Structure of Escherichia coli cytochrome bd-II type oxidase with bound aurachin D

Cytochrome bd quinol:O2 oxidoreductases are respiratory terminal oxidases so far only identified in prokaryotes, including several pathogenic bacteria. Escherichia coli contains two bd oxidases of which only the bd-I type is structurally characterized. Here, we report the structure of the Escherichia coli cytochrome bd-II type oxidase with the bound inhibitor aurachin D as obtained by electron cryo-microscopy at 3 Å resolution. The oxidase consists of subunits AppB, C and X that show an architecture similar to that of bd-I. The three heme cofactors are found in AppC, while AppB is stabilized by a structural ubiquinone-8 at the homologous positions. A fourth subunit present in bd-I is lacking in bd-II. Accordingly, heme b595 is exposed to the membrane but heme d embedded within the protein and showing an unexpectedly high redox potential is the catalytically active centre. The structure of the Q-loop is fully resolved, revealing the specific aurachin binding.

C ytochrome bd quinol:dioxygen oxidases (called bd oxidases for simplicity hereafter) are terminal reductases exclusively found in bacterial and archaeal respiratory chains. They couple quinol oxidation and release of protons to the periplasmic side with proton uptake from the cytoplasmic side to reduce dioxygen to water [1][2][3] . Thus, bd oxidases contribute to the generation of a protonmotive force (pmf) by a vectorial charge transfer 4,5 . They display a high affinity towards dioxygen, enabling growth under microaerobic conditions and endow pathogens such as Shigella flexneri 6 , Mycobacterium tuberculosis 7 , and various streptococcus species 8 with resistance to intracellular stressors such as NO and intracellular hypoxia, making them excellent drug targets. Furthermore, bd oxidases play a key role in tolerance to or degradation of NO 9 , H 2 S 10 , CO 11 and other gaseous ligands that are involved in anti-inflammatory immune response.
The Escherichia coli aerobic respiratory chain contains three terminal oxidases, cytochrome bo 3 oxidase, cytochrome bd-I oxidase (bd-I) and cytochrome bd-II oxidase (bd-II) 2,12,13 . While the cytochrome bo 3 oxidase belongs to the unrelated family of heme-copper oxidases, the bd oxidases display distinct features such as the presence of a d-type heme in the active site and the lack of copper atoms. Expression of cyoABCDE, encoding bo 3 oxidase is activated at full aerobic conditions, while cydABX, coding for bd-I, is maximally expressed at microaerobic and oxygen-rich conditions 14 . In contrast, bd-II, encoded by the appCBX operon, is expressed under complete anaerobiosis 14 and upon entry into the stationary phase and phosphate starvation [14][15][16][17] .
While the E. coli bd-II has been discovered quite recently and is relatively poorly characterized 18 , the bd-I is well-studied. It is made up of the two large subunits CydA and CydB, each comprising nine transmembrane (TM) helices, and the single TM helix subunit CydX [19][20][21][22] . CydA harbors all heme groups of bd-I, namely the low-spin heme b 558 , acting as electron input device to catalyze quinol oxidation 23,24 , the high-spin heme b 595 expected to deliver electrons to the third, the unique d-type heme, which is the site where dioxygen is reduced to water. CydB is related to CydA by a pseudo twofold symmetry and contains a structural ubiquinone-8 at a position occupied by the hemes in CydA 21,22 . CydA contains a soluble, periplasmic domain of variable length, termed Q-loop. This domain is expected to be involved in quinol binding and oxidation [21][22][23][24] . In E. coli, CydX plays a structural role as it is essential for the stability and/or the assembly of the active site 20 . Unexpectedly, E. coli bd-I contains a fourth subunit, CydY or CydH, that is encoded by the orphan gene ynhF 21,22 . CydY (or CydH) is also a single TM helix protein and probably plays a functional role by blocking substrate access to heme b 595 from the membrane 21,22 . The structure of bd-I from Geobacillus thermodenitrificans revealed that this enzyme lacks the fourth subunit and, accordingly, the positions of hemes b 595 and d are interchanged while the overall architecture of the enzyme remained unaltered 25 . To gain access to the substrate dioxygen, E. coli bd-I contains a hydrophobic channel, leading from CydB to heme d on CydA 21,22 .
By contrast, much less is known about bd-II. It is encoded by the appCBX genes, leading to the production of subunits AppC (predicted molecular mass: 58.1 kDa), AppB (42.4 kDa) and AppX (3.6 kDa) 18 . It also contains heme b 558 , b 595 and d as cofactors 26 . The two large subunits of bd-I and bd-II share high sequence similarity of about 60% 5 . The appBCX genes are most likely crucial for gut colonization by E. coli and other inflammatory Enterobacteriaceae family members 27 .
Here, we present the 3.0 Å resolution structure of E. coli bd-II with bound aurachin D, a specific inhibitor of quinol oxidation by bd oxidases, as obtained by electron cryo-microscopy (cryo-EM).
The structure comprises three subunits, AppB, AppC and AppX that show a similar architecture as the homologous subunits in bd-I. The arrangement of the heme cofactors in AppC corresponds to that of the hemes in bd-I. However, due to the lack of a fourth subunit, heme b 595 is accessible from the membrane but it is not the active site. A narrow substrate channel leads from AppB to heme d on AppC. The binding site of aurachin D is clearly resolved and aurachin binding enables structure determination of the entire Q-loop.

Results
Gene expression and protein production, purification and characterization. E. coli strain BL21*Δcyo 28 was transformed with pET28b(+) appC his BX encoding bd-II and grown under oxic conditions. Cells were disrupted and cytoplasmic membranes were prepared by differential centrifugation. Membrane proteins were extracted with the detergent lauryl maltose neopentyl glycol (LMNG) and purified by affinity-and size exclusionchromatography ( Supplementary Fig. 1). The final chromatogram showed two peaks, with the second one containing bd-II ( Supplementary Fig. 1). The major bands of the gel were identified as AppC and AppB by mass spectrometry (AppC: sequence coverage: 33%, overall score 69; AppB: sequence coverage: 3%, overall score 19). The identity of subunit AppX, migrating at an apparent molecular mass of about 4 kDa on a 16% SDS-gel (Supplementary Fig. 1), was also established by mass spectrometry (sequence coverage: 20%, overall score 25). Importantly, no other protein with a molecular mass below 10 kDa was detected in the low molecular mass region of the 16% SDS-gel by mass spectrometry.
From 9.6 L medium, 40 g cells and approximately 2 mg bd-II were obtained. The (reduced-minus-oxidized) difference spectrum of the preparation showed the typical absorbance of hemes b 558 , b 595 and d in a 1:1:1 stoichiometry using the given extinction coefficients 2 ( Fig. 1 and Supplementary Fig. 1). The single γ-peak in the Soret region at 440 nm matched the properties of hemes b 558 and b 595 . In addition, heme b 558 gave rise to absorbance at 533 and 563 nm, while heme b 595 produced another absorbance peak at 595 nm. The signal at 630 nm was finally assigned to heme d. The negative absorbance at 657 nm derived from the ferrous oxygenated heme d. Thus, the spectral properties of bd-II match those of bd-I (Fig. 1).
The UV-vis absorbance of the cofactors was used to determine their redox potentials by an electrochemical titration in a thinlayer cell ( Fig. 1 and Supplementary Fig. 2) 29 . From their absorbance at the α-band at 563 nm, the redox potential of both b-type hemes was determined to +237 mV (Fig. 1). It was not possible to further distinguish their individual redox potentials. This indicates that they exhibit similar values differing by less than 100 mV. The redox potential of heme d was determined from its absorbance at 629 nm to +440 mV. The hysteresis between oxidative and reductive titrations was <40 mV indicating the lack of cooperativity or other coupled processes (Supplementary Fig. 2). Taken together, the redox potentials of the bd-II cofactors are 60-80 mV (b-type hemes) and 180 mV (heme d) more positive than those of the corresponding cofactors in bd-I ( Fig. 1) 30 .
Structure of E. coli bd-II oxidase. According to its elution volume and BN-PAGE analyses, bd-II in LMNG eluted as a monomer from size-exclusion chromatography ( Supplementary  Fig. 1b). The protein was concentrated to 24 mg/mL, shock frozen in liquid nitrogen and stored at −80°C. Aliquots for cryo-EM preparation were thawed and diluted to 5.6 mg/mL in course of the amphipol exchange. Prior to cryo-EM analysis, purified bd-II ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-26835-2 was incubated with 160 µM aurachin D, a threefold molar excess of the specific inhibitor of bd oxidases (see below). On the cryogrid, predominantly dimeric species of bd-II were detected. Peaks of bd-II monomers and dimers were also observed by mass photometry when placing 1 µL of a concentrated sample (24 mg/ mL) on a 20 µL drop of the buffer used for grid preparation on the cover slip ( Supplementary Fig. 3) 31 . Although unlikely 31 , we cannot completely exclude the possibility that the presence of amphipol led to protein dimerization. The final cryo-EM map provided sufficient details for modeling the structure of the bd-II dimer with the bound inhibitor aurachin D at a resolution of 3.0 Å (Fig. 2 . The homo-dimer contains two oxidase hetero-trimers related by twofold symmetry (Fig. 2). Each hetero-trimer consists of the three subunits AppB, AppC and AppX of which the small AppX form the major dimer contact between the two heterotrimers. AppC and AppX establish additional contacts between the monomers through hydrophobic interactions between Val41 AppC and Trp23 AppX , as well as between Val42 AppC and Val22 AppX and Trp23 AppX , respectively.
Within each hetero-trimer, the large subunits AppB and AppC are related by a pseudo-twofold symmetry axis similar as CydA to CydB from bd-I. AppB and AppC share the same fold with nine TM helices that are arranged in two four-helix bundles and a peripheral helix. The interaction between AppB and AppC is mediated via the pseudo-symmetry related hydrophobic residues in TM helices 2, 3 and 9. Based on the structural conservation between AppB and AppC, it is reasonable to assume that appB and appC genes originated from gene duplication, although this is not reflected in their primary sequence ( Supplementary Fig. 6).
The single helix subunit AppX has a length of 30 amino acid residues and is completely buried within the membrane. It interacts with the AppC TM helices 1, 5 and 6, most likely stabilizing the protein fold to bind hemes b 595 and d. The dimerization of bd-II by AppX is mediated by several weak hydrophobic interactions, involving Met1, Leu4, Leu5, Val8, Leu11, Leu12, Ser15, Leu16, and Leu19. The Leu residues of the helix connect both monomers in a 'leucine zipper-type' fashion. Leu4 and Leu16 are both replaced by Phe residues in the homolog CydX in bd-I, while Leu5 and Leu19 are exchanged to Ala and Ile residues, respectively ( Supplementary Fig. 6). These changes might prevent dimer formation in bd-I. Indeed, the positions of all Leu residues are conserved in the homologs of AppX identified by their organization in a bd-II operon. In contrast, only the five Leu residues found in CydX are conserved among the CydX homologs, partly in different positions compared to AppX ( Supplementary Fig. 6). However, whether bd-II is also a dimer in the E. coli cytoplasmic membrane remains an open question.
AppB features a long hydrophobic patch in a cleft towards the periplasmic space filled with an extended electron density that was perfectly fitted as ubiquinone-8 (Fig. 2c). Its position follows the arrangement of the heme groups in AppC. The quinoid headgroup points away from AppC, indicating a structural role of bound quinone in stabilizing the AppB fold as reported for CydB 21,22 .
The Q-loop and binding of aurachin D. The family of bd oxidases is divided into L (long) and S (small)-subfamilies depending on the length of the periplasmic Q-loop 2 . Members of the L-subfamily harbor a C-terminal insertion in the Q-loop of about 60 residues that is essential for structural stability 32,33 . The N-terminal part of the Q-loop is involved in quinone binding 21 . The Q-loop of AppC (Fig. 3a, b) is located between TM helices 6 and 7 and extends over a length of 134 residues including the linker to the residual protein ( Fig. 3a, b and Supplementary  Fig. 7), rendering E. coli bd-II a member of the L-subfamily. Here, we resolve the entire Q-loop of bd-II covering positions 262−302 of the C-terminal region that are not found in the structures of the other bd oxidases 21,22,25 . The Q-loop consists of six small helices (Q α1 , P250-E257; Q α2 , L292-T297; Q α3 , L307-S332; Q α4 , P338-R345; Q α5 , L351-Y360, and Q α6 , A369-A379) that are connected by short loops (Fig. 3a, b). An additional density was identified in the cavity of the Q-loop that was modeled as a bound aurachin D molecule.
Aurachins inhibit electron transfer from ubiquinol-8 to heme b 558 by blocking the quinol binding site that is located close to the membrane surface ( Fig. 3a, b) [34][35][36][37] . Although aurachin C and D bind with high affinity to bd oxidases, a reliable IC 50 value for bd-II has not yet been determined 38 . Hence, we titrated the duroquinol:dioxygen oxidoreductase activity of our bd-II preparation with increasing amounts of aurachin C and D and determined the apparent IC 50 to 7.1 and 11.1 nM, respectively (Supplementary Fig. 8). Due to the extremely high affinity of aurachin D, the preparation of bd-II was incubated with only a threefold molar excess of aurachin D just prior to blotting on the cryo-EM grid.
The strong binding of aurachin is mostly due to a structural complementarity to the surface of the quinol binding site (Fig. 3c, d). Interactions with Phe269, Val271, Leu291, and Leu295 (the latter two from Q α2 ) stabilize the Q-loop and it is apparently this interaction with aurachin D that allowed for full resolution of the entire Q-loop (Fig. 3e). Aurachin D is bound by seven additional hydrophobic interactions to AppC (Fig. 3e). One H-bond of the aurachin nitrogen atom to Asp239 of AppC adds to the strong binding (Fig. 3e). This residue is conserved among bd-oxidases ( Supplementary Fig. 6). To experimentally test its importance for quinol oxidation and aurachin binding we produced the D239N AppC variant in strain CBO and purified wild-type bd-II Spectral and thermodynamic properties of the heme groups. a UVvis (reduced-minus-oxidized) difference spectra of the preparation of bd-II (black) and bd-I (red). The absorbance of the individual heme groups is indicated. b Redox potential of the heme groups of various bd oxidases. The values for the oxidases other than bd-II were taken from ref. 30 . and the D239N AppC variant from that strain by the procedure described above. Due to the presence of the chromosomal appCBX operon, the variant could not be produced in strain BL21*Δcyo. The proteins produced in strain CBO eluted at identical positions from the chromatographic columns. Remarkably, the enzymatic duroquinol oxidase activity of the D239N AppC variant was only 23% of that of the wild-type protein (2.91 vs. 12.84 µmol/mg/min). Furthermore, the IC 50 decreased from 11.1 to 60.7 nM ( Supplementary Fig. 8). Thus, Asp239 on AppC is strongly involved in aurachin binding and quinol oxidation. Binding of aurachin D completely blocks access to heme b 558 , the primary electron acceptor from the quinol (Fig. 3c, d) and, thus, prevents further electron transfer to hemes b 595 and d.
Arrangement of heme groups and access to the active site. The heme groups b 558 , b 595 , and d are positioned in between the TM helices of AppC and form a triangular arrangement (Fig. 4a). Heme b 558 is coordinated by His186 and Met393 as axial ligands, heme b 595 by Glu445 and heme d by His19. The iron−iron distance between heme b 558 and heme d is 18.8 Å, that between hemes b 558 and b 595 is 15.6 Å and the one between heme b 595 and d amounts to 11.1 Å (Fig. 4a). The position of Trp441, being discussed as important for intramolecular electron transfer 39 and formation of the F +* intermediate, is conserved in comparison with the bd-I type oxidase. Likewise, this arrangement of the heme groups in bd-II is virtually identical to that of the heme positions within E. coli bd-I. Noteworthy, hemes b 558 and b 595 are accessible from the membrane. For heme b 558 this is not surprising as it is the primary electron acceptor from the substrate quinol [21][22][23][24][25] . However, while in E. coli bd-I heme b 595 is shielded from the membrane by the additional fourth subunit CydY (or CydH), a fourth subunit was not detectable by mass spectrometry of the bd-II preparation in solution and of the low molecular weight region of a 16% SDS-gel ( Supplementary Fig. 1). It has been reported that G. thermodenitrificans bd oxidase also lacks a homolog of CydY 25 , but in comparison with E. coli bd-II, the positions of hemes b 595 and d are interchanged there. Consequently, heme b 595 of E. coli bd-II is found in a position that corresponds to the active site in the G. thermodenitrificans enzyme, rendering the membrane accessibility plausible. However, the architecture around E. coli bd-II heme d implies that this is the active site, because the TM helix 3 of AppC exerts a strong curvature (Figs. 2 and 4). This curvature is due to the insertion of the additional Leu101 also found in E. coli bd-I but lacking in G. thermodenitrificans bd oxidase ( Supplementary Fig. 6). This insertion places Glu99 in 5.1 Å coordination distance to the central Fe of heme d, thus providing space for substrate binding.
To experimentally address the question, whether heme d or b 595 is the active site, the preparation was incubated with KCN and UV-vis difference spectra were recorded over time (Fig. 4d). It is known that cyanide reacts with air-oxidized bd-I oxidase, when heme d is accessible 40,41 . The reaction results in a displacement of the oxy-complex of heme d by a cyanocomplex that is not reduced by a mild excess of reductant 40,41 . The preparation of bd-II incubated with 0.5 mM KCN clearly showed the development of the heme d cyano-complex at 652 nm (Fig. 4d). Furthermore, in agreement with the literature 40,41 , it was not possible to reduce heme d in the cyano-complex with a slight molar excess of dithionite (Fig. 4d). These data establish that heme d is indeed the active site of bd-II. This finds further support in an extra electron density located in the putative substrate binding cavity between Glu74 and Glu94 in AppC that may be caused by a dioxygen or a water molecule (Fig. 4c).
Dioxygen has no access to heme d via the solvent accessible channel to heme b 595 because heme d itself blocks a deeper penetration of substrates to the active site. In E. coli bd-I, another long substrate channel for oxygen on the opposing side was described, leading from CydB to heme d on CydA 21,22 . A similar channel is also present in bd-II, leading from AppB to Glu99 on AppC (Fig. 4b). However, this channel seems to have a smaller diameter, in particular due to the methyl group of Ala100 on AppC that narrows the channel diameter approximately by one fourth. At the homologous position, the bd-I features a Gly residue that widens the channel. To further analyze the geometry of the putative oxygen channels of both oxidases, we probed the radii with the program Caver 3.0 42 . The analysis of the derived models revealed radii of 1.4 Å for bd-I and of 1.2 Å for bd-II. Hence, despite the bd-II channel diameter being smaller, it is still sufficient to provide oxygen access to heme d.
This implies that Ala100 does not block the oxygen channel. However, if it would do so, the mutation of the homologous Gly100 in bd-I to Ala100 would significantly decrease, if not fully abolish the mutant's activity. Accordingly, we introduced the corresponding mutation (G100A bd-I ) into cydA of bd-I oxidase.
The bd-I oxidase variant was prepared from the mutant strain as described 22 . The duroquinol:dioxygen oxidoreductase activity of the preparation of G100A bd-I variant (11.7 ± 0.3 µmol/min/mg, standard deviation) was even slightly enhanced compared to that of bd-I (8.6 ± 2.2 µmol/min/mg, standard deviation). Thus, the methyl group of Ala100 does not hamper access of oxygen to heme d. The duroquinol:dioxygen oxidoreductase activity of the bd-II preparation was determined to 3.3 ± 0.2 µmol/min/mg (standard deviation), which is roughly 2.5-fold slower than that of the bd-I oxidase in agreement with the higher redox potential of the heme groups (Fig. 1).
Comparison to other bd oxidases. The structure of the E. coli bd-II monomer is very similar to that of E. coli bd-I (Fig. 5a), not surprising, considering the high sequence similarity of about 60% 2 (Supplementary Fig. 6). The positions of AppX and CydX completely overlap, emphasizing their structural role in stabilizing the binding of heme d and b 595 (Fig. 6a). The additional role of AppX in mediating the interactions between the two bd-II monomers does not lead to a change of its structure, indicating that indeed the individual residues of the TM helix contribute to dimer formation. Accordingly, the three common subunits of bd-I and bd-II align with a root-mean-squared deviation (RMSD) of 0.73 Å for the C α atoms. The preparation of bd-II lacks a homolog of the fourth subunit CydY that blocks the access to heme b 595 in bd-I. However, the relative tilt between TM helices 1 and 8 of AppC providing access to heme b 595 is lower than in bd-I and the channel to the heme is consequently more narrow. Still, it is accessible from the membrane as discussed above. Furthermore, the heme groups bind at very similar positions within the oxidases (Fig. 6b) as already indicated by the akin UV-vis redox difference spectra (Fig. 1). The same holds true for the position of the structural ubiquinone-8 that replaces the heme groups in AppB/CydB (Fig. 6c, d).
The structure of E. coli bd-II is also very similar to that of G. thermodenitrificans bd oxidase (Fig. 5b). Importantly, the access from the membrane to E. coli b 595 and G. thermodenitrificans heme d that are found in homologous positions differs from each other. According to the position of the protein backbone of the derived model, the access to the active site heme d in G. thermodenitrificans is provided by an open tunnel with a diameter of 6.6 Å (Supplementary Fig. 9). The homologous position in bd-II has a narrow keyhole shape with a maximum diameter of 5.5 Å (Supplementary Fig. 9). However, even this obstruction of the entry site is not sufficient to prevent a reaction of a substrate with heme b 595 in bd-II.
Proton pathway. The reaction of bd oxidases depends on a proton pathway that enables uptake of protons from the cytoplasmic site to heme d where dioxygen is reduced to water. TM helices 2 and 3 of AppC and of AppB, respectively, provide a broad cavity at the cytoplasmic site of the membrane (Fig. 7). Due to its hydrophilic nature this cavity is most likely filled with water molecules. A hydrophilic channel provided by the four helices protrudes from the large, hydrophilic cavity and extends nearly over the entire distance to heme d. Protons may also access heme d through Asp105 and Glu58 of AppC. From here, they can be transferred to the propionate group of heme d. The proton entry site of bd-II is much more open than those of the other oxidases (Fig. 7). The distance from Asp105 to the heme propionate group covers a distance of 13.1 Å. Interestingly, the channel to heme d is narrower in bd-I and bd from G. thermodenitrificans preventing access of water molecules as obvious from a Caver analysis 42 . Accordingly, more titratable side chains are needed to bring protons to heme d. The pathway for protons that leads along a series of titratable residues from Asp119 (bd-I) to the heme propionate covers a distance of 24.2 Å, while the corresponding one starting at His126 in G. thermodenitrificans has a length of even 30.1 Å.

Discussion
The structure of E. coli bd-II reported here shows an overall high similarity to that of E. coli bd-I 21,22 . However, there are few remarkable differences. First, bd-II consists of only three different subunits resulting in heme b 595 being accessible from the membrane (Supplementary Fig. 9). However, heme d is the active site, most likely caused by its unusually high redox potential (Figs. 1b  [14][15][16][17] . Recently, the structure of the Mycobacterium smegmatis bd-oxidase consisting only of the two major subunits was reported 43 . Two oxygen access channels were identified from the structure due to the lack of small subunits, one channel leading to heme d, the other to heme b 595 43 . Unfortunately, the redox potentials of the heme groups were not determined making it impossible to say whether both channels are used. Second, bd-II is mainly present as a dimer (Fig. 1) while bd-I is only found as a monomer 21,22 . The amino acid residues that mediate the dimer contact in bd-II are not fully conserved in bd-I ( Supplementary Fig. 6), which may indicate that bd-II is a dimer in vivo. During preparation, it turned out that bd-II, which is a monomer in LMNG ( Supplementary Fig. 1), is much less stable than bd-I in detergent. Bd-II tends to disintegrate at concentrations below 1 mg/mL and may require higher concentrations to obtain stability by dimerization. Third, the oxygen access channel has a smaller diameter in bd-II compared to bd-I (Fig. 4). Narrowing the channel in bd-I unexpectedly led to a slightly higher specific activity. Thus, a smaller channel diameter could be related to a higher substrate specificity preventing angled molecules to enter deep into the oxygen access channel. The very short proton pathway indicates that substrate protonation at heme d in bd-II is not rate limiting. Fourth, we have resolved the entire Q-loop of bd-II due to binding of aurachin D (Fig. 3). It was speculated that the N-terminal region of the Q-loop is intrinsically disordered or flexible and therefore cannot be resolved in structural analysis 43 . However, as shown here, this is not the case but it should be kept in mind that the incompletely resolved Q-loop belongs to a bd oxidase from the S-subfamily and shows in part a conformation different from that of other bd oxidases 43 . Due to its different length the Q-loop can be exchanged between E. coli bd-I and bd-II but not between E. coli bd-I and the G. thermodenitrificans enzyme 32,33 . The inhibitor aurachin is bound  in a hydrophobic pocket provided by the Q-loop and TM helices 6 and 7 of AppC (Fig. 3). The inhibitor and the protein pocket form a very nice complementary structure and binding through several hydrophobic interactions is intensified by a strong H-bond to D239 AppC (Fig. 3). This position is important for quinol and aurachin binding as the D239N AppC variant shows a decreased inhibitor sensitivity (Supplementary Fig. 8) and a clearly diminished duroquinol:oxygen oxidoreductase activity. The influence of this mutation on enzyme activity was reported for bd-I 44 .
The known structures of bacterial bd oxidases revealed a common architecture of E. coli bd-I, bd-II, G. thermodenitrificans and M. smegmatis bd oxidase. However, the enzymes seem to be involved in different physiological processes as their genes are expressed at different environmental conditions. The diverging reactivity is exerted by the addition of further, single helix subunits, by a varying arrangement of the same cofactors or by a different access to the same cofactor. Thus, diverse activities are generated by variations on a common theme. Structure determination of bd oxidases from other organisms will expand knowledge about this protein family and give rise to the assignment of new functions of these enzymes that may find their reflection in even more variations of the common bd oxidase theme.

Methods
Mutagenesis. KOD polymerase (Merck Millipore, Darmstadt, Germany) was used according to the manufacturer's specifications. Plasmid pET28b(+) was amplified with oligonucleotides pET28b(+)_fwd and pET28b(+)_rev (Supplementary Table 2). The app-genes were amplified from genomic DNA from E. coli strain BW25113 45 by standard PCR techniques, using the primers listed in Supplementary Table 2. The C-terminus of AppC was decorated with a His-tag affinity peptide. The oligonucleotides contain homologous regions to appC or appX. Initially, the appCBX genes were amplified from E. coli strain BW25113 with oligonucleotides appC_fwd and appX_rev containing homologous regions to pET28b(+) (Supplementary Table 2). The resulting PCR products were treated with DpnI and purified by agarose gel electrophoresis. For subsequent AQUA cloning 46 , plasmid and insert were mixed in a molar ratio of 1:5. The obtained vector pET28b(+) appCBX was used to introduce a sequence encoding a hexahistidine tag. Next, oligonucleotides appChis_fwd and appB_rev were used to generate the linear fragment appC his BX (Supplementary Table 2). This fragment was in turn used as megaprimer for RF cloning 47 . The plasmid encoding cydA his BX 22 was mutated by using the QuikChange protocol to generate the point mutation G100A bd-I . To generate bd-II variants, we cloned the appCBX genes from pET28b(+) into a pBAD vector 31 and transformed E. coli strain CBO with the pBAD appCBX expression plasmid 20,32 . The pBAD plasmid was mutated by using the QuikChange protocol to generate the point mutation D238N AppC . The sequences of the mutagenic primers are listed in Supplementary Table 2. Newly generated vectors were checked by sequencing (GATC Biotech, Konstanz, Germany). Restriction enzymes were obtained from Thermo Fisher Scientific (Darmstadt, Germany) or Merck Millipore (Darmstadt, Germany).
Cell growth. E. coli strain BL21*Δcyo 28 , lacking the cyoABCD genes that encode the ubiquinol: cytochrome bo 3 oxidase was used as host strain for protein production because the bo 3 oxidase has a natural His-tag that might interfere with purification of bd-II 48 . BL21*Δcyo was transformed with pET28b(+) appC his BX by electroporation. Cultures were grown aerobically in 2 L baffled flasks with 800 mL LB-medium each at 37°C. Gene expression was induced at an OD 600 of approximately 2 by the addition of 400 µM isopropyl β-D-1-thiogalacto-pyranoside. Cells were harvested 2 h after induction and stored at −80°C. To produce the bd-II D239N AppC variant, strain CBO transformed either with pBAD appCBX or with pBAD appCBX/D239N AppC and grown in LB medium as described 20,32 . Cells were harvested 2 h after induction and stored at −80°C.
6-10 g membranes (wet weight) were thawed and diluted in a sixfold volume 20 mM MOPS, 20 mM NaCl and 0.5 mM PMSF. Membrane proteins were solubilized by incubating the membrane suspension with 1% (w/v) lauryl maltose neopentyl glycol (LMNG) for 90 min and gently stirred at 4°C. After centrifugation (4°C, 50,000 rpm, 250,000 × g, 15 min, Rotor 60Ti, Optima LE80-K, Beckman-Coulter), the supernatant was immediately subjected to chromatographic purification. All chromatographic steps were carried out at 4°C using a NGC Quest 10 Plus (BioRad) chromatography system. 50 mM MOPS, 500 mM NaCl, 500 mM imidazole, 0.5 mM PMSF, 0.003% LMNG, pH 7.0 were added to the supernatant to a final imidazole concentration of 68 mM. This sample was loaded onto a nickel affinity chromatography column (ProBond, 20 mL, Life Technologies) equilibrated in 50 mM MOPS, 500 mM NaCl, 50 mM imidazole, 0.5 mM PMSF, 0.003% LMNG, pH 7.0, at a flow rate of 3 mL/min. The column was washed with the same buffer containing 140 mM imidazole until the absorbance dropped to the baseline. Bound proteins were eluted with 260 mM imidazole. Peak fractions containing bd-II were pooled and concentrated to 0.5 mL via ultrafiltration (4°C, 3800 rpm, 3000 × g, 30 min, Rotor A-4-44, Eppendorf Centrifuge 5804 R, Amico Ultra, 100,000 MWCO, Millipore). The concentrated protein solution was centrifuged (4°C, 14,000 rpm, 20,800 × g, Eppendorf Centrifuge 5417R) and applied to size exclusion chromatography, using a HiLoad 16/60 Superdex 200 pg column (GE Healthcare), equilibrated in 20 mM MOPS, 20 mM NaCl, 0.5 mM PMSF, 0.003% LMNG, pH 7.0 at a flow rate of 0.3 mL/min. Fractions containing the bd-II were pooled and concentrated to 24 mg/mL. Finally, the protein was shock frozen in liquid nitrogen and stored at −80°C. The bd-II oxidase wild-type and the D239N AppC variant were prepared according to the same protocol as the proteins produced in BL21* Δcyo. Proteins from strain CBO eluted at identical position from the columns, showed an Fig. 7 Proposed proton pathways to heme d. a shows the short channel in bd-II that begins at the end of the large, hydrophilic cavity, b the longer proton pathway along a series of titratable amino acid residues in bd-I and c the two proposed extended proton pathway in G. thermodenitrificans bd oxidase 25 . Putative proton pathways are indicated by red arrows that connect titratable residues. The proposed second pathway in G. thermodenitrificans bd oxidase is indicated by a blue arrow.
identical redox difference spectrum and the same protein band pattern after SDS-PAGE as the proteins produced in strain BL21*Δcyo. The G100A bd-I variant was prepared according to ref. 22 .
Amphipol exchange. For exchanging the detergent LMNG with amphipol, bd-II was supplemented with a threefold mass excess of amphipol A8-35 (Anatrace, Maumee, OH, USA). The sample was incubated at 4°C for 2 h while stirring. Excess detergent was removed by an addition of a 40-fold mass excess BioBeads (SM-2 resin, BioRad, Feldkirchen, Germany) and further incubation of 3 h at 4°C under stirring. Excess amphipols were removed by size exclusion chromatography (Superose 6 Increase 10/300 GL, GE Healthcare), equilibrated in 20 mM MOPS, 20 mM NaCl, pH 7.0. Peak fractions were pooled and used for further analysis. For cryo-EM the samples were supplemented with 1 μM aurachin D prior to placing them onto the cryo grids.
Cryo-electron microscopy and data analysis. Glow-discharged cryo grids were prepared with 2.5 µL protein solution (5.6 mg/mL) on R1.2/R1.3 Quantifoil holey gold grids. Grids were incubated for 45 s and blotted for 6.5 s prior to plunging into liquid ethane and flash-frozen using a Vitrobot (FEI Company). Cryo-EM recordings were made at the EM Facility at the Rudolf Virchow Center, Würzburg, Germany, with a Titan Krios G3 (Thermo Fisher Scientific) equipped with a Falcon III direct detector. Micrographs were taken at 300 kV with a ×75,000 magnification and a calibrated pixel size of 1.0635 Å. Movies with 47 frames each were collected automatically with EPU at an exposure time of 74.99 s and a total electron exposure of 79 e − /Å 2 . The final data set contained 1836 movies with a targeted under-focus between 1.4 and 2.4 µm (Supplementary Table 1). All movies were dose weighted and motion-corrected with Motioncorr 49 , CTFs were determined using CTFFIND-4.1 50 . Convolutional neuronal network-based particle picking of all micrographs was conducted with crYOLO 51 using the provided general network. All following steps ( Supplementary Fig. 3) were conducted with Relion 3.1.0 52 . The picked particle positions were imported in Relion and particle images were extracted, followed by a 2D classification. The resulting 2D classes served as template for auto-picking, which resulted in~800,000 picked particles. The particles were further 2D classified in three rounds. In the first two rounds the particles were arbitrarily separated in six particle subsets and in the third round the particles were separated in three particle subsets. After each round, classes that did neither contain contaminations or empty classes were further processed. These 2D classifications yielded a final particle set of~100,000 particles. With these particles an initial model was obtained with the InitialModel option in Relion (initial resolution 10 Å and final target resolution 5 Å to ensure the occurrence of helix densities; initial mini-batch size 1000 and final mini-batch size 10,000 for better signal-tonoise). The remaining particles were further refined and purged from residual artifact particles with a 3D classification on the initial model to a subset of~40,000 particles. The respective 3D class was taken as model for a 3D auto-refinement (in several rounds, later with an imposed C2 symmetry). The remaining~40,000 particles were 2D classified and the resulting class averages were used as template for another round of particle picking. These new particles were again 2D and 3D classified, and the resulting 120,000 particles were used for a final 3D autorefinement with following post-processing imposing C2 symmetry to a resolution of 3 Å. The cryo-EM map was taken for protein modeling in COOT 0.8.9.1 53 . The homologous bd-I structure (PDB: 6RX4) was fitted into the map as starting model and mutated to the respective sequence. The model was adjusted to the map and refined with Phenix realspace.refine 52 , followed by a validation by MolProbity 4.4 54 . Model figures were produced with PyMOL 2.4.1.
UV-vis spectroscopy. UV-vis difference spectra were recorded with a diode array spectrometer (TIDAS II, J&M Analytik AG). 200 µg protein was diluted in buffer (20 mM MOPS, 20 mM NaCl, 0.003% LMNG, pH 7.0) and a reference spectrum was recorded. The same solution was subsequently reduced with a few grains of dithionite and a UV-vis spectrum of the reduced sample was recorded. The redox spectrum was obtained by subtracting the spectrum of the oxidized sample from that of the reduced one. To determine the binding of CN − , a reference spectrum was recorded, followed by the addition of 0.5 mM KCN to the sample, while spectra were continuously recorded every 30 s. After 30 min, the cyano-complex was reduced by an 80-fold molar excess (0.5 mM) dithionite and the spectrum of the reduced cyano-complex was recorded. UV-vis differential titration. The UV-vis oxidative and reductive titrations were carried out in a thin-layer electrochemical cell as described 23,29 with a dual-beam VARIAN Cary 300 spectrometer coupled to a potentiostat. A gold grid modified with a solution of positively and negatively charged thiols (1:1 solution of cysteamine and mercaptopropionic acid) was used as working electrode, a platinum contact and an aqueous Ag/AgCl 3 M KCl served as counter and reference electrodes, respectively. The E. coli bd-II was concentrated in 100 mM KPi, 50 mM NaCl, 0.05% DDM, pH 7.0 to 0.8 mM and 1 h prior to the experiments, the solution was incubated with a mixture of 19 mediators with a final concentration of 25 μM (Supplementary Table 3). The protein sample was deposited on the gold grid and the cell was closed with CaF2 windows and filled with the buffer. All redox difference spectra were recorded at 13°C in the controlled-potential range of +500/ −500 mV vs. Ag/AgCl. The equilibration time for each potential step was at least 30 min and multiple spectra were recorded to ensure the stabilization of the redox state. The redox potentials of hemes were determined from the modified Nernst equation applied to the experimental titration curves presented as a plot of the absorbance at 562 and 629 nm, each, versus the applied potential.
Inhibition of bd-II oxidase by aurachin C and D. The duroquinol:oxygen oxidoreductase activity of isolated E. coli bd-II oxidase and its inhibition by aurachins C and D were determined by monitoring the O 2 consumption with a Clark-type oxygen electrode (Oxyview System, Hansatech Instruments). All steps were conducted at 30°C. The reaction chamber was filled with 2 mL buffer (20 mM MOPS, 20 mM NaCl, 0.003% LMNG, pH 7.0), 10 µL DTT (1 M) and 5 µL duroquinone in ethanol (100 mM) and incubated for 1 min. 10 µL bd-II in LMNG buffer (2 mg/mL) were added to start the reaction. After 2 min, the reaction was inhibited by adding various amounts (1-300 nM) of aurachin C and aurachin D, respectively. The enzymatic activity was corrected for the non-enzymatic activity that was about 5% of the entire activity. Each data point was measured in triplicates. The standard deviation of each measurement is provided.
Dynamic substrate channel analysis. As starting point for substrate channel analysis, a conformational ensemble was obtained for both E. coli bd-I and bd-II through molecular dynamics (MD) simulations with the programs of the Schrödinger software Suite 2021-1 (Schrödinger LLC, NY). First, the cryo-EM structures of bd-I (PDB ID: 6RX4) and bd-II were prepared with PrepWizard 55 by adding missing hydrogen atoms, adjusting bond orders and formal charges of smallmolecule ligands. PROPKA 56 was used to generate the ionization states of polar amino acids at pH 7.0. The structures were energy-minimized using the OPLS4 force field 57 . MD simulations were carried out with Desmond 58 . The optimized structures were embedded in POPC membranes within orthorhombic periodic boundary systems and solvated with TIP3P water molecules, while neutralizing the charge of the systems with counter-ions. After an equilibration protocol, the production of MD simulations was conducted for 100 ns in an NPT ensemble at 300 K regulated by a Nosé-Hoover thermostat and a Martyna-Tobias-Klein barostat. Atomic coordinates were recorded at an interval of 400 ps, for a total 250 frames for each trajectory. Substrate pathways to the active site were analyzed with Caver 3.0 42 via Caver Analyst 2.0 59 graphical user-interface. First, the MD frames were stripped of the counter ions, POPC membranes and TIP3P water molecules and used as input for channel computation. During the computation, all protein and ligand atoms were treated as hard spheres. The default threshold value of 0.9 Å was used as bottleneck radius to identify the most probable channels. For the identified tunnels, the residues lining them were equally listed.