Structure of a AAA+ unfoldase in the process of unfolding substrate

AAA+ unfoldases are thought to unfold substrate through the central pore of their hexameric structures, but how this process occurs is not known. VAT, the Thermoplasma acidophilum homologue of eukaryotic CDC48/p97, works in conjunction with the proteasome to degrade misfolded or damaged proteins. We show that in the presence of ATP, VAT with its regulatory N-terminal domains removed unfolds other VAT complexes as substrate. We captured images of this transient process by electron cryomicroscopy (cryo-EM) to reveal the structure of the substrate-bound intermediate. Substrate binding breaks the six-fold symmetry of the complex, allowing five of the six VAT subunits to constrict into a tight helix that grips an ~80 Å stretch of unfolded protein. The structure suggests a processive hand-over-hand unfolding mechanism, where each VAT subunit releases the substrate in turn before re-engaging further along the target protein, thereby unfolding it. DOI: http://dx.doi.org/10.7554/eLife.25754.001


Main text:
The 85 kDa VAT protein, which is composed of an N-terminal regulatory domain (NTD) and two tandem nucleotide-binding AAA+ domains (NBD1 and NBD2), assembles into a 500 kDa homohexamer ( Fig. 1 Supplement 1). Truncation of the first 182 residues of VAT removes the regulatory NTD, producing a construct (ΔN-VAT) that has both higher ATPase and unfoldase activity than the intact protein [1][2][3] . In solutions of ΔN-VAT, the enzyme itself is degraded in an ATP-and proteasome-dependent manner ( Fig. 1a-c). The 20S proteasome does not hydrolyze ATP during proteolysis 4 and therefore this observation demonstrates that ΔN-VAT complexes must unfold other copies of ΔN-VAT via a process that requires ATP hydrolysis prior to degradation by the proteasome.
Cryo-EM of full-length VAT previously showed that the enzyme adopts a helical splitring conformation in the presence of ADP 5 . ΔN-VAT particles with bound ADP assemble into long fibrils (Fig. 1d), presumably comprised of helical split-ring structures interacting with each other top-to-bottom. With the slowly-hydrolysable ATP-analogue ATPγS bound, full-length VAT exists in both the split-ring conformation and a six-fold symmetric stacked-ring conformation 5 . Similarly, ΔN-VAT with bound ATPγS forms a mixture of fibrils (Fig. 1e) as well as single particles, the majority of which appeared to be six-fold symmetric (Fig. 1e, red circles). The fibrils could be excluded easily when selecting single particle images for analysis, allowing calculation of a cryo-EM map of the six-fold symmetric stacked-ring conformation of VAT at 3.9 Å resolution. This resolution enabled construction of an atomic model for the complex (Fig. 2, Fig. Supplement 1, Table 1). In the structure, a central pore runs along the six-fold symmetry axis of the complex through the center of the stacked NBD1 and NBD2 rings (Fig. 2a, left, hexagon). The stacked-ring structure of VAT resembles recent structures of the VAT homologue p97 in the fully ATP-loaded state [6][7][8] , except for the orientations between the NBD1 and NBD2 domains in VAT monomers and the relative rotation of the NBD1 and NBD2 rings. All AAA+ unfoldases have conserved sequences that extend from the NBDs towards the center of the ring, forming 'pore loops' that are essential for unfolding substrate 1 (Fig. 2 Supplement 2). In the six-fold symmetric VAT structure, pore loop 1 of NBD1, containing conserved and essential tyrosine residues (Tyr264 and Tyr265) 1 , forms an opening that is ~25 Å across while pore loop 2 of NBD1 (residues 299 to 305) forms a slightly tighter aperture that is ~18 Å across ( including the linker region between NBD1 and NBD2 (Fig. 2c, left), and both NBD1 and NBD2 have strong density for ATPγS in their nucleotide-binding pockets (Fig. 2c, right) indicating that VAT adopts a rigid and symmetric conformation when fully loaded with nucleotide.
Unsupervised classification of ATPγS-loaded ΔN-VAT particle images using a new ab initio 3D classification algorithm 9 allowed identification of a second conformation of the protein that was significantly different from known structures. The class corresponding to this conformation, containing ~15 % of particle images, showed two ΔN-VAT complexes in close apposition to one another (Fig. 3, Fig. 3 Supplement 1). Refinement of this structure to ~4.8 Å resolution revealed it to consist of a ΔN-VAT complex in the process of unfolding a neighboring complex as substrate (Fig. 3a). The map shows clear density for the substrate protein in an extended conformation as it runs almost directly through the central pore of the active ΔN-VAT ( Fig. 3a and b, red density). Density for the substrate above NBD1 of the active ΔN-VAT complex is weak and fragmented, and is most clearly visualized without sharpening of the map (Fig. 3a, above dashed line) 10 , suggesting that outside of the active enzyme the substrate has a flexible orientation relative to the pore of ΔN-VAT. The length of unfolded protein gripped by the pore of the active ΔN-VAT is ~80 Å, which corresponds to 12 to 14 extended amino acid residues.
In both NBD1 and NBD2 five of the six VAT subunits form close contacts with the substrate through their pore loops ( Fig. 3b and c, and . The hydrophobic residues of these pore loops create a tight helical sheath around the substrate (Fig. 3a, green). Breaking the six-fold symmetry of the complex forms a 'seam' where one protomer has interfaces that differ dramatically from other protomerprotomer contacts (Fig. 3c, gold) and the seam subunit does not interact with the substrate . This helical arrangement of the pore loop residues maximizes the contact area between the unfolded substrate and the pore loops of the five subunits that form the helix (Fig. 3b, Video 1). Unfolded substrate that spans the gap between the pore loops of NBD1 and NBD2 has a low density in the cryo-EM map, likely because of flexibility of the substrate in this region. The transition from the stacked-ring to the substrate-engaged conformation of ΔN-VAT occurs with rearrangement of NBD1 relative to NBD2 in each monomer and a rotation of the NBD1 ring against the NBD2 ring ( Fig. 3 Supplement 2, Video 1 and 2). The two rings are also displaced horizontally relative to each other in the substrate-bound state and the central pore is tilted so that it passes between the rings at an angle of ~15° ( Fig. 3 Supplement 2c). This rotation and displacement brings the NBD1 and NBD2 rings closer together than in the stacked-ring conformation, decreasing the distance between their pore loops.
In this study, only the newly identified conformation described above is found with substrate bound. Consequently, the six-fold symmetric stacked-ring state with ATPγS bound and the split-ring state with ADP bound 5 likely correspond to pre-and postunfolding conformations of the enzyme. In the substrate-bound conformation, the seam protomer is positioned such that it does not interact with the substrate. Notably, recent structures of the 26S proteasome 11,12 also showed that either the Rpt5 or Rpt6 subunit in the hexameric Rpt ring of the 19S regulatory particle base is displaced from the pore, similar to the seam protomer in VAT (Fig. 3). Furthermore, the translocation mechanisms of AAA+ DNA helicases such as E1, Rho, and MCM, also involve a helical gripping state in which only five of six protomers interact with a single DNA strand [13][14][15] . The structure presented here suggests an analogous substrate translocation mechanism for VAT where the enzyme operates processively with a 'hand-over-hand' mechanism, shown schematically in Figure 4a The process repeats in a cyclical manner for additional protomers (e.g. step 3: Fig. 4a, bottom, Video 3). The power stroke for substrate unfolding is provided by ATP hydrolysis, which occurs in sequence around the VAT ring, with ADP released by the protomer adjacent to the seam (counter clockwise from the seam when viewed from NBD1 toward NBD2, Fig. 4b). Although the precise position where ATP hydrolysis and ADP release occur in the cycle is not apparent from our data, it is clear that ATP must rebind to the seam protomer as it engages with substrate at the top of the protomer helix.
As the cycle continues, each protomer pulls a single section of the substrate, corresponding to ~13 Å or approximately two extended residues, through the central pore until it becomes the lowest protomer in the helix and releases the substrate (Video 3). The cyclic nature of the ATP hydrolysis gives directionality to the substrate translocation, allowing unfolded protein to be passed to the proteasome for degradation.
The cyclic 'hand-over-hand' model requires that each protomer hydrolyze ATP in turn The slight elevation relative to the A= (1r) 6 profile in Figure 4c may reflect an 'escape' mechanism that allows stalled VAT complexes to disengage and then reengage with substrate, as seen with other molecular machines 16 . In the absence of an escape mechanism ATP hydrolysis would proceed from one protomer to the next in a cyclical manner (Fig. 4b) until a defective subunit is reached, preventing further hydrolysis and, thus, subsequent substrate unfolding. In contrast, if the machine is able to disengage at this point it would be possible for other functional subunits to hydrolyze ATP, although at reduced rates. The existence of an escape mechanism therefore has important implications for substrate unfolding, even in the context of wild type VAT, by allowing the release of stalled substrate. This release minimizes both the loss of active enzyme and increases catalytic efficiency. Our ATPase assay data in Fig. 4c. cannot distinguish between a model where each protomer hydrolyzes ATP simultaneously or a sequential binding/hydrolysis mechanism as described in Fig. 4a and b. However, previous ATPase measurements with VAT followed Michaelis-Menten kinetics 1 , with little cooperativity for ATP hydrolysis. This low cooperativity is inconsistent with simultaneous hydrolysis at all twelve nucleotide binding sites but is consistent with the proposed hand-over-hand mechanism, where only one NBD needs to bind and hydrolyze nucleotide at a time. The sequential model for ATP hydrolysis is also consistent with the asymmetry of the complex and nucleotide binding sites observed here, and the homology between VAT and other AAA+ enzymes known to hydrolyze ATP in a sequential fashion 15,17 .
The cryo-EM structure of substrate-bound ΔN-VAT shows that both NBD rings of the complex adopt similar conformations ( Fig. 3c and d). This similarity suggests that the mechanism of substrate unfolding proposed here may well operate in single-ring AAA+ unfoldases as well, including Rpt 1-6 of the 19S proteasome, and the hexameric PAN and ClpX unfoldases. Thus, a picture emerges whereby AAA+ unfoldases adopt asymmetric substrate-bound conformations, creating narrow channels that sequester unfolded stretches of target proteins. Notably, even in molecular machines comprised of homooligomers, asymmetry can play an important functional role, in the case of VAT by priming the enzyme for processive translocation of its substrate.

Video 2. Conformational changes in VAT monomers upon substrate binding.
Interpolation between the stacked-ring and substrate-bound states of VAT, shown as a ribbon diagram for two VAT monomers. Substrate is shown in red. Pore loop residues are indicated in green. Scale bar, 25 Å. Please view video as loop.

Protein expression and purification
An expression vector for the T. acidophilum VAT gene with the N-terminal 181 residues deleted (ΔN-VAT) 5

Electron microscopy image processing
Whole frame alignment of movies was performed with alignframes_lmbfgs 22 and the resulting averages of frames were used for contrast transfer function (CTF) determination with CTFFIND4 23 and automated particle image selection with RELION 24 . Particle coordinates were used to extract particle images in 256 × 256 pixel boxes from the unaligned movies, while performing individual particle movement correction and exposure weighting with alignparts_lmbfgs 22 . Magnification and CTF parameters were corrected for a previously measured 2 % magnification anisotropy 25 resulting in a calibrated pixel size of 1.45 Å. A total of 171,381 candidate particle images were subjected to two rounds of 2D classification, ignoring the CTF until the first peak, in RELION 26 . Visual inspection and selection of classes yielded 93,469 particle images, which were then transferred to the program cryoSPARC 9 for ab initio 3D classification and refinement. Multiple 3D reference-free classifications were performed with between two and six classes using the stochastic gradient descent algorithm, before selecting two classes that yielded the highest quality maps. For refinement, a class distribution threshold of 0.9 was used to define individual particle images as belonging to a class.
Refinement of the six-fold symmetric class of 75,205 particle images using C6 symmetry yielded a reconstruction at a resolution of 3.9 Å as measured by Fourier shell correlation (FSC) following a gold-standard refinement. A gold-standard refinement of the substrateengaged class of 13,238 particle images with no symmetry applied yielded a reconstruction at a resolution of 4.8 Å by FSC.

Map analysis and model building
Most of the density for loops, β-sheets and α-helices in the 3.9 Å resolution map were of sufficient quality to allow model building. A homology model of VAT that had previously been fit flexibly into a lower-resolution map 5 was rigidly fit into the map as individual domains (NBD1 and NBD2) with UCSF Chimera 27 . Final models were built with successive rounds of manual model building in Coot 28 and real space refinement in Phenix 29 , and gave an EMringer score of 1.17, which is slightly better than the typical score of 1.0 for a map at this resolution. For the substrate-bound complex, the resolution was insufficient to determine side chain orientations, and a refined structure from the symmetric class was fit into the density by molecular dynamics flexible fitting (MDFF 30 ). Density for each monomer was segmented with UCSF Chimera 27,31 . Only backbone atoms were subject to the fitting force during simulation, and optimization of geometry and rotamers was performed with Phenix 29 .

Degradation Assay
The auto-degradation of ΔN-VAT was measured following the protocol of Sauer and where Glu291 (NBD1) and Glu568 (NBD2) were substituted with Gln, followed by the unfolding/refolding protocol described above. Hydrolysis of ATP was monitored in a UV-visible light spectrometer at 340 nm, 25 °C, in the presence or absence of 30 µM GFP substrate with an 11 residue ssRA tag 4 . All measurements were done in triplicate.
Similar results were obtained in the absence of substrate.
The expected activity of VAT complexes prepared by mixing different ratios of wild type and mutant protomers can be calculated as described previously 16  . We assume that the unfolding/refolding process leads to equal probabilities of insertion of either a mutant or wild type protomer into a VAT hexamer, so that r is the probability of any protomer in the hexamer being a mutant. Under this assumption the fraction of VAT molecules with k mutant protomers is given by where n k ⎛ ⎝ ⎜ ⎞ ⎠ ⎟ = n! k! n − k ( ) ! , with the concentration of functional ATP hydrolysis sites in the protomer given by 2 ×(6-k) (two nucleotide sites per VAT). Assuming further that ATPase activity in the functional sites is not affected by the incorporation of inactive protomers until m are added, in which case all hydrolysis stops, one can write the fractional activity of a solution of VAT complexes as a function of r, as where A(r) is activity normalized relative to the expected hydrolysis rate for fully wild type VAT hexamers at an identical concentration. Our structural model predicts that a single defective protomer is enough to eliminate ATPase activity in the hexamer. In this case m=1 and A(r) = (1-r) 6 . In contrast, in the case where there is no coupling between active and inactive protomers,