Precise regulation of cell shape is vital for building functional tissues. Here, we study the mechanisms that lead to the formation of highly elongated anisotropic epithelial cells in the Drosophila epidermis. We demonstrate that this cell shape is the result of two counteracting mechanisms at the cell surface that regulate the degree of elongation: actomyosin, which inhibits cell elongation downstream of RhoA (Rho1 in Drosophila) and intercellular adhesion, modulated via clathrin-mediated endocytosis of E-cadherin (encoded by shotgun in flies), which promotes cell elongation downstream of the GTPase Arf1 (Arf79F in Drosophila). We show that these two mechanisms do not act independently but are interconnected, with RhoA signalling reducing Arf1 recruitment to the plasma membrane. Additionally, cell adhesion itself regulates both mechanisms – p120-catenin, a regulator of intercellular adhesion, promotes the activity of both Arf1 and RhoA. Altogether, we uncover a complex network of interactions between cell–cell adhesion, the endocytic machinery and the actomyosin cortex, and demonstrate how this network regulates cell shape in an epithelial tissue in vivo.
The morphogenesis of all tissues requires precise control over the shape of individual cells. In epithelia, which outlines all cavities and surfaces of animal bodies, a variety of cell shapes is observed. Cell shape is determined by mechanical properties, which define cell geometry based on intracellular and intercellular forces (Chalut and Paluch, 2016; Lecuit and Lenne, 2007). At the cell surface, mechanical properties are determined by an interplay of two factors: cortical actin and intercellular adhesion (Lecuit and Lenne, 2007; Winklbauer, 2015).
The first factor, cortical actin, is a meshwork of actin filaments crosslinked by specific crosslinking proteins and myosin motors at the cell surface (Chugh and Paluch, 2018). Cortical tension is predominantly generated by the activity of non-muscle Myosin II (MyoII) motors, which act to minimise the contact area between cells by pulling on actin filaments, although the architecture of these filaments also contributes to tension regulation (Blankenship et al., 2006; Chugh et al., 2017; Clark et al., 2014). One of the best documented regulators of cortical contractility is the GTPase RhoA (Rho1 in Drosophila; Spiering and Hodgson, 2011). Its key effector is the enzyme Rho kinase (Rok), which is recruited to membranes by the activated form of RhoA, where it phosphorylates myosin light chain, leading to activation of MyoII and an increase in actin contractility (Amano et al., 2010; Kawano et al., 1999; Kureishi et al., 1997; Leung et al., 1995).
The second factor, intercellular adhesion, is the property of one cell binding to its neighbours using specialised proteins on its surface. In epithelia, this is mediated by adherens junctions (AJs), with E-cadherin (E-cad; encoded by shotgun in Drosophila) being the principal component. This transmembrane protein binds to E-cad molecules on adjacent cells (Takeichi, 1977; van Roy and Berx, 2008). Intercellular adhesion often opposes cortical tension by increasing the contact surface between cells (De Vries et al., 2004; Lecuit and Lenne, 2007), and its strength is proportional to both the levels and dynamics of E-cad at the cell surface (Foty and Steinberg, 2005; Troyanovsky et al., 2006). The latter largely relies on the processes of endocytosis and recycling, which constantly remodel AJs (Kowalczyk and Nanes, 2012).
The p120-catenin (p120ctn) protein family are the key regulators of E-cad endocytosis in mammalian cells, through directly binding the juxtamembrane domain of E-cad (Daniel and Reynolds, 1995; Garrett et al., 2017; Ireton et al., 2002; Oas et al., 2013; Shibamoto et al., 1995; Yap et al., 1998; Yu et al., 2016). This family is represented by a single gene in invertebrates, such as Drosophila, whereas humans have seven members with different expression patterns and functional requirements (Carnahan et al., 2010; Gul et al., 2017; Hatzfeld, 2005). Most studies have focused on the founding family member, p120ctn (encoded by CTNND1 in mammals); however, other members, δ-catenin (CTNND2) and ARVCF, seem to have similar functions (Davis et al., 2003). In mammalian cells, p120ctn is required to maintain E-cad at the plasma membrane – uncoupling p120ctn from E-cad or reducing expression results in complete internalisation of E-cad (Davis et al., 2003; Ireton et al., 2002; Ishiyama et al., 2010), as the binding of p120ctn to E-cad conceals endocytosis-triggering motifs (Nanes et al., 2012; Reynolds, 2007). This model of p120ctn activity has recently been augmented, when it was found that p120ctn can also promote endocytosis of E-cad through interaction with Numb (Sato et al., 2011).
By contrast, in Drosophila and C. elegans, p120ctn was thought to play only a supporting role in adhesion, because genetic ablation failed to replicate the effects observed in mammalian systems (Myster et al., 2003; Pacquelet et al., 2003; Pettitt et al., 2003). This was thought to be due to the greater similarity of invertebrate p120ctn to mammalian δ-catenin, ablation of which is similarly viable in mice (Carnahan et al., 2010; Israely et al., 2004). However, δ-catenin expression is restricted to neural and neuroendocrine tissues (Ho et al., 2000), which is likely to explain the mildness of knockout phenotypes, whereas invertebrate p120ctn is broadly expressed in both epithelia and neurons (Myster et al., 2003), suggesting potential functional similarity with mammalian p120ctn, which shares the broad expression pattern (Davis et al., 2003). It has recently been reported that Drosophila p120ctn is required to stabilise E-cad in the pupal wing (Iyer et al., 2019) and promotes the endocytosis and recycling of E-cad in the embryo and larval wing discs (Bulgakova and Brown, 2016). This indicates an evolutionary conservation of p120ctn function where, depending upon the context, p120ctn either inhibits or promotes E-cad endocytosis.
Another protein family regulating E-cad endocytosis is the Arf GTPases, which recruit coat proteins to facilitate intracellular trafficking (Donaldson and Jackson, 2011). The first family member, Arf1 (also known as Arf79F in Drosophila), is classically viewed as Golgi resident and responsible for anterograde transport from the Golgi to the plasma membrane (Donaldson and Jackson, 2011; McMahon and Boucrot, 2011). Recently, however, Arf1 was detected at the plasma membrane, where it co-operates with Arf6-dependent endocytosis (Humphreys et al., 2013; Padovani et al., 2014). In Drosophila, Arf1 is required for endocytosis in the early syncytial embryo (Humphreys et al., 2012; Lee and Harris, 2013; Rodrigues et al., 2016), and interacts with E-cad and another component of AJs, Par-3 (Bazooka in Drosophila) (Shao et al., 2010; Toret et al., 2014).
Cortical actin and intercellular adhesion do not exist in isolation but are intimately interconnected. Intracellularly E-cad interacts with other catenin proteins: E-cad binds β-catenin (encoded by armadillo), with α-catenin binding β-catenin and actin, thus directly linking E-cad to cortical actin (Ozawa et al., 1990; Shapiro and Weis, 2009). In addition to direct linkage, cortical actin and intercellular adhesion share common regulators. In mammalian epithelial cells, RhoA localises to AJs, where E-cad complexes create local zones of active RhoA by recruiting the Ect2 guanine-nucleotide-exchange factor (GEF) through α-catenin (Priya et al., 2013; Ratheesh et al., 2012). In these cells, p120ctn has a context-dependent role in RhoA regulation: it can inhibit RhoA directly or indirectly via the GTPase-activating protein (GAP) p190RhoGAP, direct its spatiotemporal activity, or activate it (Anastasiadis et al., 2000; Derksen and van de Ven, 2017; Lang et al., 2014; Taulet et al., 2009; Yu et al., 2016; Zebda et al., 2013). The role of p120ctn in the RhoA pathway in Drosophila is unclear (Fox et al., 2005; Magie et al., 2002). However, RhoA itself regulates E-cad and is required for establishment of E-cad-mediated adhesion (Braga et al., 1997). In Drosophila, RhoA promotes the regulated endocytosis of E-cad by Dia and AP2 (Levayer et al., 2011). Conversely, RhoA activity antagonises E-cad endocytosis in the early embryo (Lee and Harris, 2013), indicating a context-dependent role of RhoA in E-cad endocytosis. Furthermore, E-cad endocytosis can direct actin remodelling (Hunter et al., 2015).
Here, we investigated the interplay between cortical actin, specifically actomyosin, and E-cad-mediated adhesion in the elongated cells of the late embryonic epidermis in Drosophila. This cellular elongation is accompanied by a reciprocal anisotropy of cortical tension and E-cad adhesion. p120ctn is directly involved in shaping these cells by both influencing the endocytosis of E-cad and modulating cortical tension. We found that the mechanism of this p120ctn dual function is dependent on interactions with two GTPase pathways: RhoA, to increase cortical tension and inhibit endocytosis; and Arf1, to promote endocytosis. Finally, we show that the interplay between these two GTPase pathways downstream of p120ctn participates in shaping the morphology of cells in vivo.
Anisotropy of actomyosin and adhesion in the epidermis of stage 15 Drosophila embryos
The epidermal cells of stage 15 Drosophila embryos are elongated in the direction of the dorsal–ventral embryo axis and have an average aspect ratio of 6.1±1.3 (ratio between lengths of the long and short cell axes, mean±s.d.; Fig. 1A–E). Consistent with the elongated shape, these cells have two distinct types of cell border: long borders, which are orthogonal to the anterior–posterior axis of the embryo (AP borders), and short borders, which are orthogonal to the dorsal–ventral axis (DV borders; Fig. 1B–D′).
These cells exhibit an asymmetric distribution of intercellular adhesion components, specifically, the levels and dynamics of E-cad. In these differentiated cells, E-cad localises asymmetrically with a 1:2 AP:DV ratio (Fig. 1F; Table S1; Bulgakova et al., 2013) in a narrow continuous band of mature AJs (Adams et al., 1996; Tepass and Hartenstein, 1994). This asymmetry is due to an accumulation at the DV borders of a specific pool of E-cad, which is dynamic due to its endocytic trafficking (Bulgakova et al., 2013).
Further to the asymmetry of E-cad, cortical tension and MyoII are also anisotropic between the AP and DV borders. MyoII-YFP (Drosophila Zipper fused with YFP) is enriched at the AP borders (AP:DV of ∼2:1; Fig. 1G,H; Table S1), consistent with previous reports (Bulgakova et al., 2013; Simoes et al., 2010). As the accumulation of MyoII has been linked to cortical tension (Priya et al., 2015; Scarpa et al., 2018; Yu and Fernandez-Gonzalez, 2016), we compared tension between the AP and DV borders using microablation and measured the initial recoil, which is proportional to the tension (Liang et al., 2016; Mao et al., 2013). We used an E-cad tagged at its endogenous locus (E-cad–GFP, Huang et al., 2009) to label cells and quantify recoil (Fig. 1I–K; Movies 1, 2). The initial recoil was positive for the AP borders, showing that they are under tension, and negative for the DV borders, suggesting that they are under compression (Fig. 1J). Intriguingly, the initial reduction in the distance between the vertices of the DV borders was followed by expansion that exceeded the original border length (Fig. 1K).
Overall, the elongated shape of these epidermal cells coincides with inverse anisotropies in actomyosin-dependent tension and the levels and dynamics of adhesion complexes. Therefore, we investigated how the interplay between cortical tension and adhesion at the cell surface produces cell shape.
p120ctn influences cell shape, RhoA signalling and cortical tension
Members of the p120ctn family regulate both the actin cytoskeleton and cadherin trafficking and are thus good candidates to mediate their interplay. We investigated whether the function of p120ctn, with only one family member in Drosophila, affected the shape of cells. We overexpressed p120ctn under the control of a UAS promoter (UAS::p120ctn) in the posterior half of each embryonic segment using the engrailed::GAL4 (en::GAL4) driver while marking the cells using UAS::CD8–Cherry (Fig. 2A). To exclude potential differences between the compartments we used an external control: an engrailed compartment expressing two copies of UAS::CD8–Cherry to balance the Gal4:UAS ratio (Fig. 2A). The cells expressing UAS::p120ctn appeared distorted and had a reduced aspect ratio (P<0.0001; Fig. 2A,B). We complemented this analysis by using a p120ctn mutant (Fig. 2A; Bulgakova and Brown, 2016). Unexpectedly, the loss of p120ctn also reduced the cell aspect ratio in the engrailed compartment (P=0.001; Fig. 2C,D). This differs from the previous report (Bulgakova and Brown, 2016), which can be attributed to inherent differences between the engrailed-positive and -negative compartments (Fig. S1A). This difference was not accounted for in the previous report and is likely to mask the effect of p120ctn depletion. Therefore, altering the levels of p120ctn changes elongation of cells in the Drosophila epidermis.
Another regulator of cell shape is cortical actomyosin, which is influenced by RhoA signalling. Because p120ctn family members regulate RhoA signalling, we examined the interaction between p120ctn and RhoA in our system using its downstream target, non-muscle MyoII. p120ctn overexpression increased the amount of MyoII–YFP at the AP borders (P=0.006; Fig. 2E,F). Conversely, in the p120ctn mutant embryos MyoII–YFP levels were reduced at the AP borders (P=0.01; Fig. 2G,H). The levels of MyoII–YFP at the DV borders were not affected in either case (P=0.47 and P=0.45; Fig. 2F,H). We complemented these experiments by using a tagged kinase-dead variant of Rok (RokKD–Venus; Simoes et al., 2010), previously used as a readout of Rok localisation and activity (Bulgakova et al., 2013; Simoes et al., 2010), and a biosensor of RhoA activity – the RhoAGTP-binding domain of anillin (RBD–GFP; Munjal et al., 2015) to directly examine RhoA activation. Both RokKD–Venus and RBD–GFP had the same localisation and were affected in the same manner as MyoII–YFP (Fig. S1B–G). Overexpression of p120ctn led to elevated RokKD–Venus and RBD–GFP at the AP borders (P=0.0043 and P=0.08, respectively; Fig. S1B,C,F,G), whereas p120ctn loss abolished RokKD–Venus asymmetry due to a reduction at the AP borders (P=0.013; Fig. S1D,E). These findings support that these three readouts can be used interchangeably. As a control we also measured total F-actin using LifeAct–GFP (Riedl et al., 2008), and found that it localised symmetrically and was not affected by altered p120ctn levels (P>0.99, P=0.63; Fig. S1H,I). These results indicated that p120ctn activates RhoA signalling specifically at the AP borders in a dose-dependent manner in epidermal cells.
Because of this correlation between p120ctn levels and RhoA signalling on the AP cell borders, we measured the cortical tension at these borders in p120ctn overexpressing and mutant cells (Fig. 2I–K; Movies 1, 3 and 4). Overexpression of p120ctn increased both the total recoil distance and the initial recoil velocity (0.36 µm/s in comparison to 0.13 µm/s, P<0.0001; Fig. 2J,K). Conversely, in p120ctn mutant cells both the total recoil distance and initial velocity (0.07 µm/s) were decreased (P=0.022; Fig. 2J,K). This demonstrated that cortical tension correlates with p120ctn levels.
Overall, the changes in p120ctn levels altered the shape of epidermal cells, with the levels of p120ctn positively correlating with the activity of RhoA signalling and cortical tension at the AP borders. The DV borders displayed no change in the activity of RhoA signalling, indicating an anisotropic action of p120ctn.
p120ctn regulates the amount and dynamics of E-cad within adhesion sites
The other factor that contributes to cell shape is intercellular adhesion. p120ctn binds to the intracellular domain of E-cad, which regulates its endocytosis (Bulgakova and Brown, 2016; Iyer et al., 2019; Nanes et al., 2012; Reynolds, 2007; Sato et al., 2011). Using a ubiquitously expressed p120ctn tagged with GFP (Ubi::p120ctn–GFP), we determined that p120ctn colocalised with E-cad, mimicking its localisation with an enrichment at the DV borders (r=0.868, P<0.00001; Fig. 3A–C). Note that in this system, the antibody against the N-terminus of p120ctn fails to reproduce the localisation of the full-length GFP-tagged p120ctn (Fig. S2A). Given this colocalisation, we examined whether changes in E-cad levels were observed when the levels of p120ctn were altered. Overexpression of p120ctn increased E-cad–GFP levels at both AP and DV borders (P<0.0001 and P=0.023, respectively; Fig. 3D,E, see also Fig. 2A). Conversely, the loss of p120ctn resulted in an isotropic decrease in E-cad–GFP at AP and DV borders (P=0.008 and P=0.035, respectively; Fig. 3F,G).
Next, we overexpressed p120ctn–GFP (UAS::p120ctn–GFP) driven by en::GAL4 to compare p120ctn and E-cad localisation in the same cell. We detected an AP:DV ratio of 2:3 for both E-cad and p120ctn (Fig. 3H–J). The lower than usual AP:DV ratio of E-cad was similar to that observed when untagged p120ctn was overexpressed. As this E-cad distribution was identical to the distribution of p120ctn itself, we concluded that additional E-cad molecules are recruited as a protein complex with p120ctn.
The strength of cell adhesion and the number of adhesion complexes are regulated by endocytosis, and p120ctn has been shown to inhibit and promote E-cad endocytosis in mammalian and Drosophila cells (Bulgakova and Brown, 2016; Ireton et al., 2002; Nanes et al., 2012; Sato et al., 2011; Xiao et al., 2003). Therefore, we used fluorescence recovery after photobleaching (FRAP), which reveals the stable fraction of the protein (which does not exchange on the timescale of the experiment) and the mobile fraction. The E-cad–GFP mobile fraction was 70% for DV borders and 60% for AP borders in control cells, with 30% and 40% of protein being immobile, respectively (Fig. 3K–M; for best-fit data see Table S1).
E-cad–GFP was less dynamic at both border types in p120ctn overexpressing cells (Fig. 3K–M). The immobile fractions were approximately 60% and 50% for AP and DV borders, respectively, which resulted in a decrease of the mobile fraction to 40% and 50% (P=0.0023 and P<0.0001 relative to control, respectively; Table S1). This is similar to the changes in the dynamics of E-cad–GFP in p120ctn mutants, where an increase in the immobile E-cad–GFP fraction is also observed (Fig. S2B–D; Bulgakova and Brown, 2016). Therefore, although p120ctn levels correlate with the levels E-cad at the plasma membrane, both the overexpression and loss of p120ctn lead to an increase of the immobile E-cad fraction. Next, we sought to determine the mechanism of these changes in E-cad dynamics and how they contribute to cell shape.
p120ctn and RhoA regulate E-cad via clathrin-mediated endocytosis
To ascertain whether the increase in immobile E-cad–GFP in response to altered p120ctn levels was due to an impairment of endocytosis, we examined clathrin. We used Clathrin light chain (CLC) tagged with GFP (UAS::CLC–GFP, Loerke et al., 2005; Wu et al., 2001b), to monitor clathrin behaviour in the plane of AJs by performing FRAP (Fig. 4A–C). CLC–GFP incorporates functionally into clathrin-coated pits (Chang et al., 2002; Gaidarov et al., 1999; Kochubey et al., 2006), and its recovery in FRAP reflects endocytic dynamics: in HeLa cells, the immobile fraction of CLC–GFP increases upon downregulation of endocytosis (Wu et al., 2001a). CLC–GFP expressed using en::GAL4 was found in spots on the plasma membrane in the plane of AJs and in the cytoplasm (Fig. 4A), a localisation consistent with its function (Kaksonen and Roux, 2018). p120ctn overexpression reduced the mobile fraction of CLC–GFP by 30% (P<0.0001; Fig. 4C,D). A similar reduction in CLC–GFP mobile fraction by 25% was found in p120ctn mutants (P<0.0001; Fig. 4C,E). These reflect the changes observed in E-cad FRAP, suggesting that CLC–GFP recovery is a valid proxy for E-cad dynamics in these cells.
Because p120ctn overexpression resulted in anisotropic activation of RhoA signalling, we asked whether this alone was responsible for the changes in E-cad. We directly inhibited the RhoA pathway using dominant negative RhoA (RhoADN). RhoADN, expressed constitutively using en::GAL4, resulted in a complete loss of E-cad at the membrane by stage 15 of embryogenesis. Therefore, we acutely induced the expression of the RhoADN using ubiquitously expressed temperature sensitive GAL80ts and a temperature shift for 4 h at 12 h after egg laying (Pilauri et al., 2005). This acute downregulation reduced the amounts of E-cad–GFP at the AP borders (P=0.0078; Fig. 4F,G). Note that in this case E-cad asymmetry was reduced in control cells, likely due to the effects of the temperature shift. Due to the punctate pattern of E-cad–GFP potentially affecting the average intensity measurements, we also measured the total protein on the plasma membrane. We found reduced E-cad–GFP total protein content following the acute downregulation of RhoA (P<0.0001; Fig. S2H).
In a complementary experiment, we expressed a constitutively active RhoACA to elevate RhoA signalling directly (Fig. 4H). The expression of RhoACA increased the amounts of E-cad–GFP, specifically at the DV but not the AP borders (P<0.0001 and P=0.34, respectively; Fig. 4I), consistent with an ectopic activation of RhoA signalling at the DV borders. We then asked whether this increase could be explained by a larger immobile fraction of E-cad–GFP. Using FRAP, we observed a significant increase in the immobile fraction of E-cad–GFP at both the AP and DV borders (P=0.02 and P<0.0001, respectively; Fig. 4J–L). To further explore whether this increase in the immobile E-cad was linked to clathrin-mediated endocytosis, we measured the dynamics of clathrin in the plane of AJs using FRAP. Indeed, the mobile fraction of CLC–GFP was reduced by 40% (P<0.0001; Fig. S2E–G). This finding was consistent with the increase of both E-cad levels and immobile fraction in cells overexpressing the RhoA activator RhoGEF2, and the opposite effect upon RhoGEF2 downregulation (Bulgakova et al., 2013).
Overall, these data suggest that p120ctn leads to an activation of RhoA signalling at the AP borders, which increases both the total amount and immobile fraction of E-cad at these borders, most likely by preventing E-cad endocytosis.
Localisation of the GTPase Arf1 at the plasma membrane depends on p120ctn and RhoA, and promotes clathrin-mediated endocytosis
Elevated RhoA signalling resulted in an increase of immobile E-cad and inhibited clathrin-mediated endocytosis; however, immobile E-cad was also increased when RhoA signalling was downregulated in p120ctn mutants. Therefore, we sought to identify the molecules responsible for this E-cad immobilisation in p120ctn mutants. We examined whether the GTPase Arf1, which has been reported to interact with E-cad (Shao et al., 2010; Toret et al., 2014), acts downstream of p120ctn using a GFP-tagged variant of Arf1 (UAS::Arf1–GFP, Lee and Harris, 2013). UAS::Arf1–GFP has a reduced affinity for ArfGAPs and ArfGEFs, and a reduced nucleotide exchange rate (Jian et al., 2010), which allowed us to study Arf1 without hyperactivating the pathway.
UAS::Arf1–GFP localised to both the Golgi apparatus and plasma membrane (Fig. 5A; Fig. S3A), consistent with previous reports (Lee and Harris, 2013; Shao et al., 2010). The Golgi-resident Arf1 appeared in large puncta throughout the cytoplasm (Fig. 5A; Fig. S3A). The localisation of Arf1–GFP at the plasma membrane was most apparent at the centre of the AJs (Fig. S3D). In the control, this Arf1–GFP localisation to the plasma membrane was symmetrical between the AP and DV borders (P=0.36; Fig. 5B). The loss of p120ctn resulted in a uniform decrease in the amount of Arf1–GFP at both borders (P<0.0001 and P<0.0001; Fig. 5B), suggesting that p120ctn promotes Arf1 localisation.
Considering the known function of Arf1 in trafficking, we tested whether the reduction in Arf1 activity was responsible for the increase in immobile CLC–GFP in the p120ctn mutants. We expressed a constitutively active Arf1 (Arf1CA) in p120ctn mutant embryos and measured the FRAP of CLC–GFP (Fig. 5E–G). In this case, the mobile fraction of CLC–GFP was no longer different from the wild-type control (P=0.19; Fig. 5G; Table S1), demonstrating that the expression of Arf1CA rescues clathrin dynamics in the p120ctn mutant. This is consistent with Arf1 acting downstream of p120ctn, providing a link between the p120ctn–E-cad complex and the clathrin-mediated endocytic machinery. Furthermore, we measured the dynamics of Arf1–GFP itself using FRAP (Fig. 5H,I): Arf1–GFP at the plasma membrane recovered almost completely within 25 s (Fig. 5I), which indicated a highly dynamic exchange, consistent with the known activation kinetics (Rouhana et al., 2013).
Curiously, the overexpression of p120ctn also reduced Arf1–GFP at the AP borders (P=0.02; Fig. 5C,D). Although the reduction of Arf1 at the DV borders was not significant (P=0.37), its distribution remained uniform (P=0.18). Because GTPases often regulate each other (Baschieri and Farhan, 2012; Singh et al., 2017), we asked whether this reduction was a consequence of p120ctn elevating RhoA signalling. We measured the membrane levels of Arf1–GFP in cells expressing RNAi against RhoGEF2 (RhoGEF2-RNAi; Fig. 6A), which reduces Rok amounts specifically at the AP borders (Bulgakova et al., 2013). The downregulation of RhoGEF2 resulted in an increase in the amount of Arf1–GFP at both the AP and DV borders (P=0.022 and P=0.049, respectively; Fig. 6B), demonstrating that RhoA signalling negatively regulated Arf1 localisation to the plasma membrane.
To further test whether the reduction in Arf1–GFP following the overexpression of p120ctn was due to elevated RhoA signalling, we simultaneously expressed UAS::p120ctn with RhoGEF2-RNAi (Fig. 6C,D). Indeed, we found an increase in Arf1–GFP at the plasma membrane at both borders in this case (AP, P=0.0014; DV, P=0.0031; Fig. 6D). To complement these experiments we hyperactivated RhoA signalling using RhoACA. We detected a reduction of Arf1–GFP localisation at the DV but not the AP cell border (AP, P=0.86; DV, P=0.025; Fig. 6E,F), providing further evidence for the negative action of RhoA signalling on the recruitment of Arf1–GFP to the plasma membrane. It was surprising that RhoGEF2-RNAi led to a uniform elevation of Arf1–GFP, because it affects Rok only at the AP borders (Bulgakova et al., 2013). Indeed, Arf1 was uniformly localised at the plasma membrane in all cases. We inferred that the effect of RhoGEF2-RNAi on Arf1 at the DV borders was indirect – reduced RhoA signalling results in elevated recruitment of Arf1 at the AP borders, followed by rapid redistribution around the cell periphery and an overall elevation of Arf1–GFP at the cell surface.
Finally, to ask whether Arf1 has any action on RhoA and actomyosin, we measured the membrane localisation of MyoII–YFP upon the upregulation of Arf1 signalling using Arf1CA (Fig. S3B). MyoII–YFP localisation was indistinguishable between control and Arf1CA-expressing cells (Fig. S3C), suggesting that RhoA signalling in the embryonic epidermis is independent of Arf1. Overall, we conclude that Arf1 was reduced at the plasma membrane in p120ctn overexpressing cells due to the elevation of RhoA signalling. The reduction of Arf1 upon loss of p120ctn appeared independent of RhoA, and we suggest that it is caused by reduction of Arf1 recruitment and/or activation by p120ctn.
Adhesion dynamics regulate cell shape
So far, we demonstrated that p120ctn regulates actomyosin via RhoA signalling, and E-cad dynamics via both RhoA and Arf1. Next, we asked how this regulatory network contributes to the cell shape changes caused by altered p120ctn levels. We first examined how the inhibition and hyperactivation of Arf1 alone affected cell shape using a dominant-negative Arf1 (Arf1DN; Fig. 7A). Prolonged exposure to Arf1DN resulted in small rounded cells (Fig. S3E) and no surviving larvae, consistent with previous reports (Carvajal-Gonzalez et al., 2015), likely due to gross perturbation of post-Golgi protein transport causing cell death (Jian et al., 2010; Luchsinger et al., 2018). Therefore, we acutely induced the expression of Arf1DN using a temperature sensitive GAL80ts. The cells that expressed the Arf1DN had a reduced aspect ratio (P=0.003; Fig. 7A,B). This suggests that the reduction of endocytosis, and therefore increased immobile E-cad, is sufficient to reduce the cell aspect ratio. This conclusion is also supported by the reduced aspect ratio observed in other phenotypes in which the immobile fraction of E-cad was increased: namely the overexpression of p120ctn and the expression of RhoACA (P<0.0001; Fig. 7C,D). To test whether an increase in E-cad immobility alone is sufficient to reduce cell elongation, we used an alternative approach to inhibit E-cad endocytosis by overexpressing a dominant-negative form of the dynamin Shibire (ShiDN; Fig. 7E). Expression of ShiDN increases the immobile fraction of E-cad at the plasma membrane similarly to the loss of p120ctn (Bulgakova et al., 2013). Indeed, cells expressing ShiDN had a reduced aspect ratio in comparison to control cells (P=0.0002; Fig. 7E,F).
Finally, to investigate whether the perturbations in Arf1 and RhoA signalling caused by changes in p120ctn levels were responsible for the defects in cell morphology, we performed genetic rescue experiments. We expressed Arf1CA in p120ctn mutant embryos and found that the aspect ratio reduction was completely rescued (P=0.57 in comparison to control; Fig. 7G,H). To complement this, we downregulated RhoA signalling using RhoGEF2 RNAi in cells overexpressing p120ctn (Fig. 7I). The aspect ratio of these cells was again indistinguishable from the control (P=0.99; Fig. 7J). Therefore, the defects in cell elongation caused by the loss or elevation of p120ctn are rescued by compensating for the Arf1 and RhoA signalling pathways, respectively.
Overall, these results indicate that the dynamics of intercellular adhesion, mediated via endocytosis of E-cad, is an important factor in determining the elongation of epidermal cells.
Epithelial cells in vitro are usually isotropic, and the application of external stretching or compressing forces induces an initial anisotropy in their shape (elongation), which is quickly resolved through cell rearrangements and divisions, or tissue three-dimensional deformation (Duda et al., 2019; Latorre et al., 2018; Nestor-Bergmann et al., 2019). By contrast, there are multiple examples of highly anisotropic elongated cells in whole organisms, including mammalian skin and the epidermal cells used in this study (Fig. 8A; Aw et al., 2016; Box et al., 2019). These elongated shapes are necessary for correct tissue and organism morphogenesis (Box et al., 2019; McCleery et al., 2019).
In this study, we focused on the regulation of such elongated cell shape through the crosstalk between E-cad-mediated adhesion and cortical actomyosin. We provide in vivo evidence that p120ctn, a known regulator of E-cad dynamics and endocytosis (Bulgakova and Brown, 2016; Ireton et al., 2002; Nanes et al., 2012; Sato et al., 2011), mediates this crosstalk and regulates cell shape. It does so by promoting the activities of at least two small GTPases with opposing effects on E-cad dynamics: RhoA, which inhibits E-cad turnover; and Arf1, which promotes it (Fig. 8B). We show an interplay between these GTPases, with RhoA preventing the localisation of Arf1 to plasma membrane (Fig. 8B). As a result, both the depletion and overexpression of p120ctn lead to an increase in immobile E-cad at the cell surface: depletion is likely to do so through directly limiting Arf1 recruitment to the plasma membrane, whereas overexpression does so through elevating RhoA activity, which then inhibits Arf1. Finally, although p120ctn normally colocalises with E-cad and is at higher levels on the DV borders, we show that it regulates RhoA activity only at the AP borders, suggesting a tension-dependent function for p120ctn.
We demonstrate that the elongated cell shape is accompanied by anisotropic forces in the epidermis: while the AP borders are under tension, the DV borders are under compression. Most previous laser ablation experiments have reported positive velocities of the initial recoil when the vertices of the manipulated junction move apart (Sugimura et al., 2016). The only exception apart from our work, to our knowledge, is the case of anisotropic tissue stress in the amnioserosa, where a similar negative recoil was observed during germ-band retraction (McCleery et al., 2019). This germ-band retraction drives the elongation of epidermal cells (Gomez et al., 2016; McCleery et al., 2019). We suggest that the anisotropic pushing by the amnioserosa is likely to be the source of the observed compression, as also suggested previously (Hirano et al., 2009).
In the case of an isotropic tissue, the behaviour of cells can be represented as those of soap bubbles (Hayashi and Carthew, 2004), which tend to minimise their surfaces. This leads to a shift of vertices after ablation to new positions such that the distance between them increases. Such a situation is usually modelled as a Kelvin–Voigt fibre (Fernandez-Gonzalez et al., 2009). In the case of anisotropic cells, which are strongly elongated, the situation can be different depending on which type of border is ablated (Fig. 8C). In the case of AP borders, the distance between the vertices increases as cells obtain freedom to minimise their surfaces (Fig. 8C). However, in the case of DV borders, the released stress enables cells anterior and posterior to the border to expand due to the same surface minimisation mechanism, decreasing the distance between vertices, as observed in the experiments (Fig. 8C). It can be concluded that the Kelvin–Voigt fibre model is not directly applicable to such anisotropic cells. We speculate that this is a strongly dynamic process: swift reduction of the distance results in over-compression, leading to expansion. We suggest that this process is governed by surface tension of cell membranes and should resemble oscillations of soap bubbles (Saye, 2017). It is still unclear whether, after expansion, a new compression stage can start.
We have shown that the increase of immobile E-cad, and thus the inhibition of junctional remodelling, following expression of Arf1DN, ShiDN, RhoACA, and loss of p120ctn (Fig. 8D), prevents the elongation of cells in the Drosophila embryonic epidermis (Fig. 8B), suggesting a central role for E-cad dynamics in cell elongation. The contractility of actomyosin normally acts to reduce the surface length, whereas the levels of adhesion acts oppositely (Lecuit and Lenne, 2007). Indeed, the strongest reduction in cell elongation was observed upon expression of RhoACA, which increased the immobile fraction of E-cad and activated actomyosin contractility. At the same time, an increase in immobile E-cad seems sufficient to reduce the junctional length, as observed both in p120ctn mutant cells and those expressing Arf1DN. Therefore, we propose that cell elongation is the product of the counteraction of E-cad dynamics and cortical contractility.
Although the roles of RhoA and p120ctn in E-cad endocytosis are long-established (Davis et al., 2003; Ellis and Mellor, 2000), the Arf1-dependent recruitment of clathrin has only been shown to occur at the Golgi by recruiting the Adaptor 1 protein (Carvajal-Gonzalez et al., 2015). The function of Arf1 at the plasma membrane has been described in dynamin-independent endocytosis, which was presumed to be clathrin-independent (Kumari and Mayor, 2008). We have shown that the capacity of Arf1 to recruit clathrin is exploited by p120ctn to facilitate the endocytosis of E-cad. Whether this requires AP2, a plasma membrane clathrin adaptor, has yet to be determined. This finding provides a mechanistic insight into the pro-endocytic activity of p120ctn (Bulgakova and Brown, 2016; Sato et al., 2011) and elaborates the number of known p120ctn interactors. The activities of Arf1 and RhoA are antagonistic, which was also seen during the cellularisation of the early embryo (Lee and Harris, 2013).
In contrast to Arf1, the regulation of RhoA by p120ctn has been shown in many studies, although the exact effects and mechanisms seem to be context-dependent (Anastasiadis et al., 2000; Derksen and van de Ven, 2017; Lang et al., 2014; Taulet et al., 2009; Yu et al., 2016; Zebda et al., 2013). We demonstrate that in the epidermal cells of Drosophila embryos, p120ctn leads to activation of RhoA at the AP but not DV borders (Fig. 8B). In contrast, we show that p120ctn loss uniformly reduces recruitment of Arf1 to the plasma membrane. However, we also find that Arf1 is very dynamic and the changes at the DV borders, for example upon the downregulation of RhoGEF2, are likely to be an indirect consequence of the effect at the AP borders (Fig. 8B). It is therefore currently unclear whether p120ctn performs any direct activity at the DV borders, or promotes RhoA and Arf1 signalling only at the high-tension AP borders. Indeed, recent evidence suggests that p120ctn has mechanosensing properties (Iyer et al., 2019). Using laser ablation we demonstrated that p120ctn modulates tension at these AP borders, providing evidence for the role of p120ctn in mechanotransduction. It is yet to be determined whether p120ctn is directly sensing the tension through a conformational change similarly to other components of cell–cell adhesion such as α-catenin and vinculin (Bays and DeMali, 2017; Yao et al., 2014), or is regulated by another mechanotransducer.
Taken together, our findings demonstrate that cell elongation in a tissue is regulated through the opposing action of actomyosin contractility and adhesion endocytosis, which are, however, closely interconnected and regulate each other. Adhesion components modify actomyosin, while the latter alters adhesion endocytosis. Considering that all of the molecules studied are expressed in all epithelia across evolution, we speculate that this system is likely to be more broadly applicable in development and a general feature of cell biology.
MATERIALS AND METHODS
Fly stocks and genetics
Flies were raised on standard medium. The GAL4/UAS system (Brand and Perrimon, 1993) was used for all the specific spatial and temporal expression of transgenic and RNAi experiments. The GAL4 expressional driver used for all experiments was engrailed::GAL4 (en::GAL4, Bloomington number 30564). The following fly stocks were used in this study (Bloomington number included where applicable): E-cad–GFP (shg::E-cad–GFP, 60584), E-cad–Cherry (shg::E-cad–Cherry, 59014), UAS::CD8–Cherry (27392), UAS::CLC–GFP (7109), UAS::Arf1–GFP (gift from Tony Harris, University of Toronto, Canada), Zipper–YFP (MyoII–YFP, Kyoto Stock Center 115082), sqh::RokK116A–Venus (gift from Jennifer Zallen, Sloan Kettering Cancer Center, New York City, USA), UAS::Arf1-T31N (DN) and UAS::Arf1-Q71 L (CA) (Dottermusch-Heidel et al., 2012), UAS::Rho1-N19 (DN) (7328), UAS::Rho1-V14 (CA) (7330), UAS::RhoGEF2-RNAi (VDRC 110577), ubi::AniRBD–GFP (RBD–GFP, Munjal et al., 2015), tubulin::GAL80TS (7017), ubi::p120ctn–GFP (7190), UAS::p120ctn–GFP (7192), UAS::ShibireK44A (ShiDN, 5822), and UAS::LifeAct–GFP (57326). The p120ctn mutant embryos (p120ctn308/Δp120) were derived from crossing two stocks: homozygously viable p120ctn308 females (Myster et al., 2003) with homozygously lethal Df(2R)M41A8/CyO, twi::Gal4, UAS::GFP males (740). Thus, the p120ctn mutants examined lacked both maternal and zygotic contributions. In all experiments when necessary, additional copies of UAS::CD8–Cherry were used to balance the Gal4:UAS ratio across genotypes in each dataset.
Embryo collection and fixation
Embryos were collected at 25°C at 3-h time intervals and allowed to develop at 18°C for 21 h to reach the desired developmental stage, except for the acute induction experiments, where embryos were allowed to develop for 13 h at 18°C and shifted to 29°C for 4 h. Then embryos were dechorionated using 50% sodium hypochlorite (Invitrogen) in water for 4 min and extensively washed with deionised water prior to fixation. Fixation was performed with a 1:1 solution of 4% formaldehyde (Sigma) in PBS (phosphate buffered saline) and heptane (Sigma) for 20 min on an orbital shaker at room temperature. Embryos were then devitellinised in 1:1 solution of methanol and heptane for 20 s with vigorous agitation. Following subsequent methanol washes, the fixed embryo specimens were stored at −20°C in methanol until required.
Embryo live imaging
Embryos were collected and dechorionated as described above. Once washed with deionised water embryos were transferred to apple-juice agar segments upon microscope slides. Correct genotypes were selected under a fluorescence microscope (Leica) using a needle. Embryos were positioned and orientated in a row consisting of 6–10 embryos per genotype. Following this, embryos were transferred to prepared microscope slides with Scotch tape and coverslip bridge and embedded in Halocarbon oil 27 (Sigma). Embryos were left to aerate for 10 min prior to covering with a coverslip and imaging.
For laser ablation, following orientation and positioning the embryos were transferred to a 60 mm×22 mm coverslip which had been prepared by applying 10 µl of heptane glue along a strip in the middle of the coverslip orientated with the long axis. The coverslip was attached to a metal slide cassette (Zeiss), and the embryos were embedded in Halocarbon oil 27 before imaging.
The p120ctn full-length cDNA was obtained from Berkeley Drosophila Genome Project (BDGP), supplied in a pBSSK vector. This was subcloned into plasmid pUAS-k10.attB (DGRC) using standard restriction digestion with NotI and BamHI (New England Biolabs) followed by ligation with T4 DNA ligase (New England Biolabs) and transformation into DH5α competent E.coli cells (Thermo Fisher Scientific). Prior to injection, plasmids were test digested and sequenced (Core Genomic Facility, University of Sheffield). Plasmids were prepared for injection using standard miniprep extraction (Qiagen) and submitted for injection (Microinjection service, Department of Genetics, University of Cambridge) into the attP-86Fb stock (Bloomington stock 24749). Successful incorporation of the transgene was determined by screening for (w+) in the F1 progeny.
The embryos were washed three times in 1 ml of PBST (PBS with 0.05% Triton) with gentle rocking. Blocking of the embryos prior to staining was done in 300 µl of a 1% NGS (normal goat serum; New England Biolabs) in PBST for 1 h at room temperature with gentle rocking. For staining, the blocking solution was removed, 300 µl of the primary antibody [1:100 rat anti-E-cad (DCAD2, DSHB), 1:10 mouse anti-engrailed (4D9, DSHB), or 1:500 anti-Golgi (Golgin-245, Calbiochem)] in fresh blocking solution was added and the embryos were incubated overnight at 4°C with orbital rotation. Then, embryos were washed three times with 1 ml of PBST. A 300 µl 1:300 dilution of the secondary antibody (goat Cy3- or Cy5-conjugated IgG, Invitrogen) was added, and the embryos incubated either overnight at 4°C with orbital rotation or for 2 h at room temperature. Then embryos were washed three times with PBST, following which they were incubated with 50–70 µl of Vectashield (Vector Laboratories) and allowed to equilibrate for a period of 2 h before being mounted on microscope slides (Thermo).
Microscopy, data acquisition and FRAP
All experiments except for laser ablation were performed using an upright Olympus FV1000 confocal microscope with a 60×/1.40 NA oil immersion objective. ShiDN-expressing embryos and the corresponding control were imaged using 100×/1.40 NA UPlanSApo objective. All measurements were made on dorsolateral epidermal cells of embryos, which were near or just after completion of dorsal closure, corresponding to the end of stage 15 of embryogenesis. For fixed samples, 16-bit images were taken at a magnification of 0.051 µm/pixel (1024×1024 pixel xy-image) or 0.062 µm/pixel (ShiDN embryos and the corresponding control) with a pixel dwell of 4 µm/pixel. For each embryo, a z-axis sectional stack through the plane of the AJs was taken, which consisted of six sections with a 0.38 µm intersectional spacing. The images were saved in the Olympus binary image format for further processing.
For E-cad FRAP (adapted from Bulgakova et al., 2013) 16-bit images were taken at a magnification of 0.093 µm/pixel (320×320 pixel xy-image). In each embryo, several circular regions of 1 µm radius were photobleached at either DV or AP junctions resulting in one bleach event per cell. Photobleaching was performed with eight scans at 2 µs/pixel at 50–70% 488 nm laser power, resulting in the reduction of E-cad–GFP signal by 60–80%. A stack of six z-sections spaced by 0.38 µm was imaged just before photobleaching and immediately after photobleaching, and then subsequently at 20 s intervals for a total of 15 min.
Because the rate of endocytosis depends on external factors, such as temperature, controls were examined in parallel with experimental conditions in all experiments with CLC–GFP. For CLC–GFP FRAP, 16-bit images were taken at a magnification of 0.051 µm/pixel (256×256 pixel xy-image). In each embryo a single plane was selected in the centre of the AJ band using E-cad–Cherry fluorescence for positioning. An area encompassing a transverse region orthogonal to the axis of the engrailed-expressing cells was selected (140×60 pixels) and photobleached with a single scan at 2 µm/pixel using 100% 488 nm laser power, resulting in reduction of CLC–GFP signal by 70–80%. Images were taken using continuous acquisition at a frame rate of 2 s−1. Prior to bleaching, a sequence of ten images was taken, and a total of 400 frames, corresponding to 3.5 min, were taken.
Data processing and statistical analysis
Membrane intensity and cell shape
Images were processed in Fiji (https://fiji.sc) by generating average intensity projections of the channel required for quantification. Masks were created by processing background-subtracted maximum intensity projections using the Tissue Analyzer plugin in Fiji (Aigouy et al., 2016). Quantification of the membrane intensity at the AP and DV borders and cell elongation (aspect ratio) was done as described previously (Bulgakova and Brown, 2016), using a custom-built Matlab script that can be found at https://github.com/nbul/Intensity. In short, cells were identified as individual objects using the created masks, and their eccentricities were calculated. The aspect ratio was calculated from the eccentricity as: , where e is eccentricity. At the same time, the individual borders were identified as objects by subtracting a dilated mask of vertices from a dilated mask of cell outlines. The mean intensity and orientation of each border were calculated. The average border intensities (0–10° for the AP and 40–90° for the DV borders, relative to cell mean orientation) were calculated for each embryo and used as individual data points to compare datasets. The average cytoplasmic intensity was used for the background subtraction. In the case of quantification of Arf1–GFP membrane intensity, due to high Arf1–GFP presence inside cells both in the Golgi and cytoplasm, the mean intensity of embryonic areas not expressing the GFP-tagged transgene were used for background subtraction. Statistical analysis was performed in Graphpad Prism (https://www.graphpad.com/scientific-software/prism/). First, the data was cleaned using ROUT detection of outliers in Prism followed by testing for normal distribution (D'Agostino–Pearson normality test). Then, the significance for parametric data was tested by either a two-way ANOVA or two-tailed t-test with Welch's correction.
A dilated mask, outlining the cell perimeter and encompassing the plasma membrane signal, was used to measure the cumulative intensity of pixels.
Images were processed by using the grouped z-projector plugin in Fiji to generate average intensity projections for each timepoint. Following this, the bleached region of interest (ROI), control ROI and background intensity were manually measured for each timepoint. This data was processed in Microsoft Excel. First, the intensity of the bleached ROI at each time point was background subtracted and normalised as follows: In=(Fn−BGn)/(FCn−BGn), where Fn is the intensity of the bleached ROI at the timepoint n, FCn is the intensity of the control unbleached ROI of the same size at the plasma membrane at the timepoint n, and BGn is the background intensity, measured with the same size ROI in the cytoplasm at the timepoint n. Then, the relative recovery at each time point was calculated using the following formula: Rn=(In−I1)/(I0−I1), where In, I1 and I0 are normalised intensities of the bleached ROI at time point n, immediately after photobleaching and before photobleaching, respectively. These values were input into Prism and nonlinear regression analysis was performed to test for the best fit model and whether recoveries were significantly different between cell borders or genotypes. The recovery was fit to either a single exponential model in the form of , or to a bi-exponential model in the form of , where Fim is the size of the immobile fraction, Tfast and Tslow are the half-times, and A1 and A2 are amplitudes of the fast and slow components of the recovery. An F-test was used to choose the model and compare datasets.
Measurements of all intensities, i.e. the bleached ROI, control ROI and the background, and normalisation were done using a custom-built Matlab script (http://github.com/nbul/FRAP) using the same algorithm as described for E-cad FRAP. Curve fitting and statistical analysis was performed in Graphpad Prism using a nonlinear regression analysis as described for E-cad FRAP.
Nanoablation of single junctions was performed to provide a measure of junctional tension. The Drosophila embryonic epidermis is uniquely suited to the study of tissue compression in vivo. Though tissue compression can lead to events such as bending, buckling and folding in epithelia, epithelia can accommodate large and rapid compressive forces extremely well (Wyatt et al., 2020). Additionally, the ability of the embryonic epidermis to buckle is abolished by the presence of vitelline membrane, which limits any tissue movement outside of its z-plane. Embryos were imaged on a Zeiss LSM 880 microscope with an Airyscan detector, an 8-bit image at 0.053 µm/pixel (512×512 pixel xy-Image) resolution with a 63× objective (1.4 NA) at 5× zoom and 2× averaging was used. An illumination wavelength of 488 nm and 0.5% laser power were used. Images were captured with a 0.5 µm z-spacing. Narrow rectangular ROIs were drawn across the centre of single junctions and this region was ablated using a pulsed TiSa laser (Chameleon), tuned to 760 nm at 45% power. 0.95 µs/pixel dwell for a single z-stack iteration was used for pulse duration. Embryos were imaged continuously in a z-stack consisting of three z-slices. The distance between the vertices at the ends of ablated junctions was measured throughout the time course of the experiment and was expressed as a proportional change in distance relative to pre-ablation length. Statistical analysis was performed in Graphpad Prism using a two-tailed t-test with Welch's correction.
The authors thank Rob Tetley and Yanlan Mao for assistance with laser ablation experiments and Nadezhda M. Bulgakova for assistance with interpreting these experiments. The authors also wish to acknowledge the advice and guidance of Professor David Strutt, University of Sheffield, and his lab. The authors thank the technical staff of the Wolfson Light Microscopy Facility (LMF) and the Drosophila Facility at the University of Sheffield.
Conceptualization: N.A.B.; Methodology: J.G., N.A.B.; Software: N.A.B.; Validation: J.G.; Formal analysis: J.G., N.A.B.; Investigation: J.G., N.A.B.; Resources: N.A.B.; Data curation: J.G., N.A.B.; Writing - original draft: J.G., N.A.B.; Writing - review & editing: J.G., N.A.B.; Visualization: J.G., N.A.B.; Supervision: N.A.B.; Project administration: N.A.B.; Funding acquisition: N.A.B.
This work was supported by a grant from the UK Biotechnology and Biological Sciences Research Council (BB/P007503/1).
Peer review history
The peer review history is available online at https://jcs.biologists.org/lookup/doi/10.1242/jcs.242321.reviewer-comments.pdf
The authors declare no competing or financial interests.