Spatiotemporal regulation of cell fusion by JNK and JAK/STAT signaling during Drosophila wound healing

Cell–cell fusion elicits dramatic changes in the genetics and physiology of the participating cells. Chromosomal instability and aberrant gene expression may occur as a result of cell fusion (Duelli and Lazebnik, 2007). Cell fusion thus needs to be under tight control to prevent spontaneous and, once initiated, excessive fusion. Cell fusion occurs widely in developmental and disease processes, including fertilization, muscle formation, bone resorption, infection and tumorigenesis (Chen and Olson, 2005; Shinn-Thomas and Mohler, 2011; Aguilar et al., 2013). However, no general picture for cell fusion has emerged to date and the mechanisms by which cell fusion is initiated, and at the same time confined to specific cells or tissue areas, has not been well studied.

Cell fusion occurs during wound healing of the Drosophila larval and adult epidermis (Galko and Krasnow, 2004; Lesch et al., 2010; Losick et al., 2013). Here, cell fusion was proposed to help clear cellular debris (Galko and Krasnow, 2004) and, along with cell migration and growth, contribute to wound closure (Losick et al., 2013). Cell fusion is promoted by JNK pathway (JNK is also known as Basket in flies) activation, which appears to be mediated by disassembly of the integrin focal adhesion complex (Wang et al., 2015). Cell fusion is also under the control of the Hippo pathway transcription factor Yki (Losick et al., 2013). More generally, the genetic mechanism of cell–cell fusion has been well characterized during Drosophila muscle differentiation. There, the interaction of cell surface receptors between two cells of different types leads to the formation of actin-based pre-fusion complexes in a manner that is mediated by Rac protein (reviewed in Rochlin et al., 2010). Here, we used Drosophila larval wound healing as a model to investigate the regulation of cell fusion because epidermal cells at this stage are huge (∼50 μm in diameter) due to extensive endoreplication, and are amenable to both cytological analysis and various conventional fly genetic analyses (Galko and Krasnow, 2004; Kwon et al., 2010).

The molecular mechanism of Drosophila wound healing is highly conserved in relation to its mammalian counterpart (Li et al., 2003; Galko and Krasnow, 2004; Mace et al., 2005; Ting et al., 2005; Belacortu and Paricio, 2011), and is analogous to dorsal closure, an epithelial cell sheet movement during embryonic development (Martin and Parkhurst, 2004; Razzell et al., 2014). For re-epithelialization, JNK is the key molecule that is absolutely required although, depending on developmental stage, noncanonical components may function in the pathway (Campos et al., 2010; Lesch et al., 2010). For regeneration of the overlying cuticle layer, the transcription factor Grh is known to be essential in the embryo. Grh induces numerous downstream target genes, including the genes encoding cuticle-synthesizing enzymes (Mace et al., 2005; Pearson et al., 2009).

The JAK/STAT signaling pathway exerts many conserved functions from insects to mammals, which include inflammation, cell proliferation, cell migration and stem cell maintenance (Hou et al., 2002; Arbouzova and Zeidler, 2006; Hombria and Sotillos, 2013; Chen et al., 2014). In Drosophila, the cytokine-like Upd ligands bind to the transmembrane receptor Dome and this event recruits Hop, the Drosophila JAK, which phosphorylates and activates Stat92E, the sole STAT transcription factor in Drosophila. JAK/STAT signaling is crucial for the migration of border cells during oogenesis (Silver and Montell, 2001) and is required for body segmentation during embryogenesis (Hou et al., 1996; Yan et al., 1996; Hombria and Sotillos, 2013). Without that pathway activation, both body patterning and muscle development are severely defective. Whether JAK/STAT is directly required for cell fusion during muscle development is unknown. JAK/STAT pathway activation is also crucial for regenerative cell proliferation in the Drosophila midgut and the mammalian liver (Sun and Irvine, 2014; Katsuyama et al., 2015), but whether the pathway has different roles in non-proliferative tissues during wound healing and regeneration is not known.

Here, we address the question of how cell fusion is regulated. Specifically, we investigated the mechanisms underlying the confinement of cell fusion to a specific area. We describe a novel role of the JAK/STAT pathway in cell fusion during Drosophila larval wound healing and provide evidence of spatiotemporal control of JNK and JAK/STAT signaling pathways, each acting counter to the other on cell fusion. Execution of cell fusion is thus determined based on the cellular balance between the activities of these two pathways.

To identify genes involved in Drosophila wound healing, we screened transgenic RNA interference (RNAi) lines (RNAi expression lines in this paper are denoted by an appended ‘-i') (Kwon et al., 2010). Components of the JAK/STAT pathway were leading candidates and noteworthy because the role of this pathway in cell migration has been well characterized (Silver and Montell, 2001; Hou et al., 2002). Larvae in which dome, hop, or Stat92E were knocked down via the larval epidermis-specific A58-GAL4 driver were able to heal a pinch wound of ∼30–40 cells in width (0.061±0.016 mm²; mean±s.d.; Fig. 1A–D). We noticed, however, that these animals frequently exhibited multinucleate syncytia that were larger than those of control larvae, as demonstrated by immunostaining for the septate junction protein Fasciclin III (FasIII) (Fig. 1D; data not shown).

To confirm the cell fusion phenotype and exclude the possibility of an off-target effect of RNAi (Mohr and Perrimon, 2012), we overexpressed a dominant-negative form of Dome (designated dome^DN) (Brown et al., 2001) in the larval epidermis and found that these larvae again developed large syncytia upon wounding (Fig. 1E,F). We next overexpressed dome^DN in the posterior compartment of each body segment via engrailed-GAL4 (en-GAL4) and made a wound selectively in an anterior (wild-type) or a posterior (JAK/STAT signaling-deficient) compartment. Large syncytia were observed only in wounded posterior compartments (Fig. 1G,H). Larvae heterozygous for hop (which encodes Drosophila JAK) showed a slightly increased level of cell fusion, compared to that seen in wild type, in wounded epidermis (Figs 1I and 5L). In contrast, hop hemizygous larvae displayed huge syncytia, similar to those in dome^DN-expressing larvae (Fig. 1J). Constitutive activation of the JAK/STAT pathway either by overexpression of upd1 or introduction of the hop^Tum allele (which is constitutively active) (Harrison et al., 1995; Luo et al., 1995), however, did not completely block cell fusion (Fig. 1K,L), indicating that a low level of cell fusion that is independent of JAK/STAT signaling also exists.

To verify that loss of FasIII immunostaining from cell boundaries indeed reflected a cell–cell fusion event, we employed two different approaches. First, we stained the plasma membrane with the lipophilic dye FM 1-43Fx, and confirmed excess cell fusion in dome^DN-expressing larvae relative to control larvae (Fig. 1M,N). Second, we examined cytoplasmic diffusion of GFP from GFP-expressing cells to non-GFP-expressing fused cells (Podbilewicz et al., 2006; Grote, 2008). For this analysis, Stat92E was knocked down in the posterior half of each body segment via en-GAL4 and this region was labeled with nuclear GFP (we used nuclear GFP instead of cytoplasmic GFP because the former might be better in delineating the position of individual cells before fusion). In Fig. 1O,P, a wound mark spans the compartmental boundary between the anterior and posterior halves of a body segment, and cell fusions were enriched in the posterior half (GFP-positive) where there is expression of Stat92E-i. We frequently observed, in this condition, that syncytial nuclei derived from the non-GFP-expressing anterior half were labeled with GFP, which is consistent with the idea that GFP diffuses from GFP-expressing cells to non-GFP-expressing cells after cell fusion.

We initially developed three different methods to quantify cell fusion (Fig. 2A–C). In the Fusion_sync method (Fig. 2B), we counted the number of syncytial cells containing two or more nuclei in a wounded ‘trapezoidal area’ (which was demarcated by tendon cells in the dorsal side of a body segment; Kwon et al., 2010). In the Fusion_ntot methods (Fig. 2C), we determined the total number of nuclei in all syncytia formed within a wounded trapezoid. However, hereafter we chose to use the Fusion_nmax method, in which the largest number of nuclei in a single syncytium was recorded. Fusion_nmax quantifies excessive cell fusion more accurately than the other two methods because it precludes the counting of frequent fusions of small numbers of cells that occur independently of experimental wounding. Wounded epidermis of dome^DN-expressing larvae contained an average of 28 nuclei in the largest syncytium, compared to 9 in control larvae (Fig. 2A). Based on these results, we hereafter define the term ‘excess cell fusion’ as the formation of a syncytium containing 15 or more cell nuclei. Excess cell fusion was observed in 21% of wounded control samples on average, whereas the number was 90% or higher in dome-i or Stat92E-i samples (Fig. 2D). Cell fusion peaked ∼16 h after injury and then decreased slightly in wild-type larvae, suggesting that some fused cells might later recellularize (Fig. 2E).

Since JAK/STAT activation was required for suppressing excess fusion in wounded epidermis, we next asked which Upd ligands were important in this process. Upon wounding, transcripts of upd1, upd2 and upd3 were increased by ∼2-, ∼14- and ∼11-fold, respectively, compared to unwounded epidermis, as measured by quantitative real-time PCR (qPCR) (Fig. 2F). We then analyzed cell fusion phenotypes in RNAi knockdown larvae for the three upd genes, in each case using two different RNAi lines. One line for upd2 and the two for upd3, among the six RNAi lines driven via the epidermal GAL4, displayed excess cell fusion phenotypes that were at least 2-fold stronger than that of control larvae. One of the two upd3 RNAi lines showed the highest level of excess fusion, which was comparable to those of dome or Stat92E knockdown (Fig. 2D,G). Based on gene induction and phenotypic strength, we conclude that Upd2 and Upd3 function together in suppressing excess cell fusion.

We examined whether the excess cell fusion was due to a secondary consequence of altered cell proliferation or cell death during wound healing. We first stained the wounded epidermis of control and dome^DN larvae using antibody against phosphorylated histone H3 (phospho-H3) to analyze mitosis (Hans and Dimitrov, 2001). Neither of the samples exhibited any signs of mitosis at 8 h after wounding, whereas the third-instar larval wing disc displayed a strong mitotic signal (Fig. S1A–C′). We next stained wounded epidermis using anti-cleaved caspase 3 (CC3) antibody to analyze apoptotic cell death (Fan and Bergmann, 2010). Wounding increased CC3 staining in wounded segments of both control and dome^DN larvae at 8 h after injury, compared to that seen in neighboring unwounded segments. Nonetheless, the staining patterns of the larvae of the two genotypes did not show any distinct differences (Fig. S1D–H′). These data indicate that the excess fusion in dome^DN larvae after wounding was not due to changes in mitosis or apoptosis.

To carefully assess a potential role of the JAK/STAT pathway in cell migration, we quantified cell shape changes during wound healing using the method we previously developed (Kwon et al., 2010). A filet of unwounded epidermis contains cells that are largely pentagonal or hexagonal with sides that are relatively linear (Kwon et al., 2010). During wound closure, cells close to the wound leading edge normally assume irregular forms with curved edges, whereas in JNK-deficient larvae, these changes are drastically reduced (Kwon et al., 2010). In control larvae, these irregular cells accounted for 79±2% (mean±s.e.m.) of all cells in the dorsal trapezoid of a wounded segment, whereas in dome^DN-expressing larvae, irregular cells accounted for 63±5% of all cells (79.8% of control level; Fig. S2A,B,D). Conversely, irregularly shaped cells in upd1-overexpressing larvae accounted for 85±2% (107.6% of control level; Fig. S2C,D). Although not as strong as in JNK-deficient larvae (Kwon et al., 2010), these phenotypic results indicate that cell shape change during wound healing is also mediated in part by the JAK/STAT pathway.

We wondered whether the reduced level of cell shape change in the dome^DN-expressing larvae was an indirect consequence of large syncytia, which might impede shape change. To address this question, we knocked down dome in the posterior half of the segment using en-GAL4 and wounded wild-type cells in the anterior half, to avoid excessive cell fusion, and then analyzed shape changes in both halves of the segment (Fig. S2E–G). In control larvae, the extent of shape change at 20 h after injury was indistinguishable in the anterior and posterior halves, both of which contained wild-type cells (Fig. S2E,G). In larvae expressing dome RNAi via en-GAL4 (denoted en>dome-i), however, the posterior (dome knockdown) half of the wounded segment still exhibited a significantly reduced level of shape change (52.7% of control level), despite the absence of excessive cell fusion (Fig. S2F,G). We conclude that the JAK/STAT pathway was required in part for cell shape change during wound healing, and that this was independent of its effect on cell fusion. We next considered the opposite argument, namely that the excessive cell fusion in JAK/STAT-deficient larvae might be a consequence of defects in cell shape change; we thought this case unlikely. Larvae with defects in wound closure, including those deficient in JNK, Pak3 or non-muscle myosin II, do not generally display excessive cell fusion during wound healing (Galko and Krasnow, 2004; Kwon et al., 2010; Baek et al., 2012).

We examined the time and location of JAK/STAT pathway activation in the epidermis during wound healing. We used 10×STAT-GFP as a reporter, based on our own and previous demonstrations that expression of this transgene faithfully reflects the activation state of the JAK/STAT pathway (Fig. S3A–E; Bach et al., 2007). As a reference, JNK activation was also monitored by using an msn-lacZ reporter (Msn is a downstream transducer of the JNK signal); JNK pathway activation in the epidermis mediates wound closure and induces cell fusion (Galko and Krasnow, 2004; Wang et al., 2015; Fig. S3J–Q). For careful observation, we removed, from epidermal filets, underlying muscle fibers that exhibited JAK/STAT signaling activation (see below; Yang et al., 2015). JNK activation was prominent at 4 h after injury (Fig. 3A–D′), peaked at ∼8 h (Fig. 3E–H′) and then gradually decreased, as demonstrated by β-galactosidase staining (Fig. 3I–Q). JAK/STAT signaling activation became apparent after 12 h, and the GFP level was still increasing at 20 h after injury (Fig. 3C,G,K,O,Q).

We next examined spatial activation patterns of JNK and JAK/STAT signaling using the same reporters. At early stages, when wound holes were still open (4 to 12 h after injury), JNK signaling was activated in cells located within a distance of approximately three cell diameters from the wound margin (Fig. 3A–L′). Interestingly, JAK/STAT signaling was gradually activated in cells located in an adjoining concentric ring (three to four cell diameters wide) that showed minimal overlap with the central JNK-active domain by 8 h after injury (Fig. 3E–H′). At 20 h after injury, when the wound was healed completely, the JNK level was low in the wound vicinity, while the JAK/STAT level was high, both in the formerly JNK-active central domain and the peripheral ring (Fig. 3M–P′).

We examined all individual cells in a wounded trapezoid to determine the relationships among the activation states of JNK and JAK/STAT signaling and the cell fusion event. There was an inverse correlation between JNK and JAK/STAT activation (Fig. 3R), suggesting that the JNK and JAK/STAT pathways are mutually exclusive. There was also a tendency for fused cells to have relatively high levels of JNK and lower levels of JAK/STAT activity (Fig. 3R). These data corroborate our conclusion that JAK/STAT signaling suppresses excess cell fusion, and further, suggest that JNK may induce cell fusion via suppression of JAK/STAT pathway activation.

We hypothesized that JNK may inhibit JAK/STAT activation, and/or that JAK/STAT inhibits JNK activation, based on the observation that activation patterns of JNK and JAK/STAT signaling were temporally and spatially distinct and appeared to be mutually exclusive. We tested this hypothesis by manipulating the activity of one pathway and examining the effect on the other. First, we analyzed JAK/STAT pathway activation by analyzing expression of 10×STAT-GFP in bsk-i larvae at 8 h after injury, the time of peak JNK activation. Induction of 10×STAT-GFP in wounded bsk-i larvae was high (156% of control) in the two or three rows of cells surrounding the wound margin, where the reporter was not normally induced at this stage (Fig. 4A–C). These results suggest that JNK normally suppresses JAK/STAT signaling during wound healing. Second, we analyzed JAK/STAT pathway activation after activating the JNK pathway. We used the following two different methods for JNK activation: overexpression of egr (encoding the Drosophila TNF homolog; Igaki et al., 2002; Moreno et al., 2002), and overexpression of hep (encoding JNK kinase, an upstream activator of JNK) combined with knockdown of puc (puc encodes the JNK phosphatase, which inhibits JNK signaling). We confirmed that these two methods markedly increased msn-lacZ expression in both unwounded and wounded epidermis (Fig. S3L–Q). We also confirmed that both of the A58>egr and A58>hep puc-i larvae exhibited increased ratios of excess cell fusion after wounding (Fig. 4D; 12.5% for control, 53.3% for A58>egr, and 100% for A58>hep puc-i larvae). In A58>egr larvae, the JAK/STAT activation level was moderately increased without wounding (see Discussion). Wounding, however, did not increase JAK/STAT activation significantly at 20 h after injury, the time point at which JAK/STAT activation was normally high (Fig. 4E,F,H; 76% of the GFP fluorescence intensity of the GAL4-only wounded epidermis). A58>hep puc-i larvae also displayed increased JAK/STAT activation without wounding, which was comparable to that of wild-type wounded epidermis (Fig. 4G,H). Upon wounding, the activation level never increased beyond that of an unwounded segment (Fig. 4G,H), despite the fact that the underlying muscles, especially the muscle fibers around the wound hole, still displayed higher induction of 10×STAT-GFP reporter (Fig. S3F–I). These results indicate that JNK normally suppresses JAK/STAT activation in the vicinity of the wound.

We next examined the possibility that JAK/STAT signaling conversely affects JNK activation during wound healing. We manipulated JAK/STAT signaling activity by overexpressing either dome^DN or upd1 and examined JNK activity by using msn-lacZ. Overexpression of dome^DN or upd1 did not significantly alter JNK activation during wound healing (Fig. S4A–D). Overexpression of upd1 increased the number of JNK-active cells; however, the JNK activity level was not significantly higher in individual cells compared to that seen in control larvae (Fig. S4D,E). The significance of these results is currently unclear.

Because the two signaling pathways both function in cell fusion during wound healing, we wanted to know whether they work in a linear manner, i.e. whether JNK induces cell fusion only by inhibiting JAK/STAT signaling, or alternatively, whether JNK directly controls the fusion process in addition to suppressing JAK/STAT signaling. Because JNK activation, but not loss of JAK/STAT signaling, induced cell fusion in the absence of wounding (Wang et al., 2015; data not shown), the former hypothesis must be rejected. To verify this idea in the context of wound healing, we blocked JAK/STAT signaling by overexpressing dome^DN and simultaneously blocked JNK by overexpressing bsk^DN. Upon wounding, A58>bsk^DN dome^DN larvae showed no increase in cell fusion compared to A58>bsk^DN control larvae (Fig. 5A–E). Data for A58>dome^DN larvae at 8 h after injury (Fig. 5A) showed that the suppression of excess fusion by bsk^DN was not due to the presence of an open wound in bsk^DN-expressing larvae. Thus, we conclude that JNK functions as an instructive signal, and loss of JAK/STAT signaling functions as a permissive signal for cell fusion during wound healing.

We sought to examine the antagonistic function of the two pathways on cell fusion more directly. We tested whether JAK/STAT pathway activation could suppress JNK-induced cell fusion by overexpressing constitutively active hep (hep^CA) and hop together in a patch of dorsal epidermal cells (Wang et al., 2015). In control groups, hop overexpression did not induce cell fusion (Fig. 5F,I) and hep^CA overexpression, as expected, induced large fusions in unwounded larvae (a Fusion_nmax value greater than 80% for the number of hep^CA-expressing cells was obtained from 21 of 34 samples) (Fig. 5G,I). In hop hep^CA-expressing samples, however, the large fusions were dramatically reduced (Fig. 5H,I); FasIII-labeled cell boundaries, albeit thin, were clearly visible, in many cases, and the fusion values did not go beyond 40% in all of the samples.

If JNK and JAK/STAT indeed act antagonistically on cell fusion (Fig. 5J), epidermal cells have to determine whether to execute fusion by assessing the status of the balance between the activation states of the two pathways. If this is the case, we should be able to observe changes in the degree of cell fusion by slightly modulating the signaling activity of either pathway. When JAK/STAT activity was slightly reduced by making the hop locus heterozygous (hop³/+), cell fusion slightly increased accordingly compared to that seen in control larvae (Fig. 5K,L). When JAK/STAT activity was reduced by the same degree (hop³/+) in a background of egr overexpression, cell fusion was again increased compared to that seen in egr-overexpressing control (Fig. 5N,O). Surprisingly, cell fusion increased dramatically when JAK/STAT signaling was blocked by dome^DN overexpression in an egr overexpression background; patchy cell fusions extended approximately two segments away from the wound margin, which corresponded to the region in which msn-lacZ was induced in egr-overexpressing larvae (Fig. 5 M,N,P; Fig. S3O). These results strongly suggest that the relative levels of JNK and JAK/STAT signaling are sensed by epidermal cells and are used to determine whether to execute cell fusion during wound healing.

JAK/STAT signaling activation induces systemic responses in several different contexts. Parasitoid wasp infection activates JAK/STAT signaling in larval somatic muscles, which is crucial for an efficient cellular immune response (Yang et al., 2015). Wounding or tumor formation in imaginal discs induces secretion of Upd ligands and activates the systemic activation loop of JAK/STAT signaling in hemocytes and fat bodies, which results in hemocyte proliferation and suppression of tumor growth (Pastor-Pareja et al., 2008). We asked whether such systemic responses are also critical for suppression of excess cell fusion.

Epidermal wounding induced 10×STAT-GFP reporter expression in muscle fibers, analyzed at 20 h after injury (Fig. S3F,G). We blocked JAK/STAT signaling in the muscles by overexpressing dome^DN via Mef2-GAL4 and examined cell fusion. The larvae closed a wound hole and did not exhibit excess cell fusion, as examined at 20 h (Fig. 6A,H). We next asked the same question, but this time in regards to hemocytes and fat bodies. We blocked JAK/STAT signaling in hemocytes by overexpressing dome^DN via two copies of HmlΔ-GAL4 in the sensitized background of Stat92E heterozygote [denoted HmlΔ (×2)>dome^DN, Stat92E⁰⁶³⁴⁶/+]. These larvae showed normal wound healing, including a normal degree of cell fusion (Fig. 6B–D,H). We then genetically ablated larval hemocytes by using HmlΔ-GAL4 to drive hid (HmlΔ >hid) and examined wound healing at 20 h after wounding. These larvae again showed normal wound healing and a normal range of cell fusion (Fig. 6E,F,H). Finally, we blocked JAK/STAT signaling in fat bodies by using Lsp2-GAL4 to drive dome^DN (Lsp2>dome^DN) and observed in these larvae normal wound healing and a normal range of cell fusion (Fig. 6G,H). Taken together, we conclude that systemic activation of JAK/STAT signaling is not required for normal wound repair, including suppression of excess cell fusion.

We asked what mediates the JAK/STAT pathway activation in order to suppress cell fusion. It has been reported recently that the focal adhesion complex involving integrin protein suppresses cell fusion in the larval epidermis (Wang et al., 2015). Interestingly, JAK/STAT pathway activation is known to increase βPS integrin (also known as Myospheroid) protein expression in somatic stem cells of the testis (Issigonis et al., 2009). We therefore manipulated JAK/STAT activity in the epidermis and examined βPS protein expression via anti-βPS antibody staining. In unwounded epidermis, overexpression of dome^DN did not decrease the level of βPS protein below that of the control sample. By contrast, overexpression of hop markedly increased the βPS protein level (Fig. 7A–C,J; 145% of the GAL4-only control). βPS was mainly localized in the cytosol and the cell boundaries. We next examined βPS integrin expression in wounded epidermis. Wounding increased βPS in the wounded area compared to that seen in a control area in the neighboring unwounded segment (Fig. 7D,G,K; 214%). Overexpression of dome^DN decreased the degree of βPS induction (Fig. 7E,H,K; 159%), whereas overexpression of hop increased the degree of βPS induction, although still most prominently in the vicinity of the wound (Fig. 7F,I,K; 251%). We wanted to examine whether integrins suppressed cell fusion during wound healing. Knockdown of the βPS gene caused severe defects in reepithelialization, which interfered with cell fusion analysis (data not shown). We performed genetic interaction studies by making double heterozygotes for the βPS gene and Stat92E or upd3 genes. Single heterozygotes for amorphic alleles of βPS, Stat92E or upd3 all displayed small cell fusions that were within a normal range (Figs 2G and 7L,M). In contrast, double heterozygotes showed higher rates of cell fusion that were, although not always present, synergistic to those of the single heterozygotes (Fig. 7L,M). We thus propose that JAK/STAT signaling suppresses cell fusion partly via the induction of integrin proteins during wound healing.

Based on our findings, we propose that a wounded larval epidermis can be divided into two concentric domains, the central JNK-activation domain formed in the vicinity of the wound and the peripheral adjoining JAK/STAT-activation domain. Wounding induces JNK activation in the wound vicinity, which peaks at ∼8 h after injury. The three upd genes, particularly upd2 and upd3, are also highly induced at this stage, beginning to activate JAK/STAT signaling. The spatiotemporal activation pattern of JAK/STAT signaling will thus be shaped by distribution patterns of Upd ligands and JNK activation, the latter of which inhibits JAK/STAT activation. Consequently, JAK/STAT signaling is active in a donut-shaped ring that exhibits minimal overlap with the JNK-active domain. JNK induces cell fusion predominantly in the wound vicinity, where fusion-inducing JNK activity is robust and fusion-inhibitory JAK/STAT activity is low. Cell fusion rarely extends to the periphery of the wounded area because JAK/STAT activity there suppresses it. We also propose that cell sheet migration, inferred from cell shape changes, is mediated by both JNK and JAK/STAT signaling. Thus, each signaling pathway has a dual role in larval epidermal wound healing. JNK induces cell fusion or shape change, possibly depending on its activation level, and JAK/STAT signaling inhibits cell fusion but, in part, promotes cell shape change.

Our results indicate that for cell fusion to occur, not only the JNK activation level, but also the balance between JNK and JAK/STAT signaling is critical. Three points can be inferred from our data and a recent report (Wang et al., 2015). First, JNK functions as an instructive signal, based on the fact that JNK is both necessary and sufficient for cell fusion induction. Second, loss of JAK/STAT pathway activation functions as a permissive condition, because loss of JAK/STAT signaling may enhance, but is not itself sufficient for, cell fusion. Third, once JNK signaling is activated, a delicate balance between the levels of JNK and JAK/STAT signaling appears to be a key determinant for the execution of cell fusion, which is based on the fact that the degree of cell fusion is sensitive to the changes in activation levels of the two pathways. However, two exceptions should be noted. First, some cell fusion occurred in bsk^DN and bsk-i larvae, indicating that JNK-independent cell fusion exists, as previously described (Galko and Krasnow, 2004). Second, the pattern of cell fusion did not perfectly overlap the JNK and JAK/STAT activation gradients; that is, small syncytia often formed in locations that were remote and isolated from the wound region, suggesting that cell fusion may also be controlled by stochastic or as-yet-unidentified factors.

The mechanisms underlying JNK-mediated suppression of JAK/STAT pathway activation and those by which the two signaling pathways converge on the cell fusion machinery are presently unknown. It should be noted, however, that JNK-mediated suppression of the JAK/STAT pathway occurs only in the context of wound healing; JNK activation via egr overexpression moderately increased JAK/STAT activation in unwounded epidermis, which is rather consistent with some previous reports regarding the relationship between the two pathways (Pastor-Pareja et al., 2008; Santabárbara-Ruiz et al., 2015). Hypothesizing that JNK might exert its effect through transcriptional induction of a negative regulator of the JAK/STAT pathway, we analyzed upstream sequences of the et, Socs36E, and Su(var)2-10 loci and found multiple highly-conserved AP-1-binding sites (Angel et al., 1987; Lee et al., 1987a,b) in the upstream sequence of Socs36E. However, Socs36E was not transcriptionally induced 8 h after wounding, the time of peak JNK activation, in Socs36E>GFP larvae (data not shown), making it unlikely that Socs36E is a downstream target of JNK signaling. One of the mechanisms by which JAK/STAT suppresses cell fusion is possibly via the increase of integrin expression. JNK activation may increase βPS integrin in wounded epidermis, as we observed that the area of βPS induction during wound healing coincided with the JNK-activation domain, consistent with a previous finding in unwounded epidermis expressing hep^CA (Wang et al., 2015). The same paper reported that βPS integrin induced by JNK activation was excluded from some membranes, presumably to allow or promote cell fusion. In contrast, the βPS protein induced by JAK/STAT pathway activation localizes mainly to the cell boundaries. More detailed analyses are necessary, but JNK and JAK/STAT may regulate cell fusion through differential subcellular localization of integrin molecules. We think that the function of JAK/STAT is to set a higher threshold for cell fusion, partly via inducing and localizing integrin proteins to the plasma membrane. This in turn may modulate the behavior of unidentified fusogenic molecules on the cell surface.

Our results and those of others highlight that, not only the induction of wound healing responses, but also the confinement of the responses to wound vicinity is under strict genetic control in flies (Juarez et al., 2011; Patterson et al., 2013). While cell fusion, in combination with cell growth, is critically required for epidermal wound closure (Losick et al., 2013), whether too much cell fusion is harmful to proper wound healing is unclear. Our studies indicate that patches of big syncytia extending over a two-segment width in both anterior and posterior directions from the wound, however, do not impose immediate harm on the survival of an organism. Ectopic expression of the fusogenic protein Eff-1 or Aff-1 causes excessive cell fusion and embryonic lethality in C. elegans, implying that uncontrolled cell fusion is indeed detrimental in some contexts (Shemer et al., 2004; del Campo et al., 2005; Sapir et al., 2007). As a more general and relevant example, fibrotic diseases cause excessive accumulation of connective tissues and mesenchymal cells resulting from various stimuli including tissue damage. Such scenarios may lead to organ dysfunction and death (Singer and Clark, 1999; Wynn, 2007).

In summary, we identified a genetic mechanism by which the wound healing response, particularly cell fusion, is restricted to a specific area. Despite being known for diverse biological roles, a role for the JAK/STAT pathway has rarely been reported in relation to cell fusion. Giant cell formation in response to foreign bodies, such as implants, is mediated by activation of fusion-promoting STAT6, which suppresses fusion-inhibitory STAT1, resulting in de-repression of the transcription of downstream target genes (Miyamoto et al., 2012). Our findings show an intriguing similarity to these data, and may aid in elucidation of the mechanism underlying macrophage fusion.

The following lines were obtained from the Bloomington Stock Center: Oregon R, w¹¹¹⁸, hop³/FM7c, hop^Tum/FM7c, Stat92E⁰⁶³⁴⁶, w* upd3^Δ, mys¹/FM4, mys¹⁰/FM7a, msn-lacZ (msn⁰⁶⁹⁴⁶), en-GAL4 (e16E), hs-GAL4 (GAL4-Hsp70.PB), Mef2-GAL4, HmlΔ-GAL4, Lsp2-GAL4, UAS-upd1(1)-i (JF03149), UAS-upd3(1)-i (HMS05061), UAS-bsk^DN, UAS-hep^CA, UAS-Dcr-2, UAS-GFP, UAS-lacZ, and UAS-GFP.nls. The following lines were obtained from the Vienna Drosophila RNAi Center: UAS-dome-i (36355), UAS-hop-i (102830), UAS-Stat92E-i (43866), UAS-upd1-i(2) (3282), UAS-upd2-i(2) (14664), and UAS-upd3-i(2) (106869). UAS-upd3-i(1) and UAS-upd3-i(2) lines had the same RNAi construct, but at different chromosomal positions. The following lines were obtained from the National Institute of Genetics (Kyoto): UAS-upd2-i(1) (5988R-3) and UAS-bsk-i (5680R-1). The following lines were obtained from private collections: A58-GAL4 (Galko and Krasnow, 2004) and Tub-GAL80^ts; pnr-GAL4 UAS-DsRed2Nuc8/TM6B (Wang et al., 2015) from Michael Galko (Department of Genetics, The University of Texas MD Anderson Cancer Center, TX); UAS-egr (Igaki et al., 2002) from Ulrich Theopold (Department of Molecular Biosciences, Stockholm University, Sweden); UAS-dome^DN (UAS-dome^ΔCYT) (Brown et al., 2001) from James C. Hombria (Centro Andaluz de Biología del Desarrollo, Spain); UAS-hop (Harrison et al., 1995) and UAS-upd1 (Harrison et al., 1998) from Martin P. Zeidler (Department of Molecular Developmental Biology, Max Planck Institute for Biophysical Chemistry, Germany); and w,UAS-hid (Pastor-Pareja et al., 2008) and 10×STAT-GFP (Bach et al., 2007) from Tian Xu (Howard Hughes Medical Institute, Yale University School of Medicine, CT). For analysis of JNK and JAK/STAT signaling activation, UAS flies were crossed to the w;10×STAT-GFP;A58-GAL4 msn-lacZ/TM6B line and the activities of the two reporters were examined in the same sample.

Mid-to-late third-instar larvae were pinch-wounded on the dorsal side of segment A2 or A3 using a pair of forceps (Fine Science Tools, Cat. No. 11295-00). After the designated time of wound recovery, larval epidermis was dissected in 1× phosphate-buffered saline (PBS) for immunohistochemistry. The samples were fixed in 4% paraformaldehyde for 30 min. For qPCR, pinching was given at five neighboring dorsal segments to maximize gene induction, and the epidermis was dissected and kept in ice-cold 1× PBS for RNA extraction. For wounding in Fig. S2E–G, the anterior half (GFP-negative) of a body segment of en>GFP.nls or en>dome-i GFP.nls larvae was pinched using forceps under the GFP dissecting microscope (Zeiss Stereo Discovery.V8 with Zeiss HBO 100). For Fig. 5F–I, Tub-GAL80^ts; pnr-GAL4 UAS-DsRed2Nuc8/TM6B flies were crossed to UAS-hop; UAS-lacZ, UAS-lacZ; UAS-hep^CA, or UAS-hop; UAS-hep^CA, respectively, and were reared at 18°C. With the larval progeny, temperature was shifted to 32°C during early-to-mid third-instar stages for 20 h before dissection.

Fixed samples were stained with antibodies for immunohistochemical analysis as reported previously (Kwon et al., 2010). Briefly, muscle fibers underlying the epidermis were carefully removed, if necessary, after fixation using finer forceps (Fine Science Tools, cat. no. 11295-10) to reduce background immunofluorescence. The following primary antibodies were used: mouse anti-Fasciclin III (1:50 dilution; Developmental Studies Hybridoma Bank, cat. no. 7G10), rabbit anti-phospho-histone H3 (1:100 dilution; Santa Cruz Biotechnology, cat. no. sc-8656-R), rabbit anti-cleaved caspase-3 (Asp175) (1:200 dilution; Cell Signaling Technology, cat. no. 9661), mouse anti-βPS (1:50 dilution; Developmental Studies Hybridoma Bank, cat. no. CF.6G11), mouse anti-β-galactosidase (1:100 dilution; Developmental Studies Hybridoma Bank, cat. no. JIE7) and chicken anti-β-galactosidase (1:1000 dilution; Abcam, cat. no. ab9361). The following secondary antibodies were used: Cy3-conjugated anti-mouse-IgG (1:500; Jackson ImmunoResearch), Alexa Fluor 488-conjugated anti-mouse IgG (1:500; Invitrogen), Alexa Fluor 405-conjugated anti-mouse IgG (1:500; Invitrogen), Alexa Fluor 546-conjugated anti-rabbit IgG (1:500; Invitrogen) and Alexa Fluor 546-conjugated anti-chicken-IgG (1:500; Invitrogen). Nuclei were counterstained with 4′,6-diamidino-2-phenylindole (DAPI; Molecular Probes, cat. no. D1306) at 1:500 dilution. Samples were mounted in VECTASHIELD (Vector Laboratories) to minimize quenching by the light source. For cell membrane staining, cold lipophilic dye FM 1-43FX (5 μg/ml in PBS) was added to un-fixed muscle-depleted samples, followed by incubation for 10 min on ice, and fixation in cold 4% paraformaldehyde for 5 to 10 min. Olympus BX40F-3 with AxioCamMRc5 and Carl Zeiss LSM 700 microscopes were used for fluorescence and confocal microscopy, respectively.

Total RNA was extracted using TRIzol (Invitrogen, cat. no. 15596026) from 13–15 epidermal filets of each condition in triplicates. cDNA was prepared using 2 μg RNA and M-MLV RT reverse transcriptase (Promega, cat. no. M1701). Quantitative real-time PCR was performed using SYBR Premix Ex Taq (Tli RNaseH Plus; Takara, cat. no. RR420A), and rp49 was used to normalize the RNA levels. Relative mRNA levels were calculated using the comparative cycle threshold (Ct) method. The following primers were used: 5′-CAGTCGGATCGATATGCTAAGCTGT-3′ and 5′-TAACCGATGTTGGGCATCAGATACT-3′ for rp49; 5′-CCACGTAAGTTTGCATGTTG-3′ and 5′-CTAAACAGTAGCCAGGACTC-3′ for upd1; 5′-TCAACGAGGCGGTCACCAAGGA-3′ and 5′-GATCCGTTGGCTGGCGTGTGAA-3′ for upd2; 5′-TGCCAGCAGTACGCATCTG-3′ and 5′-CAGACCCGTGGGCTTCAG-3′ for upd3.

For all of the three methods in Fig. 2A–C, only samples with a wound scar confined to a single body segment were taken for analysis; samples with a scar spanning two (or more) segments were discarded. This allowed us to count all cell fusions within the wounded segment and helped remove background noise generated by small, frequent cell fusions that occurred independently of experimental wounding. To count fused, multinucleate epidermal cells more reliably, A58>GFP.nls was used to label cell nuclei instead of DAPI, which also labels muscle cell nuclei.

For the cell fusion rate in Fig. 5I, larvae with 30 or more of the pnr-positive (DsRedNuc-positive) epidermal cells per segment were chosen for cell fusion analysis. Two segments from each larval filet were analyzed. The average numbers of pnr-positive cells per segment were not grossly different among the three experimental groups [mean±s.d. values are 60±23 for pnrT>hop lacZ (10 segments), 67±20 for pnrT>hep^CA lacZ (34 segments), and 57±16 for pnrT>hop hep^CA (22 segments), respectively]. For each segment, the Fusion_nmax value was divided by the number of pnr-positive cells.

Quantification of cell shape change was performed as described previously with minor modifications (Kwon et al., 2010). Briefly, each side of a polygon of cell boundaries visualized by anti-FasIII immunostaining was defined as either linear or nonlinear; a side was defined as nonlinear if the gap between the line of anti-FasIII staining and an imaginary line directly connecting two neighboring vertices of a cell was wider than the width of the FasIII band. Fused cells were excluded from quantification because they were irregular in shape and may overestimate the value. Tendon cells were also excluded, as they do not change shape during wound healing. Consistency in wounding was maintained because values may be affected by the size and shape of wounds.

Fluorescence intensities of β-galactosidase immunostaining of msn-lacZ, GFP from 10×STAT-GFP, and βPS immunostaining were quantified using ImageJ software. For msn-lacZ, all cells with increased immunofluorescence intensities were evaluated and averaged. For 10×STAT-GFP in Fig. 3, all cells in the wounded trapezoid were evaluated and averaged. For 10×STAT-GFP in Fig. 4, all cells within two-cell layers from the wound margin (for 8 h analysis) or all cells within the wound scar, demarcated by the impression left on the cuticle (for 20 h analysis) were evaluated and averaged. As a control, 30 cells in the trapezoid of a neighboring unwounded segment were chosen randomly and the fluorescence intensity values were averaged. For βPS, a sub-marginal region of the wound was selected by enlarging the wound leading edge by 200 pixels and the fluorescence intensity, excluding that from the wound hole, was measured. This value was normalized by the value obtained from 12–13 cells in the trapezoid of a neighboring non-wounded segment of the same epidermal filet.

We thank M. Galko, J. C. Hombria, U. Theopold, T. Xu, M. P. Zeidler, Bloomington Stock Center, Vienna Drosophila RNAi Center, and National Institute of Genetics in Japan for flies and reagents. We thank members of the Choe laboratory and H.-S. Lee for helpful discussions. K.-M.C. also thanks H. D. Ryoo for generosity during a sabbatical sojourn.