Alkylated green fluorescent protein chromophores: dynamics in the gas phase and in aqueous solution †

Fluorescent labelling of macromolecular samples, including using the green fluorescent protein (GFP), has revolutionised the field of bioimaging. The ongoing development of fluorescent proteins require a detailed understanding of the photophysics of the biochromophore, and how chemical derivatisation influences the excited state dynamics. Here, we investigate the photophysical properties associated with the S 1 state of three alkylated derivatives of the chromophore in GFP, in the gas phase using time-resolved photoelectron imaging, and in water using femtosecond fluorescence upconversion. The gas-phase lifetimes (1.6–10 ps), which are associated with the intrinsic (environment independent) dynamics, are substantially longer than the lifetimes in water (0.06–3 ps), attributed to stabilisation of both twisted intermediate structures and conical intersection seams in the condensed phase. In the gas phase, alkylation on the 3 and 5 positions of the phenyl ring slows the dynamics due to inertial effects, while a ‘pre-twist’ of the methine bridge through alkylation on the 2 and 6 positions significantly shortens the excited state lifetimes. Formation of a minor, long-lived ( c 40 ps) excited state population in the gas phase is attributed to intersystem crossing to a triplet state, accessed because of a T 1 /S 1 degeneracy in the so-called P-trap potential energy minimum associated with torsion of the single-bond in the bridging unit connecting to the phenoxide ring. A small amount of intersystem crossing is supported through TD-DFT molecular dynamics trajectories and MS-CASPT2 calculations. No such intersystem crossing occurs in water at T = 300 K or in ethanol at T E 77 K, due to a significantly altered potential energy surface and P-trap geometry.


Introduction
2][3] The optical absorption and emission properties of GFP are dictated by a chromophore based on the p-hydroxybenzylidene-2,3-dimethylimidazolinone anion (pHBDI À , Fig. 1) that is situated within the b -barrel structure of the protein. 4Although the Aequorea victoria jellyfish is an uncommon organism, the desirable optical properties of GFP and derivative fluorescent proteins, and the ease with which they can be deployed as optical markers in biochemical systems, has resulted in widespread adoption in photobiology, 5,6 leading to the 2008 Nobel Prize in Chemistry. 7Since the discovery and first demonstrations of GFP in photobiology, significant efforts have been devoted to understanding the detailed photophysics of the chromophore, 8 the role of the protein environment in the photophysics, and the influence of solvation on the chromophore's excited state dynamics. 9,10This understanding drives the development of improved fluorescent proteins and analytical toolkits for monitoring cellular processes. 11 notable feature of pHBDI À photophysics is that fluorescence in solution is very weak; pHBDI À has a fluorescence quantum yield of E10 À4 in water at room temperature, compared with E0.8 in the protein. 13,14The low fluorescence quantum yield for the chromophore in solution is generally attributed to the facile internal rotation of the phenoxide ring and Z-E isomerisation, with both pathways leading to internal conversion. 15,16In the protein environment, these rotations are hindered by steric and non-covalent interactions from the photoreceptor binding pocket. 17,18Attempts to observe gas-phase fluorescence from pHBDI À at T E 300 K and T E 100 K using the LUNA2 instrument at Aarhus University, including by some of the present authors, have been unsuccessful. 19,20Synthetic modification of the pHBDI chromophore to explore structure-function relationships has resulted in numerous analogues offering altered spectral properties and excited state dynamics.0][31] In one investigation, Conyard et al. 12 synthesised a series of alkylated pHBDI derivatives postulating that, in solution, the additional steric bulk should hinder both Z-E isomerisation of the methylene bond and internal rotation of the single bond.In principle, alkylation (weakly electron donating) should only perturb the electronic structure minimally and, thus, have little influence the intrinsic photophysics of the chromophore.
There have been a range of gas-phase studies on pHBDI À , mostly performed at T E 300 K.The principal appeal of studies in the gas phase is to inform on intrinsic photochemical dynamics, which can be compared with theoretical simulations in a straightforward manner.Where deviations exist, the experimental data provides direction for theoreticians to refine their models.Gas-phase investigations include using action spectroscopy strategies such as photodissociation, [32][33][34][35][36][37][38] photodetachment/photoelectron, 24,39,40 and photoisomerisation. 41When taken together, these studies have characterised many of the fundamental photophysical properties of pHBDI À , including the underlying absorption spectrum, electron binding energies and autodetachment processes, propensity for reverse Z-E photoisomerisation, and the impact of microsolvation.Time-resolved photoelectron and photodissociation strategies have shown that the lifetime of the S 1 state is several picoseconds at T E 300 K. 39,[42][43][44] Computational studies seeking to understand specific photophysical properties have been applied to several pHBDI analogues, both in isolation and in solvated environments, with alterations including halogenation [45][46][47] and alkylation 37,48,49 on the phenyl ring.Notably, there have been numerous molecular dynamics and potential energy surface investigations on pHBDI À assuming gas-phase, solution, and protein environments (see ref. 44, 50 and 51 and references therein), with the goal of establishing a theoretical understanding of the intrinsic chromophore dynamics versus environmental influences.The recent, and most robust, simulations by Martı ´nez and co-workers 15,16 used ab initio molecular dynamics with multi-state CASSCF wavefunctions to propagate the excited state dynamics of pHBDI À in isolation and also in (explicit) water.These studies highlight the importance and interplay of two S 1 /S 0 conical intersections, one associated with torsion of the phenoxide moiety (P-torsion) and the other, leading to Z-E isomerisation, associated with torsion of the imidazolone moiety (I-torsion).Both pathways contribute to deactivation of the excited state but have distinct lifetimes; in the gas phase E40% of excited molecules undergo internal conversion through P-torsion, while E60% do so through I-torsion.Timeresolved experiments are presumed to measure the combination of these pathways.Significantly, the simulations predicted that the excited-state lifetimes for pHBDI À in water are shorter than in the gas phase, due to both a stabilisation of twisted intermediate structures and a lowering in energy of the conical intersection seam for internal conversion.From a synthetic perspective, it is interesting to incorporate various functionalities onto the ring systems in order to 'tune' the conical intersection energies and topologies. 52Harnessing such synthetic control of conical intersection properties is key for the rational design of chromophores offering improved fluorescence and/or isomerisation quantum yield.
This paper details time-resolved measurements performed in the gas-phase and in water on the four chromophores shown in Fig. 1.The chromophores were probed in the gas phase using time-resolved photoelectron imaging, and in water using femtosecond fluorescence upconversion.Significantly, we found that alkylation on the 3 and 5 positions increases the excited-state lifetimes in both the gas phase and in aqueous solution compared with pHBDI À , with the gas-phase lifetimes being substantially longer than those in solution.The latter of these trends is at odds with many other organic chromophore systems, where the gasphase excited state dynamics are more rapid than those in solution due to the absence of vibrational energy transfer to solvent.We present evidence for a weak intersystem crossing (ISC) pathway in the gas phase due to a S 1 /T 1 degeneracy at the so-called P-trap on the excited-state potential energy surface; this degeneracy is lost in solution due to substantial changes to the potential energy surface associated with solvation.

Gas-phase photoelectron spectroscopy
Photoelectron spectroscopy (imaging) on the target anions was performed using an instrument that combines electrospray ionisation, ion trapping and thermalisation to T E 300 K, time-offlight m/z separation, and velocity-map imaging detection. 53,54riefly, anions electrosprayed from methanol solution were accumulated in a radio frequency ring-electrode ion trap.The trapped anions were unloaded at a 333 Hz repetition rate into a set of colinear time-of-flight optics that accelerated the ions along a 1.3 m flight region toward a continuous-mode penetrating field velocity-mapping assembly. 54Laser light was timed to interact with the mass-selected ion packet at the centre of the velocity-map imaging stack.Ejected photoelectrons were velocity-mapped onto a dual (chevron) multichannel plate detector, followed by a P43 phosphor screen, which was monitored with a charge-coupled device camera.Velocity-map images were accumulated with a 500 ns gate pulse applied to the second multi-channel plate.The velocity-mapping resolution was DE E % 5%, and the electron kinetic energy (eKE) scale was calibrated from the spectrum of I À .Velocity-map image reconstructions used antialiasing and polar onion-peeling algorithms, 55 providing the photoelectron spectra and associated photoelectron angular distributions, quantified by the conventional b 2 anisotropy parameter. 56Values of b 2 for atomic orbitals range between +2 and À1, with the limits corresponding to a cos 2 y (parallel) and a sin 2 y (perpendicular) ejected electron distribution relative to the incident laser polarisation vector, respectively.However, b 2 values for electron ejection from molecular orbitals typically span a reduced range.Higher order b 4 parameters were not statistically significant in this study.Single-colour photoelectron spectra to obtain electron detachment parameters were performed at hn = 3.44 eV (360 nm) using light from a Continuum Horizon OPO laser coupled with a Continuum Surelite Nd:YAG pump laser, E1 mJ pulse À1 (unfocussed).This photon energy was chosen based on earlier work demonstrating predominantly direct photodetachment (i.e.non-resonant) electron ejection in the spectral region between the S 1 and S 2 states. 37,39or the time-resolved experiments, femtosecond laser pulses were derived from a Spectra-Physics Ti:sapphire oscillator and regenerative amplifier.The hn = 2.50 eV (495 AE 5 nm, E20 mJ) pump pulses were produced by fourth-harmonic generation (two successive BBO crystals) of the idler output from an optical parametric amplifier (Light Conversion TOPAS-C).The photon energy of the pump pulse was chosen to be close to the maximum in the action spectra (i.e.vertical excitation energy) for the target anions. 37This choice of energy allows for direct comparison with the reference ab initio molecular dynamics studies, 16 which assume trajectories from the vertical excitation geometry.The hn = 1.55 eV (800 nm, E100 mJ) probe pulse was the fundamental output of the femtosecond laser.Pump and probe pulses were delayed relative to each other (Dt) using a motorised delay line.The pump and probe pulses were combined colinearly using a dichroic mirror and were loosely focused into the interaction region using a curved metal mirror.The pump-probe cross correlation was E60 fs; the cross-correlation was measured in a non-linear crystal and is consistent with fits to the data.
It is worth noting that previous experiments on pHBDI À photophysics following excitation of the S 1 state have been reported using the instrument described above, 42 although these earlier experiments used an older ion trap that resulted in ion vibrational temperatures greater than 300 K (likely T = 400-500 K).Since publication of that study, the ion trap was upgraded to achieve thermalisation to T E 300 K. 57

Time-resolved fluorescence upconversion
Aqueous anionic solutions of the chromophores were prepared to a concentration corresponding to a maximum absorbance of E0.1 over the S 1 ' S 0 band.Deprotonation was achieved using a drop of 1 M NaOH, and the solution spectrum (pH E 10) was confirmed to correspond to the anion by comparing neutral and anionic absorption spectra (there is a 4 70 nm red-shift upon deprotonation). 12ime-resolved fluorescence upconversion measurements on the target anions in water were performed using the instrument described in ref. 58.In brief, E800 nm light (E20 fs pulses at 80 MHz) was generated using a continuous wave Nd:YVO 4 laser driving a Kerr lens mode-locked Ti:sapphire oscillator.Focussing of the fundamental light (E840 mW), using a 150 mm focal length concave mirror, into a 50 mm thickness barium borate crystal (BBO, type I) produced the second harmonic (400 nm, E11 mW).Separation of the 800 nm and 400 nm light was achieved with a dichroic mirror, and the delay between the pump and probe light pulses was adjusted using a computercontrolled motorised delay stage (0.1 mm resolution).Temporal broadening along the beam path was minimised using chirped mirrors.A concave mirror was used to focus the 400 nm pump pulse onto the sample, contained in a static 2 mm path length quartz cell.The resulting fluorescence was focussed by a reflective microscope objective (15Â magnification) through a CG455 Schott filter and was frequency mixed (upconverted) with the 800 nm light in a 100 mm thickness BBO crystal (type I).The upconverted signal was passed through a UG11 Schott filter and monochromator (Photon Technology International Model 101, resolution 2 nm mm À1 ) prior to entering the low-noise photomultiplier (PMT, Hamamatsu R585).The PMT response was acquired by a computer-interfaced Stanford Research Systems photon counter (SR400).The instrument cross correlation was characterised at E45 fs through up-conversion of Raman scattering.

Computational details
Static electronic structure calculations were performed using the Gaussian 16.B01 and ORCA 5.0.3 software packages. 59,602][63] Optimised microsolvated geometries were computed at the MP2/aug-cc-pVDZ level of theory 64 with five explicit water molecules combined with the SMD implicit solvation model. 65The initial microsolvated geometries were determined by placing water molecules around the chromophore in several configurations, followed by optimising the geometries of the waters with a PM6 semi-empirical Hamiltonian. 66Vertical excitation energies of the lowest energy solvated structures were computed at the DLPNO-STEOM-CCSD/aug-cc-pVDZ level of theory. 67olecular dynamics trajectories for pHBDI À were carried out with Newton-X 68,69 coupled to Gaussian 16, using TD-DFT with the CAM-B3LYP functional and the 6-311G(d,p) basis set.The S 1 /T 1 degeneracy and the spin-orbit coupling was analysed for a representative trajectory with a long residence time in the Ptrap (see ESI †).Trajectories were propagated on the S 1 state with a step size of 0.5 fs, and terminated when the S 1 /S 0 gap was smaller than 0.15 eV.Initial conditions were obtained from Wigner sampling at T = 0 K and vertical excitation.MS-CASPT2 calculations were carried out with a (16,14) active space (16 p-electrons in 14 porbitals), and the ANO-S basis set (3s2p1d contraction for C, H, and N; 2s1p contraction for O) using Molcas 8.2. 70,71Three roots were calculated by state-averaging with equal weights.The default ionisation potential electron affinity (IPEA) correction of 0.25 au 72 and an imaginary level shift 73 of 0.1 au were applied.The singlet and triplet states were obtained in separate calculations.The spinorbit coupling elements were obtained from the MS-CASPT2 perturbationally-modified complete active space configuration interaction (PM-CASCI) wavefunctions.Benchmarking has shown that the CASPT2 method combined with the IPEA shift can provide pp* excitation energies and relative triplet and singlet energies to better than 0.14 eV (mean absolute energy). 74

Electron detachment parameters
Single-colour photoelectron spectra recorded at hn = 3.44 eV (360 nm) for the four anions are shown in Fig. 2. The adiabatic (ADE) and vertical detachment energy (VDE), given in parentheses in Table 1, were determined from the spectral peak maximum and the high-kinetic-energy edge, respectively (by taking the difference between the excitation energy, 3.44 eV, and the respective maximum/edge values).The experimental values are consistent with the ADE and VDE values computed at the DLPNO-CCSD(T)/aug-cc-pVTZ level of theory as given in Table 1.Our experimental values for pHBDI À (ADE = 2.72 AE 0.05 eV, VDE = 2.87 eV) are in close accord with the T 4 300 K photoelectron spectroscopy values determined by Horke et al., 75 and with a T E 20 K detachment threshold measurement at 2.73 AE 0.01 eV. 76Measured b 2 values over the photoelectron bands are negative (Fig. 2b), consistent with prompt detachment of a p-electron from a planar (or near planar) anion.
For pHBDI À , it is known that the electron detachment threshold (2.72 AE 0.05 eV in this work) occurs over the S 1 ' S 0 absorption band (maximum at 2.56 AE 0.02 eV). 37,40,77While our photoelectron spectra for the alkylated chromophore anions reveal that the electron detachment parameters are lower in energy than for pHBDI À , the maximum in the absorption spectrum for each is 2.49 AE 0.02 eV, 37 meaning that much of the absorption band (at T E 300 K) is situated above the detachment threshold and has shape resonance character 78 -this may have implications for the excited-state lifetimes.

Time-resolved photoelectron spectroscopy
Results from the time-resolved photoelectron imaging experiments on the four target anions are shown in Fig. 3.The total pumpprobe signal for each anion was fit with a three-component model: , where the fitted contributions from component 1 are blue, 2 are orange, and 3 are grey.Lifetime t 3 corresponds to loss of component 3. Inset in Fig. 3a and d are example timeresolved photoelectron spectra corresponding to DtE0 fs (associated with fit component 1, blue) and to DtE10 ps (containing signatures from components 2 and 3, grey); time-resolved spectra for pHBDI À are consistent with those reported by Mooney et al. 42 Our fittings assumed an initial Gaussian cross-correlation function (E60 fs) convoluted with exponential decay functions (and growth functions for t 2 and t 3 ).The extracted t 1 and t 2 lifetimes from the fits are summarised in We interpret the gas-phase excited-state lifetimes and dynamics in accord with the potential energy surface shown in Fig. 4a, which was adapted from ref. 16.Briefly, the potential energy surface involves two torsion coordinates, both of which lead to a conical intersection.The torsion coordinates correspond to twisting of either the single bond (P-torsion, phenoxide ring twisting) or the double bond (I-torsion, imidazolinone ring twisting) of the methine bridge.The P-torsion coordinate passes through a shallow S 1 minimum, termed the P-trap, where some fraction of the excited-state population may be temporarily contained.The P-trap has a highly twisted geometry, with the single-bond methine bridge dihedral angle of E1021. 16rthermore, the nearby conical intersection in this region has a 'sloped' (and uphill) topology, giving the tendency for excited state motions to undergo several crossings (and vibrations through the P-trap) before internal conversion.On the other hand, the I-torsion co-ordinate is associated with small barrier (e.g.][81] In our fits to the gas-phase time-resolved data, lifetime t 1 corresponds to rapid motion away from the Franck-Condon geometry (i.e.nuclear relaxation on S 1 with a concerted change in probe/photodetachment cross-section), and lifetime t 2 represents the loss of this relaxed population through the combination of internal conversion (major) and autodetachment (minor) processes.t 3 will be further discussed in Section 3.3.The ratio of rate constants for deactivation of the excited state by internal conversion (IC) and autodetachment (AD) has been estimated at k AD /k IC E 0.05 AE 0.01 (T 4 300 K, invariant to pump photon energy in the 2.39-2.57eV range), 43 implying that internal conversion outcompetes autodetachment.Fluorescence is deemed a negligible deactivation channel because, as outlined in the introduction, all attempts at probing gas-phase fluorescence on the four target anions at T E 300 K and T E 100 K were unsuccessful.It is also worth noting that calculations of oscillator strengths indicate that the P-trap is essentially non-fluorescent in the gas phase due to substantial torsion, and is weakly fluorescent in solution due to a smaller torsion angle. 15,16The similarity in t g 1 (where superscript background corrected) at Dt E 0 fs (blue) and after several picoseconds (grey) when the spectra have a significant contribution from the long-lived component 3 (Dt E 10 ps) -time-resolved eKE distributions at longer Dt for 35Me À and 35Bu À had poorer signal-to-noise.g implies gas phase) for pHBDI À and 35Me À suggests that the additional methyl groups have little effect on the prompt dynamics, i.e. the initial twisting away from the Franck-Condon geometry is fast.Conversely, the E1 ps difference in lifetime t g 2 between pHBDI À and 35Me À suggests passage through the conical intersections in 35Me À is more hindered, presumably due to inertial effects and the intramolecular vibrational energy redistribution required to access the conical intersections.The longest recorded lifetimes were for 35Bu À , with both t g 1 and t g 2 exhibiting a two-fold increase relative to pHBDI À , attributed to steric bulk and inertial effects from the tert-butyl groups (such inertial effects become quenched in solution).26Me À has the shortest lifetimes, particularly t g 1 , which is limited by the laser cross-correlation.This result is consistent with the excited-state lifetimes recorded for 26Me À in ethanol solution, 12 where the short lifetime was attributed to the non-planarity of the ground-state geometry, i.e. the geometry is already pre-twisted (P-twist angle E201 in the gas phase, see ESI †) and, thus, closer to the conical intersection geometry.This pre-twist is because of steric interaction between the methyl groups and close-lying H-atoms.A similar lifetime shortening was induced by methyl substitution at the pHBDI À central carbon, which also leads to a pre-twisted ground state. 82revious photoelectron spectroscopy measurements of S 1 excited-state lifetimes for pHBDI À (T 4 300 K) were t 1 = 330 fs and t 2 = 1.4 ps. 42The present t g 2 lifetime is roughly three-fold longer, which is attributed to some combination of: (a) differences in ion vibrational temperature between the experiments, and (b) that the earlier study assumed a bi-exponential fit of the data that did not account for the very long time component, t 3 (further discussed in Section 3.3).The time-resolved photoelectron spectroscopy experiment described in this work used an improved version of the instrument in ref. 42 incorporating a new ion trap in which ions are thermalised to T E 300 K. Thus, we consider the present gas-phase lifetimes more reliable room temperature measurements.In a different time-resolved photodissociation study utilising an ion storage ring and pumping at 480 nm (2.58 eV compared with 2.50 eV in this work), 44 reported lifetimes of t 1 = 1.3 AE 0.2 ps, t 2 = 11.5 AE 0.5 ps (T E 300 K), and the longer lifetimes at T E 100 K of t 1 = 4.6 AE 2.1 ps, t 2 = 27 AE 2 ps, and t 3 = 1.2 AE 0.1 ns assuming a tri-exponential model.No t 3 parameter was fitted to the T E 300 K data, presumably because signal level was very low.The authors assigned the t 3 E 1.2 ns lifetime at low temperature to a small fraction of the excited-state population becoming held in the Ptrap (Fig. 4a), which slowly escapes through the I-torsion coordinate.In the present work, we were unable to measure the lifetime of the long-lived component at T E 300 K due to experimental limitations, rather we give it as t 3 c 40 ps.To our knowledge, there are no reports of long-lived excited-state lifetimes for any pHBDI À derivatives in solution (aside from hot ground state dynamics).

Intersystem crossing
The minor, long-lived excited-state lifetime (t 3 ) observed for all four of the pHBDI-based anions could originate from: (1) relaxation of a small amount of S 1 population that becomes confined in the P-trap (i.e.insufficient energy to access any conical intersection), or (2) intersystem crossing (ISC) to a lower-lying triplet state (T 1 ).We Fig. 4 Summary of gas-phase pHBDI À photophysics (similar potential energy surfaces apply for the alkylated derivatives): (a) potential energy surface for pHBDI À based on molecular dynamics results adapted from ref. 16.P-torsion corresponds to rotation of the phenoxide group about the single bond on the methine bridge, and I-tosion represents rotation of the imidazolinone group about the double bond on the methine bridge.The simulations on pHBDI À by List and co-workers 15,16 determined that E40% (E34% in water) of the excited-state population passes through the S 1 minimum to reach the P-torsion S 1 /S 0 conical intersection, while E60% (E66% in water) of the excited population passes through the I-torsion S 1 /S 0 conical intersection.We propose that a small amount of intersystem crossing occurs in the S 1 minimum along the P-torsion coordinate (the so-called P-trap, where the methine bridge dihedral angle is E1021).(b) Jablonski diagram assigning the origins of the three excited-state lifetimes extracted from the time-resolved photoelectron spectra.Anions are excited to S 1 by a hn = 2.50 eV pump pulse, and the excited-state population is probed as a function of time (Dt) with 1.55 eV laser pulses.The excited-state lifetimes t 1 , t 2 , and t 3 reflect probing of the initial prompt relaxation from the Franck-Condon geometry (component 1), loss of relaxed S 1 population (component 2), and loss of T 1 population (component 3), respectively.Lifetimes t 1 and t 2 are on the order of femtoseconds to picoseconds, while t 3 is hundreds of picoseconds or longer because T 1 is situated well below the detachment threshold and because of the electron reconfiguration required to achieve the electron configuration for the neutral D 0 state.Note, in (b) the energy of the D 0 state varies with geometry, but is shown as a straight line for simplicity.
discuss these two possibilities below, referring to the Jablonski diagram shown in Fig. 4. Note that, although the Jablonski diagram in Fig. 4 is specific to gas-phase pHBDI À , the overall features apply to the alkylated derivatives.
In the time-resolved photodissociation study by Andersen and co-workers, 44 a long-lived excited-state component at T E 100 K was ascribed to possibility (1), where population in the P-trap eventually either overcomes excited-state barriers to reach the double-bond torsion (I-torsion, Fig. 4) conical intersection, or undergoes fluorescence.The subsequent ab initio molecular dynamics study by List et al. 16 highlighted the importance of the alternative, P-torsion conical intersection, close to the P-trap geometry, which E40% of population accessed at T E 300 K.In terms of the present data that provide the extra dimension of timeresolved electron kinetic energy distributions, if we assume the Ptrap interpretation for lifetime t 3 , our calculations at the DLPNO-STEOM-CCSD/aug-cc-pVDZ level of theory (assuming the P-trap geometry from ref. 16), predict a time-resolved eKE maximum at E0.8 eV.This is somewhat higher in energy than the maximum in the eKE distribution shown in Fig. 3a (inset, grey trace).Furthermore, assuming the P-trap interpretation, it is unclear why we would observe such a long timescale component in our T E 300 K experiment, considering that the internal energy exceeds the height of the barriers on the S 1 potential energy surface and autodetachment is always an open channel because the total internal energy exceeds the detachment threshold.
The second interpretation involves intersystem crossing (ISC) to form the T 1 (triplet) state, with lifetime t 3 corresponding to vibrational autodetachment from T 1 .We argue for this mechanism for several reasons.Qualitatively, P-torsion (Fig. 4a) from the S 1 Franck-Condon geometry, leading to the S 1 minimum (P-trap), effectively 'decouples' the phenoxide ring from the rest of the molecule.Once decoupled, the molecule can be thought of as two non-conjugated units, consisting of the negatively-charged phenoxide ring and the neutral imidazolinone.pHBDI À is well known to undergo a moderate degree of charge-transfer on excitation, 37 which, in this decoupled form, results in a phenoxy radical and an imidazolinone radical anion.5][86] Thus, near degenerate singlet and triplet states are expected at the P-trap geometry.Quantitatively, the S 1 /T 1 separation at the P-trap geometry, determined from MS-CASPT2 calculations at the TD-DFT optimised structure (see ESI † for details), is 0.01 eV, suggesting an ISC seam is accessible (or nearby).Furthermore, the energy profile from a molecular dynamics trajectory that stays for approximately 2 ps in the P-trap shows a small average S 1 /T 1 energy gap (0.32 AE 0.25 eV) -see ESI.† The spin-orbit coupling (SOC) element, calculated at the MS-CASPT2 level of theory for 10 points of the trajectory with the smallest S 1 /T 1 energy gap, is 0.15 AE 0.03 cm À 1 .This non-negligible SOC is consistent with a small ISC probability (e.g. a few percent as suggested from the experimental relative signal intensities) for trajectories staying in the P-trap over a timescale of the observed t 2 lifetimes.While our hybrid strategy provides qualitative evidence for ISC, it would be desirable in future work to deploy an ab initio molecular dynamics method able to directly describe intersystem crossing and could thus determine an ISC quantum yield.We note that other molecular dynamics simulations on gas-phase pHBDI À derivatives (ref.16,  50 and 51) did not consider the possibility of ISC.
Further evidence for the ISC interpretation comes from the time-resolved eKE distributions.The global minimum on the T 1 surface has a planar geometry, and our DLPNO-STEOM-CCSD/aug-cc-pVTZ calculations predict that the time-resolved eKE distribution from the relaxed T 1 state should give rise to a peak at 0.61 eV, which is in good agreement with the experimental data (Fig. 3a, inset, grey trace peaks at E0.6 eV).The T 1 potential energy minimum (planar) was calculated to lie 1.9 eV above the S 0 equilibrium geometry, and E0.4 eV below the P-trap geometry.It is worth noting that the probe (800 nm) photodetachment cross-sections from the S 1 state (P-trap) and T 1 state (planar equilibrium geometry) will likely differ, meaning that absolute quantification of the extent of ISC is difficult; we estimate the ISC contribution at E5%.However, the assigned ISC dynamics appear common to the four target chromophore anions, implying similar potential energy surfaces (Fig. 4).

Excited state lifetimes in water
To compare our gas-phase excited-state lifetimes with those recorded in solution, we studied the target anions in water using time-resolved fluorescence upconversion pumping at E400 nm.Aqueous absorption spectra are shown in Fig. 6a, with each chromophore exhibiting strong absorption at the pump wavelength.The corresponding fluorescence spectra, when exciting at E400 nm, are shown in Fig. 6b.Spectra recorded at T E 77 K in ethanol are given in the ESI, † and relevant spectral properties are summarised in Table 3. Fits of the upconverted fluorescence decay curves at T = 300 K, shown in Fig. 6c, involved a twocomponent exponential decay function (with the exception of 26Me À ).The fitted lifetimes are given in Table 2, with the data for pHBDI À closely matching earlier upconversion data, 87 while the other anions are similar to data recorded in ethanol 12 -the fitted lifetimes in ethanol and water are within 0.05 ps (t 1 ) and 0.4 ps (t 2 ) of each other.We interpret lifetime t 1 as dominated by rapid nuclear motion away from the Franck-Condon geometry convoluted with prompt solvent reorganisation, and lifetime t 2 as linked to loss of S 1 population.Because the fluorescence emission spectrum does not show any substantial variation with pump-probe delay, i.e. fluorescence lifetimes are independent of the up-converted wavelength over the emission spectrum, we interpret the fluorescence emission to occur from a broad range of geometries.This model is consistent with the idea that the P-trap potential energy minimum in solution is substantially less twisted from the Franck-Condon geometry compared with in the gas-phase and is a shallow minimum, with molecular dynamics simulations showing facile dihedral angle twisting. 15A less twisted P-trap geometry in water is associated with reduced biradical character and, consequently, the intersystem crossing seam will be shifted away from the P-trap geometry.The shorter t 2 lifetimes in solution compared with the gas phase infer that access to S 1 /S 0 conical intersection(s) is more favourable in solution.While this trend was predicted for pHBDI À from ab initio molecular dynamics, 15,16 the present work shows that same stabilisation of the twisted intermediate and conical intersections occurs for the alkylated derivatives.There is no evidence for any ISC in solution (at either T = 300 K or T E 77 K, discussed below), which would manifest as phosphorescence in the emission spectra and a long-lifetime component (phosphorescence or delayed fluorescence) in the time-resolved measurements.
The short t 1 lifetime for 26Me À (limited by cross correlation) parallels the gas-phase results, and is linked with the pre-twisted Franck-Condon geometry along the P-torsion coordinate. 12Our calculations of the Franck-Condon geometry reveal a P-twist angle for 26Me À at E201 in the gas phase and E481 in water (see ESI †).Given that, for aqueous pHBDI À , the P-torsion conical intersection occurs at a P-twist angle of E731, 15 and assuming a similar case for 26Me À , the twisted Franck-Condon geometry for 26Me À (E481) means that little motion is required to reach the P-torsion conical intersection in 26Me À .
The fluorescence emission spectra for aqueous 35Me À and 35Bu À are red-shifted and slightly narrower compared with that for pHBDI À .The red-shifted emission is attributed to stabilisation of the S 1 excited state, and a narrowing is presumably associated with restricted torsion of these alkylated molecules in their fluorescent state (consistent with longer excited-state lifetimes). 29The much broader emission spectrum for 26Me À is consistent with emission from a wider range of geometries and the shortest excited-state lifetime.
The fluorescence emission spectra for aqueous 35Me À and 35Bu À are red-shifted and slightly narrower compared with that for pHBDI À .The red-shifted emission is attributed to stabilisation of the S 1 excited state, and a narrowing is presumably associated with restricted torsion of these alkylated molecules in their fluorescent state (consistent with longer excited-state lifetimes). 29The much broader emission spectrum for 26Me À is consistent with emission from a wider range of geometries and the shortest excited-state lifetime.
Finally, we can compare our fluorescence upconversion lifetimes with those recorded for pHBDI À in water using femtosecond stimulated Raman spectroscopy (FSRS). 88The t 1 component is similar in both cases, as it reflects initial relaxation away from the Franck-Condon geometry.On the other hand, the t 2 component is longer when measured using FSRS (t 2 E 2.1 AE 0.1 ps) compared with that measured by upconversion (t 2 E 1.19 AE 0.18 ps).Presumably, the longer FSRS lifetime, if real, is linked with some fraction of excited state population which does not access fluorescent geometries.

Cryogenic measurements
To explore the influence of temperature on the absorption and fluorescence emission spectra of the target anions, we recorded fluorescence excitation, emission, and lifetime data at T E 77 K by immersing a cryogenic EPR tube containing an ethanolic solution in a liquid nitrogen bath.Cryogenic measurements in water are not possible due to volume expansion upon freezing and the formation of a poor quality glassy matrix.Wavelengths of maximum response from the fluorescence excitation (as proxies for absorption) and emission spectra are given in Table 3 (see ESI † for the T E 77 K spectra).The Stokes shifts and fitted excited-state lifetimes from the cryogenic measurements are given in the ESI.† The Stokes shifts in water at T = 300 K are substantially larger than those in ethanol, particularly for 35Bu À , which we attribute to a combination of hydrogenbonding stabilisation of the deprotonation site, which is more easily accessed in water due to the smaller molecular size of the solvent and reduced steric interaction with the alkyl groups, and because of the high dielectric constant effect from water.
Excited-state lifetimes at T E 77 K in ethanol are 2.7-3.1 ns.Significantly, there was no evidence for T 1 state formation in either the time-resolved traces or the emission spectra.Considering all data together, we conclude that component 3 with lifetime t 3 in the gas-phase time-resolved spectra is most likely associated with the T 1 state.A similar ISC process is absent in solution due to substantial changes in the potential energy surface with solvation 15 perturbing the S 1 /T 1 geometry in the P-trap.Future efforts might like to investigate pHBDI À derivatives including heavy atoms facilitating ISC, provided that the heavy atom functionality does not perturb the P-trap and T 1 state degeneracy.

Conclusions
We have reported time-resolved experiments, in both the gas phase and in aqueous solution, on a series of anions linked to the chromophore found in green fluorescent protein.The gasphase lifetimes for the S 1 state span 1.6-10 ps, while those in aqueous solution are shorter at 0.06-0.3ps.These anions present an intriguing case where the gas-phase dynamics are slowed compared to those in solution, a situation which contrasts with many other organic ions because, in the gas phase, nuclear motion is not hindered and there is no vibrational energy redistribution to the environment.The excited-state dynamics were interpreted in terms of a dual conical intersection potential energy surface possessing a shallow and twisted S 1 potential energy minimum (P-trap).Significantly, through molecular dynamics simulations, we identified an intersystem crossing seam that is degenerate with the P-trap, facilitating a small amount of intersystem crossing (ISC), consistent with a minor, long-lived (c40 ps) component in the gas phase measurements.No similar ISC was found in aqueous solution (or seen in any other solution-based study on pHBDI À ); this absence was attributed to a significant geometry change and stabilisation of the P-trap and possibly conical intersections in solution.
There are several avenues for future work.First, time-resolved photoelectron imaging measurements should be repeated at cryogenic temperatures, allowing increased spectral resolution of the electron kinetic energy distributions and also to prolong S 1 lifetimes, e.g. by exciting on the red-edge of the absorption band.Efforts to increase the S 1 excited state lifetime should result in an increase in the fraction of the population undergoing ISC.Second, the gas-phase Z-E photoisomerisation experiments on pHBDI À , reported by some of the current authors, 41 should be extended to these alkylated anions, with efforts to quantify (relative) Z-E photoisomerisation quantum yields.When such measurements are interpreted in conjunction with ab initio molecular dynamics simulations similar to those performed by Martinez and coworkers, 16 the fraction of the excited-state population passing through each conical intersection seam could be inferred.There is also the possibility to couple femtosecond pump-probe spectroscopy to such an ion mobility experiment, potentially allowing the lifetimes connected with isomerising and non-isomerising excitedstate populations to be disentangled.Third, it would be interesting to perform synthetic modifications to the six-membered ring (e.g. through addition of heavy atoms) to increase ISC coupling strengths, allowing for clear investigation of the ISC dynamics.
Table 3 Wavelength maxima (in nm) for the absorption (l abs ) and emission (l em ) spectra of pHBDI-based anions in water (T = 300 K) and in ethanol (T = 300 K and T E 77 K), and excitation (l ex ) maxima in ethanol.

Fig. 5
Fig. 5 Natural transition orbitals 83 for the S 1 ' S 0 transition in pHBDI À : (a) the P-trap minimum energy geometry, (b) schematic illustrations of the resulting ground and excited-state electronic configurations.

Fig. 6
Fig. 6 Aqueous spectroscopy of the anionic pHBDIs at T = 300 K: (a) Absorption spectra.(b) Fluorescence emission spectra (exciting at E400 nm, corresponding to the time-resolved pump wavelength).(c) Timeresolved fluorescence upconversion data, fit with a two-component kinetic model (individual data points shown for pHBDI À ).All spectra are normalised.

Table 1
Calculated (experimental values in parentheses with AE0.05 eV uncertainty) adiabatic (ADE) and vertical (VDE) detachment energies to D 0 , and from the triplet, T 1 , to D 0 , for the target anions.D E is the difference in energy between the ground-state (S 0 ) Z and E isomers, with the Z isomer being the most stable in all cases.ADE values for the T 1 state assume the planar, optimised T 1 geometries.All values are in eV and calculations were at the DLPNO-CCSD(T)/aug-cc-pVTZ level of theory

Table 2 ,
with both sets of lifetimes having the trend: 26Me

Table 2
Fitted excited-state lifetimes (in ps) from gas-phase timeresolved photoelectron imaging (g), and fluorescence upconversion in water (w) and ethanol (eth, taken from ref. 12) a 26Me À data could not be fit with a double exponential; t w journal is © the Owner 2023 Phys.Chem.Chem.Phys., 2023, 25, 23626-23636 | 23631 Uncertainties are AE1 nm JNB.LB and GRL thank the Ministerio de Ciencia, Innovacio ´n y Universidades (Spain) for project PID-2019-104654GB-I00 and the Red Espan ˜ola de Supercomputacio ´n for computational time (project QSB-2018-1-0040).EKA acknowledges a University of East Anglia Doctoral Studentship.Electronic structure calculations were carried out on the High Performance Computing Cluster supported by the Research and Specialist Computing Support service at the University of East Anglia.