Insufficient VEGFA activity in yolk sac endoderm compromises haematopoietic and endothelial differentiation

It has been demonstrated that VEGFA plays an important role in a variety of physiological and pathological conditions. To date there is a far better understanding of VEGFA function and regulation in pathological conditions – such as tumour growth and wound healing – than that of its role in embryogenesis or in normal adult tissues (for a review, see Neufeld et al., 1999). However, understanding the function of VEGFA in physiological conditions would be important for providing a better basis for the modulation of vessel formation or regression in pathological conditions such as ischaemia and cancer.

Previously performed, targeted gene inactivation studies clearly indicated an absolute requirement for VEGFA in early stages of embryonic development. These experiments showed that a 50% reduction in gene dose impairs vascular development and leads to embryonic lethality at mid-gestation (Carmeliet et al., 1996; Ferrara et al., 1996). The distinct expression pattern during organogenesis (Breier et al., 1992; Miquerol et al., 1999) suggests that VEGFA may be involved in a multitude of developmental processes, including postnatal growth and survival (Gerber et al., 1999).

Vasculogenesis, the main mechanism of vessel formation in early development, is based on the in situ formation of blood vessels from independently differentiating angioblasts. This process first takes place in the yolk sac (Risau and Flamme, 1995). The concomitant development of the first blood cells in this embryonic membrane facilitates nutrient supply to the embryo via the vitelline circulation. The simultaneous development of haematopoietic and endothelial cells, and the results of in vitro differentiation experiments (Choi et al., 1998) suggest the existence of a common precursor, the haemangioblast, for these two lineages. The precursors of the yolk sac mesodermal component are located in the proximal epiblast. From there they migrate to their final extra-embryonic position (Lawson et al., 1991) where they give rise to endothelial and haematopoietic cells. However, a crucial tissue transplantation experiment to detect such a bi-potential precursor population for these cells failed to confirm its existence, and suggested a clear segregation of the two lineages following mesodermal cell migration through the primitive streak (Kinder et al., 1999). The two lineages, however, are connected through their dependence on VEGFR2, one of the VEGFA receptors. Targeted disruption of the Vegfr2 gene (Kdr – Mouse Genome Informatics) resulted in a diverted migration of these cells to non-yolk sac mesodermal membranes such as the amnion (Shalaby et al., 1997). As a consequence, both endothelial and blood cells were missing in the mutant yolk sac. Interestingly, however, the authors showed that VEGFR2-deficient ES (embryonic stem) cells did have the potential to differentiate into blood cells in vitro. This finding has been confirmed by independent studies (Schuh et al., 1999; Hidaka et al., 1999), and extrapolation to in vivo development suggests that VEGFR2 might not be essential for the initiation of haematopoietic differentiation up to 7.5 days of embryonic development, but it is required for later lineage expansion.

After establishment of blood islands in the yolk sac, the primary function of VEGFA/VEGFR2 signalling appears to be anti-apoptotic. Although a survival factor function for VEGFA has previously been described (Alon et al., 1995), Carmeliet et al. have recently resolved a distinct antiapoptotic pathway in early vascular development involving VE-cadherin and VEGFA/VEGFR2 (Carmeliet et al., 1999a). Interference with this pathway is most probably the reason for the increased apoptosis of endothelial cells observed in the VEGFA loss-of-function mutation (Ferrara et al., 1996).

Owing to the lethal haploinsufficiency of the Vegf (previously Vegfa) knockout, it was not possible to carry out the study of the homozygous deficient phenotype with standard methods. We bypassed this obstacle by aggregating homozygous Vegf^–/– ES cells with tetraploid wild-type embryos. This approach allowed the study of a completely ES cell-derived, VEGFA-deficient embryo proper (Carmeliet et al., 1996). These embryos were supported by wild-type extra-embryonic endoderm and trophoblast of the placenta. Therefore, the assay did not permit the investigation of the role of VEGFA in the extra-embryonic tissues. Interestingly, however, the extra-embryonic endoderm is one of the first tissues to express VEGFA during development (Miquerol et al., 1999), starting at least 3 days earlier than the expression in the embryo proper.

Already in the late 1960s Miura and Wilt noticed that the extra-embryonic endoderm, as secretory epithelium was absolutely required for proper blood island formation (Miura and Wilt, 1969). This observation was confirmed by a later study that investigated Gata4^–/– embryoid bodies that could not form visceral endoderm. In this study, Bielinska et al. demonstrated that formation and organisation of blood islands was dependent on the presence of visceral endoderm (Bielinska et al., 1996). These findings were extended by a report by Belaoussoff et al. (Belaoussoff et al., 1998) showing that the visceral endoderm has an instructive function for haematopoietic and endothelial development during a narrow window of time. Only recently have the same group provided evidence that Indian hedgehog is one of the molecules derived from visceral endoderm that is responsible for the induction of endothelial and haematopoietic differentiation (Dyer et al., 2001).

We have recently reported the generation of a lacZ knock-in into the 3′untranslated region (UTR) of the mouse Vegf gene (Miquerol et al., 1999) and Vegf^lacZKI (Fig. 1A). This allele facilitated the characterisation of VEGFA expression at the cellular level both during development and in the adult organism. Interestingly, the modification of the gene resulted in a slight (two- to threefold) overexpression of VEGFA in the developing organs. The increased VEGFA expression was incompatible with normal heart development and resulted in homozygous embryonic lethality (Miquerol et al., 2000). Together with the previously observed haploinsufficiency, this phenotype, which is due to slight overexpression, was a clear demonstration of minimal tolerance to alterations in VEGFA expression levels. The study of VEGFA expression in the yolk sac showed that both the visceral endoderm and the yolk sac mesodermal sheet express VEGFA (Miquerol et al., 1999). These two layers sandwich the first site of blood and vessel formation of the embryo. This finding has raised the question of whether VEGFA expression in both layers is necessary or whether one of them would be sufficient for proper vessel and haematopoietic differentiation in the yolk sac. We address this question by a novel hypomorphic allele generated by a slightly modified insertion of lacZ into the 3′ UTR of Vegf, distinct from the lacZ insertion reported previously (Miquerol et al., 1999). The new allele (Vegf^lo; Fig. 1A) allowed heterozygous survival but conferred homozygous embryonic lethality by 9.0 dpc. In the homozygous embryos, the formation of blood islands in the yolk sac, as well as the development of the dorsal aortae, were severely defective.

A combination of chimaera studies showed that VEGFA expression in the yolk sac mesoderm sheet was not sufficient for normal vascular and haematopoietic development in this extra-embryonic membrane. However, expression in the visceral endoderm was necessary and sufficient for normal development of the yolk sac vasculature.

Vegf genomic sequences were derived from a λ genomic clone described earlier (Carmeliet et al., 1996). The 5′ homology arm, comprising exons 4-7 and the coding region of exon 8, was generated by fusing a SmaI-NcoI fragment (extending into intron 7) with a PCR amplified fragment terminating at the VEGFA stop codon. A PGK neomycin cassette flanked by loxP sites (Takeuchi et al., 1995) was inserted directly downstream of the stop codon followed by IRES-lacZ (Jang et al., 1988; Takeuchi et al., 1995) fused to the Vegf 3′UTR (Damert et al., 1997). All construction steps and exon sequences were verified by dideoxy sequencing.

The targeting vector was linearised with NotI and electroporated into R1 ES cells (Nagy et al., 1993). G418-resistant ES cell clones were screened by Southern blot using BglII digestion of the genomic DNA and a 400 bp BglII-EcoRI probe encompassing Vegf exon 3. Correct targeting was further confirmed by EcoRV digestion and hybridisation with a 1 kb HindIII probe derived from the Vegf 3′ flanking region. Several independent clones were aggregated with CD-1 morulae as described before (Nagy and Rossant, 2000). Germline transmission was obtained for two clones (B7 and C7). F₁ progeny was subsequently bred to a deletor Cre recombinase transgenic mouse to delete the selectable marker cassette (Nagy et al., 1998) and obtain Vegf^lo/+ mice. All further analysis was performed on a mixed CD-1/129 background.

Vegf^lo offspring were PCR genotyped using the following oligonucleotides: 5′ stop, 5′-GAGCATTTATTCCCATGTCTG-3′; lacPCR, 5′-CATTACCAGTTGGTCTGGTG-3′; and fusionAS, 5′-CAGGCTTTCTGGATTAAGGAC-3′. These yielded fragments of 388 bp (wild type) and 163 bp (targeted allele). To distinguish the Vegf^lo and Vegf^lo+neo alleles, a sense oligonucleotide from the PGK pA (Neo, 5′-CCACATACACTTCATTCTCAG-3′) and an antisense from lacZ (lacZ-276R, 5′-GATGGGCGCATCGTAACCGTGC-3′) were used, resulting in amplification of a 1000 bp fragment from the Vegf^lo+neo allele.

Embryos at different stages of development were dissected in cold PBS, fixed in 4% paraformaldehyde, paraffin-embedded and sectioned at 5 μm. Sections were stained with Haematoxylin/Eosin. Whole-mount immunohistochemistry using an antibody against PECAM-1 was performed as described (Davis, 1993). Procedures employed for detection of β-galactosidase activity are described elsewhere (Miquerol et al., 1999). Whole-mount in situ hybridisation and detection with AP-conjugated antibody was carried out according to the protocol published by Wilkinson and Nieto (Wilkinson and Nieto, 1993). Probes were kindly provided by J. J. Bieker (EKLF); H. Stuhlmann (VEZF) and J. Rossant (VEGFR1 and VEGFR2).

Total RNA from embryos was prepared using Trizol Reagent (Life Technologies) according to the manufacturers instructions. Reverse transcription was performed on 1 μg of RNA using the oligonucleotide lacZ-276R 5′-GATGGGCGCATCGTAACCGTGC-3′ and Superscript^TM (Life Technologies) Reverse transcriptase at 55°C. The resulting cDNA was amplified using the oligonucleotides lacZ-276R and VEGF-D, 5′-GCCCTGGAGTGCGTGCCCACGTCAGAGAGCA-3′. The PCR product was gel purified and cloned into pGEM-T-easy (Promega) for subsequent dideoxy sequencing. Protein extracts were obtained by homogenising embryos in 250 mM Tris-HCl pH 7.8 and three cycles of freezing and thawing. Protein concentration was determined using the BioRad reagent. VEGF immunoreactive material was quantified with the Quantikine ELISA kit (R&D) according to the manufacturers instructions.

Tetraploid embryos for aggregation experiments were generated as described (Nagy et al., 1993). Aggregation of diploid embryos or GFP+ ES cells (Hadjantonakis et al., 1998) with tetraploid embryos was performed according to the protocol given elsewhere (Nagy and Rossant, 2000). The resulting chimaeras were dissected at 8.5-9.5 dpc for subsequent analysis. For all chimaera experiments, small pieces of the embryo proper were removed for genotyping using the oligonucleotides given above.

An IRES (internal ribosomal entry site)-lacZ cassette was inserted into the Vegf locus by homologous recombination in ES cells, with the primary aim of generating a lacZ tagged allele where expression of the reporter gene would exactly mimic the pattern of VEGFA expression. In contrast to our recently described Vegf^lacZKI allele (Miquerol et al., 1999) the IRES-lacZ cassette was inserted directly adjacent to the STOP codon located in exon 8. The lacZ-coding sequence was followed by the Vegf 3′ UTR including the endogenous polyadenylation signals (Fig. 1A,B). We anticipated that after a Cre recombinase-mediated removal of the loxP flanked neomycin selectable marker, the targeted allele should facilitate lacZ expression under the control of all known Vegf regulatory elements, including the 3′UTR (Levy et al., 1996; Shima et al., 1996).

Electroporation into R1 ES cells (Nagy et al., 1993) yielded 12 correctly targeted clones (Southern blot analysis is shown in Fig. 1C). Germline transmission after ES cell/morula aggregation (Nagy and Rossant, 2000) was achieved for two independent ES cell clones. Heterozygous animals were viable, fertile and did not show any obvious phenotypic alterations. As the selectable marker cassette was still present in these animals, they were subsequently bred to a general deletor strain of Cre mice (Nagy et al., 1998), yielding viable offspring heterozygous for the neomycin-excised allele. Breeding these heterozygous mice with each other did not result in any live homozygous newborns. Subsequent analysis at different stages of embryonic development revealed that the homozygous embryos died between 8.5 and 9.0 dpc. Morphological evaluation of mutant embryos at 8.5-9.0 dpc showed the same defects as observed in Vegf^+/– embryos: disorganised yolk sac vasculature with large ‘lacunae’ and reduced lumina of the dorsal aortae (Fig. 2A-H). Sectioning demonstrated few and largely empty blood islands (Fig. 2I,J) and a reduction in the number of blood cells in the embryo proper, particularly noticeable in the sinus venosus (Fig. 2O,P). Development of the cardinal veins seems to proceed normally (Fig. 2M-P) and development of the heart is delayed; however, the endocardium is formed (Fig. 2M,N). To examine whether the impaired vascular development at the time of death results from regression of blood islands or defective recruitment of endothelial/haematopoietic precursors, we examined embryos at an earlier stage of development. Sections of mutant embryos at 8.0 dpc showed the same reduction in the number of blood islands devoid of primitive erythrocytes as observed at later stages, indicating that failure of precursor cell recruitment is the major cause for the defects in yolk sac vasculature (Fig. 2Q-T). The similarity to the Vegf^+/– phenotype implicates a hypomorphic function of the knock-in allele that we therefore termed Vegf low (Vegf^lo).

Analysis of marker molecules indicative for endothelial [VEGF-R1 (Fong et al., 1996), VEGFR2 (Yamaguchi et al., 1993) and VEZF (Xiong et al., 1999)] and haematopoietic [EKLF (Southwood et al., 1996)] differentiation revealed a severely reduced expression of VEGF-R1 and EKLF indicating a lack of mature endothelial and haematopoietic cells (Fig. 3). In the 8.5 dpc Vegf^lo/lo mutants, VEGFR2 expression was found in the cardinal veins and endocardium, albeit at a much lower level then in wild-type littermates (Fig. 4A,B). In the yolk sac, VEGFR2-expressing cells were found at the periphery of the rare blood islands (Fig. 4C,D). At 9.0 dpc, low level expression of VEGFR2 and VEZF was observed in the head region and in the intersomitic vessels that arise through angiogenesis (Fig. 3). This finding indicates that once vessels are formed by vasculogenesis, angiogenesis can proceed at low levels of VEGFA activity.

Taken together, these data suggest that the functionally hypomorphic Vegf allele affects differentiation of both the endothelial and haematopoietic lineages.

In order to understand the molecular mechanisms of the embryonic lethality observed, the analysis of both the RNA and protein resulting from the targeted allele was performed. RT-PCR analysis using oligo dT-primed cDNA from homozygous mutant embryos and oligonucleotides bridging Vegf- and lacZ-coding sequences yielded a 616 bp instead of the expected 1217 bp product (Fig. 5C). Furthermore, we could not detect any RT-PCR product when a reverse primer in the exon 8-coding region was used (data not shown). Cloning and sequencing the 616 bp fragment revealed the absence of the exon 8-coding region as well as of the entire IRES (Fig. 5A, lowest panel). Therefore, the mRNA produced from the Vegf knock-in allele should be translated into a modified VEGFA protein, in which the exon 8 encoded six C-terminal amino acids are replaced by 66 amino acids derived from the lacZ-coding region (Fig. 5B). As the lacZ-coding sequence present in the predicted fusion protein is not in frame with VEGFA, this protein cannot confer β-galactosidase activity. Surprisingly, although the entire IRES sequence is skipped in the process of splicing, β-galactosidase activity is detectable in the knock-in embryos and adult mice, the expression pattern corresponds to that recently described for the Vegf^lacZKI allele (data not shown) (Miquerol et al., 1999). Compared with measurements in wild-type littermates, a threefold increase in VEGFA immunoreactivity could be detected in the mutant embryos by ELISA (Fig. 5D). These results suggest that two different proteins can be translated from the chimaeric mRNA: a highly abundant C-terminally modified VEGFA and a functional β-galactosidase, probably supported by the strong lacZ Kozak sequence (Kozak, 1999).

In addition to skipping of the exon 8-coding sequence, disturbance in VEGFA alternative splicing involving exons 6 and 7 (Breier et al., 1992) was observed. Only a single mRNA containing exons 5 and 7 and thus corresponding to the VEGFA 164 isoform could be detected from the Vegf^lo allele (Fig. 5A,C).

Previously, we have demonstrated that VEGFA is expressed both in the visceral endoderm and the yolk sac mesoderm at the time when vasculogenesis occurs (Miquerol et al., 1999). Using chimaera approaches, we have addressed the requirement for both sources of this growth factor for normal yolk sac development. Green fluorescent protein (GFP)-expressing (Hadjantonakis et al., 1998) Vegf wild-type tetraploid embryos were aggregated with diploid Vegf^lo/lo embryos from the segregating F₂ generation (Fig. 6A). In these chimaeras, the tetraploid component contributed only to the yolk sac visceral endoderm and the trophoblast lineages, while the diploid component contributed to all lineages, including the embryo proper. However, we often observed a few ‘contaminating’ tetraploid cells in the gut of otherwise totally diploid embryos (Fig. 6B). Therefore, to establish the genotype of the diploid component of the chimaeras from each embryo proper recovered, we carefully removed a small part lacking GFP-positive cells for PCR analysis.

Yolk sac endoderm is one of the few lineages where the two chimaera components do not intermingle extensively (Gardner, 1984). Instead, large patches are formed, each consisting of one of the components. In the chimaeras presented here the tetraploid Vegf wild-type patches are distinguishable by GFP expression (Fig. 6B,C), whereas areas of diploid-derived Vegf^lo/lo visceral endoderm can be stained for β-galactosidase activity (Fig. 6E,I,L).

Rescue of blood island formation and vascular development was found in areas of the tetraploid (Vegf wild type) visceral endoderm (Fig. 6F-H,I,L). As the yolk sac mesoderm is Vegf^lo/lo in these chimaeras, this rescue indicates that the Vegf wild-type visceral endoderm is sufficient for normal blood and vessel formation in this membrane. In Vegf^lo/lo (β-galactosidase positive) areas of visceral endoderm, the formation of blood islands is severely compromised (Fig. 6F-I,L).

The sections show that the range of rescue by VEGFA secreted from wild-type tetraploid areas is very short, reaching a distance of only a few cell diameters (marked by an asterisk in Fig. 6L). This notion was further supported by the observation that even extensive tetraploid contribution was not sufficient to rescue the intra-embryonic phenotype. All embryos showed the same small dorsal aortae lumen (arrows in Fig. 6J) and delayed heart development (Fig. 6K) as did unmanipulated Vegf^lo/lo embryos. Interestingly, however, the haematopoietic cell population in the embryo proper increased with increasing contribution of Vegf wild-type tetraploid cells to the yolk sac visceral endoderm (red arrows in Fig. 6J,K). This finding is in accordance with the previously published notion that the yolk sac blood islands are a major source of early haematopoietic cells colonising the embryo proper.

To test whether endoderm expression of VEGFA is necessary for normal blood island formation and vessel development in the yolk sac, we performed a reverse chimaera experiment by aggregating Vegf^lo/lo tetraploid embryos with Vegf wild-type, GFP-expressing ES cells (Fig. 7A). In this case, the yolk sac endoderm was completely derived from tetraploid Vegf^lo/lo embryos (easily detectable by β-galactosidase staining, see Fig. 7C,G,H), whereas the entire extra-embryonic mesoderm, including endothelial and haematopoietic cells, was wild-type ES cell derived. As shown in Fig. 7B,C, the Vegf wild type, GFP-positive embryo proper developed normally until 9.5 dpc, the time of analysis. Histological sections revealed normal vessel formation along the entire embryo (arrowheads in Fig. 7F,I,J). However, only few blood cells could be observed in the AGM (aorta-gonad-mesonephros) region (Fig. 7F), the heart and the sinus venosus (Fig. 7I,J).

Comparison of yolk sacs derived from aggregations of tetraploid Vegf^lo/lo (Fig. 7D) and Vegf^lo/+ (Fig. 7E) embryos revealed a more uniform and intense GFP fluorescence in the latter setup. This finding already indicated proper migration of GFP positive mesodermal precursors and expansion of the lineages when the tetraploid component was heterozygous Vegf^lo/+ (Fig. 7E) that is not observed with tetraploid Vegf^lo/lo embryos (Fig. 7D). Sectioning as shown in Fig. 7G,H confirmed this notion. Appropriate yolk sac vascular development occurred only when the visceral endoderm was derived from a heterozygous Vegf^lo/+ embryo (Fig. 7G,H). In Vegf^lo/lo/wild-type ES chimaeric yolk sac only the embryo-derived vitelline vessels develop normally and show intense GFP fluorescence (Fig. 7D). We can therefore conclude that secretion of functional VEGFA from the visceral endoderm is not only sufficient, but is in fact necessary for proper blood island formation, and hence endothelial and haematopoietic differentiation in the yolk sac. The absence of blood cells in the rare blood islands corresponds well with the reduced number of blood cells in the embryo proper, highlighting the importance of the yolk sac as a source of haematopoietic cells at this developmental stage.

By insertion of an IRES-lacZ into the Vegf locus, we generated a new allele that, by comparison with the Vegf^–/– phenotype, confers a hypomorphic function. Although the exact mechanism of action requires further investigation, this allele enabled us to dissect the role of VEGFA in early embryonic development, especially in yolk sac vasculogenesis and haematopoiesis. Chimaera analysis using a combination of Vegf wild-type tetraploid and Vegf^lo/lo diploid embryos supported and extended the findings obtained from tetraploid embryo/Vegf^–/– ES cell aggregations. In particular, providing wild-type extra-embryonic membranes did not rescue the intra-embryonic vascular phenotype of embryos with insufficient level of VEGFA expression (Carmeliet et al., 1996). This might be explained by the finding that the dorsal aortae are formed by intra-embryonic vasculogenesis (Hirakow and Hiruma, 1981), which cannot be properly supported by the Vegf^–/– or Vegf^lo/lo embryo proper in these chimaeras.

Going beyond the knockout tetraploid setting, however, we created chimaeric yolk sac visceral endoderm by using Vegf^lo/lo embryos instead of ES cells as the diploid source. This approach allowed the study of wild-type and VEGFA-deficient essential lineages in the same embryo. Inspection of the chimaeric yolk sacs revealed that blood islands are only properly formed in areas where the visceral endoderm is composed of Vegf wild-type tetraploid cells, but not in Vegf^lo/lo patches of visceral endoderm. This observation already indicated that VEGFA activity derived from the visceral endoderm is a crucial element in blood island formation. Interestingly, rescue of blood island formation by wild-type visceral endoderm patches does not extend beyond a few cell diameters. Recently, Cleaver and Krieg identified Xenopus VEGF 122 as the ‘chemotactic’ isoform that acts on dorsal aorta-forming angioblasts over a considerable distance (Cleaver and Krieg, 1998). From these data it could be hypothesised that the smallest VEGFA isoform mediates chemotaxis, whereas the heparin-bound VEGFA 164 and 188 act more locally and induce differentiation once the final destination in the yolk sac is reached. In our experimental set-up, diffusible VEGFA 120 secreted from wild-type visceral endoderm might be able to facilitate precursor migration even through Vegf^lo/lo patches, resulting in an even distribution of cells over the whole yolk sac. The local action of the higher molecular weight isoforms in wild-type patches then allows further expansion of blood islands that is inhibited in areas of Vegf^lo/lo visceral endoderm. If at least one VEGFA isoform could act over a distance of more than a few cell diameters (and therefore beyond Vegf^lo/lo patches), blood island formation would be expected to be blocked close to the boundaries of the visceral endoderm. Certainly it would be interesting to see how blood island formation proceeded in a set-up where embryos that lacked a single VEGFA isoform (Carmeliet et al., 1999b) constitute the tetraploid component.

As yolk sac blood island formation was rescued by wild-type visceral endoderm, it was not surprising that the number of blood cells in the embryo proper was elevated with increasing contribution of wild-type visceral endoderm. Palis et al. have reported that definitive yolk sac derived erythroid and myeloid progenitors colonise the embryo with the onset of circulation (Palis et al., 1999). However, the still controversial issue of the relative contribution of yolk sac versus P-Sp (para-aortic splanchnopleura)/AGM-derived haematopoietic stem cells to foetal liver haematopoiesis (Keller et al., 1999) could not be addressed with this approach, as the embryos were not viable beyond 9.5 dpc because of intra-embryonic vascular malformations.

Although VEGFA was initially characterised as endothelial specific mitogen, it is now accepted that it affects haematopoietic development through VEGFR2 as well. F₁Vegf^+/– embryos display a severely reduced number of primitive erythrocytes in the yolk sac blood islands (Ferrara et al., 1996). Our analysis of the Vegf^lo/lo embryos led to similar findings. In vitro differentiation experiments have suggested an important function for VEGFA in blood cell differentiation. According to a report by Nakayama et al., emergence of progenitors for both the erythroid and lymphoid lineages in embryoid bodies is stimulated by VEGFA (Nakayama et al., 1998). Choi et al. could demonstrate that blast colonies with haematopoietic and endothelial potential were formed from VEGFA-responsive embryonic precursors in vitro (Choi et al., 1998). Additional experiments to assess the haematopoietic potential of VEGFA deficient progenitors are under way.

Failure of blood island formation in Vegf^lo/lo patches of chimaeric yolk sacs already indicated an important function of visceral endoderm-derived VEGFA. However, based on this first chimaeric approach, we could not exclude the possibility that VEGFA secreted by yolk sac mesoderm was involved in vasculogenesis in an autocrine manner. Until now Vegf^–/– or hypomorphic conceptuses, which are necessary for addressing the consequences of VEGFA malfunction in extra-embryonic tissues by tetraploid aggregations, were not available. The use of tetraploid Vegf^lo/lo embryos in combination with Vegf wild-type ES cells allowed us to investigate the effect of impaired VEGFA function in the visceral endoderm in a setting where the mesodermal target cell population was wild type. The failure of Vegf^lo/lo visceral endoderm to induce blood island formation properly in wild-type mesoderm convincingly demonstrates the intrinsic requirement for VEGFA in the yolk sac visceral endoderm. The importance of the visceral endoderm layer for proper blood island formation was discovered in the 1960s. Using chick embryos as a model system, Miura and Wilt found that although blood cells can differentiate in the absence of visceral endoderm, no endothelial cells and blood islands form. Trans-filter experiments suggested the involvement of soluble factors (Miura and Wilt, 1969). Our data now identify VEGFA as one of these factors. Using Gata4^–/– embryoid bodies that lack visceral endoderm, Bielinska et al. have reported that the visceral endoderm is essential for organisation of blood islands, but is not required for differentiation of primitive erythroblasts and endothelial cells (Bielinska et al., 1996). More recently, Belaoussoff et al. have investigated the function of visceral endoderm in mouse embryo explant cultures. They showed a requirement for visceral endoderm signals in haematopoietic and vascular development. It also became clear that this process was restricted to the onset of gastrulation, a narrow time window between 6.0 and 6.5 dpc (Belaoussoff et al., 1998). In the meantime, Dyer et al. have identified Indian hedgehog as a visceral endoderm-secreted factor that, alone, is sufficient to induce endothelial and haematopoietic differentiation in these cultures (Dyer et al., 2001). However, to date there is no direct link between signalling leading to vascular and haematopoietic development and the hedgehog cascade. Thus, the nature of the interplay between factors secreted by the visceral endoderm remains to be determined. Interestingly, Indian hedgehog is expressed in a proximal-to-distal gradient in the visceral endoderm of gastrulation stage embryos (Dyer et al., 2001) – a pattern that might be consistent with findings that a spatial (as well temporal) order of recruitment of blood island progenitors is present (Kinder et al., 1999). No such pattern has been established for visceral endoderm expression of VEGFA. Differences in the expression pattern of the VEGFA isoforms would be an attractive way in which to explain the movement of mesodermal precursor populations during gastrulation.

Regarding haematopoiesis, our chimaera approach might in perspective serve as a useful model for dissecting the contribution of extra- and intra-embryonic sites for definitive haematopoiesis. In tetraploid Vegf^lo/lo/wild-type GFP+ chimaeras the yolk sac haematopoiesis is completely abolished. Thus, it would certainly be interesting to investigate whether the embryos are viable beyond 9.5dpc, solely depending on intra-embryonic sites of definitive haematopoiesis.

Based on the data currently available on developmental expression and function of VEGFA and its receptors, as well as reports investigating the role of other molecules and tissues in gastrulation, the following model of blood island formation and vasculogenesis can be proposed. Mesodermal cells pre-destined to give rise to haematopoietic and/or endothelial cells (Nishikawa et al., 1998) leave the primitive streak attracted by VEGFA secreted from the yolk sac visceral endoderm (Shalaby et al., 1995; Shalaby et al., 1997). In the permissive environment of the yolk sac, they can complete their differentiation program. The haematopoietic and endothelial lineages can expand in response to VEGFA (Schuh et al., 1999; Hidaka et al., 1999) and their organisation into blood islands can be accomplished. During this process, VEGFR1 acts to restrict the expansion of the endothelial lineage (Fong et al., 1999).

The availability of the VEGFA hypomorphic allele described here should now facilitate further dissection of the molecular mechanisms involved in migration and endothelial/haematopoietic differentiation processes during gastrulation. The use of specific developmental restrictions of components of chimaeric embryos, such as tetraploid conceptuses and ES cells, to direct mutant and wild-type cell towards defined allocations in chimaeras was found to be a powerful tool to uncover the intrinsic requirement for VEGFA expression in the visceral endoderm.

We are particularly grateful to Sue MacMaster and Ken Harpal for their technical help, and Janet Rossant for helpful discussions and critical reading of the manuscript. We thank James J. Bieker and Heidi Stuhlmann for providing EKLF and VEZF probes, respectively. This work was supported by a grant from National Cancer Institute of Canada (NCIC). A. N. is a BMS-MRC scientist and L. M. is an NCIC postdoctoral fellow.