Pluripotent cells (stem cells) and their determination and differentiation in early vertebrate embryogenesis


  • †This article is dedicated to the memory of Dr Hildegard Tiedemann.


Mammalian embryonic stem cells can be obtained from the inner cell mass of blastocysts or from primordial germ cells. These stem cells are pluripotent and can develop into all three germ cell layers of the embryo. Somatic mammalian stem cells, derived from adult or fetal tissues, are more restricted in their developmental potency. Amphibian ectodermal and endodermal cells lose their pluripotency at the early gastrula stage. The dorsal mesoderm of the marginal zone is determined before the mid-blastula transition by factors located after cortical rotation in the marginal zone, without induction by the endoderm. Secreted maternal factors (BMP, FGF and activins), maternal receptors and maternal nuclear factors (β-catenin, Smad and Fast proteins), which form multiprotein transcriptional complexes, act together to initiate pattern formation. Following mid-blastula transition in Xenopus laevis (Daudin) embryos, secreted nodal-related (Xnr) factors become important for endoderm and mesoderm differentiation to maintain and enhance mesoderm induction. Endoderm can be induced by high concentrations of activin (vegetalizing factor) or nodal-related factors, especially Xnr5 and Xnr6, which depend on Wnt/β-catenin signaling and on VegT, a vegetal maternal transcription factor. Together, these and other factors regulate the equilibrium between endoderm and mesoderm development. Many genes are activated and/or repressed by more than one signaling pathway and by regulatory loops to refine the tuning of gene expression. The nodal related factors, BMP, activins and Vg1 belong to the TGF-β superfamily. The homeogenetic neural induction by the neural plate probably reinforces neural induction and differentiation. Medical and ethical problems of future stem cell therapy are briefly discussed.


Two topics in developmental biology have recently attracted international public interest: (i) the isolation and culture of human stem cells (Shamblott et al. 1998; Thomson et al. 1998) for potential use in transplantation medicine; and (ii) the cloning of a sheep nearly genetically identical with its parent by incorporating a nucleus from an adult cell into an enucleated oocyte (Wilmut et al. 1997). Obviously, the nucleus from the adult cell was reprogrammed by the cytoplasm of the enucleated fertilized oocyte.

The scientific foundations of these recent developments have a long history. The role of the cell cytoplasm and the nucleus and their cooperation in animal development is a fundamental question in biology. Early cell isolation experiments on sea urchin embryos identified organ-determining substances in the egg cytoplasm (Boveri 1901). The sea urchin Paracentrotus lividus contains a pigmented ring only in the vegetal part of the egg. This pigment is not involved in cell determination, but allows identification of animal and vegetal cells. Boveri found that only those cell fragments containing vegetal cytoplasm could develop to the pluteus stage. The genes are located in the cell nucleus and cytoplasmic factors are necessary to determine which genes are activated and/or repressed.

Pluripotent cells in amphibian embryos and test methods for inducers

The cleaving amphibian egg first forms a morula, followed by the blastula stage, when a cavity called the blastocoele develops. At the gastrula stage, a blastopore is formed in the presumptive endoderm. It is through the blastopore that the endoderm, the mesendoderm on the leading edge of the mesoderm and most of the mesoderm (the so-called marginal zone between the animal ectoderm and the vegetal endoderm; Fig. 1) invaginates. The invaginated dorsal mesoderm induces development of the neural plate in the overlying ectoderm. This forms the primordium of the neural anlage. If small pieces of tissue from the ectoderm and the neural anlage are exchanged in the early gastrula stage, the grafted tissue develops in conformity with its new surroundings. The grafted ectoderm becomes part of the neural plate and the grafted presumptive neural plate becomes ectoderm (Spemann 1918). These experiments indicate that the grafted tissues from the amphibian gastrula are still pluripotent.

Figure 1.

Map of the presumptive regions of the blastula of the urodele, Cynops pyrrhogaster, projected onto the surface of the embryo as seen from the left side. In the animal half higher than the limit of invagination (Eg) are the (E) epidermis, (Gl) neural crest, (N) neural plate, (T) tail somites and (t) lateral mesoderm in the tail. In the marginal zone are the (Ch) notochord, (1–9) numbering of trunk somites, (P) posterior trunk somites, (PrC) prechordal mesoderm and (Spl) lateral plate mesoderm in the trunk. In the vegetal part are the (D) endoderm of archenteron, (J) starting point of invagination, (U) boundary between the mesoderm and endoderm (prospective lateral lip of the blastopore at the beginning). After Nakamura (1942) from Annot. Zool. Jap. 21, 161.

The ectoderm of early gastrulae has been shown to form beside neural tissues somites when the presumptive mesoderm of the dorsal blastoporal lip (the organizer) is transplanted into the ventral ectoderm of a host gastrula (Spemann & Mangold 1924). The neural tube of the induced secondary axis system was derived from the host ectoderm and the chimeric somites were derived from the host ectoderm and the transplanted blastoporal lip. Holtfreter (1933a,b, 1939) then applied a sandwich technique (tissue fragments or pellets that contain inducing factors are enveloped by two pieces of explanted ectoderm from early gastrulae) to show that ectoderm can be induced by a number of other tissues. The multipotency of the ectoderm has also been used in implantation techniques (‘Einsteckmethode’; Mangold 1923) to test inducing material. In these experiments, tissue fragments or pellets containing inducing factors (highly active factors can be mixed in different proportions with a non-inducing protein, such as γ-globulin) are implanted through a slit into the blastocoele of an early gastrula. The gastrulation movements cause the pellet to come into contact with the pluripotent ventral ectoderm. However, this method is less suitable for Xenopus laevis embryos, because a splitting of the axis can occur. Substances in solution can be tested on isolated ectoderm from blastulae or early gastrulae (animal cap test; Becker et al. 1959). In another test, mRNA for the protein to be tested is injected into blastomeres, then the animal cap is excised at the blastula stage and tested for tissue-specific markers. The formation of axes after injection of RNA into blastomeres (of normal or UV-irradiated eggs; Scharf & Gerhart 1980) is a further, more complex test method.

Methods to test induction of tissue development depend on the pluripotency of the ectoderm (animal cap). The ectoderm is, however, not completely ‘undifferentiated’. Ectoderm that is isolated and cultured in salt solution becomes ciliated and develops to atypical epidermis, with an ultrastructural differentiation similar to its development in vivo (Billet & Courtenay 1973; Grunz 1973; Grunz et al. 1975). These structures, however, lack the basal lamina and the regular alignment of epidermal cells, which depends on mesenchyme (Holtfreter 1939).

The fate of ectoderm is reversible up to the early gastrula stage. When labeled single animal pole blastomeres from X. laevis blastulae are transplanted into the blastocoele of late blastula host embryos, the daughter cells of the transplanted blastomeres contribute to all three germ-cell layers, in agreement with the transplantation and induction experiments. At later stages, the fate of the transplanted animal cells becomes restricted solely to ectoderm (Snape et al. 1987). Single vegetal pole blastomeres from X. laevis morulae that are transplanted into the blastocoele of host embryos can contribute progeny to all cell layers. By the early gastrula stage, the cells contribute only to the endoderm. This process is seen in vitro when an appropriate tissue mass is present (Heasman et al. 1984; Wylie et al. 1987). The differential expression of cell surface molecules to change cell affinities is an important factor for these experiments.

The marginal zone (presumptive mesoderm) is determined to mesoderm early in the development process. Isolated marginal zones from X. laevis and Triturus pyrrhogaster have the capacity for self-differentiation from the early blastula stage onwards (Nakamura & Tarasaki 1970; Nakamura et al. 1970; Nakamura & Toivonen 1978), which shows that mesoderm is determined before the mid-blastula transition.

The attempts to characterize induction factors have been confused by inadequate methodology. This led to the still-existing prejudice that tests on Triturus alpestris, T. pyrrhogaster, Pleurodeles waltl and others led to neuralization without inducer, the so-called ‘autoneuralization’. However, autoneuralization was only observed in ectoderm explants from Ambystoma mexicanum early gastrulae in Flickinger solution as the culture medium (Holtfreter 1944; Holtfreter & Hamburger 1955; Tiedemann 1986b).

Mammalian pluripotent cells (embryonic stem cells) from blastocysts and primordial germ cells

In mammalian development, the cells of the inner cell mass of the blastocyst, the mother cells of the embryo proper, are multipotent for a short period. When these cells are transplanted to ectopic sites, for example in nude mice, they form teratocarcinomas. Invasive growth, which is characteristic for carcinomas, was also observed in a low percentage when the ectoderm of early T. alpestris gastrulae was implanted into the visceral cavity between the somatopleura and liver of adult Triturus taeniatus (Waechter 1951). Martin (1981) and Evans and Kaufman (1981) were able to maintain cells from the inner cell mass of mice in tissue culture, where they could be continuously propagated as pluripotent embryonic stem (ES) cells. If injected back into the blastocyst of a host female, the cultured stem cells can contribute to many tissues of the chimeric offspring. Subcutaneous injection of ES cells into syngeneic mice induces teratomas, which contain cells of ectodermal, mesodermal and endodermal origin (Wobus et al. 1984). Pluripotent stem cells from other mammalian species have also been maintained in tissue culture (Shamblott et al. 1998 and references therein). Mouse and other mammalian stem cells are characterized by expression of the germ-line transcription factor Oct 4 (Nichols et al. 1998), high levels of alkaline phosphatase and the expression of certain cell surface glycolipids (Solter & Knowles 1978) and glycoproteins (Andrews et al. 1984). These proteins are characteristic, but not specific, for stem cells.

Stem cells are mostly cultured in medium supplemented with 10–20% fetal bovine serum or horse serum, which contains small amounts of growth and differentiation factors. Leukemia inhibitory factor (LIF; Smith 1991) is added to the culture medium to prevent differentiation, and recombinant basic fibroblast growth factor (bFGF) is added to enhance multiplication of the stem cells. Culture in serum-free defined media is possible, but the replication of cells is retarded compared with cells cultured in media with serum.

On withdrawal of LIF, the cells aggregate to form embryoid bodies, which can be cultured under different conditions (for review see Keller 1995). Within the first few days of differentiation, the embryoid bodies express marker genes of primitive endoderm and mesoderm. After extended periods in culture, tissues of all three germ layers develop.

The induction and development of hematopoietic (see also later), muscle and neural cell lineages has been well studied. Activin and bone morphogenetic protein (BMP-4; see later) induce mesoderm formation and hematopoiesis in embryoid bodies cultured in serum-free defined medium (Johansson & Wiles 1995). A blastocyst-derived ES cell line expresses muscle-specific genes in embryoid bodies in a temporal pattern that precisely reflects the sequence in vivo in the mouse (Rohwedel et al. 1994). An enriched preparation of cardiomyocytes has been isolated by microdissection of the cardiogenic regions of embryoid bodies. In these cells, DNA synthesis is markedly decreased 21 days from induction, with a concomitant increase in multinucleated cells (Klug et al. 1995). To enrich the cardiomyocytes, multipotent ES cells were genetically modified to express both a gene construct carrying an α-cardiac myosin heavy chain (MHC) promoter–aminoglycoside phosphotransferase and a phosphoglycerate kinase–hygromycin resistance transgene. Transfected cardiomyocytes were selected by growth in the presence of hygromycin, then transplanted successfully into the hearts of adult dystrophic mice. Loss of cardiomyocytes in the mammalian heart is irreversible and can lead to diminished cardiac function; thus, cardiomyocyte transplantation has significant therapeutic potential (Klug et al. 1996).

Neuronal precursor cells and functional post-mitotic neurons have been obtained from ES cells in vitro (Okabe et al. 1996). Aggregates of mouse ES cells treated for 4 days with retinoic acid showed a high rate of neuritic outgrowth. When these cells were replated as a monolayer, many cells had a neuronal appearance and expressed neuronal markers. The cells generated action potentials and expressed tetrodotoxin-sensitive sodium channels (Bain et al. 1995). Cells expressing glial precursor markers were obtained from aggregates of mouse ES cells by propagating them sequentially in a medium containing bFGF, epidermal growth factor (EGF) and finally both bFGF and platelet-derived growth factor (PDGF). On growth factor withdrawal, the cells differentiated in either oligodendrocytes or astrocytes (Brüstle et al. 1999).

Transplantation experiments have shown that neural precursor cells generated in vitro can participate in the development of embryonic rat brain (Brüstle et al. 1997). Following transplantation of glial precursor cells into the ventricle of myelin-deficient rat brains, myelin sheets formed in various brain regions. The ethical considerations involved in the therapeutic application of these type of experiments to human beings have also been discussed (Brüstle & Wiestler 2000). A human stem-cell lineage equivalent to the mouse glial cell line may be used to treat diseases such as multiple sclerosis, which primarily affects glial cells. Systematic trials with combinations of differentiation-inducing factors to attain a special lineage pathway of stem cells will in the future be a difficult, but rewarding, task.

Human ES cells have been derived from human blastocysts. The cultured stem cells differentiate to form all three embryonic germ layers when grown to confluence. Differentiation can be induced even after several months of culture as undifferentiated cells (Thomson et al. 1998). In vitro aggregation of human stem cells induces embryoid bodies (as with other mammalian stem cells). In addition to other differentiated tissues derived from all three germ layers, a minority of embryoid bodies show rhythmic pulsing of precursor myocardial cells expressing α-cardiac actin (Itskovitz-Eldor et al. 2000).

Pluripotent stem cells can also be obtained from primordial germ cells (Matsui et al. 1991; Resnick et al. 1992). A mouse embryonic germ cell line has been cultured as embryo-like aggregates. The differentiation efficiency of this line into cardial and muscle cells was lower and the spontaneous neuronal differentiation higher as compared with an ES cell line (Rohwedel et al. 1996). To establish pluripotent stem cell lines from human primordial germ cells, gonadal ridges and mesenteries of 5–9-week-old post-fertilization human embryos (obtained as a result of therapeutic termination of pregnancy) containing primordial germ cells have been cultured on feeder layers after disaggregation (Shamblott et al. 1998). Human recombinant bFGF (see later), human LIF and forskolin were added to the culture medium. Colonies of primordial germ cells were identified by alkaline phosphatase and immunologic markers routinely used to characterize ES cells and embryonic germ cells. As with ES cells, pluripotent embryonic germ cells can aggregate to form embryoid bodies, indicating that cell–cell interaction is essential for cell differentiation. Embryoid bodies are often surrounded by a layer of endoderm and contain a mixture of differentiated cells; however, the restricted growth characteristics of differentiated cells in embryoid bodies limits their use. Human embryoid bodies grown in the presence of LIF, FGF and fetal calf serum have been enzymatically dissociated to derive clonal cell lines. The embryoid body-derived (EBD) cells show robust and long-term proliferation with a normal karyotype and the ability to be cryopreserved. They can be stably transduced with either lentiviruses or retroviruses and preferentially express neural markers. They also express endodermal, vascular, hematopoietic and muscle cell markers (Shamblott et al. 2001).

Parkinson’s disease arises from a loss of dopamine nerve terminals in the striatum. Recently, precursors of dopaminergic neurons in cultured mesencephalic cells obtained from human embryos, aborted 7–8 weeks after conception, were transplanted to Parkinson’s disease patients. Some of the implanted cells survived and differentiated (18F-fluorodopa scanning by positron emission tomography (PET)) in the absence of immunosuppressive therapy. A small improvement was observed during the first year after surgery. Thereafter, severe spontaneous dyskinesias developed as a disabling complication in five patients (15%; Greene et al. 1999; Freed et al. 2001). This trial highlighted the need to know more about the molecules synthesized by the neurons, and the proliferation, migration, differentiation and survival of the precursor cells after transplantation (Fischbach & McKhann 2001).

Experiments to elucidate the molecular mechanism of ES cell propagation in a pluripotent state have identified the importance of the transcription factor STAT3, which is activated by a LIF–glycoprotein (gp) 130 receptor complex. STAT3 has four docking sites on gp130, which are not functionally equivalent. The transcription factor is, however, also involved in the differentiation of somatic cells. Cell type-specific interpretation of STAT3 activation may explain its diverse developmental effects (Niwa et al. 1998).

Stem cells in somatic mammalian tissues and organs

Soon after the blastocyst stage, pluripotent stem cells give rise to primitive germ-line stem cells (see earlier) and somatic progenitor cells, which can only be dealt with briefly in this review. Hematopoietic stem cells are the most thoroughly investigated stem cell system (for review see Duesenbery 1998; Fuchs & Segre 2000; Weissman 2000 and references therein). Hematopoietic stem cells in the blood islands of the yolk sac provide hematopoiesis. From the embryonic loci, hematopoietic stem cells migrate to the fetal liver and from there to the spleen and the bone marrow. Long-term hematopoietic stem cells (Weissman 2000 and references therein) probably have a high turnover. Approximately 8–10% of these cells randomly enter the cell cycle per day in young adult mice. In man and mouse, chemotherapy (acute myelosuppression; Richman et al. 1976) and treatment with the cytokine granulocyte colony stimulating factor, a glycoprotein, initiate mitosis and mobilize stem cells to migrate into the blood stream (Metcalf et al. 1980; Durhsen et al. 1988; Molineux et al. 1990a,b). Blood from mice treated with granulocyte colony stimulating factor can rescue lethally irradiated mice (Molineux et al. 1990b), as seen with bone marrow transplantation.

Cell-sorter based separation of bone marrow cells labeled by monoclonal antibody or stained with fluorescent dyes led to the isolation of hematopoietic stem cells in the mouse (Spangrude et al. 1988; Goodell et al. 1996). The pluripotent stem cells pass through stages of progenitors, which are themselves still stem cells, but are more restricted in their options. Each step of stem-cell differentiation involves further maturation. At the end of this process, common lymphocyte progenitors arise, which give rise to T lymphocytes, B lymphocytes and killer cells, as well as progenitors of the myeloerythroid lineages. The latter give rise to myelomonocytic and megakaryotic/erythroid precursors. Combinatorial interactions of signal molecules and positive and negative transcription factors control the advancing differentiation, as in early embryos (see later). Renewal and differentiation of interleukin-3-dependent multipotent stem cells are modulated by stromal cells and serum factors (Spooncer et al. 1986). The G1-protein-coupled CXC chemokine receptor 4 is activated by stromal cell-derived factor 1 (SDF-1). The expression pattern of chemokine receptor 4 of X. laevis suggests that SDF-1 and the CXCR4 receptor are involved in regulating the migratory behavior of hematopoietic stem cells colonizing the larval or fetal liver (Moepps et al. 2000).

The glycoprotein erythropoietin, which is produced primarily in the adult kidney, plays an important role in the regulation of erythropoiesis (Goldwasser 1967). In embryos lacking a functional erythropoietin receptor in the yolk sac, due to targeted disruption of the receptor gene, primitive erythrocytes are produced in normal numbers, although they are reduced in size and their proliferation is retarded. In contrast, definitive erythropoiesis in the fetal liver is drastically inhibited beyond the late progenitor stage. This suggests that early and late erythropoiesis are differentially regulated. It is possible that another cytokine and receptor play a role in early hematopoiesis (Lin et al. 1996). In pluripotent hematopoietic stem cells, multilineage gene expression has been observed prior to commitment to a distinct lineage (Hu et al. 1997). However, this could to some extent depend on the culture of the cells in serum, which contains growth and differentiation factors in small concentrations.

Tissues other than blood cells in the adult can undergo self renewal from somatic stem cells (reviewed by Fuchs & Segre 2000 and references therein). These stem cells often rapidly proliferate initially and then undergo terminal differentiation. In contrast to ES cells from the inner cell mass, the developmental potency of these stem cells is limited. The epidermis maintains a basal layer of dividing cells, which then withdraw from the cell cycle and migrate outwards to the skin surface. Cultured epidermal cells are used in medicine to enhance wound healing. The characteristic structural elements of the small intestinal mucosa are ciliated villi. Between the villi, crypts protrude to the submucosa. Stem cells are located near or at the base of the crypts. Crypt cells express Tcf4, a member of the Lef/Tcf family of transcription factors in the Wnt/β-catenin pathway (see below). Tcf4 null mice have no stem cells in the intestinal crypts (Korinek et al. 1999). Cultured keratinocyte stem cells of the skin have elevated levels of activated β-catenin. Signaling via β-1 integrins and mitogen-activated protein kinase (MAPK) determine human epidermal stem cell fate (Zhu et al. 1999).

Neural crest stem cells, from which all neural crest cell lines are derived, have been isolated from rats (Morrison et al. 1999). The identification of stem cells from the adult central nervous system has also been reported (Johansson et al. 1999; reviewed by Fuchs & Segre 2000; Weissman 2000 and references therein). Stem cells from adult brain have also been obtained from bodies (Palmer et al. 2001).

Bone undergoes continuous remodeling during growth and adult life. The cells involved are chondrocytes and osteoblasts of mesenchymal origin, whereas the osteoclasts derive from bone marrow monocytes. Several BMP are thought to regulate early differentiation of mesenchymal cells to chondrogenic and osteogenic cell lineages. The expression of BMP-4 (a member of the TGF-β superfamily, see later) is increased at bone fracture. Human BMP-2 and BMP-7 are used, together with biomaterials (for example special ceramics), to adhere newly formed bone in delayed fracture healing, pseudoarthrosis and other bone diseases (reviewed by Wozney 1994; Erlebacher et al. 1995; Küswetter & Teschner 1999 and references therein).

In addition to their potential in transplantation medicine, stem cells could be useful as a gene ferry for gene therapy. Genes transferred into stem cells and expressed in the differentiated cells originating from the stem cells would be amplified together with the stem cells. One major problem with this application is the search for suitable vectors for the stable transduction of large genes.


It has been known for a long time that the different tissues of the newt eye can be transformed, a process called transdifferentiation (Lopashov 1977). For example, retina can be spontaneously regenerated from pigmented epithelium (Stone 1950). Retina from tadpoles can transform pigmented epithelium into retina with all the specific cell elements, but without normal organization, when the Bruch’s membrane (part of the mesenchymal development of the eye) is removed (Lopashov & Sologub 1972). This process is accompanied by shedding of part of the cytoplasm, including pigment granules, as seen with lens regeneration from the iris (Yamada & McDevitt 1974). Lens tissue can also be formed in cultures of cell lines originally derived from retinal pigmented cells (Eguchi & Okada 1973). Recently, the transdifferentiation of bone marrow or blood cells to donor-derived endothelial cells (angiogenesis) and somitic muscle has been reported, as well the implication of bone marrow cells in liver regeneration (Weissman 2000 and references therein).

Neural induction and differentiation

Isolation of neuralizing factors

Factors that induce the development of neural tissue have been partially purified from chicken brain (Mikhailov & Gorgolyuk 1989) and X. laevis gastrula stages (Janeczek et al. 1992; Tiedemann & Tiedemann 1999). Exclusively neural tissues of the forehead type consisting of forebrain (telencepalon and diencepalon), eyes with lenses, retina, tapetum and nose are induced by neuroblastoma cell-line homogenates (Lopashov et al. 1992).

In homogenates from X. laevis gastrula–neurula stages, neural-inducing factors are found in ribonucleoprotein (RNP) particles (~110 Å diameter), which are different from ribosomal subunits, in small vesicles and in high-speed spin supernatant (Janeczek et al. 1984). These fractions and the factors isolated from them induce foreheads, very little hindheads and no mesodermal tissues. The RNP particles are highly inductive. When tested by the ‘Einstecktest’ on T. alpestris early gastrulae, large secondary neural plates develop very early. Large areas of the ventral and the lateral ectoderm are attracted by the induction, so that the gastrulation is disturbed. In 50% of cases, embryos with two heads with eyes (‘Doppelköpfe’) and a rudimentary primary axis develop. Other embryos show large heads on the belly of the host larvae. The active component is a basic protein in a hydrophobic basic protein complex. The molecular weight of the main inducing protein estimated by size-exclusion high- pressure liquid chromatography (SE-HPLC) after reduction with mercaptoethanol, which does not inactivate neuralizing factors, is in the range of 33–50 kDa. A neuralizing factor from the supernatant was purified 800–1000-fold in the presence of protease inhibitors. After sodium dodecylsulfate (SDS)–polyacrylamide gel electrophoresis (PAGE) separation and elution, most of the activity was found from 14 to 30 kDa, with lesser activity shown at 100–110 kDa. This factor is part of an acidic protein complex, although the active protein may be basic. After complete deglycosylation with trifluoromethansulfonic acid, the inducing activity was not irreversibly lost. After ultracentrifugation of the supernatant, no activity < 10 000 kDa was found. The neuralizing factor in the small vesicles resembles that found in the supernatant. The factors could either be different proteins or part of a precursor-product line. The RNP particles and small vesicles from unfertilized oocytes show similar inducing activities to the corresponding fractions from gastrula stages, indicating that the factors in these fractions are maternal factors. The supernatant from unfertilized oocytes, however, has only a very small neural inducing activity.

In the embryo, neural inducing factors are released by the endomesoderm at the leading edge of the invaginating mesoderm and by the invaginated mesoderm (see later) to induce the dorsal ectoderm to form the neural plate. The factor(s) are released from the mesoderm in diffusible form, as shown by transfilter induction experiments (Saxén 1961). Accordingly, a small amount of neuralizing factor can be isolated from the extracellular space between the dorsal mesoderm and the overlying neural plate of very early stage neurulae (John et al. 1983). The neural inducing activity of the mesoderm is in part inhibited by actinomycin D (which inhibits RNA synthesis) and by actidion (which inhibits protein synthesis). This suggests either that neuralizing factor(s) are synthesized de novo in the mesoderm or that mRNA and protein synthesis are needed for the release of maternal factor(s) from the mesoderm (reviewed by Tiedemann et al. 1998 and references therein).

Neural inducing factors release in the dorsal ectoderm the inhibition of neuralization

Disaggregation experiments on X. laevis animal ectoderm (animal caps) have had a strong impact on the understanding of neural induction. After dissociation of the ectoderm and dispersal of the cells for 3 h prior to reaggregation, strong neuralization occurs (Grunz & Tacke 1989). However, the neuralization is blocked when extracellular matrix material, which is released to the medium after dissociation, is added before reaggregation (Grunz & Tacke 1990; Grunz 1997, 2001). This suggests that neural inducing factors have no instructive information for the induced ectoderm, but that neuralization in the ectoderm is inhibited and that the factors from the mesoderm release this inhibition. The release of inhibition depends on direct interaction of neural inducing factors with factors like BMP-4 (which is contained in the extracellular matrix in considerable amounts; W. Knöchel & H. Grunz, unpubl. data, 1995) and Xwnt-8 (Wilson & Hemmati Brivanlou 1995; Glinka et al. 1997, 1998; Piccolo et al. 1999).

In keeping with these results, the neuralizing factors from amphibian gastrula and neurula remain fully active after coupling to bromoacetylcellulose (Tiedemann & Born 1978) or bromocyano (BrCN)-sepharose beads (Born et al. 1986), which prevent their uptake into the cells. In addition, a cell surface receptor for forebrain-inducing factors cannot be detected (M. John & J. Janeczek, unpubl. data, 1982).

When ectoderm is mesodermalized by treatment with activin (see later) and then immediately combined with untreated ectoderm, it preferentially induces hindhead, trunk and tail structures. However, when the ectoderm is cultured 5–25 h before combination with untreated ectoderm, it preferentially induces forehead structures (Ariizumi & Asashima 1995). This corresponds with the results of Okada and Hama (1943), who found a temporal change in the inducing capacity of the dorsal presumptive mesoderm (upper blastoporal lip).

Masked neuralizing factors

A masked, inactive neuralizing factor is present in the ectoderm, which is induced to the neural plate (Holtfreter 1934). This factor is partially activated by freezing and thawing and completely activated by agents that denature or dissociate protein complexes (John et al. 1984). The neuralizing factor(s) themselves are not irreversibly inactivated by these agents. The induced neural plate in turn acquires neural-inducing activity (the so-called homeogenetic inducing activity; Mangold & Spemann 1927; Waechter 1953). It is likely that the homeogenetic induction does not depend on a transfer of inducing factor(s) from the invaginating endomesoderm (reviewed by Tiedemann et al. 1998 and references therein). To answer this question definitively, however, the spatial distribution of inducing factors must be known.

Because the ectoderm contains a masked neural inducing factor, several mechanisms for neural induction are possible. First, the release of the inhibition of neuralization by factors from the mesoderm and mesendoderm could directly induce neuralization and the activation of neural-specific genes. Second, it is possible that by the release of the inhibition the masked neuralizing factor(s) in the ectoderm are activated or a de novo synthesis of the factors is initiated. The factor(s) would then be released from the ectoderm or the neural plate (homeogenetic induction), respectively, and activate neural specific genes in neighboring cells. Finally, and most likely, both mechanisms could cooperate to reinforce the neuralization of the ectoderm. This cooperation is supported by the differentiation and inducing capacity of the different areas of the neural plate.

The differentiation and homeogenetic inducing capacity of the neural plate have been investigated by heteroplastic implantation of different areas of the neural plate of Triturus cristatus early neurulae (neural fold just elevated) into the blastocoele of early T. alpestris gastrulae (Einsteckmethode). The foremost transverse part of the neural fold differentiates frequently to forebrain and midbrain and also induces the development of forebrain and midbrain. The anterior median part of the neural plate develops mostly to hindbrain (rhombencephalon) and rarely to forebrain or spinal cord and induces mostly rhombencephalon development. The posterior medium part of the neural plate differentiates less frequently to rhombencephalon and more frequently to spinal cord and induces mostly spinal cord and muscle, which attends the spinal cord (Waechter 1953; Tiedemann-Waechter 1960). This muscle-inducing activity is inhibited in the embryo.

Mesodermal genes promoting neural induction

Genes that are involved in the differentiation of mesoderm, endoderm and the nervous system can act in sequence, but also in parallel (reviewed by Tiedemann et al. 1996; Grunz 1999b). This unexpected redundancy of differentiation pathways is reminiscent of Spemann’s principle of double insurance (‘doppelte Sicherung’) in biological processes (for instance in formation of the lens; Spemann 1936). However, redundant gene control mechanisms are preferentially involved in the spatial and temporal fine regulation of gene expression. The interpretation of the induction results for neural, mesoderm and endoderm differentiation is impeded by the fact that for many genes differentially expressed in development, only the spatial and temporal distribution of the mRNA, but not of the proteins encoded by the genes, which are the really effective molecules, are known.

The activated Xlim-1 gene has been implicated in the induction of muscle and head neural structures in X. laevis (Taira et al. 1994) and in mice (Shawlot & Behringer 1995). A major transcriptional activator of Xlim-1 in early development is activin or an activin-like factor (see also later). Other genes of interest in this context are chordin and noggin. The chordin gene encodes a 120 kDa secreted molecule, which dorsalizes the mesoderm (Sasai et al. 1994). The gene, which is expressed in the dorsal mesoderm, is a secondary response gene to activin. Chordin expression is activated by microinjected mRNA of the dorsal- mesodermal homeobox genes Xlim-1 and gsc (goosecoid, see later). The homeobox genes code for transcription factors. The homeobox is a highly conserved DNA sequence, which codes for a polypeptide sequence of 60 amino acids, the homeodomain (Gehring 1992). The homeodomain is one of the sequence motifs of transcription factors that bind to DNA.

Chordin, which is expressed in the dorsal mesoderm, inhibits ventral signals by binding directly to BMP-4, probably preventing BMP–receptor interaction (Piccolo et al. 1996). Microinjection of chordin mRNA into X. laevis blastomeres induces neural tissue development, as shown by histologic analysis of the isolated animal caps (Sasai et al. 1995). Chordin protein applied directly to animal caps leads to the expression of the neural marker neural cell adhesion molecule (N-CAM; Piccolo et al. 1996). Xolloid, a secreted metalloprotease, inactivates chordin by proteolytic cleavage. When chordin in inactive chordin–BMP complexes is cleaved, active BMP is released. Accordingly, injection of Xolloid mRNA leads to ventralization at the gastrula stage. In the embryo, endogenous Xolloid could spatially regulate chordin activity (Piccolo et al. 1997). Whether chordin directly induces forebrain depends on the amount of chordin expressed between the mesendoderm, the anterior dorsal mesoderm and the overlying ectoderm.

Noggin, like chordin, is a secreted protein, which dorsalizes the mesoderm (Smith & Harland 1992; Smith et al. 1993). Noggin also interacts directly with BMP-2/4 (Zimmermann et al. 1996). Transcription of noggin is induced by activin in animal caps (ectoderm) in the presence of cycloheximide, an inhibitor of protein synthesis (Sasai et al. 1994). Microinjection of chordin mRNA into blastomeres induces the expression of the neural marker N-CAM in animal cap ectoderm; in some cases, this occurs together with a marker for muscle development. Noggin binds to extracellular heparin and diffusion of noggin may only occur when the binding sites of the extracellular matrix are saturated (Zimmermann et al. 1996). The affinity of noggin for BMP is 15-fold higher than that of chordin; nevertheless, the N-CAM-inducing activity in animal caps requires 10-fold higher amounts of noggin compared with chordin (Piccolo et al. 1996).

A gene called cerberus, which codes for a secreted protein of 270 amino acids, induces forebrain, but suppresses formation of trunk–tail mesoderm. In X. laevis, zygotic expression of the gene starts at the blastula stage in the yolky anterior endomesoderm, including leading edge cells of the endomesoderm invaginating during gastrulation. At the midgastrula stage, the gene is expressed anterior and lateral to the prechordal plate (Bouwmeester et al. 1996). This corresponds with the observation that in Triturus a small area of the foremost part of the neural plate, the transverse neural fold, preferentially develops to forebrain. In mice, the anterior visceral endoderm, which corresponds to the anterior endomesoderm in X. laevis (the area of cerberus expression), is important in head formation (Thomas & Beddington 1996; Rhinn et al. 1998). Cerberus can be induced by X. laevis nodal-related factor 1 (Xnr1; see later), but can also inhibit the expression of Xnr1, potentially serving as a feedback inhibitor. Cerberus protein binds to BMP and Wnt proteins, which promote the differentiation of ventral mesoderm (see later) and inhibit neuralization in the dorsal ectoderm. It has been proposed that in order for the head territory to form, the signaling of all three factors involved in trunk development must be inhibited. A truncated form of cerberus, called cerberus short (Cer-S), interacts with and inhibits only the X. laevis Xnr1 gene (Piccolo et al. 1999). A cerberus-related gene is expressed in mouse embryos, but an in vivo requirement of the cerberus-like protein for anterior pattern formation has not thus far been found (Simpson et al. 1999).

Another protein, called Dickkopf (dkk-1), belongs to a new family of secreted proteins. It antagonizes Xwnt-8, which is expressed in ventrolateral gastrula mesoderm, and is involved in neuralization. Dkk1 induces the neural markers N-CAM and N-tubulin in animal ectoderm. This induction does not, however, lead to histologic differentiation of neural tissues (Glinka et al. 1998). Niehrs (2001) has proposed a two-inhibitor (anti-BMP and anti-Wnt) model for neural induction. This model also explains the microcephaly observed in zebrafish in which the Wnt pathway inhibitor tcf3 is mutated (Kim et al. 2000). The follistatin protein, to which neural inducing activity has been ascribed, has no neural-inducing activity when it is applied in a wide range of concentrations to animal cap ectoderm (Asashima et al. 1991a; Grunz 1996).

The proteins encoded by the genes described earlier induce only neural tissue of the forebrain. Hindbrain (rhombencephalon with metencephalon and myelencephalon) is induced by the median anterior and the adjacent posterior archenteron roof (invaginated dorsal mesoderm) of early Triturus neurulae (Ter Horst 1948). When cells of the presumptive forebrain and the axial mesoderm are disaggregated and mixed in different proportions, an increase of mesodermal cells results in an increase of hindbrain and spinal cord differentiation (Toivonen & Saxén 1968). Combinations of partially purified forebrain and mesoderm (somites, notochord) inducing factors induce hindbrain and spinal cord development, depending on the ratio of the two factors (Tiedemann & Tiedemann 1964).

Basic fibroblast growth factor (see later), together with other factors, has been implicated in the expression of more posterior brain sections (Crox & Hemmati-Brivanlou 1995; Lamb & Harland 1995). The Wnt-3a gene can alter anterior–posterior neural patterning (McGrew et al. 1995). Retinoic acid, which operates via a nuclear receptor, has no neural-inducing activity, but retinoid signaling enhances the development of posterior structures (Durston et al. 1989; Ruiz i Altaba & Jessell 1991a,b; Blumberg et al. 1997). The induction of rhombencephalon and spinal cord obviously depends on as yet unidentified additional factors. It is possible that these factors act on a distinct pathway not involving BMP and Wnt inhibition.

Dorsoventral and anterioposterior pattern formation in the developing neural system

The interesting results on pattern formation in the developing neural system can only be dealt with briefly in this review (reviewed by Lumsden & Krumlauf 1996; Gruss & Walther 1992; Krumlauf 1994). The induced neural plate is patterned through a series of inductive interactions along the dorsoventral and anterior– posterior axes. Ventral structures are induced by the underlying mesoderm and also originate within the neuroectoderm.

The zebrafish cyclops mutation leads to a deletion of the ventral-median floor plate of the spinal cord. The most severe deletions, however, are found in the ventral forebrain, especially in the midbrain, leading to cyclopia due to incomplete splitting of the eye field (Hatta et al. 1994). The cyclops locus encodes the nodal-related protein Ndr2 of zebrafish. The Ndr2 gene is first expressed in a region corresponding to the amphibian organizer (dorsal mesoderm), and is expressed subsequently in the prechordal plate and in the ventral neuroectoderm at the tail-bud stage. The cyclops gene is probably autoregulated, in that cyclops activity appears to be required for the expression of the cyclops gene in the anterior axial mesoderm and the ventral neuroectoderm. Cyclops acts as a strong neuralizing inducer in the X. laevis animal cap assay, probably by antagonizing BMP (Rebagliati et al. 1998). Additional factors, such as sonic hedgehog (Shh), are required in the patterning of the anterior ventral nervous system. Cyclops and Shh seem to act in parallel (Chiang et al. 1996).

The mouse homeobox gene Otx2 is required for the development of the forebrain, the midbrain and the anterior rhombencephalon (up to rhombomere two). The anterior neural plate can form without expressing Otx2, but important regulatory genes, including Pax2, Wnt-1 and En (a homolog of engrailed), are not expressed. Otx2 is the earliest homeobox gene to be expressed in the neuroectoderm (Rhinn et al. 1998). The X. laevis homolog of Otx2 is a maternal gene expressed at low levels. Zygotic expression starts in the blastula stage, in a region that gives rise to prechordal mesendoderm, suggesting a role in the differentiation of the most anterior structures. At gastrula stage 10.5 (Nieuwkoop & Faber 1956), transcription becomes detectable in the presumptive anterior neuroectoderm (Pannese et al. 1995). The Otx5 gene, isolated from activin-treated X. laevis ectodermal animal caps, stimulates the formation of anterior regions and represses the formation of posterior structures (Kuroda et al. 2000).

The mouse En1 and En2 genes, which code for transcription factors and depend on Otx2 expression, are also essential for the development of the midbrain and cerebellum. The En genes show overlapping expression with the Pax2, Pax5 and Pax9 genes in the developing brain. The enhancer element of the En2 gene has been shown to contain multiple positive and negative regulatory elements, including two DNA-binding sites for the Pax-2, Pax-5 and Pax-8 proteins. The Pax genes are most probably direct upstream regulators of the En2 gene (Song et al. 1996). The FGF8 gene is expressed as a ring at the posterior border of the midbrain, just posterior to the expression of the Wnt-1 gene (Crossley et al. 1996). The Wnt-1 protein does not induce midbrain, but is involved in the maintenance of En expression (Lumsden & Krumlauf 1996). In midbrain pattern formation, the size, shape and orientation of cell populations has been shown to depend on the geometry of the sonic hedgehog source (Agarwala et al. 2001).

The expression of the Pax6 homeobox gene in the chicken neural plate is inhibited by activin (Pituello et al. 1995) and FGF (including FGF8) from presomitic mesoderm. It has been proposed that the refinement of dorsoventral expression of Pax is dependent on a gradient of FGF signaling moving posteriorly to extend the limit of Pax6 expression to progress caudally (Bertrand et al. 2000). Pax6 activation depends on somitogenesis in the chicken embryo cervical cord. The maintenance of Pax6 expression also depends on somites (Pituello et al. 1999).

Many distinct neuronal cell types are determined by the patterning of the nervous system. A large number of genes, in part homologs of Drosophila melanogaster genes, are involved in the early as well as terminal differentiation of neurons (reviewed by Placzek & Furley 1996; Tiedemann et al. 1998); some are reviewed here. The X. laevis X-MyT1 gene, a C2HC-type zinc finger protein, is involved in the primary selection of neuronal precursor cells. The expression of the gene is positively regulated by the bHLH protein X-NGNR-1 and negatively regulated by the Notch/Delta signal transduction pathway. Inhibition of X-MyT1 function inhibits normal neurogenesis (Bellefroid et al. 1996). The Pax3 and Pax7 transcription factor genes share redundant functions to restrict ventral neuronal identity in the spinal cord (Mansouri & Gruss 1998). They are regulated ventrally by Shh (sonic hedgehog) and dorsally by BMP-4 and BMP-7 (Goulding et al. 1993; Liem et al. 1995, 1997; Ericson et al. 1996). The expression of two sets of homeotic transcription factors is regulated in the ventral neural tube by a gradient of Shh activity. They define the identity of neuronal progenitor cells. The expression of five class I proteins (Pax7, Irx3, Dbx1, Dbx2 and Pax6) is repressed by Shh; two class II proteins (NKx6.1 and NKx2.2) are induced by Shh. It has been proposed that cross-repressive interactions between class I and II proteins establish individual progenitor domains of neuronal cells (Briscoe et al. 2000). The protein XNLRR-1, an X. laevis homolog of the mouse neuronal leucine-rich repeat protein, is expressed in the developing X. laevis nervous system. Its similarity to other human cell adhesion molecules implicates its involvement in interactions at the neuronal cell surface (Hayata et al. 1998). The expression of the first neuronal genes immediately proceeds to the release of the inhibition of neuralization by neural induction (reviewed by Tiedemann et al. 1998). Some precursors may have a restricted neuronal fate, but most generate neurons and astroglial cells and are at least bipotential (Cochard et al. 1995). Only a limited number of neuronal genes can be activated following dissociation and reassociation of ectoderm cells (Duprat 1996).

Sox10 is the key regulator in the differentiation of peripheral glia cells in mice, which descend from pluripotent neural crest cells. In a Sox10 mutant, Schwann cells or satellite cells are not generated. Neuronal cells in the dorsal root ganglia are formed, but degenerate later. Sox10 controls the expression of ErbB3, which encodes a Neuregulin receptor. Neuregulin is a member of the EGF family. Heparin sulfate at the cell surface interferes with the free diffusion and long range signaling of Neuregulin (Meyer & Birchmeier 1995; Britsch et al. 2001).

Signal transduction in neurogenesis

Neural induction is activated via different signaling pathways, including the L-type Ca2+ channels (reviewed by Duprat 1996). An increase in internal Ca2+ mediates neural induction in P. waltl (which shows no autoneuralization). Stimulation by the L-Ca2+ channel agonist Bay K8644 induces the differentiation of neuronal and glial cell lineages. Neuronal induction of ectoderm in Holtfreter sandwiches with dorsal mesoderm (organizer) is inhibited by a Ca2+ chelator (Moreau et al. 1994).

Phorbolester (phorbolmyristateacetate (PMA, TPA) evokes large complexes of neural tissue as well some mesenchyme and melanophores in isolated T. alpestris ectoderm (which shows no autoneuralization in controls; Davids et al. 1987) and in X. laevis ectoderm (Otte et al. 1988), albeit at a higher concentration. The enzyme protein kinase C (PKC), a serine–threonine kinase, translocates to the plasma membrane in PMA-treated explants, which correlates with its activation (Otte et al. 1988). The activity of PKC is enhanced in X. laevis ectoderm explants treated with neuralizing factor (Davids 1988). PKC can upregulate or downregulate Ca2+ channels in many systems. Ca2+ channels are activated by PKC either directly by phosphorylation or indirectly by downregulation of voltage-regulated Na+ and K+ channels (Alijianian et al. 1991). To test the hypothesis that PMA may induce neural tissue by activating Ca2+ channels, Ca2+ levels have been measured during treatment with PMA. Intracellular Ca2+ was transiently increased (Leclerc et al. 1995). Staurosporin, a PKC inhibitor, prevented the Ca2+ increase (Moreau et al. 1994). These experiments suggest that an activation of L-Ca2+ channels takes part in the induction of neural tissue by PMA. Intracellular Ca2+ also increased when the lectin concanavalin A, which induces neuralization, was added to isolated ectoderm. The signal transducing Go protein was expressed simultaneously and could be colocalized with L-type Ca2+ channels (Pituello et al. 1991; Leclerc et al. 1995). The increase in Ca2+ following the activation of L-type Ca2+ channels during the differentiation of PC12 cells leads to the activation of immediate early response genes, such as c-for or Jun-B. Preliminary experiments have shown that a similar mechanism may be involved in neural induction (reviewed by Duprat 1996). The molecular interactions between L-type Ca2+ channels, inhibitors of neuralization and neural-inducing factors remain to be elucidated.

N-2-hydroxyethylpiperazine-N′-2-ethanesulfonic acid (HEPES), a zwitterionic buffer substance, triggers neuralization at a certain pH range (when the zwitterionic form prevails) in isolated ectoderm of T. alpestris (Tiedemann 1986a). HEPES activates the Na+/H+ antiport system (Altura et al. 1980). However, ethylisopropylameloride, a specific inhibitor of the antiport system, does not inhibit neuralization in a wide range of concentrations (Tiedemann 1986; Hildegard Tiedemann, unpubl. data, 1988). A change of intracellular pH is probably not sufficient to trigger neural induction. A change in the conformation of plasma membrane proteins may be involved in HEPES action.

Determination of mesoderm in the marginal zone of very early embryos without induction by the endoderm

Following fertilization, cytoplasmic streaming (cortical rotation Ancel & Vintemberger 1948; Vincent & Gerhart 1987) establishes a dorsal–ventral polarity, whereby the dorsal side of the embryo is located opposite to the random point of sperm entry. Cortical rotation leads to an upward (i.e. towards the animal pole) movement of the dorsovegetal subcortical cytoplasm relative to the egg surface. It has been shown by confocal microscopy that rotation movements are observed prior to the appearance of a detectable parallel array of microtubules inside the vegetal cortex. This suggests that the array may be oriented by preceding cytoplasmic movements and that the motor molecules may be distributed throughout the parallel array (Larabell et al. 1996). Cortical rotation modifies the animal–vegetal and dorso-anterior location of factors that are involved in the first steps of cell determination and thereby specifies the dorsoventral axis.

It has been shown that the presumptive endoderm induces mesoderm in isolated ectoderm, when presumptive vegetal endoderm (later called the Nieuwkoop center) is combined with ectoderm (Ogi 1967; Nieuwkoop 1969). This experiment has been repeated with the same result by other investigators. Tested on ectoderm, which was still competent for mesoderm induction for only 45 min, endoderm was inductive by the 64-cell stage (Jones & Woodland 1987). Nieuwkoop concluded from these experiments that the mesoderm within the embryo, which differentiates from the marginal zone (the region of the fertilized egg and early embryo between endoderm and ectoderm), is also induced by the vegetal endoderm. More recent experiments have cast doubt on this interpretation (Gallagher et al. 1991; Grunz 1994; Li et al. 1996). The four animal blastomeres, which were manually isolated from regularly cleaved X. laevis 8-cell stages as a quartet, differentiated to dorsal and dorsolateral mesodermal (notochord, somites) and (secondarily induced) neural tissues in over 50% of cases. Ventral mesodermal tissues were not induced (Grunz 1994). The yolk-rich leakage material from the dissected vegetal blastomeres, in which the animal blastomeres came to lie for 20–30 min, induced only mesenchyme and coelomic epithelium from animal cap ectoderm in control experiments. It is possible that dorsal mesodermal determinants are located in the intercellular space of 8-cell stage embryos between the animal and vegetal blastomeres. Because no general mixing of the cytoplasm occurs after cortical rotation, due to a stable gradient of yolk platelets, this would mean that mesodermal determinants are located in, and secreted from, the marginal zone adjacent to the animal blastomeres. These results suggest that the determination and early differentiation of dorsal mesodermal tissues depends on factors that are located in the marginal zone after cortical rotation. The result is in accordance with the fate map of X. laevis embryos (Nakamura & Kishiyama 1971), from which it can be inferred that the dorsal animal blastomeres of the 8-cell stage provide progeny for part of the notochord and a smaller part of the somites (Fig. 2). There is good agreement with the X. laevis fate map of Dale and Slack (1987), except in the greater contribution they attribute to somites in the second tier of the 32-cell map. In accordance with these experiments, B1 and C1 blastomeres (Fig. 2) of the 32-cell embryo can organize secondary axes, when transplanted into host embryos (Gimlich & Gerhart 1984; Gimlich 1986). This does not exclude the presence of factors in the presumptive endoderm prior to mid-blastula transition, which can diffuse into the ectoderm, are thereby diluted and induce mesoderm when the ectoderm is combined with the presumptive endoderm (see later regarding endoderm and mesoderm induction at different concentrations of activin (vegetalizing factor)).

Figure 2.

Fate map of Xenopus laevis at the 32-cell stage. After Nakamura and Toivonen (1978) from Tiedemann et al.Develop. Growth Differ. 38, 234 (1996).

To analyze the effect of the vegetal presumptive endoderm or of animal ectoderm on mesoderm differentiation, the marginal zone alone, the marginal zone together with the vegetal region or the marginal zone together with the animal region were isolated from different stages of A. mexicanum embryos. The stages were determined by measurement of the diameter of animal pole cells (Tiedemann 1993). Ambystoma mexicanum is well suited for these experiments, because of its large eggs and slow development. No differentiation occurred in the isolated marginal zone of very early blastula-stage embryos. Unexpectedly, when the marginal zone was isolated together with the animal ectodermal region from late morulae (the earliest stage in which the operation could be done), dorsal (notochord), dorsolateral (somites) and ventral (renal tubules and mesothel) mesoderm differentiated. Endoderm did not develop. When the marginal zone was isolated together with the vegetal presumptive endodermal region, notochord differentiated in the late early blastula stage and somites developed not before the late mid-blastula stage; that is, the animal region must be in contact with the marginal zone up to this stage. Ventral mesoderm did not develop. Neural tubes were induced by the primarily induced mesoderm in some samples of the animal region–marginal zone explants. Autoneuralization was not observed in the isolated animal region. Blood precursor cells developed in the late blastula stage in the animal region–marginal zone explants (Tiedemann 1993). It has been shown that, in addition to BMP (see later), factors are expressed in the ectoderm of X. laevis gastrulae that stimulate erythropoiesis in the ventral mesoderm (Maéno et al. 1994).

It has been suggested from the above-described results that in the determination of mesoderm in A. mexicanum, factors from the ectoderm could be involved. It is possible, however, that a factor (probably activin) could be present in a too-high relative concentration in the marginal zone when the ectodermal region is dissected from the marginal zone–vegetal (endodermal) region explants. This would explain why, in the marginal zone–vegetal region explants, notochord, but no ventral (renal tubules and mesothel) and (until the late mid-blastula stage) lateral (muscle) mesodermal tissues are induced. In contrast, ventrolateral tissues are induced in marginal zone–animal region explants and in animal–vegetal region combinates. Ventrolateral tissues are induced at a relatively low concentration of activin, together with FGF and BMP (Tiedemann 1993; Hildegard Tiedemann, unpubl. data, 1995).

Mesoderm and endoderm induction and differentiation: Genes and factors

Maternal secreted factors and maternal transcription factors have been found in all three germ layers. In the early gastrula, after the so-called mid-blastula transition of X. laevis (Newport & Kirschner 1982), a more complex pattern of gene expression emerges. After the mid-blastula stage, the transcription of many mRNA is enhanced. A small amount of high molecular mRNA-like RNA, however, is already synthesized during cleavage (Nakakura et al. 1987). Most eukaryotic genes contain several enhancer (and silencer) sequences for transcription factors in their regulatory regions. As well as the DNA-binding domains, factors have protein-binding domains, indicating allosteric interactions (Monod et al. 1963; Wyman 1964) with neighboring proteins.

Activins, vegetalizing factor, activin-like factors and fibroblast growth factors

A protein isolated from chicken embryos has been found to induce mesoderm and endoderm development and is called vegetalizing factor. The factor has been extracted with phenol, which dissolves protein, whereas RNA from the aqueous phase has no inducing activity (Tiedemann & Tiedemann 1956a,b). The protein has a molecular weight of 26–30 kDa and forms a homodimer with subunits of 13 kDa (Geithe et al. 1981; Schwarz et al. 1981). It has been purified ∼106-fold. The final purification was achieved by serial reversed phase (RP)-HPLC on different stationary phases (Plessow et al. 1990). Reduction with formic acid preferentially splits an exposed disulfide bond (Daopin et al. 1992; Schlumegger & Grütter 1992) between the two subunits and is in part reversible. Reduction with mercaptoethanol splits all disulfide bonds and leads to an irreversible loss of inducing activity. The factor induces all dorsal, dorsolateral and ventral mesodermal tissues in T. alpestris and X. laevis embryos in a dose-dependent manner (Grunz 1983; Plessow et al. 1990), including primordial germ cells (Kocher-Becker & Tiedemann 1971) and blood cells. In the presence of high concentrations of the vegetalizing factor, endodermal tissues (intestine, liver, pancreas) have been identified after 3–7 weeks of culture to obtain good histologic differentiation in sandwich explants of T. alpestris (Kocher-Becker & Tiedemann 1971) and in X. laevis animal cap explants (Asashima et al. 1991a). Mesodermal patterning depends on secondary inducing interactions by additional factors (Asahi et al. 1979; Minuth & Grunz 1980). When the vegetalizing factor is implanted at high concentrations into the blastocoele of an early Triturus alpestris gastrula, the already invaginated endoderm and mesoderm spreads over the induced ectoderm (exovagination, which should not be confused with exogastrulation), due to a change of cell affinities (Kocher-Becker et al. 1965). This change is mediated by the early activin response gene gsc (Niehrs et al. 1993). The change of cell affinities has been measured directly in dissociated and activin-induced cells by their patterns of sorting out following reaggregation (Kuroda et al. 1999).

The vegetalizing factor can be purified by chromatography on heparin-sepharose (Born et al. 1987). Heparin-binding growth factors (bFGF and acidic fibroblast growth factor; aFGF) have therefore been tested for their effects on induction. It has been shown that both FGF can induce trunk and tail structures with the ventral mesodermal tissues, including blood islands and heart muscle surrounded by an endothel-lined pericardial cavity, as well as somites (Knöchel et al. 1987; Slack et al. 1987; Grunz et al. 1988; Knöchel et al. 1989a). eFGF with a secretory signal sequence is a more potent inducer of mesoderm than bFGF (Isaacs et al. 1994). Whether FGF directly induce neural tissue in amphibian embryos is a controversial question (Doniach 1995). In some T. alpestris animal cap experiments, large masses of neural tissue without mesoderm have been induced by a very high concentration of recombinant bFGF (Tiedemann et al. 1994). However, this is not a physiologic process. In chicken embryos, the gene ERNI, an early response gene from Hensen’s node (equivalent to the amphibian organizer), has been used to show that FGF8 may ‘sensitize’ the epiblast to the expression of later neural markers before gastrulation. However, FGF signals are not sufficient to generate a whole nervous system (Streit et al. 2000). Activation of the extracellular signal- regulated protein kinase (ERK) in X. laevis blastulae is generated by endogeneous FGF signaling. Immunostaining suggests that FGF protein can diffuse over several cell diameters (Christen & Slack 1999).

The FGF are monomers and have a much higher affinity for heparin than the vegetalizing factor, while members of the transforming growth factor (TGF-β) family are closely related in their structure to the vegetalizing factor. Dawid and colleagues and Tiedemann and colleagues (Knöchel et al. 1987; Rosa et al. 1988; Knöchel et al. 1989b) have shown independently that TGF-β1 (2 μg/mL) induces endothel (in some explants arranged in a capillary-like network), blood precursor cells, mesenchyme and strands of segmented blastema tissue. TGF-β2 (1–3 μg/mL) induces all dorsal and ventral mesodermal tissues. Hematopoietic progenitor cells can be identified by the expression of embryonic α-globin (Oschwald et al. 1993).

In 1989, Asashima and coworkers discovered that activin A, a member of the TGF-β superfamily, has a much higher mesoderm-inducing capacity (80% induction at 1 ng/mL; Asashima et al. 1989; Asashima et al. 1990a). Activin A was originally isolated from follicle fluid as a gonadal hormone that stimulates follicle- stimulating hormone secretion. It is identical to the erythroid differentiation factor (Murata et al. 1988) and induces all mesodermal tissues (Ariizumi et al. 1991) and endoderm (Jones et al. 1993) in a dose-dependent manner. Together, Asashima and colleagues and Tiedemann and colleagues have demonstrated that the vegetalizing factor is closely related to activin (Asashima et al. 1990b). Both factors are inhibited by follistatin (Asashima et al. 1991a). Partial sequencing identified the vegetalizing factor as an activin A homolog (Tiedemann et al. 1992). A factor isolated from X. laevis transformed fibroblasts (XTC factor; Smith et al. 1990) and a factor isolated from the amniotic fluid (Chertov et al. 1990) have also proved to be either activin or activin homologs.

The FGF and activins or a related factor are expressed as maternal proteins (Kimelman et al. 1988; Asashima et al. 1991b; Isaacs et al. 1992). Activin A (βAA), activin B (βBB) and activin AB (βAB) and follistatin in abundance have been isolated from X. laevis eggs and blastulae. The activin proteins are in part present as a complex with follistatin, which inhibits the activins. Activin A and activin B have equivalent inducing activity (Asashima et al. 1991b; Fukui et al. 1994). Activin mRNA is not present in these early stages. Zygotic activin B is transcribed in X. laevis after the mid-blastula transition stage, whereas zygotic activin A is not transcribed before the end of gastrulation (Thomsen et al. 1990; Dohrmann et al. 1993). Activin A is predominantly transcribed in the follicle cells surrounding X. laevis oocytes (Dohrmann et al. 1993; Rebagliati & Dawid 1993). In mice, activin A and activin B mRNA are present in the granulosa cells of the follicles, but they are also present at lower levels in eggs and pre-implantation embryos (Albano et al. 1993).

The temporal distribution of activin mRNA in follicle cells, oocytes and embryos of Oryzias latipes (teleost fish, Japanese medaka) is very similar to the distribution of the activin mRNA in X. laevis. The spatial and temporal distribution of activin protein has been measured by a monoclonal antibody. The cytoplasm and germinal vesicles of large oocytes as well as the follicle cells were positively labeled (Ge et al. 1993; Wittbrodt & Rosa 1994). Early medaka embryos were uniformly stained from the 1-cell stage until at least the blastula stage. Two activin dominant-negative mutants were generated. Injection of RNA from one of these mutants into a medaka blastomere led to a dose-dependent loss of axial mesoderm. This mutant acts at the protein level and blocks the interaction of maternal activin protein with its receptor. The second mutant interacts at the mRNA level. It was found to deplete the activin pool when cotranslated with wild-type activin, although axis formation was not disturbed. This suggests that zygotic activin mRNA is not necessary for mesoderm differentiation (Wittbrodt & Rosa 1994). Activin also induces axial structures in the hypoblast of chicken embryos (Mitrani et al. 1990).

The X. laevis Vg1 gene, which is related to the activin genes, encodes a maternal mRNA that is localized in the vegetal region of oocytes and embryos (Tannahill & Melton 1989). Vg1 belongs, like activins, to the TGF-β superfamily and displays homology in the C-terminal region, on which the inducing activity depends (reviewed by Tiedemann et al. 1995). The protein is abundant in embryos as a precursor, which has no inducing activity. The processed active form cannot be detected. Tannahill and Melton (1989) propose, as a hypothesis, that localized processing of the precursor is the limiting step. The gene constructs Bm-Vg1 or Act-Vg1, which code for processed proteins, show similar activities to the activins (Thomsen & Melton 1993). In zebrafish, an ortholog of X. laevis Vg1 is also present as an unprocessed precursor. The zebrafish precursor is expressed to some degree as a mature protein when overexpressed in X. laevis embryos (Dohrmann et al. 1996). A X. laevis Vg1 RNA-binding protein is expressed in the vegetal region of X. laevis oocytes and early embryos. A homologous RNA- binding protein (KOC) in mammalian embryogenesis is highly expressed in the gut, pancreas and kidney and overexpressed in pancreatic cancer (Mueller-Pillasch et al. 1999). It would be interesting to know whether Vg1 is processed at later stages of development.

Follistatin, proteoglycans, inducing factors and receptors

Fukui et al. (1999) have measured the uptake and localization of labeled activin and follistatin synthesized in follicle cells, as well as vitellogenin synthesized in the maternal liver, in vitellogenic oocytes and their mutual interactions by surface plasmon resonance analysis in vitro. The data are consistent with an independent transport of the three proteins, formation of an activin–follistatin complex and binding of the complex to vitellogenin. The catabolism of yolk platelets throughout embryogenesis (also in blastomeres) could release the activin–follistatin complex (Karasaki 1963). Activin may be more tightly bound to its cell surface receptors than to follistatin (see later). Whether activin is released from follistatin by proteolytic cleavage (as is BMP from the chordin complex, see earlier) is not known (Fukui et al. 1999).

The secreted activins are bound to cell surface receptors. Activin receptors are heteromeric complexes containing at least one molecule from each of two subfamilies (I and II) of receptor serine–threonine kinases (Attisano et al. 1992; Heldin et al. 1997; Massagué 1998). A cloned type II receptor (ActR-lIβ) has been shown to have the highest affinity for activin (dissociation constant KD ~0.1 nM). The activin–follistatin complex has a lower binding affinity (KD∼2.4 nM; Fukui et al. 1999).

An activin receptor of X. laevis has been shown to be a maternal protein (Chang et al. 1997). Animal blastomeres from the 8-cell stage of X. laevis (isolated by treatment with Ca2+- and Mg2+-free solution) can be induced to form mesoderm by treatment with activin for 30–60 min, followed by treatment with a large excess of follistatin to inactivate any residual activin, which is not removed by washing the cells after treatment with activin (Kinoshita et al. 1993). This shows that functional activin receptors are present in X. laevis at the 8-cell stage. Receptors for FGF are expressed very early in development in the animal region and marginal zone (Ding et al. 1992), as is the case for activin receptors.

In embryos, the availability of activins, TFG-β, FGF and follistatin also depends on proteoglycans of the extracellular matrix and the cytoplasm, which can form low-affinity complexes with the factors. The vegetalizing factor can bind to an acidic proteoglycan in the supernatant of embryo homogenates, inhibiting the factor (Born et al. 1969, 1972b). Polyvinylsulfate competitively displaces the acidic proteoglycan, thereby activating the factor (Tiedemann et al. 1969). Proteolytic degradation of the protein core inactivates the proteoglycan (Born et al. 1972a). Cell surface proteoglycans can, in contrast, reduce their diffusion from three to two dimensions through binding the factors (Richter & Eigen 1974) and thereby increase the local concentration of the factors in the vicinity of their cell surface receptors. Betaglykan, a broadly distributed membrane proteoglycan with heparin- and chondroitin-sulfate chains attached to a 100 kDa core protein, binds TGF-β and presents the factor to the kinase subunit of the TGF-β receptor (López-Casillas et al. 1993). In rat follicular granulosa cells, where differentiation is enhanced by activin, heparan sulfate chains have been shown to associate with follistatin (Nakamura et al. 1991). The action of activins is also modulated by follistatin in these cells. Other experiments have shown that proteoglycans are needed for mesoderm formation. Removal of heparin sulfate from proteoglycans in ectoderm explants by heparinase inhibits induction of Xbra (brachyury) by activin and Xbra and cardiac actin by FGF (Itoh & Sokol 1994).

Because of the interaction of activin and follistatin with one another and with proteoglycans, the concentration of active activin is not identical to the total amount of activin in the cells. Activation of gene expression by activin depends on proteins in the signaling pathway (see later) from the plasma membrane activin receptors to nuclear factors, which bind to a DNA sequence in the enhancer region of the target gene, called the activin response element (ARE). A gene construct of the enhancer including one or multiple ARE, the promoter and a reporter gene (for example the luciferase gene) can be used as a probe for active activin. Such a probe derived from the activin-dependent gsc gene promoter has been microinjected into each single blastomere of 32-cell stage X. laevis embryos. Luciferase activity was measured during early gastrulation and therefore represents the activity distribution in early gastrulae. A uniformly high activity was found in all cells of the third and fourth tier (Fig. 2), which represents the vegetal and most of the marginal region. Low activity was found in the dorsal most blastomeres of the second tier and very little activity was found in the other animal blastomeres (Watabe et al. 1995). A graded distribution in earlier stages is not excluded. The presence of ARE in activin-dependent genes supports the regulatory function of activin or activin-like factors in early stages of embryo development.

Three arguments against activin being a mesoderm- and endoderm-inducing factor have been proposed. Truncated dominant-negative receptors disturb mesoderm induction by activins. However, activin receptors are not specific; they form signaling complexes with other factors of the TGF-β family. More recently, a receptor has been designed that selectively blocks the function of activins, but not Vg1 (Dyson & Gurdon 1997). The receptor consists only of the ligand binding extracellular domain and is therefore likely to be secreted. Injection of truncated receptor mRNA in X. laevis embryos at the 4-cell stage represses the activin-dependent pan-mesodermal gene Xbra (Xenopus brachyury) at the late blastula stage to 20% of the wild-type level. The inhibition then declines and reaches the wild-type level between the gastrula stages 10.5 and 11 (Nieuwkoop & Faber 1956). The truncated receptor obviously inhibits mesoderm induction. The decline of the inhibition could depend on proteolytic degradation of the truncated receptor, which is not inserted in the plasma membrane. In addition, the affinity of the secreted receptor for activin could be lowered. However, it is most likely that other factors, especially the nodal-related factors, maintain and enhance mesoderm differentiation in gastrulae and later stages (see later).

Injection of human follistatin mRNA up to 2 ng into fertilized X. laevis oocytes does not block dorsal or ventral mesoderm formation. Larger amounts of follistatin RNA were toxic (Schulte-Merker et al. 1994). Because follistatin protein binds and inactivates activin protein, but does not bind to processed Vg1 derived from a Vg1 construct, Schulte-Merker et al. concluded that activin does not induce mesoderm. Because follistatin binds to BMP-4 (Fainsod et al. 1997) and BMP-7 (osteogenic protein 1; Yamashita et al. 1995), and probably also to BMP-2, it could be concluded with the same justification that BMP-4 is not involved in ventral mesoderm induction. However, this is very unlikely. Xenopus laevis follistatin RNA, which is not toxic and therefore can be injected in higher amounts (up to 8 ng), inhibits activin-dependent genes. Ventral expression of the pan-mesodermal marker Xbra is strongly inhibited and the dorsal expression is inhibited in part. Xwnt-8, which is expressed in the ventral and lateral marginal zone, is more strongly repressed than gsc in the dorsal organizer mesoderm. In addition to activin, BMP are also inhibited (Marchant et al. 1998). Furthermore, it is unlikely that from 2 ng follistatin mRNA, which was injected in fertilized eggs, enough follistatin is translated up to the 8-cell stage, when activin can bind to its receptor (see earlier). It is not known in which stage enough follistatin protein is synthesized to compete with the maternal activin protein. This could be the reason that a relatively large amount of follistatin mRNA is needed to see an effect.

Deletion of the activin A and/or activin B genes in mice by gene targeting does not lead to the expected malformation of mesodermal tissues and early death of embryos (Matzuk et al. 1995). It is likely that the embryos could be rescued by activin from the ovarian follicles of their mothers (see earlier). This cannot be tested, however, because homozygous activin-deficient mice die after birth.

Dorsalization and ventralization of mesoderm

Activins generate only a coarse pattern in presumptive mesoderm and endoderm. Additional factors, which depend in part on activin, dorsalize or ventralize the mesoderm. Yamada (1940) was the first to discover a dorsalizing activity in the marginal zone. The dorsalizing factors chordin and noggin have already been discussed. Xwnt-8 is a member of the large family of Wnt/wg secreted glycoproteins. Xwnt-8 is most abundant in the future ventrolateral mesoderm of X. laevis gastrulae and is probably involved in the specification of ventral and somitic mesoderm (Christian et al. 1991; Hoppler et al. 1996). Xwnt-8 by itself cannot induce mesodermal tissues in isolated ectoderm. However, it can be induced in ventral mid-blastula ectoderm by activin (Christian & Moon 1993). Xwnt-8 is also induced by FGF in animal caps (Christian et al. 1991). An intact FGF-signaling pathway (but not FGF signaling) is needed for activin signaling (Cornell & Kimelman 1994).

Overexpression of Wnt-8 prior to the mid-blastula stage by injection of Wnt-8 mRNA into ventral blastomeres at the 16-cell stage induces a secondary axis with a second gsc domain (McMahon & Moon 1989). This result led to the supposition that Wnts are also involved in the dorsal β-catenin signaling pathway (see later). The embryonic Wnt of the dorsal pathway is not yet known, therefore this pathway is commonly called the Wnt-like β-catenin pathway.

Activation of genes in the marginal zone (the presumptive mesoderm)

The pan-mesodermal Xbra (brachyury) T-box gene can be activated in animal ectoderm by activin (Smith et al. 1991). At the blastula stage (zygotic) eFGF and Xbra can activate expression of each other, suggesting that they form a regulatory loop. In the gastrula, maintenance of Xbra expression requires FGF signaling (Isaacs et al. 1994; Schulte-Merker & Smith 1995). In high concentration, gsc appears to directly repress the transcription of Xbra (Latinkic & Smith 1999). The genes gsc, Mix.1 (Rosa 1989) and Mix.2 (Vize 1996), which are expressed in the mesoderm and part of endoderm, belong to the immediate early activin response genes. Their expression is independent of protein synthesis. Goosecoid shows a strong superinduction by activin and cycloheximide (Tadano et al. 1993). Xlim 1 is also activated by retinoic acid (Taira et al. 1994). X posterior (Xpo) and Xhox 3 are other genes expressed in posterior mesoderm; they are preferentially activated by FGF (reviewed by Asashima 1994). Activin and activin-related factors also stimulate the expression of the immediate response genes XFD-1/1′ and XFD-4/4′ (Dirksen & Jamrich 1992; Knöchel et al. 1992; Ruiz i Altaba & Jessell 1992; Köster et al. 2000), which belong to the fork head multigene family of transcription factors. The DNA-binding domain of the fork head proteins has the shape of a winged helix (reviewed by Kaufmann & Knöchel 1996). The XFD 1/1′ gene is expressed in the presumptive dorsal mesoderm of the blastoporal lip. Transcripts can be detected later in the notochord and in the neural floor plate, probably as a direct response to signals emitted from the notochord or to a common origin of notochord and floor plate cells. Like gsc (Fainsod et al. 1994), the XFD 1/1′ gene is downregulated by BMP (Clement et al. 1995; see also later). The XFD-4/4′ gene is expressed in the dorsolateral mesoderm, but not within the dorsal blastoporal lip. This is probably due to an activin-induced activation of inhibitory factors. XFD-1 overexpression inhibits an XFD-4 promoter–reporter gene (Köster et al. 2000). In contrast to other XFD genes, XFD-2/2′ genes are initially transcribed within the animal hemisphere in the mid-blastula stage and subsequently in the marginal zone (Lef et al. 1994). XSIP1, a two-handed zinc finger protein, shows marked similarities to the mouse Smad (see later) interaction protein 1 (SIP1) and may be a transcriptional repressor of an activin-dependent signaling pathway (Eisaki et al. 2000).

The spatial expression of Xegr-1, a gene encoding a zinc finger protein, is similar to that of the Xbra gene. In contrast to Xbra transcription, however, transcription factors to be phosphorylated and activated by MAPK (like the Ets proteins, which interact with serum response factor (SRF) to form ternary complexes) are required and sufficient for Xegr-1 gene expression (Panitz et al. 1998 and references therein).

Activin response elements, activin response factors, Smad and Fast proteins

The ARE of the XFD-1/1′ gene (Howel & Hill 1997), the gsc gene (Watabe et al. 1995) and the Mix.2 gene (Huang et al. 1995) are located within introns or at the 5 upstream regions. The XFD-1 gene contains an additional activin-activated element (AAE) in the upstream region, which is indirectly addressed by activin probably requiring de novo synthesis of a yet unknown additional factor (Kaufmann et al. 1996; Friedle et al. 1998). Separate sequence elements for the Wnt-like β-catenin signaling pathway (see later) have been found in the gsc enhancer proximal to the ARE (Watabe et al. 1995). Wnt, however, does not directly interact with the gsc enhancer; instead, the homeobox gene twin (twn) mediates the induction of gsc (Laurent et al. 1997). The homeobox gene siamois, which is a major mediator of the Wnt/β-catenin signaling pathway and is closely related to the twin gene, can also activate gsc (Carnac et al. 1996). In embryos hyperventralized by UV irradiation, cortical rotation does not take place (Scharf & Gerhart 1980). In these embryos, a Xtwin reporter gene is hyperinduced in vegetal pole cells. This suggests that cortical rotation distributes determinants of the Wnt/β-catenin pathway to the dorsal side, including the dorsal marginal zone of the embryo (Laurent et al. 1997).

The ARE of the homeobox Xlim-1 gene is located in the first intron of the gene (Rebbert & Dawid 1997). Xlim-1 is also activated by the proteins encoded by the Act-Vg1 construct and by nodal (see later), but not by BMP-4, Wnt-8 or members of the Wnt/β-catenin signaling pathway (see later). The ARE in the 5′ flanking regions of the gsc and Mix.2 genes function as ‘classical’ enhancers and transcription of these genes is stimulated by activin. However, Xlim-1 contains a constitutive promoter in the 5′ flanking region, which is unresponsive to activin. The ARE in intron I acts as a silencer of the gene. Binding of activin to the ARE of Xlim-1 mediates a stimulation of transcription, which is higher than the constitutive transcription of the gene in the absence of the ARE (Rebbert & Dawid 1997).

The lim-5 genes of X. laevis and zebrafish are closely related to the Xlim-1 gene in sequence, but not in expression pattern. The Xlim-5 gene is activated cell autonomously in the entire ectoderm of early gastrulae and its expression is suppressed by activin. During neurulation, the genes are rapidly restricted to the anterior region of the developing neural plate (Toyama et al. 1995).

The ARE of the different activin-responsive genes show low sequence homology. A number of proteins interact, however, in the formation of multiprotein activin response factor complexes and contribute to the specificity of activin signaling. After the activation of the cell surface receptors by binding of TGF-β factors, Smad proteins are translocated to the nucleus, where they participate in DNA binding and gene activation or repression. Based on sequence homology and biological activity, three groups of Smads can be distinguished: (i) receptor specific Smads; (ii) common mediator Smads; and (iii) inhibitory Smads (Massagué 1998; Whitman 1998). All members contain two highly conserved domains in the N- and C-terminus, separated by a proline-rich region of variable sequence and length. Binding of activins to type II receptors and, in turn, activation of type I receptors results in the phosphorylation of Smad2. The C-terminal region of Smad1 can probably act as an activator or inhibitor, depending on the region of expression in the embryo (Mueller et al. 1999). In later stages of zebrafish development, Smads determine the regionally restricted effects of TGF-β signaling.

Additional factors take part in the formation of transcription complexes. A multiprotein activin response factor (ARF) binds to the ARE of the endodermal- mesodermal Mix.2 gene. This ARF has been shown to be composed of the novel transcription factor Fast-1 and Smad 2 or Smad3 and Smad4. Fast-1 is a nuclear protein containing an N-terminal fork head binding domain and a C–terminal Smad interaction domain, which is responsible for activin-regulated association with the Smads. Fast-1 recognition of the ARE is essential for ARF binding. Smad binding enhances the binding and regulatory activity of the ARF (Chen et al. 1996; Yeo et al. 1999).

Ventralizing genes in early X. laevis embryos, the bone morphogenetic factors and the Xvent genes

Knöchel and colleagues have shown that the BMP-2 and BMP-4 genes are transcribed in X. laevis oocytes and embryos (Köster et al. 1991; Plessow et al. 1991). The BMP belong to the TGF-β superfamily and are related to activin. The BMP proteins are, besides their role in the inhibition of neuralization, the main ventralizing factors. The BMP-2 gene is upregulated during oogenesis. Maternal transcripts are abundant in cleavage stages, but decline rapidly during gastrulation. Zygotic transcription of BMP-2 starts in early neurulae. The transcripts are localized to neural crest cells, olfactory placodes, pineal gland and the heart anlage. Maternal BMP-2 transcripts show some enrichment within the animal region. Zygotic BMP-4 transcripts increase in abundance after the late blastula stage (Köster et al. 1991; Yamagishi et al. 1995) and can override dorsalizing signals (Dale et al. 1992; Jones et al. 1992; Fainsod et al. 1994; Graff et al. 1994; Clement et al. 1995). A BMP-4 protein gradient declining from ventral to dorsal may contribute to the proper formation of distinct mesodermal tissues (Dosch et al. 1997).

The Xvent-1 and Xvent-2 subfamilies of the Xvent genes are involved in ventral BMP signaling and can mediate the regulatory effects of BMP-2/4. Xvent-1 and Xvent-1B are expressed in the ventral marginal zone of early X. laevis gastrulae (Gawantka et al. 1995, 1998; Rastegar et al. 1999). Xvent-2 and Xvent-2B are expressed in the ventral marginal zone, as well as in the lateral parts of the mesoderm, with the exception of the most dorsal cells (Onichtchouk et al. 1996; Rastegar et al. 1999). Xvent-2B is directly activated by BMP-2/4 in the absence of de novo protein synthesis. Xvent-1B does not directly respond to BMP-2/4, but is activated by Xvent-2B. Cycloheximide treatment to prevent protein synthesis has shown that Xvent-2B by itself is not sufficient to activate transcription of the Xvent-1B gene; additional factors synthesized after the mid-blastula transition are required (Rastegar et al. 1999).

XFD-1′ contains an AAE that is both necessary and sufficient for transcriptional activation of reporter genes in animal cap explants. This AAE is also active within vegetal explants in the absence of added inducers (activin is present in the vegetal region, Watabe et al. 1995; see earlier). An additional inhibitory response element, acting as a silencer, prevents the transcription of the XFD-1′ gene in the ventral-vegetal region of the embryo in vivo. This element is located upstream of the AAE, responds to BMP-2- and BMP-4-triggered signals and overrides the activation by activin. Accordingly, the temporal activation and spatial restriction of XFD-1′ to the dorsal mesoderm is regulated by the antagonistic action of activin or activin-related factors and BMP (Kaufmann et al. 1996). It has been further shown that Xvent-1 mediates the suppression of XFD-1′ by using its BMP inhibitory element. The repressor domain of the Xvent-1 protein is localized in the N-terminal region (Friedle et al. 1998).

The transcription complex of the Xvent-2B gene, which is induced by BMP-4, contains a distal binding region for Smad1 with two GCAT motifs and proximal AGnC binding sites for Smad4. The latter is conserved in other TGF-β response elements. Ectopic expression of Smad1 and Smad5 mimics BMP-4 induction of ventral mesoderm. Smad2 mimics induction of dorsal mesoderm by activin and nodal-related factors (see later). Phosphorylation of the receptor-specific Smads causes hetero-oligomerization with Smad4 and translocation to the nucleus (Henningfeld et al. 2000 and references therein). Other transcription factors confer additional DNA-binding specificity for activation by BMP-4. The zinc finger protein OAZ binds directly to Smad1 to activate the Xvent-2 enhancer (Hata et al. 2000). The dorsal gsc gene can downregulate BMP-4 in a regulatory loop after the mid-blastula stage and vice versa, depending on the concentration of the gene products (Fainsod et al. 1994).

BMP-2/4 shift the response to activin from dorsal to ventral mesoderm (Jones et al. 1992; Clement et al. 1995), whereas high concentrations of activin inhibit the formation of ventral mesoderm. Activin at low concentration can, in contrast, enhance Xvent-2 (Xom) transcription (Ladher et al. 1996), which mediates the ventralization by BMP (see earlier). This implies that the type of tissue induced depends on the ratio of activin and BMP and the timing of expression of the genes. It can in part be explained in this way, that the type of mesodermal tissues induced in animal caps depends on the concentration of activin, because the animal cap contains BMP protein.

The organization of the BMP-4 gene enhancers has also been studied. Enhancer–promoter studies of the BMP-4 regulatory regions with luciferase reporter genes have revealed enhancers within the upstream region and within the second intron, which interact with a proximally located indispensable promoter. The BMP-4 gene, as other genes, is regulated by an autoregulatory loop. The autoactivatory enhancer elements are activated by BMP-4 and BMP-2. BMP-2 could therefore induce zygotic BMP-4 in embryos (Metz et al. 1998).

Endoderm: Genes and differentiation

The initial direct determination of mesoderm in the marginal zone does not negate the exchange of proteins between endoderm and mesoderm in later stages. The homeobox gene milk (mix-like) is closely related to Mix.1 and Mix.2 (see earlier; Rosa 1989; Vize 1996). Like Mix.1/2, milk can be activated by activin in animal caps and is expressed in the marginal zone (presumptive mesoderm) and the vegetal endoderm of early gastrulae. During gastrulation, its expression is restricted to the endoderm. Overexpression of milk in the marginal zone represses the expression of mesodermal genes and increases the expression of endodermin, an endodermal gene (see later). In the embryo, milk could prevent the extension of mesoderm into the endoderm (Ecochard et al. 1998). The mesoderm dorsalizers chordin and noggin (Smith & Harland 1992; Sasai et al. 1994) can also induce part of the endoderm. Chordin is expressed not only in dorsal mesoderm, but also in endoderm bottle cells adjacent to dorsal mesoderm. It induces endodermin in the embryo and in animal caps, starting at the dorsal blastoporal lip, and is expressed almost exclusively at the tail-bud stage in the endoderm (Sasai et al. 1996). That the vegetalizing factor (activin homolog) induces endoderm at high concentrations has been mentioned previously (Kocher-Becker & Tiedemann 1971).

Mixer (Mix-like endodermal regulator) is another homeobox-containing transcription factor involved in endoderm development. The gene is expressed exclusively in endoderm and can be turned on in animal ectoderm by activin-related factors. Mixer is required for the expression of the activin-inducible but not the FGF-inducible endodermal genes Xsox 17α and Xsox 17β (Hudson et al. 1997). In the embryo, all three genes can be expressed independently of one another and mixer can maintain the expression of the sox 17 genes (Henry & Melton 1998). GATA5, a zinc finger transcription factor, is expressed in the vegetal region of X. laevis embryos from the early gastrula stage on. GATA5 and GATA4 are potent inducers of endodermal marker genes in animal caps. GATA5 can be induced in the animal cap by high concentrations of activin. In the embryo, the GATA5 gene is likely to be preferentially induced by nodal-related factors (see later; Weber et al. 2000).

Explants of presumptive endoderm isolated from mid- to late blastulae and cultured until the tail-bud stage express endodermal markers differentially. Expression of the more anterior pancreas marker XlHbox-8 relies on the B-Vg1 construct, activin and FGF, while expression of the more posterior intestinal fatty acid binding protein (IFABP) does not. This suggests that a prepattern exists in the developing endoderm. Cortical rotation is required for the expression of more anterior markers, but not for those that are more posterior (Henry et al. 1996). The developmental biology of the pancreas has been reviewed by Slack (1995). Key events of pancreas formation can be induced in gut endoderm by ectopic expression of pancreatic regulatory genes. This group includes the transcription factors Pdx-1 and ngn13 and the genes Hlxb9, Pax4, NKx2.2 and NKx6.1, which are expressed in the insulin-producing β-cells (Grapin-Botton et al. 2001). Insulin is already expressed in mouse pancreas by the 20 somite stage (Gittes & Rutter 1992). The differentiation of endodermal tissues takes place at the later stages and depends on adjacent mesoderm (Mangold 1949; Okada 1954).

The Wnt/β-catenin signaling pathway

Wnt proteins are widely distributed and play a role in many signaling pathways in addition to embryogenesis. Inappropriate activation of Wnt signaling has been implicated in human cancers (reviewed by Peifer & Polakis 2000). The so-called Wnt-1/wg class of Wnt proteins can induce the formation of a secondary axis when injected in ventral X. laevis blastomeres. On formation of a complex with a receptor of the frizzled family, β-catenin, the key signal transducer, is stabilized and translocated into the cell nucleus (for review, see Gradl et al. 1999a and references therein). In the cytoplasm, the phosphoprotein dishevelled is activated and inhibits glycogen synthase-kinase-3β (gsk-3β). Additional proteins, which either inhibit or support gsk-3β activity, bind to a multiprotein complex. The suppression of gsk-3β is involved in dorsal signaling (He et al. 1995; Yost et al. 1996). It prevents phosphorylation of the N-terminus of β-catenin, stabilizing it and allowing it to escape the ubiquitin–proteasome degradation pathway. In X. laevis, targets identified thus far are the homeobox transcription factors siamois and twin and the secreted protein nodal-related 3 (Xnr3), which is involved in neural differentiation and fibronectin synthesis (reviewed by Gradl et al. 1999a,b and references therein). Xtwin and siamois are transiently expressed in the dorsal marginal zone and the vegetal region from the mid-blastula to the gastrula stage (Laurent et al. 1997).

The zebrafish bozozok gene locus encodes the dharma (dha) homeodomain protein. The activation of boz/dha depends on β-catenin. A zygotic recessive bozozok mutant shows poor development of notochord, prechordal mesoderm and forebrain (Driever et al. 1997; Fekany et al. 1999). Some genes in the presumptive dorsal mesoderm are not expressed. This shows that boz/dha is essential for only a part of dorsal mesoderm differentiation. The homeodomain gene vega1 inhibits the expression of boz/dha and causes ventralization. Vega1 is activated at the mid-blastula transition in all blastomeres, but is downregulated dorsally before gastrulation. Ubiquitous expression of vega1 is maintained in mutants in which boz/dha is disrupted, indicating a mutually antagonistic relationship between the two genes (Kawahara et al. 2000b). The N-terminal domain of boz/dha shares sequence similarity with a domain encoded by the X. laevis gsc gene, which acts as a transcriptional repressor (see earlier). In the embryo, vega1 may be eliminated by boz/dha in a limited region of the presumptive dorsal mesoderm. The expression of vega2 is initiated in a restricted ventral domain at the early gastrula stage. Together with vega1, it may negatively regulate organizer genes, including gsc, and participate in the fine regulation of the organizer region (Kawahara et al. 2000a). The zebrafish vox and vent genes (Melby et al. 2000) share substantial homology with the vega1 and vega2 genes, respectively.

The developmental importance of β-catenin was first demonstrated by its ability to induce an embryonic axis (Funayama et al. 1995). β-Catenin is associated with the cytoplasmic domains of the plasma membrane cadherin adhesion molecules. Overexpression of cadherins as well as underexpression of β-catenin (by an antisense oligodeoxynucleotide complementary to β-catenin maternal mRNA) restrict the availability of β-catenin for the signaling pathway and inhibit dorsal mesoderm induction (Heasman et al. 1994). Depletion of β-catenin in animal caps of X. laevis embryos does not inhibit mesoderm induction by activin (Heasman et al. 1994). This shows that the activin- and Wnt-like β-catenin-pathways are separate (but not completely independent, see later) signaling pathways. β-Catenin, like Xwnt-8, is not a mesoderm inducer in animal caps. Disruption of the β-catenin gene in mice prevents mesoderm formation (Haegel et al. 1995).

β-Catenin is maternally expressed at the RNA and protein levels. Zygotic transcription of β-catenin starts after mid-blastula transition (Wylie et al. 1996). In X. laevis embryos, β-catenin displays greater cytoplasmic accumulation along the entire future dorsal side by the 2-cell stage. By the 16- to 32-cell stage, β-catenin accumulates in dorsal, but not ventral, nuclei. β-Catenin is more stable at the dorsal side (Larabell et al. 1997). Experiments to show whether β-catenin is required before the mid-blastula transition have been inconclusive (Heasman et al. 1994; Wylie et al. 1996). However, the injection into C4 (ventral) blastomeres of a gsc-luciferase reporter gene containing a response element for targets of the Wnt-8 β-catenin pathway, together with a Xwnt-8 cytoskeletal actin promoter construct, which directs the expression of Xwnt-8 only after the mid- blastula stage, does not activate the reporter gene. This suggests that the maternal β-catenin is active before the mid-blastula stage (Watabe et al. 1995). In the cell nucleus, β-catenin binds to high mobility group (HMG) box transcription factors of the TCF/LEF family, which form a complex with DNA that displays an altered DNA bend (Grosschedl et al. 1994; Behrens et al. 1996). Chimeric factors constructed by fusion of β-catenin domains and the DNA binding domain of LEF 1 have shown that the C-terminal transactivation domain of β-catenin is necessary and sufficient for signaling by the LEF-1/β-catenin complex (Vleminckx et al. 1999). Based on these and other experiments, it has been proposed that the function of β-catenin in the nuclei is to recruit the transcription machinery to promoter regions of the target genes (Hsu et al. 1998; Vleminckx et al. 1999).

Sixteen different Xwnt molecules have so far been identified in X. laevis (reviewed by Gradl et al. 1999a); however, a Wnt-like factor in the dorsal β-catenin pathway (or another factor which causes the transfer of β-catenin from the cadherins to the cell nucleus) has yet to be identified. Loss-of-function experiments have not shown that Xwnt-8 or Xwnt-8a (in contrast to Xwnt-8 present in animal blastomeres) are required for endogeneous axis specification in the intact X. laevis embryo (for review see Sokol 1996; Gradl et al. 1999a).

Fritz (frzb-1) encodes a protein related to the extracellular part of the frizzled Wnt transmembrane receptor and is an antagonist of Wnt signaling. Fritz can bind and inhibit Wnt-8 and is expressed in all three germ layers in early gastrulation (Leyns et al. 1997; Mayr et al. 1997; Wang et al. 1997). In addition to the frizzled-related proteins and the Dickkopf (Dkk) protein family, another Wnt-inhibitory factor, WIF-1, has been detected in fish, amphibia and mammals. WIF-1 binds to Xwnt-8 in vitro. The gene is expressed in X. laevis embryos at the neurula stage. It is proposed to work together with the other BMP inhibitors to fine-tune the spatial and temporal pattern of Wnt activity (Hsieh et al. 1999).

Nodal and X. laevis nodal-related proteins and the maternal transcription factor VegT

Nodal was first detected in mouse embryos. The nodal gene encodes a secreted TGF-β-type protein, which is similar to the activin and BMP subgroups of the TGF-β superfamily. Unique to the nodal protein is a pair of cysteine residues separated by two other amino acids (CXXC). Transcripts of the gene can be detected in the primitive streak and the visceral endoderm of very early streak-stage embryos (Varlet et al. 1997). At gastrulation, they are localized in the anterior lip of the primitive streak surrounding the node (which is equivalent to the amphibian organizer). Mice homozygous for a mutation of the nodal gene lack the primitive streak and most of the mesoderm (Zhou et al. 1993; Coulon et al. 1994). Nodal signaling is mediated by activin receptors and transcription factors of the Smad family. Member of the ECF-CFC family (homologs to zebrafish oep and mouse crypto) code for secreted factors that regulate nodal signaling. It is possible that nodal acts only in the presence of these proteins (reviewed by Schier & Shen 2000 and references therein).

Several nodal-related genes (Xnr1–6) are expressed in X. laevis embryos. Maternal Xnr or nodal transcripts cannot be detected. Xnr1 and Xnr2 transcripts appear first in vegetal cells of the late blastula. These genes may be expressed in a dorsal to ventral gradient in endodermal cells (Agius et al. 2000). In the early gastrula, transcripts are found in the marginal zone, expression being most intense in the organizer region (presumptive dorsal mesoderm). Xnr2 is also expressed at low levels in pre-endodermal cells below the dorsal lip (Jones et al. 1995). Expression then disappears. Fugacin, a Xnr gene with the strongest sequence identity (53%) to mouse nodal, is expressed transiently in the dorsal marginal zone (Ecochard et al. 1998). Xnr4 is expressed preferentially in the endoderm and in the dorsal mesoderm (Joseph & Melton 1997; Clements et al. 1999). Xnr5 and Xnr6 are induced after mid-blastula transition in the vegetal region of the embryo, but not in the mesoderm (Takahashi et al. 2000). Xnr1 expression reappears during the tail-bud stages, initially near the posterior end of the notochord and then in a large asymmetric domain at the left lateral plate (Lustig et al. 1996a). The role of nodal signaling in left–right axis formation has been reviewed elsewhere (Schier & Shen 2000). Nodal-related gene transcriptional initiation sites used at the later stages of development are different from those used at gastrulation (Hyde & Old 2000).

Injection of Xnr1 and Xnr2 mRNA into X. laevis embryos results in substantial hyperdorsalization and increased expression of gsc and muscle actin. In animal caps, they induce notochord and muscle (Jones et al. 1995). Xnr1 and Xnr2 can also induce endoderm (Yasuo & Lemaire 1999). The main nodal-related endodermal genes are Xnr5 and Xnr6. Besides endoderm, with the merkers endodermin and Xsox 17b they induce notochord and muscle (Takahashi et al. 2000). The inducing activity of Xnr5 is similar to the inducing activity of high doses of the vegetalizing factor (activin homolog; Asashima et al. 1991c).

The regulation of Xnr genes is complex. Induction is a primary response, because de novo protein synthesis is not required. Xnr1 and Xnr2 can be induced in animal ectoderm by activin (Jones et al. 1995), but Xnr5 and Xnr6 cannot (Takahashi et al. 2000). Xnr genes are not induced by bFGF. Xnr1 and Xnr2 are also induced by an activin-like factor (processed Act-Vg1 construct) and by Xnr5 and Xnr6. The response is potentiated in β-catenin-injected explants, although β-catenin alone cannot activate Xnr1. Overexpression of VegT (a maternal transcription factor, see later) induces weak expression of Xnr1 in animal caps, which is potentiated by β-catenin (Agius et al. 2000). However, it is unlikely that VegT alone can activate the expression of Xnr1 in the vegetal region, because a reporter gene containing VegT response elements is not coexpressed (Hyde & Old 2000). Xnr5 and Xnr6 can also not be induced by β-catenin alone, but by a combination of β-catenin and VegT acting cooperatively (Takahashi et al. 2000).

The search for enhancer response elements in Xnr genes is ongoing. Besides functional response elements for VegT (Kofron et al. 1999; Hyde & Old 2000), a Fast site has been found in Xnr1 intron I, which is completely conserved between mouse nodal and Xnr1 (Hyde & Old 2000). Mutational analysis has detected two Wnt response elements in Xnr3 (McKendry et al. 1997), which itself has no VegT response elements (McKendry et al. 1997; Hyde & Old 2000). These Wnt response elements show striking homologies with elements in Xnr1 (Hyde & Old 2000). In addition, Xnr1 can regulate its own expression by a positive regulatory loop; it is probable that VegT can act together with β-catenin to induce Xnr1 expression (Hyde & Old 2000), Xnr1 can induce expression of VegT in ectodermal cells (Lustig et al. 1996b). In support of this, Derrière, another secreted protein of the TGF-β superfamily, can also be regulated via VegT by a positive regulatory loop (Sun et al. 1999).

In the embryo, Xnr1 and Xnr2 mRNA are expressed in the marginal zone of early gastrulae, when the Apod protein (expressed by VegT) is enhanced in this region (Stennard et al. 1999). The TGF-β signals for Xnr1 and Xnr2 can be zygotic activin B (Dohrmann et al. 1993) and/or Xnr5 and Xnr6 (Takahashi et al. 2000). To answer this question, the spatial distribution and concentration of the proteins must be known. Wnt/β-catenin signaling cooperates with the other factors and a positive regulatory loop can enhance Xnr1 and Xnr2 expression.

When the store of maternal VegT mRNA is depleted to 5–10% of the normal level by injection of an antisense deoxyoligonucleotide into the vegetal pole of oocytes, the fate of the different regions of the blastula is completely altered. The fate of the animal region is changed from epidermis and nervous system to epidermis only; the fate of equatorial cells is changed from mesoderm to ectoderm and ectoderm derivatives; and the vegetal cell fate is changed from endoderm to mesoderm and ectoderm. The marginal zone does not invaginate (Zhang et al. 1998). After injection of a very high dose of the antisense oligodeoxynucleotide, the embryos do not form a blastopore (Kofron et al. 1999). The dose used could have affected the specificity of the response. The low residual levels of mesodermal markers could be expressed by activin, the concentration of which is reduced, together with BMP-4, which is not reduced.

Whether VegT acts cell-autonomously or by generating signaling molecules has been investigated through cell dissociation experiments in which cell to cell signaling is prevented (Clements et al. 1999). The data suggest that endodermal markers are cell-autonomously induced from the mid-blastula transition on and that this induction must be rapidly reinforced by signaling for expression to be maintained. In contrast, mesodermal markers are entirely dependent on signaling.

Interesting results have been obtained by investigating the role of the VegT Antipodean (Apod) gene at the level of the proteins, which were identified using affinity purified antibodies (Stennard et al. 1999). Western immunoblotting revealed two proteins of different molecular weights, which could arise by alternative splicing. The larger one, referred to as VegT (62 kDa), is expressed maternally. The protein is present from the egg to gastrulation in the presumptive endoderm. The smaller one (60 kDa), referred to as Apod, is expressed only after the onset of zygotic transcription, initially in the presumptive endoderm and equatorial region, and only in the presumptive mesodermal region from the early gastrula stage (stage 10.5) on. Apod is inducible by activin (Stennard et al. 1999).The T-box of the T-box genes encodes a DNA-binding domain of ∼180 amino acids, which interacts as a dimer with the major and the minor grooves of the DNA (Mueller & Herrmann 1997).

The transcription factor XANF-1 is expressed throughout the animal region and in the upper blastopore lip. Microinjection of XNAF-1 mRNA into animal blastomeres and the marginal zone leads to a massive migration of cells to the interior of the embryo (Zaraisky et al. 1995). The protein encoded by the Xrel gene, a member of the rel family of transcription factors, has been detected in early cleavage stages and is distributed in a gradient with the most intense staining in the cytoplasm of the animal region to almost nothing in the vegetal region. Staining of the cell nuclei in the animal cap and the marginal zone becomes apparent at the early blastula stage, shortly before the onset of zygotic transcription (Bearer 1994). Loss-of-function experiments have so far not been carried out.

Cooperation of signaling pathways

The control of the activity of quite a number of genes during embryogenesis depends on several distinct signaling pathways and nuclear factors. An example of the synergy of signaling pathways is the regulation of the siamois gene. Siamois can be induced in animal ectoderm by the Wnt/β-catenin pathway, but not by the Smad1–BMP or Smad2–activin pathways. The Smad2–activin pathway, however, can cooperate with the β-catenin pathway to induce expression of siamois more strongly than the β-catenin pathway alone. The significance of this cooperation in normal embryos has been shown by blocking the activin pathway with a truncated dominant-negative activin receptor. The expression of siamois was three-fold reduced (Crease et al. 1998). In another example, the Wnt/β-catenin pathway can enhance the induction of the dorsal mesodermal gsc (see also Watabe et al. 1995) and chordin genes (Crease et al. 1998). Experiments showing that the inhibition of Wnt-8 or BMP-4/2 signaling has a similar action on the differentiation of mesoderm suggest coregulation by these factors (Hoppler & Moon 1998).

Xtwin is a target gene of β-catenin signaling mediated through the activity of the Lef1/Tcf proteins. Experiments with reporter gene constructs have shown that Smad4 binding in proximity to the Lef1/Tcf binding sites is needed for full activation. As described earlier, Smad4 is a mediator of BMP-2/4 as well as activin signaling. Lef1 immunocoprecipitates with Smad4 in vitro. A model is proposed in which Smad4 forms a complex with Lef1/Tcf and β-catenin to bind to the Xtwin promoter region (Nishita et al. 2000).

In conclusion, mesoderm determination and differentiation could take place as a two-step process in the following way. The dorsal mesoderm is determined independently of the endoderm by factors that are located in the marginal zone following cortical rotation. The factors responsible are probably BMP-2, activins (vegetalizing factor), FGF, eFGF and β-catenin, which are all maternal proteins (Fig. 3). Receptors for BMP, activin and FGF are expressed maternally (or during cleavage); β-catenin is located in cell nuclei along the whole dorsal side at the cleavage stages. Mesoderm determination could take place in a region where these factors overlap. A coarse pattern of receptor–effector complexes could be formed. The mesodermal genes are in part immediate early genes, which can be transcribed without de novo protein synthesis. The ARF (see earlier) can form as early as the 8–16-cell stage and may be a component of the ‘molecular memory’, by which pre- and mid-blastula signals are maintained until the onset of zygotic transcription (Huang et al. 1995).

Figure 3.

Simplified scheme of the distribution of differentiation factors in amphhibian embryos. (A) Maternal factors in very early cleavage to mid-blastula embryos. (B) Factors in the early gastrula stage. V, ventral; D, dorsal; AP, animal pole; VP, vegetal pole; BMP, bone morphogenetic protein; FGF, fibroblast growth factor; Xnr, Xenopus nodal-related protein.

It is possible that mesodermal genes may be expressed at a very low level prior to the mid-blastula stage, because RNA (Nakakura et al. 1987) and protein (Tiedemann & Tiedemann 1954) are synthesized in amphibian embryos at a very low level before the mid-blastula transition. To show the expression of specific genes in cleavage and morula stages (with only a low number of cells) would, however, require extremely sensitive methods.

After mid-blastula transition, the nodal-related (Xnr) genes become very important for mesoderm and endoderm differentiation and can, at least in part, replace the activins and maintain and enhance mesoderm induction. Xnr1 and Xnr2 respond to activin, the vegetal Xnr5 and Xnr6 factors, Veg T and β-catenin signaling (Fig. 3). Proteins expressed by vegetal Xnr genes could diffuse into the mesodermal marginal zone. Chordin, noggin, BMP-4, Xvents, FGF and Xwnt-8 are all involved in the development of the dorsoventral and posterior mesodermal pattern.

Determination of the vegetal endoderm before the early gastrula stage is reversible, as is the case for ectoderm (Wylie et al. 1987; see earlier). A ‘predetermination’ could occur via the presence of activin (vegetalizing factor) in high concentrations, activin-related factors, Wnt/β-catenin and the maternal VegT protein, which is still present in the presumptive endoderm after the onset of zygotic transcription until gastrulation (Stennard et al. 1999). This implicates VegT in the formation of the endoderm after mid-blastula transition, especially in the regulation of Xnr5 and Xnr6 expression (Takahashi et al. 2000). VegT appears to be important for the equilibrium between endoderm and mesoderm. It prevents the extension of mesodermal gene expression beyond the presumptive endodermal region of the embryo, but is involved, together with activin, Xnr5 and Xnr6 and β-catenin, in the expression of Xnr mesodermal genes. The ratio of the factors in the different regions of the embryo will certainly be important. It is also likely that many factors involved in the formation of the endomesodermal pattern have still to be identified.

Preferential differentiation of single organs

Heart muscle, with its typical honeycomb-like appearance surrounded by an endothel-lined pericardial cavity, can, besides skeletal muscle and ventral mesodermal tissues, be induced by recombinant bFGF at high concentration in X. laevis ectoderm explants (Knöchel et al. 1989a). Beating hearts can also be induced in 10–30% of explants by treatment of newt animal caps with a high concentration (100 ng/mL for 1 h) of activin, followed by sandwich culture with untreated ectoderm for 2 weeks. Heartbeats are regular and display the temperature-dependent frequency of the normal newt heart (Ariizumi et al. 1996; Asashima et al. 2000). The formation of heart tissue in embryos depends on endoderm. Heart is not formed in T. alpestris when the endoderm is removed at the neural plate stage (Mangold 1957). The deep dorsal endoderm contributes to the specification of the heart anlage, whereas the superficial pharyngeal ectoderm may enhance heart morphogenesis during later stages (Sater & Jacobson 1989). The induction of endoderm besides mesoderm may be also a prerequisite for heart induction. Grunz (1999a) observed a high percentage of heart-like structures and hearts when the upper blastoporal lip from early gastrulae of X. laevis, including some head endomesoderm and ectoderm, was treated with suramin (150 mM for 4 h) and cultured in vitro until normal larvae had reached the late neurula or tail-bud stages. These were then transplanted into the posterior trunk area of host embryos, the heart anlage of which was exstirpated at the late neurula stage. The host embryos could be rescued by the transplant. Suramin is a polyanion that prevents transcription of dorsal marker genes in X. laevis embryos (Oschwald et al. 1993).

The Wnt antagonists Dkk1 and crescent, a frizzle related protein, can induce heart formation in the ventral posterior mesoderm of X. laevis explants. Inhibition of BMP signaling does not prevent cardiogenesis. These and other results suggest that cardiogenesis is induced in a region of high BMP and low Wnt-8a and Wnt-3a activity (Maroon et al. 2001; Schneider & Mercola 2001).

Asashima and colleagues have induced a high frequency of pronephric tubules by treating X. laevis blastula ectoderm with 10 ng/mL recombinant activin and 0.1 mM retinoic acid for 3 h. Retinoic acid alone had no inducing activity. When the induced explants were transplanted into late neurula hosts, which were bilaterally pronephrectomized, many of the hosts survived up to 27 days. The pronephrectomized control embryos developed edema and died after 5–9 days (Chan et al. 1999).

The pancreas develops from the anterior endoderm (reviewed by Slack 1995). Contact of notochord with the dorsal pancreatic bud is required for pancreatic development. The notochord represses endodermal Shh in the dorsal pancreatic bud (Kim et al. 1997; Hebrok et al. 1998). Upper blastoporal lips of X. laevis treated with 0.1 mM retinoic acid for 3 h have been shown to differentiate into pancreas-like structures together with notochord and endodermal epithelium. The pancreas-specific markers XIHbox 8 and insulin were induced (Moriya et al. 2000a). Furthermore, isolated presumptive ectoderm from X. laevis blastulae was treated with activin and retinoic acid to induce differentiation into pancreas (Moriya et al. 2000b). The differentiation of hematopoietic precursor cells, induced by a low dose of activin (0.5 ng/mL) in X. laevis animal cap explants, to leukocyte and erythrocyte lineages was promoted by the addition of murine stem cell factor and interleukin-11 (Miyanaga et al. 1998).

Concluding remarks

The isolation of pluripotent stem cells and novel insights into embryonic differentiation provide new possibilities, especially to the field of transplantation medicine. A thorough risk assessment must be carried out for the use of stem cell therapy. It must be known in stem cell therapy studies with the same certainty as in drug studies, which risks are involved and that no harm will occur to the patients. Using long-term experiments, the possibility that teratocarcinomas or teratomas will develop should be excluded. It has not been difficult to foresee that ethical problems would arise (Tiedemann 1968). The use of human stem cells derived from human embryos, which were only grown for this purpose, has come under serious ethical scrutiny. An alternative for some tissues would be the use of somatic stem cells. Another possibility could be the use of cell lines of robust and long-term proliferation (EBD cells), which can be obtained after therapeutic termination of pregnancy. Histocompatibility could be tested, as with other transplants, by tissue typing, which can partially exclude an immunological rejection reaction. The most controversial question is the so-called therapeutic cloning, which uses similar methods as the cloning of almost identical animals. This is not far from human cloning, which is reminiscent of the nightmare of Aldous Huxley’s Brave New World (a warning to and criticism of the epoch) and which has to be absolutely refused for ethical reasons and also because of the malformations which occur. Therapeutic cloning would solve the problem of histocompatibility, but until now the frequency of mammalian nuclear transplants giving rise to blastocysts has been very low (∼1/200). If this problem is solved (first of all in animal experiments) and if public consent to the application of this method can be achieved, therapeutic cloning under very strict legal conditions could be considered as another alternative.

Our own responsibility rises with the new possibilities to interfere in human development. As Medawar (1984) expresses our obligation: ‘We must cultivate our garden’.


Our own investigations, reported in the present review, were supported by the Ministry of Education, Science, Sports and Culture of Japan, by the Deutsche Forschungsgemeinschaft and by the Fonds der Chemischen Industrie.