Quantitative cell polarity imaging defines leader-to-follower transitions during collective migration and the key role of microtubule-dependent adherens junction formation

The coordinated remodelling of epithelial sheets drives the morphogenesis of many organ systems. Epithelial plasticity can occur through epithelial mesenchymal transitions (EMTs) and mesenchymal epithelial transitions (METs), the reversible process whereby static epithelia are broken down to form individual, motile mesenchymal cells. This full cell state conversion witnesses changes in cell shape and polarity, gene expression and signalling (Thiery et al., 2009; Lim and Thiery, 2012; Nieto, 2013). Collective migration is an alternative, but related, process whereby a subset of cells at the tissue edge, termed ‘leader’ cells, displays reduced apicobasal polarity and increased dynamic actin-based protrusions, but maintains stable contacts with trailing epithelial ‘follower’ cells (Friedl et al., 2004; Lecaudey and Gilmour, 2006). Unlike EMT, collective migration does not necessarily involve a full state conversion but rather modulates the degree of epithelial organisation, which we will refer to as increased epithelialisation (Revenu and Gilmour, 2009). Collective migration therefore allows epithelial tissues to become organised while moving. However, understanding the spatiotemporal coordination of this process has been limited by a lack of quantitative analysis of cell behaviour in these three-dimensional in vivo contexts (Montell, 2008).

One dynamic readout of cell organisation is provided by the relative position of the centrosome and the nucleus, which has enabled the analysis of polarity in cells cultured in two-dimensions (Théry et al., 2006; Dupin et al., 2009; Rodríguez-Fraticelli et al., 2012). Moreover, nucleus-centrosome axes provide information on the organisation of the microtubule network, which is cell type specific (Bacallao et al., 1989; Meads and Schroer, 1995; Müsch, 2004; Brodu et al., 2010; Hong et al., 2010). It is known that microtubules regulate, and are regulated by, points of cell-cell contact (Chausovsky et al., 2000; Waterman-Storer et al., 2000; Yap and Manley, 2001; Ligon and Holzbaur, 2007), raising the possibility that microtubule network organisation is regulated at the supracellular or tissue level, as has recently been shown for the actomyosin system (Martin et al., 2009; Rauzi et al., 2010). However, there are a number of in vivo contexts where the ablation of microtubules was reported to have little influence on adherens junctions (Jankovics and Brunner, 2006; Brodu et al., 2010; Sumigray et al., 2012), indicating that the role of microtubules in tissue-shaping remains unclear.

The zebrafish posterior lateral line primordium (pLLP) provides a useful in vivo model for studying the interplay between motility and tissue assembly during collective migration. While migrating, cells undergo a stereotyped switch in organisation during their transition from less organised leading region to tightly packaged, rosette-like mechanosensory organ progenitors, the proneuromasts, that are regularly deposited. Although signalling pathways that regulate migration (David et al., 2002; Haas and Gilmour, 2006; Valentin et al., 2007; Aman and Piotrowski, 2008) and rosette assembly (Lecaudey et al., 2008; Nechiporuk and Raible, 2008; Hava et al., 2009) have been identified, and some of their downstream effectors are known (Ernst et al., 2012; Harding and Nechiporuk, 2012), much less is understood about the changes in the cytoskeleton and cell-cell contacts that execute these cellular behaviours. In this work, we present a quantitative three-dimensional cell polarity imaging approach of cell dimensions and orientations across the pLLP that allows the organisation of the migrating tissue to be assessed. We integrate these data with novel live cadherin reporters to further address changes in cell adhesion across the migrating collective and demonstrate a requirement for dynamic microtubules specifically during the leader-to-follower transition.

In order to map cell polarity and orientation across the migrating collective, we used the nucleus-centrosome axis, which provides a readout of front-rear polarity in migrating cells (Gotlieb et al., 1981; Etienne-Manneville and Hall, 2001; Gomes et al., 2005) and apicobasal polarity in epithelial cells (Rodríguez-Fraticelli et al., 2012). To achieve this, we developed methods to live label simultaneously microtubules (EMTB-3GFP), centrosomes (centrin-3BFP) and nuclei (nls-tdTomato). Mosaic expression of EMTB-3GFP allowed the segmentation of the microtubule network, its associated centrosome and nucleus of individual cells (Fig. 1). The polarity of each cell was extracted by generating vectors connecting the centre of mass of each nucleus to its corresponding centrosome. Assigning these to multiple cells provided the first quantitative information on the dimensions and orientations of cells across the primordium. As primordia at any stage show a stable configuration of a leading domain plus two or three trailing rosettes, the front-to-rear axis provides a simple and robust timeline for the organogenesis process. Therefore, cells were positioned in a reconstituted virtual primordium as a function of their distance from the leading edge, to generate a polarity map of the tissue (Fig. 1). As the exact positioning of trailing rosettes varies slightly from tissue to tissue, cells were registered to the mean position of their closest rosette centre (Fig. 1). In total, 340 cells were analysed from 127 primordia. As the average primordium comprises about 100 cells, this represents approximately three times tissue coverage.

We first used this quantitative data set to define how cell dimensions change across the tissue by analysing the nucleus-centrosome distance, i.e. the length of the polarity vectors. This revealed a progressive elongation when moving rearward from the leading edge of the tissue (Fig. 2A,B). Principal component analysis (Pincus and Theriot, 2007; Keren et al., 2008) of microtubule networks in three dimensions enabled analysis of cell body aspect ratio. EMTB labelling allows the analysis of the global cell volume while excluding the highly variable, flat, actin-based membrane protrusions. It confirmed the progressive increase in length along one cell axis, indicating that cells shift from a more spherical to an elliptical shape across the front-rear tissue axis (supplementary material Fig. S1A), consistent with the acquisition of a columnar morphology. The polarity vectors of the longest cells aligned with the migration axis of the primordium (supplementary material Fig. S1B), indicating that individual cells of the primordium become elongated in the direction of migration. However, although a global increase in length is a clear trend across the tissue, a subset of cells does not show such elongation (Fig. 2A; supplementary material Fig. S1A). One likely reason for such length heterogeneity in the forming organs is the assembly of cells into rosettes-like progenitors, which presumably constrains cell geometries. Indeed, looking at the distribution of lengths around the rosettes reveals that cells are shorter closer to the rosette centre, becoming more elongated in lateral regions (Fig. 2B; supplementary material Fig. S1C). In addition, analysing vector polar angles, which provides a measure of how ‘upright’ cells are by calculating deviation from a pole running perpendicular to the main tissue axis, shows that cells are progressively more upright towards the centre of the rosettes (Fig. 2C). Thus, cell length increases during the transition from leaders to epithelial organ progenitors with cells within the rosette assemblies showing heterogeneous lengths as they become organised in a ‘firewood stack’ configuration.

We next addressed the direction of polarity vectors to determine how the centrosome reorients relative to the nucleus as cells become organised into rosette-like organ progenitors. The azimuth angle, i.e. the orientation of the polarity vectors in the migration plane, allowed us to quantitatively track direction changes (Fig. 2D). Azimuth angles across assembled epithelial rosettes span the whole angle range, as expected for a radial organisation. By contrast, cells within the leading 15 μm of the tissue have a strict rear orientation, meaning that cells have their centrosomes located behind the nuclei (Fig. 2D), as described for cells at the very leading edge (Pouthas et al., 2008). Front-oriented centrosomes begin to appear between 15 and 45 μm from the pLLP tip, identifying a transition zone between the leaders and the radially organised followers where centrosomal-nuclear vectors show a mixture of polarities. Performing time lapses of mosaic pLLP labelled with the microtubule marker EMTB allowed this rear-to-front shift to be dynamically confirmed at the single cell level (Fig. 2E). The transition between migrating leaders and epithelial followers is thus marked by a gradual reorientation of a subset of cell centrosomes towards the front.

The described changes in length and orientation are likely driven by modifications in cell-cell interactions. We thus turned to cadherin-based junctions to assess whether changes in cell adhesion could explain these coordinated changes in the tissue. Previous work has shown that the pLLP expresses E-cadherin/cadherin1 (Cdh1) and N-cadherin/cadherin2 (Cdh2) (Liu et al., 2003). Anti-Cdh1 and anti-Cdh2 antibody stainings revealed co-expression of both cadherins in all cells of the pLLP (supplementary material Fig. S2A,B). Interestingly, anti-Cdh2 showed enhanced localisation to apical adherens junctions (AJs) that directly abut ZO1-labelled tight junctions (supplementary material Fig. S2C), a configuration that is a characteristic of epithelia. Surprisingly, Cdh2/parachute mutants show apparently normal pLLP migration and organ morphogenesis (supplementary material Fig. S2F,G), a lack of phenotype that is most likely explained by redundant expression of cadherins in the primordium (Liu et al., 2003). Nevertheless, the discrete localisation of Cdh2 makes it an excellent marker of apical AJs. In order to monitor the dynamics of AJ assembly across the migrating primordium, we generated fluorescently tagged live reporters for Cdh2 by bacterial artificial chromosome (BAC)-mediated complementation. Transgenic lines where Cdh2 was fluorescently tagged recapitulated the pattern of endogenous Cdh2 expression (Fig. 3A; supplementary material Fig. S2D) and were able to rescue the gross morphological defects of Cdh2/parachute mutants, proving that these reporters were physiologically functional (supplementary material Fig. S2E). Ratiometric imaging of Cdh2 with a uniform membrane marker, lyn-GFP, confirmed a strongly enhanced apical Cdh2 localisation in cells of the rosettes (Fig. 3B,C), demonstrating that Cdh2 apical concentration is not simply the result of increased plasma membrane density. Cdh2 thus provides a live reporter for AJs across the migrating collective. Live imaging revealed that apically enriched Cdh2 first becomes detectable as discrete puncta, which then coalesce and fuse into larger multicellular clusters (Fig. 3D; supplementary material Movie 1).

In order to assess the link between AJ formation and cell reorientation, we quantified the localisation of Cdh2 clusters across the tissue front. Interestingly, an apical enrichment of Cdh2 was undetectable within the leading 15 μm of the tissue, with cells here instead displaying more uniform plasma membrane labelling, but appeared in the transition zone between the leading domain and the first rosette constriction (Fig. 3E). Moreover, the appearance of Cdh2 clusters shows a high spatial correlation with the percentage of front-oriented polarity vectors in the leading zone (Pearson correlation of 0.9351, Fig. 3E). Imaging of clones of cells transplanted from Cdh2-GFP- and centrin-tdTomato-expressing embryos into unlabelled hosts allowed us to follow this transition directly in single cells and suggests that cell reorientation follows Cdh2 cluster formation (Fig. 3F; supplementary material Movie 2). Leader cells thus convert from a rear orientation to a radial organisation in a transition zone where cells reorient and apical AJ clusters start to form. The percentage of front-oriented cells is therefore a simple and robust proxy of epithelial organ assembly within the primordium.

Having defined the cellular changes that characterise the leader-follower transition at the tissue level, we next addressed the functional requirement of the microtubule network in this morphogenesis process. We used colcemid to prevent microtubule polymerisation, the efficacy of which was confirmed by showing loss of the microtubule plus-end tracker Eb3-tdTomato (supplementary material Movie 3), and captured the response of the tissue by time-lapse imaging. Colcemid-treated primordia displayed a reproducible phenotype where organ progenitors are no longer detectable due to the loss of rosette-like constrictions (supplementary material Fig. S3A,B). In order to exclude that these morphological changes were a consequence of mitotic arrest, we blocked cells in S-phase using the DNA synthesis inhibitor aphidicolin. Aphidicolin treatment suppressed the appearance of mitotic cells but did not prevent the loss of the rosette constrictions (supplementary material Fig. S3A). UV inactivation of colcemid (supplementary material Movie 3) resulted in surprisingly efficient tissue reassembly (supplementary material Fig. S3B and Movie 4), showing that dissociated rosette organisations can rapidly return to their normal state upon microtubule regrowth. Thus, the initial assembly of rosette-like organ progenitors is tightly coupled to the status of dynamic microtubules within the collective. In order to quantify this effect, we assigned polarity vectors to 341 cells from six samples where microtubule growth was inhibited using colcemid treatment (Fig. 4A). Measuring nucleus-centrosome distances revealed that leader cells were of normal length whereas followers were shorter compared with control (Fig. 4C). This demonstrated that cell elongation during the leader-follower transition was lost upon microtubule depolymerisation. Plotting the percentage of front-oriented vectors across control primordia showed three peaks corresponding to organ progenitors (Fig. 4B). Identical analysis of cell orientation after colcemid treatment revealed that the percentage of front-oriented vectors remained low and increased sharply only at the tissue rear (Fig. 4B). Thus, although most cells are unable to become radially organised in the absence of dynamic microtubules and display a leader-like orientation, cells at the rear of the tissue maintain their multicellular organisation, suggesting that in these regions cell-cell adhesion is maintained. Indeed, complete loss of cell cohesion induced by the calcium chelator EGTA (supplementary material Movie 5), which inactivates calcium-dependent cell adhesion molecules such as cadherins, resulted in the loss of tissue organisation (Fig. 4D), as confirmed by decreased nucleus-centrosome distances along the entire primordium (Fig. 4C) and randomised orientations across the tissue (Fig. 4D). We conclude that dynamic microtubules are not essential for the cohesion of cells of the leading domain or for the integrity of mature epithelial rosettes deposited by the migrating tissue. However, there is a strict requirement for dynamic microtubules during the leader-to-follower transition.

In order to address the impact of microtubule depolymerisation on cadherin-based cell adhesion, we performed time-lapse analysis of Cdh2-GFP in colcemid-treated embryos (Fig. 5A). Nascent Cdh2 clusters of the transition domain no longer appeared, indicating that the initial clustering was strongly dependent on microtubules. Pre-existing Cdh2-GFP apical clusters (Fig. 5A) and ZO1 apical staining (supplementary material Fig. S3C) were lost in a front-to-rear wave (Fig. 5A), as were the rosette-like membrane constrictions (Fig. 5B). By contrast, Cdh2-GFP clusters in more mature, deposited organs proved to be refractory to microtubule depolymerisation (Fig. 5A). This increased resistance to microtubule depolymerisation towards the rear of the tissue indicates that a change in cadherin complex organisation could result in increased adhesion stability across the collective.

Work in the Drosophila blastoderm has shown that trans homophilic dimers of cadherin, i.e. those engaged in cell-cell junctions, are present in stable microdomains that show greatly reduced mobility and turnover when compared with diffusing ‘free’ monomeric cadherin (Cavey et al., 2008). Thus, differences in the turnover of cadherin complexes provide a reasonable proxy for adhesion changes in living tissues. We therefore investigated cadherin turnover using a tandem fluorescent protein timer (tFT) methodology (Khmelinskii et al., 2012) that we have recently shown provides a quantitative readout of protein turnover within the primordium (Donà et al., 2013). We created a BAC transgenic line where Cdh2 was tagged with a tFT comprising the rapidly maturing superfolder GFP (sfGFP) and a slower maturing red fluorescent protein (TagRFP). Here, the red-over-green ratio, or lifetime ratio, allows direct evaluation of the age of the protein pool (Donà et al., 2013). Whole-mount ratiometric imaging of the Cdh2-tFT reporter revealed variable lifetime ratios between tissues (Fig. 5C). Performing lifetime ratio analysis within the primordium reveals that Cdh2-tFT is significantly older in apical clusters when compared with basolateral membranes of the same cells, confirming that these microdomains represent stabilised, hence functional, adherens junctions (Fig. 5E). We next used the same cadherin lifetime imaging approach to monitor changes in AJ stability at the tissue scale, by extracting lifetime ratios from apical clusters in multiple migrating primordia and deposited neuromasts (Fig. 5D). Quantifying their aging compared with corresponding membranes reveals that adherens junctions become progressively more stable across the front-rear axis of the collective during the formation and maturation of epithelial organ progenitors (Fig. 5F). Thus, monitoring of cadherin stability in living tissues provides a plausible mechanistic explanation for the differential requirement for dynamic microtubules during this morphogenesis process.

A common feature of many migrating collectives is the conversion of less-organised leader cells to differentiated followers that become assembled into multicellular tissues, such as tubes or rosettes. Here, we have taken a quantitative imaging approach to understand this transition in one model migrating collective: the zebrafish lateral line primordium. Through the generation of nucleus-centrosome vectors, we obtained a three-dimensional map of cell elongation and polarisation within the tissue. We characterised the conversion from leaders to epithelial mechanosensory organ progenitors and could show that cells gradually elongate and shift from a strict rear orientation, with centrosomes located between the nucleus and the cell rear, to a radial organisation in a transition zone that is concomitant with the appearance of defined apical AJs (Fig. 6). This analysis at single cell resolution provides a powerful, potentially widely applicable, approach to link subcellular and tissue-scale polarities during organogenesis.

Combining this approach with chemical perturbation experiments highlights a requirement for microtubules specifically in cells undergoing the leader-follower transition. The rapid loss of AJs in these cells upon microtubule depolymerisation demonstrates that the microtubule network is required for the apical concentration and maintenance of the early adhesion clusters. Obviously, microtubule depolymerisation could have further effects on epithelial cell polarity than the loss of apical cadherin localisation alone, these will be the focus of future investigation. This loss of AJs is accompanied by the loss of the radial organisation, with cells mainly displaying a rear orientation typical of leader cells. Together with the tight spatial correlation between nascent AJ clusters and the reorientation of nuclear-centrosomal axes, this result suggests that forming apical cell-cell contacts could be required for cell reorientation, possibly via the microtubule network. This is an attractive hypothesis given that the position of the centrosome depends on cortical pulling forces on microtubules anchored at the cell cortex (Burakov et al., 2003) and adherens junctions are able to anchor microtubules (Ligon et al., 2001; Ligon and Holzbaur, 2007; Schmoranzer et al., 2009). Moreover, recent work has shown that shroom 3, which regulates γ-tubulin localisation and cell apico-basal elongation (Lee et al., 2007) is also localised apically in rosettes and is involved in organ progenitor formation in the pLLP (Ernst et al., 2012). As the effects of colcemid treatments are completely reversed after washout, this approach will enable future investigations into the establishment and maintenance of cell polarity within migrating tissues.

Work from a number of laboratories has shown that the centrosome is actively repositioned between the nucleus and the leading edge in cells migrating in two-dimensional cultures, a process that has been suggested to be important for directed migration (Gotlieb et al., 1981; Etienne-Manneville and Hall, 2001; Gomes et al., 2005). However, previous work has shown that when migrating cells are plated on constrained substrates, the Golgi and associated centrosome are repositioned behind the nucleus, leading to the suggestion that it could be a response to geometrical constraints (Pouthas et al., 2008). Here, we map the polarity of all nuclear-centrosomal axes across a tissue migrating in vivo and show that centrosomes invariably orientate towards points of cell-cell contact, as recently described in non-migrating epithelial MDCK cells plated at high density (Rodríguez-Fraticelli et al., 2012). For cells of the leading domain this results in the centrosome positioned behind the nucleus. Our finding that centrosomes are positioned towards cell-cell contacts, even in the most motile cells of the collective, indicates that these could be the dominant regulators of nuclear-centrosomal axes in vivo.

One surprising finding from this study is that cell-cell contacts of the collective differ strongly in their requirement for dynamic microtubules. First, cells of the leading domain remained cohesive in the absence of dynamic microtubules. Second, mature organ progenitors deposited by the migrating primordium also proved to be resistant to the effects of microtubule depolymerisation, with their apical AJs being maintained in colcemid-treated embryos. Our finding that cell states within the same migrating tissue differ strongly in their requirement of dynamic microtubules could help explain the discrepancies in studies on microtubule network regulation of AJs using different cell culture conditions or lines (Waterman-Storer et al., 2000; Yap and Manley, 2001; Ivanov et al., 2006; Ligon and Holzbaur, 2007; Komarova et al., 2012). The lack of requirement for dynamic microtubules once organ progenitors are mature indicates that AJs in these cells are stabilised through alternative mechanisms, such as changes in the actin cortex or other maturation processes. Indeed, we could show using a tandem fluorescent timer approach that Cdh2 becomes localised to clusters that become progressively more stable across the migrating tissue. This finding is of particular relevance given that cadherin complex stability can be used as a proxy for adhesion (Cavey et al., 2008). Although the cadherin-timer approach described here does not yet provide the same level of dynamic information as FRAP and photoconversion, these latter techniques require the development of time-lapse imaging and analysis protocols that are not easily applied to three-dimensional tissues. By contrast, the ‘snapshot’ nature of the cadherin-timer approach means that it can be used to monitor cadherin stability in essentially any context. For example, here we have been able to detect differences in cadherin stability at both subcellular and tissue scales using the same tandem fluorescent cadherin timer.

The conversion of leader cells to epithelial followers is expected to involve regulated changes in cell-cell adhesion. One prediction would be that cells acquire a differential Cdh1/E-cadherin and Cdh2/N-cadherin expression pattern, depending on their differentiation state (Matsuda and Chitnis, 2010), or, in analogy with METs, that cells undergo a switch in cadherin expression (Wheelock et al., 2008). Here, we show that all cells of the primordium express both Cdh1 and Cdh2, making a role for transcriptional cadherin switching unlikely in this context. Using live reporters, we demonstrate that it is the distribution of Cdh2 in the plasma membrane that changes during the transition, and, as discussed above, probably as a consequence of a local apical stabilisation of the proteins. These AJ clusters appear only in the transition zone, suggesting that leader cells have no defined apical domain and thus are more mesenchyme like in terms of organisation. High-resolution imaging of Cdh2 dynamics reveals early clusters that could be analogous to apical membrane initiation sites or pre-apical domains (Bryant et al., 2010; Yang et al., 2013) that behave like microdomains (Cavey et al., 2008). The formation of rosette-like organ progenitor could thus be driven through establishment of apical AJs, rather than the constriction of pre-existing apical domains. The existence of apical constriction is supported by the requirement for activation of actomyosin downstream of FGF and Rho-kinase during rosette formation (Ernst et al., 2012; Harding and Nechiporuk, 2012). However, there is evidence that this pathway may also act primarily in the concentration of cadherins and the formation of cell-cell junctions consistent with what we have described here (Shewan et al., 2005; Liu et al., 2010; Hoelzle and Svitkina, 2012). Given that cells of the leading domain express Cdh2, it is intriguing that they do not show detectable apical AJs. One interesting possibility is that increased cell-matrix adhesion in these leader cells suppresses the formation of apical AJs via an integrin-cadherin negative-feedback mechanism (Ehrlich et al., 2002; Marsden and DeSimone, 2003; Sakai et al., 2003; Borghi et al., 2010; Tseng et al., 2012), an idea that could potentially explain how epithelial polarity and motility are integrated across migrating collectives.

In conclusion, this work enables us to propose a model of the transition from leaders to followers within a migrating collective (Fig. 6). Microtubules allow the apical localisation and maintenance of discrete AJ clusters in cells positioned behind the leading edge. AJs become progressively more stable as organ progenitors differentiate, with adhesion becoming independent of the dynamic microtubule network in mature deposited organs. Our work indicates that adhesive clusters could feed back on the microtubules to induce cell reorientation during the transition, an idea that will be investigated in future work. Combined, our data support a model where cell-cell adhesion within the migrating collective is primarily modulated by dynamic changes in cadherin localisation and stability, not cadherin receptor abundance. The combined quantitative analysis of cell polarity and cadherin stability described here should facilitate similar investigations into the mechanisms underlying leader-follower transitions in other in vivo contexts.

The previously described cldnb:lyn-GFP transgenic line (Haas and Gilmour, 2006) was used to label cell membranes of the pLLP. To visualise the microtubule cytoskeleton, we chose a mosaic expression system using a 6×UAS promoter and the GAL4-UAS:mCherry (ETL GA346) line provided by R. Köster (Distel et al., 2009). For the microtubule network, EMTB tagged with three GFPs (Bulinski et al., 1999) was cloned into a Gateway pENTR vector (Invitrogen) with ClaI and NotI restriction sites. Transgenic lines were generated using the Tol2 Kit (Kwan et al., 2007). The Cdh2 reporter lines were generated by replacing the stop codon of the cadherin 2 gene with a sequence encoding the eGFP or TagRFP fluorophore or the tandem fluorescent protein timer sfGFP-TagRFP (Donà et al., 2013) in the bacterial artificial chromosome (BAC) clone DKEYP-38F5 using standard ET recombineering technology (Gene-Bridges). The modified BACs containing a crystallin beta:eCFP transgenesis marker inserted in the backbone were injected in the Cdh2 mutant strain parachute (PAC, allele tm101b) (Jiang et al., 1996). To label nuclei, a BAC line expressing a nuclear localisation signal tagged with tdTomato under the Cxcr4b regulatory sequences was used (Cxcr4b:nls-tdTomato) (Donà et al., 2013). Mosaic embryos were generated by introducing cdh2:Cdh2-GFP cells injected with centrin-tdTomato mRNA into wild-type hosts, following previously described transplantation methods (Haas and Gilmour, 2006).

All constructs were created using Multisite Gateway procedures (Invitrogen) and the SP6 promoter. Zebrafish centrin1 cDNA was cloned and fused in 5′ with the fluorescent proteins tdTomato or with three TagBFPs (Subach et al., 2008). The array of three TagBFPs was created by sequentially ligating a BamHI-TagBFP-BglII PCR product into a BamHI-digested pENTR vector. The human Eb3 cDNA (Stepanova et al., 2003) was fused in 3′ with tdTomato. Capped mRNAs were transcribed using the SP6 mMessage mMachine Kit (Ambion) and injected into one-cell stage embryos.

These analyses were performed using custom codes written in MATLAB (MathWorks); scripts can be provided on request. For segmentation of individual cells, z-stacks with a sectioning of 500 nm were acquired using a 40× 1.2 NA water objective on PerkinElmer Improvision Ultraview spinning disc confocals (ERS and VOX) and deconvolved using Huygens Remote Manager (Ponti et al., 2007). Microtubule network, nucleus and centrosome were segmented in three dimensions using local background correction (disc size in the order of 5, 45 and 2, respectively), thresholding and Mexican hat filtering (standard deviation in the order of 0.7, 7 and 1, respectively). For the analysis of polarity vectors after drug treatments, nuclei, centrosome and membrane were segmented using the interactive machine learning Ilastik software (C. Sommer, C. Straehle, U. Köthe and F. A. Hamprecht, unpublished). Segmented nuclei and centrosome pairs were manually assigned using the microtubule network or the membrane to determine cell boundaries. To extract the position of each analysed cell in the corresponding primordium, the primordium was masked from the nuclei channel using an active contour method (Chan and Vese, 2001) to detect the orientation and the tip of the tissue. For the rosette based pLLP maps, the positions of the rosettes were manually assigned using the centrosomes and the nuclei channels or the membrane channel after colcemid treatment. For the percentage of front-oriented vectors and the mean length calculations, sliding windows of 10 and 30 μm, respectively, were used along the main tissue axis.

For immunohistochemistry, the following antibodies were used: mouse anti-E-cadherin (1/500; BD Transduction Laboratories), rabbit anti-cadherin 2 (1/3000) (Liu et al., 2001) and mouse anti-ZO-1 (1/200; Zymed). Approximately 32 hpf dechorionated embryos were fixed in 4% paraformaldehyde and permeabilised with PBS, 0.3% Triton X-100, 1% DMSO, 1% BSA and 0.1% Tween.

Approximately 32 hpf embryos were anaesthetised in 0.01% tricane in E3 medium and mounted on glass-bottom Petri dishes (Mattek) with 1% low melting point agarose. Imaging was performed on PerkinElmer Improvision Ultraview spinning disc confocals (ERS and VOX) or Zeiss LSM 780 confocal using Zeiss 40× or 63× 1.2 NA water objectives. Two-photon time lapses were acquired with a Zeiss LSM 780 NLO and a 63× 1.2 NA water objective. Acquired z-stacks spanning the whole primordium were processed using ImageJ and Fiji software. Images shown are generally maximal intensity projections. When appropriate, subsequent fields were stitched using the pairwise stitching plug-in (Preibisch et al., 2010). For Cdh2/lyn ratio analysis, after local background subtraction, a mask of the primordium was created using the lyn-GFP channel by Gaussian filtering, thresholding and erosion. The obtained mask was then applied on the Cdh2-TRFP channel after median filtering. The resulting image was divided by the lyn-GFP one and colour-coded with the ratio look-up table. For quantifications, a region of interest of approximately 50 μm² of masked pixels was drawn on the rosette centre. The mean ratio was measured in a z plane crossing the apical concentration and 5 μm lower for each rosette. For the analysis of apical cluster distribution, Cdh2 clusters were segmented using local background subtraction, thresholding and a minimum size filter of 0.1 μm².

Embryos were imaged on a Zeiss LSM 780 confocal microscope as previously described (Donà et al., 2013). A z-stack spanning the apical clusters and a single confocal plane in the middle plane of the primordium or of the deposited neuromasts were acquired for each sample. Owing to the extreme difference in fluorescence intensities between basolateral membranes and apical constrictions, two different acquisition settings had to be used. For each acquisition setting, background and normalisation solution (sfGFP-mCherry fusion protein) images were also acquired.

After background subtraction, apical cluster and basolateral membrane regions were manually defined on individual confocal planes using the freehand line tool in Fiji (line width=0.9 μm). To calculate lifetime ratios, mean red and green fluorescence intensities were calculated and the resulting red/green ratios were normalized by the ratio of the sfGFP-mCherry reference solution (Donà et al., 2013).

To analyse the stability of Cdh2 in the clusters relative to corresponding membranes, we analysed the data as follows. For each apical cluster, the mean ratio of the underlying basolateral membrane ratio was subtracted, in order to normalise for age differences across the anterior-posterior axis of the primordium owing to other factors affecting protein pool age, like transcription rate or cell division rate differences. For each embryo, these values were normalised to that of the front-most cluster, so that the stability of each apical cluster would be expressed relative to that one.

For EGTA treatment, embryos were washed in PBS and then mounted in agarose. PBS containing 25 mM EGTA was added in the dish just before starting the acquisition. For microtubule depolymerisation, colcemid (demecolcine, Sigma) was used at 20 μg/ml in 1% DMSO final both in agarose and in E3 medium. Control embryos were treated with 1% DMSO. For rescue experiments, the medium was washed out with 1% DMSO in E3 and the embryos were irradiated with UV for 30 seconds to inactivate the drug using a 5× objective and an HBO lamp. In order to prevent entry into S phase, embryos were treated with 75 μM aphidicolin (Sigma) and 10 mM hydroxyurea in E3 containing 1% DMSO final at least 4 hours before colcemid treatment.

We are grateful to Andreea Gruia for excellent animal care and the Advanced Light Microscopy Facility for imaging support. We thank Reinhard Koester and Martin Distel for kindly providing the lateral line GAL4-driver. We thank Jan Ellenberg, Marcus Heisler, Peter Lenart and Francesca Peri for critical reading of the manuscript and the Gilmour laboratory for stimulating discussions.