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

Using the zebrafish neural tube as a model, we uncover the in vivo mechanisms allowing the generation of two opposing apical epithelial surfaces within the centre of an initially unpolarised, solid organ. We show that NOK/Pals1/Mpp5a and Rab11a play a dual role in coordinating the generation of ipsilateral junctional belts whilst simultaneously releasing contralateral adhesions across the centre of the tissue. We show that Nok and Rab11a mediated resolution of cell-cell adhesions are both necessary for midline lumen opening and contribute to later maintenance of epithelial organisation. We propose these roles for both Nok and Rab11a operate through the transmembrane protein Crumbs. In light of a recent conflicting publication, we also clarify that the junction remodelling role of Nok is not specific to dividing cells.


Introduction
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 while in others the apical surface emerges from within a dense rod or ball of cells to generate an epithelial tube or cyst (Martin-Belmonte et al., 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 as cells at the centre of these structures have to lose inappropriate connections to generate a free surface as well as organising 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, seems to be the remodelling of junctional protein location such as Par3 and Cadherin from the developing apical surface to the apico-lateral border in order to build apico-lateral junctional belts (Grawe et al., 1996, Harris & Peifer, 2005, Morais-de-Sa et al., 2010, Tepass, 1996. To what extent similar mechanisms of apical surface development are employed at a free surface compared 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 N-cadherin (Cdh2) during generation of the apical surface of the neural tube in the zebrafish embryo in vivo. In this system the apical surface is generated de novo from within the solid neural rod and the relative accessibility and transparency of the zebrafish embryo provides an advantageous system in which to address mechanism in whole embryos. We use experimental manipulations of the NOK/Pals1/Mpp5a scaffold protein and the endocytic recycling protein Rab11a to determine the cellular and molecular mechanisms that simultaneously remodel cell-cell connections in order to release cell adhesions across the organ midline and generate the canonical apical junctional belt organisation of epithelia within a solid primordium. We compare these to previously shown mechanisms that generate apical organisation at a free surface. Our results show that both Mpp5a and Rab11a are required to remodel connections between contralateral and ipsilateral cells and we suggest they operate through apical recruitment of the transmembrane protein Crumbs.

Apical rings of Pard3 and ZO-1 are built up from the ventral floor plate
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), Zonula occludens-1 (ZO-1)) and cell adhesion proteins (such as Cdh2). We visualised this organisation prior to lumen opening in the sagittal plane of the developing zebrafish neural rod (stages 10 to 18 somites, 14 to 18 hours post fertilization, hpf) at the level of the nascent anterior spinal cord ( Figure 1A, B).
We generated a BAC transgenic fish line that reports the endogenous spatiotemporal expression of Pard3, a cardinal polarity protein, in live embryos (TgBAC(pard3:Pard3-EGFP) kg301 , referred to as 'Pard3-EGFP'). We also used an antibody for the MAGUK (membrane association and presence of the GUK domain) scaffolding molecule ZO-1 (also known as tight junction protein) on fixed tissue.
At early neural rod stages (11 somite stage), we saw that apical rings of Pard3-EGFP and ZO-1 were first established at the midline in ventral floor plate cells ( Figure 1B, e.g. arrows). In more dorsal areas of the neuroepithelium, where apical rings had not yet formed, Pard3-EGFP and ZO-1 were seen as puncta ( Figure 1B, e.g. arrowheads), reminiscent of spot adherens junctions in other systems (e.g. Tepass et al., 1996). Over the next few hours of rod stage (up to 17 somite stage) the lattice work of apical rings progressively builds from ventral to dorsal such that at any given timepoint there is a developmental gradient along the dorso-ventral axis of the nascent neuroepithelium ( Figure 1B, E). In addition, 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 expressed in a punctate manner during a very dynamic phase of cell interdigitation across the neural keel midline (current results and Buckley et al., 2013).

Pard3 puncta initially appear at cell vertices
To gain a better understanding of where in the cell Pard3 puncta first appear, we labelled Pard3-EGFP embryos with membrane-tagged mNeptune2.5 and imaged at neural keel stage prior to apical surface formation. We found Pard3-EGFP puncta were seen close to the midline around the perimeter of the cells, often at cell vertices ( Figure 1C, e.g. arrowheads). Membrane tagged mNeptune was not enriched at these vertices, indicating the Pard3-EGFP enrichment is not simply due to more membrane at these points.
Later in development, Pard3-EGFP puncta were localised progressively closer together and not confined to cell vertices ( Figure 1D). This suggests that apical rings may be built by adding Pard3 puncta at cell-cell interfaces until they coalesce to form a more or less continuous ring structure.

nok m227 mutants fail to undergo puncta to apical ring transition
To begin to uncover the mechanisms that regulate formation of the apical junctions in the sagittal plane of the neural rod we examined the apical puncta to ring transition in a mutant fish line that fails to open a lumen. Nok/Mpp5a is a MAGUK scaffolding protein (Wei & Malicki, 2002), essential to retain all forms of the transmembrane protein Crumbs at the apical surface (Zou et al., 2013) and nok mutants have previously been reported to fail to generate an open a lumen in the neural rod and to have disorganised adherens junctions (Lowery et al., 2009, Lowery & Sive, 2005. We found that although nok m227 mutants generate 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 throughout development ( Figure 1H).
Nok/Mpp5a is thus critical to build apical junctional rings.

Apical rings on either side of the tissue midline develop independently and are offset from one another
To successfully undergo de novo lumen formation, 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 may first develop at the left-right interface and then these pioneer rings might split into left and right rings.
To assess this hypothesis, we carefully analysed the development of left and right rings.
Imaging sagittal and parasagittal planes we never saw an embryo containing a single plane of apical rings ( Figure 1F, G, supplementary movie 1). Instead, apical rings were always found bilaterally, and incomplete apical rings were also found bilaterally on either side of the midline, showing left and right rings are built independently. Furthermore, we observed that the developing rings on either side of the tissue midline are always offset from one another, rather than having a mirror-symmetric arrangement that might be expected if individual rings split into two ( Figure 1G).

Non-sister connections impose cell offsets across the midline
Most neural rod cells derive from progenitors that divide at the midline at neural keel stages to generate mirror-symmetrically polarised sisters connected to each other across the midline (this division is called 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 at the neural rod stage suggests a mechanism must exist that adds offset to this initial cell alignment. To confirm that sister cells become offset, we first generated mosaically labelled embryos and used live imaging to assess the development of the midline interface between sister cells. Soon after a C-division at the neural keel stage, sister cells appeared mirror-symmetric to each other across the tissue midline ( Figure 2A, 13 somites). However, through time-lapse imaging, we observed that the shape of the nascent apical endfeet dynamically changed over time, likely in response to the movements and mitoses of neighbouring (unlabelled) cells (Figure 2A and supplementary movies 2-4). The location of the sister-cell interface in relation to the rest of the cell shape therefore also changed dynamically until the neural rod stage, when the connection between the sisters was stably offset such that the nascent apical endfeet of sister cells met at their corners ( Figure 2A, 18 somites). These observations show that an initially mirror-symmetric sister cell alignment is shifted to an offset organisation following C-division mitosis. The offset conformation between C-division sister cells suggests that contralateral cell connections across the midline are likely not 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 took advantage of a BAC transgenic line expressing the cell adhesion protein Cdh2 fused to a tandem fluorescent timer (Revenu et al., 2014) and we imaged Cdh2-GFP from this line. Cdh2-GFP is expressed in all neural rod cells over the entire plasma membrane. Prior to C-divisions, cells intercalated across the rod midline and Cdh2-GFP was enriched in this interdigitation zone, showing that cells contact and adhere to several contralateral, as well as ipsilateral, cells ( Figure 2Bi). When cells entered C-division mitosis at or close to the middle of the neural keel, we found that cell processes from neighbouring cells ingressed into the cleavage furrow and distinct enrichments of Cdh2-GFP became largely localised to the cleavage furrow between sister cells during telophase (red arrows, Figure 2Bi).
This cell ingression allowed neighbouring cells to contact and presumably adhere to both the ipsilateral and contralateral daughter 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 cartooned in figure 2Biii).
By the 17 somite stage, Cdh2-GFP is enriched at the interface between left and right cells, sometimes at the apical cell vertices (Figure 2Bii). By this stage the left-right interface is now largely a straight line at the midline of the neural rod. To see whether the offset cell conformation between cells on the left and right of the neural rod persists until lumen opening, we looked at electron micrographs taken at the very start of lumen inflation. This showed a striking zig-zag conformation between cells separating across the apical midline ( Figure 2C). It therefore 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, this data suggests 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.

A single midline interface enriched in polarity and cell adhesion proteins precedes the formation of bilateral apical rings
To better understand the transition from spot-like junctional specialisations to two independent lattices of apical rings on left and right sides of the neural rod midline we analysed the changing distribution of Pard3 and the adhesion protein Cdh2 during this transition. Since there is a developmental gradient along the dorso-ventral axis of the neuroepithelium ( Figure 1B  At the same timepoint Cdh2-GFP was expressed at low level throughout the cell membrane but also slightly elevated in the interdigitation zone, mostly in discrete spots. By the 17 somite stage To analyse the distribution of cell membranes relative to the expression of Pard3-EGFP, we revealed membranes by injecting TgBAC(pard3:Pard3-EGFP) kg301 embryos with mRNA mNeptune2.5-CAAX at the 1-2 cell stage. By 24 hpf, although spinal cord cells at almost all dorsoventral levels do still meet at the left-right interface (see membrane label in Figure 3vi), Pard3-EGFP and Cdh2-GFP no longer localise to this interface; rather their expression has fully transitioned from mediolaterally arranged puncta through a single midline expression domain and into two parasagittal domains either side of the tissue midline (Figure 3Ai, iv). These two domains are characterised by periodically elevated spots of expression that correspond to points within the left and right apical rings. Much lower levels of both Pard3-EGFP and Cdh2-GFP remain in the medial zone between the bilateral apical rings (Figure 3Ai, iv).
These data suggest left and right cells initially adhere together across the midline through Cdh2rich adhesions, but this interface is then largely cleared of both Cdh2 and Pard3 as these adhesion and polarity proteins are redeployed more basally to the apicolateral border where they contribute to adhesive junctions between ipsilateral cells. This basolateral redeployment of junctional proteins is likely to be a critical step in the process of releasing cells from their contralateral adhesions that allows lumen formation.

Nok/Mpp5a is required to redeploy adhesions to the apicolateral border
Work in Drosophila epithelia has shown that Crumbs is required for the proper localisation of Baz (Drosophila Pard3) and E-cadherin at apical junctions, displacing them from the apical-most surface of single sheet epithelial cells to the apico-lateral border of the cell (Grawe et al., 1996, Harris & Peifer, 2005, Morais-de-Sa et al., 2010, Tepass, 1996. Although eliminating Crumbs in zebrafish is difficult because there are multiple Crumbs proteins, we took advantage of the nok m227 mutant that is unable to localise Crumbs proteins to the apical surface (Zou et al., 2013).
Since we earlier demonstrated that the apical spot to ring transition did not occur in nok m227 mutants ( Figure 1H), we next wanted to test the role for Nok in displacing Pard3 and Cdh2 To understand the nok m227 phenotype in relation to Cdh2 remodelling and its relation to Crumbs expression, we examined the localisation of both proteins in wild-type and nok m227 embryos. In In nok m227 mutant embryos ( Figure 3Cii) Crumbs2a expression is absent from the midline as expected. At the neural rod 18 somite stage nok 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 expression remained as a single midline domain in nok m227 mutants and failed to undergo the transition to two lines of Cdh2-GFP either side of the tissue midline, even up to stages as late at 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 adhesion and polarity proteins are then redeployed more basally to the apicolateral border where they contribute to adhesive junctions between ipsilateral cells to build coherent, bilateral sheets of neuroepithelial cells. This basolateral redeployment of junctional proteins depends on Nok/Mpp5a and is likely to be a critical step in the process of releasing cells from their contralateral adhesions to allow lumen formation. We propose this function of Nok/Mpp5a is likely to operate through its ability to localise Crumbs proteins to the left-right interface. This is consistent with the loss of Crumbs proteins at the left-right interface and the loss of basal displacement of junctional proteins away from the left-right interface in nok m227 mutants ( Figure 3C). This allows the persistent adhesion between cells on left and right sides, which is likely to cause the inhibition of lumen formation in nok m227 mutants.

Persistent adhesions may contribute to lack of epithelial maintenance in Nok and Rab11a deficient epithelia
To test the hypothesis that Nok 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 nok morphant embryos that were unable to open a lumen.
In wild type embryos, cells had successfully separated across the midline following C-division; they maintained an elongated morphology that stretched from apical to basal surface over the course of imaging and later differentiative divisions (D-divisions) occurred at the apical (ventricular) surface ( Figure 4A). However, in nok morphant tissue, cells were not separated across the midline, often did not fully extend to the basal surface and they appeared clumped together, often in rosette-like structures. D-divisions in nok morphants occurred within these cell clumps, often at a distance from the midline of the tissue, in contrast to wild-type cells ( Figure   4B and quantified in 4C). The cell behaviour of ectopic divisions in nok morphants is accompanied by a lateral dispersal of Pard3 expression from the midline ( Figure 4D). Thus, as well as being unable to undergo the puncta-to-apical ring transition ( Figure 1H), Nok deficient cells show persistent adhesions and apical midline organisation becomes progressively fragmented over time.
Since we hypothesise that Nok functions through its ability to localise Crumbs to the nascent apical surface, we next analysed cell behaviours following another manipulation that depletes Crumbs from the apical surface. For this we expressed a dominant-negative (DN) form of Rab11a in rhombomeres 3 and 5. We previously showed that this manipulation reduces apical Crumbs 2a localisation and prevents lumen opening in these segments (Buckley et al., 2013).
Here, we show that apical Crumbs 1 is also downregulated in DNRab11a rhombomeres ( Figure   4E), confirming that this manipulation is sufficient to downregulate apical trafficking of multiple Crumbs paralogs. We found that cells in the DNRab11a rhombomeres did not separate following D-division and formed cell clumps and rosette-like structures ( Figure 4F). An example of rosette formation around a central focus of Pard3 is depicted in Figure 4G. Similar to nok morphants ( Figure 4D), at the tissue level, a lateral dispersal of Pard3-EGFP from the midline was seen, progressively fragmenting the midline organisation over time ( Figure 4H). This is in Our data suggests that persistent adhesions between cells in Nok deficient and DNRab11a embryos both inhibits lumen formation and promotes the formation of local clumps of cells. We earlier showed that adhesions between cells in Nok deficient and DNRab11a embryos are not reorganised into a lattice of ring-like adhesions between ipsilateral junctional belts ( Figure 1H), 1 0 whose role is to generate and maintain a continuous sheet of cells. We found that many cells undergo ectopic divisions in this environment, presumably further contributing to ectopic cell clump formation and loss of epithelial structure ( Figure 4B and C). However, despite persistent adhesions between DNRab11a cells they could integrate into a neuroepithelium if they had access to an apical lumen surface initiated by wild-type cells ( Figure 5). Therefore, together, our data suggests that a combination of a lack of cell separation following division and a lack of apical ring formation results in the progressive fragmentation of midline organisation over successive rounds of division in Nok deficient and DNRab11a embryos.
Nok is necessary for apical ring formation even in the absence of cell division.
A recent paper's analysis of cell divisions and lumen formation in nok m520 mutants suggested that Nok/Crumbs function in lumen formation may be specifically related to resolving connections between sister cells following the C-division (Guo et al., 2018). In the course of analysing junction formation in nok m227 mutants, we uncovered 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 & Sive, 2005, we found the loss of lumen formation was not fully penetrant and consistently found that small lumens formed at the level of the midbrain hindbrain boundary ( Figure 6A). Less frequently small lumens also formed in the dorsal aspects of the hindbrain at the level of the otic vesicles ( Figure 6B). We found that Crumbs 2a protein was present at the apical surface of these lumens ( Figure 6A), consistent with the role of Crumbs in the successful formation of an apical surface and opening of a lumen.
Since the closed-lumen phenotype is not fully penetrant in nok mutants, we suggest that using the rescue of the open lumen phenotype is problematic in nok. To better analyse whether Nok has specific functions in relation to the C-division we therefore analysed apical ring formation in the anterior spinal cord. This region is distant from the midbrain hindbrain boundary and the level of otic vesicles and provides a more reliable and precise assay of Nok function. We have shown that Nok and Rab11a are required for cells to separate across the midline following the C-division. We have previously demonstrated that left-right separation was not rescued by blocking division in DNRab11a embryos, suggesting that the role of Rab11a is not exclusively aphidicolin treated embryos ( Figure 7F and G) and the fewer, enlarged apical rings in the aphidicolin treated siblings ( Figure 7B and H). These results demonstrate that Nok function is not specifically related to remodelling adhesions between C-division sisters and show that Nok 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.

Discussion
Our work uncovers three aspects of de novo apical surface generation within a complex organ . We show this is not the case as Nok-dependent apical protein remodelling is required for apical surface generation both with and without cell divisions.

The roles of Nok and Rab11a
During epithelial development of the Drosophila ectoderm, Crumbs is required for the proper localisation of Baz (Drosophila Pard3) and E-cadherin at apical junctions, displacing them from the apical-most surface of single sheet epithelial cells to the apico-lateral border of the cell (Grawe et al., 1996, Harris & Peifer, 2005, McGill et al., 2009, Morais-de-Sa et al., 2010, Pilot et al., 2006, Tepass, 1996 and thus building the apicolateral junctional belts that tie the epithelial cells together (St Johnston & Ahringer, 2010, Wei et al., 2005.
Here, we describe a similar displacement of Pard3 and repositioning of the main zebrafish neuroepithelial cadherin, Cdh2, to the apicolateral cell borders in the zebrafish neural rod. In this case the displacement occurs at the midline plane within a solid rod of cells rather than at a free surface, but despite this fundamental difference in tissue architecture, a rather similar mechanism appears to operate to build the apicolateral junctional belts. We show that the 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 wildtype 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.

Coordinating left-right release with epithelialisation
The zebrafish neural rod is initially a solid primordium, so the displacement of Pard3 and Cdh2 is not from a free apical surface but instead occurs within the densely packed cell-cell interfaces  Figure 4D and H). We propose that this disintegration is likely due to a combination of factors; during cell division, interkinetic nuclear migration and mitotic rounding pulls cells around within the tissue. In a normal epithelium, despite these forces, 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 what should be the apical plane and cells are unable to fully re-extend to the basal side of the neural rod following division.
Additionally, a lack of cell separation following 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 nok and

The role of the C-division
A recent paper also concluded that Nok is necessary to remodel junctional proteins during epithelialisation in the neural rod (Guo et al., 2018). They however propose that Nok's role in apical remodelling is only necessary to rescue the tissue disruption caused by midline Cdivisions, a conclusion based on their reported rescue of lumen formation and apical junctional belts when cell division is blocked in the nok m520 mutant. However, we show that blocking cell division does not rescue junctional belt formation in nok m227 mutants (Figure 7), and thus has a more fundamental role in apical junction remodelling than resolving the potential disruption to tissue organisation caused by C-divisions. We suggest the results of Guo et al (2018)  and we suggest this is one mechanism that leads to staggered connections across the midline.
We suggest that this staggered alignment of endfeet may generate a morphogenetic advantage since it allows the attachment to a larger number of cells than a one-to-one 'mirror adhesion' as

Embryo Care
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.

Generation of Pard3-EGFP transgenic line
To generate the transgenic zebrafish line expressing Pard3-EGFP with endogenous spatial and temporal expression we used bacterial artificial chromosome (BAC) recombineering. We

RNA preparation and injection
Fusion constructs containing cDNA in the pCS2+ vector were linearized 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 40-100pg per embryo: Histone 2A

Chemical inhibition of cell division
Cell division was blocked by incubating dechorionated embryos in 300 µM aphidicolin (Sigma) in 4% DMSO from bud stage (10 hpf) to 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.

Immunohistochemistry
Embryos were fixed with 4% paraformaldehyde for 2 hours at room temperature prior to processing for immunohistochemistry. The following primary antibodies were used: mouse anti- Rabbit anti-Pard3 (rabbit polyclonal, Millipore 07-330, 1:100) was used following fixation in Dent's for 3 hours at room temperature and rehydration from methanol. AlexFluor secondary antibodies (Life Technologies) were used at 1:500 and Hoeshct (Life Technologies) nuclei stain was used at a final concentration of 1:10,000.

Electron microscopy
Embryos were fixed in 2% paraformaldehyde and 1.5% glutaraldehyde in 0.1 M sodium cacodylate buffer over-night 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 on a transmission electron microscope.

Confocal imaging and processing
Embryos were mounted in low melting point agarose and imaged either using a Zeiss LSM 880 Fast Airyscan microscope and water dipping x20/1.0 N.A. objective, a Leica SP5 confocal microscope and water dipping x25/0.95 N.A objective or a PerkinElmer Ultraview spinning disk microscope. Imaging at the centre of a developing live tissue mass required the improved resolution in x, y and z 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 further 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 was collected from the hindbrain and anterior spinal cord regions. Some adjacent cells have been removed from images of single cells to increase clarity of the cells of interest. Images were processed using Imaris, Volocity and Fiji/ImageJ.

Data analysis
The custom R code used to generate plots for figures 1 and 3 is available at https://github.com/andyivanhoe/pard3-analysis.
To obtain apical ring density counts across the developing neuroepithelium ( Figure 1E), Pard3-EGFP embryos were live-imaged en face between 15ss and 21ss using the Zeiss Airyscan. The dorso-ventral axis of the resulting z-stacks were divided into four quadrants (dorsal, mid-dorsal, mid-ventral, 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 6 live TgBAC(pard3:Pard3-EGFP) kg301 embryos over a period of 3 hours 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 6 embryos at 1 hour intervals (Figure 3Aii 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 6 embryos where the two lines of expression were approximately 4 µm apart (Figure 3Aiv). This analysis was done at the last timepoint of the 3-hour timeline, but at a more ventral (and hence more morphogenetically advanced, see Figure 1 and associated text) position. In this case we measured intensity along a 23 µm wide mediolateral line. Because the midline expression is not perfectly straight and the peaks of expression only 4 µm apart, for this analysis we choose a single z-level and a thinner mediolateral line than for the timeline measurements to avoid averaging out the two peaks.
Plots in figure 4C, 4J, 7G and 7H were generated using Graphpad Prism. Specific statistical analyses are described in the figure legends.  Apical ring formation illustrated in F & G was analysed from over 10 embryos. For each embryo, 10-25 apical rings were located on one side of the neuroepithelium in en face orientation and then a z-stack was taken through the middle of the neural rod until the rings on the opposite side were visible. We never saw an embryo containing a single plane of apical rings. There were always bilateral rings and these were always offset from each other across the midline.
H) Images taken from confocal time-lapse movies of wild-type and nok morphant Pard3-EGFP embryos in sagittal orientation from the 16 somite stage to 24 hpf. Comparable images were seen from 3 embryos from each genotype and demonstrate that nok morphants never progress from punctate Pard3-EGFP to apical ring formation. 3 control morphants were also assessed in the Cdh2-tFT transgenic line and all had comparable apical rings to wild types. becomes attached to two contralateral cells -one is its sister cell from the C-division and the other is one of its sister's neighbouring cells (the pink cell in this example).
ii. 5 µm projection of 17 somite stage neural rod from a Cdh2-tFT transgenic embryo. Cells are well aligned along a centrally located midline. Cdh2-GFP is upregulated along the apical midline, particularly at cell corners (red arrowheads in magnified region).
iii. Model depicting the co-ingression of neighbouring cells into the cleavage furrow during Cdivision (yellow cells) and the subsequent offsetting of sister cells from each other, which precedes apical ring formation. This is based on the images similar to that shown in figure Bi.
Pink lines and dots represent high levels of Cdh2 associated with the dividing cell. The ingression of either ipsilateral (e.g. orange) or contralateral (e.g. blue) neighbours into the cleavage furrow promotes the formation of multiple contralateral connections across the midline.
For example, the left-hand yellow sister cell becomes attached to both its right-hand yellow sister cell and the ingressing blue cell. The right-hand yellow sister cell becomes attached to both its left-hand yellow sister cell and the ingressing orange cell. This is one way in which multiple ipsilateral and contralateral adhesions might be promoted during C-division mitosis C) i. Transmission electron micrographs of a 20 hpf embryo hindbrain in transverse orientation.
The lumen has just started to open from the midline. The interface between contralateral cells has a striking 'zig-zag' pattern (3/3 19-20 hpf embryos). ii. Inset magnified region from i. Crb2a were initially colocalised at the tissue midline at the 18 somite stage (4/4 embryos) but Cdh2-GFP protein was later displaced basolaterally to form two independent stripes of expression either side of the midline by the 26 somite stage (8/8 embryos). In nok m227 mutants, Crb2a was not present at the midline at the 18 somite stage (4/4 embryos) and Cdh2-GFP remained in a single expression domain at the tissue midline even as late as 28 hpf (5/5 embryos). rhombomeres 3 and 5), 9/10 embryos from 2 of our later experiments did not have a fully penetrant DNRab11a phenotype. In these embryos some lumen opening was present, demonstrating that cells were able to separate from each other across the midline. In this case no or very little ectopic apical protein was seen. We think that this lower DNRab11a penetration was due to methylation of the Tg(UAS:mCherry-Rab11a S25N) mw35 line since, when we bred previous generations of the line, we saw fully penetrant phenotypes again. This correlation between extent of DNRab11a penetrance, lumen opening and ectopic apical proteins was also seen in nok morphant embryos (supplementary figure 1).
I) 4 µm z-projection in dorsal orientation 23 µm below the rhombomere surface of a 30 hpf DNRab11a x Krox20 embryo hindbrain at rhombomere 3, stained via IHC for Pard3 and gamma-tubulin. Pard3 is localised in a round clump, surrounded on all sides by centrosomes.
ii. aPKC is present around the whole apical surface of both wild-type and DNRab11a rhombomeres (

E)
Overall body shape of wildtype siblings and nok m227 mutants with and without aphidicolin treatment.
F) Horizontal section through neural tube showing nuclear staining in wildtype embryos treated with DMSO or aphidicolin to block C-divisions. Aphidicolin-treated embryo have larger nuclei, demonstrating that S-phase division-block was successful (quantification in G).

G)
Quantification of nucleus area of wildtype siblings and nok m227 mutants with and without aphidicolin treatment. Data from 100-150 nuclei was pooled from 2-3 embryos in each group and analysed by 2-way ANOVA. Areas of Aphidicolin treated nuclei were, on average, 30 µm 2 bigger than DMSO treated embryos (P<0.0001). There was no significant difference in nuclei area between wild type and nok m227 mutant embryos (P=0.8561). Error bars denote standard deviations.

H)
Quantification of apical ring number in wildtype siblings and nok m227 mutants with and without aphidicolin treatment. Numbers of apical rings per 1000 µm 2 were calculated from 3-4 embryos per group. No apical rings were seen in any of the nok m227 mutants. There were, on average, 40 fewer apical rings per 1000 µm 2 in aphidicolin treated wild type embryos than in DMSO treated wild type embryos (unpaired t-test, P=0.0222). Error bars denote standard deviations.