Mesenchymal-epithelial transition regulates initiation of pluripotency exit before gastrulation

A human embryo at late blastocyst stage of development contains three cell populations: the CDX2⁺ trophectoderm, GATA6⁺ primitive endoderm and POU5F1⁺ epiblast (Blakeley et al., 2015; Deglincerti et al., 2016a). Among them, the epiblast is the only pluripotent population that will give rise to all cell lineages in an adult body. Implantation takes place soon afterwards and loss of pluripotency coincides with the onset of gastrulation when the epiblast initiates lineage differentiation by generating the three definitive germ layers (ectoderm, mesoderm and endoderm). Between pre-implantation and gastrulation (about 1 week in human development), the epiblast is considered to be pluripotent throughout. But due to ethical and technical limitations, morphogenesis of the epiblast during this period of human development is poorly understood. Different states of pluripotency maintenance in vitro have been hypothesized to correspond to unique sub-stages of pre-gastrulation epiblast morphogenesis, e.g. with the naïve and primed states representing the pre-implantation and post-implantation epiblast, respectively (Nichols and Smith, 2009).

Pluripotency markers (e.g. NANOG and POU5F1) are expressed in both pre- and post-implantation epiblast and in their corresponding states captured in vitro in the mouse (Najm et al., 2011). Human epiblast is presumed to behave in a similar way, a concept partially supported by data from prolonged in vitro culture of human blastocyst and from histological analysis of rare, post-implantation stage human embryos (Deglincerti et al., 2016a; Luckett, 1975, 1978; Vogler, 1987). However, differences between human and mouse development have long been noted, e.g. in topographic arrangement of the epiblast sheet with respect to the rest of the embryo in vivo (Sheng, 2015), and in the properties of the embryonic stem cells (ESCs) and induced pluripotent stem cells (iPSCs) cultured in vitro (Schnerch et al., 2010). Furthermore, it is unclear whether epiblast intercellular organization plays any role in regulating its pluripotency in either mouse or human development. Except for germ cells, differentiation into somatic cell lineages (in each of the three germ layers) can be achieved from either the naïve or primed ESCs, suggesting that morphogenetic status of pre-gastrulation epiblast is not a key factor in its pluripotency maintenance in vitro. Yet, in all amniote species examined, including the birds and mammals, an epithelialized epiblast is a prerequisite for gastrulation to take place in vivo (Sheng, 2015), suggesting that epiblast pluripotency status is causally correlated with its morphogenesis.

Epithelialization of mouse epiblast has been shown to be dependent on cell-extracellular matrix (ECM) interactions. Its polarization is regulated by Integrin-linked kinase activity, known to bridge β1/2/3-integrin cytoplasmic tail to actin cytoskeleton (Sakai et al., 2003), and by both β1-integrin and dystroglycan, two main types of transmembrane proteins that mediate epiblast-ECM interaction (Li et al., 2017). Modulating cadherin-mediated epiblast cell-cell interactions, by deleting E-cad or replacing E-cad with N-cad, also had a profound effect on epiblast cell-ECM affinity (Basilicata et al., 2016). Weakening of integrin-mediated cell-ECM interaction led to a reduction in E-cad-mediated cell-cell adhesion strength (Martinez-Rico et al., 2010), indicating that cell-ECM and cell-cell interactions, as well as their crosstalk, are crucially involved in epiblast epithelialization (Mui et al., 2016). These lines of evidence suggest that polarization of the epiblast, including the establishment of apicobasal polarity, epiblast-ECM interaction and the modulation of epiblast adherens junction, is involved in epiblast pluripotency regulation.

To understand how epiblast morphology regulates its pluripotency, we first investigated whether its epithelialization could be viewed as a mesenchymal-epithelial transition (MET) process, the reverse of epithelial-mesenchymal transition (EMT) (Lamouille et al., 2014). Using a combination of in vivo (chicken epiblast) and in vitro [human iPSCs (hiPSCs), human ESCs (hESC) and mouse PSCs (mESCs and mEpiSCs)] models, we asked whether epiblast MET played a role in epiblast pluripotency maintenance and, if so, how this could offer us insight into normal pre-gastrulation development of the human embryo. We present data showing that amniote epiblast goes through a partial MET process that is primarily characterized by the segregation of basal and lateral plasma membrane domains of the epiblast and by the deposition of epiblast basement membrane. This partial MET regulates the initiation of pluripotency exit through activation of the integrin-mediated signaling pathway.

Using the analogy of Waddington's epigenetic landscape of lineage specification (Waddington, 1952), the loss of pluripotency in amniote development is marked by gastrulation, a process in which the epiblast gives rise to the three definitive germ layers (the ectoderm, mesoderm and endoderm). We and others have previously shown that gastrulation in chick embryos, taking place mainly at Hamburger and Hamilton stages (HH) 3 and 4 (about 12-18 h of development after egg laying), requires an epithelial-mesenchymal transition (EMT) process to convert epithelial-shaped epiblast cells into mesenchymal-shaped mesendoderm precursors (Bellairs, 1986; Hay, 1968; Nakaya et al., 2013, 2008). We and others have also reported that gastrulation EMT is associated with the loss of pluripotency markers in the epiblast at ∼HH5, when neural ectoderm fate is specified (Lavial et al., 2007; Shin et al., 2011). To investigate the dynamic relationship between epiblast EMT and its pluripotency loss, we carried out a transcriptomic analysis of pre- and peri-gastrulation stage chicken embryos (HH1-HH3; 0.5h-14 h of development) (Fig. S1A). Area pellucida (intra-embryonic) epiblast tissues, excluding the forming primitive streak, were collected and used for Affymetrix genechip-based transcriptome analysis (see Materials and Methods). Differential expression analysis (Table S1) showed that 5529 (of 38,535 in total) probe sets had statistically significant cross-stage variations (FDR<0.01), with 2376 exhibiting an increasing trend (Fig. S1B; 0001, 0011 and 0111 clusters) and 2543 a decreasing trend (Fig. S1C; 1000, 1100 and 1110 clusters). Gene ontology analysis revealed an enrichment of biological processes associated with epithelial morphogenesis from stage HH2 onwards (0001 cluster) (Table S2). Conversely, a large number of early HH1-specific genes (cluster 1000) were associated with RNA and nucleic acid binding, modification and metabolism (Table S2), likely reflecting their involvement in pluripotent maintenance at ovipositional stages as previously reported (Mak et al., 2015). This was supported by the analysis of candidate genes known to be involved in pluripotency regulation (e.g. NANOG, SOX3, FLF3, OTX2, TCF7L2, MYC, DNMT3 and LIN28) and epithelial morphogenesis (e.g. COL4 genes, SDC genes) (Fig. S1D; Table S1). Further confirmation came from re-analysis of our recently published developmental promoterome datasets (Lizio et al., 2017). Although whole embryos, including both the area pellucida and area opaca, were used in that paper, a general decrease in promoter activities of pluripotency-related genes and increase in those of epithelial genes were observed prominently in the first day of chick development (HH1-HH7) (Fig. S2). Taken together, these data suggested that, in addition to the well-known phase of pluripotency loss associated with gastrulation EMT and taking place at HH3-HH5, a decrease in pluripotency marker expression is seen much earlier from late HH1 and is possibly associated with an increase in epithelial-associated features of the epiblast.

To better characterize this epithelialization process, we analyzed HH1 embryos with epithelial polarity markers (Fig. 1). We compared embryos from freshly laid eggs, at Eyal-Giladi and Kochav (Eyal-Giladi and Kochav, 1976; Kochav et al., 1980) (EGK) X/XI sub-stages of HH1 (referred to as early HH1 in this work), with embryos from eggs incubated for 6 h, at EGK-XIII/XIV substages of HH1 (referred to as late HH1). At late HH1, epiblast polarity had already been clearly established, as evidenced by the presence of a continuous layer of basement membrane (Laminin and Agrin; Fig. 1A), presence of apical membrane domain (aPKC; Fig. 1G,H) and apical tight junctions (ZO-1; Fig. 1E,F), segregation of basal and lateral membrane domains marked by the lateral localization of E-cad (Fig. 1B), and basal enrichment of β1 integrin and dystroglycan (Fig. 1C,D), and enrichment of Golgi apparatus (GM130; Fig. 1I,J) and acetylated tubulin (Fig. 1K,L) in the cytoplasm apical to the nucleus. Interestingly, all these markers were also expressed at early HH1, but at lower levels and/or in a much less organized fashion (Fig. 1; 0 h). Laminin and agrin at early HH1 were expressed spottily under a small percentage of epiblast cells (Fig. 1A) and E-cad marked both lateral and basal regions of an epiblast cell (Fig. 1B). Both ZO-1 and aPKC exhibited apically compartmentalized uneven distribution at early HH1 (Fig. 1E,G). Taken together, these data suggested that, at early HH1, epiblast cells exhibit weak epithelial features and they become fully epithelialized within ∼6 h of development when the embryo reaches late HH1 (Fig. S3), which is ∼6 h before the onset of gastrulation at HH3. This transition from a weakly polarized organization to a fully epithelial organization can be viewed as epithelioid-to-epithelial transition (referred to as a partial mesenchymal-epithelial transition in this work; partial MET). We then used laminin expression as an indicator of the extent of this partial MET and NANOG protein expression as an indicator of epiblast pluripotency, and investigated their relative and dynamic changes from HH1 to HH3 (Fig. S4A-D). Before full epithelialization, a small increase in NANOG levels was observed from early HH1 to late HH1 (Fig. S4A,B). This likely corresponded to the final step of epiblast-hypoblast sorting through polyingression, as previously reported (Harrisson et al., 1991; Stern and Downs, 2012; Weinberger and Brick, 1982), which in the mouse is associated with a loss of GATA6 and increase of NANOG (Mathew et al., 2019). After full epithelialization, NANOG levels in epiblast cells decreased steadily from late HH1 (Fig. S4B) to late HH2 (Fig. S4C) and late HH3 (Fig. S4D). As shown previously (Shin et al., 2011), after gastrulation EMT, mesendoderm cells lost their pluripotency marker expression completely (Fig. S4D).

Similar to the avian epiblast, mammalian pluripotent cells appear also to undergo an epithelialization process in their morphogenesis, i.e. from non-epithelial epiblast precursors in the inner cell mass to an epithelial epiblast surrounding a proamniotic cavity before gastrulation (Bedzhov and Zernicka-Goetz, 2014). We hypothesized that such a partial MET, similar to the case in the chick, is crucial for mammalian pluripotency regulation, possibly marking a so far undescribed initiation checkpoint of pluripotency exit. To test this hypothesis, we used cultured human induced pluripotent stem cells (hiPSCs; 201B7 line) (Takahashi et al., 2007) as a surrogate for the human epiblast tissue. hiPSCs can be kept in a pluripotent state under maintenance conditions and can give rise to cell lineages in all three germ layers upon receiving differentiation cues. Single-cell level heterogeneity in pluripotency marker expression has been reported for hiPSCs cultured under maintenance conditions (Cahan and Daley, 2013; Chambers et al., 2007; Narsinh et al., 2011).

We first asked whether a similar heterogeneity could be observed at the macroscopic level. To reveal dynamic changes in gene expression, we used the RNA in situ hybridization method and analyzed expression patterns of key pluripotency regulatory genes POU5F1, NANOG and LEF1 in hiPSCs (Materials and Methods; Fig. 2A). For all genes, we observed two types of hiPSC colonies (Fig. 2A): one with a ubiquitous and centrally high expression [referred to as center (+) colonies in this work] (Fig. 2A, top panels; Fig. 2C); the other with a pericentrally high, but centrally low or negative expression [referred to as center (−) colonies in this work] (Fig. 2A, bottom panels; Fig. 2C). Statistical analysis revealed that center (+) colonies were generally smaller (with a colony radius of 304.6±12.3 µm and n=53 for POU5F1) than the center (−) colonies (with a colony radius of 553.7±27.6 µm and n=30 for POU5F1) (Fig. 2A,B). Similarly, a strong size-pattern correlation was observed for other genes (Fig. 2B). To understand this dynamic shift better, we divided each POU5F1-positive colony into sub-territories, as schematized in Fig. 2C, and analyzed their correlation with the colony size (Fig. 2D). The center (−) territory was strongly correlated with the overall colony size (Fig. 2D, green) (Pearson correlation coefficient r=0.88; two-tailed t-test P<0.001). A clear correlation (r=0.72, P<0.001) was also seen between the colony size and the width of POU5F1-positive territories [center (+) region in small colonies and pericenter (+) region in big colonies as shown in Fig. 2D], with an initial positive association (up to 450 µm in colony size) and a gradual shift to a steady state width of ∼150 µm (Fig. 2D blue). On the contrary, no correlation between the colony size and the edge territory was observed (r=0.22, P=ns) (Fig. 2C; Fig. 2D purple). This overall dichotomy in expression patterns was further supported by POU5F1 and NANOG immunofluorescence analysis (Fig. 2E). Taken together, these data suggested that hiPSCs cultured under pluripotency maintenance conditions undergo a macroscopically predictable change in pluripotency gene expression during their expansion from a few cells to a colony of hundreds of cells before passaging.

To test whether reduced pluripotency gene expression in center (−) colonies was indicative of an irreversible pluripotent loss, we isolated center and pericenter cells through laser microdissection (Fig. 2F) (see Materials and Methods). Isolated cells were treated with ROCK inhibitor Y-27632 for 2 h before dissociation and re-culture in normal maintenance media. Both center-derived and pericenter-derived hiPSCs were capable of growth and re-formation of the originally patterned colonies (Fig. 2G), with no significant difference in either the total cell number (Fig. 2H) or colony number (Fig. 2I) after re-culture. When large colonies were micro-dissected to remove pericenter cells, leaving only center (−) cells (Fig. 2K, top), such colonies were able to re-form large and patterned colonies after continuation of the culture without passaging (Fig. 2K, bottom). Similarly, bisected colonies (with half of the colony removed) were also able to re-establish patterning after continuation of the culture (Fig. 2J). These data showed that, despite having much reduced levels of pluripotency gene expression, hiPSCs located in the colony center still retained their pluripotency and that such pluripotency could be manifested after dissociation or disruption of original colony organization. However, center and pericenter cells did exhibit biased differentiation capability (Fig. 2L). When we collected micro-dissected center and pericenter cells for direct differentiation (see Materials and Methods), pericenter cells had the full capability (compared with non-dissected hiPSC colonies) to differentiate into the three germ layers, whereas center cells had reduced capability to differentiate into mesoderm fate (Fig. 2L).

Full pluripotency loss takes place during gastrulation EMT. Our data on the chicken epiblast suggested that an important landmark of pluripotency exit is the transition from a non-epithelial epiblast to an epithelial epiblast (i.e. epiblast MET rather than EMT). We asked whether this macroscopic shift in pluripotency gene expression in hiPSC colonies described above was correlated with changes in their epithelial status. Close observation indicated that, in small colonies, cells in the center were loosely packed (Fig. 3A,C). In large colonies, only colony-pericenter cells showed similar loose packing, whereas colony-center cells were arranged tightly (Fig. 3B,D). This was supported by histological sections of hiPSC colonies cultured on polycarbonate membrane (Fig. 3E, top and middle panels). A unilaminar (single-cell layered) organization of hiPSCs, regardless of colony size, was also evident from histology and confocal analyses (Fig. 3E,H-K). Flattened cells at the colony edge (Fig. 3A-E) were POU5F1 RNA-low cells [labeled as edge (−) in Fig. 2C]. Those cells defined colony boundary, were involved in colony size expansion, and were not studied further in this work. Intercellular space seen in loosely packed cells (center cells in small colonies and pericenter cells in large colonies) (Fig. 3A-E) was restricted to the basal side because little intercellular space was observed at the apical side of either tightly packed cells in the center (Fig. 3F, left) or loosely packed cells in the pericenter (Fig. 3F, right). Similar to the temporal progression of epithelial maturation described in the avian epiblast (Fig. 1), a spatial progression of epithelial maturation was seen in hiPSCs (Fig. 3G-K). Both ZO-1 (tight junction marker) (Fig. 3G,H) and E-cad (adherens junction marker) (Fig. 3G,I) were expressed in both center (tightly packed) and pericenter (loosely packed) cells. However, ZO-1 and E-cad were diffusely localized in pericenter cells (Fig. 3H,I), characteristic of an immature epithelium (Siliciano and Goodenough, 1988; Woichansky et al., 2016), whereas in center cells, E-cad (Fig. 3I) was localized to the lateral membrane and ZO-1 (Fig. 3H) to the apical junctions, as expected for a mature epithelial organization. The full epithelial nature of center cells was further supported by deposition of basement membrane proteins (HSPG2 in Fig. 3J; LAMA1 in Fig. 3K) under colony center cells, but not pericenter cells. Interestingly, apical membrane marker aPKC did not exhibit prominent difference in either expression level or localization between center and pericenter cells (Fig. 3L), suggesting that the pericenter cells, although not fully epithelial, exhibit partial epithelial characteristics and that these cells will progressively mature into a full epithelium as the colony expands. Collectively, these data show that hiPSCs under pluripotency maintenance culture conditions undergo a partial MET that is correlated with a reduction in their pluripotent gene expression. Taking both chick epiblast and hiPSC data into consideration, we called this phenomenon MET-associated initiation of pluripotency exit, in order to distinguish it from the loss of pluripotency associated with gastrulation EMT.

To test whether the association between MET and pluripotency is causal, we used the Rho-kinase inhibitor Y-27632 to perturb the MET process. We have previously shown that the activity of small GTPase RhoA is essential for the maintenance of epithelial status of the epiblast during gastrulation EMT (Nakaya et al., 2008). Y-27632, a potent inhibitor of Rho-associated protein kinases and of RhoA activity, was reported to increase the survival of dissociated hiPSCs without significantly altering their pluripotency status (Watanabe et al., 2007). After 48 h of Y-27632 treatment (20 µM; from day 4 to day 6 of culture), cells in the center of hiPSC colonies adopted a loosely packed morphology (Fig. 3E, bottom panel) compared with the control (Fig. 3E, middle panel), suggesting that the process of partial MET in the colony center was inhibited. Statistical analysis revealed that Y-27632 also caused a reduction in the height of hiPSCs (Fig. 4A; black control r=0.82 P<0.001, blue Y27632, r=0.50, P<0.001) and height heterogeneity (measured by max/min cell height ratio; Fig. 4B; black control, r=0.81, P<0.001; blue, Y27632, r=0.20, not significant). We then asked whether Y27632 could affect pluripotency exit as described above. After either 24 h (day 5 to 6) or 48 h (day 4 to 6) treatment with Y-27632, iPSC colonies were analyzed for POU5F1 expression. Two types of POU5F1-expressing colonies were observed in both cases (Fig. 4C), similar to the control. A dramatic increase in colony size, however, was observed after both 24 h and 48 h treatment, and in both colony types (Fig. 4D for center-positive colonies and Fig. 4E for center-negative colonies), consistent with the reduction in cell-cell compactness (Fig. 3E) and cell height (Fig. 4A,B). Moreover, in the 24 h treatment group, the relative ratio of center-negative and pericenter-positive territories in large colonies was changed (Fig. 4C,F), resulting in an increase in the POU5F1-positive pericenter territory at the expense of POU5F1-negative territory (Fig. 4C,F; green, 24 h treated; blue, control). This increase in the POU5F1-positive pericenter territory was correlated with the increase in the overall colony size (r=0.72, P<0.001) (Fig. 4F). These observations were further supported by the analysis using POU5F1 and NANOG antibodies (Fig. 4G). In the 48 h treatment group, the size of POU5F1-positive territory became variable and lost any correlation with the colony size (r=0.21, P<0.05) (Fig. 4C,F), although a mild increase throughout the colony in POU5F1 expression levels was observed for this group (Fig. 4C). To test whether such perturbation of patterning was specific to Y-27632, we cultured hiPSCs for 6 days (the entire duration of culture) under hypoxia (5% O₂), which had been reported to promote pluripotency maintenance in ESCs (Forristal et al., 2010; Mathieu et al., 2013). Hypoxia caused a small increase in overall colony size (Fig. 4C-E), but without altering the patterning of the colony (Fig. 4F; compare red and blue lines; r=0.56 and r=0.61, respectively, with P>0.001 in both conditions). Taken together, these data suggest that Y-27632 treatment disrupted the partial MET in hiPSC colonies and resulted in a delay in MET-induced pluripotency exit and a mild increase in pluripotency marker expression.

After epithelialization but before gastrulation, the epiblast in both chicken and human embryos undergoes further morphogenetic changes, among which is an increase in cell height. During gastrulation EMT (pluripotency loss), epiblast cells are typically of columnar epithelial morphology. In this study, we observed heterogeneity in hiPSC height (Fig. 4A; black control), although a simple correlation between cell height and pluripotency level could not be established. To test whether cell height plays an instructive role in pluripotency exit, we used a constitutively active version of human PAR1b (hPAR1b-TA; Materials and Methods) to regulate hiPSC height. PAR1 is a serine-threonine kinase important for cellular polarity establishment and maintenance. hPAR1b-TA (T595A), a non-phosphorylatable version of hPAR1b that renders it ‘constitutively active’, has been shown to promote basolateral membrane domain at the expense of apical domain and to increase the cell height of MDCK monolayer (Masuda-Hirata et al., 2009). As expected, expression of hPAR1b-TA in hiPSCs (analyzed 48 h after transfection) led to a dramatic increase in cell height (Fig. 4H, lateral view; Fig. 4J, quantification), as well as to a decrease in projected cell surface area (Fig. 4A, apical and basal views; Fig. 4K, quantification). However, co-immunofluorescence analysis of hPAR1b-TA (revealed using an antibody against His-tag) and NANOG showed that no significant change in pluripotency level could be associated with increased cell height (Fig. 4I). Neither wild-type hPAR1b (hPAR1b-WT) nor a kinase-dead mutant of hPAR1b (hPAR1b-KM) had a clear effect on either cell height (Fig. 4H,J,K) or pluripotency marker level (Fig. 4I). These data, together with our observation that, in both chick epiblast (Fig. 1) and hiPSC colonies (Fig. 3), the apical domain marker aPKC was already correctly localized at the immature epithelial stage, suggest that the segregation of apical and basolateral membrane domains is not a key factor in this partial MET or pluripotency exit.

We then asked whether subdivision of the basolateral domain into basal and lateral domains, mediated by dynamic interplay between cell-matrix and cell-cell contacts, could play a role in pluripotency exit. We first tested the effect of nocodazole on pluripotency gene expression. Nocodazole can destabilize microtubule network and lead to basement membrane degradation of chicken epiblast cells during gastrulation EMT at HH3/4 (Nakaya et al., 2011, 2008), potentially erasing the difference between the basal and lateral membrane domains. However, nocodazole was also shown to reduce pluripotency marker expression through an upregulation of TP53 in ESCs (Kallas et al., 2011). Treatment of HH1 embryo with nocodazole (3 h; 10 µg/ml) led to disruption of epithelial integrity and to a modest decrease in NANOG protein expression levels (Fig. S4E,F). hiPSCs treated with nocodazole (10 µg/ml) for 24 h became detached from the dish. Treatment for 3 h (10 µg/ml) caused a mild, but statistically significant, drop in NANOG and POU5F1 expression levels (Fig. 4L; Fig. 4M; red, nocodazole; blue, control; n=3). NANOG and POU5F1 intensities showed tight correlation in both control and nocodazole-treated groups (Fig. 4N), except for a small population of mitotically arrested cells (Fig. 4O, right) that had a strongly reduced level of POU5F1 expression (Fig. 4N,O). Taken together, these data suggest that microtubule destabilization has a complex, but not prominent, effect on hiPSC pluripotency exit.

We next tested whether integrin-mediated cell-matrix interaction played a role in pluripotency exit. Both chicken epiblast and hiPSCs expressed β1 integrin as the major β isoform (Fig. 1C; Fig. 5A; Table S3). The blocking antibody for β1 integrin (P5D2; Materials and Methods) would block β1 integrin-mediated cell-matrix interaction, but not the E-cad-mediated cell-cell interaction. Treatment for 24 h with the blocking antibody (1 µg/ml) led to complete detachment of hiPSCs, suggesting that it had a potent effect on cell-matrix interactions. Treatment for 2 h with β1 integrin blocking antibody (1 µg/ml) did not cause obvious morphological abnormality, but resulted in a robust upregulation of POU5F1 expression in all colonies and erased the center (−) territory in large colonies (Fig. 5B), suggesting that integrin-mediated cell-matrix interaction was a key regulator of pluripotency exit. Prolonged treatment (48 h) of hiPSCs with much lower concentrations of β1-integrin blocking antibody (10 ng/ml and 24 ng/ml), however, resulted in reduction in pluripotency marker expression (Fig. 5C). Interestingly, both concentrations (10 ng/ml and 24 ng/ml) of blocking antibody treatment also led to an increase in colony sizes [both center (+) and center (−) types] (Fig. 5D), indicating a ‘delay’ in the timing of the appearance of patterning. Conversely, treatment with an integrin β1-activating antibody (P4G11; see Materials and Methods) reduced the colony sizes when patterning started to emerge (Fig. 5E). Collectively, these data suggest that integrin-mediated signaling plays a positive role in promoting the pluripotency exit. We next asked how E-cad-mediated cell-cell interaction was involved in this process. hiPSC colonies treated with EGTA (2 mM) for 20 min stilled retained overall integrity (Fig. 5F, left panels), although cells had rounded-up morphology and longer treatment led to cell detachment. Interestingly, a 20 min EGTA treatment completely abolished patterned expression of pluripotency markers in large colonies, resulting in uniform salt-and-pepper expression of POU5F1 and eliminating the difference between central and pericentral territories (Fig. 5G,H; Fig. 5I, EGTA top). This loss of patterning was not rescued even after 3 h recovery in normal medium (Fig. 5I, EGTA bottom). Because EGTA blocks both E-cad (cell-cell) and integrin (cell-matrix) signaling, we tested the effect of EGTA treatment in the presence of Mg²⁺ (5 mM) or combined Mg²⁺ (2.5 mM) and Mn²⁺ (2.5 mM). Both Mg²⁺ and Mn²⁺ are known to specifically promote integrin-mediating signaling (Shimaoka et al., 2002). Addition of Mg²⁺ did not rescue the effect of 20 min EGTA treatment (Fig. 5G,H; Fig. 5I, top), but did rescue the control pattern after 3 h recovery (Fig. 5I, bottom). Combination of Mg²⁺ and Mn²⁺ very robustly inhibited the effect of EGTA (Fig. 5G-I), suggesting that pluripotency exit seen in the center of normal colonies was primarily mediated through integrin signaling. Further supporting our hiPSC-based observations, in early chick embryos, in which pluripotency markers were ubiquitously expressed in the epiblast (Fig. 5K, control; Fig. S3A and Fig. 5L) and pluripotent cells were of partial epithelial morphology (Fig. 1), brief activation of the integrin signaling (EGTA+ Mg²⁺+Mn²⁺) also led to dramatic reduction of NANOG expression (experimental outline in Fig. 5J; NANOG mRNA in Fig. 5K, right; NANOG protein in Fig. 5N).

The above data suggest that epiblast pluripotency regulation is tightly coordinated with epiblast MET regulation. As in any EMT/MET process, epiblast MET is presumed to be under stereotypical transcriptional regulation. Among the core EMT/MET-associated transcriptional regulators (SNAI1, SNAI2, ZEB1, ZEB2, TWIST1 and TWIST2) (Stemmler et al., 2019), data from geo-profiles suggest that SNAI1, ZEB1 and ZEB2 are highly expressed in hiPSCs (Yagi et al., 2011). We confirmed high expression of SNAI1 in hiPSCs by RNA in situ hybridization analysis (Fig. 6A). Interestingly, SNAI1 expression also showed robust patterning in large colonies (Fig. 6A, middle), similar to that of pluripotency markers. SNAI2 expression was very low or negative in hiPSCs (Fig. 6A, right). This colony size-dependent patterning of SNAI1 expression was also seen at the protein level, with pan-colony expression in smaller colonies (Fig. 6B, top) and patterned (center-low and periphery-high) expression in larger colonies (Fig. 6B, bottom). Aside from the core EMT/MET-associated transcription factors, additional transcriptional regulators are known to play a role in EMT/MET-associated in a process-specific manner and we tested three (GRHL2, YAP and TAZ) among those. GRHL2 was shown to recruit polycomb repressor complex and suppress ZEB1/2 and other EMT-related genes (Chung et al., 2016; Cieply et al., 2012; Mooney et al., 2017), and its downregulation was associated with poor prognosis in individuals with ovarian cancer (Chung et al., 2016; Nieto et al., 2016). Immunofluorescence staining showed that GRHL2 was highly expressed in hiPSCs, but without exhibiting marked regional difference (Fig. 6C). We furthermore tested YAP and TAZ, which are Hippo pathway effectors (Plouffe et al., 2018) and are known to be involved in EMT regulation (Stemmler et al., 2019). Nuclear localization of YAP was shown to positively control epiblast pluripotency in both human and mouse models (Hashimoto and Sasaki, 2018 preprint; Qin et al., 2016). Immunofluorescence staining showed that YAP was expressed and localized to the nucleus in all hiPSCs (Fig. 6D), with a slight decrease in center cells. TAZ was shown to differentially affect the self-renewal and differentiation of primed PSCs via its subcellular localization (Zhou et al., 2017). Consistent with a reported role for nuclear TAZ in promoting EMT, immunofluorescence staining with hiPSCs revealed a prominent difference in TAZ subcellular distribution, with pericenter cells having a nuclear localization and center cells having a cytoplasmic localization (Fig. 6D,E) (Lei et al., 2008). Both SNAI1 and TAZ were speculated to promote cancer EMT through their regulation of matrix metalloproteinases (MMPs), especially MMP9 (Wu et al., 2017; Xiao et al., 2015). We investigated MMP2 and MMP9 activities (both of which are expressed in hiPSCs based on microarray data) in hiPSC colonies using sequence-specific MMP fluorescent sensors (Lee et al., 2009; Sharma et al., 2013) (see Materials and Methods). MMP9 activity was high throughout the hiPSC colony (Fig. 6G), whereas MMP2 activity was relatively weak but detectable (Fig. 6G). As a negative control, MMP13 activity was not detectable in hiPSCs (Fig. 6G). However, neither MMP9 nor MMP2 exhibited prominent patterning in their activity distribution, suggesting that MMP activity is not the primary target of epiblast MET regulation. This was supported by the observation that treatment with marimastat, a pan-MMP inhibitor, did not lead to significant alteration in POU5F1 patterning in hiPSCs (Fig. 6F). Taken together, these data suggest that several EMT/MET-associated transcriptional regulators are involved in the epiblast MET process. How epiblast MET is transcriptionally regulated remains to be clarified.

We next investigated whether MET-like colony patterning could be observed in other PSCs. Human ESCs (hESCs) were reported to exhibit prepatterning when grown for 24 h in micropatterned culture condition (Warmflash et al., 2014) with colony size-dependent center-low/periphery-high intra-colony variation in pluripotency marker (NANOG, OCT4 and SOX2) expression. This observation was confirmed when hESCs were cultured in pluripotency, maintaining a micropatterned culture condition for either a shorter (11 h; MP-media11 h) or a longer (45 h; MP-media45 h) time period (Fig. 7A, top; Fig. 7B), with more prominent pre-patterning (periphery high, center low) of NANOG, POU5F1 (OCT4) than SOX2 seen after 45 h in culture (Fig. 7B). Such pre-patterning of pluripotency markers (NANOG and POU5F1) was also observed in hESCs cultured under non-micropatterning condition (Dish-media45h; Fig. 7A, bottom; Fig. 7B). These data suggest that intra-colony pre-patterning is a general behavior of human PSCs. Supporting this, hESC colonies also exhibited a prepatterning of EMT/MET marker SNAI1 (Fig. 7A,B), as seen in hiPSC colonies (Fig. 6A,B). This MET-like prepatterning of hESC colonies under maintenance conditions is likely different from EMT-like behavior of hESCs induced under BMP-induced differentiation (Deglincerti et al., 2016b; Warmflash et al., 2014). Under BMP-induced differentiation, SNAI1 was highly expressed in colony periphery, and pluripotency markers NANOG and SOX2 were markedly reduced in this region (MP-BMP45h; Fig. S5). Because SOX2 marks both the pluripotent cell population and the neuroectoderm cell lineage, a differentiating hESC colony would show an upregulation, instead of downregulation, of SOX2 in colony center. Under pluripotency maintenance conditions, reduced NANOG and OCT4 expression in colony center was prominent in both hESCs (Fig. 7A,B) and hiPSCs (Fig. 7C-E) and this was coupled with a mild decrease in SOX2 (Fig. 7) rather than differentiation-associated increase (Fig. S5), suggesting that pluripotency exit is not correlated with spontaneous differentiation.

The hiPSCs have a transcriptome profile similar to post-implantation epiblast and EpiSCs (Nakamura et al., 2016). In vivo, epithelialization of the epiblast corresponds to developmental progression from pre-implantation to post-implantation epiblast (Kinoshita and Smith, 2018). We then asked whether similar prepatterning could be observed in mouse EpiSCs (mEpiSCs). We first checked epithelial status of mEpiSC and mESC colonies (Fig. 8A,B). The mEpiSC colony (Fig. 8B) had a single cell-layered structure with well-localized ZO-1 (tight junction) and E-cad (adherens junction), with no obvious regional variation, indicating its full epithelial organization. In contrast, mESC colony had a multilayered structure with irregular E-cad and ZO-1 signals (Fig. 8A). mESCs had higher levels of expression of pluripotency markers NANOG and POU5F1 at both the protein (Fig. 8C) and RNA (Fig. 8D) levels. But no intra-colony pre-patterning could be observed in either mEpiSC or mESC colonies. Our mEpiSCs were cultured in the presence of a Wnt inhibitor that was known to reduce heterogeneity in NANOG expression levels. However, even when cultured in the absence of Wnt inhibitor, mEpiSCs did not exhibit pre-patterning in NANOG or POU5F1 levels (Fig. 8G), or show marked difference in their epithelial marker (ZO-1 and E-cad) localization (Fig. 8H). Similar to hiPSCs, GRHL2 was expressed in all cells in both mEpiSC and mESC colonies, but without any overt patterning (Fig. 8E). Interestingly, we did observe a prominent difference in SNAI1 subcellular localization, with predominantly cytoplasmic localization in mESCs and predominantly nuclear localization in mEpiSCs. Taken together, these data suggest that, in mouse PSCs under the current culture conditions, differential epithelialization status and pluripotency level are not reflected as intra-colony prepatterning, but rather as differences in colony morphology and pluripotency regulation between mESCs and mEpiSCs.

The avian embryo before gastrulation contains three lineages: the epiblast, hypoblast and area opaca (Fig. 9A, left). The first two are equivalent to their mammalian counterparts (Fig. 9A, right). The third anchors the developing embryo to the yolk and stretches the epiblast as it expands over the yolk surface. The area opaca cells are also the evolutionary origin of mammalian trophoblasts. The monotremes, the earliest branched-out extant mammals, for example, still retain an avian-like pre-gastrula organization (Sheng, 2015). Despite the superficial difference in epiblast topography between chick (exposed to the embryo exterior) and human (covered by polar trophectoderm), epiblast cells in both species undergo a very similar sequence of molecular and morphogenetic events (Fig. 9A). These events include: (1) molecular specification of epiblast precursors; (2) morphological sorting of epiblast-fated cells; (3) full epithelialization of the epiblast (subject of this study); (4) morphogenesis of epithelialized epiblast; and (5) dissolution of epithelial structure during gastrulation EMT. It is generally considered that pluripotency loss coincides with the last step in this sequence, when mesendoderm cells switch off pluripotency markers after gastrulation EMT and the remaining epiblast cells differentiate into ectoderm derivatives and, likewise, lose pluripotency. In this work, we demonstrate that there is an additional component of pluripotency regulation that is distinct from gastrulation EMT-related pluripotency loss. We termed this process ‘pluripotency exit’ and its earliest step ‘initiation of pluripotency exit’. We provide evidence that this ‘initiation of pluripotency exit’ is correlated with and regulated by epiblast partial MET (Fig. 9A). This partial MET refers to the transition from an immature epithelioid organization to a fully epithelial organization. Using a combination of avian epiblast and mammalian PSC models, our data suggest that partial MET-mediated pluripotency exit is reversible and is evolutionarily conserved. It is primarily regulated through the specification of the basal membrane domain of the epiblast and by integrin-mediated epiblast-extracellular matrix signaling.

The process of epiblast pluripotency exit described in this work is likely different from the ‘naïve-to-primed’ transition that has been reported in the mouse (Shahbazi et al., 2017) (Fig. 9B). The basic distinctions between these two in vitro (naïve and primed) states of pluripotency are their in vitro culture conditions and in vivo reconstitution potentials. The naïve state has been hypothesized to represent the pre-implantation epiblast, whereas the primed state represents the post-implantation state (Nichols and Smith, 2009). Cell biological studies suggest that this transition is associated with polarization of the mouse and human epiblast (Shahbazi et al., 2017). However, our study showed that in both chicken epiblast at EGK-X (oviposition stage) and young hiPSC colonies, the pluripotent stem cells are already polarized in their apical-basolateral axis, suggesting that the starting point of our analysis is a primed-like state of pluripotency. This hypothesis is supported by molecular data from hiPSCs (showing a primed pluripotency signature) and from our comparative analysis of pluripotency markers in two avian species: the zebra finch (oviposition at EGK-Vl; with a prominent naïve pluripotency signature) and chick (oviposition at EGK-X; with a primed pluripotency signature) (Mak et al., 2015). Despite such distinctions, the ‘naïve-to-primed’ transition and the ‘pluripotency exit’ reported here likely represent the same sequence of events in epiblast MET (Fig. 9A,B), with the former corresponding to the first half of MET (from non-polarized epiblast to polarized epiblast) and the latter to the second half (from polarized epithelioid epiblast to fully epithelial epiblast). More importantly, it is worth noting that epiblast MET, including both components discussed above, is different from the pluripotency loss seen during gastrulation EMT, and that these two morphogenetic events are separated by a wide (in developmental terms) temporal window during which epiblast cells are known to undergo additional intra- and inter-cellular morphogenesis while keeping their epithelial status. One of the most important patterning events in early amniote development, the radial-symmetry breaking that determines the future embryonic dorsal-ventral axis, takes place during this interval period. It remains to be demonstrated whether there are additional molecular checkpoints between the MET and EMT processes discussed above. A possible scenario is that, in vivo, as has been proposed for pluripotent cells cultured in vitro (Weinberger et al., 2016) and supported molecularly by the presence of many microstates during ESC differentiation into neuronal lineage (Stumpf et al., 2017), a continuum of regulatory steps will be uncovered.

A key component of this MET-associated pluripotency exit is the integrin-mediated cell-matrix signaling. Such signaling requires the deposition of basement membrane proteins, as shown in this work for both the chicken epiblast and hiPSC, and the segregation of basal and lateral membrane domains of the pluripotent cells. We have previously shown that during chicken gastrulation EMT, epiblast cells interact with their basement membrane through integrins and dystroglycan (Nakaya et al., 2011, 2013). This interaction compartmentalizes the basolateral domain of an epiblast cell into separate basal and lateral domains, and restricts the functions of cell-cell interaction molecules (e.g. E-cad) to the lateral domain and cell-matrix interaction molecules (e.g. integrins and dystroglycan) to the basal domain. We confirmed that dystroglycan was also expressed in pre-gastrulation epiblast cells (Fig. 1D) and in hiPSCs, although in this work we did not focus our study on the function of dystroglycan in pluripotency exit. Among the integrins, in both gastrulation-stage epiblast (Nakaya et al., 2011) and pre-gastrulation-stage epiblast (Lizio et al., 2017) (Table S3), ITGB1, ITGB5 and ITGB3 are the major β subunits and ITGAV and ITGA6 are the major α subunits. Interestingly, integrin profiling of the 201B7 line (Table S3) suggests that such heterogeneity is conserved in hiPSCs, which express ITGB1, ITGB5, ITGAV, ITGA6 and ITGAE as their major α and β subunits. Involvement of integrin signaling in pluripotency cell maintenance has been reported. For example, integrin-mediated focal adhesion kinase activity has been shown to protect hESCs from apoptosis and differentiation (Vitillo et al., 2016), and αvβ5 integrin supports self-renewal of hESCs (Braam et al., 2008). Furthermore, α6β1 has been shown to mediate laminin 511-supported renewal of hESCs and hiPSCs (Rodin et al., 2010), but likely through α6-mediated suppression of β1-integrin activity, as activation of β1-integrin or knockdown of α6-integrin led to a reduction of pluripotency marker expression (Villa-Diaz et al., 2016). Supporting this, inhibition of integrin signaling by disintegrin could promote pluripotency maintenance of mouse iPSCs (Higuchi et al., 2016). In our work, brief suppression of integrin signaling using high concentrations of β1-integrin blocking antibody resulted in a dramatic increase in pluripotency level and in the blockage of pluripotency exit (Fig. 5B), and brief activation of integrin signaling using Mg²⁺ and Mn²⁺ led to pluripotency exit in both hiPSCs (Fig. 5F) and chicken embryos (Fig. 5K,N), strongly suggesting that the MET-associated pluripotency exit observed in our model systems is positively regulated by integrin signaling. However, our results also point to a putatively positive role of integrin signaling in pluripotency maintenance, because long-term treatment of low concentrations of β1-integrin blocking antibody resulted in a reduction of pluripotency marker expression (Fig. 5C). Future experiments on how integrin signaling is transmitted intracellularly to regulate the pluripotency network will be needed to clarify such discrepancy.

Our work is the first to associate epiblast cell morphology with epiblast pluripotency exit in embryonic epiblast and cultured PSCs. It is unclear whether this phenomenon can be generalized to all cultured pluripotent cells that adopt an epithelial-like colony structure during their maintenance phase or whether it reflects intrinsic morphogenetic behavior of their developmental origin: the epiblast. We did not observe intra-colony MET in mEpiSCs, although these cells adopt a perfect epithelial morphology throughout the colony. Such species-specific differences may be due to differences in culture conditions, but it also suggests that morphological features are not the sole indicator of variability in the pluripotency spectrum. Patterned heterogeneity in pluripotency marker expression was proposed to facilitate higher-order patterning under differentiation conditions in human ESCs and mouse epiblast-like cells (Deglincerti et al., 2016b; Morgani et al., 2018; Warmflash et al., 2014), possibly by facilitating lineage-specific gene expression through GRHL2-mediated enhancer switching in the regulation of pluripotency genes (Chen et al., 2018). As MET-like pre-patterning was also observed in hESC colonies, it would be interesting to test in the future whether differences in differentiation potentials can also be associated with cells isolated from sub-regions of the hESC colony with different epithelialization status and EMT/MET marker expression levels. Such a hypothesis is supported by a recent paper showing that intra-colony difference in cellular adhesive properties defines a potential N-cad⁺ founder cell population within primate PSC colonies (Nakanishi et al., 2019).

More importantly, as we have demonstrated, pluripotency exit is reversible and is different from pluripotency loss seen during gastrulation in vivo or under differentiation conditions in vitro. For ectoderm-fated epiblast cells, the MET is clearly necessary because both neural ectoderm and non-neural ectoderm cells adopt a fully epithelial morphology after gastrulation. For those epiblast cells that will eventually give rise to the mesoderm and endoderm germ layers during gastrulation, they will have to undergo a MET to achieve full epithelial status only to be followed by a full EMT and loss of their pluripotency. The biological significance of such morphogenetic behavior is unclear because, in vitro, lineage differentiation into the three germ layers can be achieved from pluripotent cells regardless of their epithelialization status (Fig. 9B). Although the term epiblast generally refers to pluripotent cells in amniotic vertebrates (birds, reptiles and mammals), the pluripotency regulatory network has deep evolutionary root (Niwa et al., 2016) and similar pluripotent cells also exist in the anamniotic vertebrates [e.g. in zebrafish (Robles et al., 2011; Rosselló et al., 2013)]. It is therefore interesting to note that full epithelialization of pre-gastrulation pluripotent cells only occurs within the amniote clade (Nakaya and Sheng, 2008; Sheng, 2015). This suggests that epiblast MET is associated with certain amniote-specific developmental features that require epithelial structure-based intercellular signaling or force transmission [e.g. epiblast cell intercalation and polonaise movement reported in the pre-gastrulation chick embryo (Chuai et al., 2006; Hamidi et al., 2019; Voiculescu et al., 2007)]. These epithelial structure-based cellular events are known to be directly involved in the formation of the primitive streak (an amniote-specific feature) or, more generally, of a posterior epiblast-restricted center of mesendoderm internalization (i.e. of later-on gastrulation EMT). Therefore, epiblast MET is not only important for epiblast pluripotency regulation, as described in this work, but likely also for epiblast planar symmetry breaking. How these two phenomena are linked developmentally remains to be clarified.

Fertilized chicken eggs purchased from a local farm (Shiroyama farm, Kanagawa, Japan) were incubated without storage at 38.5°C for 0.5, 5.5, 9 and 14 h. The area pellucida (AP) region of the embryos, containing both the epiblast and hypoblast, but excluding the primitive streak region for 9 h and 14 h sample sets, was cut out in Pannett-Compton solution. Dissected AP samples of specific time points were pooled and used for RNA isolation. Two independent sets of total RNA from pooled AP samples were isolated using QIAGEN RNeasy Micro Kit. Quality checked total RNAs (100 ng) were used for array analysis, with the cDNA synthesis and cRNA labeling reactions performed according to the two-cycle protocol provided by Affymetrix. Affymetrix high-density oligonucleotide arrays for Gallus gallus (GeneChip Chicken Genome) were hybridized, stained and washed according to the Expression Analysis Technical Manual (Affymetrix), and analysis of resulting expression values was performed as described previously (Alev et al., 2010). Raw array data have been deposited in GEO under accession number GSE114476.

The hiPSC line 201B7 (purchased from RIKEN BRC Cell Bank) was used for all experiments. Cells were maintained in an undifferentiated state on iMatrix-511 (0.5 µg/cm²; Nippi; 892018)-coated dishes using the StemFit AK02N (Ajinomoto) medium and cultured at 37°C with 5% CO₂. At day 7 of culture, hiPSC colonies were dissociated using accutase (Nacalai tesque, 12679-54) and unicellular hiPSCs were then seeded on new culture dishes freshly coated with iMatrix-511, in a medium containing 10 μM Y-27632 (Wako Pure Chemicals Industries, 253-00513). All experiments in this study were performed at day 6 of culture, 1 day before normal passaging. For pattern alteration experiments, Y-27632 was used at a concentration of 20 µM, MMP broad spectrum inhibitor Marimastat (a gift from Prof. Hiroshi Sato, Kanazawa University, Japan) at 150 nM, nocodazole (Wako, 140-08531) at 10 µg/ml, EGTA (Sigma, E3889) at 2 mM, MgCl₂ (Sigma, 1901905) at 5 mM when alone or at 2.5 mM when associated with MnCl₂ (2.5 mM; Wako, 139-00722) and NaCl at 5 mM (Wako, 195-01663). Monoclonal antibody against β1-integrin (inhibiting antibody clone P5D2 or activating antibody clone P4G11; Developmental Studies Hybridoma Bank AB528308 or AB528307) was added to the culture media either at day 4 of culture at concentrations of 10 ng/ml and 25 ng/ml or for only 2 h prior to fixation at 1 µg/ml.

ES017 hESCs were grown in mTeSR1 (STEMCELL Technologies) in tissue culture dishes coated with Matrigel (Corning; 1:200 in DMEMF12) overnight at 4°C. Cells were passaged using dispase (STEMCELL Technologies) every 3 days. Cells were routinely tested for mycoplasma contamination and found negative. Micropatterning experiments were also performed with mTeSR1 following the protocol previously described (Deglincerti et al., 2016a). Where indicated, cells were treated with 50 ng/ml BMP4 to induce differentiation.

The mESC line EB5 was grown on 0.1% gelatin-coated dishes in a medium consisting of GMEM 10% knockout serum replacement, 1% FBS, 1 mM sodium pyruvate (Wako, 190-14881), 1× non-essential amino acids (Sigma M7145) and LIF 1000 U/ml. Cells were passaged every 3 days using accutase. The mEpiSC line (Tesar et al., 2007) was grown on 0.5 µg/cm² iMatrix-511 (Nippi)-coated dishes in a medium consisting of DMEM-F12, 20% knockout serum replacement, 1 mM sodium pyruvate, 1× non-essential amino acids, 10^–4 M 2-ME, 10 ng/ml activin A (R&D Systems, 338-AC), 5 ng/ml human recombinant FGF2 (R&D Systems, 3139-FB) and 10 µM of XAV939 (Abcam, ab120897). Cells were passaged every 4-5 days using accutase for single cell dissociation.

hiPSC colonies were grown on plasma-treated glass-bottomed dish coated with iMatrix-511. At day 6, the periphery and center of patterned colonies were separated using the laser beam of a Zeiss PALM microbeam. Following laser microdissection, both regions were isolated manually under a microscope. Selected cells were treated with Y-27632 for 2 h prior to dissociation and re-plated either at a density of 200 cells/cm² for pluripotency assessment experiments or at 400 cells/well (Prime Surface 96V Sumilon Sumitomo Bakelite, Wako, MS-9096V) for differentiation experiments.

After 5 days of culture, newly formed embryoid body differentiation was induced by replacing maintenance media with DMEM 10% FBS. After 3 days, samples were collected for RNA extraction (RNeasy micro Kit, QIAGEN) along with MCF-7, HepG2, HUVEC and undifferentiated hiPSC that were used as positive and negative controls. PCR was performed by using the following primers: human GAPDH, 5′-TCATCCCTGAGCTGAACGGG-3′ (forward) and 5′-TCCCCTCTTCAAGGGGTCTACA-3′ (reverse); human NESTIN, 5′-CAGGGTTGGAACAGAGGTTGG-3′ (forward) and 5′-GCATCTACAGCAGGAGAGGGTG-3′ (reverse); human SOX1, 5′-CGGAGCTCGTCGCATTTGTT-3′ (forward) and 5′-TCCCCGGGGTTCCCTTACTT-3′ (reverse); human PAX6, 5′-CACCCGCCCTGGTTGGTATC-3′ (forward) and 5′-TGAGGGCTGTGTCTGTTCGG-3′ (reverse); human FOXA2, 5′-AGCGGTGAAGATGGAAGGG-3′ (forward) and 5′-ATGGCCATGGTGATGAGCGA-3′ (reverse); human T, 5′-CCGAGAGCGCGGGAAAGAG-3′ (forward) and 5′-TCACTATGTGGATTCGAGGCTCAT-3′ (reverse); human HAND1, 5′-TTAACAGCGCATTCGCGGAG-3′ (forward) and 5′-CGTGCGATCCAAGTGTGTGG-3′ (reverse); human EOMES, 5′-ACACTTTACCTCAAGCCCGC-3′ (forward) and 5′-AGTTGCTAGGAGACAGCCGC-3′ (reverse); human GATA4, 5′-GTCCCAGTGCAGACCTGCTG-3′ (forward) and 5′-CCCTGAGGCTGTAGGTTGTGT-3′ (reverse); human CDX2, 5′-CTTCCTGCGCTTCTGGGCT-3′ (forward) and 5′-CCAGGCACTGAGGCTTGC-3′ (reverse).

Probes used for in situ hybridization were as follows: human NANOG, 5′-GTGTGGATCCAGCTTGTCC-3′ (forward) and 5′-GTCACACCATTGCTATTCTTC-3′ (reverse) 493 bp; human POU5F1, 5′-CAAGAACATGTGTAAGCTGCGG-3′ (forward) and 5′-AGGAGTACAGTGCAGTGAAGTG-3′ (reverse) 425 bp; human LEF1, 5′-CCAGACAAGCACAAACCTCTC-3′ (forward) and 5′-AGCCAAGAGGTGGGGTGATC-3′ (reverse) 420 bp; human OTX2, 5′-GAGAGGACGACGTTCACTC-3′ (forward) and 5′-TCTGACAGTGGGGAGATGG-3′ (reverse) 365 bp; human SNAI1, 5′-TGCCTCGACCACTATGCCG-3′ (forward) and 5′-AGGCTCGAAAGGCCTTCAACT-3′ (reverse) 479 bp; human SNAI2, 5′-TTCGTAAAGGAGCCGGGTGAC-3′ (forward) and 5′-ATCTTTGGGGCGAGTGAGTCC-3′ (reverse) 425 bp; human CLDN6, 5′-ACTCGGCCTAGGAATTTCCCTT-3′ (forward) and 5′-TAATCCCCGTGTGCTGGACG-3′ (reverse) 473 bp; human MMP2, 5′-TCTTTGGACTGCCCCAGACA-3′ (forward) and 5′-AGTACTCCCCATCGGCGTTC-3′ (reverse) 447 bp; human MMP9, 5′-AAGGCCAATCCTACTCCGCC-3′ (forward) and 5′-AGGGCGAGGACCATAGAGGT-3′ (reverse) 445 bp; mouse POU5F1, 5′-CAGGACATGAAAGCCCTGCAGAA-3′ (forward) and 5′-GCCCAAGCTGATTGGCGAT-3′ (reverse) 397 bp; and mouse NANOG, 5′-CTGGGAACGCCTCATCAATGC-3′ (forward) and 5′-TACTCCACTGGTGCTGAGCC-3′ (reverse) 469 bp. In situ hybridization was performed as described previously for chick embryos (Nakazawa et al., 2006) with a slight adaptation in incubation time or concentration of reagents. More specifically, cultured cells were fixed in 4% paraformaldehyde overnight at 4°C. A permeabilization step was performed using 1 µg/ml proteinase K incubation for 20 min at room temperature, followed by post-fixation for 15 min, prehybridization at 68°C for 3 h and hybridization with gene-specific antisense DIG-labeled probes at 68°C overnight. After hybridization, samples were washed in prehybridization solution at 68°C and then in TBST at room temperature. This was followed by 1 h blocking at room temperature and overnight incubation in anti-DIG antibody solution (1/2000; Roche, 11093274910) at 4°C. Finally, after several washing steps in NTMT solution, hiPSC colonies were processed for color development in NBT and BCIP at room temperature shielded from light. Whole-mount in situ hybridization with chicken embryos for NANOG mRNA detection was carried out as previously described (Nakazawa et al., 2006) and probe information for chicken NANOG has been reported previously (Shin et al., 2011).

All immunofluorescence experiments were performed using cells grown on plasma-treated glass-bottomed dish. The following primary antibodies were used for NANOG (1:500, ReproCELL, RCAB004P for hiPSCs; 1:200, R&D Systems, AF1997 for hESCs or RCAB001P for mouse PSCs), OCT3/4 (also known as POU5F1; 1:500; BD Transduction Laboratories, 611202), SOX2 (1:200; Cell Signaling, 3579 for hESCs and 1:1000 with rabbit antisera against mouse Sox2), Omni-probe D8 (1:200; Santa Cruz Biotechnology, sc-7270), E-cad (1:500; BD Transduction Laboratories, 610181), ZO-1 (1:100; Invitrogen, 40-2200), β-dystroglycan (1:100; Leica, NCL-b-DG), aPKCζ clone C-20 (1:400; Santa Cruz Biotechnology, sc-216), β1-integrin clone P5D2 (1:100; Developmental Studies Hybridoma Bank, AB528308). LAMA1 (1:200, Sigma, L9393), HSPG2 (1:200, Millipore, MABT12), GRHL2 (1:500, Sigma, HPA004820), SNAI1 (1:200, R&D Systems, AF3639), YAP (1:100, Santa Cruz Biotechnology, sc-101199) and TAZ (1:100, Atlas Antibodies, HPA007415) (both YAP and TAZ antibodies were kindly provided by Dr K. Nishiyama of Kumamoto University, Japan). The following secondary antibodies were used for multicolor detection: Alexa Fluor 488 and 568 (1:850; Invitrogen), except for hESCs, where secondary antibodies were used at a concentration of 1:500 (Alexa Fluor 488, 555 and 647). Images were acquired using an Olympus FV1200-IX-KU laser scanning confocal microscope for all samples except for 3D reconstruction, where a Leica TCS SP8 confocal microscope was used. Single cell 3D reconstruction experiments were generated using the Imaris software v8.0 (Bitplane), and all other image analyses were performed using Fiji software.

Chicken embryos were fixed in 4% PFA. For whole-mount staining of NANOG, embryos were processed for immunofluorescence staining, then cleared with SeeDB 2G tissue-clearing solution (Ke et al., 2016) before being processed for imaging. The following primary antibodies were used: E-cad (1:100; BD Transduction Laboratories, 610181), ZO-1 (1:100; Thermo Fisher, 40-2200), aPKC clone C-20 (1:100; Santa Cruz Biotechnology, sc-216), GM130 clone 35 (1:100; BD Bioscience, 610822), AcTub clone 6-11B-1 (1:1000; Sigma, T6793), β1-integrin (1:300; Chemicon, MAB13443), dystroglycan clone H242 (1:100; Santa Cruz Biotechnology, sc-28535), pan-laminin 1 clone 3H11 (1:100; DSHB, AB528342), laminin A1 (1:100; Sigma, L9393), agrin clone 6D2 (1:100; from DSHB) and NANOG (1:500; kindly provided by Dr Agata from Gakushuin University, Japan) (Nakanoh et al., 2015). The following secondary antibodies were used for multicolor detection: Alexa Fluor 488, 568 or 594 (1:300 for whole-mount embryo staining, 1:500 for frozen section staining; Invitrogen). Images were acquired using Olympus FV1000 with BX61WI upright or FV3000RS with IX83 invert laser scanning confocal microscope with UPlan-SApochromat 60×/1.2 NA, UPLSAPO 60×/1.3 NA or UPLSAPO 30XS/1.05 NA objective lenses.

hiPSC line 201B7 was cultured in the same condition as described above but on polycarbonate membrane inserts (Nunc, 140660). At day 6, membranes were washed with PBS and fixed in 4% paraformaldehyde for 15 min, followed by a second washing step using PBS membrane and coating with 1% bovine gelatin (Sigma, G9391) for 1 h at 37°C. Membranes were then washed and fixed again overnight in 4% paraformaldehyde. Counterstaining of hiPSC colonies was carried out using H&E staining reagents. Briefly, membranes were incubated with Meyer-Hematoxylin (Sakura Finetek, 8650) for 5 min and Eosin (Sakura Finetek, 8659) for 2 min. Membranes were then dehydrated using methanol and isopropanol serial baths before paraffin wax permeabilization (3×40 min each). Membranes were then cut away from the inserts using a scalpel and placed in a mold for final paraffin wax-embedded sectioning (10 µm). Sections were collected on bovine gelatin-coated slides, dewaxed and mounted. Analysis was performed using an Olympus BX51 microscope equipped with a DP70 digital camera.

The pEF-His-A-derived mammalian expression vector, T7/His epitope-tagged human wild-type PAR1b/MARK2, ‘kinase deficient’ PAR1b-KM (K49M) and non-phosphorylatable PAR1b-TA (T595A) constructs were kindly provided by Dr Shigeo Ohno (Yokohama City University, Japan) (Masuda-Hirata et al., 2009). All transfections have been performed using the Gene Juice reagent (EMD Millipore, 70967) following the manufacturer's instructions at day 4 of hiPSC culture.

The MMP sensors consist of MMPs peptide substrate, fluorescence dye (Cyanine 5.5, Ex/Em 675/690 nm) and dark quencher (BHQ-3, abs 650 nm). The MMP2, MMP9 or MMP13 sensor was prepared as previously described (Lee et al., 2009; Sharma et al., 2013). Briefly, Cy5.5 succinimide ester and BHQ-3 succinimide ester were conjugated to the MMP2 (Gly-Pro-Leu-Gly-Val-Arg-Gly-Lys-Gly-Gly), MMP9 (Gly-Lys-Gly-Pro-Arg-Ser-Leu-Ser-Gly-Lys-Gly-Gly) and MMP13 (Gly-Val-Pro-Leu-Ser-Leu-Thr-Met-Gly-Lys-Gly-Gly) substrates, respectively, and purified by reverse-phase high performance liquid chromatography (RP-HPLC). The sensor specificity was confirmed by activated recombinant enzymes (MMP1, MMP2, MMP3, MMP7, MMP9, MMP10 and MMP13) and fluorescence signal was detected using a fluorometer.

Results were expressed as mean±s.e.m. Statistical significance was determined using Student's t-test, ANOVA with Bonferroni's post-hoc test or Pearson correlation wherever appropriate. Results were considered significant when P<0.05.

We thank Ms Erike Sukowati, Dr Yuping Wu and Ms Ayako Isomura for help with chicken embryology; Dr Hiroki R. Ueda, Ms Junko Nishio and Mr. Kenichiro Uno for help with genechip data production; Dr Shigeo Ohno, Dr Tomonori Hirose and Dr Shoukichi Takahama for sharing PAR1b expression constructs; Dr Ruby Huang for sharing GRHL2 antibody; Dr Kiyokazu Agata for sharing chicken NANOG antibody; Dr Koichi Nishiyama for sharing the YAP and TAZ antibodies; Dr Hiroshi Sato for sharing the MMP inhibitor Marimastat; Ms Yangye Zhang and Mr Galym Ismagulov for help with integrin and EMT marker analysis; and Ms Marina Lizio for help with CAGE data analysis. We also thank the core imaging facility at International Research Center for Medical Sciences, Kumamoto University for help with image acquisition.