Mutational Analysis of Norrin-Frizzled4 Recognition*

Norrin and Frizzled4 (Fz4) function as a ligand-receptor pair to control vascular development in the retina and inner ear. In mice and humans, mutations in either of the corresponding genes lead to defects in vascular development. The present work is aimed at defining the sequence determinants of binding specificity between Norrin and the Fz4 amino-terminal ligand-binding domain (the “cysteine-rich domain” (CRD)). The principal conclusions are as follows: 1) Norrin binds to the Fz4 CRD and does not detectably bind to the 14 other mammalian Frizzled and secreted Frizzled-related protein CRDs; 2) Norrin and Xenopus Wnt8 recognize largely overlapping regions of the Fz4 CRD; 3) surface determinants on the Fz4 and Fz8 CRDs that allow Norrin to distinguish between these two CRDs reside within several small regions on one face of the CRD; 4) Norrin function depends critically on three pairs of cysteines that form the highly conserved trio of disulfide bonds shared among all cystine knot proteins, but the remaining two putative disulfide bonds are less important; 5) Norrin-CRD binding depends on a largely contiguous group of amino acids in the extended β-sheet domain of Norrin that are predicted to face away from the interface between the two monomers in the Norrin homodimer; 6) Norrin-CRD binding is strongly modulated by interactions involving charged amino acid side chains; and 7) Norrin-CRD binding is enhanced ∼10-fold by the addition of heparin. These observations are discussed in the context of Frizzled signaling and the structure and function of other cystine knot proteins.

"canonical" signaling pathway in conjunction with the coreceptors Lrp5 or Lrp6 (8). Wnts bind to a compact domain at the Frizzled NH 2 terminus referred to as the cysteine-rich domain (CRD) 2 (9,10), for which a high resolution crystallographic structure has been obtained (11). In mammals, closely related CRDs are present in a family of five secreted Frizzled-related proteins (sFRPs), which probably function as competitive inhibitors of Wnt-Frizzled signaling (12,13).
Recently, Norrin, a protein that is not a member of the Wnt family, has been identified as a ligand that binds the Fz4 CRD with high affinity and potently activates the canonical signaling pathway (7). Norrin is a small, cysteine-rich, secreted protein with weak homology to the transforming growth factor (TGF)-␤ family of ligands (14 -17). In humans, mutations in the corresponding gene (NDP) cause Norrie disease, an X-linked disorder characterized by hypovascularization of the retina and a severe loss of visual function (18). A milder retinal hypovascularization (familial exudative vitreoretinopathy) is seen in humans heterozygous for mutations in the genes coding for Fz4 (19,20) or the coreceptor Lrp5 (21), and severe hypovascularization is seen in humans and mice carrying a homozygous loss of function mutations in Lrp5 (22,23). Mice carrying a targeted deletion of the Ndp gene (hemizygous mutant males or homozygous mutant females) or of both copies of the Fz4 gene exhibit nearly identical retinal hypovascularization phenotypes, as well as a progressive enlargement and subsequent loss of blood vessels in the stria vascularis, the specialized epithelium that generates the endolymphatic fluid within the inner ear (7, 24 -26). Norrin binds with nanomolar affinity to the Fz4 CRD but not to several other Frizzled CRDs. Like Wnts, Norrin associates with the extracellular matrix, which limits its range of action to those target cells immediately surrounding its site of synthesis. Thus, Norrin appears to function in many respects like a Wnt despite a complete absence of primary sequence homology with the Wnt family.
In this paper, we have addressed several questions raised by the discovery that Norrin and Fz4 constitute a ligand-receptor pair. First, how selective is the binding of Norrin to the Fz4 CRD as compared with other Frizzled and sFRP CRDs? Second, which regions of the Fz4 CRD are responsible for Norrin binding, and what is the relationship of these regions to those responsible for Wnt binding? Finally, which regions of Norrin are involved in Fz4 binding and canonical pathway activation? * This work was supported by the Howard Hughes Medical Institute and the National Institutes of Health. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. □ S The on-line version of this article (available at http://www.jbc.org) contains supplemental As described below, Norrin has the ability to discriminate between the Fz4 CRD and the 14 other mammalian Frizzled and sFRP CRDs, and by site-directed mutagenesis we delimit the regions on both Norrin and Fz4 responsible for this recognition. These results have implications for other ligand-receptor families in which closely related proteins exhibit differential recognition of potential binding partners.

MATERIALS AND METHODS
Site-directed Mutagenesis-Mutations were constructed by tandem PCR. All DNA segments derived from PCR were sequenced to confirm the presence of the desired mutation and to rule out spurious mutations.
Production of Xenopus Wnt8 (Xwnt8), Norrin, and Fz4 CRD Fusion Proteins-AP-3Myc-Norrin was secreted from transiently transfected 293 cells using a vector with a cytomegalovirus immediate early gene enhancer and promotor; the AP-3Myc-Norrin was collected in Dulbecco's modified Eagle's medium/F-12 medium containing penicillin/streptomycin and 10% calf serum and stored at 4°C. Xwnt8-Myc-AP was produced under the control of a metallothionein promotor in stably transfected hygromycin-resistant Drosophila S2 cells. The secreted fusion protein was collected in Schneider Drosophila medium (Invitrogen) supplemented with penicillin/streptomycin, 10% calf serum, 50 g/ml hygromycin, and 0.5 mM CuSO 4 and stored at 4°C. To produce Fz CRD-IgG fusion proteins, the CRD (i.e. the 114-amino acid region extending from the first to the tenth conserved CRD cysteine) was inserted between MluI and ApaI sites in a vector that contains a cytomegalovirus immediate early gene enhancer and promotor, followed by DNA coding for the mouse Fz8 signal peptide, the site for CRD insertion, 25 amino acids of Fz8 "linker" sequence immediately COOH-terminal to the Fz8 CRD, and the constant region of human IgG. CRD-IgG fusion proteins were secreted from transiently transfected 293 cells, collected in serum-free Dulbecco's modified Eagle's medium/F-12 containing penicillin/streptomycin, and stored in aliquots at Ϫ80°C.
Binding to Cell Surface CRD-Myc-GPI-To display CRDs at the cell surface, each CRD was inserted between MluI and ApaI sites in a vector that contains a cytomegalovirus immediate early gene enhancer and promotor followed by DNA coding for the mouse Fz8 signal peptide, the site of CRD insertion, 25 amino acids of Fz8 "linker" sequence immediately COOH-terminal to the Fz8 CRD, a Myc epitope, and the COOH-terminal GPI-anchoring peptide from decay activating factor (10). The CRD-Myc-GPI vectors were transiently transfected with Fugene 6 into COS cells that were grown on uncoated circular glass coverslips in 24-well trays in 293 cell medium as described above. Two days after transfection, the coverslips were incubated in fresh growth medium lacking bicarbonate and containing Xwnt8-Myc-AP conditioned medium (diluted 1:2) or AP-3Myc-Norrin conditioned medium (diluted 1:5) or in fresh growth medium lacking bicarbonate and containing 0.1% calf serum with anti-Myc monoclonal antibody (ascites diluted 1:1,000). After gentle rocking at 4°C for 2 h, the coverslips were washed four times with PBS supplemented with 1 mM calcium and 1 mM magnesium (PBS/Ca/Mg), fixed in 0.5% gluteraldehyde in PBS/Ca/Mg at room temperature for 15 min, washed twice with PBS/Ca/Mg, and heated in a water bath at 65°C for 90 min to inactivate endogenous alkaline phosphatase (AP). Immobilized AP was visualized with nitro blue tetrazolium/5bromo-4-chloro-3-indolyl phosphate.
Quantitative Binding to CRD-IgG Fusion Proteins-The wells of a 96-well plate carrying immobilized protein A (Reacti-Bind plates; Pierce) were coated with CRD-IgG fusion proteins by incubating overnight at 4°C with 50 l of a 1:5 dilution of CRD-IgG in serum-free conditioned medium, an amount of CRD-IgG that appears to saturate the binding capacity of the wells. The wells were then incubated with 1% bovine serum albumin in PBS for 3 h at room temperature; washed three times with PBS/Ca/Mg, 0.1% bovine serum albumin; and incubated with serial 2-fold dilutions of Xwnt8-Myc-AP, AP-3Myc-Norrin, or AP-conjugated anti-human IgG. After a 3-h incubation at room temperature, the wells were washed three times in PBS/Ca/Mg, 0.1% bovine serum albumin and once in PBS/Ca/Mg, and the bound AP activity was measured using a soluble 5-bromo-4chloro-3-indolyl phosphate/tetrazolium enzyme assay (Blue-Phos; Kirkegaard and Perry Laboratories).
For comparisons of Fz4 CRD-IgG binding between wild type (WT) AP-3Myc-Norrin and the various AP-3Myc-Norrin mutants, the relative concentration of AP in the conditioned medium was determined for each construct using the BluePhos substrate. The relative efficiency of binding was determined by comparing the binding signal at the interpolated midpoint of the WT AP-3Myc-Norrin dilution series (i.e. between dilutions of 1:2 and 1:4; geometric mean, 1:2.8) with the binding signal at the interpolated point along the dilution series of each mutant AP-3Myc-Norrin, corresponding to an AP concentration in the conditioned medium equivalent to the 1:2.8 dilution of the WT AP-3Myc-Norrin. By performing all of the binding assays with a dilution series of the AP fusion protein, quantitative comparisons with the WT could be made despite ϳ3-fold variations in AP yield between the different AP-3Myc-Norrin conditioned media. Two independent preparations of AP-3Myc-Norrin conditioned medium were tested for each Norrin mutant.
Heparin Binding-For heparin affinity purification of AP-3Myc-Norrin, 0.25 ml of heparin-Sepharose (Sigma) was washed three times with PBS/calcium/magnesium and then incubated with 5 ml of AP-3Myc-Norrin in complete conditioned medium (described above) at room temperature for 3 h with gentle end-over-end rotation. The resin was transferred to a column and washed with PBS/calcium/magnesium, and the AP-3Myc-Norrin was eluted with PBS/calcium/magnesium containing 1 M NaCl. The partially purified AP-3Myc-Norrin was diluted with 7 volumes 10 mM NaPO 4 , pH 7.2, calcium/ magnesium supplemented with 0.01% bovine serum albumin (to reduce the final NaCl concentration to 150 mM) and stored at 4°C. Binding to Fz4 CRD-IgG that had been captured onto protein A microwells was performed in the presence of various concentrations of porcine intestinal heparin (Sigma).
Luciferase Assays-For a typical luciferase assay, a G418-resistant stable 293 cell line (STF cells) (7) carrying the Super Top Flash firefly luciferase reporter of canonical Wnt signaling (a construct with seven tandem LEF/TCF binding sites) was transfected in triplicate in a 24-well tray using Fugene 6 with the following quantities of expression plasmid DNA per well: Norrin, 50 ng; Fz4, 50 ng; Lrp6, 50 ng; Renilla luciferase, 1 ng. Two days after transfection, cells were washed with PBS and assayed using the Promega dual luciferase assay reagents. The firefly luciferase activity was normalized to the co-expressed Renilla luciferase activity, and the average of the triplicate samples was determined.
Modeling Fz4 CRD and Norrin Structures-The locations of different amino acid side chains on the Fz4 CRD surface were modeled by highlighting the corresponding residues in the high resolution Fz8 CRD crystal structure (11) using the program PyMol (available on the World Wide Web at pymol. sourceforge.net/). The amino acid sequences of the Fz4 and Fz8 CRDs align without recourse to insertion or deletion. To model the three-dimensional structure of Norrin, we first superimposed the structures of TGF-␤ family members TGF-␤2 (Protein Data Bank code 2TGI) (27), TGF-␤3 (Protein Data Bank code 1TGJ) (28), BMP7 (Protein Data Bank code 1BMP) (29), and BMP2 (Protein Data Bank code 2GOO) (30) using the program CCP4mg (31). This superposition was then used to align the amino acid sequences of these proteins. The subunits of TGF-␤ family dimers consist of two extended ␤-hairpins separated by an ␣-helical region. Each of the four ␤ strands that make up the two hairpins contains cysteines that are conserved between TGF-␤ family members and Norrin and anchor alignment of the ␤-strand regions of Norrin with the TGF-␤/BMP sequences. Branching out from these cysteines, a conserved pattern of hydrophilic and hydrophobic residues is evident between Norrin and the TGF-␤ family members in conserved secondary structure elements, which enabled reliable alignment of these regions of Norrin and TGF-␤ sequences (see Fig.  6). This sequence alignment was then used to identify sites on the BMP2 structure homologous to positions in the Norrin amino acid sequence. Since no TGF-␤ family member stands out as more related to Norrin, BMP2 was chosen for modeling because there is a relatively high resolution structure (2.2 Å), and a structure of BMP2 complexed with both types of TGF-␤ receptor allows comparison of these sites with interaction sites identified on Norrin.

RESULTS
Norrin Binds Only to the Fz4 CRD-To assess Norrin and Wnt binding to various CRD targets, we employed AP fusion proteins as probes. For Xwnt8, a fusion with a Myc epitope and AP at the COOH terminus was produced in soluble form from stably transfected Drosophila S2 cells (10); for human Norrin, a fusion with AP and three Myc epitopes at the NH 2 terminus was produced in soluble form from transiently transfected 293 cells (Fig. 1A). Both AP fusion proteins were collected in medium containing 10% serum and, unless otherwise noted, were used for binding assays without further purification. Signaling assays were performed with a ␤-catenin-responsive luciferase reporter in stably transfected 293 cells transiently expressing Fz4, Lrp6, and human Norrin carrying a COOH-terminal rhodopsin tag (Fig. 1A).
For binding assays, the CRD targets were presented in two formats ( Fig. 1). In the first format, the CRD was displayed as a Myc-tagged and GPI-anchored protein on the surface of transfected COS cells, and binding was performed with live cells. Cell surface localization and accessibility of the CRD was confirmed for each CRD construct by incubating live cells with an anti-Myc monoclonal antibody. In this format, the assay measures binding in the context of plasma membrane lipids, glycoconjugates, and cell-associated extracellular matrix molecules. Cell surface binding was scored qualitatively based on the average AP intensity per cell (Fig. 1B). In the second format, the CRD was expressed as a fusion to the constant region of human IgG, which was then immobilized in protein A-coated microwells. Binding in this context occurs free of cell-associated molecules and was scored quantitatively (Fig. 1, C and D).
To systematically assess the specificity of Norrin-Fz4 binding, each of the 10 Frizzled and five sFRP CRDs encoded in the mouse genome was displayed on the surface of transfected COS cells and probed with anti-Myc, AP-3Myc-Norrin, or Xwnt8-Myc-AP. As seen in Fig. 2 and supplemental Table 1, all 15 CRDs accumulate at the cell surface, and four of them (from Fz4, Fz5, Fz7, and Fz8) efficiently bind Xwnt8-Myc-AP. In previous work, we had assayed the CRDs of Fz2-Fz8 for AP-Myc-Norrin binding and observed binding only to the Fz4 CRD (7). Here we extend this analysis to the complete set of mammalian CRDs and observe that only the Fz4 CRD shows detectable AP-3Myc-Norrin binding. A dendrogram showing the related- Norrin and Frizzled4 FEBRUARY 9, 2007 • VOLUME 282 • NUMBER 6 ness of the 15 mouse CRD sequences does not place the Fz4 CRD sequence on an outlying branch within this family, implying that relatively modest sequence differences between Fz4 and the other 14 CRDs are responsible for the ability of Norrin to uniquely recognize Fz4.
Identifying CRD Sequences Involved in Norrin Binding-To define sequences within the Fz4 CRD responsible for its selective recognition by Norrin, we first used the high resolution x-ray structure of the Fz8 CRD to predict which Fz4 CRD amino acids are likely to be exposed at the protein surface. Since the Fz4 and Fz8 CRD sequences align without insertion or deletion and are identical at 46 of the 114 positions, including the 10 conserved cysteines, it is likely that the ␣-carbon positions of the Fz4 CRD closely match those of the Fz8 CRD (Fig. 3B). By way of comparison, the Fz8 and sFRP3 CRDs are identical at 52 of 114 positions with two insertions/deletions, and their x-ray structures show nearly identical tertiary structures with a root mean square deviation of 0.97 Å for 101 of 122 ␣-carbons (11).
Three strategies were used to create Fz4 CRD sequence variants for binding assays. In the first, amino acids predicted to reside on the surface of the Fz4 CRD were mutated to alanine in 12 groups of 2-5 residues/group (Fz4 alanine scanning; Fig. 3, A, D, and E). In the second strategy, analogous block substitutions were generated, but in this case only those amino acids that differed between Fz4 and Fz8 and were predicted to reside on the surface were mutated; at each position, the Fz8 amino acid was inserted in place of its counterpart in Fz4 (Fz8 blocks into Fz4; Fig. 3, A and C). In the third strategy, nine chimeras between Fz4 and Fz8 CRDs were generated, such that each had only a single junction between Fz4 and Fz8 sequences (i.e. the NH 2 -terminal region was derived from Fz4 and the COOH-terminal region from Fz8 or the reverse; Fz4/Fz8 chimeras or Fz8/Fz4 chimeras, respectively; Fig. 3A). In all of these strategies, the cysteines were left unaltered. Because both Fz4 and Fz8 CRDs bind to Xwnt8-Myc-AP but only the Fz4 CRD binds to AP-3Myc-Norrin, we would predict that all correctly folded CRD constructs generated with the second and third strategies would retain Xwnt8-Myc-AP binding, and, among these, differences in AP-3Myc-Norrin binding would reflect sequence differences between Fz4 and Fz8 that are relevant to the specific recognition of Fz4 by Norrin.
CRDs generated with each of the three strategies were expressed as Myc-GPI-anchored proteins and tested for binding to anti-Myc, Xwnt8-Myc-AP, and AP-3Myc-Norrin (Table  1 and Fig. 3E). All of the CRD constructs accumulate at the cell surface, and approximately half show binding to both Xwnt8-Myc-AP and AP-3Myc-Norrin that is comparable with the WT Fz4 CRD. For this subset of mutants, we can conclude that the CRD tertiary structure is largely unperturbed and that the mutated surface region is unlikely to be critically involved in AP-3Myc-Norrin or Xwnt8-Myc-AP binding. For the remaining CRD mutants that are variably defective in binding, the defects in Xwnt8-Myc-AP and AP-3Myc-Norrin binding are closely correlated (Table 1 and Fig. 3E). Although these data cannot distinguish between binding defects due to a distortion of tertiary structure and those due to defects in residues that have direct contact with the ligand, the surface locations of the alanine scanning and Fz8 block substitutions argue for the latter as the more likely explanation. We infer, therefore, that AP-3Myc-Norrin and Xwnt8-Myc-AP bind to largely overlapping sites on the CRD. The surface regions defined here for Xwnt8-Myc-AP binding are in good agreement with the regions defined by earlier mutagenesis studies using the Fz8 and Drosophila Fz2 CRDs (10,11).
As a technical point, we note that making chimeras with a single junction between the Fz4 and Fz8 CRDs (Fig. 3A), a commonly used approach that does not rely on structural information to define surface residues for mutagenesis, was minimally informative, with only 2 of 9 chimeras retaining Xwnt8-Myc-AP binding (Table 1). Presumably, packing and/or other interactions between amino acid side chains is sufficiently altered in the remaining seven chimeras that Xwnt8-Myc-AP binding is disrupted.
Although none of the Fz8 block substitution mutants shown in Table 1 exhibited an all-or-none difference in Xwnt8-Myc-AP versus AP-3Myc-Norrin binding, block substitution G showed a clear reduction in AP-3Myc-Norrin binding with little or no effect on Xwnt8-Myc-AP binding. Inferring that the region encompassed by block G might contain a subset of the residues involved in the Fz8 versus Fz4 difference in AP-3Myc-Norrin affinity, we constructed a complete set of double Fz8 block substitutions using all pairwise combinations with block G (i.e. blocks A and G, blocks B and G, blocks C and G, etc.) and tested each for binding to anti-Myc, Xwnt8-AP, and AP-3Myc-Norrin. As seen in Table 2 and Fig. 4, A and B, substitution of blocks B and G or of blocks G and H reduced AP-3Myc-Norrin binding to background levels in both the cellbased and the CRD-IgG binding assays. By contrast, although Xwnt8-Myc-AP binding was reduced, it was still readily detect-   (11) is shown below the alignment. The 12-alanine scan and 12 Fz8 block substitutions are color-coded and indicated below the CRD sequence alignments. The locations of cross-over points for nine Fz4:Fz8 chimeric CRDs are shown below the block substitutions, with the four amino acids straddling the single cross-over point of each chimera indicated above the divergent arrows. None of the constructs altered any cysteine residues or introduced any insertions or a deletions. B, the backbone of the atomic resolution Fz8 CRD structure (11) in two views that differ by a 180°rotation about a vertical axis. The amino terminus (N) of the CRD is seen at the top; the COOH terminus is seen toward the right in the front view. Disulfides are shown in yellow, ␣ helices in blue, and ␤-sheet in red. Each of the surface images of the CRD shown in Figs. 3-5 represents this pair of views. The CRD faces on the left and right of the panel are arbitrarily designated as the "front" and "back" of the CRD, respectively. To avoid uncertainties associated with modeling the Fz4 side chains onto the Fz8 CRD structure, the surface images are of the Fz8 CRD with the numerically correct residues colored. Fz4 and Fz8 CRDs align without insertion or deletion, but differences between Fz4 and Fz8 CRDs with respect to the exact location and surface exposure of side chains are unknown. C, the locations of Fz8 block substitutions, on the CRD surface; substitutions are color-coded and labeled as in A. D, the locations of alanine scanning substitutions on the CRD surface; substitutions are color-coded and labeled as in A. E, the effect of alanine scanning block substitutions in the Fz4 CRD on the binding of AP-3Myc-Norrin. Qualitative assessment of binding to Fz4-Myc-GPI fusion proteins displayed on living COS cells was performed as in Fig. 1B  able in both assays. In the CRD-IgG assay, Xwnt8-Myc-AP binding to block substitutions B and G and to G and H was reduced to ϳ25% and ϳ10%, respectively, of the level seen with WT Fz4 CRD-IgG (Fig. 4B). These blocks partially overlap the regions implicated in both Xwnt8-Myc-AP and AP-3Myc-Norrin binding based on the alanine scanning series (Figs. 3,  D and E, and 4C), and they either reside adjacent to or encompass, respectively, M105V and M157V, the two Fz4 CRD substitutions identified by Xu et al. (7) in families with familial exudative vitreoretinopathy.
In a separate analysis of the sequence requirements for Xwnt8-Myc-AP and AP-3Myc-Norrin binding to the Fz4 CRD, we examined the properties of a set of single amino acid substi-tution mutants in the Fz4 CRD in which each of the nine aspartate or glutamate residues, all of which are predicted to reside on the surface, were changed to arginine. This experiment was motivated by the observation that Norrin has a very high isolectric point (pI ϭ 8.8) and a high density of cationic amino acid side chains (17 arginine or lysine residues in the 109-residue mature polypeptide), suggesting the potential for ionic interac-

. Substitution of Fz8 blocks B and G or blocks G and H into the Fz4 CRD diminishes Xwnt8-Myc-AP binding far less than AP-3Myc-Norrin binding.
A, anti-Myc monoclonal antibody (left column; visualized with APconjugated anti-mouse secondary antibody), AP-3Myc-Norrin (center column), or Xwnt8-Myc-AP (right column) were incubated with live COS cells displaying the indicated Fz4 CRD-Myc-GPI fusion proteins, and the specifically bound AP activity was visualized with nitro blue tetrazolium/5-bromo-4-chloro-3-indolyl phosphate. Under these conditions, Xwnt8-Myc-AP and AP-3Myc-Norrin give roughly equal levels of bound AP with WT Fz4-CRD-Myc-GPI (Fig. 2). B, quantitation of AP-3Myc-Norrin and Xwnt8-Myc-AP binding to WT or the two block substitution mutants of Fz4-CRD-IgG immobilized in protein A-coated microwells as in Fig. 1C. C, locations of Fz8 block substitutions B, G, and H on the CRD surface; front and back views are shown, and the blocks are colored as in Fig. 3A.  tions between ligand and receptor. Interestingly, when these nine Fz4 CRD mutants were tested in the CRD-IgG binding assay, their effects on Xwnt8-Myc-AP and AP-3Myc-Norrin binding were nearly identical (Fig. 5). As noted above in the context of the block substitution and chimeric CRD mutants, some of these effects on ligand binding could arise from substitution-induced perturbations of the CRD structure. Despite this uncertainty, the near identity of effects on Xwnt8-Myc-AP and AP-3Myc-Norrin binding argues that if ionic interactions of the hypothesized type occur, then they are largely conserved between Norrin and Xwnt8. Identifying Norrin Sequences Involved in Fz4 CRD Binding and Canonical Signaling-As noted in the Introduction, Norrin has weak homology to the TGF-␤ family of ligands. In particular, it shares 1) the six cysteines that anchor the TGF-␤ family in the cystine knot fold (three disulfide bonds) and 2) the cysteine that mediates disulfide linkage of two monomers into the active dimer (Fig. 6A). By threading the Norrin sequence onto the x-ray structure of TGF-␤, Meitinger et al. (17) found that the four additional Norrin cysteines clustered into two pairs in the tertiary structure, suggesting that they would form two additional intrachain disulfides (Fig. 6A). Since this initial report, additional x-ray structures of TGF-␤ family members have been determined. In Fig. 6A, the Norrin sequence is shown aligned to four of these (TGF-␤2, TGF-␤3, BMP2, and BMP7), and in Fig. 6, C and D, the BMP2 ␣-carbon backbone is used as a scaffold for modeling the three-dimensional structure of Norrin. Because of the limited primary sequence homology between Norrin and the TGF-␤ family members (including BMP2), especially in those regions predicted to form loops, we have not attempted to refine this model further.

TABLE 1 Semiquantitative binding of Norrin and Xwnt8 AP fusion proteins to COS cells expressing Fz4 CRD-Myc-GPI constructs
Our mutational analysis of Norrin had three aims: 1) to explore the role of the predicted disulfide bonds in the structure and function of Norrin by substituting individual or pairwise combinations of cysteine(s) with alanine(s); 2) to define the contribution of positive charges (most of which are predicted to be on the surface) to Norrin binding and signaling by substituting glutamate in place of each arginine and lysine; and 3) to systematically map those regions of Norrin involved in binding and signaling by substituting blocks of alanine for all amino acids other than cysteine and glycine between the second residue and the third to last residue in the mature Norrin polypeptide. For each of these Norrin mutants, we measured the binding of the corresponding AP-3Myc-Norrin fusion protein to immobilized Fz4 CRD-IgG relative to that of WT AP-3Myc-Norrin. Each Norrin mutant was also assayed for canonical Wnt signaling activity in a reporter cell line in the presence of co-expressed Fz4 and Lrp6. In Fig. 6B, the binding affinities and signaling activities are displayed as a fraction of the value for the WT Norrin control. Below, each of the three sets of mutants will be described in turn.
The collection of 11 single cysteine-to-alanine substitution mutants reveals a striking relationship between the severity of the binding and/or signaling defect and the evolutionary conservation of the different cysteines with the TGF-␤ family; those cysteines that form the three evolutionarily conserved intrachain disulfides (Cys 39 , Cys 65 , Cys 69 , Cys 96 , Cys 126 , and Cys 128 ) are the only ones for which alanine substitution severely compromises Fz4 CRD-IgG binding and signaling. By contrast, CRD binding and/or signaling are generally either not decreased or are decreased to only a modest extent by mutation of the Norrin-specific cysteines (Cys 55 , Cys 93 , Cys 110 , and Cys 131 ) or Cys 95 , the cysteine involved in dimerization. In particular, mutation of either Cys 93 or Cys 131 , which are predicted to form a Norrin-specific disulfide bond, appears to have little deleterious effect. A partial exception to this pattern is seen with C55A, which exhibits substantially reduced binding but retains ϳ40% of WT Norrin signaling activity.
If the hypothesized Norrin-specific disulfide bonds between Cys 55 and Cys 110 and between Cys 93 and Cys 131 exist, then one might predict that, within a given disulfide-bonded pair, mutation of both cysteines would cause less of a disruption to protein structure and function than would mutation of a single cys-FIGURE 5. Effect of Fz4 CRD aspartate-to-arginine or glutamate-to-arginine substitutions on Xwnt8-Myc-AP and AP-3Myc-Norrin binding. A, the indicated Fz4 CRD-IgG mutants were immobilized in protein A microwells and probed with each of the AP fusions as in Fig. 1C. Probing a third set of wells with anti-human IgG revealed variation in the amount of immobilized CRD-IgG of less than Ϯ15% among the different mutants. B, locations of the substitutions and summary of binding data. Red, little or no binding (D74R, E76R, E108R, and E129R); yellow, intermediate binding (D46R and D153R); green, binding comparable with or greater than WT (E70R, E134R, and D141R). Since residue 76 in Fz8 is glycine and is not surface-exposed, the Fz8 CRD surface model does not show position 76. FEBRUARY 9, 2007 • VOLUME 282 • NUMBER 6 JOURNAL OF BIOLOGICAL CHEMISTRY 4063 teine, because in the latter case, the free thiol could potentially participate in aberrant disulfide bond formation. To test this idea, we generated double mutants C55A/C110A, C93A/ C131A, and C110A/C131A. Double mutant C55A/C110A appears to bear out the prediction by exhibiting greater binding and signaling activity than either C55A or C110A alone; double mutant C93A/C131A has roughly the same activity as C93A or C131A alone (which, as noted above, are within a factor of 2 of WT Norrin activity); and the double mutant C110A/C131A, predicted to affect cysteines involved in two different disulfides, is at least as defective as either C110A or C131A alone. Surprisingly, mutation of Cys 95 (the cysteine predicted to mediate dimerization) produces less than a 2-fold decrement in signaling and a ϳ2-fold increase in CRD binding, and simultaneous mutation of Cys 95 and Cys 131 produces a ϳ5-fold increase in CRD binding. At present, these two observations are unexplained.

Norrin and Frizzled4
The mutational analysis of the lysine and arginine residues of Norrin reveals two residues that appear to be important for Fz4 CRD-IgG binding: Arg 41 and Arg 64 (Fig. 6, A-C). Curiously, the ϳ10-fold reduction in CRD binding exhibited by the R64E mutant is accompanied by canonical signaling activity that is slightly higher than the WT control, indicating that this mutant interacts with Fz4 and/or Lrp6 to activate signaling in a manner that compensates for a lower CRD binding affinity. A second interesting feature of the lysine and arginine mutants is the large number that show enhanced CRD binding unaccompanied by enhanced signaling. Taken together, these mutants suggest that binding between Norrin and the Fz4 CRD involves an ionic interaction involving multiple arginines and lysines and that this interaction is favored by the presence of one or more anionic groups (in this case, the glutamate introduced in place of lysine or arginine). As described below, these mutants may be partially mimicking the effect of heparin binding.
The 31 Norrin alanine scanning mutants show a generally good correlation between binding and signaling activities, with alanine scanning mutants 5,6,8,11,17,25, and 29 showing a 10-fold or greater decrement in binding and a greater than 3-fold decrement in canonical signaling activity (Fig. 6, A, B, and D). As noted above for the R64E mutant, several alanine scanning mutants (mutants 7 and 30 and, to a lesser extent, mutant 12) show a substantially greater decrement in CRD binding compared with signaling. Modeling the locations of the subset of alanine scanning mutants with a 4-fold or greater decrement in Fz4 CRD binding (shown in red in Fig. 6D) shows that most of them cluster along a broad surface on the outer face of the presumptive dimer. This region may directly contact the Fz4 CRD.
Heparin Is a Cofactor for Norrin-Fz4 CRD Binding-The high pI of Norrin and the substantial effects of arginine and lysine mutations on Norrin CRD binding and signaling activities suggest that polyanions, such as heparin or heparan sulfate, may play a role in Norrin-CRD interactions. To explore this idea, we asked whether AP-3Myc-Norrin could bind to heparin under physiologic conditions. We observed that AP-3Myc-Norrin (in conditioned medium with 10% calf serum) quantitatively binds to heparin-Sepharose in 0.15 M NaCl and quantitatively elutes in 1 M NaCl. Under the same conditions, AP alone does not bind to heparin-Sepharose. The AP-3Myc-Norrin that was eluted from heparin Sepharose (and was presumably purified away from soluble serum-derived heparin) was tested for Fz4 CRD-IgG binding (Fig. 7). Without any further additions, the AP-3Myc-Norrin binding activity was roughly 10% of the level observed for the same amount of AP-3Myc-Norrin in conditioned medium containing 10% serum. Interestingly, the binding activity of this heparin-purified AP-3Myc-Norrin could be fully restored upon the addition of 100 g/ml heparin, with a half-maximal effect at ϳ20 g/ml heparin. These data implicate heparin or related polyanions as important co-factors in Norrin-Fz4 binding.

DISCUSSION
The present work represents an initial step in defining the sequence determinants of Norrin-Fz4 CRD binding specificity.
The principal conclusions are as follows: 1) Norrin binds to the Fz4 CRD and does not detectably bind to the 14 other mammalian Fz and sFRP CRDs; 2) Norrin and Xwnt8 recognize largely overlapping regions of the Fz4 CRD; 3) surface determinants on the Fz4 and Fz8 CRDs that allow Norrin to distinguish between them reside within several small regions on one face of the CRD; 4) Norrin function depends critically on the three pairs of cysteines that form the highly conserved trio of disulfide bonds shared among all cystine knot proteins, but the other two putative disulfide bonds are less important; 5) Norrin-CRD binding depends on a largely contiguous group of amino acids in the extended ␤-sheet domain of Norrin that are predicted to face away from the dimer interface; 6) Norrin-CRD binding is strongly modulated by interactions involving charged amino acid side chains; and 7) Norrin-CRD binding is enhanced ϳ10fold by the addition of heparin. Below, we discuss these obser-FIGURE 6. Effect of Norrin substitution mutations on Fz4 CRD binding and Fz4/Lrp6 activation. A, top, the amino acid sequence of mature human Norrin, shown without the putative NH 2 -terminal signal sequence based on cleavage site prediction by SignalP 3.0 (59). The numbering is from the initiator methionine. Cysteines are colored yellow, and arginines and lysines are colored blue. Center, locations of the 31 alanine scanning block substitutions, designed to alter all residues between amino acids 27 and 130 except for cysteines and glycines. Bottom, alignment of Norrin with four TGF-␤ family members; conserved cysteines and their intra-and interchain disulfide bonds are colored purple and highlighted with purple asterisks; cysteines found in Norrin but absent from the TGF-␤ family are colored black along with their predicted disulfide arrangement (27,60). Norrin cysteines are numbered (bottom). B, AP-3Myc-Norrin binding to Fz4-CRD-IgG immobilized in protein A microwells (upper plot) and Norrin-rho activation of Fz4/Lrp6 (lower plot) for the indicated Norrin substitution mutants. Binding and luciferase activity are plotted relative to WT Norrin (set at 1.0). Left, single and double cysteine-to-alanine mutations; center, arginine-to-glutamate and lysine-to-glutamate substitutions; right, alanine scanning block substitutions. For most mutants, there is rough agreement between binding and activation assays. C, the Norrin backbone threaded onto the atomic resolution backbone of one BMP2 monomer (27,60); the predicted Norrin backbone positions for cysteines are colored yellow, and those for arginines and lysines are colored blue. Disulfide bonds conserved in the TGF-␤ family are shown; note that the most COOH-terminal Norrin cysteine is not shown, because the BMP2 polypeptide terminates prior to this position. D, the Norrin backbone threaded onto the BMP2 dimer backbone and color-coded on the left monomer to indicate those alanine scanning mutants with greater than (green) or less than (red) 25% of the binding activity of WT Norrin. Most of the substitutions that disrupt binding populate a largely contiguous surface that is solvent-exposed in the dimer.
vations in the larger context of Frizzled signaling, and the structure, function, and binding interactions of other cystine knot ligands.
Frizzled CRDs and Ligand Recognition-The presence of multiple Frizzleds and Wnts in all metazoa thus far examined, together with clear evidence for Wnt-Fz specificity (as seen in the 10-fold affinity difference of Drosophila Wingless for Drosophila Fz versus Fz2 (32) and the ability of Xwnt8 to distinguish between mammalian Fz and sFRP CRDs (10) (this study)), indicates that differential ligand-receptor affinities are an ancient feature of this signaling system. In this context, the extreme specificity of Norrin is unusual, and it raises the question of whether Frizzled receptors other than Fz4 might also have "private" ligands. At present, there is no data on either the secondary or tertiary structure of Wnt proteins, and therefore the Norrin-Fz4 system represents the only experimental system in which studies of ligand-Frizzled interactions can be guided by a three-dimensional ligand model.
It is interesting that Norrin and Xwnt8, which share no discernable sequence homology, appear to recognize largely overlapping regions on the Fz4 CRD. One possible explanation for this overlap might be that it reflects a requirement for a particular geometry in the ligand-CRD complex to activate signaling, for example, by interacting with the Lrp coreceptor and/or the extracellular loops of the membrane-embedded domain of Fz4. Other models are also possible; for example, the unliganded CRD might negatively regulate signaling, and ligand binding to a particular region of the CRD might relieve that inhibition. Both models would be consistent with the binding of different ligands to similar or identical regions of the CRD. An alternate model in which binding of the ligand to the Frizzled CRD merely functions to increase the local concentrations of ligand, which then interacts with the Frizzled-Lrp complex, is supported by the observation that a Wnt binding domain from WIF-1 (Wnt-inhibitory factor-1) (33), which is unrelated in sequence to the CRD, can functionally substitute for a Frizzled CRD and support Wnt-dependent signaling (34). If this alternate model applies to Norrin signaling, then it should also be possible to activate canonical signaling by artificially tethering Norrin to the receptor complex independently of the CRD.
Comparison of Norrin with Other Cystine Knot and Heparin Binding Proteins-The cystine knot structure is found in a wide variety of extracellular proteins, including many involved in signaling. These include the COOH-terminal globular domains of mucins, the large and diverse TGF-␤ family, nerve growth factor, platelet-derived growth factor, vascular endothelial growth factor, Noggin, prepro-von Willebrand factor, and the common ␣-subunit of the gonadotropin hormones (chorionic gonadotropin, leutinizing hormone, follicle-stimulating hormone, and thyroid-stimulating hormone). Several of these cystine knot proteins have additional cysteines within the knot domain that have been demonstrated crystallographically and/or chemically (e.g. the ␣ subunit of the gonadotropins) (35,36) or are inferred by mutagenesis (the mucins) (37) to form disulfide bonds. All of the cystine knot proteins, including Norrin (38), appear to function as either homo-or heterodimers, with most but not all dimers held together by a disulfide bond.
Within the context of the shared topological arrangement of three disulfide bonds that represents the defining feature of cystine knot proteins, there is wide variation in the spatial arrangement of monomers within the dimer and in the surfaces that interact with receptors or other binding proteins. For example, complexes between BMP2 and the BMP type IA receptor, between TGF-␤3 and the TGF-␤ type II receptor, and between activin A and the activin type IIB receptor reveal substantially different arrangements of ligand subunits and points of interaction with the receptors (30, 39 -41). Additional variations in protein-protein contacts are seen in the complex between a Noggin dimer and a BMP7 dimer, both cystine knot proteins (42). The present work implicates one face of Norrin in CRD binding, but a definitive analysis will require crystallization of the complex.
With respect to the disulfide-bonded structure of Norrin, the properties of single and double cysteine mutants described here are similar to the properties of an analogous set of mutants in the ␣ subunit of the gonadotropins (43,44). Despite the near absence of sequence homology between Norrin and gonadotropin ␣ (aside from the six cysteines that form the cystine knot), in both proteins, 1) mutation of the cystine knot cysteines leads to more severe defects than mutation of cysteines involved in either of two less conserved disulfide bonds, 2) mutation of both members of a disulfide-bonded cysteine pair can produce a more modest defect than mutation of only a single member, and 3) mutation of some amino acids (e.g. Cys 59 and Cys 87 in the ␣ subunit of gonadotropin and Cys 55 , Arg 64 , and several of the alanine scanning mutants in Norrin) paradoxically lead to a large reduction in receptor binding with little or no decrement in signaling.
Heparin and heparan sulfate binding are seen in a wide variety of extracellular ligands, including approximately one-fourth of the members of the 30 TGF-␤ family, many members of the fibroblast growth factor family, Drosophila Wingless, and the FIGURE 7. Heparin stimulates AP-3Myc-Norrin binding to Fz4 CRD-IgG. AP-3Myc-Norrin, purified from serum-containing conditioned medium by heparin-Sepharose affinity chromatography (to remove free heparin), was incubated with Fz4 CRD-IgG immobilized in protein A microwells in the presence of the indicated concentrations of heparin. 100 g/ml heparin produces a 10-fold increase in AP-3Myc-Norrin binding.
cystine knot BMP2 antagonist Noggin (45)(46)(47)(48). For Wingless and the Drosophila TGF-␤ family member Dpp (Decapentaplegic), heparan sulfate proteoglycans have been shown genetically to control the diffusion of the ligands in vivo (49 -53). For basic fibroblast growth factor, heparin plays a critical role in receptor binding by functioning as a co-ligand (47, 54 -57). A role for heparin as a co-ligand has recently been demonstrated biochemically and crystallographically in a second signaling system; high affinity binding of Hedgehog to its transmembrane binding partner Ihog requires the formation of a ternary complex with heparin (58). The observation that heparin strongly enhances Norrin-CRD binding indicates that Norrin-extracellular matrix interactions not only control the spatial localization of Norrin but also play an intimate part in the Norrin-Fz4 signaling complex.