In epithelial tissues, cells tightly connect to each other through cell–cell junctions, but they also present the remarkable capacity of reorganizing themselves without compromising tissue integrity. Upon injury, simple epithelia efficiently resolve small lesions through the action of actin cytoskeleton contractile structures at the wound edge and cellular rearrangements. However, the underlying mechanisms and how they cooperate are still poorly understood. In this study, we combine live imaging and theoretical modeling to reveal a novel and indispensable role for occluding junctions (OJs) in this process. We demonstrate that OJ loss of function leads to defects in wound-closure dynamics: instead of contracting, wounds dramatically increase their area. OJ mutants exhibit phenotypes in cell shape, cellular rearrangements, and mechanical properties as well as in actin cytoskeleton dynamics at the wound edge. We propose that OJs are essential for wound closure by impacting on epithelial mechanics at the tissue level, which in turn is crucial for correct regulation of the cellular events occurring at the wound edge.
A unifying feature of metazoans is the evolution of a physical barrier between the external environment and the different compartments of the body. Epithelia are the masters in this fundamental function. Various types of intercellular junctions are responsible for establishing cell–cell contacts, polarity, and a permeability barrier in epithelial tissues. Adherens junctions (AJs) bind epithelial cells together and maintain the mechanical integrity of epithelia (Harris and Tepass, 2010). In turn, occluding junctions (OJs) control the paracellular transport of molecules across the epithelium and separate the apical and the basolateral membrane compartments. OJs are present in virtually all metazoans, although their structure varies in different organisms. Two main types of OJs have been characterized: tight junctions (TJs), present in vertebrates, and septate junctions (SJs), typical of invertebrates (Shen, 2012; Jonusaite et al., 2016). TJs and SJs show differences in their ultrastructure, molecular composition, and subcellular localization, but they share key components, namely proteins of the claudin, membrane guanylate cyclase kinase, and immunoglobulin families. Additionally, SJs present a similar ultrastructure and share many molecular components (e.g., Na+/K+ ATPase, neurexins, and band 4.1 protein) with vertebrate paranodal junctions, which connect myelinated glial cells to axons and segregate the nodes of Ranvier from adjacent myelinated domains (Harden et al., 2016).
Several studies using the fruit fly Drosophila melanogaster have identified 20 proteins localized at the SJ, all obligatory for epithelial development and conserved across Bilateria (Jonusaite et al., 2016). By the end of embryogenesis, SJ components form a large, stable, and functional molecular complex at the apical region of epithelial cells. Notably, the loss of function of a single component is enough to destabilize the complex and impair SJ function (Laval et al., 2008; Oshima and Fehon, 2011).
Although the establishment of the paracellular barrier is the canonical function of OJs, recent studies suggest they also play a role in epithelial remodeling and morphogenesis independently of their barrier function. However, the underlying mechanisms are still unknown (Wu and Beitel, 2004; Balda and Matter, 2016; Hall and Ward, 2016). In vertebrates, TJs can influence different cellular complexes such as the actin cytoskeleton and Rho-GTPase signaling. It is well known that actin can regulate mechanotransduction by interacting with AJs. However, the link between OJs and the cytoskeleton as well as the potential role of these junctions in epithelial mechanics and morphogenesis remain largely unexplored (Sluysmans et al., 2017). Interestingly, in an unbiased genetic screen in Drosophila, the SJ-associated protein Coiled (Cold) has been found to be required for wound healing in the embryonic epidermis (Campos et al., 2010). Extensive work from several groups has shown that in simple epithelia of both vertebrates and invertebrates, wound closure initiates by cytoskeleton polarization in cells at the wound margin, leading to the accumulation of F-actin and nonmuscle myosin 2 at the boundary that faces the epithelial hole (Kiehart et al., 2000; Wood et al., 2002; Garcia-Fernandez et al., 2009; Zulueta-Coarasa and Fernandez-Gonzalez, 2017). This results in the formation of a contractile supracellular cable that pulls cells together, acting as a “purse string” to quickly close the gap. In Drosophila, AJs are required for the proper assembly of this so-called actomyosin cable and hence for successful wound healing (Abreu-Blanco et al., 2012; Carvalho et al., 2014; Hunter et al., 2015; Matsubayashi et al., 2015), but the contribution of SJs remains unknown.
These observations have motivated us to investigate the role of SJs in epithelial repair. In this study, we show that 11 SJ-related proteins are indispensable for efficient wound closure in the Drosophila embryonic epidermis. Mutants for SJ components develop a dysfunctional actomyosin cable at the wound margin and display impaired wound-closure dynamics. In addition, these mutants show defects in cellular shapes and rearrangements as well as in tissue mechanical properties, suggesting that SJs regulate the cohesion and the mechanical responses of the epidermis at the tissue level. Altogether, this work uncovers a novel role for OJs in the regulation of tissue mechanics during epithelial morphogenesis and repair.
SJ core and associated proteins are required for wound healing
To investigate the role of SJs during wound healing, we determined whether mutants for the core components of SJs show a wound-closure phenotype using a previously described wounding assay (Campos et al., 2010). Briefly, we laser wounded the ventral epidermis of late-stage Drosophila embryos and scored them for open and closed wounds 16 h later (Fig. 1, A and B). Whereas only 3% of control embryos failed to close their wounds, mutants for seven SJ core components showed 80% of open wounds (Fig. 1 C). In addition, mutants for regulators of SJ assembly (Crooked) and localization (Gliotactin and Discs large [Dlg]; Oshima and Fehon, 2011) also showed a significant wound-closure phenotype (Fig. 1 C). These results show that SJs are essential for epithelial wound healing in Drosophila embryos.
SJ mutants show altered wound-closure dynamics
To understand the loss-of-function phenotype of SJ components during wound healing, we focused our analysis on the mutant kune-kune (kune), which showed a strong phenotype in our wounding assay (89% of unclosed wounds; Fig. 1 C). The protein Kune belongs to the claudin family, also essential components of vertebrate TJs (Nelson et al., 2010). We imaged live WT and kune mutant embryos expressing Cherry::Moesin to detect the F-actin cytoskeleton (Millard and Martin, 2008) and followed the formation and contraction of the actin cable at the wound edge upon laser wounding by spinning-disk confocal microscopy (Video 1). In the WT, this process occurred in a stereotypical manner as previously described (Fig. 2 A; Abreu-Blanco et al., 2012). In the first minutes after injury, cells at the wound edge began to accumulate actin at the boundary facing the hole, and ∼15–20 min later, an actin cable had assembled; the wound edge then contracted and eventually drove the hole to close. These wounds took on average 62 min to close completely (Fig. 2, D and F). In kune mutants, the dynamics of closure were dramatically impaired, and two phenotype strengths were noticed. kune embryos showing a mild phenotype (kune “mild” mutants) formed an actin cable at the wound edge and initiated contraction, but closure took much longer than in the WT (Fig. 2, B, D, and F); these embryos took on average 160 min to fully close their wounds. kune “strong” mutants exhibited a more severe phenotype: in the first 20 min after wounding, wound edge cells also assembled an actin cable and started contracting, slightly decreasing the area of the hole, but at a significantly slower rate than control embryos (closure rate of controls = 28 µm2/min; closure rate of kune strong mutants = 8 µm2/min). 20–40 min after injury, the wounds stopped decreasing size and gradually expanded, in some cases reaching 3× the initial wound size (Fig. 2, C and F); at the same time, the actin cable seemed to progressively decrease in intensity. Strikingly, 2–3 h after wounding, kune strong mutants restarted contracting and either eventually reached closure (average = 285 min; n = 3 out of 8 embryos) or failed to close the wound until the end of the time-lapse imaging (maximum 315 min after wounding; n = 5 out of 8 embryos; Fig. 2 D). Interestingly, the average initial wound area was larger in kune strong mutants than in controls and kune mild embryos (Fig. 2 E), suggesting that the observed wound-closure phenotypes are associated with defects in the epithelium before injury. To confirm whether there is a correlation between the wound size and the phenotype strength in kune mutants, we divided wounds in two groups according to their initial area: small (initial wound area = 468–1,100 µm2) and large (initial wound area = 1,101–1,820 µm2). Indeed, large wounds closed significantly later than small wounds in kune mutants (Fig. 2 G). In addition, small wounds were predominant in the mild phenotype cohort, whereas large wounds were more associated with the strong phenotype (Fig. 2 H). Similar results were obtained for the mutant sinuous (sinu), another SJ component of the claudin family, in terms of wound-closure dynamics (Fig. S1, A–D).
Altogether, these results show that SJ loss of function leads to dramatic defects in wound-closure dynamics. In particular, the wound-expansion phenotype observed in the strong mutants is remarkably different from the phenotype previously described in mutants for other wound-closure players such as AJs and actomyosin (Wood et al., 2002; Abreu-Blanco et al., 2012; Fernandez-Gonzalez and Zallen, 2013; Hunter et al., 2015; Matsubayashi et al., 2015).
SJ mutants show defective actomyosin dynamics
To determine why kune mutants do not efficiently close wounds, we first looked at the actomyosin cable that assembles at the wound edge. We quantified the intensity levels of our F-actin reporter at the wound edge in the first 30 min of wound closure, the time frame when the actin cable assembles. In the WT, F-actin levels gradually increased at the wound edge between 5 and 30 min post wounding (mpw), which is consistent with the formation of the actin cable and the beginning of wound contraction (Fig. 2 I). In contrast, in kune, F-actin levels were significantly decreased compared with the WT at 20 and 30 mpw (Fig. 2 I). These time points seem to coincide with the beginning of the expansion phenotype detected in kune strong mutants but slightly after the first defects in wound contraction (compare Fig. 2 F with Fig. 2 I). When measuring F-actin intensity in cells before wounding, we found no difference between controls and kune (Fig. 2 I), suggesting that Kune is required to form a functional actin cable but does not significantly influence F-actin dynamics in uninjured tissue.
Next, we investigated whether nonmuscle myosin-2 (hereafter called myosin), another component of the actomyosin cable required for its contractility and function (Abreu-Blanco et al., 2012; Fernandez-Gonzalez and Zallen, 2013), is also affected in kune mutants. We used the myosin heavy-chain endogenous reporter zipper::GFP (zip::GFP; Lye et al., 2014) and imaged wounded live embryos as described above. As seen for actin, myosin levels before wounding were similar between control and kune (Fig. 3 C). At 10 and 20 mpw, myosin accumulated at the wound edge in both controls and kune mutants, but at 30 mpw, the average myosin fluorescence intensity was significantly decreased in kune in comparison with controls (Fig. 3, A–C). As for F-actin, this decrease in myosin fluorescence in kune mutants seemed to occur approximately at the same time as when the wound started to expand, i.e., slightly later than when the wound contraction defects were evident (Fig. S2, A and B).
Actin-rich protrusions such as filopodia and lamellipodia are also required for proper wound-closure kinetics (Abreu-Blanco et al., 2012). Therefore, we investigated whether these protrusions were affected in kune mutants. We analyzed the area of protrusions in different time points after wounding, similar to what has been described previously (Abreu-Blanco et al., 2012; Zulueta-Coarasa et al., 2014), in both controls and kune mutants marked with our F-actin reporter (Fig. S2, C and D). We found no significant differences in protrusive activity between controls and kune mutants, suggesting that the kune wound-closure phenotype is not due to defects in these cellular structures. Altogether, we show that in the absence of functional SJs, cells at the wound edge develop impaired actomyosin dynamics shortly after the first wound-closure phenotypes appear.
Junctional dynamics at the wound edge are altered in
Next, we followed the dynamics of SJ components at the wound edge. We analyzed the localization of the transmembrane SJ proteins Neurexin-IV (Nrx-IV) and Neuroglian (Nrg), whose mammalian homologues are present at paranodal junctions (Banerjee et al., 2006). We used protein trap transgenic lines expressing the functional Nrx-IV and Nrg proteins fused to GFP under the control of their endogenous promoters (Morin et al., 2001; Buszczak et al., 2007; Oshima and Fehon, 2011) and combined them with Cherry::Moesin to compare their localization with the actin cable. As expected from previous research (Nelson et al., 2010), already before wounding, Nrx-IV–GFP and Nrg-GFP intensities were significantly decreased in kune mutants when compared with WT (Fig. 4, B, E, and G; and Fig. S3 G). Furthermore, whereas in controls, these proteins concentrated at the apical region of the membrane, in kune mutants, they spread along the apical–basal axis (Fig. 4, A′′, B′′, C′, D′′, E′′, F′, and I; and Fig. S3, A′–F and I), also corroborating previous research (Nelson et al., 2010). Notably, at 10 and 30 mpw, both in WT and kune, Nrx-IV–GFP and Nrg-GFP levels decreased significantly at the wound edge (Fig. 4, A, A′, B, B′, C, D, D′, E, E′, and F; Fig. S3, A–F; and Videos 2 and 3). To quantify this, we compared Nrx-IV–GFP levels at the wound edge to the levels before wounding (Fig. 4 H). At 10 mpw, Nrx-IV–GFP intensity exhibited a 0.6-fold decrease relative to levels before wounding both in control and kune embryos. At 30 mpw, Nrx-IV–GFP intensity decreased even further in controls (reaching a fold change of 0.35), whereas in kune, this value was significantly higher than in controls (0.54-fold change), suggesting that Nrx-IV dynamics at the wound edge is affected by Kune loss of function. Furthermore, in kune, some cells at the wound edge accumulated Nrx-IV–GFP spots that colocalized with the actin cable, whereas in controls, this was not evident (Fig. 4, A′, A′′, B′, B′′, C, C′′, D′, D′′, E′, E′′, F, and F′; and Videos 2 and 3). Similar results were obtained for Nrg-GFP (Fig. S3 H). These observations show that upon wounding, Nrx-IV–GFP and Nrg-GFP are down-regulated at the wound edge and that Kune loss of function affects the dynamics of these SJ proteins in this region.
To exclude the possibility that the ablation laser induces bleaching artifacts that lead to the observed down-regulation of these reporters at the wound edge, we stained wounded WT and kune embryos using an antibody that recognizes the SJ component Coracle and an F-actin marker. We found that similar to Nrx-IV–GFP and Nrg-GFP, Coracle was mostly excluded from the wound edge both in WT and kune, not colocalizing with the actin cable (Fig. S3, J and K).
We next investigated whether SJ loss of function affects cadherin-based AJs. AJs are also down-regulated at the wound edge, and their remodeling is required for proper actomyosin cable formation (Carvalho et al., 2014; Hunter et al., 2015; Matsubayashi et al., 2015). We imaged control and kune mutant embryos expressing the E-cadherin (E-cad) reporter ubi–E-cad::GFP (Oda and Tsukita, 1999). As previously reported, in controls, E-cad was down-regulated at the wound margin shortly upon laser wounding, only accumulating at the contact points between leading edge cells (Fig. 5, A, B, and E). In kune, E-cad intensity also decreased at the wound edge, but the fold change was significantly higher than in controls at 30 mpw (Fig. 5, C–E), suggesting that Kune loss of function interferes with E-cad remodeling at the wound edge. Interestingly, both the intensity levels and the apical localization of E-cad before wounding were unaltered in these mutants (Fig. 5 F), indicating that SJs do not affect AJ localization and polarity in the uninjured epithelium.
These results show that Kune loss of function leads to the mislocalization of both SJ and AJ components at the wound edge and that SJs are actively regulated at this region.
SJs are required both in wound edge cells and in cells not at the wound edge for efficient wound closure
We can imagine two scenarios to explain the influence of SJs on wound closure. One is that SJ loss of function impairs the adhesion between cells, thereby influencing the cohesion and mechanics of the epidermal tissue as a whole and hence its response to injury; in this scenario, the defects in actomyosin and junction dynamics would be a consequence of impaired tissue mechanics. Alternatively, SJs might influence actomyosin cable function by directly regulating actin and/or myosin. To address these hypotheses, we investigated whether the effect of kune loss of function on the actomyosin cable is autonomous to each cell or, in contrast, nonautonomous. We knocked down Kune in a specific subset of epidermal cells using RNAi by expressing double-stranded RNA (dsRNA) against kune through the Gal4/UAS system (St Johnston, 2013). We expressed Kune UAS-dsRNA under the control of the engrailed-Gal4 (en-Gal4) promotor, which induces expression in a segmented pattern in the embryonic epidermis (Brand and Perrimon, 1993). This allowed us to image the behavior of both WT and kune-deficient cells in the same embryo and observe how that affects wound closure and actin cable formation. To validate this technique, we fixed unwounded control and Kune-RNAi embryos and stained them for the SJ protein Coracle. As expected, Coracle localization was down-regulated in the cells expressing kune RNAi but not in surrounding WT cells (Fig. S4, A and B), showing that kune loss of function impairs SJ localization in a cell-autonomous manner.
Next, we followed wound closure in kune RNAi embryos. All epidermal cells expressed GFP::Moesin, an F-actin reporter (Kiehart et al., 2000), and the segmented patches of kune RNAi (or control RNAi) cells additionally expressed Cherry::CAAX, a membrane marker (Sens et al., 2010), to distinguish RNAi-expressing from RNAi-nonexpressing cells. We examined wound closure in wounds inflicted in two different regions of the epidermis: wounds located between two en-Gal4 stripes, in which wound edge cells did not express kune RNAi (Kune RNAi minus; Fig. 6, A and C; and Video 4); and wounds spanning an en-Gal4 stripe, in which some of the wound edge cells expressed kune RNAi (Kune RNAi plus; Fig. 6, B and D; and Video 5). Strikingly, both types of wounds led to closure defects similar to kune mutants. Kune RNAi plus wounds showed the strongest phenotype, resembling kune strong mutants, whereas Kune RNAi minus wounds developed a phenotype similar to kune mild mutants (compare Fig. 6 [A–F] with Figs. 2 [B–F] and S4 [C and D]). As in kune mutants, the wound size also influenced closure time in Kune RNAi embryos (Fig. 6, E and F; and Fig. S4, C and D). First, we have to consider that all minus wounds had an area 1,100 µm2 due to the small space between two en-Gal4 stripes. We found that the wound-closure time of Kune RNAi–plus large wounds (1,100 µm2) was significantly higher than the wound-closure time of Kune RNAi–plus and –minus small wounds (1,100 µm2). However, the closure time of Kune RNAi–minus and –plus small wounds was similar (Fig. S4 D). These results support the hypothesis that SJs are necessary for wound healing not only in wound-edge cells but also in the surrounding tissue.
To determine whether Kune loss of function affects the actin cable cell autonomously, we measured F-actin intensity in the plus wounds and compared en-Gal4–positive (Kune RNAi) with en-Gal4–negative (WT) regions of the cable. Interestingly, F-actin levels were not different between Kune RNAi and WT cells in the same wound (Fig. 6 G), suggesting that the effects of Kune loss of function on the actin cable are not cell autonomous. Altogether, these data indicate that SJs are not required for actomyosin cable function in a cell-autonomous manner, suggesting SJs do not directly target the cytoskeleton. This favors the hypothesis that the impact of SJs on cable contraction is a consequence of defects at the tissue level.
SJs regulate cellular shapes and rearrangements
Cell shape changes and rearrangements have been shown to promote efficient wound closure (Razzell et al., 2014). To address whether SJs regulate cell shape, we imaged unwounded ubi-E-cad::GFP transgenic embryos, which allow us to detect cell boundaries, and compared cell apical area and elongation (i.e., aspect ratio) in control with kune mutant embryos (Fig. 7, A and B). At this embryonic stage, the epidermis is composed of two types of cells that give rise to denticle-containing and denticle-free (or smooth) cells in the larva and thus exhibit distinct cell shapes (Price et al., 2006; Walters et al., 2006). We found that the average area of smooth cells in kune mutants was significantly larger than in controls (Fig. 7 C), while the average aspect ratio was significantly increased in denticle cells (Fig. 7 D). This means that kune denticle cells are more stretched along the dorsal–ventral axis than WT cells while keeping their average area; conversely, kune smooth cells maintain their aspect ratio but increase their size when compared with WT cells.
To determine whether this effect was cell autonomous, we performed the same measurements in embryos expressing kune RNAi in en-Gal4 stripes (Fig. 7, E and F) and also found cell shape phenotypes. Smooth cells showed larger areas in kune RNAi than in control embryos, whereas both smooth and denticle cells showed increased aspect ratios when compared with control embryos (Fig. 7, G and H). Interestingly, when we analyzed each cell row in separate to distinguish kune RNAi–positive from kune RNAi–negative cells, we observed that even the latter showed significant differences in cell area and aspect ratio when compared with controls (Fig. S5, C–F). Furthermore, the specific cell rows showing altered cell shapes varied between kune RNAi and kune mutants (Fig. S5, A and B). These results suggest that SJs influence epithelial cell shapes in a cell-nonautonomous manner, likely by regulating the properties of the epithelium at the tissue level.
We then checked whether cellular rearrangements are altered in kune mutants during wound closure. We measured the number of neighbor cells of each cell adjacent to the wound margin during the first 45 mpw, when wound-closure defects appear in the kune mutants. We found that cells keep the average number of neighbors approximately constant during this time period (five to six neighbors per cell; Fig. 7 K). Also, the number of cells in the first two rows closer to the wound margin does not change significantly (Fig. 7 L), suggesting that cellular rearrangements do not occur at a large scale. However, controls exhibited significantly more neighbor exchange events than kune mutants in cells at the wound edge (Fig. 7, I, J, and M). Moreover, the fraction of cells undergoing cell neighbor exchange events is higher in controls than in kune mutants (Fig. 7 N). In conclusion, these data suggest that SJs are required for proper cellular rearrangements during wound closure.
SJs influence the mechanical properties of the embryonic epidermis
It is now accepted that cell and tissue shapes highly rely on their mechanical properties (Paluch and Heisenberg, 2009). The generation and transmission of forces in an epithelial cell are mainly controlled by the actomyosin cytoskeleton and the AJs, while the contribution of OJs is still not clear (Sluysmans et al., 2017). Considering that Kune seems to regulate cell shapes and wound closure at the tissue level, we hypothesized that SJs might regulate the mechanical properties of the epithelium. To estimate tissue physical properties, we used the laser ablation technique and modeled the tissue as a viscoelastic material as described previously (Paluch and Heisenberg, 2009; Campàs, 2016). Immediately (in the first 30 s) after a small laser cut, the tissue, which is under homeostatic tension, relaxes and opens up to a maximum point before an active repair response begins. This fast response allows us to estimate not only the homeostatic tension of the tissue but also its effective viscoelastic properties at small time scales. Using control and kune mutant embryos, we performed laser cuts at the junction between two cells as described previously (Behrndt et al., 2012) and measured the displacement along time of the two vertices associated with the epidermal cells subjected to laser cutting (Fig. 8, A and B; and Video 6). We observed that kune mutant cells seemed to relax faster than control cells, whereas the maximum displacement was approximately the same (Fig. 8 C). The simplest possible viscoelastic model that describes this relation is the Kelvin-Voigt (KV) model (see Materials and methods; Fernandez-Gonzalez et al., 2009; Mayer et al., 2010). Using this model, the displacement of the vertices is given by the equation d = dM(1 − e−t/τ), where dM is the maximum displacement, and τ is the retardation time. A fit of our data gives us τ = 13.9 ± 2.2 s and dM = 2.74 ± 0.15 µm for controls versus τ = 9.0 ± 1.2 s and dM = 2.77 ± 0.09 µm for kune (Fig. 8 C). The KV model tells us that
where η is the viscous drag coefficient, G is the stiffness, and F0 is the homeostatic tension. Our results suggest that even if the tension F0 and the stiffness G are altered in the mutant, their ratio remains approximately constant. However, the smaller retardation time in the mutant can be explained either by a larger stiffness (and consequently larger tension) or by a decrease in the viscous drag coefficient.
A careful inspection of the average displacement for the initial time window (t 20 s) is also compatible with a weak power law model (Fig. 8 D; Ma et al., 2009), which can be characterized by the equation
where G′ and τ′ are scale factors for stiffness and time (see Materials and methods). When α approaches 0, the system behaves more like an elastic solid, whereas when α tends to 1, the system behaves more like a viscous fluid. Interestingly, the power law exponents for controls and kune are approximately the same (α 0.44), revealing systems with both solid and fluidic behaviors. These values are similar to the ones obtained previously for the amnioserosa cell junctions in the Drosophila embryo (Ma et al., 2009). However, the observation that the prefactor of this equation, given by F0/(G′1−αη′α), is higher in the mutant than in controls (Fig. 8 D) may be justified by a lower effective viscous drag coefficient or a higher tension in the mutant as seen with the KV model.
These results suggest that SJs significantly impact the mechanical properties of the tissue either by increasing its homeostatic tension or by reducing its effective viscosity. Although we cannot exclude any hypothesis, we strongly favor the latter as it seems more compatible with the idea of defective junctions that are less able to sustain the membrane relaxation occurring after a laser cut and with the observation that mutant cells exhibit larger apical surface areas than WT cells.
This study uncovers a novel role for SJs in regulating epithelial mechanics and efficient wound closure in the Drosophila embryonic epidermis. In simple epithelia, the main driving force to resolve a small lesion in a scar-free manner is proposed to be actomyosin contractility at the wound edge (Garcia-Fernandez et al., 2009; Zulueta-Coarasa and Fernandez-Gonzalez, 2017). Cellular rearrangements and shape changes have also been found to contribute for wound closure (Razzell et al., 2014). According to our data, SJs seem to be more important for these latter processes. Although the absence of functional SJs affects the actomyosin cable and junctional dynamics at the wound edge, these defects are milder than the ones we would expect considering the severity of the wound closure phenotype. In addition, the observed decrease in F-actin at the cable appears to be noncell autonomous, suggesting that it might be a consequence of defects at the tissue level. Finally, the wound-closure dynamics in SJ mutants is very different from the loss-of-function phenotypes of actomyosin or AJs (Wood et al., 2002; Abreu-Blanco et al., 2012; Fernandez-Gonzalez and Zallen, 2013; Hunter et al., 2015; Matsubayashi et al., 2015). Instead of the typical slower contraction described in actomyosin and E-cad mutants, in the strong SJ mutants, wounds undergo a stage of great expansion. Further supporting our hypothesis, impairing the function of the claudin Kune in only a subset of epidermal cells leads to wound-closure defects similar to the ones seen in kune mutants. Notably, a phenotype was detected even when Kune-deficient cells were part of the tissue surrounding the wound but not directly at the leading edge that forms the actomyosin cable. Recently, the classical idea that the actomyosin cable is the driving force for wound closure has been challenged. Ducuing and Vincent (2016) have proposed that this structure rather acts as a passive force field that protects cells from the forces present at the tissue scale instead of working like an active purse string. Our data are compatible with this view. Actomyosin cable contraction might not be as crucial as previously anticipated, and tissue level rearrangements and adhesive forces regulated by SJs and AJs might be key drivers of wound closure in simple epithelia.
Still, we cannot exclude that SJ dynamics at the wound edge are needed to form a proper actomyosin cable as shown for AJs (Carvalho et al., 2014; Hunter et al., 2015; Matsubayashi et al., 2015). Dlg, an SJ scaffolding protein, has been proposed to counteract myosin activity in the Drosophila embryonic epidermis (Simone and DiNardo, 2010). Similarly, the observed accumulation of Nrx-IV and Nrg at the wound edge in kune mutants could interfere with myosin activity at the cable. It has also been shown that cellular rearrangements during wound closure depend on myosin function (Razzell et al., 2014). Therefore, it is also possible that SJs control cellular rearrangements by directly regulating myosin activity. Thus, a direct molecular link between SJs and the actomyosin cytoskeleton should be tested in future studies.
Recent research in mammalian systems has already proposed a link between OJs and wound healing, but functional research has only been attempted in in vitro systems (Volksdorf et al., 2017). In this study, we show for the first time that OJs are required for wound healing of simple epithelia in vivo.
Our study is also the first evidence that SJs play an important role in epithelial mechanics in Drosophila. In fact, our data suggest that the wound-closure phenotypes observed in SJ mutants are a consequence of defects in tissue mechanics already present in the unwounded tissue. When Kune is missing, epithelial cells develop altered geometries and reduce neighbor exchange events during wound closure when compared with controls. Moreover, kune mutant cells react differently than WT cells to mechanical perturbation, leading us to propose that SJs contribute to the viscous drag component of the tissue. At the molecular and cellular level, the effective global viscosity at the cell membrane should be related to the physical properties of all junctions, e.g., their flexibility and ability of junctional molecules to bind or dissociate from each other. When SJs are defective, the physical connections between cells likely become weaker or looser, so a decrease in the global viscosity and in the stiffness of their membranes should be expected. However, our modeling is also consistent with kune mutants presenting higher tension and stiffness. Yet the observation that kune mutant cells exhibit larger apical surface areas, typically associated with lower tensions (Fanning et al., 2012; Choi et al., 2016; Tsoumpekos et al., 2018), strongly argues against this idea. Interestingly, recent studies have shown that the lack of zonula occludens (ZO) proteins at vertebrate TJs can either lead to increased tension in the Xenopus laevis embryo and in MDCK epithelial cells or to decreased tension in human endothelial cells (Tornavaca et al., 2015; Choi et al., 2016; Hatte et al., 2018). Therefore, this regulation might be context specific.
The influence of SJs on tissue mechanics is different from what has been shown for AJs: in AJ loss of function, cortical tension dramatically decreases, whereas viscoelastic properties are typically maintained (Paluch and Heisenberg, 2009). This might be, at least partially, because AJs and SJs connect differently to the cytoskeleton. Whereas AJs associate with actin, the links between OJs and the cytoskeleton are still poorly understood. In vertebrates, ZO scaffolding proteins also bind to actin, but their role in mechanotransduction is only beginning to be explored (Tornavaca et al., 2015; Choi et al., 2016; Sluysmans et al., 2017; Hatte et al., 2018). In Drosophila, the SJ components Nrg and Na+/K+ ATPase can associate with the spectrin-based cytoskeleton, which in turn plays mechanosensory functions and binds actin and myosin (Dubreuil, 2006; Meng and Sachs, 2012; Krieg et al., 2014). Therefore, it is possible that SJs are also able to transmit and/or sense mechanical cues but through different mechanisms than AJs. Finally, we should also note that the viscoelastic properties of cells and tissues in vivo have been mostly inferred indirectly, so the impact of both AJs and OJs for the viscoelastic behavior of tissues in the context of an organism is far from being characterized. Future studies combining quantitative and direct measurement methods with theoretical modeling should help us uncover these mysteries.
Materials and methods
Drosophila strains and genetics
Crosses were performed at 25°C or 29°C on standard Drosophila medium. The w1118 strain was used as control. All strains used were purchased from the Bloomington Drosophila Stock Center unless stated otherwise. Mutant alleles and chromosomal deletions used were kuneC309, sinunwu7, sinu06524, pckG0012, Nrx-IV4304, Nrx-IVEP604, Nrx-IVEY06647, nrg14, nrv2k13115, cora5, varif0033, gli1, dlg1EY05003, dlg12, dlg15, crokKG06053, coldf05607, Df(2L)BSC255, Df(2R)BSC630, Df(2R)BSC630, and Df(2R)BSC26. All alleles and deletions were crossed to balancer stocks that express GFP under the control of the twist promoter (Halfon et al., 2002), and homozygous mutant embryos were identified by the absence of GFP fluorescence. For the wounding assay, transallelic combinations or combinations with chromosomal deletions were used whenever possible to avoid effects of secondary mutations. For confocal imaging, mutant alleles were recombined with live reporter lines.
The live reporters used were UAS-Cherry::Moesin under the control of the e22c-Gal4 promoter (Millard and Martin, 2008) or sqh-GFP::Moesin (Kiehart et al., 2000) to mark F-actin; zipCPTI-100036-GFP (Lye et al., 2014) to mark myosin-2 heavy chain; UAS-mCherry::CAAX (Sens et al., 2010) or UAS-GFP to identify RNAi-expressing cells; ubi-E-cad::GFP (Oda and Tsukita, 1999) and E-cad::GFP (Huang et al., 2009) to mark E-cad; and Nrx-IV-GFP (CA06597) and Nrg-GFP (G305; Morin et al., 2001; Buszczak et al., 2007) to mark Nrx-IV and Nrg-GFP, respectively.
To knock down kune expression by RNAi, UAS-kune-dsRNA-HMS01755 (TRiP; Harvard Medical School) was expressed under the control of the en-Gal4 promoter to obtain a segmented knockdown in the epidermis. Controls were obtained by crossing en-Gal4 flies with the RNAi control line containing the same insertion sites as the RNAi line (TRiP; Harvard Medical School).
The wounding assay was performed as previously described (Campos et al., 2010; Geiger et al., 2011; Carvalho et al., 2014). Briefly, stage 15–16 embryos were collected from laying pots kept at 25°C overnight, dechorionated in 50% bleach, and rinsed extensively with water. Mutant embryos were aligned on a slide with a double-sided tape affixed to it and covered with halocarbon oil 700 (Sigma-Aldrich) and a coverslip. The embryos were wounded at 25°C by using a nitrogen laser-pumped dye laser (435 nm; Micropoint Photonic Instruments) connected to a Nikon/Andor Revolution XD spinning-disk confocal microscope with an electron-multiplying charge-coupled device (EMCCD) camera (iXon 897) using iQ software (Andor Technology) and using a 60× Plan Apochromat VC Perfect Focus System (PFS) 1.4 NA oil-immersion objective. After wounding, the coverslip was carefully removed, and the embryos were left to recover in a humid chamber at 20°C. The wounded embryos were scored under a stereo microscope for closed/open wounds 16 h later. Percent wound closure was calculated as the ratio of nearly hatching embryos with unclosed wounds over the total number of wounded embryos (dead animals were excluded). Fisher’s exact test was performed to test for statistical significance.
Immunohistochemistry and imaging of fixed samples
Embryos at stage 15 were laser wounded and placed in a humid chamber to recover. Wounded embryos were transferred to a glass vial containing a mix of 1:1 heptane and 8% formaldehyde in PBS and fixed for 40 min, manually devitellinized, washed in PBST (PBS + 0.1% Triton X-100), incubated for 1 h in blocking solution (0.3% BSA + 0.3% Triton X-100 in PBS), which was followed by primary antibody and probe incubation overnight at 4°C. The primary antibodies and probes used were mouse anti-Coracle C-terminal (1:500; Developmental Studies Hybridoma Bank), rabbit anti-GFP (1:2,000; Invitrogen), and Alexa Fluor 594–conjugated phalloidin (1:200; Invitrogen). Following incubation, embryos were washed in PBST, incubated in secondary antibodies conjugated to Alexa Fluor 488 or Alexa Fluor 568 (Invitrogen) for 2 h in blocking solution, washed in PBST, and mounted in 80% glycerol and 2% DABCO (Sigma-Aldrich). Imaging was performed on an LSM 710 confocal microscope (Zeiss) with a 63× Plan Apochromat 1.4 NA oil-immersion objective (Zeiss). Images were acquired using Zen software (Zeiss) and a step size of 0.37 µm.
Live imaging was performed as described previously (Geiger et al., 2011; Carvalho et al., 2014). Dechorionated stage 15 embryos were mounted on their ventral side on glass-bottomed culture dishes (MatTek) with embryo glue (double-sided tape diluted in heptane) and covered with halocarbon oil 27 (Sigma-Aldrich). Embryos were wounded as described above for the wounding assay except that the laser power was lower in order to inflict smaller wounds that are able to close during the imaging procedure. Time-lapse microscopy of transgenic embryos was performed at 25°C on a Nikon/Andor Revolution XD spinning-disk confocal microscope with an EMCCD camera (iXon 897) using iQ software and using a 60× Plan Apochromat VC PFS 1.4 NA oil-immersion objective, a 60× Plan Apochromat VC PFS 1.2 NA water-immersion objective, or a 100× Plan Apochromat PFS 1.4 NA oil-immersion objective (Nikon). Individual z slices with a step size of 0.36 µm (60× objectives) or 0.25 µm (100× objectives) were acquired every 1, 2.5, or 5 min for 60–300 min.
Laser-cutting experiments were performed using a UV laser cutter setup as previously described (Behrndt et al., 2012) on both WT and kune mutant embryos expressing ubi–E-cad::GFP and Cherry::Moesin at stage 15 of embryonic development. Laser cuts were performed in ventral epidermal smooth cells at the level of the AJs by applying 50 UV pulses to two equidistant sites in a 2-µm-long line perpendicular to the junction. A single cell was cut per embryo. Images were acquired every 1 s for at least 1 min in an Andor Revolution spinning-disk confocal microscope with a Zeiss 100× 1.3 NA oil-immersion lens.
Image analysis and quantifications
All images were processed and analyzed using Fiji (ImageJ; National Institutes of Health [NIH]; Schindelin et al., 2012) except when mentioned otherwise.
Fluorescence intensity measurements
To measure F-actin and myosin intensities at the wound edge, maximum Z projections of Cherry::Moesin, GFP::Moesin, and Zip::GFP stacks were used after background subtraction. The wound edge and the cortical region of epithelial cells before wounding were outlined using a 5-pixel-wide segmented line, and the average intensity was obtained using the Measure tool. To obtain intensity values before wounding, 10 cells per embryo were measured. For F-actin quantifications, cells containing actin-rich denticle precursor structures were not included in the measurements as they mask the actin present at the cable and cell cortex. For F-actin quantifications in Kune RNAi images, the ratio between F-actin intensity in en-Gal4–positive cells and en-Gal4–negative cells was calculated for each embryo.
To measure Nrx-IV, Nrg, and E-cad intensities, maximum Z projections of Nrx-IV–GFP, Nrg-GFP, and ubi–E-cad::GFP stacks were used, respectively. Background fluorescence was subtracted from each image. The Cherry::Moesin channel was used to confirm the location of the wound edge. Junctions were outlined using a 4-pixel-wide segmented line and the average intensity obtained using the Measure tool. To calculate the intensity decrease (fold change) at the wound edge, the intensity value for each wound edge junction after wounding (10 and 30 mpw) was divided by the intensity value obtained for the same junction before wounding (Nrx-IV and E-cad), or the intensity value of the whole edge was measured after wounding (10 and 30 mpw) and divided by the intensity value obtained for 10 junctions before wounding (Nrg).
To measure Nrx-IV and Nrg apical ratios, YZ sections were used. Background fluorescence was subtracted from each image. The Cherry::Moesin channel was used to confirm apical–basal limits of each cell. The whole apical–basal region was outlined using a 4-pixel-wide segmented line, and the average intensity was obtained using the Measure tool. The same procedure was applied to the apical half of the cell. The intensity at the apical region was calculated as the ratio between the intensity at the apical region and the total intensity for each cell.
Maximum Z projections were obtained from Cherry::Moesin or GFP::Moesin stacks, an ellipse was drawn along the wound edge, and the area was obtained using the Measure tool. For each embryo, the area was normalized relative to the initial wound area.
Maximum Z projections of Cherry::Moesin stacks were used, and protrusions were analyzed similar to what has been described (Abreu-Blanco et al., 2012; Zulueta-Coarasa et al., 2014). Briefly, each image was adjusted to an intensity threshold, which was the mean intensity of the image plus one SD. The area inside the wound, adjacent to the actin cable, was outlined with a 2.75-µm-wide line, and the number of pixels were measured and normalized to the total area. For this analysis, control and kune mild embryos were considered.
Cell area and aspect ratio
Maximum Z projections of ubi–E-cad::GFP images were used. Cells were outlined, and the area and aspect ratio were measured using the Measure tool. The aspect ratio reflects the ratio between the length along the dorsal–ventral and the length along the anterior–posterior axis. Cells were identified by their position relative to sensory organ precursors, which are located in the smooth cell layer 2, two layers anterior to denticle layer 1.
Neighboring cells and exchange events
Maximum Z projections of E-cad::GFP–labeled embryos during the first 45 min after wound closure, acquired with a time interval of 2.5 min, were used. An algorithm based on the invasion percolation of Ranked Surfaces (Schrenk et al., 2012) was devised to segment cells using GNU Image Manipulation Program (GIMP) and MATLAB (MathWorks), described as follows. First, a point inside the limits of every cell is marked manually. Each of these points is then uniquely identified as a cluster of pixels with a given label such that we have as many clusters as cells. All the nearest neighboring pixels of every cluster are added to a list of pixels and sorted according to their grayscale intensity. The first pixel of the list, the darkest one, is removed from the list and added to the same cluster of the neighboring pixel (meaning that it belongs to the same cell) identified with the proper label. The nearest neighbors of this new pixel that are not yet in the list are then added to the list. We proceeded iteratively in the same way, following the list of pixels sorted by their grayscale intensity until all pixels belonged to one of the clusters (cells). Since the membranes of the cells were the lightest regions, they were the last to be painted, which allowed us to trace the boundaries of the cells. These segmented images were used to quantify the number of cells and neighbors of each cell in the first two rows closer to the wound edge.
For neighbor exchange events, all cells from the first row adjacent to the wound edge were quantified. For each cell, the number of events was measured (gain or loss of neighbors). This value was normalized to the total number of cells in the first row to obtain the ratio of neighbor exchange events for each embryo. To calculate the fraction of cells undergoing neighbor exchange events, the number of cells performing these events was divided by the total number of cells in the first row adjacent to the wound margin. For this analysis, control and kune strong mutants were considered.
Displacements upon laser cutting
Single slices of the ubi–E-cad::GFP channel were used. A straight line between the two vertices associated with the cut junctions was drawn, and its length was measured using the Measure tool. To model the data, an exponential function was fit to the displacement values over time using the least-square regression method, and values for the time constant (retardation time) and the plateau (maximum displacement) were calculated using Prism (GraphPad Software). To analyze the weak power law behavior of the data, displacement values were log transformed, a linear regression was fit using the least-squares regression method, and values for the slope and intercept were calculated using Prism.
Statistical analysis was performed using Prism and Excel (Microsoft). Statistical tests, P values, sample sizes, and error bars are indicated in the respective figure legends.
The simplest possible viscoelastic model that describes our laser cut experiments (Fig. 8) is depicted in Fig. S5 G. The tissue is represented by a regular structure of cells, for which each membrane segment is associated with a KV viscoelastic element composed of a spring of stiffness G and dashpot of drag coefficient η, both in parallel. All cells were initially at rest, and the membranes were under the homeostatic tension F0. Each membrane segment of equilibrium length x had an initial length X0 = x + u0. The initial deformation of the membrane was related to the initial tension of the tissue through u0 = F0/G. If only one membrane was cut, the associated vertices would relax and move slightly toward the interior of their neighboring cells. If the inertia of this system was neglected (small Reynolds number regimen), the sum of the forces applied to each vertex must be 0, which implies F = 0 at all times. The deformation of each membrane u(t) (see Fig. S5) may be obtained through the KV constitutive equation F = Gu + ηú = 0, where the dot represents a time derivative, and which has the simple solution u(t) = u0e−t/τ, with τ = η/G as the retardation time. To simplify our analysis, if we suppose that all other vertices are fixed, the relation between the deformation u(t) of each membrane and the displacement ξ of the vertex is given by geometry. For small deformations, the displacement of the vertices was given by
The constant of proportionality varied with the geometry of the cells. As it may be seen from Fig. 8 (A and B), the geometry of the epithelial cells was irregular. To have quantifiable results, we only measured displacements in cells with similar geometries (smooth cells). As it may be seen from Eq. 1, the wound opening time, i.e., the time necessary for the vertices to reach their maximum displacement, was associated with the retardation time of the KV model, τ = η/G. The initial displacement velocity is proportional to F0/η, and the maximum displacement is proportional to F0/G.
A fit of our data gives us τ = 13.9 ± 2.2 s and dM = 2.74 ± 0.15 µm for controls versus τ = 9.0 ± 1.2 s and dM = 2.77 ± 0.09 µm for kune (Fig. 8 C). These results suggest that even if the tension F0 and stiffness G are altered in the mutant, their ratio remains approximately constant. However, the smaller retardation time in the mutant can be explained either by a larger stiffness (and consequently a larger tension) or a smaller drag coefficient.
The simple KV model described above includes only the main mechanisms involved in cell and tissue relaxation. To improve our viscoelastic model, each membrane segment may be associated with more complicated viscoelastic elements (Tschoegl, 1989), yielding a better approximation of the experimental displacement of the vertices. A careful inspection of the average displacement for the initial time window (t 20 s) is compatible with the weak power law behavior:
where G′ and τ′ = η′/G′ are scale factors for stiffness and time, and η′ a scale factor for the effective drag coefficient of the tissue. The weak power law behavior may be a consequence of not only one but a collection of connected KV elements (in series), with a spectrum of retardation modes with different time scales reflecting for instance the fractal structure of the tissue and the cell’s cytoskeleton (Patrício et al., 2015). This power law behavior has been also frequently associated to the soft glassy material model (Sollich et al., 1997) or its extended version, the soft wormlike chain model (Kroy and Glaser, 2007), in which there is a collection of relaxation modes (associated with Maxwell elements; a spring and a dashpot in series) with different time scales, reflecting a glassy behavior. The system was considered to be solidlike whenever 0 α 0.5 and fluidlike whenever 0.5 α 1.
Interestingly, the power law exponents for the WT and kune mutant are approximately the same and close to 0.44, revealing systems which have both solid and fluidic behaviors. The higher prefactor, F0/(G′1−αη′α), observed for the mutant case may be justified by a larger tension or a smaller effective drag coefficient. In analogous experiments, similar power law exponents (close to 0.4) were obtained for the amnioserosa cell tissues in the fruit fly at different stages of growth (Ma et al., 2009).
Microrheology measurements performed by attaching magnetic microspheres to the cell and applying external oscillatory magnetic fields also revealed the same type of power law behaviors (Fabry et al., 2001). The weak power law exponents, which are associated in this case with the mechanical properties of one individual cell, are smaller (more solidlike), often close to 0.2 for different types of cells, but they may take, in some occasions, higher or lower values (Kollmannsberger and Fabry, 2011; Pritchard et al., 2014).
Online supplemental material
Fig. S1 shows that the mutant sinu also develops a wound-closure phenotype. Fig. S2 shows that myosin intensity correlates with wound perimeter over time and that cellular protrusions are not affected in kune mutants. Fig. S3 shows Nrg-GFP and Coracle dynamics during wound closure in controls and kune mutants. Fig. S4 shows that Coracle is down-regulated in Kune RNAi–expressing cells and that wound-closure phenotypes in Kune RNAi embryos vary depending on wound size. Fig. S5 shows that kune mutant and RNAi affect cell shape in a nonautonomous manner and depicts a scheme on the KV model. Video 1 shows wound-closure dynamics in control, kune mild, and kune strong mutant embryos marked with an F-actin reporter. Video 2 shows Nrx-IV–GFP and F-actin dynamics in a control embryo. Video 3 shows Nrx-IV–GFP and F-actin dynamics in a kune strong mutant embryo. Video 4 shows wound-closure dynamics in control and kune RNAi minus embryos marked with an F-actin reporter. Video 5 shows wound-closure dynamics in control and kune RNAi plus embryos marked with an F-actin reporter. Video 6 shows the response to a laser cut in control and kune mutant embryos marked with an E-cad reporter.
We are indebted to Nina Matova for her insight and precious contributions; she will be remembered forever. We are grateful to T. Pereira for invaluable support on imaging and data analysis; M. Behrndt and D. Siekhaus for advice and technical help in laser-cutting experiments; C. Crespo and F. Vasconcelos for manuscript review; past and present members of the Tissue Morphogenesis and Repair Lab and R. Teodoro for helpful discussions and technical help; C. Martins for technical help; the microscopy facilities at Chronic Diseases Research Center, Instituto de Medicina Molecular, and Instituto Gulbenkian de Ciencia; the Bloomington Drosophila Stock Center (NIH P40OD018537), the TRiP at Harvard Medical School (NIH/NIGMS R01-GM084947), the Yale School of Medicine Flytrap project, and the Kyoto Stock Center (Drosophila Genomics Resource Center) for providing Drosophila lines.
This research was supported by Fundação para a Ciência e a Tecnologia (SFRH/BPD/84569/2012 to L. Carvalho, PD/BD/106058/2015 to S. Ponte, SFRH/BD/119240/2016 to A.S. Nunes, UID/FIS/00618/2013, PTDC/FIS-MAC/28146/2017, PTDC/BIA-BID/29709/2017 and CONGENTO), the European Research Council (2007-StG-208631 and ERC-2015-PoC-713735-EMODI), a European Commission Marie Curie Intra-European Fellowship (PIEF-GA-2009-255573), the Partenariat Hubert Curien Pessoa program (30920XM), the European Science Foundation (Quantitative Models of Cellular and Developmental Biology Research Networking Program, grant 6839), and the iNOVA4Health Research Unit (UID/Multi/04462/2013), cofunded by Fundação para a Ciência e Tecnologia, through national funds and by FEDER under the PT2020 Partnership Agreement.
The authors declare no competing financial interests.
Author contributions: L. Carvalho conceived, conducted, and analyzed all experiments, assembled all figures and videos (except Fig. S5 G), and wrote the paper; P. Patricio performed the theoretical modelling, helped analyzing data in Figs. 7 and 8, assembled Fig. S5 G, and wrote the paper; S. Ponte helped conducting and analyzing experiments in Figs. 1, 2, and S1 and edited the paper; C.P. Heisenberg supervised experiments in Fig. 8 and edited the paper; L. Almeida helped in the theoretical modelling and interpreting the data in Fig. 8; A.S. Nunes and N.A.M. Araújo devised the algorithm for image segmentation and analyzed the number of neighboring cells; A. Jacinto supervised the study, contributed to experimental design, and edited the paper.