X-ray structures of catalytic intermediates of cytochrome c oxidase provide insights into its O2 activation and unidirectional proton-pump mechanisms

Cytochrome c oxidase (CcO) reduces O2 to water, coupled with a proton-pumping process. The structure of the O2-reduction site of CcO contains two reducing equivalents, Fea32+ and CuB1+, and suggests that a peroxide-bound state (Fea33+–O−–O−–CuB2+) rather than an O2-bound state (Fea32+–O2) is the initial catalytic intermediate. Unexpectedly, however, resonance Raman spectroscopy results have shown that the initial intermediate is Fea32+–O2, whereas Fea33+–O−–O−–CuB2+ is undetectable. Based on X-ray structures of static noncatalytic CcO forms and mutation analyses for bovine CcO, a proton-pumping mechanism has been proposed. It involves a proton-conducting pathway (the H-pathway) comprising a tandem hydrogen-bond network and a water channel located between the N- and P-side surfaces. However, a system for unidirectional proton-transport has not been experimentally identified. Here, an essentially identical X-ray structure for the two catalytic intermediates (P and F) of bovine CcO was determined at 1.8 Å resolution. A 1.70 Å Fe–O distance of the ferryl center could best be described as Fea34+ = O2−, not as Fea34+–OH−. The distance suggests an ∼800-cm−1 Raman stretching band. We found an interstitial water molecule that could trigger a rapid proton-coupled electron transfer from tyrosine-OH to the slowly forming Fea33+–O−–O−–CuB2+ state, preventing its detection, consistent with the unexpected Raman results. The H-pathway structures of both intermediates indicated that during proton-pumping from the hydrogen-bond network to the P-side, a transmembrane helix closes the water channel connecting the N-side with the hydrogen-bond network, facilitating unidirectional proton-pumping during the P-to-F transition.

It is well-known that a one-electron reduction of O 2 is energetically unfavorable, whereas a simultaneous two-electron reduction is energetically favorable (5). This chemical property of O 2 is critical for the stability of the O 2 -bound form of hemoglobins and myoglobins, as the second electron is not readily available to the heme site (Fe 2ϩ -O 2 ) embedded within the globin moiety. O 2 binds to the O 2 -reduction site of CcO, which includes two metal sites, heme a 3 (Fe a3 ) and Cu B , only when both metal sites are in the reduced state, as in the R-form in Fig. 1A (6). The location of the two metal ions in the O 2 -reduction site of CcO strongly suggests that the initial intermediate in the O 2 -reduction process of CcO is a bridging peroxide, Fe a3 3ϩ -O Ϫ -O Ϫ -Cu B 2ϩ (5). However, extensive time-resolved resonance Raman analyses of the O 2 -reduction process have revealed that the initial intermediate is Fe a3 2ϩ -O 2 and that the second intermediate includes a ferryl oxide state (Fe a3 4ϩ ϭO 2Ϫ ), indicating that the O-O bond is broken at this stage (7). To resolve this enigma, stable Fe a3 2ϩ -O 2 and unstable Fe a3 , it is necessary to determine X-ray structures of the active sites of these intermediate species at high resolution.
The existence of a proton-conducting pathway, which is known as the H-pathway, marked by the blue and red arrows drawn on the far-left of Fig. 1B, has been revealed by X-ray structures of bovine heart CcO in various oxidation-and inhibitor-binding states (1,(8)(9)(10)(11). This pathway is composed of a hydrogen-bond network (red) and a water channel (blue) through which water molecules on the N-side gain access to the bottom end of the hydrogen-bond network, which extends to the P-side surface of the CcO molecule, as described in Fig. 1B, inset. Thus, the protons to be pumped are transferred from the N-side as hydronium ions via the water channel to the hydrogen-bond network. The hydrogen-bond network is attached to heme a (Fe a ) near its bottom end via two hydrogen bonds to the propionate group and the formyl group of the porphyrin ring of heme a, as shown in Fig. 1B (1,8,9). This structure suggests that electrostatic repulsions between the protons on the hydrogenbond network and the net positive charges created upon heme a oxidation drive the process of pumping the protons to the P-side (1,8,11).
The hydrogen-bond network of the H-pathway is attached to a proton storage area, indicated by the blue background in Fig.  1B, via a short hydrogen-bond network, indicated by the gray area in Fig. 1B (11). The proton storage area is provided by a large water cluster for at least four proton equivalents. The protons to be pumped are stored transiently in the proton storage area before oxidation of heme a. This system is necessary for effective collection of the protons to be pumped from the N-side, because proton collection could be suppressed by the net positive charges on heme a.
The tight interaction between heme a and the hydrogenbond network, which is required to drive the protons to the P-side, has recently been confirmed by a high-resolution X-ray structural analysis of the azide-bound CcO (10). However, electrostatic repulsion itself does not provide directionality to the movement of protons after repulsion. Rather, an effective system that blocks back-leakage of protons from the electrostatic repulsion site is required to provide efficient energy transduction.
Narrowing of the water channel by elimination of one of the largest cavities, including Met 383 near the P-side end of the water channel, was identified by comparison of X-ray structures of CcO forms in various oxidation-and inhibitor-binding states (1,(8)(9)(10)(11). The narrowing suggests that there is effective blockage of proton exchange between the hydrogen-bond network and the N-side, which could facilitate unidirectional proton transfer, if it occurs during the catalytic cycle (1,(8)(9)(10)(11). The structures of the water channel with and without this cavity are designated as the open state and the closed state, respectively. This unidirectional mechanism is derived from X-ray structures of stable CcO forms not involved in the catalytic turnover (1,(8)(9)(10)(11). Thus, the mechanism remains a proposal. The catalytic intermediate forms of CcO are not easily stabilized sufficiently for crystallization. A preliminary analysis of the electron density of some intermediate forms has been presented (9) but has not yet been completely refined.
We performed X-ray crystal structural analyses using crystals of a roughly equimolar mixture of two catalytic intermediate forms, the P-form and the F-form, prepared by H 2 O 2 treatment of the crystalline bovine heart CcO. The X-ray structural results show that the water channel of both forms is in the closed state, confirming experimentally the above-proposed unidirectional mechanism during the P-to-F transition.
An interstitial water molecule hydrogen-bonded to Tyr 244 was identified in the O 2 -reduction site of the intermediates. This water molecule can trigger a rapid proton-coupled electron transfer from Tyr 244 -OH to the bridging peroxide, which appears prior to the P-formation, giving a negligible level of the bridging peroxide form, consistent with the resonance Raman data reported almost 30 years ago (1,7).

Preparation of crystals of the intermediate forms of bovine heart CcO by treating crystals of the resting oxidized form with H 2 O 2 at pH 5.7
Crystals of the P-and F-forms were prepared using a method similar to that used to prepare these forms in solution from the resting oxidized form (7,12,13). The bovine heart CcO as purified is in the fully-oxidized state. The fully-oxidized preparation is known as the "resting oxidized form," because it has no proton-pumping activity, in contrast to the O-form, which is generated during catalytic turnover (Fig. 1A). It has been shown that the resting oxidized form is in a peroxide-bound fullyoxidized state (14). Treatment of the crystalline resting oxidized form at pH 5.7 with 10 mM H 2 O 2 in the presence of 50 M phenazine methosulfate (PMS) results in the gradual appearance of two bands at 607 and 580 nm in UV-visible spectra at 4°C at pH 6.0, as shown in Fig. 2A. With addition of H 2 O 2 to a concentration of 100 mM in the presence of 100 M PMS, the 607-nm band increases in intensity for 24 min and then gradually decreases, as shown in Fig. 2B. Further details in the reaction of crystalline resting oxidized CcO with H 2 O 2 is given in Text S1. The role of PMS in formation of the P-form and the F-form in the crystals is discussed in Text S2.
The position of this 607-nm band is essentially identical to the position of a similar band of the reduced heme a. In contrast, in the resting oxidized form of CcO in solution at low pH and in the presence of H 2 O 2 , the 607-nm band appears at 580 nm (12). However, the 607-nm band is unlikely to be due to heme a reduction because reduction of the resting oxidized form of CcO by H 2 O 2 is energetically unfavorable (15,16). Furthermore, even if H 2 O 2 were to cause reduction of CcO, the reduced CcO would be readily oxidized by O 2 , produced by oxidation of H 2 O 2 . Lower H 2 O 2 concentrations provide a higher intensity ratio of the 607/580-nm bands as shown in Fig.  2, A and B. This tendency is inconsistent with a scenario involving reduction of heme a to provide the 607-nm band in the crystals. Thus, it appears that the 607-nm band arises from generation of the P-form of CcO. Assuming that the extinction coefficients for the P-form and the F-form determined in solution (12) are the same in the frozen crystals, the P/F ratio is estimated to be about 1 in the crystals subsequently used for X-ray diffraction experiments (Fig. 2C). Formation of the 607-nm band at low pH in the crystals suggests that protonation of the O 2 -reduction site via the K-pathway, which is required for the spectral transition from the 607-to the 580-nm band (17), is blocked in the crystals.
The spectrum measured after X-ray diffraction experiments ( Fig. 2C) indicates that there is a significant increase in intensity of the 604-nm band relative to the 580-nm band measuredbeforetheX-raydiffractionexperiment.Thisabsorbance increment must be due to reduction of heme a by hydrated electrons produced by X-ray irradiation in the X-ray diffraction experiments at SPring-8. Under frozen conditions at 100 K, the diffusion of O 2 to the O 2 -reduction site is significantly limited, and consequently the reduced CcO species is prevented from being oxidized. Even under the frozen conditions, electron transfers within the redoxactive metal sites in each CcO molecule in the crystal are not suppressed, and reduced heme a would be readily oxidized by Fe a3 or Cu B in the P-or F-form. Thus, the detectable increment in heme a reduction upon X-ray irradiation sug-  Each spectrum is a difference spectrum representing the spectral difference before and after H 2 O 2 treatment. C, absolute absorption spectra of frozen crystals treated with 50 mM H 2 O 2 in the presence of 50 M PMS, before (blue) and after (red) X-ray exposure for X-ray diffraction experiments. The spectra were measured under N 2 flow at 100 K. The small difference in the band position of the P-form (607 nm in A and B versus 604 in C) is due to the difference in the spectral measurement. The former is the difference spectra against the resting oxidized form, and the latter shows absolute spectra.

X-ray crystal structures of cytochrome oxidase intermediates
gests that the other redox-active metal sites, Fe a , Cu A , and Cu B , are also in the reduced state. No extra absorbance band other than the 604-nm band in the ␣-band region indicates formation of the ligand-free Fe a3 2ϩ . Further discussion on this conclusion is provided in Text S1.

Assessment of multiple structures detectable in the present X-ray diffraction data set
The X-ray diffraction data set was obtained from crystals of the resting oxidized form treated with 50 mM H 2 O 2 and 50 M PMS. The structural model of the resting oxidized form of CcO (PDB code 5B1A) was found to be well-superimposed on the MR/DM map as shown in Fig. 3, A-C. The three panels display the three regions where significant structural differences are detectable between the structures of the resting oxidized form and the fully-reduced form. All three regions are involved in the H-pathway (Fig. 1B). A structural refinement converged well with the structure of the resting oxidized form. Fig. 4A shows distribution of the B-factors of the mainchain portions (-NH-C␣-CO-) of the residues in subunit I of the H 2 O 2 -treated resting oxidized CcO. The high B-factor values (Ͼ30 Å 2 ) in the two regions between residues 380 and 385, including Met 383 , and between residues 48 and 55, including Gly 49 , are not detectable in the corresponding regions of the resting oxidized CcO as shown in Fig. 4, B and C, respectively. Similarly, the significantly high average B-factor values of hydroxyfarnesyl ethyl group of heme a, compared with that of the rest of the heme a molecule, are not detectable in the same group of the resting oxidized CcO. Thus, these B-factor distributions suggest that existence of either an additional structure as a minor component or that high thermal motions in these three regions induce these high B-factor values.
For examination of the existence of a minor component, the F o Ϫ F c maps against the resting oxidized CcO model were determined. In the above three regions, the F o Ϫ F c maps exhibit significant differences in electron densities, as shown in Fig. 5, A, D, and G. The positive densities (blue) in these maps are assignable to the structures of the fully-reduced form, reported previously (11), as shown in Fig. 5, B, E, and H, whereas the negative densities (red), given in Fig. 5, C, F, and I, are assignable to the structures of the resting oxidized form. The electron density differences correspond to those induced by reduction of the resting oxidized form to the fully-reduced form. Thus, the F o Ϫ F c maps strongly suggest that the high B-factor values are induced by coexistence of the fully-reduced form as a minor component. Hereafter, the structures detectable in the resting oxidized and fully-reduced forms will be designated as the oxidized-type and reduced-type structures, respectively. The region between residues 380 and 385, including Met 383 in the oxidized-type structure, provides the closed state of the water channel of the H-pathway.
For further improvements in the X-ray structural analysis, we determined the contents of the oxidized-type and reducedtype structures by searching the content of the minor component, the reduced-type structure, that yields an identical average B-factor for the oxidized-and reduced-type structures. This determination is practical for many multiple X-ray structures, because it is reasonable to assume that B-factors of the multiple structures are identical with each other. Fig. 6 gives the average B-factor differences (⌬B) between the oxidizedtype and reduced-type structures against the content of the reduced-type structure in the three regions. The monotonous changes in ⌬B indicate that the present X-ray diffraction data allow a reliable evaluation for the content of the reduced-type structure. The contents of the reduced-type structure giving ⌬B ϭ 0 are 32% for residues 48 -55; 15% for residues 380 -385; and 35% for the hydroxyfarnesyl ethyl group of heme a as shown in Fig. 6, for monomer A, which is one of the monomers of the dimer in the asymmetric unit as defined under "Experimental procedures." Essentially identical results were obtained for monomer B.
For evaluation of the above quantification for the contents of the oxidized-type and reduced-type structures, structures representing the coexisting two structures, determined by the above ⌬B analysis, were refined. The resultant B-factor values in regions 48 -55 and 380 -385 are significantly lower than those determined assuming only the oxidized-type structure and are essentially identical to those in a structure of the resting oxidized CcO (PDB code 5B1A), as shown in Fig. S1. Residual electron densities in F o Ϫ F c maps for the coexisting structures in the three regions ( Fig. S2, B, D and F) are lower than those calculated assuming only the oxidized-type structure (Fig. S2, A, C, and E). The B-factor distributions and the F o Ϫ F c maps

X-ray crystal structures of cytochrome oxidase intermediates
indicate that the quantification of the two components by the ⌬B analysis is successful.
Ignoring these unusually-high B-factor regions as detected in this work would provide an unreal structural model that could be a weighted-average structure of the two component structures. The present B-factor analysis is applicable to X-ray structural analyses for processes of functional proteins during which multiple intermediate forms appear.

X-ray structural analyses of the O 2 reduction site
In calculation of MR/DM electron density map of the O 2 -reduction site, shown in Fig. 7A, flattening of electron densities of solvent regions as high as 70% volume of the unit cell and a local noncrystallographic 2-fold symmetry averaging in the DM procedure reduced initial model bias highly efficiently. For confirming the high efficiency, a simulated annealing composite omit map (18) was calculated by including simulated annealing

X-ray crystal structures of cytochrome oxidase intermediates
procedure for the structure without ligand atoms in the O 2 -reduction site. The simulated annealing composite omit map drawn at 2.5 ( Fig. 7B) is essentially identical with the MR/DM map given in Fig. 7A. The absence of significant difference between the two maps indicates that initial model bias has been successfully removed.
MR/DM maps of the resting oxidized CcO determined at SPring-8 ( Fig. 7C) adjusted to 1.8 Å resolution and at SACLA (Fig. 7D) determined at 1.9 Å resolution are given for comparison with the present result. X-ray diffraction data obtained at SACLA are free of X-ray damage (19). The MR/DM maps of Fig. 7, A, C, and D, are drawn at 1.7 , enabling direct comparisons among them. Although each of the resting oxidized forms in Fig. 7, C and D, provides a continuous electron density between the two metal sites, indicating the existence of a bridging peroxide anion, both of the MR/DM and the composite omit maps of the H 2 O 2 -treated crystals, as shown in Fig. 7, A and B, show that electron density located between two oxygen atoms is clearly weaker than that in the resting fully-oxidized forms. The electron densities at both Fe a3 and Cu B have protrusions indicating shorter coordination bonds of metal-oxygen and longer distance between two ligand oxygen atoms (Fig.   7, A and B) than those of the two structures of resting oxidized CcO (Fig. 7, C and D).
The structure without any ligand in the O 2 -reduction site was refined, and the F o Ϫ F c map was calculated to determine the structures of the metal-liganded oxygen atoms. The positions of the ligands (Fig. 8A) are consistent with the protrusion shapes of MR/DM map for Fe a3 and Cu B (Fig. 7A). Two peaks separated by ϳ2.51 Å were assigned to the oxygen atoms of O1 and O2 coordinated to Fe a3 and Cu B , respectively (Fig. 8A).
To confirm the structural features of the two peaks in the F o Ϫ F c map, we calculated an annealing F o Ϫ F c map and a Polder map (20). The former was calculated by including simulated annealing procedure in the refinement before assignment of ligand atoms in the O 2 -reduction site and of a water oxygen atom between two propionate groups of heme a 3 (Fig.  S3A). The latter was calculated for the final stage of the structure refinement by excluding ligand oxygen atoms in the O 2 -reduction site and a water oxygen atom between two propionate groups of heme a 3 (Fig. S3B). Both electron density maps are closely similar to the F o Ϫ F c map given in Fig. 8A. The distances between two peaks of the simulated annealing F o Ϫ F c map and

X-ray crystal structures of cytochrome oxidase intermediates
the Polder map are 2.56 and 2.48 Å, respectively, which are consistent to 2.51 Å of the F o Ϫ F c map (Fig. 8A). These two maps ensure the successful determination of the F o Ϫ F c map given in Fig. 8A.
The stable water molecule bridging the two propionate groups of heme a 3 in the F o Ϫ F c map provides an excellent internal reference for quantification of the electron density of the ligands of the two metals (8 -11). The peak heights of the internal reference water molecule, O1 and O2 for monomer A, were 1.21, 1.07, and 0.89 e/Å 3 , respectively. These heights for monomer B were 1.08, 1.05, and 0.87 e/Å 3 , respectively. These results indicate that overall ligand occupancies averaged for O1 and O2 were 0.81 and 0.89 for monomers A and B, respectively, giving the average value of 0.85. The incomplete ligand saturation in heme a 3 is consistent with the average B-factor values for the pyrrole atoms of heme a 3 (26.4 Å 2 ) and significantly higher than those of heme a (23.5 Å 2 ).
Occupancy of the oxygen-bound form (0.85) determined as above is identical with the content of the oxidized-type structure for residues 380 -385 of subunit I (85%) determined by B-factor analysis as given in Fig. 6. This strongly suggests that the oxygen-bound form has the water channel in the closed state, because residues 380 -385 in the oxidized-type structure close the water channel as described above. The absorption spectrum of the oxygen-bound form, as given in Fig. 2C, indicates that the oxygen-bound form includes the P-form and F-form in a roughly 1:1 ratio. These results prove crystallographically that both the P-form and the F-form have the water channel in the closed state.
Further structure refinements were performed using the occupancies of oxygen atoms, 0.80 and 0.90, for monomers A and B, respectively. The structure with ligand oxygen atoms was refined without any restraint for the oxygen atoms. The distances between the two ligand oxygen atoms were converged to 2.34 and 2.26 Å for monomers A and B, respectively. These distances are significantly longer than the O-O distance in a typical peroxide ion and shorter than the length of a low-barrier

X-ray crystal structures of cytochrome oxidase intermediates
hydrogen bond between two oxygen atoms (21). Because the O-O distances suggest coexistence of multiple structures consisting of a low-barrier hydrogen bond and another structure, the structure with ligand oxygen atoms was refined under the restraint of an O-O distance ϭ 2.50 Å, and the F o Ϫ F c map was calculated. A residual electron density peak is detectable at the site corresponding to the peroxide in the resting oxidized form (Fig. 8B). The peak heights at the peroxide ion and the water molecule bridging the two propionate groups are 0.23 and 1.22 e/Å 3 , respectively, for monomer A. Those for monomer B are 0.37 and 1.12 e/Å 3 , respectively. The peaks of both monomers are assignable as those of the peroxide ions of the resting oxidized form, with occupancies of 0.09 and 0.17 for monomers A and B, respectively. In further refinements, the occupancies of the peroxide oxygen atoms were fixed to 0.10 and 0.15 for the monomers A and B, respectively, and those of ligand oxygen atoms were fixed to 0.70 and 0.75, respectively. Peroxide structures of both monomers derived from the respective structures of the previous high-resolution structure (PDB code 5B1A) were not changed during the further refinements, as their positional parameters fluctuate due to the low occupancies. Structure refinement without any restraints on the ligand oxygen atoms of O1 and O2 converged to Fe a3 -O1 ϭ 1.70 Å, Cu B -O 2 ϭ 2.11 Å, and O1-O2 ϭ 2.54 Å in their averaged values of two monomers. These values of the distances were used for the restraints of respective bonds in the final stage of the structure refinement to achieve convergence.
The percentages of the reduced-type structures indicated by the regions of residues 48 -55 and 380 -385 of subunit I and the hydroxyfarnesyl ethyl group of heme a were readjusted to 33, 15, and 33%, respectively, for monomer A, and 33, 18, and 35%, respectively, for monomer B. A final structure refinement Refinement was performed on a structure in the absence of a ligand in the O 2 -reduction site and without the water molecule bridging the two propionate groups of heme a 3 . Based on the refined structure, the F o -F c map was calculated. The electron density cages were drawn at 5.0 . B, F o -F c map of the two oxygen atoms bound to Fe a3 and Cu B in monomer A at 1.80 Å resolution. Refinement was performed on a structure including two oxygen ligands at Fe a3 and Cu B under a restraint condition of 2.50 Å for the distance between two oxygen atoms and without the water molecule bridging the two propionate groups and the F o -F c map was calculated. The electron density cages were drawn at 2.50 . C, structure of the O 2 -reduction site shown as a stick model colored as described for Fig. 7. The distances and bond angles shown are averaged values between two monomers. Two ligand oxygen atoms are coordinated to Fe a3 and Cu B with distances of 1.70 and 2.11 Å, respectively. The thin stick shows the location of a peroxide ion existing partially within the O 2 -reduction site of each monomer, assigned for the residual density in B. A water molecule is hydrogen-bonded to Tyr 244 . The inset shows a schematic representation of the coordination structures of Fe a3 and Cu B .

X-ray crystal structures of cytochrome oxidase intermediates
with multiple structures for these three regions, two metalliganded oxygen atoms and the two oxygen atoms of the peroxide anion converged well: R work ϭ 0.1565; R free ϭ 0.1884; RMSD (length) ϭ 0.020 Å; and RMSD (angle) ϭ 1.84°. The residual electron density at the peroxide site, as shown in Fig. 8B, disappeared.

Structural characteristics of the O 2 reduction site
The structure of the O 2 -reduction site determined in these analyses is shown in Fig. 8C, which includes a schematic representation of the coordination geometry of the Fe a3 -Cu B site in the inset. The distances and angles given are the averaged values among the two heme a 3 structures in the asymmetric unit. The Fe a3 -O1 distance is 1.70 Å. The bond angle of N-Fe-O of His 376 -Fe a3 -O1 is 175.0°. The Cu B -O2 distance of 2.11 Å indicates Cu 2ϩ -OH Ϫ coordination. The O1-O2 distance of 2.54 Å is shorter than the O-O distance of the standard hydrogen bond, indicating formation of a low-barrier hydrogen bond (22), where a hydrogen atom is dominantly located at the Cu B side. The Fe a3 -Cu B distance of 4.61 Å is the shortest distance between these ions measured to date in bovine CcO. Other distances include 4.83 Å in the resting oxidized crystal (PDB code 5B1A) and 5.11 Å in the fully-reduced crystal (PDB code 5B1B).
The Fe a3 -O1 distance of 1.70 Å is consistent with the bond length of Fe 4ϩ ϭO 2Ϫ , and not of Fe 4ϩ -OH Ϫ , for the P-form and the F-form. Meharenna et al. (23) proposed an empirical equation relating elongation of the Fe-O distance and X-ray radiation dose, ⌬d ϭ 0.2543 ϫ Y, where ⌬d represents the elongation in Å, and Y represents the radiation dose in MGy. According to this equation, elongation of the length from the original F-O bond length by 0.04 Å would be caused by the low radiation dose of 0.17 MGy attainable under our experimental conditions. Thus, the Fe a3 -O1 length, 1.70 Å, suggests that the Fe a3 -O1 distance is 1.66 Å prior to X-ray irradiation. The distance is consistent with both Fe-O bond lengths of Fe a3 4ϩ ϭ O 2Ϫ , 1.65 Å for the P-form and 1.66 Å for the F-form, estimated from the Fe-O stretching vibrations, 804 cm Ϫ1 for the P-form and 785 cm Ϫ1 for the F-form, using Badger's rule (24). Therefore, the X-ray crystal structural measurements provide strong evidence that the P-form and the F-form, identical to those under catalytic turnover (7), are generated in the crystals. It has been shown by the absorption and resonance Raman spectral analyses (12,13) that this is also the case for the CcO in solution.
The absence of an H 2 O/D 2 O exchange effect on the resonance Raman band positions of the P-form and the F-form (25) is consistent with the absence of detectable levels of Fe 4ϩ -OH Ϫ , Fe 3ϩ -OH Ϫ , and Fe 2ϩ -OH Ϫ . Badger's rule (24) suggests that the reported longer Fe-O bonds between 1.81 and 1.92 Å for ferryl oxide intermediates of various oxygen-activating heme enzymes other than CcO are likely to be induced by X-ray irradiation damage (26) Because the Cu B site is essentially insensitive to resonance Raman excitation by visible laser light (1), the Cu B 2ϩ -OH Ϫ structure had been a proposal until the present X-ray structural study. Furthermore, the present result is clearly inconsistent with ligation of H 2 O to Cu B 2ϩ , as proposed previously (27).

Channel structures for entry of substrate (O 2 ) and release of product (H 2 O)
The product (water)-releasing channel (blue) and the substrate (O 2 )-collection channel (orange), as shown in Fig. 9A, were identified using the program CAVER (28) and using a

X-ray crystal structures of cytochrome oxidase intermediates
probe radius of 0.7 Å starting from the O 2 -reduction site. The cages with a cross-section diameter of 2.4 Å indicate locations, but not accessible surfaces. The residues surrounding the substrate (O 2 )-collection channel are hydrophobic except for Glu 242 (Fig. 9B), whereas many polarized oxygen atoms, including those of the peptide carbonyls of Leu 283 , Ile 286 , Ala 308 , and Thr 309 , the hydroxyl groups of Tyr 244 , Tyr 304 , and Thr 309 , and heme a 3 formyl, provide the product (water)-releasing channel with a highly-hydrophilic environment (Fig. 9C).
The location of the open end of the product (water)-releasing channel in the transmembrane region (Fig. 9, A and C) suggests that there is effective blockage preventing back-leakage of protons from the outside of the CcO molecule. The substrate (O 2 )collection channel extends from the O 2 -reduction site to the open end facing the O 2 pool surrounded by the seven transmembrane helices of subunit III (Fig. 9A). This is consistent with the structure reported previously (29). These channel structures suggest that O 2 is transferred through the substrate (O 2 )-collection channel to the O 2 -reduction site, not through the product (water)-releasing channel, and that the water molecules produced at the O 2 -reduction site are released to the transmembrane surface via the product (water)-releasing channel. This proposal is consistent with the observation that blockage of the substrate (O 2 )-collection channel by N,NЈ-dicyclohexylcarbodiimide completely inhibits enzyme activity (29).
An interstitial water molecule hydrogen-bonded to Tyr 244 in the O 2 -reduction site is detectable in the refined structure (Fig.  8C). It is located in the water-releasing channel (Fig. 9, C and D). Fig. 9D shows a close-up view of the water molecule that indicates an accessible surface when the water molecule is released from Tyr 244 . This model was generated with a probe radius of 1.2 Å with the program VOIDOO (30). The accessible surface indicates that the closest distances of the oxygen atoms of Fe a3 4ϩ ϭO 2Ϫ (O1) and Cu B 2ϩ -OH Ϫ (O2) from the surface are 2.7 and 4.3 Å, respectively.

X-ray structures of the O 2 reduction site suggest rapid degradation of the peroxide-bound intermediate prior to formation of the P-form
The existence of the peroxide-bound form as an intermediate preceding the P-form was recently proposed based on experimental and computational studies of a bacterial CcO (31). In the intermediate preceding the P-form, it is likely that the two oxygen atoms of the bridging O 2 2Ϫ are located quite close to the two oxide ions, O1 and O2, in the X-ray structure of the P-form. The accessible surface of the interstitial water molecule described above (Fig. 9D) indicates that the interstitial water molecule is more readily accessed by O1 (the oxygen atom bound to Fe a3 ) than by O2 (the oxygen atom bound to Cu B ). Thus, the structure strongly suggests that the interstitial water molecule identified in the P/F mixture triggers a facile proton transfer from Tyr 244 to the peroxide-oxygen bound to Fe a3 3ϩ , coupled with an electron transfer from Tyr 244 -OH (or ϪO Ϫ ) to the peroxide-oxygen bound to Cu B 2ϩ via His 240 , which is cross-linked to Tyr 244 . Protonation of the peroxideoxygen bound to Fe a3 3ϩ , and not the peroxide-oxygen bound to Cu B 2ϩ , would be critical for facile oxidation of the stable Fe a3 3ϩ to yield the high-oxidation state heme iron, Fe a3 4ϩ . The proton-coupled electron transfers provide Cu B 2ϩ -O 2Ϫ , Fe a3 4ϩ -OH Ϫ , and Tyr 244 -O ⅐ radical followed by a proton exchange between two oxides through the low-barrier hydrogen bond to yield Cu B 2ϩ -OH Ϫ and Fe a3 4ϩ ϭO 2Ϫ , as schematically shown in Fig. 1A.
The highly-conserved cross-linkage between Tyr 244 and His 240 supports this proton-coupled electron transfer (1,32). The low-barrier hydrogen bond (22) between O1 and O2 in the P-form provides a stable ligand-binding arrangement between Cu B 2ϩ and Fe a3 3ϩ upon generation of the P-form from the peroxide-bound intermediate with minimal structural change. This arrangement could also contribute to the rapid and facile degradation of the bound peroxide.
The reported X-ray structure of CO-bound CcO shows that an interaction between Cu B 1ϩ and the oxygen atom of the bound CO is significantly weaker than the interaction between the carbon atom of CO and Fe a3 2ϩ (1,9). This structure indicates uneven coordination of O 2 to the O 2 -reduction site, which would significantly lower the rate of the second electron donation to the bound O 2 from Cu B 1ϩ , thereby providing a slow two-electron reduction of O 2 to peroxide. The first electron is transferred from Fe a3 2ϩ , giving Fe a3 3ϩ -O 2 Ϫ , when O 2 is bound to Fe a3 2ϩ (7,33). The existence of the interstitial water molecule detectable in the intermediate species, together with the uneven coordination of O 2 between Fe a3 2ϩ and Cu B 1ϩ , strongly suggests that rapid degradation of the peroxide-bound form induces formation of Fe a3 3ϩ -O 2 Ϫ and Fe a3 4ϩ ϭO 2Ϫ as the first and second intermediates, respectively (7). This X-ray structural finding is a crucial structural result for resolving the enigma, namely stable Fe a3 3ϩ -O 2 Ϫ and unstable Fe a3 , identified by the resonance Raman analysis long before (7).
It is possible that this water molecule is introduced along with binding of O 2 to Fe a3 2ϩ , which provides a negativelycharged ligand-bound state, Fe a3 3ϩ -O 2 Ϫ . This state is similar to CN Ϫ -bound fully-reduced CcO (9). In most oxygenated (or O 2 -bound) hemoproteins, Fe 2ϩ -O 2 is in a resonance structure between Fe 2ϩ -O 2 and Fe 3ϩ -O 2 Ϫ . IR results suggest that the latter is predominant (1, 33).

X-ray structural basis for the resonance Raman data obtained for the ferryl site
The P-form has two resonance Raman bands at 804 and at 350 cm Ϫ1 , which are assignable to Fe a3 4ϩ ϭO 2Ϫ vibrational and HisN-Fe a3 4ϩ ϭO 2Ϫ -bending modes, respectively (34). Our X-ray structure indicates that HisN-Fe a3 4ϩ ϭO 2Ϫ has an angle of 175.0°confirming the above spectroscopic assignment nonempirically. The absorption spectrum of the crystals in Fig. 2C indicates that this X-ray structure is a mixture of the P-and F-form. Thus, the HisN-Fe a3 4ϩ ϭO 2Ϫ angle of the P-form is likely to be narrower than 175.0°, because HisN-Fe a3 4ϩ ϭO 2Ϫ of the F-form is straight, as evidenced by the lack of a resonance Raman bending mode.
Although the resonance Raman stretch band shift from 804 to 785 cm Ϫ1 upon the P-to-F transition was discovered fairly long time ago (1, 7), the structural base for the spectral change

X-ray crystal structures of cytochrome oxidase intermediates
has not been identified. The Fe-O bond length difference of 0.011 Å (1.647 Å versus 1.658 Å for the P-and F-form), estimated by Badgers' rule from the stretching vibrational band positions of the P-and F-forms (804 and 785 cm Ϫ1 ), is too small to be identified in the X-ray structures at the present resolution. However, the existence of the low-barrier hydrogen bond between Fe a3 4ϩ ϭO 2Ϫ and Cu B -OH Ϫ identified in this work suggests that Cu B , which is coordinated with His 240 crosslinked to Tyr 244 , controls the vibrational properties of Fe a3 4ϩ ϭO 2 by coordination state change in Cu B induced by hydrogenation of Tyr 244 -O ⅐ radical to Tyr 244 -OH, coupled with the P-to-F transition. The coordination state change in Cu B could also eliminate the nonstraight structure of His-N-Fe a3 4ϩ ϭO 2Ϫ of the P-form upon transition to the F-form via the tight hydrogen bond. The present findings as described above suggest an O 2 -reduction mechanism summarized schematically in Fig. 10. When the O 2 -reduction site is in the fully-reduced state (Fig.  10A), O 2 is collected through the O 2 channel to form Fe a3 3ϩ -O 2 Ϫ . The negatively-charged O 2 molecule collects the interstitial H 2 O molecule (Fig. 10B). The rather slow electron transfer from Cu B to the bound O 2 generates the peroxide-bridged form (Fig. 10C). Immediately after formation of the peroxidebridged form, the interstitial water molecule mediates the transfer of a proton from Tyr 244 -OH to the oxygen bound to Fe a3 3ϩ , concomitant with electron transfer from Tyr 244 -O Ϫ to Cu B 2ϩ via His 240 cross-linked to Tyr 244 (Fig. 10, D and E). The two electrons from Cu B 1ϩ and Fe a3 3ϩ reduce the bound O 2 2Ϫ to break the O-O bond (Fig. 10F). Facile proton exchange through the low-barrier hydrogen bond between the two oxides yields the P-form (Fig. 10G). The P-to-F transition, coupled with elimination of the bent structure of HisN-Fe a3 4ϩ ϭO 2Ϫ and the bending mode and the shift in the stretching band of Fe a3 4ϩ ϭO 2Ϫ , is induced by Cu B coordination state change triggered by reduction and protonation of the Tyr 244 radical to form Tyr 244 -OH (34).

X-ray structures of the water channel of the P-and F-forms facilitating unidirectional proton-transport through the H-pathway
The unidirectional proton-transfer mechanism facilitated by narrowing the water channel of the H-pathway has been proposed by comparison of the X-ray structures of the resting oxidized and fully-reduced CcOs as mentioned in the Introduction. These X-ray structures indicate that the open/closed structural changes in the water channel of the H-pathway are controlled by the ligand-binding and oxidation states of the O 2 -reduction site. However, these two forms are not directly involved in the catalytic cycle. The ligand-binding and oxidation states of the resting oxidized form, Fe a3 3ϩ -O 2 2Ϫ , are definitely different from those of the P-and F-forms, Fe a3 4ϩ ϭO 2Ϫ (1,7,14). Furthermore, the resting oxidized CcO does not pump protons (1,35). Therefore, the resting oxidized form is not an appropriate model for either the P-or F-forms. Thus, the present results, showing that the water channel is closed in both the P-and F-forms, are one of the critical experimental confirmations for the unidirectional proton-transfer mechanism. The closed structure of the water channel of the P-and F-forms Ϫ results in a rather slow formation of the peroxide-bridged form (C). Immediately after the peroxide-bound form is attained, the interstitial H 2 O, hydrogenbonded to Tyr 244 , takes up protons from Tyr 244 -OH and forms a hydronium ion that interacts with an oxygen atom of the bound peroxide (D). Water accessibility, as shown in Fig. 10D, strongly suggests the preference for proton transfer to the oxygen bound to Fe a3 3ϩ versus the oxygen at Cu B (D). Concomitant with the proton transfer, a facile electron transfer takes place from Tyr-O Ϫ to Cu B via His 240 (E). The resultant structure (E) triggers a simultaneous two-electron reduction of the bound peroxide from both Fe a3 3ϩ and Cu B 1ϩ , thereby breaking the O-O bond (F) to form the low-barrier hydrogen bond that induces an angled OϭFe-N structure, followed by a proton exchange between the two oxide ions, yielding the P-form (G). The F-form is produced by a proton-coupled reduction of the oxyradical of the P-form, perturbing the low-barrier hydrogen bond and yielding a straight OϭFe-N structure (H).

X-ray crystal structures of cytochrome oxidase intermediates
can provide complete blockage of proton back-leakage without any extra energy cost in the proton pump during the P-to-F transition.
Resonance Raman results indicate that the Fe a3 3ϩ -OH Ϫ species of the O-form is in the low-spin state (34), whereas the resting oxidized form has Fe a3 3ϩ -O 2 2Ϫ in the high-spin state. The results of a single-electron injection technique applied to the O-form has led to the proposal that the E-form has a lowspin Fe a3 3ϩ -OH Ϫ species and a Cu B 1ϩ species (35). It has been shown recently that Cu B is also involved in the open to closed transition of the water channel (19). Because the structures of the O 2 -reduction site of the O-and E-forms are significantly different from the structure of the resting oxidized form, the resting oxidized form cannot be an appropriate model also for the O-and E-forms. X-ray structural analyses for these intermediates are prerequisite for evaluation of the mechanism of the unidirectional proton transfer in steps other than the P-to- Recently, an XFEL analysis at 2.  (36). Although the intermediate was assigned as "Pr" (a one-electron reduced form of P) (36), the two ligand oxygen atoms were not located due to the low resolution of the X-ray diffraction data. Significant improvements in the resolution of the XFEL analysis, for which improvements in crystallization conditions appear to be most critical, are necessary for direct comparison of our static X-ray structures of the intermediate species at 1.8 Å resolution with the time-resolved results. A brief discussion on an alternative proton-pumping mechanism proposed for bacterial CcO (1, 2, 37) is given in text S3.

Preparation of H 2 O 2 -treated CcO crystals
Resting oxidized bovine heart CcO crystals were prepared as described previously (38). The final medium composition was achieved with 50 stepwise manual exchanges from the initial medium composed of 40 mM sodium phosphate buffer, pH 6.5, 0.2% decylmaltoside, 1% PEG 4000, and 2% ethylene glycol in which the crystals are stable at 4°C. Before freezing, the crystals were treated with appropriate concentrations of H 2 O 2 and PMS in the final medium. The absorption spectral changes were followed after treatment as previously described (14). When no further significant absorbance increase at 580 nm was detectable, the crystals were frozen at 90 K in 40 mM sodium phosphate buffer, pH 5.7, 0.2% decylmaltoside, 8% PEG 4000, and 40% ethylene glycol.
Spectra of the frozen crystals were acquired under N 2 flow at 100 K in a micro-loop for X-ray diffraction experiments. A white-light microbeam was irradiated vertically onto the square board plane of the crystal. Although a minimal amount of the mother liquor of the frozen crystal induces significant curvature in the baseline of the absorbance measurement (Fig. 2C), reproducible measurements sufficient for quantitative evalua-tion of the relative content of the intermediate forms were achieved by controlling the angles of the measuring beam relative to the square board plane of the crystal.
To maintain integrity of the crystals, lower concentrations of H 2 O 2 and PMS are preferable because both PMS and H 2 O 2 could cause deterioration of the protein moiety. However, concentrations of 50 mM H 2 O 2 and 50 M PMS were selected to minimize contamination of X-ray diffraction data by the resting oxidized form. PMS does not produce significant interference in absorption spectral measurements between 500 and 700 nm under the present experimental conditions.  Table S1 together with those of each crystal. Cell dimensions of a, b, and c range between 176.88 and 177.63 Å, 181.65 and 182.64 Å, and 203.34 and 205.45 Å, respectively, among 17 crystals. Statistics of cell dimensions suggest that these crystals are highly isomorphous with each other. Out of 1235 images, 897 images were selected to merge into a diffraction data set at 1.8 Å resolution. Averaged mosaicity and its e.s.d.s. estimated by the program SCALE-PACK for 897 images were 0.189 and 0.052°, respectively. R pim , I/I(I) and CC 1/2 of the highest resolution shell given in Table 1 indicate that the quality of intensity data are appropriate for structure analysis at 1.8 Å resolution.

X-ray diffraction experiments
The high yield of the purification method (about 1/2 g of the purified CcO from each batch of the purification (38)) and gradual (stepwise) manual soaking of the crystals with the medium containing a high concentration of the anti-freeze were found to be critical for providing crystals with high isomorphism as well as with low mosaicity.
All X-ray experiments were carried out at beamline BL44XU/SPring-8, which was equipped with an MX225HE CCD detector. Crystals with dimensions of ϳ700 ϫ 700 ϫ 200 m were used for diffraction experiments. For all diffraction experiments, the thin edge of a crystal was aligned parallel to the X-ray beam at a rotation angle of 0.0°. Each crystal was shot with X-rays in a helium gas stream at ϳ50 K and translated by 10 m after each shot to minimize radiation damage. Other experimental conditions used for collecting low-resolution data were as follows: X-ray beam cross-section, 25 m (v) ϫ 25

X-ray crystal structures of cytochrome oxidase intermediates
m (h) at the crystal; X-ray attenuated by aluminum foil with 0.4-mm thickness; camera distance of 431 mm; exposure period of 1.0 s; and oscillation angle of 1.0°. The conditions for collecting high-resolution data were as follows: X-ray beam crosssection of 150 m (v) ϫ 25 m (h) at the crystal, camera distance of 230 mm, exposure period of 1.0 s, and oscillation angle of 0.5°. The wavelength was 0.9 Å, and the photon number at the sample position was 4.0 ϫ 10 11 photons/s. A total of 17 crystals were used to acquire full data sets at resolutions of 1.8 Å. The radiation dose for each diffraction experiment was estimated using the program RADDOSE (39). Data processing and scaling were carried out using HKL2000 and SCALEPACK (40). A total of 897 images were successfully processed and scaled. Structure factor amplitudes (͉F͉) were calculated using the CCP4 program TRUNCATE (41)(42)(43)(44). Other statistics from the intensity data are provided in Table 1.

Structure determinations and refinements
Initial phases of the structural factors up to a resolution of 5.0 Å were calculated by the MR method (45) using a model built from a resting oxidized structure previously determined at 1.5 Å resolution (PDB code 5B1A) after removing nonprotein molecules, including water molecules, lipids, and detergents. The phases were extended to a resolution of 1.80 Å by DM (46) coupled with noncrystallographic symmetry averaging (NCS) (47,48) using the CCP4 program DM (49).
An outline of our calculation procedure for MR/DM electron density maps is as follows: electron density distributions at the low resolution were modified by flattening those of the solvent region, idealizing electron density distribution (46), and averaging those related by NCS (47,48). Phases of reflections slightly higher than the reflections used for the electron density calculation were estimated by Fourier transform of the modified electron densities. The phases were gradually extended to a resolution of 1.80 Å by repeating the density modification (DM) procedure, using the CCP4 program DM (49). The F o data of the reflections ranging from 131.6 Å resolution to 1.80 Å resolution were used for MR/DM electron density calculation. The low resolution F o data down to 131.6 Å resolution ensures high convergency of phase extension in the DM procedure. Flattening of electron densities of solvent regions as high as 70% volume of the unit cell estimated by Matthews (50) and a local 2-fold symmetry averaging in the DM procedure were performed for efficient elimination of initial model bias.
The resultant phases (␣ MR/DM ) were used to calculate the electron density map (MR/DM map) with Fourier coefficients ͉F o ͉exp(i␣ MR/DM ), where ͉F o ͉ is the amplitude of the observed structure factor. A structural model of oxidized CcO previously determined at 1.5 Å resolution (11) was revised on the MR/DM map. Refinement was carried out using phenix.refine (51). The phenix refinement was performed without simulated annealing procedure unless otherwise stated. Model building was performed using COOT (52). An asymmetric unit of the unit cell contains two monomers of A and B, each consisting of 13 different protein subunits. Each monomer in the asymmetric unit, related by NCS, was assigned to a single group for TLS refinement. The anisotropic temperature factors for the zinc, copper, iron, and magnesium atoms were imposed on the calculated structure factors. Molecules of water, ethylene glycol, lipids, and detergents were located on the MR/DM map and F o Ϫ F c maps composed of the phases calculated using atomic parameters of protein atoms and cofactors. Refinement statistics are listed in Table 2.