Proteasome substrate capture and gate opening by the accessory factor PafE from Mycobacterium tuberculosis

In all domains of life, proteasomes are gated, chambered proteases that require opening by activators to facilitate protein degradation. Twelve proteasome accessory factor E (PafE) monomers assemble into a single dodecameric ring that promotes proteolysis required for the full virulence of the human bacterial pathogen Mycobacterium tuberculosis. Whereas the best characterized proteasome activators use ATP to deliver proteins into a proteasome, PafE does not require ATP. Here, to unravel the mechanism of PafE-mediated protein targeting and proteasome activation, we studied the interactions of PafE with native substrates, including a newly identified proteasome substrate, the ParA-like protein, Rv3213c, and with proteasome core particles. We characterized the function of a highly conserved feature in bacterial proteasome activator proteins: a glycine–glutamine–tyrosine–leucine (GQYL) motif at their C termini that is essential for stimulating proteolysis. Using cryo-electron microscopy (cryo-EM), we found that the GQYL motif of PafE interacts with specific residues in the α subunits of the proteasome core particle to trigger gate opening and degradation. Finally, we also found that PafE rings have 40-Å openings lined with hydrophobic residues that form a chamber for capturing substrates before they are degraded, suggesting PafE has a previously unrecognized chaperone activity. In summary, we have identified the interactions between PafE and the proteasome core particle that cause conformational changes leading to the opening of the proteasome gate and have uncovered a mechanism of PafE-mediated substrate degradation. Collectively, our results provide detailed insights into the mechanism of ATP-independent proteasome degradation in bacteria.

Mycobacterium tuberculosis (Mtb), 3 the causative agent of tuberculosis, has a proteasome system that is essential for causing lethal infections in animals by what are likely to be a multitude of reasons (reviewed in Ref. 1). In all domains of life, proteasomes include a 20S core particle (CP), which is a barrelshaped complex formed by stacking of four heptameric rings: two ␤-rings sandwiched by two ␣-rings, with some or all of the ␤-ring subunits having catalytic activity (2)(3)(4). In Mtb, each ␤-ring is formed by seven identical PrcB subunits and each ␣-ring is composed of seven identical PrcA subunits. The Mtb 20S CP is tightly gated to prevent unregulated proteolysis (5,6). Proteasomal degradation in Mtb is promoted by two distinct pathways (reviewed in Ref . 7): an ATP-dependent pathway catalyzed by Mycobacterium proteasomal ATPase (Mpa) and an ATP-independent pathway mediated by PafE (also known as Bpa for bacterial proteasome activator) (3,8). To stimulate degradation, Mpa or PafE cap either one or both ends of a 20S CP using GQYL motifs at the C termini of each monomer of an activator complex. In addition to ATP, the Mpa-proteasome requires proteins to be post-translationally modified with the small protein Pup (prokaryotic ubiquitin-like protein) to target them for degradation (9,10). In contrast, the PafE-proteasome neither requires ATP nor interaction with a post-translational modification on a doomed protein for degradation (8,11,12). Instead, it appears that PafE can facilitate the degradation of unfolded proteins and at least one native protein, HspR, without a post-translational modification (11)(12)(13). A failure to degrade HspR by the PafE-proteasome results in slowed growth in vitro (12) and likely contributes to the attenuated growth phenotype in vivo (11).
PafE is a small protein of 174 amino acids that is functionally similar but evolutionally unrelated to the ATP-independent eukaryotic proteasome activators PA26 and PA28. Like PA26 and PA28, PafE also has a four ␣-helix bundle fold (13)(14)(15). However, PafE is unique in two ways: 1) its unstructured, proteasome-activating C terminus is 21 amino acids long, which is nearly twice the length of the unstructured C termini of PA26 and PA28; and 2) PafE monomers assemble into dodecameric rings with 12-fold symmetry (13,16), in contrast to the heptameric rings of PA26 and PA28 (17). We previously investigated the function of the extended PafE C terminus and found it con-  fers suboptimal binding to 20S CPs; shortening the C terminus while maintaining the GQYL motif results in a dramatic increase in proteolysis by Mtb 20S CPs (13). The interaction between PafE and the Mtb 20S CP was recently studied by cryo-electron microscopy (EM) (16). The structure of the PafE-20S CP complex at 3.5-Å resolution shows the asymmetric binding of PafE rings to 20S CPs as well as interactions between the GQYL motifs of PafE and the proteasome ␣-subunits. However, there are several caveats to the interpretation of this structure. The study used a mutated 20S CP to which PafE would more strongly bind: the N-terminal eight residues of the ␣-subunits (PrcA) of the 20S CP were deleted, creating an "open gate" mutant 20S CP (6); and, the catalytic threonine (Thr) of each 20S CP ␤-subunit (PrcB) was changed to alanine ("T1A" mutant), which increases binding with PafE (11). Thus, the 20S CP was used in a "pre-activated" form, preventing any conclusions from being drawn regarding the impact of PafE binding on proteasome activation. Furthermore, the site-specific cross-linker para-benzoyl-L-phenylalanine was used in place of threonine 170 in PafE, a residue immediately preceding the GQYL motif. Conceivably, the cross-linking of PafE with 20S CPs may create unintentional interactions between amino acids that are not involved in proteasome gate activation or might obscure relevant structural features (16). Although these previous studies were informative, we felt that more details on the mechanism of proteasome gate opening by PafE remained to be elucidated.
In this study, we performed cryo-EM experiments to show that, the GQYL sequence in PafE makes specific interactions with residues in PrcA to cause a rotation of Mtb 20S CP ␣-subunits. This rotation leads to an outward movement and disorder of the PrcA N termini, which opens the gate into the proteasome core protease. Finally, we found that the large hydrophobic opening of PafE may give it a previously unrecognized chaperone-like activity with the ability to capture substrates for delivery to 20S CPs for destruction.

Comparison of the 20S ␣-ring structures with and without PafE capping reveals a proteasome gate opening mechanism by PafE
We used cryo-EM to understand how PafE with an intact GQYL activation motif opens the substrate entry gate of the Mtb 20S CP. PafE rings have a low affinity for 20S CPs, but shortening the C termini of PafE without deleting the GQYL motifs (PafE(⌬155-166)) increases its affinity for 20S CPs by reducing the distance between a PafE ring and a 20S CP ␣-ring (13). To better understand the interaction between PafE rings and 20S CPs, we used PafE(⌬155-166) with 20S CPs for single particle cryo-EM analysis. Roughly 180,000 particles were automatically picked from 4,883 cryo-EM images and then submitted to reference-free 2D classification in RELION2 (18). Only 20S CPs capped by PafE(⌬155-166) rings on either or both sides were selected for 3D classification (Fig. S1). Refinement of singly capped 20S CPs without applying any symmetry yielded a 3D density map with an average resolution of 4.2 Å, as estimated by Fourier shell correlation (FSC) of two independent maps calculated with two halves of the particles ( Fig. S2 and Table 1). However, the local resolution varied considerably: the 20S CP region was in the resolution range of 4 -6 Å, with resolved secondary structures and even a few large side chains, but the PafE ring had a resolution of ϳ10 Å. The low resolution of PafE density was expected due to the flexible PafE C termini interacting with the 20S CPs (13,16). For the doubly PafEcapped 20S CP, we masked PafE densities at both ends of the CP particle, and refined the 20S CP region alone by applying D7 symmetry, which led to an improved 3D density map at 3.4 Å average resolution (Figs. S1 and S3, A and B, and Table 1). Based on a local resolution estimation, the ␤-rings of the 20S CP have a better resolution of 2.5-3 Å, and the ␣-rings at the two ends have a lower resolution of 3.2-3.8 Å (Fig. S3C).
Although the singly-capped 20S CP structure had a lower resolution than the doubly-capped 20S CP structure, it provided a unique opportunity to compare the open versus closed gates within the same structure (Fig. 1A). As expected, the EM map of PafE(⌬155-166)-bound 20S CPs shares a similar architecture with the 3D EM maps reported previously, which were either at a much lower resolution of 12 Å (13), or at a resolution of 3.5 Å but using chemically modified PafE and pre-activated 20S CPs (16). At the PafE-capped end, the first ordered residue in PrcA is Ser-8. A cloud of discontinuous densities in the center and above the ␣-ring is visible when the 3D map is rendered at a lower display threshold of 5.8 (Fig. 1B, left panel). These weak and discontinuous densities are presumably from the disordered N termini of PrcA, each consisting of seven amino acids. The discontinuous densities disappeared when the 3D map was surface rendered at the normal threshold of 7.0, revealing an opening of 27 Å in diameter (Fig. 1B, right side). The uncapped end of the 20S CP, by contrast, had the gate tightly sealed by well defined PrcA N termini (Fig. 1C). When Table 1 EM data collection and refinement statistics

Proteasome activation by PafE
the PafE-capped half of the 20S CP was aligned to the uncapped half, we found that the ␤-rings were unchanged, but each of the PrcA subunits had undergone a 3°, in-plane rotation ( Fig. 1D and Fig. S4). PrcA rotation moves its N-terminal ␣-helix 0 (H0) outward by 3 Å, thereby increasing the overall diameter of the opening by 6 Å (Fig. 1D).

PafE GQYL motif interacts with Arg-26 and Lys-52 of the 20S CP ␣-subunits
The 3.4-Å resolution of the doubly PafE-capped 20S CP structure provided a detailed description of the interactions between the activation GQYL motifs of PafE rings with PrcA subunits in the ␣-ring ( Fig. 2A). We found Tyr-173 of PafE forms a cationinteraction with Arg-26 of PrcA, whereas the carboxyl group of Leu-174 of PafE forms a salt bridge with Lys-52 of PrcA. Amino acid sequence alignment of the 20S ␣-subunits shows that Lys-52 is conserved across the domains of life, but Arg-26 is conserved only in prokaryotes (Fig. 2B). We previously demonstrated that Lys-52 of PrcA is essential for PafE-dependent degradation (13). To test the significance of the interaction between PafE Tyr-173 and PrcA Arg-26, we substituted Arg-26 to an alanine (PrcA(R26A)) and tested if the mutated 20S CP could degrade HspR. We found that 20S CPs with PrcA R26A failed to degrade HspR (heat shock protein repressor), a known PafE-proteasome substrate in Mtb (Fig. 2, C and D). This result suggests that PrcA Arg-26 is essential for PafE to stimulate gate opening to substrates. Alignment of the atomic models showed both PrcA Arg-26 and Lys-52 were pushed outwards in the presence of the PafE GQYL motif (Fig.  2E). Taking these data together, we suggest the mechanism of  (5.8) showing the weaker density of PafE ring (left) or rendered at the normal threshold (7.0) such that most PafE ring is invisible except for the well ordered C termini (right). The densities of the PafE C termini are colored in salmon. The orange and cyan circles mark the circumferences of the PafE and 20S CP, respectively. C, a view of the uncapped end of a 20S CP by 3D density map rendered at the normal threshold (7.0). D, left, superimposition of the ␣-rings of the 20S CP with (in magenta schematic) and without (in cyan schematic) the capping PafE(⌬155-166) ring (not shown), revealing a 3°in-plane rotation of each ␣-subunit. The PafE GQYL motif is shown as orange sticks. Right: a zoomed view of the area marked by the black box in the left panel. Note that the first modeled residue of the ␣-subunit is Pro-4 and Ser-8 in the uncapped end and PafE-capped end, respectively.

Proteasome activation by PafE
20S CP gate opening in Mtb is as follows: upon insertion of the PafE C-terminal GQYL motifs into the binding pockets between 20S CP ␣-subunits, PrcA Arg-26 and Lys-52 move outwards to form a cationbond and a salt bridge with the GQYL sequence, which triggers a 3°rotation of the ␣-subunits, leading to an outward movement of the PrcA H0 helices. This movement disorders the N-terminal gate-lining peptides preceding H0 helices as they lose contact with each other, widening the gate to allow substrates entry into the proteasome core.

A PafE dodecamer with six GQYL motifs is unable to activate Mtb 20S CPs
The consequences of the 12-fold symmetry of a PafE ring are 2-fold: it leads to a large, 40 Å-wide central channel, nearly twice the diameter of the PA26 and PA28 channels (14,15,17); and it has 12 C termini that can potentially interact with 20S CPs, nearly twice the number of C termini provided by PA26 and PA28. Additionally, most proteasome activators have hexameric symmetry, thus we first sought to ask if PafE could use only six of its 12 C-terminal GQYL motifs to stimulate degradation. To test this hypothesis, we made a linear PafE dimer ("PafE-PafE"). We cloned DNA encoding a C-terminal-truncated PafE (PafE(1-155)) upstream of a sequence encoding fulllength PafE (PafE(1-174)), with a linker sequence that encodes a flexible five-amino acid peptide (GSGNS) between the two genes ( Fig. 3A). Complexes of the purified fusion protein eluted from a Superose-6 gel filtration column at the same volume as WT PafE rings (Fig. S5). Negative stain EM showed that PafE-PafE assembled into rings with a diameter of 10 nm, identical to the WT PafE ring (Fig. 3A). These results indicated that we achieved our goal of acquiring a dodecameric PafE ring with only six GQYL-containing termini. We next examined if this variant PafE could activate Mtb 20S proteolysis by using HspR. In contrast to the robust degradation of HspR using WT PafE and 20S CPs, PafE-PafE could not degrade HspR (Fig. 3B). Although our recombinant PafE-PafE ring has six GQYL motifs, we do not know if all six can appropriately engage six acceptor sites in a 20S CP. Taken together, this result suggests that PafE needs more than six GQYL motifs to effectively activate the degradation of a native substrate by the Mtb proteasome.

ParP (Rv3213c) is a PafE-proteasome substrate
Although HspR is the only PafE-mediated proteasome substrate established so far, we previously identified several other proteins that accumulate in an Mtb pafE null mutant or in Mtb treated with a proteasome inhibitor (11). Furthermore, we also identified numerous proteins that specifically co-purify with a proteasome trap (11). Rv3213c is the only protein that was identified using all three approaches. Rv3213c is predicted to be a ParA-like protein with ATPase activity. ParA, which is conserved in almost all (if not all) bacterial species, works with a partner protein, ParB, and is involved in chromosome partitioning (19). Numerous ParA homologues are present in many bacterial genomes and do not necessarily work with a ParB-like partner protein. Mtb has a bona fide ParAB pair therefore Rv3213c is not likely to be involved in chromosome partitioning. Furthermore, Rv3213c does not have a conspicuous ParB partner, and it is also not predicted to be essential for viability (20). Because Rv3213c associates strongly with purified proteasomes from Mtb, we named it "ParP" for partitioning with the proteasome.

Proteasome activation by PafE
To test whether or not ParP is a substrate of PafE-mediated degradation, we tried to produce ParP-His 6 in Escherichia coli for in vitro degradation assays. However, ParP-His 6 was insoluble under numerous conditions tested (not shown). We tested if co-production with PafE could help solubilize ParP; remarkably, co-expression of parP-his 6 and pafE in E. coli resulted in the robust production of soluble ParP-His 6 (Fig. 4A). Importantly, the addition of 20S CPs to the co-purified proteins resulted in the robust degradation of ParP-His 6 (Fig. 4, B and  C). To demonstrate that the ParP degradation was truly mediated by PafE, we truncated PafE to remove the C-terminal 20 residues including the GQYL motif (PafE (⌬155-174)). Although ParP could still co-purify with PafE(⌬155-174) it could no longer be degraded by 20S CPs (Fig. 4, B and C).
Although we cannot rule out the possibility that ParP can be degraded in a PafE-independent manner, our data strongly suggest that ParP is the second, endogenous PafE-proteasome substrate identified in Mtb.

PafE substrates are held in a hydrophobic chamber prior to degradation
How an ATP-independent proteasomal activator targets a substrate for degradation by 20S CPs is incompletely under-stood. Conceivably, the activators could simply bind to the 20S CP and open it for substrate entry, thereby passively allowing the substrate to enter the 20S CP for degradation. However, the finding that PafE has defined substrates such as HspR and ParP suggest otherwise; it is possible that PafE plays a more active role in targeting its substrates for degradation. Given that PafE and ParP form a stable complex, we investigated how ParP associates with the PafE ring by using negative stain EM.
In averaged EM images obtained from reference-free image classification, we found a density or dot in the middle of the PafE ring that was absent in averaged images of PafE without ParP (11) (Fig. 5A). Therefore, the dots occupying the PafE channels were most likely ParP. ParP has a molecular mass of 28 kDa, potentially small enough to be accommodated inside the 40-Å central channel of a PafE ring.
In the absence of substrate, the 40-Å central chamber of a PafE ring is likely filled with water. We noticed that the interior surface of the chamber is lined with hydrophobic residues, Pro-65 and Phe-138, which form two hydrophobic belts inside the channel (13). We envisioned that ParP may be held in a PafE ring either by binding to the hydrophobic inner surface of the ring or by binding to a more flexible space formed by the 12 C-terminal, 21 amino acids that form a "beaded curtain."

Proteasome activation by PafE
Because we previously found that shortening the C termini by a few amino acids while keeping the GQYL motif intact does not compromise PafE function and actually enhances protein degradation (13), this finding argues against the possibility that ParP is held by the C termini.
We mutated the hydrophobic residues Pro-65 and Phe-138 to Gly and Glu, respectively (Fig. 5B). Both PafE mutants formed the same oligomers as WT protein based on size exclusion chromatography (Fig. S5). However, neither of the two mutant PafE alleles formed a complex with ParP (data not shown), indicating that these hydrophobic residues are required for binding to ParP. Because ParP could not be purified in the absence of WT PafE, we used HspR to assess the ability of the mutant PafE alleles to mediate substrate degradation by 20S CPs. As with ParP, the ability of PafE to mediate the proteolysis of HspR was abolished by either single mutation (Fig. 5, C-F). This result supports the hypothesis that PafE rings must interact with substrates via their structured hydrophobic channel, and not with the flexible 12 C termini, before delivering proteins into the proteasome core protease (Fig. 5G).

Discussion
Our structural and biochemical studies of the PafEproteasome system identified specific interactions between PafE and the 20S CP that cause conformational changes leading to the opening of the proteasome gate to substrates. Furthermore, our data suggest a mechanism of PafE-mediated degradation in which a disordered or partially disordered protein is selected by and accommodated within the hydrophobic central channel of a PafE ring. We found that PafE is lined with hydrophobic residues that are essential for mediating proteolysis, and these hydrophobic residues were critical for allowing a robust interaction with a newly identified PafE-proteasome substrate, ParP. Interestingly, this interaction occurred in the absence of proteasomes, suggesting the intriguing possibility that PafE

Proteasome activation by PafE
could act as a chaperone to deliver proteins to the proteasome for destruction (Fig. 6).
Our cryo-EM structure of the PafE ring with an intact activation motif in complex with a WT 20S CP revealed that the GQYL motif inserts into an acceptor pocket at the interface between two neighboring ␣-subunits. Lys-52 of the ␣-ring acceptor pocket is known to play a crucial role by forming a salt bridge with the terminal carboxylate of the PafE activation motif (15,(21)(22)(23). Importantly, our structure showed an additional residue, PrcA Arg-26, plays a role in gate activation by forming a cation-interaction with PafE Tyr-173 of the GQYL motif. These interactions cause an in-plane rigid-body rotation of each of the seven PrcA subunits. The net outcome of these changes is the increase of the gate diameter, leading to a loss of interaction and disorder among the seven gate-forming N-terminal peptides of the ␣-ring. To our knowledge, this is the first time the gate-opening mechanism of a bacterial 20S CP by an ATP-independent activator has been elucidated.
Structural studies of eukaryotic ATP-independent proteasomal activators showed varied molecular mechanisms of proteasome activation (reviewed in Ref. 24). Trypanosoma brucei PA26 inserts its activating C-terminal between the ␣-subunits of the Saccharomyces cerevisiae 20S CP and induces an outward displacement of a reverse turn of ␣-subunits to open the gate (14,21). Such a reverse turn is conserved in archaeal and eukaryotic 20S CPs but is absent from the Mtb 20S CP (5). In contrast, the monomeric S. cerevisiae Blm10 docks a single activation loop between the ␣5and ␣6-subunits of the yeast 20S CP to cause disorder and opening of the gate (25). Consistent with the different activation mechanisms, the last four amino acids of the ATP-independent activators are quite diverse; for example, they are HMVS in T. brucei PA26, GMIY in human PA28␣, FSMY in human PA28␤, and ETLY in human PA28␥, as compared with GQYL in PafE. Such variation may have evolved to confer specificity toward their respective 20S CPs. Despite the mechanistic variation, a prominent conserved feature across the domains of life is the formation of salt bridges between the C-terminal carboxylates of the activators and the conserved lysine of the ␣-subunits of 20S CPs.
Our observation that mutant PafE dodecamers with six GQYL motifs fail to activate 20S CPs suggests that for a PafE ring to engage sufficient 20S CP acceptor sites, it must have more than six GQYL motifs. The 12 activation motifs of a PafE ring are likely required to achieve the full occupancy of the acceptor pockets within an Mtb 20S CP. The term "full occupancy" remains to be defined; an argument against the Mtb 20S CP requiring the occupancy of all seven acceptor sites is made by the fact that the ATPase activator Mpa is functional with only six GQYL motifs (26,27). It is further important to note that just because our recombinant PafE-PafE ring has six GQYL motifs, it does not necessarily mean it can engage six acceptor sites in a 20S CP. Taken together, a PafE ring with 12 GQYL motifs may be better able to engage a sufficient number of 20S CP acceptor sites to stimulate degradation.
Despite extensive effort, robust proteasomal activation by Mpa with WT 20S CPs has not yet been demonstrated in vitro. This observation may not be surprising in light of our previous studies showing the Mpa GQYL motif is largely hidden inside the substrate translocation channel. Therefore, a large conformational change is likely required for Mpa to bind and activate

Proteasome activation by PafE
20S CPs. These activities may require assistance from additional, yet-to-be identified co-factor(s) (27). Thus, Mpa and PafE may have additional differences in their mechanisms for activating degradation by 20S CPs. There is also some consensus on ATP-independent and ATP-dependent proteasomal activation in terms of the roles of the 20S ␣-subunit residues, such as Lys-52 and Arg-26 in Mtb PrcA. A lysine is highly conserved at the equivalent position in the ␣-subunits in all domains of life, and interacts with C-terminal carboxylate groups of ATP-dependent activators, including PAN (28) and Rpts (23,29), as well as ATP-independent activators, including PA26 (14) and Blm10 (25). The cation-interaction formed between an activator's penultimate tyrosine and positively charged residue corresponding to Mtb PrcA Arg-26 in other proteasome ␣-subunits has also been observed in the archaeal PAN-20S CP interaction (28); however, this residue (Arg or Lys) in both archaeal and bacterial ␣-subunits, is less well conserved in eukaryotic proteasomes.
An important new finding was the discovery of ParP as a new PafE-proteasome substrate. Remarkably, we found that PafE holds ParP inside its hydrophobic chamber in the absence of 20S CPs. Altering the hydrophobicity of the PafE chamber abolished ParP binding as well as its degradation by 20S CPs. This discovery raises the possibility that the bacterial activator may be able to select and target specific folded substrates to 20S CPs or, alternatively, PafE may bind specifically to proteins like HspR or ParP when they are misfolded, a distinction we cannot yet quite make.
A PafE ring is formed of tightly packed ␣-helices and must be very rigid, which could impose a size limit on substrates. This observation is consistent with our earlier findings that disruption of pafE in Mtb does not result in accumulation of many proteins under normal culture conditions (11), and that the PafE proteome does not conspicuously overlap with the pupylome (30,31). Although we do not yet know the function of ParP, it is likely to have ATPase activity and is not predicted to be essential for bacterial viability in vitro (20). Due to its robust interaction with PafE, a possibility is that ParP competes with other potential substrates and could perhaps be a regulator of PafE-mediated degradation in vivo. This hypothesis may be supported by our observation that the co-production of HspR with PafE did not allow for their co-purification (data not shown), in contrast to what we observed for ParP and PafE.
Our studies have raised several fundamental questions that remain to be answered, including how does a substrate, either partially or totally disordered, get completely unfolded and threaded into a 20S CP for degradation in the absence of ATP? In addition, are other co-factors needed to help PafE target proteins to the proteasome? Finally, can PafE act as a protein chaperone for other substrates? Ongoing studies are working to answer these important questions.

Protein purification
All plasmids were made using PCR and sequenced by Genewiz, Inc. (Table S1). Co-expression of pafE and parP was achieved by insertion of a second ribosome-binding site and parP gene fragment after an XhoI site. A His 6 tag was linked to the C terminus of ParP to co-purify the complex. Plasmids were transformed into E. coli BL21(DE3) and expression was induced by 0.2 mM isopropyl 1-thio-␤-D-galactopyranoside at 16°C. Bacteria were collected after overnight induction and homogenized in buffer containing 20 mM Tris-HCl (pH 8.0) and 300 mM NaCl. The cell lysates were loaded into a 5-ml Ni 2ϩ -nitrilotriacetate acid-agarose column and target proteins were eluted with a linear gradient concentration of imidazole. Further purification was performed using a Superose 6 10/300 GL gel-filtration column. To produce the fusion protein PafE(1-155)-PafE(1-174), the gene fragment encoding PafE (1-155) plus a C-terminal "GSGNS" linker (containing an EcoRI restriction site) was cloned into pET24b(ϩ) (NdeI and NotI sites), followed by an insertion of the pafE gene between the EcoRI and NotI sites. The expression and purification methods for the fusion protein were the same as those for PafE alone. PafE variants, HspR, and 20S CPs were purified as described previously (5,13,32).

In vitro degradation assay
For ParP degradation assays, 15 g of purified WT 20S CPs were incubated at room temperature in reaction buffer containing 50 mM Tris-HCl (pH 8.0) and 5 mM MgCl 2 . The reaction was initiated by the addition of a 4-fold molar excess of purified PafE-ParP or PafE(⌬155-174)-ParP. Reaction products were collected every 30 min and mixed with SDS protein sample buffer to stop the reaction. The degradation of ParP was assessed by SDS-PAGE. For the HspR degradation assay, 15 g of purified WT 20S CPs was incubated with 10 g of WT or mutant PafE at room temperature in reaction buffer. The reaction was initiated by addition of a 30 g of HspR. Reaction products were collected every 30 min and mixed with SDS loading buffer to the stop reaction. All degradation assays were done at least three times with essentially the same results.

Grid preparation and data collection
For cryo-EM analysis of PafE-20S CP complexes, 3 l of purified sample of WT 20S CP complexed with PafE(⌬155-166) were applied to a C-flat 2/1.2 holey carbon grid, which had been glow-discharged for 1 min by PELCO easiGlow (Ted Pella). Grids were plunge-frozen in liquid ethane using FEI Vitrobot with a blotting time of 7 s. Data were collected using an FEI Titan Krios 300 kV transmission electron microscope equipped with a Gatan K2 Summit direct detector. Automated data collection was carried out in super-resolution mode at a dose rate of 11.5 electrons/physical pixel/s. For each movie, 30 frames were acquired in 6 s with a total accumulated exposure of 58 electrons/Å 2 . A dataset of ϳ5,000 micrographs was collected using a defocus range between 1.0 and 3.0 m.
For negative stain EM analysis of the PafE-PafE (PafE(1-155)-PafE(1-174)), ParP-PafE complexes, or PafE alone, 5 l of protein sample were applied to a glow-discharged 300-mesh copper EM grid. The grid was then washed with water and stained twice for 45 s by adding 5 l of 2% (w/v) uranyl acetate solution. For the analysis of PafE-PafE, negatively stained EM grids were imaged in the Tecnai Spirit G2 BioTWIN transmis-Proteasome activation by PafE sion electron microscope operating at 120 kV. Micrographs were recorded in a dose of 12 electrons/Å 2 at ϫ30,000 magnifications in a Gatan Orius 830 CCD camera. For the analysis of ParP-PafE complex, the grid was loaded on a JEOL JEM-2010F transmission electron microscope operating at 200 kV equipped with a Gatan UltraScan 4000 CCD camera. Micrographs were recorded in low-dose mode (15 electrons/Å 2 ) at ϫ50,000 magnification (4,096 ϫ 4,096 pixel), which corresponded to 2.12-Å/pixel sampling at the specimen level.
For cryo-EM analysis of PafE-20S CP, beam-induced motion correction and CTF estimation was done by MotionCor2 (34) and CTFFIND4 (35), respectively. Good micrographs were selected based on the extent and regularity of Thon rings. A total of 180,159 particles were automatically picked and applied to reference-free 2D classification using RELION2 (18). Classes with a clear density of 20S CPs, containing 141,275 particles, were selected for 3D classification. To make an initial reference, the EM structure of Mtb 20S CP with both ends capped with PafE(⌬155-166) in our previous work was low-pass filtered to 50 Å. The particles were subsequently classified into eight classes without imposing any symmetry. Two of the eight classes showed 20S CPs with one end capped with PafE. The two classes, containing 60,034 particles, were applied to highresolution refinement without imposing any symmetry, which resulted in a map with an average resolution of 4.2 Å according to the FSC ϭ 0.143 criterion. Three classes containing 51,091 particles showed maps with good density of PafE on both ends of 20S CPs. Subsequent refinement with the D7 symmetry was performed using those particles. A mask was then generated to mask the 20S CP for high-resolution refinement using the D7 symmetry. Finally, a map of 20S CP upon PafE binding was obtained with an average resolution of 3.4 Å according to the FSC ϭ 0.143 criterion. Local resolution was estimated for the reconstruction with Resmap (36). Maps were visualized using UCSF Chimera (37).
The Mtb 20S crystal structure (PDB code 3MI0) was rigid body fitted into the 3.4-Å cryo-EM map and then iteratively refined using real-space refinement in Phenix (38). The resulting model was visually inspected in Coot (39). Several secondary structures and side chain rotamers were manually adjusted to best fit the density. The last four residues of PafE were also manually added to the extra density between the ␣-subunits of the 20S CP. The quality of the final map was analyzed with Molprobity (40).

Accession codes
The doubly PafE-capped and the singly PafE-capped Mtb 20 CP cryo-EM 3D maps, respectively, correspond to atomic models deposited in the RCSB PDB with accession codes 6BGL and 6BGO.