Main

PSII is a multi-subunit pigment-protein complex embedded in the thylakoid membranes of higher plants, green algae and cyanobacteria, and is the only molecular machine capable of oxidizing water by use of visible light. Water molecules are split into electrons, hydrogen atoms and oxygen molecules at the catalytic centre of PSII, namely, the OEC, through four electron and/or proton removing steps as described in the S i -state cycle (with i = 0–4, where i indicates the number of oxidative equivalents accumulated). Because of its ability to split water, the OEC is considered a promising template for the synthesis of artificial catalysts for water-splitting aimed at obtaining clean and renewable energy from sunlight, which is considered a promising way to supplement the consumption of limited and environmentally unfriendly fossil fuels.

In order to elucidate the mechanism of the water-splitting reaction, the structure of PSII has been studied extensively by XRD, with a resolution that has gradually increased from 3.8 to 1.9 Å using synchrotron radiation (SR) X-ray sources1,15,16,17,18. In particular, the SR structure of PSII at atomic resolution revealed that the OEC is a Mn4CaO5 cluster organized into a distorted-chair shape, in which the cuboidal part is composed of Mn3CaO4 and the outer manganese is attached via two μ-oxo-bridges1. The high-resolution structure also revealed that four water molecules are coordinated to the Mn4CaO5 cluster, among which, two are coordinated to the Ca2+ ion and two are attached to the outer manganese1 (designated Mn4A; the naming of the Mn ions employed in this study combines the nomenclatures of those used in the EXAFS4 and XRD1 studies). Although these provided an important structural basis for the mechanism of water-splitting, the average Mn–ligand and Mn–Mn distances were found to be slightly longer than those deduced from EXAFS3,4,5,6,7 and from computational analysis based on the SR structure8,9,10,11,12,13,14. This has been suggested to result from radiation damage, as hydrated electrons generated by X-ray irradiation19 are able to reduce metal catalytic centres (such as the OEC in PSII) rapidly, although the SR PSII structure at 1.9 Å resolution was obtained with an X-ray dose much reduced compared to those used previously15,16,17,18. Based on the EXAFS analysis, the dose used for collecting the SR structure at 1.9 Å resolution may cause 25% of the Mn ions in the OEC to be reduced to 2+ ions2. Such radiation damage to biological samples is inevitable with SR X-ray sources, since the shortest time to collect a diffraction image is of the order of one second, which is long enough for structural changes to occur upon breakage of covalent bonds by the radicals generated by the strong X-ray irradiation.

In order to eliminate the radiation damage, a new approach employing a femtosecond X-ray free electron laser (XFEL) has been developed which demonstrated that a radiation-damage-free structure may be obtained by using the ultra-short and high-brilliance XFEL pulses to collect diffraction images before the onset of radiation damage (a ‘diffraction before destruction’ approach)20. So far XFEL has been mainly used to collect diffraction data from nanocrystals using the method of serial femtosecond crystallography (SFX), in which a stream of nanometre-to-micrometre sized crystals flows across the XFEL beam and a large number of single-crystal diffraction data are collected in random orientations at room temperature, and the data obtained are processed by a Monte Carlo approach21,22,23. However, as diffraction intensities are proportional to the diffraction volume and inversely proportional to the unit cell volume24, the resolution obtainable with this method is limited, especially for large membrane protein complexes such as PSII and cytochrome c oxidase (CcO), whose unit cell volumes exceed 6 MÅ3. In addition, the nanometre–micrometre sized crystals used for SFX are often obtained under sub-optimal conditions, and their properties are easily affected by post-crystallization handling procedures; all of these will limit the resolution of large membrane protein crystals. Very recently, it was demonstrated that a radiation-damage-free structure can be obtained from large crystals using the XFEL beam by collecting still diffraction images at a cryogenic temperature from well separated positions in a crystal and by using a number of large, highly isomorphous crystals. This method has been used successfully to obtain a radiation-damage-free structure of CcO at 1.9 Å resolution25. Here we determined the radiation-damage-free structure of PSII in the dark-stable S1 state by use of this method and hundreds of large PSII crystals with high isomorphism.

Still diffraction images were collected with XFEL provided by SACLA and large PSII crystals (1.2 × 0.5 × 0.2 mm). Each irradiation point on a crystal was separated by 50 µm (Extended Data Fig. 1a, b), which has been shown to be enough to avoid the effect of radiation damage from the previous irradiation point25 (see Methods). The crystals were rotated in steps of 0.2° following the previous irradiation, which was smaller than one-third of the typical mosaic spread of the PSII crystals; this ensured data quality high enough for the subsequent structural refinement25. Diffraction spots up to 1.85 Å resolution were observed (Extended Data Fig. 2); we collected two independent data sets, data set 1 with 5,592 still images from 254 crystals, and data set 2 with 2,058 images from 82 crystals, from which two complete data sets were processed, both to 1.95 Å resolution. The total absorbed dose and photon density of each irradiation point for both data sets were 1.4 MGy and 0.26 × 1010 photons per µm2, respectively, which were approximately equal to that each irradiation point received in the SR data set used for the analysis of the 1.9 Å structure1. Structure analysis was performed as described in Methods, and the statistics for the diffraction experiments and structural refinement are given in Extended Data Table 1. The Rfree values were 0.238 (data set 1) and 0.225 (data set 2), which, together with the 1.95 Å resolution, allowed us to build the structure of the OEC unambiguously. In this structure, the positions of heavy metals were confirmed clearly, even in the anomalous difference Fourier map (Fig. 1a, Extended Data Fig. 3a, b). In fact, the B-factors for the OEC atoms (23.6 Å2) were found to be lower than those observed in the overall protein atoms of both XFEL (35.1 Å2) and SR (35.2 Å2) structures (Table 1, Extended Data Tables 1 and 2), showing the reliability of the OEC structure. Nevertheless, we repeated the same experiments using the crystals from different preparations in the two data sets to verify the reproducibility of the bond distances within the OEC, which showed a very good agreement among the four PSII monomers in the two independent dimer structures.

Figure 1: Atomic structure of the OEC by XFEL.
figure 1

a, Electron density maps of the OEC. Grey, 2mFo − DFc map contoured at 7σ; magenta, 2mFo − DFc map contoured at 15σ; green, mFo − DFc map contoured at 6σ. The 2mFo − DFc map was calculated before placing any atoms in the OEC, and the mFo − DFc map was calculated after introducing the manganese and calcium atoms but without the oxygen atoms and water molecules. Colour codes for atoms: grey, manganese; blue, calcium; red, oxo-oxygen; yellow, O5; orange, water. Numbers 1–4 in the red spheres indicate O1–O4 atoms, and 1D, 2C, 3B and 4A indicate the four manganese ions named by a combination of crystal structure and previous EXAFS numbering. b, Mn–Mn distances in the OEC (in Å). Roman numerals in the spheres indicate the valence of manganese atoms assigned in this work. Oxygen atoms in the OEC are shown pink. c, Mn–Ca distances in the OEC (in Å). d, Mn–O, Ca–O, Mn–water and Ca–water distances in the OEC (in Å).

PowerPoint slide

Full size image
Table 1 OEC interatomic distances obtained using XFEL and SR techniques
Full size table

The overall architecture of the OEC in the XFEL structure is very similar to that of the SR structure (PDB ID 3ARC), but distinct differences were observed in Mn–Mn and Mn–O distances (Figs 1b–d and 2a, Extended Data Table 3). For comparison, we take the average distances from four monomers of the two independent dimers, and the average Mn–Mn distances were 2.7 Å for Mn1D–Mn2C, 3.2 Å for Mn1D–Mn3B, 5.0 Å for Mn1D–Mn4A, 2.7 Å for Mn2C–Mn3B, 5.2 Å for Mn2C–Mn4A and 2.9 Å for Mn3B–Mn4A (Fig. 1b, Table 1, Extended Data Table 2). All of these distances were 0.1–0.2 Å shorter than those in the SR structure, in good accordance with the results of EXAFS3,4,5,6,7 and theoretical calculations8,9,10,11,12,13,14. This clearly indicated that the X-ray damage possibly included in the previous SR structure has been eliminated. In particular, we assigned two 2.7 Å Mn–Mn vectors to Mn1D–Mn2C and Mn2C–Mn3B and one 2.8-2.9 Å vector to Mn3B–Mn4A; the presence of these three vectors has been suggested by precision EXAFS measurements, although their exact geometry may not be the same as that revealed from the crystal structure3,4,5,6,7. In contrast to the slightly shorter Mn–Mn distances, no clear differences were observed in the Mn–Ca distances compared with those in the SR structure (there were three proximal Mn–Ca distances with 3.5 Å for Mn1D–Ca, 3.3 Å for Mn2C–Ca, 3.4 Å for Mn3B–Ca, and one distal distance of 3.8 Å for Mn4A–Ca, see Fig. 1c, Table 1, Extended Data Table 3). The OEC model proposed from the EXAFS studies has possibly one to two interactions of Mn–Ca distance around 3.9 Å, which is different from the present XFEL structure4.

Figure 2: Comparison of the OEC structures obtained by XFEL and SR.
figure 2

a, Superposition of three OEC structures, two from XFEL and one from SR. The OEC is depicted as in Fig. 1a. b, Superposition of the OEC and its ligand environment from the three structures. For clarity, the structures are rotated relative to that shown in a. Additional water molecules close to the OEC are also shown as spheres. c, An enlarged view of the region boxed by red dashed lines in b. In all three panels, the structure from XFEL data set 1 is coloured by atomic element as in Fig. 1a. Structures from XFEL data set 2 and SR are coloured in green and purple, respectively.

PowerPoint slide

Full size image

A distinct feature of the SR OEC structure is the unusual position of O5, which binds weakly to Mn1D, Mn3B, Mn4A and Ca with distances ranging from 2.4 to 2.7 Å1. In the XFEL structure, the average bond distances were 2.2 Å for Mn3B–O5, 2.3 Å for Mn4A–O5, 2.7 Å for Mn1D–O5, and 2.5 Å for Ca–O5 (Fig. 1d, Table 1, Extended Data Table 3). Compared with the SR structure, three bond distances from O5 to Mn3B, Mn4A and Ca were shortened by 0.2 Å; in contrast, the distance to Mn1D was elongated by 0.1 Å. The longer Mn4A–O5 distance in the SR structure may imply that Mn4A has received some reduction by the attack of hydrated electrons generated by the radiation damage, leading to a longer distance to O5 in the SR structure; this in turn implies that Mn4A is more prone to radiation damage than Mn1D because of its dangling position which leads to a larger exposure to the solvent (and hence easier access by the free radicals), whereas Mn1D exists in the solid cuboidal part of the structure with less exposure to the solvent. Recent computational studies failed to reproduce the long bond distances from O5 to the Mn ions, and instead resulted in strong binding to either Mn1D or Mn4A (with bond lengths shorter than 1.9 Å), giving rise to either R-type (right-side open) or L-type (left-side open) structures, when O5 is assumed to be in a deprotonated state11,13. Our XFEL structure showed that O5 indeed exists in a position closer to Mn4A than to Mn1D, however, such preference for Mn4A was much less than that shown by typical theoretical analyses, and the unusual nature of the bonds involving O5 is largely preserved. Therefore, we confirmed that the longer distances of μ-oxo bridges from O5 were an intrinsic and remarkable feature of the OEC structure, not an artefact of X-ray reduction. Based on this, we propose that O5 is a hydroxide ion rather than an oxygen dianion (deprotonated oxygen) in the S1 state. Indeed, theoretical calculations based on the assumption that O5 is a hydroxide ion have yielded long bond distances between O5 and nearby Mn ions that are largely comparable with the present XFEL structure11,26.

Another feature of the Mn–O bond distances in the XFEL structure was that all of the oxo-bridges connected with Mn2C were 0.1–0.3 Å shorter than the corresponding distances in the previous SR structure1, which may indicate that Mn2C was also partially photo-reduced in the SR structure. Furthermore, structural differences were also found in the ligand environment around the OEC between the XFEL and SR structures. All of the ligand groups coordinated to the OEC had slightly shorter distances, some of them faced in slightly different directions and some water molecules existed in different positions (Table 1, Fig. 2b and 2c, Extended Data Table 3). In contrast, no apparent differences were observed in the acceptor side (the bound plastoquinone acceptors QA and QB, and the non-haem iron) or the region surrounding YD (D2-Y160) in the D2 subunit.

Based on the current XFEL structure, all of the four Mn ions are six-coordinated (Extended Data Fig. 4a–d). The averaged ligand distances, including those to amino acids, oxo-bridges and terminal water ligands, are 2.1 Å for Mn1D, 2.0 Å for Mn2C, 2.0 Å for Mn3B and 2.1 Å for Mn4A (Extended Data Table 4). Because of the longer pair of Mn1D–O5/OD2 (D1-Asp342) and Mn4A–O5/W1 (one of the two water ligands to Mn4A) distances, the average ligand distances of Mn1D and Mn4A are slightly longer than those of Mn2C and Mn3B. These longer pairs also contributed to the non-uniform distribution of the ligand distances, which is a typical characteristic of Mn ions in the 3+ oxidation state having a Jahn–Teller axis (Extended Data Fig. 5a, b, Extended Data Table 4). Based on the high oxidation scenario—that the oxidation states of the four Mn ions are (III, III, IV, IV) in the S1 state—we propose that the valences of Mn1D, Mn2C, Mn3B and Mn4A can be assigned as III, IV, IV and III, respectively. This assignment is supported by the near-centre localization of the O5 atom, as the III valence of the two Mn ions (Mn1D and Mn4A) connected to the two sides of the O5 atom would lead to a longer and almost equal distance for the two Mn–O bonds as observed in the XFEL structure (Fig. 3). The assignment of Mn1D as in a 3+ valence state is consistent with the proposal from the combined study of multi-frequency electron paramagnetic resonance, 55Mn-electron nuclear double-resonance and theoretical calculations showing that Mn1D is a 3+ ion in the S2 state27.

Figure 3: Possible mechanisms for the oxygen-evolving reaction.
figure 3

O5 in the S1 state is connected to both Mn1D(iii) and Mn4A(iii). Upon transition from S1 to S2, an electron may be removed from Mn4A, leading to a right-side open structure (‘R-type’), or from Mn1D, leading to a left-side open structure (‘L-type’). Both structures will allow insertion of a new water molecule during the subsequent S-state transition, preparing for the transition state immediately before O = O bond (indicated by a red bond in the right-hand structure in the bottom row) formation. Alternatively, the transition state ready for O = O bond formation may be formed by moving either W2 or W3 towards O5 (indicated by orange arrows in the left-hand structure in the bottom row). See text for more details. Colour code is the same as Fig. 1a.

PowerPoint slide

Full size image

The present XFEL structure provides important clues to the mechanism of water oxidation. The unusual properties of O5 and the proposal that it is an OH− ion in the S1 state suggests that it is a water molecule in the S0 state; upon transition from S0 to S1, a proton is released from the O5 site, leading to the OH− species in the S1 state. These inferences strongly suggest that O5 provides one of the substrate water molecules for the O = O bond formation, which is in agreement with recent 17O electron-electron double resonance-detected NMR experiments28. Based on the involvement of O5, formation of the O = O bond in the water oxidation reaction can proceed between O5 and another nearby oxygen atom, which could be a newly inserted water molecule, as proposed in ref. 29 and supported by a recent advanced EPR study30 (Fig. 3). It should be pointed out, however, that the second oxygen atom might instead be provided by one of the already present terminal water ligands, W2 or W3, without imposing the insertion of a new water molecule (Fig. 3), as both W2 and W3 are within hydrogen-bond distances to O5 and therefore may be able to move towards to O5 to form the O = O bond during the S-state transition. Determination of which mechanism is correct will require further XFEL experiments to analyse the intermediate structures at atomic resolution, using the approach established in this study.

Methods

Pulse parameters of SACLA

The pulse parameters of SACLA were as follows: pulse duration, 10 fs or shorter; pulse flux, 2.6 × 1010 photons; photon energy, 10.0 keV; beam size, 4.0 (H) × 2.5 (V) µm2 (FWHM) at a position 20 mm downstream from the focal position of the X-ray beam; pulse frequency, 30 Hz.

Preparation of crystals of T. vulcanus PSII in the S1 state

Highly active dimeric PSII core complexes were purified from the thermophilic cyanobacterium T. vulcanus and crystallized as described previously1,31. The crystals appeared in a few days and reached their maximum size of 1.2 × 0.5 × 0.2 mm within one week. The crystals were stored at 12 °C in the crystallization buffer1 to which glycerol, PEG 1,450 and PEG 5,000 MME were added gradually with an increment of 1% for each of the reagents every 30 min. When the cryoprotectant reagents reached the final concentrations of 25% glycerol, 10% PEG 1,450 and 10% PEG 5,000 MME, the crystals were quickly frozen in a nitrogen stream at 100 K. Purification, crystallization and freezing of the crystals were performed in the dark or under dim green light. Therefore, the crystals were fixed into the dark-stable S1 state.

Evaluation of radiation damage propagation induced by XFEL irradiation

Before image data collection, the distance that radiation damage induced by each XFEL pulse may propagate was evaluated by following the protocol developed by our group using PSII crystals25. In brief, a point on a large PSII crystal was exposed to an XFEL pulse with full intensity, followed by irradiation by weak XFEL pulses with an intensity attenuated by two orders of magnitude applied to various points separated by 1 µm to 50 µm with a step of 1 µm in both vertical and horizontal directions from the originally irradiated point. The numbers of diffraction spots of each still image obtained with the weak XFEL pulses were counted at various distances from the originally irradiated point, and the radiation damage was defined to occur when the number of diffraction spots was lower than that expected from a crystal without prior exposure to the full intensity XFEL pulse. The distance of propagation of radiation damage determined in this way for PSII crystals was very similar to those determined for other protein crystals such as lysozyme and CcO, which is around 10 µm25. No perceptible radiation damage was detected beyond 11 µm from the irradiation point with a full intensity XFEL pulse. Thus we estimated that as long as each irradiation point was spatially separated by 15 µm, they would not be affected by other XFEL pulses and the diffraction images would be essentially identical to the ones from intact PSII crystals. In order to ensure a complete avoidance of radiation damage from adjacent XFEL pulses, we adopted a distance of 50 µm (an additional 35 µm ‘safety margin’ was used) between each irradiation point to collect consecutive still images from the large PSII crystals.

Diffraction experiments

All images were collected at beamline 3 of SACLA using a Rayonix MX225-HE detector. The very strong X-ray pulse of SACLA at its focused position produced physical damage in crystals placed at that focus; some cracked or burst, making data collection difficult. To avoid such damage, the samples were set at a position 20 mm downstream from the focal position of the X-ray beam to reduce the flux density of XFEL pulses irradiated onto the crystals. This basically eliminated physical damage, thereby enabling more efficient data collection.

Still diffraction images were recorded from the large PSII crystals sequentially. Before the irradiation by an XFEL pulse, the crystal was moved by at least 50 µm and rotated by 0.2° (Extended Data Fig. 1a, b). This stepwise rotation was aimed to record consecutive still diffraction images so that they could be processed in the same manner as the conventional oscillation diffraction method. Here, the size of the rotation step should be smaller than one-third of the crystal mosaicity to ensure the consecutiveness of the images based on the studies using crystals of lysozyme and CcO25. As the typical mosaicity of the PSII crystals was 0.60° to 0.80°1, a 0.2° step was adequate.

Processing of the diffraction images

Some of the images recorded had significantly lower or higher intensities than typical images, probably due to fluctuations in the XFEL pulses. This situation was more severe for the first half of the images of data set 1 than for the rest of data set 1 and data set 2 because of a larger variation in the pulse intensities during collection of the first half of the images of data set 1. The intensities of the first half images of data set 1 were therefore scaled for each crystal on the basis of the I0 values from an X-ray beam monitor placed in the optics hutch. The standard deviations of the averaged pixel values in the groups derived from the same crystal decreased by 30% after the scaling process (Extended Data Fig. 6a, b). Among these images, (1) the images that showed characteristics of ‘not single’ or ‘cracked’ crystals, or (2) a block of images that had a significant discontinuity in the course of the crystal rotation that made the data processing as continuous rotational images impossible, were not processed. In these cases, we restarted the diffraction experiments with a new crystal to cover all the unique reflections in the asymmetric unit of the reciprocal space. The images selected based on these criteria were processed with the programs ipmosflm and scala in CCP432. The final images used for structural refinement were selected as follows: we discarded the ‘block’ of processed images whose diffraction limit was lower than 2.5 Å resolution, or the ones that made the statistics worse significantly. Those images were selected one-by-one by eye. Since we collected multiple images at similar rotational angles with multiple crystals, the selection of images with resolution higher than 2.5 Å did not affect the completeness of our data as shown in Extended Data Table 1.

The total dose was calculated with the program RADDOSE33, which showed that the total dose absorbed and photon density were 1.4 MGy and 0.26 × 1010 photons per µm2, respectively, for each irradiation point in a crystal. This total dose and photon density was approximately equal to that each point received for the data set collected previously at the SR source1. The conditions and statistics for diffraction experiments are provided in Extended Data Table 1, which showed that the unit cell constants were slightly different between the XFEL and SR data. This is because, as we have shown previously1,31, post-crystallization treatment is very important to obtain better and isomorphous diffracting PSII crystals. Our crystals are very sensitive to cryo-conditions, and the conditions (that is, the composition of the final cryo-protectant buffer) we used in this work were slightly different from those used in the previous work. We used 25% glycerol, 10% PEG 1,450 and 10% PEG 5,000 MME for the final cryo-protectant solution, whereas 25% glycerol and 20% PEG 3000 were used in the previous cases for the SR data collection. This may be the reason for the unit cell change between the XFEL and SR data.

The R-factor between data set 1 and data set 2 is 0.177. This is larger than the R-factors among sub-data sets obtained from the same data sets, which is typically less than 0.05 based on our studies at synchrotron sources. This is largely due to an insufficient isomorphism among the crystals grown from different batches of samples between data sets 1 and 2, although we cultured, purified and crystallized the sample using the same protocols.

Structural determination

The initial phase information up to 3.0 Å resolution was obtained by the molecular replacement method with the Phaser program in CCP432 using the 1.9 Å resolution structure (PDB ID 3ARC) as a search model, in which all of the cofactors including OEC, amino acid residues coordinating OEC and water molecules were omitted. The phases obtained were then extended and improved using the program DM in CCP432. Because the XFEL structure and the SR structure were essentially equivalent at 3.0 Å resolution, the phases obtained after density modification should not be biased by the SR structure. The model was built manually using COOT34 in the resulting electron density map (Fo × exp (iαDM)). The structural refinement was carried out with Phenix35. OEC was not included in the structural model until the final stage of the structural refinement which is described in the next section.

Refinement of the OEC structure

The positions of four manganese atoms, one calcium atom and ligand residues were clearly visible in the 2mFo − DFc map before placement of any of these atoms (Fig. 1a), whereas the positions of the five oxo-oxygen atoms were clearly identified in the mFo − DFc map after placement of the manganese and calcium atoms as shown in Fig. 1a. When we started to build the OEC model, we placed the Mn4CaO5 cluster manually, and performed the structural refinement with strict restraints for bond distances (Mn–O and Ca–O) that were taken exactly from the SR structure (PDB ID 3ARC). This refinement resulted in a shortening of all Mn–Mn distances compared to the SR structure, and some of the positions of the refined oxo-oxygen atoms using ‘the strict SR structure restraints’ for Mn–O and Ca–O distances did not agree well with the positions identified from the mFo − DFc map calculated before placing any oxo-oxygen atoms. Moreover, the refinement with ‘the strict SR structure restraints’ for Mn–Mn distances but loose restraints for Mn–O and Ca–O led to a significant increase in the residual density in the mFo − DFc map. Therefore, although the overall shape was very similar, the OEC in the XFEL structure was apparently different from that in the SR structure in terms of the positions of the individual atoms. Therefore, the refinements were accomplished with rounds of geometry optimizations in which the optimal geometry for the Mn4CaO5 cluster was determined from the previous refined structure under loose constraints for bond distances of Mn–O and Ca–O. Then the restrained refinement was performed successively using the modified ‘new’ library for the bond distances. This geometry optimization procedure was repeated several times until no further changes in the bond distances were observed, and the correctness of the structural refinements was confirmed by checking the decrease in the residual density of the mFo − DFc maps. In the final refined structure obtained, all of the oxo-oxygen atoms existed in the central position of the density in the mFo − DFc map calculated before placing them, and the residual electron density in the mFo − DFc map was minimized. The average and standard deviation of the distances of Mn–Mn, Mn–Ca, Mn–O and Mn–ligand were calculated from four independent PSII monomers (two dimers in each data set) and are listed in Extended Data Table 3, which shows that they had very small deviations, therefore enabling comparisons of the inter-atomic distances between the XFEL and SR structures.