ABSTRACT
This review starts from the classical standpoint that there are at least two separable processes acting with respect to axis formation and tissue specification in the early Xenopus embryo: a UV-insensitive event establishing a postgastrula embryo consisting of three concentric germ layers, ectoderm, mesoderm and endoderm, all of a ventral character; and a UV-sensitive event producing tissue of a dorsal type, including somites, notochord and neural tissue, and concomitantly establishing the dorsoventral and anteroposterior axes. The experimental evidence suggesting the molecular basis of the dorsal and ventral pathways is reviewed.
(1) INTRODUCTION
Amphibian embryos have long been favoured organisms for the study of cell interactions in early morphogenesis, particularly with respect to the questions of axis formation and mesodermal tissue specification. Reviews of this subject have been numerous (Smith, 1989; Kimelman et al., 1992; Dawid, 1994; Sive, 1993; Kessler and Melton, 1994; Lemaire and Kodjabjchian, 1996), but recent studies have begun to suggest a shift from the text-book models of how this patterning is achieved and, therefore, require a fresh reappraisal of the data.
(2) THE THREE-SIGNAL MODEL
The pioneering work of Ogi (1967, 1969), and Nieuwkoop and colleagues (Nieuwkoop, 1969; Boterenbrood and Nieuwkoop, 1973), and more recently by Slack and Smith on Xenopus embryos (Smith and Slack, 1983; Slack et al., 1984; Smith et al., 1985; Dale and Slack, 1987), led to the three-signal model of mesoderm induction (Fig. 1). According to this model, the active signalling center causing mesoderm induction is the vegetal mass of the blastula-stage embryo. The evidence for this is that vegetal explants co-cultured with animal caps are able to divert animal cells from an epidermal to a mesodermal fate. In the model, the vegetal mass is further divided into dorsal and ventral halves, with the former inducing mesoderm of a dorsal character, including notochord and somites, while the latter induces ventral mesoderm, mesenchyme, lateral plate mesoderm and blood islands. These signals are considered to occur early in development, before the onset of zygotic transcription at the mid-blastula transition (MBT). The third signal in the model is viewed as a later event, passing horizontally from the dorsal equatorial region (known as the Spemann organizer region), in recognition of the finding that, when this region is transplanted to the ventral side of an early gastrula embryo, it is able to induce an ectopic axis (Spemann and Mangold, 1924). This signal expands the size of the area of dorsal mesodermal tissue.
Over the past decade, substantial effort has been made in many laboratories to characterize the molecules involved in mesoderm induction and axial patterning, and to extend the early observations on which the three-signal model was based. However, mesoderm-inducing molecules constituting the first two signals, with the characteristics suggested by this model, remain unidentified. Several modifications have been made to the original model (Kimelman et al., 1992; Sive, 1993; Watabe et al., 1995; Sakai, 1996). While we are still far from a complete understanding of the molecular mechanisms patterning the early Xenopus embryo, significant advances have been made. This review aims to summarize these advances. In doing so, the simplicity of the old model is lost, but what emerges is the framework of molecular and cellular interactions that pattern the early embryo.
(3) THE DORSAL PATHWAY
(a) UV-irradiation experiments
Classical UV irradiation experiments have indicated that there are at least two aspects to the patterning of early amphibian embryos (Fig. 2): a UV-insensitive event establishing a postgastrula stage, consisting of three concentric germ layers, ectoderm, mesoderm and endoderm, all of a ventral pattern; and a UV-sensitive event producing tissue of a dorsal type, including somites, notochord and neural tissue, and concomitantly establishing the dorsoventral and anteroposterior axes (Grant and Wacaster, 1972; Malacinski et al., 1974). UV irradiation is effective at blocking this dorsal pathway when it is applied to the vegetal pole at two different times: either to fertilized eggs shortly after sperm entry, or to the vegetal pole of full-grown ovarian oocytes (Holwill et al., 1987; Elinson and Pasceri, 1989). This latter finding suggests that a component/s of the dorsal pathway, or a factor indirectly influencing that pathway, must be established during oogenesis and localized to the cortex of the vegetal pole before maturation. Although there has been no molecular characterization of this oocyte dorsalizing activity, its vegetal localization has been confirmed by cytoplasmic transfer experiments (Holowacz and Elinson, 1995).
The second, postfertilization, UV-sensitive period has been intensively studied (Vincent et al., 1986, 1987; Yuge et al., 1990; Fujisue et al., 1993; Sakai, 1996). Several lines of evidence indicate that, at this time, the UV target is cytoskeletal; UV disrupts cortical cytoplasmic movements by disorganizing microtubules and prevents the normal displacement of a dorsalizing activity to the dorsal side of the embryo (Gerhart et al., 1989; Houliston, 1994). The nature of this dorsalizing activity or its relationship to the oocyte factor/s is not understood, although they are clearly separable, as UV-irradiation of the oocyte does not disrupt cortical rotation after fertilization (Elinson and Pasceri, 1989). One approach to this question has been to identify molecules that are expressed in early embryos and that are able to rescue the UV-ventralized phenotype. A surprisingly large number of molecules fit this description (Table 1), and current research is directed towards understanding their relative importance and their relationship to each other in the endogenous dorsal pathway. Most of these molecules (* in Table 1) are absent or at very low abundance as maternal mRNAs, and are only strongly expressed zygotically, after MBT. It seems likely therefore that these molecules are downstream of UV-sensitive maternal events.
One standard method towards this end has been to overexpress the candidate proteins shown in Table 1 by injecting mRNAs or dominant negative forms of those mRNAs into early cleavage-stage embryos. This has suggested roles in dorsal axis formation for several molecules, including activin, goosecoid (gsc), Siamois, Xwnt-8 and Vg1. However, Xenopus embryos consist of labile, pluripotent cells until the midblastula stage (Heasman et al., 1984; Wylie et al., 1987; Snape et al., 1987) and overexpression or ectopic expression of signalling components or receptors may reveal pathways that would not normally function in undisturbed embryos. Problems in interpretation of the results may also arise if the injected dorsalizing molecules can mimic, but are not necessarily the same as, endogenous molecules in these pluripotent cells, or in the case of dominant negative constructs, act promiscuously to block signalling by related molecules(Kessler and Melton, 1994; Schulte-Merker et al., 1994). A further complication is that the overexpressed protein may have activity at any time from the cleavage to the neurula stage, which may not reflect the timing of activity of endogenous proteins. Thus this approach has not resolved the question of which of the candidate molecules are active in the embryo, or when they take part in dorsal axis formation.
(b) β-catenin
An alternative method to identify the endogenous role of maternal gene-products is to deplete the protein or mRNA from the oocyte and study the effect of that depletion on development. This method uses antisense oligodeoxynucleotides to target-specific mRNAs, which are then cleaved by endogenous RNase H (Dash et al., 1987; Dagle et al., 1991; Heasman et al., 1994a,b). It is not used to test the function of zygotic gene products, as any degraded mRNAs would be replaced by new transcription after MBT. Recently, a combination of overexpression and depletion experiments have identified some of the essential ingredients of the dorsal, UV-sensitive pathway. Specifically, dorsal axis formation depends upon cytoplasmic components homologous to those of a signalling pathway in Drosophila initiated by the Wingless secreted protein (Fig. 3). This wingless-initiated pathway has been shown to be of critical importance in the patterning of the Drosophila embryonic epidermis (DiNardo et al., 1988; Martinez-Arias et al., 1988; Bejsovec and Martinez-Arias, 1991), wing (Morata and Lawrence, 1977; Phillips and Whittle, 1993), leg (Diaz-Benjumea and Cohen, 1994) and midgut (Yu et al., 1996). Several Xenopus members of the Wnt family (Xwnt genes) are able to rescue the ventralized phenotype of UV-irradiated Xenopus embryos (see Table 1).
β-catenin is one of these essential ingredients. It was first identified as a cell membrane-associated protein in vertebrate cells, necessary for cadherin-mediated adhesion (Ozawa et al., 1989), and was later shown to be a vertebrate homolog of the Drosophila protein, Armadillo. Although armadillo was initially identified in a screen for segment polarity mutants, and is a component of the Wingless protein-initiated segment polarity signalling pathway (Peifer et al., 1992, 1994), several lines of evidence have shown that it has roles both in signalling and adhesion. In Drosophila ovary and in the early embryo, armadillo mutants have defects in cell-cell adhesion as well as signalling (Peifer et al., 1993; Cox et al., 1996). Xenopus embryos depleted of maternal β-catenin have a signalling defect: they develop without dorsal structures, including somites, notochord and neural tubes, and resemble the most severe cases of axis-deficient UV-irradiated embryos (Heasman et al., 1994b). In these experiments, cell-cell adhesion is not altered, but β-catenin is depleted only from the supernatant fraction, leaving membrane-associated β-catenin presumably available for an adhesive role (Kofron et al., 1997). Cytoplasmic β-catenin and Armadillo are thus essential for signalling, and the membrane-associassociated forms for adhesion. The distribution of β-catenin can be altered simply by overexpression of cadherin (Heasman et al., 1994b), which sequesters β-catenin at the cell membrane and depletes the cytosolic pool (Fagotto et al., 1996). However, it seems unlikely that endogenous cadherins regulate the cytosolic pool, as Xenopus embryos that have been depleted of maternal EP cadherin, and are disaggregated at the blastula stage, go on to develop with normal axes and dorsal structures when zygotic EP cadherin begins to accumulate (Heasman et al., 1994a). Recent studies suggest that cytoplasmic and membraneassociated Armadillo may be regulated separately by different kinases (Pai et al., 1997). The dorsal deficiencies of β-catenin-depleted embryos can be rescued by the injection of β-catenin mRNA, but not by Xwnt-8, even though Xwnt-8 has strong dorsalizing activity in wild-type embryos (Heasman et al.,1994b). This indicates that, as in Drosophila, β-catenin may lie downstream of a Wnt signal.
(C) Molecules upstream of β-catenin
The serine-threonine kinase zeste-white 3 zw3; (Peifer et al., 1994) acts as an inhibitor of the Wingless pathway in Drosophila, by altering the stability of soluble Armadillo (Pai et al., 1997). It also has a Xenopus homolog (glycogen synthase kinase, GSK3) that inhibits the dorsal pathway. A dominant negative form of GSK3 mRNA causes ectopic axis formation in Xenopus embryos, but is unable to rescue β-catenin-deficient embryos (Wylie et al., 1996), placing GSK3 in a similar upstream position to β-catenin, as its counterpart zw3 is to Armadillo in Drosophila (He et al., 1995; Pierce and Kimelman, 1996; Dominguez et al., 1995).
Unlike Drosophila, the components of the signalling pathway upstream of GSK3 and β-catenin have not been established in Xenopus. The Xenopus homolog of dishevelled, Xdsh, has the ability to induce complete ectopic axes (Sokol et al., 1995). A dominant negative form of this protein is active in blocking the signalling pathway stimulated by ectopically injected Xwnt-8 mRNA. Surprisingly, overexpression of this dominant negative protein alone does not prevent dorsal axis formation, suggesting that Xdsh is not an essential component of the endogenous dorsal pathway, although it may be important for convergence extension movements at the gastrula stage (Sokol, 1996).
β-catenin has many potential interactions. As well as its association with cadherin and α-catenin, it has been shown to bind to EGF receptor (Hoschuetzky et al., 1994), c-erbB-2 (Kanai et al., 1995), fascin (Tao et al., 1996) and the cytoplasmic protein adenomatous polyposis coli (APC) protein. APC is expressed in early Xenopus embryos (Vleminckx et al., 1997), where it may be involved in regulating the levels of βcatenin, as it does in cultured cells (Munemitsu et al., 1994; Papkoff et al., 1996; Hayashi et al., 1997). APC is a substrate for GSK3, and its phosphorylation regulates binding of APC to β-catenin (Rubinfeld et al., 1996). In Drosophila, however, zygotic APC is not essential for Armadillo function (Hayashi et al., 1997) and overexpression of APC in Xenopus embryos causes dorsalization, rather than acting in a negatively regulatory fashion (Vleminckx et al., 1997). A second role for APC could be in associating β-catenin with the microtubule cytoskeleton during cortical rotation in the first cell cycle, as it is known to associate with microtubules in other cell types (Munemitsu et al., 1994; Smith et al., 1994).
There is also no evidence yet that defines which Xwnt, if any, initiates the pathway in Xenopus. At least two Xwnt mRNAs are present in the oocyte and early embryo and have dorsalizing activity when injected into embryos (Xwnt 11: Ku and Melton, 1993; Xwnt-8b: Cui et al., 1995). But the roles of the endogenous molecules have yet to be defined. Recent studies have indicated that homologs of the Drosophila frizzled gene, the putative Xwnt receptor, can interact with Xwnt family members (Yang-Snyder et al., 1996; He et al., 1997). Functional studies of these frizzled family members may help answer the question of whether dorsal signalling in Xenopus involves a signal transduction cascade homologous to the complete wingless signalling pathway or not.
Thus the steps upstream of β-catenin and GSK3 are not understood, and there is no clear link yet between them and the dorsalizing cytoplasmic localizations in the oocyte and 1-to 16-cell-stage embryos. Indeed, we do not know if UV treatment of the oocyte or embryo damages the Xwnt/ β-catenin pathway directly or indirectly. Two lines of evidence point to the UV effects being indirect. Firstly, UV irradiation of the vegetal poles of fertilized eggs causes localization of Siamois, a transcription factor downstream of β-catenin, in the vegetal rather than the equatorial region of the embryo (Brannon and Kimelman, 1996). This supports the hypothesis that UV irradiation of the egg disrupts cytoplasmic movements, not the Xwnt/ β-catenin pathway directly. Secondly, UV-irradiated oocytes can be dorsalized (form exaggerated dorsal and anterior structures) in the same way as wild-type oocytes and embryos, by lithium treatment at the 32-cell stage (see below), suggesting that the dorsalizing pathway is not irreparably damaged by UV treatment of the oocyte (Elinson and Pasceri, 1989).
(d) Lithium effects
Unlike UV treatment, the molecular basis of lithium treatment has been clarified. Classical experiments showed that exposure of embryos to a pulse of lithium ions at any time between early cleavage and MBT causes dorsalization (Kao et al., 1986). Lithium has been considered to act by inhibiting the phospho-inositol (PI) cycle, as it inhibits the enzyme, inositol monophosphatase. This hypothesis was supported by the observation that coinjection of myoinositol with lithium prevents dorsalization of Xenopus embryos (Busa and Gimlich, 1989). However, a direct inhibitor of inositol monophosphatase does not dorsalize embryos (Atack et al., 1993). Recently, the dorsalization effects of lithium have been shown to be due to its effects on the Xwnt/ β-catenin pathway. Specifically, lithium inhibits the enzyme GSK3, which in turn causes the accumulation of β-catenin (Klein et al., 1996; Stambolic et al., 1996; Hedgepeth et al., 1997).
(e) Molecules downstream of β-catenin
While events upstream of beta catenin are still unclear, recent studies have succeeded in revealing the interaction immediately downstream (reviewed in Nusse, 1997). XTcf-3 is a transcription factor of the LEF/Tcf family whose association with β-catenin was first shown using the yeast two-hybrid system (Molenaar et al., 1996; Behrens et al., 1996). It is a ubiquitous protein in early Xenopus embryos. β-catenin bound to XTcf-3 is localized in nuclei, and binding of β-catenin to XTcf-3 is required for the activation of transcription in an in vitro assay. Furthermore, a dominant negative form of the mRNA coding for XTcf-3 causes ventralization of embryos similar to that caused by depletion of β-catenin (Molenaar et al., 1996). Thus the result of this dorsal signalling pathway is the localization of a specific transcription factor in nuclei. It is likely that this mechanism of signal transduction is conserved throughout metazoa. A Drosophila homolog of XTcf3, dTCF, has been shown to interact with Armadillo to affect expression of transcription factors, Engrailed and Ultrabithorax (van de Wetering et al., 1997); and a human family member hTcf-4 is expressed in colonic epithelium and transactivates transcription when bound to β-catenin (Korinek et al., 1997).
Recent studies have begun to define candidate zygotic genes that may be activated, directly or indirectly, by maternal XTcf3-β-catenin. One such candidate is the transcription factor, goosecoid (gsc). gsc is normally localized to the dorsal marginal zone, immediately zygotic transcription starts, and has head-organizing properties (Cho et al., 1991). β-catenin-deficient embryos have much reduced expression of gsc mRNA (Heasman et al., 1994b). Gsc has a proximal sensitive element in its promoter that is responsive to Xwnt signalling, defining this region as a possible target for interaction with XTcf3-β-catenin (Watabe et al., 1995). Other candidate downstream genes are suggested by mRNA overexpression experiments. β-catenin overexpression in animal caps causes the transcription of the transcription factor, Siamois, and the TGF-β class signalling molecule, Xnr3 (Carnac et al., 1996; Brannon and Kimelman, 1996; Fagotto et al., 1997). Recent dominant negative experiments also point to Siamois being downstream of β-catenin and essential for dorsal axis formation. (Fan and Sokol, 1997). Interestingly, UV irradiated embryos still express Siamois, but its expression is restricted to vegetal cells, suggesting the mislocalization of β-catenin-XTcf3 in these embryos (Brannon and Kimelman, 1996). Further studies are required to determine whether these inter-actions are direct or indirect, and whether the zygotic genes downstream of the dorsal pathway are the same or different in dorsal animal, equatorial and vegetal cells. In Drosophila development, the wingless signalling pathway has been shown to regulate the expression of several different genes in adjacent cell layers of the developing gut, ultimately determining the cell fate of endoderm and ventral mesodermal cells (Hoppler and Bienz, 1995; Yu et al., 1996). It seems likely that similar complex interactions occur in the Xenopus gastrula to specify dorsal ectodermal, mesodermal and endodermal fates.
(f) Where does the dorsal pathway act in the embryo?
The site where theβ-catenin/XTcf3 interaction happens has previously been suggested by numerous cytoplasmic and cell transfer experiments, indicating that, after the 16-cell stage, the dorsal side of the embryo contains the active dorsalizing activity. However, the exact localization of dorsalizing activity in vegetal, equatorial or animal cells remains controversial. The three-signal model emphasized the importance of the dorsal vegetal cells of the pregastrula embryo, the ‘Nieuwkoop center’. They have the capacity to induce animal cells to form dorsal mesodermal tissue, while themselves differentiating into endodermal tissue (Ogi, 1967, 1969; Nieuwkoop, 1969; Nakamura et al., 1970; Sudarwati and Nieuwkoop, 1971; Gimlich and Gerhart, 1984; Dale and Slack 1987). However, the capacity to produce dorsalizing signals is not restricted to this zone. Cell transfer experiments show that the dorsal equatorial cells from the 32-cell stage onwards have the ability to rescue UV-ventralized embryos (Gimlich, 1986) and fate map experiments show that these cells are fated to form dorsal mesodermal derivatives in the undisturbed embryo (Vodicka and Gerhart, 1995). Furthermore, experiments involving depletion and overexpression of β-catenin demonstrate that animal cells, as well as equatorial and vegetal cells, secrete dorsal signals (Heasman et al., 1994b; Wylie et al., 1996). Specifically, animal caps that are either wild-type or overexpressing β-catenin dorsalize the equatorial regions and vegetal masses of β-catenin-depleted embryos in co-culture experiments.
Recently, immunostaining studies have confirmed that the site of nuclear localization of β-catenin is on the dorsal side. One study showed localization of the protein in nuclei from the 8-cell stage onwards (Larabell et al., 1997), and a second study indicated localization in a broad dorsal area of the midblastula, centered on the equatorial region and most conspicuous in superficial nuclei (Schneider et al., 1996). Thus the asymmetry established by the wingless-type pathway may be broadly across the dorsal region of the embryo and not restricted to the Nieuwkoop center.
(4) TGF-β FAMILY GROWTH FACTORS AND MESODERM FORMATION
Family members expressed in early Xenopus embryos
Members of the TGF-β class of growth factors were first implicated in mesoderm formation by the discovery that a homolog of activin A was the active ingredient of mesoderm-inducing XTC culture medium (Smith et al., 1990). Further study suggested that activin is unlikely to play a role in mesoderm formation in vivo, as inhibition of activin activity with follistatin does not prevent mesoderm formation in early embryos (Schulte-Merke et al., 1994). Since then, two further types of TGF-β family members have been shown to be expressed maternally and have mesoderm-inducing activity; Vg1 and BMPs. Other TGF-β class growth factors, Xnr1-Xnr4 (see Table 1), are expressed after zygotic transcription starts. Only the maternal components most likely to be involved in early patterning events are considered further here. Vg1 is a vegetally localized maternal mRNA that becomes distributed in the vegetal half of the embryo after fertilization (Rebagliati et al., 1985; Weeks and Melton, 1987). While the processed, active form of the Vg1 protein cannot be detected in early Xenopus embryos, artificially processed Vg1 can rescue UV-ventralized embryos completely (Thomsen and Melton, 1993). The possibility therefore exists that localized processing of Vg1 in dorsal cells is important in dorsal axis formation. Vg1 has also been suggested to be involved in left/right axis and endoderm specification (Hyatt et al., 1996) although these experiments use overexpression of the artificially secreted form of the protein.
Several BMP (bone morphogenetic proteins) mRNAs are present in the early embryo as maternal transcripts (BMP2, BMP4 and BMP7), although no protein data are presently available (Koster et al., 1991; Nishimatsu et al., 1992; Dale et al., 1992; Jones et al., 1992). Overexpression studies show that they cause the formation of ventral mesodermal derivatives, including mesenchyme, blood islands and muscle but not notochord (Dale et al., 1992; Jones et al., 1992; Graff et al., 1994; Suzuki et al., 1994; Hemmati-Brivanlou and Thomsen, 1995). BMPs are implicated in ventral signalling events and are considered in a later section.
(b) TGF-β signal transduction
In contrast to the lack of information on signal reception in the Xwnt/ β-catenin dorsalizing pathway, a great deal is known about the interaction of TGF-βfamily members and their receptors (reviewed by Massagué, 1996; Fig. 4). Indeed, strong evidence for the involvement of TGF-βclass molecules in mesoderm formation comes from studies involving dominant negative forms of the receptors. The first Xenopus TGF-β receptor to be cloned was an activin type II receptor, whose truncated dominant negative form disrupts embryonic develpment at the gastrula stage, reducing mesoderm formation (Hemmati-Brivanlou and Melton, 1992) and enhancing neural tissue (Hemmati-Brivanlou and Melton, 1994). More recently, this mutation has been shown to block not only activin, but also Vg1 and BMP signalling, suggesting that a number of growth factors can bind this receptor (Kessler and Melton, 1994; Schulte Merker et al., 1994). In general, interpretations of the relative importance of TGF-β molecules has been hampered by the promiscuity of activity of the growth factors and their receptors in overexpression experiments. Activin and BMP have heterodimeric receptors consisting of receptor serine/threonine kinases from two subclasses. In other systems, there is evidence (1) that several type II receptors bind more than one ligand: activin and BMP7 bind ActR11 (Yamashita et al., 1995) and BMP2 and BMP7 bind BMPRII (Ten Dijke et al., 1994a; Liu et al., 1995), and (2) that type I receptors may bind more than one type II receptor: ALK2, ALK3 and ALK6 all interact with both ActRII and BMPRII (ten Dijke et al., 1994b; Liu et al., 1995; Yamashita et al., 1995). So far, the type II receptor for BMP has not been isolated in Xenopus, but activin (XALK4) and BMP type I receptors have been characterized (Chang et al., 1997; Graff et al., 1994). Expression of constitutively active forms of these receptors induces mesoderm in animal caps.
Although the relative importance of these receptors in endogenous mesoderm formation is still unclear, there is evidence that they are not functionally redundant. While dominant negative mutations of both type I and type II activin receptors block mesoderm formation in the equatorial regions of embryos, only the dominant negative ActRIIB receptor neuralizes ectodermal tissue (Chang et al., 1997). One model proposed recently to explain this difference is that activin signals through a dimeric receptor consisting of ActRIIB and ALK4 to cause mesoderm formation, while BMP2/BMP4 interact with dimers of ActRIIB and BMPR1 to induce mesoderm, and interacts with BMPR1 and an unknown type II receptor to signal ectoderm to differentiate as epidermis (Chang et al., 1997).
(c) Where and when does activin/Vg1 signalling occur in the embryo?
Although maternal as well as zygotic TGF-β family members and their receptors are implicated in roles in initiating zygotic gene expression, the extent to which they are essential for dorsal axis specification and mesoderm formation has not yet been shown, nor is their temporal and spatial activity understood. Only Vg1 is localized in the vegetal region as predicted by the three-signal model and no dorsal/ventral differences in its expression or activity have so far been detected. However, there is strong evidence that a maternal activin/Vg1 signalling pathway is important in regulating the transcription of a zygotic gene product, Mix2. Mix2 is a homeobox transcription factor that is expressed broadly across equatorial and vegetal regions at the gastrula stage (Vize, 1996). Although its function is not known, a highly related gene Mix.1, with a similar expression pattern causes ventralization of mesoderm, when overexpressed on the dorsal side of early embryos (Mead et al., 1996). An elegant series of experiments identified the signalling pathway that activates Mix2 expression (Chen et al., 1996, Fig. 4b). Activin interaction with its receptor leads to phosphorylation of the type 1 receptor and the subsequent phosphorylation of a cytoplasmic protein, Smad2, in the MAD (mothers against dpp) family. Smad2 then becomes localized in nuclei and binds to a Winged helix group transcription factor FAST1 (forkhead activin-sensitive transcription factor). FAST1 is a maternal protein and activates zygotic transcription of Mix2 in the absence of new protein synthesis.
A similar signalling cascade is also likely to exist for BMPs, as another MAD-related gene, Smad1 has been shown to be active in response to BMP signalling in Xenopus embryos (Graff et al., 1996; Fig. 4c). The transcription factors mediating BMP signalling are unknown, but the target genes are likely to include the zygotic transcription factors Xvent1 (Xom) and Xvent2 (Gawantka et al., 1995; Ladher et al., 1996; Onichtchouk et al., 1996; Papalopulu and Kintner, 1996; Schmidt et al., 1996; Tidman-Ault et al., 1996) and Mix.1 (Mead et al.,1996).
(d) What is the relationship between the Xwnt/ β-catenin/XTcf3 dorsal signalling pathway and the Smad2/FAST pathway?
While β-catenin/XTcf signalling is restricted to the dorsal segment of the early embryo, the localization of TGF-β type dorsalizing signalling pathways is not known. One important question concerns the relative roles of the dorsal β-catenin/XTcf3 signalling pathway and the activin/Vg1 pathway. Epistasis experiments have shown that BVg1 (the secretion competent-mutated form of Vg1) and activin can induce dorsal mesoderm formation in β-catenin-depleted embryos and animal caps (Wylie et al., 1996; Fagotto et al., 1997), suggesting that these embryos that lack the β-catenin/XTcf3 pathway have intact activin/Vg1 signal transduction pathways, presumably including an Smad2/FAST1 pathway. Even though the pathway is intact, no activin/Vg1 induction of dorsal mesoderm occurs in undisturbed β-catenin-depleted embryos, as they develop without forming this tissue (Heasman et al., 1994b). One model that would explain this is that, in the absence of Xwnt signalling, ventral (BMP) signals override throughout the embryo, and block or prevent the initiation of the endogenous Smad2 /FAST1 pathway.
5. THE VENTRAL PATHWAY
(a) BMPs
β-catenin-deficient and UV-irradiated embryos form mesoderm that does not contain muscle tissue or notochord, but contains mesenchyme, (expressing fibronectin) mesothelial tissue and blood tissue (expressing globin) (Smith and Slack, 1983; Heasman et al., 1994a,b). The embryos lack dorsal/ventral and anterior/posterior axes. Several growth factors have been implicated in this ventral pathway, of which the best candidates are BMPS. BMP4 mRNA is first detectable by in situ hybridization at the gastrula stage, in animal and ventral equatorial cells. UV-irradiated embryos accumulate BMP4 transcripts prematurely and throughout the equatorial zone (Fainsod et al., 1994). Furthermore, suppression of BMP4 activity in the UV-irradiated embryo using antisense BMP4 mRNA leads to a restoration of axial structures, although not head structures (Steinbeisser et al., 1995), suggesting that the overactivity of the BMP pathway is at least partly responsible for the UV phenotype. Similarly, dominant negative BMP receptor expression rescues axial structures in β-catenin- depleted embryos (Wylie et al., 1996).
More is known about the effects of BMP signalling or of blocking BMP signalling than about the temporal and spatial localizations of those signals in the embryo. Activation of BMP signalling in mesodermal tissue causes ventral mesoderm to form (Dale et al., 1992; Jones et al., 1992; Fainsod et al., 1994) and inactivation of the pathway results in notochord and somite formation rather than ventral mesodermal tissue (Graff et al.,1994, Maeno et al., 1994, Suzuki et al., 1994). There is some evidence that dorsal signalling dominates in the marginal zone of the wild-type embryo, as ventral marginal zones co-cultured with dorsal marginal zones become dorsalized (form notochord and somite), while ventral marginal zones are unable to ventralize dorsal marginal zones (Smith and Slack, 1983; Dale and Slack, 1987).
As well as BMP4, two other maternal BMPs have been described in early Xenopus embryos (see above). The relative roles of the three BMPs is not known, nor is it clear when the pathway/s act. Further work is required to determine whether a maternal BMP is essential for ventral signalling, as the maternal β-catenin/XTcf3 is for the dorsal pathway. One possibility is that the activity of both the ventralizing and dorsalizing pathways is exclusively maternal and activates zygotic genes in a specific regional fashion. Alternatively, ventral signalling may occur only after MBT to modulate mesoderm formation. Depletion experiments targetting either maternal or zygotic BMPs specifically should resolve this issue.
As with dorsal differentiation, this ventralization pathway appears to be highly evolutionarily conserved (reviewed in deRobertis and Sasai, 1996). BMPs are homologs of the dorsoventral morphogen in Drosophila, the decapentaplegic gene, dpp. (Ferguson and Anderson, 1992a,b). dpp is essential in leg, wing and gut patterning (reviewed by Bienz, 1994; Massague et al., 1994; Rusch and Levine, 1996), as well as acting as a morphogen in dorsoventral patterning of the gastrula (Ferguson and Anderson, 1992; Wharton et al., 1993). Recent studies in Xenopus indicate that BMP4 can also act as a morphogen, in that different doses of BMP pattern three distinct domains in the gastrula marginal zone (Dosch et al., 1997).
(6) THE INTERACTION OF THE DORSAL AND VENTRAL SIGNALLING PATHWAYS
(a) Antagonism between extracellular proteins
The patterning mechanism that has emerged in recent years consists of an interaction not between animal and vegetal cells, but between dorsal and ventral cells, as predicted by the third signal of the original three-signal model. When this interaction occurs is not known, because the timing of the ventral signalling pathway is not yet worked out, but it is likely to occur after MBT, when β-catenin-XTcf3 activates transcription. The secreted proteins noggin and chordin are transcribed in the Spemann organizer region and are dorsalizing molecules, in that they can convert ventral mesodermal tissue to dorsal mesodermal types when overexpressed on the ventral side of the embryo (Smith and Harland, 1992; Sasai et al., 1994). Although there is no direct evidence as yet, these proteins are generally assumed to be downstream of the Xwnt/β-catenin/XTcf3 pathway. Interestingly, noggin and chordin can also alter the pattern of differentiation of animal cells. They induce both neural tissue and endoderm to form from animal cap tissue without causing mesodermal differentiation (Smith and Harland, 1992; Sasai et al., 1994; Bouwmeester et al., 1996). Recently, evidence for the mechanism of action of noggin and chordin has been provided by the demonstration that both proteins interact directly with BMP in vitro and can block the binding of BMP to its receptor (Zimmerman et al., 1996; Piccolo et al., 1996). Thus one hypothesis is that animal cells form neural tissue if BMP signalling is blocked by noggin and chordin, and marginal cells become notochord and somite if BMP signalling is blocked (deRobertis and Sasai, 1996). This suggests that the lineage specification of blastula cells may depend on two factors: (1) whether the cells are animal or equatorial in position (set up by maternal localizations in the fertilized egg or by cleavage-stage cell interactions) and (2) whether the BMP/Smad1 signalling cascade that results in ventralization is interrupted by dorsalizing molecules such as noggin and chordin. However, at present it is not possible to assign specific functions to these proteins.
(b) Antagonism between transcription factors
A second possible mode of regulation of the dorsoventral axis is at the level of antagonism between transcription factors. This is suggested by the finding that, in vitro, Siamois protein can heterodimerize with the similar Pax-like homeodomain protein, Mix.1 (Mead et al., 1996). Mix.1 is expressed in equatorial and vegetal cells immediately zygotic transcription starts, and overexpression causes the ventralization of mesoderm. Thus one attractive mechanism for Siamois’ dorsalizing activity is that it heterodimerizes with and inactivates Mix1 in dorsal vegetal cells at the late blastula stage, and in Spemann’s organizer at the gastrula stage, and thus suppresses the ventralizing activity of Mix1 homodimers (Mead et al., 1996).
(c) Regulation of the dorsal or ventral signal transduction pathways
While the previous two regulatory mechanisms are considered to act after zygotic transcription starts, a third possible control on the extent of dorsal or ventral differentiation may be regulation of the Xwnt/β-catenin/XTcf3 and/or the Smad1/BMP signal tranduction pathways themselves. Clearly all the cells of the blastula have the potential to follow a dorsal or a ventral fate (as evidenced by lithium experiments and β-catenin depletion experiments), suggesting that one or other pathway must normally be suppressed on each side of the embryo. The first suggestion of regulatory molecules that might act to inhibit axis formation on the ventral side is the Xenopus homolog of the mouse protein, Axin, the product of the fused locus (Zeng et al., 1997). Mouse Axin inhibits the Xwnt pathway upstream of β-catenin and downstream of GSK3. Furthermore, dominant negative Axin expressed on the ventral side causes axis duplication and the functional domain deleted in the dominant negative construct is an RGS domain (Regulation of G protein Signalling). The only other evidence at the present time for the involvement of G proteins in dorsoventral patterning is the fact that artificial activation of phospholipase C in dorsal blastomeres causes ventralization (Ault et al., 1996).
(7) HOW MUCH PATTERNING HAPPENS BEFORE MBT?
For the dorsal pathway, at least three early asymmetries are known to be important before MBT. In the full-grown oocyte an as yet unidentified, UV-sensitive dorsal determinant is restricted to the vegetal pole. The second asymmetry is the dorsal cortical cytoplasmic displacement of a dorsal determinant that is generated by microtubule-based activity in the subcortical cytoplasm during the first cell cycle. Neither of these asymmetries have been explained in molecular terms. The third is the dorsal nuclear localization of β-catenin. The mechanism by which this is set up is uncertain, but is likely to involve the two previous asymmetries. However, although Xwnts are present maternally in Xenopus, there is no evidence yet of interactions involving Wingless-like molecules secreted by one cell affecting adjacent cells, as is the case in Drosophila. One Xwnt (Xwnt8b) is maternally expressed and has potent axis-forming ability, but has not been shown to be involved in the dorsal pathway (Cui et al., 1995). A second Xwnt, Xwnt 11 is vegetally localized but does not have strong axis forming capacity (Ku and Melton, 1993).
For the ventral pathway, the only known asymmetry before MBT is the animal and equatorial localization of BMP 7 mRNA (Hawley et al., 1995). This, coupled with the dorsally placed β-catenin/XTcf3 could be sufficient to dictate the spheres of influence of the two pathways before MBT and once transcription starts. Apart from these two pathways, are the cells of the early embryo naive and equivalent to one another? Single cell transplantation experiments showed that early blastula cells are pluripotent (Heasman et al., 1984). However, there is evidence for cytoplasmic localization in the full-grown oocyte, that suggests that considerable prepattern is set up before cleavage starts. For example, ligation experiments on fertilized eggs showed that components required for the activation of muscle-specific actin genes are localized in the subequatorial region (Gurdon et al., 1985). mRNAs for a variety of molecules including secreted proteins (Vg1, Xwnt11), germplasm-associated proteins (Xcat2) (Forristal et al., 1995) and transcription factors (VegT) (Zhang and King, 1996; Stennard et al., 1996; Lustig et al., 1996) are restricted to the vegetal pole of the oocyte early in oogenesis and the XreI transcription factor has been localized to nuclei in the animal hemisphere before MBT (Bearer, 1994). Recent studies on disaggregated cells suggests that there are also localizations in the equatorial zone (Lemaire and Gurdon, 1994). The ventral mesoderm marker, Xwnt-8, and the dorsal mesodermal marker, gsc, do not require cell interaction to be expressed in ventral and dorsal equatorial cells, respectively. Finally, animal cells cultured in isolation develop as neural cells, suggesting that they also have some prepattern that does not require cell interaction (Godsave and Slack, 1991). It is clear however, that for appropriate differentiation and axis formation, cell-cell interaction is essential. When do these interactions start?
(8) THE TIMING OF CELL-CELL INTERACTIONS IN DORSOVENTRAL PATTERNING
The current models of dorsoventral axis formation are based on the premise that cell interactions start early in embryogenesis (Smith and Slack, 1983; Kimelman, 1992; Sive, 1993) and, presumably, as there is no zygotic transcription until MBT, these interactions must cause post-transcriptional modifications to maternal mRNAs and proteins. Post-transcriptional regulation of expression has been shown for some maternal proteins including FGFR1 (Robbie et al., 1995) and Eg5 (Leguelle et al., 1991).
The early onset of cell interactions involved in axis formation has been suggested by at least four lines of reasoning, but none of these prove that this is the case. Nieuwkoop’s original studies showed that isolated animal caps could be diverted from their epidermal fate to form mesoderm by co-culture with vegetal masses. This suggested that the mechanism for dorsoventral patterning was that vegetal cells produce mesoderm-inducing signals. However, it has also been shown that the equatorial zones of embryos will differentiate into mesodermal tissue autonomously even when they are isolated from the vegetal masses as early as the 32-cell stage (Nakamura et al., 1970). So, if vegetal signals are the mechanism of induction of equatorial cells those signals must pass from the vegetal cells before that time. The second line of evidence stems from heterochronic Nieuwkoop recombination studies (Jones and Woodland, 1987). Here, vegetal masses from as early as the 16-cell stage were cultured with early gastrula animal caps and induced the caps to form mesoderm. The evidence is indirect, as induction could occur at any time during the co-culture period, although animal cap responsiveness has been shown to be lost during the gastrula stage. Also, it is not possible to rule out in these experiments that some mesoderm arose from the vegetal masses, particularly when these pieces were dissected from 16-cell (stage 5) embryos. Thirdly, cell transplantation experiments have shown that, when wild-type dorsal cells are transplanted into UV-ventralized embryos at the 32-cell stage, the transplanted cells can rescue the UV phenotype (Gimlich, 1986). While this experiment shows that early vegetal cells can act as a signalling center, it does not address the timing of this inductive activity, which might be any time after transplantation. Finally, the mRNA for the TGF-β growth family member Vg1 is known to be localized in the vegetal hemisphere of early Xenopus embryos. While injected BVg1 is potent in axis formation, no active Vg1 protein has been identified in early embryos. It has been reasoned that the Vg1 precursor protein may be processed in a limited, localized region of the embryo to direct dorsal axis patterning, in quantities too small to be detected by the means available.
More recent experiments provide direct evidence that inductive signals activating zygotic transcription of MyoD (a dorsal mesodermal marker) do not pass from vegetal areas to animal caps placed in contact with them until after zygotic transcription starts at MBT. In heterochronic recombinants of animal and vegetal masses cultured together for a short, one-hour culture period, vegetal masses from late blastulae are able to induce competent animal caps to transcribe MyoD. In contrast, vegetal masses from early (preMBT) blastulae do not induce identical animal caps to express MyoD over the same period of time (Wylie et al., 1996). Thus cell interactions involving signals from the vegetal mass in this case occur only after zygotic transcription has started. Whether this is true for other markers and for equatorial cell-cell interactions has yet to be shown, although the localization of β-catenin/XTcf3 in equatorial nuclei at the time of MBT suggests that XTcf3 activates zygotic genes at this time and argues against the early onset of signalling events.
(9) CONCLUSION
In summary, the three-signal model envisioned early dorsal and ventral signals emanating from vegetal cells and being acted upon by naive animal cells before MBT. Recent evidence however suggests that animal cells are not naive and that dorsal animal cells as well as equatorial and vegetal cells produce dorsalizing signals. Also, no signals have been identified that are produced specifically by dorsal or ventral vegetal cells, while evidence is accumulating for dorsal signal/s emanating from a dorsal segment after MBT, and of ventral signals produced by animal and equatorial cells. An alternative model to the three-signal model is shown in Fig. 5.
Here, the oocyte is suggested to be prepatterned into at least three zones, presaging the three germ layers. Cells inheriting animal cytoplasm become ectodermal tissue, cells inheriting equatorial cytoplasm differentiate as mesodermal tissue and cells inheriting vegetal cytoplasm differentiate as endodermal tissue. These states are however labile as, at the early blastula stage, all cells are pluripotent and single cells placed in ectopic sites will differentiate according to their new positions. After fertilization, the cylindrical symmetry is broken and a dorsal segment is established in the form of animal, equatorial and vegetal cells whose nuclei contain XTcf/β-catenin. Animal and equatorial cells also accumulate mRNA (BMPs) for future ventral signalling functions. At MBT, animal, equatorial and vegetal cells respond to ventral signals by activating zygotic transcription factors (eg Xvent1, Mix1) which direct ventral differentiation, appropriate to each germ layer. In the dorsal sector, maternal XTcf3/β-catenin directs zygotic transcription resulting in the expression of dorsalizing molecules (eg Siamois, gsc, Xnr3), which may differ in the animal, marginal and vegetal dorsal zones. One aspect of dorsal differentiation is the antagonism by dorsally secreted proteins (noggin, follistatin, chordin) of ventral BMP signals, in a germ-layer-specific manner. As well as these signals, individual cell fate may be directly or indirectly affected by the localized activity of other maternal transcripts (eg. Veg T, Xrel).
In the case of a UV-irradiated or β-catenin-deficient embryo, ventral signals predominate and the embryo does not develop D/V or A/P axes, as the oocyte’s symmetry is not broken and the dorsal sector is not established. The development of both dorsal and anterior structures depends on an intact maternal XTcf3/β-catenin signalling pathway.
This model is extremely fragile, leaving major questions unresolved including:
1. Is there a mesodermal state independent of the proposed dorsal and ventral pathways? If so, what is it?
2. Are maternal or zygotic BMPs (or both) essential for specifying ventral fates?
3. In the case of the dorsal signal, what is the primary activity that causes asymmetry of XTcfβ/beta catenin? In the case of the ventral signal, does a transcription factor similar to FAST1 bind to Smad2?
4. How do we account for the complexity of fine-grained expression patterns of transcription factors and secreted proteins in the Spemann organizer downstream of the dorsal and ventral signals?
5.What is the role of Vg1 and activin in embryonic patterning?
In conclusion, the patterning described here shows more similarities with pattern-forming mechanisms in Drosophila than were visible at the cellular level. In both Drosophila and Xenopus, maternal transcription factors specify regional activity of zygotic genes and, in both systems, similar signalling mechanisms are at work. We still do not understand how mesoderm forms in the early Xenopus embryo. Future emphasis will be placed not only on how the organizer organizes, but also on how, even without an organizer, the embryo generates three germ layers.
ACKNOWLEDGEMENTS
The financial support for this work has been provided by the Wellcome Trust, the Institute of Human Genetics, the Harrison Fund, and the NIH (1 RO1 HD33002-01). I am extremely grateful to Matt and Kelly Kofron for their help with the figures, to Steve Glaspey for assistance with references and to Chris Wylie for critical reading of the work, and essential intellectual contributions.