Activin/Nodal signalling maintains pluripotency by controlling Nanog expression

Human embryonic stem cells (hESCs) are pluripotent cells derived from embryos cultured from the blastocyst stage. Their embryonic origin confers upon hESCs the ability to proliferate indefinitely in vitro while maintaining their capacity to differentiate into the three primary germ layers, ectoderm,mesoderm and endoderm, from which all the cells of the adult body are derived. Despite their apparent common origin, recent studies have revealed that mouse and human ESCs use different signalling pathways to maintain their pluripotent status. Mouse embryonic stem cells depend on leukaemia inhibitory factor(Smith et al., 1992) and bone morphogenetic protein (Ying et al.,2003) to maintain pluripotency, whereas their human counterparts rely on Activin/Nodal (James et al.,2005; Vallier et al.,2004) and fibroblast growth factors(Xu et al., 2005). However, in vitro and in vivo studies have established that a core transcription factor circuit involving Oct4, Sox2 and Nanog is necessary for pluripotency in both species. Indeed, constitutive expression of Nanog is sufficient to prevent differentiation of mouse and human ESCs(Chambers et al., 2003; Darr et al., 2006), and loss of function confirms that Nanog is necessary to block primitive endoderm differentiation (Chambers et al.,2003; Hyslop et al.,2005; Mitsui et al.,2003). These apparent contradictory observations underline the lack of knowledge concerning the mechanisms linking extracellular signalling and the core transcriptional network (including Nanog), especially in hESCs.

Here, we show that Smad2/3, the downstream effectors of Activin/Nodal signalling, bind and directly control the activity of the Nanog gene in hESCs. Accordingly, inhibition of Activin/Nodal signalling resulted in a loss of Nanog expression while inducing differentiation toward neuroectoderm. Knockdown of Nanog expression mimicked this effect, which is strictly dependent on FGF signalling. Conversely, constitutive expression of Nanog was sufficient to maintain the pluripotent status of hESCs in the absence of Activin/Nodal signalling, by specifically blocking neuroectoderm differentiation. In addition, our biochemical analyses showed that Nanog interacts directly with Smad2/3 proteins, the direct effectors of Activin/Nodal signalling, to limit their transcriptional activity, which is crucial for the cell fate choice between pluripotency and differentiation in hESCs. Importantly, similar results were obtained using pluripotent stem cells derived from the epiblast layer of pre-gastrula stage mouse embryos (mEpiSCs)demonstrating that these mechanisms are evolutionarily conserved, consistent with the proposed homology between hESCs and EpiSCs(Brons et al., 2007; Tesar et al., 2007). Taken together, these results demonstrate that Activin/Nodal signalling blocks neuroectoderm differentiation of pluripotent cells by maintaining Nanog expression, and they also provide the basis for a model explaining for the first time how Activin/Nodal signalling can maintain the pluripotency of hESCs without inducing differentiation towards endoderm.

hESCs [H9 (WiCell, Madison, WI, USA) and hSF-6 (UCSF, San Francisco, CA,USA)] and mEpiSCs (NOD-EpiSCs and 129S2-EpiSCs) were grown in chemically defined culture conditions, as previously described(Brons et al., 2007). For embryoid body (EB) formation and differentiation, hESC colonies were grown in non-adherent conditions as described(Vallier et al., 2004). For differentiation into extraembryonic tissues, hESCs and EpiSCs were grown in CDM with 10 ng/ml BMP4 for 7 days. For neuroectoderm differentiation, hESCs and EpiSCs were grown in CDM with 10 μM SB431542 and 12 ng/ml FGF2 for 7 days. For mesendoderm differentiation, hESCs were grown in CDM with 5 ng/ml Activin but without FGF2 for 3 days, and then for 4 additional days in the presence of 100 ng/ml Activin, 10 ng/ml BMP4 and 20 ng/ml FGF2 (our unpublished results).

Microarray analyses were performed as described by Brons et al.(Brons et al., 2007). All hybridisations employed are publicly available in MIAME format from the ArrayExpress microarray repository (European Bioinformatics Institute; http://www.ebi.ac.uk/arrayexpress)under Accession Number E-MEXP-1741.

Total RNAs were extracted using the RNeasy Mini Kit (Qiagen). For each sample, 0.6 μg of total RNA was reverse transcribed using Superscript II(Invitrogen). RT-PCR was performed as described(Vallier et al., 2004). Real-time PCR reaction mixtures were prepared as described (SensiMiX protocol Quantace), denatured at 94°C for 5 minutes, cycled at 94°C for 30 seconds, 60°C for 30 seconds and 72°C for 30 seconds for 40 cycles,then subjected to a final extension step at 72°C for 10 minutes. Primer sequences have been described elsewhere(Vallier et al., 2004; Brons et al., 2007). RT-PCR reactions were performed using a Stratagene Mx3005P in triplicate and normalised to porphobilinogen deaminase (PBGD) in the same run. Data represent the mean of three independent experiments and error bars indicate standard deviation.

The nuclear extract was prepared as described(Dyer and Herzog, 1995). The final protein concentration was 2 mg/ml in HEMG110 buffer (Hepes pH 7.6, 0.5 mM EDTA, 5 mM MgCl₂, 110 mM KCl, 20% glycerol, 0.2 mM DTT, Roche Complete cocktail, 0.5 mM PMSF). In each experiment, 0.5-1.0 mg nuclear protein and 3 μl antibody [αSmad2/3 (Cell Signalling), rabbit IgG(Sigma)] were used and incubated for 3 hours at 4°C. Protein G beads (10μl, Roche) were added and samples incubated for an additional hour. Beads were washed five times with HEMG110 and proteins eluted with LDS-loading buffer at 70°C. PAGE and western blotting was performed with the Invitrogen NuPAGE system. Primary antibodies [αNanog (R&D systems),αSox2 (Abcam)] were probed with secondary HRP antibodies (Sigma) and detected using the LumilitePlus kit (Roche). The HA and FLAG immunoprecipitations (IPs) were essentially performed as described above,using Roche and Sigma M2 affinity matrixes. The Flag-HA-hNanog construct is driven by the CAG promoter and was transfected by Lipofectamine 2000(Invitrogen). Cells were harvested after 42 hours.

ChIP was carried out as previously described by Forsberg and colleagues(Okabe et al., 1996), using antibody directed against Smad2 (Abcam) or Nanog (R&D Systems). Enrichment was measured by quantitative real-time PCR using SYBR green (SensiMix Quantace). Results were normalised against control region H located in the 3′ untranslated region of the Nanog gene(Fig. 1D) and are expressed as±s.d. from three experiments. Previous studies have described the location of the Smad2/3-binding regions in the promoter of Lefty(Besser, 2004) and of Smad7(Denissova et al., 2000).

Potential Smad-binding sites in the Nanog6 promoter construct were mutated using the Quick Change II Mutagenesis Kit (Stratagene) following the manufacturer's instructions. Constructs were sequenced to confirm the presence of the desired mutation and to check the integrity of the promoter sequence. The sequence of the Smad2/3-(1) binding site was mutated from AGAC to GGCC(-310 to -307) and the sequence of the Smad2/3-(2) binding site from AGAC to GGCC (-302 to -299).

Five shRNA-Nanog expression vectors (Sigma, SHGLY-NM_024865) and one shRNA-non-targeting expression vector (Sigma, SC001) were stably transfected into H9 hESC lines in CDM supplemented with Activin and FGF2. After selection,60 puro-resistant colonies were picked (10 colonies for each shRNA-Nanog and 10 colonies for the shRNA-Non-Targeting control) and each sub-line was screened for the expression of Nanog using immunostaining analyses(Fig. 3B). Importantly, absence of Nanog protein was not observed in any of the 10 sub-lines expressing the shRNA-non-targeting control. Knockdown of Nanog expression was confirmed by real-time PCR in 12 shRNA-Nanog-hESCs sub-lines with two randomly chosen shRNA-non-targeting-hESCs used as controls(Fig. 3A).

To discover the identity of genes controlled by Activin/Nodal signalling in hESCs, we compared the gene expression profile of hESCs grown in chemically defined medium (CDM) containing Activin with that of hESCs grown for two days in CDM without Activin and instead containing the Activin receptor inhibitor SB431542. As expected, inhibition of Activin/Nodal signalling induced a strong downregulation of genes known to be downstream of this pathway, including LEFTYA (LEFTY2 - Human Gene Nomenclature Database), LEFTYB (LEFTY1 - Human Gene Nomenclature Database) and NODAL (Besser, 2004),and induced a marked increase in early markers of neuroectoderm differentiation, including GBX2, HOXA1, OLIG3 and SIX1(Fig. 1A; see also Table S1 in the supplementary material). More surprisingly, the expression of NANOG was strongly downregulated in the absence of Activin/Nodal signalling, whereas the expression of OCT4 (POU5F1 - Human Gene Nomenclature Database) and SOX2 did not change as substantially(Fig. 1A). These observations were confirmed by real-time PCR analyses showing that the expression of NANOG was downregulated by 75% after 24 hours of SB431542 treatment,similar to the decrease in expression of NODAL and LEFTYA(see Fig. S1A in the supplementary material). By contrast, the decrease in OCT4 expression occurred more progressively, while the expression of SOX2 remained relatively constant (Fig. S1A in the supplementary material). Interestingly, normal levels of NANOG, NODAL and Lefty transcripts could be re-established in hESCs by adding Activin after SB431542 treatment (see Fig. S1B in the supplementary material). Similar results were obtained either in the presence or absence of cycloheximide, indicating that reactivation of NANOG transcription by Activin/Nodal signalling did not require de novo protein synthesis. Consistent with the marked downregulation of NANOG expression resulting from the inhibition of Smad2/3-mediated TGFβ signalling, overexpression of SMAD3 in hESCs resulted in a doubling of NANOG transcription. Conversely, expression of a dominant-negative form of SMAD3 resulted in a 10-fold decrease in NANOG transcription (see Fig. S1C in the supplementary material). Finally, similar results were obtained in mEpiSCs (Fig. S1D,E in the supplementary material). Taken together, these results show that Activin/Nodal signalling is necessary for the expression of NANOG in hESCs, and that this requirement is evolutionarily conserved in pluripotent cells derived from the epiblast of post-implantation mouse embryos.

To distinguish between the hypotheses of direct transcriptional regulation of Nanog by Smad2/3 and action through an unknown intermediate, we looked for functional SMAD2/3-binding sites in the human NANOGpromoter. We first determined that a 379-bp region located upstream of the NANOG ATG was sufficient to recapitulate the transcriptional activity induced by Activin/Nodal signalling (Fig. 1B). Interestingly, this region contains binding sites for OCT4,SOX2 and Nanog (Boyer et al.,2005), and also two consensus Smad2/3-binding sites [S2/3-(1) and S2/3-(2), see Fig. S1F in the supplementary material]. Mutation of these Smad2/3-binding sites revealed that the site nearest to the Nanog-binding site, S2/3-(2), was crucial for the transcriptional activity induced by Activin/Nodal signalling (Fig. 1C, see also Fig. S1G in the supplementary material), suggesting that this Smad2/3-binding site is functional. Interestingly, sequence alignment of the human NANOG promoter region to mouse, dog and cow equivalents revealed that the mouse promoter does not contain similar Smad2/3-binding sites (Fig. S1F in the supplementary material). This provides a further indication that the location of binding sites for highly evolutionarily conserved transcription factors can vary between humans and mice (Odom et al., 2007). Finally, chromatin immunoprecipitation (ChIP) assays were performed to identify genomic regions bound by SMAD2/3 in the NANOG promoter(Fig. 1D). These analyses showed that Smad2/3 binds the same genomic region (containing the putative Smad2/3-binding sites) that was identified using luciferase assays. Taken together, these results reinforce the recent study by Thomson and colleagues(Xu et al., 2008), which showed that NANOG expression is directly controlled by SMAD2/3 in hESCs. The similar dependence of Nanog transcription on Smad2/3 in hESCs and mEpiSCs suggests that humans and mice share this direct link between extracellular growth factors and the core controlling transcriptional network despite their distinct genomic organisation.

To further understand the relationship between Activin/Nodal signalling and Nanog transcription, we generated hESCs and mEpiSCs constitutively expressing NANOG (NANOG-hESCs and Nanog-EpiSCs) (see Fig. S2A-C in the supplementary material). We then analysed the effect of Nanog overexpression when Activin/Nodal signalling was pharmacologically inhibited for 7 days. Under these conditions, wild-type cells differentiate into nearly homogenous populations of neuroectoderm progenitors(Smith et al., 2008)expressing SOX1, PAX6, SOX2, HOXA1, GBX2 and NCAM(Fig. 2A,B; Fig. S2E in the supplementary material; our unpublished results). By contrast, NANOG-hESCs grown under the same conditions did not undergo neuroectoderm differentiation,as shown by the expression of the pluripotency markers OCT4 and TRA-1-60, and by the absence of NCAM, SOX1, PAX6, HOXA1 and GBX2 expression (Fig. 2A-C; see also Fig. S2E in the supplementary material). In addition, FGF5 was not detected in NANOG-hESCs grown in the absence of Activin/Nodal signalling, thereby excluding the possibility that NANOG could block the differentiation of hESCs at a primitive ectoderm-like stage in these culture conditions (Fig. S2E in the supplementary material)(Darr et al., 2006). Interestingly, NANOG-overexpressing cells could be grown for prolonged periods in the presence of SB431542 while maintaining the expression of pluripotency markers and the absence of SOX1 and PAX6(Fig. 2A). These findings show that NANOG overexpression was sufficient to substitute for Activin/Nodal signalling in the maintenance of hESC pluripotency. Similar results were obtained with Nanog-overexpressing mEpiSCs (see Fig. S3A,B in the supplementary material). These results were also confirmed by growing hESCs as embryoid bodies (EBs), showing that NANOG inhibited neural development regardless of the culture conditions used to induce differentiation (Fig. S4A-C in the supplementary material). Finally, overexpression of OCT4 or SOX2 in hESCs was unable to prevent the neuroectoderm differentiation of hESCs grown in the presence of SB431542 (data not shown), demonstrating that the ability to inhibit hESC differentiation was unique to Nanog and was not shared by other components of the core pluripotency transcriptional circuit. Taken together, these results indicate that among the effectors of pluripotency downstream of Activin/Nodal signalling, Nanog expression is sufficient to block neuroectoderm differentiation in both hESCs and mEpiSCs.

The neural differentiation caused by inhibiting Activin/Nodal signalling in hESCs (Smith et al., 2008) led us to determine whether this effect was related to NANOG. Accordingly, we knocked down NANOG in hESCs by stably expressing a short hairpin RNA (shRNA)directed against NANOG transcripts (see Materials and methods). Immunostaining and real-time PCR analyses showed that NANOG expression was reduced by 90% (Fig. 3A-C),which represents a decrease in NANOG expression similar to that induced by the inhibition of Activin/Nodal signalling in hESCs (see Fig. 1B). OCT4expression was also decreased by 60%, whereas SOX2 expression was maintained at normal hESC levels (Fig. 3C). Importantly, shNanog-hESCs were able to proliferate almost indefinitely in CDM supplemented with Activin and FGF2 while maintaining a low level of OCT4 expression (Fig. 3D,E), confirming that Nanog is not necessary for self renewal(Chambers et al., 2007). The analysis of markers for extraembryonic tissues and for each of the three primary germ layers showed that the decrease in NANOG expression did not induce increases in primitive endoderm (SOX7, GATA6), trophectoderm(HAND1, eomesodermin) or mesendoderm (brachyury, eomesodermin, MIXL1, SOX17) markers (Fig. 3C). Notably, these experiments were performed in CDM containing Activin and FGF2, which does not possess any BMP-like activity(Vallier et al., 2005) that could interfere with the outcome of Nanog knockdown or could induce extraembryonic tissue differentiation. In striking contrast, NANOG knockdown increased the expression of neuroectoderm markers (SOX1, SIX1, GBX2, OLIG3,HOXA1; see Fig. 3C-E), showing that NANOG plays a key role in preventing neuroectoderm differentiation of hESCs. Importantly, NANOG knockdown in hESCs resulted in increases in neuroectoderm marker expression (GBX2, HOXA1 and SOX1) that were similar to those observed following pharmacological inhibition of Activin/Nodal signalling (Fig. 2A; Fig. S2E in the supplementary material). Taken together, these observations indicate that NANOG is responsible for inhibiting the neuroectoderm differentiation of hESCs, and that it mediates the effects of Activin/Nodal signalling in achieving this important element of pluripotency.

FGF signalling has been shown to be necessary for inducing neuroectoderm specification in amphibian and chick embryos(Stern, 2005), whereas BMP signalling is known to inhibit the same differentiation(Munoz-Sanjuan and Brivanlou,2002). In principle, Nanog could block neuroectoderm differentiation by interfering with an FGF inductive effect or by enhancing a BMP inhibitory effect. To distinguish between these two possibilities,wild-type hESCs and Nanog-overexpressing hESCs were grown in CDM supplemented with SB431542 and FGF2, with SB431542 and SU5402 (a chemical inhibitor of FGF receptors), with SB431542, FGF2 and BMP4, or with SB431542, FGF2 and Noggin(an inhibitor of BMPs; see Fig. 2D). The inhibition of FGF signalling resulted in a decrease in GBX2, SOX2, SOX1, PAX6, OLIG3 and HOXA1 expression in wild-type cells grown in the presence of SB431542, confirming that FGF signalling is necessary for the neuroectoderm specification of hESCs(Fig. 2D). Importantly,addition of BMP4 completely inhibited the expression of neuroectoderm markers(Fig. 2D, data not shown),confirming the inhibitory effect of BMP signalling on neuroectoderm specification. However, the presence of BMP4 did not maintain pluripotency (as shown by the decrease in OCT4 expression, see Fig. 2D), but instead drove the differentiation of wild-type hESCs and NANOG-hESCs into extraembryonic tissues(see below). These results demonstrate that BMP signalling is capable of blocking the neuroectoderm specification of hESCs but that this occurs only by promoting differentiation along the extraembryonic pathway. Moreover, the inhibition of BMPs using Noggin did not induce the expression of neuroectoderm markers in NANOG-hESCs (Fig. 2D) grown in the absence of Activin/Nodal signalling. These observations exclude the possibility that Nanog can block neuroectoderm differentiation through mechanisms involving BMP signalling. We then determined whether the induction of neuroectoderm markers induced by the knockdown of NANOG in hESCs was dependent on FGF signalling by growing shNanog-hESCs in CDM containing Activin and SU5402. The inhibition of FGF signalling strongly decreased the expression of SOX2, GBX2, SOX1, PAX6,HOXA1 and OLIG3 (Fig. 3E), confirming that FGF signalling is necessary for the expression of these neuroectoderm markers. However, the inhibition of FGF signalling did not restore pluripotency markers(Fig. 3E), confirming that FGF signalling is also involved in mechanisms controlling the pluripotency of hESCs. Taken together, these results demonstrate that Nanog is necessary to block the expression of neuroectoderm markers induced by FGF signalling, which is also necessary for hESC self renewal.

Genetic studies in the mouse have shown that the function of Nanog in mESCs and in the pre-implantation embryo is to block extraembryonic endoderm differentiation (Mitsui et al.,2003). To examine whether this function was conserved in hESCs, we analysed the effect of NANOG overexpression on extraembryonic differentiation induced by BMP4 (Xu et al.,2002) (our unpublished results). After 7 days of culture in the presence of BMP4, NANOG-hESCs adopted a homogeneous, broad cellular morphology typical of extraembryonic differentiation induced in these conditions (data not shown). BMP4-treated NANOG-hESCs became OCT4 negative(Fig. 4A,B; see also Fig. S5A-C in the supplementary material), indicating that Nanog was not sufficient to maintain pluripotency in the presence of BMP4. These observations were confirmed by FACS analyses determining the proportion of undifferentiated hESCs in culture before and after BMP4 treatment. In control conditions, 90%of both wild-type and NANOG-hESCs were positive for the pluripotency marker TRA-1-60 (see Fig. S5B in the supplementary material), whereas after 7 days of BMP4 treatment only 25-35% of either wild-type or NANOG-hESCs remained positive for TRA-1-60 (Fig. S5B in the supplementary material). In addition,expression of markers for primitive endoderm (SOX7, GATA4, GATA6, AFP,H19) and trophectoderm (CDX2, EOMES, HAND1, αhCG)could be detected in Nanog-expressing cells, as well as in wild-type cells(Fig. 4B, see also Fig. S5C in the supplementary material). Immunostaining analyses showed that NANOG and these markers (CDX2, EOMES, GATA4, GATA6) were co-expressed in the same cells(Fig. 4A), further demonstrating that Nanog did not block extraembryonic differentiation. Importantly, NANOG-hESCs also differentiated into cells expressing extraembryonic markers when induced to form EBs in a medium containing FBS(see Fig. S6A,B in the supplementary material). The extraembryonic outcome was thus independent of the culture system used to induce the differentiation of NANOG-hESCs. Finally, similar results were obtained with Nanog-mEpiSCs grown in CDM in the presence of BMP4 (see Fig. S5D,E in the supplementary material). Taken together, these results show that Nanog functions specifically to prevent neuroectoderm differentiation, rather than acting as a general blocker of differentiation.

We then analysed the effect of NANOG overexpression on mesendoderm/endoderm differentiation by using an approach that mimics in vivo development (our unpublished results). The first step of this protocol induces the differentiation of hESCs into mesendoderm cells expressing brachyury, Mixl1 and eomesodermin, and the second step drives the differentiation of these progenitors into definitive endoderm cells that nearly homogenously (∼90%of cells) express SOX17, CXCR4, MIXL1 and GSC(Fig. 5A-C; data not shown). Gene expression profiling analysis of hESCs cultured under these conditions confirms that they express a large number of known definitive endoderm markers, including GSC, LIM1, GATA4, GATA6 and FOXA2,whereas the expression of extraembryonic tissue markers (SOX7, AFP)and neuroectoderm markers (SOX1, SOX2) cannot be detected (data not shown). hESCs constitutively expressing NANOG that were grown in these conditions maintained the expression of OCT4 at levels similar to those in pluripotent cells while showing a limited induction of expression of endoderm markers (GSC, MIXL1, SOX17; see Fig. 5A,B). In addition, FACS analysis revealed that 90% of the wild-type cells expressed the definitive endoderm marker CXCR4 after differentiation compared with only 30% of the NANOG-expressing cells,suggesting that constitutive expression of NANOG could deter endoderm differentiation (Fig. 5C). However, the expression of SOX2 was strongly diminished in NANOG-hESCs subject to this protocol, whereas expression of the mesendoderm markers brachyury and eomesodermin was induced (Fig. 5B). Immunostaining analyses confirmed that NANOG protein was systematically co-expressed in the same cells as brachyury(Fig. 5A), indicating that NANOG did not prevent the onset of expression of mesendoderm markers but suggesting that it was able to interfere with further progression to definitive endoderm. This interpretation is not contradicted by the maintenance of OCT4 expression in NANOG-hESCs, as Oct4 (like Nanog) is also expressed in the mesendoderm of gastrulating mouse embryos(Hart et al., 2004). Furthermore, permissiveness for mesendoderm but not endoderm differentiation in NANOG-hESCs is also reinforced by the induction of the mesendoderm marker PDGFRα and by the relatively low number of NANOG cells expressing CXCR4(Fig. 5C). Finally, similar results were observed when Nanog-mEpiSCs were grown in endoderm-inducing culture conditions (see Fig. S7A,B). Taken together, these results indicate that rather than preventing mesendoderm specification in hESCs and mEpiSCs,Nanog limits the progression of mesendoderm progenitors towards definitive endoderm cells.

Studies in amphibians and in mice have shown that high activity of Activin/Nodal signalling is necessary to specify the endoderm germ layer(Dunn et al., 2004). In addition, BMP signalling has been shown to be essential for mesendoderm specification (Davis et al.,2004; Fujiwara et al.,2002). Therefore, the effect of Nanog on mesendoderm progression could involve modulating the activity of these signalling pathways. To address this hypothesis, NANOG-overexpressing cells were grown in culture conditions inductive for endoderm differentiation in the presence of increasing doses of Activin, BMP or FGF2 (Fig. 5D). A high dose of Activin, BMP or FGF was not sufficient to restore normal levels of endoderm markers (SOX17, GSC) in NANOG-hESCs, while expression of the mesendoderm marker brachyury was maintained in all these conditions(Fig. 5D). These observations show that an increase in extracellular factors cannot bypass the inhibitory effect of Nanog, and thus that Nanog could interfere directly with intracellular components of the Activin or BMP pathways. Interestingly, Nanog has been shown to interact directly with Smad1 to modulate mouse ESC differentiation (Suzuki et al.,2006). However, we have been unable to detect any direct interaction between Smad1 and Nanog in hESCs (data not shown), and we have shown that Nanog overexpression does not block the inductive effects of BMP extraembryonic differentiation. Therefore Nanog is unlikely to effectively regulate hESC or mEpiSC differentiation through direct interaction with Smad1,as has been reported for mESCs (Suzuki et al., 2006). However, Nanog protein itself shares homology with the common (C-) Smad4 protein (Hart et al.,2004) and thus Nanog might interact with Smad2/3. To test this hypothesis, the SBE4 luciferase reporter(Jonk et al., 1998) for Activin/Nodal-induced, Smad2/3-mediated signalling was transfected into hESCs(with or without SMAD2/3 or/and NANOG expression vectors) in the presence of increasing doses of Activin. This analysis showed that overexpression of NANOG decreased (but did not abolish) the transcriptional activity associated with increasing doses of Activin and with increased levels of SMAD2 (data not shown) or SMAD3 (Fig. 5E). This result suggests that NANOG overexpression is able to modulate the Activin/Nodal signalling pathway in hESCs. Interestingly, a similar Nanog-mediated modulation of Smad3-induced transcriptional activity was observed in mEpiSCs (see Fig. S8A in the supplementary material). Together,these observations suggest the existence of a negative-feedback loop in which Activin/Nodal-induced Nanog limits the transcriptional activity induced by Activin/Nodal signalling. To understand the mechanism of this interaction, we performed co-immunoprecipitation of NANOG or SMAD2/3, followed by western blot analyses. These revealed that the protein complexes containing NANOG also contained SMAD2/3 proteins (Fig. 3F). Importantly, these experiments were performed on nuclear extracts, and addition of SB431542 to the medium blocked the Smad2/3-Nanog interaction. This suggests that Nanog-complexed Smad2/3 must be phosphorylated and thus located in the nucleus, where it could affect the transcription of target genes. Similar Smad2/3-Nanog interactions were observed in mEpiSCs (see Fig. S8B in the supplementary material), suggesting that they are evolutionarily conserved. Finally, ChIP analyses showed that NANOG and SMAD2/3 proteins were capable of binding to the same genomic region in four different target genes of Activin/Nodal signalling, including LEFTYA, SMAD7,SnoN (SKIL - Human Gene Nomenclature Database) and NANOG itself (Fig. 1D,see also Fig. S1H in the supplementary material, data not shown), suggesting that the protein complexes containing NANOG and SMAD2/3 could regulate the promoters of target genes in hESCs. Interestingly, constitutive expression of NANOG during mesendoderm differentiation of hESCs limits the upregulation of known Smad2/3 target genes (LEFTYA and NODAL) to levels equivalent to those in hESCs (Fig. 5B). Consequently, the negative-feedback loop involving Activin/Nodal-mediated Nanog expression could act by limiting, but not entirely blocking, the transcriptional activity of Smad2/3.

The results presented here demonstrate that the Activin/Nodal signalling pathway directly controls the expression of Nanog in hESCs and in mEpiSCs, and thereby blocks their differentiation towards the neuroectoderm pathway. In turn, Nanog interacts directly with the Smad2/3 proteins to modulate their transcriptional activity. In this way, Nanog also blocks the progression of mesendoderm differentiation towards endoderm. This negative-feedback loop(Fig. 6) enforces stasis in both neuroectoderm and mesendoderm differentiation, resulting in the pluripotency of hESCs and mEpiSCs. This model represents a first step towards understanding the mechanisms by which Activin/Nodal signalling maintains the pluripotent status of hESCs and mEpiSCs, and also of pluripotent cells in the post-implantation mouse embryo. It also raises several questions concerning the mechanisms controlling early cell-fate specification. Indeed, although our results reveal the mechanisms by which Activin/Nodal signalling maintains pluripotency (i.e. by the direct control of Nanog expression), they do not explain how Activin/Nodal signalling can both maintain pluripotency and induce differentiation towards mesendoderm. However, our recent studies (our unpublished results) have demonstrated that BMP4 in combination with Activin is sufficient to drive the differentiation of hESCs into mesendoderm and then into endoderm in a chemically defined medium, suggesting that BMP4 is capable of converting Activin/Nodal signalling into an inductive signal for differentiation. Further studies will be necessary to understand the molecular interactions between these two signalling pathways during early cell-fate specification. In addition, a decrease in Nanog expression in the presence of Activin/Nodal signalling might be expected to result in mesendoderm differentiation. However, knockdown of Nanog instead increases the expression of neuroectoderm markers, suggesting that other proteins besides Nanog block the positive effect of the Activin/Nodal signalling pathway on mesendoderm differentiation.

The effect of Nanog knock down on hESC pluripotency reveals a mechanism distinct to that described for mESCs by Chambers and colleagues(Chambers et al., 2007). Their work showed that suppression of Nanog expression was not sufficient to induce full differentiation of mESCs, and suggested that Nanog acts to limit the probability of differentiation into multiple lineages induced by extracellular signals. Our results showing that Nanog blocks neuroectoderm differentiation in hESCs and mEpiSCs indicate a more lineage-specific role for Nanog than that in mESCs. Indeed, Nanog overexpression blocked neither mesendoderm differentiation nor BMP-induced extraembryonic differentiation in hESCs and mEpiSCs. Interestingly, the neuroectoderm markers analysed in our study have been shown to be direct target genes of Nanog in hESCs (SIX1, GBX2, PAX6,HOXA1, OLIG3) (Boyer et al.,2005), suggesting that Nanog could block the activity of transcription factors downstream of FGF signalling (as it does with Smad2/3 for Activin/Nodal signalling). Together, these observations suggest that Nanog acts in hESCs by blocking differentiation induced by signalling pathways that are also necessary for hESC pluripotency and self renewal (i.e. Activin/Nodal and FGF).

Therefore, the generalised function of Nanog in pluripotent stem cells is to safeguard pluripotency against the differentiation-inducing effects of essential extracellular signals. Importantly, however, Nanog does not protect hESCs, mEpiSCs and mESCs against the same differentiative events. Indeed,gain- and loss-of-function studies have shown that Nanog primarily blocks primitive endoderm differentiation of mESCs in vitro(Hamazaki et al., 2004) and of mouse inner cell mass in vivo (Mitsui et al., 2003). Conversely, Nanog blocks neuroectoderm and definitive endoderm differentiation of hESCs and of mEpiSCs. In vivo studies reinforce our results, as mouse embryos mutant for Nanog fail to develop beyond the late epiblast stage (Hamazaki et al.,2004), when Nanog expression in the epiblast becomes dependent on Nodal signalling (Mesnard et al.,2006; Mitsui et al.,2003). In addition, absence of Nodal expression and consequently of Nanog expression in post-implantation mouse embryos results in neuralisation of the epiblast, suggesting that Nanog acts in vivo to prevent neuroectoderm differentiation of the mouse epiblast before gastrulation(Camus et al., 2006), thus reinforcing our findings on its role in hESCs and mEpiSCs. The apparent similarity in the role of Nanog in hESCs, mEpiSCs and the pluripotent epiblast(and the dissimilarity of its role in mESCs) reinforces the hypothesis that hESCs and mESCs represent distinct stages of early mammalian development(Brons et al., 2007; Tesar et al., 2007). In this hypothesis, hESCs closely resemble pluripotent cells from post-implantation stages in vivo, in contrast to mESCs, which closely resemble pluripotent cells from the inner cell mass (Nichols et al.,2001). This hypothesis also implies that the functions of pluripotency factors change progressively during early mammalian development. Before implantation, the core pluripotency transcription factor circuit blocks formation of the extraembryonic lineages, whereas after implantation it blocks formation of the primary germ layers.

In this context, a recent study by Smith and colleagues(Ying et al., 2008) showed that general repression of differentiation signals by small molecules results in a `ground state of pluripotency' in mouse ESCs. This model does not appear to apply to hESCs and mEpiSCs, which strictly depend on Activin/Nodal signalling to maintain the expression of Nanog and thereby to maintain their pluripotent state. Therefore, two or more distinguishable pluripotent states relying on different growth factors but on similar core transcriptional networks appear to exist in vitro and in vivo during embryonic development. Interestingly, we recently observed that human induced pluripotent stem cells rely, like hESCs, on Activin/Nodal signalling to maintain their pluripotent state (our unpublished results). Therefore, understanding the function of pluripotency factors in each of these pluripotent states will be crucial to achieving control over the differentiation of human pluripotent cells, whether derived from mammalian embryos or by inducing pluripotency in cells of somatic origin (Takahashi et al.,2007; Yu et al.,2007).