Discovery of potent heat shock protein 90 (Hsp90) inhibitors: structure-based virtual screening, molecular dynamics simulation, and biological evaluation

Abstract

Heat shock protein 90 (Hsp90) is considered an attractive therapeutic target for cancer treatment due to its high expression in many cancers. In this study, four potent Hsp90 inhibitors (HPs 1–4) were identified using structure-based virtual screening. Among them, HP-4 exhibited the most potent inhibitory effects (IC₅₀ = 17.64 ± 1.45 nM) against the Hsp90 protein, which was about 7.7 times stronger than that of MPC-3100 (a positive inhibitor targeting Hsp90). In vitro cytotoxicity assay suggested that HP-4 could effectively inhibit the proliferation of a series of tumour cells, including HCT-116, HeLa, A549, A2780, DU145, HepG2 and A498. Furthermore, in vivo assay displayed that HP-4 had significant anti-tumour effects on HCT-116 cell-derived xenograft models. These data demonstrate that HP-4 could be a potential lead compound for the further investigation of anti-tumour drugs.

Keywords:

• Hsp90
• inhibitor
• virtual screening
• colon cancer
• biological evaluation

Introduction

Heat shock proteins (Hsps) are highly conserved in almost all organisms¹. As ATP-dependent molecular chaperones, Hsps are mainly responsible for the proper folding of client proteins, thereby regulating cellular homeostasis². According to gene homology and molecular weight, Hsps have been classified into small Hsps, Hsp40/60, Hsp70, Hsp90, and Hsp110³. Hsp90 is one of the most abundant Hsps, which accounts for 1–2% of the total cellular protein and increases to 3–5% after exposure to external stress^4,5. Hsp90 maintains the stability, conformation and function of many client proteins including EGFR, MET, HER2, AKT, CDK4 and cyclin D^6–9. Notably, most of these client proteins are overexpressed in malignant tumours and involve in the proliferation, angiogenesis, invasion and metastasis of tumour cells¹⁰. Moreover, the protein levels of Hsp90 in tumour cells are 2–10 times higher than that of their normal counterparts, suggesting its importance in tumour cell growth^11,12. Therefore, Hsp90 is considered a promising target for cancer therapy.

Hsp90 mainly exists as dimers in cells, and each monomer has three main domains including an N-terminal domain (NTD), a middle domain and a C-terminal domain (CTD)^13,14. The NTD contains an ATP-binding pocket, which is essential for the folding of client proteins by binding ATP¹⁵. The middle domain is crucial in client binding and the CTD is responsible for dimerisation¹⁶. Many classical Hsp90 inhibitors (shown in Figure 1) have been designed to target the ATP-binding pocket of NTD, and they can block ATP hydrolysis thereby leading to the degradation of client proteins¹⁷. Kim et al. reported a purine-based Hsp90 inhibitor MPC-3100, which significantly suppressed the growth of various human tumour-induced xenografts and possessed superior pharmacokinetics⁶. Moreover, MPC-3100 has successfully passed phase I clinical trials, yet poor solubility was a major obstacle to its further clinical development¹⁸. IPI-504 was a water-soluble analogue of 17-AAG (an efficient Hsp90 inhibitor) and the potential anti-tumour effects of IPI-504 have been validated in preclinical experiments^19,20. However clinical experiments revealed that the patient developed hepatic failure and ketoacidosis while receiving IPI-504 medications²¹. Sharp et al. discovered a novel 3,4-diaryl pyrazole Hsp90 inhibitor CCT018159, which possessed better potency against Hsp90 protein compared to 17-AGG with the potential advantage of water solubility²². However, the cellular inhibitory activity of CCT018159 was inferior to that of 17-AGG. NVP-BEP800 is a 2-aminothieno[2,3-d]pyrimidine Hsp90 inhibitor with nanomolar activity against a series of human tumour cell lines, yet the weak water solubility hindered its application in some contexts^23,24. So far, there are no available Hsp90 inhibitors on the market due to their poor solubility, limited efficacy, and multiple side effects^25,26. Therefore, in this study, we aimed to discover novel Hsp90 inhibitors with high efficacy and broad anti-tumour activity.

Figure 1. The chemical structures of known inhibitors of Hsp90.

Nowadays, structure-based virtual screening (SBVS), including pharmacophore mapping and molecular docking, is widely used in drug discovery due to its efficiency and convenience^27,28. Additionally, molecular dynamics (MD) simulation is usually carried out to estimate the stability of protein-ligand systems²⁹. In this study, we successfully identified four potent Hsp90 inhibitors from an internal database using SBVS. Then the results of MD simulations indicated the good stability of four protein-hit systems. Further in vitro assays revealed that the four hits exhibited strong inhibitory activities against the Hsp90 protein. Among them, the optimal inhibitor HP-4 significantly suppressed the proliferation of multiple human tumour cells and the progression of HCT-116 cell-derived xenograft.

Materials and methods

Cell culture

Human tumour cells including HCT-116, HeLa, A549, A2780, DU145, HepG2 and A498 cells were obtained from American Type Culture Collection (Manassas, VA, USA). HCT-116, DU145 and A2780 cells were cultured in RPMI-1640 medium. HeLa, A549, HepG2 and A498 cells were cultured in Dulbecco’s modified Eagle’s medium. All media were supplemented with 10% foetal bovine serum and 1% penicillin-streptomycin. Cell cultures were maintained in the humidified incubator with 5% CO₂ at 37 °C.

Generation of pharmacophore model

According to the method previously reported³⁰, the establishment of the pharmacophore model was carried out in the Molecular Operating Environment (MOE) program. The X-ray crystallographic structure of Hsp90 protein (PDB ID: 2XJX) was retrieved from the Protein Data Bank (PDB) with a resolution value of 1.66 Å. The Hsp90 protein was prepared by removing water, adding polar hydrogen bonds, merging non-polar hydrogen bonds, and calculating gasteiger charges using MOE. Then, the Ligand Interactions tool of MOE was used to analyse the interaction between Hsp90 active-site residues and the original ligand. Finally, based on the results of interaction analysis, pharmacophore features including hydrogen-bond donor, one aromatic centre, and hydrophobic feature were manually built using the Pharmacophore Query Editor of MOE.

Validation of pharmacophore model

To validate the selectivity of the generated Hsp90 pharmacophore model, the virtual screening of a testing database including 19 known Hsp90 inhibitors and 1110 decoy compounds downloaded from the enhanced Database of Useful Decoys (DUDe) was performed using the pharmacophore search protocol of MOE^31,32. Then MedCalc software was employed to analyse the receiver operating characteristic (ROC) curve³³. In ROC analysis, the area under the curve (AUC) is generally used to assess the ability of the pharmacophore model to distinguish active or inactive compounds.

Virtual screening

The virtual screening was carried out as described previously³⁴. First of all, an internal compound database containing 35 000 compounds was created by the combinatorial chemistry method, and the Energy Minimise algorithm in MOE was employed to generate three-dimensional (3D) structures of all compounds in the database. Then, 8750 compounds with drug-like properties from the database were filtered by the following modified Lipinski’s rules: molecular weight (MW) ≤ 600 Da, number of hydrogen-bond donors (nHD) ≤ 10, number of hydrogen-bond acceptors (nHA) ≤ 10, and log P_(o/w) ≤ 6. Subsequently, the Pharmaceutical Search module of MOE was used to perform a pharmacophore-based virtual screening, and the generated Hsp90 pharmacophore model served as a 3D query. The root mean square distance (RMSD) value was used to evaluate the matching degree between each compound and the pharmacophore model. In general, a lower RMSD value suggests a better match. The 27 screened compounds with RMSD < 1 Å were then subjected to molecular docking.

Molecular docking

The crystallographic structure of Hsp90 protein (PDB ID: 2XJX) from the protein data bank was prepared by removing water, adding polar hydrogen bonds, merging non-polar hydrogen bonds, and calculating gasteiger charges using MOE. In this study, molecular docking interactions were performed using the Dock tool of MOE as described previously³⁴. Each selected compound was docked into the active site of Hsp90 with the triangle matcher method, and the docking score was calculated by the dG scoring function. In general, the lower docking score indicates a stronger binding affinity between the ligand and Hsp90. Finally, four compounds were selected for further study according to the docking scores.

MD Simulation

MD simulations were performed by GROMACS 2021.5 as previously reported to explore whether each Hsp90-hit system was stable³⁵. The topol file of each hit was generated by Acpype Server (www.bio2byte.be) with GAFF force field and the topol file of Hsp90 was built by GROMACS 2021.5 with AMBER99SB-ILDN force field. A cubic box of 0.9 nm was constructed with the Hsp90-hit system placed in the centre and simple point charge (SPC) water molecules were added to solvate the Hsp90-hit system. Subsequently, appropriate Na⁺ and Cl⁻ were added to keep the system neutral, and the system was then subjected to a 5000-step energy minimisation with the steepest descent algorithm. NVT simulation was performed for 100 ps with a V-rescale thermostat to keep the system at 300K and later another 100 ps NPT simulation was conducted with a Parinello–Rahman barostat to maintain the system at 1 atm. Finally, each system run for 100 ns MD simulations, and the atomic coordinates of each system were recorded at every 10 ps for further analysis.

Fluorescence polarisation (FP) measurements

Hsp90 protein was obtained from Abcam (Cambridge, MA, USA). The Hsp90 fluorescence polarisation binding assay was carried out as described previously³⁶. Gerdanamycin (GM), a natural product that binds to the ATP pocket of Hsp90, was used to identify novel Hsp90 inhibitors based on the competition of fluorescent (BODIPY)-labelled GM with Hsp90 protein. Different concentrations of inhibitors and Hsp90 in the assay buffer were mixed in black 96-well microplates. Ten minutes later BODIPY-labelled GM was added at a final concentration of 5 nM. The binding reaction was allowed to proceed for 3 h at room temperature. Fluorescence polarisation was measured on a SpectraMax GeminiXS (Molecular Devices, Sunnyvale, CA) with excitation and emission wavelengths of 485 and 530 nm, respectively.

In vitro cytotoxicity assay

According to the experimental method reported previously³⁷, cells were plated on each well (5 × 10⁴ cells) and kept overnight. Then, different concentrations of inhibitors were added to each well and incubated at 37 °C for 72 h. Thereafter, the medium was removed, and MTT was added. Cells were incubated with medium containing MTT for an additional 4 h, the supernatant was removed by centrifugation, and 150 μL of DMSO was added to each well and gently shaken for 15 min. The absorbance of the plate was recorded at 570 nm using a microplate reader.

In vivo anti-tumour activity

All experimental protocols were approved by the Animal Ethics Committee of China Pharmaceutical University. According to the experimental method reported previously³⁷, cultured cells of HCT-116 were washed with D-Hank’s solution and prepared as a suspension at a concentration of 10⁶ cells/mL in culture medium. The cells were injected subcutaneously into the right side of BALB/c nude mice. Mice were randomly divided into three groups: control group, low-dose treatment group (5 mg/kg), and high-dose treatment group (10 mg/kg). Each group was treated with the intraperitoneal injection of inhibitors. The tumour was measured every three days, and the tumour volume was calculated using the formula (c × c × d)/2 (c, the smallest diameter; d, the largest diameter).

Results and discussion

Generation of pharmacophore model

In this study, the X-ray crystallographic structure of Hsp90 protein (PDB ID: 2XJX) complex with the original ligand was retrieved from the protein data bank. After the Hsp90 protein was prepared by removing water, adding polar hydrogen bonds, merging non-polar hydrogen bonds, and calculating gasteiger charges using MOE, this protein was used to establish a Hsp90 pharmacophore model. Based on the interaction analysis of Hsp90 and the original ligand, a pharmacophore model consisting of one hydrogen-bond donor (F1), one aromatic centre (F2), and two hydrophobic features (F3 and F4) was generated (Figure 2). The F1 feature overlapped the hydroxyl group of the original ligand, which formed the hydrogen-bond interaction with the carboxylate of Asp93. The hydrophobic feature F3 formed a hydrophobic interaction with Ala55. The F2 and F4 which mapped the aromatic ring and hydrophobic group of the ligand, respectively, formed hydrophobic interactions with the side chain benzene ring of Phe138. The pharmacophore features described above were used for subsequent virtual screening of Hsp90 inhibitors.

Figure 2. The pharmacophore model of Hsp90. The hydrogen-bond donor (F1) and aromatic centre (F2) were indicated by green and orange spheres, respectively. The two hydrophobic centroid features F3 and F4 were indicated by yellow spheres. The dashed black line represented hydrogen-bond interaction. Each key residue was annotated with a three-letter amino acid code.

Validation of pharmacophore model

For virtual screening, the specificity and accuracy of the pharmacophore model are significant. Hence, the constructed Hsp90 pharmacophore model was validated using a testing set database consisting of 19 known inhibitors of Hsp90^31,32 and 1110 decoy compounds from DUDE. Firstly, we combined the activity compounds with bait compounds and then conducted preliminary virtual screening using the Hsp90 pharmacophore model. The ROC curve was analysed to evaluate the quality of the Hsp90 pharmacophore model³³. As shown in Figure 3, the AUC value was extremely close to 1.0 (0.921), indicating the good ability of the Hsp90 pharmacophore model to differentiate true actives from decoy compounds (p < 0.001).

Figure 3. ROC curve of the generated Hsp90 pharmacophore model.

Virtual screening

Figure 4 showed the screening strategy of Hsp90 inhibitors in this study. First, the modified Lipinski’s rules were employed to filter drug-like compounds from the database, and 8750 compounds were retrieved. Then, the 8750 filtered compounds were subjected to the pharmacophore-based virtual screening, and the constructed Hsp90 pharmacophore model was used as a 3D query. In this step, 27 compounds with RMSD values lower than 1 Å were screened. Subsequently, a docking-based virtual screening was performed to predict the binding affinities between 27 compounds and the Hsp90 active site, and a classical Hsp90 inhibitor MPC-3100 was chosen as the positive control. Since the docking score of MPC-3100 was −13.31 kcal/mol, the value of −13.31 was used as the threshold for selecting candidate inhibitors of Hsp90. Finally, four top compounds (termed as HPs 1–4) with a docking score lower than −13.31 kcal/mol were chosen for further study (Figure 5 and Table 1).

Figure 4. The protocol for virtual screening and biological evaluation of Hsp90 inhibitors.

Figure 5. The chemical structures of HPs 1–4.

Table 1. The drug-like properties, docking sores and biological data of HPs 1–4 and MPC-3100.

Name	MW	nHD	nHA	Log P(o/w)	Docking Score [kcal/mol]	IC50 (nM)^a
HP-1	405.83	4	7	2.41	–13.57	23.32 ± 1.77
HP-2	361.78	4	6	2.56	–13.45	44.91 ± 2.58
HP-3	436.30	4	6	3.05	–13.39	61.69 ± 3.06
HP-4	585.01	8	9	1.41	–13.71	17.64 ± 1.45
MPC-3100	549.44	2	7	2.89	–13.31	136.16 ± 4.27

^a The inhibitory effects of HPs 1–4 and positive control MPC-3100 on Hsp90 protein.

Interaction analysis

A molecular docking study was further performed to predict the binding poses of HPs 1–4 to the Hsp90. Figure 6 showed the interactions between HPs 1–4 and Hsp90. We found that all of the HPs 1–4 created hydrogen-bond interactions with Asp93 and the benzene ring of them formed hydrophobic interactions with Ala55, which matched well with the generated pharmacophore features F1 and F3. Moreover, all of the HPs 1–4 formed hydrophobic interactions with Phe138, which corresponded to the pharmacophore features F2 and F4. Notably, both HP-1 and HP-2 formed an extra hydrogen bond with Asp 106 while HP-4 created an additional hydrogen bond with Asp 102 and Asn 51, respectively. These extra hydrogen bonds may assist in stabilising the inhibitor in the active site of Hsp90. Overall, these results indicate that HPs 1–4 can interact with critical residues of Hsp90 and could be potent Hsp90 inhibitors.

Figure 6. The docking modes of HPs 1–4 with key residues of Hsp90. (A) HP-1; (B) HP-2; (C) HP-3; (D) HP-4. The dashed black line represents hydrogen-bond interaction.

MD Simulations

Subsequently, HPs 1–4-Hsp90 complexes were subjected to 100 ns MD simulations for the evaluation of the stability of HPs 1–4 at the active site of Hsp90 evolved with time. The RMSD was calculated to determine the stability of the system during the whole MD simulations. It was observed that the RMSDs of HP-1-Hsp90 and HP-2-Hsp90 were relatively stable at about 0.45 nm, the RMSD of HP-3-Hsp90 complex was relatively stable at about 0.40 nm, and the RMSD of HP-4-Hsp90 complex was relatively stable at about 0.30 nm (Figure 7(A, C, E, G)). Moreover, we also measured the RMSDs of HPs 1–4. Figure 7(B, D, F, H) displayed the RMSD plots for HPs 1–4 during the simulation, respectively. We found that though some fluctuations were observed, the RMSDs of HPs 1–4 finally tended to equilibrium at about 0.16 nm, 0.18 nm, 0.20 nm, and 0.36 nm, respectively. These data suggest that HPs 1–4-Hsp90 complexes were fairly stable during the whole simulations and HPs 1–4 could bind stably with Hsp90. Furthermore, the mobility of backbone Cα atoms of each Hsp90 residue was depicted using the root mean square fluctuations (RMSF) values to evaluate the flexibility of HPs 1–4-Hsp90 complexes systems. A smaller RMSF value indicates less residue flexibility and the lower fluctuations suggest better protein stability. Figure 8(A–D) showed that the key residues Ala55, Asp93, and Phe138 in the Hsp90 active site of four complexes systems were all not too flexible in the whole simulations with RMSFs less than 0.10 nm, reflecting that HPs 1–4 could interact strongly with these crucial residues. Based on the above RMSD and RMSF analysis, we can conclude that HPs 1–4 could stably bind to the active pocket of the Hps90. Finally, DSSP analysis was carried out to evaluate whether the secondary structure of the Hsp90 protein was changed during the whole MD simulations. As shown in Figure 8(E–H), there were no obvious changes in structural elements, suggesting the structural stability of Hsp90.

Figure 7. RMSD of atoms during MD simulations. (A, B) RMSDs of HP-1-Hps90 complexes and HP-1, respectively; (C, D) RMSDs of HP-2-Hps90 complexes and HP-2, respectively; (E, F) RMSDs of HP-3-Hps90 complexes and HP-3, respectively; (G, H) RMSDs of HP-4-Hps90 complexes and HP-4, respectively;.

Figure 8. (A–D) RMSF of Cα atoms of Hsp90 residues in HPs 1–4-Hsp90 complexes, respectively. The colour code is HP-1-Hsp90 (yellow), HP-2-Hsp90 (red), HP-3-Hsp90 (green), and HP-4-Hsp90 (bule). (E–H) Secondary structure of Hsp90 in HPs 1–4-Hsp90 complexes, respectively.

Inhibitory activities of HPs 1–4 on Hsp90

Previous studies have confirmed that GM binds to the ATP binding site of the N-terminal domain (NTD) of Hsp90, and GM dose-dependently inhibits the ATPase activity of Hsp90³⁸. Hence, we conducted a fluorescent (BODIPY)-labelled GM-FP experiment to verify whether HPs 1–4 could competitively bind to the ATP-binding pocket of the NTD of Hsp90. The Hsp90 ATP inhibitor MPC-3100 served as a positive control. As shown in Table 1, all four compounds could competitively inhibit the ATP binding site activity of GM to Hsp90 with IC₅₀ values ranging from 17.64 to 61.69 nM, which were all smaller than 100 nM. Among them, HP-4 exhibited the most potent inhibitory effects with the IC₅₀ value of 17.64 ± 1.45 nM. Furthermore, we measured the inhibitory effects of MPC-3100 against Hsp90 protein with the IC₅₀ value of 136.16 ± 4.27 nM. By comparison, the inhibitory activity of HP-4 on Hsp90 was stronger than that of MPC-3100 by about 7.7-fold. In addition, to further verify that HPs 1–4 targeted the ATP binding site, a docking study was conducted. As shown in Figure 9, the original ligand and all HPs 1–4 were able to bind to the ATP-binding pocket of NTD of Hsp90.

Figure 9. The docking modes of HPs 1–4 and the original ligand bind to the ATP-binding pocket of NTD of Hsp90.

In vitro cytotoxicity evaluation

In view of the high potential inhibitory activities of HPs 1–4 on Hsp90 protein, a typical MTT method was used to assess their cytotoxicity against human colon tumour cells HCT-116. MPC-3100 was chosen as a positive control. As shown in Figure 10, the IC₅₀ value of MPC-3100 inhibiting the growth of HCT-116 cells was 779.59 nM. Moreover, HPs 1–4 significantly suppressed the proliferation of HCT-116 cells with IC₅₀ values ranging from 148.52 nM to 559.37 nM, which were all smaller than that of MPC-3100. Among them, HP-4 exhibited the best anti-tumour activity, which is about 5 times stronger than that of MPC-3100. Subsequently, we evaluated the inhibitory effects of HP-4 on other human tumour cells. As shown in Table 2, HP-4 exhibited good broad-spectrum anti-tumour effects on a series of tumour cells including HeLa, A549, A2780, DU145, HepG2, and A498.

Figure 10. The cytotoxicity of MPC-3100 and HPs 1–4 against human colon cancer HCT-116 cells.

Table 2. Biochemical profile of HP-4.

Assay	IC50 (nM)^a
HCT-116 cell proliferation	148.52
HeLa cell proliferation	206.39
A549 cell proliferation	210.81
A2780 cell proliferation	372.06
DU145 cell proliferation	289.14
HepG2 cell proliferation	337.92
A498 cell proliferation	254.17

^a IC₅₀ (nM) is the concentration of compound needed to reduce cell growth by 50% following 72 h cell treatment with HP-4.

In vivo anti-tumour effects of HP-4

In order to further investigate the anti-tumour efficacy of HP-4 in vivo, HCT-116 cell-derived xenograft models were established in BALB/c nude mice. The HCT-116 tumour-bearing mice were divided into three groups including HP-4 (5 mg/kg), HP-4 (10 mg/kg) and the vehicle control group. As shown in Figure 11, the tumour volume of mice after the administration of HP-4 was remarkably reduced, and HP-4 could significantly inhibit the growth of HCT-116 tumour compared with the vehicle control. These results indicate that HP-4 exhibited strong therapeutic effects on HCT-116 tumour-bearing mice.

Figure 11. In vivo anti-tumour effects of HP-4. Data are represented as mean ± SD, n = 5. ***p < 0.001.

Conclusion

Since the importance of Hsp90 in the proliferation of tumour cells, it has been a very promising target in the development of anti-tumour drugs. However, no Hsp90 inhibitors have been available on the market so far. In order to discover novel and efficient Hsp90 inhibitors, we successfully identified four compounds using a combined virtual screening strategy. In vitro test results showed that the four compounds not only had strong inhibitory activities on Hsp90 protein but also exhibited significant inhibitory effects on the proliferation of human tumour cells. Among them, the optimal inhibitor HP-4 could significantly inhibit the tumour growth of HCT-116 tumour-bearing mice. In general, HP-4 could be a promising lead compound targeting Hsp90. Structural optimisation of HP-4 will be further carried out in our laboratory for the development of novel Hsp90 drugs.

Ethical statement

All experimental protocols were reviewed and approved by the Animal Ethics Committee of China Pharmaceutical University.

Author contributions

YX, YZ and SZ performed experiments and acquired data. YX, YZ and SZ collected experimental samples. MMN and YZ performed statistical analysis. JL, YZ, ZX and LY analysed data and wrote the paper. All authors had edited and approved the final manuscript.

Disclosure statement

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Data availability statement

The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding author/s.

Additional information

Funding

This study is supported by Jiangsu Research Hospital Association for Precision Medication [JY202229] and Fengcheng Talent Plan Youth Science and Technology Talent Recruitment Project of Taizhou [202210].