Substrate-binding domain conformational dynamics mediate Hsp70 allostery

Seesaw-Like “Rigid-Body” Subdomain Rearrangements in the βSBD Link the Substrate-Binding Site and an Interdomain Interface.

Functionally, the βSBD mediates two-way negative allostery: Substrate binding is disfavored upon domain docking, and domain docking is disfavored upon substrate binding. Recent X-ray structures of ATP-bound (domain-docked) DnaK (8, 9) provide detailed structural insights into changes of substrate affinity: Significant perturbations in the substrate-binding cleft in the docked conformation suggest that this state is incompatible with efficient substrate binding (Fig. 1B). However, overall similarity between domain-undocked and -docked βSBD structures (8, 9, 19) raises the question: How does domain-docking trigger conformational changes in the substrate-binding site, and, reciprocally, how does substrate–binding disfavor domain docking?

To elucidate the origins of coupling between substrate binding and domain docking, we compared the βSBD conformations extracted from domain-undocked DnaK (PDB ID code 2KHO) (6) and domain-docked DnaK (PDB ID code 4B9Q) (8) structures using the DynDom algorithm, which enables identification of subdomains and flexible linkers in protein conformational changes (20). This analysis revealed that the structural transition between these two βSBD states can be described in terms of a semirigid body seesaw-like rearrangement of two βSBD subdomains (or “dynamic domains” as defined by the DynDom program). Subdomain I comprises residues 420–437 and 456–485 (Fig. 2A, red), and subdomain II is defined as residues 400–419 and 440–455 (Fig. 2A, blue). ATP-binding results in an 18.6° rotation of subdomain I relative to subdomain II (Table S1). Three hinge/bending regions were identified by the DynDom analysis: residues 417–420, 438–440, and 454–457 (Fig. 2A, yellow). Intriguingly, all three hinge regions have been shown to play important roles in DnaK activity, supporting their role in the βSBD allosteric transition. (i) Residues 417–420 form direct contacts with the interdomain linker (8, 9) and have been shown to be crucial for Hsp70 allostery (9, 21). (ii) Even a subtle perturbation via mutation of the highly conserved Leu454 (located in the second hinge region) to isoleucine results in conformational redistribution in the DnaK conformational ensemble (10). (iii) Residues 438–440 directly interact with substrate through the highly conserved residue Ile438 (22). Importantly, elastic network analysis (23) of a single (domain-undocked) βSBD structure predicts a very similar subdomain rearrangement in the βSBD (Fig. S1), suggesting that βSBD subdomain rotation is not entirely induced by domain docking or an artifact of crystal packing, but rather it is an intrinsic feature of the βSBD structure.

The fact that the substrate-binding site and NBD–βSBD contacts are located at interfaces between βSBD subdomains identified by DynDom and elastic network analyses suggests that βSBD subdomain rearrangement plays a direct role in βSBD-mediated allostery. First, the rotation of the subdomains significantly affects the arrangement of the substrate-binding loops ${\mathrm{ℒ}}_{12}$ and ${\mathrm{ℒ}}_{34}$ and thus should affect substrate-binding affinity and reciprocally could be regulated by substrate binding (Fig. 2A). Indeed, upon ATP binding, subdomain rotation opens the “hydrophobic arch” formed by ${\mathrm{ℒ}}_{34}$ (residue Ala429 of the subdomain I) and ${\mathrm{ℒ}}_{12}$ (residue Met404 of the subdomain II); the same arch has been shown to play a major role in substrate recognition and kinetics of substrate entry to the binding pocket (22, 24). Additionally, subdomain opening also limits substrate contacts with substrate-binding residues from ${\mathrm{ℒ}}_{34}$ and strands β3/β4 (specifically, Phe426, Ser427, Gln433, Ala435, Val436, and Ile438) and thus perturbs the substrate-binding cavity (8, 9).

Second, subdomain rotation results in rearrangements of three contact sites between the βSBD and the NBD and thus couples intradomain conformational changes in the βSBD with interdomain docking (Fig. 2A). The first NBD–βSBD binding site involves Lys414 and Asn415 from βSBD subdomain II and Asp326 from NBD subdomain IIA; the second contact involves Gln442 from the βSBD subdomain II and Asp148 from the NBD subdomain IA; finally, the third interaction site is formed by Asp481 from βSBD subdomain I and Ile168 from NBD subdomain IA (8, 9). Two of these sites are located in βSBD subdomain II, and the third site is in subdomain I. Consequently, subdomain rotation significantly changes the distance between the NBD-binding sites on the SBD: Lys414/Asn415, Gln442, and Asp481.

Interpreting the role of the βSBD in terms of a simple conformational selection model between the two states represented in the docked (ATP-bound) and undocked (isolated domain) crystal structures is confounded by observations in one region: Large structural rearrangements seen in strands β7 and β8 and loop ${\mathrm{ℒ}}_{78}$ upon ATP binding to DnaK (Fig. 2A, colored in gray) cannot be described in terms of rigid-body motions. Indeed, in the domain-undocked (ADP-bound) state, strand β8 (residues 494–502) forms an antiparallel β-sheet with strand β7. Upon domain docking, β8 dissociates from β7 and interacts with β5 (8, 9). As a result, only three β8 residues (Thr-500, Ile-501, and Lys-502) retain a stable β-strand conformation whereas the rest of β8 becomes largely unstructured, resulting in significant perturbations in a hydrophobic core that comprises highly conserved residues Leu454 (strand β5), Leu484 (strand β7), and Ile501 (strand β8). Remarkably, the β5/β7/β8 hydrophobic core does not directly interact with the NBD, but the rearrangement of strand β8 sterically prevents contacts between βSBD residue Gln442 and NBD residue Asp148 (Fig. 2B). Thus, β8 dissociation and disruption of the β5/β7/β8 hydrophobic core upon domain docking is linked to formation of a stable interdomain interface. Thus, the β5/β7/β8 hydrophobic core may act as a switch that triggers βSBD subdomain rotation.

The β5/β7/β8 Hydrophobic Core Is a Central Hub for the βSBD Allosteric Network.

We used NMR chemical shift perturbation (CSP) analysis to elucidate the role of the β5/β7/β8 core in triggering conformational transitions in the βSBD and to deduce how an allosteric signal is transduced through the βSBD. For our analysis, we examined the effect of mild (structurally nonperturbing) amino acid substitutions in apo (domain-undocked) DnaK. The mutations were made in putative allosteric “hotspots” deduced from the analysis above near the interdomain interface sites identified based on DynDom analysis: D481N, K414R, L454I, and I501A; in the interdomain linker L390V; and near the βSBD-αLid interface S505T and D511E.

CSPs from the allosteric substitutions near the βSBD–αLid interface and the β5/β7/β8 hydrophobic core revealed the allosteric network in the βSBD: Perturbations upon the K414R, S505T, and D511E substitutions (located on the βSBD–αLid interface) suggest allosteric coupling between the NBD-docking interface (the Lys-414 site), the αLid interface, and the β5/β7/β8 core (Fig. 3A, i). Perturbations in the β5/β7/β8 hydrophobic core resulted in widespread CSPs affecting almost the entire βSBD (Fig. 3A, ii), including subdomain II, the βSBD-αLid interface, the interface between subdomain I and subdomain II, and contacts between strands β5, β7, and β8 (Fig. 3A, ii). Intriguingly, a more subtle perturbation at the Ile-501 site upon the I501V substitution results in a similar CSP pattern to that observed upon the I501A substitution (Fig. 3B). Moreover, linear peak walking patterns between the WT protein and its I501V and I501A variants were observed for the majority of peaks arising from residues located distant from the I501 mutation site, suggesting that perturbations at Ile-501, the lynchpin residue of the β5/β7/β8 core, induce a conformational shift between two energetically similar βSBD conformations without significantly changing the structure, integrity, and stability of the βSBD. These results suggest that the majority of βSBD residues participate in a functional allosteric network that links the substrate-binding site and the interdomain interface and agree remarkably well with our previous statistical coupling analysis based on a large sequence alignment covering 926 Hsp70/110s (21), which suggested that more than 50% of the βSBD residues were evolutionary coupled.

In contrast, the D481N and L390V substitutions resulted in CSPs that were local to the mutated residue (Fig. 3A, iii). Interestingly, in contrast to another NBD contact residue, Gln442, Asp481 is positioned such that productive domain docking does not require large intradomain rearrangements for it to be accessible at the interface. Similarly, the linker residue Leu-390 is exposed to the solvent in the ADP-bound states and thus accessible for docking without conformational change. Interestingly, residues D481 and L390 are not only crucial for interdomain allostery (8, 9, 25), but even subtle perturbations at these sites (e.g., the D481N and L390V substitution) result in significant destabilization of domain docking in the presence of ATP (10). These facts, together with the absence of long-range CSPs in the βSBD upon the D481N substitution in the apo (domain-undocked) state, suggest that Asp481 and Leu390 are important for formation of the interdomain interface upon domain docking rather than for allosteric control of the βSBD conformational ensemble.

All together, our results experimentally demonstrated that perturbations at the interdomain interface and in the β5/β7/β8 core are propagated through the βSBD structure, suggesting allosteric networks of energetically coupled βSBD residues that link the βSBD–NBD and βSBD–αLid interfaces, the β5/β7/β8 core, and the substrate-binding site. Importantly, communication within these networks is possible even in the absence of significant changes in βSBD structure: Perturbations in the allosteric hotspots (i.e., the β5/β7/β8 core) result from redistributions in the functional ensemble of βSBD conformations that have very similar structure (Figs. 1B and 2A) but significantly different ability to bind substrate and to dock with the NBD.

Perturbations at the Substrate-Binding Site Enable Modulation of the βSBD Conformation.

Next, we asked how the allosteric signal propagates in the opposite direction: from the substrate-binding site to the interdomain interface. Our previous CSP analysis on the isolated (domain-undocked) βSBD revealed that substrate binding results in small, but significant, long-range CSPs for residues 416–420 (14), despite the fact that no significant changes were observed between the X-ray structures of substrate-unbound (PDB ID code 4R5L) (26) and substrate-bound SBD (PDB ID code 1DKZ) (19). Interestingly, our DynDom results identified this part of the protein as a hinge/bending region between two βSBD subdomains, providing a plausible explanation for these NMR perturbations: Substrate binding to domain-docked DnaK results in a small redistribution in subdomain arrangements that predominantly should affect subdomain interfaces and hinge/bending regions. To validate this model, we analyzed the effect of perturbations in loop ${\mathrm{ℒ}}_{34}$ at the substrate-binding site on the βSBD conformational ensemble. To monitor long-range changes upon perturbation of ${\mathrm{ℒ}}_{34}$, we performed CSP analysis on a previously characterized DnaK construct deficient for peptide binding (9). For this construct, called DnaK-${\mathrm{ℒ}}_{34}$, the entire ${\mathrm{ℒ}}_{34}$ (TAEDNQS) was replaced by a shorter sequence (MGG) from a DnaK homologous protein, Hsp110 (9, 27). In the domain-undocked conformation, this construct has low substrate affinity–similar to domain-docked (ATP-bound) WT DnaK (27). Intriguingly, the X-ray structure of this loop variant showed no significant changes from the domain-undocked SBD structure (rmsd ≤ 0.4 Å) apart from local perturbations in ${\mathrm{ℒ}}_{34}$ itself, which result in a slightly more open arrangement of ${\mathrm{ℒ}}_{34}$ and ${\mathrm{ℒ}}_{56}$ relative to ${\mathrm{ℒ}}_{12}$ (9). By contrast, our CSP analysis, which reports on the solution conformational ensemble, revealed widespread long-range perturbations in the βSBD upon the ${\mathrm{ℒ}}_{34}$ substitution (Fig. 4), suggesting that the ${\mathrm{ℒ}}_{34}$ substitutions alter the βSBD conformational distribution in functionally significant ways, experimentally validating two-way allosteric coupling between ${\mathrm{ℒ}}_{34}$ and β8, as predicted by our DynDom analysis. Our results fully agree with the fact that the ${\mathrm{ℒ}}_{34}$ substitution not only perturbs substrate binding locally but also offers an explanation for its facilitation of crystallization of the otherwise transient ATP-bound DnaK conformation by shifting the population of this domain to this state (9). Moreover, in the Hsp110 protein (where this ${\mathrm{ℒ}}_{34}$ substitution occurs naturally), the domain-docked conformation seems to be much more stable than it is in DnaK (27, 28).

Perturbations at the Substrate-Binding Site Enable Modulation of βSBD Conformational Flexibility.

Next, to further examine how perturbations around the substrate-binding site affect the βSBD conformation, we performed similar CSP analysis on another previously characterized mutant deficient for peptide binding (9). For this construct (called DnaK-PP), two highly conserved residues from loop ${\mathrm{ℒ}}_{56}$, Gly461, and Gly468 were substituted by prolines (9). Similar to the DnaK-${\mathrm{ℒ}}_{34}$ construct, the previous biochemical characterization of DnaK-PP revealed that it had very low substrate affinity even in the absence of ATP (9). Importantly, neither Gly461 nor Gly468 is directly involved in substrate binding, suggesting that these mutations act indirectly rather than directly perturbing interactions with substrate. As previously suggested (9), perturbations in loop ${\mathrm{ℒ}}_{56}$ somehow favor the βSBD conformation with low substrate-binding affinity even in the absence of domain docking, but they do not affect domain docking in the presence of ATP. Fully in line with this suggestion, our NMR analysis revealed that these two ${\mathrm{ℒ}}_{56}$ substitutions did not significantly affect the NBD in either the docked or undocked DnaK conformation: No significant changes were observed for NBD resonances in either the presence or the absence of ATP, suggesting that, similar to the WT protein, this variant adopts the same domain-undocked conformation in the presence of ADP and the domain-docked conformation upon ATP binding (Fig. 5A). In contrast, the ${\mathrm{ℒ}}_{56}$ substitutions drastically affect βSBD conformation (Fig. 5B): All βSBD peaks disappeared from the spectra of both domain-docked and domain-undocked states upon introduction of the ${\mathrm{ℒ}}_{56}$ substitutions, which can be explained only by enhanced μs–ms dynamics for the entire βSBD upon introduction of the$\hspace{0.17em}{\mathrm{ℒ}}_{56}$ mutations. Importantly, the fact that these enhanced dynamics are not associated with nonfunctional βSBD destabilization and/or unfolding—the ${\mathrm{ℒ}}_{56}$ construct enables highly specific and delicate NBD–βSBD contacts—suggests that the substitutions in the ${\mathrm{ℒ}}_{56}$ loop result in thermodynamic redistribution of the ensemble of functional βSBD conformations. Interestingly, a similar effect (enhanced μs–ms flexibility) was observed for all βSBD peaks in ATP-bound WT DnaK (7, 10). Moreover, the enhanced flexibility reflected in the NMR data are fully consistent with the long-time observation of enhanced proteolytic lability for the SBD in ATP-bound DnaK (15, 16), and other indications of dynamics, such as ITC results (17) (see also Introduction), suggesting that enhanced conformational flexibility is a unique and functionally important feature of the docked-like βSBD conformation.

NBD Docking, β8 Strand Rearrangement, and Substrate-Binding Control βSBD Flexibility.

To elucidate the link between ATP-induced domain docking, substrate binding, and intradomain flexibility of the βSBD, we used full-atom molecular dynamic (MD) simulations. As expected from previously reported attempts to monitor domain-docking/domain-undocking transitions in Hsp70s using full-atom MD simulations (29), no full transition between the βSBD end-point conformations (NBD-docked and -undocked) was observed in our MD simulations of the full-length DnaK. Thus, to monitor conformational transitions in the βSBD, we dissected key functional steps of the Hsp70 allosteric cycle—dissociation from the NBD, β8 rearrangement, and substrate binding—using truncations of functionally and structurally important regions that enable conformational transitions in the DnaK to occur on the MD timescale (Table S2).

First, to test whether MD can reproduce experimental results and demonstrate that substrate binding significantly reduces conformational flexibility of the substrate-binding site and favors arch closure (14, 30, 31), we performed MD simulations of the domain-undocked βSBD conformation in the presence and the absence of a short model substrate peptide NR (NRLLLTG) (32). In full agreement with experimental observations, our MD simulations demonstrated that loops ${\mathrm{ℒ}}_{34}$ and ${\mathrm{ℒ}}_{56}$ are highly dynamic in the absence of substrate but become rigid in the substrate-bound state, as indicated by the rms fluctuations (rmsf) of individual CA atoms calculated over MD trajectories (Fig. 6, Top).

Next, to monitor how NBD docking affects βSBD flexibility, we performed MD simulations on the domain-docked conformation of the βSBD in the presence and in the absence of the NBD (using as a starting structure the ATP-bound X-ray conformation, PDB ID code 4B9Q) (Table S2). NBD removal results in a very significant increase in βSBD dynamics even in the absence of significant changes in βSBD structure (Fig. 6, Bottom). The largest effects were observed for the NBD-binding loops containing residues D481 (${\mathrm{ℒ}}_{67}$) and K414 (${\mathrm{ℒ}}_{23}$), suggesting that NBD binding stabilizes the K414/Q442/D481 arrangement required for NBD–βSBD docking. Notably, NBD docking results not only in local stabilization of the NBD-docking loops, but also in long-range dynamic effects: for example, suppressing flexibility of the substrate-binding loop ${\mathrm{ℒ}}_{34}$ (in blue in Fig. 6).

To further elucidate the mechanism of coupling between structural rearrangements in the βSBD and conformational flexibility of the substrate-binding loops, we monitored the distance between two arch residues, Met404 (${\mathrm{ℒ}}_{12}$) and Ala429 (${\mathrm{ℒ}}_{34}$), as a function of simulation time for different MD trajectories (Fig. 7 and Fig. S4). In the domain-undocked, substrate-bound state, the βSBD spends the vast majority of time with stable contacts within the ${\mathrm{ℒ}}_{12}$/${\mathrm{ℒ}}_{34}$ arch [labeled “C” (closed) in Fig. 7, Left]. The fact that only very small distance variations were observed during the MD simulations suggests the presence of only very limited dynamics around the substrate-binding site. By contrast, in the domain-docked conformation, there are no stable arch contacts between loops ${\mathrm{ℒ}}_{12}$ and ${\mathrm{ℒ}}_{34}$, and the βSBD samples two distinct ${\mathrm{ℒ}}_{12}$/${\mathrm{ℒ}}_{34}$ arrangements. One of these loop arrangements is close to the ${\mathrm{ℒ}}_{12}$/${\mathrm{ℒ}}_{34}$ conformation observed in the X-ray structure of domain-docked DnaK [labeled “O” (open) in Fig. 7, Right] whereas another arrangement represents a transient partially open conformation [labeled “I” (intermediate) in Fig. 7]. The very significant variation in distance between arch residues for both loop arrangements of domain-docked βSBD indicates that this βSBD conformation is highly dynamic on the sub-ns to ns timescale. Taken together, these MD results suggest that domain-docked and -undocked βSBD conformations differ not only in the arrangement of the substrate-binding loops but also in the dynamics around the substrate-binding site (Fig. 7). Thus, the computational simulations are in excellent agreement with the surprisingly large βSBD B-factors in X-ray structures of domain-docked DnaK (8, 9) and the enhanced hydrogen–deuterium exchange observed around the substrate-binding site (7, 18).

Next, we carried out MD simulations on truncated DnaK constructs that mimic different stages in the transition between domain-docked and domain-undocked conformations. (In all of these simulations, the αLid was removed to focus on intradomain contributions.) First, we removed the NBD domain from the βSBD, using the crystal structure of domain-docked, ATP-bound DnaK as a starting point (8), and then ran MD simulations for the isolated βSBD. Strikingly, removal of the NBD contacts is adequate to cause partial closure of the substrate-binding loops (Fig. 7, red arrow). This finding suggests that interactions with the NBD are required to stabilize the fully open ${\mathrm{ℒ}}_{12}$/${\mathrm{ℒ}}_{34}$ arrangement, and in turn, destabilization/removal of NBD–βSBD contacts favors partially open intermediate [labeled “I” (intermediate) in Fig. 7]. In turn, the addition of substrate to this conformation is sufficient for formation of a stable ${\mathrm{ℒ}}_{12}$/${\mathrm{ℒ}}_{34}\hspace{0.17em}$arch (Fig. 7, red arrow).

For the domain-undocked βSBD, removal of substrate slightly enhances dynamics around the substrate-binding site (Fig. 7, blue arrow). Dissociation of strand β8 (as observed in the crystal structure of ATP-bound DnaK) and removal of loop ${\mathrm{ℒ}}_{78}\hspace{0.17em}$ were explored by virtual protein engineering: truncation to residue Gly494. Even in the domain-undocked βSBD conformation, removal of strand β8 and loop ${\mathrm{ℒ}}_{78}$ resulted in opening of the substrate-binding loops and drastic enhancement in loop flexibility around the substrate-binding site (Fig. 7, blue arrow), showing that β5/β7/β8 and ${\mathrm{ℒ}}_{78}$/${\mathrm{ℒ}}_{56}$ interactions are absolutely essential for the formation of a stable ${\mathrm{ℒ}}_{12}$-${\mathrm{ℒ}}_{34}$ arch. These results argue that formation of an interface with the NBD, binding of substrate, and specific interactions with strand β8 and loop ${\mathrm{ℒ}}_{78}$ are key trigger points that control dynamics around the substrate-binding site.