Twenty-one natural chalcones, namely chalcone, 2′-hydroxychalcone, 2-hydroxychalcone, 4-hydroxychalcone, 4-methoxychalcone, 3,4-dimethoxychalcone, 4′-hydroxychalcone, 4′-methoxychalcone, 4,4′-dimethoxychalcone, isoliquiritigenin (2′,4,4′-trihydroxychalcone), calomelanone (2′,6′-dihydroxy-4,4′-dimethoxydihydrochalcone), butein (2′,3–4,4′-tetrahydroxychalcone), flavokawain C (2′,4-dihydroxy-4′,6′-dimethoxychalcone), gymnogrammene (2′,6′-dihydroxy-4,4′-dimethoxychalcone), homobutein (2′,4,4′-trihydroxy-3-methoxychalcone), 2,3-dimethoxy-2′-hydroxychalcone, flavokawain A (2′-hydroxy-4,4′,6′-trimethoxychalcone), eriodictyolchalcone (2′,4′,6′,3,4-pentahydroxychalcone) phloretin (2′,4,4′,6′-tetrahydroxydihydrochalcone), phloridzin (phloretin-2′-O-glucoside) and marein (2′,3,3′,4,4′-pentahydroxy-4′-glucosylchalcone) were studied. Table 1 shows their chemical features and substitution patterns. Chalcone was purchased from Fluka (Steinheim, Germany), whereas all remaining chalcones were obtained from Extrasynthese (Genay Cedex, France) (purity >97%).

Human Philadelphia chromosome-positive chronic myelogenous leukemia cell line K562 was purchased from Deutsche Sammlung für Mikroorganismen und Zellkulturen (DSMZ, Braunschweig, Germany) and cultured in RPMI-1640 medium (Lonza, Verviers, Belgium) supplemented with 10% fetal calf serum (Hyclone, Perbio, Erembodegem, Belgium) and 1% (v/v) antibiotic-antimycotic (Lonza, BioWhittaker™), at 37°C, in a 5% CO₂, humidified atmosphere. Human recombinant TNFα (PeproTech, Rocky Hill, NJ, USA) was resuspended in 1× phosphate-buffered saline (PBS) sterile solution containing 0.5% bovine serum albumin (MP Biomedicals, Asse-Relegem, Belgium) to reach a final concentration of 10 μg/ml.

Direct HDAC inhibition by chalcone derivatives was estimated using K562 total extracts as an HDAC source and the enzymatic HDAC activity measurement was performed using a fluorometric HDAC assay kit (Active Motif, Rixensart, Belgium) according to the manufacturer’s instructions. Briefly, after being washed twice with ice-cold 1× PBS, cells were pelleted by centrifugation, and lysed in M-PER^® mammalian protein extraction reagent (Pierce, Erembodegem, Belgium) and supplemented with 1× protease inhibitor cocktail (Roche, Prophac, Luxembourg). The cell suspension was gently mixed on an orbital shaker for 15 min and centrifuged at 14,000 × g at 4°C for 15 min. Protein content was assessed using the Bradford assay (Bio-Rad, Nazareth, Belgium), and 10 μg of proteins were incubated with vehicle or various concentrations of the different chalcones for 1 h at 37°C in the presence of an HDAC fluorometric substrate. Subsequently, the HDAC assay developing solution was added and after 15 min of incubation at room temperature, the fluorescence was measured using a Gemini EM microplate spectrofluorometer (Molecular Devices, Belgium) with excitation at 360 nm and emission at 460 nm. The measured activities were normalized by the vehicle-treated control enzyme activities and IC₅₀ values were calculated.

Transient transfections of K562 cells were performed as previously described (26).

The in vitro growth inhibitory values of chalcone derivatives on the K562 cell line were determined as detailed previously (26).

The 2D structures of chalcone molecules were drawn using SketchEI and transferred into the VEGA ZZ molecular modeling software (27,28) to generate 3D structures. All molecules were saved into a single mol file, that was used as input for the OMEGA, OpenEye Scientific Software (Omega version 2.3.2; http://www.openeye.com) to generate a maximum of 2 low energy conformers with default values. These conformations were stored as OEB file extension format and their 3D similarity was compared using the Rocs, OpenEye Scientific Software (Rocs version 2.3.1; http://www.openeye.com). E-Dragon Software (29) was utilized to calculate constitutional and molecular property descriptors. The descriptors selected to describe the SAR were selected using Partial Least Squares regression as implemented in the PLSR module of Virtual Computational Chemistry Laboratory (29) and Gretl software was used to calculate the correlation between the logarithm of the activity and predicted molecular properties.

The molecular docking was carried out using Glide software (Grid-Based Ligand Docking With Energetics) (Schrodinger Inc., Portland, OR, USA) (30,31) after the docking targets were prepared using Protein Preparation Wizard workflow in Maestro (Schrodinger Inc.) by removing water molecules, adding the hydrogen atoms and assigning all atom force field (OPSL-2005) charges and atom types. The position of all atoms was adjusted by minimizations until the average root mean square deviation reached 0.3 Å. The crystal structures of HDAC8 wild-type and variant D101 complexed with ligands [Protein Data Bank (pdb) entries 1T69 and 3EZT] were used for molecular docking of chalcones into the protein active site. The box encompassing the active site was selected based on the position of co-crystalized ligands complex as described in a previous study (32). The crystal structure of NF-κB complexed to DNA was chosen as a target system to elucidate binding modes of chalcones (pdb entry 1NFK). Prior to docking the DNA molecule was removed and the coordinates of the enclosing box of 30 Å (center at × = −1,1958 Å; y = 9.0149 Å; z = 19,7598 Å) were encompassing the active site residues involved in hydrogen bonds with the NF-κB recognition site of DNA (Arg54, Arg56, Tyr57, Cys59, Lys241, Gln306 and Thr143) (35). Flexible ligand docking simulations were carried out with Glide using the default settings. The ten best poses obtained using the Extra-Precision Glide (Glide XP) mode were selected for analysis. The most favorable poses of molecules showing activity were subjected to further energy minimization using Macromodel 9.1 and OPLS2005 force field.

The effect of chalcone derivatives (nos. 1–21) was examined on total HDAC activity using a fluorescence HDAC assay. As shown in Table II, four chalcone aglycones, namely isoliquiritigenin (no. 10), butein (no. 12), homobutein (no. 15) and the glycoside marein (no. 21), reduced HDAC activity in a concentration-dependent manner (IC₅₀ values 60–190 μM, Fig. 1). Butein (no. 12) appeared to be the best inhibitor of HDAC activity. Other chalcone derivatives were assumed as inactive, because they were unable to provide distinct inhibitory effect even at the highest test concentration (1000 μM).

By using a luciferase-based in cellulo NF-κB reporter assay, chalcones were evaluated for TNFα-induced NF-κB transcription inhibition activity (Table II). Flavokawain C (no. 13) was the most potent NF-κB inhibitor, followed by the dihydrochalcone calomelanone (no. 11) with IC₅₀ values of 8 and 11 μM, respectively (Fig. 2). Chalcones 4-hydroxychalcone (no. 4), 4-methoxychalcone (no. 5), 4′-hydroxychalcone (no. 7), isoliquiritigenin (no. 10), butein (no. 12), homobutein (no. 15) and phloretin (no. 19) also demonstrated good potential with IC₅₀ values ranging between 24–41 μM. Three polymethoxychalcones, i.e. 4,4′-dimethoxychalcone (no. 9), gymnogrammene (no. 14), flavokawain A (no. 17), as well as cone eriodictyolchalcone (no. 18) and two chalcone glycosides, phloridzin (no. 20) and marein (no. 21), which showed no or limited inhibitory activity at 200 μM concentration, were considered as inactive. The remaining compounds, chalcone (no. 1), 2′-hydroxychalcone (no. 2), 2-hydroxychalcone (no. 3), 3,4-dimethoxychalcone (no. 6), 4′-methoxychalcone (no. 8) and 2,3-dimethoxy-2′-hydroxychalcone (16) were cytotoxic at concentrations, which inhibited NF-κB activation.

In order to shed light on the potential mode of action of chalcones, and to understand why some chalcones inhibit either NF-κB or HDACs and some inhibit both, we have carried out molecular modeling, molecular similarity and docking studies. The compounds were docked into the binding sites of HDAC8, and the best studied HDAC enzyme was selected based on the position of the co-crystalized ligand in the crystal structure of the complex (pdb entries 1T69 and 3ETZ) (32). The GlideScore values were compared to the activities that were experimentally obtained (Table III). The results of the docking indicated that all chalcones could favorably bind in the active site, although not all molecules showed activity. The most active molecule 12 had a less favorable GlideScore than chalcone 21 that exhibited the most favorable binding. The binding mode of these two molecules is different (Fig. 3) which may be the result of the large active site of HDAC8 which accommodates two (4-(dimethylamino)-N-[7-(hydroxyamino)-7-oxoheptyl]benzamide) moieties in the interior pocket of the protein. Furthermore, the docking could not distinguish the third active molecule 10 from the rest of the group. There is a clear difference between the GlideScore of the binding of 21 to 20. However, 20 binds almost as good as 12 and better than 10, resulting in the absence of the correlation between activity and binding affinity determined by Glide. We hypothesize that molecules that are not active could possibly bind preferably elsewhere on the protein surface rather than on the active site.

The docking studies of chalcones were also carried out using a crystal structure of NF-κB dimer in complex with duplex DNA. We followed a procedure reported by Piccagli et al(33) and have chosen the DNA recognition surface to define the docking target. The binding site included residues Arg54, Arg56, Tyr57, Cys59, Lys241, Gln306 and Thr143 to gain information on the interaction of our compounds with NF-κB. As shown in Table III, there is a lack of correlation between the activity and the GlideScore results. This can be due to non-specific binding. Moreover, chalcones could act on a different active site of NF-κB. To further rationalize the activity of this class of compounds we have elucidated the SARs and predicted more than 1600 molecular properties for all molecules using EDragon software (29). This analysis was not aimed to lead to the development of predictive models since the data set is small (fifteen molecules with measured NF-κB activity) and thus we did not create training and test sets. Two different sets of molecules were subjected to partial least square regression using PLSR module of the Virtual Computational Chemistry Laboratory to determine which constitutional and molecular properties correlated with the negative logarithm of activity (pNF-κB). The first group consisted of all fifteen tested molecules and the observed correlation was not satisfactory (r²=0.53), leading us to examine the second smaller group, consisting of only nine molecules that displayed activity. The PLS results indicated that a combination of nineteen descriptors correlated with activity (r²=0.99). There were some highly correlated and irrelevant descriptors selected to describe correlation. The selection of descriptors was optimized by developing least square methods using the Gretl software and removing descriptors until a minimal number of descriptors was obtained with a good correlation (r²= 0.914, s=0.250, n=9, F=3.48). The formula used was the following: pNF-κB = −79.78(±48.55) − 0.425(±.0.344)*SS + 31.582(±24.)^*Mp + 43.847(±25.978)^*ARR + 1.868 (±1.619)^*Hy + 0.265(±.468)^*MLOGP + 2.793 (±1.682)^*nBO.

The list and values of descriptors, observed and calculated are shown in Table IV. The statistics indicate reasonable descriptive value of the model that shows which molecular properties influence activity of the molecules in the NF-κB assay. Since we could not develop a satisfactory model that would differentiate the active and inactive molecules, we examined the molecular similarity and differences between most active molecules and the inactive ones. Conformational search carried out by Omega software and the Merck Molecular Force Field force revealed that all molecules can exist in several different conformations due to the free rotation around bonds between carbonyl carbon and neighboring groups. ROCS search and comparison of electrostatic forces showed that unsurprisingly molecules are similar (Shape Tanimoto coefficients are between 0.94 and 0.0.69 and Tverstsky coefficients are between 0.91 and 0.71). The highest similarity was observed between the most active chalcone molecule 13 and inactive 9, indicating that shape and electrostatic properties are not sufficient to explain the different activities of the group.