Ubiquitin Recognition by the Ubiquitin-associated Domain of p62 Involves a Novel Conformational Switch*

The p62 protein functions as a scaffold in signaling pathways that lead to activation of NF-κB and is an important regulator of osteoclastogenesis. Mutations affecting the receptor activator of NF-κB signaling axis can result in human skeletal disorders, including those identified in the C-terminal ubiquitin-associated (UBA) domain of p62 in patients with Paget disease of bone. These observations suggest that the disease may involve a common mechanism related to alterations in the ubiquitin-binding properties of p62. The structural basis for ubiquitin recognition by the UBA domain of p62 has been investigated using NMR and reveals a novel binding mechanism involving a slow exchange structural reorganization of the UBA domain to a “bound” non-canonical UBA conformation that is not significantly populated in the absence of ubiquitin. The repacking of the three-helix bundle generates a binding surface localized around the conserved Xaa-Gly-Phe-Xaa loop that appears to optimize both hydrophobic and electrostatic surface complementarity with ubiquitin. NMR titration analysis shows that the p62-UBA binds to Lys48-linked di-ubiquitin with ∼4-fold lower affinity than to mono-ubiquitin, suggesting preferential binding of the p62-UBA to single ubiquitin units, consistent with the apparent in vivo preference of the p62 protein for Lys63-linked polyubiquitin chains (which adopt a more open and extended structure). The conformational switch observed on binding may represent a novel mechanism that underlies specificity in regulating signalinduced protein recognition events.

The p62 protein (encoded by the sequestosome 1 (SQSTM1) gene) functions as a scaffold in signaling pathways downstream of the interleukin-1, tumor necrosis factor-␣, nerve growth factor, and receptor activator of NF-B 3 receptors, which ultimately lead to activation of the NF-B transcription factor, with receptor activator of NF-B signaling being a critical determinant in the regulation of osteoclast formation (1)(2)(3). Mice that are deficient in p62 show no obvious skeletal phenotype under normal conditions but exhibit defective osteoclastogenesis when challenged with bone-resorbing factors (4). Moreover, mutations affecting p62 are a common cause of Paget disease of bone (PDB), a condition associated with increased osteoclast and osteoblast activity (5)(6)(7)(8). PDB is characterized by excessive bone turnover leading to bone expansion, structural weakness, deformity, and pain (9,10).
The p62 protein has a domain structure consistent with its participation in multiple signaling complexes, although p62 also appears to be multifunctional, not least in controlling protein recruitment to endosomes (3) and proteasomal proteolysis (11). Within the p62 sequence, an N-terminal PB1 domain has been identified that binds atypical protein kinase C. In addition, a ZZ motif is evident, a binding site for the RING finger protein TRAF6, and two PEST sequences that lie adjacent to the C-terminal ubiquitin-associated (UBA) domain (Fig. 1), a motif that occurs in enzymes of the ubiquitin (Ub) conjugation pathway and in regulatory proteins involved in Ub-dependent proteolysis (12)(13)(14). Accordingly, p62 has been shown to bind noncovalently via its UBA domain to mUb and Lys 48 -or Lys 63 -linked polyUb chains (11,15,16). All of the PDB mutations identified to date cluster within or close to the UBA domain of the p62 protein, suggesting that the disease may involve a common mechanism related to alterations in the Ub-binding properties of p62 (17)(18)(19).
To rationalize the structural basis for mUb/polyUb chain recognition by p62, we report NMR structural and binding studies of a recombinant polypeptide corresponding to the C-terminal UBA domain of the human protein (Fig. 1). The isolated 50-residue polypeptide sequence forms a compact three-helix bundle with a structure analogous to the UBA domains of hHR23A and the related CUE domain of the yeast Cue2 protein, but with differences in the loop regions connecting the helices (20). Other UBA and CUE domains have been shown to bind noncovalently with low affinity to mUb through a conserved hydrophobic patch (21)(22)(23)(24)(25)(26)(27)(28), but in some cases with much higher affinity to Lys 48 -linked diUb and longer chains (29,30). We describe NMR complexation studies with the 15 Nlabeled p62-UBA domain and have mapped the interaction surface of the p62-UBA domain with both mUb and Lys 48 -linked diUb. The UBA domain in complex with mUb undergoes global chemical shift perturbations on binding, which we show results from the p62-UBA polypeptide undergoing a novel structural reorganization and repacking of the three-helix bundle. The structural rearrangement results in a large activation barrier to complexation that leads to atypical slow exchange kinetics between free and bound forms on the NMR chemical shift time scale, followed by a secondary fast Ub binding step that enables the identification of the binding surface on the rearranged UBA domain. Upon extending our NMR investigations to study the interaction of the UBA domain with Lys 48 -linked diUb, we find no evidence for linkage-dependent binding specificity, suggesting preferential binding to single Ub units, which may rationalize the apparent in vivo preference of p62 for the more open and extended Lys 63 -linked polyUb chains (11).

EXPERIMENTAL PROCEDURES
Protein Expression and Purification-The cDNA encoding the UBA domain of human p62 was amplified from the IMAGE clone 2906264 by PCR and cloned between the BamHI and XhoI sites of plasmid pGEX-4T-1 (Amersham Biosciences). All mutations were introduced by site-directed mutagenesis (QuikChange mutagenesis kit, Stratagene) and the constructs verified by sequencing. The GST-UBA fusions were expressed by growing transformed BL21 (DE3) Escherichia coli cells in LB at 30°C until the absorbance at 550 nm reached 0.7 and then inducing with isopropyl 1-thio-␤-D-galactopyranoside to a final concentration of 300 M for 20 h. The pelleted cells were stored at Ϫ20°C until required. The cells from 2 liters of growth were resuspended in 8 ml of lysis buffer (10 mM Tris, 150 mM NaCl, pH 7.5) and sonicated (MSE Soniprep 150). The lysate was clarified by centrifugation (40,000 ϫ g, 30 min) before the supernatant was applied to a 1-ml gravity column of glutathione-Sepharose 4B (GE Healthcare) pre-equilibrated with lysis buffer. The column was tumbled at 4°C for 2 h before the supernatant was allowed to flow through. The column was then washed with 10 column volumes of lysis buffer and 5 column volumes of cleavage buffer (20 mM Tris, 2.5 mM CaCl 2 , 150 mM NaCl, pH 8.4). 5 units of thrombin (Sigma) were added in 1 ml of cleavage buffer, and the column was tumbled at 4°C overnight. The column was eluted and washed with a further 2 aliquots of 0.5 ml of cleavage buffer. EDTA was added to a final concentration of 10 M.
The cDNA of yeast Ub was cloned into pKK233-3 between EcoR1 and HindIII sites. mUb was expressed by growing transformed BL21 (DE3) E. coli cells in M9 minimal media at 30°C, as described above. The cells from 4 liters of growth were resuspended in 50 ml of lysis buffer (250 mM acetic acid, 10 mM EDTA, pH 4) and sonicated (MSE Soniprep 150). The lysate was clarified by centrifugation (40,000 ϫ g, 30 min) before the supernatant was diluted 1 in 4 and loaded onto a SP-Sepharose column pre-equilibrated with 50 mM acetic acid, 2 mM EDTA, pH 4. A gradient was run to 2 M NaCl with mUb eluting at 40%. The mUb-containing fractions were concentrated by lyophilization.
All samples were subject to a final gel filtration step with a Superdex 200 (Amersham Biosciences) pre-equilibrated with 30 mM potassium phosphate, 100 mM NaCl, pH 7. The fractions containing the pure protein of interest, as determined by SDS-PAGE analysis, were concentrated by lyophilization and buffer exchanged into water using a desalt column (Amersham Biosciences) before lyophilization. Uniformly 15 N-labeled UBA and mUb were obtained by growing the bacteria in M9 minimal medium containing 1 g/liter of 15 N-labeled ammonium chloride (Isotec) and purified as described above. Doubly 13 C/ 15 Nlabeled UBA domain was prepared similarly with [ 13 C]glucose as the sole source of carbon in the same minimal medium containing 2 g/liters of [ 13 (1 mM) in the presence of 20 M Ub-conjugating enzyme E2-25K (UbcH1) and 0.1 M human Ub-activating enzyme in 50 mM Tris-Cl, pH 8.0 buffer, containing 5 mM MgCl 2 , 10 mM creatine phosphate, 0.6 units/ml inorganic pyrophosphatase, 0.6 units/ml creatine kinase, 2 mM ATP, and 0.5 mM dithiothreitol (31). Reactions were performed at 37°C for 4 h, and the diUb was purified by Q-Sepharose (removal of Ub-activating/Ub-conjugating enzymes) and SP-Sepharose (separation of differing length Ub chains) ion exchange chromatography. Presence of purified diUb was confirmed by 10% SDS-PAGE and protein concentration estimated by absorbance at 280 nm.
NMR Experiments-The NMR assignment of the unbound UBA domain of p62 was achieved using a 1 mM 13 C/ 15 N-labeled protein sample in NMR buffer (50 mM potassium phosphate, 50 mM NaCl, 10% D 2 O, 0.04% sodium azide, pH 7.0) at 298 K and subsequently in the complexed form in the presence of 6 mM Ub. For the free and bound states of the UBA domain 15 N-HSQC, 13 C-HSQC, HNCO, HN(CO)CA, CBCANH, CBCA-(CO)NH, HCCH-TOCSY, 13 C-HSQC-NOESY (mixing time 100 ms), 15 N-HSQC-TOCSY, and 15 N-HSQC-NOESY (mixing time 150 ms) two-and three-dimensional spectra were collected and the spectra processed using XWINNMR version 2.5 (Bruker) and NMRPipe (32). The assignment was carried out in CCPNMR using standard methodologies (33). Backbone 1 D NH residual dipolar couplings were obtained from the difference in 1 J scalar couplings measured from 1 H-15 N IPAP-HSQC spectra (34). The solutions were soaked into predried 6-mm 5% polyacrylamide gels (35,36), which were then compressed into a 5-mm NMR tube (Worldwide Glass Resource Ltd., Cambridge, UK). Two-dimensional spectra were collected with 64 scans and a resolution of 2048 by 1024 data points on a 1 mM 15 N-UBA sample with or without 6 eq of ubiquitin. The spectra for the free UBA domain gave RDC values in the range Ϫ8.3 to 12.2 Hz. Spectra for the complex, collected using the same alignment conditions, showed a narrower range of values (Ϫ7.7 to 4.6 Hz), which appears to correlate with small changes in the apparent size or intrinsic flexibility of the UBA domain in the bound state that affects the degree of alignment (37). These observations are consistent with analytical gel filtration results (not shown), which also suggest changes in the average molecular dimensions of the UBA domain between the free and bound conformations.
Structure Determination from NMR Restraints-The structure determination was based upon NOE distance restraints and dihedral restraints, the former derived from the 13 C-HSQC-NOESY and 15 N-HSQC-NOESY spectra. The NOEs were classified based upon their intensity into very strong (1.8 -2.5 Å), strong (1.8 -2.8 Å), medium (1.8 -4 Å), weak (1.8 -5 Å), and very weak (1.8 -6 Å). In addition, 45 backbone 1 D NH RDC parameters from the 47 HSQC cross-peaks in spectra of both the free and bound UBA domain were used as restraints using the ISAC method (38) and visualized using MODULE (39). TALOS (40) dihedral restraints with Ϯ30°limits were also used in the calculations. An initial 100 structures were produced using the standard three-step XPLOR-NIH 2.14 protocol (40). In the first stage, high temperature Cartesian dynamics was performed at 1000 K with a time step of 0.005 ps, for 20,000 steps, using the Verlet integrator. During the second cooling phase of the protocol, the temperature was reduced from 1000 to 100 K in steps of 50 K, with a time step of 5 fs, over 40,000 steps during which the relative weighting of nonbonded energy terms was increased from 10% of their default values to their force field default (41,42). These initial structures were then refined with another 10,000 cooling steps. The structures were selected based upon NOE violations Ͻ0.5 Å and dihedral violations Ͻ2°. As a further cross-validation step, we attempted to refine the structure of the bound UBA domain with RDC restraints derived from the free UBA domain. This led to significant distortions in the relative orientations of the helices and structures with much higher energies. All structures were displayed using MOLMOL (43). The structural statistics are presented in Table 1. Hydrophobic surface area measurements were performed using the program Naccess (version 2.1.1) (68).
UBA/Ub Binding Studies by NMR-A 1 mM sample of 15 N p62-UBA in NMR buffer was titrated in 20 increments with ratios of mUb to UBA in the range 0.1-6.0 and 15 N-HSQC spectra collected at each step. A second "reverse" titration was performed by titrating unlabeled p62-UBA into a 1.0 mM 15 Nlabeled sample of mUb in NMR buffer at the following ratios of UBA to mUb 0.25, 0.5, 0.75, 1, 1.25, 1.5, 2, 4, 6. All data were collected at 298 K, and changes in chemical shift and cross-peak intensity were monitored by acquisition of 15 N-HSQC spectra. The 15 N-HSQC spectra were acquired with a resolution of 2048 complex points and 112 t 1 increments over a spectral width of 9000 Hz by 2400 Hz in the direct ( 1 H) and indirect dimensions ( 15 N) with 8 scans per t 1 increment. The spectra were processed using a sine-squared apodization function shifted by /2 in both dimensions and zero-filled to 2048 by 1024 real points (F2 by F1).
Ub Binding Assays-WT-UBA domain (residues 387-436) and UBA mutants were assayed for Lys 48 -linked and Lys 63linked polyUb binding as described previously (11,20), using 1 g of polyUb chains, in 50 mM Tris, 0.1% (w/v) bovine serum albumin, pH 7.5, with all reagents maintained at the 37°C throughout the binding/washing stages. After washing, bound proteins were detected by Western blotting (anti-Ub). All experiments were repeated on at least three independent occasions and representative examples of blots are presented.

RESULTS
Structure of the UBA Domain of p62-The NMR structure of the C-terminal UBA domain of p62 (residues 387-436 in fulllength p62) has been determined previously by NMR (20), and further refined using a 13 C/ 15 N doubly labeled construct and additional 1 D NH residual dipolar coupling (RDC) restraints derived from two-dimensional 1 H-15 N HSQC spectra (see structural statistics in Table 1). The RDC data led to an improvement in secondary structure definition and a more precise relative alignment of the three helices (44 -47). The structure has an overall topology similar to other UBA and CUE domains. The structure is well defined between Pro 392 and Ile 431 (with an average backbone r.m.s. deviation to the mean structure of 0.75 Å). NMR relaxation measurements (R 1 , R 2 , and 1 H-15 N NOE), and calculated order parameters, suggest that the helical secondary structure forms a compact fold with rapid dynamics confined to the few disordered residues at the N and C termini and in the first loop (48). The three helical segments are defined by residues Pro 392 -Met 404 , Leu 413 -Lys 420 and Ile 424 -Ile 431 . The domain is compact with a well defined hydrophobic core composed primarily of the buried side chains of Met 401 , Leu 417 , Ala 427 , and Ile 431 (Fig. 1E), which is further consolidated by Leu 398 , Phe 406 , Trp 412 , Leu 413 , Ile 424 , and Leu 428 . The first loop contains the highly conserved Xaa-Gly-Phe-Xaa motif (Met-Gly-Phe-Ser in p62-UBA; residues 404 -407), which together with Asp 408 and Glu 409 form an extended single ␤-strand. However, a major difference between the p62-UBA and other UBA domains is the expansion of this connecting loop through the insertion of two glycines (Gly 410 and Gly 411 ; see Fig. 1B). The lack of NOE contacts from the H␣s of these two glycines to any adjacent residues, together with the analysis of 15 N relaxation rates (R 1 and R 2 ), heteronuclear NOE data, and fluctuations evident in the ensemble of structures sampled by molecular dynamics simulations, shows that Gly 410 and Gly 411 undergo enhanced backbone dynamics and conformational exchange (48). All of the UBA domains so far characterized in structural detail lack the di-glycine segment found in the p62-UBA domain, making loop 1 of other UBA domains considerably shorter and more highly restrained. The di-glycine insertion appears to be structurally important in the p62-UBA because a UBA mutant lacking Gly 410 and Gly 411 appears to be unfolded in solution (not shown). The presence of a proline residue in loop 1 of hHR23A UBA2 (Leu-Gly-Phe-Pro) (22) and in the CUE domain (Met-Phe-Pro) (24) forces a sharper turn. Analysis of surface hydrophobicity identifies a hydrophobic patch in the vicinity of the Met-Gly-Phe loop of the p62-UBA domain, and this appears to be common to many of the UBA domains so far characterized (23)(24)(25).
UBA Interaction with mUb from Chemical Shift Perturbation Mapping-The 1 H-15 N HSQC spectra of isotopically labeled mUb and the UBA domain were assigned using established procedures (49). In particular, we examined NMR parameters for the UBA domain over a wide range of concentrations (0.1-1.2 mM). The ratio of spin-spin (R 2 ) to spin-lattice (R 1 ) relaxation rates and chemical shifts showed no significant dependence on protein concentration at 298 K up to 1.2 mM. Consequently, all structural studies of the UBA domain were performed below this concentration. Initially, unlabeled UBA domain was titrated into a sample of 15 N-labeled mUb to characterize the chemical shift perturbations associated with binding. The free and bound forms of 15 N-mUb were in fast exchange on the chemical shift time scale ( Fig. 2A), characteristic of both fast on-and off-rates for complex formation and dissociation. The weighted average chemical shift perturbations up to saturating concentrations of UBA domain are shown in Fig. 2B with no further changes in shifts evident above a 6:1 excess ratio. The smooth change in chemical shifts was used to construct binding curves from 12 well resolved residues. Fitting these data to a 1:1 binding model yields a K d ϭ 540(Ϯ45) M with the error representing the standard deviation over the 12 data sets (Fig. 2C). This value is comparable with those reported for Mud1 (390 M), for hHR23A (500 -600 M) and for the CUE domain (155 M), demonstrating a relatively weak interaction in all cases (29). The pattern of chemical shift perturbations identifies a Ub binding epitope involving residues Leu 8 , Ile 13 , Arg 42 , Ile 44 , Ala 46 , Leu 69 , Val 70 , Leu 71 , and Arg 72 (⌬␦ Ͼ 0.1 ppm) largely located on the surface of the ␤-sheet in a predominantly hydrophobic patch (Fig. 3A), with Arg 42 and Arg 72 flanking this patch. A number of adjacent residues such as Gln 49 and Gly 53 are also perturbed, whereas Lys 48 is significantly exchange broadened and undetected. This pattern of mUb shifts is very similar to that reported previously for the binding of the UBA domains of hHR23A (30) and Mud1 (29) and also highlighted in studies of the key determinants for endocytosis and proteasomal degradation (50,51), suggesting that mUb uses a highly conserved binding epitope in protein recognition. Many much smaller shift perturbations (Ͻ0.1 ppm) are also apparent, which  suggest some small global structural readjustment occurs to optimize interactions with the UBA domain. Overall the magnitude of the shifts for mUb suggests that it is already largely preorganized for recognition and binding.
To map the binding surface of the UBA domain, we repeated the titration using 15 N-labeled p62-UBA with the addition of up to 6 eq of the unlabeled mUb. In contrast to the reverse titration where fast exchange was evident for 15 N-mUb between free and bound forms, we observed HSQC cross-peaks for both the free and bound UBA domain indicative of slow exchange between the two species. The mid-point of the titration is shown in Fig.  4A. Separate cross-peaks could be identified for free and bound forms even where chemical shift differences are comparable with NMR line widths (see Asp 421 and Asn 423 in Fig. 4A), on the basis of which we estimated a slow rate of interconversion of Ͻ10 s Ϫ1 . The slow on-and off-rates for complexation contrast markedly with NMR data for the binding of other UBA domains, including Mud1 and the UBAs of hHR23A where typically fast-on/fast-off rates are evident in titration with mUb (21)(22)(23)29). Furthermore, the magnitude of the chemical shift perturbations for 15 N-p62-UBA appeared to be generally significantly larger than observed for mUb, such that assignment of the bound state was not possible without further detailed analysis. We assigned the spectra of the bound state of the UBA domain using a 13 C/ 15 N-labeled sample titrated to saturation with 6 eq of unlabeled mUb. The plot of weighted CSP for the UBA upon binding mUb reveals a wide distribution of perturbations with a significant number Ͼ0.5 ppm, including Ile 431 which shows a free to bound shift of 3.14 ppm. The shift changes occur on such a global scale (Fig. 4B) that it was necessary to complete a full structural analysis of the bound state (Table 1; see "Experimental Procedures").
Structure of the Bound UBA Domain of p62-The structures of the free and bound states of the UBA domain (Fig. 3, D and E) were generated using the same 13 C/ 15 Nlabeled sample and assignment protocol and are of comparable quality (see Table 1). Loop 1 and the N and C termini show some flexibility in both the free and bound states. The analysis of the bound state reveals that the secondary structure content of the UBA domain is largely conserved; however, a novel realignment and packing of the three helices of the UBA domain is apparent that results in a noncanonical UBA conformation (Fig. 3, F-I). Given that the relative orientations of the helices are different in the two conformations, we used 1 D NH RDC measurements to provide further long range information to validate the change in structure (44 -47). We were able to resolve the majority of 1 D NH couplings (45 of 47 residues in the highly structured part of the UBA domain in both the free and bound states) and observed significant differences in the magnitude and sign of RDC values from the two structures (Fig. 4C). Changes in a number of vectors in helix 1 suggested a structural tightening in the bound state. Intrinsic flexibility is consistent with the observation of rapid NH-ND exchange rates for amide groups in helix 1 of the unbound UBA (not shown). Residues in helix 2 showed a similar alignment of N-H vectors in the free and bound states. However, a quite different overall alignment was evident for residues in helix 3, which also correlated with some of the largest observed changes in chemical shifts between the free and bound states (Fig. 4B). Refinement of the structure of the bound UBA domain against the RDC data produced good agreement between experimental and calculated values (see Table 1 and supplemental Fig. 1). In other reported structural studies, RDC measurements have been used in an analogous fashion to demonstrate both similarities and differences in relative orientations of helices in solution and in x-ray structures (52), for identifying periodicities and conformational distor- tions in helices (53), and in the determination of relative orientations of structural subunits (44).
Although some of the side chain packing interactions between residues in the free and bound conformations of the UBA domain are conserved, a number of residues show quite different structural environments that are evident from unique patterns of side chain NOEs. For example, in the conformation of the unbound UBA domain, Gln 400 (in helix 1) nestles in a hydrophobic pocket surrounded by side chains from loop 1 and helix 3 (Phe 406 , Ile 424 and Leu 428 ; Fig. 3F). In the Ub-bound conformation, Gln 400 is located in an entirely different pocket defined by residues in helix 2 (Trp 412 , Leu 416 , and Leu 417 ; Fig.  3G). By analogy, the side chain of Trp 412 in helix 2 shares many common contacts in both structures with residues close in sequence within loop 1 and helix 2. However, there are again some clear differences. The indole ring in the structure of the unbound UBA domain forms unique contacts with residues in loop 2 and helix 3 (Leu 417 , Asp 423 , Ala 427 , and Asp 430 ), and in the bound state unique contacts with helix 1 (Leu 398 , Gln 400 , and Met 401 ). The differences in the pattern of NOE interactions are summarized in the contact map shown in supplemental Fig. 2.
Furthermore, the 3.14 ppm shift in the proton resonance of the NH of Ile 431 is also accounted for by the structural reorganization. The Ile 431 NH is in close proximity to the indole ring of Trp 412 and experiences a substantial ring current shift in the free UBA. However, this close contact is completely lost in the bound conformation, as are contacts between the side chain of the adjacent residue Gln 432 , which occur with Phe 406 . In both cases, large differences in 1 H chemical shifts in the free and bound states are apparent. A schematic representation of the two structures is shown in Fig. 3H; the helices are represented as cylinders and the two structures are shown with a common alignment of helices 1 and 2. The overlaid structures shown in Fig. 3H show that the differences can be interpreted as the repacking of helix 3 against the other two helices. Complex

REACTION 1
Here, UBA is the unbound conformation, and UBA* is the structure that we observe in the bound state with mUb. We were unable to detect a significant population of UBA* at equilibrium in HSQC spectra, suggesting that in the absence of mUb, [UBA]/[UBA*] Ͼ10. To simplify the model, we have assumed that the UBA* in isolation has the structural characteristics found in the UBA*⅐mUb complex, with binding occurring through a process of conformational selection. However, we cannot rule out the possibility that UBA* in isolation may be partially unstructured, in which case the interaction with mUb may result in a process of induced fit to give the stable UBA*⅐mUb complex that we observed experimentally. The structural details of UBA* in isolation remain to be firmly established; however, the 1 H-15 N HSQC spectra of 15 N-mUb titrate in a linear manner upon addition of the UBA domain, suggesting that a two-state fast exchange binding process occurs. This is consistent with UBA* possessing conformational features closely akin to those observed in the bound state.
Upon titration of 15 N-UBA with mUb, we see a shift in the equilibrium from UBA to UBA*⅐mUb characterized globally by slow exchange kinetics (slow-on and slow-off rates) (Fig. 4A). However, a subset of resonances from the UBA domain showed clear evidence for further ligand-dependent chemical shift perturbations with a change in the position of HSQC cross-peaks for the bound state as the population of UBA*⅐mUb increases (54,55). These secondary shifts are associated with fast-on/fastoff rates that reflect the population-weighted average between UBA* and the UBA*⅐mUb complex. This is illustrated in Fig. 4D for the cross-peaks of Gly 405 . Cross-sections are shown through two-dimensional HSQC spectra at different UBA:Ub ratios. The 1 H chemical shift for the free UBA occurs at 7.85 ppm. The peak for the bound state first appears at 7.64 ppm and gradually shifts to higher field as the population of the UBA*⅐mUb complex increases, reaching the fully bound shift at 7.58 ppm. In contrast, titration of 15 N-labeled mUb with unlabeled UBA enabled us to observe only the fast exchange component of the two-step mechanism. The effects on NMR line shapes of this two-step binding process (shown in Reaction 1) have been reported in analogous ligand binding studies with retinol-binding protein using both experimental and simulated data (54), and in the folding-induced binding of the phosphorylated kinase-inducible activation domain in complex with the KIX domain of the cAMP-response element-binding protein (55).
The secondary fast exchange shift perturbations for 15 N-labeled UBA identify the specific residues of UBA* that are involved in complexation with mUb. The perturbations are most significant for residues Gln 400 , Leu 402 , Ser 403 , Met 404 , Gly 405 , Phe 406 , Ser 407 , Asp 408 , Ala 426 , Ala 427 , Asp 429 , Thr 430 , Gln 432 , Tyr 433 , Ser 434 , and His 436 with the side chains of Trp 412 and Gln 400 also affected. A significant line broadening effect is evident for Gly 410 within the di-glycine insertion, which suggests conformational exchange processes taking place around this loop. When the location of all of these residues is mapped on to the UBA domain, they form a contiguous binding surface involving residues at the C terminus of helix 1, within loop 1, and along helix 3 (Fig. 3C). This structural reorganization appears to optimize both surface hydrophobicity and van der Waals complementarity between the two protein surfaces. Hydrophobic surface area measurements suggest that overall there is no great difference between the free (60%) and bound (59%) states, but the surface localization does change. Around loop 1 the side chains of Phe 406 and Trp 412 both become significantly more surface-accessible in the bound conformation, generating a slightly larger hydrophobic patch. In addition, the charged residues Glu 396 , Asp 408 , and Glu 409 are brought together adjacent to the hydrophobic patch to allow for the possibility of electrostatic interactions with basic residues on the surface of mUb. Indeed, Arg 42 and Arg 72 , which are colocalized on the surface of mUb (Fig. 3A), undergo chemical shift perturbations upon binding (Fig. 2B) consistent with an electrostatic component to the binding interaction. To guide modeling studies of the complex, we attempted to identify intermolecular NOES using 13 C/ 15 N-filtered NOESY experiments using 13  Polyubiquitin Chain Recognition by p62-UBA-The interaction properties of minimal isolated UBA domains show that they divide empirically into four classes that reflect linkageselective chain recognition by binding Lys 48 -linked chains, by binding Lys 63 -linked chains, by showing no linkage-dependent discrimination, or by demonstrating an inability to bind Ub chains at all (57). It is well established that Lys 48 -linked chains are used to signal proteasomal proteolysis, whereas the Lys 63 linkage is used in a number of nonproteolytic signaling pathways, and these linkages represent two of those most commonly found in vivo. Structural studies of polyUb chains (Ub 2 , Ub 3 , and Ub 4 ) show that there are linkage-dependent interactions between Ub motifs that are likely to have some bearing on chain recognition by other binding partners (58,59). Lys 48 -linked chains form a more compact conformation in which the hydrophobic surface defined by Leu 8 , Ile 44 , and Val 70 , and identified as forming the UBA binding surface, are involved in interfacial contacts between both proximal and distal Ub molecules. This intramolecular self-association process must be competitive with binding other low affinity substrates, as illustrated schematically in Fig. 5A. Estimates from Varadan et al. (30) suggest that the equilibrium between open and closed states has a K eq ϳ6 M Ϫ1 in favor of the closed state at physiological pH. In contrast, Lys 63 -linked chains appear to be more extended with Ub motifs presented as beads on a flexible string.
The minimal p62-UBA domain has been suggested to be nonselective for polyUb chain linkage (57), although one study in assessing the role of full-length p62 protein in polyUb chain aggregate formation and cell survival indicated a clear preference for Lys 63 -linked chains (11). In contrast, earlier work also indicated a role for p62 in Lys 48 -linked polyUb recognition (15). Other studies have also implicated both intra-and intermolecular protein-protein interactions in modulating these recognition properties in full-length hHR23A (60, 61) and possibly p62 (11). We examined the binding properties of the 15 N-UBA domain in the presence of Lys 48 -linked diUb. The slow exchange between free and bound forms of the UBA domain evident in the titration with mUb was again clearly apparent. The similarity in the CSPs was such that the assignments were readily transferred between complexes. The weighted CSPs for mUb and Lys 48 -diUb binding to the 15 N-UBA are shown in Fig.  5B. The CSP difference between the two data sets shows a random distribution of small effects that demonstrate that the two titrations yield practically identical HSQC spectra of the complexes. This is in contrast to recent studies with the UBA domain of Mud1 (29), where differences between the binding of mUb and Lys 48 -diUb were clearly apparent with 0.1-0.3 ppm perturbations identifying further interactions with a second hydrophobic patch involving mainly residues within helix 2 but also Phe 330 in helix 3. The latter is conserved in the UBAs of Mud1 and hHR23A and both have been shown to form sandwich complexes with Lys 48 -diUb (29,30). These secondary interactions appear to rationalize the marked enhancement in affinity (Ͼ100-fold increase in the case of Mud1) for Lys 48 -diUb and significant line broadening effects associated with slower on-and off-rates for complex formation and dissociation.
More significantly, our NMR titration studies with p62-UBA and Lys 48 -linked diUb reveal that the binding affinity for the Ubiquitin Recognition by the p62-UBA Domain FEBRUARY 29, 2008 • VOLUME 283 • NUMBER 9 JOURNAL OF BIOLOGICAL CHEMISTRY 5433 latter is lower than for mUb. We were unable to reach a fully populated bound state of the UBA with diUb even though binding saturation was readily achieved under similar experimental conditions in the presence of mUb. With the free and bound forms in slow exchange, an estimate of the relative intensities of cross-peaks in the 1 H-15 N-HSQC spectrum during the titration enabled us to calculate an apparent K d value for the UBA⅐diUb interaction that is ϳ4 times higher than for UBA⅐mUb. The HSQC cross-peak intensities for the free and bound states for Leu 423 during the titrations of 15 N-UBA with mUb and with Lys 48 -linked diUb are shown in Fig. 5C. In each case the ratio of UBA:mUb and UBA:diUb in the titration is 1:4. Under these conditions, the UBA domain is almost fully bound with mUb (solid lines); however, the UBA⅐diUb complex is only partially populated (dotted lines) with the signal from the unbound UBA still in excess. This reduction in UBA affinity is consistent with the competitive effects on binding of a significant population of the "closed" structure of Lys 48 -linked diUb under the conditions of the titration (Fig. 5A). Our estimate of a 4-fold reduction in binding affinity for diUb by the p62-UBA is in good agreement with the self-association constant for Lys 48 -linked diUb calculated by Varadan et al. (58), suggesting that the competitive self-association property of the distal and proximal Ub domains in Lys 48linked diUb is able to account for the difference in affinities.

Effects of Mutations in the UBA Domain on Ubiquitin Binding
Affinity-We were unable to detect a secondary binding site within helix 2, or elsewhere, on the basis of comparisons of binding-induced shifts for mUb versus diUb, and conclude that there are no Lys 48 -linked diUb-specific interactions with the p62-UBA domain. We further examined our model on the basis of the effects of point mutations on Ub binding affinity. Mutations to Ala at four solvent-exposed sites within helix 2 (T414A, R415A, Q418A, and T419A; Fig. 6A) failed to produce any significant reduction in the affinity for polyUb chains in pulldown assays, consistent with the absence of primary or secondary binding sites within helix 2 in wt-UBA. However, the data for T414A at 4°C provided evidence for enhanced binding to mUb and polyUb chains (data not shown), suggesting the possibility that certain residues may have a role in negatively regulating binding specificity and affinity, as also recently proposed for a similar T1374A mutation in the Ede1 UBA domain (62). The results from our helix 2 mutants contrast with those of Seibenhener et al. (11) who demonstrated in a similar pulldown assay that the conservative L417V mutation, which is sandwiched between our own R415A and Q418A helix 2 substitutions, largely eliminated binding, but L413V had little effect. Interestingly, in our structure of the bound state, Leu 417 is sub-  stantially buried in the hydrophobic core of the UBA domain, suggesting the possibility that the effects of this mutation are mediated not by direct effects on interfacial interactions with mUb but by specific effects that destabilize the bound conformation (UBA*). Indeed, CD spectra of the L417V mutant show that the unbound structure is native-like (11) and unperturbed by the mutation. In contrast, mutations to T414A, R415A, Q418A, and T419A, which are also in helix 2, appear to have little effect on binding affinity and presumably the stability of the UBA* conformation. Selective destabilization of UBA* may also be a plausible explanation for the loss of binding affinity reported for the I431V helix 3 mutant. Although we see no secondary chemical shift perturbations to Ile 431 for wt-UBA on binding mUb, a number of adjacent residues (Thr 430 and Gln 432 ) are affected, suggesting that Ile 431 is close to the primary binding site on the UBA domain but is involved in core packing interactions rather than intermolecular contacts.
The effects of PDB mutations within the isolated UBA domain on Lys 48 -linked polyUb binding affinity have been reported for P392L, M404V, G411S, and G425R (17). These studies have been extended to include new data for P387L, S399P, and M404T (Fig. 6B) and data for all mutants binding to Lys 63 -linked polyUb chains (Fig. 6C). Although at best semiquantitative in nature, pulldown assays do not reveal any significant linkage-specific differences (Lys 48versus Lys 63 -linked Ub chains) in the apparent binding affinity for wt-UBA or for the mutant UBA domains (Fig. 6, B and C). Our results, and those reported by Seibenhener et al. (11), are summarized in Fig. 7. The M404V, M404T, and F406V mutations in loop 1 occur in the region where we observe the largest chemical shift perturbations on binding. Loss of affinity associated with these mutations is readily rationalized in terms of direct perturbations to (or close to) the Met-Gly-Phe-Ser loop sequence. Mutations within the flexible N terminus and within helix 1 (P387L, P392L, L398V, and S399P) have little effect on binding, consistent with chemical shift mapping experiments that place this part of the structure away from the binding interface. Notably however, the P387L, P392L, and S399P mutations do exert effects on the ability of the full-length p62 to bind mUb (19), implicating other components of p62 in the subtleties of Ub recognition in vivo.
We have shown that the G425R PDB mutation results in a significant loss of binding affinity in both the isolated UBA and the full-length protein (17)(18)(19). Located at the N terminus of helix 3, Gly 425 appears to be peripheral to the UBA binding surface identified from NMR chemical shift mapping experiments. Interestingly, the substitution of a Gly residue for the bulky and highly polar guanidinium group of Arg increases the stability of the UBA domain toward thermal denaturation (⌬⌬T m ϳ 9°C) as measured from CD melting curves (data not shown). Thus, the G425R mutation presents the intriguing possibility of reducing the binding affinity for mUb by stabilizing the conformation of the unbound state of the UBA and thereby further diminishing the population of UBA* at equilibrium.

DISCUSSION
Ubiquitin Binding Specificity of the p62-UBA Domain-The UBA domain sequence of the p62 protein, and those derived from other UBA domains whose complexes have been characterized structurally, including Mud1, Ede1p, UBA1, and UBA2 of hHR23A and p47, are shown schematically in Fig. 7. In each case, residues whose NMR chemical shifts are significantly perturbed on binding mUb are shown (see Fig. 7, *). The consensus suggests that residues at the end of helix 1, those within the Met-Gly-Phe-Ser loop 1 region (or equivalent), and within helix 3 form the primary binding site for mUb. In a number of cases, perturbations are observed within helix 2, which are attributed to indirect effects arising from some degree of structural rearrangement on binding. However, more recent structural investigations with the UBA2 domain of hHR23A and the UBA domain of Mud1 have demonstrated high affinity binding to Lys 48 -linked diUb through formation of a sandwich complex (Fig. 5A), in which a single UBA domain uses different binding faces to form specific contacts simultaneously between both the distal and proximal Ub motifs, thereby enhancing the binding affinity Ͼ100-fold compared with the interaction with mUb (29,30).
The absence of any significant binding-induced differences in chemical shift perturbations in complexation studies with mUb and Lys 48 -linked diUb suggests that the p62-UBA domain does not simultaneously recognize the distal and proximal Ub motifs of Lys 48 -linked diUb using two binding faces. To rationalize this behavior, the conserved residue Phe 330 in helix 3 is found in the UBAs of Mud1 and hHR23A and has been implicated in specific recognition of Lys 48 -linked diUb. However, Thr 430 is found at the equivalent position in the p62-UBA and constitutes part of the primary mUb-binding site (see Fig. 7). More generally, only a subset of UBA domains show conservation of Phe 330 , suggesting that selective recognition of Lys 48linked polyUb chains may be confined to this group of UBAs (29), from which the p62-UBA domain is excluded.
The effects on binding affinity of a significant number of p62-UBA domain mutations, several of which cause PDB, have provided further probes for examining the conformational equilibrium between the free and bound states. The effects of a number of mutations appear to be rationalized on the basis of perturbations to specific intermolecular contacts in the bound state (M404V/M404T and F406V). However, others appear to have the potential to (i) stabilize the structure of the UBA domain in its nonproductive (free) conformation (G425R), or (ii) destabilize the productive (bound) UBA* conformation, because of differences in side chain packing interactions between the free and bound structures (L417V and I431V, from the work of Seibenhener et al. (11)). The observations suggest the possibility of both direct and indirect read-out mechanisms for the effects of mutations on Ub chain recognition that we are investigating further.
In this study we have shown that binding of the p62-UBA domain to mUb is achieved by a novel structural rearrangement of the three helices, which preserves essentially the same elements of secondary structure but optimizes the binding surface for interaction with mUb. By analogy, NMR and x-ray structures of the CUE domains bound to mUb have demonstrated the possibility of distinctly different mechanisms of Ub recognition, in one case through a monomeric compact helix bundle that is structurally homologous to the UBA domains (Cue2 from yeast) (24), and in the other (Vps9p exchange factor) through formation of a novel interlocked dimer involving reciprocal exchange of helix 3 (63). The latter achieves high affinity mUb binding by increasing the interaction surface through dimerization, demonstrating the versatile use of a small motif in achieving binding specificity. Analytical ultracentrifugation studies suggest a monomer-dimer equilibrium in solution with a K d for dimerization of ϳ1 mM, with conformational selection in the presence of mUb favoring the dimer form (63).
It is possible however that the rearrangement of the p62-UBA described in this work is not a mechanism for optimizing Ub binding affinity per se but for achieving binding specificity. In the free form the basic residues on the UBA surface are dis- persed, whereas in the bound form there is a basic patch formed by residues Arg 393 , Arg 415 , and Lys 420 . Similarly, an acidic surface evident in the free form of the UBA domain is further consolidated in the bound state, running from the N-terminal end of helix 2, across the whole of helix 3, and around to helix 1. It is a possibility that the rearrangement of the UBA domain serves to expose (or conceal) other binding sites. Indeed, there is a precedent for the interaction of the p62-UBA with nonubiquitin targets, for example MuRF2, although the binding interface has not been characterized (64).
A recent structural analysis of the Par-1/MARK protein kinase has demonstrated interactions between the hMARK3 UBA domain and the kinase domain (65). The monomeric UBA domain is unusual in several intriguing ways. It adopts a noncanonical UBA fold, in which helix 3 is inverted, and appears to be in equilibrium with a partially disordered state that results in extensive line broadening associated with conformational exchange. The hMARK3 UBA binds mUb only weakly (K d ϳ2.35 Ϯ 0.5 mM), despite having the conserved Met-Gly-Tyrbinding motif (␣1-␣2 loop) common to many mUb-binding UBA domains (23)(24)(25), including the p62 UBA. Helix 3 of hMARK3 UBA appears to remain largely disordered in the complex with mUb, implying that both its structure and conformational instability may be important in regulating the binding affinity for mUb and in promoting intramolecular interactions with its catalytic kinase domain (65). These studies, and those with the p62 UBA reported here, demonstrate that the UBA domain is a versatile structural motif for which molecular recognition and binding specificity may depend on very different binding properties.
Biological Significance of the Model-We have demonstrated from NMR titration studies and from an analysis of chemical shift perturbations that the isolated UBA domain of p62 binds Lys 48 -linked diUb with marginally lower affinity than monomeric Ub units. However, this does not preclude lower affinity binding of p62 to Lys 48 -Ub chains as being physiologically relevant (15). Observations appear to suggest an apparent preference within cells for the binding of the full-length p62 to Lys 63 -linked chains (11), which adopt a more open extended conformation, analogous to mUb units assembled in a beads-on-a-string arrangement. Our NMR results qualitatively support this model by demonstrating marginally higher affinity for mUb units. The in vitro pulldown assays shown in Fig. 6 show little discernible difference in binding affinity for Lys 48versus Lys 63 -linked chains, which suggests that any selectivity is at best marginal. In contrast to the NMR titration studies, diUb and longer Ub chains are retained with higher affinity on UBA-loaded Sepharose beads in apparent contradiction of the NMR binding affinity experiments. A plausible explanation for this discrepancy is that a chelate effect operates on the surface of the bead to artificially enhance the affinity of ubiquitin chains through the interaction with multiple UBA binding partners. Any subtle differences in affinity of an individual UBA domain for Lys 48 -or Lys 63 -Ub chains are likely to be masked by this effect.
The full-length hHR23A provides a clear example where the intramolecular interactions of subdomains (UBL-UBA1) play a key role in modulating Ub linkage selectivity (60,61). In in vitro pulldown assays with Sepharose-conjugated mUb, we were able to bind full-length p62 with higher affinity than the isolated UBA domain (19). These findings provide further indications that other non-UBA domains/sequences within the full-length p62 can influence Ub recognition. Indeed, when certain PDB mutations that do not affect Ub binding in the isolated UBA domain are introduced into the full-length protein, we see a common loss of ability to bind Ub relative to wild-type p62 UBA at physiological temperatures (19). The presence of flexible unstructured regions proximal to compact UBA domains appears to be a common feature of UBA-containing proteins, implying some regulatory role that we are currently investigating for p62. It has also been proposed that different oligomerization states of p62 could be influential in imparting specificity for mUb versus polyUb chains (11), with a possible chelate-like enhancement of binding affinity available to the latter through multiple p62-UBA domains displayed on a single binding surface. This mechanism would appear to have certain similarities to the binding of polyUb chains to UBA-loaded Sepharose beads where we see preferential retention of the longer chains. This is partially supported by evidence for formation of dynamic aggregates in multimeric signaling complexes (66,67). It will be of particular interest to note whether the conforma- representation showing the UBA domain sequence of the p62 protein and other UBA motifs that have been characterized by NMR in complexation studies with Ub, including Mud1, Ede1p, UBA1, and UBA2 of hHR23A and p47. In each case, residues whose NMR chemical shifts are significantly perturbed on binding Ub are shown (*). The consensus suggests that residues at the end of helix 1, within the helix 1-helix 2 connecting loop region and within helix 3, form the primary Ub-binding site for mUb. In many cases perturbations within helix 2 appear to correlate with indirect effects arising from structural rearrangement. B, secondary structure is shown for the p62-UBA sequence as determined by NMR. The p62-UBA mutation sites so far studied from our own work (f) and that of Seibenhener et al. (11) (F) are shown. The corresponding open symbols (Ⅺ, E) represent the residues that appear to significantly reduce the affinity for binding polyUb chain mixtures in pulldown assays involving immobilized UBA domains (as shown in Fig. 6).
tional switch in the p62-UBA domain reported in these studies is in any way influenced by signaling complex assembly. As the p62-UBA has been shown to have binding partners other than Ub (64), this rearrangement could be a way for achieving Ub modulation of p62 binding partners without the need for covalent attachment of Ub. This would be an attractive biological mechanism for the explanation for the participation of p62 in multiple signaling complexes, with the possibility that other proteins may achieve specificity in binding to the UBA domain by recognizing different conformational states. There is also scope for binding to the p62-UBA domain to inhibit the Ubinduced rearrangement, perhaps serving as a negative regulator of p62-mediated signaling.