Response to the BMP gradient requires highly combinatorial inputs from multiple patterning systems in the Drosophila embryo

A recurring theme in development is the use of morphogen gradients, where different concentrations of a regulatory molecule specify different cell fates (reviewed by Ashe and Briscoe, 2006; Rogers and Schier, 2011). For example, patterning of the dorsoventral (DV) axis of the Drosophila blastoderm embryo utilizes two morphogen gradients. A ventral-to-dorsal gradient of nuclear-localized Dorsal (Dl) divides the blastoderm embryo into three basic domains: mesoderm, neuroectoderm and dorsal ectoderm; and a dorsal-to-ventral gradient of Dpp further subdivides the dorsal ectoderm into the dorsal-most amnioserosa, and dorsomedial and dorsolateral epidermis.

Dl is a transcription factor related to mammalian NFκB, whereas Dpp is an extracellular ligand of the TGFβ family and is most related to vertebrate BMP2. Dpp interaction with transmembrane receptors results in the phosphorylation of cytoplasmic Smad (Mad-P), which in combination with Smad4 (Medea) enters the nucleus and functions as a transcription factor. Hence, both the Dl and Dpp gradients are interpreted by downstream target genes. Several candidate target genes for Dl and Dpp have been identified and classified into categories depending on the level of morphogen that they respond to: high, intermediate and low, also referred to as Type I, II and III, respectively (Ashe et al., 2000; Stathopoulos and Levine, 2004).

How do target genes interpret and respond to morphogen gradients? One widely accepted mechanism is the affinity threshold model, whereby target enhancers respond to low levels of morphogen if they contain high-affinity binding sites. For example, the sog enhancer contains high-affinity Dl binding sites and is thereby activated in regions of low-level Dl protein (the lateral neuroectoderm). By contrast, the twi enhancer contains low-affinity Dl binding sites and requires high levels of Dl for activation. Conversion of the low-affinity sites in the twi enhancer to high-affinity sites prompted reporter expression in regions containing lower levels of Dl (Jiang and Levine, 1993).

The mechanisms utilized by Dpp target genes in the early embryo are not well understood. Peak levels of Dpp and Mad-P are present along the dorsal midline in a 5- to 6-cell wide domain, the presumptive amnioserosa, while lower levels are found in the 3-4 cells to either side; beyond this, Dpp and Mad-P drop precipitously to very low levels (Rushlow et al., 2001; Dorfman and Shilo, 2001; Sutherland et al., 2003; Wang and Ferguson, 2005). Using the same approach of changing binding site affinities it was shown that the target gene Race (Ance – FlyBase), which is expressed in the presumptive amnioserosa, could respond to lower levels of Dpp if high-affinity sites were introduced into its enhancer (Wharton et al., 2004), indicating that differential binding site affinity could underlie threshold responses to the Dpp gradient.

Additional studies on Race regulation by Xu et al. (Xu et al., 2005) revealed that Race activation relies on a feed-forward loop, whereby peak levels of Dpp/Smads first activate zen expression in the dorsal-most region, and then both Smads and Zen bind and activate Race. The Dpp/Zen feed-forward motif provided a mechanism by which all amnioserosa-specific genes might be activated (Xu et al., 2005).

Analysis of a second Dpp target gene, C15 (Lin et al., 2006), which is expressed in a wider domain (8-10 cells) than Race that extends into the region of intermediate Dpp levels, revealed a different mechanism for threshold responses. The C15 enhancer contains multiple clusters of Smad binding sites that contribute cumulatively to the C15 expression domain, and, importantly, although it contains higher affinity Smad sites than those in Race, abolishing the highest affinity Mad sites had little effect on the width of the expression domain. By contrast, decreasing the number of Smad sites resulted in a narrow and sporadic expression domain (Lin et al., 2006). Using multiple Smad sites to ensure robust and precise spatial expression might be a distinguishing feature of enhancers that respond to intermediate levels of Dpp.

It remains unclear how target genes respond to low levels of Dpp in the dorsolateral region. Do they also contain multiple clusters of Smad binding sites and/or do they use the affinity threshold mechanism? Here we analyze the regulation of the pannier (pnr) gene, which is expressed in a broad dorsal domain of ∼32-35 cells (Jazwinska et al., 1999a; Ashe et al., 2000). To our surprise, the pnr enhancer structure resembled that of Race, containing a similar number of Smad sites with the same relative affinity as those in the Race enhancer, suggesting that the affinity threshold mechanism plays a minimal role, if any, in the response of pnr to low levels of Dpp. Instead, we found that sequences that lie adjacent to a crucial Smad site are required for proper pnr activation, and that the extent of pnr expression is limited by Brinker (Brk)-mediated repression. We also found that the pnr expression pattern is influenced by anteroposterior (AP) genes. Thus, pnr gene regulation relies on a complex combinatorial mechanism that integrates spatial cues from diverse patterning systems.

y¹w^67c23 embryos were used as wild type. dpp^H46 is a null allele balanced over CyO23, P[dpp⁺] (Wharton et al., 1993). Homozygous dpp^H46 embryos were identified by lack of Race staining. The null alleles zen^w36, brk^M68, kni³⁰¹, tll¹, gt^x11, hb¹² and Kr¹ were balanced over lacZ-marked balancer chromosomes to distinguish hemizygous or homozygous mutant embryos from their heterozygous and homozygous balancer siblings. bcd^E1 is a null allele.

DNA fragments (summarized in Fig. 2A) were prepared by PCR using genomic DNA as template (Clontech) and the Expand High Fidelity PCR System (Roche Molecular Biochemicals). Nucleotide changes in the Mad binding sites and the HD site were introduced by PCR mutagenesis (see Figs 3 and 5 for DNA sequences). Amplified DNA fragments were first cloned into the pCRII vector using the TOPO cloning system (Invitrogen) and then subcloned into the EcoRI site of the pCasPeRhs43-β-gal transformation vector (Thummel and Pirrotta, 1992). Plasmid constructs were mixed with helper ‘turbo’ DNA and injected into yw embryos to make transgenic flies. At least three transformant lines for each construct were analyzed.

Expression plasmids encoding GST-Mad and GST-Medea fusion proteins containing the N-terminal MH1 domains were obtained from A. Laughon (Kim et al., 1997) and M. Frasch (Xu et al., 1998), respectively. GST-Brk^N contains the N-terminal 266 amino acids (Rushlow et al., 2001). GST-Zen contained the full-length protein (Xu et al., 2005). The expression and purification of the recombinant proteins were carried out as described (Kirov et al., 1993). The concentration of the isolated proteins was determined by SDS-PAGE with defined amounts of bovine serum albumin and staining with Coomassie R-250.

DNase I footprint analyses and electrophoretic mobility shift (gel shift) assays were carried out as described (Kirov et al., 1993; Rushlow et al., 2001). All DNA fragments were generated from the pnr P3 and P4 enhancers by restriction digestion or PCR (summarized in Fig. 2B) or prepared as oligonucleotides (Integrated DNA Technologies). In each gel shift reaction, 10 ng or 30 ng of GST-Mad, GST-Medea or 5 ng Brk recombinant protein was used; about ten times more was used for footprinting reactions.

Wild-type, mutant and transgenic embryos were fixed, hybridized with pnr or lacZ antisense RNA probes labeled with digoxigenin (Roche Molecular Biochemicals) and mounted in Aqua Polymount (Polysciences) as described (Rushlow et al., 2001).

pnr encodes a GATA factor that is required for several developmental processes including proper subdivision of the embryonic dorsal ectoderm (Ramain et al., 1993; Winick et al., 1993; Heitzler et al., 1996; Herranz and Morata, 2001). In addition, pnr mutant embryos do not undergo proper dorsal closure, leaving holes in the dorsal cuticle (Heitzler et al., 1996). The pnr expression pattern is consistent with these embryonic phenotypes. pnr transcripts are first detected in the dorsal region of the cellularizing embryo from future parasegment 3 (labial) to 14 (A9) (Fig. 1A) (Winick et al., 1993; Heitzler et al., 1996; Herranz and Morata, 2001). In older embryos, expression becomes segmentally repeated and is lost from the amnioserosa, but the AP limits of the pattern remain the same (Herranz and Morata, 2001). Thus, pnr appears to be highly regulated on both the DV and AP axes.

To determine how pnr responds to DV cues, we examined pnr in different mutant backgrounds. pnr expression was reduced to a few faint stripes in dpp mutants (Fig. 1B), but was only slightly affected in zen mutants (Fig. 1C). pnr expression expanded into the ventrolateral region in brk mutant embryos (Fig. 1D). Interestingly, the ‘stripey’ nature of pnr expression that can be seen in wild type (Fig. 1A) becomes more apparent in the absence of Brk. These results suggest that pnr is activated by Dpp, but that additional activators are also required, and that Brk establishes the pnr border ventrally. Snail (Sna) is thought to repress pnr in the mesoderm, as pnr expands into the entire ventral region in the absence of both Brk and Sna (Jazwinska et al., 1999a).

To study the cis-regulation of pnr, we isolated a 2.3 kb fragment from its first intron that recapitulated the wild-type blastoderm pnr pattern in transgenic reporter assays (Fig. 2A,C). We dissected the 2.3 kb enhancer into four overlapping subfragments termed P1-P4 (summarized in Fig. 2A), and two of these, P3 and P4, were able to drive blastoderm lacZ reporter expression (Fig. 2D,E), but gave different patterns. P3 gave rise to a dorsal expression domain similar to the wild-type pnr domain, and two posteriorly located stripes. P4 drove four anterior stripes that could be clearly distinguished. These results indicated that: (1) the stripe pattern can be separated from the broad dorsal pattern, which we now refer to as the dorsal patch; and (2) the stripes can be separated from each other, suggesting that they are regulated by discrete enhancers, as in the case of pair-rule genes such as hairy (h) and even skipped (eve) (Howard et al., 1988; Harding et al., 1989), although the pnr stripes are more closely spaced than pair-rule stripes.

The observation that the pnr pattern is a composite of two patterns was further substantiated by assaying the reporter genes in dpp and brk mutant backgrounds (Fig. 2F-I). In dpp mutants, the P3-lacZ dorsal patch was abolished, but the posterior stripes remained (Fig. 2F). P4-lacZ was unaffected (data not shown), indicating that only the dorsal patch in the P3 enhancer is Dpp responsive, whereas the stripes are regulated by other factors. In brk mutants, the pattern expanded ventrally (Fig. 2G-I) in a similar manner to the expansion of endogenous pnr in brk mutants (Fig. 1D), but the patch expanded to a lesser degree than the stripes to only ∼40% of the circumference, whereas the stripes expanded through the ventral neuroectoderm (Fig. 2H,I). This explains why, in the brk mutant, the expanded pnr stripes become more evident. These results indicate that the pnr pattern comprises two subpatterns of a dorsal patch and six stripes that are regulated by different factors. Dpp activates the patch, whereas AP factors are likely to direct the precise location of the stripes (see below). Interestingly, Brk represses both patterns in the ventrolateral region.

In order to identify Smad binding sites in the P3 and P4 enhancers, we performed DNA binding assays (gel shifts) with five overlapping DNA fragments (I-V; Fig. 2B) and recombinant GST-fused Mad (GST-Mad) and GST-fused Medea (GST-Medea) proteins. Two fragments (II and III) spanning 350 bp in the P3 region were able to bind Mad and Medea, whereas the fragments in P4 showed no binding (data not shown). This is consistent with the genetic analysis of P3 and P4 showing that only P3 is Dpp responsive. To search for specific sequences responsible for Smad binding, we performed gel shift assays with ten overlapping oligonucleotides (a-j) covering most of fragments II and III within P3 (Fig. 3A; binding summarized in Fig. 2B as blue ovals). Strong Mad binding was observed with oligonucleotide f (Fig. 3A), which contains the sequence GGCGCC (Fig. 3B, labeled as M3), resembling the GC-rich Drosophila Mad consensus sequence (DSC in Fig. 3B) GCCGnCG (Kim et al., 1997; Xu et al., 1998). Strong Medea binding was observed with oligonucleotide h, which contains a sequence matching the Smad binding element (SBE in Fig. 3B) CAGAC (Zawel et al., 1998), as well as a GC-rich sequence (GC in Fig. 3B).

Two Brk binding sites, B1 and B2, located in the P3 and P4 enhancers, respectively, were identified by DNase I protection (footprinting) assays (supplementary material Fig. S1; summarized in Fig. 2B as red ovals) and confirmed by gel shift (see Fig. 5F). Both sites contain the Brk recognition core GGCGC (Kirkpatrick et al., 2001; Rushlow et al., 2001; Zhang et al., 2001). B1 is a stronger Brk site than B2, probably because it contains two overlapping Brk core sites in opposite orientation (GGCGCC). Importantly, the B1 site overlaps with the Smad M3 site (Fig. 3B), suggesting that Mad-P emanating from the dorsal side and Brk from the ventral region (Jazwinska et al., 1999a) set the border of the pnr patch by competing for binding to the pnr enhancer, as has been shown for other Dpp targets (Sivasankaran et al., 2000; Kirkpatrick et al., 2001; Rushlow et al., 2001; Zhang et al., 2001).

To determine whether the Smad binding sites are important for pnr activation, mutations were introduced into the two strongest Smad sites M3 and M4. We first tested oligonucleotides containing mutated Smad sites in gel shift assays. Mutation of M3 (Fig. 3B, M3m) abolished Smad binding (Fig. 3C). For M4, we tested oligonucleotides that altered the SBE site (M4m1), the GC-rich site (M4m2) or both (M4m). M4m1 greatly reduced Smad binding, whereas M4m2 has only a slight effect; mutating both sites (M4m) abolished all in vitro Smad binding (Fig. 3C).

Since only the patch is Dpp responsive, we tested the in vivo relevance of the Smad site mutations in the context of the P3-lacZ reporter gene. Surprisingly, eliminating the two strong Smad sites M3 and M4 caused complete loss of the patch pattern (Fig. 4B), similar to what happens to P3 in dpp^– (Fig. 3F), but the two posterior stripes remained, further verifying that the stripes are regulated by different cues. The ventral expansion of these two stripes can be explained by a lack of direct Brk repression, as mutation of the M3 site also abolished Brk binding.

Mutation of the M3 and M4 sites individually caused different effects on the patch pattern, particularly on its DV width. Compared with the wild-type P3 pattern (Fig. 4A), the M4m mutation resulted in a slightly narrower patch (Fig. 4C), whereas the M3m mutation resulted in a very narrow patch domain (Fig. 4D) that resembled the pattern of Race. Hence, by abolishing a single Mad site, we converted a low-level target into a high-level target. We further examined this idea by assaying P3M3m-lacZ in different mutant backgrounds. Expression of the patch pattern was absent in dpp^H46 heterozygotes (Fig. 4E) and zen^w36 homozygotes (Fig. 4F), indicating that P3M3m-lacZ responds only to high levels of Dpp [Race expression disappears in such mutants (Xu et al., 2005)]. These results suggested that Smad sites collectively contribute to the broad domain of pnr, but that the M3 site is crucial.

The dramatic effect of mutating the M3 site suggested that a single Smad site might mediate the response to low levels of Dpp. One possibility is that M3 is a high-affinity Smad site that is capable of binding Smads when present in low concentrations. To address this, we first compared the enhancer structure of pnr with those of Race and C15, all of which give different expression boundaries. To our surprise, the Race and pnr enhancers had a similar number of Smad sites (Fig. 5A,B), whereas the C15 enhancer contained many more Smad sites spread over a broad region (Fig. 5A). We next performed gel shift assays with DNA fragments of the Race, C15 and pnr enhancers (Fig. 5C) and oligonucleotides containing individual Smad sites (Fig. 5D). Unexpectedly, the strongest Smad sites in the pnr enhancer, M3 and M4, showed a similar degree of binding to Smads as sites in the Race enhancer (Fig. 5D).

To further compare Smad sites, we replaced the Smad site in the M3 oligonucleotide with that from the Race enhancer (M3R; sequences shown in Fig. 5B). Competitive gel shift analysis showed that M3 and M3R had similar Smad binding affinities (Fig. 5E). Importantly, swapping the M3R site for M3 in the P3-lacZ transgene had little effect on patch expression (Fig. 4G); however, the posterior stripes expanded, a likely result of the observed decreased binding of Brk to M3R (Fig. 5F). Based on these results, the affinity threshold model is not sufficient to explain the broad expression domain of pnr, as there is no correlation between Smad binding affinity and the extent of the Race and pnr expression domains.

The fact that responsiveness to the Dpp gradient does not depend on differential Smad binding raised the possibility that other factors might be involved; for instance, a co-factor might interact with Mad to enhance Smad binding to M3 in vivo. A study of the BMP-responsive enhancer of Msx2, a mammalian homolog of tinman, revealed that the sequence AGCAATTAA, which includes a consensus site (AATTAA) for Antennapedia (Antp) homeodomain (HD) factors, lies adjacent to a Smad site (CGGCTCCGGC), and that both parts of this ‘bipartite’ motif are required for BMP responsiveness in mouse embryos (Brugger et al., 2004). Intriguingly, the exact same sequence AGCAATTAA lies adjacent to the M3 site (Fig. 5G). Comparison of this bipartite motif in the pnr gene among Drosophila species showed that these nine nucleotides are highly conserved (Fig. 5G) and, importantly, are not present in the enhancers of Race or C15 (Fig. 5B; data not shown).

Introducing a mutation into the bipartite site (Fig. 5G, P3HDm) in the P3-lacZ reporter gene resulted in a dramatically altered expression pattern that was sporadic and narrow (Fig. 4H). Thus, as in the case of the mouse Msx2 enhancer, both the Smad site and the HD site of the pnr bipartite motif are essential for proper response to Dpp.

The six stripes of pnr comprise 20-60% egg length (Fig. 1C) (Winick et al., 1993). The anterior-most stripe overlaps with eve stripe 2 and the posterior-most stripe abuts fushi tarazu (ftz) stripe 7; thus, pnr extends from the labial head segment through the entire abdomen (Herranz and Morata, 2001). Since pair-rule stripe borders are established by gap repressor proteins, which are localized in specific domains across the AP axis, we analyzed pnr expression in bicoid (bcd) and gap gene mutants (Fig. 6). The pnr pattern was affected in all of the mutants that we tested [bcd, hunchback (hb), knirps (kni), Kruppel (Kr), giant (gt) and tailless (tll)] and the changes correlated with what is known about the function of these genes and their cross-regulatory interactions (Fig. 6H). More specifically, pnr expression expanded anteriorly in bcd, hb and gt mutants (Fig. 6B-D), which was likely to be due to a lack of anterior Hb and Gt repression (Clyde et al., 2003). Likewise, pnr expanded posteriorly in tll^– (Fig. 6G) due to a lack of posterior Tll repressor activity (Moran and Jimenez, 2006). Note that weak derepression of pnr also occurs in the head region in tll^–. The lack of mid-body region repressors in the Kr^– and kni^– mutants caused shifting and/or expansion of gap domains (Kosman and Small, 1997); thus, both direct and indirect effects alter the middle pnr stripes in these mutants (Fig. 6E,F).

Results from these genetic experiments, together with genome-wide binding data in which all AP factors that we tested except Bcd were found to bind to the pnr enhancer (Li et al., 2008; MacArthur et al., 2009), and with binding site predictions using ClusterDrawII (Papatsenko, 2007) across P3 and P4 (not shown), strongly suggest that the gap genes establish the AP limits of the pnr domain and the stripes within the domain.

Our studies on the low-level Dpp target gene pnr revealed that: (1) the pnr expression pattern is a composite of a dorsal patch and six AP stripes; (2) the pnr patch depends on Dpp, although we predict that additional co-factors that bind adjacent to Smads enable the pnr enhancer to respond to low levels of Dpp, leading to the unique broad domain of pnr; (3) Brk repression establishes the ventral border of both the patch and stripe expression domains and thus Brk shapes the pnr expression domain on the DV axis; (4) pnr expression is precisely positioned along the AP axis by the segmentation gap genes.

Most blastoderm genes are regulated primarily on either the DV or AP axis. For example, the gap genes are expressed in one or two domains of expression along the AP axis and, although some of them may exhibit regulation along the DV axis, they are nonetheless considered AP genes. pnr represents an interesting case because although it was originally reported as a DV gene (Jazwinska et al., 1999a; Ashe et al., 2000), closer inspection of its expression pattern in wild-type and mutant embryos and detailed dissection of its cis-regulatory enhancers revealed that pnr is highly regulated by both AP and DV genes (summarized in Fig. 7). Its pattern is a composite of two superimposed patterns that each exhibit AP and DV spatial regulation: a dorsal patch and six AP stripes, which are limited to the dorsal 30% of the embryo (Fig. 1A). The patch domain, but not the stripes, disappeared in dpp mutants (Fig. 1B), whereas both the patch and stripes expand ventrally in the absence of Brk (Fig. 1D). The stripes are more sensitive to Brk repression because activation of the patch domain is limited to the region where Dpp is present dorsally, whereas the stripes can be activated along the entire DV axis. Brk in the ventrolateral region and Sna in the ventral-most region repress stripe expression (Fig. 1F) (Jazwinska et al., 1999a). Since pnr specifies dorsomedial fates, restricting its expression to the dorsal 30% of the circumference is crucial. Ectopic expression of pnr ventrally causes transformation of ventral epidermis into dorsomedial epidermis (Herranz and Morata, 2001).

Competition between Brk and Smads for binding to overlapping DNA sequences is likely to set the border of the patch domain. Two Smad sites are particularly important for patch expression (Fig. 4B), and one of these, the M3 site, is a composite site that binds both Brk and Smads (see Fig. 2B and supplementary material Fig. S1), raising the possibility that the patch border is established by competition between activating inputs from Smads in the dorsal region and repressive inputs from Brk emanating from the ventral region. Competition between Brk and Smads for overlapping binding sites has been observed for several Dpp target enhancers (Sivasankaran et al., 2000; Kirkpatrick et al., 2001; Rushlow et al., 2001; Zhang et al., 2001).

Repression of the AP stripes ventrally requires both Brk sites B1 and B2 (Fig. 2B, red ovals). The two posterior stripes driven by P3 expand to a lesser degree (Fig. 2D) than the four anterior stripes driven by P4 (Fig. 2E). This can be explained by the fact that P4 lacks Brk site B1, which is a stronger Brk site. Loss of both Brk sites would likely result in expansion to the edge of the mesoderm, as seen in embryos that lack Brk protein (Fig. 1D; Fig. 2G-I). Repression by Sna is likely to involve the Sna binding sites in the pnr enhancer, as genome-wide binding studies have shown that the pnr enhancer is bound by Sna (summarized in Fig. 6G) (Zeitlinger et al., 2007).

The positioning of the stripes, as well as of the patch, along the AP axis is regulated by the gap genes. Our results suggest that Hb, Gt and Tll set the anterior edge of the pnr domain, whereas Tll sets the posterior, and that direct and indirect interactions among the gap proteins establish the stripe borders relative to one another, as has been observed for eve (Clyde et al., 2003). For example, the broad central stripe seen in kni^– (Fig. 6F) could be explained by the lack of direct Kni repression. However, owing to the complex cross-regulatory interactions among the gap genes, it is difficult to predict which gap proteins regulate the pnr stripes directly, although genome-wide binding data of the gap factors (Li et al., 2008; MacArthur et al., 2009) support their direct binding to the pnr enhancer. Although Bcd does not appear to bind directly to the pnr enhancer, its effects are mediated through its targets Gt and Hb.

In depth studies of three genes with different boundary positions in the dorsal region, Race (Xu et al., 2005), C15 (Lin et al., 2006) and pnr (this study), indicate that complex combinatorial mechanisms are employed to establish their expression domains, with each gene having a unique regulatory network of its own. Although they all respond to Dpp signaling, their borders of expression are not set by a simple threshold response to the Dpp gradient that depends on differential binding site affinity.

The feature that has been shown to be important for high-level Dpp target expression is the feed-forward motif involving Dpp and Zen (Xu et al., 2005). High levels of Dpp/Smads first activate zen expression in the dorsal-most region, the presumptive amnioserosa, and then both Zen and Smads bind and activate the Race enhancer (Fig. 7) (Xu et al., 2005). The intermediate-level target C15 has a different enhancer structure than high-level targets, containing many Smad sites that act in a cumulative manner to drive expression in regions of intermediate Dpp levels. Mutation analysis has shown that the number of intact Smad binding sites, rather than their affinity, is important for the C15 response (Lin et al., 2006). Nevertheless, the enhancer structure of C15 might promote high levels of Smad binding in vivo, and this may increase the response to Dpp. Do all intermediate-level Dpp targets have a similar enhancer structure? We examined the enhancer that drives expression of the intermediate-level Dpp target gene tup (Zeitlinger et al., 2007) for putative Smad binding sites (SBEs and GC-rich regions), and observed multiple Smad sites across the enhancer (data not shown), similar to that seen with C15. Thus, the multiple Smad site signature might be necessary for response to lower than peak levels of Dpp. In addition, intermediate-level targets may utilize repression mechanisms to help establish their borders of expression, as was shown for C15 (Fig. 7) (Lin et al., 2006).

Our studies here revealed that the pnr enhancer resembles that of a high-level target in Smad site organization and Smad binding site affinity (Fig. 5A,B). In fact, it was surprisingly easy to convert the low-level target enhancer into a high-level target by mutating a single Smad site (Fig. 4D). This result could be easily explained if the M3 site had a higher affinity for Smads than those in Race; however, comparison of the binding sites by gel shift showed they have similar affinities (Fig. 5D). Furthermore, replacing the M3 Smad site with a Race Smad site had little effect on the expression pattern (Fig. 4G). These results suggest that activation of pnr in its broad domain has little to do with Smad binding affinity. How then does pnr respond to low levels of Dpp? One possible mechanism involves the highly conserved AGCAATTAA site that lies adjacent to the Smad sites (Fig. 5G). In the absence of this site, the P3 enhancer could not respond to low-level Dpp (Fig. 4H). It is possible that this site, when bound, leads to greater Smad binding, which would then promote pnr activation.

What factor(s) might bind to the AGCAATTAA site? ATTA is the core binding site for Antp class HD proteins. Although Zen binds to the ATTA site in vitro (supplementary material Fig. S1), neither the endogenous pnr pattern (Fig. 1C) nor P3-lacZ expression (data not shown) is significantly affected in zen mutants. To identify candidate factors, we used the TOMTOM tool (Gupta et al., 2007) at FlyFactor Survey (Zhu et al., 2011), and the best match was to the HD protein Hmx, which binds CAATTAA. However, Drosophila Hmx is expressed only in an anterior region that does not overlap with pnr (see FlyBase). Likewise, although several Antp class HD proteins were predicted to bind to the ATTA core sequence, their timing or domains of expression do not overlap ideally with those of pnr.

Brugger et al. (Brugger et al., 2004) proposed that the AGCAATTAA site in the Msx2 enhancer might bind a factor in addition to an HD protein via the 5′ half of the site, perhaps a transcriptional partner such as FAST1, which was previously shown to function with Smads (Chen et al., 1996). Although our search did not reveal any candidates, if this is the case for pnr then the bipartite motif could potentially bind four proteins: Smads, Brk, HD and ‘partner X’. The combination of these proteins in a given cell along the DV axis would determine pnr transcriptional activity. The fact that the bipartite motif is not present in the enhancers of Race or C15, or in the other pnr enhancers we identified, demonstrates the versatility of how Dpp uses different partners to establish multiple target gene domains.

Is the structure of the pnr enhancer typical for low-level Dpp targets? This is difficult to address owing to the lack of candidate low-level Dpp targets. Brk is considered a low-level Dpp target in imaginal disc development; however, Dpp represses brk, giving rise to a reciprocal gradient of the Brk repressor (Jazwinska et al., 1999b; Pyrowolakis et al., 2004). Target gene borders are thus established by competition between Smad and Brk for overlapping binding sites, as mentioned above for pnr. The brk enhancer contains multiple enhancer/silencer modules consisting of activator and repressor (Mad/Medea/Schnurri) binding sites, which contribute to threshold responses to the Dpp gradient (Yao et al., 2008), and thus it does not resemble the pnr enhancer. Although we have made good progress in understanding how pnr is expressed in regions with low levels of Dpp, learning the general rules that control broad dorsal patterns will require the analysis of more enhancer elements.

What rules do target genes for other morphogens follow? Long before the ‘feed-forward’ term was coined (Shen-Orr et al., 2002; Lee et al., 2002) it was shown that both the Dl and Bcd morphogens interact with their high-level targets, Twi and Hb, respectively, to activate downstream targets (Jiang et al., 1991; Ip et al., 1992; Small et al., 1992); thus, combinatorial motifs are generally utilized. Moreover, as more target genes of Dl and Bcd were identified and studied, it became apparent that the affinity threshold model could not explain all cases of differential response to the gradient. For example, analysis of several enhancers that drive Bcd-dependent expression in anterior regions of the embryo revealed a poor correlation between Bcd binding site affinity and the AP limits of the pattern (Ochoa-Espinosa et al., 2005). Also, although Dl targets remain archetypal examples of genes that utilize the affinity threshold mechanism, it was found that genes expressed in the lateral region also require input from the Zelda (Vielfaltig – FlyBase) transcription factor for expression in regions of low-level Dl (Nien et al., 2011). Zelda binding sites are present in target enhancers, and it was proposed that Zelda boosts Dl binding to help activate the neuroectodermal genes (Nien et al., 2011).

Downstream target gene interactions also shape domains of expression, in particular cross-repression among the targets. In both the Drosophila neuroectoderm and the vertebrate neural tube, morphogen targets are expressed in discrete domains rather than nested overlapping domains due to the repression of one target by another (reviewed by Dessaud et al., 2008; Ashe and Briscoe, 2006). This mechanism establishes sharp boundaries among the target genes.

Thus, it is clear that additional factors help morphogens set threshold responses. Given that the pnr enhancer could potentially interact with four different factors along the DV axis and at least four factors along the AP axis, several combinations of inputs could regulate other Dpp target genes. More generally, depending on the number of different factors that interact with the cis-regulatory regions of target genes, morphogen gradients could elicit multiple threshold responses, as has been seen for morphogens such as Dl in Drosophila, Activin in the Xenopus blastula (Green et al., 1992) and Shh in the vertebrate neural tube (Ericson et al., 1997), where up to seven threshold responses have been described. Only by dissecting enhancers can we fully understand how target genes integrate diverse inputs.

We thank Nikolai Kirov and Jeongsook Park for help in preparing Smad and Brk proteins; Lilly Sehgal and Jenna DeRovira for help with transgenic constructs and flies; Zhe Xu for helpful discussions; and Steve Small for insightful suggestions on the manuscript.