Matriptase activation of Gq drives epithelial disruption and inflammation via RSK and DUOX

Epithelial tissues are primed to respond to insults by activating epithelial cell motility and rapid inflammation. Such responses are also elicited upon overexpression of the membrane-bound protease, Matriptase, or mutation of its inhibitor, Hai1. Unrestricted Matriptase activity also predisposes to carcinoma. How Matriptase leads to these cellular outcomes is unknown. We demonstrate that zebrafish hai1a mutants show increased H2O2, NfκB signalling, and IP3R -mediated calcium flashes, and that these promote inflammation, but do not generate epithelial cell motility. In contrast, inhibition of the Gq subunit in hai1a mutants rescues both the inflammation and epithelial phenotypes, with the latter recapitulated by the DAG analogue, PMA. We demonstrate that hai1a has elevated MAPK pathway activity, inhibition of which rescues the epidermal defects. Finally, we identify RSK kinases as MAPK targets disrupting adherens junctions in hai1a mutants. Our work maps novel signalling cascades mediating the potent effects of Matriptase on epithelia, with implications for tissue damage response and carcinoma progression.


Introduction
The transmembrane serine protease, Matriptase, encoded by the ST14 gene, has potent oncogenic properties and is consistently dysregulated in human carcinomas. Overexpression of Matriptase in the mouse epidermis leads to epidermal papillomas, ulcerative and invasive carcinomas, and inflammation (List et al., 2005;Martin and List, 2019). These effects of Matriptase are mitigated by a cognate serine protease inhibitor, HAI-1. Clinically, an increase in the Matriptase:HAI-1 ratio has been found in a range of tumours and is predictive of poor outcome (Martin and List, 2019). Loss of mouse Hai1 leads to epidermal and intestinal barrier defects, epithelial inflammation, and failure of placental labyrinth formation, which are all due to unrestricted Matriptase activity (Kawaguchi et al., 2011;Nagaike et al., 2008;Szabo et al., 2007). The response of epithelia to unregulated Matriptase activity appears conserved across vertebrates. Mutation of the zebrafish orthologue, Hai1a, also results in epidermal defects, including loss of membrane E-cadherin, aberrant mesenchymal behaviour of keratinocytes, which form cell aggregations over the body and loss of fin fold integrity. The epidermis also displays sterile inflammation and is invaded by highly active neutrophils. Genetic ablation of the myeloid lineage demonstrated that the keratinocyte phenotypes are not a consequence of the inflammation (Carney et al., 2007). The strong hai1a fr26 allele is embryonic lethal, dying within the first week, whilst the more mild allele, hai1a hi2217 , is semi-viable, with epithelial defects resolved through sphingosine-1-phosphate-mediated entosis and cell extrusion (Armistead et al., 2020). All hai1a mutant phenotypes can be ameliorated by reduction of Matriptase levels (Carney et al., 2007;Mathias et al., 2007).
Due to the clinical implications of its dysregulation, the signalling pathways activated pathologically by Matriptase are of interest. The G-protein-coupled receptor, proteinase-activated receptor-2 (Par2), is essential for the oncogenic and inflammatory effects of uninhibited Matriptase in zebrafish and mouse (Sales et al., 2015;Schepis et al., 2018). Par2 is directly activated by Matriptase proteolysis and signals through a number of heterotrimeric Ga protein subunits. Early studies in keratinocytes linked Par2 activation with intracellular Ca ++ mobilisation via phospholipase C, thus implicating Gq subunit as the canonical target (Schechter et al., 1998). Alternate Ga subunits, including Gi, Gs, and G12/13, are now known to also be activated by Par2 (Zhao et al., 2014). Par2 displays biased agonism, and the logic of the pathway utilised depends on cell context and the activating protease.
In vitro experiments using HEK293 cells implicated both Par2 and Gi in Matriptase-mediated Nfkb pathway activation (Sales et al., 2015). Whilst this explains the inflammatory phenotype of uninhibited Matriptase, it does not address whether Par2 promotes carcinoma phenotypes directly in keratinocytes in vivo. In zebrafish, as the keratinocyte defects are not dependent on inflammation, but are dependent on Par2, it is likely that there is a direct effect of Par2 on promoting keratinocyte motility. eLife digest Cancer occurs when normal processes in the cell become corrupted or unregulated. Many proteins can contribute, including one enzyme called Matriptase that cuts other proteins at specific sites. Matriptase activity is tightly controlled by a protein called Hai1. In mice and zebrafish, when Hai1 cannot adequately control Matriptase activity, invasive cancers with severe inflammation develop. However, it is unclear how unregulated Matriptase leads to both inflammation and cancer invasion.
One outcome of Matriptase activity is removal of proteins called Cadherins from the cell surface. These proteins have a role in cell adhesion: they act like glue to stick cells together. Without them, cells can dissociate from a tissue and move away, a critical step in cancer cells invading other organs. However, it is unknown exactly how Matriptase triggers the removal of Cadherins from the cell surface to promote invasion.
Previous work has shown that Matriptase switches on a receptor called Proteinase-activated receptor 2, or Par2 for short, which is known to activate many enzymes, including one called phospholipase C. When activated, this enzyme releases two signals into the cell: a sugar called inositol triphosphate, IP3; and a lipid or fat called diacylglycerol, DAG. It is possible that these two signals have a role to play in how Matriptase removes Cadherins from the cell surface.
To find out, Ma et al. mapped the effects of Matriptase in zebrafish lacking the Hai1 protein. This revealed that Matriptase increases IP3 and DAG levels, which initiate both inflammation and invasion. IP3 promotes inflammation by switching on pro-inflammatory signals inside the cell such as the chemical hydrogen peroxide. At the same time, DAG promotes cell invasion by activating a wellknown cancer signalling pathway called MAPK. This pathway activates a protein called RSK. Ma et al. show that this protein is required to remove Cadherins from the surface of cells, thus connecting Matriptase's activation of phospholipase C with its role in disrupting cell adhesion.
An increase in the ratio of Matriptase to HAI-1 (the human equivalent of Hai1) is present in many cancers. For this reason, the signal cascades described by Ma et al. may be of interest in developing treatments for these cancers. Understanding how these signals work together could lead to more direct targeted anti-cancer approaches in the future. Par2 can also transactivate EGFR through an unknown mechanism, and inhibition of EGFR alleviates certain basal keratinocyte phenotypes of zebrafish hai1a mutants (Schepis et al., 2018). Thus, the identity, contribution, and interactions of the pathways downstream of Matriptase and Par2 remain unclear. Here through use of the zebrafish hai1a mutant, we comprehensively map the essential pathways downstream of zebrafish Matriptase and Par2, leading to inflammation and epithelial disruption.

Results
Increased hydrogen peroxide and calcium flashes contribute to inflammation in hai1a mutants Neutrophils in hai1a embryos invade the epidermis, are highly motile, but move randomly (Carney et al., 2007;Mathias et al., 2007; Figure 1A-E, Video 1). To establish the nature of their stimulus, we tested if neutrophils in hai1a altered their behaviour in the presence of a large fin wound. In wild-type larvae, neutrophils were recruited from a great distance and tracked to the wound with high directionality. However, neutrophils in the hai1a mutant appeared largely apathetic to the wound and remained migrating randomly. There was a mild increase in neutrophil speed in hai1a larvae following wounding, indicating that they retain capacity to respond to additive stimuli (Figure 1-figure supplement 1A-D, Video 2). Co-labelling of basal keratinocyte nuclei (using TP63 immunostaining), neutrophils (Tg(fli1:EGFP) y1 transgenic), and TUNEL labelling of apoptotic cells highlighted that whilst the epidermis of hai1a mutants, unlike WT, had regions of apoptosis at 24hpf (arrowhead, Figure 1-figure supplement 1E, F), neutrophils were not associated, but rather found at nascent TUNEL-negative aggregates of basal keratinocytes (arrow). We conclude that epidermal cell death does not directly contribute to inflammation and that the effector stimulating neutrophils in hai1a mutants is as, or more, potent as that of wounds.
To identify the neutrophil activator in hai1a, we employed an unbiased approach using 2D gel proteomics to compare the wild-type proteome with that of strong hai1a alleles. The dandruff (ddf) mutant has many phenotypic similarities to the strong hai1a fr26 allele (van Eeden et al., 1996). Crosses between ddf ti251 or ddf t419 and hai1a hi2217 failed to complement, and sequencing of hai1a cDNA from both ddf alleles identified a nonsense mutation in the ddf t419 allele (c.771T>G; p.Tyr257Ter) and a missense mutation of a highly conserved amino acid in the ddf ti251 allele (c.749G>A; p.Gly250Asp) ( Figure 1F, given. (G) Selected region of 2D gel of protein extracted from 24hpf embryos for hai1a t419 (left) or hai1a ti251 (right) in red, superimposed over WT protein samples (green in both). The shift in pI of peroxiredoxin4 in both alleles is indicated with an arrow. (H, I) Projected lateral confocal views of pentafluorobenzenesulfonyl fluorescein (PFBSF) staining of WT (H) and hai1a ti251/hi2217 (I) tail fins at 48hpf. (J) Box and whiskers plot of PFBSF fluorescent staining intensity of WT and hai1a fr26 mutants at 48hpf in trunk and tail. n = 9; t-test ***p<0.001. (K, L) Projected lateral confocal views of PFBSF staining of WT (K) and hai1a hi2217 (L) at 16hpf. Scale bars: (B, I, L) = 100 mm. The online version of this article includes the following source data and figure supplement(s) for figure 1: Source data 1. 2D proteomics protein ID report list of the spot identities with significant ratio changes. Source data 2. 2D proteomics protein ratio changes for each of the protein spots identified as significantly changed. Source data 3. 2D proteomics top 50 proteins significantly changed in hai1a mutants. Figure supplement 1. The epidermis of hai1a mutants displays elevated hydrogen peroxide.
Video 1. Neutrophils in WT and hai1a hi2217 4dpf larva. Projected confocal timelapses of eGFP-positive neutrophils in the tail region of 4dpf Tg(mpx:eGFP) i114 (left) and hai1a hi2217 ; Tg(mpx:eGFP) i114 (right) larvae taken every 45 s for 45 min. Scale bar: 50 mm. https://elifesciences.org/articles/66596#video1 Figure 1-figure supplement 1G-I). We used both alleles for comparative 2D protein gel analysis at 24hpf and 48hpf. Rather than finding proteins with altered molecular weight, Peroxiredoxin4 (Prdx4) was identified as having a higher pI in both hai1a t419 and hai1a ti251 mutants at 24hpf and 48hpf, indicative of a change in oxidation state ( Figure 1G, Figure 1-figure supplement 1J, K). Peroxiredoxins are hydrogen peroxide scavengers, and its altered oxidation state suggested that hai1a has higher H 2 O 2 levels, a known activator of inflammation in larval zebrafish (Niethammer et al., 2009). Pentafluorobenzenesulfonyl fluorescein (PFBSF) staining Maeda et al., 2004 demonstrated significantly higher levels of H 2 O 2 in the trunk and tails of hai1a mutants at 24hpf and 48hpf ( Figure 1H-J, Figure 1-figure supplement 1L, M). This increase in H 2 O 2 in hai1a was observed as early as 16hpf, and thus preceded presentation of hai1a phenotypes ( Figure 1K, L). To demonstrate that, as with other phenotypes, the H 2 O 2 increase in hai1a was due to unrestrained activity of Matriptase1a, we used a matriptase1a mutant allele, st14a sq10 , which prematurely terminates the protein at 156 amino acids ( Lin et al., 2019). Zygotic st14a mutants showed no overt phenotype; however, maternal zygotic mutants lacked ear otoliths ( Figure 2B, C). As expected, when crossed into the hai1a background, embryos lacking otoliths (st14a sq10 ; hai1a hi2217 double mutants) never displayed the hai1a epidermal and neutrophil phenotypes (Figure 2D-F; Table 1). Double mutants also had significantly reduced H 2 O 2 levels ( Figure 2F and neutrophil inflammation in hai1a mutants but did not rescue the epithelial defects ( Figure 2F, G). Treatment with a known Duox inhibitor, diphenyleneiodonium (DPI), also resulted in amelioration of neutrophil inflammation, but not epithelial aggregates, in hai1a mutants ( Figure 2G, Figure 2-figure supplement 1E). We conclude that Matriptase1 activity leads to excess H 2 O 2 in hai1a mutants, which partially accounts for the neutrophil inflammation, but not epidermal defects.
Duox is regulated by calcium through its EF-Hand domains, and calcium flashes are known to generate H 2 O 2 in epidermal wounds in Drosophila (Razzell et al., 2013). We injected hai1a fr26 with RNA encoding the calcium reporter GCaMP6s. Timelapse imaging at 24hpf indicated that hai1a mutants had significantly more calcium flashes in both the trunk and tail ( Figure 3A, B, E, Figure 3figure supplement 1A, B, Video 3). Increased intracellular calcium dynamics was observable as early as 16hpf, concomitant with increased H 2 O 2 , but prior to onset of hai1a phenotypes ( Figure 3G, H, Video 4). Release of calcium from intracellular stores is regulated by IP 3 receptors located on the endoplasmic reticulum. The frequency and number of calcium flashes in the trunk and tail of hai1a mutants are reduced by treatment with the IP 3 R inhibitor, 2-APB compared to control ( Figure 3C, D, F, Figure 3-figure supplement 1C, D, Video 5). Reducing calcium flashes in hai1a mutant embryos with 2-APB also significantly reduced H 2 O 2 levels ( Figure 3I, J, Figure 3-figure supplement 1E) and partially reduced inflammation; however, the epidermal defects were not noticeably rescued (imaged by DIC (Differential Interference Contrast) or labelled with the TP63 antibody) ( Figure 3I-K). We observed similar reduction in neutrophil inflammation, but not rescue of epidermal defects, in hai1a mutants treated with thapsigargin, which inhibits the replenishment of ER calcium stores by SERCA ( Figure 3K, Figure 3-figure supplement 1F, G). This suggests, in hai1a mutants, that IP 3 R-dependent calcium flashes activate Duox, flooding the epidermis with H 2 O 2 and leading to inflammation.
To demonstrate that this increase in NfkB signalling was dependent on H 2 O 2 , we injected hai1a hi2217 ; Tg(6xHsa.NFKB:EGFP) nc1 embryos with duox MO. We noted a strong reduction in NfkB pathway activation compared to uninjected hai1a hi2217 mutant controls ( Figure 4J, K). Conversely, genetic ablation of NfkB signalling, through use of the ikbkg mutant, did not reduce H 2 O 2 levels in hai1a mutants ( Figure 4-figure supplement 1F, G). Similarly, we tested if reduction of calcium flashes could also reduce NfkB signalling in hai1a mutants using 2-APB but noticed only a slight reduction ( Figure 4-figure supplement 1H, I). We propose that upon loss of Hai1a, IP 3 R-mediated release of calcium activates Duox to increase H 2 O 2 . This acts upstream of NfkB pathway activation, which occurs at later stages, and is necessary for the inflammation phenotype.
Both inflammation and epidermal aggregates of hai1a mutants are resolved by Gq inhibition IP 3 is generated from cleavage of PIP 2 by Phospholipase C. The sensitivity of the hai1a mutants to 2-APB implies that IP 3 levels are increased and therefore there may be an increase in Phospholipase C activity. Numerous attempts to inhibit PLC in hai1a mutants failed, and we were unable to find a dosage window that rescued without gross embryo deformity. Hence, we tested rescue of hai1a mutants with YM-254890, an inhibitor of the heterotrimeric G protein alpha subunit, Gq, which directly activates PLC isoforms. We found that not only did this significantly reduce neutrophil inflammation ( Figure 5D, F), but surprisingly, it also significantly rescued the epidermal defects in hai1a mutants, with a significant reduction in TP63-positive epidermal aggregates in the trunk and improved tail fin fold integrity at 48hpf ( Figure 5A-E). , and eGFP (green) antibody staining at 48hpf (middle column) with DIC imaging (right column) for hai1a hi2217/ti251 mutants (J), or hai1a hi2217/ti247 mutants treated with 2.5 mM 2-APB (I). Individuals for middle and right columns were hemizygous for the Tg(mpx:eGFP) i114 transgene. (K) Counts of eGFP-positive neutrophils in the fins at 48hpf of Tg(mpx:eGFP) i114 , or hai1a hi2217 ; Tg(mpx:eGFP) i114 treated with 0.5% DMSO, 2.5 mM 2-APB, or 6.5 mM thapsigargin. n = 20; t-test; ***p<0.001. Scale bars (A-D) = 50 mm; (G, I, J) = 100 mm.
The online version of this article includes the following figure supplement(s) for figure 3: Figure supplement 1. Thapsigargin and 2-APB reduce neutrophil inflammation but not epidermal defects in hai1a mutants.
Video 3. Calcium dynamics in WT and hai1a fr26 embryos at 24hpf. Projected confocal timelapses of eGFP in the trunks (left side) and tails (right side) of a 24hpf WT (top row) and hai1a fr26 (bottom row) embryos injected with GCaMP6s RNA, indicating calcium dynamics. Scale bar: 50 mm.
https://elifesciences.org/articles/66596#video3 PMA treatment phenocopies the hai1a mutant As IP 3 R inhibition only blocks inflammation in hai1a mutants, but an inhibitor of a PLC activator (Gq) additionally reduces the epidermal defects, we considered that diacyl glycerol (DAG) might contribute to the epidermal defects as the second product of PIP 2 cleavage (along with IP 3 ). Indeed, treating WT embryos from 15hpf to 24hpf with 125 ng/ml phorbol 12-myristate 13-acetate (PMA), a DAG analogue, resulted in embryos with striking similarities to strong hai1a mutants, including a thin or absent yolk sac extension, lack of head straightening, lack of lifting the head off the yolk, and multiple epidermal aggregates on the skin ( Figure 6A-C). These aggregates were due, at least partially, to displacement of basal keratinocytes as shown by TP63 staining where the basal keratinocyte nuclei lost their uniform distribution ( Figure 6D, E). Treatment from 24hpf to 48hpf with 125 ng/ml PMA led to a fin defect similar to the dysmorphic hai1a mutant fin ( Figure 6F, G). It has been shown that the basal keratinocytes in hai1a lose their epithelial nature and adopt a partially migratory phenotype (Carney et al., 2007;Video 6). We treated Tg(krtt1c19e:lyn-tdtomato) sq16 larvae (Lee et al., 2014) with 37.5 ng/ml PMA for 12 hr and imaged the basal epidermis at 3dpf by light-sheet timelapse. Whilst the DMSO-treated transgenic larvae had very stable keratinocyte membranes and shape, PMA treatment led to a less stable cell membrane topology and dynamic cell shape, similar to hai1a mutants ( Figure 6H, Videos 7 and 8). Kymographs taken from Video 7 highlighted both the more labile and weaker cell membrane staining following PMA treatment ( Figure 6I). The potency of PMA was dependant on region and reduced with age. Most PMA-treated Tg(mpx.eGFP) i114 larvae at 48hpf also had more neutrophils in the epidermis than untreated controls, which were highly migratory ( Figure 6F-G, J-K 0 , Video 8). We determined H 2 O 2 levels in PMA-treated embryos using PFBSF staining and found that it was significantly increased in both trunk and tail at 24hpf ( Figure 6L-O, R). In contrast, when we treated GCaMP6s RNA-injected embryos with PMA, we failed to see an increase in calcium flashes, as seen in hai1a ( Figure 6P, Q, S). To see if the heightened H 2 O 2 and inflammation was also correlated with increased NfkB signalling, we treated Tg(6xHsa.NFKB:EGFP) nc1 embryos with 125 ng/ml PMA. There Video 4. Calcium dynamics in WT and hai1a hi2217 embryos at 16hpf. Projected light-sheet timelapses of eGFP in WT (left side) and hai1a hi2217 (right side) embryos at 16hpf. Both embryos carried the Tg(actb2: GCaMP6s, myl7:mCherry) lkc2 transgene reporting calcium dynamics, which were higher in the hai1a mutant, particularly over the yolk. Images were taken every 45 s for 19 min. Scale bar: 50 mm. https://elifesciences.org/articles/66596#video4 Video 5. Calcium dynamics in DMSO and 2-APBtreated hai1a fr26 embryos at 24hpf. Projected confocal timelapses of eGFP signal in the trunks (left side) and tails (right side) of 24hpf hai1a fr26 embryos injected with GCaMP6s RNA and treated with 0.03% DMSO (top row) and 2.5 mM 2-APB (bottom row), indicating reduced calcium dynamics following 2-APB treatment. Scale bar: 50 mm.
The phenocopy and the rescue of hai1a by PMA and Gq inhibition respectively imply that DAG is elevated in hai1a mutants. Elevated cellular DAG leads to relocalisation of Protein Kinase C isoforms to the plasma and nuclear lipid membranes where they bind DAG and become activated. Using a  GFP-tagged PKCd fusion protein (Sivak et al., 2005), we showed that in the WT embryo there was largely diffuse cytoplasmic PKCd-GFP signal, however, it translocated to plasma and nuclear membranes in hai1a mutants, indicating increased levels of DAG ( Figure 7A, B, Figure 7-figure supplement 1A, B). This is indeed relevant to the epidermal defects, as treatment of hai1a hi2217 embryos with the PKC inhibitor, GFX109203, reduced the epidermal aggregates and disruption of fin morphology as imaged by DIC or immunostaining for TP63 ( Figure 7C-H). Neutrophil inflammation in the epidermis was somewhat reduced, but not to a significant degree ( Figure 7E-I). Thus, these experiments strongly suggest that epithelial defects of hai1a are due to DAG generation and PKC activation.

Elevated MAPK signalling generates epithelial defects in hai1a
We next sought to determine which pathways downstream of PKC are responsible for the epidermal defects. The MAPK pathway is a major target pathway of multiple PKC isoforms, and activation of this pathway in zebrafish epidermis has previously been shown to induce papilloma formation which have very similar attributes to hai1a mutant aggregates (Chou et al., 2015). Although whole embryo western analysis of hai1a mutants failed to show an overall increase in pERK (Armistead et al.,  To establish that this is an early marker of aggregate formation, and not a sequela, we stained hai1a mutant embryos at earlier timepoints. We found that at 16hpf regions of the epidermis have pERK staining before overt aggregation formation ( Figure 8G-H), whilst nascent aggregates also contain pERK staining which increases in number over time (Figure 8-figure supplement 1G-L).
To determine if elevated pERK is causative of epidermal defects, we attempted to rescue using pERK inhibitors. Initially we used PD0325901; however, this appeared to give fin fold deformities, even in WT embryos (Anastasaki et al., 2012), precluding ability to assess rescue in hai1a, although there was a noticeable reduction in epidermal aggregates forming under the yolksac extension (data not shown). Instead, we tried U0126 and CI-1040, other well-known pERK inhibitors (Allen et al., 2003;Favata et al., 1998). Both inhibitors showed a significant reduction in hai1a mutant epidermal aggregates under the yolk, and restoration of the overall and tail epithelial morphology, with embryos showing a hai1a phenotype class significantly reduced ( Figure 9A-G, Figure 9-figure supplement 1A-F). Similarly, the epidermal defects of the trunk, yolk, and tail following PMA treatment were also ameliorated by concomitant U0126 treatment ( Figure 9H, I, Figure 9-figure supplement 1G, H). Rescue of aggregates and tail morphology following PMA treatment or in hai1a mutants could be visualised by immunolabelling TP63 in basal keratinocyte nuclei ( Figure 9J-O, Video 6. Basal keratinocyte membrane and neutrophil dynamics in 3dpf wild-type and hai1a hi2217 larvae carrying the Tg(krtt1c19e:lyn-tdtomato) sq16 and Tg (mpx:eGFP) i114 transgenes. Projected light-sheet timelapses of the trunk of 3dpf WT (left) and hai1a hi2217 (right) larvae with neutrophils and basal keratinocyte membranes labelled by eGFP and lyn-tdTomato, respectively. Both larvae carried the Tg(krtt1c19e:lyntdtomato) sq16 ; Tg(mpx:eGFP) i114 transgenes. The hai1a mutants have highly dynamic neutrophils and keratinocyte membrane dynamics. Scale bar: 20 mm. https://elifesciences.org/articles/66596#video6 Treatment with U0126 did not significantly reduce neutrophil inflammation of hai1a mutants or PMA treatment ( Figure 9L-P). This suggests that the inflammation phenotype is not simply a consequence of the epidermal defects. Furthermore, dye penetration assays showed that the epithelial barrier was not globally and overtly compromised in hai1a, underscoring that inflammation is not simply a consequence of epithelial defects (Figure 9-figure supplement 2A-H). It has been shown that the epidermal defects in hai1a are associated with loss of E-cadherin from adherens junctions (Carney et al., 2007). As there was a rescue of the epithelial phenotype following pERK inhibition, we looked at the status of the adherens junction marker b-catenin. Whilst the WT basal epidermal cells of the 48hpf tail showed strong staining at the membrane, hai1a mutants and PMA-treated embryos showed a significant loss of b-catenin at the membrane and increase in the cytoplasm ( Figure 9Q-V, Y, Z). Treatment of hai1a mutants with U0126 restored the membrane localisation of b-catenin ( Figure 9W, X, AA).
Phosphorylation of cytoplasmic RSK by pERK leads to loss of E-cadherin at the hai1a keratinocyte membrane As increased pERK appeared to contribute strongly to loss of adherens junctions and removal of E-cadherin/b-catenin from the membrane, we sought to determine how pERK signalling might affect adherens junctions. We predicted that this would occur through a cytoplasmic target of pERK as we have previously shown that there is no transcriptional downregulation of E-cadherin levels in hai1a, making a nuclear transcription factor target less likely to be relevant (Carney et al., 2007). The p90RSK family of kinases represents direct cytoplasmic targets of Erk1/2 phosphorylation which regulate cell motility, and thus were good candidates for mediators disrupting cell-cell adhesion (Č áslavský et al., 2013;Tanimura and Takeda, 2017). We determined that at least RSK2a (=p90RSK2a, encoded by rps6ka3a) is expressed in basal keratinocytes at 24hpf ( Figure 10A, B). To gauge if there was an alteration in phosphorylation of RSK family members in the epidermis of hai1a mutants, we used an antibody which detects a phosphorylated site of mouse p90RSK (Phospho-Thr 348 ). This site is phosphorylated in an ERK1/2-dependent manner (Romeo et al., 2012). We noticed a substantial increase in cytoplasmic signal in both hai1a mutants and PMA-treated embryos. Where p90RSK-pT 348 signal was largely nuclear in both basal and periderm cells in WT, it was more broadly observed in hai1a mutant fins, with an increase in the cytoplasm leading to a more uniform staining ( Figure 10C-D 0 ). This increase in cytoplasmic levels of p90RSK-pT 348 was observable at 17hpf prior to epithelial defects ( Figure 10-figure supplement 1A-C). p90RSK cytoplasmic signal Video 7. Basal keratinocyte membranes in DMSO and phorbol 12-myristate 13-acetate (PMA)-treated 3dpf Tg (krtt1c19e:lyn-tdtomato) sq16 larvae. Zoomed projected light-sheet timelapses of basal keratinocyte membranes labelled by lyn-tdTomato in the trunk of 3dpf Tg(krtt1c19e:lyn-tdtomato) sq16 larvae treated with 0.1% DMSO (left) and 37.5 ng/ml PMA (middle and right) for 18 hr. Membranes are stable in DMSOtreated larvae but were dynamic in PMA-treated larvae. Images were captured every 20 s. Scale bar: 10 mm. https://elifesciences.org/articles/66596#video7 Video 8. Neutrophils and basal keratinocyte membranes in DMSO and phorbol 12-myristate 13acetate (PMA)-treated 3dpf Tg(krtt1c19e:lyntdtomato) sq16 ; Tg(mpx:eGFP) i114 larvae. Lateral projection of light-sheet timelapse of neutrophils labelled by eGFP and basal keratinocyte cell membranes labelled by lyn-tdTomato in the trunks of 3dpf Tg(krtt1c19e:lyn-tdtomato) sq16 larva treated with 0.1% DMSO (left) and 37.5 ng/ml PMA (right) for 18 hr. PMA treatment leads to slightly dynamic cell membranes and motile neutrophils. Images were captured every 20 s for 30 min. Scale bar: 50 mm.
If phosphorylation of an RSK protein is required for mediating the pERK epidermal defects in hai1a mutants, then inhibition of RSK should rescue the epidermal defects. As morpholino-targeted inhibition of rps6ka3a was unsuccessful, we employed established pan-RSK inhibitors BI-D1870 and dimethyl fumarate (Andersen et al., 2018;Sapkota et al., 2007). Dimethyl fumarate treatment reduced the extent of cytoplasmic p90RSK-pT 348 in hai1a (Figure 10-figure supplement 1F, G). We noted that both inhibitors were able to reduce epidermal aggregates in hai1a mutants and restore fin morphology when visualised by DIC or TP63 immunofluorescence ( Figure 10K-N, Figure 10-figure supplement 1H, I, K, L). Reduction of mutant phenotype classes was significant at both 24hpf and 48hpf ( Figure 10-figure supplement 1J). We then assayed if RSK inhibition can reduce the aberrant cytoplasmic E-cadherin staining in hai1a mutant basal keratinocytes and observed that dimethyl fumarate treatment restored membrane localisation of E-cadherin in the mutants ( Figure 10O-Q 0 ). Thus, phosphorylation of RSK proteins is altered in hai1a mutants, and their inhibition appears to restore E-cadherin to the membrane and reduce epidermal aggregate formation.

Discussion
There are a number of similarities between loss of Hai1a in zebrafish and overexpression of Matriptase in the mouse epidermis, including inflammation, hyperproliferation, and enhanced keratinocyte motility, suggesting conservation of downstream pathways. What the conserved ancestral role of the Matriptase-Hai1 might have been is unclear. Matriptase dysregulation in the mouse is associated with cancer progression (Martin and List, 2019). Tumours have long been considered to represent non-healing wounds, and the cellular-and tissue-level phenotypes of hai1a have similarities to tumours. Epidermal cells in zebrafish transformed by MAPK activation both promote and respond to inflammation through similar mechanisms to wound responses (Feng et al., 2010;Schäfer and Werner, 2008). Further, tissue damage of the zebrafish epidermis perturbs osmolarity and releases nucleotides, leading to inflammation and epithelial cell motility, with the resulting phenotypes strikingly similar to hai1a mutants (de Oliveira et al., 2014;Enyedi and Niethammer, 2015;Gault et al., 2014;Hatzold et al., 2016). Indeed, the tissue responses initiated by loss of zebrafish Hai1a have been previously suggested to represent an early injury response (Schepis et al., 2018), whilst PAR2 synergises with P2Y purinergic and EGF receptors to promote cell migration in scratch assays (Shi et al., 2013). Thus our analysis supports the previous hypothesis of the Hai1-Matriptase system as a component of tissue injury responses (Schepis et al., 2018), which, if inappropriately activated, promotes carcinoma.
The various molecular pathways known to be activated by Matriptase have not been fully delineated or integrated. Par2 has previously been shown to be required for the hai1a phenotype in zebrafish and contributes to the phenotypes of Matriptase overexpression in the mouse. Exactly which heterotrimeric G-protein Par2 is activating in vivo and how this links to phenotypes has not been identified. Our analyses allow us to propose a pathway downstream of Par2 which accounts for both the inflammatory and the epidermal phenotypes ( Figure 11). Firstly, inhibition of Gq rescued both the inflammation and epithelial defects. PAR2 activation of Gq has been documented to occur in many cell types including keratinocytes, where inhibition of Gq and PKC reduces PAR2-mediated Nfkb signalling (Bö hm et al., 1996;Goon Goh et al., 2008;Macfarlane et al., 2005). Although we were unable to rescue hai1a phenotypes with a PLC inhibitor due to toxicity, genetic sensors demonstrated increased levels of Ca ++ and DAG in hai1a epidermis. Our analysis demonstrated that the different products of PIP2 hydrolysis appear to invoke the two main hai1a phenotypes to different  extents. IP 3 R-dependent calcium release in hai1a epidermis was required for Duox activity, high hydrogen peroxide levels, and, later, increased NfkB signalling. Reduction of these attenuated the inflammatory, but not epithelial, defects. Conversely, inhibiting the DAG receptor, PKC, rescued the epithelial phenotypes, and the inflammation slightly. The DAG analogue, PMA, phenocopied the epidermal defects of hai1a mutants but also increased H 2 O 2 , NfkB, and neutrophil inflammation, indicating that PKC activation may be sufficient, but not necessary, for inflammation. This is in line with known activation of Duox and IKK by PKC (Rigutto et al., 2009;Turvey et al., 2014). In addition, expression of activated Ras in zebrafish keratinocytes has been shown to lead to H 2 O 2 release and neutrophil attraction (Feng et al., 2010). Thus, there is likely to be dual contribution to the inflammatory phenotype from IP3 and DAG. It is important to stress however that the inflammation is not simply a result of epithelial defects or an overt loss of barrier. Firstly, we see increase in Ca ++ and H 2 O 2 very early in the epidermis prior to skin defects. Secondly, barrier assays failed to conclusively show a broad increase in permeability. Finally, rescue of epithelial defects by PKC and pERK inhibition did not fully rescue the inflammation. We conclude in our model that DAG contributes to both aspects of the phenotype, but IP 3 promotes only the inflammation.
Seminal experiments in transgenic mice overexpressing Matriptase in the epidermis and treated with a DMBA/PMA regime concluded that Matriptase and PMA activate functionally similar carcinoma promoting pathways (List et al., 2005). Our subsequent analysis suggests that this would include the MAPK pathway as we see increased phosphorylated-ERK in the epidermis of both hai1a mutants and also PMA-treated embryos. That we can rescue the epithelial defects using a MEK inhibitor indicated that this increase in epidermal pERK is likely critical to the phenotype. The MAPK pathway is known to regulate cell motility (Tanimura and Takeda, 2017). In the zebrafish epidermis, misexpression of activated MEK2 generated papillomas with remarkable resemblance to the epidermal aggregates in hai1a mutants (Chou et al., 2015), and which are not overtly proliferative. In astrocytes and oesophageal or breast tumour cell lines, PAR2 stimulates migration and invasiveness through MAPK/ERK, activation of which required Gq and PIP2 hydrolysis (Jiang et al., 2004;McCoy et al., 2010;Morris et al., 2006;Sheng et al., 2019).
One of the main molecular defects defined for zebrafish hai1a is the removal of adherens junction proteins from the membrane (Carney et al., 2007). MAPK signalling has been shown to reduce E-cadherin expression at adherens junctions and promote cytoplasmic accumulation through phosphorylation of the effector, RSK (Č áslavský et al., 2013). Like Matriptase, activation of RSK2 is associated with tumour progression, promoting invasiveness and metastasis of glioblastomas and head and neck squamous cell carcinomas (Kang et al., 2010;Sulzmaier et al., 2016). Promotion of invasiveness has also been noted for activated RSK1, which promotes invasion of melanoma clinically as well as in vitro and zebrafish melanoma models (Salhi et al., 2015). Intriguingly, proximity protein labelling has identified p120-catenin as a target of RSK phosphorylation. This catenin promotes cellcell adhesion by stabilising cadherins at junctions, a function inhibited by RSK phosphorylation (Méant et al., 2020). More broadly, RSK2 activity promotes cell motility through other mechanisms, including inactivation of Integrins and activation of the RhoGEF, LARG (Gawecka et al., 2012;Shi et al., 2018). Thus, we propose that pERK signalling, through RSK members, significantly contributes to dissolution of adherens junctions and the hai1a epidermal phenotype. We observed increased pERK in the cytoplasm and also the nucleus of keratinocytes, with comparatively more nuclear levels in periderm cells. Thus, whilst RSKs are phosphorylated by pERK, it is also likely that other cytoplasmic and also nuclear targets, such as cFos and Ets transcription factors, may also be activated, and that there are underlying transcriptional changes in hai1a mutants. It is not clear why pERK shows slightly different subcellular localisation patterns between the two different epidermal layers, but the two layers do respond differently to ErbB2 inhibition (Schepis et al., 2018), whilst calcium is recently described to alter nuclear shuttling of pERK (Chuderland et al., 2020).
Our model for how Matriptase invokes cellular responses is highly likely to be incomplete. Indeed, others have indicated MMPs, HB-EGF, EGFR, and AKT and are downstream of Matriptase and PAR2 function (List et al., 2005;Schepis et al., 2018;Darmoul et al., 2004;Chung et al., 2013;Rattenholl et al., 2007). Furthermore, Matriptase promotes HGF-cMet signalling in mouse (Szabo et al., 2011). We do not think that these conflict with our model but will interface with it. A number of reports have demonstrated that PI3K/AKT and MEK/ERK function in parallel downstream of PAR2 (Sheng et al., 2019;Tanaka et al., 2008;van der Merwe et al., 2009). Furthermore, there is evidence that PKC activates both MEK/ERK and EGFR independently following PAR2 stimulation, and that PI3K is activated by PAR2 via Gq (Wang and DeFea, 2006;Al-Ani et al., 2010). Cell identity, subcellular localisation, b-arrestin scaffolding, and biased agonism/antagonism are known to generate alternative downstream outputs from PAR2 (Zhao et al., 2014). To understand fully the roles of Matriptase and PAR2 in epithelial homeostasis and carcinoma, it will be critical to map how, when, and where they activate different downstream pathways.

Materials and methods
Key resources table Reagent type (species) or resource Designation Source or reference Identifiers Additional information Strain, strain background (Escherichia coli)
Genomic DNA and RNA extraction, reverse transcription, and PCR Adult fin clips or embryos were isolated following anaesthesia, and genomic DNA extracted by incubation at 55˚C for 4 hr in Lysis buffer (10 mM Tris pH 8.3, 50 mM KCl, 0.3% Tween20, 0.3% Nonidet P-40, 0.5 mg/ml Proteinase K). PCRs were performed using GoTaq (Promega) on a Veriti thermal cycler (Applied Biosystems) and purified with a PCR purification kit (Qiagen). TRIzol (Invitrogen) was used for RNA extraction following provided protocol, and cDNA generated from 1 mg total RNA using SuperScript III Reverse Transcriptase (Invitrogen) with Oligo(dT)12-18 primer. For qPCR, iTaq SYBR green (Bio-Rad) was used to amplify, with reaction dynamics measured on a Bio-Rad CFX96 Real-Time PCR Detection System. For measuring nfkbiaa mRNA by qPCR, the following primers (5 0 to 3 0 ) were used to amplify a region encoded on exons 4 and 5: F-AGACGCAAAGGAGCAGTGTAG, R-TGTGTGTCTGCCGAAGGTC. Reference gene was eef1a1l1 and the primers used amplified between exon 3 to 4: F-CTGGAGGCCAGCTCAAACAT, R-ATCAAGAAGAGTAGTACCGCTAGCA TTAC.

RNA synthesis
RNAs for GCaMP6s and PKCd-GFP were synthesised from pCS2-based plasmids containing the respective coding sequences (Sivak et al., 2005;Chen et al., 2017). These were linearised with NotI (NEB), and RNA in vitro transcribed with mMESSAGE mMACHINE SP6 Transcription Kit (Ambion). RNA for Tol2 was generated from the pT3Ts-Tol2 plasmid, linearised with SmaI (NEB), and transcribed with the mMESSAGE mMACHINE T3 Transcription Kit (Ambion). RNA for injection was purified by lithium chloride precipitation.

Embryo injection and morpholino
Embryos were aligned on an agarose plate and injected at the one-cell stage with RNA or morpholino diluted in Phenol Red and Danieau's buffer using a PLI-100 microinjector (Harvard Apparatus). Injection needles were pulled from borosilicate glass capillaries (0.5 mm inner diameter, Sutter) on a Sutter P-97 micropipette puller. The Duox morpholino (AGTGAATTAGAGAAATGCACCTTTT) was purchased from GeneTools and injected at 0.4 mM with 0.2 mM of the tp53 morpholino (GCGCCA TTGCTTTGCAAGAATTG).

TALEN mutagenesis
To generate the ikbkg mutant, TALEN vectors targeting the sequence ATGGAGGGCTGG in second exon were designed and constructed by ToolGen (http://toolgen.com). TALEN vectors were linearised with PvuII (NEB) and purified using a PCR purification kit (Qiagen), and then used for in vitro transcription with the MEGAshortscript T7 kit (Ambion). About 170-300 pg of supplied ZFN RNAs or purified TALEN RNAs were then injected into one-cell stage WT zebrafish embryos, which were raised to 24 hr, then genomic DNA extracted. For detection of fish with edited loci, PCR was performed on genomic DNA of injected fish with primers flanking the target site, cloned by TA cloning into pGEMT-Easy (Promega) or pCR2.1-TOPO-TA (Invitrogen) and individual clones sequenced to establish efficiency. Other embryos were raised to adulthood and their offspring were similarly genotyped to identify founder mutants.

In situ hybridisation
A probe corresponding to the final 1078 bp of rps6ka3a (RSK2a; NM_212786.1) was generated by cloning a PCR-derived cDNA fragment into in pGEMT-Easy (Promega), linearising with ApaI (NEB) and transcribing a DIG probe with SP6 RNA polymerase (Roche). Whole-mount in situ hybridisation developed with NBT/BCIP (Roche) was performed as described (Thisse and Thisse, 2008).

Immunofluorescent, dye staining, and TUNEL
For antibody staining, embryos were fixed in 4% paraformaldehyde overnight at 4˚C and then washed in PBT (0.1% Triton in PBS), permeabilised in À20˚C acetone for 7 min, washed in PBT, Figure 11. Model of pathway-activated downstream of Hai1 and Matriptase. Proposed model of pathways downstream of Hai1 which drives chronic and acute sterile inflammation (red) and epithelial motility (blue). A previously defined transactivation of EGFR is also integrated. Other pathways known to act downstream of Matriptase, involving cMet, PI3K, AKT, and mTOR, are not shown.
To stain hydrogen peroxide, embryos were incubated for 60 min at room temperature with 12.5 mM PFBSF (#10005983, Cayman Chemicals), then rinsed in Embryo Medium, anaesthetised, and imaged.
Fluorescent TUNEL staining was performed using the Fluorescein In Situ Cell Death Detection Kit (11684795910, Roche), with the fluorescein detected by antibody staining using rabbit anti-FITC, and co-immunostained for TP63 and eGFP. Epidermal permeability assays were conducted by immersing 36hpf embryos in 2.5 mg/ml fluorescein isothiocyanate-dextran 3-5 kDa (Sigma) or 0.075% methylene blue for 30 min and then destained in E2 medium.

Microscopy and statistical analysis
Still and timelapse imaging was performed on upright Zeiss AxioImager M2, Zeiss Light-sheet Z.1, upright Zeiss LSM800 Confocal Microscope or Zeiss AxioZoom V16 microscopes. Embryos were mounted in 1.2% Low Melting Point Agarose (Mo Bio Laboratories) in 0.5Â E2 medium in 35 mm glass-bottom imaging dishes (MatTek) or in a 1 mm inner diameter capillary for light-sheet timelapse. When imaging was performed on live embryos, the embryo media were supplemented with buffered 0.02% Tricaine and imaging conducted at 25˚C. Image processing was done using Zen 3.1 software (Zeiss), Fiji (ImageJ, ver. 1.52p), or Imaris (Bitplane) and compiled using Photoshop 2020 (Adobe). Neutrophils were tracked with TrackMate in Fiji or using the Spot function in Imaris. Kymographs were generated using the Reslice function in Fiji following generation of a line of interest across image. Fluorescence intensities were calculated using the Average Intensity function in Fiji following generation of a Region of Interest and masking of the DAPI channel to exclude the nucleus when required. In statistical analyses, n = number of embryos or cells measured, and as defined in the figure legend. GraphPad Prism was used for statistical analyses and graph generation. In all statistical tests, *p<0.05, **p<0.01, ***p<0.001. Tests used are indicated in the associated figure legend and were Student's t-test, Chi-squared test, Mann-Whitney test, or ANOVA with Bonferroni or Dunn's post-tests. The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication. Additional files

Supplementary files
. Transparent reporting form

Data availability
All data generated or analysed during this study are included in the manuscript and supporting files.