Molecular architecture of the 26S proteasome holocomplex determined by an integrative approach

Abstract

The 26S proteasome is at the executive end of the ubiquitin-proteasome pathway for the controlled degradation of intracellular proteins. While the structure of its 20S core particle (CP) has been determined by X-ray crystallography, the structure of the 19S regulatory particle (RP), which recruits substrates, unfolds them, and translocates them to the CP for degradation, has remained elusive. Here, we describe the molecular architecture of the 26S holocomplex determined by an integrative approach based on data from cryoelectron microscopy, X-ray crystallography, residue-specific chemical cross-linking, and several proteomics techniques. The “lid” of the RP (consisting of Rpn3/5/6/7/8/9/11/12) is organized in a modular fashion. Rpn3/5/6/7/9/12 form a horseshoe-shaped heterohexamer, which connects to the CP and roofs the AAA-ATPase module, positioning the Rpn8/Rpn11 heterodimer close to its mouth. Rpn2 is rigid, supporting the lid, while Rpn1 is conformationally variable, positioned at the periphery of the ATPase ring. The ubiquitin receptors Rpn10 and Rpn13 are located in the distal part of the RP, indicating that they were recruited to the complex late in its evolution. The modular structure of the 26S proteasome provides insights into the sequence of events prior to the degradation of ubiquitylated substrates.

In eukaryotes, the ubiquitin-proteasome pathway (UPP) is essential for proteostasis: Misfolded proteins or otherwise defective proteins as well as short-lived regulatory proteins are eliminated by degradation (1). The UPP regulates many fundamental cellular processes, such as protein quality control, DNA repair, and signal transduction (for review see ref. 2). The 26S proteasome is the most downstream element of the UPP, executing the degradation of polyubiquitylated substrates (3–5). It consists of the barrel-shaped core particle (CP; approximately 700 kDa), which sequesters the proteolytically active site in its central cavity, and the regulatory particle (RP; approximately 900 kDa), which is attached at either one or both ends of the CP and prepares substrates for degradation (6).

The RP consists of 19 different canonical subunits, including six regulatory particle AAA-ATPase subunits (Rpt1-6) and 13 regulatory particle non-ATPase subunits (Rpn1-3, Rpn5-13, and Rpn15). The integral ubiquitin (Ub) receptors Rpn10 and Rpn13 recognize polyubiquitylated substrates (7–9). Alternatively, polyubiquitylated substrates can be recruited by shuttling Ub-receptors (Dsk2, Rad23, Ddi2), which bind to substrates with their Ub-associated domain, and to Rpn1, Rpn10, or Rpn13 at their Ub-like domain (5). The metalloprotease Rpn11 deubiquitylates substrates prior to their degradation (10, 11). The functions of the other Rpn subunits remain to be established. The AAA-ATPases form a hexameric ring that unfolds substrates, opens the gate to the CP, and eventually translocates the substrates to the CP.

Electron microscopy (EM) (12) and X-ray crystallography studies of the CP revealed an evolutionary highly conserved structure, namely a stack of four seven-membered rings (α1-7; β1-7; β1-7; α1-7) (13, 14). The chamber formed by the β-subunits houses the proteolytic sites and the outer α-rings constitute two antechambers (15) serving as holding compartments (16). Most of our structural knowledge on the AAA-ATPases so far stems from the archaeal homolog of the RP, the proteasome-activating nucleotidase (PAN) (17, 18). The six AAA-ATPase subunits assemble into two stacked rings: The ring formed by the AAA-domains (AAA-ring) is pseudo sixfold symmetrical and positioned proximal to the CP, whereas the distal N-terminal domains form a pseudo threefold symmetric ring (trimer of dimers) with coiled-coil pairs protruding from it (N ring). The N ring unfolds substrates whereas the AAA ring is believed to actively translocate substrates into the CP in an ATP-dependent manner (17, 19). The topological order of the AAA-ATPase subunits is Rpt1/Rpt2/Rpt6/Rpt3/Rpt4/Rpt5, as determined based on protein–protein interactions (20) and engineered disulfide cross-links (21). Integration of a cryoelectron microscopy (cryo-EM) map at 9.1-Å resolution, protein–protein interactions, and site-specific intermolecular cross-links revealed the relative positions of the AAA-ATPase ring and the CP (20, 22), which was subsequently corroborated by engineered disulfide cross-links between the AAA-ATPase C termini and residues in the pockets between the α-subunits of the CP (23). In contrast, the spatial configuration of the non-ATPase subunits and their role in the preparation of substrates for degradation remains largely unknown.

The RP has been refractory to any single structure determination method, presumably due to sample heterogeneity caused by the RP’s tendency to disassociate into heterogeneous subcomplexes during purification and concentration as well as the presence of proteasome interacting proteins (PIPs). Moreover, some RP subunits may also exhibit a significant degree of structural variability. As a consequence, X-ray crystallographic analysis of the entire RP has not yet been accomplished, although the atomic structures of a few subunit fragments have been determined (9, 24–26). To overcome limitations in sample preparation and individual structure-determination techniques, we have applied an integrative structure determination approach in which a variety of different datasets were computationally integrated into a more accurate and higher-resolution structural model than is possible with any of the individual datasets (27–29). In particular, we have determined the structure of the complete 26S proteasome from Schizosaccharomyces pombe, based on a new 8.4-Å cryo-EM map, residue-specific intersubunit cross-links, crystallographic structures of RP subunit homologs, as well as previously published protein–protein interaction data from a variety of species. The modular architecture of the RP provides unique insights into the sequence of events prior to degradation.

Results and Discussion

8.4-Å Resolution Cryo-EM Map of the 26S Proteasome.

We have used cryo-EM to structurally characterize the 26S proteasome holocomplex from Schizosaccharomyces pombe (Fig. 1A). Notably, we have approximately doubled the number of particles previously used for a single-particle reconstruction (22), to 375,000. This increase allowed us to improve the resolution from 9.1 to 8.4 Å (Fig. S1). The resolution of the cryo-EM map varies locally (Fig. 1B); the CP is better resolved than the RP, which exhibits “hot spots” of structural variability. From the cryo-EM map we segmented the CP and the AAA-ATPase modules based on the CP and PAN crystal structures (20, 22) (Fig. 1A). The remaining cryo-EM density corresponds to the Rpn subunits. Numerous helical elements, many of which resemble bihelical repeats, are clearly visible in the 8.4-Å resolution map.

Fig. 1.

Cryo-EM map of the S. pombe 26S proteasome. (A) The single-particle cryo-EM density map of the 26S proteasome from S. pombe at 8.4-Å resolution is shown in two views, related by a 90° rotation around the pseudo-sevenfold axis of the CP (CP: red; AAA-ATPase hexamer: blue; Rpn subunits: gold). (B) The isosurface of the cryo-EM map is colored according to the local resolution in Å, as specified in the color bar.

Subunit Interactions Revealed by Chemical Cross-Linking/Mass Spectrometry (XL/MS).

To obtain information on protein–protein proximities at the peptide level, we chemically cross-linked the purified S. pombe 26S proteasomes using the lysine reactive cross-linker disuccinimidyl suberate (DSS), subjected the trypsin-digested samples to liquid chromatography-tandem mass spectrometry (LC-MS/MS), and identified the cross-linked peptides from MS/MS spectra at a false-discovery rate (FDR) of less than 5% (Materials and Methods) (30). Thirty-two cross-links were obtained within the AAA-ATPase hexamer and four between the CP and the AAA-ATPases, involving nine and three different subunit pairs, respectively (Fig. 2 and Table S1); all of these cross-links are consistent with our previously determined subunit organization (20). An additional 15 intra-Rpn and eight Rpt/Rpn cross-links involved seven out of the 13 Rpn subunits and 15 different subunit pairs (Fig. 2 and Table S1). Notably, these cross-links indicate proximity of the lysine residues in the N-terminal segments of the non-ATPases Rpn5 and Rpn6 to the AAA domains of Rpt3 and Rpt4.

Fig. 2.

Chemical cross-links for the S. pombe and S. cerevisiae 26S proteasomes. Fifty-five (21) pairs of cross-linked lysines from the S. pombe (S. cerevisiae) 26S proteasome subunits are shown as an interaction network (Table S1). Multiple edges between a pair of subunits indicate multiple cross-linked lysine pairs.

While we used the data from S. pombe 26S proteasome for integrative structure determination (below), we also performed cross-linking experiments with the 26S proteasome from Saccharomyces cerevisiae for validation. Not surprisingly, four of the 11 S. cerevisiae cross-links apply to the same Rpn/Rpn and Rpn/Rpt pairs as in the S. pombe dataset, indicating their accuracy. Nevertheless, some cross-links observed in the S. pombe dataset were not observed in S. cerevisiae and vice versa. Thus, complementary information can be obtained by cross-linking protein complexes of related organisms. The differences in cross-link coverage are attributed to the differences in the lysine residue locations between the two species (only two of the 11 Rpn/Rpt and Rpn/Rpn cross-links occurred at corresponding lysine pairs) and to technical limitations of XL/MS.

Integrative Structure Determination of the RP Architecture.

To determine the positions, orientations, and conformations of the Rpn subunits and the AAA-ATPase hexamer, we applied an integrative approach implemented in our open source Integrative Modeling Platform (IMP) package (http://salilab.org/imp/) (27, 28, 31). The process consists of four stages: gathering information about the structure of the 26S proteasome, choosing how to represent the system and how to translate the information into spatial restraints, calculating an ensemble of structures that satisfy these restraints, and analyzing the ensemble (Fig. 3). The approach has been previously used for determining the positions of 456 subunits in the eightfold symmetric ring of the yeast nuclear pore complex (31, 32). However, the 26S proteasome has presented new challenges—namely, the asymmetry of the assembly and the interpretation of the structure at the amino acid level. These challenges required integration of new types of data, a multiscale model representation including the atomic level, as well as a more powerful structural sampling scheme.

Fig. 3.

Integrative structure determination of the RP. First, structural data and information are generated by various experiments and computational methods, respectively. Second, the AAA-ATPase hexamer and the Rpn subunits are each represented as a string of beads or by atoms, and the data are translated into spatial restraints on their arrangement. Third, an ensemble of structures that satisfy the data are obtained by inferential sampling. Fourth, the ensemble is clustered into distinct subsets of structures, and analyzed in terms of geometry and accuracy.

Stage 1: Gathering Information.

Five different types of information were used for structure determination: (i) an 8.4-Å cryo-EM map of the RP from S. pombe; (ii) 12 residue-specific cross-links between the AAA-ATPase and Rpn subunits as well as three cross-links between the Rpn subunits from S. pombe (Table S1); (iii) densities of Rpn10 and Rpn13 in the 26S holocomplex as revealed by cryo-EM of the S. cerevisiae 26S proteasomes from ΔRpn10 and ΔRpn13 deletion strains (33); (iv) 47 direct interactions and proximities between the RP subunits determined by various proteomics studies, including two-hybrid assays, chemical cross-linking, and immunoprecipitations (Table S2); as well as (v) atomic structures of the individual Rpn subunits and their homologs (Fig. S2 and Table S3).

Stage 2: System Representation and Translation of Information into Spatial Restraints.

Two representations of a sequence segment, subunit, or a subcomplex were used, depending on the data available, spatial restraints applied, and sampling performed (Table S4). The first, coarse-grained representation depicted each approximately 50-residue fragment of the segment by a bead. If the atomic structure of a subunit or subcomplex (i.e., the AAA-ATPase hexamer) was predicted by comparative modeling, the beads were fixed to mimic the shape of the atomic model; otherwise, the string of beads was allowed to flex. The second, atomic representation specified atomic coordinates for segments with available comparative models.

The spatial restraints encoded the different types of information gathered (Table S5). First, a measure of cross-correlation between the cryo-EM map and the atoms in an RP model restrained the shape of the model. Second, upper distance bounds enforced proximity between pairs of cross-linked atoms or subunits, depending on the representation used. Third, the “connectivity” restraints (31) enforced direct protein interactions and proximities among the subunit sets in the RP subcomplexes (Table S2). In addition, we applied a radius-of-gyration restraint to each flexible subunit in the coarse-grained representation (31), an excluded-volume restraint to each bead and/or atom pair, and a geometric-complementarity restraint for each pair of interacting subunits in the atomic representation.

Stage 3: Ensemble Sampling.

Models of the RP that satisfy the data as well as possible were obtained in three steps. In the first, subunit localization step, best-scoring models were enumerated in the coarse-grained representation by aligning the RP interaction network implied by the residue-specific cross-links (Table S1) and subunit interactions (Table S2) into the cryo-EM density map. The sampling started by discretizing the density map into a graph of 238 nodes, with graph nodes corresponding to the density map regions and graph edges connecting the neighboring regions. Possible localizations for each subunit were then found efficiently by applying a “message passing” algorithm to the graph (34). The sampling explored only localizations that were consistent with the subunit shapes and localization data. In particular, the positions of the Rpn10 and Rpn13 beads in the RP density were constrained by the corresponding difference cryo-EM maps. Similarly, the positions of the AAA-ATPases hexamer (20) and Rpn6 (26) were fixed based on previous fitting of their atomic models into a cryo-EM map. The number of candidate localizations for each nonlocalized subunit ranged from 10³ for the small Rpn subunits to 10⁵ for the large Rpn1 and Rpn2 subunits (Table S4). Approximately 5·10⁵ best-scoring coarse-grained RP models that significantly violated at most five restraints were finally found by enumerating combinations of subunit localizations, again using message passing.

In the second, subunit fitting step, multiple comparative models were computed for each of the S. pombe Rpn subunits (Materials and Methods and Fig. S2 and Table S3). Alternative combinations of comparative models of the PCI subunits were then simultaneously fitted into the region of the cryo-EM density map identified uniquely in the previous subunit localization step using MultiFit (34) (Materials and Methods and Fig. S3). Finally, additional high-confidence models of Rpn8, Rpn10, and Rpn2 were fitted to the map (Fig. S4) using MOLMATCH (35).

In the third, subunit refinement step, the atomic coordinates of the PCI subunits were refined by flexible fitting using molecular dynamics flexible fitting (MDFF) (36).

Stage 4: Assessment of Data and Structure.

We analyzed the ensemble of approximately 5·10⁵ best-scoring RP coarse-grained models in terms of subunit positions, contacts, and configuration. The models were clustered according to the subunit centroid rmsd, revealing three large clusters of models with an average centroid rmsd within each cluster of only 5 Å (Fig. 4 A and B). The largest variability in the ensemble corresponds to a swap between Rpn8 and Rpn11 as well as between Rpn7 and Rpn3 (Figs. 4 C and D). Random subsets of 10% of the best-scoring models share a similar clustering pattern, confirming that our sampling procedure has indeed enumerated the best-scoring models, given the data and the sampling resolution.

Fig. 4.

Analysis of the ensemble of 26S proteasome structures. (A) Distribution of scores for the ensemble of models. 5·10⁵ of the nearly 250 million sampled models violated at most five of the 98 cross-linking and proteomics restraints. (B) Clustering of the top scoring models that violate at most five cross-linking and proteomics restraints. The models were clustered by their centroid rmsd into three main clusters with the maximum rmsd from the cluster centroid of 5 Å. (C) Conservation of subunit positions in the ensemble. For each subunit, the average cross-correlation coefficient between pairs of subunit clusters is displayed. The plot shows localization variability for Rpn3, Rpn7, Rpn8, and Rpn11. (D) Localization probabilities for each subunit in the cluster displayed as a heat map. Each voxel of the density map is colored by its localization probability, defined as the percentage of cluster models that localized a subunit in the voxel. A gradient color from blue to red indicates that the subunit was localized to a voxel in 5% to 100% of the models; a gray color indicates 0%.

The confidence in the accuracy of the ensemble is based on three considerations (27, 31). First, the ensemble models satisfy most of the independently determined data points. All good scoring models violate the same small subset of restraints, including the residue-specific cross-links Rpn10:30/Rpt5:54 and Rpn6:272/Rpt4:164, the Rpn10-Rpn1 interaction (37, 38), and the Rpt1/2,Rpn2/8/12/13 subcomplex (39). The first three violations could be explained by conformational variability (see below) and the fourth one by an error or incompleteness of high-throughput detection. Second, structural similarity among the models in the ensemble, as described above, is sufficient to determine the positions of most subunits and orientations for subunits with atomic models. Third, the ensemble is consistent with the structural data not used to compute it, including 21 residue-specific intersubunit cross-links from S. cerevisiae, the PCI domain contacts implied by the Rpn6 crystal contacts (26), the high-confidence fits of the atomic models of the leucine rich repeat (LRR) region of Rpn2 (40) (Fig. S4), the peripheral position of Rpn12 from the whole complex mass spectrometry (41), the Rpn9 interaction with the Ub receptor Rpn10 from a two-hybrid assay (42), and the subcomplexes formed during lid assembly (43). Moreover, the ensemble is consistent with > 95% of the XL/MS data at the residue level, although cross-links were imposed only at the subunit-level (Fig. S5). Finally, even after omitting one restraint at a time, the resulting jackknife ensembles remain similar to the original ensemble (Fig. S6).

AAA-ATPase Hexamer.

To analyze the AAA-ATPase hexamer in more detail, its comparative model based on the PAN structures (22) was subjected to flexible fitting into the EM map (36). The refined model reveals substantial deviations from the approximate sixfold rotational symmetry of the PAN AAA ring (Fig. 5A). In particular, the Rpt3, Rpt6, and Rpt4 subunits vary from those in the symmetrical starting model. Moreover, the principal axes of the N ring and the AAA ring are not in register. Thus, the evolutionary diversification of the Rpt subunit functions (44) has also resulted in a major divergence of their structures. The cryo-EM map allowed us to resolve the C termini of the Rpt2 and Rpt3 subunits, which are absent in the PAN crystal structures (Fig. 5B). These two C termini are located in the pockets between the α3/α4 and α1/α2 CP subunits, respectively. Rpt2 and Rpt3 both contain the C-terminal HbYX motif that has been shown to induce gate opening of the CP (45). We also observed density in the pocket formed by the α5/α6 pocket, which likely corresponds to the C terminus of Rpt5, the third Rpt subunit containing HbYX motif. However, this density is weaker than that observed in the α3/α4 and α1/α2 pockets. A recent cross-linking study, in which cysteines were introduced into the AAA-ATPase C termini, was consistent with our observations (23). Specifically, Rpt2 and Rpt3 were found to localize in the pockets formed by the CP α3/α4 and α1/α2 interfaces, respectively, whereas Rpt5 was not confined to a specific pocket and was found to occupy both α5/α6 and α6/α7 pockets.

Fig. 5.

Structure of the AAA-ATPase hexamer. (A) Atomic representation of the CP and AAA-ATPase hexamer after flexible fitting into the cryo-EM map, colored by their C^α atom rmsd compared to the initial comparative models based on the S. cerevisiae CP (13) and the AAA/N-rings from PAN (18). (B) Near the C termini of Rpt2 (left) and Rpt3 (right), we observe density in the pockets formed by the α3/α4 and α1/α2 interfaces, respectively (cyan).

Modular Architecture of Non-ATPases.

We determined the configuration of all Rpn subunits relative to the AAA-ATPase and CP. Based on sequence similarity, the Rpns can be divided into four categories (46): (i) the proteasome cyclosome (PC) repeat containing subunits Rpn1 and Rpn2, (ii) the MPN (Mpr1 and Pad1 in the N terminus) domain containing subunits Rpn8 and Rpn11, (iii) the proteasome-COP9-eIf3 (PCI) domain containing subunits Rpn3/5/6/7/9/12, and (iv) the Ub receptors Rpn10 and Rpn13. These evolutionary relationships are reflected in the modular structure of the RP.

PC-Repeat-Containing Subunits.

The hallmarks of the Rpn1 and Rpn2 sequences are nine-repeat sequences (PC repeats), which also occur in the cyclosome/anaphase-promoting complex subunits (47). The PC repeats are predicted to resemble bihelical LRRs, which assemble into a torus (40, 48). In our structure, we find Rpn1 at the periphery of the Rpt1/2 dimer (Fig. 6A). The density corresponding to Rpn1 is highly variable (Fig. 1B), as already observed (49), which makes it impossible to discern secondary structure elements. This variability may be due to Rpn1’s function as a recruitment factor for shuttling Ub receptors and the DUB Ubp6, which dock to the LRRs of Rpn1 (50, 51).

Fig. 6.

Modular architecture of the 26S proteasome. (A) From left to right: segmented densities of the PC-repeat containing proteins, Ub receptors (Rpn13 is taken from the S. cerevisiae 26S proteasome because it binds substoichmetrically in S. pombe) (33), the MPN-domain-containing proteins, and the PCI-domain-containing subunits. (B) The segmented 26S proteasome density in three different views. (C) Hybrid representation of the 26S proteasome, indicating fitted available comparative models for the CP (red), the AAA-ATPase heterohexamer (blue), the PCI heterohexamer (green), Rpn8 MPN domain (pink), Rpn2 LRR domain (yellow), and Rpn10 VWA domain (purple). (D) Close-up view of the heterohexamer formed by the PCI domains.

In contrast to Rpn1, Rpn2 is structurally well defined. With its domain proximal to the CP, it contacts the Rpt3/6 dimer and projects to the distal part of the RP. Short helices are discernible at many positions in the Rpn2 density, suggesting that Rpn2 consists largely of helical repeats. Interestingly, a segment of Rpn2 has a toroidal shape, as predicted for the PC-repeat region (40, 48). Indeed, the proposed atomic model (40) matches the cryo-EM density map very well in this region (Fig. S4). Rpn2 contacts the Rpn8/11 dimer at its torus-shaped region and it interacts with Rpn12 and Rpn3 at its distal end. The Ub receptor Rpn13 also binds to the distal, probably C-terminal domain of Rpn2 (9).

MPN-Domain-Containing Subunits.

Rpn8 and Rpn11 both contain an N-terminal MPN domain. The MPN domain of Rpn11, in contrast to that of its paralog Rpn8, contains a zinc-binding JAMM (JAB1/MPN/Mov34 metalloenzyme) motif, which is responsible for its deubiquitylation activity (24, 52, 53). Rpn8 and Rpn11 form a heterodimer, which is positioned above the opening of the AAA-ATPase ring and incorporated into the scaffold provided by the PCI-domain-containing subunits (Fig. 6). While the heterodimer was localized with high confidence, the individual Rpn8 and Rpn11 positions remain ambiguous (only the MPN domain of Rpn8 can be positioned accurately) (Fig. S4).

PCI-Domain-Containing Subunits.

We recently solved a protein structure containing a proteasomal PCI-domain (26): Rpn6 consists of several bihelical repeats that assemble into a solenoid fold followed by a PCI domain (Fig. S4). Structure prediction suggests that all PCI-domain-containing subunits have the same fold (Fig. S2). Protein–protein interaction data show that the PCI-containing subunits colocalize in the same region of the RP (46). Some of these interactions have been narrowed down to their PCI domains (26, 54). Our structure reveals that the six PCI subunits are arranged in a horseshoe-like fashion forming a “roof” covering a large part of the AAA-ATPase heterohexamer (involving Rpt3, Rpt6, and Rpt4) (Fig. 6). The subunit order in the horseshoe is Rpn9/Rpn5/Rpn6/Rpn7/Rpn3/Rpn12. The PCI-heterohexamer embraces the Rpn8/11 heterodimer and contacts Rpn10 (via Rpn9) and Rpn2 (via Rpn12) at its ends. The scaffold of the horseshoe is made of the interacting PCI domains, while the α-solenoid domains project away from the horseshoe (Fig. 6D). Interestingly, Rpn6 monomers contact each other through their PCI domains in the Rpn6 crystal and form a screw (26), which resembles the horseshoe observed for the PCI-heterohexamer in the RP. The solenoid N termini of the PCI-domain-containing subunits all form right-handed superhelices, but their respective curvatures vary substantially.

Ub Receptors.

Rpn10 and Rpn13 are structurally unrelated and recognize Ub through different binding motifs (8). From their peripheral localization at the distal end of the 26S proteasome, we conclude that both receptors were added late in the evolution of the RP. Rpn13 interacts solely with Rpn2, which is consistent with numerous interaction studies (9, 55–57). We anticipated that the cryo-EM map could only resolve the C-terminal von Willenbrand factor A (VWA) domain of Rpn10, due to the structural variability of the N-terminal ubiquitin interacting motif (UIM) (58), which is consistent with the fit for Rpn10 (Fig. S4D). Rpn10 interacts with Rpn9 and the Rpn8/11 heterodimer. We do not observe the Rpn10–Rpn1 interaction, which was implied by immunoprecipitation (38, 59); nevertheless, this interaction may occur transiently. The Rpn10 UIM is flexible, and in the cryo-EM map we observe Rpn1 to be variable. Hence, Rpn1 and the Rpn10 UIM may bridge the distance of approximately 70 Å between Rpn1 and the Rpn10 VWA domain temporarily, which is not resolved in our map.

Structure and Function of the Lid Subcomplex.

The subcomplex formed by the MPN- and PCI-domain-containing subunits was termed the “lid” (60), indicating that it was positioned distally from the CP. However, it turns out that the lid instead flanks the side of the AAA-ATPase hexamer and even contacts the CP directly; the N termini of Rpn6 and to a lesser extent Rpn5 physically interact with the CP. The only currently known function of the lid is a deubiquitylating activity. Rpn11 is only active within the assembled RP and its deubiquitylating activity is ATP-dependent (10, 11). We find the Rpn8/Rpn11 heterodimer close to the AAA-ATPase hexamer, which may explain the ATP-dependence of the Rpn11 function. The positioning of Rpn8/Rpn11 suggests that Rpt3 and Rpt6 may be critical for activation of Rpn11. Nevertheless, higher resolution studies are required to precisely identify the mechanism by which the AAA-ATPases activate Rpn11 and to elucidate the role of Rpn8.

The AAA-ATPase heterohexamer is thought to undergo large-scale motions during its functional cycle (61), probably similar to the ClpX hexamer (62). One function of the hexamer formed by the PCI-containing subunits seems to be that of a scaffold, which positions Rpn11 near the mouth of the AAA-ATPase. Pivotal for this scaffolding function is Rpn6, which maintains the contact to the CP through its solenoid N terminus. A further role of the PCI submodule at the top of the AAA-ATPase module could be that of a shield: The AAA-ATPases are expected to have a propensity to bind to secondary degradation signals (degrons) of substrates (see below) and their shielding could help to minimize uncontrolled degradation.

It is unlikely that a complex structure such as the PCI horseshoe has evolved to solely function as a scaffold or a shield. For example, the 71-residue subunit Rpn15 (Sem1), which is associated with Rpn3 and Rpn7, would not be required if the PCI subunits did not have further functions (63). The PCI domain contains a winged-helix motif, which typically binds nucleotides (64). Interestingly, when the winged-helix motif of Rpn6 is superposed on that of the transcription factor E2F-DP (PDB ID code 1CF7), the bound DNA molecule would not clash with the surrounding proteasome subunits (Fig. S7). DNA association with the 26S proteasome could be required for its role in transcription (65), but no experimental evidence for recruitment of the 26S proteasomes to DNA or RNA exists to date. Interestingly, for subunit eIF3c of the lid paralog eIF3, RNA interaction of its PCI domain has been shown recently (66).

Model for the Initial Protein Degradation Steps.

Based on the molecular architecture of the RP and biochemical findings, we put forward a model for the early steps of protein degradation by the 26S proteasome (Fig. 7). Genetic studies revealed a functional linkage between Rpn10 and Rpn13 (8). In NMR spectroscopy experiments, purified Rpn13 and Rpn10 were found to simultaneously bind to K-48 linked di-ubiquitin (67). Together, these findings suggest that Ub receptors may bind polyubiquitin tags “in concert” when recruiting substrates to the 26S proteasome (33). This hypothesis is consistent with the positions of Rpn10 and Rpn13 found in our structure. The distance between both subunits is approximately 90 Å, which could be bridged by a tetraubiquitin moiety (33). In fact, this relative placement of Ub receptors offers an explanation why polyubiquitin chains need to comprise at least four Ub subunits to function efficiently as a degradation signal (68, 69). Thus, we hypothesize that Rpn13 and Rpn10 both bind to the polyubiquitin tag of their respective substrates with a preference of Rpn13 for the proximal Ub moiety (67) (Fig. 7I).

Fig. 7.

Mechanistic model for the early steps of protein degradation by the 26S proteasome. Polyubiquitylated substrates first bind to the Ub receptors Rpn10 and Rpn13 in concert (33) (I). During the subsequent commitment step (II), the substrates are more tightly bound by the AAA-ATPase coiled coils, which may perform swinging motion. The coiled coils further transfer the substrates to Rpn11, where they are deubiquitylated to enable Ub recycling (III).

Initial binding of polyubiquitylated substrates is followed by a “commitment” step (70); the substrate is bound more tightly to the proteasome via a loosely folded domain in an ATP-dependent manner. Rpn10 and Rpn13 are positioned above the coiled coils of the Rpt4/5 and Rpt1/2 dimers, respectively. The structure of the archaeal homolog PAN suggests that the N-terminal coiled coils are flexibly linked to the N ring of the ATPase (17). We envisage that they act as “swinging arms” sampling the space above the ATPase. Such swinging arms have been implicated in the transfer of substrates in multienzyme complexes to couple multistep reactions (71). Indeed, analysis of the space accessible to the coiled-coils pairs shows that the Rpt1/2 and Rpt4/5 coils can undergo substantial swinging motions without clashes with surrounding subunits (Fig. S8); our structural data indicate that the Rpt1/2 and Rpt4/5 coiled coils can rotate by up to approximately 40° around the hinge residues corresponding to Pro62 in PAN. This flexibility enables them to grab substrates bound to Rpn10 and Rpn13 via tetraubiquitin chains (33) and dangling into the space accessible to the swinging arms (Fig. 7II). It has already been shown that the PAN coiled coils bind substrates (72, 73), which suggests that the coiled coils have the capability to bind substrates with appropriate degrons (74).

Subsequently, the coiled coils expose the substrates to Rpn11 (Fig. 7III), which is located above the central opening of the AAA-ATPase. The substrates are deubiquitylated by Rpn11 before being unfolded in the upper AAA-ATPase cavity (N ring) and translocated to the CP where degradation occurs. The coiled-coil motions might trigger gate opening to the CP because the coiled coils are positioned above the Rpt subunits (Rpt2, Rpt3, and Rpt5) whose C termini open the gate (45).

Materials and Methods

Sample Preparation, Cryo-EM, and XL/MS.

Sample preparation, cryo-EM single-particle analysis, and XL/MS were performed as described in ref. 22 and explained in more detail in SI Methods.

Integrative Structure Determination.

In brief, multiple comparative models were built for each subunit, fitted into the cryo-EM map, and the best-scoring configurations were subjected to flexible fitting (SI Methods).