Bone morphogenetic protein 4: a ventralizing factor in early Xenopus development

The mesoderm in amphibian embryos such as Xenopus laevis arises through an inductive interaction in which cells of the vegetal hemisphere of the embryo act on overlying equatorial and animal pole cells (Nieuwkoop, 1969; reviewed by Smith 1989). Embryological experiments suggest that only two types of mesoderm are initially induced: the dorsalmost quadrant of the vegetal hemisphere induces dorsal ‘organizer’ mesoderm, while the remaining ventral and lateral quadrants induce ventral mesoderm (Dale et al., 1985; Dale and Slack, 1987). The dorsal organizer then acts upon the ventral mesoderm to generate the typical vertebrate mesodermal pattern, a process known as dorsalization (Smith and Slack, 1983; Dale and Slack, 1987; Stewart and Gerhart, 1990).

Recently, progress has been made in identifying ‘mesoderm-inducing factors’ (MIFs), and the candidates fall into three groups (reviewed by Whitman and Melton 1989; Dawid and Sargent, 1990; New and Smith, 1990; and see Smith and Harland, 1991; Sokol et al., 1991). The first group consists of members of the fibroblast growth factor family, such as basic fibroblast growth factor (bFGF). These induce ventral mesoderm when added to isolated animal caps. A second group comprises members of the transforming growth factor-/? (TGF-/J) superfamily, of which the most potent are activins A and B (Asashima et al., 1990; Smith et al., 1990; Sokol et al., 1990; Thomsen et al., 1990; van den Eijnden Van Raaij et al., 1990). These latter proteins induce dorsal mesoderm when added to isolated animal caps. The final group includes members of the Wnt family of proteins. These are not available in soluble form, but when synthetic RNA encoding one of the members of the family is injected into the ventral-vegetal quadrant of an early Xenopus embryo, it induces a second dorsal axis (McMahon and Moon, 1989; Christian et al., 1991; Smith and Harland, 1991; Sokol et al., 1991).

At present, it is not known which members of the FGF, TGF-β and Wnt families, if any, are the natural endogenous inducing factors, although the evidence is good that an FGF-like molecule does play a role in mesoderm formation. bFGF mRNA and protein are present in the Xenopus egg and early embryo (Kimel-man and Kirschner 1987; Kimelman et al., 1988; Slack and Isaacs 1989), and recent work by Shiurba et al. (1991) has shown that both basic and acidic FGF are concentrated in the vegetal hemisphere and marginal zone of the blastula. In addition, injection of RNA encoding dominant negative mutations of the FGF receptor into fertilized Xenopus eggs causes posterior defects in the resulting tadpoles (Amaya et al., 1991), a result consistent with the proposed role of FGF as an inducer of posterior mesoderm (Green et al., 1990; Ruiz i Altaba and Melton, 1989; Cho and De Robertis, 1990).

In this paper we concentrate on members of the TGF-β family. One strong candidate for a TGF-β-like natural inducing factor is the protein product of Vg1, a vegetally localised maternal mRNA whose sequence is most closely related to the bone morphogenetic proteins (Rebagliati et al., 1985; Weeks and Melton 1987). However, there is no evidence that Vg1 protein has mesoderm-inducing activity, and indeed very little, if any, of the 40 000 relative molecular mass (M_r) immature protein is processed into the mature 17 000 M_T form (Dale et al., 1989; Tannahill and Melton 1989). Another candidate is activin B, transcripts for which are first detected at the mid-blastula transition in Xenopus, when zygotic transcription begins (Thomsen et al., 1990). However, work by Jones and Woodland (1987) suggests that mesoderm induction in vivo begins at the 64-cell stage, at least 3 hours before the mid-blastula transition, indicating that the natural mesoderm-inducing factor(s) should be maternally encoded. Consistent with this, Asashima et al. (1991) have recently shown that the Xenopus egg and early embryo contains activin-like inducing activity.

Two approaches might be adopted to identify and characterize the endogenous activin-like mesoderm-inducing factors. The first involves purifying and sequencing the proteins, and work of this kind is under way in one of our laboratories (G.-D. Guex and J.C.S., unpublished data). The alternative is to clone members of the TGF-β family that are expressed during early development in Xenopus, to study their expression patterns, and then discover whether they have inducing activity. In this paper, which is an example of the second approach, we describe the cloning and expression pattern of Xenopus bone morphogenetic protein-4 (XBMP-4), a member of the TGF-β family, and one which is particularly closely related to Drosophila decapentaplegic (dpp), Xenopus Vg1, mouse Vgr-1, and BMPs -2 and -3 (Padgett et al., 1987; Weeks and Melton, 1987; Wozney et al., 1988; Lyons et al., 1989). We also investigate the inducing activity of XBMP-4 by injecting mRNA into the animal hemisphere of Xenopus eggs and dissecting animal caps from the resulting blastulae. Using this technique, we find that XBMR-4 preferentially induces ventral mesoderm, unlike the closely related activin which induces axial tissues. This result confirms those of Köster et al. (1991), who treated animal caps with recombinant Xenopus BMP-4. We also find, however, that the action of XBMP-4 is ‘dominant’ over that of activin. In this respect, XBMP-4 differs from basic FGF, another ventral inducer, where simultaneous treatment with FGF and activin results in activin-like responses (Cooke, 1989). The dominance of XBMP-4 over activin may account for the ability of injected XBMP-4 RNA completely to ‘ventralize’ intact Xenopus embryos. It is interesting, however, that blastopore formation in such embryos can occur perfectly normally. This contrasts with embryos ventralized by UV irradiation and suggests that XBMP-4-induced ventralization occurs after the onset of gastrulation.

To identify additional members of the TGF-β activin family that have mesoderm-inducing activity, we designed degenerate primers based on the known sequences of mammalian activins A and B for use in the polymerase chain reaction (PCR). The sequences of the primers are shown below:

The first primer corresponds to the N-terminal region of the mature activin A protein, and the second to the C-terminal region (Fig. 1C). DNA to be amplified was Xenopus genomic DNA (1 μl of 1 mg ml^-1 DNA) or cDNA made from poly(A)⁺ RNA derived from the Xenopus XTC cell line (5 μ l of reaction product from a reverse transcription reaction using 3 μg poly(A)⁺ RNA in a volume of 30 μ l). For each PCR reaction, we mixed DNA with 0.5 μl of 5 U ml^-1 Taq DNA polymerase, 10 μ of 5 × Taq polymerase buffer (0.25 M TrisHQ pH 8.3, 0.375 M KC1,15 mM MgQ₂, 50 mM DTT, 0.625 mM dNTPs) and 1 × l of each primer at 1 mg ml^-1. The final volume was adjusted to 50 ×l with water. Thirty cycles of amplification were performed, each consisting of denaturation at 94°C for 1 minute, annealing at 52°C for 1.5 minutes and elongation at 72°C for 1.5 minutes. The final elongation step was extended to 10 minutes.

PCR products were approximately 260 nucleotides. They were blunt-end cloned into the vector pSP72 and sequenced. The 86 amino acids encoded by one cloned fragment differed by only one amino acid from human bone morphogenetic protein 4 (BMP-4), and this was used for screening an XTC cDNA library (see below).

A Xenopus XTC cell cDNA library prepared in λ ZAP (the kind gift of Dr I. B. Dawid, NIH, Bethesda) was plated on E. coli BB4 cells. 8 × 10⁵ plaques were screened with the cloned PCR fragment. Three positive clones were isolated and sequenced, of which the longest was about 1.3 kb. This clone, pA, contained all but the first 50 amino acids of the coding region of Xenopus BMP-4 (XBMP-4). To obtain a full-length cDNA, we used this clone to screen a stage 15/16 Xenopus cDNA library prepared in λ gtl0. 10⁶ plaques were screened, and those with the longest inserts were selected as described by Elliott and Green (1989). Two cDNAs were obtained, one of which was 1.9 kb (XBMP-4.1) and the other 2.3 kb (XBMP-4.4). Both these cDNAs were subcloned into pBlue-script KS+ and XBMP-4.1 was also cloned into pSP64T (Krieg and Melton, 1984) in both the sense and antisense orientations to yield, respectively, pSP64T-XBMP-4+ and pSP64T-XBMP-4 –.

XBMP-4.1 was sequenced using a Pharmacia T7 sequencing kit.

RN Aase protections were carried out essentially as described by Green et al. (1990). An XBMP-4 antisense probe was prepared by digesting pA with Hpal, which was then transcribed with T7 RNA polymerase to give a probe of approximately 360 bases and a protected fragment of 300 bases, consisting of 200 bases of 3’ untranslated region and 100 bases of the C terminus of XBMP-4 (see Fig. 1). Próbes for Xhox3 (Ruiz i Altaba and Melton, 1989) and muscle specific actin (Mohun et al., 1984) were also used. As a loading control in RN Aase protections, we used an antisense probe for EF-lœ, which is expressed in all embryonic cells in the absence of induction (Krieg et al., 1989; see Sargent and Bennett, 1990).

RNA was extracted from Xenopus embryos and analysed by northern blotting as described by Sambrook et al. (1989). For XBMP-4 the probe was a Sacl/HinCH fragment containing most of the non-mature and about half of the mature region of the protein (see Fig. 1C). To detect actin, RNA blots were probed with an antisense RNA probe spanning the 3’ UTR and the last protein coding exon of muscle-specific αactin (Mohun et al., 1984). To detect globin transcripts, an antisense RNA probe spanning the 5’ HaeIII fragment of the Xenopus tadpole β Tl-globin gene was used (Banville et al., 1983).

Embryos of Xenopus laevis were obtained by artificial fertilization as described by Smith and Slack (1983). They were chemically dejellied using 2% cysteine hydrochloride (pH 7.8-8.1), washed and transferred to Petri dishes containing 10% normal amphibian medium (NAM: Slack, 1984). The embryos were staged according to Nieuwkoop and Faber (1967). Embryos were dissected using sharpened forceps and electrolytically sharpened tungsten needles. Treatment of embryos with LiCl was as described by Cooke and Smith (1988).

Recombinant human activin A was the generous gift of Dr G. Wong (Genetics Institute, Massachusetts). Pure recombinant Xenopus bFGF was prepared from an expression plasmid kindly provided by Drs D. Kimelman and M.W. Kirschner (University of California, San Francisco; see Kimelman et al., 1988).

Capped synthetic XBMP-4 mRNA was synthesized from pSP64T-XBMP-4+ and pSP64T-XBMP-4 – to give both the sense and antisense transcripts. Polyacrylamide gel electrophoresis of the products of in vitro translation of sense RNA resulted in a band of the expected size (M_r 43 × 10³), while no band was visible with antisense RNA. Different concentrations of RNA were injected into Xenopus embryos between the one-cell and four-cell stages. Some embryos were then allowed to develop until stage 38-40, while the animal pole regions of others were dissected and allowed to develop in the presence or the absence of inducing factors, as appropriate.

Embryos and explants were fixed and processed for histology as described by Green et al. (1990). 7 μ m sections were stained by the Feulgen/Light Green/Orange G technique of Cooke (1979).

In an attempt to discover potential mesoderm-inducing factors, we used the polymerase chain reaction to search for activin-related molecules expressed during early Xenopus development. Our primers were based on the N- and C-terminal regions of mammalian activins, which also resemble members of the BMP family. PCR of both Xenopus genomic DNA and cDNA from the XTC cell line gave fragments which showed high homology to mammalian BMP-4, and we went on to screen two libraries in order to obtain full-length cDNAs for Xenopus BMP-4 (see Materials and Methods).

XBMP-4.1 was completely sequenced, and was found to be homologous, over most of its length, to the Xenopus BMP-4_II of K ö ster et al. (1991). The sequences diverged, however, 5’ of nucleotide 295 of our sequence (see Fig. 1A, showing the first 360 nucleotides of XBMP-4.1). This 5’ sequence includes an in-frame ATG at nucleotide 267 (italics), which would add 12 amino acids to the proXBMP-4 of K öster et al. (1991) which begins at our nucleotide 303. We suggest, however, that XBMP-4.1 is derived from an unprocessed RNA. Firstly, there is a 3’ splice-site consensus sequence 8 nucleotides upstream from nucleotide 303 (bold). Secondly, sequence analysis of XBMP-4.4, which is longer than XBMP-4.1, shows that the extreme 5’ region of this cDNA is homologous to human BMP-4 but that it then diverges in the same way as XBMP-4.1 (data not shown). It is possible that XBMP-4.4 contains the entire intron; our method for selecting the largest inserts (see Materials and Methods) may inadvertently have led us to isolate cDNAs representing unprocessed transcripts.

The amino acid sequence of XBMP-4 (Fig. 1B) shows 80% similarity with the human protein (66 mismatches out of 398), with the highest homology in the C-terminal TGF-β -like domain (2 mismatches out of 102). Both XBMP-4 and its human homologue possess an N-terminal signal sequence, characteristic of secretory proteins, and all four potential asparagine-linked glycosylation sites are conserved, with an additional site in Xenopus at Asn 237. Consistent with the expected secretion of this protein, when synthetic XBMP-4 RNA is translated in a message-dependent Xenopus egg cell-free system (Matthews and Colman, 1991), the resulting protein is both segregated within the endoplasmic reticulum and glycosylated (data not shown). Active human BMP-4 has been isolated and shown to comprise the C-terminal 116 amino acids (Hammonds et al., 1991), with cleavage from proBMP-4 occurring after the amino acid sequence RAKR. Since a similar sequence is found in our clone at amino acids 284 –287 (RSKR), we expect mature XBMP-4 to comprise a dimer of the C-terminal 114 amino acids. Fig. 1C shows a map of XBMP-4 indicating the probes used for RN Aase protection analyses and northern blots.

RN Aase protection analysis (Fig. 2A) showed that zygotic expression of XBMP-4 begins at the mid-blastula transition and persists at least until stage 34. As expected from PCR analysis and screening of the XTC cDNA library, transcripts are also present in XTC cells. Further analyses showed that low levels of maternal XBMP-4 mRNA are also detectable (Fig. 2B). This expression pattern differs slightly from that of BMP-4_I’ (K öster et al., 1991), where there is only a slight increase in RNA levels after the mid-blastula transition.

XBMP-4 transcripts were studied by northern blot analysis of RNA samples from fertilized eggs and late gastrula-staged embryos (Fig. 2C). As expected from the RN Aase protection analysis described above, we obtained a much stronger signal with late-gastrula RNA than with unfertilized egg RNA. Two transcripts were detected, one with an apparent size of 1.9 kb, and one of about 3.1 kb. This larger RNA may be the result of alternative splicing or may represent a close homologue of the 1.9 kb transcript. Such a homologue has been identified by K öster et al. (1991) and, from the sequence of this gene, we might anticipate hybridisation with our probe. However, Kôster et al. (1991) have provided evidence that the RNA for their clone is similarly represented in both the maternal and zygotic pool, while the larger transcript in Fig. 2C is clearly more abundant in the zygotic pool than in the maternal pool.

The spatial expression of XBMP-4 at the early gastrula stage was studied by dissection of embryos into five regions (animal pole, vegetal pole, dorsal marginal zone, lateral marginal zone and ventral marginal zone) followed by RN Aase protection analysis. Although in one experiment levels were lower in the animal pole, and in another in the lateral marginal zone, overall our results suggest that expression occurs throughout the early gastrula (Fig. 3).

To investigate the function of XBMP-4 during Xenopus development, we injected 1.5 -2.0 ng XBMP-4 RNA into the animal hemispheres of fertilized eggs. All embryos injected in this way developed normally up to the early gastrula stage and formed a dorsal lip, but the development of some of those receiving the higher concentrations became slower at this stage, and the blastopore did not close (Fig. 4A, B). Sections of embryos (Fig. 4C, D) in which the progress of gastrulation was delayed resembled those of embryos in which ectopic mesoderm had formed in the animal cap, in response to injection of Brachyury RNA (V. T. Cunliffe and J.C.S., unpublished data), or to intra-blastocoelic injection of inducing factors (Cooke et al., 1987; Cooke and Smith, 1989). Even when this did not happen, however, by stage 40 virtually all injected embryos appeared extremely abnormal (Fig. 4E, F). Superficially they resembled the ‘grade 0’ embryos caused by UV irradiation of the vegetal hemisphere of fertilized eggs, and we were able to use the ‘Dorso-anterior Index’ (DAI) of Kao and Elinson (1988) to reveal a graded response resulting from injections of different amounts of RNA (Table 1). In this index, ‘5’ represents a normal embryo, ‘0’ a completely ventra-lized case and TO’ a completely dorsalized individual, and our results showed that a 100-fold dilution of injected RNA (1500 pg to 15 pg) increased the mean DAI of the resulting embryos from 0.39 to 3.32. Control injections of antisense XBMP-4 RNA (see Fig. 4 and Table 1), and RNA encoding Vg1 (Dale et al., 1989) and prolactin (Leaf et al., 1990) did not perturb development; in each case the resulting embryos had a mean DAI of greater than 4.75 (data not shown).

More direct evidence that injection of XBMP-4 causes ventralization comes from histological examination of ‘grade 0’ injected embryos. In these cases neither notochord nor muscle could be identified; rather, they appear to produce an excess of ventral mesoderm including large numbers of cells resembling red blood cells (Fig. 4G, H). Lack of notochord and muscle was confirmed by whole-mount immunocytochemistry with the monoclonal antibodies MZ15 (Smith and Watt, 1985) and 12/101 (Kintner and Brockes, 1984) (data not shown), and a large reduction in muscle was also indicated by RN Aase protection analysis of injected embryos at stage 20 (data not shown) as well as northern blot analysis performed at stage 40 (Fig. 8). To determine whether XBMP-4 RNA injection causes production of additional ventral mesoderm, as well as loss of dorsal tissue, we examined levels of Xhox3 (Ruiz i Altaba and Melton, 1989) and globin (Banville et al., 1983) mRNA in injected embryos. Expression of both these genes is restricted to posterior and/or ventral mesoderm in normal embryos and the expression of both is elevated in UV-irradiated embryos (Cooke and Smith, 1987; Ruiz i Altaba and Melton, 1989), and we find that the same is true in embryos injected with XBMP-4 RNA (Figs 5 and 8).

These results indicate that injection of XBMP-4 RNA causes ventralization of Xenopus embryos, with a reduction in the amount of dorsal mesoderm being complemented by an increase in ventral mesoderm. One prediction from this is that injection of RNA into the dorsal side of the embryo should have a more dramatic effect than injection into the ventral side. We therefore injected 0.5 ng RNA into either a single dorsal or a single ventral blastomere at the 4-cell stage. Table 2 shows that injection into a dorsal blastomere did result in more extreme ventralization; whereas embryos resulting from dorsal injections had a mean DAI of 0.7, those resulting from ventral injections had a mean DAI of 2.67. Once again the mean DAI of control injections was greater than 4.75 irrespective of the site of injection.

In the experiments described above, XBMP-4 might cause ventralization of Xenopus embryos by interfering with the action of an endogenous dorsal mesoderm-inducing factor, and one plausible candidate for such a molecule is activin. When added to isolated animal pole cells in vitro activin induces a wide range of mesodermal tissues, including notochord and muscle, in a concentration-dependent manner (Green et al., 1990). By contrast, we would not expect XBMP-4 to inhibit the action of a ventral mesoderm-inducing factor such as bFGF (Slack et al., 1987; Green et al., 1990; Amaya et al., 1991). To examine this, animal caps derived from embryos injected with either sense or antisense XBMP-4 RNA, or from uninjected embryos, were treated with activin, FGF, or a control solution.

The first indication that injection of XBMP-4 RNA inhibits the response to activin was that the elongation of animal pole regions that is observed in response to this factor (Symes and Smith, 1987) was inhibited (not shown). When such animal caps were allowed to develop further, histological analysis showed that they did not form significant amounts of muscle or notochord but either appeared uninduced or resembled the weak ‘ventral’ inductions formed in response to low concentrations of FGF (Fig. 6 and Table 3). By contrast, injection of XBMP-4 RNA appeared to have little effect on induction by FGF, where the animal caps formed structures similar to those differentiating in response to FGF alone. Animal caps derived from embryos injected with XBMP-4 RNA and then allowed to develop in the absence of inducing factors appeared uninduced or showed weak ventral-like inductions, which were distinct from those obtained with FGF alone (Fig. 6 and Table 3). As controls, animal caps derived from uninjected embryos or embryos receiving antisense XBMP-4 RNA were treated with activin, FGF or control solutions; the results were similar to those reported by Green et al. (1990) (see Fig. 6 and Table 3).

The above data were confirmed by molecular analysis of treated animal caps. Induction of muscle-specific actin in response to activin was greatly reduced in animal caps derived from embryos receiving injections of XBMP-4 RNA (Fig. 7A). Animal caps derived from embryos injected with XBMP-4 RNA, but not exposed to exogenous growth factors, express very low levels of muscle-specific actin (Fig. 7A). As with the histological data, this suggests that XBMP-4 is itself an inducer of mesoderm. This is confirmed by data presented in Fig. 7B. Xhox3 is most strongly expressed in ventral and posterior mesoderm, and animal caps from embryos injected with XBMP-4, but not exposed to exogenous growth factors, express levels of Xhox3 similar to those occurring in response to bFGF. This suggests that XBMP-4 is an inducer of ventral mesoderm. These experiments also confirmed that XBMP-4 mRNA injection ‘ventralizes’ the response to activin, because such animal caps express high levels of Xhox3.

Xenopus embryos can also be ventralized by UV-irradiation of the vegetal hemisphere shortly after fertilization (Scharf and Gerhart, 1983), a procedure that blocks the formation of the dorsal mesoderm-inducing centre in the vegetal hemisphere (Gimlich and Gerhart, 1984). This block can be ‘reversed’ by exposing UV-irradiated embryos to solutions of lithium chloride during early cleavage stages (Kao et al., 1986). As a result, embryos become ‘hyperdorsal’, with dorsal mesoderm differentiating from all sectors of the marginal zone (Kao and Elinson, 1988). The first signs of hyperdorsalization are observed at the early gastrula stage, when the blastoporal lip of the invaginating mesoderm is found all round the vegetal hemisphere of LiCl-treated embryos but localized to the dorsal quadrant in normal embryos. LiCl is believed to act by sensitizing the response of animal cap cells to ventral mesoderm-inducing signals (e.g. FGF or low concentrations of activin), such that they respond as though they were exposed to a dorsal signal (e.g. a high concentration of activin: Kao and Elinson, 1988; Slack et al., 1988; Cooke et al., 1989).

We examined whether embryos ventralized by injection of BMP-4 RNA could be rescued in this way by exposing them to 0.3 M LiCI at the 64- to 128-cell stage. The results were assessed at stage 38 by the DAI index (Table 4) and by northern blot analysis using probes for actin and globin (Fig. 8). Both methods of analysis showed that LiCI treatment cannot rescue BMP-4 injected embryos. Table 4 shows that antisense-injected embryos were hyperdorsalized by exposure to LiCI, with a DAI of 7.78 compared with a control value of 4.94. Histological examination showed ‘that these embryos differentiated more notochord at the expense of muscle (data not shown), and a severe reduction in muscle differentiation is reflected in the reduced levels of muscle-specific actin expression shown in Fig. 8. By contrast, exposure of XBMP-4 injected embryos to LiCI had no effect on the DAI index (0.69 in the absence of LiCI compared to 0.39 in the presence of LiCI), and histological examination confirmed that the embryos remained ventralized, differentiating blood and mesothelium at the expense of muscle and notochord (data not shown). This is also confirmed in Fig. 8, which shows that exposure to LiCI has no effect on the enhanced expression of a tadpole βglobin in XBMP-4 injected embryos. Interestingly, however, in embryos injected with both sense and antisense XBMP-4 RNA, the blastopore lip appeared around all regions of the marginal zone as if presaging hyperdorsal development. Like our earlier observation that XBMP-4 injection can cause ventralization even though the blastopore lip forms normally (see above), this result suggests that XBMP-4 acts late to produce ventralized embryos.

The results described in this paper add a further level of complexity to the analysis of mesoderm induction in Xenopus. In a search for additional mesoderm-inducing factors, we have used the polymerase chain reaction followed by cDNA library screening to obtain a full-length cDNA clone for Xenopus bone morphogenetic protein 4 (XBMP-4). Like K öster et al. (1991) we observe low levels of maternal mRNA for XBMP-4, and this is followed by a dramatic increase in RNA at the late blastula stage. The increase reported by K öster et al. (1991) was less dramatic than that seen in our Fig. 2, and this may be because their expression study concentrated on BMP-4_r, while our sequence resembles their BMP-4_n. At the early gastrula stage XBMP-4 is expressed at similar levels in all regions of the embryo.

Bone morphogenetic proteins were originally identified in bone extracts by their ability to induce the formation of ectopic cartilage and bone following implantation in rats (Wang et al., 1988). BMP-4 (formerly called BMP-2b) was cloned because of its homology to the closely related BMP-2, a component of bone extracts (Wozney et al., 1988), and was subsequently shown to have similar activity (Hammonds et al., 1991). A recent analysis using in situ hybridization has implicated BMP-4 in diverse processes during murine embryogenesis, including development of the limbs, heart, face and pituitary gland (Jones et al., 1991). Although expression of this RNA in pregastrula mouse embryos was not identified, in early neurulae it was localized to the’ posterior and ventral mesoderm suggesting a role for BMP-4 in the development of mesoderm in vertebrates. The results we have presented for Xenopus suggest that this is indeed the case.

Using morphological, histological and molecular criteria, we show that overexpression of XBMP-4 by microinjection of transcripts into the animal hemisphere of the fertilized egg causes ‘ventralization’ of the resulting embryo. In these embryos, dorsal mesodermal tissues such as notochord and muscle are severely reduced and replaced by ventral mesodermal tissues such as blood. Consistent with this, injection of transcripts into the dorsal half of the early embryo has a more dramatic effect than injection into the ventral half. Furthermore, like K öster et al. (1991), we show that XBMP-4 can act as a ventral mesoderm-inducing factor. Thus, injection of XBMP-4 mRNA into the animal hemisphere of the Xenopus embryo followed by dissection of the animal cap at the mid-blastula stage results in the formation of ventral mesodermal cell types, as assayed both by conventional histology and by the level of activation of Xhox3, which in normal embryos is most highly expressed in posterior and ventral mesoderm (Ruiz i Altaba and Melton, 1989). In this respect XBMP-4 differs from the activins, to which it is closely related, because the activins preferentially induce anterior and dorsal mesoderm (Green et al., 1990). Rather, the action of XBMP-4 more resembles that of members of the FGF family (Slack et al., 1987; Kimelman and Kirschner, 1987).

It is possible that the ability of XBMP-4 to induce ventral mesoderm is responsible for its ability to ventralize whole embryos. In order to cause ventralization in this way it is necessary that the ventralizing effect of XBMP-4 is ‘dominant’ over treatment with dorsal mesoderm-inducing factors such as activin. This is observed in experiments in which animal caps derived from embryos receiving injections of XBMP-4 RNA are treated with activin, and the effect is demonstrated most dramatically by the observation that injection of XBMP-4 RNA can even ventralize embryos that subsequently are treated with the dorsalizing agent LiCl. This dominant effect of a ventral mesoderm-inducing factor contrasts with work in which activin and FGF are applied simultaneously to animal cap tissue; here, the ability of activin to induce animal pole cells to form Spemann’s organizer is not compromised by the ventral mesoderm-inducing factor bFGF (Cooke, 1989).

Superficially, XBMP-4-injected embryos resemble those resulting from UV irradiation of the vegetal hemisphere soon after fertilization (Scharf and Gerhart, 1983). UV irradiation blocks the cortical rotation which is responsible for the formation of a dorsovegetal inducing centre, which in turn induces dorsal organizer mesoderm from the overlying marginal zone (Gimlich and Gerhart, 1984). However, there are clear differences in the gastrulation movements of UV-irradiated and XBMP-4-injected embryos. UV-irradiated embryos form a ‘ventral-type’ blastoporal lip which appears late and synchronously around the whole circumference of the embryo (Scharf and Gerhart, 1983). By contrast, XBMP-4 injected embryos form a dorsal lip at the same time as controls. Although in some cases gastrulation then arrests, due to formation of ectopic mesoderm in the animal hemisphere of the embryo (see Cooke et al., 1987; Cooke and Smith, 1989; V. T. Cunliffe and J.C.S., unpublished data), in most cases the blastopore closes on a time scale similar to that of embryos injected with antisense RNA. XBMP-4 injected embryos also differ from UV-irradiated embryos in that they cannot be ‘rescued’ by exposure to LiCl, even though in both types of embryo LiCl causes the formation of a precociously complete blastoporal lip around the circumference of the entire vegetal hemisphere.

Together, these results suggest that the ventralization caused by XBMP-4 is a late event, occurring after the onset of gastrulation. The mechanism by which this occurs is unclear, although one possibility is that the action of XBMP-4 resembles that of polysulphonated compounds such as trypan blue and suramin. If injected into the blastocoels of embryos when the dorsal lip first appears, these compounds inhibit convergent extension movements while allowing blastopore closure. The resulting embryos fail to differentiate dorsal mesodermal tissues and become ventralized (Gerhart et al., 1989). If, however, these agents are injected after the end of gastrulation, when convergent extension has been completed, they have no effect on the differentiation of dorsal mesoderm. Like XBMP-4, but unlike UV-irradiation, both trypan blue and suramin can reverse the hyperdorsalizing effects of LiCl.

It is not clear how trypan blue and suramin inhibit convergent extension and cause ventralization, although suramin is known to disrupt the binding of some growth factors to their receptors. These include members of all three families of mesoderm-inducing factor such as FGF and TGF-β (Coffey et al., 1987), and members of the Wnt family (Papkoff and Schryver, 1990; Chakrabarti et al., 1992). It is possible that the continued presence of a member of one of these families is required for convergent extension and dorsal development and that an excess of exogenous XBMP-4 competes for the receptor for such a protein. This is under investigation.

The role of XBMP-4 in Xenopus development will remain unknown until it becomes possible to eliminate the protein or its receptor from the embryo. It is clear, however, that the action of XBMP-4 must be considered in the context of the other inducing factors known to be present in the embryo, such as members of the activin, FGF and Wnt families as well as other bone morphogenetic proteins. It will be a formidable task to understand how the functions of all these factors are coordinated to form the normal embryo.

This work is supported by the Medical Research Council. L.D. thanks Professor Alan Colman for support. J.C.S. thanks Dr Igor Dawid for the XTC cDNA library. We also thank Drs Rob Grainger and Margaret Saha for the Xhox3 probe and Roger Patient and Adnan Ruazi for the globin probe.

The nucleotide sequence described in this paper has been submitted to the EMBL database under the accession number X64538.