Cryo-EM structures of intermediates suggest an alternative catalytic reaction cycle for cytochrome c oxidase

Cytochrome c oxidases are among the most important and fundamental enzymes of life. Integrated into membranes they use four electrons from cytochrome c molecules to reduce molecular oxygen (dioxygen) to water. Their catalytic cycle has been considered to start with the oxidized form. Subsequent electron transfers lead to the E-state, the R-state (which binds oxygen), the P-state (with an already split dioxygen bond), the F-state and the O-state again. Here, we determined structures of up to 1.9 Å resolution of these intermediates by single particle cryo-EM. Our results suggest that in the O-state the active site contains a peroxide dianion and in the P-state possibly an intact dioxygen molecule, the F-state may contain a superoxide anion. Thus, the enzyme’s catalytic cycle may have to be turned by 180 degrees.

C ytochrome c oxidases (CcOs), members of the hemecopper superfamily of terminal oxidases, are among the most fundamental enzymes of life. Located in the inner membrane of mitochondria or of some prokaryotes they reduce molecular oxygen (dioxygen) to water and generate an electrochemical proton gradient across the membrane by using electrons from the external and protons from the internal membrane side as well as by proton pumping [1][2][3] . In order to understand these processes, it is essential to know the structures of the intermediates of the catalytic cycle which comprises the oxidized form (O-state), the reduced CcO (R-state), the P-state (after reaction with dioxygen), then the F-state and again the O-state after successive electron transfers (Fig. 1).
In the conventional view (see Fig. 1) the reaction catalyzed by CcOs starts with electron transfer from a reduced cytochrome c to Cu A . Electrons are subsequently transferred to the binuclear centre (BNC) via the low spin heme a. The BNC is composed of the high-spin heme a 3 and the Cu B center and constitutes the active site of the enzyme 4,5 . The oxidized BNC [O (Fe a3 3+ , Cu B 2+ )] is reduced by the first electron leading to the appearance of the E-state (E = electronated) which is converted to an R-state [R (Fe a3 2+ , Cu B + )] by further electron transfer. The R-state binds dioxygen forming compound A 6 . The original believe that two electrons are transferred onto the bound dioxygen creating a peroxide in the BNC, the P-state, (see e.g., ref. 7 ) has been challenged on the basis of the results of resonance Raman spectroscopy and magnetic circular dichroism spectroscopy 8,9 . These results indicated that the dioxygen bond is already broken in the P-state, one oxygen atom has been bound to the heme a 3 -Fe via a double bond, thus forming an oxoferryl moiety whereas the second oxygen atom is released as water. However, for the splitting of the dioxygen bond four electrons are required. Two of them would be provided by the heme a 3 -Fe, one by Cu B . The potential donors of the missing fourth electron are shown (Fig. 1). Currently, a tyrosine residue cross-linked to a histidine ligand of Cu B is favored. A simultaneous transfer of four electrons onto dioxygen is supposed to lead to the splitting of the dioxygen bond without the danger of forming reactive oxygen species. Whether a truly peroxidic P-state is part of the reaction cycle has been discussed controversially. Early Raman spectroscopy studies had suggested the presence of a peroxidic P-state intermediate while the later results have been interpreted in favor of the presence of an oxoferryl moiety in the BNC [8][9][10] . The proposal that the P-state already contains an oxoferryl moiety has found rapid acceptance because Weng and Baker had claimed that the 606-(P-state) and 580 nm (F-state) species do not differ in oxidation state, and the F-state, obtained after the input of the third electron, is generally believed to contain an oxoferryl moiety 11 . More recently there has been a comeback of a true peroxidic P-state intermediate 12 . Input of the fourth electron leads to formation of a water molecule (or a hydroxide) closing the cycle.
In this work we used single-particle cryo-electron microscopy (cryo-EM) to determine high-resolution structures of biochemically defined and adjusted intermediate states (O, R, P CO (carbon monoxide induced), F) of the wild type four-subunit CcO from Paracoccus denitrificans in lipid nanodiscs complexed with an Fvantibody-fragment. These intermediate states refined locally to 1.9 Å (O), 2.6 Å (R), 1.9 Å (P CO ) and 2.3 Å (F) resolution. Our density maps provided insights into the active site configuration during the catalytic cycle revealing expected and unexpected rearrangements of dioxygen species in proximity to the BNC for the O, P, and F intermediates. In addition, we found densities for many dioxygen molecules in particular in the O-state CcO. Collectively, our results suggest that the established CcO catalytic cycle may have to be revised.

Results and discussion
We recorded UV/visible absorption spectra throughout the specimen preparation to confirm the presence of the desired specific electron-induced catalytic intermediate states (Fig. 2, Supplementary Fig. 1). The O-state CcO showed a Soret peak with an Fig. 1 The structure of oxygen intermediates in the catalytic cycle of cytochrome c oxidase. As the chemical composition of the P-state remains unsettled, alternative proposed structures are presented (intact dioxygen bond left, split bond right). Por • and res • denote radicals of the porphyrin ring and a spatially close amino acid residue, respectively. absorption maximum at 426 nm, while that of the R-state is redshifted to 448 nm. The difference absorption spectrum (P CO minus O) for the two-electron induced P CO -state (CO + 2OH -→ 2e -+ CO 2 + H 2 O) showed a Soret peak at 433 nm (α-band at 607 nm) which is distinct from the three electron-induced F state with its maximum at 438 nm (α-band at 585 nm).
Identification of the dioxygen channel. We start with a description of the O-state ( Supplementary Fig. 2-7). Our density map allowed us to model a continuous cluster of dioxygen molecules, starting at the hydrophobic membrane interface near Val 42.I of TMH1.I and Leu 149.I of TMH3.I, ending at Val 279.I in proximity to the dioxygen binding site of heme a 3 and Cu B (Fig. 3a/3b). Comparable corresponding densities were not observed for the fully reduced enzyme, however, they are in excellent agreement with the molecular dynamics results of Hofacker and Schulten 13 . We performed computational assisted calculations (MOLE 2.5 14 ) to map possible channels within the O-state structure. We identified a continuous hydrophobic pathway that overlaps with the densities assigned as oxygen molecules (Fig. 3c/3d). This pathway is in good agreement with the previously characterized oxygen channel of a B-type cytochrome c oxidase 15 . The entry site of our proposed channel is defined by a cavity at the interface between the transmembrane region and the hydrophobic membrane environment located in subunit I. Site-directed mutagenesis experiments have demonstrated a significantly higher K M for oxygen when replacing Val 279.I by isoleucine 16,17 . We conclude that impairment of oxygen binding and reduction in the V279I mutant is caused by the introduced bulky residue at the end of our identified oxygen channel restricting access to the dioxygen reduction site (Supplementary Fig. 8).
O-state. In the O-state structure, we were able to identify a prominent density within the binuclear center (BNC) between heme a 3 and Cu B . Careful refinements with respect to distances to neighbouring molecules led to the assignment of a bridging peroxide species for this density in agreement with previous X-ray O 20 . Neither a single nor two hydroxide ions could efficiently reside in this location since the density of one anion would be too weak while individual hydroxide ions couldn't be accommodated due to insufficient space in proximity to both metals. The Fe a3 -Cu B distance is 4.7 Å with the heme a 3 iron (93°) coordinated almost in-plane to the porphyrin scaffold calculated from the merged bond angles heme a 3 N-Fe-His 411 . The existence of peroxide in the BNC as observed in various X-ray structures has not found much acceptance despite the fact that it has been shown that six electrons are required to fully reduce the as isolated oxidized CcO 18,21,23 . Four electrons would be needed to reduce the four prosthetic groups of CcO and two to reduce the peroxide. A peroxide bridge would be ideally suited to electrostatically compensate the positive charges at Fe a3 and Cu B . The peroxide has been claimed to be an artifact of X-ray radiation by X-ray induced reduction of dioxygen molecules present in the crystals 21,23,24 . The observation of a density for a dioxide molecule, presumably peroxide, by cryo-EM definitely rules out such explanations. In close proximity to the peroxide dianion (3 Å), we observed a distinct density feature which most likely represents a  TMH1   TMH2   TMH3  TMH4   TMH5   TMH6 heme R-state. The structure of the fully reduced R-state CcO does not show any density between Fe a3 and Cu B nor does it show densities for bound oxygen in the substrate conducting channel (Fig. 5). The absence of the densities assigned to dioxygen molecules in the O-state after the dithionite treatment supports the assignment of the observed densities in the channel of the Ostate to dioxygen molecules. An interesting distinction between the O-state and the R-state is the different side-chain orientation of the prominent residue Lys 354.I of the K-pathway (Fig. 6). In the fully reduced R-state the side chain is found closer to the then more negatively charged BNC that might indicate that the lysine side chain is protonated, or becomes protonated during reduction. Figure 6 also presents the water molecules which form part of the proton transfer network and of the pump loading site.
P-state. P-states can be induced by a number of different methods 25 . We chose to generate the P-state by limited exposure to carbon monoxide in the presence of oxygen thus denoted here as P CO -state. This is a classical procedure discovered by Peter Nicholls and was used also in previous experiments 25,26 . The presence of the P CO -state was unambiguously confirmed by uv/ vis spectroscopy. The structure in and around the BNC, obtained at a local resolution of sub 2 Å, met us with a big surprise. We identified a specific density between the two metal centers of the active site while, compared to the O-state, the nearby dioxygen molecule appears to be absent and to be replaced by a water molecule or a hydroxide ion (Fig. 7).   Fig. 9). The density for the diatomic species in the P-state appears to be more parallel to the heme a 3 macrocycle than in the O-state. As outlined above the view based on resonance Raman spectroscopy that the dioxygen double bond has been already broken in the P-state appears to be widely accepted. However, one has to note that the species providing the signal in Raman spectroscopy may not reflect the most abundant population in the sample 27 . In contrast, particle averaging and class sorting in cryo-EM result in weighted map densities with the most populated species having the strongest contribution to the map features.
What about the chemical experiment providing evidence for a split dioxygen bond in P CO -state? There, the P CO -state was generated by flashing off carbon monoxide from mixed-valence CO bound CcO in the presence of 18 O 2 . Approximately half of the 18 O-label, compared to the amount of P CO -state CcO, was found as water 28 . However, the reaction of CO with dioxygen yielding CO 2 and water would provide the same experimental result.
What then can be the nature of the diatomic species between the two metals of the BNC? We can exclude carbon monoxide on the basis of the spectroscopic features of the state generated and investigated. A peroxide dianion would be a possibility. However, the BNC of the O-state contains two electrons more than that of the P CO -state (see Fig. 1). If one accepts that the O-state BNC contains a peroxide dianion, the BNC of P CO -state should contain a neutral dioxygen molecule! In agreement with this proposal, quantum chemical calculations have indicated that binding of dioxygen to the oxidized BNC is energetically favorable 23 .
F-state. In order to gain insights into the previously unknown ligand arrangement of the F-state BNC, we treated the CcO with an excess of hydrogen peroxide as published 25 . We obtained the structure of the F-intermediate at a resolution of 2.3 Å. The space between Cu B and heme a 3 does not contain a distinct density (see Fig. 8) as observed for the P CO state (Fig. 7). Hence, the absence of a dioxygen species is apparent. Based on the expectations we modelled a geometrically optimized oxoferryl group in the center of heme a 3 in agreement with previously postulated structures, despite the fact, that established X-ray crystallographically determined electron densities of oxoferryl heme moieties in proteins reveal a much better shape for the oxoferryl moiety 29 . However, modeling simply a ferric iron into the central heme a 3 density is at least equally well possible (Fig. 8d). Modelling of a water molecule as a potential fourth Cu B ligand, providing a hydrogen bond to the oxoferryl oxygen, was not possible because of insufficient space. The Fe a3 -Cu B distance of 4.5 Å is similar to that in the O/P CO state while the coordinated iron is slightly more displaced from the center of the heme macrocycle (95°) compared to the O state (93°). Furthermore, the cross-linked His 276.I -Tyr 280.I distance has increased by more than 1 Å (2.7 Å) compared to the P co state (1.6 Å). Clearly, the covalent crosslink between the N ε2 of His 276.I and C ε2 of Tyr 280.I appears to be subject of structural rearrangements based on the interatomic distance. This observation also means that a reinvestigation of whether the crosslink is opened in the F-state is required. In addition, the bonding distance between His 411.I -Fe a3 is shortened by~0.3 Å (2.2 Å) which can be explained by a different protonated state as outlined previously 30 . Intriguingly, a prominent and undescribed density is located between His 326.I and Val 279.I in close proximity to Cu B (3.3 Å). Its shape and location argue for a dioxide species, dioxygen or superoxide. After reduction the latter could form a peroxide bridge between Fe a3 and Cu B as observed in the O-state. We find a similar but slightly more displaced density also in the O-state in proximity to Val 279.I  in a more distant position (~0.4 Å, 3.7 Å) towards the Cu B . The presence of superoxide at this position in the F-state agrees well with complementary EPR and ligand-mimicking studies, which have shown that treatment of the F-state CcO with catalase leads to an O-state like UV/vis spectrum and the appearance of a tyrosine radical 25 . This observation can be explained if a superoxide present takes an electron from the CcO thus being converted to a peroxide that leaves the enzyme and is split by the catalase.
As mentioned above and presented in Fig. 8, modelling of the central density of heme a 3 iron as a ferric iron atom is at least equally well possible. The experimental result mentioned above that treatment of the F-state CcO with catalase leads to the formation of a CcO form with UV/vis spectral properties very similar to the O-state 25 . The only difference to the O-state spectrum is the appearance of a broad absorption band around 630 nm. This band has been considered to be a charge transfer band of the ferric heme a 3 iron 31 . How could the simple addition of catalase lead to the removal of an oxygen atom from a central ferryl iron? We also would like to remind the reader of the postulate by Weng and Baker that the 606-(P-state) and 580 nm (F-state) species do not differ in oxidation state 11 .
The scenario which fits our experimental results best is the following: The classical P-state is an oxygenated oxidized form of CcO. Upon reduction of the bound dioxygen by one electron a superoxide anion is formed which is observed near Cu B . This form corresponds to the classical F-state. Upon input of another electron, the superoxide is reduced to a peroxide, leading to the O-state with the peroxide bridging the heme a 3 iron and Cu B .
For a complete cycle (Fig. 9) the observed O-state has to be reduced by two more electrons which would lead to the formation of two water molecules (or one water molecule and one hydroxide ion) which would have to be replaced by a dioxygen molecule forming the P-state again. We look with great interest into the structure determination of the one-electron reduced E-state.

Methods
Production of cytochrome c oxidase from Paracoccus denitrificans. The cytochrome c oxidase (cytochrome aa 3 ) from Paracoccus denitrificans was produced in P. denitrificans AO1 cells transformed with the pUB39 (pBBR1 MCS derivative) plasmid and RP-4 helper strain via triparental mating according to a standardized in-house protocol 32,33 . A pre-culture was set up by adding 0.1 ml of 50% glycerol stock to 50 ml succinate (25 µg/ml kanamycin, 25 µg/ml streptomycin) growth medium. Cells were incubated at 32°C while shaking (180 rpm) overnight. An intermediate-culture was started by inoculating 500 ml succinate (25 µg/ml kanamycin, 25 µg/ml streptomycin) growth medium with 50 ml of the pre-culture. Subsequently, cell production was carried out by growing the main culture with 2.5 l by adding 200 ml from the intermediate-culture at 32°C while shaking (150 rpm) for approximately 8 hours. After harvest and homogenization, cells were disrupted using a high-pressure homogenizer (Constant Systems Ltd.) with an applied pressure of 25 kpsi. The cell lysate was centrifuged at 6,000 × g at 4°C for 60 m followed by an ultra-centrifugation step with a speed of 205,000 × g at 4°C for 12 h. Collected membrane pellets were suspended in 50 mM KPi (pH 8) with a concentration of 20 mg/ml.  plasmid pASK68 which possess the ability to bind to SU II of CcO and carries a C-terminal Strep-tag 34,35 . For production, a pre-culture was inoculated with a volume of 0.1 ml 50% glycerol stock and grown in a 200 ml LB (100 µg/ml ampicillin) growth medium at 30°C (180 rpm) overnight. The 2 L main culture (100 µg/ml ampicillin) was grown at 23.5°C while shaking (165 rpm) to an OD600 of 0.5 before recombinant antibody-fragment production was induced by the addition of 0.5 mM IPTG for additional 3 h. Collected cell material was centrifuged at 3500 × g for 15 m at 4°C and pellets were subsequently mixed with 500 mM sucrose, 35 mM KPi (pH 8) for 20 m on ice. The disrupted cell lysate was centrifuged at 24,000 × g at 4°C for 60 m and the supernatant was collected.
Purification of cytochrome c oxidase from P. denitrificans. For preparation of the four-subunit cytochrome c oxidase, pelleted membranes were solubilized at pH 8 with 1 % dodecyl-β-maltoside (DDM) with a resulting mass ratio of 1 mg detergent per 5 mg of membrane proteins. In a first step, solubilized CcO was incubated with the strep-tagged Fv fragment for 30 min at 4°C while stirring to allow efficient binding. Final mixture was centrifuged at 235,000 × g for 60 m at 4°C and the supernatant was loaded on a Strep-Tactin ® XT Superflow ® high capacity column (IBA GmbH) to perform affinity chromatography. Production and purification of MSP1D1. The production of the N-terminal 7x histidine-tagged MSP1D1 (Membrane scaffold protein) was done according to a standardized protocol 36 . Therefore, BL21Gold (DE3) E. coli cells were transformed with pMSP1D1 and grown in LB-Kan medium (50 µg/ml kanamycin) at 37°C. By reaching an OD 600 of 0.6, induction with 1 mM IPTG was carried out to start the MSP1D1 production for 4 h before cell harvest. The pellet from 4 L cell culture was resuspended in a cell disruption buffer (20 mM Tris-Cl pH 8, 1 mM PMSF) and afterwards supplemented with Triton X-100 (1% (v/v)). The resulting lysate was centrifuged and filtered. In a final step, the cell solution was supplemented with 500 mM NaCl. The lysate was mixed with 6 mL Ni-NTA agarose (Thermo Fisher) which had been equilibrated with the washing buffer (20 mM Tris-Cl pH 8, 500 mM NaCl, 1% Triton X-100). The mixture was incubated for 12 hours while stirring at 4°C. Next, the solution was loaded, washed and eluted with 500 mM imidazole. Eluted fractions were pooled and dialyzed against buffer (20 mM Tris-Cl, pH 7.5, 150 mM NaCl) and finalized samples were concentrated to 5 mg/mL, frozen in liquid nitrogen and stored at -80°C.
Reconstitution of cytochrome c oxidase in MSP1D1 and lipids. The reconstitution of cytochrome c oxidase into lipid nanodiscs was accomplished by using the aforementioned protocols 36 . Here, mixing POPC, MSP1D1 and the CcO was done in a molar ratio of 500  (Thermo Fisher Scientific). The blotting time was set to 4 s before plunge freezing in liquid ethane.
Image recording. Images for each data set were recorded by using a Titan Krios G3 microscope operated at 300 kV (Thermo Fisher Scientific). Coma and beam tilt correction was done by using EPU (Thermo Fisher Scientific) and data collection was conducted by operating in an electron counting mode with a Falcon III direct electron detector at a nominal magnification of x96,000, corresponding to a calibrated pixel size of 0.833 Å. An accumulated dose of 0.98 e -/Å 2 per fraction was used for the 30 dose-fractionated frames (~total dose 30 e -/Å 2 ). Defocus values were applied in a range from 0.5 to 2.5 µm.
Image processing. Collected data in MRC format were consistently processed with RELION-3.1 37,38 and motion-corrected with the MotionCor2 algorithm 39 . Initial CTF parameters from each dose-weighted image were determined by using CTFFIND4 40 and particle images were automatically selected followed by initial model building, 3D classification, CTF refinement, Bayesian polishing and final map reconstruction using RELION-3.1. Unfiltered maps were sharpened by applying different b-factors for building and visual improvement purposes. The final overall resolution was estimated by using the gold-standard Fourier shell correlation (FSC 0.143 ) which was calculated from two independently refined data sets. Global resolution anisotropy was examined by RELION-3.1. Observable map quality improvements were achieved by applying a density-modification procedure to each data set in a consistent way 41 . For this purpose, both half maps and a sequence file were used together with a model file, a full map file and a mask file. Significant map improvements were already observed after one cycle and were repeated until no further improvement was observed ( Supplementary Figs. 2-7).
Model building and geometry refinement. For building the four-subunit atomic model of the cytochrome aa 3 oxidase, the Protein Data Bank (PDB) submission by Harrenga et al. (PDB 1QLE) was used as a template structure 42 . After manual backbone fitting and correct fitting of side chains in the respective map densities, real-space refinement was done using Phenix (version 1.18) 43 . As a final building step, density-modified maps were used for more precise molecule allocation around the catalytic center. Each finalized model was validated by the MolProbity online server 44 . A final summary for each model parameter and its corresponding cryo-EM statistics can be found in Supplementary Table 1. Finalized models were visualized by using Chimera 45 and ChimeraX 46 .
Visualization of an oxygen channel. The proposed oxygen channel was mapped with MOLE 2.5 (bottleneck radius: 1.5 Å, bottleneck tolerance 3 Å, origin radius 5 Å, surface cover radius 10 Å, max tunnel similarity 1) 14 and visualized with Chimera. Oxygen molecules were manually placed with Coot into corresponding densities observed in the O-state map.
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The generated cryo-EM maps of the aa 3 cytochrome c oxidase from Paracoccus denitrificans have been deposited at the Electron Microscopy Data Bank under accession codes EMD-11921, EMD-11922, EMD-11924 and EMD-11925. The models of the aa 3 cytochrome c oxidase structures were submitted to the PDB data bank with the accession numbers: 7ATE, 7AU6, 7ATN, and 7AU3. 1QLE was used as a template. All other data is presented in the main text or supplementary information. Reductive phase Fig. 9 A potential catalytic cycle of CcO based on our results. The classical P-state is an oxygenated oxidized form of CcO. Input of one electron reduces the bound dioxygen to superoxide which remains bound in the BNC near Cu B (classical F-state). Input of another electron reduces the superoxide to a peroxide which forms a bridging ligand in the BNC of the O-state. The next input of an electron leads to the formation of the classical E-state whose structure is still unknown. Input of the fourth electron leads to the formation of a 2-electron R-state which binds a dioxygen molecule closing the cycle. Two water molecules or one water molecule plus one hydroxide ion are formed during the input of the third and fourth electrons in the classical reductive phase.