Ab initio model for the chlorophyll-lutein exciton

coupling in the LHCII complex

Daniil Khokhlov, Aleksandr Belov
Department of Chemistry, Lomonosov Moscow State University, Leninskie gory 1-3,
Moscow, Russia, 119991

Abstract

2A−
g state of lutein plays a crucial role in photoprotection of higher plants.

Due to its multiconfigurational nature, accurate description of this electronic

state and respective transition properties is a formidable task. In this paper,
applicability of various CASSCF and RASSCF formulations for description
of the 2A−
g state is discussed. It is shown that inclusion of the entire π-system

of lutein into the active space is required for accurate calculation of transition
properties. Exciton coupling in the chlorophyll-lutein dimer involved in non-
photochemical quenching in the LHCII complex was calculated to provide a
connection between pigment interactions and non-photochemical quenching
regulation.
Keywords: Xanthophyll, lutein, LHCII, non-photochemical quenching,
exciton coupling, MCSCF

1. Introduction

Non-photochemical quenching of chlorophyll fluorescence is an important

part of the protection of photosynthetic apparatus from degradation under
excessive light. In such conditions, reaction centers are unable to convert

Preprint submitted to Biophysical Chemistry December 27, 2018

 5 all excitation energy into chemical one which leads to formation of various
strong oxidizers such as the long-lived special pair radical P+•
680 and singlet

oxygen 1 ∆g O2 [1, 2]. They could irreversibly damage protein environment

and/or pigments thereby disrupting normal functioning of the photosynthetic
apparatus. In photosystem II (PSII) of higher plants, almost all peripheral
10 antennae such as CP24, CP26, CP29, and LHCII proteins could serve as flu-
orescence quenching sites. The latter acts as a light-harvesting antenna most
of the time while under high-light conditions it is an important fluorescence
quencher.
All these proteins bind various xanthophylls (oxygenated carotenoids)
15 such as lutein, zeaxanthin, neoxanthin and violaxanthin. Xanthophylls play
at least three different roles in photosynthetic complexes: they stabilize pro-
tein structure, absorb light in blue spectral region, and take part in non-
photochemical quenching (NPQ) of chlorophyll fluorescence [3]. Two notable
properties of xanthophylls make them powerful quenchers of chlorophyll flu-
20 orescence. First, energy-wise the S1 state (2A−
g assuming ideal C2h polyene

symmetry [4]) of some xanthophylls such as lutein and zeaxanthin lies be-
low the S1 state of chlorophyll a (Qy transition in Gouterman model [5]),
thus making the excitation energy transfer from the pool of chlorophylls a
to the lutein molecule possible [6, 7]. Second, the 2A−
g state of xanthophylls

25 promptly (lifetime ∼10-25 ps [8]) decays to the ground state (1A−

g ) via radi-

ationless internal conversion providing the way to trap excessive excitation

energy. Therefore, the excitation pathway may look as follows [9]: rapid en-
ergy transfer from chlorophylls b (less than 1 ps [10]) and energy equilibration
in the pool of chlorophylls a (several ps [11]) are followed by a relatively slow

2
30 energy transfer (from 130 ps for quenched LHCII to 1 ns for unquenched
one [12]) to one of the xanthophylls in the complex with a subsequent fast
relaxation to its 1A−
g state.

For the LHCII complex, this model was extensively developed by the-
oretical means by Duffy and coworkers [13, 14, 15]. The closely packed
35 heterodimer of lutein #620 and chlorophyll a #612 (further denoted as
Lut620/Chla612 dimer, hereinafter numbering according to PDB ID: 1RWT
[16]) is supposed to be the quenching site in the LHCII protein due to a rel-
atively large interaction energy between the ground and the lowest excited
states of both pigments in the pair (exciton coupling ∼14 cm−1 [13]). Ex-
40 citon couplings of Lut620 with other chlorophylls can be considered to be
negligible. Further, this model was expanded to the entire LHCII complex
by including both chlorophylls and xanthophylls into the exciton Hamilto-
nian. It confirmed the previously supposed quenching pathway [14]. LHCII
contains yet another closely coupled Lut621/Chla603 dimer (fig. 1) which
45 can take part in energy trapping. However, this lutein is coupled with chloro-
phylls a #602/603/604 with higher excitation energies than chlorophylls
#610/612/613 which are known to have the lowest excitation energies in
the complex [17]. Thus, taking into account rapid excitation equilibration
in the chlorophyll a pool, the quenching pathway through Lut620/Chla612
50 looks more probable. Moreover, the same model was successfully used to
describe switching in LHCII from the light-harvesting state to the quenching
one [15]. Mutual rotations of the pigments in the dimer lead to significant
change of transition density overlap and, thus, may promote NPQ in LHCII.

3
 Figure 1: Structure of the LHCII complex of Spinacia oleracea (from PDB ID: 1RWT).
Protein is shown in gray, chlorophylls a and b – in green and cyan, respectively. Neoxanthin
and violaxanthin are shown in orange. The two closely packed dimers Lut620/Chla612
and Lut621/Chla603 are shown in magenta and violet, respectively. Image was prepared
in VMD program [18]

55 While the Qy state of chlorophyll a is dominated by HOMO → LUMO

excitation [19] and, thus, can be accurately described by many quantum
chemical approaches from time-dependent functional theory (TD-DFT) to
various formulations of multireference perturbation theories, the 2A−
g state

of lutein has a more complicated electronic structure due to its multiconfigu-

60 rational nature [4, 20]. Semiempirical multiconfigurational methods has been
used for calculations in the early works of 70s and 80s [21, 4] and are still in
use [20, 22, 15]. The canonical TD-DFT approach yields a relatively good

4
 excitation energy (2.18 eV as compared to the experimental value of 1.76 eV
[7]) due to a fortunate cancellation of errors [23] but a poor transition dipole
65 value (11.9 times higher than DFT/MRCI reference in the same work) since
this method is single reference by design. A more sophisticated semiempri-
cal DFT/MRCI approach adopting DFT terms in the MRCI Hamiltonian
was successfully used for calculation of excitation energies of xanthophylls
[24] including lutein [25, 23]. Ab initio methods based on density matrix
70 renormalization group ansatz (DMRG) which accounts for static correlation
in large active spaces were successfully applied to large polyenes including
carotenoids. They give correct energetic ordering of dark excited states [26]
but are not invariant to orbital rotations, thus small transition dipoles can
vary significantly depending on orbital choice. Lastly, with regard to pho-
75 tosynthesis, it should be noted that all current NPQ models in the LHCII
[13, 14, 15] complex rely on semiempirical CAS-CI calculations for the lutein
molecule and, thus, may be refined by ab initio means.
In this paper, an ab initio study of the exciton coupling in the Lut620/Chla612
pigment pair is presented. The paper is organized as follows. First, impact of
80 the active space choice and wavefunction size in the MCSCF calculation on
the electronic properties of the 2A−
g state of the lutein molecule is discussed.

Second, exciton coupling in the Lut620/Chla612 pair is calculated basing on

the results for the individual pigments. Last, the influence of mutual ori-
entations of the pigments in the pair on the exciton coupling is studied for
85 internal coordinates which may be involved in NPQ regulation in the LHCII
complex. The obtained ab initio results are compared to the semi-empirical
ones following the approaches reported previously.

5
 2. Computation details

2.1. Individual pigments

90 2.1.1. Lutein
Initial positions of heavy atoms in the lutein molecule were taken from
the X-ray structure of the LHCII complex of Spinacia oleracea (PDB ID:
1RWT [16]), namely the lutein #620 was used. After addition of hydrogen
atoms, geometry was optimized at DFT/B3LYP/6-31G* level. RMSD of
95 conjugated bond lengths compared to the X-ray structure of lutein ester [27]
is 8.4 · 10−3 Å with maximum deviation 1.8 · 10−2 Å.
Starting molecular orbitals (MOs) for the MCSCF calculations were pre-
pared in the following way. In order to obtain initial approximation, Hartree-
Fock calculation was carried out in minimalistic ANO-S-MB basis set lacking
100 orbitals with angular momentum quantum number higher than 1 to facili-
tate identification of virtual orbitals of π-symmetry. Further, MP2 correction
was applied to the previous calculation, and natural orbitals were obtained
from the MP2 reduced density matrix. We selected MOs corresponding to
the conjugated π-system of the lutein molecule and projected them onto the
105 larger ANO-L-VDZP basis set designed for accurate description of electronic
properties with correlated wavefunctions [28]. Thus, the obtained MOs had
proper symmetry but they were not optimized in the new basis set. State-
specific (1A−
g state only was included) restricted active space SCF (RASSCF

[29]) calculation including 20 electrons on 20 orbitals was performed to op-

110 timize MOs and protect them against symmetry breaking. In the RASSCF
method, the active space is generally split into three subspaces: RAS1 con-
taining only occupied orbitals with limited number of possible holes, complete

6
 active space RAS2, and RAS3 containing only virtual orbital with limited
number of allowed electrons. For the calculations described in the paper,
115 we use the following notation for RASSCF active spaces: [A,B,C,X] means
that A occupied orbitals are included into RAS1 subspace; B electrons on B
orbitals are included into RAS2 subspace; C orbitals are included into RAS3
subspace; X stands for the maximum excitation level allowed from RAS1 and
to RAS3 (“s” for single excitations,“sd” for singles and doubles). Following
120 this notation, in preparation of MOs, active space of the type [10,0,10,sd]
was used. The optimized MOs obtained in this procedure (fig. S1) were used
as a starting guess in all further calculations for the lutein molecule.
Complete active space SCF (CASSCF) and RASSCF with various active
spaces were used to calculate transition dipole moments and density matri-
125 ces corresponding to the transition from the 1A− −
g to the 2Ag state. State-

averaged formulation of MCSCF was used to provide a balanced description

of 1A− −
g and 2Ag states simultaneously; both states had equal weights in

the averaging. Within the CASSCF approach we studied symmetric active

spaces (referred to as [Nel , Norb ]) with sizes from [6,6] to [14,14]. The latter
130 space is close to the current computational limit and, thus, CASSCF does
not allow inclusion of the entire conjugated π-system of lutein which would
be desirable. RASSCF is able to overcome this limitation, so it was used
to study effects of static correlation in the entire π-system. Each choice of
active space for RASSCF contained 20 electrons on 20 orbitals shared by
135 three RASSCF subspaces: RAS2 contained equal number of electrons and
orbitals and the remaining orbitals were equally distributed between RAS1
and RAS3 subspaces. We used two different sets of active spaces in RASSCF:

7
 in the first one (referred to as RAS(s)), only single excitations were allowed
from RAS1 and to RAS3; in the second one (referred to as RAS(sd)), single
140 and double excitations were allowed from RAS1 and to RAS3. So, the active
spaces used were of the type [10-N/2,N,10-N/2,s or sd] where N varied from
4 to 8.

2.1.2. Chlorophyll a
Geometry optimization procedure was the same as for the lutein molecule;
145 chlorophyll a #612 was taken from the same PDB. Nonpolar phytyl tail was
replaced by methyl group in all quantum chemical calculations in order to
accelerate them since it does not affect the π-system. RASSCF was used for
calculation of electronic properties of the chlorophyll in order to achieve con-
sistency with quantum chemical description of the lutein molecule. Starting
150 MOs for MCSCF calculations were prepared by projecting the Hartree-Fock
orbitals obtained in ANO-S-MB basis set into ANO-L-VDZP. 20 orbitals
corresponding to the conjugated π-system of the chlorin ring (fig. S2) were
selected. RASSCF active space was [8,4,8,sd]. The three lowest electronic
states (ground, Qy , and Qx ) were included in state-averaging with equal
155 weights. Resulting transition dipole between the ground and the Qy states
is 5.46 D and agrees well with the experimental one (4.49 D) obtained from
extrapolation of experimental transition dipoles to vacuum permittivity [30].
Therefore, transition densities were used further without any rescaling. How-
ever, it should be noted that proper rescaling of chlorophyll a dipole can lead
160 to decreasing of the exciton coupling by a factor of 0.82.

8
 2.2. Semi-empirical CAS-CI

Semi-empirical AM1-CAS-CI calculations were performed in MOPAC2016

[31] on the same geometries of pigments as for ab initio calculations; com-
165 plete active space included 6 electrons on 6 orbitals for both pigments. In the
calculation of the lutein molecule, the SCF procedure that preceded CI was
carried out with unitary occupancies for molecular orbitals of CI active space
following the computation protocol described in [22]. Since semi-empirical
methods lack explicit analytic form for wavefunctions, the respective TrESP
170 charges were calculated on wavefunctions obtained via projecting the implied
orthogonal basis onto STO-6G basis set using Löwdin ”deorthogonalization”
procedure in the same way as it is implemented in MOPAC for ESP charges
[32]. Rescaling of transition densities was not made to be consistent with our
ab initio results.

175 2.2.1. Software

Geometry optimizations were performed using GAMESS-US [33]. Prepa-
ration of MOs and MCSCF calculations were performed in OpenMolcas
(build #180705-1750), the developer’s version of MOLCAS package [34].
In the OpenMolcas calculations, Cholesky decomposition (using default de-
180 composition threshold 1.0·10−4 ) of two-electron integrals in combination with
atomic compact auxiliary basis set [35] was used to accelerate calculations.
Transition dipole moments were calculated in dipole-length formulation.

2.3. Structure of the chlorophyll-lutein pigment pair

Initial structure of the subunit of the LHCII complex (chain A) was taken
185 from PDB ID: 1RTW [16]. Hydrogen atoms were added; protonation states of

9
 amino acids were the default ones for physiological pH; all histidine residues
were in δ-configurations. After addition of hydrogen atoms, protein was
placed in the explicit 10 Å truncated octahedral water box; the system was
neutralized by adding 12 sodium ions to the simulation box. Solvent box
190 was optimized (atoms of the complex and ions were frozen) using 10000
steps of steepest descent algorithm. Further optimization was carried out for
the entire system using conjugated gradient algorithm with double accuracy.
RMSD between the coordinates of heavy atoms of the dimer in the optimized
structure and that in the X-ray structure was 0.66 Å. All molecular mechan-
195 ics calculations were performed in AMBER18 package [36], AMBER10 force
field was used for proteins, AMBER-compatible force fields for chlorophylls
and xanthophylls [37, 38], generalized AMBER force field [39] for LHG phos-
pholipid, and TIP3P model [40] for water.

2.4. Exciton coupling between pigments

200 Exciton coupling VLut−Chl between the two transition densities from the
ground to the first excited states for both pigments reads

VLut−Chl = hψeLut ψgChl |V̂ |ψgLut ψeChl i, (1)

where V̂ – Coulomb interaction operator, ψgLut and ψeLut – wavefunctions of

the 1A− − Chl
g and 2Ag states of lutein, respectively, ψg and ψeChl – wavefunctions
of the ground and the Qy states of chlorophyll, respectively. For coupling
205 calculation, an approximate approach based on effective TrESP (transition
ESP) charges [41] located on the nuclei and fitted to non-diagonal matrix
elements of electrostatic potential operator was used. This approach for cou-
pling calculation was chosen for the sake of transferability of the charges to

10
 another geometries of the pigment (provided they are not too different) which
210 is an advantage over other methods such as transition density cubes [42] or
cumulative atomic multipole moments [43]. Although such transferability is
limited to small distortions of the structure of pigment, the reported transi-
tion charges can be useful for further NPQ studies. The resulting drawback
i.e. inability to describe the component of transition dipole which is orthog-
215 onal to the π-system plane is not significant since the transition density is
symmetric with respect to the plane for π-π transitions. In this method,
exciton coupling can be evaluated as Coulomb interaction energy between
two sets of TrESP charges q Lut and q Chl , corresponding to transitions under
consideration in lutein and chlorophyll:
X qiLut qjChl
VLut−Chl ' , (2)
~ Lut − R
|R ~ Chl |
i,j i j

220
~ iLut and R
where summation is carried out over all charges in the set, R ~ iChl –

positions of the corresponding charges. TrESP charges were evaluated by fit-

ting electrostatic potential calculated on a geodesic grid surrounding nuclei
using in-house software and transition density matrices from MCSCF cal-
culations. Effective charges calculated for the in vacuo optimized structures
225 were placed in the positions of the corresponding atoms in the MM structure.
Direct fitting of the QM structures of the entire pigment molecules into the
MM geometry of the dimer (as is and with internal rotations, see section 3.3)
was also tested (see Supplementary, section S2) However, sterical clashes of
the side groups lead to unstable behavior at large rotation angles.

11
230 3. Results and discussion

3.1. Transition properties of the lutein molecule

Since we are interested in exciton coupling between the 1A− −

g → 2Ag

transition in the lutein and the S0 → Qy transition in the chlorophyll a, we

started with the study of transition electronic properties of the individual
235 pigments. Excited state of chlorophyll a can be described with appropriate
accuracy with a modest computational effort (section 2.1.2), therefore, we
focused on the lutein molecule. First, systematical study of the impact of
the active space size on the electronic properties of the lutein molecule within
MCSCF framework was done. CASSCF in relatively small active spaces (up
240 to [14,14]) is promising from the viewpoint of further improvement of ener-
gies and, to a much lesser extent, of transition properties using multireference
perturbation theories. 2A−
g state wavefunction is dominated by three CSFs

(φ2101 , φ1210 , and φ2020 ; index in the subscript consists of the two highest oc-
cupied orbitals and the two lowest unoccupied) which can be captured even
245 by a small active space. However, expansion of the active space does not lead
to convergent transition dipole values which do not have monotonous depen-
dence on the active space size (table 1). Deviation from C2h symmetry in the
lutein molecule leads to orbital symmetry breaking (deviation from or com-
plete absence of the symmetry of an orbital as compared to the expected one
250 of a linear polyene) for several occupied and unoccupied MOs which seems
to be the main reason for such behavior. MOs from 6 to 4 (occupied orbitals
are numbered from 10 to 1 in ascending order of orbital occupancy in the
initial SS-RASSCF, unoccupied orbitals – from 1’ to 10’ in descending order
of orbitals occupancy in the same calculation) and the corresponding unoccu-

12
 1, Au 2, Bg 3, Au

4, bs 5, bs 6, bs

7, Bg 8, Au 9, Bg 10, Au

Figure 2: Occupied active space MOs of the lutein molecule obtained by SS-RASSCF.
The same numbering as in the text were used, approximate symmetry in C2h group is
indicated. Bs means broken symmetry. MOs were rendered in LUSCUS program [44].

13
255 pied ones are the product of mixing of MOs of ideal 9-ene with the π-orbital
of cyclohexene ring (fig. 2) which is pulled out of the conjugated π-system
by 48◦ and, thus, has a small overlap with it. The [6,6] active space con-
tains doubly occupied MOs from 3 to 1 and does not have any orbitals mixed
with the out-of-plane π-orbital. Thus, the results of MCSCF closely resemble
260 (fig. S3) that for an ideal polyene with the forbidden 1A− −
g → 2Ag transition

which leads to underestimation of transition dipole value (9·10−3 D). Further

expansion of the active space gradually incorporates MOs with broken sym-
metry which, in absence of lower lying orbitals, destabilize MCSCF solution
which produce MOs substantially localized in the cyclohexene π-system (fig.
265 S4). In this case, the wavefunction of the 2A−
g state is contaminated by φ2110

CSF of Bu symmetry (table S1) which has non-zero contribution to transi-

ˆ 2110 i. This affects the transition dipole
tion dipole matrix elements hΨ1A−g |d|φ
value which gradually increases upon inclusion of new orbitals of broken
symmetry(from 0.61 to 0.71 D). The orbital 7 (fig. 2) has a nearly ideal C2h
270 symmetry due to small contribution from the cyclohexene π-orbital. When
included into the active space (CASSCF[14,14]), it changes the behavior of
MCSCF procedure again decreasing transition dipole down to 6.2·10−2 D.
Basing only on CASSCF results, any reliable conclusions could not be made
about the transition dipole value due to the technical impossibility of fur-
275 ther expansion of complete active space. The main question is whether the
transition dipole value is close to the CASSCF[8,8]-CASSCF[12,12] results
or the CASSCF[14,14] result. The lack of experimental data does not allow
to make a conclusion at this step, thus, active space should be expanded in
some way. Even in the case of the largest CASSCF[14,14] the occupancies of

14
280 the orbitals with the highest occupancies and with the lowest ones for 2A−
g

state are far (table S2) from common criteria for exclusion of orbitals from
active space (> 1.98 and < 0.02, respectively [45]) which emphasizes the
importance of further active space expansion.

CAS RAS(s) RAS(sd)

CAS size RAS2 size
∆E01 , eV |d~01 |, D ∆E01 , eV |d~01 |, D ∆E01 , eV |d~01 |, D
6 4.38 9.11·10−3 4 3.45 0.156 3.71 0.257
−2
8 4.40 0.612 6 3.36 2.44·10 3.47 6.76·10−2
10 4.35 0.643 8 3.32 3.81·10−2 3.35 5.59·10−2
12 4.26 0.714
-
14 3.87 6.20·10−2

Table 1: Excitation energies and norms of transition dipole moments between the ground
and the 2A−
g electronic states of the lutein molecule obtained in various active spaces.

The entire conjugated π-system of lutein was treated using RASSCF for-
285 malism which can deal with large active spaces with restricted electronic ex-
citations. RAS2 active was systematically expanded while RAS1 and RAS3
contained the remaining MOs (for details see section 2.1.1). MCSCF cal-
culations with the smallest RAS2 active space (4 orbitals) produced much
higher dipole values than the others possibly due to impossibility to incorpo-
290 rate all significant CSFs. Expansion of RAS2 leads to substantial decrease
of transition dipole moment which has the same order of magnitude as in
CASSCF[14,14]. Also, in all studied active spaces (table 1), RASSCF with
higher number of holes in RAS1/electrons in RAS3 gave higher dipole values
reflecting higher contribution from broken symmetry orbitals. In the absence
295 of experimental reference, it is interesting to compare the MCSCF transition
dipole values with those from other computational methods. Values obtained

15
 by RASSCF method with large active spaces (0.03-0.07 D) deviate signifi-
cantly (by one order of magnitude) from the DFT/MRCI result (0.767 D
[23]). However, DFT/MRCI is known to yield the results different from
300 those of the state-of-the-art references such as multireference perturbation
theories, especially for dipole-forbidden transitions, and such references are
absent for xanthophylls. It should be noted that RASSCF can not serve as
such reference. First, it does not account for static correlation in the π-system
to the same extent as the CASSCF with the entire π-system included into
305 the active space due to the limited number of excitations from the RAS1 to
the RAS3 subspace. Second, it lacks any dynamic correlation contribution.
The latter factor, however, has more effect on the energy value and less so
on the electron density.

3.2. Exciton coupling in the pigment pair

310 Exciton coupling VLut−Chl in the Lut620/Chla612 dimer relevant to NPQ

in LHCII was calculated as described in section 2.4. The coupling depends
on the active space size to a much lesser extent than the transition dipole
moment (fig. 3) which is due to close proximity of the pigments in the dimer.
The pigments reside in parallel planes with interplane distance being approx-
315 imately 4 Å. In this case, dipole-dipole approximation is not applicable so the
coupling is defined by spatial distribution of transition densities. Therefore,
similar coupling values indicate that all calculations produce qualitatively
the same transition density between the 1A− −
g and the 2Ag states of lutein in

the center of the molecule where the distance between the pigments is the
320 smallest. Exciton coupling significantly depends on the distance between the
pigments as expected. Increasing the distance between pigments’ planes by

16
 Figure 3: Exciton coupling VLut−Chl calculated on the various TrESP charges for lutein.
Red line corresponds to the MM equilibrium geometry of the pigment dimer in the LHCII
complex. Blue line corresponds to the pigments in the dimer pair pushed apart by 1 Å
with respect to equilibrium geometry, green one – pushed by 3 Å. Active space size in
CASSCF and RASSCF is given in brackets.

3 Å is enough to decrease the coupling below 10 cm−1 .

As can be seen from the fig. 3, the coupling value converges to a limit
with the increasing active space size. Results obtained by RASSCF method
325 within the RAS(s) formulation converge monotonously to the value of 19.2
cm−1 . The same is valid for the RAS(sd) results with the limit coupling value
being 21.9 cm−1 . Although the used active spaces are much smaller than
the CASSCF limit for the entire lutein π-system, the nature of convergence
implies that the coupling in the limit case would not be very different from
330 the results obtained by RASSCF with large active spaces. Such relatively

17
 large value (for a coupling with a dipole-forbidden state) gives additional
support to the hypothesis that LHCII is in a highly quenched state in its X-
ray conformation [46]. Direct comparison of exciton coupling with previous
papers is not possible due to slightly different pigment pair geometries. This
335 value is much higher than the value of 2 cm−1 reported by Balevicius et al.
[15] and higher than the same coupling calculated with MNDO-CAS-CI [13]
by a factor of 1.6 which can be explained by the difference in computational
methods and the dimer geometries. To eliminate the geometry factor, semi-
empirical calculations on the same dimer geometries were carried out; the
340 results are reported in section 3.3.3.

3.3. Chlorophyll - lutein exciton coupling as a function of internal coordinates

Dependence of the coupling in the Lut620/Chla612 dimer on the relative

positions of the pigments is essential for NPQ regulation mechanism. Two
internal movements in the dimer, namely the rotation of the lutein molecule
345 in the plane parallel to the chlorophyll molecule and the angle change between
the lutein axis and the chlorophyll plane, are supposed to be involved in
NPQ control in the LHCII antenna [15]. Basing on the ab initio model
for the coupling, its dependence on three internal coordinates including the
two noted above is discussed further (for exact definitions of rotation angles
350 see Supplementary, section S4). Apart from regulatory implications of such
dependence, the study is aimed to discriminate between various MCSCF
formulations with regards to the transition density distributions along the
lutein molecule.

18
 3.3.1. Rotation of the chlorophyll molecule
355 The first coordinate is the angle of rotation of the chlorophyll a molecule
within its plane around the axis centered at magnesium atom. This coordi-
nate is not directly related to NPQ activation but has a significant impact on
the coupling (fig. 4) due to that the chlorophyll transition density is mostly
located on the B ring close to the lutein molecule (the corresponding TrESP
360 charges are provided in Supplementary, fig. S7).
Exciton couplings calculated basing on CASSCF exhibit inconsistent be-
havior (fig. 4b) upon expansion of the active space. Two active spaces,
namely [8,8] and [12,12], produce similar transition density distributions as
indicated by similar coupling plots. Three other spaces, namely [6,6], [10,10],
365 and [14,14], produce the angle-coupling functions dissimilar to one another
as well as to the previous two. This can be viewed as an additional evidence
that CASSCF with small active space is not capable to describe the 2A−
g

state of the lutein molecule properly. Therefore, any CASSCF-based con-

clusions regarding the NPQ rate in the chlorophyll-lutein dimer should be
370 treated with caution.
The angle-coupling dependences for RASSCF density matrices exhibit a
more stable behavior upon expansion of the active space (fig. 4c). Within
the rotation angle range of ±90 degrees, the mean deviation between the
couplings for RAS spaces of different size is 6.46 ± 2.68 cm−1 . If the two
375 smallest RASSCF spaces, namely [8,4,8,s] and [8,4,8,sd], are excluded, the
deviation decreases to 3.57 ± 1.03 cm−1 . This shows that RASSCF method
for active spaces of different sizes gives internally consistent results which
approach a limit when the number of CSFs grows thus providing reliable

19
 b)

a) c)

Figure 4: Internal coordinate #1 – in-plane rotation of the chlorophyll a molecule. a)

Structure of the Lut620/Chla612 dimer. Chlorophyll is shown in green, lutein – in red.
Arrow shows the direction of the pigment rotation. b,c) Exciton coupling as a function of
the chlorophyll rotation angle calculated based on CASSCF (b) or RASSCF (c)

transition densities.

380 3.3.2. Rotations of the lutein molecule

Two internal coordinates were studied: a) angle of rotation of the lutein
molecule in the plane of its own π-system with respect to the axis centered
on the middle of the C15-C35 bond and b) tilt angle of the lutein π-system
axis with respect to the chlorophyll molecule plane (see fig. 5a,c). These two
385 coordinates are supposed to be involved in NPQ activation [15].

20
 a) b)

c) d)

Figure 5: Internal coordinates #2 and #3 – in-plane and out-of-plane rotation of the lutein
a molecule. a,c) Structure of the Lut620/Chla612 dimer. Chlorophyll is shown in green,
lutein – in red; translucent lutein molecules indicate the range of angles. b,d) RASSCF-
based exciton couplings as a function of b) in-plane rotation angle and d) out-of-plane tilt
angle.

Considering the instability of CASSCF-based results shown previously,

we focused on RASSCF. It can be seen from fig. 5b,d that exciton couplings
have a moderate dependence on these internal coordinates and behave con-
sistently for active spaces of different sizes. In-plane rotation of lutein by
390 35◦ leads to significant increase of the exciton coupling (up to 32 cm−1 in

21
 [6,8,6,sd]) but it is not clear whether the dimer can really adopt such con-
formation. Assuming that ±20◦ range of rotation and tilt angles from the
equilibrium value is accessible in the real protein (for the tilt angle even
that would be an overestimation due to Chl-Lut sterical clashes), the range
395 of possible coupling values in the dimer is rather narrow. Within the said
range of the in-plane rotation angle, with the RASSCF active spaces [8,4,8,s]
and [8,4,8,sd] excluded, the coupling value falls in the range between 19
and 24 cm−1 . Within the same range of the tilt angle, the coupling stays
in the range from 18 to 24 cm−1 . This result contrasts with a more pro-
400 nounced angle-coupling dependence obtained by the AM1-CAS-CI method
[15]. Moreover, rotation and tilt-related variations of exciton coupling are
much lower than that introduced by the change of the interplane distance
which is likely accompany any structural change within the Lut620/Chla612
dimer. Nevertheless, all internal movements in the pigment pair combined
405 can have a significant impact on the exciton coupling and, thus, underlie the
NPQ regulation in the LHCII complex.

3.3.3. Comparison with the semi-empirical model

The same coupling-angle plots were calculated using AM1-CAS-CI to
410 estimate the impact of the computational method on the model of NPQ
process. The obtained semi-empirical coupling-angle plot for chlorophyll ro-
tation (fig. 6, red line) is in a qualitative agreement with the one provided in
the paper of Balevicius et al. [15]: it reproduces all specific features such as
the minimum near −120 degrees, the broad maximum near 70 degrees, and
415 the plateau near 0 degrees. The difference of the maximal positive coupling

22
 value (∼ 13 cm−1 versus ∼ 20 cm−1 ) arises predominantly from the absence
of rescaling in our model and much less from the difference of geometries.
Indeed, if rescaling factor 1.67 (taken from supplementary of [15]), maximal
positive value of the coupling becomes 22 cm−1 which is much closer to the
420 cited value. Since the geometry difference cannot be eliminated completely,
further discussion will focus on the results obtained using our geometry.

Figure 6: Exciton coupling as a function of rotation angle of the chlorophyll molecule

(as described above, fig. 4a). A/B denotes computational method for lutein/chlorophyll,
respectively. AM1 stands for AM1-CAS-CI, ab i. stands for RASSCF in the largest active
space [6,8,6,sd].

In the range of physiologically accessible rotation angles AM1-CAS-CI

and RASSCF couplings (fig. 6, red and purple lines, respectively) behave
differently. While the semi-empirical values are close to zero and can vary
425 severalfold, the ab initio couplings are already relatively large (for this sys-
tem) in the equilibrium geometry and can vary by no more that two times.

23
 This facts support the hypothesis that the change of the interpigment dis-
tance in the dimer is important for NPQ. However, it should be noted that
the volume change corresponding to the transition to the quenched state is
430 very small (∼ 0.006% [47]) which limits the contribution of the interpigment
distance change in the NPQ activation.
It can be useful to calculate exciton couplings basing on ab initio TrESP
charges for one pigment and semi-emprical ones for another pigment. Al-
though, it produces inconsistent and, thus, unphysical coupling it can be
435 used to find out which pigment requires computationally demanding ab ini-
tio treatment to produce accurate results. If the chlorophyll only is described
by AM1-CAS-CI (fig. 6, green line), the coupling-angle dependence is much
closer to the RASSCF plot than one for the semi-empirically modeled lutein.
This is in complete agreement with the fact that the HOMO→LUMO tran-
440 sition in chlorophyll is much easier to describe accurately than the transition
from the ground to the multiconfigurational 2A−
g state of the lutein molecule

and emphasizes the importance of the accurate model for electronic structure
of lutein. The two methods produce transition densities which are located
differently within the lutein molecule. For AM1, transition density is lo-
445 calized on one end of the π-system which is far from the Chl-Lut contact
region. So, a significant rotation angle is required to achieve considerable
coupling. For RASSCF, transition density is distributed more evenly, has
an inversion pseudosymmetry and produces the TrESP charges which are
nonzero in the region of close contact. The visualization of TrESP charges
450 derived from AM1 and the largest RASSCF for both pigments is provided
in the Supplementary (fig. S7). The plots for the two remaining internal

24
 coordinates (rotations of lutein) exhibit the behavior similar to the one for
chlorophyll rotation (near-zero values for AM1 and nonzero values for ab ini-
tio at equilibrium geometry) and are less representative, thus they are shown
455 in Supplementary (fig. S8).

4. Conclusion

Multiconfigurational nature of the dark 2A−

g of the lutein molecule makes

it necessary to use multireference methods for its accurate description. As

shown in this paper, it is also necessary to include the entire conjugated π-
460 system of lutein into the active space which makes the computational effort
required to perform a CASSCF calculation prohibitive. Thus, multirefer-
ence methods using incomplete active spaces with an appropriate method
for CSF selection are necessary for such calculations. With active spaces of
moderate sizes, the RASSCF method yields consistent 1A− −
g →2Ag transi-

465 tion properties including transition dipole and transition density. Basing on
the RASSCF results, the value of exciton coupling in the Lut620/Chla612
dimer in the LHCII complex was found to be 21.9 cm−1 . This value is much
higher than the MNDO/AM1-CAS-CI results reported previously [13, 15].
Moreover, the sensitivity of this coupling to internal movements in the dimer,
470 namely the lutein rotation and tilt angles, is quite different within RASSCF
and semi-empirical approaches. Due to high sensitivity of lutein transition
properties to the choice of computational method, any conclusions regard-
ing exciton coupling value should be made cautiously. Since the excitation
energy transfer rate has a quadratic dependence on the coupling, the lat-
475 ter consideration is also important in studying of NPQ regulation. Despite

25
 the supposed accuracy of the reported ab initio methods, they have a major
drawback, namely the prohibitive computational complexity when applied
to multiple frames of MD trajectories. We suppose that this problem can be
addressed either by some parametrization of transitional electronic density of
480 the pigments which accounts for changes of their geometries or by reducing
the phase space in MD to a limited set of conformations. For this, further
research is required.

5. Acknowledgements

This work was supported by Russian Foundation for Basic Research [grant
485 number 18-34-00700]. Authors thank Dr. Christopher Duffy (Queen Mary
University of London) for insightful discussion about NPQ model and Dr.
Ilya Glebov (Lomonosov Moscow State University) for the help with the
interpretation of MCSCF results.

[1] C. Triantaphylidès, M. Havaux, Singlet oxygen in plants: production,

490 detoxification and signaling, Trends in Plant Science 14 (4) (2009) 219–
228. doi:10.1016/j.tplants.2009.01.008.

[2] I. Vass, Molecular mechanisms of photodamage in the Photosystem II

complex, Biochimica et Biophysica Acta - Bioenergetics 1817 (1) (2012)
209–217. doi:10.1016/j.bbabio.2011.04.014.

495 [3] A. Young, G. Britton, Carotenoids in Photosynthesis, Springer Sci-

ence+Business Media, 1993. doi:10.1007/978-94-011-2124-8.

[4] P. Tavan, K. Schulten, Electronic excitations in finite and infinite

26
 polyenes, Physical Review B 36 (8) (1986) 4337–4358. doi:10.1103/
PhysRevB.36.4337.

500 [5] M. Gouterman, G. H. Wagnière, L. C. Snyder, Spectra of porphyrins:

Part II. Four orbital model, Journal of Molecular Spectroscopy 11 (1-6)
(1963) 108–127. doi:10.1016/0022-2852(63)90011-0.

[6] K. Furuichi, T. Sashima, Y. Koyama, The first detection of the

3Ag-state in carotenoids using resonance-Raman excitation profiles,
505 Chemical Physics Letters 356 (5-6) (2002) 547–555. doi:10.1016/
S0009-2614(02)00412-8.

[7] P. J. Walla, P. A. Linden, K. Ohta, G. R. Fleming, Excited-state kinetics

of the carotenoid S1 state in LHCII and two-photon excitation spectra of
lutein and β-carotene in solution: Efficient Car S1 Chl electronic energy
510 transfer via hot S1 states?, Journal of Physical Chemistry A 106 (10)
(2002) 1909–1916. doi:10.1021/jp011495x.

[8] T. Polı́vka, V. Sundström, Ultrafast Dynamics of Carotenoid Excited

States From Solution to Natural and Artificial Systems, Chemical Re-
views 104 (2004) 2021–2071. doi:10.1021/cr020674n.

515 [9] K. F. Fox, V. Balevičius, J. Chmeliov, L. Valkunas, A. V. Ruban,

C. D. P. Duffy, The carotenoid pathway: what is important for excitation
quenching in plant antenna complexes?, Physical Chemistry Chemical
Physics 19 (34) (2017) 22957–22968. doi:10.1039/C7CP03535G.

[10] H. van Amerongen, R. van Grondelle, Understanding the energy transfer

520 function of lhcii, the major light-harvesting complex of green plants,

27
 The Journal of Physical Chemistry B 105 (3) (2001) 604–617. doi:
10.1021/jp0028406.

[11] R. Agarwal, B. P. Krueger, G. D. Scholes, M. Yang, J. Yom, L. Mets,

G. R. Fleming, Ultrafast energy transfer in lhc-ii revealed by three-
525 pulse photon echo peak shift measurements, The Journal of Physical
Chemistry B 104 (13) (2000) 2908–2918. doi:10.1021/jp9915578.

[12] A. V. Ruban, R. Berera, C. Ilioaia, I. H. Van Stokkum, J. T. Ken-

nis, A. A. Pascal, H. Van Amerongen, B. Robert, P. Horton, R. Van
Grondelle, Identification of a mechanism of photoprotective energy dis-
530 sipation in higher plants, Nature 450 (7169) (2007) 575–578. doi:
10.1038/nature06262.

[13] C. D. Duffy, J. Chmeliov, M. Macernis, J. Sulskus, L. Valkunas, A. V.

Ruban, Modeling of fluorescence quenching by lutein in the plant light-
harvesting complex LHCII, Journal of Physical Chemistry B 117 (38)
535 (2013) 10974–10986. doi:10.1021/jp3110997.

[14] J. Chmeliov, W. P. Bricker, C. Lo, E. Jouin, L. Valkunas, A. V. Ruban,

C. D. Duffy, An ’all pigment’ model of excitation quenching in LHCII,
Physical Chemistry Chemical Physics 17 (24) (2015) 15857–15867. doi:
10.1039/c5cp01905b.

540 [15] V. Balevičius, K. F. Fox, W. P. Bricker, S. Jurinovich, I. G. Prandi,

B. Mennucci, C. D. Duffy, Fine control of chlorophyll-carotenoid interac-
tions defines the functionality of light-harvesting proteins in plants, Sci-
entific Reports 7 (1) (2017) 1–10. doi:10.1038/s41598-017-13720-6.

28
 [16] Z. Liu, H. Yan, K. Wang, T. Kuang, J. Zhang, L. Gui, X. An,
545 W. Chang, Crystal structure of spinach major light-harvesting com-
plex at 2.72 Å resolution, Nature 428 (6980) (2004) 287–292. doi:
10.1038/nature02373.

[17] V. I. Novoderezhkin, M. A. Palacios, H. Van Amerongen, R. Van Gron-

delle, Excitation dynamics in the LHCII complex of higher plants: Mod-
550 eling based on the 2.72 Å crystal structure, Journal of Physical Chem-
istry B 109 (20) (2005) 10493–10504. doi:10.1021/jp044082f.

[18] W. Humphrey, A. Dalke, K. Schulten, VMD: Visual molecular dy-

namics, Journal of Molecular Graphics 14 (1) (1996) 33–38. doi:
10.1016/0263-7855(96)00018-5.

555 [19] Z.-L. Cai, M. J. Crossley, J. R. Reimers, R. Kobayashi, R. D. Amos,

Density Functional Theory for Charge Transfer: The Nature of the N-
Bands of Porphyrins and Chlorophylls Revealed through CAM-B3LYP,
CASPT2, and SAC-CI Calculations, The Journal of Physical Chemistry
B 110 (31) (2006) 15624–15632. doi:10.1021/jp063376t.

560 [20] J. H. Starcke, M. Wormit, J. Schirmer, A. Dreuw, How much double

excitation character do the lowest excited states of linear polyenes have?,
Chemical Physics 329 (1-3) (2006) 39–49. doi:10.1016/j.chemphys.
2006.07.020.

[21] K. Schulten, M. Karplus, On the origin of a low-lying forbidden transi-

565 tion in polyenes and related molecules, Chemical Physics Letters 14 (3)
(1972) 305–309. doi:10.1016/0009-2614(72)80120-9.

29
 [22] M. Macernis, J. Sulskus, C. D. P. Duffy, A. V. Ruban, L. Valkunas, Elec-
tronic Spectra of Structurally Deformed Lutein, The Journal of Physical
Chemistry A 116 (40) (2012) 9843–9853. doi:10.1021/jp304363q.

570 [23] O. Andreussi, S. Knecht, C. M. Marian, J. Kongsted, B. Mennucci,

Carotenoids and light-harvesting: From DFT/MRCI to the Tamm-
Dancoff approximation, Journal of Chemical Theory and Computation
11 (2) (2015) 655–666. doi:10.1021/ct5011246.

[24] M. Kleinschmidt, C. M. Marian, M. Waletzke, S. Grimme, Parallel mul-

575 tireference configuration interaction calculations on mini-β-carotenes
and β-carotene, The Journal of Chemical Physics 130 (4) (2009) 044708.
doi:10.1063/1.3062842.

[25] C. M. Marian, M. Kleinschmidt, A. R. Holzwarth, E. Ostroumov, M. G.

Mu, Electronic Coherence Provides a Direct Proof for Energy-Level
580 Crossing in Photoexcited Lutein and β-Carotene 108302 (September)
(2009) 1–4. doi:10.1103/PhysRevLett.103.108302.

[26] D. Ghosh, J. Hachmann, T. Yanai, G. K.-L. Chan, Orbital optimiza-

tion in the density matrix renormalization group, with applications to
polyenes and β-carotene, The Journal of Chemical Physics 128 (14)
585 (2008) 144117. doi:10.1063/1.2883976.

[27] A. Linden, B. Bruno, C. H. Eugster, Confirmation of the Structures

of Lutein and Zeaxanthin 87 (2004) 1254–1269. doi:10.1002/hlca.
200490115.

30
 [28] P.-O. Widmark, B. J. Persson, B. O. Roos, Density matrix averaged
590 atomic natural orbital (ano) basis sets for correlated molecular wave
functions, Theoretica chimica acta 79 (6) (1991) 419–432. doi:10.1007/
BF01112569.

[29] P. A. Malmqvist, A. Rendell, B. O. Roos, The restricted active space self-

consistent-field method, implemented with a split graph unitary group
595 approach, The Journal of Physical Chemistry 94 (14) (1990) 5477–5482.
doi:10.1021/j100377a011.

[30] R. S. Knox, B. Q. Spring, Dipole Strengths in the Chlorophylls, Pho-

tochemistry and Photobiology 77 (5) (2007) 497–501. doi:10.1562/
0031-8655(2003)0770497DSITC2.0.CO2.

600 [31] J. J. P. Stewart, Mopac2016, stewart Computational Chemistry, Col-

orado Springs (2016).
URL http://openmopac.net

[32] B. H. Besler, K. M. Merz, P. A. Kollman, Atomic charges derived from

semiempirical methods, Journal of Computational Chemistry 11 (4)
605 (1990) 431–439. doi:10.1002/jcc.540110404.

[33] M. W. Schmidt, K. K. Baldridge, J. A. Boatz, S. T. Elbert, M. S.

Gordon, J. H. Jensen, S. Koseki, N. Matsunaga, K. A. Nguyen, S. Su,
T. L. Windus, M. Dupuis, J. A. Montgomery Jr, General atomic and
molecular electronic structure system, J. Comput. Chem. 14 (11) (1993)
610 1347. doi:10.1002/jcc.540141112.

31
 [34] F. Aquilante, J. Autschbach, R. K. Carlson, L. F. Chibotaru, M. G.
Delcey, L. De Vico, I. Fdez. Galvn, N. Ferr, L. M. Frutos, L. Gagliardi,
M. Garavelli, A. Giussani, C. E. Hoyer, G. Li Manni, H. Lischka, D. Ma,
P. . Malmqvist, T. Mller, A. Nenov, M. Olivucci, T. B. Pedersen,
615 D. Peng, F. Plasser, B. Pritchard, M. Reiher, I. Rivalta, I. Schapiro,
J. Segarra-Mart, M. Stenrup, D. G. Truhlar, L. Ungur, A. Valentini,
S. Vancoillie, V. Veryazov, V. P. Vysotskiy, O. Weingart, F. Zapata,
R. Lindh, Molcas 8: New capabilities for multiconfigurational quantum
chemical calculations across the periodic table, Journal of Computa-
620 tional Chemistry 37 (5) 506–541. doi:10.1002/jcc.24221.

[35] F. Aquilante, R. Lindh, T. B. Pedersen, Unbiased auxiliary basis sets for

accurate two-electron integral approximations, The Journal of Chemical
Physics 127 (11) (2007) 114107. doi:10.1063/1.2777146.

[36] D. Case, S. Brozell, D. Cerutti, T. Cheatham III, V. Cruzeiro, T. Dar-

625 den, R. Duke, D. Ghoreishi, H. Gohlke, A. Goetz, D. Greene, R. Har-
ris, N. Homeyer, S. Izadi, A. Kovalenko, T. Lee, S. LeGrand, P. Li,
C. Lin, J. Liu, T. Luchko, R. Luo, D. Mermelstein, K. Merz, Y. Miao,
G. Monard, H. Nguyen, I. Omelyan, A. Onufriev, F. Pan, R. Qi, D. Roe,
A. Roitberg, C. Sagui, S. Schott-Verdugo, J. Shen, C. Simmerling,
630 J. Smith, J. Swails, R. Walker, J. Wang, H. Wei, R. Wolf, X. Wu,
L. Xiao, D. York, P. Kollman, Amber 2018, university of California, San
Francisco (2018).

[37] A. M. Rosnik, C. Curutchet, Theoretical Characterization of the Spec-

tral Density of the Water-Soluble Chlorophyll-Binding Protein from

32
635 Combined Quantum Mechanics/Molecular Mechanics Molecular Dy-
namics Simulations, Journal of Chemical Theory and Computation
11 (12) (2015) 5826–5837. doi:10.1021/acs.jctc.5b00891.

[38] I. G. Prandi, L. Viani, O. Andreussi, B. Mennucci, Combining classical

molecular dynamics and quantum mechanical methods for the descrip-
640 tion of electronic excitations: The case of carotenoids, Journal of Com-
putational Chemistry 37 (11) (2016) 981–991. doi:10.1002/jcc.24286.

[39] J. M. Wang, R. M. Wolf, J. W. Caldwell, P. Kollman, D. Case, Devel-

opment and testing of a general amber force field, J. Comput. Chem.
25 (9) (2004) 1157–1174. doi:10.1002/jcc.20035.

645 [40] W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, M. L.

Klein, Comparison of simple potential functions for simulating liquid
water, The Journal of Chemical Physics 79 (2) (1983) 926–935. doi:
10.1063/1.445869.

[41] M. E. Madjet, A. Abdurahman, T. Renger, Intermolecular coulomb

650 couplings from ab initio electrostatic potentials: Application to opti-
cal transitions of strongly coupled pigments in photosynthetic antennae
and reaction centers, Journal of Physical Chemistry B 110 (34) (2006)
17268–17281. doi:10.1021/jp0615398.

[42] B. P. Krueger, G. D. Scholes, I. R. Gould, G. R. Fleming, Carotenoid

655 mediated B800B850 coupling in LH2, PhysChemComm 2 (8) (1999)
34–40. doi:10.1039/A903172C.

33
 [43] B. Blasiak, M. Maj, M. Cho, R. W. Góra, Distributed Multipolar Expan-
sion Approach to Calculation of Excitation Energy Transfer Couplings,
Journal of Chemical Theory and Computation 11 (7) (2015) 3259–3266.
660 doi:10.1021/acs.jctc.5b00216.

[44] G. Kovačević, V. Veryazov, Luscus: molecular viewer and editor for

MOLCAS, Journal of Cheminformatics 7 (1) (2015) 16. doi:10.1186/
s13321-015-0060-z.

[45] V. Veryazov, P. Å. Malmqvist, B. O. Roos, How to select active space

665 for multiconfigurational quantum chemistry?, International Journal of
Quantum Chemistry 111 (13) (2011) 3329–3338. doi:10.1002/qua.
23068.
URL http://doi.wiley.com/10.1002/qua.23068

[46] A. A. Pascal, Z. Liu, K. Broess, B. van Oort, H. van Amerongen,

670 C. Wang, P. Horton, B. Robert, W. Chang, A. Ruban, Molecular basis of
photoprotection and control of photosynthetic light-harvesting, Nature
436 (7047) (2005) 134–137. doi:10.1038/nature03795.

[47] S. Santabarbara, P. Horton, A. V. Ruban, Comparison of the thermody-

namic landscapes of unfolding and formation of the energy dissipative
675 state in the isolated light harvesting complex II, Biophysical Journal
97 (4) (2009) 1188–1197. doi:10.1016/j.bpj.2009.06.005.

34