Fast dynamics shape the function of the AAA+ machine ClpB: lessons from single‐molecule FRET spectroscopy

It has been recently shown that in some proteins, tertiary‐structure dynamics occur surprisingly fast, that is on the microsecond or sub‐millisecond time scales. In this State of the Art Review, we discuss how such ultrafast domain motions relate to the function of caseinolytic peptidase B (ClpB), a AAA+ disaggregation machine. ClpB is a large hexameric protein that collaborates with cellular chaperone machinery to rescue protein chains from aggregates. We used single‐molecule FRET spectroscopy to capture the dynamics of essential structural elements within this machine. It was found that the middle domain of ClpB, known to act as its activator, toggles between two states much faster than the overall activity cycle of the protein, suggesting a novel mode of continuous, tunable switching. Motions of the N‐terminal domain were observed to restrict the conformational space of the M domain in the absence of a substrate protein, thereby preventing it from tilting and spuriously activating ClpB. Finally, microsecond dynamics of pore loops responsible for substrate pulling through ClpB's central channel, together with their response to specific perturbations, point to a Brownian‐ratchet mechanism for protein translocation. Based on our findings, we propose a two‐time‐scale model for the activity of ClpB, in which fast conformational dynamics affect slower functional steps, determined by ATP hydrolysis time. Future work on this and other proteins is likely to shed further light on the role of ultrafast dynamics on protein function.


Introduction
Proteins often function as machines, consuming energy in order to proceed through a well-defined set of intermediate states that lead to the execution of a particular function. The proper operation of such protein machines, particularly in relation to the function of other proteins in the cell, requires intricate regulation. Such regulation is achieved through the operation of allosteric effectors, namely biomacromolecules or small molecules that bind specifically to a machine and modulate its activity. The concept of allostery was introduced 60 years ago by Monod and coworkers [1], and the various mechanisms involved in its mediation are still under intense scrutiny. Classically, allostery was considered to be mediated through conformational changes within a protein [1]. More recently, the modulation of protein dynamics in the absence of conformational changes was proposed as an additional allosteric mechanism [2]. In either case, the outcome is the coupling of two remote sites on a protein and the unlocking of elaborate functional control processes. The elucidation of these control mechanisms is essential for our understanding of the functional cycles of protein machines.
An important and remarkable property of proteins, already alluded to above, is the availability of internal motions on multiple time scale. A growing body of scientific work [3] suggests that large-scale motions of tertiary-structure elements (e.g. domains) within proteins may occur on the microsecond or sub-millisecond time scales. This is important as these time scales are much faster than typical functional cycles of protein machines. The implication is that there is no one-to-one coupling of the dynamics to the function. Rather, a more complex relation that transduces fast motions into effectors of slower functions must exist. The ability to observe fast domain motions rests on the introduction of new experimental techniques in recent years, from NMR fast relaxation methods [4] to single-molecule FRET (smFRET) techniques [5]. In this State of the Art Review, we focus on our application of the latter to the disaggregation machine caseinolytic peptidase B (ClpB), which serves as a wonderful test case for the observation of fast dynamics and their importance for functional mechanisms. We put our studies in the context of the rich literature on the structure and function of this protein and others belonging to the same family.

ClpB structure and function
Hsp104 of yeast and ClpB of bacteria are homologous ATP-dependent molecular chaperones belonging to the Clp/Hsp100 subfamily of the AAA+ (ATPases associated with various cellular activities) superfamily [6][7][8][9]. They are categorized as type II ATPases, as each of their subunits contains two ATPase domains, as opposed to type I proteins, such as ClpX, whose subunits contain only a single ATPase domain [10]. Their disaggregation activity is driven by ATP consumption and is conducted in cooperation with the co-chaperones Hsp70/DnaK and Hsp40/DnaJ [11,12]. Hsp104 and ClpB form a homohexameric ring structure in a nucleotide-dependent manner [12]. The monomer structure of the ClpB from Thermus thermophilus (TT), which we study, was solved early on in complex with the ATP analogue AMP-PNP using X-ray crystallography [13], and was observed more recently in cryo-electron microscopy (cryo-EM) structures (e.g. [14][15][16][17]   The NTD is distantly related to that of other Clp/ Hsp100 members and is suggested to be very dynamic and flexible since it shows different orientations in the crystal structure, and it is frequently undetectable in cryo-EM reconstructions [14][15][16][17]. Some recent studies did manage to capture partial configurations of the NTD which appeared to be more mobile than other parts of the machine [18,19]. The NTD has long been assumed unnecessary for ClpB function based on in vivo and in vitro studies [20][21][22][23]. Interestingly enough, a truncated NTD version of ClpB was found to exist physiologically in Escherichia coli as a mixture with the full length protein [24] and to function similarly to the full-length version. However, some studies found the NTD to be essential for cooperative substrate handling and for enhancing the disaggregase activity [25,26], while others showed it to serve a regulatory role in protein disaggregation [27,28], a role also supported by studies of the NTD of ClpA [29,30]. It was also suggested that the function of the NTD is substrate dependent [31]. Recently, smFRET spectroscopy with the ability to follow domain motions in real time revealed that the NTD acts as a regulatory entity while exhibiting ultrafast (microsecond) dynamics, as will be further discussed in this review [32].
The two tandem Walker-type ATP binding domains (AAA+ domains) [8], termed NBD1 and NBD2, serve as the central building blocks for oligomerization and cavity formation and are considered to be the energy source for machine activity, being responsible for ATP binding and hydrolysis. Mutational studies in Hsp104 showed that NBD1 is mainly involved ATP hydrolysis while NBD2 is also responsible for oligomerization [33], though other studies suggested that in ClpB and other type II AAA+ machines NBD2 is the main contributor to the machine activity and that both NBDs contribute to protein assembly [22,34].
The M domain (or coiled-coil domain) is rather unique to the substrate-translocating chaperone family ClpB/Hsp104; other Clp/Hsp100 proteins, such as ClpA, have no M domain and ClpC has a shorter one [35][36][37]. The 85 A-long M domain is inserted into NBD1 and consists of two antiparallel coiled coils, thus resembling a 'two-blade propeller' [13] (Fig. 1A). It was found that the M domain plays an important role in the disaggregation process [38][39][40]. The M domain was also found to be pivotal to the interaction with the adaptor protein DnaK, which recruits ClpB molecules to aggregate structures in order to initiate the disaggregation activity [39,[41][42][43][44]. Based on these studies, it has been suggested that the M domain acts as a regulatory switch that alternates between active and inactive conformations. When the M domain is in its active conformation, it can bind DnaK, leading to recruitment of ClpB to aggregates, while in its inactive conformation Dnak cannot bind, and therefore disaggregation is repressed [15,18,38,39]. The M domain was only partially resolved in Hsp104 cryo-EM structures [45,46], as well as in ClpB structures [16], indicating that this domain is flexible and mobile, and may adopt multiple conformations [13,14,[47][48][49][50][51], as supported also recently by hydrogen-exchange mass spectrometry measurements [52]. Significantly, it was proposed that this mobility is essential for disaggregation function [13,49].
The cryo-EM structure of Yokom et al. [45] at 5. 6 A showed that Hsp104 in complex with AMP-PNP forms a left-handed two-turn spiral asymmetric hexameric structure with a shift of 44 A between protomers 1 and 6, resulting in an interaction between NBD2 and NBD1 of these seam protomers. Interestingly, the spiral architecture was found to form a large opening pore along the hexamer (thus termed the 'open' conformation), and the pore loops (which are not fully resolved) arrange in a staircase form, proposed as a way to optimize and guide polypeptide threading across the channel [45]. A higher resolution cryo-EM study of Hsp104 in complex with ATPcS and the polypeptide casein as a substrate, showed that Hsp104 adopts two different right handed asymmetric conformations, 'closed' and 'extended' [46]. The two conformations mainly differ in their seam promoters, namely protomer 1 and protomer 6. In the closed state, the pore loops of protomers 1-5 directly contact the substrate via tyrosine residues, making contact with almost every second amino acid [46]. However, in the extended state, protomer 6 rotates towards the axial channel and contacts the substrate directly, whereas protomer 1 rotates counterclockwise forming a new interaction between its NBD2 and the NBD2 of protomer 2. The cycle involving these two states was suggested to generate a mechanism for sequential threading the substrate along the channel [46]. A similar switching mechanism of the hexameric structure from left-handed open conformation to right-handed closed conformation was also observed in a study of ClpB [16]. Recent reconstructions of a catalytically inactive Hsp104 in complex with ATP confirm the observed open, closed and extended conformations [54]. However, these conformations were obtained even without casein as a substrate, suggesting that they exist in equilibrium, and that the massive change between the closed and extended conformations is thermal rather than ATP hydrolysis-driven [54]. This conclusion was also supported by a study of the AAA+ protein NSF [60]. Asymmetric right-handed hexameric structure was also observed in E. coli ClpB in complex with ATPcS [18]. However, in contrast to Hsp104 [45], the ATPcS-bound state did not form a continuous spiral arrangement, but rather a flat asymmetric hexamer with canonical interface interactions. A 60°rotation of the NBD1 ring with respect to NBD2 placed NBD2 of one promoter below the NBD1 of adjacent protomer, forming a different pore-loop arrangement [18].

Mechanism of disaggregation
Several modes of disaggregation by ClpB have been proposed in the literature, such as the 'crowbar' mechanism, subunit exchange mechanism and the threading mechanism [63]. So far, no clear evidence for the first two of these was provided, whereas several biochemical and structural studies of the pore loops of NBD1 and NBD2 supported the threading mechanism [17,19,[64][65][66][67][68][69], which was also supported by the studies of other AAA+ members such as ClpA [70], NSF [60], CdC48 [61,71] and its homologue VAT [58]. Interestingly, some studies indicated that ClpB or Hsp104 perform partial threading rather than complete threading of soluble substrate, a mechanism that might allow the disaggregases to dissociate and bind to other regions of a polypeptide [66,72,73]. Another recent study suggested that Hsp104 can also act as a 'holdase', capturing soluble forms of amyloid substrates, a function which is different from its well-established disaggregation activity [74]. The threading mechanism of protein substrates is facilitated by pore loops along the axial channel of the machine. Pore loops exist and are conserved in almost all AAA+ machines, especially unfoldases, and perform the most important and complicated function of these machines by pulling client proteins through their central channels [8,12,[75][76][77]. Protein members of the AAA+ type I family usually contain a conserved aromatic hydrophobic motif 'GYVG' in their pore loop, which is suggested to be responsible for the binding of client proteins [8,78]. A pore loop carrying a similar motif also exists in the AAA+ type II family members, mainly in NBD2 rather than in NBD1, and it is known as pore loop 3 ( Fig. 1C) [64,79,80]. Biochemical and single-molecule force spectroscopy studies of pore loops in ClpX [79,[81][82][83][84], and other studies on FtsH [85], ClpA [70], Hsp104 and ClpB [64,65,[67][68][69][86][87][88], showed that the conserved tyrosine residue is crucial for polypeptide pulling and for the activity of the machine.
In addition to pore loop 3 (PL3), AAA+ type II members contain two pore loops in their NBD1 domains, known as pore loop 1 (PL1) and pore loop 2 (PL2, Fig. 1C) [19,70,80,89]. PL1 is very well studied structurally and biochemically in ClpB, Hsp104 [35,65,80,88,89], ClpA [70] and NSF [60]. Interestingly, this pore loop contains a conserved motif 'AGAKYR' which also harbours a tyrosine amino acid [80]. While mutation of the tyrosine in pore loop 3 seems to cause a complete loss of the machine activity, a variant with the same mutation in PL1 still maintains some activity [69,70,89]. Binding measurements showed that substrate engagement occurs mainly with a functional pore loop 1, suggesting that it is responsible for the initial interaction with the substrate and translocation [69,70,89].
In contrast to PL1 and PL3, PL2 is less well studied. Interestingly, it contains a conserved 'GAG' motif in AAA+ type II prokaryotic proteins and a similar 'GxG' motif in the eukaryotic variants (x refers to any amino acid) [80]. A study of PL2 in ClpA showed that a single point mutation at position 293 from alanine to aspartate led to the loss of machine activity [70]. A more recent study on Hsp104 by Lee et al. [89] showed that this pore loop is insensitive to single point mutations, and requires as many as four mutations to abolish machine activity. Pore loop 2 was observed in two conformations, up and down, which together with its role in machine activity suggested that it is responsible for substrate pulling and translocation across the axial channel of the machine [89].
The coupling between the pore loops within the protomers in AAA+ type II machines was studied in ClpA, ClpB and Hsp104 [69,70,89]. These studies suggested that the role of PL1 is mainly in substrate engagement, while PL2 and PL3 apply force to pull the protein across the axial channel. Moreover, abolishing the function of any one out of three pore loops led to loss of disaggregation function, suggesting a non-overlapping function of the pore loops, and that all of them are essential for substrate coordination and translocation [69,70,89]. Subunit mixing of PL1 mutants with non-mutants suggested a strong cooperativity between the protomers in processing both soluble and aggregated substrates.
Recently, multiple high-resolution cryo-EM structural studies reported on the conformational changes of the pore loops that occur upon substrate binding in various AAA+ protein members [19,[57][58][59][60][61]71,90]. Interestingly, most of these studies described a staircase arrangement of the pore loops along the central channel of the machine, suggesting polypeptide threading in a hand-overhand manner, facilitated by rigid-body movement of the protomers with a step size equivalent to two amino acids per ATP hydrolysis cycle. Furthermore, the fact that they observed partial nucleotide occupancy in the NBDs ruled out a concerted ATPase activity mechanism. Instead, it suggested a sequential mechanism of ATP hydrolysis, which in turn governs the engagement of the seam protomers with the substrate in the hand-over-hand translocation mechanism [17,19,46,[57][58][59][60][61]71,90].
Some recent studies challenged the hand-over-hand mechanism of substrate pulling by ClpB and similar machines. In particular, a recent single-molecule force spectroscopy study showed that translocation by ClpB occurs at the impressive rate of $ 240-450 amino acids per second, and demonstrated bursts of 14-28 amino acids [73]. Studies on ClpX [91] and ClpA [92] also identified substrate translocation with large step sizes. The measured rates of translocation are much larger than expected based on the hand-over-hand mechanism, in which (the slow) ATP hydrolysis should be the limiting factor. Surprisingly, it was also found that these machines were active in translocation even when several of their six subunits were rendered inactive [81,93], a finding that contradicts the sequential nature of hand-over-hand activity. Atomic force spectroscopy experiments on the histone chaperone Abo1, another AAA+ machine, also showed that ATPase activity is probabilistic rather than sequential [94]. The strong disagreement between models based on structural studies and the results of real-time measurements clearly calls for further studies to clarify the nature of the function-related dynamics in these machines.
Our recent smFRET work on ClpB, discussed below, was geared to study and attempt to answer some of the conundrums introduced above. We demonstrated in several ways the involvement of ultrafast large-scale domain motions in ClpB function, showing how fast dynamics create new and unique regulation pathways that otherwise could not be allowed. Our studies shed new light on the activity of the M domain [95], on the role of the pore loops in substrate threading [96] and on the important function of the NTD as an entropic inhibitor of the M domain [32]. Each section below is dedicated to one of these questions.

The M domain is a tunable allosteric switch
The M domain samples its conformational states on the microsecond time scale As noted in the Introduction, the M domain in ClpB is considered to act as the activation switch of the whole machine. In order to decipher its conformational dynamics using smFRET spectroscopy, we prepared three constructs, each with one fluorescent probe on the M domain and another on a fixed position on NBD2 [95]. Double-labelled variants were assembled in such a way that there would be essentially a single subunit within each ClpB hexamer that contains the FRET pair. To this end, ClpB monomers were reassembled by mixing labelled subunits with cysteine-less (unlabelled) subunits at a ratio of 1 : 100. The distribution of labelled vs. unlabelled protomers in mixed ClpB molecules was calculated using a binomial distribution [95]. Based on this ratio, the probability to find one labelled protomer in a hexamer was 5.7%, while the probability of incorporation of two labelled promoters in the same hexamer was expected to be 0.15% only. Thus, the vast majority of the complexes we studied contained only a single labelled subunit.
Labelled ClpB hexamers were monitored in the presence of 2 mM ATP, a condition that ensures that they are correctly assembled and functional. In each smFRET experiment, freely diffusing molecules of labelled ClpB emitted bursts of photons as they passed through a focused laser beam, and the arrival time and colour of each photon was registered in the donor and acceptor detection channels. FRET efficiency histograms were then constructed from the bursts collected from each variant, as seen in Fig. 2A, which shows the histogram of the variant S428-S771. This broadened FRET histogram suggests immediately the existence of two or more conformations that undergo fast exchange dynamics. A closer look at binned photon trajectories originating from bursts ( Fig. 2B) clearly shows FRET efficiency fluctuations, reinforcing this observation. H 2 MM, a powerful photon-byphoton hidden Markov model algorithm developed recently in our lab [97], was used to extract information on fast dynamics of the M domain. Our analysis optimizes model parameters based on a large number of burst-related photon trajectories, typically between 5000 and 10 000 per data set. Using H 2 MM, we found that the minimal number of states required to fit our single-molecule trajectories is three: two major populations at FRET efficiencies of 0.8 AE 0.01 (state 1) and 0.47 AE 0.01 (state 2), and a minor population at a FRET efficiency of 0.15 AE 0.01 (state 3), as marked by the arrows on Fig. 1A. The relative populations of these states were 0.43 AE 0.01, 0.42 AE 0.01 and 0.15 AE 0.01, respectively. Interestingly, the three states were found to exchange in a sequential manner, with high transition rates between states 1 and 2, and lower transition rates between states 2 and 3. Repeating the

Effect of nucleotide binding on M-domain dynamics
A Walker A mutation in NBD1 (A À A + ), which abolished ATPase activity in this domain, led to a shift of the FRET efficiency histogram to higher values compared to the WT (Fig. 2C). This shift arose from a reduction of the transition rate from the active conformation (state 1) to the inactive conformation (state 2) of the M domain, without any change in the transition rate in the reverse direction [95]. The change in transition rates led to an increase in the active/inactive state ratio, from 1.00 AE 0.01 in the WT to 1.63 AE 0.02. The A À A + mutant showed only a $ 30% disaggregation activity and a weak ATPase activity, which was somewhat enhanced upon binding to the substrate j-casein.
In contrast, an A + A À mutant with abolished ATP binding to NBD2, showed a lower active/inactive state ratio than the WT, a very weak ATPase activity that was not enhanced by j-casein binding and a lower disaggregation rate. These results show that nucleotide binding has a significant allosteric effect on M-domain dynamics: nucleotide binding to NBD2 stabilizes the M domain in the active state, whereas nucleotide binding to NBD1 stabilizes the inactive state. Coupling of the M-domain structural changes to nucleotide binding to the NBDs has been also observed in a recent study by Sugita et al. [98], who reported that ATP binding to NBD1 shifts the M domain to a tilted conformation. Our smFRET studies clearly demonstrate that this coupling is exerted through an effect on Mdomain dynamics. Together, the two NBDs regulate M-domain dynamics as follows: at a relatively low ATP concentration, where NBD2 binds the nucleotide better than NBD1, the M domain is preferentially in its active state, while at a higher ATP concentration NBD1 also binds ATP well, thus pushing the M domain to its inactive state. Indeed, we found that due to these two contrasting effects, at saturating ATP concentrations, the M domain in WT ClpB spends $ 50% of the time in each of the two states. DnaK is the main component of the co-chaperone system in the disaggregation process, and binds to the M domain at motif 2 [42]. smFRET measurements of M-domain dynamics showed that the fast exchange between states 1 and 2 was retained in the presence of DnaK. However, the population ratio of states 1 and 2 was found to increase with the co-chaperone concentration, suggesting that DnaK binding does not simply stabilize the M domain in its active state, as previously suggested [38,39], but rather changes the dynamic equilibrium between the active and inactive states in favour of the former. Studying M-domain dynamics in the presence of the soluble substrate j-casein and aggregated substrates such as G6PDH showed that, while k 12 decreased as a function of substrate concentration, k 21 increased, leading to the accumulation of active state population (Fig. 2D). At the same time, the ATPase activity of ClpB increased significantly. This enhancement is likely due to the relative increase of the M-domain active state population.
Further smFRET studies pointed to the major role that the M domain plays through its ultrafast (microsecond) dynamics. In particular, substrate binding to ClpB, which involves binding sites on the NTD [28] and also at the central pore of NBD1 [88], exerts an allosteric effect that shifts the M-domain dynamics towards the active conformation, enhancing the probability of DnaK binding to initiate the disaggregation activity (Fig. 2E). Remarkably, this is done while the M domain continues to sample both active and inactive states, in fact at an increasing rate compared to the substrate-unbound state. Overall, it is found that, rather than a static population of the active and inactive states of the M domain, a dynamic equilibrium between these two states is used as a signal to activate or repress the machine. Even though the M domain continues to sample two discrete states, the fact that it toggles $ 4-5 orders of magnitude faster than the whole machine activation cycle, enables it to act as a continuous rather than an on-off switch. This behaviour facilitates a positive feedback mechanism, in which ATP binding to NBD1 or NBD2 allosterically regulates M-domain dynamics while M-domain dynamics in turn regulate and enhance NBD activity.

Brownian-ratchet mechanism for substrate translocation
The three pore loops in ClpB manifest microsecond dynamics As discussed in the Introduction, recent structural studies suggested a universal hand-over-hand protein translocation mechanism, in which pore loops are moving rigidly in tandem with their corresponding subunits, though functional and biophysical studies are in discord with this model. Using smFRET spectroscopy, we probed the real-time dynamics of the pore loops of ClpB and their response to substrate binding. It was found that all pore loops undergo large-amplitude fluctuations on the microsecond time scale, and change their conformation upon interaction with substrate proteins [96], a result confirmed in molecular dynamics simulations carried out by Stan and coworkers [99]. To probe pore-loop motions along the axial channel using smFRET, each pore loop was labelled, one at a time, together with a reference point located at the centre of the vertical axis of the ClpB protomer. As in M-domain experiments, a variant labelled with donor and acceptor dyes was constructed and assembled such that only a single subunit within each ClpB hexamer contained the fluorescent probes. In the presence of 2 mM ATP, FRET efficiency histograms of the variant labelled on PL1 showed a major population at a value of 0.65 AE 0.01. This major peak was again found to be much broader than expected based on shot noise, indicating dynamic heterogeneity. Indeed, donor-acceptor fluorescence cross-correlation spectroscopy pointed to motion with a characteristic time of a few tens of microseconds. In the presence of the soluble substrate j-casein, a shift of the FRET efficiency histograms was detected towards lower values, pointing to a major conformational change, which was accompanied with enhanced ATPase activity. In the same manner, looking at the dynamics of PL2 and PL3, FRET efficiency histograms dramatically changed upon the addition of j-casein, and the donor-acceptor fluorescence crosscorrelation functions again pointed to microsecond dynamics. H 2 MM was used to analyse the smFRET data under the assumption of a Markov model that involves a large number of sequentially connected states [100]. The idea behind this exercise was to model the pore loops dynamics in terms of a one dimensional free-energy surface of an arbitrary shape. Interestingly, two well-defined potential wells were retrieved in each case, with microsecond time-scale jumps between them. Estimation of the amplitude of motion of each pore loop using the FRET efficiency values obtained from the analysis suggested a motion of more than 10 A in all cases, corresponding to the translocation of two substrate-protein residues. These large fluctuations are likely to contribute to substrate threading on a much faster time scale than expected based on ATP hydrolysis rates. The ability of smFRET experiments to follow such fast pore-loop motions in real time allows us to address additional questions regarding dynamics and function. The first of these questions is the relation between nucleotide binding and hydrolysis and poreloop motion.
PL1 dynamics were essentially similar in the presence of either ATP, ADP or the slowly hydrolysable analogue ATPcS. In contrast, PL2 responded much more strongly to ATP than ADP, while ATPcS did not elicit any substrate-induced change in the FRET efficiency histogram. PL3 showed a large response to substrate addition in the presence of ATP, and significantly smaller responses with ADP or ATPcS. Taken together, ATP hydrolysis and likely the presence of the product Pi seem to be required for a significant shift of the free energy surface of PL2 and PL3 by the substrate, while PL1 is completely nucleotide-type independent.
To further investigate the correlation of dynamics to machine activity, we turned to investigate PL dynamics in ClpB constructs with mutations of the tyrosine residues on PL1 and PL3. Such mutationsare known to reduce machine activity, especially in the case of PL3 [69,87,89]. Three ClpB variants were examined [96]: PL1 with a substitution of tyrosine 243 to alanine (Y243A, termed Y1), PL3 in which tyrosine 643 was mutated to alanine (Y643A, termed Y3), or a doublemutant, Y243A-Y643A (termed Y1-Y3) in which both PL1 and PL3 carried the tyrosine mutation. All three mutants showed a reduction in disaggregation activity [69,100]; Y3 showed a stronger activity reduction than Y1, while the double mutant Y1-Y3 had almost no disaggregation activity. A high correlation between PL1 and PL3 conformational changes and disaggregation activity was observed (Fig. 3A,B), with PL2 showing a much lower correlation with activity. The high correlation of PL3 dynamics with disaggregation activity is in agreement with previous experiments suggesting that NBD2 is the main contributor to ClpB activity [17,87]. However, the high correlation found for PL1 was less anticipated. 'Substrate-response factors' that quantified the PL conformational changes in response to substrate addition (for a definition see [96]), clearly showed a strong correlation between PL1 and PL3, with an R 2 value of 0.99 (Fig. 3C).
To interpret the above findings, we first recall two general models for protein machine function, the power stroke and the Brownian ratchet [101]. In a power stroke mechanism, a large conformational change follows immediately the hydrolysis of ATP. The hand-over-hand translocation mechanism of ClpB, in which subunits move rigidly following ATP hydrolysis, is clearly reminiscent of a power stroke. On the other hand, in a typical Brownian-ratchet model (Fig. 3D), ATP hydrolysis switches the machine between a state of free diffusion and a state with a pawl-like free-energy surface that makes the overall motion unidirectional. The Brownian ratchet is therefore characterized by two time scales, a fast time scale of diffusion and a slow time scale of free-energy surface switching. We propose that our findings match a Brownian ratchet mechanism for substrate threading by ClpB, as ilustrated in Fig. 3E and explained below.
The strong correlation between the conformational changes of PL1 and PL3 and the disaggregation rate suggests that they are active in substrate pulling. At the same time, the requirement for ATP hydrolysis for PL2 and PL3 conformational changes implies that these pore loops serve as ratchet pawls that rectify substrate motion at different stages of translocation through the channel. PL2 acts first as a pawl when the substrate interacts with NBD1 and ATP has been hydrolyzed (Fig. 3E, step 3). Similar events at NBD2 then engage PL3 as a pawl (step 4). Disengagement of these pawls may allow looped polypeptide segments to escape after partial threading [66,73]. It is likely that the protein harnesses the power of asynchronous pulling by neighbouring subunits to generate rapid translocation events in a Brownian-ratchet like mechanism. Given the length of the central channel of ClpB, such a processive translocation mechanism may explain the translocation steps observed in the optical tweezers study of Avellaneda et al. [73]. The fast fluctuations of the pore loops allow them to reconfigure along a protein substrate, facilitating proper gripping and pulling and likely preventing premature stalling. The Brownian-ratchet mechanism may operate in parallel to the hand-over-hand mechanism discussed in the Introduction.

N-terminal domain and entropic inhibition
An additional path of allosteric regulation in ClpB rests on fast dynamics that involve both the M domain and the NTD. This novel autoinhibitory function, which we termed entropic inhibition [32], will be described in detail in this section.
In the crystal structure of the full-length ClpB [13], the NTDs are connected to the neighbouring NBD1 domains by disordered linkers, suggesting they are highly mobile. Indeed, cryo-EM structures of ClpB hexamers often do not resolve the ring of NTD domains [14][15][16][17], though Gates et al. did manage to resolve six NTDs in one class of Hsp104 structures (out of three) in their study [46], and some studies resolved individual ClpB NTDs [18,102]. Further, analysis of the NTD dynamics in a related AAA+ hexameric machine, p97, by solution NMR spectroscopy suggested microsecond motions [103]. We followed NTD fluctuations by smFRET spectroscopy using (D) in a Brownian ratchet, an effective pawl periodically switches the molecular dynamics between a flat free-energy surface and a structured surface, promoting unidirectional motion. (E) Model for a potential Brownian-ratchet action of pore loops. As a substrate is engaged, pore loops gradually change their average conformation even while continuing to fluctuate on the microsecond time scale between two conformational states. The change in the population ratio of the two states of PL2 and PL3, which likely takes place only upon hydrolysis of ATP, is equivalent to a shift between two free-energy surfaces, and turns them into effective pawls that rectify substrate motion through the central channel. At the same time, PL1 and PL3 function in pulling the substrate. ClpB labelled constructs in which one label was inserted into the NTD and a second dye into NBD1 (Fig. 4A) [32]. Broad FRET efficiency histograms indicated multiple conformations of the NTD and fast fluctuations. In the presence of the substrate j-casein, the NTD got positioned closer to the central channel as was revealed by a shift in FRET efficiency histograms to lower values, in line with previous proposals that the NTDs direct bound substrate-proteins towards the central pore [27]. Fast dynamics of the NTD were verified by fluorescence lifetime correlation spectroscopy (FLCS) [104]. The presence of j-casein accelerated NTD dynamics. Strikingly, the amplitude of the NTD motion was calculated from FRET values to be $ 28 A. Experiments with a truncated form of ClpB (DNClpB, first 140 residues are missing) clearly indicated that the NTD suppresses both ATPase and disaggregation activities [21,32,105]. However, in the presence of the soluble substrate j-casein, both the full length ClpB and the truncated form DNClpB showed similar maximal hydrolysis rates (Fig. 4B), suggesting that in the absence of a protein substrate, DNClpB is dysregulated. In the presence of G6PDH or firefly luciferase aggregates, DNClpB displayed a higher disaggregation rate (Fig. 4C). Probing M-domain motions using smFRET spectroscopy revealed activated M domains in the truncated form (Fig. 4D). Indeed, the active/inactive state ratio of the M domain was calculated to be 1.30 AE 0.05 in DNClpB, instead of 1.00 AE 0.01 in the full-length protein (Fig. 4D). The activation of the M domain in DNClpB was further enhanced by j-casein. Taken together, these findings indicate that the M domain is activated upon the deletion of the NTD and additional activation can take place in response to substrate binding. These results were further substantiated by experiments in which quenching of the dye Atto 655 located on the M domain by a tryptophan residue inserted into the NTD was observed [32]. Considering that the dye was located on the tip of the M domain, the quenching results indicated that the NTD can make contacts with the M-domain. Moreover, removing the entire a-helix A1 from the NTD by deleting residues 8-25 of the fulllength ClpB, resulted in an activated M domain, as was detected from smFRET experiments, showing that indeed the interaction between the M domain and NTD significantly affects its conformational transitions. Finally, the allosteric pathway connecting ATP binding sites and the M domain was tested in the DNClpB variant. As discussed above, it has been previously shown that ATP binding to NBD1 decreased the population of the active state of the M domain, whereas ATP binding to NBD2 had the opposite effect [95]. We found that this regulation was retained in DNClpB, suggesting that NTD removal does not affect the allosteric regulation of the M-domain conformations upon nucleotide binding.
In summary, our smFRET studies, which were able to capture the ultrafast dynamics of the NTD, exposed an unexpected auto-inhibitory mechanism, in which this domain allosterically regulates M-domain dynamics, holding it in a less activated state and eventually preventing ClpB from futile disaggregation activity in the absence of substrate-protein and DnaK binding.
Substrate binding to ClpB, as well as removal of the NTD, alleviated this regulation. Since this mechanism relies on the fluctuations of the NTD, rather than on direct binding to its target, we termed it entropic inhibition [32]. Apparently, the NTD's ultrafast dynamics enable a reversible inhibitory pathway that controls M-domain conformational states, and therefore the activation state of the whole machine.

Conclusions and perspectives
This State of the Art Review presents a set of new and unexpected smFRET results on the dynamics of different parts of the disaggregation machine ClpB. The common denominator of all of these findings is very fast, microsecond time-scale motion between conformational states, which is indeed much faster than the biological function of the protein. In addition to revealing the ultrafast conformational dynamics, to which we will return below, our studies have also shed light on multiple allosteric communication channels within this machine that regulate its function (Fig. 5). One such allosteric channel involves the effect of nucleotide binding to the NBDs on the conformation of the M domain. Our investigation of the M-domain dynamics on the single molecule level showed that the coupling between NBDs and M-domains is exerted through an effect on the rate in which the M-domain shifts from its inactive, horizontal conformation to its active, tilted conformation. The increased rate makes the active state more populated, shifting the whole ClpB machine towards a more active functional state. Another example of an allosteric channel that utilizes domain dynamics is the way the NTD regulates the machine activity. In this case, ultrafast motions of the NTD limit the conformational space of the M domain, making it less likely to occupy the tilted, active state. This type of auto-inhibition does not shut the activity down completely, but rather creates finetuning of ClpB function, when the Mdomain keeps toggling between its two activation states and DnaK can potentially bind. The NTD was also found to allosterically affect NBD function, as a DNClpB variant was shown to exhibit enhanced ATPase activity in the absence of the soluble K-casein. Overall, this suggests a double inhibitory effect of NTDs on the overall function of ClpB; one through the M domain and another through the NBDs.
A close look at the dynamics of ClpB's pore loops led to the proposal of a unique Brownian-ratchet mechanism for substrate-protein translocation. Though all pore loops undergo large-amplitude fluctuations on the microsecond time scale and change their conformations upon interaction with substrate proteins, the Brownian-ratchet mechanism relies on the differences in their sensitivity to the nucleotide states of the NBDs, yet another allosteric effect. Indeed, PL2 and PL3 conformational changes require ATP hydrolysis, suggesting that these pore loops act as the ratchet pawls and ensure unidirectional translocation of substrate proteins through the channel.
There are still multiple exciting open questions to explore in relation to the mechanism of action of ClpB. A most revealing experiment to carry out would be the direct observation of substrate threading by ClpB. Further, other AAA+ proteins, which like ClpB are threading machines for protein unfolding, have intriguing differences in their shape and domain arrangement, such as the presence of only a single ATPase domain in the subunits of type-I proteins or of M domains with different lengths. It will be imperative to understand how these differences shape the dynamics and regulation of these machines. Such studies will allow to both enhance our understanding of the AAA+ protein family and also to refine our experimental methodology in order to obtain more detailed information on protein dynamics on a broad range of time scales.
Finally, we return to the question of the temporal mismatch between conformational dynamics and functional steps in some proteins, which was alluded to in the Introduction. The activity cycle of a protein machine defines a time scale, which is limited by a 'timer' event such as ATP hydrolysis or product release [106]. What we and others [3] find to be quite abundant are structural fluctuations that are much faster than the timer's time scale. These fast fluctuations therefore constitute a second time scale, which is used by the protein to affect its activity cycle. Thus, for example, the fast fluctuations of the M domain in ClpB imply that this domain resides only part of the time in its active state, and as we have found, that that fraction of the time can be tuned by external factors (such as ATP or DnaK concentrations) or internal factors (such as the state of the NTD). In turn, this leads to a tunable, continuous control of the overall activity of the machine, rather than a two-state, digital control. Similarly, the fast fluctuations of pore loops facilitate a Brownian-ratchet like protein substrate threading; fast, microsecond motions push and pull the substrate within the central channel of ClpB, while on a much slower time scale a pawl is engaged, rectifying the motion and preventing back-slipping of the substrate.
This mode of action, involving two very different time scales, seems to describe well the activity of different parts of ClpB, as well as of other proteins. For example, in the enzyme adenylate kinase the two substrates are brought together for their reaction by a domain-closure conformational transition, which was shown to be two orders of magnitude faster than its catalytic turnover time [107]. Ultrafast large-scale conformational changes were also detected in membrane proteins by Shi et al. [108], who have shown, using relaxation-dispersion NMR measurements, that the rhomboid protease GlpG undergoes opening and closing transitions on a time scale of 40 ls. The two-time-scale feature leads to a significantly increased flexibility and facilitates continuous tunability of allosteric interactions. It is likely that similar activity patterns will be found in additional protein machines in the near future.