Automated AI algorithms enable us to quantify each of the three cell behaviours that contribute to wound closure

We tracked and quantified the three cell behaviours that contribute to wound re-epithelialisation – cell migration, cell shape changes and cell divisions – by means of automated algorithms. We used the plugin TrackMate to track the nuclei of cells over time in 3D utilising the Histone2-RFP channel data (Tinevez et al., 2017). Cell migration was measured by averaging the radial component of the velocity of a cell’s nucleus at given times after wounding and at different distances from the leading edge (Fig. 1D). Next, we applied the first of our deep learning models, U-NetBoundary (Turley et al., 2023), to detect cell boundaries (from the E-cadherin-GFP channel), and thus identify individual cells (Fig. 1E). Cell elongation relative to the wound was calculated by computing individual q-tensors for each cell from their cell boundaries (Olenik et al., 2023). These q-tensors comprise a component for cell elongation and another for a cell’s orientation (see Methods for further details). A heatmap of cell elongation relative to the wound indicates which cells are elongated perpendicular to (pointing towards) the wound edge in red and those aligned along the wound margin appear blue (Fig. 1B’). To quantify the shape of cells adjacent to the wound, we averaged the cells’ q-tensors, binning them into groups of cells as a function of distance from the wound, and the time after wounding.

Similarly, we detected cell divisions within the wounded epithelium with a deep learning model called U-NetCellDivision (Turley et al., 2023), which utilises dynamic cell boundary and nuclear information from the GFP and RFP channels (Fig. 1F). We binned these dividing cells into radial bands (annuli) extending out from the wound edge, and for various time points post wounding, to determine division density for each tissue annulus back from the wound edge.

Epithelial cells extending several cell diameters back from the leading edge contribute to wound closure

To determine how wound size impacts the cell biology of repair, we generated wounds of two different sizes; large wounds had an initial area between 700-1100μm², and small wounds were between 200-400μm² (Turley et al., 2023). The large wounds took, on average, 50mins to close to 20% of their initial size, and small wounds healed the same amount in 25mins (Fig. 2A). At this late stage of wound closure, both wound sizes have leading edge cells that make contact with one another to seal the wound closed (Abreu-Blanco et al., 2012; Wood et al., 2002). This last part of wound closure is visually very noisy, and the leading edge becomes unclear. Hence, we define a wound as closed at this point as the majority of re-epithelialisation has occurred (see Fig.2B-B” for time course of a large wound healing and Supplemental Movies S1 and S2 for repair of small and large wounds, respectively).

Open in new tabFigure 2: Quantifying cell migration, shape changes and divisions around small and large wounds

A) Quantification of the wound area over time for both small and large wounds. The red dashed line is the 9-10 minute point where we classify the size of wounds. B-B”) A large wound as visualised over 3 timepoints 5mins, 25mins, and 45mins. Yellow shading highlights non-epithelial tissue in images to illustrate wound closure over time. C,F) Spatio-temporal heatmaps of cell velocity relative to wound centre for small and large wounds, respectively. Blue regions indicate cells are migrating away from wounds and red cells are migrating towards the wound. D,G) Spatio-temporal heatmaps of cell shape elongation relative to wounds for small and large wounds, respectively. Blue regions indicate that cells are elongated perpendicular to wounds and red cells are oriented towards the wound. E,H) Heatmaps of the division density for small and large wounds. (n=8 small wounds and n=9 large wounds)

We next quantified the spatiotemporal distribution of each of the three contributing cell behaviours during wound closure. Our heatmaps indicate the mean velocity of cell nuclei towards the wound as a function of distance from the wound edge and time after wounding, with red highlighting cells migrating towards the wound and blue away from it (Fig. 2C, F). For small wounds, cell rows extending up to 50μm (or 10+ rows of cells), back from the leading edge, migrate toward the wound; this migration persists throughout the 25 mins of healing time. For larger wounds, cells migrate faster and for a longer time, which is as expected due to the longer distance needed to be travelled for effective closure of these wounds (Fig. 2C, F). Interestingly, there are two distinct phases of migration observed during closure of large wounds; for the first (approximately 30 mins post-wounding) cells migrate rapidly, but subsequently migration slows over the next ≈15 minutes as the wound closes (Fig. 2C, F).

Next, we examined the cell shape changes during wound closure. In the cell shape heatmaps we calculated the mean q-tensor of the cells for each distance from wound and time after wounding. This was used to determine cell elongation relative to the wound, with cells elongating towards the wound centre indicated by red in the heatmap, and those elongating along the wound edge highlighted in blue (Fig. 2D, G, and see Methods for details). The q-tensor is a dimensionless quantity, and we have normalised its value using the maximum absolute value of cell elongation around large wounds. For small wounds, cells at the leading edge, and one row back from this edge, rapidly elongated perpendicular to the wound margin (Fig. 2D); these cells subsequently became rounder and were no longer elongated after 10 minutes; this shape reversion contributes to wound closure. In large wounds too, cell elongation occured in leading edge cells after wounding. But here, cell elongation propagated back much further into the neighbouring tissue, with cells up to 30μm (approx. 7-8 rows) back from the leading edge affected (Fig. 2G). Similar to small wounds, cells rapidly reverted back to a rounder shape such that by 25 mins post wounding the mean cell shape had returned to the unwounded tissue average (Fig. 2G). This initial period of cell shape change aligns with the early ‘fast phase’ of cell migration (Fig. 2F).

A synchronised surge in cell divisions occurs after wounding in a zone back from the leading edge

The third cell behaviour contributing to wound re-epithelialisation is cell division. We quantified the density of proliferation around wounds as a function of space and time, and this data is presented as heatmaps in (Fig. 2E, H and (Turley et al., 2023)). During the initial period, when cells are rapidly migrating and changing their shapes, we see a clear suppression of cell division in the wound edge epithelium, reaching a minimum at 60mins. This suggests that cell proliferation and the other contributing wound cell behaviours might be mutually exclusive, which is also suggested in studies of mammalian wounds (Aragona et al., 2017; Park et al., 2017). For small wounds, the zone of suppressed cell divisions extends 30μm back from the wound edge and lasts until 80 minutes after wounding (Turley et al., 2023). In larger wounds, this suppression of cell divisions has a similar time course, but extends over a much larger area, stretching back over 100μm (>25 cell rows) back from the wound, beyond the field of view. After these periods of reduced cell division, we observe a synchronised burst of proliferation beginning 90 minutes after wounding and lasting for an hour (Turley et al., 2023). This occurs in an annulus ranging from 20 to 70μm back from the small wound sites and 20 to >100μm for large wounds (Fig. 2E, H). This surge of proliferation could be a consequence of synchronised wound signals inducing cells to divide to replenish cells lost during wounding, and/or cells that would have divided if not restricted by wounding becoming synchronised as suppression ends.

A rapid calcium wave travels back through the wounded epithelium that is required for optimal wound-induced cell migration, cell shape changes and cell divisions

After characterising re-epithelialisation of wounds in wild-type pupal wing tissue, we generated transgenic flies in which key wound healing signals had been perturbed. We began by blocking the well-characterised calcium wave, which is the first damage signal released after tissue injury (Razzell et al., 2013; Weavers et al., 2019). This was achieved by expressing trpm^RNAi to knock down the Trpm calcium channels through which this signal propagates (Razzell et al., 2013); this effectively inhibits the spread of the wound Ca²⁺ wave (Fig. 3E,E’). As ubiquitous expression of trpm^RNAi is partially lethal we temporally controlled its expression using the Gal4-UAS system with a temperature sensitive Gal80^ts (McGuire et al., 2004). We kept embryos and larvae at 18°C during early development (to prohibit RNAi expression before pupal stages) and raised the temperature to 29°C once at the pre-pupal stage to activate trpm^RNAi expression. Since flies develop faster at higher temperatures, pupae were raised for 14.75hours at 29°C (instead of 18hours at 25°C) to achieve an equivalent level of development. To confirm consistency of developmental stage, we used cell elongation as a proxy for wing development (Athilingam et al., 2021; Etournay et al., 2015), and found no significant difference between the temperature shift paradigms (Fig. S1).

Open in new tabFigure 3: Blocking the wound calcium wave and quantifying the resulting changes in cell behaviours

A) Spatio-temporal heatmaps of cell velocity relative to wounds for control 14.75h APF large wounds. Blue regions indicate those cells that are migrating away from the wound and red those migrating towards the wound. B) Spatio-temporal heatmaps of cell shape elongation relative to wounds for control 14.75h APF large wounds. Blue regions indicate cells elongated perpendicular to the wound and red are cells oriented towards the wound. C) Heatmaps of cell division density for control 14.75h APF large wounds. D) Quantification of the wound area over time as large wounds close. Red dashed line is the 9-10 minute point where we classify the size of wounds. E-E’) Still images from gCAMP movies at 0 and 26 sec post wounding of control versus Trpm^RNAi wing tissues to show blocked calcium wave around Trpm^RNAi wounds. White asterix indicates location of wound. F) Spatio-temporal heatmaps of cell velocity relative to wound centre for large wounds. Blue regions indicate cells that are migrating away from wounds and red are those migrating towards the wound. G) Spatio-temporal heatmaps of cell shape elongation relative to wounds for large wounds. Blue regions indicate cells that are elongated perpendicular to wounds and red are cells oriented towards the wound. H) Heatmaps of cell division density for large wounds. I) Spatio-temporal heatmaps of the change in cell velocity for control versus wounds with a blocked calcium wave. J) Spatio-temporal heatmaps of the change in cell shape elongation for control versus calcium wave blocked wounds. K) Heatmaps of the change in division density for control versus calcium wave blocked wounds. (n=5 control 14.75h APF wounds, and n=4 Trpm^RNA’ wounds). Scale bars are 50μm.

We first quantified the area of wounds over time by comparing control (which had been raised with the same temperature shifts) and trpm^RNAi fly lines and observed a slight reduction in the healing rate (Fig. 3D). We went on to investigate whether this gross defect in healing might be linked to any particular changes in the contributing cell behaviour(s). The control (14.75h APF) pupae exhibited a typical wound response with regards to cell migration and shape changes and a clear (delayed) burst of divisions back from the leading edge (Fig. 3A-C). However, all three cell behaviours were altered in trpm^RNAi expressing pupae (Fig. 3F-H). Epithelial cell migration at the wound edge was initially slower for 30 mins after wounding, but migration subsequently persisted for longer than in control wounds, perhaps to compensate for the initially slow start (Fig. 3I). Cell shape changes were also altered, with cells at the leading edge remaining highly elongated for a more extended period (Fig. 3G, J). These two changes in cell behaviour could explain the slower healing of wounds lacking a fully functional calcium wave. We also observe that the synchronised burst of proliferation we observe in wild-type wounds is considerably reduced to approximately half the number of dividing cells in wounded trpm^RNAi pupae (Fig. 3H, K). For all of these genetic perturbation experiments we measured the effect on large wounds since they generally displayed a more extensive cell behavioural response (see earlier), and so we reasoned that any differences after genetic manipulation might be clearer.

JNK signalling perturbation primarily impacts wound closure by inhibiting changes in cell shape

Next, we disrupted JNK signalling, which is known to be key in developmental morphogenetic episodes such as dorsal closure (Tafesh-Edwards & Eleftherianos, 2020), and in wound healing, both in the formation of actomyosin cables and for associated cell shape changes (Kwon et al., 2010; Lee et al., 2019).

JNK signalling was blocked by expressing a dominant negative mutant version of Drosophila JNK (termed basket, Bsk^DN) in the pupal epithelium as previously utilised for inhibiting JNK signalling (Lee et al., 2019; Tafesh-Edwards & Eleftherianos, 2020; Weavers et al., 2019). We showed that this knockdown was effective by live imaging JNK activity via TRE-DsRed, a transgenic reporter that indicates active JNK signalling, around control and Bsk^DN laser wounds (Fig. 4B,B’). JNK signalling is known to be downstream of the Ca²⁺ wave (Weavers et al., 2019) and so, unsurprisingly, we see a similar, but not identical effect on aspects of wound closure. Again, there was a reduction in overall rate of wound closure with wounds taking 80 minutes to close to 20%, which is half an hour more than for wild type pupal wounds (Fig. (4A)). However, whilst our automated tracking data revealed that rates of cell migration in Bsk^DN pupae were similar to WT tissues (Fig. 4C, F), cells in the wound epithelium appeared unable to effectively change their shape (Fig. 4D, G). Cell elongation around the wound margin failed to propagate back through the tissue. Cells at the leading edge did elongate initially, similar to wild type wounds, but this elongation then took 80 minutes to resolve, rather than only 25 mins in control pupae. This failure of cells to change shape, and thus not contribute to wound closure could be a major reason for the slower healing in pupae with JNK signalling blocked. We also observed a loss of post-wound suppression of cell divisions after JNK knockdown from 30-100mins, as previously described for larval cell death lesions (Cosolo et al., 2019). Our data indicate that while the timing of the wound induced cell proliferative burst is similar to controls the number of dividing cells is reduced (Fig. 4E, H) although to a lesser degree than for the trpm^RNAi pupae.

Open in new tabFigure 4: Knockdown of JNK signalling and quantification of how this alters cell behaviours in wound healing

A) Quantification of the wound area over time for control versus JNK knockdown tissues. Red dashed line is the 9-10 minute point where we classify the size of wounds. B-B’) JNK signalling reporter TRE-DsRed in control versus Bsk^DN wing tissues at 5hours post wounding to show dampened JNK signalling around the Bsk^DN wound. White asterix indicates location of wound. C) Spatio-temporal heatmaps of cell velocity relative to wounds for large wounds. Blue regions indicate cells that are migrating away from the wound and red are cells migrating towards the wound. D) Spatio-temporal heatmaps of cell shape elongation relative to wounds for large wounds. Blue regions indicate cells that are elongated perpendicular to wounds and red are cells oriented towards the wound. E) Heatmaps of the division density for wounds. F) Spatio-temporal heatmaps of the change in cell velocity for control versus JNK knockdown wounds. G) Spatio-temporal heatmaps of the change in cell shape elongation for control versus JNK knockdown wounds. H) Heatmaps of the change in cell division density for control versus JNK knockdown wounds. (n=5 control 14.75h APF wounds and n=6 Bsk^DN wounds). Scale bars are 50μm.

Blocking the wound inflammatory response reduces both cell migration and cell proliferation

Lastly, we ablated the flies’ innate immune cells (termed hemocytes) to block the wound inflammatory response (Fig.5). The inflammatory response is believed, through studies in various model organisms, to both clear away wound debris but also release signals that aid in orchestrating various aspects of wound closure including wound re-epithelialisation (Antsiferova & Werner, 2012; Weavers & Martin, 2020). Hemocytes were ablated in pupae by expressing the pro-apoptotic gene, reaper (rpr) specifically in hemocytes using srp-Gal4 in combination with a temperature sensitive Gal80^ts, similar to that performed previously (Weavers et al., 2016); for this genetic manipulation, the shift to 29°C for 14.75hours (to reach an equivalent developmental stage to 18h APF at 25°C) was lethal, and so, instead, we maintained pupae at 18°C for 21hours then shifted them to 29°C for 5hours. Once again, we confirmed there were no statistical differences in developmental stage of control versus “immune ablated” wings (Fig. S1).

Open in new tabFigure 5: Ablation of immune cells reduces cell migration and decrease divisions globally

A) Spatio-temporal heatmaps of cell velocity relative to wounds for control 26h APF large wounds. Blue regions indicate cells that are migrating away from wounds and red are cells migrating towards the wound. B) Spatio-temporal heatmaps of cell shape elongation relative to wounds for control 26h APF large wounds. Blue regions indicate that cells are elongated perpendicular to wounds and red are cells oriented towards the wound. C) Heatmaps of the division density for control 26h APF large wounds. D) Quantification of the wound area over time for wounds in control versus immune ablated tissues. Red dashed line is the 9-10 minute point where we classify the size of wounds. E-E’) Red-stinger nuclei of innate immune cells recruited to a control wound but absent from immune cell ablated wing tissues. White asterix indicates location of wound. F) Spatio-temporal heatmaps of cell velocity relative to wounds. Blue regions indicate cells that are migrating away from wounds and red cells are migrating towards the wound. G) Spatio-temporal heatmaps of change in cell shape elongation relative to wounds for large wounds. Blue regions indicate cells that are elongated perpendicular to wounds and red are cells oriented towards the wound. H) Heatmaps of the spatial division densities for wounds. I) Spatio-temporal heatmaps of the change in cell velocity for control versus immune ablated wounds. J) Spatio-temporal heatmaps of the change in cell shape elongation for control versus immune cell ablated wounds. K) Heatmaps of the change in cell division density for control versus immune ablated wound. (n=5 control, 26h APF wounds and n=5 rpr wounds). Scale bars are 50μm.

We confirmed that this genetic perturbation successfully killed the pupal hemocytes and so prevented the standard immune response to wounds (Fig. 5B, B’). We find that wounds in these immune ablated pupae are also slower to heal, but the contributing disruptions in cell behaviour are subtely different to those where calcium signalling or JNK signalling are blocked (Fig. 5A). This time, cell shape changes remained unaffected, whereas we see that cell migration is severely hindered. We observe almost no cell migration for the first 40 mins and a slow mean nuclei velocity even after 45 mins, by which time control wounds have ceased their cell migration (Fig. 5C, F). In contrast to our studies of repair in JNK knockdown wounds, we observed that cell shape changes are relatively normal, with little difference between control and immune ablation flies (Fig. 5D, G). Wound-induced cell proliferation in these pupae shows the same spatial temporal pattern to that of control wounds but with a clear overall reduction in divisions (Fig. 5E, H). These results indicate that signals from hemocytes appear to be needed for both epithelial cell migration and to stimulate cell division also.