Intermolecular Hydrogen Bonding Modulates O‐H Photodissociation in Molecular Aggregates of a Catechol Derivative

The catechol functional group plays a major role in the chemistry of a wide variety of molecules important in biology and technology. In eumelanin, intermolecular hydrogen bonding between these functional groups is thought to contribute to UV photoprotective and radical buffering properties, but the mechanisms are poorly understood. Here, aggregates of 4‐t‐butylcatechol are used as model systems to study how intermolecular hydrogen bonding influences photochemical pathways that may occur in eumelanin. Ultrafast UV‐visible and mid‐IR transient absorption measurements are used to identify the photochemical processes of 4‐t‐butylcatechol monomers and their hydrogen‐bonded aggregates in cyclohexane solution. Monomer photoexcitation results in hydrogen atom ejection to the solvent via homolytic O‐H bond dissociation with a time constant of 12 ps, producing a neutral semiquinone radical with a lifetime greater than 1 ns. In contrast, intermolecular hydrogen bonding interactions within aggregates retard O‐H bond photodissociation by over an order of magnitude in time. Excited state structural relaxation is proposed to slow O‐H dissociation, allowing internal conversion to the ground state to occur in hundreds of picoseconds in competition with this channel. The semiquinone radicals formed in the aggregates exhibit spectral broadening of both their electronic and vibrational transitions.

Intermolecular hydrogen-bonding interactions between catechol functional groups are central to some of the supramolecular motifs proposed to occur in eumelanins (2,3,17). Aggregated domains of dihydroxy-containing molecules play a photoprotective role in melanin chemistry (14), but the precise mechanisms remain elusive. Uncertainty about the chemical structures found in eumelanins and the likely presence of significant heterogeneity (1,15,18) complicate the interpretation of spectroscopic measurements (19)(20)(21)(22). For these reasons, a number of investigators have studied simplified model systems containing chemical groups in arrangements thought to be present in eumelanin (20,21,23). These studies have addressed the impact of covalent bonding within dimers and oligomers in addition to solute-solvent interactions, but studies isolating the effects of intermolecular hydrogen-bonding interactions in aggregates are lacking to the best of our knowledge.
Here, we investigate the effects of intermolecular hydrogen bonding on UV-initiated photochemical and photophysical relaxation pathways in aggregates of 4-t-butylcatechol (4-tBuC, Scheme 1) as a step toward understanding photoprocesses in natural and synthetic eumelanins. This catechol derivative was chosen as an alternative to 5,6-dihydroxyindole and other eumelanin precursors because the latter compounds spontaneously oxidize in solution (22,24). Our measurements were performed in cyclohexane solutions to isolate intermolecular hydrogen bonding from solute-solvent interactions. A t-butyl-substituted catechol was chosen because the alkyl side chain significantly enhances its solubility in cyclohexane. Horbury et al. (25) studied O-H bond photodissociation of monomeric 4-tBuC molecules in several solvents. Inspired by this work, our investigation seeks to understand how intermolecular hydrogen bonding interactions within 4-tBuC aggregates modulate their photochemistry.
Steady-state UV and FTIR spectroscopy show that hydrogenbonded aggregates form by self-association of 4-tBuC molecules in cyclohexane solution. Electronic and vibrational transient absorption measurements are combined to identify and compare the photochemistry of isolated molecules to the aggregates. In agreement with Horbury et al. (25), the primary photochemical reaction of 4-tBuC in a weakly interacting, nonpolar solvent is homolytic O-H bond dissociation, resulting in hydrogen atom ejection to the solvent and formation of a neutral semiquinone radical. In contrast, intermolecular hydrogen bonding in 4-tBuC aggregates fosters excited state structural relaxation and slows the O-H bond dissociation that ultimately gives rise to hydrogenbonded neutral semiquinone radicals. Intermolecular hydrogenbonding interactions and the associated changes in molecular conformation reveal potential mechanisms for photochemical modulation in hydrogen-bonded aggregates.
Steady-state spectroscopy. UV absorption spectra were collected using a Cary 5000 UV-Vis-NIR Spectrometer (Agilent; Santa Clara, CA). Solutions were measured in fused silica cuvettes (for 0.5 and 10 mm path lengths) or CaF 2 demountable liquid cells (for 12-250 lm path lengths), depending on the concentration. For all spectra, the maximum absorbance was kept below 1.0.
Vibrational absorption spectra were collected using an FTIR spectrometer (FT/IR-4200, JASCO; Easton, MD). Samples were loaded into a custom-made demountable liquid cell with CaF 2 windows. Teflon spacers with various thicknesses (ranging 100-500 lm) were used depending on the solution concentration.
Transient absorption spectroscopy. Transient UV-Vis spectroscopy was performed using an ultrafast pump-probe spectrometer. The pump was generated by driving an optical parametric amplifier (TOPAS Prime, Coherent; Santa Clara, CA) tuned to 265 nm with the output of ã 100 fs Ti:Sapphire laser amplifier (Astrella, Coherent; Santa Clara, CA). The probe was a white light supercontinuum that was generated by focusing a portion of the Ti:Sapphire output onto a CaF 2 window. The excitation density at the sample position was~620 and~290 lJ cm À2 for 5 and 75 mM samples respectively. The absorbed energy density was 310 and 260 lJ cm À2 for the 5 and 75 mM samples respectively. Solution samples were flowed during all measurements in a demountable CaF 2 flow cell (TFC-M25-3, Harrick Scientific Products; Pleasantville, NY), which was pumped using a peristaltic pump (Masterflex, Cole-Parmer; Vernon Hills, IL). The path lengths used were 500 and 100 lm for the 5 and 75 mM samples respectively. Samples were replaced after every~10 min of scanning time in order to prevent the build-up of o-quinone photoproducts. The transmitted probe beam was dispersed and detected using a prism spectrometer (2.0 VIS, Stresing Entwicklungsb€ uro; Berlin, Germany).
Time-resolved vibrational spectroscopy was performed using an ultrafast time-resolved infrared (TRIR) pump-probe spectrometer. The pump was generated by driving an optical parametric amplifier (OPerA Solo, Coherent; Santa Clara, CA) tuned to 265 nm with the output of ã 100 fs Ti:Sapphire laser amplifier (Libra HE, Coherent; Santa Clara, CA). The probe was generated by driving an optical parametric amplifier (TOPAS-C/nDFG, Coherent; Santa Clara, CA), which was tuned between 5.5-8 lm. The excitation density at the sample position was 0.55 mJ cm À2 and the absorbed energy density was~0.45 mJ cm À2 . Solution samples were recirculated during all measurements through the same flow cell used for the transient UV-Vis measurements using a peristaltic pump (Masterflex L/S; Cole-Parmer, Vernon Hills, IL). The path lengths used were 500 and 100 lm for the 5 and 75 mM samples respectively. Samples were replaced every~10 min of scanning time in order to prevent measuring the build-up of o-quinone photoproducts. The transmitted probe beam was dispersed using a spectrograph (Triax, Horiba; Kisshoin, Japan) and detected by a dual-row, 64-element/row, liquid N 2cooled HgCdTe array (MCT-10-128, Infrared Associates; Stuart, Florida), and read out by a high speed integrator (FPAS-0144, Infrared Systems Development; Winter Park, FL).
UV-irradiation experiments and in-situ absorption spectroscopy. UV irradiation experiments were performed using a homebuilt system. UV radiation from a deuterium lamp (Q Series, Newport; Irvine, California) was collected and focused onto the sample using parabolic reflectors. The beam was filtered using a 1 cm chloroform filter (cut-on was~250 nm). In-situ absorption spectra were collected by measuring the transmitted light using a fiber-coupled CCD spectrometer (Flame-S, Ocean Optics; Largo, FL).
Quantum chemical computations. Calculations on the 4-tBuC monomer and dimer were carried out with density functional theory (DFT) and its time-dependent variant (TD-DFT) for the excited state. We calculated the ground-and excited-state structures and the excited state O-H dissociation paths. These paths were calculated as relaxed scans, freezing one internal coordinate to a fixed value and optimizing all of the remaining ones. We used the long-range corrected CAM-B3LYP functional, which avoids the appearance of spurious charge transfer excited states (26), and the 6-311G** basis set. These calculations were carried out with Gaussian 16 (27).
To support the identification of the neutral semiquinone radical we also calculated the vibrational frequencies of this species and its electronic spectrum. For the vibrational spectrum, we identified the most stable conformer and benchmarked the CAM-B3LYP/6-311G** level of theory with additional calculations on the phenoxyl radical. The electronic spectrum was calculated at the MS-CASPT2 level of theory (multistate complete active space second order perturbation) with the ANO-L atomic natural orbitals basis set, using Molcas 7.8 (28). More computational details are in Appendix S7.

UV absorption and FTIR spectra of 4-t-butylcatechol in cyclohexane
The UV absorption spectrum of a 6 mM solution of 4-tBuC in cyclohexane exhibits pronounced vibronic structure (Fig. 1a), reminiscent of that of 4-tBuC in the gas-phase (25). As the Scheme 1. Molecules discussed in the text. concentration is increased above~20 mM, the spectrum loses vibronic structure and blue shifts slightly. These high-concentration spectra resemble those of 4-tBuC in the hydrogen-bonding solvents water and acetonitrile ( Figure S1 in Appendix S1).
FTIR spectra were recorded from catechol solutions with concentrations of 5 to 214 mM (Fig. 1b). A dilute, 5 mM solution exhibits two narrow and intense absorption bands at 3576 and 3624 cm À1 . As the concentration is increased to 38 mM, a broad absorption band appears at~3440 cm À1 . Further increasing the concentration results in an even broader and more red-shifted absorption band at~3400 cm À1 . Changes in the FTIR spectra are fully reversible, indicating that the spectral features are not due to irreversible chemical changes. The band assignments are listed in Fig. 1d and possible hydrogen bonding patterns responsible for these frequencies are shown in Fig. 1e. As in past work on cyclohexane solutions of phenol (29-32), a model was used to estimate the fraction of monomers and the fraction of dimers and higher-order aggregates (Fig. 1c).
Transient UV-Vis spectroscopy of monomers and aggregates of 4-tBuC UV-Vis transient absorption measurements were performed on 5 mM solutions (predominantly monomer-containing) and 75 mM solutions (primarily aggregate-containing, see discussion section) of 4-tBuC in cyclohexane with excitation at 265 nm (Fig. 2). These concentrations represent the lowest and highest concentrations that are feasible for our instrumentation. The signals for the 5 mM solution agree very well with ones reported by Horbury et al. (25), while signals of 4-tBuC aggregates are reported here for the first time.
The transient absorption spectrum of the 5 mM sample at the earliest delay times spans our entire detection window from 320 to 650 nm and consists of several broad and overlapping peaks. During the first 50 ps, the signals decay to reveal strong photoinduced absorption (PIA) below 400 nm, which includes a very narrow absorption band at~390 nm. An absorption onset appears below 340 nm. Across the entire measured spectral range, an additional PIA is observed (clearly evident in the 50-60 ps and 100-250 ps traces). This component matches the spectrum recorded from cyclohexane under identical measurement conditions (black curve in Fig. 2a). There is an additional weak, but detectable, absorption band occurring around 600 nm (see 2000-3000 ps trace in Fig. 2a) that is distinct from the spectrum measured from neat cyclohexane.
The 75 mM sample also exhibits a prompt, broad PIA feature, which spans the entire measured spectral region and is very similar to the spectrum recorded at 1-2 ps for the 5 mM solution (Fig. 2b). This feature decays more slowly than in the 5 mM sample. At the longest time delays (i.e. 2000-3000 ps), a broad PIA band is observed in the 320-400 nm spectral range that differs in several respects from the band seen in the 5 mM solution. In particular, the narrow peak centered at 390 nm in the 5 mM solution is not observed in the 75 mM solution. A maximum is also seen in the 75 mM solution near 350 nm, while the signal in the 5 mM solution rises monotonically below 380 nm. Unlike the 5 mM solution, the 75 mM solution exhibits a transient band near 625 nm, which is readily seen between 50 and 250 ps (Fig. 2b).
Kinetic traces for the 5 and 75 mM samples are compared in Fig. 3. The traces were generated by averaging the signals over the indicated wavelength ranges. Three key features in the kinetics of the 5 mM sample are observed. First, there is a fast decaying component present for all three wavelength ranges shown, which is primarily associated with the PIA feature spanning 320-660 nm. Second, the signal in the 350-370 nm trace persists beyond the longest recorded delay time of~3 ns. Third, an additional slowly decaying component can be seen at times greater than~40 ps in the 480-500 nm and 600 to 650 nm traces. Matching decay kinetics are seen over the same time interval when exciting neat cyclohexane under identical experimental conditions (gray curve in Fig. 3a).
The kinetic traces from the 75 mM sample differ in several important aspects from the ones measured for the 5 mM solution. First, the PIA feature spanning 320-660 nm decays more slowly in the more concentrated solution as seen by comparing the 480-500 nm traces in Fig. 3. Second, an additional kinetic component that decays on a hundreds of picoseconds timescale is seen in the 600-650 nm trace of the 75 mM solution, but not in the 5 mM sample. This decay component is likely associated with the band centered at~625 nm that is observed in the 75 mM solution but absent in the 5 mM solution. Last, the signal arising from excitation of the solvent contributes marginally to the observed kinetics for the 75 mM sample because a much smaller path length was used. It is also important to note that the neat solvent signal represents an upper limit to the signal contribution in solutions where the solute absorbs a large fraction of the pump pulse. Similar to the 5 mM sample, there is a persistent signal observable in the 350-370 nm trace.

Time-resolved vibrational spectroscopy of monomers and aggregates of 4-t-butylcatechol
For time-resolved infrared (TRIR) measurements, the minimum concentration that could be studied was 10 mM, and not 5 mM, due to the detection limit of our TRIR spectrometer. It should also be noted that a broad positive absorption signal arising from solvent excitation was removed from the data by subtraction as described in Appendix S3. The results are shown in Fig. 4 for both 10 mM (primarily monomer) and 75 mM (primarily aggregate) solutions of 4-tBuC in cyclohexane.
Immediately after photoexcitation, the 10 mM sample exhibits ground state bleaches (GSBs) at 1290, 1508, 1527, and 1610 cm À1 , and PIAs at 1341, 1406, <1485, 1552, and 1575 cm À1 . The PIA bands decay on a time scale similar to that of the S 1 ?S n absorption feature measured using transient UV-Vis spectroscopy. As these PIA bands decay, a pair of positive bands grow in at 1632 and 1662 cm À1 . Additional weaker PIA bands are observed at 1552 and 1587 cm À1 . These four PIA bands persist out to the nanosecond timescale.
The 75 mM sample exhibits a similar set of negative and positive absorption bands immediately following 265 nm photoexcitation, including GSBs at 1289, 1509, 1526, and 1609 cm À1 , and PIAs at 1339, 1406, 1481, 1548, and 1582 cm À1 . These PIA features decay over tens of picoseconds, or somewhat more slowly than those in the 10 mM sample. An intense, broad PIA band grows in around 1645 cm À1 . Additional broad PIA bands are revealed at <1467 and~1566 cm À1 . These PIA bands persist into the nanosecond timescale. Notably, the 1566 and 1645 cm À1 bands observed in the 75 mM sample are located at approximately the mean position of the corresponding bands observed in the 10 mM sample, between the 1587 and 1552 cm À1 bands, and the 1632 and 1662 cm À1 bands, respectively.
Kinetic analysis of the time-resolved vibrational spectra provides further insight into the population dynamics. The kinetics corresponding to three of the most dominant features in the transient absorption spectra are shown in Fig. 5. The traces in the figure were obtained by averaging over the spectral ranges indicated by boxes of the same color in Fig. 4. For both solute concentrations, the kinetics for the absorbing species that appears immediately upon photoexcitation is isolated by averaging thẽ 1341 and 1406 cm À1 PIA bands to improve signal-to-noise (red curves in Fig. 5). The kinetics for the positive absorbing species at later time delays is obtained from the band present around 1650 cm À1 (blue curves). The recovery of the ground state population for both samples was followed using the GSB peak at 1290 cm À1 , which is devoid of spectral contamination by PIA features.
For the 10 mM sample, the PIA decays biexponentially with a fast~12 ps decay component and a slow~180 ps decay component. The PIA signal at~1650 cm À1 (blue curve) also exhibits biphasic behavior, quantitatively mirroring the two components of the GSB decay kinetics. The GSB signal (black squares, Fig. 5c) does not decay within the entire time window.
For the 75 mM sample, the PIA bands decay more slowly over the first 100 ps than is the case for the 10 mM sample. The kinetics resemble the S 1 ?S n PIA decay kinetics for the 75 mM solution measured using transient UV-Vis spectroscopy (green and red curves in Fig. 3b). The growth kinetics of the 75 mM sample (blue curve in Fig. 5b) also exhibits complex behavior, and the growth at early time delays does not match the decay of the PIA signal (red curve in Fig. 5b) at early time delays. Unlike for the case of the 10 mM solution, the GSB signal of the 75 mM solution decreases in absolute value over the 10-100s of picoseconds timescale, indicating partial recovery of the ground state population. This GSB decay trace (gray squares, Fig. 5c) resembles the decay of the PIA feature (red trace in Fig. 5b).

UV-irradiation of 4-t-butylcatechol solutions
Absorption spectra recorded from samples exposed to continuous UV irradiation from a filtered deuterium lamp for up to 30 min are shown in Fig. 6 for the 5 and 75 mM 4-tBuC solutions. In both solutions, the absorption band at~281 nm decreases and a broad absorption band appears between 350 and 450 nm with increasing irradiation time, demonstrating that photoproducts accumulate in significant quantities after several minutes of UV irradiation.

O-H dissociation and H-atom transfer calculations
Density functional calculations (CAM-B3LYP functional, see Materials and Methods) were performed on the 4-tBuC monomer and dimer to gain insight into the reaction pathways (Fig. 7). Exploratory calculations showed that there are several dimer conformations with different H-bonding patterns within a narrow energy range, but in every structure each of the OH groups participates in an intermolecular hydrogen bond. Here we focus on the structure displayed in Fig. 7b, which is also representative of calculated trimers with similar H-bonding patterns. It has three intermolecular H-bonds, two weakened intramolecular H-bonds compared to the monomer (O-H distances increase from 2.11 to 2.37-2.38 A), and a free O-H group. The calculations suggest that the excitation is localized on a single catechol molecule. Upon relaxation on S 1 , there are significant changes in the intermolecular H-bonds, and one of these bonds stretches from 2.01 to 3.12 A.
In our calculations, we estimated the energetics of the excitedstate dissociation of the free O-H bonds of the monomer and dimer and, as an alternative pathway, photoinduced H-atom transfer along hydrogen bonds. The latter reaction was considered both for H-atom transfer along the intramolecular hydrogen bond in the monomer as well as along the intermolecular hydrogen bond joining two catechol units in the dimer. The calculated activation energy barriers for these pathways, which correspond to weakly avoided crossings between the S 1 and S 2 adiabatic states (33), are collected in Table 1, and the potential energy profiles are described in Appendix S7.
The dissociation reactions of free O-H groups in the monomer and dimer have almost equal activation energy barriers. This suggests that any tunneling contribution to the O-H dissociation (33) will be similar for the two species. In contrast, the activation energy barriers for photoinduced transfer of an H atom along a hydrogen bond, whether this occurs intramolecularly in a single 4-tBuC molecule or between two 4-tBuC molecules connected by an intermolecular hydrogen bond, are~5-7 times higher compared to the dissociation of a free O-H bond.

Self-association and intermolecular hydrogen bonding in 4-t-butylcatechol solutions
At low concentrations, 4-tBuC molecules do not form hydrogen bonds significantly with each other and a gas-phase-like UV  absorption spectrum with sharp vibronic structure is observed (Fig. 1a). As previously observed for phenol (29)(30)(31)(32), 4-tBuC forms aggregates at high solute concentrations in cyclohexane. Aggregates form when the total concentration of 4-tBuC exceeds approximately 5 mM, according to analysis of the electronic and vibrational absorption spectra in Fig. 1a,b. The loss of vibronic structure with increasing 4-tBuC concentration reflects intermolecular hydrogen bonding between 4-tBuC molecules.
FTIR spectroscopy provides a direct means of assessing intramolecular and intermolecular hydrogen bonding through changes in O-H absorption lineshapes. The FTIR spectrum of the 5 mM solution (Fig. 1b) features two narrow and intense absorption bands at 3576 and 3624 cm À1 , which are assigned to the intramolecularly hydrogen-bonded ("b") and free ("f") O-H local stretching modes (Fig. 1e), respectively, based on catechol studies (34)(35)(36). CAM-B3LYP calculations on the monomer support this assignment, giving frequencies of 3826 and 3892 cm À1 for the b and f modes. These values are in good agreement with experiment when a scaling factor of 0.93 is applied. Note that our use of the term, "free O-H", signifies that the O-H group does not donate a hydrogen bond, although it may still accept a hydrogen bond as this does not lead to a frequency shift of the O-H stretch as seen in other hydrogen-bonded alcohols, such as methanol (37). The observation of intramolecular hydrogen bonding of 4-tBuC supports the work of Horbury et al. (25), which deduced that the molecules primarily assume an intramolecularly hydrogen-bonded ("closed") form in cyclohexane. As the concentration is increased much broader lower-frequency bands appear with maxima around 3440 and 3400 cm À1 , indicating the formation of intermolecularly hydrogen-bonded aggregates (29).
Both the f and b bands disappear in unison as the concentration of 4-tBuC is increased. This differs from the behavior observed when a base such as diethyl ether is added to catechol in carbon tetrachloride where the intensity of the free O-H stretching band is diminished before a decrease is observed in the intramolecularly hydrogen-bonded band (38). This is because the base initially hydrogen bonds to the free O-H group, and the intramolecular hydrogen bond remains intact. A similar phenomenon was observed when adding DMSO to 3,5-di-t-butylcatechol in carbon tetrachloride (39).
The difference in behavior may be because the addition of a 4-tBuC monomer to another monomer or aggregate not only causes the loss of a free OH bond, but also simultaneously significantly perturbs the intramolecular hydrogen bond. The calculated dimer structure shown in Fig. 7 shows how this might happen. In this structure, one O-H group on each molecule forms a bifurcated hydrogen bond: The intramolecular hydrogen bond weakens and becomes longer, but the same hydrogen participates in a second, stronger intermolecular hydrogen bond (see also the bifurcated aggregate structure in Fig. 1e). The latter effect is expected to redshift the frequency of the O-H group that formerly participated only in intramolecular hydrogen bonding. Consistent with this idea, CAM-B3LYP calculations for the dimer shown in Fig. 7b give stretching frequencies for the two bifurcated O-H groups of 3412 and 3468 cm À1 , applying the correction factor of 0.93 suggested by the calculations for the monomer.
The self-association of 4-tBuC in cyclohexane leads to spectral changes in the O-H stretching region of the FTIR spectrum (Fig. 1b) that closely resemble changes arising from the aggregation of phenol in cyclohexane (29,31). Extensive modeling of cyclohexane solutions of phenol has revealed a complex equilibrium between monomers, dimers, and multimers (29)(30)(31)(32). Adapting from these models, the fractions of monomers and aggregates containing at least two 4-tBuC molecules were determined (Fig. 1c). Further details are provided in Appendix S3.

UV-irradiation of 4-t-butylcatechol solutions
The photoproducts of UV-irradiated 4-tBuC were identified using time-dependent UV-Vis spectroscopy (see Fig. 6) in order to assist interpretation of the time-resolved spectroscopy results. The Supporting Information includes an expanded analysis of the  time-dependent absorption spectra and further details about the experimental design. In both the 5 and 75 mM samples, new redshifted absorption bands appear over time. These bands are assigned to o-quinones, the fully oxidized form of catechol (40)(41)(42). This assignment is supported by mass spectrometry measurements made with the 5 mM sample ( Figure S9 in Appendix S5). Note that the quinone bands in the 5 and 75 mM samples are centered at 370 and 400 nm, respectively. The redshift and broadening of this band is attributed to hydrogen bonding between the quinone and catechol species (see Figure S7 in Appendix S4). The quantum yields of quinone formation were determined to be 2.5% and 0.4% for the 5 and 75 mM solutions respectively (see the Supporting Information for details). Despite the differences in transient absorption signals observed for 4-tBuC monomers and aggregates in cyclohexane, photoexcitation of each leads to similar photoproducts, albeit with a lower quantum yield for the 75 mM solution.

Photochemistry of 4-t-butylcatechol monomers
We first discuss the time-resolved spectroscopy of the 4-tBuC monomer because its electronic spectroscopy aids analysis of the photochemical dynamics of the aggregates, which are presented here for the first time. The solutions were excited at 265 nm, a wavelength known to directly populate the S 1 state, which has 1 pp* character at the Franck-Condon geometry (25). The PIA feature appearing at 1-2 ps in the 5 mM sample (red trace in Fig. 2a) was previously assigned by Horbury et al. (25) to S 1 ? S n absorption by 4-tBuC. This PIA spectrum evolves into a feature below 400 nm possessing a sharp peak at~390 nm, which Horbury et al. (25) assigned to a neutral 4-tBuC semiquinone radical (structure shown in Scheme 1). This feature resembles the sharp band observed at 383 nm in the absorption spectrum of the neutral semiquinone radical of catechol in liquid paraffin (43). A weak feature around 600 nm is also expected for the semiquinone radical based on the absorption spectrum of the phenoxyl radical (44). MS-CASPT2 calculations predict absorption peaks at 349 nm and 583 nm for the 4-tBuC semiquinone radical (see Table S3 in Appendix S7), supporting our assignment. The additional absorption onset appearing below 340 nm is tentatively assigned to a further photoproduct (see below). The UV photochemistry of 4-tBuC monomers in cyclohexane mirrors that of catechol (33,45) and phenol (46,47). Excitation to the S 1 state, which has 1 pp* character initially, causes O-H dissociation via tunneling to the dissociative 1 pr* state (33). This occurs at a weakly avoided crossing where S 1 changes from 1 pp* to 1 pr* character. Horbury et al. (25) observed that photoinduced O-H bond dissociation of 4-tBuC upon 267 nm excitation occurs with a time constant of 18 AE 1 ps in cyclohexane and 4.9 AE 0.6 ps in the gas phase. A somewhat shorter time constant of 12.15 AE 0.03 ps was observed in our study for 4-tBuC in cyclohexane solution following 265 nm excitation (see Fig. 2 and Supporting Information for a detailed target analysis). The slower dissociation time in ref. (25) could be due to the seven-fold higher solute concentration compared to ours (35 vs 5 mM). According to our analysis of the data in Fig. 1, a concentration of 35 mM is high enough to have a significant population of aggregates, which result in more slowly decaying signals (see below).
Despite using moderate excitation conditions (i.e. 620 lJ cm À2 ), the long path length of 500 lm needed to obtain an appreciable UV optical density for the 5 mM sample increases the signal contribution from solvent ionization (48). This produces a PIA feature spanning the entire spectral range measured for the 5 mM sample, which is identified in a neat cyclohexane solution measured under identical conditions (black trace in Fig. 2a). The solvent-only signal is successfully accounted for in the target analysis model used to obtain the time constant for O-H dissociation in the 5 mM solution (see Figure S14 in Appendix S8 for further details). Because of this, and because its signal strength contributed only modestly to the total observed sample signal, we did not subtract it prior to kinetic analysis.
The TRIR spectrum of the 4-tBuC monomer in cyclohexane exhibits several prompt PIA features (see the red trace in Fig. 4a) that decay with a time constant of~12 ps, a value that matches the lifetime of the S 1 ?S n feature determined from transient UV-Vis spectroscopy. As such, we assign these vibrational features to the S 1 electronic state. Note that this S 1 vibrational spectrum of 4-tBuC resembles that reported for phenol in cyclohexane (46). The occurrence of these intense, low-frequency PIA bands (relative to the C=C ring stretching GSB bands) is reflective of a break in ring planarity (25,49,50) of the S 1 state of 4-tBuC.
As the PIA bands of the S 1 state decay, new PIA features appear around 1662, 1587, and 1552 cm À1 . The most prominent band at 1650 cm À1 has an initial formation time constant of~12 ps, which matches the decay time of the S 1 absorption bands observed in the TRIR spectrum, and the O-H dissociation time constant determined using transient UV-Vis spectroscopy. Combining the vibrational and electronic data, we assign the features at 1662, 1587, and 1552 cm À1 to neutral semiquinone radicals of 4-tBuC, formed from O-H bond dissociation. Our DFT calculations (see Appendix S7) support this assignment. The formation kinetics of the~1662 cm À1 band exhibits biexponential growth, with a slower component occurring on the tens to hundreds of picosecond timescale in addition to the fast~12 ps component.
There are three possible explanations for this kinetic behavior. First, we considered that hot ground electronic states might produce the slowly decaying positive feature observed near 1650 cm À1 . However, the GSB kinetics at 1290 cm À1 in the 10 mM solution (see Fig. 5c) exhibit practically no change in intensity across the entire time window, indicating that the depleted ground state population remains constant and is not restored via internal conversion. Additionally, temperature-dependent FTIR experiments (see Figure S17 in Appendix S9) show that the probed C=C ring stretching mode at~1610 cm À1 is insensitive to temperature change and cannot explain the positive feature at~1650 cm À1 .
Second, we considered the possibility that aggregates contribute to the spectrum because approximately 25% of the 4-tBuC molecules in the 10 mM sample are present as dimers or higher aggregates. The S 1 PIA decay kinetics and the 1662 cm À1 growth kinetics are biexponential, with weighting factors of 0.75 and 0.25 for the fast and slow components respectively. The relative amplitude of the fast and slow component is in excellent agreement with the fraction of monomer (75%) and aggregates (25%) present in 10 mM solution, based on our self-association modeling (see Fig. 1). The fast component decays with a time constant of~12 ps in agreement with the lifetime of the 4-tBuC monomer observed in the transient UV-Vis experiment. The longer-lived component decays on a similar timescale as the excited singlet state in the aggregates (see 75 mM data in Figs. 2  and 3). This evidence thus suggests that there is a small population of longer-lived excited states of aggregates present in the 10 mM solution that contributes to the total TRIR spectrum. In this case, the slow growth of the 1662 cm À1 band is assigned to UV-excitation of 4-tBuC aggregates.
Third, it was considered that some of the ejected H-atoms might recombine geminately with semiquinone radicals, producing cyclohexadienone photoproducts like the ones shown in Scheme 1. If these adducts possess vibrational spectra similar to that of the semiquinone radical, then the slower growth at 1650 cm À1 could simply reflect the time scale of geminate recombination, which commonly occurs on the tens to hundreds of picoseconds timescale (51)(52)(53)(54). For example, the formation of cyclohexadienone via recombination is observed after photodissociation of phenol in cyclohexane. Harris et al. (46) assigned a feature at 1671 cm À1 in an 8 mM solution of phenol dissolved in cyclohexane to either a cyclohexa-2,4-dienone or a cyclohexa-2,5-dienone. Isolation of unsubstituted cyclohexadienones for spectroscopic study is very difficult due to their instability (55,56). Nonetheless, prior experimental and computational reports (55,57) suggest that they contain intense vibrational peaks around 1670 and 1636 cm À1 , corresponding to C=O and C=C stretching modes (57). In addition, the FTIR spectrum of a brominated cyclohexa-2,5-dienone derivative ( Figure S18 in Appendix S10) has a pair of peaks that resembles the 1000-3000 ps TRIR spectrum of the 10 mM sample, although they appear at somewhat higher frequencies (1675 and 1685 cm À1 ) likely due to different aromatic substituents.
In the transient UV-Vis spectrum, the absorption onset below 340 nm for the 5 mM solution (Fig. 2a) is not seen in the neutral semiquinone radical (43). Instead, this absorption onset qualitatively matches that of the brominated cyclohexa-2,5-dienone derivative (see Figure S18 in Appendix S10). Previous pulse radiolysis studies of OH adducts of phenol (58), catechol (59), and 4-tBuC (60) reveal that their hydroxycyclohexadienyl radicals, which are similar in chemical structure to cyclohexadienones, possess absorption spectra similar to those of cyclohexa-2,4-dienone (61) and cyclohexa-2,5-dienone (62) derivatives. The absorption spectrum of each of these compounds also qualitatively matches the absorption below 340 nm in monomers of 4-tBuC. However, while cyclohexadienone formation explains the second growth component seen for thẽ 1662 cm À1 PIA feature, it cannot explain the slow decay of the S 1 PIA features at~1341 and~1406 cm À1 (red trace in Fig. 5a). We therefore favor the second explanation above in which a minor amount of aggregates results in a slowly forming population of semiquinone radicals. Regardless of whether dimers or cyclohexadienones contribute to the PIA band at~1650 cm À1 , all of the PIA bands in the 1000-3000 ps trace in Fig. 4a, including the~1650 cm À1 feature, must be at least partially attributable to neutral semiquinone radicals of 4-tBuC.

Photochemistry of 4-t-butylcatechol hydrogen-bonded aggregates
Transient absorption by the initially photogenerated state in the 75 mM solution resembles that observed in the 5 mM solution (compare 1-2 ps spectra in Fig. 2). Based on these similarities and our TD-DFT results, we propose that the initially photogenerated S 1 state in aggregates is a locally excited singlet state, which is minimally perturbed by neighboring molecules. This PIA feature decays more slowly in the aggregates, evolving on the tens of picoseconds timescale into a red-shifted spectrum with the longest wavelength peak appearing beyond 600 nm. This indicates structural relaxation of the excited state. At later times, the spectrum features a peak around 355 nm that persists for nanoseconds. This feature resembles that observed for the neutral semiquinone radical in the 5 mM sample, but lacks the sharp band at~390 nm. The appearance of semiquinone radicals indicates that O-H dissociation also occurs in aggregates. The assignments of these features will be discussed later. The first obvious difference in the transient spectra is that the S 1 feature decays more slowly in the 75 mM solution compared to the 5 mM solution, with a fast time constant of at least~24 ps and a slow time constant of at least~154 ps (depending on the target model used; see Figures S15 and S16 in Appendix S8). Excited state relaxation in aggregates occurs more slowly than O-H dissociation in monomers, suggesting that differences in the aggregate excited state potential energy landscape hinder this reaction. However, the lifetime of the S 1 feature (~154 ps, see Figure S15 in Appendix S8) in the 75 mM solution is still much shorter than the lifetime of 1.7 ns measured by Horbury et al. for 265 nm excitation of 4-tBuC dissolved in acetonitrile (25), indicating that nonradiative decay proceeds at a higher rate in the hydrogen-bonded aggregates.
The second obvious difference in the spectral dynamics for the 75 mM solution is that the initial PIA feature evolves on the tens of picoseconds timescale, revealing a low-energy peak at~625 nm (see 50-60 ps spectrum in Fig. 2b). We assign this new spectrum to locally excited S 1 states that have undergone structural relaxation, leading to a change in electronic density in the ring system, based on the following considerations. First, this spectrum resembles that of the initially photogenerated S 1 state absorption. Second, its spectral shape is highly reminiscent of the S 1 absorption spectrum of dilute solutions of phenol in cyclohexane excited at 265 nm (47) or 267 nm (46). Phenol contains one fewer hydroxy group (Scheme 1), suggesting that there is less electron density on the ring than in 4-tBuC. Third, the S 1 state of guaiacol in cyclohexane has been shown to have an absorption peak beyond 600 nm (63). Guaiacol contains an ortho-positioned hydroxy and methoxy group (Scheme 1), suggesting that the aromatic ring may possess an electron density in-between that of phenol and 4-tBuC. Altogether, we propose that the locally excited S 1 states in 4-tBuC aggregates relax because of structural rearrangements within the hydrogen-bonded aggregates, changing the S 1 ?S n spectrum. Consistent with our spectroscopic assignment, the calculated changes in O-H hydrogen bond distances between the ground and excited states of 4-tBuC dimers (Fig. 7b) indicate that the aggregates may undergo significant conformational rearrangements after excitation.
Note that phenol excimers have been reported to absorb in the visible spectral region (47), however we do not observe a clear signature of excimers in 4-tBuC aggregates. This difference in behavior can be rationalized by considering the much longer S 1 lifetime (i.e. 2.1 ns) of the phenol monomer, and the even longer lifetime of the phenol excimer (47). In 4-tBuC, excimer formation is unlikely to compete kinetically with the decay of S 1 , which has a lifetime approximately one order of magnitude shorter than in analogous phenol aggregates. Furthermore, substantial rearrangements of the molecules are expected in order to transition from the predicted ground state geometry (e.g. the dimer structure in Fig. 7b) to the expected excimer geometry, which in the case of phenol involves a pseudo-parallel p-stacked geometry (64).
The TRIR spectrum of the 75 mM solution (see Fig. 4) closely resembles that of the 10 mM solution, but the decay and growth of the vibrational features are slower. In particular, the vibrational spectrum measured 1 ps after photoexcitation contains the same PIA bands seen in the 10 mM sample, which are assigned to the S 1 electronic state. These bands decay on the tens-to-hundreds of picoseconds timescale, without significant changes in lineshape. Excimers and exciplexes have been shown in the past to produce TRIR spectra with frequencies and lineshapes that are readily distinguishable from those of locally excited or ionic states (65)(66)(67). In particular, the broadened distribution of nuclear geometries that accompanies excimer or exciplex formation yields significantly broadened vibrational spectra (65)(66)(67). Because the vibrational bands do not exhibit any peak shifting or broadening during the timescale in which the excited electronic state relaxes (several tens of picoseconds), it is unlikely that excimers are formed. Instead, it is more likely that the excited singlets of the 4-tBuC aggregates remain as locally excited states during structural relaxation.
The TRIR spectra of the 75 mM sample at later delay times resemble the 1000-3000 ps spectrum of the 10 mM sample, except with significantly broadened absorption peaks at approximately 1566 and 1645 cm À1 . The similar vibrational frequencies of the bands observed for both the 10 and 75 mM samples suggest common transient species, namely, semiquinone radicals of 4-tBuC. Furthermore, both monomer and aggregate solutions produce o-quinones in the UV-irradiation experiments, suggesting that the semiquinone radical is a common reaction intermediate. The broadening of the semiquinone vibrational spectrum is proposed to be a consequence of aggregation, whereby the radical species experience intermolecular hydrogen bonding like ground state molecules. Note that an analogous vibrational peakbroadening phenomenon has also been reported for exciplexes in solution (66).
As indicated in Fig. 1c, the 75 mM solution contains~25% monomers and so it is reasonable to believe that the semiquinones may form exclusively from monomers. It is difficult to discern the monomer contribution to the transient UV-Vis spectra (Figs. 2 and 3) both visually and using target analysis due to multiple overlapping features. However, the TRIR kinetics of the semiquinone band at~1645 cm À1 (blue trace in Fig. 5b) clearly show a continuous growth of its population over the entire time window. Such behavior cannot be explained by monomers, which yield a single lifetime of~12 ps.
Transient absorption spectroscopy clearly reveals that O-H dissociation is slower in 4-tBuC aggregates than in monomers. While the FTIR results suggest that the aggregates are composed of extended hydrogen-bonded networks of 4-tBuC molecules that exhibit hydrogen bonds to both O-H groups, it is instructional to consider a 4-tBuC dimer. Our TD-DFT calculations on the dimer predict a~5-fold increase in activation energy barrier for H-atom transfer compared to the dissociation of a free O-H bond. Poterya et al. considered analogous pathways in their computational study of hydrogen-bonded clusters of phenol in the gas phase (68). Their calculations predict a larger barrier for the Hatom transfer pathway (1.0 eV) than for the dissociation of a free H-bond (0.9 eV), although the difference in barriers is much less pronounced than in our calculations for 4-tBuC. In fact, the barrier height of 0.51 eV (Table 1) calculated for H-atom transfer along an intermolecular H bond is too large to explain the appearance of the semiquinone radical in our measurements in no more than several hundred ps. Of course, tunneling under the barrier could take place, but further study is warranted of how hydrogen bonding in 4-tBuC aggregates retards OH bond dissociation.
Our observations are consistent with a previous study of phenol dimers, which attributed the observed increase of the S 1 lifetime to a decrease in O-H dissociation efficiency compared to monomers (47). The authors considered that hydrogen bonding between phenol molecules may cause the repulsive S 2 state ( 1 pr*) to become bound at larger O-H bond distances (47), based on calculations of the excited state of dimeric water (69). Our experimental and computational evidence for S 1 structural relaxation, in addition to the measured partial GSB recovery in 4-tBuC aggregates, suggests that intermolecular hydrogen bonding hinders O-H dissociation by a similar mechanism. Nevertheless, O-H dissociation still takes place in 4-tBuC aggregates, and further study is warranted to determine whether dynamic fluctuations of the hydrogen bonds or other factors regulate this process.
We tested two possible kinetic models for describing excited state processes in 4-tBuC aggregates using target analysis (see Figures S15 and S16 in Appendix S8). In the sequential decay scheme, the S 1 states undergo structural relaxation preceding O-H dissociation. In the parallel decay scheme, O-H dissociation Scheme 2. Proposed kinetic models for monomers and aggregates of 4-tBuC. 1 (4-tBuC)* relax represents a structurally relaxed S 1 state. The time constants shown are from target analysis of the transient UV-Vis data. occurs only from S 1 states that have not had time to structurally relax. While both schemes fit the data equally well, we propose that only the sequential pathway shown in Scheme 2 is adequate. According to this model, structural relaxation of the S 1 state occurs with a time constant of 23.6 AE 0.2 ps, and is followed by competitive deactivation via internal conversion (154 AE 2 ps) or O-H bond dissociation (190 AE 20 ps). Using these time constants, 55% of the excited population is predicted to decay by internal conversion in excellent agreement with the loss of 57% of the initial bleach amplitude seen in the TRIR GSB signal in Fig. 5c. On the other hand, the parallel model is less satisfactory as it predicts that 81% of the population decays by internal conversion.
Examination of the excited state decay times of 4-tBuC monomers and aggregates provides further insights. Horbury et al. (25) proposed that because the ground electronic state of 4-tBuC is planar, the break in ring planarity of the S 1 state of 4-tBuC results in symmetry-enhanced tunneling that promotes O-H bond dissociation. This phenomenon is favored in cyclohexane, in which 4-tBuC exists predominantly in an intramolecularly hydrogen-bonded (or "closed") geometry (25). In contrast, 4-tBuC exists predominantly in a non-intramolecularly hydrogen-bonded (or "open") geometry in polar solvents like acetonitrile, resulting in a planar S 1 state that is unlikely to undergo O-H dissociation (25).
Our FTIR results support their work by providing direct evidence for intramolecular hydrogen bonding in 4-tBuC monomers in cyclohexane. Our transient UV-Vis data demonstrate that 4-tBuC aggregates in cyclohexane possess a longer S 1 lifetime than monomers, by at least an order of magnitude. Variations in the O-H stretch lineshape with concentration (Fig. 1b,c) suggest that molecules in aggregates exist in a slightly less "closed" form in the ground state than monomers. Additionally, the changes in hydrogen bond distances predicted by 4-tBuC dimer computations (see Fig. 7) suggest that intramolecular hydrogen bonds are weakened in aggregates. This change in ground state conformation relative to the monomer leads to the prediction of a more planar S 1 state, which could further inhibit O-H dissociation due to closer symmetry to the S 0 state. These observations form an additional potential mechanism for inhibiting O-H bond dissociation and allowing internal conversion to proceed in aggregates (recall the observation of partial ground state recovery in Fig. 5c).

CONCLUSION AND OUTLOOK
To date, most spectroscopic work has investigated the photochemistry and photophysics of monomeric models of eumelanins (19,22,25) and small oligomer and polymer models (20,21), in addition to eumelanin itself (23,70). Additionally, solventmediated excited state deactivation has been studied (20). While the participation of solvent on the excited state dynamics following photoexcitation of these model systems is generally understood, less attention has been paid to understanding how direct intermolecular interactions between eumelanin units affect excited state relaxation. Past studies have emphasized that covalent dimerization and oligomerization of eumelanin building blocks is important for shortening their excited state lifetimes in aqueous solution (20,21). However, our results show that turning on intermolecular interactions from monomer to aggregates of 4-tBuC elongates its excited state lifetime. Because dihydroxy functionalized subunits in the eumelanin polymer may be located within domains that are hardly influenced by the solvent, alternative excited state deactivation pathways and their photochemical consequences should be considered.