Structures of the ATP-fueled ClpXP proteolytic machine bound to protein substrate

ClpXP is an ATP-dependent protease in which the ClpX AAA+ motor binds, unfolds, and translocates specific protein substrates into the degradation chamber of ClpP. We present cryo-EM studies of the E. coli enzyme that show how asymmetric hexameric rings of ClpX bind symmetric heptameric rings of ClpP and interact with protein substrates. Subunits in the ClpX hexamer assume a spiral conformation and interact with two-residue segments of substrate in the axial channel, as observed for other AAA+ proteases and protein-remodeling machines. Strictly sequential models of ATP hydrolysis and a power stroke that moves two residues of the substrate per translocation step have been inferred from these structural features for other AAA+ unfoldases, but biochemical and single-molecule biophysical studies indicate that ClpXP operates by a probabilistic mechanism in which five to eight residues are translocated for each ATP hydrolyzed. We propose structure-based models that could account for the functional results.

Introduction AAA+ motors harness the energy of ATP hydrolysis to carry out mechanical tasks in cells (Erzberger and Berger, 2006). In the ClpXP protease, for example, AAA+ ClpX ring hexamers bind target proteins, unfold them, and translocate the unfolded polypeptide through an axial channel and into the peptidase chamber of ClpP, which consists of two heptameric rings ( Figure 1A; Wang et al., 1997;Grimaud et al., 1998;Ortega et al., 2000;Ortega et al., 2002;Sauer and Baker, 2011). In the absence of ClpX or another AAA+ partner, small peptides diffuse into the ClpP chamber through narrow axial pores, but larger peptides and native proteins are excluded and escape degradation (Grimaud et al., 1998;Lee et al., 2010a). The substrates of Escherichia coli ClpXP include aberrant ssrA-tagged proteins, produced by abortive translation, and normal cellular proteins synthesized with degradation tags that program rapid turnover Keiler, 2015).
ClpX subunits consist of a family specific N-terminal domain, which is dispensable for degradation of ssrA-tagged proteins, and large and small AAA+ domains, which contain sequence motifs that mediate ATP binding and hydrolysis, ClpP binding, and substrate recognition ( Figure 1B; Baker and Sauer, 2012). Ring hexamers of ClpX bind the ssrA tag within an axial channel (Martin et al., 2008a). Following degron binding, ATP-fueled power strokes pull on and eventually unfold attached native domains (Kenniston et al., 2003). Single-chain ClpX DN pseudohexamers, containing six 'subunits' linked by genetically encoded tethers, support ClpP degradation of ssrA-tagged substrates at rates similar to wild-type ClpX (Martin et al., 2005). Eliminating ATP hydrolysis in four or five subunits of single-chain pseudohexamers slows but does not prevent ClpP-mediated degradation, suggesting that ATP hydrolysis in any one of multiple subunits in the ClpX ring is sufficient to power unfolding and translocation (Martin et al., 2005). In optical-trapping experiments using single-chain ClpX and ClpP, the smallest observed translocation steps correspond to movement of five to eight amino acids of the substrate, and kinetic bursts of power strokes produce fast translocation steps two, three, or four-fold larger in terms of the number of residues translocated (Aubin-Tam et al., 2011;Maillard et al., 2011;Sen et al., 2013;Cordova et al., 2014;Olivares et al., 2017). Such bursts do not occur in repeating patterns, supporting probabilistic but coordinated ATP hydrolysis within the ClpX ring.
We describe here near atomic-resolution single-particle cryo-EM structures of single-chain ClpX pseudohexamers bound to ClpP and protein substrates. These structures show how asymmetric hexameric rings of ClpX dock with symmetric heptameric rings of ClpP and reveal how the pore-1, pore-2 and RKH loops of ClpX function in substrate binding. ClpX adopts a spiral conformation, with neighboring pore-1 loops interacting with every two residues of substrate in the axial channel, as observed in other AAA+ unfolding and remodeling machines (Puchades et al., 2017;de la Peña et al., 2018;Dong et al., 2019;Majumder et al., 2019;Ripstein et al., 2017;Gates et al., 2017;Zehr et al., 2017;Han et al., 2017;Monroe et al., 2017;Su et al., 2017;Sun et al., 2017;Yu et al., 2018;White et al., 2018;Cooney et al., 2019;Rizo et al., 2019;Shin et al., 2019;Twomey et al., 2019). Based on these structural features, strictly sequential models of ATP hydrolysis and a power stroke that moves two residues of the substrate per translocation step have been  proposed. As noted, however, ClpX does not need to operate in a strictly sequential manner and takes translocation steps substantially longer than two residues. Thus, an apparent incongruity exists between the structural and functional studies. We discuss this conflict and propose structure-based translocation models that reconcile how ClpX might use probabilistic ATP hydrolysis to take larger translocation steps of varying length.

Cryo-EM structures
For cryo-EM studies, we used epitope-tagged variants of Escherichia coli ClpP and a single-chain variant of E. coli ClpX DN with an E185Q mutation to eliminate ATP hydrolysis without compromising nucleotide, ClpP, or substrate binding (Hersch et al., 2005;Martin et al., 2005). Single-chain ClpX DN was used to ensure subunits of the pseudohexamer do not dissociate during sample preparation. This enzyme has been used previously for many biochemical and single-molecule studies (Martin et al., 2005;Martin et al., 2007;Martin et al., 2008a;Martin et al., 2008b;Aubin-Tam et al., 2011;Maillard et al., 2011;Glynn et al., 2012;Sen et al., 2013;Stinson et al., 2013;Cordova et al., 2014;Iosefson et al., 2015a;Iosefson et al., 2015b;Olivares et al., 2017;Rodriguez-Aliaga et al., 2016;Amor et al., 2019;Bell et al., 2018). We purified enzymes separately and incubated ClpX DN (4 mM pseudohexamer), ClpP (2 mM 14-mer), and ATPgS (5 mM) for five min before vitrification. Experiments using ATP instead of ATPgS resulted in fewer ClpXP complexes. Imaging revealed complexes with ClpP bound to one or two ClpX DN hexamers (Figure 1-figure supplement 1A-C). We analyzed the more abundant doubly capped complexes. As single-particle data processing with C 2 symmetry produced maps with conformational heterogeneity, we used signal-subtraction methods to allow independent classification and refinement of ClpX DN or ClpP density (Figure 1-figure supplement 1D-F). We calculated a D 7 symmetric map for ClpP and parts of ClpX DN making symmetric contacts, four different classes of symmetry-free ClpX DN maps, and then extended each asymmetric ClpX DN map to include one heptameric ClpP ring. Final structures have good stereochemistry with resolutions from 3.2 to 4.3 Å (Figure 1-figure supplement 2; Supplementary file 1- Table S1). Substrates were observed in all ClpX DN structures and probably represent bound endogenous peptides/proteins or partially denatured portions of ClpP or ClpX DN . Figure 1C shows density for a composite ClpX DN /ClpP/substrate complex. The six subunits of ClpX DN formed a shallow spiral (labeled ABCDEF from top to bottom) in all structural classes, which differed largely in substrate density or nucleotide state, and the hexameric ClpX DN and heptameric ClpP rings were slightly offset ( Figure 1D). The structure of the ClpX DN hexamers in the class-1, class-3, and class-4 EM structures were very similar to each other (pair-wise Ca RMSDs 1.2-1.9 Å ). The hexamer in the class-2 structure was generally similar (pair-wise Ca RMSDs 2.6 Å ) but differed in the position of the 'seam', a dilated inter-subunit interface that occurs as a result of ring closure. This seam was located between subunits A and B in the class-2 structure and between subunits F and A in the class-1, class-3, and class-4 structures ( Figure 1E). As discussed below, this difference appears to be related to the identity of the nucleotide bound in subunits A or F.

ClpX docking with ClpP
Our D 7 -symmetric map included ClpP 14 and symmetric interface contacts with ClpX DN (Figure 2A). In both heptameric ClpP rings, the N-terminal residues of each subunit formed a collar of b-hairpins creating a pore into the degradation chamber ( Figure 2A ClpP had essentially the same structure in the D 7 and symmetry-free ClpXP maps. ClpX IGF loops, named for an Ile 268 -Gly 269 -Phe 270 sequence, are critical for ClpP binding Joshi et al., 2004;Martin et al., 2007;Amor et al., 2019) and were responsible for most contacts in our structures ( Figure 2C). These interactions included packing of the Ile 268 , Phe 270 , and Val 274 side chains into pockets at interfaces between ClpP subunits ( Figure 2D). I268L, F270L, and V274A variants have severe ClpP binding defects (Amor et al., 2019). Asymmetric ClpX-ClpP docking relies on conformational adjustments in the N-and C-terminal residues of individual IGF loops, allowing the central portion of each IGF loop to contact the flat ClpP ring despite projecting from a spiral ( Figure 2E). Changes in IGF-loop length decrease ClpP affinity and degradation activity (Amor et al., 2019), suggesting that these loops act as shock absorbers to maintain ClpP contacts In our structures, the side chain of Arg 192 in ClpP appeared to hydrogen bond to the backbone of the IGF loop ( Figure 2F). Unlike wild-type ClpP, R192K ClpP neither bound ClpX DN in pull-down assays ( Figure 2G) nor degraded protein substrate in the presence of ClpX DN ( Figure 2H). Smallmolecule acyldepsipeptides (ADEPs) bind in the same ClpP pockets as the ClpX IGF loops and kill bacteria by opening the ClpP pore to facilitate rogue degradation of unstructured proteins (Brö tz-Oesterhelt et al., 2005;Kirstein et al., 2009;Li et al., 2010;Lee et al., 2010b). ADEPs stimulated R192K and wild-type ClpP decapeptide cleavage to similar extents ( Figure 2I), establishing that R192K ClpP has normal peptidase activity. Thus, Arg 192 is critical for ClpX binding but not for ADEP binding or ClpP pore opening. Modifying ADEPs to interact with Arg 192 may increase affinity for ClpP and improve their efficacy as antibiotics.
Crystal structures show that ADEP binding to ClpP results both in opening of the axial pore and assembly of its N-terminal sequences into a collar of b-hairpins (Li et al., 2010;Lee et al., 2010b). Based on a recent cryo-EM structure of Listeria monocytogenes ClpXP, it was concluded that ClpX binding does not induce the same widening of the ClpP pore as does ADEP binding (Gatsogiannis et al., 2019). By contrast, our results support the opposite conclusion, as the~30 Å diameter and overall structure of the ClpP pore in our E. coli ClpXP structures were extremely similar to the crystal structure of ADEP-activated E. coli ClpP (Li et al., 2010), with an overall RMSDs of 0.8 Å for all Ca's in a single ClpP ring and 0.6 Å for all Ca's of the seven N-terminal b-hairpins and adjacent a-helices that define the pore diameter.

Substrate binding
All ClpX DN maps contained substrate density within the axial channel, which we generally modeled as an extended poly-alanine chain ( Figure 1D), although additional side-chain density of the substrate was modeled as  arginine (residue 5) in the class-1 structure and histidine (residue 3) in the class-3 structure . Substrate was built with the N terminus facing ClpP in the class-2 and class-4 structures and in the opposite orientation in the class-1 and class-3 structures. ClpXP can translocate substrates in either the N-to-C or C-to-N direction (Olivares et al., 2017). In all structures, the top of the ClpX channel was most constricted by tight packing between ClpX side chains and substrate ( Figure 3A-B), providing a structural basis for experiments showing that interactions with substrate near the top of the ClpX channel are most important for unfolding grip (Bell et al., 2019). The class-1 and class-3 structures also had density corresponding to a roughly globular native domain above the channel ( Figure 1C, Figure 4-figure supplement 1B).
In our structures, the pore-1, pore-2, and RKH loops of ClpX contacted substrate in accord with genetic and biochemical studies (Siddiqui et al., 2004;Farrell et al., 2007;Martin et al., 2008a;Martin et al., 2008b;Iosefson et al., 2015a;Iosefson et al., 2015b;Rodriguez-Aliaga et al., 2016). For example, substrate was contacted by the pore-1 loops of subunits B/C/D/E in all classes, Depending on the class, from three to five pore-2 loops also contacted substrate, with Val 202 making many of these interactions ( Figure 4A, B and D, Figure 4-figure supplement 1C-D; Video 2).
The RKH loop, which has not been visualized in previous structures and is unique in ClpX-family enzymes , consisted of an antiparallel b-ribbon stem and a short helix that includes part of the conserved Arg 228 -Lys 229 -His 230 motif ( Figure 4E). When contacting the globular portion of substrate above the pore, the RKH loops mimicked adjustable structural jacks supporting a house during foundation repair ( Figure 4E). The RKH loops alter substrate specificity (Farrell et al., 2007;Martin et al., 2008a), but whether they are critical determinants of substrate binding is unknown. To test this possibility, we constructed and assayed RKH-loop mutants. The most severe mutant (RKHfiAAA) did not support ClpP degradation of an ssrA-tagged substrate and hydrolyzed ATP~5 fold faster than the parent ( Figure 4F-G). Changing RKH to AKH, KHR, or KRH slowed degradation ( Figure 4F). A variant [RKH 2 -AAA] 2 pseudohexamer with four wild-type RKH loops and two AAA mutations had a~3 fold higher K M for substrate degradation than the parent

Nucleotide binding and motor conformations
We observed density for ATPgS in five subunits of each ClpX DN hexamer ( Figure 5A; Supplementary file 1- Table S2). In the sixth subunit, bound nucleotide density fit best as ADP, which is a 10-15% contaminant in ATPgS preparations, but was slightly less convincing than ATPgS density in other subunits ( Figure 5B; Supplementary file 1- Table S2), raising the possibility of averaging of multiple nucleotide-binding states. 'ADP-bound' subunits had poorly structured channel loops, made fewer contacts with substrate and with neighboring subunits, Video 2. Interaction of substrate with ClpX DN . Substrate in the axial channel of the class-4 ClpX structure adopts an extended conformation and is contacted by the pore-1, pore-2, and RKH loops of ClpX, which are arranged into a spiral and pack against each other. https://elifesciences.org/articles/52774#video2   Table  S3).
Side-chain contacts with nucleotide included ClpX residues Val 78 -Ile 79 (box-II), Lys 125 -Thr 126 -Leu 127 (Walker A), Asp 184 -Gln 185 (Walker B with E185Q mutation), Arg 307 (arginine finger), and Arg 370 (sensor-II) ( Figure 5A and B, Figure 5-figure supplement 1A). V78A/I79A, E185Q, and R370K mutations eliminate ATP hydrolysis or weaken nucleotide binding (Joshi et al., 2004;Hersch et al., 2005;Martin et al., 2005;Stinson et al., 2013). We found that K125M, T126A, L127A, D184A, and R307A mutations also severely inhibited ATP hydrolysis ( Figure 5-figure supplement 1B). ClpX does not have a traditional sensor-I residue, which in many AAA+ enzymes positions a water molecule for hydrolysis of ATP bound to the same subunit (Erzberger and Berger, 2006). For ATPgS bound in subunits A-E of our structures, the side chain of Glu 303 was close to the g-thiophosphate, at a sensor-I-like position ( Figure 5A). We found that an E303A mutation severely inhibited ATP hydrolysis, whereas E303Q was partially defective (Figure 5-figure supplement 1B-C). Based on these results, we refer to Glu 303 as a sensor-I* element. As Glu 303 is in the same helix as Arg 307 , the arginine-finger residue, these residues could coordinate structurally to activate ATP hydrolysis in a neighboring subunit.
A short hinge connects the large and small AAA+ domain of each ClpX DN subunit. In all of our structures, the conformations of the small domains were very similar to each other, as were those of the large domains after residues 65-114, which comprise a subdomain, and the pore-1, pore-2, RKH, and IGF loops were removed (Figure 5-figure supplement 2A). The relative orientations of the large and small AAA+ domains were similar across all four structural classes for subunits at equivalent spiral positions, but conformational changes in the hinge resulted in different orientations of the large and small domains at many positions in the spiral for each hexamer ( Figure 5F). Changes in hinge length or deletion of one hinge largely eliminate ClpX function (Glynn et al., 2012;Bell et al., 2018). Thus, the conformational changes associated with a power stroke likely arise from changes in hinge conformations, whereas movements of different large-domain loops or the 65-114 subdomain mediate ring closure and asymmetric contacts with ClpP and substrate.
ATP hydrolysis requires proper positioning of the box-II, Walker-A, and Walker-B elements in the large AAA+ domain of one ClpX subunit, the sensor-II arginine from the small AAA+ domain of the same subunit, and the arginine-finger/sensor-I* element from the large domain of the clockwise subunit (viewed from the top of the ring). These structural features depend on how the small AAA+ domain of each subunit packs against the large AAA+ domain of its clockwise neighbor, which was similar for units A/B, B/C, C/D, D/E, and E/F, suggesting that the associated ATP-binding sites are hydrolytically active. By contrast, the structure of the FA interface was different as a consequence of changes in rotation of the 65-114 subdomain in subunit A relative to the rest of the large AAA+ domain and changes in a loop that contains the sensor-II arginine in subunit F ( Figure 5G). These changes resulted in disengagement of the Arg-finger and sensor-I* side chains in subunit A from the nucleotide bound to subunit F ( Figure 5B; Supplementary file 1- Table S4). Thus, in each of our structures, the nucleotide-binding site in subunit F appears to be catalytically inactive.

ClpX interactions with ClpP and substrates
Our cryo-EM structures provide snapshots of the binding of E. coli ClpX DN to ClpP, to protein substrates, and to nucleotides. In each of our structures, the six subunits of the ClpX ring hexamer are arranged in a shallow spiral. Slightly altered orientations of the large and small AAA+ domains in each ClpX subunit allow the hexameric ring to remain topologically closed with each large domain contacting the small domain of one neighbor. These structural results are consistent with biochemical experiments that show that ClpX is fully functional when all of its subunit-subunit interfaces are covalently crosslinked (Glynn et al., 2012) and support a model in which the architectural changes in the spiral that drive a power stroke result from changes in the conformations of the hinges connecting the large and small AAA+ domains of each subunit. Our structures show that relatively flexible interactions between IGF loops of ClpX and binding pockets on ClpP heptamers allow docking of these symmetry-mismatched partners. Although IGF-ClpP contacts are highly dynamic in solution (Amor et al., 2016), they were well defined in our structures and revealed essential interactions.
Protein substrates were observed in the axial channel of ClpX in all structures and above the channel in two structures. The ClpX pore-1 and pore-2 loops were responsible for most substrate contacts in the channel, with a periodicity of two substrate residues per ClpX subunit. The RKH loops of ClpX, which we found play critical roles in substrate recognition and control of ATP-hydrolysis rates, also contacted substrate near the top and above the channel. Substrate contacts near the top of the channel are most important in determining substrate grip during unfolding (Bell et al., 2019), and substrate-ClpX contacts were tightest in this region in our structures. Finally, we observed ATPgS bound to five of the six ClpX subunits with ADP likely bound to the sixth subunit, and suggest how ClpX functions without a traditional sensor-I residue.
In contrast to our present cryo-EM structures, crystal structures of E. coli ClpX DN pseudohexamers (Glynn et al., 2009;Stinson et al., 2013) do not form spirals, bind only four nucleotides, and have conformations incompatible with ClpP and substrate binding. These observations highlight the fact that low-energy conformations observed by any structural method may or may not be relevant to biological function. A recent cryo-EM structure of Listeria monocytogenes ClpXP (Gatsogiannis et al., 2019), which was crosslinked to stabilize complexes, differs from our E. coli structures in having full-length ClpX bound to just the ClpP2 ring of heteromeric ClpP1/ClpP2, in ClpP2 having a narrower pore than ClpP in our structures, in not containing protein substrate, and in displaying head-to-head dimerization interactions between the N-terminal domains of two ClpX hexamers. Full-length E. coli ClpX is more aggregation prone than the DN variant, and negative-stain EM analysis of full-length E. coli ClpX bound to ClpP fails to support the importance of specific head-to-head dimerization between ClpX hexamers (Grimaud et al., 1998;Ortega et al., 2000;Ortega et al., 2002). Recent cryo-EM structures of Neisseria meningitidis ClpX DN /ClpP (Ripstein et al., 2020) are generally similar to those presented here but show fewer contacts between substrate and the pore-2 or RKH loops.

Proteolytic motors from different AAA+ clades have similar structures
AAA+ protease motors belong to either the classic or HCLR clades (Erzberger and Berger, 2006). For HCLR-clade members ClpX and Lon, the spiral hexamer architectures and pore-1-loop interactions with substrate in the axial channel are similar to those for the classic-clade YME1/FtsH and the proteasomal Rpt 1-6 /PAN motors (Puchades et al., 2017;de la Peña et al., 2018;Dong et al., 2019;Majumder et al., 2019;Shin et al., 2019). Thus, from a structural perspective, it is reasonable to suggest that motors from different clades may operate by a common fundamental mechanism. A sequential translocation model has been proposed for Yme1 and the Rpt 1-6 ring of the 26S proteasome, based on placing distinct cryo-EM structures with different nucleotide and substrateengagement states in a defined kinetic pathway, (Puchades et al., 2017;de la Peña et al., 2018;Dong et al., 2019). In this model, ATP hydrolysis in the fifth spiral subunit (subunit E in ClpX) drives a power stroke that moves each subunit from its previous location to a position offset by one subunit in the clockwise direction, generating a two-residue translocation step ( Figure 6A). Vps4 was proposed to remodel proteins by a similar mechanism, but with ATP hydrolysis in the D subunit driving sequential progression of subunits into new spiral positions . We refer to this general translocation model as SC/2R (Sequential Clockwise/2-Residue Step). Similar models have been proposed for Lon, for PAN/20S, and for the AAA+ protein-remodeling machines ClpB/ Hsp104, and CDC48/p97 (

Does ClpX function by an SC/2R mechanism?
The ClpX structures reported here resemble those of Yme1 and Rpt 1-6 in the spiral architecture of the hexamer, in the positions of subunits that contain a nucleoside triphosphate, in the interaction of successive pore-1 loops with two-residue segments of substrate, and in the patterns of substrate engaged and disengaged pore-1 and pore-2 loops in the ring (Puchades et al., 2017;de la Peña et al., 2018;Dong et al., 2019). Thus, based on structural considerations alone, ClpXP might also be expected to operate by an SC/2R mechanism. We cannot exclude this possibility. However, as discussed below, predictions of the SC/2R-model conflict with the results of multiple ClpXP experiments.
One major issue involves the length of translocation steps. The SC/2R model predicts a two-residue step. By contrast, the basic translocation step of ClpXP measured in optical-trapping experiments appears to be~6 residues, with some steps as much as fourfold larger (Aubin-Tam et al., 2011;Maillard et al., 2011;Sen et al., 2013;Cordova et al., 2014;Iosefson et al., 2015b;Rodriguez-Aliaga et al., 2016;Olivares et al., 2017). These single-molecule experiments measure average length changes in the portion of a protein substrate outside of the axial channel as a consequence of ClpXP unfolding (increased distance) or translocation (decreased distance). For example, ClpXP unfolding of filamin-A domains results in 14-19 nm distance increases (Aubin-Tam et al., 2011), with the distance being greater at higher force in accord with the worm-like chain model (Schlierf et al., 2007;Ferrer et al., 2008). As a filamin-A domain consists of~100 residues, 14-19 nm corresponds to 0.14-0.19 nm/res or~5-7 res/nm. Thus, the unfolded polypeptide outside the channel is partially compact, as a fully extended conformation would be 0.34 nm/res (~3 res/nm), similar to the extension of the substrate in the axial channel of our cryo-EM structures. During ClpXP translocation in the optical trap, distance decreases in steps of approximately 1, 2, 3, or 4 nm. These results indicate that the basic translocation step is~6 residues, with longer steps presumably representing kinetic bursts of this fundamental step.
A second important issue involves translocation kinetics. The sensitivity of optical trapping precludes direct identification and quantification of translocation steps as short as two residues, and thus it might be argued that~6 residue steps consist of three unresolved SC/2R sub-steps. Consideration of rates, however, makes this possibility unlikely. For example, the time from the beginning to the end of each translocation step is less than 0.1 s in the optical trap. Under similar conditions, the steady-state rate of ClpX DN /ClpP ATP hydrolysis is 3.6 ± 0.1 s -1 (Figure 6-figure supplement 1), corresponding to a time constant of 0.28 s. The SC/2R model predicts that each sub-step would require an independent ATP hydrolysis event, including ADP/P i dissociation and ATP rebinding, and thus that an average of~0.8 s would be required to take three sub-steps. This kinetic problem becomes worse for translocation bursts that move 12 to 24 residues in less than 0.1 s. Thus, ClpXP translocation occurs at rates~8 to~32 fold faster than predicted by the SC/2R mechanism and the experimentally determined rate of ATP hydrolysis. Another issue sometimes raised is that ClpX step sizes could be longer under tension in optical-trapping experiments, or that ClpXP uses a different translocation mechanism under these conditions. We consider both possibilities unlikely as the distribution of translocation step lengths shows little dependence on trap force. Moreover, ClpXP unfolding and translocation rates measured by optical trapping are similar to those determined in singlemolecule fluorescence assays (Shin et al., 2009).
A third issue involves the prediction of the SC/2R model that ATP hydrolysis occurs at only one position in the ClpX spiral during translocation. If this prediction were correct, then single-chain ClpX variants containing subunits that cannot hydrolyze ATP should stall when an ATPase inactive subunit moves into the hydrolysis position. However, ClpX DN hexamers with just two ATPase-active subunits support ClpP degradation at~30% of the wild-type rate and with the same thermodynamic efficiency as wild-type ClpXP (Martin et al., 2005). Could thermal motions move an ATPase-inactive subunit out of the hydrolysis position in the ClpX spiral and an active subunit into it, allowing continued function? This possibility seems unlikely, as the variant with just two ATPase active subunits translocates substrates against force in optical-trap experiments (Cordova et al., 2014). As we discuss below, non-SC/2R models that permit probabilistic hydrolysis by subunits at different positions in the spiral could explain how stalling is prevented when multiple ClpX subunits are hydrolytically inactive and could also provide a mechanism for rapid bursts of translocation steps.

Structure-based translocation models for ClpXP
To account for the experimentally determined translocation properties of ClpXP, any model needs to explain: (1) how structural changes in the spiral result in a fundamental translocation step of~6 residues; (2) how kinetic bursts could generate very fast translocation of as many as~24 residues without requiring multiple ADPdissociation and ATP-rebinding events; and (3) how the ClpX motor functions efficiently without strict requirements for ATP hydrolysis in any specific subunit in the spiral.
Modification of the SC/2R model could, in principle, address the step-size issue. For example, upon ATP hydrolysis in a unique subunit in the spiral, a clockwise three-subunit shift could Video 3. Transition between class-2 and class-4 structures of ClpX DN /ClpP. A conformational morph between the class-2 and class-4 structures of ClpXP is shown. To generate this morph, ClpX subunits in the class-2 structure were renamed from ABCDEF to FABCDE and aligned to the class-4 ClpXP structure in two steps to maximize overlap between the ClpP portions of the two structures. The ADP-bound ClpX subunit (colored aquamarine) moves most dramatically, from the top to the bottom of the ClpX spiral. In the ADP-bound subunit, residues of the pore-1 loop move~1.6-1.9 nm. https://elifesciences.org/articles/52774#video3 generate a six-residue power stroke in a SC/6R model ( Figure 6B). Potential issues with this mechanism include that substrate contacts with multiple subunits would need to break during each translocation step, the structural features that would drive a three-subunit rearrangement are unclear, and this SC/6R model fails to account for kinetic bursts resulting in longer translocation steps.
To account for single~6 residue translocation steps and longer kinetic bursts, we propose a Probabilistic Anti-clockwise Long-Step (PA/LS) model ( Figure 6C and D), which has two major components. First, a power stroke in the anti-clockwise direction moves the top subunit in the ClpX spiral to the bottom position, thereby translocating substrate by~1.7 nm. A conformational change of this type is captured in a morph in which subunit A in class two is aligned with subunit F in class 4 (Video 3); a similar transition has been considered for Vps4 . We assume that ATPhydrolysis in subunit A of ClpX powers this step, although this is not critical for the general model. Repeating this cycle sequentially would account for~6 residue steps, assuming that the polypeptide is translocated in an extended conformation, but not for bursts of longer steps. Thus, we also posit that initial ATP hydrolysis could occur with some probability in ClpX subunits B through E, with these events causing strain in the spiral that is released in a burst of power strokes when ATP in subunit A is eventually hydrolyzed or sufficient strain accumulates. By allowing coupling of multiple ATP-hydrolysis events prior to translocation, this model obviates the need to release ADP and rebind ATP prior to each step, allowing fast kinetic bursts. Moreover, from a structural perspective, five of the six nucleotide-binding sites in our ClpX structures have similar geometries in at least one structural class, and thus could plausibly hydrolyze ATP.
The PA/LS model implies that subunit A can bind substrate at the top of the spiral and then drag or push it to the bottom. In this regard, we note that the pore-1 loop of subunit-A contacts substrate in our class-4 structure, with these contacts buttressed by interactions with nearby RKH loops. This model would also require transient breaking of contacts between the substrate and other spiral subunits during the power stroke, but this could be done in a zipper-like fashion to reduce kinetic barriers.
By allowing ATP hydrolysis at multiple spiral positions, probabilistic variations of the SC/2R and SC/6R models (PC/2R and PC/6R) can also be formulated, which would allow the possibility of kinetic bursts. We note that probabilistic models do not preclude neighboring subunits in the ring from firing sequentially, but rather exclude any mechanistic requirement for sequential action. It has been argued that probabilistic or stochastic models imply that subunits in the AAA+ ring must be completely independent (Smith et al., 2011), but this is not true. Communication between ClpX subunits and cooperative ATP hydrolysis has been documented (Hersch et al., 2005;Martin et al., 2005) and is compatible with probabilistic models.
It is not clear how many structural changes are involved in the complete reaction cycle for ClpXP or related AAA+ ATPases. For example, crystal structures of ClpX reveal rotations between the large and small AAA+ domains so large that nucleotide cannot bind some subunits (Glynn et al., 2009;Stinson et al., 2013). Although subunits of this type are not present in current cryo-EM structures, they might represent transient functional conformations, as crosslinks engineered to trap these hexamer conformations form in solution and prevent ClpXP degradation but not ATP hydrolysis (Stinson et al., 2013;Stinson et al., 2015). Additional substrate-bound conformations of ClpXP probably also await discovery, as it is unclear from current structures how ClpX initially recognizes substrate degrons. Moreover, ClpXP degrades disulfide-bonded and knotted proteins in reactions that require simultaneous translocation of two or more polypeptides (Burton et al., 2001;Bolon et al., 2004;San Martín et al., 2017;Sivertsson et al., 2019). In our cryo-EM structures, a single polypeptide strand fills the axial channel at its narrowest point, which would have to expand to accommodate multiple strands during translocation. Vps4 accommodates two polypeptides in its axial channel without major distortion of its hexameric spiral . A similar mechanism might allow ClpX to translocate two polypeptide chains, although it is difficult to imagine accommodation of three or more chains. Finally, we note that the SC/2R mechanism requires substrates to be translocated in an extended conformation, whereas ClpXP translocates poly-proline tracts that are unable to form this structure (Barkow et al., 2009). Other AAA+ unfolding/remodeling machines and SC/2R mechanisms ClpAP takes~1 nm and~2 nm residue translocation steps Olivares et al., 2017), suggesting that it can also operate by a non-SC/2R mechanism, but we are unaware of experiments that establish the translocation step size for other AAA+ protein unfolding or remodeling enzymes. Like ClpX, some of these enzymes do not stall if an ATPase-inactive subunit occupies the hydrolysis position. In the Rpt 1-6 hexamer of the yeast 26S proteasome, for example, ATPasedefective Rpt 3 , Rpt 4 , or Rpt 6 subunits cause virtually complete loss of degradation activity, whereas ATPase-defective Rpt 1 , Rpt 2 , and Rpt 5 subunits do not (Beckwith et al., 2013). In the alternating Yta10/Yta12 hetero-hexamer (a Yme1 homolog), a Walker-B mutation in Yta12 prevents ATP hydrolysis in Yta10, as expected for tight subunit-subunit coupling in a strictly sequential mechanism, but a Walker-B mutation in Yta10 does not prevent ATP hydrolysis in Yta12 (Augustin et al., 2009). Given the strong similarities between the structures and substrate interactions of a large number of AAA+ unfolding and remodeling machines and the fact that the SC/2R mechanism does not account for ClpX and ClpA experimental results, it will be important to continue to use structural and functional studies to investigate the molecular mechanisms by which these AAA+ machines function.

Protein expression and purification
The single-chain ClpX variant used for microscopy contained six copies of E. coli E185Q ClpX DN (residues 62-424), with neighboring units connected by six-residue linkers of variable composition, and a C-terminal TEV cleavage site and His 6 tag. The E. coli ClpP variant consisted of an N-terminal propeptide, which was removed auto-proteolytically during expression/purification, residues 1-193 of mature ClpP, and the TEV-His 6 epitope. Both proteins were expressed separately in E. coli strain ER2566 and purified as described (Stinson et al., 2013). After purification, TEV protease was used to remove the His 6 tags. TEV protease and uncleaved proteins were removed by Ni 2+ -NTA affinity chromatography, and purified proteins were flash frozen in storage buffer (20 mM HEPES, pH 7.5, 300 mM KCl, 0.5 mM EDTA, 10% glycerol) and stored at -80˚C.

Sample preparation, data acquisition, and image processing
To assemble complexes, the ClpX DN pseudohexamer and ClpP 14 were diluted into EM buffer (20 mM HEPES, pH 7.5, 100 mM KCl, 25 mM MgCl 2 , 5 mM ATPgS) to final concentrations of 4 mM and 1.8 mM, respectively. After 5 min at 25˚C, 3 mL of the mixture was applied to glow discharged R1.2/ 1.3 400 mesh grids (Quantifoil). Grids were blotted with filter paper 494 (VWR) and plunged into liquid ethane using a Cryoplunge-3 system (Gatan). Electron micrographs were collected using a Talos Arctica with a Gatan K2-Summit direct electron detector in super-resolution mode. High-resolution movies were recorded at a magnification of 36000X (0.58 Å pixel size). Each movie was composed of fifty frames (200 ms per frame) and a total dose of~58 e -/Å 2 per movie. The final dataset consisted of 3657 movies recorded in two separate sessions. Frames in each movie were 2X binned, aligned, gain-corrected, and dose-weighted using Motioncor2 (Zheng et al., 2017), to generate a single micrograph. The contrast transfer function (CTF) was estimated using CTFFIND4 (Rohou and Grigorieff, 2015). Unless noted, Relion 2.0 (Kimanius et al., 2016) was used for 2D/3D classification and refinement. We first attempted to construct a density map of doubly capped ClpX-ClpP-ClpX using standard protocols. 1.4 million doubly capped particles were automatically picked and filtered by 2D classification. 443,717 'good' particles were selected for 3D map reconstruction. To generate an initial model for 3D refinement, the crystal structures of a ClpX DN hexamer (PDB 3HWS; Glynn et al., 2009) and a ClpP tetradecamer (PDB code 3MT6; Li et al., 2010) were merged in PyMOL and lowpass filtered to 40 Å . 3D refinement with C 2 symmetry yielded a map with a resolution of~4 Å , but the quality of this map was poor and interpretation of secondary structure elements was impossible. Using C 1 or C 7 symmetry did not improve the map.
To minimize problems caused by the symmetry mismatch between ClpX and ClpP, we treated ClpX and ClpP separately before the last step of refinement. For the ClpP reconstruction, we applied a soft circular mask, including ClpP and the tips of the ClpX IGF loops, to the 3D reference and 2D images, respectively. Because we observed predominantly side views of ClpXP (perpendicular to the axial channel; Figure 1-figure supplement 1A-C), this simple masking of the particle images allowed us to remove most of the ClpX density and to focus particle alignment on ClpP. Starting from a low-pass filtered ClpP structure as an initial reference, 3D refinement with D 7 symmetry yielded a 3.2 Å resolution map with clear secondary structure and side-chain features. For the ClpX reconstruction, we extracted ClpX sub-particles from both ends of ClpP based on the previously determined ClpP alignment using a python script and IMOD (Mastronarde and Held, 2017). We prepared an original and a ClpP-signal-subtracted ClpX particle stack. To remove misaligned particles, the ClpX stacks were 2D classified without alignment. Particles from 2D classes that showed clear secondary structure were used in subsequent 3D-classification and reconstruction steps.
Clean ClpX sub-particles were 3D classified (K = 6) and refined, resulting in four distinct ClpX classes with resolutions ranging from 3.9 to 4.2 Å . To recover the ClpX DN -ClpP interface, we reextracted ClpXP sub-particles using a larger box that included the cis ClpP ring. Alignment and classification of each ClpX DN sub-particle was transferred to corresponding ClpX DN /ClpP sub-particle and four classes were refined with local alignment optimization, resulting in four ClpX DN /ClpP maps (resolutions 4.0 to 4.3 Å ).
To test the robustness of this workflow, we performed 3D classification of ClpX sub-particles multiple times using K = 4 or K = 8. The quality of maps suffered slightly, but the overall structures of four predominant classes of ClpX hexamers remained unchanged.

Model building and refinement
A ClpP tetradecamer (PDB 3MT6; Li et al., 2010) was docked into the D 7 map using Chimera's 'fit to map' function (Pettersen et al., 2004). For ClpX DN , six copies of the large and small AAA+ domains (PDB 3HWS; Glynn et al., 2009) were docked into the map sequentially and refined for three iterations in Chimera. Real-space refinement of docked ClpP and ClpX was performed using PHENIX (Adams et al., 2010), and model building was performed using Coot (Emsley and Cowtan, 2004). ChimeraX (Cordova et al., 2014), Chimera (Pettersen et al., 2004), and PyMOL were used to create figures and movies.

Biochemical assays
Assays were conducted at 37˚C in PD buffer (25 mM HEPES-KOH, pH 7.5, 5 mM MgCl 2 , 200 mM KCl, 10% glycerol). Experiments were performed in triplicate and reported values are averages ± SD. Degradation of CP7 GFP-ssrA (15 mM) by ClpX DN (0.3 mM hexamer) and ClpP 14 (0.9 mM) was assayed in the presence of 5 mM ATP and was monitored by loss of fluorescent signal (excitation 467 nm; emission 511 nm). For RKH loop mutants with reduced substrate affinity, degradation was measured at high concentrations of fluorescent substrate, with excitation at an off-peak wavelength (excitation 420 nm; emission 511 nm). Degradation of fluorescein-labeled Arc-st11-ssrA by ClpX DN (0.3 mM hexamer) and ClpP 14 (0.9 mM) was assayed as described (Bell et al., 2018). K M and V max were determined by fitting the average values of replicates to a hyperbolic equation. ATP hydrolysis rates were measured using a coupled-NADH oxidation assay as described (Martin et al., 2005), using ClpX DN (0.3 mM hexamer), with or without ClpP 14 (0.9 mM), and 5 mM ATP. Activation of decapeptide cleavage by ClpP or variants was performed as described (Lee et al., 2010a), using ClpP 14 (50 nM), RseA decapeptide (15 mM), ATP (5 mM) and a regeneration system, and either ClpX DN (0.5 mM hexamer) or ADEB-2B (100 mM), which was a generous gift from J. Sello (Brown). Pull-down experiments were performed in 25 mM HEPES-KOH (pH 7.5), 5 mM MgCl 2 , 150 mM KCl, 20 mM imidazole, 10% glycerol, 500 mM dithiothreitol and 2 mM ADP or ATPgS. 40 mL of a mixture of ClpX DN C169S (1 mM hexamer concentration) and ClpP (1 mM 14-mer concentration) were mixed, incubated for 10 min at 30˚C, and then added to 20 mL of Ni 2+ -NTA resin (Thermo Scientific) equilibrated in the same buffer. Binding reactions were incubated for 15 min at room temperature with rotation, centrifuged for 1 min at 9400 x g, and the supernatant was discarded. Reactions were washed, centrifuged, and the supernatant discarded three times. Bound protein was eluted with 40 mL of buffer supplemented with 300 mM imidazole for 15 min with rotation. Reactions were then centrifuged again and the eluant collected. Input and elution samples for each reaction were resolved by SDS-PAGE on a 10% Bis-Tris/MES gel run at 150V and visualized by staining with Coomassie Brilliant Blue R250. Results were validated in three independent replicates. The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.