Characterization of Tiki, a New Family of Wnt-specific Metalloproteases*

The Wnt family of secreted glycolipoproteins plays pivotal roles in development and human diseases. Tiki family proteins were identified as novel Wnt inhibitors that act by cleaving the Wnt amino-terminal region to inactivate specific Wnt ligands. Tiki represents a new metalloprotease family that is dependent on Mn2+/Co2+ but lacks known metalloprotease motifs. The Tiki extracellular domain shares homology with bacterial TraB/PrgY proteins, known for their roles in the inhibition of mating pheromones. The TIKI/TraB fold is predicted to be distantly related to structures of additional bacterial proteins and may use a core β-sheet within an α+β-fold to coordinate conserved residues for catalysis. In this study, using assays for Wnt3a cleavage and signaling inhibition, we performed mutagenesis analyses of human TIKI2 to examine the structural prediction and identify the active site residues. We also established an in vitro assay for TIKI2 protease activity using FRET peptide substrates derived from the cleavage motifs of Wnt3a and Xenopus wnt8 (Xwnt8). We further identified two pairs of potential disulfide bonds that reside outside the β-sheet catalytic core but likely assist the folding of the TIKI domain. Finally, we systematically analyzed TIKI2 cleavage of the 19 human WNT proteins, of which we identified 10 as potential TIKI2 substrates, revealing the hydrophobic nature of Tiki cleavage sites. Our study provides insights into the Tiki family of proteases and its Wnt substrates.

Wnt signaling plays pivotal roles in embryogenesis, homeostasis, and regeneration, and its dysregulation is implicated in many human diseases, notably cancer and metabolic disorders (1)(2)(3)(4). The human genome contains 19 WNT genes that encode cysteine-rich ligands with a unique lipidation-palmitoleoylation, essential for their biogenesis and engagement with the Frizzled receptors (4). Wnt ligands appear to exhibit poorly defined specificity for Frizzled and several different Wnt co-receptors (LRP5/6, ROR1/2, RYK, and PTK7), which in combinations largely dictate the recruitment of intracellular proteins and downstream Wnt signaling outputs (5,6). Wnt signaling is tightly regulated by various Wnt inhibitors (7). Most Wnt inhibitors, such as sFRP, DKK, WIF and SOST, antagonize Wnt signaling by binding to Wnt ligand or receptors and preventing Wnt-receptor interactions (7). We have recently identified a novel family of Wnt inhibitors, Tiki, named after the largeheaded Polynesian statues and in reference to the Tiki overexpression phenotype in the frog (8). Our study suggested that Tiki proteins act as membrane bound proteases to cleave the amino-terminal region of Wnt proteins, thereby inactivating them. Tiki and another hydrolytic enzyme, Notum, which acts to remove a lipid modification of Wnt proteins (9,10), represent a new mode of Wnt antagonism by modifications and inactivation of Wnt proteins.
Tiki is a metalloprotease that is inhibited by divalent metal chelators (EDTA or 1,10-phenanthroline) and shows a dependence on Mn 2ϩ or Co 2ϩ , whereas Ni 2ϩ , Cu 2ϩ , or Zn 2ϩ does not promote enzymatic activity (8). Tiki cleavage sites for mouse Wnt3a (major, SL 8 2AV; and minor, YS 15 2SL or SS 16 2LG; the number indicates the residue position in the mature Wnt3a protein) and Xwnt8 (LT 17 2YS and SA 19 2SV) were determined by mass spectrometry and amino-terminal sequencing, revealing a relatively degenerate (although possibly hydrophobic) sequence requirement without a clear consensus for the Tiki protease. Thus, Tiki is able to remove the first 8, 15, or 16 amino acids from Wnt3a and the first 17 or 19 amino acids from Xwnt8 (8). The 19 human WNT proteins share little sequence identity in the amino terminus prior to the first common conserved cysteine (11), confounding the definition of a consensus Tiki proteolytic cleavage site. Tiki is probably an endopeptidase given that the inclusion of an amino-terminal HA epitope tag in HA-Wnt3a does not interfere with Tiki specificity or cleavage position (8). We also observed Tiki-mediated amino-terminal cleavage of mouse Wnt5a but not Xwnt11, suggesting Tiki specificity toward Wnt proteins (8).
Most vertebrates contain two Tiki genes (TIKI1 and TIKI2 in human) with a single, clearly identifiable homologue in primitive metazoans, including the Nematostella sea anemone and the Amphimedon sponge. The core 330-amino acid TIKI domain is around 25% identical with the uncharacterized GumN proteins from Gram-negative bacteria (Pfam protein family PF07446). Subsequently Pfam GumN merged into the TraB family (Pfam PF01963), joined chiefly by the presence of two conserved GXXH motifs. In Gram-positive bacteria, the plasmid encoded TraB gene functions to antagonize mating pheromone-induced conjugation and self-induction (12). PrgY from Enterococcus faecalis is the best studied TraB homologue (12)(13)(14). Based on their kinship with Tiki, we proposed that TraB/PrgY proteins inactivate mating pheromones composed of short hydrophobic peptides via proteolytic cleavage (15).
Tiki and TraB/PrgY proteins share limited sequence identity, but their predicted secondary structures reveal a similar placement of ␤-strands and ␣-helices to anchor several potential conserved catalytic residues (15), consistent with a common evolutionary origin. To gain insights into the novel TIKI/TraB/ PrgY enzymatic domain, previously we used a sensitive fold recognition and structure prediction approach to build a threedimensional template for the TIKI domain based on distantly related proteins with known crystal structures (15). Using these techniques, we and others constructed a model for the TIKI fold modeling on structures of bacterial proteins from the EreA/ChaN-like superfamily (15,16). Six ␤-strands (␤A-␤F) in parallel/anti-parallel orientation formed the ␤-sheet core of our predicted TIKI ␣ϩ␤-fold, aided by four key ␣-helices (␣1, ␣6, ␣11, and ␣12), to bring together the conserved potential catalytic residues, including the two GXXH motifs, to one side of the threedimensional model. The remaining eight predicted ␣-helices, contributing to nearly half of the TIKI domain and containing several conserved cysteine residues, do not directly contribute to the putative "catalytic surface" but may have roles in the integrity of the TIKI fold or specialized functions for each of the Tiki/TraB/PrgY proteins (15). These analyses provide informative clues for further characterizing the Tiki family of metalloproteases.
In the current study, we used mutagenesis approaches to characterize Tiki enzymatic activity and specificity. We also established an in vitro assay to monitor Tiki proteolytic function using FRET peptide substrates. Finally, we analyzed Tiki specificity toward all 19 human Wnt proteins.

Experimental Procedures
Plasmids-HA-Wnt3a, TIKI2N-FLAG-His (TIKI2 extracellular domain (amino acid residues 1-494) fused with a FLAG tag and a His 6 tag), and KRM2N-FLAG-His (Kremen2 extracellular domain (amino acid residues 1-363) fused with a FLAG tag and a His 6 tag) expressing constructs were described previously (8). TIKI2N mutants were generated via a PCR-based sitedirected mutagenesis method using a TIKI2N-FLAG-His construct as the template. The HA-and V5-tagged human WNTs (HA-WNT-V5) were generated by introducing a HA tag after the predicted signal peptide of each WNT that already had a carboxylterminal V5 tag in the pcDNA3.2/V5-DEST vector (11).
Immunoblotting and Antibodies-Immunoblotting were performed as described previously (8 Protein Purification and FRET Assay-Secreted KRM2N, TIKI2N, and TIKI2N mutants were purified from the conditioned medium of transfected HEK293T cells following the protocol described elsewhere (8). Wnt3a and Xwnt8 FRET peptides were synthesized by Cambridge Research Biochemicals, Ltd. The control peptide, 520 MMP FRET substrate II, was from Anaspec Inc. (catalog No. 60569-01). For the FRET assay, purified TIKI2N or KRM2N (a series of concentrations or 1 g/ml) from the conditioned medium was incubated with FRET peptide (100 ng/ml final concentration) in PBS (50 l) for 2 h at room temperature. The fluorescent signals (excitation 488 or 520 nm) were measured by a 96-well plate reader.
TIKI2 Enzyme Kinetics Measurement-A series of concentrations (in 1:2 serial dilutions with a starting concentration of 165 nM) of Wnt3a FRET peptide was incubated with excessive amount of TIKI2N (700 nM) in 50 l of PBS for 2 h at 37°C, and the fluorescent signals were measured. A standard curve for fluorescent signals (the blank background was subtracted) versus Wnt3a FRET peptide concentrations was plotted. For the enzyme kinetics assay, a series of concentrations of Wnt3a FRET peptide (in 1:2 serial dilutions with a start concentration of 1.65 M) was incubated with TIKI2N (138 nM) or control buffer (serving as negative controls for background measurements) in 50 l of PBS for 30 min at 37°C, and the fluorescent signals were measured. The net fluorescent signal (the background signal was subtracted) for each substrate concentration was used in the standard curve to calculate the amount of FRET peptide cleaved by TIKI2N. The velocity (nM/min) at each substrate concentration was calculated. The velocity versus substrate concentration curve was plotted, and K m , K cat , and K cat /K m values were calculated using Prism 6 (GraphPad Software, Inc., San Diego) and a Michaelis-Menten model.

Results
Characterization of Tiki Enzymatic Active Site-Our previous results suggest that Tiki is a Mn 2ϩ -or Co 2ϩ -dependent metalloprotease (8). However, the sequence and predicted structure of the Tiki protein do not match any of the previously known metalloproteases (8). Our early interrogation of the potential Tiki active site focused on the two GXXH motifs, which were universally conserved in all Tiki and bacterial GumN proteins. GXXH motifs were initially classified as an extension of the Zinc-dependent HEXXH motif found in metzincin enzymes (HEXXHXXGXXH) (17). As Tiki proteins lack HEXXH motifs and are not zinc-dependent metalloproteases, the GXXH motifs of Tiki are distinct in context and function. Using the secreted form of the TIKI2 extracellular domain (TIKI2N-FLAG), which is necessary and sufficient for the Tiki protease function, we first mutated the histidines (His-58 and His-331) to alanines in the GXXH motifs plus an additional histidine, His-254, that is conserved down to nematodes (but not to bacteria) (Fig. 1A). We tested the TIKI2N enzymatic function by coexpression with an expression vector for HA-Wnt3a, which contains an amino-terminal HA and is active for Wnt signaling and suitable for Tiki-mediated cleavage as we demonstrated previously (8). TIKI2N cleaved the amino-terminal region of HA-Wnt3a as indicated by the HA removal and the faster mobility of the remaining Wnt3a protein detected by an anti-Wnt3a antibody (Fig. 1C). TIKI2N H58A and H331A mutants were not able to either remove the HA tag or change the Wnt3a mobility; by contrast TIKI2N H254A was able to cleave the HA tag and increase the mobility similar to TIKI2N (Fig. 1C). Therefore each of the histidines of the two GXXH motifs is essential for the protease activity, but His-254 is not (and thus serves as a control for other experiments). We further analyzed TIKI2N-mediated inactivation of Wnt signaling using the SuperTopFlash Wnt reporter. Compared with the high activation levels from Wnt3a alone, coexpression of Wnt3a and TIKI2N resulted in a dramatic reduction of Wnt signaling (Fig. 1D). Consistent with HA-Wnt3a cleavage results, TIKI2N H58A and H331A mutants were unable to inhibit Wnt signaling, whereas the H254A mutant inhibited Wnt3a in a manner similar to TIKI2N (Fig. 1D). As a result of our functional characterization, we refer to His-58 and His-331 as the H1 and H2 3 positions, respectively, in the context of the TIKI domain (Fig. 1, A and B). Using structure-guided searches for distant homologies to the TIKI domain, we and others have identified bacterial ␣ϩ␤fold protein structures from the EreA/ChaN-like superfamily: EreA-like Bacillus cereus proteins BcR135 and BcR136 (erythromycin esterase, Pfam PF05139, PDB codes 3B55 and 2QGM), ChaN (Campylobacter heme acquisition iron transporter, Pfam PF04187, PDB code 2G5G) and the C2 domain of PMT (Pasteurella multocida toxin, PDB code 2EBF) (15,16). Importantly, a single metal ligand (likely Ca 2ϩ ) was observed between the histidine residues analogous to H1 and H2 in the BcR135 structure (PDB code 3B55, deposited by the Northeast Structural Genomics Consortium) (Fig. 1B). Based on our predicted TIKI domain model, the GXXH motifs are located at the edge of two ␤-strands (H1 after ␤B and H2 after ␤E), which are parallel and adjacent to each other within the ␤-sheet core (15). At least four other residues (Glu-85 (E1), Glu-205 (E2), Arg (R), and Asn (N)) ( Fig. 1A) converge with the H1 and H2 residues in our predicted model, as shown in the context of the analogous active site residues in the BcR135 and ChaN structures (Fig. 1B). Two conserved glutamate residues at the end of ␤C (Glu-85 or E1) and ␤D (Glu-205 or E2) adopt a similar orientation to continue the ␤-sheet, with H1, H2, and E1 coordinating the metal ligand and E2 serving as catalytic residue to activate a water nucleophile (15). Tiki proteins also contain a highly conserved arginine-asparagine pair (Arg-303 and Asn-304 in TIKI2), analogous to known substrate stabilizing residues found in funnelin-type metalloproteases (17). Arg-303 and Asn-304 are located in a predicted extended ␣-helix that we juxtaposed between H2 and E2 in our model (15) (Fig. 1B). Therefore we generated additional mutants with alanine substitutions to test these potential metal-binding, catalytic, and substrate-stabilizing residues in our model: E1 (Glu-85), E2 (Glu-205), R (Arg-303), and N (Asn-304). A highly conserved threonine (Thr-56) preceding the H1 is present in Tiki/TraB proteins to form a consensus GTXH. This position, which is occupied by a glutamate residue in the distantly related bacterial proteins (GEXH in ChaN and PMT and GEXXH in EreA/BcR), appears to orient the H1 toward the metal-binding active site. The conserved TIKI2 Asp-187 within the predicted ␣-helix 6 is more difficult to model based on the existing structures. Alignments anchored by secondary structure point to acidic residues BcR135 Glu-145 and ChaN Asp-119 as possible analogous matches, although these residues are at the periphery of the presumptive active sites (Fig. 1B). Another candidate location for TIKI2 ASP-187 could be the aromatic residues BcR135 Tyr-141 and ChaN Trp-115, which orient toward the H1 and E1 residues (Fig. 1B). Therefore, at the moment it is unclear whether Asp-187 (D) contributes to the catalytic core or performs a role similar to TIKI2 Thr-56 in orienting the H1 and/or E1 residues. We note that in an alternate model of the Tiki active site, Asp-187 is shown in place of E2 (16). Given the importance of the H1 position, we also analyzed alanine mutations at the T (Thr-56) and D (Asp-D187) positions to examine whether they act as non-catalytic but active site stabilizing residues.
Similar to the TIKI2N H1 and H2 mutants, alanine mutations at E1, D, or E2 were unable to cleave HA-Wnt3a or change Wnt3a mobility (Fig. 1C). Mutations at the T, R, and N posi-tions resulted in detection of residual HA-tagged Wnt3a, with a majority of the Wnt3a proteins exhibiting faster mobility (Fig.  1C). Therefore, mutations at the H1, E1, D, E2, or H2 position each abolished TIKI2 protease activity, whereas mutations at T, R, or N reduced but did not abolish it.
We next examined the ability of these additional TIKI2N mutants to inhibit Wnt3a signaling (Fig. 1D). Mutation at E2, like those at H1 and H2, completely abolished TIKI2-mediated inhibition of Wnt3a signaling in the SuperTopFlash reporter assay (Fig. 1D). Mutation of E1 or D significantly compromised TIKI2N function in Wnt3a inhibition, whereas mutation at R had a moderate effect (Fig. 1D). On the other hand, mutation at T or N only mildly affected TIKI2N function, permitting significant inhibition of Wnt3a when compared with TIKI2N WT or the control H254A mutant (Fig. 1D).
Thus these data on Wnt cleavage and functional inhibition by Tiki support our structural modeling of the conserved TIKI catalytic domain (15), and identify key residues for the Tiki protease activity. Given that metalloproteases often use His, Glu, and Asp residues as metal ligands and use one Glu as the catalytic residue (17,18), these results are consistent with our earlier findings that Tiki family proteins are metalloproteases.
Monitoring Tiki Enzymatic Activity in Vitro Using FRET Peptide Substrates-To monitor Tiki activity in vitro, we designed a FRET peptide cleavage assay. We purified TIKI2N and mutant proteins from conditioned medium of transfected HEK293T cells using a FLAG tag-based affinity purification approach (8) (Fig. 2A). The Tiki cleavage sites in Wnt3a and Xwnt8 have been determined by mass spectrometry and Edman sequencing (8). We synthesized two peptides that cover the cleavage sites in Wnt3a (YPIWWS2LAVGPQY) and Xwnt8 (AYLT2 YS2ASVAVGAQ) and a Wnt3a mutant peptide (YPIWWD-DAVGPQY, underlined aspartate substitutions) that has an altered cleavage site and does not permit Tiki cleavage of Wnt3a (8). Each of these peptides was conjugated with an amino-terminal QSY-7 succinimidyl ester (quencher) group and a carboxyl-terminal Alexa Fluor 488 (AF488, fluorophore) group (Fig. 2B). The recombinant TIKI2N protein and the FRET peptides were used in the in vitro reaction. TIKI2N cleaved Wnt3a and Xwnt8 peptides at an EC 50 below 0.5 g/ml, whereas KRM2N, which corresponds to the extracellular domain of human Kremen2 and was used as a negative control (19), was unable to cleave either Wnt3a or Xwnt8 peptide (Fig. 2, C and  D). Neither TIKI2N nor KRM2N cleaved the Wnt3a mutant peptide (Fig. 2C). TIKI2N did not cleave a control FRET peptide that was a substrate for the metalloprotease MMP (Fig. 2E), further indicating Tiki enzymatic specificity in this in vitro assay. Next, we examined the TIKI2N mutants by using the Wnt3a FRET peptide (Fig. 2F). Consistent with the results using HA-Wnt3a as the substrate (Fig. 1B), TIKI2N and the H254A mutant cleaved the Wnt3a peptide with identical efficiency (Fig. 2F). Mutation at the H1, E1, D, E2, or H2 site resulted in complete loss of TIKI2 cleavage activity. The TIKI2N T mutant retained approximately half of the peptide cleavage activity. Somewhat surprisingly, the R mutant was almost completely inactive in the FRET peptide assay (Fig. 2F) whereas it had partial activity in HA-Wnt3a cleavage and Wnt3a inhibition assays (Fig. 1, C and D). On the other hand, the neighboring N mutant was largely active in these three assays (Figs. 1, C and D, and 2F). Overall, the activity data assessed by the in vitro Wnt3a peptide assay correlated strongly with the results of HA-Wnt3a cleavage and functional Wnt inhibition.
To assess the enzymatic efficiency of TIKI2 in Wnt cleavage, we determined the enzyme kinetics parameters using the  (20). Taken together, these results support the characterization of Tiki proteins as Wnt-specific metalloproteases.
Analysis of Conserved Cysteines in TIKI Domain-Secreted proteins often contain disulfide bonds that are important for protein folding and stability. Sequence alignment indicates that TIKI domain contains four conserved cysteines, which in human TIKI2 are Cys-29 (c1), Cys-100 (c2), Cys-211 (c3), and Cys-257 (c4). The first conserved cysteine is in the unstructured amino terminus, and the other three follow helices ␣2, ␣7, and ␣9, respectively (Fig. 4B). The cysteine-containing regions within the TIKI domain could not be modeled using templates from the bacterial protein EreA-like, ChaN, or PMT structure. To assess the role of these cysteines in Tiki enzymatic activity and potential disulfide bond formation, we generated a series of single and double cysteine mutants and coexpressed each of them with Wnt3a in HEK293T cells (Fig. 4A). Human TIKI2 contains an additional extracellular cysteine (Cys-323) at the beginning of ␤E (on the opposite side from the second GXXH motif), although this is not conserved among Tiki proteins or even in the frog Tiki2. To test the importance of this non-conserved cysteine we mutated human TIKI2 to resemble frog Tiki2, IC322SF (c5), and used it together with a TIKI2N inactive mutant, H58A/H331A, as control. TIKI2N and the c5 mutant efficiently cleaved Wnt3a and resulted in significant Wnt3a band shifts on Wnt3a Western blot as compared with Wnt3a coexpressed with the H58A/H331A mutant (Fig. 4A). The activity of the four conserved cysteine mutants was reduced to different extents (Fig. 4A). Among the double mutants, c1-c2, c1-c3, c2-c4, and c3-c4 were largely inactive, whereas c1-c4 and c2-c3 were almost fully active (Fig. 4A). These results are consistent with the possibility that there are two disulfide bonds, one formed by c1-c4 and one by c2-c3 (Fig. 4B). This also appears to be consistent with our predicted TIKI domain-folding topology, with ␤C and ␤D strands in close proximity to accommodate a c2-c3 disulfide bond and intercalation of ␤E closer to the amino-terminal ␤-strands to allow a c1-c4 disulfide bond (15). Although the Tiki ␣-helical bundles (␣2-␣5 and ␣7-␣10) share no similarity to the bacterial EreA/ChaN-like structures, ChaN contains analogous ␣-helical bundle insertions (Fig. 4C, insert 1 in green and insert 2 in wheat), which are spatially located above and around the presumed active site. Superimposing the locations of the Tiki cysteines on the ChaN structure illustrates the potential for these disulfide bond connections and provides further support for likely Tiki c1-c4 and c2-c3 disulfide bond organization (Fig. 4C). The two potential disulfide bonds may contribute to the stability of the TIKI domain structure or may assist in Wnt protein substrate recognition in the Tiki-specific ␣-helical regions.
Characterization of Tiki Specificity toward Wnt Proteins-We have shown that Tiki cleaves mouse Wnt3a, mouse Wnt5a, and Xwnt8 but not Xwnt11, suggesting that Tiki specifically cleaves some but not all Wnt proteins (8).
To further study Tiki specificity, we systematically examined TIKI2 cleavage of each of the 19 human WNT proteins. A HA tag and a V5 tag were fused to the amino and carboxyl termini, respectively, of each WNT, and each of these double-tagged WNT proteins was expressed in HEK293T cells alone or with TIKI2N. Whole cell extracts were analyzed by Western blotting with anti-HA and anti-V5 antibodies. WNT cleavage was assessed by the reduction of HA signal or by mobility shift of the V5 signal (Fig. 5A). The results indicate that TIKI2 cleaved WNT2B, -3, -3A, -4, -5A, -5B, -6, -8A, -8B, and -16 but not WNT1, -2, -7A, -7B, -9A, -9B, -10A, or -10B (Fig. 5B). Therefore TIKI2 appears to be able to cleave 10 of the 19 WNT proteins, and the cleavage occurs strictly at the WNT amino terminus.
Wnt proteins are conserved in amino acids throughout most of the molecules, but their amino-terminal sequences are variable among family members although conserved among paralogues (11). We noticed previously that the sequences flanking the Tiki cleavage sites in Wnt3a and Xwnt8 are very different (Fig. 2B) (8). Comparisons among Tiki-cleavable Wnt versus non-cleavable Wnt did not appear to yield a consensus cleavage sequence but seemed to reinforce the notion that the amino termini of Tiki-cleavable Wnt are relatively hydrophobic, whereas the amino termini of non-cleavable Wnt often contain more charged and polar residues (11). Consistent with this notion, Tiki did not cleave FLAG-Wnt3a (the FLAG tag DYK-DDDDK is at the amino terminus) (data not shown), the Wnt3a mutant that harbors four aspartate substitutions of the major and minor cleavage sites (SYPIWWDDAVGPQYSDDGS . . . ) (8), or Wnt3a mutant peptide (YPIWWDDAVGPQY) (Fig. 2, B  and C).
Indeed although WNT2 and WNT2B are closely related and their amino-terminal sequences are relatively conserved, TIKI2 cleaved WNT2B but not WNT2 (Fig. 5A). The amino-terminal sequence alignment of WNT2 and -2B indicates that two residues in WNT2, Arg-40 and Thr-42, make the WNT2 amino terminus more hydrophilic than that of WNT2B, in which the two corresponding residues are Gly and Leu, respectively (Fig.  5C). We hypothesized that these two residues might determine the Tiki cleavage specificity. We generated three mutants of WNT2, R40G, T42L, and R40G/T42L, in which one of these residues was changed to Gly and Leu, respectively, or both were changed to Gly and Leu together. We examined the TIKI2 cleavage of these WNT2 mutants by coexpressing each with TIKI2N. As shown in Fig. 5D, TIKI2N cleaved the WNT2 R40G/T42L mutant and also the WNT2 R40G mutant, albeit less effectively. These results suggest that the hydrophobicity of the Wnt amino terminus is a major determinant of Tiki cleavage susceptibility.

Discussion
Tiki proteins form a new family of Wnt antagonists and act as metalloproteases that cleaves the Wnt amino terminus (8). In this study, we analyzed potential enzymatic active sites of TIKI2 and characterized the Wnt substrate specificity of TIKI2. Our study establishes that the Tiki metalloprotease family has unique enzymatic characteristics, supporting its classification in MEROPS database as metallopeptidase family M96.
Tiki/TraB/PrgY family proteins do not harbor known metalloprotease motifs but contain two conserved GXXH motifs (8,15). Two previous studies used bioinformatics approaches to model the enzymatic fold of the TIKI domain (15,16). Both of those studies suggest that the TIKI domain is related to bacterial proteins from the EreA/ChaN-like superfamily, including the erythromycin esterase family members BcR135 and BcR136, the heme-binding ChaN iron transporter, and the uncharacterized C2 subdomain from PMT. Of these structures, a metal ligand was only found in the BcR135 presumptive active site (PDB code 3B55). No metal was present in the BcR136 structure (PDB code 2QGM), consistent with functional data showing that hydrolase activity has been observed for BcR136 and EreB but not EreA in the presence of chelators (21).  Wnt3a and TIKI2N or TIKI2N mutants were coexpressed in HEK293T cells, and the resulting conditioned medium (CM) and whole cell lysates were analyzed by Western blotting. The differential migration of Wnt3a indicates TIKI2N cleavage activity. Note that the H58A/H331A double mutant is functionally inactive as is each single mutant; the TIKI2-specific c5 (IC322SF) mutant is fully active; the single cysteine c1, c2, c3, and c4 mutants are partially active; and only the double cysteine c1-c4 and c2-c3 are active. B, predicted secondary structure of Tiki. ␤-Strands A-F make up the core ␤-sheet of the TIKI domain flanked by darkly shaded ␣-helices 1, 6, 11, and 12. Lightly shaded ␣-helices 2-5 and 7-10 within the TIKI domain are unplaced in the current simulated model for active sites. Located outside of the TIKI domain and omitted from the diagram are an additional predicted ␤-strand (␤G) and ␣-helix (␣13) of unassigned functions. Two potential disulfide bonds are indicated by connecting lines, and putative enzymatic active sites are also indicated. C, schematic model of ChaN (PDB code 2G5G, residues 9 -263) with corresponding TIKI2 residues analyzed in this study are highlighted according to descriptions in the legend for Fig. 1. ChaN ␣-helical insert 1 (green, residues 84 -113) and insert 2 (wheat, residues 140 -182) are analogous to the Tiki ␣2-␣5 and ␣7-␣10 inserts, respectively. The relative positions of the Tiki c1-c4 and c2-c3 disulfide bond connections are overlaid on the ChaN structure.
Although it is unclear whether EreB acts as a metalloprotease, EreB mutants for the corresponding active site positions H1 (H46A), E1 (E74A), and H2 (H288A) display loss of function in an erythromycin cleavage assay (21). Partial loss of function was also observed for EreB E43A mutant in the GEXXH motif, analogous to TIKI2 GTXH.
Based on the known structural information for this group of proteins, a TIKI domain topology was predicted in which the His residue in two GXXH motifs and conserved Glu, Asp, Arg, and Asn residues between the GXXH motifs are mapped to the putative enzymatic core. Although mutation at Thr-56 resulted in partial loss of function, this data effectively rules out Thr-56 as a candidate nucleophile. The results from the mutagenic analysis of TIKI2 suggest that both conserved His residues (H1 and H2) and the conserved Glu and Asp residues highlighted by the two predicted models (15,16) (Fig. 1, E1, D, and E2) are required for TIKI2 activity (Figs. 1 and 2). The only disagreement between these two predictions is whether a second Glu (E2, Glu-205 in TIKI2) or an Asp (D, Asp-187 in TIKI2) residue is an active site (15,16). Because we observed an equivalent loss of function for the D and E2 residues, our functional data do not favor one model over the other. However it does put in question how many metal ions are within the active site. These widely spaced residues in the TIKI domain may come together in the three-dimensional structure and coordinate one or two metal ions (Co 2ϩ or Mn 2ϩ ) (8) to form the catalytic core. Although this pattern of the enzymatic active site in the TIKI domain is unique among metalloprotease families, there is precedence for co-catalytic Co 2ϩ /Mn 2ϩ -dependent metalloproteases (18). A crystal structure of the TIKI domain will be critical to further define the enzymatic activity of Tiki family metalloproteases.
Our in vitro data for the TIKI2 mutants using the Wnt3a FRET peptide are largely consistent with cleavage of HA-Wnt3a and functional inactivation of Wnt signaling. The only exception was for the Arg (R303A) mutant that was largely dead in the FRET assay but remained partially active in the other two assays. Although completely conserved in Tiki homologues, the Arg/Asn residues are variable compared with that of TraB/PrgY (RD), ChaN (KD), EreA (RD), EreB (KD), and BcR (HD). Notably, TIKI2 Asn-304 is unique, and the lack of conservation may support a role for substrate recognition or stabilization of the active site. These requirements may be less stringent for a short Wnt3a FRET peptide than for the fulllength Wnt3a protein.
Bacterial TraB and PrgY are plasmid-encoded transmembrane proteins with extracellular TraB domains that are homologous to the TIKI domain (15). TraB/PrgY proteins function to reduce the amount of the autocrine pheromone by an unknown mechanism to prevent self-induction (12). Two possible models of TraB/PrgY function have been proposed: by binding the pheromone and delivering it to a cell wall protease for degradation or by acting as a protease and directly degrading bound pheromone (12). Our studies support the idea that TraB/PrgY proteins are metalloproteases that directly degrade bound pheromone. Consistent with this model, mutagenic analysis of PrgY has demonstrated that several conserved residues in the TraB domain are required for its function, including the H1, E1, D, E2, and R equivalents in TIKI2 (Fig. 1A) (13). Our study also suggests that the TraBD (TraB domain-containing) protein, which exists in most organisms in metazoan, may act as a metalloprotease. However, TraBD is unlikely to cleave Wnt pro- FIGURE 5. The specificity of Tiki for 19 human WNT proteins. A, each of the 19 human WNT with an HA and a V5 tag on each terminus was expressed alone or coexpressed with TIKI2N, and whole cell lysates were analyzed by Western blotting with the indicated antibodies. Reduction of the HA signal and a mobility shift of the V5 signal in the presence of TIKI2N expression indicate an amino-terminal cleavage of the WNT protein by TIKI2N. B, summary of Tiki-cleavable and non-cleavable WNT. C, sequence alignment of the amino terminus of WNT2 and WNT2B. Arrows indicate WNT2 (non-cleavable) Arg-40 and Thr-42 residues that were altered to corresponding WNT2B (cleavable) residues. D, WNT2B, WNT2, and WNT2 mutants were each expressed alone or with TIKI2N in HEK293T cells, and whole cell lysates were analyzed by Western blotting. teins, as it lacks a predicted signal peptide and likely functions as a cytoplasmic protein.
We reported previously that Tiki cleaves mouse Wnt3a, mouse Wnt5a, and Xwnt8 but not Xwnt11 (8). We have systematically analyzed TIKI2 cleavage of the 19 human WNT proteins. Based on the Tiki cleavage susceptibility, we have classified WNT proteins into Tiki-cleavable and non-cleavable groups. TIKI2 cleavage does not appear to follow a stringent consensus sequence, except that the sequences that flank cleavage sites are relatively hydrophobic. It is interesting to note that all known bacterial sex pheromones, which are putative TraB/ PrgY substrates, are hydrophobic hepta-or octapeptides, such as LVTLVFV and LFSLVLAG (22). We also note that in addition to the 13-residue FRET Wnt3a peptide (YPIWWS-2LAVGPQY), shorter FRET Wnt3a peptides, IWWS-2LAVGP and WWS2LAVG, also displayed similar TIKI2N cleavage (data not shown), suggesting that TIKI, like TraB/ PrgY, is capable of cleaving short hydrophobic peptides analogous to bacterial pheromones. However, TIKI2 cleavage of the 10 cleavable WNT proteins is strictly at the amino terminus, as we noted previously (Fig. 5A) (8). We have shown that TIKI2 binds to mouse Wnt3a, Wnt5a, Xwnt8, and Xwnt11, but it cleaves Wnt3a, Wnt5a, and Xwnt8 but not Xwnt11 (8). These results taken together suggest two steps of Tiki cleavage of Wnt proteins: Tiki interacts with a conserved region within Wnt proteins, thereby possibly orienting the Wnt amino terminus for Tiki cleavage, which requires the hydrophobic nature of the cleavable amino terminus. Our results will guide future studies on the regulation of Wnt functions by the Tiki metalloprotease family.