Complex I is the first enzyme of the respiratory chain and has a central role in cellular energy production, coupling electron transfer between NADH and quinone to proton translocation by an unknown mechanism. Dysfunction of complex I has been implicated in many human neurodegenerative diseases. We have determined the structure of its hydrophilic domain previously. Here, we report the α-helical structure of the membrane domain of complex I from Escherichia coli at 3.9 Å resolution. The antiporter-like subunits NuoL/M/N each contain 14 conserved transmembrane (TM) helices. Two of them are discontinuous, as in some transporters. Unexpectedly, subunit NuoL also contains a 110-Å long amphipathic α-helix, spanning almost the entire length of the domain. Furthermore, we have determined the structure of the entire complex I from Thermus thermophilus at 4.5 Å resolution. The L-shaped assembly consists of the α-helical model for the membrane domain, with 63 TM helices, and the known structure of the hydrophilic domain. The architecture of the complex provides strong clues about the coupling mechanism: the conformational changes at the interface of the two main domains may drive the long amphipathic α-helix of NuoL in a piston-like motion, tilting nearby discontinuous TM helices, resulting in proton translocation.
Complex I (NADH:ubiquinone oxidoreductase) is one of the largest membrane protein assemblies known and has a central role in energy production by the mitochondrial respiratory chain, providing about 40% of the proton-motive force required for the synthesis of ATP1,2,3,4,5. Many mutations in complex I subunits have been associated with human neurodegenerative diseases5,6. Complex I has also been suggested to be a major source of reactive oxygen species in mitochondria, which can damage mitochondrial DNA and may be one of the causes of Parkinson’s disease7 and ageing8. Mitochondrial complex I consists of 45 subunits with a combined mass of about 980 kDa9. The prokaryotic enzyme is simpler, consisting of 14 ‘core’ subunits conserved from bacteria to humans, with a total mass of about 550 kDa1,2,5. The mitochondrial and bacterial enzymes contain equivalent redox components and have a similar L-shaped structure2,10,11. The hydrophobic arm is embedded in the membrane and the hydrophilic peripheral arm protrudes into the mitochondrial matrix or the bacterial cytoplasm2,5. The high degree of sequence conservation of ‘core’ subunits indicates that complex I mechanism is likely to be the same throughout the species, and so the bacterial enzyme represents a ‘minimal’ model of human complex I.
The complete structure and the coupling mechanism of this large molecular machine are currently unknown. Our crystal structure of the hydrophilic domain (eight subunits, 280 kDa) of complex I from T. thermophilus established the electron transfer pathway from NADH to the primary electron acceptor flavin mononucleotide (FMN) and then through seven conserved iron-sulphur (Fe-S) clusters to the likely quinone binding site (Q-site) at the interface with the membrane domain12,13. The membrane spanning part of the enzyme lacks covalently bound prosthetic groups, but it must contain the proton translocating machinery. Its atomic structure is currently unknown. The three largest hydrophobic subunits of complex I, NuoL/Nqo12, NuoM/Nqo13 and NuoN/Nqo14 (E. coli/T. thermophilus nomenclature) are homologous to each other and to the Na+ or K+/H+ antiporter family (Mrp)14,15, and so are likely to participate in proton translocation. The transfer of two electrons from NADH to quinone is coupled to the translocation of four (current consensus value) protons across the membrane1,2,3,5. Although T. thermophilus and some other bacteria use lower potential menaquinone instead of ubiquinone, it is likely that H+/e− stoichiometry of the enzyme is the same across the species, as discussed previously5. Two possible mechanisms of coupling have been proposed: direct (redox-driven) and indirect (conformation-driven)2,3,5,16.
Structure of the membrane domain
Crystallization of the intact complex I from E. coli has not yet produced high quality crystals, possibly owing to the flexibility of the connection between the two arms of the enzyme17,18. However, we have isolated and crystallized the membrane domain of complex I from E. coli (see Methods). The crystals contain the six subunits, NuoL, M, N, A, J and K (total mass 222 kDa), and lack NuoH (Supplementary Fig. 1a), which dissociates readily from the complex19. After extensive optimization, including post-crystallization dehydration, the crystals diffract to 3.9 Å resolution (Supplementary Table 1). Phasing using Ta6Br12 cluster derivatives and density modification with multi-crystal averaging led to good quality maps showing clear electron density for transmembrane (TM) helices (Supplementary Fig. 2). An α-helical model (Cα trace) was built into this density. It consists of 55 TM helices (Fig. 1), in agreement with most predictions for the NuoL/M/N/A/J/K subunits14,15,19,20,21,22. The extent and the arrangement of the helices in this model are unambiguous, but the directionality and the connectivity are not established because of limited resolution. The domain is curved, if viewed both in plane and perpendicular to the membrane surface (Fig. 1 and Supplementary Fig. 3), consistent with previous electron microscopy studies19,23. It is about 160 Å long and about 40 Å thick, matching the average thickness of a lipid bilayer.
A pattern of 14 TM helices repeats three times in the structure, which allowed us to assign the helices to homologous subunits NuoL, M and N (Supplementary Fig. 4). Not only do all the helices align very well (Fig. 2), but also each subunit contains, in the same positions, two characteristic ‘broken’ helices (shown in red and orange in Fig. 1 and indicated in Fig. 2). Such discontinuous TM helices, interrupted by a loop (extended peptide) in the middle of the lipid bilayer, are thought to be important for the function of many transporters and channels24 (see below). Four helices form a core of each subunit, surrounded by a ring of ten helices (including two discontinuous), tilted up to about 25° in the same direction relative to the centre of the subunit (Fig. 2c). Thus, all the antiporter-like subunits in complex I contain 14 conserved TM helices. This is in contrast to labelling studies, which suggested only 12 conserved TM helices15,25 (probably, the introduction of the label in a sensitive position can disrupt the folding of these subunits). The assignment of the individual subunits within the domain (Fig. 1) is based on fragmentation and electron microscopy studies26,27. Because subunit NuoL (or subunits NuoL and M jointly) can be removed from the otherwise intact complex26,27, NuoL is located at the distal (from the hydrophilic domain) end of the membrane domain, followed by NuoM.
Subunit NuoL contains a carboxy-terminal extension, absent in NuoM and N, with two TM helices predicted at the beginning and the end of the extension (Supplementary Fig. 4). This is consistent with the position of NuoL at the end of the domain, where there is one additional TM helix at the very tip. Thus, this helix can be identified as the fifteenth TM helix of NuoL. In the electron density (Supplementary Fig. 2b), this helix is continued by very strong α-helical density (helix HL in Fig. 1), running along almost the entire length of the domain and ending with another TM helix, which is likely to be the sixteenth and C-terminal TM helix of NuoL. There is a kink in the middle of helix HL (near subunit NuoM) and so it is modelled as discontinuous in that area. However, the electron density does connect two halves of the helix (Supplementary Figs 2 and 6), so this connection is likely to be rigid and we refer to a single helix HL. Consistently, secondary structure predictions strongly indicate (in all species examined by us) a near-continuous α-helix in this part of NuoL (residues 515–590, Supplementary Fig. 4). It is amphipathic, consistent with its position near the expected surface of lipid bilayer (Fig. 1). The number of residues (75) is just sufficient to form an α-helix of the observed length (110 Å). This is an unexpected and unusual feature of subunit NuoL, which is likely to be important for the mechanism, as discussed below. Additionally, such an arrangement anchors NuoL by its last TM helix in the vicinity of the interface with the hydrophilic domain, which may be beneficial for the stability of the complex.
Subunits NuoA/J/K must occupy the TM helices of the model unaccounted for by NuoL/M/N. There are 11 such helices, consistent with topology studies (NuoA22 and NuoK20 are predicted to contain three helices each and NuoJ five21). On the basis of mutagenesis studies it was suggested that subunits NuoA/J/K may form a multi-helix bundle involved in the coupling mechanism28. This is consistent with their joint position here at the interface with the hydrophilic domain.
Structure of the entire complex I
It is clearly important to know exactly how membrane domain structure fits within the entire complex. We purified intact complex I from T. thermophilus (Methods). The preparation shows high activity with decyl-ubiquinone (19 μmoles NADH oxidized per minute per mg of protein), which is more than 90% inhibited by rotenone or piericidin A, indicating that the complex is fully active and intact. We crystallized it in two different crystal forms (P21 and P212121, Methods), containing all 15 known subunits (14 ‘core’ and the frataxin-like subunit Nqo15, specific to T. thermophilus12; Supplementary Fig. 1b). The P21 crystals were of better quality (although pseudo-merohedrally twinned), and after extensive optimization diffracted to 4.5 Å resolution (Supplementary Table 2). The T. thermophilus hydrophilic domain model (PDB 3I9V, ref. 13) and the E. coli membrane domain model were used for molecular replacement. Similar solutions, yielding an L-shaped molecule, were obtained for both crystal forms (Supplementary Fig. 5). The electron density in the P21 crystal form was of sufficient quality to allow us to build an α-helical model for subunit Nqo8/NuoH (Supplementary Fig. 6), which was lacking in the E. coli crystals. It contains eight TM helices (one strongly bent), consistent with predictions and topology studies for most species29. In T. thermophilus, sequence analysis indicates one additional TM helix at the C terminus, but there is no clear evidence for it in the current electron density. The individual subunits of T. thermophilus membrane domain were refined as rigid bodies. No further rebuilding was done in the absence of independent phase information. Thus, the model reflects the overall arrangement of membrane subunits. Compared to the E. coli membrane domain, the T. thermophilus membrane domain is less curved when viewed both in the lipid plane and perpendicular to it (Supplementary Fig. 6).
Thus, the current model of complex I from T. thermophilus (total mass 523 kDa) consists of an atomic model for the hydrophilic domain and an α-helical model for the membrane domain (Fig. 3). In total, the membrane domain contains 63 TM helices, one of the largest hydrophobic protein complexes yet described. As expected from electron microscopy studies10,11, the complex is L-shaped, with the membrane arm about 180-Å long and the peripheral arm extending about 130 Å over the lipid bilayer. Comparison to recent three dimensional electron microscopy reconstructions10 shows that only the mirror image of the electron microscopy density fits well with the crystal structure (Supplementary Fig. 7), indicating that the hand of these electron microscopy reconstructions needs to be reversed.
The overall orientation of the hydrophilic domain relative to the membrane domain is unexpected. Helix H1 from subunit Nqo6, rather than extending over the surface of the membrane domain as suggested previously5, points sideways at about 90° (Fig. 3). The contacts between the two arms of the complex are mediated mostly through subunit Nqo8 of the membrane domain and subunits Nqo4/6 of the hydrophilic domain. One of subunits in the Nqo7, 10, 11 (NuoA/J/K) bundle also contributes by contacting the amino-terminal part of Nqo4. Consistently, Nqo7 can be cross-linked to Nqo4 and Nqo6 (ref. 30). Subunit Nqo8 contains very large cytoplasmic loops with many conserved charged residues, which may form part of the interface with the hydrophilic domain. These loops are not resolved and so the contacts between the two domains do not seem to be extensive in the current model.
The last in the chain of Fe-S clusters in the hydrophilic domain is thought to be the donor of electrons to the quinone5,12. This high-potential cluster N2 is near the cavity between subunits Nqo4 and Nqo6, which forms part of a larger cavity at the interface of these subunits with hydrophobic subunits Nqo8 and the Nqo7, 10, 11 bundle (Fig. 3d). The N-terminal β-sheet of Nqo4 (residues 26–60) forms part of this cavity, consistent with labelling of the corresponding region of bovine 49 kDa subunit with azidoquinazoline31. Subunit Nqo8 was also labelled with quinone-like inhibitors32 and is thought to form a part of quinone-binding pocket along with subunits Nqo4 and Nqo6 (refs 2, 3, 5, 12). Thus, cumulative evidence indicates that this cavity at the interface of subunits Nqo4, Nqo6, Nqo8 and the Nqo7, 10, 11 bundle provides a quinone-binding site. It is a large cavity, up to about 30 Å across, consistent with it providing a large common binding domain, with partially overlapping sites, for various quinone-like inhibitors3,32. A quinone head group, bound in the cavity, can easily reach to within about 14 Å (required for physiological electron transfer33) or to about 12 Å (suggested by N2-semiquinone interactions34) from the edge of cluster N2. This cluster is about 20–25 Å from the expected surface of the lipid bilayer (Fig. 3), thus some movement of the quinone out of the membrane is likely to be required to approach N2, and its extent will depend on the exact location of the bound quinone in the membrane domain. There are several possible sites for accommodating the isoprenoid tail, formed either by subunit Nqo8 alone or at its interface with the Nqo7, 10, 11 bundle. In the latter case quinone will need to move about 20 Å to approach cluster N2, whereas in the former the movement does not need to exceed about 10 Å, which means that the tail of the quinone can still reside mostly within the membrane, as expected for a highly hydrophobic moiety. This is in contrast to suggestions that cluster N2 is 30–60 Å away from the membrane and the quinone moves to it along a hydrophobic “ramp”10. Additional quinone-binding sites in subunits NuoL35 and NuoM36 have been proposed. There is no clear evidence for them in the structure, although it cannot be excluded that the quinone binds at the surface of these subunits.
The overall architecture of complex I has strong implications for the coupling mechanism. All antiporter-like subunits, including Nqo14, are separated from the quinone-binding site and the interface with the hydrophilic domain by a belt of two layers of α-helices in the Nuo7, 10, 11 bundle. This arrangement precludes a direct link between electron transfer and proton translocation via the antiporter-like subunits. Many cross-linking studies have indicated conformational changes in complex I upon reduction23,37,38,39. In our ‘combined’ mechanism proposed previously, in the course of the catalytic cycle, one proton is translocated by ‘direct’ coupling via cluster N2 and its unusual coordinating tandem cysteines and the rest are driven by conformational changes involving the four-helix bundle of Nqo4 and helix H1 of Nqo6, which shift upon reduction13. We can see now that the four-helix bundle is in direct contact with the Nqo8 subunit, and that helix H1 is also likely to contact the longest helix in Nqo8 (TM1, Fig. 3d). Thus, these Nqo4/6 helices are well positioned to transmit conformational changes to Nqo8 and the Nqo7, 10, 11 bundle (cross-links between Nqo7 and Nqo10 change upon reduction37). The six outermost helices of Nqo8 (including TM1) are highly tilted (by up to about 40°) relative to the lipid bilayer normal, which may reflect the role of Nqo8 in conformational coupling. Furthermore, conformational changes in Nqo8 and the Nqo7, 10, 11 bundle may be aided by quinone movement to and from cluster N2 on binding, reduction and release. This would agree with the suggestion that most of the energy from NADH is released only on delivery of its second electron to quinone40.
The unusual, conserved, long amphipathic helix HL of NuoL/Nqo12 (Figs 1 and 3) is well positioned to act as a mechanical link transmitting conformational changes from near the Q-site to the antiporter-like subunits. Helix HL is in direct contact with one of the discontinuous helices in each of the three antiporter-like subunits (red in Fig. 1 and HD1 in Fig. 2). Such helices are thought to be involved in ion translocation by introducing a charge (and some flexibility) into the core of the membrane24. The other three discontinuous helices (orange in Fig. 1 and HD2 in Fig. 2) may not be involved directly in the translocation as they are farther away from helix HL. A possible mechanism (Fig. 4) is that conformational changes in Nqo8 and the Nqo7, 10, 11 bundle affect the nearby C-terminal helix of Nqo12 (Fig. 3) and a tilt of this helix leads to a piston-like motion of helix HL along the membrane domain. This movement synchronously tilts the three nearby discontinuous helices, resulting in proton translocation. From charged residues predicted to be located deep within the membrane, one glutamate and one lysine are conserved in all three antiporter-like subunits and they are essential for energy transduction (Glu 144 and Lys 234 for E. coli NuoM; Supplementary Fig. 4)25,41,42. Glu 144 is located in TM helix 5, for which there are weak α-helical prediction scores in the middle and it may therefore be discontinuous. This glutamate may thus be located in the discontinuous helices contacting helix HL. During turnover, an ionizable residue in such a position, if exposed in the proton channel with suitable environment, could bind and release protons on the opposite sides of the membrane via conformational changes driven by helix HL. Without such a mechanical link, it would be difficult to imagine how conformational changes might propagate over the considerable length of the membrane domain.
A net result of such conformationally driven catalytic cycle would be three translocated protons (one per antiporter-like subunit). The fourth proton can be translocated via cluster N2 and its tandem cysteines13, if an appropriate network of charged residues is formed in the loops between the two main domains (not resolved in the structure). Alternatively, in view of the elevated position of cluster N2 over the membrane (Figs 3 and 4), the translocation of this proton may also be conformation-driven, via subunits Nqo7, 10, 11 (ref. 28) (or Nqo8). Subunit Nqo11 also contains a conserved and essential for energy-transducing activities glutamate within predicted TM helix28,43.
Multi-subunit Mrp antiporters are thought to represent a common ancestor of complex I and membrane-bound hydrogenases15, and the essential Glu 144 is also conserved in these enzymes25,44. The antiporter subunit MrpA, hydrogenase-4 subunit HyfB and subunit NdhF from the chloroplast Ndh complex, related to NuoL, all contain, according to secondary structure predictions, a long cytoplasmic helix between their fifteenth and sixteenth TM helices. Therefore, ion-translocating membrane protein complexes related to complex I may operate using a mechanical link similar to helix HL.
The overall architecture of this large molecular machine is now clear. F-ATPase has been compared to a turbine. In a similar vein, complex I seems to resemble a steam engine, where the energy of the electron transfer is used to move a piston, which then drives, instead of wheels, a set of discontinuous helices. The full mechanistic details remain to be clarified by atomic structures of the membrane domain and the entire complex.
Complex I was purified from E. coli strain BL21 as described previously17. Membrane and hydrophilic domains were separated by treatment with high concentration of Mg2+, and the membrane domain was purified further on a Mono-S HR5/5 column. For crystallization, a protein solution (10–11 mg ml−1 in 10 mM sodium acetate pH 4.8, 50 mM NaCl and 0.05% n-dodecyl-β-maltoside (DDM)) was mixed first with 0.5% n-heptyl-β-glucoside and 2 mg ml−1 lipids and then, in sitting drop or microbatch plates, with 0.1 M sodium acetate pH 4.8, 0.8–1.0 M sodium or potassium formate and 9% PEG4000. The diffraction properties were improved by dehydration of crystals in 40% PEG4000 before cryo-cooling. Phasing was performed with Ta6Br12 cluster derivatives. Electron density was improved by multi-crystal averaging and density modification, using three phased and one unphased non-isomorphous data sets. The α-helical model of the membrane domain was built automatically initially and completed manually.
Intact complex I from T. thermophilus was purified by a series of three ion-exchange columns and a gel-filtration column. For crystallization in P21 space group, complex I (20 mg ml−1 in 20 mM Bis-Tris pH 6.0, 0.002% PMSF, 2 mM CaCl2, 25% glycerol, 100 mM NaCl and about 1% n-tridecyl-β-maltoside) was mixed, in sitting drop plates, with 100 mM Bis-Tris pH 6.0, 100 mM KCl, 100 mM glutaric acid pH 6.0, 16% (w/v) PEG 4000 and 2.04 mM n-octyl-β-maltoside fluorinated. P212121 crystals were grown under broadly similar conditions, except that protein solution was mixed with E. coli polar lipids at 9:1 (w/w) ratio before set-up. Phasing was performed by molecular replacement, using the previously determined structure of the hydrophilic domain of T. thermophilus complex I (refs 12, 13) and the model of the membrane domain of E. coli complex I described here.
E. coli complex I was purified from strain BL21 as described45. The membrane and hydrophilic domains were separated by treatment with a high concentration of Mg2+: purified complex I (2.5 mg ml−1) was incubated in 20 mM Bis-Tris pH 6.0, 5% (v/v) glycerol, 400 mM MgCl2, 12.5 mM NaCl, 0.5 mM CaCl2 and 0.25% (w/v) n-dodecyl-β-maltoside (DDM, Glycon) for 2 h on ice. The protein solution was diluted with buffer EA (20 mM Bis-Tris pH 6.0, 0.05% DDM), loaded onto a Mono-S HR 5/5 column and eluted with linear gradient of buffer EB (1 M NaCl in buffer A). Fractions containing the membrane arm were pooled, diluted fivefold with 10 mM sodium acetate pH 4.8, 50 mM NaCl, 0.05% DDM, concentrated to 10–11 mg ml−1 using 100 kDa MWCO concentrators (Ultracel-100K, Amicon) and used immediately for crystallization trials.
Intact complex I from T. thermophilus was purified by a novel procedure. All steps were carried out at room temperature. Membranes were prepared from 150 g of cells46, resuspended in 50 mM Bis-Tris pH 6.0, 2 mM CaCl2, 100 mM NaCl, 10% (v/v) glycerol, 0.002% (w/v) PMSF, and protease inhibitor cocktail (EDTA-free, Roche), and solubilized for 1 h with 1% (w/v) n-tridecyl-β-maltoside (TDM, Glycon). Solubilized material was clarified by centrifugation (150,000g, 1 h) and filtration (0.45 μm) and passed through a series of three ion-exchange columns (HiLoad 26/10 Q-Sepharose, HiPrep 16/10 ANX FF and Bio-Scale DEAE 20). All ion-exchange columns were equilibrated with buffer TA (20 mM Bis-Tris pH 6.0, 0.002% PMSF, 2 mM CaCl2, 10% glycerol and 0.05% TDM) and the protein was eluted with a linear gradient of buffer TB (1 M NaCl in buffer A). Fractions containing complex I were identified by NADH:FeCy activity, pooled and diluted in buffer TA before applying to the next column. Pooled fractions from the third ion-exchange column were concentrated to about 1 ml using 100 kDa MWCO concentrators and applied to a HiLoad 16/60 Superdex 200 gel-filtration column, equilibrated with GF buffer (buffer TA with 100 mM NaCl). Fractions were pooled on the basis of purity (assessed by SDS–PAGE), concentrated to about 15 mg ml−1 and an additional 15% glycerol was added for storage in liquid nitrogen. NADH:decyl-ubiquinone activity was measured at 50 °C and in the presence of asolectin.
Crystals of the membrane domain of E. coli complex I were grown using sitting drop or microbatch crystallization under paraffin oil. First, 0.5% n-heptyl-β-glucoside and 2 mg ml−1 lipids (1,2-dimyristoleoyl-sn-glycerol-3-phosphocholine:E. coli polar lipids 3:1 (w/w), Avanti Polar Lipids) were added to the protein solution, then it was mixed 1:1 (v/v) with crystallization reagent, comprising 0.1 M sodium acetate pH 4.8, 0.8–1.0 M sodium or potassium formate and 9% (w/v) PEG4000. Crystals grew at 10 °C for between 3 weeks and 2 months and were rod-like (about 30 × 50 × 200–400 μm). Initially the crystals diffracted to about 7 Å resolution. Diffraction was improved by slow exchange of mother liquor (using microdialysis buttons at 4 °C) to 0.1 M sodium acetate, 0.1 M sodium or potassium formate and 40% PEG4000 over a period of 2–4 days, increasing PEG4000 concentration in 5% steps. The simultaneous decrease in salt concentration together with increase in PEG concentration was essential for the success of this dehydration treatment. Crystals were plunge-frozen in liquid nitrogen without additional cryo-protection. For heavy atom derivatization, mother liquor was exchanged to 0.1 M sodium citrate pH 5.2, 15% PEG4000 and crystals were soaked in Ta6Br12 solution overnight at 4 °C. Subsequently PEG4000 concentration was raised to 40% and crystals were plunge-frozen in liquid nitrogen.
Crystals of intact T. thermophilus complex I were grown at 23 °C using sitting drop crystallization. The P21 crystals were grown by mixing complex I (20 mg ml−1) with a crystallization reagent comprising 100 mM Bis-Tris pH 6.0, 16% (w/v) PEG4000, 100 mM KCl, 100 mM glutaric acid pH 6.0 and 2.04 mM n-octyl-β-maltoside (OM) fluorinated at 1:1 ratio. These conditions were obtained after extensive optimization that improved the resolution from 8 Å to about 4.5 Å through use of glutaric acid as a small molecule additive and OM fluorinated as a detergent additive. The P212121 crystals were grown by first incubating complex I with E. coli polar lipids (prepared in 10 mM Bis-Tris pH 6.0 and 2% (w/v) DDM; Avanti Polar Lipids) at 9:1 (w/w) ratio, and then mixing this solution with a crystallization reagent comprising 100 mM Bis-Tris pH 6.0, 18% PEG4000, 100 mM KCl, 100 mM glutaric acid pH 6.0, 1.02 mM OM fluorinated and 10 mM spermidine tetra-hydrochloride at 1:1 ratio. These crystals were not twinned, but diffracted only to 8.5 Å resolution even after optimization. Further post-crystallization treatments did not improve resolution limits for both crystal forms, unlike for the membrane domain crystals. The P21 crystals were rod-like (about 20 × 20 × 200–300 μm) and the P212121 crystals were plate-like (about 30 × 50 × 300–400 μm). Crystals were fully grown within 1 week and were cryo-cooled by plunging into liquid nitrogen, using 25% (v/v) glycerol as the cryoprotectant.
Data collection and processing
Data were collected at 100 K with an ADSC Q210 detector at beamline ID29, with a Marmosaic 225 detector at beamline ID23-1, with a Marmosaic 225 detector at beamline ID23-2 (using helical strategy for collection along the crystal to minimize radiation damage) at the European Synchrotron Radiation Facility (ESRF, Grenoble) and with a Marmosaic 225 detector at the microdiffractometer MD2 of beamline X06SA (using several points along the crystal to minimize radiation damage) at the Swiss Light Source (Villigen). Image data were processed with MOSFLM and SCALA from the CCP4 suite47.
Membrane domain of E. coli complex I crystallized in the orthorhombic space group P212121 with one molecule per asymmetric unit (ASU) (Supplementary Table 1). The crystals were strongly non-isomorphous. The locations of TaBr clusters (six sites), used for phasing, were determined using SHELXD48 on the anomalous data from the Ta absorbance peak. In total, seven different crystals were used for phasing: four native and three TaBr derivatives. First, the best TaBr crystal was used with the most isomorphous native crystal for phasing in SHARP49 and density modification with SOLOMON, PIRATE and DM from the CCP4 suite. A preliminary α-helical structure was built (mostly automatically using the find_helices_strands module of PHENIX50) and used for molecular replacement with PHASER51 on the other five data sets. This allowed us to determine the three most isomorphous native/TaBr pairs for three independent phasing runs in SHARP. The phasing pairs were not isomorphous to each other owing to differences in unit cell dimensions and the position of the molecule in the unit cell. Therefore, DMmulti program from the CCP4 suite was used, with the molecular masks and transformation matrices obtained from molecular replacement solutions. Multi-crystal electron density averaging and modification (with histogram matching and solvent flattening) was performed using three phased data sets and one additional not phased native data set (Supplementary Table 1). This resulted in marked improvement of the electron density (Supplementary Fig. 2), so that the model could be manually updated and completed in O52 and COOT53. In the crystals, the membrane domains are stacked vertically, interacting via hydrophilic surfaces, with no visible ‘side’ contacts (Supplementary Fig. 3), leading to high solvent content (about 73%) and anisotropic diffraction. It is possible that some of side contacts are mediated by lipids, which are necessary for crystallization.
For the intact complex I from T. thermophilus, crystals in the monoclinic P21 space group were obtained under a range of conditions and all were subject to the rare case of pseudo-merohedral twinning54, imitating space group C2221. The twin fraction was high, in the range of 0.46 to 0.49 for several data sets, as estimated in PHENIX using twin refinement target. Two sets of clear molecular replacement solutions (related by the twinning operator) were obtained in PHASER with the T. thermophilus hydrophilic domain (PDB 3I9V, ref. 55) and the E. coli membrane domain model, resulting in two molecules of intact complex per ASU (68% solvent content; Supplementary Fig. 5b). A clear solution for one molecule per ASU (72% solvent content) was obtained in P212121 crystal form (Supplementary Fig. 5c). When P21 and P212121 solutions are aligned by the hydrophilic domains, the membrane domains align as well (Supplementary Fig. 5a), confirming the correctness of molecular replacement solutions.
The P21 solutions were then further processed in PHASER by allowing the membrane domain subunits to refine as independent rigid bodies. This resulted in slight shifts of subunits relative to the E. coli model (Supplementary Fig. 6bc). Consistently, the shifts were similar for both molecules in the ASU. The electron density for the membrane domain improved, including the density for the subunit NuoH/Nqo8, lacking in the E. coli model (Supplementary Fig. 6a). Despite twinning, the density was somewhat improved by solvent flattening and NCS averaging in DM. The α-helical model of subunit NuoH/Nqo8 was built in COOT using the DM modified and the original density (both from PHASER and from PHENIX, calculated allowing for twinning).
This work was funded by the Medical Research Council. We thank the European Synchrotron Radiation Facility and the Swiss Light Source for provision of synchrotron radiation facilities. We are grateful to the staff of beamlines ID23, ID29 (ESRF, Grenoble) and X06SA (Swiss Light Source, Villigen) for assistance. We thank J. E. Walker and A. Leslie for discussions of the manuscript.
This file contains Supplementary Figures 1-7 with legends, Supplementary Tables 1-2 and References.