Medicinal Chemistry Research

, Volume 19, Issue 8, pp 864–901

Quantitative structure–activity relationships for estimating the aryl hydrocarbon receptor binding affinities of resveratrol derivatives and the antioxidant activities of hydroxystilbenes

Authors

    • Department of ChemistryUniversity of Winnipeg
    • Ecologica Research
  • Charles D. Goss
    • Department of ChemistryUniversity of Winnipeg
  • Kaya Forest
    • Department of ChemistryOkanagan College
  • Ken J. Friesen
    • Department of ChemistryUniversity of Winnipeg
Original Research

DOI: 10.1007/s00044-009-9236-2

Cite this article as:
Rayne, S., Goss, C.D., Forest, K. et al. Med Chem Res (2010) 19: 864. doi:10.1007/s00044-009-9236-2

Abstract

Quantitative structure–activity relationships (QSARs) were developed for the aryl hydrocarbon receptor (AhR) binding affinity of non-fluorinated and fluorinated cis and trans 3,4′,5-substituted resveratrol derivatives. Lower quality QSAR fits were found when all compounds were modeled together, in contrast to strong correlations with Hammett substituent constants and atomic charges for separate non-fluorinated/fluorinated and cis/trans groupings. The collective findings suggest little promise in developing new resveratrol derivatives with significantly higher AhR binding affinity beyond the range already mapped by the available experimental data sets. The results also suggest that future QSAR searches for AhR binding affinity among resveratrol derivatives will likely need to proceed in a more constrained class-by-class fashion owing to the potentially different mechanistic implications from changes in the types/positions of aromatic substitution and the cis/trans geometrical isomerism. The antioxidant activity of hydroxystilbenes can be accurately modeled using high-quality empirical univariate relationships with various molecular descriptors. In comparison, correlations between theoretical bond dissociation enthalpies and ionization energies of the molecular and dissociated forms yielded lower quality QSAR fits with the antioxidant behavior.

Keywords

StilbenesResveratrol derivativesAryl hydrocarbon receptorBinding affinityHydroxystilbenesAntioxidant activityQuantitative structure–activity relationships

Introduction

Stilbenes are diarylethene compounds (Fig. 1) produced in plants via the metabolism of phenylalanine into cinnamic acid, which is then polymerized by the enzyme stilbene synthase via the Shikimic pathway into the first member of the series, trans-resveratrol (trans-1-(4′-hydroxyphenyl)-2-(3,5-dihydroxyphenyl)-ethene; compound 1) (Langcake and Pryce, 1976; Langcake, 1981; Aggarwal et al., 2004). Trans-resveratrol and its derivatives exist in a variety of food sources, including grapes, wines, berries, nuts, and herbal plants (Baur and Sinclair, 2006), and agricultural wastes, such as grapevine prunings (Rayne et al., 2008; Karacabey and Mazza, 2008). Given the inherent complexities, wastes involved, and lack of green chemistry approaches regarding conventional laboratory methods for synthesizing stilbenes (Guiso et al., 2002; Jeffery and Ferber, 2003; Ferre-Filmon et al., 2004), there also is interest in obtaining this compound directly from renewable sources and using it as a platform molecule in medicinal chemistry (Rayne, 2008). In addition to trans-resveratrol, a large number of other stilbenes have been reported in plant extracts, and since 1995, more than 400 new naturally occurring stilbenes were isolated and identified (Shen et al., 2009).
https://static-content.springer.com/image/art%3A10.1007%2Fs00044-009-9236-2/MediaObjects/44_2009_9236_Fig1_HTML.gif
Fig. 1

General structures and numbering system for stilbenes

However, trans-resveratrol remains perhaps the most high-profile member of this natural product class, particularly due to its known high bioactivity for a variety of endpoints (Wu et al., 2001; Savouret and Quesne, 2002; de la Lastra and Villegas, 2005; Baur and Sinclair, 2006; Athar et al., 2007; Pirola and Frojdo, 2008; Guerrero et al., 2009). In particular, trans-resveratrol and its derivatives are known to compete with 2,3,7,8-tetrachlorodibenzo[1,4]dioxin, benzo[a]pyrene, and other environmental contaminants for binding to the aryl hydrocarbon receptor (AhR), thereby preventing subsequent toxic effects from this biomolecular interaction (Ciolino et al., 1998; Casper et al., 1999; Ciolino and Yeh, 1999; Singh et al., 2000; Revel et al., 2001; Revel et al., 2003). Furthermore, a number of studies have investigated several endpoints regarding the antioxidant activities for various hydroxystilbenes (Murias et al., 2005; de la Lastra and Villegas, 2007; Fang and Zhou, 2008). As well as the direct bioactivities of trans-resveratrol in vivo, this compound can be rapidly metabolized (Joseph et al., 2008; Remsberg et al., 2008) by cytochrome P450 enzymes into higher hydroxylated analogues (Potter et al., 2002; Piver et al., 2004), oligomerized, converted into sulfate and glucuronic acid derivatives, or undergo hydrogenation of the aliphatic double bond (Walle et al., 2004). Consequently, there is a need to understand better the potential range and associated bioactivities of trans-resveratrol derivatives produced in vivo (Pirola and Frojdo, 2008), as well as related compounds that can be produced by synthetic methods. To help answer such questions, the present study sought to investigate and develop quantitative structure–activity relationships (QSARs) for predicting the AhR binding affinity and antioxidant activity of trans-resveratrol, its derivatives, and related stilbene compounds.

Materials and methods

AhR binding affinity QSAR development

Hammett substituent constants and resonance and field parameters were taken from Hansch et al. (1991). Two-dimensional molecular structures from de Medina et al. (2005) were drawn in ACD/ChemSketch v. 11.02 (Build 25941, 21 May 2008; Advanced Chemistry Development, Inc., Toronto, ON, Canada) and exported in SMILES file format (Weininger, 1988; Weininger et al., 1989). The two-dimensional structures were converted to three-dimensional structures and conformationally minimized with the Frog online software program (http://bioserv.rpbs.jussieu.fr/cgi-bin/Frog; Bohme Leite et al., 2007) using the following parameters: output = mol2; produce = single; #conf = 50; Emax = 100; #mc steps = 1000. The output geometries from the conformer minimization efforts were converted from Sybyl Mol2 format to MOPAC internal format using Open Babel v. 2.1.1 (http://openbabel.org/wiki/Main_Page) and used as inputs for gas and aqueous phase semi-empirical PM6 (Stewart, 2007) geometry optimizations in MOPAC 2009 (v. 9.045; James J. P. Stewart, Stewart Computational Chemistry).

Gas phase geometry optimizations in MOPAC 2009 were conducted with the following keywords in the input file header: PM6 BONDS CHARGE = 0 SINGLET LET GNORM = 0 CYCLES = 10000 GRAPHF. Aqueous phase geometry optimizations in MOPAC 2009 with the COSMO solvation model (Klamt and Schuumann, 1993) were conducted with the following keywords in the input file header: PM6 RSOLV = 1.3 EPS = 80.1 BONDS CHARGE = 0 SINGLET LET GNORM = 0 CYCLES = 10000 GRAPHF. PM6 optimized geometries were visualized using Jmol v. 11.6 (http://www.jmol.org/; Herraez, 2006; McMahon and Hanson, 2008). The gas and aqueous phase PM6 calculations yielded the following three-dimensional molecular properties, which were included in the QSAR development approach: standard state enthalpy of formation; total energy, electronic energy; core-core repulsion energy; Connolly molecular area (aqueous phase only); Connolly molecular volume (aqueous phase only); dipole; ionization energy; energies of the highest occupied molecular orbital (EHOMO) and lowest unoccupied molecular orbital (ELUMO) and the energy difference between the EHOMO and ELUMO; and partial Mulliken charges on all carbon atoms.

Molecular structures in SMILES (Weininger, 1988; Weininger et al., 1989) format were input to the E-DRAGON 1.0 software program (Dragon v.5.4 [28 March 2006]; http://www.vcclab.org/lab/edragon/; Tetko, 2005; Tetko et al., 2005). For each congener, 48 constitutional descriptors, 119 topological descriptors, 47 walk and path counts, 33 connectivity indices, 47 information indices, 96 two-dimensional autocorrelations, 107 edge adjacency indices, 64 Burden eigen values, 21 topological charge indices, 44 eigen value-based indices, 41 Randic molecular profiles, 74 geometrical descriptors, 150 RDF descriptors, 160 three-dimensional MoRSE descriptors, 99 WHIM descriptors, 197 GETAWAY descriptors, 154 functional group counts, 120 atom centered fragments, 14 charge descriptors, and 31 molecular properties were generated. The SPARC software program (http://ibmlc2.chem.uga.edu/sparc/; August 2007 release w4.0.1219-s4.0.1219) was used to estimate octanol-water partitioning constants (log P) (Hilal et al., 2007) using SMILES input structures.

Unsupervised forward selection (UFS; http://www.vcclab.org/lab/ufs/start.html) was used to produce the following reduced descriptor data set for pKi, which contained maximal linearly independent sets of descriptor columns with a minimal amount of multiple correlation (Whitley et al., 2000) (r-values against pKi in parentheses): RCON (−0.65); Ke (0.64); E2m (0.64); P1s (0.63); P1m (0.61); Mor26v (−0.60); P2m (−0.59); E1m (0.58); H1e (−0.58); Mor28p (0.56); L1s (0.56); RDF055v (−0.55); and RDF045e (−0.54). Variable acronym definitions are available in the E-DRAGON for VCCLAB User Manual (http://michem.disat.unimib.it/chm/Help/edragon/index.html). Stepwise forward multiple linear regression of the reduced data sets using Fin/Fout criteria of 0.2 and 0.1, respectively (Bendel and Afifi, 1977) was conducted with KyPlot (v.2.b.15; Dr. K. Yoshioka, Tokyo Medical and Dental University, Tokyo, Japan) against the experimental pKi values to produce the final QSAR model. QSAR model development was performed consistent with the guidelines set forward by Dearden et al. (2009).

Antioxidant activity QSAR development

Molecular structures in SMILES (Weininger, 1988; Weininger et al., 1989) format were input to the E-DRAGON 1.0 software program (Dragon v.5.4 [28 March 2006]; http://www.vcclab.org/lab/edragon/; Tetko, 2005; Tetko et al., 2005). Equivalent molecular descriptors were generated as described in the “AhR binding affinity QSAR development” section. The SPARC software program (http://ibmlc2.chem.uga.edu/sparc/; August 2007 release w4.0.1219-s4.0.1219) was used to estimate acidity constants (pKa values) for all phenolic groups (Hilal et al., 1995, 2007) using SMILES input structures.

Two-dimensional molecular structures were drawn in ACD/ChemSketch v. 12.01 (Build 30815, 10 February 2009; Advanced Chemistry Development, Inc., Toronto, ON, Canada), converted to three-dimensional structures using the 3D structure optimization tool, and exported as MDL mol files (v2000). The resulting structures were subsequently converted from MDL mol format to MOPAC internal format using Open Babel v. 2.2.0 (http://openbabel.org/wiki/Main_Page). Gas phase semiempirical PM6 (Stewart, 2007) calculations were performed using MOPAC 2009 (v. 9.126; James J. P. Stewart, Stewart Computational Chemistry, Colorado Springs, CO, USA, http://OpenMOPAC.net). Gas phase PM6 calculations on the molecular phenols were performed with the following keywords in the input file header: PM6 BONDS CHARGE = 0 LET DDMIN = 0.0 GNORM = 0.0 XYZ. Gas phase PM6 calculations on the phenoxyl radicals were performed with the following keywords in the input file header: PM6 UHF BONDS CHARGE = 0 LET DDMIN = 0.0 GNORM = 0.0 XYZ. Gas phase PM6 calculations on the phenolate anions were performed with the following keywords in the input file header: PM6 BONDS CHARGE = -1 LET DDMIN = 0.0 GNORM = 0.0 XYZ. PM6 calculations were also conducted using benzene as a model solvent (ε/εo = 2.3, solvent radius = 2.60 Å; Benitez and Goddard, 2005) with the COSMO solvent module (Klamt and Schuumann, 1993) in MOPAC 2009, resulting in the addition of the following keywords to the respective phenol/phenoxyl radical input file headers given above: RSOLV = 2.6 EPS = 2.3.

Density functional theory (DFT) calculations were conducted using Gaussian 03 (Frisch et al., 2004) with the high-performance computing resources on the Western Canada Research Grid (WestGrid; www.westgrid.ca/; project #100185; K. Forest). Semiempirical PM6 gas phase geometry optimized structures were used as the inputs for DFT calculations. Gas phase optimizations and frequency calculations were performed using the B3LYP hybrid functional (Lee et al., 1988; Becke, 1993) and the 6-31++G(d,p) and 6-311++G(d,p) basis sets (Hariharan and Pople, 1972; Hehre et al., 1972; Rassolov et al., 2001). Linear regression analyses were conducted with KyPlot (v.2.b.15; Dr. K. Yoshioka, Tokyo Medical and Dental University, Tokyo, Japan) at a significance level of 0.05 and a confidence level of 0.95.

Results and discussion

Aryl hydrocarbon receptor binding affinities of resveratrol derivatives

Quantitative structure-activity relationships (QSARs) were developed for the aryl hydrocarbon receptor (AhR) binding affinity of a series of trans- and cis-substituted resveratrol derivatives (compounds 1 through 24 in Table 1). Using all available E-DRAGON, PM6, and SPARC molecular descriptors, training of the pKi QSAR model via stepwise forward linear regression of the UFS reduced data set gave the following two variable predictive Eq. 1,
$$ \begin{aligned} {\text{pK}}_{\text{i}} \left( {\text{nM}} \right) & = - 3.0 1 6 \left( { \pm 0. 3 1 9; \pm {\text{SE }}\left[ {\text{standard error}} \right]; p < 10^{ - 8} } \right) \\& \quad + 1. 7 7 2 \left( { \pm 0. 4 6 3; p < 0.00 1} \right) \times Ke + 1. 9 4 9\hfill \\ & \quad \quad \left( { \pm 0. 5 2 8; p = 0.00 1} \right) \times {\text{E2m}} \hfill \\ \end{aligned} $$
(1)
$$ n = 24; r^{{ 2 }} = 0.648; s = 0.370; F_{{2,21}} = 19.4; p < 0.0001$$
in which Ke is the K global shape index weighted by atomic Sanderson electronegativities and E2m is the second component accessibility directional WHIM index weighted by atomic masses (Fig. 2). Multicollinearity was not present among the final variables (Dillon and Goldstein condition number < 30) with the corresponding partial correlation matrix containing all r values <|0.25| between the independent descriptors. The QSAR statistical quality of fit included an r value of 0.805 (r2 = 0.648; \( r_{\rm adj}^{2} = 0.615 \)), a standard error of 0.370, a coefficient of variation of −0.323, a predicted residual sum of squares of 3.83, an Akaike’s information criterion of 25.1, and p(Fcalc = 19.4 > F0.05 = 3.5) < 10−4. No curvature was observed in the residuals plot against predicted pKi values (Fig. 2 inset [lower right]; pm=0 = 1, pb≠0 = 1), but the residuals plot against experimental pKi values has both a linear regression y-intercept and slope not equal to zero (Fig. 2 inset [upper left]; pm=0 = 0.002, pb≠0 < 0.006). The variation inflation factor (VIF; VIF = 1/(1 − r2), in which r is the correlation coefficient of multiple regression between one independent variable and others in the equation; VIF = 1 indicates no self-correlation, 1 < VIF < 5 is acceptable, and VIF > 10 indicates unstable regression (Wei et al., 2001)) was 1.1, indicating an acceptable level of self-correlation in the model.
Table 1

Compound numbering system, geometrical isomerism, and aryl substitution patterns for the stilbene derivatives under consideration

ID

Geometrical isomerism

Substitution

2

3

4

5

6

2′

3′

4′

5′

6′

1

trans

–H

–OH

–H

–OH

–H

–H

–H

–OH

–H

–H

2

trans

–H

–OCH3

–H

–OCH3

–H

–H

–H

–OCH3

–H

–H

3

trans

–H

–Cl

–H

–Cl

–H

–H

–H

–Cl

–H

–H

4

trans

–H

–F

–H

–F

–H

–H

–H

–OCH3

–H

–H

5

trans

–H

–F

–H

–F

–H

–H

–H

–F

–H

–H

6

trans

–H

–CF3

–H

–CF3

–H

–H

–H

–CF3

–H

–H

7

trans

–H

–OCH3

–H

–OCH3

–H

–H

–H

–F

–H

–H

8

trans

–H

–OCH3

–H

–OCH3

–H

–H

–H

–OEt

–H

–H

9

trans

–H

–OCH3

–H

–OCH3

–H

–H

–H

–OBun

–H

–H

10

trans

–H

–Cl

–H

–Cl

–H

–H

–H

–OCH3

–H

–H

11

trans

–H

–Cl

–H

–Cl

–H

–H

–CF3

–H

–H

–H

12

trans

–H

–Cl

–H

–Cl

–H

–H

–OCH3

–H

–H

–H

13

cis

–H

–OCH3

–H

–OCH3

–H

–H

–H

–OCH3

–H

–H

14

cis

–H

–Cl

–H

–Cl

–H

–H

–H

–Cl

–H

–H

15

cis

–H

–F

–H

–F

–H

–H

–H

–OCH3

–H

–H

16

cis

–H

–F

–H

–F

–H

–H

–H

–F

–H

–H

17

cis

–H

–CF3

–H

–CF3

–H

–H

–H

–CF3

–H

–H

18

cis

–H

–OCH3

–H

–OCH3

–H

–H

–H

–F

–H

–H

19

cis

–H

–OCH3

–H

–OCH3

–H

–H

–H

–OEt

–H

–H

20

cis

–H

–OCH3

–H

–OCH3

–H

–H

–H

–OBun

–H

–H

21

cis

–H

–Cl

–H

–Cl

–H

–H

–H

–CF3

–H

–H

22

cis

–H

–Cl

–H

–Cl

–H

–H

–H

–OCH3

–H

–H

23

cis

–H

–Cl

–H

–Cl

–H

–H

–CF3

–H

–H

–H

24

cis

–H

–Cl

–H

–Cl

–H

–H

–OCH3

–H

–H

–H

25

trans

–H

–OH

–H

–OH

–H

–H

–H

–H

–H

–H

26

trans

–H

–OH

–H

–OH

–H

–H

–OH

–H

–OH

–H

27

trans

–H

–OH

–OH

–H

–H

–H

–OH

–H

–OH

–H

28

trans

–H

–OH

–OH

–OH

–H

–H

–H

–OH

–H

–H

29

trans

–H

–OH

–OH

–OH

–H

–H

–OH

–H

–OH

–H

30

cis

–H

–OH

–H

–OH

–H

–H

–H

–OH

–H

–H

31

trans

–H

–OH

–H

–OH

–H

–H

–H

–OCH3

–H

–H

32

trans

–H

–OCH3

–H

–OCH3

–H

–H

–H

–OH

–H

–H

33

trans

–OH

–H

–OH

–H

–H

–H

–H

–H

–H

–H

34

cis

–H

–OCH3

–H

–OCH3

–H

–H

–H

–OH

–H

–H

35

trans

–H

–OCH3

–H

–OCH3

–H

–H

–OCH3

–OH

–H

–H

36

cis

–H

–OCH3

–H

–OCH3

–H

–H

–OCH3

–OH

–H

–H

37

trans

–H

–OCH3

–H

–OCH3

–H

–H

–OH

–OH

–H

–H

38

cis

–H

–OCH3

–H

–OCH3

–H

–H

–OH

–OH

–H

–H

39

trans

–H

–OCH3

–H

–OCH3

–H

–H

–But

–OH

–But

–H

40

cis

–H

–OCH3

–H

–OCH3

–H

–H

–But

–OH

–But

–H

41

trans

–H

–OCH3

–H

–OCH3

–H

–H

–But

–OH

–OH

–H

42

cis

–H

–OCH3

–H

–OCH3

–H

–H

–But

–OH

–OH

–H

43

trans

–H

–OH

–OH

–OH

–H

–H

–OH

–OH

–OH

–H

44

trans

–H

–OH

–OH

–H

–H

–H

–H

–OH

–H

–H

45

trans

–H

–H

–OH

–H

–H

–H

–H

–OH

–H

–H

46

trans

–H

–OH

–OH

–H

–H

–H

–H

–H

–H

–H

47

trans

–H

–OH

–OH

–H

–H

–H

–OH

–OH

–H

–H

48

trans

–OH

–H

–OH

–H

–H

–H

–H

–OH

–H

–H

49

trans

–H

–OCH3

–OH

–H

–H

–H

–OCH3

–OH

–H

–H

50

trans

–H

–OH

–OH

–OH

–H

–H

–OH

–OH

–H

–H

51

trans

–H

–OH

–OH

–OH

–H

–H

–OH

–OH

–Br

–H

52

trans

–H

–OH

–H

–(3,4,5-trihydroxybenzoyl)

–H

–H

–H

–OH

–H

–H

53

trans

–H

–(3,4,5-trihydroxybenzoyl)

–H

–(3,4,5-trihydroxybenzoyl)

–H

–H

–H

–OH

–H

–H

54

trans

–H

–OH

–H

–(3,4–dihydroxycinnamoyl)

–H

–H

–H

–OH

–H

–H

55

trans

–H

–OH

–H

–(3,4–trihydroxycinnamoyl)

–H

–H

–H

–OH

–H

–H

56

trans

–H

–H

–OH

–H

–H

–H

–H

–H

–H

–H

57

trans

–H

–OH

–OH

–OH

–H

–H

–H

–H

–H

–H

https://static-content.springer.com/image/art%3A10.1007%2Fs00044-009-9236-2/MediaObjects/44_2009_9236_Fig2_HTML.gif
Fig. 2

Comparison between predicted (y-axis) and experimental (x-axis) AhR binding affinity pKi values using the QSAR regression equation given in the text on compounds 1 through 24. A 1:1 line (dashed) and a linear regression (solid line) of the form pKi,pred = 0.65 (±0.10; ±SE) × pKi,expt – 0.40 (±0.13) (r = 0.805, pm=0 < 10−5, pb=0 = 0.006) are shown. Insets show plots of residual pKi,pred prediction errors over the range of pKi,pred and pKi,expt values

The QSAR was limited to two independent variables (22 = 4), even though three (23 = 8), four (24 = 16), and possibly five (25 = 32) variables could have been used without exceeding overfitting criteria (2N < n; where N is the number of independent variables and n is the size of training sample data set [n = 24]). Increasing the number of variables from two to three (H1e was the third chosen variable using stepwise regression) only improved the r2 of the QSAR by 0.02 (Fin = 1.2 > Fin,crit = 0.20) compared with a Δr2 of 0.23 for N = 1 → N = 2 (Fin = 13.6). Moreover, H1e did not display a significant regression coefficient (p = 0.28), indicating that its addition to the QSAR would constitute overfitting. Further inclusion of RDF055v as a fourth variable in the QSAR (Δr2 of 0.015 for N = 3 → N = 4 [Fin = 0.90]) also did not significantly improve the quality of fit with an associated nonsignificant regression coefficient (p = 0.35), and introduced multicollinearity among the variables (Dillon and Goldstein condition number > 30). The pKi QSAR model was validated using both the leave-one-out and N-fold (divide-in-half) cross-validation approaches for the training set compounds (Dogra, 2009). Reasonable agreement was observed between the experimental and predicted pKi values for all validation combinations, with low average signed and unsigned prediction errors, respectively, for the leave-one-out (0.00 and 0.29) and two alternate divide-in-half (0.03 and 0.20/0.02 and 0.33) validations and a cross-validated r2 value, q2, of 0.570 and a \( q_{\text{adj}}^{ 2} \) of 0.560 (Fig. 3).
https://static-content.springer.com/image/art%3A10.1007%2Fs00044-009-9236-2/MediaObjects/44_2009_9236_Fig3_HTML.gif
Fig. 3

Comparison between predicted (y-axis) and experimental (x-axis) AhR binding affinity pKi values towards during the leave-one-out (open circles) and two alternate divide-in-half (open squares and open triangles, respectively) cross-validation exercises for the QSAR regression equation given in the text on compounds 1 through 24. A 1:1 line (dashed) is shown

In comparison with our QSAR, Tripathi and Saxena (2008) used hydrophobic (π1 and π2), electronic (σ1 and σ2), steric (MR1 and MR2), resonance (RR1 and RR2), and field effect (f1 and f2) parameters to develop a suite of two parameter QSARs on the same data set. Using an optimized equation with π2 and I (a binary indicator parameter with values of 1 for trans and 0 for cis configurations about the ethene function), a QSAR with an r value of 0.881 and standard error of 0.279 was obtained by these researchers. These results are moderately better than our corresponding r value of 0.805 and standard error of 0.370. Collectively, our results and those of Tripathi and Saxena (2008) have now investigated the entire effective domain of molecular descriptors available for QSAR modeling using a variety of two- and three-dimensional descriptors, and neither approach yields a QSAR with satisfactorily high predictive power for estimating the AhR binding affinity of resveratrol derivatives.

To understand better the underlying difficulties in developing a AhR binding affinity QSAR for this suite of compounds, we divided the data set into subclasses of both trans- and cis-substituted and fluorinated (i.e., −F and −CF3 substituents) and nonfluorinated 3,4′,5-substituted derivatives. Our further analysis of this data set suggested that separate reliable predictive relationships can be developed between the sum (Σ) of the Hammett substituent constants (σ values) on both aryl rings (Σσr1+r2; r1 contains substituents numbered 2 through 6, r2 contains substituents numbered 2′ through 6′) and the pKi data for nonfluorinated trans-resveratrol derivatives (compounds 13 and 810; Fig. 4) and their cis-substituted counterparts (compounds 13, 14, 19, 20, and 22) having 3,4′,5-substitution patterns. Both relationships suggest that increasing electron withdrawing substituents on the stilbene moiety increase the pKi values, but that no significant increases in pKi will occur with Σσr1+r2 > 0.4. The biochemical interpretation of these fits suggest that more electron withdrawing substituents on the trans-stilbene framework (i.e., increasing Σσr1+r2) increases the AhR binding affinity pKi, with a maximum potential pKi of approximately −0.1 (0.8 nM). Electron donating substituents (e.g., para-alkoxy groups) greatly decrease the pKi, and if the net effect of all stilbene substituents is electron donating (Σσr1+r2 < 0), the pKi rapidly declines. trans-Stilbene compounds with a set of substituents having a composite net electron donating character higher than that of 1 (i.e., Σσr1+r2 < −0.15) would be expected to have a negligible pKi. The findings also suggest that the 3,4′,5-trichlorinated trans-stilbene derivative 3 (pKi = −0.08) synthesized and evaluated by de Medina et al. (2005) is at the maximum possible pKi value. Thus, additional efforts on synthesizing more potent AhR antagonists using replacements of the phenolic groups on 1 with other more electron withdrawing moieties (e.g., Br, NO2, CCl3, CN, etc.) may not prove worthwhile if the Hammett-based structure–activity model holds for Σσr1+r2 > 1.0 (the current limit of the literature data set).
https://static-content.springer.com/image/art%3A10.1007%2Fs00044-009-9236-2/MediaObjects/44_2009_9236_Fig4_HTML.gif
Fig. 4

Relationships between the sum of the Hammett substituent constants on both aryl rings (Σσr1+r2) and the experimental AhR binding affinity pKi values for (a) nonfluorinated and (b) fluorinated trans- (circles) and cis- (squares) 3,4’,5-substituted resveratrol derivatives

By comparison, the nonfluorinated cis-stilbene 3,4′,5-substituted derivatives display a lower apparent maximum potential pKi of approximately −1.1 (12 nM). When the fluorinated (trans-substituted, 47; cis-substituted, 1518 and 21) and 3,3′,5-substituted (11, 12, 23, and 24) derivatives are analyzed (Fig. 4b), there are no coherent relationships between the Σσr1+r2 and pKi data between or within the trans-/cis-groupings. Tripathi and Saxena (2008) also found a lower predictive ability for the fluorinated derivatives in their combined modeling efforts on developing 2D-QSAR models for all 24 compounds, but did not probe separate QSARs for cis/trans and/or non-fluorinated/fluorinated subclasses. In our model, all fluorinated derivatives have weaker than expected pKi values based on the fit for the corresponding nonfluorinated analogs. AhR binding affinity pKi correlations also were considered with ΣσF and ΣσR (for individual rings and the sums), Σσr1 and Σσr2, and using σ+ and σ values. However, the quality of fits was generally poor, and no fits among the compound subgroupings were comparable to the Σσr1+r2 analysis presented in Fig. 4.

Although Σσr1+r2 values appear to control the pKi values for both trans- and cis-substituted nonfluorinated stilbene derivatives, the rationale for the lower bioactivity of the cis-derivatives seems to be related to the molecular size and the ability of the stilbene group to achieve planarity. It is believed that the AhR ligand binding site accommodates only planar ligands with maximum dimensions of 14 Å × 12 Å × 5 Å (Waller and McKinney, 1995). In contrast to the claims of de Medina et al. (2005), not all of compounds 1 through 24 appear to fall within the AhR binding site size domain. Both gas and aqueous phase semiempirical PM6 calculations suggest that the cis-substituted compounds 13, 1620, and 2224 have molecular “heights” >5 Å (5.3, 5.2, 5.3, 5.7, 7.2, 6.4, 5.5, 6.2, and 6.1 Å, respectively). We also note that para-ethoxy and para-n-butoxy substituted trans-derivatives 8 and 9, respectively, also have estimated molecular “lengths” >14 Å (15.4/15.6 Å and 17.9/16.6 Å in the gas/aqueous phases, respectively). Similarly, the molecular modeling shows that all trans-substituted derivatives can achieve stilbene function planarity, although both gas and aqueous phase models show that the planar conformation in these compounds is not always the lowest energy. Molecular dynamics and docking studies would be needed to understand better the potential energetics behind the interaction of planar trans-substituted stilbenes binding with the AhR. In contrast, the computational results show that none of the cis-substituted derivatives can achieve planarity without a significant degree of molecular strain.

For the fluorinated 3,4′,5-substituted trans- and cis-derivatives, the only reliable correlation we found was between the gas phase charge on the stilbene carbon adjacent to ring 2 (qstilbene C, ring 2) and the pKi values for the trans-substituted derivatives 47 (Fig. 5). As may be expected, the Σσr2 also displayed a strong correlation with pKi values for this subset, demonstrating the linkage between the empirically derived Hammett substituent constant and the semiempirical PM6 calculated substituent effects on charge distributions within these molecules. The aqueous phase qstilbene C, ring 2 values had a poorer correlation with the pKi values. No coherent correlations were found between the corresponding fluorinated 3,4′,5-substituted cis-derivatives 1518 and 21 and any molecular descriptor. Based on these findings, and similar to the corresponding nonfluorinated analogs, there appears to be little promise in developing new fluorinated trans- or cis-stilbene leads with significantly higher AhR binding affinity pKi values beyond the range already mapped by the available experimental data sets. We also note that the 3,3′,5-substituted derivatives 11, 12, 23, and 24 did not fit into any sets of correlations that we observed for their 3,4′,5-substituted counterparts. Further synthetic and testing efforts will be needed to generate additional 3,3′,5-substituted derivatives to investigate whether this series of compounds, as with the trans- and cis-substituted non-fluorinated and fluorinated analogs, display their bioactive properties in a manner relatively independent from closely related analogs. Our collective findings also suggest that future QSAR searches for improved AhR binding affinity among stilbene derivatives will likely need to proceed in a more constrained class-by-class fashion owing to the potentially different mechanistic implications from changes in the positions of aromatic substitution and the cis/trans geometrical isomerism.
https://static-content.springer.com/image/art%3A10.1007%2Fs00044-009-9236-2/MediaObjects/44_2009_9236_Fig5_HTML.gif
Fig. 5

Relationships between the charge on the stilbene carbon adjacent to ring 2 (qstilbene C, ring 2) and the experimental AhR binding affinity pKi values for trans- (circles) and cis- (squares) 3,4’,5-substituted fluorinated resveratrol derivatives

Antioxidant activities of hydroxystilbenes

A large number of studies have developed QSARs for predicting the antioxidant activity of various phenolic classes using a range of molecular descriptors, reaction pathways, and theoretical and empirical frameworks (Lien et al., 1999; Gupta et al., 2006; Khlebnikov et al., 2007). However, to date, only semiquantitative SARs have been developed regarding the antioxidant activity of resveratrol and related hydroxystilbenes (Ovesna and Horvathova-Kozics, 2005). A review of the literature provides nine distinct studies that each have investigated the antioxidant activities of resveratrol and at least several other hydroxystilbene derivatives (Wang et al., 1999; Stivala et al., 2001; Gokaraju et al., 2002; Fang et al., 2002; Amorati et al., 2004; Cai et al., 2004; Murias et al., 2005; Cheng et al., 2006; Fang and Zhou, 2008). Cai et al. (2004) simply provide the composite relative antioxidant activity of the various hydroxystilbene derivatives given in Fang et al. (2002); thus, this manuscript does not appear to present new experimental data. Similarly, Fang and Zhou (2008) also presents duplicate antioxidant data from Fang et al. (2002). Murias et al. (2005) determined the antioxidant activity of various hydroxystilbenes, but the raw data behind their figures was not given in the manuscript. The logarithmic scale figures in this publication were not suitable for accurate digitization and numerical extraction, precluding inclusion in our quantitative investigations. Consequently, the literature provides a reduced set of six separate studies and 14 individual data sets amenable to QSAR analysis (Table 2). For reliable QSAR assessments, each reported data set is best treated individually, because pooled data sets across different researchers and/or research groups—despite measuring the same endpoint—may introduce false-positive or -negative conclusions resulting from inherent experimental differences. Of the 33 hydroxystilbenes (2557) that have been the subject of these studies, no compound (other than 1) has been the object of all eight studies, and most derivatives have only been examined for one or two antioxidant assays (compound 25 is the most investigated derivative, and has been studied by six of the eight groups).
Table 2

Summary of previous studies that have reported on the antioxidant activity of resveratrol and multiple hydroxystilbene derivatives

 

(Wang et al., 1999)

(Stivala et al., 2001)

(Fang et al., 2002)

Percent inhibition of DPPH free radical (mean ± SD; n = 3)

Induction time of lipid oxidation by Rancimat method (hours; mean ± SD; n = 3)

Citronellal thermal oxidation inhibition efficiency quantity (EQ; mean ± SD; n = 3)

Effective concentration giving 50% inhibition of lipid peroxidation in rat liver microsomes (EC50; μM; mean ± SD; n = 3)

Concentration required to decrease initial quantity of DPPH by 50% (EC50; μM; mean ± SD; n = 3)

Inhibition rate constant for AAPH initiated peroxidation of linoleic acid in SDS micelles (kinh; 104 L mol−1 s−1; n = 3)

Inhibition period for AAPH initiated peroxidation of linoleic acid in SDS micelles (tinh; 103 s; n = 3)

Inhibition rate constant for AAPH initiated peroxidation of linoleic acid in CTAB micelles (kinh; 104 L mol−1 s−1; n = 3)

Inhibition period for AAPH initiated peroxidation of linoleic acid in CTAB micelles (tinh; 103 s; n = 3)

1

55.1 ± 2.5

12.3 ± 0.7

135 ± 9

0.77 ± 0.08

24.5 ± 1.5

1.3

3.1

0.7

2.7

25

29.8 ± 0.6

2.3 ± 0.1

   

1.7

1.5

0.5

2.4

26

46.5 ± 0.5

3.7 ± 0.2

       

27

67.1 ± 0.6

60.3 ± 1.5

       

28

92.9 ± 0.2

43.0 ± 1.0

       

29

92.9 ± 0.2

48.1 ± 0.7

       

30

  

241 ± 38

1.10 ± 0.09

24.1 ± 1.2

    

31

  

661 ± 20

2.40 ± 0.34

48.6 ± 4.2

    

32

  

355 ± 7

1.22 ± 0.18

30.1 ± 2.1

    

33

         

34

         

35

         

36

         

37

         

38

         

39

         

40

         

41

         

42

         

43

         

44

     

nd

7.2

nd

5.1

45

     

1.4

3.2

0.8

2.6

46

     

2.9

4.9

0.9

4.1

47

         

48

         

49

         

50

         

51

         

52

         

53

         

54

         

55

         

56

     

1.2

2.1

1.0

1.7

57

     

3.1

2.4

1.3

2.9

 

(Gokaraju et al., 2002)

(Amorati et al., 2004)

(Cheng et al., 2006)

 

Superoxide scavenging activity in riboflavin- light-NBT system (IC50; μg mL−1)

Inhibition rate constants for reaction with peroxyl radicals (kinh; 104 L mol−1 s−1)

Inhibition time in Cu2+ induced peroxidation of human LDL (tinh; min; mean ± SD; n = 3)

Inhibition time in AAPH induced peroxidation of human LDL (tinh; min; mean ± SD; n = 3)

Inhibition rate constant in AAPH induced peroxidation of human LDL (kinh; 105 L mol−1 s−1)

1

106.0

14

58 ± 1

23.8 ± 1.3

4.4

25

  

43 ± 1

21.7 ± 1.9

3.3

26

     

27

     

28

     

29

     

30

 

7.4

   

31

     

32

 

13

   

33

  

53 ± 2

17.5 ± 1.9

6.2

34

 

6.3

   

35

 

10

   

36

 

5.7

   

37

 

110

   

38

 

76

   

39

 

6.8

   

40

 

2.3

   

41

 

240

   

42

 

nd

   

43

0.9

    

44

  

105 ± 4

37.2 ± 1.2

4.2

45

  

77 ± 2

18.5 ± 0.5

8.2

46

  

106 ± 3

39.0 ± 2.1

4.9

47

  

107 ± 4

51.0 ± 2.1

3.7

48

  

43 ± 1

27.5 ± 1.6

3.3

49

  

88 ± 3

51.0 ± 2.3

4.1

50

3.0

    

51

8.1

    

52

16.0

    

53

17.5

    

54

9.2

    

55

10.0

    

56

     

57

     

DPPH 2,2-diphenyl-1-picrylhydrazyl, AAPH 2,2′-azobis(methylpropionamidine) dihydrochloride, SDS sodium dodecyl sulfate, CTAB cetyl trimethylammonium, LDL low density lipoprotein, NBT nitro blue tetrazolium

Before a more detailed consideration of any potential antioxidant activity QSARs for resveratrol and related hydroxystilbenes, we investigated the likely speciation of each compound at a physiologically relevant pH of 7.4. The phenolic groups on each compound are acidic, and the speciation in vivo will play a major role in the mechanism(s) and rate(s) of antioxidant behavior. Such pH/pKa effects on antioxidant behavior are not always considered in the literature, but they are critical to a full understanding of any related bioactivities. To the best of our knowledge, 1 is the only compound in the data set that has reliable experimentally determined pKa values, which range from 8.8 for the 4′-OH moiety to 9.8 and 11.4 for the 3/5-OH groups (Lopez-Nicolas and Garcia-Carmona, 2008). Before the work of Lopez-Nicolas and Garcia-Carmona (2008), Galeano Diaz et al. (2007) reported pKa values of 8.2, 9.7, and 10.5 for 1, Takagai et al. (2005) reported values of 8.1 and 9.9, whereas Cao et al. (2006) reported a value of 9.5. The SPARC software program accurately models the acidity constants reported by Lopez-Nicolas and Garcia-Carmona (2008), generally considered the most reliable, yielding corresponding values of 9.0, 9.8, and 11.8, respectively. We note that Stojanovic et al. (2001) reported pKa values of 6.4, 9.4, and 10.5 for 1 without any experimental details. The value of 6.4 reported by Stojanovic et al. (2001) is clearly in error, because well known structure–property relationships for pKa values of substituted phenols require the presence of one or more strongly electron withdrawing groups to be present on the trans-resveratrol framework to achieve such a high acidity constant.

With this validation of the SPARC pKa prediction module for hydroxystilbenes, we proceeded to estimate the pKa values of all phenolic groups on compounds 2557 (Table 3). For <10% ionization to occur in vivo, the pKa of the most acidic phenolic group would need to be >8.4; similarly, for <1% ionization in vivo, all phenolic moieties would need to have pKa values >9.4. Thus, resveratrol derivatives 2729, 43, 46, 47, 50, 51, and 57 are expected to be substantially ionized (>10%) at pH 7.4, whereas all compounds with the exception of 31, 39, and 40 will be at least 1% dissociated at pH 7.4. As shown in Table 3, the range of percent dissociations expected at physiologically relevant pH values varies greatly (from ~0 to >61%), and some compounds are expected to be present in significant quantities as dianions in vivo (e.g., 43, 51).
Table 3

SPARC estimated pKa values of phenolic moieties in the hydroxystilbenes under consideration

ID

Phenolic-OH substitution

2

3

4

5

3′

4′

5′

1

 

9.8 (0.4%)

 

11.8 (~0%)

 

9.0 (2.5%)

 

25

 

9.4 (1.0%)

 

11.5 (~0%)

   

26

 

9.1 (2.0%)

 

11.3 (~0%)

9.8 (0.4%)

 

12.0 (~0%)

27

 

13.3 (~0%)

8.4 (10.0%)

 

9.7 (0.5%)

 

11.8 (~0%)

28

 

11.9 (~0%)

8.1 (16.6%)

11.9 (~0%)

 

9.4 (1.0%)

 

29

 

11.5 (~0%)

8.0 (20.1%)

11.5 (~0%)

9.7 (0.5%)

 

12.1 (~0%)

30

 

9.8 (0.4%)

 

11.8 (~0%)

 

9.0 (2.5%)

 

31

 

9.7 (0.5%)

 

11.8 (~0%)

   

32

     

9.1 (2.0%)

 

33

10.7 (0.1%)

 

8.6 (5.9%)

    

34

     

9.1 (2.0%)

 

35

     

9.1 (2.0%)

 

36

     

9.1 (2.0%)

 

37

    

13.1 (~0%)

8.5 (7.4%)

 

38

    

13.1 (~0%)

8.5 (7.4%)

 

39

     

11.3 (~0%)

 

40

     

11.3 (~0%)

 

41

     

13.8 (~0%)

9.3 (1.2%)

42

     

13.8 (~0%)

9.3 (1.2%)

43

 

11.5 (~0%)

7.9 (24.0%)

11.5 (~0%)

12.2 (~0%)

8.6 (5.9%)

12.2 (~0%)

44

 

13.4 (~0%)

8.5 (7.4%)

  

9.5 (0.8%)

 

45

  

8.9 (3.1%)

  

9.6 (0.6%)

 

46

 

13.1 (~0%)

8.4 (10.0%)

    

47

 

13.1 (~0%)

8.3 (11.2%)

 

13.8 (~0%)

9.1 (2.0%)

 

48

11.2 (~0%)

 

8.7 (4.8%)

  

9.7 (0.5%)

 

49

  

8.9 (3.1%)

  

9.7 (0.5%)

 

50

 

11.9 (~0%)

8.0 (20.1%)

11.9 (~0%)

13.2 (~0%)

8.9 (3.1%)

 

51

 

11.7 (~0%)

8.3 (11.2%)

11.7 (~0%)

12.6 (~0%)

7.2 (61.3%)

 

52a

 

9.7 (0.5%)

   

8.9 (3.1%)

 

53b

     

9.0 (2.5%)

 

54c

 

9.7 (0.5%)

   

8.9 (3.1%)

 

55d

 

9.7 (0.5%)

   

8.9 (3.1%)

 

56

  

9.0 (2.5%)

    

57

 

11.6 (~0%)

7.9 (24.0%)

11.6 (~0%)

   

Percent dissociation at a physiologically relevant pH of 7.4 is shown in parentheses

aThe pKa values on the benzoyl moiety and percent dissociation at pH 7.4 are 13.5 (3/5-OH; ~0%) and 6.7 (4-OH; 91%)

bThe pKa values on the benzoyl moieties and percent dissociation at pH 7.4 are 13.5 (3/3′/5/5-OH; ~0%) and 6.7 (4/4′-OH; 91%)

cThe pKa values on the cinnamoyl moiety and percent dissociation at pH 7.4 are 12.7 (3-OH; ~0%) and 7.7 (4-OH; 33%)

dThe pKa values on the cinnamoyl moiety and percent dissociation at pH 7.4 are 13.9 (3/5-OH; ~0%) and 7.2 (4-OH; 61%)

These issues raise questions regarding the applicability of in vitro testing results toward in vivo conditions. For example, Wang et al. (1999) used methanolic solutions for examining the inhibitory effects of 1 and 2529 on lipid oxidation by the Rancimat method and ethanolic solutions for determining the scavenging effects of these compounds on 2,2-diphenyl-1-picrylhydrazyl (DPPH) radicals. However, even the most acidic protons in this group of compounds (the 4-OH group of 2729, with estimated aqueous pKa values of 8.0–8.4) are expected to have pKa values at least several units higher in these alcoholic solutions because proton transfer to, and solvation by, the less polar organic solvent is less energetically favorable than in aqueous solutions. In Stivala et al. (2001), for compounds 1, 25, and 3032, chlorobenzene (which would effectively suppress phenolic dissociation) was used as the solvent for their citronellal thermal oxidation tests and methanol was used for the DPPH reduction method. Amorati et al. (2004) also used chlorobenzene for their studies regarding the reaction of 1, 30, 32, and 3442 with peroxyl radicals. Murias et al. (2005) used methanol for their DPPH assays on 1, 2629, and 43. In these collective in vitro studies, the resveratrol derivatives would be expected to exert their antioxidant activity exclusively via the molecular form, although some of these compounds would correspondingly be expected to be at least partially dissociated in vivo. Consequently, if the ionized forms have significantly different antioxidant activities from the molecular forms, the in vitro results could have little relevance in vivo.

In contrast, Stivala et al. (2001) examined the inhibition of lipid peroxidation in rat liver microsomes using a phosphate buffer at pH 7.5 for compounds 1, 25, and 3032. Similarly, the studies of Fang et al. (2002) (compounds 1, 25, 4446, 56, and 57), Cai et al. (2004) (compounds 1, 25, 33, 4446, 56, and 57), and Fang and Zhou (2008) (compounds 1, 25, 45, 46, and 56) on 2,2′-azobis (methylpropionamidine) dihydrochloride (AAPH) initiated peroxidation in sodium dodecyl sulfate (SDS) and cetyl trimethylammonium (CTAB) micelles were conducted in a pH 7.4 phosphate buffer designed to mimic the microenvironment of biomembranes. The work of Cheng et al. (2006) on the inhibition of Cu2+-induced peroxidation of human low density lipoprotein (LDL) by compounds 1, 25, 33, and 4449 was performed at pH 7.5 in phosphate buffer, as were their AAPH studies (pH 7.4). The patent of Gokaraju et al. (2002) does not specify the phosphate buffer pH for the superoxide scavenging activity in the riboflavin-light-nitro blue tetrazolium (NBT) system on compounds 1, 43, and 5055, but a value of 7.4 is likely based on the well-known use of this assay. By cross-referencing these compounds with the corresponding pKa values in Table 3, it is apparent that for each test, ambiguity exists about whether the reported antioxidant activity for each compound is dominantly contributed by the molecular or ionized form, or some unknown combination of antioxidant activities from each species.

Furthermore, Lopez-Nicolas and Garcia-Carmona (2008) also found that 1 forms aggregates at concentrations higher than 12–13 μM, with this critical aggregation concentration increasing with increasing pH. The concentrations at which the other hydroxystilbenes cease to be molecularly dispersed have, to the best of our knowledge, not been reported. Fang et al. (2002), Cai et al. (2004), and Fang and Zhou (2008) conducted their duplicated work at hydroxystilbene concentrations of approximately 12 μM, and their data might have been impacted by aggregation effects. Similarly, Stivala et al. (2001) performed their lipid peroxidation in rat liver microsome experiments with hydroxystilbene concentrations ranging from 1 to 100 μM, well into the range of potential aggregation effects, but the IC50 values obtained were all in the low μM range below aggregation thresholds. Cheng et al. (2006) appear to have worked with hydroxystilbene concentrations well below the corresponding aggregation thresholds. The patent by Gokaraju et al. (2002) gives IC50 concentrations for their test compounds in the low mM range, suggesting that this data set also could represent aggregation effects. Future work in this field will need to take special note regarding potential aggregation effects on reported in vitro bioactivities, as this also might complicate extrapolating the laboratory test work to in vivo conditions.

Our initial explorations into potential QSAR development for the antioxidant activity of various hydroxystilbenes used the E-DRAGON descriptors. These types of descriptors have been previously used to investigate other classes of wine polyphenols (Rastija and Medic-Saric, 2009). Because of the small size of all data sets in Table 2 (n = 4 to 11), attempting to develop multivariable regressions for each data set, starting with a large number (>1650) of potential descriptors, led to significant issues regarding multicollinearity of independent variables, overfitting of the data set, and loss of statistical significance with inclusion of multiple descriptors. Thus, in anticipation of future additions to these small data sets, we decided to report the univariate correlations between the top three descriptors for each data set and the log10 transformed antioxidant activities reported for each compound (Table 4). In many cases, these individual univariate equations have sufficient quality of fit (i.e., |r| to 0.9996) that they can already be used to reliably estimate the corresponding antioxidant activity of hydroxystilbenes. Alternatively, as these data sets expand to include new hydroxystilbenes due to future research, our information can be used to help guide development of multivariable QSARs.
Table 4

Univariate linear regression statistics for the log10 transformed antioxidant activities given in Table 2 from various literature data sets against the top three E-DRAGON descriptors for each endpoint and the corresponding ΔBDE, ΔIEET-PT, and ΔIESPLET estimates

References

Endpoint

na

Variable

mb

bc

rd

se

CVf

F

p

Wang et al., (1999)

DPPH % inhibition

6

CSI

0.00678 ± 0.00030

−0.203 ± 0.088

0.996

0.019

0.011

F1,5 = 513

<0.0001

ECC

0.0134 ± 0.0007

−0.119 ± 0.097

0.995

0.022

0.012

F1,5 = 381

<0.0001

Mor04p

1.47 ± 0.08

1.24 ± 0.03

0.994

0.023

0.013

F1,5 = 353

<0.0001

ΔIEET-PT

nsg

ns

−0.806

ns

ns

ns

>0.05

ΔIESPLET

ns

ns

−0.545

ns

ns

ns

>0.05

ΔBDE

−0.0383 ± 0.0096

1.76 ± 0.04

−0.894

0.095

0.054

F1,5 = 16

<0.05

Rancimat induction time

6

TIC2

0.116 ± 0.009

−10.1 ± 0.9

0.988

0.10

0.087

F1,5 = 171

<0.0001

ITH

0.107 ± 0.011

−4.52 ± 0.59

0.980

0.14

0.12

F1,5 = 95

<0.0001

EEig03d

9.47 ± 1.26

−20.7 ± 2.9

0.966

0.18

0.15

F1,5 = 56

<0.001

ΔIEET-PT

−0.101 ± 0.024

1.32 ± 0.12

−0.902

0.29

0.25

F1,5 = 18

<0.05

ΔIESPLET

ns

ns

−0.444

ns

ns

ns

>0.05

ΔBDE

−0.134 ± 0.016

1.13 ± 0.07

−0.972

0.16

0.14

F1,5 = 69

<0.01

Stivala et al., (2001)

Citronellal thermal oxidation inhibition EQ

4

Mor16e

−9.64 ± 0.84

4.04 ± 0.14

−0.993

0.043

0.018

F1,3 = 132

<0.01

ISH

ns

ns

0.939

ns

ns

ns

>0.05

TI2

ns

ns

0.936

ns

ns

ns

>0.05

ΔIEET-PT

ns

ns

−0.860

ns

ns

ns

>0.05

ΔIESPLET

ns

ns

−0.946

ns

ns

ns

>0.05

ΔBDE

ns

ns

0.747

ns

ns

ns

>0.05

EC50 of lipid peroxidation in rat liver microsomes

4

VEA1

3.77 ± 0.84

−14.5 ± 3.3

0.954

0.075

0.75

F1,3 = 20

<0.05

IC4

2.59 ± 0.57

−10.6 ± 2.4

0.954

0.075

0.75

F1,3 = 20

<0.05

EEig02r

ns

ns

0.934

ns

ns

ns

>0.05

ΔIEET-PT

ns

ns

−0.789

ns

ns

ns

>0.05

ΔIESPLET

ns

ns

−0.883

ns

ns

ns

>0.05

ΔBDE

ns

ns

0.879

ns

ns

ns

>0.05

Concentration to decrease DPPH by 50%

4

IC4

1.90 ± 0.04

−6.37 ± 0.16

0.9996

0.0050

0.0034

F1,3 = 2478

<0.001

EEig02r

5.08 ± 0.13

−14.4 ± 0.4

0.9993

0.0065

0.0043

F1,3 = 1488

<0.001

VEA1

2.76 ± 0.10

−9.26 ± 0.38

0.999

0.0088

0.0059

F1,3 = 803

<0.01

ΔIEET-PT

ns

ns

−0.866

ns

ns

ns

>0.05

ΔIESPLET

ns

ns

−0.719

ns

ns

ns

>0.05

ΔBDE

ns

ns

0.855

ns

ns

ns

>0.05

Fang et al., (2002)

AAPH kinh in SDS micelles

6

HATS5p

8.97 ± 0.67

−0.599 ± 0.065

0.989

0.029

0.12

F1,5 = 180

<0.001

HATS5e

1.60 ± 0.16

−0.433 ± 0.069

0.981

0.038

0.15

F1,5 = 104

<0.001

R3 m+

6.03 ± 0.78

−0.0684 ± 0.0465

0.968

0.050

0.20

F1,5 = 59

<0.01

ΔIEET-PT

ns

ns

−0.099

ns

ns

ns

>0.05

ΔIESPLET

ns

ns

−0.763

ns

ns

ns

>0.05

ΔBDE

ns

ns

−0.532

ns

ns

ns

>0.05

AAPH tinh in SDS micelles

7

RDF095p

−0.740 ± 0.073

1.31 ± 0.08

−0.977

0.054

0.11

F1,6 = 103

<0.001

RDF095v

−0.917 ± 0.094

1.28 ± 0.08

−0.975

0.056

0.11

F1,6 = 95

<0.001

RDF090p

0.972 ± 0.102

−0.691 ± 0.126

0.973

0.057

0.12

F1,6 = 92

<0.001

ΔIEET-PT

ns

ns

−0.733

ns

ns

ns

>0.05

ΔIESPLET

ns

ns

−0.387

ns

ns

ns

>0.05

ΔBDE

ns

ns

−0.705

ns

ns

ns

>0.05

AAPH kinh in CTAB micelles

6

EEig08r

0.589 ± 0.162

−0.620 ± 0.151

0.876

0.075

−0.92

F1,5 = 13

<0.05

EEig08x

0.620 ± 0.173

−0.870 ± 0.221

0.874

0.076

−0.93

F1,5 = 13

<0.05

PJI3

2.18 ± 0.65

−2.11 ± 0.61

0.857

0.080

−0.98

F1,5 = 11

<0.05

ΔIEET-PT

ns

ns

−0.700

ns

ns

ns

>0.05

ΔIESPLET

ns

ns

−0.704

ns

ns

ns

>0.05

ΔBDE

−0.0380 ± 0.0089

−0.133 ± 0.030

−0.905

0.067

−0.83

F1,5 = 18

<0.05

AAPH tinh in CTAB micelles

7

RDF090v

0.879 ± 0.151

−0.351 ± 0.141

0.934

0.061

0.13

F1,6 = 34

<0.01

RDF050e

0.145 ± 0.028

−1.90 ± 0.46

0.918

0.068

0.15

F1,6 = 27

<0.01

RDF060p

1.73 ± 0.36

−1.78 ± 0.47

0.905

0.073

0.16

F1,6 = 23

<0.01

ΔIEET-PT

ns

ns

−0.393

ns

ns

ns

>0.05

ΔIESPLET

ns

ns

−0.471

ns

ns

ns

>0.05

ΔBDE

ns

ns

−0.533

ns

ns

ns

>0.05

Gokaraju et al., (2002)

Riboflavin-light-NBT IC50

8

R5u

−4.79 ± 0.60

5.31 ± 0.55

−0.956

0.19

0.20

F1,7 = 64

<0.001

R5e

−4.95 ± 0.85

5.58 ± 0.79

−0.922

0.25

0.26

F1,7 = 34

<0.01

MATS3 m

−5.61 ± 1.13

0.859 ± 0.104

−0.897

0.29

0.30

F1,7 = 25

<0.01

ΔIEET-PT

ns

ns

0.637

ns

ns

ns

>0.05

ΔIESPLET

ns

ns

0.596

ns

ns

ns

>0.05

ΔBDE

0.206 ± 0.076

1.65 ± 0.29

0.744

0.43

0.45

F1,7 = 7.4

<0.05

Amorati et al., (2004)

Peroxyl radical kinh

11

MAXDN

3.13 ± 0.64

−3.74 ± 1.02

0.850

0.35

0.29

F1,10 = 24

<0.001

Gs

9.26 ± 2.38

−0.794 ± 0.526

0.792

0.40

0.34

F1,10 = 15

<0.01

G3 s

1.52 ± 0.39

0.674 ± 0.183

0.790

0.40

0.34

F1,10 = 15

<0.01

ΔIEET-PT

ns

ns

0.059

ns

ns

ns

>0.05

ΔIESPLET

ns

ns

−0.533

ns

ns

ns

>0.05

ΔBDE

ns

ns

−0.555

ns

ns

ns

>0.05

Cheng et al., (2006)

Cu2+ induced peroxidation tinh of human LDL

9

MATS4 m

−0.673 ± 0.124

1.87 ± 0.03

−0.898

0.078

0.042

F1,8 = 29

<0.001

R7e+

−28.9 ± 5.8

2.51 ± 0.13

−0.884

0.083

0.045

F1,8 = 25

<0.01

RDF065 m

−0.289 ± 0.063

4.48 ± 0.57

−0.867

0.088

0.048

F1,8 = 21

<0.01

ΔIEET-PT

ns

ns

−0.309

ns

ns

ns

>0.05

ΔIESPLET

−0.0295 ± 0.0098

1.69 ± 0.07

−0.751

0.12

0.063

F1,8 = 9.0

<0.05

ΔBDE

−0.0329 ± 0.0103

1.75 ± 0.05

−0.771

0.11

0.061

F1,8 = 10

<0.05

AAPH tinh of human LDL

9

H3 m

1.33 ± 0.18

0.742 ± 0.102

0.941

0.066

0.045

F1,8 = 54

<0.001

H5e

0.777 ± 0.109

1.07 ± 0.06

0.937

0.067

0.046

F1,8 = 51

<0.001

Mor03 m

−1.16 ± 0.19

−0.526 ± 0.331

−0.916

0.077

0.053

F1,8 = 37

<0.001

ΔIEET-PT

ns

ns

−0.076

ns

ns

ns

>0.05

ΔIESPLET

ns

ns

−0.528

ns

ns

ns

>0.05

ΔBDE

ns

ns

−0.663

ns

ns

ns

>0.05

AAPH kinh of human LDL

9

GATS4 m

−0.905 ± 0.142

1.55 ± 0.14

−0.924

0.051

0.075

F1,8 = 41

<0.001

MATS7 m

2.00 ± 0.44

1.08 ± 0.09

0.865

0.067

0.098

F1,8 = 21

<0.01

R8 m

−7.14 ± 1.93

1.50 ± 0.22

−0.814

0.078

0.11

F1,8 = 14

<0.01

ΔIEET-PT

ns

ns

0.441

ns

ns

ns

>0.05

ΔIESPLET

ns

ns

0.021

ns

ns

ns

>0.05

ΔBDE

ns

ns

0.554

ns

ns

ns

>0.05

aNumber of experimental data points

bLinear regression slope ± SE

cLinear regression y-intercept ± SE

dCorrelation coefficient

eStandard error of the linear regression

fCoefficient of variation = standard error/mean y

gRegression was not significant at α = 0.05

At least three general mechanisms have been postulated to represent the radical (R·) scavenging properties of phenolic antioxidants (ArOH), such as hydroxystilbenes. Direct hydrogen atom transfer (HAT) to the radical (R·) (2), electron transfer-proton transfer (ET-PT) (3), and sequential proton loss-electron transfer (SPLET), which occurs once the anion (ArO) has been formed (4). All three mechanisms given in Eqs. 24 may occur in parallel, but with different rates (Wright et al., 2001; Zhang et al., 2003; Mayer and Rhile, 2004; Musialik and Litwinienko, 2005; Nakanishi et al., 2005; Zhang and Ji, 2006; Estevez and Mosquera, 2008).
$$ {\text{R}} \cdot + {\text{ ArOH}} \to {\text{ ArO}} \cdot + {\text{ RH}} $$
(2)
$$ {\text{ArOH}} + {\text{R}} \cdot \to {\text{ArOH}}\cdot^{+} + {\text{ R}}^{ - } \to {\text{ArO}} \cdot + {\text{ RH}} $$
(3)
$$ {\text{ArOH}} \to {\text{ArO}}^{ - } + {\text{ H}}^{ + }; {\text{ ArO}}^{ - } + {\text{R}} \cdot \to {\text{ArO}} \cdot + {\text{ R}}^{ - }; {\text{R}}^{ - } + {\text{H}}^{+}\rightarrow {\text{RH}} $$
(4)
Reaction (2) is controlled by the relative bond dissociation enthalpies (ΔBDE) of ArOH and RH. For a suite of various ArOHs, the relative rates of reaction (2) are governed by the ΔBDE of the ArOHs. Reaction (3) involves a presumably rate limited electron transfer. Thus, this reaction should be controlled by the relative ionization energies (ΔIEET-PT) of ArOH and R. For a suite of various ArOHs, the relative rates of reaction (3) are governed by the ΔIEET-PT of the ArOHs. For reaction (4), it is believed that both the proton dissociation enthalpy (PDE; for O–H bond heterolysis) of ArOH and relative ionization energies of ArO and R (ΔIESPLET) are governing factors. For small experimental data sets, developing multivariate QSARs for modeling reaction (4) are difficult due to the unknown relative contributions of the ΔPDE and ΔIESPLET among the compounds of interest, as well as the unknown extent of dissociation that is occurring in the reaction systems containing other components besides ArOH and R· (e.g., micellar systems, etc.). In addition, for all three mechanisms, steric effects may play significant roles, but these cannot readily be incorporated into the thermodynamically based QSARs with ΔBDE, ΔIEET-PT, and ΔIESPLET. Previous work has shown that while BDE correlates with the log10 free radical scavenging rate constants of various compounds (Van Acker et al., 1993; Migliavacca et al., 1997; Zhang, 1998) because of a good correlation between BDE and the activation energy for hydrogen atom abstraction (Tomiyama et al., 1993), there has generally been difficulty reported in the literature in using static molecular descriptors, such as the O–H bond length, O–H Mulliken population, O–H charge difference, and EHOMO for predicting antioxidant activity (Zhang et al., 2000). As a result, QSARs using these additional descriptors were not considered in the present study.
To determine whether more mechanistically based QSARs using ΔBDE and ΔIE could be developed for the available hydroxystilbene antioxidant activity data sets, we first validated the PM6 semiempirical method for estimating ΔBDE (relative to the parent unsubstituted phenol) of phenolic groups against both known experimental values and the B3LYP/LDBS (locally dense basis set) density functional theory (DFT) level calculations conducted by Wright et al. (2001) (Fig. 6). It appears that ΔBDE for phenolics can be reliably estimated by the PM6 method, which is less computationally expensive than ab initio and DFT approaches, making the PM6 method particularly suitable for investigating the antioxidant behavior of larger phenolics, such as resveratrol derivatives, which can have a number of different phenolic groups that require independent BDE calculations.
https://static-content.springer.com/image/art%3A10.1007%2Fs00044-009-9236-2/MediaObjects/44_2009_9236_Fig6_HTML.gif
Fig. 6

Comparison of differences in bond dissociation enthalpies (ΔBDE) for variously substituted phenols relative to the parent unsubstituted phenol using the experimental data set (from Wright et al., 2001; see Table 3 in this reference for compound identities and raw data) and semiempirical PM6 method (squares) and the B3LYP/LDBS DFT level calculations (from Wright et al., 2001) (circles). 2-Methoxyphenol (ΔBDEpred = −7.4 kcal mol−1) and 3-methoxyphenol (ΔBDEpred = +3.7 kcal mol−1) were outliers for the PM6 calculations (closed squares), while 2-t-butylphenol (ΔBDEpred = −10.6 kcal mol−1) was an outlier (closed circle) reported by Wright et al. (2001) for their B3LYP/LDBS DFT level calculations. A 1:1 line (dashed) is shown, as are corresponding regression lines for the B3LYP/LDBS calculations (ΔBDEpred,B3LYP/LDBS = 1.18 (±0.05; ±SE) × ΔBDEexpt + 0.41 (±0.28); r = 0.990, pm=0 < 10−10, pb=0 = 0.16) and PM6 calculations (ΔBDEpred,PM6 = 0.89 (±0.06) × ΔBDEexpt + 0.00 (±0.37); r = 0.976, pm=0 < 10−7, pb=0 = 0.99). Error bars for experimental ΔBDE values are those reported in Wright et al. (2001)

In contrast to the strong agreement between the experimental, B3LYP/LDBS, and PM6 ΔBDE data sets for various phenolics, we did not find good agreement between the B3LYP/LDBS ΔIEET-PT data from Wright et al. (2001) and PM6 estimated ΔIEET-PT (relative to the parent unsubstituted phenol), but found good agreement between the PM6 estimated ΔIEET-PT and corresponding experimental values for a wide range of variously substituted phenols as reported in the NIST Chemistry WebBook (Fig. 7). The poor agreement between the B3LYP/LDBS and PM6 computationally derived data sets is perhaps not unexpected, given the extended discussion accorded by Wright et al. (2001) to addressing potential concerns regarding ΔIEET-PT values derived via the B3LYP/LDBS method, as well as the established large general disagreements between the B3LYP and PM6 methods for absolute IEET-PT values (Puzyn et al., 2008). Together, the good agreements we found between the experimental and PM6 ΔBDE and ΔIEET-PT suggest the semiempirical PM6 method may be both an accurate and computationally inexpensive means for probing the antioxidant activity of phenols. We also stress that this appears to be the first report validating the semiempirical PM6 method for obtaining reliable ΔBDE and ΔIEET-PT estimates for phenolics.
https://static-content.springer.com/image/art%3A10.1007%2Fs00044-009-9236-2/MediaObjects/44_2009_9236_Fig7_HTML.gif
Fig. 7

Comparison of differences in ionization energies (ΔIE) between B3LYP/LDBS DFT level calculations (squares; from Wright et al., 2001) and experimentally determined values (circles; from the NIST Chemistry WebBook; http://webbook.nist.gov/chemistry/) relative to the parent unsubstituted phenol and corresponding semiempirical PM6 estimated ΔIE for a range of variously substituted phenols

To further validate our PM6 calculations, we also conducted DFT calculations for the absolute BDE of the unsubstituted parent phenol at the B3LYP/6-31++G(d,p) and B3LYP/6-311++G(d,p) levels and for the three possible phenolic BDEs of 1 at the B3LYP/6-31++G(d,p) level. Our BDEs of 85.6 and 86.5 kcal mol−1 for phenol at the B3LYP/6-31++G(d,p) and B3LYP/6-311++G(d,p) levels, respectively, were in good agreement with the B3LYP/LDBS value of 87.1 kcal mol−1 reported by Wright et al. (2001), and higher than the PM6 value of 80.0 kcal mol−1. In general, we found that the PM6 method underestimated absolute BDEs of the phenols reported in Table 3 of Wright et al. (2001) (average difference of −5.7 kcal mol−1 between the gas phase PM6 and gas phase B3LYP/LDBS methods and −7.6 kcal mol−1 between the gas phase PM6 and experimental values [obtained in benzene]), but was comparable in accuracy to DFT methods for calculating ΔBDE. Because the experimental values in Table 3 of Wright et al. (2001) were obtained in benzene, we also repeated our PM6 calculations in this solvent model. There was no improvement in the agreement between the PM6 and experimental data (average difference of −7.7 kcal mol−1) or the PM6 calculations in benzene and the gas phase B3LYP/LDBS results (average difference of −5.8 kcal mol−1). For 1, the PM6 method estimated absolute BDEs of 81.5, 83.1, and 77.0 kcal mol−1 for the 3-, 5-, and 4′-OH groups, respectively, in good agreement with our corresponding B3LYP/6-31++G(d,p) estimates of 83.2, 84.9, and 78.3 kcal mol−1, with the B3LYP/LDBS value of 78.9 kcal mol−1 reported by Wright et al. (2001) for the 4′-OH group, and with the BLYP/DND values of 84.4, 84.8, and 78.2 kcal mol−1 reported by Xu et al. (2007). Similarly, Wright et al. (2001) reported a B3LYP/LDBS BDE of 71.5 kcal mol−1 for the 3-OH group of 27, in good agreement with our corresponding PM6 value of 73.3 kcal mol−1.

Stivala et al. (2001) also showed, using semiempirical PM3 calculations, that hydroxyl groups in the para positions (4 and 4′) on the stilbene framework conferred additional radical stabilization, compared to the meta positions (3, 3′, 5, and 5′), due to the ability to delocalize the radical density from para-phenoxyl radicals to the ortho-/para-positions across both benzene rings. In contrast, meta-phenoxyl radicals cannot delocalize their radical density to the other ring. This work has received computational support at the DFT level by Amorati et al. (2004) and Rossi et al. (2008). As was noted by both Stivala et al. (2001) and Amorati et al. (2004), the lack of coplanarity for cis-hydroxystilbene derivatives hinders delocalization of radical density across both rings even for the 4 and 4′ positions. This also is apparent in our PM6 calculations, where we find the 4′-OH BDE to be 0.9 kcal mol−1 higher in the cis-resveratrol 30 compared with its trans-counterpart 1, a value that is in excellent agreement with the PM3 estimate by Stivala et al. (2001) of a 0.8 kcal mol−1 resonance stabilization, and also with the B3LYP/6-31G* calculations by Amorati et al. (2004), which gave a 1.2 kcal mol−1 higher BDE for the 4′-OH group of 30 versus 1. Amorati et al. (2004) also experimentally determined the BDE values of 37, 39, 40, and 41, yielding ΔBDE (relative to 1) of −4.4, −5.1, −3.3, and −6.0 kcal mol−1, respectively, which compare favorably to our corresponding PM6 estimates of −4.2, −3.5, −2.9, and −4.9 kcal mol−1 (average unsigned difference of 0.8 kcal mol−1 between the two data sets, which is within the calculation error).

Also consistent with the theoretical calculations of Wright et al. (2001) and the empirical analysis of antioxidant activities by Fang et al. (2002), we find lower estimated ΔBDE values where two phenolic groups are adjacent. As discussed by Wright et al. (2001), the intramolecular hydrogen bonding between the phenoxyl radical and the adjacent undissociated phenol group further stabilizes radical formation, thereby lowering the BDE and enhancing hydrogen abstraction. However, Fang et al. (2002) and Fang and Zhou (2008) erroneously used a pKa of 6.4 for trans-resveratrol, and thereby assumed a far higher level of dissociation (approximately 90%) under their experimental conditions of pH 7.4 than was actually the case using the correct pKa of 8.8 (which yields a percent dissociation of only 3.8% at pH 7.4). Consequently, the mechanistic analysis by Fang et al. (2002) and Fang and Zhou (2008), whereby the dominant mechanism was assumed to be SPLET, is likely in error. If trans-resveratrol reacts via electron transfer as the major antioxidant mechanism at pH 7.4, it is instead likely via the molecular form, followed by proton transfer from the resulting radical cation (i.e., ET-PT mechanism).

Following these ΔBDE and ΔIEET-PT validation steps, we used the PM6 method to estimate ΔIEET-PT, ΔIESPLET (ionization energy of the first phenolate expected to form based on the relative pKa values of all phenolic groups for a particular hydroxystilbene), and ΔBDE for all phenolic groups in compounds 1 and 2557 (Table 5). Note that ΔIEET-PT, ΔIESPLET, and ΔBDEmin (the minimum ΔBDE for each compound) in Table 5 are relative to compound 1, rather than the parent unsubstituted phenol. ΔIEET-PT and ΔBDE values were found to be significantly correlated (r = 0.590, p < 0.001), as were ΔIEET-PT and ΔIESPLET (r = 0.411, p = 0.02), and ΔIEET-PT and ΔBDEmin (r = 0.418, p = 0.01). These significant intercorrelations preclude any development of multiple linear regressions for predicting antioxidant activity using these three independent variables. Previous work (Cheng et al., 2002) has developed multiple linear regression-based QSARs for antioxidant activities using ΔBDE, ΔIE, and other descriptors that are likely highly intercorrelated. The subsequent use of such modeling results to determine normalized correlation coefficients for inferring the relative importance of HAT and ET-PT mechanisms may be in error, as multicollinearity could result in spurious interpretations of any QSARs. For these reasons, caution needs to be exercised in ensuring that all antioxidant QSARs meet the basic non-multicollinearity requirements before using any results for mechanistic inferences. In our case, we chose not to attempt any mechanistic interpretations from such multivariate QSARs based on IE and BDE values with multicollinearity issues, or even to develop such models, because the results could inevitably be called into doubt.
Table 5

Estimated relative molecular ionization energies (ΔIE) for the ET-PT and SPLET mechanisms and homolytic bond dissociation enthalpies (ΔBDE) of phenolic O–H bonds for the HAT mechanism in hydroxystilbenes using the semiempirical PM6 method

ID

ΔIEET-PTa (kcal mol−1)

ΔIESPLETa (kcal mol−1)

ΔBDEb (kcal mol−1)

2

3

4

5

3′

4′

5′

minc

1

0.0

0.0

 

4.5

 

6.1

 

0.0

 

0.0

25

6.9

−4.9

 

4.5

 

6.3

   

4.5

26

9.0

−2.1

 

5.3

 

6.8

6.8

 

5.3

5.3

27

−2.5

−3.9

 

−3.7

−1.1

 

5.6

 

4.0

−3.7

28

−5.1

−13.4

 

−3.6

−4.9

0.5

 

−2.0

 

−4.9

29

−0.2

−6.7

 

−3.0

−3.6

0.9

5.9

 

4.4

−3.6

30

0.5

−3.2

 

30.0

 

30.5

 

0.9

 

0.9

31

−4.2

−4.9

 

6.1

 

4.2

   

4.2

32

−3.5

−3.8

     

−0.4

 

−0.4

33

−8.5

−0.7

−0.6

 

−1.9

    

−1.9

34

−2.3

−4.7

     

0.6

 

0.6

35

−9.5

−9.8

     

−4.8

 

−4.8

36

−7.8

−7.5

     

−0.2

 

−0.2

37

−5.8

−8.0

    

−4.2

−1.4

 

−4.2

38

−5.1

−8.8

    

−4.5

−0.8

 

−4.5

39

−10.8

−5.3

     

−3.5

 

−3.5

40

−11.5

−6.2

     

−2.9

 

−2.9

41

−8.5

−12.0

     

−2.5

−4.9

−4.9

42

−10.1

−8.8

     

−2.0

18.7

−2.0

43

−2.3

−6.9

 

−3.6

−5.2

0.4

0.4

−5.2

−3.6

−5.2

44

−7.6

−11.3

 

−4.5

−2.8

  

−2.6

 

−4.5

45

−7.4

−7.8

  

−2.1

  

−2.1

 

−2.1

46

−5.5

−10.8

 

−4.3

−2.6

    

−4.3

47

−6.2

−4.9

 

−0.2

−7.8

 

−4.8

−3.2

 

−7.8

48

−12.2

−1.2

−1.1

 

−2.3

  

−2.0

 

−2.3

49

−11.1

−7.1

  

−8.1

  

−6.5

 

−8.1

50

−4.6

−9.8

 

0.1

−5.5

−4.0

−4.3

−2.9

 

−5.5

51

−1.4

1.6

 

0.1

−5.5

−3.9

−4.4

−3.1

 

−5.5

52

−0.7

2.0

 

4.4

   

−0.5

 

−2.2d

53

1.8

9.9

     

−0.6

 

−3.6e

54

−2.3

0.2

 

4.4

   

−0.6

 

−0.6f

55

−1.4

1.2

 

4.4

   

−0.5

 

−3.7g

56

−4.4

−7.1

  

−1.6

    

−1.6

57

−3.0

−12.9

 

−3.5

−4.7

0.7

   

−4.7

aΔIE calculated relative to trans-resveratrol (1)

bΔBDE calculated relative to the 4′-OH in trans-resveratrol (1)

cmin = lowest ΔBDE for each compound relative to the 4′-OH in trans-resveratrol (1)

dThe O–H ΔBDE on the benzoyl moiety are −2.2 (3-OH), −2.4 (4-OH), and −2.2 (5-OH) kcal mol−1, giving a ΔBDEmin of −2.2 kcal mol−1 rather than −0.5 kcal mol−1

eThe O–H ΔBDE on the benzoyl moieties are −1.1 (3-OH), −2.4 (4-OH), −1.9 (5-OH), −3.6 (3′), 2.0 (4′), and 2.0 (5′) kcal mol−1, giving a ΔBDEmin of −3.6 kcal mol−1 rather than −0.6 kcal mol−1

fThe O–H ΔBDE on the cinnamoyl moiety are 17.3 (3-OH) and 15.6 (4-OH) kcal mol−1

gThe O–H ΔBDE on the cinnamoyl moiety are −3.0 (3-OH), 15.9 (4-OH), and −3.7 (5-OH) kcal mol−1, giving a ΔBDEmin of −3.7 kcal mol−1 rather than −0.5 kcal mol−1

The ΔBDE calculations suggest the following order of decreasing antioxidant activity among the 34 hydroxystilbenes, assuming that reaction (2) is dominant and that—in solution—the overall antioxidant activity is dominated by the undissociated form of each compound: 49 > 47 > 50 = 51 > 43 > 28 = 41 > 35 > 57 > 44 = 38 > 46 > 37 > 27 = 55 > 29 = 53 > 39 > 40 > 48 > 52 > 45 > 42 > 33 > 56 > 54 > 32 > 36 > 1 > 34 > 30 > 31 > 25 > 26. In comparison, the ΔIEET-PT calculations suggest the following order, assuming that reaction (3) is dominant and that the dominant reactivity is from the ionized forms: 48 > 40 > 49 > 39 > 42 > 35 > 41 = 33 > 36 > 44 > 45 > 47 > 37 > 46 > 28 = 38 > 50 > 56 > 31 > 32 > 57 > 27 > 43 = 54 = 34 > 51 = 55 > 52 > 29 > 1 > 30 > 53 > 25 > 26. For reaction (4), and assuming deprotonation is not controlling, such that ΔIESPLET governs the resulting reactivity, the ΔIESPLET calculations suggest the following antioxidant activity order if phenolate ionization is dominant: 28 > 57 > 41 > 44 > 46 > 35 = 50 > 42 = 38 > 37 > 45 > 36 > 49 = 56 > 43 > 29 > 40 > 39 > 47 = 31 = 25 > 34 > 27 > 32 > 30 > 26 > 48 > 33 > 1 > 54 > 55 > 51 > 52 > 53. A rank analysis (1 = highest predicted antioxidant activity; 34 = lowest predicted antioxidant activity) of the hydroxystilbenes based on the relative PM6 estimated ΔBDE and ΔIEET-PT yields only modest correlation (r = 0.34, rankΔIE,ET-PT = 11.6 + 0.33 × rankΔBDE; p = 0.05), as does ΔIEET-PT and ΔIESPLET (r = 0.50, rankΔIE,SPLET = 8.6 + 0.50 × rankΔIE,ET-PT; p < 0.01) and ΔBDE and ΔIESPLET (r = 0.43, rankΔIE,SPLET = 9.9 + 0.43 × rankΔBDE; p = 0.01). These results indicate that, despite the collinearity of these antioxidant activity descriptors (precluding multivariate analyses), a purely thermodynamic-based antioxidant activity analysis using relative ΔBDE, ΔIEET-PT, and ΔIESPLET (HAT, ET-PT, and SPLET mechanisms, respectively) would yield generally different expectations of antioxidant behavior for these hydroxystilbenes. Although the actual antioxidant mechanism(s) for these compounds are unknown, Wright et al. (2001) suggest that HAT would be dominant for hydroxystilbenes, whereas Nakanishi et al. (2007) have suggested that the ET-PT mechanism may be dominant (at least for 1).

Murias et al. (2005) determined the concentrations of 1, 2629, and 43 causing 50% quenching (IC50) of DMPO(5,5-dimethyl-1-pyrroline-N-oxide)/·OOH signals in electron spin resonance (ESR) experiments and the second order rate constants for hydrogen atom abstraction by DPPH radicals (k9). As noted, the absence of tabular raw data from this publication precludes a quantitative analysis. The relative orders of IC50 and k9 reported by Murias et al. (2005) were as follows (these orders as given from left to right also reflect expected decreasing antioxidant activity): IC50, 27 < 43 < 29 < 28 < 1 < 26; k9, 28 > 27 > 29 > 1 > 26 > 43. In comparison, the ΔBDE and ΔIEET-PT values predict the following orders of decreasing antioxidant activity among these six compounds: ΔBDE, 43 > 28 > 27 > 29 > 1 > 26; ΔIEET-PT, 28 > 27 > 43 > 29 > 1 > 26. Neither computational approach fully predicts the relative antioxidant activity order given by the IC50 or k9 values, although the ΔIEET-PT approach correctly predicted the order of 5 of the 6 compounds for both endpoints (failing only to predict the correct placement of 28 for the IC50 data set and 43 for the k9 data set). The ΔBDE approach correctly predicts the order for 4 of the 6 IC50 values and 5 of the 6 k9 values (failing to predict the correct placement of 43 and 28 for the IC50 data set and 43 for the k9 data set). Consequently, our computational methods appear to offer some utility in estimating the general order of antioxidant activities measured by the quenching of DMPO/·OOH signals by ESR and by the rate constants for hydrogen atom abstraction by DPPH radicals. It is important to note that 28 and 43 are the two most acidic compounds among this data set, with the following acidity order (the lowest estimated phenolic pKa value is given in parentheses), 43 (7.9) > 28 (8.1) > 27 (8.4) > 1 (9.0) > 26 (9.1) > 29 (9.7), although the DMPO/·OOH/ESR and DPPH test work was performed in DMSO and methanol solvents, respectively, which should have suppressed ionization under the test conditions (i.e., making the SPLET mechanism inoperative; consistent with the rank order of ΔIESPLET only correctly predicting the IC50 order for 3 of the 6 compounds and the k9 order for 4 of the 6 compounds).

Compared with the E-DRAGON descriptors, the correlations between the ΔIEET-PT, ΔIESPLET, and ΔBDE values and the various log10 transformed antioxidant activities reported in each of the six studies given in Table 2 were generally lower and included a number of nonsignificant correlations (Table 4). At present, even with the future addition of new compounds to the existing data set, there appears less promise in this type of IE- and BDE-based first principles attempt to estimate the antioxidant activities of various hydroxystilbenes compared with the less mechanistically satisfying two- and three-dimensional molecular descriptor approach. One particular complexity for a theoretical approach are the multiple potential initial oxidative reaction paths for each compound (including different phenolic groups, the benzene rings, and the double bond), as well as multiple reaction sites on a single molecule that contribute toward the overall measured activity. For example, ortho- and para-phenolic groups (including orientations across the ethene linkage) can undergo sequential oxidation to form relatively stable ortho- and para-quinones, respectively (Cai et al., 2004). Consequently, more complete product studies need to be conducted on model hydroxystilbenes by various antioxidant assays to understand better the relative contributions of the numerous mechanistic pathways toward overall antioxidant behavior. As well, the rank ordering of both ΔIEET-PT, ΔIESPLET, and ΔBDE values and corresponding antioxidant activities did not result in any combination of IE/BDE-antioxidant activity combination accurately predicting the correct experimental rank order.

Conclusions

A general quantitative structure-activity relationship (QSAR) was developed for the aryl hydrocarbon receptor (AhR) binding affinity of a series of 24 cis- and trans-substituted resveratrol derivatives using a starting database of >1650 molecular descriptors from the E-DRAGON software program. The optimum two descriptor QSAR had a high statistical significance (>99.99%) and a modest correlation coefficient (r = 0.805). Collectively, our results and those of other researchers have investigated the entire effective domain of molecular descriptors available for QSAR modeling using a variety of two- and three-dimensional descriptors. No approaches to date yield a general QSAR with satisfactorily high predictive power for estimating the AhR binding affinity of both cis- and trans-resveratrol derivatives. Improved predictive power can be obtained by dividing the experimental data set into both cis- and trans-isomers and fluorinated and nonfluorinated subclasses, after which Hammett substituent constants and atomic charges can be used to reliably estimate AhR binding affinities. The lower AhR activity of cis-resveratrol derivatives seems to be related to the larger molecular size and the inability of the stilbene group to achieve planarity relative to the more active trans-isomers.

A variety of empirical descriptors can be used to estimate the antioxidant activity of hydroxystilbenes via univariate correlations. Once additional hydroxystilbene derivatives are obtained and subjected to antioxidant testing, there seems to be significant promise in obtaining high-quality multivariate QSARs for predicting the antioxidant activity of this compound class. With the limited available experimental data set at present, a mechanistically based theoretical analysis of hydroxystilbene antioxidant activity QSARs using bond dissociation enthalpies and ionization energies for the molecular and dissociated forms yields comparatively lower quality QSAR fits.

Acknowledgments

S. Rayne thanks the Natural Sciences and Engineering Research Council (NSERC) of Canada for partial financial support at the University of Winnipeg. The Western Canada Research Grid (WestGrid) provided computational support under project #100185 at Okanagan College (K. Forest).

Copyright information

© Birkhäuser Boston 2009