|Home | About | Journals | Submit | Contact Us | Français|
As a signal for substrate targeting, polyubiquitin meets various layers of receptors upstream to the 26S proteasome. We obtained structural information on two receptors, Rpn10 and Dsk2, alone, and in complex with (poly)ubiquitin or with each other. A hierarchy of affinities emerges with Dsk2 binding monoubiquitin tighter than Rpn10 does, whereas Rpn10 prefers the ubiquitin-like domain of Dsk2 to monoubiquitin, with increasing affinities for longer polyubiquitin chains. We demonstrated the formation of ternary complexes of both receptors simultaneously with (poly)ubiquitin and found that, depending on the ubiquitin-chain length, the orientation of the resulting complex is entirely different, providing for alternate signals. Dynamic rearrangement provides a chain-length sensor, possibly explaining how accessibility of Dsk2 to the proteasome is limited unless it carries a properly-tagged cargo. We propose a mechanism for a malleable ubiquitin-signal that depends both on chain-length and combination of receptors to produce tetra-ubiquitin as an efficient signal threshold.
Post-translational modification of cellular proteins by ubiquitin influences protein-protein interactions, alters recognition by binding partners, and serves to target proteins to cellular compartments (Hicke and Dunn, 2003; Ikeda and Dikic, 2008; Mukhopadhyay and Riezman, 2007; Pickart, 2004). The best-studied outcome of ubiquitination is targeting to the proteasome, which is responsible for the degradation of most cytosolic and endogenous proteins (Cuervo and Dice, 1998; Glickman and Ciechanover, 2002; Hershko and Ciechanover, 1986; Orlowski, 1990; Sahagian and Novikoff, 1994; Thrower et al., 2000). The majority of substrate proteins are tagged not by a single ubiquitin (Ub), but by a polyubiquitin (polyUb) chain in which Ub monomers are linked to each other at a lysine residue on the preceding Ub. It was shown that Lysine-48 linkages are among the most prevalent in the cytosol (Matiuhin et al., 2008). Since polyUb chain lengths of n ≥ 4 are preferred for recognition and efficient degradation by the 26S proteasome (Thrower et al., 2000), K48-linked polyUb is thought to act as a universal signal for targeting proteins to the 26S proteasome. Once bound by the proteasome, substrate proteins are deubiquitinated, unfolded, and subsequently degraded. Even though the general pathway of the ubiquitin-proteasome system is charted, there are still many critical unanswered questions regarding the mechanism of Ub-signal recognition, transport of the ubiquitinated substrates to the proteasome, and their subsequent processing for degradation.
Several protein families have drawn attention due to their ability to recognize and bind both polyUb and proteasomes, therefore, they may possibly function as shuttling factors for polyubiquitinated substrates. One of these putative shuttles is Rpn10, a polyUb-binding protein distributed between the proteasome-bound and unassociated forms (or “pools”) (Fu et al., 1998; Glickman et al., 1998; van Nocker et al., 1996). Other proteins that have garnered a lot of attention as shuttles of polyUb-conjugates have a Ub-like (UBL) domain at (or near) the N-terminus, and a Ub-associated (UBA) domain at the C-terminus. UBL domains can interact with subunits in the proteasome (Elsasser et al., 2004; Elsasser et al., 2002; Mueller and Feigon, 2003; Rosenzweig et al., 2008; Sakata et al., 2003; Seeger et al., 2003; Upadhya and Hegde, 2003), whereas the UBA domain has an affinity for Ub or polyUb, or both (Bertolaet et al., 2001; Hofmann and Bucher, 1996; Kang et al., 2006; Kim et al., 2004; Mueller et al., 2004; Raasi et al., 2004; Raasi et al., 2005; Varadan et al., 2004; Varadan et al., 2005; Wilkinson et al., 2001; Zhang et al., 2008). Some of these UBL-UBA proteins have more than one UBA domain. Prominent members of the UBL-UBA family are Rad23, Dsk2, and Ddi1 (and their mammalian homologues hHR23, hPLIC/ubiquilin, and hDDI1, respectively) (Elsasser et al., 2004; Elsasser and Finley, 2005; Hartmann-Petersen and Gordon, 2004; Hartmann-Petersen et al., 2003; Matiuhin et al., 2008; Verma et al., 2004a; Wilkinson et al., 2001). Several proteasomal subunits have been identified as interacting directly with the UBL domains of Rad23 or Dsk2: Rpn10/S5a, Rpn1, Rpn2, Rpn13, Pre1, and Pre2 subunits (Elsasser et al., 2002; Husnjak et al., 2008; Ishii et al., 2006; Rosenzweig et al., 2008; Saeki et al., 2002), although the primary proteasome UBL-receptor is apparently Rpn1. Because of the bi-functional nature of recognition signals that the UBL-UBA proteins carry, they can bridge ubiquitinated substrates and the proteasome, possibly triggering subsequent events at the proteasome leading to substrate degradation. Underlying their importance, the correct levels of UBL-UBA proteins must be tightly controlled as both hyper- and hypo- levels lead to alterations in the cellular Ub-landscape, proteolytic defects, and in the case of Dsk2 even cell cycle arrest and cell death (Matiuhin et al., 2008). This recent observation singled out Dsk2 as a potent polyUb-binding protein influencing the global Ub-landscape, chain specificity, and rates of cellular protein turnover, although it did not provide the mechanism for how abundance of such a protein leads to pervasive deleterious effects.
Not only is the precise role of polyUb-shuttles still vague, the proteasome itself also has an inherent capacity to bind polyUb chains and polyubiquitinated cargo without need for attachment of auxiliary Ub-binding proteins (Bech-Otschir et al., 2009). Two subunits responsible for anchoring polyUb chains at the proteasome are Rpn10 (Verma et al., 2004a; Young et al., 1998) and Rpn13 (Husnjak et al., 2008; Seong et al., 2007b). The integral proteasomal subunit Rpn10 (and its human orthologue S5a) was the first proteasomal subunit documented to have an affinity for polyUb and as such may function as a direct receptor for polyUb at the proteasome (Deveraux et al., 1994; Elsasser et al., 2004; Fu et al., 2001; Fu et al., 1998; Glickman et al., 1998; van Nocker et al., 1996; Verma et al., 2004a). The Ub-binding activity of Rpn10/S5a has been mapped to a Ub-interacting motif (UIM) (Deveraux et al., 1994; Haririnia et al., 2007; Wang et al., 2005). Even though alternate pathways exist, a significant portion of cellular proteins targeted to the proteasome require this ubiquitin-binding capacity of Rpn10 for efficient turnover by the proteasome (Mayor et al., 2007; Mayor et al., 2005). An early study showed that the UIM-containing part of Rpn10 is dispensable in budding yeast, but is required for degradation of a subset of polyubiquitinated targets (van Nocker et al., 1996). For example, turnover of targets such as Sic1 or Gcn4 in vivo is inhibited in Δrpn10 (Glickman et al., 1998; Seong et al., 2007a; Verma et al., 2004a). Likewise while wild type proteasome is able to degrade polyubiquitinated Sic1 in the absence of UBL-UBA proteins, proteasomes purified from Δrpn10 are defective at degrading Sic1, or polyUb conjugates in vitro (Glickman et al., 1998; Verma et al., 2004a). That Rpn10 can bind polyUb conjugates supports its role as a direct Ub receptor in the proteasome (Glickman et al., 1998; Seong et al., 2007a; Verma et al., 2004a), however, the role of Rpn10 is not limited to anchoring of Ub-conjugates since Rpn10ΔUIM proteasomes degrade some of the same substrates that are stable with proteasomes from Δrpn10 (Fu et al., 1998; Verma et al., 2004a). Moreover, the roles of Rpn10 and Rad23 are not fully redundant since mutations in Ub (L69S, L67S) that abolish Ub binding to Rpn10/S5a but do not affect its binding to Rad23/hHR23A, result in accumulation of non-degraded polyubiquitinated substrates and lethality of yeast cells (Haririnia et al., 2008).
The convoluted network of interactions between substrates and the proteasome is a consequence not only of the multitude of overlapping delivery proteins and receptors, but is also an outcome of the capacity of these Ub-binding proteins to interact with each other. The dual nature of most cargo-delivery proteins – harboring both Ub-like domains and Ub-binding domains – begets the potential for many dimeric partners at or upstream to the proteasome. In this context, S5a had been intensively investigated for interactions with Ub or UBL domains. S5a contains two UIMs, and both have a single-helix structure (Fujiwara et al., 2004; Hofmann and Falquet, 2001; Mueller and Feigon, 2003; Wang et al., 2005; Young et al., 1998), although UIM-2, the one nearest the C-terminus, appears to have a higher affinity for Ub and UBLs (Hiyama et al., 1999; Wang et al., 2005; Young et al., 1998). Hydrophobic residues on UIMs of S5a (Hurley et al., 2006; Wang et al., 2005) recognize the complementary so-called hydrophobic patch (L8, I44, H68, V70) on one face of Ub (Beal et al., 1996; Pickart and Fushman, 2004). The single UIM located at the C-terminus of Rpn10 has high sequence homology to UIM-1 of S5a, however the residues participating in binding of polyUb have not been mapped at the molecular level, and the structure of Rpn10 has not been determined. The two properties of Rpn10, binding polyUb chains and attaching to monoUb-like domains may be mapped to two pools of Rpn10 as an integral proteasome subunit and in the unassociated “free form” (Glickman et al., 1998). Thus, the polyUb-binding role seems to map to the proteasome-associated pool of Rpn10 (Verma et al., 2004a), whereas the affinity for UBL domains has so far been documented for the unassociated pool (Matiuhin et al., 2008).
That UIM-containing proteins from the S5a/Rpn10 family can bind to the Ub signal directly and also to Ub-shuttles of the UBL-containing family, such as hHR23 (Kang et al., 2007) or Dsk2 (Matiuhin et al., 2008) raises interesting possibilities. For instance, proteasomal targeting may not follow independent parallel pathways, but rather a convoluted network, a product of cross-talk between different components. In a recently proposed model (Kang et al., 2007) hHR23A mediates the interaction between polyUb and S5a resulting in a ternary complex. This naturally raises the philosophical question of why would a UBL-containing protein be needed to mediate polyUb targeting? Furthermore, how does the sequence diversity in UBL-containing shuttles complement the linkage/length diversity of their polyUb cargo? In a broader sense, it is still unclear where the polyUb chain-length requirement comes from, or how is tetra-Ub an optimal signal? And ultimately, what is the primary signal that is actually recognized by the proteasome (i.e. chains vs. their shuttles)? In the current study, we chose the three-way relationship between Rpn10, Dsk2, and (poly)Ub as a paradigm for understanding what shapes the ubiquitin signal.
As a first step in addressing the preceding questions we screened the UBL domains of the known Ub shuttles in yeast for their ability to bind the UIM-containing region of Rpn10 (Rpn10204-268): Rad23, Dsk2, Ddi1, Ubp6, and monoUb (as control). Of the four UBLs, only the Dsk2 and Rad23 UBLs showed distinct binding to 15N-labeled Rpn10 in our NMR experiments (Figs. 1A–E, S1). Likewise, the addition of unlabeled Rpn10204-268 to 15N-Dsk2-UBL or 15N-Rad23-UBL at a 1:1 molar ratio caused perturbations in distinct NMR signals of these UBLs (Fig. 1F,H), indicating a specific interaction. Based on the overall smaller magnitude of shifts in 15N-Rpn10 signals caused by Rad23-UBL compared to those caused by Dsk2-UBL (Figs. 1A,B, S1), we conclude that the Rad23-UBL/Rpn10 interaction is relatively weaker. However, in order to directly compare the affinities of Rad23 and Dsk2 for Rpn10, we performed a competition assay (see Figs. 1F-I and S2). Indeed, the addition of Dsk2-UBL at a 1:1 molar ratio to a preformed complex of Rad23-UBL/Rpn10 caused Rpn10-perturbed 15N-Rad23-UBL signals to shift back to their free-state positions, indicating that Dsk2 outcompetes Rad23 for Rpn10 association (Fig. 1H–I). In line with this conclusion, in a reverse competition assay, 15N-Dsk2-UBL signals remained at their Rpn10-bound positions after Rad23-UBL was added in equimolar ratio to 15N-Dsk2-UBL/Rpn10 (Figs. 1F–G, S2). Taken together, these two screens substantiate the unique relationship between Rpn10 and Dsk2.
To verify that an isolated UBL domain possesses the same Rpn10-binding properties as full-length Dsk2, we performed a similar assay, this time with 15N-labeled full-length Dsk2. Note that intramolecular UBL-UBA interactions cause slight shifts of the UBL signals in the context of full length Dsk2 compared to the corresponding signals of free UBL (Fig. 1J–L). However, once in the complex with Rpn10, the NMR signals of the UBL domain of full-length Dsk2 superimpose well with the corresponding peaks of UBL alone in complex with Rpn10 (Fig. 1J–L). This indicates that isolated UBL is recognized by Rpn10 in the same manner as the UBL domain in the context of full-length Dsk2. Furthermore, the NMR spectrum of 15N-labeled full-length Dsk2 in complex with full-length Rpn10 superimposes well with that in complex with Rpn10204-268 (Fig. S3), strengthening the conclusion that it is the UIM-containing region of Rpn10 that comes in contact with Dsk2. These results justify the use of the UBL and UIM domains as a means to study the nature of the interaction between Dsk2 and Rpn10.
These findings agree with the previous study which showed that Dsk2 interacts with Rpn10 through its UBL domain (Matiuhin et al., 2008). Both proteins can also interact with ubiquitin, therefore, in order to understand the competition of these two receptors for each other and for ubiquitin, we first obtained structural information on Rpn10 and Dsk2 independently, mapped out their binding interface, and modeled their binary complexes.
Taking advantage of their modular architecture, we initially focused on structural characterization on the co-interacting domains of Dsk2 and Rpn10; the UBL and UIM domains respectively. Complete NMR resonance assignment for 1H, 15N, and 13C nuclei of Dsk2-UBL domain was carried out previously (Chen et al., 2008). The chemical shifts, NOESY spectra, 15N relaxation data, and residual dipolar couplings measured in Dsk2-UBL all are in general agreement (Fig. S4) with predictions from the crystal structure of Dsk2-UBL (PDB code 2BWF (Lowe et al., 2006)); therefore we conclude that the structure determined in the crystal form is largely preserved in solution, and our NMR signal assignments ((Chen et al., 2008) and Fig. S4) can be used to map the binding interactions.
No 3D structure is available for Rpn10. Using NMR measurements on the UIM-containing construct (Rpn10204-268), we obtained an almost complete assignment of the backbone resonances for the stretch of residues from G204 to L246 (except for Prolines P222 and P226, and, due to signal overlap, Glutamates E237 and E238). Of multiple UIM-containing proteins, the UIM in Rpn10 is most similar to UIM-1 of its orthologue S5a (Hofmann and Falquet, 2001). In fact, the two stretches - residues F218-E244 of Rpn10 and F206-E232 of S5a - share 81% identity and 100% sequence similarity (Fig. S5). Residues E215-E245 within UIM-1 of S5a, form a single α-helix (Wang et al., 2005), and it is natural to expect a similar structure for the homologous region of Rpn10. Indeed, based on the multiple lines of evidence presented in Supplemental Data, the UIM region of Rpn10 (V220-E244) is structurally similar to that of UIM-1 of S5a (V208-E232), with residues E227-E244 defining the hallmark α-helix in Rpn10. We note that the regions N- and C-terminal to the UIM region in the Rpn10204-268 fragment are largely flexible (Fig. S6).
Rpn10 interacts with both Dsk2-UBL (Fig. 1 and (Matiuhin et al., 2008)) and Ub (Figs. 1C, S1), though in fact Rpn10/S5a was initially identified as a multiUb-chain binding protein (MCB1; (Deveraux et al., 1994; van Nocker et al., 1996)). With the structures of Dsk2-UBL and Rpn10-UIM, we set to determine the binding interface of UIM with UBL and compare it to that of UIM with monoUb or K48-linked chains. Since resonance frequency of each nucleus in a protein is sensitive to its local electronic environment, NMR is a sensitive method for mapping interacting sites and for affinity determination (Zuiderweg, 2002). Perturbations in the local electronic environment upon binding cause changes in the NMR spectra in the form of shifts (called chemical shift perturbations, CSPs) or broadening (attenuation) of the signals. Both CSPs and signal broadening are a direct consequence of changes in the local environment due either to direct involvement of the corresponding group (residue) in the binding interaction or to a secondary effect of structural rearrangements. Strong signal attenuation as opposed to CSP usually reflects intermediate or slow (on the NMR chemical shift time scale) exchange between the free and ligand-bound states of the protein, due to slow on/off kinetics. In addition, the broadening effect could also be nonspecific, due to an increase in the apparent size of the molecule (slower tumbling).
Titration of 15N-Rpn10204-268 with monoUb, polyUb, and Dsk2-UBL, allowed us to map and compare the Rpn10 residues involved in binding to these ligands (Fig. 2). In general, the binding surface on Rpn10204-268 comprises a predominantly hydrophobic stretch spanning the canonical LAMAL motif at the beginning of the UIM’s α-helix (L228-S235) as well as some of the disordered residues N-terminal to it (G219-P226).
The surface of Rpn10204-268 perturbed by Ub chains (Ub2 or Ub4) is essentially the same as that involved in monoUb binding (Fig. 2). This contact area is generally similar to the recognition surface on each of the two UIMs in S5a (determined in complex with monoUb or di-Ubs (Haririnia et al., 2007; Wang et al., 2005; Zhang et al., 2009); see also Fig. S7). Even though polymerization leads to tighter Rpn10 binding (evident from strong signal attenuations, Fig. 2), it does not engage additional sites on the Rpn10 surface. This finding suggests that Rpn10 interacts with individual Ub units within the chain. That the mode of recognition remains the same, whether Rpn10 binds mono- or polyUb, is further corroborated by the fact that the NMR signals of Rpn10 shift in the same directions upon titration with monoUb or Ub2 (Fig. S7) indicating a similar change in the chemical environment.
Association of Rpn10 with Dsk2, which also appears tighter compared to monoUb (strong signal attenuations, Fig. 2) expands the interaction region beyond the canonical LAL/MAL motif and includes a stretch of polar α-helical residues (Q240, Q241, R242) C-terminal to it (Fig. 2). In addition, stronger CSPs were observed in the disordered region (e.g. M216, D221) N-terminal to the helix. This suggests that Rpn10-Dsk2 is a tighter partnership compared to Rpn10-monoUb; which will be put to test in the subsequent sections.
The site-specific character of the observed perturbations in 15N-Ub or Ub2 (15N-labeled on proximal Ub) upon titration with Rpn10204-268 (Fig. 3A,B) clearly points to the L8-I44-H68-V70 hydrophobic patch of Ub as the central surface of contact (Fig. 3G). The Rpn10-interacting surfaces on Ub and Ub2 are generally consistent with those involved in S5a binding (Haririnia et al., 2007; Wang et al., 2005; Zhang et al., 2009). It is striking how similar the Rpn10-binding surface of the Ub is, whether as a free unit or conjugated on K48 (Fig. 3D,E); this finding further supports the conclusion from the previous section that the polymerization does not alter the mode of recognition by the Rpn10 receptor. Combined with the fact that the dimensions of the receptor (the UIM unit) are basically identical to a single Ub unit (Fig. 3I), this indicates that the unit of recognition is a single Ub and not a polyUb. Nevertheless the larger number of strongly attenuated residues in Ub2 indicates stronger Rpn10 binding (slower off-rates) in the case of Ub chains. This could be due to a “local concentration” effect, which makes re-binding to the same chain after dissociation more likely.
A similar titration of 15N-Dsk2-UBL with the UIM region of Rpn10 resulted in site-specific perturbations in its NMR spectra covering the surface of the protein centered around residues I45, H69, and V71 (Fig. 3C,F,H). These residues superimpose well with their conserved counterparts I44, H68, and V70 that define the canonical hydrophobic patch on Ub (Fig. 3G,H). However, compared to Ub (Fig. 3A,D), several additional sites on the Dsk2-UBL participate in Rpn10 binding. These include the H61-D64 stretch in the β4/β5 loop and residues K72-R74 C-terminal to β5. We recall that in the reverse titration (15N-Rpn10204-268/Dsk2-UBL, Fig. 2), additional perturbations were also observed in Rpn10 residues Q240-R242 and M216, D221. Both these results point to larger binding surfaces involved in pairing of Rpn10 with Dsk2-UBL than with Ub (Fig. 3D,F). In fact, Dsk2/Rpn10 association is apparently tight given that signals of several UBL (Fig. 3C) and Rpn10 (Fig. 2) residues were severely attenuated.
A model of the Rpn10/Dsk2 complex (Fig. 3J) highlights the complementarity of these additional contacts, most notably between H61-D64 of Dsk2 and N240-R242 of Rpn10. These additional interactions are primarily electrostatic or polar in nature and extend the axis of contact along the α-helix in Rpn10 on one face and the β5 strand (beyond the V71-H69 stretch) in Dsk2-UBL on the matching face (Fig. 3H). These additional contacts could account for the apparently stronger affinity of Dsk2-UBL/Rpn10 vs. Ub/Rpn10 binding.
To validate our homology-based models of the Rpn10/Ub and Rpn10/Dsk2 complexes (Fig. 3I,J), we used site-directed spin labeling, which provides information on the orientation and distance between the two proteins in the complex. Briefly, a nitroxyl spin label, (1-oxy-2,2,5,5-tetramethyl-3-pyrroline-3-methyl) methanesulfonate (MTSL), was covalently attached to Rpn10204-268 at position 247 through disulfide bond to the side chain of a Cysteine introduced as a substitute for R247 (details in Supplemental Data). We chose residue 247 for spin-labeling because it is positioned outside the region involved in interactions with Ub or Dsk2, and its location near the C-terminus of the α-helix should allow unambiguous determination of the UIM’s orientation with respect to its binding partner (see Fig 3). Note that our NMR data presented in Table S1 and Fig. S6 indicate that Rpn10’s α-helix remains intact in the Dsk2-bound state. The unpaired electron spin of MTSL causes distance-dependent attenuation of NMR signals that can be detected at distances up to ~20 Å. This allowed us to reveal residues that are in close proximity to the spin label. Indeed, strong signal attenuations observed in both Ub and Dsk2-UBL upon binding to MTSL•Rpn10 were highly site-specific and all could be mapped to the part of Ub or Dsk2 surface facing the C-terminus of the UIM helix, see Figs. 3K,L and S8. Moreover, as shown in Supplemental Data, the actual location of the unpaired electron of the spin label can be calculated from the collective signal attenuations measured in Ub or UBL. For both complexes, the result (gold ball in Figs. 3K,L and S8) agrees well with the expected location of Rpn10’s residue 247 extrapolated from the orientation of the UIM-1 α-helix (cyan ball, Figs. 3K,L and S8). Therefore we conclude that these spin labeling data provide independent experimental evidence for Rpn10/Ub and Rpn10/Dsk2 complexes, shown in Fig. 3I,J.
Having obtained evidence for stable complexes between UIM and a number of ligands, we set out to quantify the stoichiometry and strength of interactions. In the case of Rpn10 binding to Dsk2-UBL or monoUb, the magnitude of CSPs of residues participating in interactions from either binding partner saturate at approximately 1:1 molar ratio (Fig. 4A,B). Furthermore, the striking similarity of the binding curves observed in the forward and reverse NMR titrations indicates a 1:1 stoichiometry of these complexes.
To quantify the binding interactions, we derived the dissociation constants from the NMR titration data (Fig. 4A–C) and independently by Surface Plasmon Resonance (SPR) measurements (Fig. 4D–F) (details in Supplemental Data). The data from both methods, summarized in Table 1, are in good agreement and indicate that Rpn10 has strong binding preference for Dsk2-UBL over monoUb. The broader interaction surface of Dsk2-UBL may be the underlying mechanism for its slower exchange (i.e. on/off kinetics; Fig. 3A,C) and tighter affinity constant compared to monoUb (Table 1, Fig 4D, E).
The binding surfaces in the Rpn10/Dsk2 complex (summarized in Figs 3J, ,4H)4H) raise an interesting hypothesis that the electrostatic interaction between the charged side chains of D64 on Dsk2 and R242 on Rpn10 accounts for a great deal of the tighter binding measured for Rpn10 with Dsk2 over that with Ub. An alignment of Dsk2-UBL with Ub showed that D64 in Dsk2 corresponds to K63 in Ub (see Table S2). Therefore we mutated D64 to a Lysine. Indeed, a single-point substitution mutation D64K resulted in a significant reduction in the strength of Rpn10/UBL binding (Table 1, Fig. 4D), essentially converting the UBL into a “Ub” molecule (as far as the Rpn10 binding is concerned). This supports an electrostatic contribution to the enhanced affinity of Rpn10 for UBL over that for monoUb (highlighted in Fig. 4G,H).
The tighter binding constant of Dsk2-UBL/Rpn10 association suggests that Dsk2 would outcompete monoUb for binding to Rpn10. Indeed, this property was verified by a direct NMR competition assay (Fig. 5A). At this stage we note that polymerization of Ub has been shown to enhance its affinity for the proteasome, and in particular for the S5a (Beal et al., 1998; Thrower et al., 2000). Interestingly, in a similar assay, Dsk2-UBL was unable to efficiently displace Ub2 on Rpn10 (Fig. 5B), and neither could Ub2 fully displace Dsk2-UBL even when in excess, thus suggesting comparable affinities of the two proteins. This prompted us to rigorously quantify binding affinities of Rpn10 for Ub and polyUb chains relative to that for UBL under identical experimental conditions.
Equilibrium constants determined by SPR for association of Ub(n) with Rpn10-UIM show that affinity increases over two orders of magnitude with increasing chain length from mono to tetra-Ub (Table 1, Fig. 4E,F). Although Dsk2 has a single UBL unit, its affinity for Rpn10-UIM is comparable to that of Ub2, fully five-times tighter than monoUb (Table 1 and Fig 4D,E). As mentioned above, strong NMR signal attenuation (Fig. 2) also pointed to an increase in the affinity of Rpn10 with the length of Ub chain (though strong signal attenuations preclude accurate quantification by NMR). However, NMR competitions confirm that the preferences of Rpn10 in solution are the same as those determined by SPR for immobilized Rpn10-UIM: Ub2 binds Rpn10-UIM comparably to Dsk2-UBL (Fig. 5A,B) while Ub4 can displace Dsk2-UBL quite efficiently (Fig. 5B).
The same trend in the binding preferences holds for full-length Rpn10, as evident from a separate direct competition assay using biochemical pulldowns (Fig. 5E). Extensive washes prior to elution render biochemical pulldowns sensitive to fast off-rates. Thus, monoUb is not retained at all on an affinity column of immobilized Rpn10 (Fig. 5E) in line with the low affinity and fast exchange rates measured for this pair (Figs. 2–5 and Table 1). However, the more “sticky” substrates, UBL and Ub4, are efficiently isolated via the Rpn10 column. When present simultaneously in solution, Ub4 outcompetes UBL (Fig. 5E), supporting the results from NMR competitions (Fig. 5B). Together, the affinities measured by SPR or NMR titrations (summarized in Table 1), NMR competition studies, and protein pulldowns, demonstrate a hierarchy in affinities for Rpn10 association: Ub4 > Ub2 ≈ Dsk2-UBL > monoUb.
At this point we recall that Dsk2 can interfere with Ub/Rpn10 binding not only by presenting a UBL domain to the UIM motif (this study and (Matiuhin et al., 2008)) but also by its intrinsic affinity for Ub through its UBA domain (Funakoshi et al., 2002; Rao and Sastry, 2002). It has already been shown that UBA domains from the Dsk2 family are among the tighter Ub recognition elements (Ohno et al., 2005; Raasi et al., 2005; Zhang et al., 2008). Indeed, our SPR measurements confirm the Dsk2-UBA’s strong affinity even for monomeric Ub (Fig. 4F, Table 1). Therefore we performed NMR competition assays to compare the relative affinities of UBA and Rpn10 for polyUb species (Ub2 and Ub4). In these experiments the UBA of hPLIC-1/Ubiquilin-1 (UQ1), the human version of Dsk2, was used, the structure of which is remarkably similar to that of Dsk2-UBA (Zhang et al., 2008). UQ1-UBA has been extensively studied by NMR in our laboratory (Raasi et al., 2005; Zhang et al., 2008). Most importantly, the UBA domains from the two orthologues (UQ1 and Dsk2) have almost identical polyUb binding properties (Raasi et al., 2005) and both have strong affinities for monoUb (Ohno et al., 2005; Raasi et al., 2005; Zhang et al., 2008). It is important to note that the surface of Ub recognized by either Rpn10-UIM (Fig. 3) or Dsk2/UQ1 UBAs (Ohno et al., 2005; Zhang et al., 2008) is the same and centered around the same hydrophobic patch.
That both Rpn10 and Dsk2 interact with the same residues in Ub precludes the possibility that they could share the same Ub unit. Indeed, and in agreement with the higher affinity of Ub for UBA over Rpn10 (Table 1), adding UBA to a preformed complex of monoUb and 15N-labeled Rpn10 essentially caused a complete transfer of the Ub molecule to UBA (Figs. 5C, S9). Interestingly, in a similar assay involving an equimolar ratio of UBA, Ub2, and Rpn10, the UBA also efficiently trapped Ub2 (Figs. 5D, S9). For now we conclude that binding of UBA, with the higher affinity, will preclude sharing of di-Ub with Rpn10, probably due to steric occlusion. Apparently, a single UBA domain blocks Rpn10 access to both Ub moieties in Ub2. This then raises the question: if the chain were longer, would it be possible for the UBA and UIM to share the same chain?
Based on the preceding observation we hypothesize that a long enough chain will provide a sufficient number of unbound Ub units for co-binding of two or more receptors. To test this hypothesis, we first demonstrated the formation of a ternary complex Dsk2UBA/polyUb/Rpn10 by biochemical pulldowns. The addition of Ub4 at equimolar ratio to the UBA-containing region of Dsk2 (Dsk2ΔUBL) retains both proteins to the Rpn10-column, whereas Dsk2ΔUBL alone had no affinity for Rpn10 (Fig. 6A). We conclude that a polyUb chain can link the UBA and Rpn10 in a ternary complex. Similar results were obtained with Rpn10204-268 (Fig. 6B), pointing to the role of the UIM as the polyUb-receptor within the Rpn10 protein.
In order to obtain further information on the orientation and stoichiometry of this ternary complex, we took a more pinpointed approach. Given that the UBA and UIM separately displayed dramatically different affinities for Ub (Table 1), we initially designed several experiments in order to obtain detailed information on the order of binding to a single polyUb chain. Upon mixing 15N-Rpn10204-268 and Ub4 in a 1:0.9 molar ratio (Fig. S10), many of the Rpn10 NMR signals attenuated beyond detection, reflecting formation of the complex as we described earlier (Fig. 2). Although UBA exhibits significantly higher affinity for polyUb (Table 1 and Figs. 4, ,5),5), addition of UQ1-UBA at a 1:1 and then 2:1 molar ratio to Ub4 did not significantly affect the spectra of 15N-Rpn10204-268, indicating that Rpn10 remained in the Ub4-bound state (Fig S10). This could occur if the two receptors co-bind a single chain. Only when the UBA was present in a 4:1 molar ratio to Ub4 (i.e. one UBA per Ub) did most attenuated peaks in Rpn10 become fully visible again. The resulting Rpn10204-268 signals overlapped very well with those of free Rpn10204-268 (Fig. S10), suggesting that all Ub units in the mix were sequestered by the tighter binder of the two, UBA. This observation is in line with the ability of UBA to outcompete Rpn10 for Ub and Ub2 binding (Fig. 5C,D).
We then performed a reverse titration experiment, starting with a 1:1 mixture of 15N-labeled UQ1-UBA and Ub4 (i.e. four Ub units per UBA) (Fig. S11). The CSPs and signal attenuations indicated that UQ1-UBA was in the bound state (Zhang et al., 2008). Addition of Rpn10204-268 to this sample in a 2:1 molar ratio to Ub4 had no major impact on the UBA spectrum, indicating that UBA remained in the bound state. These reciprocal assays demonstrate the possibility that chains with unmasked Ub units are capable of being shared simultaneously by different receptors.
In order to directly demonstrate simultaneous binding of Rpn10 and UBA to a single chain, 15N-Rpn10204-268 and 15N-UQ1-UBA were mixed in an equimolar ratio, and unlabeled Ub4 was later added to this sample in small increments. Note that the NMR spectrum recorded prior to addition of Ub4 superimposes perfectly with the spectra of the two receptors recorded separately, indicating that there is no interaction between Rpn10204-268 and UQ1-UBA (Fig. S12). The spread in the NMR signals (Fig. S12) allowed us to dissect changes in either protein upon titration with Ub4 in the same sample (Fig. S13). The first to show attenuation were signals of UQ1-UBA residues already at the first steps of Ub4 titration, followed by attenuation in the Rpn10 signals only at higher concentrations of Ub4 (Figs. S13) reflecting the stronger affinity of UBA for (poly)Ub. At a molar ratio of 1:1:1 Ub4:UBA:Rpn10, strong signal attenuations were observed both in UQ1-UBA and Rpn10-UIM residues, indicating that at these conditions both receptors were bound to Ub4.
All these results clearly demonstrate that despite the higher affinity of UBA for Ub, and although UBA outcompetes Rpn10 for binding to shorter chains, Rpn10 and Dsk2 can co-bind to Ub4 or longer chains, provided there are enough Ub moieties in the chain unmasked by UBA. Together these results provide direct evidence for a single polyUb chain shared between multiple receptors.
Having dissected possible interactions between either of the two receptors with mono or polyUb, or with each other, we are now finally in a position to consolidate. From all our data presented above it is evident that Dsk2 and Rpn10 are mechanistically linked in two distinct ways: directly, via the UBL/UIM association, and indirectly, through a shared Ub chain. The various modes of mutual competition raise the question: “what is the resulting orientation of Dsk2/polyUb/Rpn10 in the ternary complex?”. As we described above, in a binary complex of Rpn10 with full-length Dsk2, the UBL is bound to the UIM (Fig. 1J-L) and therefore masked from competing interactions (see also (Matiuhin et al., 2008)). To this complex we added polyUb chains and tracked NMR signals that we could positively identify as emanating from 15N-labeled UBL domain (e.g. W14, A34, G48, Y60, and V71 in Fig. 6C,D). The addition of polyUb caused shifts in the UBL signals to their free state, indicating the unmasking of UBL due to the preference Rpn10-UIM has for polyUb over that for Dsk2 (Fig. 6C). At the same time, the UBA signals of Dsk2 strongly attenuated (e.g. XUBA disappeared in Fig. 6C), confirming that UBA also enters into a tight complex with polyUb. The resulting ternary complex in which polyUb is shared by UBA and Rpn10-UIM and the UBL is unmasked is shown schematically in Figs. 6E (bottom right) and S15A. By contrast, while also binding to the UBA domain, resulting in CSPs of UBA signals (XUBA in Fig. 6D), monomeric Ub caused only insignificant shifts in the UBL signals (Fig. 6D) even when added in eight-fold excess to the Dsk2/Rpn10 complex, in line with the stronger preference of Rpn10-UIM for UBL over monoUb (Table 1; Fig. 5). Thus we conclude that even in excess, monoUb is unable to pry apart the UBL from its complex with the UIM although it associates with Dsk2 by binding to its UBA domain. The resulting ternary complex in which Dsk2 bridges Rpn10 and monoUb is shown schematically in Figs. 6E (top right) and S15B.
To summarize, two very different orientations of ternary complexes form between Dsk2-Rpn10 and (poly)Ub depending on the length of the polyUb chain (Fig. 6E).
The comparison of Ub-binding affinities suggests (this study and (Raasi et al., 2005)) that Dsk2-UBA is one of the tightest Ub binders and that in fact it may be adapted to bind monoUb. This predicts that Dsk2 may somehow be involved in controlling the levels of monoubiquitinated proteins. Due to their high affinity for Ub even in its monomeric form, abnormally elevated levels of Dsk2 may trap proteins with single Ub modifications (monoubiquitinated proteins) interfering with the ubiquitin system from properly extending the chains or from properly targeting these conjugates to their cellular destinations. Indeed, overexpression of DSK2 uniquely alters the cellular Ub-landscape causing a sharp increase in absolute levels of Ub conjugates both in the form of multiple monoUb modifications or as polyUb chains (Matiuhin et al., 2008). The total level of multiple mono-ubiquitination increased three-folds in whole cell extracts upon induction of Dsk2. Of the remaining high MW conjugates, a threefold increase of Lysine-48 linkages over Lysine-63 linked chains was measured. The stabilization of ubiquitin-conjugates coupled with the slower turnover of ubiquitinated substrates and increased protein stability lead to severe growth phenotypes and cell death (Matiuhin et al., 2008). These deleterious effects associated with elevated Dsk2 explain the need to regulate its cellular levels and filter its accessibility to the proteasome. Overexpression of other polyUb-shuttles that were tested did not lead to similar effects. Moreover, only full-length Dsk2 displayed such cytotoxicity when overly abundant, whereas a version lacking the proteasome targeting signal, the UBL domain, was innocuous pointing to involvement of the functional protein able to target cargo to the proteasome (Matiuhin et al., 2008).
The results in the current study provide a mechanistic explanation to the previous biological observations. The tight affinity of Dsk2-UBA for monoUb and for short chains necessitates filtering mechanisms to limit premature cargo delivery. Through its ability to differentiate between Dsk2 and Ub, Rpn10 is able to mask the UBL domain of Dsk2 and thereby impose a threshold on the access of Dsk2 to downstream targets such as the proteasome. Once Dsk2 associates with longer chains, the resulting complex is able to bypass the filter through a fascinating mechanism of dynamic rearrangement that leads to exposure of the UBL domain. TetraUb chains are the shortest chain-length that can drive this rearrangement efficiently. Together, these studies begin to shed light on the complexity of the Dsk2-interaction network and how through competing interactions, full-length Dsk2 can help shape the Ub-signal. The small, yet significant, differences in their sequence and binding surfaces make it likely that other substrate-delivery proteins, polyUb-shuttles, and assorted ubiquitin-domain proteins are also subject to dedicated filters and selection processes.
One of the intriguing aspects of the Ub-proteasome system (UPS) is that substrates are targeted to the proteasome by means of polymeric chains. What is the benefit of polymeric modifications and why is post-translational modification by a mere Ub unit insufficient? One of the obvious answers is that chain length could enhance the efficiency of targeting, therefore diversity in chain length and linkage type may provide for hierarchy in targeting. It comes, however, as somewhat of a puzzle, that some of the Ub receptors are incredibly efficient in recognizing already a single Ub moiety. For instance, the UBA domain of Dsk2 (see above) or Pru of Rpn13 (Husnjak et al., 2008) bind monomeric Ub tighter than some other receptors bind polyUb. Moreover, some proteins (e.g. the UBL-UBA shuttle family) are targeted to the proteasome via a single Ub-like domain. Clearly, Ub (and its kin) can serve as efficient signals even as monomers. In fact, monoubiquitination may even serve as a proteasome targeting signal under some conditions (Guterman and Glickman, 2004; Hershko and Heller, 1985; Kravtsova-Ivantsiv et al., 2009). Another benefit of a polymeric modification is that it could provide a flexible signal. Various layers of shuttles/receptors together with chain extenders (E4s) and chain trimmers (DUBs), by either enhancing the signal or interfering with it, give the UPS a means for quality control upstream to the irreversible step of degradation by the proteasome.
On the receiving end, a plethora of Ub-binding proteins displaying an array of affinities for various chain-length and linkage types await these chains. These Ub-binding proteins are thought to shuttle substrates to the proteasome, raising yet another question: why are Ub-conjugates funneled through another layer of selection when they have an intrinsic capacity to bind to the proteasome?
The current study reveals how Rpn10 and Dsk2 function together as a Ub chain-length sensor which provides a possible mechanism for selection of linkages above a certain threshold simultaneous with filtering uncharged shuttles (Fig. 6E). Through competing interactions and by sharing a Ub-chain, the resulting ternary complex of Dsk2/polyUb/Rpn10 is dynamic and can be rearranged depending on the length of the chain. The UBA domain of Dsk2 outcompetes the Rpn10-UIM for binding to mono- or to di-Ub, therefore at a 1:1 ratio only chains of three units or longer have “overhanging” Ub units capable of binding to Rpn10. However, since the UBL domain of Dsk2 binds to the same UIM of Rpn10 in this complex stronger than monoUb does and comparable to Ub2, even a tri-Ub chain should not provide an “overhang” sufficient to compete with the UBL for Rpn10-binding. The result is that in the presence of Dsk2, tetra-Ub is the shortest chain with sufficient Ub capacity to extend beyond the grip of the UBA and bind Rpn10 (Fig. 6E). This observation nicely complements the early findings that polyUb chains of n ≥ 4 are efficient signals for proteasomal targeting (Thrower et al., 2000).
For multiple ligands to play musical chairs with their receptors, the same surface on a given receptor must be capable of distinguishing from among a broad range of potential ligands. As a case in point, a single amino acid residue can account for a 5-fold difference in binding affinity of two similar ligands: Ub and Dsk2-UBL (Fig. 4, Table 1). Thus the D64K mutation essentially converts the affinity of Dsk2-UBL for Rpn10 into that of Ub for the same receptor. Contribution of a hydrophilic interaction to binding of a member of the Ub-family (Dsk2-UBL in this study) alters our current understanding of Ub recognition. Until now, most studies identified the canonical hydrophobic patch and matching hydrophobic residues in the receptor as critical for recognition of Ub and most UBL domains by the major classes of receptors (whether UBA, UIM, Cue, and others (Hurley et al., 2006)) (Figs. 2–4). The protocols researchers use to affinity purify, trap, isolate, or wash Ub-family members will have to be carefully assessed if they want to study the unique properties of these signals that differ in their binding properties. Interestingly, in the strength of its interaction with the UIM of Rpn10, Ub falls between the UBLs of Dsk2 and Rad23. Other Ub-binding proteins may vary in their binding preferences. It may be concluded, that without accurate data on the comparative properties and binding affinities for various Ub-family members, merely stating that a protein has a Ub-binding domain might be insufficient to explain its overall role in the context of the UPS.
While shuttles of Ub-conjugates support degradation, they can also, paradoxically, oppose UPS activity ((Chen and Madura, 2002; Chen et al., 2001; Funakoshi et al., 2002; Hartmann-Petersen et al., 2003; Kleijnen et al., 2000; Matiuhin et al., 2008; Raasi and Pickart, 2003)). High concentrations of Ub-binders “decorating” a polyUb “tree” could make the Ub signal inaccessible, thus their biological abundance must be under very strict control (as in the case of ubistatins (Verma et al., 2004b), or Dsk2 (Matiuhin et al., 2008)). In vitro, Rad23 displays a concentration-dependent inhibitory effect on polyUb-chain formation (Ortolan et al., 2000). Likewise, both Rpn10 and Rad23 effectively protect model substrates from proteolysis ((Deveraux et al., 1995; Raasi and Pickart, 2003; Verma et al., 2004a)). In vivo, overexpression of RAD23 can inhibit the degradation of model substrates in yeast (Ortolan et al., 2000), while hPLIC can prevent degradation of physiological substrates including p53 and IκB in mammalian cells (Kleijnen et al., 2000). PolyUb-conjugates accumulate upon overexpression of DSK2 and half-life of short-live proteins is extended eventually leading to cell death (Matiuhin et al., 2008). These references together with the data presented in this manuscript uphold a model whereby Ub-binding proteins may function in more ways than merely cargo-shuttles.
Extrapolating beyond the interesting relationship we uncovered between Rpn10 and Dsk2, we expect that other Ub-binding partners will participate in shaping the Ub signal by serving as filters and/or enhancers. An additional example whereby a Ub chain may relay between multiple receptors is the case of hHR23A/S5a (Kang et al., 2007), where hHR23A recognizes and binds polyUb via the UBA-2 domain and docks via its UBL domain on UIM-2 of S5a. Subsequently, the UIM-1 of S5a recognizes and binds polyUb to complete the transition. Undoubtedly these examples only scrape the tip of the iceberg. A broad interaction sphere “decorating”, masking, or chaperoning Ub-conjugates adds to the heterogeneity of the signal in terms of chain length and linkage type. The combinatorics of Ub-binding proteins and other factors (DUBs, E2s, E3s, etc) contributes to the richness of the Ub-signal landscape.
It is by extending the complexity of previous studies that we show how two receptors working in tandem (Rpn10 and Dsk2) are able to discriminate between Ub chains below and above the threshold of four. Neither of them is able to do so alone. These findings alter our conception of the ubiquitin signal: (1) Ub might not be the ultimate signal but apparently serves in coordination with an assortment of receptors and shuttles that “decorate” the chain; and (2) depending on the length of the chain, this assemblage reorients presenting alternative signals to the downstream components, such as the proteasome.
Full-length Rpn10 and Dsk2 constructs from Saccharomyces Cerevisiae were expressed and purified as described previously (Matiuhin et al., 2008). The Rpn10 C-terminal construct (referred to as Rpn10204-268) used in this study contained residues G204-K268 from the Rpn10 sequence. Residue Q261 was mutated to Tyr for quantification purposes. UBL domains from four yeast proteins were used in this study: Dsk2-UBL, Rad23-UBL, Ubp6-UBL, and Ddi1-UBL. Details on cloning, expression and purification of the proteins and biochemical characterization of binding are in Supplemental Data.
Samples for NMR studies were prepared in 20 mM phosphate buffer, pH 6.8, containing 7% D2O and 0.02% (w/v) NaN3. All NMR data were acquired at 22–23 °C on a Bruker Avance 600 MHz spectrometer equipped with a cryoprobe. The experiments and data analysis are detailed in Supplemental Data.
Surface Plasmon Resonance (SPR) measurements were performed using ProteOn XPR36 instruments developed by Bio-Rad Haifa (Haifa, Israel) (see Supplemental Data).
Supported by NIH grant GM065334 to D.F. and by ISF and BSF grants to M.H.G. The sabbatical of D.F. at the Technion, was supported in part by a fellowship from the Lady Davis Foundation. We thank Dr. Ananya Majumdar (Johns Hopkins University) for help with setting up triple-resonance NMR experiments and Noa Reis for help with cloning and design of constructs.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.