Sizzled Is Unique among Secreted Frizzled-related Proteins for Its Ability to Specifically Inhibit Bone Morphogenetic Protein-1 (BMP-1)/Tolloid-like Proteinases*

Background: Xenopus and zebrafish BMP-1/tolloid-like proteinases (BTPs) are inhibited by sizzled, a secreted frizzled-related protein (sFRP) not present in mammals. Results: Xenopus sizzled is a very potent inhibitor of human BTPs, whereas mammalian sFRPs have no effect. Conclusion: Regulation of BTP activity by sFRPs is not conserved in mammals. Significance: Sizzled is the most potent exogenous inhibitor of human BTPs. BMP-1/tolloid-like proteinases (BTPs) are major enzymes involved in extracellular matrix assembly and activation of bioactive molecules, both growth factors and anti-angiogenic molecules. Although the control of BTP activity by several enhancing molecules is well established, the possibility that regulation also occurs through endogenous inhibitors is still debated. Secreted frizzled-related proteins (sFRPs) have been studied as possible candidates, with highly contradictory results, after the demonstration that sizzled, a sFRP found in Xenopus and zebrafish, was a potent inhibitor of Xenopus and zebrafish tolloid-like proteases. In this study, we demonstrate that mammalian sFRP-1, -2, and -4 do not modify human BMP-1 activity on several of its known substrates including procollagen I, procollagen III, pN-collagen V, and prolysyl oxidase. In contrast, Xenopus sizzled appears as a tight binding inhibitor of human BMP-1, with a Ki of 1.5 ± 0.5 nm, and is shown to strongly inhibit other human tolloid isoforms mTLD and mTLL-1. Because sizzled is the most potent inhibitor of human tolloid-like proteinases known to date, we have studied its mechanism of action in detail and shown that the frizzled domain of sizzled is both necessary and sufficient for inhibitory activity and that it acts directly on the catalytic domain of BMP-1. Residues in sizzled required for inhibition include Asp-92, which is shared by sFRP-1 and -2, and also Phe-94, Ser-43, and Glu-44, which are specific to sizzled, thereby providing a rational basis for the absence of inhibitory activity of human sFRPs.

Bone morphogenetic protein-1 (BMP-1) 3 is the prototype of a small group of zinc metalloproteinases, the BMP-1/tolloidlike proteinases (BTPs), which include members in species ranging from Drosophila to human (1). In mammals, there are four isoforms of BTPs (supplemental Fig. S1): BMP-1 and mTLD (mammalian tolloid), which are alternative spliced products of the same gene (2), and also mTLL-1 and mTLL-2 (mammalian tolloid-like 1 and 2) (3,4). These proteinases are characterized by a highly similar domain structure encompassing a signal peptide, a propeptide that maintains latency of the enzyme (5), a catalytic domain belonging to the astacin subfamily (M12A according to MEROPS database) (6), and a variable number of CUB (Complement, Uegf, BMP-1) and epidermal growth factor (EGF) auxiliary domains. Other members of the astacin subfamily in humans include meprins ␣ and ␤ and ovastacin (supplemental Fig. S1). BTPs have been found to be active on more than 20 substrates, including components of the extracellular matrix, growth factors, and pro/anti-angiogenic factors and are thought to play major roles in development (7), tissue repair, and related pathological conditions such as fibrosis (8,9).
An interesting feature of the BTPs is the fact that their regulation seems to rely mainly on specific enhancers which can direct activity toward distinct subsets of substrates (10). For example, procollagen C-proteinase enhancers (PCPEs) enhance the cleavage of fibrillar procollagens (11)(12)(13), whereas twisted gastrulation and ONT-1 (Olfactomedin Noelin Tiarin factor 1) are both activators of chordin cleavage (14,15), and periostin enhances the maturation of prolysyl oxidase (16). However, it is not always known if these effects are achieved through direct or indirect mechanisms. In contrast to other extracellular metalloproteinases of the metzincin family like matrix metalloproteinases (MMPs; supplemental Fig. S1) for which well established specific endogenous inhibitors have been described (e.g. TIMPs, tissue inhibitors of matrix metalloproteinases), the data concerning specific endogenous inhibitors of human BTPs are presently limited and controversial. In 2009, Lee et al. (17) showed that BMP-4, a member of the transforming growth factor-␤ superfamily, inhibits BTPs in a noncompetitive manner by binding to their CUB domains. However, this result is questioned by a previous study showing that mature BMP-1 is unable to interact with BMP-4 (18). Equally controversial is the potential role of secreted frizzled-related proteins (sFRPs) in the regulation of BTP activity.
The family of sFRPs includes five members in mammals (sFRP-1 to sFRP-5) and additional members in some species of vertebrates (sizzled, crescent, and Tlc). They are characterized by the presence of two domains, a netrin-like (NTR) domain, homologous to the N-terminal domain of TIMPs and to the C-terminal domain of PCPEs, and a frizzled (Fz) domain structurally related to the extracellular Wnt binding domain of frizzled receptors. Due to this homology, sFRPs have been mainly studied for their role as antagonists of Wnt signaling. However, recent studies have shown that sFRPs are not only Wnt-binding proteins but have a wide range of biological activities (19). In support of this idea was the finding that sizzled is an inhibitor of Xenopus and zebrafish BTPs (20,21).
Sizzled was initially identified in Xenopus embryos as a putative Wnt8 antagonist (22), but subsequent studies showed that sizzled was unable to block Wnt signaling in vivo (23)(24)(25). Rather, sizzled was shown to inhibit BMP signaling, thereby playing an important role in dorso-ventral patterning of Xenopus and zebrafish embryos (24 -26). It does this by inhibiting BTPs and thus stabilizing chordin, which is both a BMP antagonist and a BTP substrate. Further biochemical studies suggested that the Fz domain of sizzled binds and competitively inhibits non-mammalian BTPs (20,21). More recently, Xenopus crescent, which is the closest relative of sizzled, was also shown to inhibit BTPs (27) but remains capable of interfering with Wnt signaling. These findings reveal a new role for sFRPs as inhibitors of BTPs, but because sizzled and crescent genes are not present in mammals, this raises the question of whether or not this mechanism of inhibition is conserved in this class of vertebrates.
In recent years much attention has been paid to mouse sFRP-2 and its activity on Xenopus and human BTPs. Although it was shown to potently inhibit Xenopus Xlr in the initial study by Lee et al. (20), Kobayashi et al. (9) reported a few years later that, far from inhibiting human BMP-1, sFRP-2 was a specific enhancer of procollagen I cleavage and that this enhancement could promote fibrosis in a rat model of infarcted heart. In contrast, He et al. (28) showed that high concentrations of sFRP-2 inhibited the cleavage of a fluorogenic peptide and of procollagen I by human BMP-1, and using a similar model, they concluded that direct injection of sFRP-2 in the infracted heart could inhibit fibrosis and improve cardiac function. Finally, a third study from von Marschall et al. (29) found that sFRP-2 had no effect on the procollagen-C-proteinase activity of BMP-1.
Given the apparent conflict between the data, it appeared necessary to carefully re-evaluate whether the inhibition of BTPs is a general property of all sFRPs. In this study, we used five different substrates of human BMP-1 to check if its activity was inhibited or activated by three mammalian sFRPs (sFRP-1, sFRP-2, and sFRP-4). All tests led to the same conclusion that none of the tested sFRPs is able to enhance or inhibit BMP-1 activity in the conditions used for the activity assays. In contrast, Xenopus sizzled proved to be a selective and potent inhibitor of human BMP-1, with an inhibition constant in the low nanomolar range. Based on the sequence differences between sizzled and sFRP-1 and -2, its closest human relatives, we used site-directed mutagenesis to map the first interaction surface for sizzled and thus propose a model for the mechanism of inhibition of BMP-1.

EXPERIMENTAL PROCEDURES
Materials-Recombinant human Wnt3a, mouse sFRP-2, human sFRP-4, and full-length human MMP-1, -8, and -13 were obtained from R&D Systems. Human sFRP-1 was either from R&D Systems or produced as described below. All other proteins were produced and expressed as described below.
Molecular Biology-The cDNA for Xenopus laevis sizzled in the pCS2ϩ plasmid (gift from E.M. De Robertis, Los Angeles, CA) was amplified by PCR and inserted in-frame with a C-terminal V5-His 6 tag in the pcDNA3.1/V5-His-TOPO vector (Invitrogen). Pfx platinum polymerase (Invitrogen) was used for the PCR reaction, and the primers were as follows (the sequence used to insert a thrombin cleavage site between the sequence of sizzled and the V5-His 6 tag is underlined): 5Ј-GAACATGTCCGGAGTCTTCCTGCT-3Ј and 5Ј-ACTA-CCGCGTGGCACCAGACATTTATGATGTCTCCACCTC-CTTG-3Ј. The cDNA for sizzled in-frame with the V5-His 6 tag was then excised with KpnI/PmeI and subcloned into the pCEP4 vector (Invitrogen) digested with KpnI/NaeI. The cDNA for human sFRP-1 in the pBlueScriptR plasmid (ImaGenes) was subcloned into pcDNA3.1/V5-His-TOPO and pCEP4 vectors using the same strategy with the following primers: 5Ј-GGCATGGGCATCGGGCGCAGCG-3Ј and 5Ј-ACTACCGCGTGGCACCAGCTTAAACACGGACTGA-AAGGTGGG-3Ј.
Sizzled, sFRP-1, and their mutants were also produced in 293-EBNA cells after transfection with the pCEP4 constructs described in the previous section using Lipofectamine (Invitrogen) as the transfection agent. After 24 h, hygromycin B (300 g/ml) was added to the medium, and selection was maintained for the duration of cell amplification. Protein production was performed in Cellstack culture chambers (Corning) using Dulbecco's modified Eagle's medium (PAA) containing 50 g/ml gentamycin. Medium was collected every 2 days and stored at Ϫ80°C. Pooled medium was then centrifuged to remove cell debris and incubated with nickel nitrilotriacetic acid-agarose (Qiagen, 15 ml/liter) for 3 h at 4°C. The resin was then packed into a column, rinsed with 50 mM imidazole in 50 mM NaH 2 PO 4 , 0.3 M NaCl, pH 8, and His-tagged proteins were eluted with 250 mM imidazole in the same buffer. Pooled fractions were concentrated using Amicon Ultra centrifugal filter devices (Millipore), and imidazole was removed using PD-10 desalting columns (GE Healthcare) according to the manufacturer's instructions. Human soluble meprins ␣ and ␤ were produced in insect cells, purified by streptactin or Ni-NTA chromatography, and trypsin-activated as described (34,35). BMP-1cat (36) and recombinant astacin (37,38) were expressed in Escherichia coli, purified, and re-folded as reported. The catalytic domains of MMP-9 and -12 were produced as described (39,40). Mass Spectrometry-Sizzled was digested with sequencing grade trypsin (Promega; 20/1, protein/enzyme) for 1 h at 50°C with continuous stirring without reduction/alkylation. Liquid chromatography-mass spectrometry (LC-MS) experiments were performed at the Centre Commun de Microanalyse des Protéines of UMS3444 on a LTQ Velos mass spectrometer (Thermo Scientific) equipped with an Ultimate 3000 nanoLC pump (Dionex). A C 4 column (Dionex, 75-m inner diameter ϫ 15 cm, 300 Å pore, and 5-m particle size) was coupled online to the mass spectrometer through a nanospray needle (New Objective). A 20-min linear gradient was applied using mobile phase A (0.1% formic acid and 10% acetonitrile in water) and mobile phase B (0.1% formic acid and 80% acetonitrile in water) at 300 nl/min. The mass spectrometer was operated in MS mode only using zoom scan.
Activity Assays-BMP-1 activity in the presence of its substrates [ 3 H]procollagen I, procollagen III, pN-collagen V, prolysyl oxidase, and mini-procollagen III was measured as previously described (12) at 37°C in 50 mM HEPES, 0.15 M NaCl, 5 mM CaCl 2 , 0.02% Brij, pH 7.4 (buffer A) with incubation times and protein concentrations as indicated. Cleavage of mini-procollagen III by mTLD, mTLL-1, and meprins (41)  Surface Plasmon Resonance-The interaction between sizzled and BMP-1 was studied by surface plasmon resonance experiments using a Biacore T100 instrument (Plateau Production et Analyse des Protéines; UMS 3444). Sizzled was immobilized in 10 mM HEPES, 0.15 M NaCl, pH 7.4, on a CM5 sensorchip (series S) using amine coupling chemistry. The control channel was prepared with the same activation/deactivation procedure except that the protein solution was replaced by buffer alone. Before injection, analyte buffer was exchanged with 10 mM HEPES, 0.15 M NaCl, 5 mM CaCl 2 , 0.05% P-20, pH 7.4, using a Zeba Desalt Spin Column (Thermo Scientific). The same buffer was used as running buffer, and sensorgrams were recorded at 25°C with a flow rate of 30 l/min. Regeneration of active and control channels was performed using 2 M guanidinium chloride. Other surface plasmon resonance experiments were performed in a similar manner except that a different running buffer was used in experiments involving Wnt3a (10 mM HEPES, pH 7.4, 0.15 M NaCl, 3 M EDTA, 0.05% P-20, 1% CHAPS). Also, immobilization of BMP-1 or the anti-His antibody (R&D Systems) was performed at pH 4.5 (in 10 mM sodium acetate buffer), and corresponding interactions were regenerated with different solutions (50% ethylene glycol then 2 M NaCl for interactions involving BMP-1, 10 mM glycine, pH 2, to release sizzled captured on the anti-His antibody). Kinetic data were analyzed using Biacore T100 Evaluation software 2.0.2.
Molecular Modeling-The model for the Fz domain of sizzled was obtained by comparative homology modeling using the crystal structures of the cysteine-rich domain of secreted frizzled related protein 3 (PDB code 1IJX) and of the cysteinerich domain of mouse frizzled 8 (PDB code 1IJY) as templates (with a sequence identity of 35 and 32%, respectively) (42). Sequences were aligned using MAFFT with manual adjustments. Based on this multiple alignment, the program Modeler (43) was used to generate a three-dimensional representation of the Fz domain (from residues 26 to 136 of sizzled). The loop between residues 91 and 97 was subjected to several rounds of refinement as implemented in Modeler (43). The geometry of the final model was checked with MolProbity (44).

Sizzled, but Not Mammalian sFRPs, Can Inhibit Human
BMP-1-To evaluate which human sFRP is the most likely to inhibit human BMP-1, we compared their Fz domains with the Fz domain of sizzled (Fig. 1A). For sFRP-1, sFRP-2, and sFRP-5, the percentage of sequence identity is around 50 (50, 47.2, and 47%, respectively), whereas it is only around 30% for sFRP-3 and sFRP-4 (32.2 and 29%, respectively). We decided to analyze the effect on BMP-1 activity of one member of each subgroup, namely sFRP-1 and sFRP-4, as well as of sFRP-2 for which the largest amount of data is available. As most regulators of BTPs are substrate-specific (10, 12, 14 -16, 45), it seemed important to test the effect of sFRPs on the cleavage of several substrates.
The first substrates to be tested were the fibrillar procollagens I and III for which removal of C-propeptide by BTPs is necessary for incorporation into fibrils (46). First, the conversion of [ 3 H]procollagen I to pN-collagen I and C-propeptide was monitored by a quantitative assay based on the detection of the radioactivity emitted by the soluble C-propeptide after precipitation of procollagen and pN-collagen (12,47). In this assay human sFRP-1, mouse sFRP-2, and human sFRP-4 (100 nM) had no effect on BMP-1 activity, whereas the same concentration of sizzled from X. laevis led to a decrease in BMP-1 activity to 23 Ϯ 7% of its initial value ( Fig. 2A). A previous study reported that low concentrations of sFRP-2 (10 -20 nM) could enhance the cleavage of procollagen I by BMP-1, whereas higher concentrations (100 -200 nM) led to inhibition (28). To check for a possible concentration-dependent effect, we analyzed the cleavage of procollagen I in the presence of increasing concentrations of sFRP-2 (10, 50, 200, and 1000 nM) in conditions leading to 45% processing of the substrate by BMP-1. In our hands, however, sFRP-2 was still unable to inhibit or enhance BMP-1 activity (supplemental Fig. S2A) throughout this concentration range. Similarly, the processing of procollagen III into pN-collagen III was unaffected by mammalian sFRPs but was inhibited by sizzled (Fig. 2B). To eliminate the possibility that the lack of activity of recombinant sFRP-1 and -2 was due to denaturation or aggregation of these proteins, we checked their ability to interact with Wnt3a, one of their well known ligands (48), by surface plasmon resonance. As shown in supplemental Fig. S3A, human Wnt3a interacted well with both sFRPs when the latter were immobilized on a sensorchip.
Another substrate of BMP-1 is procollagen V, which regulates the size of collagen I fibrils. In contrast to procollagens I and III, BMP-1 processing can occur in both the N-and C-propeptide regions of procollagen V (49 -52). Here we have used the pN-collagen form of the procollagen ␣1(V) homotrimer, which is converted by BMP-1 to mature collagen by release of the N-propeptide. Only sizzled was able to inhibit this conversion, as can be seen by the decreased intensity of the band corresponding to the N-propeptide (Fig. 2C). The last physiological substrate to be tested was prolysyl oxidase, whose mature form is responsible for the stabilization of collagen and elastin networks through the initiation of the formation of covalent cross-links (53). As shown in Fig. 2D, the addition of BMP-1 led to an increase in intensity of the band corresponding to mature lysyl oxidase, which could be prevented only by sizzled.
In addition to these physiological substrates, we also tested the ability of sFRPs to inhibit the cleavage of mini-procollagen III, a model substrate derived from procollagen III (which consists of the C-propeptide, the C-telopeptide, and a very short triple helix made of the last 33 triplets of procollagen III (12)) and of the fluorogenic peptide Mca-YVADAPK(Dpn)-OH, originally described for caspases (20,54). In the presence of 100 nM sizzled, the residual activity of BMP-1 was only 4 Ϯ 3% that of the control activity with mini-procollagen III (Fig. 2E) and 27 Ϯ 3% that of the control activity with the fluorogenic peptide (Fig. 2F). As for sFRP-1, sFRP-2, and sFRP-4, these proteins neither inhibited nor enhanced BMP-1 activity with these substrates. Because other authors may have used slightly different buffer conditions, we checked whether small variations of pH (between 7.0 and 8.0), HEPES concentration (10 -200 mM), calcium concentration (0 -50 mM), or replacement of HEPES buffer by Tris might modify any effects of sFRP-2 on cleavage of mini-procollagen III by BMP-1 but found no differences (data not shown).
Finally, we tested the possibility that sizzled and the three sFRPs of interest could be substrates of BMP-1, but none of them was found to yield detectable cleavage products by SDS-PAGE when incubated for 3 h at 37°C in the presence of 100 nM BMP-1 (data not shown). Neither were they cleaved by the two other main human astacin-like proteases, meprins ␣ and ␤ (55), which were recently demonstrated to have procollagen C-proteinase activity (41). Because one report suggested that sFRP-2 selectively targets this activity of the BMP-1/tolloid-like proteinases (9), we also verified that cleavage of mini-procollagen III by meprins was not affected by sFRP-1 and sFRP-2 (supplemental Fig. S2B).
Altogether, these studies indicate that sFRP-1, sFRP-2, and sFRP-4 are neither inhibitors, nor enhancers, nor substrates of human BMP-1, at least in our conditions. In contrast, it appears that Xenopus sizzled is an inhibitor not only of Xenopus BTPs but also of human BMP-1.
Sizzled Is a Potent Inhibitor of Human BMP-1-We next performed a kinetic analysis to determine the inhibition constant (K i ) for sizzled on BMP-1. For this purpose we used the fluoro-genic peptide as a substrate. Preliminary experiments indicated that the IC 50 value of sizzled increased with concentration of BMP-1 in a linear manner (data not shown). This is a distinctive characteristic of tight binding inhibitors, i.e. inhibitors for which K i is at or below the enzyme concentration used in the assay (56). As a consequence, we used the mathematical treatment of tight binding inhibitors described by Morrison (57) (supplemental Table S1). When the fractional velocity of BMP-1 was plotted as a function of sizzled concentration for various substrate concentrations, nonlinear regression of these curves using the Morrison equation can be performed to determine apparent K i values (K i app ; Fig.  3A). A secondary plot of K i app values as a function of substrate concentration shows a linear dependence characteristic of competitive tight binding inhibitors (Fig. 3A) and gives a K i value of 1.5 Ϯ 0.5 nM.  (71). Residue numbering refers to the sequence of X. laevis sizzled (beginning at the signal peptide), and residues mutated in this study are highlighted in yellow. Also shown are secondary structure elements found in the structure of the mouse sFRP-3 Fz domain and in the model of the sizzled Fz domain as well as the positions of cysteines involved in intradomain disulfide bonds. The figure was prepared using ESPript (72). xl, X. laevis; am, Ambystoma mexicanum; dr, Danio rerio; h, human; m, mouse. B, shown is a schematic representation of the model of the sizzled Fz domain. Disulfide bridges and amino acids studied by site-directed mutagenesis are represented by sticks (colored by atom type). C, a surface representation of the model of the sizzled Fz domain is shown. Coloring based on site-directed mutagenesis data shows the effects of mutations on sizzled inhibiting activity (light blue ϭ no effect; light orange ϭ some effect; orange ϭ strong effect).
We also studied the interaction between sizzled and BMP-1 by surface plasmon resonance. As shown in supplemental Fig.  S3B, BMP-1 was found to bind immobilized sizzled in a concentration-dependent manner. Incidentally, the interaction was found to be reversible and could be regenerated with 2 M guanidinium chloride. The curves were best fitted with the "heterogeneous ligand" model, assuming that there are two binding sites for BMP-1 on sizzled, and the two calculated dissociation constants (K D ) were 6.0 nM (accounting for 58% of the signal) and 37 nM (accounting for 42% of the signal). This heterogeneity was induced by the coupling procedure as it disappeared when sizzled was captured with an anti-His antibody, leading to a single K D of 14 nM (by single-cycle analysis; Fig. 3B). In addition, the fast on-rate of the BMP-1/sizzled interaction (2.5 10 5 M Ϫ1 s Ϫ1 ) together with the absence of effect of enzyme preincubation with inhibitor in activity assays (data not shown) indicate that sizzled does not behave as a "slow-binding" inhibitor.
These two complementary approaches both showed that sizzled strongly binds and inhibits human BMP-1 in a reversible manner with an inhibition constant in the nanomolar range. More unexpectedly, when the interaction of BMP-1 with sFRP-1, -2, and -4 was tested by surface plasmon resonance (supplemental Fig. S3C), we found that sFRP-1 and -2, but not sFRP-4, were also able to bind BMP-1.
Sizzled Is a Specific Inhibitor of BTPs-We next investigated if sizzled was able to inhibit two other human BTPs: mTLD and mTLL-1. These proteinases were also able to cleave mini-procollagen III. In the presence of increasing concentrations of sizzled, mTLD and mTLL-1 were both inhibited to the same extent as BMP-1 (Fig. 4A). Then, the effect of sizzled was tested on other members of the astacin family, including crayfish astacin, the founder member of the family, and human meprins ␣ and ␤ (6). Using the fluorogenic peptide substrate, we found no evidence of inhibition of crayfish astacin, meprin ␣, nor meprin ␤ by sizzled, even at a concentration of 1 M (Fig. 4B). This result was confirmed for meprins on the mini-procollagen III substrate with 100 nM sizzled (supplemental Fig. S2B). The remarkably high specificity of sizzled was also shown with a panel of five representative MMPs (MMP-1, MMP-8, MMP-9, MMP-12, and MMP-13), which were found to be essentially unaffected by sizzled at concentrations more than 100 times greater than the K i found for BMP-1 (Fig. 4C). We conclude that sizzled is a potent and specific inhibitor of BTPs.
Sizzled Inhibits the Catalytic Domain of BMP-1 by Its Fz Domain-According to their finding that sizzled inhibited Xenopus BTPs in a competitive manner, Lee et al. (20) suggested that sizzled bound the catalytic domains of these enzymes. In contrast to this proposal, Kobayashi et al. (9) showed data from pulldown assays indicating that sizzled  , n ϭ 3). B, procollagen III (44 nM) was incubated with BMP-1 (23 nM) for 3 h, and the products were analyzed by SDS-PAGE (6% acrylamide in the separating gel, reducing conditions) followed by staining with EZ Blue. C, partially purified pN-collagen V homotrimer was incubated with BMP-1 (48 nM) for 6 h, and release of the N-propeptide was analyzed by SDS-PAGE (10% acrylamide, reducing conditions) followed by staining with EZ Blue. D, prolysyl oxidase was incubated with BMP-1 (19 nM) for 2 h. Analysis was by SDS-PAGE (10% acrylamide, reducing conditions) followed by Western blotting with an anti-V5 antibody (Invitrogen). E, mini-procollagen III (330 nM) was incubated with BMP-1 (16 nM) for 1.5 h, and the products were analyzed by SDS-PAGE (8% acrylamide, non-reducing conditions) followed by staining with SYPRO Ruby (Sigma) and detection/quantification on a Storm 860 Imager. F, the fluorogenic quenched peptide (20 M) was incubated with BMP-1 (21 nM), and the fluorescence emitted by the products was monitored for 20 min. Activities are expressed relative non-reducing to activity in the absence of any sFRP and correspond to the means Ϯ S.D. (n ϭ 4).
bound BTPs through the non-catalytic domains. To further investigate this question and confirm our kinetic experiments suggesting a competitive inhibition, we tested the ability of sizzled to inhibit the catalytic domain of BMP-1 (BMP-1cat) produced in bacteria (36). Using the fluorogenic peptide as a substrate, a BMP-1cat concentration 27-fold higher than the concentration of BMP-1 was necessary to reach the same reaction rate. This observation can be explained either by the fact that the catalytic domain alone is less efficient than full-length BMP-1 in cleaving the peptide or by the fact that a fraction of BMP-1cat produced in bacteria is inactive. As shown in Fig. 5A, sizzled inhibited BMP-1cat in a dose-dependent manner, with an IC 50 value of 94 Ϯ 1 nM. This experiment proves that sizzled acts directly on the catalytic domain of BMP-1, albeit that additional interactions with non-catalytic domains cannot be excluded.
To determine which domain of sizzled is responsible for BTP inhibition, we produced and purified its two separate domains. Residues Cys-115 in the Fz domain and Cys-159 in the NTR domain were also mutated to serines to avoid the presence of free cysteines (see below). As shown in Fig. 5B, the Fz domain alone was sufficient to inhibit the cleavage of the fluorogenic peptide by BMP-1. In contrast, when BMP-1 was incubated with the NTR domain, no visible inhibitory effect was observed. This is consistent with the observations of Lee et al. (20) on the selective binding of the Fz domain to Xenopus Xlr.
In summary, sizzled competitively inhibits the catalytic domain of BMP-1 by its Fz domain. Given the fact that the Fz domain of sFRP-1 and -2 is 50% identical to that of sizzled, it is  Incubation times were calculated so that around 70% of mini-procollagen III was converted to C-propeptide and N-terminal cleavage product (50 min for BMP-1, 3 h for mTLD, and 5 h for mTLL-1). Analysis was by SDS-PAGE (4 -20% acrylamide, reducing conditions) followed by staining with Coomassie Blue. quite surprising that these sFRPs were unable to inhibit human BTPs. In an attempt to explain why sizzled is an inhibitor of BTPs whereas human sFRP-1 and mouse sFRP-2 are not, we next analyzed differences in sequences and secondary structures between these three proteins.
Sizzled Has an Interdomain Disulfide Bridge in Contrast to Mammalian sFRPs-One of the major differences between sFRPs and sizzled is found in their disulfide bonding patterns. The pattern of disulfide bonds in human sFRP-1 has previously been determined by Chong et al. (58); there are five disulfide bridges in the Fz domain and three in the NTR domain. Noteworthy, sizzled presents an additional cysteine residue (Cys-115) in its Fz domain and a missing cysteine residue in its NTR domain, leaving Cys-159 unpaired (supplemental Fig. S4). Chong et al. (58) suggested that cysteines 115 and 159 might actually form an interdomain disulfide bridge. Another hypothesis would be that these two cysteines remain in their reduced state, thereby being potentially able to coordinate the active site zinc atom of BTPs or to react with a putative vicinal disulfide bridge located above the active site of BMP-1 (59).
To investigate these two possibilities, we cleaved sizzled with trypsin and analyzed the tryptic fragments by electrospray ionization-LC-MS. If one assumes that the disulfide bonds of sFRP-1 are conserved in sizzled and that sizzled contains an additional interdomain disulfide bridge between Cys-115 and Cys-159, tryptic digestion should lead to three major peptides (calculated masses 2155.5, 6046.1, and 9353.6 Da) (supplemental Fig. S4). These three peptides were indeed found by electrospray ionization LC-MS analysis ( Table 1). Among them, the 9353.6-Da peptide includes five cysteine residues from the Fz domain and three from the NTR domain. To check that the interdomain disulfide bridge is formed between Cys-115 and Cys-159, we mutated these two residues to serines. This mutant as well as other mutants described below was characterized by MS (supplemental Table S2), and circular dichroism analysis showed that the mutations did not induce major changes in secondary structure elements (supplemental Fig. S5). Tryptic digestion of this mutant form of sizzled no longer generated the 9353.6-Da peptide but, rather, two shorter peptides (calculated masses 4205.8 and 5117.6 Da) corresponding to fragments from its NTR and Fz domains, respectively (Table 1). Altogether, these data indicate that sizzled contains an interdomain disulfide linkage between residues Cys-115 and Cys-159 that is not present in sFRP-1. However, using mini-procollagen III as a substrate, we found that mutation of these two cysteines had no effect on the ability of sizzled to inhibit BMP-1 ( Table 2).
Analysis of Differences in Sequence between Sizzled and sFRP-1 and -2-We next sought to identify residues specific to sizzled that might be critical for BMP-1 inhibition. As previously reported (9,20,21) and as shown in Table 2, mutation of residue Asp-92 to asparagine dramatically decreased sizzled inhibitory activity. Sequence alignment of Fz domains from sizzled and human sFRPs (Fig. 1A) shows that this aspartate is conserved in sFRP-1, -2, and -5. Taking this into account, plausible explanations for the inability of sFRP-1 and -2 to inhibit BMP-1 are: 1) this aspartate may be hindered by other parts of the sFRP that prevent its access to BMP-1, and 2) there may be other residues specific to sizzled that are important for the inhibition.
To test the first hypothesis, we produced two constructs derived from sFRP-1, one without the NTR domain (sFRP-1-  Fz) and one without the N-terminal extension (sFRP-1-⌬Nterm). However, both these constructs proved to be unable to inhibit BMP-1 ( Table 2). To explore the second hypothesis, we searched for residues that are conserved in sizzled but not in sFRP-1 and -2 and that are located in putative surface-exposed loops. For this purpose, we modeled the structure of the Fz domain of sizzled based on the known three-dimensional structures of the mouse sFRP-3 and frizzled-8 Fz domains (42) (Fig.  1, B and C). Using this model and the sequence alignment in Fig.  1A, we identified four promising target residues: Phe-94 (located in the same loop as Asp-92), Ser-43, Glu-44, and Lys-26 (located in nearby loops). Another target was Gln-96, in the same loop as Asp-92, which is also present in sFRP-2 but replaced by tyrosine in sFRP-1. All these residues were individually mutated to alanines, with the exception of Ser-43 and Glu-44, which were mutated as a pair. Among residues that are located in the same loop as Asp-92, we found that the Q96A mutation only led to a small decrease of inhibition, whereas sizzled-F94A lost most of the inhibitory potency of sizzled with an IC 50 more than 50 times higher ( Table 2). This suggests that several residues in the loop encompassing Asp-92 may play an important role in BMP-1 inhibition. To determine if the absence of sFRP-1 inhibiting activity is due only to its different amino acid sequence in this loop, we replaced it by that of sizzled. As shown in Table 2, this mutant of sFRP-1 (sFRP-1-R124T/P125F/Y127Q) remained unable to inhibit BMP-1.
This above result suggests that residues located in other putative loops might also play a role in BMP-1 inhibition. We found indeed that the double mutation S43A/E44A led to a severe drop in sizzled inhibiting activity, with an IC 50 value almost 30-fold higher than that of wild-type sizzled. As for the mutation K26A, this had no effect on the ability of sizzled to inhibit BMP-1 ( Table 2). In conclusion, comparison of amino acid sequences of sizzled and sFRP-1 and -2 led us to identify new residues in sizzled involved in BMP-1 inhibition, both in the same loop as Asp-92 and in a nearby loop.

DISCUSSION
Until recently, BTPs had remained without any known endogenous specific inhibitor despite their implication in important processes during development and tissue repair. The unexpected discovery that sizzled acts as a very potent inhibitor of BTPs during dorso-ventral patterning in Xenopus and zebrafish (24 -26) raised the possibility that a similar regulatory mechanism might also be at work in mammals. Obvious candidates were mammalian sFRP-1-5, which have been extensively studied (9,28,29) but for which only one (sFRP-2) has emerged as a potential regulator of mammalian BTPs. However, the results are strongly divergent from one study to another and range from inhibition (28) to enhancement (9) or absence of any effect (29). The main basis of these studies was SDS-PAGE analysis of the effect of sFRP-2 on procollagen I processing by BMP-1 in vitro, with all the pitfalls inherent to gel-based quantification and limited demonstration of the reproducibility of the results.
For these reasons we compared the effects of three different mouse or human sFRPs (-1, -2, and -4) and of Xenopus sizzled on procollagen I processing using a quantitative assay. In these conditions we could unambiguously demonstrate that none of the tested mammalian sFRPs, not even sFRP-2, was capable of inhibiting or enhancing procollagen I processing, whereas sizzled led to a marked decrease of BMP-1 activity on this substrate. sFRP-2 was also tested at several concentrations ranging from 10 to 1000 nM and clearly appeared to have no significant effect on the procollagen C-proteinase activity of BMP-1, thus confirming the results obtained by von Marschall and Fisher (29). Two other precursor forms of fibrillar collagens, procollagen III and pN-collagen V, together with the cross-linking enzyme prolysyl oxidase were also tested for the first time in the presence of sFRPs, and these experiments confirmed the absence of effect of the mammalian sFRPs on BMP-1. Finally, another quantitative assay using a quenched fluorescent peptide showed a similar behavior with no effect of sFRP-1, -2, and -4 and more than 80% inhibition in the presence of 100 nM sizzled. Altogether, these results question previous studies showing a direct effect of sFRP-2 on the procollagen C-proteinase activity of human BMP-1 and together with the results reported for chordin (9), dentin matrix protein-1, and dentin sialophosphoprotein (29) suggest that the absence of effect of mouse and human sFRPs can be extended to all BMP-1 activities.
It remains that interesting biological effects have been attributed to sFRP-2 and -3 in fibrotic diseases where excessive deposition of disorganized extracellular matrix, especially collagen, is observed. However, here again, controversial results have been obtained by the two groups who studied the effect of sFRP-2 in models of myocardial infarction, showing that the presence of sFRP-2 either aggravates or reduces the extent of fibrosis in relation to the level of collagen deposition (9,28). We note that sFRPs seem to be extremely versatile molecules with biological effects that are highly context-dependent (19). For example, there is little certainty concerning their active forms in tissues (alone or in complex with other proteins), and two reports at least suggest that sFRP-1 and -3 could be present as monomers or form dimers through their Fz domains (42,60). Thus, subtle differences in the state/conformation of sFRP-2 might explain some of the discrepancies between studies, and further insights into the structural properties of sFRPs would certainly help to settle some of the controversies. Moreover, in

values of sizzled, sFRP-1, and their mutants
Mini-procollagen III (340 nM) was incubated for 2 h at 37°C with BMP-1 (17 nM) in the absence or presence of increasing concentrations of sizzled, sFRP-1, or their mutants. The products were analyzed by SDS-PAGE (4 -20% acrylamide, reducing conditions) followed by staining with SYPRO Ruby (Sigma). Densitometric analysis of the gels was performed on a Storm 860 imager (GE Healthcare) and quantified using ImageQuant. Relative activity of BMP-1 as a function of inhibitor concentration was plotted. The curves were fitted using GraphPad Prism 5.

Sizzled
both the above-mentioned studies, sFRP-2 seems to affect the total amount of procollagen I expressed by cells, complicating the analysis of the level of processing and suggesting that pathways unrelated to BMP-1 activity could also play roles in the infarcted heart. Indeed, sFRP-2 is known mainly as a Wnt antagonist, and as such, has been implicated in the control of cell apoptosis, proliferation, and differentiation (19). In agreement with this, mesenchymal stem cells overexpressing sFRP-2 seem to have enhanced ability to promote myocardial regeneration through increased proliferation and survival involving both Wnt and BMP signaling (61,62). In addition, sFRP-2 has been shown to be up-regulated in hypertrophic scars resulting from abnormal healing of the skin (63), and sFRP-2 silencing in hypertrophic scar-derived fibroblasts promoted their apoptosis and led to down-regulation of ␣-smooth muscle actin, procollagen I, and procollagen III (64). In conclusion, although sFRP-2 clearly appears as an important player during healing of several organs, including heart and skin, its effects can also be mediated by the modulation of signaling pathways not directly related to BTPs. Noteworthy, although sFRP-3 seems to be without effect on the procollagen C-proteinase activity of BMP-1 (28), its therapeutic use for the prevention of hypertrophic scars has been patented (65).
Using surface plasmon resonance, we observed that sFRP-2 binds BMP-1, in agreement with previous studies (9,28). Interestingly, sFRP-1 showed an even greater ability to interact with BMP-1, whereas sFRP-4 was unable to interact with the proteinase. Also, we found that sizzled lost its ability to interact with BMP-1 when the latter was immobilized on the sensor chip (data not shown), whereas sFRP-1 and -2 did not. This observation suggests that the three proteins do not share the same binding site on BMP-1 and may explain why they are not equally able to regulate BMP-1 activity. However, the potential role of the interaction between BMP-1 and sFRP-1 and -2 needs further clarification and should be carefully evaluated in the numerous biological contexts where they are co-expressed. These interactions could serve to control important aspects of the biological activity of BMP-1 or sFRPs, which are difficult to assess in vitro, like their diffusion or localization in specific micro-environments. Similar mechanisms have already been shown to regulate the activity of Wnt molecules, which can diffuse more efficiently in the presence of sFRPs (66).
Because the frizzled domain of sizzled displays around 50% identity and 77% similarity with the frizzled domains of human sFRP-1 and -2, the different behaviors of the three proteins must be explained by subtle modifications in sequences or structures of the three proteins. Careful examination of the amino acid composition and of a model of the frizzled domain of sizzled led us to identify three possible causes for the absence of activity of sFRP-1 and -2. The first was an additional cysteine residue in the Fz domain of sizzled which was unambiguously demonstrated in this study to be involved in an interdomain disulfide bridge and to have no effect on the inhibitory activity of sizzled. The second major difference between sizzled and sFRP-1 is a long N-terminal extension that is found before the Fz domain itself and is specific to sFRP-1 and, to a lesser extent, to sFRP-2. However, no appearance of sFRP-1 activity toward BMP-1 was observed upon deletion of this N-terminal extension.
Much more conclusive was the comparison of the amino acids found in putative surface-exposed loops in sizzled and sFRP-1 or -2. As expected from previous reports (9,20), the Asp-92 residue in sizzled, which is also present in sFRP-1 and -2, proved crucial for the inhibition of human BMP-1. This mutation is indeed responsible for the ogon mutation in zebrafish (25,26), resulting in non-functional sizzled protein. This points to the loop surrounding Asp-92 as an important determinant of sizzled activity, as was confirmed by the decreased activity of the F94A mutant of sizzled. However, contributions from the non-conserved residues in the loop were not sufficient to explain the differences between sizzled and sFRP-1 as replacement of the three residues that were different in a region spanning six residues before and after Asp-92 did not confer inhibitory activity to sFRP-1. Further analysis revealed three other residues located in two different loops that are present in sizzled and absent in sFRP-1 and -2. Interestingly, residues Ser-43 and Glu-44 (also conserved as a pair in Xenopus crescent) but not Lys-26 seem to be involved in the interaction surface between sizzled and BMP-1, thereby delineating the first map of the binding site of BMP-1 on sizzled. As shown in Fig. 1, B and C, all the residues that were shown here to be important for this interaction are clustered in a defined region of the model of sizzled involving the loop preceding helix ␣1 and the loop between helices ␣3 and ␣4. Most importantly also, we have shown that both conserved and non-conserved residues were required to allow the inhibitory complex with BMP-1 to form, explaining why sFRP-1 and -2 cannot form "productive" complexes with BMP-1. Interestingly, Ser-43 and Glu-44 in sizzled align with the KK motif in the short sequence 74 YKKM of sFRP-1, which has been shown to be necessary for Wnt3a antagonist activity (67), pointing to the important functional role played by this loop.
Our study also sheds light on how sizzled inhibits BMP-1. First of all, the absence of inhibition by the NTR domain of sizzled shows that the mechanism of BMP-1 inhibition by sizzled differs significantly from the inhibition of MMPs by TIMPs, which involves binding of the NTR domain of TIMPs in the catalytic pocket of MMPs (68). However, we demonstrate here for the first time that sizzled, like TIMPs, behaves as a tight binding inhibitor. Using the appropriate formalism for tight binding inhibitors, we showed that sizzled is a competitive inhibitor of BMP-1 with an inhibition constant as low as 1.5 Ϯ 0.5 nM. In addition, inhibition by sizzled of the activity of the catalytic domain of BMP-1 confirms that sizzled mainly binds the catalytic domain of BMP-1, in agreement with the initial study by Lee et al. (20) on Xenopus tolloid Xlr that was then contradicted by a subsequent study involving human BMP-1 and mTLL-1 (9). An interesting consequence of this finding is that the conserved aspartate, which is crucial for sizzled activity (Asp-92), is probably located in the active site of BMP-1, but despite the strong preference of BTPs for aspartate in P1Ј, this does not trigger cleavage of sizzled (nor of any of the tested sFRPs) by BMP-1. This observation is reminiscent of the mechanism used by crayfish astacin, the founder member of the astacin family, to achieve latency. Indeed, the three-dimensional structure of the pro-astacin solved by Guevara et al. (69) reveals that the carboxyl group of the side chain of a conserved aspartic residue from the propeptide coordinates the catalytic zinc ion, thereby positioning the propeptide in a reverse orientation compared with the "normal" orientation seen with substrates. This precludes access of the substrate to the active site and activation of the water molecule required to achieve proteolysis. Because it is very likely that the equivalent aspartate in BMP-1 and other BTPs plays a similar role (69), we can also speculate that Asp-92 in sizzled could coordinate the zinc and thereby prevent its cleavage by the protease. Also, the threedimensional structure of the catalytic domain of BMP-1 (59) revealed the presence of an arginine residue (Arg-176 corresponding to Arg-302 in the full-length BMP-1) in the S1Ј pocket that would be well positioned to interact with Glu-44 in sizzled and to further stabilize the enzyme-inhibitor complex.
Another interesting feature of sizzled that was also demonstrated here is its high specificity for BTPs. Although all three BTPs tested (BMP-1, mTLD, and mTLL-1) were inhibited by sizzled, other members of the astacin family, such as astacin itself and human meprins ␣ and ␤ (that demonstrate the same preference for acidic residues in P1Ј as BTPs (70)), were not inhibited or enhanced by sizzled. Similarly, sizzled had no effect on five well characterized MMPs (-1, -8, -9, -12, and -13) and, in contrast to other sFRPs including crescent (27), also seems unable to modulate Wnt signaling (23)(24)(25). Consequently, it seems that thanks to its potency and specificity, sizzled could be an appropriate tool to study the biological roles of BTPs in complex systems and potentially to derive novel classes of BTP inhibitors that could be used to prevent fibrotic disorders.