The primitive streak gene Mixl1 is required for efficient haematopoiesis and BMP4-induced ventral mesoderm patterning in differentiating ES cells

Gastrulation in vertebrates is a tightly coordinated process in which the pluripotent epiblast is converted into the three embryonic germ layers(Robb and Tam, 2004; Tam and Behringer, 1997; Tam et al., 2001; Tam et al., 2003). In mice,the earliest visible indicator of gastrulation is the formation of the primitive streak, a transient structure at the presumptive posterior of the embryo from which mesodermal and endodermal cells develop(Tam and Behringer, 1997). Several families of growth factors have been implicated in the regulation of gastrulation in the mouse, including fibroblast growth factors and members of the transforming growth factor β (TGFβ) superfamily(Conlon et al., 1994; Sun et al., 1999; Winnier et al., 1995).

Key targets of TGFβ signalling in gastrulating Xenopus laevisembryos include the Mix/Bix homeobox genes that regulate mesoderm and endoderm formation in response to nodal/activin and BMP4(Rosa, 1989; Vize, 1996). Mix.1 is induced by BMP4 and can ventralise mesoderm(Mead et al., 1996), while a number of other Mix/Bix genes induce endoderm(Henry and Melton, 1998; Latinkic and Smith, 1999; Lemaire et al., 1998; Tada et al., 1998). The two zebrafish Mix-related homeobox genes, bon and mezzo(og9x - Zebrafish Information Network) are immediate-early targets of nodal signalling that are transiently expressed in precursors of mesoderm and endoderm (Kikuchi et al.,2000; Poulain and Lepage,2002; Trinh et al.,2003). The avian Mix gene is expressed in the epiblast and posterior marginal zone endoderm just prior to gastrulation and in the primitive streak, excluding Hensen's node(Peale et al., 1998; Stein et al., 1998). Similarly, expression of the single mouse Mix gene homologue, Mixl1,is restricted to the visceral endoderm of the pre-gastrulation embryo and the primitive streak (Pearce and Evans,1999; Robb et al.,2000). Indeed, gene targeting has confirmed the important role that Mixl1 plays during gastrulation. Mixl1-null mutants display an enlarged primitive streak and subsequently exhibit abnormalities in axial morphogenesis and formation of definitive endoderm that result in death at embryonic day (E) 8.5 (Hart et al.,2002).

In mice, the first mesoderm to emerge from the primitive streak migrates extra-embryonically and forms the blood islands of the yolk sac at E7-7.5(Kinder et al., 1999). The first wave of haematopoiesis occurs concurrently with the formation of extra-embryonic vasculature, consistent with the development of these two lineages from a common progenitor (Keller et al., 1999; Lacaud et al.,2001). The in vitro differentiation of embryonic stem (ES) cells recapitulates many aspects of early haematopoietic development and represents a valuable model system to study a process occurring at a relatively inaccessible period of embryogenesis (Dang et al., 2002; Desbaillets et al., 2000; Keller,1995; Maye et al.,2000; Takahashi et al.,2003). Indeed, the Flk1 (Kdr - Mouse Genome Informatics)-positive blast colony forming cell (BL-CFC) capable of giving rise to both haematopoietic and endothelial lineages was first isolated from embryoid bodies (EBs), thus providing tangible evidence for the existence of an haemangioblast (Choi et al.,1998; Kennedy et al.,1997; Nishikawa et al.,1998).

We have used mouse ES cell lines in which Mixl1-coding sequences on one (Mixl1^GFP/w) or both(Mixl1^GFP/GFP) alleles were replaced by the gene encoding green fluorescent protein (GFP) (Hart et al., 2002) to investigate the role of Mixl1 in ventral mesoderm patterning and haematopoiesis. We have shown that a large proportion of differentiating Mixl1^GFP/w ES cells transiently expressed both GFP and Flk1 and that this doubly-positive population was enriched for BL-CFCs. However, in differentiating Mixl1-null ES cells, Flk1 expression was delayed and reduced and the frequency of haematopoietic CFCs was decreased. Differentiation of ES cells under serum-free (SF) conditions demonstrated that induction of Mixl1- and Flk1-expressing haematopoietic mesoderm required medium supplemented with BMP4 or activin A. Therefore, this study has revealed an important role for Mixl1 in haematopoietic development and demonstrated the utility of the Mixl1^GFP/w ES cells for evaluating growth factors influencing mesendodermal differentiation.

A cassette encoding enhanced GFP and a loxP-flanked neomycin resistance gene was knocked into the first exon of Mixl1 by homologous recombination as described (Hart et al.,2002). Two karyotypically normal heterozygous ES lines (M147 and M114) were chosen for subsequent experiments. To generate the Mixl1-null line M916, a plasmid encoding cre recombinase (pMC1Cre,provided by K. Rajewsky) (Gu et al.,1994) was transiently transfected into the M147 line to remove the neomycin resistance cassette prior to a second round of gene targeting using the original Mixl1 targeting vector. The Mixl1-null cell line M3C5 was created by targeting the wild-type allele of M114 with a Mixl1GFP targeting vector that incorporated a puromycin resistance cassette.

ES cells were cultured as described(Barnett and Köntgen,2001) and differentiated using the method of Kennedy et al.(Kennedy et al., 1997). For differentiation of ES cells under SF conditions, ES cells were resuspended at 5000 cells/ml in modified Chemically Defined Medium (CDM)(Johansson and Wiles, 1995)comprising IMDM/Ham's F12 with Glutamax (Gibco) supplemented with 5 mg/ml bovine serum albumin (Sigma), 1 U/ml LIF (Chemicon), 4.5×10^-4M α-MTG (Sigma), a 1% Chemically-Defined Lipid Concentrate (Gibco), 1%Insulin-Transferrin-Selenium-X Supplement (Gibco) and antibiotics. Activin A(0.1-100 ng/ml) or BMP4 (0.5-20 ng/ml) (R&D Systems) were added at the time of cell plating. Cultures were maintained at 37°C in a humidified environment of 8% CO₂ in air. Phase-contrast and fluorescent microscopy images of EB cultures were acquired using a Zeiss Axiocam mounted on an Axiovert 200 microscope and processed with Axiovision software. Single optical sections of EBs were taken using a Leica confocal scanning microscope.

Embryoid bodies were dissociated to single cells using trypsin/EDTA (Gibco)containing 1% chicken serum (Hunter). Cells were resuspended in a block solution (phosphate-buffered saline supplemented with 2% FCS, 1% goat serum and 1% rabbit serum) and incubated with primary antibodies directed against E-cadherin (ECCD-2, Zymed), FLK1 (VEGF-R2, Ly-73) conjugated to phycoerythrin(PE) (Avas 12α1, BD Biosciences), Ter-119 (Ly-76) conjugated to PE (BD Biosciences) and CD34 (RAM34) conjugated to biotin (BD Biosciences). Anti E-cadherin antibodies were detected with either PE or allophycocyanin(APC)-conjugated anti-rat IgG (BD Biosciences) while biotinylated anti-CD34 antibodies were detected with streptavidin-conjugated PE or APC. Cells were analysed using a FACSCalibur (Becton Dickinson) running CellQuest software(Becton Dickinson). For cell-sorting and reculture experiments,differentiating Mixl1^GFP/w EBs were dissociated, stained with antibodies against FLK1 and sorted according to GFP and FLK1 expression using a FACStar Plus (Becton Dickinson).

Haematopoietic colonies were generated by plating 2.5×10⁴-10⁵ dissociated EB cells into 1.5 ml of 1%methylcellulose in IMDM supplemented with 10% FCS, 25% D4T endothelial cell conditioned medium, 25 μg/ml ascorbic acid, 2 mM L-glutamine(Kennedy et al., 1997). To assay BL-CFCs, methylcellulose cultures were supplemented with 5 ng/ml Vascular Endothelial Growth Factor (VEGF) (R&D systems) and 50 ng/ml Stem Cell Factor (SCF) (R&D systems). For detection of BL-CFCs and primitive erythroid colonies (EryP), the growth factor combination used was 5 ng/ml VEGF, 50 ng/ml SCF, 5 U/ml EPO (Janssen Cilag) and IL3 (1% of a supernatant from a cell line producing mIL3)(Karasuyama and Melchers,1988). Colonies were scored after 5-7 days. Blast colonies were expanded in liquid culture supplemented with the same combination of growth factors and analysed after 7 days. Morphology was assessed by May-Grunwald-Giemsa staining of cytocentrifuge preparations and endothelial cells were identified by staining wells fixed with 4% paraformaldehyde (Sigma)with anti-PECAM1 antibodies (MEC13.3, BD Biosciences).

Total RNA was prepared using an RNeasy kit (Qiagen) according to the manufacturer's instructions. DNase I treated samples were reverse transcribed using Superscript II (Invitrogen) and the resultant cDNA preparations standardized as described (Elefanty et al., 1997). Primer sequences and annealing temperatures are shown in Table 1. References for Brachyury, FLK1, βH1 globin and HPRT primer sequences can be found in Elefanty et al. (Elefanty et al.,1997). PCRs were performed for 30 cycles with reaction conditions as described (Elefanty et al.,1997). PCR products were analysed by electrophoresis on a 2%agarose gel.

Consistent with the findings of others(Lacaud et al., 2002; Mohn et al., 2003; Robertson et al., 2000), gene expression analysis of differentiating mouse ES cells demonstrated the progressive downregulation of stem cell genes such as Rex1(Zfp42 - Mouse Genome Informatics), Oct4 (Pou5f1 -Mouse Genome Informatics), E-cadherin (Cdh1 - Mouse Genome Informatics) and Sox2 accompanied by the sequential acquisition of molecular markers of epiblast, primitive streak and mesoderm(Fig. 1). Expression of the epiblast marker Fgf5 peaked at day 2 followed by transient expression of the primitive streak genes brachyury, Mixl1 and Gsc at days 3 and 4. Transcription of Flk1, an early marker of ventral mesoderm,was up regulated from day 3, and βH1 globin, which reflects the development of primitive erythroblasts, was expressed from day 4. Thus, the kinetics of Mixl1 expression in differentiating ES cells was consistent with its restricted expression during primitive streak formation in vivo, and marked the period of transition between the epiblast and mesoderm.

ES cells in which GFP was expressed from one(Mixl1^GFP/w) or both (Mixl1^GFP/GFP)alleles of the Mixl1 locus have been described(Hart et al., 2002). Previous analysis had shown that Mixl1^GFP/w mice were phenotypically normal and that GFP expression was observed in the primitive streak, validating the use of GFP as a reporter of Mixl1-expressing cells (Hart et al., 2002). This analysis also confirmed that the neomycin resistance cassette did not affect the distribution of GFP in these animals or in the differentiating ES cells in vitro (data not shown).

Examination of differentiating Mixl1^GFP/w ES cells revealed that most day 3 and 4 embryoid bodies (EBs) expressed GFP, coinciding with endogenous Mixl1 expression (Figs 1, 2 and data not shown). Confocal images demonstrated that GFP was present in the outer, flattened endoderm-like cells and in the inner core of the EBs(Fig. 2C,D).

GFP expression during differentiation of Mixl1^GFP/wcells was examined by flow cytometry. In the representative experiment shown in Fig. 3, GFP expressing cells were first evident at day 2.5 and their frequency rapidly increased to 43.5%at day 3 and peaked at 85.2% at day 4 of differentiation. Although Mixl1 RNA was not detected after day 5(Fig. 1), GFP expression did not wane until after day 6, consistent with the long half life of this reporter protein (Corish and Tyler-Smith,1999). A similar time course of GFP expression was observed in two independent Mixl1^GFP/w ES lines (data not shown).

Mixl1-null EBs displayed a similar profile of GFP expression to heterozygous Mixl1^GFP/w ES cells with the frequency of GFP-positive cells peaking at ∼90% at day 4(Fig. 3). A higher intensity and prolongation of GFP expression in Mixl1^GFP/GFP EBs was consistently observed, probably reflecting the presence of two copies of GFP in the null cells.

Examination of the gene expression profile of differentiating ES cells demonstrated that day 3 EBs simultaneously expressed stem cell, primitive streak and mesodermal genes (Fig. 1). We used flow cytometry to correlate the expression of GFP with that of the stem cell marker E-cadherin (E-cad) and the ventral mesoderm marker FLK1 in differentiating Mixl1^GFP/w and Mixl1^GFP/GFP ES cells. In differentiating EBs from all Mixl1 genotypes, E-cad was expressed in over 90% of cells up until day 2 of differentiation (Fig. 4A and data not shown), consistent with its expression in the inner cell mass, epiblast and primitive streak in vivo(Ciruna and Rossant, 2001; Huber et al., 1996). Between day 2 and day 4, the proportion of E-cad⁺ cells fell to under 20%,reflecting loss expression in emerging mesoderm(Fig. 4B). The first cells expressing GFP were invariably E-cad⁺, consistent with the expected phenotype of primitive streak cells (Ciruna and Rossant, 2001; Huber et al., 1996) (Fig. 4B). As differentiation progressed, an increasing proportion of GFP⁺ cells lost E-cad expression, suggesting that they had `passed through' the streak. There was no significant difference in the E-cad expression profiles between Mixl1^GFP/w and Mixl1^GFP/GFP EBs (Fig. 4A and data not shown).

The profile of Flk1 expression was examined to determine the relationship between Mixl1 expression and the earliest ventral mesoderm cells (Fig. 5A). Trace amounts of Flk1 expression were first detected on day 2 of differentiation. In wild-type ES cells, the frequency of Flk1-expressing cells increased rapidly between day 3 and day 4 to∼70%, then fell to ∼20% by day 6. In both Mixl1^GFP/w and Mixl1^GFP/GFP EBs, a small population of GFP^-FLK1⁺ cells was consistently detected at day 2.5 (Fig. 5A). By day 3, differences between the Flk1 expression of Mixl1^GFP/w and Mixl1^GFP/GFP lines became apparent, with Mixl1^GFP/GFP EBs showing a reduced proportion of FLK1⁺ and GFP⁺FLK⁺ cells. This difference was more evident at day 4 of differentiation. Although the frequency of Flk1⁺ cells was similar in Mixl1^w/w (69.5%) and Mixl1^GFP/w(57.8%) EBs, the frequency of Flk1⁺ cells was lower in Mixl1^GFP/GFP EBs (28.2%)(Fig. 5A). Moreover, in the example shown, only 29% of GFP⁺ cells from day 4 Mixl1^GFP/GFP EBs expressed Flk1 compared with 61%from Mixl1^GFP/w EBs, suggesting that Mixl1 was required to generate normal proportions of FLK1⁺ cells(Fig. 5A). These observations were confirmed in experiments using independently derived Mixl1heterozygous and null ES cell clones (Fig. 5B,C). In wild-type and Mixl1^GFP/w EBs, the frequency of Flk1-expressing cells declined after day 4. By comparison, the highest frequency of Flk1⁺ cells in Mixl1-null EBs occurred 1 day later, at day 5 of differentiation(Fig. 5A).

In order to ascertain whether the abnormal Flk1 expression profile in Mixl1^GFP/GFP EBs foreshadowed a later haematopoietic defect, day 6 EBs were analysed for expression of the haematopoietic stem cell/endothelial marker CD34 and the late erythroid marker Ter119(Fig. 6A). Although there was no significant difference in the frequency of CD34-expressing cells, EBs derived from Mixl1-null ES cells invariably contained very small numbers of Ter119-positive cells and did not undergo visible haemoglobinization, indicating reduced formation of terminally differentiated erythroid cells.

To determine whether the reduced ability of Mixl1-null EBs to generate haemoglobinised erythroid cells reflected an underlying defect in haematopoietic progenitors, the frequency of haematopoietic colony-forming cells (CFCs) was examined in day 4 EBs(Fig. 6B). In experiments using two independent Mixl1^GFP/w and Mixl1^GFP/GFP ES cell lines, we found that Mixl1-null EBs contained significantly fewer blast (BL)-CFCs and primitive erythroid progenitors (EryP) than did Mixl1-heterozygous EBs. This reduction in the frequency of CFCs was also evident in Mixl1-null EBs at day 3, and persisted in day 6 and day 7 EBs despite an increase in the frequency of FLK1⁺ cells at these later time points (data not shown). These data confirmed that the absence of Mixl1 impaired haematopoiesis in differentiating ES cells.

In order to further explore the relationship between Mixl1expression and haematopoietic differentiation, the blast colony-forming ability of sorted fractions of anti-FLK-1 labelled Mixl1^GFP/w day 3 and day 4 EBs were compared. FACS sorted GFP^-FLK1^-, GFP⁺FLK1^-,GFP⁺FLK1⁺ and GFP^-FLK1⁺populations were cultured in methylcellulose in the presence or absence of vascular endothelial growth factor (VEGF) and stem cell factor (SCF) in order to detect haematopoietic blast colonies(Kennedy et al., 1997). Interestingly, BL-CFCs (Fig. 7A-D) were cultured from all sorted fractions from day 3 EBs,although the highest frequency was observed in the GFP⁺ fractions(219 colonies/5×10⁴ GFP⁺FLK1^- cells and 225/5×10⁴ GFP⁺FLK1⁺ cells)(Fig. 7E). Irrespective of their population of origin, all the blast cell colonies displayed similar morphology, dependence on VEGF and SCF and ability to give rise to haematopoietic and adherent vascular cells. The endothelial nature of the adherent cells was verified by staining with antibodies to PECAM1 or FLK1 (see Fig. 7A-D; data not shown). Consistent with the transient nature of BL-CFCs(Kennedy et al., 1997), their frequency was considerably decreased in day 4 EBs(Fig. 7F). At this time, the majority of BL-CFCs were found in the GFP⁺FLK1⁺fraction, although some were present in the GFP^-FLK1⁺population (Fig. 7F).

In order to dissect the developmental relationship between the GFP- and FLK1-expressing cells and to rationalize the emergence of haematopoietic blast colonies from FLK^- populations, sorted fractions taken at different time points were recultured and analysed. We wished to determine whether cells in the GFP⁺FLK1^- and GFP^-FLK1^-fractions from early timepoints, in which we had already found BL-CFCs(Fig. 7E), could differentiate into FLK1⁺ cells. This would allow us to postulate that a subset of the FLK1^- cells from day 3 EBs cultured in methylcellulose in VEGF and SCF upregulated FLK1 and responded to the factors by forming BCs. At day 2.8, EBs contained only 12.0% GFP⁺ cells and FLK1 expression was virtually absent (Fig. 8A). When overnight cultures of GFP positive and negative fractions were reanalysed, the GFP^- cells had differentiated into GFP⁺FLK1^- (10.7%), GFP⁺FLK1⁺(9.3%) and GFP^-FLK1⁺ (25.7%) cells, and 22.7% of the initial GFP⁺ cells had become GFP⁺FLK1⁺. These experiments demonstrated that day 2.8 GFP^- cells had considerable differentiative capacity and that the GFP⁺FLK1⁺ population enriched for BL-CFCs could arise from GFP^-FLK1^- or GFP⁺FLK1^- cells. GFP^-FLK1^- cells from day 3 EBs also possessed the capacity to give rise to GFP^-FLK1⁺,GFP⁺FLK1^- and GFP⁺FLK1⁺ cells(Fig. 8B). Interestingly, it appeared that some FLK1⁺ cells arose directly from the GFP^-FLK1^- population without passing through an intermediate GFP⁺ stage. FLK1 expression was also observed in 6.3%of re-cultured GFP⁺FLK1^- cells, indicating the persistence of a subset of the GFP⁺ cells with the ability to develop into GFP⁺FLK1⁺ precursors. Conversely,GFP^-FLK1⁺ cells rarely turned on GFP expression. Consistent with the transient nature of FLK1 expression during EB differentiation (see Fig. 5A),the majority of recultured GFP⁺FLK1⁺ and GFP^-FLK1⁺ populations actually lost FLK1 expression. Reanalysis of sorted day 4 EBs, revealed that the GFP^-FLK1^- population was no longer capable of differentiating into GFP⁺ or FLK1⁺ cells(Fig. 8C). One third of recultured GFP⁺FLK1^- cells lost GFP expression and this population displayed minimal ability to develop into new GFP⁺FLK1⁺ cells. Although most GFP⁺FLK1⁺ cells no longer expressed FLK1 after overnight culture, a distinct subset of GFP^-FLK1⁺ cells was seen,indicating that some GFP^-FLK1⁺ cells derived from GFP⁺FLK1⁺ precursors. This experiment also showed that∼70% of recultured GFP^-FLK1⁺ cells lost FLK1 expression by day 5.

These data demonstrated that GFP⁺ and FLK1⁺ cells were only generated between d2.8 and day 4 of ES cell differentiation, the same time frame during which BL-CFCs develop from FLK1⁺ precursors(Chung et al., 2002; Fehling et al., 2003). Although most of the BL-CFCs were cultured from GFP⁺FLK1⁺ cells in day 4 EBs, the considerable capacity of GFP^-FLK1^- and GFP⁺FLK1^- cells at earlier time points to differentiate into FLK1⁺ cells provided an explanation for our ability to derive VEGF-dependent BL-CFCs from all day 3 EB fractions (Fig. 7E) and reconciled our findings with data of others who had shown that BC-CFCs expressed FLK1 on their cell surface (Chung et al., 2002; Fehling et al.,2003).

Studies in Xenopus laevis demonstrated that Mix-family genes were directly induced by both activin and BMP4(Mead et al., 1996; Rosa, 1989; Vize, 1996). Therefore, we examined the ability of these factors to induce GFP expression in Mixl1-heterozygous and Mixl1-null ES cells. ES cells differentiated in SF medium (Johansson and Wiles, 1995) in the absence of growth factor supplements failed to express GFP or FLK1, but lost expression of the stem cell marker E-cad(Fig. 9C,E and data not shown),consistent with a default to neurectodermal differentiation(Wiles and Johansson,1999).

Supplementation of the SF medium with activin A at increasing concentrations resulted in a small, dose-dependent increase in the percentage of GFP-expressing cells in day 4 EBs (Fig. 9A). By contrast, BMP4 was far more potent at inducing GFP expression. More than 50% of Mixl1^GFP/w EB cells exposed to 1 ng/ml BMP4 were GFP⁺ at day 4 compared with a maximum of 10%GFP⁺ cells seen when EBs were cultured in the presence of 100 ng/ml activin A (Fig. 9A,B).

Comparison of GFP induction in Mixl1^GFP/w cells in serum containing and SF media revealed that the appearance of GFP-expressing cells was delayed in SF media supplemented with BMP4 (SF/BMP4) and that the peak percentage of GFP⁺ cells was reduced(Fig. 9C). Interestingly,supplementation of SF medium with BMP4 significantly increased the viability of cells at day 4 and 5 of differentiation(Fig. 9D). Moreover, the augmentation of cell survival in SF medium by BMP4 occurred prior to the induction of GFP expression (data not shown).

Induction of GFP expression was examined in Mixl1^GFP/wand Mixl1^GFP/GFP EBs cultured in SF medium in the absence or presence of BMP4 (Fig. 9E). GFP expression was seen in 46% of Mixl1^GFP/w and 55% of Mixl1^GFP/GFP cells in day 4 EBs cultured in SF/BMP4. FLK1 expression was reduced and delayed in the Mixl1^GFP/GFPcells, as observed in serum containing medium. In contrast to the results for serum-containing cultures in which the highest percentage of FLK1⁺cells coincided with maximal GFP expression(Fig. 5A), the frequency of FLK1⁺ cells in SF/BMP4 medium continued to increase for at least 3 days after the peak in GFP expression (Fig. 9E). In day 4 EBs, clonogenic haematopoietic progenitor cells were∼30-fold less frequent in SF/BMP4 than in serum differentiated EBs,further indicating that the addition of 5 ng/ml BMP4 was insufficient to completely replicate the effects of serum (compare Fig. 6B with Fig. 9F). One tenth the number of CFCs were observed in the Mixl-null cell lines, consistent with the results obtained from experiments using serum containing medium(Fig. 6B). Despite the observation that, by day 7, Mixl1^GFP/w and Mixl1^GFP/GFP EBs contained a similar percentage of FLK1⁺ cells, the frequency of CFCs was still lower in EBs derived from Mixl1-null cells (Fig. 9F). These data establish a link between activin A and BMP4 signalling and Mixl1 expression in mammalian cells. Furthermore,analysis of Mixl1-null cells in SF medium demonstrates that Mixl1 is required for normal BMP4-dependent expression of Flk1 and development of haematopoietic CFCs.

In accordance with the results of others(Keller et al., 1993; Lacaud et al., 2002; Robertson et al., 2000), we found that EB formation was characterised by the sequential expression of genes associated with the initial developmental stages of postimplantation embryos. The expression of stem cell genes was gradually lost as primitive streak genes, such as Mixl1, brachyury and goosecoid, were induced. The restricted expression of these genes between day 3 and day 4 indicated that the ES cells underwent a process of `molecular gastrulation', during which genes normally expressed during embryonic gastrulation were transiently expressed. This was followed by the upregulation of genes associated with ventral mesoderm and haematopoiesis. The kinetics of GFP induction in differentiating Mixl1<sup>GFP/w</sup> EBs was consistent with the PCR gene expression data. There was a higher proportion of Mixl1-expressing cells in EBs (85% at day 4) compared with an embryo at the equivalent developmental stage(Hart et al., 2002). In fact,studies by Keller and colleagues (Fehling et al., 2003) demonstrated that a similarly high percentage of differentiating ES cells expressed another primitive streak gene, brachyury. Also consistent with the results of others(Chung et al., 2002; Kabrun et al., 1997; Nishikawa et al., 1998), we observed that a large percentage of differentiating ES cells (up to 70%)expressed the ventral mesoderm marker Flk1, consistent with the majority of primitive streak cells being directed towards the haematopoietic lineages. These data indicate that, in serum containing cultures, the differentiation of mouse ES cells is biased towards the formation of`primitive streak' cells and, subsequently, ventral mesoderm.

The appearance of GFP⁺ cells in Mixl1-null EBs mirrored that seen in EBs from the Mixl1-heterozygous line, consistent with the assumption that `pre gastrulation' events would be unchanged in the absence of Mixl1. The higher peak percentage of GFP⁺ cells and the prolongation of GFP expression seen in Mixl1-null cells could be explained by the presence of two GFP alleles. However, an alternative hypothesis is that these features represented an in vitro correlate of the expanded primitive streak and delayed egress of GFP⁺ nascent mesendodermal cells observed in Mixl1-null embryos(Hart et al., 2002). Definitive exclusion of this scenario will require analysis of Mixl1-null ES cells that carry only one GFP allele.

In order to place Mixl1 expression during ES cell differentiation into a developmental context, flow cytometry was used to correlate the expression of GFP in Mixl1^GFP/w EBs with the expression of E-cadherin, a cell-adhesion molecule expressed in the epiblast of the mouse embryo (Burdsal et al., 1993). Experiments by Ciruna and Rosssant showed that expression of E-cadherin was downregulated as cells passed through the primitive streak and was lost by the time they underwent the epithelial mesenchymal transition required for migration from the streak (Ciruna and Rossant, 2001). Therefore, cells coexpressing E-cadherin and Mixl1 during EB differentiation could be regarded as a primitive streak like population. Indeed, flow cytometric analysis of E-cadherin and GFP during differentiation of Mixl1^GFP/w ES cells allowed the identification of cells reminiscent of the embryonic stages of epiblast(GFP^-E-cad⁺), primitive streak(GFP⁺E-cad⁺) and nascent mesoderm(GFP⁺E-cad^-). By day 4 of ES differentiation, most GFP⁺ cells expressed the haematopoietic marker Flk1 and only ∼20% of cells were still E-cad⁺. In fact, as reported by Nishikawa et al. (Nishikawa et al.,1998), we found that FLK1⁺ and E-cad⁺populations were mutually exclusive (data not shown).

In heterozygous Mixl1^GFP/w ES cells, Flk1 was expressed immediately following GFP and both genes were maximally expressed in day 4 EBs. In Mixl1-null cells, the onset of Flk1 expression was delayed and, even though the frequency of FLK1⁺ cells increased by day 4, normal haematopoiesis was not established, irrespective of whether the cells were differentiated in serum-based or SF medium (compare Fig. 5 with Fig. 9). This was most obviously evidenced by the failure of visible haemoglobinisation of day 7 Mixl1-null EBs. Also, the large reduction in the frequency of blast and erythroid colonies generated from day 4 Mixl1^GFP/GFPEBs suggested that the haematopoietic defect could not simply be explained by the lower percentage of FLK1⁺ cells at that time. This conclusion is consistent with analysis of Flk1-null ES cells showing that Flk1 was not necessary for the development of haematopoietic cells in vitro (Ema et al., 2003; Schuh et al., 1999). These data indicated that Mixl1 expression was required for the efficient generation of haematopoietic cells and placed Mixl1 functionally upstream of a FLK1⁺ haematopoietic precursor. Interestingly, recent studies showed that the haematopoietic defect associated with deficiency of the transcription factor Scl could only be rescued by expressing Scl by day 3 of ES cell differentiation, contemporaneous with the expression of Mixl1 but antedating expression of Flk1(Endoh et al., 2002). Collectively, these data predict that the specification of haematopoietic precursors occurs early in gastrulation and requires the input of a number of transcription factors, including Mixl1 and Scl.

The consequences of Mixl1-deficiency were reminiscent of the phenotype observed in the absence of Fgfr1 or Fgf8, the predominant FGF family member expressed in the gastrulating embryo. Specifically, embryos lacking Fgfr1 exhibited an enlarged primitive streak due to a failure of progenitor cell migration(Ciruna and Rossant, 2001; Ciruna et al., 1997; Deng et al., 1994; Yamaguchi et al., 1994) and Fgfr1-deficient ES cells displayed a marked impairment in haematopoietic colony formation in vitro(Faloon et al., 2000). Similarly, Fgf8-null embryos also displayed a cell migration defect in the primitive streak and showed evidence of perturbed haematopoiesis, with reduced expression of the erythroid and endothelial markers Fog and PECAM in the yolk sac (Sun et al.,1999). Collectively, these data speak to the requirement for both Mixl1 expression and FGF signalling in normal streak morphogenesis and haematopoietic specification. As the expression patterns of Fgf8and Mixl1 overlapped both in the primitive streak(Crossley and Martin, 1995; Pearce and Evans, 1999; Robb et al., 2000) and in differentiating EBs (C. Hirst, A. Mossman and A.G.E., unpublished), we are investigating the possibility of a direct link between these two molecules.

Primitive erythroid and endothelial cells arise from a common mesodermal precursor, the haemangioblast, most convincingly identified as a transient FLK1⁺ population in day 2.75-4.00 EBs(Choi et al., 1998; Chung et al., 2002; Fehling et al., 2003; Kennedy et al., 1997; Nishikawa et al., 1998). Furthermore, Fehling et al. (Fehling et al., 2003) showed that in day 3.5 EBs, haemangioblasts arose exclusively from brachyury-positive precursors and that brachyury⁺FLK1⁺ cells from day 2.5 EBs were enriched for BL-CFCs. Given the overlapping expression of brachyury and Mixl1 in differentiating ES cells (Kubo et al.,2004) (this study) and during embryogenesis, we anticipated that haemangioblasts would arise from a Mixl1-positive precursor and be enriched in GFP⁺FLK1⁺ cells from Mixl1^GFP/w EBs. Indeed, when cells were sorted from day 4 Mixl1^GFP/w EBs, most BL-CFCs were found in the GFP⁺FLK1⁺ fraction. Interestingly, some BL-CFCs were detected in the GFP^-FLK1⁺ population from day 4 EBs. This might have represented an artefact of the sorting process, with some GFP-dull cells inadvertently included in the GFP^-FLK1⁺fraction. Alternatively, BL-CFCs did not express Mixl1 and might have expressed alternate `gastrulation' genes such as brachyury. Indeed, in ES cells in which GFP was targeted to the brachyury locus, essentially all the FLK1⁺ cells were GFP⁺(Fehling et al., 2003).

The frequency of BL-CFCs was much higher in cells sorted from day 3 EBs but there was not a statistically significant enrichment in the GFP⁺FLK1⁺ population. Because overnight culture of sorted cells demonstrated a narrow window between day 2.8 and day 4, during which GFP^-FLK1^- and GFP⁺FLK1^-cells could continue to differentiate, cells that fell into the either of these populations at the time of sorting at day 3 could have upregulated Mixl1 and/or Flk1 expression after seeding in methylcellulose and produced VEGF-dependent blast cell colonies. However, in other studies that compared the BL-CFC content of FLK1⁺ and FLK1^- populations sorted from day 3 EBs, a clearer enrichment of BL-CFC in the FLK1⁺ cells was observed(Chung et al., 2002; Faloon et al., 2000; Fehling et al., 2003). It is possible that differences in the kinetics of FLK1 expression between the various ES cell lines used or differences in the ability of the methylcellulose cultures to support further differentiation may explain this discrepancy.

We used a SF culture system in order to determine which growth factors regulate `molecular gastrulation' and subsequent haematopoietic commitment in differentiating ES cells. In this context, the Mixl1^GFP/wES cells provided a simple means to identify live cells at the primitive streak stage by fluorescence microscopy and by flow cytometry. Previous studies in Xenopus have shown that Mix.1, the closest homologue to mammalian Mixl1, was induced by either activin A or BMP4(Mead et al., 1996; Rosa, 1989). In fact, more recently, Kubo et al. (Kubo et al.,2004) used a two-step SF culture system to show that Mixl1 RNA could be induced by activin A. In our culture system,although both activin and BMP4 induced GFP with similar kinetics in Mixl1^GFP/w EBs, the percentage of GFP⁺ cells was much greater in BMP4-treated cultures. The modest expression of Mixl1 in response to activin A contrasted with the robust induction of brachyury GFP by this factor reported by Keller and colleagues(Fehling et al., 2003). Indeed, in their original description of the SF medium used in our studies,Johansson and Wiles demonstrated that brachyury was more readily induced by activin than by BMP4 (Johansson and Wiles,1995). These data suggest that aspects of TGFβ signalling may be conserved between mouse ES cells and Xenopus embryos, in which activin induced both brachyury and Mix.1, and BMP4 treatment of animal pole cells induced Mix.1(Cunliffe and Smith, 1992; Mead et al., 1996; Smith et al., 1991).

The addition of BMP4 to ES cells differentiated in SF cultures led to an increase in total cell numbers and an improvement in cell viability that antedated Mixl1 expression (Fig. 9D) (E.S.N. and A.G.E., unpublished), corresponding to the growth-promoting effect of BMP4 on epiblast cells prior to gastrulation(Beppu et al., 2000; Mishina et al., 1995; Winnier et al., 1995).

Park et al. (Park et al.,2004) recently described Flk1 induction by BMP4 in a different SF medium. However, because brachyury was expressed in their EBs prior to the addition BMP4, it was unclear whether BMP4 induced mesoderm or simply acted to ventrally pattern existing mesoderm. We also observed the emergence of Flk1⁺ ventrally patterned mesoderm in Mixl1^GFP/w EBs cultured in SF/BMP4 medium. Our data showed unambiguously that induction of the primitive streak marker Mixl1 was completely dependent on the inclusion of BMP4 in the culture medium,strengthening the link between BMP4 and mesoderm induction in mammalian cells suggested by Johansson and Wiles(Johansson and Wiles, 1995; Wiles and Johansson, 1997). Nevertheless, as others have shown (Park et al., 2004), additional growth factors, such as VEGF, are required to efficiently generate haematopoietic CFCs from BMP4-induced FLK1⁺ ventral mesoderm.

In conclusion, we have shown that differentiating ES cells express Mixl1 as they transit through a stage equivalent to the primitive streak of the gastrulating mouse embryo. The most primitive haematopoietic BL-CFCs arise from Mixl1-expressing cells and absence of Mixl1 disrupts normal haematopoiesis. We have demonstrated that BMP4 augments survival, induces Mixl1 expression and ventrally patterns mesoderm in differentiating EBs. Finally, the Mixl1^GFP/wES cell lines will be valuable tools in further elucidating the factors that regulate mesoderm and endoderm formation.

We thank Dr Suzanne Micallef and Mary Janes for their valuable contribution to the establishment of SF differentiation culture conditions, and Dr Frank Battye and the staff of the WEHI flow cytometry laboratory for expert technical assistance. This work was supported by the Australian Stem Cell Centre, the Juvenile Diabetes Research Foundation, the National Health and Medical Research Council of Australia, and ES Cell International. L.R. and A.G.E. are NHMRC Senior Research Fellows.