A role for E-cadherin in ensuring cohesive migration of a heterogeneous population of non-epithelial cells

Collective cell migration is a key process underlying the morphogenesis of many organs as well as tumour invasion, which very often involves heterogeneous cell populations. Here we investigated how such populations can migrate cohesively in the Drosophila posterior midgut, comprised of epithelial and mesenchymal cells and show a novel role for the epithelial adhesion molecule E-cadherin (E-Cad) in mesenchymal cells. Despite a lack of junctions at the ultrastructure level, reducing E-Cad levels causes mesenchymal cells to detach from one another and from neighbouring epithelial cells; as a result, coordination between the two populations is lost. Moreover, Bazooka and recycling mechanisms are also required for E-Cad accumulation in mesenchymal cells. These results indicate an active role for E-Cad in mediating cohesive and ordered migration of non-epithelial cells, and discount the notion of E-Cad as just an epithelial feature that has to be switched off to enable migration of mesenchymal cells.

T he fundamental role of cell migration in development and homeostasis has been recognized for quite some time now. In particular, the fact that static cells acquire migratory capacity and become motile at very precise times and settings and, conversely, that inappropriate migration is associated with many pathologies. Very often, both in normal and disease conditions, collective migration involves heterogeneous cell populations with distinct mesenchymal and epithelial features. However, it is poorly understood how such populations can migrate cohesively.
We have addressed this issue by analysing endoderm migration in the process of gut formation. The endoderm of Drosophila is subdivided into three populations of cells before migration that differ morphologically and genetically ( Fig. 1a; refs 1,2). Throughout migration, principle midgut epithelial cells (PMECs) are apicobasally polarized, columnar and regular in shape, although without adherens junctions (Fig. 1a,b; refs 2,3). Previous work has shown that these cells are in direct contact with the neighbouring mesoderm, and require the mesoderm as a substrate for migration 2 . In contrast, interstitial cell precursors (ICPs) and adult midgut precursors (AMPs) have been shown to be mesenchymal throughout migration, as seen by their irregular morphology, lack of apicobasal polarity and extensive protrusive activity (Fig. 1b,c;, and this is particularly apparent in live movies where they are seen to extend and retract many protrusions, constantly making and breaking contacts with each other and the surrounding PMECs ( Fig. 1c; Supplementary Movie 1). Migration of midgut cells is highly coordinated; indeed, ICP and AMPs require PMECs for migration, as when PMECs are genetically ablated, ICPs and AMPs completely fail to migrate 1 . However, similar experiments showed that migration of PMECs does not rely on interactions with the ICPs 1 . While it is known that the coordination of PMEC behaviour with the mesoderm is mediated by integrins 5,6 , how ICP and AMP behaviour is coordinated with PMECs, is completely unknown.
We previously identified a set of specific GATA factors in Drosophila and mammals that are responsible for inducing epithelial cells towards a migratory endoderm behaviour 7 and, interestingly, these mesenchymal cells keep low levels of E-cadherin (E-Cad) protein throughout migration. While it is widely recognized the fundamental impact of E-Cad on cell behaviour, there is an important debate about its functional role. Considered for a long time to be a protein that assured the static behaviour of epithelial cells, with the repression of E-Cad long considered a necessary or even a sufficient step for epithelial cells to become migratory through an epithelial-to-mesenchymal transition (EMT). Indeed, the switch from E-Cad to an alternative cadherin protein has been claimed to be a critical event in such processes 8 , with cadherins such as N-Cad playing active roles in mesenchymal cell migration 9 . Thus, it is intriguing that mesenchymal-endodermal cells express E-Cad, but is unclear whether this is simply a remnant of an incomplete EMT or whether E-Cad may actually play an active role in the migrating endoderm.
Thus, we decided to investigate the putative role of E-Cad in the heterogeneous population of endodermal cells in the Drosophila midgut, where E-Cad is expressed not only in the polarized PMECs but also in the non-polarized ICP and AMPs throughout migration ( Fig. 1b; Supplementary Fig. 1). Our results show a functional requirement for E-Cad for the cohesive migration of mesenchymal cells. Furthermore, we have also found that the recycling mechanisms and polarity proteins thought to be specific to polarized epithelial cells can be shared by non-epithelial and non-apicobasally polarized cells. Thus, in this regard, our data challenge the paradigm of E-Cad as just an epithelial feature that has to be switched off to enable the migration of mesenchymal cells.

Results
E-Cad is required for the collective migration of midgut cells.
To examine a possible role for E-Cad in the behaviour of the mesenchymal cells during midgut migration, we focused our studies on the larger ICPs, which are clearly distinguishable by their big nuclei, and because they take up stereotypic positions during migration. It is not possible to analyse gut migration in the complete absence of E-Cad, due to the requirement for its maternal contribution during oogenesis and very early embryonic development, and because its cell-specific downregulation by RNA interference does not work at this early embryonic stage. Thus, we chose to focus our analysis on the strong allele shg G317 , which shows a phenotype stronger than the zygotic null mutant alleles, and an extensive genetic characterization suggests this is due to a dominant-negative effect on the maternal contribution 10 . This was reinforced by staining for E-Cad in shg G317 embryos, which showed that E-Cad expression is reduced to almost undetectable levels in the migrating posterior midgut (PMG) (Supplementary Fig. 2a). Furthermore, this allele was previously used to uncover a role for E-Cad in the repolarization of the endoderm after migration, and shown to give similar, but more defined phenotypes as seen in the genetic null allele shg IH (ref. 2).
In wild-type embryos, midgut cells are highly organized throughout migration with PMECs forming two pseudostratified layers that adhere to the mesoderm on their basal sides, and sandwich the ICPs apically, which are tightly packed together (Fig. 1d,f). In E-Cad mutants, this stereotypical organization of midgut cells is lost with gaps appearing between the ICPs (Fig. 1e). Furthermore, ICPs are found both intermingled with and on the basal side of the PMECs (Fig. 1g), indicating perturbed migration. To further investigate cell dynamics, we labelled midgut cells with a marker for actin 11 and followed their behaviour in vivo using two-photon microscopy. We found that in E-Cad mutants, the gaps and lack of cohesion both between the ICPs and between ICPs and PMECs increase and are more frequent as migration proceeds (Fig. 1h,i; Supplementary Movies 2 and 3), suggesting a continuous role for E-Cad in mediating attachment between migrating midgut cells.
E-Cad is required for cohesion of the migrating midgut cells. The E-Cad requirement for the precise alignment of midgut cells throughout migration suggested the presence of intercellular junctions. However, ultrastructural analysis showed only a few scattered spot adherens junctions (sAJs) in a very small number of cells ( Supplementary Fig. 3a), with the clear majority of midgut cells possessing no visible junctions (Fig. 1j), as was reported in a previous electron microscope study of cellular junctions in the early embryo 3 . Despite this lack of junctions, all midgut cell membranes in the wild type are extremely closely aligned, with ICPs protrusions tightly associated with the contours of neighbouring cells (Fig. 1j). Conversely, in E-Cad mutants, the cohesion between midgut cells is markedly altered; cell membranes are no longer aligned and large gaps appear between ICP cells (Fig. 1k), despite the fact that occasional sAJs can still be found ( Supplementary Fig. 3b). Its worth noting that the lack of cohesion between ICPs occurs with no detectable change in the cohesion between the PMECs. Together, these results show that E-Cad is required for the proper adhesion of migrating midgut cells in the absence of distinct morphological junctions. This is in contrast with the adhesive function of E-Cad in epithelial cells, where it is most often associated with zonula or sAJs 12,13 . cells using a combination of custom-built ImageJ macros that allowed the automated three-dimensional (3D) tracking and manual validation of each track (see Methods). These analyses revealed a remarkable coordination of PMG cell migration as the ratio of velocity between ICPs and PMECs within an embryo is always close to 1 ( Supplementary Fig. 4). In wild type, both PMECs and ICPs migrate very directionally, with a high degree of coordination and a velocity of 1.7 mm min À 1 (Fig. 2a Supplementary Fig. 4), and consequently the ICPs end up mispositioned (Fig. 2i,j). This result suggests that ICPs have a higher intrinsic migration capability than PMECs, and that E-Cad-mediated adhesion between ICPs and PMECs lowers ICP speed and helps coordinate their joint migration.
To ensure that the effects on midgut migration are due to a specific requirement for E-Cad within PMG cells, we expressed a dominant-negative E-Cad construct (DE-cad ex (ref. 14)), using the driver 48Y-Gal4, which targets it specifically to the PMG cells. This construct has previously been extensively characterized and shown in vivo to act as a strong dominant negative when expressed in the background of endogenous E-Cad, without any 'side effects' on Wingless signalling 14 . Similar to E-Cad mutants, we find that this produces gaps between ICP cells, indicating a loss of adhesion (Supplementary Finally, to ensure that changes in ICP migration are due to a specific requirement in these cells, and not just due to subtle undetected changes in the PMECs, we expressed dominantnegative E-Cad using the driver insc-Gal4, which targets it specifically to the ICPs (Fig. 2k). Similar to E-Cad mutants, the stereotypical organization of the ICPs is lost, and they are found both intermingled with and on the basal side of the PMECs (Fig. 2l, compare with E-Cad mutant in Fig. 1g). Thus, these results support a role for E-Cad specifically in the mesenchymal ICP cells, to coordinate their migration with both themselves and the surrounding PMECs.

High levels of E-Cad slows migration of both ICPs and PMECs.
On specification, midgut cells lower their levels of E-Cad as compared with the neighbouring epithelial cells 7 . To test the functional relevance of the levels of E-Cad, we overexpressed E-Cad throughout the midgut during migration (see Methods). Indeed, higher levels of E-Cad caused gross defects as the midgut loses its bilateral symmetry, and shows a significant delay in fusing with the anterior midgut ( Fig. 3a-d). A detailed in vivo analysis showed that higher levels of E-Cad reduce the speed of migration in both PMECs and ICPs, without affecting their coupling (Fig. 3e-h; Supplementary Fig. 4; Supplementary Movie 7). Increased E-Cad also causes a decrease in the coordination and directional efficiency of both cells types, which reflects the change of the migration path from straight to more jiggly (Fig. 3e-h). To ensure that the effects on ICP migration are not just a consequence of changes in PMEC behaviour, we overexpressed E-Cad using the driver insc-Gal4 and found it sufficient to slow down ICPs, which no longer migrate in parallel with the PMECs, and are found lagging behind (Fig. 3j). This further supports the notion that ICPs undergo an active migration and that levels of E-Cad-mediated adhesion ensure a balance between coupling and efficient migration of the two cell types.
E-Cad undergoes endocytic trafficking in PMG cells. In epithelial cells, E-Cad is dynamically endocytosed and recycled to modulate cell-cell contacts while maintaining overall adhesion [15][16][17] . We reasoned that a similar mechanism might also operate in the non-epithelial midgut cells, as E-Cad is very dynamic in both PMECs and ICPs, forming punctate spots on the membranes that are constantly appearing and disappearing 7 . To test this, we used a dominant-negative form of dynamin (Shi DN ), which is required for the scission of endocytic vesicles and recycling endosomes [18][19][20] . In fact, midgut expression of Shi DN caused a marked upregulation of membrane E-Cad in both PMECs and ICPs (Fig. 4b). Conversely, the levels of E-Cad are greatly lowered throughout the midgut on expression of a dominant-negative form of Rab11, which impairs the recycling of endocytosed proteins 21 (Fig. 4c). These results thus indicate that also in the non-epithelial midgut cells E-Cad is dynamically endocytosed and recycled.
While blocking endocytosis or recycling is likely to alter more than E-Cad trafficking, we also analysed the functional consequences of blocking endocytosis. We found a significant delay in PMG migration on expression of the Shi DN construct (Fig. 4e). Indeed, live-cell tracking revealed that blocking endocytosis causes a decrease in velocity of both cell types, and a loss of coordination and directional persistence, thus having a milder yet similar effect to overexpression of E-Cad (Fig. 4h,i; Supplementary Fig. 5a,b; Supplementary Movie 8). It is interesting to note that either overexpression of E-Cad or Shi DN causes a delay in not only ICP migration but also in the PMECs (Fig. 5b,c). Conversely, blocking recycling with the Rab11 DN construct either in the whole midgut or just ICPs alone caused gaps to appear between the cells (Figs 4g, 5e). However, while blocking endocytosis mimics the effects of E-Cad overexpression, live imaging revealed that blocking recycling on Rab11 DN expression affects PMG migration behaviour in a different manner to E-Cad mutants, as both ICP and PMEC migration is stalled by Rab11 DN expression (Fig. 4h,i; Supplementary Fig. 5c,d; Supplementary Movie 9).
From these phenotypes, it is clear that in addition for regulating E-Cad levels, another protein or proteins required for the active migration of both cell types are likely to be trafficked through recycling endosomes. It has previously been shown that the coordination of PMEC behaviour with the mesoderm is mediated by integrins 5,6 Fig. 6a-c). However, in both Shi DN -and Rab11 DN -expressing PMGs, this high concentration at the border is lost and bPS is found more uniform throughout the PMECs. Taken together, these data support a model whereby E-Cad and other proteins required for PMG migration, such as the bPS integrin subunit, are constantly endocytosed and recycled in PMG cells, thereby maintaining a balance of transient adhesive interactions both between the midgut cells, and between PMECs and the mesoderm, that act to   couple migration of these different cell types, without impeding it (Fig. 6i).
Baz is required to recruit E-Cad to the membrane of ICPs. In the Drosophila embryonic epithelium, one of the earliest cues for recruitment of E-Cad to the apical membrane is the Par3 homologue Bazooka (Baz) 22 , which is itself apicolaterally localized through interactions with other epithelial polarity cues 23,24 . It was thus intriguing to note that Baz is also expressed in non-epithelial cells of the midgut throughout migration and shows a strong overlap with E-Cad ( Fig. 6f; Supplementary Fig. 1). Furthermore, we found that overexpression of Baz-green fluorescent protein (GFP) in the midgut causes a significant increase in the levels of E-Cad in both PMECs and ICPs, which co-localizes precisely with exogenous Baz-GFP (Fig. 6d,g) and that levels of midgut E-Cad are greatly reduced in baz mutant embryos (Fig. 6e)    ARTICLE speeds of both cell types as measured in vivo ( Fig. 6b;  Supplementary Fig. 7; Supplementary Movie 10) and causes a specific delay in ICPs migration when triggered only in these cells (Fig. 5d). In contrast, when E-Cad is overexpressed, there is no effect on the levels of Baz (Fig. 6h), suggesting that while Baz is both necessary and sufficient to recruit E-Cad to the cell membrane, E-Cad does not affect Baz levels.
Interestingly, despite an almost complete loss of E-Cad in the PMGs of baz mutants (Fig. 6e), migration of the PMG is severely delayed, it fails to elongate along the visceral mesoderm, and remains very rounded throughout germband retraction ( Fig. 6e;  Supplementary Fig. 6f). Given the similarity of this phenotype to that of rab11 dominant negative, we decided to investigate the expression of bPS in baz mutant PMGs. We found that bPS no longer localizes to the PMG-visceral mesoderm interface, but is found all around the cell surface ( Supplementary Fig. 6f). Thus, while it is clear that correct levels of Baz are required in PMG cells to maintain precise levels of membrane E-Cad, it also plays an additional role in localizing bPS to the PMEC-visceral mesoderm border, which is in turn required for the normal migration of the PMECs.
Finally, due to the co-localization of Baz and E-Cad in PMG cells, we next investigated whether Baz is also trafficked through the endocytic pathway. We found that similar to E-Cad, Baz levels are upregulated when endocytosis is blocked in the PMG using ShiDN (Supplementary Fig. 6h), and, conversely, Baz levels are decreased in Rab11DN conditions ( Supplementary  Fig. 6I). These data further demonstrate the key role of endocytic trafficking in PMG migration, regulating the levels/ localization not only of E-Cad and the bPS integrin subunit but also of Baz (Fig. 6i).

Discussion
Previous studies on the role of E-Cad have focused on homogeneous populations of epithelial cells. E-Cad has long been considered to play a role in their static behaviour, mediating homophilic attachments through adherens junctions and its repression a necessary or even sufficient step for epithelial cells to become migratory in the EMT 8,25 . While mutants for E-Cad had been shown to affect some cell migration events 26,27 , it has only been very recently that the analysis of cell-specific requirements has shown E-Cad to play an active role in the migration of a certain type of epithelial cells, the Drosophila border cells, mediating adhesion, polarizing the cluster and orchestrating the directional migration of the group 28 . Our results indicate a further role for E-Cad as an adhesive factor in the migration of non-epithelial cells based on the following observations. First, in embryos mutant for E-Cad, there is a clear effect on the mesenchymal ICPs, without any obvious defect in the PMECs. Second, in these same embryos, the defects in ICPs are associated with the cell membranes between the ICPs no longer being closely aligned, with large gaps appearing between them. And third, the organization of non-epithelial cells during migration is perturbed when E-Cad function is disrupted just in these cells. Indeed, this could be quite a general role for E-Cad in heterogeneous populations of non-epithelial cells. In fact, the mesendoderm cells of zebrafish have been shown to actually upregulate E-Cad before migration and when E-Cad function in the whole embryo is compromised, they fail to migrate efficiently along the epiblast 27 . Our model of E-Cad recycling could account for this phenotype (Fig. 6h).
Our work also shows that the same mechanisms that act on the epithelial control of E-Cad membrane accumulation are also  active in non-epithelial cells. This notion is strengthened by our results showing that a 'polarity protein' such as Baz/Par3 is instructive in the recruitment of E-Cad in non-apicobasally polarized cells. Together, these results indicate that the adhesive role of E-Cad and the mechanisms regulating its recycling to the membrane could be shared by both epithelial and non-epithelial cells whether or not E-Cad is organized into junctional structures. Indeed, the adhesive function of E-Cad in these endodermal cells is not localized to a specific domain, as our ultrastructural analysis shows that in the absence of E-Cad, the tight membrane alignment of ICP cells is perturbed all around the cell circumference. We suggest that the epithelial specificity for junctions could instead be more related to 'extra-adhesive functions' of E-Cad such as cell polarization and intercellular cytoskeleton coupling as suggested in the review 12 . Finally, we would like to note that the morphogenesis of many endodermal organs, such as the thyroid, liver and pancreas, critically depends on interactions between endodermal and mesenchymal cells types 29,30 . Furthermore, almost all human tumours comprise heterogenous populations of epithelial and non-epithelial cells. Thus, our study suggests a potential role for E-Cad in coupling the behaviour of both cell types in many different developmental and pathogenic contexts.

Methods
Fly strains and genetics. Details for all genotypes and transgenes can be found in flybase (http://flybase.org) or in references listed here. Unless otherwise noted, stocks were obtained from the Bloomington Stock centre. Wild-type embryos were from yw stocks. The Gal4/UAS system 31 was used to drive the expression of transgenes. Transgenes were driven in the PMG using either the hkbGal4 or the 48Y-Gal4 drivers recombined with UAS-GMA-GFP, UAS-Stinger or UAS-Brainbow. To misexpress in the ICPs, an insc-Gal4 recombined with UAS-SrcGFP was used. Within the midgut, the insc-Gal4 driver is specific for the ICPs and AMPs, but it also drives expression in other tissues such as the nervous system.   Electron microscopy. Embryos were staged to 7.5-8 h, dechorionated, fixed with 25% glutaraldehyde in heptane, hand peeled in PBS and then prepared for transmission electron microscopy using standard techniques 33 . Ultrathin sections were analysed using a Tecnai SPIRIT transmission electron microscope.
Live imaging. Embryos were dechorinated using bleach and stage 10 embryos were manually picked from an agar plate using a Leica fluorescent dissecting microscope. The selected embryos were dorsally or laterally orientated and mounted on a coverslip coated with heptane glue to prevent drift during imaging. A drop of oil was placed on the embryos to maintain their survival. Embryos were imaged using an inverted Leica SP5 2-Photon microscope at 890-nm wavelength, using a 60 Â oil immersion lens. Multi-position time-lapse stacks of 20-25 mm, and z-depth of 1.5 was acquired at 2-min intervals over a period of 60 min. Six movies per condition were selected for analysis, and the starting point was defined as the initiation of germband retraction, which is unaffected in the different conditions.
Time-lapse preprocessing. The cells under study can exhibit a fast directed movement and are densely packed, making their tracking challenging. The overall movement can be broadly estimated and advantageously compensated for to lower the burden on the tracker. This was performed using a custom ImageJ macro that incrementally and uniformly shifts all the images from each frame to partially cancel out the estimated overall displacement. The shifts applied to the images are stored and accounted for at a later stage.
Nuclei tracking. The nuclei tracking was performed from the fluorescent signal of the nuclei using FIJI plugin Trackmate. This plugin implements a blob detector based on an adjusted 3D Laplacian of Gaussian filter followed by 3D local minima detection. The candidate nuclei are selected based on the intensity at the local minima and the distance to the closest local minima: if this distance is closer to the characteristic radius of the nuclei, only the strongest minima is kept. This last criterion vastly improved the results of the tracking and was implemented as a custom extension to Trackmate. To link the nuclei and build their tracks, the plugin then relies on a frame to frame constrained linear assignment.
Tracks visualization and correction. The nuclei tracks were exported from Trackmate to spreadsheets and the original overall movement was numerically restored by accordingly shifting the positions of the detected blobs. Each track was then manually checked. To this end, we developed a custom ImageJ macro allowing the overlay of tracks originating from a specific region of interest to the original movie. With this tool, we were able to follow a nucleus along a track in 3D, as the z-slice is automatically adjusted to the detected position. Tracks from ICPs and PMECs were selected according to the nuclear diameter: PMECs o3.5 mm and ICPs 45.5 mm. Only the valid tracks starting inside the specified regions and spanning the whole movie were kept, yielding an average of 15 tracks of each cell type per movie, corresponding to some 300 nuclei positions defined over time per embryo.
Tracks statistics. From the selected tracks, the cells instantaneous speeds were estimated from the nuclei frame to frame displacements. A metric of cohesion of the local movements was estimated as the correlation of a nucleus instantaneous speed direction (and magnitude) and that of the nucleus from the closest track (track distance evaluated at the first time point; Fig. 2e,f). From this information, the track average velocity, coordination and directional persistence were derived. Finally, the average and s.d. of these measurements were computed over all the tracks of all the movies from the same condition.