Schnurri transcription factors from Drosophila and vertebrates can mediate Bmp signaling through a phylogenetically conserved mechanism

Bone Morphogenetic Proteins (Bmps) are secreted growth factors of the Tgfβ superfamily that regulate a wide range of biological processes,including specification of embryonic axes, cell fate determination,proliferation and apoptosis. Bmp ligands and receptors have been identified in organisms from sponges and Cnidaria to humans, and elicit their effects in a similar fashion across phyla (Holstein et al., 2003; Massague et al.,2005). Signaling is initiated by ligand binding to a complex of type-I and type-II transmembrane serine/threonine kinase receptors, enabling the type-II receptor to phosphorylate the type-I receptor. The type-I receptor in turn phosphorylates a receptor-regulated member of the Smad (RSmad) family of intracellular signal transducers. In vertebrates, Bmp signaling is mediated by Smad1, Smad5 and Smad8, whereas in Drosophila the sole Bmp-specific Smad is Mothers against dpp (Mad). The activated R-Smad associates with the co-Smad, Smad4 [Medea (Med) in Drosophila],forming a heteromeric complex that translocates into the nucleus. R-Smads and co-Smads are structurally similar, possessing amino-terminal MH1 and carboxyterminal MH2 domains separated by a flexible linker. A conservedβ-hairpin motif in the MH1 domain mediates DNA binding, enabling the R-Smad/Smad4 complex to direct transcriptional responses. Co-Smads recognize the sequence GTCT (the `Smad-binding element' or SBE), whereas Smad1 and Mad preferentially bind GCCGNC (Gao et al.,2005; Johnson et al.,1999; Kim et al.,1997; Xu et al.,1998). Smads bind to these low complexity sites with weak affinity and it is thought that Smad-DNA interactions are enhanced by partnership with other transcription factors that are crucial for promoter selectivity. In support of this idea, reporter constructs containing multimers of SBEs or Smad1 sites alone are incapable of replicating the expression patterns of endogenous Bmp target genes in vivo, although they confer Bmp responsiveness in cell culture (Shi and Massague,2003; von Bubnoff et al.,2005).

Drosophila Schnurri (Shn) is a large DNA-binding transcription factor with multiple, widely separated zinc finger domains, and was one of the first partners identified for Bmp-specific R-Smads(Dai et al., 2000; Udagawa et al., 2000). Shn localizes to the nucleus and interacts with Mad and Med in response to Dpp signaling. Genetic and phenotypical analysis has established that shnis essential for Dpp signaling at many developmental stages and in diverse tissues (Arora et al., 1995; Grieder et al., 1995; Staehling-Hampton et al.,1995). At the molecular level, Shn uses two modes to regulate Dpp target genes: transcriptional activation and repression. Shn can directly activate a small subset of Dpp-responsive genes, but exerts most of its effect through a relief-of-repression mechanism. Shn inhibits the expression of a transcriptional repressor Brinker (Brk) that, in turn, negatively regulates the expression of most Dpp target genes. Therefore, Dpp signaling leads to the inhibition of brk expression and the derepression of batteries of Dpp-responsive genes (Affolter et al.,2001; Marty et al.,2000; Torres-Vazquez et al.,2001; Torres-Vazquez et al.,2000). Shn represses brk via the formation of a Shn/Mad/Med complex on defined elements in the brk promoter(Muller et al., 2003; Pyrowolakis et al., 2004). Given the prominent role of Shn in Dpp signaling, an inevitable question is whether the conservation encountered at other levels of the pathway extends to the cis-regulatory elements and nuclear factors that mediate the transcriptional response to Bmps. Furthermore, brk, a primary target of Shn activity, is absent from genomes outside the arthropods, raising the possibility that the involvement of Shn in Bmp signaling may be unique to insects.

Unlike brk, Shn-related genes are found in other phyla. In C. elegans the lone Shn homolog SMA-9 has been implicated in Bmp signaling,although its mechanism of action is unknown(Liang et al., 2003). Three Shn homologs are found in vertebrates that also contain widely spaced, paired zinc finger domains with high sequence similarity to the fly protein. Shn proteins share additional features, including an unusually large size, overall structural organization and the presence of acidic domains. Human members of this family have been variously referred to as Shn1/HIVEP1/MBP-1/PRDII-BF1/ZAS1, Shn2/HIVEP2/MBP-2/ZAS2 and Shn3/HIVEP3/KRC/ZAS3 (Liang et al.,2003; Wu, 2002). No direct evidence ties vertebrate Shn1 or Shn3 to the Bmp pathway, but Shn2 has recently been shown to promote Bmp-responsive transcription of the PPARγ gene in mouse embryonic fibroblasts. However, Shn2 acts through sites that bear no resemblance to the sequences recognized by the Drosophila Shn/Smad complex, and requires co-operativity with the transcription factor C/EBPα (Jin et al., 2006). Thus, whether the mechanism of action of Shn proteins in Bmp signaling is phylogenetically conserved remains an unanswered question.

Here, we demonstrate an unusual and unexpected conservation of the cis- and trans-regulatory elements involved in the response to Bmp signaling. We found that an element from the Xenopus Vent2 (Xvent2) promoter that confers sensitivity to Bmp signaling in vertebrate embryos also directs a Dpp-dependent response in Drosophila. However, although the element mediates transcriptional activation in Xenopus, it directs transcriptional repression in Drosophila. We demonstrate that the closest human Shn homolog, hShn1, can mediate signaling-dependent transcriptional activation through this element in vertebrate cells,implicating it in Bmp signal transduction. Remarkably hShn1 can also mediate signaling-dependent repression in Drosophila and rescues patterning defects in shn mutant embryos, whereas fly Shn can activate transcription in Xenopus assays. This switch in transcriptional output between repression and activation is not a function of differences in the cis-regulatory elements or an inherent property of the human and Drosophila proteins, indicating that Shn/Smad complexes are likely to recruit co-activators and co-repressors in a context-dependent manner.

Grh-lacZ (provided by Scott Barolo) contains three 8-bp Grh sites upstream of lacZ in pH-Pelican(Barolo et al., 2000). Xvent2-BRE-lacZ contains five 23-bp Xvent2-BREs between NotI and BamHI sites in Grh-lacZ. The 6.0 brk-lacZ transgene contains -8.3 to -2.7 kb of the brkpromoter driving lacZ (L.-C.Y., unpublished). All other stocks are described in FlyBase. Embryos and imaginal discs were stained as described(Torres-Vazquez et al.,2000).

Full-length Shn and hShn1 (Fan and Maniatis, 1990) were subcloned into pUAST(Brand and Perrimon, 1993). shnP4738 is a null allele (Arora et al., 1995). Progeny from UAS-Shn/UAS-hShn1,shnP4738/CyO flies crossed to shnP4738/CyO; HS-Gal4were aged to 8-10 hours and received two 1-hour heat shocks at 37°C,separated by 25°C rest periods. brk expression was monitored using the enhancer trap X-47(Campbell and Tomlinson, 1999). Whole-mount cuticles were prepared as described(Arora et al., 1995). Homozygous mutants were identified by the absence of wg-lacZ or CyO, Kr>GFP.

pXvent2(-150)LUC and pXvent2-5×BRE(-150)LUC have been described previously (von Bubnoff et al., 2005). The oligonucleotide 5′-CGGCAGACAGGTTGGAGCCAGCTCGGCAGACAGGTTGGAGCCAGCT-3′containing a BRE dimer (mutations underlined) was used to generate a PCR product that was subcloned into pXvent2(-150)LUC to create pXvent2-5×BRE(sub2)(-150)LUC. pXvent2-6×BRE(del2)(-150)LUC was similarly created using 5′-CGGCAGAC_GGTGGAGCCAGCTCGGCAGAC_GGTGGAGCCAGCT-3 (underscores indicate deleted bases). pXvent2-5×BRE(-150)GFP3,pXvent2-5×BRE(sub2)(-150)GFP3 and pXvent2-6×BRE(-150)(del2)GFP3 contain BamHI-HindIII fragments from the respective reporter plasmids in pCXGFP3 (a gift of Enrique Amaya, University of Manchester). Full-length Shn and hShn1 were subcloned into pBSII-KS, containing the SV40 polyadenylation signal. Capped mRNAs were synthesized using the T3 mMessage Machine Kit (Ambion), including 1 μl GTP/20 μl reaction, diluted into single use `injection cocktails' containing reporter genes, stored at-80°C and used within 24 hours.

mRNA concentrations/embryo were: luciferase reporters, 200 pg;dominant-negative (Suzuki et al.,1994) and constitutively active(Candia et al., 1997) BmpR,0.1-2.0 ng; and hShn1 and Shn, 2 ng. Wild-type and mutant BRE reporters were microinjected into the animal poles of two- to four-cell stage embryos and incubated until blastula stages 8-9. Mid-blastula stage animal caps were dissected and homogenized for luciferase assays when sibling embryos reached stages 10.5-11 (Watabe et al.,1995). In situ hybridization and generation of transient transgenic embryos using the sperm nuclear transfer technique have been described previously (von Bubnoff et al.,2005).

N-terminal FLAG-Mad and FLAG-Med, TkvA and ShnCT plasmids have been described previously (Gao et al.,2005). PCR fragments 1 (hShn1^1-599), 2(hShn1^1-702), 3 (hShn1^496-1121), 4(hShn1^1002-1635), 5 (hShn1^1496-2213), 6(hShn1^1756-2544) and 7 (hShn1^1756-2717) were subcloned into the pAWM vector containing an Actin5C promoter and C-terminal 6×Myc epitopes (a gift of T. D. Murphy, Carnegie Institution, Baltimore). Extracts from Drosophila S2 cells transfected with individual hShn1 constructs and TkvA/Mad/Med or TkvA/hSmad1/hSmad4 were incubated with anti-FLAG antisera. Immunoprecipitates were run on 4-12% gradient SDS polyacrylamide gels before western blotting and visualization of interacting protein with anti-Myc. Gel shift assays were performed as described(Gao et al., 2005; Kirov et al., 1994; Pyrowolakis et al., 2004). Extracts from Drosophila S2 cells transfected with different plasmid combinations were used to shift a Xvent2 BRE probe(5′-CTAAGAGCTGGCTCCACCATGTCTGCCGTTAGTTGGCTCA-3′). For supershifts, 60 ng of anti-FLAG (M2, Sigma) or 100 ng of anti-MYC (9E10,Santa Cruz) were added to the binding reactions. GST pull-downs were carried out as described (Dai et al.,2000). Human Smad1 and Smad4 linker+MH2 regions subcloned in pGEX-4T-1 were expressed in E. coli BL21. PCR fragments of hShn1 were subcloned into pCITE4A (Novagen) before in vitro translation using the TNT-T7 Coupled Transcription/Translation Kit (Promega).

Shn proteins used were: D. melanogaster DMU31368; X. tropicalis Shn1 genome v4.1 gw1.33.114.1, Shn2 gw1.274.19.1, Shn3 ensemble.c_scaffold_ 478000040; Mouse Shn1 NM_007772, Shn2 NM_010437, Shn3 NM_010657; Human Shn1 NM_002114, Shn2 NM_006734, Shn3, NM_024503. CLUSTALW analysis was performed using the San Diego Supercomputer Center's Biology Workbench.

To determine whether conservation in the Bmp pathway extends to the cis-regulatory elements and transcription factors involved in the nuclear response, we examined whether a vertebrate Bmp response element (BRE) is sensitive to Bmp signaling in Drosophila. The Xenopus Vent2gene is one of the best-characterized direct transcriptional targets of Bmp signaling in vertebrates, and is essential for patterning ventral mesoderm and preventing neuralization in the embryo(Onichtchouk et al., 1996). Classical promoter analyses have identified a 53-bp Bmp responsive module located at -191 relative to the initiator ATG(Candia et al., 1997; Hata et al., 2000; Rastegar et al., 1999). Phylogenetic footprinting and in vivo reporter gene analyses have further refined the ciselement to a 23-bp BRE that is both necessary and sufficient for Bmp-dependent activation in animal cap ectodermal explants and transgenic embryos (von Bubnoff et al.,2005). To test whether the 23-bp Xvent2-BRE can mediate Bmp responsiveness in Drosophila, we generated transgenic lines containing five tandem BREs upstream of lacZ in the 3×Grh vector (henceforth referred to as Xvent2-BRE-lacZ; Fig. 1A). As the Dpp pathway utilizes both direct and indirect relief-of-repression mechanisms to activate gene expression, we chose a single vector that would allow us to assay both outputs. 3×Grh-lacZ contains binding sites for Grainyhead(Grh), a transcriptional activator ubiquitously expressed in the ectoderm after embryonic stage 11 (Bray and Kafatos,1991; Furriols and Bray,2001). Therefore, 3×Grh-lacZ is transcriptionally silent until extended germ band stage, but then drives uniform lacZexpression through later embryogenesis and in larval tissues(Fig. 1B-E). Thus, enhancers mediating activation can be assayed prior to stage 11, whereas elements mediating repression can be detected at later stages.

Dpp signaling is crucial in early embryogenesis for the specification of dorsal cell fates and is detected by stage 4 (syncytial blastoderm) in the dorsal embryo (Dorfman and Shilo,2001). Subsequently, dpp is involved in many other aspects of embryonic and larval patterning(Raftery and Sutherland,1999). We analyzed Xvent2-BRE-lacZ transgenic lines and found no reporter gene expression prior to stage 11, suggesting that this element cannot direct Dpp-dependent activation in early embryos (data not shown). However, from stage 11 onwards β-galactosidase could be detected in a distinct pattern in both embryos and imaginal discs(Fig. 1F-I). Interestingly, the reporter was downregulated where Dpp signaling is active(Dorfman and Shilo, 2001; Sutherland et al., 2003; Tanimoto et al., 2000). In stage 13 embryos (germband retraction), dpp is expressed in the leading edge of the dorsal ectoderm and in the boundary between dorsal and ventral ectoderm, both sites where reporter expression is absent (compare Fig. 1F with 1N). Likewise, lacZ is repressed in the medial cells of the wing imaginal disc, and regions of eye-antennal and leg discs subject to Dpp signaling (compare Fig. 1G-I with 1O-Q). At all locations, the spatial and temporal pattern of reporter expression closely replicates that of brk, a negatively regulated target of Dpp signaling (Fig. 1J-M). These remarkable parallels strongly suggest that the Xvent2-BRE, which activates transcription in vertebrates, mediates transcriptional repression in response to Bmp signaling in Drosophila.

During the course of these experiments, an analysis of the Drosophila brk gene revealed that its expression pattern results from two inputs:ubiquitous activation and Dpp-dependent repression(Muller et al., 2003). Repression elements in the brk promoter and several other direct Dpp targets contain a GRCGNC(N5)GTCTG consensus, with binding sites for both Mad and Med separated by a spacer of 5 bp (see Fig. 2A). This precise spacing is required for Shn to dock to Mad/Med complexes on the DNA and mediate repression (Gao et al., 2005; Pyrowolakis et al., 2004). A comparison between the Xenopus BRE and the Drosophilaconsensus sequence shows a near perfect match (see Fig. 2A). Although this finding suggested a simple explanation for the ability of the Xenopus BRE to function in Drosophila, there is a single nucleotide difference between the BRE and the Brk consensus (G→T at position 4) that has been reported to prevent Mad/Med binding and formation of the repression complex in flies (Pyrowolakis et al.,2004). This change in a crucial residue raised the possibility that the sequence similarity was purely coincidental and that the BRE-dependent expression pattern arises through a mechanism that is independent of Shn. To test whether BRE function requires Shn, we examined Xvent2-BRE-lacZ expression in shn^-embryos. Xvent2-BRE-lacZ is derepressed throughout the ectoderm from stage 11 onwards, indicating that Shn is crucial for restricted BRE expression (data not shown). Next, we tested whether the XenopusBRE can nucleate assembly of a Shn/Mad/Med complex required for repression. Drosophila S2 cells were transfected with constitutively activated Dpp type-I receptor Thickveins (TkvA), epitope-tagged Mad, Med, and a C-terminal fragment of Shn (ShnCT) that has been shown to participate in complex formation (Pyrowolakis et al.,2004). Radiolabelled BRE oligonucleotides were incubated with cell extracts and subjected to gel electrophoresis. We found that Mad and Med decreased the mobility of the BRE fragment(Fig. 1R, lane 2), and addition of Shn further decreased its migration(Fig. 1R, lane 4). Furthermore,both Mad/Med and the Shn/Mad/Med complexes could be supershifted by antisera against FLAG and Myc epitopes on the Smads and Shn, respectively (compare lane 2 with 3, and 4 with 5), confirming the presence of these proteins in the shifted bands. The ability of the Xvent2-BRE to assemble a Shn/Mad/Med complex despite the difference in sequence between the Drosophila and Xenopus elements may be due to compensatory effects from adjacent or flanking residues. This biochemical analysis indicates that the Xvent2-BRE can mediate Bmp signaling-dependent transcriptional repression through a mechanism similar to that used by the brk element.

The similarities between the Xenopus BRE and the Drosophila brk element, both in sequence and in the ability to assemble a signaling-dependent repression complex, raise an intriguing question. Do these parallels reflect an underlying conservation of the Bmp-responsive transcriptional regulators? To address this issue, we first tested whether mutations that abolish repression by the brk element in Drosophila correspondingly alter the ability of Xvent2-BRE to respond to Bmp signaling in Xenopus. Substitutions of individual residues in the Mad and Med sites of the brk element interfere with complex formation and affect repression in vivo(Pyrowolakis et al., 2004). In previous studies, we and others have shown that Smad4 binds the BRE and that mutations that compromise the stimulation of Xvent2 reporters map to regions corresponding to the Smad sites(Hata et al., 2000; Henningfeld et al., 2000; Karaulanov et al., 2004; von Bubnoff et al., 2005). A distinguishing feature of the Drosophila element is the requirement for an invariant spacing between the Mad and Med sites that is crucial for the recruitment of Shn to the Smad-DNA complex(Pyrowolakis et al., 2004). If a similar complex binds the Xenopus BRE, the spacer sequences could serve as docking sites for a Smad co-factor, such as a vertebrate Shn (see Fig. 1R). We therefore generated BRE-luciferase reporters that alter either the length or the sequence of the Xvent2 spacer and tested their ability to respond to Bmp signaling in Xenopus animal cap assays(Fig. 2B, see Materials and methods). The wild-type Xvent2-BRE drives high-level luciferase expression in response to endogenous Bmps that is further increased upon co-injection of a constitutively active Bmp receptor (CABR), demonstrating that the reporter responds to changes in activity of the Bmp pathway(Fig. 2C). Substitution of 2 bp within the spacer without altering its length (Xvent2-sub2), does not affect Bmp responsiveness (Fig. 2A,C). By contrast, shortening the spacer by deleting 2 bp(Xvent2-del2) rendered the reporter insensitive to stimulation(Fig. 2A,C). These results demonstrate that the Xvent2-BRE responds to Bmp stimulation in a manner identical to that observed for the brk element in Drosophila.

To examine whether the response to endogenous Bmp signaling places similar constraints on BRE architecture, we assessed the activity of these constructs in transgenic Xenopus (Kroll and Amaya, 1996; von Bubnoff et al., 2005). Wild-type and mutant BRE multimers were cloned upstream of the Xvent2 minimal promoter driving GFP, and reporter gene expression was examined by in situ hybridization. The wild-type BRE drives expression at several sites where Bmp4 is transcribed, including the eye, otic vesicle, ventral branchial arches, and the proctodeal region(Fig. 2C)(Karaulanov et al., 2004; von Bubnoff et al., 2005). The Xvent2-sub2 mutant drives a similar expression pattern. By contrast,transgenic embryos carrying the Xvent2-del2 mutant show random expression patterns, presumably as a result of integration site-dependent position effects. These data reinforce the idea that in vivo (as in animal cap assays) maintenance of an invariant spacer length is crucial for the BRE to respond positively to Bmp signaling in Xenopus. Thus, BRE function has the same architectural requirements in Xenopus and Drosophila, and this constraint on the organization of the element suggests that the recruitment of a Smad-interacting protein analogous to Shn could be essential for BRE activity in vertebrates.

Shn proteins have high homology in the paired C2H2 zinc finger domains but show limited sequence identity outside of these regions(Fig. 3A) (reviewed by Wu, 2002). Generation of a molecular phylogeny of fly, worm, frog, mouse and human proteins indicates that Drosophila Shn is most closely related to vertebrate Shn1(Fig. 3B). Therefore, to investigate whether Shn proteins play mechanistically similar roles in the Bmp pathway, we examined whether human Shn1 (hShn1) could stimulate the BRE-mediated response to Bmp signaling in animal cap assays(Fig. 3C). We found that expression of hShn1 alone resulted in an approximately 4-fold stimulation of the reporter that was further enhanced to 16-fold in the presence of an activated Bmp receptor. This result suggests that vertebrate Shn-related proteins could play an important role in mediating Bmp signaling.

Because the Xenopus BRE was Dpp responsive in Drosophilaembryos, we were interested in determining whether Drosophila Shn can also elicit a response in Xenopus animal caps. We found that Shn alone did not significantly stimulate the reporter(Fig. 3D). Remarkably,co-injection of Shn and activated Bmp receptor significantly elevated reporter activity (Fig. 3D). Because fly Shn is capable of participating in a Smad complex on the Xvent2-BRE in vitro (see Fig. 1R), these data suggest that Shn can function as a transcriptional activator in Xenopus embryos. This result is striking as, in Drosophila,Shn mediates repression via the same cis-element. This difference in modality suggests that whether Shn drives transcriptional activation or repression is context-dependent rather than an inherent property of the protein.

The functional similarities between hShn1 and Shn suggest that vertebrate Shn proteins act as Smad co-factors on the Vent2 BRE. However docking of Shn to the Mad/Med complex requires zinc fingers 6 and 8(Pyrowolakis et al., 2004),which are not represented in vertebrate Shn proteins (see Fig. 3A). However, we, and others, have shown that Drosophila Shn contains at least two additional Smad-interacting domains (Dai et al., 2000; Udagawa et al.,2000), raising the possibility that interactions through other regions contribute to hShn1-Smad complex formation on the Xvent2-BRE. We therefore sought to identify and localize potential Smad-interacting regions within hShn1 by co-expressing hShn1 fragments with hSmad1 and an activated Bmp receptor. Co-immunoprecipitation assays identified two separate domains (hShn1^1-599 and hShn1^1756-2544) that mediate Smad interactions (Fig. 4A). The ability of hShn1 to bind hSmad1 was independently confirmed in GST pull-down assays. As demonstrated in Fig. 4B, fragments hShn1^1-702 and hShn1^1756-2544bind GST-Smad1 strongly and GST-Smad4 with lower affinity, suggesting that Shn/Smad interactions are conserved in vertebrate proteins. To determine whether hShn1 binding to Smads results in the formation of a DNA-protein complex on the Xvent2-BRE, different hShn1 polypeptides were subjected to gel shift analysis using Drosophila Smads. We found that hShn1^1-599 and hShn1^1496-2213 could bind and supershift a BRE oligonucleotide (Fig. 4A,C, lanes 4 and 6). These polypeptides correlate well with fragments that tested positive in the co-immunoprecipitation and GST pull-down experiments, and suggest that the Smad1-interaction domains are located in the region of overlap. Collectively, these data strongly suggest that, despite the difference in modality, vertebrate Shn proteins use a conserved molecular mechanism to mediate the transcriptional response to Bmp signaling.

Our data provide strong evidence that parallel cis-regulatory elements and trans-acting factors can mediate Bmp-dependent activation of the Vent2 gene in Xenopus and Dpp-dependent repression in Drosophila. Because Drosophila Shn activates BRE-reporters in Xenopus embryos, but functions primarily as a repressor in flies,we sought to determine whether the converse is also true, by assaying the ability of hShn1 to substitute for Drosophila Shn in vivo. We focused on two Dpp-dependent events: down-regulation of brk expression in dorsal tissues and differentiation of dorsal epidermis in embryos. In the absence of Shn function, Dpp signaling cannot repress brktranscription and brk-lacZ is upregulated in all tissues(Fig. 5A,C). In addition, shn mutants display a `dorsal open' phenotype, as the dorsal epidermis fails to differentiate owing to impaired Dpp signaling(Fig. 5B,D). We generated transgenic lines containing hShn1 cDNA downstream of UAS sites and used a heat-shock Gal4 (HS>Gal4) driver to express hShn1 ubiquitously in shn<sup>-</sup> mutant embryos. We found that hShn1 effectively represses brk-lacZ in regions of the embryo that are subject to high levels of Dpp signaling (Fig. 5G). Thus, although hShn1 is required for transcriptional activation in vertebrate cells, it can direct signaling-dependent repression in a different context. The analysis of whole-mount cuticle preparations reveals that hShn1 expression allows the correct differentiation of the dorsal epidermis and the consequent rescue of the `dorsal hole' phenotype of shn mutants (Fig. 5H). Overall, the severity of the patterning defects is significantly reduced and the morphology of these animals is much closer to that of wild type. Notably,the extent of rescue by hShn1 is comparable to that encountered in control experiments with UAS-driven Drosophila Shn(Fig. 5E,F). The ability of hShn1 to compensate for the loss of Shn function in Drosophilademonstrates that the human protein can: (1) respond correctly to endogenous Bmp signals to regulate gene expression; (2) do so with enough fidelity to substantially restore embryonic patterning and development; and (3) mediate repression rather than activation depending on the cellular context. These data establish that hShn1 can act in Bmp signaling in the context of an organism, and suggest that it participates in an evolutionarily conserved mechanism for the transcriptional response to Bmp growth factors.

The Bmp pathway is an ancient highly conserved signaling module that performs diverse functions in different organismal contexts. Underscoring this conservation, Bmp ligands from humans and the coral Acropora can rescue patterning defects in Drosophila embryos lacking the Dpp ligand, and fly Dpp can trigger ectopic bone formation in rodents(Hayward et al., 2002; Padgett et al., 1993; Sampath et al., 1993). In this study, we show that both Drosophila Shn and its vertebrate homolog Shn1 can use a common mechanism involving interaction with Smads and similar cis-regulatory elements to transduce Bmp signals. Another important finding is that Shn and hShn1 can mediate either transcriptional activation or repression through identical cis-elements, depending on the cellular context. These results demonstrate that Shn proteins are not obligate transcriptional repressors but rather that they recruit co-activators or co-repressors in a cell-type specific fashion.

Our data indicate that a common set of simple parameters - the presence of Smad1 and Smad4 binding sites with a 5-bp separation - govern the ability of the Xvent2-BRE to respond to Bmp signaling in both vertebrate cells and in Drosophila. We establish that the BRE can assemble a Smad/Shn complex despite a G→T substitution at position 4 (see Fig. 2A). This was unexpected,as the change affects one of the two GNCN sites in the brk consensus site that are thought to bind Mad (Gao and Laughon, 2006; Gao et al.,2005). One possibility is that protein-protein interactions render the presence of both GNCN sites unnecessary. Additionally, it has recently been shown that an A→C substitution at position 5 in the brkelement enhances Mad binding (Gao and Laughon, 2006). Interestingly the Xvent2-BRE also contains a C at this position that could play a compensatory role and aid complex formation. Thus, our study suggests a modified consensus recognition site GRCKNC(N5)GTCTG for the Shn/Smad complex. The conservation of both sequence and function in the BRE is unexpected. For example, even skipped (eve) stripe 2 enhancers from the related species D. yakuba and D. melanogaster, which drive expression in similar patterns and are functionally equivalent, show considerable sequence divergence, suggesting that `turnover' in transcription factor binding sites and their relative spacing is likely to be common, with co-evolutionary changes being necessary to retain essential functions(Ludwig, 2002; Tautz, 2000). The stability of the Shn/Smad site differs from the eve element and is reminiscent of the cross-phyla conservation of enhancers in the Dorsal/NFκB and wound healing pathways in Drosophila and vertebrates(Gonzalez-Crespo and Levine,1994; Mace et al.,2005).

Incorporation of the Shn/Mad/Med sequence into target gene promoters appears to be a widely used strategy in Drosophila. Genome-wide searches reveal that in addition to brk, the motif is present flanking a number of genes that are downregulated by Dpp. In two instances, bag of marbles and gooseberry, functional studies have validated that the brk consensus is crucial for repression(Pyrowolakis et al., 2004). The extent to which the BRE confers Bmp responsiveness in vertebrates remains to be established. In addition to Xvent2, this element is precisely retained in the promoters of the related Xvent1B and Xvent2Bgenes, as well as in the human and zebrafish homologs VENTX2, Vox and Vent. The BRE consensus is also present in the Bmp-responsive enhancers of the id1, id2, id3 and id4 genes(Karaulanov et al., 2004; von Bubnoff et al., 2005). These elements in the id genes are also conserved between humans,rats, mice, fish and frogs, which strongly suggests that the BRE represents an ancient and functionally important motif. Limited searches have identified BRE-like sequences in several other Bmp-responsive genes (Q. Zeng and K.W.Y.C., unpublished), although more extensive bioinformatics combined with in vivo verification will be required to gauge the full spectrum of genes regulated by Shn proteins. It is likely that Shn proteins regulate only a subset of Bmp targets through BRE-like motifs, as close matches to the consensus are absent from several characterized Bmp-responsive promoters. This is not unexpected, as an important function of Smad-interacting cofactors would be the provision of cell, tissue and promoter specificity.

A previous study has implicated another Smad-interacting factor, OAZ, in activation of Xvent2 transcription(Hata et al., 2000). Although it remains possible that OAZ may contribute to Xvent2 expression in some tissues, the OAZ expression pattern is insufficient to explain all aspects of Xvent2 synexpression with Bmps. In addition, the OAZ site GCTCCA (that overlaps with the GRCKNC Smad1 motif) was retained in the Xvent2-del2 mutation that abolished Bmp responsiveness in animal caps and embryos (see Fig. 2), which suggests that OAZ may not have a crucial role in Xvent2regulation.

A striking feature of the Shn complex assembled at the BRE is that it leads to opposite outputs in Drosophila and Xenopus. In Drosophila, the brk element appears to exclusively mediate repression both in cell culture and in vivo(Gao et al., 2005; Muller et al., 2003). Repression is dependent upon recruitment of Shn to a Mad/Med complex, and this suggested that Shn might function as an obligate repressor(Pyrowolakis et al., 2004). Consistent with this view, a 1-bp increase or decrease in the brkspacer prevents Shn recruitment and changes the response from repression to activation in cell culture assays (Gao et al., 2005). The results presented in this work do not fit such a model. Although Shn/Mad/Med bound to the Xenopus BRE mediates repression in Drosophila (see Fig. 1), in Xenopus Shn elicits activation through the same response element (see Fig. 3D). Thus the crucial determinant of whether the complex mediates activation or repression appears to be the cellular milieu rather than the presence of Shn. Human Shn1 similarly behaves as a transcriptional repressor in flies and as an activator in Xenopus (see Fig. 3B, Fig. 4) further emphasizing the fact that the transcriptional polarity of Shn proteins may depend on differential interaction with co-repressors and co-activators(Fig. 6). Furthermore, there is evidence that Drosophila Shn need not be an obligate repressor in flies. We have previously shown that Shn can directly stimulate the expression of Ubx, a Dpp target gene, both in vivo and in cell culture(Dai et al., 2000) (see Fig. 6). Likewise, Shn proteins may not be obligate activators in vertebrate cells. Mouse Shn1 represses expression of collagen2α1 through the recruitment of SPEN, a protein that interacts with co-repressors SMRT/NCoR and HDAC(Yang et al., 2005). Thus, the outcome of Shn interactions with the BRE may not be a species-specific phenomenon. Instead, our observations suggest that the modality of Shn is regulated in a celltype specific manner. This has several ramifications. First, it identifies Shn co-activators and co-repressors as additional, and as yet uncharacterized, players in the Bmp pathway. Second, the ability of human Shn proteins to repress genes in Drosophila leads us to speculate that at least some of the co-repressors are evolutionarily conserved, and that vertebrate Shn proteins may also be involved in the downregulation of Bmp targets.

In Shn, zinc fingers 6 and 8 that mediate complex formation with Mad/Med and an adjacent 113 amino acid domain are essential for repression(Pyrowolakis et al., 2004). Surprisingly, hShn1 that does not contain either the Smad interaction domain or the `repression domain' can nevertheless effectively repress brk-lacZ expression in flies (see Fig. 5). As we have demonstrated, hShn1 can assemble a complex on the BRE through two independent Smad-interacting regions (see Fig. 4C), thus it appears that co-repressor recruitment may depend on structural motifs/charge interactions that are not apparent from the primary sequence.

Our results show that Shn1 can transduce a Bmp signal in Xenopusand Drosophila embryos, as well as in cell culture (see Figs 3, 5; S.P. and R.W., unpublished). The transcript distribution of Shn1 in Xenopus has significant overlap with, and could explain, the Bmp-dependent synexpression pattern of the Xvent2 gene from which the BRE was derived(Durr et al., 2004; Niehrs and Pollet, 1999; Onichtchouk et al., 1996). However, a complete understanding of the role of Shn proteins in vertebrates will require loss-of-function analyses. Recent studies of knockouts for the closely related Shn2 gene has shown that it contributes to Bmp-responsive transcriptional activation of PPARγ2, a key regulator of adipocyte differentiation. Consistent with this, although Shn2 null mice are viable and fertile, they have reduced adipose tissue. Interestingly,although Shn2 also forms a complex with Smad1/Smad4 in response to Bmp signaling, it mediates transcriptional activation via a distinct mechanism(Jin et al., 2006). The sequence recognized by the complex does not resemble the BRE and in fact lacks a Smad1 site. Instead, activation of PPARγ2 expression by Shn2 requires sites for Smad4 and the transcription factor C/EBP-α (see Fig. 6). Mice lacking Shn3 are also viable and fertile, but they show defects in immune function. Shn3 has been shown to act downstream of the T cell and Tnf receptors (Oukka et al., 2002; Oukka et al., 2004), but whether these events are linked to Bmp signaling remains to be explored.

All three vertebrate Shn proteins contain the zinc finger domains implicated in Smad interaction(Pyrowolakis et al., 2004; Jin et al., 2006) (this work),raising questions about how related Shn proteins contribute to Bmp signaling. One possibility is that all Shn proteins can interact with Smads but that each paralog recruits different sets of cofactors. Thus, each Shn protein would regulate a distinct subset of targets, allowing Bmp signaling to control a wide range of target genes in a selective manner. Alternatively, vertebrate Shn proteins may share the ability to interact with common co-factors and have redundant or partially overlapping functions. In light of this, it is interesting that Shn2 and Shn3 mutant mice are viable, in contrast to the embryonic lethality of mutations in Bmp ligands, receptors and Smads (reviewed by Zhao,2003). Mouse Shn1 and Shn2 are widely expressed in overlapping tissues during embryogenesis and in adults, whereas Shn3 shows restricted expression, suggesting that it may have a more limited role (reviewed by Wu,2002). A more complete understanding of how Shn proteins contribute to Bmp signaling in mice will require the analysis of mouse Shn1 knockouts, as well as the generation of double and triple null animals. Our attempts to knockdown Shn1 function in X. tropicalisusing splice-site directed morpholino oligonucleotides have been unsuccessful(I.L.B. and K.W.Y.C., unpublished). Given that all three genes are ubiquitously distributed in gastrula stage embryos and continue to be expressed throughout early development(Durr et al., 2004), one possibility is that vertebrate Shn proteins are functionally redundant. Alternatively, because the morpholinos do not target mature mRNA, the absence of a phenotype may be due to perdurance of the maternal message. The conservation of the size and organization of Shn-like proteins suggests that their structural properties may contribute to their function. One possibility is that they act as molecular scaffolds for the assembly of the transcriptional machinery that mediates Bmp responsiveness (see Fig. 6). Shn proteins could also serve an architectural function by promoting bending or looping of the DNA, by binding at more than one location.

We gratefully acknowledge Rene Bernards, Tom Maniatis, Scott Barolo and Alan Laughon for generously sharing reagents, and Douglas Bornemann for insightful discussion and comments on the manuscript. Support was provided by grants from the NIH (GM-55442) and March of Dimes (FY2-06-135) to K.A., the NICHD (HD-29507) to K.W.Y.C., and the UC Cancer Research Coordination Committee (CRCC 34384) to R.W.