A phylogenetically conserved cis-regulatory module in the Msx2promoter is sufficient for BMP-dependent transcription in murine and Drosophila embryos

A striking property of morphogens is that they act through a small number of signaling pathways, yet elicit a wide variety of cellular responses in different developmental settings. How this diversity is achieved remains an unanswered question. It is clear that part of the solution is in the complement of transcription factors present in a responding cell. Such factors, through interactions with cis-regulatory elements, transform a morphogen signal into a pattern of gene expression. To understand morphogen action, it is therefore crucial to identify these transcription factors and to understand how their interactions on target promoters control gene expression. Here, we address the problem of morphogen specificity by investigating a BMP-responsive enhancer of the murine Msx2 gene.

BMPs are a large class of TGF-β related ligands with diverse activities in embryonic development. They can function as classical morphogens, eliciting changes in cell fate in a concentration-dependent manner(Gurdon and Bourillot, 2001). Interaction of BMPs with their cognate receptors leads to the phosphorylation of members of the receptor-regulated Smad family (R-Smads)(Heldin et al., 1997; Massague, 1998). Upon phosphorylation, R-Smads associate with Smad4 and translocate to the nucleus,where they serve as effectors of transcription(Kretzschmar et al., 1997; Lagna et al., 1996; Liu et al., 1997). Smads 1, 5 and 8 function in the BMP pathway, Smads 2 and 3 in the activin and TGF-βpathways, and Smad4 in all three branches of the TGF-β superfamily(Hoodless et al., 1996; Liu et al., 1996; Massague and Chen, 2000; Shi and Massague, 2003; Suzuki et al., 1997; Wiersdorff et al., 1996).

Although Smads are crucial for the transcriptional activation of BMP target genes, they bind DNA weakly and are expressed broadly in embryonic tissues(Shi et al., 1998; Wrana, 2000). Therefore, the tissue-specific activation of BMP target genes probably involves sequence-specific transcription factors that cooperate with Smads. Despite the importance of such factors, few have been identified and little is known about how they interact with Smads to regulate transcription(Hata et al., 2000; Henningfeld et al., 2002). The Msx2 promoter provides an ideal starting point in the search for such factors.

Msx2 is one of three closely related genes in mammals(Bell et al., 1993; Davidson, 1995; Maxson et al., 2003; Pollard and Holland, 2000). A well-documented BMP target, Msx2 is expressed in a subset of known domains of BMP function (Furuta and Hogan,1998; Graham et al.,1994; Hollnagel et al.,1999; Vainio et al.,1993). Targeted mutations in Msx2 have verified that it acts downstream of BMPs and is required for a subset of BMP-mediated cellular responses (Satokata and Maas,1994; Satokata et al.,2000). The regulatory elements controlling Msx2expression thus provide a useful model for investigating how BMP signals are processed at the level of the promoter to produce different BMP responses in specific developmental settings.

Here, we report the identification and fine-structure analysis of a BMP-responsive enhancer (BMPRE) located upstream of the murine Msx2gene. We demonstrate that consensus Smad elements and a consensus homeodomain element control the temporal and spatial pattern of BMP-dependent transcription directed by this enhancer. We show that the nucleotide sequence of this enhancer is highly conserved within mammals, but not in other vertebrate classes or nonvertebrates. Despite this lack of conservation outside mammals, this Msx2 BMPRE accurately interprets Dpp signals in Drosophila embryos and imaginal discs. Furthermore, as in mouse embryos, it exhibits a dependence on the consensus homeodomain as well as consensus Smad sites. These results demonstrate that the Msx2 BMPRE can respond to BMP/Dpp signals in two widely divergent animal groups, and suggest the functional design represented by this enhancer originated early in the evolution of the bilateria.

For transient transfections, segments of the murine Msx2 5′flanking sequence were fused with a TK minimal promoter and subcloned into PGL2 (Promega). The Δ3Msx2luciferase construct contained a 1.8 kb BamHI fragment located between –5082 and –3298 bp upstream of the translation start site. The 560 bpΔ 4Msx2luciferase construct contained a fragment (–3862 to–3303) generated by PCR. The insert in the 560Δ52bpMsx2luciferase construct was created by ligating two PCR products together, with the 52 bp BMP-responsive region being omitted. The 560 bp homeodomain mutant (hdm) insert was generated using the QuickChange Site-Directed Mutagenesis Kit (Stratagene) and the primers 5′-GCCGGGCCGAGAGCTCGTACCGCGGCTCCGGCGCG-3′ (forward) and 5′-CGCGCCGGAGCCGCGGTACGAGCTCTCGGCCCGGC-3′ (reverse). This insert was subcloned into TK-PGL2 to generate 560bpMsx2-hdm-luciferase.

For the production of transgenic mice, Msx2 promoter fragments were fused with the hsp68 minimal promoter and a lacZreporter (Kothary et al.,1989). The 1.8 kb Δ3Msx2-hsplacZ and 560 bpΔ 4Msx2-hsplacZ transgenic mice have been described(Kwang et al., 2002). TheΔ 5Msx2-hsplacZ construct contained a 220 bp PCR fragment located between –3521 and –3310 bp upstream of the Msx2translation start site. The 52 bp Δ6Msx2-hsplacZ construct contained a tetramer repeat of the sequence between –3523 and–3471. This construct, and those bearing mutations in the Smad and homeodomain sites, were generated using dimer repeat oligos that were ligated together immediately upstream of the hsp68-lacZ-SV40 cassette. The 560Δ52bpMsx2-hsplacZ and 560bpMsx2-hdm-hsplacZ transgenes were generated as described previously.

Msx2 promoter fragments used for producing transgenic flies were cloned into the CPLZN P-element vector. For the 560, 480 and 220 bp promoter fragments, inserts were generated by PCR. The 52 bp transgene consisted of a tetramer repeat.

10T1/2 cells were propagated in DMEM with 10% FBS. The cells were transfected with the reporter constructs using Superfect (Qiagen) and 24 hours later induced with BMP4 (60ng/ml final) (R&D). After another 24 hours in culture, the cells were harvested and luciferase assays performed using the dual luciferase system (Promega).

Affigel agarose beads (BioRad) were washed in PBS prior to incubating in 0.1% BSA with 100 ng/μl BMP4 (R&D), 5 ng/μl TGFβ1 (R&D) or BSA (0.1%) for 30 minutes at 37°C. Beads were placed on tissues in transwell membranes, immersed in DMEM. For limb-bud implantations, BMP4 beads were placed on the dorsal side of left forelimbs prior to overnight incubation in 6% CO₂. Tissues were then fixed for 10 minutes in 4%paraformaldehyde and stained for lacZ activity.

MLB13MYC-clone-14 (C-14) cells (Rosen et al., 1994) were cultured in DMEM (10% FBS) and induced with 60 ng/ml BMP4 (R&D) for 30 minutes prior to fixation with formaldehyde (1.0%final concentration) at room temperature for 10 minutes. ChIP assays were performed as described (Ma et al.,2003) using an anti Phospho-Smad1 antibody (Upstate). PCR amplification was performed using primers that span the 52 bp BMP-responsive region of the Msx2 promoter (32 cycles). Primer sequences were: BMPRE(forward) 5′-TCT GCC CAG TTG GAG GTT TGA-3′ and (reverse)5′-GCC GCG TTA ATT GCT CTC G-3′; upstream control (forward)5′-GCA ACA AAC ATC CCT GAG A-3′ and (reverse) 5′-CTG CCT CCT AAC CTT CAT AG-3′.

GST-Smad4 (MH-1 domain) fusion proteins were expressed in BL21 bacteria and purified by chromatography on glutathione beads (Pharmacia). Prx1b protein was expressed from a mouse cDNA (Norris and Kern, 2001) by coupled in vitro transcription-translation (TNT kit, Invitrogen). All probes used were labeled withα-³²P-dCTP. EMSA was performed as previously described(Kwang et al., 2002).

Mouse embryos were fixed in 4% paraformaldehyde at 4°C prior to staining for lacZ (Liu et al.,1994). Whole-mount in situ hybridization on mouse embryos was performed as described previously (Hogan et al., 1994; Kwang et al.,2002). In situ hybridization and detection of β-galactosidase expression in Drosophila embryos was carried out as described in Arora et al. (Arora et al.,1994). To detect phosphorylated R-Smads, tissue was fixed in 4%paraformaldehyde and equilibrated in 30% sucrose prior to freezing in Histo Prep (Fisher). Frozen sections were cut (8 μm) and fixed in cold acetone for 10 minutes. Primary anti phospho-Smad1, 5, 8 antibody (Cell Signaling Technology) was diluted 1/50 in PBST/1.0% BSA and incubated at 4°C overnight. The secondary antibody, anti-Rabbit Rhodamine (Calbiochem), diluted 1/200 in PBST/1.0% BSA, was incubated for 1 hour at room temperature. Sections were counterstained with DAPI and cover slipped using Vectashield (Vector Labs). lacZ staining of frozen sections was performed as described in Ishii et al. (Ishii et al.,2003).

Transgenic mouse embryos and lines were generated by pronuclear injections as described by Liu et al. (Liu et al.,1994). Transgenic flies were produced by germline transformation as described by Arora et al. (Arora et al.,1994).

DNA was prepared from mouse tails or embryo yolk sacs as described by Hogan(Hogan, 1994). Msx2-lacZ transgene genotypes were determined by PCR using primers to lacZ as described by Kwang et al.(Kwang et al., 2002).

Previously we identified a 1.8 kb fragment of the 5′ flanking sequence of Msx2 (Fig. 1A, Δ3Msx2) that was sufficient to confer BMP responsiveness on a reporter gene in cultured cells(Daluiski et al., 2001). A closely related fragment directed reporter gene expression to most sites in the developing embryo where endogenous Msx2 is expressed(Kwang et al., 2002). A 560 bp subfragment of the 1.8 kb fragment (Fig. 1A, Δ4Msx2) responded to BMP4 stimulation in 10T1/2 cells to approximately the same extent as the parental 1.8 kb fragment(Fig. 1B). A transgene comprising the 560 bp fragment fused with hsp68-lacZ also was sufficient for correct spatiotemporal expression in embryos(Fig. 1C-F).

To determine whether the 560 bp fragment was able to respond to BMP signals in murine embryos, we carried out BMP bead implantations in mice bearing the 560bpMsx2-hsplacZ transgene (Fig. 1E,G). Beads soaked in BMP4 were implanted on E11.5 limb buds, and after 16 hours in culture, lacZ activity was assessed. A halo of lacZ activity appeared around implanted beads within 15 minutes of staining and grew significantly darker by 1 hour(Fig. 1G). Control beads soaked either in BSA or TGFβ1 did not elicit a lacZ signal(Fig. 1G), demonstrating the response was specific for BMP. When bead implantations were performed at E12.5, a weaker response was evident (Fig. 1G), indicating the response was stage-specific. BMP4-soaked beads also stimulated β-galactosidase activity in the head, mandible and tail(data not shown).

Alignments of Msx2 loci from mouse, rat, rabbit, human and chimpanzee revealed a 480 bp block of homology positioned between 3 and 4 kb upstream of the translation start sites(Fig. 2). Located within the 560 bp BMP-responsive region, this block exhibited an average identity between different orders of mammals of ∼90%. By contrast, the 3.3 kb promoter region located between the conserved block and the translation start site had an average identity of 56%, suggesting that selection has maintained the sequence of the 480 bp block. No significant matches with this 480 bp region were evident in non-mammals. Nor did mammalian sequences outside the Msx2 locus match the 480 bp sequence. A survey (MatInspector 7.0,2003) of consensus cis-regulatory elements within the conserved region revealed several consensus binding sites with potential roles in BMP signaling. These included Smad1 and Smad4 elements, as well as sites for OAZ,Creb, Tcf/Lef1 and Hnf/Fox (Fig. 2) (Ionescu et al.,2004; Ishida et al.,2000; Johnson et al.,1999; Rice et al.,2003; Shim et al.,2002; Zawel et al.,1998).

We tested a series of Msx2 promoter fragments for their ability to mediate BMP responsiveness in transgenic embryos(Fig. 3A). A 220 bp subfragment of the 560 bp fragment drove reporter expression in a near-normal spatial pattern in embryos, with the only significant difference being in the posterior limb bud (Fig. 3C; Table 1). Like the 560 bp fragment, this 220 bp fragment conferred BMP responsiveness on a lacZreporter in limb buds (Fig. 3F), as well as in tails and mandibles (data not shown). Limb buds bearing BSA and TGFβ1 control beads were negative (data not shown).

Fragments responding to BMP treatment had in common a 52 bp region(Fig. 3A, asterisk). This sequence included Smad and homeodomain consensus sites(Fig. 2). To determine whether this sequence was important for BMP responsiveness, we deleted it in the context of the 560 bp fragment, and tested the activity of the mutant transgene (560Δ52bpMsx2) in transgenic mice and 10T1/2 cells. In transgenic embryos at E11.5 (eight independent lines), the mutated 560 bp fragment drove lacZ expression at much lower levels(Fig. 3D) compared with the parental fragment (Fig. 3H), as evidenced by the greatly extended staining time required for the 560Δ52bpMsx2 transgene to achieve a similar relative intensity as the transgenes containing the 560 or 220 bp fragments(Fig. 3B,C). Expression in both the limb bud mesenchyme and AER was reduced substantially (compare Fig. 3G with Fig. 1G). Bead implantations on limbs carrying the mutant transgene elicited no detectable lacZactivity (Fig. 3G), whereas implantations on limbs carrying the wild-type 560bpMsx2 transgene resulted in robust lacZ signals(Fig. 3E). In 10T1/2 cells, the 52 bp deletion caused a near-complete loss of BMP responsiveness without affecting the basal level of promoter activity(Fig. 3I). Together, data from transgenic embryos and cultured cells suggested that the 52 bp sequence is required for BMP responsiveness in limb bud mesenchymal cells and in 10T1/2 cells.

We next asked whether the 52 bp sequence was sufficient to respond to a BMP signal in vivo. To minimize position effect, we produced a tetramer of this sequence, which we inserted upstream of the hsp68-lacZreporter. In transgenic embryos, this construct was expressed in a subset of the sites where the parental 560bpMsx2-hsplacZ transgene and the endogenous Msx2 gene are expressed(Fig. 4A-I; Table 1) (10 independent lines). These included the cardiac outflow tract, pharyngeal arches, genital region, allantois, limb buds, mandible, nasal process, midbrain, eye, otic vesicle and hair follicles. An extended staining time was also required to achieve signal intensities comparable with those of larger Msx2constructs, indicating reduced transgene activity.

That the 52 bp fragment responded to BMP4 in a tissue-restricted manner compared with the parental 560 bp fragment was evident from experiments in which beads were placed in various tissues and at different locations in limb buds of E11.5 transgenic embryos. A BMP4-soaked bead placed in the center or anterior of the limb elicited a lacZ signal comparable to that of the 560bpMsx2 transgene (Fig. 4J,K; 60 minutes). When a BMP4-soaked bead was placed in either the proximal or posterior regions of the limb, it did not stimulate lacZ activity (Fig. 4K; data not shown). By contrast, the 560bpMsx2 transgene responded to BMP4 beads placed in each of these locations(Fig. 4J, Fig. 6N). Furthermore, while the 560 bp fragment responded to beads placed on the head or tail, the 52 bp fragment did not. The BMP responsiveness of the 52bpMsx2 transgene was also restricted temporally. Whereas the 560 bp fragment responded to beads implanted in E12.5 limbs (Fig. 1G), the 52 bp fragment did not(Fig. 4K). TGFβ1 beads elicited no lacZ response (Fig. 4K), demonstrating the 52bpMsx2 transgene remained specific for the BMP pathway.

When transfected into 10T1/2 cells, a multimerized 52 bp fragment did not respond to BMP stimulation (data not shown), suggesting that 10T1/2 cells,like much of the limb mesenchyme, lacked the transcriptional machinery necessary to activate expression of the BMP-responsive element. Transgenic and transfection approaches thus showed that the 52 bp fragment directed expression and mediated BMP responsiveness in a subset of tissues and developmental stages compared with the parental 560 bp fragment.

That the 52 bp transgene is expressed at sites of active BMP signaling was confirmed by staining with an antibody against the phosphorylated (active)form of the BMP restricted Smads 1, 5, 8 (R-Smads). Limbs exhibited nuclear staining in a halo surrounding implanted BMP4-soaked beads(Fig. 5D,E), coincident with lacZ expression (Fig. 5B,C). Nuclear staining was also evident in the anterior limb,overlapping with 52bpMsx2 transgene expression(Fig. 5A,E, arrows). Similarly,in the cardiac region, nuclear staining was apparent in pharyngeal arch mesenchyme and the outflow tract (Fig. 5H,J). Thus, in the pharyngeal region the 52 bp fragment was expressed in a subset of R-Smad positive cells(Fig. 5G,I).

The correlation between R-Smad nuclear localization and Msx2transgene expression suggested that Msx2 is an immediate early target of R-Smads. If so, R-Smads should associate with the Msx2 BMPRE in a BMP-dependent manner. To test this hypothesis, we carried out a chromatin immunoprecipitation (ChIP) analysis on C14 limb bud mesenchymal cells(Rosen et al., 1994). We chose these cells because of their tissue of origin and because their endogenous Msx2 mRNA is upregulated robustly by BMPs(Fig. 5K). An anti Phospho-Smad1 antibody was used to immunoprecipitate chromatin from BMP-treated and control cells. Semi-quantitative PCR showed that a 350 bp fragment encompassing the 52 bp BMPRE (Fig. 5L) was enriched several fold in chromatin derived from cells treated with BMP4 compared with untreated cells(Fig. 5M), consistent with the view that BMP treatment causes association of BMP R-Smads with the Msx2 BMPRE.

The 52 bp fragment contained a GC-rich partial inverted repeat flanking an AT-rich region (Fig. 2, Fig. 6A). Each repeat contained two GCCG sequences, which resembled the GCCGNCGC consensus sequence to which the Drosophila Mad protein binds(Kim et al., 1997; Raftery and Sutherland, 1999; Szuts et al., 1998; Xu et al., 1998). The GCCG sites flanked the sequence AGAGCAATTAACG, which matched closely the consensus site for several Antennapedia superclass homeodomain proteins, as well as paired class homeodomain proteins(de Jong et al., 1993; Florence et al., 1991; Gehring et al., 1994; Laughon, 1991) (MatInspector 7.0, 2003). As homeodomain proteins are candidates for tissue- and stage-specific modulators of the BMP response(Marty et al., 2001), the AATTAA site was of particular interest.

To test the function of the GCCG and AATTAA sequences in vivo, we produced two mutant 52bpMsx2-hsplacZ transgenes. One contained mutations in the four GCCG sequences, the other a mutation in the AATTAA site(Fig. 6A). Gel shift experiments verified that mutations in the GC-rich inverted repeats significantly reduced binding by the Smad4 MH1 domain, a DNA-binding motif conserved among several Smad proteins(Massague, 1998). Low protein inputs produced a single complex, higher inputs an additional, more slowly migrating complex (Fig. 6B). The latter complex probably resulted from the binding of additional Smad4 molecules. Mutation of the four GCCG sequences abolished the slower migrating(2°) complex and reduced substantially (60-90%) the faster migrating(1°) complex (Fig. 6B,C). That the primary complex was not abolished may have been due to an additional GCCG sequence located 3′ of the inverted repeat, or to several GNCT motifs, both of which might bind Smad4(Fig. 6A)(Ishida et al., 2000; Johnson et al., 1999; Kusanagi et al., 2000; Zawel et al., 1998). Despite this residual binding of Smad4 in vitro, the transgene bearing mutations in the four GCCG sites (7 independent lines) was expressed at much lower levels and in fewer regions than its non-mutated counterpart(Fig. 6E,F; Table 1). In addition, the mutant transgene did not respond to BMP4-soaked beads implanted on the limb bud (Fig. 6I) or mandible (data not shown).

Protein titrations showed that oligonucleotides containing the TTAATT site and flanking sequence were capable of binding the homeodomain proteins Prx1b(Fig. 6D) and Msx2 (data not shown). A mutation in the TTAATT site (Fig. 6A) reduced the binding of Prx1b to an undetectable level(Fig. 6D), prompting us to test this same mutation in vivo. Analysis of 14 independent lines revealed a dramatic reduction in transgene expression in the limb bud and in most other sites (Fig. 6E,G; Table 1). Bead implantations in the limb bud showed that the mutant transgene was not BMP responsive(Fig. 6J). This profound change in the expression of the 52bpMsx2 transgene led us to test the role of the TTAATT element in a larger genomic context. We used the 560 bp fragment because it contained all or most of the cis-regulatory information needed for expression and BMP responsiveness of Msx2(Kwang et al., 2002).

In 10T1/2 cells, the homeodomain site mutation resulted in near-complete loss of BMP-inducibility of a transfected 560 bp construct(Fig. 6K), similar to the effect of deleting the entire 52 bp sequence in the 560 bp fragment(Fig. 3I). Analysis of six transgenic lines bearing 560bpMsx2-hdm-hsplacZ mutant transgenes showed that the mutation resulted in a general reduction of transgene expression (compare Fig. 6Mwith Fig. 3H) as well as a site-selective loss of expression (Table 1). For example, expression was reduced substantially in limb bud mesenchyme but retained in the AER (Fig. 6M, arrow). Bead implantations showed that the mutation abolished BMP-responsiveness in limb bud mesenchyme of E11.5 embryos(Fig. 6O). We conclude that the TTAATT element is crucially important for BMP responsiveness of Msx2both in cultured cells and in a subset of sites in murine embryos.

That the homeodomain and Smad consensus sites are required coordinately for the function of the Msx2 BMPRE in murine embryos, and that the sequence of the BMPRE is highly conserved among mammalian groups, raised the question of whether this combination of homeodomain and Smad consensus sites reflects an ancient mechanism for BMP-dependent transcriptional activation. Cis-regulatory elements can undergo mutational turnover yet maintain their function (Ludwig et al.,2000). Thus, despite the apparent lack of sequence conservation of the BMPRE outside mammals, the possibility remained that it might function in a more diverse group of organisms.

To test this idea, we asked whether the Msx2 BMPRE was capable of responding to Dpp signals in Drosophila. We chose Drosophilabecause its distant relation to mouse would provide a stringent test of the hypothesis that the BMPRE can function over a large phylogenetic distance. In addition, a large number of mutants in components of the Dpp pathway are available, enabling us to rigorously test whether the murine element was responding appropriately to Dpp signals.

We produced transgenic flies bearing the 560, 480, 220 and 52 bp sequences driving nuclear-localized lacZ. We examined expression in imaginal discs and embryos, both in wild-type flies and in mutants in which Dpp signaling was perturbed. Examples of transgene expression are shown in Fig. 7. The 480bpMsx2-lacZ transgene was expressed in wing imaginal discs in a pattern that closely resembled that of vestigial, a known Dpp target(Fig. 7B,C). Ectopic activation of Dpp signaling with A9Gal4>TkvA resulted in a corresponding expansion of Msx2 transgene expression(Fig. 7D). In stage 13 embryos,the 220bpMsx2-lacZ transgene was expressed in parasegments (ps) 3 and 7 of embryo visceral mesoderm in a pattern closely matching that of dpp (Fig. 7E,G). This pattern was expanded throughout the gut in embryos in which dppexpression was driven ectopically by a heat shock promoter(Fig. 7I). In dpp(S11/S22) regulatory mutants, in which dpp signaling is lost specifically in ps3 (Fig. 7F,arrow), the 220bpMsx2-lacZ transgene was also downregulated in ps3(Fig. 7H, arrow).

The 52 bp sequence drove expression on the dorsal side of early blastoderm embryos in a pattern that matched Dpp signaling(Fig. 7J,K). Expanded transgene expression was evident in dorsal mutant embryos(Fig. 7L), consistent with the finding that dorsal represses dpp transcription in ventral cells,thus restricting Dpp signaling to the dorsal region of the embryo(Huang et al., 1993). Expression was also lost in screw mutant embryos(Fig. 7M), in which levels of Dpp signaling are reduced (Arora et al.,1994). Conversely, ectopic activation of Dpp signaling using Tub Gal4>UAS Dpp resulted in expansion of transgene expression in the dorsal half of the embryo (Fig. 7N). Similarly, embryos mutant for brinker, a repressor of Dpp signaling (Jazwinska et al.,1999), exhibited expanded lacZ activity(Fig. 7O).

When we tested the expression of 52bpMsx2-lacZ transgenes bearing the same GCCG or TTAATT site mutations used for transgenic mice, we found that each mutation resulted in a profound loss of transgene expression(Fig. 7P,Q) (five independent lines each). Similar results were apparent in stage 13 embryos bearing these mutant transgenes (data not shown). Together, these data show: (1) that sequences within the Msx2 promoter can respond to Dpp signaling in Drosophila embryos and wing imaginal discs; and (2) that both TTAATT and GCCG sites are crucial for expression of the Msx2 BMPRE in Drosophila embryos as they are in mouse embryos.

Although several BMP responsive elements have been characterized in cultured cells (Benchabane and Wrana,2003; Henningfeld et al.,2000; Hullinger et al.,2001; Jonk et al.,1998), few have been studied in a developing mammalian embryo(Liberatore et al., 2002; Lien et al., 2002; Theil et al., 2002). The Msx2 BMP-dependent enhancer is of particular interest because of the importance of Msx2 in the BMP pathway and because Msx2 is frequently employed as a readout of BMP action, despite the fact that the mechanisms controlling its BMP responsiveness have not been elucidated.

A long-term goal in the analysis of the Msx2 BMPRE has been to identify trans-regulators that cooperate with Smads to modulate BMP responsiveness in different developmental settings. We provide evidence here that a homeodomain consensus site is required, together with Smad consensus sites, for BMP responsiveness of Msx2 transgenes, and that these sites function together with other cis-regulatory elements to control differential BMP responsiveness and expression in subregions of limb mesenchyme and in other structures in the developing embryo. That the consensus homeodomain site plays such a major role in the BMP responsiveness of the Msx2 promoter was unexpected. Although Smad sites are typically found to be functionally crucial in vertebrate BMP responsive elements (Lopez-Rovira et al.,2002; Park and Morasso,2002), similar crucial roles for homeodomain sites have not been documented. We note, however, that Smad1 interacts with Hoxc8 during activation of the osteopontin promoter(Shi et al., 1999) and that in Xenopus embryos, the Xvent-2 homeodomain protein can act as a Smad1-specific co-activator during maintenance of its own transcriptional regulation (Henningfeld et al.,2002).

It is clearly of interest to know the identities of proteins that interact with the TTAATT site in the Msx2 BMPRE. Homeodomain-containing proteins of the Nk (including Msx), paired and Hox classes are capable of binding this element in vitro (Fig. 6; S.M.B. and R.M., unpublished). Representatives of each class are expressed in patterns consistent with a role in the regulation of Msx2, though genetic proof that any such proteins participate in the BMP-dependent activation of Msx2 is lacking. Interestingly, Xenopus Xmsx-1 has been shown to physically associate with pathway-restricted Smads and Smad4(Yamamoto et al., 2001),suggesting that like Xvent2, Msx proteins may participate in an autoregulatory loop.

In addition to homeodomain-containing proteins, members of the Fox family of forkhead/winged helix transcription factors have emerged as candidate regulators of the BMP-responsiveness of Msx2. Foxc1 is required for the BMP-dependent induction of Msx2 in calvarial tissues of murine embryos (Rice et al., 2003). Inspection of the sequence around the TTAATT site revealed a close match (5/7)to a consensus derived for Hnf3/Fox class proteins(Fig. 2)(Gao et al., 2003). Whether Fox proteins can bind this sequence, and, if so, whether such binding is disrupted by mutations in the TTAATT site are unanswered questions. It is intriguing that Fox proteins can form heterodimers with homeodomain-containing proteins (Foucher et al.,2003), and that the Fox protein Fast1 can interact with Smads and modulate the activity of TGFβ-responsive promoters(Chen et al., 1996; Chen et al., 1997; Weisberg et al., 1998).

Deletion analysis showed sequences flanking the 52 bp element control the location, sensitivity and timing of BMP responsiveness in limbs and other sites in mouse embryos. Within these sequences are consensus binding sites for Creb, OAZ and Tcf/Lef1 (Fig. 2), each implicated in BMP signaling in widely divergent organisms(Hata et al., 2000; Ionescu et al., 2004; Theil et al., 2002). In the Drosophila Ubx promoter, for example, Dpp responsiveness in parasegments 3 and 7 of the embryonic midgut depends on a CRE site acting together with Mad sites (Waltzer and Bienz, 1999). In addition, Tcf/Lef1 sites located 270 and 420 bp upstream (at positions –4130 and –4280) of theΔ 4Msx2 fragment, participate in BMP responsiveness in ES cells(Hussein et al., 2003). Mutation of these sites in mouse embryos will provide insight into their roles in modulating the BMP responsiveness of Msx2 in different tissues and cell types. We showed previously that Pax3 acts through a site in the 560 bp fragment to repress Msx2 expression in the dorsal neural tube(Kwang et al., 2002). We have not yet identified sequences required for upregulated neural tube expression of Msx2 in the absence of Pax3 function, but it is possible that these sequences include the Msx2 BMPRE. Finally, we note that YYI has been implicated in the activation of Msx2 expression in craniofacial structures, but in a manner independent of the BMP pathway(Tan et al., 2002).

In Drosophila, brinker and schnurri are crucial for the Dpp-dependent expression of a number of genes. Brinker (Brk) acts through GC-rich sites to repress target gene expression(Rushlow et al., 2001; Sivasankaran et al., 2000). Schnurri represses brinker expression in a Dpp-dependent manner(Marty et al., 2000; Torres-Vazquez et al., 2000). That Msx2 transgene expression in Drosophila embryos expands in a brinker mutant is consistent with results showing that Brk acts through Mad-like binding sites for its repressive function(Kirkpatrick et al., 2001; Saller and Bienz, 2001), and with our findings that the Msx2 BMPRE is probably a direct target of Smad1 (Fig. 5). Our demonstration that mutations in Mad/Smad1-binding sites cause loss of transgene expression in Drosophila, together with the expansion of transgene expression in the brinker mutant, suggest that the Msx2 BMPRE is subject to positive and negative regulation in Drosophila, as expected for direct targets of Mad.

Whether Msx2 transgenes are similarly regulated by positive and negative inputs in mouse embryos is not clear. Orthologs of brinkerhave not been found in vertebrates. However, Zeb-2, a zinc-finger transcription factor, has been identified as a negative regulator of BMP signaling (Postigo, 2003). This protein binds to a GC-rich sequence and can interact with Smads(Verschueren et al., 1999), as well as inhibit expression of the BMP target, Xmsx1, in Xenopus (Postigo et al.,2003).

Despite the lack of demonstrable sequence homology between the Msx2 BMPRE and Dpp-responsive enhancers in flies, our transgenic experiments show that the Msx2 element provides an accurate readout of Dpp activity in Drosophila. This is not merely a result of the activity of Mad, as a mutant transgene bearing a disabled TTAATT sequence but intact GCCG consensus sites was not expressed in Drosophila embryos. We note that the requirement of the TTAATT site for BMP responsiveness of Msx2 transgenes in mice and flies is reminiscent of the dependence of the labial, tinman and even-skipped Dpp-activated enhancers of Drosophila on the homeodomain proteins Labial and Tinman(Knirr and Frasch, 2001; Marty et al., 2001; Xu et al., 1998).

That several BMP-responsive enhancers spanning a wide phylogenetic distance exhibit a dual requirement for homeodomain and Smad sites suggests synergistic interactions between homeodomain and Smad proteins is an ancient feature of the BMP/Dpp pathway. Such interactions may represent a general mechanism for integrating BMP/Dpp signaling inputs with tissue-specific transcription factors.

Few cis-regulatory elements are known to function over an evolutionary distance comparable with that separating mammals and insects. One such element is the eye enhancer of the Drosophila eyeless gene, a Pax6ortholog. In transgenic mice, this enhancer directs expression that partially recapitulates that of endogenous Pax6(Xu et al., 1999). Another is the murine Hoxa2 rhombomere 2-specific enhancer, which is expressed in head segments of transgenic flies, paralleling the expression of the Hhoxa2 ortholog, proboscipedia(Frasch et al., 1995). Similarly, a Hox4b cis-regulatory element mediates correct spatial expression in Drosophila (Malicki et al., 1992), and the orthologous Drosophilaautoregulatory element from the homeotic gene, Deformed, functions appropriately in mice (Awgulewitsch and Jacobs, 1992).

These examples notwithstanding, careful comparative analyses within nematode and echinoderm lineages suggest that cis-regulatory elements usually evolve rapidly, becoming non-alignable within 35-50 million years(Ruvinsky and Ruvkun, 2003; Wray et al., 2003). Intriguingly, such divergent promoter elements frequently maintain their function, suggesting they are subject both to drift and stabilizing selection,which promote rapid sequence divergence while maintaining regulatory function(Ludwig et al., 2000; Ruvunski and Ruvkun, 2003; Wray et al., 2003). It seems probable that the ability of the Msx2 BMPRE to function in flies,despite its lack of significant sequence identity to known Dpp-responsive enhancers, can be explained by a similar process of drift and stabilizing selection. We suggest this process differs from that operating on a typical promoter element only in its slow pace, reflecting the fundamental and conservative role of the machinery that controls the transcription of crucial effectors of the BMP pathway.

This work was supported by NIH Grants DE12941 and DE12450 to R.M., and AR44528 to K.M.L.; by March of Dimes grant 6-F402-135 to K.A.; and by CRCC grant 33348 to R.W. S.B. was supported by NIH, National Research Service Award T32HL07895, from the National Heart, Lung and Blood Institute. A.M. was supported by a grant 0010040Y from the American Heart Association. We thank Youzhen Yan and Yazmin Santiago for excellent technical assistance.