Thirty-six synthetic chromone derivatives (indicated as compounds 1–36) were assessed for their antioxidant activities by DPPH radical scavenging assay. As shown in Tables 1 and 2, various chromones exhibited different levels of activity, ranging from EC₅₀ = 2.58 to 182.77 μM which are more potent than the well known natural antioxidants, e.g., quercetin and luteolin which possessed IC₅₀ = 10.89 and 11.04 μM, respectively [24]. Structure-radical scavenging activity relationship demonstrated that the 7,8-dihydroxy-2-phenyl-3-benzoyl substituted compounds (compounds 29, 30 and 36) exhibited a strong antioxidant activity with low log EC₅₀. This indicated that dihydroxy substitution (cathecol group) on ring A was essential for radical scavenging activity. The presence of benzoyl group at position 3 confers a high degree of stability toward the phenoxy radicals by participating in electron delocalization and thus is an important feature for potential antiradical property. The proposed model for the progression of successive dehydrogenation from a hydroxyl chromone molecule using adjacent OH-containing aromatic ring is shown in Figure 2. The initial dehydrogenation occurs on the para-OH group. If this is the case, prototropy from an adjacent OH group will be easy. This semiquinone type radical is more reactive than the original phenol molecule, so the second hydrogen liberation proceeds rapidly, thus resulting in biradical changes into a quinone.

The MFA model of 35 chromone derivatives (30 compounds in a training set; 5 compounds in a test set) was developed using field fit alignment. The most active compound, 7,8-dihydroxy-2-(4′-trifluoromethylphenyl)-3-(4″-trifluoromethylbenzoyl)chromone 29 was used as a template model for superimposing the rest of the molecules. Superimposition of the aligned molecules is shown in Figure 3.

The steric (CH₃) and electrostatic (H⁺) descriptors in the MFA-QSAR equations specify the regions where variations in the structural features (steric or electrostatic) of different compounds in the training set, leading to either an increase or a decrease in activities. The steric descriptor with positive or negative coefficients shows a region where bulky substituent is favored or disfavored, respectively. The electrostatic descriptor with a positive coefficient indicates a region favorable for electropositive group, while a negative coefficient indicates that an electronegative (electron-withdrawing) group is required at the position. The numbers accompanying descriptors in the equations represent their positions in the three-dimensional MFA grid (Figure 4). The MFA-QSAR equation is expressed as follow: (1) Activity = 2.86587 - 0.021102   ( H + / 314 ) - 0.004635   ( Vm ) + 0.010606   ( H + / 336 ) - 0.01133 ( CH 3 / 663 ) - 0.009739 ( CH 3 / 670 )n = 30, r = 0.932, r² = 0.868, r²_cv = 0.771, PRESS = 1.022, r²_BS = 0.857, LSE = 0.02, N = 5, r²_pred = 0.924

The presence of two steric descriptors (CH₃/663) and (CH₃/670) with negative coefficients, indicates that bulky substituents are disfavored. The presence of electrostatic descriptor (H⁺/336) with a positive coefficient on phenyl ring suggests that electropositive groups are favored, while (H⁺/314) with negative coefficients indicates that electronegative groups should be substituted on benzoyl ring. Appearance of descriptor (Vm) with a negative coefficient demonstrates that larger molecule decrease activity. Figure 4 illustrates the regions around the molecule (compound 29) corresponding to the MFA model. This proposed model can be accounted for the lowest activity of compound 31 (EC₅₀ = 182.77) which possessed no electronegative group in region 3 and electropositive group in region 4.

A QSAR equation is generally acceptable if the correlation coefficient (r) is approximately 0.9 or higher. The r value is a relative measure of the quality of fit of the model. Its value depends on the overall variance of the data. An r²_cv, a squared correlation coefficient generated during a cross-validation procedure, is used as a diagnostic tool to evaluate the predictive power of an equation. Cross-validation is often used to determine how large a model (number of terms) can be used for a given data set. Equation 1 explains 86.8% variance in the activity with respect to the steric and electrostatic fields and molecular volume while leave-one-out cross-validation power of prediction was found to be 77.1%. An r²_BS value of 0.857 is an average squared correlation coefficient calculated during the validation procedure. The predictive power of the model was calculated by using the following equation (2) r 2 pred = ( SD − PRESS ) SDwhere SD is the sum of the squared deviations between the biological activities of each molecules and the mean activity of the training set of molecules and PRESS is the sum of squared deviations between the predicted and actual activity values for every molecule in the test set.

The calculated activity obtained from equation 1 and actual activity of the training set and test set molecules are summarized in Table 1 and 2. Scattered plots of calculated and actual activities and the plot of residuals for the training set and the test set molecules are shown in Figure 5 and 6, respectively. Most of the molecules showed residual values less than 0.2. The outlier molecule was 7,8-dihydroxy-2-(4′-nitrophenyl)-3-(4″-nitrobenzoyl)chromone 36 (residual = 0.947). The obtained MFA model shows good statistical results with r²_cv = 0.771, conventional r² = 0.868 and r²_pred = 0.924.

Chromone derivatives were synthesized by one-pot cyclization reaction with 1,8-diazabicyclo[5,4,0]undec-7-ene (DBU) as catalyst [25]. The antioxidant activities of the synthesized compounds were assessed on the basis of the radical scavenging effect on the DPPH free radicals as described previously [16]. The concentrations of test samples required to scavenge 50% of DPPH free radicals (EC₅₀ μM) were converted into corresponding log EC₅₀ values.

The molecular structures of chromone derivatives were modeled with SYBYL 7.0 molecular modeling program (Tripos Associates, Saint Louis, MO) on an Indigo Elan workstation (Silicon Graphics Inc., Mountain View, CA) using the sketch approach. The fragment libraries in SYBYL database were used as building blocks for construction of larger images. Firstly, each structure was energy minimized using the standard Tripos force field (Powell method and 0.05 kcal/mol.Å energy gradient convergence criteria) and electrostatic charge was assigned by the Gasteiger-Hückel method. Further, geometry optimization was then carried out with the MOPAC 6 package using the semi-empirical PM3 with Gasteiger-Hückel for charges calculation. The SMILESes forms of all structures are shown in Table 3.

The field fit alignment method was used for MFA. All molecules were submitted to the CONFORMER SEARCH module within Cerius² to generate 150 conformers of each molecule using Boltzman jump method [26]. The lowest energy conformer of each molecule was selected. All the selected conformers were aligned using field fit alignment method in the QSAR module. The most active compound, 7,8-dihydroxy-2-(4′-trifluoromethylphenyl)-3-(4″-trifluoromethylbenzoyl)chromone 29, was used as a template model for superimposing the rest of the molecules.

MFA studies were performed with the QSAR module of Cerius². The molecular field was created using CH₃ and H⁺ as probes representing steric and electrostatic fields, respectively. The steric and electrostatic fields were sampled at each point of regularly spaced grid of 2 Å. In addition, numerous spatial and structural descriptors such as polarizability, dipole moment, radius of gyration, molecular area, molecular dimension, density, principal moment of inertia, molecular volume, molecular weight, number of rotatable bonds, hydrogen bond donors and acceptors, log P, molar refractivity and others were also calculated and considered as independent variables. Only 10% of the total descriptors with the highest variance were considered for further analysis. Regression analysis was carried out using genetic partial least squares (G/PLS) method consisting of 5000 generations with a population size of 100. The optimum number of components was set to 4 based on better r² and r²_cv values for a given training set. An energy cutoff of ± 30.0 kcal/mol was set for both steric and electrostatic contributions. The smoothing parameter, d, was set to 1.0 to control the bias in the scoring factors between equations with different number of terms. The length of the final equation was fixed to five descriptors. The linear option was used in the equation creation. Cross validation was performed with the leave-one-out procedure. The PLS analysis was set to no scale.