Structural Analysis of Siah1-Siah-interacting Protein Interactions and Insights into the Assembly of an E3 Ligase Multiprotein Complex*

Siah1 is the central component of a multiprotein E3 ubiquitin ligase complex that targets β-catenin for destruction in response to p53 activation. The E3 complex comprises, in addition to Siah1, Siah-interacting protein (SIP), the adaptor protein Skp1, and the F-box protein Ebi. Here we show that SIP engages Siah1 by means of two elements, both of which are required for mediating β-catenin destruction in cells. An N-terminal dimerization domain of SIP sits across the saddle-shaped upper surface of Siah1, with two extended legs packing against the sides of Siah1 by means of a consensus PXAXVXP motif that is common to a family of Siah-binding proteins. The C-terminal domain of SIP, which binds to Skp1, protrudes from the lower surface of Siah1, and we propose that this surface provides the scaffold for bringing substrate and the E2 enzyme into apposition in the functional complex.

Recently, we discovered a novel pathway for ␤-catenin degradation involving a complex formed by Siah1, SIP, the adaptor protein Skp1 that is common to the SCF complex, and the F-box protein Ebi that binds ␤-catenin independent of phosphorylation (31). Siah1 expression is upregulated by p53, revealing a link between genotoxic injury and destruction of ␤-catenin, reduced Tcf/LEF activity, and cell cycle arrest (31). Siah1 is a dimeric protein that contains an N-terminal RING domain (an E2 binding domain) followed by two zinc finger motifs and a C-terminal dimerization domain. The crystal structure of a major fragment of Siah1a lacking the RING domain (Siah1⌬R) has been determined (32); in this structure, the first zinc finger is highly mobile, whereas the second packs tightly against the C-terminal domain, forming a dimeric substrate-binding domain. In the multiprotein E3 complex, Siah1 plays an analogous role to the cullin-1 and Rbx1 domains in the classic SCF complex, and Siah-interacting protein (SIP) provides a novel link between Siah1 and Skp1 (31). We previously showed that SIP binds to Siah1 via determinants in its N-terminal domains (residues 1-72), whereas its C terminus (residues 73-228) binds to Skp1 (31). The murine ortholog of SIP (which is 93% identical) was initially characterized as a calcyclin-binding protein (33,34). Calcyclin binding does not appear to compete with Skp1 binding, and its functional role has not been determined.
A consensus motif for Siah-binding proteins, comprising the sequence PXAXVXP, has been identified (35). This motif is present in 15 of 27 known, functionally diverse, Siah-binding proteins, some of which are targets of degradation. The motif is also present in SIP. Here we present a structural and functional characterization of the interaction between human Siah1 and SIP. Our work demonstrates how SIP interacts with Siah1 in the context of the multiprotein complex. It also provides the first structure of Siah1 bound to the consensus PXAXVXP motif, providing a basis for probing a broad range of Siah functions.  (TABLE I) was collected at 100 K at the SSRL facility beamline 9.1 and processed with the HKL package (36). The structure was solved by multiwavelength anomalous dispersion using a crystal soaked for 30 s in cryoprotectant solution saturated with NaBr. A fourwavelength multiwavelength anomalous dispersion data set (TABLE I) was collected at beamline 9.2 to a resolution of 1.5 Å. Initial phases were obtained with SOLVE (37), and the density was improved with ACORN (38) and DM (39). A main-chain model was built with ARP/wARP (40). Manual model adjustments using O (41) and refinement with Refmac5 (42) to R work ϭ 17.4, R free ϭ 19.8 yielded the final model (TABLE I). The model contains three polypeptide chains per asymmetric unit; chains A and B form a noncrystallographic dimer, whereas C forms a dimer across a crystallographic dyad. The model includes residues 1-43 for chains A and C and 1-47 for chain B, 230 water molecules, one SO 4 2Ϫ ion, one CAPS molecule, and one or two nonnative N-terminal residues. The stereochemistry is excellent as judged by PROCHECK (43). Coordinates and structure factors of SIP-(1-47) have been deposited in the Protein Data Bank with code 2A26.

EXPERIMENTAL PROCEDURES
Crystallization and Structure Solution of Siah1⌬R/SIP-(58 -70)-Synthetic peptides (DNEKPAAVVAP-NH 2 and EKPAAVVAPITTG-NH 2 ) were purchased at 95% purity from the Louisiana State University Health Sciences Center Core Laboratories. Siah1-(90 -282) was mixed at 6 mg/ml in 15 mM Tris-HCl, pH 8.0, 30 mM NaCl, 10 mM dithiothreitol, with 1 mM peptide and crystallized by sitting drop vapor diffusion. The reservoir solution was 1.4 M NaH 2 PO 4 /K 2 HPO 4 in a 9:1 to 8:2 ratio, 10 mM dithiothreitol. Crystals appeared in 3-5 days, growing to a typical size of 150 ϫ 150 ϫ 400 m 3 . They adopt space group C222 1 with unit cell dimensions a ϭ 47.7, b ϭ 107.7, c ϭ 78.3 Å. Crystals were flashcooled in liquid nitrogen using the crystallization solution with 20% glycerol. A complete data set to 2.2 Å resolution (TABLE I) was collected on a Rigaku FR-E generator equipped with Osmic optics and an R-axis IV image plate detector and processed with the HKL package (36). The structure was solved by molecular replacement using residues 150 -282 of a Siah1 monomeric subunit (Protein Data Bank code 1k2f) as a search model using MOLREP (39). Refinement was carried out using Refmac5 (42) with simulated annealing in CNS (44) and model building in O (41). Electron density for seven residues in the peptide (PAAVVAP) was clearly visible in 2 F o Ϫ F c maps in the early stages of the refinement, and the refined model is fully consistent with density in a 5000°C simulated annealing omit map. The final model (R work ϭ 0.215, R free ϭ 0.254; see TABLE I) consists of Siah1 residues 125-282, SIP residues 59 -67 (KPAAVVAPI), one Zn 2ϩ ion, and 100 water molecules. Electron density for residues in turns 132-133, 172-175, and 197-202 was weak and is not included in the model. No electron density was observed for residues 90 -124. Residues 162-282 superpose with the corresponding residues in the Siah1 probe model with a root mean square deviation (C␣) of 0.46 Å. All stereochemical parameters are excellent as analyzed with PROCHECK (43). Coordinates and structure factors of Siah1⌬R/SIP-(58 -70) have been deposited in the Protein Data Bank with code 2A25.
NMR Measurements-Sequential assignments of SIP-(1-47) were obtained with standard techniques using appropriately labeled SIP-(1-47) at 0.5-2 mM and pH 5.1. All two-and three-dimensional spectra used for resonance assignments were recorded on a Varian UNITY PLUS 500-MHz spectrometer and processed and analyzed using FELIX (Accelrys Inc.). Sequential resonances for residues 4 -47 were unambiguously assigned and have been deposited in the BioMagnRes data bank. All other spectra were collected at 298 K on a Bruker Avance 600-MHz spectrometer equipped with a 5-mm triple resonance probe and z axis pulsed field gradients. 1 H, 15 N HSQC and transverse relaxation optimized spectroscopy (TROSY) spectra were recorded with 2048 (H N )*128 or *256 ( 15 N) data points and a total of 4 or 16 transients per t 1 increment. Proton and nitrogen spectral widths were set to 12 and 26 ppm, respectively. Spectra were recorded on a sample of 500 M uniformly 15 N/ 2 H-labeled SIP-S in 100 mM NaCl, 20 mM phosphate buffer H 2 O/D 2 O (90:10) pH 6.7, 10 M Zn(OAc) 2 . 13 C HMQC experiments were acquired in the same buffer on a sample of 13 C ␥ , 1 H ␥ -Thr, 13 C ␦ 1, 1 H ␦ 1-Ile, 2 H-labeled SIP-S (200 M) in the presence of increasing amounts of unlabeled Siah1⌬R protein. 13 C, 1 H HMQC spectra were recorded as 8192*150 data points with 8 transients per t 1 increment; proton and carbon spectral widths were set to 12 and 20 ppm, respectively. Two-dimensional NOESY spectra were acquired with mixing times of 150 and 200 ms (spectral width 12 ppm along both f 1 and f 2 , 4096*320 data points in t 2 and t 1 , respectively, and 80 scans per t 1 increment). Water suppression was achieved by the WATERGATE PFG technique. All spectra were processed with Bruker software (Xwinnmr 3.5) and analyzed with XEASY version 1.3.13 (45) and Mestrec (available on the World Wide Web at qobrue.usc.es/). For K d evaluation, normalized intensities from ] 13 C, 1 H HMQC spectra were plotted as a function of Siah1 concentration (10, 50, 100, 150, 200, 300, and 400 M). Normalized intensities were evaluated according to the relationship I free Ϫ I obs /I sat Ϫ I free , where I obs is the measured value of the intensity, I free is the value in the absence of Siah1, and I sat is the value at saturation (400 M Siah1). Intensities were evaluated as average values of the observed peaks with the errors set to the S.D., using XEASY (45). Data were fitted to the equation, where B represents the concentration of SIP-S (200 M), and A is the K d obtained from the fit.
Docking-AutoDock version 3.0.5 (46) was used for the docking simulation. We used the Lamarckian genetic algorithm for SIP-(1-47) positional searching. The protonation states of Siah1 and SIP were set to pH 7.4. SIP-(1-47) domain was kept as a rigid body. The docking parameters were as follows: trials of 100 dockings, population size of 150, random starting position and conformation, translation step ranges of 1.5 Å, rotation step ranges of 35°, elitism of 1, mutation rate of 0.02, crossover rate of 0.8, local search rate of 0.06, and 10 million energy evaluations. Final docked conformations were clustered using a tolerance of 1.5 Å root mean square deviation.
Plasmids, Transfections, and Cell Culture-Mutations in SIP were generated by two-step PCR-based mutagenesis using a full-length human SIP cDNA (31) as a template. Products were purified by Qia-Quick gel extraction kit (Qiagen), digested with EcoRI and XhoI, and then directly subcloned into the EcoRI and XhoI sites of pcDNA3 plasmid (Invitrogen) with an N-terminal Myc epitope tag (MEQKLI-SEEDL), thus creating pcDNA3-Myc. Alternatively, the cDNAs were subcloned into yeast two-hybrid plasmids pGilda and pJG4 -5, which produce fusion proteins with a LexA DNA-binding domain or a B42 transcription activation domain, respectively, at the N terminus, under the control of a GAL1 promoter. HEK293T cells were maintained in high glucose Dulbecco's modified Eagle's medium containing 10% fetal calf serum, 1 mM L-glutamine, and antibiotics. For transient transfections, cells (ϳ5 ϫ 10 5 ) in 6-well plates were transfected with plasmid DNAs using Lipofectamine Plus (Invitrogen).
Isothermal Titration Calorimetry-SIP-S and Siah1⌬R were extensively dialyzed against a buffer containing 100 mM NaCl, 20 mM Tris-HCl, pH 7.5, 10 M Zn(OAc) 2 with or without 10 mM MgCl 2 and concentrated to 1.20 -2.90 mM and 100 -115 M, respectively. 7-15-l aliquots of SIP-S were added to the initial 2 ml of Siah1 in the cell of a Microcal (Northampton, MA) VP-ITC. Values for K d , ⌬H, and ⌬S were calculated using the Microcal Origin software provided by the manufacturer. For each titration point, the heat of reaction was plotted against the molar ratio, and the data were fitted as described (47). The SIP-S concentration and SIP-S/Siah1 molar ratio at the end point of the titration were at least 210 M and 4.2, respectively. Each measurement was performed independently in duplicate and was reproducible within a 4% error margin (K d ).

SIP-S Comprises an N-terminal Dimer with a Pair of Widely Separated Flexible Tails-
The gene for human SIP encodes two alternatively spliced proteins 228 and 80 amino acids long, called SIP and SIP-S (S for short) (31). SIP-S is identical to SIP in its first 72 amino acids (Fig. 1A). Using NMR techniques, we probed the structure of SIP-S and obtained sequential assignments for residues 4 -47. The chemical shifts and the NOE patterns are consistent with a folded domain with ␣-helical structure. The remaining residues of SIP-S were not assigned, due to over-  :10), pH 6.7, and 0.05% NaN 3 ) in the absence (black) and presence (red) of 750 M unlabeled Siah1⌬R. Peaks that move and/or disappear after the addition of Siah1 are indicated by sequence numbers or by arrows for residues in the SIP tail. The asterisks indicate nonnative residues at the N terminus from the protein construct. F, one-dimensional 1 H traces from two-dimensional 13 C, 1 H HMQC spectra recorded with 13 C-1 H␥-Thr, 13  lapping resonances at typical random coil values and the lack of sequential NOEs, suggesting that the tail is disordered.
Attempts to crystallize full-length SIP or SIP-S were unsuccessful. Limited proteolysis identified a protease-resistant core (residues 4 -43), consistent with the NMR data, and we made a new construct encoding SIP residues 1-47 that crystallized and diffracted X-rays to high resolution. We solved the crystal structure of SIP-(1-47) at 1.2 Å resolution (Fig. 1, TABLE I). The protein is dimeric, as expected based on our previous biophysical and biochemical studies (31). There are two independent copies of the dimer within the crystal lattice with essentially identical structure. The dimer forms a curved cylinder, ϳ50 Å long with a maximum diameter of ϳ20 Å. Each monomeric subunit forms a helical hairpin in which the two helices, ␣ 1 (residues 2-20) and ␣ 2 (residues 24 -47), are connected by a tight 3-residue turn. Hairpins from two monomers associate as a four-helix bundle (Fig. 1, C and D), with the two C termini at opposite ends of the cylinder, 50 Å apart. The bundle is stabilized by a hydrophobic core across the 2-fold axis that includes hydrophobic residues (Val 13 , Leu 16 , Val 25 , and Leu 29 ) from each chain; further stability arises from a number of surface-exposed salt bridges, burying a total of 1700 Å 2 . The DCOMPLEX server (available on the World Wide Web at phyyz4.med.buffalo.edu/czhang/complex.html) (48) attempts to discriminate between biological dimers and crystal packing artifacts. It predicts that SIP-S forms a true biological dimer and estimates a K d of Ϫ15.2 kcal/mol (compared, for example, with Ϫ18.6 kcal/mol for the Siah1 dimer). The dimerization motif has a novel topology according to SCOP (49) and most closely resembles the dimeriza-  (50), and the authors proposed that SIP-S is monomeric based on the absence of putative intermolecular NOEs in a sample comprised of labeled and unlabeled SIP-S at 1:1 ratio. However, it is not clear if mixing equal amounts of labeled and unlabeled samples is sufficient to obtain a differentially labeled dimer, especially if the dissociation constant is in the nanomolar range. Also, in comparing the NMR and x-ray structures, there are significant differences in the relative disposition of the two helices that might be explained if some of the intermolecular NOEs have been interpreted as intramolecular. Clearly, further biophysical data are required to settle this issue.

SIP-S Engages Siah1 via the PXAXVXP Consensus Motif within Its
Flexible Tails-We previously showed that SIP binds to Siah1 via its N-terminal 72 residues (31). We carried out a series of binding studies of SIP-S to Siah1⌬R. Siah1⌬R binds SIP-S with a K d of 10 Ϯ 5 M as determined by isothermal titration calorimetry and NMR. Using NMR, we mapped the interactions between 15 N-SIP-S and Siah1⌬R. Initial spectra showed extreme broadening of most resonances due to the large size of the resulting protein complex (ϳ60 kDa). This problem was largely resolved by using a combination of deuteration and TROSY (51). Subsequently, 1 H, 15 N TROSY spectra of 15 N/ 2 H-SIP-S were recorded upon titration with unlabeled Siah1⌬R (Fig. 1E) in order to map the intermolecular interactions (52). Two types of changes were observed. The more obvious one was the extreme broadening of a subset of the tail resonances, consistent with residues in the C-terminal segment of SIP-S interacting with Siah1. In addition, several resonances within the ordered domain are shifted, due to a second site of interaction (see below).
SIP-S contains the core of the consensus Siah-binding motif, 60 PAAVVAP 66 , within its flexible tail. To further characterize SIP-Siah1 interactions, we prepared a selectively labeled sample of SIP-S in which its six Thr residues are 1 H-and 13 C-labeled, the five Ile residues are 1 H and 13 C labeled in the ␦ 1 position only, whereas all other residue types are labeled with 12 C and 2 H. A 13 C, 1 H HMQC spectrum of such a sample yields cross-peaks in the aliphatic region corresponding to ␥-methyl groups of Thr residues and ␥ 1 methyl groups of Ile residues. We selected these two amino acid types because they occur in both the structured domain and the tail region. We monitored the perturbations in their resonances upon titration with Siah⌬R (Fig. 1F). Resonance broadening upon complexation was most evident for the threonine residues in the tail region that are close to the 60 PAAVVAP 66 motif, whereas the side chains of other labeled residues remained largely unaffected, supporting a role for the tail region and, in particular, the PAAVVAP motif in binding to Siah. Although we were unable to co-crystallize Siah1⌬R with either SIP or SIP-S, we did succeed in crystallizing Siah1⌬R in complex with a synthetic peptide comprising SIP residues 58 EKPAAVVAPITTG 70 . The peptide binds to Siah1⌬R with a K d of 24 Ϯ 4 M as determined by isothermal titration calorimetry. We solved the structure at 2.2 Å resolution by molecular replacement (TABLE I, Fig. 2). The crystal symmetry and packing are different for the peptide-bound versus unliganded Siah1⌬R; nevertheless, the overall structure and organization are very similar, and as observed in the unliganded structure, the first zinc finger is poorly ordered. Good electron density is observed for the 60 PAAV-VAP 66 motif, and some density is observed for one flanking residue at each end of the motif (Fig. 2B). NMR-transferred NOE measurements using a similar peptide confirmed that the peptide binds to Siah1⌬R in solution and adopts an extended conformation as indicated by strong sequential H ␣ -H N NOEs that span from residue Glu 58 to Ala 65 (Fig. 2C). One SIP peptide binds at each edge of the saddle-shaped upper surface of the Siah1 dimer, which is formed by an eight-stranded antiparallel ␤-sheet, with the N termini of the two peptides separated by 35 Å (Fig.  2D). Each 60 PAAVVAP 66 motif forms a ␤-strand that augments the ␤-sheet, making parallel ␤-strand main-chain interactions from Pro 60 to Val 64 with strand ␤ 1 of Siah1; in addition, Ile 67 , outside the central motif, makes a main-chain hydrogen bond to Asp 177 in strand ␤ 2 , part of the lower sheet of the Siah ␤-sandwich (Fig. 2E). The side chains of the consensus residues Ala 62 and Val 64 pack against the hydrophobic core exposed at the edge of the ␤-sandwich. The main chain of the tripeptide 60 PAA 62 packs around an exposed hydrophobic group, Leu 158 , from strand ␤ 0 of Siah; this strand is the connector between the ␤-sandwich and the second zinc finger motif. When compared with the unliganded structure, peptide binding induces small (1-2 Å) changes in the local structure of Siah1⌬R in this region. The conserved Pro 66 stacks against the side chain of Trp 178 (strand ␤ 2 ) in Siah1. Residues Ala 61 and Val 63 point out into solution, consistent with their lack of conservation among Siah-binding proteins.
High affinity peptides derived from phyllopod and plectin have been shown to bind very strongly to Siah1 (K d ϳ 100 nM) and to compete effectively with a range of Siah-binding proteins, including SIP (35). The phyllopod and plectin sequences contain arginines flanking the central motif that have been shown to be important for binding. We note that two acidic patches on Siah1 (Glu 161 /Asp 162 and Asp 177 /Glu 194 ) are appropriately placed to make salt bridges with the arginines; these additional interactions may explain their stronger binding.
The PXAXVXP Motif Is Required for ␤-Catenin Regulation-Consistent with our structural results, it has previously been shown that mutation of the invariant residues within the 60 PXAXVXP 66 motif, Val 64 and Pro 66 , reduced the binding of SIP to Siah1 in an in vitro pull-down assay (35). We confirmed the importance of these residues by probing the binding of SIP and Siah1 in co-immunoprecipitation assays in HEK293T cells. As shown in Fig. 3A, Siah1 co-immunoprecipitated with wild-type but not with mutant SIP. Similar results were obtained using yeast two-hybrid assays (Fig. 3B). Overexpression of Siah induces FIGURE 3. The PXAXVXP motif within SIP is essential for Siah1 binding and ␤-catenin regulation. A, co-immunoprecipitation. HEK293T cells were transiently transfected with plasmids encoding Myc epitope-tagged full-length wild-type (wt) or mutant SIP (mt3; V64N and P66N) as indicated and hemagglutinin (HA)-tagged Siah1. Lysates were normalized for total protein content and immunoprecipitated using agarose-conjugated anti-Myc monoclonal antibody (9E10). After recovering immune complexes and washing, samples were analyzed by SDS-PAGE/immunoblotting using an anti-HA monoclonal antibody. An aliquot of the cell lysate (0.02 volumes) was used as a control. Immunoblot analysis confirmed that the expected ϳ30-kDa Myc-SIP wild type and mutant were produced at readily detectable levels in transiently transfected HEK293T cells. Siah1-(81-282) was used for all immunoprecipitation and yeast two-hybrid experiments, since its RING domain causes rapid turnover of the protein (61). The apparent increased mobility in SDS-PAGE of the charged-toalanine SIP mutants is presumably due to differential detergent binding, because no truncation or deletion was introduced. B, yeast two-hybrid experiments were performed by transforming EGY191 (strain) cells. Wild-type or mutant SIP as indicated was expressed from plasmid pGilda or pJG4-5 and tested for interactions with Siah1 as detected by trans-activation of LEU2 reporter genes. Growth on leucine-deficient medium at 30°C was examined 4 days later. Plasmid combinations that resulted in growth on leucine-deficient medium within 4 days were scored as positive. The ␤-galactosidase activity of each colony was tested by filter assay and scored as blue (ϩ) or white (Ϫ) after 1 h. C, ␤-catenin assays. HEK293T cells were transiently transfected with 0.2 g of plasmids encoding Myc-␤-catenin and full-length wild-type Siah1 (0.5 g) in combination with wild-type (wt) or mutant SIP (total DNA amount normalized). After 24 h, cell lysates were prepared from duplicated dishes of transfectants, normalized for total protein content (20 g/lane), and analyzed by SDS-PAGE/immunoblotting using antibodies specific for Myc-␤-Catenin (top), Myc-SIP, or endogenous c-Myc as a control (bottom), with ECL-based detection. D, in vivo Tcf/LEF activity measurements. HEK293T cells were transiently transfected with 0.1 g of a plasmid containing a Tcf/LEF-responsive element cloned upstream of a luciferase reporter gene (54,55), together with 0.01 g of pCMV-␤-gal as a transfection efficiency control and 0.1 g of plasmids encoding ␤-catenin, Siah1, and wild-type or mutant SIP as indicated, normalizing by adding empty pcDNA3 as necessary. Luciferase activity was measured in cell lysates 24 h later and normalized relative to ␤-galactosidase (mean Ϯ S.D.; n ϭ 3).
ubiquitin-dependent degradation and reduction of ␤-catenin levels in cells, which can be inhibited by co-expression of loss-of-function mutants of SIP (31). We took advantage of this knowledge by monitoring ␤-catenin levels in cells transfected with Siah1 and wild-type or mutant SIP. HEK293 cells were transiently transfected with plasmids encoding Siah1, alone or in combination with wild-type or mutant fulllength SIP. As shown in Fig. 3C, overexpression of full-length Siah1 caused a marked reduction in cellular levels of ␤-catenin. Co-transfection of wild-type SIP with Siah1 had no substantial effect on ␤-catenin levels, since endogenous SIP is abundant in HEK293 cells. In contrast, a mutant SIP with Asn substitutions in the Siah-binding site acted as a dominant negative mutant abrogating the effect of Siah1, perhaps because it retains Skp1-binding activity. Since ␤-catenin is a co-factor for the transcriptional activator Tcf/LEF (53), the effects of wild-type and mutant SIP on Tcf/LEF activity were probed using transient transfection reporter gene assays (54,55). Whereas transfection with ␤-catenin induced a Ͼ10-fold increase in Tcf/LEF activity in HEK293T, and co-transfection of an equivalent amount of a Siah1-encoding plasmid reduced such enhancement by about half, co-transfection of wild-type SIP with Siah1 did not lead to a significant change in Tcf/LEF activity (Fig. 3D). In contrast, co-transfection of mutant SIP failed to suppress ␤-catenin-mediated activation of Tcf/LEF and rather increased transactivation of the Tcf/LEF-responsive reporter gene plasmid. Taken together, these results show that Val 64 and Pro 66 are required for interaction of SIP with Siah1 in vitro and for the function of the full-length proteins in cells.
The SIP Dimerization Domain Sits across the Siah1 Saddle-The location of the 60 PAAVVAP 66 motifs, with their N termini protruding at the two edges of the central ␤-sheet of Siah, places strong constraints on the location of the SIP N-terminal dimerization domain. Consideration of shape complementarity suggested to us that the SIP dimerization domain might use its concave face to sit across the saddle formed by the ␤-sheet on the upper surface of Siah1. Indeed, our NMR data show that the SIP residues most affected by binding to Siah1 map to its lower concave surface (Fig.  4A), and we previously showed by mutagenesis that acidic residues on the surface of the Siah1 saddle are important for SIP-Siah1 association (56). Computational molecular docking further supports this location: the coordinates of the SIP dimer were subjected to a conformational search over the entire surface of the Siah1⌬R dimer. In the top solution, the concave surface of SIP-(1-47) docked onto the Siah1 saddle with the 2-fold axes of SIP and Siah1 approximately aligned (no symmetry restraints were applied).
To explore this hypothesis, we mutated lysines and arginines protruding from the surface of the concave dimerization domain of fulllength SIP. Most single site point mutations had a limited effect (not shown), although a K35A mutation reduced Siah1 binding significantly. However, a triple mutant, K23A/R24A/R26A, completely abrogated binding to Siah1⌬R in co-immunoprecipitation experiments (Fig. 4B). Similar results were obtained using yeast two-hybrid assays (Fig. 4C). The triple mutant is folded correctly, as judged by its ability to homodimerize and to heterodimerize with wild-type SIP (Fig. 4C). Therefore, our data support a function for the SIP N-terminal domain in stabilizing the interaction with Siah1 in addition to its role as a dimerization module.
The SIP-Saddle Interaction Is Also Required for Cellular Function-To assess the importance of the SIP-saddle interaction for ␤-catenin regulation in cells, we examined the effect of the mutant SIP, K23A/ R24A/R26A, on ␤-catenin levels in transient transfection assays in HEK293T cells. Overexpression of Siah1 alone markedly reduced levels FIGURE 4. The dimerization domain of SIP is directly involved in Siah1 binding and is essential for ␤-catenin regulation. A, structure of SIP-S with residues affected by binding to Siah1⌬R shaded in red as determined by NMR (see Fig. 1E). The proposed binding site for the Siah1 saddle is indicated. The C-terminal tails are shown schematically in white (disordered) and blue (ordered in the presence of Siah1⌬R). B, full-length wild-type (wt) SIP or mutant mt1 (K23A/R24A/R26A) or mt2 (K35A) was assayed for binding to Siah1 by co-immunoprecipitation. Details for B-E are explained in the legend to Fig. 3, A-D, respectively. C, yeast two-hybrid analysis. D, ␤-catenin degradation. E, in vivo Tcf/LEF activity.
of ␤-catenin (Fig. 4D). Co-transfection of wild-type SIP with Siah1 did not substantially alter Siah1-mediated degradation of ␤-catenin, since endogenous SIP is abundant in HEK293 cells. In contrast, mutant SIP abrogated the effect of Siah1 overexpression, acting in a dominant negative fashion. Similarly, in a transient transfection reporter gene assay, co-expression of wild-type SIP with Siah1 had no effect on the suppression of ␤-catenin-induced Tcf/LEF activation compared with Siah1 alone (Fig. 4E), whereas co-expression of mutant SIP (K23A, R24A, R26A) with Siah1 largely ablated the effect of Siah1, restoring Tcf/LEF transcriptional activity. These results thus suggest that a second SIP-Siah1 interface is required for the assembly and function of the Siah1-SIP-Skp1-Ebi complex.
A Different Surface of Siah1 May Direct the Assembly of the E3 Ligase Complex-The C termini of the SIP-S dimer protrude on the lower surface of Siah1 (i.e. opposite from the saddle that binds the SIP dimerization domain) adjacent to the second zinc finger domain and several conserved loops (Fig. 5A). We previously showed that the C-terminal domain of full-length SIP, residues 73-228, contains the binding determinants for Skp1 (31). To test the hypothesis that this lower surface is involved in assembly of the E3 multiprotein complex, we made mutations in two of the surface loops. The mutants retained the ability to bind SIP, as revealed by yeast two-hybrid assays and co-immunoprecipitation (Fig. 5, B and C), indicating that they do not adversely affect folding. However, the mutants failed to mediate degradation of ␤-catenin or to suppress Tcf/LEF-dependent transcription (Fig. 5, D and E), suggesting that they play a functional role in directing the ubiquitination reaction.

DISCUSSION
By analogy to the well studied SCF multiprotein E3 ligase complex, the Siah1-SIP-Skp1-Ebi complex is believed to function by assembling a scaffold that orchestrates the ubiquitination reaction by bringing the E2 enzyme and its substrate into apposition with the appropriate geometry. Our results allow us to develop a model of the structure and function of the multiprotein Siah1 complex.
We identified two binding surfaces for SIP on Siah1, both of which are necessary for the Siah1-SIP complex to form, and showed that disruption of either of these two interactions abrogates functional activity in a cellular context, indicating that the full interaction is required for the assembly of the Siah-SIP-Skp1-Ebi E3 ligase complex. The helical N-terminal domain of SIP/SIP-S forms a dimer, and our data are consistent with a model in which the SIP-S dimer sits atop the saddle formed by the large ␤-sheet on the upper surface of Siah1, making chiefly ionic interactions, while two "legs" containing the 60 PAAVVAP 66 sequence pack against both sides of the Siah1 dimer, making specific ␤-sheet interactions (Fig. 6A). The binding mode, in which a ␤-strand from the ligand augments a ␤-sheet, is reminiscent of the interactions between PTB and PDZ domains and their ligands, as well as those of the structurally related tumor necrosis factor receptor-associated factor with ligands (32,57). The distance (ϳ18 Å) between the C termini of the SIP dimer (residues 1-47) and the N termini of the SIP peptides (residues 59 -67) can be readily bridged by the 11-residue linker (residues 48 -58).
Our structural data provide a rationale for the role of the PXAXVXP motif in Siah1-SIP interactions. More than half of Siah-binding proteins identified thus far contain this motif (Fig. 6B). High affinity peptides FIGURE 5. Residues on the "lower" surface of Siah1 are required for function in ␤-catenin regulation. A, the "lower face" of Siah1⌬R (i.e. opposite from the one involved in binding the SIP dimerization domain) viewed along the 2-fold axis. The mobile first zinc finger is in yellow. Helix ␣ 2 is cyan. Disordered loops are in blue (residues 197-202), green (172-175), and magenta (132-133). The SIP PXAXVXP peptide is in red. Mutated residues are indicated. B, full-length SIP was assayed for binding to wild-type (wt) or mutant Siah1 by co-immunoprecipitation. The mutants are mtA (K198G/Y199D/D200G), mtB (M252D), and mtC (M252K). Details for B-E are explained in the legend to Fig. 3, A-D, respectively. C, yeast two-hybrid analysis. D, ␤-catenin degradation. E, in vivo Tcf/ LEF activity. OCTOBER 7, 2005 • VOLUME 280 • NUMBER 40 from phyllopod and plectin containing the motif have been shown to compete effectively with a range of Siah-binding proteins, including SIP (35). It thus seems very likely that many Siah-binding proteins will bind in the same mode that we have observed for SIP. The importance of this interaction is underscored by studies of the AF4 gene, which is disrupted in childhood leukemia (58) and mutated in the "robotic" mouse (59). The human AF4 gene binds Siah1 (23), and genetic mapping identified a point mutation (V280A) in AF4 within the consensus PXAXVXP motif (22,23) that correlated with accumulation of the AF4 proteins and increased transcriptional activity in cells, suggesting that the AF4 activity is controlled by Siah1-mediated degradation.

Siah-SIP Complex Structure
We previously showed that the C-terminal domain of full-length SIP, residues 73-228, contains the binding determinants for Skp1 (31). In our model of SIP-Siah1 binding, the C termini of SIP-S emerge on the "lower" surface of the Siah1 dimer, on the opposite side from the binding site for the SIP dimerization domain. This suggests that the C-terminal domain of full-length SIP (residues 68 -228) and its ligand Skp1 lie adjacent to this lower surface. This surface includes three prominent loops that are highly conserved but mobile in the Siah1 crystal structures (except when stabilized by fortuitous crystal contacts), as well as an ␣-helix (␣2) that packs against the ␤-sheet. We showed that mutagenesis of one of the mobile loops, as well as of a conserved methionine residue exposed on the surface of helix ␣ 2 , abrogated ␤-catenin degradation without affecting SIP binding. Thus, our studies support a role for the lower surface of Siah1 in organizing the E3 ligase assembly into a functional complex. Indeed, a conservative point mutation (I208L) in a patient with gastric cancer causes inactivation of Siah1 and maps to the ␤ 4 strand, where it is likely to subtly alter the packing of helix ␣ 2 against the ␤-sheet (60).
Finally, we note that the inherent binding affinity of SIP for Siah1 (ϳ11 M) is relatively low, consistent with the ability of peptides derived from phyllopod to readily disrupt this interaction. Therefore, we propose that the selection between construction of a single protein E3 ligase versus an SCF-like multiprotein complex could be made at the level of competition between SIP and substrate proteins binding at the PXAX-VXP site on Siah1 and modulated by the availability of other components of the multiprotein complex that would enhance binding.