Using NMR to identify binding regions for N and C-terminal Hsp90 inhibitors using Hsp90 domains

Allosteric inhibitors of Hsp90 have potential as anti-cancer agents without the side-effects that arise from targeting ATP-binding site in the N-terminal domain. This study gives NMR information on binding of allosteric inhibitor compounds to Hsp90.

Abstract

We present the first NMR study of the interaction between heat shock protein 90 (Hsp90) and amino (N)-terminal inhibitors 17-AAG, and AUY922, and carboxy (C)-terminal modulators SM253, and LB51. We show that the two ATP mimics, 17-AAG and AUY922, bind deeply within the ATP binding pocket of the N-terminal domain, consistent with the crystal structures. In contrast, SM253, a C-terminal Hsp90 modulator, binds to the linker region between the N and middle domains. We also show that C-terminal inhibitor LB51 binds to the C-terminus with a more significant spectroscopic change than previously reported using NMR binding studies of C-terminal inhibitors novobiocin and silybin. These data provide key insights into how the allosteric inhibitor SM253 controls the C-terminal co-chaperones and confirms the binding domain of LB51.

Introduction

Chaperones control the protein-folding machinery and are responsible for ensuring effective proteostasis within cells. Deregulation of the chaperone function produces many diseases and is believed to be a major contributor to cancer development in cells. Cancer cells are required to be highly efficient at protein folding in order to promote their rapid metabolism. Inhibiting the folding process inhibits malignancy progression. Building new small molecule drugs that regulate individual chaperones is challenging. Heat shock protein 90 (Hsp90) performs the final protein folding step on 70% of all proteins in a cell.1 It has 3 domains, the amino (N), middle (M), and carboxy (C) domains (Fig. 1). Approximately 15 families of Hsp90 inhibitor compounds have been assessed in ∼200 cancer clinical trials and all of these trials have failed when these drugs are used as single therapies (www.clinicaltrials.gov). One reason suggested for their failure is that these drugs are non-selective because they target the ATP binding site, which is common to many proteins. The non-specific binding to other proteins is thought to produce a cellular stress response (heat shock response, HSR) because it inhibits multiple unrelated cellular pathways.2 The HSR is designed to increase the concentration of heat shock proteins in order to promote protein folding, thereby protecting the cell. The HSR, therefore counters the function of the clinical Hsp90 inhibitors, producing drug resistance.2d

Fig. 1. Structure of Hsp90 (N = amino domain, M = middle domain, and C = carboxy domain) and Hsp90 inhibitors.

The recent development of chemical scaffolds designed to interact with the C-terminal domain of Hsp90 by several labs2c,3 have confirmed that targeting this region on Hsp90 is more effective than targeting the ATP binding site for regulating cancer cell growth.2–4 C-terminal modulators should be more efficacious than the ATP mimics for the following reasons: first, the timing, types of proteins being folded by Hsp90, and protection capacity, are dictated by co-chaperones that bind to Hsp90’s C-terminus. Thus, targeting the C-terminus is more efficient for killing cells, because it directly regulates these events. Second, C-terminal modulators do not activate the HSR,2 therefore, unlike the ATP mimics, there is no cell protection mechanism to overcome.

In this report we utilize two well established ATP mimics that bind to the N-terminal domain, 17-AAG and AUY922, and examine them using protein NMR. Although the binding sites for both compounds have been identified using crystal structures, there is no NMR information available.5 Chemical shift perturbation data are available on Hsp90’s N-terminal domain in the presence of ADP, AMP, geldanamycin and radicicol.6 We also report the binding interactions between Hsp90 and two C-terminal modulators: the SM,2 and LB inhibitors.3,7,8 Neither of these molecules induce the HSR, and both control protein binding events at the C-terminus of Hsp90. Both the SM and LB series have been analyzed for their impact on co-chaperones. These data suggest that SM molecules require both the N and M domains while allosterically controlling the C-terminus.2,4 The LB series may bind directly to the C-terminus.3,7,8 However, to date there has been no NMR binding evidence for these molecules.

Results and discussion

Herein we use protein NMR (¹H, ¹⁵N TROSY) to determine the Hsp90 binding sites of four molecules: 17-AAG, AUY922, SM253 and LB51 (Fig. 1).9 Isolated domains of Hsp90 [N-terminal (N), amino-middle (NM), and C-terminal (C)] were prepared with ¹⁵N labelling. The NM domain was also deuterated since this construct was large (63 kDa) for NMR experiments that examine binding events between protein and small molecules.

During this study we assessed the chemical shift perturbations (CSP) that occur when molecules bind to proteins, causing observable shifts in the protein's NMR spectra; the extent of CSP for the resonance of a given nucleus in the protein is indicative of a changed environment due to the binding of the added molecule.10 The resonances with the highest perturbations usually indicate the binding site of the small molecule on the protein.10b Our analysis of the interaction sites and binding affinity of the N-terminal inhibitors 17-AAG and AUY922 with ¹⁵N-labelled N domain utilized a set of previously-determined resonance assignments;9 spectra were acquired under identical conditions to those reported,9 allowing the published assignments to be mapped directly onto our spectra.

Increasing amounts of each drug were added to ¹⁵N-labeled N-domain, and a spectrum was acquired after each addition until a 1 : 1 ratio of protein to N-terminal inhibitor was achieved. At this ratio, saturation of the binding occurred, and no further chemical shift changes were observed with addition of further compound. A portion of the ¹H–¹⁵N TROSY- HSQC spectrum of Hsp90 N domain is shown in Fig. 2a, superimposed with the spectra of the N domain in the presence of 1 : 1 ratio of 17-AAG (red) and AUY922 (green). The full superimposed spectra are shown in Fig. S1 and S2 of the ESI.† Because the formation of the complex is in slow exchange on the NMR chemical shift time-scale (see following paragraph), K_ds could not be calculated for these drugs.

Fig. 2. a) Superposition of a portion of the ¹H–¹⁵N TROSY-HSQC spectra of the N domain of Hsp90A (black) and the N domain in the presence of a 1 : 1 ratio of 17-AAG (red) and of AUY922 (green). Full spectra showing the complete titrations are provided in Fig. S1 and S2.† b) Superposition of the same portion of the ¹H–¹⁵N TROSY-HSQC spectrum of the N domain of Hsp90A (black) with that of the N domain in the presence of a 1 : 1 ratio of SM253 (cyan). c) Chemical shift perturbation for the N domain of Hsp90 with 17-AAG (red) and AUY922 (green). The black horizontal line represents the mean Δδ_av value (0.074 for 17-AAG and 0.071 for AUY922); the orange line represents the mean + 1 standard deviation (0.19 for 17-AAG and 0.19 for AUY922); the red line represents the mean + 2 standard deviations (0.30 for 17-AAG and 0.31 for AUY922). d) Ribbon structure of the crystal structures of the N domain of Hsp90 with the 17-AAG analogue 17-DMAG (PDB ; 1OSF)11 and the AUY922 analog VER-49009 (PDB ; 2UWD).6,12 The backbone of each structure is colored red for residues with Δδ_av values above 0.30 and orange for values between 0.19 and 0.30. The compound structure is green.

The exchange between free Hsp90 and the bound states with either 17-AAG or AUY922 is slow. Upon addition of sub-stoichiometric amounts of drug, cross peaks corresponding to the drug-binding site disappear, while new peaks corresponding to the drug-bound state appear elsewhere in the spectrum (illustrated in Fig. S3, ESI†). A ‘minimum chemical shift method’11 was employed to assign the resonances in slow exchange; this method measures the distance between the original free peak and the nearest new peak. The chemical shift perturbation (or CSP) for each residue (Fig. 2c) is calculated using the formula .

Cross peaks with the most significant perturbation upon addition of 17-AAG (above red line, Fig. 2c) include Ser53, Gly97 Thr99, Ile104, Ile128, Gln133, Gly137, Phe138, Tyr139, Thr184 and Lys185. The backbones of these residues, together with residues with perturbations between 1 (orange) and 2(red) standard deviations above the mean, are plotted onto the crystal structure of the N domain of Hsp90 bound to 17-dimethylaminoethylamino-17-demethoxy geldanamycin (17-DMAG)6,12 in Fig. 2d. The perturbations we identified by NMR are located close to the drug binding site identified in the crystal structure (drug in green).12

A very similar set of residues is perturbed by the addition of AUY922, but the chemical shift changes caused by the two compounds were frequently in different directions (Fig. 2a). This is likely due to differences in local binding modes and chemical differences between the compounds. However, the overall magnitude and location of the most-perturbed resonances are similar (Fig. 2c). The residues most perturbed (>mean + 2 standard deviation) by AUY922 are Ser53, Asp54, Leu56, Gly95, Gly97, Ile104, Asp127, Met130, Ile131, Val136, Ser140, Ala141 and Lys185, which correspond well with the binding site shown in the crystal structure of the Hsp90 N-domain with the analogue VER-49009 (structurally similar to AUY922). Thus, in solution both AUY922 and 17-AAG interact with the N-domain in a manner similar to that reported in the crystal structures.

Pulldown experiments with the SM series and each domain or multidomain fragment of Hsp90 (N, M, NM, C, and MC), have suggested that this class of molecules bind to the multidomain fragment N–M domain.4a,4b,13 Using similar NMR titration experiments to those involving AUY922 and 17-AAG we added increasing amounts of SM253 to a solution of the N-domain until saturation of the solution was reached and SM253 became insoluble. The solubility limit for SM253 in water is ∼200 μM and at these concentrations there were no significant CSPs (Fig. 2b). This result is consistent with pulldown data showing that SM253 does not bind to the N-domain alone but requires the flexible linker between the N and M domains.4a,13

The binding of SM253 to a ²H/¹⁵N-labeled construct containing the N domain, the flexible, disordered linker region and the M domain (i.e. the NM construct), was assessed by addition of increasing amounts of the drug. The NM domain peaks have not been assigned as it is relatively large (63 kDa), however, it was possible to identify cross peaks that have been assigned in the N-domain spectrum. Two types of cross peaks are visible in the TROSY-HSQC spectrum of the NM domain (Fig. 3), weak, broad peaks belonging to the structured N and M domains, which likely interact to form a globular folded entity (Fig. 3a), and strong, sharp peaks belonging to the 60-residue disordered linker sequence between the domains (Fig. 3b), (see also Fig. S4, ESI†).

Fig. 3. Portions of a superposition of the ¹H–¹⁵N TROSY-HSQC spectra of 100 μM Hsp90 NM domain (black) and in the presence of 40 μM (red), 80 μM (green) and 120 μM (blue) SM253. a) Region plotted at lowest contour level 0.4575. Assignments inferred from those of the N domain are labelled b) region containing resonances of the disordered linker (plotted at lowest contour level 4.3); these resonances have not been assigned. Colors represent the same concentration ratios as a).

Similar to data produced when adding SM253 to the N domain alone (Fig. 2b), there was little to no chemical shift perturbation in the resonances of the structured and folded N and M domains (Fig. 3a, and Fig. S4†). In contrast, the resonances corresponding to the disordered linker region are impacted by the addition of SM253 (Fig. 3b, red, green, and blue shifts) and there is a clear concentration dependent shift of several cross peaks. These data are consistent with data from pulldowns, where both the N, linker, and M domains must be present for binding of SM253.4a,4b,13

The SM series is well documented for inhibiting binding between all 9 tetratricopeptide repeat (TPR) -containing co-chaperones and Hsp90’s C-terminus.4 Thus, we also examined whether SM253 may also bind to the C-terminus. Previous reports discussed the binding of two C-terminal Hsp90 inhibitors, novobiocin and silybin. However only when these compounds were used at very high concentrations (500 μM) did they perturb residues on the C-domain.14 Unfortunately, the NMR spectrum of the C-terminal domain has not been assigned, thus, the exact residues involved in the binding of these inhibitors were not identified.9,15

The C-domain of Hsp90 contains 180 residues and constitutes the dimerization domain of Hsp90. Thus, the isolated domain, if it is dimeric, would have a molecular weight of 40 kDa, which is quite large for assessing NMR peak shifts. The ¹H–¹⁵N HSQC spectrum of the ¹⁵N-labeled C domain is superimposed with the spectrum that contains the C domain and SM253 in a 1 : 1 ratio (Fig. S5, ESI†).

The spectrum contains less than half of the cross peaks expected for a 180-residue domain, and the cross peaks that appear are relatively strong, concentrated in the center of the spectrum. Such behavior, where the peaks are strong, and concentrated in the center, is characteristic of unfolded or disordered regions of globular proteins. Indeed, these peaks of the C-domain display behavior similar to that of the linker region in the N–M domain construct described earlier. These strong cross peaks indicate that either the C-terminal domain is incompletely folded (perhaps under the experimental conditions), or these regions of the domain are normally disordered, similar to the linker region. The superposition (Fig. S5, ESI†) shows that SM253 has little effect on the observable C-domain resonances. These data are consistent with the pulldown data, where SM253 acts via an allosteric mechanism on Hsp90, impacting the C-terminus, but not binding directly to this region.

Several elegant studies have shown that Hsp90 activity is modulated by the linker region between the NM domains, where the linker allows the necessary flexibility for the N-domain to rotate, thereby inducing a long-range conformational change, and exposing the C-domain.16 Thus, the linker is critical for controlling the timing and binding events of both clients and co-chaperones at the C-domain.16 Our NMR binding data indicate that SM253 impacts the binding of co-chaperones at the C-terminus via conformational changes caused by binding to the linker region between the NM domain. SM253 likely stops the linker from allowing rotation, thereby inhibiting appropriate presentation of the C-terminal domain to co-chaperones, acting via an allosteric mechanism.2,4

The LB inhibitors were designed to bind to Hsp90’s C-terminal domain3c and inhibit C-terminal binding co-chaperones.7,8 The cyclized peptide LB51 exclusively bound Hsp90 in cell lysate (assessed via pulldown assays),7 and was highly effective at inhibiting the interaction between the C domain and co-chaperones Cyp40, FKBP38, FKBP51, and HOP.3,7,8 Thus, we examined whether LB51 bound to the C-domain via NMR. Addition of a 1 : 1.6 molar ratio of C domain protein: LB51 (100 μM: 160 μM) caused splitting, perturbation and attenuation of multiple peaks within the C-domain spectrum (Fig. 4). The ratio and concentration used of LB51 is much lower than the 1 : 2.5 ratio of protein: compound (200 μM: 500 μM) required to cause perturbation when using novobiocin or silybin.

Fig. 4. Superposition of the ¹H–¹⁵N HSQC spectra of 100 μM Hsp90 C domain (black) and in the presence of 160 μM LB51 (red). Portions of each spectrum are enlarged for clarity.

The peaks for the C-domain are not assigned, however, the alterations observed upon the addition of LB51 are greater than those observed when novobiocin or silybin are added to the C-domain.15 Mapping the fast exchange CSPs caused by increasing amounts of LB51 provided the binding affinity to the C-domain, which has a K_d of 19.8 ± 3.5 μM (K_d = ∼45 μM in biochemical binding assays). This K_d is significantly lower than the values for C-terminal Hsp90 inhibitors novobiocin and silybin (K_d values of ∼500 μM).14

Experimental

Materials and methods

Protein expression and purification

Hsp90 domain expression

The sequence for the constructs of the Hsp90 domains used here are: N-terminal domain (1–226), the NM-domain construct (1–554), and the C-domain (553–732). The N 215 and C domain were generous gifts from Dr. S. Snyder and the NM domain was prepared as described previously.9 All constructs were tagged with a 6× histidine tag at the N-terminus. The constructs were transformed into the E. coli host BL21(DE3) (DNAY) cells by combining 80 μL of bacteria with 1 μL of plasmid and cooling the mixture on ice for 20 minutes. The mixture was then subjected to heat shock at 42 °C for 45 seconds followed by 2 minutes on ice. Then 800 μL of luria broth (LB) was added to the mixture and it was allowed to shake at 37 °C for 45 minutes.

After this incubation an LB/agar plate treated with carbenicillin and kanamycin was inoculated with 50 μL of the LB plamid mixture and incubated at 37 °C overnight. One culture was selected and grown up in M9 minimal media. Proteins were expressed in M9 minimal medium in the E. coli host BL21(DE3) (DNAY) with induction initiated with 1 mM IPTG at 15 °C for 12–20 h. Cells were lysed by sonication in 25 mM Tris buffer (pH 8.0) supplemented with 300 mM NaCl and a tablet of protease inhibitor cocktail (Roche). The soluble fraction of the cell lysate was applied to 12 mL of His-60 nickel resin, washed with a 300 mM NaCl, 10 mM imidazole solution and then eluted with 300 mM imidazole. Fractions containing the target protein were determined with 4–20% tris–glycine gels electrophoresis and then those fractions were loaded onto a 5 ml HiTrap-Q FF column equilibrated with 25 mM tris (pH 6.0), 50 mM NaCl and proteins eluted with a linear gradient to 1 M NaCl. Clean fractions were then dialyzed into NMR buffer (25 mM sodium phosphate pH 7.2, 100 mM NaCl, 1 mM EDTA, and 4 mM DTT) and concentrated using Centriprep10 (Amicon). Expression of ¹⁵N-labeled proteins was carried out in M9 minimal medium supplemented with ¹⁵NH₄SO₄ (1 g l^–1). Expression of ¹⁵N, ²H-labeled proteins was carried out in M9 minimal medium in D₂O supplemented with ¹⁵NH₄SO₄ (1 g l^–1).

NMR spectra acquisition

¹H–¹⁵N transverse relaxation-optimized (TROSY) NMR spectra were acquired with 100 μM protein samples in NMR buffer supplemented with 5% D2O at 298 K on an Avance 900 MHz spectrometer equipped with a cryoprobe. Titration experiments with the Hsp90 inhibitors were done with the following ratios of protein to compound:

17-AAG: 1 : 0, 1 : 0.4, 1 : 0.6, 1 : 1

AUY922: 1 : 0, 1 : 0.1, 1 : 0.2, 1 : 0.4, 1 : 0.6, 1 : 1, 1 : 1.6

SM253 1 : 0, 1 : 0.2, 1 : 0.4, 1 : 0.6, 1 : 0.8, 1 : 1, 1 : 2

LB51 1 : 0, 1 : 0.2, 1 : 0.4, 1 : 0.8, 1 : 1.2, 1 : 1.6

17-AAG, AUY922, and SM253 were dissolved in DMSO for the experiments and thus, their spectra were compared to the DMSO control, where DMSO percentage was equivalent to the titration end point DMSO percentage. LB51 was dissolved in water and thus, the spectra for this compound was compared to the protein alone as a control. Assignments to the cross peaks were made by mapping the data from Park et al.9 onto the spectra here, using NMRviewJ.

Conclusions

In conclusion, we report NMR binding studies between 17-AAG and AUY922 with the Hsp90 N-domain. The NMR spectra show that the N domain residues that were significantly impacted upon compound binding were the same residues observed to bind in the crystal structure. We also show NMR data that strongly indicate that SM253 binds between the N and M domains, specifically impacting residues in the disordered linker region, which connects these two domains. These data support the mechanism proposed for SM253, where this SM series regulates the C-terminal domain interactions by allosterically inhibiting the long-range movement that exposes the binding site to co-chaperones (Fig. 5). These NMR data also support that inhibiting movement between the N–M domains, SM253 acts by blocking the binding between Hsp90 and HOP4b,4c thereby inhibiting the protein folding function of Hsp90 (Fig. 5).

Fig. 5. Model system showing how SM253 and LB51 function in the protein folding model.

We also demonstrated via NMR that the C-terminal inhibitor, LB51, impacts residues in the C-domain, directly inhibiting C-terminal binding-co-chaperones. Importantly, the K_d for LB51 with the C-domain is significantly lower than the previously reported C-terminal binding Hsp90 inhibitors, and it matches the K_d identified using biochemical binding assays. Modulating the C-terminus is likely the most effective strategy in developing future Hsp90 inhibitors because it is the C-terminal domain that controls the protein's timing and protein folding function. Directing the domain that controls most of Hsp90’s functions will likely alleviate side effects and produce a more efficient inhibitor than the current clinical molecules.

Conflicts of interest

The authors declare that there are no conflicts.