Basal-type lumenogenesis in extraembryonic endoderm stem cells models the early visceral endoderm

Rat primitive extraembryonic endoderm stem cells (pXEN cells) are normal, naturally immortal blastocyst-derived cell lines that show features of the primitive endoderm (PrE) and contribute exclusively to the yolk sac endoderm in chimerism experiments (Debeb et al., 2009; Lo Nigro et al., 2012). They exhibit a mesenchymal morphology under standard culture conditions (Lo Nigro et al., 2012), but undergo epithelialization when cultured on feeders (Debeb et al., 2009) or in the presence of the GSKb inhibitor Chir99021 (Chir) (Chuykin et al., 2013).

A notable characteristic of rat pXEN cell lines is their tendency to form floating multicellular vesicles de novo when treated with Chir under matrix-free, low-adhesion conditions (see figure S2 of Chuykin et al., 2013). Similar vesicles were obtained with mouse pXEN cells (Zhong et al., 2018), a mouse teratocarcinoma-derived visceral endoderm (VE)-like cell line (Adamson et al., 1985) and mouse embryonic stem cell-derived PrE-like cells (Anderson et al., 2017). In unpublished work, we also obtained floating vesicles with mouse XEN cells (Kunath et al., 2005), which are highly similar to pXEN cells (Zhong et al., 2018). The rat pXEN cell vesicles were never analyzed, but the vesicles from mouse cell lines were found to express markers consistent with a VE-like identity (Adamson et al., 1985; Anderson et al., 2017; Zhong et al., 2018). It thus appears that rat pXEN cell vesicles are also VE-like and that the ability to form floating VE-like vesicles is characteristic at least of rodent extraembryonic endoderm stem cell lines. However, no comprehensive gene expression or structural analysis has been performed for any of the VE-like floating vesicles and it remains possible that they exhibit artefactual features.

Floating vesicles that form de novo in a matrix-free environment were also reported with some other cell lines; where tested, these vesicles showed an apical-outside/basal-inside polarity (Chambard et al., 1981; Wang et al., 1990; Rivron et al., 2018), suggesting the same for pXEN cell vesicles. In contrast, vesicles that originate within an extracellular matrix (ECM) show a basal-outside/apical-inside polarity (Chambard et al., 1981; Wang et al., 1990); this vesicle type reproduces the polarity of most natural cavities and its origin has therefore been studied in detail (Datta et al., 2011; Overeem et al., 2015; Taniguchi et al., 2017; Jewett and Prekeris, 2018). The apical-outside/basal-inside polarity shown by free-floating vesicles is rare in vivo; a notable example is the blastocyst, the lumen of which is formed by trophoblast cells (Aziz and Alexandre, 1991). Fittingly, trophoblast cell lines easily form floating vesicles (Shimada et al., 2001; Ezashi et al., 2011; Rivron et al., 2018), making us wonder whether the equally remarkable pXEN cell vesicles also reproduce a natural polarity. Based on electron microscopy, blastocyst lumenogenesis appears to involve lysosome-associated vacuoles (Aziz and Alexandre, 1991) but, beyond this observation, the mechanisms underlying de novo vesiculogenesis in ECM-free environments remain unstudied.

Here, we clarify the nature and mechanism of origin of floating pXEN cell vesicles. We optimize conditions for their formation and maturation and show that the vesicular lumina exhibit a basal-type polarity. We provide arguments that this configuration is not a culture artifact, but reflects the specific anatomical and physiological relationship of the VE with the epiblast during development. As a result, we present a convenient in vitro model for studying basal-type lumenogenesis and the physiology of the early visceral yolk sac endoderm.

The central processes in vesiculation are lumenogenesis and polarization. In the present study, we first elucidated the principal mechanism of lumenogenesis by pXEN cell vesicles and then followed the evolution of cell polarity throughout this process. A potential clue to lumenogenesis was provided by our observation that pXEN cells often showed conspicuous vacuoles during routine culture. In fact, those vacuoles had previously been documented (see movies 1 and 2 in Chyukin et al., 2013), although they were not explicitly mentioned. Similar vacuoles were seen in the closely related XEN cells (see figure 1E in Kunath et al., 2005), but not further studied. It is known that vacuoles can contribute to lumenogenesis (Aziz and Alexandre, 1991; Davis and Camarillo, 1996; Taniguchi et al., 2017), so we suspected that pXEN cell vacuoles are related to the ability of these cells to form floating vesicles. Therefore, we started this study by taking a closer look at pXEN cell vacuoles.

We routinely maintained rat pXEN cells on fibronectin-coated plates at moderate density in the presence of several growth factors and 2% fetal bovine serum (FBS) (Lo Nigro et al., 2012) (condition A). Depending on the batch of FBS, up to 10% of the cells formed (mostly small) vacuoles. When we raised the FBS concentration to 16% (condition A plus 14% FBS), the vacuoles grew larger within a few hours and became giant after 1-2 days; higher FBS concentrations were not more effective (Fig. S1A,B). Conversely, complete removal of serum eliminated the vacuoles (Fig. S1A,B), but long term culture was not possible (data not shown). Postnatal bovine calf serum (BCS) was as effective as FBS in vacuole induction (Fig. 1A), but could not substitute for FBS in plating assays (Fig. 1B). Hence, the vacuole-inducing and mitogenic components of FBS were not identical. The plating assays also revealed that pXEN cells could not proliferate at the higher serum levels (Fig. 1B), suggesting that vacuolization interfered with proliferation. We observed similar vacuole formation in other rat pXEN cell lines, mouse pXEN cells and mouse XEN cells, but not in fibroblasts or embryonic stem cells (ESCs) (Fig. S1C). Vacuoles were also easily induced in pXEN-like cell lines established from rat and mouse yolk sacs, in primary pXEN cells emerging from cultured rat blastocysts and (less efficiently) in primary cultures of dissociated rat and mouse yolk sacs (Fig. S1C).

We wondered whether the vacuole-promoting activity of serum could be replaced by a popular serum substitute called KOSR. This was indeed the case (not shown). Moreover, the main component of KOSR, a lipid-rich bovine serum albumin (BSA) preparation (Albumax), also induced vacuoles (Fig. 1A), and this ability was retained in a >50 kDa filtrate (Fig. S1D). In contrast, BSA Cohn fraction V (BSA FV), linoleic acid–BSA and fatty acid-free BSA were ineffective (Fig. 1A, and results not shown). Thus, a BSA-bound lipid other than linoleic acid might be the active principle in the serum.

We also determined extracellular factors that counteracted vacuole formation. First, when the vacuoles were induced with FBS, they later disappeared when the cultures had reached a high cell density (Fig. S1A,B; Movie 1). Second, when cultured on feeder (rat embryonic fibroblast cell line Li1), pXEN cells did not show vacuoles despite the presence of 16% FBS (Fig. 1C). Third, several extracellular matrix molecules or preparations, including type 1 collagen, type 4 collagen, laminin, fibrin, Geltrex and fibronectin (at an increased concentration compared with condition A), inhibited vacuolization (Fig. 1C). Thus, cell–cell and cell–matrix interactions can counteract the vacuole-inducing effect of serum.

Because we knew that pXEN cells can produce collagen (Debeb et al., 2009), we tested the role of endogenous collagen production. We partially inhibited collagen synthesis and maturation with cis-L-hydroxy-proline (CHP) (Wang et al., 1990). Strikingly, at 0.3 mM CHP (the maximum nontoxic concentration), cells with large vacuoles, which were a small minority (10-12%; Fig. 1D; Fig. S1B) in the routine culture (condition A), now became dominant (54±6%; Fig. 1D). We obtained a similar result (28±4%) after treating the cultures with 0.1 mg/ml of collagenase 4 (Fig. 1D), showing that the CHP effect was collagen specific. Thus, in the routine culture, the intrinsic tendency for vacuolization is held in check by endogenous collagen production.

Next, we tested signaling drugs for their effects on vacuolization, hoping that the results would contribute to mechanistic information. Indeed, several signaling drugs induced giant vacuoles, including the GSK3b inhibitor Chir, the protein kinase C (PKC) inhibitor Bim1 and the atypical PKC inhibitor Gö6983 (Fig. 1A).

Using FBS or Chir as vacuole inducers, we tried to determine signaling pathways that would antagonize vacuolization. Because Chir is a known Wnt agonist, we first tested the Wnt/β-catenin inhibitors Xav939, IWP2 and calphostin C. However, these inhibitors did not reduce vacuole formation (Fig. S1E; IWP2 and calphostin C data not shown). In contrast, the PKC activator phorbol 12- myristate 13-acetate (PMA) blocked vacuole formation (Fig. S1E), mirroring the vacuole-inducing effect of PKC inhibitors. Additional drugs that prevented or reduced vacuole formation included forskolin (PKA activator), Y-27632 (inhibitor of ROCK, Rho-associated protein kinase), cytochalasin D (actin polymerization inhibitor), PD0325901 (MEK inhibitor), ZSTK474 (PI3K inhibitor) and imatinib (receptor tyrosine kinase inhibitor) (Fig. S1E). The effects of all of these drug were identical in both serum- and Chir-treated pXEN cells, indicating that serum and Chir induced the same type of vacuoles.

Having determined parameters that control vacuole formation, we studied the identity of the vacuoles. It is known that the VE, which blastocyst-injected pXEN cells can colonize, contains large apical vacuoles that are acidic and exhibit β-galactosidase activity (Wada, 2013; Huang and Rivéra-Perez, 2014). However, the serum-induced vacuoles of pXEN cells showed neither β-galactosidase activity (not shown) nor an acidic pH: Staining with Neutral Red or Lysotracker revealed many small lysosomes (especially in the maintenance condition), but vacuoles were stained only occasionally. Instead, vacuoles were stained green by intracellular pH Probe 1 (Lee et al., 2013), FITC-dextran and an indicator of alkaline phosphatase (AP) activity, each time implying a slightly alkaline pH (Fig. S1B,F,G). Because this appeared more typical for the extracellular milieu, we also incubated the cells with a calcium indicator dye, Fluo-3am, and observed that the vacuoles were stained intensively and selectively, indicating a high calcium concentration (Fig. S1F).

We then transfected the cells with constructs expressing fluorescently tagged organelle markers, including Sec61 (endoplasmic reticulum), Rab5 (early endosome), Rab7 (late endosome), Lamp1 (lysosome), B4galt1 (Golgi) and Cox8a (mitochondria) (Wu et al., 2016; Vashi et al., 2017; Nemani et al., 2018). None of the constructs stained the vacuoles (Fig. 1E), apart from a small proportion of vacuoles (<1%) that were stained with Lamp1, in agreement with the observations made using Neutral Red.

In view of their rapid growth, giant size, lack of organelle markers and extracellular-like milieu, we suspected that vacuoles grow by macropinocytosis. We confirmed this using two kinds of markers: 70 kDa rhodamine B (Rho B)-dextran and 0.5 µm or 1 µm microbeads (Fig. 1F,G). In line with macropinocytosis, electron microscopy images showed abundant microvilli and membrane ruffling on the cell surface (Fig. 1H). These images also showed that the vacuoles grew to giant size by merging with smaller vacuoles that, in turn, contained smaller membrane vesicles (Fig. 1I). The merger of large vacuoles into giant vacuoles was also directly observed in videos (Fig. 1J; Movie 2). Furthermore, when we pulse-labeled the cell surface with fluorescent CF488A-labeled wheat germ agglutinin (CF-WGA) and induced vacuolization, the vacuoles became labeled (Fig. 1K). Hence, at least in part, the vacuolar membranes were directly derived from the plasma membrane, again consistent with macropinocytosis.

To clarify the molecular identity of vacuolated pXEN cells, we performed RNA-seq. We compared routinely cultured (condition A or ‘ground state’) pXEN cells with the following vacuolated samples: (1) cultures treated with BCS (used rather than FBS because BCS lacked the proliferative effect), (2) cultures treated with Chir and (3) AP-labeled cell fraction purified from Chir-treated cultures by FACS. These vacuolated samples shared 81 upregulated and 53 downregulated genes compared with routinely cultured pXEN cells (Fig. 2A; Table S10A,B). Among the upregulated genes, the main GO terms and KEGG pathways were related to endoderm development, tubal and epithelial morphogenesis, PI3K signaling and apoptosis (Fig. 2A; Table S11A).

On the other hand, there were very few differences between the vacuolated and nonvacuolated samples with respect to early embryonic lineage markers (i.e. markers of early ICM/epiblast, PrE and trophoblast). On the heatmap, we noticed some increase in the VE marker Ihh and a decrease in the PrE marker Gata6 in Chir-treated samples (Fig. 2B). Furthermore, although not visible in the heatmap, quantitative RT-PCR revealed a tendency toward decreased PrE marker expression (especially Oct4/Pou5f1 and to lesser degrees Gata4, Gata6, Sox7 and Sox17) in BCS- and especially Chir-treated samples (Fig. 2C). Immunocytochemistry for Oct4 and Gata4 revealed corresponding intensities (Fig. 2D).

As noticed before (Fig. 1B), pXEN cells did not form colonies when condition A was supplemented with a vacuole-inducing FBS concentration. To assess the proliferative properties of vacuolated cells, we performed additional plating assays. When transferred from routine maintenance (2% FBS) to vacuole-inducing (16% FBS) conditions, cells lost the ability to form colonies (Fig. 3A). However, when cells with giant vacuoles were plated into the routine maintenance condition, they easily formed colonies of nonvacuolated cells (Fig. 3A). Reversal to the nonvacuolated ground state was directly observed in videos (Fig. 3B; Movie 3).

In the vacuole-promoting condition, flow cytometry revealed a significant decrease in the proportion of G1 cells and a corresponding increase in the proportion of G2 cells (Fig. 3C). Live cell imaging revealed that many vacuolated cells started cytokinesis, but failed to complete it and merged back (Fig. 3D; Movie 4). The resulting binucleated vacuolated cells repeatedly tried to divide again, but finally burst (Fig. 3D; Movie 4). However, some of the vacuolated cells managed to complete cell division (Fig. 3E; Fig. S2, Movie 5). In type 1 divisions, the vacuole became a large bleb on the surface of one daughter cell and was then internalized again (Fig. 3E top; Movie 6). In type 2 divisions, both daughter cells inherited a vacuole (Fig. 3E middle; Movie 7). In type 3 divisions, which were the most frequent successful divisions, only one daughter cell kept the vacuole (Fig. 3E bottom; Movie 8). In summary, when vacuole-promoting conditions were maintained, most vacuolated cells unsuccessfully tried to divide and died, but some managed to complete division.

Next, we determined conditions that facilitate vesicle formation. Adding Chir to the routine maintenance condition (condition A) was not effective for producing vesicles; however, vesicles formed at high frequency when (in addition to adding Chir) the serum was omitted or a low concentration of collagenase 4 (0.1 mg/ml) was added, or if cells were cultured on low-attachment plates or on the nonadhesive substrate agarose (Fig. S3A). When we simply placed the cultures on a shaker, we observed vesicle formation even in the absence of drugs, albeit at reduced frequency (Fig. S3A). Thus, low-adhesion conditions generally facilitate vesiculation.

We noticed, however, that in batch culture a complete lack of adhesion could counteract vesiculation. After treatment with a high concentration of collagenase 4 (20 mg/ml), cells floated up and formed dense multicellular aggregates (Fig. S3B). This result indicated the importance of an effective cell density; indeed, we obtained the same result by increasing cell density. We obtained vesicles in abundance by seeding 2500 cells/cm² on non-cell-culture grade plastic in the presence of Chir, but only large aggregates when seeding 20,000 cells/cm² (Fig. S3C). In 256-well agarose plates seeded with 5 cells/well, we obtained vesicles at tenfold higher efficiency than when seeding 40 cells/well (Fig. S3D). Under comparable conditions but embedded in Matrigel, single cells easily formed large colonies; however, vesicles were seen only rarely and had a very different morphology so we did not pursue this condition further.

We then explored the effects of signaling drugs on vesiculation. At least on agarose, the PKC inhibitors Bim1 and Gö6983 could substitute for Chir (Fig. S3E). On the other hand, the Chir effect was blocked by the Wnt pathway inhibitors Xav939, IWP2, and calphostin C as well as the PKA pathway stimulator forskolin, the ROCK inhibitor Y-27632 and the PKC stimulator PMA (Fig. S3E; IWP2 and calphostin C data not shown). The latter result mirrors the vesicle-inducing effect of the PKC inhibitors, indicating that Chir- and PKCi-induced vesicles are identical. Note that the Chir-antagonizing drugs had identical effects in routine maintenance medium (on agarose) and in serum-free N2B27-based medium (on low-level fibronectin) (Fig. S3E). Drugs that did not inhibit vesiculation included the MEK inhibitor PD0325901, the PI3K inhibitor ZSTK474 and the receptor tyrosine kinase inhibitor imatinib. In summary, vesiculation was most efficient when the cells were triggered to become epithelial (high concentration of Chir), were at a density favoring microaggregation and were prevented from adhering too tightly to the substratum.

How do the vacuoles fit into this scenario? To clarify this, we observed vesicle formation through time. When cells were cultured in condition B (which included 3 µM Chir) on (low-level) fibronectin, they first attached onto the dish, then floated and started to aggregate (days 5-6). Characteristically, three (sometimes four or five) vacuolated cells formed nascent vesicles by merging their vacuoles into a single lumen (Fig. 4A). Live cell imaging allowed us to observe directly the transition from vacuoles to lumen (Fig. S3F, Movies 9 and 10). Over the next month or so, the number of cells per vesicle increased and the vesicles grew to a diameter of about 1 mm (Fig. 4A); mitotic figures in DAPI-stained vesicles indicated that this involved cell proliferation (Fig. S3G). As described above for vacuoles, Rho B-dextran (70 kDa), FITC-dextran (10 kDa) and the calcium dye Fluo-3am were rapidly taken up into the vesicle lumina, suggesting pinocytosis/transcytosis and revealing a neutral to mildly alkaline pH and a high calcium concentration (Fig. S3G). Electron micrographs showed rich microvilli and membrane ruffles on the vesicle outside, plenty of vacuoles in the cytoplasm and discharges of membrane vesicles into the lumen (Fig. 4B). In summary, under low-adhesion and epithelium-promoting conditions, the vesicle lumina originated from vacuoles of three-cell microaggregates and grew by both membrane-mediated transcellular transport and an increase in cell number.

Electron micrographs revealed that the apical surface of the vesicle epithelia was on the outside (Fig. 4B). To obtain molecular confirmation, trace the origin of polarity and characterize the process of epithelialization, we performed immunochemistry or cytochemistry of established polarity and epithelial markers throughout the whole process of vesicular morphogenesis.

We started by co-staining for the apical membrane marker podocalyxin (Pdxl) (Bryant et al., 2014) and the basal membrane marker sodium/potassium ATPase (Na/K ATPase) (Farr et al., 2009). Even in the single cell prevacuolar ground state, Pdxl clearly marked the surface membrane and only the surface membrane (Fig. 4C). The side view revealed that only the membrane facing the medium, not the substrate-facing lower side, was positive for Pdxl (Fig. 4G,I). However, upon vacuolization, the cells often rounded up and then Pdxl marked the whole plasma membrane (Fig. S3H). In all cases, Pdxl remained strictly surface associated and never marked the vacuolar membrane. By the nascent vesicle stage, Pdxl marked the outer vesicle surface and kept doing so in the bona fide vesicle (Fig. 4C). By contrast, Na/K ATPase caused a punctuate pattern in the cytoplasm of the prevacuolar stage and was absent from the plasma membrane (Fig. S3H); however, upon vacuolization, Na/K ATPase marked the vacuolar membranes, then the luminal membranes of the nascent vesicles and eventually the luminal membrane of the vesicle itself (Fig. 4C; Fig. S3H).

Lipoprotein-related protein 2 (Lrp2, also known as megalin) is a known apical marker of the VE (Maurer and Cooper, 2005). We found it in various inner regions of prevacuolar pXEN cells without an obvious rule. However, from the nascent vesicle stage on, Lrp2 ‘condensed’ toward the outer (apical) membrane and then clearly marked the apical membrane of the true vesicles (Fig. 4D). F-actin is a cytoskeletal protein located near the apical pole in epithelia (Yano et al., 2017). Although it was seen in pseudopodia and lamellipodia of nonvacuolated and vacuolated pXEN cells, it became preferably (but not exclusively) associated with the apical side of nascent and true vesicles, and especially with regions of cell–cell contact (Fig. 4E). E-cadherin (Cdh1) is a component of epithelial adherens junctions (van Roy and Berx, 2008). It was diffusively distributed in the single cell stages (prevacuolar and vacuolar), started to condense toward the cell–cell junctions in the nascent vesicles and clearly marked cell–cell contacts (together with actin) in the true vesicle stage (Fig. 4E). Tight junction protein 1 (Zo1) is a marker of epithelial tight junctions (González-Mariscal et al., 2000). In prevacuolar and vacuolar pXEN cells, it was distributed in a somewhat punctuated pattern through the cytoplasm but became localized to the cell–cell junctions in the three-cell state and then in the mature vesicle, in a manner that overlapped with F-actin (Fig. 4F).

Collagen 4 and laminin B2 are components of the basement membrane on which epithelia rest with their basal side (Pozzi et al., 2017). In the prevacuolar ground state, both proteins were located on the bottom of the dish, sometimes directly below the cells and sometimes in between, probably as a result of cell migration (Fig. 4G,I). Upon vacuolization, however, the two proteins appeared in the vacuoles. In a typical cell population, about a quarter of the vacuoles contained collagen and about half contained laminin (Fig. S3I). The stains were not associated with the membrane but filled the vacuoles or formed blobs (Fig. 4H,J; Fig. S3I). By the nascent vesicle stage, all lumina contained laminin and collagen, but still not in association with the membrane. In true vesicles, laminin and collagen lined the inner surface, indicating that a basement membrane had formed (Fig. 4H,J).

Together, these results show that, under routine maintenance conditions, single pXEN cells are already prepolarized in accordance with their potential epithelial fates and that pXEN cell vesicles possess all the structural elements of a true epithelium.

To help understand the identity of the vesicles, we used RNA-seq to compare vacuolated cells and the developmental stages of the extraembryonic endoderm. Vesicles were induced by 3 µM Chir in the presence of either maintenance culture medium with 2% FBS or serum-free N2B27. Only vesicles >300 µm in diameter were collected. Compared with the vacuolated cells, both vesicle versions shared 862 upregulated and 888 downregulated genes; more up- or downregulated genes were shared than were not, indicating a high similarity of the vesicles (Fig. S4A, Table S10C,D). Among the shared upregulated genes, the dominating GO terms included epithelial differentiation, tubular morphogenesis, VE differentiation and Wnt signaling (Table S11B). On tSNE plots, the order of the ground state pXEN cells, vacuolated pXEN cells and vesicles paralleled in vivo differentiation: The ground state cells were horizontally aligned with E3.5 PrE, the sorted vacuolated cells with E4.5 PrE, and the vesicles with E9.5 VE (Fig. 5A). However, the heatmap of VE markers revealed a more nuanced picture. Some VE markers, such as Ihh, were increased even beyond the E6.5 level; some, like Hnf4a, were comparable; and others, such as Afp, ApoB and Ttr, were not higher than in the PrE. The early PrE markers Gata6 and Oct4 were downregulated in vesicles, whereas epithelial markers Cdh1 and Epcam were well expressed (Fig. 5B-E; Fig. S4B-C).

Overall, the picture in the vesicles was one of limited VE differentiation. We suspected that the continued presence of Chir was a limiting factor. Therefore, after the vesicles were established (within about 5-10 days), we transferred them into a Chir-free condition for 10 days and then performed quantitative RT-PCR for selected VE and epithelial markers (Paca et al., 2012; Chuykin et al., 2013). In addition, we verified two markers (Afp and Cer1) by immunostaining and stained vesicles with the lectin ligand Dolichos biflorus agglutinin (DBA) (Noguchi et al., 1982; Kimber, 1986). We observed a significant, in part dramatic, increase in lineage-specific VE differentiation products (especially Afp, Ttr, ApoA1, ApoB and DBA receptor), some markers of the anterior VE (Cer1, Lefty1 and Otx2) and epithelial markers (especially Ocln and Cdh1) (Fig. 5C-F). Furthermore, we noticed that the epithelial cells, which in the immature vesicles remained somewhat spindle-like, assumed a more cuboidal shape (Fig. S4D). However, we could not maintain epithelial integrity without Chir for a prolonged culture period.

Our observations collectively indicate that, even in the ground state, pXEN cells are prestructured to form basal-type lumina in a self-driven process.

First, we found that routinely cultured pXEN cells (i.e. disconnected cells with a ‘mesenchymal’ appearance) showed signs of an apical–basal polarity. That is, even though many cells exhibited front–rear polarity (as demonstrated by morphology and actin staining), the medium-facing membranes of all cells (regardless of morphology) were exclusively stained for the apical marker Pdxl, but not for the basolateral marker Na/K ATPase. Importantly, these membranes continuously remained positive for Pdxl until they became the outer (apical) membranes of the vesicles. This is a difference from the well-studied process of apical-type lumen formation by MDCK cells, which initially also contain Pdxl in the outer membrane, but then relocate it to the cell–cell contact region at the end of the first cell division (Bryant et al., 2014). It is also different from a kind of apical-type lumen formation by endothelial cells, where Pdxl is recruited to the cell–cell contact region from intracellular locations (Strilić et al., 2009).

Second, ground state pXEN cells showed an intrinsic tendency toward vacuolization. We exposed this tendency by inhibiting collagen production with CHP, which implies that the cells have to suppress vacuolization actively in order to maintain their ground state. The collagen specificity of this effect was supported by the observation that treatment with collagenase also stimulated vacuolization. Because various exogenously added matrix proteins (including collagen) were also able to block vacuolization (when it was ‘induced’ by serum), it appears that cell–matrix interaction in general antagonizes vacuolization.

Third, the properties of vacuoles pre-empted their destiny as the lumen of the vesicles. That is, the vacuolar contents appeared similar to the extracellular milieu, and the vacuolar membranes already showed the polarity that corresponded to their fate as a luminal membrane (i.e. basal polarity; staining for Na/K ATPase but not Pdxl). The basal-luminal fate of the vacuoles was further in line with the accumulation of laminin and collagen, two key components of the future basement membrane. Importantly, however, the intravacuolar accumulation of collagen and laminin followed, rather than preceded, membrane polarization and these proteins were not initially associated with the vacuolar membrane. Hence, although both collagen and laminin can induce polarization and (apical-type) lumenogenesis in epithelial cells (Lee and Streuli, 2014; Overeem et al., 2015), they do not seem to play this role in pXEN cells. Of course, this does not exclude the possibility that the intraluminal ECM proteins later reinforce polarization once they had formed a basement membrane. We speculate that the matrix-independent polarization of pXEN cells avoids a conflict resulting from the ability to secrete collagen and laminin both on the cell surface and into the vacuoles. Indeed, the ground state pXEN cells also deposited collagen and laminin below their bottom membrane (which, however, did not contain the basal membrane marker Na/K ATPase).

For the vacuolization to progress into vesiculation, a finely adjusted cell–cell contact is key. At low cell density (i.e. in the absence of cell–cell contact), the vacuoles grew to giant size, which disrupted cytokinesis and the cells died. This was reminiscent of the previous observation of giant vacuoles with basolateral characteristics in the trophectoderm of blastocysts when cell–cell contact was compromised by knockout of Cdh1 (Stephenson et al., 2010). At the opposite extreme (i.e. at high cell density), pXEN cell vacuoles did not form or they reverted, because cell–cell contact counteracted vacuolization. However, a balance was reached at a certain intermediate density. In very small cell aggregates, three to five (typically three) appropriately positioned cells undergoing vacuolization formed cell–cell junctions containing Cdh1 and Zo1, which allowed the vacuoles to extrude their contents into a diffusion-restricted intercellular space. This mode of lumen initiation fits the mechanism of intracellular hollowing, perhaps with some cord hollowing, as previously seen in endothelial cells (Sigurbjörnsdóttir et al., 2014). At this nascent vesicle stage, F-actin and Lrp2 became associated with the apical pole, whereas collagen 4 and laminin started to associate with the basal pole. Interestingly, although the ECM was able to prevent vacuolization and morphogenesis, a complete lack of substrate adhesion (e.g. high concentration of collagenase) tended to let the cells float up too easily, which favored the formation of multicellular aggregates that did not produce vesicles. Very gentle adhesion provided an anchor that helped the cells in the microaggregates to interact. Hence, cell–substrate adhesion was clearly not an initiator of lumenogenesis, but could support it in a subtle way.

Once formed from only three to five cells, the vesicles expanded, which required a simultaneous increase in both luminal volume and cell number. Interestingly, the experiments with labeled FITC-dextran, calcium imaging and electron microscopy suggest that the same process that drove vacuolization (namely, pinocytosis) became extended into performing transcellular transports to support the volume increase. At the same time, the discharge of vacuolar contents into the growing lumen prevented the deadly vacuole gigantism, thus allowing the cell cycle to continue (as evidenced by mitoses). The near inability of single vacuolated pXEN cells to divide might also prevent the start of lumenogenesis from the two-cell stage, indicating another difference from the lumenogenesis of MDCK cells.

Taken together, our results reveal that pXEN cells are prepolarized and prestructured for basal-type lumenogenesis and do not depend on the classical inducers, cell-matrix or cell–cell interaction (Lee and Streuli, 2014; Sigurbjörnsdóttir et al., 2014; Manninen, 2015; Overeem et al., 2015; Ebnet et al., 2018) to trigger this process. Rather, they start the first stage of lumenogenesis (vacuolization) by default once an inhibitor is removed. Interestingly, this inhibitor is the ECM (endogenous collagen or exogenous ECM) (Fig. S5A). In contrast, the second stage (vesiculation) of vesicular morphogenesis requires a positive trigger, namely a finely adjusted cell–cell interaction, which leads to lumen initiation, epithelialization and mitosis. In this second stage, the vacuole mechanisms still operate but in controlled form, supporting expansion of the vesicles.

Future work must underpin the above mechanism with molecular details such as the surface receptors, intracellular transport and signaling pathways involved. As to the latter, our drug experiments so far suggest that the vacuolization stage is under the control of PI3K signaling, whereas the vesiculation stage appears to be controlled by Wnt signaling (Fig. S5B).

Because of the close anatomical, developmental and, consequently, physiological relationship between VE and epiblast, we found it interesting to compare pXEN cell vacuoles with the recently described apicosomes of epiblast stem cells, which create an in vitro equivalent of the pro-amniotic cavity (Taniguchi et al., 2017). In egg cylinder stage embryos, epiblast and VE (represented by pXEN cell-derived multicellular vesicles) are sandwiched and their basal sides face each other (Bielinska et al., 1999).We showed that, like apicosomes, pXEN cell vacuoles lack standard organelle markers, exhibit a nonacidic pH, contain calcium ions at a high level, derive their membranes from the plasma membrane, grow by macropinocytosis and contribute to vesicular morphogenesis by providing the luminal membrane and fluid. Moreover, both vacuolated pXEN cells and apicosome-containing epiblast stem cells exhibited signs of basal–apical polarity at the single cell stage, albeit in opposite orientations. Thus, apicosomes and pXEN cell vacuoles might be two variants of the same lumenogenic organelle, distinguished by the orientation of their polarity. It will therefore be interesting to compare these organelles in more detail. An obvious difference is that apicosomes are associated with a primary cilium, which pXEN cells almost certainly lack (Bangs et al., 2015).

As mentioned in the Introduction, basal-type lumenogenesis is not the norm in mammals. So, do pXEN cell vesicles represent the natural polarity of the VE? We argue that, in this specific case, it depends on what is being studied. Unusual for an epithelium, the early VE can be viewed as part of two vesicular ‘organs’, the yolk sac with its yolk sac cavity and the egg cylinder with its pro-amniotic cavity (Fig. S6, left). Hence, VE vesicles with apical-type lumina are more representative of the yolk sac, whereas VE vesicles with basal-type lumina (present study) should be more representative of the egg cylinder (Fig. S6, right).

In vivo, the VE probably works in tandem with the epiblast. The epiblast creates, in a noncavitational process, the pro-amniotic cavity (Bedzhov and Zernicka-Goetz, 2015). Anatomically, it seems inevitable that the necessary fluid enters through the VE layer. However, although moving overall in one direction (from the yolk sac cavity toward the pro-amniotic cavity), fluid transport must go from apical to basal in the VE, but from basal to apical in the epiblast. This could occur in both cell types through similar mechanisms, involving apicosomes in the epiblast (Taniguchi et al., 2017) and oppositely polarized equivalents in the VE. It seems reasonable to expect that VE vacuoles are similar to pXEN cell vacuoles, as the latter are similar to apicosomes (present study) and transcellular transport in pXEN cell vesicles is probably the extension of the vacuole-based vesicle-forming mechanism(s) (supported by the experiments with labeled dextran, calcium imaging and electron microscopy). In further support, in nascent and growing vesicles we observed the expression of Lrp2 and Dab2, which have been associated with transcytosis in the VE (Maurer and Cooper, 2005).

Vacuoles have been observed in the VE at the egg cylinder stage. However, they usually are considered to be of the acidic (lysosomal) type (Kawamura et al., 2012; Wada, 2013; Huang and Rivéra-Perez, 2014), whereas apicosomes and pXEN cell vacuoles are not acidic. Most probably, the vacuolar population within VE cells is complex; even in vacuolated pXEN cells, we occasionally observed acidic vacuoles along with the nonacidic ones (for an example, see Fig. 1E, Lamp1). Regardless of whether lysosomes participate in fluid transcytosis, as might be the case in blastocysts (Aziz and Alexandre, 1991), our experiments with FITC-dextran confirmed that the lumina of our vesicles were not acidic. It will therefore be of interest to compare the vacuolar populations of pXEN cell vesicles and the VE.

We cannot advocate our pXEN cell multicellular vesicles as an ‘egg cylinder-type’ model of the VE without considering their gene expression patterns. Here, we characterize for the first time the whole gene expression spectrum of the vesicles. The data indicated a close similarity with the early VE, and we observed an even higher expression of differentiated VE markers when we withdrew Chir from the established vesicles. Notably, we saw a dramatic induction of Afp, Ttr and DBA staining and a significant increase in ApoA1, ApoA4 and ApoB, which all are well-established VE markers.

Thus, pXEN cell vesicles with basal-type lumina provide an opportunity to study developmental and physiological aspects of the early VE in isolation. The apical-outside/basal-inside polarity of this model should be particularly useful for studying the transport of molecules and fluid related to the nutritional role of the visceral yolk sac and its support of the epiblast.

We present three main conclusions: (1) pXEN cells provide an excellent model for dissecting the poorly understood basal-type lumenogenesis. We resolved this process into molecularly defined stages and determined the principal cellular mechanism (Fig. 6A). We found that single pXEN cells are remarkably prestructured for this process, which proceeds in a self-driven manner but is controlled by ECM and cell–cell contact (Fig. S5A). (2) Our results suggest that pXEN cell vacuoles (this study) and apicosomes (Taniguchi et al., 2017) together define a class of organelles that play important roles in the generation and expansion of lumina, but exist in alternative polarities. We propose that these organelles operate in tandem in the VE and epiblast, respectively (Fig. 6B). (3) The in vivo-like polarity and VE-like differentiation of pXEN cell-derived multicellular vesicles make them attractive for studying the physiology of the early visceral yolk sac endoderm in vitro (Fig. 6; Fig. S6). The relevance of this model is emphasized by the fact that pXEN cells can contribute to the VE in vivo (Debeb et al., 2009).

Key cell culture materials included Albumax1 (11020-021; Thermo Fisher Scientific, Waltham, MA, USA), bovine fibronectin (Thermo Fisher Scientific, 33010018), BCS (S003-05; Welgene, Gyeongsan, Korea), B27 without vitamin A (Thermo Fisher Scientific, 12587-010), FBS (2916754; MP biomedicals, Santa Ana, CA, USA), Glutamax (Thermo Fisher Scientific, 35050061), KOSR (Thermo Fisher Scientific, 10828-010), low glucose DMEM (D6046; Sigma-Aldrich, St Louis, MO, USA), DMEM/F12 (Thermo Fisher Scientific, 21331-020), Neurobasal (Thermo Fisher Scientific, 21103-049), N2 (Thermo Fisher Scientific, 17502-048), penicillin/streptomycin (Thermo Fisher Scientific, 10378016), trypsin/EDTA (Welgene, LS015-09), β-mercaptoethanol (Sigma-Aldrich, M7522).

Drugs used included bisindolylmaleimide1 (Bim1) (B-6191; LC Laboratories, Woburn, MA, USA), calphostin C (AG-CN2-0430-C100; Adipogen, San Diego, CA, USA), cis-L-hydroxy-proline (H1169; Tokyo chemical, Nihonbashi-honcho, Tokyo, Japan), Chir99021 (4423; Tocris, Bristol, UK), collagenase 4 (LS004186; Worthington, Lakewood, NJ, USA), forskolin (Tocris, 1099), Gö6983 (LC Laboratories, G-7700), imatinib (LC Laboratories, I-5508), IWP2 (6866167; Peprotech, RI, USA), PD0325901 (Tocris, 4192), phorbol 12-myristate 13-acetate (PMA) (LC Laboratories, P-1680), Y-27632 (A3008; Apexbio, Boston, MA, USA), Xav939 (13596; Cayman, Ann Arbor, MI, USA) and ZSTK474 (S1072; Selleck, Houston, TX, USA).

For a complete list of media, additives and chemicals see Tables S1, S2.

The blastocyst-derived rat pXEN cell lines have been previously described (Debeb et al., 2009; Lo Nigro et al., 2012). pXEN-like yolk sac stem cell lines were derived from yolk sacs as follows: Yolk sacs (E10.5, mouse; E11.5, rat) were minced with scissors and incubated with 1 ml of 0.05% trypsin/EDTA in a 37°C water bath for 1 h and then washed with DMEM containing 10% (v/v) FBS and 1% (v/v) penicillin/streptomycin. The pXEN and pXEN-like cell lines were maintained as previously described (Lo Nigro et al., 2012) at a low density on fibronectin-coated plates in medium containing a low (2%) concentration of FBS; we refer to this as condition A, and to the pXEN cells as being in a ‘ground state’ (the culture conditions used in this study are summarized in Table S3). Mouse XEN cell lines (Kunath et al., 2005; Debeb et al., 2009), the rat feeder cell line Li1 (Debeb et al., 2009) and the mouse ESC line HM1 (Magin et al., 1992) were maintained as previously described. All cell lines used in this study, except for the ESCs, were derived in our laboratory (see Tables S4, S5). Using DAPI (Sigma-Aldrich, D9542) staining and confocal microscopy, we ascertained that our cell lines were free of mycoplasma contamination.

pXEN cell lines carrying fluorescent organelle markers were generated by separately transfecting the following constructs obtained from Addgene (Watertown, MA, USA): B4GALT1-pmTurquoise2 (Golgi, 36205), Lamp1-RFP (lysosomes, 1817), Sec61-mCh (ER, 49155), Rab7-GFP (late endosomes, 12605), Rab5-mRFP (early endosomes, 14437) and COX8A-pmTurquoise2 (mitochondria, 36208); see Table S6 for more details. Then, 1 μg of the given plasmid in 50 μl NaCl and 1 mg/ml polyethylenimine (23966; Polyscience, Warrington, PA, USA) in 50 μl NaCl were mixed, incubated for 10 min at room temperature and added to the cells. On the next day, cells were washed three times with PBS and incubated in fresh medium containing 1 μg/ml puromycin (ant-pr-1; Invivogen, Pak Shek Kok, Hong Kong) for 3-4 days, after which colonies were collected. Confocal images were obtained with a Zeiss LSM 800 confocal microscope (Carl Zeiss, Oberkochen, Germany) using a 63× objective.

pXEN or pXEN-like cells were seeded (day 0) into four-well plates at 2500 cells/cm² and cultured for 2 days as in condition A, but with the additional presence of a vacuole inducer. The kinds and concentrations of the inducers used are specified in the figure legends. To characterize the vacuole-inducing activity of serum, FBS was centrifuged (4000×g) through a 50 kDa filter (UFC805008; Merck Millipore, Burlington, MA, USA) for 15 min at 4°C. Mouse XEN cells (in their routine culture condition) were supplemented with an additional 20% FBS, resulting in a serum concentration of 30%.

When testing drugs for their vacuole-inhibiting ability, they were added on day 1 (1 day after seeding the cells) and the experiment terminated on day 2. In these experiments, either Chir or FBS was used as inducer and was added on day 0 or day 1, respectively. To test the effect of cell–cell contact, vacuoles were induced by FBS (added on day 0) and the proportion of vacuolated cells was quantified at specified time points by flow cytometry. To test the effects of different culture substrates, cells were seeded in condition A but with modified coating and the additional presence of 14% FBS (added on day 0); the culture was continued for a further 2 days. Coatings included type 1 collagen (39024; SPL, Pocheon, Korea), type 4 collagen (SPL, 39424), laminin (Thermo Fisher Scientific, 23017015), poly-D-lysine (SPL, 39224), 10 μg/ml fibronectin, 10 mg/ml fibrin (SC-473503; Santa Cruz Biotechnology, Dallas, TX, USA), 0.16 mg/ml Geltrex (Thermo Fisher Scientific, A1413201) and 1.5×10⁴ mitomycin C (Tocris, 3258)-treated Li1 cells/cm².

To induce vesicles, rat pXEN cells were subjected to one of the following five procedures: (1) 2500 cells/cm² were cultured under condition B, defined as follows: DMEM/F12, Neurobasal, N2, B27 without vitamin A, 1% penicillin/streptomycin, 1% (v/v) Glutamax, 100 μM β-mercaptoethanol, 10 ng/ml LIF (01-A1140; ORF Genetics, Kopavogur, Iceland) in four-well plates coated with 100 ng/ml bovine fibronectin and additionally supplemented with 3 μM Chir. In some cases (RNA-seq and immunostaining experiments), the vesicles were transferred into 12-well plates for further growth (typically for ∼5 days). (2) On day 0, cells were seeded as in condition A, but into 256-well microwell plates (ten cells/well). After 30 min (when the cells had settled down), an additional 800 μl of the same medium was added. On day 1, the vesiculation-promoting drug was added (3 μM Chir, 3 μM Bim1 or 5 μM Gö6983) and the culture terminated 3-4 days later. In some cases (RNA-seq experiments), the vesicles were transferred into 12-well plates for further growth (typically for ∼5 days). The microwell plates were prepared on day −1 by pouring 3% agarose (50004; TAKARA, Kusatsu, Japan) in water into a silicone mold (Sigma-Aldrich, Z764000). After 10 min, the agarose plate was transferred into a 12-well plate and 1 ml of condition A medium was added; the plate was then transferred into the incubator. On the following day (day 0), the medium was replaced with 1 ml of condition A medium containing the cells. (3, 4) Cells were plated at 2500 cells/cm² in regular (3) or low attachment (4) (SPL, 39706) four-well plates in condition A, but without FBS. Inhibitors were added as in procedure 2. (5) The pXEN cells were pre-aggregated in agarose microwells (50 cells/well seeded) for 3 days and then transferred to a 12-well plate on a shaker (5-10 days, condition A but without fibronectin).

To differentiate vesicles, they were first produced by procedure 1 and then transferred into condition C, defined as Advanced DMEM/F12 (Thermo Fisher Scientific, 12634-010), N2, B27 without vitamin A, 1% penicillin/streptomycin, 1% (v/v) Glutamax, 1.25 mM N-acetylcysteine (Sigma-Aldrich, A7250), 500 nM A83-01 (Peprotech, 9094360) and 50 ng/ml EGF (130-093-825; Miltenyi Biotec, Bergisch Gladbach, Germany). Cells were cultured for another 10 days in the absence of Chir.

For inhibition of vesicles, inhibitors were added on day 4 during procedure 1 or on day 3 during procedure 2 (with 3 μM Chir) and incubated for 1-2 days.

Note that several (but not all) conditions (e.g. condition A plus Chir) were able to support the entire morphogenesis from the ground state to the true vesicle, but with different efficiencies for individual steps. If not stated otherwise, we used the optimal condition for each step: A for the ground state and B for the three-cell stage and vesicles. Vacuoles were originally induced with A plus 14% serum but later more often with A plus 0.5 µM Chir (both conditions created the same type of vacuoles; see section ‘Signaling drugs can induce and inhibit vacuole formation’).

Three types of assays were used: (1) Cells or vesicles, in four-well plates, were incubated with FITC-dextran (Sigma-Aldrich, FD10S) or Rho B-dextran (Sigma-Aldrich, R9379) (40 μg/ml each) for 6 h on day 2 (vacuole experiments) or day 5 (vesicle experiments), followed by a medium change. Living cells were observed with an inverted fluorescence microscope (DMIL LED/DFC295, Leica Microsystems, Wetzlar, Germany) with a 10× or 20× objective. (2) Microbeads (3.64×10⁷ of 1 μm or 0.5 μm; Polyscience, 19391-10 and 15700-10, respectively) were added to vacuolated cells on day 1. On day 2, the cells were washed with PBS three times and fixed for 15 min with 4% paraformaldehyde at pH 7.4 (58295-1201; JUNSEI, Tokyo, Japan); nuclei were then stained with 4 μg/ml DAPI for 10 min. (3) CF488A-labeled wheat germ agglutinin (CF-WGA) (4 μg/ml; 29022-1; Biotium, Fremont, CA, USA) was added to vacuolated cells for 10 min at 37°C and the cells were washed, fixed and DAPI-stained immediately or, after a medium change, 1 day later. An LSM 800 Zeiss confocal microscope with a 63× or 40× objective was used for the microbeads or WGA experiments. In assays 1 and 2, vacuoles were induced with FBS and in assay 3 with either Chir or FBS (with identical results). In assay 1, vesicles were induced using vesicle induction procedure 1.

Vacuoles were produced over 2 days in condition A plus 0.5 μM Chir in the absence or presence of either 300 μM cis-L-hydroxy-proline (CHP) or 0.1 mg/ml collagenase 4. Vesicles were formed over 5 days in condition B in the absence or presence of 20 mg/ml collagenase 4. A Leica inverted fluorescence microscope with 10× objective was used.

Vacuoles were prepared in condition A plus 14% FBS over 2 days; vesicles were prepared in condition B over 5 days. After this time, while still in the incubator, vacuoles and vesicles were labeled with the indicated dyes as follows: 4 μg/ml Neutral Red (Sigma-Aldrich, N7005) for 10 min; 1 μM lysotracker (Thermo Fisher Scientific, L12492) for 30 min; 40 μg/ml FITC-dextran for 6 h; 20 μM Probe 1 (pH indicator) (Lee et al., 2013) (the kind gift of Prof. Chulhun Kang, Kyung Hee University, Korea) overnight; and 4.5 μM Fluo-3am (C213; ABP Biosciences, Beltsville, MD, USA) for 1 h. Before photography, cells were washed three times with PBS, except for calcium imaging, where the medium was changed to fresh medium and the incubation continued for an additional 20 min. Images were taken with a Leica inverted fluorescence microscope using a 20× (for vacuoles) or 10× (for vesicles) objective.

Cells were plated at 50 cells/cm² and cultured for 6 days under the conditions specified in the figure legends. They were then photographed with a 2.0 MP USB digital microscope (Digitech industries, Sheung Wan, Hong Kong) and the colonies counted.

We determined the percentage of vacuolated cells by visual count or flow cytometry. For the visual count, about 1000 cells total for each data point were counted in five independent experiments directly in the culture dish with an inverted phase contrast microscope (10× objective). For flow cytometry, the cells were trypsinized and stained with 0.1 μM alkaline phosphatase (AP) live stain dye (Thermo Fisher Scientific, A14353) in 500 μl low-glucose DMEM for 30 min (see photo in Fig. S1B), then immediately run without a washing step through a Becton Dickinson FACSCalibur device (BD Biosciences, San Jose, CA, USA). Because the dye tended to leak out again, we limited the run time per sample to ∼1 min (∼80,000 cells). Up to ten samples were run in one comparison (∼10 min total). In this time frame, the AP dye did not significantly leak out, as confirmed by running the same samples a second time with the same result. Where applicable, both methods gave equal results in direct comparison.

Trypsinized cells were resuspended in ice cold 70% ethanol for 1 h at 4°C, washed once with PBS and then resuspended in 500 μl propidium iodide solution (10 μg/ml; Sigma-Aldrich, P4170) in PBS containing 0.1% Triton X-100 (T/3751/08; Thermo Fisher Scientific, Hampton, NH, USA) and 100 μg/ml RNAse A (43014; m.biotech, Hanam, Korea) and analyzed immediately (∼80,000 cells per run) or stored at 4°C. A FACSCalibur device was used for the analytical runs. Cell cycle analysis was performed with ModFit LT software (Verity Software House, Topsham, ME, USA).

To stain adherent cells, they were cultured in four-well plates and fixed with 4% paraformaldehyde (pH 7.4) for 15 min, permeabilized with 0.1% Triton X-100 in PBS for 15 min, rinsed three times with PBS and incubated for 1 h in blocking solution of 5% donkey serum (GTX73205; Genetex, Irvine, CA, USA) in PBS. Cells were then incubated with the primary antibody (in blocking solution) overnight at 4°C, washed three times for 5 min each with blocking solution and incubated for 1 h with the secondary antibody (1:200 in blocking solution). The antibody-stained cells were incubated with DAPI for 10 min, rinsed three times for 5 min each with blocking solution and mounted in Vectashield (H-1000; Vector Laboratories, Burlingame, CA, USA) for 10 min. For nonadherent samples (nascent vesicles and vesicles), the same procedure was followed, except that the samples were handled using a glass Pasteur pipette (D-34323; Hilgenberg, Rheinstetten, Germany). Except for the primary antibody incubations, all procedures were performed at room temperature. Antibodies and their dilutions are listed in Table S7. Images were obtained with a Zeiss confocal microscope (Zeiss, LSM 800) with a 20× or 40× objective.

Vacuolated cells were prepared by treating the cells in condition A plus 14% FBS for 2 days; vesicles were produced by treating the cells in condition B (which includes 3 μM Chir) for 10 days. Sample preparation followed the specimen protocol of the National Instrumentation Center for Environmental Management (Seoul National University): Samples were fixed for 2 h in 2% paraformaldehyde and 2% glutaraldehyde and postfixed for 1 h using 2% osmium tetroxide in 0.1 M cacodylate buffer. Samples were then stained in 0.5% uranyl acetate overnight and dehydrated in a series of EtOH solutions (30, 50, 70, 80, 90 and 100%) for 10 min each. Finally, the samples were imbedded in Spurr's resin, sectioned at 70 nm with a Leica ultramicrotome (Leica, EM UC7), and photographed using a 80 kV JEOL JEM 1010 transmission electron microscope (JEOL, Tokyo).

To determine the fate of vacuolated cells under vacuole-inducing conditions, randomly selected cells were numbered and traced individually until death or the end of the filming period (days 1-4). The tracing graph was generated using MATLAB vR2018b software (MathWorks, Natick, MA, USA). To study the reversibility of vacuolization, vacuolated cells were collected with a glass pipette, transferred to condition A and filmed for 1-2 days. To film the transition from vacuoles to vesicles, microaggregates were collected from condition B at day 4 or 5 and cultured in the same medium for a further 1-2 days. All videos were taken at 37°C, one photo every 30 s, in a JuLI Fl live image system (JuLI-FLG04; NanoEnTek, Seoul, Korea) and edited with Adobe Premiere CS6 (Adobe Systems, San Jose, CA, USA) and Vapmix (VAPSHION, Seoul, Korea).

In this study, a Zeiss LSM 800 confocal microscope (Plan-Apochromat 20×/0.8 M27, C Plan-Apochromat 40×/1.30 or C Plan-Apochromat 63×/1.40) and an inverted DMIL LED/DFC295 fluorescence microscope (Hi Plan 4×/0.10, Hi Plan 10×/0.25 PH1, N Plan L 20×/0.35 PH1 or N Plan L 40×/0.55 CORR PH 2) were used with Zen software (v2.3, Zeiss) and the Leica application suite (v4.3.0, Leica), respectively. The images were improved with Photoshop CS6 (Adobe) in accordance with the journal's guidelines.

Total RNA was extracted with the Hybrid-R kit (305-101; GeneAll, Seoul, Korea). SYBR green (RT500M; Enzynomics, Daejeon, Korea) was used in combination with a reverse transcriptase kit (DR22-R10k; Solgent, Daejeon, Korea) and QuantStudio3 Real-Time PCR system (Thermo Fisher Scientific, A28572). Fold differences in transcript levels were determined with the 2^-ΔΔCt method, using β-actin mRNA for normalization. Primers for PCR are listed in Table S8.

For RNA-seq, the rat pXEN cell line CX1 was cultured under the conditions defined in the figure legends and described in detail in the foregoing paragraphs. For each condition, samples were generated from three independent cell culture experiments, performed on different days and using different passage numbers. RNA was isolated from whole cultures, except for AP-stained cultures. To isolate AP-stained cells (considered as vacuolated), ∼1,600,000 cells (cultured in 100 mm dishes) were trypsinized and stained as described, suspended in 4 ml PBS, filtered (35 µm) and sorted with a Becton Dickinson FACSAria III cell sorter (BD Biosciences) for 30-50 min. About 4 ml PBS was used for collecting the sorted cells, which were then centrifuged and dissolved in Trizol (TAKARA, 9190) for RNA-seq analysis.

Transcriptome sequencing was performed as previously described (Kim et al., 2016). Briefly, total RNA was extracted using Trizol and the QIAGEN RNeasy Mini Kit (74104; QIAGEN, Hilden, Germany). For ribosomal RNA (rRNA) depletion, the RiboMinus Eukaryote kit (Thermo Fisher Scientific, A1083708) was used according to the manufacturer's instructions. The RNA library was prepared with the NEBNext UltraTM directional RNA library preparation kit for Illumina (E7420S; New England BioLabs, Ipswich, MA, USA). After cDNA synthesis, cDNA fragmentation and adaptor ligation, the products were purified with AMPure XP beads (Beckman Coulter, Brea, CA, USA) and amplified by PCR (15 cycles) to generate the final cDNA library. The RNA library was sequenced using the Illumina HiSeq2500 (Illumina, San Diego, CA, USA). The RNA-seq raw data underwent quality check and control using FastQC v0.11.5 (Babraham Bioinformatics, UK) and Trimmomatic v0.36 (Bolger et al., 2014). Published data on mouse embryos (E3.5-E6.5 embryos; epiblast and ExEn cells; E9.5, E12.5 and E16.5 visceral yolk sac cells) and rat pXEN cell lines were included in the bioinformatics analysis (see Table S9). Mouse and rat quality-controlled FASTQ files were aligned by STAR 2.5.3a (Dobin et al., 2013) with the UCSC reference genome (http://genome.ucsc.edu/) mm10 and rn6, respectively. The gene annotation was based on Ensembl Release 94 (http://www.ensembl.org). The integration of cross-species data was restricted to the overlapping set of genes. The data analysis was performed in RStudio v1.1383 (RStudio, Boston, MA, USA) and packages were downloaded from either CRAN (https://cran.r-project.org/web/packages) or Bioconductor (http://www.bioconductor.org). t-SNE plots were made using Rtsne. The expression matrix was produced by normalizing the combined counts of the previously published (see above) and new (current study) data using Limma-voom (Law et al., 2014), resulting in a normalized gene expression profiling data table (Table S12, related to Fig. 2B, Fig. 5B, Fig. S4B), which in turn was used for pairwise comparisons by Limma (Fig. 2A, Fig. S4A, Table S10, Table S11). Differentially expressed genes were cut by a log2 fold change of >1.5 and P<0.05, then GO and KEGG annotations were produced using the ClusterProfiler package (Yu et al., 2012). Heatmaps were created with the Pheatmap package; markers of the extraembryonic endoderm were selected from previous publications (Paca et al., 2012; Chuykin et al., 2013). Venn plots were created by VennDiagram.

All quantitative experiments were independently performed for the number of times indicated in the figure legends (n indicates the number of independent experiments). Prism 8 (GraphPad Software, La Jolla, CA, USA) and Excel (Microsoft, Redmond, WA, USA) were used for the statistical analyses, including calculation of mean±s.e.m. (for error bars) and P-values using the two-tailed Student's t-test (for comparisons of two groups). Statistically significant differences are indicated as *P<0.05, **P<0.01, ***P<0.001 or ^$P<0.0001.

We thank Prof. Chulhun Kang (Kyung Hee University, Korea) for providing the pH Probe 1.