Oxygen-evolving photosystem II structures during S1–S2–S3 transitions – Nature

Sample preparation

Samples of the PSII microcrystals were prepared as in the previous SFX studies conducted at room temperature^4,25, with a few minor adjustments. In brief, cells of the thermophilic cyanobacterium Thermosynechococcus vulcanus were grown in a previously described medium^41,42,43,44 in eight 5-l bottles, to a density of OD_730 nm = 2.5–3.0, and collected as described previously^41,42,43,44. The cells were resuspended in a buffer of 40 mM KH₂PO₄-KOH (pH 6.8) and 0.4 M mannitol, and treated with 1.21 g l⁻¹ lysozyme (FUJIFILM Wako Pure Chemical Corporation) at 37 °C for 90 min with constant shaking. The treated cells were pelleted by centrifugation at 13,700g for 15 min, suspended in 25% (w/v) glycerol, 20 mM HEPES-NaOH (pH 7.0) and 10 mM MgCl₂ (buffer A), and stored at −80 °C until use.

The frozen cells were thawed, to which ten folds of a buffer containing 30 mM HEPES-NaOH (pH 7.0) and 10 mM MgCl₂ were added to disrupt the cells by freeze-thawing and osmotic shock. After centrifugation at 13,700g for 15 min, pelleted thylakoids were suspended in 5% (w/v) glycerol, 20 mM HEPES-NaOH (pH 7.0) and 10 mM MgCl₂. Crude PSII particles were obtained from the thylakoids by a two-step solubilization with a detergent N,N-dimethyldodecylamine N-oxide (LDAO) (Sigma-Aldrich, 40236-250ML). In the first step, the thylakoids were treated with 0.16% (w/v) LDAO for 5 min on ice, and centrifuged at 43,200g for 60 min. The pellet obtained was suspended in buffer A, and treated with 0.27% (w/v) LDAO for 5 min again. The mixture was centrifuged at 100,000g for 1 h, and the supernatant was recovered. After the addition of 50 (w/v) polyethylene glycol (PEG) 1450 to a final concentration of 15%, crude PSII particles were recovered by centrifugation at 100,000g for 30 min, and resuspended in buffer A^41,42,43,44.

The PSII crude particles were treated with 1.0% n-dodecyl-β-D-maltoside (β-DDM) (FUJIFILM Wako Pure Chemical Corporation, D316) for 5 min, and loaded onto a Q-Sepharose high-performance column (Cytiva) pre-equilibrated with 5% (w/v) glycerol, 30 mM MES-NaOH (pH 6.0), 3 mM CaCl₂ and 0.03% β-DDM (buffer B) in a cooled chamber at 6 °C. The column was washed with eight to ten folds of the column volume of buffer B containing 170 mM NaCl, and eluted with a liner gradient of 12.5 folds of the column volume of 170–300 mM NaCl in buffer B. Elution peaks first appeared for PSII monomer, followed by PSII dimer and PSI monomer, among which PSII dimers were collected. The PSII dimers collected were diluted threefold by buffer B without DDM, and PEG 1450 was added to a final concentration of 13%. The PSII dimers were centrifuged at 100,000g for 30 min, and the pellet was suspended in buffer B without DDM and stored in liquid nitrogen until use^41,42,43,44.

To make microcrystals of the PSII dimer, the sample was diluted with 20 mM MES-NaOH (pH 6.0), 40 mM MgSO₄, 20 mM NaCl and 10 mM CaCl₂, followed by additions of n-heptyl-β-D-thioglucopyranoside (HTG) (FUJIFILM Wako Pure Chemical Corporation, H015) and PEG 1450 to final concentrations of 0.85% (w/v) and around 5.50–5.75% (w/v), respectively, at a final concentration of 2.25 mg chlorophyll per ml (refs. ^4,6). Microcrystals were grown in a 2.0-ml glass vial (J.G. Finneran Associates, 9800-1232), and 150 μl PSII dimer sample was put into each vial. After standing for 20–30 min at 20 °C, the solution was mixed gently and left to stand for another 10–30 min to allow the microcrystals to grow. In cases in which microcrystals did not appear or appeared in small numbers, the mixing-and-standing step was repeated until enough microcrystals appeared.

After the microcrystals appeared, they were allowed to grow to a maximum size of 100 μm in length for several hours to overnight, following which 150 μl of a crystal storage buffer containing 7% (w/v) PEG 1450, 20 mM MES-NaOH (pH 6.0), 20 mM NaCl, 10 mM CaCl₂ and 0.85% (w/v) HTG was added to stop the growth of the microcrystals. After collection of the microcrystals, the supernatant was discarded, and the microcrystals were stored in the crystal storage buffer at 20 °C until the X-ray free electron laser (XFEL) experiments. It is important to store the microcrystals in the crystal storage buffer for more than 24 h to ensure high resolution, and they are stable in the crystal storage buffer for at least three days but not more than seven days^4,6.

Before conducting the diffraction experiment, a 10 mM potassium ferricyanide solution was added to the PSII microcrystal solution under dim green light, and one pre-flash was given at 20 °C with a laser at a wavelength of 532 nm and an energy of 52 mJ cm⁻². The microcrystals were subsequently transferred to 7% (w/v) PEG 1450, 20 mM MES-NaOH (pH 6.0), 20 mM NaCl, 10 mM CaCl₂, 0.85% (w/v) HTG, 2% dimethyl sulfoxide (DMSO) and 10 mM potassium ferricyanide, and incubated for 10 min at 20 °C. The solution was finally replaced by a cryoprotectant solution containing 10% (w/v) PEG 1450, 10% (w/v) PEG monomethyl ether 5000, 23% (w/v) glycerol, 20 mM NaCl, 10 mM CaCl₂, 0.85% (w/v) HTG, 2% DMSO and 10 mM potassium ferricyanide for six steps, with each step for 10 min at 20 °C (refs. ^4,6).

After replacement of the solution with the cryoprotectant solution, PSII microcrystals were gently mixed with a vacuum grease of a nuclear power grade (Super Lube, 42150)⁴⁵. The ratio of grease to microcrystals was 200 μl to 50 μl (obtained from 4–5 mg chlorophyll), and to avoid physical damage to the microcrystals, the mixing was conducted gently for 2 min. The mixture was exposed to air at 20 °C for around 30–60 min to dehydrate further, before being used for the diffraction experiments at room temperature in darkness⁴. The total time from cryoprotectant replacement to XFEL experiments was one to two hours.

Diffraction experiment

The dark and 1F data, as well as the 1F and 2F time-delayed data, were collected in two independent experiments, resulting in a total of 14 experimental datasets (Extended Data Table 1). Unless otherwise stated, the experimental set-ups were identical for both beamtimes. Diffraction images were obtained using single-shot XFELs collected at the BL2 beamline in the SPring-8 Ångstrom Compact Free Electron Laser (SACLA)⁴⁶. The parameters of the XFEL pulses were as follows: pulse duration 10 fs, energy 10 keV, beam size 3.0 μm (H) × 3.0 μm (W) and repetition rate 10 Hz (ref. ⁴). The PSII microcrystals were excited using pump lasers with the following parameters: pulse duration 6 ns (FWHM, Gaussian), energy 42 mJ cm⁻², focused spot size 240 μm (top-hat), wavelength 532 nm and frequency rate 10 Hz (ref. ⁴). To ensure efficient excitation, one laser beam was split into two beams that focused on the same point of the sample from two different directions separated by an angle of 160° (ref. ⁴).

The injector containing the mixture of PSII microcrystals and grease was carefully inserted into a sample chamber, in which the mixture was ejected from the injector using liquid pressure, ultimately forming a micrometre-sized liquid stream^47,48.

The sample flow rate is regulated by adjusting the fluid pressure in the injector. For the ‘dark’ sample, the flow rate is 1.99 μl min⁻¹, whereas for the ‘light’ samples, it is 7.80 μl min⁻¹. As described previously, by maintaining this flow rate, contamination from the prior lasers is effectively avoided²⁵. The dark dataset was obtained by directly exposing the sample stream to XFELs, whereas the 1F and 2F datasets were acquired by illuminating the sample stream with the pump laser first, followed by exposure to the XFELs after a specified delay time Δt. The values of Δt1 and Δt2 were set to 20 ns, 200 ns, 1 μs, 30 μs, 200 μs and 5 ms, respectively (Extended Data Fig. 1d). In addition, in the 2F time-delay experiment, the time interval between the first and second flash was set to 5 ms (Extended Data Fig. 1d), which is enough to fully transform the S₁ state to the S₂ state after 1F. The focal centres of the lasers and XFELs were the same for data with a Δt of 20 ns–200 μs, but for data with a Δt of 5 ms, the focal centres of lasers were set 60 μm higher than those of the XFELs to prevent the light-excited microcrystals from escaping the XFEL irradiations after a Δt of 5 ms. Diffraction spots were recorded using a Rayonix MX300-HS detector, which was positioned 240 mm from the sample.

Data processing

During the beamtime, we used Cheetah⁴⁹ (https://github.com/keitaroyam/cheetah) and CrystFEL (v.0.6.3)^50,51 to observe and analyse the diffraction images. The analyses provide hit rates, the number of indexed images and approximate resolutions for each dataset, which greatly aided us in devising an effective data-collection strategy. For the processing of diffraction images at the beamline, we at first used approximately 10,000 indexed diffraction images from lysozyme crystals to determine the beam centre and camera length accurately. These parameters were then supplied to CrytFEL for processing the PSII diffraction images. The PSII diffraction images were indexed with ‘indexamajig’, using the Dirax^50,51 indexing method with unit-cell parameters of a = 124.7 Å, b = 229.89 Å, c = 285.5 Å, α = β = γ = 90° adopted from PDB code 5WS5 (ref. ⁴). The resulting individual intensities were merged using ‘process_hkl’ and the reflection data were evaluated using ‘compare_hkl’ (refs. ^50,51).

After data collection, ‘cctbx.xfel’ was used for the indexing and integration of diffraction images, as well as for merging reflections^52,53. The accuracy of the beam centre and camera length obtained from CrystFEL were verified by using the program ‘cspad.cbf_metrology’ (refs. ^52,53). The PSII diffraction images were indexed and integrated using ‘dials.stills_process’ (ref. ⁵⁴), incorporating the determined detector information and targeted unit-cell parameters mentioned above. Individual reflections were merged by the program ‘cxi.merge’ (refs. ^52,53) with the post-refinement rs2 algorithm, and a filter based on the value of I/sigma was not applied so as to include weak signals at high resolutions. The average unit cell, calculated from all of the datasets collected in the same experiment, was used to merge each individual dataset once again. All datasets were processed to a resolution of 2.15–2.30 Å on the basis of the criteria of CC_1/2 of around 50% (Extended Data Table 1).

Structural refinement for the dark and 1F datasets

Molecular replacement for the dark data was performed using Phaser-MR from PHENIX⁵⁵ with the PSII structure solved at 2.35-Å resolution and at room temperature (PDB code: 5WS5) as the search model, in which water molecules and the OEC were removed⁴. Next, rigid body refinement was applied to the resultant model for one cycle. Subsequently, the B factor was set to 20 for all atoms in the model, and the atomic coordinates and temperature factors of atoms were refined by ‘Phenix.refine’ in the resolution range of 2.15–20.0 Å, in conjunction with manual modifications by Coot⁵⁶. We iteratively carried out reciprocal space refinement using ‘Phenix.refine’ and real-space refinement using Coot until the structures of residues and cofactors were confined. Then, the OEC and water molecules were added to the model. Geometric restraints of the OEC are based on the Mn₄CaO₅ cluster solved at 2.15 Å using microcrystals at cryo-temperature (PDB code: 6JLJ)⁶, with a loose distance restraint of σ = 0.06 Å on Mn–O and Ca–O distances, whereas no restraints were provided for the Mn–Mn and Mn–Ca distances. Any pre-existing water molecules exhibiting negative mF_o-DF_c signals or lacking 2mF_o-DF_c signals were removed from the model. New water molecules were constructed at the positions of positive spherical mF_o-DF_c signals over 4σ, and these water molecules were examined after subsequent rounds of reciprocal and real-space refinements to confirm. Finally, a TLS refinement was applied.

For the refinement of the 1F model in the two-flash time-delay experiments, we assigned a single conformation to the OEC and ligands, considering that the geometry of the OEC does not differ much between S₁ and S₂ states. During the refinement process, the Mn–Mn and Mn–Ca distances were not restrained, whereas the distances of Mn–O and Ca–O were restrained to the values observed in the 1F model solved at 2.15 Å (PDB code: 6JLK)⁶, and refined with a loose restraint (σ = 0.06 Å). W16 was removed from the model owing to the emergence of a negative mF_o-DF_c signal when W16 was present, even at low occupancy. Conversely, W10 was retained because its deletion resulted in a significant positive mF_o-DF_c signal at the corresponding location.

Difference-map calculations and structural refinement of intermediates

The phases obtained from the well-refined dark and 1F models were used to calculate isomorphous-difference Fourier maps between dark and 1F time-delayed data, and between 1F and 2F time-delayed data, respectively. Substantial difference densities were detected in the Q_A–Q_B, P680, Y_Z and OEC channel regions at each time point, with their locations dynamically varying over time (Figs. 1–4 and Extended Data Fig. 7). To refine the dynamic intermediate structures conveniently and effectively, we devised double conformations for all residues, water molecules and ligands within a spherical range of 20 Å centred on the Ca of the OEC and the non-haem iron, with A and B conformations corresponding to structures of the ground state and intermediate state, respectively. In this case, unstable water molecules and residues in the intermediate state were also built into the structures. Whether to preserve or delete these water molecules is decided by examining the mF_o-DF_c signal. For example, in the case of W16, which became very unstable after 1F, building two conformations resulted in a strong negative signal on W16. Therefore, we deleted the B conformation of W16. On the other hand, for other unstable water molecules, such as W7 and W10, building two conformations did not result in a particularly strong negative mF_o-DF_c signal, so their B conformations were preserved. Populations of S_i state in PSII crystals were estimated to be 0.4/0.6 for S₁/S₂ after 1F and 0.49/0.51 for S₂/S₃ after 2F, on the basis of flash-induced Fourier transform infrared (FTIR) measurements^4,6,57. On the basis of these ratios, we constructed the 1F structure by adopting two conformations for those atoms or residues that showed structural changes between S₁ and S₂. The S₂-state structure was refined against the density map, whereas the S₁-state structure was taken from the dark structure solved in the present study. On the other hand, in the 2F data, the structure of PSII that does not advance to the S₃ state is a mixture of S₁ and S₂. Owing to the small structural changes between S₁ and S₂, we fixed the structure to the S₂ state for PSII that does not advance to the S₃ state after 2F, and refined the S₃-state structure against the density map. These assignments do not pose major problems for modelling the structures according to the densities obtained. We refined the xyz coordinates of the B conformation, followed by refining the B factors of both the A and the B conformation, and applied TLS refinement at last.

O6* was modelled as a water molecule with an occupancy of 0.51, without imposing artificial constraints on its distance to Ca and the nearby water molecules. The structures of the OEC containing O6 at Δt2 = 200 μs and Δt2 = 5 ms were investigated using three different O5–O6 distances: 1.9 Å, 2.2 Å and 2.4 Å, as indicated by theoretical calculations^37,58. The optimal distance was determined by assessing the magnitude of the adjacent mF_o-DF_c signals (Extended Data Fig. 6).

We need to point out that, although the XFEL data collected in the present study have a high quality, and the resolutions obtained are high, uncertainties exist with regard to the subtle structural changes that occur during S₁–S₂–S₃ transitions, and it is important not to overinterpret the crystallographic data presented in this study.

Estimation of errors in inter-atomic distances

To estimate the errors in the inter-atomic distances, we used the resampling method, creating ten substructures with reduced data multiplicity. Subsequently, we calculated the standard deviations of atom–atom distances within these ten substructures. We resampled our XFEL data by the jackknifing method⁵⁹. We began with a dataset consisting of 100% images and created ten sub-datasets by merging 75% randomly selected images. Subsequently, we refined ten substructures against these sub-datasets. To initiate the refinement of the substructures, we used the well-refined structure derived from the 100% image dataset as our starting model, resetting the temperature factors of all atoms to 20 Ų and applying simulated annealing. After this, we performed refinements on the rigid body, atom position coordinates, temperature factors and TLS. The standard deviations of atom–atom distances were calculated across the ten substructures, which were used as estimates of the errors associated with the corresponding atom–atom distances in the determined structures (Extended Data Fig. 3).

Density functional theory calculations

An OEC model of the S₃ state for density functional theory (DFT) calculations was constructed from the XFEL model (monomer A) of PSII (PDB code: 6JLL)⁶. This model comprises 408 atoms, including the inorganic Mn₄CaO₅ cluster, 4 terminal aqua/hydroxo ligands at Ca and Mn4, 15 crystal waters along with one extra hydroxide anion referred to as O6*, one chloride anion, and the following amino acid residues: D1-D61, D1-N87, Yz, D1-Q165, D1-S169, D1-D170, D1-N181, D1-V185, D1-F182 (backbone only), D1-E189, D1-H190, D1-N296, D1-N298 (fragment), D2-K317 (fragment), D1-H332, D1-E333, D1-A336, D1-H337, D1-D342, D1-A344 (C terminus), CP43-E354, CP43-R357, CP43-L401, CP43-V410 and CP43-A411. The revision made to the previous computational model^6,37 involves augmenting it with the incorporation of five water molecules next to O6*, called a ‘water wheel’, along with four supporting amino acid residues (D1-N296, CP43-L401, CP43-V410 and CP43-A411). Geometric optimizations for the hydroxo form of O6* bound to the Ca site of (Mn^IV)₃Mn^IIICaO₅ were carried out at multiplicity 14 (M_S = 13/2) using the B3LYP hybrid functional⁶⁰ augmented with the D3 version of Grimme’s empirical dispersion correction and the Becke–Johnson damping function^61,62, in combination with the Los Alamos (LANL2DZ) pseudopotential basis set for Ca and Mn and 6-31G(d) for all other atoms^63,64,65,66. A crucial requirement for the production of meta-stable Ca²⁺-bound hydroxo form of O6*, as displayed in Extended Data Fig. 5a,b, is the absence of a Y_Z radical (Tyr_Z–O^•…⁺HN–His190), as the pK_a value of Ca²⁺-bound water (around 12.7 in aqueous solution)^67,68 might be much higher than that of the histidine residue (6.0) (ref. ⁶⁹), even within the protein environment.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.