Characterization of mesendoderm: a diverging point of the definitive endoderm and mesoderm in embryonic stem cell differentiation culture

Existence of mesendoderm that can give rise to both endoderm and mesoderm has been reported from C. elegans to Xenopus(Rodaway and Patient, 2001). In early zebrafish development, the marginal zone bordering on the vegetal margin contains precursors for endoderm as well as mesoderm. Both brachyury and Gata5, which are specific markers for mesoderm and endoderm precursors,respectively, are co-expressed in this marginal zone(Rodaway et al., 1999). The combination of fate maps and gene expression analyses indicates that endoderm and mesoderm are derived, at least in part, from the bi-potent mesendodermal population (Kimelman and Griffin,2000).

During early mouse development, endoderm precursors initially arise from the anterior primitive streak, which corresponds to early and mid gastrula organizer (EGO and MGO) (Kinder et al.,2001; Lawson et al.,1991; Wells and Melton,1999). Fate analyses by the transplantation of these organizers(Kinder et al., 2001) and fate maps defined by cell labelling experiments(Lawson et al., 1991) revealed that cells in two organizers subsequently gave rise to anterior definitive endoderm and axial mesoderm, including prechordal and notochordal plates. Although the results from tissue fate analyses strongly suggested that the organizer contains both endoderm and mesoderm precursors, the presence and the differentiation course of bi-potent mesendoderm are still elusive because of the difficulty in isolating a sufficient number of cells for further analyses. As mesendoderm is implicated as the major source of anterior endoderm in mammalian development, a part of which can eventually differentiate into hepatocytes and pancreatic β-cells(Wells and Melton, 1999), its characterization is essential for elucidating the differentiation pathway leading to anterior definitive endoderm and also for developing methods to induce endodermal cells for regenerative medicine.

In mouse development, endoderms are divided into two types: one is visceral endoderm, which diverges directly from the inner cell mass and gives rise to extra-embryonic endoderm; and the other is definitive endoderm(Lu et al., 2001). It is still difficult to judge whether endodermal cells generated in ES cell cultures represent definitive or visceral endoderm, owing to lack of molecular markers distinguishing two endodermal lineages(Grapin-Botton and Melton,2000; Tam et al.,2003). This deficit of molecular markers may be circumvented if the history of the resulting endodermal cells can be defined, as anterior definitive endoderm is known to be derived from endoderm precursors present in EGO and MGO (Kinder et al.,2001). Recently, Kubo and his colleagues used brachyury (T) as a marker for primitive streak mesoderm and showed that T-GFP-expressing cells in ES cell cultures could give rise to cells expressing endodermal markers(Kubo et al., 2004). Given that visceral endoderm is not derived from T⁺ primitive streak,this is the first unequivocal demonstration that definitive endoderm cells can be generated in ES cell cultures. As T is one of early markers for mesoderm,this suggests that endoderm cells are derived from mesoderm expressing T. In addition, because T is expressed in both node and early mesodermal lineages(Showell et al., 2004), cells marked by T cannot distinguish mesendoderm from other mesoderm cells. Thus, it is still necessary to establish new markers that enable us to specifically track the differentiation course of mesendoderm in an in vitro ES cell culture.

The goosecoid gene (Gsc) is an ideal marker for mesendoderm, as it is expressed specifically in the organizer region from which definitive endoderm arises (Blum et al.,1992). In addition, Gsc-null mice show no obvious abnormality during gastrulation (Yamada et al., 1995), indicating that insertion of a marker gene into a Gsc allele may not affect the differentiation process.

The central aim of this study is to identify and characterize bi-potent mesendoderm and its differentiation process in in vitro ES cell differentiation system. To achieve this, we have established ES cell lines that contain a Gsc allele to which an enhanced green fluorescence protein (gfp) gene (Zhang et al.,1996) is knocked-in by homologous recombination. We demonstrate that a defined culture condition allows us to induce almost pure Gsc⁺ population. This study shows that bi-potent mesendoderm cells do exist in our ES cell differentiation culture and are defined as a Gsc⁺E-cadherin(ECD)⁺PDGFRα(αR)⁺population that subsequently diverges to Gsc⁺ECD⁺αR^- and Gsc⁺ECD^-αR⁺ intermediates that eventually differentiate into definitive endoderm and mesodermal lineages,respectively.

The 5′ and 3′ arms of the targeting construct homologous to Gsc sequences were isolated from the 129 mouse strain(Yamada et al., 1995). The gfp gene was fused in-frame with the ATG start codon. A PGK-Neo cassette flanked by loxP sites was inserted downstream of the gfpgene (Fig. 1A). After electroporation into EB5 ES cells, 150 G418 resistant clones were selected. Three correctly targeted clones were confirmed by Southern blot analysis with an upstream genomic probe (HindIII-EcoRI, 1.2 kb fragment). HindIII-digested genomic DNA generated the wild type 6.6 kb and targeted 5.5 kb fragments. Transient transfection with a Cre recombinase expression vector was performed to remove the PGK-Neo cassette. The absence of PGK-Neo cassette was confirmed by Southern blot analysis using the neo cassette as probe (Fig. 1A).

EB5 (a kind gift from Dr Hitoshi Niwa, RIKEN, Kobe, Japan) is a subline derived from E14tg2a ES cells. This line was generated by targeted integration of an Oct3/4-IRES-BSD-pA vector into the Oct3/4 allele and carries the blasticidin S-resistant selection marker gene driven by the Oct3/4 promoter, which is active under the undifferentiated status(Niwa et al., 2000). Undifferentiated ES cells were maintained on gelatin-coated dishes in Glasgow minimum essential medium (G-MEM; GIBCO-BRL) supplemented with 1% foetal calf serum (FCS), 10% knockout serum replacement (KSR; GIBCO-BRL), 0.1 mM nonessential amino acids, 1 mM sodium pyruvate (GIBCO-BRL), 0.1 mM 2-mercaptoethanol (2ME), 1000 U/ml leukemia inhibitory factor (LIF)(GIBCO-BRL) and 20 μg/ml blasticidin S to eliminate differentiated cells. OP9 stromal cells were maintained in α minimum essential medium (MEM)(GIBCO-BRL) supplemented with 20% FCS (Era and Witte, 2000).

Under serum-containing condition, induction of ES cell differentiation was carried out as described previously(Nishikawa et al., 1998). Briefly, undifferentiated ES cells were transferred to type IV collagen-coated dishes (BioCoat; Becton Dickinson Labware) or non-coated dishes and incubated in α MEM supplemented with 10% FCS and 50 μM 2ME in the absence of LIF.

For the induction under serum-free condition, ES cells were seeded onto type IV collagen-coated 10 cm dishes at a density of 1×10⁵cells per dish in SF-O3 medium (SFM) (Sanko Junyaku)(Takakura et al., 1996a)supplemented with 0.1% bovine serum albumin (BSA) and 50 μM 2ME. For the formation of EB, ES cells were seeded on non-coated 6 cm petri dishes at a density of 3×10⁴ cells per dish. In some experiments,recombinant human activin A, BMP4 and recombinant mouse Nodal (R&D Systems) were added from the beginning of the culture: activin A 10 ng/ml,Bmp4 10 ng/ml and nodal 1000 ng/ml.

For re-culture experiments, Gsc⁺ECD⁺ and Gsc⁺ECD^low cells in Gsc^gfp/+ES cell culture or ECD⁺αR⁺ cells in unmanipulated EB5 ES cell culture were sorted at day 4 and re-cultured in SFM with activin A on type IV collagen-coated dishes. For the induction to albumin-producing cells,day 6 Gsc⁺ECD⁺ cells were sorted and re-cultured for 3 days on type I collagen-coated dishes (BioCoat; BD Labware) in SFM with 20 ng/ml Egf (R&D Systems), 20 ng/ml Bmp4, 10 ng/ml acid Fgf (R&D Systems) and 5 ng/ml basic Fgf (R&D Systems)(Jung et al., 1999; Zaret, 2002). The Gsc⁺ECD⁺-derived cells were stained by anti-albumin antibody. The expression of endoderm, hepatic and pancreatic markers was analyzed by RT-PCR.

Gsc⁺ECD^- cells were sorted at day 6 and used for the analyses of gene expression and cell fate. For osteogenesis, sorted cells were re-cultured on gelatin-coated dish in KnockOut D-MEM (GIBCO) supplemented with 10% FCS, 0.1 μM dexamethasone, 50 μM ascorbic-acid-2-phosphate (Sigma),10 mM β-glycerophosphate (Sigma) and 10 ng/ml Bmp4(Pittenger et al., 1999). On day 28, Gsc⁺ECD^--derived cells were harvested and stained by Alizarin Red (Muraglia et al.,2003). In addition, total RNA was isolated to examine the expression of bone-specific markers. For vasculogenesis, 500 Gsc⁺ECD^- cells were re-cultured onto a confluent OP9 cell layer in each well of 24-well plate with α MEM containing 10% FCS and 50 μM 2ME (Hirai et al.,2003). Two days later, the number of endothelial colonies was enumerated after staining wells with either anti-VE-cadherin or anti-Pecam1 antibodies.

Gsc⁺ECD⁺ cells were sorted at day 6 and re-cultured in α MEM supplemented with 10% FCS and 50 μM 2ME on type I collagen-coated dishes. The passage of cells was performed every 3 days.

Preparation of the rat monoclonal Abs APA5 (anti-PDGFRα)(Takakura et al., 1996b),ECCD2 (anti-ECD) (Shirayoshi et al.,1986), AVAS12 (anti-Vegfr2)(Kataoka et al., 1997) and APB5 (anti-PDGFRβ) (Sano et al.,2001), and their labelling by either biotin or allophycocyanin were carried out as described previously(Nishikawa et al., 1998). Cultured cells were harvested with cell dissociation buffer (GIBCO-BRL). Cell staining, analysis by FACS Calibur and cell sorting by FACS Vantage or FACS Aria (Becton Dickinson) were as previously described(Nishikawa et al., 1998).

Total RNA was prepared from sorted cells or cultured cells using TRIZOL Reagent (Invitrogen) according to manufacturer's protocol. First-strand cDNA was prepared using Superscript First-Strand Synthesis System for RT-PCR. The PCR cycling conditions were as follows:1 cycle of 94°C for 4 minutes; 30 cycles of 94°C for 30 seconds, 55°C for 30 seconds, 72°C for 1 minute; and 1 cycle of 72°C for 7 minutes. β-Actin was used as the invariant control. The sequences of primers used are as follows (forward and reverse): Snail, CTCCACAAGCACCAAGAGTCTG and TCCAGTAACCACCCTGCTGAG; Brachyury,AACTTTCCTCCATGTGCTGAGAC and TGACTTCCCAACACAAAAAGCT; Sox17,TTTGTGTATAAGCCCGAGATGG and AAGATTGAGAAAACACGCATGAC; Hex,AGACTCAGAAATACCTCTCCCCA and TTTATCCCCCTCGATGTCCA; Mixl1,ACTTTCCAGCTCTTTCAAGAGCC and ATTGTGTACTCCCCAACTTTCCC; E-cadherin,AAACTTGGGGACAGCAACATCAG and TCTTTTGGTTTGCAGAGACAGGG; Goosecoid,ATGCTGCCCTACATGAACGT and CAGTCCTGGGCCTGTACATT; Cytokeratin 8,AGATGAACCGCAACATCAACC and TCAATCTTCTTCACAACCACAGC; Cytokeratin 18,GATTGACTGTGGAAGTGGATGC and GTTTGCATGGAGTTGCTGGA; Cytokeratin 19,TGCGCGACAAGATTCTTGGT and TGACTTCGGTCTTGCTTATCTGG; Prox1,TTCAACAGATGCATTACCTCGC and TGACATCACAACATATCCATGCC; Afp,TGCAGAAACACATCGAGGAGAG and GCTTCACCAGGTTAATGAGAAGCT; Gata6,AGACATAACATTCCTTCGATGCG and TTCCAAGTGACCTCAGATCAGC; Pdgfra,AATCCTGCAGACGAGAGCAC and GCCACCAAGGGAAAAGATTT; Gata4,CGAGATGGGACGGGACACT and CTCACCCTCGGCCATTACGA; β-Actin,CCTAAGGCAACCGTGAAAAG and TCTTCATGGTGCTAGGAGCCA; Pdgfrb,GTCTGGTCTTTTGGGATCCT and AAGGCTGGTTACAGTTTGGC; Cadherin 11,TCAGGGAACATTCATGCCAC and TTCTATGCCGTCTCCATCAAC; Vegfr2,AAGGAACTAGAATGCGGGCT and ACTCCCTGCTTTTACTGGGC; Claudin 6, ACAAAGCTGACCGAGCACT and AGCAGCAAAAGGCCTGAG; Foxa2, TGGCTGCAGACACTTCCTACT and CAACATCAGTACAACCCTCTGGT; Lim1, GGTTGTCTTCAGAAGTCATCCC and AATCCGGAGATAAACTAGGGTCAC; Chordin, TGGTTCCCAGAGAATCAGAGCT and CCAAAAGAAAACAGGGCAGG; Nodal, CCGTCCCCTCTGGCGTACATG and GACCTGAGAAGGAATGACGG;Cerberus l, GGAAGAAACCTGAGACCGAAT and AGTCCAGGGATGAAGGAACC; Hnf6,AGATCAATACCAAAGAGGTGGCG and TTTGGTACAAGTGCTCGATGAGG; Albumin,CCACTGTTGAAGAAAGCCCA and CAGATAGTCTTCCACACAAGGCA; Pdx1,CCGGACATCTCCCCATACGAA and GAGGTCACCGCACAATCTTGC; Osteopontin,GCTGTGTCCTCTGAAGAAAAGG and AGACTTGGTTCATCCAGCTGAC; Bglap1,TCTGACCTCACAGATGCCAA and ACATCCATACTTGCAGGGCA; Bglap2,AGCAGGAGGGCAATAAGGTAGT and GCTGCTGTGACATCCATACTTG; Hnf4,TTTGATCCAGATGCCAAGGG and TTGCTTGGTGATCGTTGGCT; Sox7,TCATGTCAGGAGGAGCATGG and CAGGACTGAGATGAGGCTGGT; Pthr1,AGATGCTCTTCAACTCCTTCCA and GTGACACCATTGGGCCATAG.

For immunohistochemistry, cultured cells were fixed with 2%paraformaldehyde in PBS. The following antibodies were used: ECCD2, 1:1000;AVAS12, 1:1000; APB5, 1:1000; anti-HNF3β (Foxa2) (goat polyclonal) (Santa Cruz), 1:100; anti-Gata4 (goat polyclonal) (Santa Cruz), 1:500; anti-brachyury(goat polyclonal) (Santa Cruz), 1:250; anti-Pecam1 (rat monoclonal) (BD Pharmingen), 1:500; VECD1 (anti-VE-Cadherin) (rat monoclonal)(Matsuyoshi et al., 1997),1:500; anti-albumin (goat polyclonal) (Bethyl), 1:500; anti-cytokeratin 18(mouse monoclonal) (Progen), 1:200; anti-GFP (rabbit polyclonal) (Molecular Probes), 1:500. Appropriate HRP-conjugated secondary antibodies were selected,and the signals were detected by DAB-Ni(Takakura et al., 1996b). For immunofluorescence staining, samples were stained by appropriate fluorescence-tagged secondary antibodies and examined on Radiance2100 (BioRad)confocal imaging system.

Tetraploid embryo complementation was performed as described(Nagy et al., 1993; Tam and Rossant, 2003), using C57BL/6 zygotes. Briefly, two-cell stage embryos were electrofused and developed in vitro to 4N blastocyst stage. Gsc^gfp/+ ES cells were injected into blastocysts and transferred to pseudopregnant ICR females. The expression of GFP was examined in mid-streak stage embryos.

Single Gsc⁺ECD⁺ cell in day 4 culture was seeded into individual wells of 96-well collagen IV-coated plates (Becton Dickinson) by FACS Vantage or Aria equipped with single cell deposition device. Sorted single cell was cultured in serum-containing medium supplemented with 50 ng/ml Activin A, 5 ng/ml basic Fgf and 1 mM LiCl for 3 days. Cells were stained by an antibody cocktail of anti-Foxa2 Ab, APB5 and AVAS12. HRP-conjugated anti-goat IgG (Sigma) and ALP-conjugated anti-rat IgG (Jackson ImmunoResearch)were used as secondary antibodies. The substrate used was DAB-Ni in HRP staining. ALP staining was performed by the Vector Red substrate kit I(VECTOR), according to manufacturer's protocol.

Endoderm cell line was maintained in α MEM media supplemented 10% FBS and 2ME. Cells of this cell line (1×10⁶) were injected into either the renal capsule or the spleen of scid/scid immunodeficient mice. Every week from 2 to 5 weeks after transplantation, mice were sacrificed and the serial sections of transplanted tissues were investigated.

Fig. 1A illustrates our strategy to establish ES cell lines containing a Gsc-gfpknock-in allele (Gsc^gfp/+ ES) by homologous recombination. Three ES cell lines harbouring GFP gene were established and used for further experiments.

In order to examine whether or not Gsc^gfp/+ ES cell lines can be used for monitoring the differentiation of Gsc⁺ cells,we cultured Gsc^gfp/+ ES cells under conditions that have been used for inducing mesoderm cells from ES cells(Nishikawa et al., 1998). GFP⁺ cells were detected after 4 days of differentiation in serum-containing medium (SCM) either on collagen IV-coated dishes or using embryoid body (EB) formation (Burkert et al., 1991) (Fig. 1B, part i). Cell-sorting and RT-PCR analysis of GFP⁺and GFP^- cells demonstrated that GFP expression correlated with that of endogenous Gsc. Moreover, expression of other molecular markers of EGO and MGO, such as Foxa2(Monaghan et al., 1993), Lim1 (Shawlot and Behringer,1995) and chordin (Sasai et al., 1994), was detected only in GFP⁺ population, and expression of T in this population suggests the possibility that it is not derived from visceral endoderm(Showell et al., 2004; Tam et al., 2003; Wilkinson et al., 1990)(Fig. 1B, part ii). To check GFP expression in vivo, we injected Gsc^gfp/+ ES cells into blastocysts of tetraploid embryos (Nagy et al., 1993; Tam and Rossant,2003). In the mid-streak stage embryo, GFP expression was detectable not only in the gastrula organizer, but also in the newly formed mesoderm (Fig. 1B, part iii). The expression pattern is consistent with a previous study of Gsc^lacz expression in mouse embryo(Kinder et al., 2001). This result, together with gene expression pattern of Gsc⁺ cells in culture, indicates strongly that Gsc⁺ cells correspond to the organizer region of actual embryos.

Under the culture condition that we have been using for inducing mesoderm cells (Nishikawa et al.,1998), the proportion of GFP⁺ cells did not exceed 3%on day 4 (Fig. 1B, part i).

As activin was shown to induce the expression of Gsc in Xenopusanimal cap cells (Cho et al.,1991; Symes et al.,1994), we investigated the effect of activin. The differentiation of ES cells were induced on collagen IV-coated dishes in SF-O3 serum-free medium (SFM) with 10 ng/ml of activin. Activin acts as a potent inducer of Gsc under serum-free culture condition, with 60% and nearly 100% of cells being GFP⁺ by day 4 and day 6, respectively(Fig. 1C, part i). The expression analysis by RT-PCR showed that a set of molecules specific for the organizer, including Gsc, Foxa2, Lim1 and chordin, was exclusively expressed in GFP⁺ population and in the same population under serum-containing conditions (Fig. 1C, part ii). Though our ES cell line can differentiate into neuroectodermal cells by treatment with retinoic acid(Ying et al., 2003), the expression of neuroectodermal markers such as Pax6(Hill et al., 1992) and Sox1 (Wood and Episkopou,1999) was not induced in this culture system (data not shown). Next, we examined whether or not visceral endoderm marker expression was induced in this serum-free culture, as Gsc is known to be expressed also in some visceral endoderm cells (Perea-Gomez et al., 1999). The expression of visceral endoderm markers such as Sox7 (Kanai-Azuma et al.,2002), Hnf4 (Duncan et al., 1994) and Pthr1(Verheijen et al., 1999) was not induced in serum-free culture with activin(Fig. 1C, part iii). Moreover,we recently established a culture condition that permits only differentiation into visceral endoderm (M. Yasunaga and S.T., unpublished). Although Sox7,Hnf4, Pthr1, Pem (Tam et al.,2004) and Msg1(Dunwoodie et al., 1998) were expressed in cells induced under this selective condition for visceral endoderm, these markers were not in Gsc⁺ population induced under activin containing serum-free condition (M. Yasunaga and S.T., unpublished). These results, together with the result that T was expressed in both GFP⁺ and GFP^- population(Fig. 1C, part iv), strongly suggest that GFP⁺ population produced under this serum-free condition does not contain visceral endoderm.

Though somewhat lower in potency, nodal, which shares a common signal pathway with activin, was also effective in Gsc induction(Fig. 1D). Conversely, the addition of Bmp4 suppressed the differentiation of Gsc⁺ cells, even in the presence of activin (Fig. 1E). Although the serum-free culture condition allows us to induce pure Gsc⁺ cells by activin, it was difficult to obtain the same effect of activin in SCM or in EB cultures(Fig. 1F). This suggests that the exclusion of undefined factors such as sera and complex cellular architecture formed in EBs should be actively pursued to obtain useful steerage of in vitro differentiation of ES cells into Gsc⁺ cells. This serum-free condition (Fig. 1C) was used for inducing Gsc⁺ cells throughout this study.

As FACS sorting is an essential technology to isolate as well as to define cells in ES cell cultures, we searched for appropriate surface markers to describe differentiation of Gsc⁺ cells and found that E-cadherin(ECD) and PDGFRα (αR) are useful for this purpose. It was shown that ECD expression distinguishes exfoliated mesoderm cells from ectoderm and endoderm in mouse embryo (Huber et al.,1996). Moreover, the expression of αR was known to be induced by activin in dissociated Xenopus animal cap cells(Symes et al., 1994).

Gsc⁺ cells were induced from ES cells, and surface expression of ECD and αR was analyzed at various time points. Initially, most nascent GFP⁺ cells co-expressed ECD and αR, but diverged quickly to ECD⁺αR^- and ECD^-αR⁺populations (Fig. 2A,B). ECD^-αR⁺ cells were expressing other mesodermal lineage markers such as PDGFRβ (βR)(Betsholtz et al., 2001) and Vegfr2 (Kataoka et al., 1997)(Fig. 2C).

To confirm the sequence of events occurring during the differentiation of Gsc⁺ECD⁺ cells, we purified GFP⁺ECD^high, GFP⁺ECD^low and GFP^- cells from day 4 cultures and re-cultured them. In contrast to the Gsc⁺ECD^high population that could give rise to both Gsc⁺ECD⁺ and Gsc⁺ECD^- cells, the Gsc⁺ECD^low fraction could generate only ECD^low/- cells. (Fig. 2D). The Gsc^- population contained immature cells that could differentiate into both Gsc⁺ECD⁺ and Gsc⁺ECD^- populations, but such immature cells disappeared by day 5 of the culture (data not shown). This re-culture experiment in combination with our analysis of αR expression indicates an order of differentiation in which Gsc^-ECD⁺ immature cells give rise to Gsc⁺ECD⁺αR⁺ cells that subsequently diverge to Gsc⁺ECD⁺αR^- and Gsc⁺ECD^-αR⁺ cells.

We next characterized in detail both ECD⁺ and ECD^-populations on day 5 by RT-PCR and immunohistochemical staining. RT-PCR analysis showed that ECD⁺ population expressed claudin 6, which is involved in tight junction of epithelium(Sousa-Nunes et al., 2003),and other endodermal markers such as Foxa2(Ang et al., 1993) and Sox17 (Kanai-Azuma et al.,2002; Tam et al.,2003) (Fig. 3A). By contrast, ECD^- cells were expressing a set of mesodermal markers,including βR, Cad11 (Kimura et al., 1995) and Vegfr2(Fig. 2C, Fig. 3A). Two morphologically different cell types were recognized in day 5 cultures: one composed of epithelial sheets and the other containing dispersed cells(Fig. 3B). Co-existence of Gsc⁺ECD⁺Foxa2⁺ epithelial-like cells and Gsc⁺ECD^-Foxa2^- non-epithelial-like cells in the same culture was confirmed by fluorescent immunohistochemistry(Fig. 3B).

The expression of Sox17 and the cytokeratins Krt2-8,cytokeratin 18 and Krt119 (Owens and Lane, 2003) was maintained during further culture of Gsc⁺ECD⁺ population in SCM, whereas the expression of Gsc and Mixl1, which is one of mesendoderm markers(Hart et al., 2002), was successively downregulated and α-fetoprotein (Afp)(Koike and Shiojiri, 1996) was newly expressed (Fig. 3C). To further characterize Gsc⁺ECD⁺ cells in day 6 cultures,we attempted to induce albumin-producing hepatocytes and pancreaticβ-cells. The albumin-producing cells that were also expressing other hepatic markers such as Hnf6(Landry et al., 1997; Rausa et al., 1997) could be generated from Gsc⁺ECD⁺ cells(Fig. 3D). These results suggest that the Gsc⁺ECD⁺ population in our defined ES cell culture contains precursors that can give rise to more mature endodermal lineages. At present, however, we could not determine conditions for inducing other endoderm-derived cells, such as pancreatic cells.

We next examined whether or not day 6 Gsc⁺ECD⁺ cells can give rise to mesodermal lineages, such as osteocytes and endothelial cells, and have confirmed that this population has no potential to differentiate into mesoderm (data not shown). At this stage,Gsc⁺ECD⁺ cells are likely to be committed to endodermal lineages. The expression of endodermal markers together with the defined differentiation course of day 6 Gsc⁺ECD⁺ cells indicated that they should correspond to precursors of definitive endoderm.

To evaluate the differentiation potential of Gsc⁺ECD^-cells that diverge from nascent Gsc⁺ECD⁺ cells, day 6 Gsc⁺ECD^- cells were sorted and cultured under selective conditions for inducing mesodermal lineages such as osteoblasts and endothelial cells. This ECD^- population could give rise to osteocytes and endothelial cells, which are typical progenies of mesoderm(Fig. 4A). Though almost all of Gsc⁺ECD^- cells expressed Vegfr2, the efficiency of endothelial differentiation from Gsc⁺ECD^- population was one-tenth that in Vegfr2⁺ lateral mesoderm cells [reported in our previous study (Yamashita et al.,2000) (data not shown)].

The above data suggest that the first Gsc⁺ECD⁺population generated in ES cell differentiation culture represents presumptive mesendoderm with potential to give rise to both endoderm and mesoderm. This potential was vividly illustrated by immunofluorescent images of day 6.5 cultures in which Gsc⁺ECD⁺Foxa2⁺ cells were wedged between Gsc^-ECD⁺Foxa2⁺ endoderm cells and Gsc⁺ECD^-Foxa2^- mesoderm cells(Fig. 4B).

Though the experiments described above suggest that the day 4 Gsc⁺ECD⁺ population contains mesendodermal cells that can give rise to both endoderm and mesoderm, it is necessary to perform a clonal analysis in order to prove the multipotency of an individual cell. To obtain the efficient growth in single cell culture, we modified our culture conditions into that including 10% FCS, LiCl, activin and basic Fgf, although it is very difficult to judge whether these culture conditions are optimal for this clonogenic assay.

Gsc⁺ECD⁺ cells from day 4 cultures were seeded by single cell deposition. Following a 3-day incubation, all wells were stained with anti-Foxa2 and a mixture of anti-Vegfr2 and anti-βR mAbs to specify endoderm and mesoderm, respectively. The frequency of Gsc⁺ECD⁺ cells that undergo clonogenic growth was only about 0.5% under this condition (Table 1). From 64 wells in which we could observe more than three cells after staining, 45 contained only mesoderm cells and 19 contained both mesoderm and endoderm (Table 1; Fig. 4C). We could not detect any wells containing only endoderm cells. Co-existence of mesoderm cells may be required for the growth of endoderm. In conjunction with results of our mass culture of Gsc⁺ECD⁺ cells, this single cell assay demonstrates that bi-potent mesendodermal cells exist in our mouse ES cell culture.

We noticed that Gsc⁺ECD⁺ cells harvested from the day 6 culture showed a remarkable proliferative activity in SCM. Indeed, by attempting to establish cell lines, Gsc⁺ECD⁺ cells undergo sustained growth maintaining the initial phenotype(Fig. 5A,D). The resulting cells were homogeneous in morphology (Fig. 5B) and maintained the expression of a series of endodermal markers (Fig. 5C,D). We could maintain some of cultures by serial passage without change of phenotype for more than 6 months (data not shown). When these cells were transplanted into either renal capsules or spleens of scid/scid immunodeficient mice,no obvious tumour formation was observed. So far, in both renal capsules and in vitro cultures, we could not induce further maturation of endoderm cell lines.

As almost all of day 4 Gsc⁺ECD⁺ cells induced in our defined culture condition co-express αR,ECD⁺αR⁺ cells in day 4 culture theoretically correspond to Gsc⁺ECD⁺ population. In order to investigate this possibility, we cultured unmodified EB5 ES cells under our defined condition and ECD⁺αR⁺ cells were purified from day 4 culture (Fig. 6A). The sorted cells were further incubated for 2 days and ECD⁺(population B in Fig. 6A) cells were isolated for the analysis of endoderm marker expression and for the establishment of endoderm cell lines.

A set of endodermal markers, including claudin 6, Foxa2 and Sox17, was expressed in ECD⁺ population but not in ECD^- population, whereas mesoderm markers were expressed in ECD^- population (Fig. 6B). Moreover, we observed that Gsc was expressed in both ECD⁺ and ECD^- populations(Fig. 6B). These results indicate that ECD⁺ and ECD^- population derived from day 4 ECD⁺αR⁺ cells in genetically unmanipulated ES cell cultures correspond to ECD⁺ and ECD^- population in day 6 Gsc⁺ population, respectively. Moreover, ECD⁺endoderm cell lines have been established from unmanipulated ES cells by the same protocol as that used for Gsc^gfp/+ ES cells(Fig. 6C). These results suggest that the method developed in this study is powerful for the purification of mesendodermal cells in ES cell differentiation culture that can give rise to both endodermal and mesodermal cells.

In this study, we have developed in vitro ES cell differentiation system that enables us to identify and characterize mesendoderm cells with the combination of genetic marking by knock-in method. Using this system, we have demonstrated bi-potent mesendoderm is represented as a Gsc⁺ECD⁺αR⁺ population in ES cell differentiation culture. The presence of bi-potent mesendoderm in vivo, which can give rise to both endoderm and mesoderm, has been established in both C. elegans and zebrafish. Its presence has been implicated also in mouse embryo by experiments using cell labelling, cell transplantation and whole-embryo cultures (Kinder et al.,2001; Lawson et al.,1991). Our study has demonstrated the presence of Gsc⁺bi-potent mesendoderm in ES cell differentiation culture at a single cell level.

According to previous studies, Gsc is strongly expressed in EGO and MGO of gastrula stage embryos (Belo et al.,1998; Kinder et al.,2001). Gsc⁺ cells generated under our culture condition express a set of organizer specific genes such as Foxa2, Lim1 and Chordin, as well as T (a pan-mesoderm marker), and can give rise to both endodermal and mesodermal cells. Although Gsc is also known to be expressed in some visceral endoderm cells (Perea-Gomez et al.,1999), the expression of visceral markers such as Sox7and Hnf4 are not induced in our defined culture condition. These results suggest that ES cell-derived Gsc⁺ cells correspond to cells in EGO and MGO that give rise to anterior endoderm, cranial mesenchyme and somites (Kinder et al.,2001).

The defined condition together with surface markers exploited in this study allowed us to delineate the differentiation course of definitive endoderm via Gsc⁺ mesendoderm in vitro. Fig. 7 summarizes our understanding of the differentiation of mesendoderm from ES cells, which is the first detailed model of the differentiation course of mesendoderm. It is likely that mesendodermal cells in EGO are ECD⁺ (Huber et al.,1996), as it is a part of embryonic ectoderm. The expression of ECD in epiblasts disappears along with their differentiation to mesoderm after the exfoliation from the primitive streak. Thus, ECD is a useful marker for distinguishing mesoderm from other germ layers. In our study, mesendoderm cells diverge to Gsc⁺ECD⁺ endoderm and Gsc⁺ECD^- mesoderm precursors(Fig. 7). The Gsc⁺ECD⁺ endoderm precursors differentiate into definitive endoderm, whereas Gsc⁺ECD^- mesoderm precursors can give rise to mesodermal lineages. Our previous observations that both αR and Vegfr2 are expressed in anterior migrating mesoderm cells agree with the surface phenotype of Gsc⁺ mesoderm cells in culture (Kataoka et al.,1997). Thus, as is often the case with other cell processes such as haematopoietic differentiation (Nakano et al., 1996), the process of mesendodermal differentiation, even under such a selective culture condition, appears to correlate well with that in the actual embryo.

We have also demonstrated that Gsc⁺ECD^- population can give rise to endothelial cells, but the ability to generate them is much less than that of Vegfr2⁺ lateral mesoderm(Yamashita et al., 2000). Although endoderm derived from mesendoderm was shown to differentiate into the primitive gut in both C. elegans and zebrafish, the fate of mesoderm cells derived from mesendoderm is variable among species. In C. elegans, it has been shown that mesendoderm gives rise to pharynx and muscles (Maduro et al., 2001),whereas to heart and muscles in zebrafish(Warga and Nusslein-Volhard,1999). However, the fate of Gsc⁺ mesoderm in mice remains uncharacterized. In this study, we have revealed that Gsc⁺mesoderm cells have potential to differentiate into endothelial cells and bone cells in vitro. However, we could not induce heart cells or haematopoietic cells under conditions that support their differentiation from other cell sources (data not shown). Nonetheless, our observation that Gsc⁺cells representing cells in EGO and MGO, the most anterior end of primitive streak, could give rise to endothelial cells strongly suggests that endothelial cells can be differentiated from most regions of the primitive streak of mouse. This is consistent with cell labelling experiments in zebrafish showing that endothelial cells are generated from all areas of the blastoderm rim (Kimmel et al.,1990).

The fresh endoderm precursors defined by the expression of Gsc and ECD on day 6 could give rise to HNF6⁺ albumin-producing cells. However, we could not induce cells displaying pancreatic markers from Gsc⁺ECD⁺ population, although we applied usual pancreatic induction methods in which EB culture was used(Hori et al., 2002; Lumelsky et al., 2001). We speculate that unknown combination of growth factors or co-cultures with other lineage cells are required for the differentiation from highly purified Gsc⁺ECD⁺ cells to pancreatic cells. During long-term propagation in vitro, the potential of differentiation into albumin⁺ cells is lost and the culture is dominated by the cells expressing molecules specific for immature endodermal lineage, such as Foxa2,Sox17 and Gata4, but not markers for more mature stage, such as albumin and Pdx1. Loss of differentiation potential into mature endoderm in this cell line may result from other processes such as overdomination of a particular cell type. Our result shows for the first time that some of ES cell-derived endoderm cells bear an ability to undergo sustained proliferation in vitro. This ability is not due to the malignant transformation, because the tumour formations were not observed by transplantation into immunodeficient mice. To our disappointment, however, we have no evidence that these cell lines can undergo further differentiation into mature endodermal cells such as pancreatic β-cells. Further studies are necessary to determine whether endoderm cells from the actual embryo has a similar ability to undergo sustained proliferation or whether such a high proliferative activity is specific to ES cell-derived endoderm cells.

One of ultimate goals for in vitro ES cell differentiation is to prepare a sufficient number of pure cells by controlling ES cell differentiation. There are thought to be two ways to achieve this. One is to consider that ES cell differentiation requires the environments which are similar to those present in the actual embryo. The most typical example of such a method is EB culture system for ES cell differentiation (Burkert et al., 1991). The other is based on the idea that steering ES cell differentiation is attained only by the highly selective culture condition that excludes differentiation into unnecessary lineages. In this study, we sought tested the latter possibility, and have succeeded in determining a defined condition that can generate almost pure Gsc⁺population without using EB formation method. We have demonstrated that EB culture, even using the same defined culture condition, is less efficient in inducing Gsc⁺ cells than the two-dimensional (2D) culture on collagen IV-coated dishes. This result indicates an inherent limitation of EB system in guiding ES cell differentiation, as uncontrollable complexity is inevitably associated with three-dimensional architecture in EB.

We realize that what is taking place in ES cell differentiation culture under a defined condition does not necessarily recapitulate the differentiation process of the actual embryo. However, it is also clear that in vitro differentiation, even though induced under a highly artificial condition, proceeds under similar constraints to that in the differentiation process of the actual embryo. In this study, we have shown that nodal and activin can selectively induce ES cell-derived Gsc⁺ mesendoderm. This result is consistent with previous reports showing that nodal is an indispensable molecule for inducing the organizer in both Xenopus and mouse (Varlet et al., 1997; Zhou et al., 1993). Mutant analyses of nodal-related genes in zebrafish and Xenopus have indicated that they are required for the generation of both endoderm and mesoderm (Osada and Wright,1999; Rodaway et al.,1999). Moreover, it has been shown that Nodal signalling has an essential role in specification of anterior definitive endoderm and a part of mesoderm in mouse (Lowe et al.,2001; Vincent et al.,2003). Our observation that Activin/nodal signal can induce endoderm and mesoderm differentiation in ES cell culture is consistent with these in vivo studies. In the actual embryo, however, it has been difficult to specify at cellular level which differentiation process is regulated by nodal or nodal-related molecules. In this sense, our method of inducing pure Gsc⁺ population in ES cell cultures should complement previous in vivo studies and provide a clearer view on the process of mesendoderm differentiation.

In this study, we also aimed at developing a method to define mesendoderm without using genetically manipulated ES cells. We have shown here that mesendoderm is defined as ECD⁺αR⁺ when these ES cells are cultured under the selective culture condition containing activin. As mesendoderm is one of the precursors of definitive endoderm(Lu et al., 2001; Martinez Barbera et al.,2000), isolation and characterization of mesendodermal cells are essential steps for generating more mature cells of definitive endodermal cell lineage. We validated this method by establishing endoderm cell lines from genetically unmanipulated ES cells. Indeed, this experimental system allowed us to establish endoderm cell lines that are similar to those established from Gsc^gfp/+ ES cell lines. Moreover, pure mesendoderm population is also useful for screening defined culture conditions required for subsequent differentiation steps towards mature endoderm cells. The trials on this line are currently in progress.

The demonstration that activin can induce Gsc and αR in ES cell culture is consistent with previous studies that activin induced the expression of both Gsc and αR in Xenopus animal cap cells(Gurdon et al., 1995; Jones et al., 1993; Symes et al., 1994). This similarity between evolutionally distant species suggests that our defined culture system, which can induce pure Gsc⁺ mesendoderm from mouse ES cells, is also likely to be applicable to differentiation of human ES cells.

Supplementary material for this article is available at http://dev.biologists.org/cgi/content/full/132/19/4363/DC1

We thank Dr L. M. Jakt for critical reading of the manuscript and Dr G. Yamada for providing the genomic DNA of goosecoid. This work was supported by grants (No. 12219209 to N.S. and No. 16606005 to E.T.) from the Ministry of Education and Science of Japan and by a grant (to N.S.) from the project for realization of regenerative medicine.