Discovery of new molecular entities able to strongly interfere with Hsp90 C-terminal domain

Heat shock protein 90 (Hsp90) is an ATP dependent molecular chaperone deeply involved in the complex network of cellular signaling governing some key functions, such as cell proliferation and survival, invasion and angiogenesis. Over the past years the N-terminal protein domain has been fully investigated as attractive strategy against cancer, but despite the many efforts lavished in the field, none of the N-terminal binders (termed “classical inhibitors”), currently in clinical trials, have yet successfully reached the market, because of the detrimental heat shock response (HSR) that showed to induce; thus, recently, the selective inhibition of Hsp90 C-terminal domain has powerfully emerged as a more promising alternative strategy for anti-cancer therapy, not eliciting this cell rescue cascade. However, the structural complexity of the target protein and, mostly, the lack of a co-crystal structure of C-terminal domain-ligand, essential to drive the identification of new hits, represent the largest hurdles in the development of new selective C-terminal inhibitors. Continuing our investigations on the identification of new anticancer drug candidates, by using an orthogonal screening approach, here we describe two new potent C-terminal inhibitors able to induce cancer cell death and a considerable down-regulation of Hsp90 client oncoproteins, without triggering the undesired heat shock response.

it represents a challenging target for structural analysis. In addition to the structural complexity of this protein, the absence of crystal structures of C-terminal Hsp90-inhibitor complexes, represents the main drawback for progress in the field.
Despite the above-mentioned difficulties and the lack of a convincing grasp regarding the exact structural requirements for C-terminal domain interactions, recently we reported the identification of new potent dihydropyrimidinone based Hsp90 inhibitors that target the C-terminal binding pocket [25][26][27][28] .

Results and Discussion
In order to continue our research program aimed at expanding the number of Hsp90 C-terminal inhibitors, we decided to utilize the surface plasmon resonance (SPR) [25][26][27][28][29] assay for screening a collection of low molecular weight synthetically accessible compounds, selected in order to explore the chemical space encoded by different scaffolds. In more details, a set of forty-eight commercially available small molecules (Table S1, Supplementary Material), endowed with different structural features (Fig. S1, Supplementary Material) was subjected to SPR screening on recombinant Hsp90α for testing their ability to bind to the immobilized protein.
Based on this assay, sixteen compounds with novel chemical scaffolds Fig. 1 have been identified as high affinity leads for the Hsp90 chaperone with low K D values (see surface plasmon resonance analyses in material and methods section).
In light of these results, we decided to evaluate their potential anti-proliferative activity against Jurkat (human T-lymphocyte) and U937 (human monocyte form histiocytic lymphoma) cell lines, that were exposed to increasing concentrations of compounds 1, 4-5, 7-15, 20-22, 26 or novobiocin, a well characterized Hsp90 C-terminal inhibitor 30 . Cells viability was evaluated at 24 or 48 h by MTT assay and IC 50 values are reported in Tables S2 and S3 (Supplementary Material), respectively. Data showed the best results for 7 and 10 that affected the cell viability in both leukemia-derived cell lines with the lowest IC 50 values, so they were selected for a deeper biological evaluation. Notably these compounds do not affect the proliferation of non-tumor cell line PHA-stimulated human peripheral blood mononuclear cells (PBMC). In fact, the non-viable cells percentage after 24 or 48 h of treatment with compound 7 (25 µM) (about 6 ± 1.8%, 8 ± 1.5%, respectively), or with with compound 10 (25 µM) (about 7 ± 1.3%, 9 ± 1.8%, respectively), was similar to the value observed in DMSO treated control cells (about 5 ± 0.9%).
Since Jurkat (human leukemic T lymphocyte) cells were about two-fold more susceptible than U937 (human monocytic cell line from histiocytic lymphoma) cells, they were used for additional analysis with the aim of characterizing the mechanism of the observed anti-proliferative effect. In order to discriminate if 7 and 10 affected cell cycle progression and/or inducing cell death, the cells were incubated for 24 or 48 h with a concentration of these two compounds close to their IC 50 values or two and four times higher. Novobiocin was used at the same conditions with concentrations close to its IC 50 value (200 µM) or at lower doses (50 or 100 μM).
Cell cycle distribution analysis showed that 7 and 10 moderately prevented the cycle progression by arresting the cells in G 0 /G 1 or S phase, respectively, after 24 h ( Fig. 2A,B) without any significant increase of subG 0 /G 1 cell fraction, indicative of apoptotic/necrosis cell death (Fig. 2D,E). These dose-dependent cytostatic effects of 7 and 10 were retained even after 48 h ( Fig. 2A,B). Conversely, the hypodiploid cells (subG 0 /G 1 DNA content) increased at 10%, after 48 h cell treatment, with 7 or 10 used at high concentrations (100 µM), as shown in Fig. 2D,E. The novobiocin cell treatment triggered different responses with respect to 7 and 10, indeed, when it was used at low concentration (50 and 100 µM) for 24 or 48 h, it did not affect cell cycle progression. On the other hand, the exposure to high dose of novobiocin (200 µM), caused cell accumulation in G 0 /G 1 , or S phase (Fig. 2C) with a 14% or 30% of hypodiploid cells after 24 h or 48 h, respectively (Fig. 2F). These data indicate that 7 and 10 induced cytostatic effect after 24 h-exposure and cytostatic/cytotoxic effect after 48 h-exposure on leukemic cells at low doses, compared to novobiocin.
In addition we analyzed the cell cycle cyclin-dependent kinases, Cyclin A, Cyclin D, CDK2, and CDK4 following the treatment with compounds 7 and 10. Cyclin D and CDK4 regulates the G0/G1 phase cycle, and their expression levels were slightly down-regulated in Jurkat cells with compound 7 treatment (see Fig. 3A). Similarly compound 10 caused the decrease of Cyclin A and CDK2, involved in normal S-phase progression.
In order to further characterize the most effective compounds and to provide evidence of a bona fide inhibition of Hsp90′s activity, the levels of some representative oncogenic Hsp90-client proteins from Jurkat cell lysates, treated with the selected compounds (7 and 10), were verified through western blot analysis. As shown in Fig. 3B incubation with 7 and 10 induced a significant degradation (50-60%) of Hsp90-dependent client proteins Raf-1, p-Akt, p-Erk and p53 in a concentration-dependent way, without affecting actin levels, which is not dependent upon Hsp90; a minimal reduction of Hsc70 (15-20%) was observed. Furthermore, the treatment of our compounds did not induce ER chaperone grp94 expression.
Notably, our data demonstrate that these new potent Hsp90 inhibitors did not induce any considerable increase in Hsp90 and Hsp70 protein levels which, conversely is a hallmark resulting from Hsp-90 N-terminal inhibition (Fig. 3). Indeed, 17-AGG, a well-known N-terminal inhibitor 31 , caused a significant up-regulation of Hsp90 and Hsp70 expression levels (Fig. 3).
Thus, basing on these data, we can infer that our compounds have a different modulation effect on Hsp90 than 17-AAG, compatible with Hsp90 C-terminal inhibition. In order to characterize the precise protein region involved in 7 or 10 recognition, a mass spectrometry assisted limited proteolysis strategy was used.
By this approach, through the differences in the proteolytic patterns observed in presence or in absence of putative protein ligands, it is possible to identify the protein regions involved in the molecular interactions 25,32,33 . Limited proteolysis experiments were performed on Hsp90α and on Hsp90α/7 and on Hsp90α/10 complexes; moreover, the same experiment was also carried out on the protein interacting with novobiocin, selected as positive control 27,28 .
The preferential trypsin or chymotrypsin cleavage sites, detected for Hsp90α and for the different Hsp90α complexes, identified on the basis of MALDI analysis of the respective digestion mixture, are shown in Fig. 4. A comparison between the results achieved in these experiments suggests that the interaction of Hsp90α with 7 and with 10, led to a reduced protease accessibility of Lys484, Lys498 and Lys656, thus indicating that the middle and C-terminal domains of Hsp90α are preferentially involved in the molecular binding. In addition, the conformational changes of Hsp90, induced by compound 7 or 10, is similar to the cleavage sites we detected on Hsp90α/ novobiocin 30,34,35 .
Finally, the molecular basis behind the observed Hsp90 inhibitory activity of the compounds was clarified at molecular level using docking experiments. All the compounds, filtered out by SPR experiments, were docked onto closed active crystal structure of Hsp90α homologue (PDB code: 2CG9) 36 . In more details, we have used the results of limited proteolysis experiments reported above to disclose the domain interacting with our tested compounds, and, during the conformational searches, we have considered a region including the Arg670 Hsp82 (Lys656 Hsp90 ) that is involved in the molecular binding (see above). During our in silico analysis, we referred to sequence alignment with the human chaperone protein, reported by Lee et al. 37 . In order to assess the binding of tested compounds on Hsp90α, we have performed molecular docking experiments using the induced fit docking protocol (Schrödinger Suite) 38-40 , to account for flexibility of ligands and receptor 41 . In light of the surface plasmon resonance results and the biological data (see Table S3, Supplementary Material), we have disclosed two novel not nature-inspired Hsp90 C-terminal inhibitors (7 and 10). For these reasons, during our analysis, we have compared: 7 with its structural related compounds 8 and 9, and 10 with respect to 11-14. Considering the first group of molecules, from the structural point of view, we have attributed the different biological activity of 7 with respect to 8 and 9 to the diverse substitutions at C-5 of thiazole ring. All the three compounds, in fact, occupy the same region of C-terminal domain, establishing a similar pattern of H-bonds between their phenylamino acetamide and Leu671 and Leu674 of both the chains (A and B, Fig. S2 and S3 Supplementary Material), while the 3,4-dichlorobenzyl moiety of 7 is further involved in a peculiar halogen bond with the side chain of Arg670 ChainA (See Fig. 5A). On the other hand, for the second group of molecules (10-14), we noticed   In addition, its bromide atom forms a halogen bond with the CO of Thr638 ChainB .
On the other hand , we observed a hydrogen bond between the quinazolinedione and Arg670 ChainA and Leu674 ChainA (see Fig. 5B and S5, Supplementary Material). For the above considerations, the substitution on N-hexanamide is fundamental for biological activity of this type of molecules, thanks to the additional interactions established between this molecular portion and the receptor counterpart (see Table S3 and Fig. S4, Supplementary Material). Considering these two different groups of novel chaperone binders, our structural results suggest that the precise pattern of hydrogen bonds with the leucine aminoacids (Leu671, Leu674, Leu676) direct their placement at the interface of the two Hsp90 A and B chains. Notably, the halogen interactions with Thr638 ChainB for 7 and Asp641 ChainA for 10 have been confirmed as fundamental interactions suitable for the design of novel more effective C-terminal Hsp90 inhibitors [26][27][28] .
For the considerations above, the best poses of the most active compounds, 7 and 10, in complex with Hsp90 were used as a starting point for molecular dynamics (MD) simulations (100 ns) in explicit solvent (Desmond software) 42-44 , in order to capture the dynamic nature of protein-ligand interactions and to investigate the ligand affinity and selectivity towards the target 24,45 . In particular, for the analysis of the trajectories we have used the simulation interactions diagram (SID) (Maestro version 10.2) 46 a tool for exploring protein-ligand interactions. During the simulation, 7 and 10 are located at the interface of the two chains of the C terminal domain, maintaining a good number of interactions (Fig. 6) with the amino acids singled out in the induced fit docking results analysis (see above). From the analysis of the protein-ligand (P-L) contacts (Fig. 6), in addition to further interactions of 7 with Hsp90 (namely with Leu676, Glu624, Asp641) maintained for ≈50% of the simulation time, 7 establishes hydrophobic contact and hydrogen bonds with Leu671 (>80%), with respect to 10, that mainly interacts with Leu674 (>70%). Hence, the pattern of hydrogen bonds and the number of hydrophobic contacts both with the C-terminal domain (Fig. 6) seem to be the driving forces of the target-ligand complexes.
Therefore, basing on the data here discussed and already reported by us 25,27,28,41 and by other research groups 45   second nucleotide pocket on Hsp90-C terminus, for which to date, only computational hypothesis can be proposed because its precise physical location in the CTD as well as its exact physiological role are still unclear 49 .
In conclusion, in the course of our investigations we succeeded to disclose two new structurally unrelated attractive hits with a high binding affinity for C-terminal Hsp90 through a multidisciplinary approach, including SPR protein-ligand interaction measurement, limited proteolysis, molecular docking, molecular dynamics receptor exploration, which allowed us to identify the region of binding of these novel entities. These molecules showed to potently down-regulate the oncogenic clients of the chaperone, the cell cycle cyclin-dependent kinases, inducing cancer-cell death without activating the deleterious HSR response.
These findings are of great interest for progress in the field as they provide an excellent opportunity to expand the chemical space associated with Hsp90 C-terminal inhibition, after a careful molecular optimization process; indeed, the identification of new selective C-terminal modulators are greatly desired because they can represent both attractive anticancer drug candidates and may serve as chemical probes to better clarify the structural requirements needed for an optimal protein-ligand interaction. Such studies are currently underway and will be reported in due course.

Materials and Methods
General information. All compounds  are available at the stocks of the Otava Chemicals, Otava Ltd.  [25][26][27][28]32,50 . Surface Plasmon Resonance Spectroscopy (SPR) analyses were performed to determine binding of various molecules to full length Hsp90α using a Biacore 3000 optical biosensor equipped with research-grade CM5 sensor chips (GE Healthcare). Hsp90α was coupled to the surface of a CM5 sensor chip using standard amine-coupling protocols, according to the manufacturer's instructions.
The protein (100 µg mL −1 in 10 mM CH 3 COONa, pH 5.0) was immobilized on individual sensor chip surfaces at a flow rate of 5 µL min −1 to obtain densities of 8-12 kRU. For the experiments a recombinant Hsp90α surface, a BSA surface and one unmodified reference surface were prepared for simultaneous analyses. Compounds 1-48, as well as 17-AAG were dissolved, to obtain 4 mM solutions, in 100% DMSO and diluted 1:200 (v/v) in PBS (10 mM NaH 2 PO 4 , 150 mM NaCl, pH 7.4) to a final DMSO concentration of 0.5%. For each molecule a six-point concentration series were set up, spanning 0 -0.02 -0.08 -0.25 -1 µM, and for each sample the complete binding study was performed using triplicate aliquots. Changes in mass, due to the binding response, were recorded as resonance units (RU). To obtain the dissociation constant (K D ) (Table 1), these responses were fit to a 1:1 Langmuir binding model by nonlinear regression using the BiaEvaluation software program provided by GE Healthcare. Simple interactions were suitably fitted to a single-site bimolecular interaction model (A + B = AB), yielding a single K D . (Table 1). SPR experiments were performed at 25 °C, using a flow rate of 50 µL min −1 , with 60 s monitoring of association and 300 s monitoring of dissociation.
Anti-proliferation experiments. Jurkat and U937 (2 × 10 4 /well) cells were seeded in triplicate in 96 well−plates and incubated for the 24 h or 48 h in the absence or presence of different concentrations of compounds 1, 4-5, 7-15, 21-22, 26 (concentration between 10 µM to 100 µM) or Novobiocin (concentration between 50 µM to 500 µM). Stock solutions of compounds (100 mM in DMSO) were diluted just before addition to the sterile culture medium and the final concentration of DMSO was 0.15% (v/v).
The number of viable cells was determined by using a [3-4,5-dimethyldiazol-2-yl]-2,5-diphenyl tetrazolium bromide (MTT, Sigma-Aldrich) conversion assay, according to the method described by Terracciano et al. 2016 27,28 . After the indicated treatment, the cells were incubated for additional 3 h at 37 °C, with 25 µL of MTT (5 mg/mL in PBS). The formazan crystals thus formed were dissolved in 100 μl of buffer containing 50% (v/v) N,N-dimethylformamide, 20% SDS (pH 4.5). The absorbance was read at 570 nm using an Absorbance Microplate Reader (Titertek multiskan MCC7340, LabSystems, Vienna, VA, USA) equipped with a 620 nm filter. The cell population growth inhibition was also tested by cytometric counting (trypan blu exclusion). IC 50 values were calculated from cell viability dose−response curves and defined as the concentration resulting in 50% inhibition of cell survival, compared to control cells treated with DMSO (see Supplementary Material Tables S2 and S3).
Human peripheral blood mononuclear cells (PBMC), induced to proliferate by phytohemagglutinin (PHA) (10 µg/ml), were used to evaluate cytotoxic effects by trypan blue count of compound 7 and 10. Human peripheral blood mononuclear cells (PBMC) were isolated from buffy coats of healthy donors (kindly provided by the Blood Center of the Hospital of Battipaglia, Salerno, Italy) by using standard Ficoll−Hypaque gradients. Freshly isolated PBMC contained 92.8 ± 3.1% live cells, were incubated with DMSO or compounds 7 and 10 used at 25 µM (concentrations close to IC 50  Western Blot. The Jurkat cells were seeded in RPMI medium with DMSO or compounds 7 or 10 (25 or 50 µM) or the Hsp90 inhibitor 17-AAG for 24 h. Following the treatment, the cells were trypsinized and homogenized on ice in RIPA lysis buffer (50 mM Hepes, 10 mM EDTA, 150 mM NaCl, 1% NP-40, 0.5% sodium deoxycholate, 0.1% SDS, pH 7.4), supplemented with protease inhibitors cocktail (Sigma-Aldrich). After centrifugation at 4 °C to removed cell debris, an equal protein amount (40 µg) was separated by SDS-PAGE under denatured reducing conditions and was then transferred to nitrocellulose membranes. The blots were blocked with 3% BSA and were incubated at 4 °C overnight with primary antibodies: anti-Hsp 70; anti-Hsp 90α/ß, anti-Raf1, anti-pAkt, anti Erk, anti pErk, anti-p53, anti Hsc70, anti GRP94, anti cyclin A antibodies (Santa Cruz Biotechnology, Inc., Delaware, CA, USA), anti-actin antibody (Sigma-Aldrich) anti CDK2, anti Cyclin D1 and anti CDK4 antibodies (Cell Signaling Technology, Leiden, Netherlands). Appropriate peroxidase-conjugate secondary antibodies were used to detect primary antibodies, and enzymatic signals were visualized by chemiluminescence reagent Pierce ECL (Thermo Scientific, Rockford, IL, USA). Quantitative densitometry analyses were performed using a ImageQuant LAS 4000 system (GE Healthcare Life Sciences,NY, USA).
Limited Proteolysis. The assays 27,28,50-52 were conducted on recombinant Hsp90α at 37 °C; PBS 0.1% DMSO, the proteolytic agents trypsin or chymotrypsin, 30 μL of a 3 μM Hsp90α solution for each experiment were used. Hsp90α complexes were obtained by incubating the protein with a 5:1 molar excess of each ligand (7,10 or Novobiocin) at 37 °C for 15 min. Protein and complexes were digested using a 1:100 (w/w) enzyme to substrate ratio. The extent of the proteolysis was monitored on a time-course basis by sampling the reaction mixture at different time intervals (5,15, and 30 min of digestion). Samples were analyzed by MALDI-TOF/MS instrument using a MALDI micro MX (Waters).
Mass data were processed using the Masslynx software (Waters Preferential hydrolysis sites on Hsp90α under different conditions were identified on the basis of the fragments released during enzymatic digestions. Statistical analysis. All the reported data represent the mean ± standard deviation (SD) of at least two independent experiments, performed in triplicate. Where necessary, data were statistically compared by Student's t-test; the statistical significance of DNA content between cells group was examined in the two-way analysis of variance (ANOVA) with Bonferroni post-test analysis using GraphPad Prism 5 software. Differences were considered significant if p < 0.05. 46 was used to prepare the ATP-bound active state of Hsp82, a yeast Hsp90α homologue (PDB code: 2CG9) 37 removing the water molecules that were found 5 Å or more and cap termini were included. Additionally, all hydrogen atoms were added and bond orders were assigned. Chemical structures of investigated compounds were built with Maestro's Build Panel (version 10.2) 46 and subsequently processed with LigPrep (version 3.4) 53 in order to generate all the possible tautomers and protonation states at a pH of 7.4 ± 1.0; the resulting ligands were finally minimized employing the OPLS 2005 force field.

Molecular docking studies. Schrödinger Protein Preparation Wizard workflow
Induced Fit Docking. The Induced Fit Docking protocol is composed of job sequence in which ligands are docked with Glide (first step), then Prime Refinement is used to allow the receptor to relax (second step), and the ligands are redocked into the relaxed receptor with Glide (third step).
Binding site for the first Glide docking phase (Glide Standard Precision Mode) of the Induced Fit Workflow (Induced Fit Docking, protocol 2015-2, Glide version 6.4, Prime version 3.7, Schrödinger) 38-40 is calculated on the 2CG9 structure 37 , mapping onto a grid with dimensions of 36 Å (outer box) and 20 Å (inner box), centered on residues 628-630, 640-641, 670-675 (Hsp90 residues numbering as in the PDB entry 2CG9). Maestro's default protocol was used for the first (Initial Glide docking) and the second step (Prime Induced Fit) considering 20 poses per ligand; these poses were retained from the initial docking and then were passed to Prime (Prime version 3.7, Schrödinger 2015), for the Prime refinement step. Finally, the ligands were re-docked (third step) into their corresponding low energy protein structures (Glide Extra Precision Mode) with resulting complexes ranked according to GlideScore.

Molecular Dynamics Simulations.
The starting structures for the Molecular Dynamics (MD) simulations were prepared with the System Builder in Desmond 42, 43 . Na + ions were added to system the to ensure electroneutrality, and the SPC 54 (simple point charge) water model was used for solvation in a rectangular box with a 10 Å buffer distance, resulting in a system with approximately 178000 atoms. OPLS-2005 force field parameters available in the Schrödinger Suite was used for the entire system (protein and substrates). The MD simulation workflow was run with the default parameters in the Maestro interface to Desmond 42,43 , accounting for a total simulation time of 100 ns, using a recording interval of 1.2 ps, and an ensemble class NPT (300 K and 1.01 bar). Before the simulation, a relaxation of the system was performed using the default equilibration protocol in Desmond 42,43 , with multisim procedure, which is a vital step to prepare a molecular system for production-quality MD simulation. In particular, Maestro's default relaxation protocol was used, which have included two stages of minimization (restrained and unrestrained) followed by four stages of MD runs with gradually diminishing restraints.