Cryo-EM structure of the NDH–PSI–LHCI supercomplex from Spinacia oleracea
Main
Photosynthesis is the vital process through which green plants, algae and cyanobacteria harness solar energy to power cellular metabolism and assimilate carbon dioxide. The initial light reactions in the thylakoid membrane convert solar energy into the chemical energy of a phosphate bond in ATP (adenosine triphosphate) and generate NADPH (nicotinamide adenine dinucleotide phosphate). The subsequent dark reactions of the Calvin cycle in the chloroplast stroma fix CO2 and produce organic compounds. The two processes are coordinated by linear or cyclic electron flow. In linear electron flow, electrons are transported through a sequence of redox-active electron transfer complexes, most notably photosystems (PSs) I and II, generating a proton gradient across the thylakoid membrane and producing NADPH. The proton gradient drives ATP production by ATP synthase. ATP and NADPH are produced at a ratio of approximately 1.3 (refs. 1,2). Cyclic electron flow regulates the ATP:NADPH ratio and protects the photosynthetic machinery from stress-induced damage, cycling electrons around PSI3,4. In the thylakoid membranes of plants and cyanobacterial chloroplasts, NADPH dehydrogenase (NDH) is homologous to the mitochondrial NADH:ubiquinone (Q) oxidoreductase (complex I). NDH transfers electrons from PSI to the plastoquinone (PQ) pool and cytochrome b6f complex, pumping protons across the thylakoid membrane. Cyclic electron flow enhances ATP production by adding to the proton-motive force.
The 11 subunits of plant NDH are homologous to the core subunits of complex I5. The sequences of subunits NdhA–NdhG in the NDH membrane arm resemble the proximal (PP) and distal (PD) modules in the membrane arm of mitochondrial complex I (ND1–ND6). Subunits NdhH–NdhK in the peripheral arm of NDH are homologous to the Q reduction module of complex I (NDFUS2, NDFUS3, NDFUS7 and NDFUS8; Extended Data Fig. 1). Unlike complex I, NDH lacks an NADH oxidation (N) center and uses ferredoxin (Fd) as the electron donor instead. Subunits NdhL–NdhO and NdhS–NdhV of plant NDH are specific to oxygenic photosynthesis. Subunits PnsL1–PnsL5 and PnsB1–PnsB5 are chloroplast specific.
The existence of the NDH–PSI–light-harvesting complex I (LHCI) supercomplex in plants has been demonstrated by blue native PAGE and immunoblotting6. Mass spectrometry indicated the subunits that are crucial for supercomplex formation6. Lhca5 and Lhca6 have key roles in supercomplex stability. Electron microscopy (EM)7 and cryo-EM8,9 revealed that one or two copies of PSI are attached to one NDH complex in angiosperms (Extended Data Fig. 2). Recent cryo-EM structures of the NDH–PSI–LHCI supercomplex from Hordeum vulgare8, Arabidopsis thaliana9 and maize9 achieved resolutions of 4.5, 3.9 and 4.4 Å, leaving details of critical subunits, cofactors and indeed the entire peripheral arm largely unresolved (Extended Data Fig. 2b,c).
Here, we present the structure of the NDH–PSI–LHCI supercomplex from Spinacia oleracea at 3–3.3-Å resolution (Extended Data Figs. 3b and 4 and Table 2). We refer to this complex as NDH–PSI–LHCI-2 (Extended Data Fig. 2). Our cryo-EM map reveals detailed subunit interactions at the level of side chains. The peripheral redox arm of NDH is well resolved and contains a density near the entry of the PQ cavity that fits a bound PQ molecule. The diverse roles of NDH in electron transfer and proton translocation become clear when we compare our map to recent high-resolution structures of mitochondrial complex I. Understanding the structure and function of NDH and its interaction with PSI in detail has important implications for photosynthesis in plants and their adaptation to changing environmental conditions.
Results
The NDH–PSI–LHCI-2 supercomplex
Our cryo-EM map of the supercomplex has an average resolution of 3.2 Å (according to local refined map resolutions; Extended Data Fig. 4) but well-ordered regions are locally resolved up to 3 Å (Extended Data Fig. 4). The supercomplex consists of one NDH and one PSI–LHCI unit (Fig. 1a–c), interacting directly at the terminus of the transmembrane module of NDH and Lhca6 of PSI–LHCI-2 (refs. 7,9). Our cryo-EM density map revealed 26 of the 29 NDH subunits10 and all 16 subunits of the PSI–LHCI-2 complex11, including 154 chlorophylls, 37 carotenoids and 46 lipids. NDH is organized into five modules: SubA, SubB, SubE, SubL and SubM. Modules SubA (NdhH–NdhO) and SubE (NdhU) form the peripheral arm. In the SubE module, which is specific to oxygenic photosynthesis, we identified subunit NdhU, while NdhS, NdhT and NdhV12,13,14,15 were absent.
a–c, Lateral (a,b) and top (c) views of the cryo-EM map of the supercomplex (contour level: 0.38). Shades of cyan (module SubA, NdhH–NdhO) and magenta (module SubE, NdhU) indicate the NDH peripheral arm. The membrane arm is drawn in shades of blue (module SubM, NdhA–NdhG). Lumenal SubL subunits (PnsL1–PnsL5) are shown in shades of purple and subunits of the stromal SubB (PnsB1–PnsB5) in shades of gray. PSI–LHC-2 is colored in shades of green. Cryo-EM densities of lipids are salmon (PG), violet (MGDG), pink (SQDG) or green (DGDG). Carotenoids are orange and chlorophylls are bright green. The detergent belt is transparent gray. d, The peripheral arm contains a row of three 4Fe–4S iron–sulfur clusters (N6a, N6b and N2; Fe, orange; S, yellow) closely spaced for direct electron transfer. N2 is next to the PQ cavity that bifurcates into the PQ entry channel and the E-channel. The distance between the putative PQ molecule bound in the entry channel and N2 is too far for direct electron transfer. e, Structure of the PSI cofactors with corresponding map densities (contour level: 0.22). Subunits participating in electron transfer are indicated. PSI contains three 4Fe–4S centers (FB, FA and FX). The protein scaffold holds the special-pair chlorophylls (P700), pheophytins (A1A/B and A0A/B) and PhyQ in two near-symmetrical membrane-spanning branches that converge on Fe–S cluster Fx.
The membrane arm extends from SubB (PnsB1–PnsB5) on the stromal side to SubL (PnsL1–PnsL5) on the lumenal side. These subunits are unique to chloroplast NDH and absent in cyanobacteria10,16,17,18,19,20. On the stromal side, PnsB1 interacts with NdhB, NdhD, NdhH, NdhF, PnsB4 and PnsB5 (Fig. 2a–e). PnsB3 has four conserved cysteine residues (Extended Data Fig. 5d) expected to bind an iron–sulfur cluster21, although we did not find an Fe–S density in this position.
a, Lateral view of the supercomplex. The subunits interacting with PnsB1 are color-coded as in Fig. 1. PnsB1 is shown as a ribbon; other subunits are shown as an outline. The highlighted region illustrates the points of contact between the long N-terminal loop and the subunits indicated. b, Main contact points between the PnsB1 N-terminal loop and the subunits below. Each group of electrostatic interactions is marked with a letter and shown in the corresponding panels (c–e).
The peripheral arm of plant NDH binds three iron–sulfur clusters precisely spaced for efficient electron transfer. NdhI coordinates two 4Fe–4S clusters and NdhK coordinates one (Fig. 1d). PQ binds in the PQ cavity at the junction between the peripheral and membrane arms (Fig. 1d).
Spinach PSI–LHCI-2 closely resembles the Pisum sativum PSI–LHCI complex22,23, consisting of 12 core subunits (PsaA–PsaL) and four LHCI antenna complexes (Lhca1, Lhca3, Lhca4 and Lhca6; Fig. 1b,c). As observed in NDH–PSI–LHCI-2 supercomplexes from other plants8,9, Lhca6 interacts with NDH, replacing Lhca2 (Fig. 1a–c). The PSI core subunits, PsaA and PsaB, form a conserved heterodimer22,23,24, with the reaction center situated at the interface of their C termini. This assembly includes two membrane-spanning branches, A and B, each featuring one chlorophyll P700, two pheophytins (A1A/B and A0A/B) and a phylloquinone (PhyQ) (Fig. 1e). The terminal electron acceptors are three 4Fe–4S clusters (FX, FA and FB), which subsequently reduce Fd. FA and FB are bound to PsaC, while FX is coordinated by four pseudosymmetrical cysteine residues from PsaA and PsaB (Extended Data Fig. 5c). On the stromal side, peripheral subunits PsaC, PsaD and PsaE form the Fd-docking site. PsaG and PsaI are unique to angiosperms and algae25. PsaF, PsaA and PsaB together create a potential docking site for the soluble electron transfer protein plastocyanin26,27.
In our PSI–LHCI-2 density map, we observed 154 chlorophylls, matching the count in PSI–LHCI-2 from A. thaliana but exceeding the 148 chlorophylls in PSI–LHCI-2 from H. vulgare (Supplementary Fig. 1a,c). Additionally, PSI–LHCI-2 from S. oleracea has more chlorophylls than PSI–LHCI-1 from both A. thaliana (152) and barley (148) (Supplementary Fig. 1d,e). This discrepancy is most apparent in subunit Lhca5, which in PSI–LHCI-1 interacts with NDH (chlorophylls labeled a, b and z in Supplementary Fig. 1d,e), suggesting that differences in pigment content reflect the variation in subunit composition and that the presence or absence of particular chlorophylls might be required for energy transfer in PSI–LHCI-1.
NDH–PSI–LHCI-2 interaction
In the spinach supercomplex, Lhca6, NdhF and PnsB5 stabilize the connection between NDH and PSI–LHCI-2 (Fig. 3a,b). Numerous electrostatic, polar and hydrophobic interactions contribute to supercomplex formation and stability, primarily on the stromal side (Fig. 3c–h). Lhca6 interacts with NdhF mainly through its transmembrane helix 2 (TMH2)–TMH3 loop (Fig. 3c–e). One electrostatic interaction was found on the lumenal side between the TMH1–TMH2 loop of Lhca6 and the N-terminal helix of NdhF (Fig. 3f). The extensive N-terminal stromal loop of PnsB5 engages with both PnsB1 (Extended Data Fig. 6a,b) and NdhD (Extended Data Fig. 6a,c,d), passing above NdhF (Extended Data Fig. 6a,c–e) and forming a hook around the N-terminal stromal loop of Lhca6 (Fig. 3g and Extended Data Fig. 6e). A detergent molecule, likely replacing a lipid, appears to reinforce the connection between TMH3 of Lhca6, the N-terminal stromal loop of PnsB5 and the α-helix motif linking TMH1 and TMH2 of NdhF (Fig. 3h). The loop connecting β-strands 4 and 5 of PnsB2 closely approaches the TMH2–TMH3 loop of Lhca6 (Fig. 3c). Contrary to expectations28, we did not observe direct interactions of NDH PnsB2 and PnsB3 with PSI–LHCI-2.
a, Top view of the supercomplex. The subunits connecting the two complexes are shown as ribbons (NdhF, dark blue; PnsB5 dark gray; Lhca6, spring green) while the other subunits are shown as outlines. Key subunits are color-coded as in Fig. 1. b, Highlighted region illustrating the points of subunit contacts. c–h, Interactions in detail.
The NdhU subunit
The local resolution of the peripheral arm of spinach NDH was 3.3 Å, enabling detailed mapping of subunit interactions at the level of individual side chains in the SubA and SubE modules. Biochemical assays12,29,30 and cryo-EM structures of cyanobacterial NDH31,32 both revealed that SubE is necessary for the interaction of the complex with Fd. In addition to the subunits found in cyanobacterial NDH, we discovered a previously unresolved subunit (Fig. 4a and Extended Data Fig. 2) that we identified on the basis of its resolved side-chain sequence (Fig. 4a) as NdhU of the SubE module12. AlphaFold33 models for NDH subunits previously identified by mass spectrometry12,13,14 (Fig. 4b,c) suggested that either NdhT (Fig. 4c, yellow) or NdhU (Fig. 4c, magenta) was compatible with this new EM density (Fig. 4e). Both models featured a J-shaped domain that was visible in the map but side-chain densities of aromatic residues identified the subunit as NdhU unambiguously (Fig. 4d–h).
a, Cryo-EM map of the unidentified subunit. b, List of nuclear-encoded subunits and protein factors with established interactions with NDH observed in the model organism A. thaliana. c, Panel with the computational models of the proteins listed in b. The designated subunits were produced with AlphaFold33 from S. oleracea input sequences. d–h, On the basis of the AlphaFold predictions and the results from Yamamoto et al.12, we attempted to fit both NdhT and NdhU into the cryo-EM map shown in a. Side chains of NdhU (magenta) fit the density map well, firmly establishing the identity of this subunit. The cryo-EM map shown in a,e,g is drawn at a contour level of 0.3. In d,f,h the contour level is 0.45 to show the side chains more clearly.
NdhU has a flexible N-terminal loop and a J-shaped domain composed of four α helices (α1–α4), followed by a short helix (α5) (Extended Data Fig. 7a). An extended loop connects α5 to the transmembrane domain (Extended Data Fig. 7a). NdhU is centrally placed in the SubE module, establishing electrostatic interactions and salt bridges with subunits NdhI, NdhH, NdhK and NdhJ on the stromal side and potentially with NdhA on the lumenal side (Extended Data Fig. 7e). Its long N-terminal loop and J-shaped domain interact with NdhJ forming an aromatic cluster with NdhK, anchoring it to the peripheral arm (Extended Data Fig. 7b,c). Additionally, the loop connecting the soluble part to the NdhU TMH interacts with the surface of the SubA module (Extended Data Fig. 7d,f).
Bound lipids
We modeled a total of 46 lipids into distinct densities in the spinach supercomplex map, of which 23 are in the NDH membrane arm and 23 are in PSI–LHCI-2 (Extended Data Fig. 8a,b). In NDH, we found densities for 14 phosphatidyl glycerol (PG), 5 monogalactosyl diacylglycerol (MGDG) and 3 sulfoquinovosyl diacylglycerol (SQDG) lipids (Extended Data Fig. 8a) but, surprisingly, no digalactosyl diacylglycerol (DGDG). Most lipids are associated with transmembrane subunits (Fig. 5, middle) and many fill gaps between neighboring subunits, acting as a hydrophobic adhesive (Fig. 5a–i). The negatively charged PG head groups connect loops and helices (Fig. 5a–c,g–i). Two adjacent PG molecules link the N-terminal loop of NdhL to NdhA (Fig. 5b), as in cyanobacterial NDH32 (Supplementary table 1a). Nested within the NDH core, we detected three well-defined SQDG molecules at the juncture between NdhB, NdhD and PnsB1 (SQDG 522/B; Fig. 5d), between NdhD and NdhF (SQDG 806/D; Fig. 5e) and between NdhF and PnsB5 (SQDG 1015/F; Fig. 5f). SQDG 1015/F, together with two PGs and a molecule of β-carotene8,32,34 (BCR 1003/D), connect NdhD, NdhF, PnsB5 and PnsB4 (Fig. 5f).
Middle, subunits of the NDH membrane arm in outline, with NdhA, NdhB, NdhD and NdhF color-coded. Stromal and lumenal subunits were removed for clarity. Lipid densities are salmon (PG), violet (MGDG) or pink (SQDG). Detergent (t-PCCαM (A1H1M)) density is turquoise. a–i, Several lipids interact with more than one subunit, presumably enhancing supercomplex stability.
Of the 23 lipids modeled into the PSI–LHCI-2 complex, PG is the most abundant, followed by MGDG (Extended Data Fig. 8b). The lipid distribution within PSI–LHCI-2 is striking. While PG and DGDG are located at the interface between LHCIs and PSI or within the PSI core (Extended Data Fig. 9, middle), MDGD and SQDG are confined to the border between the LHCs and PSI (Extended Data Fig. 9, middle; violet and pink densities respectively). At the interface with Lhca1, Lhca6 and Lhca3 (Extended Data Fig. 9a,c,f,h), a prominent string of lipids connects the LHCI belt to PSI. MGDG in this region is consistent with the crystal structure of P. sativum PSI–LHCI22,23 (Supplementary table 1b). Some PG lipids bind closely to chlorophyll, creating a ligand for the central Mg2+ (Extended Data Fig. 9b–e). The symmetrical arrangement of one PG (PG 1064/a) and one DGDG (DGDG 850/b) in the PSI core is conserved across species (Extended Data Figs. 8b and 9e).
The PQ-binding pocket
At the juncture of the membrane arm and the peripheral arm, the HOLLOW software tool35 revealed a bifurcated cavity. This cavity comprises the PQ entry channel, E-channel and PQ-binding pocket, closely resembling the homologous Q entry channel, E-channel and Q-binding pocket of complex I (Figs. 1d and 6a,b). Within the PQ entry channel, we observed a distinct nonprotein density of approximately 26 Å in length (Fig. 6b, red) in a position that corresponds to the shallow site in mitochondrial complex I36,37. Guided by previously published structures of cyanobacterial NDH32 and complex I36,38, we modeled a PQ molecule in the density (Fig. 6c). The PQ sits at the entrance of the cavity surrounded by several charged residues, interacting with Y259 of NdhA through its carbonyl group (Fig. 6c). The PQ head group is directed toward the N2 Fe–S cluster at an approximate distance of ~24 Å (Fig. 1d), too far for direct electron transfer. Although we cannot rule out t-PCCαM (4-trans-(4-trans-propylcyclohexyl)-cyclohexyl α-maltoside), the detergent used for solubilizing and purifying the supercomplex, the density in the PQ pocket looks too long and bulky for a detergent molecule and PQ fits it almost perfectly (Fig. 6c).
a, Lateral view of the peripheral arm with subunits forming the PQ chamber (NdhH, NdhK and NdhA) shown as cylinders and surrounding subunits in outline. b, Secondary-structure elements defining the PQ pocket. The PQ density is red and the cavity is shown as a gray surface. c, Close-up view of the PQ molecule modeled into the density (red) visualized at a contour level of 0.23. PQ interacts with Y259. d, TMH5–TMH6 loop of NdhA and the density map (light gray) at a contour level of 0.4. Residues making intrasubunit and intersubunit interactions are shown as sticks. e, Salt bridges formed between the NdhC TMH1–TMH2 stromal loop and surrounding subunits. The density of the NdhC TMH1–TMH2 loop is shown in gray at a contour level of 0.4. Backbone carbonyl groups involved in side-chain interactions are highlighted red in d,e.
The PQ-binding pocket is formed by subunits NdhA, NdhH and NdhK (Fig. 6a,b). As in cyanobacterial NDH32, the PQ pocket is lined by the four-helix bundle and the loop connecting the first two β-strands (β1–β2 loop) of NdhH. The highly conserved Y72 residue approaches Fe–S cluster N2 closely (Fig. 6b). The PQ pocket is further lined by the second helix (α2) and the loops α1–α2 and α3–α4 of NdhK and the long loop connecting TMH5 and TMH6 of NdhA. This loop engages in numerous intrasubunit and intersubunit interactions, including three salt bridges between NdhA and NdhH (Fig. 6d). Note that the TMH5–TMH6 loop is not fully resolved in PQ-bound cyanobacterial NDH32.
The special confinements of the PQ pocket are lined with bulky hydrophobic residues (Fig. 6b). In the NdhH β1–β2 loop, the highly conserved residues H23 and H19 are oriented outward (Fig. 6b), leaving the path to Y72 open. The tyrosine residue (Y39) in the NdhK α1–α2 loop is not conserved in complex I but present in both plant and cyanobacterial NDH.
The PQ-binding pocket is formed by subunits NdhC, NdhI, NdhJ, NdhL, NdhM, NdhO and NdhU (Fig. 6a,e), of which NdhC, NdhI, NdHL and NdhU link the membrane arm to the hydrophilic arm of NDH. The long stromal loop connecting the first and second TMHs of NdhC (TMH1–TMH2 loop; Fig. 6e) connects to NdhA and the soluble subunits NdhK, NdhH and NdhM with multiple salt bridges, reinforcing the interactions between the two arms of NDH (Fig. 6e).
The transmembrane proton pathway
The transmembrane module of NDH comprises seven subunits (NdhA–NdhG). Of these, NdhA, NdhC, NdhE and NdhG form the E-channel37 (Fig. 7a,b) and NdhB, NdhD and NdhF correspond to the antiporter-like subunits of complex I38 (Fig. 7a). Together, they constitute the four proton-pumping units known from complex I5,37,39. Each antiporter-like subunit contains two discontinuous helices (TMH7 and TMH12, Fig. 7a) that are integral to the formation of conserved internal cavities within the membrane arm (Fig. 7a, gray surfaces).
a, Side view of NDH. From right to left, NdhA helices and NdhH and NdhK loops contributing to the formation of the PQ-binding pocket (gray surface), NdhC TMH1–TMH2, NdhG TMH3, NdhE TMH2–TMH3 and discontinuous TMHs of NdhB, NdhD and NdhF are shown as solid cylinders. Other subunits are shown as outlines. The 4Fe–4S N2 cluster is indicated. Charged residues of NdhA–NdhG and residues involved in proton translocation that are conserved between spinach NDH and A. thaliana complex I are shown as stick models and are color-coded. Positively charged residues are blue, negatively charged residues are red and key aromatic residues are white. PQ pocket, voids, channels and depressions within NDH that are likely to be relevant for proton translocation are shown as gray surfaces. The light-blue arrow delineates a continuous hydrophilic path from the Q-binding site to NdhF. Stromal half-channels above NdhB, NdhD and NdhF are indicated as dotted cyan arrows and labeled 1–3; the channel between NdhB and NdhD (magenta) is labeled 4. b, Top view of the E-channel with the relative map of voids presumably occupied by water molecules. Key residues are labeled. c, TMH3 of NdhG. Conserved residues in A. thaliana complex I important for proton translocation are shown in stick representation. The cryo-EM map (transparent cyan) is at a contour level of 0.4. d, Top view of the channel entry above NdhB shown as a surface (1 in a). Subunits of to the entry channel are shown as electrostatic potential surfaces (red, negative; blue, positive). The PnsB1 N-terminal loop is shown in cartoon representation. e,g, Side (e) and top (g) views of channel NdhB-D (4 in a). Polar residues defining the putative hydrophilic channel are shown as sticks. In e, the asterisk indicates the constriction point shown in f. f, Close-up view of the constriction in channel NdhB–NdhD. As in e, the asterisk indicates the constriction point. In g, elements from PnsB1, PnsB4 (not visible), PnsB5, NdhB and NdhD form a funnel, depicted as electrostatic potential surfaces (negative, red; positive, blue) above the stromal entry. The subunits involved in forming the funnel are shown in cartoon representation. Charged residues are represented as sticks.
As in cyanobacterial NDH32, we observed that NdhA, NdhH and NdhK subunits enclose a branched cavity (Fig. 7a, gray surface). The primary branch lined by NdhA helices forms the access channel for PQ toward the PQ entry channel and the PQ-binding site in subunits NdhH and NdhK (Fig. 7a). The second branch forms the E-channel37, a passage extending approximately 32 Å through NdhA toward subunits NdhC, NdhG and NdhE in the membrane arm (Fig. 7a,b). The E-channel is defined by multiple glutamate residues (Fig. 7b), most of which are strictly conserved between NDH and complex I (Supplementary Figs. 2–5). Among these residues, the functionally important glutamates of complex I40 E148ND1_At (E174NdhA) and E197ND1_At (E227NhdA) are conserved, while E232ND1_At is substituted by a serine in NDH (S262NdhA; Fig. 7b and Supplementary Fig. 2h, yellow arrows). In the E-channel, protons injected from NdhA initially encounter the NdhG subunit (homologous to ND6 of mitochondrial complex I; Supplementary Fig. 3). In complex I, TMH3 of ND6 forms a π-bulge that opens and closes the aqueous passage through the membrane arm37,41. Our map shows no evidence of a π-bulge in the corresponding helix (Fig. 7c). At this position the E-channel is continuous but the potential proton passage is interrupted at the antiporter-like subunit NdhB (Fig. 7b). As in complex I37,42,43,44, the subunits responsible for proton translocation define a chain of buried, charged residues along the potential proton passage (Fig. 7a, gray surfaces) from the E-channel to the half-channels in NdhB, NdhD and NdhF (Supplementary Fig. 6–8).
We mapped three shallow half-channels at the stromal side above the discontinuous helix TMH7 and the broken helix TMH8 of NdhB, NdhD and NdhF (Fig. 7a; labeled 1, 2 and 3 respectively). A network of conserved charged and polar residues connects the stromal half-channels to the buried half-channels37,42,43,44. The chloroplast subunit PnsB1 sits exactly above the half-channel on the NdhB stromal site (Fig. 7d). The PnsB1 long N-terminal loop rich in charged residues defines part of a funnel leading to the conserved residues E258NdhB, K321NdhB and R332NdhB that are implicated in proton translocation in complex I37,39,42,44. A t-PCCαM detergent molecule (A1H1M 1020/F; Fig. 7d) and one MGDG (MGDG 1016/F) contribute to the formation of the funnel by blocking the lateral exit, interacting with NdhB, NdhE, the transversal helix of NdhF and PnsB1 (Fig. 5c).
We observe a potential proton channel at the interface between NdhB and NdhD, which we refer to as the NdhB–NdhD channel (Fig. 7a,e–g). The NdhB–NdhD channel is lined by TMH12–TMH13 of NdhB and TMH5–TMH6 of NdhD. It is blocked by two negatively charged lipids (Figs. 7e–g and 5e) in positions that are conserved from complex I to NDH (Supplementary Table 1a). This wide tunnel opens toward the stroma through a network of charged residues (Fig. 7e,g) and is closed on the lumenal side by two tryptophan side chains (Fig. 7e). Surface analysis indicates that the subunits above the NdhB–NdhD channel together form a charged funnel, potentially facilitating proton translocation (Fig. 7g). A triad of nonpolar residues shields the passage from the charged residues in the proton translocation pathway (Fig. 7f). It is worth noting that all our maps of NDH indicate nonprotein densities within the membrane arm channels (Extended Data Fig. 10, cyan), which are most likely water molecules.
Discussion
Our analysis of the spinach NDH–PSI–LHCI-2 supercomplex reveals surprisingly close similarities but also some unique features compared to complex I. The map enables a precise identification of NDH modules at the side-chain level, especially in the previously unresolved peripheral arm. SubA extends toward the stroma. While some of its subunits (NdhH–NdhK) share homology with those that form the complex I Q module (Extended Data Fig. 1), others (NdhL–NdhO)16 are confined to organisms involved in oxygenic photosynthesis. Chloroplast-specific subunits SubB and SubL are thought to stabilize NDH, enhancing its activity27,45. At this stage, their precise function in proton translocation and photosynthesis remains unclear, necessitating further research. Previously, the extended N-terminal loop of PnsB1 was assigned to an unknown protein8 (Supplementary Fig. 9a,f). We now find it interacts with stromal-exposed charged residues of NdhB, suggesting a role in proton translocation across the thylakoid membrane (Figs. 2c–e and 7d). Even though we isolated the supercomplex under reducing conditions, some cysteines in subunit PnsB3 formed disulfide bonds (Extended Data Fig. 5a,b,d,e), in contrast to the Arabidopsis9 and barley8 supercomplexes that were purified without reducing agents but show an Fe–S density in this position.
The 4Fe–4S clusters in the peripheral arm of plant NDH resemble those in cyanobacterial NDH31,32,34,46. Notably, their arrangement mirrors that of the clusters N6a, N6b and N2 in the peripheral arm of complex I47,48,49, suggesting a pathway for electron transfer from Fd to PQ (Fig. 1d) and underscoring their functional importance in photosynthetic electron transfer.
Our PSI–LHCI-2 contains the same number of chlorophylls (154) as PSI–LHCI-2 from A. thaliana9 but more than PSI–LHCI-2 from H. vulgare8 (148) (Supplementary Fig. 1a,c). We attribute this difference first and foremost to the substantially higher resolution of our map (EMD-19248, 3.09 Å) compared to barley8 PSI–LHCI-2 (EMD-31350, 3.88 Å). At a resolution of around 3.9 Å, pigment densities are not very clear and can be easily missed or mistaken. Second, both PSI–LHCI-1 and PSI–LHCI-2 complexes from barley8 lack the subunit PsaG (Supplementary Fig. 1c), which, as in our spinach structure and in A. thaliana9, contains several chlorophylls and carotenoids. Certain chlorophylls are observed only in barley8 (x and y in Supplementary Fig. 1a,d). These disparities in the number and position of chlorophylls in barley may be species specific. This may reflect the evolutionary distance between barley, a monocotyledon, and spinach or A. thaliana, which are dicotyledons.
Lhca6 of PSI is known to be required for interaction with NDH6,50. Our study offers clear evidence that Lhca6 interacts with the SubB module of NDH, whereas earlier conclusions28,51 were speculative. Some studies7,8,9 found that NDH can bind a second PSI–LHCI assembly (that is, PSI–LHCI-1), through interactions of the PnsB1 β3–α5 and β5–β6 loops and the Lhca5 TMH2–TMH3 loop (Extended Data Fig. 2b,c and Supplementary Fig. 9b–e). Sequence alignments of spinach PnsB1 and Lhca5 with the corresponding subunits in A. thaliana and H. vulgare showed no notable amino acid differences in the interacting interface, ruling out sequence variations as the cause for the observed 1:1 ratio of NDH to PSI–LHCI (Supplementary Fig. 9g–i). The abundance of interactions between Lhca6 and NDH (Fig. 3c–h) potentially stabilizes the formation of the smaller supercomplex, while transient interactions may favor formation of the larger supercomplex with two PSI assemblies8,9. Our findings support previous studies indicating that particles with a 1:1 NDH-to-PSI–LHCI ratio predominantly preserve PSI–LHCI-2 (refs. 7,8,9). PSI–LHCI-2 refers to the position of PSI relative to NDH as illustrated in Extended Data Fig. 2. The higher stability of the smaller supercomplex may be the result of evolutionary adaptation, suggesting that angiosperms evolved to stabilize the NDH complex by forming the Lhca6-dependent supercomplex instead of producing new NDH complexes in mature leaves28.
Immunoblotting and mass spectrometry consistently detected Lhca6 in the NDH–PSI–LHCI-2 supercomplex6 but not in PSI monomers from various plants11, thus suggesting a key role of Lhca6 in supercomplex formation. However, variations in the number of bound PSI–LHCIs may be influenced by environmental factors or leaf maturity52,53. Further investigation is needed to determine whether the reversible formation of the supercomplex is dependent on environmental stress, light intensity or leaf maturity.
Of the 23 lipids modeled into the PSI–LHCI-2 complex, PG and MGDG have key roles in photosynthetic activity54,55 and LHCI stability56. Depletion of MGDG in A. thaliana seedlings has been shown to disrupt the assembly of the PSI–LHCI complex and the formation of LHCI aggregates57. Collectively, these findings underscore the pivotal role of MGDG in PSI–LHCI complex assembly. The direct interaction of PG with chlorophylls may explain the reduced chlorophyll accumulation in the PG-lacking A. thaliana pgp1–pgp2 mutant54. Moreover, the symmetrical positioning of PG (1064/a) and DGDG (850/b) within the PSI core (Extended Data Fig. 9e) may be important for the assembly of the PSI reaction center, given that PG depletion leads to decreased photosynthetic activity in A. thaliana54,55.
In summary, our detailed insights into NDH–PSI–LHCI-2 supercomplex architecture, subunit interactions and shared characteristics with complex I pave the way for understanding the mechanisms underlying electron transfer and supercomplex formation in photosynthetic organisms. Previously published in vitro assays demonstrated the proton-pumping activity for NDH experimentally4. Furthermore, the presence of charged residues in the core subunits of complex I, known to be vital for proton translocation activity39,58,59, are conserved in NdhA–NdhG of NDH (Supplementary Figs. 2–5). Collectively, our structural findings indicate that chloroplast NDH and mitochondrial complex I function in essentially the same way. This finding holds promise for future advancements in optimizing the photosynthetic process for applications from agriculture to renewable energy.
Methods
Isolation of NDH–PSI–LHCI-2 from spinach leaves
Preparation of thylakoid membranes from young leaves of market spinach (S. oleracea) and membrane protein solubilization were carried out as described previously60. The suspension was diluted 1:1 with suspension buffer (300 mM sorbitol, 150 mM NaCl, 10 mM Tris-HCl pH 8.0, 0.5 mM MgCl2, 2 mM KCl, 2 mM DTT and 2 mM PMSF) and sonicated for 120 s on ice with an ultrasonic cell disruptor (Sonifier s-250d digital, Branson Ultrasonics). Grana stacks were sedimented by centrifugating thylakoid suspension for 15 min at ~12,500g (9,000 rpm; JA14.50, Beckman Coulter). Supernatant containing the stromal lamella fraction was carefully removed and diluted 1:1 with extraction buffer (30 mM HEPES pH 7.8, 2 mM MgCl2, 2 mM KCl, 0.5 mM EDTA, 50 mM NaCl, 1% (w/v) t-PCCαM (Glycon Luckenwalde), 10% glycerol and 2 mM DTT) and incubated for 30 min at 4 °C in dark. Next, the sample was centrifuged for 30 min at 40,000 rpm (Ti45, Beckman Coulter). Supernatant was loaded on a POROS GoPure HQ 50 anion-exchange column (Life Technologies) equilibrated with buffer A (30 mM HEPES pH 7.8, 2 mM MgCl2, 2 mM KCl, 0.5 mM Na2-EDTA, 100 mM NaCl, 2 mM DTT and 0.04% (w/v) t-PCCαM) and gradually eluted with 10–60% buffer B (buffer A with 1 M NaCl) for three column volumes using an Äkta Explorer chromatography system (GE Healthcare) at 4 °C (Supplementary Fig. 10a). Peak fractions were pooled and loaded on a clear native PAGE to confirm protein content (Supplementary Fig. 10b). Fractions containing NDH1–PS1 were concentrated to 0.5 ml and loaded on a 16/300 Superose-6 (GE Healthcare Life Sciences) gel-filtration column equilibrated with buffer A (Supplementary Fig. 11a). Samples were eluted with at 0.5 ml min−1. Each fraction was screened with negative-stain EM in an FEI Tecnai Spirit Bio-Twin transmission electron microscope to identify fractions with highest NDH1–PS1 content (Supplementary Fig. 11b). Fractions eluting at ~9–10 ml were concentrated to 1–2 mg ml−1 and used for cryo-EM sample preparation (Supplementary Fig. 11c). Protein concentration was determined using the BCA assay (Thermo Fisher Scientific, Pierce).
Cryo-EM preparation and EM
First, 3 μl of sample was applied to freshly glow-discharged Quantifoil R1.2/1.3 grids and plunge-frozen in liquid ethane using a Vitrobot (Thermo Fisher Scientific, FEI). Micrographs were recorded in a Titan Krios G2 microscope operated at 300 kV (Thermo Fisher Scientific, FEI) on a K3 direct electron detector in electron counting mode at a pixel size of 0.837 Å, calibrated with apoferritin and an X-ray model as the reference. A total of 55,704 dose-fractionated videos with 40 fractions and a total dose of ~42 e− per A2 in a defocus range of −0.7 to −1.7 μm were recorded using EPU and aberration-free image shift (Thermo Fisher Scientific, FEI). Details are shown in Table 1.
Image processing for single-particle cryo-EM
Images were processed using cryoSPARC60. A detailed description of data processing is shown in Supplementary Figs. 12 and 13. Briefly, videos were motion-corrected and contrast transfer function (CTF) parameters were initially estimated as implemented in cryoSPARC. Particle-picking models were manually built using the Topaz61 wrapper implemented in cryoSPARC and subsequently applied to the datasets. After extraction, particles were subjected to several rounds of two-dimensional classification followed by homogeneous refinement and three-dimensional (3D) classification. The particles from the best classes were merged and, after an initial 3D nonuniform refinement, 3D-refined particles together with a real-space mask around the whole supercomplex were used as inputs for 3D variability analysis62. This yielded an initial set of 107,752 particles. Particle coordinates were exported to RELION63 and underwent 3D refinement, multiple rounds of CTF refinement and Bayesian polishing63. Subsequent 3D refinement was carried out in RELION and cryoSPARC, with cryoSPARC producing slightly better results. Next, these particles were subjected to heterogeneous refinement, isolating a subset of 88,276 particles.
The 3D-refined particles with a real-space mask around the peripheral arm were used as inputs for 3D variability analysis to separate particles that showed a well-defined peripheral arm from those where it was damaged (Supplementary Fig. 13c–e). A final set of 38,385 particles was again subjected to nonuniform refinement in cryoSPARC, resulting in a map of 3.34-Å resolution (Extended Data Fig. 3a–c). Local refinement of the various supercomplex components used different masks, one around the NDH peripheral arm (Extended Data Fig. 4a), one around the NDH membrane arm (Extended Data Fig. 4b), one around the border region between NDH and PSI–LHCI (Extended Data Fig. 4g) and one around the PSI–LHCI complex (Extended Data Fig. 4h). Local refinement with these masks improved the overall resolution of the supercomplex to an average of 3.19 Å.
Real-space masks were automatically generated by cryoSPARC. For local refinement, we created a mask base starting from molecular models using the molmap command in UCSF ChimeraX64. The mask base was then converted into a mask using volume tools in cryoSPARC. In this process, we set the threshold values chosen in UCSF ChimeraX (which varied from mask to mask), the dilation radius to 0 pixels and the soft padding width to 8 pixels. All other parameters were left at their default settings.
The four locally refined maps were used to generate a final composite map in UCSF ChimeraX64. Local resolution estimation was performed in cryoSPARC. All resolutions were estimated according to the Fourier shell correlation (FSC) 0.143 cutoff criterion of two independently refined half-maps65.
Model building and analysis
For the initial construction of the model of S. oleracea NDH–PSI supercomplex we relied on the structural blueprints provided by the pea PSI–LHCI complex crystal structure (Protein Data Bank (PDB) 4Y28)22 and the cyanobacteria NDH1L cryo-EM structure (PDB 6KHJ)32. Predictions for the plant-specific chloroplastic subunits SubB and SubL of spinach, as well as NdhU, were generated with AlphaFold33 (Supplementary Figs. 14–16, left). Protein models predicted by AlphaFold are color-coded according to a per-residue model confidence score called the predicted local distance difference test (pLDDT) (Supplementary Figs. 14a, 15a and 16a). The pLDDT score reflects the predicted accuracy of the model in the LDDT at Cα atoms, which measures the reliability of the structural prediction at a local level. The pLDDT score ranges from 0 to 100. Regions with a pLDDT score below 50 may be unstructured. AlphaFold ‘predict align error’ plots indicate confidence in the relative positioning of two residues, showing domain reliability. Green shades represent the expected distance error (0–31 Å), with the color at (x, y) reflecting the predicted error in residue x’s position when aligned to residue y (Supplementary Figs. 14–16, right).
Subsequently, the models were initially docked into the various locally refined maps using UCSF ChimeraX64. Where necessary, amino acid residues of specific subunits were modified on the basis of the spinach sequence in Coot66. Initial model fitting was conducted using ISOLDE67 and iterative refinement was performed with Coot in conjunction with PHENIX68. Final model fitting was performed using the composite map (Extended Data Fig. 3d; EMD-51527) in Coot in conjunction with PHENIX. The quality of the models was assessed with MolProbity69 (Table 1).
To identify tunnels, cavities and voids in the NDH complex, we used the software tool HOLLOW35 with default values (that is, probe radius set to 1.4 Å and grid spacing set to 0.5 Å). Sequence alignments shown in Supplementary Fig. 9 were performed with ClustalOmega70. Figures showing molecular structures and density maps were created with UCSF ChimeraX64. Graphical elements and icons were created with BioRender.com.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Responses