Chem. Res. Toxicol. 1996, 9, 1305-1312
1305
A Quantum Chemical Explanation of the Antioxidant Activity of Flavonoids Saskia A. B. E. van Acker,*,†,‡ Marcel J. de Groot,† Dirk-Jan van den Berg,†,‡ Miche`l N. J. L. Tromp,† Gabrielle Donne´-Op den Kelder,† Wim J. F. van der Vijgh,‡ and Aalt Bast† Leiden/Amsterdam Center for Drug Research, Department of Pharmacochemistry, Faculty of Chemistry, Vrije Universiteit, De Boelelaan 1083, 1081 HV Amsterdam, The Netherlands, and Department of Medical Oncology, Free University Hospital, De Boelelaan 1117, 1081 HV Amsterdam, The Netherlands Received June 13, 1996X
Flavonoids are a group of naturally occurring antioxidants, which over the past years have gained tremendous interest because of their possible therapeutic applicability. The mechanism of their antioxidant activity has been extensively studied over several decades. However, there is still much confusion about the molecular mechanism of radical scavenging and the relationship between structure and activity. Therefore, we have calculated the heat of formation and the geometry of both the parent compound and the corresponding radical using the ab initio program GAMESS. We have compared their differences in energy in order to gain insight into the stability of the radical and the ease with which it is formed. We have also investigated the spin density of the radical, to determine the delocalization possibilities. These calculated data were compared with experimental data from ESR (hyperfine coupling constants) and electrochemical oxidation (Ep/2) and were found to be in good agreement. By comparing the geometries of several flavonoids, we were able to explain the structural dependency of the antioxidant action of these compounds. The extremely good antioxidant activity of the flavonols could be explained by the formation of an intramolecular hydrogen bond.
Introduction Flavonoids are a group of polyphenolic compounds ubiquitously found in fruits and vegetables. Because of their broad pharmacological activity, they have recently gained tremendous interest. In many traditional medicines, part of the therapeutic effect may be ascribed to the flavonoids. The pharmacological effect can be explained by their inhibition of certain enzymes and their antioxidant activity (1). Many authors have attempted to elucidate the structure-activity relationships (SAR)1 of this antioxidant activity (2-9). However, this is hampered by the fact that antioxidant activity can be considered to be determined by several factors, of which lipophilicity (and thus uptake into the membranes, which are often the site of action), iron chelation, and scavenging of free radicals are the most important. There appears to be agreement now on the SAR of the free radical scavenging activity: hydroxyl groups in ring B, preferably a catechol moiety, are required for good scavenging activity and a C2-C3 double bond in combination with a hydroxyl at C3 can further increase the scavenging activity (Figure 1). There is, however, no molecular explanation as to why these structural features are important. Usually (10, 11), conjugation effects are mentioned, but this does not explain why a double bond without a hydroxyl at C3 does not have an increasing * To whom correspondence and requests for reprints should be addressed at the Department of Pharmacochemistry, Faculty of Chemistry, Vrije Universiteit, De Boelelaan 1083, 1081 HV Amsterdam, The Netherlands. † Vrije Universiteit. ‡ Free University Hospital. X Abstract published in Advance ACS Abstracts, November 1, 1996. 1 Abbreviations: SAR, structure-activity relationship; QSAR, quantitative structure-activity relationship; RHF, restricted Hartree Fock; UHF, unrestricted Hartree Fock; DMA, distributed multipole analysis; LPO, lipid peroxidation.
S0893-228x(96)00096-3 CCC: $12.00
Figure 1. The structural subclasses of the flavonoids.
effect. Several other compounds which do not meet these structural requirements are unexpectedly active, such as catechin (cyanidanol), which lacks the C2-C3 double bond and the 4-keto function. Thus until now, SAR has only been descriptive, not explanatory by means of, for example, a quantitative relationship (QSAR).1 © 1996 American Chemical Society
1306 Chem. Res. Toxicol., Vol. 9, No. 8, 1996
The aim of the present investigation was to find a possible explanation for the observations described above and understand the effects of the structure on the scavenging activity in a molecular way. Therefore, we have used quantum chemical calculations to optimize the geometry of both the molecule and the corresponding radical and to compare the heat of formation. This can give information about the ease with which the radical is formed. The spin distribution gives insight into the degree of delocalization and thus conjugation, which is a measure for the stability of the radical.
Materials and Methods Chemicals. Hesperidin (97%), diosmin (95%), fisetin, naringin, phloridzin (99%), phloretin (98%), and galangin were obtained from Aldrich (Milwaukee, WI), and hesperetin, naringenin (95%), taxifolin, NADP, and horseradish peroxidase were obtained from Sigma (St. Louis, MO). Rutin was obtained from Merck (Darmstadt, Germany), and quercetin, myricetin (>97%), pelargonidin chloride, apigenin (98%), and kaempferol (96%) were purchased from Fluka (Buchs, Switzerland). Cyanidin chloride and luteolin (90%) were purchased from Roth (Karlsruhe, Germany). The hydroxyethyl rutosides, trihydroxyethyl quercetin, and (+)-catechin (cyanidanol) were a generous gift from Zyma (Nyon, Switzerland). Hydrogen peroxide (30%) was obtained from Baker (Deventer, The Netherlands). All other chemicals were of the highest grade of purity available. Half Peak Oxidation Potentials. The pH dependence of the Ep/2 of monoHER and kaempferol was measured at several pH values ranging from pH 2 to pH 13. The pH was changed by adding either NaOH or HCl to a 50 mM phosphate buffer. Flavonoids were dissolved in DMSO, diluted 1:1 with nanopure water, and further diluted in buffer, until a final concentration of 100 µM. This buffer solution was gassed with oxygenfree nitrogen for 5 min before each measurement, and nitrogen was led over the solution while measuring. A platinum electrode was used as a working electrode, with a platinum counter electrode and a saturated calomel reference electrode. A scan was made from -0.2 to 0.6 V with a scanning speed of 20 mV/s on a PSTAT 10 potentiostat connected to an Autolab (Eco Chemie, Utrecht, The Netherlands) and controlled by the program General Purpose Electrochemical System 3.0 (Eco Chemie, Utrecht, The Netherlands) run on an Olivetti PC M240. Electron Spin Resonance. The ESR experiments were performed on a Bruker ESP300 ESR spectrometer. Modulation frequency was 100 kHz, the amplitude 1 G. Microwave frequency was set at 9.79 GHz, with a power of 20 mW. Whenever necessary, spectra were simulated using the program EPRcalc, which was included in the operating software of this spectrometer (see below). The flavonoid solution in 100% DMSO (650 µL) was added to a flat cell, and after placing it in the cavity, 250 µL of saturated solution of KOH was added. This resulted in spontaneous oxidation by air as the oxidation potential was lowered considerably. Alternatively, flavonoids were dissolved in DMSO and diluted with phosphate buffer, pH 7.4, rendering a solution with less than 4% DMSO. To this solution H2O2 and horseradish peroxidase were added to initiate an oxidation reaction. Both experiments were performed in the presence and absence of Mg2+, which is known to stabilize catechol radicals. This was done to increase the signal to noise ratio and to obtain spectra of higher resolution. The resolution of ESR spectra at pH 13 was considerably better than that of spectra measured under physiological conditions. Electron Spin Resonance Simulations. ESR spectra were also simulated by the program EPRcalc implemented on a Bruker ESP300 ESR spectrometer. Hyperfine coupling constants (a) were taken from the experimental spectra, from Kuhnle et al. (12), and from the calculated spin densities (see below).
van Acker et al. Quantum Chemical Calculations. (A) Structural Optimization (Minimal Energy Conformation). Flavonoid X-ray structures were taken from the Cambridge Structural Database (13) at CAOS/CAMM Centre in Nijmegen. Quercetin and (+)-catechin were used as found. Rutin was found without hydrogen atoms, which were added using the molecular modeling program package ChemX (14). Taxifolin was constructed in ChemX starting from the X-ray structure of naringenin. An estimate for the global minimum structure was determined by conformational analysis with the software package Macro Model (15-17) installed at the CAOS/CAMM Centre (KUN, Nijmegen) by rotating around C2-C1′ and C3-sugar axes and around axes of OH groups not involved in an H-bond (Figure 1). Therefore, first the most suitable Macro Model forcefield for these structures was determined. The AMBER forcefield (18, 19) was found to have the least number of illdefined parameters, and was therefore used for the conformational analysis. For taxifolin and (+)-catechin the “ring-flip” was included in the analysis by using the “closure bond” option. In order to include this flexibility into the saturated C ring, the O1-C2 bond was opened and rotation around C4-C3 and C3C2 was performed. The lowest energy conformation was taken, except for (+)catechin, where the conformational analysis led to the wrong stereoisomer around C2. In this case we therefore used both the lowest energy conformation (the (-) isomer) and the lowest energy conformation of the (+) isomer, which was the third lowest energy conformation. The aforementioned conformations were optimized using the quantum chemical program package GAMESS-UK (20, 21) at the Restricted Hartree Fock (RHF)1 level with the STO-3G (Slater type orbitals comprised of 3 Gaussians) minimal basis set (22) on a CRAY-98 supercomputer facility. On the resulting geometry, a single point RHF calculation in a SV (split valence) 6-31G (23, 24) basis set was performed to derive the energy. From these GAMESS-optimized structures, the structures of apigenin, naringenin, galangin, kaempferol, diosmin, hesperetin, and 3-OMe-quercetin were constructed by addition or removal of OH or OMe groups. These structures were also optimized using the procedure described above. (B) Calculation of Spin Distribution and Differences in Heat of Formation between Parent Molecule and Corresponding Radical. A theoretical, quantum chemically determined, suitable parameter for describing the abstraction of a H• from an O-H bond is the difference in heat of formation between the flavonoid and its corresponding radical (∆∆Hf). The delocalization possibilities within the radical flavonoids will largely contribute to the corresponding ∆∆Hf value. In order to gain insight into the delocalization possibilities of flavonoids (which are the reason for their antioxidant activity), spin distributions were calculated for the radicals of several main classes. Ep/2 and ∆∆Hf both involve the breaking of an O-H bond either hetero- or homolytically (see reaction 1 in discussion) and are therefore expected to correlate well, but only in case the proposed reaction mechanism is correct. Geometry optimizations for the radicals were performed at the unrestricted Hartree Fock (UHF) level with the STO-3G basis set. To form the radical, an H• was removed from the 4′hydroxyl or, if this was not present, from the 3- or 3′-hydroxyl which are also, but less, likely to be involved in a scavenging reaction. The ∆Hf of the radical and the unpaired spin distribution were calculated using a single point (open shell RHF) energy and distributed multipole analysis (DMA) calculation (25) in a SV-631G basis set. The use of stabilization energies (∆Hr, our ∆∆Hf) for the prediction of the reaction of a flavonoid with a radical in hydrogen abstraction reactions is illustrated by Korzekwa et al. (26). They found that a linear (Brønsted) relationship exists between stability of radicals (experimental bond dissociation energy data) and activation energies (∆Hq) of hydrogen abstraction for similar reactions in a series of analogous substrates; the relative order of hydrogen atom abstraction can be obtained
Antioxidant Activity of Flavonoids
Chem. Res. Toxicol., Vol. 9, No. 8, 1996 1307
Table 1. Flavonoids Used in This Study Divided into Subclasses
Table 2. Torsion Angles (deg) of Ring B in both Parent Compound and Radical Relative to O1
substituents class flavonols
compound
quercetin fisetin rutin kaempferol galangin monoHER flavanon(ol)s naringenin hesperetin taxifolin flavones apigenin diosmin luteolin flavanols (+)-catechin anthocyanidins cyanidin a
3
5
7
3′
4′
5′
OH OH ORua OH OH ORu H H OH H H H OH OH
OH H OH OH OH OH OH OH OH OH OH OH OH OH
OH OH OH OH OH OEtOH OH OH OH OH ORu OH OH OH
OH OH OH H H OH H OH OH H OH OH OH OH
OH OH OH OH H OH OH OMe OH OH OMe OH OH OH
H H H H H H H H H H H H H H
a
flavonoid
molecule
radical
quercetin luteolin (+)-catechin apigenin diosmin galangin kaempferol taxifolin 3-OMe-quercetin hesperetin naringenin rutin
-0.29 16.29 35.64 16.48 15.54 -0.27 -0.14 -27.64 -23.58 -42.28 -42.73 27.17
-0.19 0.04 39.19 -0.05 0.00 0.07 0.00 -37.53 0.04 -41.74 -41.34 NDa
ND: not determined.
Ru: rutinose ()glucose-rhamnose).
by simply calculating the energy difference between a compound and its potential radical (∆E ) our ∆∆Hf).2 The relatively tedious task of searching for and optimizing transition states can thus be avoided (26). On the basis of these observations, the molecule radical couple having the lowest ∆∆Hf value in our calculations will most easily allow hydrogen atom abstraction by a radical, which is likely to be an important factor in the scavenging reaction. The resulting flavonoid radical will then be able to scavenge another radical; thus predictions will depend at least on two phenomena: ease of hydrogen atom abstraction (depending on the stabilization energies) and ease of second radical scavenging (depending on the spin distribution in the radical). The first process is probably rate limiting; the second process is a critical one, because a very reactive flavonoid radical (high amount of localized spin) will be able to start a radical chain reaction and thus be a pro-oxidant. Structures which were used for the calculations were quercetin, luteolin, (+)-catechin, taxifolin, rutin, kaempferol, galangin, naringenin, apigenin, diosmin, and hesperetin (Figure 1, Table 1). Furthermore, a hypothetical structure of quercetin with a 3-OMe moiety was included. Structural variations are C2-C3 double bond or saturated ring C, keto group or H at C4, H, OH, or ORutinose at C3, and different hydroxylation patterns on ring B, which are the most common structural differences between flavonoid subclasses. ESR Simulations with Calculated Spin Densities. Spin densities as calculated by GAMESS were converted into a values by O’Connell’s equation:
a ) QFπ where a is the hyperfine splitting constant, Q is O’Connell’s constant which O’Connell determined to lie between 22.5 and 29 for aromatic compounds, and Fπ is the spin density in the π-orbitals of the particular atom to which the hydrogen is attached (27). The a values are entered into EPRcalc implemented on a Bruker ESP300 ESR spectrometer for spectrum simulation.
Results Calculations. (A) Flavonoids. It was found that the optimized structure of quercetin was completely planar; i.e., the torsion angle of ring B with the rest of the molecule was close to 0° (see Table 2 and Figure 2). This means that the molecule is completely conjugated. As there was one report in the literature (28) stating that in the very similar flavone (no substituents, basic struc2 A Brønsted relationship is a linear correlation between ∆H and r ∆Hq. These relationships are observed in a series of similar reactions and suggest that a constant fraction of effects that stabilize or destabilize the reactants or products is present in the transition state.
Figure 2. 3-D structures of quercetin (A) and luteolin (B) after optimization using GAMESS.
ture of the subclass, see Table 1) this torsion angle was about 20° and thus ring B is not completely conjugated to the rest of the molecule, we investigated whether the removal of the 3-OH would induce such a torsion angle. The 3-OH was removed from Macro Model optimized quercetin, rendering luteolin. It appeared that the torsion angle of ring B with the rest of the molecule was 17° and thus in agreement with flavone, which is a similar structure, at least in ring C. Comparable results were obtained for galangin, kaempferol, apigenin, and diosmin, of which compounds with a 3-OH are planar, and the ones lacking a 3-OH moiety are twisted, meaning that it can be stated in general that flavonols are planar and that in flavones ring B is slightly ((20°) twisted relative to the plane of ring A and C (Figure 2). In (-)-catechin the 3-OH was found to have an H-bond interaction with the hetero oxygen in ring C. H-bonds were generally also found between the hydroxyl groups on ring B (the catechol moiety) and between 5-OH and 4-keto and 3-OH and 4-keto when appropriate (Figure 3). These hydrogen bonds were indicated by the program ChemX. (B) Flavonoid Radicals. The radicals of the flavones were, in contrast to the parent compounds, planar and thus completely conjugated. Within a subclass, the same behavior in geometry change can be observed when going from molecule to radical. The ∆∆Hf per molecule radical couple is given in Table 3 for the situation when there is an H-bond in the
1308 Chem. Res. Toxicol., Vol. 9, No. 8, 1996
van Acker et al.
Figure 3. Hydrogen bonds as indicated by the program ChemX. Table 3. Radical Stability Compared with Oxidation Potential and Influence of a H-Bond in the Catechol Moiety in Ring B
molecule
∆Hf molecule (au)a
∆∆Hf mol to rad (kJ/mol)
∆∆Hf no H-bond (kJ/mol)
Ep/2 (mV)b
quercetin luteolin (+)-catechin apigenin diosmin galangin kaempferol taxifolin 3-OMe-quercetin hesperetin naringenin rutin
-1097.40 -1022.58 -1024.89 -947.76 -1061.59 -947.12 -1022.58 -1098.56 -1136.40 -1062.75 -948.92 -1704.41
1667.3519 1672.8599 1623.2839 1676.9872 1676.5389 1683.5924 1671.5921 1646.1754 1675.9057 1671.5796 1650.3027 ND
1684.2315 1697.8856 1648.2500 NDc ND -c 1670.3985 ND ND ND
30 180 160 >1000 >1000 320 120 150 ND 400 600 180
a 1 au (atomic unit) ) 2625.5 kJ/mol. b From (29). c Not present; ND: not determined.
Figure 4. Correlation of ∆∆Hf with Ep/2 for quercetin, kaempferol, luteolin, and galangin.
catechol moiety in ring B and for the situation when there is no H-bond. It can be seen that the H-bond interaction has a large stabilizing effect on the catechol radical. When comparing ∆∆Hf and Ep/2, it can be seen that there is a trend to correlate within each subclass, but only a quantitative correlation for the flavon(ol)s (Figure 4) and not for the whole group of flavonoids (Table 3). An explanation may be that the Brønsted equation only holds within a subclass. From the spin distributions (Figure 5) it can be seen that when oxidation takes place in ring B as is usually the case, almost all spin remains in ring B, even in case of a completely conjugated molecule such as quercetin. Surprisingly, about 84% of all spin density remains on the O from where the H• is removed. This is quite unexpected, as flavonoids are usually considered good
Figure 5. Spin distribution of the quercetin and taxifolin radical.
radical scavengers because of their excellent delocalization possibilities. However, it can also be seen that delocalization is larger in compounds with a higher degree of conjugation, e.g., quercetin vs taxifolin. Apparently, this small difference in delocalization has a large influence on antioxidant activity, as these highly conjugated flavonoids have a higher in vitro activity than their less conjugated counterparts (29). Electron Spin Resonance. (A) Spectra Obtained at pH 13. The luteolin spectrum consists of 6 peaks with intensities 1:3:4:4:3:1 (Figure 6). An explanation is two overlapping quartets, with the last two peaks of the first quartet overlapping with the first two peaks of the last quartet. Spectra of quercetin are difficult to interpret. At different scan speeds different spectra were found, indicating that several consecutive reactions occur. Quercetin is easily oxidized, even at physiological pH. Lowering the Ep/2 for other parts of the molecule (increasing the oxidizability) by increasing the pH may give rise to several different radicals, which are being formed at different speeds. The spectrum of fisetin is very similar to that of quercetin. Myricetin gives a triplet caused by coupling of 2′-H and 6′-H. The taxifolin spectrum shows two overlapping quartets (Figure 6). It appears that also the 2-H is involved in coupling. From the quantum chemical calculations it appears that this 2-H has a spin density close to that of 2′-H and 6′-H (Figure 5).3 This signal is split by 5′-H which has a relatively high spin density. (+)-Catechin shows a similar spectrum, consisting of what appears to be two overlapping quartets, with the last peak of the first quartet overlapping with the first peak of the last quartet. Interpretation is the same as for taxifolin. Kaempferol shows a spectrum which appears to be two overlapping triplets (1:2:2:2:1) (Figure 6). Interpretation is difficult, as there are 2 × 2 equivalent protons in ring B. Probably at this pH ring B is deprotonated and oxidation occurs by a different mechanism than at physiological pH, as can be seen from the pH dependence 3 Note that under the experimental conditions the catechol moiety in ring B is deprotonated, which gives rise to an increase in spin on H2′, C4′ and H6′.
Antioxidant Activity of Flavonoids
Chem. Res. Toxicol., Vol. 9, No. 8, 1996 1309
A
B
Figure 7. Relationship between Ep/2 and pH for kaempferol.
A C
B D
Figure 8. Influence of Mg2+ on the ESR spectrum of monoHER, pH 7.4. Figure 6. ESR spectra of luteolin (A), taxifolin (B), and kaempferol at pH 13 (C) and 7.4 (D).
of the Ep/2 of kaempferol (Figure 7). It appears that at pH >9, no protons are involved in the reaction, indicating oxidation by removal of an electron only and not an H•. The spectrum might be interpreted as 2 equivalent protons, possibly 6-H and 8-H, and a proton on one of the hydroxyl moieties in ring A, possibly 5-OH. Another possibility is an H-bond between 2′-H or 6′-H and 3-OH, which would also render 2 equivalent and one other proton. This theory is further discussed below. Rutin and monoHER show spectra consisting of two triplets caused by the two equivalent protons 2′-H and 6′-H, and the proton at 5′, which has the highest spin density. No spectra could be obtained from all other flavonoids. Probably their Ep/2 values were too high for them to be spontaneously oxidized under these conditions. Ep/2 values for the flavonoids which could be oxidized were
