Negative Autoregulation of the Organizer-specific Homeobox Gene goosecoid *

The homeobox gene goosecoid has been implicated to play a central role in the Spemann organizer tissue of the vertebrate embryo. Misexpression of goosecoid on the ventral side of a Xenopus laevis gastrula embryo was shown to result in a partial duplication of the primary body axis, reminiscent of the Spemann organizer graft. Normal embryonic development thus requires tight temporal and spatial control of genes instrumental for organizer function. In the present study we investigated the transcriptional control of goosecoid gene expression. Sequence analysis of the mouse and human promoter region revealed the presence of two palindromic binding elements for homeobox genes of the prd type to which goosecoidbelongs. We show that Goosecoid protein can bind to these sitesin vitro. By using reporter gene constructs of the human and mouse promoter, we demonstrate that Goosecoid can act as a repressor of its own promoter activity in transient co-transfection experiments in mouse P19 cells and in Xenopus embryos. Autorepression depends on the presence of the homeodomain and is mediated through the prd element more proximal to the transcriptional start site. Our results suggest a role forgoosecoid in restricting organizer activity in the vertebrate gastrula embryo.

The most acclaimed classical experiment in embryology, the organizer transplantation, was performed by Spemann and Mangold in the 1920s (1). Through this experiment a crucial role for the development of the primary body axis was assigned to the dorsal lip of the early newt gastrula embryo. When it was grafted to the ventral side of a recipient newt gastrula, a twinned embryo developed, in which a complete second body axis had formed. The dorsal lip was named the organizer by Spemann and Mangold to reflect its ability to not only autonomously differentiate into notochord but to change the fate of the neighboring mesodermal cells into dorsal characteristics and to induce a neural axis in the overlying ectoderm.
In an attempt to molecularly characterize the organizer phe-nomenon, the homeobox gene goosecoid was cloned from a dorsal lip cDNA library made from Xenopus gastrulae (2). Homologous goosecoid genes have been cloned in mouse (3), chick (4), zebrafish (5,6), human (7), and Drosophila (8,9). In all vertebrate gastrula embryos goosecoid expressing cells mark the equivalent of Spemann's organizer (10). Ectopic expression of synthetic goosecoid RNA in ventral blastomeres was able to cause axis duplications in frogs (11), mimicking the organizer graft to some extent. Further support for an essential role of goosecoid in the organizer came from functional experiments that implicated goosecoid in cell migration and dorsalization of mesodermal cells at the gastrula stage (12,13). Surprisingly, goosecoid null mutations generated by gene targeting (14,15) displayed no obvious gastrulation defects in mouse embryos, indicating that a related gene might provide a mechanism for a functional complementation. Mutant mice died perinatally of craniofacial defects related to the second phase of goosecoid expression during organogenesis (16). 1 Some aspects of the regulation of goosecoid expression have been studied. In Xenopus and mouse it was shown that the mesoderm-inducing growth factor Activin induced goosecoid gene expression in Xenopus in the absence of protein biosynthesis (3,11). Two elements in the Xenopus goosecoid promoter were identified that mediate the induction by the Activin and Xwnt-8 growth factors through their receptor-mediated signaling cascades (17). How gene expression is maintained in the node region and the prechordal plate and excluded from the more posteriorly located regions of the primitive streak has not been addressed as yet.
In this study we have investigated autoregulation of goosecoid. Our experiments were spurred by our finding that two potential goosecoid binding sites were located in the promoter region of the mouse and human goosecoid genes (18). These elements are contained in the activin and wnt response elements (17). We show that in vitro translated full-length Goosecoid protein can bind to these sites. Our experiments in Xenopus embryos and mouse P19 teratocarcinoma cells show that Goosecoid regulates its own transcription by a negative autoregulatory feedback loop. We demonstrate that autorepression is exclusively mediated through the element proximal to the transcriptional start site and that it is dependent on the presence of the homeodomain. Our results suggest that goosecoid is involved in the control and modulation of organizer activity both spatially and temporally during embryogenesis in mouse and frog. ADMP, a BMP 2 -3 related gene in Xenopus, was re-* The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The cently cloned and shown to provide anti-organizer activity in the organizer itself (19). Taken together this indicates that control mechanisms by which the vertebrate embryo restricts the organizer activity temporally and spatially during gastrulation are a function of the organizer itself.

EXPERIMENTAL PROCEDURES
Sequencing of the Mouse and Human Goosecoid Promoter and Cloning of Reporter Gene Constructs-Restriction fragments of genomic clones were subcloned and sequenced using standard procedures. Reporter gene constructs were generated by subcloning genomic fragments or fragments amplified by PCR into the luciferase vector pT81LUC (20), which contains the minimal promoter of the thymidine kinase gene (positions Ϫ81 to ϩ52) linked to the firefly luciferase gene. Cloning made use of a conserved SalI restriction site 7 bp upstream of the TATA box in both human and mouse goosecoid (Fig. 1A). All PCR clones were verified by sequencing.
Cloning of Goosecoid cDNA Clones-About 500 early mouse egg cylinder stage embryos (E6.4 -6.8) were isolated and frozen in liquid nitrogen. Polyadenylated RNA was prepared using standard procedures. cDNA synthesis and cloning into MOSSlox (21; Amersham Corp.) followed the procedures suggested by the manufacturer. The library was titrated and estimated to consist of about 500,000 independent clones with an average insert size of 1.3 kb. The unamplified library was plated and screened with a 909-bp PstI-HincII genomic goosecoid fragment containing most of exon 2, intron 2, and exon 3 (probe 2 in Ref. 3). After two rounds of rescreening, positive clones were subcloned and sequenced.
Determination of the Transcriptional Start Site of the Murine goosecoid Gene-The initiation site of transcription was determined by 5Ј-RACE (rapid amplification of cDNA ends). Polyadenylated RNA (2 g), isolated from E12.5 mouse embryos using a commercial kit (Life Technologies, Inc.), was hybridized with a goosecoid-specific primer (5Ј AGAAGTCTCCAAGTGGTGTTGTTTGGGGTG), which was derived from a sequence 234 bp downstream of the presumed TATA box (nucleotides 1173-1201 in Fig. 1C). Experimental conditions for RACE were as published (22). RACE products were cloned using the TA cloning kit (Invitrogen) and sequenced.
Mobility Shift Assay-Full-length Goosecoid protein was synthesized from cDNA-E in vitro using a reticulocyte lysate in vitro transcription and translation kit (Promega) and verified by 10% SDS-polyacrylamide gel electrophoresis. Oligonucleotides were annealed and radioactively labeled by fill-in reaction with Klenow DNA polymerase, followed by purification through a Sephadex G-50 column. Binding reactions (volume ϭ 20 l) containing Goosecoid protein (2.5% of the in vitro synthesis reaction), labeled double-stranded oligonucleotides (5 ng/reaction), poly(dI/dC) (Sigma; final concentration 50 g/ml), 15 mM Tris-HCl (pH 7.5), 60 mM KCl, 0.5 mM dithiothreitol, 0.25 mg/ml bovine serum albumin, 0.05% Nonidet P-40, 7.5% glycerol were incubated for 30 min at 25°C and electrophoresed on 4% acrylamide gels in 0.25 ϫ TBE (22.5 mM Tris borate, 0.5 mM EDTA (pH 8.0)) at 150 V for 1.5 h at room temperature. For competition experiments 100-fold molar excess of non-radioactive double-stranded oligonucleotides were added to the binding reactions. Prior to loading gels were prerun for 1 h. Gels were dried and exposed on x-ray film. The sequences for the top strands of the probes were as follows: consensus, TCGACTGAGTCTAATCCGATTAC-TGTACA; mouse DE, TCGACAATAGTATTAATAAGATTAACCTG; mouse PE, TCGAGATTAGGTTAATTTCATTAATTCTCAAT; and mouse mutated DE, TCGACAATAGTATTGACAAGGTCAACCTG.
Transient Transfections and Luciferase Assay-P19 cells were cultured on gelatinized dishes in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum. Cells (5 ϫ 10 5 per 100-mm culture dish) were transfected with 1 nmol of reporter plasmid (6.6 g of mouse long reporter, 5.4 g of human long reporter, 4 g of short reporter constructs), 1 g of effector plasmid (pRC/CMV-cDNA-E), and 1 g RSV-LacZ plasmid, using the calcium phosphate co-precipitation method. After 16 h the precipitate was removed by washing the cells twice with phosphate-buffered saline. Cells were harvested 40 -46 h after transfection. Following two rinses with Ca 2ϩ -Mg 2ϩ -free phosphate-buffered saline, cells were scraped off the dish in 0.5 ml of ice-cold lysis buffer (0.1 M Tris acetate (pH 7.5), 2 mM EDTA, 1% Triton X-100). Cells were collected by centrifugation (10,000 ϫ g, 4°C, 5 min), and the protein concentration was determined. Approximately 200 g of protein were assayed for luciferase reporter gene activity (23). To account for differences in transfection efficiency between samples, values were corrected by determining the ␤-galactosidase activity (24). Each experiment was repeated at least three times with at least three different preparations of plasmid DNA. Point mutations in the Goosecoid effector protein were introduced by site-directed mutagenesis of cDNA-E using the PCR-based overlap extension protocol (25). Mutations were verified by sequencing.
Injection Experiments in Xenopus Embryos-Adult frogs were purchased from the African Xenopus facility C. C. (South Africa). Eggs were obtained from females after injection of 500 units of human chorionic gonadotropin (Serva) into the dorsal lymph sac. Eggs were fertilized in vitro with macerated testes. After fertilization the jelly layers were removed by incubation in 2% cysteine solution (pH 7.9). Embryos were injected with RNA or DNA solutions (8 -10 nl/embryo) at the 2-4-cell stage and cultured until they reached the gastrula stage (NF stage 10; Ref. 26).
Capped Xenopus sense RNA was transcribed in vitro from plasmid pspgsc linearized with EcoRI using SP6 Polymerase (13). Control goosecoid mRNA lacking the homeobox was transcribed from plasmid p⌬gsc linearized with XhoI by using T7 RNA polymerase (11). For the synthesis of full-length Xenopus goosecoid antisense T7 RNA polymerase was used on plasmid pspgsc linearized with EcoRI (11). All transcription reactions were performed with the message machine kit (Ambion) according to the manufacturer's instructions.
Southern Blot Analysis-Genomic DNA (10 g/reaction) was digested with various restriction enzymes, electrophoresed on a 0.8% agarose gel, and transferred to a nylon membrane (Hybond; Amersham Corp.). The blot was first hybridized with random-labeled probe 2 (GC content 62%) in 1 M NaCl, 10% dextran sulfate, 1% SDS, 1 mg/ml yeast tRNA at 65°C; final wash: 1 ϫ SSC, 1% SDS at 65°C for 30 min. After obtaining several exposures the blot was stripped by boiling in 5 mM EDTA for 30 min. No signal was detectable after overnight exposure of the stripped membrane. Subsequent hybridization with random-primed probe 1 (71% GC content) was performed under identical conditions, except that hybridization and wash temperatures were raised to 67°C. Probe 1 was a 582-bp SalI-ApaI fragment corresponding to nucleotides 1-582 in the mouse goosecoid sequence deposited in the data base (accession number M85271). This fragment contained 295 bp upstream of the putative initiator methionine and 287 bp of exon 1 coding sequence. Probe 2 was a 909-bp PstI-HincII restriction fragment, corresponding to nucleotides 1248 -2157 (accession number M85271); it contained sequences derived from exon 2, intron 2, and exon 3.

Sequence Analysis of the Mouse and Human Goosecoid Promoter Reveals the Presence of Potential Goosecoid Binding
Sites-As a first step to investigate transcriptional regulation of goosecoid, we subcloned and sequenced genomic DNA fragments upstream of the translational start site of the human and mouse goosecoid genes (3, 7). Fig. 1A shows a schematic representation of the two genes. In the case of both the mouse and human gene more than 1 kb were sequenced. The start site of transcription of the murine goosecoid gene was mapped by 5Ј-RACE using RNA from E12.5 mouse embryos. It is indicated in the alignment of the promoter sequences which is depicted in Fig. 1C. The two sequences are highly similar to each other, with an identity of 75.1% in the stretch of 1276 bp shown in Fig. 1C.
The most striking feature of the two sequences was the presence of two palindromic motifs of the type TAAT-NNNATTA (Fig. 1B), which are also present in the published partial promoter sequence of Xenopus goosecoid (17). These palindromes were contained within sequences identified by Watabe et al. (17) as responsible for mediating the Activin and Wnt signals that induce goosecoid gene expression in Xenopus (see Fig. 1C and "Discussion"). In accordance with the nomenclature proposed by Watabe et al. (17) these elements will be addressed as distal element (DE) and proximal element (PE).
Palindromes of this type were previously identified as binding sites for homeobox proteins of the so-called prd type, named after the Drosophila homeobox gene paired (27). The homeobox proteins in this class are the only ones known to bind as homoor heterodimers to palindromic binding sites (18). It has been shown by Wilson et al. (18) that the amino acid residue in position 50 of the homeodomain plays the decisive role for the type of palindrome that was preferred by the different proteins. In case of a serine at this position a palindrome with a 2-bp spacing was found, whereas a 3-bp spacing was identified for lysine and glutamine. Examples of genes of the glutamine group are rpx/hesx1 (28,29), mhox (30), mix-1 (31), and orthopedia (32); the serine group contains paired and the pax genes  (18). Y ϭ C or T; R ϭ G or A; N ϭ A or T or G or C. C, alignment of human (top) and mouse (bottom) goosecoid promoter and 5Ј leader sequences. The palindromic prd elements are shaded. Additional potential binding sites for homeobox proteins (TAAT or ATTA) are underlined. Double-headed arrows indicate Activin and Wnt response elements identified by Watabe et al. (17). The prd site contained within the Activin response element (ARE) represents the distal element (DE), and the one in the Wnt response element (WRE) represents the proximal element (PE). TATA box and translation start codon are marked by boxes. The start site of transcription and the 5Ј ends of cDNA-D and cDNA-E are indicated below the murine sequence.
pax-3, pax-6, and pax-7 (33) (among others), and the most prominent members of the lysine group are represented by otx-1 and otx-2 (34) and, most notably, goosecoid itself. Fig. 1B shows an alignment of the palindromic prd sites in the goosecoid promoter of human, mouse, and Xenopus together with the consensus sequence identified for goosecoid by Wilson et al. (18). The presence of these two sites prompted us to study a possible autoregulatory feedback loop as one component of goosecoid gene regulation.
Isolation of Mouse Goosecoid cDNA Clones-To study autoregulation in the mouse system we cloned a mouse goosecoid cDNA to use as effector in DNA binding and gene regulation studies. A cDNA library was constructed from mouse gastrula egg cylinder embryos (E6.4 -6.8) and screened with a genomic goosecoid fragment (see "Experimental Procedures" for details). Five positive clones were recovered after the second rescreen. Sequence analysis revealed that the two clones designated cDNA-D and cDNA-E (Fig. 2) contained the complete coding sequence as predicted from the genomic sequence (3). The other three represented truncated clones and one partial clone fused to an unrelated cDNA (not shown). For further studies, mouse goosecoid cDNA was subcloned into the eukaryotic expression vector pRC/CMV which allows expression of genes under the control of the cytomegalovirus promoter.
Binding of the Mouse Goosecoid Protein to prd Sites in Its Own Promoter-To test if the full-length mouse Goosecoid protein can bind to the prd sites in its own promoter, we performed in vitro binding studies. Wilson et al. (18) have previously shown that a peptide containing the Xenopus Goosecoid homeodomain fused to the glutathione S-transferase protein can bind to the consensus palindromic sequence TAATCCGATTA as a homodimer in gel mobility shift assays. A full-length Goosecoid protein was synthesized by in vitro transcription and translation of clone cDNA-E and verified by SDS-polyacrylamide gel electrophoresis, which resulted in a band of the expected size of 28 kDa. Gel shift assays were performed with different templates. Specific binding was obvious in reactions using the consensus sequence (Fig. 3, lane 2), DE (Fig. 3, lane 5), and PE (Fig. 3, lane 8). Binding was abolished upon addition of 100-fold excess of nonradioactive template (Fig. 3, lanes 3, 6,  and 9). The specificity of this binding was further tested by using a mutated binding site (mDE) as template (Fig. 4F). In such reactions no specific mobility shift was observed (Fig. 3,  lane 10). In addition, binding to the consensus site, PE, and DE could not be competed by the addition of 100-fold molar excess In lanes 1, 4, and 7 binding reactions were performed with the reticulocyte extract without Goosecoid; lanes 3, 6, and 9 show binding reactions that were competed by the addition of 100-fold molar excess the of non-radioactive binding sites, and in the reactions in lanes 11-13 100-fold molar excess of mDE was used as competitor. Note that Goosecoid specifically binds to PE, DE, and the consensus element, but not to mDE, and that mDE cannot compete the specific binding. of non-labeled mDE (Fig. 3, lanes 11-13). These experiments demonstrate that in vitro a full-length mouse Goosecoid protein can bind to the two palindromic prd type elements present in its own promoter.
Goosecoid Acts as a Repressor of Its Own Transcription in Mouse P19 Cells-To analyze the goosecoid promoter activity in vitro, a series of reporter gene constructs were generated. Promoter sequences from the mouse and human goosecoid genes were cloned in front of the minimal promoter of the thymidine kinase gene linked to the firefly luciferase reporter gene (vector pT81Luc; Ref. 20). A reporter construct containing the endogenous mouse goosecoid promoter up to the translational start site, which was cloned into the promoterless luciferase vector pXP2, behaved in an identical manner to the thymidine kinase-Luciferase constructs. 1 Therefore, the heterologous thymidine kinase promoter was used throughout this study.
In fection of goosecoid cDNA-E in several cell lines tested. However, we noticed a further reduction of the already low reporter activities in the presence of the effector plasmid (data not shown). These experiments indicated a negative autoregulatory effect of goosecoid on its own transcription.
To test this option we screened mouse cell lines for high reporter activity levels under conditions of transient transfections without co-transfection of the goosecoid effector plasmid. The mouse teratocarcinoma cell line P19 (35) proved to be most efficient in this respect. In four experiments average induction factors of 7-fold were observed for the 4.2-kb mouse long reporter construct and of 9.8-fold for the 2.3-kb human long reporter, as compared with transfection of the empty reporter plasmid pT81LUC (Fig. 4A). When cDNA-E or a full-length Xenopus cDNA clone was co-transfected, this activity was repressed for both the mouse and human long reporters (Fig. 4, A  and C). The dependence of this repression effect of the human long reporter on the concentration of the effector plasmid pRC/CMV-cDNA-E is shown in Fig. 4E. A more than 3-fold repression was observed with 1 g of effector plasmid (Fig. 4, A  and E). In another titration experiment we tested the influence of the CMV promoter on the activity of the reporter plasmids. Lower reporter levels due to squelching effects started to become evident at concentrations of Ͼ2 g/per dish of the CMV effector plasmid (not shown). All transient transfection experiments reported here were therefore performed with 5 g of reporter and 1 g of effector plasmid per 100-mm culture dish. From this set of experiments we conclude that Goosecoid can efficiently repress its own promoter activity in mouse P19 cells.
To determine which parts of the Goosecoid protein were required for autorepression, we tested a Xenopus goosecoid cDNA clone which was truncated before the homeodomain and which was shown to be inactive in the axis duplication assay (11). No repression of the long human and mouse reporters was found when this plasmid was co-transfected (Fig. 4, A and C 36 -38). To test the involvement of specific residues in autorepression, we constructed a series of point mutations in cDNA-E. When the asparagine in position 51 of the homeodomain (position 210 of the protein) was changed into glycine, autorepression was abolished (mutant mgscN210G in Fig. 4B). Repression levels were reduced by mutating valine in position 28 of the homeodomain into asparagine (4.3-fold as compared with 9.8-fold for the wild-type Goosecoid protein; mutant mgscV187R in Fig. 4B). These mutants demonstrate the dependence of autorepression on DNA binding, although cooperative dimerization seemed to be involved but not essential for autorepression under the experimental conditions of transient co-transfections. When the lysine in position 50 of the Goosecoid homeodomain was mutated into a glutamine, the repression effect was still significant but markedly reduced (3.8-fold as compared with 9.8-fold for the wild-type Goosecoid protein; mutant mgscK209Q in Fig. 4B). Co-transfection of the homeobox gene mix-1, however, which belongs to the prd class like goosecoid and which has a glutamine in position 50 of the homeodomain, showed no repression effect (Fig. 4B). mix-1 was shown to bind to the consensus prd type binding element TAATTGAATTA in vitro (18). These results suggested that residues outside of the homeodomain might contribute to the observed autorepression.
A recent publication reported the presence of a highly con-served region of seven amino acids, shared between the homeoproteins Engrailed, Goosecoid, NK1, NK2, and MSH (39). This sequence motif is located in the amino-terminal part of the Goosecoid protein and is indicated in Fig. 2. In in vivo experiments in Drosophila it was shown that this domain was essential for the repressor function of Engrailed (39). To test a possible involvement of this motif in autorepression, we constructed two amino-terminal truncations of cDNA-E which lacked the first 12 (mgsc⌬12) and 23 (mgsc⌬23) amino acids, respectively. Although repression was virtually unaffected with mgsc⌬12, it was reduced to 4.1-fold with mgsc⌬23. Taken together, our experiments with mutated Goosecoid proteins demonstrate the requirement of DNA binding for autorepression and suggest an involvement of cooperative dimerization, whereas the suspected repression domain was not essential.
To determine the role of the prd type binding sites in the goosecoid promoter, we constructed short reporter plasmids of 139 bp for human and 142 bp for mouse goosecoid that consisted of DE and PE and the intervening sequences (Fig. 1C). Although the basal activity of these constructs was about 4-fold below that of the respective long reporters, co-transfection of mouse goosecoid cDNA-E resulted in repression close to background levels (Fig. 4B). These experiments indicated that while sequences upstream of the prd sites contribute to the basal activity of the promoter in P19 cells, repression was mediated through fragments harboring just the two palindromic binding sites. As DE and PE in mouse and human are quite similar, and Goosecoid can bind to both elements equally well (Fig. 3), we investigated whether both contributed equally to the observed repression effect. Multiple point mutations were introduced in either element changing pyrimidine into purine residues (Fig. 4F). Co-transfection experiments of mutated elements in the context of the 142-bp fragments together with mouse goosecoid cDNA-E showed that while a mutation of DE did not alter the repression pattern, it was completely abolished by mutating PE (Fig. 4, B and D). The basal activity of a double mutant was not elevated in comparison to pT81LUC (Fig. 4D). From this set of experiments we conclude that autorepression was mediated via PE, and that one of the elements has to be functional for the basal activity of the short promoter fragment. Taken together the experiments in P19 cells show that Goosecoid can efficiently autorepress its own transcription and that this activity is dependent on binding of the homeodomain to the proximal prd type element (PE).
Autorepression of Goosecoid in Xenopus Embryos-In the following experiment we asked whether the autorepression of goosecoid which was seen in P19 cells could also be observed in Xenopus embryos. The regulation of goosecoid during the gastrulation process is important for the correct establishment of the anterior posterior body axis, and autorepression could be needed to spatially restrict the expression domain of goosecoid. Two luciferase reporter constructs containing a 2.3-kb or a 279-bp human goosecoid promoter fragment were injected into Xenopus embryos at the 2-4-cell stage. The embryos were cultured until early gastrula stage (NF stage 10), and luciferase activity was measured in embryo extracts. As an internal standard constant amounts of ␤-galactosidase mRNA were coinjected, and the enzyme activity was quantitated in the embryonic extracts and used for normalization of luciferase activity.
The long as well as the short human reporter construct showed activity in Xenopus in contrast to the empty luciferase vector which remained inactive (Fig. 5). When high amounts of synthetic goosecoid mRNA (180 pg/embryo) were co-injected with the reporter plasmids, the activity of the reporter constructs was reduced by over 90%. The same amount of anti-sense goosecoid RNA did not significantly alter the activity of the human reporter constructs. In agreement with the previous finding that the homeobox of Goosecoid was required for autorepression of the goosecoid promoter in mouse P19 cells, we found no or only weak repression in Xenopus embryos when mRNA for a truncated protein lacking the homeodomain was co-injected with the human reporter constructs. When mutant reporters mDE and mPE were co-injected with wild-type goosecoid RNA repression was seen with mDE and abolished with mPE (data not shown). As in P19 cells a double mutant of DE and PE was reduced to background levels (not shown). The experiments in frog embryos corroborate the results obtained in the mouse system, demonstrating that the homeobox of the Goosecoid protein is able to interact with its own promoter resulting in a repression of goosecoid transcription.
Goosecoid-related Sequences in Mouse Genomic DNA-The experiments described so far clearly demonstrate that Goosecoid can act as a strong transcriptional repressor. In the light of this finding the lack of a gastrula phenotype of goosecoid knock-out mice (14,15) suggests that another homeobox gene with similar DNA binding specificity, which is also able to act as a repressor, can complement the loss of goosecoid function. Many vertebrate homeobox genes outside of the Hox clusters exist in two or more related copies. In the mouse there are, for example, two engrailed-like genes, en-1 and en-2 (40), two evenskipped-like genes, evx1 and evx2 (41,42), and two each of the orthodenticle (otx1 and otx2; Ref. 34) and empty spiracle class (emx1 and emx2, Ref. 34). To test if a gene related to goosecoid was present in the mouse genome, we performed a detailed analysis. Genomic DNA was digested with a series of restriction enzymes and analyzed on a Southern blot by sequential hybridization under reduced stringency conditions with two non-overlapping genomic goosecoid probes (Fig. 6A). Probe 1 was derived from exon 1 and contained the engrailed homology domain, whereas probe 2, derived from exon 2, intron 2, and exon 3, contained the homeobox. All radioactive label was removed after the first round of hybridization.
The two probes detected fragments of the expected sizes, as deduced from the genomic map of goosecoid shown in Fig. 6A. Hybridization with probes 1 and 2 is shown in Fig. 6, B and C, respectively. Both probes detected fragments of 12.5 kb in digests with BamHI, 17.5 kb with EcoRI, 14.8 kb with EcoRV, 6.5 kb with HindIII, 11.7 kb with KpnI, 15.8 kb with NcoI, 11.0 kb with SalI, and 3.2 kb with SmaI. The expected different sized fragments were detected in digests using ApaI (3.0 kb with probe 1, 4.7 kb with probe 2) and PstI (2.8 kb with probe 1, 4.7 kb with probe 2). In addition to the expected fragments, both probes recognized multiple further bands. In many digests these additional bands were identical with both probes. This may be best seen with HindIII (9.1-kb fragment, additional fragment(s) marked with arrowheads), NcoI (5.4 kb), PstI (fragments of 7.7 and 5.6 kb), SalI (6.1 kb), and SmaI (8.2 and 4.9 kb). This result strongly suggests the presence of at least one additional goosecoid-related gene in the mouse genome.

DISCUSSION
The Goosecoid Promoter-The most striking feature of goosecoid promoter sequences of the mouse and human genes are two palindromic elements, DE and PE, reminiscent of the P3 element identified as binding site for homeobox proteins of the prd type such as goosecoid (18). An aspect of the two elements that deserves to be mentioned is that one of the half-sites overlaps with a second canonical homeobox binding site TAAT (Fig. 1C). The sequence ATTAAT . . . ATTA as such can be read as a palindrome of the prd type, TAAT . . . ATTA, or as a direct repeat, ATTA. . . . .ATTA, with a 5-bp spacing. The significance of this arrangement has not been addressed in the present study; however, it indicates the possibility of a complex regulation of the goosecoid promoter through homeobox proteins. Other typical promoter elements were not identified in the sequenced stretches of more than 1 kb of mouse and human 5Ј-flanking sequences, except for several SP1 sites in a region of about 300 bp upstream of the TATA box (not highlighted in Fig. 1C). The presence of the prd elements prompted us to study autoregulation, an idea that was supported by the in vitro binding of full-length Goosecoid protein in gel mobility shift assays (Fig. 3). We therefore designed in vivo experiments to study the physiological relevance of this interaction.

Goosecoid Acts as a Transcriptional Repressor of Its Own Promoter-The experiments in both mouse P19 cells and in
Xenopus embryos argue that Goosecoid acts as a repressor of its own transcription, an effect that was dependent on the presence of the homeodomain, and was mediated through DNA binding to the proximal prd element PE. Our preliminary analysis of the Goosecoid effector protein shows that DNA binding is a prerequisite, and cooperative dimerization may add to the observed autorepression. The homology region to the homeobox protein Engrailed, however, which comprises seven amino acids in the amino-terminal part of the protein and which in case of Engrailed was shown to mediate repression function in vivo (39), was dispensable for autorepression of goosecoid. We are presently mapping the domain responsible for autorepression further by deletion and mutation analysis of Goosecoid.
Not much is known to date with respect to target genes of goosecoid. Experiments in Xenopus have suggested that ectopic goosecoid expression could result in activation or repression of other genes. Transcription of bmp-4 and the two ventral homeobox genes xvent-1 and xvent-2 were shown to be downregulated upon ectopic goosecoid expression (43)(44)(45), whereas expression of chordin was induced in Xenopus embryos. To reconcile these conflicting results two possibilities can be considered.
First, as was argued for engrailed (39), activation of target genes might be a consequence of the inactivation of a repressor. Second, interaction with an as yet unidentified co-factor may inhibit goosecoid's repressor activity and lead to increased transcription, or, third, may be able to switch goosecoid to an activator of transcription. The demonstration of cooperative interactions of several of the Hox proteins with the divergent homeobox protein Extradenticle/PBX1 provides a precedent for modulation of transcriptional activity of homeobox proteins through co-factors (46,47), and in the case of the Drosophila homeobox gene deformed it has been recently suggested that these co-factors are necessary for transcriptional activation of target genes (48). It remains to be seen if the homology region to engrailed and several other homeobox genes (39), which plays no role for autorepression, may have a function in modulating the activity of the transcription factor goosecoid.
Goosecoid and the Organizer: A Model-In Xenopus and mouse it was shown that the mesoderm-inducing growth factors Activin, a member of the transforming growth factor-␤ family, and Xwnt8, a member of the wnt family of secreted factors, can induce zygotic goosecoid transcription (3,11). Response elements in the goosecoid promoter were identified that mediate this response (17). Most interestingly, the two palindromes that were the subject of the present study are located within these regions ( Fig. 1C; Ref. 17). It was shown previously that point mutations of the homeobox binding sites abolished activation through Activin and Wnt8. Factors through which the downstream signaling cascades of Activin and Wnt exert their effects therefore may include homeobox proteins, possibly of the prd type. This interpretation is supported by our finding that mutations of both DE and PE led to a decrease of reporter gene activity to background levels in both P19 cells and Xenopus embryos (Fig. 4B and data not shown). Candidate genes for positive regulators that might mediate the induction of zygotic goosecoid transcription through these regulatory elements include siamois (49), otx-2 (34), and xlim-1 (50).
Once goosecoid expression during gastrulation is induced and Goosecoid protein is translated, the autorepression described in this study becomes relevant. Repression may involve a competition mechanism with transcriptional activators like otx-2 for DE and PE. Autorepression seems to be a mechanism to keep goosecoid expression at a moderate level. In chick and mouse embryos the strongest goosecoid mRNA expression was seen at a stage when the primitive streak was fully extended. At later stages signals in in situ hybridization experiments became weaker, probably reflecting the onset of autorepression of transcription (3,4). In Xenopus embryos it was shown that overexpression of goosecoid mRNA in dorsal cells, where the gene is normally expressed, altered cell fate and migration behavior during gastrulation such that injected cells ended up in more anterior positions (12). This experiment reflects a situation in which autorepression of goosecoid was overcome. The phenotypic consequences of misdirecting small groups of dorsal cells by microinjection of goosecoid into single blastomeres of the 32-cell embryo were not addressed in detail. However, one can assume that proper development of the prechordal plate mesoderm would be grossly affected if all of the goosecoid positive cells of the organizer would undergo such changes in cell fate and migration. Severe phenotypes were reported for several knock-out mutants in the mouse that affected the region of the prechordal plate, i.e. hnf3␤ (51,52), lim-1 (53), and sonic hedgehog (54). Taken together, this suggests a physiological role for the observed autorepression of goosecoid.
The recently cloned secreted growth factor ADMP, a BMP-3 related protein of the transforming growth factor-␤ family, is expressed in the dorsal lip and behaves like a typical dorsal gene in Xenopus gastrula embryos but provides potent antidorsalizing activity (19). The organizer tissue thus synthesizes transcription and growth factors that counteract its own activity locally. This might represent a regulatory mechanism by which the organizer activity is modulated and balanced for proper embryonic development to occur. Our results suggest that goosecoid has a dual function in the organizer. It can promote organizer function via repression of the ventral genes xvent-1, xvent-2, and bmp-4, and by inducing chordin, directly or indirectly. On the other hand these functions can be counteracted in the embryo by the autorepression described here, providing a means for negative regulation of the organizer.
The scenario presented above describing the potency of goosecoid in gain-of-function studies contrasts sharply with the apparent ease with which the embryo deals with loss of goosec-oid in the mouse (14,15). Together with experiments in which the organizer region was ablated in Xenopus, chick, and zebrafish (55)(56)(57), without affecting normal embryogenesis, they demonstrate the plasticity of early vertebrate embryos and their regulative potential. In molecular terms other genes must be able to compensate for the loss of goosecoid function and must be able to reset gastrulation in the absence of the primary organizer. We have presented evidence that at least one gene related to goosecoid exists in the mouse genome. Recently, a goosecoid-related gene was described in the chick (58). This gene shows high homology to goosecoid in the homeobox and in the engrailed homology region and an initially overlapping expression pattern in Koller's sickle (58). The cloning of the homologous gene in the mouse, which is presently underway in our laboratory, and genetic analysis in knock-out and double knock-out mice will clarify the role of goosecoid genes during vertebrate gastrulation.