GATG respectively; lanes 6-7). In experiments with the larger 122 bp fragment that contains seven Smad sites(Fig. 4A), mutation of the two CAGAC sites (boxes d and e in Fig. 3A and Fig. 4A)prevented Medea binding (data not shown), indicating that Medea binds the CAGAC site, as was recently shown by Pyrowolakis et al.(Pyrowolakis et al., 2004). Zen forms two complexes with the 42 bp oligonucleotide that have different mobilities from the Smad complexes (Fig. 4B, lane 4), and does not bind to oligonucleotides in which the core sites (ATTA, TAAT) were mutated (TAGA, TCGA, respectively; Fig. 4B, lane 12). When a concentration of Mad that does not produce detectable Mad complexes was incubated with varying Zen protein concentrations and the122 bp DNA fragment,the intensity of the Zen complexes increased compared with when Mad was not added to the reactions (Fig. 4C, lanes 10-12 compared to lanes 6-8). Complexes that could be interpreted as Mad/Zen supershifts were not clearly visible as distinct complexes. Similar experiments using the DNA fragments with mutated Smad sites showed little if any enhancement of Zen binding by small amounts of Mad (data not shown), indicating that the observed cooperative binding depends on Mad interaction with DNA.

In the reverse experiment, incubating Zen protein at a concentration that does not produce detectable complexes with varying amounts of Mad, neither the formation of a new complex nor any increase in Mad complexes was observed(data not shown). We did, however, observe an enhancement of Zen binding in the lanes with low Zen and increasing Mad concentrations (data not shown),consistent with the results from the previous experiment.

To test for direct physical protein-protein interactions between Zen and Mad proteins, we performed GST pull-down assays with GST-Mad^N and in vitro translated full-length and truncated Zen proteins,(Fig. 5A,B). Full-length Zen clearly interacts with Mad (Fig. 5C, lanes 1-2). Our results confirm the recent report of the Mad-Zen interaction found in a genome-wide yeast two-hybrid screen(Giot et al., 2003). By testing a series of truncated Zen proteins(Fig. 5C, lanes 3-14), we mapped the domain of the Zen protein involved in the interaction with Mad to be within the 48 amino acids C-terminal to the homeodomain (amino acids 152-199). Removal of this region by an internal deletion(Fig. 5A, Zen-Del) also abolishes the ability of Zen to interact with GST-Mad^N(Fig. 5C, lanes 15-16). Though the recombinant Zen protein with this deletion, Zen-Del, binds DNA similarly to full-length Zen (Fig. 4D,lanes 6-9), the cooperative binding to the wild-type DNA fragment containing Mad- and Zen-binding sites is not observed(Fig. 4D, lanes 10-13).

To test the functional relevance of the Mad-Zen physical interaction, we ubiquitously expressed the Zen-Del protein in the early embryo. As mentioned before, UAS-zen driven by maternal-Gal4 can ectopically activate Race in the posterior region of the dorsal midline, as well as more laterally along the DV axis(Fig. 6A). By contrast, the deletion construct UAS-zen-Del fails to elicit any ectopic Race expression (Fig. 6B). Indeed, Race expression in maternal-Gal4/UAS-zen-Del embryos is indistinguishable from that in wild-type embryos, demonstrating that the Mad-Zen interaction is required for ectopic Race expression.

As a second test for functional relevance of the Mad-Zen interaction, we examined whether the loss of Race expression in zen mutants could be rescued by ubiquitously expressed Zen, and if so, was the Mad interaction domain required. Wild-type UAS-zen driven by the maternal-Gal4 driver was able to restore Race expression in zen^w36 null mutants(Fig. 6C). By contrast, UAS-zen-Del failed to rescue Race expression(Fig. 6D). This result suggests that the Mad-Zen protein interaction is necessary for endogenous Raceactivation. In summary, both in vitro and in vivo evidence suggests that a physical interaction between Mad and Zen is essential to the underlying mechanism involved in Race activation.

We have studied the role of the BMP morphogen in regulating high level target genes in the early Drosophila embryo. Strikingly, our results present evidence that the primary role, and possibly the only role, of the peak level of the BMP gradient is to activate zen. Once zenexpression becomes restricted to the presumptive amnioserosa in the cellular blastoderm embryo, Zen protein and the BMP signal-transducing Smads act in concert to activate the high level BMP target gene Race. We have shown that Zen and Smads bind directly to adjacent sites in a short regulatory module in the Race enhancer, and this interaction with DNA is essential for Race activation. We also found that Raceactivation depends on direct protein-protein interaction between Mad and Zen. Our results showing enhanced DNA binding of Zen to the Race enhancer in the presence of Mad provide a further explanation for this interaction, to ensure the synergistic action of Zen and Smads necessary to activate high level target genes in the presumptive amnioserosa.

Specific activation or repression of transcription by a combination of transcription factors is a common theme in the regulation of developmentally important genes (Howard and Davidson,2004). The results from our genetic analysis and the molecular dissection of the Race enhancer clearly show that Race is activated by the combined action of Smads and Zen. Although Smads can single handedly activate Race when overexpressed(Fig. 2I), under normal circumstances concurrent Zen activity is required. Why are both Smads and Zen necessary?

Zen may act to restrict target gene expression specifically to the presumptive amnioserosa. As the Dpp pathway is used repeatedly during development, other factors must function in combination with Dpp to ensure tissue specificity (see Affolter and Mann,2001; Reim et al.,2003). The ectopic expression studies support this idea. In normal embryos, Race is activated only in regions where there are peak levels of PMad and Zen. In embryos where Zen is ubiquitously expressed, Race can now be activated in regions where there are lower levels of PMad (Fig. 2L), indicating that high level PMad is not the determining factor for amnioserosa tissue specificity. Rather PMad allows expression, and Zen determines the border of expression. We interpret the overexpression studies where Dpp can activate Race alone (Fig. 2I)(Rusch and Levine, 1997) to be situations where there are such high levels of Smads that Race and hnt become activated promiscuously, and hence differential regulation is lost. In normal embryos, the combination of Smads and Zen ensures that the high level target genes are activated only in the presumptive amnioserosa.

On the other hand, why the need for Smads? One role for Smads is suggested from the observation that Smads facilitate the binding of Zen to the Race enhancer (Fig. 4). It is well established that Hox proteins often require co-factors for DNA binding to target enhancers (reviewed by Mann and Affolter, 1998). For example, composite sites that also bind the co-factor Extradenticle (Exd)ensure a greater selectivity for binding over the higher frequency Hox core site such as TAAT. In other examples, binding sites for signaling pathway effectors lie close to Hox/Selector-binding sites (see Guss et al., 2001; Affolter and Mann, 2001). The closely apposed Zen- and Smad-binding sites in the Race enhancer is one such scenario, as Zen can be thought of as a Selector gene(Rushlow and Levine, 1990). Our studies add to this idea of Smads and Selector cooperativity by demonstrating enhanced binding of Zen in the presence of Smads. Though the enhancement we observe in our in vitro assays is not dramatic, it is possible that in the embryo a moderate enhancement is functionally significant as is the twofold doubling of the dpp dose.

Another potential role of Smads was suggested from previous overexpression studies (Rusch and Levine,1997). Zen was only able to activate Race in the absence of Dpp if fused to a strong activation domain derived from VP16. This suggests that Smads provide a transactivation function different from that of Zen. The Smad MH2 domain has been shown to interact with the transcriptional co-activators CBP and p300 (Waltzer and Bienz, 1999; Shi and Massagué, 2003). Zen has not yet been analyzed for interaction with transcriptional co-activators, however, the activation domain of Zen lies within the C-terminal 119 amino acids(Han et al., 1989), and does not overlap with the homeodomain or the Mad interaction domain(Fig. 5). Mechanistically, the difference in the activation potential between Zen and Smads could be due to their ability to recruit different co-activators to the transcriptional machinery.

Gradients of morphogens provide positional information to the cells by activating different genes at different threshold concentrations. In early Drosophila embryos, the transcriptional threshold responses to the Bicoid (Driever et al., 1989)and Dl (reviewed by Stathopoulos and Levine, 2002) morphogens have been extensively studied. The major mechanisms by which thresholds are established exploits the DNA-binding affinities of Bcd and Dl to their operator sites, as well as synergistic interactions with other transcription factors bound to the cis-regulatory sequences.

The BMP morphogen gradient also elicits different threshold responses from its targets, and, as discussed above, a combinatorial mechanism is used to activate Race, a high level Dpp target. Our genetic results indicate that Race (Fig. 1),and also another high level target hnt (M.X., unpublished), are activated only when a specific threshold of Zen and Smad activities are reached. In sog mutant embryos, Zen and Smad concentrations are relatively high, though below peak levels(Rushlow et al., 2001), and there is just enough of their combined activity to weakly activate Race (Fig. 1C) and hnt (data not shown). By contrast, in the double heterozygous embryos dpp^hr4/+; zen^w36/+,Race is not activated (Fig. 1E) because Zen and Smad concentrations are below the threshold levels required for activation.

A simple way to explain these results is if the Race enhancer has low affinity to Zen and Smad proteins in vivo. To transcribe Raceeffectively would then require relatively high concentrations of the proteins,which are indeed reached in the dorsalmost cells. It has been known for some time that the enhancers of the high level Dl targets contain binding sites with lower affinity for Dl compared with genes responding to lower levels of Dl (reviewed by Stathopoulos and Levine,2002). Recently, it has been shown that increasing the affinities of Smad-binding sites in the Race enhancer broadens the Raceexpression domain, which argues that the affinities of the Smad-binding sites in this high level Dpp target gene enhancer are low(Wharton et al., 2004). Our results suggest that cooperative binding between Smads and Zen, which is dependent on their physical interaction, should increase their binding to the Race enhancer (Figs 4, 5, 6). It is possible that interacting with Smads at the protein level either increases the binding affinity of Zen or effectively increases the local concentration of Zen when Smads bind the adjacent sites. This in turn leads to a robust transcriptional response of Race. The overexpression results are consistent with such a model. Ectopic Zen can only activate Race if some detectable level of PMad is present (Fig. 2L),and in addition Zen must contain the Smad interaction domain(Fig. 6B).

How are the lower level target genes activated? ush and rho are expressed in a broader domain, the border coinciding exactly with that of low level PMad staining. The Zen domains, however, do not;refined zen is not broad enough, while early Zen is too broad encompassing the entire dorsal domain, though Zen could possibly be graded in this region (Rushlow et al.,1987b). Thus, it is possible that this class of target genes relies on a mechanism that uses numerous high-affinity Smad sites, and/or synergistic action of Smads with other co-factor(s) besides Zen. Such a mechanism resembles the activation of target genes in the neurogenic ectoderm of the embryo by Dl (reviewed by Stathopoulos and Levine,2002). It has been shown that the threshold responses from these genes depend on high-affinity Dl-binding sites, as well as synergistic interactions of Dl with bHLH transcription factors(Ip et al., 1992a; Jiang and Levine, 1993).

The pnr expression domain, which is about three times broader than ush, may represent a third threshold of Dpp activity. However, pnr is a different type of target gene compared with the prior classes in that it is repressed by Brk, which is present in a reverse gradient to Dpp (Jazwinska et al.,1999a; Ashe et al.,2000; Müller et al.,2003). In brk mutants, pnr expands into the ventral region, while Race and ush, for example, are unchanged. We expect that the pnr gene enhancer contains Brk-binding sites, whereas we observed in this study that the Race enhancer does not (M.X., unpublished). Brk binding sites often overlap with GNCN sites(Sivasankaran et al., 2000; Rushlow et al., 2001; Zhang et al., 2001), and it is possible that in the embryo, as in the wing disc, a concentration-dependent competition between Smads and Brk establishes the expression domains of the target genes regulated by both inputs(Rushlow et al., 2001). However, whether direct competition for binding can generate threshold responses remains to be seen. In summary, it appears that different classes of Dpp target genes are regulated by different combinations of transcription factors.

One of the simple regulatory motifs used in transcriptional networks is the feed-forward (Lee et al.,2002) or self-enabling (Kang et al., 2003) mechanism, whereby one regulator controls a second regulator and then both bind a common target gene. It has been shown both in prokaryotes (Shen-Orr et al.,2002) and yeasts (Lee et al.,2002) that this mode of regulation appears relatively frequently and is favored over others, e.g. autoregulation motifs, single input motifs in which one regulator controls several genes, or regulator chain motifs whereby one gene regulates a second which regulates a third, and so on. Such an over-representation of the feed-forward motif is probably due to its potential to provide enhanced sensitivity and temporal control to the transcriptional response. The feed-forward loop is especially suitable for eliciting precise threshold responses of morphogen targets as it allows a strong response of the target gene to small changes in the activity of the regulator that initiates the loop (Dpp), because of the combined action with the second regulator(Zen). In fact Bcd and Dl use mechanisms that are reminiscent of the feed-forward loop to activate their high level targets. Bcd regulates zygotic hunchback (hb) and together Bcd and Hb activate the downstream target even-skipped (eve) stripe 2(Small et al., 1992), and Dl activates sna with the help of Twi(Ip et al., 1992b). It is striking that the three morphogen gradients involved in specifying the Drosophila embryonic axes use the feed-forward strategy to regulate downstream target genes.

An unexpected implication from our results concerns the role of the high end of BMP morphogen gradient. In Drosophila embryos, the refined zen domain depends on peak levels of BMP activity, and we have shown that Zen can activate high level targets as long as there is some level of PMad present to facilitate DNA binding. It can be then concluded that, for the high level targets, the role of Dpp is twofold: to set the domain of zen, which we can refer to as a primary target gene; and then to act in combination with Zen to activate the other, secondary, target genes such as Race and hnt. In addition, with respect to the BMP gradient in the Drosophila embryo, we further propose that the sole purpose of the peak of the gradient is to set up the zen domain.

The authors thank Mike Levine for the Race enhancer DNA and Peter ten Dijke for the anti-phospho-Smad 1 antibodies. We are indebted to Michael Klein, Yelena Shusterman and Nzovu Ulenga for help in making and analyzing transgenic embryos. This work was supported by a grant from the National Institutes of Health (GM63024). This investigation was conducted in a facility constructed with support from Research Facilities Improvement Grant C06 RR-15518-01 from the National Center for Research Resources, National Institutes of Health.