Observation of dissipative chlorophyll-to-carotenoid energy transfer in light-harvesting complex II in membrane nanodiscs

Plants prevent photodamage under high light by dissipating excess energy as heat. Conformational changes of the photosynthetic antenna complexes activate dissipation by leveraging the sensitivity of the photophysics to the protein structure. The mechanisms of dissipation remain debated, largely due to two challenges. First, because of the ultrafast timescales and large energy gaps involved, measurements lacked the temporal or spectral requirements. Second, experiments have been performed in detergent, which can induce non-native conformations, or in vivo, where contributions from homologous antenna complexes cannot be disentangled. Here, we overcome both challenges by applying ultrabroadband two-dimensional electronic spectroscopy to the principal antenna complex, LHCII, in a near-native membrane. Our data provide evidence that the membrane enhances two dissipative pathways, one of which is a previously uncharacterized chlorophyll-to-carotenoid energy transfer. Our results highlight the sensitivity of the photophysics to local environment, which may control the balance between light harvesting and dissipation in vivo.

I n green plants, the light-harvesting machinery is a complex network of multiple antenna complexes that absorb sunlight and funnel the solar energy to the reaction center, where charge separation takes place to initiate the chemical reactions of photosynthesis 1,2 . In parallel to its primary light-harvesting functionality, the protein network has evolved to react sensitively to fluctuating light conditions in order to prevent the generation of deleterious photoproducts. In the presence of excess light, the network transitions reversibly and rapidly from a fully light-harvesting to a photoprotective state, where harmful excess energy is dissipated as heat in a process called non-photochemical quenching [3][4][5] . The individual antenna complexes exhibit photophysics that include energy transfer, dissipative, and deleterious pathways. The timescales and amplitudes of these pathways are known to vary with conformation for these complexes. This complexity, along with the complexity intrinsic to a multi-protein network, has made it difficult to determine the balance of energy transfer and dissipation, as well as the underlying mechanisms.
The antenna complexes are membrane proteins that bind a dense network of primary (chlorophylls, Chls) and accessory (carotenoids, Cars) light-harvesting pigments. The electronic interactions between the Chls and the Cars give rise to rapid and efficient energy transfer, which provides the power for chemical reactions [6][7][8] , and, in parallel, dissipative pathways. However, proposals as to the nature and dynamics of these pathways vary widely. The four primary proposals are (1) energy transfer from the Chl Q y to the Car S 1 state 9,10 ; (2) excitonic states constructed from a linear combination of the Chl Q y and the Car S 1 states 11,12 ; (3) charge transfer from the Car S 1 to the Chl Q y state 13,14 ; and (4) charge transfer among Chls 15,16 . In the first two proposals, the short-lived dark Car S 1 state mediates dissipation, which has a picosecond-order lifetime 17 . While the first proposal had been suggested as the most likely pathway, previous ultrafast experiments were unable to observe energy transfer 18,19 . Instead, the measured dynamics were consistent with an excitonic state, leading to the development of the second proposal 11,12 . However, the ambiguity of the Car S 1 energy due to its low oscillator strength has made it difficult to characterize these two proposals. The third proposal is supported by spectroscopic signatures of the Car radical cation 20 , although their small amplitude has prevented clear analysis. In the fourth proposal, the states with charge transfer character are thought to appear as redshifted fluorescence peaks 15,16 , yet recent results indicate that the redshifted and the quenched species are distinct 21 . This series of observations and their associated limitations highlights the challenges in understanding the photophysics in green plants.
The primary antenna complex in green plants is lightharvesting complex II (LHCII), and therefore its photophysics have been the most extensively characterized. Previous investigations on LHCII suggested that a conformational change of the antenna complexes is an important trigger for the transition into the dissipative state 10,[22][23][24][25] . This transition is thought to leverage the sensitivity of the electronic interactions to the relative orientation and distance between the Chls and Cars, and so various conformational changes of the Cars have been proposed 10,22,23 . Several strategies were used to induce conformational changes, involving dramatically different local environments for LHCII ranging from crystals 22,26 to protein aggregates 10,27 to whole leaves 10 . While the results provided some insight into dissipative pathways, the multiplicity of environments is a contributor to the multiplicity of proposed conformational and photophysical mechanisms of photoprotection. For example, the fluorescence lifetime of LHCII has been reported to be different in a lipid environment as compared to in detergent micelles 28,29 . The in vitro environments, which employ detergent or crystallization, may introduce additional, non-native conformational changes that could alter or even denature the functional structure of membrane proteins [30][31][32] . In contrast, in vivo spectroscopy on whole leaves provides physiological information [33][34][35] . However, identifying the photophysical pathways in each of the homologous antenna complexes is not possible. Furthermore, in vivo transient absorption measurements have been shown to inevitably lead to laser-induced artifacts, such as singlet-singlet annihilation in the measured photophysics due to the large absorption cross-section of the intact protein network 34,36 . Due to these challenges and limitations, a simple, yet physiological environment has been lacking, leaving the photophysical pathways of individual antenna complexes undetermined.
In this work, we benchmark the photophysics of individual LHCIIs in a membrane disc, known as a "nanodisc", using ultrabroadband two-dimensional electronic spectroscopy (2DES). In nanodiscs, the membrane protein of interest is embedded in a discoidal lipid bilayer membrane, providing a well-controlled, near-native membrane environment without the complexity of the intact protein network 37,38 . Our experiments show differences of up to 40% in the energy transfer timescales between the two environments, including an enhancement of two dissipative pathways in the membrane. Conformational changes of two Cars at the periphery of the LHCII trimer increase energy transfer to the dissipative Car S 1 state via two parallel pathways, rapid internal conversion from the Car S 2 state and energy transfer from Chls. While the latter energy transfer pathway had been proposed based on indirect evidence, we report direct observation of this dissipative pathway. Furthermore, the measured subpicosecond timescale implies energy transfer between strongly coupled states. Our results demonstrate the ability of the local environment to control the photophysical pathways in LHCII, which may be used to balance light harvesting and dissipation in the native thylakoid membrane.

Results
Membrane-induced conformational changes in LHCII. The linear absorption spectra of LHCII (Fig. 1a) in detergent and in nanodiscs confirm its successful incorporation into nanodiscs in intact trimeric form, based on similar overall peak location and profiles (Supplementary Note 4, Supplementary Table 2 and Supplementary Fig. 11). A closer inspection of the spectra shows subtle changes in peak position and/or intensity in the Car S 2 states (470-510 nm) as well as the two Q y bands of Chls (640-690 nm), suggesting changes in the arrangement of both the Cars and Chls resulting from introduction of the membrane environment.
Circular dichroism (CD) spectra of LHCII in detergent and in discs provide a sensitive measure of the spatial configuration of pigments bound to the complex, because CD peak shape and intensity are directly related to the mutual orientation of the transition dipoles and the strength of their interactions 39,40 . Comparison of the CD spectra reveals two differences involving two peripheral Cars, neoxanthin (Neo), and lutein 1 (Lut1, Fig. 1b, c). First, the relative peak intensities between 474 and 492 nm (494 nm) change, which has been reported to originate from the interactions between the Soret band of the Chls b and Neo ( Fig. 1d and Supplementary Note 5, Supplementary Fig. 12) 30,39,41 . A similar change in the peak ratio was previously observed in LHCII nanodiscs 42 . Second, the negative 492 nm peak redshifts by 2 nm, reported to originate from the interactions between the high-energy lutein (Lut1) and the Soret band of Chl a612 39 . Thus, the observed changes point to alterations in the spatial arrangement of Neo and Lut1 caused by the membrane. In contrast, we do not observe any difference in the CD signal at 500-510 nm, where the lower-energy lutein (Lut2) absorbs. CD in the Chl Q y region (Fig. 1e) reveals a slight broadening of the 653 nm peak and a 2 nm redshift of the 682 nm peak, attributed to excitonic interactions between Chl a604-Chl b606 and Chl a610-Chl b608, and between Chls a611 and a612, respectively 39 . These are the Chls that are strongly coupled with Neo (Chl a604, Chl b606, Chl b608) and Lut1 (Chls a610, a611, a612) 43 , illustrated as domains I and II, respectively, in Fig. 1b, c. Given that Neo and Lut1 are the two Cars impacted upon incorporation into the membrane, we speculate that the observed changes in the rotational strengths of the Chls could arise from changes in their excitonic interactions with the neighboring Cars, rather than independent structural reorganization of the Chls in the membrane.
Perturbation of domain II, which contains the three lowestenergy Chl a pigments that form the emissive locus (Chls a610, a611, a612) 44 , is further supported by a reduction of fluorescence in the membrane. The steady-state fluorescence quantum yield and fluorescence lifetime are reduced in the membrane discs by 17% and 18%, respectively (Supplementary Note 6, Supplementary Figs. 13, 14 and Supplementary Table 3). The slight quenching of the fluorescence upon membrane insertion is consistent with previous results on LHCII nanodiscs 28 . The observed fluorescence lifetime (2.8 ns) is still significantly longer than that measured in vivo (<2 ns) 45 or in crystals (1 ns) 22 , suggesting additional interactions are present in these systems due to the presence of multiple antenna complexes.
Domains I and II are located at the periphery of the trimeric LHCII complex (Fig. 1a, b). Compared to the counterparts located closer to the core that are shielded by the surrounding pigments and protein matrix, these domains are more exposed to the lipid bilayer. Thus, they are more susceptible to structural changes induced by the membrane, consistent with our results. In particular, a significant part of the conjugated chain of Neo protrudes outward from the protein matrix, which may allow severe twisting of the chain by environmental interactions. Such a distortion in the conjugated chain of Neo has, in fact, been predicted theoretically 24 .
Energetics and ultrafast dynamics of the peripheral Cars. Ultrabroadband 2DES was employed to determine the impact of the membrane on the photophysical pathways in LHCII. By using a laser spectrum with a significantly broader bandwidth than that in conventional 2DES 46 , we map out energy transfer and dissipation across the broad range of Car and Chl excited states. Supplementary Fig. 15, Supplementary Note 7 shows a representative ultrabroadband 2D spectrum of LHCII with the main spectral features labeled. Figure 2a compares the 2D spectra of LHCII in the detergent and the membrane environment (T = 533 fs) in the frequency range of the Car S 2 states. Two major changes are observed. The first is increased transfer of the Car S 2 population into the dark S 1 state (S 2 → S 1 internal conversion), which results in decreased energy transfer to the lower-lying Chls, the competing pathway . Lut2 is located at the inner core of the trimeric LHCII ( Fig. 1a, b), and thus relatively protected from direct exposure to the protein-lipid interface, as mentioned earlier. This may be the origin of its environment-independent dynamics. The Car-Chl cross peaks directly visualize energy transfer from the Car S 2 to the lowerlying Chl Q states, and so further report on Car S 2 dynamics. The cross-peak intensities decrease by 35% in the membrane (Fig. 2d, e), consistent with the increased S 1 to S 2 ratio shown in Fig. 2b.
The second major change is a blueshift of the Car S 1 ESA bỹ 200 cm −1 , indicating that the S 1 → S N energy gap increases in the membrane (Fig. 2a,  . This blueshift can originate from either a redshift in S 1 energy or a blueshift in S N energies. The former is more likely, because S N is a broad manifold of multiple higherlying states that are unlikely to all shift in a correlated manner, especially given the environment-independent transition energy of the S 2 state. This energy level shift is an environment-induced static effect present at all waiting times, separate from a dynamic shift due to vibrational cooling of the hot S 1 state 48,49 . We do additionally observe dynamic shifts in the S 2 −S 1 zero-crossing frequency in the initial 500 fs, where the contribution from vibrational cooling is significant (Supplementary Note 7, Supplementary Fig. 21). These dynamic effects are independent of environment. In contrast to the S 1 → S N transition, no energy shift is observed for the S 2 states.
Along with the changes in spectral features, we observe an acceleration of the decay of the S 1 population of Neo/Vio (54%) and Lut1 (53%) in the membrane (Fig. 2c and Supplementary  Fig. 18  NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-15074-6 ARTICLE originate from two different processes: a decrease in the S 1 −S 0 energy gap, which speeds up non-radiative decay, or an increase in energy transfer to the energetically close-lying Chl Q y states, which accelerates the depletion of the S 1 population 11 . We attribute the acceleration of the decay to the former mechanism, faster non-radiative decay, based on two results. First, an increase in energy transfer from Car S 1 to Chl Q y would result in an increase in magnitude of the Car-Chl cross peaks on the timescale of the S 1 decay, and no such feature is observed. Second, the S 1 state likely redshifts in the membrane, as discussed above. Consistent with the trend observed in the S 1 to S 2 ratio, the kinetics of Lut2 is independent of environment (Supplementary Note 7, Supplementary Fig. 18).
Chl b to Chl a energy transfer. The relaxation dynamics of the Chls reveal two prominent changes in the membrane environment (Fig. 3, Supplementary Note 7). First, the energy transfer from Chl b to Chl a 50-52 is slowed down in the membrane (Fig. 3a- . The timescales of the energy transfer pathways, obtained by fitting the initial rise time of the cross peaks, become longer in the membrane, from 80(±20) to 132(±22) fs (Chl b → Chl a H ) and from 130(±20) to 225(±20) fs (Chl b → Chl a L ), indicating a 39-42% reduction in the energy transfer rates and resulting in diminished cross peak intensities in the membrane. The same trend is observed in the kinetics of the Chl b diagonal peak, which decays 40% slower in the membrane due to the decreased rate of energy transfer to Chl a (Fig. 3b). The energy transfer between the high-energy and low-energy Chl a pools (Chl a H and Chl a L ) is also slowed down, but to a much lesser extent (14%, Supplementary Fig. 23, Supplementary Note 7). The specific pigment structural changes responsible for the observed deceleration of Chl b → Chl a energy transfer cannot definitively be identified. Although LHCII is thought to compact overall in the membrane environment as compared to in a detergent micelle, the slower Chl b → Chl a energy transfer observed here suggests that the specific pigments involved actually move further apart. As discussed above, several Chl bs form a strongly coupled pigment cluster with Neo (domain I in Fig. 1b, c), the Car that is positioned to most easily undergo large structural motions 24 , which may induce displacement of these Chl bs. Even minor perturbations to inter-pigment distances can significantly change the dynamics due to the nonlinear relationship between distance and energy transfer rate 53,54 .
Low-energy Chl a to Car S 1 energy transfer. The second prominent change appears on the red side of the lower-energy Chl a pool (a L ). This pool consists of the three Chl as in domain II that interact strongly with Lut1 and form the terminal locus of energy, collecting energy from higher-lying states and emitting fluorescence in isolated LHCIIs 44,55 . The waiting time traces of the red half of the Chl stimulated emission (SE) reveal pronounced rapid decay components with time constants and amplitudes of 350 (±30) fs (39%) in detergent and 270(±20) fs (53%) in the membrane, followed by slower decays of several ps. In LHCII, there are picosecond-timescale vibrational relaxation processes 56,57 as well as the nanosecond-timescale fluorescence. Because of the limited Intensity of the Car S 1 ESA relative to the initial Car S 2 population at T = 533 fs in detergent (gray) and in membrane discs (green). The relative S 1 intensity was obtained by normalizing the S 1 ESA intensity to the initial S 2 GSB intensity immediately after photoexcitation (T = 30 fs). c Comparison of Car S 1 ESA decay constants in detergent (gray) and in membrane discs (green). Due to the limited temporal window of our 2DES measurement (T = 0−8 ps), we are unable to determine the accurate S 1 lifetimes and therefore confine our discussion to relative changes in these timescales. d Absorptive 2D spectrum of the Car-Chl cross peak region at T = 300 fs (in detergent). Colored sticks indicate the energy levels of the Car S 2 and Chl Q states. e Ratio of Car-Chl cross peak intensity obtained by dividing the sum of all cross peak intensities in the membrane by that in detergent. Error bars in b, c, and e are s.d. from three independent measurements. f Projection of the 2D spectra shown in a onto the ω t -axis for a 600 cm −1 ω τ interval centered at ω τ = 20,000 cm −1 (gray: detergent, green: membrane). g A closer view of the boxed region in f, where both traces are normalized to the same scale to emphasize the energy shift.
temporal range of our 2DES apparatus, we do not fully characterize these slower processes and thus the collectively fit them as a single long-timescale component (Supplementary Note 7 and Supplementary Fig. 26). A representative time trace from the center of this region is shown in Fig. 3d. The amplitude of the sub-ps decay component increases as the emission frequency decreases, and is non-negligible only when the red side of the Chl a L band is probed, which corresponds to the red half of the Chl a emission (Supplementary Note 7, Supplementary Fig. 24). The biexponential decay kinetics of Chl a L imply two subpopulations with different levels of quenching, likely reflecting a quenched conformation and an unquenched one 58,59 . Recent transient absorption studies on CP29, a minor antenna complex homologous to LHCII, found a similar biexponential decay of the terminal Chl a excited state, which was attributed to the coexistence of quenched and unquenched conformations 60 . The coexistence of multiple conformations with distinct photophysics is further supported by single-molecule fluorescence measurements that identified unquenched and quenched conformations of LHCII 61,62 and other homologous complexes 63,64 . The presence of a rapid, sub-ps decay component points towards an energy sink that accepts energy from the terminal locus. Notably, we find concurrent rise at the excitation frequency of Chl a L and emission frequency of Car S 1 ESA, which indicates that the Car S 1 states are the energy sink populated by energy transfer from the terminal Chl as (Fig. 3e). Although this region of the 2D spectrum contains a contribution from Chl ESA 18 , the absence of an increase in Chl a population on the corresponding timescale supports the assignment that the rise originates from the ESA of the Car S 1 instead of Chl states (Supplementary Note 7, Supplementary Fig. 25). Following energy transfer from the Chls, the Car S 1 state dissipates the excitation energy via a picosecond non-radiative decay process, as mentioned earlier. This is a clear and direct observation of the dissipative energy transfer pathway from the emissive Chl a locus into the dark S 1 state of the Cars, one of the mechanisms of photoprotection proposed but not well understood 10,18,19,25,65 . Correlated decay of Chl a and rise of Car S 1 , similar to those identified here but on a slower timescale (2.1 ps), have been observed in a high lightinducible protein (Hlip), a cyanobacterial ancestor of plant antenna complexes, and assigned to Chl-to-Car energy transfer 65 . In contrast, in previous experiments on LHCII, differences in the kinetics of unquenched and quenched samples were observed, yet no rise of the Car S 1 ESA was detected, which was attributed to excitonic mixing of the Chl and Car states 19 or inverted kinetics 10,25 following data processing and/or kinetic modeling.
While the terminal Chl a → Car S 1 energy transfer pathway is present for LHCII in both environments, the amplitude of the component increases by 14% for the Chl a L decay, and consistently, by 12% for the Car S 1 ESA rise in the membrane. This is qualitatively in agreement with the observation of increased fluorescence quenching in the membrane discussed above. The enhancement of this Chl a → Car S 1 pathway could arise from the redshift of the Car S 1 states discussed earlier.
Although the exact energy gap between the Chl a Q y and Car S 1 states cannot be determined, such a redshift could bring the two states closer to resonance, and thus increase the rate of energy transfer.

Discussion
The mechanism of photoprotective quenching has been extensively debated in the field. One of the likely mechanisms, the Chl Q → Car S 1 energy transfer observed here, appears spectroscopically as a rise in the Car S 1 population after Chl excitation. While differences in the long-time decay dynamics of the Car S 1 have been reported for unquenched and quenched LHCII, an instantaneous initial rise of the Car S 1 population was seen, potentially due to limitations in temporal or spectral resolution 10,18 . This led to the development of an excitonic mixing model between Chl Q y and Car S 1 states 12,19 , where the observed instantaneous Car S 1 rise was attributed to strong excitonic interactions between the Chl and the Car states. Here, as illustrated in Fig. 4a, we resolve the initial rise of Car S 1 ESA, characteristic of a directional energy transfer rather than a delocalized Car-Chl excited state of the excitonic mixing model. Theoretically predicted timescales for this energy transfer pathway are >20 ps due to the optically forbidden nature of the Car S 1 state 43,66,67 , which is two orders of magnitude longer than the sub-ps (<400 fs) timescale observed in our experiment. This discrepancy suggests that a more complex picture is required, such as directional Chl → Car energy transfer mediated by partial mixing of the excited states, along the lines of previous proposals 11,12,19 . The comparison between detergent and membrane environments presented here demonstrates that the local environment is able to impact photophysical pathways in plants, including altering the balance between light harvesting and photoprotection. To illustrate the dynamics in both environments, we have constructed a kinetic model of the photophysical pathways using the time constants extracted from our spectra (Supplementary Note 8,Supplementary Figs. 27,28). Our model shows that the two major energy transfer pathways for efficient light harvesting in LHCII, Car S 2 → Chl Q and Chl b Q y → Chl a Q y energy transfer, are both suppressed in the membrane. The Chl a Q y states then transfer energy to neighboring proteins for transport towards the reaction center. Consistently, our model also shows that two dissipative pathways, Car S 2 → Car S 1 and Chl a → Car S 1 energy transfer, are both enhanced in the membrane. The short-lived dark S 1 state of the Cars then rapidly quenches the excitation via non-radiative decay. Collectively, these changes enhance the dissipative pathways relative to light harvesting ones by increasing the relative population of the Car S 1 state and decreasing that of the Chl Q states, as illustrated in Fig. 4b. The quantitative agreement between the experimental and simulated populations illustrates that the minimal set of photophysical pathways included here is sufficient to describe the observed dynamics.
Our data, both in the steady state (CD) and on ultrafast timescales (2DES), suggest that the two peripheral pigment domains (Neo and Lut1 and the Chls strongly coupled to them (Fig. 1b, c)) are the molecular origin of the observed energetic and dynamical changes in the membrane. On the other hand, Lut2 is found to be completely immune to the introduction of the lipid bilayer, maintaining its light-harvesting role as the principal energy donor to Chls 8 . While our nanodisc platform cannot fully replicate the complex architecture of the native thylakoid membrane, these observations show that the Car conformation is readily modulated by interaction with the surrounding local environment, which can impact the excited-state dynamics, and potentially enhance dissipative pathways. Consistent with these experimental results, a recent theoretical work found that even a 5−10°tilt in the backbone of the luteins causes a 50% drop in the fluorescence lifetime of LHCII, highlighting the integral role of Car conformations on LHCII photophysics under varying light conditions 67 .
It is interesting to note that the two strongly perturbed peripheral domains identified in this work correspond to two of the proposed photoprotective quenching sites in LHCII from previous work, and here we similarly observe a correlation between these perturbations and quenching. Twisting of the Neoconjugated chain has been postulated as a potential mechanism for quenching in crystals of LHCII based on a correlation between the twist and quenching 22 . Lut1 was speculated to undergo a conformational change that opens up a quenching site with the terminal Chl as in oligomeric LHCIIs 10,23 . We observe quenching even in the non-aggregated, individual trimeric LHCIIs through the reduced fluorescence lifetime of 2.8 ns and the associated dissipative photophysics. This suggests that the native structure of LHCII trimers enhances quenching upon environmental perturbation, which may be a similar effect to that observed in LHCII aggregates. Considering that the two strongly perturbed pigment domains identified herein would be located near the interface of trimeric LHCIIs in vivo, protein-protein interactions in the native system may introduce a similar effect and further amplify the structural reorganizations observed here. These interactions may be either between multiple LHCIIs or between LHCII and the photosystem II subunit S (PsbS), which is a non-pigmentbinding protein required for quenching in vivo, potentially via induction of a conformational change in LHCII 26,68 . In order for a dissipative pathway to be relevant for photoprotection, it must be activable under high light conditions, and these interactions may be the mechanism behind activation.
In this work, we benchmark the dynamics and pathways of light harvesting and dissipation in LHCII embedded within a near-native membrane. We characterize two dissipative pathways, both of which utilize the dark Car S 1 state as energy sink. One of the dissipative pathways, sub-picosecond energy transfer from the terminal Chl locus to the Car S 1 state, is uncovered through our ultrafast time resolution. The observation of this predicted, but previously uncharacterized dissipative pathway opens the door to studies of its role in photoprotection. Our measurements provide evidence that dissipation is enhanced in the membrane, likely through an increase in the population of a quenched conformation. These results point to the ability of the local environment to determine the conformation and dynamics-and therefore function-of the photosynthetic apparatus in green plants.

Methods
Sample preparation. Detailed information on sample preparation including production and characterization of the nanodisc sample can be found in the Supplementary Information (Supplementary Notes 1-3, Supplementary Figs. 1-6, 9, 10, and Supplementary Table 1). The final optical density (OD) of both samples was 0.45 (per 0.2 mm) at 675 nm for the dataset obtained with spectrum 1 and 0.2 for the dataset obtained with spectrum 2 (Supplementary Note 2, Supplementary  Fig. 7). In all 2DES measurements, the samples were circulated in a 0.2-mm pathlength flow cell with a peristaltic pump to prevent photodegradation and repetitive excitation of the same spot. The sample reservoir was kept at 4°C throughout the measurement with a home-built water jacket cooling system.
Ultrabroadband 2DES. Details of the ultrabroadband 2DES apparatus are provided in the Supplementary Information and in ref. 46 . Glass filters with different cutoff wavelengths were chosen for each dataset in order to tune the spectrum for optimal excitation of the Car S 2 /Chl Q x and Chl Q x /Q y regions, respectively (Supplementary Note 2, Supplementary Fig. 7a). Spectrum 1 (primarily Car/Chl Q x excitation) was centered at 550 nm (18,182 cm −1 ) with a full-width at halfmaximum (FWHM) of 113 nm (3,819 cm −1 ), and spectrum 2 (primarily Chl Q x / Q y excitation) was centered at 614 nm (16,287 cm −1 ) with a FWHM of 168 nm (4,807 cm −1 ). The final spectra were compressed with chirped mirror pairs (Ultrafast Innovations GmbH) to 6.2−6.9 fs pulses as characterized with transient grating frequency-resolved optical gating (TG-FROG, Supplementary Fig. 7b, c in Supplementary Note 2) 69 . Coherence time (τ) was sampled in 0.4 fs steps in the range of −200 to 200 fs, resulting in a 43.8 cm −1 resolution of the excitation frequency (ω τ ) axis. Waiting time (T) was incremented in steps of 10 fs for T = 0 −100 fs, 33 fs for T = 100−467 fs, 67 fs for T = 467 fs-1 ps, and 1 ps for T = 1−8 ps (dataset with spectrum 1) or T = 1-10 ps (dataset with spectrum 2). The resolution of the emission frequency (ω t ) axis was 4.2 cm −1 . The data were measured with allparallel pulse polarization. A pulse energy of 10 nJ was employed for all 2DES measurements with a beam waist of 150 μm at the sample position, corresponding to an excitation density of 3.9−4.4 × 10 13 photons per pulse per cm 2 , previously reported to be in the linear regime 52 . Each dataset was collected three times, on separate days with freshly prepared samples, to ensure reproducibility of the data. The integrity of the sample was confirmed by comparing the linear absorption spectra and fluorescence decay profiles before and after each set of measurement (Supplementary Note 2, Supplementary Fig. 8).
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The source data underlying Figs. 2b, c, e, 4b and Supplementary Figs. 2b, 3b, 4c, 5a are provided as a Source Data file. Other data are available from the corresponding author upon reasonable request.