Conformational Plasticity Enables Functional Switching in Diatom Light-Harvesting Complexes

Conformational Plasticity Enables Functional Switching in Diatom Light-Harvesting Complexes | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Conformational Plasticity Enables Functional Switching in Diatom Light-Harvesting Complexes Theofani-Iosifina Sousani*, Boutheina Zender*, Sayan Maity, Ulrich Kleinekathöfer, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7326805/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 05 Dec, 2025 Read the published version in Communications Chemistry → Version 1 posted You are reading this latest preprint version Abstract Biomolecules exhibit a fundamental correlation between structure and function, which can be modulated by environmental factors. Deciphering this relationship remains a central and long-standing challenge for many protein families. In this study, we investigate such a correlation in the light-harvesting complexes (LHCs) of diatoms; unicellular, photosynthetic organisms that thrive in marine ecosystems. Using μs-long molecular dynamics simulations and machine learning, we reveal that all experimentally resolved LHC configurations correspond to a few distinct interconverting states linked to an intrinsic transition between light-harvesting and photoprotective mode; a property that can be tuned or engineered. Thus, we provide an original view on the plethora of experimentally resolved structures. Our model strongly correlates with experimental findings on the effect of the photoprotective protein LHCX1 and the xanthophyll cycle on the FCP acclimation states. *Theofani-Iosifina Sousani & Boutheina Zender contributed equally to this work. Physical sciences/Chemistry/Chemical biology/Biophysical chemistry Biological sciences/Structural biology/Molecular modelling Figures Figure 1 Figure 2 Figure 3 Figure 4 Introduction Photosynthesis is a vital natural process through which solar energy is captured and transformed into chemical energy, defining the basis of life on Earth 1 . Organisms capable of photosynthesis are generally grouped into green and red evolutionary lineages 2 . Despite their diversity, green plants and algae share common structural features in their photosynthetic systems. Light energy, in the form of photons, is initially absorbed by pigments like chlorophylls and carotenoids found in light-harvesting complexes (LHCs) – protein-pigment assemblies located within the thylakoid membrane of chloroplasts 1 Varying environmental factors such as changes in light intensity or spectrum, often influenced by daily light cycles, can affect this process. To manage these fluctuations and avoid damage from excess light, green plants and algae use (down) regulatory strategies that adjust light absorption. One key protective response is non-photochemical quenching (NPQ), a mechanism by which surplus excitation energy is safely dissipated as heat to prevent photodamage 3 . Diatoms in the red lineage of photosynthetic organisms are highly important as primary producers in marine ecosystems, supplying organic substrates and oxygen, by rapidly adjusting light harvesting and photochemistry to cell productivity. Therefore, they are considered an ideal model organism for studying photosynthetic regulation 4 . Diatoms express fucoxanthin (Fx) and chlorophyll-a/c (Chl) binding proteins (FCPs) as LHCs. FCPs are responsible for the exceptional light-harvesting capabilities of diatoms in the blue-green region, which is available underwater. However, they also have a robust intrinsic photoprotective mechanism that enables them to adapt to fluctuating light at the ocean surface 5,6 . The details of this mechanism at all-atom resolution are still obscure 6 . Regulation of photosynthesis is a promising area of research for improving crops and increasing biomass or boosting the atmospheric CO 2 assimilation to ease the strangle against global warming 7–10 . The fundamental research on the downregulatory mechanisms of photosynthesis could lay the groundwork for artificial photosynthesis 11 . and could also provide crucial knowledge towards the increased production of biofuels and nutrients. FCPs come in monomers or various oligomeric states such as dimers, trimers, tetramers, or even pentamers 5,12–18 . The plethora of structural insights into FCPs calls for a comprehensive investigation to answer the following question: What is the connection between FCP structure and function for a remarkable adaptability? To answer this question, we implemented a comprehensive analytical framework which integrates extensive conformational sampling at the classical Molecular Dynamics (MD) level, exciton coupling calculations between pigments, statistical tools like Markov State Modeling (MSM), data from experimentally resolved structures and machine learning (ML). This integrated computational approach allows for a quantitative assessment of how FCP conformation changes in different acclimation states. Results The configurational space of light harvesting antennas We have complemented previous MD simulations on the FCP complex from Ph. Tricornutum 14,19 with dynamics of FCP complexes from Ch. Gracilis (Sm1 complex in monomeric and tetrameric forms and m2 complex in monomeric forms) 13 . The setup and run of the MD simulations of Ch. Gracilis followed exactly the protocols in our previous studies of Ph. Tricornutum 19,20 . All FCPs were sampled at two different pH states (neutral and acidic (pH ~5.5) lumenal environments) 19,20 and different multimerization states resulting in a cumulative simulation time of 80μs monomer-equivalent dynamics for the FCPs from the two species. For details refer to the Methods section. Consistency was ensured in the analysis by truncating all trajectories to a common core of 111 residues (444 atoms in the mainchain – backbone) that is shared among both diatoms, identified by a multi-sequence alignment method (MUSTANG) 21 . Only residues aa 30-51, 54-116, 132-157 were considered for the FCP from Ph. Tricornutum 14 . For Ch. Gracilis residues aa 58-79, 83-145, 155-180 were considered for the Sm1 in monomeric and tetrameric forms and aa 58-79, 83-145, 158-183 for the m2 monomer 13 . Interestingly, this minimum core of 111 residues matches a continues sequence in the dimeric FCP fold (H1/ H2) found in the diatom Cy. meneghiniana (aa 64-174) 22 . All residue numbers reported hereafter refer to the numbering in the sequence of H1/ H2. Cryo-EM (cryogenic electron microscopy) and X-ray crystallographic data were also collected from the protein data bank (pdb). Structures from diverse FCPs in various oligomeric states are available from different diatom species (Ph. tricornutum: 6A2W; Ch. gracilis: 6J3Y, 6J3Z, 6J40, 6JLU, 6L4U, 7VD5, 7VD6, 8WCK, 8WCL; Th. pseudonana: 8IWH, 8J0D, 8JP3; Cy. meneghiniana: 8J5K, 8J7Z, 8W4O, 8W4P; Ch. roscoffensis: 9KQB). Dimers, tetramers, or trimers were disassembled into 118 different monomeric units before the analysis. Thus, two primary datasets will be analyzed in the following, i.e., set1: MD frames and set2: experimental structures. A Markov state model (MSM) was constructed from the MD trajectories, which can predict the long-time scale behavior of a biomolecular system 23 . For further details refer to the Methods. The reweighted free energy surface (FES) was projected onto the slowest degrees of freedom of the FCP common core, determined by the time-lagged independent component analysis (tICA) procedure and described by the two vectors IC-1 and IC-2. This is shown in Fig. 1A , alongside the positions of the four identified kinetically distinct states 1-4. State-2 is shared between the FCPs of Ph. tricornutum and Ch. gracilis, states-1 and- 3 are assigned to Ph. tricornutum whereas state-4 to Ch. gracilis. Four ensemble averaged conformations (macrostates) that are associated with states 1-4 were also predicted by MSM (C1 to C4). Superimposed conformations of the common core for C1-4 are shown in Fig. 1B . Therein, the FCP scaffold clearly shows a progressive expansion from conformation C1 and C4 to C2 and finally to C3. The mechanism and functional significance of this expansion are not readily available. However, the transitions should be related to pH differences at the lumen FCP side or oligomerization states, given that the MD trajectories sampled FCP complexes at different pH values and oligomeric states (see Methods for FCP model setup). Given also that the combination of MD-MSM provides insight into kinetically distinct conformations of a biomolecular system over the long-time scale 23 , C1-4 should indicate key conformational transitions of the FCP scaffold not explicitly sampled by classical MD. The tICA components or vectors cannot be easily understood in terms of physical parameters that can describe conformational changes of the FCP protein scaffold. We thus have employed an alternative to the tICA dimensionality reduction approach. The configurational space of the MD set1 was reduced into two dimensions: (a) the average (avg.) angle between helices in the common FCP core and (b) the average of the tilts of the same helices with respect to the Z axis defined as the normal on the thylakoid membrane (see Fig. 1B for definitions). In detail, helical segments were identified sequentially by the DSSP method (Define Secondary Structure of Proteins) implemented within the biopython toolbox (Bio.PDB.DSSP) and based on α-helical secondary structure annotations. For each helix, a principal component analysis (PCA) was applied to the Cα atom coordinates employing the scikit-learn toolbox to determine the dominant helical axis. Subsequently, the inter-helical angles and tilt angles with respect to the z-axis were computed. The probability distribution P(x,y) of the MD data set1 over the two reduced dimensions is converted to -ln(P(x,y)) in energy units [kT] (Fig. 1C). To elucidate the relationship between the predicted C1-4 conformations and the experimentally FCP resolved structures across species, we repeated the procedure for data set2 and the results are shown in Fig. 1D . The position of the MSM-predicted conformations or macrostates C1-4 are shown for reference on both plots (Fig. 1C-D). Remarkably, the MSM predictions quantitatively match the minima in the distribution of the experimentally resolved FCP structures from the protein databank (set2). The C1-4 conformations do not correspond to minima sampled by MD (Fig. 1C), as they are long timescale "projections" based on the MD-MSM combination that, however, remarkably fit the experimental data, even for different species of diatoms. In Fig. 1E we show the distribution of each experimentally resolved structure colored by species over the avg. angle and avg. tilt dimensions. As can be observed, there is no dependency of the C1 to C4 conformations on the diatom species. We have to note that the distribution of Fig. 1E is also independent of the resolution of the experimentally resolved structures. The scikit-learn toolbox was employed for the subsequent analysis and machine learning model training. For details refer to the Methods section. The experimentally resolved set2 structures (Fig. 1E) were classified into four clusters identified by color (Fig. 1F) and based on their structural diversity. C1, C2 belong to the same cluster (blue), whereas C4 to the red cluster and C3 to the green cluster. This unsupervised classification was performed by a simple k-means clustering over the avg. angle and avg. tilt features. The classification is not always obvious by just placing an unknown structure alongside the data of Fig. 1F . However, a more robust Random Forest (RF) supervised machine learning model was trained on the experimentally resolved structures along with the four MSM-predicted conformations (C1-4) and verified the original classification (Fig. 1F). The C1 to C4 MSM-predicted conformations, are thus classified into three main clusters (red, blue and green). This latter classification was performed at a higher level of accuracy, employing further dimensionality reduction by Uniform Manifold Approximation and Projection (UMAP), compared to the simpler k-means on the un-reduced data (see Methods for details). The trained RF ML model can thus be employed to characterize accurately any future FCP structure providing also percentages of “blue”, “red”, “orange” or “green” character. Structure and function correlation Could the distinct C1-4 conformations be associated with a robust transition of the FCP scaffold between an efficient light-harvesting and a downregulated photosynthetic state? In order to answer this question, we focus on a special pigment pair in the FCP structure of Ph. tricornutum (Chl-a 409 and carotenoid Fx-301) that has been proposed computationally and experimentally to be involved in such a transition 19,24,25 . Under NPQ conditions, energy transfer from Chl-a to Fx enables a quick dissipation of the excess absorbed energy as heat, through a short-lived excited state of Fx. The excitonic coupling value between Chl-a (m) and Fx (n) is a measure of the efficiency of this energy transfer, with the rate of transfer given by 19 , where is the excitonic coupling between m and n. The values for the pigments of this pair have been calculated in an earlier study on the FCP from Ph. tricornutum from a 860 snapshot trajectory 19 and herein for Ch. gracilis Sm1 monomer (861 snapshots randomly extracted). Low values (10 cm -1 ) are characteristic for photoprotective states (NPQ) 19 . A gradient boost regression (GBR) algorithm was trained on the features of the combined data set (1721 snapshots) and the associated excitonic couplings for Chl-a 409/ Fx-301 as output target. Feature-target relationships were visualized by plotting the predicted versus the actual values for 1721 snapshots of both species 19 as shown in Fig. 2A . This plot indicates a very strong correlation between geometric characteristics of the helices (angles, tilts) and the excitonic couplings, at least for the FCP models in the two species of the MD data set1 19,20 . The trained GBR model was employed to predict “equivalent” coupling values based on the two structural features (avg. angles and avg. tilts) for the whole experimental data set2 in this study (Fig. 2B) as well as the C1-4 conformations. A clear trend is identified with an increase of excitonic couplings in the transition from C4 (7.65 cm -1 ) to C2 (9.48 cm -1 ), to C1 (13.07 cm -1 ) and to C3 (26.41 cm -1 ). The coupling values roughly follow the progression of the FCP scaffold expansion as depicted in Fig. 1B (C1-C4 to C2 and to C3). It is possible that other inter-pigment excitonic couplings within FCPs are also important for the transition of the complex between different light acclimation states. The assignment of excitonic coupling values for other pigment pairs within FCPs, or the calculation of excited state lifetimes is, however, beyond the scope of the current study and impossible at present as the exact NPQ sites have not been identified for all the different FCP variants. Thus, our current model might not be able to unambiguously assign all experimentally resolved FCP structures to an exact acclimation state. However, it is remarkable that a general trend is obvious with our model capturing quantitative general patterns for the FCP conformation, that can be qualitatively correlated with excitonic couplings for a specific pair of pigments. An intrinsic flexibility of the FCP scaffold For further analysis, the MSM-predicted C1-4 conformations mapped onto the sequence of the minimum core of the H dimer in Cy. meneghiniana (aa 64-174) were fed into ProteinMPNN; a deep learning protein design algorithm 26 . The algorithm predicts new protein sequences that fold into the desired 3D protein structure (C1-4). Only the aa 109-131 (sequence (seq): VAAINAIPALGWAQIIFAIGAVD) were allowed to be designed (mutated) so that the end sequence could fold to the associated C1-4 conformation with contracted or expanded protein scaffolds. The resulting sequences of the highest score were folded by AlphaFold 27 as shown in Fig. 3A and referred to as AF1 (seq: PRGVLAFPPALQALLAPVAALLH), AF2 (seq: PARHAILDPWAVLPAVLALLVVL), AF3 (seq: PLALLALTPEQRALLAAILGALL) and AF4 (seq: LAALPYIDPALWERVAAALAELE) for C1, C2, C3 and C4 based predictions, respectively. Despite the failure in reproducing the exact expansion – contraction trend of C1-4 in Fig. 1B , the predictions show clearly contracted and expanded scaffolds with the same variability. RF/ GBR models predict that AF4 belongs to the orange cluster (Fig. 3B) with an excitonic coupling at 16.01 cm -1 , just in-between C1-C2-C4 and C3 (Fig. 1F). “Equivalent” excitonic couplings are also predicted for AF1 at 12.48 cm -1 , AF3 at 9.14 cm -1 , and AF2 at 17.77 cm -1 . Despite the supposed expansion of the AF1 scaffold (shown in blue in Fig. 3A ), expansion is only occurring on the stromal side. In contrast, the lumenal side appears to be contracting. This AF1 conformation is classified within the same cluster as AF3. The ability of ProteinMPNN to design distinct sequences compatible with the different C1-4 conformations suggests that these conformations are not only physically plausible but also intrinsically supported by the FCP protein scaffold. Given that the ProteinMPNN and AlphaFold are trained on a wide range of experimentally validated structures, this result points to an inherent structural plasticity of the FCP fold likely enabling conformational tuning through sequence variation, pH, or oligomerization state. This result highlights the intrinsic conformational versatility of the FCP fold that stably adopt its various experimentally observed conformations. Proof of concept We must note that the NPQ mechanism in diatoms is highly dependent on the xanthophyll cycle and the presence of the LHCX family of photoprotective proteins 28,29 . Our MD models do not consider these effects thus far, yet we could reproduce all different FCP conformations found in the protein data bank. This could imply that the xanthophyll cycle and LHCX proteins fine tune the populations of these FCP conformations in vivo. To prove our case, we have run additional simulations for the FCP from Ph. tricornutum in complex with the photoprotective LHCX1 protein and we also exchanged diadinoxanthin (Ddx) to diatoxanthin (Dtx) at low lumenal pH to take into consideration the xanthophyll cycle 30,31 . The Ddx-Dtx exchange was only considered in the models where FCP interacts with LHCX1 at low pH. For FCP-LHCX1 (Dtx/ Ddx) model setup please refer to the Methods section. In Fig. 4 we show the distribution of FCP conformations for isolated FCP at both low pH and neutral pH (Ddx present) and for FCP-LHCX1 at neutral pH (Ddx present) and at low pH (Dtx present). A comparison of Fig. 4 and Fig. 2B shows that the LHCX1 protein and the xanthophyll cycle are necessary for NPQ induction by shifting the FCP population to expanded FCP scaffolds and increased excitonic coupling values for the Chl-a 409/ Fx-301 pigment pair (cm -1 ) in the diatom Ph. tricornutum, consistent with experimental studies in the literature 6,30–33 . Discussion This study addresses the fundamental biochemical relationship between protein structure and function in photosynthesis. Specifically, it seeks to answer the question: how are photosynthesis and light absorption regulated in terms of the conformations of (LHCs)? Using microsecond-scale all-atom molecular dynamics, Markov state modeling, and machine learning (ML), we demonstrate that all experimentally determined fucoxanthin and chlorophyll a/c-binding protein (FCP) structures, the LHCs in diatoms, correspond to only a few interconverting conformational states. We have identified that the light harvesting antennas in diatoms (FCP scaffold) undergo expansion and contraction in both Molecular Modeling for just two species (Ph. tricornutum and Ch. gracilis) and for all experimentally resolved structures across species. These conformational changes appear to be associated with inter-pigment excitonic couplings, at least within FCPs from Ph. tricornutum and Ch. gracilis and correlate with experimental works on the effect of the photoprotective LHCX1 protein and the xanthophyll cycle. We propose that experimentally resolved structures capture intermediate or transition states of FCP protein dynamics as a fingerprint of the FCP acclimation state. This is the first demonstration that the structural heterogeneity of FCPs reflects an intrinsic, functionally tunable conformational landscape. In the context of this study, we have developed and trained a machine learning (ML) model in a combination of Randrom Forest and Gradient Boosting Regression that can be used to classify FCP structures resolved experimentally into different clusters, or to evaluate those predicted computationally from different diatom species. They can also be used to predict 'equivalent' excitonic coupling values and assign acclimation states for FCP structures. The model can be refined as more MD or crystallographic data become available. It can also be used in terms of workflow to identify key conformations in other protein families. It should be noted that the correlation between structure and function (acclimation state) for FCP structures has been established based on the calculated excitonic coupling of a Chlorophyll-fucoxanthin pigment pair within FCPs from Ph. tricornutum and Ch. gracilis. This approach may be limited in its application to FCP complexes from other diatom species. Nevertheless, there is a strong correlation between the computational and experimental results regarding the effect of the photoprotective LHCX1 protein and the xanthophyll cycle on the acclimation state of the FCP in Ph. tricornutum, which positively validates our approach. Our findings highlight the need for further investigation. Methods Setup of FCP Models The crystal structure of the fucoxanthin chlorophyll a/c -binding (FCP) protein from the diatom Ch. Gracilis (pdb: 7vd5) 13 was used as the initial coordinates to build the monomer models (Sm1 and m2). To construct the S-tetramer, we used the monomer structure of Sm1 and superimposed it with the tetramer chains Sm1, Sm2, Sm3, Sm4 from the Nagao et al. structure 13 . Although chain Sm2 in their structure exhibits slight differences compared to the other three, a previous study 15 refers to the S-tetramer as an homotetramer. All polypeptide chains were described using the Amber ff14SB force field 34 Amber-compatible parameters for fucoxanthin and DD6 were taken from different studies 34–36 , while chlorophyll a and c parameters were adopted from two separate studies 37,38 . The missing phytyl tails of chlorophyll a were modeled using Schrödinger Maestro to restore the complete molecular structure. Protonation states for the Sm1 monomer were assigned as follows: all lumen-facing Asp and Glu residues were protonated to reflect pH 5.5 19 . In contrast, at pH 7.0, the same Asp and Glu residues are treated as deprotonated in correlation with a previous study on Ph. tricornutum 19 . For the m2 model, Glu86 is protonated, and the remaining Asp and Glu residues follow the same protonation pattern as in the Sm1 monomer. His84 was protonated at Nε, while all other His residues at Ν δ sites. In both Sm1 and m2 models, Chl-c 304 was treated protonated at lumenal pH 5.5 and deprotonated at lumenal pH 7, as it faces the lumenal side of the membrane with the acrylate group exposed to the acidic lumen. A lipid bilayer patch of approximately 350 thylakoid lipids 19 , described by the AMBER force field 34 , was used to embed each all-atom model. Lipid composition was based on the thylakoid membrane model of Chryasfoudi et al. 24 containing 45% MGDG, 25% DGDG, 25% SQDG, and 5% PG—reflecting an elevated SQDG/PG content (30%) relative to plant thylakoids (15–20%). 39 The MGDG-DGDG lipid content is at 70% to simulate high-light adapted diatoms compared to low-light grown diatoms (50%) 39 . An amount of around 50000 TIP3P water 40 was used for solvation, and each system included ~150 mM KCl with additional K⁺ ions to neutralize protein and lipid charges. The equilibrated unit cell dimensions of each model were 16.3 × 15.6 × 8.5 nm 3 (monomer) as well as 18.8 × 17.0 × 8.8 nm 3 (tetramer). Setup of FCP-LHCX1 models The crystal structure of the dimer fucoxanthin chlorophyll a/c -binding (FCP) protein from the diatom Ph. tricornutum (PDB ID: 6A2W) served as the starting template for building FCP-LHCX1 dimer model. The LHCX1 protein 3D structure from Ph. Tricornutum (uniport code B7FYL0) was predicted by RosettaFold (robetta.bakerlab.org) using the default parametrization. The FCP-LHCX1 model was generated by structurally aligning the predicted LHCX1 onto the dimeric FCP scaffold resolved experimentally 14 , by ChimeraX software and replacing the aligned FCP monomer by LHCX1. In the following we provide the LHCX1 sequence from the uniport database. >tr|B7FYL0|B7FYL0_PHATC Protein fucoxanthin chlorophyll a/c protein OS=Phaeodactylum tricornutum (strain CCAP 1055/1) OX=556484 GN=Lhcx1 PE=3 SV=1 MKFAATILALIGSAAAFAPAQTSRASTSLQYAKEDLVGAIPPVGFFDPLGFAD KADSPTLKRYREAELTHGRVAMLAVVGFLVGEAVEGSSFLFDASISGPAITHL SQVPAP FWVLLTIAIGASEQTRAVIGWVDPADAPVDKPGLLRDDYVPGDLGF DPLGLKPSDPEELITLQTKELQNGRLAMLAAAGFMAQELVNGKGILENLQG The resulting FCP-LHCX1 complex closely matches the model previously proposed in the literature 24 . All polypeptide chains and pigments were described using the same force fields as in the previous section. The Dtx (diatoxanthin) pigment was parametrized by modifiying the Ddx (diadinoxanthin) parameters specifically by removing one oxygen atom and adapting accordingly. The LHCX1 was considered non-pigmented 41 , Protonation states for the FCP and LHCX1 monomers were assigned as follows: all lumen-facing Asp and Glu residues were protonated to reflect protonation state at pH 5.5. In contrast, at pH 7.0, the same Asp and Glu residues were treated as deprotonated. For the LHCX1 monomer, Glu72, Glu189, and Asp79 were trated as protonated at low pH. All Histidine residues were protonated at Ν δ sites. Classical Molecular Dynamics Following established protocols, all systems were gradually relaxed and equilibrated by progressively releasing positional restraints on the heavy backbone atoms of the protein 19 . During a sequence of simulations in the NVT and NPT ensembles (constant volume and pressure, respectively), the system temperature was gradually raised from 100 K to 303 K before entering the production phase. Classical molecular dynamics (MD) simulations were carried out using the leapfrog integrator available in GROMACS 2021 42 with a 2.0 fs integration time step. The production runs were performed in the constant pressure NPT ensemble with semi-isotropic couplings in the xy membrane plane and in the z-direction (compressibility at 4.5 × 10 –5 bar –1 ). Furthermore, the van der Waals interactions were smoothly shifted to zero between 1.0 and 1.2 nm using the Verlet cutoff scheme. Short-range electrostatics were cut off at 1.2 nm, while long-range electrostatic interactions were computed using the particle mesh Ewald (PME) method 43,44 . All bonds between hydrogen atoms and heavy atoms were constrained using the LINCS algorithm 45 . The v-rescale thermostat was used 46 (303 K, temperature coupling constant 0.5) along with the C-rescale [44] for equilibration, while the Parrinello–Rahman barostats 47 was used for production (1 atm, pressure coupling constant 2.0). Independent trajectories (replicas) were initialized from structures extracted at 10 ns intervals during the final phase of equilibration. Simulation parameters were otherwise consistent with those applied in a prior study of P. tricornutum 19 . The total simulation time for the FCP models amounts to 12 μs, including four independent 0.5 μs trajectories for the tetramer model and four independent 0.5 μs trajectories for each monomeric model (Sm1 and m2), for two distinct pH protonation states (neutral – low). This simulation time can be translated in monomer-equivalent dynamics of: 2 pH states × (4 monomers × 2 μs + 1 monomer × 2 μs) = 20 μs sampling for the Sm1 monomer and 2 pH states × (1 monomer × 2 μs) = 4 μs sampling for the m2 FCP monomer in the different multimeric states. The simulations sum to 24 μs monomer-equilvalent dynamics for Ch. Gracilis. The first 100 ns from each trajectory were considered as further equilibration, and the analysis was only performed for the final 400 ns of each independent trajectory. Structures were collected every 1.0 ns for all the trajectories. The total simulation time for the FCP-LHCX1 models amounted to 8 μs, including: four independent trajectories of 1 μs for the low pH and four independent trajectories of 1 μs for the neutral pH. For details on the simulations for the isolated FCP from Ph. tricornutum (neutral-low pH) and the method of calculation of excitonic couplings between Chl-a 409/ Fx-301 refer to a previous study. 19 Markov State modeling analysis (MSM) Dimers and tetramers were disassembled into different monomeric units before the analysis. The first 100-200ns were disregarded from each individual trajectory based on the time when the backbone root-mean-square-deviation (RMSD) reaches a plateau, resulting cumulatively in 66.3 μs for two diatoms: Ch. Gracilis and Ph. Tricornutum. Only the FCP complex (common minimum core, without protons, ions or pigments) was extracted from the trajectories and all frames (1 ns -1 ) were structurally aligned based on Ca atoms by the GROMACS toolbox (trjconv -fit rot+trans) on a reference common core to assure consistency in the analysis. The PyEMMA package in Jupyter notebooks was employed 48 . All backbone torsional angles of residues aa 109-131 were chosen as input features. The dimensionality of the configurational space sampled in MD was further reduced. This was achieved by the time-lagged independent component analysis (tICA) to remove any redundant information. A 6-component tICA space and a time lag of 50ns was used for coarse graining the degrees of freedom and identify a set of the slowest modes among all the initial input features (6 vectors). Then different MSMs were constructed with their slowest implied timescales to converge quickly and to be constant within a 95% confidence interval for lag times above 50ns. The MSM passed the Chapman–Kolmogorov test at 95% confidence. The validation procedure is a standard approach in the field 23 . A lag time of 50 ns was thus selected for Bayesian MSM model construction. tICA components are the optimal linear combination of input features which maximizes their kinetic variance. The conformations of the FCPs (common core) were projected on the first two tICA vectors (IC-1, IC-2) and the trajectory frames were clustered into 100 cluster-centers (macrostates) by k-means clustering, as implemented in PyEMMA. The resulting macrostates were further coarse grained into a smaller number of four macrostates using PCCA++ as implemented in PyEMMA (conformations C1, C2, C3 and C4). Random Forest and Gradient Boost Machine Learning Models Due to the size of our data set for the experimentally resolved structures (118), the Random Forest approach is chosen for model training and subsequent predictions. The scikit-learn toolbox was employed for the analysis. The experimentally resolved set2 structures were clustered into four clusters by the k-means algorithm. A Random Forest (RF) machine learning model was trained on the experimentally resolved structures along with the four MSM-predicted conformations (C1-4) with avg. angle, avg. tilt as input features and the classification into the k-means clusters as output targets. Prior to training, feature dimensionality was further reduced using Uniform Manifold Approximation and Projection (UMAP) with Euclidean distance, 10 nearest neighbors, and a minimum distance of 0.2 to enhance classification performance. The resulting model was evaluated using a standard train-test split and classification metrics. The RF f1-scores were at 1.00 (red), 0.80 (orange), 0.73 (blue) and 0.92 (green) with an overall accuracy of 0.88. Without the use of the UMAP dimensionality reduction, the overall f1-scores are 1.00 (red), 0.83 (orange), 0.80 (blue) and 0.67 (green) with the overall accuracy at 0.84. The Gradient Boost model was trained on 80% of data (input features/ target excitonic couplings) and evaluated on the remaining 20%, yielding performance metrics including mean squared error (MSE) and the coefficient of determination (R²). Declarations Acknowledgments The work is funded by the Hellenic Foundation for Research & Innovation (H.F.R.I) in the context of the call “Basic Research Financing (Horizontal support for all Sciences), National Recovery and Resilience Plan (Greece 2.0) for the project number 014775, with acronym “SUNDIAL”. This project has received funding from the European Union’s Horizon Europe Research and Innovation Program under the Marie Skłodowska-Curie grant agreement No 101119442. Moreover, this work was also supported by computational time granted from the National Infrastructures for Research and Technology S.A. (GRNET S.A.) in the National HPC facility - ARIS - under project ID „FCPC”. Part of the simulations were performed on a compute cluster funded through the DFG project INST 676/7-1 FUGG. Author contributions: Conceptualization: V.D.; Methodology: V.D., U.K.; Investigation: T.-I.S., B.Z., S.M.; Visualization: V.D., T.-I.S., B.Z.; Supervision: V.D.; Writing-original draft: V.D.; Writing-review & editing: V.D., U.K.; Molecular Dynamics & Machine Learning: I-T.S., B.Z.; Calculation of Excitonic Couplings: S.M. Competing interests : All other authors declare they have no competing interests. Data and materials availability : All data, code details, and material details used in the analyses are available in the main manuscript. Scripts (python codes) for analysis and the datasets used in this study have been deposited in an open access github repository (https://github.com/vdas-upatras/fcp_diatoms). References Croce, R. & van Amerongen, H. Light harvesting in oxygenic photosynthesis: Structural biology meets spectroscopy. Science (1979) 369 , eaay2058 (2020). Giovagnetti, V. & Ruban, A. V. The evolution of the photoprotective antenna proteins in oxygenic photosynthetic eukaryotes. Biochem Soc Trans 46 , 1263–1277 (2018). Pascal, A. A. et al. Molecular basis of photoprotection and control of photosynthetic light-harvesting. Nature 436 , 134–137 (2005). Falciatore, A. & Mock, T. The Molecular Life of Diatoms. Volpe, C. & Büchel, C. Function, Structure and Organization of Light-Harvesting Proteins in Diatoms. in Diatom Photosynthesis 191–215 (2024). doi:https://doi.org/10.1002/9781119842156.ch6. Giossi, C. E., Bitnel, D. B., Wünsch, M. A., Kroth, P. G. & Lepetit, B. Synergistic effects of temperature and light on photoprotection in the model diatom Phaeodactylum tricornutum. Physiol Plant 177 , e70039 (2025). Morosinotto, T., Perin, G. & Gambaro, F. Knowledge of regulation of photosynthesis in outdoor microalgae cultures is essential for the optimization of biomass productivity. Front Plant Sci 751 (2022). Perin, G., Bellan, A., Lyska, D., Niyogi, K. K. & Morosinotto, T. Modulation of xanthophyll cycle impacts biomass productivity in the marine microalga Nannochloropsis bioRxiv 2022.08.16.504082 (2022) doi:10.1101/2022.08.16.504082. Vecchi, V., Barera, S., Bassi, R. & Dall’osto, L. Potential and challenges of improving photosynthesis in algae. Plants vol. 9 Preprint at https://doi.org/10.3390/plants9010067 (2020). Kromdijk, J. et al. Improving photosynthesis and crop productivity by accelerating recovery from photoprotection. Science (1979) 354 , 857–861 (2016). Shi, Y. et al. Harnessing the Power of Photosynthesis: from Current Engineering Strategies to Cell Factory Applications. Small Methods n/a , 2402147 (2025). Feng, Y. et al. Structure of a diatom photosystem II supercomplex containing a member of Lhcx family and dimeric FCPII. Sci Adv 9 , eadi8446 (2023). Nagao, R. et al. Structural basis for different types of hetero-tetrameric light-harvesting complexes in a diatom PSII-FCPII supercomplex. Nat Commun 13 , 1764 (2022). Wang, W. et al. Structural basis for blue-green light harvesting and energy dissipation in diatoms. Science (1979) 363 , eaav0365 (2019). Pi, X. et al. The pigment-protein network of a diatom photosystem II–light-harvesting antenna supercomplex. Science (1979) 365 , eaax4406 (2019). Nagao, R. et al. Structural basis for energy harvesting and dissipation in a diatom PSII–FCPII supercomplex. Nat Plants 5 , 890–901 (2019). Nagao, R. et al. Comparison of oligomeric states and polypeptide compositions of fucoxanthin chlorophyll a/c-binding protein complexes among various diatom species. Photosynth Res 117 , 281–288 (2013). Büchel, C. Light harvesting complexes in chlorophyll c-containing algae. Biochimica et Biophysica Acta (BBA) - Bioenergetics 1861 , 148027 (2020). Daskalakis, V., Maity, S. & Kleinekathöfer, U. An Unexpected Water Channel in the Light-Harvesting Complex of a Diatom: Implications for the Switch between Light Harvesting and Photoprotection. ACS Physical Chemistry Au 5 , 47–61 (2025). Maity, S., Daskalakis, V., Jansen, T. L. C. & Kleinekathöfer, U. Electric Field Susceptibility of Chlorophyll c Leads to Unexpected Excitation Dynamics in the Major Light-Harvesting Complex of Diatoms. J Phys Chem Lett 15 , 2499–2510 (2024). Konagurthu, A. S., Whisstock, J. C., Stuckey, P. J. & Lesk, A. M. MUSTANG: A multiple structural alignment algorithm. Proteins: Structure, Function, and Bioinformatics 64 , 559–574 (2006). Zhao, S. et al. Structural insights into photosystem II supercomplex and trimeric FCP antennae of a centric diatom Cyclotella meneghiniana. Nat Commun 14 , 8164 (2023). Husic, B. E. & Pande, V. S. Markov State Models: From an Art to a Science. J Am Chem Soc 140 , 2386–2396 (2018). Chrysafoudi, A., Maity, S., Kleinekathöfer, U. & Daskalakis, V. Robust Strategy for Photoprotection in the Light-Harvesting Antenna of Diatoms: A Molecular Dynamics Study. J Phys Chem Lett 12 , 9626–9633 (2021). Agostini, A., Bína, D., Carbonera, D. & Litvín, R. Conservation of triplet-triplet energy transfer photoprotective pathways in fucoxanthin chlorophyll-binding proteins across algal lineages. Biochimica et Biophysica Acta (BBA) - Bioenergetics 1864 , 148935 (2023). Dauparas, J. et al. Robust deep learning–based protein sequence design using ProteinMPNN. Science (1979) 378 , 49–56 (2022). Akdel, M. et al. A structural biology community assessment of AlphaFold2 applications. Nat Struct Mol Biol 29 , 1056–1067 (2022). Falciatore, A., Jaubert, M., Bouly, J.-P., Bailleul, B. & Mock, T. Diatom Molecular Research Comes of Age: Model Species for Studying Phytoplankton Biology and Diversity[OPEN]. Plant Cell 32 , 547–572 (2020). Croteau, D., Jaubert, M., Falciatore, A. & Bailleul, B. Pennate diatoms make non-photochemical quenching as simple as possible but not simpler. Nat Commun 16 , 2385 (2025). Lepetit, B. et al. The diatom Phaeodactylum tricornutum adjusts nonphotochemical fluorescence quenching capacity in response to dynamic light via fine-tuned Lhcx and xanthophyll cycle pigment synthesis. New Phytologist 214 , 205–218 (2017). Kuczynska, P. et al. The xanthophyll cycle in diatom Phaeodactylum tricornutum in response to light stress. Plant Physiology and Biochemistry 152 , 125–137 (2020). Taddei, L. et al. Dynamic changes between two LHCX-related energy quenching sites control diatom photoacclimation. Plant Physiol 177 , 953–965 (2018). Buck, J. M. et al. Lhcx proteins provide photoprotection via thermal dissipation of absorbed light in the diatom Phaeodactylum tricornutum. Nat Commun 10 , 4167 (2019). Wang, J., Wolf, R. M., Caldwell, J. W., Kollman, P. A. & Case, D. A. Development and testing of a general Amber force field. J Comput Chem 25 , 1157–1174 (2004). Wang, Y., Mao, L. & Hu, X. Insight into the structural role of carotenoids in the Photosystem I: a quantum chemical analysis. Biophys J 86 , 3097–3111 (2004). Prandi, I. G., Viani, L., Andreussi, O. & Mennucci, B. Combining classical molecular dynamics and quantum mechanical methods for the description of electronic excitations: The case of carotenoids. J Comput Chem 37 , 981–991 (2016). Ceccarelli, M., Procacci, P. & Marchi, M. An ab initio force field for the cofactors of bacterial photosynthesis. J Comput Chem 24 , 129–142 (2003). Zhang, L., Silva, D. A., Yan, Y. & Huang, X. Force field development for cofactors in the photosystem II. J Comput Chem 33 , 1969–1980 (2012). Lepetit, B., Goss, R., Jakob, T. & Wilhelm, C. Molecular dynamics of the diatom thylakoid membrane under different light conditions. Photosynth Res 111 , 245–257 (2012). Mark, P. & Nilsson, L. Structure and dynamics of the TIP3P, SPC, and SPC/E water models at 298 K. J Phys Chem A 105 , 9954–9960 (2001). Giovagnetti, V. et al. Biochemical and molecular properties of LHCX1, the essential regulator of dynamic photoprotection in diatoms. Plant Physiol 188 , 509–525 (2022). Berendsen, H. J. C., van der Spoel, D. & van Drunen, R. GROMACS: A message-passing parallel molecular dynamics implementation. Comput Phys Commun 91 , 43–56 (1995). Yeh, I.-C. & Berkowitz, M. L. Ewald summation for systems with slab geometry. J Chem Phys 111 , 3155–3162 (1999). Darden, T., York, D. & Pedersen, L. Particle mesh Ewald: An N⋅ log (N) method for Ewald sums in large systems. J Chem Phys 98 , 10089–10092 (1993). Hess, B., Bekker, H., Berendsen, H. J. C. & Fraaije, J. G. E. M. LINCS: a linear constraint solver for molecular simulations. J Comput Chem 18 , 1463–1472 (1997). Bussi, G., Donadio, D. & Parrinello, M. Canonical sampling through velocity rescaling. J Chem Phys 126 , 14101 (2007). Parrinello, M. & Rahman, A. Polymorphic transitions in single crystals: A new molecular dynamics method. J Appl Phys 52 , 7182–7190 (1981). Scherer, M. K. et al. PyEMMA 2: A Software Package for Estimation, Validation, and Analysis of Markov Models. J Chem Theory Comput 11 , 5525–5542 (2015). Additional Declarations There is NO Competing Interest. Cite Share Download PDF Status: Published Journal Publication published 05 Dec, 2025 Read the published version in Communications Chemistry → Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7326805","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":498619438,"identity":"ac95319e-b176-4cbb-802e-01cf0f29005c","order_by":0,"name":"Theofani-Iosifina Sousani*","email":"","orcid":"","institution":"Department of Chemical Engineering, University of Patras, Patras 265 04, Greece","correspondingAuthor":false,"prefix":"","firstName":"Theofani-Iosifina","middleName":"","lastName":"Sousani*","suffix":""},{"id":498619439,"identity":"80e02073-15df-48c2-873b-757900ee9d9e","order_by":1,"name":"Boutheina Zender*","email":"","orcid":"","institution":"Department of Chemical Engineering, University of Patras, Patras 265 04, Greece","correspondingAuthor":false,"prefix":"","firstName":"Boutheina","middleName":"","lastName":"Zender*","suffix":""},{"id":498619440,"identity":"73124ac6-71cf-48e6-96ed-9fa382c05aec","order_by":2,"name":"Sayan Maity","email":"","orcid":"","institution":"School of Science, Constructor University, 28759 Bremen, Germany","correspondingAuthor":false,"prefix":"","firstName":"Sayan","middleName":"","lastName":"Maity","suffix":""},{"id":498619441,"identity":"e52a7389-de9f-4e34-8ae2-dbc5dbb94df4","order_by":3,"name":"Ulrich Kleinekathöfer","email":"","orcid":"","institution":"School of Science, Constructor University, 28759 Bremen, Germany","correspondingAuthor":false,"prefix":"","firstName":"Ulrich","middleName":"","lastName":"Kleinekathöfer","suffix":""},{"id":498619437,"identity":"c41a3bc6-3ae0-4cc7-be77-fc69efbb1f37","order_by":4,"name":"Vangelis Daskalakis","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABAklEQVRIie3RP0vDQBzG8d9xkCx3ZH0OCr6FEyE6+OetRDJkySa4VFQQ0kWcU4p9DYpLBwchYJaAL6DLBSEuFRpE6SRWB8HBI90E77s9w2d6iFyuP1lE7JS2CJyYWYXgk3C9nFiBEHnoRHQZP7b5EdbUwG/6crJ7vDkoDXu9tZCqCdXVPdZHXIRTWcXoVanmveZ3ovIoZMYDG3PhTWXGAUqJ485GkpfWvGNvzP3mQGYnQPBkrCRAqtV1hv0Rp5DLrAAQaTa3ETE7VMMLxMMzsaEuq1LlmOmCLMTzk5v2/G17J38o6/nzpB8gSOp6YSE/Wz70VSG6im/CFp2Jy+Vy/YM+AO1RTf1ClebwAAAAAElFTkSuQmCC","orcid":"https://orcid.org/0000-0001-8870-0850","institution":"Department of Chemical Engineering, University of Patras, Patras 265 04, Greece","correspondingAuthor":true,"prefix":"","firstName":"Vangelis","middleName":"","lastName":"Daskalakis","suffix":""}],"badges":[],"createdAt":"2025-08-08 11:30:42","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7326805/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7326805/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s42004-025-01774-x","type":"published","date":"2025-12-05T05:00:00+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":88872772,"identity":"fcab068e-273e-4f94-9f87-63bc3937a591","added_by":"auto","created_at":"2025-08-12 09:31:52","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":3171556,"visible":true,"origin":"","legend":"\u003cp\u003eThe configurational space of the diatom light harvesting complexes. (\u003cstrong\u003eA\u003c/strong\u003e) Free Energy of the protein scaffold spanned over the tICA components IC-1 and IC-2. The four main energy states (1 to 4) are highlighted with colored labels denoting their position. The energy values are given in units of kT with k being the Boltzmann constant and T the temperature. (\u003cstrong\u003eB\u003c/strong\u003e) The FCP monomer core in the different C1−4 conformations associated with states 1-4 color coded as in panel A. The avg. angle \u003cimg width=\"9\" height=\"30\" src=\"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAkAAAAeCAMAAADJnMQBAAAAAXNSR0IArs4c6QAAAEtQTFRFAAAAAAAAAAA6AABmADpmADqQAGa2OgAAOgA6OpDbZgAAZrb/kDoAkLbbkNv/tmYAttv/tv//25A625Bm2////7Zm/9uQ//+2///b1MRjywAAAAF0Uk5TAEDm2GYAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAAZdEVYdFNvZnR3YXJlAE1pY3Jvc29mdCBPZmZpY2V/7TVxAAAAS0lEQVQoU2NgoCtghAGCtoqyM7LwgVSJcfAyCDELAln8bAwM4lzcDAwSPJzIXKgE2GiYErAYVB2qYlEmoHmsIiCjhYF2CBB0CVwBAArmAr+caeX1AAAAAElFTkSuQmCC\"/\u003e\u0026nbsp;is defined as the average of the angles θ\u003csub\u003e1\u003c/sub\u003e, θ\u003csub\u003e2\u003c/sub\u003e and θ\u003csub\u003e3\u003c/sub\u003e defined by the vectors: v\u003csub\u003e1\u003c/sub\u003e aa 64-85, v\u003csub\u003e2\u003c/sub\u003e 109-116, v\u003csub\u003e3\u003c/sub\u003e 117-131 and v\u003csub\u003e4\u003c/sub\u003e for aa 149-174. The avg. tilt is calculated as the average of the angles formed by each vector v\u003csub\u003e1\u003c/sub\u003e to v\u003csub\u003e4\u003c/sub\u003e with respect to the membrane normal Z. The residue numbering refers to the sequence of the H (H1/ H2) dimer of Cy. meneghiniana. (\u003cstrong\u003eC\u003c/strong\u003e) Energy distributions of the FCP scaffold conformations spanned over the avg. angle and tilt dimensions for the MD data set1 and (\u003cstrong\u003eD\u003c/strong\u003e) the experimentally resolved structural data set2. (\u003cstrong\u003eE\u003c/strong\u003e) The distribution of experimentally resolved structures per diatom species. (\u003cstrong\u003eF\u003c/strong\u003e) Clustering of the experimentally resolved structural data set2 by the k-means algorithm into four color-coded clusters. In panels C to F, the position of C1-4 conformations is given for reference.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7326805/v1/6c002ec8fdd2fa9ec7c2e81a.png"},{"id":88872777,"identity":"314539ae-b18f-4116-a9c7-a256e2b6294f","added_by":"auto","created_at":"2025-08-12 09:31:52","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":663828,"visible":true,"origin":"","legend":"\u003cp\u003eCorrelation between structure and function. (\u003cstrong\u003eA\u003c/strong\u003e) Evaluation of the regression by gradient boosting for the calculated excitonic couplings for the Chl-a 409/ Fx-301 pigment pair (cm-1) in the MD data set1. (\u003cstrong\u003eB\u003c/strong\u003e) Predicted distribution of excitonic couplings spanned over the avg. angle and avg. tilt dimensions for the experimentally resolved structural data set2.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7326805/v1/dbca789b00ebea0096909c0a.png"},{"id":88873630,"identity":"8bba030d-e895-4dd1-82c1-d26a9be3d25a","added_by":"auto","created_at":"2025-08-12 09:39:52","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":213604,"visible":true,"origin":"","legend":"\u003cp\u003eSequence dependent configuration. (\u003cstrong\u003eA\u003c/strong\u003e) Equivalent to C1-4 conformations, the AF1-4 conformations predicted by ProteinMPNN and AlphaFold, are shown along with their position (\u003cstrong\u003eB\u003c/strong\u003e) on the two dimensions (avg. angle, avg. tilt). The coloring is identical to the one used in \u003cstrong\u003eFig. 1B\u003c/strong\u003e.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7326805/v1/e92772c873a1b1611c7d6e0b.png"},{"id":88874037,"identity":"f43aab5d-9b7e-466b-bf34-149d8762547e","added_by":"auto","created_at":"2025-08-12 09:47:52","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":708709,"visible":true,"origin":"","legend":"\u003cp\u003eThe effect of LHCX1 and the xanthophyll cycle. (A) The structure of the FCP-LHCX1 complex. Key pigments are shown for reference (diadinoxanthin (Ddx), diatoxanthin (Dtx), Chlorophyll (Chl) 409 and Fucoxanthin (Fx) 301). (B) Energy distributions of the FCP scaffold conformations spanned over the avg. angle and tilt dimensions for models of data set1 and FCP-LHCX1 (Dtx - diatoxanthin) at low pH and FCP-LHCX1 (Ddx - diadinoxanthin) at neutral pH complexes.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7326805/v1/679398b68dd93640d65ef146.png"},{"id":97577113,"identity":"eae0bc2c-70d7-4a06-86b8-78e8ee98ead2","added_by":"auto","created_at":"2025-12-06 08:06:17","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":5040811,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7326805/v1/d1f9d63e-44f7-4a02-964a-3cc397fbf633.pdf"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Conformational Plasticity Enables Functional Switching in Diatom Light-Harvesting Complexes","fulltext":[{"header":"Introduction","content":"\u003cp\u003ePhotosynthesis is a vital natural process through which solar energy is captured and transformed into chemical energy, defining the basis of life on Earth \u003csup\u003e1\u003c/sup\u003e. Organisms capable of photosynthesis are generally grouped into green and red evolutionary lineages \u003csup\u003e2\u003c/sup\u003e. Despite their diversity, green plants and algae share common structural features in their photosynthetic systems. Light energy, in the form of photons, is initially absorbed by pigments like chlorophylls and carotenoids found in light-harvesting complexes (LHCs) \u0026ndash; protein-pigment assemblies located within the thylakoid membrane of chloroplasts \u003csup\u003e1\u003c/sup\u003e Varying environmental factors such as changes in light intensity or spectrum, often influenced by daily light cycles, can affect this process. To manage these fluctuations and avoid damage from excess light, green plants and algae use (down) regulatory strategies that adjust light absorption. One key protective response is non-photochemical quenching (NPQ), a mechanism by which surplus excitation energy is safely dissipated as heat to prevent photodamage \u003csup\u003e3\u003c/sup\u003e. Diatoms in the red lineage of photosynthetic organisms are highly important as primary producers in marine ecosystems, supplying organic substrates and oxygen, by rapidly adjusting light harvesting and photochemistry to cell productivity. Therefore, they are considered an ideal model organism for studying photosynthetic regulation \u003csup\u003e4\u003c/sup\u003e. Diatoms express fucoxanthin (Fx) and chlorophyll-a/c (Chl) binding proteins (FCPs) as LHCs. FCPs are responsible for the exceptional light-harvesting capabilities of diatoms in the blue-green region, which is available underwater. However, they also have a robust intrinsic photoprotective mechanism that enables them to adapt to fluctuating light at the ocean surface \u003csup\u003e5,6\u003c/sup\u003e. The details of this mechanism at all-atom resolution are still obscure \u003csup\u003e6\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eRegulation of photosynthesis is a promising area of research for improving crops and increasing biomass or boosting the atmospheric CO\u003csub\u003e2\u003c/sub\u003e assimilation to ease the strangle against global warming \u003csup\u003e7\u0026ndash;10\u003c/sup\u003e. The fundamental research on the downregulatory mechanisms of photosynthesis\u003csup\u003e\u0026nbsp;\u003c/sup\u003ecould lay the groundwork for artificial photosynthesis \u003csup\u003e11\u003c/sup\u003e. and could also provide crucial knowledge towards the increased production of biofuels and nutrients. FCPs come in monomers or various oligomeric states such as dimers, trimers, tetramers, or even pentamers \u003csup\u003e5,12\u0026ndash;18\u003c/sup\u003e. The plethora of structural insights into FCPs calls for a comprehensive investigation to answer the following question: What is the connection between FCP structure and function for a remarkable adaptability? To answer this question, we implemented a comprehensive analytical framework which integrates extensive conformational sampling at the classical Molecular Dynamics (MD) level, exciton coupling calculations between pigments, statistical tools like Markov State Modeling (MSM), data from experimentally resolved structures and machine learning (ML). This integrated computational approach allows for a quantitative assessment of how FCP conformation changes in different acclimation states.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003e\u003cem\u003eThe configurational space of light harvesting antennas\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe have complemented previous MD simulations on the FCP complex from Ph. Tricornutum \u003csup\u003e14,19\u003c/sup\u003e with dynamics of FCP complexes from Ch. Gracilis (Sm1 complex in monomeric and tetrameric forms and m2 complex in monomeric forms) \u003csup\u003e13\u003c/sup\u003e. The setup and run of the MD simulations of Ch. Gracilis followed exactly the protocols in our previous studies of Ph. Tricornutum \u003csup\u003e19,20\u003c/sup\u003e. All FCPs were sampled at two different pH states (neutral and acidic (pH ~5.5) lumenal environments) \u003csup\u003e19,20\u003c/sup\u003e and different multimerization states resulting in a cumulative simulation time of 80\u0026mu;s monomer-equivalent dynamics for the FCPs from the two species. For details refer to the Methods section. Consistency was ensured in the analysis by truncating all trajectories to a common core of 111 residues (444 atoms in the mainchain \u0026ndash; backbone) that is shared among both diatoms, identified by a multi-sequence alignment method (MUSTANG) \u003csup\u003e21\u003c/sup\u003e. Only residues aa 30-51, 54-116, 132-157 were considered for the FCP from Ph. Tricornutum \u003csup\u003e14\u003c/sup\u003e. For Ch. Gracilis residues aa 58-79, 83-145, 155-180 were considered for the Sm1 in monomeric and tetrameric forms and aa 58-79, 83-145, 158-183 for the m2 monomer \u003csup\u003e13\u003c/sup\u003e. Interestingly, this minimum core of 111 residues matches a continues sequence in the dimeric FCP fold (H1/ H2) found in the diatom Cy. meneghiniana (aa 64-174) \u003csup\u003e22\u003c/sup\u003e. All residue numbers reported hereafter refer to the numbering in the sequence of H1/ H2.\u003c/p\u003e\n\u003cp\u003eCryo-EM (cryogenic electron microscopy) and X-ray crystallographic data were also collected from the protein data bank (pdb). Structures from diverse FCPs in various oligomeric states are available from different diatom species (Ph. tricornutum: 6A2W; Ch. gracilis: 6J3Y, 6J3Z, 6J40, 6JLU, 6L4U, 7VD5, 7VD6, 8WCK, 8WCL; Th. pseudonana: 8IWH, 8J0D, 8JP3; Cy. meneghiniana: 8J5K, 8J7Z, 8W4O, 8W4P; Ch. roscoffensis: 9KQB). Dimers, tetramers, or trimers were disassembled into 118 different monomeric units before the analysis. Thus, two primary datasets will be analyzed in the following, i.e., set1: MD frames and set2: experimental structures.\u003c/p\u003e\n\u003cp\u003eA Markov state model (MSM) was constructed from the MD trajectories, which can predict the long-time scale behavior of a biomolecular system \u003csup\u003e23\u003c/sup\u003e. For further details refer to the Methods. The reweighted free energy surface (FES) was projected onto the slowest degrees of freedom of the FCP common core, determined by the time-lagged independent component analysis (tICA) procedure and described by the two vectors IC-1 and IC-2. This is shown in \u003cstrong\u003eFig. 1A\u003c/strong\u003e, alongside the positions of the four identified kinetically distinct states 1-4. State-2 is shared between the FCPs of Ph. tricornutum and Ch. gracilis, states-1 and- 3 are assigned to Ph. tricornutum whereas state-4 to Ch. gracilis. Four ensemble averaged conformations (macrostates) that are associated with states 1-4 were also predicted by MSM (C1 to C4). Superimposed conformations of the common core for C1-4 are shown in \u003cstrong\u003eFig. 1B\u003c/strong\u003e. Therein, the FCP scaffold clearly shows a progressive expansion from conformation C1 and C4 to C2 and finally to C3. The mechanism and functional significance of this expansion are not readily available. However, the transitions should be related to pH differences at the lumen FCP side or oligomerization states, given that the MD trajectories sampled FCP complexes at different pH values and oligomeric states (see Methods for FCP model setup). Given also that the combination of MD-MSM provides insight into kinetically distinct conformations of a biomolecular system over the long-time scale \u003csup\u003e23\u003c/sup\u003e, C1-4 should indicate key conformational transitions of the FCP scaffold not explicitly sampled by classical MD.\u003c/p\u003e\n\u003cp\u003eThe tICA components or vectors cannot be easily understood in terms of physical parameters that can describe conformational changes of the FCP protein scaffold. We thus have employed an alternative to the tICA dimensionality reduction approach. The configurational space of the MD set1 was reduced into two dimensions: (a) the average (avg.) angle \u003cimg width=\"9\" height=\"22\" src=\"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAkAAAAWCAMAAAAlz0ZsAAAAAXNSR0IArs4c6QAAAEtQTFRFAAAAAAAAAAA6AABmADpmADqQAGa2OgAAOgA6OpDbZgAAZrb/kDoAkLbbkNv/tmYAttv/tv//25A625Bm2////7Zm/9uQ//+2///b1MRjywAAAAF0Uk5TAEDm2GYAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAAZdEVYdFNvZnR3YXJlAE1pY3Jvc29mdCBPZmZpY2V/7TVxAAAATUlEQVQYV2NgwAsYYQC/MqCsKDsjCx9IlRgHL4MQsyCQxc/GwCDOxc3AIMHDicyFSoCNhikBi0HVoSoWZQKaxyoCMloYaIcAQZfAFQAACpYCv1c02MwAAAAASUVORK5CYII=\" alt=\"image\"\u003e between helices in the common FCP core and (b) the average of the tilts of the same helices with respect to the Z axis defined as the normal on the thylakoid membrane (see \u003cstrong\u003eFig. 1B\u003c/strong\u003e for definitions). In detail, helical segments were identified sequentially by the DSSP method (Define Secondary Structure of Proteins) implemented within the biopython toolbox (Bio.PDB.DSSP) and based on \u0026alpha;-helical secondary structure annotations. For each helix, a principal component analysis (PCA) was applied to the C\u0026alpha; atom coordinates employing the scikit-learn toolbox to determine the dominant helical axis. Subsequently, the inter-helical angles and tilt angles with respect to the z-axis were computed. The probability distribution P(x,y) of the MD data set1 over the two reduced dimensions is converted to -ln(P(x,y)) in energy units [kT] (Fig. 1C). To elucidate the relationship between the predicted C1-4 conformations and the experimentally FCP resolved structures across species, we repeated the procedure for data set2 and the results are shown in \u003cstrong\u003eFig. 1D\u003c/strong\u003e. The position of the MSM-predicted conformations or macrostates C1-4 are shown for reference on both plots (Fig. 1C-D). Remarkably, the MSM predictions quantitatively match the minima in the distribution of the experimentally resolved FCP structures from the protein databank (set2). The C1-4 conformations do not correspond to minima sampled by MD (Fig. 1C), as they are long timescale \u0026quot;projections\u0026quot; based on the MD-MSM combination that, however, remarkably fit the experimental data, even for different species of diatoms. In \u003cstrong\u003eFig. 1E\u003c/strong\u003e we show the distribution of each experimentally resolved structure colored by species over the avg. angle and avg. tilt dimensions. As can be observed, there is no dependency of the C1 to C4 conformations on the diatom species. We have to note that the distribution of \u003cstrong\u003eFig. 1E\u003c/strong\u003e is also independent of the resolution of the experimentally resolved structures.\u003c/p\u003e\n\u003cp\u003eThe scikit-learn toolbox was employed for the subsequent analysis and machine learning model training. For details refer to the Methods section. The experimentally resolved set2 structures (Fig. 1E) were classified into four clusters identified by color (Fig. 1F) and based on their structural diversity. C1, C2 belong to the same cluster (blue), whereas C4 to the red cluster and C3 to the green cluster. This unsupervised classification was performed by a simple k-means clustering over the avg. angle and avg. tilt features. The classification is not always obvious by just placing an unknown structure alongside the data of \u003cstrong\u003eFig. 1F\u003c/strong\u003e. However, a more robust Random Forest (RF) supervised machine learning model was trained on the experimentally resolved structures along with the four MSM-predicted conformations (C1-4) and verified the original classification (Fig. 1F). The C1 to C4 MSM-predicted conformations, are thus classified into three main clusters (red, blue and green). This latter classification was performed at a higher level of accuracy, employing further dimensionality reduction by Uniform Manifold Approximation and Projection (UMAP), compared to the simpler k-means on the un-reduced data (see Methods for details). The trained RF ML model can thus be employed to characterize accurately any future FCP structure providing also percentages of \u0026ldquo;blue\u0026rdquo;, \u0026ldquo;red\u0026rdquo;, \u0026ldquo;orange\u0026rdquo; or \u0026ldquo;green\u0026rdquo; character.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eStructure and function correlation\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCould the distinct C1-4 conformations be associated with a robust transition of the FCP scaffold between an efficient light-harvesting and a downregulated photosynthetic state? In order to answer this question, we focus on a special pigment pair in the FCP structure of Ph. tricornutum (Chl-a 409 and carotenoid Fx-301) that has been proposed computationally and experimentally to be involved in such a transition \u003csup\u003e19,24,25\u003c/sup\u003e. Under NPQ conditions, energy transfer from Chl-a to Fx enables a quick dissipation of the excess absorbed energy as heat, through a short-lived excited state of Fx. The excitonic coupling value between Chl-a (m) and Fx (n) is a measure of the efficiency of this energy transfer, with the rate of transfer given by \u003cimg width=\"92\" height=\"19\" src=\"data:image/png;base64,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\" alt=\"image\"\u003e \u003csup\u003e19\u003c/sup\u003e, where \u003cimg width=\"27\" height=\"19\" src=\"data:image/png;base64,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\" alt=\"image\"\u003e\u0026nbsp;is the excitonic coupling between m and n. The \u003cimg width=\"27\" height=\"19\" src=\"data:image/png;base64,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\" alt=\"image\"\u003e values for the pigments of this pair have been calculated in an earlier study on the FCP from Ph. tricornutum from a 860 snapshot trajectory \u003csup\u003e19\u003c/sup\u003e and herein for Ch. gracilis Sm1 monomer (861 snapshots randomly extracted). Low \u003cimg width=\"27\" height=\"19\" src=\"data:image/png;base64,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\" alt=\"image\"\u003e values (\u0026lt;10 cm\u003csup\u003e-1\u003c/sup\u003e) indicate a light-harvesting state and higher values (\u0026gt;10 cm\u003csup\u003e-1\u003c/sup\u003e) are characteristic for photoprotective states (NPQ) \u003csup\u003e19\u003c/sup\u003e. A gradient boost regression (GBR) algorithm was trained on the features \u0026nbsp;of the combined data set (1721 snapshots) and the associated excitonic couplings for Chl-a 409/ Fx-301 as output target. \u0026nbsp;Feature-target relationships were visualized by plotting the predicted versus the actual values for 1721 snapshots of both species\u003csup\u003e19\u003c/sup\u003e as shown in \u003cstrong\u003eFig. 2A\u003c/strong\u003e. This plot indicates a very strong correlation between geometric characteristics of the helices (angles, tilts) and the excitonic couplings, at least for the FCP models in the two species of the MD data set1\u003csup\u003e19,20\u003c/sup\u003e. The trained GBR model was employed to predict \u0026ldquo;equivalent\u0026rdquo; coupling values based on the two structural features (avg. angles and avg. tilts) for the whole experimental data set2 in this study (Fig. 2B) as well as the C1-4 conformations. A clear trend is identified with an increase of excitonic couplings in the transition from C4 (7.65 cm\u003csup\u003e-1\u003c/sup\u003e) to C2 (9.48 cm\u003csup\u003e-1\u003c/sup\u003e), to C1 (13.07 cm\u003csup\u003e-1\u003c/sup\u003e) and to C3 (26.41 cm\u003csup\u003e-1\u003c/sup\u003e). The coupling values roughly follow the progression of the FCP scaffold expansion as depicted in \u003cstrong\u003eFig. 1B\u003c/strong\u003e (C1-C4 to C2 and to C3). It is possible that other inter-pigment excitonic couplings within FCPs are also important for the transition of the complex between different light acclimation states. The assignment of excitonic coupling values for other pigment pairs within FCPs, or the calculation of excited state lifetimes is, however, beyond the scope of the current study and impossible at present as the exact NPQ sites have not been identified for all the different FCP variants. Thus, our current model might not be able to unambiguously assign all experimentally resolved FCP structures to an exact acclimation state. However, it is remarkable that a general trend is obvious with our model capturing quantitative general patterns for the FCP conformation, that can be qualitatively correlated with excitonic couplings for a specific pair of pigments.\u003cstrong\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eAn intrinsic flexibility of the FCP scaffold\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFor further analysis, the MSM-predicted C1-4 conformations mapped onto the sequence of the minimum core of the H dimer in Cy. meneghiniana (aa 64-174) were fed into ProteinMPNN; a deep learning protein design algorithm \u003csup\u003e26\u003c/sup\u003e. The algorithm predicts new protein sequences that fold into the desired 3D protein structure (C1-4). Only the aa 109-131 (sequence (seq): VAAINAIPALGWAQIIFAIGAVD) were allowed to be designed (mutated) so that the end sequence could fold to the associated C1-4 conformation with contracted or expanded protein scaffolds. The resulting sequences of the highest score were folded by AlphaFold \u003csup\u003e27\u003c/sup\u003e as shown in \u003cstrong\u003eFig. 3A\u003c/strong\u003e and referred to as AF1 (seq: PRGVLAFPPALQALLAPVAALLH), AF2 (seq: PARHAILDPWAVLPAVLALLVVL), AF3 (seq: PLALLALTPEQRALLAAILGALL) and AF4 (seq: LAALPYIDPALWERVAAALAELE) for C1, C2, C3 and C4 based predictions, respectively. Despite the failure in reproducing the exact expansion \u0026ndash; contraction trend of C1-4 in \u003cstrong\u003eFig. 1B\u003c/strong\u003e, the predictions show clearly contracted and expanded scaffolds with the same variability. RF/ GBR models predict that AF4 belongs to the orange cluster (Fig. 3B) with an excitonic coupling at 16.01 cm\u003csup\u003e-1\u003c/sup\u003e, just in-between C1-C2-C4 and C3 (Fig. 1F). \u0026ldquo;Equivalent\u0026rdquo; excitonic couplings are also predicted for AF1 at 12.48 cm\u003csup\u003e-1\u003c/sup\u003e, AF3 at 9.14 cm\u003csup\u003e-1\u003c/sup\u003e, and AF2 at 17.77 cm\u003csup\u003e-1\u003c/sup\u003e. Despite the supposed expansion of the AF1 scaffold (shown in blue in \u003cstrong\u003eFig. 3A\u003c/strong\u003e), expansion is only occurring on the stromal side. In contrast, the lumenal side appears to be contracting. This AF1 conformation is classified within the same cluster as AF3. The ability of ProteinMPNN to design distinct sequences compatible with the different C1-4 conformations suggests that these conformations are not only physically plausible but also intrinsically supported by the FCP protein scaffold. Given that the ProteinMPNN and AlphaFold are trained on a wide range of experimentally validated structures, this result points to an inherent structural plasticity of the FCP fold likely enabling conformational tuning through sequence variation, pH, or oligomerization state. This result highlights the intrinsic conformational versatility of the FCP fold that stably adopt its various experimentally observed conformations.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eProof of concept\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe must note that the NPQ mechanism in diatoms is highly dependent on the xanthophyll cycle and the presence of the LHCX family of photoprotective proteins \u003csup\u003e28,29\u003c/sup\u003e. Our MD models do not consider these effects thus far, yet we could reproduce all different FCP conformations found in the protein data bank. This could imply that the xanthophyll cycle and LHCX proteins fine tune the populations of these FCP conformations in vivo. To prove our case, we have run additional simulations for the FCP from Ph. tricornutum in complex with the photoprotective LHCX1 protein and we also exchanged diadinoxanthin (Ddx) to diatoxanthin (Dtx) at low lumenal pH to take into consideration the xanthophyll cycle \u003csup\u003e30,31\u003c/sup\u003e. The Ddx-Dtx exchange was only considered in the models where FCP interacts with LHCX1 at low pH. For FCP-LHCX1 (Dtx/ Ddx) model setup please refer to the Methods section. In \u003cstrong\u003eFig. 4\u003c/strong\u003e we show the distribution of FCP conformations for isolated FCP at both low pH and neutral pH (Ddx present) and for FCP-LHCX1 at neutral pH (Ddx present) and at low pH (Dtx present). A comparison of \u003cstrong\u003eFig. 4\u003c/strong\u003e and \u003cstrong\u003eFig. 2B\u003c/strong\u003e shows that the LHCX1 protein and the xanthophyll cycle are necessary for NPQ induction by shifting the FCP population to expanded FCP scaffolds and increased excitonic coupling values for the Chl-a 409/ Fx-301 pigment pair (cm\u003csup\u003e-1\u003c/sup\u003e) in the diatom Ph. tricornutum, consistent with experimental studies in the literature \u003csup\u003e6,30\u0026ndash;33\u003c/sup\u003e.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study addresses the fundamental biochemical relationship between protein structure and function in photosynthesis. Specifically, it seeks to answer the question: how are photosynthesis and light absorption regulated in terms of the conformations of (LHCs)? Using microsecond-scale all-atom molecular dynamics, Markov state modeling, and machine learning (ML), we demonstrate that all experimentally determined fucoxanthin and chlorophyll a/c-binding protein (FCP) structures, the LHCs in diatoms, correspond to only a few interconverting conformational states. We have identified that the light harvesting antennas in diatoms (FCP scaffold) undergo expansion and contraction in both Molecular Modeling for just two species (Ph. tricornutum and Ch. gracilis) and for all experimentally resolved structures across species. These conformational changes appear to be associated with inter-pigment excitonic couplings, at least within FCPs from Ph. tricornutum and Ch. gracilis and correlate with experimental works on the effect of the photoprotective LHCX1 protein and the xanthophyll cycle. We propose that experimentally resolved structures capture intermediate or transition states of FCP protein dynamics as a fingerprint of the FCP acclimation state. This is the first demonstration that the structural heterogeneity of FCPs reflects an intrinsic, functionally tunable conformational landscape.\u003c/p\u003e\u003cp\u003eIn the context of this study, we have developed and trained a machine learning (ML) model in a combination of Randrom Forest and Gradient Boosting Regression that can be used to classify FCP structures resolved experimentally into different clusters, or to evaluate those predicted computationally from different diatom species. They can also be used to predict 'equivalent' excitonic coupling values and assign acclimation states for FCP structures. The model can be refined as more MD or crystallographic data become available. It can also be used in terms of workflow to identify key conformations in other protein families.\u003c/p\u003e\u003cp\u003eIt should be noted that the correlation between structure and function (acclimation state) for FCP structures has been established based on the calculated excitonic coupling of a Chlorophyll-fucoxanthin pigment pair within FCPs from Ph. tricornutum and Ch. gracilis. This approach may be limited in its application to FCP complexes from other diatom species. Nevertheless, there is a strong correlation between the computational and experimental results regarding the effect of the photoprotective LHCX1 protein and the xanthophyll cycle on the acclimation state of the FCP in Ph. tricornutum, which positively validates our approach. Our findings highlight the need for further investigation.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003e\u003cstrong\u003e\u003cem\u003eSetup of FCP Models\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe crystal structure of the fucoxanthin chlorophyll a/c -binding (FCP) protein from the diatom Ch. Gracilis (pdb: 7vd5) \u003csup\u003e13\u003c/sup\u003e was used as the initial coordinates to build the monomer models (Sm1 and m2). To construct the S-tetramer, we used the monomer structure of Sm1 and superimposed it with the tetramer chains Sm1, Sm2, Sm3, Sm4 from the Nagao et al. structure \u003csup\u003e13\u003c/sup\u003e. Although chain Sm2 in their structure exhibits slight differences compared to the other three, a previous study \u003csup\u003e15\u003c/sup\u003e refers to the S-tetramer as an homotetramer. All polypeptide chains were described using the Amber ff14SB force field \u003csup\u003e34\u003c/sup\u003e Amber-compatible parameters for fucoxanthin and DD6 were taken from different studies \u003csup\u003e34\u0026ndash;36\u003c/sup\u003e, while chlorophyll a and c parameters were adopted from two separate studies \u003csup\u003e37,38\u003c/sup\u003e. The missing phytyl tails of chlorophyll a were modeled using Schr\u0026ouml;dinger Maestro to restore the complete molecular structure. Protonation states for the Sm1 monomer were assigned as follows: all lumen-facing Asp and Glu residues were protonated to reflect pH 5.5 \u003csup\u003e19\u003c/sup\u003e.\u003csup\u003e\u0026nbsp;\u003c/sup\u003eIn contrast, at pH 7.0, the same Asp and Glu residues are treated as deprotonated in correlation with a previous study on Ph. tricornutum \u003csup\u003e19\u003c/sup\u003e. For the m2 model, Glu86 is protonated, and the remaining Asp and Glu residues follow the same protonation pattern as in the Sm1 monomer. \u003csup\u003e\u0026nbsp;\u003c/sup\u003eHis84 was protonated at N\u0026epsilon;, while all other His residues at \u0026Nu;\u003csub\u003e\u0026delta;\u003c/sub\u003e sites. In both Sm1 and m2 models, Chl-c 304 was treated protonated at lumenal pH 5.5 and deprotonated at lumenal pH 7, as it faces the lumenal side of the membrane with the acrylate group exposed to the acidic lumen.\u003csup\u003e\u0026nbsp;\u003c/sup\u003eA lipid bilayer patch of approximately 350 thylakoid lipids \u003csup\u003e19\u003c/sup\u003e, described by the AMBER force field \u003csup\u003e34\u003c/sup\u003e, was used to embed each all-atom model. Lipid composition was based on the thylakoid membrane model of Chryasfoudi et al. \u003csup\u003e24\u003c/sup\u003e containing 45% MGDG, 25% DGDG, 25% SQDG, and 5% PG\u0026mdash;reflecting an elevated SQDG/PG content (30%) relative to plant thylakoids (15\u0026ndash;20%).\u003csup\u003e39\u003c/sup\u003e The MGDG-DGDG lipid content is at 70% to simulate high-light adapted diatoms compared to low-light grown diatoms (50%) \u003csup\u003e39\u003c/sup\u003e. An amount of around 50000 TIP3P water \u003csup\u003e40\u0026nbsp;\u003c/sup\u003ewas used for solvation, and each system included ~150 mM KCl with additional K⁺ ions to neutralize protein and lipid charges. The equilibrated unit cell dimensions of each model were 16.3 \u0026times; 15.6 \u0026times; 8.5 nm\u003csup\u003e3\u003c/sup\u003e (monomer) as well as 18.8 \u0026times; 17.0 \u0026times; 8.8 nm\u003csup\u003e3\u003c/sup\u003e (tetramer).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eSetup of FCP-LHCX1 models\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe crystal structure of the dimer fucoxanthin chlorophyll a/c -binding (FCP) protein from the diatom Ph. tricornutum (PDB ID: 6A2W) served as the starting template for building FCP-LHCX1 dimer model. The LHCX1 protein 3D structure from Ph. Tricornutum (uniport code B7FYL0) was predicted by RosettaFold (robetta.bakerlab.org) using the default parametrization. The FCP-LHCX1 model was generated by structurally aligning the predicted LHCX1 onto the dimeric FCP scaffold resolved experimentally \u003csup\u003e14\u003c/sup\u003e, by ChimeraX software and replacing the aligned FCP monomer by LHCX1. In the following we provide the LHCX1 sequence from the uniport database.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026gt;tr|B7FYL0|B7FYL0_PHATC Protein fucoxanthin chlorophyll a/c protein OS=Phaeodactylum tricornutum (strain CCAP 1055/1) OX=556484 GN=Lhcx1 PE=3 SV=1\u003c/p\u003e\n\u003cp\u003eMKFAATILALIGSAAAFAPAQTSRASTSLQYAKEDLVGAIPPVGFFDPLGFAD KADSPTLKRYREAELTHGRVAMLAVVGFLVGEAVEGSSFLFDASISGPAITHL SQVPAP FWVLLTIAIGASEQTRAVIGWVDPADAPVDKPGLLRDDYVPGDLGF DPLGLKPSDPEELITLQTKELQNGRLAMLAAAGFMAQELVNGKGILENLQG\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe resulting FCP-LHCX1 complex closely matches the model previously proposed in the literature \u003csup\u003e24\u003c/sup\u003e. All polypeptide chains and pigments were described using the same force fields as in the previous section. The Dtx (diatoxanthin) pigment was parametrized by modifiying the Ddx (diadinoxanthin) parameters specifically by removing one oxygen atom and adapting accordingly. The LHCX1 was considered non-pigmented \u003csup\u003e41\u003c/sup\u003e, Protonation states for the FCP and LHCX1 monomers were assigned as follows: all lumen-facing Asp and Glu residues were protonated to reflect protonation state at pH 5.5. In contrast, at pH 7.0, the same Asp and Glu residues were treated as deprotonated. For the LHCX1 monomer, Glu72, Glu189, and Asp79 were trated as protonated at low pH. All Histidine residues were protonated at \u0026Nu;\u003csub\u003e\u0026delta;\u003c/sub\u003e sites.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eClassical Molecular Dynamics\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFollowing established protocols, all systems were gradually relaxed and equilibrated by progressively releasing positional restraints on the heavy backbone atoms of the protein \u003csup\u003e19\u003c/sup\u003e. \u0026nbsp;During a sequence of simulations in the NVT and NPT ensembles (constant volume and pressure, respectively), the system temperature was gradually raised from 100 K to 303 K before entering the production phase. Classical molecular dynamics (MD) simulations were carried out using the leapfrog integrator available in GROMACS 2021 \u003csup\u003e42\u003c/sup\u003e with a 2.0 fs integration time step. The production runs were performed in the constant pressure NPT ensemble with semi-isotropic couplings in the xy membrane plane and in the z-direction (compressibility at 4.5 \u0026times; 10\u003csup\u003e\u0026ndash;5\u003c/sup\u003e bar\u003csup\u003e\u0026ndash;1\u003c/sup\u003e). Furthermore, the van der Waals interactions were smoothly shifted to zero between 1.0 and 1.2 nm using the Verlet cutoff scheme. Short-range electrostatics were cut off at 1.2 nm, while long-range electrostatic interactions were computed using the particle mesh Ewald (PME) method \u003csup\u003e43,44\u003c/sup\u003e. All bonds between hydrogen atoms and heavy atoms were constrained using the LINCS algorithm \u003csup\u003e45\u003c/sup\u003e. The v-rescale thermostat was used \u003csup\u003e46\u0026nbsp;\u003c/sup\u003e(303 K, temperature coupling constant 0.5) along with the C-rescale\u003csup\u003e[44]\u003c/sup\u003e for equilibration, while the Parrinello\u0026ndash;Rahman barostats \u003csup\u003e47\u0026nbsp;\u003c/sup\u003ewas used for production (1 atm, pressure coupling constant 2.0). Independent trajectories (replicas) were initialized from structures extracted at 10 ns intervals during the final phase of equilibration. Simulation parameters were otherwise consistent with those applied in a prior study of P. tricornutum \u003csup\u003e19\u003c/sup\u003e. The total simulation time for the FCP models amounts to 12 \u0026mu;s, including four independent 0.5 \u0026mu;s trajectories for the tetramer model and four independent 0.5 \u0026mu;s trajectories for each monomeric model (Sm1 and m2), for two distinct pH protonation states (neutral \u0026ndash; low). This simulation time can be translated in monomer-equivalent dynamics of: 2 pH states \u0026times; (4 monomers \u0026times; 2 \u0026mu;s + 1 monomer \u0026times; 2 \u0026mu;s) = 20 \u0026mu;s sampling for the Sm1 monomer and 2 pH states \u0026times; (1 monomer \u0026times; 2 \u0026mu;s) = 4 \u0026mu;s sampling for the m2 FCP monomer in the different multimeric states. The simulations sum to 24 \u0026mu;s monomer-equilvalent dynamics for Ch. Gracilis. The first 100 ns from each trajectory were considered as further equilibration, and the analysis was only performed for the final 400 ns of each independent trajectory. Structures were collected every 1.0 ns for all the trajectories. The total simulation time for the FCP-LHCX1 models amounted to 8 \u0026mu;s, including: four independent trajectories of 1 \u0026mu;s for the low pH and four independent trajectories of 1 \u0026mu;s for the neutral pH.\u003c/p\u003e\n\u003cp\u003eFor details on the simulations for the isolated FCP from Ph. tricornutum (neutral-low pH) and the method of calculation of excitonic couplings between Chl-a 409/ Fx-301 refer to a previous study.\u003csup\u003e19\u003c/sup\u003e\u003cstrong\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eMarkov State modeling analysis (MSM)\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eDimers and tetramers were disassembled into different monomeric units before the analysis. The first 100-200ns were disregarded from each individual trajectory based on the time when the backbone root-mean-square-deviation (RMSD) reaches a plateau, resulting cumulatively in 66.3\u0026nbsp;\u0026mu;s for two diatoms: Ch. Gracilis and Ph. Tricornutum. Only the FCP complex (common minimum core, without protons, ions or pigments) was extracted from the trajectories and all frames (1 ns\u003csup\u003e-1\u003c/sup\u003e) were structurally aligned based on Ca atoms by the GROMACS toolbox (trjconv -fit rot+trans) on a reference common core to assure consistency in the analysis. The PyEMMA package in Jupyter notebooks was employed \u003csup\u003e\u003cspan lang=\"EN-US\"\u003e48\u003c/span\u003e\u003c/sup\u003e. All backbone torsional angles of residues aa 109-131 were chosen as input features. The dimensionality of the configurational space sampled in MD was further reduced. This was achieved by the time-lagged independent component analysis (tICA) to remove any redundant information. A 6-component tICA space and a time lag of 50ns was used for coarse graining the degrees of freedom and identify a set of the slowest modes among all the initial input features (6 vectors). Then different MSMs were constructed with their slowest implied timescales to converge quickly and to be constant within a 95% confidence interval for lag times above 50ns. The MSM passed the Chapman\u0026ndash;Kolmogorov test at 95% confidence. The validation procedure is a standard approach in the field \u003csup\u003e\u003cspan lang=\"EN-US\"\u003e23\u003c/span\u003e\u003c/sup\u003e. A lag time of 50 ns was thus selected for Bayesian MSM model construction. tICA components are the optimal linear combination of input features which maximizes their kinetic variance. The conformations of the FCPs (common core) were projected on the first two tICA vectors (IC-1, IC-2) and the trajectory frames were clustered into 100 cluster-centers (macrostates) by k-means clustering, as implemented in PyEMMA. The resulting macrostates were further coarse grained into a smaller number of four macrostates using PCCA++ as implemented in PyEMMA (conformations C1, C2, C3 and C4). \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eRandom Forest and Gradient Boost Machine Learning Models\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eDue to the size of our data set for the experimentally resolved structures (118), the Random Forest approach is chosen for model training and subsequent predictions. The scikit-learn toolbox was employed for the analysis. The experimentally resolved set2 structures were clustered into four clusters by the k-means algorithm. A Random Forest (RF) machine learning model was trained on the experimentally resolved structures along with the four MSM-predicted conformations (C1-4) with avg. angle, avg. tilt as input features and the classification into the k-means clusters as output targets. Prior to training, feature dimensionality was further reduced using Uniform Manifold Approximation and Projection (UMAP) with Euclidean distance, 10 nearest neighbors, and a minimum distance of 0.2 to enhance classification performance. The resulting model was evaluated using a standard train-test split and classification metrics. The RF f1-scores were at 1.00 (red), 0.80 (orange), 0.73 (blue) and 0.92 (green) with an overall accuracy of 0.88. Without the use of the UMAP dimensionality reduction, the overall f1-scores are 1.00 (red), 0.83 (orange), 0.80 (blue) and 0.67 (green) with the overall accuracy at 0.84.\u003c/p\u003e\n\u003cp\u003eThe Gradient Boost model was trained on 80% of data (input features/ target excitonic couplings) and evaluated on the remaining 20%, yielding performance metrics including mean squared error (MSE) and the coefficient of determination (R\u0026sup2;). \u0026nbsp;\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe work is funded by the Hellenic Foundation for Research \u0026amp; Innovation (H.F.R.I) in the context of the call \u0026ldquo;Basic Research Financing (Horizontal support for all Sciences), National Recovery and Resilience Plan (Greece 2.0) for the project number 014775, with acronym \u0026ldquo;SUNDIAL\u0026rdquo;. This project has received funding from the European Union\u0026rsquo;s Horizon Europe Research and Innovation Program under the Marie Skłodowska-Curie grant agreement No 101119442. Moreover, this work was also supported by computational time granted from the National Infrastructures for Research and Technology S.A. (GRNET S.A.) in the National HPC facility - ARIS - under project ID \u0026bdquo;FCPC\u0026rdquo;. Part of the simulations were performed on a compute cluster funded through the DFG project INST 676/7-1 FUGG.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions:\u0026nbsp;\u003c/strong\u003eConceptualization: V.D.; Methodology: V.D., U.K.; Investigation: T.-I.S., B.Z., S.M.; Visualization: V.D., T.-I.S., B.Z.; Supervision: V.D.; Writing-original draft: V.D.; Writing-review \u0026amp; editing: V.D., U.K.; Molecular Dynamics \u0026amp; Machine Learning: I-T.S., B.Z.; Calculation of Excitonic Couplings: S.M.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e: All other authors declare they have no competing interests.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData and materials availability\u003c/strong\u003e: All data, code details, and material details used in the analyses are available in the main manuscript. Scripts (python codes) for analysis and the datasets used in this study have been deposited in an open access github repository (https://github.com/vdas-upatras/fcp_diatoms).\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eCroce, R. \u0026amp; van Amerongen, H. Light harvesting in oxygenic photosynthesis: Structural biology meets spectroscopy. \u003cem\u003eScience (1979)\u003c/em\u003e \u003cstrong\u003e369\u003c/strong\u003e, eaay2058 (2020).\u003c/li\u003e\n\u003cli\u003eGiovagnetti, V. \u0026amp; Ruban, A. V. The evolution of the photoprotective antenna proteins in oxygenic photosynthetic eukaryotes. \u003cem\u003eBiochem Soc Trans\u003c/em\u003e \u003cstrong\u003e46\u003c/strong\u003e, 1263\u0026ndash;1277 (2018).\u003c/li\u003e\n\u003cli\u003ePascal, A. A. \u003cem\u003eet al.\u003c/em\u003e Molecular basis of photoprotection and control of photosynthetic light-harvesting. \u003cem\u003eNature\u003c/em\u003e \u003cstrong\u003e436\u003c/strong\u003e, 134\u0026ndash;137 (2005).\u003c/li\u003e\n\u003cli\u003eFalciatore, A. \u0026amp; Mock, T. The Molecular Life of Diatoms.\u003c/li\u003e\n\u003cli\u003eVolpe, C. \u0026amp; B\u0026uuml;chel, C. Function, Structure and Organization of Light-Harvesting Proteins in Diatoms. in \u003cem\u003eDiatom Photosynthesis\u003c/em\u003e 191\u0026ndash;215 (2024). doi:https://doi.org/10.1002/9781119842156.ch6.\u003c/li\u003e\n\u003cli\u003eGiossi, C. E., Bitnel, D. B., W\u0026uuml;nsch, M. A., Kroth, P. G. \u0026amp; Lepetit, B. Synergistic effects of temperature and light on photoprotection in the model diatom Phaeodactylum tricornutum. \u003cem\u003ePhysiol Plant\u003c/em\u003e \u003cstrong\u003e177\u003c/strong\u003e, e70039 (2025).\u003c/li\u003e\n\u003cli\u003eMorosinotto, T., Perin, G. \u0026amp; Gambaro, F. Knowledge of regulation of photosynthesis in outdoor microalgae cultures is essential for the optimization of biomass productivity. \u003cem\u003eFront Plant Sci\u003c/em\u003e 751 (2022).\u003c/li\u003e\n\u003cli\u003ePerin, G., Bellan, A., Lyska, D., Niyogi, K. K. \u0026amp; Morosinotto, T. Modulation of xanthophyll cycle impacts biomass productivity in the marine microalga \u0026amp;lt;em\u0026amp;gt;Nannochloropsis\u0026amp;lt;/em\u0026amp;gt; \u003cem\u003ebioRxiv\u003c/em\u003e 2022.08.16.504082 (2022) doi:10.1101/2022.08.16.504082.\u003c/li\u003e\n\u003cli\u003eVecchi, V., Barera, S., Bassi, R. \u0026amp; Dall\u0026rsquo;osto, L. Potential and challenges of improving photosynthesis in algae. \u003cem\u003ePlants\u003c/em\u003e vol. 9 Preprint at https://doi.org/10.3390/plants9010067 (2020).\u003c/li\u003e\n\u003cli\u003eKromdijk, J. \u003cem\u003eet al.\u003c/em\u003e Improving photosynthesis and crop productivity by accelerating recovery from photoprotection. \u003cem\u003eScience (1979)\u003c/em\u003e \u003cstrong\u003e354\u003c/strong\u003e, 857\u0026ndash;861 (2016).\u003c/li\u003e\n\u003cli\u003eShi, Y. \u003cem\u003eet al.\u003c/em\u003e Harnessing the Power of Photosynthesis: from Current Engineering Strategies to Cell Factory Applications. \u003cem\u003eSmall Methods\u003c/em\u003e \u003cstrong\u003en/a\u003c/strong\u003e, 2402147 (2025).\u003c/li\u003e\n\u003cli\u003eFeng, Y. \u003cem\u003eet al.\u003c/em\u003e Structure of a diatom photosystem II supercomplex containing a member of Lhcx family and dimeric FCPII. \u003cem\u003eSci Adv\u003c/em\u003e \u003cstrong\u003e9\u003c/strong\u003e, eadi8446 (2023).\u003c/li\u003e\n\u003cli\u003eNagao, R. \u003cem\u003eet al.\u003c/em\u003e Structural basis for different types of hetero-tetrameric light-harvesting complexes in a diatom PSII-FCPII supercomplex. \u003cem\u003eNat Commun\u003c/em\u003e \u003cstrong\u003e13\u003c/strong\u003e, 1764 (2022).\u003c/li\u003e\n\u003cli\u003eWang, W. \u003cem\u003eet al.\u003c/em\u003e Structural basis for blue-green light harvesting and energy dissipation in diatoms. \u003cem\u003eScience (1979)\u003c/em\u003e \u003cstrong\u003e363\u003c/strong\u003e, eaav0365 (2019).\u003c/li\u003e\n\u003cli\u003ePi, X. \u003cem\u003eet al.\u003c/em\u003e The pigment-protein network of a diatom photosystem II\u0026ndash;light-harvesting antenna supercomplex. \u003cem\u003eScience (1979)\u003c/em\u003e \u003cstrong\u003e365\u003c/strong\u003e, eaax4406 (2019).\u003c/li\u003e\n\u003cli\u003eNagao, R. \u003cem\u003eet al.\u003c/em\u003e Structural basis for energy harvesting and dissipation in a diatom PSII\u0026ndash;FCPII supercomplex. \u003cem\u003eNat Plants\u003c/em\u003e \u003cstrong\u003e5\u003c/strong\u003e, 890\u0026ndash;901 (2019).\u003c/li\u003e\n\u003cli\u003eNagao, R. \u003cem\u003eet al.\u003c/em\u003e Comparison of oligomeric states and polypeptide compositions of fucoxanthin chlorophyll a/c-binding protein complexes among various diatom species. \u003cem\u003ePhotosynth Res\u003c/em\u003e \u003cstrong\u003e117\u003c/strong\u003e, 281\u0026ndash;288 (2013).\u003c/li\u003e\n\u003cli\u003eB\u0026uuml;chel, C. Light harvesting complexes in chlorophyll c-containing algae. \u003cem\u003eBiochimica et Biophysica Acta (BBA) - Bioenergetics\u003c/em\u003e \u003cstrong\u003e1861\u003c/strong\u003e, 148027 (2020).\u003c/li\u003e\n\u003cli\u003eDaskalakis, V., Maity, S. \u0026amp; Kleinekath\u0026ouml;fer, U. An Unexpected Water Channel in the Light-Harvesting Complex of a Diatom: Implications for the Switch between Light Harvesting and Photoprotection. \u003cem\u003eACS Physical Chemistry Au\u003c/em\u003e \u003cstrong\u003e5\u003c/strong\u003e, 47\u0026ndash;61 (2025).\u003c/li\u003e\n\u003cli\u003eMaity, S., Daskalakis, V., Jansen, T. L. C. \u0026amp; Kleinekath\u0026ouml;fer, U. Electric Field Susceptibility of Chlorophyll c Leads to Unexpected Excitation Dynamics in the Major Light-Harvesting Complex of Diatoms. \u003cem\u003eJ Phys Chem Lett\u003c/em\u003e \u003cstrong\u003e15\u003c/strong\u003e, 2499\u0026ndash;2510 (2024).\u003c/li\u003e\n\u003cli\u003eKonagurthu, A. S., Whisstock, J. C., Stuckey, P. J. \u0026amp; Lesk, A. M. MUSTANG: A multiple structural alignment algorithm. \u003cem\u003eProteins: Structure, Function, and Bioinformatics\u003c/em\u003e \u003cstrong\u003e64\u003c/strong\u003e, 559\u0026ndash;574 (2006).\u003c/li\u003e\n\u003cli\u003eZhao, S. \u003cem\u003eet al.\u003c/em\u003e Structural insights into photosystem II supercomplex and trimeric FCP antennae of a centric diatom Cyclotella meneghiniana. \u003cem\u003eNat Commun\u003c/em\u003e \u003cstrong\u003e14\u003c/strong\u003e, 8164 (2023).\u003c/li\u003e\n\u003cli\u003eHusic, B. E. \u0026amp; Pande, V. S. Markov State Models: From an Art to a Science. \u003cem\u003eJ Am Chem Soc\u003c/em\u003e \u003cstrong\u003e140\u003c/strong\u003e, 2386\u0026ndash;2396 (2018).\u003c/li\u003e\n\u003cli\u003eChrysafoudi, A., Maity, S., Kleinekath\u0026ouml;fer, U. \u0026amp; Daskalakis, V. Robust Strategy for Photoprotection in the Light-Harvesting Antenna of Diatoms: A Molecular Dynamics Study. \u003cem\u003eJ Phys Chem Lett\u003c/em\u003e \u003cstrong\u003e12\u003c/strong\u003e, 9626\u0026ndash;9633 (2021).\u003c/li\u003e\n\u003cli\u003eAgostini, A., B\u0026iacute;na, D., Carbonera, D. \u0026amp; Litv\u0026iacute;n, R. Conservation of triplet-triplet energy transfer photoprotective pathways in fucoxanthin chlorophyll-binding proteins across algal lineages. \u003cem\u003eBiochimica et Biophysica Acta (BBA) - Bioenergetics\u003c/em\u003e \u003cstrong\u003e1864\u003c/strong\u003e, 148935 (2023).\u003c/li\u003e\n\u003cli\u003eDauparas, J. \u003cem\u003eet al.\u003c/em\u003e Robust deep learning\u0026ndash;based protein sequence design using ProteinMPNN. \u003cem\u003eScience (1979)\u003c/em\u003e \u003cstrong\u003e378\u003c/strong\u003e, 49\u0026ndash;56 (2022).\u003c/li\u003e\n\u003cli\u003eAkdel, M. \u003cem\u003eet al.\u003c/em\u003e A structural biology community assessment of AlphaFold2 applications. \u003cem\u003eNat Struct Mol Biol\u003c/em\u003e \u003cstrong\u003e29\u003c/strong\u003e, 1056\u0026ndash;1067 (2022).\u003c/li\u003e\n\u003cli\u003eFalciatore, A., Jaubert, M., Bouly, J.-P., Bailleul, B. \u0026amp; Mock, T. Diatom Molecular Research Comes of Age: Model Species for Studying Phytoplankton Biology and Diversity[OPEN]. \u003cem\u003ePlant Cell\u003c/em\u003e \u003cstrong\u003e32\u003c/strong\u003e, 547\u0026ndash;572 (2020).\u003c/li\u003e\n\u003cli\u003eCroteau, D., Jaubert, M., Falciatore, A. \u0026amp; Bailleul, B. Pennate diatoms make non-photochemical quenching as simple as possible but not simpler. \u003cem\u003eNat Commun\u003c/em\u003e \u003cstrong\u003e16\u003c/strong\u003e, 2385 (2025).\u003c/li\u003e\n\u003cli\u003eLepetit, B. \u003cem\u003eet al.\u003c/em\u003e The diatom Phaeodactylum tricornutum adjusts nonphotochemical fluorescence quenching capacity in response to dynamic light via fine-tuned Lhcx and xanthophyll cycle pigment synthesis. \u003cem\u003eNew Phytologist\u003c/em\u003e \u003cstrong\u003e214\u003c/strong\u003e, 205\u0026ndash;218 (2017).\u003c/li\u003e\n\u003cli\u003eKuczynska, P. \u003cem\u003eet al.\u003c/em\u003e The xanthophyll cycle in diatom Phaeodactylum tricornutum in response to light stress. \u003cem\u003ePlant Physiology and Biochemistry\u003c/em\u003e \u003cstrong\u003e152\u003c/strong\u003e, 125\u0026ndash;137 (2020).\u003c/li\u003e\n\u003cli\u003eTaddei, L. \u003cem\u003eet al.\u003c/em\u003e Dynamic changes between two LHCX-related energy quenching sites control diatom photoacclimation. \u003cem\u003ePlant Physiol\u003c/em\u003e \u003cstrong\u003e177\u003c/strong\u003e, 953\u0026ndash;965 (2018).\u003c/li\u003e\n\u003cli\u003eBuck, J. M. \u003cem\u003eet al.\u003c/em\u003e Lhcx proteins provide photoprotection via thermal dissipation of absorbed light in the diatom Phaeodactylum tricornutum. \u003cem\u003eNat Commun\u003c/em\u003e \u003cstrong\u003e10\u003c/strong\u003e, 4167 (2019).\u003c/li\u003e\n\u003cli\u003eWang, J., Wolf, R. M., Caldwell, J. W., Kollman, P. A. \u0026amp; Case, D. A. Development and testing of a general Amber force field. \u003cem\u003eJ Comput Chem\u003c/em\u003e \u003cstrong\u003e25\u003c/strong\u003e, 1157\u0026ndash;1174 (2004).\u003c/li\u003e\n\u003cli\u003eWang, Y., Mao, L. \u0026amp; Hu, X. Insight into the structural role of carotenoids in the Photosystem I: a quantum chemical analysis. \u003cem\u003eBiophys J\u003c/em\u003e \u003cstrong\u003e86\u003c/strong\u003e, 3097\u0026ndash;3111 (2004).\u003c/li\u003e\n\u003cli\u003ePrandi, I. G., Viani, L., Andreussi, O. \u0026amp; Mennucci, B. Combining classical molecular dynamics and quantum mechanical methods for the description of electronic excitations: The case of carotenoids. \u003cem\u003eJ Comput Chem\u003c/em\u003e \u003cstrong\u003e37\u003c/strong\u003e, 981\u0026ndash;991 (2016).\u003c/li\u003e\n\u003cli\u003eCeccarelli, M., Procacci, P. \u0026amp; Marchi, M. An ab initio force field for the cofactors of bacterial photosynthesis. \u003cem\u003eJ Comput Chem\u003c/em\u003e \u003cstrong\u003e24\u003c/strong\u003e, 129\u0026ndash;142 (2003).\u003c/li\u003e\n\u003cli\u003eZhang, L., Silva, D. A., Yan, Y. \u0026amp; Huang, X. Force field development for cofactors in the photosystem II. \u003cem\u003eJ Comput Chem\u003c/em\u003e \u003cstrong\u003e33\u003c/strong\u003e, 1969\u0026ndash;1980 (2012).\u003c/li\u003e\n\u003cli\u003eLepetit, B., Goss, R., Jakob, T. \u0026amp; Wilhelm, C. Molecular dynamics of the diatom thylakoid membrane under different light conditions. \u003cem\u003ePhotosynth Res\u003c/em\u003e \u003cstrong\u003e111\u003c/strong\u003e, 245\u0026ndash;257 (2012).\u003c/li\u003e\n\u003cli\u003eMark, P. \u0026amp; Nilsson, L. Structure and dynamics of the TIP3P, SPC, and SPC/E water models at 298 K. \u003cem\u003eJ Phys Chem A\u003c/em\u003e \u003cstrong\u003e105\u003c/strong\u003e, 9954\u0026ndash;9960 (2001).\u003c/li\u003e\n\u003cli\u003eGiovagnetti, V. \u003cem\u003eet al.\u003c/em\u003e Biochemical and molecular properties of LHCX1, the essential regulator of dynamic photoprotection in diatoms. \u003cem\u003ePlant Physiol\u003c/em\u003e \u003cstrong\u003e188\u003c/strong\u003e, 509\u0026ndash;525 (2022).\u003c/li\u003e\n\u003cli\u003eBerendsen, H. J. C., van der Spoel, D. \u0026amp; van Drunen, R. GROMACS: A message-passing parallel molecular dynamics implementation. \u003cem\u003eComput Phys Commun\u003c/em\u003e \u003cstrong\u003e91\u003c/strong\u003e, 43\u0026ndash;56 (1995).\u003c/li\u003e\n\u003cli\u003eYeh, I.-C. \u0026amp; Berkowitz, M. L. Ewald summation for systems with slab geometry. \u003cem\u003eJ Chem Phys\u003c/em\u003e \u003cstrong\u003e111\u003c/strong\u003e, 3155\u0026ndash;3162 (1999).\u003c/li\u003e\n\u003cli\u003eDarden, T., York, D. \u0026amp; Pedersen, L. Particle mesh Ewald: An N\u0026sdot; log (N) method for Ewald sums in large systems. \u003cem\u003eJ Chem Phys\u003c/em\u003e \u003cstrong\u003e98\u003c/strong\u003e, 10089\u0026ndash;10092 (1993).\u003c/li\u003e\n\u003cli\u003eHess, B., Bekker, H., Berendsen, H. J. C. \u0026amp; Fraaije, J. G. E. M. LINCS: a linear constraint solver for molecular simulations. \u003cem\u003eJ Comput Chem\u003c/em\u003e \u003cstrong\u003e18\u003c/strong\u003e, 1463\u0026ndash;1472 (1997).\u003c/li\u003e\n\u003cli\u003eBussi, G., Donadio, D. \u0026amp; Parrinello, M. Canonical sampling through velocity rescaling. \u003cem\u003eJ Chem Phys\u003c/em\u003e \u003cstrong\u003e126\u003c/strong\u003e, 14101 (2007).\u003c/li\u003e\n\u003cli\u003eParrinello, M. \u0026amp; Rahman, A. Polymorphic transitions in single crystals: A new molecular dynamics method. \u003cem\u003eJ Appl Phys\u003c/em\u003e \u003cstrong\u003e52\u003c/strong\u003e, 7182\u0026ndash;7190 (1981).\u003c/li\u003e\n\u003cli\u003eScherer, M. K. \u003cem\u003eet al.\u003c/em\u003e PyEMMA 2: A Software Package for Estimation, Validation, and Analysis of Markov Models. \u003cem\u003eJ Chem Theory Comput\u003c/em\u003e \u003cstrong\u003e11\u003c/strong\u003e, 5525\u0026ndash;5542 (2015).\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"nature-portfolio","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"","title":"Nature Portfolio","twitterHandle":"","acdcEnabled":false,"dfaEnabled":false,"editorialSystem":"ejp","reportingPortfolio":"","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-7326805/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7326805/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eBiomolecules exhibit a fundamental correlation between structure and function, which can be modulated by environmental factors. Deciphering this relationship remains a central and long-standing challenge for many protein families. In this study, we investigate such a correlation in the light-harvesting complexes (LHCs) of diatoms; unicellular, photosynthetic organisms that thrive in marine ecosystems. Using μs-long molecular dynamics simulations and machine learning, we reveal that all experimentally resolved LHC configurations correspond to a few distinct interconverting states linked to an intrinsic transition between light-harvesting and photoprotective mode; a property that can be tuned or engineered. Thus, we provide an original view on the plethora of experimentally resolved structures. Our model strongly correlates with experimental findings on the effect of the photoprotective protein LHCX1 and the xanthophyll cycle on the FCP acclimation states.\u003c/p\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e*Theofani-Iosifina Sousani \u0026amp; Boutheina Zender contributed equally to this work.\u003c/strong\u003e\u003c/p\u003e","manuscriptTitle":"Conformational Plasticity Enables Functional Switching in Diatom Light-Harvesting Complexes","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-08-12 09:31:48","doi":"10.21203/rs.3.rs-7326805/v1","editorialEvents":[],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"communications-chemistry","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"commschem","sideBox":"Learn more about [Communications Chemistry](http://www.nature.com/commschem/)","snPcode":"","submissionUrl":"","title":"Communications Chemistry","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Communications Series","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"285844ad-eaab-480d-b0b0-6590298b5457","owner":[],"postedDate":"August 12th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":52953362,"name":"Physical sciences/Chemistry/Chemical biology/Biophysical chemistry"},{"id":52953363,"name":"Biological sciences/Structural biology/Molecular modelling"}],"tags":[],"updatedAt":"2025-12-06T08:06:06+00:00","versionOfRecord":{"articleIdentity":"rs-7326805","link":"https://doi.org/10.1038/s42004-025-01774-x","journal":{"identity":"communications-chemistry","isVorOnly":false,"title":"Communications Chemistry"},"publishedOn":"2025-12-05 05:00:00","publishedOnDateReadable":"December 5th, 2025"},"versionCreatedAt":"2025-08-12 09:31:48","video":"","vorDoi":"10.1038/s42004-025-01774-x","vorDoiUrl":"https://doi.org/10.1038/s42004-025-01774-x","workflowStages":[]},"version":"v1","identity":"rs-7326805","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7326805","identity":"rs-7326805","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}