Near-atomic resolution structural model of the yeast 26S proteasome

Abstract

The 26S proteasome operates at the executive end of the ubiquitin-proteasome pathway. Here, we present a cryo-EM structure of the Saccharomyces cerevisiae 26S proteasome at a resolution of 7.4 Å or 6.7 Å (Fourier-Shell Correlation of 0.5 or 0.3, respectively). We used this map in conjunction with molecular dynamics-based flexible fitting to build a near-atomic resolution model of the holocomplex. The quality of the map allowed us to assign α-helices, the predominant secondary structure element of the regulatory particle subunits, throughout the entire map. We were able to determine the architecture of the Rpn8/Rpn11 heterodimer, which had hitherto remained elusive. The MPN domain of Rpn11 is positioned directly above the AAA-ATPase N-ring suggesting that Rpn11 deubiquitylates substrates immediately following commitment and prior to their unfolding by the AAA-ATPase module. The MPN domain of Rpn11 dimerizes with that of Rpn8 and the C-termini of both subunits form long helices, which are integral parts of a coiled-coil module. Together with the C-terminal helices of the six PCI-domain subunits they form a very large coiled-coil bundle, which appears to serve as a flexible anchoring device for all the lid subunits.

The 26S proteasome is a 2.5 MDa molecular machine designed for the controlled degradation of proteins marked for destruction by the covalent attachment of polyubiquitin chains [for reviews see (1–3)]. It is composed of two copies, each of 33 canonical subunits, as well as some proteasome interacting proteins (PIPs). The 26S holocomplex comprises two types of subcomplexes: the cylindrical 20S core particle (CP) harbouring the proteolytic chamber and the two 19S regulatory particles (RPs), which attach to opposite ends of CP cylinder. The RPs have multiple roles in preparing substrates for degradation: They recognize and bind ubiquitylated proteins, they deubiquitylate them followed by their unfolding, and they control the opening of the gate which gives access to the interior of the CP.

While the structure of the 20S core complex was determined by X-ray crystallography almost two decades ago (4, 5), the structure of the 26S complex remained recalcitrant to crystallization attempts, presumably due to its conformational and compositional heterogeneity. Recently, the subunit architecture of the holocomplex has been determined by cryo-electron microscopy (EM) single particle analysis (SPA and ref. 6, 7) independently by two groups using different approaches for the assignment of RP subunits. Lander, et al. (6) obtained a 9 Å resolution map (Fourier-shell correlation, FSC = 0.5) of the 26S Saccharomyces cerevisiae proteasome and they determined the subunit positions by means of fusion constructs and automated segmentation methods. Lasker, et al. (7) performed an exhaustive computational search of possible subunit configurations within the boundaries of an 8.5 Å map of the 26S proteasome from Schizosaccharomyces pombe scoring possible configurations against the large number of experimental restraints. These approximate subunit positions guided the docking of comparative models of the subunits into the EM map, which were then refined by flexible fitting.

Subsequently, and in light of the now established subunit architecture, another group proposed a structure for the human 26S complex based on rigid body docking of existing crystal structures of subunits and homology models into a 9 Å EM map of the Homo sapiens complex (8). The authors suggested that “redefinition” of the subunit architecture was necessary. Specifically, three subunits (Rpn12, Rpn8, Rpn11) out of 12 non-ATPase subunits were assigned to positions different from those in the aforementioned structures (6, 7).

Here we present a higher-resolution map of the S. cerevisiae 26S complex. This structure was generated from over two million particles, which allowed us to reach 7.4 Å resolution without imposing C2 symmetry onto the reconstruction. Given the presence of nonstoichiometric PIPs it is to be expected that the C2 symmetry does not extend beyond the CP and deviations from symmetry may well be functionally important. The nonsymmetrized map enabled us to account for subtle conformational differences between the two RPs. α-helices, the predominant secondary structure elements of the RP subunits, could be identified with a high degree of confidence throughout the entire map and were assigned with the help of atomic structures of subunits (9–13) or homology models (14). To improve the fit of atomic models to the density we used molecular dynamics flexible fitting (15). Similar to X-ray crystallographic refinement conformations are retrieved that not only comply with the experimental density but also molecular dynamic force fields, which makes interpretation beyond the nominal resolution of the map possible (16, 17). While the resulting model of the 26S holocomplex might not be as accurate as models derived from high-resolution crystallographic data, it nevertheless provides a basis for a deeper understanding of the sequence of events prior to substrate degradation in the 20S core particle.

Results and Discussion

Automated Single Particle Acquisition.

26S proteasomes were purified from S. cerevisiae cells with a C-terminal 3xFLAG tag at Rpn11. To obtain large data sets as needed for high-resolution structure determination by SPA, we implemented an automated pipeline for data acquisition and processing. Automated data acquisition was performed on a FEI Titan Krios using the TOM² package (18). Of a total of 120,000 8k × 8k pixel images of 26S proteasomes more than 63,000 images of suitable quality were selected for reconstruction (Table S1). From this dataset we extracted more than 2.4 million individual particles and used them for structure determination.

C2 Symmetry of 26S proteasome is Broken.

Initially, we reconstructed a 3D density map of the 26S proteasome using C2 symmetry as done previously (6–8). According to the FSC = 0.5 criterion the resolution was 6.8 Å. Using the less stringent FSC = 0.3 measure, the global resolution of our map was determined to be 6.0 Å. However, the resolution of the two RP volumes was significantly lower than that of the CP (Fig. S1). In an effort to improve the resolution of the RP we decided not to impose C2 symmetry. While the nominal resolution decreased to approximately 7.4 Å (or 6.7 Å, Fig. S2), a higher level of detail became visible in parts of the RP and the reconstruction revealed significant structural differences between the two RPs (Fig. 1). The helices are resolved better in one RP compared to its counterpart, which is most pronounced in its distal parts (Fig. 1 D and E). An accurate assessment of the local resolution was obtained by local FSC (Fig. 1): the resolution of the 26S proteasome varies locally from approximately 6.5 Å to up to 12 Å in the most variable parts of the RP. For a detailed analysis of the holocomplex we focused on the better resolved RP.

Fig. 1.

7.4 Å resolution EM single particle reconstruction of S.cerevisiae 26S proteasome without imposed symmetry. The density is displayed as an isosurface from two different views, on the right colored according to the local resolution as indicated by the color key. Five different slices from the upper RP (A–E) are shown and compared to the counterparts in the lower RP (A’–E’) by difference images.

Structural Organization of the Non-ATPases in the Base Complex.

In the light of the recent subnanometer resolution structures the definition of “base” and “lid” subcomplexes of the RP (19), must be revisited. These terms no longer accurately describe subcomplexes as positioned proximally (base) and distally (lid) with respect to the CP; they rather refer to assembly modules [reviewed in (20)]. The base consists of the six RP AAA-ATPases (Rpt1-6) and the RP Non-ATPases Rpn1, Rpn2, and Rpn13. Rpt1-6 assemble into a heterohexamer, which is responsible for substrate unfolding, for translocation of substrates into the CP, and for opening of the gate formed by the α-subunit rings of the CP.

We fitted the recently determined crystal structure of Rpn2 (11) and the crystal structure of Rpn13 (12) as rigid bodies into the cryo-EM map (Fig. 2A, Movie S1). The crystal structure of Rpn2 fits well into the density map at the previously suggested positions (6–8). Subsequent flexible fitting resulted in only minor structural changes (C-α RMSD < 2.2 Å), mostly in the N-terminal domain (Fig. 2B). In contrast, we obtained ambiguous fits of Rpn13 around its previously suggested location (6, 7), which is due to the protein’s small size (18 kDa) and the low resolution of the density map in the corresponding region (Fig. 1). Of the five configurations with similar cross-correlation coefficients, we chose one that is consistent with the Rpn2-exposed interface as determined by NMR spectroscopy (12).

Fig. 2.

Atomic model of the 26S holocomplex fitted into the cryo-EM map. (A) Holocomplex seen from three different views. In the model, the PCI subunits are colored in different shades of green (order from left to right in middle panel: Rpn9/5/6/7/3/12), the MPN subunits in light (Rpn8) and dark (Rpn11) magenta, the Ub receptors Rpn10 and Rpn13 in purple, Rpn1 in brown, Rpn2 in yellow, the AAA-ATPase hexamer in blue, and the CP in red. (B) Subunit models fitted into their corresponding experimental densities. The ubiquitin receptors Rpn10 and Rpn13 were omitted due to their small size and due to the low resolution of the EM map in the corresponding areas.

For Rpn1, we built a comparative model using the Rpn2 structure as a template and fitted it as a rigid body into its corresponding EM density, which led to substantial differences between the map and the model. By dividing the Rpn1 model into two rigid parts, the central proteasome cyclosome (PC) domain and a segment comprising the N- and C-terminal domains, we obtained an improved rigid body fit (Fig. 2). Here, the PC torus is positioned somewhat differently in Rpn1 compared to Rpn2. However, the Rpn1 density is relatively poorly resolved, presumably due to the conformational heterogeneity of this subunit. Therefore, we refrained from using MDFF methods to refine the model (Fig. 2).

Structural Organization of the AAA-ATPases.

We previously presented a preliminary model of the AAA-ATPase module in complex with the CP based on low- and medium- resolution cryo-EM data, the domain structures of the homohexameric PAN complex (the archaeal homolog of the Rpt module, refs. 21, 22), as well as on protein-protein interaction data (23–25). The heterohexamer forms a trimer of dimers with the pairs Rpt1/Rpt2, Rpt6/Rpt3, and Rpt4/Rpt5 each held together by coiled coils formed by near N-terminal regions of these subunits. The coiled coils protrude from a ring formed by oligosaccharide binding (OB) domains of the six subunits (N-ring). The N-ring is positioned atop the hexameric ring formed by the AAA-ATPase domains (AAA-ring), but the two rings are not in register.

Our previous comparative model served as the starting model for the AAA-ATPase module (7). Bioinformatics predictions indicate that the coiled coils of the three AAA-ATPase pairs differ substantially in length: They consist of 27, 44, and 33 residues for Rpt1/Rpt2, Rpt6/Rpt3, and Rpt4/Rpt5, respectively. We created a template covering the coiled coils over their entire predicted length by N-terminal extension of the helices assuming a 7-residue periodicity of coiled coils (Fig. S3A). Subsequently, we refined the model by flexible fitting into the cryo-EM density using MDFF (15). The correctness of the resulting model at the secondary structure element level is reassured by the coincidence of helices within the model with clearly recognizable helical density elements in the EM map (Fig. 3).

Fig. 3.

Structure of the AAA-ATPase hexamer. (A) The atomic model of the AAA-ATPase (dark blue: Rpt1/6/4; light blue: Rpt2/3/5) is displayed together with the segmented cryo-EM density in isosurface representation. The Walker A and Walker B motifs are colored in red and orange, respectively. The densities marked by the dashed ovals corresponds to the likely disordered N-termini. (B) Slice through the N-ring as indicated in A. (C) Slice through the AAA-fold.

The N-terminal end of the coiled-coil pair of the AAA-ATPases Rpt6/Rpt3 makes contact with the PC-domain of Rpn2. Moreover, the Rpt6/Rpt3 coiled coil interacts with a large coiled-coil bundle consisting of the C-termini of the lid subunits (see below). This may restrict the freedom of the Rpt6/Rpt3 coiled-coil pair to move in a “swinging arm” mode (7). The 27-residue Rpt1/Rpt2 coiled coils appear to interact with Rpn1 in a similar fashion as Rpt6/Rpt3 with Rpn2, but the Rpt1/Rpt2/Rpn1 interaction is resolved less well. The 33-residue Rpt4/Rpt5 coiled coil does not interact with non-ATPases and also is resolved relatively poorly indicating a high degree of conformational freedom.

The approximately 50 N-terminal residues of all ATPases are predicted to be disordered like the N-termini of their archaeal and bacterial counterparts PAN and ARC. Consistent with these predictions, we observe only diffuse densities at the N-terminal ends of the coiled coils. The N-terminal of approximately 100 residues (comprising disordered residues and coiled coils) of the bacterial homolog of the 26S AAA-ATPases, ARC, have been shown to be essential for substrate recognition (26). We hypothesize that the unfolded N-termini bind to secondary degradation signals (typically loosely folded domains) when they irreversibly commit for degradation (27).

The N-ring is only loosely connected to the AAA-ring and its pore axis does not coincide with that of the AAA-ring. The linkers connecting N-ring and AAA-ring are resolved for Rpt1, Rpt3, Rpt4, and Rpt5. The loose interaction is consistent with the ATP-independent chaperone activity of the N-ring (21).

A characteristic feature of AAA-ATPases is a central pore, which is formed by loops from each subunit (the “pore loops”). The primary structures of the pore loops are highly conserved throughout the Rpts (Fig. S3B). Rigid body fitting of the PAN structure into the 26S proteasome density suggested that the pore loops of Rpt3, Rpt4, Rpt5, Rpt1, and Rpt2 are arranged in a regular “staircase” fashion and Rpt6 adopts an intermediate position between the top (Rpt3) and bottom (Rpt2) of the staircase (6). The flexibly fitted model confirms the staircase arrangement and moreover reveals that the helix immediately preceding the pore loop of Rpt6 is tilted by approximately 32° compared to its counterpart in PAN whereas the corresponding helices of the other five Rpt subunits are similar to PAN (Fig. S4, Movie S2). Thus, the flexibly fitted model suggests that the structure of the Rpt6 pore loop is substantially different from those subunits forming the staircase.

In the AAA-ring of our model the ATP binding sites consisting of Walker A and Walker B motifs of all subunits show the canonical structure and appear accessible suggesting that all subunits are capable of hydrolyzing ATP (Fig. 3 and Fig. S4). This finding is consistent with the high sequence conservation of the Walker A and Walker B motifs in the Rpt subunits (Fig. S3B) and the observation that the Walker A and Walker B motifs of all six Rpts are essential (28). A further remarkable feature of the Rpts is that the orientations of the small AAA+ domains with respect to the large AAA+ domains vary substantially (Fig. S4). Hinge-motions of the two AAA+ domains with respect to each other can be a result of different nucleotide binding states (29), but the resolution of our reconstruction is not sufficient to localize nucleotides.

The C-termini of Rpt2, Rpt3, and Rpt5 contain a hydrophobic-tyrosine-X (HbYX) motif, which has been shown to induce opening the gate to the CP (30, 31). In the EM map, the densities corresponding to the C-termini of all of these subunits are well resolved in the EM map (Figs. 1E and 3A). Interestingly, the Rpt5 density is slightly weaker than those of Rpt2 and Rpt3. In fact, some promiscuity of the Rpt5 C-terminus was observed in cross-linking experiments (32). For the remaining three ATPase C-termini we could not observe any density suggesting that they are disordered.

Organization of the PCI Subunits.

Six lid subunits, Rpn3, Rpn5, Rpn6, Rpn7, Rpn9, and Rpn12, share a proteasome-cyclosome-initiation factor (PCI) domain. According to the model of the lid subcomplex determined by Lander, et al. (6) and Lasker, et al. (7), the PCI-domains form a horseshoe. Each PCI domain is preceded by solenoids, which project away from the horseshoe scaffold. Interestingly, the N-terminal solenoids of Rpn6, and to a lesser extent Rpn5, make contact to the CP. The proposed arrangement is consistent with the locations of MBPs fused to the termini of the subunits as well as with site-specific cross-linking data. Moreover, the arrangement of the PCI domains is similar to the one observed in the crystal contacts in the Rpn6 crystal structure (9). We positioned comparative models at their respective sites and refined them by MDFF flexible fitting (Fig. 2). Prior to flexible fitting, the PCI domains yielded good agreement of helices visible in the density map and all the subunit models, whereas more substantial differences between the N-terminal solenoids in the initial model and the map were observed. After refinement helices in the models and in the map coincided throughout the modeled length indicating that our models are accurate at this level of detail (Fig. 2). In summary, we have strong evidence to believe that the reassignment of Rpn12 (8) is erroneous and this subunit is positioned as originally proposed (6, 7). A comparison of single particle reconstructions to the crystal structure of Rpn6 (Fig. S5) suggests that the high noise level might be responsible for misassignment in (8).

Organization of the MPN Domains and Rpn10.

The metalloprotease Rpn11 and the homologous but catalytically inactive Rpn8 share an MPN domain. Both subunits have been suggested to form a heterodimer positioned above the mouth of the AAA-ATPase heterohexamer near Rpt3 and Rpt4 (6, 7). The structural organization of the lid suggests that the PCI subunits form a scaffold that positions the Rpn8/Rpn11 heterodimer in close vicinity to the mouth of the AAA-ATPase and “clamps” the RP to the CP.

We have previously located the Rpn8 MPN domain by fitting its crystallographic structure into the EM map of the S. pombe 26S proteasome (7). In our S. cerevisiae map, the Rpn8 MPN domain is at approximately the same position (Fig. 4 and Fig. S6). The correlation coefficient of this fit is markedly higher than that of alternative positions and helices in template and density coincide, which supports this positioning of the Rpn8 MPN domain over alternative assignments in previous studies (6, 8).

Fig. 4.

Organization of Rpn8/Rpn11 dimer and lid C-termini. (A) Localization of the MPN domains (colored tan) and the coiled-coil bundle likely formed by the lid C-termini (green) in the context of the 26S holocomplex (red: CP, blue: AAA-ATPase hexamer, grey: remaining subunits) in a side view. (B) Same seen from top. (C) Magnified top-view of MPN densities and coiled-coil bundle. The atomic models of Rpn8 (light magenta) and Rpn11 (magenta) are fitted into the density. The interface of the MPN domains of Rpn8 and Rpn11 is formed by helices ⁸H1/¹¹H1, ⁸H2/¹¹H3, and ⁸H4/¹¹H5. In the coiled-coil bundle Rpn11 helices ¹¹H6 and ¹¹H7 and Rpn8 helix ⁸H5 could be traced. (D) Rpn8/Rpn11 dimer in the context of AAA-ATPases and coiled-coil bundle.

We used a hypothesis inspired by X-ray crystallography studies of the Rpn8 MPN domain to localize the structurally similar Rpn11. The isolated Rpn8 MPN domain forms homodimers in solution and it was suggested that the structure of the homodimer is similar to that observed in crystal contacts (10). Indeed, when we positioned a comparative model of the Rpn11 MPN domain relative to the Rpn8 MPN domain mimicking the crystal contact the corresponding density displayed a similar fold. Upon flexible fitting, the positions of the secondary structure elements changed slightly, but the overall fold remained the same (Fig. 4C). The Rpn11 MPN domain was localized essentially where the Rpn11 N-terminus was localized previously (6). This placement of the Rpn11 MPN domain is also supported by residue-specific cross-links (7): Rpn11 residue K150 is cross-linked to K59 of Rpt6, which is at a Cα–Cα distance of approximately 13 Å.

We interpret the possible deubiquitylation mechanism of Rpn11 in the light of the crystal structure of the deubiquitylating enzyme (DUB) AMSH-LP, which was crystallized in complex with K-63 linked diubiquitin (33). Sequence analysis suggests that binding of the distal ubiquitin (Ub) and DUB catalysis is similar for Rpn11 and AMSH-LP (33). The binding site of proximal Ub is not conserved suggesting that proximal Ub may bind differently in AMSH-LP (which is specific for K-63 linked polyubiquitin) and Rpn11 (which is likely to be less specific). We superposed the crystal structure onto our model of the Rpn11 MPN domain. The distal Ub moiety becomes positioned near the Rpt4/Rpt5 coiled coils near the surface of the 26S holocomplex and the proximal Ub is positioned in the center of the OB-ring of the AAA-ATPase module (Fig. 4D). The positioning of the distal Ub does not cause steric clashes with the 26S holocomplex and the proximal Ub exhibits only minor clashes with the N-ring (Rpt6). This suggests the following scenario: In order to be deubiquitylated a substrate must be committed for degradation and deubiquitylation occurs immediately before substrate unfolding in the OB-ring. Binding of Ub at the proximal Rpn11 site might be relatively unspecific, as suggested by the sequence comparison to AMSH-LP (33). The committed substrate then pushes the Ub moiety to the site for the distal Ub where specific binding and cleavage of the Ub chain occurs. The distal Ub is released and another Ub moiety may be bound for deubiquitylation. However, the positioning of the Rpn11 active site would also comply with a recently proposed model that substrates are pulled into the AAA-ATPase by the deubiquitylating Rpn11 (6).

We confirmed the previously suggested position of Rpn10 (6–8) by fitting the crystal structure of the Rpn10 von Willebrand domain into the density map (Fig. S6). Rpn10 is associated with the 26S holocomplex primarily via Rpn8.

A Coiled-Coil Bundle Formed by the C-terminal Helices of the Lid Subunits.

Adjacent to the heterodimer formed by the MPN domains of Rpn8 and Rpn11 a hitherto unassigned density consisting of long twisted helices (a coiled-coil bundle) was observed. In the EM density we could trace Rpn11 into this density towards its C-terminus using the model building program COOT (34). The C-terminus of Rpn11 consists of two long helices (¹¹H6 and ¹¹H7 with 41 and 29 residues, respectively) connected by a linker of approximately 6 residues. The observed length of both helices agrees well with predictions (Fig. S7). ¹¹H6 is kinked and its C-terminal segment forms a coiled coil with ¹¹H7. The linker of ¹¹H6 and ¹¹H7 is positioned in close proximity to the Rpt6/Rpt3 coiled coil. Residue-specific cross-links of the linker and Rpt3 in its coiled coil (residue 87) as well as its N-ring domain (residue 132) strongly support our assignment of Rpn11 (7). Moreover, the C-terminus of Rpn11 is accessible from the outside, which explains why C-terminal fusion constructs can be used for purification of intact 26S proteasomes (35) and the position is consistent with cryo-EM data of previous antibody labels of the Rpn11 C-terminus (25). Similarly, we attempted to trace Rpn8 towards its C-terminus. We could only assign a 30-residue helix ⁸H5 that follows the MPN domain. Secondary structure prediction suggests that ⁸H5 is followed by at least one further helix with approximately 50 residues. However, we could not unambiguously trace Rpn8 beyond ⁸H5 because many short helical elements are present in the EM density near the C-terminus of ⁸H5.

The volume of the coiled-coil bundle is too large to be formed by the C-terminal domains of Rpn8 and Rpn11 alone and there is evidence that the PCI subunits also contribute to it. Secondary structure prediction indicates that the C-termini of all PCI subunits form long helices (Fig. S7) with high coiled-coil formation propensity for many of them (14). These C-terminal helices are shown to mediate the interactions between the PCI subunits, as for example shown for Rpn6 and Rpn7 (9). This has been further validated by genetics experiments showing that C-terminal deletions cause assembly defects in the lid. For example, deletions of C-terminal Rpn11 residues result in accumulation of partially assembled proteasomes and the resulting growth defects that can be rescued in trans (36, 37). Likewise, the C-terminal helix of Rpn12 is essential for proteasome assembly (38).

At this point, we cannot unambiguously localize the C-termini of the PCI-domain-containing subunits. However, we can provide some clues for their approximate localization. The Rpn12 C-terminus has been suggested to be localized adjacent to Rpt3 (38); i.e., in proximity to the Rpn11 C-terminus. Secondary structure prediction indicates that the Rpn3 C-terminus consists of two long helices. Cross-linking has shown that residue 487 of Rpn3 and residue 278 of Rpn11 are in close proximity (7). Moreover, strong interaction was reported for Rpn3 and Rpn11 (39), which must occur at the C-termini because the other domains are too far apart in the 26S holocomplex.

Low-resolution negative-stain electron microscopy has indicated that the structure of the lid changes conspicuously upon integration into the RP, in particular in the vicinity of the Rpn11 MPN domain (6). Indeed, the local resolution assessment of our map as well as preliminary classification results suggest that the lid adopts different conformational states. Thus, the coiled-coil bundle formed by the C-termini of the lid subunits holds the lid together while allowing for large movements of its subunits.

A Possible Regulatory Function of the Coiled-Coil Bundle.

We hypothesize that the C-terminus of Rpn11 and the coiled-coil bundle into which it is integrated may serve yet another function. It is a common feature of regulated proteases that they are activated only when assembly is complete and subunits have found their designated position. We propose that, when Rpn11 is not integrated into the RP, its C-terminus blocks the active site, which would explain why integration of Rpn11 is essential for its activation and why it is not accessible to common DUB inhibitors (40, 41). Thus, the large structural changes in the isolated lid compared to its structure in the 26S holocomplex might be correlated with activation of Rpn11 (6). Our hypothesis could also explain why deletion of the C-terminal Rpn11 helix (m-Rpn11) results in defects in mitochondrial morphogenesis (37). The phenotype appears similar to that observed for deletion of MARCH5, an E3 ubiquitin ligase that ubiquitylates Fzo1 (mitofusin), a mitochondrial protein promoting mitochondria fusion (42, 43). A possible explanation for the similar phenotype of the Rpn11 mutant would be that unregulated deubiquitylation of Fzo1 by Rpn11 reverts ubiquitylation by MARCH5, which would be supported by the enrichment of mutant m-Rpn11 bound to mitochondria compared to wildtype Rpn11 (37). Interestingly, mutations in Rpn3 result in a very similar phenotype as observed for m-Rpn11 and MARCH5 deletion (39). This finding suggests that a disruption of the Rpn3-Rpn11 interactions at the C-terminal coiled coils also renders Rpn11 active even when not integrated into the 26S proteasome.

Conclusions

High-throughput cryo-electron microscopy enabled us to collect micrographs of almost 2.5 million particles, which allowed us to obtain a 7.4 Å resolution map of the 26S proteasome without imposing C2 symmetry. The model corroborates the positioning of subunits in recent models (6, 7) and it refutes claims of a redefinition (8). The refined model clarifies the structural organization of the Rpn8/Rpn11 heterodimer and allows us to put forward a hypothesis for the mechanism of deubiquitylation. A previously unassigned density was identified as a large coiled-coil bundle serving as a flexible anchoring device for the lid subunits. The size of the single particle data set will allow us to perform a deep classification in future studies and thereby provide a first glimpse of the dynamics of the 26S proteasome during its functional cycle.

Materials and Methods

Sample Preparation and Cryo-EM.

26S proteasomes were purified from ^3xFLAGRpn11 S. cerevisiae cells as described previously (35). The sample was applied to holey carbon grids (Quantifoil Micro Tools, Germany), vitrified, and imaged using a Titan Krios electron microscope (FEI, Netherlands). Data were automatically acquired using TOM² (18). Images were corrected for the contrast transfer function in TOM (44), particles were localized automatically based on (45), and reconstruction was performed using XMIPP (46).

Modeling.

Initial models of S.cerevisiae CP and Rpn2 were their crystal structures (PDB ID code 1RYP and 4ADY, respectively). Comparative models initially represented the remaining subunits. Templates were obtained from HHpred (47) or from our previous atomic model (7). Models were built using MODELLER (48). Segments without structural templates were removed from the models. These initial models were first rigidly fitted into the EM map with Chimera (49) followed by flexible fitting with MDFF (15) using a simulated annealing protocol and implicit solvent. Special restraints for β-strands were introduced for the PCI domains to capture the crystal contacts of Rpn6 D.melanogaster (PDB code 3TXN) as described previously (7). AAA-ATPases, PCI-domain-containing subunits and MPN-domain-containing subunits were first fitted separately while keeping the other components fixed. Finally, these subunits were fitted simultaneously. Rpn1, Rpn10 and Rpn13, which were located in comparably poorly resolved parts of the map, as well as the unassigned helices in the coiled-coil bundle were not treated flexibly. The accuracy of the model was assessed by local cross-correlation of map and model (Fig. S8).