Matriptase-dependent epidermal pre-neoplasm in zebrafish embryos caused by a combination of hypotonic stress and epithelial polarity defects

Aberrantly up-regulated activity of the type II transmembrane protease Matriptase-1 has been associated with the development and progression of a range of epithelial-derived carcinomas, and a variety of signaling pathways can mediate Matriptase-dependent tumorigenic events. During mammalian carcinogenesis, gain of Matriptase activity often results from imbalanced ratios between Matriptase and its cognate transmembrane inhibitor Hai1. Similarly, in zebrafish, unrestrained Matriptase activity due to loss of hai1a results in epidermal pre-neoplasms already during embryogenesis. Here, based on our former findings of a similar tumor-suppressive role for the Na+/K+-pump beta subunit ATP1b1a, we identify epithelial polarity defects and systemic hypotonic stress as another mode of aberrant Matriptase activation in the embryonic zebrafish epidermis in vivo. In this case, however, a different oncogenic pathway is activated which contains PI3K, AKT and NFkB, rather than EGFR and PLD (as in hai1a mutants). Strikingly, epidermal pre-neoplasm is only induced when epithelial polarity defects in keratinocytes (leading to disturbed Matriptase subcellular localization) occur in combination with systemic hypotonic stress (leading to increased proteolytic activity of Matriptase). A similar combinatorial effect of hypotonicity and loss of epithelial polarity was also obtained for the activity levels of Matriptase-1 in human MCF-10A epithelial breast cells. Together, this is in line with the multi-factor concept of carcinogenesis, with the notion that such factors can even branch off from one and the same initiator (here ATP1a1b) and can converge again at the level of one and the same mediator (here Matriptase). In sum, our data point to tonicity and epithelial cell polarity as evolutionarily conserved regulators of Matriptase activity that upon de-regulation can constitute an alternative mode of Matriptase-dependent carcinogenesis in vivo.


Introduction
The genetic basis of carcinogenesis is highly complex and variable. Depending on context and tissue, dysregulated oncogenes can lead to different outcomes, sometimes even involving different downstream signaling pathways. For example, the type II serine transmembrane protease Matriptase-1, also named Suppressor of tumorigenicity 14 (St14), has been reported as a potential oncogene in a wide range of epithelial derived cancers, including breast cancer, colorectal cancer, and squamous cell carcinomas [1,2,3]. Aberrant Matriptase activity is associated with tumor initiation as well as progression and metastasis. Underlying pathways are diverse and can include different growth factors and cell surface receptors such as Hepatocyte growth factor (HGF) and its receptor c-Met, Macrophage-stimulating protein (MST) and its receptor Ron, and/or the proteolysis-activated receptor Par2, as well as different intracellular signal transducers such as phosphatidylinositol-3-kinase (PI3K), MAP-kinase or the transcription factor NFkB [1,4,5,6]. Matriptase is synthesized as a zymogen and trafficked to the cell surface, where it undergoes tightly controlled auto-activation [7]. In cell culture systems and biochemical in vitro assays, respectively, this auto-activation has been shown to be influenced by the subcellular localization of Matriptase [8,9] and by pericellular environmental conditions, with highest activation levels at mildly acidic pH and low ionic strength [10,11,12]. In healthy tissue, Matriptase activity is usually restrained by its cognate inhibitors, the Kunitz-type serine protease inhibitors Hai-1 and Hai-2, also named Spint1 and Spint2, while an imbalanced ratio of Matriptase and its inhibitors results in Matriptase dysregulation and pathology [13]. However, other modes of pathological / oncogenic Matriptase hyperactivation are feasible as well, but, at least with respect to their in vivo relevance, little understood.
In recent years, not only adult but also embryonic zebrafish have evolved as animal models of carcinogenesis [14,15,16,17,18], in line with the concept of oncofetal reprogramming in tumor development and progression [19]. Several zebrafish mutants have been identified that display early stages of carcinogenesis in the bi-layered embryonic epidermis, characterized by keratinocyte hyper-proliferation and aggregate formation, as well as partial or complete epithelial-mesenchymal transitions (EMTs) of keratinocytes. Among them are loss-of-function In this light, we wondered whether hypotonicity has a similar impact on epidermal aggregate formation caused by loss-of-function mutations in three other zebrafish genes, hai1a, encoding the cognate inhibitor of the type II transmembrane serine protease Matriptase-1/ St14a [21], clint1a, encoding a clathrin-interacting protein involved in endocytosis [22], and epcam, encoding an epithelial cell adhesion molecule [42]. For this purpose, embryos from incrosses of parents heterozygous for the respective mutation were raised in E3 medium, which, with an osmolality of 12 mOsm, is highly hypotonic to the zebrafish embryo, or in E3 supplemented with 250 mM Mannitol to increase the osmolality to isotonic conditions (270 mOsm) [26]. Roughly 25% of embryos obtained from clint1a and epcam heterozygote incrosses developed indistinguishable epidermal phenotypes in both hypotonic and isotonic medium (Fig 1A). In contrast, hai1a mutants or morphants (wild-type embryos injected with a hai1a antisense morpholino oligonucleotide / MO) raised in isotonic medium displayed a reduction in the severity of the epidermal aggregate phenotype compared to embryos raised in hypotonic medium. However-as opposed to atp1b1a mutants (Fig 1A)-the rescue was not complete (Fig 1A and 1F). Similarly, the expression of mmp9 (Fig 1G), indicative of partial EMT of basal keratinocytes, as well as epidermal proliferation rates (Fig 1H), were attenuated in hai1a mutants or morphants raised in isotonic medium compared to their siblings in hypotonic medium, but-in contrast to atp1b1a mutants-without reaching wild-type levels (Fig 1G, [26]). Together, this indicates that hypotonic stress specifically enhances epidermal pre-neoplastic transformations caused by the loss of ATP1b1a or Hai1a, yet does not influence transformations caused by the loss of Clint1a or EpCAM.
But what do ATP1b1a and Hai1a have in common that makes them susceptible to environmental tonicity? In line with the role of Hai1 as the cognate inhibitor of Matriptase-1, the epidermal defects in hai1a mutants are completely restored to wild-type conditions by inactivating Matriptase-1 with st14a MOs (Fig 1A) [21]. As former reports have demonstrated a positive effect of low ionic strength on the proteolytic activity of Matriptase-1 in vitro [10,11,12], we wondered whether the partial rescue of hai1a mutants by isotonicity might be due to an isotonicity-induced reduction in Matriptase activity. Conversely, the epidermal defects of atp1b1a mutants might be due to a hypotonicity-induced increase in Matriptase activity, whereas the defects of clint1a and epcam mutants are Matriptase-independent. To look into the latter, offspring from incrosses of clint1a, epcam, atp1b1a or hai1a heterozygotes were injected with st14a MO to inactivate endogenous Matriptase-1 and raised in hypotonic E3 medium. Indeed, st14a MO-injected atp1b1a mutants displayed a complete rescue of the epidermal aggregate phenotype (Fig 1A, 1I, 1J, 1L and 1M; but persistent pericardial edema, see below), similar to the response of hai1a mutants ( Fig 1A) [21], whereas clint1a and epcam mutants did not respond (Fig 1A). To validate these data, we also generated a genetic st14a null mutant (st14a fr56 ), using CRISPR/Cas9 technology (S1A-S1C Fig), to genetically delete Matriptase-1 function in atp1b1a mutants. Strikingly, atp1b1a-/-; st14a-/-double mutants (Figs 1K and 1N and S1D and S1E), as well as embryos homozygous for atp1b1a but heterozygous for the mutant st14a fr56 allele (S1H, S1I and S1J Fig), did not develop any epidermal aggregates, implying a haploinsufficiency of st14a in this context. Similarly, only 38% of st14a +/-embryos injected with hai1a MO developed aggregates, compared to 0% of injected st14a-/-embryos (S1F, S1G, S1H and S1K Fig) and 100% of injected embryos homozygous for the wild-type st14a allele (S1H and S1K Fig).
Whereas these two phenotypic traits of atp1b1a mutants persist upon concomitant loss of St14a, components of the ATP1b1a-dependent oncogenic pathway downstream of the convergence point of the epithelial polarity and osmoregulatory branches are affected [26] (Fig 2A). atp1b1a single mutants display elevated levels of pAKT in basal cells as a consequence of upregulated PI3K signaling, which in turn leads to increased mTORC1 and NFkB activity [26] (Fig 2A). Concomitant loss of Matriptase-1 activity, however, reduces AKT phosphorylation (Fig 3A-3C; n = 41-46), mTORC1 activity (reflected by pRPS6 immunofluorescence as a readout for mTORC1 signaling; Fig 3D-3F; n = 27-32), as well as transcriptional NFkB activity (reflected by the intensity of eGFP in a NFkB-RE:eGFP transgenic line; Fig 3G-3I, 3P), back to wild-type levels. Furthermore, loss of Matriptase-1 function in atp1b1a mutants blocks downstream effects of NFkB, such as the hyper-proliferation in the epidermis (Fig 3J-3L, 3Q) and the strongly elevated expression of mmp9, indicative of partial EMT (Fig 3M-3O, 3R), both of which contribute to the pre-neoplastic transformation of basal keratinocytes caused by the loss of ATP1b1a [26].

Par2b is partially required for epidermal carcinogenesis in atp1b1a mutants
In the context of the loss of its cognate inhibitor Hai1, Matriptase-1 has been shown to act via the G-protein coupled receptor Par2. Upon proteolytic cleavage of its ectodomain by

PLOS GENETICS
Matriptase-1, Par2 can activate different intracellular signal transduction pathways [43,44,45,46]. Accordingly, in zebrafish hai1a mutants, simultaneous knockdown or knockout of Par2b can almost completely restore the wild-type phenotype [20,23,25]. In atp1b1a mutants, concomitant loss of Par2b function via par2b MOs or a genetic par2b fr57 null mutation generated via CRISPR/Cas9 technology (S2J-S2L Fig) also attenuates epidermal aggregate formation and mmp9 expression in basal keratinocytes in a dose-dependent manner (S2A- S2I  Fig). However, in contrast to hai1a mutants treated in parallel (S2M- S2Q Fig), this rescue is incomplete, and wild-type conditions are not fully restored. We conclude that also in atp1b1a mutants, the oncogenic effect of Matriptase-1 is-at least in part-mediated by Par2b. However, in contrast to hai1a mutants and the activation of the EGFR-PLD pathway [20], additional Matriptase-1 targets seem to be involved to fully activate the ATP1b1a-specific oncogenic pathway via PI3K and AKT.

In cultured human cells, hypotonic medium and loss of the epithelial cell polarity regulator Scribble increase Matriptase-1 activity
We next aimed to elucidate the molecular and cellular mechanisms by which loss of ATP1b1a leads to aberrantly increased Matriptase-1 activity. In zebrafish hai1a mutants and multiple human malignancies [1,13], this hyper-activity most likely results from an imbalanced ratio between Matriptase-1 and its cognate inhibitor Hai1. In zebrafish atp1b1a mutants, however, st14a and hai1a transcript levels are not significantly changed compared to their siblings ( Fig 4A and 4B), pointing to other mechanisms. If Matriptase-1 is indeed positioned downstream of the hypotonic stress induced by the loss of ATP1b1a, as proposed above (Fig 2A), its activity should-even in systems with wild-type ATP1b1 function-be influenced by the ionic strength of the environment and by epithelial cell polarity regulators. To analyze this and to look into the evolutionary conservation of this mechanism, we turned to mammalian cell culture systems.
To test whether hypotonic stress increases the activity of zebrafish Matriptase-1 towards cleavage of its likely in vivo substrate Par2b, we applied a formerly established HEK293 cell culture assay. Hereby, the activity of Matriptase-1 is measured by the cleavage of the extracellular domain of its substrate Par2b fused to Alkaline Phosphatase (AP), quantifying released AP in the supernatant via an enzymatic reaction [25,44,47]. Previous work via this assay has demonstrated that, similar to mammalian counterparts, zebrafish Matriptase-1a cleaves Par2b mainly at the canonical cleavage site [25]. Here, cells co-transfected with plasmids encoding zebrafish Matriptase-1a and AP-Par2b were grown in isotonic medium for 24 hours and then exposed to media of different osmolalities (320 mOsm as the standard cell culture condition control, 270 mOsm corresponding to the isotonic medium zebrafish embryos were exposed to (Fig 1), and 230 mOsm and 150 mOsm as hypotonic conditions) for 15 and 45 minutes. Compared to isotonic 320 mOsm medium, a reduction of the medium osmolality to 230 mOsm resulted in a moderate increase of AP activity, and a reduction to 150 mOsm in a strong increase of AP in the supernatant already after 15 minutes, which persisted until the 45 minutes end point (Figs 4C-4E and S3). Of note, when as a control, Par2b was transfected alone, the hypotonicityinduced increase of AP release (possibly mediated by endogenous Matriptases of the HEK293 cells) was much weaker, indicating that approximately 80% of the hypotonicity-induced increase of AP release in the co-transfected samples is indeed due to an hypotonicity-induced increase in the proteolytic activity of the co-transfected zebrafish Matriptase-1a (S3 Fig).
Consistent and even more direct data pointing to a positive effect of low ionic strength on the activity of endogenous Matriptase-1 were also observed in the breast epithelial line MCF-10A, which in contrast to breast cancers [13,48,49,50] normally only contains low levels of active Matriptase-1 [9]. Active Matriptase resulting from autocatalytic cleavage of the fulllength zymogen can be directly detected by immunoblot as a 26 kDa product [51]. When culturing MCF-10A cells for 24 hours in hypotonic medium, we found the levels of cleaved / cleaves AP-Par2b more efficiently at lower osmolalities. HEK293 cells were transfected with empty pcDNA3, pcDNA3 +AP-Par2b, and pcDNA3+St14a. After 24 hrs in regular / isotonic medium (tonicity of 320 mOsm), cells were exposed to media of 320 mOsm, 270 mOsm, 230 mOsm, and 150 mOsm for 15 and 45 minutes, respectively. (C) At 15 min, absolute luminescence values of AP released into the supernatant progressively increase with increasing hypotonicity / lower tonicity. (D,E) Ratios of luminescence between isotonic and hypotonic media indicate an up to 3.79-fold increase (in 150 mOsm medium) after 15 min (D) and a 2.51-fold increase after 45 min (E). Ratios were determined from values as shown in (C), deducting baseline luminescence (bar 1 in C) from the luminenscences of co-transfected samples obtained at different osmolarities (bars 3-6 in C), normalized against the value at 320 mOsm (isotonic); n = 5, significances were determined via a one-way ANOVA and Tukey's post hoc test; columns with same superscript letter (a,b,c) are not significantly different (p>0.05). (F) Immunoblot analysis for processed / activated Matriptase-1 (ST14) showing that culturing of MCF-10A cells for 24 hours in hypotonic medium (150 mOsm) leads to a 3.78-fold increase in active endogenous Matriptase-1 compared to cells cultured in isotonic medium (320 Osm). Bar diagram displays mean value of proteins normalized to loading control GAPDH (n = 3); ns, not significantly different (p>0.05); **, ***, significantly different with p<0.01, p<0.001, respectively. (G-H) Scribble knockout (SCRIB KO) in human MCF-10A cells does not affect protein levels of full length ST14 or HAI1. Representative western blots show unaltered amounts of endogenous full-length ST14 (G) and HAI1 (H) proteins in SCRIB-KO cells compared to knockout cells with reintroduced Scribble (SCRIB KO + SCRIB), cultured in media of 320 mOsm, 230 mOsm, or 150 mOsm for 1 hr. Bar diagrams display mean value of proteins normalized to loading control GAPDH (n = 2); all differences are not statistically significant (p>0.05). (I) Loss of Scribble increases active ST14 in media of different osmolalities. Representative western blots showing active ST14 and GAPDH of SCRIB-KO and SCRIB-KO + SCRIB cells cultured in media of 320 mOsm, 230 mOsm, and 150 mOsm for 24 hrs, with highest numbers in cells lacking the epithelial polarity protein Scribble protein and exposed to hypotonic stress. Bar diagram displays mean value of proteins normalized to loading control GAPDH (n = 3). Significances were determined via a one-way ANOVA and Tukey's post hoc test; columns with same superscript letter (a,b,c,d) are not significantly different (p>0.05).
Together, these results imply hypotonic stress as a factor enhancing the proteolytic activity of both zebrafish and human Matriptase-1, consistent with former data obtained for the activity of mammalian Matriptase-1 towards other substrates [10,11,12], and suggesting that in zebrafish atp1b1a mutant keratinocytes in vivo, Matriptase-1a activity may also be increased due to the low ionic strength conditions in the interstitial fluid of the epidermis.
Furthermore, to investigate the impact of epithelial cell polarity on human Matriptase-1 activity in this MCF-10A system, we knocked out the epithelial polarity regulator Scribble (SCRIB). Scribble is part of the Scrib/Dlg/Lgl module regulating epithelial cell polarity and has also been implicated with tumor suppression [34,37]. SCRIB knockout (KO) cells lack Scribble protein (Fig 4G), however, they do not show significant differences in the levels of full-length Matriptase-1 or HAI-1 compared to SCRIB-KO cells in which Scribble was re-introduced via transfection with the corresponding expression plasmid (SCRIB KO + SCRIB; Fig 4H). Intriguingly, after 24 hours of incubation in different osmolalities, the positive effect of hypotonic stress on Matriptase-1 activity was significantly stronger in SCRIB KO cells compared to SCRIB KO + SCRIB cells ( Fig 4I). These data implicate that also in mammalian cells, disrupted epithelial cell polarity has Matriptase-activating capacity, which synergizes with the effect of hypotonic stress, similar to their proposed cooperation in zebrafish atp1b1a mutants.

Matriptase-1a functions in the periderm to activate oncogenic signaling in the underlying basal layer
To understand by which means epithelial polarity affects Matriptase function, we returned to the zebrafish embryos. Previous studies have shown that in the epidermis of wild-type embryos, the Matriptase-1 encoding st14a gene is expressed in both layers of the epidermis [21,25]. In contrast, atp1b1a is only expressed in the outer peridermal layer, in which it, however, is required to assure proper epithelial polarity in keratinocytes of both layers [26]. In this light, we first sought to identify the epidermal layer in which Matriptase-1 activity needs to be hyper-activated to mediate the oncogenic effects caused by the loss of ATP1b1a function. For this purpose, in a first set of experiments, we performed cell autonomy studies with mosaic embryos. Two types of chimeric embryos were generated via transplantation of ventral ectodermal cells, the progenitors of basal cells (Fig 5A), yielding clones of basal keratinocytes lacking both ATP1b1a and Matriptase-1 positioned underneath an ATP1b1a-deficient periderm containing Matriptase-1a (Fig 5B), and vice versa ( Fig 5C). In both cases, transplanted basal cells at 58 hpf display mmp9 expression levels according to the genetic constitution of the host periderm. Thus, loss of Matriptase-1 in the host / the periderm (atp1b1a MO, st14a MO) rescues clones of underlying basal cells (atpb1a MO) to wild-type condition, even though they do contain Matriptase-1 (no mmp9 expression in basal cells in brown to label eGFP protein; Fig 5C), whereas loss of Matriptase-1 in clones of basal cells themselves (atp1b1a MO, st14a MO) leaves them pre-neoplastic (strong mmp9 expression in basal cells labelled in brown; Fig 5B), as long as the host / the overlying periderm contains (unrestrained) Matriptase-1 activity (atpb1a1 MO).
Together, these data indicate that st14a, although it affects basal keratinocytes, does not act in basal keratinocytes themselves, but in other cells, most likely in the overlying periderm. To directly prove the latter, we employed a transgenic approach (peri:Gal4 [52];UAS:gfp-st14a) to specifically re-introduce Matriptase-1a fused with GFP into peridermal cells of atp1b1a,st14a double-deficient embryos. Immunofluorescence analysis confirmed that GFP-Matriptase1a was solely present in the periderm, where it is localized in lateral and basal plasma membrane  (Fig 5D). Driving peridermal expression of gfp-st14a in otherwise wild-type embryos has a mild effect on the epidermis, with minor aggregate formation (Fig 5E, 5F and 5M) and undetectable mmp9 expression (Fig 5I, 5J and 5N). Expression of gfp-st14a in atp1b1a,st14a double morphants, however, which per se also have an epidermis of wild-type appearance (Fig 5G, 5K, 5M and 5N), leads to strong aggregate formation and mmp9 expression in basal keratinocytes (Fig 5H, 5L, 5M and 5N), comparable to the epidermal defects in regular atp1b1a mutants (compare with Figs 1M and 3N).
Together, these data suggest that loss of ATP1b1a function in peridermal cells leads to increased signaling of Matriptase-1 from peridermal cells to underlying basal keratinocytes, thereby activating the oncogenic pathway and mmp9 expression in basal cells. This would be in line with cell non-autonomous functions of Matriptase described before in other contexts, for example in oral squamous cell carcinomas (OSCC), in which deregulated Matriptase from the cancer cells was suggested to activate Par2 on adjacent tumor-associated fibroblasts [53].

Epithelial polarity defects result in disrupted subcellular localization of Matriptase-1 in the periderm and affect Par2b in basal keratinocytes
But how does loss of ATP1b1a in peridermal cells cause increased Matriptase-1 signaling to underlying basal keratinocytes? Could the loss of Atp1b1a as a regulator of epithelial cell polarity affect Matriptase-1 localization in peridermal cells and thereby Par2b in underlying basal keratinocytes? In mammalian cells, Matriptase-1 has been shown to be targeted to basolateral membranes [54,55], co-localizing with E-cadherin and basolateral determinants like Na + K + -ATPase. To analyze Matriptase-1 localization in the zebrafish periderm, 1-cell stage embryos were injected with a krt4:egfp-st14a -containing plasmid to express the GFP-Matriptase-1 fusion protein (Fig 5D) specifically in single peridermal cells (Fig 6A-6D). Orthogonal views and quantifications on sum of slices projections demonstrate that in wild-type embryos, the fusion protein is mainly localized at lateral and less at basal cell membranes (Fig 6A',6A" and 6D), whereas in atp1b1a mutants (Fig 6B',6B" and 6D) and in morphants lacking the epithelial cell polarity regulator Lgl2 [38] (Fig 6C',6C" and 6D), relatively higher amounts of GFP are detected at the basal side. This indicates that polarity defects in the periderm as caused by the loss of ATP1b1a can result in an accumulation of Matriptase-1 on the basal side of peridermal cells, thereby bringing more of it into direct contact with potential substrates present on the apical surface of underlying basal keratinocytes.
According to our functional data described above (Figs S2 and 4), the Proteolysis-activated receptor Par2b could be one such substrate. To investigate whether it is indeed affected in basal keratinocytes of atp1b1a mutants, we generated a stable tp63:par2b-gfp fr59Tg transgenic line expressing a Par2b-GFP fusion protein specifically in basal keratinocytes. In 54 hpf wildtype embryos, Par2b-GFP mainly localizes to the plasma membrane of keratinocytes (Fig 6E  and 6H). Strikingly, in atp1b1a mutants, Par2b-GFP was additionally observed in internal controls (E,F) and injected with atp1b1a and st14a morpholinos (atp1b1a MO, st14a MO) (G,H); lateral views of entire embryos (E-H), and magnified views of tail region of same embryos (E'-H'). In contrast to atp1b1a MO, st14a MO embryos without transgenic re-introduction of st14a (G,G'), the atp1b1a MO, st14a MO embryos transgenic for peri: Gal4 zc1044a ; UAS:eGFP-st14a displays epidermal aggregates (H,H'), comparable to global atp1b1a single mutants (compare with Fig 1L'

PLOS GENETICS
vesicle-like structures of some, but not all (see Discussion) basal cells (Fig 6F and 6H). This subcellular distribution is similar to that reported for zfPar2b-GFP expressed in HEK293 cells once cells are treated with trypsin to activate Par2b, most likely reflecting activation-induced endocytic internalization of the receptor [56]. In atp1b1a-/-, st14a morphant zebrafish embryos, however, no such Par2b-GFP internalization was observed (Fig 6G and 6H), indicating that the effect depends on Matriptase-1. Together with the data described above identifying the periderm as the site of Matriptase-1 action (Fig 5), this suggests that aberrant activation of Matriptase-1 caused by the loss of ATP1b1a in peridermal cells does indeed lead to increased Par2b activation in underlying basal cells (and thereby their pre-neoplastic transformation).

Epithelial cell polarity defects combined with hypotonic stress causes Matriptase-dependent pre-neoplastic epidermal transformations in zebrafish embryos with wild-type ATP1b1a
Our results described thus far have revealed that hypotonic stress and epithelial polarity defects are necessary for Matriptase-dependent pre-neoplastic epidermal transformations in the context of ATP1b1a loss of function. But are they also sufficient to do so per se, and can this be a more general mode of oncogenic Matriptase activation? To find out, we finally studied whether epithelial polarity defects and hypotonic stress per se have similar Matriptase-dependent oncogenic effects on the embryonic zebrafish epidermis even in the presence of functional ATP1b1a. To induce systemic hypotonic stress, wild-type embryos were injected with antisense MOs against pax2a, a transcription factor required for proper pronephros function. Zebrafish pax2a mutants have been described to display pronephros-specific disruption of the Na + /K + -ATPase α subunit localization. In addition, most likely as a consequence of pronephric malfunction and the resulting systemic hypotonic stress, they develop pericardial edema [57], thus defects very similar to the osmoregulatory traits of atp1b1a mutants. To induce epithelial polarity defects, we again targeted the cell polarity determinant Lgl2 [38,39], whose subcellular localization is disrupted in keratinocytes of atp1b1a mutants [26] (Fig 2H, 2I, 2K and 2L), by morpholino injection as well as using the CRISPR/Cas9 system to create F0 mutants (crispants). When injecting pax2a MO alone, all embryos develop pericardial edemas at 54 hpf, whereas only a small portion also develops few epidermal aggregates (Figs 7B, 7K and S4K, S4M). Similarly, knockdown or knockout of lgl2 by itself does not result in any obvious epidermal phenotype at 54 hpf (Figs 7A, 7K and S4J, S4M), although the protein is absent (S4D-S4F Fig). In contrast, combined knockdown of pax2a and lgl2 leads to a strong increase in epidermal aggregate formation (Figs 7C, 7K and S4L,S4M), and also induces mmp9 expression in basal keratinocytes (Fig 7H and 7L), as observed in atp1b1a mutants (Fig 3N), whereas single loss of pax2a or lgl2 does not (Fig 7F, 7G and 7L). When pax2a and lgl2 double morphant embryos are raised in isotonic E3 containing 250 mM Mannitol, pericardial edema formation is strongly suppressed and epidermal aggregate formation and mmp9 expression are completely abrogated (Fig 7D, 7I, 7K and 7L), whereas concomitant genetic or MO-induced loss of st14a suppresses epidermal aggregate formation and mmp9 expression in basal keratinocytes only, but not edema formation (Fig 7E, 7J, 7K and 7L). Furthermore, basal keratinocytes of pax2a, lgl2 double morphants show increased pAKT levels, indicative of PI3K signaling, while blockage of PI3K by Wortmannin or LY294002 restores wild-type pAKT levels (Fig 7M-7O) and rescues epidermal aggregate formation and mmp9 expression (Fig 7P-7R, 7T-7V, 7X and 7Y). Similarly, inhibiting mTORC1 signaling by Rapamycin prevents epidermal aggregate formation and mmp9 expression in basal keratinocytes (Fig 7S, 7W, 7X and 7Y). Together, this indicates that although induced by other means, loss of epithelial polarity in combination with systemic hypotonic stress activates the same Matriptase-PI3K-AKT-mTORC1 pathway resulting in epidermal hyperplasia and mmp9 expression, as in the case of loss of ATP1b1a. This points to St14/Matriptase-1 as an oncogene that can be activated in vivo by multiple means and downstream of different mono-or multigenic insults to exhibit its carcinogenic properties.

Discussion
Matriptase-1 plays a major role in epithelial development and homeostasis from zebrafish to mice and humans. It also can act as an oncogene, making its tight regulation crucial, for instance via its tumor-suppressing cognate inhibitor Hai1 [1,2]. Also in zebrafish embryos, deregulated Matriptase due to the loss of Hai1a results in preneoplastic transformations in

The activity of Matriptase can be increased by hypotonic stress
Matriptase is synthesized as a zymogen, which can be activated via proteolytic cleavage by other proteases, for example by Tmprss2, enhancing its effect on the growth and invasiveness of prostate cancer [58]. However, in most cases, Matriptase-1 is suggested to primarily undergo autoactivation [7]. Several non-protein factors have been shown to promote this autoactivation, including small molecules like the phospholipid sphingosine-1-phosphate [8], the small molecule suramin [59], but also reactive oxygen species (ROS) [60]. Interestingly, the cellular inorganic environment also has a crucial impact on Matriptase-1 autoactivation, which is highest in mildly acidic and low ionic strength conditions [10,11,12]. Here, we present evidence that the pericellular tonicity has a substantial effect on Matriptase activation and activity, both in zebrafish embryos in vivo as well as in cultured human embryonic kidney (HEK293) and breast epithelial (MCF-10A) cells (Fig 4). Hypotonic growth media substantially increased zebrafish Matriptase-1a-mediated cleaveage of zebrafish Par2b in HEK293 cells, and increased amounts of processed / active endogenous Matriptase-1 in MCF-10A cells, implicating a conserved regulation of Matriptase activity by ionic strength. In vivo, epidermal tumor formation in zebrafish atp1b1a mutant embryos suffering from compromised kidney function is fully dependent on Matriptase activity. Unfortunately, due to their small size, it is technically impossible to directly measure the tonicity of the interstitial fluid within zebrafish embryos and their epidermis; yet, systemic tonicity most likely is strongly reduced in atp1b1a mutants and in pax2a, lgl2 double morphants, as reflected by the formation of pericardial edema, a characteristic trait for zebrafish mutants with compromised kidney function [61,62]. Accordingly, incubation of atp1b1a mutants and pax2a, lgl2 double morphants in isotonic medium not only heals their pericardial defects (thus interstitial tonicity), but also epidermal aggregate formation (Figs 1 and 7). In sum, these data indicate that a hypotonic pericellular environment can promote Matriptase activation and hence, constitutes a risk factor in Matriptase-dependent tumorigenesis.

Matriptase activity is regulated by the epithelial cell polarity control system affecting its subcellular localization
However, hypotonicity by itself only has very minor oncogenic effects on the embryonic zebrafish epidermis, indicated by very low rates of epidermal aggregate formation and mmp9

PLOS GENETICS
expression in basal keratinocytes of pax2a morphant embryos. Rather, hypotonic stress can only induce tumor formation in combination with loss of epithelial polarity in epidermal cells (Fig 7 and [26]). It has been known for decades that carcinogenesis usually results from mutations in more than one gene [63], and disturbed polarity has been reported to contribute to tumorigenesis in several cases [32,33,34,35,36,37]. However, a functional connection between epithelial polarity regulators and Matriptase-1 in the context of carcinogenesis had, at least to our knowledge, not been reported thus far.
We show here that in MCF-10A breast epithelial cells, Matriptase activity is significantly increased upon the loss of Scribble (Fig 4), a regulator of epithelial cell polarity promoting the basolateral domain, like ATP1b and Lgl. This is in line with the increased Matriptase activity formerly reported for breast cancer cells (which have largely lost their epithelial polarity) in comparison to untransformed MCF-10A breast epithelial cells [9], and with our data obtained in zebrafish embryos, where loss of the epithelial cell polarity regulators atp1b1a and lgl2 both promote Matriptase-mediated pre-neoplastic processes (Figs 1 and 7).
Of note, however, loss of atp1b1a and lgl2 in zebrafish does not seem to lead to a general increase of Matriptase-1a protein levels or activity, but to changes in its subcellular localization. In wild-type peridermal cells, it is predominantly localized at lateral domains, whereas it shows a relatively higher abundance in basal domains upon loss of atp1b1a or lgl2 (Fig 6A-6D). It is not clear how epithelial / apico-basal cell polarity contributes to this differential lateral versus basal regulation. With respect to epithelial cell polarity, the bi-layered embryonic zebrafish epidermis constitutes an intermediate between mono-layered simple epithelia and multi-layered stratified epithelia [64,65,66]. Thus, a distinct apical domain and tight junctions are only present in the outer / peridermal cells, and a distinct basal domain with hemidesmosomes adhering to the underlying basement membrane only in basal keratinocytes. All other domains of cells in both layers display similar "lateral-like" properties, characterized by E-cadherin-dependent cell-cell contacts [67]. Yet, even these "lateral-like" domains display a certain epithelial polarity, with highest levels of the cell-cell adhesion molecule E-cadherin levels at the interface of the two cell layers [67]. However, loss of atp1b1a or lgl2 does not affect this E-cadherin polarity [67] or even causes decreased E-cadherin levels [23]. Therefore, the aberrant accumulation of Matriptase-1 in atp1b1a mutants on the basal side of peridermal cells most likely is not mediated via E-cadherin [68,69] but might be directly influenced by the disrupted cell polarity.

Matriptase can act in trans between different cell types and different cell layers to promote tumorigenesis
Strikingly, this shift of Matriptase-1a towards the basal side of peridermal cells of atp1b1a mutants might enable exposure of the protease to target substrates on basal cells that are normally constrained from cleavage due to the lack of direct physical contact (Fig 8C). Thus, in contrast to wild-type siblings, in which a transgene-encoded Par2b-GFP fusion protein was almost exclusively found at the cell surface of basal keratinocytes, a large fraction of basal keratinocytes of atp1b1a mutants displayed additional, Matriptase-dependent, localization of Par2b-GFP in intracellular vesicle-like structures. This implies that Par2b gets proteolytically activated and internalized in basal keratinocytes by Matriptase (Fig 6E-6H). In addition, our cell autonomy studies with chimeric embryos indicated that Matriptase is required in the peridermal cells to mediate the effect of loss of ATP1b1a function in inducing basal cell transformation (Fig 5). Together, this strongly suggests that in atp1b1a mutants, a shift towards the basal side of peridermal cells allows Matriptase to act in trans to cleave substrates such as Par2b on the underlying basal cells (Fig 8). Similar non-cell-autonomous functions of Matriptase have been reported before; for example Matriptase located on thymic epithelial cells can cleave the receptor tyrosine kinase Tie2 on endothelial cells, thereby inducing PI3K signaling in the target cells [70]. Moreover, in a human invasive oral squamous cell carcinoma line, deregulated Matriptase activity due to loss of Hai1 induces Par2 signaling in Par2-expressing cancer-associated fibroblasts, thereby promoting cancer progression [53]. These findings, together with our zebrafish data, show the potential of Matriptase to not only promote tumorigenesis in a cell-autonomous manner, cleaving targets in cis on the tumor cells themselves, but also in a non-cell-autonomous manner, cleaving target proteins on other cells in trans. Of note, the latter can occur in two ways, with Matriptase on tumor cells cleaving targets on the surface of other cells of the tumor microenvironment [53], or with Matriptase from cells of the tumor microenvironment cleaving targets on the prospective tumor cells, as demonstrated here for outer and (pre-neoplastic) basal keratinocytes of the embryonic zebrafish epidermis.

Aberrant regulation of Matriptase-1 in hai1a and atp1b1a mutants leads to the activation of different oncogenic pathways and responses
Aberrant Matriptase activity is associated with tumor initiation as well as progression and metastasis in a variety of epithelia-derived cancers [2], mediated by different pathways. For example, in some breast cancers and in human squamous cell carcinomas, Matriptase is responsible for cancer progression by increasing cell proliferation and invasiveness via an HGF (Hepatocyte growth factor)-cMET-pAKT pathway [50,71]. In other breast cancer cell lines, it signals via a Par2-phospholipase C γ2 (PLCγ2)-protein kinase C (PKC) pathway, resulting in the upregulation of NFkB and subsequent mmp9 expression responsible for metastasis [49].
In zebrafish embryos, deregulated Matriptase due to the loss of Hai1a has previously been shown to act mainly via Par2b, whereas downstream of Par2b, several pathways diverge to lead Mild hypotonicity (small blue star), most likely due to compromised epidermal integrity, further enhances Matriptase activity levels. Note that additional pathways downstream of Par2b have been described, which lead to additional preneoplastic events, like sterile inflammation; however, they do not include PI3K [23]. (C) More extreme hypotonicity in the pericellular space (due to loss of ATP1b1a or Pax2a; large blue star) causes (moderate) Matriptase activation even in the presence of Hai1a. This, however, only has subtle effects on epidermal cells (Fig 7K,7L), unless occurring in conjunction with the loss of epithelial polarity (due to loss of ATP1b1a or Lgl2), allowing Matriptase to shift towards the basal side of peridermal cells, thereby getting into physical contact and to cleave / activate Par2b and other, not yet identified targets (indicated by?) on underlying basal keratinocytes in trans. These targets activate a PI3K-pAKT-mTORC1-NFkB pathway in basal cells resulting in pre-neoplastic events like hyperproliferation, EMT and, in contrast to hai1a mutants, strong invasiveness of basal cells. https://doi.org/10.1371/journal.pgen.1010873.g008

PLOS GENETICS
to different pre-neoplastic events. One of them, an EGFR-Phospholipase D (PLD) pathway, branches off at the level of the PLD product phosphatidic acid and promotes 1) mTORC1 signaling and thereby epidermal cell proliferation and mmp9 expression, and 2) sphingosine kinase activity and thereby S1P production and apical cell extrusion of epidermal cells [20]. Additionally, Matriptase-activated Par2b in hai1a mutants can activate PLC, leading to sterile inflammation via PLC-induced IP3 signaling as well as interference with cell adhesion via PLC-induced DAG-MAPK-RSK signaling [23].
In the atp1b1a mutant, however, Matriptase induces an oncogenic pathway that is clearly distinct from those initiated in hai1a mutants. Thus, epidermal defects of both hai1a and atp1b1a mutants are fully dependent on Matriptase-1a, whereas pharmacological inhibition of EGFR and PLD only rescued hyperproliferation and mmp9 expression in hai1a, but not in atp1b1a mutants. Conversely, pharmacological inhibition of PI3K and NFkB only rescued atp1b1a, but not hai1a mutant epidermal defects ( [20] and this work). These observations demonstrate that Matriptase has the capacity to induce different downstream pathways also in the zebrafish epidermis. Whether aberrant Matriptase-1a activity in atp1b1a also induces PLCdependent pathways as in hai1a mutants [23], and whether these pathways can contribute to the atp1b1a mutant phenotype remains to be determined.

Future perspectives
Future studies also need to elucidate in more detail the molecular mechanisms via which the different modes of aberrant Matriptase-1 regulation lead to the activation of different oncogenic pathways and different cellular responses. Of note, whereas oncogenic effects in the hai1a mutant seem to fully depend on the direct Matriptase-1 target Par2b [20,23], loss of par2b only partially blocks tumorigenesis in atp1b1a mutants. This suggests that in atp1b1a mutants, Par2b might act in partial functional redundancy with other Matriptase-1 targets to preferentially promote the oncogenic PI3K-AKT-mTORC1-NFkB pathway. Potentially, these other targets can only be activated in atp1b1a mutants as here, disrupted cell polarity might enable their contact with Matriptase-1a, whereas in hai1a mutants, in which cells are still polarized, this contact is spatially restricted. Preliminary data obtained via combined rescue experiments of atp1b1a mutants point to cMET-related receptor tyrosine kinases [50, 71,72] as such potential additional Matriptase-1 targets with partial functional redundancy to Par2b, possibly contributing to the more invasive properties of transformed basal keratinocytes in atp1b1a [26] compared to hai1a [21] mutants. If so, blockade of these receptor tyrosine kinases should alleviate the highly invasive phenotype of keratinocytes in atp1b1a mutants to the milder "hai1a"-like phenotype mediated by Par2b only.
Transcriptomics analyses are in progress to further elucidate the oncogenic outcomes downstream of Matriptase and its different pathways in hai1a versus atp1b1a mutants and their potential impact on the higher aggressiveness of transformed keratinocytes in the atp1b1a mutant. Of note, the Matriptase-encoding gene st14a is not only expressed in peridermal cells, but also in basal cells themselves [21]. In this light, it will be interesting to investigate whether Matriptase-1 takes over a crucial role in transformed basal keratinocytes themselves during later steps of carcinogenesis, as reported formerly in the context of metastasizing human squamous cell carcinomas [53]. In addition, single cell RNAseq approaches might help to elucidate potential epigenetic heterogeneities [73,74] among the basal keratinocytes in each mutant itself. Indeed, although all cells in the affected relatively young embryos should be genetically identical, epidermal aggregates of atp1b1a mutants are formed as discrete cell clusters, and intracellular Par2b, the likely direct result of increased Matriptase activity (Fig 6E-6H), as well as increased NFkB activity downstream of Matriptase (Fig 3G-3I and [26]), were observed in similar patterns. In this light, it is tempting to speculate that differential levels of Matriptase-1 activation might contribute to such intratumoral heterogeneity.

Ethics statement
All zebrafish experiments were approved by the local and federal animal care committees

Generation and genotyping of st14a fr56 CRISPR/Cas mutant
The guide RNA GGTGATCCTGGCAGCTGTTT targeting exon 3 of st14a was designed using the webtool at http://crispr.mit.edu/. The specific Alt-R CRISPR-Cas9 crRNA and the universal Alt-R TM tracrRNA were obtained from IDT, Belgium, and annealed in equal amounts of 30 μM each in Nuclease-Free Duplex Buffer (IDT) by heating to 95˚C for 5 min and subsequent cooling to room temperature. 9 μM of complexed RNA was injected together with 150 ng/μl of Cas9 mRNA generated via the SP6 mMessage mMachine kit (Ambion) from plasmid pCS2-Cas9 (gift from Alex Schier; Addgene plasmid # 47322; http://n2t.net/addgene:47322; RRID:Addgene_47322) [77] in Danieau buffer into the zygote. Embryos were raised and their progeny screened for germline transmission by PCR with the primers 5'-TTG ATT TCA GAG AGA CCG GAA T-3' and 5'-TCT CCT TAT TTA GAT TAA GGC AAA AC-3', followed by a T7 endonuclease assay (NEB) to detect indels and to establish the line st14a with a 5 bp deletion in exon 3 (see Results). Genotyping was performed using the same primers followed by MwoI digestion, which results in a cleaved wild-type PCR product and non-cleaved mutant product.

Generation and genotyping of par2b fr57 CRISPR/Cas mutant
The par2b mutant was generated applying the same method, using as guide RNA TGGCGGTGTCCGAGAGCTAC targeting exon 1 of par2b. Embryos were raised and their progeny screened for germline transmission by PCR using the primers 5'-CGG CAG AAC TCA ACG CTT C-3' and 5'-AGA GCA ACG CAC AAA ACA GG-3' followed by a T7 endonuclease assay (NEB) to detect indels and to establish the line par2 fr57 with a 10 bp deletion in exon 1 (see Results). Genotyping was performed using the primer 5'-AGC TGG ATC TGA CTG GAT CG-3' in combination with the primer 5'-AAA TCC TGT AGC TCT CGG AC-3' to detect the wild-type allele and in combination with the primer 5'-AAT AAT AAA ATC CTG ACA CC-3' to detect the mutant allele.

Generation of lgl2 CRISPR/Cas9 crispants
The guide RNAs GCTTCACGACGAGAATGCGG and ATGAAAACCCGCTGAACCCG targeting exon 4 and 7 of lgl2 were designed using the webtool at http://chopchop.cbu.uib.no and obtained as above. 7.5 μM of complexed RNA of both gRNAs were incubated with 10 μg Alt-R S.p. HiFi Cas9 Nuclease V3 (IDT, Belgium) for 5 min at RT and injected either with or without pax2a MO into the zygote. To test for efficiency, PCR with the primers 5'-CGT GAT CCA TTA CAC CCC TAT T-3' and 5'-AAG CC ACA GGA TTA AAA GTC CA-3' for gRNA #1 and 5'-TTC CCA GAG TTC CAG AGG ATT A-3' and 5'-GAA ACA TTG CGA AAT ACT GCA C-3' for gRNA#2 was performed on DNA from 1 dpf embryos, followed by a T7 endonuclease assay (NEB).

Morpholinos
The following morpholinos were obtained from GeneTools (Philomath, CA), diluted in Danieau buffer to the indicated concentration, and 3 nl were injected into 1-cell stage embryos according to standard protocols.

Quantitative-RT-PCR (RT-qPCR)
Total RNA was isolated from 10-15 embryos at 54 hpf using Trizol (Thermo Fisher Scientific) following standard procedures, followed by a DNaseI treatment (Roche). First-strand cDNA synthesis was performed using reverse transcription (Promega). RT-qPCR was performed in triplicates with Sybr Select Master Mix (Life Technologies, Thermo Fisher Scientific) on an ABI-Prism 7500 Fast Detect system, and relative expression levels were calculated by the ΔΔCt method with gapdh as the control gene. Data are presented as fold change relative to the rele-

Plasmid generation and injection, and generation of transgenic lines
The construct uas:egfp-st14a was generated using the Tol2 kit [82], with the described uas promoter in the p5E vector, the egfp sequence in the pME vector, and the st14a cDNA sequence amplified with the following primers: 5'-GGG GAC AGC TTT CTT GTA CAA AGT GGC CAT GGA CCC TAT GGA TGG AGG A-3' and 5'-GGG GAC AAC TTT GTA TAA TAA

PLOS GENETICS
AGT TGC TTA CAC TCC CGT CTT CTC CTT G-3' in p3E. The obtained plasmid was injected to generate the stable transgenic line Tg(uas:gfp-st14a) fr58Tg that was used in conjunction with the periderm-specific driver Et(Gal4-VP16) zc1044a [52] (peri:Gal4; Fig 5) for exclusive expression of GFP-labelled Matriptase-1 in the outer epidermal cell layer. For mosaic expression of egfp-st14a in peridermal cells (Fig 6), a krt4:egfp-st14a construct was obtained as above, with the described krt4 promoter [83] instead of uas and injected into 1-cell stage wild-type embryos at a concentration of 10 ng/μl. Embryos were analysed at 48 hpf using confocal microscopy. The construct ΔNp63:par2b-egfp was generated by first using the Tol2 kit to C-terminally fuse the par2b coding sequence with egfp (the following primers were used to amplify par2b cDNA for insertion into pME:  [20]. The obtained plasmid was used to obtain the line Tg(ΔNp63:par2b-egfp) fr59Tg by standard injection and screening procedures, which was then used to analyse Par2b-eGFP subcellular localization.

Cell transplantations
Ventral ectodermal cells from Tg(Ola.Actb:Hsa.hras-egfp) vu119 transgenic donor embryos, either injected with atp1b1a MO or atp1b1a MO and st14a MO were transplanted at 6 hpf into the ventral ectoderm of atp1b1a, st14a double morphant or atp1b1a single morphant recipients. Chimeric recipients were raised in E3 embryo medium and fixed at 58 hpf.

Whole-mount in situ hybridization (WISH) and colorimetric immunostainings
Embryos were fixed in 4% paraformaldehyde (PFA) and WISH was performed as previously described [84]. mmp9 cDNA templates and DIG-labeled probes were synthesized with the Roche digoxygenin RNA synthesis kit, as described [21,38]. Combined colorimetric WISH and immunostainings were performed as described [21] using the primary antibodies mouse anti-GFP (Invitrogen; A10262, 1:300), or mouse anti-Tp63 BC4A4 (Zytomed, 1:200), and the secondary biotinylated horse anti-mouse IgG (Vector Laboratories). ABC amplification was performed using the Vectastain Elite ABC Peroxidase kit (Vector Laboratories), and embryos were incubated in DAB substrate (Sigma-Aldrich) and H 2 O 2 until development of a signal.

Microscopy
Images were taken using a Zeiss Confocal (LSM710 META), an AxioImager microscope (Zeiss) equipped with an Apotome, using the AxioVision software (Zeiss), an Axioplan2 microscope (Zeiss) using AxioVision software (Zeiss), or a Leica M165FC stereo microscope with DFC425C camera and the Leica Application Suite V3.8 software. Images were processed using the ImageJ software and Adobe Photoshop.

HEK293 cell culture and Alkaline Phosphatase (AP) release assay
Plasmids for the zebrafish Matriptase activity test were generated as follows: for pcDNA3.1 +st14a, the st14a cDNA was amplified using the oligos 5'-GGC CGG ATC CAC CAT GGA CCC TAT GGA TGG AGG AAT-3' and 5'-GGC CGC GGC CGC TTA CAC TCC CGT CTT CTC CTT-3', cloned into the EcoRV site of pBluescript SK(-) (Addgene), and from the obtained plasmid, a KpnI-NotI fragment containing the st14a cDNA was subcloned between the KpnI and NotI sites of pcDNA3.1. To obtain pcDNA3+AP-par2b, the SEAP coding sequence was cut out of pCMV-SEAP (Addgene) with EcoRI and XbaI and inserted into the EcoRI and XbaI sites of pCDNA3.1. Subsequently, the par2b coding sequence was amplified using the primers 5'-GGA GGC CTC CGA CTA CAA AGA CGA TGA CGA CAA GGA CGC CCA GCC AGG CAA AAA TGG-3' and 5'-GGC CGT TAA CTC AGC AAG TGC TGG TTT CCG TGT T-3' adding a flag tag as linker sequence between SEAP and aa21 of Par2b, and this PCR product was cut with StuI and HpaI and cloned into the HpaI site at the 3' end of SEAP, removing its stop codon.
The AP release assay was done as described by [25] with slight modification. Briefly, HEK 293T cells were plated in DMEM (Gibco) with 10% FBS at a density of 0.8 × 10 5 cells/well of a 24-well plate. Approximately 24 h later, cells were left untransfected, or transfected by using Lipofectamine 3000 reagent (Invitrogen) with 500 ng pcDNA3.1-AP-Par2b and 4 ng empty pcDNA3.1 or 500 ng pcDNA3.1-AP-Par2b and 4 ng pcDNA3.1-st14a. Approximately 24 hours later, wells were washed once with 0.5 ml DMEM medium containing 0.1% BSA and 20 mM Hepes, and 300 μl medium with different osmolarity was added to each well. The medium with isotonicity (320 mOsmol/l was prepared by adding 20% FBS to DMEM and diluted 1 to 1 with 150 mM Na-Glutamate. Media diluted 1 to 1 with 110 mM, 70 mM and 0 mM Na-Glutamate and measured by osmometer have an osmolarity of approximately 270, 230 and 150 mOsmol/l respectively. After being incubated for 45 min at 37˚C, 200 μl medium was removed and transferred to labeled tubes kept on ice. The samples were centrifuged at 13,000 rpm for 5 min, and 25 μl of each sample was added to 50 μl of assay buffer from the NovaBright Phospha-Light EXP Assay Kit (Invitrogen) and heated for 5 min at 65˚C. Samples were then cooled to room temperature, and moved into a 96-well plate. 50 μl reaction buffer containing AP chemiluminescent substrate from the kit was added to each well and incubated for 20 min. Chemiluminescence was measured in a microplate luminometer (Infinite 200 Pro, Tecan). All samples were in triplicate.

Generation of SCRIB knockout and rescue MCF10A cells
Scribble KO cells were engineered using the CRISPR/Cas9 gene editing approach with guide arms targeting Scribble's exon 1 (5'-GAA GCG GCA CTG TTC GCT GC-3'). Scribble guide arms were cloned into the pLX-sgRNA vector (Addgene #50662) and transfected into 293T cells along with the pCW-Cas9 vector (Addgene #50661) and lentiviral packaging vectors to produce lentivirus for co-transduction into MCF10A cells. Positive cells were treated using 2 μg/mL puromycin for pCW-Cas9 and 5 μg/mL blasticidin for pLX-sgRNA selection for 7 days. Following antibiotic selection, pCW-Cas9 only and Scribble KO (pCW-Cas9 and pLX-sgRNA containing) clonal cell lines were generated using serial dilution to isolate colonies. Scribble KO for each clone was confirmed using western blot and immunofluorescence microscopy. Full length SCRIB was cloned into a retroviral murine stem cell virus (MSCV) backbone containing a bicistronic internal ribosome entry site (IRES) for stable expression in MCF10A cells.

Statistics
Quantitative experiments were repeated at least three times, reaching similar results. Mean values and standard deviations of all individual specimens (biological samples; n) from one representative or all independent experiments (N) are presented, as specified in the respective figure legends. Statistical analysis was performed using Graph Pad Prism software. For comparison of multiple groups, one-way ANOVA with post-hoc Tukey's test was used; for comparison of two groups, an unpaired two-tailed Student's t-test was used to determine significance, for comparison of distributions of phenotypes, a Chi-square test was used. Obtained p-values are mentioned in the respective figure legends and provided in S1 Appendix.

S3 Fig. Increased AP release upon hypotonicity is dependent on Matriptase.
Reporter assay for Matriptase activity towards Par2 cleavage showing that increased AP release at low osmolalities is dependent on Matriptase. HEK293 cells were transfected with pcDNA3+AP-Par2b, and pcDNA3+AP-Par2b together with pcDNA3+St14a. After 24 hrs in regular / isotonic medium (tonicity of 320 mOsm), cells were exposed to media of 320 mOsm and 150 mOsm for 15 RNAs. (B-C). Images of ethidium bromide agarose gels showing PCR fragments obtained from DNA of single embryos and digested by T7 Endonuclease I. PCR fragments from embryos injected with both gRNA#1 and #2 (*) are cut compared to non-injected controls (no *), indicative of occurring indels. (D-F) Immunofluorescence for Lgl2 (red) and p63 (green) on the trunk region of whole mount 54 hpf embryos. Lgl2 is present at cell borders of basal cells in wt embryos (D) but absent from lgl2 morphants (E) and lgl2 crispants (F), revealing efficient disruption of the lgl2 gene (n = 8-10) Scale bar: 20 μm. (G-I) Brightfield images of a live 6 dpf wt embryo (G, G'), lgl2 morphant (H, H'), and lgl2 crispant (I, I'); magnified views of tail regions (G'-I'). Whereas the transient knockdown of lgl2 by morpholinos is not able to induce late epidermal defects, F0 Crispr/Cas9 mutants develop epidermal phenotypes starting at 5-6 dpf in about 60% of injected embryos (N = 3, n = 32-40), reminiscent of the lgl2/penner mutant (39, Fig 2B) and demonstrating loss of lgl2 function. (J-L) Brightfield images of live 54 hpf embryos crispant for lgl2 to induce polarity defects, morphant for pax2a to induce hypotonic stress, or both; overviews of entire embryos (J-L) and magnified views of tail regions (J'-L'). Whereas lgl2 knockout by Crispr/Cas9 alone does not result in epidermal defects at early stages, it causes epidermal aggregate formation when combined with the knockdown of pax2a by morpholinos (L,L'; indicated in L' with arrows) comparable to the knockdown of lgl2 by morpholinos (Fig 7A-7C, 7K), Scale bar: 500 μm in overview, 100 μm in magnified image. (K) Quantification of phenotypes of embryos with different genotypes / morphant conditions (N = 3, n = 27-53). Significances were determined via a one-way ANOVA and Tukey's post hoc test by comparing fraction of embryos displaying an epidermal phenotype (mild, medium, and strong) with no epidermal phenotype; columns with same superscript letter (a,b,c) are not significantly different (p>0.05). (TIF) S1 Appendix. Numerical data and statistical tests.