Structure and Mechanism of Respiratory III-IV Supercomplexes in Bioenergetic Membranes

In mammalian cells, cardiolipin is found primarily in the mitochondrial inner membrane where the weight fraction of the lipid is ∼18%193 (16% in the S. cerevisiae inner mitochondrial membrane197). In addition, the lipid may be enriched in the inner leaflet of the inner mitochondrial membrane,191 and it has been suggested to be involved in shaping the cristae.52 Cardiolipin has been identified as an integral part of many membrane proteins,198,199 and the enzymatic activities of, for example, detergent-solubilized mitochondrial cyt. bc1 and CytcO are dependent on the presence of bound cardiolipin200,201 (this effect is not observed with the R. sphaeroides CytcO202). In addition, cardiolipin is involved in apoptosis, where one step in the cascade of signaling reactions involves formation of a co-complex between the lipid and cyt. c, which results in cyt. c acquiring peroxidase activity.203

A discussion on the role of cardiolipin in supporting enzymatic activities of the respiratory complexes and its involvement in apoptosis is beyond the scope of this review. Instead, we discuss briefly cardiolipin’s role in maintaining supramolecular interactions between cyt. bc1 and CytcO in supercomplexes. The lipid is enriched in both the mammalian I1III2IV1204 and S. cerevisiae III2IV1/2205 supercomplexes. In the presence of cardiolipin the fraction of supercomplexes is larger than in its absence.71,204−208 Recent cryo-EM structures of the S. cerevisiae III2IV1/2 supercomplexes showed that a cardiolipin and presumably a phosphocholine are found at the cyt. bc1–CytcO interface. Two other cardiolipins are found in the vicinity where they also may contribute to supporting the cyt. bc1–CytcO interaction34 (Figure 8A). The lipid is suggested to mediate interactions between cyt. bc1 and CytcO acting as a “glue”209 by simultaneously binding to specific sites at each of these two complexes.199,210

Involvement of cardiolipin in stabilizing binding of cyt. bc1 to CytcO may, at least in part, explain why the fraction of supercomplexes and free complexes depends on S. cerevisiae growth conditions,17 which often influence the lipid composition of mitochondria. Furthermore, it is likely that the fraction of the two supercomplex forms, i.e., III2IV1 and III2IV2, is not only determined by the concentration of the cyt. bc1 and CytcO components in the membrane,33 but also by the presence of cardiolipin,189,205,206,208,209 which would modulate the cyt. bc1–CytcO binding affinity.

In the obligate III2IV2 supercomplexes in M. smegmatis and C. glutamicum three cardiolipins are found at the interface of complexes III and IV (Figure 8B).44,45 Similarly, to the S. cerevisiae supercomplex, the head groups of all these cardiolipin molecules face the n side of the membrane.

Respiratory supercomplex factors, Rcf1 and Rcf2, physically associate with cyt. bc1 and CytcO. Both Rcf1 and Rcf2 contain a hypoxia-induced gene domain 1 (HIGD1), which is conserved in a wide range of organisms.211−214 In Rcf1, the HIGD1 is in the N terminus and the C terminus has a fungi-specific domain, composed of approximately 60 amino acid residues. In Rcf2, which is a fungi-specific protein, the HIGD1 is located at the C terminus, preceded by a subdomain composed of ∼100 amino acid residues, which forms two transmembrane helices.215,216 The Rcf2 protein has been shown to be proteolytically processed to yield a stable C-terminal fragment that associates with CytcO.217

Data from early studies of the functional role of Rcf1 and Rcf2 were interpreted to indicate that these factors are required for formation of the cyt. bc1–CytcO supercomplexes in S. cerevisiae.73,188,214,217−220 The conclusion is in part based on observations that the ratio between supercomplexes and free components decreased upon genetic removal of Rcf1, which was also interpreted to suggest that this factor acts as a bridge between the components of the supercomplex. However, Rcf1 interacts with the Cox3 subunit and possibly also Cox13,214,219,221−223 but the recently determined supercomplex structures show that these subunits are found at the opposite side of CytcO from the III2–IV interaction surface (Figure 2A).33−35,37 Hence, Rcf1 cannot bridge supramolecular interactions between cyt. bc1 and CytcO. Similarly, a recently determined cryo-EM structure suggested binding of Rcf2 at the distal side of the supercomplex.33

More recent studies suggest that Rcf1 is instead involved in assembly of CytcO (reviewed in refs (54,224)) and incompletely assembled CytcO would result in a smaller fraction of supercomplexes. In other words, the cyt. bc1–CytcO supercomplexes can form also in the absence of Rcf1, but when Rcf1 is removed, a fraction of CytcO is modified, which yields less supercomplexes. Similarly, the Aim24 protein in S. cerevisiae(225) and mammalian homologue of Rcf1, HIGD2A, have recently been shown to be involved in the assembly of CytcO.226,227 It is interesting to note that data from recent studies indicate that removal of Rcf1 or Rcf2 affects the ability of the CytcO to maintain a proton electrochemical potential across the membrane, possibly due to proton leaks across the incorrectly assembled fraction of CytcO in the absence of Rcf.228

Genetic deletion of Rcf1 yields a subpopulation of CytcO that is incorrectly assembled and a subpopulation that is correctly assembled.219,229−231 In the absence of Rcf1, the correctly assembled CytcO subpopulation displays a lower activity and a modified heme a3-CuB catalytic site.229−231 The activity of this subpopulation could be restored upon addition of recombinantly expressed Rcf1,232 which suggests that in the correctly assembled CytcO reversible binding of Rcf1 can modulate the CytcO activity. This finding is further supported by recent data showing that Rcf1 positively modulates CytcO activity also in the intact mitochondrial membrane.221

Deletion of Rcf2 alone has a small effect on CytcO turnover,214,219,221,233,234 but more recent data indicate that binding of Rcf2 results in lowering the CytcO activity.221 Collectively, these data suggest that, in addition to being involved in assembly of CytcO, the binding of the Rcf proteins is linked to changes in the turnover activity.

Mass spectrometry revealed the presence of Rcf1 and Rcf2 in preparations of purified S. cerevisiae supercomplexes, but these proteins were not resolved in the first cryo-EM structures.34,35 As indicated above, more recent cryo-EM data show additional density in a pocket formed by Cox1, Cox3, Cox12, and Cox13 that in the supercomplex containing the Cox5B isoform could be assigned to the processed C terminus (HIGD1) of Rcf2.33 In CytcO containing the Cox5A isoform, the additional density could not be assigned with confidence. As the HIGD1 fragment is conserved to both Rcf1 and Rcf2, but is found in the C terminus of Rcf2 or the N terminus of Rcf1, the interaction between this segment and a putative conserved CytcO site would expose the remaining parts of the two Rcf proteins to different sides of HIGD1 (discussed in more detail in ref (215)). In other words, any additional interactions with the supercomplex would be very different for the Rcf1 and Rcf2 proteins. This observation reveals how binding of Rcf1 and Rcf2 could differently modulate the activity of CytcO or the supercomplex. An interaction between the homologous bovine HIGD1A protein and bovine CytcO was also observed.235 Furthermore, formation of the mammalian III2IV supercomplexes is dependent on another protein factor, COX7A2L.27,28

It is also interesting to note that interaction of Rcf1 with subunit Cox3 (subunit III) may modulate O2 binding at catalytic site221,228,234 because Cox3 harbors the lipid-containing V-shaped cleft suggested to be used for O2 diffusion from the membrane phase into the CytcO catalytic site. Data from earlier studies with the R. sphaeroides CytcO showed that changes in lipid molecules in this cleft result in changes of the CytcO catalytic site.236

As evident from the discussion above, the Rcf proteins determine the structure and function of complex IV of the S. cerevisiae respiratory chain, however, their role at the molecular level is complex and presently not fully understood.

A copper-containing superoxide dismutase (SodC) dimer subunit was found to be bound in the M. smegmatis III2IV2 supercomplex, near the cyt. cc head domain of the QcrC subunit43,44 (Figure 2B). As other SOD enzymes, it catalyzes the dismutation of the O2•– radical to H2O2 and O2:4a4b

The functional role of this SodC is unknown. Because the semiquinone formed as an intermediate at the QP site of complex III may react with O2 to form superoxide,81,91,92 association of a SodC with the respiratory supercomplex could allow detoxification near the O2•– generation site.44 In addition, the product H2O2 released by the SodC is a substrate for CytcO, which upon transfer of two electrons from cyt. c reduces H2O2 to water.237 Alternatively, the reduced Cu+ formed in SodC in the first reaction step (eq 4a) may transfer an electron to cyt. cc and then to CuA in CytcO, where it would enter the respiratory chain thereby bypassing formation of H2O2.44 In some anaerobic organisms, an essentially opposite reaction is catalyzed by a superoxide reductase, which reduces O2•– to H2O2 upon electron transfer from an external donor.238 Recently, an integral-membrane superoxide oxidase was discovered in E. coli.239 The M. smegmatis SodC has a similar orthologue in M. tuberculosis, where the subunit could remove O2•– generated by the host as a defense mechanism in the phagolysosomes of macrophages.44

In mitochondria cyt c is a small, typically ∼12 kDa, water-soluble protein that diffuses in the three-dimensional (3D) intermembrane space (Figure 1B). Cytochrome c has a dipole moment and a net positive charge.240,241 The edge of the heme group is positioned toward the positively charged protein surface, which docks either to cyt. c1 or near CuA at negatively charged surfaces of cyt. bc1 or CytcO, respectively.242−244 The orientation of cyt. c is the same when binding to either cyt. bc1 or CytcO.245,246

It is generally assumed that the intracellular ionic strength is relatively high (80–150 mM), and it has been shown that at this ionic strength a major fraction of cyt c diffuses in three dimensions.16,51 However, a recent analysis revealed that only the cation concentration is kept at high concentration, while the concentration of small anions is much lower and the remaining negative charges are found at the surfaces of polyanionic macromolecules.247 As a consequence, the Debye screening radius in the intracellular medium is larger than that obtained for a monovalent salt electrolyte at 80–150 mM. Oliveberg, Wennerström, and coauthors estimated that a more reasonable mimic of the intracellular environment is the equivalent of ∼20 mM of a 1:1-electrolyte. As a consequence, the electrostatic interactions between the positively charged cyt. c, and its negatively charged interaction partners are likely to be much stronger than those observed when mimicking the intracellular environment in a solution containing 80–150 mM monovalent salt. Below, we discuss the consequence of supercomplex-cyt. c interactions for electron transfer between complexes III and IV in supercomplexes, but first we briefly describe data from studies of interactions of cyt. c with complexes III2 and IV, respectively.

Early data from steady-state turnover measurements with the mammalian cyt. bc1 suggested that cyt. c binds at a single site near cyt. c1.246 More recent data from NMR studies of the plant complex III identified an additional low-affinity distal binding site.248 In the crystal structure of the S. cerevisiae cyt. bc1–cyt. c co-complex, cyt. c was found bound to cyt. c1.243,249 In the structure of the S. cerevisiae III2IV1/2 supercomplex–cyt. c co-complex (see inset to Figure 4A), the position of cyt. c at cyt. bc1 was only slightly shifted compared to that observed in the crystal structure.37

Interactions of cyt. c with CytcO are more complex. Results from studies of the steady-state turnover rate of mammalian CytcO were interpreted to indicate two cyt. c binding sites in CytcO.245,250,251 This observation does not automatically imply the presence of two independent binding sites from which an electron is transferred to CuA. The same data could also be explained in terms of “nonproductive” binding of cyt. c that interferes with the “productive” binding site.252 However, results from other experiments suggested that two cyt. c molecules can simultaneously bind to a monomer of the mammalian CytcO, with KD values of ∼10 nM and ∼1 μM, respectively.245,250,251 Furthermore, covalent cross-linking of a cyt. c at the high-affinity site only had a minor effect on binding of a second cyt. c at the low-affinity site.253 Binding at each site presumably results in electron transfer from cyt. c to CuA, but electron transfer from cyt. c at the high-affinity site is slower than that from the low-affinity site.253

Studies of the steady-state activity of the S. cerevisiae CytcO were initially interpreted to suggest binding of two cyt. c molecules with equal affinities, KM ≅ 100 nM.254 However, more recent data revealed an additional KM of ∼30 μM,177 indicating a similar mechanism of cyt. c binding to the mammalian and S. cerevisiae CytcO.

The cryo-EM structure of the III2IV1/2 supercomplex-cyt. c co-complex in S. cerevisiae(37) showed that the cyt. c binding is similar to that seen in the crystal structure of the equivalent co-complex with the bovine CytcO244 (see inset to Figure 5A).

It is clear that association of cyt. bc1 and CytcO to form a supercomplex leads to a decrease in the intercomplex distance. The distance between the electron donor site at cyt. bc1 and the acceptor site near CuA at CytcO within the S. cerevisiae supercomplex is ∼60 Å (Figure 3A)34,35 (see also refs (72,74)), i.e., too long to yield a catalytically relevant electron-transfer rate through docking of a single cyt. c between the electron donor and acceptor sites.255 Thus, the question arises whether or not a shorter diffusion distance via the water phase of the intermembrane space (defined as 3D diffusion) would result in a higher QH2:O2 oxidoreductase activity.36,50 Considering a reasonable average distance between independently diffusing cyt. bc1 and CytcO in the membrane (∼50 nm, see Figure 1B), the 3D diffusion time of cyt. c between these complexes is in the order of 10 μs.50 Hence, diffusion of cyt. c cannot be rate limiting for electron transfer from QH2 to O2 because the maximum turnover (kcat) of cyt. bc1 and CytcO in S. cerevisiae is ∼102 s–1 and ∼103 s–1, respectively.17 Furthermore, the overall electron flux through the respiratory chain in vivo is lower than the lowest kcat value of the involved components, in the range 40 s–1 to 140 s–1 (Michel Rigoulet, personal communication). Nevertheless, the QH2:O2 oxidoreductase activity is dependent on the concentration of externally added cyt. c to mitoplasts36 or purified supercomplexes at a cyt. c:supercomplex ratio similar to that found in vivo,37 suggesting that the cyt. c-mediated electron transfer is rate limiting.

Results from a recent theoretical study showed that the electron flux between cyt. bc1 and CytcO, mediated by 3D diffusion of cyt. c, is determined by the equilibration time of cyt. c with the cyt. c pool in the intermembrane space, rather than by the cyt. c diffusion time constant itself.50 Furthermore, the data showed that this equilibration time increases with decreasing cyt. c concentration, i.e., the lower the cyt. c concentration, the stronger the distance dependence on activity. For freely diffusing components, a cyt. c:supercomplex ratio of 2–3 and an average cyt. bc1–CytcO distance of 50 nm (Figure 1B), this scenario yields a cyt. c-mediated QH2:O2 oxidoreductase activity that is slower than the turnover of cyt. bc1 and is dependent on the average cyt. bc1–CytcO distance. Interestingly, on the basis of the data in ref (256), Maldonado et al. estimated that in plant mitochondria the cyt. c:supercomplex ratio is one,32 suggesting an even stronger cyt. bc1–CytcO distance dependence on the QH2:O2 oxidoreductase activity than in S. cerevisiae mitochondria. Taking into consideration the recent finding that the salt concentration equivalent of the intracellular environment is estimated to be ∼20 mM247 rather than the 150 mM used in the theoretical study,50 the diffusion coefficient for cyt. c in mitochondria would be a factor of ∼102 lower51 than that used in the theoretical study in ref (50). This effect further emphasizes the kinetic advantage in forming supercomplexes, under the assumption that electron transfer occurs via 3D diffusion.

Many Gram-negative bacteria, e.g., R. capsulatus, R. sphaeroides, and P. denitrificans harbor a membrane-anchored cyt. cy in addition to a water-soluble cyt. c.46,257,258 A cyt. cy homologue is the only cyt. c present in Rickettsia prowazekii.257,259 Restriction of cyt. c diffusion to the two-dimensional (2D) space of the membrane surface yields shorter diffusion times than for 3D diffusion at the same concentrations of the involved components.50 Furthermore, integration of a membrane-anchored cyt. c into a cyt. bc1–CytcO supercomplex allows direct electron transfer from the donor at cyt. bc1 to the acceptor at CytcO,59,260 even though the linker between the membrane domain and the cytochrome domain in cyt. cy is too long to distinguish between 2D and restricted 3D diffusion. In a recent study, the normally water-soluble cyt. c was attached to a membrane-bound protein in S. cerevisiae mitochondria, which allowed electron transfer between complexes III and IV over a time scale similar to that in vivo.261

Some Gram-positive bacteria, which lack an outer membrane, harbor membrane-associated cyt cs that are attached either via a transmembrane polypeptide or a lipid anchor.262 In Bacillus PS3, a supercomplex composed of cyt. bc1, CytcO and a cyt. c was identified and shown to display quinol oxidase activity, i.e., electron transfer from quinol to oxygen.68 In the Gram-positive actinobacteria from, e.g., M. smegmatis and C. glutamicum electron transfer between cyt. bcc and CytcO occurs via the diheme cyt. c ectodomain of the QcrC subunit of the cyt. bcc complex (Figures 2B and 3B). Because these bacteria lack any water-soluble or membrane-anchored free cyt. c, a supercomplex composed of cyt. bcc and CytcO is required for electron transfer from MQH2 to dioxygen.62,64,65,263 Disruption of the supercomplex using detergent results in a decrease in activity.263

Electron transfer between cyt. bc1 and CytcO by 2D diffusion of cyt. c that is bound to the supercomplex surface or weakly associated with the membrane has been discussed also in organisms that harbor a water-soluble cyt. c(37,50,77,264−268) (see also ref (53)). The surface between the cyt. c-binding sites at cyt. bc1 and CytcO in the S. cerevisiae supercomplex is negatively charged (Figure 9A), and one cyt. c per CytcO is tightly bound to the supercomplex204,234,269in situ (but not in purified complexes). Assuming the same scenario in plant mitochondria, an estimated cyt. c:supercomplex ratio of one in V. radiata(32) suggests that the entire cyt. c pool would be associated with supercomplexes but presumably at equilibrium. Recent Cryo-EM structures of the supercomplex with added cyt. c revealed distinct states where cyt. c is bound either to cyt. bc1 or CytcO, or resides at intermediate positions at the supercomplex surface.37 Measurement of the supercomplex activity as a function of the concentration of added cyt. c yielded apparent KM values of ≤6 nM and ∼1.7 μM, i.e., much smaller than those obtained with isolated S. cerevisiae CytcO (∼100 nM and ∼30 μM, respectively, see above). These data suggest a stronger binding to the supercomplex than to CytcO, which is consistent with the large negatively charged binding surface for cyt. c between cyt. bc1 and CytcO. The QH2:O2 oxidoreductase activity of the supercomplex is ∼20 e–/s for a supercomplex with a single bound cyt. c. This rate decreased upon dissociation of the supercomplex, i.e., when increasing the average distance between cyt. bc1 and CytcO. Collectively, the structural and kinetic data showed that electron transfer within the supercomplex is mediated by 2D diffusion of a single surface-associated cyt. c. It is also interesting to note that the rate of electron transfer between cyt.

bc1 and CytcO with a single bound cyt. c is near the lower limit of the electron flux through the respiratory chain in vivo. It is also worth mentioning that the above-described experiments were performed at the assumed near-physiological monovalent salt concentration of ∼150 mM, which was also required to prevent protein aggregation on the cryo-EM grids.37 Considering the novel finding that a better mimic of physiological conditions is 20 mM monovalent salt,247 the cyt. c–supercomplex interactions are most likely even stronger in vivo than those experimentally observed.37

In conclusion, the combined cryo-EM and kinetic data show that supercomplex formation in S. cerevisiae does not result in increasing the electron transfer rate by decreasing the cyt. c 3D diffusion distance, as recently suggested.36 Rather, formation of III2IV1/2 supercomplexes in S. cerevisiae results in switching to a different mechanism that involves 2D diffusion from the electron donor to the electron acceptor.37 In other systems electron transfer between complexes III and IV may occur by 3D diffusion and the theoretical studies show that also under these conditions, there is a kinetic advantage in decreasing the intercomplex distance by formation of supercomplexes.50 The 2D-diffusion mechanism in S. cerevisiae is similar to that suggested for electron transfer from cyt. bc1 to the cbb3 CytcO via a movable membrane-anchored cyt. cy domain in R. capsulatus.46

Electron transfer from cyt. bc1 to CytcO by 2D diffusion of cyt. c along the supercomplex surface resembles a “substrate channeling” model, which has been criticized based on the finding that cyt. c diffusion in S. cerevisiae is unrestricted.269 However, 2D diffusion of cyt. c is not in conflict with this finding because it assumes only weak electrostatic interactions between cyt. c and the supercomplex surface, and cyt. c remains in equilibrium with the cyt. c pool during the electron-transfer process16 (see Figure 10).

In mammalian mitochondria, complexes III2 and IV are not only part of respirasomes but also assemble independently to form III2IV supercomplexes.27,28,272 The structure of these supercomplexes is presently not known. Figure 9B shows the electrostatic potential surface of the cyt. bc1–CytcO part of the mammalian respirasome. As seen in the figure, the negatively charged cyt. c binding sites at cyt. bc1 and CytcO are less connected by negative charges on the surface in between the sites than in the S. cerevisiae supercomplex (Figure 9A). This difference in charge distribution may reflect the much lower fraction of CytcO that is part of supercomplexes in mammalian (15–30%54) than in S. cerevisiae yeast (∼90%,72) mitochondria. In other words, in the mammalian respiratory chain electron transfer between cyt. bc1 and CytcO occurs primarily via 3D diffusion.

It is also interesting to note that in the M. smegmatis III2IV2 supercomplex, interactions between the movable cyt. cc domain44 (see Figure 2B) and complex IV most likely occur by electrostatic interactions between positive charges on the cyt. cc surface and negative charges at complex IV (Figure 9C). However, the extracellular surface of complex III is positively charged, which indicates that the cyt. cc domain is held in place by its TM α-helix rather than by electrostatic interactions.