Inactivation of serine protease Matriptase1a by its inhibitor Hai1 is required for epithelial integrity of the zebrafish epidermis

Serine proteases regulate diverse cellular behaviors in numerous contexts,including in development, tumorigenesis and wound healing (reviewed in Kataoka et al., 2002). This is achieved either by cleaving a pro-protein to a functional form or by the inactivation/degradation of a substrate. Strict control of protease activity and expression is thus crucial for proper tissue ontogeny and homeostasis, and deregulated activity of a number of proteases is associated with human disease states.

Matriptase1 [also known as Matriptase, membrane-type serine protease 1(MT-SP1) and Suppression of tumorigenicity 14 (St14)] is a type II transmembrane serine protease, first identified in human breast cancer cells(Lin et al., 1999b; Shi et al., 1993), and is expressed in a broad range of epithelia(Kim et al., 1999; Oberst et al., 2001; Oberst et al., 2003; Takeuchi et al., 1999). Increasing interest in this enzyme has centered on its strong correlation with epithelial tumor progression and the deleterious effects resulting from altering its activity levels in vivo (reviewed in List et al., 2006). One crucial regulator of the proteolytic activity of Matriptase1 has been shown,in vitro, to be the membrane-associated serine protease inhibitor Hepatocyte growth factor activator inhibitor 1 (Hai1, also known as Spint1)(Benaud et al., 2001; Lin et al., 1999a), which is almost invariably co-expressed with Matriptase1(Oberst et al., 2001) and was first described as an inhibitor of the circulating serine protease Hepatocyte growth factor activator (Hgfa, also known as Hgfac)(Shimomura et al., 1997). The inhibitory activity of Hai1 is conferred via its extracellular Kunitz domains,which bind directly to the protease sites(Denda et al., 2002; Kirchhofer et al., 2003). Via the inhibition of both Matriptase1 and Hgfa, Hai1 is thought to be able to attenuate cell motility via limiting serine protease-mediated activation of the potent motogen Hepatocyte growth factor (Hgf, also known as Scatter factor, Sf) (Parr and Jiang,2006). In addition to Hgf, Matriptase1 has been shown to activate other proteins and zymogens, such as urokinase plasminogen activator (uPA,also known as Plau), protease activated receptor 2 (PAR2, also known as F2rl1), matrix metalloprotease 3 (MMP3), insulin-like growth factor binding protein-related protein-1 (IGFBP-rP1, also known as IGFBP7) and CUB domain containing protein 1 (CDCP1) (Ahmed et al.,2006; Bhatt et al.,2005; Jin et al.,2006; Lee et al.,2000; Takeuchi et al.,2000). Furthermore, Matriptase1 can degrade extracellular matrix and structural proteins, including Laminin and Fibronectin(Satomi et al., 2001). In vitro models have shown that Matriptase1 induces cell scattering and invasion via both Hgf-Met-dependent and -independent mechanisms(Forbs et al., 2005); however,the in vivo relevance of these downstream pathways remains largely untested.

Knockout of matriptase1 in the mouse produced epidermal structural and barrier defects due to insufficient processing of profilaggrin (Flg) to a mature form in cornifying outer keratinocytes(List et al., 2002; List et al., 2003). The hai1-knockout mouse dies in utero because of placental defects underscored by a loss of epithelial integrity of chorionic trophoblasts(Tanaka et al., 2005) and the underlying basement membrane (Fan et al.,2007). These placental defects are due to deregulated matriptase1 activity, as confirmed by a double-knockout strategy(Szabo et al., 2007). Similarly, squamous cell carcinogenesis caused by overexpressing matriptase1 in the mouse skin was rescued by the concomitant overexpression of Hai1(List et al., 2005). However,it remains to be demonstrated whether Hai1 repression of Matriptase1 is necessary for normal epidermal development.

Here, we describe the roles of two Hai1 and a Matriptase1 homologues during skin development in the zebrafish (Danio rerio), based on mutant and antisense-mediated loss-of-function studies. During embryonic and larval stages, the zebrafish epidermis is bilayered, consisting of a basal and an outer layer of keratinocytes (Le Guellec et al., 2004). As in mammals, basal keratinocytes are attached to the basement membrane via hemi-desmosomes and to each other via desmosomes(Sonawane et al., 2005), and are characterized by the expression of the p53-related transcription factorΔNp63 (also known as Tp73l - Zebrafish Information Network)(Bakkers et al., 2002). Zebrafish hai1a (also known as spint1la - Zebrafish Information Network; GenBank accession number NM_213152), hai1b(spint1lb; GenBank accession number EF424430) and matriptase1a are expressed in the developing basal layer of the epidermis. Via live imaging and marker analysis, we describe both epithelial and inflammatory skin phenotypes caused by loss of Hai1 activity, and thus provide direct genetic evidence for an essential role of Hai1 to maintain epithelial integrity of the epidermis during development. Furthermore, by inactivating both Hai1 and Matriptase1a, we demonstrate that Hai1 fulfills its epidermal function by blocking Matriptase1a. Finally, we show that deregulated Matriptase1a activity does not require the Hgf receptor Met or related receptors to induce the observed scattering of hai1 mutant keratinocytes, indicating that other Matriptase1 target proteins must be involved.

The hai1a allele hi2217 was isolated during an insertional mutagenesis screen (Amsterdam et al., 1999). hai1a mutant embryos were obtained from heterozygous or homozygous parents. Embryos lacking early Smad5 function and specification of ectodermal precursor cells were obtained by crossing mothers heterozygous for the dominant-negative allele dtc24 with wild-type males (Hild et al., 1999). The Tg(bactin:hras-egfp) (allele vu119) and the Tg(fli1a:egfp) (allele y1) transgenic lines have been described previously (Cooper et al.,2005; Lawson and Weinstein,2002).

Clusters of mGFP-labeled basal keratinocytes were obtained by homotypic(wild type→wild type; hai1a morphant→hai1amorphant; hai1a+hai1bmorphant→hai1a+hai1b morphant) transplantation of non-neural ectodermal cells from Tg(bactin:hras-egfp)transgenic donor embryos into non-transgenic hosts. Recipients were developed to 24 hours post fertilization (hpf), mounted in 1.5% low melting point agarose under E3 medium and analyzed by time-lapse confocal microscopy.

Total RNA was isolated from embryos using Trizol-LS (Invitrogen, CA) and cDNA synthesized with SuperscriptII reverse transcriptase (Invitrogen). Sequences corresponding to zebrafish orthologs of hai1a, hai1b,prgfr3 and hgfa were obtained from the zebrafish genome(Ensembl, Sanger Center), and were amplified via reverse transcriptase(RT)-PCR. To determine the full 5′ sequence of hai1b and prgfr3 cDNAs, 5′RACE was performed using the SMART RACE kit (BD Biosciences, CA).

Zebrafish cDNA fragments for hai1a, hai1b and prgfr3obtained by RT-PCR were cloned into pCRII-TOPO (Invitrogen) or pGEM-T Easy(Promega, WI). cDNA clones of matriptase1a (IMAGp998P136587), matriptase1b (IMAGp998H0614296) and ron (IMAGp998M1514792)were obtained from RZPD (Berlin, Germany). The matriptase1a insert was shuttled into pCRII-TOPO. For probe synthesis, plasmids were linearized with NcoI (hai1b, hgfa), NotI (matriptase1a,hai1a, prgfr3), EcoRI (matriptase1b) or SmaI(ron). Probes for hai1a, hai1b, matriptase1a, prgfr3 and hgfa were synthesized using Sp6 RNA polymerase (Roche, Mannheim,Germany), and probes for matriptase1b and ron with T7 RNA polymerase.

For matriptase1a overexpression studies, the heat-inducible construct pTol2-hse-GFP/matriptase1a was generated, cloning the matriptase1a cDNA into the bicistronic vector pSGH2(Bajoghli et al., 2004), which,in addition to matriptase1a, drives expression of GFP under the control of heat-shock elements (hse). Subsequently, the cassette was ligated into vector pT2AL200R150G, which contains Tol2 recognition sites to allow early genomic integration and widespread expression upon co-injection with Tol2 transposase mRNA (Kawakami et al.,2004; Urasaki et al.,2006). For transgene activation, injected embryos were transferred from 28°C to 39°C from 20-22 hpf.

In situ hybridizations were performed as previously described(Hammerschmidt et al., 1996),using probes for hai1a, hai1b, matriptase1a, myoD(Weinberg et al., 1996), met (Haines et al.,2004), pu.1 (spi1)(Lieschke et al., 2002), lcp1 (Herbomel et al.,1999), mpx (Lieschke et al., 2001) and e-cadherin (cdh1)(Babb et al., 2001). Whole-mount antibody stainings were visualized with the Vectastain ABC kit(Axxora) as described (Hammerschmidt et al., 1996), or with fluorescent secondary antibodies. For combined colorimetric in situ hybridizations and immunostainings(Fig. 1F,G,I and Fig. 7A), embryos underwent standard in situ hybridization, followed by a fixation for 4-6 hours in 4%paraformaldehyde/PBS, and a standard immunostaining(Hammerschmidt et al., 1996). Antibodies, dilutions used and sources were as follows: 4A4 anti-p63 (1:200,Santa Cruz), anti-GFP (1:400, Invitrogen), anti-E-cadherin (1:200, BD Biosciences), Alexa-Fluor-546 goat anti-mouse (1:400, Invitrogen) and Alexa-Fluor-488 goat anti-rabbit (1:400, Invitrogen).

Protein extracts of embryos were separated by 8% SDS-PAGE under reducing conditions and transferred to nitrocellulose membrane. Anti-E-cadherin antibody was used at 1:5000 dilution, and secondary HRP-coupled goat anti-mouse antibody (Dianova, Hamburg, Germany) at 1:10000.

MOs were obtained from Gene Tools (Philomath, OR) and diluted in Danieau's buffer (Nasevicius and Ekker,2000). MO solution (1.5 nl) was injected per embryo. hai1a MO (5′-ACCCTGAGTAGAGCCAGAGTCATCC-3′) and matriptase1a MO (5′-AACGCATTCCTCCATCCATAGGGTC-3′) were injected at 100 μM; ron MO(5′-AACAAGGTCTTTGGGCTGATGAACA-3′) and prgfr3 MO(5′-CTAAATGAGTGGCCCAATGGACCAT-3′) at 200 μM; and hai1bMO (sequence 5′-CACCACGAACCCATTTTTGATTGAT-3′), the met MO(Haines et al., 2004) and pu.1 MO (Rhodes et al.,2005) at 500 μM.

Epidermal cell proliferation was assessed by BrdU incorporation followed by combined α-p63 and α-BrdU antibody detection as described(Lee and Kimelman, 2002). Apoptotic cells were visualized by acridine orange staining as described(Furutani-Seiki et al.,1996).

Fluorescent images were taken with a Zeiss Confocal microscope (LSM510 META); all other microscopy was performed on a Zeiss Axiophot.

To identify genes with essential functions during zebrafish skin development, we performed an antibody-based screen on a previously described bank of insertional mutants (Amsterdam et al., 1999), staining larvae at 48 hours post fertilization (hpf)for p63 protein, a marker of basal keratinocytes(Lee and Kimelman, 2002). One mutant that we identified, hi2217, displayed aggregations of p63-positive keratinocytes (Fig. 1B) compared with the even distribution of these cells in wild-type siblings (Fig. 1A). Such aggregates were most prominent on the yolk sac and yolk extension, and were already visible at 24 hpf (see below). In addition, skin cells were seen floating in the chorion of mutant embryos from 24 hpf(Fig. 1C).

By determining the flanking sequences of the viral DNA in hi2217mutants (Amsterdam et al.,2004), we found that the mutagenic insertion had occurred in intron 1 of the hai1a gene, which is homologous to the hepatocyte growth factor activator inhibitor 1 (Hai1, Spint1) gene in mammals. Because the insertion is upstream of the first coding exon(Fig. 2A), it most probably leaves the structure and function of the Hai1a protein intact. However, it attenuated hai1a transcription to below detection thresholds. At 24 hpf and 48 hpf, wild-type embryos displayed strong hai1a expression in epithelia such as the epidermis, the olfactory epithelium, the otic vesicle, the gut, and the lateral line primordium and neuromasts(Fig. 1D,E). Combination with anti-p63 immunostaining revealed that, within the epidermis, hai1atranscripts were present around the p63-positive nuclei, pointing to expression in the basal layer (Fig. 1F,G). However, in contrast to wild-type siblings(Fig. 2B), no hai1aexpression could be detected by in situ hybridization in hai1amutants between 24 and 72 hpf (Fig. 2C; and data not shown). Consistent results were obtained using semi-quantitative reverse transcriptase (RT)-PCR of extracts from whole-mutant and wild-type sibling embryos at 24 hpf(Fig. 2D), indicating that the viral insertion strongly abrogates hai1a transcription or transcript stability. As further evidence that the viral insertion was causative of the observed skin phenotype, we injected an antisense morpholino oligonucleotide(MO) targeting the hai1a translation start site (hai1a MO)into wild-type embryos, which led to a robust phenocopy of the epidermal defects (compare Fig. 3G-I with 3A-F).

In addition to the epidermis, hai1a mutants also displayed disrupted morphology of the olfactory epithelium (see Fig. S1A-C in the supplementary material), another ectodermal derivative displaying prominent hai1a expression (Fig. 1D), whereas, in the pronephric ducts and the gut, hai1a-positive epithelia (Fig. 1D,F) derived from the mesoderm or endoderm, respectively,appeared unaffected by the mutation (see Fig. S1D-G in the supplementary material).

By searching genomic and EST databases of the zebrafish, we identified a second zebrafish hai1 gene, named hai1b. On the amino acid level, Hai1a and Hai1b are 43% identical, and both are more similar to mouse Hai1 (34 and 37% identity, respectively) than to mouse Hai2 (also known as Spint2 - Mouse Genome Informatics; both 14% identity), indicating that they are true paralogs that might have arisen from the genome duplication that has occurred during teleost evolution(Postlethwait et al.,1998).

Whole-mount in situ hybridizations revealed that, as is hai1a(Fig. 1D,G), hai1b is expressed in the basal epidermis (Fig. 1H,I), raising the possibility that Hai1a and Hai1b might have partly redundant functions. Therefore, we inactivated hai1b by injecting a specific hai1b MO into wild-type or hai1a mutant or morphant embryos. hai1b morphant embryos displayed completely normal morphology and normal p63 expression(Fig. 3J-L), whereas the skin defects of hai1a, hai1b double-morphant embryos(Fig. 3M-O) were much more severe than in hai1a single mutants(Fig. 3D-F) or morphants(Fig. 3G-I). In 24 hpf hai1a single mutants, aggregates of basal keratinocytes were largely restricted to the yolk sac, the yolk extension and the ventral trunk, whereas cell shedding mainly occurred at the border between the yolk sac and yolk extension (Fig. 3D,E). By contrast, hai1a, hai1b double-morphant embryos displayed keratinocyte aggregations and cell shedding in all regions of the skin(Fig. 3M-O). At 24 hpf, many embryos had already lysed (Table 1) or were about to die, most probably due to the loss of functional skin, whereas hai1a mutants often recovered quite well,with skin aggregations and lesions only remaining in the forming body fins(data not shown). Together, these data indicate that Hai1b, although dispensable during normal skin development, can, to some extent, compensate for the loss of Hai1a.

To study whether the formation of keratinocyte aggregates in hai1amutants might be caused by an over-proliferation of basal cells, we carried out BrdU pulse labeling at different time points of development, combined with anti-p63 immunostainings to specifically mark basal keratinocytes. BrdU labeling for 4 hours from 20 hpf and staining at 24 hpf revealed a moderate number of proliferating cells among the evenly spaced basal keratinocytes of wild-type siblings (Fig. 4A). In hai1a mutants, the number of BrdU and p63 double-positive cells was not noticeably increased (Fig. 4B). Importantly, nascent aggregates of basal keratinocytes consisted solely of BrdU-negative cells(Fig. 4B). However, at 48 hpf,similar 4 hour BrdU pulse experiments revealed a significant increase in the number of proliferating basal keratinocytes in hai1a mutants(Fig. 4D), whereas epidermal proliferation had largely ceased in wild-type siblings(Fig. 4C). In conclusion, hai1a mutant basal keratinocytes display hyper-proliferation, which,however, occurs too late to account for the initial aggregate formation.

Another explanation for aggregate formation could be a redistribution of basal cells. In line with this notion, p63-positive cells adjacent to the aggregates appeared sparser than in unaffected skin (compare Fig. 4B with 4A). Such redistribution would require that cells give up their initial epithelial organization and acquire mesenchymal-like properties. To examine this possibility, we carried out time-lapse confocal microscopy of clusters of fluorescently labeled basal keratinocytes in wild-type and hai1amutant or morphant embryos from 24 hpf. Regardless of the recorded region,wild-type cells showed typical epithelial characteristics. They were of regular polygonal shapes and remained tightly associated with their neighbors and in fixed locations relative to each other during the entire recorded time(90 minutes; Fig. 5A; in 4/4 movies of transplanted cells). Depending on the region within the embryo,cells from hai1a morphant basal keratinocytes displayed a strikingly different behavior. On the yolk sac (data not shown) and in the ventral trunk dorsal to the yolk sac extension, they had less regular and more fibroblastoid cell shapes, changed positions relative to each other, broke contacts to adjacent cells, and formed highly dynamic lamellipodia-like cell protrusions,as normally seen in mesenchymal cells (Fig. 5B; in 5/5 movies; see Fig. 6G for an indication of recorded region). However, this behavior was only observed in regions that are highly susceptible to aggregate formation, whereas regions of hai1a morphants that appeared morphologically normal, such as the head and the dorsal trunk, displayed normal epithelial characteristics of recorded basal keratinocytes(Fig. 5C; 4/4 movies). By contrast, mesenchymal behavior of basal keratinocytes started significantly earlier (18-20 hpf) in hai1a, hai1b double morphants, was more pronounced and was seen throughout the entire epidermis(Fig. 5D; 6/6 movies). We frequently observed keratinocytes crawling on top of each other(Fig. 5E; n=10). We conclude that loss of Hai1 compromises the epithelial integrity of the epidermis and causes keratinocytes to attain an inappropriate mesenchymal-like character and motility, which might account for the observed aggregate formation.

To address epithelial versus mesenchymal properties of epidermal cells with molecular markers, we also carried out immunostainings for the transmembrane adhesion protein E-cadherin. In basal keratinocytes of wild-type embryos at 24 hpf, E-cadherin was localized at the membrane(Fig. 5F). By contrast, in hai1a mutant embryos, little to no membranous E-cadherin was detected in basal keratinocytes of the yolk sac or the yolk extension, whereas more E-cadherin signals were obtained in cytoplasmic locations(Fig. 5G). Whole-mount in situ hybridizations and western blotting experiments showed that total E-cadherin mRNA and protein levels were unaltered even in hai1a mutants injected with hai1b MO. By contrast,E-cadherin protein levels were strongly reduced in smad5^dtc24 mutants(Hild et al., 1999), which lack basal keratinocytes, demonstrating that they are the major source of E-cadherin at this stage (Fig. 5H; and data not shown). This suggests that E-cadherin in basal keratinocytes of hai1 mutants has been redistributed, indicating the loss of epithelial and the incomplete acquisition of mesenchymal properties.

In mammals, diseases of the epidermis are often either due to, or invoke,an excessive inflammatory response(Thivolet et al., 1990). To determine whether hai1a mutants display similar defects, we examined leukocytes using the in situ markers leukocyte-specific-plastin(lcp1) (Herbomel et al.,1999; Meijer et al.,2007) and myeloid-specific peroxidase (mpx)(Lieschke et al., 2001). At 24 hpf, wild-type embryos showed a few leukocytes below the epidermis of the yolk sac and the anterior trunk (Fig. 6A; and data not shown), in concordance with previous reports(Herbomel et al., 2001). hai1a mutant embryos at this stage, by contrast, had a strong accumulation of leukocytes over the yolk sac and yolk extension, corresponding to sites of epidermal aggregate formation(Fig. 6B; and data not shown). At later stages (48 hpf), enhanced inflammation was also seen at other sites in hai1a mutant embryos, such as the posterior trunk and fins(compare Fig. 6E with 6D).

In order to determine whether this inflammation is causative of the epithelial defects described above, we genetically ablated innate immune cells in hai1a mutants by injecting MOs against pu.1, which is required for the specification of the myeloid lineage(Rhodes et al., 2005). In situ hybridizations revealed a complete loss of leukocytes in pu.1morphant embryos (Fig. 6C). However, epidermal defects of hai1a mutants were not ameliorated(Fig. 6C), suggesting that inflammation does not cause, but might instead be induced in parallel to or by, the epidermal defects of hai1a mutants.

In several instances, inflammation is induced by dying cells to ensure the proper clearance of apoptotic cell debris by innate immune cells (reviewed in Henson and Hume, 2006). The same might be true for the epidermal inflammation of hai1a mutants. As early as 24 hpf, epidermal aggregates of hai1a mutants contained a high number of acridine orange-positive apoptotic keratinocytes(Fig. 6G,H), whereas keratinocyte death was not evident in the epidermis of wild-type embryos(Fig. 6F). Similar results were obtained by TUNEL labeling (data not shown). Time-lapse movies of hai1a mutants carrying a fli1:EGFP transgene(Redd et al., 2006) further showed that yolk sac leukocytes were strongly attracted by epidermal aggregates with acridine orange-positive cells(Fig. 6J), whereas, in wild-type controls, leukocytes moved more slowly and were less directed(Fig. 6I). In summary, this suggests that epidermal inflammation of hai1a mutants is secondarily caused by the death of keratinocytes.

Hai1 is known to inhibit a number of serine proteases, including the membrane bound protease Matriptase1. To ascertain whether this target was implicated in the hai1 mutant phenotype, we identified and cloned two zebrafish matriptase1 orthologs - matriptase1a(st14a; GenBank accession number BC115342) and matriptase1b(st14b; GenBank accession number BC125837). While matriptase1b was expressed in the central nervous system (data not shown), matriptase1a showed strong epidermal expression starting at approximately 18 hpf, immediately prior to the onset of the hai1mutant phenotype. Expression at 24 hpf appeared notably similar to that of hai1a, with transcripts in the basal layer of the epidermis, the otic vesicle, the olfactory epithelium and the gut(Fig. 7A,B).

To test whether the epidermal defects of hai1 mutants might be due to a lack of Matriptase1a inhibition, we attempted to phenocopy and rescue the defects by Matripase1a overexpression or inactivation, respectively. Heat-treated wild-type embryos that had been injected with pTol2-hse-GTP/matriptase1a DNA (see Materials and methods)displayed severe dissociation of GFP-positive keratinocytes and skin aggregate formation both in dorsal (data not shown) and ventral(Fig. 7C; n=7/8 embryos with GFP-positive keratinocytes) positions, similar to the phenotype of hai1 mutants/morphants (Fig. 3). Defects were rescued to wild-type condition upon co-injection of pTol2-hse-GTP/matriptase1a with an MO targeting the matriptase1a translation start site(Fig. 7D; n=0/6 embryos with GFP-positive keratinocytes), indicating that they are indeed due to a gain of Matriptase1a activity. Similarly, whereas the knockdown of Matriptase1a in wild-type embryos had no effect (compare Fig. 7F with Fig. 7E), hai1a,matriptase1a double- and hai1a, hai1b, matriptase1atriple-mutant/morphant embryos displayed a robust rescue of the hai1mutant phenotype, with epidermal morphology indistinguishable from that of wild-type siblings (compare Fig. 7G,H with Fig. 3E,N; Table 1). In particular, basal keratinocytes did not form aggregates and were not shed, but maintained their even distribution, as was seen in wild-type embryos (compare Fig. 7I with Fig. 3F,O). Similarly, in vivo imaging revealed that keratinocytes dorsal of the yolk extension of matriptase1a-MO-injected hai1a mutants maintained epithelial properties (Fig. 7K; 4/4 movies of transplanted cells), in contrast to the acquisition of mesenchymal-like characteristics in hai1a mutants at this location, as described above(Fig. 5B). Furthermore,apoptosis of keratinocytes was suppressed (compare Fig. 7J with Fig. 6G,H), as was skin inflammation (compare Fig. 7M with 7L). In conclusion, all phenotypic traits caused by the loss of Hai1 activity could be rescued by concomitant inactivation of Matriptase1a,indicating that, during normal development, Hai1 promotes skin homeostasis by blocking Matriptase1a activity.

Biochemical studies have identified multiple extracellular target proteins of Matriptase1. A prominent example is Hgf, which, upon binding to its receptor, Met, can induce epithelial scattering and epithelial-mesenchymal transitions (EMT). In vitro, Matriptase1 can proteolytically cleave the biologically inactive Hgf precursor protein, thereby releasing mature Hgf. Inactivation of the zebrafish Met or Hgf orthologs via MO or antibody injection abrogates the migration of pectoral fin muscle precursors and presumptive neuromasts of the lateral line system(Haines et al., 2004). To study whether elevated Hgf signaling might account for the loss of epithelial integrity seen in the hai1a mutant epidermis, we injected hai1a mutants with met MO. However, in contrast to Matriptase1a (Fig. 7G),knockdown of Met activity did not alleviate epidermal aggregate formation(compare Fig. 8C with 8A,B) or inflammation (data not shown), although embryos lacked pectoral fin muscles(compare Fig. 8F with 8D,E),demonstrating that the MO was functional. To discount potential redundancies with other related receptors, we also cloned zebrafish orthologs of the closely related mammalian Ron receptor (GenBank accession number CF997695) and a Ron-relative most similar to avian Sea and Fugu rubripes, Prgfr3[Plasminogen related growth factor receptor 3(Cottage et al., 1999);GenBank accession number EF424429]. However, we were unable to detect expression of any of these receptor genes in the zebrafish epidermis at 24 hpf(data not shown). Furthermore, co-injection of MOs against all three members of this receptor family (Met, Ron and Prgfr3) failed to rescue the skin defects of hai1a mutants (see Fig. S2E-H in the supplementary material), although all MOs efficiently suppressed the translation of their respective mRNA upon co-injection (see Fig. S2A-D in the supplementary material). We conclude that the defects caused by the loss of Hai1 and gain of Matriptase1a activity are not due to enhanced signaling via the Hgf receptor Met or any of its relatives.

In the mouse, Hai1-mediated inhibition of Matriptase1 is required for proper placenta formation, with mutant embryos resorbing at embryonic day(E)12 (Tanaka et al., 2005). These early defects have precluded studies on possible other essential roles of Hai1 during later development, such as in the skin. The recent observation that Hai1, Matriptase1 double-mutant mice develop to term(Szabo et al., 2007) and only show skin defects associated with the loss of Matriptase1(List et al., 2002; List et al., 2003) could have two meanings. It could indicate either that Hai1 is dispensable for other developmental processes, or that, as in the placenta, the role of Hai1 in such other tissues is to block Matriptase1. Tissue-specific knockout of the mouse Hai1 gene would be necessary to distinguish between these two possibilities.

Here, we have analyzed the skin phenotype caused by loss of Hai1 function in a non-placental vertebrate, the zebrafish. Skin defects of hai1amutants are most prominent around 24-26 hpf, but recover to almost wild-type condition during further development. By contrast, combined loss of Hai1a and its paralog, Hai1b, which, per se, is dispensable, causes a much stronger phenotype, characterized by a dissociation of the entire epidermis and embryonic death between 18 and 26 hpf. Both the moderate defects of hai1a mutants and the stronger defects of hai1a mutants or morphants injected with hai1b MO were rescued upon injection of a matriptase1a-specific MO, indicating that Hai1a and Hai1b act in partial redundancy and by blocking Matriptase1a activity. Our data are in line with results obtained upon transgene-driven overexpression of Matriptase1 in mouse keratinocytes, which leads to epidermal hyperplasia, skin inflammation and heightened tumorigenicity in the adult epidermis, all of which can be suppressed upon concomitant skin-specific overexpression of Hai1(List et al., 2005). In addition to providing the first direct proof for an indispensable role of Hai1 in the vertebrate epidermis, our data also highlight that the primary function of Hai1 is concerned with the maintenance of epithelial integrity within the basal epidermal layer, whereas skin inflammation and epidermal hyperplasia appear to be secondary consequences.

At 24 hpf, aggregation and shedding of keratinocytes in hai1a^-/- homozygotes was largely restricted to the ventral side of the embryo, and most prominent on the yolk sac and the forming yolk extension, whereas, at 36 hpf, aggregates were mainly found on the outgrowing body fins. These are the regions of the embryo with the most-pronounced morphological changes, suggesting that here, the epidermis is exposed to highest mechanical stress and undergoes tissue remodeling even during normal development. Such remodeling processes most probably involve a transient loss of epithelial properties of keratinocytes, which might be driven by Matriptase1a. If so, however, it remains unclear why the loss of Matriptase1a activity does not cause developmental defects. Functional redundancy with other serine proteases could be one explanation. In addition, Matriptase1a might only become essential for tissue remodeling that occurs during pathological conditions, such as wound healing. To test this notion, we compared wound healing in wild-type and matriptase1a morphant larvae,which, at this stage, normally involves actin cable-driven purse-string contractions of keratinocytes, as in mammalian embryos(Redd et al., 2004). This process occurred normally in wounded matriptase1a morphants, (T.J.C. and M.H., unpublished observations). Further analyses are necessary, including wound-healing studies of matriptase1a mutants during adulthood, when cutaneous wound closure involves EMT of keratinocytes (T.J.C., Krasimir Slanchev and M.H., unpublished observations).

Nonetheless, our in vivo time-lapse recordings of fluorescently labeled hai1a mutant keratinocytes revealed that, during development, these cells have the potential to become mesenchymal. We observed such behavior,localized to ventral regions, in hai1a single mutants at 24 hpf,whereas the loss of both Hai1a and Hai1b led to the enhanced, earlier and more-widespread acquisition of mesenchymal-like properties of basal keratinocytes. In these movies, we also observed mutant basal keratinocytes crawling on top of each other. Furthermore, mesenchymal-like behavior was observed in keratinocytes far removed from aggregates, suggesting that the acquisition of fibroblastoid properties and motility is causative, rather than a secondary consequence, of aggregate formation.

It is important to note that hai1 mutant keratinocytes did not become completely mesenchymal. Thus, they lacked the mesenchymal marker Vimentin (T.J.C. and M.H., unpublished data). In addition, mRNA and protein levels of the epithelial marker E-cadherin remained unchanged. However,E-cadherin protein appeared to be redistributed, indicating a loss of epithelial polarity. Acquisition of fibroblastoid shapes and motility under such conditions is sometimes termed `scattering', in reference to its initial discovery as an effect induced by Scatter factor (Sf, identical to Hgf), and is opposed to complete EMTs, which require additional and/or prolonged signaling (for a review, see Grünert et al., 2003).

In addition to keratinocyte scattering, loss of Hai1 activity leads to keratinocyte death, enhanced skin inflammation and epidermal hyperplasia. Although the phenotype caused by the combined loss of Hai1a and Hai1b was too severe to dissect these traits, analyses of hai1a single mutants suggest that all are secondary consequences of the loss of epithelial integrity. Thus, apoptosis of keratinocytes was usually observed only in the most severely affected regions and within cell aggregates, whereas keratinocyte scattering was also recorded in regions devoid of dying cells. This suggests that keratinocyte death might be a secondary consequence of their detachment from the basement membrane when they pile up on each other, a phenomenon called anoikis, which has been well characterized in vitro(Gilmore, 2005). This death of keratinocytes might in turn induce skin inflammation, characterized by enhanced numbers of leukocytes in the hai1a mutant skin. Strikingly,the spatial pattern of inflammation resembled that of keratinocyte apoptosis(compare Fig. 6B with 6G). Whereas apoptosis preceded inflammation and had its peak between 24 and 28 hpf, inflammation continued beyond 48 hpf. Furthermore, time-lapse recordings revealed that innate immune cells were strongly attracted by apoptotic keratinocytes. Ultimate proof for a causative relationship between keratinocyte death and inflammation would require analysis of inflammation in hai1a mutants after the suppression of cell death. For this purpose,we injected mutants with MOs targeting the pro-apoptotic transcription factor p53 (Plaster et al., 2006) or with mRNA encoding the anti-apoptotic Bcl-2 protein(Langenau et al., 2005). However, both treatments failed to lower the number of dying keratinocytes in hai1a mutants (T.J.C. and M.H., unpublished data), suggesting that keratinocyte death is accomplished independently of the intrinsic, and possibly driven by the extrinsic, apoptosis pathway [compare with Rytomaa et al. (Rytomaa et al., 1999)]. Although it cannot be ruled out that inflammation in hai1a is induced by other means in parallel to dying keratinocytes, we can rule out that inflammation is causative of the epithelial defects, because they persisted in hai1a mutants after genetic ablation of all innate immune cells by pu.1-MO injection.

Finally, our BrdU-incorporation studies indicate that proliferation of keratinocytes is only secondarily affected, most probably due to the loss of epithelial integrity and possibly contact inhibition. Thus, at 24 hpf,keratinocyte aggregates in hai1 mutants were solely composed of non-proliferating cells, whereas keratinocyte hyper-proliferation was only seen later and after the onset all other phenotypic traits (48 hpf). Similarly, transgenic mice overexpressing Matriptase1 in keratinocytes display late epidermal hyperplasia in vivo, although isolated keratinocytes show normal proliferative behavior when cultured in vitro(List et al., 2005), arguing for an indirect effect of the Hai1-Matriptase1 system on cell proliferation.

Biochemical analyses have identified multiple Matriptase1 substrate proteins; among them is Hgf, which is involved in the regulation of several epithelial-mesenchymal transitions during vertebrate development (for a review, see Birchmeier and Gherardi,1998) and which requires Matriptase1- or Hgfa-mediated proteolytic cleavage to become biologically active. This suggests that the epithelial defects of hai1 mutants might be due to Matriptase1-mediated upregulation of Hgf activity. Furthermore, applied Hgf has been shown to attract innate immune cells to skin wounds(Bevan et al., 2004),suggesting that elevated Hgf activity might also account for the observed skin inflammation of hai1 mutants. Additionally, Hai1 and Matriptase1 could act via other related ligands such as Mst1 (Macrophage-stimulating protein), shown to stimulate chemotactic migration of macrophages(Leonard and Skeel, 1978) as well as keratinocyte mobility (Santoro et al., 2003). Indeed, a zebrafish mst1 homolog has been reported to be highly expressed on the yolk sac and the yolk extension(Bassett, 2003), the two most strongly affected sites of hai1a mutants. However, knockdown of the Hgf receptor Met (Haines et al.,2004), the Mst1 receptor Ron(Gaudino et al., 1994) and the related Plasminogen related growth factor receptor 3(Cottage et al., 1999) failed to rescue the epidermal defects of hai1a mutants. This strongly suggests that Hai1 and Matriptase1a regulate skin homeostasis and remodeling via other or additional proteins.

A recent report on the placental phenotype of Hai1 mutant mice has described compromised basement membrane integrity(Fan et al., 2007), suggesting that epithelial defects might be due to increased degradation of Laminins,other described Matriptase1 substrate proteins (see Introduction). However,whole-mount immunostainings and western blotting experiments of hai1amutant and hai1a, hai1b double-deficient embryos revealed unchanged Laminin protein levels and cleavage patterns (T.J.C. and M.H., unpublished data), suggesting that the epidermal defects are not due to direct alterations in Laminin processing.

In conclusion, although we can exclude some of the most obvious candidates as the prime or sole Matriptase1 target proteins, the crucial players downstream of Matriptase1, accounting for the epidermal defects caused by the loss of Hai1 activity, remain elusive. Systematic antisense-mediated knockdowns of other candidates in hai1a zebrafish mutants, as well as genetic hai1a suppressor screens, are in preparation to hopefully reveal the nature of these proteins in the future.

We are very grateful to John Kanki and A. Thomas Look (mpx, pu.1 MO); Thomas Czerny (pSGH2); Koichi Kawakami (pT2AL200R150G); and Christine and Bernard Thisse (lcp1) for sending reagents. The Tg(bactin:hras-egfp) line was generated by J.T. while he was a postdoctoral fellow in the laboratory of Lilianna Solnica-Krezel. Work in the laboratory of M.H. was supported by the Max-Planck Society, by the European Union (6^th framework integrated project `Zebrafish models for human development and disease') and by the National Institutes of Health (NIH grant 1R01-GM63904). Work in N.H.'s laboratory was supported by a grant from the National Center for Research Resources at the National Institutes of Health(2R01-RR012589-6).