Introduction

In photosynthesis, sunlight absorbance results in the accumulation of low potential electrons that drive metabolic processes, leading to the storage of energy in chemical bonds1. Attempts to hijack photosynthesis to instead produce valuable fuel products have a rich history2, principally as it pertains to sustainable energy initiatives3,4. Photosynthetic biohybrids are especially attractive due to their ability to couple nature’s highly evolved, fine-tuned light capture and conversion processes with synthetic catalytic functionality. A prime example is the coupling of metal nanoparticle (NP) catalysts to the photosynthetic machinery to produce H2, a clean and renewable energy source2,5,6,7,8,9,10,11,12,13,14. In this approach, the abiotic NP catalyst is physically attached to the protein scaffold of one of the photooxidoreductases involved in the light reactions of photosynthesis, called photosystem I (PSI)9. This allows the low potential electrons generated by PSI to be directly shunted to the NP catalyst to produce H2.

PSI is a large, integral membrane protein complex that coordinates cofactors used for light absorption, energy transfer, and electron transfer. In cyanobacteria, PSI is trimeric, each “monomer” being composed of 10–12 subunits (depending on the organism) that coordinate over 100 cofactors, most of which are chlorophyll molecules used for light-harvesting15. At the center of each PSI monomer are 11 cofactors involved in electron transfer, referred to as the electron transfer chain. These include a pair of chlorophylls near the lumenal side of the complex called P700, four monomeric chlorophylls (A−1A, A−1B, A0A, and A0B)16, two quinone molecules (A1A and A1B)17, and three [4Fe–4S] clusters (FX, FA, and FB) (Supplementary Fig. 1)18. Upon excitation energy arriving at the electron transfer chain, charge separation occurs where an electron is transferred down either of the two pseudo-C2-symmetric branches of electron transfer chain cofactors that converge at FX, and the electron is subsequently transferred to FA and FB19,20, thus stabilizing the charge-separated state. In vivo, this charge separation powers transmembrane electron transfer from water-soluble electron donors plastocyanin or cytochrome c6 in the lumen to water-soluble electron acceptors ferredoxin (Fd) or flavodoxin (Flv) in the stroma21,22,23. PSI is an ideal candidate for photocatalytic H2 production due to: (a) its relatively broad absorbance cross-section, (b) its quantum yield of nearly 1.0, and (c) the long-lived and low potential redox couple of its terminal charge-separated state P700+FB− (~60 ms, −580 mV vs. NHE, pH independent) to drive alternative chemistry (Em = −370  mV for H2 at pH 6.3)9.

In one of the most effective PSI-based biohybrid solar fuel catalysts reported to date, PSI self-assembles with mercaptosuccinic acid-coated crystalline platinum nanoparticles (PtNPs) to catalyze H2 production. It has been suggested that the PtNPs bind near three protein subunits on the stromal surface of PSI that harbor the terminal [4Fe–4S] electron acceptors7. This is supported by electron paramagnetic resonance (EPR) experiments showing PtNPs functionally mimic acceptor protein binding to PSI as the same altered charge separation is clearly observed when PtNP or Flv are bound to PSI7,24. Additionally, EPR studies showed that when PtNPs are bound to PSI, electron transfer to excess Flv in solution is inhibited7. Scanning transmission electron microscopy showed the localization of PtNPs in trimeric arrangements that mimic the trimeric configuration of Flv and Fd11 observed in the cryo-EM and crystal structures of PSI in complex with native electron acceptors Fd or Flv25,26,27,28. Despite these initial characterizations, a thorough understanding of how and where the PtNPs bind to PSI is still unknown.

The lack of structural information significantly hinders a deeper understanding of the interplays of structural factors in photon-driven H2 generation and the nature of PtNP interactions with PSI. To understand the molecular basis for the self-assembling PSI–PtNP photocatalyst, we have determined its molecular structure using single-particle cryogenic electron microscopy (cryo-EM). The structure provides insight into the molecular interactions between the stromal surface of PSI and the PtNPs bound to it. The analysis opens the door for altering these interactions to design similar biohybrid solar fuel systems in the future. This high-resolution structure of a photosynthetic biohybrid provides a direct look at a nano-bio interface in a functional photons-to-fuel system that couples nature’s optimized photochemistry to an abiotic catalyst.

Results

Preparation and Cryo-EM

Mercaptosuccinic acid-stabilized PtNPs of ~1.8 nm in diameter (Supplementary Fig. 2) were synthesized and bound to detergent-solubilized PSI complexes isolated from Synechococcus lividus PCC 6717 (hereafter S. lividus) as described in the “Methods” section (Supplementary Figs. 3 and 4). The typical mixed chlorophyll spectrum observed for PSI is maintained after PtNP binding (Supplementary Fig. 3). Photocatalytic H2 production was detected at a rate of 6070 mol H2 mol PSI−1 h−1 and turnover number of 36,600 in 400 min (Supplementary Fig. 5). The PSI–PtNPs were plunge-frozen and screened for cryo-EM (Supplementary Fig. 6). A data set was collected for single particle analysis without imposing symmetry that resulted in a structure of global resolution 2.27 Å (Supplementary Figs. 7,  8, Supplementary Table 1, and Supplementary Movie 1). Example map regions are shown in Supplementary Fig. 9. The map revealed a trimeric PSI structure consistent with most other cyanobacterial PSI structures15. All subunits identified are shown in Supplementary Fig. 10. Because there is presently no structural data on PSI from S. lividus, we performed comparisons of sequence identities (Supplementary Fig. 11), analyzed sequence alignments (Supplementary Figs. 12 and 13), and generated Cα superpositions with subunits from other PSI structures (Supplementary Table 2). The data showed that S. lividus PSI is highly similar to the well-characterized Thermosynechococcus vestitus PSI.

PtNP-binding sites

In addition to the signal corresponding to PSI in the cryo-EM map, two high-signal ellipsoidal map regions were identified per PSI monomer, which were assigned as PtNP-binding sites (Fig. 1). Note that signal corresponding to individual, discrete Pt atoms in these sites are not observable. This suggests that although PtNPs bind in these locations, the orientation and exact positions are variable among the particles that comprise the ensemble cryo-EM data. These sites are hereafter referred to as NP1A, NP1B (the two PtNPs bound to monomer 1), NP2A, NP2B (the two PtNPs bound to monomer 2), NP3A, and NP3B (the two PtNPs bound to monomer 3) (Fig. 1a). The PtNP map signals are located on the stromal surface of PSI nearby subunits PsaA, PsaB, PsaE, PsaF, PsaJ, and PsaX (Fig. 1b and c). When the map is viewed at a low threshold, the PtNP signals also appear to interact with the detergent micelle (Fig. 1d and f). In each monomer, the closest PtNP to a [4Fe–4S] cluster is between the NP in site A and FB, ~14 Å (Supplementary Table 3). Because these distances are very similar, we focus hereafter on monomer 1.

Fig. 1: Locations of bound PtNPs to each PSI monomer.
figure 1

a Two views of the PSI trimer showing the map corresponding to PtNPs (gray transparent map). There are two sites on each PSI monomer, referred to as sites A (close to the [4Fe–4S] cluster FB) and B (further from FB). FB, from which electron transfer most likely proceeds, is shown as yellow glowing spheres. b Close-up of the PtNP-binding sites on monomer 1. c Protein subunits near the PtNP-binding sites. Dashed lines correspond to the PtNP-binding sites in panel b. d Signal of the PtNP in site A of monomer 1 (“NP1A”) at 2.5σ. Note the possible interaction with the nearby detergent micelle based on its signal in the map. e Signal of NP1A at 4.9σ. In d and e, the dimensions of the signal at that threshold are labeled, and the shortest distance to FB (including the coordinating Cys thiolate) is shown. f Signal of the PtNP in site B of monomer 1 (“NP1B”) at 2.5σ. Like site A, note the possible interaction with the nearby detergent micelle based on its signal in the map. g Signal of NP1B at 4.9σ. In panels f and g, the dimensions of the signal at that threshold are labeled, and the shortest distance to FB is shown.

Full size image

For each pair of PtNP map signals (one pair per PSI monomer), site B is smaller and more spherical than the PtNP signal in site A (Fig. 1d–g and Table 1). To visualize a PtNP in the corresponding map signal, we generated a geometry-optimized PtNP model (Supplementary Data 1) using density functional-based tight binding (DFTB) as described in the “Methods” section. A single PtNP of 405 atoms is ~22 Å in diameter, which is consistent with the size distribution maximum determined from STEM HAADF imaging measurements (Supplementary Fig. 2). When the map is viewed at 2.5σ, a single PtNP model fits nicely in the smaller and more spherical B sites, but in the elongated and larger A sites, two PtNP models fit that overlap somewhat (Fig. 2). This suggests higher positional specificity in site B, but lower in site A. In site A, it seems likely that in the ensemble cryo-EM data, particles are bound at various locations along the continuum described by the elongated map region, the two extremes of which are estimated by the position of two particles in that site (Fig. 2). Importantly, however, only PtNPs bound nearest to FB (green PtNP in Fig. 2), at distances in the range of ~14 Å (Supplementary Table 3), would allow efficient electron transfer to occur to drive light-driven H2 production29. When the PtNP is modeled in the more peripheral position of site A, the nearest distance from that to a [4Fe–4S] cluster (FB) is ~30 Å, which is too long for efficient electron transfer. It should be noted that these observations seem reasonable based on the fitting of our spherical 405-atom PtNP model, but this does not consider other particle sizes (Supplementary Fig. 2) or possible non-spherical PtNPs (e.g., perhaps two smaller PtNPs could bind in site A).

Table 1 Approximate areas of PtNP signals
Full size table
Fig. 2: Fitting of PtNP models into the cryo-EM map and most likely PtNP position to accept electrons from the PSI electron transfer chain.
figure 2

A peripheral view of a PSI monomer is shown with selected subunits labeled. NP1A was fit with two PtNP models to show possible alternate positions, and NP1B was fit with one PtNP model. The PtNP model closest to a [4Fe–4S] cluster is shown in green and corresponds to the likely PtNP position to which electrons can be transferred from FB.

Full size image

To gain a better understanding of what interactions allow for the binding of PtNPs to PSI, we generated a surface electrostatics map (Fig. 3a and b) and analyzed what residues are nearest the PtNP map signals (Fig. 3c–h). The electrostatic surface suggests that the regions binding PtNPs are slightly positively charged, which is perhaps unsurprising because the PtNPs should be coated in negatively charged mercaptosuccinic acid. However, we cannot account for whether there is full coverage of mercaptosuccinic acid on the PtNPs, nor how the mercaptosuccinic acid might rearrange upon PtNP-binding to PSI. Furthermore, the sidechains of amino acids that potentially interact with the PtNPs are not well resolved. It should also be noted that surface residues are often the most flexible and, therefore, difficult to precisely position in a cryo-EM map. Coupled with the lower local resolution at the periphery of the structure, and with the probable positional heterogeneity of the PtNPs, it is therefore unsurprising that interactions between amino acids and PtNPs are challenging to determine. That said, numerous residues from subunits PsaA, PsaE, and PsaJ are close enough to the PtNP in site A to propose specific interactions (Fig. 3c). Those residues are the sidechains of PsaA-Glu26, Lys30, Arg36, PsaE-Arg4, Arg39, PsaJ-Lys4 and His5, and the backbone amide nitrogen of PsaJ-Met1 (Fig. 3d–g). Of those residues, the closest to FB are PsaE-Arg4 and Arg39 (Fig. 3f), so when the PtNP in site A occupies a position able to accept electrons from the electron transfer chain (green nanoparticle model in Fig. 2), these Arg sidechains probably provide important charge–charge interactions with mercaptosuccinic acid on the surface of the PtNP. Interestingly, the PsaE residues analogous to these in structures with the native acceptors bound form molecular interactions between the PSI surface and Fd or Flv. Furthermore, it was shown that the mutation of PsaE-Arg39 strongly influences PSI-Fd dissociation30. Thus, PsaE-Arg4 and Arg39 are probably important for binding both the negatively charged PtNPs and native acceptors, alike.

Fig. 3: Surface electrostatics and possible residues that interact with PtNPs.
figure 3

a Surface electrostatics of the S. lividus PSI structure. b Magnified view of the PtNP-binding sites in monomer 1. c Same view as panel b, but instead of surface electrostatics, it shows a cartoon representation of the protein and the features of the PSI structure that appear to interact with the PtNPs (yellow glow). Note that the FB cluster is also labeled. In panels a–c, the dashed circles correspond to the PtNP-binding sites. d and e PsaA residues nearby the PtNP in site 1A. f PsaE residues nearby the PtNP in site 1A. g PsaJ residues nearby the PtNP in site 1A. h PsaX and PsaF residues, and chlorophylls (abbreviated Chl in the figure) nearby the PtNP in site 1B and unidentified (“UI”) map regions. The observations depicted in panels b–h are shown for monomer 1 only, but these are also consistent with monomers 2 and 3.

Full size image

The residues that may interact with the PtNP in site B are less clear but have a relatively positive charge as in site A. The only residue clearly nearby that site is PsaX-Thr9. Specifically, its backbone amide nitrogen is directed toward the NP. PsaX residues 1–8 could not be clearly identified in the cryo-EM map, and therefore, it is not known whether they interact with the PtNP in site B. Notably, a similar number of PsaX residues also are not modeled in numerous other cyanobacterial PSI structures from different species, so this region is probably highly disordered even without a bound NP. Although few modeled residues are observed to be interacting with the PtNP in site B, there are two unmodeled map regions containing unidentified signals (labeled UIA and UIB in Fig. 3g) that appear to bridge interactions from the protein to the nanoparticle. These map regions interact with the backbone carbonyl oxygen atoms of PsaF-Leu147, Ala148, and Ser149, the sidechain of PsaF-Thr153, the sidechain of PsaX-Tyr10, and two chlorophyll molecules in sites B27 and B28 (chlorophyll nomenclature based on that assigned from the first high-resolution structure of PSI18). The most likely candidates for these unidentified signals are lipids or detergents. Also, note that both PtNP sites exhibit signals near the disordered detergent micelle comprising n-dodecyl β-d-maltoside (β-DDM) molecules (Fig. 1d–g). If the unidentified signals correspond to polar detergent headgroups, they are probably involved in H-bonding interactions with the PtNP in site B. Note, too, that steric interactions may provide binding site specificity: the NPs reside in depressions in the protein surface (Fig. 3a, b).

Comparison of PtNP-binding sites with those of the native electron acceptor proteins

On the stromal side of PSI, three extrinsic subunits called PsaC, PsaD, and PsaE form a stromal ridge (Supplementary Figs. 1 and 14) that creates a positively charged cavity to which negatively charged regions of electron acceptors Fd or Flv bind. It has been shown that PtNPs bound to PSI block electron transfer to Flv7. In addition, the binding of PtNPs to PSI mimics the effect that bound Flv has on low-temperature PSI electron transfer. Time-resolved high-frequency EPR studies show that both symmetric branches of cofactors, the A branch and the B branch, are open to low-temperature electron transfer when either Flv or a PtNP is bound to PSI.7 Furthermore, spectral analysis showed nearly 100% quantum yield for the interprotein electron transfer reaction in the PSI–Flv complex24. It was hypothesized that the enhanced electron flow induced by Flv binding to PSI could explain the high rate of H2 production if PtNPs bind to PSI in a similar manner7. To determine the basis for these observations, we superimposed PSI structures with Fd bound (PDB 7FIX26) or Flv bound (PDB 6KIF25) onto the S. lividus PSI structure and generated a low-resolution map of the Fd or Flv, comparing those sites to where the PtNPs are bound (Fig. 4). The PtNP in site A overlaps strongly with both the Fd- and Flv-binding sites. Thus, although the PtNP (~20 Å in diameter) does not bind within the stromal ridge as do the similarly sized Fd (~30 Å in diameter) and Flv (~35 Å in diameter), the stromal ridge is blocked by the PtNP in site A, providing a definitive structural basis for the observation of blocked electron transfer to Flv in EPR studies as well as revealing a new binding position that can readily accept light-generated electrons from PSI for photocatalysis.

Fig. 4: Overlap of PtNP-binding site A with ferredoxin- and flavodoxin-binding sites.
figure 4

a Comparison of Fd-binding site (blue) with PtNP-binding sites (gray). b Comparison of Flv-binding site (blue) with PtNP-binding sites (gray). A stromal side view of the PSI complex is shown. The Fd and Flv maps were created by superimposing a structure of cyanobacterial PSI with Fd bound (PDB 7FIX) or Flv bound (PDB 6KIF) with the structure of PSI with PtNPs bound (reported here), and an 8 Å map of Fd or Flv was generated. The overlapping regions of the PtNP-binding site A and the known Fd/Flv-binding sites are marked with a red X, showing how PtNP binding in that site prohibits Fd/Flv-binding in the PtNP–PSI biohybrid complexes.

Full size image

Discussion

The self-assembling strategy for PSI–PtNP complexation was designed to mimic native acceptor protein binding, where the NPs are negatively charged and of a size similar to Fd and Flv, ~2–3 nm. The initial hypothesis was that the NPs would, therefore, bind within the stromal ridge like Fd and Flv (Supplementary Fig. 14), positioning them to readily receive the light-generated electrons from PSI. Functional studies with EPR spectroscopy and the observed high photocatalytic rates achieved supported this premise7. The high-resolution structure presented herein, however, shows that two PtNPs can bind per PSI monomer, and although one PtNP overlaps the Fd- and Flv-binding sites, they do not exactly bind within the stromal ridge. Despite this, the structure shows that PSI’s light-generated electrons can still be used to drive catalysis from a site not previously reported.

Based on our structure, the binding of PtNPs does not appear to influence the integrity of PSI, i.e., the PtNP does not disrupt or displace any subunits or cofactors. The blockage of electron donors after NP binding7 implies that the PtNPs bind with higher affinity than native electron acceptors, which is reasonable because the native electron acceptors must diffuse readily to drive downstream metabolic processes. Based on our structural analysis, the molecular interactions between PSI and at least the PtNP in site A appear to be primarily charge–charge interactions but may also include polar interactions. It is, therefore, surprising that other sites on the surfaces of PSI do not bind PtNPs with high occupancy, such as the positively charged regions on the stromal side near monomer–monomer interfaces (Fig. 3a). On the other hand, close inspection of the PSI–PtNPs in the cryo-EM micrograph images suggest significant heterogeneity in PtNP-binding sites (Supplementary Fig. 15).

A challenge in data interpretation herein is that the process of particle classification and reconstruction in the cryo-EM data processing workflow reveals only the most common PtNP binding sites in the sample. Fractional occupancies of PtNPs seem likely because despite the 405-atom PtNP model fitting reasonably well into the map signals and that size being consistent with STEM HAADF imaging (Supplementary Fig. 2), we also measured only 400 ± 28 Pt atoms per PSI monomer (see the “Methods” section) directly with ICP-AES analysis. This suggests that about one PtNP is bound per PSI monomer on average despite two PtNP signals being present per monomer in the cryo-EM data. In other words, each PtNP site may be occupied ~50%. This hypothesis is supported by HAADF STEM imaging in which trigonal and linear arrangements of PtNPs corresponding to one PtNP bound per monomer of PSI were observed for the biohybrid11. It is possible that PSI complexes are present with PtNPs bound in other locations not represented in the final cryo-EM maps (Supplementary Fig. 15), which would probably have very low occupancy. Further insight could be provided in the future by more advanced cryo-EM data processing techniques that are specifically designed to analyze heterogeneity. Importantly, the revelation of two PtNP-binding sites per monomer in the cryo-EM structure prioritizes optimizing PtNP-binding site occupancy in further biohybrid development strategies.

The structural resolution of a biohybrid complex reveals important protein features that can be genetically targeted to build optimized systems for solar-driven H2 production. We note that the only high occupancy site that probably does not interact with the detergent micelle is that which is nearest, at 14 Å distance, to the PSI electron transfer chain (green site in Fig. 2). It is of interest to perform future mutagenesis experiments that probe the roles of specific amino acids in the binding of PtNPs. It would be especially interesting to further enhance the specificity of PtNPs for binding to the productive site by mutating surface residues on PSI to include additional positively charged sidechains to those already observed in the native system (e.g., Fig. 3e). There is presently no genetic system established for S. lividus but based on the high similarity of S. lividus PSI subunits to those from T. vestitus PSI (Supplementary Figs. 11–13), we would expect that PtNPs would readily self-assemble in a similar fashion to the latter, which does have an established genetic system. This engineering of PtNP specificity could lead to increased rates of H2 photocatalysis.

Alternatively, synthetic and materials chemistry approaches provide opportunities to chemically tune NP size, surface chemistries, and compositions which could enhance PSI–NP interactions and catalytic efficiency31,32. The importance of synthetic tuning on biohybrid formation has been demonstrated in complexes that link light-activated nanomaterials to biological enzyme-driven catalysis33,34. NPs, with excited state lifetimes typically in the tens of nanoseconds to microsecond range35 and considerably shorter than the millisecond charge-separated state lifetime of PSI, have been successfully complexed with hydrogenases for H2 photoreduction36,37,38,39, carbon monoxide dehydrogenase40 and formate dehydrogenase41 for CO2 photoreduction, and nitrogenase for photocatalytic reduction of N2 to ammonia42. Most of these catalytic NP-enzyme systems rely on self-assembly and non-covalent interactions similar to PSI–PtNP, yet none have been structurally characterized. Thus, this direct look at a nano-bio interface that mediates electron transfer informs on these other systems and, importantly, shows that the nanoparticle does not perturb the biological system. The structural discernment of bioenergy at the atomic level increases our understanding of fundamental structure–function relationships in a photosynthetic biohybrid system and highlights the opportunities for future designs that interface nature’s evolved protein architecture with synthetic materials in solar fuels catalysis.

Methods

Thylakoid preparation

Thylakoids were prepared from S. lividus cells grown in AC medium at 48 °C under fluorescent lighting. Cells (10 g) were washed and resuspended in 30 mM Tricine–NaOH pH 8.0, 300 mM sucrose, and 15 mM NaCl. The cell resuspension was placed in a pre-chilled Bead-Beater (BioSpec Products, Inc.) with 0.5 mm glass beads. The sample was beaten for 5 × 1 min bursts, with 5 min cooling in between runs with a surrounding ice bath. The solution was decanted and spun at 599 × g for 2 min in a Beckman Coulter Avanti JLA 16.25 rotor at 4 °C. The supernatant was discarded. The pellet was resuspended in 30 mM Tricine pH 8.0, 300 mM sucrose, and 150 mM NaCl and incubated on ice for 30 min. The resuspension was then pelleted by ultracentrifugation for 2 h at 203,529 × g in a Beckman L-60 ultracentrifuge with a 60 Ti rotor.

Photosystem I purification

PSI was isolated from S. lividus thylakoids by a slight modification of reported procedures43. Thylakoid membranes (1.0 mg chlorophyll mL−1) were solubilized with 1.5% β-DDM (Affymetrix) for 30 min on ice (β-DDM to chlorophyll ratio of 15:1). Insoluble material was removed by centrifugation (60 Ti rotor, 203,529 × g, 30 min). Solubilized protein complexes were loaded onto a Toyopearl DEAE 650-C column. The column was washed with low salt and protein eluted with a 15–350 mM NaCl gradient in buffer containing 30 mM Tricine–NaOH, pH 8.0, 0.2% β-DDM. The dark green fraction was collected and concentrated using a 50 kDa MWCO microconcentrator (Millipore) and loaded onto a 15–35% sucrose density gradient prepared in 20 mM Tricine–NaOH, pH 8.0, 75 mM NaCl, 0.05% β-DDM. The dominant trimer green band was collected after ultracentrifugation (141,000 × g for 16 h). The sample was further concentrated using a 100-kDa MWCO microconcentrator and used for further analysis.

Photosystem I–PtNP biohybrid preparation

The synthesis of mercaptosuccinic acid-stabilized PtNPs was carried out according to literature procedures44. Scanning transmission electron microscopy (STEM) characterization confirmed the PtNPs to be crystalline, with a size distribution of 1.8–2.0 nm11. Mercaptosuccinic acid-stabilized PtNPs (~7.2 µM for diameter = 2.0 nm PtNPs) were added to the purified PSI (~3 µM monomer) in a volume/volume ratio of 2:1 with an estimated 4.6 mol equivalents of PtNPs to one mole PSI monomer in solution in buffer containing 20 mM Tricine (pH 8.0) and 0.04% β-DDM7. Note that this ratio should be considered a rough estimate because PtNPs exhibit a heterogeneous size population, and the estimated concentration of PtNPs in solution is dependent on the NP size. The mixture was tumbled overnight in the dark at 4 °C. Unbound PtNPs were removed from protein-bound nanoparticles by microfiltration (100 kDa MWCO), and the filtrate was a light-brown color of the PtNPs. PSI–PtNP complexes were washed multiple times by repeated resuspension/concentration steps using microfiltration (100 kDa MWCO) with buffer containing 20 mM Tricine (pH 8.0) and 0.03% β-DDM until the filtrate was clear. Protein concentration was determined based on chlorophyll content in 100% methanol by measuring the absorbance at 665 nm45. Inductively coupled plasma-atomic emission spectroscopy (ICP-AES) on a ThermoScientific iCAP 6000 spectrometer was used to determine the Pt and Fe content in the PSI–PtNP complex. The PSI protein concentration was determined by Fe analysis assuming 12 Fe per PSI monomer. Based on this, an average of 400 ± 28 Pt atoms per PSI monomer was determined for the PSI–PtNP sample used for cryo-EM.

Photocatalytic hydrogen generation

Photocatalytic H2 production was initiated using a white light LED (Solis-3C, Thorlabs) and performed as previously described7 utilizing 33 nM PSI monomer in 10 mM MES, pH 6.03 buffer with 10 µM cytochrome c6, 100 mM sodium ascorbate and 0.03% β-DDM in a total sample volume of 3.0 mL. A Varian CP-4900A micro gas chromatograph with a 10 m, 5 Å molecular sieves column with a thermal conductivity detector and UHP N2 carrier gas was used to detect H2 from samples of the headspace. H2 calibration curves were constructed using injections of 3% H2 in N2 as a known standard.

Cryo-EM grid preparation and data collection

3 µL of the PSI–PtNP preparation at ~1 mg Chl mL–1 was applied to a gold Quantifoil 2/1 300-mesh electron microscopy grid (Electron Microscopy Sciences) that had been glow-discharged for 30 s at 25 mA. The grid was blotted and plunged frozen into liquid ethane using a Thermo Fisher Vitrobot Mark IV system at 4 °C and 100% humidity with a blot force of zero. Grids were transferred to liquid nitrogen until imaging. Initial cryo-EM screening was performed on a Glacios transmission electron microscope (Thermo Fisher/FEI) (Supplementary Fig. 6). Grids were transferred to a 300 kV Titan Krios G2 transmission electron microscope (Thermo Fisher/FEI) equipped with a Gatan K3 direct electron detector and a Gatan BioQuantum energy filter with a slit width of 15 eV for high-resolution data collection. The data were collected in super-resolution mode. The nominal magnification was ×105,000, the defocus range was set to −0.8 to −2.2 μm, and the super-resolution pixel size was 0.4125 Å. The total dose was 40 e− (Å)−2 delivered over 1.8 s. EPU was used to collect 13,349 micrograph movies with 40 frames per stack.

Cryo-EM data processing

All data processing was performed in Relion 3.146. Micrograph movies were corrected, aligned, and dose-weighted using MotionCor247. The contrast transfer function was estimated using Ctffind-4.1.1348. 837 particles were manually selected to create 2D class templates for autopicking. Autopicking yielded 1,334,173 particles. Two rounds of 2D classification yielded a particle set of 1,220,272 particles. It should be noted that the selection of 2D classes was challenging because particles aligned on the high-signal PtNPs (Supplementary Fig. 7). The 3D classification was run on those particles using an Initial Model generated from the manually selected 837 particles, and one class containing 189,655 particles clearly resembled the PSI–PtNP complex. Rounds of 3D refinement, CTF refinement, and Polishing resulted in a map at 2.27 Å global resolution based on the Gold-standard Fourier shell correlation (0.143) cutoff criterion46,49 (Supplementary Fig. 8). C1 symmetry was used to process the data throughout the workflow because C3 symmetry resulted in a clearly incorrect structure, probably due to the high signal of PtNPs. Class images and processing workflow are shown in Supplementary Fig. 7, and resolution plots, the local resolution map, and angular completeness are shown in Supplementary Fig. 8. Note that in the masked and sharpened map, the PtNPs are masked out. This was to (a) consider only the protein that would result in the best structure of S. lividus PSI possible and (b) avoid oversharpening in the PtNP signal regions that are low resolution.

Model building

An initial model was constructed by creating homology models of each subunit from templates of the T. vestitus PSI structure (PDB 1JB0)18 using SwissModel50. The complete homology model was fit into the PSI–PtNP map using UCSF Chimera51. Manual editing was performed using Coot52. Automated refinement was performed using real_space_refine53 in the Phenix software suite54.

Construction of a PtNP model

A platinum crystal was built using a lattice parameter of a = 3.966 Å and cut into the shape of a Pt nanocrystal consisting of 405 atoms (Pt405), exposing (111), (100), and (110) facets as given in reference55. Using the DFTB+ program (version 22.2)56, the nanoparticle was optimized from this nanocrystalline cutout. Due to the large size of the nanoparticle, the semi-empirical SCC-DFTB57,58 method was employed, as it has recently been found to be the best-performing semi-empirical method for predicting the structure of Pt nanoparticles in a benchmark study by Ricchebuono et al.59. Specialized Slater–Koster parameters for the Pt–Pt interaction were used as developed for the Pt55 nanoparticle by Van den Bossche60 but shown to perform well for larger Pt nanoparticles in the previously mentioned benchmark study59. The geometry was optimized until the gradient was below 10−4Eh/a0, with a tolerance of 10−5Eh for the self-consistent charge iterations. The orbitals were filled using Fermi smearing of a width of 0.043 eV. Implicit water was modeled by using the conductor-like screening model (COSMO)61 as implemented in the DFTB+ program, using a dielectric constant of 80.2 and a density of 1 g/mL.

Reporting summary

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