Lmo2 and Scl/Tal1 convert non-axial mesoderm into haemangioblasts which differentiate into endothelial cells in the absence of Gata1

In mammalian and avian embryos, primitive erythrocytes form in the extraembryonic yolk sac where clusters of mesodermal cells give rise to both haematopoietic cells (HCs) and angioblasts, early progenitors for endothelial cells (ECs). Inside the embryo in the aorta-gonad-mesonephros (AGM) region,definitive HCs develop in association with the floor of the dorsal aorta. This close relationship of HCs and ECs suggests the existence of a common progenitor, the haemangioblast (Keller,2001).

The concept of the haemangioblast has gained support from several findings. Firstly, angioblasts and haematopoietic progenitors share the expression of several genes (Keller, 2001). Secondly, targeted disruption of the gene for the vascular endothelial growth factor (VEGF) receptor-2, flk1, in the mouse(Shalaby et al., 1995) and mutation of the zebrafish cloche locus affect both the blood and endothelial lineages (Stainier et al.,1995). Thirdly, single Flk1-positive (Flk1⁺) cells (so called blast colony forming cells; BL-CFCs) derived from embryonic stem (ES)cells can give rise to blast colonies (BL-C) that contain both blood and endothelial progenitors in vitro (Choi et al., 1998; Faloon et al.,2000; Nishikawa et al.,1998). However, haemangioblasts have not yet been isolated from a vertebrate embryo. The best in vivo evidence for the existence of the haemangioblast comes from lineage labelling studies in the chick(Jaffredo et al., 1998; Jaffredo et al., 2000). These studies also suggest that definitive HC clusters directly differentiate from the ventral endothelial lining of the dorsal aorta. Consistent with this idea,ECs isolated from the AGM region and the vitelline and umbilical arteries of murine embryos possess haematopoietic stem cell activity(North et al., 2002), and ECs selected from the AGM region, the foetal liver and the foetal bone marrow of human embryos develop into myelo-lymphoid cells in culture(Oberlin et al., 2002).

In zebrafish, primitive erythrocytes and ECs of the major trunk vessels,the dorsal aorta and the axial vein, form in close association in the intermediate cell mass (ICM) in the posterior midline of the embryo 1 day after fertilisation. These cells originate from the posterior lateral mesoderm(PLM) of the post-gastrula embryo (Zhong et al., 2001) (Fig. 1D). Here, overlapping expression patterns of blood and endothelial genes suggest the existence of haemangioblast-like cells(Brown et al., 2000; Gering et al., 1998; Thompson et al., 1998). Expression patterns of blood and endothelial genes also overlap in the anterior lateral mesoderm (ALM, Fig. 1D) which gives rise to ECs(Roman and Weinstein, 2000)and myeloid HCs (Hsu et al.,2001). In this paper, we will refer to such cells co-expressing blood and endothelial genes as haemangioblasts.

One of the genes expressed in both the ALM and the PLM is the stem cell leukaemia/T-cell acute lymphoblastic leukaemia gene scl/tal1, which encodes a basic helix-loop-helix transcription factor(Gering et al., 1998). It was first discovered through its involvement in chromosomal translocations in leukaemic T-cells. It has since been found to be expressed in early blood progenitors and in cells of the erythrocyte, megakaryocyte and mast cell lineages, as well as in endothelial and neuronal progenitors(Baer, 1993; Begley and Green, 1999). scl/tal1^-/- mouse embryos die at day 9.5 from a complete lack of primitive haematopoiesis (Robb et al., 1995; Shivdasani et al.,1995) and scl/tal1^–/– ES cells cannot take part in either primitive or definitive haematopoiesis in chimaeric animals (Porcher et al., 1996; Robb et al., 1996). Furthermore, a conditional knockout of scl/tal1 in the adult revealed that scl/tal1 is important for the specification of the haematopoietic stem cell rather than its maintenance(Hall et al., 2003; Mikkola et al., 2003). The loss-of-function studies did not reveal an essential role for scl/talin vasculogenesis. However, although ECs appear to develop normally, they fail to form a primary vascular plexus and to undergo subsequent angiogenic remodelling (Elefanty et al.,1999; Visvader et al.,1998).

Gain-of-function studies suggest an earlier role for scl/tal1consistent with its expression in early lateral mesoderm prior to the separation of the blood and endothelial lineages. We reported previously that Scl/Tal1 can induce ectopic development of cells co-expressing both early blood and endothelial genes mainly from the paraxial mesoderm and that these cells appeared to give rise to blood and later some ECs(Gering et al., 1998). In a parallel study, forced expression of scl/tal1 was shown to partially rescue expression of blood and endothelial genes in the zebrafish cloche mutant (Liao et al.,1998). Both studies suggest that Scl/Tal1 can specify blood and ECs possibly through the specification of the haemangioblast.

Scl/Tal1 forms complexes with other transcription factors. It heterodimerises with ubiquitously expressed bHLH transcription factors encoded by the E2a gene, which bind E-boxes in regulatory sequences and activate gene transcription (Hsu et al.,1991; Hsu et al.,1994). Scl/Tal1 also interacts with the LIM domain transcription factors Lmo1 and Lmo2, with whom it shares involvement in T-cell leukaemias(Rabbitts, 1998; Valge-Archer et al., 1994; Wadman et al., 1994). Like scl/tal1, lmo2 is expressed in endothelial and haematopoietic progenitors and in cells of the erythroid and megakaryocytic lineage(Warren et al., 1994), and is essential for primitive and definitive haematopoiesis, as well as vascular remodelling (Warren et al.,1994; Yamada et al.,1998; Yamada et al.,2000). In erythroid cells, Lmo2, which itself does not bind DNA,acts as a bridging molecule in a complex that contains Ldb1, Scl/Tal1, E2a and Gata1 and binds composite E-box-GATA sequence motifs(Wadman et al., 1997). In early blood progenitors, Gata2 can replace Gata1, and Sp1 takes the complex to an essential Sp1 site within the c-kit promoter(Lecuyer et al., 2002). Later in erythroid maturation, the retinoblastoma protein has been reported to replace the GATA factor (Vitelli et al.,2000). Consistent with an important role for the Lmo2-Scl/Tal1-Gata1/2 complex in erythropoiesis, forced expression of scl/tal1, lmo2 and gata1 can turn activin or FGF-treated Xenopus animal caps into erythrocytes and can induce widespread erythropoiesis in Xenopus embryos(Mead et al., 2001). However,these studies were carried out at concentrations sufficient to induce Bmp4 and the embryos were ventralised. In addition, only differentiated erythroid cells were monitored and the effects on other blood cell types or endothelial cell differentiation were not addressed. Whether the red cells followed their normal differentiation pathway, developing from haemangioblasts, was also not determined.

In this report, we describe a role for a synergistic action of Lmo2 and Scl/Tal1 in the early mesoderm. We show that scl/tal1 specifies haemangioblast-like cells in the very tissue in which it can ectopically induce the expression of its interaction partner Lmo2. Co-injection of scl/tal1 and lmo2 mRNAs extends the effect of Scl/Tal1 to other non-axial mesodermal tissues inducing the early blood and endothelial transcription programmes throughout the anteroposterior axis. Blood differentiation, however, is restricted to the pronephric mesoderm where Lmo2 and Scl/Tal1 are able to induce gata1 expression. In the absence of Gata1, Lmo2-Scl/Tal1-induced haemangioblasts differentiate into ECs.

Breeding zebrafish were maintained and embryos were raised and staged according to Westerfield (Westerfield,1993).

cDNAs for murine Lmo1, 2 and 4, and zebrafish Gata1, 2 and 5 were cloned into the expression vector pβUT2-MT(Gering et al., 1998). Oligos were used in PCR reactions on plasmid DNA to introduce suitable restriction sites at the ends of open reading frames:

• murine Lmo1: 5′GATTCTAGACGCCTGGGGACTATGGTTTTGGAC3′ and 5′GATCTCGAGCCTGAACCTGGGATTCAAAGGTGCC3′

• murine Lmo2: 5′GATTCTAGACCAATGTCCTCGGCCATCGAAAGG3′ and 5′GATCTCGAGCGATGATCCCATTGATCTTGGTCCA3′

• murine Lmo4: 5′GATTCTAGAACCATGGTGAATCCGGGCAGCAGC3′ and 5′GATCTCGAGCGCAGACCTTTTGGTCTGGCAGTAG3′

• zebrafish gata1: 5′CTCCTCTAGAGCCACCATGGAGAACTCCTCTGAGCCTTC and 5′CTCCCTCGAGACACTAGTGTGGGCATCATGCC3′

• zebrafish gata2: 5′GCCGATATCCACTTGAGATGGAGGTTG3′ and 5′CGGCTCGAGAGTGTGGTTCGGCCAG3′ (construct made by A. R. F. Rodaway)

• zebrafish gata5: 5′CTCCTCTAGATATATCATGTATTCGAGCCTG3′and 5′CTCCCTCGAGACGCTTGAGACAGAGCACAC3′ (construct made by A. Gibson).

PCR fragments were digested with XbaI and XhoI, and cloned in frame with the Myc-tag of pβUT2. To produce mRNA encoding zebrafish Scl/Tal1, β-galactosidase, nuclear localised GFP and murine MyoD, pFC6 (Gering et al.,1998), pCS2lacZ, pCSnlsGFP2 (a linker encoding a nuclear localisation signal with the sequence 5′GAATTCCCCAAAAAAGAAGAGAAAGGTAGAATTC3′ was introduced into the EcoRI and XbaI sites of pCS2mt-SGP(Rubenstein et al., 1997)(construct made by Maz O'Reilly) and MmyoD.R1/Bam(Theze et al., 1995) were used. mRNAs were transcribed from linearised templates using mMessage-mMachine-Kits (Ambion, USA). mRNAs were characterised spectrophotometrically, on agarose gels and by in vitro translation.

mRNAs (100 pg of each mRNA in all experiments except the gata1-scl/tal1-lmo2 injections where 25 pg of each mRNA were injected) were injected in 0.5 nl into one cell at the two or four cell stage using a Picospritzer II microinjector (Parker Instrumentation). gfpor lacZ mRNA (20 pg) were co-injected as tracers. The progeny of the injected cell occasionally came to lie on one side of the embryo, leaving the uninjected side as an internal control.

scl/tal1^end. and gata1^end.expression were determined with probes complementary to the 3′UTRs,which were not in the injected mRNAs. pZE62 contains 0.8 kb of the scl/tal1 3′UTR (Gering et al., 1998). peG1 contains the last 0.3 kb of the 3′UTR of gata1. It was generated by a HindIII cut-and-shut of the original plasmid, pGATA1 (Detrich et al.,1995). The zebrafish lmo2 probe did not hybridise significantly to the injected mouse Lmo2 mRNA and could therefore be used to detect endogenous transcripts. Antisense RNAs for in situ hybridisation were transcribed from linearised templates using Promega's T3,T7 and SP6 RNA polymerases in the presence of digoxigenin (DIG)- or fluorescein-labelled nucleotides (Roche Ltd., UK). Detection of the DIG/fluorescein antibody-alkaline phosphatase conjugate was done using BM-Purple and Fast Red (Roche Ltd., UK).

Single and double whole-mount in situ hybridisation on zebrafish embryos was carried out as previously described(Jowett and Yan, 1996). For sectioning, embryos were transferred into ethanol, embedded in JB4 methacrylate (Agar Scientific Ltd., UK) and sectioned on a microtome (Leica Jung RM2165). All sections shown are 10 μm.

Co-expression of scl/tal1 and lmo2 in blood and endothelial progenitors of the ALM and PLM(Fig. 1Aa-b)(Gering et al., 1998; Thompson et al., 1998) raises the possibility that Scl/Tal1 and Lmo2-containing transcription factor complexes may be active outside the erythroid lineage. We have shown previously that ectopic expression of scl/tal1 alone in zebrafish induces haemangioblast formation in the SPM(Fig. 1D)(Gering et al., 1998). To determine if the restriction of this induction to the SPM correlated with expression of Lmo2, we examined lmo2 expression in scl/tal1-injected zebrafish embryos. We found that lmo2 was like endogenous scl/tal1 (scl/tal1^end.)[Fig. 1Ab,e, arrow; 1D, n=15/20(Gering et al., 1998)] and the endothelial gene fli1 (Brown et al., 2000; Thompson et al.,1998) (Fig. 1Ac,f;arrow, n=35/55), ectopically expressed in the SPM(Fig. 1Ad; arrow; n=18/22; 1D), but not in the head paraxial mesoderm or in the nkx2.5⁺(Chen and Fishman, 1996)cardiogenic mesoderm (Fig. 1Ad)that lies between the ALM and the PLM (Fig. 2A). Thus, the tissue in which scl/tal1 induces ectopic lmo2 expression is the very tissue in which it can switch on the haematopoietic and endothelial transcription programmes. This suggests that Scl/Tal1-induced blood and endothelial development might be spatially restricted by the ability of Scl/Tal1 to activate endogenous lmo2expression.

To test whether providing Lmo2 exogenously through coinjection could spread the effect of Scl/Tal1 to other parts of the mesoderm, we co-injected mRNAs coding for zebrafish Scl/Tal1 and murine Lmo2 into zebrafish embryos and examined the expression of scl/tal1^end., fli1 and lmo2^end.. While control injections [gfp(n=26), lmo2 (n=37)] had no effect and scl/tal1 injection caused ectopic expression only in the SPM, scl/tal1-lmo2-co-injection had far more dramatic consequences. All three genes were now expressed laterally throughout the anteroposterior axis including the heart region (Fig. 1Ag-i,brackets;D) and in a wider domain in the head(Fig. 1Ag-i, arrowheads; D), in addition to their ectopic expression in the SPM(Fig. 1Ag-i, arrows; 1D; n_scltal1=15/15, n_fli1=34/35, n_lmo2=14/14). Ectopic expression coincided with maximal expression of the co-injected lacZ lineage marker(Fig. 1Ba,b). Transverse sections through a flat-mounted embryo confirmed that the ectopic expression of scl/tal1^end. was in the mesodermal layer(Fig. 1Cc-h). This suggests that ectopic expression of scl/tal1 and lmo2 together significantly extends the range of mesodermal territories in which the blood and endothelial transcription programmes can be induced.

This phenotype was specific to the injection of scl/tal1 and lmo2 mRNAs. Replacing Scl/Tal1 with the murine myogenic bHLH protein MyoD [which induced ectopic anterior expression of the muscle marker desmin (Xu et al.,2000)] in 12-somite stage embryos (n=13/15) induced no ectopic scl/tal1^end. expression (n=12) and reduced normal expression in 5/12 embryos. Likewise, embryos co-injected with Scl/Tal1 and another LIM-only protein, murine Lmo4(Grutz et al., 1998; Kenny et al., 1998), did not display widespread ectopic expression of scl/tal1^end.(n=20). In contrast, Lmo1, a Lim-only protein that, like Lmo2 and Scl/Tal1, is involved in chromosomal translocations in leukaemic T-cells(Rabbitts, 1998) and that like Lmo2 has been reported to interact with Scl/Tal1 in vitro(Valge-Archer et al., 1994),could replace Lmo2 in our assay (n=17). Thus, Lmo2 and Lmo1 are functionally equivalent in this assay as well as during tumorigenesis(Rabbitts, 1998).

Ectopic expression of scl/tal1 and lmo2 abolished normal cardiac gene expression. At early somite stages, nkx2.5(n=17/19), gata4 (Reiter et al., 1999) (n=10/16), gata6(n=12/22) and tbx5.1(Begemann and Ingham, 2000)(n=6/12) were down-regulated in scl/tal1-lmo2-injected embryos (Fig. 2A,C-I, arrows). Only gata5 (Brown et al.,2000; Reiter et al.,1999) expression appeared unchanged (weak reduction in 2/18; data not shown). Control embryos injected with gfp, lmo2 or scl/tal1 displayed very little reduction in nkx2.5 (0/12,0/20, 0/21) or gata4 (1/16, 0/20, 1/15) expression at this time. Consistent with the loss of early heart gene expression, we found a reduction in ventricular tissue just prior to the fusion of the bilateral heart fields(Fig. 2J,K, arrow; n=8/17) and a reduction in the overall size of the myocardium by 34 hpf (Fig. 2L,M; n=14/17). These results suggest that blood and endothelial development occurred at the expense of the myocardial fate.

To see whether the induced progenitors develop into erythrocytes, we examined the expression of two erythroid genes, beta embryonic globin 1 (βE1) (Quinkertz and Campos-Ortega, 1999) and alas2(Brownlie et al., 1998), in scl/tal1-lmo2-injected embryos. At 22 hpf, erythrocytes are usually located in the ICM in the posterior midline of the embryo(Fig. 3A). In scl/tal1-lmo2-injected embryos, mesodermal cells expressingβ E1 and alas2 were still restricted to the posterior part of the embryo, albeit remaining in more lateral positions(Fig. 3B, 10/10, 13/13). However, the numbers of βE1⁺ cells were increased compared to uninjected embryos. This expansion of erythroid cells requires the co-injection of lmo2 because there was only minimal expansion with scl/tal1 alone (Gering et al.,1998) (data not shown). Therefore, although many of the early blood and endothelial progenitors induced in scl/tal1-lmo2-injected embryos did not become erythrocytes, some local expansion was apparent.

To characterise this expansion more precisely, we analysed the expression of the early erythroid gene gata1(Detrich et al., 1995), which at the 10-somite stage (14 hpf) is expressed in the lateral cells that later give rise to erythocytes (Fig. 3C, blue stain, 3 hours). Its expression normally reaches as far anterior as somite 5 or 6 (Fig. 3C, arrows), however, in scl/tal1-lmo2-injected embryos, gata1 expression was expanded to somite 1(Fig. 3D arrows; H; n=32/36). The lateral mesoderm adjacent to somites 1 to 5 usually expresses the pronephric gene wt1 at the 6 somite stage (13 hpf; Fig. 3E,H)(Drummond et al., 1998; Weidinger et al., 2002). In scl/tal1-lmo2-injected embryos, wt1 expression was reduced at the same stage (Fig. 3F,G,H; n=8/14) suggesting that erythropoiesis had been induced at the expense of pronephros development. As seen for the later erythroid markers, gata1 expression was excluded from ectopic fli1⁺cells in the head (Fig. 4D,green arrow), the heart (Fig. 4D, bracket) and the SPM (Fig. 4D, black arrow) (n=12).

The expression pattern of gata1 in scl/tal1-lmo2-injected embryos was also 5-7 cells wide instead of 2-4 cells wide(Fig. 3C,D,H; n=33/36). gata1 expression usually only overlaps the medial aspect of the fli1 expression pattern(Fig. 4B, blue arrow), but in injected embryos it was expanded into the lateral aspect(Fig. 4D, blue arrows). This lateral aspect of the fli1 expression domain normally coincides with expression of the pronephric duct (PND) marker pax2.1(Fig. 4E,F)(Krauss et al., 1991),suggesting that these lateral fli1⁺ cells are PND progenitors and that gata1 expression had been induced in them by scl/tal1-lmo2-injection (Fig. 4I). Indeed two out of three PND markers were affected in scl/tal1-lmo2-injected embryos, with pax2.1 and pax8 (Pfeffer et al.,1998) expression reduced or missing at 14 hpf (pax2.1: Fig. 4G,H; n=15/20; pax8: n=13/19) but lim1 appeared to be unaffected(n=12). These results suggest that ectopic expression of scl/tal1 and lmo2 in PND progenitors induces ectopic expression of erythroid genes while compromising early PND gene expression. In contrast, scl/tal1 alone had no effect on early pax2.1expression, only on its maintenance in differentiating PND cells(Gering et al., 1998).

In summary, our data show that ectopic haemangioblasts induced by Scl/Tal1 and Lmo2 only undergo erythroid differentiation in pronephric mesoderm and that this correlates with the induction of gata1 expression. Importantly, the timing of both the haemangioblast and erythroid gene expression programmes corresponds to the naturally induced cells.

To determine if the addition of gata1 to scl/tal1 and lmo2 could overcome the tissue restrictions to erythroid differentiation seen in our system, we co-injected mRNAs for all three transcription factors into zebrafish embryos. By 24 hpf, widespread development of round, βE1⁺ (data not shown; n=13/13), alas2⁺(Fig. 5B, black arrowhead; n=13/13) and haemoglobin-containing (diaminofluorene⁺;data not shown; n=10/10) cells likely to be erythrocytes was seen in the heart and head regions. Earlier, at the 10 somite stage (14 hpf),endogenous gata1 (gata1^end.) also displayed ectopic anterior expression (Fig. 5E,F; n=9/9). Ectopic expression of alas2 and gata1^end. was not observed after injection of either gata1 alone (n_alas2=20, n_gata1=19) or with scl/tal1(Fig. 5C; n_alas2=20, n_gata1=35). In fact, gata1 injection causes morphologic malformations(Lyons et al., 2002) and often led to reduced blood marker expression(Fig. 5C). The induction of the erythroid programme is Gata1-specific because Gata1 could not be replaced by Gata2 (Fig. 5D, n=23)or Gata5 (data not shown; n=19) in this assay. The gata2 and gata5 mRNAs were active as gata2 reduced chordinexpression in the embryonic shield (n=7/29)(Sykes et al., 1998) and gata5 induced a slight increase in endodermal sox17⁺ cells at 90% epiboly (n=11/20)(Reiter et al., 2001; Weber et al., 2000) (data not shown).

Mead et al. (Mead et al.,2001) suggested that in Xenopus embryos gata1-scl/tal1-lmo2 injection induces erythroid development through the induction of bmp4 expression and ventralisation of the embryo. In zebrafish, however, injection of 25 pg of each mRNA neither activated ectopic bmp4 expression at tailbud stage (10 hpf; Fig. 5G,H; n=20) nor obviously ventralised/posteriorised the embryo, yet was sufficient to cause anterior erythropoiesis (Fig. 5B). Head structures, as illustrated by the pax2.1expression pattern in the brain (Fig. 5B, red arrows), were clearly present. Only injection of 100 pg of each mRNA caused wider bmp4 expression in half of the embryos(n=10/20) at the same stage. However, this expression was ventral not dorsal (Fig. 5I,J, arrowheads mark ventral expression), suggesting that embryos were retarded and displayed earlier normal expression rather than ectopic expression. Since eve1(Joly et al., 1993) expression in ventroposterior mesoderm was also normal (data not shown, n=11),we conclude that even the 100 pg-injected embryos were not ventralised/posteriorised, suggesting that anterior erythropoiesis was not a consequence of gross ventralisation/posteriorisation of the embryo, but rather a specific induction of the erythroid programme by Gata1 in haemangioblasts induced by Scl/Tal1 and Lmo2.

To determine the potential of the induced haemangioblasts to make blood lineages other than erythroid, expression of early myeloid genes was examined in scl/tal1-lmo2-injected embryos at the 10-12 somite stage (14-14.5 hpf). The four early myeloid genes analysed, pu.1(Lieschke et al., 2002), runx1 (Kalev-Zylinska et al.,2002), c-myb(Thompson et al., 1998) and dra (Herbomel et al.,1999), were expressed in the ALM that gives rise to myeloid cells(Fig. 6A,C,E,G, arrowheads). They are also expressed in the PLM caudal to somite 5, which mainly gives rise to red blood cells (Fig. 6A,C,E,G, black arrows). In addition, dra is expressed in early ventral mesodermal progenitors around the tail bud(Fig. 6G, green arrow). In scl/tal1-lmo2-injected embryos, the expression of these four myeloid genes was not dramatically expanded in the head(Fig. 6,B,D,F,H, arrowheads). pu.1⁺ and c-myb⁺ cells appeared disorganised (Fig. 6B,F,arrowheads; n=23, n=17), while the expression of runx1 and dra was unaffected or even reduced compared to uninjected embryos (Fig. 6D,H; n=19, n=15). c-myb and dra were induced in the cardiogenic mesoderm (Fig. 6F,H, brackets; n=14/17, n=11/15), however, runx1 and pu.1 were not expressed there(Fig. 6B,D; n=0/19, n=0/23). In addition, the number of L-plastin⁺(Herbomel et al., 1999)macrophages that develop from the ALM was either unchanged or slightly reduced but never increased in 20 hpf scl/tal1-lmo2-injected embryos(Fig. 6I,J; n=18).

We also studied myeloid gene expression in the PLM. The posterior expression of runx1, c-myb and dra was expanded anteriorly(Fig. 6D,F,H, blue arrows; n=10/19, n=14/17, n=11/15) and laterally(Fig. 6D,F,H, purple arrows; n=13/19, n=13/17, n=11/15), but these cells also expressed gata1, suggesting that they were undergoing erythroid rather than myeloid differentiation. Furthermore, Pu.1, a transcription factor known to interfere with erythropoiesis(Rekhtman et al., 1999; Zhang et al., 2000), did not display a similar expansion of its expression pattern(Fig. 6B, black arrow). Interestingly, dra expression was also expanded in the posterior SPM(Fig. 6H, green arrow; n=11/15), suggesting an expansion of early ventral progenitors that have not yet switched on early blood and endothelial genes like scl/tal1,lmo2 and fli1 (Fig. 6H and Fig. 1Ad-i). We conclude that regional controls over myeloid differentiation are not over-ridden by expression of scl/tal1 and lmo2.

In the absence of erythroid or expanded myeloid differentiation in the head and heart mesoderm, and in the SPM in scl/tal1-lmo2-injected embryos,we found that the expression of the early endothelial genes flk1(Fouquet et al., 1997; Liao et al., 1997; Sumoy et al., 1997) and flt4 (Thompson et al.,1998) was expanded in the head(Fig. 7C,F, arrowheads; n=20/20, n=15/19), into the heart mesoderm(Fig. 7C,F, brackets; n=20/20, n=6/19) and into the SPM(Fig. 7C,F; arrows; n=14/20, n=15/19) at 10 somites (14 hpf)(Fig. 7N). While the expansion into the head and heart mesoderm required the coinjection of lmo2with scl/tal1, the latter alone was sufficient to induce flk1 and flt4 in the SPM(Fig. 7B,E, arrows; N).

The endothelial nature of these cells was further substantiated by two findings. Firstly, we found ectopic expression of the late endothelial genes tie1 (data not shown) and tie2(Lyons et al., 1998) in the somitic mesoderm of scl/tal1-injected embryos(Fig. 7G,H; n=11/15)and all along the anteroposterior axis of scl/tal1-lmo2-injected embryos (Fig. 7G,I-K; n=17/17) at 22 hpf. This suggests that many of the ectopic flk1⁺ and flt4⁺ cells identified in injected embryos at 14 hpf differentiated into tie1- and tie2-expressing ECs by 22 hpf. Secondly, these cells appeared to have lost their haemangioblast character in that they no longer expressed scl/tal1^end. (Fig. 7K,M). While anterior ectopic tie2 expression was prominent (Fig. 7K; arrowhead),ectopic expression of scl/tal1^end. was very limited(Fig. 7L,M). The posterior expression pattern of scl/tal1^end. closely resembled that of βE1 (Fig. 3B)in that it marked ventrolateral cells that had not migrated towards the midline (Fig. 7M, red arrows). By contrast, in the same anteroposterior position, tie2 expression was found in the midline (Fig. 7K, red arrow). The only region of the injected embryo where scl/tal1^end. and tie2 appeared to be co-expressed ectopically was in the posterior SPM (Fig. 7K,M, brackets). We assume that these cells represent the less mature descendants of the expanded posterior, dra⁺,ventral mesodermal progenitors that we identified at 14 hpf(Fig. 6H, green arrow). Thus,we conclude that many of the early progenitors identified at 14 hpf develop into ECs. These results suggest that in the absence of haematopoietic induction cues, the endothelial transcription programme matures and becomes stabilised in Scl/Tal1-Lmo2-induced haemangioblasts.

In this report, we show that the transcription factors Scl/Tal1 and Lmo2 can synergistically act to specify haemangioblasts from non-axial mesoderm in zebrafish embryos. Our data thereby identify a potential novel role for Lmo2-Scl/Tal1-containing transcription factor complexes in early post-gastrula mesoderm, additional to their already established functions in erythropoiesis and leukaemogenesis. We show that while ectopic expression of scl/tal1 is sufficient to induce ectopic expression of early blood and endothelial genes in the somitic paraxial mesoderm (SPM), induction in other parts of the mesoderm requires the co-injection of lmo2 mRNA. In the presence of Lmo2, Scl/Tal1 induces early haematopoietic and endothelial transcription programmes (including the genes fli1,scl/tal1^end. and lmo2^end.) throughout the lateral mesoderm, in a wider domain in the head and in the SPM. Such a role is consistent with the co-expression of these two genes in the ALM and the PLM which are thought to harbour haemangioblasts.

In scl/tal1-lmo2-injected embryos, ectopic blood and endothelial development occurs at the expense of other mesodermal tissues, in particular the somitic, pronephric and cardiogenic mesoderm. Scl/Tal1 and Lmo2 appear to interfere with the endogenous transcription programmes, activating early blood and endothelial genes instead. Since Scl/Tal1 and Lmo2 activate these genes synergistically, and since they are known to form transcription factor complexes (Rabbitts, 1998), we assume that gene activation involves recruitment of components of the complex endogenously. Absence of such endogenous transcription factors is likely to render ectopic Scl/Tal1 and Lmo2 ineffective. Obvious candidates are proteins already known to bind Scl/Tal1 and Lmo2, such as the E proteins and Gata factors, respectively. Their absence might explain why the notochord is always unaffected (data not shown).

Repression of endogenous transcription programmes could result from activation of repressors or direct repression by Scl/Tal1-Lmo2 complexes. Alternatively, Scl/Tal1 and Lmo2 could scavenge transcription factors that are essential for an alternative transcription programme. Scl/Tal1 is believed to scavenge E proteins normally required to interact with myogenic bHLH proteins(Goldfarb and Lewandowska,1995; Hofmann and Cole,1996) or for the expression of E protein target genes in pre-leukaemic T-cells, requiring Lmo1/2 in the latter context(Herblot et al., 2000; Larson et al., 1996). Gata factors are known to interact with Lmo2 and other Lim-only proteins(Chang et al., 2003; Osada et al., 1995). In the PLM, gata1 and gata2 are expressed. In the SPM, gata2 is induced ectopically by injection of scl/tal1 and scl/tal1-lmo2 (M.G. and R.K.P., unpublished). In the ALM, gata2 is expressed together with gata4, 5 and 6,which are also expressed in the cardiogenic mesoderm(Reiter et al., 1999; Rodaway et al., 1999) (A. Gibson and R.K.P., unpublished). Although scl/tal1-lmo2 injection down-regulates gata4 and 6 in the heart region, gata5 expression is almost unchanged. It is possible that Scl/Tal1 and Lmo2 recruit heart Gata factors, leading not only to induction of the early blood and endothelial transcription programmes but also to a blockade of heart-specific gene expression. It remains to be seen whether Gata4, 5 and 6 can interact with Lmo2 and which interaction partners Scl/Tal1 and Lmo2 require in the various mesodermal tissues in which they can ectopically switch on blood and endothelial genes.

While induction of the haemangioblast programme requires both Scl/Tal1 and Lmo2 in the head and the heart mesoderm, injection of scl/tal1 mRNA is sufficient to induce it in the SPM. We show here that only in this tissue can Scl/Tal1 induce ectopic lmo2 expression. This suggests that once lmo2 is activated, the two partners can synergistically drive the expression of genes like fli1, flk1, flt4, tie1 and tie2. It does, however, not explain why Scl/Tal1 can induce lmo2 in the SPM in the first place. It is possible that a Lmo2 paralogue is expressed in the SPM and assists Scl/Tal1. Alternatively, Scl/Tal1 could simply scavenge E proteins. This might be enough to abrogate somitic development given the vast array of bHLH proteins it depends on and, in the absence of other developmental options, SPM cells may undergo endothelial development. It should be noted that in mammalian and avian embryos endothelial development occurs from somites after the onset of blood circulation(Ambler et al., 2001; Beddington and Martin, 1989; Pardanaud et al., 1996; Wilting et al., 1995). It is therefore possible that ectopic scl/tal1 expression induces earlier expression of the intrinsic endothelial programme.

Erythropoiesis only proceeds from Scl/Tal1-Lmo2-induced haemangioblasts located in the PLM, not from the heart, head or somitic mesoderm. This correlates with where gata1 expression was induced. Consistent with a previous study (Mead et al.,2001), we show that co-injection of Scl/Tal1, Lmo2 and Gata1 causes widespread induction of erythropoiesis. However, we also demonstrate that this is due to relief of the posterior restriction, leading to induction of gata1^end. expression and erythropoiesis anteriorly. We additionally show that other Gata factors, in particular Gata2, cannot replace Gata1. Furthermore, gata1-scl/tal1-lmo2-injected embryos are not ventralised/posteriorised as eve1 expression is unchanged and bmp4 expression is not obviously activated, suggesting that gata1-scl/tal-lmo2-injection can induce anterior erythropoiesis downstream of the dorsal-ventral patterning of the mesoderm.

In our hands, injection of gata1 alone was not sufficient to induce either anterior gata1^end. expression or head erythropoiesis. This result conflicts with a recent paper showing that injected gata1 mRNA can induce expression of a gfp reporter gene under the control of the gata1 promoter in transgenic zebrafish(Kobayashi et al., 2001). We assume that the difference lies with the transgenic reporter construct used or its chromosomal integration site. Nevertheless, our data confirm that Gata1 can auto-regulate its own expression, albeit with the help of Scl/Tal1 and Lmo2.

We observed no increase in the number of myeloid cells in scl/tal1-lmo2-injected embryos, whereas erythropoiesis was expanded laterally and anteriorly. The anterior expansion occurred into the region of the pronephros adjacent to somites 1-5. This is surprising given that scl/tal1 and lmo2 are expressed in this region at the 10 somite stage (14 hpf). However, at the 6 somite stage (13 hpf), they are not expressed in this part of the PLM but are restricted to the PLM posterior to somite 6 (data not shown). Whether their more anterior expression at the 10-somite stage is a result of cell migration or de novo expression of scl/tal1 and lmo2 is currently unknown. Nevertheless, our data argue that premature/ectopic expression of scl/tal1 and lmo2 in the PLM adjacent to somites 1-5 is sufficient to induce ectopic erythropoiesis in a region of the embryo that normally forms the pronephros.

The lateral expansion occurred at the expense of the PND. We showed previously that PND development was compromised in scl/tal1-injected embryos, however, initial expression of the PND gene pax2.1 was normal at the 10-somite stage and only displayed gaps by 22 hpf(Gering et al., 1998). In contrast, in scl/tal1-lmo2-injected embryos, pax2.1 and pax8 expression was already lost at the 10-somite stage. Thus,Scl/Tal1 and Lmo2 together are affecting the PND-specific transcription programme at a much earlier stage than Scl/Tal1 on its own. Loss of pax2 and pax8 in the mouse completely abrogates pronephric development (Bouchard et al.,2002). Interestingly, lim1 expression was not lost in the PLM of scl/tal1-lmo2-injected embryos. In the mouse embryo, lim1 appears to be a competence factor that marks the territory competent for pronephric induction. Its expression is initially independent of pax2/8 which is consistent with its continued expression in scl/tal1-lmo2-injected zebrafish embryos. The continued expression of lim1 further suggests that Scl/Tal1 and Lmo2 have no negative influence on its expression and that lim1 expression does not interfere with erythropoiesis.

Scl/Tal1 and Lmo2 activated some but not all blood genes in the heart, head and somitic mesoderm. In the absence of induction of the complete myeloid or erythroid transcription programmes, blood cells did not develop from these tissues. Instead, the cells first switched on flk1 and flt4,later expressed tie1 and tie2 and eventually down-regulated scl/tal1^end. Thus, many Scl/Tal1-Lmo2-induced progenitors appeared to differentiate into ECs, suggesting that endothelial development may be the default fate of the haemangioblast.

Since yolk sac ECs develop in lmo2^–/–(Warren et al., 1994; Yamada et al., 2000) and scl/tal1^–/–(Elefanty et al., 1999; Visvader et al., 1998) mice,it would appear that Lmo2 and Scl/Tal1 are dispensible for angioblast formation and only confer haematopoietic competence to the mesodermal tissue they are expressed in. However, results from studies on scl/tal1^–/– ES cells(Chung et al., 2002; Faloon et al., 2000; Robertson et al., 2000) and the absence of contributions to adult haematopoiesis or angiogenesis by lmo2-null ES cells in chimeric mice(Yamada et al., 1998; Yamada et al., 2000), suggest that these proteins are needed for correctly programmed haemangioblast formation. Haemangioblast-like BL-CFCs derived from ES cells are highly enriched in the scl/tal1-positive fraction of Flk1⁺ cells(Chung et al., 2002), and BL-CFCs do not form in the absence of a functional scl/tal1 gene(Faloon et al., 2000; Robertson et al., 2000). Ubiquitous scl/tal1 expression in differentiating ES cells increases the number of Flk1⁺ cells and subsequently the number of primitive blood progenitors among the Flk1⁺ cells as well as the number of VE-Cadherin⁺ definitive blood progenitors (haemogenic ECs)(Endoh et al., 2002). scl/tal1 also negatively influences smooth muscle cell differentiation from early ES cell-derived Flk1⁺ cells in favour of more haemangioblasts (Ema et al.,2003). However, ECs do develop from scl/tal1^–/– ES cells(Faloon et al., 2000; Robertson et al., 2000) and from lmo2^–/– ES cells (M. M. McCormack and T.H.R., unpublished) indicating that endothelial development from ES cells can occur independently of haemangioblast formation and of lmo2 or scl/tal1 expression. Yet, ECs formed in the absence of lmo2or scl/tal1 are not normal as they fail to form a primary plexus or undergo vascular remodelling in the embryo(Elefanty et al., 1999; Visvader et al., 1998; Yamada et al., 2000; Yamada et al., 2002). The molecular basis of these defects is not known. Interestingly, rescue experiments show that primitive as well as definitive HCs only form from scl/tal1^–/– ES cells if Scl/Tal1 function is restored at the stage when the cells are still of mesodermal character,suggesting that even the haemogenic ECs that give rise to definitive HCs need to have experienced scl/tal1 expression long before they give rise to HCs (Endoh et al., 2002). These results suggest that the early lack of Lmo2 or Scl/Tal1 in the mesoderm can have very late consequences in the endothelial lineage and the data presented here support a model whereby these two factors cooperate to specify normal haemangioblasts from early mesodermal progenitors.

R.K.P., M.G., T.H.R. and Y.Y. were supported by the Medical Research Council and Y.Y. also by the National Foundation for Cancer Research. We would like to thank A. Forster for his help in this work and R. F. Bachvarova, A. Ciau-Uitz, A. D. Johnson, M. Walmsley and the reviewers for critical comments on the manuscript. We are grateful to T. Mohun, A. R. F. Rodaway, M. O'Reilly and A. Gibson for expression constructs. We would also like to thank L. Zon,M. Hammerschmidt, B. Thisse, M. Fishman, D. Stainier, G. Lieschke, P. Ingham,M. Busslinger, I. Drummond, S. Wilson, J. Campos-Ortega and K. Crosier for probes.