Structural analysis of the dodecameric proteasome activator PafE in Mycobacterium tuberculosis

Significance

Mycobacterium tuberculosis (Mtb) has evolved a sophisticated toolkit to cope with the harsh environment inside its natural host, the human macrophage. Macrophages are immune cells that normally kill invading microbes; however, Mtb has a proteasome system that allows it to persist and cause lethal infections in animals. Although the Mtb proteasome core particle is evolutionally related to its eukaryotic counterpart, factors involved in targeting doomed proteins to the mycobacterial proteasome appear to be distinct. A prime example is the bacterial pupylation pathway, which is biochemically unrelated to the eukaryotic ubiquitylation system. Here, we describe a second example: a bacterial proteasome activator called PafE (Rv3780), which is structurally unlike any previously characterized proteasome activator in biology.

Abstract

The human pathogen Mycobacterium tuberculosis (Mtb) requires a proteasome system to cause lethal infections in mice. We recently found that proteasome accessory factor E (PafE, Rv3780) activates proteolysis by the Mtb proteasome independently of adenosine triphosphate (ATP). Moreover, PafE contributes to the heat-shock response and virulence of Mtb. Here, we show that PafE subunits formed four-helix bundles similar to those of the eukaryotic ATP-independent proteasome activator subunits of PA26 and PA28. However, unlike any other known proteasome activator, PafE formed dodecamers with 12-fold symmetry, which required a glycine-XXX-glycine-XXX-glycine motif that is not found in previously described activators. Intriguingly, the truncation of the PafE carboxyl-terminus resulted in the robust binding of PafE rings to native proteasome core particles and substantially increased proteasomal activity, suggesting that the extended carboxyl-terminus of this cofactor confers suboptimal binding to the proteasome core particle. Collectively, our data show that proteasomal activation is not limited to hexameric ATPases in bacteria.

Although the ubiquitin proteasome pathway plays essential roles in eukaryotes (reviewed in refs. 1 and 2), most bacterial species do not have proteasome systems and instead degrade proteins using ATP-dependent proteases like ClpP, Lon, and HslUV (reviewed in refs. 3 and 4). However, bacteria of the orders Actinomycetales and Nitrospirales also encode proteasomes that are structurally highly similar to eukaryotic and archaeal proteasomes (reviewed in refs. 5 and 6). Importantly, the human pathogen Mycobacterium tuberculosis (Mtb), an Actinomycete, requires proteasomal function to cause lethal infections in mice (7). Ablation of proteasomal degradation sensitizes bacteria to nitric oxide, an antimicrobial free radical made by macrophages and other cell types, and attenuates bacterial growth in mice (7–9). The potential to target persistent or latent bacteria has made the Mtb proteasome system a prioritized target for the development of antituberculosis drugs (10, 11). Indeed, Mtb-specific proteasome inhibitors have been identified that may provide a promising lead for new drugs to treat tuberculosis (12, 13).

There are numerous similarities and differences between eukaryotic and bacterial proteasomes. The 20S proteasome core particle (20S CP), which consists of two seven-membered β-rings between two seven-membered α-rings, is highly conserved structurally between prokaryotes and eukaryotes (14–16). However, the accessory factors that associate with the 20S CPs quickly diverge among the domains of life. Both bacteria and eukaryotes use a covalent small protein modification to mark substrate proteins for degradation; however, the eukaryotic ubiquitin tag is a well-folded protein whereas the Mtb Pup (prokaryotic ubiquitin-like protein) tag is intrinsically disordered (17, 18). Furthermore, degradation of ubiquitylated proteins by eukaryotic 20S CPs largely relies on a complex regulatory particle that caps one or both ends of the 20S CP and includes a heterohexameric ring of adenosine triphosphatases (ATPases) for substrate recognition and unfolding (reviewed in refs. 19 and 20). In contrast, the mycobacterial 20S CP uses a homohexameric ATPase ring called Mpa (mycobacterial proteasome ATPase) for both the recognition and unfolding of pupylated proteins (18, 21, 22).

In addition to the ATPase activators, proteolysis by eukaryotic proteasomes can also be stimulated by several ATP-independent factors, such as the 11S activators PA26 and PA28, as well as Blm10 (23–28). We and another group recently discovered that Mtb has an analogous factor encoded by Rv3780 that we call PafE (proteasome accessory factor E; also known as Bpa for bacterial proteasome activator), which stimulates the degradation of small peptides and β-casein in vitro (29, 30). Both studies also showed that a carboxyl (C)-terminal glycine-glutamine-tyrosine-leucine (GQYL) motif is essential for interacting with and activating 20S CPs, and the penultimate tyrosine residue contributes to activation similarly to tyrosines observed in the “HbYX” (hydrophobic-tyrosine-any amino acid) motif in other characterized proteasome activators (reviewed in ref. 28). Our work further showed that PafE promotes the degradation of at least one native Mtb protein substrate, heat-shock protein repressor (HspR), and that an Mtb pafE mutant is sensitive to heat shock and is attenuated for growth in mice (30). Importantly, PafE-mediated degradation does not require pupylation. Thus, there appear to be at least two independent paths for targeting proteins to the mycobacterial proteasome for degradation.

Like the eukaryotic 11S proteasome activators, PafE does not require ATP to stimulate proteolysis. However, it was unknown if PafE formed heptameric complexes like PA26 or PA28. In this work, we show that PafE monomers assume a four-helix bundle structure that is similar to that found in 11S activators, but assemble differently into an unprecedented dodecameric ring structure with 12-fold symmetry. We used isothermal titration calorimetry, cryo-electron microscopy (cryo-EM), and X-ray crystallography to analyze interactions between PafE and 20S core particles, and found that PafE binding induces a larger gate-opening change than has been described for other organisms. We also found that PafE has an extended C terminus that limits the ability of PafE to activate proteasomal degradation in vitro and in vivo.

Results

PafE Forms Dodecameric Rings with 12-Fold Symmetry.

We previously showed that a WT PafE complex forms dodecameric rings using EM and size-exclusion chromatography coupled with multiangle light scattering (SEC-MALS) (30). However, we did not know how the 12 subunits assembled to form an active complex. To determine how PafE multimerized, we solved the structure of the Mtb PafE complex. PafE is a 19-kDa protein of 174 residues and is predicted to form four α-helices with disordered regions at both the amino (N) and C termini (Fig. 1A). To facilitate crystallization, we made two mutant PafE proteins with deletions at both the N and C termini: PafE_ΔΝ14ΔC21 and PafE_ΔΝ43ΔC21 (SI Appendix, Table S1–S3). We isolated the WT and truncated recombinant Mtb proteins from Escherichia coli at high purity (Fig. 1B). We solved the PafE_ΔΝ14ΔC21 structure at 2.88 Å resolution with R_work = 19.6% and R_free = 21.5%, with experimental phases derived from selenium (Se) and platinum (Pt) anomalous signals (SI Appendix, Table S4). We found that PafE_ΔΝ14ΔC21 crystallized as a dodecameric ring with a diameter of ∼10 nm, which is the same diameter we observed for the WT PafE ring (30) (Fig. 1 C and D and SI Appendix, Figs. S1 and S2 A and B). In solution, the PafE_ΔΝ14ΔC21 proteins formed a double-dodecamer by two single rings associating via the flexible N and C termini at the exposed proximal surface (“proximal” refer to the surface that is expected to interact with 20S CP) (SI Appendix, Fig. S2 A and F). Each ring could shift and rotate with respect to the other, which likely resulted in only one ring being ordered and creating an unresolved density gap in the crystal lattice (SI Appendix, Fig. S2 A and C–E). Because WT PafE behaved and bound to 20S CPs as a single dodecameric ring (30), we presumed that the PafE_ΔΝ14ΔC21 double-dodecamer was artificial and that each of the two rings in the crystal structure represented the solution structure of a single WT dodecameric PafE ring. The quality of the electron density map of the PafE_ΔΝ14ΔC21 dodecamer was demonstrated by the clarity of most of the amino acid side chains (SI Appendix, Fig. S3). Location of anomalous scatterers of selenium and platinum confirmed that the atomic model was correct (SI Appendix, Fig. S4).

Fig. 1.

PafE forms dodecameric rings. (A) PafE contains four α-helices (cyan cylinders), as well as an additional N-terminal short H0 helix and C-terminal 21-residue peptide, including a terminal GQYL motif (red). In addition to the full-length protein, we also produced two N- and C-terminal truncation constructs PafE_ΔN14ΔC21 and PafE_ΔN43ΔC21. (B) SDS/PAGE analysis of the purified proteins described in A. (C) Crystal structure of the dodecameric PafE_ΔN14ΔC21 in top (Left) and side (Right) views. (D) Surface potential of PafE_ΔN14ΔC21. The positive and negative charges are colored blue and red, respectively.

The crystal structure revealed three unexpected features of PafE. First, in contrast to the previously known oligomeric proteasome activators that have six- or sevenfold symmetry, PafE formed a dodecamer with 12-fold symmetry (Fig. 1C). Second, PafE monomers formed a four-helix bundle structure similar to that of the eukaryotic PA26 and PA28 activator proteins, despite the absence of sequence homology between PafE and these proteins (24, 25). However, we observed that the length of the PafE α-helices on average is much shorter, and their packing is different from that of PA26 and PA28 in 11S activators (SI Appendix, Fig. S5 A and B). Third, we found that the orientation of a PafE subunit within the ring structure is different from that of PA26 or PA28 (SI Appendix, Fig. S5C): PafE helices (H) H1 and H4 face the interior channel but in PA26 and PA28, H3 and H4 face inside. The central channel of the PafE ring is 40 Å wide, which is larger than the 20–30 Å channel observed among the helical regions of the PA26 and PA28 activators (23–25), and significantly wider than the elongated 9 Å × 18 Å channel in Blm10 (27). Interestingly, the distal as well as the side surfaces of PafE are highly negatively charged (Fig. 1D). The central channel of the PafE ring is lined largely by several hydrophobic residues [Leucine (Leu)58, Alanine (Ala)64, Proline (Pro)65, Ala113, Leu137, Phenylalanine (Phe)138], a feature compatible with its putative function for threading unfolded polypeptides into a 20S CP (29, 30). These features are unique to PafE, as the central channels as well as the outer surfaces are charged in PA26 and PA28 heptamers (23–25).

An N-Terminal Helix and a Double-GXXXG Motif Are Essential for PafE Ring Formation.

We next compared the crystal structures of the two truncated PafE proteins, which multimerized very differently from each other and native PafE (Fig. 1C and SI Appendix, Figs. S2 and S6, and Table S4). Comparison of the crystal structure of PafE_ΔΝ14ΔC21 with that of PafE_ΔΝ43ΔC21 revealed the importance of the short N-terminal H0 helix (residues 36–43) in mediating subunit interactions. The shorter of the two proteins, PafE_ΔΝ43ΔC21, lacks the first helix H0 and crystallized as tetramers, each consisting of a dimer of dimers (SI Appendix, Fig. S6). In the PafE_ΔΝ14ΔC21 dodecamer ring, H0 was located between H2, H3, and H4. In contrast, deletion of H0 resulted in an intersubunit H4–H4 interaction, creating the artificial dimer structure of PafE_ΔΝ43ΔC21. Thus, removing H0 enabled tight packing of H4 helices between two subunits (SI Appendix, Fig. S6B). PA26 and PA28 subunits lack a corresponding short H0 helix, such that the equivalent H4 is able to tightly pack against H3 (SI Appendix, Fig. S5B). Thus, the PafE N-terminal H0 is critical for the formation of the dodecameric ring structure.

We observed that a PafE ring consists of two concentric shells, each with 24 α-helices; helices H1 and H4 line the inner surface and helices H2 and H3 make up the outer shell (Fig. 2A). Measured from the center of the α-helices, the inner shell is 60.7 Å in diameter and the outer ring is 84.2 Å in diameter. This translates to ∼10.8 Å between H2 and H3 for both intra- and intersubunit helix packing in the outer shell. In the inner shell, the distance between H1 and H4 is 10.5 Å and 6.5 Å for intra- and intersubunit helical packing, respectively. The unusually short distance between helices H1 and H4 was made possible by the presence of a Gly-XXX-Gly-XXX-Gly (GXXXGXXXG) motif in H4 (Fig. 2B, Left). The GXXXG motif is known to be particularly important for tight α-helix packing in membrane proteins (31). We found this motif was essential for maintaining the dodecameric ring structure: mutating any of the three glycines disrupted PafE ring formation (Fig. 2B, Right). The three glycine-to-isoleucine PafE mutant full-length proteins eluted similarly to that of the PafE_ΔN43ΔC21 tetramer. In addition to the tight helix packing, the subunit interface is further stabilized by a hydrogen bond between Gln56 and Glu127 of the inner shell H1 and H4 helices, and two salt bridges between the outer shell H2 and H3 helices formed between Glu84-arginine (Arg)99 and Arg49-Glu96 (Fig. 2C, Left). Breaking the two salt bridges by mutating Glu96 or Arg99 to leucine resulted in dissociation of the ring structure into oligomers with a molecular mass slightly larger than that of a tetramer (Fig. 2C, Right). However, disrupting the hydrogen bond in the Gln56Ile mutant retained the ring structure, although the ring was much less stable. Therefore, the electrostatic H2–H3 interactions in the outer shell of the PafE ring complement the inner shell H4–H1 interactions that are largely mediated by the two GXXXG motifs in H4.

Fig. 2.

Subunit interface is essential for the integrity of the PafE dodecamer structure. (A) Two neighboring PafE monomers in the dodecameric ring viewed from the top distal surface in rainbow cartoon view. Two concentric arcs are part of the inner and outer shells of the PafE ring. The dashed vertical line marks the subunit interface. Note the close intersubunit H1/H4 packing. (B, Left) Viewed from inside the channel, showing the GXXXGXXXG motif of helix H4 that enables H1/H4 tight packing. The three Gly are shown as spheres. Small or flexible side chains in helix H1 important for the tight packing are indicated. Gln56 and Glu127 form H-bonds. (Right) SEC profiles of WT PafE (blue) in comparison with that of three Gly single-mutation PafE proteins (yellow, PafE_G124I; green, PafE_G128I; and purple, PafE_G132I). (C, Left) Viewed from the surface proximal to 20S CP, residues involved in intersubunit interaction in the outer shell are shown as sticks. Arg99 and Glu84 form two H-bonds, and Arg49 and Glu96 form two H-bonds. (Right) Gel-filtration profiles showed mutagenesis of PafE residues involved in H-bonding affect or abolish native ring formation. Right panels in B and C used the same HiLoad 16/60 Superdex 200-pg column. Size standards are marked between the two chromatographic panels.

PafE Enlarges the Substrate Entrance of Mtb CPs.

Unlike most known eukaryotic proteasome-activator complexes, Mtb 20S CPs do not interact robustly with the cofactors Mpa or PafE in vitro (30, 32), thus making the detailed analysis of bacterial activator–CP interactions challenging. Previously, we circumvented this problem by using mutant forms of the Mtb 20S CP that bind more robustly than WT 20S CPs (20S_WT) to both Mpa and PafE. To begin to determine how PafE activates Mtb CPs, we quantified the binding of PafE with different 20S CPs by gel filtration and isothermal titration calorimetry (ITC). We tested the ability of PafE to interact with a catalytically inactive 20S CP in which the active site threonine was mutated to alanine (20S_T1A). Unlike 20S_WT CPs, the 20S_T1A CP can robustly copurify with PafE and Mpa in vitro (30). In addition, we used an “open gate” 20S CP (20S_OG) mutant that lacks amino acids 2–9 from the α-subunit PrcA (33); these CPs previously allowed us to visualize interactions with PafE and Mpa by EM (30, 32).

Based on SEC profiling (SI Appendix, Fig. S7A), full-length PafE bound tightly to both 20S_T1A and 20S_OG CPs but not to 20S_WT CPs. However, we observed by EM that ∼2.3% of the 20S_WT CPs interacted coaxially with WT PafE rings (SI Appendix, Fig. S7B). In ITC experiments, addition of PafE to 20S_WT, 20S_OG, or 20S_T1A CPs resulted in an overall evolution of heat, indicated by the negative peaks in the raw heat plots (Fig. 3). Nonlinear least-square fitting of the data showed that PafE rings bound to 20S_WT, 20S_OG, and 20S_T1A CPs with binding constants of 9.01, 0.25, and 1.54 μM, respectively, and with greater than one-to-one stoichiometry, which indicated that in solution some 20S CPs were capped by PafE rings at both ends (Table 1).

Fig. 3.

In vitro binding of PafE to Mtb 20S CPs as measured by isothermal titration calorimetry. The Upper panels show typical raw calorimetry data when PafE was injected at 25 °C into 20S CPs: (A) 480 μM PafE into 33 μM Mtb 20S_WT, and (B and C) 85 μM PafE into 6.7 μM Mtb 20S_OG and 20S_T1A, respectively. Higher protein concentrations were needed to measure the weak binding between PafE and 20S_WT CPs. In the Lower panels, each datapoint represents the reaction heat normalized to the amount of PafE injected and is corrected for the heat of dilution. The solid line is the nonlinear least-square fitting of the data. n = binding sites; K_d, dissociation constant.

Table 1.

ITC binding parameters of various PafE constructs to Mtb 20S CPs

Titrant	n	Kd (μM)	ΔH ×103 (kcal⋅mol−1)	ΔS (kcal⋅mol−1⋅K−1)
PafE to 20SWT	1.92	9.01 (1.23)	−1.72	17.3
PafE to 20SOG	1.29	0.25 (0.01)	−40.80	−107.0
PafE to 20ST1A	1.04	1.54 (0.06)	−25.70	−59.7
PafEΔ155–166 to 20SWT	1.77	4.61 (0.36)	−6.09	4.0
PafEΔ155–166 to 20SOG	1.99	0.05 (0.01)	−33.36	−78.5
PafEΔ155–166 to 20ST1A	1.67	0.89 (0.05)	−29.50	−71.3

n refers to number of PafE dodecamer rings bound per 20S CP.

We studied by cryo-EM and single-particle 3D reconstruction the stable and active PafE:20S_OG CP complex. Raw electron micrographs and the averaged particles showed that the PafE:20S_OG CP complexes were not homogeneous after gel filtration. We observed three major species in a single gel-filtration peak sample: 20S_OG CPs only; 20S_OG CPs with one PafE ring (20S_OG:PafE), and 20S_OG CPs with two PafE rings (PafE:20S_OG:PafE) (SI Appendix, Figs. S7 and S8). Reference-free 2D classification and unsupervised 3D classification confirmed the existence of all three complexes (SI Appendix, Fig. S8 B and C). After 3D classification, the original data were divided into three major classes, each of which was refined independently to obtain the final 3D density maps (Fig. 4A and SI Appendix, Fig. S8C). The reconstructed cryo-EM 3D density maps at ∼13 Å resolution agreed well with the Mtb 20S CP crystal structure (Fig. 4B) (15).

Fig. 4.

Cryo-EM of full-length PafE:20S_OG complexes. (A) Surface-rendered side views of the three complexes that coexisted in solution: Mtb 20S_OG alone without PafE (Left, light blue); 20S_OG CPs capped at both ends by PafE dodecamer rings (Center, gray); and 20S_OG CPs capped at one end by a PafE dodecamer ring (Right, brown). The 3D maps were low-pass–filtered to ∼13 Å. The PafE density is smaller in the singly capped complex, which may reflect increased flexibility. (B) Surface-rendered and semitransparent 3D map of the Mtb 20S_OG in side (Left) and top (Right) views docked with the crystal structure of Mtb 20S proteasome (PDB ID code 3MI0). (C) Superimposition of the 3D maps of Mtb 20S_OG (light blue) and 20S_OG CPs capped at one end by the PafE ring (brown). The substrate entrance in the α-ring of the proteasome was enlarged by ∼30% (6 Å) at the PafE-capped end (Left), whereas the entrance size was unchanged at the uncapped end (Right). (D) Superimposition of the 3D maps of the Mtb 20S_OG (light blue) and Mtb 20S_OG capped at both ends by PafE rings (gray). The substrate entrance gates in the two α-rings of the open-gate proteasome were both enlarged by ∼35% or 7 Å. (Right) A section of the 3D difference map between 20S_OG with and without PafE binding (magenta) superimposed on the 20S_OG map (light blue) in top views. The sectioning plane is marked by the dashed line in the left panel. The difference density peaks from the PafE C termini are located between α-subunits of the proteasome.

Next, we determined if binding of PafE to 20S_OG CPs caused any structural changes in the CP. We overlapped the 20S_OG CP EM density with that of 20S_OG CP:PafE (Fig. 4C) or PafE:20S_OG CP:PafE (Fig. 4D), and found that the gate of the 20S_OG CP was enlarged by an additional 6–7 Å in the PafE-capped side of the CP, whereas the gate size was unchanged at the CP end without PafE (Fig. 4 C and D). A difference map showed extra density between the α-subunits of the 20S_OG CP (Fig. 4D, Right), which is similar to the region in which activators bind to 20S CPs in eukaryotes (reviewed in ref. 28).

A C-Terminal Linker Region in PafE Modulates Activation of 20S CPs.

Thus far, our results suggested that to obtain a robust interaction of an Mtb proteasome activator with a 20S CP, we needed to alter the gate or catalytic activity of the CP. We then wondered if there were residues in the activator that could also be altered to increase its interaction with 20S_WT CPs. We noticed that PafE complexes observed by EM were not always flush with 20S_WT CPs (SI Appendix, Fig. S8), resulting in a reduced resolution of the PafE ring region in the reconstructed 3D maps. Despite the limited resolution, the crystal structure of the processed PafE_ΔΝ14ΔC21 dodecamer fit well in the PafE EM density (Fig. 5A, Left) but the PafE ring appeared to float above the 20S_WT CP with a gap of ∼20 Å. In contrast, this gap is absent from the crystal structure of the Trypanosoma brucei PA26 in complex with the Saccharomyces cerevisiae 20S CP (Fig. 5A, Right) (24). PafE conspicuously featured a 21-residue unstructured C terminus that is twice the length of those of PA26 and PA28 (Fig. 5B). We hypothesized that this extension served as a “spacer” region that created the 20 Å gap observed between the CP and PafE ring, possibly limiting activation of the 20S CP.

Fig. 5.

Shortening PafE C-terminal (CT) linker increases binding affinity to 20S CPs. (A) Zoomed side view of the 3D map of PafE:20S_OG docked with the crystal structures of PafE (magenta) and Mtb 20S (PDB 3MI0, yellow) (Left) or docked with the PA26-proteasome complex structure (PDB ID code 1FNT, purple ribbon) (Right). PA26 is ∼20 Å closer to the proteasome than PafE, indicating an extended linker region between the helical domain and the CT GQYL motif in PafE (magenta dashed lines, Left). (B) PafE has a four-helix bundle fold similar to that of PA26 and PA28 subunits. The PafE CT activation peptide is double the length of that of PA26 and PA28. (C) Schematic diagram of the PafE constructs with different C-terminal domains. (D) SDS/PAGE of gel filtration peaks of 20S-PafE complexes showing the binding of PafE CT peptide truncations to 20S_T1A CPs. (E) Electron micrograph of PafE (Left) or PafE_Δ155–166 (Right) incubated with Mtb 20S_OG CPs. Nearly 90% of the 20S_OG particles were doubly capped by the PafE_Δ155–166 rings. (F) Zoomed side views of 3D maps of 20S CPs (PDB ID code 3MI0, yellow ribbon) docked with WT and PafE truncation mutants (magenta ribbon).

We produced PafE with the C-terminal region shortened by five to 15 residues while retaining the terminal GQYL motif that is essential for gate opening (Fig. 5C) (29, 30). All of the truncated PafE alleles formed rings and retained their ability to interact with 20S_T1A CPs as shown by SDS/PAGE of the gel-filtration peaks (Fig. 5D and SI Appendix, Fig. S9). We chose to use PafE_Δ155–166 for further characterization by EM, as larger deletions began to disrupt binding to the 20S CP. Surprisingly, we found that virtually all 20S_OG CPs (96%) were capped by PafE_Δ155–166 rings, with ∼90% of the 20S_OG CPs doubly capped (Fig. 5E). In comparison, only 51% of the 20S_OG CPs were capped by WT PafE, with 37% of them uncapped. We subsequently determined cryo-EM 3D structures of PafE_Δ155–166 in complex with 20S_OG or 20S_WT CPs, respectively, and found that truncation of the PafE C terminus brought the dodecameric PafE ring ∼10 Å closer to the 20S CPs (Fig. 5F), indicating that the extended C terminus of WT PafE is responsible for the large gap between the activator ring and 20S CP.

ITC experiments showed that PafE_Δ155–166 had a markedly increased proteasome-binding affinity, by factors of 5- and 1.7-fold for 20S_OG and 20S_T1A CPs, respectively, compared with full-length PafE (Table 1 and SI Appendix, Fig. S10). Most importantly, PafE_Δ155–166 had a significant affinity for 20S_WT CPs, with a K_d of 4.6 μM. In agreement with the increased affinity, the binding stoichiometry of PafE_Δ155–166 to 20S_OG, 20S_T1A, and 20S_WT CPs were well above 1:1, indicating that many of these CPs had PafE_Δ155–166 rings capping at both ends. We next asked how shortening of the activation peptide affected the gate opening activity of PafE. Consistent with the increased binding affinity to the CPs, PafE_Δ155–166 rings stimulated proteolysis of a peptide reporter by 20S_WT CPs by over an order-of-magnitude more than the WT PafE rings (Fig. 6). Thus, the 21-residue C-terminal tail of PafE weakens both its binding to the 20S CP and its in vitro activation of proteasomal peptidase activity.

Fig. 6.

PafE mutants with shortened C termini have dramatically increased 20S activation capacity. Peptide degradation assays using PafE truncation mutants. **P < 0.005; ***P < 0.0005; ****P < 0.0001.

The discovery of a pafE mutant allele that dramatically enhanced the activation of Mtb 20S CPs provided us a unique opportunity to examine how this ATP-independent activator opens the gate of a 20S_WT CP. Electron micrographs showed that PafE_Δ155–166 formed tight complexes with both 20S_WT and 20S_OG CPs (Fig. 7A and SI Appendix, Fig. S11). The PafE_Δ155–166 rings opened the gates of both 20S_OG and 20S_WT CPs to 29 and 27 Å wide, respectively, which was similar to what we observed with WT PafE rings interacting with 20S_OG CPs (Fig. 4). Notably, the gate of 20S_WT CPs was opened only at the PafE_Δ155–166-capped end, whereas the gate remained fully closed at the end without a PafE_Δ155–166 ring (Fig. 7B, Right).

Fig. 7.

Cryo-EM of Mtb PafE _Δ155–166:20S_WT complexes. (A) Surface-rendered side views of the three complexes that coexisted in solution: Mtb 20S CP capped at both ends by PafE_Δ155–166 dodecamer rings (Left, light green); Mtb 20S CP alone without PafE_Δ155–166 (Center, gray); and 20S CP capped at one end by a PafE_Δ155–166 dodecamer ring (Right, brown). The 3D maps were low-pass–filtered to ∼12 Å. (B) End-on views of α-rings of proteasome complexes. The substrate entrance in the α-ring of 20S CP was open (∼27 Å in diameter) at the PafE-capped end (Left and Upper Right), whereas the entrance was closed at the uncapped end (Center and Lower Right). (C) Superimposition of the 3D maps of the Mtb 20S CP (gray) and Mtb 20S CP capped at one end by PafE_Δ155–166 rings (brown, Left). (Right) A section of the 3D difference map between 20S with and without PafE_Δ155–166 binding (magenta) superimposed on the 20S CP map (gray) in top views. The dashed line marks the sectioning plane in the Left panel. The difference density peaks from the PafE C termini are located between α-subunits of the proteasome. (D, Left) The PA26 C-terminal motif (PDB ID code 1FNT, magenta spheres), modeled in the predicted Mtb 20S CP activation pocket between two α-subunits (PDB ID code 3MI0, green and gray surfaces, respectively). The difference density, as shown in C, is rendered semitransparent (magenta) for clarity. The amino acid, shown as red spheres, at the bottom of the pocket marked by an arrow is Lys52 in the right α-subunit (gray surface). (Right) PafE-activated peptide degradation activities of Mtb 20S CPs with and without the α-subunit PrcA_K52A mutation.

The difference map between PafE_Δ155–166:20S_WT CP and WT 20S_WT CP alone showed extra density located between neighboring α-subunits of the 20S_WT CP, similar to what we observed with WT PafE binding to 20S_OG CPs (compare Fig. 4D to Fig. 7C). This density could have been to the result of a combined effect of the PafE C-terminal tail and conformational changes in the CP, or because of the PafE C termini alone. The density is at a location where PA26 and PA28 CT activating peptides bind to the archaeal or eukaryotic 20S CP (Fig. 7D, Left) (28). We modeled the PA26 activating peptide in the corresponding pocket between neighboring Mtb 20S α-subunits, and found the modeled peptide overlapped with the difference density map. Because each PafE ring has 12 C-terminal tails, the seven binding pockets at the end of a 20S CP have the potential to be fully occupied by PafE C termini.

A conserved lysine found in the α-subunits of eukaryotic proteasomes is essential for binding by the PA26 and PA28 activators. The C-terminal carboxylates of the activators form salt bridges with the α-subunit lysines to facilitate proteasomal activation (reviewed in ref. 28). We mutated the equivalent lysine (Lys, K52) to alanine in the Mtb 20S α-subunit PrcA and found that the mutant 20S CP (20S_K52A CP) could not be activated by PafE to degrade a peptide substrate (Fig. 7D, Right). Taken together, these data suggest that PafE activates the Mtb 20S CP by binding within pockets between the α-subunits that are similar to those found in the archaeal and eukaryotic 20S CPs.

Truncated PafE Is More Active than Full-Length PafE in Mtb.

We next tested if the C-terminally truncated PafE variants had increased activity in vivo. To this end, we took advantage of our previous observation that complementing a pafE deletion-disruption mutant with a WT pafE allele only partially restored PafE protein levels and thus only partially complemented the growth defect and heat-shock sensitivity phenotypes of this mutant (30). We wondered if complementation of the pafE mutant with a hyperactive pafE allele identified in this study could fully reverse the growth defects of this mutant. We introduced a single copy of either of two truncated pafE alleles (pafE_Δ155–159 and pafE_Δ155-164) expressed from the native pafE promoter into the chromosome of an Mtb pafE-null mutant. Immunoblotting demonstrated that although the truncated PafE variants were produced at levels below the endogenous amounts of PafE in the parental strain (Fig. 8A), both of the truncated pafE alleles, but not WT pafE, fully restored growth to normal levels (Fig. 8B). Moreover, complementation of a pafE-null mutant Mtb strain with the pafE_Δ155–164 allele, and not the WT pafE allele, restored heat-shock resistance to normal levels (Fig. 8C). Taken together, these data show that the residues preceding the GQYL motif are inhibitory, and that deletion of these residues enhances PafE activity in vivo.

Fig. 8.

C-terminally truncated PafE variants show increased activity in vivo. (A) Complementation of an Mtb pafE mutant partially restores PafE protein levels. A single copy of WT or truncated pafE was introduced into the chromosome. Total cell lysates were prepared from each strain, separated by 13% (wt/vol) SDS/PAGE, and analyzed by immunoblotting using antibodies raised against PafE. (B) Truncated PafE variants complement the growth defect of a pafE mutant. The indicated strains were diluted in triplicate to OD₅₈₀ = 0.025, and OD₅₈₀ was measured at the indicated time points. (C) A truncated PafE variant fully complements the heat-shock resistance of a pafE mutant. The indicated strains were diluted in triplicate to OD₅₈₀ = 0.08, incubated at 45 °C for 24 h, and inoculated onto Middlebrook 7H11 agar to enumerate surviving bacteria. Statistical analysis was done by nonparametric Student’s t test. For the pafE mutant and complemented strain, an asterisk indicates these strains were significantly more sensitive to heat shock than the parental strain with a P < 0.05, but the strain complemented with WT pafE is significantly improved compared with the mutant (P = 0.0183).

Discussion

To better understand the mechanism of ATP-independent proteasomal degradation in bacteria, we solved the crystal structure of Mtb PafE. Although PafE folds into a four-helix bundle like the eukaryotic 11S activator subunits PA26 and PA28, we found that PafE assembles into rings with 12-fold symmetry. The formation of this dodecameric ring structure required a GXXXGXXXG motif that is similar to a motif involved in the tight packing of membrane proteins, but has not been previously observed in proteasomal activators from any domain of life. We also demonstrated that PafE widens the substrate entrance of Mtb 20S CPs. We also determined that PafE has an extended C terminus that prevents optimal proteasomal activation; short deletions in the C terminus of PafE resulted in stronger interactions with 20S CPs, more robust proteasomal peptide degradation in vitro, and improved complementation of an Mtb pafE-null mutant.

All known ATP-dependent proteasomal activators, such as the archaeal PAN (proteasome activating nucleotidase), the eukaryotic Rpts (regulatory particle ATPases) 1–6, bacterial ARC (ATPase forming ring-shaped complexes), and Mpa form hexamers without exception (2, 28, 34–37). Previously characterized ATP-independent proteasomal activators are generally heptameric, with the exception of monomeric Blm10 and tetrameric PbaB (38). The elucidation of the structure of PafE, the only known ATP-independent activator in bacteria, further supports the notion that proteasome activators can be found with different symmetries. Because the 20S CP is composed of heptameric rings, if all seven potential PafE binding sites are occupied, five additional PafE C termini would remain unbound. It is possible that the presence of 12 available binding sites in PafE helps to increase avidity for what is otherwise a weak binding event; alternatively, these “free” C termini may contribute to other interactions important for either binding or degradation.

Our current and previous studies highlight the curious observation that bacterial proteasome cofactors interact poorly with 20S CPs. The eukaryotic 19S RP and 11S activators are capable of remaining bound to the 20S CP while undergoing relatively stringent purification processes; in contrast, attempts to copurify Mpa or PafE with 20S CPs requires the use of modified 20S CPs that enhance binding. The finding that residues near the C terminus of PafE inhibit its interaction with 20S CPs suggests that bacteria use suboptimal binding to regulate proteolysis. Perhaps the inhibitory residues at the C terminus of PafE allow this activator to disengage more easily from 20S CPs so that degradation products can be released. Alternatively, the long C termini of a PafE ring may prevent Mpa from competing for access to 20S CPs under certain growth conditions. Additional studies will be needed to understand the biological significance of this inhibitory region.

Why does PafE bind to mutant 20S CPs so much better than to the WT CPs? Our previous crystallographic studies of the Mtb 20S proteasome may provide some clues (39). Compared with the more rigid, symmetrical 20S_WT CP, the Mtb 20S_OG CP α-rings are more flexible, and are able to rotate by as much as 3.3° and move radially by 1.4 Å to 2.2 Å. In the case of the Mtb 20S_T1A CP structure, the seven α-subunits move outwards radially by 1.5 Å. Conceivably, the perturbation of the α-ring in the 20S_T1A and 20S_OG CPs may facilitate the insertion of the PafE GQYL sequence, resulting in a higher occupancy of the seven activation sites in an α-subunit ring of the 20S CP.

Gate opening in the archaeal and eukaryotic 20S CPs involves reconfiguration of an N-terminal reverse turn that is absent from Mtb 20S CPs (39); therefore, the gate-opening mechanism must be different in the mycobacterial system. Eukaryotic activators appear to open 20S CP gates by inserting their C termini into pockets between CP α-subunits, thereby stabilizing the shifted conformation of the reverse turn at the N-terminal regions of the α-subunits (24, 27). We have shown that PafE inserts into a location between two α-subunits, similar to that of the eukaryotic activators, and either opens the closed gate of 20S_WT CPs by 27 Å, or widens the 20-Å gate of 20S_OG CPs by nearly 35% (Figs. 4 and 7). Because the density change upon opening of the Mtb gate is substantial, we speculate that the gate opening of Mtb 20S CPs is not limited to the movement of the N termini, but may also involve a small rotation of the α-subunits as observed for the binding of the Methanococcus jannaschii PAN C termini to Thermoplasma acidophilum 20S CPs (40). The current resolution of our cryo-EM reconstruction is insufficient to determine if a similar movement occurs for the Mtb proteasome.

Our analysis of PafE has provided, to our knowledge, the first structural information for a dodecameric proteasome activator and raises questions about why bacteria use proteasomal activators that do not bind to 20S CPs as well as their eukaryotic counterparts. Moreover, the discovery of hyperactive PafE truncation mutants will be extremely useful for the exploration of the structural details of activator–20S CP interactions. Importantly, these studies will continue to enhance our knowledge of Mtb physiology, and may provide important details toward understanding the structural constraints of proteasome activation in general.

Materials and Methods

Purification of PafE.

Gene fragments encoding two truncated PafE proteins (PafE_ΔN14ΔC21 and PafE_ΔN43ΔC21) were amplified with C-terminal His₆-tags and cloned into pET24b(+) vector (Novagen) at the NdeI and NotI restriction sites (SI Appendix, Table S1). For Mtb proteasome binding, activation, and ITC assays, the full-length and internal deletion mutant pafE alleles were cloned into pET24b(+) with a N-terminal His₆-tag followed by a thrombin cleavage site. The pafE mutants were constructed using the Stratagene QuikChange Mutagenesis kit (SI Appendix, Table S2 and S3). Plasmids were transformed into E. coli BL21(DE3) after sequencing to confirm the veracity of the cloned DNA. Bacteria were cultured in LB medium at 37 °C and gene expression was induced by isopropyl β-d-1-thiogalactopyranoside (IPTG) at a final concentration of 0.5 mM. The cell lysates were loaded to 5 mL Ni²⁺-nitrilotriacetate acid (Ni-NTA) agarose column and target proteins were eluted with a linear gradient concentration of imidazole. Fractions containing target proteins were pooled, concentrated, and further purified either in a HiLoad 16/60 Superdex 200 pg or a Superose 6 10/300 GL gel-filtration column. Finally, PafE_ΔN14ΔC21 was concentrated to 20 mg/mL in buffer containing 10 mM Tris⋅HCl (pH 8.0) and 50 mM NaCl, whereas PafE_ΔN43ΔC21 was in 10 mM Hepes (pH 7.4) and 50 mM NaCl. The purity of protein was examined by SDS/PAGE. Protein concentration was determined using a Nanodrop ND-1000 Spectrophotometer (A₂₈₀) or by Bradford assay. Selenomethionine-substituted proteins were prepared as for the native proteins, except that the bacteria were grown in M9 medium.

Crystallization, Data Collection, and Structure Determination.

Crystallization was performed using the sitting-drop vapor-diffusion method at 20 °C. For dodecameric PafE_ΔN14ΔC21, crystals appeared after 2 d from drops consisting of 1 µL of protein solution and 1 µL of reservoir solution containing 0.2 M ammonium acetate, 0.1 M sodium citrate pH 5.9, and 29% (vol/vol) 2-methyl-2,4-pentanediol. Selenomethionine-incorporated crystals were obtained under the same conditions. Platinum (Pt) derivatives were prepared by soaking native crystals for 30 min in the mother liquor containing 5 mg/mL K₂PtCl₄. Selenium (Se) and Pt single-wavelength anomalous dispersion (SAD) and native diffraction data were collected from flash-cooled crystals at 100 K at beamline 31-ID-D, Advanced Photon Source, Argonne National Laboratory, and processed using iMOSFLM (41) and SCALA (42). The space group and data quality were determined and evaluated using POINTLESS and phenix.xtriage. The Se and Pt positions were determined with the program phenix.hyss. The identified selenium and Pt²⁺ sites were then refined, and initial phases were calculated in the program phenix.phaser. An initial structure model was built using phenix.autobuild. The high-resolution Se SAD dataset was used for structure refinement. We used Coot, phenix.refine, and CCP4 in subsequent iterative model building and refinement until R_work = 19.6% and R_free = 21.5% were obtained (SI Appendix, Table S4).

We found that two dodecameric rings stacked head-to-head in the crystal lattice in the c direction, forming a double-dodecameric ring. In this structure, the rigid and well-ordered surfaces of each of the two rings interact, whereas the disordered N and the C termini face outward. This arrangement made the next interface in c direction partially disordered because two flexible surfaces contact each other. We found that between two layers of well-ordered doubled dodecamer densities, there was a space with very weak densities, indicating the presence of an intervening layer of disordered double dodecamers (SI Appendix, Fig. S1). Flexible intervening molecular layers in a crystalline lattice are not unprecedented as this phenomenon was also observed in the crystal lattice of the calcium channel YetJ (PDB ID code 4PGW) (43) and outer membrane porin OmpG (PDB ID code 2IWV) (44). Furthermore, the high variability of the c-axis of unit cell by as much as 5–10% in different PafE_ΔN14ΔC21 crystals is another indication that crystallographic packing in the c direction contained flexible elements (SI Appendix, Table S5). To verify the structural solution based on Se anomalous signal, we also collected a diffraction dataset with Pt²⁺ anomalous signal. Colocalization of the two different anomalous scatterers at the Methione sites in the electron density map (Se and Pt) demonstrated that our atomic model was correct (SI Appendix, Fig. S3).

For tetrameric PafE_ΔN43ΔC21, 1 µL of protein solution was mixed with 1 µL of reservoir solution containing 0.1 M Tris⋅HCl pH 8.3 and 2 M ammonium sulfate. The crystals were transferred to cryoprotectant containing 1.5 M lithium sulfate before flash-freezing in liquid nitrogen. Selenium SAD and native diffraction data were collected at beamline X25, National Synchrotron Light Source, Brookhaven National Laboratory, and processed using HKL2000 (45). The initial SAD phase of PafE_ΔN43ΔC21 was calculated in PHENIX (46). Partial structures were built automatically in PHENIX and subsequently improved manually using Coot (47). Phases were improved by molecular replacement using the initial model and the higher-resolution native data. The structure was iteratively improved by refinement in PHENIX and manual building in Coot. Further model refinement was carried out using Refmac5 of the CCP4 suite (42) and PHENIX until R_work = 25.8% and R_free = 28.2% were obtained.

In Vitro Binding of Purified PafE Rings to Mtb 20S CPs.

The full-length and internal deletion mutant PafE proteins were purified as described above. WT 20S, 20S_OG, and 20S_T1A CPs were purified as described previously (39). In 20S_OG CPs, the N-terminal eight residues of the α-subunit were deleted to produce a CP with a constitutively open gate. In 20S_T1A CP, the active site Thr-1 of the β-subunit was replaced by Ala to abolish protease activity. Purified 20S_OG or 20S_T1A CPs were incubated with various PafE preparations (full-length or internal-deletion mutants) at molar ratios of ∼1:3 at 4 °C for 3 h, and the mixture was further purified in a Superose 6 10/300 gel-filtration column in a buffer containing 20 mM Tris (pH 8.0) and 150 mM NaCl.

ITC.

For all ITC, binding of PafE rings to 20S CPs was analyzed in 20 mM Tris (pH 8.0), 150 mM NaCl at 25 °C using a VP-ITC microcalorimeter (MicroCal). We titrated 480 μM full-length PafE into 33 μM 20_WT CP, and 85 μM full-length PafE into 6.7 μM 20S_OG and 20S_T1A, respectively, and titrated 85 μM PafE_Δ155–166 into 6.7 μM 20S_T1A, 95 μM PafE_Δ155–166 into 4.5 μM 20S_WT and 5 μM 20S_OG, respectively. We fitted the titration curves in program Origin 7.0.

Peptide Degradation Assay.

For the peptide degradation assay, 500 ng of purified Mtb 20S CP was used per reaction, and triplicate reactions were set up for each condition. 20S CPs were incubated at room temperature in activity assay reaction buffer (50 mM Tris pH 8.0, 5 mM MgCl₂) containing 20 µM of the fluorogenic nonapeptide LF-2 (30). Where indicated, a 10-fold molar excess of purified PafE dodecamers was added; 20S CP and PafE:20S CP complexes were preincubated at 37 °C for 30 min to promote complex formation and then allowed to return to room temperature before addition of substrate. Peptide degradation was assessed by measuring the change in fluorescence over time (λ_ex = 340 nm, λ_em = 405 nm), and reaction rates were calculated by determining relative fluorescence units generated per minute.

Negative-Staining EM and Initial Model.

We used negative-stain EM to characterize PafE_ΔN14ΔC21 and various Mtb PafE:20S CP complexes. For carbon-coated EM grid preparation, we first evaporated a layer of carbon onto a piece of freshly cleaved mica in an Edwards vacuum evaporator (<10⁻⁵ Torr). We floated the carbon film off the mica on the surface of deionized water and then deposited the film onto an array of 300-mesh copper grids. The dried and carbon-coated grids were glow discharged in a 0.39-mbar air atmosphere for 1 min using PELCO easiGlow (Ted Pella). Next, 4–5 μL of sample were pipetted onto a freshly treated EM grid and incubated for 1 min. Excess solution was removed by blotting the edge of the grids with filter paper, and the grids were then washed with a droplet of water, followed by two rounds of staining for 30 s with 5 μL 2% (wt/vol) uranyl acetate solution. Negatively stained EM grids were imaged in a JEOL JEM-2010F transmission electron microscope (JOEL) operating at 200 kV. Micrographs were recorded in a low-dose mode (15 e⁻/Ų) at 50,000× microscope magnification in a Gatan UltraScan 4000 CCD camera (4,096 × 4,096 pixels), which corresponded to 2.12 Å per pixel sampling at the specimen level.

For particle selection and image processing, we used EMAN and EMAN2 packages (48, 49). Raw particle images were selected in a semiautomatic manner with e2boxer.py in EMAN2. For 3D reconstruction of various PafE:20S CP complexes, we generated an experimental EM-based starting model from 20S_OG CPs images selected from the raw images of PafE:20S_OG complex, instead of using a low-pass–filtered crystal structure of the Mtb 20S CP. The 20S particles were manually inspected to remove “bad” particles that were partially disassembled, having low contrast, or contacting other particles. The contrast transfer function (CTF) was first determined with raw images and corrected for by flipping the phases. Phase-flipped data were then subject to reference-free 2D classification to generate a set of high-contrast class averages that were representative of views present in the raw particle images. The 3D starting model was calculated by e2initialmodel.py using reference-free 2D averages as input, and was refined against the phase-flipped dataset. D7 symmetry was applied during iterative refinement. The final 3D density map was then low-pass–filtered to 60 Å and used as an initial model in the following processing of cryo-EM datasets of various PafE:20S complexes.

Three-Dimensional Cryo-EM of Various PafE:20S Complexes.

We performed cryo-EM and 3D reconstruction on the following three purified PafE:20S CP complexes: PafE:20_OG, PafE_Δ155–166:20S_OG, PafE_Δ155–166:20S_WT. We used lacey carbon-coated 300-mesh copper grids purchased from Structure Probe, Inc. (Prod # 3830C). Grids were glow-discharged in a 0.39-mbar air atmosphere for 1 min by using PELCO easiGlow (Ted Pella) before use. Cryo-EM grids were prepared in an FEI Vitrobot at 11 °C with the relative humidity set to 90% and the blotting pad height offset to −1.0 mm. Next, 3.0 μL of protein sample (∼0.4 mg/mL) were pipetted onto a freshly glow-discharged lacey carbon grid. The sample solution was incubated on the EM grid for 30 s, blotted for 6 s before being plunged into liquid ethane that was precooled by liquid nitrogen. The cryo-EM grids were then transferred to and stored in liquid nitrogen. The cryo-EM grids were transferred in liquid nitrogen into a Gatan 626 cryo-specimen holder and then inserted into the microscope. The specimen temperature was maintained below −170 °C during data collection. Cryo-EM imaging was performed in JEOL JEM-2010F TEM operating at 200 kV. Cryo-EM images were recorded in the low-dose mode (15 e⁻/Ų) at 50,000× microscope magnification on a Gatan UltraScan 4000 CCD camera (4,096 × 4,096 pixel), corresponding to a 2.12 Å per pixel sampling at the specimen level.

Particle selection and image processing were done using the RELION (50) software package. The CTF was first determined with raw micrographs and corrected in ctffind3 (51) and particle selection was performed using semiautomated selection algorithms recently implemented in RELION (52). In brief, after CTF correction ∼1,000 particle images were manually picked from several micrographs and subjected to reference-free 2D classification to generate 10–15 classes. Representative views of protein complexes were then selected and used as templates in the following automated particle selection on all micrographs. Particles were extracted from all micrographs, normalized, and saved in a single data file. Reference-free 2D classification was then performed on all extracted particles. Particles that belong to good 2D classes were extracted and stored in a new particle data file for the following 3D classification and refinement. Next, the selected particles were subjected to unsupervised 3D classification to generate 3D classes, with the low-pass–filtered negative stain of the 20S_OG CP 3D map as the start model. Three-dimensional classes that show identical architectures of proteasome complexes were combined to generate three classes that include particles of 20S CPs only, 20S CP in complex with one PafE ring, and 20S CP in complex with two PafE rings respectively. Particles that do not belong to any of these three 3D classes were discarded at this stage. Each 3D class was subjected to high-resolution 3D autorefine with or without a binary mask of the 20S_OG CP, followed by postrefine processing that included automasking and B-factor sharpening. D7, C7, and D7 symmetry were applied during the refinements of the WT 20S CP only, 20S CP in complex with one PafE ring, and 20S CP in complex with two PafE rings, respectively. Enforcing sevenfold symmetry blurred out the PafE ring in 3D maps, but led to better defined 20S CP structures that were of most interest because our goal was to reveal how the 20S CP structure responded to PafE binding and activation. The binary mask that covered the volume of the 20S CP was calculated in postrefined processing of the 20S CP only. The resolution of the final 3D map was estimated by the gold-standard Fourier shell correlation method at the threshold of 0.143 (53). The number of particles used for 3D reconstruction and estimated resolution are summarized in SI Appendix, Table S6. The difference map was calculated by subtracting the 20S CP from the 20S CP in complex with a PafE ring using the volume operation command vop in Chimera. All image-processing was done either on an 8-CPU Dell Linux workstation or a multinode Dell Linux cluster.

M. tuberculosis Growth Conditions and Heat-Stress Sensitivity Assay.

M. tuberculosis cultures were grown to an OD₅₈₀ ∼1.0 in fresh Middlebrook 7H9 medium (Difco). For growth curves, cultures were diluted to OD₅₈₀ = 0.025 and grown in triplicate, with OD₅₈₀ monitored daily. For heat-stress sensitivity assays, cultures were diluted to OD₅₈₀ = 0.08, a total of 1 mL was transferred to a 2-mL O-ringed tube and incubated for 24 h at 45 °C, and bacterial survival was assessed by plating serial dilutions on Middlebrook 7H11 agar to enumerate colony forming units. Statistical analysis was performed by using a nonparametric Student’s two-tailed t test.