Regulation of cell surface protease matriptase by HAI2 is essential for placental development, neural tube closure and embryonic survival in mice

The several hundred proteases that reside within the extracellular and pericellular space in vertebrates are essential modulators of cell behavior in the context of development, tissue homeostasis and tissue remodeling. Chief functions of this large class of enzymes include pruning of the extracellular matrix, activation of hormones, growth factors and cytokines, and ectodomain shedding of cell surface receptors and adhesion molecules. A common feature of proteases is their synthesis as inactive zymogens that are activated by endoproteolytic cleavage within a highly conserved activation site. This activation step is irreversible, and the active proteases are then regulated by cognate macromolecular protease inhibitors that usually bind directly to the active site, thereby blocking proteolytic activity and frequently causing internalization and degradation of the protease-inhibitor complex(Herz and Strickland, 2001; Lopez-Otin and Bond, 2008; Page-McCaw et al., 2007; Puente et al., 2003; Rau et al., 2007).

A potentially dramatic illustration of the requirement for postactivation regulation of protease activity during tissue morphogenesis was given by the recent discovery that humans homozygous or compound heterozygous for an assortment of SPINT2 mutations suffer from an exceptionally broad range of developmental abnormalities that include duplication of internal organs, duplication and abnormal location of digits, craniofacial dysmorphisms, anal and choanal atresia, fistulas and harmatoma. Defects in postnatal epithelial function, such as persistent sodium diarrhea, corneal erosions and abnormal hair, are also associated with SPINT2deficiency (Heinz-Erian et al.,2009). The SPINT2 gene encodes a serine protease inhibitor, hepatocyte growth factor activator inhibitor (HAI) 2 (also known as placental bikunin; hereafter HAI2)(Kawaguchi et al., 1997; Marlor et al., 1997), and reduced expression of HAI2 or synthesis of HAI2 variants with reduced in vitro inhibitory activity is common to all the mutated SPINT2 alleles(Heinz-Erian et al., 2009). This raises the possibility that key morphogenic processes, such as organ and digit specification and tubulogenesis, may require the fine-tuning of pericellular proteolytic activity through HAI2 silencing of unknown target proteases. Alternatively, HAI2, a transmembrane protein, could have signaling or cell adhesion functions that are independent of its protease inhibitory activity.

HAI2 is a Kunitz-type serine protease inhibitor named after its homology to hepatocyte growth factor activator inhibitor 1 and its first isolation from human placenta, respectively (Kawaguchi et al., 1997; Marlor et al.,1997). It is an integral type I transmembrane glycoprotein consisting of a short cytoplasmic tail, a single transmembrane domain and an extracellular tandem array of Kunitz inhibitory domains. The isolated Kunitz inhibitory domains of HAI2 display potent inhibitory activity towards more than a dozen trypsin-like serine proteases when tested in cell-free in vitro systems. These include both membrane serine proteases such as hepsin,matriptase and prostasin, as well as secreted proteases such as tissue kallikrein, plasma kallikrein, hepatocyte growth factor activator, and coagulation factors IXa, Xa, XIa and XIIa(Delaria et al., 1997; Herter et al., 2005; Kirchhofer et al., 2005; Qin et al., 1998; Szabo et al., 2008). This list of in vitro targets may be even larger, as HAI2 has been tested for inhibitory activity towards only a small fraction of the trypsin-like serine proteases encoded by the human genome. This suggests that inhibitory target selectivity by HAI2 is achieved primarily through temporal and spatial restrictions, such as the site of expression, membrane anchorage and subcellular localization,thus complicating the identification of key developmental and postnatal HAI2 targets using standard biochemical approaches. However, SPINT2orthologs are found in the genomes of all mammals analyzed, indicating a conserved function in development (Quesada et al., 2009), which makes target identification amenable to a combination of cell biological, genomic and genetic approaches, such as epistasis and complementation analysis in mice.

ST14 encodes the type II transmembrane serine protease,matriptase, that is expressed in many epithelial tissues and plays a crucial postnatal role in the terminal differentiation of the epidermis and other epithelial tissues in both humans and rodents(Alef et al., 2008; Basel-Vanagaite et al., 2007; List et al., 2006a; List et al., 2003). Previous studies by our group and by others have shown that dysregulation of matriptase in intact tissues by ectopic expression or loss of inhibition can perturb homeostasis by derailing proliferative and migratory cellular circuits(Carney et al., 2007; List et al., 2005; Mathias et al., 2007).

In this study, we show that HAI2 is a potent modulator of cell behavior that is governed by resident trypsin-like serine proteases. The use of transcriptomic profiling, combined with epistasis and complementation analysis, identified matriptase as a key target for HAI2 during tissue morphogenesis, and revealed a complex redundancy of HAI2 and the related HAI1 in matriptase regulation. This study establishes that tissue morphogenesis requires postactivation regulation of membrane-associated serine proteases by a network of partially redundant Kunitz-type transmembrane protease inhibitors.

Type I and type II Madin-Darby canine kidney (MDCK) cells were kindly provided by Dr Jay Chiorini (NIDCR, Bethesda, MD, USA). The cells were maintained in Dulbecco's Modified Eagle Medium (DMEM) containing high glucose and supplemented with 10% fetal bovine serum, penicillin/streptomycin and 2 mM L-glutamine (all Gibco, Invitrogen, Carlsbad, CA, USA), and were cultured in a 5% CO₂-buffered tissue culture incubator. To generate cells with cytomegalovirus promoter-driven expression of full-length human HAI2 protein,the 1.3 kb EcoRI fragment of the pOTB7 vector containing the entire coding region of the human SPINT2 cDNA (IMAGE clone 3952243,nucleotide -293 to +1035 of the SPINT2 cDNA) was cloned into the mammalian expression plasmid pLenti-EF-1-SfiI-shuttle-IRES-PuroR (a kind gift from Drs Daniel Martin and Silvio Gutkind, NIDCR). MDCK type I cells were transfected with pLenti-EF-1-SfiI-shuttle-IRES-PuroR/hSpint2 or control pLenti-EF-1-SfiI-shuttle-IRES-PuroR plasmid DNA using PolyFect Reagent(Qiagen, Valencia, CA, USA) following the manufacturer's instructions. Transfected cells were selected by adding 5 μg/ml of puromycin (Sigma, St Louis, MO, USA) into the culturing medium, using untransfected cells as a negative control. Stably transfected HAI2-expressing and mock-transfected clones (several hundred) were each pooled and the expression of HAI2 was determined by western blot as described below.

MDCK I cells stably transfected with HAI2-expressing plasmid or control plasmid were grown on 100 mm cell culture dishes to ∼90% confluency,washed with PBS and lysed in buffer containing 62.5 mM Tris-HCl pH 6.8, 2% SDS and 10% glycerol, and subjected to a western blot analysis as described previously (Szabo et al.,2005). Anti-human HAI2 primary antibody (R&D Systems,Minneapolis, MN, USA) and anti-goat secondary antibodies (Sigma) conjugated with alkaline phosphatase were used.

MDCK type I cells (25,000 per well) expressing full-length human HAI2 or mock-transfected control cells (see above) were seeded in triplicate into 6.5 mm Transwell chambers (Corning, Lowell, MA, USA). The transepithelial ion permeability of monolayers was determined every 24 hours by measuring electric resistance across the epithelial monolayer using the EVOM Epithelial Voltohmmeter (World Precision Instruments, Saratosa, FL, USA). The medium in both internal and external compartments of each Transwell was changed every 2 days.

MDCK type II cells (5000 per well) were seeded into 12-well plates and allowed to form isolated colonies in serum-containing DMEM medium. Cells were then washed once with PBS and incubated in serum-free DMEM at 37°C for 1 hour. The medium was then replaced with serum-free DMEM containing 100 ng/ml of pro-HGF/SF or 100 ng/ml active HGF/SF (R&D Systems) and the soluble extracellular domain of human recombinant HAI2 protein (R&D Systems) was added to a final concentration of 0, 5, 20 or 60 nM. Aprotinin (1 μM) was added to cells as a positive control for the inhibition of pro-HGF/SF activation. The experiments were performed at least twice and in duplicates.

MDCK I cells stably transfected with a human HAI2 expression vector or an empty expression vector were seeded onto 60 mm plates (10,000 cells) and grown in serum-containing DMEM medium with high glucose. The number of cells in three separate areas of approximately 1 cm² each was counted every 24 hours for period of 7 days.

To determine the effect of HAI2 expression on the overall cell size and apoptotic rate of MDCK I cells during the acquisition of transepithelial electrical resistance (TEER), 4×10⁵ MDCK I cells expressing full-length human recombinant HAI2 protein or MCDK I control cells were seeded in triplicate into 24 mm Transwell chambers and incubated in serum-containing DMEM medium. The transepithelial ion permeability of monolayers was measured every 24 hours as described above. At the onset of TEER acquisition (96 hours after seeding) and once the cells reached maximum TEER values (168 hours after seeding), the relative rate of apoptosis was determined by measuring caspase 3 activity using a Caspase 3 Fluorometric Assay (R&D Systems) following the manufacturer's instructions. To measure the relative size of the cells, cells grown in Transwell chambers for 96 and 168 hours were detached from the bottom of the chamber by the addition of 0.05% Trypsin-EDTA (Gibco), centrifuged for 4 minutes at 500 g, and resuspended in PBS (Gibco) with 0.5%bovine serum albumin (Sigma). Cells of each sample(5×10⁴-10×10⁴) were then subjected to flow cytometry using forward light scatter on a BD FACS Calibur System (BD Biosciences, San Jose, CA, USA), and the relative size of cells in each sample was analyzed using CellQuest Pro 5.2 software (BD Biosciences).

All procedures involving live animals were performed in an Association for Assessment and Accreditation of Laboratory Animal Care International-accredited vivarium, following Institutional Guidelines and standard operating procedures. The generation of mice carryingβ-galactosidase-tagged Spint2, Spint1 null, St14 null and β-galactosidase-tagged St14 alleles has been described previously (List et al., 2002; List et al., 2006b; Mitchell et al., 2001; Szabo et al., 2008; Szabo et al., 2007). Mice were genotyped for the presence of the Spint2 null allele by PCR using HAI2-geo51 (5′-ATCTGCAACCTCAAGCTAGC-3′) and HAI2-geo31(5′-CAGAACCAGCAAACTGAAGG-3′) primers, and for the presence of the Spint2 wild-type allele using HAI2-wt51(5′-AACACATTTCACCACCATGC-3′) and HAI2-wt31(5′-CCAGACTTTCCTAAGTGGG-3′) primers. PCR screening for Spint1 null, β-galactosidase-tagged St14 and St14 null alleles has been described previously(List et al., 2002; List et al., 2006b; Szabo et al., 2008). All mice used in this study were in a mixed 129/C57BL/6J/BlackSwiss genetic background. Inherent to the lethality of mice homozygous for Spint2, Spint1 and St14 null alleles, all studies were littermate-controlled.

Breeding females were checked for vaginal plugs in the morning and the day on which the plug was found was defined as the first day of pregnancy (E0.5). Pregnant females were euthanized in the mid-day at designated time points by CO₂ asphyxiation. Embryos were extracted by Cesarian section and the individual embryos and placentae were dissected and processed as described below. Visceral yolk sacs of individual embryos were washed twice in PBS,subjected to genomic DNA extraction and genotyped by PCR. Newborn pups were euthanized by CO₂ inhalation at 0°C. For histological analysis,the embryos and newborn pups were fixed for 24 hours in 4% paraformaldehyde(PFA) in PBS, processed into paraffin, sectioned into parallel sagittal- or cross-sections and stained with Hematoxylin and Eosin (H&E), or used for immunohistochemistry as described below.

The Spint2-deficient and Spint2-expressing littermate control embryos were isolated as described above. Total RNA was prepared by extraction in TRIzol reagent (Gibco, Carlsbad, CA, USA), as recommended by the manufacturer. First-strand cDNA synthesis was performed from 1 μg of total RNA using a RETROscript Kit (Ambion) and an oligo-dT primer according to the manufacturer's instructions. The subsequent PCR (denaturation temperature 92°C, annealing temperature 55°C, extension temperature 72°C, 35 cycles) was carried out using HAI2-RT51 (5′-CTGGCAGCAGGTCATGG-3′)and HAI2-RT31 (5′-TTGCCATTGCCTTCACAGCC-3′) primers. To verify the identity of the RT-PCR products, the amplified DNA was excised from the agarose gel following electrophoresis, purified and analyzed by DNA sequencing.

Tissues were fixed overnight in 4% PFA, embedded in paraffin and cut into 5μm-thick sections. Antigens were retrieved by incubation for 10 minutes at 37°C with 10 μg/ml proteinase K (Fermentas, Hanover, MD, USA) for HAI1 staining, or by incubation for 20 minutes at 100°C in 0.01 M sodium citrate buffer pH 6.0, for all other antigens. The sections were blocked with 2% bovine serum albumin in PBS and incubated overnight at 4°C with anti-laminin (Sigma), anti-CD31 (Santa Cruz Biotechnology, Santa Cruz, CA,USA), anti-HAI1 (R&D Systems), or anti-matriptase (R&D Systems)primary antibodies. Bound antibodies were visualized using biotin-conjugated anti-mouse, -rabbit, -sheep or -goat secondary antibodies (Vector Laboratories, Burlingame, CA, USA) and a Vectastain ABC Kit (Vector Laboratories), using 3,3′-diaminobenzidine as the substrate (Sigma). All microscopic images were acquired on a Olympus BX40 microscope using an Olympus DP70 digital camera system (Olympus, Melville, NY, USA).

Tissue samples were fixed for 30 minutes in 4% PFA in PBS, rinsed in PBS and stained with a β-galactosidase staining kit (Roche, Indianapolis, IN,USA) for 6 hours at 37°C using X-Gal as the substrate. The tissues were postfixed overnight in 4% PFA, embedded in paraffin, sectioned and stained with nuclear Fast Red. All analysis included parallel staining of wild-type littermate control mice to ensure the specificity of the staining.

Spint2 heterozygous females bred to Spint2 heterozygous males were checked for vaginal plugs. Folic acid (10 mg/g body weight, Sigma)in PBS was administered to pregnant females by intraperitoneal injection every 24 hours starting on E0.5. Embryos were extracted on E10.5 by Cesarian section, genotyped as described above and scored for neural tube closure defects.

The cranial region of the neural tube with adjacent surface epithelium was microdissected from wild-type E8.5 embryos. Total RNA was extracted from the dissected tissue and reverse-transcribed into cDNA as described above. PCR amplification was performed using the degenerate oligonucleotide primers mSPD52 [5′-GT(G/T)(C/G)T(G/T)-(A/T)C(A/T)GCTGC(A/T/C)CACTG-3′] and mSPD32 [5′-(A/T)GGGCC(A/T)CC(A/T/G)GAGTC(A/T)CC-3′], corresponding to the conserved amino acid sequences Val-Leu-Thr-Ala-Ala-His-Cys and Gly-Asp-Ser-Gly-Gly-Pro-Leu, respectively, as described previously(Takano et al., 2004). Fragments of 400-500 base pairs (bp) in size were recovered from the PCR products by agarose gel electrophoresis. The PCR products of the expected size of 400-500 bp were extracted from agarose gels following electrophoresis,subcloned into the TOPO TA vector (Invitrogen) and the identity of the inserted DNA was then determined by sequencing. A total of 600 plasmid-transformed bacterial clones with an identified PCR insert were screened.

The potent and promiscuous activity of the Kunitz-type domains of HAI2 towards trypsin-like serine proteases suggests that Spint2 supports development and maintains epithelial homeostasis by regulating the pericellular activity of one or more members of this large class of enzymes that regulates multiple cellular processes, including the bioavailability of growth factors, ion fluxes and paracellular permeability(Bhatt et al., 2007; Carattino et al., 2008; Chin et al., 2008; Cohen et al., 1985; Lee et al., 2000; Leyvraz et al., 2005; Mizuno et al., 1994; Myerburg et al., 2006; Naldini et al., 1992; Picard et al., 2008; Scudamore et al., 1998; Shimomura et al., 1993; Vallet et al., 1997; Wang et al., 1994a; Wang et al., 1996; Wang et al., 1994b). To survey the portfolio of biological activities that can be modulated by HAI2, we first manipulated the levels of the protease inhibitor either through overexpression or exogenous addition of a soluble bioactive version. For this purpose, we chose Madine-Darby canine kidney (MDCK) epithelial cells, as the aforementioned processes are regulated in a trypsin-like serine protease-dependent manner in this cell type. Exposure of MDCK type II cells to pro-hepatocyte growth factor/scatter factor (pro-HGF/SF) caused a well-described and conspicuous dissolution of cell-cell junctions of monolayers of cells and the subsequent migration of individual cells out of the cluster (Fig. 1A,B). Interestingly, this classic response was inhibited in a dose-dependent manner by the presence of nanomolar concentrations of the soluble extracellular domain of HAI2 (Fig. 1C,D; see Fig. S1A in the supplementary material). However, the scattering process proceeded normally in the presence of soluble HAI2 when cells were exposed to equimolar concentrations of proteolytically activated HGF/SF(Fig. 1E,F), showing that HAI2 might act on cell motility by regulating the bioavailability of HGF/SF at the level of its proteolytic conversion by endogenous MDCK-expressed trypsin-like serine proteases. MDCK type I cells, when unexposed to motility-inducing growth factors, form functional tight junctions whose ion permeability is modulated by proteolytic activity generated by trypsin-like serine proteases(Scudamore et al., 1998). To determine the potential effects of HAI2 levels on protease-modulated tight junction function, we measured the acquisition of transepithelial electrical resistance (TEER) in MDCK type I cells with increased expression of HAI2(Fig. 1G). The proliferative capacity, outwards morphology, size and rate of apoptosis of the cells were unaffected by the level of HAI2 (see Fig. S1B-F in the supplementary material;data not shown). However, MDCK type I cells with increased levels of HAI2 displayed a pronounced acceleration of the rate of acquisition of functional tight junctions, reaching a maximum TEER 72 hours earlier than control MDCK type I cells (Fig. 1H). Taken together, these studies reveal HAI2 as a modulator of the capacity of trypsin-like serine proteases to regulate basic cellular responses, such as cell motility and transepithelial barrier formation.

We next used mouse reverse genetics to understand HAI2 function in the context of the developing embryo. In a preliminary phenotypic screen of a large array of randomly targeted mouse genes, homozygosity for a lacZ-tagged Spint2 insertion was reported to result in embryonic lethality at gastrulation before embryonic day (E) 8, due to severe clefting of the embryonic ectoderm(Mitchell et al., 2001). Genomic analysis showed that the lacZ insertion was associated with the deletion of coding exons 2 through to 7, thus giving rise to a Spint2 null allele (Fig. 2A). In agreement with the original analysis, our interbreeding of mice that were heterozygous for this Spint2 allele (hereafter called Spint2^+/-) failed to produce any Spint2^-/- pups among term offspring(Fig. 2B). However, mendelian analysis of multiple staged embryos showed that about half of the Spint2^-/- embryos developed past E8 and proceeded to the headfold stage, although their survival decreased rapidly after E9.5, and none of the Spint2^-/- embryos was found to live past E14.5(Fig. 2C). Macroscopic inspection of Spint2^-/- embryos that survived past E9.5 revealed severe defects in the development of the neural tube. Thus, 96% of Spint2^-/- embryos extracted at E10.5 or later suffered from incomplete closure of the neural tube in the cranial region (exencephaly; Fig. 2D,D′), 89% of embryos extracted at E11.5 or later presented with a curly tail phenotype(Fig. 2F), indicative of a defect in secondary neurulation in the rostral region of the neural tube(Embury et al., 1979), and one Spint2^-/- embryo suffered from an incomplete closure of the neural tube in the back region (spina bifida; Fig. 2F). None of the Spint2^+/+ or Spint2^+/- mice was affected (Fig. 2E,E′,F′). These defects in neurulation were not rescued by the supplementation of dietary folate (data not shown). Whereas the genesis of the medial hinge of the developing neural tube appeared to proceed normally, the morphogenesis of the dorsolateral hinges of the neural epithelium was revealed to be Spint2-dependent(Fig. 2G,G′; data not shown).

Despite their severity, defects in neural tube closure are not known to abolish term development in either mice or humans(Kibar et al., 2007). However,careful macroscopic and histological inspection of Spint2^-/- embryos did not reveal any additional morphological abnormalities in the embryo proper (data not shown), suggesting a possible additional requirement of Spint2 for the development of extraembryonic tissues (yolk sac or placenta). Although no obvious effect of Spint2 deficiency was observed on the outwards or histological appearance of the visceral yolk sac at E9.5 or E10.5 (data not shown), the development of the placental labyrinth was strongly impaired by homozygous Spint2 deficiency (Fig. 3A,A′). Immunohistological staining for HAI1, a specific marker of chorionic and labyrinthine trophoblasts(Hallikas et al., 2006; Szabo et al., 2007)(Fig. 3B,B′), and CD31(also known as PECAM), a marker for endothelial cells(Albelda et al., 1990) invading the labyrinth (Fig. 3C,C′), revealed a pronounced reduction in the thickness of the labyrinth (Fig. 3D) and a corresponding diminution of fetal vessel permeation of the labyrinthine space(Fig. 3E). Further analysis documented that most Spint2^-/- placentae suffered from a pronounced defect in the polarization of the chorionic epithelium, as shown by delocalized deposition of the basement membrane protein laminin(Fig. 3F-H′). Placental abnormalities were not detected in Spint2^+/- embryos at any stage, indicating that Spint2 heterozygosity is permissive for normal placental development. These findings demonstrate that Spint2is essential to establish the functional maternal-fetal interface required for term development of the embryo, and that Spint2^-/- embryos that succeed in clearing the first developmental hurdle prior to E8.0 succumb to oxygen and nutrient starvation at mid-gestation associated with placental failure.

To identify potential HAI2 targets during morphogenesis, tissues from the cranial region of the developing neural tube of wild-type mice were isolated by microdissection and used for the generation of a cDNA library. Next, we used a previously devised strategy (Takano et al., 2004), which uses degenerate primers designed to be capable of amplifying trypsin-like serine proteases, to perform PCR amplification of the cDNA library and DNA sequencing of the ensuing amplicons. This analysis revealed the expression of genes encoding 17-19 different secreted or membrane-bound trypsin-like serine proteases during neurogenesis(Table 1). In addition to a number of genes encoding granzymes, these included genes encoding proteases known to be expressed in the developing central nervous system, such as tissue-type plasminogen activator, kallikrein-related protease 8 (neuropsin)and PRSS12 (neurotrypsin), and also the transmembrane serine proteases TMPRSS2, TMPRSS4, matriptase and matriptase 3, as well as the GPI-anchored protease prostasin. Notably, two of these transmembrane serine proteases,matriptase and prostasin, were recently reported to be efficiently inhibited by HAI2 in vitro and, thus, represented potential HAI2 targets during neural tube closure (Coote et al.,2009; Szabo et al.,2008) (see below).

To gain additional mechanistic insights into Spint2-dependent neural tube closure and placental branching morphogenesis, we took advantage of the insertion of the promoterless lacZ marker gene, encodingβ-galactosidase, in the Spint2 null allele(Fig. 2A) to determine the localization of Spint2 expression in the two developmental processes by X-Gal staining of embryos carrying one wild-type and one lacZ-targeted allele. This analysis revealed a spatial and temporal correlation between neural and placental morphogenesis and the expression of Spint2, consistent with the notion that HAI2 is acting in the local pericellular microenvironment to facilitate development. Thus, Spint2expression was abundant in the developing neuroepithelium of the cranial region of the embryo, as well as at the caudal extremity of the embryo that develops in the tail region, at least as early as E7.5 and persisting until E9.5 (Fig. 4A-D; data not shown), thereby spanning the critical developmental window in which neural tube closure occurs (Kaufmann and Bart,1999). Furthermore, histological analysis revealed that Spint2 is expressed in cells at the point of contact between the two sides of the neural epithelium before neural tube closure occurs(Fig. 4E). Likewise, suggesting a localized mode of action in placental formation, Spint2 was expressed specifically in chorionic trophoblasts and in the underlying trophoblast cells within the labyrinth, commencing as early as E8.0 and persisting for the duration of placental labyrinth development(Fig. 4F-H; data not shown).

One protease of particular interest that was identified in the transcriptomic analysis was matriptase, encoded by the St14 gene. First, microarray analysis has shown that matriptase is expressed in MDCK cells (Hellman et al., 2008). Second, matriptase can proteolytically process and activate pro-HGF and the epithelial permeability-inducing PAR2 (F2RL1)(Lee et al., 2000; Takeuchi et al., 2000). Third,in addition to its expression in the developing neural tube, we have previously shown that matriptase is also expressed in chorionic trophoblasts during branching morphogenesis (Szabo et al., 2007). Fourth, soluble HAI2 displays potent inhibitory activity towards endogenous matriptase located on the surface of the monocytoid cell line THP-1 (Kilpatrick et al., 2006; Szabo et al.,2008). To directly query the specific contribution of matriptase suppression by HAI2 to embryonic development, we took advantage of the availability of mice carrying an St14 null allele, and the fact that St14 by itself is dispensable for mouse term development(List et al., 2002). Our rationale was that if HAI2 suppression of matriptase was crucial for embryonic development, then the genetic ablation of St14 should eliminate one or more of the developmental defects associated with Spint2deficiency. To test this hypothesis, we first bred Spint2^+/- mice to St14^+/- mice, and next interbred the ensuing Spint2^+/-;St14^+/-double-heterozygous mice to generate offspring carrying one or two wild-type Spint2 and one or two wild-type St14 alleles (hereafter termed Spint2⁺;St14⁺), two Spint2 null alleles and one or two St14 wild-type alleles (hereafter termed Spint2^-/-;St14⁺), one or two Spint2wild-type alleles and two St14 null alleles (hereafter termed Spint2⁺;St14^-/-), and two Spint2 null alleles and two St14 null alleles (hereafter termed Spint2^-/-;St14^-/-). In agreement with data presented above, no Spint2^-/-;St14⁺ mice were identified among 226 term offspring from these crosses(Fig. 5A). Surprisingly,however, Spint2^-/-;St14^-/- pups were found in the expected mendelian frequency in the term offspring from the Spint2^+/-;St14^+/- double-heterozygous crosses(Fig. 5A). This demonstrates that both the early (pre-E8.0) demise that befalls about half of the Spint2 null embryos and the mid-gestation lethality of the remaining Spint2 null embryos were dependent on St14 activity.

We next extracted embryos with single and combined loss of Spint2and St14 at different embryonic stages to directly determine the effect of St14 ablation on the morphogenesis of Spint2-dependent tissues. Consistent with the re-establishment of term development, histological analysis of placental tissues from Spint2^-/-;St14^-/- embryos showed normal development of the placental labyrinth, as evidenced by both the restored thickness of the labyrinth layer (Fig. 5B,C; compare with Fig. 3B,B′), normal branching of fetal vessels into the labyrinthine layer (Fig. 5D,E;compare with Fig. 3C,C′)and the normal deposition of laminin by chorionic epithelial cells(Fig. 5F; compare with Fig. 3F,G). In agreement with our previous study (Szabo et al.,2007), St14 expression could be detected specifically in chorionic trophoblasts as early as E8.0 and colocalized with Spint2for the duration of labyrinth development, from E8.5 through to E11.5(Fig. 6A,B; data not shown;compare with Fig. 4F,G). These data show that a local matriptase-dependent proteolytic pathway is the only essential target for inhibition by HAI2 during placental branching morphogenesis, and identifies an excess of matriptase activity by chorionic trophoblasts as the underlying cause of placental defects in Spint2-deficient embryos.

Analysis of the macroscopic appearance of pre-term embryos and newborn Spint2^-/-;St14^-/- mice revealed that the overall frequency of neural tube closure-related defects presenting as exencephaly (Fig. 6C,C′;data not shown) was reduced more than fivefold in Spint2^-/-;St14^-/- offspring, compared with their Spint2^-/-;St14⁺ siblings(Fig. 6F). Caudal neural tube development in Spint2-deficient embryos was less affected by the ablation of St14, with a 30% reduction in neural tube defects presenting as curly tail (Fig. 6D,F; data not shown) and no appreciable effect on the frequency of spina bifida (Fig. 6E,F). This indicates that the inhibition of a matriptase-dependent proteolytic pathway is one crucial molecular function of HAI2 in neural tube development. In contrast to placental development, however, the elimination of St14 does not restore normal neural tube formation of all Spint2 null embryos, indicating that HAI2 suppression of additional protease targets, besides matriptase, is functionally relevant to the processes of neurulation.

The requirement of functional alleles of both Spint2 (this study)and Spint1 (Szabo et al.,2007) for the developmental suppression of St14 during placental morphogenesis is counterintuitive in light of the high structural homology of the protease inhibitors encoded by the two genes, and their conspicuous colocalization in developing placental tissues. To investigate this apparent paradox, we took advantage of the localization of Spint2,Spint1 and St14 on separate chromosomes (7, 2 and 11,respectively) and, thus, their independent segregation, to perform a combined complementation and epistasis analysis to determine their interrelationship. Unexpectedly, this study demonstrated that the combined heterozygosity of Spint2 and Spint1 was incompatible with term development. Thus, no Spint2^+/-;Spint1^+/-;St14^+/+pups were identified in 76 offspring from crosses of Spint2^+/-;Spint1^+/+;St14^+/-mice with Spint2^+/+;Spint1^+/-;St14^+/+mice (Fig. 6G). This unusual outcome, which is termed nonallelic noncomplementation, has been described previously in mice, Drosophila, C. elegans and yeast (e.g. Hays et al., 1989; Rancourt et al., 1995; Rine and Herskowitz, 1987; Varkey et al., 1993). It shows that the expression of any three of the combined four Spint1 and 2 alleles provides sufficient proteolytic inhibitory activity for term development, whereas the expression of any two of the four is insufficient. Even more informatively, this combined heterozygosity of Spint2 and Spint1 became permissive for term development once St14 expression was reduced by the elimination of one functional St14 allele. Indeed, Spint1^+/-;Spint2^+/-;St14^+/- siblings were found at the expected frequency in term offspring from these crosses(Fig. 6G) and were outwardly normal (data not shown). Taken together(Fig. 6H), this analysis demonstrates that a delicate balance of the activity of the transmembrane serine protease matriptase and the inhibitory activities of HAI1 and HAI2 establishes a local pericellular proteolytic microenvironment that is conducive to successful tissue morphogenesis.

The current study shows that the pericellular availability of HAI2 is a potent determinant of protease-guided activities of cultured epithelial cells,including migratory responses and the acquisition of a diffusion barrier of confluent epithelial cell monolayers. The study also shows that focal inhibition of proteolytic activity by HAI2 is essential for fundamental developmental processes, such as branching morphogenesis and neural tube closure, and specifically identifies matriptase-dependent proteolysis as one crucial HAI2 target. Furthermore, our epistasis and complementation analysis revealed that a delicate balance of proteolytic activity, provided by matriptase and other serine protease(s), and serine protease inhibitory activity, provided by HAI2 and HAI1, creates a pericellular microenvironment that is permissive for the completion of these morphogenic processes. Thus,whereas Spint2 null embryos failed to develop a functional labyrinth,form a neural tube and develop to term, the imposition of a simultaneous St14 deficiency enabled placental development, greatly reduced the incidence of neural tube closure defects and enabled development to term. Additionally, combined heterozygosity for Spint2 and Spint1was incompatible with development to term, whereas triple heterozygosity for Spint2, Spint1 and St14 enabled term development and postnatal survival. We propose that HAI2 acts by modulating protease-dependent migratory and homeostatic circuits in developing tissues by regulating the focal activity of matriptase and other trypsin-like serine proteases. In this regard, it is somewhat surprising that matriptase itself is dispensable for term development in humans, mice and zebrafish(Basel-Vanagaite et al., 2007; Carney et al., 2007; List et al., 2002). However,matriptase is one of three closely related proteases(Hooper et al., 2003; Szabo et al., 2005; Velasco et al., 2002). Moreover, matriptase 2 (TMPRSS6) and matriptase 3 display some overlap with matriptase in the pattern of expression during embryonic development(Hooper et al., 2003) (this study, and R.S. and T.H.B., unpublished data). It is therefore plausible that one or more of the three matriptases have redundant functions during development and that single deficiency in each is compatible with term development due to the compensatory activity of the others. Studies to experimentally test this hypothesis have been initiated.

The residual neural tube defects in combined Spint2;St14 null mice reveal that neural tube closure is facilitated by HAI2 suppression of at least one additional protease besides matriptase. Our targeted transcriptomic analysis of the trypsin-like serine protease complement expressed during this morphogenic process revealed several potential candidates. Most notable among the additional 16-18 trypsin-like proteases expressed in the developing neural tube were the close matriptase relative, matriptase 3(Szabo et al., 2005) and the type II transmembrane serine proteases TMPRSS2(Paoloni-Giacobino et al.,1997) and TMPRSS4 (Wallrapp et al., 2000). Because the HAI2 Kunitz-type inhibitor domains are highly promiscuous in purified systems (see Introduction), the definitive identification of the residual morphogenic HAI2 targets will probably require exhaustive genetic complementation analysis.

Matriptase is an autoactivating protease that has been proposed to serve as an initiator of proteolytic cascades in adult tissues(Kilpatrick et al., 2006; Lee et al., 2000; Netzel-Arnett et al., 2006; Takeuchi et al., 2000). It is formally possible that HAI2 and HAI1 therefore do not target matriptase directly during development, but rather target proteases that are located downstream of matriptase. However, the potent inhibitory activity of HAI1 and HAI2 Kunitz-type inhibitory domains towards matriptase in vitro(Lin et al., 1999; Szabo et al., 2008) and the developmental colocalization of the two protease inhibitors with matriptase make this scenario unlikely.

The developmental consequences of Spint2 deficiency at a superficial glance appear to manifest at an earlier stage in mice than humans and seem to be more severe. Importantly, however, to date, the known complement of mutated SPINT2 alleles all either directed the production of HAI2 proteins at lower levels or with lower inhibitory activity(Heinz-Erian et al., 2009),whereas the phenotypic consequence of complete Spint2 deficiency has been analyzed only in mice. Also, as only SPINT2-deficient individuals that completed term development have been available for clinical evaluation, it is premature to discuss potential species-specific variations in the developmental manifestations of the loss of HAI2. In addition to abnormal morphogenesis, HAI2 deficiency in humans is associated with distinct defects in postnatal epithelial homeostasis, as shown by severe corneal erosions, abnormal hair, persistent sodium diarrhea and poor postnatal survival. The abundant expression of matriptase in the epithelial compartment of all of these tissues (List et al.,2006b; Oberst et al.,2003) is a conspicuous indication that these postnatal manifestations might also be causally related to the loss of matriptase inhibition in these tissues.

In summary, this study establishes that a network of partially redundant Kunitz-type transmembrane protease inhibitors enables tissue morphogenesis through the local postactivation regulation of membrane-associated serine proteases.