Dorsal-Ventral Patterning and Neural Induction in Xenopus Embryos

Inductive Properties of Dorsal Mesoderm

The starting point was provided by an experiment carried out by Hans Spemann and Hilde Mangold in 1924 in which they transplanted a small fragment of dorsal mesoderm into ventral mesoderm of the salamander gastrula (Spemann & Mangold 1924) and found it could induce a twinned embryo. They grafted a mesoderm fragment of different pigmentation, as in the experimental reproduction shown in Figure 1. The transplanted cells changed the differentiation of neighboring ventral cells in the host, inducing CNS, somite, and other axial components. In other words, both the ectoderm and the mesoderm became dorsalized. Research stimulated by this experiment has continued to this day, in particular with the molecular exploration of Spemann's organizer phenomenon (De Robertis & Aréchaga 2001, Grunz 2004, Stern 2004).

Open in a separate window

Figure 1

The Spemann-Mangold organizer experiment repeated in Xenopus laevis. (Top) control swimming tadpole; (bottom-right) Spemann organizer graft at the same stage. In the bottom-left embryo the size of the graft can be visualized as a white patch (Spemann organizer from an albino donor embryo) at early gastrula (vegetal view). The dorsal lip of the blastopore may be seen as a thin crescent opposite the graft.

Chordin, Noggin, and Xnr3

Chordin and Noggin are expressed in the Spemann organizer at gastrula and in notochord and prechordal plate at later stages. They encode BMP antagonists that bind to BMPs in the extracellular space, blocking binding to BMP receptors (Piccolo et al. 1996, Zimmerman et al. 1996). Work on the Noggin and Chordin secreted proteins has provided a molecular framework for understanding the function of many extracellular proteins. The three-dimensional structure of Noggin bound to BMP7 revealed that this BMP antagonist contains a cystine knot (Groppe et al. 2002). The cystine knot structural motif is found in many extracellular proteins, such as the TGF-β superfamily, luteinizing hormone (LH), follicle stimulating hormone (FSH), platelet-derived growth factor (PDGF), nerve growth factor (NGF), and others (Avsian-Kretchmer & Hsueh 2004). Because BMP and Noggin share cystine knots and conserved protein folds, it has been proposed that the ligand and its antagonist may have evolved from ancestrally related proteins (Groppe et al. 2002). Chordin is a large protein of about 1000 amino acids containing four cysteine-rich domains of about 70 amino acids each (called CR1 to CR4), which constitute BMP binding modules (Larraín et al. 2000). The structures of Chordin or CR domains have not been solved, but some predictions indicate a possible cystine knot (Avsian-Kretchmer & Hsueh 2004). CR repeats, also called von Willebrand factor C domains (vWF-C), are present in a large number of extracellular proteins, many of which have been found to regulate BMP or TGF-β signaling. Proteins containing CR domains include CTGF (connective tissue growth factor), Procollagens, Amnionless, Chordin-like/Ventroptin/Neuralin-1 and -2, CRIM-1 (cysteine-rich motor neuron protein), Nel (neural tissue protein containing Egf-Like domains), Nel-like 1 and 2, Keilin, and Crossveinless-2 (Garcia-Abreu et al. 2002).

On the dorsal side of the Xenopus embryo, Chordin protein is present at concentrations of 6 to 12 nM in the extracellular space (Piccolo et al. 1996). Because BMPs are expressed in the picomolar range in the embryo, Chordin alone should suffice to block BMP signaling on the dorsal side. However, the knockout of the Chordin gene in mouse causes only a small percentage of embryos to become ventralized at the gastrula stage (Bachiller et al. 2003). These infrequent embryos have a small neural plate and embryonic region, and an enlarged allantois (ventral mesoderm). Most chd^−/− embryos have a normal CNS and die at birth, mimicking a human malformation called DiGeorge syndrome, which is caused by the lack of Chordin in pharyngeal endoderm at a later developmental stage (Bachiller et al. 2003). The mouse Noggin knockout has normal gastrulation and neural plate formation, and a strong skeletal phenotype later on (McMahon et al. 1998). Mouse embryos mutant for HNF3β, in which Hensen's node–the equivalent of Spemann's organizer–does not develop, do not express Chordin or Noggin at gastrula but still form a neural plate (Klingensmith et al. 1999). Because these two BMP antagonists are required for neural plate development, they must also function at earlier stages before HNF3β is expressed. Indeed, chordin;noggin double mutant embryos display a loss of the prosencephalic vesicle, lack of anterior notochord (ventralization of the mesoderm), and randomization of heart left-right asymmetry (Bachiller et al. 2000). Thus the Chordin and Noggin BMP antagonists have redundant functions and are required for the patterning of the three embryonic axes of the mouse.

In Xenopus, knockdown of Chordin expression is achieved using chordin antisense morpholino oligos (Chd-MO) (Oelgeschläager et al. 2003a). The phenotype obtained is very similar to that of the chordino zebrafish mutant (Schulte-Merker et al. 1997), displaying a reduction of the size of the neural plate and eventually CNS tissue, and an expansion of ventral mesoderm. When the Xenopus embryo is experimentally manipulated, strong requirements for Chordin are observed. For example, the dorsalizing effects of lithium chloride (LiCl), a treatment that stabilizes β-Catenin, can be completely blocked by Chd-MO (Oelgeschläager et al. 2003a). When the Spemann organizer transplantation experiment is repeated using dorsal lip explants injected with Chd-MO, the grafts completely lose their inducing activity (Oelgeschläager et al. 2003a). Organizer grafts probably require a full complement of BMP antagonists, and even the loss of a single one, Chordin, has profound effects.

Xnr-3 encodes a Nodal-related protein that lacks mesoderm-inducing activity, presumably because it is mutated in a critical cysteine residue of the cystine knot. Xnr3 is able to induce neural differentiation when overexpressed in animal caps (Hansen et al. 1997) and is able to antagonize BMP signaling through its amino-terminal proregion (Haramoto et al. 2004). Xnr3 homologues have not been found in any other vertebrates, but in Xenopus it is, after Chordin, the gene most strongly induced by the early β-Catenin signal in genome-wide studies (Wessely et al. 2004).

The Wnt Antagonists: Frzb-1, Crescent, sFRP-2, and Dickkopf

Secreted Frizzled–related proteins (sFRPs) constitute a large family of Wnt antagonists that encode secreted forms of the amino-terminal, cysteine-rich domain of the Wnt receptor Frizzled (Kawano & Kypta 2003). They bind Wnt proteins in the extracellular space and prevent them from signaling (Leyns et al. 1997). The Xenopus gastrula expresses high levels of sFRPs; a screen for cDNAs encoding secreted proteins resulted in a surprising 24% of isolates encoding sFRPs (Pera & De Robertis 2000). The Spemann organizer expresses Frzb-1/sFRP3, Crescent and sFRP2 (Figure 3).

Xenopus Dickkopf-1 (Dkk-1) is a secreted inhibitor of Wnt signaling that functions through a different and interesting molecular mechanism. It encodes a cysteine-rich secreted protein expressed in dorsal endomesoderm that defines a new protein family (Glinka et al. 1998). Dkk-1 binds to a Wnt coreceptor called LDL receptor-related protein-5/6 (LRP5/6) (Mao et al. 2001). Wnt binds to both Frizzled and LRP6, forming a ternary receptor complex on the cell surface (Tamai et al. 2000). This triggers phosphorylation of the intracellular domain of LRP5/6 at conserved PPPSP sites, causing the recruitment of the β-Catenin destruction complex protein Axin to the cell membrane and inhibition of β-Catenin degradation (Tamai et al. 2004). Thus LRP5/6 specifically links Wnt signaling to β-Catenin stabilization. Dkk-1 binds not only to LRP5/6 but also to a second transmembrane protein called Kremen. The resulting trimolecular complex of LRP5/6, Dkk, and Kremen is endocytosed, resulting in the depletion of LRP5/6 coreceptor from the plasma membrane (Mao et al. 2002). This provides an elegant molecular explanation for how Dkk selectively inhibits the action of Wnt on the canonical β-Catenin Wnt pathway without affecting other aspects of Wnt signaling. In Xenopus, Dkk1 neutralizing antibodies inhibit head and prechordal plate formation (Glinka et al. 1998, Kazanskaya et al. 2000). In the mouse, Dkk-1 homozygous mutants lack CNS structures anterior to the midbrain (Mukhopadhyay et al. 2001), and heterozygotes show strong cooperation with noggin in head formation (del Barco et al. 2003).

Crossveinless-2, Twisted Gastrulation, Xolloid-Related, and Bambi

Drosophila Crossveinless-2 (CV-2)

is a gene required for the formation of the wing crossveins, structures that require Dpp signaling during development. Drosophila and vertebrate CV-2 contain five CR domains of the types present in the BMP-binding modules of Chordin, as well as a von Willebrand factor-D domain (Figure 4A) (Conley et al. 2000, Coffinier et al. 2002). Mouse and Xenopus CV-2 are expressed in ventral mesoderm and ectoderm and when overexpressed behave as BMP antagonists (Coffinier et al. 2002, Moser et al. 2003, Binnerts et al. 2004; C. Coffinier & E.M. De Robertis, unpublished observations). Whereas Chordin expression is repressed by BMP4, CV-2 expression is upregulated by BMP4.

Open in a separate window

Figure 4

Cysteine-rich (CR) BMP-binding modules in Chordin and Crossveinless-2. (a) Chordin contains four CR modules; Crossveinless-2 (CV-2) has five. Chordin contains four additional repeats (ovals)in the region between CR1 and CR2 designated CHRD domains. These immunoglobin-like beta-barrel domains are also present in Short-gastrulation and some secreted bacterial proteins (Hyvönen 2003). CV-2 contains a von Willebrand factor-D domain (vWF-D) and a trypsin-like cysteine-rich domain (TIL) at its carboxy-terminal end. (b) A molecular complex involving BMP, Chordin, and Tsg regulates the D-V activity gradient of BMP4 in Xenopus. Chordin binds BMP4 through cysteine-rich domains CR1 and CR3. After cleavage of Chordin by the Xolloid-related (Xlr) zinc metalloprotease, the affinity between CR modules and BMP4 is greatly reduced, and BMP4 protein is able to bind to and activate BMP receptor (BMPR) on the cell surface.

The Neural Induction Default Model

Early efforts to identify neural inducers used ectodermal explants exposed to many substances, such as dead organizers, methylene blue, sterols, fatty acids, and even sand particles. All were found to neuralize embryonic ectoderm. Gradually, the search for the Spemann organizer neural inducer became a funeral march for the field (reviewed by Holtfreter & Hamburger 1955). Eventually it was realized that neuralization of axolotl ectoderm could be obtained in the complete absence of any inducer simply by culturing animal caps in an inadequate saline solution (Barth 1941). This effect could be mimicked in other amphibians by partial cell dissociation with citrate, oxalate, or low pH treatments that received the unfortunate name of “sublethal cytolysis” (Holtfreter & Hamburger 1955). Xenopus ectoderm is relatively resistant to neuralization, but neural differentiation can be elicited by cell dissociation and culture for several hours (reviewed by Weinstein & Hemmati-Brivanlou 1999, Munoz-Sanjuan & Brivanlou 2002). This neuralization can be reversed by adding BMP4 to the culture medium, which led to the proposal that during dissociation BMP4 protein is diluted by diffusion into the culture medium (Wilson & Hemmati-Brivanlou 1995). BMP acts within the ectoderm to induce epidermis, and it was proposed that when it diffuses away in dissociated cells, neural differentiation by a default pathway would ensue. However, given the recent realization that activation of MAPK signaling can downregulate the BMP signaling pathway at the level of Smad1 phosphorylation (see below), alternative interpretations for why cell dissociation and abnormal substances have neural-inducing activities in ectodermal cells will have to be explored.

Although the BMP default neural induction model has generally received support in Xenopus (Harland 2000), work in other model systems has highlighted the role of FGF and Wnt signals in neural induction and de-emphasized a role for BMP (reviewed in Wilson & Edlund 2001, Streit & Stern 1999, Stern 2002, Lemaire et al. 2002). In chick, FGF can initiate ectopic expression of neural markers but Chordin and Noggin cannot. However, Chordin can stabilize expression of transiently induced neural markers, expand an already formed neural plate, and induce ectopic primitive streaks (Streit & Stern 1999). Explants of the medial chick epiblast differentiate into neural tissue, but when BMP4 is added develop into epidermis instead. Lateral chick epiblast explants develop into epidermis and do not respond to FGF or BMP antagonists; however, if Wnt signaling levels are lowered by treatment with Wnt antagonists, then both FGFs and BMP antagonists can induce neural differentiation in lateral epiblast cells (Wilson & Edlund 2001). Additional lines of evidence indicate that Wnt signals are involved in the choice between epidermal and neural differentiation. A functional screening of cDNA for genes able to cause differentiation of mouse embryonic stem cells into neurons resulted in the isolation of the sFRP-2 Wnt antagonist (Aubert et al. 2002). In Xenopus animal cap explants, Wnt antagonists are able to transiently induce neural markers (Glinka et al. 1997), and FGFs function as neural inducers (Harland 2000). Taken together, the available evidence suggests that multiple signaling pathways are involved in neural induction in all vertebrates. Rather than emphasizing the differences between organisms, the field now needs ways of integrating these diverse neural-inducing signaling pathways. One such integration of cell-cell signals may occur at the level of Smad1 phosphorylation.

Integrating Signals at the Level of Smads

At any given time, cells are exposed to a multitude of cell-cell signals. Understanding how multiple signaling pathways are integrated is a major challenge in cell biology. The differentiation of the ectoderm provides a good material because at the gastrula stage cells must choose between two different fates, epidermis or neural tissue, for which excellent molecular markers exist. Smads 1/5/8 transduce signals of the BMP serine/threonine kinase receptors through phosphorylation of carboxy-terminal SXS sequences, thus triggering nuclear translocation (Shi & Massagué 2003). BMP antagonists, such as Chordin and Noggin, inhibit this carboxy-terminal phosphorylation and promote neural gene expression by decreasing Smad1 activity (Figure 5). Positively acting neural inducers such as FGF8 (Hardcastle et al. 2000) and IGF (Pera et al. 2001, Richard-Parpaillon 2002) signal through tyrosine kinase (RTK) transmembrane receptors. It has recently been shown that FGF and IGF induce, via mitogen-activated protein kinase (MAPK), phosphorylation in the linker (middle) region of Smad1 at four conserved PXSP sites (Pera et al. 2003). Linker phosphorylation prevents nuclear translocation and has an inhibitory effect on Smad activity (Figure 5). This effect had been initially observed in cultured cells treated with EGF (epidermal growth factor) or HGF (hepatocyte growth factor), but its physiological significance had remained unclear (Kretzschmar et al. 1997, Massagué 2003). Overexpression of Smad1 in Xenopus has little ventralizing (pro-BMP) effect except in the case of mutant proteins that cannot be phosphorylated by MAPK (Pera et al. 2003, Sater et al. 2003). This indicates that MAPK signals are active in the developing embryo and inhibit Smad activity. The emerging picture indicates that neural genes are expressed only at very low levels of Smad1 activity, which requires both low BMP levels and high MAPK signals (Figure 5). In agreement with this view, neural induction by Chordin can be blocked by agents that inhibit FGF or IGF signaling (Pera et al. 2003).

Open in a separate window

Figure 5

Integration of multiple signaling pathways at the level of Smad1 phosphorylation during neural induction. Neural induction requires extremely low levels of Smad1 activity that are reached through the combination of two signaling systems. One is inhibition of binding between BMP4 and its serine/threonine kinase (RS/TK) receptor by anti-BMP molecules such as Chordin and Noggin. The other is inhibitory phosphorylation of Smad1 in the linker region by receptor tyrosine kinase (RTK) signals such as FGF, IGF, HGF, and EGF mediated by activation of MAPK. MH1 and MH2 are evolutionarily conserved globular Mad-homology domains; MH1 contains the DNA-binding domain and MH2 multiple protein interaction sites.

In the mouse, knock-in of Smad1 forms that are insensitive to MAPK phosphorylation in the linker region exhibit phenotypes in gastrointestinal epithelium and germ cells (Aubin et al. 2004). These animals express normal levels of mutant Smad1 (as well as of wild-type Smad5 and Smad8), supporting the view that integrating BMP and MAPK signals at the level of Smad1 is required in vivo.

The finding that FGF signaling can cause inhibition of signaling by BMP Smads via the hard-wired mechanism shown in Figure 5 may help explain other situations in which FGF and BMP signals oppose each other during development. A classical example is the antagonism between FGF4 and BMP2 in limb development (Niswander & Martin 1993). Similarly, opposing effects of FGFs and BMPs have been reported in lung morphogenesis, cranial suture fusion, and tooth development (Weaver et al. 2000, Warren et al. 2003, Thesleff & Mikkola 2002). In future, one aspect of the neural induction default model that should be reinvestigated is whether animal cap dissociation, in addition to lowering BMP levels, causes the activation of other signaling pathways such as MAPK.

Inversion of the D-V Axis in Evolution

As shown in Figure 7, Sog is expressed ventrally in Drosophila embryos (first in the ventral two thirds of the blastoderm, then in neurogenic ventral ectoderm, and finally in ventral midline cells), whereas Chordin is expressed dorsally in the vertebrates. An argument against homologous roles for Chordin and Sog was that one was expressed in mesoderm and the other in neuroectoderm. The realization that Chordin is expressed in the BCNE center, which gives rise to the brain and floor plate (Kuroda et al. 2004), now removes this objection because in both animals the initial expression is found in neuroectoderm. A number of recently identified extracellular regulators of BMP signaling—Xld, CV-2 and Tsg—are expressed in the ventral side of the Xenopus embryo as part of the BMP4 synexpression group. Their Drosophila counterparts—Dpp, Tld, CV-2, and Tsg—are expressed on the dorsal side (Figure 7). As additional genes are discovered in future, the case for a unified mechanism for D-V patterning in evolution will become more persuasive. At later stages, once the ventral nerve cord or the neural plate is formed, additional D-V conservations become apparent between Drosophila and the vertebrates. These include Netrin and Slit in the midline, and three D-V columns of homeobox gene expression in the neurogenic regions (vnd/Nkx2.2, ind/Gsh and msh/Msx) (Arendt & Nubler-Jung 1999, von Ohlen & Doe 2000). Although considerable similarities exist between the zygotic D-V patterning systems, the initial signals that set up these patterns of gene expression appear to be very different: In Drosophila the maternal signal is provided by a Dorsal/NFκB signal and in Xenopus by the early β-Catenin signal (Lall & Patel 2001). Detailed promoter studies, perhaps on the Chordin BCNE enhancer, may help determine whether any common upstream signals exist.

Open in a separate window

Figure 7

Inversion in the course of evolution of a conserved network of extracellular regulators of BMP/Dpp signaling involved in D-V patterning.

Whereas the extracellular Chd/Sog BMP/Dpp system has been remarkably conserved, other vertebrate D-V genes are not conserved in Drosophila. The sFRP Wnt antagonists, Dkk, Cerberus, and Bambi are not present in the Drosophila genome. It is possible that in the course of evolution growth factor antagonists may be more easily generated than, say, new signaling pathways. Within the vertebrates, the ventral gene sizzled and its dorsal homologue crescent are found in zebrafish, Xenopus, and chick but have not been detected in the mouse, rat, or human genomes. An extreme example of innovation within the vertebrates is the case of Xnr3, which is only found in Xenopus.

We thank B. Reversade, H. Lee, A. Ikeda, L. Fuentealba, A. Mays, J. Kim, L. Zakin, and D. Geissert for comments on the manuscript. Our work is supported by the NIH and the HHMI. The authors are Investigator and Associate, respectively, of the Howard Hughes Medical Institute.