Structure of a Conserved Dimerization Domain within the F-box Protein Fbxo7 and the PI31 Proteasome Inhibitor*

F-box proteins are the substrate-recognition components of the Skp1-Cul1-F box protein (SCF) E3 ubiquitin ligases. Here we report a structural relationship between Fbxo7, a component of the SCFFbxo7 E3 ligase, and the proteasome inhibitor PI31. SCFFbxo7 is known to catalyze the ubiquitination of hepatoma-up-regulated protein (HURP) and the inhibitor of apoptosis (IAP) protein but also functions as an activator of cyclin D-Cdk6 complexes. We identify PI31 as an Fbxo7·Skp1 binding partner and show that this interaction requires an N-terminal domain present in both proteins that we term the FP (Fbxo7/PI31) domain. The crystal structure of the PI31 FP domain reveals a novel α/β-fold. Biophysical and mutational analyses are used to map regions of the PI31 FP domain mediating homodimerization and required for heterodimerization with Fbxo7·Skp1. Equivalent mutations in Fbxo7 ablate interaction with PI31 and also block Fbxo7 homodimerization. Knockdown of Fbxo7 does not affect PI31 levels arguing against PI31 being a substrate for SCFFbxo7. We present a model for FP domain-mediated dimerization of SCFFbxo7 and PI31.


F-box proteins are the substrate-recognition components of the Skp1-Cul1-F box protein (SCF) E3 ubiquitin ligases.
Here we report a structural relationship between Fbxo7, a component of the SCF Fbxo7 E3 ligase, and the proteasome inhibitor PI31. SCF Fbxo7 is known to catalyze the ubiquitination of hepatoma-up-regulated protein (HURP) and the inhibitor of apoptosis (IAP) protein but also functions as an activator of cyclin D-Cdk6 complexes. We identify PI31 as an Fbxo7⅐Skp1 binding partner and show that this interaction requires an N-terminal domain present in both proteins that we term the FP (Fbxo7/PI31) domain. The crystal structure of the PI31 FP domain reveals a novel ␣/␤-fold. Biophysical and mutational analyses are used to map regions of the PI31 FP domain mediating homodimerization and required for heterodimerization with Fbxo7⅐Skp1. Equivalent mutations in Fbxo7 ablate interaction with PI31 and also block Fbxo7 homodimerization. Knockdown of Fbxo7 does not affect PI31 levels arguing against PI31 being a substrate for SCF Fbxo7 . We present a model for FP domain-mediated dimerization of SCF Fbxo7 and PI31.
The levels of many regulatory and misfolded proteins are controlled by the ubiquitin-proteasome system (1,2). A series of enzymes termed E1 (ubiquitin-activating enzyme), E2 (ubiquitin carrier protein), and E3 6 (ubiquitin ligase) act in the ubiquitin-proteasome system as part of a concerted cascade to activate and conjugate ubiquitin via an isopeptide linkage to protein substrates (2). Ubiquitin modification can affect the activity, localization, sorting, and stability of protein substrates (2)(3)(4). Recognition and recruitment of protein substrates to the ubiquitin-proteasome system machinery resides within the multisubunit E3 ubiquitin ligases, the largest group of which is the Skp1-Cullin1-F-box protein (SCF) family (1,5,6). All known SCF E3 ligases bind an E2 (ubiquitin carrier protein) enzyme through a RING finger-containing subunit (Rbx1) and utilize their F-box subunit to recruit substrates. Although classified by the presence of an F-box (Fbx) motif (7), F-box proteins can be further divided into three classes depending on additional structural elements: Fbxw (WD40 motifs), Fbxl (with leucine-rich repeats), and Fbxo (F-box domain only) (8,9). Recent structural data have shown how the Fbxw protein CDC4 (Fbw7) uses the center of the WD40 propeller to recognize the phosphorylated epitope of its substrate, cyclin E (10,11). Structural data also exist for the Fbxl protein Skp2 (Fbl1) as part of the SCF Skp2 complex bound to phosphorylated p27 (12). In contrast, little is known about how Fbxo proteins recognize their substrates or indeed whether other substrate-targeting domains exist within this subgroup.
Fbxo7 has been characterized recently as a member of the Fbxo subgroup and is conserved among higher eukaryotes (8,13). As well as an F-box motif (residues 329 -375), which interacts with Skp1 (8,9,13,14), Fbxo7 has an N-terminal ubiquitinlike (Ubl) domain (residues 1-78) (13,15) (Fig. 1A), which is thought to mediate interactions with ubiquitin receptor proteins bearing ubiquitin binding domains (16). The C terminus of Fbxo7 contains a proline-rich region (PRR) but lacks any predicted secondary structure. HURP (hepatoma up-regulated protein) and cIAP1 (an inhibitor of apoptosis protein) have both been reported as substrates for SCF Fbxo7 -mediated ubiquitination leading to proteasome-mediated degradation (17,18). Each binds to Fbxo7 through its C-terminal proline-rich region (17,18). Fbxo7 also acts as an assembly factor for cyclin D-Cdk6 complexes by virtue of its interaction with D-type cyclins and Cdk6 (13).
In this study a combination of structural, biochemical, and genetic approaches were used to identify and characterize regions of Fbxo7 involved in protein interaction as a means to identify putative substrates of SCF Fbxo7 . Unexpectedly we uncovered a structural link between a domain from Fbxo7 and a related domain from PI31, a regulatory subunit of the immunoproteasome which is an in vitro inhibitor of the 20 S proteasome (19 -22). The crystal structure of this domain from PI31 revealed a novel ␣/␤-fold and two distinct intermolecular contact surfaces. We used this structure to guide the design of * 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. The atomic coordinates and structure factors (code 2VT8) have been deposited in site-specific substitutions to probe surfaces of the FP domain from PI31 responsible for homodimerization and for heterodimeric interaction with Fbxo7. We discuss a possible role for FP domain-mediated dimerization for SCF Fbxo7 and PI31 function.
The expression plasmids described above were transformed into Escherichia coli strain BL21(DE3), and cultures were grown in Luria broth at 37°C to an A 600 of 0.6. Recombinant protein expression was induced by the addition of isopropyl-␤-D-1-thiogalactopyranoside to a final concentration of 250 M for 3 h at 37°C. Bacteria were harvested by centrifugation and resuspended in lysis buffer (100 mM NaCl, 50 mM Tris-HCl, pH 8, 10 mM benzimidine, 5 mM ␤-mercaptoethanol, 1 mM phenylmethylsulfonyl fluoride). Proteins were purified from the soluble fraction using either glutathione-Sepharose (Fbxo7⅐Skp1) or nickel-nitrilotriacetic acid affinity beads (PI31) by batch purification and cleaved from the beads with thrombin at 4°C. PI31 and Fbxo7⅐Skp1 proteins were then purified by size exclusion chromatography using Superdex 75 and Superdex 200 columns, respectively. Columns were equilibrated and run in 20 mM Tris-HCl, 50 mM NaCl, pH 8, and 1 mM dithiothreitol. Selenomethionine was incorporated by transforming the expression plasmid into the E. coli methionine auxotroph strain B834(DE3) grown on minimal medium supplemented by selenium methionine using standard procedures. Selenomethionine-labeled protein and mutants were prepared using a similar procedure to the native protein as described above.
Crystallization and Data Collection-The PI31 FP domain was crystallized by vapor diffusion in sitting drops containing 2 l of protein (8 mg/ml in 20 mM Tris-HCl, pH 8.0) and 2 l of well solution (20% polyethylene glycol 3350 and 0.1 M ammonium iodide) at room temperature. Crystals grew within 1 week to roughly 0.2 mm. Selenomethionine-substituted wild type FP domain and L7M mutant protein crystallized under the same conditions as the native protein. Data were collected for crystals of human PI31 FP domain, selenomethionine-substituted FP domain, and the selenomethionine-substituted L7M mutant (Table 1). In each case a single crystal was flash-cooled to 100 K in a nitrogen gas stream using 20% (v/v) ethylene glycol as a cryoprotectant. The CCP4 suite of programs (25) were used for data processing, and REFMAC and COOT were used for refinement and model building (26,27).
Structure Determination-Initial attempts to solve the structure using selenomethionine single wavelength anomalous diffraction data were unsuccessful due to the centrosymetric arrangement of the two selenium sites from two methionines in native PI31 FP domain (data not shown). A third methionine site was engineered by an L7M mutation leading to crystals of selenomethionyl-labeled protein, and subsequently datasets were collected at the selenium peak wavelength ( Table 1, datasets 2 and 3). The selenium substructure from dataset 2 was found by SHELXD/E and used to calculate SAD phases with a contrast of 0.403 and connectivity of 0.877 (incorrect hand had a contrast of 0.314 and connectivity of 0.845). Phases were improved by 2-fold non-crystallographic averaging within dataset 2 (initial FOM 0.45, final FOM 0.64) using DM (28) and cross-crystal averaging with dataset 3 ( Table 1, datasets 2 and 3) and density modification using RESOLVE (29). To obtain the native PI31 FP domain structure free of introduced mutations, a partial model was built into this map and was subsequently positioned by molecular replacement into the native unit cell (dataset 1). Refinement against this native dataset collected in house produced the final model (Table 1, dataset 1). The final model lacks 9 residues from the C termini of both copies (chain A and B) and residue Ser-75 in chain A. Residues in loop 72-76 and 96 -98 are relatively mobile as reflected in high temperature factors and poor electron density. Chain A and B are essentially identical (root mean square deviation of 0.419 Å over 141 C␣ residues). Four of the cysteine residues in each chain appear to have undergone partial oxidation and have been modeled as sulfenic acids. Coordinates have been deposited in the Protein Data Bank data base with accession code 2VT8.
Isothermal Calorimetry-Isothermal calorimetry experiments used a MicroCal VP-ITC according to manufacturer's instructions. All proteins analyzed by isothermal calorimetry (ITC) were purified by size exclusion chromatography and dialyzed into the same buffer (20 mM Hepes, pH 8, 50 mM NaCl, and 1 mM ␤-mercaptoethanol) before measurement of their molar concentration. The Fbxo7⅐Skp1 concentration within the sample cell was 20 M, and the PI31 titrant concentration was 100 M. The experiments were carried out at 30°C with 28 injections each of 10 l measured with stirring, and all settings were constant for both wild type and mutant experiments. Results were analyzed using the ORIGIN7 software provided by MicroCal and using a one-binding-site model.
Analytical Ultracentrifugation-For analytical ultracentrifugation (AUC) experiments, recombinant proteins were stored and analyzed in a buffer containing 20 mM Tris-HCl, pH 8, 50 mM NaCl, and 5 mM ␤-mercaptoethanol. A Beckman XL-I analytical ultracentrifuge was used to measure both absorbance and interference data according to the manufacturer's instruc-tions. Protein parameters used for data analysis and the calculation of initial experimental parameters were calculated using the program SEDNTERP (30). Data were analyzed using the AUC machine software and the programs SEDFIT and SEDPHAT (31) using the model monomer-dimer equilibrium in the final analysis.
Cell Culture-Human osteosarcoma (U2OS) and Jurkat E6 cell lines were obtained from Cancer Research UK (LRI Cell Production). U2OS cells were grown in Dulbecco's modified Eagle's medium, whereas Jurkat E6 cells were grown in RPMI media. Both media were supplemented with 10% fetal calf serum (Helena Biosciences), 2 mM glutamine, 100 units/ml penicillin, and streptomycin at 37°C in a humidified 5% CO 2 atmosphere. U2OS cells were seeded the day before transfection using FuGENE (catalog #1814443, Roche Applied Science). To visualize the subcellular localization of Fbxo7, cells were transfected with a construct encoding a fusion of dsRED with full-length Fbxo7. 24 h post-transfection, live cells were visualized by confocal microscopy. For immunofluorescence assays on PI31, cells were grown on glass coverslips, fixed in 3% paraformaldehyde, permeabilized with 0.1% Triton X, and blocked with 5% fetal bovine serum before incubation with primary antibody against PI31 (1:200) and secondary rhodamineconjugated anti-mouse antibody (1:500). The fluorescent dye 3,3Ј-dihexyloxacarbocyanine iodide (DiOC 6 (3)) (Invitrogen) was included in the final wash to permit visualization of the endoplasmic reticulum by confocal microscopy. For co-immunofluorescence assays, cells were grown on glass coverslips, fixed in 70% ethanol, permeabilized with 0.1% Triton-X, and blocked with 0.1% gelatin before staining with affinity-purified Fbxo7 polyclonal antibody (1:40) and PI31 monoclonal antibody (1:10). Secondary antibodies sheep anti-rabbit conjugated to AlexaFluor 488 (Molecular Probes) and goat anti-mouse conjugated to Cy5 (DAKO) were both used at 1:300. Coverslips were mounted with Vectashield (Vector) mounting media containing 4Ј,6-diamidino-2-phenylindole. Cells were visualized by epifluorescence microscopy.
Fractionation, Immunoprecipitation, and Western Blotting-U2OS cells were fractionated by resuspending in 100 mM Tris-HCl, pH 7.4, 100 mM NaCl, 25 mM MgCl 2 , and adding 40 g/ml digitonin and incubating on ice for 10 min to lyse plasma membrane. Nuclei were separated from cytoplasmic fraction by centrifugation at 3000 rpm for 10 min at 4°C and lysed in radioimmune precipitation assay buffer (50 mM Tris, pH 7.5, 150 mM NaCl, 1% Nonidet P-40, 0.5% sodium deoxycholate, 0.1% SDS, protease inhibitor mixture). For immunoprecipitation, cells were lysed in hypotonic lysis buffer (10 mM Tris-HCL, pH 7.5, 10 mM NaCl, 2 mM EDTA, 0.5% Triton X-100, 1 mM phenylmethylsulfonyl fluoride) and protease inhibitor mixture for 10 min on ice before the addition of NaCl to a final concentration of 150 mM. Lysates were centrifuged for 15 min at 13,000 rpm and then immunoprecipitated with anti-FLAG-M2-agarose slurry for 3 h at 4°C on a rotating wheel. Agarose beads were then washed 4 times with NET2 buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, and 0.05% Triton X-100) before being resuspended in 4ϫ Laemmli buffer and resolved by SDS-PAGE. The following antibodies were used for immunoprecipitation or Western blotting: Cdk6 (SC177) (Santa Cruz Biotechnology Inc.), FLAG (F3165, Sigma), and T7 (catalog #69522-3, Novagen). The antibody against Fbxo7 has been previously described (13). The monoclonal antibody raised against the FP domain of PI31 was made by Monoclonal Antibody Service, Cancer Research UK. Anti-rabbit IgG and anti-mouse IgG antibodies conjugated to horseradish peroxidase were obtained from Santa Cruz Biotechnology, Inc. and Jackson ImmunoResearch Laboratories.

Mapping Protein Interaction Sites within Fbxo7 and Identification of PI31 as an Fbxo7
Binding Partner-To map regions within Fbxo7 responsible for mediating its in vivo interactions with known protein partners, we prepared several deletion constructs based on predicted domain boundaries of Fbxo7 and incorporated an N-terminal T7 epitope. These included constructs lacking the N-terminal Ubl domain, the F-box motif, or the C-terminal PRR (Fig. 1A). We first sought to map the binding site for Cdk6, a validated binding partner for Fbxo7 (13). Cdk6 was efficiently co-immunoprecipitated by full-length Fbxo7-(1-522) and by a mutant of Fbxo7 lacking the F-box domain (⌬F-box) indicating that the interaction did not require the F-box (Fig. 1, A and B). We next deleted the first 129 amino acids of Fbxo7 which includes the Ubl domain, designated as Fbxo7-(129 -522). However, by SDS-PAGE, this mutant co-migrated with IgG heavy chain (molecular mass, 50 kDa), making it unusable in our coimmunoprecipitation assays. Fbxo7-(129 -398) (lacking both the Ubl domain and the PRR region) was also able to co-immunoprecipitate Cdk6 in vivo (Fig. 1, A and B), showing neither region was essential for binding. Because Fbxo7-(129 -398) was predicted to have an unstructured sequence of 40 amino acids at the N terminus (129 -169), we therefore, truncated the construct further to produce Fbxo7-(169 -398). This protein was now unable to bind Cdk6 (Fig. 1B), indicating that within the context of Fbxo7-(129 -398), amino acids 129 -169 were necessary to bind Cdk6.
We then made further deletions from the N terminus including Fbxo7-(239 -522), which surprisingly, was able to bind Cdk6 despite the absence of residues 129 -169. This indicated another region within 239 -522 could mediate Cdk6 interaction. This region contains the F-box domain and the PRR sequence. We deleted the unstructured C terminus of Fbxo7, creating Fbxo7-(239 -381) and also attempted to express the C terminus, Fbxo7-(419 -522), but this protein could not be produced (Fig. 1, A and B). Fbxo7-(239 -381) did not interact with Cdk6, suggesting that the sequences from 381-522, which includes the PRR, previously proposed to mediate substrate recognition for HURP and c-IAP, could also contribute to Cdk6 interaction (17,18). Our data indicate that two regions from Fbxo7 can independently bind to Cdk6, suggesting Fbxo7 has a bipartite Cdk6 binding site. By mapping the binding site of a known Fbxo7-interacting partner to domains within Fbxo7, we val-idated the use of this panel of Fbxo7-truncated constructs to search for new protein partners of Fbxo7 and, thus, for SCF Fbxo7 .
Closer examination of Fbxo7 sequence conservation and predicted secondary structure identified a putative globular domain (residues 180 -324) that precedes the F-box and follows the Ubl and the N-terminal Cdk6 binding sequence. We also identified a highly conserved R(Ar)DP motif (where Ar indicates any aromatic amino acid) within residues 466 -496 of the PRR, with an as yet undetermined function. To investigate whether residues 180 -324 could function as a protein interaction module and to identify putative partner proteins, we prepared GST-tagged recombinant Fbxo7-(129 -398)⅐Skp1 complexes for ex vivo affinity pulldown experiments from Jurkat cell lysates as described under "Experimental Procedures" (33). Skp1 was co-expressed to engage the F-box motif and was used as a positive control to demonstrate proper folding of Fbxo7 (Fig. 1C). After incubation of recombinant GST-Fbxo7-(129 -398)⅐Skp1 immobilized on glutathione-Sepharose beads with the Jurkat cell lysate, the beads were washed and treated with thrombin to release Fbxo7-(129 -398)⅐Skp1. Co-eluting proteins were then identified by mass spectrometry. Using this approach we identified PI31 (proteasome inhibitor 31kDa) as co-eluting with Fbxo7⅐Skp1 ( Fig. 1C) but not from a GST control elution. In addition, the known Skp1-binding protein Cul-lin1 and endogenous Skp1 also co-eluted with Fbxo7⅐Skp1. The latter was present as excess recombinant Fbxo7 was present in the protein preparations.  129-522 nd FIGURE 1. Mapping Fbxo7 protein interaction sites. A, schematic of motifs within Fbxo7 and the deletion constructs used for co-immunoprecipitations with Cdk6. ϩ, detectable Cdk6 binding; Ϫ, no detectable Cdk6 binding; nd, not determined. B, in vivo co-immunoprecipitation (IP) assays using Fbxo7 deletion mutants. Lysates were immunoprecipitated with antibodies to Cdk6 and analyzed for the associated proteins as indicated. C, Coomassie-stained SDS-PAGE gel of binding proteins from Jurkat cell lysates isolated by affinity-tagged recombinant Fbxo7-(129 -398)⅐Skp1. Proteins were identified by mass spectrometry. D, yeast-2-hybrid assay for proteins interacting with the indicated bait proteins. Yeast cells were transformed as indicated (pGBD bait ϩ pGAD prey) and plated on media selecting for the plasmids (SC-Ura-Leu) and on media selecting for activation of reporter genes (SC-Ura-Leu-His-Ade).
In parallel to the affinity pulldown experiments, we undertook yeast two-hybrid screens of a human cDNA library using Fbxo7-(129 -398) and Fbxo7-(129 -522) as bait (Fig. 1D). In both screens more than 60% of the library plasmids that were isolated had PI31 fused with Gal4 activation domain (GAD), whereas 7% of the clones contained Skp1. Assaying for activation of the HIS3 and ADE2 reporter genes tested the specificity of the pGAD-PI31 interaction with the pGBD-Fbxo7 bait plasmid. Yeast co-transformed with Fbxo7 plasmid and pGAD-PI31 grew on media lacking histidine and adenine. However, yeast transformed with either plasmid alone or a different bait plasmid encoding a viral cyclin fused to GBD together with pGAD-PI31 failed to grow under the same conditions (Fig. 1D). Taken together, the results from these independent screens and affinity pulldown experiments indicate that PI31 can interact with Fbxo7 and is a previously uncharacterized binding partner for Fbxo7.
Fbxo7 and PI31 Are Structurally Related-After the identification of PI31 as a putative Fbxo7 binding partner, we analyzed sequences of Fbxo7 and PI31 and noticed that these proteins are related at both a sequence and structural level despite their involvement in quite different multiprotein complexes. Both contain a domain equivalent to residues 180 -324 of Fbxo7 and an unstructured C-terminal rich in proline ( Fig. 2A). The sequences of Fbxo7 (residues 180 -324) and PI31 (residues 1-151) are 24% identical and 45% similar (Fig. 2B). Further searches with sequence databases found no further examples of this domain, which we refer to as the FP domain (Fbxo7 and PI31). We also found that the R(Ar)DP motif originally identi-fied in the PRR of Fbxo7 is absolutely conserved in the PRR of all PI31 sequences identified to date (Fig. 2C). Thus, Fbxo7 and PI31 share a structural and evolutionary relationship despite their distinct biological functions.
The FP Domain Adopts an ␣/␤-Fold -We then determined the crystal structure of the human PI31 FP domain at 2.64 Å by SAD phasing (Table 1). Experimental phases were obtained using data collected from crystals of selenomethionine-labeled L7M mutation as described under "Experimental Procedures." Representative electron density for the refined structure is shown in Fig. 3A. The FP domain has approximate dimensions of 40 ϫ 28 ϫ 25 Å and consists of an ␣/␤-fold with a central five-stranded anti-parallel ␤-sheet flanked by two N-terminal ␣ helices and three C-terminal ␣ helices (Fig. 3B). All the helices pack against one face of the ␤-sheet, leaving the opposite side more accessible to solvent. Helix ␣2 is almost completely buried within the FP domain and contains two highly conserved characteristic residues, Asp-20 and His-27 (Fig. 2, B and C). This DX 7 H motif is intimately associated with both tyrosine residues on the YXLXY motif of strand ␤2 (residues 62-71) and contributes to a network of conserved hydrogen bonds. These include hydrogen bonds from His-27 side chain to both the Tyr-69 hydroxyl and to the main-chain carbonyl of Glu-52; OD1 of Asp-20 hydrogen bonds to Tyr-65 hydroxyl and the NZ atom of Lys-62, whereas OD2 hydrogen bonds to the mainchain amide functions of Thr-16 and Cys-17. A further feature of the FP domain structure is a striking hydrophobic patch comprising Leu-64, Ile-83, Val-85, and Ile-90 on the exposed surface of the central ␤-sheet, which is discussed later (Fig. 3D).

Pongo_pygmaeus_Fbxo7
D P P R F G P P G P G E T P S Q F P L R P I . . L P G P N P I L P Macaca_fascicularis_Fbxo7 D P P R F G P P G P G E T P S Q F P L R P I . . L P G P N P I L P Homo_sapiens_Fbxo7 D P P R F G P P G P G E T P S Q F P L R P V . . L P G P N P I L P Bos_taurus_Fbxo7 D P P R F G P P G P G E T P S Q F P L R P V . . L P G P N P I L P Canis_familiaris_Fbxo7 D P P R F G P P G P G E T P S Q F P L R P V . . L P G P R P T L P Gallus_gallus_Fbxo7 D P P H F G S P G P G E A P G Q F P F R P I . . L P G A N P T L P Mus_musculus_Fbxo7 D P P R F D P P R P G E L P G Q F R L R P V . . L P G P H S L L P    (38). Residues targeted by mutation in this study are indicated by a red star above the sequence. Secondary structure from the PI31 FP domain structure is indicated above the sequence. C, sequence conservation within the C-terminal R(Ar)D motif of PI31 and Fbxo7.
An irregular ␤1-␤2 loop unique to PI31 meanders alongside and shields the ␤1 edge strand from solvent and is stabilized by a salt bridge formed by residues Asp-48 and Arg-68.
Searches with secondary structure matching failed to reveal any other protein with a topology exactly matching that of the FP domain. A remote similarity was found with the E. coli protein CyaY (Protein Data Bank code 1EW4) (34), which is closely related to frataxin (Fig. 3E). CyaY and the PI31 FP domain superpose with a root mean square deviations of 2.85 Å (68 C␣ atoms). Although topologically similar with a single anti-parallel ␤-sheet and a helix equivalent to ␣2, their respective hydrophobic cores are quite distinct, indicating that no evolutionary relationship is likely.
Dimeric Arrangement of the PI31 FP Domain in Solution and within Crystals-Characterization of recombinant PI31 FP domain in solution by size exclusion chromatography indicated that the domain migrated with an apparent molecular mass of 32 kDa, consistent with a dimeric structure (Fig. 4A). Sedimentation equilibrium analysis by analytical ultracentrifugation under physiological buffer conditions also indicated a monomer-dimer equilibrium with an affinity constant (K d ) of 4.85 M. Two distinct dimeric arrangements of the PI31 FP domain are present within the crystal lattice, both with sufficiently large buried surfaces to be candidates for the solution dimer interface. One surface involves predominantly hydrophobic residues from helix ␣1 centered on Val-6 and buries an area of 557.2 Å 2 per protomer (Fig. 4, A, right panel, and B). These residues are generally conserved within PI31 sequences, and we refer to this as the "helical interface" (Fig. 4B). A second dimeric contact involves the solvent-accessible surface of the ␤-sheet, which is also relatively hydrophobic and covers 279.9 Å 2 per protomer (Fig. 4A, left panel). This "␤ interface" is centered on residues Ile-83 and Ile-90 ( Fig. 3D and 4B). Each of the two molecules within the asymmetric unit forms equivalent crystal-lographic dimers, suggesting these arrangements are not solely a result of lattice contacts.
The PI31 FP Domain Homodimerizes through a Helical Interface-To determine which of the two interfaces was required for PI31 homodimerization, disrupting substitutions were introduced into the dimeric interface of both the helical (V6R) and the ␤ interface (I83E/I90E) (Fig. 4, A and B). Both the V6R and I83E/I90E mutants expressed at similar levels to the wild type protein and purified under identical conditions (data not shown). We probed the oligomeric state of the mutated PI31 proteins by size exclusion chromatography and AUC. V6R migrated with a molecular weight by size exclusion chromatography consistent with a monomer (Fig. 4B), and AUC data could not be fitted with a monomer-dimer equilibrium model, indicating that this mutant is monomeric. In contrast, the double mutant I83E/I90E eluted at the same volume as wild type PI31 by size exclusion chromatography, and AUC gave a calculated K d of 3.95 M similar to the wild type PI31, indicating the dimeric structure was not perturbed by these point mutations. The monomeric nature of the V6R mutant demonstrates that the PI31 FP domain homodimerizes via the helical interface, the larger of the two buried surfaces observed within the crystal lattice.
The FP Domain of Fbxo7 and PI31 Heterodimerize through a ␤ Interface-We investigated whether the ␤ interface of the PI31 FP domain was involved in mediating heterodimerization with the Fbxo7 FP domain by using GST-Fbxo7-(169 -398)⅐Skp1 complex containing the Fbxo7 FP domain. The affinity of the FP domain of PI31 for Fbxo7-(169 -398)⅐Skp1⌬H8 was measured by ITC (Fig. 4C). The assumption that there is one binding site on Fbxo7⅐Skp1 per PI31 protomer gave the best fit to the experimental data. A K d of 0.37 M (error 12%) was determined for the binding of PI31 FP domain to Fbxo7-(169 -398) ⅐Skp1⌬H8. This is an ϳ10-fold higher affinity than that determined for the PI31 homodimer. Similar ITC experiments using the dimeric PI31 I83E/I90E double mutant showed no detectable binding as isotherms were identical to diluting sample with buffer (Fig. 4C). These in vitro data indicated that the ␤ interface of the PI31 FP domain forms a dominant part of the heteromeric interaction with Fbxo7⅐Skp1.
To test this interaction in vivo, double point mutations I83E/ I90E were engineered into full-length of PI31 tagged at the N terminus with a FLAG epitope and expressed in U2OS cells.
Immunoprecipitates from these cell lysates using anti-FLAG antibodies were subsequently analyzed by Western blotting for the presence of Fbxo7. Endogenous Fbxo7 co-immunoprecipitated with wild type PI31; however, despite expression at equivalent levels, the I83E/I90E mutations in PI31 did not interact with Fbxo7 in vivo (Fig. 4D). We noted that these point mutations reduced the mobility of PI31 compared with the wild type.
We also determined whether equivalent mutations within the FP domain of Fbxo7 would ablate its heterodimerization with PI31. Based on structural alignments of the two ␤ interfaces, Val-253 in the FP domain of Fbxo7 is equivalent to Ile-83 in PI31 (Fig. 2B). V253E Fbxo7 mutant was expressed and purified as a GST fusion protein as outlined above. Fbxo7-(169 -398 V253E)-Skp1⌬H8 behaved similarly to the PI31 I83E/I90E mutant in ITC experiments (Fig. 4C), giving no detectable binding. Similarly, we used in vitro binding assays using GST-PI31 to pull down in vitro transcribed and translated full-lengthT7-tagged Fbxo7 alleles (Fig.  4E). This confirmed that Fbxo7-(169 -398 V253E)-Skp1⌬H8 was severely impaired in binding PI31 (Fig. 4E). We also tested whether Fbxo7⅐Skp1 is able to homodimerize and whether V253E mutation affects this interaction. Indeed Fbxo7-(169 -398)-Skp1⌬H8 does pull down wild type Fbxo7 but not Fbxo7 (V253E) mutant (Fig. 4F). This indicates that Fbxo7 does indeed homodimerize and that this interaction also involves the Fbxo7 ␤ interface centered on Val-253. Therefore, the binding site for PI31 overlaps with a region of Fbxo7 required for homodimerization. We note that as the in vitro Fbxo7 expression and purification relied upon co-expression of Skp1, a role for Skp1 in these assays cannot be formally excluded.
PI31 and Fbxo7 Cellular Localization-Our data suggest that the FP domains of PI31 and Fbxo7 would be capable of interacting in vivo, which may have functional significance, so we tested whether they share a subcellular location. When U2OS cell lysates were separated into nuclear and cytoplasmic fractions and probed for Fbxo7 and PI31, both proteins were found to be cytoplasmic, and a portion of Fbxo7   was present in the nucleus (Fig. 5A). In live U2OS cells, Fbxo7 also localized predominantly within the cytosol, with a small proportion localized in the nucleus, in agreement with the fractionation (Fig. 5B). The subcellular localization of PI31 was investigated using an antibody generated against the FP domain of PI31. Immunofluorescence assays on fixed U2OS cells demonstrated that the majority of PI31 was perinuclear and colocalized with a lipophilic dye that stains the endoplasmic reticulum (Fig. 5C). We also found that detectable amounts of Fbxo7 and PI31 overlap in merged images indicating they can be localized to discrete subcellular compartments (Fig. 5D). Because Fbxo7 is a subunit of an SCF-type E3 ubiquitin ligase and it co-localizes in immunofluorescence assays and co-immunoprecipitates with PI31, we tested whether reducing Fbxo7 affected PI31 levels. U2OS cells were transfected with control double-stranded (ds)RNA or dsRNA against Fbxo7, as previously described (13), and cell lysates were analyzed by Western blotting for the presence of endogenous Fbxo7 and PI31. PI31 levels, however, were unaffected by decreases in Fbxo7 levels, suggesting that it does not precipitate the ubiquitin-mediated degradation of PI31 (Fig. 5E).

DISCUSSION
To understand how F-box proteins engage SCF substrates and other potential regulatory proteins through protein-protein interaction, we focused our analysis on the Fbxo7 subunit of the SCF Fbxo7 . This SCF-type E3 ubiquitin ligase is reported to promote the ubiquitin-mediated degradation of HURP and c-IAP (17,18), and the Fbxo7 subunit acts as a specific enhancer of cyclin D-Cdk6 complexes (13,35). To identify further putative substrates and protein-interactions, we combined structural, biochemical, and genetic approaches to study regions of Fbxo7 required for protein interaction. Previously published data implicated the C-terminal PRR as being required for binding to HURP and c-IAP (17,18). We find that two separate epitopes on Fbxo7 are required for its in vivo interaction with Cdk6, one within amino acids 129 -169 and the second spanning the C-terminal PRR. This suggests that Fbxo7 has a bipartite interaction with Cdk6.
We have identified a globular domain of ϳ150 amino acids, defined here as the FP domain, located between the Ubl and F-box of Fbxo7. Using affinity purification coupled with mass spectrometry and yeast two hybrid techniques, we separately identified PI31, a regulator of proteasome assembly, as interacting with Fbxo7⅐Skp1and also having an FP domain. These two proteins have not been previously linked. Furthermore, PI31 and Fbxo7 share not only an FP domain in common but also a highly conserved R(Ar)DP motif embedded near the prolinerich C terminus. Fbxo7 appears to be an elaborated variant of PI31, possessing additional domains (Ubl and F-box), which presumably provide additional functionalities through regulatory and protein interaction sites.
The PI31 FP domain structure reveals an ␣/␤-fold with no close structural relatives to date. The domain associates as a dimer both in solution and within crystals. By structure-guided mutation of surface residues, we found that the PI31 FP domain homodimerizes through a predominantly helical interface and heterodimerizes with the Fbxo7 FP domain through contacts between their ␤ sheets. Previous studies from McCutchen-Maloney (21) showed that full-length PI31 forms homodimers with an apparent K d of 6.25 M. Our determination of a K d of 4.85 M for the isolated FP domain suggests that PI31 dimerization is mediated primarily through its FP domain. Size exclusion chromatography of cell lysates indicated that most of the PI31 in cells was present in complexes of ϳ60 kDa, which is consistent with dimer formation (data not shown). Whether the dimeric organization of PI31 is important for its inhibitory function of immunoproteasome assembly is unclear at present.
We also demonstrate that Fbxo7 and PI31 heterodimerize via a ␤-sheet surface on their FP domains and with an apparent affinity constant of 0.37 M. In vivo, only a fraction of the two proteins occupy the same subcellular compartment. This did not exclude the possibility that a proportion of either protein could associate transiently and may lead to ubiquitin-proteasome system-mediated degradation. We, therefore, tested whether reducing Fbxo7 protein expression levels by RNA-mediated interference affected the levels of PI31 protein; however, no changes were observed, which argues against PI31 being a direct substrate for an SCF Fbxo7 E3 ligase. Alternatively, this interaction may be relevant in different cell types or under different growth conditions (e.g. cell stress or viral infection).
Both Fbxo7 and PI31 associate as homodimers. This oligomeric state may be functionally relevant. For example, other F-box proteins, such as Fbw7/Cdc4, ␤-TrCP, and Pop1/Pop2 homologues, have D box domains that mediate homodimerization, and the structure of one of these domains has been recently solved (36). D boxes are small 45-amino acid motifs closely juxtaposed to the F-box domain, analogous to the FP domain of Fbxo7. Dimerization of SCF Cdc4 has been shown to be required for its in vivo function and may affect efficient ubiquitin chain elongation rather than substrate binding per se (10,36). Similarly FP domain-mediated dimerization of SCF Fbxo7 might also be required for E3 ligase function possibly by enabling cyclinD-Cdk6 complex formation. The overlap between the PI31 binding site on Fbxo7 and Fbxo7 homodimerization surface suggests an intriguing possibility that PI31 could modulate SCF Fbxo7 function by antagonizing Fbxo7 homodimerization, a model we are actively exploring.
This study and others have presented evidence that the C terminus of Fbxo7 mediates binding to its substrates (17,18), and the ability of PI31 to bind and inhibit proteasome activity has also been mapped to the C terminus (21). Within this region is the most highly conserved sequence between the two proteins, the R(Ar)DP motif. Although its function is undetermined, the R(Ar)DP motif in PI31 partly overlaps a PPGXR consensus binding site. In fact, PI31 has two such consensus sites in its C terminus. These act as binding sites for GYF domains, which facilitate protein interactions with proline-rich sequences (37). Fbxo7, however, does not contain the PPGXR motif, suggesting the R(Ar)DP derives from a common ancestor of Fbxo7 and PI31 before their divergence to evolve different biological roles. Furthermore, Ubl domains, such as that found at the N terminus of Fbxo7, have also been reported to bind to ubiquitin receptors, including subunits of the proteasome (16), providing a possible functional link with PI31 as both proteins are capable of directly interacting with the proteasome (19 -21).
In conclusion, we have mapped regions required for proteinprotein interaction within Fbxo7 and identified a novel dimerization domain that is also present within the structural-related PI31. The proposed link between Fbxo7 and PI31 at both a structural and functional level is further consolidated by the observed direct interaction between these proteins. Site-specific substitutions within both proteins ablate this interaction, and this knowledge can be exploited in vivo to probe the function of these proteins, their interaction, and the functional significance of their respective dimeric arrangements.