Fat2 polarizes Lar and Sema5c to coordinate the motility of collectively migrating epithelial cells

Epithelial cells migrate collectively during animal development, wound healing, intestinal turnover and cancer metastasis (Friedl and Gilmour, 2009; Rørth, 2012; Theveneau and Mayor, 2013; Shaw and Martin, 2016; Krndija et al., 2019). To do so, they must polarize within the epithelial plane at both the individual and tissue scales. At the individual scale, cells polarize along a leading-trailing axis. Protrusion and adhesion formation are biased to the leading edges of cells, and contractility and adhesion removal to their trailing edges, much as in cells migrating solo (Sherrard et al., 2021; Stock and Pauli, 2021; Soans et al., 2022). At the tissue scale, cells throughout the epithelium are polarized, such that their leading edges preferentially point in the direction of migration, and trailing edges in the opposite direction, a form of planar cell polarity. At the intersection of these scales are the cell–cell interfaces that link the trailing edge of one cell to the leading edge of the cell behind. These leading-trailing interfaces can act as sites of biochemical or mechano-chemical signaling that polarize motility behaviors across the interface (Venhuizen and Zegers, 2017; Jain et al., 2020; Gupta and Yap, 2021; Roberto and Emery, 2022). However, we know little about how signaling proteins become organized along interfaces to accomplish this feat.

The rotational migration of the follicle cells in Drosophila melanogaster has proven to be a fruitful system for identifying signaling mechanisms that coordinate epithelial cell movements. Follicle cells are somatic cells of the egg chamber, the multicellular structure within the ovary that gives rise to an egg. They form a continuous monolayer epithelium around a central cluster of germ cells, and they are surrounded in turn by a basement membrane extracellular matrix that encapsulates the entire egg chamber. The apical surfaces of the follicle cells adhere to the germ cells, and their basal surfaces adhere to and crawl along the basement membrane (Haigo and Bilder, 2011; Cetera et al., 2014). Migration in this topologically closed configuration causes the entire egg chamber to rotate within the stationary basement membrane. This motion changes the structure of the basement membrane, ultimately helping give the egg its elongated shape (Gutzeit et al., 1991; Haigo and Bilder, 2011; Isabella and Horne-Badovinac, 2016; Crest et al., 2017). Follicle cell migration requires WAVE complex-dependent lamellipodia, which are polarized to the leading edge of each cell and planar-polarized across the epithelium (Gutzeit et al., 1991; Cetera et al., 2014) (Fig. 1A,B). Polarity emerges tissue-autonomously, without input from extrinsic directional cues. This simplifying feature allows us to more easily isolate the contribution of within-group coordination to collective migration. It likely also makes these cells particularly reliant on such coordination for movement.

Two biochemical signaling modules operate at leading-trailing interfaces, where they coordinate the migratory behaviors of neighboring follicle cells. The first module is composed of the atypical cadherin Fat2 (also known as Kugelei) and the receptor tyrosine phosphatase Leukocyte-antigen-related-like (Lar). Fat2 is enriched along the trailing edge of each cell, where it acts in trans to concentrate Lar and the WAVE complex across the cell–cell interface, at the leading edge of the cell behind (Viktorinová and Dahmann, 2013; Barlan et al., 2017; Williams et al., 2022) (Fig. 1C,D). Lar also contributes to WAVE complex localization, but not as strongly as Fat2, implying the existence of additional unidentified Fat2 effectors (Squarr et al., 2016; Barlan et al., 2017). Together, these proteins restrict cell protrusive activity to a single leading-edge domain and orient the protrusions from all the cells in a uniform direction across the tissue (Williams et al., 2022). The second module is composed of a transmembrane semaphorin (ligand) and plexin (receptor) pair, Semaphorin 5c (Sema5c) and Plexin A (PlexA), which are enriched at leading and trailing edges respectively (Stedden et al., 2019) (Fig. 1C,D). In other contexts, semaphorin–plexin signaling can lower integrin-based adhesion and/or inhibit protrusivity on the plexin-containing cell side (Hung and Terman, 2011; Alto and Terman, 2017; Verhagen and Pasterkamp, 2020). Similarly, overexpression of Sema5c in one follicle cell reduces the protrusivity of its neighbors in a PlexA-dependent manner (Stedden et al., 2019). This led to the model that Sema5c signals through PlexA to maintain a non-protrusive state at the trailing edges of cells.

Despite their distinct depletion and overexpression phenotypes, several lines of evidence suggest that the Fat2–Lar and Sema5c–PlexA modules function within one interface-polarizing signaling system. A series of pairwise comparisons show that Fat2, Lar and Sema5c all colocalize with the WAVE complex in interface-spanning puncta that sit at the tips of filopodia within a broader lamellipodium (Squarr et al., 2016; Barlan et al., 2017; Stedden et al., 2019; Williams et al., 2022) (Fig. 1E,F). For reasons that are not yet clear, PlexA only rarely colocalizes with the other proteins (Stedden et al., 2019). Loss of Lar also reduces the enrichment of Sema5c at leading edges (Stedden et al., 2019), implying functional interaction between the two modules in addition to their shared spatial organization. In this study, we investigated the hierarchy of interactions between Fat2, Lar, and Sema5c by which they form interface-spanning puncta, and asked how the three proteins work together to promote collective migration.

We find that Fat2 forms the core of both the Lar and Sema5c-containing signaling modules, concentrating Sema5c at leading edges in trans as it was previously shown to do for Lar. Conversely, Lar and Sema5c play little or no role in the localization of Fat2. Using fluorescence recovery after photobleaching (FRAP) and acute inhibition experiments, we show that Fat2 resides more stably at trailing edges than do Lar or Sema5c at leading edges, and that Fat2 is likely continuously required to maintain the enrichment of Lar and Sema5c at leading edges in the face of their ongoing turnover. We further find that Lar and Sema5c act in parallel to promote collective migration. From these data, we propose that Fat2 acts as a central organizer of the follicle cell interface-polarizing signaling system, serving to couple and polarize the distributions of multiple effectors that together align the motility machinery of neighboring cells.

Fat2, Lar and Sema5c all colocalize in interface-spanning puncta, with Fat2 at trailing edges and Lar and Sema5c at leading edges (Viktorinová and Dahmann, 2013; Barlan et al., 2017; Stedden et al., 2019). Fat2 acts in trans to concentrate Lar at leading edges (Barlan et al., 2017), a result we confirmed in this study using a new endogenous 3×GFP tag on Lar (Fig. S1). To ask whether Fat2 has the same effect on Sema5c, we generated epithelia with mosaic expression of fat2-RNAi, which allowed us to analyze Sema5c–3×GFP levels along leading-trailing interfaces at the basal surfaces of epithelia in which tissue-wide planar polarity was preserved. Sema5c levels were reduced wherever a fat2-RNAi cell was present ahead of the interface, regardless of the genotype of the cell behind the interface (Fig. 2A,B), again indicating a local trans interaction. This effect was sufficiently strong to eliminate all enrichment of Sema5c at leading-trailing interfaces relative to side interfaces (Fig. S2A). Given that Lar also contributes to the enrichment of Sema5c at leading edges (Stedden et al., 2019), we asked whether Fat2 localizes Sema5c indirectly through Lar. We confirmed that Sema5c levels were indeed reduced at leading-trailing interfaces between control cells and cells with a null lar allele (lar^13.2, Fig. 2C,D). However, this reduction was not as great as at interfaces between fat2-RNAi cells (Fig. 2E). In addition, if Fat2 localized Sema5c through Lar, because Fat2 concentrates Lar and Sema5c in trans to leading edges, we would expect Lar to concentrate Sema5c in cis. However, Sema5c levels were only reduced at leading-trailing interfaces between two lar^13.2 cells, indicating that the localization relationship between Lar and Sema5c is more complex. We conclude that Fat2 acts in trans to concentrate Sema5c at leading edges at least partly independently of Lar (Fig. 2F). We also found that Fat2-induced localization of Sema5c is not mediated by the WAVE complex or protrusions (Fig. S2B,C).

We next asked whether Fat2 is continuously required for the enrichment of Lar and Sema5c at leading edges, or whether it becomes dispensable after interface polarity is established. Because Ca²⁺ ions are integral to the structure of cadherin extracellular domains (Nagar et al., 1996; Kim et al., 2011; Tsukasaki et al., 2014), Ca²⁺ removal should disrupt any Fat2 function that depends directly or indirectly on its extracellular domain. Treating egg chambers with the cell-impermeable Ca²⁺ chelator ethylene glycol-O,O′-bis(2-aminoethyl)-N,N,N′,N′-tetraacetic acid (EGTA, 20 mM) for 5 min reduced the levels of protrusion-associated F-actin (Fig. 3A), consistent with loss of Fat2 function (Squarr et al., 2016; Barlan et al., 2017) (compare to Fig. 2A, Fig. S1D). It did not, however, cause other obvious changes to cell morphology, and protrusions returned within 1 h after EGTA washout (Fig. 3A), suggesting that this method rapidly and reversibly inhibits Fat2 without generally disrupting tissue architecture. However, it is likely that other cadherins and extracellular Ca²⁺-dependent molecules are also being affected by EGTA treatment, which could contribute to observed phenotypes. Notably, the 5-min EGTA treatment substantially reduced Lar and Sema5c levels along leading-trailing interfaces (Lar fluorescence intensity by 50%, Sema5c by 25%; Fig. 3B,C). Fat2 levels were nearly unchanged after 5 min, but its distribution became even more punctate, and by 30 min of EGTA treatment its levels had substantially decreased (Fig. 3B–D). Thus, Fat2 appears to be continuously required to maintain the enrichment of Lar and Sema5c at leading edges, although more specific methods for acute inhibition of Fat2 will be required to confirm this conclusion.

To probe the limits of the capacity of Fat2 to position the leading-edge proteins, we used GrabFP-A^Int to create an ectopic Fat2 population (Harmansa et al., 2017) (Fig. 4A), and asked whether this could relocalize Lar. At present, we lack a functional Sema5c antibody, and so could not check for its relocalization. Fat2 and Lar normally colocalize in puncta at the basal surface and along tricellular junctions that span the apical-basal axis (Viktorinová et al., 2009; Bateman et al., 2001). Expression of GrabFP-A^Int caused Fat2–3×GFP to accumulate all around adherens junctions as well, and increased Fat2–3×GFP levels overall (Fig. 4B–F). The ectopic Fat2 had no effect on the distribution of Lar, which remained restricted to tricellular junctions in the adherens junction plane despite Fat2 being all around the cell perimeter (Fig. 4D–F). These data show that although Fat2 is absolutely required for enrichment of Lar at leading edges, it is limited in its ability to recruit Lar, implying that other proteins or cell features also help shape the distribution of Lar.

Given the importance of Fat2 for localizing Lar and Sema5c, we asked whether they play a reciprocal role in localizing Fat2. We previously used a lar mutant condition to show that Lar plays little or no role in the localization of Fat2. Here, we confirmed this finding using lar-RNAi, and then examined the effect of loss of Sema5c or both proteins together on the localization of Fat2. To this end, we measured the average level of Fat2–3×GFP along all cell–cell interfaces at the basal surface of epithelia lacking Lar (lar-RNAi), Sema5c (Sema5c^K175) or both proteins together (lar-RNAi, Sema5c^K175) and compared these values to control epithelia. Full basal cell perimeters were used instead of leading-trailing interfaces because Fat2 loses its planar polarization when migration is disrupted (Barlan et al., 2017), as occurs in some of these backgrounds (see Fig. 7). Fat2 levels were normal in epithelia lacking Lar and mildly reduced in epithelia lacking Sema5c (Fig. 5A–C). Importantly, Fat2 levels were not further reduced in epithelia lacking both proteins compared to those lacking Sema5c alone (Fig. 5A,B). So, whereas Fat2 plays a major role in localizing Lar and Sema5c, Lar and Sema5c play at most a minor role in localizing Fat2 (Fig. 2H).

We next investigated the dynamics of Fat2, Lar and Sema5c turnover along leading-trailing interfaces, predicting that as a more central organizer, Fat2 would be more stable than the other proteins. To test this, we used FRAP to measure the turnover rate of each protein along leading-trailing interfaces at the basal surfaces of follicle cells. Within 6 min after photobleaching, Fat2–3×GFP fluorescence had only recovered to 29±10% of its initial levels, whereas Lar–3×GFP and Sema5c–3×GFP fluorescence had recovered to 70±21% and 90±26%, respectively (mean±s.d.; Fig. 6A,B,D; Movie 1). During a longer recovery period of 28 min, Fat2–3×GFP fluorescence reached 59±25% of initial levels, still lower than the maximum recovery seen for Lar and Sema5c (Fig. 6C,E; Movie 2). These data show that Fat2 molecules reside at leading-trailing interfaces for tens of minutes, and Lar and Sema5c only minutes, consistent with the more central role for Fat2 in interface organization.

We then asked whether Fat2 stabilizes Lar and Sema5c at interfaces by performing the same FRAP experiment in epithelia lacking Fat2, predicting that the turnover of Lar and Sema5c would increase in its absence (Fig. S3A–F). We found that in fat2^N103-2 epithelia, the average time to half recovery of both Lar–3×GFP and Sema5c–3×GFP decreased (Fig. S3C,F). However, this decrease was not statistically significant for Lar–3×GFP, and in the case of Sema5c–3×GFP it was accompanied by a decrease in total recovery as well as an increase in the rate of recovery. Sema5c–3×GFP levels are lower than Lar–3×GFP under normal circumstances (Fig. S3G,H), and Lar–3×GFP and Sema5c–3×GFP levels at interfaces are reduced in the absence of Fat2 (see Fig. 2, Fig. S1), which might have limited our ability to resolve changes in their dynamics. Further work is needed to determine whether Fat2 stabilizes Lar and Sema5c at leading edges or concentrates them there through a different mechanism.

Fat2, Lar and Sema5c are all required for normal collective migration, but whereas loss of Fat2 prevents migration entirely, loss of Lar only slows migration, and loss of Sema5c both slows it and delays its onset (Viktorinová and Dahmann, 2013; Barlan et al., 2017; Stedden et al., 2019) (Fig. 7A,B). These milder migration phenotypes suggest that Lar and Sema5c might act in parallel downstream of Fat2 to regulate cell motility. If true, analysis of the cell-scale phenotypes caused by loss of Lar or Sema5c could help us understand the two ‘arms’ of the interface signaling system that Fat2 assembles. Fat2 promotes epithelial migration both by increasing cell protrusivity and by polarizing protrusions of cells in the direction of tissue movement (Squarr et al., 2016; Barlan et al., 2017;Williams et al., 2022), so we asked whether Lar or Sema5c play similar roles in protrusion formation and polarity. These roles have previously been studied using F-actin labeling (Squarr et al., 2016; Barlan et al., 2017; Stedden et al., 2019), but we recently found that this approach misses much of the protrusive activity in fat2^N103-2 epithelia (Williams et al., 2022). We therefore used live imaging and membrane labeling to re-examine the protrusion phenotypes caused by loss of Lar or Sema5c alone, and to build on this by determining the effect of losing both proteins together. We performed automated segmentation of protrusions to measure the average cell protrusivity of epithelia and the polarity of those protrusions, as first described in Williams et al. (2022).

First, we examined the role of Lar in protrusion formation and polarity. F-actin staining had suggested that cells lacking Lar are less protrusive than normal (Squarr et al., 2016; Barlan et al., 2017). However, using membrane labeling, we detected no reduction in protrusivity in lar^13.2/bola1 epithelia (Fig. 7C,D; Movie 3). This suggests that Lar increases F-actin enrichment within protrusions, but is not required for protrusion formation per se. Loss of Lar did, however, disrupt protrusion polarity. Whereas control epithelia had most protrusions pointed in the direction of migration, slowly migrating lar^13.2/bola1 epithelia had a sizeable minority pointed rearwards, and their protrusions were less aligned with migration overall (Fig. 7C,E; Fig. S4, Movie 3). We did not observe an increase in rearwards protrusions in slowly-migrating epithelia of other genotypes (see below; compare protrusion alignment of migratory lar^13.2/bola1 and migratory Sema5c^K175 epithelia in Fig. 7E), so it is unlikely that the increase in rearwards protrusions is an indirect effect of the slow migration of the cells, but rather a local (cell or interface-scale) effect of loss of Lar. These data show that Lar helps polarize protrusions across the tissue and suggest that the role of Fat2 in this process is due in part to its ability to localize Lar.

We then performed the same analysis for Sema5c. Sema5c^K175 epithelia initiate migration later in development than normal, and their migration is slow and difficult to preserve ex vivo (Stedden et al., 2019). Consistent with this, about half of the Sema5c^K175 epithelia were migrating detectably, albeit slowly, and several others had a degree of protrusion polarity that likely indicated they had been migrating prior to dissection (Fig. 7A,B; Fig. S4B). In agreement with previous findings based on F-actin staining (Stedden et al., 2019), we did not observe a difference in the membrane protrusivity of Sema5c^K175 epithelia. In the detectably migrating subset of epithelia, protrusion polarity also appeared normal (Fig. 7C–E; Fig. S4, Movie 3). The slow migration speeds of these epithelia therefore do not seem to be caused by protrusion defects, although it is possible that Sema5c regulates protrusions in ways we did not detect. Alternatively, Sema5c might primarily regulate other aspects of cell motility such as contractility or adhesion.

Finally, we asked whether Lar and Sema5c work in parallel to promote collective migration by expressing lar-RNAi in follicle cells of Sema5c^K175 egg chambers. These epithelia failed to migrate and had unpolarized protrusions in all cases, and a significant speed reduction compared to epithelia lacking Lar or one of the two conditions lacking Sema5c (Fig. 7A,B), which was indistinguishable from fat2^N103-2 epithelia (Fig. 7A-C,E; Fig. S4; Movie 3). We conclude that Lar and Sema5c act in parallel to regulate cell motility, likely through distinct mechanisms. These data further suggest that Fat2-mediated polarization of both Lar and Sema5c to leading edges is a major means by which it regulates the collective migration of follicle cells.

The follicle cells use biochemical signaling across their leading-trailing interfaces to polarize their migration machinery at interface, cell and tissue scales. Here, we have shown that Fat2 is a central organizer of this signaling system. Fat2 acts at the trailing edge of each cell to concentrate both Lar and Sema5c at the leading edge of the cell behind. By contrast, Lar and Sema5c play at most minor roles in the localization of Fat2. In this way, Fat2 coordinates the activities of two effector proteins with distinct functions, allowing them to work synergistically to promote highly persistent collective migration.

One defining feature of this Fat2-based signaling system is that most of the component proteins colocalize in interface-spanning puncta. Cadherins often self-organize into clusters (Truong Quang et al., 2013; Wu et al., 2015; Brasch et al., 2019; Stahley et al., 2021), making it likely that this punctate organization stems from Fat2. However, whether Fat2 concentrates Lar and Sema5c in the puncta through direct binding or through intermediary proteins is unknown. An ectopic Fat2 pool did not cause redistribution of Lar, suggesting that additional inputs also contribute to the localization of Lar. We also do not know how the receptor for Sema5c, PlexA, fits into this model. Like Fat2, PlexA is enriched at the trailing edges of cells and helps to localize Sema5c, and yet antibody staining indicates that PlexA only partially colocalizes with the Fat2-based puncta (Stedden et al., 2019). The biggest open question is how Fat2 becomes localized to the trailing edge, as this appears to be the key event that polarizes the entire signaling system. Mechanical feedback from collective migration itself is required for the polarization of Fat2 to trailing edges (Barlan et al., 2017), but the nature of this feedback, and whether Fat2 has a trans binding partner that further stabilizes its localization, will be important areas for future investigation.

This study includes the first measurements of the dynamics of planar signaling proteins in follicle cells, which is an important step towards understanding their polarization mechanism. Using FRAP, we found that Fat2 is a more stable resident of the leading-trailing interface-spanning puncta than are Lar or Sema5c, consistent with its more central role in maintaining these structures. Based on these FRAP data, as well as the rapid redistribution of Lar and Sema5c upon extracellular Ca²⁺ chelation, we hypothesize that Fat2 maintains the leading edge enrichment of Lar and Sema5c by slowing their turnover within the puncta, thereby concentrating them at the leading edge. However, the continuous requirement of Fat2 for the maintenance of polarization of Lar and Sema5c awaits confirmation with a more specific method of acute Fat2 inhibition, as extracellular Ca²⁺ chelation is a blunt tool, and it is possible that Fat2-independent effects, such as the disruption of another cadherin, contributed to the rapid localization changes of Lar and Sema5c. Further comparison of the dynamics of Lar and Sema5c with and without Fat2 will also be needed to determine whether Fat2 concentrates them through local stabilization (for example through direct or indirect binding) or by a different mechanism, such as increasing their rate of arrival at leading edges.

This work also sheds light on how Lar and Sema5c work together to promote collective migration. We have previously shown that loss of either protein alone impairs migration but does not stop it (Barlan et al., 2017; Stedden et al., 2019). By contrast, we now find that removing both proteins together fully blocks migration in a way that is indistinguishable from loss of Fat2. These data suggest that once Fat2 concentrates Lar and Sema5c to the leading edges of cells, they then act in parallel to promote collective migration, likely by polarizing distinct aspects of the migration machinery.

What aspects of the migration machinery do Lar and Sema5c each control? In the case of Lar, it seems to be part of the bridge between Fat2 and WAVE complex-dependent protrusions — both Fat2 and Lar increase protrusive F-actin enrichment at leading edges (Fat2 in trans and Lar in cis) (Squarr et al., 2016; Barlan et al., 2017), and both also help polarize protrusions in the direction of migration (Williams et al. (2022) and this study). In addition, Lar acts in trans to promote retraction of the trailing edge of the cell ahead (Barlan et al., 2017), but the mechanistic basis for this trans function and the degree to which it is separable from the cis function of Lar, remain undetermined. In the case of Sema5c, cell-scale loss-of-function phenotypes have proven more elusive, but enrichment of PlexA at trailing edges and the ability of overexpressed Sema5c to suppress protrusion in trans both point to the trailing edge as the likely site of regulation (Stedden et al., 2019).

By positioning both Lar and Sema5c, Fat2 integrates two features of interface signaling systems known to operate in other collectively migrating cell types but not yet seen together. The first feature is the use of a trailing edge-associated mechanical cue to orient protrusions in the cell behind (Das et al., 2015; Hayer et al., 2016). Mechanical localization of Fat2 to trailing edges polarizes protrusions in the following cell, in part by localizing Lar (Barlan et al., 2017). The second feature is the use of contact inhibition of locomotion (Carmona-Fontaine et al., 2017; Smeets et al., 2016; Zimmermann et al., 2016), which causes cells to polarize away from one another by suppressing protrusion and/or increasing contractility at the point of contact (Roycroft and Mayor, 2016; Stramer and Mayor, 2017). By localizing Lar and Sema5c to leading edges, Fat2 positions them to enforce trailing edge behavior at the contacting edge of the cell ahead. Therefore, Fat2 translates a mechanical cue into bi-directional signaling across leading-trailing interfaces to coordinate cell migratory behaviors for collective migration.

The sources and references for all Drosophila melanogaster stocks used in this study are listed in Table S1 and the genotypes of the lines corresponding to each figure panel are listed in Table S2. Newly generated plasmids and Drosophila strains are available upon request to the corresponding author. Drosophila were raised on a diet of cornmeal molasses agar food at 25°C. Experimental females were collected 0–3 days post-eclosion and aged in the presence of males with a diet of the same cornmeal molasses agar supplemented with finely powdered active dry yeast before dissection. In most experiments, they were aged for 2–3 days at 25°C. Higher temperatures were used to increase the expression of Gal4-driven transgenes in some cases, and longer aging times (up to 5 days) were used to obtain more FRT recombination events in some mosaic epithelia. The temperatures and yeasting times used in each experiment are listed in Table S3. Epithelia mosaic for RNAi expression were generated using the Flp-out system with Gal4 expressed under control of a heat-shock promoter. To heat shock, experimental females were aged while undergoing the following 12 h temperature cycle: 1 h 37°C, 1 h 25°C, 1 h 37°C, 9 h 25°C. After 5 cycles (2.5 days), they were then transferred to 29°C and given fresh yeast for an additional 12 h prior to dissection to increase RNAi expression strength and uniformity across clones.

Endogenous Lar was tagged C-terminally with three enhanced GFP proteins in tandem separated by short linker sequences (3×GFP) using CRISPR following the general approaches described in Gratz et al. (2013, 2014). The same 3×GFP sequence had been used to generate Fat2–3×GFP and Sema5c–3×GFP proteins in previous studies (Barlan et al., 2017; Stedden et al., 2019). The guide RNA target sequence 5′-GCTTCTGCTATCGCGCTGCACTGG-3′ was chosen with flyCRISPR Target Finder (Gratz et al., 2014). The underlined sequence was cloned into the pU6-BbsI-chiRNA plasmid (Gratz et al., 2013; Addgene #45946), the upstream G added for efficient U6-driven expression, and the bold sequence is the adjacent PAM motif. The homologous recombination donor plasmid contained homology arms ∼2 kb in length flanking the insertion target site, acquired by amplification of genomic DNA from the y1 Mnos-Cas9.PZH-2A w* (nanos-Cas9) background (Port et al., 2014). The 3×GFP sequence was amplified from the pDsRed-Fat2-3xGFP plasmid (Barlan et al., 2017). A linker with a sequence encoding the amino acids GSGGSTVPRARDPPVAT connects the Lar C-terminus with the N-terminus of 3×GFP. Homology arms, linker and 3×GFP DNA fragments were inserted into donor plasmid pDsRed-attP, which contains 3×P3-DsRed flanked by loxP sites for insertion screening and subsequent removal (Gratz et al., 2014). The linker-3×GFP insertion was made immediately before the Lar stop codon. Guide and homologous recombination plasmids were injected by Genetivision in the nanos-Cas9 background, providing a germline Cas9 source. F1 males were screened for 3xP3-DsRed, which is expressed in the eye, and then 3xP3-DsRed was excised by crossing to Cre-expressing flies (MKRS hsFLP/TM6b Cre). Successful insertion was confirmed with genomic DNA sequencing of the lar C-terminus and 3×GFP insertion region following 3xP3-DsRed excision. This insertion strategy leaves a loxP ‘scar’ downstream of the lar protein-coding region.

Ovaries were dissected into 500 µl of live imaging medium (Schneider's Drosophila medium with 15% fetal bovine serum and 200 µg/ml insulin) in a spot plate using one set of Dumont #55 forceps in the experimenters' dominant hand and 1 set of Dumont #5 forceps in the non-dominant hand. A more detailed description and images of this dissection protocol are available in Cetera et al. (2016). Ovarioles were removed from both the ovary and their individual muscle sheaths by pinching the germarium with forceps and pulling anteriorly. For fixed imaging, egg chambers older than stage 9 were removed prior to fixation. For live imaging, egg chambers older than the egg chamber to be imaged were removed from the ovariole strands by slicing through the adjoining stalk cells with a 27-gauge hypodermic needle. Removal of older egg chambers results in more compression of the younger egg chambers between the slide and the coverslip.

Immediately after dissection, ovarioles were transferred to a fresh well of live imaging medium using a p10 pipettor. For samples with membrane staining, CellMask Orange (Thermo Fisher Scientific, Waltham, MA, USA; 1:500) was added and samples incubated for 15 min, and then samples were washed in live imaging medium to remove excess dye before mounting. For samples treated with CK-666, following plasma membrane staining, ovarioles were instead transferred to live imaging medium with 750 µM CK-666 (Millipore Sigma, St Louis, MO, USA) and then mounted immediately after in the same CK-666-containing medium. Samples were mounted along with 51 µm beads on standard glass slides, and covered with 22×22 mm #1.5 coverslips. The beads support the coverslip, controlling the amount of tissue compression. Some compression enables imaging of the basal surface of a field of cells in a single plane, but too much interferes with the migration process. Coverslip edges were sealed with petroleum jelly to prevent evaporation during imaging. Samples were checked for damage using the membrane stain or other available fluorescent markers as indicators, and excluded if damage was present. Slides were imaged for 1 h or less.

Egg chambers were dissected into live imaging medium, and then half were transferred into spot plate wells with 250 µl of live imaging medium (the ‘control’ treatment), and the other half into 250 µl of live imaging medium containing 20 mM EGTA (Thermo Fisher Scientific). They were incubated for 5 or 30 min, depending on the experiment, with wells covered by a slide to slow evaporation, and then fixed and stained with phalloidin as detailed below.

After dissection, ovarioles were fixed at room temperature for 15 min in 4% EM-grade formaldehyde in PBT (phosphate-buffered saline with 0.1% Triton X-100 detergent), and then washed three times for 5 min each time in PBT. For antibody staining, egg chambers were incubated in primary antibody (mouse anti-Lar monoclonal, 1:200, or mouse anti-Dlg monoclonal, 1:20; see Table S1 for full details) overnight at 4°C, washed three times for 5 min each time in PBT, and then incubated in secondary antibody (donkey anti-mouse polyclonal, 1:200; Table S1) for 2–3 h at room temperature. For phalloidin staining, samples were incubated in TRITC–phalloidin (Millipore Sigma, 1:250) for 15 min at room temperature, or with 647 AlexaFluor–phalloidin (Thermo Fisher Scientific, 1:100) for 2–3 h at room temperature. For a more detailed phalloidin staining protocol, see Anderson et al. (2023). Phalloidin and secondary antibody incubations were performed concurrently where applicable. Samples were then washed three times for 5 min each time in PBT and mounted on a slide in a drop (∼40 µl) of SlowFade Diamond antifade (Invitrogen), covered in a 22×50 mm #1.5 coverslip sealed with nail polish, and stored at 4°C until imaged. This mounting strategy results in compression of stage 6–7 egg chambers, allowing imaging of the basal surfaces of many cells in a single plane.

All micrograph images collected are of egg chambers at either stage 6 or 7, as assessed by their size and by the size and shape of their oocyte in cross-section. Images were acquired using a Zeiss LSM800 upright laser scanning confocal microscope with either a 40×/1.3 NA EC 386 Plan-NEOFLUAR or 63×/1.4 NA Plan-APOCHROMAT oil immersion objective. The system was controlled with Zen 2.3 Blue acquisition software (Zeiss). Imaging was performed at room temperature. All images and measurements are of a single Z-slice of basal surfaces of the follicle cells unless otherwise noted. Exceptions are Fig. 4B and the bottom rows of Fig. 3A and Fig. S1A, which show cross-sections through the apical-basal axis; Fig. 4D, which show a near-apical plane (the plane of the adherens junctions); and the timelapse movies used to measure the migration speeds plotted in Fig. 7B on a white background, which were taken at a medial plane through the cells with respect to their apical-basal axis. Unless noted, imaging settings were held constant for all images of the same fluorescent protein shown in the same image panel and any corresponding plot panels, but settings affecting image brightness were optimized separately for each fluorophore when multiple are present in the same panel, or for data shown in separate panels. The exceptions are in Fig. 2A,B and C,D, which were collected with the same imaging settings to enable the comparison in 2E, and Fig. S3G,H, in which Fat2–3×GFP, Lar–3×GFP and Sema5c–3×GFP were imaged and displayed in the same manner to enable comparison of their levels.

Timelapse movies of 20 min in length were acquired of stage 6–7 egg chambers stained with CellMask membrane dye. These were taken at either the medial epithelial plane (used for migration speed measurement only, conditions with white background in Fig. 7B) or the basal plane (for migration speed and protrusion trait measurement, conditions with gray backgrounds). In either case, cells were segmented in each frame using the pretrained ‘cytoplasm’ model in Cellpose (Stringer et al., 2017). Subsequent analysis steps were performed in Python using scikit-image (https://scikit-image.org/) and scipy (https://scipy.org/) libraries (van der Walt et al., 2014; Virtanen et al., 2020). The gaps between segmented cells were closed with a watershed algorithm seeded by the initial segmented regions. Segmented cells were then tracked by linking regions of high overlap in consecutive frames. Errors in segmentation and tracking were corrected manually using napari (https://napari.org/). Migration speed was measured from a 22 µm-wide band of cells around the ‘equator’ of the egg chamber, half-way between the anterior and posterior poles, where linear cell migration speed is fastest in migratory epithelia. The displacement vector as calculated, and these vectors averaged to obtain a vector whose length and direction were used to determine the tissue migration speed and direction. See also ‘Code’ section below.

Protrusions were identified from timelapse movies of epithelia stained with Cell Mask membrane dye, which yield rich information about protrusion characteristics including their sizes and distributions, and often allow detection of more protrusive activity than is detectable by F-actin labeling. Automated protrusion segmentation was performed using a watershed-based approach, and the average lengths and orientations of these protrusions measured, as described in Williams et al. (2022) and briefly here. Protrusion segmentation and trait measurement were performed in Python using scikit-image and scipy libraries. Following cell segmentation, the region of high membrane fluorescence at the interface between each pair of neighboring cells was segmented. This region includes the cell–cell interface and any protrusions that extend across it from either of the neighboring cells. We then approximated the cell–cell interface position running through this region by drawing the shortest path through the region that connects its bounding vertices. This divides the region into two portions belonging to each of the neighboring cells, which we call ‘edges’.

To obtain a benchmark for the width of non-protrusive cell edges, we identified the average width (area divided by interface length) of edges from CK-666-treated epithelia, which are nearly non-protrusive. Edges from all conditions were considered protrusive if their width (corresponding with their average length of membrane extension from the interface) was greater than that of 98% of those of CK-666-treated epithelia. We then calculated the average cell protrusivity of each epithelium, defined as its ratio of protrusive to total edges.

To measure protrusion polarity, we took the protrusive edges (‘protrusions’), identified their ‘tip’ and ‘base’ as in Williams et al. (2022), and then assigned their orientation as the direction of the vector from base to tip. As a metric for protrusion alignment across an epithelium, we found the dot product of each pair of protrusions present in the same frame, and then took the mean of all dot products. For each protrusion pair, the dot product will be 1 if they have the same orientation, 0 if they are orthogonal to one another, and −1 if they have opposite orientations. Averaging these for an epithelium would yield an alignment score of 1 if all protrusions were perfectly aligned (high vectorial polarity), but could have a score of zero either if protrusions were randomly oriented (no polarity), or if half pointed in each direction along one axis (high axial polarity, no vectorial polarity). We report alignment of protrusions with one another, rather than with the direction of migration, because this is still meaningful in non-migratory epithelia (whether truly non-migratory or only non-migratory ex vivo). In migrating epithelia, protrusion–protrusion alignment and protrusion–migration alignment measurements were highly correlated. See also ‘Code’ section below.

Egg chambers expressing Fat2–3×GFP were stained with anti-Dlg antibody to label cell edges, stage 6–7 egg chambers were imaged, and the cells were segmented using the pretrained ‘cytoplasm’ model in Cellpose applied to the Dlg channel. Subsequent analysis steps were performed in Python using scikit-image and scipy libraries. Gaps between segmented cells were closed with a watershed algorithm seeded by the initial segmented regions, and segmentation errors were then manually corrected using napari. Mean fluorescence intensity was calculated at cell–cell interfaces (boundaries between segmented cells dilated by 10 pixels) and the entire in-focus basal surface. See also ‘Code’ section below.

Leading-trailing interface regions were annotated by hand based on phalloidin staining using Fiji (ImageJ; Schindelin et al., 2012, 2015) as follows: 10 pixel-wide segmented lines were drawn along leading-trailing interfaces and then the mean fluorescence intensity across those interfaces was calculated. In experiments with non-mosaic epithelia, at least ten leading-trailing interfaces were measured. For more information about measurement in mosaic epithelia, see the next section. Note that no correction was performed for background fluorescence, so the proportional relationship between fluorescence intensity measurements should not be interpreted as corresponding directly to the proportional relationship between protein levels.

Analysis of protein localization in genetically mosaic epithelia was used to disentangle cell- and interface-scale protein-localizing interactions from the tissue-wide effects of genotype on planar polarity and migration. It was also used to determine the cell autonomy of localizing interactions. Interpretations of the cell autonomy of effects on the localization of Lar or Sema5c were made based on the much greater enrichment of Lar and Sema5c at leading edges than at trailing edges in normal circumstances (Barlan et al., 2017; Stedden et al., 2019) (see Fig. 1C). In epithelia mosaic, for a loss-of-function mutation (generated using the Flp/FRT system), cells with the chromosome containing the wild-type allele were marked with a nucleus-enriched fluorophore, and ‘control’ cells include both homozygous wild-type and heterozygous cells. In epithelia mosaic for RNAi expression (generated using the Flp-out system), RNAi-expressing cells were marked with a nucleus-enriched fluorophore. F-actin staining with AlexaFluor 647–phalloidin was used to visually assess whether epithelia were planar-polarized and migratory (based on stress fiber alignment across the epithelium), their migration direction (based on the orientation of leading edge protrusions), and to check for regions of tissue damage that could otherwise be misinterpreted as genetic clones. Egg chambers were included in subsequent analysis if they were within stage 6–7, had planar-polarized actin stress fibers, and had both control and mutant or RNAi-expressing cells in view with at least three leading-trailing interfaces between cells of each genotype combination. For measurement of fluorescence levels at leading-trailing interfaces between pairs of cells of each genotype combination, mean fluorescence intensity at individual leading-trailing interfaces was measured as described in the previous section, with all leading-trailing interfaces at genotype boundaries, and a similar number of clone-internal ones, included. Plots show intensities rescaled per sample, such that the mean fluorescence intensity at leading-trailing interfaces between control cells is equal to 1. For comparison of fluorescence levels at leading-trailing versus side interfaces, the mean fluorescence intensity at leading-trailing and side interfaces between pairs of control cells and between pairs of mutant or RNAi-expressing cells was measured as described in the previous section except that tricellular junctions were not included in the interface regions. Plots show intensities rescaled such that the mean fluorescence intensity at side interfaces between control cells is equal to 1. Plots were generated using GraphPad Prism 9.

Plots of ‘unrolled’ cell perimeters were used to display the distributions of Fat2, Lar and GrabFP-AInt fluorescence around cell perimeters and the correspondence between those distributions. 10 pixel-wide segmented lines were drawn around the perimeter of cells in Fiji, and fluorescence intensities measured. Using Python, intensities were rescaled for ease of comparison and plotted. Specifically, Fat2–3×GFP and anti-Lar intensities were rescaled such that they had a range of 0 to 1 in the control dataset. GrabFP-A^Int was rescaled proportionally to Fat2–3×GFP.

Photobleaching was performed on samples mounted and imaged as described in previous sections using the timed bleaching tool in Zen Blue. Focusing at the basal surface plane, leading-trailing interface regions between two vertices, excluding those vertices, were selected for bleaching. Lar–3×GFP, fat2^N103-2 and Sema5c–3×GFP, fat2^N103-2 epithelia were non-migratory and lacked planar polarity, so any interfaces were used. Up to four interfaces were bleached per egg chamber, always from non-neighboring cells. Two frames were acquired before bleaching, bleaching was performed once, and then timelapse acquisition continued for a total of either 8 min (at 10 s/frame) or 30 min (at 30 s/frame). Bleached interfaces were excluded from analysis if the majority of the interface length was not bleached, the bleached region encompassed a cell vertex, or the bleached interface was not in view throughout the timelapse. Fluorescence intensity measurements were made using Fiji. Rectangular regions were drawn around the bleached interfaces and the same number of non-bleached interfaces, the positions of these regions were translated in each time point as needed to follow interface movement, and the mean intensity of each was measured. Subsequent analysis and plotting were performed using Python. The mean fluorescence per time point was calculated separately for bleached and unbleached regions, and the mean of the unbleached regions used to rescale the bleached regions to correct for the effects of progressive photobleaching and region selection inaccuracy. The bleach-corrected values were then rescaled so that the time point just before bleaching had a value of one, and the time point just after bleaching zero. Standard deviations were calculated for each time point from the mean intensities of individual bleached regions after they were rescaled in the same manner. For calculation of the time to half recovery, single exponential curves of the form I(t)=A(1−e^−u*t) were fit to data from individual interfaces, where I is fluorescence intensity, t is time, and the parameters A (the immobile fraction) and τ are fit. Time to half recovery (τ_1/2, ⁠) was then calculated for each interface. In each of the Sema5c–3×GFP and Lar–3×GFP fat2^N103-2 datasets there was one interface with a time to half recovery that was higher than the others by 3–4 orders of magnitude, and these were excluded.

For timelapse movies (but not analysis) of FRAP at cell–cell interfaces (Movies 1, 2), bleach correction was performed on the full tissue timelapse using the ‘simple ratio’ method in ImageJ prior to cropping and movie generation, and the cropped region was moved to follow the interface. For all movies, margins, scale bars and text labels were first added to TIF image stacks using ImageJ or Adobe Illustrator 2021, and then the files exported from ImageJ as uncompressed AVI files. These were encoded as 1080 p30 MP4 files with H.264 (×264) video encoder using HandBrake 1.4 (https://handbrake.fr/).

Two or more biological replicates were performed for each experiment, with each containing egg chambers pooled from multiple flies, and results confirmed to be qualitatively consistent between them. Visibly damaged egg chambers were excluded from all analyses. No randomization of treatment groups was performed, and experiments were not performed with blinding. Sample sizes were not predetermined using a statistical method. The number of egg chambers, interfaces and/or filopodia (n) analyzed for each experiment can be found in the associated figure or figure legend. Statistical tests performed and their significance can also be found in figures or figure legends. An alpha of 0.05 was used to determine significance in all cases, but we have also attempted to plot data distributions in ways that allow the reader to weigh their similarities and differences for themselves. Statistical testing was conducted using GraphPad Prism 9. Based on visual inspection, all datasets appeared normally distributed, and statistical tests assuming normalcy were used throughout. All t-tests were two-tailed. A one-way ANOVA was used when more than two conditions were compared. A repeated measures one-way ANOVA was used for analysis of genetic mosaic epithelia, in which multiple genetic conditions were present in the same tissue. Welch's corrections were performed and Dunnet's T3 multiple comparison tests used for datasets in which the variance did not appear consistent between conditions (indicated in figure legends). Otherwise, post-hoc Tukey's multiple comparison tests were used. Initial and post-hoc multiple comparisons testing was conducted on all data present in the corresponding plots with two exceptions: in Fig. 3C, only data including the same fluorophore were compared, and in Fig. 7B, data from conditions with gray and white background shading were collected and analyzed separately.

Code used in data analysis is described where applicable in the previous sections and is available at https://github.com/a9w/epithelial_migration_signaling.

We thank members of the Horne-Badovinac and Munro Labs, Michael Glotzer, Ellie Heckscher for discussion, Seth Donoughe for image analysis advice, and the University of Chicago DNA Sequencing Facility for sequencing data acquisition.

Image and numerical data and plasmid maps generated in this study are available at https://doi.org/10.6084/m9.figshare.23564223.

The peer review history is available online at https://journals.biologists.com/jcs/lookup/doi/10.1242/jcs.261173.reviewer-comments.pdf