Introduction

A 2D electronic spectrum is presented in two frequency dimensions, which can be termed the excitation frequency and the emission frequency. Two-dimensional electronic spectroscopy (2DES)1,2,3 measures a two-point correlation plot, and is an excellent tool to measure single-step excitation energy transfer (EET) processes as the appearance and evolution of off-diagonal cross peaks provide a direct observation of electronic coupling between the donor and acceptor pigment molecules, by reading off the corresponding frequencies on the excitation and emission axes, respectively4,5,6,7. However, multistep EET processes are only indirectly inferred from the 2DES measurements8. Since it is clear that complex multistep EET network is present in most light-harvesting complexes9,10,11,12, higher-order multidimensional electronic spectroscopy will be needed to directly measure multistep EET processes.

Fifth-order 3D optical spectroscopy is an extension of third-order 2D optical spectroscopy13,14,15,16,17,18. Applications include the probing of higher states of vibrational systems in order to better characterize the anharmonicity 14,19, the testing of non-Markovian processes20 and the characterizing of electron transfer systems21.

In fifth-order 3D optical spectroscopy, a spectrum S(5)(ω1, ω3, ω5; t2, t4) with three frequency axes parametrized with two population times t2 and t4, is measured. A two-step EET process that proceeds from state A to state B over time t2 and from state B to state C over time t4 results in the emergence of a 3D cross peak (ω1, ω3, ω5)=(ωA, ωB, ωC) on a 3D spectrum, where ωA, ωB and ωC are the absorption frequencies of state A, B and C, respectively. In comparison, a conventional 2D spectrum is presented as S(3) (ω1, ω3; t2) with a population time t2. In addition, if the conventional 2D spectrum S(3)(ω1, ω3; t2) is measured for a series of population times t2 and Fourier transformed about t2, one can obtain a data set S(3)(ω1, ω3; ω2). As there are three frequency axes involved, it has been commonly termed 3D spectroscopy as well22,23. This third-order 3D optical spectroscopy S(3)(ω1, ω3; ω2) is different from the fifth-order 3D optical spectroscopy S(5)(ω1, ω3, ω5; t2, t4), which is the subject of this article.

There are several other fifth-order multipulse nonlinear optical experiments such as the family of pump-dump-probe (PDP) and pump-repump-probe spectroscopies24 and the multiple population-period transient spectroscopy25. In these experiments, the system interacts at two moments in time (separated by a first population period) with the light field. This is followed by a second population period before a transient absorption is obtained. Therefore these experiments give a transient spectrum as a function of the two population time S(5)(ω5; t2, t4). There is no frequency resolution (apart from the coarse resolution determined by the laser-pulse spectral bandwidth) for the two interactions. The fifth-order 3D optical spectroscopy described in this paper can be viewed as a general form of these fifth-order experiments. In addition to the two controllable population periods, fifth-order 3D optical spectroscopy has three coherence periods that give rise to three frequency dimensions. This gives the three-point frequency–frequency correlation as a function of two population times S(5)(ω1, ω3, ω5; t2, t4). Another kind of fifth-order optical experiment, 2D Raman spectroscopy was plagued with spurious signal arising from third-order cascading processes26. It has been shown in control tests performed for fifth-order resonant optical experiments (such as the fifth-order 3D electronic spectroscopy experiments discussed in this article) that there is no significant contamination from third-order cascading signals15,19,27. If for some systems or conditions that the cascading signals become significant, procedures are available to isolate the contaminant signals28.

A fifth-order 3D optical spectroscopy experiment involves the control of up to six (including local oscillator pulse) phase-locked pulses and is extremely challenging. Purely absorptive fifth-order 3D optical spectroscopy was first demonstrated in the mid-infrared29 and only recently did we first demonstrate a purely absorptive fifth-order 3D electronic spectroscopy in the visible (three-dimensional electronic spectroscopy (3DES)) based on a pulse shaping pump-probe setup17,27.

Plant light-harvesting complex II (LHCII)—the most abundant membrane protein in nature, is responsible for the absorption of about half of the radiation used for plant photosynthesis and plays key roles in regulating the energy input to Photosystem II to optimize photosynthetic efficiency while avoiding damage30,31.

In this article, we report the use of 3DES to directly observe two-step EET processes in LHCII. We observe 3D cross peaks that represent a two-step EET process from the chlorophyll b (Chl b) manifold to the low-energy level chlorophyll a (Chl a) via mid-level Chl a energy states. The population time dependence of the 3D cross peak amplitude corresponds well with corresponding 2DES measurements. We provide possible assignments of the observed spectral features based on published structure-based models.

Results

Fifth-order 3DES

In fifth-order 3DES, five laser pulses, with inter-pulse delay times t1, t2, t3 and t4, interact with the sample and a fifth-order optical signal is emitted at time t5. The pulse sequence scheme is shown in Fig. 1a. The periods t1, t3 and t5 are the coherence times where the excitons oscillate at their transition frequencies. The periods t2 and t4 are the population times, during which the population transfers between different excitons. The final coherence over t5 results in an optical signal that is emitted in a phase-matched direction and can be measured by frequency-resolved heterodyne detection (along the ω5 axis). At given population times t2 and t4, the experiment is scanned over the coherence periods t1 and t3. The data are processed and then Fourier transformed about coherence times t1 and t3 to obtain a 3D spectrum S(5)(ω1, ω3, ω5; t2, t4). The 3D cross peak feature that represents a two-step EET process that proceeds from state A to state B over time t2 and from state B to state C over time t4, can be outlined using the Liouville pathways diagrams (double-sided Feynmann diagrams)32 in Fig. 1b. The labels for the processes R1, R2, R3 and R4 follow the convention of Hamm13. The shaded areas denote the population periods where EET processes proceed between the states separated by the dashed lines.

Figure 1: Three-dimensional electronic spectroscopy (3DES) pulse sequence and Liouville pathways diagrams.
figure 1

(a) Pulse sequence used in coherent fifth-order 3D optical spectroscopy performed in a pulse-shaper assisted pump-probe geometry. The pulse-shaper creates a four-pulse pump sequence (red) with controllable delays and relative phases followed by the probe (green) that interacts at a small angle to emit the signal (blue). (b) A 3D cross peak (ω1, ω3, ω5)=(ωA, ωB, ωC) on a 3DES spectrum is a combination of fifth-order optical processes outlined in the Liouville pathways (double-sided Feynman diagrams) depicted above16. The shaded areas denote the population periods where EET processes proceed between the states separated by the dashed lines. The label of the processes R1, R2, R3 and R4 follows the convention of Hamm10.

Full size image

Linear spectroscopy and 2DES of LHCII

According to the crystallographic structure of LHCII33, each monomer in the trimeric pigment–protein complex binds 14 chlorophylls (8 Chl a and 6 Chl b33). The 14 excitons are closely spaced, which makes it hard to elucidate the different timescales of every energy migration dynamics in the multistep EET processes. Henceforth we will adopt the more often used wavelength unit, λ instead of the frequency unit ω. The linear absorption spectrum (Fig. 2a) can be divided into three regions: high-energy exciton region (λH<660 nm), encompassing transitions to Chl b and high-energy Chl a exciton states, the mid-energy region (λM=660–675 nm), which contains mid-energy Chl a states, and the low-energy exciton region (λL>675 nm), which contains the lowest-energy Chl a exciton states. In the following, we will demonstrate how 3DES can be employed to observe the multistep flow of energy between the three wavelength regions λH→λM→λL.

Figure 2: Linear and 2DES spectrum of LHCII.
figure 2

(a) The experimental linear absorption spectrum of LHCII (solid blue) and the excitation pump spectrum (solid filled pink). (b) Two-dimensional spectrum of LHCII at population time t2=0.3 ps, with the white line indicating the diagonal of the spectra. (c) The amplitude evolution of the integrated exciton cross peaks 657→670 nm and 657→678 nm as a function of population time t2. A multistep EET process can be inferred from the associated decay of the 657→670 nm cross peak and the rise of the 657→678 nm cross peak.

Full size image

As a reference, 2DES of the EET process was performed as well, and a representative 2D spectrum at population time t2=300 fs is shown in Fig. 2b. More 2D spectra at various t2 have been included in Supplementary Fig. 1. Two cross peak regions, H/L and H/M, are labelled in Fig. 2b, which is ascribed to EET from the high-energy exciton region to the low-energy region (λH→λL) and to the mid-energy region (λH→λM), respectively. The cross-peak amplitudes at H/L (655→678 nm) and H/M (655→670 nm) are plotted against t2 in Fig. 2c. The H/M amplitude initially rises rapidly on a timescale of ∼300 fs due to EET from Chls b to mid-energy Chls a. The H/M cross peak subsequently decays on a longer timescale of ∼2 ps, which is a result of EET out of the mid-energy Chl a region. The H/L cross peak amplitude shows only an increasing trend. Apart from indicating the direct λH→λL transfer, a multistep EET process λH→λM→λL can be indirectly inferred from the correlated timescale of the decay of the H/M peak and the rise of the H/L cross peak, as it has been previously reported8,34. It is evident that the cross peaks may not be readily resolved in the 2D spectra, due to the broad inhomogeneous widths of the spectral features. In this regard, 3DES has the advantage of being able to spectrally resolve closely spaced or overlapping exciton bands in the additional frequency axis ω5, and also by correlating energy transfers with the three frequency axes.

Fifth-order 3DES of LHCII

In the 3D spectra of LHCII presented in this article, the population time t2 was kept fixed at 300 fs, and spectra were taken at different population times t4 from 150 fs to 8 ps. A graphical representation of the 3D spectrum at t2=300 fs, t4=800 fs is shown in Fig. 3. The rendered isosurface represents amplitude values of 0.1 relative to the global maximum. The maximum along λ1 and λ3 is found near the excitation wavelength, 668 nm; the peak along the detection wavelength λ5 coincides with the pump-probe maximum, 678 nm. A cutaway is made at λ5=682 nm to better illustrate the features and structure of the 3D spectra. The amplitude of the 3D spectrum on this cutaway slice is colour coded. The amplitude is concentrated on the λ3=λ1 diagonal plane, representing all the states whose population has not decayed during t2. It is evident from the figure that the 3D spectrum features additional cross-correlation amplitudes off the λ3=λ1 diagonal plane, particularly in the region λ3<λ1. This indicates downhill energy transfer during the population time t2. One prominent feature to note is a ridge (or 3D cross peak) along the λ5 axis, around λ1=655 nm, λ3=670 nm. The ridge’s shape and feature evolves with the different population times t4 in our experiments and is the main subject of our discussion below. For comparison, representative 3D spectra at t2=300 fs and t4=150 fs and 5 ps are presented in Supplementary Fig. 2. Qualitatively, one can discern the evolution of the 3D cross peaks clearly with different delays t4.

Figure 3: Three-dimensional electronic spectroscopy (3DES) spectrum of LHCII.
figure 3

Spectrum recorded at population times t2=0.3 ps and t4=800 fs. The isosurface represents amplitude values of 0.1 relative to the global maximum. The cutaway at λ5=685 nm to better illustrate the features and structure of the 3D spectra. The prominent feature to note is the ridge along the λ5 axis, around λ1=655, λ3=670. The ridge’s shape and feature evolves with the different population times t4 in our experiments.

Full size image

To better visualize the time evolution of the 3D cross peaks, slices of the 3D spectra at wavelengths λ1=655 nm and λ1=665 nm for three different population times t4 are shown as quasi-2D spectra with wavelength axes λ3 and λ5 (Fig. 4). The spectra demonstrate the potential of 3DES to selectively probe EET pathways with high resolving power. The slices at λ1=655 nm feature two well-separated cross peaks, at λ3∼655 nm and λ3∼670 nm, whereas the slices at λ1=665 nm have only one band around 665 nm. The most significant feature on the presented slices is the 3D cross peak 655→670→678 nm, which can be denoted H/M/L, and interpreted as follows. In a 2D spectrum at population time t2, the H/L cross peak amplitude is the conditional population of state L (absorbing at λL) after population time t2, given that the initial excitation is at H (absorbing at λH). Extending the interpretation to 3DES, the 3D cross peak amplitude at H/M/L is the conditional population of L after population time t2+t4 given that (i) the initial excitation is at H (λH∼655 nm) and (ii) the population after time t2 is at M (λM∼670 nm). This gives a direct measurement of a two-step EET process between three different states: H→M→L. The other features of interest in the spectra are along λ1=655 nm, that is, on the diagonal plane λ3=λ1, denoting EET from states H to states L (H/H/L cross peak) and from H to M (H/H/M cross peak).

Figure 4: Two-dimensional slices of 3DES spectra.
figure 4

Slices of the 3D spectra of LHCII at λ1=655 nm and λ1=665 nm to obtain quasi-2D spectra, with selected quasi-2D spectra shown at population times t2=0.3 ps and t4=0.2, 0.8 and 5 ps (indicated on the plots). The colour scale represents amplitudes relative to the respective 3D spectrum’s global maximum.

Full size image

Owing to the separation over a third frequency (wavelength) dimension, the spectral regions H and M are fully resolved in the 3D spectra, whereas in the 2D spectra (Fig. 2b) they overlap. Moreover, by virtue of using two population times, the kinetics of the direct H→L and multistep H→M→L pathways can be unambiguously resolved. The rise dynamics of the H/H/L cross peak amplitude depends solely on the H→L process, and the rise of the H/M/L peak depends solely on the M→L process, allowing us, in principle, to determine the respective rate constants directly from the experimental data without the need for soft modelling and the inherent uncertainty associated with it. Figure 5 plots the experimental amplitudes of the H/M/L cross peak as a function of population times t4. We simulate the H/M/L cross peak amplitudes by solving a rate equation for a system comprising H, M and L states (Supplementary Note 2). As parameters for the simulations, we use the EET transfer timescales values that have been recently obtained for LHC II trimers34, where the Chl b excitonic states (associated with the H states here) transfer energy to Chl a states (associated with the M and L states here) at a timescale of 300 fs and the intermediate Chl a states (associated with the M states here) to low-energy Chl a states (associated with the L states here) at a 2.3 ps timescale. There is then a >20 ps relaxation to ground state g. The simulated values (red line in Fig. 5) are consistent with the trend of the experimental values, as can be seen in Fig. 5. The line essentially traces a 2.3 ps timescale rise that represents the M→L transfer and a subsequent >20 ps decay representing the decay to ground state.

Figure 5: Population time dependence of 3D cross peak amplitude.
figure 5

Experimental values (blue circles) and simulation trace (blue line) of the H/M/L cross peak amplitudes in the quasi-2D spectra (λ1=656 nm slice; t2=0.3 ps) as a function of the second population time t4.

Full size image

Discussion

Assignment of spectral features to structural moieties in the pigment–protein complex is possible by use of structure-based models of the exciton energies and transfer kinetics9,10,12,35,36. On the basis of more recent model by Müh et al.36, the H state (λH∼655 nm) can be contributed by either Chl 608 (∼658 nm) in the stromal layer of Chls b, or Chl 607 (∼654 nm), in the lumenal Chl b cluster. Although both states are within the relevant wavelength, Chl 607 is incompatible with the observed H/H/L cross peak at t2=300 fs. EET from this Chl b, on the lumenal side, to the L state, which is identified with the Chl 610/611/612 cluster on the stromal side, must occur over a much longer time scale. In contrast, rapid EET (≤1 ps) from Chl 608/609 to the 610/611/612 cluster, is expected by the structural models of Novoderezhkin et. al.10,12, which is in line with the dynamics of the 3D cross peak. As to the identity of the M state (λM∼670 nm), several pigments may contribute to this spectral region but only the Chl 602/603 pair (also according to Novoderezhkin et al.10) would have the longer lifetime compatible with the 3DES data. The available structural models provide different time constants for EET depending on the chosen theoretical formalism. The EET from Chl 602/603 to the lowest-energy Chls is estimated to be from subpicoseconds10 to tens of picoseconds12. The kinetics of the H/M/L peak in our experiment is consistent with the rate constant of M→L transfer between 2 and 6 ps which falls closer to the more recent calculation by Renger et al.12

The results presented in this report demonstrate for the first time that direct observation of multistep EET processes in LHCII can be achieved using 3DES. The method allows us to selectively probe EET between single exciton states with defined transition energies, isolated from the congested spectrum of the exciton network. In principle, with two independent population times, the two rate constants of a two-step EET process can be directly obtained from experimental data by measuring the population time-dependent amplitudes of appropriate 3D cross peaks. Fifth-order 3DES represents a valuable extension to 2D electronic spectroscopy, which is rapidly becoming a standard tool in photosynthesis research, and will undoubtedly help in unravelling the complex energy transfer processes in biological light-harvesting systems. As a much longer experimental time needed to collect a 3D spectrum compared with a 2D spectrum, it may not be practical to obtain 3D spectra for an exhaustive range of population times t2 and t4. It is envisioned that 2DES will first be deployed to obtain as much information as possible about the EET rates. 3DES can then be performed for selected population times t2 and t4 to resolve specific exciton states involved in the multistep EET processes.

Methods

Sample preparation

The LHCII trimers sample was purified from spinach thylakoid membranes and solubilized with α-dodecyl maltoside (α-DM) using sucrose-gradient ultracentrifugation37. The sample was mixed with nitrogen-flushed 0.06% α-DM and 10 mM HEPES buffer solution, and placed in a 1 mm sapphire cell at room temperature. The absorbance of the sample at 674 nm was measured to be 0.2. All linear, 2D and 3D spectra were acquired at room temperature.

Data acquisition

Our implementation of third-order 2DES38 and fifth-order 3DES17,27 has been described in detail elsewhere. In brief, in the pump-probe geometry39,40,41,42 implemented 3DES, the pulse shaper was used to create a four-pulse ‘pump’ train (1 kHz laser pulses from an optical parametric amplifier centred at 668 nm. Pulse duration is 43 fs; see Fig. 2a) while a white light continuum pulse is used as the ‘probe’ (that provides the fifth interaction). In the present setup, the pulse shaper is able to produce pulse trains whereby the maximum delay between the first and fourth pulses is about 1 ps. Both coherence times t1 and t3 were increased in 7 fs time steps from 0 fs to 140 fs. The maximum population time t2 for our present setup is therefore about 600 fs. The delay for the other population time t4 is achieved by a conventional optomechanical delay stage, and the maximum achievable t4 is determined by the length of the delay stage. To obtain the signal from the desired processes (Fig. 1b), phase cycling43,44,45 is necessary. A 3 × 3 × 3 × 1 phase cycling scheme27,45 was used to obtain the purely absorptive 3D spectra. For the experiments reported in this article, the population time t2 was set at 300 fs, and population times t4 were measured at 150, 200, 300, 500, 800, 2,000, 5,000 and 8,000 fs. The 3D spectra were normalized to the maximum intensity of the 2D projection at frequency axis (λ1, λ3). The pulse shaper was referenced to a carrier frequency of 420 THz and the signal was collected in a partially rotating frame. The ‘probe’ beam is measured with a spectrometer (Acton SP2300, Princeton Instruments) and recorded with a 100-pixel × 1,340-pixel charge-coupled device camera (PIXIS 100B, Princeton Instruments) giving the wavelength axis λ5. The data is Fourier transformed about t1 and t3 to give the frequency axes ω1 and ω3, which was then converted to wavelength axes λ1 and λ3. Scatter subtraction and Fourier transformation into the frequency spectrum was performed as reported previously27,38. For the accompanying 2DES, a two-pulse ‘pump’ train with a 3 × 1 phase cycling scheme38 is used. The population times t2 were recorded at a series of delays from 120 fs to 10,000 fs. In 2D optical spectroscopy, a common method to characterize possible experimental artefacts is to apply the 2D projection-slice theorem39. In this method, the 2D spectrum is integrated over one frequency axis and compared with the transient absorption spectrum. The 2D projection-slice theorem states that the 2D spectrum should be similar to the 1D spectrum. Analogous to the 2D counterpart, the 3D projection-slice theorem has also been developed13. We compare the experimental 3D spectra integrated along one of the frequency axes with the corresponding experimental 2D spectra to verify that our 3D spectra are largely free of experimental artefacts. The details of the comparison are provided in the Supplementary Note 1.

Additional information

How to cite this article: Zhang, Z. et al. Direct observation of multistep energy transfer in LHCII with fifth-order 3D electronic spectroscopy. Nat. Commun. 6:7914 doi: 10.1038/ncomms8914 (2015).