The 70 kDa heat shock protein (Hsp70) family of molecular chaperones is crucial for biogenesis and protein homeostasis1–4. The Hsp70s account for up to 4% of total cellular protein mass, making them one of the most abundant proteins in the cell5–7. Hsp70s are involved in diverse cellular processes 3,8–12, including de-novo protein folding at the ribosome13,14, protein translocation through pores15,16 and solubilization of protein aggregates1,17,18. Consistently, Hsp70s are connected to multiple pathophysiological conditions including cancer and neurodegenerative diseases19–21. In the endoplasmic reticulum (ER), BiP (Binding Immunoglobulin Protein) is the sole Hsp70 isoform in all eukaryotes22,23 and the most abundant ER chaperone7,24. BiP is the central functional hub of the ER chaperone network that ensure protein folding homeostasis in the “folding factory of the cell”. It consequently binds to most of the proteins that are processed in the ER, promoting their folding and preventing their aggregation25,26. Additionally, it also acts as the central regulator of the unfolded protein response (UPR) by binding to the UPR sensors in a stress-dependent manner27–29. Furthermore, BiP targets unfolded protein to the degradation machinery associated with the ER-associated degradation pathway (ERAD)30–32. Moreover, BiP is overexpressed in many human cancers, making it a major therapeutic target33–36.
Hsp70 chaperones comprise two distinct domains, the nucleotide-binding domain (NBD) and the substrate-binding domain (SBD) that are connected by a flexible linker37,38. The SBD is sub-divided into the subdomains SBDα and SBDβ enclosing the client binding site12. The NBD has a clamp-like shape with two lobes I and II12. Each lobe is further subdivided into two subdomains A and B. A nucleotide binding site is located in the center of the domain in the cleft between lobes I and II. Hsp70 chaperones go through a functional cycle encompassing ATP-binding, ATP-hydrolysis, and ADP-Pi release, all of which takes place in the NBD39. ATP binding leads to a rearrangement of lobe I, which triggers the SBD to open the client binding site40–42. Following ATP hydrolysis, the NBD is in an ADP-bound state that leads to the closure of the client binding site. From there, ADP is released at one point, resulting in the apo form, to which a new ATP molecule binds to restart the cycle. The overall Hsp70 chaperone activity resulting from this fundamental cycle dependents on the cellular context and manifests into a diverse set of effective functions such as a foldase, holdase, translocase, unfoldase or disaggregase11,12,43. Importantly, these functions are fundamentally regulated by the timing originating from nucleotide processing in the NBD and understanding the Hsp70 functional cycle thus requires elucidating the nucleotide reaction steps and their modulation by environmental change.
Interestingly, despite exhaustive biochemical characterization of Hsp70 proteins, measurements of the functional cycle kinetic parameters have so far been possible only by isolating individual steps and not at atomic resolution44–46. In this study, we develop a method that combines the power of methyl NMR spectroscopy to resolve atomic siteswith a temporal dimension to resolve the reaction kinetics of the cycle individual steps. We benchmark the method on the example of the NBD of the human Hsp70 chaperone BiP from the endoplasmic reticulum (ER) and compare the results from the in cyclo experiments with classical single turn-over and ADP release experiments. We find that the functional cycle is completely independent from the bulk concentration of free inorganic phosphate (Pi) and characterize absence of any significant effect of the concentration of calcium. The method established here provides a technology platform for a fundamental understanding of the Hsp70 functional cycle at the atomic level, also in the context of the full-length Hsp70s and their regulation by co-chaperones.
NMR resonance assignment of NBD methyl groups
As a first step towards observing individual atomic sites during the functional cycle of BiP NBD, we established a highly pure and homogenous sample preparation. Since we want to be able to detect conformational sub-states of the protein with potentially low populations, the preparation needs to be free of any, specifically or non-specifically bound nucleotides and other contaminants. Such contaminants have been reported as a major cause of concern in the literature47. We thus purified the protein using well-established protocols and then added an affinity column purification step under denaturing conditions of 8 M Urea. The protein is entirely unfolded under these conditions and thus loses the affinity for bound impurities, which are washed through the column. After elution from the column, the protein was slowly refolded via dialysis. The refolded BiP NBD was analyzed by SEC-MALS showing a perfect overlap with BiP NBD purified without unfolding/refolding step (Supplementary Fig. 1a,b). Its NMR spectrum was 100% homogenous, evidencing the absence of any significant amounts of ligands. The proper conformation of the refolded protein was assessed by a comparison of the NMR spectra of the BiP NBD with and without denaturation (Supplementary Fig. 1c). The two spectra overlapped perfectly, demonstrating that the BiP NBD reaches its native state after denaturation and refolding. This high purification standard was kept in all subsequent experiments.
In a next step, we isotope-labelled the methyl groups of three amino acids, Ile-[13C1H]δ1, Met-[13C1H]ε and Val-[13C1H]g1/g2, on an otherwise deuterated background. This is a well-established technique to allow atomic resolution NMR studies even at large molecular sizes up to several 100 kDa48. 2D [13C,1H]-methyl-TROSY spectra49 with high sensitivity can be recorded in times as short as 5 minutes at protein concentrations of 100 μM. This high sensitivity is key to detect also minor sub-states of the cycle when longer experiment times are used. With the purification and preparation steps, NMR spectra of the protein in thermodynamic equilibrium were recorded in the apo form or in presence of ADP. The resulting equilibrium spectra serve as references for the respective conformational states of the protein. In presence of 5 mM ADP-Pi, we observed a homogeneous spectrum with a single set of 102 resonances, precisely matching the 102 resonances expected from the chemical structure of the molecule (Fig. 1a). We established sequence-specific assignments of these resonances using a strategy that combines single point-mutagenesis and NOESY experiments. As a first step, we established the assignments of all 6 methionine residues by single point mutagenesis (M148L, M153L, M196L, M263L, M332L, M339L). Then, using these anchor points, we expanded the assignment using a 3D 13C, 13C-resolved [1H, 1H]-NOESY experiments that we manually curated against a published crystal structure of the BiP NBD (PDB 5EVZ) (Fig. 1b and Supplementary Fig. 2a). The methyl groups of Met, Val and Ile have unambiguously separated chemical shift ranges for their NMR signals, permitting a direct identification of the amino acid type for a given signal and thus to easily distinguish different NOESY network. We resolved around 4 NOE contacts per residue, guaranteeing unambiguous assignments as a network effect. We additionally exploited the good correlation between the NOE cross peaks intensities and the calculated distance in the BiP NBD structure to confirm the correctness of these network (Supplementary Fig. 2b). Overall, we could resolve 100 NOESY contacts up to 5 Å, 255 NOESY contacts in the distance range 5–8 Å and 91 NOESY contacts in the range 8–10 Å, leading to a high-confidence assignment (Fig. 1c). As a final validation step, we selected 7 individual residues at the core of large NOESY networks and confirmed the correctness of their assignment by single-point mutagenesis. In total, this approach led to the stereospecific assignment of 60 residues, thereof 32/35 valines, 22/25 isoleucine, 6/6 methionine. Among the assigned 32 valines, 24 had both methyl Cg1 and Cg2 assigned and 8 one of them, resulting in a total of 84 observable NMR signals (Fig. 1a and d, Supplementary Fig. 3). For the subsequent experiments, we thus have 84 atomic level reporters that we can observe simultaneously and with high sensitivity for local structural changes.
Setup of the functional cycle with the ATP regeneration system
Based on these prerequisites of a highly pure preparation and near-complete resonance assignments, we set up the experiment to monitor the BiP NBD functional cycle. We selected a buffer composition corresponding to optimal Hsp70 activity (25 mM HEPES pH 7.5, 150 mM KCl and 10 mM MgCl2), and the physiological temperature of 37°C. Magnesium is required for the Hsp70 ATP hydrolysis as it coordinates the ATP b and g phosphate50. The potassium concentration was chosen to match previous reports that it acts as a cofactor of the Hsp70 hydrolysis of ATP increasing the activity by 5-fold in the optimal range of concentration between 100 and 150 mM51.
Notably, while it is possible to prepare pure apo and pure ADP-bound state, it is not possible to prepare a stable ATP-bound state, due to the catalytic activity of the protein. For example, the addition of 5 mM ATP to 100 μM of the BiP NBD leads to the significant accumulation of ADP already in the first few minutes and results in a non-equilibrium situation with rapidly changing ADP/ATP concentration ratio (Supplementary Fig. 4). In order to create a stable steady state condition, we implemented an ATP regeneration system inside the NMR tube that steadily converts ADP into ATP52. This system exploits the activity of pyruvate kinase, at catalytic amounts, to combine phosphoenol pyruvate (PEP) and ADP to form a novel ATP molecule (Fig. 2a). ATP and ADP molecules can be unambiguously distinguished by their H8 adenine proton in 1D 1H NMR experiments, PEP by its methylene protons and pyruvate by its methyl protons (ATP: 8.554 ppm, ADP: 8.558 ppm, PEP: 5.424 ppm, pyruvate: 2.424 ppm). Monitoring of ATP, ADP, PEP and pyruvate concentrations by 1D 1H NMR spectra shows that the ATP regeneration system keeps the concentrations of ATP and ADP effectively constant with no detectable signal for the ADP ([ATP] = 5 mM and [ADP] < 5 μM) (Fig. 2b,c), while the PEP concentration is linearly decreasing and the pyruvate concentration linearly increasing with the same rate (Fig. 2d,e). Thereby, the linear increase of pyruvate corresponds stoichiometrically to the ATP consumption and thus directly allows to calculate the ATP hydrolysis rate of the BiP NBD (Fig. 2e). To distinguish this experiment from equilibrium experiments, we refer to this setup with ATP regeneration as the “in cyclo” NMR experiment. The ATP consumption in our setup was 0.087 ± 0.003 mM min-1, in a sample with 100 μM BiP NBD (Fig. 2f), which corresponds to a molecular hydrolysis rate of khydr = 0.95 ± 0.03 min-1. Mechanistically, this number is the sum of the ATP hydrolysis rate kcat and the ADP release koff(ADP). The inverse of this rate, t = khydr-1 = 63 ± 2 s is the length of the functional cycle of BiP NBD.
Notably, the hydrolysis rate determined in the in cyclo experiment matches classical experiments. A typical assay is the NADH-coupled ATPase assay53. This assay exploits the enzymatic activity of lactate dehydrogenase to turn pyruvate into lactate by coupled NADH oxidation, which can be monitored by photospectrometry (A340) to measure the ATP consumption. We determined the rate with this assay and found perfect correspondence within error under the steady state conditions of our in cyclo NMR experiment (khydr = 0.92 ± 0.11 min-1) (Fig. 2f). Our assay thus faithfully reproduces established properties of the cycle, while simultaneously allowing atomic resolution observations. Our experimental setup shows an excellent match in terms of its consumption rate with conventional ATP hydrolysis assay. The crucial advantage is however that it now allows the simultaneous observation of conformational states at atomic resolution.
Direct observation of the ATP-bound state
In a next step, we wanted to exploit these properties to get atomic level insights. 2D [13C,1H]- methyl-TROSY spectra recorded in cyclo with active ATP regeneration show that the resonances split into two distinct sets of NMR signals. The relative amounts of the two signals is very similar for all residues and these signals thus correspond to two conformational states of the BiP NBD under the steady state conditions in cyclo. One of the two states matches perfectly with the NMR signals in 5 mM ADP-Pi in equilibrium experiment, unambiguously identifying this resonance set as the ADP-Pi-bound state (Supplementary Fig. 5a). The second set of NMR signals did neither match apo BiP NBD, ADP-Pi BiP NBD nor ADP BiP NBD (Fig. 3a,b and Supplementary Fig. 5b,c). It was detected only in presence of the ATP regeneration system or in the early phase of non-equilibrium experiments with pure ATP added (Supplementary Fig. 5b,c). This set of NMR signals therefore corresponds to the ATP-bound state. We term the BiP NBD ADP-Pi-bound state the D state and BiP NBD ATP-bound state T state.
Since we did not yet have sequence-specific resonance assignment for the ATP-bound state, we established them by direct transfer from the ADP-Pi-bound state for all signals in unambiguous spectral regions (Fig. 3a,b) and further established and confirmed assignment by the same 13 single point mutants that had previously been used for the assignment of the ADP-Pi-bound state. In total, 74 unambiguous assignments for the ATP-bound state were established (Fig. 3a and Supplementary Fig. 6).
These assignments now give us valuable information about the ATP-bound state. On the one hand, we note that for every single residue in the entire BiP NBD, we observe a distinct signal for the ADP-Pi-bound state and the ATP-bound state. This leads to the conclusion that the structural rearrangements of the BiP NBD to adapt between the ATP and ADP-Pi molecules involve the entire BiP NBD. Second, we can interpret the chemical shift differences between the two states in a structure-specific manner. The chemical shift differences were largest for residues located in the vicinity of the bound nucleotide, as expected (Fig. 3c and Supplementary Fig. 7a). Additional large chemical shift differences were however also observed in the lobe IA. These reflect the rotation of the lobe IA resulting from ATP binding, which in the full-length protein result in docking of the SBD12.
The availability of assignments of the ATP-bound state allowed us also to assess the effects of slow-hydrolysable ATP analogs AMPPNP (adenylyl imidodiphosphate), AMPPCP (adenylyl methylenediphosphate) and ATP-γ-S (adenosine 5’-(gamma-thiotriphosphate)). These analogs are frequently used to mimic the ATP-bound state54. The comparison with the 2D [13C,1H]-methyl-TROSY spectra fingerprints for the three commons ATP analogs shows large chemical shift deviations for the entire BiP NBD (Supplementary Fig. 7b). The shift differences for AMPPNP and AMPPCP shows that these two analogs shift the BiP NBD conformation in a state that is more similar to the ADP-Pi-bound state than ATP-bound state, i.e. their NMR fingerprints are closer to the one of the ADP-Pi than ATP-bound state (Supplementary Fig. 7b). The ATP-γ-S fingerprint spectrum features three NMR signals per residue, two major states that resemble the ADP-Pi-bound state, and one minor state that is closer to the ATP-bound state but does not overlap with it (Supplementary Fig. 7b). Therefore, ATP-γ-S leads to the formation of an heterogenous mix of conformations that do not represent the ATP-bound state. This fully explains why neither of these three analogs induces the expected Hsp70 interdomain conformational change that is triggered by ATP binding as it has been reported in the literature44,55,56. As a consequence, our in cyclo experiment is the only setup that enables studies of the native ATP-bound state at atomic resolution.
Combined measurement of the functional cycle kinetic parameters
With the assignments of the two functional states at hand, we could next determine the complete kinetic parameters of the functional cycle. The NMR signal intensities of each of the two states is proportional to their relative population and thus to the kinetic rate constants that connect the two states. We integrated 51 non-overlapping, independent methyl groups to quantify the population ratio pD/pT. Along the entire protein, this ratio showed very little variation, clearly establishing that BiP NBD completely splits into two independent states (Fig. 3d). The relative populations are T: 21 ± 4% and D: 79 ± 4% (Fig. 3d,e). Because the length of a complete cycle is 63 ± 2 s, these population levels correspond to mean-lifetimes tT = 13.4 ± 2 s for the ATP-bound T state and tD = 49.7 ± 3 s for the ADP-Pi-bound D state (Fig. 3f). The inverse of these life times thus correspond to the catalytic ATP hydrolysis rate kcat = tT-1 = 0.075 s-1 and the ADP release rate koff = tD-1 = 0.02 s-1 during the functional cycle of NBD.
The ATP lifetime in cyclo is identical to classical single turn-over experiments
Conventional approaches do not allow the determination of the ATP hydrolysis rate kcat from a steady-state experiment, but require separate single turn-over experiments. We wanted to benchmark the ATP lifetime obtained in cyclo by comparison with standard single-turn over experiments44,46. The single-turn over experiments were performed according to the standard protocol adapted from Theyssen et al.44. First, the NBD ATP complex was formed by incubation of an excess of ATP with BiP NBD at 4°C. At this temperature, it is generally assumed that the hydrolysis rate can be neglected. Next, the complex was separated from unbound nucleotide by gel filtration columns and then incubated at 37°C for variable time, until the reaction was stopped by addition of HCl. From the resulting samples, the [ATP]/[ADP] ratio was determined by anion exchange chromatography57. The data were fitted to a monoexponential, which corresponds to the ATP half-life time (Fig. 4a). The single turn-over experiments showed an excellent agreement with the measurements in cyclo (tT = 14.3 ± 2 s (single turn-over) vs tT = 13.4 ± 2 s (in cyclo)) (Fig 4b). The in cyclo experiment thus faithfully reports the ATP lifetime of the system in a steady state experiment, while simultaneously also giving atomic level structural information.
ADP-Pi release is the limiting step of the BiP NBD functional cycle
We next wanted to benchmark also the second kinetic rate obtained with the in cyclo experiment by classical experiments. The standard assay in the Hsp70 field to measure ADP lifetimes is the ADP displacement assay44. This assay relies on a fluorescently labelled ADP derivate, N8-(4-N'-methylan-thraniloylaminobutyl)-8-aminoadenosine 5'-diphosphate (MABA-ADP), that shows a substantial increase of fluorescence by 140% upon binding to nucleotide-free Hsp7044. With this reagent, nucleotide release is determined by real-time measurements of the fluorescence signal of Hsp70 with bound MABA-ADP in the presence of an excess of non-fluorescent ADP. The MABA-ADP unbinds over time, leading to a decrease in fluorescence intensity, because the non-fluorescent ADP prevents re-binding. Fitting the data with a mono-exponential gives the ADP half-life time. The MABA-ADP release times and the ADP release in the functional cycle should perfectly match and this is what is generally assumed in the literature when interpreting the MABA-ADP results. Direct measurements of ADP release during the functional cycle have so far not been accessible. Strikingly, our measurements show a large and highly significant deviation between the two experiments (tD = 13.6 ± 2 s (MABA-ADP) vs tD = 49.7 ± 3 s (in cyclo)) (Fig 5a,b).
To resolve this conundrum, we realized that the functional state encountered in the functional cycle is the ADP-Pi-bound state, while the MABA-ADP experiment is phosphate-free and its lifetime thus corresponds to the ADP-bound state. It is well established that there is a difference in lifetime between the two cases and the presence of Pi in the buffer increases the ADP affinity and lifetime58,59. We therefore measured the MABA-ADP release at Pi concentrations from 0.1 mM to 30 mM, leading to an increase of the ADP lifetime from tD = 13.6 ± 2 s in the absence of Pi to tD = 153.2 ± 5 s in the presence of 30 mM Pi (Fig. 5a,b). Thereby, the effect of Pi concentration corresponds to an IC50 of 5.9 ± 0.2 mM. To provide a structural rationale for this observation, we compared the ADP-bound state and the ADP-Pi-bound states in cyclo of BiP NBD using the NMR chemical shifts of our 84 atomic probes. We identify large chemical shift differences between the two states, which are localized in the nucleotide binding pocket and in the vicinity of the phosphate binding site (lobe IIA) (Supplementary Fig. 8a,b).
Because the ADP lifetime in equilibrium experiments is strongly correlated to the bulk phosphate concentration, we wanted to assess the presence of the same effect on the functional cycle. We thus performed the in cyclo experiment at different phosphate concentration in the range from 0.1 mM to 30 mM (Supplementary Fig. 8c,d). Strikingly, the kinetic parameters of the functional cycle, and in particular, the ADP lifetime were completely inert to the bulk Pi concentration (Fig. 5b). This finding leads us to the conclusion that the sole determinant of the long-lived ADP-Pi-bound state during the functional cycle is the phosphate generated upon ATP hydrolysis in the BiP NBD nucleotide binding pocket. This phosphate remains inside the nucleotide binding pocket, “gluing” ADP into the BiP NBD and leaving the pocket only concomitantly. Rebinding of phosphate molecules from the bulk plays no role for the functional cycle of the BiP NBD and the cycle is thus robust to fluctuations of the phosphate concentration. The in cyclo experiment thus overcomes the limitations of static experiments.
Ca2+ does not affect the functional cycle
Ca2+ is a key ER stress marker and plays a fundamental role in regulating the activity of multiple ER proteins60. While the role of magnesium ions for the catalytic activity of Hsp70 is well established50, the potential role played by calcium on the Hsp70 functional cycle is not yet well understood61. This question is of special interest for the BiP as the ER can show large variations in calcium, during protein folding homeostasis and stress, with concentrations ranging from 0.1 mM under homeostatic conditions up to 0.8 mM in stress conditions62,63. It has been proposed that the calcium concentration might decrease the ATP hydrolysis rate of BiP by 2-fold at physiological Ca2+ levels compared to no calcium50,64 and increase the ADP-bound lifetime in a concentration-dependent manner (4-folds at physiological concentration61). Published crystal structures of the nucleotide bound BiP NBD in presence of calcium show that it binds in the nucleotide binding site61, in which the ADP-bound state calcium contacts both phosphate groups (a and b) while magnesium only contact the b phosphate (Supplementary Fig. 9a). This might suggest a mechanism for the variation of the kinetic parameters, if Ca2+ replaces Mg2+ ions. Importantly, the Ca2+ concentration is always lower than the Mg2+ concentration also in the ER65,66. Accordingly, to test the effect of calcium on the kinetic parameters of the BiP NBD in the following experiments, we keep the magnesium concentration at 10 mM and vary the calcium concentration as indicated for each experiment.
First, we probed the presence of calcium binding to the ADP-bound state by NMR equilibrium experiments. Upon addition of 3.3 mM Ca2+ to ADP-bound BiP NBD in presence of 10 mM Mg2+ ([Mg2+]/[Ca2+]=3) large CSPs were observed, clearly confirming the Ca2+ binding (Supplementary Fig. 9b). Mapping of these CSPs on the structure showed changes consistent with a calcium binding site in lobes IA and IIA as expected from the published crystal structure of NBD bound to ADP-Ca2+ (PDB 6ZYH)61 (Supplementary Fig. 9c). Next, we used MABA-ADP release to measure the ADP mean life as a function of calcium concentration at a fixed concentration of magnesium. We observed a strong dependence of the ADP lifetime on the calcium concentration with an IC50 of 212+7 μM, in excellent agreement with the literature61 (Fig. 6a,b and Supplementary Fig. 10a). Since the BiP NBD nucleotide binding pocket is always occupied by an ADP and a phosphate moiety during the functional cycle, we wanted to explore the effect of calcium on the ADP-Pi-bound state and thus repeated the experiment in the presence of 0.5 mM Pi. Strikingly, this experiment showed that the lifetime of bound ADP-Pi is completely independent of the calcium concentration (Fig. 6a,b and Supplementary Fig. 10b).
The data thus readily suggest that the presence of calcium should also have no effect on the functional cycle. And indeed, in the presence of 3.3 mM calcium in the in cyclo setup, the ATP consumption is 0.090 ± 0.003 mM min-1, which corresponds to a molecular hydrolysis rate of khyd = 0.99 ± 0.03 min-1 (Fig. 5c). This is essentially equivalent to the molecular hydrolysis rate in the absence of calcium khyd = 0.95 ± 0.03 min-1. Also the relative populations of the two states, with T: 22 ± 4% and D: 78 ± 4% were the same as in the absence of calcium (T: 21 ± 4% and D: 79 ± 4%) (Fig. 5d,e and Supplementary Fig. 11a). Together, this determined the lifetimes of bound nucleotides to tT = 13.3 ± 3 s and tD = 47.4 ± 3 s, which are similar to the lifetimes in the absence of calcium (tT = 13.4 ± 2 s and and tD = 49.7 ± 3 s) (Fig. 5f-h). These data thus completely exclude any effect of calcium on the native functional cycle, in the cellular concentration range as well as up to several mM.
Notably, the in cyclo experiment now allows us also to use our structural probes to probe the effect of calcium binding during the functional cycle. The 2D NMR experiments showed no significant CSP, neither for the ATP nor the ADP-Pi-bound state, evidencing an absence in structural changes, in full agreement with the finding that the kinetics are unchanged. (Supplementary Fig. 11c). The data thus clearly establish that calcium ions do bind to the ADP-bound state of the BiP NBD, but not to the ADP-Pi-bound state, and consequently, since the ADP-bound state is not part of the functional cycle of the BiP NBD, physiological variations of calcium have no effect on the BiP NBD functional cycle.