Coordinated assembly and release of adhesions builds apical junctional belts during de novo polarisation of an epithelial tube

Epithelia are one of the fundamental tissue types of the body, forming protective sheets of cells around the outside of the organism and lining the inside surface of many organs or parts of organs. All epithelia are polarised and have an apical and a basal surface that are molecularly and functionally distinct. In some cases, the organisation of the apical surface of an epithelium is generated at the free surface of an embryonic sheet of cells, whereas in others the apical surface emerges from within a dense rod or ball of cells to generate an epithelial tube or cyst (Martin-Belmonte and Mostov, 2008; Datta et al., 2011). Generating an apical surface from within a rod or ball of cells adds an extra layer of complexity to this process, because cells at the centre of these structures have to lose inappropriate connections to generate a free surface. They must also organise that free surface into a typical apical structure comprising a lattice of closely adherent cells connected by apicolateral belts of specialised junctions (typically adherens junctions and tight junctions in vertebrates). A key step, elucidated in epithelial development in flies, is the Crumbs-dependent redeployment of junctional proteins, such as Baz (Drosophila Pard3) and E-Cadherin from the apical-most surface of single-sheet epithelial cells to the apicolateral border of the cell (Grawe et al., 1996; Harris and Peifer, 2005; McGill et al., 2009; Morais-de-Sá et al., 2010; Pilot et al., 2006; Tepass, 1996). This builds the apicolateral junctional belts that tie the epithelial cells together (St Johnston and Ahringer, 2010; Wei et al., 2005). The extent to which similar mechanisms of apical surface development are used at a free surface in comparison to within a rod or sphere primordium, and how these can be coordinated with the remodelling of cell-cell connections necessary for de novo surface generation, is poorly understood, especially in vertebrates in vivo.

Here, we analyse the spatial deployment of the cardinal polarity protein Partitioning-defective 3 (Pard3) and the cell adhesion protein Cadherin 2, type 1, N-cadherin (neuronal) (Cdh2) during generation of the apical surface of the neural tube in the zebrafish embryo in vivo. Here, the apical surface is generated de novo from within the solid neural rod. The relative accessibility and transparency of zebrafish embryos provides an advantageous system in which to address the mechanism in whole embryos. We use experimental manipulations of the Mpp5a (also known as Nok or Pals1) scaffold protein and the endocytic recycling protein Rab11a. We determine the cellular and molecular mechanisms that release cell adhesions across the organ midline whilst simultaneously generating the canonical apical junctional belt organisation of epithelia within a solid primordium. We compare these with previously shown mechanisms that generate apical organisation at a free surface. We show that both Mpp5a and Rab11a are required to remodel connections between contralateral and ipsilateral cells and suggest that they operate through apical recruitment of the transmembrane protein Crumbs.

The apical surface of epithelia is characterised by a lattice-like arrangement of polarity and scaffolding proteins [such as Pard3, atypical protein kinase C (aPKC) and zona occludens 1 (ZO-1; also known as Tjp1a)] and cell adhesion proteins (such as Cdh2). We visualised this organisation before lumen opening in the sagittal plane of the zebrafish neural rod [stages 10-18 somites, 14-18 h postfertilisation (hpf)] at the level of the nascent anterior spinal cord (Fig. 1A,B). We generated a bacterial artificial chromosome (BAC) transgenic fish line that reports endogenous spatiotemporal expression of Pard3, a cardinal polarity protein, in live embryos [TgBAC(pard3:Pard3-EGFP)^kg301, referred to as ‘Pard3-EGFP’]. At early rod stages (11 somites), apical rings of Pard3-EGFP and ZO-1 were first established at the midline in ventral floor plate cells (Fig. 1B). In more dorsal areas of neuroepithelium, where apical rings had not yet formed, Pard3-EGFP and ZO-1 were seen as puncta (Fig. 1B, e.g. arrowheads), reminiscent of spot adherens junctions in other systems (e.g. Tepass, 1996). Over the next few hours of rod stage (≤17-somite stage) the lattice work of apical rings progressively builds from ventral to dorsal such that at any time point there is a developmental gradient along the dorsoventral axis (Fig. 1B,E). In contrast to a previous study (Guo et al., 2018), our results suggest that Pard3 is an early component of the nascent apical junction organisation, with both Pard3 and ZO-1 being expressed in a punctate manner during a dynamic phase of cell interdigitation across the neural midline (present results and Buckley et al., 2013).

To understand where in the cell Pard3 puncta first appear, we labelled Pard3-EGFP embryos with membrane-tagged mNeptune2.5 and obtained images before apical surface formation. Pard3-EGFP puncta were seen close to the midline around the perimeter of cells, often at cell vertices (Fig. 1C). Membrane-tagged mNeptune was not enriched at vertices, indicating that Pard3-EGFP enrichment is not simply attributable to more membrane at these points. Later, Pard3-EGFP puncta were localised progressively closer together and not confined to cell vertices (Fig. 1D). This suggests that apical rings might be built by adding Pard3 puncta at cell-cell interfaces until they coalesce to form a more-or-less continuous ring structure.

To uncover mechanisms that regulate formation of apical junctions in the neural rod, we examined puncta-to-apical ring transition in a mutant fish line that fails to open a lumen. Mpp5a is a MAGUK (membrane associated guanylate kinase) scaffolding protein (Wei and Malicki, 2002), essential to retain all forms of the transmembrane protein Crumbs at the apical surface (Zou et al., 2013). mpp5a mutants fail to generate and open a lumen in the neural rod and have disorganised adherens junctions (Lowery et al., 2009; Lowery and Sive, 2005). Although mpp5a^m227 mutants generated Pard3-EGFP puncta in the sagittal plane at the 16-somite stage, these puncta failed to undergo the puncta-to-apical ring transition and remained as punctate deposits (Fig. 1H). Mpp5a is thus crucial to build apical junctional rings.

To undergo de novo lumen formation successfully, the zebrafish must generate not one, but two lattice-like apical surfaces in the middle of a solid tissue, one on the left side and one on the right side of the future lumen. One possibility is that a single plane of apical rings might first develop at the left-right interface and then these pioneer rings split into left and right rings. To assess this, we analysed development of left and right rings. Imaging sagittal and parasagittal planes, we never saw an embryo containing a single plane of apical rings (Fig. 1F,G; Movie 1). Instead, rings were always found bilaterally, and incomplete apical rings were also found bilaterally on either side of the midline, showing that left and right rings are built independently. Furthermore, the developing rings on either side of the midline are always offset from one another, rather than having a mirror-symmetric arrangement that might be expected if individual rings split into two (Fig. 1G).

Most neural rod cells are derived from progenitor divisions that generate mirror-symmetrically polarised sisters connected to each other across the midline (the C-division; Buckley et al., 2013; Kimmel et al., 1994; Tawk et al., 2007). This process will, at least temporarily, align pairs of sister cells across the midline. However, the offset organisation of left and right apical rings later in development suggests that a mechanism exists to add offset to this initial alignment. To confirm that sister cells become offset, we generated mosaically labelled embryos and used live imaging to assess the interface between sister cells. Soon after a C-division, sister cells appeared mirror-symmetric to each other across the tissue midline (Fig. 2A, 13 somites). However, the shape of nascent apical endfeet changed over time, probably in response to movements and mitoses of neighbouring (unlabelled) cells (Fig. 2A; Movies 2-4). The location of the sister-cell interface in relationship to the rest of the cell shape therefore also changed until the neural rod stage, when the connection between the sisters was stably offset such that nascent apical endfeet of sister cells met at their corners (Fig. 2A, 18 somites). Thus, an initially mirror-symmetric sister-cell alignment is shifted to an offset organisation after C-division mitosis. The offset conformation suggests that contralateral cell connections across the midline are likely not to be confined to their C-division sisters.

To examine the dynamics of ipsilateral and contralateral cell adhesions during the process of the midline C-division, we used a BAC transgenic line expressing the cell adhesion protein Cdh2 fused to a tandem fluorescent timer (Revenu et al., 2014). Cdh2-GFP is expressed in neural rod cells over the entire plasma membrane. Before C-divisions, cells intercalated across the midline, and Cdh2-GFP was enriched in this interdigitation zone, showing that cells contact and adhere to several contralateral, in addition to ipsilateral, cells (Fig. 2Bi). When cells entered C-division mitosis near the midline, cell processes from neighbouring cells ingressed into the cleavage furrow, and distinct enrichments of Cdh2-GFP became largely localised to the cleavage furrow during telophase (red arrows in Fig. 2Bi). This cell ingression allowed neighbouring cells to contact and, presumably, adhere to both ipsilateral and contralateral daughters of the C-division. Cell ingression into the cleavage furrow therefore provides a potential mechanism to promote connections between multiple contralateral and ipsilateral cells (as shown schematically in Fig. 2Biii).

By the 17-somite stage, Cdh2-GFP was enriched at the interface between left and right cells, sometimes at the apical cell vertices (Fig. 2Bii). By this stage, the left-right interface was largely a straight line at the midline. To see whether the offset cell conformation between cells on the left and right persists until lumen opening, we examined electron micrographs taken at the start of lumen inflation. These showed a striking zig-zag conformation between cells separating across the midline (Fig. 2C). It appears that the offset conformation between left and right cells persists until the lumen inflates, separating the left and right sides of the neural rod.

In summary, these data suggest that the C-division can promote connections between multiple ipsilateral and contralateral cells during epithelialisation. We suggest that the maintenance of these connections until rod stage would promote an offset conformation between sister cells.

To understand the transition from spot-like junctional specialisations to two independent lattices of apical rings on left and right sides of the midline, we analysed the distribution of Pard3 and Cdh2 during this transition. Given that there is a developmental gradient along the dorsoventral axis of the neuroepithelium (Fig. 1B,E), we focused on a mid-dorsoventral plane of the anterior spinal cord. We imaged Pard3-EGFP and Cdh2-GFP from neural keel to neural rod stages in the horizontal plane (Fig. 3Ai-Aiii). Pard3-EGFP and Cdh2-GFP initially had different patterns of expression. At 11 somites (14 hpf), puncta of Pard3-EGFP were localised broadly around the midline, along interfaces between interdigitating cells. These correspond to the spot distribution of Pard3-EGFP seen sagittally (Fig. 1B,C). At the same time point, Cdh2-GFP was expressed at low levels throughout the cell membrane but also slightly elevated in the interdigitation zone, mostly in discrete spots. By 17 somites (17.5 hpf), both Cdh2-GFP and Pard3-EGFP were clearly enriched at the neural rod midline in a near-continuous, fairly straight expression domain at the interface between left and right cells. Consistent with a developmental gradient along the dorsoventral axis of the neuroepithelium, dorsal to this plane, Pard3-EGFP was still expressed in a discontinuous mediolateral stripe pattern, and ventral to this it appeared in two parasagittal domains either side of the midline (Fig. 3Av). The intermediate level that shows a single left-right interface of Pard3-EGFP expression occupies a small and transient dorsoventral domain that moves dorsally as the assembly of bilateral rings progresses from ventral to dorsal.

To analyse the distribution of cell membranes relative to the expression of Pard3-EGFP, we revealed membranes by injecting embryos with mRNA mNeptune2.5-CAAX. By 24 hpf, although spinal cells at almost all dorsoventral levels still met at the left-right interface (see Fig. 3vi), Pard3-EGFP and Cdh2-GFP were no longer localised to this interface; instead, their expression had transitioned from mediolaterally arranged puncta through a single midline domain and into two parasagittal domains either side of the midline (Fig. 3Ai,Aiv). These two domains were characterised by periodically elevated spots of expression that corresponded to points within the left and right apical rings. Much lower levels of Pard3-EGFP and Cdh2-GFP remained in the medial zone between bilateral apical rings (Fig. 3Ai,Aiv).

Although eliminating Crumbs in zebrafish is difficult because there are multiple Crumbs proteins, we took advantage of the mpp5a^m227 mutant that is unable to localise Crumbs proteins to the apical surface (Zou et al., 2013). Given that we demonstrated that the apical spot-to-ring transition did not occur in mpp5a^m227 mutants (Fig. 1H), we wanted to test the role of Mpp5a in displacing Pard3 and Cdh2 proteins from a single line at the left-right interface to the bilateral expression domains seen at later time points. For this, we generated a mpp5a^m227 mutant; Cdh2-tFT (Cdh2-tandem fluorescent protein timer) compound transgenic line. The ‘wild-type’ Cdh2-tFT embryos opened a lumen in the hindbrain and anterior spinal cord (Fig. 3Bi), whereas mpp5a^m227; Cdh2-tFT embryos failed to open a lumen (Fig. 3Bii), in line with previous observations of mpp5a mutant embryos (Lowery and Sive, 2005).

To understand the mpp5a^m227 phenotype in relationship to Cdh2 remodelling and its relationship to Crumbs expression, we examined their localisation in wild-type and mpp5a^m227 embryos. In wild-type embryos (Fig. 3Ci), Crb2a was first expressed at the neural rod 17- to 18-somite stage at the left-right interface, coincident with a single line of expression of Cdh2. However, by 26 somites, when Crb2a was still localised to the left-right interface, Cdh2-GFP protein was displaced to form two slightly more basolateral lines of expression. Cdh2-GFP expression was located immediately lateral to Crb2a, with little overlap of expression.

In mpp5a^m227 mutant embryos (Fig. 3Cii), Crb2a was absent from the midline, as expected. At the 18-somite stage, mpp5a^m227 mutant Cdh2-tFT embryos showed a single domain of enrichment of Cdh2-GFP at the left-right interface, similar to wild-type Cdh2-GFP embryos. However, unlike wild-type embryos, Cdh2-GFP remained as a single midline domain in mpp5a^m227 mutants, failing to undergo the transition to two lines of Cdh2-GFP either side of the midline, even up to stages as late as 28 hpf.

Together, these data show that left and right cells initially adhere together across the midline through contralateral adhesions. However, both Cdh2 and Pard3 are then redeployed more basally to the apicolateral border, where they contribute to adhesive junctions between ipsilateral cells to build coherent, bilateral sheets of neuroepithelium. This redeployment of junctional proteins depends on Mpp5a and is likely to be a crucial step in releasing cells from their contralateral adhesions to allow lumen formation.

To test whether Mpp5a-deficient cells retain persistent adhesions across the midline, we analysed the behaviour of clonally related cells in the hindbrain regions of 25 hpf wild-type embryos that had open lumens and mpp5a morphant embryos that were unable to open a lumen. In wild-type embryos, cells successfully separated across the midline after C-division; they maintained an elongated morphology that stretched from the apical to the basal surface, and later differentiative divisions (D-divisions) occurred at the apical surface (Fig. 4A). However, in mpp5a morphant tissue, cells were not separated across the midline, they often did not extend fully to the basal surface and they appeared clumped together, often in rosette-like structures. D-divisions in mpp5a morphants occurred within these cell clumps, often at a distance from the midline of the tissue (Fig. 4B,C). The behaviour of ectopic divisions in mpp5a morphants is accompanied by a lateral dispersal of Pard3 from the midline (Fig. 4D). Thus, in addition to being unable to undergo the puncta-to-apical ring transition (Fig. 1H), Mpp5a-deficient cells show persistent adhesions, and apical midline organisation becomes progressively fragmented over time.

Given that we hypothesise that Mpp5a functions through its ability to localise Crumbs to the nascent apical surface, we next analysed cell behaviours after another manipulation that depletes Crumbs from the apical surface. We expressed a dominant-negative (DN) form of Rab11a in rhombomeres 3 and 5. We previously showed that this manipulation reduces apical Crb2a localisation and prevents lumen opening in these segments (Buckley et al., 2013). Here, we show that apical Crb1 is also downregulated in DNRab11a rhombomeres (Fig. 4E), confirming that this manipulation is sufficient to downregulate apical trafficking of multiple Crumbs paralogues. Cells in DNRab11a rhombomeres did not separate after D-division and formed cell clumps and rosette-like structures with a central focus of Pard3 (Fig. 4F,G; Fig. S1). Similar to mpp5a morphants (Fig. 4D), a lateral dispersal of Pard3-EGFP from the midline was seen, progressively fragmenting midline organisation over time (Fig. 4H). This supports our previous observations of ZO-1 dispersion in DNRab11a segments (Buckley et al., 2013). Staining of DNRab11a rhombomeres revealed ectopic clumps of Pard3 and centrosomes, sometimes arranged in rosette-like structures, with centrally located Pard3 surrounded by centrosomes (Fig. 4I,J).

Interestingly, although cells situated in the middle of closed-lumen DNRab11a rhombomeres 3 and 5 formed clumps, cells at the edges of these rhombomeres often had a typical elongated neuroepithelial morphology. These were able to build an aPKC-positive apical surface that was contiguous with the wild-type apical surface in adjacent open-lumen wild-type rhombomeres 2, 4 and 6 (Fig. 5A). Edge cells were oriented obliquely along the anterior-posterior axis of the embryo, apical towards the open lumens (e.g. arrows in Fig. 5Aii). To determine how this morphology arose, we made time-lapse movies of DNRab11a/wild-type interface neuroepithelial cells. We found that interface cells remained connected across the midline as the neighbouring wild-type lumen inflated, acting like a hinge and reorienting cells at the DNRab11a/wild-type rhombomere boundary towards the opening lumen (Fig. 5Bi). Despite persistent connections between cells, the DNRab11a luminal surface expanded via cell division and subsequent reintegration into the epithelium (Fig. 5Bii). Cell divisions in the centre of DNRab11a rhombomeres were disorganised and did not orient along the tissue midline, whereas those occurring at the DNRab11a luminal surface occurred with parallel orientation (Fig. 5Biii). These results demonstrate that DNRab11a cells at the DNRab11a/wild-type interface are able to reintegrate successfully into the epithelium after division if they have access to an apical lumen surface.

Our data suggest that persistent adhesions between cells in Mpp5a-deficient and DNRab11a embryos both inhibit lumen formation and promote the formation of local clumps of cells. This results in the progressive fragmentation of midline organisation over successive rounds of division in Mpp5a-deficient and DNRab11a embryos.

A recent analysis of cell divisions and lumen formation in mpp5a^m520 mutants suggested that Mpp5a function in lumen formation might be related specifically to the resolution of connections between sister cells after the C-division (Guo et al., 2018). When analysing junction formation in mpp5a^m227 mutants, we discovered that the loss of lumen formation itself in these mutants was more complicated than expected. In line with previous observations (Lowery et al., 2009; Lowery and Sive, 2005;), we found the loss of lumen formation was not fully penetrant and we consistently found that small lumens formed at the level of the midbrain-hindbrain boundary (Fig. S2A). Less frequently, small lumens also formed in the dorsal aspects of the hindbrain at the level of the otic vesicles (Fig. S2B). We found that Crb2a protein was present at the apical surface of these lumens (Fig. S2A), consistent with the role of Crumbs in the successful formation of an apical surface and opening of a lumen. The ventricular walls of these small lumens were lined with Pard3-EGFP apical rings (Fig. S2B). We suggest the results of Guo et al. (2018) might be a consequence of their use of the weaker mpp5a^m520 mutant allele having non-penetrant phenotypes, as they previously published (Zou et al., 2013). We also suggest that the mini lumen at the midbrain-hindbrain boundary that we observed in mpp5a^m227 mutants (Fig. S2) is attributable to the expression of a second Mpp5a-like gene expressed in this region (such as mpp2b, Thisse et al., 2004). Mpp5a is a member of the Mpp MAGUK protein family, and there are six MPP genes (mmp2, mmp3, mmp4, mmp5, mmp6 and mmp7) with conserved protein domain structure in mammals. All these proteins contain a PDZ (PSD-95/Dlg-A/ZO-1) domain that, in the case of Mpp5, has been shown to be the specific domain that binds Crumbs (Roh et al., 2002). Little is known about the other MPP genes in zebrafish and this could form the basis of further study.

To address whether Mpp5a function might be related specifically to the resolution of connections between sister cells (as suggested by Guo et al., 2018), we analysed the key events of apical ring formation in mpp5a^m227 mutant embryos with and without C-divisions. Given that the closed-lumen phenotype is not fully penetrant in some hindbrain regions of the mpp5a mutants (Fig. S2), we suggest that using apical ring formation in the anterior spinal cord, rather than lumen opening in more anterior regions, is a better test of function. We blocked C-divisions using the S-phase inhibitor aphidicolin and analysed Pard3-EGFP in the anterior spinal cord region. Apical rings of Pard3-EGFP were not rescued by blocking C-divisions in mpp5a^m227 mutants (Fig. 6). The efficacy of the division block was confirmed by the enlarged nuclear size in aphidicolin-treated embryos (Fig. 6F,G) and by the fewer enlarged apical rings in the aphidicolin-treated siblings (Fig. 6B,H). These results demonstrate that, although necessary for cell-cell separation after division, Mpp5a function is not specifically related to remodelling adhesions between C-division sisters. Instead, it plays a more fundamental role in the formation of apical junctional rings and release of contralateral adhesions during de novo polarisation of an epithelial tube.

The development and remodelling of apical epithelial junctions is likely to be regulated by a complex integration of molecular interactions involving polarity proteins and adhesion proteins (Martin-Belmonte and Mostov, 2008; St Johnston and Ahringer, 2010; Datta et al., 2011), together with biomechanical forces that interpret tissue tensions generated by the actomyosin cyctoskeleton (Pinheiro and Bellaiche, 2018). Here, we have investigated some of the cellular and molecular features that underlie de novo apical surface generation within the solid rod primordium of the embryonic zebrafish central nervous system. Our work uncovers three aspects of de novo apical surface generation within a complex organ in vivo. First, despite the very different cellular organisations, apical surface generation at the centre of a solid organ primordium follows a strikingly similar sequence of spot-to-ring junction building to those previously described for apical surface generation at a free surface of a sheet of cells. Second, the displacement of proteins from the nascent apical surface to the nascent apicolateral junctional belt domain serves two purposes in a solid organ primordium; simultaneously, it contributes to the building of epithelial junctional belts and releases adhesions with contralateral cells that would otherwise inhibit the generation of a free surface and formation of a lumen. Third, neuroepithelial cell divisions across the nascent apical plane (C-divisions) do not promote exclusive connections between contralateral sister cells. Instead, connections between multiple ipsilateral and contralateral partners are promoted by neighbour ingression into the cleavage furrow of the C-division and appear to contribute to the offset alignment of contralateral cells across the midline. Through this realignment, sister cells do not remain connected ‘en face’, as previously suggested (Guo et al., 2018), but instead the sisters become attached in a configuration offset from one another across the midline. In addition, we address a recent suggestion that the MAGUK scaffolding protein Mpp5a is only required to remodel apical protein location after cell divisions across the nascent apical plane (Guo et al., 2018). We show that this is not the case, because Mpp5a-dependent apical protein remodelling is required for apical surface generation both with and without cell divisions.

We describe a similar displacement of Pard3 and Cdh2 to the apicolateral border of zebrafish neuroepithelial cells to that seen in Drosophila ectodermal cells (Grawe et al., 1996; Harris and Peifer, 2005; McGill et al., 2009; Morais-de-Sá et al., 2010; Pilot et al., 2006; Tepass, 1996). In the zebrafish neural rod, the displacement occurs at the midline within a solid rod of cells rather than at a free surface, but despite this fundamental difference in tissue architecture, a similar mechanism appears to build the apicolateral junctional belts. We show that the scaffold protein Mpp5a is necessary for the displacement of Pard3 and Cdh2 and for the formation of apical rings. In line with the Drosophila literature mentioned above, we propose that Mpp5a operates through its role in scaffolding Crumbs to the apical surface. This is consistent with the appearance of Crumbs at the nascent apical surface when Pard3 and Cdh2 are displaced from this surface and the lack of Crumbs localisation to the nascent apical surface in mpp5a mutants (Fig. 3C; Zou et al., 2013). A direct test of Crumbs function is difficult in zebrafish because three different Crumbs proteins (Crb1, Crb2a and Crb2b) are present at this stage in the developing neural rod. Previous research has shown that neuroepithelial organisation was altered only when all paralogues of Crumbs protein were knocked down by multiple morpholinos (Zou et al., 2013) and, consistent with our hypothesis, this appears to include a lack of lumen formation. Rather than transiently knocking down all paralogues of Crumbs proteins in this study, we used a Mpp5a mutation (mpp5a²²⁷) that downregulates all paralogues of Crumbs protein from the nascent apical surface. In this case, neither Pard3 nor Cdh2 was displaced from the nascent apical surface, and this resulted in persistent adhesions between left and right sides of the neural rod and a loss of lumen formation. Additionally, we found that a stable genetic manipulation of Rab11a that also lacks Crumbs at the nascent apical surface (Roeth et al., 2009; Buckley et al., 2013) also showed persistent adhesions between contralateral cells, a lack of lumen opening and, at later stages, comparable disorganisation of neuroepithelial tissue to mpp5a mutants (Fig. 4).

Despite its function in mediating epithelial cohesion via homophilic interactions of its extracellular domain (Thompson et al., 2013), the role of Crumbs protein in allowing cells to resolve their apical adhesions is emerging as another important mechanism for the control of epithelial morphogenesis. For example, a recent study demonstrated that a secreted version of Crb2 acted in a non-cell-autonomous manner to cause delamination of neuroepithelial cells during dorsal collapse of the spinal cord central canal in mouse embryos (Tait et al., 2020). By studying dominant-negative Rab11a cells in an environment where they have access to wild-type neighbours, we uncovered that cells depleted of Crumbs from their apical surface can nonetheless form an epithelial surface when adjacent to wild-type cells. Thus, in the fish neuroepithelium, Rab11a and Crumbs appear to be required for de novo epithelial surface generation but dispensable for integration next to an already generated apical surface. This suggests that non-cell-autonomous rescue from neighbouring wild-type cells is able to initiate epithelialisation in the absence of Rab11a/Crumbs. It would be interesting to determine whether secreted Crumbs protein plays a role in this rescue.

The zebrafish neural rod is initially a solid primordium, meaning that the displacement of Pard3 and Cdh2 is not from a free apical surface but instead occurs within the densely packed cell-cell interfaces that lie at the tissue centre. Cell adhesions between contralateral cells across the neural rod initially bind left and right halves of the neural rod together, but these adhesions need to be released to generate the lumen and apical surface of the neuroepithelium. We show that Mpp5a neatly coordinates the release of contralateral connections with the strengthening of ipsilateral connections, thus leading both to the generation of a free surface within the centre of an initially solid tissue and to junctional belt formation and epithelialisation. Mpp5a-deficient cells were unable to form a coherent sheet of apical rings (Fig. 1H). Additionally, in the absence of Mpp5a or the presence of DNRab11a, cells were unable to release cell-cell adhesions. During epithelialisation, these defects first resulted in closed lumens (Figs 3B and 4Hi). This is in line with our previous findings (Buckley et al., 2013) and the findings of others that apical localisation of Rab11 is necessary for cell abscission and normal opening of Kupffer's vesicle in zebrafish (Rathbun et al., 2020). It would be interesting to know whether similar mechanisms also operate in the caudal segments of avian and mammalian spinal cord. The neural tube in these segments forms through the process of secondary neurulation that also involves the de novo formation of a central lumen within a condensing rod of cells (see Schoenwolf, 1984, Dady et al., 2014), rather than epithelial folding that generates the neural tube in more rostral segments of these animals.

At later stages of development in the zebrafish neural tube, when neuroepithelial cells undergo interkinetic nuclear migration and D-divisions, the absence of Mpp5a or the presence of DNRab11a resulted in progressive disintegration of midline organisation (Fig. 4D,H). This might be attributable to a combination of factors; during cell division, interkinetic nuclear migration and mitotic rounding pull cells around within the tissue. In a normal epithelium, the apical surface would be stabilised by the lattice of ring-like junctional belts that hold the apical surface together. But when junctional belts between ipsilateral cells are not present, the tissue fails to hold cells at the apical plane, and cells are unable to re-extend fully to the basal side of the neural rod after division. Additionally, a lack of cell separation after division promotes the formation of local clumps of cells. Together, these defects cause divisions to occur in ectopic locations, and the midline organisation disintegrates.

Our results suggest that an abnormal persistence of cell adhesions is a key part of the Mpp5a and DNRab11a phenotypes and suggests that a lack of cell-cell separation might contribute to other epithelial disorganisation phenotypes in Crumbs- or Rab11a-deficient epithelia (Das and Knust, 2018; Roeth et al., 2009).

A recent paper also concluded that Mpp5a is necessary to remodel junctional proteins during epithelialisation in the neural rod (Guo et al., 2018); however, these authors propose that a role of Mpp5a in apical remodelling is necessary only to rescue potential tissue disruption caused by midline C-divisions, a conclusion based on their reported rescue of lumen formation and apical junctional belts when cell division is blocked in the mpp5a^m520 mutant. However, we show that blocking cell division does not rescue junctional belt formation in mpp5a^m227 mutants (Fig. 6) and has a more fundamental role in apical junction remodelling than resolving cell connections during C-divisions.

We previously demonstrated that C-divisions at the neural rod midline are an important morphogenetic force during neural tube formation. Ectopic divisions are able to organise ectopic neural lumens and duplicate the neural tube (Tawk et al., 2007), whereas neural tubes generated in the absence of C-divisions are less efficient at resolving cell interdigitation across the midline and have a disorganised morphology (Buckley et al., 2013). Here, we show that connections between ipsilateral and contralateral cells are pulled into the cleavage furrow of C-divisions (Fig. 2). This ingression of neighbours into the cleavage furrow is reminiscent of neighbour ingression during mitoses in Drosophila epithelia (Herszterg et al., 2013). Invasion of neighbours into the cleavage furrow also occurs during chick gastrulation, where it can be regulated developmentally to promote or impede cell intercalation between daughter cells, hence contributing to the control of morphogenetic movements (Firmino et al., 2016). In zebrafish, we show that multiple contralateral cell contacts are promoted through ingression of neighbouring cell processes into the cleavage furrow. We propose that the maintenance of connections with multiple contralateral cells through C-divisions ensures that cells do not connect solely to their sisters and suggest that this is one mechanism leading to staggered connections across the midline. This staggered alignment of endfeet might generate a morphogenetic advantage by allowing attachment to a larger number of cells than a one-to-one ‘mirror adhesion’ and therefore helps to colocalise apical junctions from multiple neighbouring cells to the midline. The localisation of junctional proteins at cell corners might also be important for allowing flexibility of cell movement during tissue remodelling (Finegan et al., 2019).

It is also likely that actomyosin-generated forces at the neural rod midline will contribute to apical junction assembly. Both actin (Yang et al., 2009) and non-muscle myosin (Gutzman and Sive, 2010) accumulate at the developing apical plane, and there is substantial evidence that biomechanical forces play a role in adherens junction dynamics in other systems (Pinheiro and Bellaiche, 2018). Recent findings also show that actomyosin-mediated tension is necessary for assembling ZO-1 at tight junctions during zebrafish gastrulation (Schwayer et al., 2019). In addition to the role of microtubules (Buckley et al., 2013), it would therefore be interesting to determine the role of actomyosin mechanics in the assembly of apical junctions within the zebrafish neural rod.

All embryos were collected, staged and cultured according to standard protocols (Kimmel et al., 1995). All procedures were carried out with UK Home Office approval and were subject to local Ethical Committee review.

To generate the transgenic zebrafish line expressing Pard3-EGFP with endogenous spatial and temporal expression, we used BAC recombineering. We replaced the stop codon of the pard3-003/ASIP transcript (von Trotha et al., 2006), in the BAC clone DKEY-71E21 (Source Bioscience). This BAC clone contains ∼71 kb upstream of the pard3-003/ASIP start codon and ∼30 kb downstream from the targeted stop codon. We followed the protocol of Bussmann and Schulte-Merker (2011), with some modifications. The original protocol leaves a kanamycin selection cassette in the final BAC construct. However, we found that this led to cell death and a lack of endogenous Pard3-EGFP expression. To remove the kanamycin selection cassette, we substituted two reagents with those from the recombineering protocol described by Sarov et al. (2006) and added an extra ‘flip out’ step to remove the kanamycin cassette. Specifically, we used the pRedFlp4 and R6k-EGFP-FRT-kanR-FRT cassettes from Sarov et al. (2006). The EGFP-FRT-kan-FRT targeting cassette was PCR amplified with 50 bp homology arms at either end, to target it to replace the pard3-003/ASIP stop codon during BAC recombineering: forward primer, 5′-CACAGAAGCAGAACGGACGCAATGGACACCCCTCCACTTCAGACAGGTACAGCTCAGGAGGTAGCGG-3′; reverse primer, 5′-AATTGAGTTTCATGATAGAACTTTGTATTTCTGCAATTCTGAAAAGCTGAGGCAGATCGTCAGTCAG-3′. Nucleotides in bold amplify the EGFP-FRT-kan-FRT targeting cassette. The nucleotides in regular text are the 50 bp homology arms that target the cassette to replace the pard3-003/ASIP stop codon during BAC recombineering.

The protocols for pRedFlp4 transformation, targeted recombination steps and removal of the kanamycin resistance selection marker by pRedFlp4-induced flipase were carried out as previously described (Sarov et al., 2006), with slight modifications. Briefly, the recombineering pipeline was modified to have two rounds of tagging by Red/ET recombination; the first was insertion of iTol2_amp into the BAC backbone and the second insertion of EGFP-FRT-kanR-FRT. Successful recombination was assessed by colony PCR followed by full sequencing of the sites of recombination. Preparation and injection of BAC DNA and screening for transgenic founders was all performed according to standard protocols (Bussmann and Schulte-Merker, 2011; Suster et al., 2011).

The TgBAC(pard3:Pard3-EGFP)^kg301 was generated as detailed above. The TgBAC(cdh2:Cdh2-tFT) (Revenu et al., 2014) was kindly provided by Darren Gilmour. The TgBAC(cdh2:Cdh2-tFT) line expresses both Cdh2-sfGFP and Cdh2-tagRFP, but we image only Cdh2-sfGFP expression in our experiments and refer to this as Cdh2-GFP for simplicity. These lines were bred with the Mpp5a^m227 mutant fish line (Wei and Malicki, 2002) to generate compound Mpp5a^m227;Cdh2-tFT and Mpp5a^m227;Pard3-EGFP fish. To prevent ventricle opening in rhombomeres 3 and 5, we crossed the UAS (upstream activating sequence)-inducible dominant-negative Rab11a line, Tg(UAS:mCherry-Rab11a S25N)^mw35 (Clark et al., 2011), with Tg(Egr2a:RFP-KalTA4), which drives the optimised Gal4-activator, KalTA4, only in Egr2a⁺ rhombomeres 3 and 5 (Distel et al., 2009). This resulted in the expression of Rab11a-S25N specifically in rhombomeres 3 and 5, as previously described (Buckley et al., 2013).

Fusion constructs containing cDNA in the pCS2+ vector were linearised, and mRNA was synthesised using the SP6 mMessage mMachine kit (Ambion, AM1340). RNA for the following constructs was injected using standard protocols (Westerfield, 2000) at 40-100 pg per embryo: Human Histone 2A tagged with GFP (H2A-GFP), Human Histone 2B tagged with RFP (H2B-RFP), Human CAAX membrane moiety tagged with EGFP, mCherry or mNeptune2.5 (EGFP-CAAX, mCherry-CAAX or mNeptune2.5-CAAX), Dominant-negative Human rab protein 11a tagged with EGFP (RAB11A-S25N-EGFP) and Zebrafish partitioning-defective 3 tagged with GFP or RFP (Pard3-GFP/RFP). For ubiquitous distribution of mRNA, embryos were injected at the one- to two-cell stage. For mosaic labelling, a single blastomere of a 16- to 64-cell-stage embryo was injected.

Co-injection of mpp5a splice blocking morpholinos (donor: 5′-GTTTATGACACCCACCTAGTAAAGC-3′ and acceptor: 5′-CTCCAGCTCTGAAAGTACAAACACA3′) was made into one-cell-stage embryos (Hsu et al., 2006). Full loss of Crb2a from the midline and associated phenotypes were seen with ∼1.7 nl of 300 mM (0.5 pM) of each morpholino, which closely phenocopied the mpp5a mutant phenotype (Fig. S3; Hsu et al., 2006). Mild to intermediate loss of Crb2a from the midline was seen with ∼0.5-1.0 nl of 200-300 mM (0.1-0.3 pM) of each morpholino. The level of Crb2a loss was correlated with the extent of associated phenotypes (see Fig. S3). The standard morpholino sequence was 5′-CCTCTTACCTCAGTTACAATTTATA-3′.

In order to image cell behaviour in a Mpp5a-deficient background, we sometimes used mpp5a morpholino knockdown rather than mpp5a mutant embryos because we could not identify mutant embryos before imaging. We therefore confirmed that the phenotypes were comparable between mpp5a morphants and mutants (Fig. S3). We used a previously characterised morpholino (Hsu et al., 2006) and found that a morpholino concentration of 0.5 pM resulted in the loss of detectable Crb2a protein from the nascent apical surface and recapitulated the closed lumen phenotype of mutant embryos.

Cell division was blocked by incubating dechorionated embryos in 300 µM aphidicolin (Sigma) in 4% dimethyl sulfoxide (DMSO) from the bud stage (10 hpf) to the 28-somite stage (23 hpf). Control embryos were incubated in 4% DMSO only. Embryos were maintained on a bed of agarose in 12-well plates in the dark throughout the drug incubation.

Embryos were fixed with 4% paraformaldehyde for 2 h at room temperature before processing for immunohistochemistry. The following primary antibodies were used: mouse anti-Crb2a (mouse monoclonal, ZIRC, zs-4, 1:200), rabbit anti-Crb2a and anti-Crb1 (a kind gift from the Wei laboratory, 1:350), mouse anti-ZO-1 (mouse monoclonal, Invitrogen, 33-9100, 1:500), rabbit anti-aPKC (rabbit polyclonal, Santa Cruz, sc-216, 1:300), mouse anti-gamma-tubulin (mouse monoclonal, MilliporeSigma, T6557, 1:200) and mouse anti-ZO-1 (Invitrogen, 33-910, 1:300). Rabbit anti-Pard3 (rabbit polyclonal, Millipore, 07-330, 1:100) was used after fixation in Dent's fixative for 3 h at room temperature and rehydration from methanol. Alexa Fluor secondary antibodies (Life Technologies) were used at 1:500, and Hoechst (Life Technologies) nuclei stain was used at a final concentration of 1:10,000.

Embryos were fixed in 2% paraformaldehyde and 1.5% glutaraldehyde in 0.1 M sodium cacodylate buffer overnight and then processed through 1% osmium and stained in 0.5% uranyl acetate. They were then dehydrated in ethanol and processed through propylene oxide into 100% resin, embedded and baked. Ultra-thin sections were taken in transverse orientation and imaged in a transmission electron microscope.

Embryos were mounted in low-melting-point agarose and imaged using a Zeiss LSM 880 Fast Airyscan microscope and water-dipping ×20/1.0 NA objective, a Leica SP5 confocal microscope and water-dipping ×25/0.95 NA objective or a PerkinElmer Ultraview spinning disc microscope. Imaging at the centre of a developing live tissue mass required the improved resolution in x, y and z coordinates obtained by using the Fast Airyscan mode of the Zeiss LSM 880 Fast Airyscan microscope. By using this microscope, high-resolution images of both en face Pard3-EGFP rings and the spatial location of Crb2a, aPKC and Cdh2-EGFP in the developing embryo were obtained. To improve lateral imaging, fish were mounted on an agarose bed with small depressions to hold the yolk of the embryo in place. These agarose beds were made using a custom mould (a kind gift from Andrew Oates). Data were collected from the hindbrain and anterior spinal cord regions. In Fig. 4F, some overlying cells have been manually edited from image to increase clarity of the cells of interest. Unedited versions of these images can be seen in Fig. S1. Images were processed using Imaris, Volocity and Fiji/ImageJ.

The custom R code used to generate plots for Figs 1 and 3 is available at https://github.com/andyivanhoe/pard3-analysis.

To obtain apical ring density counts across the developing neuroepithelium (Fig. 1E), Pard3-EGFP embryos were live-imaged en face between the 15- and 21-somite stages using the Zeiss Airyscan. The dorsoventral axis of the resulting z-stacks was divided into four quadrants (dorsal, mid-dorsal, mid-ventral and ventral), by applying a grid in Fiji. Apical rings were counted when all sides of the cell border had Pard3-EGFP signal along them.

To quantify the changing distribution of Pard3-EGFP across the neural rod midline over time, we imaged six live TgBAC(pard3:Pard3-EGFP)^kg301 embryos over a period of 3 h using Zeiss Airyscan confocal acquisition and processing. We then made a maximum-intensity projection of a 10-µm-deep volume at a mid-dorsoventral level of the neural rod. A mediolateral line of 116 µm width was then drawn across the basal-to-basal width of the developing neuroepithelium and the mean fluorescence intensity at each pixel along that line measured using Fiji. This was repeated for six embryos at 1 h intervals (Fig. 3Aii,Aiii; 0 h is 16-somite stage). All line profiles from different embryos and time points were aligned on the maximum intensity value for each line profile. The central 70 µm of the line profiles, centred on the maximum intensity values, were then used to calculate the mean intensity and standard deviation at each point along the line for each time point before plotting. We aligned on the maximum intensity value for each line profile because the exact maximum intensity of Pard3-EGFP relative to the tissue midline exhibits slight biological variation between embryos.

To quantify the distribution of Pard3-EGFP at the midline at the fully neuroepithelial stage (after the single midline peak had resolved into what appeared to be two near-parallel lines of expression on either side of the midline), we analysed a single z-level in the same six embryos where the two lines of expression were ∼4 µm apart (Fig. 3Aiv). This analysis was done at the last time point of the 3 h timeline, but at a more ventral (hence, more morphogenetically advanced; see Fig. 1 and associated text) position. In this case, we measured intensity along a 23-µm-wide mediolateral line. Given that the midline expression was not perfectly straight and the peaks of expression were only 4 µm apart, for this analysis we chose a single z-level and a thinner mediolateral line than for the timeline measurements to avoid averaging out the two peaks.

Plots in Figs 4C,J and 7G,H were generated using Graphpad Prism. Specific statistical analyses are described in the figure legends.

We thank Rachel Moore, Christopher Rookyard, Vineetha Vijayakumar and Xuan Liang for helpful discussions on this work. We thank Florent Campo-Paysaa for providing some of the movies analysed in Fig. 4H and 5B. We thank the Centre for Ultrastructural Imaging at King's College London and the Cambridge Advanced Imaging Centre for their help with processing electron microscopy samples. We thank the zebrafish facility at King's College London for their help in maintaining zebrafish stocks. We thank Darren Gilmour for the Cdh2:(Cdh2-tFT) line, Gwyn Gould for the RAB11A-S25N-EGFP plasmid, Xiangyun Wei for the Crb2a antibody, and Stefan Schulte-Merker and Mihail Sarov for BAC plasmid reagents.

The peer review history is available online at https://dev.biologists.org/lookup/doi/10.1242/dev.191494.reviewer-comments.pdf