Mapping the evolution of mitochondrial complex I through structures

Abstract

Respiratory complex I (CI) is a multi-subunit membrane protein complex important to produce ATP via the oxidative phosphorylation pathway. The structure of CI is roughly conserved across species and is composed of subunits that are either embedded in the membrane or are exposed to the aqueous environment that together form an overall L-shaped “boot”. The conserved core of CI is generally composed of fourteen subunits. Across species, various less conserved “supernumerary” or “accessory” subunits have been added. Accessory subunits vary in number across species and can include proteins that are unique to specific lineages. Additionally, there are structural variations in the core subunits between clades. In this review, we compare seven representative CI structures from divergent eukaryotic lineages to identify what aspects of the CI core subunits are susceptible to variation and classify eukaryotic accessory subunits into conserved from the last eukaryotic common ancestor (LECA) or lineage specific.

Introduction

Respiratory complex I (CI) is a proton-pumping NADH:quinone (coenzyme Q, CoQ) oxidoreductase central to the metabolism of diverse organisms, including prokaryotes and eukaryotes [1–3]. In eukaryotes, CI, along with complexes II, III₂, and IV (CII-CIV), comprises the electron transfer chain (ETC) of the mitochondrial oxidative phosphorylation pathway. CI is often found as the largest ETC complex and consists of subunits that can be grouped into either the conserved core or the more variable accessory or supernumerary subunits (Fig. 1) [4–7]. Core subunits are essential for the electron-transfer-coupled proton pumping activity of CI. Accessory subunits, though scarce in prokaryotic CIs, are needed for CI’s assembly, function, and stability in eukaryotes [8]. Along distinct prokaryotic lineages, it is common to find unique sets of few accessory subunits (Fig. 2) [9–12], but only the α-proteobacteria, such as Paracoccus denitrificans, have CI accessory subunits homologous to those found in eukaryotic CI [10]. This demonstrates the close relationship between the α-proteobacteria and the bacterial ancestor of the mitochondria [13]. Accessory subunits are prevalent in eukaryotes, and their numbers vary widely across species. For example, there are 31 accessory subunits in human CI compared to 34 in Arabidopsis thaliana and 52 in the ciliate Tetrahymena thermophila [14–18].

Figure 1.

The surface representation of CI modules from P. denitrificans showing only the core subunits [10]. Subunits are labeled, colored individually and highlighted with squares in the left panel based on the modules: N-module (periwinkle), Q-module (emerald, green), P_P-module (rose pink), and P_D-module (burnt orange). The right panel is shown with 20% transparency. For clarity, residues 170 to 200 in ND6 are not shown. Cofactors are shown in the right inset and labeled. Quinone binding site is indicated (black dotted circle). Residues involved in the hydrophilic axis identified in ref. [10] are shown as sticks and colored by atom (carbon grey, oxygen red and nitrogen blue). Visualization throughout was performed with UCSF ChimeraX.

Figure 2.

A comparison of bacterial CI structures with the core shown as cartoons (grey) and accessory subunits shown as surfaces. Shown are CI from (A) E. coli (PDB ID: 7P62) with residues 173–211 connecting S2 and S3 (NuoCD) shown as surfaces (turquoise), (B) M. smegmatis (PDB ID: 8E9G) with its accessory subunit MSMEG_2064 labeled (turquoise), (C) T. thermophilus (PDB ID: 6I1P) with accessory subunits Nqo15 (rosy finch) and Nqo16 (turquoise), and (D) P. denitrificans (PDB ID: 8QBY) with the observed ancestral subunits S4 (emerald green), S6 (plum), and A12 (sandy brown). A9 is also observed in ref. [53] when in a supercomplex with CIII₂ and CIV; its observed location is indicated (red circle). PIMT subunit unique to P. denitrificans CI is also shown (turquoise).

With some exceptions resulting from gene fusions and fissions [11,16,19], the CI core is formed by 14 conserved subunits arranged in a peripheral arm responsible for electron transfer and a membrane arm responsible for proton-pumping, making a “boot”-like shape (Fig. 1). CI can further be divided into four modules based on function and location (Fig. 1, left). The peripheral arm is divided into two modules, the NADH-oxidizing N-module and the CoQ-reducing Q-module [20]. The N-module is comprised of core subunits NDUFV1, NDUFV2, and NDUFS1; the Q-module is comprised of NDUFS2, NDUFS3, NDUFS8, and NDUFS7. The N-module contains one flavin mononucleotide (FMN) cofactor, two 2Fe-2S clusters (N1a and N1b), and three 4Fe-4S clusters (N4, N5, and N3) [20]; the Q-module contains three 4Fe-4S clusters (N6a, N6b, and N2). Together, the N- and Q-modules form the ~100 Å long electron transfer pathway that conducts electrons from NADH at the FMN site in NDUFV1 to the quinone-binding site in NDUFS2 adjacent to the membrane (Fig. 1) [20,21].

Core subunits in the membrane arm are structurally homologous to bacterial cation/H⁺ antiporters and are responsible for transporting H⁺ across the membrane [20,22]. The membrane arm is divided into two pumping modules, one proximal (P_P) and another distal (P_D) to the peripheral arm (Fig. 1). During a single catalytic turnover, CI in the mitochondria pumps four protons across the inner mitochondrial membrane (IMM) from the matrix side to the intermembrane space (IMS) [23]. Although the CI mechanism of electron-transfer-coupled proton pumping is still debated [14,24], protons are thought to travel along a water-mediated network within the membrane arm core subunits. Antiporter-like subunits ND5, ND4, and ND2 have highly conserved transmembrane helices (TMHs) containing key Lys and Glu residue pairs thought to aid proton translocation across the membrane [25,26]. These subunits also form a Grotthuss-competent network that spans the entire membrane arm known as the hydrophilic axis (Fig. 1) [1,26], but its exact role in proton translocation is debated [24].

In eukaryotic CI, numerous accessory subunits surround the conserved core, some of which are conserved across multiple lineages [27]. For example, all eukaryotic CIs characterized so far share 26 accessory subunits (including two copies of NDUFAB1) (Fig. 3). Ten conserved accessory subunits are in the peripheral arm, including the four accessory subunits found in α-proteobacteria. The remaining 16 are in the membrane arm. Additionally, there are accessory subunits with mixed distribution across lineages or specific to individual lineages (Table 1).

Figure 3.

Shared subunits of mitochondrial CI from the seven species explored in this review, compared to P. denitrificans CI. Lineage-specific subunits are not shown. Subunits are grouped as follows: present (dark blue), not present (white circle), truncation observed (light blue), extension observed (red), potential substitutes (grey), and duplicates (green). Subunit splits (half-sectors) and heterotrimers (one-third sectors) are also indicated. Subunits that are expressed in the mitochondrial genome are also indicated (gold box), and classifications are based on protein-nucleotide BLASTs. Since the genome annotation for P. parva was incomplete, the classification of proteins encoded in the mitochondria in C. reinhardtii was used as an additional reference. Several subunits from E. gracilis gave partial BLAST hits for both the nuclear and mitochondrial genomes (dashed gold box). The figure was prepared using TBtools-II [128].

Table 1.

Classification of CI core and accessory subunits

Module	Core	Accessory subunits
		Ancestral*	Conserved Across Eukaryotes	Opithokonts		SAR	Dicoba	Chloripastida
				Bilaterians	Fungi	Tetrahymena	Euglena	Algae	Plants
N	V1, V2, S1	S6	A2	V3		T2	EG1, EG2, EG3, EG5, EG12
Q	S2, S3, S7, S8	A9, A12, S4	A5, A6, A7, AB1		ST1	T1, T15	EG4, EG6
PP	ND1, ND2, ND3, ND4L, ND6		A1, A3, A8, A11, A13, C2, S5	A10, C1	NUXM	γ-CA, J1, TX, PH1, PH2, T6, T7, T10, T12, T13, T14, NUXM, CI-FDX	γ -CA, EG6, EG10, UNK1, NUXM, CI-FDX	γ -CA, P9, NUXM, CI-FDX	γ -CA, P2, P9, NUXM, CI-FDX
PD	ND4, ND5		B2, B3, B4, B6, B7, B8, B9, B10, B11, AB1	B1, B5	B5	T3, T4, T5, T8, T9, T11, T16, T17	EG7, EG8, EG11, EG13, B5	P4, P7, P10, P1 (B5)	P1 (B5)

^*Present in α-proteobacteria

It is unclear how and why these accessory subunits vary in structure and number across eukaryotes. To date, there are CI structures from 19 different species: four prokaryotes [9–12] (Fig. 2) and 15 eukaryotes (Table 2). Mapping the eukaryotic structures onto a simplified tree of eukaryotic biodiversity allows us to put these structures into their evolutionary context (Fig. 4). The last eukaryotic common ancestor (LECA) marks the divergence into two large clades known as the Opimoda+ and Diphoda+ (Fig. 4) [13]. Within the Opimoda+ and Diphoda+, eukaryotes can be further organized into supergroups whose exact number and organization are still being worked out [13,28]. Figure 4 gives a simplified schematic of the eukaryotic tree based on recent studies [13,28–30]. When we map the currently available structures of mitochondrial CI to the eukaryotic supergroups, we find that we have representative structures from single clades within only four supergroups: Amorphea, Discoba, Archaeplastida, and SAR (Fig. 4 and Table 2). Of the available species with CI structures, eight are from the Opisthokonta clade within Amorphea [31–39], which includes all fungi and metazoans [30]; five are from the Chloroplastida clade within Archaeplastida [15,40–43], which includes all green algae and land plants [13,30]; one is from Euglenozoa within Discoba [19], which is a highly biodiverse clade of planktonic protists including Euglenida, Diplonemea, and Kinetoplastea [44]; and one is from the Alveolata clade within SAR [16–18], which includes all ciliates, dinoflagellates, and apicomplexa [13,30,44] (Table 2). Thus, there is an over-representation of two clades, Opisthokonta and Chloroplastida in CI structures (Table 2). Further, while there are structures from three clades, Discoba, Chloroplastida, and Alveolata, from the Diphoda+, we only have structural information from a single clade, Opisthokonta, from Opimoda+. This demonstrates that although the field has come a long way since the first structures of mitochondrial CI nearly a decade ago [35,45], the structural understanding of this important core metabolic complex remains in its infancy.

Table 2.

Classification of eukaryotic species with CI structure information

Species	PDB (ref)	Class (Order)	Kingdom/Division	Clade	Supergroup	Clade
Homo sapiens	5XTD (ref [39])	Mammalia (Primates)	Metazoa (Bilaterians)	Opisthokonta	Amorphea	Opimoda+
Bos taurus *	8Q48 (ref [33])	Mammalia (Artiodactyla)
Sus scrofa	7V2C (ref [34])	Mammalia (Artiodactyla)
Ovis aries	5LNK (ref [35])	Mammalia (Artiodactyla)
Mus musculus	6ZR2 (ref [38])	Mammalia (Rodentia)
Drosophila melanogaster *	8ESZ (ref [36])	Insecta (Diptera)
Yarrowia lipolytica *	6YJ4 (ref [31])	Saccharomycetes (Saccharomycetales)	Fungi
Chaetomium thermophilum	7ZMG (ref [32])	Ascomycota (Sordariomycetes)
Polytomella parva *	7ARD (ref [42])	Chlorophyceae (Chlamydomonadales)	Chlorophyta (green algae)	Chloroplastida (Viridiplantae)	Archaeplastida	Diphoda+
Chlamydomonas reinhardtii	9F5Y (ref [15])	Chlorophyceae (Chlamydomonadales)
Arabidopsis thaliana	7ARB (ref [42])	Rosids† (Brassicales)	Streptophyta Or Embryophyta (land plant)
Brassica oleracea	7A23 (ref [40])	Rosids† (Brassicales)
Vigna radiata *	8E73 (ref [41])	Rosids† (Fabales)
Tetrahymena thermophila *	7TGH, 8GYM (refs [16,17])	Oligohymenophorea (Hymenostomatida)	Ciliate	Alveolata	SAR
Euglena gracilis *	8J9J (ref [19])	Euglenida (Euglenales)	Euglenozoa	Discicristata	Discoba

^*Species chosen as examples in this study

^†Clade

Figure 4.

Simplified schematic of the eukaryotic tree of life showing the early Diphoda+ and Opimoda+ split and eukaryotic supergroups. The relative positions on the tree are schematically represented based on ref. [13] and [28]. Dashed lines indicate a high degree of uncertainty. Clades where CI structures are known are marked with a yellow star.

Here, we compare the structure of CI from seven species, each representative of a different major lineage of eukaryotes (Table 2 and Fig. 5) [16,17,19,31,33,36,41,42]. Our goal is to use these structures to identify evolutionary constraints on the variability of the core subunits and map the lineage-specific acquisition of accessory subunits. By comparing the structures from these diverse lineages, we also aim to deduce the subunit composition of CI from LECA (Fig. 4). The structure of P. denitrificans CI, with its four ancestral accessory subunits shared by all eukaryotic CI structures to date, is used as an ancestral blueprint of the core subunits and the likely CI composition in the first eukaryotic common ancestor (FECA). We use human gene names (e.g., NDUF) for CI subunits to minimize confusion from employing different naming schemes that exist across species [4]. For simplicity, after introducing the subunit, we remove the common NDUF root and refer to it with its unique designation, i.e., NDUFA6 will be referred to simply as A6. We also identify subunit naming discrepancies based on structural homology to better organize and understand subunit conservation across eukaryotes. The purpose of this review is not to perform a detailed phylogenetic analysis of each subunit to track its evolution and distribution, but to identify patterns from the structures that are worth pursuing with more detailed phylogenetics. Across large evolutionary distances, respiratory complex structures have proven more informative than sequence analysis alone for identifying homologues. For example, proteomics identified several subunits of CIV as unique to plants [46,47]; however, when the structure of plant CIV was determined, these subunits were found to be homologues of subunits found in other organisms [48]. This is particularly true of small hydrophobic subunits where structural homology can be maintained while sequence homology is lost. Thus, structures are needed to generate evolutionary and phylogenetic hypotheses across vast timescales of biodiversity.

Figure 5.

CI structures from the seven selected species, classified into either Opimoda+ or Diphoda+ clades. Shown are core subunits (grey cartoons) with accessory subunits highlighted as surfaces, grouped into eukaryotic-conserved subunits (neon green), mixed-lineage subunits (cornflower blue), and lineage-specific subunits (orange-red).

Bacterial accessory subunits and eukaryotic ancestry

Accessory subunits are not unique to eukaryotic CI. Of the four known prokaryotic CI structures, three have accessory subunits in addition to the core (Fig. 2). Escherichia coli CI is an exception, as it has no accessory subunits, but instead is distinct in its fusion of the NDUFS2 and NDUFS3 subunits [11] (NuoCD, Fig. 2A). The E. coli S2 and S3 subunits are linked by a long extension that occupies the same space on the complex as accessory subunits in other species [4] (Fig. 2). CI structures from Mycobacterium smegmatis [12] and Thermus thermophilus [9] have one and two accessory subunits, respectively (Fig. 2B and C). Each of these accessory subunits is derived from globular, well-structured protein domains that associate with the peripheral arm of the complex (Fig. 2). In the case of M. smegmatis CI, its accessory subunit MSMEG_2064 is a two-component response regulator protein, whose cognate kinase is unknown but whose phosphorylation site is conserved [12]. Importantly, its phosphorylation site is buried at the interface between MSMEG_2064 and the CI core, indicating that its association may be dependent on phosphorylation state, potentially allowing MSMEG_2064 to act as a dynamic regulator of M. smegmatis CI assembly [12]. In T. thermophilus, the accessory subunit Nqo15 is structurally homologous to the frataxin family of iron chaperones that are involved in the maturation of Fe-S and heme-containing proteins, and has thus been suggested to be the first example of a frataxin-like protein permanently bound to its target [49,50]. The second T. thermophilus accessory subunit Nqo16 is a member of the DUF3197 superfamily of proteins with no known function; it is not required for the activity of CI in T. thermophilus, but is encoded in the nqo operon, and its association is thought to stabilize the peripheral arm of CI [51].

Thus far P. denitrificans CI has the largest number of accessory subunits observed in prokaryotes: NDUFA9 (A9), NDUFA12 (A12), NDUFS4 (S4), and NDUFS6 (S6) (Fig. 2D), as well as a lineage-specific PIMT subunit [10,52,53]. The structures of P. denitrificans A9, A12, S4, and S6 are conserved across eukaryotic lineages [4] (Fig. 3 and Table 1), indicating that these subunits were present in the bacterial mitochondrial ancestor, retained from FECA to LECA, and maintained in extant mitochondrial CIs. P. denitrificans A9 was not observed in the isolated structure of CI, but was seen in the structure of supercomplexes formed between CI, CIII₂, and CIV [10,53]. The PIMT subunit is homologous to enzymes that catalyze repair of degraded aspartate and asparagine residues [54] and is thought to have been associated with P. denitrificans CI along the α-proteobacterial lineage after divergence from the proto-mitochondrion [10].

Interestingly, although the bacterial accessory subunits evolved independently along specific bacterial lineages, they are found in similar locations on the CI structure, namely the PA often bridging the N- and Q-modules (Fig. 2). This suggests that bacterial accessory subunits and subunit extensions are examples of convergent evolution stabilizing the complex in response to specific selection pressures, as has been proposed previously [4]. Further, this suggests a natural tendency for the accretion of subunits on CI, as this has occurred independently along several prokaryotic lineages. The somewhat vague notion of “CI stability” may provide selective advantage that partially explains the tendency of CI to accumulate subunits. However, the accretion of subunits may also be examples of constructive neutral evolution (CNE) [55,56], which explains how increases in complexity can arise that are not necessarily advantageous or require positive selection. Whatever may be the driving principle behind the accumulation of subunits on CI, it appears to have greatly accelerated upon the evolution of eukaryotes.

Variability in the core across eukaryotes

CI is composed of 14 canonical core subunits. Seven hydrophobic subunits (ND1, ND3, ND6, ND4L, ND2, ND4 and ND5) form the membrane arm of CI (Fig. 1 and Table 1) and are mostly found encoded in the mitochondrial genome (Fig. 3) [57–63]. Seven hydrophilic subunits (NDUFV1, NDUFV2, NDUFS1, NDUFS2, NDUFS3, NDUFS7, and NDUFS8) make up the peripheral arm of CI and vary across eukaryotes in terms of gene location between nuclear and mitochondrial genomes [1,57,64]. For example, in mammals, only the seven membrane core subunits are encoded by the mitochondrial DNA [57,58], while the nuclear genome encodes the rest (Fig. 3) [1,57,64]. In contrast, the nuclear genome encodes the hydrophobic subunit ND1 in the algae Chlamydomonas reinhardtii [65] while the hydrophilic subunits S2 and S3 are encoded by the mitochondrial genome in vascular plants (Fig. 3) [66]. The mitochondrial genome of Amoebozoa Acanthamoeba castellanii encodes for 10 core subunits, excluding S7, S8, V1, and V2 [59] and In a marked example, the mitochondrial genome of the discobid Andalucia godoyi encodes for 12 of the core subunits, excluding only V1 and V2 [60], a pattern observed in other eukaryotes from both Opimoda+ and Diphoda+ [61–63]. This suggests that in LECA 12 core subunits were encoded in the mitochondrial genome and that V1 and V2 had already transferred to the nuclear genome along with the four α-proteobacteria derived accessory subunits (S4, S6, A9, and A12).

The core subunits are present in all CI structures and are needed for function (Table 2), yet variation exists across species. There are multiple facets where even the highly conserved core subunits can undergo significant changes. Here we discuss the observed structural variations of selected core subunits, ND1, ND2, ND5, and S1.

ND1.

ND1 is located on the P_P-module and is composed of eight TMHs[1,20,51,67], with its fold homologous to a half-channel of the antiporter-like subunits ND2, ND4 and ND5 (Fig. 6A) [1,51,67]. ND1 forms the CI CoQ-binding site with S2, S7, and ND3. Specifically, TMH 1 from ND3, TMHs 1 and 6 from ND1, and an amphipathic helix connecting TMHs 1 and 2 form the entrance to the hydrophobic quinone cavity [1,20]. In the Alveolate T. thermophila, ND1 is split into two subunits, ND1a and ND1b, containing the ND1 equivalent TMHs 1–6 in ND1a with the quinone binding site and TMHs 7–8 in ND1b [16] (Fig. 6B). To date, splitting of the ND1 subunit has only been observed in T. thermophila [16], indicating that ND1 was encoded by a single gene in LECA. The location of the split in ND1 along Alveolata or more broadly in SAR remains an open question.

Figure 6.

ND1 is split in T. thermophila. (A) P. denitrificans and (B) T. thermophila, ND1 from matrix view showing the ND1 TMHs and their numbering. Location of the ND1 subunit in relation to the entire CI is shown in (A; matrix up, IMS down). The approximate location of the perpendicular slice is indicated (red dashed line). For (B) T. thermophila, the split chains ND1B and ND1B are shown. Equivalent TMH numbering with P. denitrificans for T. thermophila ND1b are shown in parentheses.

ND2.

ND2 is responsible for proton translocation across the membrane together with other core membrane subunits of the P_P-module (Fig. 1 and Table 1). ND2 is one of three antiporter-like subunits structurally homologous to the bacterial cation/H⁺ Mrp antiporter subunits (MrpD and MrpA, the other two being ND4 and ND5) [20,22,68]. The structures of these antiporter-like proteins contain 14 TMHs, of which TMHs 4–13 are highly conserved [20]. TMHs 4–8 and TMHs 9–13 each form half-channels containing solvent-accessible cavities that open to one side of the membrane [69]. These channels contain a spatially conserved Glu/Lys pair in TMHs 5 and 7, Lys or His on TMH 8, and Lys or Glu on TMH 12 that behave as pH-dependent conformational switches for unidirectional proton translocation [20,26].

Among the conserved antiporter proteins, ND2 varies the most (Fig. 7). TMHs 1–3 are absent in Bos taurus [33] and Drosophila melanogaster [36,37] CIs (Figs 7G and H). The loss of these N-terminal helices is specific to the metazoan lineage [70,71], as CI structures from other Opisthokont lineages maintain all 14 TMHs, i.e., fungal species Yarrowia lipolytica and Chaetomium thermophilum [31,32,71]. Although structures of non-bilaterian metazoan CIs are currently unavailable, sequence analysis suggests that truncation of ND2 is specific to the bilaterians, as the sister clade Cnidaria maintains the ancestral “long” ND2 sequence [71]. Despite this extensive truncation, TMHs 4–13 that form the solvent channels are still conserved (compare Figs 7A with G and H). Sequence alignments and high-resolution structures show that conserved Lys-Glu/His residue pairs are still present in bilaterian ND2 antiporter subunits [26,71]. Therefore, these structural changes are unlikely to have influenced the proton translocation function of CI [23], and the possible vestigial nature of the TMHs 1–3 has been suggested [72].

Figure 7.

ND2 subunits are split or truncated in different lineages. The location of ND2 in the CI structure is shown (top middle) and cartoons of ND2 subunits for (A) P. denitrificans, (B) P. parva, (C) V. radiata, (D) T. thermophila, (E) E. gracilis, (F) Y. lipolytica, (G) D. melanogaster, and (H) B. taurus are shown as central slices through the TMHs (left) and from the side view (right; matrix up, IMS down) in neon blue and lava red for split segments highlighted with surfaces shown. The NUXM subunit is also shown if present (light pink). In (A) TMHs in ND2 are labeled (red for TMHs absent in metazoan structures). The approximate line of the TMH slice is indicated in (A) by the red dotted line. For E. gracilis, a putative NUXM subunit paralog, EG10 is shown (green).

However, if the missing helices were truly vestigial, it would be expected that they would be absent in other eukaryotic lineages as well, which has not been observed to date. Thus, peculiarities in the bilaterian CI structure or composition may have relaxed selection pressure on the structure of ND2, allowing these TMHs to be lost. One clear difference in bilaterian CI is the presence of the lineage-specific NDUFA10 (A10) subunit, which binds on the matrix side of ND2. Sequence analysis suggests that A10 evolved early in metazoans [73]. Still, structures of non-bilaterian metazoan CI are needed to confirm the extent of its presence in metazoans and to map possible changes in its interactions with ND2 that could explain the relaxation of ND2 structural integrity. Interestingly, A10 occupies the space normally occupied by the γ-carbonic anhydrase (γ-CA) subunits found in all non-Opisthokont CI structures to date. Thus, the incorporation of A10 into the CI structure was contingent on the loss of the γ-CA subunits, which is thought to have occurred early in Opisthokonta (see below for further discussion on the γ-CA subunits) [74–76].

Besides truncation, ND2 subunits in both T. thermophila [16–18] and E. gracilis [19] have been split into two separate subunits, ND2a and ND2b. These two subunits come together to form a canonical ND2 structure (Fig. 7). In T. thermophila [16–18], the ND2a subunit contains TMHs 1–10, and ND2b contains TMHs 11–14 (Fig. 7B). Despite being composed of two separate proteins, the overall alignment of the TMHs in T. thermophila ND2a and ND2b, particularly those containing the conserved charged-residue pairs, is comparable to that of other eukaryotes and prokaryotes (compare Figs 7D to A–C, E–F). Like T. thermophila [16–18], the ND2 functional unit in E. gracilis [19] is composed of two subunits ND2a and ND2b. However, ND2a contains TMHs 1–11, and ND2b contains TMHs 12–14. The differences in the location of the ND2 split (between TMHs 10 and 11 in T. thermophila [16–18] vs. between TMHs 11 and 12 in E. gracilis [19]) and the location of the genes suggest that ND2 split independently along these different eukaryotic lineages. Hence, ND2’s structural integrity is under less selection pressure than that of ND4, which to date has only been observed as a single polypeptide.

The structural variations observed in the ND2 subunit demonstrate that flexibility exists in the structure of ND2 across species. Additional factors that may select for variation in ND2, i.e., allowing the CI membrane arm (MA) to adapt to different membrane curvature [77], or relax selection for ND2 structural integrity, i.e., compensatory interactions with lineage-specific accessory subunits, remain to be fully elucidated. Furthermore, these structural differences in the number and location of TMHs have implications for their interaction with the accessory subunit NUXM, which is discussed below. Overall, the structure of ND2 seen in P. denitrificans [10], with all 14 TMHs encoded in a single gene, remains the best model for the ND2 subunit present in LECA. This indicates that the ND2 structure remained stable from FECA to LECA and diverged later along specific eukaryotic lineages.

ND5.

Like ND2, ND5 is an antiporter-like protein in the P_D-module (Fig. 1 and Table 2) [20]. The antiporter-like domain is made up of 15 TMHs and is followed by a long lateral amphipathic α-helix that extends from TMH 15 and inserts a TMH (TMH 16) adjacent to ND2 in the P_P-module (Fig. 8A) [20]. The lateral helix runs alongside ND4, bridging the three antiporter-like subunits (Fig. 8).

Figure 8.

ND5 is split or C-terminally truncated across lineages. Side views showing cartoons and surfaces of subunits ND2 (neon blue, split ND2B subunits in lava red when present), ND4 (vivid yellow), and ND5 (bright green, split ND5B subunits in magenta when present) for (A) P. denitrificans, (B) P. parva, (C) V. radiata, (D) T. thermophila (PDB ID: 8GYM was used for this figure as the ND5 N-terminal domain was well resolved), (E) E. gracilis, (F) Y. lipolytica, (G) D. melanogaster, and (H) B. taurus. A slice across the TMHs parallel to the membrane plane is shown for (A) P. denitrificans (indicated in red dotted line), with the TMHs numbered in ND5 (black, bolded), along with TMHs in ND4 and the three TMHs in ND2 that are truncated in some species (grey). The location of ND5 in CI is also highlighted (top center).

Structural variations in ND5 are observed in the main antiporter domain, the lateral helix, and the C-terminal TMHs. In D. melanogaster [36], Vigna radiata [41], B. taurus [33], and Y. lipolytica [31], the lateral helix extends from TMH 15 and inserts TMH 16 adjacent to ND2 (Fig. 8). However, in Y. lipolytica [31] and V. radiata [41], there is an additional TMH on the C-terminus relative to what is seen in P. denitrificans CI or other prokaryotic CI structures [9–12] (Figs 8C and F). In E. gracilis [19], the C-terminal TMH is not observed, and the lateral helix extends only to ND4 (Fig. 8E). Like ND2, ND5 is also split and encoded by separate genes in some organisms. In T. thermophila, ND5 is split into two subunits [16], ND5a and ND5b. ND5a comprises the antiporter-like domain with an extended N-terminal domain on the IMS side of the IMM, and ND5b comprises the lateral α-helix and the C-terminal TMH equivalent to TMH 16 in P. denitrificans [10] and B. taurus [33] (Fig. 8D). In the green algae Polytomella parva (P. parva, subunit names based on ref. [78]) and C. reinhardtii [15], ND5 is also split into two domains. However, unlike T. thermophila [16], the split occurs midway along the lateral helix, and ND5b (termed P8) contains two TMHs adjacent to ND4 as seen in V. radiata [41] and fungi [31,32] (compare Figs. 8B and D).

The role of the lateral helix in the CI coupling mechanism has been much debated [69,79–82]. Originally, it was proposed to act as a conformational coupling element, analogous to the coupling rod of a locomotive, that transmits conformational changes between the CoQ reduction site and the antiporter-like subunits [69,79]. However, mutagenesis and cross-linking studies in E. coli CI suggest that it plays less of a functional role and more of a structural role, acting as a “clamp” that holds the CI membrane arm together [80–82]. The variability in the structure of the lateral helix across eukaryotes is consistent with the addition of accessory subunits rendering the ND5 lateral helix partially redundant and relaxing selection for its stabilizing role.

Comparison of the structures across eukaryotes suggests that LECA’s ND5 was encoded by a single subunit with two C-terminal TMHs. Two lines of argumentation support this hypothesis. First, Y. lipolytica and V. radiata are species from Opimoda+ and Diphoda+ [30], respectively, on either side of the early eukaryotic divergence from LECA [13,28] (Fig. 4). This indicates one of two possibilities, either: 1) LECA ND5 had two C-terminal TMHs and one helix has been lost in multiple lineages, as in Discoba and metazoans, or 2) LECA ND5 had a single C-terminal TMH, and convergent evolution in plants and fungi both led to the addition of a second TMH. Assuming its structural role, divergence from a common “two C-terminal TMH” ancestor appears more likely than convergence, as changes in accessory subunits likely relaxed selection for the lateral helix’s structural role. Further, the C-terminal TMH of Y. lipolytica [31] and V. radiata [41] ND5 interacts closely with the accessory subunit NDUFA11, suggesting it plays a role in the recruitment of this accessory subunit. A11 is a four TMH accessory subunit with homology to the mitochondrial translocation machinery, specifically Tim17, Tim22, and Tim23 [83]. In the three structures with the second ND5 C-terminal TMH, the additional helix binds in the center of the cavity formed by the A11 THMs, which is analogous to the putative protein translocation path in Tim17 [84]. This suggests that the second ND5 C-terminal TMH recruited A11 to CI by mimicking a peptide transport substrate. Given that A11 is present in all current eukaryotic CI structures (Fig. 3), we can infer that A11 was present in LECA CI. If so, the ancestral form of eukaryotic ND5 had two C-terminal TMHs, despite P. denitrificans CI [10] and thus likely FECA CI only having one. This suggests that ND5 in LECA was like that of Y. lipolytica [31] and V. radiata [41], with the final TMH interacting intimately with A11 (Figs 8C and F). In distinct lineages, changes in nearby subunits, including A11, may have relaxed selection for the lateral helix’s structural role and the need for the final TMH to support A11 binding, resulting in the observed structural divergence. Structures of CI from more eukaryotic lineages are needed to confirm this hypothesis.

NDUFS1.

NDUFS1 (S1) is the largest core subunit. S1 is a part of the electron transfer chain between NADH and CoQ, and its N-terminal domain contains the FeS clusters N1a, N4, and N5 (Fig. 9A) [85]. The global structure of S1 is conserved across organisms and consists of two regions: a smaller N-terminal domain containing three FeS clusters (with an additional “G-loop” domain in bacteria), and a larger C-terminal Rossmann-fold domain thought to play a structural role through its interactions with other subunits [4,86].

Figure 9.

The S1 subunit is split in E. gracilis. Cartoon representations of S1 from (A) P. denitrificans and (B) E. gracilis are shown. In (A), the location of S1 in relation to the entire CI structure is shown (matrix up/IMS down), with the FeS clusters N1b, N4, and N5 labeled. The N-terminal domain and C-terminal Rossman fold domain are indicated. For the S1 subunit in (B) E. gracilis, the split S1 subunit into S1A (red, N-terminal domain) and S1B (purple, C-terminal domain) are indicated.

Accessory subunits S4 and S6 interact with the N-terminal FeS cluster domain, and subunits A2, A6, and A9 interact with the C-terminal Rossmann fold domain. Although the A2 subunit is missing in the CI structure of D. melanogaster [36], this may be a biochemical artefact. Interaction between A2 and S1 increases the stability of CI, as knockouts of A2 in mammalian cell lines lead to the accumulation of subcomplexes [4,8]. S1 from D. melanogaster lacks a helix in the A2 binding site compared to other eukaryotes [36], suggesting conformational flexibility in this region. Such changes in D. melanogaster S1 may highlight the changing dependence on A2 for stability or may explain why A2 falls off more easily during sample preparations.

Among the seven species highlighted here, the most extensive S1 structural variation is observed in the discobid E. gracilis [19]. Unlike other eukaryotes, S1 is split into two subunits (S1a and S1b), each encoded by different genes in the nuclear genome [19] (Figs. 3 and 9B). All other CI structures determined to date have a single gene encoding S1 [15,16,31–43], indicating that the splitting of the S1 gene in Discoba occurred after divergence from LECA. The gene is divided at the junction between the two structural domains, separating the FeS cluster domain (now S1a) from the Rossmann fold domain (now S1b) [19] (Fig. 9B). While the FeS-containing S1a is structurally conserved, S1b has notable differences compared to equivalent S1 domains in other species. For example, S1b in E. gracilis contains extended alpha helices between the β-sheets of the Rossman fold [19]. In E. gracilis CI, S1b is encased by five additional Euglena-specific subunits (E1, E2, E3, E5, and E12) homologous to enzymes related to fatty acid synthesis and anaerobic wax fermentation [19]. These additional subunits may provide extra stability to the peripheral arm, alleviating S1’s role in stabilizing CI and creating a permissive environment for the neutral evolution of S1b.

Accessory subunits

The total number of accessory subunits varies depending on the organism (Fig. 5). Among all known eukaryotic CI structures, 26 accessory subunits are consistently present, while 6 are found in some, but not all, lineages. Additionally, some subunits are observed only in specific species or lineages, adding significant variability to eukaryotic CI structure.

Conserved eukaryotic-specific accessory subunits

Besides the four ancestral accessory subunits, 22 eukaryotic-specific accessory subunits (including two copies of NDUFAB1) are observed in all available eukaryotic CI structures (Figs. 3 and 5). These subunits are A1, A3, A8, A11, A13, C2, S5, B2, B3, B4, B6, B7, B8, B9, B10, B11, A2, A5, A6, A7, and AB1 (Fig. 10 and Table 1). Due to their presence in both Opimoda+ and Diphoda+, these subunits, along with the four that are also present in P. denitrificans, were likely present in LECA CI. The core structural features of these 26 subunits are highly conserved across eukaryotes (Fig. 10). Yet, they are subject to extensions on the N- and C-termini, which can alter their inter-subunit interactions (Fig. 5). Below, we briefly discuss the interesting case of NDUFAB1.

Figure 10.

Conserved accessory subunits in eukaryotic CI. Accessory subunits that lacked extensions or truncations to the basic subunit core structure were chosen as representative structures. S4 and A12 from P. denitrificans (PDB ID: 8QBY); S6, A9, A7, A5, A2, B11, B8, B7, B6, B4, S5, C2, A11, A8, and A1 from V. radiata (PDB ID: 8E73); B5, A3, A13, B2, B3, B9, B10, A6, AB1 from Y. lipolytica (PDB ID: 6YJ4). Core subunits, except for ND6 (from V. radiata), are from P. denitrificans in this representation.

NDUFAB1.

The acyl carrier protein (ACP) subunit NDUFAB1 (AB1) is always found on CI paired with a leucine-tyrosine-arginine motif (LYRM) protein. There are two copies of AB1 in CI, named NDUFAB1-α (AB1-α) and NDUFAB1-β (AB1-β). AB1-α is located at the Q-module, and AB1-β is located at the P_D-module [4]. These subunits are paired with different LYRM subunits: AB1-α with NDUFA6 (A6) and AB1-β with NDUFB9 (B9) [4]. The loss or mutation of A6 has been reported to severely impact CI activity [2,87], and the mutation of B9 reduces both the amount and activity of CI [88]. These two ACP/LYRM subunit pairs are found in all current structures of eukaryotic CI, indicating that they were present in LECA CI.

Despite the structural conservation across eukaryotes, the two AB1 subunits can be either identical copies encoded by a single gene or distinct ACP paralogs encoded by separate genes (Fig. 3) [4,7,89–98]. In metazoan B. taurus [7,33] and D. melanogaster [36], AB1-α and AB1-β are identical proteins sharing the same sequence. In D. melanogaster, there are alternative transcription splice variants for ACP expression from a single gene (Gene ID: 38154) [90,99,100], only one of which matches the sequence of AB1 from D. melanogaster CI [36,37]. These findings suggest that D. melanogaster CI selects for two copies of the same ACP. In all other eukaryotic CIs structurally studied so far, the two AB1 subunits are encoded by separate paralogs [89,92,93,96,98] that differ in sequence [16,19,31,41,42]. This suggests that the two copies of AB1 in LECA CI were paralogs encoded by separate genes, and that in the metazoan lineage lost ACP diversity.

Mixed-lineage subunits

Subunits NUXM, NDUFB5 (B5), the ferredoxin subunit (CI-FDX), and those in the γ-CA domain are not found in every eukaryotic CI structure, but are widely present in many different eukaryotic species (Fig. 11 and Table 2). This presents two possibilities: 1) these subunits were components of LECA CI that were lost in some lineages, or 2) they were acquired early after the split from LECA at the root of specific lineages. Given the sparse sampling of structures (Fig. 4), especially for clades within the Opimoda+, structural analysis does not allow to robustly differentiate between the above options at this point. Nevertheless, we develop hypotheses regarding the likelihood that these mixed lineage subunits were present in LECA CI.

Figure 11.

Mixed-lineage accessory subunits in eukaryotic CI. Mixed-lineage subunits were chosen as representative structures. γ -CA, CI-FDX, and NUXM from V. radiata (PDB ID: 8E73); B5 from Y. lipolytica (PDB ID: 6YJ4). Core subunits, except for ND6 (from V. radiata), are from P. denitrificans in this representation.

NUXM.

NUXM is found in all eukaryotic CI structures except [33–39]. The presence (or absence) of the NUXM subunit is related to the truncations in the ND2 antiporter subunit (discussed above). TMHs 1 and 2 in ND2 form the binding interface with NUXM (Fig. 7). In species that lack TMHs 1–3 in ND2 (D. melanogaster and B. taurus) [33,36], NUXM is absent. In species with TMHs 1–3 in ND2 (Y. lipolytica [31], V. radiata [41], T. thermophila [16], and P. parva [42]), NUXM is always observed. This pattern is a clear example of what would be expected for a subunit that was present in LECA and lost in a specific eukaryotic lineage —in this case the metazoans (likely specifically bilaterians, see ND2 discussion above).

This simple picture is complicated by the E. gracilis CI structure [19]. In E. gracilis, EG10, a Euglena-specific subunit, is bound to ND2 TMHs 1 and 2 instead of NUXM [19]. Whereas NUXM is a two TMH membrane protein with both termini in the IMS, EG10 is a single TMH protein with a matrix domain consisting of a bundle of short, solvent-exposed helices and loops. Three possible explanations for this observation are: 1) NUXM was not present in LECA and lineages independently acquired a NUXM-like subunit, which was subsequently lost in the bilaterians; 2) NUXM was present in LECA, independently lost along the E. gracilis and bilaterian lineages, then subsequently replaced with a new subunit EG10 along the E. gracilis lineage; or 3) in E. gracilis, NUXM underwent significant structural changes in which TMH2 lost its hydrophobic nature and became a matrix domain, thus the EG10 gene evolved directly from an ancestral NUXM gene. Parsimony suggests that scenario 3 is most likely. Thus, the absence of NUXM may be restricted to a subset of the Opisthokonts correlated with the loss of N-terminal ND2 TMHs (Fig. 7). If correct, NUXM should be considered a LECA CI subunit, but additional structures from sister clades of E. gracilis and non-bilaterian metazoans are needed to fully confirm this hypothesis.

NDUFB5.

Located on the side of the membrane arm opposite the ND5 lateral helix, NDUFB5 (B5) is a bridging subunit connecting the P_P-module with the P_D-module via a long α-helix that spans the membrane arm on the IMS side (Fig. 12) [4,101]. In Opisthokonta, B5 is characterized by a single TMH near its N-terminus followed by its long IMS α-helix (Fig. 12). Protein sequence searches suggested that a gene product in A. thaliana was a B5 homolog [27], but comparison with recently resolved structures confirmed that the sequence was for subunit P9, a Chloroplastida specific subunit that binds CI on the opposite side of the membrane arm relative to B5 (see discussion below) [43]. However, structural studies from P. parva, V. radiata, A. thaliana, and Brassica oleracea identify subunit P1 (also named NUOP5 in P. parva) as a structural homolog of B5 (Fig. 12) [40–43]. Specifically, in Chloroplastida CI structures[15,42], P1 is a long IMS α-helix analogous to B5, but without the N-terminal TMH. A structural homolog of B5 is absent in T. thermophila CI [16–18], leading to the proposal that B5 was an Opisthokonta-specific subunit [27,102]. However, the recent structure of E. gracilis CI has shown that B5 also exists in Discoba (Fig. 12) [19] and thus Diphoda+ (Fig. 4), raising the possibility that B5 is ancestral to LECA CI and has been truncated or lost in the different Diphoda+ lineages. This hypothesis posits that P1 and B5 are both descended from a common LECA B5 ancestor, the subunit was truncated along the Chloroplastida lineage, losing the TMH, and was lost completely in the Alveolata lineage. Additional structures of CI from diverse Archaeplastida, SAR, and other Diphoda+ clades are needed to fully confirm this hypothesis.

Figure 12.

Surface representation of B5 and P1 subunits (Chloroplastida) on overall CI structures (grey cartoons).

Gamma Carbonic Anhydrase (γ-CA).

The γ-CA domain is attached to the matrix side of the P_P-module and is made up of 3 distinct subunits (CA1, CA2, and CA2L) [103]. The CA family of proteins catalyzes the reversible hydration of CO₂ to $\text{HC}{\text{O}}_3^{-}$ and is important for cellular functions [104,105]. For example, in plants, the CA domain is related to CO₂ recycling [106], plant embryogenesis [107], CI expression level [108], and pH regulation [109]. Bacterial γ-CAs are homotrimers, and each monomer has a canonical left-handed parallel three-sheet β-helix fold and catalytically active Zn²⁺ ions bound between the monomers, usually by His residues whose conservation varies across species [75]. However, γ-CA bound to CI is a heterotrimer with structural differences, including in the active sites, with several missing the catalytic Zn²⁺ (Fig. 3 and 11). The γ-CA subunits in CIs that retain the catalytic Zn²⁺ (V. radiata [41], B. oleracea [40], and A. thaliana [42]) have the potential to bind bicarbonate [106], but their enzymatic activity has not been demonstrated [19,101]. Subunits structurally homologous to γ-CA are present in four of the seven eukaryotic species highlighted in this review: P. parva [42], V. radiata [41], T. thermophila [16–18], and E. gracilis [19]. γ-CA is also suggested to be a subunit of CI in diatom Phaeodactylum tricornutum based on 2D Blue-native page gels [75], and mass spectrometric characterization suggests homologues of γ-CAs are also present in CI of amoeboid Acanthamoeba castellanii [76]. Since these organisms are from both Diphoda+ and Opimoda+ clades, it has been persuasively argued that LECA CI had a γ-CA domain [75,110]. Currently, all Opimoda+ CI structures are limited to those from Opisthokonta and all lack the γ-CA domain [31–39]. Thus, more structural data from non-Opisthokont Opimoda+ clades are needed to fully establish the γ-CA domain as ancestral to LECA CI.

Ferredoxin subunit (CI-FDX).

CI-FDX is located on the P_P-module, it is required for CI assembly [111] and plays a role in regulating the CI activity of plants [42]. The general architecture of the FDX domain is conserved across P. parva [42], V. radiata [41], T. thermophila [16–18], and E. gracilis [19]. Although structural alignments show that the CI-FDX is structurally homologous between species, sequence alignments highlight differences between organisms. Cys ligands required to coordinate 2Fe-2S clusters (or including other common ligands for 2Fe-2S clusters such as Asp, His, and Arg) [112,113] are absent in V. radiata [41], P. parva [42], and E. gracilis [19]. In contrast, the CI-FDX in T. thermophila CI contains a 2Fe-2S cluster in the active site [16]. The loss of the 2Fe-2S cluster binding site in plants and algae implies the loss of the enzymatic activity of the CI-FDX [16]. Sequences of the 2Fe-2S binding site vary across species, but the presence of a FeS cluster coordinating ligands does not guarantee the presence of a cofactor. In CI-FDX from A. thaliana’s CI, three of four Cys residues in the FeS-binding site are conserved, but the coordinating metal is a single metal ion instead of a typical 2Fe-2S cluster [42]. These observations suggest that the active site of CI-FDX, like that of γ-CA, is not necessarily conserved.

CI-FDX is present in all current Diphoda+ CI structures, and it sits above ND2, adjacent to the γ-CA domain. There are direct protein-protein interactions between CI-FDX and the γ-CA domain, suggesting that the γ-CA domain may be needed for CI-FDX association. This hypothesis is supported by the lack of CI-FDX in Opisthokont CI structures, which also lack the γ-CA domain. If, as has been argued [75,110], the γ-CA domain was present in LECA CI, it follows that CI-FDX was also present in LECA CI (Fig. 11). Another possibility is that CI-FDX associated with the complex after LECA in an early ancestor of the Diphoda+. More structural data from Opimoda+ clades outside of Opisthokonta are needed to distinguish between these two possibilities.

Lineage-specific subunits

In addition to accessory subunits that are either universally conserved amongst eukaryotes or shared amongst different lineages, some are uniquely found within a lineage of eukaryotes. Here, we outline the lineage-specific subunits observed from CI structures in Opisthokonta, Chloroplastida, Alveolata, and Discoba (Fig. 13 and Table 1).

Figure 13.

Lineage-specific subunits present in eukaryotic CI, shown as surface representations. For Y. lipolytica (PDB ID: 6YJ4), ST1 (PDB ID: 6YJ5) overlayed with the CI structure based on ref. [31]. For T. thermophila, the CI structure from PDB ID: 8GYM was used for the image.

Yarrowia specific subunit ST1.

In Y. lipolytica CI, a catalytically active sulfur transferase, ST1, binds to the Q-module of CI at sub-stoichiometric ratios [31,114] (Fig. 13). However, ST1 was not observed in the fungus C. thermophilum CI structure [32]. Whether ST1 is a fungal-specific accessory subunit or if its association with CI is specific to a specific fungal clade remains to be determined.

Metazoan specific subunits.

Phylogenetic analyses suggested that several CI subunits (NDUFA10, NDUFB1, NDUFB6, and NDUFC1) are of metazoan origin [115]. However, B6 structural homologues are present in Chloroplastida (V. radiata [41]), Discoba (E. gracilis [19]), and Alveolata (T. thermophila [16]), confirming that identification of lineage-specific subunits through sequence homology alone is insufficient. Structural comparisons identify the following four metazoan-specific subunits: A10, B1, C1, and NDUFV3 (V3; Fig. 13).

A10 is found directly above the matrix-exposed surface of ND2, where γ-CA is found in non-Opisthokonta lineages. In Y. lipolytica CI [31], as well as in C. thermophilum CI [32], the binding site of the γ-CA and A10 is empty (Fig. 13). This is consistent with the loss of the γ-CA domain in a common ancestor of Opisthokonta and the specific acquisition of A10 only along the metazoan lineage (Fig. 3). As all structures of metazoan CI to date are from the bilaterian lineage and A10 is homologous to the nucleoside kinase family of proteins, which are ubiquitous, without structural confirmation it is not clear if A10 is present in the sister clades of bilaterians within metazoans, i.e. Cnidaria. In mammals, A10 is known to bind dGTP [116], is required for CI assembly [8], and has been hypothesized to regulate CI activity [4]. Interestingly, all CI structures that have A10 also lack the first three TMHs of ND2 and accessory subunit NUXM (Fig. 7) [33–39]. The N-terminal portion of A10 dips into the membrane where these ND2 and NUXM structural elements are found in other CI structures. This correlation predicts that Cnidarian CI, which has not lost the ND2 TMHs [71], would retain NUXM and not have A10. However, a structure of a Cnidarian CI is needed to confirm this hypothesis.

C1 is a single TMH subunit interacting with NDUFC2 (C2), a conserved accessory subunit composed of two TMHs that bind to ND2 and A10 (Figs. 10 and 13). Thus far, C1 is observed only in mammals [33–35,38,39]. This is consistent with the gene identification results [115], where C1 is classified as vertebrate-specific. C1 has been hypothesized to influence CI assembly by recruiting a structural cardiolipin molecule adjacent to C2 and ND2 TMH 11 [4]. ND2 TMH 11 is shorter in mammalian CI [33–35,38,39] compared to D. melanogaster CI [36,37]. Both CIs have A10, but D. melanogaster CI does not have C1 [36,37], suggesting that C1 recruitment to the complex went to fixation in vertebrates by specific destabilizing changes in ND2 TMH 11. B1 is also a single TMH adjacent to TMH 1 of the core ND4 subunit (Fig. 13). It is hypothesized to stabilize structural lipids [4]. From gene analyses, B1 is classified as a bilaterian-specific protein [115] and is important for CI assembly [8].

The V3 subunit is a helical protein wedged between the V1 and V2 core subunits at the tip of the peripheral arm (Fig. 13). V3 is present in mammals in different isoforms depending on the tissue [117,118]. In D. melanogaster, V3 has been suggested to resemble a fragment of fat cadherin (Ft4) [117], a protein that has be proposed to regulate CI activity [119]. Structures of D. melanogaster CI show partial occupancy for a subunit bound in the equivalent position of V3, but the resolution was insufficient to identify the subunit from the map [36]. Interestingly, structural comparison between CI from B. taurus [33], D. melanogaster [36], and P. denitrificans CI [10] shows that V3 is situated in the same location as the two C-terminal α-helices of core subunit V2 in P. denitrificans CI (Fig. 14). These two α-helices are missing from all eukaryotic CI structures [15,16,19,31–42] and the C-terminus of V2 extends along the surface of V1 instead of folding back into this groove (Fig. 14). This suggests that V2’s C-terminal structure was altered before LECA and that the open cleft left by the missing helices was co-opted in metazoans to bind V3.

Figure 14.

V3 occupies a groove occupied by V2 C-terminus in P. denitrificans. Cartoon and surface representations are shown for V1 (blue zircon) and V3 (purple) with surface representations of the adjacent V2 subunit (mist blue) for (A) P. denitrificans, (B) Y. lipolytica, (C) D. melanogaster, (D) B. taurus, (E) P. parva, (F) V. radiata, (G) T. thermophila, and (H) E. gracilis. For P. denitrificans (A), residues 196–236 that occupy the same site as the V3 subunit in B. taurus and D. melanogaster are highlighted in yellow.

Subunits unique to photosynthetic organisms: P-family proteins

P-family of proteins are subunits observed only in Chloroplastida CI to date [15,40–43]. Based on currently resolved CI structures, these proteins are P2, P4 (NUOP4), P7 (NUOP7), P9, and P10. As mentioned above, P1 is likely a homolog of B5 (Fig. 12), and P8 likely evolved from a lineage-specific splitting of ND5 in Chlorophyta (Fig. 8B). Within Chloroplastida, land plants are Embryophyta, a clade within Streptophyta, and the Chlamydomonadales green algae are within the Chlorophyta clade (Table 2)[120]. Streptophyta and Chlorophyta diverged early after the divergence of Chloroplastida within the Archaeplastida supergroup (Fig. 4) [121–124]. Genetic analysis confirms that Penium margaritaceum, a charophyte green alga within the Zygnematophyceae clade of Streptophyta, represents the closest sister lineage of land plants [124]. Thus, land plants and Chlamydomonadales are distant sister lineages that developed in parallel.

Subunit P9 is observed in all Chloroplastida CI to date: V. radiata (labeled P4 in the PDB structure) [41], A. thaliana [42,43], P. parva [42], and C. reinhardtii [15]. It is composed of a single TMH and is located adjacent to the Q-entry tunnel near core subunits ND4L and ND6 (Fig. 13). Thus far, structural homologs of P9 have not been observed in any other lineages outside Chloroplastida. Therefore, P9 is a Chloroplastida-specific subunit that emerged early in Chloroplastida before the divergence of Streptophyta and Chlorophyta. Additional structures from sister clades of Chloroplastida within Archaeplastida, such as Rhodophyta, are needed to confirm this hypothesis.

P2 is a single TMH subunit adjacent to the NUXM subunit. Unlike P9, P2 is not observed in all Chloroplastida lineages. Among the available structures, it is only observed in the land plants V. radiata [41], B. oleracea [40], and A. thaliana [42,43], suggesting it associated with CI after the divergence of Streptophyta and Chlorophyta (Table 1), indicating that it is a Streptophyta-specific subunit. Its presence in CI from other Streptophyta clades outside of Embryophyta, such as Charophyta, will need to be determined.

Chlorophyte-specific subunits are P4, P7, and P10, which are to date only observed in CI structures from C. reinhardtii [15] and P. parva [42] (Fig. 13). P4 is located on the lateral-helix side of ND5 and ND4 and contains a single TMH with the extended N-terminus on the matrix side of the membrane arm. P7 and P10 are both single TMH subunits present at the tip of the P_D-module bound to ND5 adjacent to TMH 14 and 15 (Fig. 13). P7 forms a contact site between CI and CIII₂ in C. reinhardtii [15], suggesting its role in the formation of supercomplexes.

Alveolata and Discobid-specific subunits

To date, CI from E. gracilis [19] and T. thermophila [16–18] have the most lineage-specific accessory subunits. T. thermophila has the most unique accessory subunits in all eukaryotic CI structures known to date, with a total of 21 lineage-specific subunits (Fig. 13); of these, 18 are part of the membrane arm [17,18] (Figs. 1 and 13). Three subunits are structurally homologous to lipid-A-disaccharide synthases (T1) and acyl-CoA synthetase/AMP-acid ligase II (T2 and T15), whose function has not yet been established [16,17]. In E. gracilis, 14 unique accessory subunits are evenly distributed in both the membrane and peripheral arms (Fig. 13), two of whose identities are yet to be assigned [19]. Of these, subunits E2, E3, and E5 in E. gracilis are homologous to proteins involved in fatty acid synthesis, whose functional activity in CI has not yet been established [19]. As we only have single structures from Aveolata and Euglenida, which are our sole examples of CI from the SAR and Discoba supergroups, no strong conclusions can be made regarding the order of accretion of these accessory subunits. More structures from these, as well as additional supergroups, are needed to fully map the evolution and biodiversity of CI in eukaryotes.

Discussion

In this review, we describe the variation of subunits across CI structures (Fig. 1) and hypothesize the evolutionary relationship between subunits seen in different eukaryotic lineages (Fig. 5). Accessory subunits are observed in prokaryotic CI (Fig. 2); however, these accessory subunits evolved independently along distinct lineages and only the α-proteobacterial lineage shares any accessory subunits with eukaryotic CI. Thus, FECA CI was probably α-proteobacteria-like with four accessory subunits: A9, A12, S4, and S6 (Fig. 15). The available structures of mitochondrial complexes are currently biased towards two major supergroups of eukaryotes: Opisthokonta and Chloroplastida (Fig. 4, Table 2). This bias stems from the primacy of medical research, well established model organisms from these clades and the availability of established protocols for mitochondrial isolation at scale. However, advances in cryoEM have made structural work with much less abundant samples possible, opening new sources of mitochondria for structural characterization and facilitating the exploration of CI structural biodiversity across eukaryotes.

Figure 15.

Proposed structure of FECA CI (top) and LECA CI (bottom). Shown are representative subunits from CI models of P. denitrificans, V. radiata, and Y. lipolytica. Subunits used to model FECA and LECA CI are indicated in Table 3. The estimated time between FECA and LECA are indicated, based on ref. [125].

What we have learned thus far is that core subunits are generally conserved across species, with some structural variation usually at the periphery. Common differences observed involve extensions or truncations at termini as well as gene splitting. The core subunit most seen to vary is the antiporter-like membrane subunit ND2 (Fig. 7), suggesting that it is under weaker selection pressure than other core subunits. For accessory subunits we show that the current structures can provide us with a model of LECA CI (Fig. 15 and Table 3), composed of the 14 core subunits (ND1, ND2, ND3, ND4, ND4L, ND5, ND6, S1, S2, S3, S7, S8, V1, V2) and 32 accessory subunits (A9, A12, S4, S6, A1, A3, A8, A11, A13, C2, S5, B2, B3, B4, B6, B7, B8, B9, B10, B11, A2, A5, A6, A7, two copies of AB1, B5, NUXM, the three γ-CA subunits, and CI-FDX). This suggests that between FECA and LECA, the accessory subunit count of CI increased by 28 proteins (Fig. 15). Throughout this review we propose key lineages that need to be structurally characterized to address open questions regarding LECA CI and lineage specific subunit additions. Importantly, CI structures are needed from non-Opisthokont clades in Opimoda+ to further constrain the subunit composition of LECA CI.

Table 3.

Categories of accessory subunits

Paralog of interaction domain, carrier, putative assembly factor or import machinery	Paralog of enzyme	Single TMH Protein (>100 a.a.)	SEP (at least one homolog <100 a.a.)	Single TMH & SEP	Exceptions
A2 and TX (thioredoxin)* A11 (Tim23) A6/AB1-α (LYRM/ACP) B9/AB1-β (LYRM/ACP) S6 (IscU) CI-FDX (ferredoxin) EG6 and J1 (DnaJ domain) EG11 (COX17) EG12 (ACP) PH1 and PH2 (PH domain) TT6 (Nuclear receptor ligand binding domain)	γ-CA (carbonic anhydrase trimer) A9 and EG4 (short-chain dehydrogenase/ reductase) A10 (Nucleoside kinase) EG1 (Adenylate Forming Domain) EG2 (Phosphoenolpyruvate carboxykinase EG3 and EG5 (Medium-chain dehydrogenase/reductase) EG7 (Phospholipase A2) TT1 (Lipid-A-disaccharide synthase) TT2 and TT15 (Acyl-CoA synthetase) TT3 (RNase III domain) ST1 (Sulfur transferase)	A13 P7 P2 EG8 EG10 EG13 TT8 TT10	A7 A8 B7 B10 C2 S5 TT14	A1 A3 B1 B2 B3 B4 B5 [P1]# B6 B8 B11 C1 P4 P9 P10 TT11 TT12 TT13	A12 (small folded domain) S4 (small folded domain) A5 (3 helix bundle) NUXM (2 TMHs) V3 (short/long isoforms) TT4 (2 TMHs) TT5 (2 TMHs) TT6 (3 helix bundle) TT7 (2 TMHs) TT9 (4 helix bundle) TT16 (2 TMHs) TT17 (2 TMHs)
Accessory subunits proposed here to be present in LECA
A2 (Vr) A11 (Vr) A6/AB1-α (Yl) B9/AB1-β (Yl) S6 (Vr) CI-FDX (Vr)	γ-CA (Vr) A9 (Vr)	A13 (Yl)	A7(Vr) A8 (Vr) B7 (Vr) B10 (Yl) C2 (Vr) S5 (Vr)	A1 (Vr) A3 (Yl) B2 (Yl) B3 (Yl) B4 (Vr) B5 (Yl) B6 (Vr) B8 (Vr) B11 (Vr)	A12 (Pd) S4 (Pd) A5 (Vr) NUXM (Vr)

^*Top the parentheses indicate the homologous domain or enzyme or provide some context for the exceptions, for subunits present in LECA (below) the parentheses indicate the source organism use to build the LECA complex I model shown in Figure 15. Pd represents P. denitrificans, Vr represents Vigna radiata, Yl represents Yarrowia lipolytica.

^#We argue that B5 and P1 are paralogues. See Figure 12.

Recently, LECA was proposed to be a heterotrophic, flagellated excavate that made its living by eating bacteria and other cells [13]. This metabolic strategy would depend heavily on OXPHOS and hence CI activity for ATP production, maintaining strong selection pressure for CI. This suggests that any protein that became important for CI stability and function would rapidly go to fixation in the population. Although this does not rule out possible selective advantages provided by the addition of accessory subunits, it does evoke a suitable scenario for CNE [55,56]. Given recent estimates of ~240–740 million years between FECA and LECA (Fig. 15) [125], we can define a “CI-clock” for CNE, in which a new subunit was added to CI every ~10–30 million years (assuming the γ-CA trimer was added in a single event). This CI-clock can be used to constrain quantitative theories of CNE.

In general, the accessory subunits fall into three categories: 1) proteins that had pre-existing functions in protein assembly or maintenance that have subsequently gotten stuck to the complex, likely after gene duplication followed by neofunctionalization (e.g. Nqo15 in T. thermophilus or NDUJ1 in T. thermophila); 2) larger soluble proteins homologous to mitochondrial enzymes, that were already likely present at relatively high concentrations, associating with CI and getting stuck, again likely after gene duplication and neofunctionalization; and 3) small hydrophobic proteins, generally with a single TMH (Table 3). These small open reading frame (sORF)-encoded peptides (SEPs) have recently been shown to be abundantly expressed by mammals with a disproportionate amount targeted to the mitochondria [126]. Some SEPs have become essential for the assembly and stability of the respiratory complexes [127]. Interestingly, many of the most recently evolved lineage-specific subunits, such as B1, C1, P2, P4, P7, P9, P10, T8, T10, T11, T12, T13, E10, and E13, are SEPs (Fig. 13; Table 3), suggesting that this is a relatively rapid path for CNE to add subunits. In this scenario, non-functional genetic elements that encode for hydrophobic SEPs surreptitiously become associated with essential metabolic complexes and, through mutation in the complexes, become indispensable for their assembly and stability, driving the SEP genes to fixation. Thus, many of the accessory subunits of CI can be viewed as parasitic genetic elements, not providing any selective advantage, but through random mutation becoming indispensable for CI function.

Understanding CI structural biodiversity is a window into the mechanisms of molecular evolution, its dependence on contingency, and the balance between the evolution of complexity through selective pressures vs. random processes. We are just at the beginning of mapping the molecular biodiversity of macromolecular complexes across organisms, and what we will learn will undoubtedly revolutionize our understanding of biological complexity.

List of abbreviations:

ACP

Acyl carrier protein

ATP

Adenosine triphosphate

CI

Complex I

CII

Complex II

CIV

Complex IV

CIII₂

Complex III dimer

CNE

Constructive neutral evolution

CoQ

Coenzyme Q

FDX

Ferredoxin

FECA

First eukaryotic common ancestor

γCA

Gamma carbonic anhydrase

IMM

Inner mitochondrial membrane

IMS

Intermembrane space

LECA

Last eukaryotic common ancestor

LYRM

Leucine-tyrosine-arginine motif

PA

Complex I peripheral arm

MA

Complex I membrane arm

NADH

Reduced form of nicotinamide-adenine dinucleotide

P_P

Proximal proton-pumping module of the membrane arm

P_D

Distal proton-pumping module of the membrane arm

SEP

Small ORF-encoded peptides

sORF

Small open reading frame

TMH

Transmembrane helix