Spiral architecture of the Hsp104 disaggregase reveals the structural basis for polypeptide translocation

Hsp104 hexamers adopt an asymmetric, spiral architecture

We determined the cryo-EM structure of the intact, wild-type Hsp104 complex in the presence of AMP-PNP, a nonhydrolyzable ATP analog, to mimic the substrate-binding ATP state⁸. Purified Hsp104 was determined to be functionally active via established ATPase and disaggregase assays (Supplementary Fig. 1a, 1b). Cryo-EM images of Hsp104-AMP-PNP show homogeneous oligomeric complexes (Supplementary Fig 1c) and reference-free 2D classification reveals a variety of distinct views with well-resolved secondary structural features (Supplementary Fig 1d). In these 2D averages a well-defined three-tier architecture is observed and top-view and side-view projections show a clear asymmetry and a unique protomer arrangement compared to what has been described previously (Figure 1b)^27,31. The final cryo-EM reconstruction following 3D refinement with no imposed symmetry was achieved at an indicated resolution of 6.5 Å for the unsharpened map and 5.6 Å for the sharpened map (Supplementary Fig. 1e). The final map is comprised of 65% of the total data, and 85% of the data following sorting by 2D classification. No additional conformations or oligomeric states were identified by 3D classification (data not shown), thus the final map represents the predominant form of Hsp104 in the dataset. The angular distribution of the particles that went into the final map showed multiple preferred orientations including distinct equatorial and axial views, while additional, less populated orientations were also present that aided the asymmetric refinement (Supplementary Fig 1f). 2D projections of the 3D map match the reference-free averages, confirming the overall asymmetric architecture (Supplementary Figure 1g).

The reconstruction identifies Hsp104-AMP-PNP as a ring-shaped hexamer with three distinct domain-layers that correspond to the NTD, NBD1-MD and NBD2, consistent with other studies (Figure 1c and Supplementary Video 1) ^27,28. Surprisingly, the hexamer adopts an apparent helical spiral arrangement where the NBD rings connect together at a distinct asymmetric seam. The NBD1 and NBD2 show the highest resolution density, at approximately 5 Å, while the NTDs as well as regions at the hexamer seam are more flexible and at ~6–7 Å resolution (Supplementary Fig. 1h). Density for the MD coiled-coil is partly resolved on one face of the hexamer where it wraps around the outside of the hexamer adjacent to the NBD1s (Figure 1c). The overall dimensions are similar to those of previous studies with a 145 Å outer diameter and 140 Å length. The central channel is approximately 25–30 Å in diameter but, surprisingly, opens to form a wide cleft at the hexamer seam due to the spiral offset of the adjacent protomers.

The crystal structure from bacterial (T. thermophilus) ClpB²⁵, which has ~45% sequence identity, was used for initial fitting and to determine a homology model for Hsp104. Helices and β-strands are designated as they are for the ClpB structure (Supplementary Fig. 2a) The reconstruction was docked by rigid body fitting of the domains followed by flexible fitting to achieve an atomic model with the highest correlation to the map, at 0.92. The protomers are designated 1 through 6, with protomer 1 (P1) in the highest position in the side-view orientation, and proceeding counterclockwise when viewed from the NTD-face (Figure 2a, b). The NBDs fit well, with a cross-correlation value of 0.94 between the map and model, and identify the conserved AAA+ subdomain architecture with α-helical and β-sheet regions that are well resolved for each of the protomers (Figure 2c and Supplementary Video 2). For P2-P5, density that likely corresponds to a bound nucleotide is observed in all the NBDs. The NBD1s show a canonical AAA+ arrangement, including expected positions for the Walker A (K218), Walker B (E285), and sensor-1 (T317) residues. Putative Arg finger residues R334 and R333²⁷ are within 7 and 10 Å, respectively, of the nucleotide pocket of the adjacent protomer, indicating a catalytically active arrangement. The NBD2s are similarly well defined showing conserved interprotomer interactions and positions for the Walker A (K620), Walker B (E687), sensor-1 (N728) and sensor-2 (R826) residues, and the proposed Arg finger (R765)⁴⁰ is near the nucleotide pocket at less than 10 Å, based on our model. Protomers P1 and P6 that make up the unusual hexamer seam are in a different conformation and are discussed in further detail below.

Density corresponding to the MD coiled-coil that is partly resolved for P3-P6 was docked with residues 409–467 that correspond to helix L1 and part of helix L2 that connect to the NBD1 large subdomain (Figure 2c). Density for the C-terminal-half of the extended, 85 Å helix L2 as well as helix L3 and helix L4, which includes Hsp70 interaction sites¹⁸, was unable to be modeled, indicating this portion of the MD remains flexible. The NTDs form globular lobes of density that interact and follow the helical arrangement of the NBDs, forming a broad entrance to the channel (Figure 1c and 2a). Despite the NTD flexibility, the NTD-NBD1 linkers are well defined, enabling unambiguous connection to the corresponding protomer. NTD structures were able to be modeled for protomers P2-P4 (Figure 2b), which exhibit the best-resolved features, and the cross correlation of the fit was determined to be 0.91. Further details of the MD and NTD conformations are discussed below.

A large protomer offset and an unusual NBD1-NBD2 interaction define the hexamer seam

The most striking structural feature of the Hsp104 hexamer is the helical-like arrangement of the protomers. While canonical AAA+ interactions are generally maintained around the hexamer, each protomer is tilted slightly, resulting in an approximate 7-fold helical symmetry. Going from P1 to P6, the protomers each rise nearly 10 Å and rotate 53°, on average (Figure 3a). The individual protomers vary in conformation and helical change, indicating the complex is indeed asymmetric (Supplementary Fig. 3a). When the NBD1s are superimposed, continuous conformational changes are observed from P1-P6 (Supplementary Fig 3b), with the greatest protomer-protomer differences between P1 and P6 at the hexamer seam (Supplementary Fig. 3c). The conformational differences primarily involve rotations of the NBD1 and NBD2 small subdomains, which move inward relative to the channel axis by approximately 10° and 20°, respectively, resulting in a more compact P1 protomer compared to P6 (Supplementary Fig. 3c).

The helical shift of the protomers results in a 44 Å-offset between P1 and P6, yet these protomers interact together, covering a 100° rotation along the channel axis to complete the hexamer ring (Figure 3b). Remarkably, this large offset positions the P1-NBD2 adjacent the P6-NBD1, resulting in an unusual interaction between the two different AAA+ domains. Thus, the AAA+ domains form a continuous two-turn spiral around the hexamer (Figure 3c). Although at a slightly lower resolution compared to other protomers, electron density is well defined in this region and reveals interactions between the P1-NBD2 and P6-NBD1 that are comprised of the small and large AAA+ subdomains, respectively (Figure 3d). While precise contacts were unable to be modeled, the interaction appears to be mediated by the B3-b2 and B6-b5 connecting loops from the P6-NBD1 and helices E2 and E3 from the P1-NBD2. Density corresponding to the P1-CTD also contacts the P6 NBD1, towards the outside of the channel, potentially stabilizing the interaction.

Despite the unusual NBD2-NBD1 interaction, density is present that corresponds to bound nucleotide for the P1-NBD2, indicating a potentially active catalytic site (Figure 3d). As indicated above, ATP hydrolysis by the AAA+ domains requires interactions with the adjacent protomer that includes contact by the Arg finger to stabilize the bound nucleotide. While the NBD2-NBD1 interface is different compared to the canonical AAA+ inter-protomer interface, putative Arg finger residues R334 and R333 in the P6-NBD1 are localized approximately 10–12 Å away from the P1-NBD2 nucleotide pocket. Interestingly, another highly conserved Arg residue, R307 in P6, is within 5 Å of the P1-NBD2 pocket. Thus, either a small conformational change that repositions R334/R333 or activation by R307 could trigger hydrolysis at this heteromeric NBD2-NBD1 site. In contrast, density is not observed in the P1-NBD1 pocket, indicating nucleotide is not present (Figure 3d). However, because of the protomer offset, P1-NBD1 is not supported by an adjacent AAA+ domain from P6, thus the nucleotide pocket is likely destabilized and inactive. Despite the adjacent P5-NBD1 in an apparent canonical orientation, density in the P6-NBD1 pocket appears weak relative to the other protomers, indicating partial or no nucleotide occupancy. Finally, density for nucleotide is observed for the P6-NBD2; this site is supported by the adjacent P5-NBD2, which is in a similar arrangement compared to what is observed for other protomers. Overall, comparison of the AAA+ domains for protomers 1 and 6 reveals a mix of nucleotide states and an asymmetric structural organization with distinct interactions relative to the rest of the hexamer that potentially support unique functions during disaggregation.

Arrangement of the NTD, CTD and MD highlight distinct functions in the hexamer

The conserved NTDs bind hydrophobic regions of substrates and are proposed to direct polypeptides to the NBD pore loops as well as enhance cooperative substrate binding required for amyloid disaggregation^15,29. In our Hsp104 map the NTDs interact together, forming a broad, open-ended funnel that mirrors the helical arrangement of the NBD rings (Figure 4a). The P1-P5 NTDs surround the axial channel, forming a C-shaped entrance that is approximately 35 Å in diameter. The opening widens substantially to approximately 50 Å moving towards the P6 NTD at the hexamer seam, creating a large cleft that opens along the side of the hexamer. Although the P6-NTD is more flexible and less well resolved, interactions between the adjacent P5-NTD as well as an unusual interaction with P1-NBD2 are identified when viewed at a reduced threshold for the electron density (Figure 4a). Because of this NTD separation, the P1-NBD1 largely defines an equatorial channel entrance. Based on the arrangement in the other protomers, the P1-MD would extend across the cleft towards the P6-NTD, however density was not sufficient for localization due to its flexibility.

Modeling the P2-P4 NTDs was facilitated by localization of the well-defined NTD-NBD1 connecting density as well as regions that could be attributed to the larger A1 and A6 helices that make up the hydrophobic substrate-binding cleft¹⁵ (Figure 4b). Based on our model only the P3 NTD appears to be positioned with its substrate-binding surface facing towards the channel, while the P2 and P4 NTDs each appear to be in different orientations. Thus, while the NTDs bind one another to form a defined ring at the channel entrance, the interaction interfaces are variable and multiple conformations are adopted around the hexamer.

Conversely, the channel exit on the opposite face of the hexamer is largely defined by the NBD2, however additional density is observed, which we predict corresponds to the 38-residue CTD (Figure 4c). While the CTD is considered to be disordered based on sequence⁴¹, this distinct lobe of density is clearly observed in all the protomers, indicating the CTD is partially structured and interacts with the NBD2 domains around the hexamer. The density extends from the NBD2 C-terminal residue that was modeled based on ClpB and extends across, forming a defined bridge connection to the adjacent protomer, contacting helix D11 in the NBD2 (Figure 4d). The interaction likely stabilizes the hexamer, supporting previous studies ²⁴ and the CTD position around the channel exit also rationalizes previously proposed functions in cochaperone binding²³, and thus could serve as a docking point for substrate hand-off to downstream chaperone systems.

The fit of the MD L1-L2 helices for P3-P6 revealed a substantial, 47° rotation of the MD compared to the crystal structure that enables direct connection to the adjacent protomer (Figure 4e). In this conformation, the MD helix L1 and L2 are positioned alongside the NBD1 of the clockwise neighboring protomer, and together with helix C3, form a 60 Å-long strap across the small and large subdomains (Figure 4f). This interaction is with the opposite protomer that makes up the canonical AAA+ interface, such that each NBD1 nucleotide pocket is supported by the MD of one neighbor and the NBD1 of the other neighbor. Density in this region is well defined for helix L1, which contacts helix C1, the C1-C2 connecting loop and part of helix B3 in the neighboring protomer. Specifically, we identify that conserved charged residues, R419, E427, and D434, that have previously identified functional roles^27,42,43, are positioned adjacent the conserved residues E190, R353 and R366 in the NBD1, respectively, potentially making stabilizing salt-bridge interactions (Supplementary Fig. 4).

Spiral arrangement of the pore loops reveals a directional NBD1-NBD2 substrate translocation mechanism

Each NBD contains flexible loop regions that harbor the conserved Tyr residues (Tyr 257 for NBD1 and Tyr 662 in NBD2) that bind polypeptide substrates and are required for disaggregation functions^5,6,39,44. The pore loops are proposed to face the axial channel in cryo-EM models²⁶ and coordinate substrate threading based on asymmetric structures of the related ClpX single-ring AAA+ complex³². However, the pore loops for both NBD1 and NBD2 are not resolved in the ClpB structure²⁵ and how polypeptides could be actively transferred either across the two distinct AAA+ domains or between protomers has been a mystery. Based on our flexible fitting we are able to model pore loop residues up to 252 and 259 for NBD1 and 659 and 667 for NBD2 within the large lobes of density that extend from these positions (Figure 5a). The NBD1 and NBD2 pore loop densities are separated by 35–40 Å, depending on the protomer, and indeed project toward the channel axis in a staggered, counter clockwise arrangement looking down the N-terminal channel entrance (Figure 5b).

Seen from inside the channel, the NBD1 and NBD2 pore loops form a striking two-turn spiral staircase (Figure 5c and Supplementary Video 3). The P1-NBD1 pore loop is nearest to the NTD channel cleft and each subsequent loop is 15–20 Å away, moving further into the channel towards the P6-NBD2 (Figure 5d). The distances between the pore loops vary, with greater separation near the hexamer seam, particularly between P5 and P6, which are separated by 20 Å, reflecting conformational differences in the protomers. Notably, we observe additional lobes of density located beneath the NBD1 pore loops, corresponding to residues 288–298, which could serve as additional substrate binding surfaces via electrostatic interactions (Supplementary Fig 5). Indeed, this corresponding region in ClpA, a related Hsp100, has been shown to interact directly with substrate⁴⁵. The position of NBD1 loops are approximately 10 Å away from the respective NTDs for each protomer, thus substrates could be transferred to the NBD at multiple sites around the channel. Remarkably, the heteromeric NBD1-NBD2 interaction at the hexamer seam positions the respective pore loops 17 Å apart, similar to the other distances, revealing a continuous substrate transfer pathway between the AAA+ domains.

Protein Purification, ATPase and Luciferase Reactivation Assay

Wild type Hsp104 was purified as described⁵⁶. Hexameric Hsp104 (0.042 μM) was incubated with ATP (1 mM) at 25°C before monitoring the release of inorganic phosphate over 5 minutes using a malachite green kit (Innova). ATPase turnover was measured at a maximum rate of ~14/min, which corresponded to a functional, ATPase-active complex. Disaggregase function was measured by a luciferase reactivation assay as previously described³⁵. Aggregated luciferase (50nM) was incubated with Hsp104 (0.167μM hexamer) in the presence of equimolar Hsc70 and Hdj2 (Enzo Life Sciences) plus ATP (5.1 mM) and an ATP regeneration system (1 mM creatine phosphate, 0.25μM creatine kinase) for 90 min at 25°C.

Cryo Sample Preparation and Data Collection

WT Hsp104 (8.5 mg/mL ) was incubated with AMP-PNP (5 mM ) for 20 minutes at 25°C. Samples were diluted to 0.7 mg/ml in 40 mM HEPES pH=7.5, 40 mM NaCl, 10 mM MgCl₂, 1 mM DTT, 5mM AMP-PNP and 3.5 μl was applied to plasma cleaned C-Flat 2/2 holey carbon grids (Protochips). Vitrification was performed using a Vitrobot (FEI Company) and samples were blotted for 1.5–2 seconds prior to plunge freezing in liquid ethane. Of note, during initial attempts at imaging Hsp104 routinely dissociated into monomers and significant optimization of buffer conditions, sample concentration and freezing conditions was required to achieve proper ice thickness and a homogeneous spread of hexameric particles. Furthermore, the presence of DDM detergent was tested to increase the angular distribution but resulted in dissociation of the hexamer at a variety of concentrations (data not shown).

Samples were imaged using a Titan Krios TEM (FEI Inc.) operated at 300 kV. Images were recorded on a Gatan K2 Summit direct electron detector operated in counted mode at 50,000X nominal magnification corresponding to a calibrated to 1.00 Å/pixel. Dose fractionated imaging was performed by semi-automated collection methods using UCSF Image 4⁵⁷ with a defocus range of 1.5–3 um. Total exposure time was 8 seconds with 0.2 second frames with a cumulative dose of ~45 e⁻ per Ų for 40 frames. Motion corrected frames were summed with the first two frames excluded, and the FFT was visually inspected for sufficient Thon rings prior to additional processing ⁵⁸.

Image Processing and 3D Refinement

All micrographs were CTF corrected using CTFFIND4⁵⁹ and poorly corrected micrographs were removed following visual inspection of the FFT and CTF estimation. An initial single particle dataset was achieved by manual particle picking using e2boxer (EMAN2) ⁶⁰, which yielded ~50,000 particles from 1731 micrographs. Well-populated reference-free 2D class averages, determined with Relion⁶¹, were used for templated automated particle picking with the Template Picker in Appion⁶² to achieve a ~200,000 single particles dataset from 1930 micrographs. The total dataset of ~250,000 particles was initially sorted following 2D classification by removing particles images from poorly resolved class averages, resulting in a total dataset of ~190,000 particles. All subsequent 3D processing was performed using Relion⁶¹ with no symmetry imposed based on the asymmetric arrangement identified in the 2D class averages (Figure 1b). A previously determined cryo-EM 3D reconstruction of ATP-Hsp104 (EMDB: 1600) ³¹, low pass filtered to 50Å, served as an initial model for 3D classification and refinement. The 8 models generated from 3D classification appeared homogenous (data not shown), therefore, the full dataset was used for initial gold-standard 3D refinement. The resulting 3D model refined to an estimated resolution of 7.5 Å (data not shown). Z-score parameters based on this refinement, defined in Relion, were written for the data set and the dataset was trimmed to the highest Z-score for 160,000 particles. Additional 3D refinement was performed with this trimmed data set and resulted in a final model with an estimated resolution of 6.54 Å using the FSC = 0.143 FSC criterion (Supplementary Fig 1e). Additional trimming based on the Z-score showed no improvement and extensive 3D classification and refinement of individual classes as well as local refinement using 3D masking was tested and did not did not yield improvements in the map. Thus, the resolution is likely limited by flexibility of multiple regions in the complex (Supplementary 1h) and moderate preferred orientation (Supplementary Fig. 1f). For the final sharpened map, the “Post-processing” procedure was used to generate a soft mask for the two half maps prior to FSC estimation, which was determined to be 5.64 Å (Supplementary Fig 1e). Automated B-factor sharpening was carried on the combined map with an estimated -188 B-factor. The local resolution was estimated using ResMap on the unsharpened map⁶³.

Modeling

This final sharpened map was used for all rigid body docking and flexible fitting of atomic structures. Rigid body fitting was performed with UCSF Chimera⁶⁴. Initial comparisons were performed by fitting individual protomers with the protomers in T. thermophilus ClpB crystal structure (PBD_ID=1qvr), but resulted in a low cross correlation value and clear conformational differences were apparent. Individual subdomains (NTD, NBD1 large, NBD1 small, MD (residues 409–467), NBD2 large, NBD2 small) from the crystal structure were then docked as rigid bodies and resulted in an improved fit, however a number of AAA+ domain helices did not align with the density. Therefore, to more accurately interrupt the map, a homology model of Hsp104 (residues 6–857) was determined from the ClpB structure (residues 4–850) using SWISS_MODEL⁶⁵ followed by rigid body docking the subdomains. Flexible fitting of the protomer sub volumes and the entire hexamer map was then performed using phenix. real_space_refine⁶⁶ using backbone carbons of our homology model. The real space refinement was implemented at 6σ, resolution 6Å and the cross correlation improved from 0.72 to 0.9 for the complete map after refinement. Nucleotide density was seen for 10 of the nucleotide binding pockets, therefore this molecule was included in the model and docked based on the ClpB structure. Manual modeling of the tyrosine loop regions was performed in COOT⁶⁷. Loop regions within the NBD1 domain (248–260,287–300) were modeled by backbone positioning within the electron density; residues 251–259, 292–297 were unable to be modeled in the density. Of note, a major shift was required to model the NBD2 tyrosine loop (656–669), which showed a distinct curve away from the flanking helices. Similarly, the most central portion of the loop lacked density and was not modeled (659–666). All images were generated using UCSF Chimera⁶⁴.

Accession codes

The cryo-EM electron density map and molecular model have been deposited in the Electron Microscopy Data Bank (EMDB-3416).