Integrin-based adhesions promote cell–cell junction and cytoskeletal remodelling to drive embryonic wound healing

Embryos heal wounds rapidly, with no inflammation or scarring (Rowlatt, 1979; Whitby and Ferguson, 1991), in a process driven by collective cell movements (Hunter and Fernandez-Gonzalez, 2017). Cell movement and coordination during tissue repair require cytoskeletal polarization. In the cells adjacent to the wound, actin and the molecular motor non-muscle myosin II accumulate at the wound edge (Martin and Lewis, 1992; Brock et al., 1996), forming a supracellular cable that coordinates cell movements (Wood et al., 2002; Zulueta-Coarasa and Fernandez-Gonzalez, 2018). In parallel, the adherens junction proteins E-cadherin, β-catenin and α-catenin are internalized from the wound edge (Brock et al., 1996; Wood et al., 2002; Abreu-Blanco et al., 2012; Zulueta-Coarasa et al., 2014; Hunter et al., 2015; Matsubayashi et al., 2015), and accumulate at former tricellular junctions along the wound margin (Wood et al., 2002; Abreu-Blanco et al., 2012; Zulueta-Coarasa et al., 2014). Disrupting junctional remodelling prevents actomyosin cable assembly and slows down wound closure (Hunter et al., 2015; Matsubayashi et al., 2015).

Cell migration in many systems involves adhesion to the extracellular matrix (ECM) (Fraley et al., 2010; Charras and Sahai, 2014). Cells adhere to the ECM through integrins, transmembrane receptors formed by α- and β-subunits that link the ECM to the actin cytoskeleton. In two-dimensional (2D) cell cultures, integrin-based adhesions (IBAs) remodel in response to mechanical cues from actomyosin contractility and substrate stiffness (Geiger et al., 2009; Fraley et al., 2010). Similar to the process of embryonic repair, cells in 2D tissue cultures assemble a contractile actomyosin cable around wounds (Bement et al., 1993). In the presence of ECM, the cable relays tension to the substrate through IBAs (Brugués et al., 2014). In addition, cell–cell and cell–ECM adhesions are mechanically linked both in tissue culture and in vivo (Bécam et al., 2005; Borghi et al., 2010; Jülich et al., 2015; Goodwin et al., 2017), suggesting that IBAs might contribute to morphogenetic processes that require cell–cell adhesion remodelling. Here, we investigate the role of integrins in Drosophila embryonic wound repair, and we reveal an interplay between cell–cell and cell–ECM adhesion to promote the cytoskeletal rearrangements that drive rapid wound healing.

To determine whether wound healing in the Drosophila embryonic epidermis occurs within an ECM, we examined the presence of ECM proteins midway into embryonic development (stages 14–15), when epidermal cells are already differentiated (Payre, 2003). Through GFP immunofluorescence, we found that epidermal cells were surrounded by the ECM proteins Collagen IVa2 (encoded for by the viking gene) and Laminin A (Fig. 1A,B). These results suggest that the Drosophila embryonic epidermis is embedded within a multi-protein ECM.

To determine whether wound healing is associated with ECM remodelling, we fixed wounded embryos expressing GFP-tagged Viking or GFP-tagged Laminin A. Viking accumulated within and around the wound, potentially deposited by wound-associated macrophages, although the Viking signal was apical to the individual macrophages (Fig. 1C). Laminin was present throughout the epidermis, with a modest accumulation at the wound edge (Fig. 1D). Thus, our data suggest that the ECM is remodelled during wound closure. Consistent with this, a GFP-tagged form of Talin, a protein that links IBAs to the actin cytoskeleton (Bulgakova et al., 2012; Goodwin et al., 2016), accumulated at the wound edge by 3±1-fold (mean±s.d.) over the first 40 min of wound healing (n=6 wounds, P=0.03; Fig. 1E, Movie 1). Talin localized apically along the wound edge, on the same apical-basal plane as E-cadherin (Fig. 1F,F′, Movie 2), suggesting that IBAs form at the wound front and might be interacting with adherens junctions. Taken together, our data suggest that cell–ECM adhesion might contribute to embryonic wound closure.

To establish whether cell–ECM adhesion is necessary for rapid wound repair, we quantified the dynamics of wound closure in embryos in which integrins had been knocked down by RNA interference (RNAi). Integrin βPS, encoded by the gene myospheroid (mys) is the predominant β-subunit in the Drosophila embryonic epidermis (Brown et al., 2000). We used the UAS-Gal4 system (Brand and Perrimon, 1993) to drive a transgenic RNAi line against mys. daughterless-Gal4 (da-Gal4)-driven RNAi was expressed throughout the embryo and reduced Mys levels by 33% [3217±508 arbitrary units (a.u.) in controls versus 2169±400 a.u. in mys RNAi embryos, P=0.03; mean±s.d.; Fig. S1A–C], resulting in 100% embryonic lethality. mys RNAi embryos repaired wounds 54% slower than mCherry RNAi controls (P=8×10⁻³; Fig. 2A–D; Movie 3). mys RNAi reduced the recruitment of macrophages to the wound (Fig. S1D,E). However, macrophages are dispensable for embryonic wound closure (Wood et al., 2002) and ablation of macrophages in the Drosophila embryo accelerates wound repair (Weavers et al., 2019), suggesting that reduced macrophage recruitment is not the reason that wound healing is slower in mys RNAi embryos. We did not find any effects of mys RNAi on muscle architecture or attachment as assessed by Myosin heavy chain staining (Fig. S1F,G). Thus, our results indicate that reduced integrin levels in the epidermis slow down wound repair.

To further confirm the role of integrins in wound healing, we disrupted integrins using a temperature-sensitive allele of myspheroid (mys^ts1). mys^ts1 reduces integrin levels at temperatures above 25°C (Beumer et al., 1999). We exposed control embryos expressing wild-type mys and embryos expressing mys^ts1 to the restrictive temperature of 35°C for 90 min prior to wounding. Embryos expressing mys^ts1 and exposed to the restrictive temperature closed wounds 46% slower than controls (P=0.03, Fig. S2A–C). Thus, our data indicate that integrins are necessary for rapid embryonic wound healing.

Integrins can bind to the ECM. To determine whether integrin–ECM binding is necessary for rapid wound healing, we disrupted integrin–ECM interactions by injecting embryos 10 min before wounding with 1 mM Arg-Gly-Asp-Ser (RGDS), a peptide that blocks integrin binding to ECM proteins (Naidet et al., 1987; Yasothornsrikul et al., 1997) (Fig. S2H,I; Movie 4). RGDS slowed wound closure by 45% (P=0.02; Fig. S2J,K). Together, our results show that integrin-based cell–ECM adhesion is necessary for rapid embryonic wound healing.

To investigate how integrin disruption affects wound healing, we quantified the formation of the actomyosin cable in embryos expressing myosin II tagged with GFP (Fig. 2A; Movie 3) or mCherry (Fig. S2H, Movie 4), or expressing the actin-binding domain of Utrophin tagged with GFP as a reporter of filamentous actin (F-actin) (Fig. 2B; Fig. S2A,I, Movies 5–7). We did not find a significant difference in myosin accumulation at the wound edge in mys RNAi embryos (Fig. 2E,F; Movie 3) or in RGDS-treated embryos compared to in control embryos (Fig. S2L,M, Movie 4). However, we measured a significant, 38% reduction in actin polarization to the wound edge 40 min after wounding in mys RNAi embryos (P=0.03, Fig. 2G,H; Movie 5), a 62% reduction in mys^ts1 embryos (P=0.01, Fig. S2D,E, Movie 6) and a 40% reduction in RGDS-injected embryos (P=0.04, Fig. S2N,O, Movie 7). The defect in actin polarization when integrin levels or ECM binding were disrupted was independent of the delay in wound healing – when we restricted our analysis to large (130–170 μm in perimeter), medium (90–130 μm) or small (50–90 μm) wounds, regardless of time point, we found that mys RNAi embryos had lower actin accumulations at the wound edge than did controls for all groups (Fig. S3A,B). Similar effects were observed for RGDS (Fig. S3D,E). Overall, our data indicate that integrins promote actin polarization during wound closure.

The actin cable around wounds is remodelled during repair, with the actin distribution starting out heterogeneous and becoming more uniform in response to increasing tension (Zulueta-Coarasa et al., 2014; Kobb et al., 2019; Rothenberg et al., 2023). To establish whether integrins are necessary for actin dynamics during wound healing, we measured the heterogeneity of the actin cable, quantified as the ratio between the standard deviation and the mean pixel values (Zulueta-Coarasa and Fernandez-Gonzalez, 2018). F-actin intensity along the wound edge became more uniform over time in controls, but not in mys RNAi embryos: the distribution of actin 40 min after wounding was 91% more heterogeneous in mys RNAi embryos compared to controls (P=0.04, Fig. 2I–K). We obtained similar results in mys^ts1 embryos (P=0.03, Fig. S2F,G) and with RGDS (P=0.02, Fig. S2P–R). The defects in actin redistribution in mys RNAi embryos and in RGDS-treated embryos were independent of the delay in wound healing (Fig. S3C,F). Together, our results indicate that actin polarization and remodelling at the wound edge require mechanical engagement of integrins with the ECM. Reduced actin levels could affect the generation of contractile force and delay wound healing (Kobb et al., 2019). Additionally, integrins are necessary for the transmission of forces from cell to cell. In the Drosophila amnioserosa, IBAs limit apical deformation by transmitting actin-based forces basally across cells (Goodwin et al., 2016). Disruption of cell–ECM adhesion in the amnioserosa increases apical force transmission and changes actin organization (Goodwin et al., 2017). Our findings suggest that IBAs might contribute to tension generation, force transmission and actin organization at the wound edge.

To determine whether integrins regulate tension at the wound edge, we used laser ablation to quantify the mechanical properties of the actomyosin cable. We severed the cable when wounds closed to 50% of their maximum area (Fig. 3A,B). The initial recoil velocity after ablation is proportional to the tension sustained (Hutson et al., 2003). We found no difference in the recoil velocity after ablation of the actomyosin cable for mys RNAi embryos (Fig. 3C; Movie 8) or for RGDS-treated embryos (Fig. S4A–C; Movie 9), suggesting that integrins are dispensable for contractility at the wound edge. To determine whether integrins control the viscoelastic properties of the actomyosin cable, we fitted the laser ablation data using a Kelvin–Voigt model to estimate a relaxation time that represents the viscosity-to-elasticity ratio (Kumar et al., 2006; Zulueta-Coarasa and Fernandez-Gonzalez, 2015). The relaxation time was similar to controls in both mys RNAi embryos (Fig. 3D) and RGDS-injected embryos (Fig. S4D), suggesting that integrins do not regulate viscoelasticity at the wound margin.

Coordinated cell movements in epithelia are affected by the mechanical properties of the tissue (Tetley et al., 2019). To determine whether the baseline mechanical properties of the epidermis changed when we disrupted integrins, we used laser ablation to sever individual cell–cell junctions in non-wounded epidermis. mys RNAi reduced the recoil velocity after ablation by 28% (P=0.03, Fig. 3E–H; Movie 10), indicating that the baseline epidermal tension decreased at lower integrin levels. Relaxation times after ablation were not affected by integrin knockdown (Fig. 3H). RGDS treatment did not alter recoil velocities or relaxation times (Fig. S4E–H; Movie 11). Notably, mys RNAi also reduced overall E-cadherin levels in the epidermis by 17% (P=2×10⁻⁹, Fig. 3E,I), and α-catenin levels by 60% (P=0.03, n=4 embryos in each group, Fig. S1F,G), whereas RGDS treatment did not affect E-cadherin levels (Fig. S4I,J), suggesting that differences in adherens junction protein levels could explain the reduction in baseline epidermal tension in mys RNAi embryos. The different effects of integrin knockdown versus blocking integrin binding on epidermal tension and E-cadherin levels suggest that integrin expression, but not mechanical engagement, is necessary for epidermal homeostasis. Alternatively, the prolonged effect of the integrin knockdown by RNAi could lead to changes in E-cadherin expression that the acute inhibition using RGDS does not have time to induce. In addition, integrins can bind ECM proteins in a non-RGD-dependent manner (Naidet et al., 1987). Finally, in Drosophila embryos integrins can adhere to the vitelline membrane (Bailles et al., 2019; Münster et al., 2019), suggesting that alternative binding partners might contribute to integrin signalling during wound healing.

To investigate whether integrins regulate adherens junction remodelling during wound closure, we quantified E-cadherin dynamics at bicellular and tricellular junctions (BCJs and TCJs, respectively) at the wound edge (Fig. 4A,B; Movie 12). In controls, E-cadherin–GFP at BCJs decreased by a maximum of 46±23% (mean±s.d.) within 15 min after wounding (P=9×10⁻⁶, Fig. 4C), and increased by 25±60% at TCJs in the same time period (P=0.004, Fig. 4D). mys RNAi did not affect the depletion of E-cadherin from BCJs (51±21%, P=7×10⁻⁵, Fig. 4C), but prevented the accumulation of E-cadherin at TCJs (−1±56%, P=0.003, Fig. 4D). We obtained similar results in mys^ts1 embryos (Fig. S5A–D, Movie 13) and in RGDS-treated embryos (Fig. S5E–H, Movie 14). Taken together, our results reveal a previously unrecognized interplay between cell-cell and cell-ECM adhesion to drive the collective cell movements that promote embryonic wound closure.

Crosstalk between cell–cell and cell–ECM adhesions occurs both in tissue culture and in vivo, and has been proposed to be bidirectional (Bécam et al., 2005; Borghi et al., 2010; Jülich et al., 2015; Hadjisavva et al., 2022). We find that integrins specifically control TCJ remodelling around embryonic wounds. TCJs at the leading edge of collectively migrating cells have been proposed to act as hubs for the nucleation of supracellular actin cables (Kaltschmidt et al., 2002; Matsubayashi et al., 2015). TCJs might also transmit forces necessary to stabilize actin around the wound (Kobb et al., 2019). Experiments investigating how integrin disruptions or ECM protein knockdowns affect E-cadherin trafficking, actin turnover and the localization and dynamics of different actin regulators will shed light on how IBAs contribute to embryonic wound repair.

Animals were maintained and mated at 18°C or 23°C (room temperature). Drosophila strains were kept in plastic vials or bottles on fly food provided by a central kitchen operated by H. Lipshitz at the University of Toronto. Stage 14–15 embryos (10–12 h after egg laying) were collected from apple juice-agar plates kept overnight at room temperature (23°C) on collection cages. For immunofluorescence staining, we used yellow white, viking:GFP (Morin et al., 2001), or laminin-A:GFP flies (Sarov et al., 2016). For live imaging, we used the following markers: ubi-talin:GFP (Yuan et al., 2010), endo-E-cadherin:tdTomato (Huang et al., 2009), endo-E-cadherin:GFP (Huang et al., 2009), sqh-sqh:GFP (Royou et al., 2004), sqh-sqh:mCherry (Martin et al., 2009) and UtrophinABD:GFP (Rauzi et al., 2010). We used UAS-myospheroid RNAi (Perkins et al., 2015) to knock down integrins, and UAS-mCherry RNAi (BL #35785) as a control. We also used a mys^ts1 temperature-sensitive allele to reduce integrin levels (Wright, 1968; Beumer et al., 1999). UAS constructs were ubiquitously driven with daughterless-Gal4 (Perrin et al., 2003).

Drosophila embryos were dechorionated in 50% bleach for 2 min and rinsed with water. Embryos were aligned ventral-lateral side up on an apple juice-agar pad and transferred to a coverslip coated with heptane glue. For injections, embryos were dehydrated for 7–10 min by placing the coverslip in a plastic container with silica beads (Drierite). Subsequently, embryos were covered with a 1:1 mix of halocarbon oil 27 and 700 (Sigma-Aldrich) (Scepanovic et al., 2021).

Injections were conducted using a Transferman NK2 micromanipulator (Eppendorf) and a PV820 microinjector (World Precision Instruments) coupled to a spinning disc confocal microscope. Embryos were injected with 100–200 pl of 1 mM RGDS peptide (Tocris Bioscience) in water or with water as a control. Injections were into the perivitelline space and were followed by an incubation at room temperature for 10 min before imaging. Drugs are predicted to be diluted by 50-fold in the perivitelline fluid (Foe and Alberts, 1983).

mys^ts1; UtrophinABD:GFP, E-cadherin:tdTomato embryos were dechorionated and transferred to a coverslip, covered with 1:1 halocarbon oil 27:700, and heated at 35°C for 90 min on a hot plate. Embryos were then wounded and imaged. Control embryos expressing wild-type mys were heated and imaged under the same conditions as mutant embryos. Temperature shifts resulted in 100% embryonic or early larval lethality in embryos expressing mys^ts1, but not in controls.

Imaging was performed at room temperature using a Revolution XD spinning disc confocal microscope (Andor Technology) with an iXon Ultra 897 camera (Andor Technology), a 60× oil-immersion lens (NA 1.35; Olympus), and Metamorph software (Molecular Devices). 16-bit Z-stacks were acquired at 0.5 µm steps every 4–30 s (11–21 slices per stack). Maximum intensity projections were used for analysis.

Wounds on the embryonic epidermis were created using a pulsed Micropoint nitrogen laser (Andor Technology) tuned to 365 nm. Ten laser pulses were delivered in six spots along a 13-μm line to generate a wound. Each embryo was wounded only once. To ensure consistency, we restricted our analyses to wounds larger than 700 μm². For spot laser ablations, ten pulses were delivered at a single point over the course of 670 ms to release tension at the wound margin or in the intact epidermis. Samples were imaged immediately before and 1.72 s after spot ablations. After ablation, embryos were imaged every 4 s for spot ablations or 30 s for wound healing assays.

For immunofluorescent staining, we dechorionated stage 14–15 embryos as above. Embryos were fixed for 20 min in a 1:1 mix of heptane and 37% formaldehyde in phosphate buffer. Embryos were manually devitellinized or popped using 90% ethanol, stained with primary antibodies for 2 h at room temperature or overnight at 4°C and fluorescently labelled with secondary antibodies incubated for 1 h at room temperature. Antibodies used were mouse anti-GFP (1:50; DSHB, #DSHB-GFP-12A6), rabbit anti-GFP (1:200, Torrey Pines Biolab, TP401), mouse anti-mCherry (1:50; DSHB, #DSHB-mCherry-3D5), mouse anti-integrin βPS (0.3 μg/ml; DSHB, #CF.6G11), mouse anti-Myosin heavy chain (1:100; DSHB, #3E8-3D3), rat anti-α-catenin (1:20; DSHB, #DCAT-1), Alexa Fluor 488 goat anti-mouse IgG (1:500; Invitrogen), Alexa Fluor 488 goat anti-rabbit IgG (1:500; Invitrogen), Alexa Fluor 555 goat anti-rat IgG (1:500; Invitrogen), Alexa Fluor 568 goat anti-mouse IgG (1:500; Invitrogen), and Alexa Fluor 568 goat anti-rabbit IgG (1:500; Invitrogen) antibodies. Filamentous actin was stained with Alexa Fluor 647-conjugated phalloidin (1:500; Invitrogen). Embryos were mounted in ProLong Gold (Molecular Probes) between two coverslips. Stained embryos were imaged on an Olympus FV3000 laser-scanning confocal microscope with a 60× oil-immersion lens (NA 1.35; Olympus), and FluoView software (Olympus). 16-bit Z-stacks were acquired at 0.5 µm or 0.2 µm steps (21–50 slices per stack).

Image analysis was performed using PyJAMAS (Fernandez-Gonzalez et al., 2022), an image analysis platform developed by our laboratory, and custom scripts (available at https://www.quantmorph.ca or upon request) written in Python or in MATLAB (Mathworks). To analyse wound-closure phenotypes, we traced wounds using the semi-automated LiveWire method that uses Dijkstra's optimal path search algorithm to find and trace the brightest path of pixels between two manually selected points (Dijkstra, 1959).

To create kymographs of the wound edge, annotations from the first 20 min after wounding were circularly shifted until the start points were aligned. The intensity signal along the wound edge was interpolated to obtain 1500 equally spaced points. The intensity signal was corrected for background intensity and photobleaching before being smoothed with a Gaussian kernel with a size of 5 pixels (1.33 μm). Each row of the kymograph displays the intensity of the wound edge at a single time point. Three-dimensional reconstructions of E-cadherin and Talin were created using the ClearVolume plugin (Royer et al., 2015) for Fiji (Schindelin et al., 2012).

Annotations of TCJs along the wound edge were generated with a semi-automated machine-learning method. A support vector machine was trained using PyJAMAS on positive and negative training sets of one hundred 30×30-pixel images each. The positive training set contained images of different TCJs imaged in embryos expressing E-cadherin:tdTomato at various stages of wound closure. The negative training set was created from background regions, bicellular junctions away from the wound, and the auto-fluorescent scars from the wounding laser on the vitelline membrane. The support vector machine was then applied to classify 30×30-pixel regions (5.3×5.3 μm²) of the image as either positive or negative for TCJ detection. The detected TCJs were filtered by first applying non-maximum suppression to include only the top 300 regions with highest confidence, followed by removing any regions not intersecting the wound edge. An average of seven to eight TCJs were detected per frame and classification accuracy after non-maximum suppression and restriction to the wound edge was 83% true positive. Detected TCJs were automatically tracked in time based on the closest Euclidean distance. The position of the TCJ was established at the pixel of maximum intensity within the 30×30-pixel region. Automated TCJ positions were accurate to 1.8±0.1 μm with respect to manual annotations. Automated TCJ positions were visually inspected and manually corrected as necessary. The TCJ detection pipeline, including the support vector machine classifier, is available at https://bitbucket.org/raymond_hawkins_utor/tricell_automation_plugin/.

To measure BCJ intensity over time, individual interfaces between wounded and adjacent cells were delineated as segments of the identified wound edge between two adjacent TCJs (Zulueta-Coarasa et al., 2014). Each interface was divided into 1000 evenly spaced points and fluorescence was quantified using linear interpolation. The BCJ intensity was calculated as the mean of the central 200 points. Intensity values were corrected and normalized as above. To calculate the maximum intensity change, we obtained the minimum BCJ intensity within 5–15 min after wounding and normalized to the intensity immediately prior to wounding.

To measure retraction velocity after laser ablation, the positions of the two tricellular vertices connected by the ablated structure (wound edge segment or cell–cell junction) were manually tracked in PyJAMAS. We quantified the retraction velocity of TCJs after ablation as a proxy for mechanical tension (Hutson et al., 2003; Zulueta-Coarasa and Fernandez-Gonzalez, 2015). We used a Kelvin–Voigt mechanical equivalent circuit, which models junctions as a combination of a spring and a dashpot configured in parallel, to estimate the viscosity-to-elasticity ratio (Zulueta-Coarasa and Fernandez-Gonzalez, 2015).

The number of samples analysed corresponds to the reported n value in all cases. When multiple samples were collected from a single embryo, we verified that the variance within embryos was at least as large as the variance across embryos. To measure the significance of changes across two unpaired samples, we used a non-parametric Mann–Whitney test. To measure the significance of change of a single sample relative to zero, we used a one-sample Wilcoxon signed-rank test. A two-way analysis of variance (ANOVA) was conducted to examine the effects of treatment group and wound perimeter size on the actin accumulation and heterogeneity at the wound edge. We then used post-hoc tests including Tukey's honestly significant difference (HSD) test and Mann–Whitney to determine significant difference between and within groups. In plots, shaded areas (boxes) show the s.e.m., bars indicate the s.d., and the central line is the mean. In time-course plots, error bars represent the s.e.m. ns, not significant, *P<0.05, **P<0.01, ***P<0.001.

We thank Guy Tanentzapf for reagents. Flybase provided important information for this study. We are grateful to Ana Maria do Carmo and Gordana Scepanovic for comments on the manuscript.

The peer review history is available online at https://journals.biologists.com/jcs/lookup/doi/10.1242/jcs.261138.reviewer-comments.pdf