Plant supercomplex I + III₂ structure and function: implications for the growing field

The mitochondrial electron transport chain (mETC) couples the transfer of electrons from NADH and succinate to molecular oxygen to the pumping of protons across the inner mitochondrial membrane, from the mitochondrial matrix into the inter-membrane space (IMS). This generates an electrochemical proton gradient that is then used by ATP synthase to generate ATP for the cell. The mETC is composed of four multi-subunit membrane complexes (complex I–IV, CI–IV) as well as electron carriers quinone (in the membrane) and cytochrome c (in the IMS) [1]. While this canonical respiratory chain is present across free-living eukaryotes, plants (and other organisms) also contain an alternative respiratory chain that transfers electrons from NADH to oxygen via quinone without proton pumping [2]. Thus, although the alternative chain does not directly contribute to the proton gradient, it does affect mitochondrial NADH and quinone levels.

The canonical mETC complexes can exist in isolation or assemble into higher-order supercomplexes. Biochemical and in situ electron tomography studies suggest that supercomplexes may be the most abundant and physiologically relevant form of the mETC [3–5], although the advantages of assembling supercomplexes remains poorly understood. In plants, the most abundant supercomplex is the assembly between CI and CIII₂, i.e. supercomplex I + III₂ (SC I + III₂), containing 50–90% of CI [4,6]. Plant CI is a 48-subunit complex with 14 core subunits with catalytic roles and 34 accessory subunits with incompletely understood assembly, stability, or regulatory roles [3,7–9] (Figure 1a,b). The subunits are arranged into multiple modules that form a membrane arm (embedded in the inner mitochondrial membrane) and a peripheral arm that protrudes into the matrix (Figure 1c). The peripheral arm contains redox active sites in the N module (where NADH binds) and a Q module (where quinone binds). The membrane arm contains the proton pumps arranged in a proximal (P_P) and distal (P_D) module, as well as a γ-carbonic anhydrase (γCA) domain attached on the end of the P_p. Additionally, a more recently discovered bridge domain links plant CI's arms [9] (Figure 1c,d). The γCA and bridge domain are inferred to be ancestral domains present in the last eukaryotic common ancestor and lost in CI of mammals and fungi [10,11]. Plant CIII₂ is a 20-subunit obligate dimer (10 subunits per protomer) composed of a membrane domain (where electron transfer occurs) and a matrix domain composed of subunits of the mitochondrial processing peptidase (MPP) family [12] (Figure 1e). In plants, CIII₂’s MPP domain is active and provides the MPP activity that cleaves the import signal from most proteins that are imported into the mitochondria [13–17]. Thus, in plants, CIII₂ is a dual respiratory and peptidase enzyme.

The mechanistic study of mETC complexes and supercomplexes is an active area of research across organisms. In this article, I review recent structural and functional investigations of plant SC I + III₂ from Arabidopsis thaliana [7] and Vigna radiata [8] (mung bean). I summarize key results from each paper and then discuss commonalities, differences, open questions and implications for CI and supercomplex mechanisms in plants and in non-plant organisms. I focus on CI's γCA anhydrase and bridge domains, catalytic loops and active-to-deactive transition, as well as on CIII₂’s MPP isoforms, and then discuss studies on the roles of supercomplexes.

The structural and functional studies of SC I + III₂ from A. thaliana and V. radiata were published simultaneously [7,8]; data had not been exchanged between the groups before publication, to maximize the independence of the studies. The cryo-electron microscopy (cryoEM) structures of these plant SC I + III₂ are generally in excellent agreement, particularly for CI and the supercomplex interfaces (Figure 1). In both structures, the binding of CIII₂ to CI's membrane arm led to the stabilization of and improved cryoEM densities for this region. This allowed for the identification of NDUP9 (a newly identified plant-specific subunit of CI), the modeling of previously missing segments of core subunits (e.g. ND5, ND6), as well as the retention and modeling of subunit NDUA11, which was missing in all previous plant CI structures. Thus, both structures provided the full complement of CI subunits (48) as well as the specific protein:protein interactions between CI and CIII₂. Three interfaces were observed: one in the mitochondrial matrix between NDUB9 from CI and MPP-α and MPP-β from CIII₂, one in the IMS between NDUP9 and NDUA11 from CI and QCR6 from CIII₂, and one in the transmembrane region between NDUA11 from CI and QCR8 from CIII₂. The bridge domain does not form part of any supercomplex interface. Both groups also observed a decreased flexibility between the matrix and membrane/IMS regions of CIII₂, in contrast to that previously seen in free V. radiata CIII₂ [12].

In their study, Klusch and team investigated the structure of SC I + III₂ from an A. thaliana cell suspension culture and obtained a 2.1 Å resolution structure [7]. This high resolution allowed for the modelling of ∼3000 well-ordered water molecules within the region of CI's central axis, E-channel and ubiquitin-binding channel (PDB: 8BPX). Well-ordered water molecules are relevant for CI because they provide the medium for proton conductance via hydrogen-bonded networks (Grotthus mechanism) [18]. As such, they play an essential role in CI's redox-coupled proton pumping mechanism from the matrix into the IMS, which remains incompletely understood [19–21]. The structure identified three potential proton entrances from the matrix into CI's central aqueous passage (the central channel) at core subunits ND2, ND4 and ND5 in the membrane arm. It also identified two potential proton exits into the IMS at ND2 and ND5. Within CI's γCA domain, the high-resolution structure revealed a water molecule in tetrahedral coordination by the Zn⁺² atom between γCA1 and γCA2, confirming the plant γCA domain as a member of the CamH subclass and further suggesting the domain is catalytically active (yet to be confirmed experimentally). The structure also allowed for the assignment of a density between subunits γCA2 and γCAL-2 as butyryl- or crotonyl-CoA (CoA-X, the only undiscernible difference being a single or a double bond) (Figure 1d). This potential new active site has implications for a possible carbon-assimilation role for plant CI.

In our study, we obtained an overall 3.3 Å structure of SC I + III₂ from etiolated V. radiata hypocotyls [8]. Whereas the water molecules seen in A. thaliana were not observable in the V. radiata structure due to its lower overall resolution, density consistent with CoA-X is present in the V. radiata reconstructions but was not modeled due to insufficient resolution [8,22]. In the V. radiata SC I + III₂ structure, we observed differences in the CIII₂ isoforms found within SC I + III₂ compared to free CIII₂ and to CIII₂ in a supercomplex with complex IV (SC III₂+ IV, [12]). For CI, we examined the conformations of several catalytically relevant loops in the vicinity of the quinone-binding channel (ND1 TMH5–6, NDUFS2 β1–2, NDUFS7 α1–2 and α2–β1) and at the interface between CI's arms (Nad3 TMH2–3, Nad6 TMH3–4). Broadly speaking, these loops have been seen to be ordered in closed and active turnover states of CI and disordered in open and deactive states of CI from mammals, yeast and bacteria. Whether the different conformations of these loops represent active and deactive states of CI or different points in the catalytic cycle remains actively debated [19,20]. Most of the loops in V. radiata were disordered or in a mixed or intermediate state between those previously seen in other organisms, without neatly aligning with any one single state. The presence of a π-bulge in ND6's transmembrane helix 6 (TMH6) suggested our structure was in a resting state, consistent with its purification in the absence of substrate. Therefore, we concluded that resting conformation of the quinone-binding site of plant CI adopts a more ordered conformation than mammalian deactive CI. We also investigated the existence of a potential ‘active-to-deactive’ (A/D) transition in plant CI. This mechanism, observed in mammals and other vertebrates [23–25], is a transition that places CI in an off-pathway conformation in the absence of substrate. Similar to the intermediate state observed with the key loops, our functional results showed that V. radiata CI displays some but not all features of the A/D transition (susceptibility to NEM, but no change in susceptibility upon thermal deactivation or substrate pre-activation).

Carbonic anhydrases catalyze the conversion of CO₂ and H₂O into HCO₃⁻ and a proton (H⁺) via a catalytic Zn²⁺ ion. Based on the geometry of the putative active site between γCA1 and γCA2 (three histidine residues and an ordered water molecule coordinating a Zn²⁺ in tetrahedral geometry, as seen in the high-resolution SC I + III₂ structure of A. thaliana), it has been assumed that plant CI's γCA domain is catalytically active [7,26]. Moreover, it is hypothesized that the γCA hydration of CO₂ into HCO₃⁻ contributes to the refixation of carbon in the Calvin-Benson–Bassham cycle of photosynthesis in plants [27–30]. This is in line with the Zn²⁺ catalytic site being absent in Polytomella (non-photosynthetic green alga) [9], Tetrahymena thermophila (non-photosynthetic alveolate) [10] and Euglena gracilis (mixotrophic euglenid) [11] (Table 1). The finding of the CoA-X ligand in the γCA domain and the hydrated half-channel on the matrix side of ND2 (the core subunit to which the γCA domain is anchored, Figure 1c) in the recent A. thaliana SC I + III₂ structure [7] add layers to the hypothesized bioenergetic roles of the γCA domain (Figure 1d).

First, given crotonyl-CoA's interactions with CO₂ in the carboxylation reaction of crotonyl-CoA carboxylase/reductase [32], it is now hypothesized that this co-factor helps position the CO₂ molecule to enable γCA's reaction in plants [7,26]. This is in line with density consistent with CoA-X being present in the A. thaliana, V. radiata and cauliflower CI structures [7–9, 22, 33] — although not previously assigned due to insufficient resolution — and absent in the CI γCA domain of T. thermophila, a non-photosynthetic alveolate [10]. A density consistent with CoA-X is also present in Polytomella sp., perhaps as a vestigial remnant [9,26]. No relevant density is present in mixotrophic organism E. gracilis, implying that the crotonyl-CoA connection is either only present in the green lineage, or only in purely photosynthetic organisms (photoautotrophs) rather than in mixotrophs (Table 1). Interestingly, E. gracilis shows a phosphoethanolamine phospholipid at the interface equivalent to CAL-2 and γCA1, suggesting that the interfaces between γCA subunits may be prone to binding hydrophobic molecules. Nevertheless, E. gracilis CI does have a different connection to crotonyl-CoA: it contains a new fatty acid synthesis (FAS) domain with homologs to NADH-dependent trans-2-enoyl-coenzyme A/acyl-carrier-protein reductases, which can reduce crotonyl-CoA into butyryl-CoA during mitochondrial anaerobic wax fermentation [34,35]. Although the enzymatic activity of this domain has not been shown yet, it does suggest a wider (more ancient) potentially direct relationship between CI and crotonyl-CoA, as well as between oxidative phosphorylation and FAS.

The molecular details of the putative new plant CoA-X/CO₂ active site, how it compares to crotonyl-CoA carboxylase/reductases [32], and how the CO₂ in this site would be transferred to the carbonic anhydrase Zn⁺²-binding site have not been determined. We propose another non-mutually exclusive role for the CoA-X co-factor. CoA-X may be required for the assembly and integrity of the γCA domain, an early essential step in the assembly of the entire CI [36–41]. Linking CI assembly to the availability of crotonyl/butyryl-CoA may be a way to regulate the abundance of CI — and electron transfer through the canonical mETC — by the status of amino acid and fatty acid metabolism in different tissues or developmental stages [42]. For instance, this CoA-X ‘sensor’ would be particularly important during germination, where both high levels of CI assembly [38] and high levels of crotonyl/butyryl-coA from amino acid and fatty acid catabolism [42] are expected. This is in line with the increasingly appreciated roles of FAS in mitochondrial biogenesis and oxidative metabolism [43,44]. A re-examination and further testing of existing plant γCA deletion and point mutants [30,39,45–50] in light of the now known γCA domain composition, and the creation of further point mutants lacking the CoA-X binding sites will shed light on these hypotheses.

Second, the γCA functional hypothesis now also includes the H⁺ that is produced by the CO₂ hydration reaction [26]. It is hypothesized that these protons generate a local gradient that facilitates H⁺ entry into the matrix half-channel on ND2 (Figure 1c,d); proton entry would also be facilitated by electrostatic repulsion from the nearby Fe cation in the ferredoxin subunit of CI's bridge domain (NDUFX, see below) (Figure 1c,d). The ND2 half-channel identified in the A. thaliana SC I + III₂ was not seen in other recent high-resolution structures of CI from heterotrophic organisms [51–57]. However, this may be due to the ‘snapshot’ nature of cryoEM data collection, particularly from CI samples without substrate, as excellently discussed elsewhere [21]. Moreover, whereas NDUFX likely contains a Fe ion in vascular plants [9,58], it contains a 2Fe2S co-factor in T. thermophila [10] (non-photosynthetic) and no co-factor in Polytomella [9] (non-photosynthetic). In E. gracilis (mixotrophic), NDUFX has a truncated ferredoxin fold that is lacking the co-factor-coordinating elements altogether [11] (Table 1). Therefore, the question remains whether the ND2 half-channel is an underlying universal feature of CI's H⁺ pumping mechanism or whether it is an adaptation of organisms with (presumed) γCA activity and Fe-containing co-factors in its vicinity (so far limited to vascular plants, given the Polytomella and E. gracilis findings).

In cyanobacteria, carbonic anhydrase activity is used as a carbon-concentrating mechanism to aid photosynthesis by certain versions of photosynthetic complexes evolutionarily related to CI (photosynthetic CI, PS-CI, a.k.a. NDH-1) [59,60]. PS-CI contains homologs of CI subunits in the membrane arm and Q module but lack the N module. Thus, these complexes use ferredoxin (rather than NADH) as the electron donor to quinone and couple this redox reaction to proton pumping across the membrane [61–64]. What can these PS-CI teach us about carbonic anhydrase and proton pumping activity in respiratory CI? In the carbon-concentrating PS-CIs studied so far [31], the reaction is carried out by a Zn²⁺-binding protein (CupA) that is on the ‘end’ of PS-CI's membrane arm (rather than anchored to the ND2 homolog) and that is not related to canonical, multimeric α, β or γ carbonic anhydrases. Therefore, the potential carbonic anhydrase activity of respiratory CI might be an example of convergent evolution. In terms of proton pumping, although it is commonly assumed that PS-CI pumps protons through half-channels in membrane arm, including through the ND2 homolog, this has only been inferred from nearby arrangement of protonatable residues [65–67] or molecular dynamics simulations [31] rather than from the observation of Grotthus-competent ordered water molecules in high-resolution structures of PS-CI. Thus, further structural and functional work on PS-CI is needed. If an ND2-like proton pathway were present in PS-CI, it would predict its universal presence in respiratory CI as well. Moreover, if the ND2 half-channel were universal in all CIs, this would open additional mechanistic and evolutionary questions regarding the proposed γCA-induced local proton gradient and repulsion by the bridge ferredoxin co-factor, as CI's γCA and bridge domains are believed to have been present in the last eukaryotic common ancestor [10,26,68].

Determining the γCA activity, proton pathways and pumping mechanism(s) of CI and evolutionarily related complexes such as PS-CI in a variety of photosynthetic, mixotrophic and heterotrophic organisms is critical to understand how the enzyme works and whether it indeed has a universal mechanism.

The existence of a bridge domain that links CI's membrane and peripheral arms was initially discovered in A. thaliana and Polytomella sp. [9]. In plants, the bridge domain is formed by subunit NDUA6 (B14) connected to the peripheral arm, acyl-carrier-protein subunit NDUAB1α (SDAP2) and NDUFX, a ferredoxin-like subunit that binds ND2 (membrane arm) and γCAL-2 (γCA domain). A larger bridge was later seen in T. thermophila [10] and E. gracilis [11]. In the initial A. thaliana CI structural analysis [9], CI was found in different structural classes corresponding to ‘closed’ CI with a full bridge (with an inter-arm angle of 106°) or ‘open’ CI with a partial bridge (inter-arm angle of 112°). Given that at the time the inter-arm angle of CI was thought to hold biological significance, it was hypothesized that the bridge may have a regulatory role on CI's mechanism. The structures of plant SC I + III₂, together with findings in other organisms that the inter-arm angle of CI is not critical for function [10,56,57], have updated this view. Now, both groups have concluded that the bridge domain of CI has structural rather than regulatory roles. The bridge domain restricts the angles between the CI arms, as shown by the fact that the SC I + III₂ cryoEM particles are classified into a single class or very few highly similar classes with a narrow range of angles between the CI arms. Similar findings were seen for the bridge domains of T. thermophila [10] and E. gracilis [11]. The ‘open’ CI particles in the initial A. thaliana CI structure [9] had a partial bridge and were also lacking accessory subunit NDUA11. Similar ‘bridge-less’ CI was observed in V. radiata’s SC I + III₂. Here, the lack of the bridge domain was always accompanied with the lack of subunits (e.g. NDUA11) and/or the disordering of catalytically relevant regions (e.g. TMHs in ND5 and ND6). This is highly reminiscent of that seen in partially degraded samples of mammalian CI [54], suggesting that the bridge-less particles in plants correspond to non-functional CI that likely arises from detrimental interactions between the sample and the cryoEM grid.

Additional evidence for the roles of the bridge comes from functional studies of NDUFX knockout mutants in A. thaliana [58]. These bridge-less mutants contain intact CI but are not able to form SC I + III₂. Therefore, the bridge-less supercomplex particles observed in the cryoEM data are more likely degradation products than assembly intermediates. Moreover, the bridge-mutant studies suggest that the bridge plays a key role in the correct assembly of CI's membrane arm and, consequently, its interactions with CIII₂. This has led to the hypotheses that the roles of the bridge domain are to (1) correctly position the γCA domain to enable efficient assembly of CI's P_d domain onto CI* (the last plant CI assembly intermediate, which contains the N, Q and P_P modules and lacks the P_d domain [38]) and/or (2) properly place NDUA11 to allow for the necessary CI:CIII₂ interactions in SC I + III₂ [58]. It is important to note that the bridge-less CI in the NDUFX mutants is not equivalent to bridge-less CI in the SC I + III₂ studies: the first cannot bind to CIII₂ but contains NDUA11, whereas the second most likely represents a previously fully bridged CI in SC I + III₂ that subsequently lost its bridge and NDUA11. Thus, the bridge-less SC I + III₂ structures cannot be used to test the above hypotheses. Testing these hypotheses will require the structural examination of CI from bridge-less mutants [58].

The conformational state of several key CI loops around the quinone-binding site has garnered attention as a way to define the active/deactive state of CI and draw inferences on its catalytic cycle [19]. Close inspection of the key CI loops shows excellent agreement between the A. thaliana (PDB: 8BPX, 8BQ5, 8BQ6) and the V. radiata SC I + III₂ models (PDB: 8E73) for all loops except for the NDUS2 (49 kDa) β1–2 loop, which contains a key histidine residue that binds the quinone headgroup in the quinone-binding pocket [7,8]. Although the loop was fully disordered in V. radiata (similar to murine deactive CI (PDB: 6G72) [69]), it was ordered in A. thaliana. Here, the loop resembled (but did not fully match) the ‘retracted’ conformation seen in ovine closed CI (PDB: 6ZKO) [52] and yeast CI in turnover (active) conditions (PDB: 7O6Y) [53] (Figure 2). The differences between V. radiata and A. thaliana may be due to small sequence differences surrounding the loop (the loop itself is perfectly conserved across V. radiata, A. thaliana, Mus musculus and Yarrowia lipolytica) or to the different purification protocols used. Either way, the comparison further supports the finding that the quinone-binding site of the resting state of plant CI can adopt more ordered conformations than those typically seen in deactive mammalian CI (the ordered S2 loop was also seen in yeast deactive CI, PDB: 7O71, [53]). Moreover, the ordered NDUS2 loop in A. thaliana further shows that changes in CI's inter-arm angle are not required to organize the loops.

These plant findings, as well as more recent observations of ordered loops in E. gracilis CI in the absence of substrate [11], emphasize that different species and clades can have different conformations for their resting-state CI. Therefore, caution should be taken when extrapolating aspects of CI based on features seen in mammals. The observations from plants and other non-opisthokont organisms need be considered in proposals of the universal mechanism of CI. Alternatively, it is possible that the details of CI's mechanism are different in different eukaryotic supergroups. The structural characterization of a broader set of organisms will benefit our understanding of CI function.

Similar to the finding that V. radiata’s and A. thaliana’s catalytic loops were in conformations that did not neatly align with findings from mammals or yeast, the V. radiata A/D assays also showed a ‘non-canonical’ behavior for plant CI. In this assay, CI is exposed to NEM a thiol reactant that modifies a cysteine residue in ND3's TMH1–2 loop, which is part of the quinone-binding site. In mammals, the cysteine is exposed in the deactive state and buried in the active/turnover state. Modification by NEM prevents the loop from going into its active conformation, thereby blocking CI in a deactive state and reducing the overall NADH-quinone oxidoreductase rate [70,71]. In organisms with a classic A/D transition, NEM effects are increased when CI is exposed to high temperature (37°C, thermal deactivation) and minimized when CI is pre-exposed to a low amount of NADH to put the enzyme in turnover conditions (pre-activation) [23]. Whereas V. radiata CI does show sensitivity to NEM, the sensitivity does not deepen with thermal deactivation and it does not lessen upon pre-activation [8]. This suggests that CI's active and deactive states are not biochemically distinct in plants as in mammals, but rather closer to each other. Likely, the active state conformation — in particular, the ordering of the ND3 TMH2–1 loop — is more accessible in the absence of substrate in plants than in other organisms (and/or the deactive state may also be more accessible in the presence of substrate). That is, the A/D functional assay also points to plant CI being capable of partial loop ordering in the resting state, in line with that seen in the V. radiata and A. thaliana structures.

Questions remain as to whether the presence of the bridge is what causes this A/D behavior, or whether it is mere correlation. The other organisms with a bridge in which the A/D transition was tested are T. thermophila and E. gracilis. In T. thermophila, the A/D transition is absent, with no effect of NEM on NADH oxidation rates [10]. In E. gracilis, very similar results to V. radiata were obtained: CI is sensitive to NEM, but this sensitivity is unaffected by NADH pre-activation (the effect of thermal deactivation was not examined) [11]. The A/D transition has also been examined in Drosophila melanogaster [72]. Although this organism does not have a bridge, it also shows a ‘non-canonical’ A/D behavior of a different kind: CI does show a lag phase in its activation when substrate is re-introduced (suggesting that D. melanogaster CI does go into an off-pathway resting state), but the activation is insensitive to the presence of NEM. This behavior correlates with a novel helix in NDUFS4 and a novel conformation for the C-terminal loop of NDUFA9 that lock the CI arms and hold several quinone-binding-site loops under a ‘latch’ [72]. These structural features may be part of a regulatory mechanism controlling the transitions between the CI active and resting states [72]. Together, this growing evidence suggests that the A/D transition may not be a binary feature of CI (absence/presence) and/or that the path from active to deactive may be differently traversed and regulated in different organisms. Therefore, studies of CI structure and A/D transition in as many and as biodiverse organisms as possible are needed.

When the quinone pool is over-reduced and the protonmotive force is high (high quinol/quinone and NAD⁺/NADH ratios), there might be sufficient thermodynamic force to allow reverse electron transfer (RET) from quinol to NAD⁺ by CI. This would be detrimental, as it would lead to high levels of reactive oxygen species (ROS). To prevent this, mammalian CI undergoes deactivation [70,73]. It was recently shown in a reconstituted system that deactive CI is not able to perform RET [74], confirming the protective nature of the A/D transition in mammals. It is currently unknown whether plant CI is capable of RET and whether a deactivation-like behavior of plant CI may protect against it. We previously hypothesized that the presence of the alternative oxidative phosphorylation chain in plants likely pushes CI closer to a RET regime (high quinol/quinone, high NAD⁺/NADH ratios, see [22] Appendix 1) and that, therefore, a protective deactivation mechanism for CI is expected [22]. Yeast also contain an alternative oxidative phosphorylation chain with NDH and AOX homologs [75]. Interestingly, whereas CI from yeast Y. lipolytica is incapable of RET, CI from yeast Pichia pastoris performs RET at rates indistinguishable from those of bovine CI [74]. The inability of Y. lipolytica CI to perform RET is likely related to is tendency to easily enter the deactive state, as seen in functional and structural studies [23,51]. These differences suggest that the relationships between the presence of an alternative chain and CI structure and A/D dynamics will require careful examination. It is critical to determine the RET capabilities of plant CI in the active and deactive states, whether plant CI displays novel structural mechanisms for deactivation and how the activity of the alternative respiratory chain affects these issues.

In V. radiata’s SC I + III₂ structure, we observed differences in the MPP-α isoforms present in the CIII₂ protomers. The protomer closest to CI corresponded to gene LOC106765382, which is different from LOC106774328, which we had previously observed in SC III₂+ IV and free CIII₂ from V. radiata [12]. The other protomer had more ambiguous density, with most positions more closely fitting LOC106774328, but some more closely resembling LOC106765382. Moreover, both protomers lacked their catalytic Zn²⁺ in MPP-β due partial disorder in the helix containing one of the coordinating residues (Glu217). Therefore, these structures correspond to a peptidase-inactive CIII₂. In contrast, both Zn²⁺ atoms had been visible in V. radiata SC III₂+ IV and free CIII₂ [12]. Although their peptidase activities were not tested, they are predicted to be active peptidases. Fittingly, CIII₂ exhibits significant conformational flexibility as a free complex or within SC III₂+ IV [12] but not within SC I + III₂ [7,8,12], likely due to the restraints imposed by the matrix interface with CI. Therefore, the current V. radiata structures suggest that CIII₂ may be an active peptidase in SC III₂+ IV or as an active complex, but not in SC I + III₂. In the A. thaliana SC I + III₂ structure both MPP-α isoforms were identical to each other, and both contained an ordered catalytic site in MPP-β with density for their coordinated Zn²⁺ (including a potentially coordinating water molecule), predicting an active CIII₂ in SC I + III₂. Atomic models for A. thaliana SC III₂+ IV or free CIII₂ are not available. Moreover, although A. thaliana contains two MPP-α isoforms, both exhibit higher sequence similarity to V. radiata LOC106774328 (isoform in SC III₂+ IV) than to LOC106765382 (isoform in SC I + III₂). This inconclusive homology precludes straightforward inferences about A. thaliana’s MPP isoforms.

However, the V. radiata results raise multiple questions. Is CIII₂’s peptidase activity different in different supercomplex assemblies? If so, is the sorting based on the MPP isoforms? Are as-yet unidentified supercomplex assembly factors involved in differential sorting? Or is the isoform irrelevant and it is rather the interactions with CI that regulate MPP activity (e.g. MPP activity only from the MPP protomer not contacting CI)? Are there major differences between plant species? Support for the concept of differential sorting into separate supercomplexes comes from mammals, where, for instance, CI, CIII₂ and CIV can assemble into different respirasomes (C-MRC, S-MRC) based on the differential incorporation of COX7A isoforms [76] (see below). There is also some initial evidence for potential supercomplex assembly factors, although their existence remains unclear [77]. Answering these questions will require the functional, structural and complexome study of mutants of the various MPP-α and -β isoforms in multiple plant species.

The physiological roles of organizing the mETC into supercomplexes over having only individual complexes remain unclear. The fact that the available SC I + III₂ structures show interfaces involving different species-specific details and subunits across four eukaryotic supergroups (mammals [78], alveolates [10], euglenids [11], plants [7,8] in Opisthokonta, TSAR, Excavata and Arachaeplastida, respectively) while sharing an overall similar location, orientation and architecture suggests that there is functional significance through convergent evolution [8,10]. What is the advantage of assembling respiratory complexes into supercomplexes?

Current hypotheses for the ‘structural’ role(s) of supercomplexes include to aid the assembly and stability of the individual complexes, to limit aggregation in the crowded environment of the cristae and to provide stoichiometric balance between complexes to prevent local differences in the quinol/quinone ratio thereby reducing ROS [79–81]. Evidence has been growing for a cooperative-assembly role for supercomplexes [77]. In this model, a key role of supercomplexes is to assist and promote the assembly and stabilization of the individual complexes. For instance, CI, CIII₂ and CIV assembly intermediates have been shown to associate in supercomplexes before (rather than after) the full assembly of the individual complexes [77]. There is also some initial evidence that supercomplex-specific assembly factors may exist to regulate the assembly of complexes via distinct supercomplex-first or individual complex-first biogenesis pathways [82,83].

Recent papers have also investigated the ‘functional’ consequences of an inability to form SC I + III₂. As discussed above, in A. thaliana the assembly of CI and CIII₂ into SC I + III₂ was almost completely disrupted by knocking out CI subunits NDUFX (in the bridge domain) or NDUA11 (interface subunit) [58]. The SC-less plants did not have phenotypes under the standard conditions assayed; however, stress conditions were not reported in that study. In mouse, the matrix interface (UQCRC1:NDUFB4) was disrupted by mutating three key residues on UQCRC1 (MPP-α) [84]. Conversely, the same interface was disrupted in human HEK293 T cells by mutating two key residues in NDUFB4 [85]. In both cases, the levels of CI-containing supercomplexes decreased ∼70–90%. With these low levels of SC I + III₂, under the conditions tested, the UQCRC1-mutated SC-less mice showed normal bioenergetic capacity (NADH oxidation, ROS production by CI, oxygen consumption rates, mitochondrial ATP production/oxygen consumption); in contrast, the NDUFB4-mutated SC-less human cells showed significantly reduced ATP-linked respiration and a compensatory shift toward succinate (CII)-linked respiration. From this contradiction, it is clear that further examination of the physiological effects of SC I + III₂ disruption under various conditions is needed.

Furthermore, it was recently shown that two SCI + III₂+ IV assemblies (C-MRC and S-MRC) with different isoforms of a CIV subunit (COX7A1/2 vs SCAFI) co-exist in human cells [76]. C-MRC vs S-MRC levels also vary in different human tissues [76]. CRISPR knockouts cells with only one type of the above supercomplexes showed the same respiratory rates but different net OXPHOS capacity, suggesting different abilities of C-MRC vs S-MRC to couple electron transfer to ATP synthesis. The authors also proposed that a metabolic shift from glycolytic to oxidative metabolism can reversibly regulate the relative abundance of C-MRC vs S-MRC [76]. Another recent paper investigated the roles of SC I + III₂ in D. melanogaster through a different approach. Fly mitochondria normally contain very low levels of SC I + III₂, likely owing to the truncation of NDUFB4's N-terminus [72,86]. Therefore, rather than assessing the consequences of disrupting supercomplexes, the study investigated the effect of increasing the level of SC I + III₂ by perturbing assembly factors of CI or CIII₂ or decreasing the levels of CI itself [87]. Mitochondria from flies with increased SC I + III₂ levels did not show respirometric differences from controls (beyond that from increases in protein levels of the individual complexes) or a preferential use of CI-linked electrons by CIII₂ (no increase in electron transfer efficiency between CI and CIII₂). Although this argues against a catalytic role for the organization into SC I + III₂, many questions remain — not least, whether D. melanogaster findings can be extrapolated to other organisms given that it does not normally form abundant supercomplexes.

Overall, given the different experimental settings and assays, as well as the contradictory results, it is too early to conclude that SC I + III₂, and respiratory supercomplexes in general, do not have roles ‘beyond’ structural ones. It is also possible that supercomplexes have different roles in different organisms and physiological settings. For instance, could supercomplexes be providing a kinetic advantage through complexes I–IV to regulate electron flow through the canonical vs alternative chains, in organisms like plants and fungi that possess both [88]? It also remains to be determined why there would exist regulatory supercomplex assembly factors (proposed to regulate the biogenesis of the OXPHOS complexes via supercomplexes or via individual complexes [77]), or regulated OXPHOS chain organizations into C- vs S-MRC supercomplexes (proposed to have different preponderance regulated by the extent of glycolytic vs oxidative metabolism [76]) if supercomplexes did not have catalytic roles. Why regulate something that provides no advantage?

The holy grail in the field seems to be to determine whether supercomplexes have a ‘functional’ (catalytic) role beyond ‘merely structural’ ones. Given that structural roles are functional roles, the functional vs structural differentiation may be a false dichotomy. For instance, even if the ‘only’ role of SCI I + III₂ were to promote proper and timely assembly of CI and CIII₂, this would be a significant functional role — regardless of whether the supercomplex organization also affected catalytic rates or the respiratory capacity of the cell.

Overall, studies, interpretations and proposed models for the functions and mechanisms of supercomplexes and the individual complexes should account for observations across the tree of life when proposing universal mechanisms and consider that functions may be different in different clades. Studies across further clades will enhance our understanding of the mechanisms of respiratory complexes and supercomplexes.

• Respiratory supercomplexes are the main physiological form of respiratory CI in plants (and other organisms), yet the roles of the supercomplexes and the mechanisms of the individual complexes remain incompletely understood.

• Recent structural and functional studies of plant supercomplex I + III₂ from two different species allow for further insight into CI, CIII₂ and supercomplexes by comparing to each other and to non-plant organisms.

• Many open questions remain. How does subunit composition diversity contribute to CI's assembly and regulation? What are CI's proton pumping pathways, and how conserved are they? Does CI's γCA domain contribute to proton pumping (in what organisms)? How does CI's bridge affect CI assembly, deactivation and supercomplex formation? Are CI's loop conformations related to its catalytic cycle or to its active-to-deactive transition? Should we re-conceptualize the active-to-deactive transition? Can plant CI carry out RET? If so, what is the relationship with deactivation and the alternative respiratory chain? Is plant CIII₂’s MPP domain active in supercomplexes? Do MPP isoforms lead to differential supercomplex sorting? Do supercomplexes affect respiratory function? Are supercomplex roles different in different clades?

The author declares that there are no competing interests associated with this manuscript.

Open access for this article was enabled by the participation of University of California in an all-inclusive Read & Publish agreement with Portland Press and the Biochemical Society under a transformative agreement with UC.

I thank James Letts, Hans-Peter Braun and reviewers for their helpful comments on the manuscript.

γCA

γ-carbonic anhydrase

cryoEM

cryo-electron microscopy

IMS

inter-membrane space

mETC

mitochondrial electron transport chain

MPP

mitochondrial processing peptidase

RET

reverse electron transfer