SOX9 accelerates ESC differentiation to three germ layer lineages by repressing SOX2 expression through P21 (WAF1/CIP1)

The identity of cells can be altered by the forced induction of combination of transcription factors (TFs) (Takahashi and Yamanaka, 2006; Vierbuchen et al., 2010; Ieda et al., 2010; Sekiya and Suzuki, 2011; Huang et al., 2011; Hiramatsu et al., 2011), the forced induction of single TFs (Davis et al., 1987; Nishiyama et al., 2009; Correa-Cerro et al., 2011; Yamamizu et al., 2013) or by the repression of single TFs (Skarnes et al., 2004; Ivanova et al., 2006; Collins et al., 2007; Nishiyama et al., 2013). As an aid to analyze the effects of TF manipulation on mouse embryonic stem cell (ESC) differentiation, we have established the NIA Mouse ESC Bank (Nishiyama et al., 2009; Correa-Cerro et al., 2011), in which each of 137 TFs, i.e. 7-10% of all TFs encoded in the mouse genome (Kanamori et al., 2004), can be induced in a tetracycline-regulatable manner. We have measured the global gene expression profiles (i.e. transcriptome) of these ESC lines 48 h after overexpressing each TF (Correa-Cerro et al., 2011; Nishiyama et al., 2009). By comparing these transcriptome data to the publicly available expression profiles of a variety of cell types (Su et al., 2002; Wu et al., 2009), we generated a correlation matrix that can help to predict the TF-induced direction of ESC differentiation (Correa-Cerro et al., 2011). Based on predictions, we have successfully directed cell differentiation into target organ cells such as myocytes, hepatocytes, blood cells and neurons (Yamamizu et al., 2013).

Here, we have attempted an alternative use of the transcriptome data sets obtained by overexpressing each of 137 TFs in mouse ESCs. We selected the 36 ESC lines that individually showed the greatest degree of transcriptome perturbations and analyzed their early differentiation. As we expected, most TFs direct the ESC differentiation into cells ordinarily derived from one of the embryonic germ layers, but Sry (sex determining region Y) box 9 (SOX9), a member of the Sry-related high-mobility group (HMG) box transcription factors, is an exception. SOX9 had already been shown to have pivotal roles in embryonic development of multiple organs, including testis, chondrocytes, heart, lung, pancreas, bile duct, hair follicles, kidney, inner ear, retina and the central nervous system (Stolt et al., 2003; Chaboissier et al., 2004; Vidal et al., 2005; Akiyama et al., 2005; Seymour et al., 2007; Furuyama et al., 2011). Recent studies have shown that Sox9 is expressed in progenitor cells of various organs and Sox9-expressing progenitors differentiate into multiple mature organ cells (Vidal et al., 2005; Seymour et al., 2007; Furuyama et al., 2011; Kadaja et al., 2014). However, molecular mechanisms for SOX9 function in progenitors are unclear. Here, we show that, in contrast to all other TFs tested, SOX9 can accelerate the differentiation of ESCs to cell types of all three germ layers. We have investigated the detailed molecular mechanism of this exceptional differential capacity of SOX9 in mediating ESC differentiation, and report the involvement of the Cdkn1a (P21/WAF1/CIP1)-Sox2 pathway.

Previously, we have reported global gene expression profiles of mouse ESC lines that were generated 48 h after overexpressing 137 TFs individually (Fig. 1A-C) (Nishiyama et al., 2009; Correa-Cerro et al., 2011). From a list of 137 TFs sorted by the magnitude of transcriptome perturbation, we arbitrarily selected the top 36 TFs (Fig. 1A-C) and analyzed systematically the differentiation into three germ layers using FACS, with FLK1, FOXA2 and PSA-NCAM as markers for mesoderm, endoderm and ectoderm, respectively. The ESC lines seemed to be differentiated into mixtures of cells of three germ layers, as these markers were not co-expressed in the same cells in most cases, according to the FACS and immunostaining analyses (supplementary material Fig. S1).

For mesoderm differentiation, overexpression of T (brachyury), which is known as a mesoderm inducer, increased the number of FLK1-positive mesoderm cells most efficiently after 5 days of differentiation (Fig. 1D,E). Sox9 is a second-ranked TF, the overexpression of which significantly increased the number of mesoderm cells from ESCs. Furthermore, we identified novel mesoderm inducers: Ascl2, Tcf3, Esx1 and Tcfap2 (Tfap2e – Mouse Genome Informatics). For endoderm differentiation, overexpression of Hnf4a and Foxa1, both of which are known as endoderm inducers, most efficiently increased the number of FOXA2-positive endoderm cells after 5 days of differentiation (Fig. 1F,G). Sox9 is the third-ranked TF, the overexpression of which efficiently increased endoderm cells from ESCs. For ectoderm differentiation, overexpression of Ascl1 most efficiently increased PSA-NCAM-positive ectoderm cells after 6 days of differentiation (Fig. 1H,I). Sox9 is the fourth-ranked TF, the overexpression of which significantly increased the number of ectoderm cells from ESCs. These results indicate the utility of our approach to identify potent TFs such as T, Hnf4a, Foxa1 and Ascl1 for lineage-specific cell differentiation. Furthermore, among 36 ESC lines examined here, Sox9 stood out as the only TF gene that ranked near the top of the list of genes necessary for the efficient differentiation of ESCs into all three germ layers.

Unique ability of SOX9 to efficiently direct differentiation into three germ layers prompted us to focus on Sox9. First, we confirmed Sox9 expression with qPCR and western blotting. In the control condition [doxycycline (Dox+)], endogenous Sox9 levels increased from 3 days after differentiation (D3), preceding the emergence of the three germ layers (around D4) (Fig. 2A,B). By contrast, withdrawal of Dox on D0 increased the expression of exogenous Sox9 from D1 (Fig. 2A,B). Immunostaining and FACS analyses showed that the Dox withdrawal on D0 markedly increased the proportion of cells expressing each of one of the three germ layer markers: FLK1 (KDR – Mouse Genome Informatics) and T (mesoderm); FOXA2 and SOX17 (endoderm); and NES (nestin) and PSA-NCAM (ectoderm) from D4 at comparable levels (Fig. 2C-G). Analyses by qPCR also showed that the forced overexpression of SOX9 increased the expression of mesoderm marker genes (Flk1, T and Pdgfra), endoderm marker genes (Foxa2, Sox17 and Gsc) and ectoderm marker genes (Nes and Fgf5) from D2 or D3 (Fig. 2H). These results indicate that the forced expression of SOX9 in early differentiating cells regulates the generations of three germ layers from ESCs.

To further examine ESC differentiation into organ cells, we used a lineage-specific medium with or without Dox from D3 to D7 (Fig. 3A). Under the control condition (Dox+), endothelial cells, hepatocytes and neurons did not appear even if cells were treated with lineage-specifying growth factors or cultured with lineage-specifying media (Yamamizu et al., 2012b,, 2013). By contrast, the overexpression of Sox9 combined with a treatment of vascular endothelial growth factor (VEGF) dramatically increased the proportion of endothelial cells stained with antibodies against CD31 and CDH5 (VE-Cadherin) by day 7 (Fig. 3B-D). Using hepatocyte-specific medium from D3, the overexpression of Sox9 rapidly and dramatically increased the proportion of the hepatocytes, detected by the production of albumin (ALB) with immunostaining, by Periodic acid-Schiff (PAS) staining and by the uptake of low-density lipoproteins (LDL) (Fig. 3D). Furthermore, as early as D7, ALB production was detected in Sox9-overexpressing cells, but not in the Dox+ control cells, which became more prominent by D14 (Fig. 3E). Use of neuron-specific medium increased the Sox9-mediated efficiency of neural differentiation (Fig. 3G). Neurons induced by Sox9 expressed a variety of neural markers: pan-neural markers (TUJ1 and MAP2); a dopaminergic neuron marker [tyrosine hydroxylase (TH)]; a motor neuron marker (ISL1/ISL2); and an inhibitory neurotransmitter (GABA) (Fig. 3H). Taken together, Sox9 accelerates the differentiation of ESCs into progenitors of the three germ layers, thereby enhancing generation of organ cells.

Next, we examined markers of undifferentiated ESCs during differentiation induced by the overexpression of Sox9. Under the control condition (Dox+), alkaline phosphatase (AP)-negative differentiated colonies appeared from D5. By contrast, the overexpression of Sox9 (Dox−) induced the appearance of AP-negative colonies from D4 (Fig. 4A,B). FACS analysis for a pluripotent marker, SSEA1, revealed that the overexpression of Sox9 significantly decreased SSEA1-positive undifferentiated cells from D3 (Fig. 4C,D). Immunostaining and qPCR analyses for pluripotent markers – POU5F1, NANOG, SOX2 and SSEA1 – also revealed an earlier disappearance of these proteins, when Sox9 is overexpressed. These results suggest that Sox9 could be involved in the reduction of pluripotency that occurs during ESC differentiation.

To investigate roles of endogenous Sox9 in early differentiating ESCs, we blocked Sox9 expression by treating cells with an siRNA against Sox9. As shown in Fig. 2B, endogenous Sox9 increased from D3, preceding the emergence of the three germ layers (around D4) (Fig. 2B). We treated cells with three independent siRNAs against Sox9 from D0 (Fig. 5; supplementary material Fig. S2). Analyses by FACS, immunostaining and qPCR clearly showed that the treatment with siRNAs against Sox9 significantly inhibited the expression of mesoderm markers (FLK1, T and PDGFRα), endoderm markers (FOXA2, SOX17 and GSC) and ectoderm markers (NES and FGF5). Furthermore, treating cells with siRNAs against SOX9 maintained the expression of pluripotency markers (SSEA1, POU5F1, NANOG and SOX2), which are otherwise downregulated, in early differentiating ESCs (Fig. 5C-O; supplementary material Fig. S2B,C). These results indicate that the expression of endogenous SOX9 in early differentiating cells is required for both the generation of three germ layers from ESCs and the repression of pluripotency.

As proliferation and differentiation are considered to be closely related to each other, we examined cell proliferation during SOX9-induced ESC differentiation. The overexpression of Sox9 significantly decreased cell numbers from D3 (Fig. 6A). FACS analysis also revealed that the cell population in G1 phase increased by Sox9 overexpression from D2 (Fig. 6B; supplementary material Fig. S3). Previous studies have shown that, although pluripotency is associated with a specific cell-cycle profile characterized by a shortened G1 phase, pluripotent stem cells exhibit a lengthening of the G1 phase during differentiation produced by G1-checkpoint genes, such as p21 and p27 (Becker et al., 2006; Egozi et al., 2007). Accordingly, we investigated the expression of genes related to the G1 phase. We found that in the Dox− condition (i.e. the overexpression of Sox9), the level of p21 was markedly increased from D2, whereas in the Dox+ condition (control), the level of p21 was increased from D4 (Fig. 6C). The increase in p21 levels from D4 is most likely caused by the upregulation of endogenous SOX9 expression from D3 (Fig. 2A), because blocking the activation of endogenous Sox9 with an siRNA against Sox9 inhibited the increase of p21 expression that normally occurs from D4 (supplementary material Fig. S4). The expression levels of p27, but not of p53 or Rb, were slightly increased by the overexpression of Sox9.

Western blotting analysis revealed an early increase of P21 and a reciprocal decrease of SOX2 levels from D2 caused by the overexpression of Sox9 (Fig. 6D,E). Decreases in levels of POU5F1 and NANOG occurred later than the decrease in the levels of SOX2. It has been shown that P21 can modulate transcription by directly binding to specific promoters in a cell cycle-independent manner (Devgan et al., 2005; Besson et al., 2008). A recent study has shown that P21 directly binds to a Sox2 enhancer and negatively regulates Sox2 expression in neural stem cells (Marqués-Torrejon et al., 2013). The main regulatory element, named SRR2, that is responsible for the expression of Sox2 in pluripotent stem cells is located ∼4 kb downstream of the single Sox2-coding exon (Fig. 6F) (Tomioka et al., 2002; Sikorska et al., 2008; Miyagi et al., 2004). To test the possibility of a direct binding of P21 to the SRR2 enhancer, we performed chromatin immunoprecipitation (ChIP) assays with an anti-P21 antibody. As expected, we observed a specific binding of P21 to the SRR2 enhancer of Sox2 on D5 under the control conditions (Dox+), whereas P21 did not bind to the UTR of Sox2 or the Nanog promoter. Interestingly, the overexpression of Sox9 (Dox−) significantly enhanced the direct binding of P21 to the SRR2 enhancer as early as D1 (Fig. 6G), which preceded the decrease of Sox2 and the accelerated differentiation into three germ layers.

To investigate whether the regulation of accelerated differentiation by Sox9 is mediated by P21, we blocked the expression of p21 by treating cells with a siRNA against p21. Western blotting and qPCR analyses showed that treating cells with a siRNA against p21 decreased the expression of p21 by 80% by D4 (Fig. 7A,B). Significant increases of mesoderm markers (T and FLK1), endoderm markers (SOX17 and FOXA2) and ectoderm markers (NES and PSA-NCAM) by Sox9 overexpression were almost completely abolished by the inhibition of p21 (Fig. 7C-K). However, early disappearance of pluripotency gene expression (SOX2 and SSEA1) induced by Sox9 overexpression was cancelled by inhibiting p21 (Fig. 7L-N). Furthermore, a ChIP assay revealed that direct binding of P21 in SRR2 enhancer of Sox2 genes was completely abolished by treatment of siRNA against p21.

Next, we investigated whether the forced expression of Sox2 on D2 can abolish the ESC differentiation induced by the overexpression of Sox9. Transfection of a plasmid vector expressing Sox2 significantly abolished Sox9-elicited acceleration of the differentiation into the three germ layers from ESCs and also cancelled the early disappearance of the expression of pluripotency-related genes (supplementary material Fig. S5). These results indicate that the overexpression of Sox9 accelerates the differentiation into the three germ layers from ESCs through the p21-Sox2 cascade (Fig. 7P).

Finally, we examined whether the knockout of Sox2 recapitulates phenotypes of Sox9 overexpression using an inducible Sox2-null ESC line. Previous studies have shown that the knockout or knockdown of Sox2 in ESCs differentiates ESCs into trophectoderm-like cells (Masui et al., 2007; Sharov et al., 2008; Nishiyama et al., 2013). Our results did not seem to be consistent with these well-established earlier findings, as the repression of SOX2 by the overexpression of SOX9 did not differentiate ESCs into trophoblast cells, but into three germ layers in our hands. Considering that the ability of early embryonic cells to give rise to the trophectoderm lineage is diminished at the epiblast stage of development, we hypothesized that the delay in Sox2 knockout is responsible for the generation of three germ layers, but not trophectoderm-like cells. To test this notion, we compared two different protocols of Sox2 knockout: for protocol 1 (Dox+, D0), Dox treatment is given from the start of differentiation (D0); for protocol 2 (Dox+, D2), Dox treatment is given from D2 (Fig. 8A-C). In protocol 1, the expression of Sox2 diminished on D1, whereas in protocol 2 the expression of SOX2 diminished on D3 (Fig. 8A-C). The knockout of Sox2 significantly decreased cell numbers during ESC differentiation (supplementary material Fig. S6A). FACS analysis also revealed that an increase in the cell population in G1 phase occurred in response to Sox2 knockout from D2 (protocol 2; supplementary material Fig. S6B,C), showing similarity to Sox9-overexpressing effects during ESC differentiation. We then investigated the pluripotency state of these cells and their differentiation into trophectoderm cells or into three germ layers. Morphologically, many cells showed differentiation into trophectoderm-like cells with Dox treatment from D0 (Dox+, D0; protocol 1). These cells were positive for the expression of trophectoderm markers keratin 7 (KRT7) and cadherin 3 (CDH3). By contrast, the control condition (Dox−) and Sox2 knockout from D2 (Dox+, D2; protocol 2) did not produce cells with trophectoderm-like morphology (Fig. 8D). Levels of pluripotency markers (SOX2 and NANOG) completely diminished by knockout of Sox2 on D4 (Fig. 8E-G). Interestingly, SOX2 knockout from D0 (protocol 1) significantly increased SOX17- and FOXA2-positve endoderm cells, but suppressed the appearance of T- and FLK1-positive mesoderm cells and NES- and PSA-NCAM-positive ectoderm cells on D4 (Fig. 8E-G). Sox2 knockout from D2 (protocol 2) significantly increased the numbers of T- and FLK1-positive mesoderm cells, SOX17- and FOXA2-positive endoderm cells, and NES- and PSA-NCAM-positive ectoderm cells on D4 (Fig. 8E-G). These results suggest that timing of Sox2 knockout during cell differentiation is crucial for cell fate determination and that the delayed knockout of Sox2 enhances the differentiation of ESCs into three germ layers (Fig. 8H).

Here, we have demonstrated a molecular mechanism involved in Sox9-mediated regulation of early differentiation of ESCs into three germ layers. Sox9 stood out in unbiased analyses using 137 ESC clones of NIA mouse ESC bank. We showed the following functional links: Sox9 increased p21 expression; and P21 then inhibited Sox2 expression via direct binding in SRR enhancer. Although it has been shown that the repression of SOX2 in ESCs generates trophectoderm cells and endoderm cells (Masui et al., 2007; Nishiyama et al., 2009), the repression of SOX2 somewhat later during differentiation generates cells in all three germ layers (Fig. 8H), indicating that the timing of repression or activation of TFs is a crucial parameter in the determination of cell fate. This scenario may help to understand the initiation of stem cell differentiation and the finely controlled progenitor property.

In developmental biology, it is well established that TFs generally work in a cascade-like manner: early acting TFs initiate the differentiation, mid-acting TFs specify the cell lineage and late-acting TFs complete the process to finally form the maturely differentiated cells (Murry and Keller, 2008; Zaret and Grompe, 2008; Snykers et al., 2009). Although we and others have identified master and ancillary TFs for the determination of cell fate during the differentiation of ESCs (Vierbuchen et al., 2010; Ieda et al., 2010; Sekiya and Suzuki, 2011; Huang et al., 2011; Hiramatsu et al., 2011; Yamamizu et al., 2013), our approach using ESC lines with 137 TF transgenes successfully found the striking broad effects of SOX9 in promoting differentiation of ESCs into three germ layers by SOX9. Such a broad differentiation spectrum is most likely mediated by the SOX9-P21-SOX2 cascade, which seems to be a unique and distinct mechanism through which SOX9 acts. By contrast, the overexpression of other TFs (e.g. Cdx2) causes the differentiation into a rather specific lineage (e.g. trophectoderm), which is mediated by the direct activation of lineage-specific genes and the simultaneous repression of many downstream targets of pluripotency genes (e.g. Pou5f1, Sox2 and Nanog) (Nishiyama et al., 2009). Obviously, it remains to be investigated whether the regulatory mechanisms found in ESCs also function during normal animal development.

Although SOX9 and SOX2 belong to the same Sox gene family, they seem to have distinct functions. SOX2, which is classified as a member of the SoxB1 family, interacts with other TFs that have POU domain proteins, including POU5F1 for pluripotency and POU3F2 for neural progenitors (Kondoh and Kamachi, 2010). The expression of most pluripotency-associated genes, including Fgf4, Utf1, Foxo15, Lefty and Nanog, is regulated by an enhancer containing SOX2- and POU5F1-binding motifs, which is highly active in undifferentiated ESCs but is no longer active in differentiated cells (Masui et al., 2007). In the neural stem cells of the developing embryo, co-expressing SOX2 and POU3F2 contributes to the maintenance of neural stem cell state by regulating NES expression (Tanaka et al., 2004; Miyagi et al., 2006). Thus, the combination of SOX2 with partner factors provides a code of cell specification. By contrast, SOX9, which is classified as a SoxE family member, has a dimerization domain and mainly functions in organ development as a dimer (Kondoh and Kamachi, 2010; Lefebvre et al., 2007). Although the SOX9 dimer directly binds to promote transcription of specific genes, in depth investigations of the molecular interactions and post-transcriptional regulation of SOX9 are needed to determine how it activates different groups of genes in different cell types. In this sense, several post-translational modifications of SOX9 can help to understand the level of its activity. Phosphorylation by protein kinase A (PKA) on two sites that flank the HMG box domain result in an increase of SOX9-binding affinity to DNA (Huang et al., 2000). PKA signaling accelerates the onset of differentiation into three germ layers through expression of the H3K9 methyltransferase G9a (Yamamizu et al., 2012a). Furthermore, PKA turns on in the early blastocyte stages of development and regulates early differentiation of the mouse embryo (Yamamizu et al., 2012a). Furthermore, SOX9 can be sumoylated, with variable consequences for protein stability, nuclear localization, DNA-binding efficiency and transactivation potential (Taylor and Labonne, 2005; Hattori et al., 2006). PKA and sumoylation add to the network that regulates SOX9 function in various cell types.

The paradigm identified here in ESC differentiation may apply to other stem cells and progenitors, and also to cancer cells. SOX9 is highly expressed in cancers of the skin, prostate, lung, colon and brain (Kordes and Hagel, 2006; Vidal et al., 2008; Wang et al., 2008; Jiang et al., 2010; Matheu et al., 2012). However, whether SOX9 has an effect on tumor growth is still controversial. SOX9 has shown several pro-oncogenic properties, including the ability to promote proliferation, to inhibit senescence and to collaborate with other oncogenes in neoplastic transformation (Wang et al., 2008; Jiang et al., 2010; Matheu et al., 2012). But some reports demonstrated that overexpression of SOX9 in human melanoma or human endometrial carcinoma, as well as in chondrocytes, inhibited cell proliferation through P21 expression in cooperation with p53 (Panda et al., 2001; Passeron et al., 2009; Saegusa et al., 2012). Variable outcomes are likely to be correlated with the relative activation of P21. P21 regulates functions of many progenitors and stem cells, including cancer stem cells, neural stem cells (NSCs), hematopoietic stem cells (HSCs), hair follicle stem cells, mesenchymal stem cells and ESC/iPSCs (van den Bos et al., 1998; Kippin et al., 2005; Orford and Scadden, 2008; Moriguchi et al., 2010; Dolezalova et al., 2012; Lee et al., 2013). P21 governs self-renewal of colon cancer-initiating cells through cell-cycle restriction and protection from DNA damage accumulation (O'Brien et al., 2012). In NSCs and HSCs, cell cycle restriction by P21 appears to be crucial for self-renewal, as P21-deficient NSCs and HSCs proliferate more actively but are depleted over time (Kippin et al., 2005; Orford and Scadden, 2008).

Overall, SOX9 has two special features. First, its widespread positive transcription factor activity. It is noteworthy that, in spite of its extensive function, Sox9 haploinsufficiency results in campomelic dysplasia, which is mainly characterized by skeletal malformation, XY sex reversal and neonatal lethality (Foster et al., 1994; Wagner et al., 1994; Bi et al., 2001; Chaboissier et al., 2004). This may indicate that one active allele is sufficient in cell types that have a lesser requirement for SOX9 activity. The large spectrum of action of SOX9 is logically related to its second feature: the general inhibitory action on SOX2 that we describe. Our work is consistent with a recent study showing that P21 control adult neural stem cell expansion by negatively regulating SOX2 induction by direct binding of SRR2 enhancer (Marqués-Torrejon et al., 2013). In the more extensive study here, the attendant release from the pluripotential state initiates differentiation of ESCs. To this end, SOX9 itself must be transcribed efficiently at an early phase in ESC differentiation and thereafter in many tissues. This transcription now becomes the target of further work.

Mouse ESC lines carrying a tetracycline-regulatable TF (Nishiyama et al., 2009; Correa-Cerro et al., 2011) and tetracycline-regulatable Sox2-null ESCs (Masui et al., 2007) were cultured and differentiated as described previously (Yamamizu et al., 2009,, 2012a). The organ-specific cell culture media have also been described previously (Yamamizu et al., 2013). Differentiated cells were examined by immunostaining and flow cytometric analysis.

Cultured cells were harvested in differentiation medium (DM) (Yamamizu et al., 2009,, 2012a) or in DM and culture medium for neurons (Yamamizu et al., 2013) at 4, 5 and 6 days after differentiation, and stained with APC-conjugated anti-FLK1 antibody MoAb (eBioscience, 17-5821; 1:50), APC-conjugated anti-PSA-NCAM antibody MoAb (Miltenyi Biotec, 130-093-273; 1:50), PE-conjugated anti-SSEA1 antibody MoAb (R&D Systems, FAB2155P; 1:50) or APC-conjugated anti-CDH5 antibody MoAb (eBioscience, FAB761A; 1:50) and then subjected to analysis using FACS Canto II (Becton Dickinson) (Yamamizu et al., 2013). For intracellular proteins, cultured cells were fixed with 4% paraformaldehyde and washed by PBS with 5% FCS and 0.75% saponin (Sigma-Aldrich) (Yamamizu et al., 2012b,, 2013). Fixed cells were stained with PE-conjugated anti-FOXA2 antibody (Bioss, bs-2358R-PE; 1:50) and then subjected to analysis using FACS Canto II (Becton Dickinson).

For cell cycle analysis, the differentiated cells were fixed in 70% ethanol and 100 µg/ml RNase (Sigma-Aldrich) was added for removal of RNA. The cells were stained with 5 µg/ml propidium iodide (Roche) at 4°C for 1 h and then subjected to analysis using FACS Canto II (Becton Dickinson).

Immunostaining for cultured cells was carried out as described previously (Yamamizu et al., 2009,, 2012a). Primary antibodies were as follows: mouse anti-POU5F1 (Santa Cruz, 5279; 1:200); rabbit anti-NANOG (ReproCell, RCAB0002P; 1:500); rat anti-SOX2 (eBioscience; 1:500); mouse anti-SSEA1 (R&D Systems, FAB2155P; 1:500); rat anti-FLK1 (eBioscience, 14-5821; 1:500); goat anti-T (R&D Systems, AF2085; 1:500); rat anti-CD31 (eBioscience, 557355; 1:50); rat anti-CDH5 (eBioscience, MAB3226; 1:500); rabbit anti-FOXA2 (Millipore, 07-633; 1:500); goat anti-SOX17 (R&D Systems, AF1924; 1:500); mouse anti-ALB (R&D Systems, MAB1455; 1:200); mouse anti-NES (StemCell Technologies, 01418; 1:500); mouse anti-βIII-TUBULIN (TUJ1) (Covance, MMS-435P; 1:500); rabbit anti-TUJ1 (Covance, MRB-435P; 1:500); mouse anti-MAP2 (Sigma-Aldrich, M4403; 1:500); mouse anti-ISL1/ISL2 (DSHB, 81.5C10-c; 1:500); and rabbit anti-GABA (Sigma-Aldrich, A2052; 1:500).

For alkaline phosphatase staining, differentiated cells were stained using an alkaline phosphatase kit (Sigma-Aldrich) according to the manufacturer's instruction. Undifferentiated ESC colonies, partially differentiated cell colonies and totally differentiated cell colonies were evaluated by AP staining as follows: undifferentiated, AP-positive cells only; partially differentiated, mixture of AP-positive and -negative cells; totally differentiated, AP-negative cells only (Yamamizu et al., 2012a).

Western blot was performed as previously described (Yamamizu et al., 2010). Briefly, undifferentiated and differentiated cells were lysed in lysis buffer, and the samples were run on SDS/polyacrylamide gel electrophoresis, using a gradient gel (Invitrogen) followed by electrophoretic transfer onto nitrocellulose membranes with iBlot (Invitrogen). After the blots were incubated for 1 h in 5% skim milk, they were incubated overnight with the respective primary antibodies at 4°C. Primary antibodies were as follows: rabbit anti-SOX9 (Abcam, ab3697; 1:1000); mouse anti-FLAG (Sigma-Aldrich, F1804; 1:1000); mouse anti-POU5F1 (Santa Cruz, 5279; 1:1000); rabbit anti-NANOG (ReproCell, RCA0002P; 1:1000); rat anti-SOX2 (eBioscience, 14-9811-82; 1:1000); rabbit anti-P21 (Invitrogen, ab7960; 1:500); and mouse anti-β-actin (Sigma-Aldrich, A5441; 1:5000). Anti-mouse, anti-rabbit or anti-rat IgG antibodies conjugated with horseradish peroxidase (HRP) were used as secondary antibodies (1:10,000). Can Get Signal Immunoreaction Enhancer solution kit (Toyobo) was used for signal enhancement. Immunoreactivity was detected with the enhanced Pierce ECL Western Blotting Substrate (Thermo Scientific). Signal intensity was calculated with Scion Image software (Scion).

Total RNA was isolated from cells in undifferentiated ESCs and differentiated cells, using Trizol (Invitrogen), according to the manufacturer's instructions. Reverse transcription was performed with the SuperScript III first-strand synthesis system (Invitrogen). Quantitative PCR (qPCR) was performed using Power SYBR Green PCR Master Mix (Applied Biosystems) and an ABI7300 system (Applied Biosystems). The amount of target RNA was determined from the appropriate standard curve and normalized relative to the amount of Gapdh mRNA. Primer sequences are shown in supplementary material Table S1.

Microarray analyses were carried out as described previously (Nishiyama et al., 2009; Correa-Cerro et al., 2011).

Cells on day 7 of differentiation were fixed by 4% paraformaldehyde and stained by periodic acid-Schiff (PAS, Sigma) according to the manufacturer's instruction. LDL uptake by cells was assessed by fluorescence microscopy after incubation of the differentiated cells with 10 µg/ml acetylated LDL labeled with 1,1′-dioctadecyl-3,3,3′,3′-tetramethylindo-carbocyanine perchlorate (DiI-Ac-LDL) (Biomedical Technologies) for 4 h at 37°C and DAPI. The amounts of mouse ALB secreted in the culture media [DMEM without glucose or Phenol Red (Invitrogen) supplemented with 2 mM sodium pyruvate (Sigma-Aldrich) and 20 mM sodium lactate (Sigma-Aldrich)] were measured after culture of differentiated cells from ESCs or adult mouse primary hepatocytes for 24 h using a Mouse Albumin ELISA Kit (Bethyl Laboratory) according to the manufacturer's instructions. The absorbance was measured with a VICTOR³ V microplate reader (PerkinElmer).

siRNA targeting mouse SOX9, P21 and negative control were purchased from Invitrogen (Stealth RNAi). Stealth RNAi for SOX9 or P21, or control siRNA (10 nM) were transfected at D0 and D2 using Lipofectamine RNAiMAX (Invitrogen) according to the manufacturer's instruction.Differentiated cells were examined by immunohistochemistry, FACS analysis and qPCR.

After 1, 3 or 5 days, differentiated cells were subjected to crosslinking with 1% formaldehyde. Chromatin was digested in the buffer containing 0.1% sodium deoxycholate and then sheared to DNA fragments with an average length of 100-500 bp. Sonicated DNA was subjected to immunoprecipitation using anti-P21 antibody (Invitrogen). Immunoprecipitated DNA was reverse-crosslinked and qPCR was performed using Power SYBR Green PCR Master Mix (Applied Biosystems). Sets of primers were used to amplify DNA sequences (supplementary material Table S2) (Marqués-Torrejon et al., 2013; Li et al., 2012). PCR amplification was conducted with a variable number of cycles (94°C for 30 s, 60°C for 30 s and 72°C for 30 s).

At least three independent experiments were performed. Statistical analysis of the data was performed with ANOVA. P<0.05 was considered significant. Values are reported as mean±s.e.m.

The GEO accession number for microarray data reported in this paper is GSE31381.

We thank our lab members for the production and early analyses of the NIA mouse ESC bank. We also thank Dr Hitoshi Niwa for providing an inducible SOX2-null ESCs.