The 1.3-Å resolution structure of bovine cytochrome c oxidase suggests a dimerization mechanism

Highlights

• •Optimization of the cryoprotectant soaking method for protein crystals
• •Improvement of the crystallographic resolution for cytochrome c oxidase dimer
• •Remodeling of weakly bound detergents and lipids involving dimerization
• •A reversible monomer/dimer transition may contribute to enzymatic regulation

Abstract

Cytochrome c oxidase (CcO) in the respiratory chain catalyzes oxygen reduction by coupling electron and proton transfer through the enzyme and proton pumping across the membrane. Although the functional unit of CcO is monomeric, mitochondrial CcO forms a monomer and a dimer, as well as a supercomplex with respiratory complexes I and III. A recent study showed that dimeric CcO has lower activity than monomeric CcO and proposed that dimeric CcO is a standby form for enzymatic activation in the mitochondrial membrane. Other studies have suggested that the dimerization is dependent on specifically arranged lipid molecules, peptide segments, and post-translationally modified amino acid residues. To re-examine the structural basis of dimerization, we improved the resolution of the crystallographic structure to 1.3 Å by optimizing the method for cryoprotectant soaking. The observed electron density map revealed many weakly bound detergent and lipid molecules at the interface of the dimer. The dimer interface also contained hydrogen bonds with tightly bound cholate molecules, hydrophobic interactions between the transmembrane helices, and a Met–Met interaction between the extramembrane regions. These results imply that binding of physiological ligands structurally similar to cholate could trigger dimerization in the mitochondrial membrane and that non-specifically bound lipid molecules at the transmembrane surface between monomers support the stabilization of the dimer. The weak interactions involving the transmembrane helices and extramembrane regions may play a role in positioning each monomer at the correct orientation in the dimer.

1. Introduction

Cytochrome c oxidase (CcO) in the respiratory chain catalyzes oxygen reduction by coupling electron and proton transfer through the enzyme and proton pumping across the membrane. CcO isolated from the mitochondrial membrane consists of core subunits I, II, and III, which are essential for the enzymatic function, and other subunits that surround these core subunits. Electrons transferred from cytochrome c are received at the Cu_A site of subunit II and transferred to the oxygen reduction site through heme a in subunit I. The substrate dioxygen is transferred through the transmembrane region of subunit III and bound to the heme a₃–Cu_B site. Protons for oxygen reduction are transferred through the hydrogen bond network of the K- and D-pathways in subunit I [1]. It has been proposed that pumped protons are transferred through the H- or D-pathway [2], [3], [4].

Although the functional unit of CcO is a monomer, the enzyme in the mitochondrial membrane exists as three states: monomer, dimer, and a supercomplex with respiratory complexes I and III [4]. Supercomplex formation is thought to contribute to efficient electron transfer, stabilization of individual enzyme complexes, and inhibition of reactive oxygen species (ROS) generation [5], [6], [7], [8], [9], [10], [11], [12]. However, the roles of the monomer and dimer forms, as well as the transition mechanism between these three states, remain unclear. A recent study showed that dimeric CcO has lower activity than monomeric CcO and proposed that dimeric CcO is a standby form for enzymatic activation in the mitochondrial membrane [13]. Comparison of the crystal structures of dimeric and monomeric CcO revealed that a molecule of cholate (CH), which was used as a detergent to solubilize CcO, was bound to the interface of dimeric CcO near the entry site of the proton transfer K-pathway, whereas no cholate was associated with monomeric CcO. This result suggests that the binding of cholate or structurally similar physiological ligands are involved in dimerization. On the other hand, in a previous analysis that revealed the structure of dimeric CcO at 1.8-Å resolution (PDB ID: 2DYR), four phospholipids [one cardiolipin (CL1), two phosphatidylethanolamines (PE1, PE3), and one phosphatidylglycerol (PG3)] were modeled at the interface of the dimer to bridge the two monomers. In addition, the N-terminal segment of subunit VIa is bound to another CcO monomer [14]. Unfortunately, however, the electron densities of the lipids and N-terminal segment were less clear. According to previous mass spectrometric results, seven or eight phospholipids or three or four CLs could be contained within the cavity between the monomers in dimeric CcO, and the lysine residues in subunits III and Vb (K77^III, K57^Vb, and K68^Vb) that are acetylated by post-translational modification (PTM) are located at the dimer interface, suggesting a role for these species in the regulation and stabilization of the dimer [15]. To date, however, the structures of lipid molecules contained between the monomers and the amino acid residues subjected to PTM remain to be determined. Therefore, to understand the structural basis of the dimerization mechanism, it is necessary to determine more accurate structures by improving the crystallographic resolution.

To improve the resolution, crystallographic analysis of CcO must be performed using X-ray diffraction data obtained under cryogenic conditions. To conduct such measurements, the crystals must be frozen without the formation of ice, which would destroy the crystal packing and generate misleading diffraction data. In general, to prevent ice growth, a cryoprotectant such as polyethylene glycol (PEG), ethylene glycol (EG), or glycerol is introduced into the crystal [16]. Although larger crystals of bovine CcO yield higher-resolution X-ray diffraction data [17], the introduction of a cryoprotectant into such large crystals often results in damage due to osmotic shock when the crystal composition changes abruptly. On the other hand, if the introduction of a cryoprotectant causes effective dehydration, the enzymes will be more efficiently packed within the crystals, and the resolution of the diffraction data should be improved. Hence, we optimized the type and concentration of the cryoprotectant, as well as the soaking method for cryoprotectant introduction. As a result of this optimization, the resolution of the X-ray diffraction data from crystals of CcO improved to 1.3 Å.

Structural analysis at 1.3-Å resolution revealed numerous weakly bound detergent and lipid molecules at the dimer interface. Furthermore, we found that the dimer interface also contained hydrogen bonds to the tightly bound cholate molecule, hydrophobic interactions between the transmembrane helices, and a Met–Met interaction between extramembrane regions. On the other hand, the previous structural models of CL1 and the N-terminal segment of subunit VIa at the dimer interface did not fit the current electron density data. We could not observe clear electron densities for any known PTM groups, with the exception of two formyl groups modifying N-terminal methionines. On the basis of these structural results, we discuss the dimerization mechanism of bovine CcO.

2. Results and Discussion

2.1. Purification and mass spectrometric characterization of bovine CcO

To investigate the oligomerization states (monomer, dimer, and supercomplex) of CcO in the mitochondrial membrane, we solubilized the membranes from bovine heart using detergents, digitonin, lauryl maltose neopentyl glycol (LMNG), n-dodecyl-β-d-maltoside (DDM), and n-decyl-β-d-maltoside (DM). BN-PAGE analysis revealed that monomeric complex I, monomeric complex V, dimeric complex III, and monomeric complex IV were present in all solubilized fractions (Fig. 1A, lanes 1–4). The solutions with digitonin and LMNG also contained the supercomplex. On the other hand, when CcO was purified from the cholate-solubilized fraction, most of the CcO in the cholate solution was obtained in the dimeric state (Fig. 1B). During the purification process, CcO formed a dimer upon ammonium sulfate fractionation in the presence of cholate (Fig. 1B, lanes 1–4). However, the abundance of monomeric CcO increased after replacement of cholate with DM (Fig. 1B, lane 5), and almost all molecules of CcO were in the monomeric state after ammonium sulfate fractionation in the presence of DM followed by dialysis (Fig. 1B, lane 6). Finally, CcO formed a dimer again during sample washing by repeated concentration and dilution of the sample with 40 mM Na-Pi buffer (pH 6.8) containing 0.2% DM (Fig. 1B, lane 7). Concentration of about 800 mg of purified sample, obtained from 1.1 kg of bovine heart, resulted in the growth of about 300 mg of micro-crystals of dimeric CcO with a high level of reproducibility. These micro-crystals were dissolved in 40 mM Na-Pi buffer (pH 6.8) containing 0.2% DM (this solution is referred to as “crystalline sample”), and used in the biochemical and crystallographic analyses described below. As shown in Fig. 1B (lane 8), the CcO in the crystalline sample was mainly in the dimeric state, although its dimerization/monomerization was dependent on the concentration of CcO and the pH value of the solution [13]. During the purification process, the amount of cholate molecules in the sample decreased in a stepwise manner; however, the final crystalline sample still contained 14.1 cholate molecules per CcO monomer (Table 1).

Fig. 1.

BN-PAGE of bovine CcO during the purification process. The labels Super, I₁, V₁, III₂, IV₂, and IV₁ indicate the supercomplex, monomeric complex I, monomeric complex V, dimeric complex III, dimeric CcO (complex IV), and monomeric CcO, respectively. (A) Comparison of the compositions of bovine mitochondrial proteins solubilized with digitonin (lane 1), LMNG (lane 2), DDM (lane 3), and DM (lane 4). Approximately 30 μg of protein from each condition was loaded in each lane. Lane 1: Supernatant obtained after solubilization of bovine mitochondria in digitonin at 10 times the amount of protein. Lane 2: Supernatant obtained after solubilization of 6 mg/mL bovine mitochondria in 2% LMNG. Lane 3: Supernatant obtained after solubilization of 6 mg/mL bovine mitochondria in 2% DDM. Lane 4: Supernatant obtained after solubilization of 6 mg/mL bovine mitochondria in 2% DM. (B) Purification profile of complex IV according to BN-PAGE. During the purification steps, the level of purity was determined using BN-PAGE. The positions of complexes IV₁ and IV₂ are shown on the left. Lane 1: 33–50% saturation AS fractionation in the presence of 3.2% cholate [purification step (i) described in Section 4.1]. Lane 2: Re-solubilized precipitate dissolved in 2% cholate buffer after dialysis against 40 mM Na-Pi buffer (pH 7.4) [step (ii)]. Lane 3: 25–45% saturation AS fractionation in the presence of cholate [step (iii)]. Lane 4: 25–40% saturation AS fractionation in the presence of cholate [step (iv)]. Lane 5: 40–60% saturation AS fractionation in the presence of DM [step (vi)]. Lane 6: After dialysis against 10 mM Na-Pi buffer (pH 7.4) [step (ix)]. Lane 7: After washing using ultrafiltration membrane 4 [step (xiii)]. Lane 8: Micro-crystals solubilized in 40 mM Na-Pi buffer (pH 7.4) containing 0.2% DM.

Table 1.

Change in the number of cholate molecules per CcO during purification.

No.	Purification step	Cholate/CcO (mol/mol)
Iv	AS fractionation in the presence of cholate: 25–40% sat.	1,050.2
V	AS fractionation in the presence of cholate: 25–35% sat.	901.5
Vi	AS fractionation in the presence of DM: 40–60% sat.	382.6
Vii	AS fractionation in the presence of DM: 50–70% sat.	139.4
viii	AS fractionation in the presence of DM: 55–70% sat.	73.3
Ix	After dialysis against 10 mM Na-Pi buffer	32.5
X	Washing using ultrafiltration membrane 1	26.6
Xi	Washing using ultrafiltration membrane 2	24.3
Xii	Washing using ultrafiltration membrane 3	21.6
xiii	Washing using ultrafiltration membrane 4	19.7
	Mother liquor	31.5
	Micro-crystals	14.1

The details of each purification step are described in Section 4.1.

Mass spectrometric analysis of the crystalline sample revealed the peaks for all 13 subunits, several of which were altered by PTMs. In addition to the subunit VIa peak, the mass spectra contained several lower-mass peaks in the m/z=8,900–9,600 region (Fig. S1). Together with the mass calculated based on the N-terminal amino acid sequence, this is consistent with the proposal that a truncated subunit VIa lacking the three N-terminal residues is >3-fold more abundant than intact subunit VIa, and that molecules of subunit VIa lacking 1, 4, 5, or 7 residues are also present, albeit in relatively small amounts (Table 2). The mass spectra also displayed a peak consistent with subunit VIb containing one acetylated residue and two peaks for subunit VIc, one without PTM and one in which a single residue was acetylated (Table 2). The peak intensity for acetylated subunit VIc was significantly higher than that for unmodified subunit VIc.

Table 2.

Mass spectrometric data.

UniProt ID	Subunit	Modification	Measured mass	Average mass	Difference
P00396	I	Formylation (M1)a	57,012.1	57,032.3	−20.2
P68530	II	Formylation (M1)b	26,025.4	26,021.5	3.9
P00415	III		29,838.3	29,932.7	−94.4
P00423	IV		17,142.8	17,152.6	−9.8
P00426	Va		12,428.8	12,436.1	−7.3
P00428	Vb		10,664.2	10,670.0	−5.8
P07471	VIa	Mature	9,527.1	9,532.8	−5.7
		Missing one N-terminal residue	9,454.4	9,461.7	−7.3
		Missing three N-terminal residues	9,298.5	9,303.6	−5.1
		Missing four N-terminal residues	9,229.4	9,232.5	−3.1
		Missing five N-terminal residues	9,100.4	9,104.3	−3.9
		Missing seven N-terminal residues	8,927.7	8,932.2	−4.5
P00429	VIb	Acetylation (A1)c	10,058.0	10,025.2	32.8
P04038	VIc		8,475.8	8,479.0	−3.2
		Acetylation	8,516.5	8,479.0	37.5
P07470	VIIa		6,671.2	6,673.7	−2.5
P13183	VIIb		6,354.8	6,357.2	−2.4
P00430	VIIc		5,440.0	5,441.4	−1.4
P10175	VIII		4,960.6	4,961.8	−1.2

The average mass values were calculated from the peptide sequence using PeptideMass on the ExPASy website (https://web.expasy.org/peptide_mass/).

^a[30].

^b[31].

^c[32].

2.2. Crystallization of dimeric CcO and procedure for cryoprotectant soaking

To crystallize the dimeric CcO, 70–95 mg/mL CcO was dissolved in 40 mM Na-Pi buffer (pH 6.8) containing 0.2% (w/v) DM. Evaluation of crystallization conditions at various pH values revealed the growth of many crystal nuclei at lower pH and no crystal growth at higher pH. Therefore, we concluded that a pH of 6.8 was optimal. Batch crystallization was performed using a concave microscope slide (depression: 50–120 μL) with the depression covered by a coverslip (Fig. 2A). This technique preserved the crystals without exposing the enzyme to air. Upon adding PEG 4k, a precipitating agent, at a final concentration of 1% or lower, dimeric CcO crystals were obtained with a high level of reproducibility. The crystals obtained through batch crystallization exhibited no apparent damage after 10 days, whereas those prepared by hanging-drop crystallization under the same conditions began to degrade within 5 days. Batch crystallization resulted in the formation of crystal nuclei at well-dispersed positions, which kept the crystals from overlapping. This approach allowed individual crystals to grow. A microscope slide with a depression of 120 μL permitted the growth of up to 30 crystals with dimensions of 0.4–1.0 mm per side; these crystals were used in the X-ray diffraction measurements. In addition, the use of a cryoprotectant allowed the crystals to precipitate on the concave microscope slide without the slide needing to be moved. This point will be discussed further below.

Fig. 2.

Crystallization and cryoprotectant soaking procedures. (A) Batch crystallization. (B) Replacement of the mother liquor by a solution containing 3% PEG 4k. (C) Nine-step soaking. The solutions used to replace the mother liquor and freeze the crystals were mixed in nine different ratios. In each step, half of the volume of the soaking solution was replaced with a solution containing a 5% higher concentration of EG, and the sample was agitated then left to stand for 30 min until the next step. (D) Eighteen-step soaking. The solutions used to replace the mother liquor and freeze the crystals were mixed in nine different ratios. One-fifth of the volume of the soaking solution was replaced with a solution containing a 5% higher concentration of EG, and the sample was agitated then left to stand for 30 min until the next step, when one-fifth of the volume of the solution was replaced again. These steps were repeated nine times over 2–3 days. (E) Fifty-step soaking. The solutions used to replace the mother liquor and freeze the crystals were mixed in 50 different ratios. In each step, half of the volume of the soaking solution was slowly replaced with a solution containing a <1% higher concentration of EG; the solution mixture was slowly pipetted and left to stand for 30 min.

A subset of the CcO crystals grown in the mother liquor started to dissolve after 10 days. To stabilize the crystals, we replaced the mother liquor with a solution of 40 mM Na-Pi buffer (pH 6.8), 0.2% DM, and 3% PEG 4k (Fig. 2B). When EG was added as a cryoprotectant, all of the crystals dissolved. However, dissolution of the crystals could be prevented by lowering the pH to 5.7. Therefore, the final composition of the solution used to freeze the crystals was 40 mM Na-Pi buffer (pH 5.7), 0.2% DM, 5% PEG 4k, and 45% EG. Introduction of 45% EG was initially performed by a nine-step soaking method (Fig. 2C). Although many of the crystals cracked upon the addition of increasing concentrations of EG, as well as during agitation, X-ray diffraction data could still be collected at 1.8-Å resolution (PDB ID: 2DYR). An 18-step soaking method decreased the number of damaged crystals and improved the resolution to 1.5 Å (Fig. 2D; PDB ID: 5B1A). Nonetheless, some crystals were still damaged.

To optimize the procedure for introducing the cryoprotectant, we devised a new approach (50-step soaking) to keep the differences in the EG concentration to less than 1% per step (Fig. 2E). To compare the 18- and 50-step soaking methods, we assessed the quality of the crystals obtained from both methods based on the mosaicity values calculated from the diffraction data. As shown in Fig. 3A (methods 1 and 2), the 50-step soaking decreased the mosaicity from 0.28 to 0.18, indicating better crystallinity than the 18-step soaking. Consequently, we used the 50-step approach for the following experiments.

Fig. 3.

(A) Cryoprotectant soaking conditions for the CcO crystals and the mosaicity of the crystals. (B) Optimized conditions for the crystallization and cryoprotectant soaking.

2.3. Composition of the solution used for cryoprotectant soaking

To optimize the composition of the solution used for freezing the crystals, we tested glycerol, diethylene glycol, triethylene glycol (TEG), and propylene glycol in place of EG. Preliminary experiments using the 18-step soaking method indicated that EG and TEG caused less apparent damage to the crystals than the other cryoprotectants. Hence, we also compared EG and TEG in the 50-step soaking method. Specifically, we evaluated the quality of the crystals treated with EG and TEG based on the resolution of observed diffraction spots and mosaicity. As shown in Fig. 3A (methods 2 and 3), the EG solution afforded higher maximum resolution and lower mosaicity than the TEG solution. Accordingly, EG was used in subsequent experiments.

Although the primary aim of cryoprotectant soaking is to prevent the formation of ice in the frozen crystals, dehydration should also result in more efficient packing of the enzymes within the crystals, leading to improved resolution. Thus, we examined in detail the concentration of EG and the type and concentration of PEG in the cryoprotectant solution. Fig. 4 shows the maximum resolution and mosaicity measured from crystals treated with PEGs of various molecular weights. The use of PEGs with molecular weights exceeding PEG 4k was not feasible, as the resolution and mosaicity varied widely (Fig. 3A, methods 4A–4D, and Fig 4A). In the case of lower-molecular-weight PEGs, although the resolution varied irrespective of the PEG used, the mosaicity was significantly lower than that observed with higher-molecular-weight PEGs (Fig. 3A, methods 6A–6D, and Fig. 4B). The maximum resolutions among the lower-molecular-weight PEGs were almost identical; hence, PEG 4k, which was used for crystallization, was also used in the following experiments.

Fig. 4.

Resolution, mosaicity, and b-axis length of the crystals prepared by 50-step soaking using various solutions for freezing the crystals. X-ray diffraction data were collected using at least three crystals prepared under the same conditions. (A) The solution for freezing the crystals contained 40 mM Na-Pi buffer (pH 5.7), 0.2% DM, 35% EG, and 15% of the indicated type of PEG (PEG 4k, PEG 8k, PEG 10k, or PEG 20k). Crystals obtained from six different purification batches were treated with each type of PEG. To avoid abrupt changes in the type of PEG, the solutions for replacement of the mother liquor contained PEG 4k at the same concentration as the crystallization conditions. The concentration of EG and type of PEG were changed over 50 steps. (B) The solution for freezing the crystals contained 40 mM Na-Pi buffer (pH 5.7), 0.2% DM, 40% EG, and 8% of the indicated type of PEG (PEG 400, PEG 1k, PEG 1.5k, or PEG 4k). Crystals obtained from three different purification batches were treated with each type of PEG. For replacement of the mother liquor, a solution containing 40 mM Na-Pi buffer (pH 6.5), 0.2% DM, 2% EG, and 1% (w/v) PEG 4k was used. (C, D) The solution for freezing the crystals contained 40 mM Na-Pi buffer (pH 5.7), 0.2% DM, and various concentrations of EG and PEG 4k. Crystals obtained from 26 different purification batches were treated with each combination. Prior to soaking with cryoprotectant, the unit cell parameters were a = 189 Å, b = 210 Å, and c = 178 Å.

We optimized the combination of EG and PEG 4k concentrations by examining nine different conditions (Fig. 3A, methods 2, 4A, and 5A–5D, and Fig. 4C and D). Crystal quality was evaluated based on resolution, mosaicity, and b-axis length in the unit cell parameters. Packing of CcO in the crystal resulted in the least contact between CcO molecules along the b axis. Dehydration due to the soaking shortened the b axis. The results revealed that crystals diffracting to a higher resolution possessed a b-axis length of 203–204 Å, which was 6–7 Å shorter than the length of 210 Å prior to soaking. As shown in Fig. 4C, the b-axis lengths varied widely from 204 to 207 Å in 45% EG and 5% PEG 4k, decreased to as low as 203 Å in 35% EG and 15% PEG 4k, and were consistent at 204 Å in 40% EG and 8% PEG 4k. When the b-axis length was shorter than 204 Å, the mosaicity was significantly increased (Fig. 4D), indicating that excessive dehydration may hamper crystal packing. A crystal with a b-axis of ~204 Å exhibited low mosaicity with little variation, indicating a high level of isomorphism. Thus, we concluded that 40% EG and 8% PEG 4k was the optimal combination for freezing our crystals.

In the first step of cryoprotectant soaking, a solution containing 40 mM Na-Pi buffer (pH 6.8), 0.2% (w/v) DM, and 3% (v/v) PEG 4k was used to replace the mother liquor (Fig. 3A, method 5D). To reduce crystal damage to the greatest extent possible, when replacing the mother liquor, 2% EG was added before crystallization. In addition, the PEG concentration in the solution used to replace the mother liquor was decreased to 1% and the pH was lowered from 6.8 to 6.5 to decrease the solubility of the enzyme in the crystals. Thus, the solution used to replace the mother liquor was 40 mM Na-Pi buffer (pH 6.5) containing 0.2% (w/v) DM, 2% (v/v) EG, and 1% PEG 4k. After replacing the mother liquor, the EG concentration was brought to 40% and the PEG 4k concentration to 8% over 50 steps. The resultant crystals exhibited an extremely high level of isomorphism with low mosaicity and a b-axis length of ~204 Å (Fig. 3A, method 6A, Fig. 3B, and Fig. 4C and D).

2.4. Structural analysis of lipids, detergents, and PTM sites at 1.3-Å resolution

As described above, we examined in excess of 300 diffraction data sets obtained using more than 20 compositions of cryoprotectant solution and crystals prepared from over 40 different purification batches. On the basis of these results, we performed X-ray diffraction measurements to collect a full data set with high resolution. Among 22 data sets, the highest resolution was obtained from a crystal prepared under the conditions shown in Fig. 3A, method 6A; this crystal was analyzed at 1.3-Å resolution (Table 3). In the electron density map obtained from this data set, long and thin electron densities were observed on the transmembrane surface of CcO and the membrane region between each monomer of dimeric CcO, although most of the densities were poor. These electron densities indicate weak binding of lipids and detergents to CcO. On the basis of the detailed analysis described below, we modeled two phosphatidylglycerols (PGs), one phosphatidylethanolamine (PE), three cardiolipins (CLs), 13 hydrocarbon tail fragments of unidentified lipids, and 17 DMs per monomer of dimeric CcO (Fig. 5).

Table 3.

Statistics of crystallographic analysis.

	High-resolution data	Anomalous data
Diffraction data
Space group	P212121
Unit-cell parameters (Å)	a = 182.0, b = 204.2, c = 177.8
Resolution (Å)	200.0–1.30 (1.31–1.30)	200.0–1.79 (1.84–1.79)
σ cut off	−3.0	−3.0
Observed reflections	21,435,910	171,250,826 (3,189,309)
Independent reflections	1,596,005 (39,643)	1,193,955 (84,589)
Redundancy	13.4 (11.0)	143.4 (37.7)
Completeness (%)	100 (100)	99.6 (95.3)
<I/error>	39.7 (1.0)	63.62 (2.42)
Rpim (%)	3.1 (93.6)	Rmeas (%) 8.0 (198.6)
CC1/2	0.55	0.82
Wilson B factor (Å2)	20.4	29.6
Refinement
Resolution (Å)	40–1.30 (1.34–1.30)
R	0.149 (0.37)
Rfree	0.171 (0.37)
r.m.s.d., bonds (Å)	0.018
r.m.s.d., angles (degrees)	2.11

Fig. 5.

Whole structure of dimeric CcO and bound lipid and detergent molecules. The whole structure of dimeric CcO is represented as a cartoon model and a molecular surface. Each CcO monomer is shown in gray or green. The hydrocarbon tails of the lipids and DMs are colored yellow and orange, respectively. Cholate is represented by a dark gray stick model. Hemes a and a₃ are represented by a magenta stick model. (A) View from the membrane. The top of the panel represents the intermembrane space and the bottom is the matrix space. (B) View from the intermembrane space. (C) View from matrix space. (D–G) The lipids and detergents bound to the monomer in dimeric CcO. The top of the panel represents the intermembrane space and the bottom is the matrix space.

To identify phosphorus atoms in phospholipid structures, we measured the anomalous difference electron density using X-ray diffraction data at 1.75 Å and analyzed at a resolution of 1.8 Å. In this map, we observed three strong anomalous peaks inside the transmembrane region of subunit III. These peaks indicated phosphorus atoms in two PGs (PG1, PG2) and one PE (PE1), which were already reported in previous studies [14]. In the previous structural analysis of dimeric CcO, we modeled one CL (CL1) at the transmembrane region of the dimer interface, suggesting that CL1 was involved in the dimerization of CcO monomers [14]. On the other hand, in a recent crystallographic analysis of monomeric CcO, the CL1-binding site was involved in crystal packing and bound to another CcO, indicating that binding of CL1 was impossible in this crystal [13]. Because other lipid-binding sites are not involved in crystal packing, the number of phosphorus atoms bound to the CcO monomer in the monomeric CcO crystal was expected to be two less than the number in the dimeric CcO crystal owing to the absence of one CL. However, biochemical analysis of both monomeric and dimeric CcO crystals revealed no significant difference in phosphorus atom contents, i.e., the phosphorus atoms per CcO monomer in the dimeric and monomeric CcO crystals were 12.9 and 12.3, respectively (Table 4). Therefore, we carefully re-examined the CL1 site in the electron density map of a dimeric CcO crystal at 1.3-Å resolution. The electron density at the CL1 site was weak relative to the peptide region, and the anomalous difference electron density was not observed at this site. Weakly bound DM and lipid fragments provide a better fit to the electron density than the previous CL1 model (Fig. 6A), although we cannot completely exclude the possibilities that CL is bound at this site with very low occupancy or non-specifically bound in a highly flexible conformation.

Table 4.

Phosphorous content of crystalline bovine heart CcO.

Detergent used to stabilize CcO	Phosphorous atoms per CcO monomer
DM	12.9 ± 1.9 (n=40)
3OM	12.3 ± 1.5 (n=6)

Fig. 6.

(A) Partial structure of DMs at the dimer interface. The previous CL model (PBD ID 2DYR) was replaced by this model. (B) CL2 structure. (C) CL3 structure. The previous triglyceride (TG) model (PBD ID 2DYR) was replaced by this model. (D) CL4 structure. The previous TG model (PBD ID 2DYR) was replaced by this model. The models are colored as described in Fig. 5. The 2F_O–F_C electron density maps at the 0.5σ level are shown as blue cages. The anomalous difference electron density maps at the 3.0σ level are shown as magenta cages. The top of each panel represents the intermembrane space and the bottom is the matrix space.

Small anomalous peaks were observed at three sites on the transmembrane surface. These sites are identical to the CL2, CL3, and CL4 sites reported in a recently determined structure of monomeric CcO [13]. Anomalous peaks corresponding to two phosphorus atoms in CL2 were observed at the matrix side of the transmembrane surface of subunit III (Fig. 6B). The binding geometry of the phosphorus atoms was essentially the same as in the monomeric CcO. An anomalous peak for one of the two phosphorus atoms was observed at the CL3- and CL4-binding sites (Fig. 6C and D). Although another phosphorus atom in CL3 and CL4 could not be identified based on the anomalous peak, the electron densities for the four hydrocarbon tails of CL3 and CL4 were observed. The conformations of the hydrocarbon tails of CL3 and CL4 were essentially the same in monomeric CcO, in which the anomalous peaks for both phosphorus atoms in CL3 and CL4 were observed [13]. This indicates that some of the head groups of CL3 and CL4 possess more flexible conformations in the dimeric CcO crystal. The dimeric CcO crystal was grown in the presence of DM, whereas the monomeric CcO crystal was grown with 3-oxatridecyl-α-d-mannoside (3OM) as the detergent. Differences in the environment of the transmembrane surface surrounded by detergent may affect the flexibility of the head groups of CL3 and CL4. A previous study demonstrated that when the structure of CcO was superimposed onto the respiratory supercomplex, CL3 was situated between CcO and complex I while CL2 was located in the space between CcO and complex III, suggesting that both CL2 and CL3 are involved in supercomplex formation [13]. However, CL2, CL3, and CL4 are not involved in CcO dimerization.

One of the tails of CL3 was located near a terminal region of the hydroxyfarnesylethyl moiety of heme a. The current electron density map at 1.3-Å resolution revealed that the terminal portion of the farnesyl group exhibited multiple conformations (Fig. 7). One conformation was identical to that in a previous model of the resting oxidized structure at 1.8-Å resolution (PDB ID: 2DYR). In the second conformation, a terminal part of the farnesyl group was rotated. Although the hydroxyfarnesylethyl moiety is known to change conformation depending on the redox state of CcO [18], the second conformation was distinct from that observed in the fully reduced structure. The hydroxyl group in the second conformation was the same as that in the resting oxidized structure, whereas it was altered in the fully reduced structure. The distance between the C25 atom of heme a and the C61 atom of CL3 was 4.7 Å, and the electron density of the tail region of CL3 was lower than that of the surrounding amino acid residues, indicating that the lower occupancy of CL3 may be related to the multiple conformations of the farnesyl group of heme a. The electron density at 1.3-Å resolution also indicated numerous other conformations of the amino acid residues, although most of these were identified in the previous structural analysis performed at 1.5-Å resolution (PDB ID: 5B1A) [19].

Fig. 7.

Structure of the tail region of CL3 and the hydroxyfarnesylethyl group of heme a. The hydrocarbon tail of CL3 is represented by a yellow stick model. One of the conformers of heme a, which is the same as that in the previous model of the resting oxidized structure (PDB ID: 5B1A), is represented by a magenta stick model. The hydrocarbon tail of another heme a conformer is represented by a green stick model. The 2F_O–F_C electron density map at the 0.5σ level is represented by a blue cage.

No anomalous peak was observed with the other long and thin electron densities, indicating that these represent the hydrophobic tails of DMs or lipids whose head groups possess flexible conformations. In a previous analysis, several phospholipids were modeled for these electron densities [14]. However, because the previous electron density at 1.8-Å resolution was not clearer than the electron density in this study, we remodeled the DM and lipid tail structures. Numerous long and thin electron densities were observed at the membrane region between the two CcO monomers. By comparing these electron densities with the length of the hydrocarbon tail and the shape of the head group, we modeled 7 DMs and 13 lipid fragments in this region (Fig. 5D). This result indicates that, under physiological conditions, many lipid molecules could be placed between the two monomers but no specific phospholipid is required for dimer formation. We also newly modeled 10 DMs for the electron densities found outside the transmembrane surface of the CcO dimer (Fig. 5E–G). At several sites, the DM or the lipid molecule was located at the same place with low occupancy.

In the previous crystal structure of dimeric CcO at 1.8-Å resolution, the N-terminal segment of subunit VIa was modeled in the transmembrane region between two CcO monomers [14]. On the basis of this model, it was suggested that the dimer form was stabilized by the binding of this segment to the transmembrane surface of another monomer, although the model atoms for this segment displayed a high temperature factor (Fig. 8A). In the electron density map at 1.3-Å resolution presented in Fig. 8B, no clear density was observed from Ala1^VIa to phosphothreonine 11^VIa (Tpo11^VIa), and no significant anomalous signal was observed at the phosphorus atom position in the previous model structure of Tpo11^VIa. These results indicate that the N-terminal segment of the subunit VIa structure is non-specifically located with a flexible conformation, which is possibly attributable to the glycine-rich sequence of (Gly9–Gly10–Tpo11–Gly12)^VIa in this segment. As described above, the mass spectra indicate that one to seven residues in the N-terminal segment are easily truncated by PTM (Fig. S1). In addition, a previous mass spectrometric study [15] did not identify a signal (mass peak) corresponding to subunit VIa in which one amino acid residue was phosphorylated. These results are consistent with our crystallographic results in this study. Therefore, we conclude that this segment is not involved in the dimerization of CcO and that phosphorylation of Thr11 is not evident. We did not include the N-terminal 11 residues in the model structure; instead, we modeled weakly bound lipid and detergent tails near this site on the basis of the weak electron density observed in the current map.

Fig. 8.

(A) Temperature factor of the N-terminal amino acid residues of subunit VIa in the previous model (PBD ID 2DYR). The blue and orange curves represent subunit VIa for each monomer of dimeric CcO. (B) The N-terminal structure of subunit VIa is represented by a stick model. The molecular surface of the CcO monomer, which is distinct from the CcO monomer containing subunit VIa represented by the stick model, is shown in gray. The 2F_O–F_C electron density map at the 2.0σ level is represented by a blue cage.

In our previous study, a total of 13 lipid structures, including two CLs, one PC, three PEs, four PGs, and three TGs, were modeled per monomer of dimeric CcO (PDB ID: 2DYR) [14]. These lipid models were used for subsequent structural analyses of dimeric CcO (PDB IDs: 2EIJ, 5B1A, etc.). The previous determination of the lipid structures was based on the electron density map at 1.8-Å resolution and the biochemical studies detailed below. The lipids (PG, PE, CL, PC, and TG) were identified by two-dimensional thin-layer chromatography [14]. Analysis of the phosphorus content of the crystalline sample revealed that the number of phosphorus atoms per CcO monomer was 12.9 (Table 4) [14]. Gas chromatographic analysis of the lipids extracted from CcO revealed that the total number of fatty acyl groups, except for arachidonate, per CcO monomer was 30.6 [14]. However, the number of phosphorus atoms includes not only those in phospholipids but also residues phosphorylated by PTM; numerous residues have been reported as possible phosphorylation sites [15,20]. Furthermore, as shown by the electron density, most lipids are not bound to all of the CcO molecules, i.e., lipids do not fully occupy specific sites. Therefore, it is difficult to estimate the exact number of lipid-binding sites on the CcO molecule. Our current crystallographic analysis permitted the identification of binding sites for two PGs, one PE, and three CLs on the CcO monomer, indicating that each CcO monomer contains nine phosphorus atoms and 18 fatty acyl groups, assuming full lipid occupancy. Because several lipid-like low electron densities were also found in the current map, the CcO molecule may contain more binding sites where lipids are weakly bound. Biochemical analysis revealed that the relative ratio of phospholipid contents was PG:PE:CL:PC = 2:1.3:3.9:1.0 (Table 5), indicating that, in addition to the crystallographically identified phospholipids, one more of each of CL, PE, and PC is possibly contained in CcO, or that some more of these phospholipids partially occupy several binding sites on CcO. Furthermore, phosphatidylserine (PS), phosphatidylinositol (PI), and phosphatidylglycerol phosphate (PGP) were also identified by mass spectrometric analysis [15]. Therefore, the weakly bound lipids not determined for the structure in this study may also include PS, PI, PGP, and TG.

Table 5.

Relative amounts of phospholipids in crystalline bovine heart CcO.

Phospholipid	Sample 1	Sample 2	Ave.
PG	2	2	2
PE	0.9	1.6	1.3
CL	4.6	3.2	3.9
PC	1.1	0.9	1.0
Total phospholipids	8.6	7.8	8.2
Total phosphorus atoms	13.2	11.0	12.1

The values indicate the relative ratios of the amounts of phospholipids and phosphorus atoms, in which the amount of PG is assumed to be 2.

Previous biochemical and mass spectrometric studies have indicated numerous sites of PTM, including acetylation and phosphorylation, in bovine CcO [15,20]. Most of these were identified in the peripheral subunits, especially subunits IV, Va, and Vb. As described above, our mass spectrometric measurements revealed one acetylation site in each of subunits VIb and VIc. Among the reported PTM sites, the acetylated lysine residues K77^III, K57^Vb, and K68^Vb were located at the dimer interface. To investigate the roles of these residues in dimerization, we analyzed the electron density around these residues. Although the acetyl groups are expected to be located at the termini of K77^III, K57^Vb, and K68^Vb, the electron densities in these regions were unclear (Fig. 9), and these lysine residues displayed high temperature factors (Table 6) that indicate flexible conformations. In the vicinities of these residues, we observed no obvious electron density corresponding to an interacting molecule. These results suggest that the acetylation sites do not specifically or tightly interact with lipids contained between the CcO monomers, although a non-specific interaction (e.g., an electrostatic interaction) between the acetylation sites and lipids remains possible. We also analyzed the electron densities for other known PTM residues and the anomalous difference electron densities for the phosphorus atoms in known phosphorylated residues (Table 6) [15,20]. Clear electron densities for the formyl groups were observed at the N-terminal methionines of both subunits I and II. However, acetylation of Ala1^VIb could not be confirmed owing to the poor electron density of the N-terminal segment; likewise, the acetylation site in subunit VIc could not be identified. In other possible PTM sites, some residues exhibited clear electron densities and low temperature factors (Table 6), but we observed no clear electron density for PTM groups or anomalous peaks for the phosphorus atoms of phosphoryl groups, indicating a very low level of PTMs in the crystallized CcO.

Fig. 9.

Structures of possible PTM residues at the dimer interface. The structures of potentially acetylated lysine residues (K77 in subunit III, K57 in subunit Vb, and K68 in subunit Vb) are represented by stick models. Each CcO monomer is shown in gray or green. The 2F_O–F_C electron density map at the 2.0σ level is represented by a blue cage. View from the membrane.

Table 6.

Temperature factors of residues known to undergo post-translational modification.

Subunit	Residue		PTMa	Atomb	Temperature factorc
I	MET	1	F	CF	38	38
	MET	1	F	OF	34	36
	SER	115	P	Oγd	19	26
	SER	115	P	Oγd	24	28
	SER	116	P	Oγ	27	32
	TYR	304	P	Oη	21	22
II	MET	1	F	CF	86	32
	MET	1	F	OF	87	30
	SER	126	P	Oγ	38	44
	LYS	171	A	Nζ	20	25
	TYR	218	P	Oη	33	61
III	LYS	77	A	Nζ	65	76
IV	LYS	7	A	Nζ	79	–e
	SER	8	P	Oγ	36	–e
	TYR	11	P	Oη	42	50
	LYS	31	A	Nζ	77	97
	SER	36	P	Oγ	35	78
	LYS	43	A	Nζ	39	70
	LYS	45	A	Nζ	50	65
	SER	47	P	Oγ	34	60
	SER	50	P	Oγ	38	60
	SER	52	P	Oγ	31	70
	LYS	63	A	Nζ	35	47
	SER	67	P	Oγ	24	43
Va	THR	17	P	Oγ1	39	42
	LYS	21	A	Nζ	37	43
	LYS	31	A	Nζ	36	38
	THR	35	P	Oγ1	27	29
Vb	LYS	26	A	Nζ	83	78
	LYS	37	A	Nζ	88	89
	SER	40	P	Oγ	39	36
	LYS	43	A	Nζ	86	73
	THR	53	P	Oγ1	43	30
VIa	LYS	58	A	Nζ	35	41
VIb	SER	51	P	Oγ	81	75
	SER	70	P	Oγ	42	44
VIc	LYS	6	A	Nζ	44	54

^aPossible PTMs reported in [15,20]. F: formylation; P: phosphorylation; A: acetylation.

^bC^F and O^F indicate the formyl carbon and oxygen atoms, respectively. The other cases indicate the atoms that may bind PTM groups.

^cEach value is the temperature factor of the indicated atom in each monomer of dimeric CcO.

^dThe atoms in two conformers of 0.5 occupancy.

^eThe structural models were not constructed owing to poor electron density.

2.5. Structural analysis of dimerization sites

The dimer interface of CcO consists of the transmembrane surface and a small region of extramembrane surface on the mitochondrial intermembrane side (Fig. 5). Most of the transmembrane interface is mediated by interactions among detergents and lipids weakly bound between the CcO monomers, as described above. In addition, two tightly bound cholate molecules (CH1 and CH2) and hydrophobic interactions between the transmembrane helices form the transmembrane dimer interface (Fig. 10A and B). CH1 was stabilized at the dimer interface by five hydrogen bonds with one monomer and one hydrophobic (aromatic–aliphatic) interaction (3.7-Å distance) with the side chain of Leu127^III in the other monomer. CH2 was stabilized by two hydrogen bonds with Glu62^II and Thr63^II and one hydrophobic (aromatic–aromatic) interaction (4.1-Å distance) with Trp275^I in one monomer, and four hydrogen bonds with Arg14^VIa and Arg17^VIa and one hydrophobic (aromatic–aromatic) interaction (4.1-Å distance) with Phe18^VIa in the other monomer (Fig. 10C). Previous structural analysis of monomeric CcO revealed that CH1 was still stabilized by five hydrogen bonds, whereas CH2 was completely dissociated from CcO [13]. Together with the biochemical observation of the effect of cholate on the dimerization (Fig. 1), these findings imply that the binding of CH2 triggers the dimerization. As depicted in Fig. 10C, the hydrophobic interactions between the transmembrane helices involve four leucine residues of subunit VIa (L37, L33, L30, and L23) and residues of subunits I and II (L78^II, I74^II, I311^I, and F282^I) in another monomer. The distances involved in these hydrophobic interactions were 4–5 Å, and the electron densities were less clear at the upper part than at the bottom part in the vicinity of CH2. In addition to these hydrophobic interactions, a hydrogen bond was formed between Thr15^VIa and Tyr179^I in another monomer, whose electron densities were clear. The above interactions may play a role in controlling the orientation of each monomer at the dimer interface.

Fig. 10.

(A) CH1 site and (B) CH2 site. The cholate structure is shown in dark gray. The 2F_O–F_C electron density map at the 1.0σ level is represented by a blue cage. The molecular surface of each CcO monomer is shown in gray or green. The hydrocarbon tails of the lipid and DM are colored yellow and orange, respectively. (C) The transmembrane helices of each CcO monomer are shown in gray or green. The carbon, oxygen, and nitrogen atoms of the amino acid residues and CH2 at the dimer interface are colored purple-blue, red, and blue, respectively. Heme a₃ is represented by a magenta stick model. The hydrocarbon tails of the lipid and DM are colored yellow and orange, respectively.

In the extramembrane region, the VIb subunits of each monomer face one another (Fig. 11A). In functional terms, this subunit may be involved only in dimerization, as it is located far from the catalytic center, the cytochrome c binding site, and the electron- and proton-transfer pathways. As shown in Fig. 11B, the backbones of the loop segments of (Lys46–Gly47–Gly48–Asp49)^VIb in the VIb subunits of dimeric CcO were situated parallel to each other to form the dimer interface. The side chain of Lys46 could be modeled within hydrogen-bonding distance of the side chain of Asp49 in the other VIb subunit. However, the electron densities of these residues were relatively low, indicating that the loop conformation is not rigid and that the interactions between these loop segments are not strong. On the other hand, clear electron density was observed for Met43 near the loop segments. The distance between the sulfur atom of Met43 and the corresponding atom in the other monomer was 4.0 Å, indicating that the dimer interface is partially formed by a Met–Met interaction, although this interaction is weaker than a typical hydrogen bond [21]. Together, these results indicate that the VIb subunits are not tightly bound to each other; however, the parallel backbone structure may play a role in correcting the orientation of the two monomers. The interactions involved in dimerization are summarized in Table 7.

Fig. 11.

Dimer interface structure of the extramembrane domain. The structures of Lys46, Met43, and Asp49 in subunit VIb of both monomers are represented by stick models. The molecular surface of each CcO monomer without subunit VIb is shown in gray or green. The 2F_O–F_C electron density map at the 1.5σ level is represented by a blue cage. (A) View from the membrane. (B) View from the intermembrane space.

Table 7.

Amino acid residues involved in dimerization.

Subunit	Residuea		Conservationb
Residues interacting with another monomer
I	TYR	179	15/15
	PHE	282	15/15
	ILE	311	15/15
II	ILE	74	13/15
	LEU	78	15/15
VIa	THR	15	15/15
	LEU	23	15/15
	LEU	30	14/15
	LEU	33	14/15
	LEU	37	13/15
VIb	MET	43	15/15
Residues interacting with CH2
I	TRP	275	15/15
II	GLU	62	15/15
	THR	63	14/15
VIb	ARG	14	4/15
	ARG	17	15/15
	PHE	18	3/15

^aBovine residue names and numbers.

^bConservation of the residues among the following 15 mammalian species: Bos taurus (bovine), Physeter macrocephalus (sperm whale), Equus caballus (horse), Felis catus (cat), Phoca vitulina (harbor seal), Manis pentadactyla (Chinese pangolin), Artibeus jamaicensis (Jamaican fruit bat), Mus musculus (mouse), Rattus norvegicus (rat), Homo sapiens (human), Microcebus murinus (gray mouse lemur), Dasypus novemcinctus (nine-banded armadillo), Choloepus didactylus (southern two-toed sloth), Loxodonta africana (African elephant), and Vombatus ursinus (common wombat). The amino acid sequence alignments are shown in Fig. S2.

According to the mass spectrometric results, in the mitochondrial membrane, CcO exists as three states: monomer, dimer, and a supercomplex with respiratory complexes I and III [5]; however, it is difficult to estimate the quantitative ratio of these three states from the results. Our analysis of solubilized bovine heart membrane with several types of detergent revealed that, with the exception of cholate, the proportion of CcO in the dimeric form was markedly lower than the proportions in the monomeric and supercomplex forms (Fig. 1). The dimeric form is also not well known in the mitochondria of any other species or organs. Previous biochemical analyses demonstrated that dimerization was induced by the addition of cholate or its analogs [deoxycholate and 3-[(3-cholamidopropyl)dimethylammonio]-1-propanesulfonate (CHAPS)] into the CcO solution [22,23] and when CcO was reconstituted into mixed-lipid liposomes [24]. Thus, it is suggested that, in the mitochondrial membrane, dimerization may be transiently triggered and stabilized by some specific physiological condition. Our current work indicates that the structural basis of dimer formation is as follows: numerous detergent and lipid molecules are weakly bound at the dimer interface, which also consists of hydrogen bonds to tightly bound CH1 and CH2, hydrophobic interactions involving the transmembrane helix of subunit VIa, and the Met–Met interaction between the extramembrane regions of subunit VIb of both CcO monomers. Accordingly, the binding of physiological ligands structurally similar to cholate could trigger the dimerization in the mitochondrial membrane, in which case the binding of non-specific lipid molecules at the transmembrane surface between monomers would support the stabilization of the dimer. The interactions among the transmembrane helices and extramembrane regions may play a role in correctly orienting each monomer within the dimer.

On the basis of this proposed mechanism, we examined whether well-known physiological molecules possessing a similar molecular size or shape to cholate could function as alternatives to cholate. Thus, ATP, ADP, or cholesterol (final concentration: 0.2–2.0 mM, 0.2–2.0 mM, or 1–500 μM, respectively) was added to the mitochondrial fractions during solubilization with LMNG or DM; alternatively, these compounds were separately incubated with the solubilized fraction. No obvious effect indicating stabilization of the dimeric form was observed for any of the molecules tested (Fig. S3). These results also suggest that the dimeric form is relatively unstable even if these molecules trigger dimerization, and that dimerization in the mitochondrial membrane may occur under some specific condition that has yet to be identified. A relatively unstable and transient dimerization mechanism could possibly contribute to the regulation of respiratory activity according to changes in physiological conditions, because it would enable transition between the monomeric form with higher activity and the dimeric form with lower activity [13].

3. Conclusion

In this study, the resolution of the crystallographic structure of CcO was improved to 1.3 Å by optimizing the method for cryoprotectant soaking. The improved electron density map revealed numerous weak densities for detergent and lipid molecules on the transmembrane surface between monomers that were not identified in the previous electron density map at 1.5-Å resolution. The current study has also elucidated the structural basis underlying the dimerization of CcO, including interactions involving the hydrophobic tails of non-specific lipids (detergents), hydrophobic interactions between the transmembrane helices, and a Met–Met interaction between the extramembrane regions. The cholate molecule tightly bound at the dimer interface suggests that the binding of physiological ligands structurally similar to cholate could trigger dimerization in the mitochondrial membrane. On the other hand, the current results do not support the involvement of CL, the N-terminal segment of subunit VIa, or the PTM sites in the dimerization. The weak and non-specific interactions of the lipid tails and hydrophobic residues may enable a reversible transition between the monomer and dimer and contribute to enzymatic regulation.

4. Materials and methods

4.1. Purification of mitochondrial CcO from bovine heart

Cytochrome oxidase was purified from bovine heart according to the protocol described below. The temperature was maintained at 0–4°C to the extent possible. After careful removal of fat and connective tissues, bovine heart muscle was minced to prepare 1,100 g of minced tissue, and a 550-g portion was suspended in 2,750 mL of 20 mM Na-Pi buffer (pH 7.4) at 0°C and homogenized for 10 min at 13,000 rpm in a homogenizer (Nihon Seiki), followed by centrifugation for 20 min at 2,200 × g in a large-scale refrigerated centrifuge (Kubota Model 9810) equipped with an RS-6600 rotor. The precipitate was suspended in 1,500 mL of 20 mM Na-Pi buffer (pH 7.4) and re-homogenized, followed by centrifugation as described above. The other 550-g portion was also subjected to the same procedure. The combined supernatants were treated with 30% acetic acid to decrease the pH to 5.15, followed by centrifugation for 15 min at 2,200 × g. The precipitates were collected and homogenized, suspended in 3,000 mL of water, and centrifuged under the same conditions. The precipitates were collected in 100 mL of 100 mM Na-Pi buffer (pH 7.4) and homogenized for 3 min at 5,000 rpm in a homogenizer (Nihon Seiki). The homogenate was centrifuged for 10 min at 2,100 × g to eliminate bubbles, and then the volume was measured. An additional 100 mL of 100 mM Na-Pi buffer (pH 7.4) was used to rinse the homogenizer cup and centrifuge tubes. The total volume of the mitochondrial membrane fraction (Keilin–Hartree submitochondrial particles) was adjusted to 576 mL with 100 mM Na-Pi buffer (pH 7.4) and stored overnight at 0°C.

Fifty milliliters of 40% (w/v) sodium cholate (Dojindo, the highest grade) was added to the mitochondrial membrane fraction, and then ammonium sulfate was added to 33% saturation at 0°C with stirring (30 min). During this process, the pH of the mixture was maintained between 7.3 and 7.4 by the addition of 3 M NaOH. The volume of the supernatant obtained after centrifugation for 20 min at 25,000 × g was measured, and ammonium sulfate was added to 50% saturation. For the precipitation of proteins including CcO, ammonium sulfate was added without checking the pH. After centrifugation for 25 min at 25,000 × g, the precipitate was dissolved in 100 mM Na-Pi buffer (pH 7.4) containing 0.5% (w/v) sodium cholate, and the volume was adjusted to 220 mL (i). This solution was dialyzed (dialysis membrane pore size: 14,000) against 3,000 mL of 40 mM Na-Pi buffer (pH 7.4) for 90 min, followed by centrifugation at 100,000 × g for 40 min. The precipitate was re-solubilized in 100 mM Na-Pi buffer (pH 7.4) containing 2.0% (w/v) sodium cholate using a glass/Teflon Potter homogenizer, the volume was adjusted to 200 mL (ii), and ammonium sulfate was added to 25% saturation. The pH was maintained between 7.3 and 7.4 by the addition of 3 M NaOH, and the precipitate solution was kept at 0°C for 20 min. The solution was centrifuged at 35,000 × g for 10 min. The ammonium sulfate saturation of the supernatant was increased to 45%, followed by centrifugation at 35,000 × g for 10 min. The precipitate was dissolved in 200 mL of 100 mM Na-Pi buffer (pH 7.4) containing 0.5% (w/v) sodium cholate (iii). Ammonium sulfate fractionation at between 25% and 40% saturation was then performed as described above. The resultant precipitates were then dissolved in 130 mL of 100 mM Na-Pi buffer (pH 7.4) containing 0.5% (w/v) sodium cholate (iv). Ammonium sulfate fractionation at between 25% and 35% saturation was performed as described above. The resultant precipitates were then dissolved in 200 mL of 100 mM Na-Pi buffer (pH 7.4) containing 0.34% (w/v) DM (v), and the detergent was switched from sodium cholate to DM. Ammonium sulfate fractionation at between 40% and 60% saturation was performed as described above, except that the 20-min incubation step was omitted. The resultant precipitates were dissolved in 100 mL of 100 mM Na-Pi buffer (pH 7.4) containing 0.2% (w/v) DM (vi) and stored overnight at 0°C.

After adjusting the volume to 220 mL with the same buffer containing 0.2% DM, ammonium sulfate fractionation at between 50% and 70% saturation was performed. The resultant precipitates were dissolved in 200 mL of 100 mM Na-Pi buffer (pH 7.4) containing 0.2% (w/v) DM (vii), and ammonium sulfate fractionation at between 55% and 70% saturation was performed. The resultant precipitates were dissolved in 10–15 mL of 100 mM Na-Pi buffer (pH 7.4) containing 0.2% (w/v) DM (viii) and then dialyzed against 10 mM Na-Pi buffer (pH 7.4; replaced three times: the first liter for 1 h, the second liter for 2 h, and the third liter for 3 h, for a total of 6 h) (ix).

After dialysis, the sample was centrifuged at 35,000 × g for 20 min and the supernatant was concentrated to 5 mL with an Amicon Diaflow apparatus using an ultrafiltration membrane (Advantec) with a pore size of 200,000. Twenty-five milliliters of 20 mM Na-Pi buffer (pH 7.4) containing 0.2% (w/v) DM was added to the concentrated sample, which was again concentrated to 5 mL (x). Twenty-five milliliters of 40 mM Na-Pi buffer (pH 6.8) containing 0.2% (w/v) DM was then added, and the sample was again concentrated to 5 mL (xi). This step was performed three times (xii, xiii). After the third (final) buffer addition, the sample was concentrated to a volume no greater than 3 mL; the protein concentration at this point was 140 mg/mL or higher. When the sample was concentrated in this manner, micro-crystals formed to afford a cloudy solution, with a large quantity of micro-crystals on the ultrafiltration membrane. The solution containing these micro-crystals was transferred to a centrifuge tube and centrifuged for 5 min at 35,000 × g to yield crystals and mother liquor. The micro-crystals on the ultrafiltration membrane were dissolved in 1 mL of 60 mM Na-Pi buffer (pH 6.8) containing 0.2% (w/v) DM. These crystalline samples were transferred to a centrifuge tube containing the collected crystals. In addition, 0.5–0.8 mL of solution was used to dissolve all of the remaining micro-crystals on the ultrafiltration membrane, and the crystalline samples were transferred to a centrifuge tube. The crystals collected in the centrifuge tube were completely dissolved in 60 mM Na-Pi buffer (pH 6.8) containing 0.2% (w/v) DM and then centrifuged for 5 min at 35,000 × g and the insoluble material was removed. The dissolved crystalline sample was transferred to another container and the volume was recorded. The concentration of the Na-Pi buffer was determined on the basis of the volume of 60 mM Na-Pi buffer (pH 6.8) containing 0.2% (w/v) DM that had been added and the quantity of crystals in the 40 mM Na-Pi buffer (pH 6.8) containing 0.2% (w/v) DM. The concentration of CcO was determined (ε^{red(604–630)}=46.6 mM⁻¹cm⁻¹) from the absorption spectra of the reduced enzymes.

4.2. Mass spectrometry

To prepare the sample for mass spectrometry, 1 μM crystalline bovine cytochrome c oxidase was mixed with a matrix solution containing 10 mg/mL sinapinic acid, 0.1% trifluoroacetic acid, and 30% acetonitrile in a ratio of 1:7. Three microliters of this sample solution was loaded onto the sample plate and dried prior to measurement. The measurement was performed using a MALDI-TOF mass spectrometer (Voyager DE PRO system).

4.3. Determination of the contents of cholate molecules and phosphorus atoms per CcO

The amount of cholate per CcO in each purification step was calculated using the ¹⁴C-labeled internal standard cholate and the concentration of CcO. Radioactivity was measured with an Aloka LSC-701 liquid scintillation spectrometer

The phosphorus content of CcO stabilized by the detergent was analyzed directly without the need for extraction with an organic solvent, as previously described [25,26].

4.4. Determination of the relative amounts of phospholipid molecules in CcO

The phospholipids were extracted from the CcO solution with DM and isolated by thin-layer chromatography on silica gel plates. The amount of phosphorus present in the silica gel was analyzed for each phospholipid as previously described [14]. In the case of 3OM, this experiment failed because the 3OM prevented adequate separation of the phospholipids during the chromatography step.

4.5. Crystallization of dimeric CcO

Na-Pi buffer at a concentration of 40 mM, DM at a concentration of 0.2%, EG at a concentration of 2%, and protein at a concentration of 70–95 mg/mL were used to crystallize the crystalline sample. Twenty percent (w/v) PEG 4k (Merck, polyethylene glycol 4000 for gas chromatography) dissolved in 40 mM Na-Pi buffer (pH 6.8) containing 0.2% (w/v) DM was added to the enzyme solution. The resultant mixture was shaken well in a tube with a rounded bottom. Batch crystallization was then performed by placing the solution on a concave microscope slide (depression, 50–120 µL), which was covered with a coverslip to prevent air entering and left to stand at 4°C.

4.6. Cryoprotection to prevent ice crystal formation

This step was deemed optimal as described in the Results and Discussion section. Solutions were prepared as follows to serve as cryoprotectants. Two buffers were prepared: a replacement buffer for replacement of the mother liquor [40 mM Na-Pi buffer (pH 6.5) containing 0.2% (w/v) DM, 2% (v/v) EG (Hampton Research), and 1% (w/v) PEG 4k] and a cryoprotectant buffer to prevent the formation of ice crystals during freezing [40 mM Na-Pi buffer (pH 5.7) containing 0.2% (w/v) DM, 40% (v/v) EG, and 8% (w/v) PEG 4k]. These two solutions were mixed in 50 different ratios, i.e., 1:49, 2:48, 3:47, …, 50:0.

Following crystal precipitation, the coverslip was removed from the concave microscope slide, the mother liquor was removed to the maximum degree possible, and replacement buffer was promptly added. The mother liquor around the crystals was removed to the extent possible by replacing the solution a number of times. The depression in the microscope slide contained 500 μL of replacement buffer when using a microscope slide with a depression of 120 μL. After the removal of 250 μL of the solution, 250 μL of the first batch of solution for preventing ice crystal formation (as described in the previous paragraph) was added to the depression. The two solutions were slowly mixed and the mixture was left to stand for 20 min. Next, 250 μL of the mixture was removed, and 250 μL of the second batch of solution was added. This mixture was again left to stand for 20 min. These steps were repeated until the addition of the 50th batch (i.e., cryoprotectant buffer only). This procedure typically required 2–3 d. The 50th batch of solution was replaced twice. These steps were performed in a cold room at 4°C. After finishing the soaking treatment with the 50th batch of solution, the crystals were removed with a cryo-loop and flash-cooled in a stream of liquid nitrogen. The frozen crystals were stored in liquid nitrogen.

4.7. X-ray crystallographic analysis

X-ray diffraction measurements were performed at SPring-8 using the BL44XU beamline with the MX225HE CCD detector (Rayonix), or the EIGER X 16M detector (Dectris) for the anomalous diffraction data. To evaluate the resolution, the X-ray beam entered parallel (0°) to the crystal b axis, and after translation and rotation of the crystal, the beam entered perpendicular (90°) to the b axis. The maximum resolution of the diffraction spots was determined by observing the diffraction images. To evaluate the mosaicity and unit cell parameters, the X-ray beam entered the crystal at an angle that was varied from 0° to 90° in a stepwise manner. To reduce the effect of X-ray radiation damage, the crystal was translated two times at rotation angles of 30° and 60°. The experimental conditions are summarized in Table 8. The mosaicity and unit cell parameters were obtained by processing the diffraction data using the Denzo and Scalepack software [26].

Table 8.

Experimental conditions for the X-ray diffraction measurements.

	Evaluation of the resolution	Evaluation of the mosaicity and b-axis length
Wavelength (Å)	0.9	0.9
Beam size (μm)	50 × 50	50 × 50
Temperature (K)	50	50
Resolution range (Å)		200–2.00
Crystal–detector distance (mm)	116	230
Exposure period (s)	10	2.5
Oscillation angle (degree)	0.25	0.75
Total angular range (degree)	0–1.0, 90.0–91.0	0–90.0
Translation (μm)	100	100
Number of crystals used	1	1
Number of images	8	120

For the collection of high-resolution data, diffraction images up to 1.30-Å resolution were measured. Because the intensity of the diffraction spots in the low-resolution range exceeded the detector limit, diffraction images up to 2.00-Å resolution were also measured by decreasing the exposure period. To minimize the effect of X-ray radiation damage, the crystal was translated after each X-ray exposure. For the collection of the data set containing anomalous diffraction data, an X-ray beam with a wavelength of 1.75 Å was used. The experimental conditions for the data collection are summarized in Table 9. The data sets were processed using the Denzo software and merged using the Scalepack software [27]. The structure-factor amplitude (F_O) was calculated using the CCP4 program TRUNCATE [28]. For structural refinement with the high-resolution data, the atomic coordinates of CcO in the fully oxidized state (PDB ID: 5B1A) were used as the initial model. The refinement was performed using the program Refmac5 [29] with anisotropic B-factors. The statistics of the diffraction data and refinement are summarized in Table 3.

Table 9.

Experimental conditions for the X-ray diffraction measurements.

	High-resolution data		Anomalous data
Wavelength (Å)	0.9	0.9	1.75
Beam size (μm)	50 × 50	50 × 50	ϕ70
Temperature (K)	50	50	60 or 90
Resolution range (Å)	200–2.00	200–1.30	200–1.79
Crystal–detector distance (mm)	230	131	120
Exposure period (s)	1	10	0.3
Oscillation angle (degree)	0.75	0.25	0.1
Translation (μm)	1	10	0.1
Number of crystals used	3	41	12
Number of images	345	1,273	86,400

Data availability

The structure factor at 1.3-Å resolution and the atomic coordinates of the structural model are deposited at the Protein Data Bank (PDB, www.rcsb.org) under the accession code 7COH.

Funding and additional information

This work was supported by JSPS KAKENHI grants JP17H03646 (to K. S.-I.) and JP22370060 and JP18K06162 (to K. M.). Diffraction data were collected at the Osaka University beamline BL44XU at SPring-8 (Harima, Japan) (Proposal Nos.: 2011B6500, 2012A6500, 2018A6854, 2018B6854, 2019A6954, 2019B6954, and 2020A6552).

Declaration of Competing Interest

The authors declare that they have no conflict of interest.