A folding nucleus and minimal ATP binding domain of Hsp70 identified by single-molecule force spectroscopy

Significance

Proteins with a similar structure can have largely different folding properties. Although some fold readily, others can only assume their native structure through the help of chaperone proteins. Partially folded intermediates play a key role in defining those folding differences. However, owing to their transient nature, they are not amenable to the structural investigation. Using a combination of single-molecule mechanics, protein engineering, and crystallography, we identified a stable native-like functional nucleus, which is a critical intermediate for spontaneous folding of the Hsp70 nucleotide-binding domain. Based on our findings, we engineered a chimera turning a homologous but folding-incompetent protein into a spontaneously folding protein that is enzymatically active. Our results have implications for the folding of actin from the same superfamily.

Abstract

The folding pathways of large proteins are complex, with many of them requiring the aid of chaperones and others folding spontaneously. Along the folding pathways, partially folded intermediates are frequently populated; their role in the driving of the folding process is unclear. The structures of these intermediates are generally not amenable to high-resolution structural techniques because of their transient nature. Here we employed single-molecule force measurements to scrutinize the hierarchy of intermediate structures along the folding pathway of the nucleotide binding domain (NBD) of Escherichia coli Hsp70 DnaK. DnaK-NBD is a member of the sugar kinase superfamily that includes Hsp70s and the cytoskeletal protein actin. Using optical tweezers, a stable nucleotide-binding competent en route folding intermediate comprising lobe II residues (183–383) was identified as a critical checkpoint for productive folding. We obtained a structural snapshot of this folding intermediate that shows native-like conformation. To assess the fundamental role of folded lobe II for efficient folding, we turned our attention to yeast mitochondrial NBD, which does not fold without a dedicated chaperone. After replacing the yeast lobe II residues with stable E. coli lobe II, the obtained chimeric protein showed native-like ATPase activity and robust folding into the native state, even in the absence of chaperone. In summary, lobe II is a stable nucleotide-binding competent folding nucleus that is the key to time-efficient folding and possibly resembles a common ancestor domain. Our findings provide a conceptual framework for the folding pathways of other members of this protein superfamily.

Modern structural biology has provided us with more than 100,000 3D structures of proteins. On the basis of their structural similarity, they are categorized into so-called protein superfamilies. Even though a large amount of valuable information can be obtained from these static structures, they miss a crucial piece of information about how these structures are formed during the folding process. Often, on the pathway from the unfolded to the folded state, transiently populated structures, called folding intermediates, are populated (1–6). Structural information on folding intermediates is very rare mainly because of their elusive nature. However, the folding intermediates can have a dramatic effect on the timescales over which folding occurs. Interestingly, these intermediates can be differently populated even among homologous, structurally highly similar proteins. Hence, structurally similar proteins can have dramatically different folding rates; such cases are exemplified by the Im7 and Im9 proteins (7); spectrin repeats R15, R16, and R17 (8, 9); and the β2-microglobulin and antibody light chain (10). As reported in these cases, the refolding time of the homologous proteins slows down because of a different extent of folding intermediate population.

The sugar kinase family of proteins also contains members with very diverse folding properties. Examples are important proteins such as actin, hexokinase, or the nucleotide binding domain (NBD) of Hsp70 proteins (11). The overall architecture of these proteins is composed of two subdomains named lobe I and lobe II. These lobes can further be subdivided into subdomains a and b (lobe Ia and Ib and lobe IIa and IIb). The two lobes are engaged via a C-terminal helix and a loop that connects lobes directly. The sugar kinase family proteins bind Mg-ATP in a highly conserved ATP binding site, which is mainly located between the interface of lobe IIa and IIb (12, 13). Sequence and structural alignment studies suggested that these homologous proteins share common evolutionary history (11, 14) and a common ancestral domain consisting of only one lobe. In this evolutionary picture, a stable independently folding unit, which can bind ATP, can dimerize and thus form the ancestral protein domain. After gene duplication, two lobes have been combined in a single polypeptide chain, which has been further evolved in the process of divergent evolution to an arsenal of structures with different functionalities, as we know it today (11).

Hsp70 NBD from Escherichia coli DnaK (EcNBD, 44 kDa) and mitochondrial Hsp70 NBD from Saccharomyces cerevisiae (mtNBD, 44 kDa) belong to the sugar kinase family. Using single-molecule mechanical force experiments in combination with MD simulations, we previously showed that under an applied force, EcNBD and mtNBD unfold through an identical series of unfolding intermediates. The first unfolding intermediate corresponds to unraveling of the C-terminal helix that guarantees the high stability of both proteins (15). The unfolding sequence of the remaining two lobes depends on the presence of nucleotide; the binding of ligand switches the unfolding pathway by providing more stabilizing interactions with lobe II compared with lobe I. In contrast, early qualitative data indicated a significant difference in refolding behavior of EcNBD and mtNBD: EcNBD refolds fast in the absence of chaperone (16), whereas yeast mtNBD alone does not refold in vitro, and its refolding critically depends on Hep1 and ATP (17, 18). Why do these homologous proteins fold on different timescales, and what are the detailed folding pathways?

In this report, we show that lobe II is a critical folding intermediate that already can bind nucleotide and promotes successful folding of the EcNBD to the native state. We used these insights to create a folding-competent bacterial-yeast chimera by replacing lobe II of mtNBD with the respective EcNBD sequence (50% mtNBD). We show that isolated lobe II is a stable protein and determined its crystal structure in the presence of nucleotide. Because lobe II alone fulfills structural and certain functional requirements, it is a potential candidate for a common ancestor domain of the sugar kinase family.

Results

Single-Molecule Force Experiments Reveal the Refolding Pathway of EcNBD of DnaK.

To study refolding pathways of the nucleotide binding domain of DnaK (EcNBD, aa 1–394), we introduced cysteines at either terminus (EcNBD-nc) that served as attachment sites for maleimide-functionalized single-stranded oligonucleotides in an optical trap assay. These single-stranded oligonucleotides hybridize with the complementary 3′ single-stranded overhang of DNA handles containing either biotin or a digoxigenin group at the 5′ end (Fig. 1A and SI Appendix, SI Methods). To produce bead–DNA–protein–DNA–bead tethers, the construct was incubated with streptavidin- and α-digoxigenin-coated silica beads. For single-molecule force experiments, two single beads were trapped in two separated laser foci (called mobile and fixed traps); the position of the mobile trap was controlled by a steerable mirror. After the beads’ surfaces got in contact, they were separated for verification of a single tether connection between the beads. A single tether shows the expected DNA contour length and the typical force-extension curve for DNA (19). In a first set of experiments, we continuously stretched and relaxed the tethered single EcNBD-nc molecules at a pulling velocity of 200 nm/s with a 1-s waiting time between the individual cycles. The sequence of stretching cycles of a single NBD molecule showed two patterns of force-extension curves with different frequency. Pattern 1 (Fig. 1B, 22/100, pink) exhibited unfolding forces of 33 ± 4 pN and contour length increases of ∼134 nm, which is the unfolding pattern for native EcNBD-nc (15). Pattern 2 (Fig. 1B, 78/100, orange/red/gray) exhibited two unfolding steps at forces ∼5 pN and ∼7 pN with a combined contour length increase of ∼61 nm, which we attributed to the unfolding of refolding intermediates. From the frequency of occurrence of pattern 1, the folding kinetics to the native state can be estimated by plotting the fraction of natively refolded EcNBD-nc (pink traces) in dependency of the waiting time at zero force (Fig. 1C), yielding a folding rate of 0.03 ± 0.02 s⁻¹. To understand the relation between patterns 1 and 2, we overlaid force-extension curves for the native NBD and refolding intermediates (Fig. 1D). We observed that both patterns have the same contour length in their unfolded states, but different refolded lengths. Thus, pattern 2 reflects the unfolding of two refolding units of EcNBD-nc, which we call RFI1 and RFI2. In summary, the refolding process of EcNBD-nc involves refolding intermediates (Fig. 1B) and needs 30–40 s for completion (Fig. 1C). Several stretching/relaxing attempts and long waiting times at zero force are necessary to observe the native state (Fig. 1C and SI Appendix, Fig. S1A).

Fig. 1.

Refolding of NBD investigated by optical tweezers experiments. (A) Optical tweezers assay setup. EcNBD protein is tethered N and C terminally to dsDNA handles between two trapped beads. The mobile trap can be moved to exert a force on the molecule. (B) Series of measured stretching cycles: a fingerprint of the native protein (pink), and refolding intermediates (RFI1, red circle; RFI2, yellow star). Force-extension traces were recorded at a pulling speed of 200 nm/s and waiting time at zero force of 1 s. (C) The plot of natively refolded fraction dependent on the waiting time at zero force [EcNBD-nc (pink) and mtNBD (green)]. Fitting the data with a simple exponential equation gave a refolding rate for EcNBD-nc of 0.03 ± 0.02 s⁻¹; mtNBD does not refold in our experiments. (D) Contour lengths measured for the refolding intermediates RFI1 (red circle) and RFI2 (yellow star). Native unfolding (gray) shows, first, stretching of DNA (L_C = 0 nm), followed by the whole NBD unfolding (L_C = 134 nm). (E) Scatter plot of all unfolding events (n = 10 molecules). Unfolding of mature NBD (pink) is characterized by an unfolding force of 34 pN and a contour length change of ∆L_C = 134 nm (15). The refolding intermediates RFI1 (red circle) and RFI2 (yellow star) unfold at ∼5 and ∼7 pN, respectively. (F) Force-extension traces of the unfolding of isolated lobe II (aa 183–383). Unfolding of intact lobe II (pink, left), the next stretching shows the RFI1 and RFI2 unfolding (red circle and yellow star, right). (G) Scatter plot of the unfolding events in NBD 183–383. (H) Illustration of analyzed loop insertion variants Ins183 (K183-L20), Ins290 (A290-L20), and Ins364 (D364-L20). (I) Contour length transformation plots of EcNBD-nc, Ins290, Ins183 and Ins364 NBD variant. Ins183 and Ins364 show the same contour length for RFI1 and RFI2 as the EcNBD protein. The contour length differences for U-RFI1 in InsA290 are increased compared with EcNBD-nc (30.3 vs. 23.9 nm).

Unfolding of EcNBD-nc refolding intermediates yielded contour lengths of L_C = 73.0 ± 7.2 nm for RFI2 (orange, star), L_C = 110.1 ± 4.7 nm for RFI1 (red, circle), and L_C = 134.3 ± 1.9 nm for the unfolded protein (gray; Fig. 1D, scatter plot Fig. 1E). Although RFI1 and RFI2 unfold typically at 6.7 ± 2 pN and 4.6 ± 2 pN, respectively, the native state is more stable and unfolds at 33 ± 4 pN. Also, we observed rarely populated RFI3 states (SI Appendix, Fig. S1 C and D). However, these RFI3 states are not obligatory on a productive folding pathway, as RFI2 can directly proceed to the native state (SI Appendix, Fig. S1F).

Next, we asked whether RFI1 and RFI2 can form independently or whether they follow an obligatory sequential pathway, where RFI1 is needed to fold RFI2. To populate refolding intermediates, we performed constant velocity experiments at different pulling speeds and without waiting time at zero force. Although RFI1 was present in all cycles, RFI2 was sometimes not populated and only formed when RFI1 was already folded (SI Appendix, Fig. S1 A and B). We conclude that these intermediates follow an obligatory sequential pathway U → RFI1 → RFI2, indicating that the RFI2 comprises folded RFI1.

Structural Assignment Identifies Lobe II as a Structured Portion of the Refolding Intermediate.

In the next set of experiments, we set out to determine the structural identity of the refolding intermediates RFI1 and RFI2. Judging from the contour length changes measured for unfolding of RFI1 and RFI2 (ΔL_C = 23.9 ± 1 nm and 37.1 ± 1 nm, respectively), we estimated that 68-aa residues form the folded portion of RFI1, and 106-aa residues form the folded core of RFI2. RFI1 and RFI2 could, therefore, comprise either lobe I (residues 1–183) or lobe II (residues 184–383). To investigate whether lobe II comprises RFI1 and RFI2, we designed a protein variant including only aa 183–383 (EcNBD-lobeII; Fig. 1F). For EcNBD-lobeII, the unfolding of the native state occurs at high load around 33 pN (pink trace in Fig. 1F), similar to the EcNBD-nc. The contour length gain measured on unfolding is shorter, reflecting the stretch of 200 aa residues between the pulling positions (183 and 383). Subsequent stretch-and-relax cycles revealed that both unfolding events correspond to RFI1 and RFI2 (Fig. 1G) with the same contour length increases and forces as for pattern 2 in the complete EcNBD-nc (Fig. 1E). In summary, EcNBD-lobeII is a well-behaved protein, and the folded portions of both refolding intermediates (total 174 residues) lie entirely in the lobe II region (193 residues).

To further characterize specific regions within EcNBD that are part of the folded portions of RFI1 and RFI2, we used previously designed and characterized loop insertion variants of EcNBD-nc (15). The three variants, Ins183, Ins290, and Ins364, carry a 20 Gly-Ser-rich insert (20) at aa positions 183, 290, and 364, respectively (Fig. 1H). Those variants help to define the boundaries for RFI1 and RFI2: Ins183 probes the domain interface between lobe I and II, Ins290 probes lobe IIb, and Ins364 probes the C-terminal region of lobe IIa. If the insert lies within the region folded in RFI1 or RFI2, we expect an increase of 7 nm in contour length for the respective unfolding peak because of the 20 aa inserts. For a better visualization of the contour length increases, we transformed the force-extension traces into contour length vs. time trace plots (SI Appendix, Fig. S2) (21). Here, all protein states with the same contour length of the unfolded polypeptide are on the same horizontal line, independent of the applied force. Compared with EcNBD-nc, Ins183 and Ins364 exhibited identical contour lengths for both RFI1 and RFI2, demonstrating that both RF intermediates lie between those two positions (Fig. 1I) and do not involve the C-terminal helix (residues 369–383). For Ins290 (red), the contour length increases upon unfolding of RFI1 by 7 nm (ΔL_C = 30.3 ± 0.7 nm vs. ΔL_C = 23.9 ± 1 nm (EcNBD-nc), indicating that RFI1 comprises the region around residue 290. We conclude that RFI1 comprises lobe IIb (residues 229–306), and RFI2 comprises the entire folded lobe II (residues 184–363) (unfolding lengths are listed in SI Appendix, Tables S1 and S2).

Kinetic Analysis of the Refolding Pathway of EcNBD.

To obtain microscopic rates of the transitions along the refolding pathway, we performed passive mode experiments on the EcNBD-nc in which we held the two trap positions constant and allowed the protein to fluctuate between accessible folded conformations. The observed sequential order of folding/unfolding events of RFI1 and RFI2 validates our previous conclusions from constant velocity cycles (SI Appendix, Fig. S1 A and B). At high forces (6.0 pN; upper trace in Fig. 2A), we found the exchange between the unfolded state and RFI1 (gray and red, respectively). When lowering the force to 4.4 pN, all three states are populated (Fig. 2A, Middle). RFI2 always exchanges with RFI1, but never transitions directly to the unfolded state. In addition to the refolding intermediate RFI2 (orange), we found an additional state, which is of similar contour length as RFI2 but with much faster unfolding rate (green), which we call RFI2f (low population probability of <1%). RFI2f forms only from RFI1, and does not bind nucleotides (Fig. 2 B and C). The force-dependent folding/unfolding kinetics for RFI1 and RFI2 are plotted in Fig. 2B. Detailed zero-force folding and unfolding rates of all populated states are listed in SI Appendix, Table S3 (SI Appendix). Force-dependent population probabilities (Fig. 2B) yield folding free energies of −15.2 k_BT for RFI1, −6.6 k_BT for RFI2, and −3.6 k_BT for RFI2f (SI Appendix, Table S4).

Fig. 2.

Refolding kinetics of EcNBD-nc and micromolar ATP binding capability of RFI2. (A) Refolding trajectories acquired at different force biases. (B) Chevron plots of all force dependent refolding transitions (circles) in apo with extrapolation to zero force rates (lines). Force-dependent probability plot with extrapolation to zero force. (C) Passive mode experiments at varying ATP concentrations at 4.0 pN show the binding of ATP to RFI2, which results in a longer RFI2 life-time. (D) Mg²⁺ATP interaction in the NBD structure using LigPlot software (38). (E) Chevron plot of the RFI1 ↔ RFI2 transition. Fits represent extrapolations of the force-dependent rates to zero force. Data points: force-dependent rates in the presence of 200 µM ATP and 5 mM MgCl₂ (RFI2 → RFI1, orange hexagons; RFI1 → RFI2, red hexagons; ATP-free, gray). The presence of ATP slows down the RFI2 unfolding rate (yellow hexagon). (F) Dependence of the transition rate (RFI2 → RFI1, yellow circles) on the ATP concentration at 4 pN. A fit to the data with the binding-unfolding-model (yellow line, fit, SI Appendix) yields a K_D of 33 µM, which is verified by simulations (brown circles).

Refolding Intermediate 2 Is Natively Folded and Binds Nucleotides.

Nucleotides specifically bind to lobe II (Fig. 2D) in the NBD, which can be used to probe how native-like RFI1 and RFI2 intermediates are. We conducted passive mode experiments at 4.0 pN in the presence of varying Mg-ATP concentrations (Fig. 2C). RFI2 is stabilized significantly in the presence of Mg-ATP (Fig. 2 E and F), whereas the RFI1 (red) and RFI2f (green) lifetimes are not affected (SI Appendix, Fig. S3), confirming that RFI2 comprises the natively folded lobe II.

To determine the ATP binding affinity of RFI2, we analyzed the RFI2 unfolding rate as a function of the ATP concentration in an equilibrium model (SI Appendix, Fig. S4A), assuming that unbinding of ATP is much faster than unfolding of RFI2 (RFI2/ATP-bound ↔ RFI2/ATP-free ↔ RFI1; for details, see SI Appendix, Table S5). The model yielded an ATP affinity to RFI2 of 33 ± 10 µM (Fig. 2F), which is about 50 times lower than for EcNBD (K_D = 0.7 µM) (22, 23).

To justify the approximation of fast exchange and obtain information on the nucleotide binding kinetics, we performed kinetic simulations of the lifetime distributions at different ATP concentrations, based on a microscopic model for folding/unfolding and nucleotide exchange (SI Appendix, Fig. S4A and SI Methods). Our model qualitatively reproduced measured trajectories in the absence (compare SI Appendix, Figs. S2A and S4C) and presence (compare Fig. 2C and SI Appendix, Fig. S4D) of nucleotide, and reproduced the measured nucleotide-dependent unfolding kinetics of RFI2 (Fig. 2F and SI Appendix, Fig. S4 E and F). We used the kinetic model to deduce information on the ATP exchange kinetics. Although in the case of a fast exchange, all instances of RFI2 become stabilized by ATP, slow binding of nucleotide results in the emergence of an additional population of RFI2 that unfolds before it can bind ATP, resulting in double-exponential distribution RFI2 lifetimes (SI Appendix, Fig. S4G). Observing that the experimentally determined lifetimes of RFI2 are single-exponential, we thus infer that the exchange of ATP is fast. By comparing model predictions with experimentally observed lifetime distributions for RFI2 (SI Appendix, Fig. S4H), we conclude that k_{on, ATP} > 10^4.1 M⁻¹⋅s⁻¹ (P < 0.05, Kolmogorov-Smirnov test).

Previous experiments with native EcNBD-nc showed different binding affinities for ATP, ADP, and AMP (24). We tested whether the RFI2 folding intermediate can discriminate nucleotides depending on their number of phosphate groups, but could not detect any differences (SI Appendix, Fig. S3), suggesting that lobe II cannot discriminate among ATP, ADP, or AMP.

Lobe II Is a Properly Folded Minimal ATP Binding Domain.

From the results obtained so far, we have concluded that lobe II alone forms a minimal ATP-binding domain. To determine whether lobe II can exist as an independently folded ATP binding domain, we produced a protein construct comprising aa 183–383 of EcNBD (Mini-NBD; Fig. 3A). Mini-NBD is a stable and well-behaved protein and shows the same refolding intermediates, RFI1 and RFI2, as in EcNBD-nc, identified by optical tweezers experiments (Fig. 3 B and D). In the presence of Mg-ATP, RFI2 in Mini-NBD is stabilized and shows a clear shift to higher unfolding forces (Fig. 3C) and an increased lifetime (Fig. 3E) compared with EcNBD-nc (SI Appendix, Fig. S1E). Our experiments show that isolated Mini-NBD is a natively folded ATP binding domain. We determined the crystal structure of Mini-NBD in the presence of AMP-PCP at 2.9-Å resolution, with R_work/R_free values of 24.4/26.0% (Fig. 3F and SI Appendix, Table S7). The Mini-NBD structure is very similar to full-length Mg-ATP-bound DnaK (PDB: 4B9Q). They superimpose with C_α rmsd of 0.7 Å (25) (Fig. 3 G and I). Minor deviations are only found for the N- and C-terminal residues (region A and D in Fig. 3G), as well as for flexible loops (region B and C in Fig. 3G). Residues involved in nucleotide binding are structurally well conserved in Mini-NBD (Figs. 2D and 3 H and I, red arrows), whereas the residues, which contact lobe I in EcNBD, have higher dynamics.

Fig. 3.

Lobe II is an independent stable folding domain and able to bind nucleotide. (A) Design of Mini-NBD (aa 183–383). (B and C) The sequence of N-/C-terminal stretching cycles with a pulling speed of 200 nm/s and waiting time at zero force for 1 s in the absence of ATP (B) and the presence of 200 µM ATP and 5 mM MgCl₂ (C). (D and E) Passive mode experiments at 3.7 pN in the absence (D) and presence of 200 µM ATP and 5 mM MgCl₂ (E) RFI1 (red), RFI2 (orange) and RFI2f (green) are the populated states in this experiment with contour lengths of 36 nm (RFI1) and 0 nm (RFI2/RFI2f). (F) Crystal structure of AMP-PCP bound Mini-NBD at a resolution of 2.9 Å. (G) Superposition of the isolated lobe II domain with DnaK (aa 1–394) (PDB: 4B9Q) (rmsd = 0.7 Å). Flexible regions A, B, C, and D are indicated in circles. (H) LigPlot (38) analysis of AMP-PCP bound to Mini-NBD. The numbering is according to the EcNBD (PDB: 4B9Q). (I) Rmsd per residue plot. In gray, the unresolved structure is marked, and residues marked in bold (red ellipse) are involved in the nucleotide coordination. The overall rmsd of the structure alignment is 0.7 Å (25).

NBD of a Mitochondrial Hsp70 Chaperone Is Not Able to Refold in Vitro.

A close evolutionary relative of EcNBD is mitochondrial Hsp70 NBD [mtNBD; >60% sequence identity, emboss needle online tool (26)]. This protein is known to fold only in the presence of the cochaperone Hep1 in vivo (27), and requires specialized yeast chaperones for functional recombinant bacterial expression (18). Indeed, in a single-molecule unfolding/refolding experiment, we found that mature mtNBD, after unfolding, never folds back to the native state (Fig. 4A). In contrast to EcNBD-nc from DnaK (yellow and red dots in Fig. 4B), we did not find defined refolding intermediates, as seen previously for EcNBD-nc, but rather a continuum of weakly collapsed states that scatter over a large range of forces and contour lengths (Fig. 4B, green triangles). Interestingly, the addition of ATP and/or cochaperone Hep1 did not improve the folding yield of mtNBD in our optical tweezers assay (SI Appendix, Fig. S5 E and F).

Fig. 4.

Insertion of lobe II of EcNBD can create a folding-competent mtNBD structure by acting as a folding nucleus. (A) Force-extension traces of mtNBD at a pulling velocity of 500 nm/s. Shown are only stretching cycles. First, natively folded mtNBD unfolds, followed by the unfolding of weakly formed structures that do not fold back to the native fold even after 10 min. (B) Scatter plot of force vs. contour length shows the broad scattering of refolding intermediates for mtNBD (green). To compare, the events for EcNBD are shown in red and orange (RFI1 and RFI2). (C) Design of the 50% mtNBD chimeric protein consisting of lobe II from EcNBD, lobe I, and C-terminal helix from mtNBD. (D) Force-extension stretching traces at a pulling speed of 200 nm/s show the characteristic refolding events of lobe II. After ∼230 s, the chimeric protein can refold completely to its stable native fold.

Rescue of a Refolding Incompetent mtNBD by Replacing Lobe II.

We wondered whether the refolding-incompetent mtNBD could be rescued by replacement of lobe II with the corresponding EcNBD sequence. We produced a variant consisting of lobe I of mtNBD and lobe II from EcNBD (SI Appendix, Fig. S5A), which we term 50% mtNBD (Fig. 4C). Single-molecule force experiments showed that this chimera exhibits the stable native conformation with an average unfolding force of (33 ± 2 pN, n = 4), which is the same as observed for the native state of EcNBD-nc and mtNBD (15). Moreover, we observed the defined refolding intermediates RFI1 and RFI2. After several refolding attempts, the chimeric protein refolded robustly into the native state (Fig. 4D). In addition, we also observe the nucleotide-dependent unfolding pathways and intermediates, as reported previously (15) (SI Appendix, Fig. S5B). The overall refolding rate into the native state is accelerated in the presence of nucleotide [0.005 ± 0.0003 s⁻¹ (n = 3) for apo and 0.02 ± 0.005 s⁻¹ (n = 9) for 1 mM ATP, 5 mM MgCl₂]. We conclude that E. coli lobe II acts as a folding seed for the mtNBD structure.

To further narrow down the size of the folding seed, we designed three additional chimeric proteins. First, we tested whether E. coli lobe IIb is already enough to fully recover the refolding-incompetent mtNBD (75% mtNBD). However, this chimera did not result in a folded, functional protein (SI Appendix, Fig. S6 A–C). Substituting only subdomain IIb together with an E. coli-specific stretch of 11 aa from lobe IIa (75% mtNBDΔ ²³⁶NGVFEVKS²⁴³) or the 11 aa from lobe IIa only (mtNBDΔ ²³⁶NGVFEVKS²⁴³) did result in a soluble but not natively folded protein (SI Appendix, Figs. S5G and S6 D and E). These results demonstrate that at least residues 184–363 are necessary to promote the folding of NBD.

Discussion

NBD Folds via a Hierarchical Multistep Pathway.

Of the two major Hsp70 domains (NBD and SBD), NBD is structurally the more complex one, responsible for ATP-regulated function. In the nucleotide-free and ADP-bound states, NBD and SBD act independently, and hence our results obtained on isolated NBD have relevance also for the full-length protein.

Using high-resolution force spectroscopy optical tweezers, we identified a collection of refolding intermediates of EcNBD along the way from the fully stretched unfolded chain to the completely folded native state. First, lobe IIb folds rapidly in a single step. Next, the folded lobe IIb supports the refolding of lobe IIa, leading to the completion of the entire lobe II within milliseconds. This hierarchical refolding pathway is necessary because of the sequence topology of EcNBD (Fig. 1A). Lobe IIb is inserted into lobe IIa, and hence its refolding brings the ends of lobe IIa into proximity, which permits the folding of lobe IIa. In contrast, lobe I does not fold independently. Folded lobe II serves as a nucleus for lobe I folding and the integration of the stabilizing C-terminal helix, which is a 1000 times slower process (overall folding rate, ∼0.03 s⁻¹). The folding of lobe II is an obligatory, on-pathway intermediate, possibly also in other sugar kinase proteins. Moreover, we found that lobe II is an independently folding domain and binds ATP with micromolar affinity. Although assembly of lobe II mostly seems to occur sequentially, we cannot exclude that short-lived nonnative intermediates such as RFI2f also form and decay on the way to the folded state, consistent with a more complicated underlying energy landscape (28), leading to a multistate kinetic partitioning (29).

Consistent with our findings, Shomura et al. (30) reported that lobe II of human Hsp70 forms a stable complex with the cochaperone BP1c after proteolytic digestion of full-length NBD. They could derive a crystal structure of human Hsp70 lobe II in complex with HspBP1 in the AMP-PCP bound state (PDB: 1XQS) (30). However, we find that lobe II from DnaK-NBD folds spontaneously even in isolation.

What is the energy contribution for ATP binding of lobe I? Comparing ATP binding K_D values of a construct containing lobe I (K_{D EcNBD} = 0.7 µM) and a construct without lobe I (K_{D Mini-NBD} = 33 µM), ATP binding free energies under standard conditions can be approximated: −14.2 k_BT for EcNBD and −10.4 k_BT for Mini-NBD. The energetic difference of both can be attributed to the presence or absence of lobe I residues and their interactions with the nucleotide. Thus, assuming the binding energies are additive, the contribution of lobe I in EcNBD to ATP binding is −3.8 k_BT.

We demonstrate that lobe II is sufficient to bind nucleotides and folded lobe I is not necessarily required for binding. In addition to ATP binding, lobe II is also able to bind ADP and AMP with similar affinity as it binds ATP (SI Appendix, Fig. S4). Such low discrimination among mono-, di- and triphosphates contrasts with the nucleotide selectivity of full-length NBD (22). Apparently, lobe II does not distinguish between ATP and its hydrolysis products. The contact map between ATP and its coordinating residues of EcNBD (Fig. 2D) offers an explanation: Although the adenosine moiety is exclusively contacted by residues from lobe II [red (IIb) and orange (IIa) subunits of EcNBD], residues from lobe I (blue) only contact the phosphates. The precise kinetic characterization of on and off rates for the nucleotides to RFI2 remained elusive; our kinetic simulations suggest a lower boundary for the off-rates of 3 s⁻¹. The reported off-rate for ATP from full-length EcNBD is 0.019 s⁻¹ (31). The lack of folded lobe I, therefore, increases the off-rate by about 100-fold.

E. coli Lobe II Accelerates Refolding of the Refolding-Incompetent mtNBD Protein.

The refolding pathway we deciphered for EcNBD proceeds via structuring of lobe II and folding of lobe I onto the preformed lobe II together with the insertion of the C-terminal helix to a natively folded EcNBD protein. The last two steps are rate-limiting and occur at zero force. In contrast, the nucleotide binding domain from mitochondrial Hsp70 (mtNBD) cannot refold in our in vitro experiments (green symbols Figs. 1C and 4A), despite its structural similarity to EcNBD, probably because of the absence of RFI1 and RFI2. Hence, lobe II does not fold robustly. In contrast to a study by Blamowska et al. (32), we did not observe any native refolding in the presence of Hep1 and ATP (SI Appendix, Fig. S5 E and F), which might be a result of a different experimental setup. Aggregation-prone regions in lobe II of mitochondrial Hsp70 have been identified in an in vivo import assay. This assay is based on the observation that after import into the mitochondrion, mtHsp70 aggregates when Hep1 is absent. Thus, the assay can identify variants with improved solubility, and hence are presumably natively folded (33). In that study, the authors found that the exchange of only an 11-aa-long stretch in lobe IIa with the respective sequence from EcNBD was enough to obtain a soluble mtHsp70 variant. In our experiments, the same 11-aa exchange in the mtNBD yielded soluble but folding-incompetent protein. Similarly, the substitution of lobe IIb and substitution of lobe IIb and the 11-aa-long stretch did not result in a natively folded protein (SI Appendix, Fig. S6). A chimera variant of EcNBD and mtNBD folded successfully only when the complete lobe II region was replaced (Fig. 4 C and D). Apparently, the preformed native lobe II is an important nucleus that critically drives successful folding of the chimeric protein. We conclude that the key to the successful folding of EcNBD and mtNBD lies in the folding ability of the lobe II region.

The Physiological Relevance of Different Protein Refolding Timescales.

Why has nature evolved a largely different timescale for folding of EcNBD and mtNBD? We speculate that the different refolding timescales may be a result of their distinct functional requirements and their respective subcellular localization in the cytosol and mitochondrial matrix.

For DnaK, robust, hierarchical, and well-defined (un)folding pathways are advantageous under stress conditions. Winter et al. (34) showed that oxidative stress and elevated temperatures result in the oxidation of a cysteine residue in lobe I of DnaK, which leads to inactivation. This inactivation can be easily and efficiently rescued by strong reductants and high ATP concentrations. The same inactivation process was obtained on glutathionylation as a result of oxidative stress and heat shock (35). Our data now shed light on possible reasons of such fast rescue: Lobe II, the folding nucleus, has remained intact and hence can significantly support efficient and nondelayed protein reactivation. In contrast, yeast mtHsp70 lacks this cysteine in lobe I. Hence, this oxidative deactivation mechanism is not required for yeast mtHsp70, and lobe II is not required to provide fast refolding times.

Here we identified and characterized a stable folding nucleus for NBD of Hsp70, which is capable of binding nucleotide. We propose that this independently folding unit is a putative candidate for a common ancestral domain of the sugar kinase family. In an evolutionary context, gene duplication and evolutionary modification led to the divergent development of lobe I and II, with lobe I adopting responsibility for the discrimination between nucleotides and allosteric coupling to the substrate binding domain.

An evolutionary scenario of the development of sugar-kinases family members was proposed by Flaherty et al. and Bork et al. (11, 12) in the 1990s: an ancestral ATP-binding domain was the first protein that built all the members of this family by gene duplication and divergent evolution of the two domains forming the lobe–lobe architecture known today. A similar picture for glucose kinases is proposed by Kawai et al. (36). Our experiments locate lobe II as a potential candidate for this ancestral domain because of its stability and nucleotide binding affinity, which would be beneficial for positive selection during the evolutionary process.

In the case of actin, another member of the sugar kinase family, lobe I evolved to promote polymerization and the formation of filaments. Our results identifying lobe II as a seed that can rescue folding-incompetent variants of Hsp70 NBDs may in the future help engineer a fully folding-competent actin variant, independent of cochaperones. This would greatly facilitate many in vitro structure function studies of the actin cytoskeleton.

Methods

All experiments were performed using a custom-built, high-resolution back focal plane detection optical tweezers setup, as published previously (15, 37). Briefly, the constructs were genetically modified to serve cysteine residues for the attachment of the required double-stranded DNA handles (185 nm). These DNA handles carried the respective modifications on each end to ensure coupling to the 1 µm functionalized beads. Those beads could be trapped in our optical tweezers setup and manipulated in a so-called passive mode or constant velocity measurement mode. From constant velocity measurements, we were able to identify the unfolding and refolding pattern of our proteins and the corresponding contour length of each event. Using passive mode measurements, we were able to extract equilibrium energetic and kinetic parameters of our protein system without and in the presence of nucleotides. All experiments were performed in 50 mM Tris⋅HCl, 150 mM KCl at pH 7.5, and if necessary, 5 mM MgCl₂, 10 µM–1 mM ATP or ADP or AMP was added. For a more detailed description of materials and methods used in this work, see SI Appendix.