Structural basis for ubiquitin recognition and autoubiquitination by Rabex-5

Structure of the Rabex-5 A20 ZnF and IUIM domains

The structure of Rabex-5 residues 9–73, comprising the A20 ZnF and IUIM domains (Fig. 1a), was determined in complex with ubiquitin in two different space groups: at 2.8 Å in a P6₁ lattice containing one complex per asymmetric unit, and at 2.5 Å in a C2 lattice containing three complexes per asymmetric unit. The structure was determined initially in the P6₁ lattice by molecular replacement (MR), using ubiquitin as the search model. Density modification was used to determine the structure of the A20 ZnF-IUIM portion of the complex (Fig. 1b). The structure of the complex in the P6₁ lattice was re-determined de novo by single anomalous dispersion (SAD) using the native Zn ion, and the resulting Fourier synthesis was found to be consistent with the MR solution. The C2 structure was determined by molecular replacement with the partially refined P6₁ structure. Together the two crystal forms present images of four copies of the complex.

The structure reveals that the A20 ZnF and IUIM domains are fused into a single relatively rigid unit (Fig. 1c, d). Based on sequence analysis, the A20 ZnF has nominal boundaries from residues 13 to 47, with the IUIM comprising the remainder. The structure shows that residues 36–47 of the A20 ZnF, together with the entire IUIM, form a single contiguous helix. The non-helical N-terminal part of the A20 ZnF contains three of the zinc ligands: Cys19, Cys23, and Cys35. The fourth zinc ligand, Cys38, is located in the first turn of the helix (Fig. 1e). The zinc thus bridges the N-terminal non-helical part of the A20 ZnF with the helix, rigidly connecting the two. Trp31 of the non-helical part of the A20 ZnF stacks against Trp39 of the helix, further stabilizing the orientation of the helix with respect to the A20 ZnF (Fig. 1f).

The A20 ZnF has a striking hydrophobic patch on its surface, centered on a pair of highly solvent-exposed Tyr residues, Tyr25 and Tyr26 (Fig. 1f). Leu17 and Leu18 are contiguous with one side of the Tyr pair. The stacked aromatic residues Trp31, Trp39, and Tyr43 are contiguous with the other side, forming a hydrophobic strip of roughly 25 Å along its longest dimension. The hydrophobic strip is broken by a polar patch on the helix from His44 to Gln50. The surface of the C-terminal portion of the IUIM helix presents a second hydrophobic strip, consisting of the exposed residues Ile51, Trp55, Leu57, Ala58, Leu61 and Phe69 (Fig. 1f and g). This IUIM hydrophobic strip is flanked by the acidic residues Glu53, Asp54, Glu56, Glu59, Glu64, Glu65, Glu66, and Glu67.

Structure of the Rabex-5:ubiquitin complex

Ubiquitin and the Rabex-5 (9–73) fragment are present at a 1:1 stoichiometry in both crystal forms. The lattice in both crystal forms is packed such that the two molecules interact at two distinct interfaces. In the first interface, ubiquitin binds to Rabex-5 through extensive hydrophobic and polar interactions with the Rabex-5 IUIM. Ubiquitin binds through its surface hydrophobic patch centered on Ile44, Leu8, and Val70 (Fig. 2a). The IUIM binding site for ubiquitin is centered on Ile57 and Ala58, and includes all of the hydrophobic strip residues on the IUIM (Fig. 2b). The hydrophobic interactions are supplemented by hydrogen bonds between the side-chain of Rabex-5 Arg47 and the main chain carbonyl of ubiquitin Lys63 (Fig. 2c); the side-chain of Rabex-5 Asp54 and the main chain amides of ubiquitin Ala46 and Gly47; and a salt bridge between Rabex-5 Glu65 and ubiquitin Arg42 (Fig. 2d). Two potential sites of ubiquitin chain extension, Lys48 and Lys63, are near the interface (7 to 9 Å from Lys Nζ to nearest Rabex-5 atom; Fig. 2a). Neither side-chain participates directly in the interface, and both are solvent accessible in the complex. The shape complementarity score ²⁵ is 0.76, which is considered highly complementary. The complex buries 780 Ų and 710 Ų of solvent accessible surface area, respectively, for Rabex-5 and ubiquitin.

The A20 ZnF of Rabex-5 forms a second interface with a molecule of ubiquitin that is distinct from the one bound to the IUIM (Fig. 2e, f, h). At the closest point of approach, these two ubiquitin molecules are 20 Šapart. On the A20 ZnF, the two hyperexposed Tyr residues 25 and 26 at the center of the hydrophobic patch are the main locus of the interaction. Several other residues of the A20 domain also interact: Asn28, Trp31, Ser36, and Lys37 (Fig. 2f). The interaction site on ubiquitin consists of the polar residues Arg54, Thr55, Ser57, Asp58, Tyr59, and Asn60 (Fig. 2e). Aliphatic and aromatic carbon atoms on these side-chains interact with the di-Tyr motif on the A20 ZnF domain. Rabex-5 Ser36 forms two hydrogen bonds with ubiquitin Asp58 (Fig. 2g), one of which is a short-range and presumably strong interaction between the two side-chains. The shape complementarity score is 0.74, nearly identical to the high score obtained for the ubiquitin/IUIM interaction. Roughly 420 Ų of solvent accessible surface area is buried on each molecule.

Rabex-5 contains an inverted UIM

The IUIM of Rabex-5 binds to the Ile44 patch on ubiquitin in a manner very akin to the binding of the Vps27 UIM-1 ²⁶. UIMs are ~25 residue, single helix motifs that were first discovered through bioinformatics analysis of the sequence of the polyubiquitin binding site in the proteasome subunit S5a ²⁷. UIMs typically bind monoubiquitin with 200 – 2000 μM affinity ²⁸. UIMs within the endocytic proteins epsin, eps15, eps15R, and Vps27 ^29–32 both bind to monoubiquitin and promote the monoubiquitination of the proteins that contain them. This monoubiquitination requires an E3 ligase, as UIMs are not themselves E3 ligases.

The Rabex-5 and Vps27 UIM-1 (pdb entry 1Q0W) complexes were compared by superimposing the ubiquitin molecules, which differ by 1.0 Å r.m.s.d. over the first 73 Cα positions (Fig. 3a). Vps27 residues 257–275 are helical and in contact with the same surface on ubiquitin as Rabex-5 residues 49–67. The conserved central Ala266 of the Vps27 UIM-1 corresponds to the central Ala58 of the Rabex-5 IUIM (Fig. 3b-d). There is a one-to-one correspondence between ubiquitin-interacting residues in the C-terminal two-thirds of the Rabex-5 IUIM and the N-terminal two-thirds of the Vps27 UIM-1. The close relationship breaks down only for the N-terminal five residues of Rabex-5. The C-terminal acidic cluster of Rabex-5 matches the N-terminal acidic cluster of Vps27 UIM-1, for example. Asp-54 of Rabex-5 makes interactions with the main chain of ubiquitin residues 46 and 47 much like those reported for its counterpart in Vps27, the conserved Ser270 ²⁶. For this reason, we designated the Rabex-5 ubiquitin binding helix as an “inverted ubiquitin interacting motif” (IUIM).

A template for the A20 ZnF family

The A20 ZnF is a compact unit built around two pairs of Cys residues and a single zinc ion. The coordination of the zinc by four Cys (Fig. 1e), and the helical conformation of the last Cys, fit the pattern of the classic CCCC finger ³³. This class of zinc finger is found in the GATA, LIM, and PHD domain-containing DNA-binding proteins, C1 and FYVE domain-containing peripheral membrane proteins, and represents the first zinc site in the RING domain E3 ligases ³³. The A20 ZnF is unique in that the first two Cys are separated by three residues, as compared to two residues in other CCCC fingers. The hyperexposed Tyr-Tyr pair on the Rabex-5 A20 ZnF is its most prominent surface feature. These two aromatic residues are often Phe or Tyr in the sequences of other A20 ZnF domains (Fig. 4a,b). The Tyr-Tyr pair is surrounded by other conserved residues (Fig. 4b). The A20 ZnF can be overlaid on the Cbl RING structure based on the similarity between the A20 zinc site and the first zinc site in the RING domain. The E2 binding site on the Cbl RING domain is known from the crystal structure of its complex with UbcH7 ²⁴. This site overlays a non-conserved polar surface on A20 that is distal to the Tyr-Tyr patch. This is consistent with the classification of the A20 ZnF E3 ligases as a separate group from the RING E3 ligases. Another zinc finger domain, the NZF domain, also binds ubiquitin. The NZF domain contains no α-helix and binds ubiquitin through its Ile44 patch³⁴, and thus has little in common with the A20 zinc finger.

Mutational analysis of ubiquitin binding sites

In order to evaluate the energetic contribution of the A20 ZnF and IUIM ubiqiutin interaction sites observed in the structure to ubiquitin binding, these regions were mutagenized alone and in combination. The effects on binding were measured in two ways. Isothermal titration calorimetry (ITC) was used to determine the effect on binding to ubiquitin in free solution (Fig. 5a). Surface plasmon resonance (SPR) was used to determine the effect on binding to ubiquitin immobilized on a planar surface (Fig. 5b). Wild-type Rabex-5 (9–73) binds to ubiquitin in solution with an apparent affinity of K_D = 12 μM when the ITC data are fit assuming there is only one type of binding site. These data fits had relatively high residuals, suggesting that the single site model was inappropriate. It was not possible to obtain consistent and statistically significant improvement by fitting the ITC data to a two-site model, so we pursued a mutational approach to estimating the ubiquitin affinities of the two sites. The mutations L57D and A58D in the IUIM reduced affinity to K_D = 22 and 29 μM, respectively. The mutation Y26A reduced affinity to K_D = 28 μM. The double mutants Y25A/A58D completely abolished binding at achievable concentrations. From these results we infer that the most deleterious point mutants in each site are capable of completely abrogating binding to that site. We infer that the A20 site has an affinity of ~ 29 μM and the IUIM site has an affinity of ~ 28 μM in free solution.

Wild-type Rabex-5 (9–73) binds to immobilized ubiquitin with apparent affinities of K_D = 1.3 μM and 37 μM when the SPR data are fit with a two site model (Fig. 5c, Table 2). GST-ubiquitin was immobilized to the sensor chip. Since GST is a dimer, two ubiquitin molecules are expected to be presented to the Rabex-5 molecule on average. The higher apparent affinity of wild-type Rabex-5 (9–73) for dimeric immobilized ubiquitin as opposed to ubiquitin in solution suggests that avidity plays a role in the first case. The most deleterious mutant tested in the A20 ZnF domain, Y25A, had K_D = 29 μM, while the most deleterious mutant in the IUIM, A58D, had K_D = 22 μM. The double mutant Y25A/A58D had essentially no detectable binding. The SPR data thus suggest that the A20 ZnF and IUIM sites have K_D values of ~22 and 29 μM, respectively. The agreement between the ITC and SPR results for the two sites is remarkably good in view of the differences in the presentation of the molecules in the two experiments.

In order to determine which interfaces of ubiquitin were involved in functional interactions, the ubiquitin mutants I44D and D58A were generated. The binding of these mutants to wild-type and mutant Rabex-5 constructs was assessed to determine which interfaces were functional, an approach known as “double mutant cycle” analysis. The I44D mutant was made to disrupt the canonical hydrophobic patch on the surface of ubiquitin that interacts structurally with the IUIM. The ubiquitin D58A mutant was designed to disrupt the two hydrogen bonds between ubiquitin Asp58 and the Rabex-5 Ser36 side-chain and main-chain. Ubiquitin I44D binds to wild-type Rabex-5 and its IUIM mutant A58D with very similar affinity (K_D = 20–23 μM; Fig. 5d; Table 2). This is consistent with the expectation that either mutant should completely abrogate the Rabex-5 IUIM/ubiquitin Ile44 patch interaction, yet have no effect on the A20 ZnF/ubiquitin Asp58 patch interaction. Ubiquitin I44D binds with sharply reduced affinity to Rabex-5 A20 ZnF domain mutant Y25A, and not at all to Y25A/Y26A. This confirms that the residual interaction of I44D with Rabex-5 requires an intact A20 ZnF domain binding site. Ubiquitin D58A binds to wild-type Rabex-5 and its A20 ZnF domain mutant Y25A with very similar affinity (K_D = 26–28 μM; Fig. 5e; Table 2), consistent with the structural finding that these two regions directly interact. In contrast, mutating the Rabex-5 IUIM (A58D) almost completely blocks binding. Thus the residual Rabex-5 binding of the D58A mutant occurs entirely through the IUIM. The double mutant cycle results show that both of the ubiquitin:Rabex-5 interfaces seen in the crystal structure are functional.

Mechanism for E3 ligase activity

In order to probe the mechanism for ubiquitin E3 ligase activity by Rabex-5, we tested the ability of Rabex-5 (9–73) and its A20 ZnF domain mutant Y25A/Y26A to recruit ubiquitin conjugating enzymes (Ubcs; Fig. 6a). Both ubiquitin loaded and unloaded UbcH5 and Ubc7 were generated and tested. To assess specificity, Rabex-5 was compared to a known Ubc7-binding protein, gp78 ³⁵ and a known UbcH5-binding E3, RNF25 ³⁶. Ubiquitin-loaded UbcH5 binds to wild-type Rabex-5 (9–73), but not to the A20 ZnF domain Y25A/Y26A mutant nor to gp78 (Fig. 6a). Unloaded UbcH5 does not bind to Rabex-5 (9–73). Ubc7 binds to gp78 with roughly equal affinity in both its ubiquitin-loaded and unloaded states (Fig. 6a). Thus Rabex-5 specifically recruits at least one Ubc, UbcH5, in a manner that depends on the interaction between ubiquitin and the A20 ZnF domain. The presence of ubiquitin alone is insufficient for recruitment, at least at concentrations tested, since ubiquitin-loaded Ubc7 is not recruited. The E3 ligase activity with respect to GST-Rabex-5 (9–73) was tested for wild-type and mutant constructs. The A20 ZnF domain Tyr mutants tested blocked catalytic activity, consistent with the requirement for E2 binding (Fig. 6b). The IUIM mutant A58D did not block activity, but rather enhanced activity several-fold.

Cloning and sample preparation

DNA coding for the fragment of bovine Rabex-5 spanning residues 9–73 was cloned into parallel GST2 vector ³⁹ using a PCR-based cloning strategy. Site-directed single and double mutants were generated using QuikChange Mutagenesis Kit (Stratagene). All the constructs were verified by DNA sequencing. Rabex-5(9–73) was overexpressed as an N-terminal glutathione S-transferase (GST) fusion protein in Escherichia coli strain BL21(DE3). ZnCl₂ was added to terrific broth media at final concentration of 0.1 mM. Cells were grown at 37°C, induced with 0.5 mM isopropyl-β-D-thiogalactopyranoside at optical density of 1.0 and further incubated at 20°C overnight. Harvested cells were resuspended in buffer A (50 mM TrisHCl, pH 7.4, 150 mM NaCl), lysed and centrifuged. Supernatant was applied to a glutathione-Sepharose column (Amersham) in buffer A and washed with buffer B (50 mM TrisHCl, pH 7.4, 500 mM NaCl). Tobacco etch virus protease was added to cleave GST and removed by passing through TALON resin (BD Bioscience) in the buffer A. The cleaved protein was concentrated and applied to a Superdex 200 column (Amersham) in the buffer C (50 mM TrisHCl, pH 7.4, 150 mM NaCl, 5 mM dithiothreitol). To form a complex with ubiquitin, bovine ubiquitin (Sigma) was added to the purified rabex-5(9–73) and the resulting complex was separated using a Superdex 200 column in the buffer C. The pooled complex solution was concentrated using a VivaSpin concentrator (Viva Science). All mutants were purified as described above except that no reducing agent was used in buffers. DNA coding for human ubiquitin was amplified by PCR and cloned into the pGST2 vector. Recombinant GST and GST-ubiquitin for SPR studies were produced from BL21 (DE3) cells and purified using glutathione-sepharose affinity column followed by a Superdex S200 (GE Health Sciences) gel filtration chromatography. GST and GST-ubiquitin samples were dialyzed against 10 mM HEPES (pH 7.4), 150 mM NaCl, and 0.5% surfactant P20 (HPS-P).

Crystallization, data collection and structure determination

Crystals of the complex were grown by hanging drop method at 22°C. The reservoir solution contained 0.1 M sodium citrate, pH 5.6, 20–25% polyethylene glycol 4000, and 0.2 M lithium sulfate. Diffraction-quality crystals were obtained by streak seeding. Crystals appeared in 1–2 days and were flash frozen under liquid nitrogen. Data were collected using a Riguaku rotating anode home source and an R-AxisIV detector at 100 K and processed with HKL2000 (HKL Research). An initial molecular-replacement (MR) solution was obtained using ubiquitin as a search model using MOLREP ⁴⁰ in space group P6₁. Solvent flattening by the prime-and-switch algorithm in RESOLVE ⁴¹ yielded an electron density map sufficient to locate the missing Rabex-5(9–73). Single anomalous dispersion (SAD) at the Zn edge, λ = 1.28 Å, was used to re-determine the structure of the complex de novo using SOLVE/RESOLVE ⁴². The structure by SAD confirmed that the model by MR was correct. The structure in space group C2 was determined by molecular-replacement using the refined model in P6₁. Iterative manual model building and initial refinement were performed using O ⁴³ and CNS ⁴⁴, respectively. Final refinement was done using REFMAC5 ⁴⁵ with TLS parameters ⁴⁶ incorporated. Statistics for data processing and refinement are listed in Table 1. All the structural figures were generated using PyMol (www.pymol.org).

Surface plasmon resonance

The binding of wild-type and mutant Rabex-5 proteins to ubiquitin was monitored using SPR with a BIAcore T100 (BIAcore AB, Sweden), at 25⁰C, with a flow rate of 20 ml min⁻¹. Approximately 10,000 RU of GST, GST-ubiquitin, and GST-I44D and D58A ubiquitin mutants were immobilized on a CM5 surface that was stable for more than 60 experimental cycles. Briefly, a CM5 chip was activated using NHS/EDC (1:1) at a flow rate of 5 ml min⁻¹ for 1200 s. GST (10 μM) and GST-Ubiquitin (10 μM) in 10 mM Acetate buffer (pH 5.5) was passed over separate flow cells at 5 μl min⁻¹ for 2400 s, followed by a blocking step using ethanolamine (1M pH 8.5) at 5 μl min⁻¹ for 1200 s. All binding experiments were performed in HPS-P. Binding of wild-type and mutant Rabex-5 proteins to ubiquitin was measured simultaneously by passing Rabex-5 over GST-, GST-ubiquitin, and GST-ubiquitin mutant- coupled flow cells with association and dissociation times of 100 s and 300 s, respectively. Between subsequent injections of Rabex-5 proteins, surfaces were regenerated with an injection of HPS-P supplemented with 500 mM NaCl for 15 s at 100 ml min⁻¹.

We initially tried to fit our data assuming a 1:1 binding stoichiometry for a Rabex:Ubiquitin complex. However, poor fitting statistics and subsequent mutational data led us to define 2 independent binding sites on Rabex-5 for ubiquitin as follows:

Where R is Rabex-5 and U is ubiquitin. Corresponding data were fit to the following equation:

Where [Rabex] is the protein concentration of the flowing analyte, K_D1 and K_D2 are the dissociation constants for site 1 and site 2, respectively, Rmax¹ and Rmax² are the relative maximal change in response levels for site 1 and site 2, respectively, and RI is the residual or background signal. Fitting was performed using the BiaEvalution software, with globally floating K_D, Rmax, and RI values.

Isothermal titration calorimetry

The Rabex-5 (9–73) fragments were dissolved in 50 mM Tris/HCl pH 7.4, 150 mM NaCl. The same buffer was used to prepare 0.3–0.75 mM solutions of bovine ubiquitin (Sigma), which was the injectant. Rabex-5 (9–73) was placed in the sample cell at 10–30 μM. The concentrations of the ubiquitin and Rabex-5 (9–73) were calculated by measuring absorbance at 280 nm and extinction coefficients of 1814 M⁻¹ for ubiquitin and 21,175 M⁻¹, 19,892 M⁻¹, 19,892 M⁻¹, 21,193 M⁻¹, 21,141 M⁻¹, 21,180 M⁻¹, 21,127 M⁻¹, 21,180 M⁻¹, 21,185 M⁻¹ and 19,759 M⁻¹ for Rabex-5 (9–73) wild type and the Y26A, Y43A, R47A, D54A, L57D, A58D, L61D, E65A, Y25A, A58D mutants. Titrations (42 injections of 5 μl each) were performed at 30°C using a VP-ITC Microcalorimeter (MicroCal, LLC), and data were analyzed using Origin software (Origin Lab).

Interaction with E2 enzymes

UbcH5C was expressed using the plasmid pET15-UbcH5C ⁴⁷. DNA coding for MmUbc7 was subcloned into the NcoI/BamHI sites of pET15. The human E3 gp78 encodes a specific MmUbc7 binding site between amino acid 574 – 643. A fragment coding for gp78 amino acid 574–643 was described previously ⁴⁸, as was the GST-RNF25(86–363) construct ³⁶. [³⁵S]-Labeled UbcH5C and MmUbc7 were translated in vitro using the S30 T7 bacteria lysate system (Promega). For ubiquitin thioester formation, reactions contained 12 μL E2 translation mix, 100 nM murine E1 and 10 μg ubiquitin 80 μL reaction buffer (20 mM Tris, pH 7.5, 100 mM NaCl, 4 mM ATP, 2 mM MgCl₂). The reactions were carried out at 30^o C for 10 min. To produce free E2, the same reaction conditions were used except that E1 was omitted. An aliquot containing 2 μg of GST fusion protein immobilized on glutathione-Sepharose 4B (Amersham) was incubated with 20 μL thioester reaction mix in 100 μL of binding buffer (PBS containing 0.2% Triton X-100 and 0.2 mM ZnCl₂) for 2 h at 4°C with constant mixing. Equal loading of all samples was confirmed by SDS-PAGE and Coomassie blue staining (not shown). The immobilized proteins were collected by centrifugation and washed three times with 100 bed volumes of binding buffer. The mixture was quenched in SDS sample buffer (with or without DTT), separated by SDS-PAGE and processed for visualization on a Storm PhosphorImager (Amersham).

In vitro ubiquitination assays

Reaction mixtures contained 2 μg GST fusion proteins immobilized on glutathione-sepharose 4B (Amersham), 100nM recombinant murine E1 (expressed in Sf9 cells), 40 nM recombinant UbcH5C (expressed in E.coli), 2μMFlag-Ub, 0.5–1μM ubiquitin aldehyde in a final volume of 50μL reaction buffer (20mM Trisp H7.5, 100mM NaCl, 4mM ATP and 2mM MgCl2). Reactions were incubated at 30°C for 90m in with agitation, washed three times with 20 bed volumes of wash buffer (20mM Tris, pH7.5, 100mM NaCl, 1 mM DTT) and quenched with SDS sample buffer. Samples were subjected to SDS-PAGE and processed for immunoblot with anti-Flag antibody. The recombinant E2 enzymes were prepared as described⁴⁹.