Brachyury - a gene affecting mouse gastrulation and early organogenesis

At present the mouse is pre-eminent amongst vertebrate experimental organisms as a source of developmental mutants. For a start, there is a relatively large repertoire of spontaneous or physically induced developmental mutants (Lyon and Searle, 1989), although the catalogue of homozygous embryonic lethals is somewhat biased towards genes that are semi-dominant, and have relatively obvious heterozygous phenotypes in viable offspring. This list of developmental mutants is now rapidly being augmented by mutations generated by transgenesis, and in particular by mutations created in embryonic stem (ES) cells (Evans and Kaufman, 1981; Martin, 1981). ES cells provide an opportunity not only for generating random mutations by transgene insertion (some of which may affect development) but also for disrupting previously identified genes, whose genomic sequence is known and which have been shown to be expressed during particular stages of embryogenesis (see Reith and Bernstein, 1991). As a result of this new technology mouse developmental biology is entering a new era of extensive genetic analysis. However, in all cases an intact mutant embryo, while identifying genes necessary for normal development, often provides only gross information regarding the developmental consequences of gene malfunction: the primary perturbation may be obscured by subsequent defective tissue interactions producing a complex phenotype (see Beddington et al., 1991).

In this paper we will describe experiments aimed at further resolving the developmental effects of a well-known mouse mutant, Brachyury (T). The first part will be devoted to a brief review of previous work describing the morphological, genetic and molecular basis of the Brachyury¹ phenotype. This will provide both a context for our own work and also illustrate some of the strengths and weaknesses of descriptive studies, most of which were performed before it was possible to recognise homozygous mutants prior to the inception of abnormalities. The data that we present comprise an analysis of the behaviour of T/T ES cells in chimeric embryos throughout gestation, and demonstrate that the dynamic behaviour of mutant cells intermixed with wild-type cells can reveal subtle alterations in morphogenetic movements.

Brachyury was first recognised almost 70 years ago by Dobrovolskaïa-Zavadskaïa (Dobrovolskaïa-Zavadskaïa, 1927) because heterozygous animals have short, and often slightly kinked, tails. Subsequently, it was found that homozygous embryos die at mid-gestation, about 10.5 day post coitum (dpc), and have distinctive caudal abnormalities. The allantois, which should form a major component of the chorioallantoic placenta, fails to extend and traverse the exocoelom (Gluecksohn-Schoenheimer, 1944). Consequently, the embryo is denied a placental connection and is deprived of adequate nutritive supply. This is probably the physiological cause of embryonic death. However, embryonic pattern posterior to the forelimb region is also disturbed. Somites posterior to the seventh pair of somites are absent or abnormal, the neural folds fuse but the neural tube is severely kinked in the caudal region and the surface ectoderm tends to form large fluid filled blisters. Central features of T/T embryos are the apparent absence of a notochord and profound thickening of the primitive streak (Chesley, 1935; Gluecksohn-Schoenheimer, 1938; Gruneberg, 1958). In addition, the node at the extreme anterior of the primitive streak, the normal origin of the notochord, is less distinct than in wild-type embryos (Fujimoto and Yanagisawa, 1983). Careful descriptive studies over the last 65 years, together with a wealth of evidence implicating the notochord in the patterning of the neural tube and possibly the somites (e.g. Clarke et al., 1991; Hemmati-Bravanlou et al., 1990; Kitchin, 1949; Placzek et al., 1990; Smith and Schoenwolf, 1989; van Straaten et al., 1985; Yamada et al., 1991), have pointed to the defective notochord being a prime cause of many of the embryonic abnormalities. There is some debate as to whether the notochord fails to form altogether or in fact does delaminate, but subsequently degenerates or fuses with the adjacent gut or ventral neural tube, thereby becoming unrecognisable in histological sections (Chesley, 1935; Gruneberg, 1958; Spiegelman, 1976; Yanagisawa, 1990). Immunocytochemical staining of 10 dpc T/T embryos using an antibody raised against cellular retinol binding protein (Maden et al., 1990), which is present in the notochord, suggest that a notochord may be present in caudal regions (P. Rashbass, V. Wilson, M. Maden and R. Beddington, in preparation). Certainly, the absence of notochord is an improbable explanation for the failure of allantois to differentiate normally and the allantoic defects are more likely to result from abnormal deployment of cells emerging from the primitive streak, or inappropriate differentiation of cells once incorporated into the allantoic bud.

The observation that the mesoderrmectoderm ratio is elevated in the caudal 15% of 8 dpc putative T/T embryos but reduced compared to wild-type embryos in the region immediately anterior to the primitive streak (Yanagisawa et al., 1981) would support the notion that morphogenetic movements are abnormal during gastrulation. The mesoderrmectoderm ratio is normal in the anterior half of the embryo (Yanagisawa et al., 1981) and there is no significant difference in either mitotic index (Yanagisawa et al., 1981) or the incidence of [³H]thymidine labelling (Yanagisawa and Fujimoto, 1977b) in different axial regions. Furthermore, no increase in cell death was noted in the posterior region underlying the primitive streak. This argues that it is migration of mesoderm away from the streak which is compromised in the latter stages of gastrulation. Direct measurements of active mesoderm migration on extracellular matrix in vitro reveal that T/T cells from 8–9 dpc embryos have a slightly but significantly reduced migration rate (e.g. 8 dpc, 39.4 ± 11 μm h⁻¹) compared to wild-type mesoderm (8 dpc, 52.8±22.6 μm h⁻¹) (Hashimoto et al., 1987).

In the heterozygote the tail is short and often kinked. Again the notochord in the caudal region (usually confined to the tail but sometimes extending as far forwards as the cloaca) is abnormal during embryonic development. It may be branched, improperly separated from the hindgut or neural tube and often has a prominent central lumen. From studies of both homozygotes and heterozygotes Grüneberg concludes that “a common denominator for all the abnormalities of the notochord may be a change in surface properties” (Gruneberg, 1958).

That the surface of mutant cells may be altered has been tested directly by comparing the ability of wild-type or T/T mutant cells to form aggregates (Yanagisawa and Fujimoto, 1977a). Cells from the trunk, head and forelimb bud were disaggregated and the diameter of aggregates formed in suspension culture measured. Mutant cells, from any one of these regions, consistently formed smaller aggregates and this implies a difference in cell surface adhesive properties. However, the true relevance of these data is not clear since cells were isolated from both affected and unaffected tissues. Glycosyltransferase activity has also been shown to be reduced in T/T mutants (Shur, 1982) and abnomalities of the extracellular matrix have been described (Jacobs-Cohen et al., 1983).

Several experiments have shown that T/T cells are capable of differentiating into a wide variety of mature differentiated tissues. Initially, Ephrussi explanted mutant embryonic tissues in vitro and showed that these could survive beyond the time of embryonic death and could differentiate into an array of cell types comparable to those formed by wildtype embryos (Ephrussi, 1935). Subsequently, ectopic transfer of posterior regions recovered from 8.5 – 9.5 dpc T/T embryos showed that a diverse mixture of mature differentiated tissues, representative of derivatives of all three germ layers, could develop in the resulting experimental teratomas (Bennett et al., 1977). There was no predisposition for T/T embryos to give rise to teratocarcinomas containing undifferentiated embryonal carcinoma (EC) cells indicating that the epiblast, the progenitor of EC cells (Diwan and Stevens, 1976), matured at an equivalent rate to wild-type embryos. When anterior and posterior regions were compared a decrease in the frequency, but not the complete absence, of bone and cartilage was detected in tumours formed from posterior regions (Fujimoto and Yanagisawa, 1979). Taken together, these results suggest that the embryonic defects seen in T/T conceptuses stem from an organizational failing rather than an inability to differentiate into specific tissues. However, it should be noted that neither notochord nor allantoic differentiation can be recognised by histological inspection of experimental teratomas.

The original Brachyury mutation (7) has been shown to be a large deletion (160–200 kb). Two other mutants have been described which exhibit an identical phenotype. One, T^2J, is also a large deletion (81–110 kb) and the other T^kt1 (Justice and Bode, 1990) was induced by ethylnitrosourea and likely to be a point mutation or small deletion. No other gene has been identified in the Brachyury deletion and if T^kt1 is indeed a point mutation, then this argues that the observed phenotype stems from the absence of a single gene.

There are three further T alleles that have similar phenotypes to each other but which differ from that of the original Brachyury mutant. In T^Wis (Shedlovsky et al., 1988) there is an insertion of a retroviral-like element in the seventh exon (Herrmann et al., 1990) and in T^c (Searle, 1966) there is a 19 bp deletion in the last exon (Herrmann et al., 1990). T^{C −2H} has a frameshift mutation affecting the same region of the carboxy terminus as the T^c deletion (B. G. Herrmann, personal communication). What is interesting about these mutations is that they present a more severe phenotype than deletion of the T gene. In both heterozygotes and homozygotes abnormalities occur at a more rostral axial level (Herrmann, 1991; Searle, 1966). Thus heterozygotes usually have no tail and homozygotes show no sign of somites. The boundary of embryonic defects in the homozygotes is shifted rostrally to the cervical region.

Evidence that T^c is an antimorph comes from dosage studies. T alleles that delete the T gene can be complemented by the T locus duplication t^wLub2 (MacMurray and Shin, 1988; Winking and Silver, 1974). However, 7^e, while not being independent of wild-type gene copy number, is only partially complemented by this duplication, which indicates that the T’ product serves to antagonise wild-type activity (MacMurray and Shin, 1988). In addition it has been shown that T^c/T embryos have a less severe phenotype than T /T (Searle, 1966). However, since T is a deletion this amelioration of the T^c effect cannot be due to residual T activity. It is more likely that the T⁰ gene product acts like a dominant negative mutation. If the T gene product is only active in association with a second gene product (Herrmann, 1991, 1992; Lyon and Meredith, 1964), perhaps as a dimer, the T^c protein may interfere with this association or affect the biological activity of resulting protein complexes.

Whatever the mode of action of such an antimorph there emerges a compelling gradient of phenotype where the more severe alleles (T^c and T^wis) affect more rostral levels in both homozygotes and heterozygotes than the weaker deletion mutants (T and T^2J). A similar phenomenon has been observed when comparing the severity of tail defects in T/+ and T/t embryos (Yanagisawa, 1990). On the face of it, this indicates that there is an increased requirement for T activity as one moves caudally along the axis, the cranial region being independent of T activity but tail formation requiring high levels. As the anteroposterior axis is laid down sequentially in the mouse this can be viewed as an increased requirement for T activity with time rather than with distance. In other words, late gastrulation and tailbud differentiation have higher demands for T activity than does early gastrulation. Alternatively, the level of T activity may be constant but the consequences of defective gastrulation accumulate with time. Thus, a more severe effect on gastrulation will culminate in abnormalities earlier, and therefore more rostrally, than less severe disruption. The chimeric analysis presented in this paper (see below) is consistent with a progressive accumulation of defective cells caudally rather than an increased demand for T activity.

The sequence of the cloned mouse T gene does not immediately reveal the nature of the protein product (Herrmann et al., 1990). The sequence has an open reading frame of 436 amino acids and shows limited homology to MyoDl (Willison, 1990). Antibodies raised against the Brachyury homologue in zebrafish demonstrate that the protein is localised to the nucleus (see Herrmann, 1992). Therefore, at present, all available data are consistent with the T protein being a transcription factor.

The expression pattern of the T gene in mouse embryos is largely consistent with the observed pattern of abnormalities seen in mutants (Herrmann, 1991; Wilkinson et al., 1990). Furthermore, the expression of T, or its homologue, in zebrafish (see Herrmann, 1992), Xenopus, and mouse are directly comparable with respect both to embryonic stage and position. In the mouse it is first expressed in the primitive streak at the onset of gastrulation (Fig. 1A), and the Xenopus homologue of T has been shown to be induced, in the absence of protein synthesis, by the mesoderm inducing activity of peptide growth factors (Smith et al., 1991). Expression continues in the primitive streak throughout gastrulation and can be detected in ectoderm adjacent to the streak and nascent mesoderm underlying the streak (Herrmann, 1991; Fig. 1B). However, expression in the mesoderm disappears as the cells move away from the streak and assume their lateral, paraxial or extraembryonic positions. Only the head process and notochord continue to express high levels of T. Interestingly, the allantois, with the possible exception of a very early basal component (A. McMahon and J. McMahon, unpublished data) does not express T. Thus, the expression pattern of the gene supports the notion that its primary sites of action are the primitive steak and the notochord, but does little to explain the defects in allantois development.

In T^wis/ T^wis embryos, the mutant Tgene is expressed normally during the early stages of gastrulation (Herrmann, 1991). However, expression declines rapidly at about 8 dpc, expression being lost first in the head process and notochord precursor and anterior part of the primitive streak and finally at the posterior end of the streak. This loss of expression, particularly in the primitive streak, cannot be accounted for solely by cell death but suggests instead that the normal pattern of T gene expression is dependent, either directly or indirectly, on normal T protein activity (Herrmann, 1992).

All the studies on Brachyury to date suggest a strong correlation between expression of the T gene, whose product is probably a transcription factor, and the normal differentiation or survival of axial mesoderm: the head process and notochord. Expression in the primitive streak may affect cell survival but it also has an influence on the ability of nascent mesoderm to move away from the streak, although this may only be true for the latter stages of gastrulation since the cranial region forms normally. The defects in the allantois may be a consequence of this abnormal migration or they may indicate that for normal allantoic differentiation T expression is required as cells pass through the streak.

Mixing wild-type and mutant cells together in a chimera allows the cell autonomous function of a gene to be assessed. Furthermore, it may allow the analysis of mutant cell behaviour in embryos surviving beyond the stage at which intact mutant embryos die. In mouse, chimerism in all tissues can only be achieved by the addition of cells to the preimplantation embryo. However, there are no morphological criteria for identifying T/T preimplantation embryos. Cloning of the T gene makes genetic characterisation theoretically possible, but would involve laborious polymerase chain reaction assays on biopsies of individual embryos (Handyside et al., 1990). The alternative, of isolating and genetically characterising T/T ES cells is appealing for several reasons. First of all, once estabished, such lines provide a continuous source of mutant cells whose development can be monitored either in vivo or in vitro. Secondly, the availability of ES cells null for the T gene presents an ideal substrate for genetic manipulation of T expression or of genes acting downstream of it.

We have isolated and genetically characterised several T/T, T/+ and +/+ ES cell lines from blastocysts derived from heterozygous BTBR T/+ matings (Rashbass et al., 1991).

These cells are all homozygous for the glucose phosphate isomerase-la gene (Gpi-l^a). These lines have similar morphological characteristics and growth rates in vitro. The only minor difference observed is that T/T embryoid bodies take longer to disaggregate in trypsin than do heterozygous or wild-type lines. Following inoculation under the testis capsule all lines, regardless of genotype, form teratocarci-nomas. Inspection of the differentiated tissues present in these tumours does not show any consistent, qualitative differences, although one T/T cell line (BTBR6) failed to give rise to bone and cartilage and a second null line (BTBR 10) gave a lower incidence of these tissues (Table 1). This is reminiscent of the results from ectopic transfer of embryonic fractions (Bennett et al., 1977; Fujimoto and Yanagisawa, 1979).

When introduced into wild-type embryos, which are Gpi- l^b/Gpi-l^b, a significant difference was observed between T/T and T/+ lines (Beddington et al., 1991; Rashbass et al., 1991). Heterozygous lines formed normal chimeric conceptuses at 8.5 –9.5 dpc. On the other hand, two independent T/T cell lines formed predominantly abnormal chimeras at midgestation (Rashbass et al., 1991 ; V. Wilson and R. Beddington, unpublished observations). Typically, these T/T ↔ +/+ chimeras have a phenotype which mimics that of the intact T/T mutant (Table 2). Chimeric embryos with an ES cell contribution greater than 70%, as judged by ES cell glucose phosphate isomerase (GP1) isozyme activity, appear almost indistinguishable from intact mutants. Embryos with a lower but detectable ES cell contribution almost invariably have an abnormal allantois, some thickening of the primitive streak and sectioned material shows regions of necrotic or absent notochord (Rashbass et al.. 1991). The extent of neural tube disruption and the number of somites formed is more variable in embryos that are less than 70% chimeric. In the absence of a single cell marker, which distinguishes all mutant cells from wild-type cells, it is impossible to be dogmatic about the cell autonomous nature of these effects but the cell death seen in the notochord and beneath the primitive streak suggest that T may have a cell autonomous function in these tissues. Furthermore, chimerism in the allantois invariably results in defective development of this extraem-bryonic tissue and the severity of the abnormalities correlates with the degree of chimerism. This too could be a cell autonomous effect if there is a requirement for allantoic precursors to express T as they invaginate through the streak.

This analysis has now been extended to examine what happens to T/T cells in low level chimeras later in gestation (V. Wilson, P. Rashbass and R. Beddington, unpublished data). Initially, blastocysts injected with T/Tor 77+ ES cells were reimplanted in pseudopregnant recipients and allowed to develop to term. Table 3 shows the frequency of liveborn young and the incidence and level of chimerism. As expected, T/+ ES cells could contribute to viable liveborn chimeras. This had already been witnessed in preimplantation embryo aggregation chimeras where the liveborn T/+ ↔ +/+ chimeras were genotyped retrospectively according to tail length and transmission of the mutant T allele to offspring (Bennett, 1978). Blastocysts injected with T/T ES cells gave rise to liveborn young but no coat colour chimerism was evident in those pups that survived. Five of these surviving offspring had short or deformed tails. A further 3 neonates, which died shortly after birth, had short, curly or absent tails. Surprisingly, only a trace of GPI IA activity was found in the tail remnant of one of these animals (Table 3). Nonetheless, the striking tail abnormalities strongly suggested that these 8 offspring had been chimeric at some stage of their development. However, despite the lack of coat colour chimerism, it was possible that all 8 might simply be extremely low level chimeras (less than 2% contribution) which could not be detected by the GPI assay. Alternatively, they might once have contained detectable levels of ES cell contribution which had been selected against during the latter part of gestation. To resolve this question we examined embryos injected with T/T or T/+ ES cells on the 11 th and 12th days of gestation.

Table 4 shows the incidence of normal and abnormal chimeras recovered on the 11 th or 12th day of gestation. In general, T/+ ↔ +/+ chimeras either appeared grossly normal or displayed a high incidence of mild tail defects, such as kinking or slight irregularity of the tail tip. The allantois had fused with the chorion and as expected there were no visible abnormalities in the cranial or trunk region of the embryo reminiscent of the homozygous phenotype. In contrast, T/T ↔ +/+ chimeras show a range of defects extending from a severe phenotype, similar to dying intact homozygous mutants, to localised defects in the allantois or distal tail region. The tail defects included truncation, branching and the appearance of blood filled sacs at the caudal extreme of the tail. Normal and abnormal embryos were subdivided into different axial regions and the extent of chimerism measured following GPI gel electrophoresis.

Embryos classified as normal were not chimeric, with the exception of a single embryo, which proved to have an ES contribution of approximately 10%. On the other hand, over 85% of the abnormal embryos contained T/T ES cell descendants (Table 4). A very interesting axial pattern of chimerism was evident (Fig. 2). In all chimeras there was an increase from head to tail in the T/T cell contribution. Indeed, in some embryos a contribution was only evident in caudal structures (Fig. 2) or in the allantois (data not shown). Enzymatic separation (Beddington, 1987; Levak-Svajger and Svajger, 1971) of forelimb bud and hindlimb bud regions of some abnormal embryos into separate neural tube and paraxial and lateral mesoderm fractions also showed a consistent trend. The neural tissue was almost invariably more chimeric than mesodermal tissues in T/T ↔+/+ (Fig. 3). This was not the case in T/+ ↔+/+ chimeras w’here there was a more or less equivalent level of chimerism in the neural tube and mesoderm fractions.

There are three possible explanations for this rostrocaudal gradient of T/T ES cell colonisation. Embryos earlier in development could have been equally chimeric throughout, but there was subsequently strong selection against T/T cells in rostral regions. This would seem unlikely since the gene is not expressed in the head and no defects are apparent in this region, which might be expected if there were extensive cell death. Certainly, no such cell death has been observed in intact T/T mutants. Moreover, high level chimeras contain significant populations of T/T cells in the cranial region (e.g. Embryo 20, Fig. 2). The second possibility is that T/T ES cells assume a very abnormal position in the epiblast before or during gastrulation, compared to the random distribution observed with wild-type ES cells (R.Beddington, unpublished observations; Suemori et al., 1990), which leads to this preferred caudal fate. Again the high level chimeras make this improbable because T/T cells can be distributed throughout the axis. Furthermore, inspection of separated germ layer derivatives (Fig. 3) would argue that T/T cells were present in anterior regions of the epiblast which give rise to neurectodermal tissues (Beddington, 1981, 1982; Lawson et al., 1991). The third, and most plausible, explanation is that T/T cells accumulate at the caudal end of the embryo because once invaginated during gastrulation they fail to migrate away from the streak efficiently. Consequently, T/T mesodermal cells emerging from the streak will tend to end up in the cloacal region (the conventional anatomical boundary between primitive streak and tailbud derived embryonic tissues). They will also make a relatively higher contribution to the tail bud when it begins to form at the caudal end of the embryo during the early forelimb bud stage. This will lead to an elevated level of T/T chimerism in the tail compared to more rostra) levels of the embryo. Proliferation of the mesodermal cells in the tail bud is responsible for elongation of the tail (Tam, 1984). If T/T cells are compromised in their ability to move away from this growth zone then they will further accumulate at the distal tip of the growing tail. The restriction of tail abnormalities to this region in T/T ↔+/+ chimeras endorses this interpretation.

A prediction that follows the hypothesis that T/T cells fail to migrate away from the streak would be that the level of chimerism in the neurectoderm, a tissue that never invaginates, should always be higher than the level of chimerism in the paraxial and lateral mesoderm (Fig. 4). This prediction is bome out in T/T ↔+/+ chimeras when compared to 77+ +++/+ embryos (Fig. 3). This bias against T/T cells colonising mesodermal tissues rostral to the primitive streak also implies that this impairment of movement is a cell autonomous effect, rather than being caused by extracellular matrix defects, since wild-type cells can evidently move away from the streak normally. Therefore, we would predict that a gene or genes activated by T expression would code for intracellular or cell surface components involved in active movement or in cell adhesion.

A possible model for the mode of action of the T gene can be made. Assuming, on the basis of the dominant negative mutant phenotypes (see above), that T acts as part of a transcription factor dimer or complex (Herrmann, 1992) then one function of this complex would be to regulate a gene, or genes, whose products are cell autonomous components of adhesion or migration processes. In T^Wîs/T^Wîs or Tc/Tc embryos the abnormal T protein can still form a dimer or protein complex but this binding of the mutant T protein results in an inactive complex. Consequently, a subset of adhesion or motility proteins would be altered in the cells beneath the streak, thereby seriously compromising their morphogenetic movements. In T/T embryos, the absence of T protein does not affect the availability of the other component(s) of the transcriptional regulator and there is enough residual activity in a transcription complex lacking T protein to produce low level transcription of downstream genes. As a result, morphogenetic movements of the mesoderm are still affected but less severely than in T^WîsT^Wîs or Tc/Tc In the chimera where T/T cells are in competition with wild-type ones these motility defects result in accumulation of T/T cells at the caudal end of the streak or tail.

This scenario does not fully explain either the defects in the notochord or the abnormalities in the allantois. The sustained expression of T in the notochord after gastrulation has ceased indicates that the gene product may have additional function in this tissue and the cell death observed in the notochord of T/T ↔ +/+ chimeras (Rashbass et al., 1991) would be consistent with the T protein being required for cell survival. The effects on the allantois are more mysterious. Analysis of in vitro chimeras, formed by grafting [³H]thymidine-labelled cells into the primitive streak of late-streak or early-somite stage embryos, indicate that the allantois forms as a self-contained population of cells in less than 24 hours (Tam and Beddington, 1987). Chimeras created at the late-streak stage contain labelled cells in the allantois whereas those grafted at the early-somite stage do not. It is possible that accumulation of T/T cells at the caudal end of the embryo physically blocks the recruitment of cells into the allantoic bud. However, the degree of chimerism expected to produce such an efficient blockage might be expected to result in concomitant tail chimerism. This is not always the case; some T/T <->+/+ chimeras, albeit a low percentage, are abnormal only in the allantois and contain detectable numbers of T/T cells only in this tissue. A more compelling argument against simple physical blockage is that in chimeras there does not appear to be an under representation of T/T cells in the allantois. More over, the effect must be more complex than a simple ‘logjam” effect because there is a characteristic alteration in the morphology and behaviour of allantoic cells. They show a pronounced preference to spread over the amnion rather than cohere and traverse the exocoelom. Thus, the absence of T expression results in the altered phenotype of cells which even in wild-type embryos are not expressing the gene. It is possible, therefore, that T expression is an obligatory component of the initial specification of normal allantoic tissue within the streak region, or that it is an essential early step in the subsequent differentiation of allantoic precursors.

This work was supported by the Agriculture and Food Research Council and the Imperial Cancer Research Fund. We would like to thank Linda Manson and Louise Anderson for their valuable technical assistance.