Cryo-EM Structures of Respiratory bc1-cbb3 type CIII2CIV Supercomplex and Electronic Communication Between the Complexes

The respiratory electron transport complexes convey electrons from nutrients to oxygen and generate a proton-motive force used for energy (ATP) production in cells. These enzymes are conserved among organisms, and organized as individual complexes or combined forming large super-complexes (SC). Bacterial electron transport pathways are more branched than those of mitochondria and contain multiple variants of such complexes depending on their growth modes. The Gram-negative species deploy a mitochondrial-like cytochrome bc1 (Complex III, CIII2), and may have bacteria-specific cbb3-type cytochrome c oxidases (Complex IV, CIV) in addition to, or instead of, the canonical aa3-type CIV. Electron transfer between these complexes is mediated by two different carriers: the soluble cytochrome c2 which is similar to mitochondrial cytochrome c and the membrane-anchored cytochrome cy which is unique to bacteria. Here, we report the first cryo-EM structure of a respiratory bc1-cbb3 type SC (CIII2CIV, 5.2Å resolution) and several conformers of native CIII2 (3.3Å resolution) from the Gram-negative bacterium Rhodobacter capsulatus. The SC contains all catalytic subunits and cofactors of CIII2 and CIV, as well as two extra transmembrane helices attributed to cytochrome cy and the assembly factor CcoH. Remarkably, some of the native CIII2 are structural heterodimers with different conformations of their [2Fe-2S] cluster-bearing domains. The unresolved cytochrome c domain of cy suggests that it is mobile, and it interacts with CIII2CIV differently than cytochrome c2. Distance requirements for electron transfer suggest that cytochrome cy and cytochrome c2 donate electrons to heme cp1 and heme cp2 of CIV, respectively. For the first time, the CIII2CIV architecture and its electronic connections establish the structural features of two separate respiratory electron transport pathways (membrane-confined and membrane-external) between its partners in Gram-negative bacteria.


Introduction
Mitochondrial and bacterial respiratory chains couple exergonic electron transport from nutrients to the terminal acceptor oxygen (O 2 ) through a set of enzyme complexes. Concomitantly, they generate a proton motive force used for ATP synthesis and other energy-dependent cellular processes. The mitochondrial respiratory chain consists of four complexes. Complex I (NADH dehydrogenase) and Complex II (succinate dehydrogenase) are the entry points of reducing equivalents (NADH and FADH 2 ) derived from nutrients into the chain. They reduce the hydrophobic electron carrier quinone (Q). Reduced quinone (QH 2 ) moves rapidly within the membrane to Complex III (cytochrome (cyt) bc 1 or CIII 2 ) which oxidizes it and reduces the electron carrier cyt c. The reduced cyt c diffuses to Complex IV (cyt c oxidase or CIV) which oxidizes it and subsequently reduces the terminal electron acceptor oxygen to water (Nicholls and Ferguson, 2013) (Fig. 1A).
Respiratory complexes are evolutionarily conserved among organisms, but bacterial enzymes are structurally simpler than their mitochondrial counterparts, consisting mainly of the catalytic subunits.
However, bacterial respiratory chains are more elaborate than those of mitochondria, since they contain various complexes forming branched pathways to accommodate their diverse growth modes (Melo and Teixeira, 2016). In facultative phototrophs, the mitochondrial-like bc 1 -type CIII 2 is central to respiratory and photosynthetic electron transport pathways. CIII 2 is a dimer with each monomer comprised of three subunits: the Rieske FeS (FeS) protein with a [2Fe-2S] cluster, cyt b with hemes b H and b L , and cyt c 1 with heme c 1 cofactors (Fig. 1A,B). The FeS protein external domain (FeS-ED) is mobile between the b (close to heme b L ) and c (close to heme c 1 ) positions (Darrouzet et al., 2001;Esser et al., 2006). Some species such as Rhodobacter sphaeroides contain a mitochondrial-like aa 3type and a bacteria-specific cbb 3 -type CIV, a monomer comprised of four subunits: CcoN with heme b and heme b 3 -Cu binuclear center, CcoO with heme c o , CcoQ, and CcoP with hemes c p1 and c p2 cofactors (Fig. 1A,B). Other species such as Rhodobacter capsulatus (Khalfaoui-Hassani et al., 2016) 5 and pathogens like Helicobacter pylori and Campylobacter jejuni (Smith et al., 2000), Neisseria (Aspholm et al., 2010) have only a high oxygen affinity cbb 3 -type CIV to support their microaerophilic growth.
Conversely, Gram-positive bacteria are devoid of freely diffusing electron carriers. Instead, they may have additional cyt c domains fused to their CIII 2 (i.e., bcc-type) such as in Mycobacterium smegmatis (Kim et al., 2015) and Corynebacterium glutamicum (Niebisch and Bott, 2003) or CIV (i.e., caa 3type) such as in Bacillus subtilis (Winstedt and von Wachenfeldt, 2000) and Bacillus stearothermophilus (Sakamoto et al., 1996). Bacterial electron carrier cyts c are involved in multiple metabolic pathways. Both the diffusible cyt c (e.g., R. capsulatus cyt c 2 or its homologs) and the membrane-anchored cyt c (e.g., R. capsulatus c y or its homologs) electronically connect CIII 2 to the photochemical reaction center in photosynthesis (Daldal et al., 1986), and to CIV in respiration (Hochkoeppler et al., 1995). In species like R. sphaeroides, cyt c 2 functions in both photosynthesis and respiration, while cyt c y is restricted to respiration (Myllykallio et al., 1999).
As of yet, no respiratory SC structure has been determined for Gram-negative bacteria, the evolutionary precursors of mitochondria. Furthermore, SCs containing ancient forms of CIV (i.e., cbb 3 -type) representing primordial features of respiratory chains with multiple electron carriers are unknown (Ducluzeau et al., 2008). Structural studies of such SCs have been hampered due to unstable interaction between CIII 2 and CIV, hence their trace amounts in nature. We have overcome this hurdle using a genetic approach, yielding large amounts of SCs from the Gram negative facultative phototroph R. capsulatus. Here, we report the first cryo-EM structure of a respiratory bc 1 -cbb 3 type SC (CIII 2 CIV, at 5.2Å resolution), as well as several conformers of native CIII 2 (at 3.3-4.2Å resolution). We define the interaction regions of cyt c 2 and cyt c y within the SC by combining cryo-EM, cross-linking mass spectrometry (XL-MS) and integrative structure modeling. We propose that the membrane-bound cyt c y donates electrons to heme c p1 , while the diffusible cyt c 2 transfers them to heme c p2 , of CcoP subunit of CIV. For the first time, this work establishes the structural features of CIII 2 CIV and its two distinct respiratory electron transport pathways (membrane-confined and membrane-peripheral) connecting its partners in Gram-negative bacteria.

Stabilization, isolation, and composition of functional fused SCs.
Earlier studies on soluble cyt cindependent electron transport pathways have indicated that in some species (e.g., R. capsulatus (Myllykallio et al., 2000)), CIII 2 , CIV, and the membrane-anchored cyt c y are in close proximity to each other. BN-PAGE of membranes from a wild type strain of R. capsulatus, overstained for CIVspecific in-gel activity, showed barely detectable bands around ~450 kDa M r (Fig. S1A). The masses of these bands were larger than that of the CIV monomer (~100 kDa, running as ~230 kDa on BN-PAGE) or CIII 2 dimer (~200 kDa, running as >250 kDa on BN-PAGE), suggesting the occurrence of large SCs. However, these entities were of low abundance and highly unstable, rendering their study difficult. In our earlier work, translationally fusing cyt c 1 of CIII 2 to cyt c y had produced an active bcc-type CIII 2 (i.e., cyt bc 1 -c y fusion) , suggesting that this approach might also be used to stabilize the interactions between CIII 2 and CIV.
During the assembly processes of CIII 2 and CIV, cyt c 1 interacts with cyt b to form a cyt b-c 1 subcomplex (Davidson et al., 1992), and CcoP associates with CcoNOQ subcomplex to yield an active CIV (Kulajta et al., 2006). We thought that translationally fusing the C-terminus (C-ter) of cyt c 1 to the N-terminus (N-ter) of CcoP, which are on the inner (n) side of the membrane, forming a bipartite cyt c 1 -CcoP fusion protein might produce a stable bipartite bc 1 -cbb 3 type SC (left panels of Fig. 1C,D). Furthermore, adding the natural 69-residue linker (L) and the 100-residue cyt c domain of c y to the C-ter of cyt c 1 -CcoP, which is on the outer (p) side of the membrane, to form a tripartite cyt c 1 -CcoP-c y fusion protein might yield a tripartite bc 1 -ccbb 3 type SC with an attached electron carrier (right panels of Fig. 1C,D). This approach (see Supplemental Information, Methods, for details) yielded fusion constructs (Fig. S1B) that functionally complemented a mutant lacking CIII 2 and CIV for photosynthesis-proficiency (i.e., CIII 2 activity) and CIV activity (Fig. S1C).

8
The His-tagged bipartite and Flag-tagged tripartite SCs were purified from detergent-dispersed membranes by tag-affinity and size exclusion chromatography (SEC) (SI, Methods) ( Fig. 2A,B). BN-PAGE of isolated proteins showed that the A-1 and B-1 fractions contained mostly the large entities of M r ~450 kDa range ( Fig. 2A,B, insets), and SDS-PAGE revealed that they had the cyt c 1 -CcoP (~65 kDa) or cyt c 1 -CcoP-c y fusion proteins (~80 kDa) (Fig. 2C). All protein bands seen in Fig. 2C were identified by mass spectrometry (MS) ( Table S3) and assigned to the subunits of CIII 2 and CIV.
Remarkably, the cyt c domain of c y fused to cyt c 1 -CcoP transferred electrons from CIII 2 to CIV.

Structures of the tripartite SCs.
We first focused on cryo-EM analysis of the tripartite SC preparations that were more stable and abundant than the bipartite SCs (Fig. 2B, fraction B-1). Initial 3D classes were of primarily two different sizes (Fig. S3, Box 1, left). The smaller (~180Å length) particles were asymmetrical, and their size and shape suggested that they may correspond to a dimeric CIII 2 associated with a single CIV. Focused classification and processing of the subclass containing ~62,000 particles with the highest initial resolution, and best discernable features, led to a tripartite CIII 2 CIV map (SC-1A, EMD-22228) at 6.1Å resolution (Fig. S3A, see SI Methods for details), while another dataset yielded a slightly lower resolution map (SC-1B, EMD-22230) at 7.2Å ( Fig. S3B) ( Table 1). The larger particles (~250Å length, Fig. S3, Box 1) were more symmetrical and represented a dimeric CIII 2 flanked by two CIV (i.e., CIII 2 CIV 2 ), as expected based on two c 1 -CcoPc y subunits per CIII 2 . However these particles were rare (~5,000) and their map (SC-1C) could not be refined beyond ~10Å resolution (Fig. S3C).
The R. capsulatus cbb 3 -type CIV is highly homologous to that of P. stutzeri but not identical (see Methods for details). Thus, a homology model of CIV was built using the P. stutzeri structure (PDB: 3MK7; 3.2Å resolution) as a template and validated (Table S7) (SI, Methods). In addition, the existing CIII 2 model (PDB: 1ZRT; 3.5Å resolution) was further refined (PDB: 6XI0; 3.3Å resolution) using our cryo-EM data (see below and Table 2). These models were fitted as rigid bodies into the maps SC1-A with a correlation coefficient CC box of 0.75 and SC-1B with a correlation coefficient CC box of 0.71 ( Fig. S4A) ( Table 1). The [2Fe-2S] clusters of the FeS proteins of CIII 2 could be recognized closer to heme b L (b position) than to heme c (c position), but had lower occupancy and resolution likely due to conformational heterogeneity (Fig. S4B). In particular, the heterogeneity of the FeS-ED in monomer A (i.e., adjacent to CIV) was more pronounced than that in monomer B (i.e., away from CIV) of CIII 2 . Lower resolutions of the FeS-ED portions were anticipated because of their mobility (Darrouzet et al., 2001;Esser et al., 2006). Details of the tripartite CIII 2 CIV structure are described below together with the bipartite SC, which has a higher resolution.
Superimposition of the CIII 2 portions of SC-1A and SC-1B maps showed that CIV was in different orientations in different maps (Fig. S4C). The two extreme locations of CIV with respect to CIII 2 were displaced from each other by a translation of ~3Å and a rotation of ~37 degrees (Fig. S4D, E; SC-1A in red, and SC-1B in blue). Other subclasses identified in 3D classifications showed CIV in slightly different orientations between those seen in SC-1A and SC-2B maps. This variable rotation of CIV around CIII 2 is attributed to the limited interaction interface between the CcoP (N-ter TMH) of CIV and the cyt b (TMH7) of CIII 2 (see Fig. 3C), indicating that the CIII 2 CIV interface is flexible.
In the interface regions of SC-1A and SC-1B maps, additional weaker features that are not readily attributable to CIII 2 and CIV structures were also observed. Intriguingly though, no membraneexternal features corresponding to cyt c domain of c y , which is an integral part of the cyt c 1 -CcoP-c y subunit of tripartite CIII 2 CIV, could be discerned in these maps.
Structure of bipartite SC supplemented with cyt c y . In an attempt to locate the cyt c domain of c y , the bipartite SC preparations devoid of it ( Fig. 2A, fraction A-1) were supplemented with either purified full-length cyt c y , or with its soluble variant lacking the TMH (i.e., cyt S-c y ) , to yield the bipartite SC+c y and SC+S-c y samples. Following SEC, the elution fractions analyzed by SDS-PAGE showed that only the intact cyt c y , but not the cyt S-c y , associated with the SC (Fig. S5A). Thus, the cyt c domain of c y does not bind tightly to, and its TMH is required for association with, this SC.
The cryo-EM analyses of the bipartite SC+c y samples were carried out as above, and yielded a map (SC-2A, EMD-22227) at 5.2Å resolution (Fig. S6A,B), with local resolutions ranging from 4.3-8.0Å (Fig. S7A,C). The homology model of CIV and the refined model of CIII 2 (PDB: 6XI0) were fitted as rigid bodies into SC-2A with a correlation coefficient CC box of 0.74 ( Fig. 3A) ( Table 1).
Comparison of SC-2A (bipartite CIII 2 CIV) with SC-1A (tripartite CIII 2 CIV) maps showed that they were highly similar with RMSD of 1.6 Å. They are collectively referred to as CIII 2 CIV, irrespective of their bipartite or tripartite origins.
The dimensions of the slightly curved CIII 2 CIV structure (~155x60x90Å, LxWxH) were consistent with a CIII 2 dimer associated with one CIV. On SC-2A map at 5.2Å resolution, some large aromatic side chains could be discerned (Fig. 3B), and of the TMHs seen, 34 accounted for by two FeS proteins, two cyts b and two cyts c 1 (2, 16 and 2 TMHs per dimer, respectively) of CIII 2 , and single CcoN, CcoO and CcoP (12, 1 and 1 TMHs, respectively) of CIV (Fig. 3C). The features corresponding to the heme cofactors of CIII 2 CIV were readily attributed to hemes b H and b L of cyt b, heme c 1 of cyt c 1 , and to hemes b and b 3 of CcoN, heme c of CcoO and hemes c p1 and c p2 of CcoP proteins. As seen with the tripartite maps, the [2Fe-2S] clusters of CIII 2 could be recognized closer to heme b L (in b position), but had lower resolution because of conformational heterogeneity.
An additional TMH was observed at the distal end of CIV (Fig. 3A, rotated 180 degrees in Fig.   4A) close to CcoN TMH3 and TMH4 (Fig. 3C). Due to its location, this TMH (depicted in Fig. 3 and Fig. 4 as an ab initio model of the CcoN Arg25-Leu48 residues generated by I-TASSER ) was tentatively attributed to the extra N-ter TMH (i.e., TMH0) of CcoN (Fig. 4B).
The interface of CIII 2 CIV is roughly delimited by CcoN TMH8 and TMH9, CcoP TMH, cyt b-TMH5 and TMH7, and cyt c 1 C-ter TMH of monomer A, with the closest interaction being between CcoP TMH and cyt b TMH7 (Fig. 3A,C). Two highly confined inter-complex connections and two interacting TMHs of unknown identities were present at the interface (Fig. 4A, red and blue TMHs).
One such connection was at the n face of the membrane, near the cyt c 1 and CcoP TMHs (Fig. 4C, Lys257 c1 and Thr13 CcoP ). These subunits being covalently linked, the connecting feature in the map was tentatively attributed to their junction linking CIII 2 and CIV.

12
The assembly factor CcoH and cyt c y TMHs are located at CIII 2 CIV interface. The identities of the unknown TMHs at the interface of CIII 2 CIV ( Fig. 3 and Fig. 4) were sought using a co-evolution based approach, RaptorX-ComplexContact (Zeng et al., 2018), predicting the residue-residue contacts in protein-protein interactions. All known single TMH containing CIV-related proteins (i.e., CcoQ subunit, CcoS and CcoH assembly factors (Koch et al., 2000) and cyt c y (Myllykallio et al., 1997)) were analyzed against all subunits of CIII 2 and CIV. Significant predictions of interacting residue pairs (confidence value >0.5) were observed only between CcoN (primarily TMH9) and the putative CcoH N-term TMH (Table S4). An ab initio model of CcoH TMH (Fig. S8A, residues 11 to 35) was generated by I-TASSER , and docked onto CIV using PatchDock (Schneidman-Duhovny et al., 2005) with the predicted residue-residue contacts as distance restraints (15 Å threshold) and without using the corresponding cryo-EM maps (SI, Methods). The top scoring models converged to a single cluster around the location of the unknown TMH, close to CcoN TMH9 at CIII 2 CIV interface (Fig. S8B). Close examination of the interactions between CcoH TMH and CcoN TMH9 showed that multiple co-evolutionarily conserved residues are in close contacts ( Fig.   S8C). Earlier studies had suggested that CcoH is near the CcoP and CcoN, to which it can be crosslinked by disuccinimidyl suberate (DSS, spacer length ~11Å) (Pawlik et al., 2010). Thus, the unknown TMH located close to CcoN TMH9 ( Fig. 3C and Fig. 4C, blue TMH) was tentatively assigned to the assembly factor CcoH.
An important difference between the maps of the bipartite CIII 2 CIV+c y (SC-2A) and tripartite CIII 2 CIV (SC-1A) was in the features corresponding to the unidentified TMHs at the interface. These densities were barely visible in SC-1A, but highly enhanced in SC-2A (Fig. 4C), indicating higher occupancy. The observation that only the native cyt c y binds to bipartite SC via its TMH (not its cyt c domain, i.e., cyt S-c y ), suggested that the TMH (red in Fig. 4C), next to CcoH TMH (blue in Fig.  4C), may correspond to the membrane-anchor of cyt c y . This explanation is most plausible since the bipartite CIII 2 CIV+c y samples were supplemented with full-length cyt c y while the tripartite samples contained only the fused cyt c domain but not the TMH. Indeed, landmark densities corresponding to the helix-breaking Gly11 and two correctly spaced bulky sides chains of Phe15 and Tyr21 of cyt c y TMH (NH2-xxxGly11xxxPhe15xxxxxTyr21-COOH) were discerned (Fig. 4D).
Additionally, some CIII 2 CIV+c y subclasses exhibited a weak feature on the p side of the membrane that may reflect the cyt c domain of c y (Fig. S6G, SC-2B). However, this feature could not be refined to high resolution, consistent with the weak binding of cyt c domain of c y to CIII 2 CIV (Fig.   S5A). Moreover, the predominant conformation of CIV in the bipartite CIII 2 CIV+c y (Fig. S6A,B, SC-2A) shifted towards that found in SC-1A map of tripartite SC (Fig. S3A), with no major class corresponding to SC-1B. This suggested that the local interactions between the CcoH and cyt c y TMHs and CIV decreased the interface flexibility of CIII 2 CIV (Fig. 4C).
Cryo-EM structures of R. capsulatus native CIII 2 . During this study we noted that the bipartite SC+c y samples contained large amounts of smaller particles (~110Å length, Fig. S3, Box 2) that were the size of CIII 2 (Fig. S6C,D). Analyses of these particles using C2 symmetry led to the map CIII 2 at 3.3Å resolution for native CIII 2 (Fig. S6E), with local resolutions ranging from 3.0 to 4.0Å (Fig.   S7B,D) ( Table 2). The FeS-ED parts showed a lower occupancy and resolution compared to the rest of the map, indicating conformational heterogeneity. Interestingly, when similar analyses were carried out without imposing C2 symmetry, three distinct maps (CIII 2 c-c, CIII 2 b-c and CIII 2 b-b) for CIII 2 were obtained at 3.8, 4.2 and 3.5Å resolutions, respectively (Fig. S6F). These maps were superimposable with respect to cyt b and cyt c 1 subunits, except for the FeS-ED portions. The CIII 2 structures depicted by the CIII 2 b-b ( Fig. 5A-C) and CIII 2 c-c (Fig. 5D) maps exhibited overall C2 symmetry, but in the former the FeS-EDs were located in b, whereas in the latter they were in c 14 position (Esser et al., 2006). Notably, the third structure (Fig. S6F, CIII 2 b-c) was asymmetric, with the FeS-ED of one monomer being in c, and the other in b positions (Fig. 5E). Such asymmetric structures of native CIII 2 have been rarely seen using crystallographic approaches, although proposed to occur during QH 2 oxidation by CIII 2 (Castellani et al., 2010;Cooley et al., 2009;Covian and Trumpower, 2005). Similar low occupancy and resolution of the FeS-EDs, suggesting conformational heterogeneity, were also seen with the CIII 2 CIV maps.
Interactions of cyt c 2 and cyt c y with CIII 2 CIV. The interaction interfaces between CIII 2 CIV and its physiological electron carriers were pursued using cross-linking mass spectrometry (XL-MS) (Gotze et al., 2015;Slavin and Kalisman, 2018). First, the co-crystal structure (PDB: 3CX5) of yeast CIII 2 with its soluble electron carrier iso-1 cyt c (Solmaz and Hunte, 2008) was used as a template (homology between yeast cyt c 1 and R. capsulatus cyt c 1 : 31% identity and 58% similarity; iso-cyt c and cyt c 2 : 25% identity and 56% similarity) to model the binding of cyt c 2 on bacterial CIII 2 . As the co-crystal structure contains only one iso-1 cyt c bound to one of the two cyt c 1 of yeast CIII 2 , R. capsulatus cyt c 1 (PDB: 6XI0) and cyt c 2 (PDB: 1C2N) structures were superimposed with their counterparts on the co-crystal structure, and a model with a single cyt c 2 docked to one monomer of CIII 2 was generated. To experimentally verify this model, the protein cross-linker 4-(4,6-dimethoxy-1,3,5-triazin-2-yl)-4-methyl-morpholinium chloride (DMTMM) was used with R. capsulatus cyt c 2 bound to CIII 2 (SI, Methods). Multiple intra-subunit cross-links (XLs) within CIII 2 CIV detected in several experiments served as controls (Table S5 and Fig. S9A). High-confidence XLs were obtained using both FindXL (Kalisman et al., 2012) and MeroX (Iacobucci et al., 2018) search engines, and only those identified by both were retained. The three XLs between cyt c 1 and cyt c 2 provided distance restraints (~30Å for DMTMM) for docking cyt c 2 to CIII 2 using PatchDock (Table S5 and Fig. S9B).
The docking models clustered at a single region per monomer of CIII 2 (Fig. 6A, right), which overlapped with the binding site of cyt c 2 defined by the model generated by alignment to the yeast co-crystal structure (Fig. S9C). The distance from cyt c 2 heme-Fe to cyt c 1 heme-Fe is ~16.8Å for the co-crystal derived model, while comparable distances between ~13.8 -20.4Å were obtained with the docking models. Thus, docking with Patchdock integrating XL-MS based distance restraints defined reliably, but with limited accuracy, the interaction region of cyt c 2 on CIII 2 .
No information about the binding sites between cyt c 2 and cbb 3 -type CIV was available, so the XL-MS with DMTMM was extended to this case. Similarly, the XLs found between the proteins (1 between cyt c 2 and CcoP, and 8 between cyt c 2 and CcoO) provided distance restraints for docking cyt c 2 to CIV via Patchdock ( Table S5). The cyt c 2 docking models also clustered in a single region of CIV (Fig. 6A, left), closer to heme c p2 (c 2 heme-Fe to c p2 heme-Fe: ~15.2 to 35.6Å) than heme c p1 (c 2 heme-Fe to c p1 heme-Fe: ~23.0 to 42.0Å) of CcoP subunit (Fig. 7). Surface charge complementarities between the positively charged face of cyt c 2 and the negatively charged likely binding regions on both CIV and on CIII 2 are seen (Fig. S10A). These two cyt c 2 binding regions on CIII 2 CIV are distant from each other (closest c 2 heme-Fe on CIII 2 to that on CIV is ~69Å) (Fig. 7A).
Next, the binding interactions between cyt c domain of c y and CIII 2 CIV were addressed using DMTMM and disuccinimidyl dibutyric urea (DSBU) as cross-linkers. Similar to DMTMM, DSBU yielded multiple intra-subunit XLs within the subunits of CIII 2 CIV, providing experimental controls (Table S6 and Fig. S9D). Six XLs (five cyt c y to cyt c 1 and one cyt c y to FeS protein) with DMTMM (Table S5) and four XLs (only cyt c y to FeS protein) with DSBU (Table S6) were identified. Although chemically different cross-linkers were used, XLs were observed only between cyt c y and CIII 2 , and not with CIV, suggesting that this cyt c domain is closer to CIII 2 in CIII 2 CIV. Using the XLs as distance restraints (~35Å for DSBU and ~30Å for DMTMM) PatchDock generated two binding clusters for cyt c domain of c y on each CIII 2 monomer of SC. One of the clusters was on cyt c 1 , overlapping with the binding region of cyt c 2 (Fig. 6B), whereas the other one was located between cyt c 1 and the FeS-ED near the inter-monomer region of CIII 2 (Fig. S11). To further support these binding locations obtained by XL-MS-based docking, we sought classes that have extra densities corresponding to cyt c domain of c y in our cryo-EM datasets, and found a minor 3D class containing ~18,000 particles (Fig. S6G), which has an extra feature between CIV and CIII that may be attributable to this domain (Fig. S11). The two docking clusters, clearly visible in top view (Fig.   S11C), were more spread out compared with those of cyt c 2 (Fig. 6A, Fig. 7A,C), with the distances between cyt c y heme-Fe and cyt c 1 heme-Fe of CIII 2 monomer A being between 13.8 to 47.1Å, consistent with the weak binding of cyt c domain of c y .
Patchdock mediated docking of cyt c domain of c y was also performed with the same XLs as above but using the conformers of native CIII 2 with differently located FeS-EDs ( Fig. 5C-E, CIII 2 bb, c-c and b-c). The data showed that when the FeS-EDs are in c position (CIII 2 c-c), the docking models gathered as a single cluster on cyt c 1 , slightly displaced towards the FeS-ED of the same monomer ( Fig. S12A-C). However, when the FeS-EDs are in b position (CIII 2 b-b), such models were more spread out (Fig. S12D-F). The third model with one FeS-ED in c and the other in b positions showed the expected clustering pattern depending on the local FeS-ED conformation. As in the SC both FeS-EDs appear to be in the b position, we assume that the docking pattern of cyt c domain of c y is like that seen with CIII 2 b-b. Thus, the relatively spread docking position observed with SC ( Fig.   7, Fig. S11) was attributed to variable conformations of the FeS-EDs on CIII 2 . Furthermore, since heme c 1 , and not the FeS protein, is the electron exit site of CIII 2 (Crofts et al., 2008;Osyczka et al., 2005), the cluster on cyt c 1 was taken as the productive binding region of cyt c domain of c y .
Examination of all pertinent distances between the cofactors of CIII 2 CIV (Fig. 7A) indicates that the binding region of cyt c domain of c y near heme c 1 of CIII 2 is far away from the expected electron entry point(s) of CIV. The large distance (~50.8Å) separating cyt c 1 heme-Fe of CIII 2 monomer A from CcoP c p1 heme-Fe (the closest compared with heme c p2 of CIV) renders it impossible to define a location for cyt c y close enough to heme c 1 reducing it, and heme c p1 oxidizing it, to sustain productive electron transfer from CIII 2 to CIV. This distance constraint, the inability to resolve the cyt c domain of cyt c y by cryo-EM, and the higher frequency of XLs to CIII 2 strongly infer that the cyt c domain of c y must oscillate to carry out soluble carrier-independent electron transfer within CIII 2 CIV to couple QH 2 oxidation to O 2 reduction (Fig. 8).

Discussion
Prior to this work, no structural information was available on any bacterial cbb 3 -type CIV containing SC, or on its interactions with its physiological redox partners. Here, we describe the first cryo-EM structures of CIII 2 CIV, a bc 1 -cbb 3 type respiratory SC from the Gram-negative, facultative phototroph R. capsulatus. We also define the likely binding regions of the electron carriers cyt c 2 and cyt c y to CIII 2 CIV, and report the structures of both homo-and hetero-dimeric conformers of native CIII 2 .
Although X-ray based structures of bacterial bc 1 -type CIII 2 are available, native CIII 2 heterodimers have not been observed frequently. Similarly, only a single structure, that of P. stutzeri (Buschmann et al., 2010), was available for cbb 3 -type CIV. Members of this subfamily of heme-Cu:O 2 reductases are widespread among bacteria and essential for major micro-aerobic processes, including anaerobic photosynthesis, nitrogen fixation, symbiosis and bacterial infection (Khalfaoui-Hassani et al., 2016).
Unlike the obligate CIII 2 CIV 2 SC of Actinobacteria, which is rigid and devoid of a free electron carrier (Gong et al., 2018;Wiseman et al., 2018), the R. capsulatus facultative CIII 2 CIV is naturally of low abundance and flexible, limiting its structural resolution. The dual function of bacterial CIII 2 interacting with both the photochemical reaction center in photosynthesis, and cyt c oxidase in respiration, may necessitate this natural plasticity to allow swift metabolic adaptations. Similar flexibilities have also been seen with the yeast and human SCs (Sousa and Vonck, 2019).
Isolation of CIII 2 CIV was only possible using a genetically modified strain carrying a translational fusion between CIII 2 and CIV (SI, Methods). Despite the complexity of translocation, maturation and assembly processes of multi-cofactor containing membrane complexes, this fusion approach is of general use. Our fused SC preparations were compositionally heterogeneous, containing mixtures of CIII 2 CIV 2 , CIII 2 CIV and CIII 2 particles. The basis of this heterogeneity is unclear, though it may stem from subunit sub-stoichiometry, incomplete assembly, or higher susceptibility to degradation during sample preparations. Insertion of different spacers at the cyt c 1 -CcoP fusion junction, overexpression of the subunits and the related assembly components could not overcome the heterogeneity (SI, Methods). Consequently, structural studies required extensive data collections and limited structural resolutions, but allowed analyses of fragmented particles.
Structures of CIII 2 CIV. The structures of the tripartite CIII 2 CIV or bipartite CIII 2 CIV+c y at subnanometer resolution (~5.2 to 7.2Å) were highly similar. Limited protein-protein interaction between the subunits of CIII 2 and CIV was seen at the interface where the TMHs of cyt c y and CcoH were located (Fig. 4), limiting the flexibility of CIII 2 CIV. Another helix-like feature found at the exterior edge of CIV was attributed to the extra N-ter helix (TMH0) unique to R. capsulatus CcoN. However, due to the limited resolutions of the structures, these attributions are tentative. Limited resolution also precluded identification of non-protein constituents at the CIII 2 CIV interface. In this respect, R. Ornithine lipid-less mutants contain very low amounts of CIV and CIII 2 , and if any SC is unknown.
Previously, neither the exact location nor the mobility of cyt c y , which is the basis of the "soluble carrier-independent" electron transfer from CIII 2 to CIV, were known. The SC structure shows that locking the N-terminal TMH of cyt c y at the interface allows mobility of its cyt c domain (Fig. 8).
The linker region attaching the TMH to cyt c domain remains unresolved, but is long enough to allow oscillations between CIII 2 and CIV. Earlier studies with R. capsulatus cyt c y had shown that a fulllength linker is needed for rapid (< ~50 µsec) electron transfer from CIII 2 to the photosynthetic reaction center in photosynthesis (Myllykallio et al., 1998). In contrast, a shorter linker (~45-residue instead of 69) is fully proficient for respiratory electron transfer to CIV .

Structures of bacterial native CIII 2 . In native CIII 2 conformers, different positions of the [2Fe-
2S] cluster bearing FeS-EDs were seen. Crystallographic structures have often depicted bacterial CIII 2 as symmetrical homodimers (Berry et al., 2004;Esser et al., 2006;Xia et al., 2008). These structures 20 were obtained in the presence of inhibitors constraining FeS-EDs near heme b L or used mutants stabilizing it on cyt b surface. Alternatively, they contained crystal contacts restricting the FeS-ED movement . To our knowledge, no native heterodimeric CIII 2 structure of bacterial origin with different conformations of its FeS-EDs has been reported. Only recently, the cryo-EM structures of mitochondrial SCs with different maps for CIII 2 FeS-EDs have been reported (Letts et al., 2019;Sousa et al., 2016). Thus, native CIII 2 is not always a symmetric homodimer, and the FeS-ED of each monomer is free to move independently from each other, which has functional implications. The Q-cycle models describe the mechanism of CIII 2 catalysis by two turnovers of a given monomer (Crofts and Berry, 1998;Crofts et al., 2008;Osyczka et al., 2005). The mobility of the FeS-ED between the b and c positions is essential for QH 2 oxidation, and the different positions of the FeS-ED protein are often attributed to different catalytic steps (Esser et al., 2006). Emerging asymmetric structures of bacterial and mitochondrial native CIII 2 obtained by cryo-EM in the absence of inhibitors or mutations, combined with the well-established inter-monomer electron transfer between the heme b L of the monomers (Lanciano et al., 2013;Lanciano et al., 2011;Swierczek et al., 2010;Yu et al., 2002), start to provide a glimpse into plausible "heterodimeric Q cycle" mechanism(s) (Castellani et al., 2010;Cooley, 2010;Cooley et al., 2009), at least when CIII 2 is a part of SCs.
Accordingly, CIII 2 may cycle between homo-and hetero-dimeric conformations in regards to its FeS-EDs during catalysis. These mechanistic implications remain to be studied.

Electronic communication between CIII 2 CIV partners. Earlier, binding interactions between
CIII 2 CIV and its physiological electron carriers were unknown. Here we defined the likely interaction regions between the cyt c 2 or the cyt c y and CIII 2 CIV (Fig. 8). The CIII 2 CIV structure indicates that the distances separating heme c 1 of CIII 2 monomer A and hemes c p1 and c p2 of CIV are too large (Fig.   7) for direct microsecond scale electronic communication (Moser et al., 1992) to sustain the turnover rate of CIII 2 CIV. Thus, even when CIII 2 and CIV form a SC, a freely diffusing cyt c 2 or a membraneanchored mobile cyt c y , is required for QH 2 :O 2 oxidation.
The binding region of cyt c 2 on CIII 2 was identified earlier (Solmaz and Hunte, 2008), but that on CIV was unknown. The binding location of cyt c 2 on CIV determined in this study, the redox midpoint potentials (E m ) of the cofactors and the distances separating them (Fig. 7A) suggest that cyt c 2 would confer electrons to the closer heme c p2 , rather than the more distant heme c p1 , of CcoP. This will then initiate canonical electron transfer via heme c p1 , heme c o and heme b to heme b 3 -Cu B site for O 2 reduction (Brzezinski and Gennis, 2008;Wikstrom et al., 2018) (Fig. 8). For purified R. capsulatus proteins, the E m value of cyt c 2 is ~350 mV (Myllykallio et al., 1999), while those of CIII 2 heme c 1 and CIV heme c o are ~320 mV (Valkova-Valchanova et al., 1998) and ~210 mV (Gray et al., 1994), respectively. The E m values of R. capsulatus CIV hemes c p1 and c p2 are unknown, but based on similar E m values of heme c o for B. japonicum (200 mV) and R. capsulatus (210 mV), they are expected to be close to those of B. japonicum c p1 (~300 mV) and c p2 (~390 mV) (Verissimo et al., 2007).
In the case of cyt c y , its interaction region on CIV remains less well defined. Of the two binding regions of cyt c y on CIII 2 , that on cyt c 1 was taken as the most likely functional site. This binding region on cyt c 1 is close to that of cyt c 2 , but cyt c y has less complementary surface charges (Fig.   S10B), consistent with its weaker binding to CIII 2 CIV. Anchoring cyt c y to the membrane, next to its redox partners, might have enhanced its electron transfer efficiency while minimizing its electrostatic interactions with its partners.
The distance separating the redox centers is a major factor that controls the rate of electron transfer (Moser et al., 1992). The binding region of cyt c domain of c y on CIII 2 suggests that reduced cyt c y , upon its movement to CIV, might preferentially convey electrons to the closer heme c p1 than heme c p2 of CcoP (Fig. 8). If so, then under physiological conditions, heme c p1 would be the primary receiver of electrons derived from QH 2 oxidation by CIII 2 , forming a fully membrane-confined electronic 22 wiring within CIII 2 CIV. In contrast, cyt c 2 carries electrons from heme c 1 to heme c p2 via free diffusion.
Significantly, this membrane-external pathway might accommodate electrons not only from QH 2 but also from other donors distinct from CIII 2 . As such, reduction of cyt c 2 during methylamine oxidation (Otten et al., 2001), or degradation of sulfur containing amino acids, converting toxic sulfite (SO 3 2-) to sulfate (SO 4 -2 ) by sulfate oxidase (Kappler and Dahl, 2001) might provide electrons to CIV, contributing to cellular energy production.
In summary, for the first time, the architecture of CIII 2 CIV SC along with its dynamics and interactions with its physiological redox partners established salient structural features of two distinct respiratory electron transport pathways (membrane-confined and membrane-external) that operate between CIII 2 and CIV in Gram-negative bacteria.

Data deposition
The following R. capsulatus structures and the corresponding cryo-EM maps are deposited to PDB and EMDB with the accession codes listed in the table below: The raw XL-MS data deposited to PRIDE repository (http://www.ebi.ac.uk/pride/archive/) with the dataset identifier PXD020038 Table 1: Statistics of data collection and 3D reconstruction of CIII 2 CIV SC. Six and two individual datasets were combined for the tripartite (~c y , fused cyt c domain of cyt c y ) and the bipartite (+ c y , supplemented with native cyt c y ) SCs, respectively. For the combined datasets, the parameters for data collection were identical, except the exposure time. For all combined datasets, the range of exposure times and the corresponding dose rates are provided. Models were obtained by fitting the high-resolution model of CIII 2 (PDB: 6XI0) and the homology model of CIV (Table S7) into the maps. See these tables for refinement statistics.  (Table 1) were used for the 3D reconstruction of CIII 2 . The model (PDB: 6XI0) was refined in map CIII 2 (EMD-22189) and then used for rigid body fitting in maps CIII 2 c-c (EMD-22224), CIII 2 b-c (EMD-22225) and CIII 2 b-b (EMD-22226).

Data Collection
See Table 1, bipartite CIII 2 CIV + c y

3D Reconstruction
Map Name CIII 2 CIII 2 c-c CIII 2 b-c        XL-MS guided docking, and the subunits of CIII 2 CIV are colored as in Fig. 3, except that the monomer B of CIII 2 is shown in light grey for clarity. Only binding regions on monomer A are shown. A. Cyt c 2 (PDB: 1C2N) was docked onto CIII 2 and CIV using Patchdock with the DMTMM generated XLs as distance restraints, and yielded one cluster of models on CIV and one per monomer of CIII 2 . B. A model of cyt c domain of c y , generated using P. denitrificans cyt c 552 structure (PDB: 3M97) as a template (RMSD between template and model: 0.2 Å) was docked on CIII 2 as in A, except that both DMTMM and DSBU generated XLs provided distance restraints.

36
Two binding clusters for cyt c domain of c y per monomer of CIII 2 were found. These two clusters are located behind each other on a side view, but they are clearly visible on top views (Fig. S11C,   labelled 1 and 2). Here, only cluster 1 which is closer to cyt c 1 and overlapping with the binding region of cyt c 2 is shown. In all cases, the top 10 representative models are shown to depict the clusters of binding models. No binding region for cyt c y on CIV could be defined since no XL was found between these proteins.  Similarly, cyt c 2 which is peripheral to CIII 2 CIV also receives an electron from heme c 1 , diffuses away to reach CIV and conveys it to heme c p2 . The canonical electron transfers occurring from QH 2 to heme c 1 in CIII 2 , and from heme c p1 to O 2 in CIV, are indicated by thinner arrows. The double headed dashed black arrow depicts the movement of the [2Fe-2S] of FeS protein from the b position (b-pos, in black) to the c position (c-pos, in grey) in CIII 2 during QH 2 oxidation. Electron equilibration between the two heme b L of CIII 2 is indicated by double arrows, and the electron transfer steps subsequent to heme b H reduction are not shown for the sake of clarity.

Bipartite SC Tripartite SC
CIV-His CIV-c y -Flag