E‐cadherin mediates apical membrane initiation site localisation during de novo polarisation of epithelial cavities

Abstract Individual cells within de novo polarising tubes and cavities must integrate their forming apical domains into a centralised apical membrane initiation site (AMIS). This is necessary to enable organised lumen formation within multi‐cellular tissue. Despite the well‐documented importance of cell division in localising the AMIS, we have found a division‐independent mechanism of AMIS localisation that relies instead on Cadherin‐mediated cell–cell adhesion. Our study of de novo polarising mouse embryonic stem cells (mESCs) cultured in 3D suggests that cell–cell adhesion localises apical proteins such as PAR‐6 to a centralised AMIS. Unexpectedly, we also found that mESC clusters lacking functional E‐cadherin still formed a lumen‐like cavity in the absence of AMIS localisation but did so at a later stage of development via a “closure” mechanism, instead of via hollowing. This work suggests that there are two, interrelated mechanisms of apical polarity localisation: cell adhesion and cell division. Alignment of these mechanisms in space allows for redundancy in the system and ensures the development of a coherent epithelial structure within a growing organ.


Introduction
Most organs in the body arise from tubes or cavities made from polarised epithelial cells. These cells have a strict apico-basal orientation; they align their apical ends along a centrally located lumen. Some tubes, such as the anterior neural tube in amniotes, arise via folding and closure of an already polarised epithelial tissue, through mechanisms such as actomyosin-mediated apical constriction (Nikolopoulou et al, 2017). However, many tubes and cavities, such as the posterior neural tube, mammary acini, kidney tubules and mammalian epiblast, arise via apical-basal polarisation within the centre of an initially solid tissue. The mechanisms by which such "de novo" polarisation is coordinated within dynamically growing tissue have been the focus of a significant body of research from several different models and have relevance both for understanding polarity-associated diseases and for directing organ bioengineering approaches.
Although the exact mechanisms are still under debate and may differ in different epithelia (Buckley & Johnston, 2022), Laminin, Integrin β1 and RAC1 signalling from the extracellular matrix (ECM) is now well established to be necessary for directing the overall apico-basal axis of polarisation of internally polarising tubes (Yu et al, 2004;Akhtar & Streuli, 2013;Buckley et al, 2013;Bedzhov & Zernicka-Goetz, 2014;Bryant et al, 2014;Mol e et al, 2021). What is less clear is how the precise localisation of the apical membrane initiation site (AMIS) is directed at the single-cell level and how this is coordinated between neighbouring cells. The AMIS is a transient structure, marked by the scaffolding protein Partitioning-defective-3 (PAR-3) and tight junctional components such as Zonula occludens-1 (ZO-1), that defines where apically targeted proteins will fuse with the membrane, therefore determining where the lumen will arise (Bryant et al, 2010;Blasky et al, 2015). It is important that the subcellular localisation of the AMIS is coordinated between cells during morphogenesis to enable organised lumen formation.
The current literature suggests that cell division plays an important role in AMIS localisation. In particular, the post-mitotic midbody has been shown to anchor apically directed proteins (Schl€ uter et al, 2009;Li et al, 2014;Wang et al, 2014;Luj an et al, 2016;Rathbun et al, 2020). However, studies within the zebrafish neural rod showed that, whilst misorientation of cell division results in disruption of the apical plane at a tissue level, these phenotypes can be rescued by inhibiting cell division (Ciruna et al, 2006;Tawk et al, 2007;Quesada-Hernandez et al, 2010;Žigman et al, 2011).
We also previously demonstrated that individual zebrafish neuroepithelial cells were able to recognise the future midline of the neural primordium and organise their intracellular structure around this location in advance and independently of cell division. This resulted in the initiation of an apical surface at whichever point the cells intersect the middle of the developing tissue, even if this is part way along a cell length (Buckley et al, 2013). This suggests that, whilst cell division is undoubtably a dominant mechanism, there must be another overlying mechanism driving AMIS localisation during de novo polarisation. The earliest indication of midline positioning in the zebrafish neural rod is the central accumulation of the junctional scaffolding protein Pard3 (PAR-3) and the adhesion protein N-cadherin (Buckley et al, 2013;Symonds & Buckley et al, 2020). This led us to hypothesise that cell-cell adhesions could direct the site for AMIS localisation during de novo polarisation. In line with this hypothesis, β-catenin-mediated maturation of N-cadherin was found to be necessary for the recruitment of the PAR apical complex protein atypical protein kinase C (aPKC) in the chick neural tube (Herrera et al, 2021). Opposing localisations of ECM and Cadherin proteins were also found to be sufficient to specify the apical-basal axis of hepatocytes in culture (Zhang et al, 2020).
To test the role of cell-cell adhesions in AMIS localisation, we turned to mouse embryo stem cell (mESC) culture in Matrigel, which has been used as an in vitro model for the de novo polarisation of the mouse epiblast (Bedzhov & Zernicka-Goetz, 2014;Shahbazi et al, 2017;Kim et al, 2021;Mol e et al, 2021). This allowed us to study the initiation of apico-basal polarity of embryonic cells alongside the first cell-cell contacts between isolated cells and small cell clusters. It also allowed us to determine within a mammalian model whether division-independent polarisation is a conserved feature of de novo polarising structures. Unlike vertebrate epithelial cell culture models such as Madin-Darby canine kidney (MDCK) cells, which initiate lumenogenesis as early as the 2-cell stage when cultured in Matrigel (Bryant et al, 2010;Blasky et al, 2015), mESC cells in Matrigel only form lumens at the multi-cellular stage after 48-72 h in culture, coinciding with an exit in pluripotency (Bedzhov & Zernicka-Goetz, 2014;Shahbazi et al, 2017). This results in a relatively clear separation of the stages of de novo polarisation ( Fig 1A). Previous literature suggests that the AMIS is formed at the 2-cell stage, around 24-36 h after culture in Matrigel, as denoted by membrane-localised PAR-3 and ZO-1 and sub-apical localisation of apical proteins such as Podocalyxin (PODYXL) (Shahbazi et al, 2017). The pre-apical patch (PAP) stage is formed after 36-48 h in culture, as denoted by the fusion of apical proteins such as PODXYL, PAR-6 and aPKC to the apical membrane and the displacement of junctional proteins PAR-3, ZO-1 and E-cadherin to the apico-lateral junctions (Shahbazi et al, 2017;Kim et al, 2021), following which lumenogenesis is initiated after 48-72 h in culture.
To determine the role of cell division and of cell adhesion in mESC AMIS localisation, we analysed mESC cells cultured in Matrigel at the AMIS stage with and without cell division in wildtype and E-cadherin knockout cell lines. We then further analysed polarisation and lumenogenesis in the absence of E-cadherin. Our results suggest that there is a division-independent mechanism of AMIS localisation that relies instead on E-cadherin-mediated cellcell adhesions.

Cell division is dispensable for AMIS localisation
First, we tested whether cell division was necessary for AMIS localisation in mESC rosettes. We cultured naïve, unpolarised mESCs (ES-E14 cells) in 2D on gelatin with 2i/LIF and then treated them with mitomycin C to block cell division. We then isolated single cells and seeded them into Matrigel without 2i/LIF, in N2B27 differentiation medium ( Fig 1B). Cell divisions were efficiently blocked during the first 24-h post-seeding, during which time individual cells contacted each other and formed cell clusters in the absence of cell division (Movie EV1; Fig EV1A and B).
To assess AMIS localisation, we carried out immunofluorescence (IF) for PAR-3 and ZO-1 at 24-h post-seeding. As previously published (Shahbazi et al, 2017), in addition to several puncta at the cell peripheries, both PAR-3 and ZO-1 localised to the membrane at the centre of control cell-cell contacts, marking the AMIS in the majority of cell clusters (Fig 1C(i) and H(i)). Interestingly, divisionblocked cells also localised PAR-3 and ZO-1 to the central membrane (Fig 1C(ii) and H(ii), quantified in Fig 1D-G and I). In both control and division-blocked cell clusters, there was a small proportion that had not yet fully localised the AMIS at the 24-h stage (Fig 1  D and I), where PAR-3 was either not localised (Appendix Fig S1A) or was only weakly present at cell-cell interfaces (Appendix Fig S1B). E-cadherin was upregulated along the whole length of the cell-cell interfaces in both dividing and non-dividing cell clusters, with a higher level of E-cadherin at the cell-cell interface relative to the cell-matrix interface (Figs 1C and EV1C). To quantify the subcellular localisation of PAR-3 in each cell cluster, we carried out intensity profiles across the cell-cell interface of 2-cell doublets (Fig 1E(a) and F) and calculated the ratio between centralised and surrounding non-centralised PAR-3 in multi-cellular clusters (Fig 1  E(b) and G). This confirmed that PAR-3 localised to a small central area at the cell-cell interface in both control and division-blocked 2cell doublets and multi-cellular clusters. Golgi apparatus and centrosomes were also localised to the centre of cell-cell contacts both in dividing and non-dividing conditions Appendix Fig S1C), confirming that mESCs were polarised in the absence of cell division. mESC clusters also centrally localised PAR-3, ZO-1 and polarised the Golgi apparatus when cell division was blocked using an alternative compound, aphidicolin (Fig EV1D-H).
Together, these results demonstrate that cell division is dispensable for de novo AMIS localisation in polarising mESCs.

Cell-cell contact directs PAR-6 localisation
To understand the dynamics of apical protein polarisation in the absence of cell division, we generated a mESC stable cell line expressing mCherry-PAR-6B and imaged cells live. In line with the previous characterisation of PAR-6 by IF (Shahbazi et al, 2017;Kim et al, 2021) in control dividing cells, mCherry-PAR-6B localised to the apical membrane by the PAP stage at 48 h and to the luminal apical membrane from 72 h (Fig 2A; Appendix Fig S2A and B). In addition, the transgene allowed us to better visualise PAR-6B puncta at earlier 24-h AMIS stages of development. At this stage, mCherry-PAR-6B was localised sub-apically, polarised towards the central region of cell-cell contact (Fig 2A; Appendix Fig S2B). Some Timeline of experiment setups to assess AMIS formation. C Immunofluorescence of PAR-3 and E-cadherin. PAR-3 localisation was concentrated at the centre of 2-cell mESC doublets and 3or 4-cell mESC clusters, whilst Ecadherin was localised along the whole length of cell-cell interfaces in both control and mitomycin C-treated conditions. D Quantification of the frequency of cell clusters with a strong polarised PAR-3 centre. See the representative mESC clusters without the strong polarised PAR-3 in Appendix Fig S1A and B. E Illustrations of PAR-3 analysis in 2-cell (a) and >2-cell (b) clusters. The line-scan analysis was performed from the yellow arrow to the orange arrow at the cell-cell interface between two cells in (a). The average pixel intensity was analysed at central (red region, inner dotted line) and surrounding non-central regions (orange region, between the inner and outer dotted lines) in (b). F Line-scan profiles of PAR-3 at the cell-cell interface of 2-cell mESC doublets. Line-scans were sectioned and fitted to each 5% along the cell-cell interface length. Source data are available online for this figure.
Ó 2022 The Authors The EMBO Journal 41: e111021 | 2022 mCherry-PAR-6B puncta appeared to be associated with the Golgi network. However, a large proportion of mCherry-PAR-6B was in the cytoplasm sub-apically (Appendix Fig S2C), suggesting that PAR-6B puncta were in the process of being delivered to the apical membrane at 24 h. A similar polarised distribution of PAR-6B was observed in both control and division-blocked cells (Fig 2B and C; Appendix Fig S2C), demonstrating that cell division is dispensable for apical protein polarisation.
We next assessed the dynamics of PAR-6B polarisation. In both control and division-blocked cells, non-cortical mCherry-PAR-6B puncta were visible at the single-cell stage. In control cells, these puncta relocalised to the abscission plane, following cell division  These results suggest that cell-cell contact directs PAR-6B localisation at the central AMIS, independent of cell division.

E-cadherin adhesions are necessary for AMIS localisation
The above results suggest that there is a division-independent mechanism of AMIS localisation that relies instead on cell-cell adhesions. Since E-cadherin is the predominant adhesion molecule in nonneural epithelia, we hypothesised that it might be important in AMIS localisation. To achieve a full removal of E-cadherin, we employed an E-cadherin knockout (Cdh1 KO) mESC line (Appendix Fig S3A; Larue et al, 1996).
To assess AMIS localisation, we again carried out IF for PAR-3 and ZO-1 at 24-h post-seeding. As seen earlier (Fig 1), both PAR-3 and ZO-1 localised to the central region of cell-cell contact within wild-type (W4 cells) doublets/clusters with and without division. However, PAR-3 and ZO-1 localisation was strongly inhibited in the absence of E-cadherin ( To investigate AMIS localisation at a single-cell level, we cocultured division-blocked wild-type (ES-E14) and Cdh1 KO cells and analysed chimeric mESC doublets, comprising one wild-type and one Cdh1 KO cell. Whilst homogenous control doublets localised PAR-3 to the central region of the cell-cell interface, heterogeneous chimeric doublets did not localise PAR-3 centrally (Fig 3E and F). The same result was seen in E-cadherin RNAi chimeric doublets (Appendix Fig S3B(b)). Golgi and centrosome localisation towards the cell-cell interface suggested that the overall axis of polarity was maintained, even in the absence of both cell division and E-cadherin (Fig 3G-I; Appendix Fig S3D). These results demonstrated that Ecadherin is necessary for AMIS localisation.
Since E-cadherin is localised along the whole cell-cell interface but PAR-3 and ZO-1 localise at central cell-cell interfaces, we next used fluorescence recovery after photobleaching (FRAP) to compare the stability of E-cadherin protein at central and side regions in 2-cell mESC clusters (Fig EV2A and B, regions illustrated in Fig  EV2C). E-cadherin-eGFP ( Fig EV2B) and E-cadherin immunofluorescence ( Fig EV2C and D) levels were the same at these two regions. However, FRAP of E-cadherin-eGFP showed that the mobile fraction of E-cadherin-eGFP was lower in the central region than the side regions ( Fig 3J and K). Therefore, E-cadherin junctions are more stable at the centre-most region of the cell-cell interface, which may provide at least a partial explanation for why AMIS localisation occurs precisely at this region.
It has previously been demonstrated that a reduction in Ecadherin can slow pluripotency exit (Soncin et al, 2009). However, pluripotency exit was previously shown not to alter AMIS formation (Shahbazi et al, 2017). In support of these results, we also found that cells maintained in the pluripotent state when cultured in Feeder Cell medium provided with 2i/LIF still localised the AMIS, with and without cell division (Fig EV3A-C). However, in line with our results showing lack of AMIS localisation in Cdh1 KO cells cultured in the absence of 2i/LIF (Fig 3), cells cultured in the presence of 2i/LIF also could not localise an AMIS in the absence of Ecadherin ( Fig EV3A-C). Despite this result, we wanted to check whether the stage of pluripotency exit differed between WT and Cdh1 KO cells in our experiments since this might indicate a different speed of maturation. We, therefore, carried out IF for Orthodenticle Homeobox 2 (OTX2) protein, which is necessary for pluripotency exit, and the pluripotency marker protein Nanog. Although, as expected, the overall level of nuclear OTX2 increased and Nanog decreased over the 24-h course of development, we found no significant difference in post-mitotic levels of OTX2 and Nanog between WT and Cdh1 KO cells ( Fig EV3D). This result suggests that there was no difference in the stage of pluripotency exit in the cell clusters that we analysed during this study, and this is therefore unlikely to play a role in the lack of AMIS localisation seen in Cdh1 KO cells.
These results demonstrate that E-cadherin adhesions between cells are necessary for AMIS localisation but not for the overall axis of polarity. They also demonstrate that ECM in the absence of Ecadherin is insufficient for AMIS localisation. Adhesion molecules P-cadherin, JAM-A and Nectin-2 are not necessary for AMIS localisation E-cadherin is not the only form of adhesion molecule that is expressed at cell-cell contacts. A complex network of interactions between the PAR-complex, adhesion molecules, MAGUK scaffolding proteins and the actin cytoskeleton is responsible for building cellcell junctions (Buckley & Johnston, 2022). Of relevance to this study, PAR-3 has been found to directly bind to transmembrane Junctional Adhesion Molecules (JAMs) and Nectin proteins in mammals (Ebnet et al, 2001;Itoh et al, 2001;Takekuni et al, 2003). In the mammalian embryo, JAM-A and Nectin-2 adhesion molecules are expressed between inner cell mass cells in the mouse blastocyst (Thomas et al, 2004). We found that JAM-A and Nectin-2, as well as P-cadherin, were expressed at cell-cell contacts in 2D cultured mESCs ( Fig EV4A). We, therefore, carried out IF of 2-cell mESC clusters cultured in Matrigel for 24 h to determine the localisation of P-cadherin, JAM-A and Nectin-2 at the AMIS stage. Whilst the majority of 2-cell and 4cell clusters formed a polarised PAR-3 centre, P-cadherin was uniformly expressed along cell-cell interfaces (Figs 4A and EV4B). JAM-A was uniformly expressed along cell-cell interfaces at the 2cell stage ( Fig 4D) and concentrated towards the centre of 4-cell mESC clusters (Fig EV4C). At the 2-cell stage, Nectin-2 was expressed along the whole cell-cell interface with a sight concentration towards the central regions where PAR-3 was localised ( Fig  4G). At the 4-cell stage, Nectin-2 was concentrated towards the centre of the clusters (Fig EV4D). These results suggest that PAR-3 localises at the AMIS before JAM-A or Nectin-2.
Next, we used siRNA KD of protein function in dividing and division-blocked cells to test whether P-cadherin, JAM-A or Nectin-2 proteins were necessary for AMIS localisation. However, following siRNA for each of these proteins, PAR-3 was still polarised to the centre of cell-cell contacts (Figs 4A-I and EV4B-D). Compared with the loss of PAR-3 polarisation upon E-cadherin KO or KD (Fig 3, Appendix Fig S3), the results demonstrate that P-cadherin, JAM-A and Nectin-2 are not necessary for AMIS localisation. Indeed, when the centralised PAR-3 localisation was lost in the E-cadherin KO mESC 2-cell clusters (Fig 3A-C), P-cadherin, JAM-A and Nectin-2 were still expressed at the cell-cell interface between the mESCs (Fig EV4E-G). This suggests that E-cadherinbased adhesions might be specifically responsible for mediating AMIS localisation.

E-cadherin adhesions are sufficient to initiate AMIS localisation, independent of ECM signalling and cell division
As discussed, ECM-mediated signalling plays an important role in orienting the axis of polarisation within de novo polarising systems. Recently, the apico-basal axis of cultured mature hepatocytes was established by a combination of ECM signalling and immobilised Ecadherin (Zhang et al, 2020). However, PAR-3 has also recently been shown to polarise in mESCs lacking functional Integrin-β1 or cultured in agarose in the absence of ECM proteins (Mol e et al, 2021). Our current study shows that the AMIS can localise in the absence of cell division but not in the absence of E-cadherin. We, therefore, wanted to explore the relative roles of ECM, cell division and E-cadherin in AMIS localisation.
We first eliminated the influence of ECM by culturing divisionblocked mESCs (ES-E14) in 0.5% agarose and carried out IF for PAR-3 after 30 h in culture. These cells were still able to polarise PAR-3, even in the absence of both cell division and ECM proteins (Fig 5A and B). However, in line with our earlier results (Fig 3), PAR-3 localisation was strongly inhibited in Cdh1 KO cells (Fig 5A  and B). These results suggest that AMIS localisation occurs independently of both ECM signalling and of cell division, relying instead on E-cadherin.
To further test the sufficiency of E-cadherin adhesions in initiating AMIS localisation, we cultured individual division-blocked mESCs (ES-E14) onto either E-cadherin recombinant protein or Fibronectin precoated glass, then topped the cells with N2B27 medium, with or without 20% Matrigel and carried out IF for PAR-3 after 24 h in culture. Like results from hepatocytes (Zhang et al, 2020), cells plated on E-cadherin and topped with Matrigel localised PAR-3 to the centre of the cell-cadherin interface (Fig 5C-F). However, this central PAR-3 localisation was significantly reduced when cells were plated on fibronectin (Fig 5C-F). Interestingly, cells cultured on E-cadherin but in the absence of Matrigel still localised PAR-3 to the centre of the cell-cadherin interface (Fig 5C-F).
These results demonstrate that E-cadherin adhesions are both necessary and sufficient for initiating AMIS localisation, whilst ECM is not necessary or sufficient for AMIS localisation.

E-cadherin is necessary for hollowing lumenogenesis
We next wanted to test the importance of E-cadherin-mediated AMIS localisation in lumenogenesis. We, therefore, cultured WT (ES-E14) and Cdh1 KO mESCs and fixed them at the AMIS 24-h stage, PAP 48-h stage and lumen 72 and 96-h stage. We then carried out IF for PAR-3 and ZO-1 to label AMIS/apical-lateral junctions and PODXYL to label apical proteins. Whilst most WT cell clusters had a centralised apical domain or small lumen after 48 h in culture, very few Cdh1 KO cell clusters had made a centralised apical domain by the 48-h PAP stage (Fig 6A-D). In line with our earlier findings at the 24-h AMIS stage (Fig 3), this provides further evidence that Ecadherin is necessary for centralised AMIS localisation. However, we noticed that a small percentage of Cdh1 KO cell clusters at 48 h had formed an open "cup-shape," with apically localised PODXYL ( Fig 6B) and apico-laterally localised junctional PAR-3 (e.g. Fig 6  A(iii)) and ZO-1 (Fig EV5A). We termed these "open cavities" (Fig 6  C). Surprisingly, by the 72-h lumen stage, approximately 75% of Cdh1 KO cell clusters had formed polarised cavities, approximately 50% of which were open cavities and 50% were closed (Fig 6D and  E). Over the course of 48-96 h in culture, the overall percentage of polarised cavities increased (Fig 6D) as did the proportion of these structures that were "closed" (Fig 6E). This suggested that these cavities might form via gradual "closure" of the tissue, rather than via hollowing. Both "open" and "closed" cavities were surrounded by polarised Golgi apparatus, demonstrating that the overall apicobasal axis of cells was intact. (Fig EV5B).
To further assess the morphogenetic mechanism by which cavities form in Cdh1 KO cells, we generated WT (ES-E14) and Cdh1 KO mESC lines labelled with LifeAct-mRuby (Appendix Fig S4). We visualised the process of lumenogenesis within mESCs cultured in Matrigel via live imaging (Fig 6F; Movies EV3 and EV4). This 8 of 17 The EMBO Journal 41: e111021 | 2022 Ó 2022 The Authors confirmed that, whilst the WT cell clusters made a central lumen (8/8 movies on day 2) and then expanded this already central lumen (8/8 movies on day 3), cdh1 KO cell clusters first generated an open cup-shape cavity (3/3 movies on day 2), which then gradually closed, eventually generating a centralised lumen-like structure without hollowing at a later stage of development (3/3 movies on day 3). These results demonstrate that, in the absence of E-cadherinmediated AMIS localisation, cell clusters do not hollow but instead generate lumen-like cavities via a closure mechanism. Our results also demonstrate that E-cadherin and centralised AMIS localisation are not required for apical membrane formation. In the absence of E-cadherin, an apical surface is still formed but this occurs later in development so appears less efficient.

Epithelial cells can polarise de novo in the absence of cell division
Both AMIS-associated proteins PAR-3/ZO-1 and apical polarity protein PAR-6B localised similarly in WT and division-blocked Ó 2022 The Authors The EMBO Journal 41: e111021 | 2022 mESCs (Figs 1, 2, and 7A). This finding supports our previously published zebrafish neuroepithelial cell in vivo analysis, which demonstrated the division-independent localisation of Pard3 (PAR-3) and ZO-1 at the neural rod primordial midline (Buckley et al, 2013). Together, this demonstrates that although division is an important contributor to AMIS formation, a division-independent mechanism of de novo polarisation and AMIS localisation can occur in both in vivo and in vitro conditions. Whilst disorganised lumen formation can also occur in the absence of division in the zebrafish neural rod (Buckley et al, 2013

E-cadherin-based cell-cell contacts are necessary and sufficient to initiate AMIS localisation
AMIS localisation in Cdh1 KO cell clusters is strongly inhibited (Fig 3), and individual mESC cells can localise their AMIS to the central region of the cell-cell interface independently from ECM signalling ( Fig 5). Together, this demonstrates that the formation of Ecadherin-based cell-cell contacts is both necessary and sufficient for initiating AMIS localisation and that ECM is insufficient to direct AMIS localisation in the absence of E-cadherin. Our results, therefore, suggest that Cadherin-based cell-cell adhesion provides the spatial cue required for AMIS localisation during de novo polarisation. This in turn localises apical proteins such as PAR-6B to a centralised region of cell-cell contact (Fig 2), determining where the lumen will arise. Whilst we demonstrate that AMIS localisation can occur independently from cell division, the importance of abscission and midbody formation in apical protein targeting has been robustly demonstrated and the molecules involved are now starting to emerge (Schl€ uter et al, 2009;Li et al, 2014;Wang et al, 2014Wang et al, , 2021Klinkert et al, 2016;Luj an et al, 2016;Mangan et al, 2016;Rathbun et al, 2020). Rather than acting as the initial symmetrybreaking step in AMIS localisation, we suggest that tethering of apically directed proteins to the midbody might instead act to transiently align cell division, cell adhesion and the forming apical domain, therefore enabling an organised structure to be generated in the presence of dynamic cell movement and tissue growth (Buckley & Johnston, 2022). The localisation of scaffolding and tight junction-associated proteins such as PAR-3 and ZO-1 at the AMIS might aid in this alignment. For example, Cingulin is a tight junctional protein that has been shown to bind both to the midbody and to FIP5, which is important for the apical targeting of vesicles containing apical proteins (Mangan et al, 2016). During zebrafish neural rod development, cell adhesion and cell division align to allow an organised structure to arise from dynamically reorganising cells (Symonds & Buckley et al, 2020) and loss of Ncadherin results in mis-oriented cell divisions and a disrupted ◀ Figure 6. Lumenogenesis in wild-type and E-cadherin knockout mESC cultures.

A PAR-3 immunofluorescence in wild-type (ES-E14) and E-cadherin knockout (Cdh1 KO) mESCs cultured in Matrigel for 48 and 72 h. At 48 h, most WT cell clusters had formed a polarised rosette or small lumen (i). However, most Cdh1 KO cells had not formed a central AMIS (ii). A small percentage of Cdh1 KO cells had formed an
open "cup-shape," with apically localised PAR-3 (iii). By 72 h, a significant proportion of Cdh1 KO cells had formed polarised "cavity-like" structures, about half of which were configured in an open "cup-shape" and half as closed "lumen-like" structures (iv). B Podocalyxin immunofluorescence in WT and Cdh1 KO mESCs cultured in Matrigel from 1-4 days. See Fig EV5A for ZO-1 staining. C Examples of masked Cdh1 KO cell cluster surfaces from "open" and "closed" cavities. The cavities were categorised based on Podocalyxin signals. D Percentage of cell clusters with different cavities relative to total cell clusters at different time points. The analysis was compared between the closed/rosette category among the conditions. E Percentage of cell clusters with closed cavities relative to total cell clusters with cavities calculated from (D). Whilst we have demonstrated that E-cadherin directs AMIS localisation, we do not yet have a full explanation for why AMIS proteins localise at the central-most point of cell-cell contact in the absence of divisions, despite E-cadherin localisation all along the cell-cell interface. As mentioned, PAR-3 has been shown to directly bind to the transmembrane JAMs and Nectin proteins (Ebnet et al, 2001;Itoh et al, 2001;Takekuni et al, 2003). However, these proteins localised to the AMIS later than PAR-3 and were not necessary for AMIS localisation (Fig 4). PAR-3 and PAR-6 have been found to be directly recruited to Cadherin proteins within endothelial cells (Iden et al, 2006). Recently, opposing actin flows in migrating cells as they first encounter each other were found to be responsible for regulating the first AJ deposition via tension-mediated unfolding of a-catenin and further clustering of surface E-cadherin molecules (Noordstra et al, 2021). Together, this could provide an explanation for how the first contacts between cells could act as an apical "seed," therefore defining the position of the AMIS within multi-cellular tissues. This could, therefore, explain why we have previously seen an upregulation of N-cadherin at the zebrafish neural rod midline, where cells growing from either side of the organ primordium meet (Symonds & Buckley et al, 2020). However, it is still unclear how this might regulate the subcellular localisation of the AMIS to the centre of cell-cell contacts. Our FRAP results demonstrate that E-cadherin is more stable at the central-most point of contact between two adhering cells (Fig 3J and K). This might suggest that E-cadherin is more stably bound via its downstream partners to the internal actin cytoskeleton at this point, which might help to stabilise or to localise AMIS proteins. In line with this hypothesis, previous publications have demonstrated an upregulation of phosphorylated MYOSIN-II (p-MLC) at the AMIS (Mol e et al, 2021), which is suggestive of higher actomyosin-mediated tension. Uncovering the mechanisms directing adhesion-dependent AMIS localisation precisely to the midpoint of cell-cell adhesions will be an interesting area for future studies. In addition, a recent study of chick neural tube polarisation (where N-cadherin is the dominant Cadherin) has demonstrated that the interaction of β-catenin with pro-Ncadherin in the Golgi apparatus is necessary for the maturation of N-cadherin, which is in turn important for apical-basal polarity establishment (Herrera et al, 2021). This provides the possibility that the polarised Golgi apparatus that we observe in the mESC clusters might be directionally delivering mature E-cadherin to the central-most region of cell-cell contact.
In the absence of an AMIS, lumens form via "closure" rather than hollowing The centralised localisation of an AMIS appears necessary to enable lumen hollowing within multi-cellular clusters. Cdh1 KO cells lack AMIS localisation at the 24-h AMIS stage (Fig 3). However, they still retain their apico-basal polarity axis (as denoted by Golgi apparatus and centrosome localisation, Fig 3G-I) and form apico-lateral junctions at luminal stages of development (Fig 6). Therefore, Cdh1 KO cells appear to still make an apical membrane (presumably directed by ECM-mediated signalling) but do so more slowly than in WT cells and without going through a centralised AMIS stage. This suggests that the role of E-cadherin in de novo polarisation is specifically to localise the AMIS, which enables the integration of individual cell apical domains to a centralised region preceding lumen hollowing. The lack of a centralised AMIS in E-cadherin deficient cells could also explain the multiple-lumen (but otherwise polarised) phenotypes previously seen in E-cadherin deficient MDCK cells cultured on collagen (Jia et al, 2011). Although the other adhesion molecules we have tested (P-cadherin, JAM-A and Nectin-2) did not contribute to centralised AMIS formation, mESCs cultured in Matrigel and mouse inner cell mass cells only become fully epithelialised and start to generate the central cavity once they have exited pluripotency and there are multiple cells in the structures (Shahbazi et al, 2017;Kim et al, 2021). Thus, whilst Ecadherin appears to be essential for AMIS localisation, other adhesion molecules may be important at later polarisation and lumenogenesis stages.
A surprising observation was the ability of Cdh1 KO mESC clusters, in the absence of AMIS localisation, to instead form "lumen-like" structures via a "closure" process (summarised in Fig 7B). Our movies of Cdh1 KO cell clusters (Movies EV3 and EV4) confirmed conclusions from fixed data (Fig 6) that Cdh1 KO cell clusters first generate a polarised, open cup-shape cavity, before "closing." Due to phototoxicity, we only had limited sample size and movie lengths; thus, we were not able to fully exclude the possibility that the hollowing lumenogenesis occurs to some small extent in parallel, but our data are not suggestive of hollowing lumenogenesis in the Cdh1 KO cell clusters. We do not currently know the mechanism by which such "closure" occurs in Cdh1 KO cell clusters. However, the presence of F-actin and p-MLC-rich cable-like structures in "cup"-shaped open cavities is potentially suggestive of a contractile process (Fig EV5). Understanding the relative roles of mechanics in localisation of the AMIS and in "opening" vs. "closing" tubes is an important future research goal, as is the potential role of cell geometry in mediating such differences. Additionally, collective cell migration could play a role in this "closure" mechanism. Collective inwards migration of cells caused lumen formation via a folding mechanism when MDCK monolayers were overlaid with a soft collagen gel (Ishida et al, 2014). A similar collective process could be occurring in the Cdh1 KO cell clusters from our study, which were cultured in a soft (10%) Matrigel and formed loosely connected cell clusters, which then "closed" to make a centralised lumen.
In summary, our work suggests that Cadherin-mediated cell-cell adhesion directs AMIS localisation during de novo polarisation of epithelial tubes and cavities. Our work also suggests that ECM is insufficient to direct AMIS localisation in the absence of Cadherin. In parallel with the well-described role of the midbody in tethering apical proteins, this suggests that there are two, interrelated mechanisms of AMIS localisation: cell adhesion and cell division. The alignment of these cellular processes allows for redundancy in the system and provides an explanation for how an organised epithelial structure can arise within the centre of a proliferating organ primordium.

Cell cultures and treatment
Cell lines used in this study are listed in Table 1. mESC carriers were maintained in Feeder Cell medium in Corning cell culture dishes precoated with 0.1% Gelatin (ES-006-B, Sigma-Aldrich), at 37°C suppled with 5% CO 2 at one atmospheric pressure. The culture medium was renewed every 3 days. The cells were trypsinised when reaching confluency to be passaged or subjected to experiments. The cells were regularly checked to be mycoplasmacontamination-free.
For 3D cultures of wild-type and Cdh1 KO mESCs, 20 μL of Matrigel (356231,Corning,Lot 354230,354234,356231) was spread evenly to the bottom of each well in a μ-slide 8-well dish (80821, Ibidi). The dish was left on ice for 10 min to flatten the Matrigel surface and then was left at 37°C for 10 min to solidify the Matrigel. mESCs were trypsinised, pipetted thoroughly and passed through a cell strainer (431750, Corning) to isolate cells into single cells. Singled mESCs were suspended in the N2B27 medium and seeded onto the solidified Matrigel. The seeded density was: control, 14 cells/mm 2 ; mitomycin C-treated, 227 cells/mm 2 . The cells were left at 37°C for 15 min when over 95% of the cells attached to the Matrigel, then the culture medium was renewed to 10% Matrigel/N2B27 medium with or without 2i/LIF.
For control and Cdh1 KO mESC chimeric cluster cultures, wildtype and Cdh1 KO mESCs were mixed at 1:4 ratio and co-cultured in 2D in the Feeder Cell Medium. They were then treated with mitomycin C for 2 h, trypsinised and seeded for 3D Matrigel culture at 227 cells/mm 2 .
For mESC cultured in agarose, 5,000 control or 125,000 mitomycin C-treated cells were suspended in a 37°C warmed 20 μL 0.5% low melting point agarose (16520, Invitrogen) droplet at the bottom of the μ-slide 8-well dish. The dish was left at room temperature for 5 min to solidify and topped with the N2B27 medium. The cells were then cultured at 37°C, 5% CO 2 until analysis.
For cells cultured on E-cadherin and fibronectin-coated glass, the μ-slide 8-well dish was incubated with nitrocellulose/methanal at 37°C for 3 h and left to air dry. The dish was then incubated with 40 μg/ml mouse E-cadherin recombinant protein (8875-EC-050, Bio-Techne) or 40 μg/ml fibronectin (F1141, Sigma-Aldrich) at 4°C overnight. The dish was briefly washed with water. Mitomycin C pretreated ES-E14 cells were seeded onto the dish at 14 cells/mm 2 in N2B27 medium. The cells were allowed to attach to the glass at 37°C for 1 h, and then the medium was renewed to N2B27 medium with 20% Matrigel. The cells were fixed 24-h post-Matrigel introduction.
For mitomycin C treatment, the cells were incubated with 10 μg/ ml mitomycin C (J63193, Alfa Aesar) in culture media at 37°C for 2 h. The mitomycin C-contained media were removed, and the cells were washed with PBS briefly. Then, the mitomycin C-treated cells were trypsinised and subjected to further experiments.
Mouse PAR-6B coding DNA sequence (cDNA, GenBank: BC025147.1) was assembled with mCherry by Gibson assembly, and LifeAct-Ruby cDNA were sub-cloned from an existing pRN3P-LifeAct-Ruby plasmid. The cDNAs were cloned into pDONR221 plasmid and introduced into the PB-Hyg-Dest plasmid using Gateway technology (Thermo Fisher Scientific). The PB-Hyg-Dest-mCherry-PAR-6B or the PB-Hyg-Dest-LifeAct-Ruby plasmid was cotransfected with the piggyBac plasmid using Lipofectamine 3,000 to generate Hygromycin B resistant stable cell lines. The mCherry-PAR-6B or LifeAct-Ruby expressing mESC stable cell line was created via 10 μg/ml Hygromycin B selection and single-cell colonies expansion. Primers used for cloning are listed in Table 2.

siRNA transfection
To achieve protein knockdowns, siRNA was transfected using Lipofectamine RNAiMAX according to the manufacturer's instructions. mESCs cultured in 2D on the gelatin were transfected with 100 nM pre-designed siRNA (s63752, Silencer Select). siRNA target mRNA sequences were as follows: E-cadherin, GAAGAUCACGUAUCGG AUU; P-cadherin, CGAAAGAGAGAGUGGGUGA; JAM-A, GCCUUU GAUAGUGGUGAAU; Nectin-2, GGACUACUGAAUUCUUUUA. Two days after the first transfection (Fig EV4A), the cells were cultured with or without mitomycin C and seeded into 10% Matrigel. The cells were cultured in Matrigel with Lipofectamine RNAiMAX and 100 nM siRNA for another 24 h and then were subjected to further experiments and analysis.

Immunofluorescence
Cells cultured in a μ-slide 8-well dish were fixed with 4% paraformaldehyde (J61899, Alfa Aesar) for 30 min at room temperature and then were permeabilised with 0.5% Triton X-100 for 15 min at room temperature. The cells were blocked with the incubation buffer (0.5% BSA, 0.1% Tween in PBS) for 2 h and then were incubated with primary antibodies diluted in the incubation buffer at 4°C overnight on an orbital shaker. The primary antibodies were washed off with PBS and then were incubated with secondary antibodies diluted in the incubation buffer at room temperature for 2 h. The secondary antibodies were washed off with PBS. Samples cultured in Matrigel were kept in PBS; samples cultured in agarose were sealed in 200 μl 0.5% agarose. The samples were imaged shortly after. Antibodies and dilutions were as listed in Table 3.

Microscope imaging
Live cell imaging was carried out on the PerkinElmer UltraVIEW spinning disc system fitted on an Olympus IX80 confocal microscope with a 37°C and 5% CO 2 chamber. Images were captured with the 40× 1.

Image and data analysis
The central section images were projected from raw images in the Fiji software by maximum-value projection of the whole z stacks to produce the 3D projection images, or of the central three images of the z stacks to produce the central section images. The whole z stacks projections were applied to count cell cluster percentages. The central section images were applied for line-scans or analysis of region of interest to determine protein signals. The mESCs with positive or negative protein centres were manually determined and counted. The percentage was calculated with the number relative to the number of total cell clusters captured in each condition. The mean percentages from three independent experiments were compared with Student's t-test or one-way ANOVA specified in figure legends using the GraphPad Prism software. Sample sizes are specified in figure legends.

5-GGGGACCACTTTGTACAAGAAAGCTGGGTGTTACCCTCCGCCCAGGCCGG-3
To quantify PAR-3 signal along the cell-cell interface, a 0.8 μm width line was drawn along the cell-cell interface by using F-actin (labelled by Phalloidin) or E-cadherin signal as the path. PAR-3 and F-actin pixel values along the path were extracted. The two peaks of F-actin signals at two ends of the path were determined as the start and end of the cell-cell interface, and the positions in-between were defined as 1× cell-cell interface. The corresponding PAR-3 pixel to the F-actin peak positions was identified, and the PAR-3 line profile between the two positions was sectioned into 20 sections. PAR-3 pixel value in each section was averaged to be the PAR-3 signal in 5% of the cell-cell interface. The values from 10 to 20 cells were plotted as line graphs.
To compare the level of PAR-3 at the centre and outer regions in multi-cell mESC clusters, a 4-μm-diameter circular area was created in the Fiji software and placed over the multi-cellular joint in a cell cluster, and the average PAR-3 fluorescence values in the circle were measured. The boundary of the cell cluster was drawn by the free-hand tool in Fiji by using the correspondent F-actin signal, and the average PAR-3 fluorescence values between the circle and the boundary were measured. The ratio between the PAR-3 values in the circle and outer regions was then calculated.
To generate the PAR-3 signal heatmap, based on the F-actin signal, a squared region of interest was created in the Fiji software for every cell, which the main cell body of a cell was fitted into. The PAR-3 image in each squared region was transformed into a 10 × 10-unit matrix by using the R-language. Each unit was created by averaging the PAR-3 fluorescence intensity in every 10% length along the X or Y axis of the squared region of interest extracted from the original image. A serial of images from one condition of an experiment was stacked, and the averaged pixel value at each position was calculated to generate an averaged 10 × 10 matrix. The final matrix was transformed into a heatmap in the GraphPad Prism software.
To compare the level of PAR-3 in the core areas in the cells culture on the glass, a 6-μm-diameter circular area of interest was created in the Fiji software to cover the PAR-3 core area in a cell, and the average PAR-3 fluorescence values in the circle were measured. The boundary of the cell was drawn by the free-hand tool in Fiji, and the average PAR-3 fluorescence values in the boundary were measured. The ratio between the PAR-3 values in the circle and inside the boundary was then calculated.
Sample sizes and statistical analysis are detailed in the figure legends. No blinding was done for the experiments in this study. Instead, we included direct continuous measurements (e.g. of fluorescent intensity) alongside categorical analyses. To assess AMIS localisation independently to cell division, we did not include control cells that were undergoing mitosis (identified by chromosome and cell morphologies) into the quantifications. We checked that excluding this data made no difference to the outcome of the analyses.

FRAP
FRAP of E-cadherin-eGFP was performed to measure E-cadherin dynamics at the cell-cell interface. An E-cadherin-eGFP expressing stable ES-E14 line was used to perform the experiment (Mol e et al, 2021). After 24-h culturing in Matrigel, two-cell clusters with long axis parallel to the surface of the culture dish, hence the cellcell interfaces were aligned to the axis of the objective were used for FRAP. A region of interest (ROI) of 2 × 1 μm along xy-axis was chosen at the approximately centre-most region or side regions at each cell-cell interface. The ROI in a series of z-axis stacks with 0.3-μm intervals totalled to 2.1 μm in depth was bleached with three iterations of the 488 nm laser with 100% transmission. This resulted in a photobleaching of over 80% in a 2 × 2.1 μm (xz-axis) region at the cell-cell interface. Time-lapse images were acquired before (three frames) and after (30 frames) photobleaching with an interval of 10s per frame. Average fluorescence intensity values F(t) in the bleached area within the centre z-stack were analysed with ImageJ.
The mean values of the three frames before bleaching were used as the pre-bleached value F(i). The value of the first frame after bleaching was defined as F (0). FRAP values were then calculated and plotted over time in Fig 3J as: The FRAP values were fitted using a non-linear regression and the exponential one-phase association model using Y0 = 0 and where mobile fraction corresponds to the plateau value in the GraphPad Prism software. For Fig 3K, the mobile fraction from each FRAP profile was pooled and compared between conditions.
Expanded View for this article is available online.