Ab initio Study of Potential Ultrafast Internal Conversion Routes in Oxybenzone, Caffeic Acid and Ferulic Acid: Implications for sunscreens

Tolga N. V. Karsilia,b*, BarbaraMarchettia, Michael N. R. Ashfolda* and Wolfgang Domckeb

aSchool of Chemistry, University of Bristol, Bristol, BS8 1TS, United Kingdom

bDepartment of Chemistry, TechnischeUniversitätMünchen, Lichtenbergstr. 4, 85747 Garching, Germany

Number of Figures: 11

Number of Tables: 2

ABSTRACT

Oxybenzene (OB) and ferulic acid (FA) both find use in commercial sunscreens; caffeic acid (CA) differs from FA by virtue of an –OH group in place of a –OCH3 group on the aromatic ring. We report the results of ab initio calculations designed to explore the excited state non-radiative relaxation pathways that provide photostability to these molecules and the photoprotection they offer towards UV-A and UV-B radiation. In the case of OB, internal conversion (IC) is deduced to occur on ultrafast timescales, via a barrierless electron driven H atom transfer pathway from the S1(11n*) state to a conical intersection (CI) with the ground (S0) state potential energy surface (PES). The situation with respect to CA and FA is somewhat less clear cut, with low energy CIs identified linking excited states to the S0 state following photoexcitation and subsequent evolution along (i) a ring centred out-of-plane deformation coordinate, (ii) the E/Z isomerism coordinate and, in the case of CA, (iii) an O–H stretch coordinate. Analogy with catechol suggests that the last of these processes (if active) would lead to radical formation (and thus potential phototoxicity), encouraging a suggestion that FA might be superior to CA as a sunscreen ingredient.

  1. INTRODUCTION

Jablonski diagrams1invariably show internal conversion (IC) as a probable spin allowed radiationless pathway for photoexcited molecules but it is only now, with the advent of high level electronic structure calculations, that we are in a position to understand and to predict the detailed nuclear dynamics by which IC occurs in any given molecule. Conical intersections (CIs) between potential energy surfaces (PESs) are now recognised as crucial to enabling population transfer between different electronic states,2,3and this article illustrates some of the diverse range of nuclear distortions that can drive IC – in the context of photoprotection mechanisms that prevail in selected molecules with real (or potential) application in sunscreens.

Sunscreens are designed to attenuate the UV-A and UV-B components within the solar spectrum (i.e. wavelengths longer than ~280 nm), thereby reducing the probability of skin cancers like melanomaor squamous cell carcinoma.4, 5 Commercial sunscreens typically comprise a heavy carrier oil, photostable inorganic particulates (TiO2 and/or ZnO, designed to reflect, scatter and absorb some of the incident radiation) and conjugated organic molecules that offer photoprotection without photodegrading or inducing unwanted phototoxicity. Most organic molecules used in sunscreens comprise aromatic rings conjugated to carbonyl groups (e.g. avobenzene, oxybenzone, cinnamates6 (including caffeic and ferulic acid) andsalicilates)7. These molecules offer photoprotection by virtue of their high UV absorption cross-sections and high IC efficiencies; UV absorption results in electronic excitation, but the dominant (ideally the exclusive) fate of this electronic excitation is rapid conversion to vibrational energy which is then dissipated (as heat) within the sunscreen.

The present study explores, theoretically, the detailed anatomy of several possible IC routes in three molecules that either are, or potentially could be, used in sunscreens: caffeic acid (3-(3,4-dihydroxyphenyl)-2-propenoic acid, henceforth often written simply as CA), its close relative ferulic acid (3-(4-hydroxy-3-methoxyphenyl)-2-propenoic acid, henceforth FA) and oxybenzone ((2-hydroxy-4-methoxyphenyl)-phenylmethanone, henceforth OB). CA and FA are also intermediates in the synthesis of monolignols (the monomers of lignin).8, 9 The minimum energy structures of each of these are depicted in fig. 1, with the constituent atoms numbered for future reference.

Figure 1: (a) MP2 ground state minimum energy geometry of the most stable isomer of caffeic/ferulic acid. (b) CAM-B3LYP minimum energy geometry of oxybenzone. The constituent atoms are numbered for future reference, and the dotted lines indicate hydrogen bonds.

CA and FA can both be viewed as substituted phenols or as substituted cinnamic acids. Adding the hydroxy/methoxy groups to the ring (cf. bare cinnamic acid) has the effect of red-shifting the UV absorption spectrum so as to match better with that needed for sunscreen applications. Previous studies of phenol itself have identified possible IC pathways from the first excited (S1 state) as a result of distortions along both the O–H stretch coordinate, and an out-of-plane ring deformation.10, 11 The side chain in CA and FA also contains a C=C double bond, that has the potential to isomerise upon UV photoexcitation. Tuna et al.12recently reported a detailed theoretical study of IC enabled by just such a photo-induced isomerism (henceforth photoisomerism) in urocanic acid – a UV filter that is naturally present in human skin. Other well documented examples of molecules that undergo efficient E/Zphotoisomerismupon UV excitation include stilbenes13 andretinal.14 OBserves to illustrate another potential IC route (which could also operate in urocanic acid). OB contains an H atom donor (a hydroxyl group) and an acceptor (the carbonyl oxygen atom)aligned so as to encourage intramolecular hydrogen transfer. Such electron-driven Hatom transfer processes on excited state PESs have been identified as key relaxation pathways underpinning the photostability of many biomolecules.15-19 The present work constitutes a computational study of the energetics associated with each of these various pathways in CA, FA and OB, the results of which enables prediction of the most probable IC pathway(s) in each case.

  1. COMPUTATIONAL METHODOLOGY

The calculations for CA/FA and for OB used different methodologies and in some cases sought to address different questions. Key points are summarised below, and further details are provided as Supporting Information (SI).

i) Caffeic/Ferulic acid

Ground state conformers. The ground (S0) state minimum energy geometries of selected low lying conformational and E/Zisomers of CA and FA were optimised in two ways. One involved use of Molpro 2010.1,20andMøller-Plesset second order perturbation theory (MP2) coupled with the aug-cc-pVDZ21 basis set with extra sets of even-tempered s and p diffuse functions added to the O atoms. No symmetry constraints were used during this optimisation. The second employed Gaussian 09,22density functional theory (DFT), the Coulomb Attenuated Model Becke-3rd parameter-Lee-Yang-Parr (CAM-B3LYP) functional and the aug-cc-pVDZ basis set.

RO-H/RO-CH3stretch coordinate: Using Molpro 2010.1, unrelaxed (rigid body) potential energy functions for the S0 and first few excited singlet states of CA and FA were calculated using complete active space with second order perturbation (CASPT2) theory, based on a state-averaged (SA) reference wavefunction (SA5-CASSCF) and assuming CS(i.e. planar) geometry. The active space for these calculations consisted of twelve electrons in eleven orbitals (12/11)and comprised the following orbitals: three  and three * orbitals, the 2pxorbitals on O(10) and O(12), and theO atom centred 3s Rydberg orbital and the and* orbitals associated with the O–H/O–CH3 stretch coordinate of interest. An imaginary level shift of 0.5Eh was used in all CASPT2 calculations in this publication to aid convergence and avoid intruder state problems. These calculations were based on the MP2 optmised geometry and utilised the aug-cc-pVDZ basis set

Out-of-plane ring deformation and E/Z isomerisation coordinates: The lowest energy CIs along these two coordinateswere optimised at the CASSCF / 6-31G(d)23 level of theory using Gaussian 09and a 6/6 active spaceinvolving three  and three * orbitals. In each case, two independent linear interpolations in internal coordinates (LIICs) were then constructed linking the optimised S0state geometries of both the E and the Z isomer and the optimised CI geometry. The energies of the S0 and first few excited singlet states were then calculated (using MolPro) at various points along each LIIC at the CASPT2/cc-pVDZ level of theory using a 14/12(for CA, 12/11 for FA) active space based on a SA4-CASSCF reference wavefunction. The 12 orbitals used in these calculations for CA werefouroccupied  and threeunoccupied* orbitals, 2px orbitals on O(10) and O(12),  and * orbitals associated with the carbonyl C=O bond and the 2py lone pair on O(21). In the case of FA, theC=O centred  orbital was excluded to aid convergence; subsequent test calculations for CA using this reduced active space showed its exclusion made negligibledifference to the calculated vertical excitation and CI energies.

ii) Oxybenzone

The ground state minimum energy geometries of selected conformers were optimised using the CAM-B3LYP and M06-2X functionals embedded within DFTin Gaussian 09, with the cc-pVDZ basis set. Minimum energy pathwaysfor electron-driven H atom transfer on the S0 and S1 PESs of OB were explored using the O(23)–H(24) stretch as the driving coordinate. An oxetane-type photoproduct was also identifiedin these calculations. This was optimised using the DFT/CAM-B3LYP/cc-pVDZ level of theory, and the variations in the potential energies of the S1 and S0 states calculated along a LIIC linking this structure to that at which the S0 and S1 states become near degenerate (RO(23)–H(24) ~2 Å).

Low energy CIs with prefulvenic geometries were optimised in Gaussian 09 using the CASSCF(10,8)/3-21G level of theory. PECs for the first four singlet excited states along the LIIC linking the lowest energy CI to the ground state minimum energy geometry were calculated having first re-optimised the latter at the MP2/cc-pVDZ level of theory. In Molpro, CASPT2(10,8)/cc-pVDZ was then used for the potential energy calculations. An active space comprising ten electrons in eight orbitals (thefour highest the three lowest * and the O(2py) orbitals) was used in both the CASSCF and CASPT2 calculations.

  1. RESULTS and DISCUSSION

3.1 Ground State Conformational Stabilities

Each of the molecules considered in this publication support different conformational and E/Z isomers, and we here focus just on the sub-set of lower energy isomers depicted in figure 2. Their respective stabilities were explored with MP2, DFT and, in some cases, CASPT2, and the (relative) energies so derived are presented in Table 1. The minimum energy isomers of CA/FA and OB in their respective ground electronic states are identified as structures Aand Fin fig. 2, and Cartesian coordinates for the minimum energy geometries of conformers A – D of both CA and FA are listed in Tables S2 and S3 in the SI.

Figure 2:Selected isomers of caffeic/ferulic acid (A-E) and oxybenzone(F-H), the most stable of which are, respectively, Aand F.

Focussing first on CA and FA, we see that rotation about the C(14)=C(16) double bond in the side chain enables two geometric isomers: E (trans)and Z (cis). The Eisomer (A) is more stable than the Z isomer (B) in both molecules, reflecting the greater steric hindrance in the latter. Changing the relative orientations of the O-H/O-CH3moietiesattached to the ring (i.e. rotation about the C(6)-O(12) and/or C(5)-O(10) bonds) gives rise tosyn/ anti rotational isomers. As Table 1 shows, the most stable isomers (A and C) are those that benefit from intramolecular H-bonding between O(10) and H(13). Rotational isomerism can also occur via rotation around the C(16)C(18) bond. Structure C is calculated to be marginally less stable in both CA and FA (Table 1). Again, its geometric isomer (D) is significantly less stable on account of increased steric hindrance. The present results are in good accord with several previous studies regarding the relative stabilities of the E and Z isomers of CA and FA,24,25though one earlier study of FA26(which employed a smaller basis set) placed conformer C just below A. OB can also exist as two rotational isomers (F and G), the former of which is favoured by the strong intramolecular hydrogen bond that exists between O(22) and H(24). Rotation about the C(4)–C(10) bond leads to further (sterically disfavoured) isomers, the lower energy of which (H) is calculated to be marginally more stable than G.

3.2 Vertical Excitation Energies and Oscillator Strengths

Table 2 lists calculated vertical excitation energies and oscillator strengths of transitions to the 11*, 11n* and 21* electronic states of different isomers of CA, FA and OB. The associated orbitals and orbital promotions for CA and OB are shown in figure 3.

Figure 3: Orbitals and orbitalpromotions involved in forming the first three excited singlet states of E-caffeic acid (left) and oxybenzone (right). Note that the energetic ordering of the second and third promotions in Z-caffeic acid is reversed (these transitions are labelled in parentheses, and the corresponding orbitals are displayed in fig. S1). The orbitals and electronic transitions shown in the left hand column are extremely similar to those for structures A and B in ferulic acid, which are thus not displayed here but can be found in the SI (figure S1).

Caffeic/Ferulic Acid

For both molecules, the first three singlet excited states upon vertical excitationare1* (two) and1n* (one) in nature. As in molecules like phenol and catechol,27, 28 the 11* state is the S1state for all of the isomers investigated. The participating orbitals (LUMOHOMO in fig. 3) show good spatial overlap, and the S1–S0 transitions have appreciable oscillator strengths – as do transitions to the 21* state (see Table 2).29-31 Forming the 11n* state, in contrast, involves electron promotion between orbitals that have little spatial overlap; the calculated oscillator strength for the 11n*S0 transition is orders of magnitude smaller.

The energetic ordering of these states inthe lowest energy E isomer (A) is11*11n*21*. The calculated 11*S0excitation energiesare isomer-independent, implying that the HOMO (and LUMO (orbitals are destabilised by similar amounts upon EZ isomerisation. Such is not surprising, given that both are predominantly ring centred orbitals.

The 11n* state arises from a LUMOHOMO1excitation. The HOMO–1 is dominated by the 2py orbital on O(21) (fig. 3). Table 1 shows that EZ isomerisation reverses the relative ordering of the 11n* and 21* states in both rotational isomers, that the 11n*S0 excitation energies are similar in A and C, and that the corresponding transition in Dis substantially red-shifted compared with B. These differences can be rationalised by considering the likely consequences of steric crowding between the COOH moiety and the ring. In the E isomers (A and C), the distance between these groups is sufficiently large that the relative orientation of the COOHgroup is unimportant whereas, in the Z isomers, a logicalstarting premise might be that steric interaction between the ring (particularly H(8)) and the COOH group would substantially destabilise the HOMO–1 and thus reduce the 11n*S0gap. Such expectations are largely borne out in the case of D, but not by B. These isomers are distinguished by having the C=O group oriented towards (B) or away from (D) the ring. As fig. S1 (in the SI) shows, in the specific case of isomer B, this interaction polarises the LUMO–1 (the 2py orbital on O(21)), which increases the strength of the H-bond between O(21) and H(20) and provides some counteracting stabilisation of the n orbital.

The 21*S0 excitations show obvious red-shiftsupon EZ isomerisation. As fig. 3 shows, these are best viewed as LUMO+1HOMO promotions and, asnoted previously, EZ isomerisation destabilises the HOMO. Thus the calculated energies imply such isomerisation also induces a (relative) stabilisation of the LUMO+1. These trends can also be attributed to steric crowding. Aplausible starting premise in this case would be that the steric crowding accompanying EZ isomerisation destabilises the valence orbitals (including the* orbital of interest in this transition). That being the case, the reduced excitation energies of the Z (cf.E) isomers must reflect some preferential stabilisation of the LUMO+1, most notably in the case of isomer D. This can be explained as follows: the terminal COOH group is electron withdrawing, and thus attracts π density from the adjacent C=C double bond, thereby destabilising the LUMO+1 (since it is antibondingin this region). So any geometric rearrangement that reduces the electron withdrawing ability of the COOH group (such as unfavourable electrostatic interaction between the ring π system and the ‘acceptor’ C=O π* orbital in the case of the Z isomers) will act to stabilise the LUMO+1 and reduce the 21*S0 transition energy (most particularly in the case that the C=O group is directed towards the ring (i.e. in isomer B)).

Oxybenzone

Table 2 also shows calculated vertical excitation energies to the first three excited singlet states of our chosen isomers of OB. In all cases, the 11n* state formed by electron promotion from the O(2py) orbital to the LUMO lies lowest in energy. The calculated oscillator strength of this S1–S0 transition is fairly low (reflecting the indifferent spatial overlap of the combining orbitals). This state develops increasing charge transfer (CT) character upon extending the O(23)H(24) bond, and the n* description progressively evolves towards n* with the * orbital localised on this bond. All higher excited states of OB investigated in this work are of *character.

3.3 Possible Internal Conversion Pathways

3.3.1 Caffeic/Ferulic Acid

OH/OCH3 Bond Extension

Much recent interest has focussed on the RO-Hstretch coordinate in photoexcited phenolic molecules.10, 32 Figure 4(a) displays PECs for the S0, 11*, 21* and 11* states of the most stable isomer of CA along the lowest energy (O(10)H)bond extension coordinate. Note that the11n* state was excluded from these calculations in order to aid convergence, and that the corresponding PECs along RO-H for other isomers of CA are shown in fig.S2 of the SI. The PEC for the 11* state shows a CI with that of the dissociative 11* state at short RO-H (~1.2 Å) and, as in phenol,27catechol28and related systems,30, 33 it is reasonable to assume that molecules excited to low vibrational levels of the 11* state of CA could tunnel through the barrier under this 11*/11* CI.27, 28,34, 35 The11* PEC experiences another CI, with the diabatic S0 PEC, at extended RO-H (~1.8 Å). As noted previously,36non-adiabatic coupling at this 11*/S0 CI could provide a route for IC to the ground state PES. Again, analogy with phenol, catechol, etc,27, 28, 35,37suggests that excitation at energies above the 11*/11* CI is likely to result in prompt O(10)–H bond fission.

The lowest energy1* state in FA, in contrast, arises as a result of a *O(10)–CH3← excitation and, as fig. 4(b) shows, is dissociative (at long range) along RO–CH3. (Readers are referred to fig. S3for the corresponding PECs for other isomers of FA and for PECs along the lowest energy RO-Hstretch coordinate). The asymptotic limit of the 11*O(10)–CH3 PEC in FA lies ~1 eV lower than the corresponding limit for the 11*O(10)–H PEC in CA; the implied difference in O–H and O–CH3 bond strengths accords well with that found in, for example, 4-methoxyphenol, for which D0(H3COPhOH) = 3.55 eV 38cf. D0(H3C–OPhOH) = 2.7 eV.39 The 11*O(10)–CH3 PEC in FA also exhibits CIs with the 11* and S0 states (at RO–CH3~1.7 Å and 2.5 Å, respectively), but shows clear differences with the 11*O(10)–H PEC in CA. The obvious minimum at RO–CH3~1.7 Å is attributable to increasedRydberg/valence mixing.31, 40The resulting barrier in this diabatic PEC, the magnitude of the barrier under the 11*O(10)–H/11* CI and the mass of the departing CH3 group all encourage the view that tunnelling along RO–CH3is an improbable loss mechanism for FA molecules following photoexcitation to the 11* state. Such a view accords with the results of a recent study of the near UV photochemistry of guaiacol(2-methoxyphenol),which found no translationally excited CH3fragmentssuch as would be expected in the event of O–CH3 bond fission on the 11* potential following excitation to the analogous 11* state.9