Regulation of human embryonic stem cell differentiation by BMP-2 and its antagonist noggin

The development of human embryonic stem (ES) cells (Reubinoff et al., 2000; Thomson et al., 1998) has opened up exciting new opportunities for basic research and regenerative medicine (reviewed by Pera et al., 2000). To exploit the potential of human ES cells, it will be essential to understand the molecular control of their growth and differentiation. Under certain conditions in vitro, human ES cells differentiate spontaneously into a wide range of somatic and extra-embryonic cell types (Itskovitz-Eldor et al., 2000; Reubinoff et al., 2000). It is logical to assume that ES cell differentiation in vitro mimics in a chaotic way the inductive events seen in the peri-implantation embryo in vivo. While our understanding of these processes in mammals is still incomplete, studies of mouse development have identified several secreted polypeptide factors which mediate critical events in cellular commitment and development (reviewed by Beddington and Robertson, 1999).

Members of the transforming growth factor beta superfamily play a prominent role in driving cell commitment events in early development across the animal phyla. Bone morphogenetic protein-2 (BMP-2) is a member of the transforming growth factor beta superfamily implicated by gene ablation studies in several critical processes in early mouse development (Ying and Zhao, 2001b; Zhang and Bradley, 1996). Studies of mouse ES cell differentiation in vitro adduced evidence for a critical role for BMP-2 in the differentiation of extra-embryonic endoderm, and in the cavitation of the embryo, the process whereby programmed cell death in a subpopulation of the pluripotent stem cells of the inner cell mass leads to the formation of the egg cylinder (Coucouvanis and Martin, 1999). This study further showed that BMP-2 transcripts were expressed in the extra-embryonic endoderm, a finding that has recently been confirmed in a study which defined a role for this molecule in the induction of primordial germ cells (Ying and Zhao, 2001b).

Before the development of human ES cells, human embryonal carcinoma (EC) cells were used as a model to study cell differentiation in early human development. Human EC cells resemble primate ES cells in their morphology, surface marker and gene expression, and growth properties (Pera et al., 2000). A screen for factors affecting growth and differentiation of human pluripotent EC cells revealed that BMP-2 induced differentiation of the cells into a flattened, epithelial squamous cell displaying an immunophenotype and gene expression profile similar to extra-embryonic endoderm (Pera and Herszfeld, 1998). Treatment with retinoic acid had previously been shown to induce EC cells to undergo a very similar program of differentiation (Roach et al., 1994), and cells with similar morphology and marker expression commonly appear in EC cell cultures undergoing spontaneous differentiation. Recently, Xu et al. (Xu et al., 2002b) reported that BMP-4 induced the differentiation of human ES cells grown in serum-fee medium in the presence of FGF-2 into a different extra-embryonic lineage, the trophoblast.

Cells that morphologically resemble the flat epithelial cells induced in human EC cultures by BMP-2 also appear spontaneously in human ES cultures grown under standard conditions and are different in appearance to the cells described by Xu et al. (Xu et al., 2002b); under suboptimal conditions this differentiated cell grows rapidly and often overtakes the culture, leading to the elimination of stem cells and other differentiated progeny. We speculated that the commitment to undertake primitive endoderm differentiation might be driven by a positive feedback loop involving BMP-2, and that modulation of this pathway might facilitate ES stem cell renewal or differentiation into embryonic lineages, as suggested by studies in the mouse (Lake et al., 2000; Niwa et al., 2000). In this study we sought to characterize this form of differentiation and assess the role of BMP-2 in directing ES differentiation along this lineage.

Human EC cells and ES cells were cultured as described in previous publications (Pera et al., 1989; Reubinoff et al., 2000). All experiments were carried out at least twice on cell lines HES-2 and HES-3. The mouse embryo fibroblasts used throughout most of this study came from two different lots, and they were plated at a density of 7.5×10⁴/cm² for routine ES cell maintenance and experimental protocols. With one lot of embryo fibroblasts, BMP-2 effects were seen in ES cells plated onto feeder cell layers at this density, but with the second lot, effects were only apparent when ES cells were cultured on a feeder layer prepared at a lower density (see Results). Effects of noggin were apparent using either lot of feeder cells at the density normally employed for ES cell maintenance. Human EC cell differentiation was induced by all-trans-retinoic acid or BMP-2 treatment as described elsewhere (Pera and Herszfeld, 1998; Roach et al., 1994). Treatment of ES cells with noggin or bone morphogenetic proteins was begun one day following routine subculture and continued for 5-14 days. The medium used for conversion of noggin monolayer cells to neural progenitors (neural progenitor medium) was the same as that used previously to grow neural progenitors from spontaneously differentiating ES cell cultures (Reubinoff et al., 2001).

Recombinant human BMP-2, BMP-4 and recombinant mouse noggin were obtained from R&D Systems. Mammalian expression plasmids containing DAN or Cerberus cDNA were described elsewhere (Biben et al., 1998; Stanley et al., 1998). These proteins were expressed in 293 T cells and purified by affinity chromatography using a FLAG affinity column.

The coding sequences of BMP-2, BMP-4 and BMP-7 were amplified from a mouse embryonic day 7 cDNA library using a PCR-based approach. This process yielded two fragments for each of BMP-4 and -7, representing the pro-domain and cystine knot. For BMP-2, only the cystine knot region was amplified. The primers used to generate these fragments were: BMP-4 Pro-domain (CATGGCGCGCCTGATGATTCCTGGTAACCGAATG and ACGCGTCTTGGGACTACGTTTGGCCCT), BMP-7 Pro-domain (ACGCGTATGCACGTGCGCTCGCTGCGCGCTG and ACGCGTCCAAAGAACCAAGAGGCCCTG), BMP-2 cystine knot (GGGACGCGTAAGCGCCTCAAGTCCAGCTGC and CAACGCGTTGCTGTGCTAACGACACCCGCAG), BMP-4 cystine knot (CTGACGCGTAGGAAGAAGAATAAGAACTGC and TCCGCCCTCCGGACTGCCTGATCTC), BMP-7 cystine knot (ACGCGTCCAAAGAACCAAGAGGCCCTG and GACGCGTGAAGAGCTAGTGGCAGCCACAGG). MluI digested DNA fragments encompassing sequences encoding the pro-domains of BMP-4 and BMP-7 were ligated into the AscI site of pEFBOS (Mizushima and Nagata, 1990) that had been previously modified to include myc or glu/glu epitope tags. Recombinant plasmids resulting from this ligation contained the pro-domains N-terminal to the epitope tag. These plasmids were subsequently digested with MluI and ligated to MluI fragments representing the cystine knot regions of BMP4, 2 and 7 to yield a series of vectors of the configuration BMP4-Pro-domain-myc-BMP-4, BMP-4-Pro-domain-myc-BMP-2 and BMP-7-Pro-domain-glu-BMP-7. Western blot analysis of supernatants from 293T cells transfected with these vectors contained myc- and glu-tagged proteins of approximately 20 kDa (data not shown), the predicted size of processed monomeric BMPs. These supernatants also possessed BMP activity as adjudged by their ability to induce alkaline phosphatase activity in C2C12 cells (Katagiri et al., 1994) (data not shown). To generate supernatants containing BMP-2/-7 heterodimers, 293T cells were transfected with the vectors BMP-4-Pro-domain-myc-BMP-2 and BMP-7-Pro-domain-glu/glu-BMP-7, whereas supernatants containing BMP homodimers were derived from cells transfected with only one of the above expression vectors. Formation of heterodimers was confirmed by affinity copurification of myc- and glu-tagged proteins and by the much higher biological activity of the heterodimers in the C2C12 bioassay compared with homodimers. Purified follistatin (native protein from follicular fluid) was obtained from David Phillips of this Institute.

Noggin-treated cells were harvested after 10-14 days using dispase and were further subcultured on monolayers of mouse embryo fibroblasts, as described for human ES cells, or under similar conditions in the absence of a feeder cell layer, or as neurospheres (Reubinoff et al., 2001). To determine the proportion of noggin or control ES cell cultures that could be converted to neural progenitors, colonies were grown in control or noggin containing medium for 5 days, dissected into pieces approximately 0.5 μm in diameter and transferred to neural progenitor medium. After 1 week growth in suspension culture, the embryoid bodies or neurospheres were replated as described previously (Reubinoff et al., 2001) on to laminin-coated dishes in neural precursor medium lacking growth factors, allowed to attach and grow for 2 days and then stained for nestin. The proportion of colonies showing uniform positive staining for nestin was determined by inspection and counting using indirect immunofluorescence.

RNA isolation on magnetic oligo dT beads or on oligo dT Separose was carried out as described elsewhere (Reubinoff et al., 2000; Roach et al., 1994). RT-PCR was carried out as described previously (Reubinoff et al., 2000). Product sizes, annealing temperatures and primers for PCR reactions for all gene products examined in this study are listed in Table 1. Thirty cycles of PCR were carried out for all reactions. All products were sequenced, and identity with the expected human cDNA was confirmed in all cases. All RT-PCR reactions were repeated on three separate occasions with consistent results.

Northern analysis of human EC cell RNA for BMP-2 transcripts was carried out as described (Roach et al., 1994) using a random primed ³²PcDNA probe consisting of a partial cDNA clone corresponding to the region from 677-1333 bp of the cDNA sequence for human BMP-2 (GI4557368).

ES cell cultures 7-14 days old containing stem cells and differentiated cells were extracted in organ culture dishes in situ for 30 minutes into buffer containing 150 mM NaCl, 50mM Tris and 1% Nonidet P-40, pH 8. The lysate, and the nuclei and cytoskeletal material detached with it, were removed from the dishes, and the remaining proteins on the monolayer were harvested into 100-200 μl of SDS-PAGE sample buffer, and 20 μl was added to each lane of a gel. Four to six organ culture dishes were used in each experiment. The samples were run on 6% or 12% polyacrylamide gels under reducing and denaturing conditions, and the proteins were transferred to Immobilon membranes, which were probed with mouse anti-BMP-2 (12% gel) or GCTM-2 (6% gel). Detection was carried out using rabbit anti-mouse immunoglobulin conjugated to horseradish peroxidase and enhanced chemiluminescence.

The sources and methods for indirect immunofluorescence microscopy with antibodies GCTM-2, anti-desmin and neural markers, were as described previously (Reubinoff et al., 2000). The same methods were used with these antibodies: TG343, reactive with the surface proteoglycan recognized by GCTM-2 and TRA1-60, from this laboratory (Cooper et al., 2002); TG42.1 reactive with a 25 kDa surface protein found on stem cells (this antigen copurifies with the GCTM-2 antigen and is identical to the tetraspannin CD9) (L. Stamp, A. Laslett and M.F.P., unpublished); Oct-4, mouse monoclonal against a human Oct-4 sequence from Santa Cruz; rabbit antiserum against Sox-2 from Robin Lovell-Badge at the National Institute for Medical Research, London; GCTM-5 reactive with a surface protein found on differentiating cells of human ES and EC cultures and with fetal hepatocytes, from this laboratory (L. Stamp, H. A. Crosby, S. M. Hawes, A. J. Strain and M.F.P., unpublished); antibodies to Notch-1 and Notch-2, from the laboratory of Professor Spiros Artavanis; anti-human BMP-2 from R&D Systems; antibody PHM-4, reactive with Class I HLA surface antigens, from the Department of Nephrology, Monash Medical Centre (Hancock et al., 1982); mouse monoclonal H17 against placental alkaline phosphatase from Jose Luis Millan at the Burnham Institute, La Jolla Ca.; mouse monoclonal antibody against SPARC from Hematologic Technologies, Essex, Vermont. Assays for the beta subunit of human chorionic gonadotrophin were performed as described elsewhere (Reubinoff et al., 2000).

Quantitative analysis for GCTM-2 staining in control, BMP-2 and noggin-treated cultures was carried out by harvesting colonies first with dispase, then dissociating them to single cells with trypsin. Cells were stained in suspension using the primary antibody and anti-mouse immunoglobulins conjugated to FITC, after which they were counted under the fluorescence microscope to determine the proportion of positive cells. Five hundred cells from each group were counted. Alternately, when larger numbers of colonies were harvested, analysis was carried out by flow cytometry (below).

For staining with antibodies to phosphorylated Smad1, ES cell suspensions were plated onto glass coverslips coated with 0.1% gelatin, grown overnight then transferred to ES medium supplemented with 0.5% FCS for 4 hours, after which 50 ng/ml BMP-2 was added for 1.5 hours. Cells were fixed in 2% paraformaldehyde, rinsed three times in PBS, permeabilized in 0.1% Triton, rinsed again, then incubated in anti-phosphorylated-Smad1 (1:50) (Persson et al., 1998)or GCTM-2 (1:10) in PBS with 4% normal goat serum for 1 hour at room temp. Secondary antibodies (goat anti-rabbit-cy3-IgG 1:250 and goat anti-mouse IgM-FITC 1:100, respectively) were added for 1 hour after washing three times in PBS/0.05% Tween. Coverslips were mounted in Mowiol and viewed in a confocal laser scanning microscope.

Control cells or cells treated with 25 ng/ml BMP-2 were harvested after 5 days of treatment following subculture. The cells were carefully trypsinized to yield single cell suspensions, which were incubated with monoclonal antibody GCTM-2 or an isotype-matched control antibody on ice for 30 minutes, followed by several washes in phosphate buffered saline, then a 30 minute incubation with anti-mouse immunoglobulins conjugated to fluorescein isothiocyanate. The cells were rinsed with phosphate buffered saline and fixed in 0.4% paraformaldehyde before analysis. Cells were also incubated with isotype matched control primary antibody followed by secondary antibody.

BMP-2 treatment of human ES cell cultures grown in medium containing fetal calf serum in the presence of a feeder cell layer induced differentiation into flat, squamous epithelial cells (Fig. 1A, control; B, treated cells); these cells sometimes formed fluid-filled cysts within the culture dish. They were very similar in appearance to those commonly observed during spontaneous differentiation of human ES cells (Fig. 1C). The appearance of a sheet of spontaneously differentiating cells that was scraped from the culture surface and processed using routine histological methods is shown in Fig. 1D (BMP-treated cells had a similar appearance). The flattened cells with abundant eosinophilic intracellular material resembled cells found in the primary yolk sac of the primate embryo and in certain morphological variants of yolk sac carcinomas; in the presence of BMP-2 the entire culture eventually took on this appearance. By contrast, under optimal growth conditions, spontaneously differentiating ES cell cultures grown in monolayer gave rise not only to these cells but also to a wide range of additional cell types.

Indirect immunofluorescence examination revealed that the BMP-2-treated cells or spontaneously differentiating cells resembling them expressed low molecular weight cytokeratins and abundant extracellular matrix proteins (Fig. 1E,F, cytokeratin; G,H, laminin; I, SPARC staining at an early stage of differentiation). Particularly strong staining for alphafetoprotein was observed in cystic cells and cells at the base of cysts (Fig. 1J,K). These flat epithelial cells were not stained with the ES stem cell antibody GCTM-2, nor were they reactive with antibodies to class I major histocompatibility complex (MHC) molecules. The effect of BMP-2 on the expression of the stem cell marker GCTM-2 could be shown quantitatively by flow cytometric analysis, with a lower proportion of any cells showing staining after 5 days of BMP-2 treatment compared with control cultures (negative cells increased 1.6-fold relative to controls, Fig. 2).

Xu et al. (Xu et al., 2002b) recently showed that treatment of the human ES cell line H-1 with BMP-4 induced differentiation into trophoblast cells. Under our conditions of treatment, very few cells expressing placental alkaline phosphatase or human chorionic gonadotrophin were observed in control or treated cultures, and the morphology of the BMP-2- or BMP-4-treated cells was very different to that observed by Xu et al. (data not shown). Nevertheless, we occasionally observed foci of cells resembling the trophoblast precursors described by these workers; these cells appeared in less than 5% of colonies treated with BMP-2, BMP-4 or BMP-2/-7 under our culture conditions.

Initially our experiments used BMP-2 only and were performed with ES cells grown on a specific lot of mouse embryo feeder cells. When another lot of feeder cells was used, the effects of BMP-2 and other bone morphogenetic proteins were minimal or nonexistent. Antagonism of the action of the exogenous bone morphogenetic proteins appeared to account for the lack of a BMP effect on ES cells cultured using the second lot of feeder cells, as further examination showed clearly that the outcome of treatment was entirely dependent on feeder cell density. When the second batch of feeder cells was used at the density employed for routine ES maintenance, the effect of BMP-2 was variable, but a threefold reduction in feeder cell density revealed a strongly inhibitory effect of the protein on stem cell maintenance (Fig. 3A; the experiment was repeated three times on HES-2 and HES-3 with similar results and each experiment analyzed duplicate wells containing at least four colonies per treatment). Using this density of the second lot of mouse embryo feeder cells, we observed that other bone morphogenetic proteins, including BMP-4, BMP-7 and BMP-2/BMP-7 heterodimers, could induce the same effect as BMP-2. To determine the possible mechanism whereby feeder cell layers might inhibit the action of bone morphogenetic proteins, we carried out RT-PCR analysis for expression of known antagonists of BMP in this system. Examination of embryo fibroblast feeder cells for transcripts encoding known inhibitors of bone morphogenetic protein revealed that the antagonist gremlin was transcribed in these cells (Fig. 3B).

Analysis of gene expression in the BMP-2-treated cultures consisting mainly of the flat epithelial cells by RT-PCR (Fig. 4A) showed reproducible loss of stem cell markers, and upregulation of a range of markers characteristic of endoderm, including transcription factors (HNF3-α, HNF-4, GATA-4 and GATA-6) and genes encoding secreted products and extracellular matrix molecules expressed in parietal endoderm in the mouse and visceral endoderm in mouse and human. RNA from stem cells and BMP-2-treated cells yielded similar amounts of RT-PCR product for beta actin. Some transcripts characteristic of endoderm differentiation were observed at lower levels in control ES cell cultures, indicative of low levels of spontaneous differentiation into this lineage.

The response of human ES cells to BMP-2, and the presence of cells resembling those in BMP-2-treated cultures in untreated ES cell cultures undergoing spontaneous differentiation, suggested that endogenous BMP-2 production might be modulating the spontaneous differentiation observed. We carried out RT-PCR analysis for BMP-2, and its receptors BMPR1A, BMPR2 and the activin receptor II beta, and showed expression of transcripts for all these genes in stem cells and in cultures consisting predominantly of flat squamous epithelial cells (Fig. 4B).

We speculated that BMP-2 production might be activated at early stages of ES cell differentiation, driving a positive feedback loop towards extra-embryonic differentiation. To assess this possibility we studied BMP-2 expression in differentiating cultures of human EC cells. Pluripotent human EC cell line GCT 27X-1 resembles ES cells in morphology, marker expression and growth requirements. However, EC cells are easier to grow in large quantities as pure stem cell population. Cell line GCT27X-1 was cultured in the absence of a feeder cell layer and differentiation was induced by treatment with all-trans retinoic acid or BMP-2 as described previously (Pera and Herszfeld, 1998; Roach et al., 1994). Spontaneously differentiating cultures (controls) and retinoic acid treated cultures both show increased levels of transcripts for BMP-2, but a very striking increase in these transcripts is seen 12-48 hours after BMP-2 treatment (Fig. 4C). Thus, treatment of human pluripotent cells with BMP-2 leads to the accumulation of transcripts for this factor, consistent with a positive feedback model.

We also determined whether or not BMP-2 protein was present in ES cell cultures. Attempts to identify the protein in culture supernatant or cell lysates prepared using nonionic detergents failed. Following lysis of spontaneously differentiating ES cell monolayers with nonionic detergent and removal of the lysate, we extracted the remaining protein on the monolayer with reducing SDS-PAGE sample buffer. Immunoblotting of this material revealed the presence of BMP-2 protein of the expected size, and a strong band corresponding to a doublet (Fig. 4D); despite the use of reducing sample buffer, it was difficult to convert all this material to monomeric form. BMP-2 protein was also detected by immunostaining in differentiating colonies of human ES cells (Fig. 4E-G, cells grown in the absence of a feeder cell layer). Recent cell biological (Suzawa et al., 1999) and genetic (Arteaga-Solis et al., 2001) studies have highlighted the activity of bone morphogenetic proteins bound to pericellular matrix.

Examination of feeder cell layer by RT-PCR or by immunostaining failed to detect BMP-2 or BMP-4 transcripts, or BMP-2 immunoreactivity (data not shown).

To determine whether BMP-2 addition to human ES cell cultures resulted in activation of the Smad signal transduction pathways, and to evaluate the extent to which this signaling operated spontaneously in ES cells, we analyzed cells for the presence of phosphorylated Smad-1 protein in the nucleus. Before BMP2 treatment, phosphorylated Smad-1 staining was found at low levels in stem cell nuclei (Fig. 5A-D), whereas after BMP2 addition, much brighter staining was detectable in the nuclei of undifferentiated cells (Fig. 5C-I). Thus, stem cells appeared to activate the Smad-1 pathway under basal conditions, and BMP-2 treatment enhanced nuclear localization of this signal transduction molecule.

The results above suggested that stem cell maintenance might be a dependent on a balance between expression of the BMP antagonist gremlin by the fibroblast feeder cell layer and the endogenous production of BMP-2 by the differentiating stem cells. We postulated that by strongly interfering with endogenous BMP-2 action, it would be possible to prevent the differentiation of the cells into the extra-embryonic phenotype, leading either to enhancement of stem cell renewal, or to an enhancement of differentiation into embryonic lineages. We tested several natural and synthetic antagonists of BMP for activity on human ES cell cultures. Follistatin, DAN, and Cer-1 were without any obvious morphological effect, as was a soluble form of the BMPRIA ectodomain. However, treatment with a recombinant form of mouse noggin had a profound effect on human ES cell morphology in the dose range 100-500 ng/ml. After approximately 5 days in culture, noggin-treated cultures consisted of colonies of small round cells, which were different in appearance from ES cells, and, in contrast to control cultures, colonies in noggin-treated dishes contained no flat squamous epithelial cells or cystic structures similar to those seen after BMP-2 treatment (Fig. 6A-C). The size of the noggin-treated colonies was smaller than those of controls, although the same number of cells was present in each (not shown). In noggin-treated cultures, the percentage of cells positive for the stem cell marker GCTM-2 was much lower than controls (Fig. 6D).

The immunophenotype of the noggin-treated cells was distinguished by their lack of expression of several markers characteristic of ES cells or differentiated cells found spontaneously at early time points (approximately 5-7 days) following ES cell subculture under standard conditions (Table 2). Some markers of early neural differentiation were found in noggin-treated colonies.

RT-PCR analysis confirmed that the noggin-treated cells expressed Oct-4 transcripts at low levels, if at all (Fig. 6E, comparison of noggin-treated culture with ES cells). The noggin-treated cells did express transcripts for markers characteristic of early neuroectoderm such as Pax-6 or nestin. The noggin cells did not express brachyury, characteristic of early mesoderm, and finally, they did not express gene products characteristic of extra-embryonic endoderm, which are found in spontaneously differentiating control cultures or in BMP-2-treated cultures.

The noggin-treated cells were further characterized in biological studies. The addition of 25 ng/ml recombinant human BMP-2 to ES cell cultures along with 250 ng/ml noggin led to the appearance of squamous cells and cysts characteristic of spontaneously differentiating ES cell cultures or BMP-2-treated cultures, indicating that BMP-2 could antagonize the noggin effect (not shown). The noggin-treated cells could be subcultivated under standard conditions for ES cell culture in the presence of a feeder cell layer, and retained their distinctive morphology under these conditions for at least several passages (Fig. 6C). To compare the neurogenic potential of control and noggin-treated ES cells, colonies of either cell type were dissected under microscopic control and transferred to medium designed to support the growth of neural stem cells (Fig. 7A-D). Control cells formed embryoid bodies that displayed variable morphology, were often cystic and frequently reattached to the culture surface (Fig. 7A). By contrast, noggin-treated cells formed spheres which remained floating in suspension and could be serially cultivated (Fig. 7B). The fate of either control ES cells or noggin-treated ES cells following culture in serum-free medium was examined by allowing them to reattach to a poly-L-lysine-coated surface, then staining them with nestin (Fig. 7C,D). The proportion of structures forming neurospheres after 5 days of noggin treatment varied between 40-70%, but, in any given experiment it was at least fivefold higher than that of untreated ES cells (Fig. 7E). Nestin-negative colonies from noggin-treated cells retained the appearance of noggin cells grown on feeder layers, whereas nestin-negative colonies from control cultures had a varied appearance typical of mixed embryoid bodies allowed to reattach to the culture surface. After 2 weeks of treatment, the proportion of noggin-treated colonies that could be converted to nestin-positive spheres rose to 90% or higher (B.R., unpublished; G. Peh, S. Hawes and M.F.P., unpublished).

When the spheres derived from noggin-treated cells were allowed to attach to the dish, cells forming elongated processes migrated out onto the monolayer (Fig. 8A). These cells displayed an immunophenotype consistent with their identification as mature neurons, including expression of 200 kDa neurofilament protein and MAP2-a,b (Fig. 8B-E). However, if noggin-treated cells were cultivated as monolayers in the absence of a feeder cell layer and in the presence of serum, they gave rise to cultures consisting of cells with fibroblastoid morphology. At least 50% of the cells in these cultures stained with antibodies against glial fibrillary acidic protein and vimentin (Fig. 8F,G).

Several groups have now reported the spontaneous differentiation of human ES cells (Itskovitz-Eldor et al., 2000; Levenberg et al., 2002; Reubinoff et al., 2000; Schuldiner et al., 2000; Xu et al., 2002a), and most of these have used embryoid body formation to begin the initial process of differentiation. ES cells allowed to differentiate on plastic surfaces or embryoid bodies give rise to a mixture of cells, a significant number of which express markers of extra-embryonic endoderm. The regulation of this process of spontaneous differentiation is poorly understood. Schuldiner et al. (Schuldiner et al., 2000) showed that growth factor treatment could influence some of the outcome of spontaneous differentiation of ES cells but, like other studies to date, this one did not directly elucidate the factors involved in spontaneous differentiation. BMP-2 has a role in visceral endoderm differentiation in the mouse ES cell system, and it is expressed in the extra-embryonic endoderm in this species (Coucouvanis and Martin, 1999; Ying and Zhao, 2001a). We previously observed that BMP-2 could induce differentiation of human EC cells into a cell with patterns of gene expression resembling extra-embryonic endoderm (Pera and Herszfeld, 1998).

Here we showed that BMP-2 has a similar effect in human ES cell cultures, and that this cell type is also seen during spontaneous differentiation of ES cells. The identification of these differentiated epithelial cells is based chiefly on their patterns of gene expression and on immunostaining. Thus, the cells are epithelial cells expressing low molecular weight cytokeratins, laminin, alpha-fetoprotein, transferrin, SPARC and vitronectin but not class I HLA molecules. Morphologically, the BMP-treated cells resemble a cell found in the human peri-implantation embryo, which forms a mesh-like network within the blastocoel cavity. This cell was known classically in the human embryology literature as a mesoblast, but on comparative anatomical grounds Luckett (Luckett, 1978) argued that these cells were more likely to represent extra-embryonic endoderm. Several further morphological studies of the primate embryo have documented the development of cells resembling parietal endoderm and visceral endoderm from flattened cells formed below the epiblast (Enders et al., 1990; Enders and Schlafke, 1981; Enders et al., 1986). Although detailed studies of marker expression in these cells in the primate embryo have not been carried out, the properties of the BMP-induced cell are consistent with their identification as extra-embryonic endoderm. Formation of this cell in vitro may provide a source of multiple signals that influence human ES cell differentiation, and controlling these signals may be critical to directing differentiation of the ES cells themselves.

Xu et al. (Xu et al., 2002b) have shown that a different type of extra-embryonic cell, the trophoblast, can be induced by BMP-4 treatment of ES cells cultured in serum-free medium in the presence of FGF-2 (Fig. 9). Although we have not routinely observed cells of this phenotype in our experiments, we have occasionally observed small foci of cells resembling those shown in the study of Xu et al. (Xu et al., 2002b). It is possible that differences between the cell lines studied by our two groups could account for differences in response to these factors. It is also possible that serum components influence the outcome of differentiation, but it is interesting that activation of the same signal transduction pathway can result in the formation of either extra-embryonic endoderm or trophoblast. It has been shown that mouse ES cells undergoing differentiation can follow either of these pathways depending on whether Oct-4 levels rise or fall (Niwa et al., 2000).

The data reported here strongly suggest that there is a positive feedback loop involving BMP-2, which acts to drive extra-embryonic endoderm differentiation of human pluripotent stem cells. Our data show that BMP-2 can induce its own expression in human EC cells. Both human EC and ES cells, and the cells that differentiate from them, express transcripts for receptors that enable them to respond to the factor. Consistent with the RNA analyses, BMP-2 protein is present in the pericellular matrix of spontaneously differentiating ES cells. Data on nuclear localization of phosphorylated Smad-1 show that under standard culture conditions some nuclear staining is observed in ES cells; the intensity of the nuclear staining is greatly increased following treatment with BMP-2. All these data are consistent with an endogenous signaling system based on BMP-2 that drives extra-embryonic endoderm differentiation.

There are several natural and synthetic antagonists to the BMPs. We showed that transcripts for the antagonist gremlin are produced by mouse embryo fibroblasts, and previous studies have shown that rat embryo fibroblasts produce gremlin protein, much of which is cell associated and not soluble (Topol et al., 2000). Gremlin may be one of a number of factors produced by feeder cells that are required for stem cell renewal. We speculate that high local concentrations of cell-associated gremlin or other feeder cell products may have a spatially limited effect in blocking stem cell differentiation. We show elsewhere (S. M. Hawes and M.F.P., unpublished) that the zones of human ES cell colonies at some distance from the feeder cell layer undergo differentiation first.

We hypothesize that stem cell fate may be determined in part by a balance between feeder cell inhibition of differentiation, mediated by gremlin, and by the production of BMP-2, which drives differentiation. We tested the effect of the addition of several BMP antagonists in the human ES cell system and obtained a specific effect with noggin. This effect was evident after a short period of treatment of ES cell colonies, and appeared to affect most cells in the culture. Noggin is known to play a role in the induction of the nervous system in several vertebrate model systems by antagonizing BMPs (Bachiller et al., 2000; Smith and Harland, 1992) (reviewed by Streit and Stern, 1999). The noggin-treated cells can give rise to both neuronal and glial lineages, but they themselves do not appear to be equivalent to the neural progenitors that we and others have previously derived from human ES cells. Noggin-derived neural cells have properties expected of mature neurons and glial cells, and the noggin neural progenitors have been shown to engraft into the nervous system of experimental animals and undergo appropriate differentiation (B.R., T. Ben-Hur, E. Reinhartz, A. Itzik, M. Idelson and M.F.P., unpublished), as shown previously for neural progenitors derived from spontaneously differentiating ES cell cultures (Reubinoff et al., 2001).

Because antagonism of BMP signaling blocks extra-embryonic differentiation, it is interesting that noggin treatment does not enhance stem cell renewal. One plausible hypothesis to account for our results is that extra-embryonic cells, in addition to secreting bone morphogenetic proteins and other factors that can induce differentiation, also produce factors that help maintain ES cells. In the mouse embryo, the extra-embryonic endoderm produces a variety of factors that act locally; some are postulated to induce ectoderm, others mesoderm, and some are known to induce germ cell formation. Thus, this tissue may have multiple effects on embryonic stem cells. If the extra-embryonic factors are not present to maintain stem cells, then the stem cells may default towards neuroectoderm differentiation as depicted in the scheme shown in Fig. 9. In this scheme, feeder cells produce gremlin or other factors that act locally to maintain stem cell renewal but allow for some degree of extra-embryonic differentiation. Complete blockade of extra-embryonic differentiation by exogenous noggin removes a source of factors that can support stem cell renewal and factors that drive other differentiation pathways, resulting in default differentiation into the neural lineage.

Further work will be required to determine the developmental potential of the noggin cells isolated here. It appears clear that they have neurogenic potential. By bypassing extra-embyonic endoderm differentiation, it may be possible to direct the fate of human ES cells into specific somatic lineages, as suggested for mouse ES cells (Lake et al., 2000). The results here indicate that BMP-controlled extra-embryonic differentiation is an important regulatory node in ES differentiation, show that addition of polypeptide regulators of early mammalian development can direct the fate of human ES cells, and provide a facile route to the generation of neural precursors from ES cells.

At Monash University, this work was supported by grants from the National Health and Medical Research Council, and the National Institutes of Health (NIGMS GM68417), and by a sponsored research agreement with ES Cell International Pte. We gratefully acknowledge the expert assistance of Jacqui Johnson, Irene Tellis, Karen Koh and Linh Nguyen with human ES cell culture. We thank Peter ten Dijke for the antibody to phosphorylated Smad-1and Leon Tertoolen for help with the confocal microscopy.