Otx2 is an intrinsic determinant of the embryonic stem cell state and is required for transition to a stable epiblast stem cell condition

Understanding the genetic and epigenetic mechanisms that control the initial state and differentiation capability of pluripotent stem cells is essential for the comprehension of mammalian development and for the design of experimental protocols for the controlled generation of cell types of therapeutic interest (Hanna et al., 2010; Rossant, 2008; Murry and Keller, 2008; Niwa, 2007). Mouse embryonic stem cells (ESCs) have been isolated from the inner cell mass (ICM) of the blastocyst, may generate chimeric embryos at high efficiency, and their undifferentiated state depends on a self-maintaining network of core transcription factors [Oct4 (Pou5f1), Sox2, Nanog and Klf4/2/5] and signaling pathways (LIF, WNT and BMP4), which ensure self-renewal and protection from FGF-mediated lineage commitment (Silva and Smith, 2008; Lanner and Rossant, 2010; Hanna et al., 2010; Silva et al., 2009; Chambers et al., 2007; Niwa, 2011; Niwa et al., 2009; Ying et al., 2008; Ying et al., 2003a; Wray et al., 2011; Yi et al., 2011; Lyashenko et al., 2011; ten Berge et al., 2011; Nichols et al., 2009; Matsuda et al., 1999). The ESC population is not homogeneous, but rather is characterized by cell-to-cell spontaneous and reversible differences in the expression level and sensitivity to specific transcription factors and signaling pathways, which together maintain the balance between self-renewal and susceptibility to differentiation. This condition has been defined as the metastable state of ESCs (Niwa et al., 2009; Miyanari and Torres-Padilla, 2012; Niwa, 2007; Silva and Smith, 2008; Silva et al., 2009; Chambers et al., 2007; Toyooka et al., 2008).

Mouse epiblast stem cells (EpiSCs) are derived from the epiblast of pre-streak embryos, express Oct4, Sox2 and Nanog (at a low level compared with ESCs) but are silent for Klf factors, exhibit specific markers such as Fgf5 and brachyury (T), are highly inefficient in chimera formation and their self-renewal and undifferentiated states depend on FGF and activin A signaling pathways (Hanna et al., 2010; Lanner and Rossant, 2010; Brons et al., 2007; Tesar et al., 2007; Najm et al., 2011). However, most of these signaling factors have several functions; for example, FGF signaling is also required to inhibit neural differentiation of EpiSCs and to prevent their reversion to a preimplantation ESC-like state, and BMP4 is required to suppress neural fate and allow differentiation toward non-neural lineages (Lanner and Rossant, 2010; Greber et al., 2010; Greber et al., 2011; LaVaute et al., 2009; Kunath et al., 2007; Zhang et al., 2010; Di-Gregorio et al., 2007; Linker and Stern, 2004).

The ESC state corresponds to the naïve ground state of the preimplantation epiblast, whereas the EpiSC state corresponds to the primed state of the postimplantation epiblast (Hanna et al., 2010; Niwa, 2007; Lanner and Rossant, 2010; Nichols and Smith, 2009), thus implying that ESCs should be converted to a primed epiblast-like state before definitive differentiation into germ layers occurs. However, despite significant advancements, the regulatory mechanisms that control the state and differentiation capability of ESCs and EpiSCs remain to be fully clarified. Here, we have studied whether the transcription factor Otx2, which is required at multiple steps in brain development and neuronal differentiation (Simeone et al., 1992; Simeone et al., 2002; Simeone et al., 2011), might be functionally relevant in ESCs and EpiSCs, where it is expressed. We found that Otx2 is crucially required to maintain the ESC metastable state by opposing self-renewal and predisposing the cells to differentiation. Moreover, Otx2 stabilizes the transition from naïve ESCs to primed EpiSCs in cooperation with BMP4 and Fgf2 and is required for telencephalic and mesencephalic differentiation of ESC-derived neural progenitors.

The ESC lines were generated in E14Tg2a cells. In the Otx2^–/– ESC line, the coding exons of Otx2 alleles have been replaced with lacZ and GFP (Acampora et al., 1995; Acampora et al., 2009) (supplementary material Fig. S1A-C). In the R26^Otx2/Otx2 ESC line, a cassette including full-length Otx2 coding cDNA followed by an IRES GFP sequence is inserted into the Rosa26 locus (Di Salvio et al., 2010) (supplementary material Fig. S1D-G). The Otx2^–/–;R26^GFP/+ ESC line carries an Otx2^floxed-out null allele and the Otx2^lacZ null allele; in addition, the GFP gene was inserted into the Rosa26 locus for chimerism studies (supplementary material Fig. S1H-O). The R26^GFP/+ ESC line carries the GFP gene in the Rosa26 locus and was used as control for chimerism experiments (supplementary material Fig. S1M-O). The Otx2^flox/–;R26^CreER/+ ESC line carries an Otx2^flox allele in combination with the Otx2^lacZ null allele (Otx2^flox/–). An inducible CreER-puro cassette was introduced into the Rosa26 locus (supplementary material Fig. S1P-S).

Serum-free floating embryoid body-like aggregates (SFEBs) were generated from ESCs as reported (Watanabe et al., 2005). Mesendoderm differentiation in adherent conditions was obtained under serum-free conditions in the presence of activin A for 4 days (d) (Waese and Stanford, 2011; Izumi et al., 2007). For neural differentiation, SFEBs were cultured up to d5 without exogenous factors or with Dkk1 (from d1 to d5) (R&D Systems; 500 ng/ml), followed by adherent culture for 5 more days (Watanabe et al., 2005); alternatively, neural differentiation was induced using adherent monoculture conditions (Ying et al., 2003b).

Otx2^–/– ESCs were transfected using Lipofectamine 2000 reagent (Invitrogen) with a plasmid overexpressing Otx2 cDNA under a CMV-chicken β-actin promoter (CAG). Six hours after transfection, ESCs were seeded in ESC medium or differentiated into SFEBs.

ESCs were seeded at a density of 3×10³/cm² in N2B27 supplemented with 20 ng/ml activin A (R&D Systems) and 12 ng/ml Fgf2 (Peprotec) and cultured for 6 days (Brons et al., 2007; Tesar et al., 2007; Zhang et al., 2010), which corresponded to passage (p) 0. For further passages, EpiSCs were cultured in the same conditions. For Otx2 conditional inactivation, Otx2^flox/–;R26^CreER/+ EpiSCs were administered with 4-hydroxy-tamoxifen (4-OHT) (125 nM) at p1 and kept in 40 nM 4-OHT throughout subsequent passages. Epiblasts were dissected from single embryonic day (E) 5.75 mouse embryos and incubated in EpiSC medium.

LIF and FGF response was monitored as previously described (Mitsui et al., 2003; Zhang et al., 2010). For experiments involving BMP4 (recombinant human; R&D Systems), the factor was administered on d1 to ESC- or EpiSC-derived SFEBs (Zhang et al., 2010) at 0.5 and 10 ng/ml; SFEBs were then cultured up to d5. EpiSCs induced with Fgf2 and activin A were also administered with BMP4 (10 ng/ml) at p1 and until p4. Experiments involving activation or inhibition of the FGF and BMP pathways were performed in three conditions: (1) ESCs kept in 15% knockout serum replacement (KSR) medium with LIF were administered with BMP4 (10 ng/ml) or Fgf2 (12 ng/ml) or the BMP inhibitor dorsomorphin (DM) (0.5 μM; Calbiochem) or the Fgf/Erk inhibitor PD325901 (PD) (1 μM; Calbiochem) or with both inhibitors for 12 hours; (2) ESCs kept in ESC medium were cultured for the last 12 hours in 5% KSR medium plus BMP4 or Fgf2 or their inhibitors; (3) EpiSCs in N2B27 containing Fgf2 and activin A were cultured for 12 hours in N2B27 plus activin A only or also supplemented with BMP4 or DM, PD, or DM plus PD.

ALP staining was performed on EpiSCs at p0 and later passages or on ESCs in three different growth conditions at clonal density (1×10³ cells/10 cm²): in ESC medium (with LIF), without LIF, or without LIF plus JAK inhibitor 1 (0.6 μM; Calbiochem). ALP activity was revealed by the naphthol/Fast Red Violet reaction.

SFEBs treated with BMP4 (10 ng/ml) from d1 were trypsinized at d2 and reseeded in EpiSC medium (without BMP4) at a cell density of 1×10⁵/10 cm². Immunostaining for Oct4 was performed on d6 to unambiguously identify EpiSC colonies.

Chimeras were obtained by injecting 13-16 ESCs or Otx2^–/– EpiSCs into C57BL/6 blastocysts and colonization was evaluated by assessing GFP expression on E11.5, E9 or E8. Teratomas were generated by subcutaneous injection of 1.5×10⁶ ESCs or EpiSCs into NOD SCID mice. For Otx2^–/– ESCs, teratoma occurrence was also assayed by injecting 6×10⁶ ESCs.

SFEBs and postimplantation mouse embryos were embedded in paraffin and processed as described (Acampora et al., 2009). For immunohistochemistry on ESCs and EpiSCs cultured on Permanox chamber slides, paraformaldehyde-fixed cells were blocked in 2% skimmed milk powder and 0.3% Triton X-100. Morulae and blastocysts were blocked in 1% BSA, 10% FBS and 0.2% Triton X-100. Antibodies raised in rabbit were directed against Otx2 (1:3500), Sox2 (1:500) (both gifts of G. Corte, CBA, Geneva), Nanog (1:600; Calbiochem), Foxa2 (1:1000; Abcam), Foxg1 (1:500) and Eomes (1:1000) (Chemicon); antibodies raised in goat were against Otx2 (1:100; R&D Systems), Sox1 (1:100; Santa Cruz Biotechnology), T (1:150; Santa Cruz Biotechnology) and GFP (1:200; Abcam); antibodies raised in mouse were against Oct4 (1:250; Santa Cruz Biotechnology), Cdx2 (1:5; BioGenex), nestin (1:1000; Chemicon), Pax6 (1:150; Developmental Studies Hybridoma Bank), Nkx2.1 (1:200; Santa Cruz Biotechnology), αMHC (1:500; Developmental Studies Hybridoma Bank), Gfap (1:200; Chemicon) and Tuj1 (Tubb3 – Mouse Genome Informatics) (1:5000; Covance). Propidium iodide or DAPI counterstaining was performed as necessary.

RT-PCR was performed in non-saturating conditions using the primers and cycles listed in supplementary material Table S1. Western blots were probed with rabbit antibodies against Erk1,2 (Mapk3/1 – Mouse Genome Informatics) (1:1500), p-Erk1,2 (1:350), Stat3 (1:1000), p-Stat3 (1:350), Smad1 (1:1000), p-Smad1,5,8 (1:350) (Cell Signaling); β-actin antibody (1:6000; Sigma) was raised in mouse.

Cell counting was performed manually on immunohistochemistry images printed in A4 format or using ImageJ software (NIH). Standard deviation was calculated from four independent experiments.

In preimplantation mouse embryos, Otx2 was expressed in late morula blastomeres co-expressing Nanog or the trophectoderm determinant Cdx2 and, during blastocyst development, Otx2 colocalized with Cdx2 in trophectoderm and with Nanog in a fraction of cells of the ICM (Fig. 1A). Upon embryo implantation, robust Otx2 expression was detected in epiblast and visceral endoderm. At E6.5, Otx2 was downregulated in posterior epiblast, where the second wave of Nanog expression was activated (Fig. 1B). In ESCs, 46% of the Oct4⁺ cells co-expressed Otx2, whereas only a fraction of Oct4⁺ cells colocalized with Nanog and Klf4 (Fig. 2A-C). Cell counting on cytospun ESCs showed that Otx2⁺ cells were almost equally distributed between those expressing high or moderate levels of Nanog and those with low or no expression of Nanog (Fig. 2D-G; supplementary material Table S2). Moreover, ESCs exhibiting high levels of Otx2 preferentially expressed low levels of Nanog, whereas those with low levels of Otx2 exhibited high Nanog expression (Fig. 2D-G). In EpiSCs, Otx2 was detected in all Oct4⁺ cells at a level that varied between cells and was frequently complementary to that of Nanog (Fig. 2H). These data indicate that Otx2 is a molecular correlate of different pluripotent cell types in vivo and in vitro.

ESC differentiation may be activated by LIF withdrawal and/or low concentrations of serum, and early postimplantation embryonic development can be mimicked by ESC-derived embryoid bodies. In these culture conditions, Otx2 expression was highly responsive to ESC differentiation. Indeed, LIF withdrawal and/or diminished concentrations of KSR medium induced a rapid and generalized activation of Otx2 and a corresponding reduction in the number of Nanog⁺ ESCs (supplementary material Fig. S2A-D). Similarly, in differentiating SFEBs, Otx2 rapidly spread to all the Oct4⁺ cells, mirrored the downregulation of Nanog, Klf4 and Rex1 (Zfp42 – Mouse Genome Informatics), and anticipated the induction of the epiblast markers Fgf5 and Cer-l (Cer1 – Mouse Genome Informatics) and the generation of T⁺ primitive streak-like and Sox1⁺ neural cells (supplementary material Fig. S2E-H, Table S3). These observations support the possibility that Otx2 is required in ESCs to control their state and/or promote their differentiation.

To investigate the role of Otx2 we generated mutant ESC lines that lack (Otx2^–/–) or ubiquitously and constitutively express (R26^Otx2/Otx2) Otx2 (supplementary material Fig. S1). First, we studied functional parameters and markers related to the undifferentiated state or predisposition to differentiation. Compared with wild type (wt), virtually all the Otx2^–/– ESC colonies exhibited a sphere-like morphology, uniform ALP staining, ubiquitous distribution of Nanog and Klf4 and higher expression of Rex1 (Fig. 3A-D; supplementary material Tables S4, S5). During the earliest stages of wt ESC colony formation, Otx2 and Nanog exhibited complex expression profiles with variable degrees of complementarity, whereas Otx2^–/– ESC colonies exhibited constitutively high expression of Nanog (supplementary material Fig. S3). These data suggest that, in the absence of Otx2, the fluctuating expression of Nanog is severely affected and, therefore, that Otx2 is directly or indirectly required to prevent its constitutive expression.

In Otx2^–/– ESCs, the endogenous activities of LIF and FGF were respectively enhanced and severely decreased, as monitored by the level of the phosphorylated, active forms of the LIF signaling transcriptional mediator Stat3 (p-Stat3) and the extracellular signal-related kinases 1 and 2 (p-Erk1,2) (Fig. 3E,F). Furthermore, compared with wt, LIF addition to LIF-deprived ESCs generated a substantially higher response in Otx2^–/– ESCs (Fig. 3E), and LIF deprivation and blockade of Stat3 phosphorylation by JAK inhibitor 1 were not sufficient to disrupt the undifferentiated state of Otx2^–/– ESCs (Fig. 3G,H; supplementary material Table S4). Since this phenotype may also depend on loss of autocrine Fgf4 activity (Kunath et al., 2007) or abnormal expression of NuRD complex components (Reynolds et al., 2012; Kaji et al., 2006), the expression of Fgf4, Fgfr1-4, Mbd3, Mta1 and Mta2, Rbap46 (Rbbp7 – Mouse Genome Informatics) and Hdac1 was analyzed, and showed a very mild reduction in Otx2^–/– ESCs only for Fgf4 (supplementary material Fig. S4).

Chimerism was studied using a second, independent Otx2^–/– ESC line carrying a GFP constitutive reporter in the R26 locus (supplementary material Fig. S1). Lack of Otx2 only impaired chimerism in the forebrain and midbrain (Fig. 4A-C), whereas the frequency and growth of Otx2^–/– ESC-derived teratomas were substantially affected (Fig. 4D,E; supplementary material Tables S6, S7).

Next, we investigated whether ubiquitous and constitutive expression of Otx2 was sufficient to affect the ESC state by promoting differentiation. R26^Otx2/Otx2 ESCs exhibited a fairly flat morphology, a low frequency of uniformly ALP-stained colonies, a reduced percentage of ESCs co-expressing high levels of Nanog and Klf4, decreased expression of Rex1 (Fig. 3A-D), a heavily attenuated response to LIF and substantially enhanced activity of endogenous FGF (Fig. 3E,F). Remarkably, R26^Otx2/Otx2 ESCs exhibited the typical signature of primed EpiSCs, such as the expression of Fgf5 and Cer-l, generation of T⁺ cells (Fig. 3A,D) and very poor ability to generate chimeras, while retaining high efficiency in teratoma formation (Fig. 4A,C-E). Thus, Otx2 ubiquitous expression in ESCs is sufficient to induce stable molecular and functional features of the EpiSC state.

Collectively, these findings suggest that Otx2 is required to maintain the integrity of the ESC state by controlling the dynamic balance between pluripotency with high self-renewal and susceptibility to differentiation (Fig. 3I). In this context, Otx2^–/– ESCs resemble ESCs cultured in 2i, which exhibit severe downregulation of Otx2 (Marks et al., 2012).

We then studied Otx2 requirement in ESC differentiation by monitoring sequential steps marking pluripotency transition from ESCs to EpiSCs. First, we analyzed the response to Fgf2, which primes ESC differentiation. Compared with wt, Otx2^–/– ESCs howed a severe reduction in the p-Erk1,2 level, whereas in R26^Otx2/Otx2 ESCs, which already exhibited strong endogenous FGF activity, the response to Fgf2 was substantially higher (Fig. 5A). In the absence of any growth factor or cytokine, Otx2^–/– differentiating SFEBs showed after d1 a rapid loss of Oct4⁺ cells, failed to efficiently downregulate Nanog, Klf4 and Rex1, did not activate Fgf5 and Cer-l, and prematurely differentiated exclusively into Sox1⁺ neural cells co-expressing nestin and Sox2 (Fig. 5B-D; supplementary material Fig. S5A, Tables S3, S8). Conversely, R26^Otx2/Otx2 SFEBs showed a more graded decrease of Oct4⁺ cells, efficiently repressed Nanog, Klf4 and Rex1, expressed high levels of Fgf5 and Cer-l over the entire timecourse and differentiated prevalently into T⁺ primitive streak-like cells (Fig. 5B-D; supplementary material Fig. S5A, Tables S3, S8). Compared with d3.5 control SFEBs, these R26^Otx2/Otx2 SFEBs exhibited a substantial increase in the number of Sox1^– Foxa2⁺ cells (Fig. 5D), which are likely to correspond to T⁺ cells generated before d3.5 and differentiated into endoderm. Further mesendoderm determinants, such as Mixl1, Eomes, Gsc and Tbx6 (Izumi et al., 2007), showed that, as for T, their expression was virtually lost in Otx2^–/– SFEBs (supplementary material Fig. S5B). Thus, in the absence of any added factor, Otx2 is required in ESCs to promote their exit from the undifferentiated state and to protect committed EpiSCs from a premature neural fate.

To better evaluate the pluripotency and lineage potential of Otx2 mutant ESC lines, we analyzed Otx2^–/– and R26^Otx2/Otx2 ESC-derived teratomas (n=3 per genotype). Compared with wt, Otx2 mutant teratomas showed no obvious difference in the generation of neurons (neurofilament⁺), glial cells (Gfap⁺), muscle-like (αMHC⁺) and endoderm-like (Foxa2⁺) structures; however, Otx2^–/– teratomas exhibited a widespread abnormal distribution of Sox1⁺ nestin⁺ neural rosette-like and Oct4⁺ Nanog⁺ pluripotent-like cell aggregates (supplementary material Fig. S6), suggesting that Otx2 prevents the accumulation of neural progenitors and pluripotent-like cells.

To investigate whether the abnormalities described for Otx2^–/– ESCs were caused by a secondary adaptive effect triggered by Otx2 ablation, we first analyzed whether reintroduction of Otx2 is sufficient to rescue some of the major phenotypic abnormalities and, second, inactivated Otx2 in a 4-OHT conditionally inducible ESC line (Otx2^flox/–;R26^CreER/+) (supplementary material Fig. S1). Immunostaining of Otx2 and Nanog in Otx2^–/– ESCs transfected with a pCAG-Otx2 plasmid showed that most of the cells expressing high levels of Otx2 lacked or exhibited severely reduced expression of Nanog; moreover, d3.5 SFEBs generated after the pCAG-Otx2 transfection partially recovered the differentiation of T⁺ and Eomes⁺ cells (supplementary material Fig. S7A,B). In the second experiment, 48 hours after 4-OHT administration the Otx2^flox/–;R26^CreER/+ ESC colonies showed a compact morphology, ubiquitous expression of Nanog, increased levels of endogenous p-Stat3 and a moderate decrease in p-Erk1,2; SFEBs derived from 4-OHT-treated ESCs almost completely lacked T⁺ and Eomes⁺ cells (supplementary material Fig. S7C-E). Thus, the major phenotypes of Otx2^–/– ESCs are not due to adaptive and irreversible effects triggered by Otx2 inactivation.

To investigate which, if any, of the Otx2 requirements described in differentiating SFEBs might be detected in vivo, we revisited the phenotype of Otx2^–/– embryos. Otx2^–/– embryos showed widespread derepression of Nanog in the epiblast of pre-gastrula and gastrulating embryos (Fig. 6A-C), while the generation of T⁺ cells, although delayed, was still efficient and premature differentiation of Sox1 neural progenitors was not observed (Fig. 6; data not shown) (Acampora et al., 2009; Acampora et al., 1995; Ang et al., 1996). We hypothesized that in vivo, with the exception of Nanog, embryonic extrinsic signals might circumvent abnormalities detected in Otx2^–/– differentiating SFEBs. We examined whether Otx2^–/– ESCs may be converted into EpiSCs when provided with Fgf2 and activin A, which promote ESC transition into EpiSCs and maintain their undifferentiated state. Moreover, FGF activity is required in EpiSCs also to suppress neuroectoderm differentiation and prevent regression to a preimplantation ESC-like state (Lanner and Rossant, 2010; Greber et al., 2010; Greber et al., 2011; Kunath et al., 2007).

Otx2^–/– EpiSC colonies at p0, which corresponds to d6 of differentiation, showed spotted ALP staining and a mild increase in Sox1 expression but fairly normal expression of Nanog, Fgf5, Cer-l and T (supplementary material Fig. S8A-C). Suspecting these abnormalities as initial signs of a much more serious instability, Otx2^–/– EpiSCs were analyzed at subsequent passages. Strikingly, at p5, Otx2^–/– EpiSCs showed intense ALP staining and high levels of Nanog, mild reduction of Fgf5 and Cer-l transcripts in Oct4 normalized RNA samples, increased generation of Sox1⁺ nestin⁺ cells and a corresponding decrease of T⁺ cells (Fig. 7A,B). This phenotype worsened severely over time and, at p10, the great majority of Otx2^–/– EpiSC colonies showed domed or sphere-like morphology with uniform ALP staining, strong Nanog immunoreactivity, severe decrease in Fgf5 and Cer-l expression, almost complete extinction of T⁺ cells, substantial reduction in the expression of Eomes, Mixl1, Gsc and Tbx6, and massive derepression of the neural lineage (Fig. 7A,B). However, p12 Otx2^–/– EpiSCs can be propagated and maintained in EpiSC medium, thus excluding the possibility that they might have regressed to LIF dependence (data not shown). Accordingly, p5 Otx2^–/–;R26^GFP/+ EpiSCs failed to generate chimerism in E9 embryos (n=24; data not shown). However, p0 and p5 Otx2^–/– EpiSCs were able to form teratomas and, as for those derived from Otx2^–/– ESCs, they generated neuronal, glial, muscle-like and endodermal derivatives and numerous Sox1⁺ nestin⁺ rosette-like neural progenitors and Oct4⁺ Nanog⁺ cell aggregates (supplementary material Fig. S9, Table S7).

Next, we analyzed the effect of Otx2 ablation or constitutive expression following a different protocol using only activin A for mesendoderm differentiation (Waese and Stanford, 2011; Izumi et al., 2007), and found that lack of Otx2 primarily affected the expression of Fgf5, Eomes, Mixl1, Gsc and Foxa2 and, interestingly, induced Sox1 and nestin in numerous Oct4⁺ patches; conversely, mesendoderm markers were all upregulated in R26^Otx2/Otx2 cells (supplementary material Fig. S10). Together, these data indicate that a high dosage of Fgf2 and activin A apparently compensates for the Otx2 requirement only for the initial transition of ESCs to primed EpiSCs, but not for maintenance of EpiSC identity. To assess whether Otx2 is an intrinsic stabilizer of the EpiSC state, we analyzed at p4 and p8 the phenotype of Otx2^flox/–;R26^CreER/+ conditional EpiSCs administered at p1 with 4-OHT, and found that they exhibited the same abnormalities described for Otx2^–/– EpiSCs at similar passages (Fig. 7C-E). Of note, Otx2^flox/–;R26^CreER/+ EpiSC colonies that were not treated with 4-OHT also showed mild impairments at p8, suggesting an Otx2 dosage requirement for maintenance of the EpiSC state (Fig. 7C-E).

Finally, we tested our hypothesis that, in vivo, extrinsic factors might circumvent the requirement for Otx2 in EpiSCs. EpiSCs isolated from the egg cylinder of Otx2^–/– embryos developed severe abnormalities that were undistinguishable from those of Otx2^–/– ESC-derived EpiSCs (Fig. 7F). These findings indicate that Otx2 is an intrinsic determinant required to stabilize the EpiSC state by repressing the progressive switch of mesendoderm to neural fate and suggest that factor(s) other than, or together with, Fgf2 and activin A are required to stabilize the EpiSC state in cooperation with Otx2.

Previous studies have demonstrated that in ESCs BMP4 maintains pluripotency and antagonizes neural fate (Ying et al., 2003a; Zhang et al., 2010). Moreover, during embryoid body development, BMP4 first inhibits ESC transition into EpiSCs and, subsequently, in EpiSCs represses neural differentiation and promotes non-neural fate (Zhang et al., 2010). We therefore studied potential functional interactions between Otx2 and BMP4 signaling. First, we administered a high dosage (10 ng/ml) of recombinant BMP4 at d1 to wt and Otx2^–/– SFEBs and analyzed at d5 Oct4⁺ Sox1^– (pluripotent), Oct4^– Sox1⁺ (neural) and Oct4^– Sox1^– (non-neural and non-pluripotent) cell compartments. In Otx2^–/– SFEBs, BMP4 efficiently repressed neural fate, but was unable to promote non-neural differentiation and caused an expansion of Oct4⁺ cells (Fig. 8A-C; supplementary material Fig. S11A, Table S9). Importantly, compared with wt, d5 Otx2^–/– BMP4-treated SFEBs showed increased numbers of Klf4⁺ Oct4⁺ ESC-like cells, high expression of Rex1 and low levels of Fgf5 transcripts among Oct4 normalized RNAs (supplementary material Fig. S11A,B). Accordingly, compared with wt, d2 Otx2^–/– BMP4-treated SFEBs generated many fewer EpiSC colonies when dissociated and cultured in EpiSC medium (Fig. 8D).

To assess the hierarchical position of Otx2 in this process, we analyzed its expression in response to BMP4 or Fgf2 and their inhibitors (DM and PD, respectively) in ESCs cultured in 15% KSR with LIF and in ESCs primed to differentiate in 5% KSR. The Otx2 response was also compared with that of Nanog. Fgf2 induced robust activation of Otx2 in ESCs cultured in 15% KSR with LIF and even stronger activation in those kept in 5% KSR; accordingly, treatment with the Fgf inhibitor PD generated the opposite effect, and treatment with BMP4 or its inhibitor DM revealed a moderate repressive effect of BMP4 on Otx2 expression (supplementary material Fig. S12A-C). The effects on Nanog expression mirrored those on Otx2. Together, these findings suggest that, during the initial ESC transition to EpiSCs, Otx2 antagonizes the inhibitory action of BMP4 through a positive loop with Fgf2 (supplementary material Fig. S11E).

Then, we tested whether BMP4 requires Otx2 to suppress neural fate in SFEBs, this time derived from p0 EpiSC colonies. Compared with wt, the BMP4 antineuralizing activity was less efficient in Otx2^–/– EpiSC-derived SFEBs (Fig. 8A-C; supplementary material Fig. S11C). We reasoned that if the inefficient neural suppression by BMP4 reflected lack of cooperation with Otx2, then the BMP4 anti-neuralizing activity should be potentiated in R26^Otx2/Otx2 ESC-derived SFEBs. Indeed, R26^Otx2/Otx2 SFEBs cultured with a very low concentration of BMP4 (0.5 ng/ml) that is unable to suppress neural fate in wt SFEBs, fully suppressed Sox1⁺ neural cells and generated almost exclusively Oct4^– Sox1^– cells (Fig. 8A-C; supplementary material Fig. S11D). Accordingly, when Otx2^–/– EpiSCs induced by Fgf2 and activin A also received BMP4 at p1, the derepression of neural fate was inhibited and EpiSC identity was maintained (Fig. 8E,F). This also suggests that, in vivo, the same compensatory mechanism might stabilize the identity of the epiblast of Otx2^–/– embryos up to gastrulation.

To gain more insight into this aspect, we investigated Otx2 expression in EpiSCs in response to Fgf2, BMP4 and their inhibitors. The expression of Otx2 detected in the presence of Fgf2 and activin A was unaffected by Fgf2 withdrawal or Fgf2 withdrawal and addition of BMP4 (supplementary material Fig. S12D,E). However, treatment with DM in the presence of activin A induced a moderate activation of Otx2, which was enhanced by PD or DM plus PD; of note, Otx2 activation correlated with downregulation of Nanog and Oct4 (supplementary material Fig. S12D,E). Thus, in contrast to ESCs, inactivation of FGF and/or BMP4 signaling in EpiSCs upregulates Otx2 and downregulates pluripotency factors. These data are similar to those reported in human ESCs showing that inhibition of FGF signaling is reflected in OTX2 activation, repression of NANOG and OCT4 and OTX2-mediated activation of PAX6 (Greber et al., 2011). Collectively, these experiments suggest that, in EpiSCs, Otx2 expression is maintained at a relatively low level at which it synergizes with Fgf2 and BMP4 to stabilize the EpiSC state by suppressing neural fate (supplementary material Fig. S11E).

Previous studies indicated the crucial role played by Otx2 in the specification, regionalization and differentiation of the rostral neuroectoderm (Simeone et al., 2002; Acampora et al., 2009; Simeone et al., 2011). We examined whether the identity of neural progenitors is affected by Otx2. First, we induced the differentiation of neural progenitors enriched in telencephalic precursors (Watanabe et al., 2005). Compared with wt, in Otx2^–/– neural cells, the expression level of the pan-telencephalic marker Foxg1 was severely diminished, even when telencephalic differentiation should be enhanced by treatment with the WNT antagonist Dkk1; conversely, in R26^Otx2/Otx2 neural progenitors the expression of Foxg1 was moderately increased (Fig. 9A). Otx2^–/– neural cells retained expression of En1 and En2, and essentially lacked that of telencephalic, diencephalic and mesencephalic markers; conversely, R26^Otx2/Otx2 neural cells showed increased expression of anterior neuroectoderm markers and attenuated expression of En1 and En2 (Fig. 9A).

To assess the subregional identity of presumptive telencephalic precursors we determined the percentage of Foxg1⁺ cells co-expressing the pallial marker Pax6 and the subpallial marker Nkx2.1. Wt Foxg1⁺ cells showed a prevalent subpallial (Nkx2.1⁺) identity, which was converted to pallial by Dkk1 treatment (Fig. 9B,C; supplementary material Table S10). Interestingly, the rare Otx2^–/– Foxg1⁺ patches exhibited a fully penetrant subpallial identity that was unresponsive to Dkk1, whereas R26^Otx2/Otx2 Foxg1⁺ cells showed a pallial (Pax6⁺) identity even in the absence of Dkk1 (Fig. 9B,C).

A different protocol of ESC differentiation into neuroectoderm precursors in adherent monoculture (Ying et al., 2003b) confirmed that lack of Otx2 results in decreased expression of telencephalic, diencephalic and mesencephalic markers and revealed increased expression of metencephalic and spinal cord markers; the opposite expression profile was observed in R26^Otx2/Otx2 ESC-derived neural cells (Fig. 9D). Thus, these findings indicate that, as in the embryo, Otx2 is crucially required to confer anterior character to ESC-derived neuroectoderm progenitors.

Numerous studies have shown that ESC metastability is defined by the opposing and dynamic action of specific signaling pathways and transcription factors (Hanna et al., 2010; Silva and Smith, 2008; Lanner and Rossant, 2010; Niwa, 2007). Signaling molecules such as LIF, but also BMP4 and WNT, together with the transcription factor Nanog, control ground state pluripotency by protecting ESCs from FGF-mediated differentiation commitment (Chambers et al., 2007; Niwa et al., 2009; Ying et al., 2008; ten Berge et al., 2011; Kunath et al., 2007; Niwa, 2011; Wray et al., 2011; Yi et al., 2011; Lyashenko et al., 2011). This antagonism generates in ESCs a continuum of fluctuating and interconvertible states which, if perturbed by genetic modifications or chemical inhibitors, may rapidly drift toward irreversible differentiation or the fully undifferentiated state (Hanna et al., 2010; Niwa, 2007; Silva and Smith, 2008; Lanner and Rossant, 2010; Guo et al., 2009).

In this study, we have investigated for the first time whether Otx2, which has been extensively investigated in the context of brain development, is required in ESCs and EpiSCs. Otx2 is expressed in a large subset of ESCs that defines a cell population showing graded transition from those expressing high levels of Nanog to those exhibiting low or no Nanog expression. ESC culture conditions promoting differentiation (LIF withdrawal and/or Fgf2 addition and/or BMP4 inhibition) induce rapid activation of Otx2, which is prevented by FGF inhibition. Based on these observations, we hypothesized that genetic ablation of Otx2, or its constitutive and ubiquitous expression, might affect the ESC state and corrupt the metastable condition. Lack of Otx2 indeed causes severe abnormalities consisting of ubiquitous and constitutive high level expression of Nanog and Klf4, increased LIF signaling and weakened activity of FGF signaling; conversely, constitutive and ubiquitous activation of Otx2 causes a substantial reduction in the number of ESCs co-expressing high levels of Nanog and Klf4, a robust increase of FGF activity, induction of epiblast markers and a very poor ability to generate chimeric embryos. These data strongly suggest that Otx2 is a novel intrinsic determinant of the ESC state, controlling the balance between self-renewal and differentiation. We propose that Otx2 protects the susceptibility to differentiation by counter-balancing signaling pathways and transcription factors that promote ground state pluripotency. Our findings suggest that Otx2 and Nanog, by promoting antagonistic conditions, together contribute to define the ESC metastable state (Fig. 3I). Of note, lack of Otx2, or its ubiquitous activation, generate ESC abnormalities that are similar to those exhibited by ESCs overexpressing or lacking Nanog, respectively (Silva and Smith, 2008; Chambers et al., 2007; Chambers et al., 2003; Mitsui et al., 2003).

This study shows that Otx2 is also required for ESC transition into EpiSCs and maintenance of the EpiSC state. In the absence of any added factor, Otx2^–/– ESCs fail the transition to EpiSCs and prematurely differentiate only into neural cells. When provided with Fgf2 and activin A, transition of Otx2^–/– ESCs to EpiSCs is initially achieved but maintenance of their state in terms of identity and fate is gradually lost through a progressive mesendoderm-to-neural fate switch accompanied by loss of EpiSC markers, strong activation of Nanog and intense ALP reactivity. Similar abnormalities, with early derepression of neural fate and selective impairment of the endoderm lineage, were also observed in Otx2^–/– ESCs induced only with activin A. Thus, Otx2 represents a key factor that is intrinsically required to maintain EpiSC identity, primarily by suppressing neural fate and preventing the adoption of features resembling naïve pluripotency. We speculate that Otx2^–/– EpiSCs might progressively regress to an early postimplantation-like epiblast stage, similar to that at which SFEBs prematurely derepress neural fate. In the absence of Otx2, maintenance of EpiSC identity and suppression of neural fate require a high dosage of both Fgf2 and BMP4 in the presence of activin A. Like Fgf2, BMP4 also plays a dual role, first opposing ESC transition into EpiSCs and then suppressing neural fate in EpiSCs (Zhang et al., 2010). Our data suggest that, during the first stage Otx2 antagonizes BMP4-mediated inhibition of EpiSC derivation and, subsequently, in EpiSCs Otx2 synergizes with BMP4 for suppression of neural fate (supplementary material Fig. S11E).

Otx2 has been extensively studied in the context of its essential role in the development and differentiation of anterior neuroectoderm. Our data indicate that Otx2 is crucially required in ESC-derived neural cells to specify the identity of telencephalic and mesencephalic territories and to allow chimerism in the neuroectoderm rostral to the isthmic organizer.

In summary, this study has revealed that Otx2 is a novel determinant controlling the ESC state and its transition to a stable EpiSC condition. Subsequently, as in vivo, Otx2 is required to specify the anterior identity of ESC-derived neural progenitors.

We thank G. Cossu, T. Perlmann, T. Russo and C. Stern for helpful discussion and constructive criticism; G. Gagliardi for typing and formatting the manuscript; and the confocal microscopy facility of CEINGE Biotecnologie Avanzate.