|Home | About | Journals | Submit | Contact Us | Français|
The molecular chaperone DnaK is essential for the survival of bacterial pathogens in the hostile environment of the host. Hence, it is in principle a promising target for drug design but for which no current inhibitors are available apart from certain antimicrobial peptides. To this end, we have screened libraries of small molecules for their ability to interact with the substrate-binding domain of DnaK. The most promising hit from the screen was synthesized and along with its analogs subjected to further assays to determine their binding affinity and ability to interfere with bacterial growth. This work resulted in the identification of a number of compounds that bind with submicromolar affinity and capable of inhibiting Yersinia pseudotuberculosis growth more effectively than the previously characterized peptides.
Chaperones provide crucial cellular functions across all domains of life (1). Hsp70 homologs are involved in such essential processes as protein folding, translocation, degradation, and even gene expression (2–7). The multiple homologs and the wide array of co-chaperones for Hsp70 underscore their importance and functional diversity (1,8,9). Hsp70s, including DnaK, the Escherichia coli Hsp70 (10–14) investigated here, have been extensively studied and have greatly increased our understanding of allostery. DnaK is composed of two structural domains (Figure 1A and B), an N-terminal nucleotide-binding ATPase domain (NBD) and a C-terminal substrate-binding domain (SBD), which can be further subdivided into an all β-strand region that binds substrate (β-domain) and an α-helical lid region that can close over bound substrate (11,12,15). DnaK cycles between two main states: an ATP-bound state with low affinity for substrate that is characterized by a tight interaction of the two domains, and an ADP-bound state that has high substrate affinity with less interaction between the domains (16–18). Structural and biochemical data have shown that residues from both domains (including loop L2,3 of the β-domain), as well as from the linker between them are required for communicating the nucleotide or substrate occupancy to the other respective domain (10,12,14,19,20).
Chaperones are essential for survival under myriad stress conditions. Of particular interest for drug discovery are the essential roles of Hsp70 in pathogenic microorganisms growing intracellularly in a host (21,22), and in cancer cells proliferating under the stressful environment of a tumor (23). Recently, there has been much progress in targeting the Hsp90 class of chaperones with potential anti-cancer compounds (24,25). With the experience of protein kinase inhibitors in mind (26), screens for Hsp70 inhibitors have focused on compounds targeting the ATP-binding site. While this approach has been successful for Hsp90 (27), the binding of ATP to Hsp70 is drastically different from that of kinases (28). In protein kinases, most inhibitors take advantage of a hydrophobic pocket and a set of highly conserved hydrogen bond interactions available at the site. In contrast, the ATP-binding site of Hsp70 is rather hydrophilic, and most interaction energy between the nucleotide (ATP or ADP) and the protein is derived from the phosphate groups (29,30). As a consequence, high throughput screening approaches targeting the ATP-binding pocket of Hsp70s are unlikely to produce viable hits, although a recent structure-based design study was successful in identifying submicromolar affinity binders (31).
We have taken the novel approach of directly targeting the SBD of DnaK. Previous work on a class of antimicrobial peptides from insects that target DnaK via the SBD helps to validate our approach (32–35). These SBD-targeting peptides show significant efficacies against a series of clinical pathogens, including fluoroquinolone-resistant strains (35). Furthermore, at least one of the peptides did not bind to human Hsp70 or alter its function (32), suggesting that effective and safe antimicrobial drugs could arise from targeting the SBD of DnaK.
Using a construct of DnaK corresponding to the β-domain only (residues 393–507) for which solution structures with and without bound peptide substrate are known (36,37) and an NMR-based screen, we have identified a scaffold compound capable of binding the β-domain. Our results led to the identification of a compound series that occupies a binding pocket near a proposed allosteric site of the SBD, loop L2,3 (Figure 1A and B). We report structure–activity relationship (SAR) studies, in silico docking, isothermal titration calorimetry (ITC), and antimicrobial activity of the most promising compounds of the series. Given the dire need for new classes of antibiotics, we hope that these compounds can eventually stimulate further research in this area may lead to clinically useful drugs.
The gene coding for the E. coli DnaK substrate-binding (β) domain (393–507) was amplified by PCR and subcloned into pET21a using the NdeI and BamHI cloning sites. The resulting protein contains 17 extra amino acid residues (MGSSHHHHHHGLVPRGS) at the N-terminus. The protein was expressed in the E. coli strain BL21(DE3) pLysS and purified using Ni2 + affinity chromatography. Uniformly, 15N-labeled DnaK was produced by growing the bacteria in M9 minimal media containing 15NH4Cl as the sole nitrogen source. The NMR samples were dissolved in 20 mM sodium phosphate buffer (pH 7.5) containing 90%/10% (H2O/D2O) or 99.5% D2O. DnaK, human Hsp70, DnaJ, and GrpE were purchased from Stressgen (Ann Arbor, MI, USA).
Apidaecin1a (GKPRPYSPRPTSHPRPIRV), pyrrhocoricin (VDKGSYLPRP-TPPRPIYNRN), drosocin (GNNRPVYIPGPRPPHPRI), and the model substrate NRLLLTG were synthesized by the Medical College of Wisconsin (Milwaukee, WI, USA).
Spectra were acquired on a 600-MHz Bruker Avance spectrometer (Bruker BioSpin Corp., Billerica, MA, USA) equipped with a TCI cryoprobe. Ligand binding was monitored by comparing the aliphatic region of 1D 13C-filtered 1H NMR spectra of 20 μM protein solutions (containing 90%/10% H2O/D2O or 99.5% D2O buffered with 20 mM sodium phosphate, pH 7.5; T = 300 K) in the presence or absence of 80 μM mixtures of 10 compounds, and then individual compounds from mixtures that caused significant perturbations in the spectra were identified.
The binding affinity of each ligand was initially estimated with a single titration point according to the following equation:
where To and Lo are total concentration of target protein and ligand, respectively. The parameter p represents the fractional population of bound versus free species at equilibrium, which for fast exchanging ligands is measured as:
where δobs is the observed protein chemical shift during the titration, and δfree and δsat are the chemical shifts for the protein in the fully unbound and fully bound (saturated) states, respectively.
Titrations were performed using a VP-ITC instrument from MicroCal (Northampton, MA, USA). Full-length DnaK or β-domain were used at 10–100 μM in 20 mM sodium phosphate buffer (pH 7.4) and 5% dimethyl sulphoxide (DMSO). Compounds were used at 15-fold excess in the same buffer. Data were analyzed using MicroCal Origin software provided by the ITC manufacturer (MicroCal, Northampton, MA, USA).
Escherichia coli cultures were seeded in 96-well plates by mixing 90 μL aliquots of an OD600 = 0.001 culture with 10 μL of compound solutions at various concentrations and grown overnight. Optical density (OD) at 530 nm was measured using a Perkin Elmer Victor 2 Multilabel plate reader (PerkinElmer, Shelton, CT, USA). Testing was carried out in 50% Mueller-Hinton II broth (BBL) in H2O. Wild-type Yersinia pseudotuberculosis YP126 (38) was grown overnight at 25 °C and approximately 106 bacteria were inoculated into each 1 mL of broth containing serial dilutions of the compound to be tested. After overnight incubation at 40 °C, the OD was read at 600 nm and compared with the growth without inhibitor. The minimal inhibitory concentration (MIC) was read as the concentration producing greater than fourfold reduction in final OD600.
To a stirred solution of free amine (1.0 equiv.) and Et3N (2.0 equiv.) in 5 mL dichloromethane (DCM) was added a solution of thiophene-2-carbonyl chloride (1.1 equiv.) in 5 mL DCM at −30 °C (Scheme 1). The resulting solution was stirred for 1 h and then allowed to warm to room temperature. After removal of the solvent, the residue was purified by flash column chromatography in hexane-ethyl acetate or DCM-methanol to provide the correspondent product (yield 75–95%). Synthesis scale was 100 mg, for which about 0.5 mmol of starting material was required.
N-(naphthalen-1-ylmethyl)thiophene-2-carboxamide 1H NMR (CDCl3, 600 MHz): δ 8.17 (s, 1H), 7.87 (s, 1H), 7.73 (s, 1H), 7.46 (m, 6H), 6.99 (s, 1H), 6.39 (s, 1H), 5.01 (s, 2H); HRESI-TOF-MS: calcd for C16H13NOS 268.0791 [M + H]+, found 268.0796.
N-(1H-indol-5-yl)thiophene-2-carboxamide 1H NMR (CDCl3, 600 MHz): δ 8.19 (s, 1H), 7.92 (s, 1H), 7.71 (s, 1H), 7.62 (s, 1H), 7.52 (s, 1H), 7.33 (m, 2H), 7.23 (s, 1H), 7.12 (s, 1H), 6.54 (s, 1H); HRESI-TOF-MS: calcd for C13H10N2OS 243.0587 [M + H]+, found 243.0592.
N-(quinazolin-2-yl)thiophene-2-carboxamide 1H NMR (CDCl3, 600 MHz): δ 9.35 (s, 1H), 7.90 (m, 3H), 7.61 (m, 4H), 6.89 (s, 1H); HRESI-TOF-MS: calcd for C13H9N3OS 256.0539 [M + H]+, found 256.0545.
N-(benzo[b]thiophen-7-ylmethyl)thiophene-2-carboxamide 1H NMR (CDCl3, 600 MHz): δ 8.35 (s, 1H), 7.82 (m, 5H), 7.47 (s, 1H), 6.97 (s, 1H), 5.30 (s, 2H); HRESI-TOF-MS: calcd for C14H11NOS2 274.0355 [M + H]+, found 274.0361.
N-(3,4-dichlorobenzyl)thiophene-2-carboxamide 1H NMR (DMSO-d6, 600 MHz): δ 9.11 (s, 1H), 7.79 (m, 2H), 7.60 (d, J = 7.2 Hz, 1H), 7.56 (s, 1H), 7.30 (d, J = 7.2 Hz, 1H), 7.16 (s, 1H), 4.43 (s, 2H); HRESI-TOF-MS: calcd for C12H9Cl2NOS 285.9855 [M + H]+, found 285.9861.
N-(2,6-dichlorobenzyl)thiophene-2-carboxamide 1H NMR (DMSO-d6, 600 MHz): δ 9.11 (s, 1H), 7.79 (m, 2H), 7.51 (s, 1H), 7.35 (m, 2H), 7.17 (s, 1H), 4.44 (s, 2H); HRESI-TOF-MS: calcd for C12H9Cl2NOS 285.9855 [M + H]+, found 285.9860.
N-(2-hydroxybenzyl)thiophene-2-carboxamide 1H NMR (DMSO-d6, 600 MHz): δ 8.96 (s, 1H), 8.09 (s, 1H), 8.03 (s, 1H), 7.78 (s, 1H), 7.39 (m, 2H), 7.28 (m, 2H), 7.14 (s, 1H), 4.42 (s, 2H); HRESI-TOF-MS: calcd for C12H11NO2S 232.0438 [M − H]−, found 232.0440.
N-(3,5-dichlorobenzyl)thiophene-2-carboxamide 1H NMR (DMSO-d6, 600 MHz): δ 8.67 (s, 1H), 7.79 (s, 1H), 7.75 (s, 1H), 7.51 (m, 2H), 7.38 (s, 1H), 7.11 (s, 1H), 4.66 (s, 2H); HRESI-TOF-MS: calcd for C12H9Cl2NOS 285.9855 [M + H]+, found 285.9862.
(4-(2-hydroxyethyl)piperazin-1-yl)(thiophen-2-yl)methanone 1H NMR (DMSO-d6, 600 MHz): δ 7.75 (s, 1H), 7.38 (s, 1H), 7.11 (s, 1H), 3.61 (m, 4H), 3.51 (m, 2H), 2.42 (m, 6H); HRESI-TOF-MS: calcd for C11H16N2O2S 241.1005 [M + H]+, found 241.1012. Commercially available BI-88B12 analogs were purchased from Chembridge Corp., (San Diego, CA, USA). Structures for purchased and synthesized compounds are shown in Table 1.
We have assembled a scaffold library composed of ~4000 compounds that were selected based upon their anticipated use as building blocks or scaffold components of further optimized molecules. The library has been acquired from three different sources (39), and chemical structures of the compounds have been deposited into PubChem (http://pubchem.ncbi.nlm.nih.gov/). In addition to this small molecule selection, we have also included a collection of 602 natural products (MicroSource, Gaylordsville, CT, USA) that was also screened by NMR.
To screen such a relatively high number of compounds with minimal protein and spectrometer time, we used simple 1D 1H NMR experiments of the protein measured in the presence and absence of mixtures of potential ligands, observing the aliphatic region of the spectra (δ < 1 ppm). This experiment requires relatively low protein concentrations (10–20 μM) and can be conducted in a relatively short time (30 min to 1 h by using a spectrometer operating at 600 MHz 1H frequency and equipped with a cryoprobe). Briefly, chemical shifts were monitored in the presence and absence of mixtures of 10 compounds at a fourfold excess to the protein. Mixtures causing spectral perturbation above our threshold of 0.08 ppm were deconvoluted. The compound BI-88B12 emerged as a hit from the screen (Figure 1C). Selected commercially available BI-88B12 analogs were purchased and then qualitatively assessed for their ability to bind the β-domain of DnaK by the magnitude of the chemical shifts induced upon complexation on the 13Cδ,1Hδ resonances of Leu484 (1:10 protein to ligand ratio, at 70 μM protein concentration) by 2D [13C, 1H] correlation spectra. Hit compounds with estimated Kd values of 200 μM or less were selected for further SAR studies including Kd determination by ITC. These data were further analyzed to design additional analogs for chemical synthesis (‘Materials and Methods’).
Based upon the NMR experiments and assays described below on the BI-88B12 analogs three compounds of interest emerged. BI-88B12 was the initial hit and causes the greatest NMR chemical shift change upon binding and was thus used as a comparison for all other compounds. BI-88D7 (synthesis described in ‘Materials and Methods’) was the strongest binder as determined by ITC. The commercially available BI-88E3 was the most potent inhibitor of microbial growth and thus the bulk of the discussion here will focus on BI-88B12, BI-88D7, and BI-88E3. Findings for all other compounds to come through the NMR screen with an estimated Kd < 200 μM are summarized in Table 1.
After the initial identification of BI-88B12, chemical shift mapping studies with 13C/15N-labeled protein were performed. Based upon these experiments, we discovered that BI-88B12 and its analogs do not bind to the substrate-binding pocket but rather to a pocket located around loop L2,3 on the opposite side of the SBD (Figures 1A, B and and2;2; Figure S1). Molecular docking studies then revealed the presence of a deep pocket centered between residues Leu484 and Pro419, which likely constitutes the binding site of these compounds (Figure 2). Residues in this region were previously reported to be essential for the allosteric communication between the substrate binding and the ATPase domains (10,12,14,19,20). Gold docking software (40) was used to predict the binding poses of compounds BI-88B12 and BI-88D7 near residues Leu484 and Pro419 (pdb: 1DKX) (41).
Most interestingly, residues in this pocket are among the least conserved between bacterial and mammalian Hsp70s (Figure 1D), furthering the potential of our compounds as non-toxic antibiotics. Figure S2 further shows that that the conservation of this site among bacterial DnaKs reaches across many other important bacterial pathogens, both Gram-negative and Gram-positive while differing significantly from mammalian sequences in this region.
Molecular modeling studies suggest that, despite the shared thiophene, the compounds might adopt two different positions. Both maintain a hydrogen bond with Gly482 but while BI-88B12 places its thiophene ring outside the pocket, this moiety is inserted into the pocket by BI-88D7. For BI-88B12, the fluorobenzene group is inserted deep into the hydrophobic pocket as evidenced by the strongest perturbation of residues there. With its thiophene group in the pocket, the indole ring of BI-88D7 is also able to interact with Asp481, located further out near the edge of the pocket. BI-88D7 causes equal perturbation of both Leu484 and Asp481, possibly because of its larger size that allows it to bridge both residues. These results may explain the much lower Kd of BI-88D7 as determined by ITC.
Isothermal titration calorimetry was performed on the most promising binders from the first set of BI-88B12 analogs (Figure 3C). These data were used in the design of a number of compounds which were not commercially available, including BI-88D7. As seen in Figure 3B and Table 1, the designed compounds exhibit increased binding affinity for DnaK by an order of magnitude compared with BI-88B12. The Kds determined by ITC for BI-88B12 and the earliest derivatives tested were in the range of ~10–70 μM while many of the optimized compounds exhibited single digit to submicromolar range dissociation constants.
Using ITC, we also investigated how binding of a substrate mimic peptide or ATP affected the binding of a compound and vice versa. This thermodynamic information may help explain how the compounds bind and how they may be affecting allosteric communication between the domains. In the context of the full-length protein, which includes the NBD, SBD, and the α-helical lid domains as shown in Figure 1, the substrate mimic peptide was found to bind stronger by an order of magnitude if the protein was presaturated with compound (Table 2) with the increased affinity mostly entropically driven.
This is possibly because of bound compound causing hydrophobic surface area in the substrate-binding pocket to become exposed. Furthermore, the SBD is known to be highly dynamic (36), and if compound binding is able to reduce the conformational exchange of the SBD, the peptide should then be able to bind without the normal entropic penalty of binding to a highly dynamic apo-SBD. The priming of the SBD by these compounds may be an example of the model of Swain et al. (12), in which prevention of docking between the two domains by the interdomain linker, specifically residues 389–392 (VLLL), would leave the β-domain in a state of high substrate affinity.
All compounds were first tested for their ability to limit cell growth of E. coli. Compounds were tested initially in mixtures of five over a range of concentrations. Mixtures that restricted growth were de-convoluted to determine efficacies of individual compounds (Figure 4A). Compound BI-88E3 demonstrated the greatest ability to inhibit growth, followed closely by BI-88D7 (Table 3). Both compounds were found to have similar MICs to apidaecin1a, the most effective peptide tested in this study (Table 3). The substrate mimic peptide NRLLLTG was not inhibitory in this assay upto a concentration of 100 μM. We also found no inhibitory effect under our experimental conditions from pyrrhocoricin, a result at odds with previous results (42). This difference may be possible because of the fact that we are using a different strain of E. coli as the peptide displayed tight binding by ITC.
As DnaK is required for growth under heat shock conditions, we tested the compounds’ ability to inhibit growth of E. coli at 42 °C. Only apidaecin1a showed any increased efficacy at the higher temperature, and its MIC value was about twofold better (80 μM at 37 °C to 40 μM at 42 °C). To determine whether these compounds have potential against an intracellular pathogen, we conducted preliminary sensitivity testing using Y. pseudotuberculosis (Figure 4B; Table 3). Compound BI-88E3 demonstrated a greater than 20-fold decrease in the MIC when tested over a temperature range of 25–40 °C. These results were consistent with inhibition of DnaK, as loss of DnaK function in bacteria is known to produce a temperature-dependent effect on growth.
By attempting to target the substrate-binding β-domain of DnaK, we have discovered compounds that bind not in the substrate-binding pocket but rather in a cavity involved in allosteric communication between protein domains. This series of compounds will help us gain further insight into the allosteric mechanism of DnaK by determining more precisely how compound binding affects substrate peptide and possibly ATP binding to DnaK and inhibit cell growth. Our preliminary data (not shown) indicate that many of the compounds inhibit both the ATPase and protein refolding activities of DnaK but more work is necessary. More importantly, we have identified a compound, BI-88E3, which is capable of inhibiting growth of E. coli and is even more potent than apidaecin1a against Y. pseudotuberculosis. Further work is needed to fully characterize the mode of action of the compounds in the presence of co-chaperones and unfolded proteins rather than a model peptide. However, we believe that the discovered compounds with submicromolar affinity for DnaK have the potential to be both biochemical tools for further understanding DnaK function and lead compounds for developing sorely needed antibacterial drugs with a novel target.
Figure S1. NMR-based characterization of BI-88D7. (A) Overlay of 2D 1H–15N HSQC of DnaK with (blue) and without (red) BI-88D7. (B) 1H–13C HSQC titration of DnaK with increasing amounts of BI-88D7. The peak corresponding to L484 shows the greatest movement while the peak for L441 moves very little over the course of the titration. Compound: protein molar ratios: 3.6:1 (green), 7.3:1 (red), 14.3:1 (blue), and 28.6:1 (yellow). Protein concentration is 70 μM for all points except the 28.6:1 where protein was 35 μM.
Figure S2. Alignment of multiple bacteria, bovine, and human Hsp70 sequences from Ile501–Leu507 based on numbering from Escherichia coli.
This work was supported by NIH Grants HG003916, AI055789, AI059572.
Please note: Wiley-Blackwell is not responsible for the content or functionality of any supporting materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.