Solution Structure of RING Finger-like Domain of Retinoblastoma-binding Protein-6 (RBBP6) Suggests It Functions as a U-box*

Background: U-box-containing proteins cooperate with chaperones in ubiquitinating irreversibly unfolded proteins. Results: Retinoblastoma binding protein-6 (RBBP6) contains a zinc-binding U-box-like domain and interacts directly with chaperones. Conclusion: RBBP6 may play a role in protein quality control. Significance: U-boxes should be classified in terms of their interaction with chaperones and not their zinc binding properties. Retinoblastoma-binding protein-6 (RBBP6) plays a facilitating role, through its RING finger-like domain, in the ubiquitination of p53 by Hdm2 that is suggestive of E4-like activity. Although the presence of eight conserved cysteine residues makes it highly probable that the RING finger-like domain coordinates two zinc ions, analysis of the primary sequence suggests an alternative classification as a member of the U-box family, the members of which do not bind zinc ions. We show here that despite binding two zinc ions, the domain adopts a homodimeric structure highly similar to those of a number of U-boxes. Zinc ions could be replaced by cadmium ions without significantly disrupting the structure or the stability of the domain, although the rate of substitution was an order of magnitude slower than any previous measurement, suggesting that the structure is particularly stable, a conclusion supported by the high thermal stability of the domain. A hallmark of U-box-containing proteins is their association with chaperones, with which they cooperate in eliminating irretrievably unfolded proteins by tagging them for degradation by the proteasome. Using a yeast two-hybrid screen, we show that RBBP6 interacts with chaperones Hsp70 and Hsp40 through its N-terminal ubiquitin-like domain. Taken together with the structural similarities to U-box-containing proteins, our data suggest that RBBP6 plays a role in chaperone-mediated ubiquitination and possibly in protein quality control.

Retinoblastoma-binding protein-6 (RBBP6) is a multifunctional protein found ostensibly in all eukaryotes but not in bacteria, which is implicated in a diverse set of cellular functions, including mRNA metabolism (1)(2)(3), regulation of the cell cycle (4 -8), tumorigenesis (9), and development (10,11). It forms part of the pre-mRNA 3-end processing complex (3) in humans and the cleavage and polyadenylation factor complex in yeast (2) and has been shown to interact directly with tumor suppressors p53 and the retinoblastoma gene product (1,13).
RBBP6 contains an N-terminal ubiquitin-like domain (14) and a cysteine-rich RING finger-like domain, through which it promotes the ubiquitination of p53 by Hdm2 in an E4-like manner (10). Also through its RING finger-like domain, RBBP6 interacts directly with the pro-proliferative transcription factor Y-box-binding protein-1 (YB-1), and overexpression of RBBP6 in cultured mammalian cells leads to suppression of YB-1 in a proteasome-dependent manner (15). Whether this represents an example of E4-or E3-like behavior has yet to be determined. However, because it down-regulates both the pro-apoptotic p53 and the anti-apoptotic YB-1, the effect of RBBP6 on tumorigenesis is likely to be highly complex.
The cysteine-rich domain of RBBP6 has been classified both as a RING finger, due to the presence of eight conserved cysteine residues, and as a U-box, due to a conserved pattern of hydrophobic residues (16). RING fingers are small domains, ϳ70 residues in length, that fold independently with the help of two Zn 2ϩ ions. The ions are coordinated by four pairs of cysteine or histidine residues in a so-called "cross-braced" fashion, meaning that one of the ions is coordinated by the first and third Cys/His pairs and the other ion by the second and fourth Cys/His pairs. RING fingers are classified according to the pattern of zinc coordinating residues, with C3HC4 being the most common, although C3HHC3, C2H2C4, and C4C4 RING fingers are also found. The presence of eight conserved cysteine residues therefore suggests that RBBP6 contains a C4C4 RING finger domain binding two zinc ions.
RING fingers all share the same basic fold consisting of two large loops, each stabilized by a single zinc ion, lying parallel to an ␣-helix and together packing against a short three-stranded ␤-sheet. An identical fold is found in the U-box family, although there the coordination of zinc ions is replaced by a scaffold of salt bridges and hydrogen bonds (17). Most U-boxes and some RING fingers have been shown to form homodimers in solution, and in some cases dimerization has been shown to be required for ubiquitination activity (18,19). Dimerization takes place along the ␤-sheet and is stabilized by interactions between the N and C termini of both monomers; the RING domains from the BRCA1/BARD1 heterodimer each contain N-and C-terminal helices that interact to form a 4-helix bundle. In the U-box domains from CHIP and PUB14, the N-terminal helices are replaced by structured loops that pack against and C-terminal helices, which constitute the bulk of the interface.
The primary function of RING fingers and U-boxes is to recruit the ubiquitin-conjugated E2 so that ubiquitin can be transferred from the E2 to the substrate, which in its turn is recruited by the substrate-binding domain of the E3. In addition to their characterization as "zinc-less RING fingers," U-boxes were originally considered to be the hallmark of E4 ligases (also called elongation factors), which catalyze the addition of ubiquitin to pre-existing polyubiquitin chains but not to the substrate itself. However, many U-boxes have subsequently been shown to be capable of catalyzing polyubiquitination without the need for a separate initiating E3 ligase, rendering the association with E4 ligases problematic. More recently, U-boxes have become associated with chaperone-mediated ubiquitination and protein quality control (20). Many U-boxes have been shown to interact directly with chaperones or cochaperones (21), the most well characterized being the C terminus of Hsp70 interacting protein (CHIP), which interacts with Hsp70 and Hsp90 through its tetratricopeptide domain, ubiquitinating unfolded proteins presented to it by Hsp70 (22)(23)(24). To resolve the question of whether RBBP6 contains a RING finger or a U-box domain, and to provide a structural framework for future interaction studies, we set out to determine the structure of the human domain using solution-state NMR spectroscopy.
Bacterial Expression and Purification-DNA sequences incorporating BamHI and XhoI restriction sites were amplified from a full-length cDNA clone of RBBP6 (15) and cloned into the corresponding sites of pGEX-6P-2 expression plasmids (GE Healthcare). 15 N, 13 C-Enriched GST fusion proteins were expressed as described previously (14), with the addition of 2 mM ZnSO 4 to the culture media following induction with isopropyl 1-thio-␤-D-galactopyranoside. With the exception of the concentration series, all concentrations used for NMR analysis were in the range 1-2 mM.
Mass Spectrometry-Non-denaturing mass spectrometry was carried out on an LTQ Orbitrap Velos mass spectrometer equipped with electrospray source by direct injection of a 250 M protein sample in 10 mM ammonium acetate, 0 mM NaCl, pH 6.0 (pH adjusted with ammonium hydroxide and acetic acid). The flow rate was set to 10 l/min, the source voltage to 4.0 kV, the S-lens to 60 V, and the capillary temperature to 200°C. Denaturing conditions were produced by addition of 0.1% v/v formic acid.
NMR Data Collection and Analysis-All NMR work was carried out at 25°C in 50 mM sodium phosphate buffer, 150 mM NaCl, 1 mM DTT, 0.02% sodium azide, pH 6.0. Cd 2ϩ /Zn 2ϩ exchange rates were measured by adding 113 Cd-EDTA to a final concentration of 4 mM to a protein sample of ϳ1 mM and recording a series of 1 H-15 N HSQC 3 spectra over a 2-week period. A 40 mM stock solution of 113 Cd-EDTA was prepared by suspending 1 mg of 113 CdO (Cambridge Isotope Laboratories, Andover, MA) in 100 l of a 100 mM solution of HCl, followed by addition of 100 l of a 100 mM solution of NaOH containing 80 mM EDTA and adjusting the pH with HCl or NaOH. 4 Unless otherwise specified, NMR data were collected on a 600 MHz UNITY INOVA spectrometer (Varian Inc, Palo Alto, CA), using a room temperature triple resonance probe. Directly detected one-dimensional 113 Cd spectra were recorded using a 5-mm broadband probe, and spectra were referenced relative to 113 CdSO 4 by setting the chemical shift of 113 Cd-EDTA to 90 ppm (27). 1 H-113 Cd HMQC spectra without gradient enhancement were recorded using a 5-mm broadband probe; 32 t 1 increments of 1000 transients each were recorded with a transfer delay of 12.5 ms, corresponding to an effective coupling constant of 40 Hz. The concentration series was recorded on a Bruker Avance 600 MHz spectrometer equipped with a triple resonance cryoprobe.
NMR data were processed using NMRPipe (28) and visualized using NMRView (29). H 2 O/D 2 O and Cd 2ϩ /Zn 2ϩ decay rates were fitted to a two-parameter exponential using MAT-LAB (The Mathworks, Natick, MA). Chemical shift perturbations induced by changes in protein concentration [P] were fitted to the three-parameter expression as shown in Equations 1 and 2, from which values for ⌬␦ tot , K D , and d were extracted for each residue. ⌬␦ represents the composite chemical shift perturbation corresponding to perturbations ⌬␦⌯ in the 1 H and ⌬␦⌵ in the 15 N dimensions, respectively, and the factor of 6.5 accounts for the relative widths of 1 H and 15 N chemical shift distribu-tions (30). ⌬␦ tot is the total chemical shift change between monomer and dimer for each residue, and d corresponds to the residual chemical shift change between the lowest accessible concentration and the zero concentration limit (pure monomer). The curves in supplemental Fig. 5 have been shifted vertically by an amount d to reflect perturbations relative to zero concentration. Curve plotting and fitting were carried out using ProFit (QuantumSoft, Uetikon am See, Switzerland). Thermal Stability-The state of folding was monitored qualitatively using a series of 1 H-15 N HSQC recorded at 5°C intervals between 25 and 85°C. Significant loss of dispersion between 75 and 80°C was taken as evidence of thermally induced unfolding. On return to 25°C, the original spectrum was restored, with negligible associated precipitation of the sample.
NMR Relaxation Measurements-The oligomeric state of RBBP6(249 -335) was monitored as a function of concentration using 15 N relaxation measurements. The ratio of transverse and longitudinal relaxation rates, R 2 /R 1 , was measured using a protocol that uses an a priori estimate of the correlation time and measures the deviation of the true value from this estimate (31). A set of three 1 H-15 N HSQC spectra was measured at each concentration, using a sequence that incorporates an R 1 relaxation delay as follows: (a) 125.4 ms, (b) 376.2 ms, or (c) R 2 relaxation delay of 25.8 ms. Peak intensities yield R 2 /R 1 values according to Equation 3. 5 To assess the oligomeric state of wild type and mutant RBBP6(249 -335), expected R 2 /R 1 values for monomeric and dimeric proteins were estimated using an empirical relationship between the number of amino acids and the correlation time for overall rotation described by Daragan and Mayo (32). These values were then used to determine 15 N R 1 and R 2 relaxation delays that should yield identical peak intensities if the true R 2 /R 1 value matches the expected value. One-dimensional 15 N-edited 1 H spectra were recorded with an R 2 relaxation delay of 51.56 ms and R 1 relaxation delays of 200.9 ms (monomer) and 723.2 ms (dimer) and compared.
Structure Determination-Assignment of backbone and side-chain resonances was carried out using standard triple resonance procedures (33). NOEs were collected from homonuclear NOESY, 15 N-edited NOESY, and 13 C-edited NOESY spectra centered on the aliphatic and aromatic regions of the 13 C spectrum, respectively. Restraints on backbone torsion angles were generated from backbone resonance assignments using the TALOS algorithm (34). H N exchange times were measured by resuspending a sample previously lyophilized out of H 2 O in D 2 O and recording a series of 1 H-15 N HSQC spectra over the course of 2 days. Peaks intensities plotted as a function of time were fitted to a two-parameter exponential decay function using ProFit (QuantumSoft, Uetikon am See, Switzerland).
Simulated annealing with automated assignment of NOEderived restraints was carried out using CYANA (35,36). Zinc coordination was imposed by restraining the distance between the zinc ion and the sulfur atom of each cysteine residue to between 2.15 and 2.45 Å. Tetrahedral geometry was imposed by restraining the distance between each pair of sulfur atoms to between 3.55 and 3.95 Å and the distances between the zinc ion and the H ␤ protons of the coordinating cysteine to between 3.0 and 3.6 Å (37). Hydrogen bond restraints were added if the corresponding H N resonance exhibited a measurable H 2 O/ D 2 O exchange time with Ͻ20% fitted error, and if the bond was present in 90% of unrestrained structures.
Docking of monomers was carried out in HADDOCK using the standard parameter set augmented with C 2 symmetry, with the imposition of inter-molecular ambiguous interaction restraints (AIRs) but no NMR-derived intra-molecular restraints. AIRs were generated by specifying residues with dimerization-induced chemical shifts greater than 0.1 ppm as "active" and the complete interface as "passive." Following calculation of the structure of the homodimer using the CANDID protocol in CYANA, the structures were returned to HAD-DOCK and recalculated as done previously, but this time with imposition of all unambiguous NOE-derived restraints from the CANDID runs and AIRs as described above. 400 HAD-DOCK structures were generated of which 32 structures were selected on the basis of low HADDOCK energy, low NOE violations (Ͻ0.5 Å), and validation scores from WHATCHECK (38) and PROCHECK NMR (39).
Protein-Protein Interactions-The yeast two-hybrid screen was carried out as described previously (15) using a cDNA bait corresponding to residues 1-81 of human RBBP6. His 6 -tagged proteins IntC and toxin A were used as dummy interactors in the His 6 pulldown assay, and GST-tagged peroxiredoxin Prx2 from Xerophyta viscosa was used in the GST pulldown assay. Assembly of the pGFP-RBBP6 construct was described previously (15).

RESULTS
Sequence Analysis-An alignment of the RING finger-like sequences from orthologues of human RBBP6 is shown in supplemental Fig. 1, along with a number of other U-box and RING finger domains. In the overwhelming majority of eukaryotic genomes, the domain forms part of a single copy gene alongside an N-terminal DWNN domain and a CCHC zinc knuckle domain (see Fig. 1A). However a distantly related form of the protein found in the primitive single-celled eukaryotes Tetrahymena thermophila (XP_001020977) and Paramecium tetraurelia (XP_001434055) contains the DWNN domain fused directly to the RING finger domain, followed by a YT521-B homology (YTH) domain, which, interestingly, is also found in splicing factors (40). YTH domains bind single-stranded mRNA (41), and it is therefore possible that they play the same role as the CCHC zinc knuckle in these proteins. A number of fungal genomes contain a closely related copy of the domain at the N terminus of an otherwise apparently unrelated protein.
The primitive eukaryote Giardia lamblia is the only fully sequenced eukaryotic genome in which we were unable to identify an RBBP6-like protein.
The pattern of eight cysteine residues is highly conserved across the RBBP6 family, suggesting that the domain binds two zinc ions in common with RING fingers. The second pair of cysteine residues forms a highly characteristic "CC" motif that is almost completely conserved across the RBBP6 family and differs from the CX 1-2 H found at this position in C3HC4 RING fingers or the CX 1-2 C found in C4C4 RING fingers. The only exceptions we were able to find were the primitive single-celled eukaryotes Encephalitozoon cuniculi and Enterocytozoon bieneusi, in which the C4C4 motif reverts to the canonical C3HC4 motif. In a number of fungi, including yeasts Saccharomyces cerevisiae and Pichia pastoris, the first and third pair of cysteine residues, which together coordinate the same zinc ion, have been lost, suggesting that these proteins bind only one zinc ion, with the other ion possibly being replaced by a hydrogen bond network similar to that found in U-boxes. In other fungi, such as the fungus Aspergillus niger, the second cysteine has been replaced by aspartic acid.
In RBBP6 sequences from higher eukaryotes an asparagine residue is found at the position occupied by histidine in C3HC4 RING fingers (see supplemental Fig. 1), raising the question as to whether it is involved in zinc coordination (42). Surveys of metal-binding proteins in the Protein Data Bank suggest that direct coordination of zinc ions by asparagine residues (socalled "first shell" ligands) is highly unlikely (43); however, they are found to stabilize zinc-binding sites by forming hydrogen bonds with the sulfur atoms of coordinating cysteine residues, so-called "second shell" ligands.
Bacterial Expression and NMR Resonance Assignment-A peptide corresponding to residues 236 -335 of human RBBP6 was expressed in Escherichia coli, and sequence-specific assignment was performed according to standard triple resonance NMR procedures (33). A 1 H-15 N heteronuclear NOE spectrum (44) showed that residues 251-333 constitute a rigid domain, with residues 236 -248 forming a flexible N-terminal extension (Fig. 1B). A shorter peptide corresponding to residues 249 -335 was therefore expressed and used for all subsequent analyses. With the exception of the excluded residues, the 1 H-15 N HSQC spectrum of the new fragment was essentially the same as that of the previous fragment (data not shown); however the 1 H-15 N HSQC spectrum of a shorter fragment beginning at residue 255 revealed it to be unfolded (data not shown), leading us to conclude that residues 249 -335 constitute the minimal independently folding domain. The domain was found to be highly stable in solution, with a number of samples remaining stable for more than a year at 4°C with minimal deterioration of the 1 H-15 N HSQC spectrum. Preliminary NMR data suggest that the domain has high thermal stability, with at least some elements of structure surviving up to 75°C, as shown in supplemental Fig. 2. The H hydroxyl resonance of Tyr 279 , which is usually not observed due to exchange with the solvent, appears at the unusual shift of 6.0 ppm (compared with an expected shift of 9.13 Ϯ 2.70 ppm), suggesting that it is protected from the solvent by hydrogen bonding. Similarly, the H 1 and H 2 resonances of Arg 285 , which are usually not observed due to exchange with the solvent, appear as broadened peaks in the 1 H-15 N HSQC spectrum at (␦H, ␦N) ϭ (7.0, 72.1) and (6.7, 72.1) ppm, respectively (boxed in Fig. 1C), suggesting that they are also involved in hydrogen bonding or salt bridges.
Metal Binding-Electrospray ionization mass spectrometry revealed a mass difference of 127 Da between the native and the denatured states of the protein (Fig. 2), which is consistent with binding of two Zn 2ϩ ions (M r 65.4 each) by the native state.
In the presence of EDTA, the protein did not immediately unfold but precipitated slowly over the course of 1-2 weeks, suggesting that the zinc ions are very tightly bound. The protein remained unfolded even after removal of the EDTA, as determined from its 1 H-15 N HSQC spectrum, but it folded immedi-ately on addition of 2 mM ZnSO 4 , yielding a spectrum indistinguishable from the original (data not shown). Investigation of the pH dependence of folding by monitoring the one-dimensional proton NMR spectrum revealed that the protein unfolded almost completely when the pH was reduced from 6.0 to 5.5 (data not shown), which is consistent with protonation of the cysteine thiolate groups and corresponding loss of Zn 2ϩ ions at lower pH (45,46).
Following addition of 2 mM CdCl 2 , the 1 H-15 N HSQC spectrum of the zinc-bound form was replaced by a cadmiumbound form over the course of a 2-week period (Fig. 3A). The stability of the cadmium-bound form appeared to be comparable with that of the zinc-bound form, surviving for a number of weeks in cadmium-free buffer. Following dialysis into a buffer containing 2 mM ZnCl 2 , the 1 H-15 N HSQC spectrum returned to its original zinc-bound form over approximately the same time period, indicating that the exchange was fully reversible.
The 1 H-15 N HSQC spectrum of the cadmium-bound form was assigned by comparing 15 N-edited NOESY spectra recorded on each form, although in the majority of cases the shifts were sufficiently small to make this unnecessary. The magnitudes of the 1 H and 15 N chemical shift changes are shown as a function of residue number in Fig. 3B. Peak intensities were extracted as a function of time (supplemental Fig. 3) and used to determine the metal ion exchange rate for each residue. Two different rates are readily discernible in Fig. 3C, corresponding to binding Sites 1 (0.22 Ϯ 0.06 day Ϫ1 ) and 2 (0.35 Ϯ 0.10 day Ϫ1 ), respectively. These rates are approximately an order of magnitude slower than those obtained for other C4C4 RING domains (see Table 1), despite the fact that the measurements were made at pH 6 (as compared with pH 7 for the other two proteins in Table 1), where the rates might be expected to be faster in view of the fact that the protein starts to unfold just below pH 6. Cd 2ϩ /Zn 2ϩ exchange was monitored directly using a directly detected one-dimensional 113 Cd NMR spectrum. The peak at 90 ppm (see Fig. 4A) corresponds to 113 Cd-EDTA, and those at 729 and 711 ppm to 113 Cd 2ϩ ions bound into the two sites of the protein, respectively. 113 Cd chemical shifts in the range 700 to 750 ppm correspond to tetrahedral coordination, with octahedral coordination in the vicinity of Ϫ110 ppm (47), from which we conclude that the Cd 2ϩ ions are tetrahedrally coordinated in both sites. During the Zn 2ϩ /Cd 2ϩ exchange experiment, the amplitudes of the 113 Cd-EDTA peak decreased, and the two protein-bound peaks increased at a rate qualitatively consistent with those measured using 1 H-15 N HSQC spectra (data not shown), with the slower exchange rate corresponding to the resonance at 711 ppm. We conclude that the resonance at 711 ppm corresponds to the ion coordinated by the first and third pairs of cysteine residues (Site 1) and 729 ppm to the ion coordinated by the second and fourth pairs (Site 2).
To confirm the identities of the residues coordinating each ion, a 1 H-113 Cd HMQC spectrum was recorded with the aim of observing correlations between the H ␤ nuclei belonging to the coordinating cysteine residues and the bound 113 Cd 2ϩ ions. Fig.  4B shows that there is coherence transfer between the two cadmium nuclei and a number of 1 H nuclei with resonances around 3 ppm, which is characteristic of the H ␤ nuclei of cysteine residues. Closer analysis allowed assignment of Cys 262 , Cys 280 , and Cys 283 in Site 1 and Cys 274 , Cys 275 , and Cys 299 in Site 2. Although Cys 259 and Cys 296 were not observed in the 1 H-113 Cd HMQC spectrum, they were assumed to be involved in coordination due to their conservation and the fact that the 113 Cd chemical shifts are both consistent with tetrahedral coordination. Furthermore, because both cysteine residues in the "CC" motif (Cys 274 and Cys 275 ) were directly observed in the 1 H-113 Cd HMQC, we conclude that RBBP6(249 -335) is indeed a C4C4 RING finger domain.
Homodimerization-To investigate whether the RBBP6 RING finger-like domain homodimerizes in vitro, we performed a concentration series to observe the effect of concentration on backbone chemical shifts and relaxation properties. Fig. 5 shows a superposition of 1 H-15 N HSQC spectra recorded at concentrations ranging from 1000 to 25 M. There is a marked concentration-dependent shift in a number of residues and little or no change in the rest, consistent with a change in the oligomeric state of the molecule rather than an overall conformation change. 15 N relaxation measurements, which provide an estimate of the correlation time for isotropic tumbling in solution (48), confirmed that at the lowest concentrations the molecule is almost completely monomeric, although at higher concentrations it is almost completely dimeric (supplemental Fig. 4).
Chemical shift changes were plotted as a function of concentration (supplemental Fig. 5) and fitted to a homodimeric binding curve (Equation 1 under "Experimental Procedures"), from which the effective K D value of dimerization was extracted for  15 N chemical shift changes resulting from exchange of cadmium for zinc. Nitrogen chemical shift changes are scaled as described under "Experimental Procedures" to represent the true biological significance. Circles above the bars indicate slowly exchanging backbone amides. C, rates of decay of the zincbound form as a function of residue number. Two rates are broadly discernible, correlating with residues close to Site 1 (gray; slowly exchanging) or Site 2 (sepia; rapidly exchanging). On the basis of their rapid exchange, the sidechain NH 2 resonances of Asn 277 can be assigned to Site 2. Solution Structure-Because chemical shift perturbations suggested that dimerization did not significantly alter the structure of the monomer, monomeric structures were first calculated using CYANA and then "soft-docked" using HADDOCK (49) with the aid of AIRs generated from the chemical shift perturbations. On the basis of the docked homodimers, a number of unambiguous inter-molecular NOEs were identified that were used to guide a full CYANA calculation of the homodimer structure. Finally, the structures were returned to HADDOCK and subjected to refine-ment in explicit solvent (not available in CYANA) using the complete set of unambiguous restraints generated by CYANA, including 146 intermolecular NOE-derived restraints.
Superposition of the 32 best structures yielded average r.m.s.d. values of 0.89 Ϯ 0.23 Å for backbone heavy atoms and 1.21 Ϯ 0.25 Å for all heavy atoms in structured regions. A summary of restraints and structural statistics is presented in Table  2. A superposition of the backbone traces and a representative schematic are shown in Fig. 6, A and B, respectively. Analysis of a representative structure by the PISA server revealed that the homodimer buries 787.9 Å 2 of surface area with a free energy change of ⌬G i ϭ Ϫ8.3 kcal⅐mol Ϫ1 . The p value of 0.18 indicates that the homodimer most probably  Arrows point in the direction of decreasing concentration. In many cases the spectrum of the K313E mutant (in green) corresponds to the low concentration limit of the wild type spectra, indicating that K313E is monomeric. The horizontal lines in the right-hand panel join the side-chain NH 2 resonances of Asn 312 , which are severely broadened in the wild type (dotted line; highest concentration) but much narrower in the mutant (solid line). As with the backbone amides, the position of the side-chain NH 2 of Asn 312 in the mutant corresponds to the low concentration limit of the wild type spectra, indicating that K313E is monomeric. results from a specific interaction (p Ͼ 0.5 corresponds to a random or nonspecific interaction).
The secondary structure includes the ␤-␤-␣-␤ motif typical of RING finger domains, with three short ␤-strands forming a triple-stranded anti-parallel sheet (␤ 1 , residues 270 -272; ␤ 2 , residues 276 -279; ␣ 1 , residues 281-290; and ␤ 3 , residues 310 -311). The core is made up of the side chains of Val 270 , Ile 272 , Tyr 279 , Ile 284 , Leu 288 , Val 304 , and Leu 309 , which correspond to conserved hydrophobic residues in most RING domains and U-boxes (see supplemental Fig. 1). In addition, the domain contains the C-terminal helix (␣ 2 , residues 314 -325) found in many U-boxes, which packs against a structured N-terminal loop (residues 252-257). This creates a second core region made up of the hydrophobic side chains of Ile 253 , Pro 254 , Leu 257 , and Ile 265 , contributed by the N-terminal strand, and Val 319 and Phe 322 , contributed by the C-terminal helix, all of which are conserved in RING and U-box domains featuring the C-terminal helix. The packing of Ile 253 and Pro 254 against the side chain of Phe 322 accounts for the fact that deletion of these residues in RBBP6(255-335) resulted in an unfolded protein.
The structure contains a large number of protected amides, represented by open circles in Fig. 3B. Although some of these can be explained on the basis of hydrogen bonding in secondary structural regions (shown in schematic form at the bottom of Fig. 3B), others such as Leu 261 , Cys 262 , Cys 264 , Cys 275 , Asn 277 , Cys 280 , and Thr 298 cannot be accounted for in this manner. Furthermore, as can seen from Fig. 3B, these correspond to the residues with the largest 1 H chemical shift changes on cadmium binding. Analysis of the structures suggests that these correspond to hydrogen bonds between backbone/side-chain amides and the S ␥ atoms of cysteine residues involved in zinc coordination. A corresponding pattern of hydrogen bonds between backbone amides and residues taking the places of the cysteine residues is found in U-boxes, where they provide the

Solution Structure of RING Finger-like Domain of RBBP6
MARCH 2, 2012 • VOLUME 287 • NUMBER 10 stabilization required to make up for the lack of zinc coordination. Our data therefore suggest that the domain is stabilized by a similar network of hydrogen bonds to that found in U-boxes, in addition to the stabilization provided by zinc binding. To our knowledge, this has not been previously reported in other RING finger domains. A search of the Protein Data bank by the Dali server (50) using the structure of the monomer (residues 252-326) found that the structure most closely resembles a number of U-boxes, including those from murine CHIP (PDB code 2C2L, Z-score 11.1, r.m.s.d. 1.6 Å) and yeast Ufd2p (PDB code 3M62, Z-score 10.9, r.m.s. d. 1.8 Å). The highest ranking RING finger proteins were TRIM37 (PDB code 3LRQ, Z-score 10.4, r.m.s.d. 2.1 Å) and the Polycomb Group RING finger proteins RING1b and Bmi1 (PDB code 2CKL, Z-score 10.1, r.m.s.d. 1.9 Å), all of which have been described as U-box-like (51). CHIP and TRIM37 form homodimers and RING1b and Bmi1 a heterodimer, whereas Ufd2p is monomeric. Superposition of the full homodimer on the U-boxes from CHIP, TRIM37, and RING1b/Bmi gives Z-scores of 19.9, 16.7, and 16.1, respectively.
An overlay of the backbone schematics of RBBP6 and CHIP is shown in Fig. 6C and the corresponding structural alignment in Fig. 6D. Equivalent residues in the interface are listed in supplemental Table 2. The largest part of the interface is formed by the two ␤-sheets, which come together to create a pseudo sixstranded ␤-barrel burying hydrophobic residues Val 271 , Pro 273 , Gly 276 , and Ile 310 from each monomer. Phe 314 docks into a deep pocket in the opposite monomer formed by Leu 257 , Leu 260 , Met 266 , and Leu 315 , a conclusion supported by the presence of a number of inter-molecular NOEs and the broadening of the backbone resonances of Phe 314 .
In a number of U-boxes a pivotal role is played by an asparagine (e.g. Asn 284 in murine CHIP) positioned close to the 2-fold axis, which forms a reciprocal pair of hydrogen bonds with its counterpart in the opposite monomer (52). Asn 312 occupies the equivalent position in this case; large dimerization-induced perturbations of the chemical shifts of the backbone H N and side-chain H ␦2 resonances (supplemental Table 1) suggest that it makes similar reciprocal hydrogen bonds as found in the U-boxes. The structures also provide evidence for a hydrogen bond between Ser 278 and Asn 312 , which is supported by broadening of all resonances belonging to Ser 278 . The equivalent interaction is observed in crystal structures of CHIP (PDB codes 2F42 and 2C2L, although Ser 278 is substituted by a threonine residue), where it is described by Xu et al. (53) as "orienting" the asparagine for its reciprocal interaction across the interface. In the hydrophobic context of the rest of the interface, this network of polar interactions is likely to exercise a strong stabilizing effect on the homodimer. Mutation of Asn 312 to aspartic acid (N312D) abolished the homodimer completely, yielding a 1 H-15 N HSQC spectrum closely approximating the low concentration limit of the concentration series (data not shown).
Severe broadening of all resonances of Lys 313 suggested that it may participate in a stable intermolecular salt bridge with the acidic patch formed by Asp 255 and Glu 256 on the opposite monomer. Mutation of Lys 313 to glutamic acid (K313E) also completely abolished the homodimer, yielding a 1 H-15 N HSQC spectrum closely approximating the low concentration limit of the concentration series (shown in green in the overlay in Fig. 5). The monomeric nature of both N312D and K313E mutants was confirmed using 15 N relaxation (supplemental Fig. 4).
The H resonance of Tyr 279 is protected from exchange with the solvent, indicating that it is involved in a stable hydrogen bond. The structure confirms that Tyr 279 is completely buried, extending through the core from the N-terminal end of helix ␣ 1 into loop 2. In RING fingers this residue is almost always large and hydrophobic (see supplemental Fig. 1), but in both the RBBP6 family and the U-box family tyrosine is most common at this position. Formation of a hydrogen bond at the very end of the side chain enables Tyr 279 to play the role of a reinforcing girder, contributing to the rigidity of the molecule. Although the identity of the hydrogen bond donor is not certain, possible candidates include the S ␥ atom of Cys 296 and the N ␦2 side-chain atom of Asn 277 . In crystal structures of CHIP, the H atom of the corresponding tyrosine residue forms a hydrogen bond with the residue corresponding to Cys 296 (Tyr 253 H -Asn 269 O ␦1 in mouse, PDB code 2C2L; Tyr 233 H -Asp 249 O ␦2 in zebrafish, PDB 2OXQ), suggesting that Cys 296 may indeed be the acceptor in this case.
The structures suggest that the side-chain NH 2 group of Asn 277 forms a hydrogen bond with the S ␥ atom of Cys 275 in Site 2, providing a possible explanation for its highly unusual chemical shifts. The chemical shift perturbations on cadmium exchange (Ͼ0.1 ppm for both H ␦2 protons) suggest a stable association with the coordination center. Interestingly, the orthologous domain from S. cerevisiae (see supplemental Fig. 1) contains an asparagine in a similar position showing virtually identical chemical shifts, 6 which suggests that the effect may be a feature of side-chain NH 2 groups involved in NH-S hydrogen bonds. Ongoing structural analysis of other members of the RBBP6 family may shed more light on this.
Analysis of the structures indicates a possible long range salt bridge between Asp 268 and Arg 316 , which is also present in PUB14 (54), where arginine is replace by lysine (Asp 263 -Lys 308 ), and possibly also in CHIP. This interaction would account for the high level of conservation at both of these positions in the RBBP6 and U-box families. We see no evidence for the salt bridges corresponding to Lys 262 -Glu 271 and Lys 262 -Glu 321 seen in PUB14 (54).
The appearance in the 1 H-15 N HSQC spectrum of both sidechain NH 2 groups of Arg 285 (see Fig. 1C) indicates that these groups are involved in stable interactions. Candidate interactions include a long range hydrogen bond with Pro 306 O and a short range salt bridge with Asp 281 , which is adjacent to Arg 285 on helix ␣ 1 . Both of these interactions would require Arg 285 to lie parallel to the surface of the molecule, presumably contributing to its stability. In support of this, crystal structures of CHIP (PBD codes 2OXQ and 2C2L) show the equivalent salt bridge involving an arginine oriented parallel to the surface (Arg 255 and Glu 259 in murine CHIP), although the positions of the interacting residues are swapped (see supplemental Fig. 1). Following mutation of Asp 281 to alanine (D281A) the side-chain NH 2 groups of Arg 285 were no longer observable in the 1 H-15 N HSQC spectrum.
Interaction with Chaperones-RBBP6 orthologues in lower eukaryotes consist only of the DWNN domain, the CCHC zinc knuckle, and the RING finger-like domain, suggesting that this may contain the core ubiquitin-ligase function of the protein.
Because the zinc knuckle is likely to function as an RNA-binding domain, the DWNN domain remains as the most likely substrate-binding domain.
To identify potential substrates for ubiquitination, a yeast two-hybrid screen with the human DWNN domain as bait was used to screen a human testis expression library as described previously (15). Five potential interacting clones were identified, which encoded two copies of the last 82 residues of heat shock 70-kDa protein 14 (HSPA14), one copy of the last 72 residues of heat shock protein 40 (DNAJB1), two copies of the core splicing protein SmG (full-length), and one each of RANBP9, Gametogenin 1a, and Niemann-Pick disease type C2 (see supplemental Table 3). In the light of the similarities between RBBP6 and the U-box proteins reported above, the interactions with heat shock proteins were considered to be highly significant. The other hits will be pursued elsewhere.
HSPA14 is a close homologue of heat shock protein 70 (HSPA1A), containing the N-terminal ATPase domain and the substrate-binding domain but lacking the 10-kDa C-terminal subdomain (55,56). Because the last 82 residues of HSPA14 fall within the substrate-binding domain of HSPA1A, which also forms part of Hsp70 (HSP1A), we surmised that it should therefore also bind to Hsp70 (HSPA1A). To confirm this, we expressed His 6 -tagged Hsp70 and GST-tagged DWNN (RBBP6(1-81)) in E. coli and carried out His 6 affinity pulldown assays, detecting with a monoclonal antibody against DWNN (Fig. 7A). Hsp70 was able to precipitate the DWNN domain but two unrelated His 6 -tagged proteins were not.
When the reciprocal assay was carried out, precipitating GST and detecting with an antibody against Hsp70 (Fig. 7B), RBBP6 fragments containing the DWNN domain were able to precipitate Hsp70, but an unrelated GST fusion protein was not.
Finally, to investigate whether the interaction is realized in vivo and in the context of the complete RBBP6 protein, GFPtagged RBBP6 was transfected into mammalian cells and its ability to precipitate endogenous Hsp70 investigated. Fig. 7C shows that GFP-RBBP6 was able to precipitate endogenous Hsp70, but GFP alone was not.

DISCUSSION
The cysteine-rich region of retinoblastoma binding protein-6 (RBBP6(249 -335)) has previously been characterized both as a zinc-binding RING finger and as a U-box. Our results confirm that the domain binds two zinc ions in the characteristic "cross-braced" fashion and that the Zn 2ϩ ions could be replaced by Cd 2ϩ ions without significantly affecting either the stability or the conformation. However, the rates of exchange were found to be more than 10-fold slower than any previous measurement, suggesting that the structure is unusually stable. The presence of scalar coupling between the cadmium ion in Site 2 and the H ␤ protons of both Cys 274 and Cys 275 , combined with lack of evidence of coupling between the cadmium ion and the side-chain NH 2 group of Asn 277 , suggests that the structure should be classified as a C4C4 RING finger rather than a C3NC4 RING finger. Asn 277 most probably fulfils the role of a second shell ligand, donating a HN-S hydrogen bond to the S ␥ of Cys 275 .
The structure contains a large number of hydrogen bonds, including a number of NH-S hydrogen bonds between backbone NH groups and S ␥ atoms of cysteines coordinating zinc ions. These NH-S bonds correlate well with NH groups with large chemical shift perturbations on Cd 2ϩ /Zn 2ϩ exchange, a relationship that has been noted previously (57). Hydrogen  2). Equal amounts of GST-DWNN were present in each lane. WB, Western blot. B, in the reciprocal GST affinity assay, GST-DWNN and a larger GST-tagged fragment of RBBP6 (RBBP6(1-335)) were able to precipitate Hsp70-His 6 (lanes 2 and 3), but an unrelated GST-fusion protein was not (lane 1). Equal amounts of Hsp70-His 6 were present in each lane. C, following transfection of GFP-tagged full-length RBBP6 into mammalian cells (HEK293), GFP-RBBP6 was able to precipitate endogenous Hsp70 (as detected using an antibody against Hsp70), whereas GFP alone was not. bonding networks of this kind are hallmarks of U-boxes, in which they compensate for the lack of zinc coordination. It therefore appears that the RBBP6 domain may have inherited both mechanisms of stabilization, resulting in its being doubly stabilized. This would account for the slow rate of metal ion exchange, the high thermal stability, and the long term stability in solution.
It may also explain the fact that zinc coordination appears to be at least partially dispensable in some organisms, including S. cerevisiae and P. pastoris, where the first zinc-binding site has been lost, and A. niger, where one of the coordinating cysteines has been replaced by an aspartic acid, making it likely that the corresponding zinc ion has either been lost or else is only weakly bound. The family therefore represents a structural intermediate between the RING finger family and the U-box family, containing examples intermediate between the two. It is tempting to speculate that the family also represents an evolutionary intermediate on the path from RING fingers to U-boxes, but a more detailed genomic analysis will be needed to substantiate this notion.
The solution structure closely resembles those of the U-box family, in particular the U-box from the C terminus of Hsp70 interacting protein (CHIP). The domain dimerizes along the identical interface to that found in CHIP, and many of the residues in the interface are conserved, including Asn 312 and Ser 278 (substituted by threonine in CHIP), which form a network of inter-molecular polar interactions in the middle of an otherwise highly hydrophobic interface. Asn 312 is highly conserved in higher eukaryotes but is frequently replaced by aspartic acid in lower eukaryotes. In view of the fact that the N312D mutant disrupts the homodimer, it is tempting to propose asparagine at this position as an indicator of whether a particular orthologue will homodimerize. In support of this model, preliminary investigations suggest that the domain from S. cerevisiae, which has aspartic acid at the corresponding position (see supplemental Fig. 1), is monomeric in solution. 7 However, two U-boxes which are known to be monomeric, yeast Ufd2p, and mammalian E4B (58), both have asparagine at this position so the presence of asparagine is not sufficient for dimerization.
The K D value of homodimerization is found to be on the order of 100 M, which corresponds to a weak interaction. This value is similar to those reported for U-boxes Prp19 (59) and PUB14 (54), both of which exhibited evidence of a mixed monomer/dimer at NMR concentrations. Although such a weak interaction is unlikely to be physiologically significant, the isolated domain does not represent a physiologically realistic scenario; it is possible that the complete protein homodimerizes with greater affinity. Because homodimerization has been shown to be necessary for the ubiquitination activity of CHIP (18), and for native functioning of Prp19 (59), it will be of interest to determine whether homodimerization has functional significance for RBBP6. The K313E and N312D mutants reported here, both of which abolish homodimerization, will be useful reagents in establishing this.
The E2-binding site on U-boxes and RING fingers is located in a hydrophobic groove lying between helix ␣ 1 and loops L1 and L2, which contain the two zinc ions in the case of RING fingers. In human RBBP6, the residues surrounding the groove are Ile 261 , Ala 287 , and Pro 297 , of which Ile 261 and Pro 297 are highly conserved across the RBBP6 family and in U-boxes. In CHIP and BRCA1, mutation of the residue corresponding to Ile 261 to alanine leads to the abolition of E2 binding and consequent loss of activity, without affecting the dimerization state of the molecule (60,61). Based on our structure, we therefore predict that mutating Ile 261 to alanine is likely to abolish ubiquitination activity.
U-boxes are currently classified primarily as "non-zinc-binding" RING fingers. From a functional point of view, U-boxes are differentiated from RING fingers by their interaction with chaperones, most notably with Hsp70, Hsp40, and Hsp90, and their involvement in protein quality control. The strong structural similarities presented here between the cysteine-rich domain of RBBP6 and U-boxes, combined with its association with chaperones Hsp70, Hsp40, and Hsp90, suggests that it should be classified as a U-box, notwithstanding the fact that it binds two zinc ions. Two similar examples have already been reported, the RING1b/Bmi1 heterodimer and the TRIM37 homodimer, both of which bind zinc ions despite being structurally more similar to U-boxes.
The structure has been deposited in the Protein Data Bank (PDB) under accession number 3ZTG. During the preparation of this work, the NMR structure of residues 249 -309 of human RBBP6 was deposited in the PDB under accession number 2YUR without associated report. 2YUR is similar in many respects to the structure presented here, but the expressed fragment lacks the third strand of the triple-stranded ␤-sheet and the entire C-terminal helix and is therefore unlikely to be able to homodimerize. The Z-score between 2YUR and this structure is 6.8, compared with 11.1 between this structure and CHIP.
In conclusion, the strong similarities between the RING finger-like domain of RBBP6 and members of the U-box family, combined with its association with heat shock proteins, lead us to propose that RBBP6 plays a role in chaperone-mediated ubiquitination of unfolded protein substrates. Additional investigations will be needed to establish the validity of this hypothesis. In addition, the growing number of zinc-binding U-boxes suggests that new criteria, based ideally on function rather than zinc binding, are required to distinguish between U-boxes and RING fingers.