The haem–copper cytochrome oxidases are terminal catalysts of the respiratory chains in aerobic organisms. These integral membrane protein complexes catalyse the reduction of molecular oxygen to water and utilize the free energy of this reaction to generate a transmembrane proton gradient. Quinol oxidase complexes such as the Escherichia coli cytochrome bo belong to this superfamily. To elucidate the similarities as well as differences between ubiquinol and cytochrome c oxidases, we have analysed two‐dimensional crystals of cytochrome bo by cryo‐electron microscopy. The crystals diffract beyond 5 Å. A projection map was calculated to a resolution of 6 Å. All four subunits can be identified and single α‐helices are resolved within the density for the protein complex. The comparison with the three‐dimensional structure of cytochrome c oxidase shows the clear structural similarity within the common functional core surrounding the metal‐binding sites in subunit I. It also indicates subtle differences which are due to the distinct subunit composition. This study can be extended to a three‐dimensional structure analysis of the quinol oxidase complex by electron image processing of tilted crystals.
Haem–copper oxidases are terminal enzymes in the respiratory chain and catalyse the reduction of molecular oxygen to water by electrons originally deriving from the reduction of NADH or succinate by the electron transfer chain. During this reaction, protons are pumped across the membrane, leading to the generation of a chemiosmotic proton gradient which drives ATP synthesis (Trumpower and Gennis, 1994). The superfamily of haem–copper oxidases is divided into several branches (Castresana et al., 1994; García‐Horsman et al., 1994; Castresana and Saraste, 1995). Mitochondrial cytochrome oxidases and most oxidases from aerobic bacteria use cytochrome c as a substrate, but many bacteria express homologous enzymes that use quinols as electron donor. The eukaryotic cytochrome c oxidase complex contains up to 13 different subunits, three of which are encoded by the mitochondrial DNA. The corresponding bacterial enzymes have a much simpler subunit composition, typically consisting of three or four subunits that always include one or more homologues of the mitochondrially coded eukaryotic proteins (Saraste, 1990).
The Gram‐negative eubacterium Escherichia coli possesses a branched respiratory chain. Depending on the growth conditions, ubiquinol‐8 is oxidized by molecular oxygen either by the cytochrome bd or the cytochrome bo complex. Only the cytochrome bo ubiquinol oxidase (hereafter referred to as Cyt bo) is a member of the haem–copper oxidase family. Neither cytochrome c reductase nor cytochrome c oxidase are present in E.coli (Anraku and Gennis, 1987). Similarly to the mitochondrial cytochrome c oxidase, Cyt bo generates a proton motive force (Kita et al., 1982) by a combination of a scalar mechanism and a proton pump coupled to the redox chemistry (Puustinen et al., 1989).
The genes for Cyt bo are organized in a single operon. Five open reading frames, cyoABCDE, encode the subunits II, I, III and IV of the complex and a protohaem IX farnesyltransferase, respectively (Chepuri et al., 1990; Saiki et al., 1992). Subunits I–III are homologous (10–40% identity) to the major subunits of the mitochondrial and bacterial cytochrome c oxidases (Chepuri et al., 1990; Saraste, 1990), but subunit IV shows no homology to any component of the latter (Chepuri et al., 1990; Haltia et al., 1994).
Cyt bo can be purified as an intact protein complex, containing all four subunits (Matsushita et al., 1984; Georgiou et al., 1988). It contains two haem groups, which are haems B and O (Puustinen and Wikström, 1991). Additionally, about one copper atom is present per complex (Kita et al., 1984). Haem B is in a low‐spin configuration, and the high‐spin haem O is anti‐ferromagnetically coupled to the copper atom, called CuB (Salerno et al., 1990). The iron of haem O and CuB form a binuclear centre in which oxygen is reduced. All metal centres reside in subunit I, CyoB (see Brown et al., 1993; García‐Horsman et al., 1994). Using site‐directed mutagenesis, the axial ligands to the metal centres have been identified as six invariant histidines in this subunit (Minagawa et al., 1992; Lemieux et al., 1992; Calhoun et al., 1993; Uno et al., 1994; Tsubaki et al., 1994).
The three‐dimensional structures are known for the cytochrome oxidases from Paracoccus denitrificans (Iwata et al., 1995) and bovine heart mitochondria (Tsukihara et al., 1995, 1996). The primary electron acceptor site CuA centre is present in the subunit II of cytochrome c oxidases but not in the corresponding subunit of quinol‐oxidizing enzymes. However, it has been possible to engineer this metal centre into subunit II (CyoA) of Cyt bo (van der Oost et al., 1992). The crystal structure of the periplasmic domain of CyoA, with and without the engineered copper site, has been determined at high resolution (Wilmanns et al., 1995). Recent studies indicate that electron transfer from quinol to the low‐spin cytochrome b in subunit I follows a similar path as from CuA to cytochrome a in cytochrome c oxidases (Sato‐Watanabe et al., 1995; Puustinen et al., 1996).
The function of subunit III is unspecified for all haem–copper oxidases. However, subunit III appears to be at least involved in the assembly of the complex (Haltia et al., 1989). It has been suggested to play a role as an oxygen store (Tsukihara et al., 1996), and to be part of the putative oxygen channel that leads to the binuclear centre (Riistama et al., 1996). Likewise, the role of subunit IV is unknown. Preparations of bacterial cytochrome oxidase that consist of only subunits I and II are fully active (Hendler et al., 1991). However, deletion of the gene encoding subunit IV impairs functional expression, probably due to defective assembly of the binuclear centre (Saiki et al., 1996).
Structural studies on two‐dimensional crystals of cytochrome c oxidases have been carried out for many years (see references in Frey, 1994). Three‐dimensional models for the mitochondrial enzyme complex have been obtained at low resolution both in negative stain and amorphous ice (Deatherage et al., 1982; Valpuesta et al., 1990). Until now, structural details of a quinol oxidase complex have not been reported, apart from the three‐dimensional structure of the periplasmic domain of CyoA (Wilmanns et al., 1995). Here, we describe the projection structure of the entire Cyt bo complex determined by cryo‐electron microscopy. The projected density in the membrane plane can be compared easily with the similarly oriented structure of the bacterial cytochrome c oxidase (Iwata et al., 1995).
Crystallization of Cyt bo and image processing
Two‐dimensional crystallization studies on haem–copper oxidases have been carried out with cytochrome c oxidases purified from P.denitrificans and Rhodobacter sphaeroides and Cyt bo from E.coli. These enzymes form membraneous crystals which diffract to a resolution of ∼25 Å in negative stain (Warne et al., 1995). The best results in terms of crystal size and intrinsic order could be obtained with Cyt bo in which subunits II and I are genetically fused.
We have studied large flattened tubular crystals with dimensions of 0.5 μm in width and 2–8 μm in length by cryo‐electron microscopy. Optical diffraction of the micrographs revealed reflections to a resolution of ∼8 Å (Figure 1). After filtering and unbending of the crystal lattice (see Materials and methods), the Fourier transform of a single image could be obtained (Figure 2). Reflections from only one layer of the flattened tubular crystal can be observed from this image. The most likely reason for this on a holey carbon grid is that one of the two layers is disordered. In the air–ice interface this could be caused by a partial denaturation of the protein, while in the layer opposite to the air–water interface disorder could be caused by insufficient flattening of the lattice due to missing surface tension. In contrast, images taken from specimens in thinner ice on continuous carbon films show diffraction from both layers which can be processed separately (data not shown).
In Figure 2, reflections with an IQ value (see legend) of 3 or better are visible to a resolution of 5 Å. Comparison of the phases of all structure factors revealed the planar space group to be p22121. The amplitudes and phases of the image were then averaged with symmetry applied, and amplitudes sharpened by a resolution‐dependent scale factor (see Materials and methods). Table I shows the mean phase error of the averaged data in resolution ranges for 255 unique structure factors. The average error of the phases is less than random to a resolution of 6.0 Å.
Structure of the p22121 crystal form
Figure 3 shows the projection map calculated using data obtained from a single two‐dimensional crystal. Our interpretation is that, as the positive density represents protein mass, the asymmetric unit contains one Cyt bo complex (four asymmetric units per unit cell). The oxidase complexes seem to form dimers related by a crystallographic 2‐fold axis perpendicular to the membrane plane. In the lattice, these dimers are oriented alternatively top‐up and bottom‐up, resulting in screw axes parallel to the membrane plane.
In Figure 4A, the projection map of the monomeric Cyt bo complex is shown. Although the identification and the assignment of secondary structure elements can be difficult in a projection map at medium resolution, the high quality of the data and the comparison with the X‐ray structure of cytochrome c oxidase from P.denitrificans (Figure 4B; Iwata et al., 1995) allows us to interpret the map as follows.
The complex consists of four subunits, namely the gene products of cyoABCD. The largest continuous density area in the map (Figure 4A) originates from subunits I and II (CyoB and CyoA). This density is recognizable by the pseudo 3‐fold symmetry of subunit I (shaded yellow) that is known for cytochrome c oxidases (Iwata et al., 1995; Tsukihara et al., 1996). Within subunit I, several regions with negative density are found. They clearly correspond to the three pores found in cytochrome c oxidase (Iwata et al., 1995). Pore A (bottom left) is blocked by mostly conserved aromatic residues, pore B (right) contains the binuclear centre and is covered by the periplasmic domain of subunit II, and pore C (top left and centre) ends at the cytochrome b centre (cytochrome a in cytochrome c oxidase). Positive density peaks typical for separate α‐helices can hardly be resolved within subunit I, probably due to the fact that almost all α‐helices in this subunit are tilted with respect to the membrane normal (cf. Iwata et al., 1995). The most prominent density that resembles an α‐helix which is perpendicular to the membrane plane, divides pore C into two halves, and includes helix X and haem b. On the top‐right of the molecule, two features are found that resemble a perpendicular and a largely tilted α‐helix, respectively (shaded purple). Their arrangement matches almost exactly that of the two membrane‐spanning helices of subunit II in cytochrome c oxidase (Figure 4B).
Assignment of the density found in the bottom part of the projection map (Figure 4A) is less straightforward. Analysis of the predicted sequences has shown that CyoB possesses one additional helix at the N‐terminus (helix 0) and two additional helices in the C‐terminus (helices XIII and XIV) in comparison with its homologue in cytochrome c oxidase (Chepuri and Gennis, 1990; Chepuri et al., 1990). On the other hand, subunit III of Cyt bo (CyoC) consists of only five helices, whereas this subunit has seven helices in cytochrome c oxidase. Therefore, it is reasonable to assume that the arrangement of the last two C‐terminal helices of subunit I in Cyt bo and the first two helices of subunit III in cytochrome c oxidase is similar. Hence, the density found in the bottom‐left region of the map is assigned to helices XIII and XIV of CyoB (shaded yellow). Consequently, the right bottom part of the density probably corresponds to the five helices of CyoC (shaded blue).
The remaining density may be interpreted as subunit IV (CyoD, shaded green) which is predicted to have three membrane‐spanning α‐helices (Chepuri and Gennis, 1990). This is attached to subunits I and III in the region where the bacterial cytochrome c oxidase also contains an additional small subunit with a single transmembrane span (Iwata et al., 1995; Figure 4C).
The molecular anatomy of the complexes belonging to the two main branches of the haem–copper oxidase superfamily is very similar. In particular, the arrangement of metal centres and their surrounding transmembrane helices in subunit I and the folding of subunit II are well conserved. However, the fact that these two oxidase subgroups utilize completely different electron‐donating substrates raises a question about structural differences between the quinol‐ and cytochrome c‐oxidizing complexes. Cytochrome c is a soluble protein with a mol. wt of 12 000–14 000 Da. In contrast, the physiological electron donor for Cyo bo, ubiquinol, is a small, hydrophobic molecule with a mol. wt <1000 Da (727 Da for ubiquinol‐8). Moreover, cytochrome c is an one‐electron donor while ubiquinol is a two‐electron donor.
The first proof for the high degree of similarity between cytochrome c and ubiquinol oxidases came from protein engineering studies on CyoA (van der Oost et al., 1992). This was confirmed by the subsequent crystallographic studies (Iwata et al., 1995; Tsukihara et al., 1995, 1996; Wilmanns et al., 1995). In relation to these studies, we have attempted to survey further structural similarities and possible differences by an electron crystallographic study on Cyt bo.
The projection structure of the entire Cyt bo at 6 Å resolution shows that the common subunit I–III segment has a similar fold to cytochrome c oxidase. The structure is very complex (25 predicted transmembrane helices; Chepuri and Gennis, 1990), and without additional information it would be extremely difficult to interpret. However, the known structure of cytochrome c oxidase from P.denitrificans (Iwata et al., 1995) can be used as a template for analysis of the projection map. The comparison of both structures in projection (Figure 4) has made it possible to identify all four subunits of Cyt bo and, moreover, it allows the assignment of single α–helices. This is especially important in places where major differences occur between the primary structures. The first difference appears in subunit I (Figure 4A). This subunit in cytochrome c oxidase consists of 12 α‐helices, whereas the corresponding CyoB subunit has been predicted to contain 15 transmembrane helices (Chepuri et al., 1990; Ma et al., 1993), which is supported by gene fusion experiments (Chepuri and Gennis, 1990). It is difficult to assign a particular density to the additional helix 0 at the N‐terminus, although we assume that a weak peak of extra density on the left side of CyoB corresponds to this helix (Figure 4A, above left of pore A). It has been suggested that the two C‐terminal helices of subunit I replace the first two helices of subunit III which are present in cytochrome c oxidase but missing in Cyt bo (Ma et al., 1993). Although there is virtually no sequence homology between helices XIII and XIV of CyoB and helices I and II of subunit III in cytochrome c oxidase, the superposition of both projections (Figure 4C) indicates an almost identical arrangement of these helices in the membrane. This is supported by the fact that the Cyt bo complex, in which subunits I and III have been genetically fused, is fully active (Ma et al., 1993). This swapping of a pair of transmembrane helices between subunits I and III is also known for the cytochrome c and quinol oxidases from Bacillus (Ishizuka et al., 1990; Saraste et al., 1991; Santana et al., 1992). Finally, in the case of the cytochrome oxidases in Thermus thermophilus (Mather et al., 1993) and Sulfolobus acidocaldarius (Lübben et al., 1994), a natural genetic fusion between subunits I and III has occurred, possibly reflecting the high temperatures under which these thermophilic organisms live. The possibility of these fusions and swapping of terminal helices from subunit I to subunit III reflects the vicinity of the termini in the canonical 12‐helix subunit I and the 7‐helix subunit III.
The second difference between Cyt bo and bacterial cytochrome c oxidases is predicted to involve subunit IV (CyoD). Subunit IV of cytochrome c oxidase from P.denitrificans contains a single transmembrane helix (Iwata et al., 1995), whereas hydropathy analysis and topological studies on CyoD have indicated three membrane‐spanning helices (Chepuri and Gennis, 1990; Chepuri et al., 1990). Cross‐linking studies propose a close proximity of subunit IV to the subunits I and III (Saiki et al., 1996).
Superposition of the lower regions of Cyt bo and cytochrome c oxidase shows indeed that after the assignment of the C‐terminal part of CyoB and five helices of CyoC, extra density remains which can account for the three α‐helices of CyoD (Figure 4C). One of these density patches superimposes almost exactly with the position of the single transmembrane helix of subunit IV of cytochrome c oxidase. The remaining density appears to be closely attached to subunit III and is assigned to the other transmembrane helices of CyoD. Between the C‐terminus of subunit I and subunit III, a peak of density remains unassigned at the resolution of 6 Å (Figure 4C).
The detectable density for CyoA shows a striking similarity to the structure of the corresponding subunit II in cytochrome c oxidase. Subunit II consists of two membrane‐spanning helices plus a large periplasmic domain (Iwata et al., 1995). In both structures, the transmembrane helices appear to be oriented either tilted (helix I) or perpendicular (helix II) with respect to the membrane normal. There is no distinctive density for the cytoplasmic part of subunit II. This part of the polypeptide is folded as an 11‐stranded β‐sandwich followed by three short α–helices (Wilmanns et al., 1995). While no distinct secondary structure can be expected for a β‐sheet at 6 Å resolution, α‐helices can already be resolved at medium resolution (Unwin and Henderson, 1975; Unwin, 1993). Extending the information obtained from the X‐ray structures (Iwata et al., 1995; Wilmanns et al., 1995), our electron crystallographic data confirm that the transmembrane part of subunit II is also structurally conserved. Additionally, the mutual arrangement of subunits I and II is likely to be the same in both oxidase subfamilies.
Cytochrome c binds to the periplasmic surface of the cytochrome c oxidase. A part of its binding site is known to reside on the exposed region of subunit II (Taha and Ferguson‐Miller, 1992; Lappalainen et al., 1995). For ubiquinol oxidases the situation must be different. Accounting for the hydrophobic nature of the substrate, the binding site is thought to be located within the hydrophobic interior of the membrane bilayer. Studies with a photoreactive radiolabelled substrate analogue have indicated that subunit II is at least partially involved in ubiquinol binding (Welter et al., 1994). Our future experiments will aim at the identification of this substrate‐binding site by using gold‐labelled ubiquinol derivatives. The combination of electron crystallography and gold‐labelled substrates or polypeptides has been used already in studies on the localization of the cytochrome c binding site of mitochondrial oxidase (Frey and Murray, 1994) and the location of subunit III in the same complex (Crum et al., 1994). Furthermore, our structural analysis will be extended to the recording of tilt series in order to obtain three‐dimensional data of the Cyt bo complex. A three‐dimensional model even at medium resolution will facilitate the interpretation of density maps and render a more accurate comparison possible with the available X‐ray data of cytochrome c oxidases.
Phylogenetic analysis (Castresana et al., 1994; van der Oost et al., 1994; Castresana and Saraste, 1995) has led to the suggestion that the first cytochrome c oxidase containing B‐ and C‐type haems evolved from a structurally similar nitric oxide reductase already under the primordial conditions when only a limited amount of free oxygen was present. This enzyme evolved to the mitochondrial‐type cytochrome c oxidase by the incorporation of A‐type haems and introduction of the CuA centre into subunit II. In a more recent step, the CuA centre was lost when the quinol oxidases developed in Eubacteria and Archaea. Our results show that both branches of the haem–copper oxidase family share the same fold, which confirms their common evolutionary history.
Materials and methods
Purification and crystallization
Purification and crystallization of Cyt bo from the E.coli strain RG129/pRCO1 was done as described (Warne et al., 1995), with modifications. The strain overproduces an engineered Cyt bo with a genetic fusion between cyoA and cyoB encoding subunits II and I, respectively (Ma et al., 1993). Two‐dimensional crystallization was carried out by microdialysis (Warne et al., 1995). Briefly, Cyt bo (1 mg/ml) was added to a 2.6:1 mixture of egg phosphatidylcholine and crude phosphatidylserine from bovine brain (Sigma, Germany) solubilized in Triton X‐100 with a lipid to protein ratio of 0.29 by weight. Dialysis of 0.2% (w/v) detergent against 25 mM Tris, pH 9.0, containing 0.5 mM EDTA and 1 mM sodium azide at 37°C resulted in the formation of tubular crystals 2–8 μm in length. The use of pH 9.0–10.0 was essential for the formation of the tubes. The crystals were stable for at least 12 months at 4°C.
Aliquots (2.0 μl) of the crystal suspension were pipetted onto the grid side of copper grids covered with a holey carbon film. The excess liquid was then removed by blotting with filter paper from the carbon side, ensuring the trapping of the crystals across the holes (Toyoshima, 1989). Immediately, grids were plunged into a liquid ethane/liquid nitrogen slurry at a temperature of 93 K (Dubochet et al., 1988). In the case of continuous carbon grids, 0.5 μl aliquots were applied to grids which had been glow‐discharged in a pentylamine atmosphere. The specimens were then blotted and frozen as above. Finally, the grids were transferred under cryo conditions to either a Gatan‐ or an Oxford cold stage and examined on a JEOL 2000EX electron microscope operating at an accelerating voltage of 200 kV.
Images were selected by optical diffraction, based on a maximum resolution of 8–10 Å. Between 5% (holey carbon specimens) and 15% (continuous carbon grids) of ∼200 recorded images were suitable for further processing. Areas showing good isotropic resolution were marked and digitized on a Perkin‐Elmer 1010 M flat‐bed microdensitometer. The size of the selected area was 3000×3000 pixels scanned with a circular aperture of 15 μm at a step size of 10 μm, corresponding to a pixel size of 2.5 Å (magnification 40 000×) at the specimen.
The digitized images were processed using the MRC image processing programs (Crowther et al., 1996). Briefly, the images were corrected for distortions of the crystal lattice, contrast transfer function (CTF), astigmatism and specimen drift (Henderson et al., 1986). Manual indexing of the Fourier transform of the images revealed an orthorhombic planar space group with cell constants of a = 200.65 Å and b = 85.92 Å. This is in good agreement with the previous results with negatively stained crystals (space group p22121, a = 210 Å, b = 90 Å, Warne et al., 1995). The observed flattened tubes were not large enough for collecting electron diffraction data (width of the tubes ∼0.5 μm). Therefore, the Fourier amplitudes were estimated from the image amplitudes. Comparison of the internal phase residual in all 17 space groups (Valpuesta et al., 1994) revealed p22121 symmetry for one image (No. 1415) and p121 symmetry for all other images examined so far (15 micrographs). The internal order of the lattice does not seem to be influenced by the space group since all processed images revealed structure factors to 6–8 Å. After refinement of the phase origin of image 1415 (using ORIGTILTD, Henderson et al., 1986), the amplitudes and phases of symmetry‐related reflections were averaged. The phases were rounded to either 0 or 180°, as the projection structure is centro‐symmetric in this plane group. Finally, the image amplitudes were rescaled as a function of resolution to compensate for the loss of contrast caused by the limited intrinsic order of the crystals. The scale factors were derived from a comparison with the average amplitude of bacteriorhodopsin in projection at a given resolution (Havelka et al., 1995).
Calculation of the projection map
With the amplitudes and phases obtained after the image processing, Fourier projection maps were calculated using the CCP4 program package (Collaborative Computational Project, No. 4, 1994). The structure factors were weighted by their figures of merit (Bullough and Tulloch, 1991).
We would like to thank M.Auer, M.Cyrklaff, B.Gowen, V.Guénebaut, W.Kühlbrandt, H.Savage and D.N.Wang at EMBL for their assistance and advice with the early stages of electron microscopy and image processing. We are grateful to R.Henderson and N.Grigorieff at MRC‐LMB, Cambridge, for discussion and invaluable comments, and to H.Michel, MPI für Biophysik, Frankfurt, for kindly making the co‐ordinates of cytochrome c oxidase from P.denitrificans available to us. U.G. acknowledges financial support from EMBL.
- Copyright © 1997 European Molecular Biology Organization