Regulation of dorsal-ventral patterning: the ventralizing effects of the novel Xenopus homeobox gene Vox

Fertilization of the Xenopus egg initiates and orients a rotation of the cortex relative to the cytoplasm, initiating a signaling cascade on the future dorsal side of the embryo (reviewed by Kimelman et al., 1992; Kessler and Melton, 1994). Additional signals arising from the vegetal hemisphere induce mesoderm at the equator of the embryo. Both signaling events are believed to transpire prior to zygotic transcription, imparting a primary pattern to the blastula-stage embryo. This primary pattern appears after the onset of zygotic transcription (the mid-blastula transition; MBT), when several genes are expressed differentially in either the presumptive dorsal or ventral regions of the embryo (Cho et al., 1991; Christian et al., 1991; Jones et al., 1995; Lemaire et al., 1995; Smith et al., 1995).

Spemann and Mangold demonstrated in transplantation experiments that the organizer, the region of cells above the dorsal lip, is capable of instructing cells in the ventral ectoderm and mesoderm to adopt neural and dorsal mesodermal fates, respectively, whereas in normal development the ventral ectoderm forms epidermis and the ventral mesoderm forms mesothelium and blood (Spemann and Mangold, 1924). Thus, in the traditional view of axis formation in amphibians, the tissue above the dorsal lip was thought to hold the communicable information to pattern the entire embryo. This view has been most recently summarized in the three signal model, in which the organizer region releases a diffusable signal that dorsalizes mesodermal cells and ‘neuralizes’ ectodermal cells (Slack, 1994). Epidermis and ventral mesoderm, in this model, are default fates from which cells are redirected by active signaling from the organizer.

Recent observations suggest an alternative to this model, in which the ventral side of the embryo plays an active and vital role in patterning the embryo (Sive, 1993; Fainsod et al., 1994; Graff et al., 1994; Suzuki et al., 1994; Schmidt et al., 1995b; Steinbeisser et al., 1995). The signaling molecule BMP-4 has received considerable attention with respect to this role. BMP-4 is expressed during gastrulation (Dale et al., 1992) in the ventral marginal zone and ectoderm (Fainsod et al., 1994; Schmidt et al., 1995b), and has been shown to be critical for the ventral patterning of the embryo (Graff et al., 1994; Suzuki et al., 1994; Schmidt et al., 1995b; Steinbeisser et al., 1995), possibly by antagonizing dorsal signaling molecules such as chordin (Sasai et al., 1994) or noggin (Smith and Harland, 1992). Thus, in an emerging view of axis formation in amphibians, pre-MBT inductive signals establish two approximate territories, dorsal and ventral, which produce opposing dorsalizing and ventralizing signals during gastrulation. This view is borne out by experiments in which BMP-4 is ectopically expressed, causing a significant reduction in the effectiveness of the dorsal organizer (Dale et al., 1992; Jones et al., 1992; Fainsod et al., 1994; Schmidt et al., 1995b). In addition, pattern refinement and elaboration in the embryo comes about gradually. This is evident during gastrulation in which gene expression patterns reveal the dynamic nature of regionalization; at the beginning of gastrulation, gene expression patterns appear unrefined, and by the end of gastrulation, boundaries of gene expression within specific tissue primordia become quite distinct. For example, prior to gastrulation Xnot is expressed throughout the embryo, and by the end of gastrulation is restricted to the presumptive notochord (von Dassow et al., 1993). This suggests that the combined activity of signaling molecules and regulators of gene transcription, in a dynamic process, leads eventually to a definitive pattern and to subsequent tissue differentiation within distinct regions of the embryo.

We have identified a novel homeobox-containing gene, Vox, which is first expressed throughout the blastula-stage embryo. During gastrulation, Vox is eliminated from the future notochord and neural plate. We show that ectopic expression of Vox on the dorsal side of embryos negatively regulates the expression of prechordal, notochordal and neural gene expression, while positively regulating the expression of the paraxial mesodermal marker MyoD and the ventral marker Bmp-4. In a reciprocal fashion, ectopic dorsal expression of Bmp-4 maintains the expression of Vox on the dorsal side of the embryo, from where it is normally eliminated. Furthermore, embryos injected with Vox RNA, when allowed to develop further, appear completely ventralized. These results provide evidence for a network of regulation in which reciprocal interactions result in the definitive patterning of the embryo. Vox may play an active role in establishing the ventral fates of mesoderm and ectoderm and in limiting the extent of dorsalization.

Vox was isolated using a polymerase chain reaction (PCR)-based approach as previously described (Northrop and Kimelman, 1994). The PCR fragments were used as probes to isolate full-length cDNAs from a stage 17 phage library (Kintner and Melton, 1987), which were inserted into the EcoRI site of the Bluescript SK vector (Stratagene). The complete nucleotide sequences of the cDNAs were determined and the nucleotide sequence of the Vox-15 cDNA was deposited in GenBank under accession number U53529 and the Vox-1 cDNA sequence under accession number U53528. A truncated Vox construct was produced using PCR, which contains the Vox coding region from the initiation codon through the WFQNRR sequence within the homeodomain. Both the normal and truncated Vox cDNAs were inserted into the CS2+ expression vector (Turner and Weintraub, 1994) to produce Vox-174 and Vox-165, respectively.

Fertilized Xenopus embryos were prepared as described previously (Newport and Kirschner, 1982). After the jelly coat was removed with 2% cysteine (pH 7.8), the eggs were washed in 0.1× MMR (1× MMR is 0.1 M NaCl, 2 mM KCl, 1.0 mM MgSO₄, 2.0 mM CaCl₂, 5.0 mM Hepes, and 0.1 mM EDTA). Embryos were incubated in 0.1× MMR at 14-23°C. Embryos were staged according to Nieuwkoop and Faber (1967). After removal of the vitelline envelope, injected and uninjected stage 12-13 embryos were placed in 1× MMR and dorsal and ventral quarters were cut with a 90° arc centered on the dorsal or ventral midline of the embryo using a fine wire knife.

RNAs were synthesized using the mMessage mMachine kit (Ambion). The RNAs were purified by one extraction with phenol:chloroform (1:1) followed by two rounds of concentration and separation in Microcon 100 microconcentrators (Amicon) to separate the RNA from unincorporated nucleotides. RNA was microinjected as previously described (Moon and Christian, 1989). 10 nl of RNA were injected per blastomere. Embryos were injected in 2 blastomeres of 4-cell embryos. The dorsal side of four-cell embryos was identified by pigment and cell size differences between dorsal and ventral blastomeres at this stage (Nieuwkoop and Faber, 1967). Embryos were fixed in 1× MEMFA (Harland, 1991).

Whole-mount in situ hybridization was performed using digoxigenin-labeled antisense RNA probes (Knecht et al., 1995), except that glass vials were used instead of baskets.

After MEMFA fixation and storage in methanol, embryos were embedded in Paraplast. Embedded embryos were sectioned at 10 μm and mounted in Permount (Kelly et al., 1991). Whole-mount embryos and sections were photographed using Kodak Ektachrome 160T film.

RNA was prepared by homogenization in an SDS-Proteinase K buffer (Cornell and Kimelman, 1994) and analyzed by the RNase protection assay (Melton et al., 1984). Probes for Vox, Bmp-4 and EF-1α were prepared as previously described (Dale et al., 1992; Cornell and Kimelman, 1994). A probe for Vox was produced by digesting Vox-15 with BglII and transcribing with T7 RNA polymerase. Probes were hybridized with RNA samples overnight at 45°C and then treated for 1 hour at 30°C with 40 μg/ml RNase A. Protected fragments were separated on 8% acrylamide-urea gels and exposed to film. For northern blots, poly(A) RNA was separated on a 1% agarose gel and transferred to Nylon. A Vox probe was produced from Vox-15 by labeling with ³²P by random priming.

We originally isolated Vox as a novel homeobox gene by PCR amplification of animal cap RNA using degenerate oligonucleotides directed against conserved regions of the homeobox. We isolated two full-length Vox cDNAs from a neurula-stage cDNA library, Vox-1 and Vox-15 (Vox = Ventral homeobox), both approximately 1.3 kb, with extensive nucleotide homology throughout the coding and untranslated regions (data not shown). Since the presence of two very similar Vox genes is likely to be due to the duplication of the Xenopus genome (Kobel and DuPasquier, 1986), we have studied only Vox-15, which we refer to here as Vox.

The Vox homeodomain has only limited similarity to homeodomains in previously identified homeobox-containing genes. The most closely related homeobox sequences are of the Gbx class; the homeodomain of the Xenopus gene Xgbx-2 (von Bubnoff et al., 1996) is shown in Figure 1B and has only 56.7% amino acid identity with the Vox homeodomain. Other homeodomains are even less similar, and most of the shared amino acids are highly conserved among all homeodomains (Fig. 1B). Vox therefore represents a new family of homeobox-containing genes.

Northern blot analysis revealed the presence of a single Vox transcript in the gastrula and neurula stages (Fig. 2A). The exact onset of Vox expression was determined by RNase protection analysis (Fig. 2B). Vox was very weakly detected in unfertilized eggs and early cleavage-stage embryos. At 7 hours postfertilization, strong expression of Vox was first detected. Since transcription of genes in the Xenopus embryo begins at the midblastula transition, 6 hours postfertilization, Vox is among the first genes to be zygotically expressed. The expression of Vox was maintained at a constant level throughout the gastrula and mid-neurula stages, but progressively declined during the late neurula and tailbud stages (Fig. 2B and data not shown).

The spatial localization of Vox was determined using whole-mount in situ hybridization (Harland, 1991). At stage 9, prior to gastrulation, Vox is expressed throughout the animal hemisphere and equatorial region of the embryo (data not shown). As RNAs present in the vegetal hemisphere are frequently not detected by whole-mount in situ hybridization, we analyzed the RNA levels present in animal versus vegetal hemispheres by microdissection and RNase protection. We detected equivalent levels of Vox in the animal and vegetal hemispheres at stage 10 (data not shown). Thus, Vox expression is activated uniformly at the MBT. At the onset of gastrulation (stage 10), Vox expression is cleared from a small region on the dorsal side of the embryo; this region corresponds to the location of Spemann’s organizer (Fig. 3A,B). By mid-gastrulation (stage 11), the clearing approximates both the region of the future anterior neural plate and the presumptive notochord (Fig. 3C). Near the end of gastrulation (stage 12), Vox expression is significantly cleared from the anterior dorsal region of the embryo, and the posterior paraxial expression develops distinct boundaries with the adjacent future notochord and floorplate regions (Fig. 3D). By the beginning of neurulation (stage 13-14), the posterior paraxial borders of Vox expression are sharp, and Vox expression in the anterior region is further reduced (Fig. 3E). Also in the early neurula, two new spots of expression appear in the anterior region of the embryo, which eventually becomes the dorsal side of the eye primordium in the tailbud and tadpole stages. At the end of neurulation (stage 19-20), Vox is expressed primarily on the ventral side of the embryo, and paraxial expression is limited to the posterior region of the embryo (Fig. 3F). Expression above the eye primordia is more distinct at this stage. By stage 22, posterior Vox expression is mainly limited to the tip of the tailbud flanking the notochord and floorplate (Fig. 3G). In stage 27 embryos, posterior expression is found only in the tail tip around the developing region of the notochord (Fig. 3H).

Because Vox is expressed in both the marginal zone and animal cap before gastrulation, we wished to compare the expression in the involuted versus noninvoluted cells during and after gastrulation. Sections of stage 11 embryos confirmed that Vox expression is absent at this stage in the region of the future neural plate (Fig. 4B) and along the future dorsal axis of the embryo (Fig. 4A) in both involuted and noninvoluted cells. Sections of stage 12 embryos stained for Vox expression revealed a significantly higher level of expression in the noninvoluted ventral marginal zone than in the involuted ventral mesoderm (Fig. 4C,C′,D). On the dorsal side of the embryo, sagittal sections revealed significantly diminished levels of expression along the dorsal midline, both in the involuted and noninvoluted tissues (Fig. 4C,C′′). In parasagittal sections, Vox expression was evident in both dorsal involuted and noninvoluted marginal zone tissue, specifically in the posterior region of the embryo, near the blastopore lip (Fig. 4D,D′). Anteriorly in those same sections Vox expression was very weak or absent in the dorsal ectoderm and mesoderm (Fig. 4D,D′).

The expression pattern of Vox bears a striking complementarity to the expression patterns of chordin (Sasai et al., 1994) and Xnot (Gont et al., 1993; von Dassow et al., 1993), which are both expressed in the dorsal region of the embryo. Both Xnot and Vox are expressed throughout the embryo at stage 9. At the onset of gastrulation, however, Xnot becomes restricted to the presumptive notochord, whereas Vox is eliminated from this region. This is seen clearly at stage 11 when the expression of Xnot in the presumptive notochord becomes pronounced and the absence of Vox from the presumptive notochord is distinct (Fig. 5A). At this stage, chordin is expressed within the presumptive notochord and is also expressed in the anterior region of the embryo in the future neural plate (Fig. 5A). This contrasts precisely with the loss of Vox expression at this stage within the future anterior neural plate (Fig. 5A). Also within the gastrula embryo, Vox and Bmp-4 expression overlaps; however, Vox expression extends more dorsally than that of Bmp-4, which is expressed on the ventral side of gastrula-stage embryos up to the border of the future neural plate (Fainsod et al., 1994; Schmidt et al., 1995b).

The complementarity of expression between chordin, Xnot and Vox continues into late gastrulation, and the Vox expression domain continues to overlap and extend beyond that of Bmp-4 (Figs 3 and 7; Schmidt et al., 1995b). In the early tailbud stages, chordin is expressed throughout the notochord (Fig. 5B, third embryo from top), whereas Xnot is limited to the posterior tip of the notochord and floorplate (Fig. 5B, top embryo). At this stage, the complementarity is most striking between Xnot and Vox, as Vox expression flanks the notochord and floorplate only in the posterior region of the embryo (Fig. 5B, second embryo from top). In the posterior of the embryo, Bmp-4 is restricted to a more ventral region than Vox, as well as to ectodermally derived regions of the tail fin (Fig. 5B, bottom embryo). At a slightly later tailbud stage, the complementarity between Xnot and Vox is clearly seen (Fig. 5C). By stage 31, chordin, Xnot and Vox are all restricted to the posterior tail region of the embryo and the complementarity between Vox and both of these axial genes is apparent (Fig. 5D, top three embryos). At this stage, Bmp-4 is expressed in the ventral posterior portion of the embryo as well as the tail fin of the embryo, including the tail tip (Fig. 5D, bottom embryo). Thus, not only do the expression patterns of these genes show striking complementarity, the complementarity is maintained during many stages of development.

Restriction of Vox expression from the dorsal region of the embryo suggests that elimination of Vox from this region is important for the development and patterning of the embryo. RNA encoding Vox was injected into the dorsal region of early cleavagestage embryos. Embryos injected dorsally at the 4-cell stage with 4 ng Vox RNA began gastrulation coincident with uninjected controls, and completed gastrulation with a slight delay in the closure of the blastopore relative to control embryos. However, as early as the late gastrula and early neurula-stages, Vox-injected embryos were morphologically distinguishable from uninjected controls and also later clearly failed to neurulate (data not shown). When allowed to develop until the tadpole stages, embryos injected dorsally with Vox RNA lacked axial structures and appeared in most cases to be completely ventralized (avg. DAI = 0.26, n=206, Fig. 6A; scored as 0 on the DAI scale where normal embryos are given a score of 5; Kao and Elinson, 1988). Lower doses of Vox RNA caused less severe effects. Ventral injections of Vox had very little or no effect on development; the predominant effect of ventral injection of Vox RNA, if any, was a slight abnormality in the ventral region of the tadpole stage embryo near the proctodeum, indicating that the dorsal side of embryos is much more sensitive to Vox RNA injections than the ventral side (avg. DAI = 4.3, n=74, Fig. 6B). As a control for these injection experiments, we made a construct encoding a truncated version of the Vox protein lacking a small portion of the C-terminal region of the homeodomain (including two DNA contacting residues thought to enhance specificity; Kissinger et al., 1990) as well as the C-terminal domain of the protein. Injection of RNA encoding this truncated form of the Vox protein into the dorsal region of early Xenopus embryos produced normal and nearly normal embryos (avg. DAI=4.2, n=81). The defective embryos in these experiments had a slight reduction in anterior development and/or a curved axis, which may have been due to an artifact of RNA injection (Moon and Christian, 1989), or residual activity of the mutant protein.

We also injected RNA encoding the homeobox-containing protein, Xhox3, expressed from the same vector. Ectopic expression of Xhox3 has been shown to cause truncation and ventralization of Xenopus embryos (Ruiz i Altaba and Melton, 1989b). Injection of 4 ng of Xhox3 RNA into the dorsal side of early embryos caused truncation of tadpole stages as reported earlier (avg. DAI = 2.4), but we did not observe any completely ventralized embryos as we observed with Vox RNA (see above). Injection of a two-fold higher dose of Xhox3 also did not produce any completely ventralized embryos, indicating that, although ectopic expression of both genes ventralizes Xenopus embryos, the effects of misexpression of Vox are more severe.

To better understand the abnormalities observed in embryos injected dorsally with Vox RNA, we examined the expression of genes transcribed during gastrulation and patterned during and after the onset of Vox expression in gastrula-stage embryos. Since Vox is cleared from the dorsal axial region of the embryo, we reasoned that expression of Vox within this region may adversely affect the expression of dorsal axial genes during gastrulation. To examine this, embryos injected dorsally with 4 ng Vox RNA were fixed and stained at the gastrula stage for goosecoid (gsc; Cho et al., 1991; Fig. 7A) and for Xnot (Fig. 7C) expression. Expression of both of these dorsal axial genes was eliminated (Fig. 7B,D). Since Xnot expression was eliminated from injected embryos, we looked at the expression of Xbra, which is also found within the notochord (Fig. 7E; Smith et al., 1991). In stage 12.5 embryos injected dorsally with Vox RNA, notochordal Xbra expression was eliminated (Fig. 7F). This may be a consequence of the elimination of Xnot expression in Vox-injected embryos, since floating head (the zebrafish homologue of Xnot) is required for the notochordal expression of no tail (the zebrafish homologue of brachyury), as shown by Talbot et al. (1995). We also examined the expression of MyoD, which is first expressed during midgastrulation (Harvey, 1990; Frank and Harland, 1991; Rupp and Weintraub, 1991). MyoD is expressed within the paraxial mesoderm and marks the presumptive somitic region of the embryo (Fig. 7G; Frank and Harland, 1991). In embryos injected dorsally with Vox RNA, the expression of MyoD was no longer absent from the presumptive notochordal region of the embryo, but instead circumscribed the embryo (Fig. 7H).

As embryos injected dorsally with Vox RNA appeared in most cases completely ‘ventralized’, we examined the expression of Xwnt-8, a marker of early lateral mesoderm (Christian et al., 1991). Xwnt-8 expression is absent from the dorsal side of stage 10 embryos (Fig. 7I). Dorsal injection of Vox RNA had no effect on the expression of Xwnt-8. The dorsal gap in expression appeared to be the same size in injected as in uninjected control embryos (Fig. 7I and J). This result demonstrated that dorsal injection of Vox RNA could not induce the expression of Xwnt-8 on the dorsal side of stage 10 embryos. However, injection of Vox RNA into the ventral marginal zone of 4-cell embryos eliminated the ventral expression of Xwnt-8 (Fig. 7L). These results indicate that Vox does not simply promote ventralization at the expense of dorsal gene expression, but affects gene expression selectively.

Because embryos injected dorsally with Vox RNA were ventralized, yet did not express Xwnt-8 on the dorsal side, we examined the expression of Bmp-4 during gastrulation in injected and uninjected embryos. In uninjected, stage 12 embryos, Bmp-4 expression is absent from a broad region approximating that of the presumptive neural plate. Expression of Bmp-4 appears strongest in the anterior ventral region and near the ventral side of the blastopore (Fig. 7M and Fainsod et al., 1994; Schmidt et al., 1995b). In embryos injected dorsally with Vox RNA, Bmp-4 expression was no longer absent from the dorsal side of the embryo, but instead was radially expressed, surrounding the entire blastopore (Fig. 7N). To quantitate the change in Bmp-4 expression, we dissected dorsal and ventral quarters of injected and uninjected embryos at stage 12-13, and analyzed the levels of expression of Bmp-4 by RNase protection (Fig. 8). Whereas control embryos had five-fold more Bmp-4 RNA on the ventral side than on the dorsal side of the embryos, nearly equal levels of Bmp-4 RNA were found on both sides of embryos injected dorsally with Vox RNA (Fig. 8). These results suggest that ectopic expression of Vox may lead to ventralization, in part, by inducing Bmp-4 on the dorsal side of the embryo during gastrulation.

Since Vox expression is eliminated from a region approximating that of the anterior neural plate and since embryos injected dorsally with Vox RNA do not appear to neurulate, we examined genes expressed within regions of the presumptive neural plate in embryos injected dorsally with Vox RNA. chordin is normally expressed within the presumptive notochord and anterior neural plate in stage 12 embryos (Fig. 7O; Sasai et al., 1994). Neither the presumptive notochordal nor anterior neural plate expression of chordin was detected in embryos injected dorsally with Vox RNA (Fig. 7P). Xotx2, a marker of the anterior neural plate (Fig. 7Q; Blitz and Cho, 1995; Pannese et al., 1995), was also absent in embryos injected dorsally with Vox RNA (Fig. 7R). Hairy II is expressed in a stripe that borders the presumptive neural plate as well as in the floorplate in late gastrula-stage embryos (Fig. 7S; Turner and Weintraub, 1994). Hairy II expression in stage 12 embryos injected dorsally with Vox RNA was greatly reduced (Fig. 7R). Thus, the disruption of neural development in embryos injected dorsally with Vox RNA is likely due to the suppression of genes important for neural plate specification.

Ectopic expression of BMP-4 ventralizes Xenopus embryos, reducing or eliminating all axial structures (Dale et al., 1992;Jones et al., 1992; Fainsod et al., 1994; Schmidt et al., 1995b). Since ectopic expression of Vox produces embryos that are also ventralized and lack axial structures, we examined the expression of Vox in embryos injected dorsally with Bmp-4 RNA. Vox expression in uninjected, stage 11 embryos is cleared from the dorsal side of the embryo (Fig. 9A). In embryos injected dorsally with Bmp-4 RNA, Vox was no longer cleared from the dorsal side but rather, was radially expressed (Fig. 9B). Therefore, BMP-4 is capable of inducing the expression of Vox on the dorsal side of the embryo.

By the late blastula stage, early gene expression patterns reveal some semblance of dorsal-ventral patterning in the embryo. During gastrulation additional genes are expressed, and as gastrulation commences, the expression patterns of these genes become increasingly defined relative to one another and relative to specific regions of the embryo. We describe here the isolation of Vox, a novel homeobox gene, which has an expression pattern that reflects the dynamic regulation of gene expression during gastrulation. Vox and the previously described gene, Xnot (von Dassow et al., 1993), are first expressed ubiquitously after the MBT and subsequently become restricted to specific regions of the embryo during gastrulation. Furthermore, the expression patterns of these genes are strikingly complementary. Xnot expression is restricted to the presumptive notochord during gastrulation, and Vox is excluded from the notochord. This complementarity suggests that these genes may function to regulate each other as well as the expression of other axial and paraxial genes. Although the complementarity of Xnot and Vox is striking, there is an additional feature of Vox expression that is not reciprocally represented in the expression of Xnot. Not only is the expression of Vox cleared from the presumptive notochord, it is also cleared from the presumptive neural plate during gastrulation. At this time, chordin is strongly expressed in both the notochord and the presumptive anterior neural plate (Sasai et al., 1994). This suggests an additional function for Vox in limiting the extent of the neural plate and a possible regulatory relationship between these genes.

The striking complementarity of the expression of these genes persists during the patterning and growth of the tail in the tailbud and tadpole stages, reflecting the patterning of the late gastrula organizing region (Gont et al., 1993; this paper). In particular, Vox and Xnot expression becomes progressively restricted to the posterior region of the tailbud embryo (e.g. stage 26). The restriction of genes expressed during gastrulation to corresponding regions in the tailbud is not unique to Xnot and Vox, but the timing of this restriction is precisely coincident for these two genes. At these stages, chordin is expressed throughout the notochord and is not yet restricted to the posterior region. However, during tadpole stages (e.g. stage 31), chordin, as well as Vox and Xnot, becomes restricted to the tip of the tail (Gont et al., 1993; von Dassow et al., 1993; Sasai et al., 1994; this work), suggesting that there may be a continuous regulatory relationship that persists well into later development.

In addition, the expression patterns of chordin and Bmp-4 are somewhat reciprocal. The complementarity of these two genes is perhaps most evident in the ectoderm, where Bmp-4 is expressed adjacent to the anterior border of the neural plate (Fainsod et al., 1994; Schmidt et al., 1995b), and chordin is expressed within the anterior neural plate (Sasai et al., 1994). However, the expression of Bmp-4 within the ventral marginal zone appears quite distant from the expression of genes such as chordin within the presumptive notochord. Thus, Bmp-4 may be important in normal development for the continued expression of genes such as Vox in ventral and paraxial regions; Vox in turn may directly regulate dorsal axial genes such as chordin and Xnot.

After submission of this manuscript, another Xenopus homeobox gene, Xvent-1, was described, which is very similar in sequence to Vox within the homeodomain and also has ventralizing properties when ectopically expressed (Gawantka et al., 1995). Since the sequences of Xvent-1 outside of the homeobox share very little similarity with Vox, it may be that the ventral region of the embryo is under the control of a set of genes with similar DNA-binding characteristics.

Heretofore a number of genes have been found that are expressed on the dorsal side of the early embryo, and are therefore putative regulators of organizer specification. Some of these genes induce the formation of a secondary axis when expressed on the ventral side of the embryo. Relatively few genes have been described thus far that promote the development of ventral tissues; Vox and Bmp-4 are among these few. BMP-4 ventralizes embryos when ectopically expressed on the dorsal side (Dale et al., 1992; Jones et al., 1992; Schmidt et al., 1995b), and Vox, when expressed ectopically on the dorsal side, also ventralizes embryos such that little or no axial development is apparent. This indicates that elimination of Vox expression from the neural plate, notochord and organizer is critical for the formation of axial structures.

Closer examination of Vox-injected embryos by in situ hybridization reveals that the ventralization takes place during gastrulation as a result of the loss of expression of axial genes such as gsc and Xnot, and the induction of non-axial genes such as MyoD and Bmp-4 on the dorsal side of the embryo. The induction of MyoD in Vox-injected embryos suggests that Vox is important for establishing MyoD expression in the paraxial mesoderm. Vox may achieve this either by restricting the expression of axial genes from the paraxial region, thereby allowing the expression of MyoD, or by positively regulating the expression of MyoD in the paraxial mesoderm.

An unexpected result was that ectopic Vox expression on the ventral side of embryos resulted in the elimination of ventral Xwnt-8 expression. This seems contradictory since both Vox and Xwnt-8 are both expressed within the ventral region of the embryo. However, an examination of the localization of ventral Vox expression demonstrated that Vox is primarily expressed in the non-involuted layer, and that Vox expression is down-regulated once the tissue (the prospective mesoderm) involutes. Xwnt-8, in contrast, is expressed within the involuted layer (Christian et al., 1991). Thus, the level of Vox expression in the ventral mesoderm, as a result of RNA injections, may be much higher than the levels found normally in that region. Consequently, Xwnt-8 expression is repressed in Vox-injected embryos, whereas in normal circumstances, Xwnt-8 and Vox are coexpressed in the ventral mesoderm.

Ectopic expression of Vox on the dorsal side of the embryo also eliminates or greatly reduces the expression of neural genes such as chordin, Hairy II and Xotx2. This suggests that the clearing of Vox expression from the neural plate is important for the development of the neuroectoderm. Ectopic expression of BMP-4 similarly eliminates or greatly reduces the expression of neural plate markers (Schmidt et al., 1995b). What relationship do these antagonists of neural induction have to one another? Ectopic Vox expression in early embryos induces Bmp-4 such that it is expressed in nearly equal amounts on both the dorsal and ventral sides of the embryo. In addition, ectopic expression of BMP-4 radializes Vox expression, eliminating the dorsal clearing. Vox expression normally precedes that of Bmp-4 (Dale et al., 1992) and therefore Vox might initially activate Bmp-4. However, Vox is eliminated from the organizer region at approximately the same time as Bmp-4 is first expressed, and Bmp-4 expression is absent from a broader dorsal domain than Vox. At this stage the ventral expression of Vox may become dependent on BMP-4 signaling for its continued expression. A previous study showed that ectopic expression of exogenous Bmp-4 in whole embryos enhanced the expression of endogenous Bmp-4 (Jones et al., 1992). Our results suggest that this may have occurred by the activation of endogenous Vox expression.

Patterning of the embryo during gastrulation is, not surprisingly, a complex phenomenon. Searching for regulatory hierarchies and linear relationships between genes expressed during early development may not be sufficient to explain embryonic patterning. Previous studies have shown that chordin and BMP-4 may be functional antagonists in the early embryo (Sasai et al., 1995). While chordin can induce a secondary axis on the ventral side of the embryo, thereby dorsalizing the character of the tissue into which it has been introduced (Sasai et al., 1994), inhibiting BMP-4 signaling on the ventral side of embryos can also impart a dorsal character to the injected region (Graff et al., 1994; Suzuki et al., 1994; Schmidt et al., 1995b; Steinbeisser et al., 1995). BMP-4 can eliminate all visible dorsal characters from the embryo when injected on the dorsal side (Dale et al., 1992; Jones et al., 1992), and can eliminate the expression of all axial genes tested (Schmidt et al., 1995b). Vox can eliminate chordin expression and induce Bmp-4 expression when ectopically expressed on the dorsal side, and BMP-4 can induce the expression of Vox on the dorsal side of gastrula embryos. BMP-4 thus both regulates and is regulated by Vox either directly or indirectly; each may be required for the expression of the other on the ventral side of the embryo (Fig. 10). By extension, chordin may both regulate and be regulated by both Vox and BMP-4. In this model, chordin antagonizes BMP-4 action, thus destabilizing Vox expression, and relieving its own repression (Fig. 10). This reciprocal regulation may be part of the mechanism by which embryonic patterning becomes defined during gastrulation.

The positive regulatory loop between Vox and BMP-4 may establish a stable ventral regulatory region within the embryo. In order for dorsal tissue to form on the ventral side, this positive regulatory loop must be either destabilized or prevented from forming. The ability of several genes to induce complete or incomplete secondary axes may reflect the ability of these genes to prevent formation of the ventral regulatory loop. Dorsal genes capable of inducing a complete secondary axis may act in two ways, by inducing the expression of other dorsal axial genes and by inhibiting, directly or indirectly, Vox and/or Bmp-4. Dorsal genes which induce an incomplete secondary axis may interfere with the ventral regulatory loop without the induction of a sufficient set of dorsal axial genes, and dorsal genes that do not induce a secondary axis may fail to do so because they fail to interfere with the loop. Siamois, a homeobox-containing gene expressed very early in the organizer region, can induce a complete axis when ectopically expressed on the ventral side of the embryo (Lemaire et al., 1995). As a possible scenario, siamois may directly suppress ventral regulatory genes or may induce the expression of chordin. The presence of chordin may prevent formation of the ventral regulatory loop. In addition, siamois may induce the expression of dorsal axial genes such as Xnot and gsc, which may in turn stabilize chordin expression (Sasai et al., 1994) and elicit additional dorsal gene expression. This constitutes the establishment of a dorsal regulatory loop (Fig. 10). It seems likely that cortical rotation initiates a series of events that bias the conflict between dorsal and ventral regulatory loops toward the dorsal genes in a single region of the embryo. Subsequent gene induction and reciprocal regulation generate, maintain and refine the patterning of the embryo.

We wish to thank Hank Farr, Kathryn Helde, Lauro Sumoy and Cynthia Yost for critical comments on this manuscript, and Eddy DeRobertis and Ken Cho for providing chordin and Xotx2 plasmids, respectively. J. S. is a recipient of a Howard Hughes predoctoral fellowship. G. v. D. is a recipient of a National Science Foundation predoctoral fellowship. This work was supported by a grant FY94-0474 from the March of Dimes to D. K.

Vox was also isolated as Xbr-1 in a screen for homeobox genes expressed in the eye.