Intermediates in allosteric equilibria of DnaK–ATP interactions with substrate peptides

ATP and client peptides interact allosterically to effect the molecular-chaperone activity via which Hsp70 proteins promote protein folding. Here, two DnaK–ATP structures are described: one with implications for a transition in which peptides are released upon the binding of ATP and one with relevance for the rebinding and capture of client polypeptides.

Abstract

Hsp70 molecular chaperones facilitate protein disaggregation and proper folding through iterative cycles of polypeptide binding and release that are allosterically coupled to ATP binding and hydrolysis. Hsp70s are ubiquitous and highly conserved across all of life; they bind ATP at an N-terminal nucleotide-binding domain (NBD) and client peptides in the substrate-binding domain (SBD). The NBD and SBD are connected by a highly conserved linker segment that is integrated into the NBD when ATP is bound but is flexible when the NBD is nucleotide-free or bound with ADP. Allosteric coupling is lost when the linker is flexible, and the freed SBD binds peptide clients with high affinity. It was recently discovered that Hsp70–ATP is in an equilibrium between a restraining state (R) with little affinity for peptides and a low ATPase activity, and a stimulating state (S) that binds peptides efficiently, but with rapid kinetics, and has a relatively high ATPase activity. While attempting to characterize the S state, crystal structures of DnaK–ATP were obtained that demonstrate intrinsic Hsp70 plasticity that affects binding interactions with substrate peptides. These structures provide insights into intermediate states along transition pathways in the Hsp70 chaperone cycle.

1. Introduction  

Molecular chaperones are proteins that promote the folding and assembly of client macromolecules without making covalent modifications or being self-incorporated. Many molecular chaperones are under the control of the heat-shock response, and are therefore known as heat-shock proteins (Hsps), although the controlling transcription factors can be activated by diverse stresses including oxidants, heavy metals and inflammation (Morimoto, 2011 ▸). Heat-shock proteins of 70 kDa (Hsp70s) are pre-eminent in that they cooperate with many partners in chaperone activity, including those in the Hsp60 and Hsp90 chaperone pathways and Hsp100 and ubiquitination-based protein-degradation systems (Hartl et al., 2011 ▸). Hsp70s are ubiquitous, abundant and highly conserved across all of life, and family members derive from essential, constitutively expressed genes as well from stress-induced genes (Daugaard et al., 2007 ▸). Hsp70 chaperones are important in human disease, providing protection against neuro­degeneration (Witt, 2013 ▸) and nurturing tumors and cancer (Murphy, 2013 ▸).

Hsp70 chaperones act in ATP-dependent cycles of binding and release of client substrates, typically hydrophobic segments of partially denatured proteins. ATP binds to the N-terminal nucleotide-binding domain (NBD) and client peptides bind to a following substrate-binding domain (SBD). A short, poorly ordered C-terminal segment often includes targeting sequences. Crystal structures of the constituent domains provide a framework for biochemical understanding of Hsp70 chaperone activity. The prototype NBD structure is that of bovine Hsc70 (Flaherty et al., 1990 ▸), which has an actin-like fold, while the prototype SBD structure is that of the bacterial Hsp70 DnaK from Escherichia coli (Zhu et al., 1996 ▸), and is to this day unique. A highly conserved segment fosters allosteric communication between the NBD and SBD when the NBD is bound with ATP. This linkage is uncoupled and flexible when nucleotide-free or bound with ADP, and this uncoupled state has been characterized from NMR measurements of DnaK (Bertelsen et al., 2009 ▸). A crystal structure of the yeast Hsp110 Sse1 (Liu & Hendrickson, 2007 ▸), a remote Hsp70 homolog, inspired the generation of protein constructs that led to structures of DnaK–ATP (Kityk et al., 2012 ▸; Qi et al., 2013 ▸).

Biochemical tests of interface mutants produced unexpected results and motivated a theoretical treatment of allosteric regulation in Hsp70 chaperones (Hendrickson, 2020 ▸; Wang et al., 2021 ▸). The theory features conformational equilibria between alternative ATP-bound states: a restraining state, Hsp70_R–ATP, now identified with the earlier DnaK–ATP structures, has a low rate of ATP hydrolysis and little or no affinity for substrate peptides, whereas a stimulating state, Hsp70_S–ATP, hydrolyzes ATP rapidly and binds substrates with high intrinsic affinity. We have recently performed biochemical tests of the theory and determined crystal structures of DnaK–ATP in the predicted stimulating state as well as in the restraining state (Wang et al., 2021 ▸). The new structures corroborate and elaborate the allosteric theory. Together with descriptions of the uncoupled (U) states, Hsp70_U–ADP and apo Hsp70_U (nucleotide-free), they also define the major structural elements of the Hsp70 chaperone cycle (Fig. 1 ▸).

Figure 1.

Hsp70 chaperone cycle. Major conformations in the Hsp70 chaperone cycle depicted as structures of E. coli DnaK. Left: U (apo, P_C), with DnaK in the uncoupled state, nucleotide-free (apo) and bound to a client polypeptide (P_C). The NBD (PDB entry 1dkg) and SBD (PDB entry 1dkz) are disposed as in PDB entry 2kho. Center: R (ATP, no client), with DnaK in the restraining state bound to ATP after release of the peptide (PDB entry 7krt). Right: S (ATP, P_C′), with DnaK in the stimulating state bound to ATP and a new client peptide (P_C′) (PDB entry 7krw). Hydrolysis of ATP then yields U (ADP, P_C′), and subsequent release of ADP and phosphate completes the cycle back to U (apo, P_C′). The NBD is in blue, SBDβ in green, the NBD–SBD linker in purple, SBDαA in yellow, the remainder of SBDα in red, ADP in yellow, Pγ in orange and the client polypeptide in black. Adapted from Fig. 7(a) of Wang et al. (2021 ▸).

Iterations of peptide binding and ATP hydrolysis in the Hsp70 cycle promote the disaggregation of aberrant protein masses and proper protein folding. U-state Hsp70–ADP–peptide complexes are relatively stable, and nucleotide-exchange factors, such as Hsp110 for eukaryotic Hsp70s, typically assist in displacing ADP to advance the chaperone cycle; ATP rebinding after ADP dissociation generates R-state Hsp70–ATP with concomitant rapid peptide release; allosteric equilibration with the S state allows the binding of client peptides into Hsp70–ATP–peptide complexes; and ATP hydrolysis, which may be further stimulated by Hsp40-activating factors, regenerates the long-lived U-state Hsp70–ADP–peptide complexes. Intermediate substates must necessarily occur during transitions between the major states. In the course of our efforts to produce the postulated stimulating state of DnaK, we also obtained additional DnaK–ATP crystal structures in conformations suggestive of such intermediates. In this report, we describe two DnaK–ATP structures that inform about intermediates in transition pathways of the Hsp70 chaperone cycle.

2. Materials and methods  

2.1. Protein production and crystallization  

The cloning, expression and purification procedures used to produce DnaK proteins for this study were the same as those described previously for our analysis of the DnaK–ATP structure (Qi et al., 2013 ▸), referred to here as DnaK_R–ATP. In brief, specified constructs were cloned into pSMT3 vector, expressed as the corresponding Smt3 fusion proteins, purified by nickel-affinity chromatography (HisTrap column; GE Healthcare), separated from the Smt3 tag by Ulp1 protease cleavage and elution through the nickel-affinity column, further purified by ion-exchange chromatography (HiTrap Q column; GE Healthcare) and size-exclusion chromatography (SEC; Superdex 200 Increase column; GE Healthcare), and typically concentrated to approximately 10 mg ml⁻¹ in a buffer consisting of 10 mM HEPES–KOH pH 7.5, 40 mM KCl, 100 mM NaCl. Crystallizations were performed by vapor diffusion with a Mosquito nanolitre pipetting robot (TTP Labtech), typically using 600 nl protein solution plus 600 nl precipitant from The JCSG Core Suite (Qiagen). Trials were performed at 4°C and 20°C, but crystals were only obtained at 4°C for these DnaK–ATP complexes. Initial crystal hits were optimized in grid searches against pH and precipitant concentration.

2.1.1. DnaK₅₄₀::NR  

For the production of this protein, DNA sequences corresponding to DnaK_2–540 bearing a T199A mutation, a –TTGSG– linker and an NRLLLTG heptapeptide were joined and cloned into the Smt3 system. Since Ulp1 cleavage adds an N-terminal serine, the resulting product is an M1S mutant, although native E. coli DnaK would have a cleaved N-terminal methionine since the next residue is a glycine. The successful condition from crystallization screening had the initial composition 0.1 M NaCl, 0.1 M HEPES pH 7.5, 1.6 M ammonium sulfate. Crystals were obtained after concentrating the SEC-eluted protein by centrifugation to ∼10 mg ml⁻¹ and proved to be in the lattice-swapped Q state described here. Legitimate S-state crystals (Wang et al., 2021 ▸) were obtained by direct crystallization after SEC elution, without buffer exchange or a centrifugal concentrating process. The initial precipitant concentrations in this latter case were very different (0.04 M HEPES pH 7.5, 25.5% PEG 4000, 15% glycerol, 0.17 M ammonium sulfate, 0.1 M KCl) and supersaturation was achieved by using a 3:1 volume ratio of protein solution to well solution.

2.1.2. DnaK_609-QQQ::NR  

For the production of this protein, DNA sequences corresponding to DnaK_2–609 (again bearing the T199A mutation), a –GGSAQQQTTGSG– linker and an NRLLLTG heptapeptide were joined and cloned into the Smt3 system. The initial precipitant concentrations in this case were 0.1 M HEPES pH 7.5, 2.0 M ammonium sulfate, 2% PEG 400, 0.15 M sodium citrate. The initial crystals diffracted to 3.4 Å resolution; however, the diffraction improved to 2.8 Å resolution after dehydration in 2.6 M ammonium sulfate for two weeks.

2.2. Diffraction measurements and data reduction  

Crystals were mounted on LithoLoops (Molecular Dimensions), cryoprotected with 20–25% glycerol or LV CryoOil (MiTeGen), plunge-cooled in liquid nitrogen and stored in ALS-style pucks (MiTeGen). The pucks were loaded into a dry-shipper canister and sent to the Northeast Collaborative Access Team (NE-CAT; http://necat.chem.cornell.edu) for remote data collection on the NE-CAT beamlines of the Advanced Photon Source (APS) at Argonne National Laboratory (ANL). The diffraction measurements for both crystals were made on beamline 24-ID-C equipped with a PILATUS 9M detector (Dectris). The X-ray energy was set to the Se K-edge energy of 12.6 keV for the Q-state DnaK₅₄₀::NR structure and to 7 keV for the DnaK_609-QQQ::NR structure. The crystals were maintained at 100 K in N₂ gas vaporized from liquid nitrogen. The diffraction data were indexed, integrated and scaled using XDS (Kabsch, 2010 ▸) and were merged and truncated using AIMLESS (Evans & Murshudov, 2013 ▸).

Diffraction from the DnaK_609-QQQ::NR crystals proved to be appreciably anisotropic, extending to higher angles in the unique c* direction than in the perpendicular a*/b* plane. Following initial AIMLESS processing to a limit of CC_1/2 = 0.3, the data were further analyzed by STARANISO (Global Phasing Ltd) to truncate to an ellipsoidal diffraction volume extending to comparable limits of d _min = 2.82 Å in the c* semi-axis and to d _min = 3.48 Å in the perpendicular a*/b* semi-axes (Table 1 ▸).

Table 1. Diffraction data-collection and refinement statistics.

Both of the DnaK constructs include the hydrolysis-impairing T199A mutation.

	DnaK540::NR (Q state)	DnaK609-QQQ::NR†
Data collection
Beamline	NE-CAT 24-ID-C	NE-CAT 24-ID-C
Wavelength (Å)	0.9792	1.7712
Bragg spacings (Å)	47.51–2.15 (2.23–2.15)	49.75–2.82 (2.92–2.82)|49.75–2.82/3.48 (2.97–2.82)
Space group	P6322	P6422
a, b, c (Å)	145.14, 145.14, 122.21	121.52, 121.52, 457.88
α, β, γ (°)	90, 90, 120	90, 90, 120
Za‡	1	2
Solvent content (%)	59.9	65.2
Total reflections	869726 (83098)	1825816 (151035)|1398563 (62590)
Unique reflections	41734 (4092)	48737 (4394)|38530 (1927)
Multiplicity	20.8 (20.3)	37.5 (34.4)|36.3 (32.5)
Completeness (%)	99.9 (99.4)	97.0 (91.2)|92.6 (71.1)
Mean I/σ(I)	15.3 (0.7)	21.2 (0.2)|24.1 (1.0)
R merge	0.187 (1.77)	0.160 (9.59)|0.117 (5.32)
R meas	0.192 (1.82)	0.162 (9.73)|0.119 (5.41)
R p.i.m.	0.042 (0.44)	0.026 (1.63)|0.020 (0.93)
CC1/2	0.999 (0.37)	0.999 (0.38)|0.999 (0.69)
CC*	1.000 (0.74)	1.000 (0.74)
Refinement
Resolution (Å)	2.15	2.82/3.48§
Reflections used for R work	39718 (3832)	36583 (1206)
Reflections used for R free	1996 (192)	1926 (63)
R work	0.205 (0.319)	0.255 (0.465)
R free	0.245 (0.338)	0.273 (0.510)
No. of non-H atoms
Total	4228	9233
Macromolecules	4022	9140
Ligands	49	85
Solvent	157	8
Protein residues	532	1206
R.m.s.d. ideality, bonds (Å)	0.011	0.013
R.m.s.d. ideality, angles (°)	0.91	1.86
Ramachandran favored (%)	98.5	97.1
Ramachandran allowed (%)	1.5	2.9
|Rama-Z| (whole structure)	0.03	0.33
Rotamer outliers (%)	0.7	0.3
Clashscore	3.31	10.22
Average B factor (Å2)
Overall	61.7	71.5
Macromolecules	62.3	71.7
Ligands	47.6	57.1
Solvent	51.4	43.9
PDB code	7kzu	7kzi

^†Certain merging statistics for DnaK_609-QQQ::NR are reported in two fields separated by |. The field before | reports the statistics from spherical truncation by AIMLESS and the field after | reports the statistics from ellipsoidal truncation by the STARANISO server. The ellipsoid is specified by |S _max| = 2sinθ/λ = 1/2.82 Å along c* and 1/3.48 Å in the a*/b* plane.

^‡Molecules per asymmetric unit.

^§DnaK_609-QQQ::NR was refined against anisotropically truncated data processed by the STARANISO server.

2.3. Structure determination and refinement  

Structures were determined by molecular replacement using Phaser (McCoy et al., 2007 ▸) and were refined with Phenix (Liebschner et al., 2019 ▸). Structure validation was performed with MolProbity (Chen et al., 2010 ▸).

2.3.1. Q-state DnaK₅₄₀::NR  

The NBD from an Hsp70_R–ATP structure (PDB entry 4jne; Qi et al., 2013 ▸) was first positioned by molecular replacement, and the SBD elements and NR peptide were built manually and iteratively into the emergent density features. Refinement was then carried out to completion (Table 1 ▸).

2.3.2. DnaK_609-QQQ::NR  

The NBD from an Hsp70_R–ATP structure (PDB entry 4jne; Qi et al., 2013 ▸) and the SBD from the SBD_U–NR structure (PDB entry 1dkz; Zhu et al., 1996 ▸) were placed by initial molecular replacements. The α-lid domain in this model deviated from the emergent density features, which were of somewhat lower resolution than for the other domains. This α-lid portion of the structure was then refitted as a rigid body as guided by features from Met588 and Met599 in the 7 keV Bijvoet difference map, which showed density corresponding to all 28 S atoms in the protein plus all three ATP P atoms and four sulfate ions. Modeling and refinement were then completed for adequately ordered portions of the structure (Table 1 ▸).

Structural figures were prepared with PyMOL (DeLano, 2002 ▸; Schrödinger).

3. Results  

3.1. Structure of lattice-swapped DnaK₅₄₀::NR fusion protein  

Our theoretical treatment of allosteric regulation in Hsp70 proteins postulates a conformational equilibrium between a restraining state R, represented by the known structures of DnaK–ATP (Kityk et al., 2012 ▸; Qi et al., 2013 ▸), and a predicted stimulating state S. The equilibrium strongly favors the R state in the absence of client peptides, based on our analysis of ATP-hydrolysis data, and the balance tips towards the S state as the peptide concentration increases, according to the theory (Hendrickson, 2020 ▸). In our efforts to obtain an S-state structure, we produced fusion proteins with the heptapeptide substrate NRLLLTG (called NR for short) coupled by a flexible linker to the C-terminus of our DnaK constructs.

The DnaK₆₀₉::NR fusion protein (DnaK₆₀₉-GGSGSG-NRLLLTG) showed the expected S-state properties (Wang et al., 2021 ▸), but when the resulting crystals diffracted insufficiently well (d _min = 7.7 Å) we adopted the strategy of deleting the α-lid domain (αB2–αE) to produce DnaK₅₄₀::NR. The hydrolysis-impaired T199A variant of this protein eluted as a monomer from size-exclusion chromatography, and the initial crystals obtained from it diffracted quite well (d _min = 2.15 Å; Table 1 ▸); however, the structure proved to be in the lattice-swapped state described here. We subsequently adjusted the conditions for crystallogenesis and obtained an authentic S-state structure (d _min = 1.82 Å; PDB entry 7kru; Wang et al., 2021 ▸).

The structure reported here has one molecule in the asymmetric unit of a P6₃22 lattice. [Incidentally, DnaK associates across dyad or quasi-dyad axes in all R- and S-state DnaK structures (Wang et al., 2021 ▸), in this case at a crystallographic dyad (Fig. 2 ▸).] The NR peptide is bound to SBDβ as in the isolated SBD structure (Zhu et al., 1996 ▸); however, this NR peptide derives from another molecule in the lattice. SBDα is displaced and its αA/B1 helix lies across the NBD at the IA–IB interface, just as in R-state structures. Thereby, the helix extends towards SBDβ of a 6₃-screw related molecule in the lattice (Fig. 2 ▸). Residues 527–540, which would be part of αB1, and the –GGSGSG– linker segment are all disordered; however, the span from C^α(526) to the N-terminus of the adjacent NR peptide (24.9 Å) readily accommodates the 20 disordered residues, whereas that to NR in its own SBDβ is too long (59.2 Å). Besides being lattice-swapped, DnaK here appears to be an unanticipated hybrid of R-like and S-like features, which we call the quasi-intermediate state Q, DnaK_Q–ATP.

Figure 2.

Crystal structure of lattice-swapped Q-state DnaK₅₄₀::NR fusion protein. The 6₃ crystallographic screw axis presents the NR peptide from one DnaK subunit into SBDβ of the next, as indicted by the dotted connections. Crystallographic P6₃22 dyad axes relate subunits through face-to-face NBD contacts as shown for the dimer centered at c = 0, where the dyad (diamond) is directed away from the 6₃ axis at φ = 0°. Successive dyads at −c/2, +c/2 and +c are shown as arrows directed away from the 6₃ axis at φ = −60°, +60° and +120°, respectively, counterclockwise from above. For clarity, the diagram propagates only the protomer on the right from the central dimer. Arrows are drawn with depth-cued red shading. Domain coloring is as in Fig. 1 ▸ except that αA/B is all red.

3.1.1. Comparison of the SBD–NBD interfaces of DnaK_Q–ATP with DnaK_R–ATP and DnaK_S–ATP  

Structures of DnaK–ATP in the Q, R and S states are shown in Figs. 3 ▸(a), 3 ▸(b) and 3 ▸(c), respectively, where each is superimposed on the NBD of DnaK_R–ATP for comparison since the NBD is quite similar in all three cases. The elements of secondary structure as delineated in Fig. S5 of Wang et al. (2021 ▸) also apply to the Q state, and selected elements are labeled in Figs. 3, 4, 6 and 9.

Figure 3.

Comparison of SBD–NBD interfaces. (a) Stereo drawing of lattice-swapped Q-state DnaK₅₄₀::NR (PDB entry 7kzu). (b) Stereo drawing of R-state DnaK₆₀₀::NR (PDB entry 7krt; Wang et al., 2021 ▸). (c) Stereo drawing of S-state DnaK₅₄₀::NR (PDB entry 7kru; Wang et al., 2021 ▸). In (a), (b) and (c) the backbone ribbon is colored blue for NBD, purple for the NBD–SBD linker, green for SBDβ and red for SBDα; the NR peptide and ATP are colored orange. Each NBD has been superimposed onto the Q-state NBD, which is oriented with its dyad axis vertical and viewed looking into IA and IB.

Clearly, the lattice-swapped Q state of DnaK₅₄₀::NR binds the NR peptide. Moreover, the mode of binding in SBDβ_Q is precisely the same as that in S-state DnaK₅₄₀::NR (Fig. 4 ▸ a). Yet, despite SBDβ_Q having the peptide-complexed conformation of DnaK_S–ATP, its interface with the NBD is largely similar to that in DnaK_R–ATP structures (Fig. 4 ▸ b). SBDβ_Q loops , and are nearly superimposed onto their DnaK_R–ATP counterparts when the NBDs are superimposed (Fig. 3 ▸ a versus Fig. 3 ▸ b). Due to the changes within SBD_Q relative to SBD_R, differences do occur in the NBD–SBD interface at the β8–αA junction and at , which is pulled away together with for peptide association. The truncated αA/B1 helix of SBD_Q is precisely superimposable onto that of SBD_R at the N-terminus, and diverges only slightly over its length as it crosses NBD αE. Internal differences within the SBD complicate overall comparison, but with the NBDs aligned a rotation of only 7.7° is needed to superimpose the sandwich/barrel axes. Although the conformation of SBD_Q differs appreciably from that of SBD_R, the NBD-proximal loops within SBD_Q are remarkably similar to those within SBD_R (Fig. 4 ▸ c).

Figure 4.

Comparison of SBD conformations. (a) Stereo drawing of SBDβ_Q (lime) superimposed on SBDβ_U (PDB entry 1dkz; yellow) with superposition based on the C^α atoms of residues 399–501. (b) Stereo drawing of SBD_Q superimposed on SBD_R (PDB entry 7krt; pink) with superposition based on the NBD. (c) Stereo drawing of SBD_Q superimposed on SBD_R with superposition based on the C^α atoms of SBD inner loops and (residues 399–412 and 443–452). The Q-state structure is oriented the same in each panel, and the view is roughly into the SBD_Q-bound NR peptide (orange).

Since the orientations of SBD_Q and SBD_S relative to the NBD are very different (Fig. 3 ▸ a versus Fig. 3 ▸ c), their interfaces with the NBD are essentially unrelated.

3.1.2. Comparison of the Q-state NBD linker to DnaK_R–ATP  

The NBD domain of Q-state DnaK₅₄₀::NR is very similar to that of the DnaK_R–ATP structures, including the association of the NBD–SBD linker with the β12–β13–β14 sheet. The relative dispositions of the subdomains are slightly changed (IA–IIA, IA–IB and IIA–IIB by 1.7°, 3.3° and 2.6°, respectively), but this leaves the distinguishing features of the R state essentially identical. Namely, the linker segment is associated in the same way (Fig. 5 ▸ a), the R–S switch segment has the same conformation (Fig. 5 ▸ b) and the distinctive R-specific hydrogen bonding of Arg71 to Asp231 and the π–π stacking of Tyr145 with peptide 122–123 are also conserved (Fig. 5 ▸ b).

Figure 5.

Comparison of NBD_Q (lime) with NBD_R (pink) demonstrating an R-like conformation in the Q state. (a) NBD–SBD linker. (b) R–S switch segment, ₁₉₆GGGT₁₉₉ segment, Lys70 and Arg71/Glu231/C=O198 latch.

3.1.3. Comparison of the Q-state SBD–NR to NR complexes in SBD_U and DnaK_S–ATP  

The SBD–NR peptide association in Q-state DnaK₅₄₀::NR is nearly identical to that in isolated SBD_U complexes (Fig. 6 ▸ a), as is also the case for DnaK_S–ATP (Fig. 4 ▸ a); the only noticeable differences are in loop , which is not directly involved in peptide binding, and in strand β8, which makes a distinguishing β8–αA junction in the S state compared with the R state. Nevertheless, β8 remains completely hydrogen-bonded to β7 in the Q state, as in the S state, rather than engaging β7 only at its N-terminal end and β5 at its C-terminal end as in the R state.

Figure 6.

Comparison of SBD_Q with SBD_U and SBD_R. (a) Superposition of Q-state SBD (lime) with isolated SBD_U (yellow). The Q-state NBD is surface-rendered and semi-transparent. ATP is in stick representation. (b) Comparison of SBD_Q (lime) with SBD_R (pink) based on superposition of the NBDs as in Fig. 4 ▸(b). (c) Comparison of Q-state SBDβ–αA/B with the R-state SBD superimposed on inner loops and as in Fig. 4 ▸(c). SBD_Q in each panel is oriented as in Fig. 4 ▸(b).

3.1.4. Conformational change induced in SBD_R on binding to the NBD  

Since DnaK_Q–ATP seems to be a hybrid of DnaK_R–ATP and DnaK_S–ATP, with an NBD_Q like NBD_R and an SBD_Q like SBD_S, we have the opportunity to examine how it is that the NBD–SBD interface in DnaK_R–ATP effects the conformational change in SBD_S–NR that leads to the ejection of the bound peptide as ATP binds to uncoupled NBD(apo)–SBD(peptide) complexes. The inner loops , and and helix αA superimpose quite precisely when their NBDs are superimposed (Figs. 4 ▸ b and 6 ▸ b); at the same time, however, the strands diverge and outer loop displacements are substantial. Loops and are spread wide apart from the NBD-proximal loops and in apo SBD_R compared with peptide-bound SBD_Q. The NBD-proximal loops and in all SBD–NR complexes (SBD_U, SBD_S or SBD_Q) have the same conformation as those of SBD_R, and the opened binding site is clearly evident when the inner loops are superimposed (Figs. 4 ▸ c and 6 ▸ c).

Where the SBD–NBD interfaces of DnaK_Q–ATP and DnaK_R–ATP differ is in how DnaK_R–ATP additionally engages loop (Fig. 3 ▸ a versus Fig. 3 ▸ b). It is as if the interactions of Gln442 from SBD_R loop with the NBD backbone at Asp148 work against the fulcrum of contacts at –– to twist the β-sandwich of SBD_S into the β-barrel of SBD_R, thereby opening the site and ejecting the peptide.

3.2. Structure of the DnaK_609-QQQ::NR fusion protein  

While attempting to improve upon our poorly diffracting initial crystals of DnaK₆₀₉::NR, which showed an S state biochemically, we investigated variously lengthened linkers. One, with the sequence DnaK₆₀₉-GGSAQQQTTGSG-NR, which we denote DnaK_609-QQQ::NR for simplicity, was well behaved after biochemical purification and produced acceptable crystals as the hydrolysis-impaired T199A mutant (d _min = 2.82 Å; Table 1 ▸).

The structure is in a P6₄22 lattice that has two molecules in the asymmetric unit, and these are related by approximate twofold symmetry (Fig. 7 ▸ a). Contacts between the crystallo­graphically independent molecules include unnatural interactions of the QQQ linker segment of one with the NBD of the other (Fig. 7 ▸ b). Each NR peptide is bound to the SBD as expected; however, interactions between the NBD and SBD are quite limited (Fig. 8 ▸ a), with SBD loop binding to Asn170 and Thr173 on the NBD as the only direct SBD inter­action with the NBD.

Figure 7.

Crystal structure of DnaK_609-QQQ::NR fusion protein. (a) Dyad-symmetric pseudo-dimeric asymmetric unit. (b) Interface between the QQQ linker segment and the partner NBD. The region of this interface in the pseudo-dimer is indicated by an orange circle in (a).

Figure 8.

QQQ structure and comparison of its NBD linker and state-distinguishing NBD elements with NBD_R and NBD_S. (a) Overall structure. Domain coloring is as in Fig. 3 ▸. (b) Comparison of NBD–SBD linker segments from QQQ (white) and S-state (light blue) structures. (c) Comparison of state-distinguishing elements from NBD_QQQ (white), NBD_R (pink) and NBD_S (light blue) structures: R–S switch segment, ₁₉₆GGGT₁₉₉, Lys70 and Arg71.

The NBD–SBD linker is only partially engaged with the NBD, making just two hydrogen bonds with NBD β14 compared with four in DnaK_S::NR (Fig. 8 ▸ b) and five in DnaK_R (Fig. 5 ▸ a). The NBD conformation here is somewhat intermediate between the NBDs in DnaK_R and DnaK_S (Fig. 8 ▸ c). It is more S-like than R-like (notably, lacking Arg71–Asp231 hydrogen bonding and having Tyr145 active-site engagement); however, its switch segment is not as cleanly aligned with that of DnaK_S as seen for DnaK_Q and DnaK_R (Fig. 5 ▸ b), and the Lys70 and Arg71 residues are intermediate between those of the R and S states (Fig. 8 ▸ c).

The paucity of inter-domain contacts in this Hsp70 conformation seems to be incompatible with allosteric communication. This problem suggested to us that a natural S state has not been realized here, perhaps because it is overcome by the QQQ linker contacts, and it prompted our continued efforts to produce authentic S-state structures (Wang et al., 2021 ▸). Indeed, we now see that the SBD in this QQQ structure is displaced relative to the NBD by ∼70° from that in the S-state DnaK₅₄₀::NR structure.

Nevertheless, this QQQ structure proved to be instructive. Despite having incomplete linker engagement, the NBD–ATP conformation shows S-state hallmarks while departing from being fully S-like (Fig. 8 ▸ c), and the SBDβ–NR structure is nearly identical to that of the isolated SBD_U–NR complex (Zhu et al., 1996 ▸), as are the αA and αB1 portions of SBDα (Fig. 9 ▸ a). However, the QQQ structure differs from that of SBD_U after the middle of αB, which is kinked such that the αB2 segment diverges from that in SBD_U by 23.2° (Fig. 9 ▸ b). This kinking is similar to that found in one copy of the DnaK₆₀₉::NR structure; however, whereas that DnaK_S kinking is caused by NBD–SBD contacts that reorient αA and αB1 relative to SBDβ (Wang et al., 2021 ▸), here it is interference from the SBD–NR linker segment that precludes SBD_U-like extension. The α-lid domains are poorly ordered having high B factors for modeled atoms and unbuildable density for the αB–αC loop and for helix αE (Fig. 9 ▸ a). Kinking of αB in the vicinity of residue 537 was initially seen in the type II SBD–NR structure of Zhu et al. (1996 ▸), and it offers an explanation for the rapid kinetics of peptide association and disassociation from DnaK–ATP (Schmid et al., 1994 ▸).

Figure 9.

Comparison of the SBD portion of the QQQ structure with isolated SBD_U. (a) Side view as in Fig. 6 ▸(a). SBD_QQQ (SBDβ, green; SBDα, red; QQQ linker–NR peptide, gray–orange) is superimposed onto SBD_U (SBDβ, yellow; αA and αB, yellow). The α-lid subdomain of SBD_U is deleted for clarity. Disordered portions of the α-lid subdomain of SBD_QQQ are indicated by dashed lines. Surface renderings of NBD_QQQ and SBD_QQQ are semi-transparent. (b) Top view of (a), except that only SBD_QQQ is shown surface-rendered.

4. Discussion  

4.1. Conformational changes in the Hsp70 chaperone cycle  

With our discovery and characterization of the stimulating state, we now have a refined picture of the Hsp70 chaperone cycle and the allosteric interplay between ATP and client peptide-binding sites (Wang et al., 2021 ▸). A state complexed with ADP and polypeptide has the NBD and SBD in the uncoupled state UP, Hsp70_U (ADP, peptide), with a conformational disposition as in Hsp70_U (apo, peptide) (Fig. 1 ▸, left), and we imagine that all Hsp70_U proteins, whether with a client polypeptide or empty, are generally similar in overall conformation. When ATP displaces ADP or adds to nucleotide-free Hsp70_U, the UP state is transformed to R (Fig. 1 ▸, center) and the peptide is released if present. The R state is in equilibrium with the S state, which has the capacity to bind client polypeptides to yield the SP state (Fig. 1 ▸, right). As for the U and UP states, we imagine that the S and SP states have generally similar conformations. ATP hydrolysis by stimulating the S or SP state generates the U or UP state, respectively, and the cycle repeats. Ultimately, after repeated chaperone cycles, the released polypeptide may be a natively folded protein.

Conformational changes between major Hsp70 states are quite dramatic, as is evident from the disposition of the SBD subdomains relative to the NBD in Fig. 1 ▸. The subdomains move as quasi-rigid bodies, for the most part involving defined ordered-to-ordered transformations. The NBD–SBD linker is an exception; it is poorly ordered in Hsp70_U structures and is integrated with the NBD in Hsp70_R and Hsp70_S structures. Although less readily apparent, changes within the NBD are also substantial between ATP- and ADP-bound or nucleotide-free states and, albeit somewhat subtle, NBD differences in the R state compared with the S state are highly consequential in that they directly restrain hydrolytic activity. Similarly, less apparent changes within the SBD subdomains also have important functional impact; for example, SBDβ_S can bind peptides, whereas SBDβ_R cannot.

Conformational distinctions between Hsp70 states are accommodated by flexibility in subdomain junctures, plasticity at interfaces between subdomains and some intra-subdomain malleability. Similar variations are known or can be expected to occur in distinguishing otherwise similar states, such as U (ADP) versus U (apo) and S versus SP. It is also to be expected that transitions between states may entail passage through well defined intermediate states. The structures that we are reporting here have implications for state-dependent Hsp70 properties and for transitions between the states.

4.2. Implications for the transition from Hsp70_U–peptide complexes to Hsp70_R–ATP  

The DnaK₅₄₀::NR structure reported here is clearly aberrant since the protein purifies as a monomer but yet crystallizes as an entangled, lattice-swapped network. Moreover, sensible crystals of S-state monomers were achieved when the conditions were adjusted to maintain the monomeric state during vapor diffusion. Nevertheless, despite the nonphysiological lattice swapping, the intimacy of NBD–SBD contacts and the striking similarity of its NBD and SBDβ domains to those of DnaK_R–ATP and DnaK_S–ATP, respectively, suggest to us that this hybrid arrangement actually represents a natural intermediate. We know from DnaK (ADP) that the SBD_U–NR component in Hsp70_U (ADP, peptide) is isostructural with the isolated SBD–NR complex (Bertelsen et al., 2009 ▸), and we know from our recent S-state structures that SBDβ_S–NR is also isostructural with isolated SBD–NR (Wang et al., 2021 ▸). Thus, plausibly, as an initial step in the transition from Hsp70_U (ADP, peptide) or Hsp70_U (apo, peptide) to Hsp70_R (ATP), SBDβ preserves its S-state conformation as it engages NBD (ATP) in an R-state conformation. That is, this intermediate may have the hybrid Q-state conformation fortuitously adopted in the present crystal (Fig. 10 ▸).

Figure 10.

Hypothetical Hsp70_Q-mediated transition from an Hsp70_U peptide complex to peptide-discharged Hsp70_R–ATP. Schematic drawings of peptide sites in apo Hsp70_U (left), Hsp70_Q (center) and Hsp70_R (right). Segments from the NBD–SBD linker through β8 are drawn faithfully within schematic SBDβ outlines, which are oriented to give side views along the bound peptide axis. Domain outlines are colored as in Fig. 1 ▸(a), with NBD lobe I in lighter blue and lobe II in darker blue. The NBD and SBDα are oriented faithfully relative to each SBDβ in accord with PDB entries 2kho, 7kzu and 7krt for Hsp70_U, Hsp70_Q and Hsp70_R, respectively. Bound peptides are indicated by orange dots. Adapted from Fig. 7(b) of Wang et al. (2021 ▸).

As the NBD of Hsp70_U binds ATP, it necessarily draws SBD_U–NR into contact with the NBD. Kinetic measurements based on FRET-labeled cysteine mutants indicate that the conformational changes consequent to ATP binding by peptide-free DnaK occur most rapidly at the linker site (107 s⁻¹ at 30°C), then at the SBDβ–NBD interface (27.9 s⁻¹), and most slowly for SBDα–NBD docking (8.3 s⁻¹), with the latter being further slowed to 0.7 s⁻¹ for NR-bound DnaK (Kityk et al., 2012 ▸). The transition from Hsp70_U to Hsp70_R necessarily expels the peptide, if present. The peptide can be retained along the path; however, the α-lid subdomain needs to be displaced for this to happen, as is achieved in the Q-state NBD_R–SBDβ_S–SBDα_R interface, and displacement will be impeded with SBDα–SBDβ latch interactions in place, as for the SBD_U–NR conformation expected in peptide-complexed Hsp70_U. Since the α-lid is flexibly associated in peptide-free SBD, the rate-limiting step of SBDα displacement is understandably slower when the SBD is peptide-bound.

In DnaK, ATP binding can be followed by the quenching of fluorescence from Trp102, its single tryptophan residue. We attribute this quenching to R-state contacts of Trp102 with Leu507 and Met515 (Liu & Hendrickson, 2007 ▸; Qi et al., 2013 ▸), which are interactions that are also in place in the Q state. Using this non-invasive probe of conformation, Slepenkov & Witt (1998 ▸) found that ATP binds to a nucleotide-free DnaK–peptide complex (UP→R) with single-exponential kinetics (4.7 s⁻¹ at 25°C), whereas in the superficially simpler case of nucleotide-free and peptide-free DnaK (U→R) ATP binds with double-exponential kinetics (18.1 s⁻¹ fast phase). The Q-state intermediate may provide an explanation for these puzzling observations. As the SBD is drawn to the NBD when ATP binds, SBDα must disengage from SBDβ to form either the R or Q interface. We speculate that positioning of the peptide-complexed SBD relative to the NBD occurs simultaneously with SBDα displacement, which leaves helix αA/B uniquely oriented for its association with the NBD, hence the single-exponential kinetics. For peptide-free SBD, SBDα–SBDβ engagement is loose and SBDα may make initial Trp102-quenching contacts contemporaneously with SBDβ association, after which αA/B consolidation into the IA–IB interface might occur more slowly.

4.3. Implications for peptide-binding affinity in Hsp70_S–ATP  

The substantially reduced affinity of Hsp70 for client peptides when in the presence of ATP relative to ADP or the absence of nucleotide is a hallmark of Hsp70 allostery. Our theoretical model for Hsp70 allostery (Hendrickson, 2020 ▸) describes both the dependence of ATP hydrolysis on peptide binding and the dependence of peptide binding on ATP binding. The theory proved to be satisfyingly effective for ATP hydrolysis, where data on hydrolysis as a function of substrate-peptide concentration were fitted quantitatively by the R-to-S equilibrium model, with the hydrolytic rate constants (k ⁰ and k′, respectively) and intrinsic peptide dissociation constants for the two states (K _d ^0R and K _d ^0S, respectively) and the equilibrium constant (K _eqS) as definable parameters. Optimal fitting was achieved with no R-state peptide affinity (K _d ^0R = ∞; Wang et al., 2021 ▸). However, the analysis of ATP-dependent peptide affinity was less straightforward.

The simple R-to-S equilibrium model explains the apparent peptide affinity in ATP (defined through K _d ^app) as reduced from the intrinsic affinity in relation to the relative abundances of the species, which is simply K _d ^app = K _d ^0S/(K _eqS + 1) for the case of no R-state peptide affinity. For our particular experimental conditions, we found K _d ^app = 36.7 µM, whereas we deduced K _d ^0S(ATP) = 1.73 µM for the intrinsic affinity in ATP, comparable to K _d ^U(ADP) = 1.64 µM in ADP. Then, from the hydrolysis-fitted value of K _eqS = 65.5, the calculated K _d ^app is 115.0 µM; alternatively, the observed K _d ^app would be explained if, in contradistinction to the hydrolysis fitting, R-state binding occurred with K _d ^0R = 53.0 µM. The existence of the Q state affords an explanation; in it, peptide binding can then take place while still restricting hydrolysis to the basal R-state rate of k ⁰. The data do not suffice to determine the intrinsic Q-state affinity, K _d ^0Q; however, if plausible estimates are taken, the observed K _d ^app constant of 36.7 µM can be explained with the inclusion of a Q-state fraction. The fraction depends on the peptide concentration as well as the equilibrium constants; however, the impact of an R–Q–S equilibrium can be illustrated at zero peptide concentration, [P] = 0. In the absence of peptide, the S-state fraction here is Q _S([P]) = Q _S(0) = 1.5% independent of the Q-state inclusion. The Q-state fraction, Q _Q, ranges from 3% to 18% as K _d ^0Q ranges from K _d ^0S to 5.66K _d ^0S (Hendrickson, 2020 ▸).

4.4. Implications for peptide-binding kinetics in Hsp70_S–ATP  

The kinetics of peptide association and disassociation from DnaK are much faster when in ATP than when in ADP or with no nucleotide (Schmid et al., 1994 ▸; Farr et al., 1995 ▸). This is the case despite the fact that the SBD of DnaK makes nearly identical interactions with an associated NR peptide whether the NBD is complexed with ATP or ADP or is simply not present. This is true when comparing the isolated SBD_U structure (Zhu et al., 1996 ▸) or the DnaK–ADP structure (Bertelsen et al., 2009 ▸) with our DnaK–ATP_S structures (Wang et al., 2021 ▸) or with the QQQ and Q-state structures reported here. What differs is the nature of SBDα association with SBDβ, which affects access to the peptide-binding site.

It was immediately apparent from the crystal structure of DnaK_609-QQQ::NR that its SBD conformation is markedly distorted from that of isolated SBD_U–NR. Although the kink in αB is at nearly the same place here (near C^α 537) as for the isolated SBD in the type II lattice (Zhu et al., 1996 ▸), the α-lid subdomain is more deviated and has greater mobility in this QQQ structure. As it happens, these features are in common with those of the DnaK₆₀₉::NR structure, which we solved subsequently but at lower resolution (PDB entry 7krw; Wang et al., 2021 ▸). This structure has the same NBD linker–SBDβ–αA–αB1 core as found in the high-resolution DnaK₅₄₀::NR structures (PDB entries 7kru and 7krv), but only one of the four α-lid subdomains (αB2–αE) was sufficiently well ordered to be modeled. The kinking of αB in S-state structures seems to be provoked by NBD–SBD contacts that redirect SBD αA and αB relative to SBDβ such that latch contacts cannot form between the outer loops of SBDβ and αB2 of the α-lid subdomain. S-state NBD–SBD contacts have not formed in this QQQ structure and αA and αB1 remain as in SBD_U; however, aberrant contacts with the QQQ linker segment are disposed to block a natural extension of αB2, and kinking occurs in the inherently labile middle of αB.

We expect that S-state-induced kinking of αB in natural Hsp70–ATP complexes will elicit free-ranging rigid-body flexibility of the associated α-lid domains. This cannot happen as freely in DnaK₆₀₉::NR or DnaK_609-QQQ::NR because the NR heptapeptide, as tethered to the α-lid, constrains this end to the peptide site in SBDβ. Consequently, even weak lattice contacts can stabilize the α-lid domain into an essentially arbitrary, but specifically defined, orientation. To our good fortune, in DnaK_609-QQQ::NR, as for the fortuitous protomer of DnaK₆₀₉::NR, a well defined α-lid domain confirms that lid domains move largely as quasi-rigid bodies.

Supplementary Material

PDB reference: DnaK₅₄₀::NR, 7kzu

PDB reference: DnaK_609-QQQ::NR, 7kzi

Funding Statement

This work was funded by National Institutes of Health, National Institute of General Medical Sciences grant R01 GM107462.