To examine the importance of Ub-binding sites 1 and 2 to the ability of ataxin-3 to cleave poly-Ub chains independently from the C-terminal UIMs, we carried out enzymatic Ub cleavage studies using the isolated Josephin domain (amino acids 1–182). We followed established protocols [14], [18]. In parallel we examined the cleavage properties of wild-type Josephin versus two mutants (Jos_I77KQ78K and Jos_W87K) that have been shown to individually abolish binding to site 1 or site 2, respectively [17] (Fig. 1A). Jos_I77KQ78K mutates two exposed residues that are close to the active site and are in the interface with the Ub molecule which binds in site 1. Jos_W87K mutates an exposed hydrophobic tryptophan found to be essential for binding both Ub and the Ubl of hHR23B to site 2 [18].

Our results show that Ub-binding site 1 is necessary for the Josephin domain to cleave K63-linked hexa and K48-linked pentaUb chains, as well as mixed-linkage chains (Fig. 1B,C,D). In contrast, site 2 shows a more nuanced involvement in Ub cleavage by the Josephin domain: site 2 is dispensable for cleavage of K63-linked chains (Fig. 1B), but mutating site 2 does lead to a reduction in cleavage of K48-linked and mixed-linkage polyUb chains (Fig. 1C,D).

These data indicate that site 1 is critically involved in any DUB activity of the Josephin domain, and suggest an involvement of site 2 in specific linkage recognition or preference by the Josephin domain.

We next compared our results for the isolated Josephin domain with those obtained for full-length, non-expanded (Q22) ataxin-3. Independent of the Jos_I77KQ78K and Jos_W87K mutants tested above, we generated I77AQ78A and W87A mutants in full-length ataxin-3. Replacing the bulky hydrophobic residues in sites 1 and 2 with a smaller hydrophobic residue rather than a charged residue represents a milder mutation. Even so, we obtained results similar to those observed with the isolated Josephin domain: site 1 proved to be essential for the ability of full-length ataxin-3 to cleave K63-linked or K48-linked chains (Fig. 2). In contrast, site 2 mutations did not alter the ability of full-length ataxin-3 to cleave K63-linked chains but did impair cleavage of K48-linked chains.

Together, these results confirm the crucial importance of site 1 in cleaving all Ub chains and a role of site 2 in conferring linkage preference to ataxin-3.

To gather structural perspective into the role of sites 1 and 2 in Ub chain cleavage by the isolated Josephin domain, we first tested whether the Josephin enzymatic activity presents any preference towards polyUb chains of different lengths and/or linkage. We first incubated Josephin with either K48-linked or K63-linked diUb chains. We observed virtually no change in diUb levels or accumulation of monoUb in either case (Materials and Methods; Fig. 3A, B). The same behavior was observed for full-length ataxin-3 (data not shown). As a control, we verified that these diUb chains are cleavable by an unrelated DUB, USP28 [19]. In the presence of USP28, both K48- and K63-linked diUb chains were cleaved quickly and efficiently (Fig. 3C). These data indicate that diUb chains, whether K63- or K48-linked, are poor Josephin domain and ataxin-3 substrates.

To clarify linkage preferences of the Josephin domain in the absence of the C-terminal region of ataxin-3, we incubated isolated Josephin with equal amounts of K48-linked or K63-linked pentaUb chains (Fig. 3D). The isolated Josephin domain shows preference for K48 linkages compared to K63-linked chains, suggesting that the UIMs of ataxin-3 confer a protective effect toward K48-linked chains.

Finally, we systematically compared Josephin-mediated cleavage of K48-linked polyUb chains of increasing length. Longer chains are cleaved more efficiently than shorter ones, though in general the first cleavage at any length seems to be more efficient than successive ones (Fig. 3E).

To explain the above findings mechanistically, we sought to obtain a better structural description of the possible complexes involved. We used nuclear magnetic resonance (NMR) to study the effects of titrating the isolated Josephin domain with K48- or K63-linked diUb chains and to collect distance restraints which could be translated into structural models. A C14A mutant was used to be sure that no enzymatic cleavage could occur. We observed similar chemical shift perturbation (CSP) effects for both diUb linkages (Fig. 4A,B). Diagnostic CSP effects were those observed for the amide resonances of residues 27, 28, 86 and 87 which are in site 2, and of residues 16, 76, 77, 78, 122 and 133 which are in site 1 [18]. A small number of resonances disappeared or broadened significantly during the titration, indicating an intermediate exchange rate. Among these are 116 and 136 in site 1 (close to the catalytic triad) and 35, 36, and 38 in binding site 2.

We conclude that both Ub binding sites are populated during the titrations in agreement with what observed for monoUb [18]. Since, however, the two sites could be in principle occupied by Ub subunits from different polyUb chains, we clarified the stoichiometries by estimating the correlation times of the complexes by NMR. This is a very sensitive parameter that reflects directly the molecular weight of the species in solution. By reporting the correlation times as a function of different Ub:Josephin molar ratios, we observed that the complex with K48-linked diUb has always shorter values which tend to reach a plateau, albeit at high diUb:Josephin molar ratios (Fig. 4C). Close to plateau, the degree of saturation assuming dissociation constants around 30–60 µM [18] is ca. 85%. Under these conditions, the observed correlation time corresponds to a protein of 31–34 kDa, in excellent agreement with the molecular weight expected for a 1∶1 complex of Josephin with diUb (35 kDa). The complex with K63-linked diUb is instead always comparatively longer, and the curve does not seem to reach a plateau.

These results indicate that Josephin dictates preferential binding properties for different Ub linkages.

To rationalize these observations, we translated the distance information into molecular models using the biomolecular docking program, HADDOCK [20], [21]. We performed three docking runs based on the NMR CSP data. In the first run, we imposed a K48-diUb linkage in combination with the ambiguous interaction restraints (AIRs) defined from the CSP data. This resulted in two ensembles of solutions with similar scores (Fig. 5A). Site 1 (or proximal) Ub shares the same orientation in all solutions, suggesting that this site is overall better defined. Upon binding, the hairpin bends on one side to allow space for the Ub linkage and wraps around the surface of proximal Ub (i.e. the Ub in site 1 which has the C-terminus free) which contains the β-sheet. The bending is much more pronounced than in free Josephin [18], supporting the hypothesis that this secondary structure element helps determine binding specificity. Two contiguous binding surfaces seem instead to be equally compatible with the experimental restraints for Ub binding to site 2 (distal). One surface is similar to that observed in the Josephin/mono-Ub complex [17]; the other is formed by residues in the β1/β2 and α/β3 loops. In both clusters, the aromatic Josephin side-chains of Y27, F28 and W87 contribute to the interface. Strikingly, both clusters contain solutions with the C-terminus of site 1 Ub at close proximity to the Josephin active site, even though no explicit distance restraints were defined to position a Ub close to the active site of Josephin. Comparison of the model of the K48-linked diUb/Josephin complex with other diUb structures shows that, to bind both sites, the diUb chain needs to have an extended linker, adopting a conformation much more open than that observed for the K48-linked diUb complex with the UBA domain of HHR23A [22]. K48-linked polyUb chains are known to exist in solution as a fast dynamic equilibrium between open and closed conformations [23], [24] (Fig. 5B). The closed conformation is predominant at neutral pH and in the absence of binding partners. Other diUb complexes do not, however, show the open conformation necessary to accommodate Josephin.

In the second run, the same docking procedure was followed but this time imposing a K63 Ub linkage (Fig. 5C). The solutions were more scattered and no biologically meaningful structures could be obtained (i.e. none having the Ub C-terminus close to the catalytic center). Even in the best case (cluster 2), the C-terminus of site 1 Ub is close only to H119 but not to C14 or N134. To investigate if this could be an artifact of our docking procedure we repeated the calculations with an additional distance restraint to position the C-terminus of the site 1 Ub into the Josephin active site. No solution was found to satisfy this restraint (data not shown) indicating that the two subunits of a K63-linked diUb cannot be accommodated simultaneously in both Josephin sites.

Finally, we performed a run in which the linkage preference was left ambiguous by defining a linkage restraint including both K48 and K63. Although in principle both linkage types were obtained at the rigid-body docking stage, the only selected solutions for the subsequent semi-flexible refinement correspond to the K48 linkage.

These results indicate that the relative position of the Ub binding sites on Josephin dictate different binding specificities. The structure of K63-linked diUb is incompatible with the geometric requirements imposed by the relative position of the two Josephin binding sites and that only K48-linkage diUb is able to occupy both Ub-binding sites of Josephin simultaneously.

We tested whether the presence of hHR23A has any effect on ataxin-3 cleavage of polyUb chains. Ataxin-3 is known to interact in cells and in vitro with the Ubl domain of hHR23A [18], [25], [26]. This interaction is centered on residue W87 of ataxin-3, which is also the critical residue for Ub binding by site 2 [17]. Consistent with this, HHR23B Ubl was shown to compete monoUb in vitro [18].

The ability of ataxin-3 to cleave polyUb chains is not visibly affected by the presence of either full-length hHR23A or its isolated Ubl domain (Figs. 6A and C). However, full-length hHR23A does reduce the ability of the isolated Josephin domain to cleave polyUb chains (Figs. 6B and C, block arrows). In contrast, the isolated Ubl domain of hHR23A does not affect cleavage by the isolated Josephin domain.

Together, these data argue against a significant impact of hHR23A/ataxin-3 interaction on the ability of full length ataxin-3 to cleave chains in this specific setup. This becomes particularly evident when considering that the Ubl domain of hHR23A, which interacts with ataxin-3, does not have a detectable effect on Ub cleavage.

The following primers were used to generate missense mutations of specific residues of ataxin-3 into alanine (Quickchange Mutagenesis, Stratagene): 5′gccttctggaaatatggatgacagtggttttttctctgctgcggttataagcaatgccttg (forward) and 5′caaggcattgcttataaccgcagcagagaaaaaaccactgtcatccatatttccagaaggc (reverse) for the Iso77-Ala/Glu78-Ala mutant, 5′ggttataagcaatgccttgaaagttgcgggtttagaactaatcctg (forward) and 5′caggattagttctaaacccgcaactttcaaggcattgcttataacc (reverse) for the W98A ataxin-3 mutant. Mutations of the equivalent Josephin residues into lysines were obtained as previously described [17]–[18], [35]. We observed similar results in our enzymatic assays whether these specific residues were mutated to lysine or to alanine. The primers: 5′actatccatggagtccatcttccacgagaaacaagaaggctcacttgctgctcaacattg (forward) and 5′atcttgcggccgcttacctaatcatctgcaggagttggtcagcttcgcaatctggcagatcacc (reverse) were used for a C14A Josephin mutant (cloned using NcoI and NotI restriction sites).

The proteins were expressed in E. coli BL21(DE3) and purified as previously described [35]. Labeled proteins were obtained by growing E. coli in synthetic medium containing ¹⁵NH₄Cl as the sole source of nitrogen. The K48-linked diUb used for structural studies was produced as described elsewhere [36]. Briefly, recombinant Ub mutants (UbD77 and UbK48C) were produced and incubated overnight with E1 (Boston Biochem), Cdc34, ATP, phosphocreatine, creatine phosphokinase and inorganic pyrophosphate at 37°C. The protein was then purified using a monoS cation exchange column after stopping the reaction by adding acetic acid at pH 4.5. K63-linked diUb chains were generated with the same protocol but using UbD77 and UbK63R mutants and replacing E2 with Ubc13 and Uev1a enzymes. The purity and chemical identity of the resulting proteins were examined by SDS polyacrylamide gel electrophoresis and mass spectrometry.

PolyUb chains (250–1000 nM; Boston Biochem) were incubated with full-length ataxin-3 or isolated Josephin domain species (100–400 nM) in DUB buffer (50 mM HEPES, 0.5 mM EDTA, 1 mM DTT, 0.1 mg/ml ovalbumin, pH 7.5) at 37°C. Fractions were collected in 2% SDS, 100 mM DTT sample buffer, boiled for 1 min and electrophoresed in 15% SDS-PAGE gels or in 4%–20% Ready Gels (BioRad). SDS-PAGE electrophoresis and western blotting were conducted as previously described [14], [37]. In-gel protein staining by Sypro Ruby (Invitrogen) was conducted per the manufacturer's instructions, and imaged in a VersaDoc MP5000 imager (BioRad) with UV trans-illumination. Rabbit anti-Ub (1∶500; DAKO), mouse anti-ataxin-3 (1H9, 1∶1000; generous gift from Yvon Trottier), rabbit anti-MJD (1∶20,000) [38], goat anti-GST (1∶20,000) (GE Healthcare) antibodies were used for protein detection. High molecular weight (HMW) ubiquitin species observed by western blotting in enzymatic assays are believed to be multimers of the respective substrates [14].

Although we routinely observe monoUb generated during our enzymatic assays, monoUb is less readily detected by western blotting techniques than longer Ub chains. Thus, for diUb cleavage assays, to ascertain that diUb is indeed not a substrate for ataxin-3 we used a combination of western blotting and in-gel protein detection (as described above), as well as positive controls with the unrelated DUB USP28 with which the monoUb reaction product is easily detected. This approach, together with data from cleavage of polyUb chains by ataxin-3 (in which we detect monoUb) ensures us that the absence of monoUb from diUb cleavage assays is due to the actual absence of monoUb generated during the reaction rather than to a lack of detection on immunoblots.

For semi-quantification of western blots, images were collected below-saturation levels using VersaDoc 5000MP (BioRad). Quantification was conducted with Quantity One (BioRad). Background was subtracted equally among all lanes and the intensity of each band was integrated as the area under each intensity curve. The sum of intensities of all product bands was divided by the sum of intensities of all bands in that lane. For example, for cleavage of Ub5 chains, the equation was: Σ(I_Ub4, I_Ub3, I_Ub2, I_Ub1)/Σ(I_Ub5, I_Ub4, I_Ub3, I_Ub2, I_Ub1). Statistical analyses of data collected from independent experiments were conducted using two-tailed student t-tests. All graphical data are shown as means +/− standard deviation.

For structure calculation purposes the biomolecular docking program, HADDOCK [20], [21] was used based on CSP data analyzsed with a recently developed method, SAMPLEX [39]. SAMPLEX identifies the residues that reflect significant chemical shift changes through an unbiased and automated methodology; it outputs three types of residues: residues classified as unaffected (core family), interfacial (interface family) and intermediate. SAMPLEX was run to determine the significant changes in the CSP data of the two binding sites obtained from NMR titrations of Josephin constructs with a mutation at either site 1 (I77K-Q78K) or site 2 (W87K). The active residues for docking were defined based on the SAMPLEX selection (including the intermediate class for Josephin to increase the possible interface region). For Josephin, the solvent accessible neighbors of the active residues within a 10 Å cut-off were selected as passive residues in order to increase the ambiguity, resulting thus in a larger interface coverage. Passive residues on the Ub surface were defined in the same manner using a 5 Å cut-off.

Three different three-body (two Ub molecules and Josephin) docking runs were performed to test the ability of K48 and K63 linked diUb to interact with Josephin: K48 or K63 linkages were imposed in two of them. In a third run, ambiguous linkage restraints were imposed to both K48 and K63 to see which one would be preferentially selected during docking. In all three runs, 5000 rigid-body docking solutions were generated and the best 200 were subjected to semi-flexible refinement in torsion angle space followed by refinement in explicit water. Random removal of restraints was turned off (noecv = false) and additional center-of-mass restraints were defined between the various molecules to ensure compact solutions. All other parameters were left to their HADDOCK default values. The C-terminal tail of each Ub was defined as fully flexible together with K48 and/or K63 depending on the linkage type.

The isopeptide bond between the two Ubs was defined by imposing a loose distance restraint (10 Å) between the side chain ammonium group of either K48 or K63 and the carboxyl atom of G76. In a third docking run with an ambiguous linkage definition, an ambiguous distance restraint was defined including both K48 and K63. A final additional water refinement was performed in which the distance restraint was decreased to 1.5 Å in order to re-establish the iso-peptide bond.

To check if positioning of Ub into the catalytic site could be consistent with the measured chemical shift data, the three runs were repeated with one additional distance restraint (10 Å) between the Josephin side chain sulphur of C14 and the C-terminal carbon of site 2. This distance was shortened to 5 Å for the additional water refinement step.

The resulting models of each docking run were clustered with a 5 Å cut-off and the clusters ranked based on the average HADDOCK score calculated over the best four members of each cluster.

NMR experiments were performed at 298 K on a Bruker Avance spectrometer operating at 600 MHz and equipped with cryoprobe. Titrations of ¹⁵N labelled Josephin or C14A Josephin mutant (0.05 mM in 20 mM sodium phosphate pH 6.5, 2 mM DTT) with diUB chains were performed with stepwise additions of concentrated stock solutions of K48-linked or K63-linked diUb chains (1.47 mM and 0.4 mM, respectively) up to a 10 and 2.5 fold molar excess, respectively. Combined measurements of T1 and T2 relaxation times were performed to estimate the correlation times using a home modified Bruker sequence (hsqct2etf3gpsi3d). Overall correlation time values were estimated from the R2/R1 ratio using the program R2R1_TM available from the website of A.G. Palmer's group (A.G. Palmer III, Columbia University) [40]. The values were compared to a literature calibration curve [41].