PAPC mediates self/non–self-distinction during Snail1-dependent tissue separation

Cell-impermeable boundaries are essential for the maintenance of tissue integrity. One distinct type of boundary is characterized by a narrow cleft between tissues, such as that which separates newly formed somites or hindbrain rhombomeres. Often, these clefts mature into ECM-filled spaces during development (Tepass et al., 2002). Another example of this type of boundary is Brachet’s cleft, which separates the ectoderm from the mesoderm in Xenopus laevis gastrulae. It permits the mesodermal cell mass to migrate across the multilayered ectodermal blastocoel roof (BCR) without invading it (Winklbauer, 2009). In zebrafish gastrulae, the mesendodermal hypoblast is similarly separated from the ectodermal epiblast (Kimmel et al., 1995).

Eph/ephrin signaling is required for tissue separation at Brachet’s cleft. Eph receptor tyrosine kinases generally interact with membrane-linked ephrin ligands, initiating receptor forward signaling or reverse signaling through the ligand. In Xenopus, numerous Ephs and ephrins are coexpressed in the ectoderm and mesoderm, and antiparallel forward signaling from the ectoderm to the mesoderm and vice versa induce cell repulsion at the boundary (Rohani et al., 2011; Park et al., 2011). The coexpression of Ephs with their ligands raises the question of why repulsion occurs only at the boundary but not also within each tissue, which would compromise tissue integrity. In part, this is caused by quantitative differences in receptor and ligand isoform expression that, in combination with the selectivity of receptor–ligand interactions, confer the highest levels of Eph/ephrin signaling to the ectoderm–mesoderm boundary (Rohani et al., 2014). Here, we show that an additional mechanism is required for boundary formation.

Paraxial protocadherin (PAPC) is a central factor in this mechanism, which acts in parallel to Eph signaling. PAPC (Kim et al., 1998; Hukriede et al., 2003; Medina et al., 2004; Schambony and Wedlich, 2007) and a PAPC effector, ankyrin repeat domain protein 5 (xANR5; Chung et al., 2007), have been shown previously to effect tissue separation in collaboration with an Xfz7-PKC noncanonical Wnt pathway (Winklbauer et al., 2001). In the present study, we show that Fz7 also signals through a Dvl-RhoA-JNK-cJun cascade to promote the expression of Snail1 (Sargent and Bennett, 1990; Hammerschmidt and Nüsslein-Volhard, 1993). Snail transcription factors affect adhesion, motility, epithelial–mesenchymal transition, and tumor progression (Thiery, 2002; Barrallo-Gimeno and Nieto, 2005; Speirs et al., 2010), and we show here that Snail1 enables PAPC to function in tissue separation. PAPC down-regulates planar cell polarity (PCP) signaling (Mlodzik, 2002) at the tissue boundary, apparently to reduce adhesion across Brachet’s cleft. PAPC also controls a self/non–self-recognition process that establishes an adhesive, yet flexible, contact between ectoderm and mesoderm cells. Our findings integrate new and previously identified control modules and outline the mechanism underlying tissue separation at the prechordal mesoderm.

Xsnail1 is expressed in the mesoderm of Xenopus gastrulae (Essex et al., 1993). To identify its function, morpholino antisense oligonucleotides (MOs) were injected into dorsal blastomeres. In uninjected or 5-mismatch control MO (5mis-MO)–injected embryos, Brachet’s cleft separated the prechordal mesoderm from the ectodermal BCR (Fig. 1 A). Injection of Snail1-MO eliminated cleft formation in this region (Fig. 1 B), and coinjection of Xsnail1 mRNA rescued it (Fig. 1 C), suggesting that Xsnail1 is required for tissue separation. Cleft defects in Xsnail1 morphants were not accompanied by any noticeable changes in mesoderm specification (Fig. S1, A–D).

Tissue separation can be tested on explanted BCR by using a standard assay (Fig. 1, D–G; Winklbauer and Keller, 1996; Wacker et al., 2000; Winklbauer et al., 2001). Normally, prechordal mesoderm explants remain on the BCR surface, showing separation behavior, whereas ectodermal BCR explants sink into the BCR. In accordance with the gastrula phenotype, control 5mis-MO had no effect on separation behavior (Fig. 1, D and H), whereas Xsnail1-MO–injected mesoderm integrated into the BCR (Fig. 1, E and H). Furthermore, separation behavior was rescued by coinjection of Xsnail1 mRNA (Fig. 1, F and H). These results indicate that Xsnail1 is necessary for ectodermal–mesodermal tissue separation.

Snail1 is also essential for tissue separation in zebrafish gastrulae. Brachet’s cleft, which forms between the ectodermal epiblast and mesendodermal hypoblast was disrupted by injection of Snail1a-MO, but not by a 5mis-MO (Fig. 1 J). Moreover, when pieces of mesoderm or epiblast were placed on epiblast explants in a manner analogous to the Xenopus BCR assay, epiblast aggregates sunk in reliably, whereas mesoderm aggregates remained on the surface (Fig. 1, I and K). Snail1a-MO caused the mesoderm to sink into the epiblast, whereas 5mis-MO had no such effect (Fig. 1 I and Fig. S1, E and F). Coinjection of Snail1a mRNA with Snail1a-MO rescued separation behavior (Fig. 1 I and Fig. S1 G).

We next asked how Xfz7 signaling (Winklbauer et al., 2001) relates to Xsnail1 function during tissue separation. We confirmed that Xfz7b-MO diminishes separation behavior (Fig. 2 A). Coinjection of Xsnail1 mRNA rescued separation (Fig. 2 A), placing Xsnail1 downstream of Xfz7. Indeed, Xsnail1 expression was regulated by Xfz7b (Fig. 2, B and C). Xsnail1 is expressed in the involuting mesoderm. Xfz7b-MO reduced its expression, and this effect was reversed by Xfz7b mRNA. Constitutively active (CA) JNK also rescued Xsnail1 expression, and the JNK inhibitor SP600125 diminished it, indicating that Xfz7b acts through JNK. A target of JNK is the transcription factor c-Jun; MOs against two closely related c-Jun genes (Knöchel et al., 2000) inhibited Xsnail1 expression except in the most posterior region. c-Jun mRNA rescued Xsnail1 expression in c-Jun–MO and in Xfz7b-MO embryos, placing c-Jun downstream of Xfz7-JNK. Xsnail1 expression appeared mislocalized after some treatments because of aberrant gastrulation movements (Fig. 2 B). The JNK inhibitor SP600125 strongly affected separation, and injection of Xsnail1 mRNA mitigated this effect (Fig. 2 A). Similarly, c-Jun–MO diminished separation, and both c-Jun and Xsnail1 mRNAs rescued it. CA-JNK and c-Jun mRNA rescued separation in Xfz7b-MO–injected mesoderm (Fig. 2 A). Together, these results indicate that Xfz7/JNK/c-Jun signaling controls Xsnail1 expression, which is in turn essential for separation behavior.

During noncanonical Wnt signaling, Fz activates JNK through Dvl and any of the GTPases RhoA, Rac, or Cdc42 (Boutros et al., 1998; Bikkavilli et al., 2008 and references therein). We characterized this link in our system. Knockdown of Xenopus Dvl2 strongly reduced Xsnail1 expression; coinjection of CA-JNK restored this expression, confirming that Dvl acts through JNK (Fig. 2 B). Of the dominant-negative (DN) forms of Rac, Cdc42, and RhoA, only DN-RhoA inhibited Xsnail1 expression (Fig. 2 B). Expression could be rescued by CA-JNK (Fig. 2 B), placing RhoA upstream of JNK. Because of the multiple functions of RhoA and Dvl2, their complex effects are treated separately below.

In the noncanonical Wnt pathway, which controls convergent extension of Xenopus chordamesoderm, the formin protein Daam1 links Dvl to RhoA activation (Habas et al., 2001; Liu et al., 2008; Tanegashima et al., 2008). However, neither N-Daam1 nor DN T-Daam1 inhibited Xsnail1 expression (Fig. 2 B and not depicted), nor did Daam1-MO or N-Daam1 diminish separation behavior (Fig. 2 A). Thus, Daam1 does not mediate the signal from Dvl2 to RhoA to regulate Xsnail1 expression.

In the zebrafish, the upstream control of Snail1a is similar to that in Xenopus. DN-RhoA reduced Snail1a expression, and expression could be rescued by CA-JNK (Fig. 2 D). In addition, an MO directed against the zebrafish dvl2 isoform diminished Snail1a expression (Fig. 2 D). Although knockdown of zebrafish Fz7b had no effect, expression of the cytoplasmically truncated Xfz7ΔC (Winklbauer et al., 2001) reduced Snail1a expression (Fig. 2 D). These data are consistent with Snail1a being regulated through an Fz7-Dvl2-RhoA-JNK module, although the Fz7 isoform involved has yet to be determined.

Together, these results suggest that Fz7 acts through Dvl and the activation of RhoA, JNK, and c-Jun to up-regulate Snail1 expression (Fig. 2 E). This cascade is distinct from the noncanonical pathway that regulates chordamesoderm convergent extension in Xenopus (Wallingford et al., 2002; Kim and Han, 2005; Roszko et al., 2009), but similar to the Fz/JNK/Jun cascade that controls PCP in the Drosophila melanogaster eye.

We showed previously that Xfz7-Gαβγ-PKC signaling also controls tissue separation (Winklbauer et al., 2001). Inhibition of G protein function with pertussis toxin (PTX; Winklbauer et al., 2001) or treatment with the PKC inhibitor chelerythrine did not affect Xsnail1 expression, and Xfz7b-MO–reduced Xsnail1 expression could not be rescued by PKC mRNA (Fig. 2 B). Thus, Xfz7-Gαβγ-PKC signaling is necessary for separation but not for Xsnail1 expression. However, Xsnail1 and CA-JNK each rescued tissue separation upon Xfz7b-MO knockdown or PTX inhibition (Fig. 2 A), suggesting that the Gαβγ-PKC branch acts upstream of Xsnail1 and involves JNK activation (Fig. 2 E).

PAPC is proposed to act synergistically with Xfz7 during tissue separation (Medina et al., 2004). PAPC-MO indeed reduced separation behavior, but morphants could not be rescued by CA-JNK or Xsnail1 mRNA (Fig. 3 A), and PAPC-MO did not diminish Xsnail1 expression (Fig. 2, B and C). Moreover, PAPC coexpression was not essential for induction of Xsnail1 expression in the BCR by Xfz7b (Fig. 3 C). Thus, PAPC does not act on Xsnail1 expression through the Xfz7/JNK/Jun pathway. Conversely, PAPC expression was not induced by Xsnail1 mRNA in the BCR (Fig. 3 D), indicating that Xsnail1 and PAPC are regulated independently and that both factors are required for tissue separation. In the zebrafish, knockdown of the PAPC homologue pcdh8 also inhibited separation behavior, and separation could be rescued by pcdh8 mRNA (Fig. 3 B).

Cooperation of Snail1 and PAPC was shown in gain-of-function experiments (Fig. 3 E). Coexpression of Xfz7b and PAPC induced separation behavior in the ectodermal BCR (Fig. 3 E; Medina et al., 2004). This effect depended on Xsnail1, as coinjection of Xsnail1-MO strongly reduced separation (Fig. 3 E). Xsnail1 mRNA, at a dose which rescued separation behavior in the mesoderm, had no effect in the BCR. PAPC mRNA alone was also ineffective, but coinjection of Xsnail1 and PAPC mRNA induced separation behavior (Fig. 3 E). These results imply that Xfz7b and PAPC interact indirectly.

PAPC function in tissue separation is mediated by XANR5 (Chung et al., 2007). Its baseline expression in the BCR was not up-regulated by Xsnail1 (Fig. S2 A), and adding XANR5 alone or together with PAPC barely induced separation behavior in the BCR (Fig. 3 E). Thus, a normal level of XANR5 in the ectoderm is sufficient, but PAPC function requires an additional Xsnail1-dependent factor. In contrast, the membrane-tethered extracellular domain of PAPC, M-PAPC, is sufficient to cause cell sorting (Kim et al., 1998; Ninomiya and Winklbauer, 2008), rescue tissue separation upon Xsnail1 knockdown (Fig. 3 A), and induce separation ectopically in the BCR (Fig. 3 M). These data imply that XANR5 activity is equivalent to the removal of the PAPC cytoplasmic domain, e.g., by blocking an autoinhibitory function. Snail1 could control PAPC at the same level by regulating the expression of a PAPC-interacting factor. In conclusion, PAPC is essential for tissue separation and Snail1 enables full-length PAPC function (Fig. 3 F), whereas M-PAPC acts in a Snail1-independent manner.

To further define the role of PAPC, we induced ectopic cleft formation in the BCR (Fig. 3, G–L). Expression of either Xsnail1 or PAPC in half of the BCR did not induce boundary formation (Fig. 3, G and H); however, coexpression of these factors caused tissue separation along the expression boundary (Fig. 3 I), confirming the co-requirement for these factors. The Xsnail1-independent M-PAPC construct also induced tissue separation (Fig. 3 J). Separation was not caused by altered levels of Ephs and ephrins; no differential regulation of any of these factors was observed in BCR expressing Xsnail1 and PAPC (Fig. S2, B and C). Moreover, EphB4 overexpression or ephrinB1/ephrinB2 knockdown in half of the BCR did not induce a cleft (Fig. S2, D and E). Thus, PAPC/Xsnail1 can elicit ectopic boundary formation without modulating the Eph/ephrin complement of the BCR.

When PAPC was expressed throughout the BCR and Xsnail1 was coexpressed on one side, no boundary formed (Fig. 3 K), but separation occurred in the converse experiment (Fig. 3 L). Thus, apposition of PAPC-expressing (PAPC⁺) and -nonexpressing (PAPC⁻) cells is required to determine the boundary, whereas Snail1 enables PAPC function. In agreement with this notion, ectopic PAPC in the BCR attenuated its separation from the mesoderm, which endogenously expresses PAPC and Snail1 (Fig. 3 M and Fig. S2 F). Although no cleft formed, expression of PAPC and Xsnail1 in adjacent halves of the BCR led to the formation of a sorting boundary that often prevented cell mixing (Fig. S3). Such an effect could be relevant when a boundary is determined by the apposition of Snail1-expressing and nonexpressing tissue, such as in the case of the notochord-somite boundary in the ascidian Ciona (Fujiwara et al., 1998).

Despite its homophilic binding ability (Kim et al., 2011), PAPC does not promote adhesion (Chen and Gumbiner, 2006). Moreover, M-PAPC⁺ and M-PAPC⁻ cells sort out when mixed, but have no strong preference for interior or surface positions, respectively (Kim et al., 1998; Ninomiya and Winklbauer, 2008), and we found that PAPC indeed does not generate an adhesion difference: tissue surface tension, a measure of cell–cell adhesion (Manning et al., 2010; David et al., 2014), was not altered by M-PAPC expression in the BCR (Fig. 3 N). This argues against a differential adhesion mechanism of boundary formation and suggests a localized modulation of adhesive interactions between M-PAPC⁺ and M-PAPC⁻ cells.

To examine this interaction, we reaggregated mixtures of M-PAPC⁺ and M-PAPC⁻ BCR cells (Fig. 4 A; Fig. S4, A and B; and Videos 1 and 2). Cells of the same type attached rapidly. M-PAPC⁺ and M-PAPC⁻ cells also formed stable contacts, but appeared to remain separated by a narrow gap when cells were visualized by injected fluorescent dextran (Fig. 4, A, C, and E; and Fig. S4 B). Although these heterotypic contacts resisted pulling, cells slipped easily past each other at these gap-like interfaces (Fig. 4 A and Videos 1 and 2). Thus, although tight contacts between cells of the same type appeared similar (Fig. 4, A, C, and E), contacts between different cell types appeared cleft-like, yet adhesive and permitted rapid lateral exchange of cell neighbors. Herein, we refer to these adhesive structures as cleft contacts.

Cleft contacts depended on Eph signaling. Gaps between M-PAPC⁺ and M-PAPC⁻ cells were stable, but EphB4 knockdown in M-PAPC⁺ cells caused prolonged gap occlusions (Fig. 4, C and E; Fig. S4 C; and Video 3). Cleft contacts also formed between prechordal mesoderm and BCR cells, with lateral gliding occurring at rates of up to 2–3 µm/min and cells separating completely on occasion (Fig. 4, B, C, and E; Fig. S4 C; and Video 4). Once initiated, cleft contacts expanded rapidly in size during cell attachment, consistent with their adhesive role (Fig. 4 D). Contacts were sensitive to PAPC-MO (Fig. 4, C and E; Fig. S4 C; and Videos 5 and 6). Thus, PAPC function is both necessary and sufficient for cleft contacts to form between cells with different PAPC expression, ascribing to PAPC a role in self/non–self-recognition.

To analyze downstream effects of PAPC/Snail1 during tissue separation, we explored a seemingly paradoxical finding: Dvl2 was required for Xsnail1 expression, but Dvl2-MO did not block separation behavior (Fig. 5 A). One possibility is that Dvl2-MO inhibited Xsnail1 expression but rescued separation further downstream; indeed, Dvl2-MO rescued separation in Xsnail1-MO–injected as well as PTX-treated mesoderm (Fig. 5 A). The PCP component Prickle (Pk1) is up-regulated in Xenopus mesoderm (Takeuchi et al., 2003; Veeman et al., 2003) and in the axial mesoderm of zebrafish (Carreira-Barbosa et al., 2003). Pk1 knockdown, which neither blocked separation (Fig. 5 A) nor reduced Xsnail1 expression (Fig. S5 A), also rescued separation. The PCP component Vangl2/Stbm had similar effects (Fig. 5 A). In the zebrafish, Dvl2-MO and Pk-MO likewise rescued tissue separation (Fig. 5 B and Fig. S1 H).

In Xenopus two Xfz7 isoforms exist, Xfz7a and Xfz7b (Wheeler and Hoppler, 1999; Djiane et al., 2000; Medina et al., 2000). Xfz7b is required for Xsnail1 expression, and its knockdown does not rescue the loss of Xsnail1 function. However, knockdown of closely related Xfz7a rescued separation in Xsnail1-MO embryos (Fig. 5 A). Thus, interference with main components of the PCP complex is sufficient to promote separation behavior in the mesoderm.

PAPC/Snail1 did not noticeably affect expression of PCP components (Fig. S5, B and C), suggesting that it inhibited PCP signaling functionally (Fig. 5 C). To characterize PCP inhibition, we reconstituted the mesoderm–BCR boundary in vitro (Fig. 5 D; Rohani et al., 2011). Dvl2-GFP or Pk1-venus localized to membrane puncta in the mesoderm (Fig. 5, E and F). Knockdown of Pk1 reduced Dvl puncta and vice versa (Fig. S5, D and H), indicating a mutual dependence. Strikingly, both Dvl and Pk puncta were diminished in the mesoderm at membranes bordering the BCR (Fig. 5, E and F). In contrast, Dvl2-GFP puncta did form on the BCR side of the boundary (Fig. 5 I and Fig. S5 E).

In the mesoderm, Dvl2-GFP puncta appeared at the boundary, but moved rapidly away from the membrane and vanished (Fig. 5, G and I; and Video 7). PAPC-MO stabilized puncta (Fig. 5, H and I; and Video 8), suggesting that PAPC normally prevents their accumulation at the boundary. Likewise, in Xsnail1 morphants, Pk1 and Dvl2 puncta increased in number in the mesoderm adjacent to the BCR, and cells of the two tissues interdigitated (Fig. 5 I and Fig. S5, F and G). Coinjection of Dvl2-MO with Xsnail1-MO reduced Pk1 puncta across the entire mesoderm (Fig. S5 H), and their absence from the boundary (Fig. 5 I) is consistent with the rescue of separation. Knockdown of XANR5 increased puncta numbers at the boundary, and the effect of Xsnail1-MO on puncta was partially reversed by XANR5 mRNA (Fig. 5 I and Fig. S5, J and K), showing partially redundant XANR5 and Snail1 functions. Puncta were localized on explant surfaces (Fig. 5 I and Fig. S5 L), suggesting that their absence from the boundary is not a consequence of, but rather a prerequisite for cleft formation. M-PAPC expression in the BCR also diminished Dsh puncta at the induced cleft (Fig. 5 I and Fig. S5 M).

PCP signaling has been implicated in cell adhesion (see Discussion). When we knocked down Dvl2 or Pk1 in mesoderm explants or in the embryo, cell separation generated pronounced retraction fibers (Fig. 6, A–F), reminiscent of the membrane tethers between detaching mesoderm and BCR cells at Brachet’s cleft (Rohani et al., 2011). In addition, tissue surface tension, a measure of cell adhesion, was diminished in the mesoderm by Pk1-MO (Fig. 6 G), and Pk1-MO–injected cells showed reduced mixing with uninjected cells (Fig. 6 H). In contrast, injection of ephrinB1-MO increased surface tension (Fig. 6 G), consistent with Eph/ephrin signaling normally lowering mesoderm cohesion (Rohani et al., 2014). We propose that PCP puncta strengthen adhesion within the mesoderm to counteract the effects of Eph/ephrin and that their down-regulation at the mesoderm–ectoderm boundary permits tissue separation. Removal of Dvl puncta from the boundary-apposed membranes of cells, but not from interior membranes, is compatible with the simultaneous requirement for Dvl2 in Snail1 expression. Its role as an EphB-interacting protein (Tanaka et al., 2003) in the context of tissue separation remains to be elucidated.

PCP inhibition is not sufficient for boundary formation. In gain-of-function experiments, coinjection of Dvl2-MO and PAPC, but not Dvl2-MO alone, induced separation (Fig. 7 A). Moreover, when Dvl2-MO was expressed in half of the BCR, no boundary formed, but a cleft appeared when PAPC was coexpressed; the boundary vanished when PAPC was simultaneously expressed in the other half of the BCR (Fig. 7, B–D). These results confirm that Xsnail1 can be functionally replaced by Dvl2 inhibition. Importantly, they also show that PAPC has a function beyond PCP inhibition. This was also evident in reaggregation experiments. Dvl2-MO–injected BCR cells attached to uninjected cells to form close contacts, and the coexpression of PAPC was required to induce cleft contact formation (Fig. 7, E and F). The function of PAPC in this process does not depend on Snail1.

Eph/ephrin function is likewise required in addition to PCP inhibition. When EphB4 was knocked down in the mesoderm such that tissue separation was inhibited (Rohani et al., 2011), Dvl puncta remained sparse at the boundary (Figs. 6 I and 7 G). This result confirms that reduced puncta density is not a consequence of cleft formation, and it places EphB signaling in parallel to PCP inhibition. Consistent with this notion, injection of Xsnail1 mRNA or of Dvl2-MO did not rescue separation upon EphB4 knockdown, whereas separation behavior induced by Dvl2-MO/PAPC expression in the BCR was inhibited by EphB4-MO (Fig. 7 H). In conclusion, PCP inhibition is required for cleft formation but is not sufficient: PAPC has an essential role in addition to its PCP-inhibitory function, a role in which Eph/ephrin signaling is required. Our overall findings on the regulation of tissue separation at Brachet’s cleft are summarized in a model (Fig. 7 I) discussed below.

To analyze cell–cell interactions across ectoderm–mesoderm boundaries at high resolution, we performed transmission EM (TEM). At low magnification, Brachet’s cleft appeared as an ECM-filled space of irregular width (Fig. 8 A). Large micrometer-sized gaps were intermixed with narrower contacts (Fig. 8 B), which in turn consisted of intermediate contacts 100–300 nm wide and adherens junction–like close contacts <30 nm wide (Fig. 8, C–E), and compatible with cadherin adhesion (Farquhar and Palade, 1963; Miyaguchi, 2000; Tepass et al., 2000). Close contacts occupied 17 ± 1.5% of the cleft length (n = 12 embryos). At intermediate contacts, membranes of adjacent cells typically ran in parallel, bending sharply when they opened into larger spaces, suggesting that cell adhesion occurred across these contacts.

Between ectoderm–mesoderm cell pairs reaggregated in vitro for 20 min (Fig. 9 A), a similar pattern of contacts was observed. Large gaps were interspersed with intermediate and close adherens junction–like contacts (Fig. 9, B–D). Regularly spaced membranes of intermediate contacts diverged abruptly at the surface of cell pairs or at large gaps (Fig. 9 D), again suggesting an adhesive role for these structures. Patches of ECM decorated free cell surfaces and spanned between membranes up to 0.5 µm apart at contacts (Fig. 9, E and F). Knockdown of PAPC generated long stretches of super-close contacts with separation distances of 10 nm (Fig. 9, G and H). In occasional gaps, dense material resembling compacted ECM was present (Fig. 9 H). In PAPC morphant embryos, the fraction of close contacts at Brachet’s cleft increased to 48 ± 4.2% (n = 4 embryos).

Ultrastructural gaps between cell pairs were often narrow and perhaps inconsistent with the pronounced gaps observed by light microscopy. We therefore asked whether exclusion of fluorescent dextran from the cell periphery contributed to how wide gaps appeared. In reaggregated cell pairs, phalloidin staining revealed prominent F-actin cortices (Fig. 10, A–D). Between ectoderm–ectoderm or mesoderm–mesoderm pairs, no gaps were observed using dextran labeling, and F-actin staining at the interface was very weak (Fig. 10, A and B), consistent with a reduction in cortex density upon adhesion in these cells (David et al., 2014). At ectoderm–mesoderm contacts, F-actin staining overlapped considerably, though not completely, with the dextran-contrasted gaps between cells (Fig. 10, C and C′), in agreement with the presence of both large and narrow gaps at the ultrastructural level. F-actin staining was also peripheral to the dextran label at free cell surfaces above contacts (Fig. 10 D). Exclusion of dextran from the cortices of ectoderm–mesoderm pairs was also observed in living cells labeled with LifeAct (Fig. 10 E).

In conclusion, cleft contacts comprise a small fraction of adherens junction–like contacts located between stretches of intermediate-sized gaps, which appear to be adhesive, and large micrometer-scale gaps. All gaps are filled with ECM, and their normal maintenance requires PAPC expression. In contrast to contacts between cells of the same type, strong cell cortices are maintained at cleft contacts, which apparently exclude labeled dextran.

We identified two roles for PAPC in tissue separation at Brachet’s cleft. First, Fz7-dependent Snail1 expression enables PAPC to down-regulate PCP function in the mesoderm specifically at its boundary to the ectoderm, and we argue that this reduces adhesion across the boundary. Second, PAPC is necessary for the formation of a distinct type of cell–cell attachment: cleft contacts. This role is independent of Snail1 but requires Eph signaling and inhibition of PCP function. We synthesized these findings into a model of tissue separation at the ectoderm–mesoderm boundary (Fig. 7 I).

In Xenopus and zebrafish, C- and E-cadherin, respectively, ensure cohesion of gastrula tissue on either side of Brachet’s cleft (Kühl and Wedlich, 1996; Montero et al., 2005). In Xenopus mesoderm, this basic cell–cell adhesion is modulated. Eph/ephrin signaling lowers tissue cohesion, PCP signaling increases it, and the eventual strength of mesoderm cohesion is the result of these opposite effects. Evidence for adhesion promotion by PCP components was presented previously. In the zebrafish, interaction of Wnt11, Fz7, Dvl, and Flamingo (Fmi) increases cell contact persistence at respective puncta (Witzel et al., 2006), and E-cadherin adhesion is strengthened by Pk1a (Oteiza et al., 2010). In Xenopus, the PCP protein Fritz strengthens adhesion (Kim et al., 2010). Although we focused on Dvl2 and Pk1 puncta in our analysis, we cannot exclude that adhesion is also affected by diffuse PCP proteins at the cell membrane (Panousopoulou et al., 2013). Cohesion of Xenopus chordamesoderm is also modulated by PAPC (Chen and Gumbiner, 2006; Kraft et al., 2012), but our surface tension measurements indicated no general effect of M-PAPC on cohesion. We propose that cadherin-mediated adhesion, its reduction by Ephs/ephrins, and its compensatory up-regulation by PCP signaling represent a basic cohesion module in the mesoderm, and perhaps in the ectoderm (Fig. 7 I).

Boundary formation at Brachet’s cleft is not caused by general adhesion differences between tissues, i.e., by differential adhesion (Winklbauer, 2009). Thus, M-PAPC has no effect on tissue cohesion despite the strong induction of cell sorting (Ninomiya and Winklbauer, 2008) and ectopic boundary formation. Conversely, differential adhesiveness in Xenopus gastrulae induced by manipulations of cadherin expression does not promote cell sorting in the intact embryo and generates indistinct sorting boundaries in vitro (Ninomiya et al., 2012). Moreover, Eph/ephrin signaling is strongest and PCP puncta are diminished specifically at the cleft. Together, these findings suggest that tissue separation relies on local cell interactions at the boundary and not on overall cohesion differences.

At the core of boundary cell interaction lays a self/non–self-recognition function of PAPC (Fig. 7 I). We showed that the apposition of PAPC⁺ and PAPC^– cells is not only essential for normal tissue separation, but also at ectopic boundaries in the ectoderm, and in the chordamesoderm (Ninomiya et al., 2012). Apparently, PAPC does not mediate the interaction between specific cell types, but controls a self/non–self-recognition process that discriminates between PAPC⁺ and PAPC^– cells, leading to the formation of close contacts between cells of the same type and of cleft contacts between different types.

We propose that a difference between trans-binding and free PAPC is essential for recognition (Fig. 7 I). PAPC–PAPC interaction can occur within the mesoderm, but not at contacts with nonexpressing cells. In a population of Snail1-enabled PAPC cells, non-enabled PAPC in adjacent cells suffices to disrupt their separation. Even the secreted extracellular domain of PAPC prompts cell dispersal in an M-PAPC–expressing patch (Kim et al., 1998), suggesting that any occupancy of the PAPC extracellular domain by a homophilic binding partner disrupts its function in separation.

PCP puncta are present on the free surface of explants. Apparently, cells distinguish between neighbors lacking PAPC and the absence of adjacent cells altogether. Because ectopic boundaries can be induced in ectoderm and chordamesoderm, the required interaction between tissues should not be very specific. In Xenopus, PAPC interacts indirectly with ubiquitously expressed C-cadherin (Chen et al., 2009), and seeding M-PAPC–expressing BCR cells on a substratum of purified C-cadherin permits M-PAPC to reduce adhesion to this substratum (Chen and Gumbiner, 2006), probably mimicking conditions at the boundary. Thus, cadherin interaction at the boundary may be required to alert PAPC⁺ cells to the presence of PAPC^– cells (Fig. 7 I).

The PAPC recognition process triggers two independent effects (Fig. 7 I). First, PCP signaling is inhibited and Dvl and Pk1 puncta are reduced on the PAPC⁺ side of the tissue interface. A possible mechanism could be based on the physical interaction between the extracellular PAPC domain and Fz7 (Medina et al., 2004; Kraft et al., 2012). Binding of Fz7a to free PAPC could remove it from its PCP function. Such an effect could become independent of Snail1 at high PAPC concentrations and be responsible for PAPC-mediated BCR cell sorting (Chen and Gumbiner, 2006). Together with elevated Eph/ephrin signaling, local inhibition of the PCP adhesive function is bound to create a line of weak tissue cohesion. This is not sufficient, however, as no ectopic boundary forms between Dvl-depleted and normal ectoderm. Second, PAPC is needed to establish cleft contacts (Fig. 7 I). This function does not require Snail1 to enable PAPC; addition of PAPC to Dvl-MO cells is sufficient. Moreover, PCP inhibition does not require Eph signaling, whereas cleft contact formation does, confirming that the two effects of PAPC recognition are independent processes.

PAPC-dependent cleft contact formation constitutes a peculiar adhesive interaction where cells attach but can easily slip past each other. Adherens junction–like close contacts should contribute to cell attachment and allow for Eph/ephrin signaling across clefts. These contacts are infrequent, however, and they could permit lateral cell gliding by forming and dissolving rapidly. Pk/Dvl puncta have been implicated in contact persistence in zebrafish (Witzel et al., 2006), and we speculate that they could stabilize cadherin-based close contacts within a tissue, but not at the boundary where puncta are reduced. The increase of close contacts at the boundary upon PAPC knockdown increases puncta frequency, supporting this hypothesis. The simultaneous appearance of large blobs of extracellular material in gaps between cells suggests that PAPC may also regulate ECM distribution on cell surfaces at the cleft.

Actin cortex density is inversely correlated with adhesiveness in Xenopus gastrula cells (David et al., 2014). The presence of a stronger and perhaps more contractile cortex between mesoderm and ectoderm cells, as compared with homotypic cell pairs, would also be consistent with less adhesive and more dynamic cell contacts at the mesoderm–ectoderm interface, as indicated by lateral gliding of cells or their spontaneous detachments in mixed ectoderm/mesoderm cell reaggregates.

Clustering of cell adhesion receptors is often driven by anti-adhesive steric hindrance of close contacts in the regions between adhesion domains (Sackmann and Smith, 2014). Interestingly, however, the geometry of intermediate contacts suggests that they mediate adhesion, and lanthanum/alcian blue staining is consistent with an adhesive ECM or glycocalyx. However, we cannot exclude cell–cell attachment by very large nonclassical cadherins (e.g., Kazmierczak et al., 2007). Cell–ECM adhesion is common, but cell–cell adhesion mediated by an intervening thin layer of ECM is not well described other than in sponges, where large extracellular proteoglycan complexes are bound by cell surface receptors (Bucior and Burger, 2004; Bucior et al., 2004). Nonetheless, the combination of facilitated shear movement and substantial adhesion at cleft contacts seems well suited for mesoderm translocation on the BCR during gastrulation, and intermediate contact adhesion may essentially contribute to this property.

The PAPC/Snail1 module, PCP inhibition, and Eph signaling seem to collaborate during tissue separation at other cleft-like boundaries. PAPC and Snail1 are involved in notochord–somite boundary formation in Xenopus and zebrafish (unpublished data), and Snail1 expression determines this boundary in Ciona (Fujiwara et al., 1998). The requirement for Eph/ephrin signaling in Xenopus (Fagotto et al., 2013) and the absence of PCP puncta from this boundary in zebrafish (Yin et al., 2008) and Ciona (Jiang et al., 2005) are consistent with this hypothesis. PAPC, Snail1, and Eph/ephrin signaling are also required for somite segmentation (Holley, 2007). However, the specific roles of these components may differ in the different systems. For example, Snail1 expression on both sides of the boundary interferes with notochord–somite separation (Fujiwara et al., 1998), whereas PAPC does not (unpublished data). Examining these differences may lead to a deeper understanding of Snail1 and PAPC function during tissue separation at Brachet’s cleft.

Xenopus embryos were obtained and injected as described previously (Luu et al., 2008). Xenopus embryos obtained from in vitro fertilized eggs were dejellied with 2% cysteine in 1/10× modified Barth’s solution (MBS; 88 mM NaCl, 1 mM KCl, 2.4 mM NaHCO₃, 0.82 mM MgSO₄, 0.33 mM Ca(NO₃)₂, 0.41 mM CaCl₂, 10 mM Hepes [+NaOH], 1% streptomycin, and 1% penicillin, pH 7.4) with the pH adjusted to 8. After washing, embryos were cultured in 1/10× MBS in a 15°C incubator and staged according to Nieuwkoop and Faber (1967). Embryos at the 4-cell stage were injected in both dorsal blastomeres or animally into all four blastomeres using a Nanoinject II (Drummond Scientific Company). Embryos were kept in 4% Ficoll during injections. Transgenic CLGY21 (Ellingsen et al., 2005) and AB zebrafish stocks were maintained at 28°C and bred as described previously (Westerfield, 2000). CLGY21 fish carry a retrovirally inserted enhancer detection vector consisting of a murine leukemia virus retroviral vector (pCL) carrying the gene for yellow fluorescent protein and a 1-kb proximal promoter of the zebrafish GATA2 gene. The transgene is inserted 14,041 bp downstream of the snail1a gene on chromosome 11 (Ellingsen et al., 2005). Embryos were obtained from natural spawnings and staged as described previously (Kimmel et al., 1995). Embryos were maintained in embryo medium in a 28.5°C incubator. Microinjections were performed at room temperature, using an MPP1-2 pressure injector (ASI), into the yolk at the 1-cell stage, and embryos were then returned to the 28.5°C incubator (Bruce et al., 2003). Before microsurgery, embryos were manually removed from the chorion in embryos medium using forceps.

Microsurgery and the BCR assay in Xenopus and zebrafish were performed in MBS at room temperature. Vitellin membranes were removed manually with forceps. In Xenopus, the BCR assay was performed as described previously (Wacker et al., 2000): prechordal mesoderm or inner BCR test explants were placed on explanted BCR (typically five test explants per BCR), and after 1 h, test explants were scored for having remained on the surface of the BCR, using Stemi SV11 (Carl Zeiss) or MZ16F (Leica) stereomicroscopes. Each assay was repeated at least six times. The number of explants tested (n) is indicated in the respective figures. Statistical significance of differences was confirmed using the paired Student’s t test (Rohani et al., 2011). For zebrafish, the interior surface of deyolked shield stage CLGY21 embryos served as the ectodermal (epiblast) substratum, and control ectoderm test aggregates were acquired by scraping cells off the inner surface. For mesoderm aggregates, 70% epiboly stage CLGY21 embryos were bisected with a tungsten wire and deyolked, and dorsal anterior mesoderm cells were scraped off. Aggregates were placed onto prepared ectodermal caps, coverslipped, and scored as in Xenopus, using an Imager Z1 compound microscope (Carl Zeiss) with a 20× Plan-Apochromat objective, numerical aperture 0.75, and an Orca-ER camera (Hamamatsu Photonics) using Volocity 6.0 (PerkinElmer) software.

Excised tissues from fluorescein-dextran (FDA)–, cascade blue–dextran-, or rhodamine-dextran (RDA)–injected embryos were dissociated in Ca²⁺-free MBS, and cells were mixed and filmed (two to three independent repetitions per treatment) while reaggregating in MBS in BSA-coated dishes at room temperature using an Axiovert 200M inverted microscope with a 20× Plan-Neofluar objective, numerical aperture 0.5, an AxioCam MRm, and AxioVision 4.8 image processing software (all Carl Zeiss).

Tissue surface tension was measured using axisymmetric drop shape analysis (ADSA; del Río and Neumann, 1997; Luu et al., 2011). At equilibrium, the drop shape of a tissue explant reflects a compromise between tissue surface tension and gravity and is described by the Laplace equation of capillarity. The ADSA program numerically integrates the Laplace equation to generate theoretical drop shapes for different hypothetical surface tensions and optimizes the surface tension to find the shape that best fits an experimental drop profile. To obtain explant profiles, we combined small excised pieces of tissue collected from one (ectoderm) or three (mesoderm) embryos in a plasticine well for 1 h to form an aggregate, placed the aggregates (10–37 ectoderm and 4–12 mesoderm aggregates per treatment) on nonadhesive poly-2-hydroxyethyl methacrylate substrate for 2 h at room temperature, fixed them in 4% formaldehyde in MBS, and imaged the drop-shaped aggregates from the side in a 45° mirror using a Stemi SV11 stereomicroscope and AxioCam MRc camera (both Carl Zeiss). We determined the density of explants from different regions according to Ninomiya and Winklbauer (2008). Explants were placed in MBS layered on top of a mixture of MBS and Ficoll (19%, 20%, 21%, 22%, or 23%) in 1.5-ml centrifugation tubes and centrifuged at 100 g for 3 min. Explants were either sinking to the bottom or floating at the interface depending on the Ficoll mixture, whose density was determined by weighing a defined volume. We fitted explant profiles and explant densities using ADSA to obtain surface tension values.

For morpholinos, constructs, and concentrations used, refer to Tables S1–S4. The Xsnail1-MO overlaps the ATG start codon. It fully recognizes the mRNAs of both closely related (98% identity) Xenopus Snail1 genes (X53450.1, BC056857.1 and NM_001086456.1), but not the in vitro synthesized Xsnail1 mRNA used for injection. Chelerythrine was used at 6 µM and SP600125 at 10 µM.

In situ hybridization on Xenopus embryos was modified after Harland (1991). gsc pBluescript SK(−) (H. Steinbeisser, Heidelberg University, Heidelberg, Germany) was linearized with EcoR1 and Xbra DB30 pSP73 (M. Sargent, Medical Research Council National Institute for Medical Research, Mill Hill, London, England, UK) with Bgl2 (GE Healthcare), and digoxigenin-labeled antisense probe was synthesized using T7 RNA polymerase (mMessage mMachine; Ambion). Sna probe was generated using Bgl2 (GE Healthcare)-digested pMX51 and SP6 RNA polymerase (mMessage mMachine). Embryos were fixed in MEMFA for 1–2 h at room temperature, dehydrated in ethanol, and stored at −20°C. They were rehydrated in PBS/Tween, incubated with Proteinase K, rinsed with triethanolamine, refixed with 4% formaldehyde, prehybridized for 4 h at 60°C in 50% formamide, 5× SSC, 1 mg/ml Torula RNA, 100 µg/ml heparin, 1× Denhart’s, 0.1% Tween-20, 0.1% CHAPS, and 10 mM EDTA, and incubated with probe overnight at 60°C in hybridization buffer (prehybridization buffer plus 0.5 mg/ml tRNA). After washes in 2× SSC solutions and maleic acid buffer, probe was detected by incubating embryos with a 1:2,000 dilution of anti-digoxigenin AP, fab fragments of antibody from sheep (Roche) for 4 h, and washing with maleic acid buffer overnight and briefly with alkaline phosphatase buffer. Embryos were incubated in BM purple (Roche), and the color reaction was monitored and stopped with MEMFA. Between 18 and 58 specimens from at least two batches of embryos were examined for each treatment, 127 from uninjected controls, and photographed using a Stemi SV11 stereomicroscope and AxioCam MRc camera. Whole-mount in situ hybridization on zebrafish embryos was performed as described previously (Jowett and Lettice, 1994). Antisense probe was synthesized from EcoRI-digested pBS-snail1a (G. Kelly, Western University, London, Ontario, Canada) transcribed with T3 RNA polymerase. Embryos were fixed overnight at 4°C in 4% PFA, rinsed in PBT, and stored overnight at −20°C in 100% methanol. After rehydration into PBT, embryos were post-fixed for 20 min at room temperature in 4% PFA followed by PBT washes. Embryos were prehybridized for 3 h at 70°C in 65% formamide, 5× SSC, 50 µg/ml heparin, 0.1% Tween-20, and 9 mM citric acid, incubated with probe overnight at 70°C in hybridization buffer (prehybridization buffer plus 0.5 mg/ml tRNA), and washed in 2× SSC solutions and PBT. Probe was detected by incubating embryos overnight at 4°C with a 1:5,000 dilution of anti–digoxigenin-AP, fab fragments of antibody from sheep (Roche). After PBT washes, embryos were incubated with NBT/BCIP (Roche), and the color reaction was monitored and stopped by PBT washes. Between 35 and 67 embryos from at least two batches were examined for each treatment and 191 uninjected controls. Embryos in 80% glycerol were photographed at 5× on an MZ16F stereomicroscope with a Plan-Apochromat 1.0× objective using a MicroPublisher 3.3 camera (QImaging) using Openlab.

RT-PCR was performed at least twice for each experiment. For primers and temperatures used, see Table S3.

Xenopus and zebrafish embryos were processed for SEM as in Damm and Winklbauer (2011). Embryos were fixed in 2.5% glutaraldehyde (GA) in 0.1 M sodium cacodylate buffer (CB) overnight at 4°C, post-fixed in osmium tetroxide (Structure Probe) for 1 h at 4°C, washed in distilled water for 30 min, and dehydrated in an ethanol dehydration series (50% ethanol in 0.1 M cacodylate for 20 min, 100% ethanol twice for 20 min) and in hexamethyldisilazane (a 1:1 dilution of hexamethyldisilazane and ethanol for 30 min followed by 100% hexamethyldisilazane twice for 30 min). The specimens were left to dry overnight, mounted on SEM stubs using conductive carbon tape (Structure Probe), and sputter coated with gold-palladium (60%/40%) for 50 s. Specimens were imaged with a Hitachi S-2500 scanning electron microscope.

Stage 11 Xenopus embryos were fixed overnight at 4°C in 3% GA and 2% PFA in 0.05 M CB, pH 7.0, either with (seven embryos) or without (seven uninjected and four PAPC-MO–injected embryos) 1% lanthanum nitrate (LN) as an ECM stain (Johnson, 1977). Embryos were then rinsed in 0.1 M CB, pH 7.0, cut in half along the sagittal plane, and fixed overnight at 4°C in 0.1 M CB containing 1% OsO₄ or 1% OsO₄ and 1% LN if previously treated with LN. After rinsing with 0.1 M CB, embryos were dehydrated in a graded series of ethanol, embedded in 100% Spurr’s resin, and cured at 65°C for 24 h. Ultrathin (90–100 nm) sections were obtained using an EM UC6 microtome (Leica) and stained with 3% uranyl acetate in methanol for 1 h and Reynold’s lead citrate for 10 min. About 10–15 ectoderm–prechordal mesoderm cell contacts were present per embryo.

Cell reaggregate sample preparation was modified from a yeast cell protocol by Wright et al. (1988). BCR and mesoderm from stage 11 Xenopus embryos were dissociated and reaggregated as described in the “Reaggregation” section. Cells were allowed to reaggregate for 20 min on a glass-bottom Petri dish coated with 1% BSA. Cell reaggregates were then fixed in 0.05 M CB containing 3% GA and 2% PFA, either with or without 1% alcian blue as an enhancer for LN ECM staining, for 30 min at room temperature (Shea, 1971). Cells were rinsed with 0.1 M CB and embedded in 3% low-melt agarose overnight at room temperature. Slices containing embedded cells were excised, placed bottom-up in a Petri dish, and covered with a thin layer of 3% low-melt agarose with a glass coverslip on top. After 4 h at 4°C, slices containing embedded cells were excised and fixed for 1 h at room temperature in 1% OsO₄ in 0.1 M CB or 1% OsO₄ and 1% LN in 0.1 M CB if previously treated with alcian blue. Samples were then washed with 0.1 M CB, dehydrated in a graded ethanol series, and embedded in 100% Spurr’s resin. Ultrathin (90–100 nm) sections were obtained using an EM UC6 microtome. Two re-embedded agarose blocks were sectioned for each treatment (normal cells with and without LN/alcian blue staining and cells from PAPC-MO–injected embryos). Ultrathin sections were stained with 3% uranyl acetate in methanol for 1 h and Reynold’s lead citrate for 10 min. TEM images were collected on a Hitachi HT7700 microscope at 80.0 kV. 15 LN/alcian blue–stained ectoderm–prechordal mesoderm cell pairs, 6 unstained pairs, and 10 pairs with PAPC-MO–injected prechordal mesoderm were inspected closely. Ectoderm and mesoderm cells were distinguished by the size of yolk platelets (Nakatsuji, 1975) and ectodermal pigment granules.

To visualize Pk1 or Dvl2 puncta, embryos were injected with Pk1-venus, Dvl2-GFP, or membrane-RFP with or without Xsnail1-, Pk1-, or Dvl-MOs. Mesoderm pieces were explanted at stage 10, covered with ectodermal BCR, and gently pressed down with a piece of coverslip. Images were taken at room temperature from an Axiovert 100M LSM 510 confocal microscope (Carl Zeiss) at 40× magnification (Plan-Neofluar 40×/1.3 oil objective) using a PMT and LSM 3.2 WNT4.0 software (Carl Zeiss) to process images. The F-actin cortex in reaggregating cells was visualized by FITC-phalloidin staining of fixed cells (four experiments) or by expression of LifeAct in living cells (four experiments). For image collection, a TCS SP8 confocal microscope (Leica) with 20×/0.75 oil immersion, 63×/1.4 oil immersion, and 100×/1.4 oil immersion objectives and LasAF 3.2.0.9652 software or an Axiovert 100M LSM 510 confocal microscope at 40× magnification (Plan-Neofluar 40×/1.3 oil objective) using a PMT and LSM 3.2 WNT4.0 software was used. To visualize ectopic boundaries in the BCR, the border between uninjected and injected sides was visualized under incident light illumination, and in parallel by fluorescence microscopy to detect injected RDA, using an Axiovert 200M microscope with 20× Plan-Neofluar/0.5 objective, AxioCam MDm camera, and AxioVision 4.8 image processing software (all Carl Zeiss). To visualize the cell membranes of migrating prechordal mesoderm cells, embryos were injected with mRNA encoding membrane-RFP, and explants of the mesoderm were combined with BCR explants to induce directional cell migration (Damm and Winklbauer, 2011). Cells were observed using an Axiovert 100M LSM 510 laser-scanning confocal microscope at 40× magnification (Plan-Neofluar 40×/1.3 oil objective), using a PMT and LSM 3.2 WNT4.0 software to process images.

Fig. S1 shows that mesoderm marker gene expression is not affected by Xsnail1 knockdown and that snai1a mediates tissue separation in the zebrafish gastrula. Fig. S2 shows that PAPC and Snail1 control tissue separation without modulating Eph/ephrin expression. Fig. S3 demonstrates that complementary ectopic Snail1 and PAPC expression can generate a sorting boundary but not a cleft-like boundary. Fig. S4 provides details on cell sorting and gap contact formation between PAPC-expressing and -nonexpressing cells. Fig. S5 shows additional data on PCP inhibition at the tissue boundary. Tables S1 and S2 list the morpholinos and constructs used in this study, respectively. Table S3 lists the RT-PCR primers and annealing temperatures used in this study. Table S4 lists the constructs used for Xenopus and zebrafish experiments. Video 1 shows reaggregating BCR cells injected with M-PAPC mRNA and RDA (red) or FDA (green). Video 2 shows the same sequence in grayscale mode. Video 3 shows reaggregating cells similar to Video 1, except that M-PAPC–expressing cells were coinjected with EphB4-MO. Video 4 presents reaggregating prechordal mesoderm cells (RDA injected, red) and ectodermal BCR cells (FDA, green). Video 5 shows reaggregating mesoderm and BCR cells as in Video 4, but with PAPC-MO having been injected into mesoderm cells. Video 6 shows the same sequence as Video 5 but in the grayscale mode. Video 7 shows the behavior of Dvl2-GFP puncta (green) in mesoderm cells at the boundary to the ectodermal BCR where cell membranes are labeled with membrane-RFP (red). Video 8 shows the behavior of Dvl2-GFP puncta as in Video 7, but with PAPC-MO being injected into mesoderm cells.

We thank F. Fagotto, R. Habas, G. Kelly, R. Moon, M. Sargent, H. Steinbeisser, and E. Nishida for reagents and Henry Hong and Audrey Darabie of the CSB Imaging Facility for help with TEM and confocal microscopy.

This work was supported by grant MOP-53075 from the Canadian Institutes of Health Research to R. Winklbauer, grant 458019 from the Natural Sciences and Engineering Research Council of Canada to A.E.E. Bruce, and grant 019355 from the Canadian Cancer Society to R. Winklbauer and A.E.E. Bruce.

The authors declare no competing financial interests.