Genome Analysis of Structure–Function Relationships in Respiratory Complex I, an Ancient Bioenergetic Enzyme

Abstract

Respiratory complex I (NADH:ubiquinone oxidoreductase) is a ubiquitous bioenergetic enzyme formed by over 40 subunits in eukaryotes and a minimum of 11 subunits in bacteria. Recently, crystal structures have greatly advanced our knowledge of complex I but have not clarified the details of its reaction with ubiquinone (Q). This reaction is essential for bioenergy production and takes place in a large cavity embedded within a conserved module that is homologous to the catalytic core of Ni–Fe hydrogenases. However, how a hydrogenase core has evolved into the protonmotive Q reductase module of complex I has remained unclear. This work has exploited the abundant genomic information that is currently available to deduce structure–function relationships in complex I that indicate the evolutionary steps of Q reactivity and its adaptation to natural Q substrates. The results provide answers to fundamental questions regarding various aspects of complex I reaction with Q and help re-defining the old concept that this reaction may involve two Q or inhibitor sites. The re-definition leads to a simplified classification of the plethora of complex I inhibitors while throwing a new light on the evolution of the enzyme function.

Introduction

Respiratory complex I is the common name of a large enzyme complex that transports electrons from NADH to ubiquinone (Q), NADH:ubiquinone oxidoreductase (Nuo), EC. 1.6.5.3. Contrary to other NADH dehydrogenases, complex I couples the transport of electrons to the generation of membrane potential by pumping two protons per electron across the membrane in which the enzyme complex resides (either the bacterial plasma membrane or the inner membrane of mitochondria; Brandt 2006; Sazanov 2015). Complex I can also catalyze NAD+ reduction from reduced Q, QH₂ (Gutman et al 1970; Dupuis et al 1998; Grivennikova et al 2003), consuming membrane potential to drive the energetically unfavorable reaction (the redox midpoint potential of NADH being −0.34 V, whereas that of Q/QH₂ normally around +0.1 V; Brandt 2006; Sazanov 2015).

The recent three-dimensional (3D) structural information produced by Sazanov and co-workers on bacterial complex I (Sazanov and Hinchcliffe 2006; Baradaran et al. 2013; Sazanov 2015) and by others on mitochondrial complex I (Vinothkumar et al. 2014; Zickermann et al. 2015) has confirmed previously noted homologies with bacterial enzymes such as hydrogenases (Fearnley and Walker 1992; Oh and Bowien 1998; Baumer et al. 2000; Friedrich and Scheide2000), reinforcing the concept that complex I has a modular design (Moparthi and Hägerhäll 2011; Sazanov 2015). Three physically distinct modules are named after the enzyme substrates (N for NADH and Q for quinone) and the proton pumping function in the membrane (P for proton) (Moparthi and Hägerhäll 2011; Marreiros et al. 2013; Sazanov 2015). The evolution of complex I is thought to have followed a progressive assembly of these modules derived from diverse enzyme systems, with the combination of the Q and P module resembling the structure of membrane-bound Ni–Fe hydrogenases (Friedrich and Scheide 2000; Hedderich 2004; Efremov and Sazanov 2012; Marreiros et al. 2013). The subsequent acquisition of the N module would have produced the minimal enzyme structure conforming to mitochondrial complex I, which in bacteria is coded by nuo operons of 13 or 14 subunits (Dupuis et al. 1998; Friedrich and Scheide 2000; Moparthi and Hägerhäll 2011; Sazanov 2015; Spero et al. 2015). In eukaryotes, many accessory subunits have been added to facilitate the assembly and regulation of mitochondrial complex I (Fearnley and Walker 1992; Brandt 2006; Yip et al. 2011; Vinothkumar et al. 2014; Zickermann et al. 2015). However, the redox and bioenergetic function is essentially the same in the bacterial and mitochondrial enzyme complexes (Yagi et al. 1998).

Even if bacterial Nuo complexes reflect the core structure of mitochondrial complex I (Friedrich and Scheide 2000; Brandt 2006; Sazanov 2015), the 3D information that is currently available shows a remarkable difference in the binding pocket for Q and its antagonist inhibitors (Sazanov 2015). The crystal structure of the complex from Thermus, a thermophilic bacterium of the ancient Deinococcus/Thermus phylum (Segata et al. 2013), has revealed an unusually large reaction chamber with a narrow entrance within the Q module lying at the membrane periphery, in which exogenous Q (decyl-ubiquinone, DBQ) is hydrogen-bonded to conserved H38 and Y87 of the NuoD subunit (Baradaran et al. 2013). These structural features are illustrated in fig. 1A, which highlights key residues of the Q reaction chamber that will be analyzed here. The potent Q antagonist inhibitor piericidin A (Takahashi et al. 1965) is bound in the same way within the Q reaction chamber of Thermus (Baradaran et al. 2013; Sazanov 2015) (fig. 1A, right panel). However, the structure of mitochondrial complex I from the fungus Yarrowia shows a narrower Q binding pocket, in which the Q antagonist inhibitor DQA (2-decyl-4-quinazolinyl amine, previously known as SAN 547; Hollingworth et al. 1994; Degli Esposti 1998; Okun et al. 1999) is hydrogen-bonded to a histidine residue of the 49-kDa subunit that corresponds to H38 in the NuoD of Thermus, but is far away from the tyrosine residue equivalent to Thermus Y87, as sketched in fig. 1A, right panel (Zickermann et al. 2015). The possibility that this binding difference is owing to the so-called deactive state in which the Yarrowia complex is isolated (Zickermann et al. 2015 cf. Galkin et al. 2008) apparently contradicts biochemical results showing very strong resistance toward DQA inhibition in Yarrowia and E. coli complex I after mutation of the residues equivalent to Thermus Y87 (Kashani-Poor et al. 2001; Tocilescu et al. 2010b; Sinha et al. 2015).

Fig. 1.—

Structure and evolution of complex I. (A) 3D image of Q-interacting subunits and residues in complex I. The ribbon model of the subunits forming the Q reaction chamber in Thermus complex I (top panel on the left, cf. Baradaran et al. 2013) is zoomed in and slightly tilted to the left in the bottom panel. It is then enlarged in the right panel, in which the overall structures of NuoD and NuoB were removed, while rendering in cylinder mode most residues that interact with Q or form its reaction chamber (Sazanov 2015). The residues belonging to the NuoD subunit are colored in blue or brown, whereas those of NuoB and NuoH are in red. The position of the quinone head of bound DBQ is indicated by the orange hexagon, with the dashed lines indicating the approximate position of the hydrogen bonds between the quinone carbonyls and Y87 and H38 (Baradaran et al. 2013). The dashed oval indicates instead the approximate position of the Q antagonist DQA bound to Yarrowia complex I (Zickermann et al. 2015). The brown hexagons on top of residues D401 and M85, also colored in brown, indicate the involvement of such residues in natural or induced resistance to the Q antagonist piericidin A (table 3), even if they were not listed among the Q-interacting amino acids in Thermus complex I (Baradaran et al. 2013). At the bottom of the right panel, A63 is shown; it frames the entrance into the reaction chamber from the wormhole through which Q penetrates the complex from the membrane (Baradaran et al. 2013). (B) Phylogenetically wide tree of the NuoD/Nad7/49 kDa subunit. The neighbor-joining (NJ) tree was initially obtained using the deltaBLAST program (Boratyn et al. 2012; Degli Esposti et al. 2015) using NuoD of Nitrospira defluvii (accession number: CBK40385) as the query and was extended to 5,000 sequences from bacteria and archaea possessing simplified Nuo operons. Note that after the early, well-separated branching groups of sulfur-reducing anaerobes (such as Nautiliales) and mbx-like complexes (which appear to be sister groups in trees of other Nuo subunits), as well as that of Fpo complexes, the proteins from other bacterial phyla are not well-resolved from each other, thus forming a semi-comb pattern of clades (Bapteste et al. 2008). Only the central group of alpha, beta, and gamma proteobacteria, corresponding to clade A and B in Spero et al. (2015) and encompassing also the eukaryotic homologs, is resolved as an independent branch. Nevertheless, the NuoD of Chlorobi, some gram-positive bacteria such as Bacillus cereus and Clostridium pasteurianum, as well as of plastidial NDH-1 group together in an early diverging branch containing Q-reacting NUO11 complexes. The tree is rooted on group 3 hydrogenases from E. coli (Vignais and Billoud 2007), while ech hydrogenases form the immediate precursor branch for complex I clades.

The simplest explanation for the different binding of Q and its antagonists in the structure of bacterial (Thermus) and mitochondrial (Yarrowia) complex I (cf. fig. 1A) would be that the Q module can simultaneously bind two molecules of Q or antagonists such as DQA and related quinazoline inhibitors (Degli Esposti et al. 1999; Murai et al. 2009; Verkhovsky et al. 2012). This possibility would be consistent with a wealth of biochemical data and with the structural evidence that the Q reaction chamber in Thermus complex I is wide enough to accommodate two short-chain Q homologs such as Q-1 or Q-2 (Baradaran et al. 2013). These quinones retain functional activity after mutation of Y87 (Tocilescu et al. 2010b; Sinha et al. 2015) and consequently can bind to another side of the Q chamber. The same chamber could equally accommodate two molecules of the Q antagonist piericidin, which has a structure very similar to Q-2 (Takahashi et al. 1965; fig. 6)— even if the conditions of crystallization may allow saturation only of the tighter of the two sites in Thermus (Baradaran et al. 2013). For a long time, piericidin has been known to bind to two sites in mammalian and fungal complex I (Gutman et al. 1970; Coles et al. 1974; Singer and Ramsay 1994) and to potently inhibit also bacterial complex I (Friedrich et al. 1994; Dupuis et al. 1998). This evidence suggested that mitochondrial complex I may have two Q reaction sites (Friedrich et al 1994; Degli Esposti et al. 1996; Magnitsky et al. 2002; Ohshima et al. 1998; Verkhovsky et al. 2012), a concept that has been subsequently weakened by mutagenesis (Brandt 2006; Tocilescu et al. 2010b) and structural (Baradaran et al. 2013) studies.

Fig. 6.—

Mapping Q-antagonist inhibitors around the Q reaction chamber of complex I. The residues associated with resistance toward naturally occurring and quinazoline inhibitors of complex I (table 3) are mapped on the model representing the Q-reacting subunits (fig. 1A), modified from that presented in fig. 5 in the overlapping of M51 of the NuoB subunit over the NuoD subunit to introduce two additional residues of the latter protein, M404 and G39. Although not listed among the Q-interacting sites by Baradaran et al (2013), these residues form part of the Q cavity in the crystal structure of Thermus complex I (see right panel) and are associated with resistance to Q antagonist inhibitors (table 3). Moreover, M404 has been found crucial for DBQ activity (table 3, cf. Tocilescu et al. 2007; Fendel et al. 2008). Graphical symbols indicate inhibitor resistance—either natural or experimentally induced by mutagenesis—as follows: brown hexagons, piericidin A as in fig. 1A; green squares, rotenoids; light green square, deduced rotenone resistance in E. coli (Friedrich et al. 1994); red squares, DQA and other quinazoline inhibitors without cross-resistance to rotenone (cf. Fendel et al. 2008); blue triangle, myxococcal antibiotics (e.g., myxothiazol and stigmatellin); orange inverted triangles, capsaicin and other type C inhibitors such as C12E8 (cf. Tocilescu et al. 2010b); gray star, position binding an acetogenin analog that is displaced by bullatacin and quinazoline inhibitors (Masuya et al. 2014). The sponge-produced mycothiazole is tentatively considered to overlap both rotenone and type C inhibitors (table 3). The residues associated with resistance towardsthe first site of piericidin interaction are highlighted in light brown (left), whereas those that may form the second piericidin site are highlighted in gray (within NuoD only, right). Contrary to this second site, the first piericidin site is separated from the distribution of the residues associated with resistance toward myxococcal antibiotics. Note that the sites associated with rotenone and quinazoline resistance accumulate in the central part of the diagram, overlapping the resistance sites for both piericidin and myxococcal antibiotics. The panel on the right shows a 3D image of the Q reaction chamber of Thermus complex I as presented in fig. 1A, but with a downward tilt to better visualize the position of many residues presented in the scheme on the left. H38 and Y87 binding to Q are colored in red, whereas other residues of the same NuoD subunit are in blue, as in fig. 1A. Conversely, residues involved in inhibitor resistance that belong to the NuoB and NuoH subunit are colored in black and tan, respectively. Blue triangles indicate residues involved in the resistance to myxococcal antibiotics, as in the scheme of the left panel. The pale blue dashed triangle indicates the position of T135, which is not shown for sake of clarity. The chemical structures of piericidin A, DQA, and stigmatellin—a representative of myxococcal antibiotics—are shown in the bottom panel and compared with that of Q-2.

Mutually exclusive binding of complex I inhibitors can now be rationalized on the basis of the unique features of the Q reaction chamber. The structure of both bacterial (Baradaran et al. 2013) and mitochondrial (Zickermann et al. 2015) complex I shows a narrow entry channel for Q coming from the membrane. Hence, the long isoprenoid tail of natural Q substrates locks them within the wormhole-like entry site from the membrane, whereas their hydrophilic head (and initial prenyl groups) acquires freedom of movement upon penetrating the wide reaction chamber formed by the subunits of the Q module. Movements of Q within the chamber are likely to contribute to the conformational changes that couple redox reactions to proton pumping (Sazanov 2015; Sharma et al. 2015). Most antagonist inhibitors mimic the hydrophilic head of Q substrates but match only partially their hydrophobic tail, as they usually are not much longer than Q-2 (Ohshima et al. 1998; Murai et al. 2009, cf. fig. 6). Notably, Q-2 acts a product-like inhibitor in mammalian complex I (Lenaz et al. 1968; Degli Esposti 1998; Ohshima et al. 1998). Hence, multiple Q antagonist molecules may penetrate one at a time the wormhole-like entry in complex I and then accumulate inside the Q-reacting chamber.

The considerations just discussed have stimulated the present work that has followed genomic and evolutionary approaches to deduce crucial structure–function relationships of complex I interaction with quinones. The analysis has been integrated with functional studies to answer the fundamental question of how Q reactivity emerged along evolution. Because crystal structures of complex I show a single Q binding pocket that is enclosed within the NuoD protein at its interface with the NuoB and NuoH subunits (fig. 1A, cf. Sazanov 2015), it can be assumed that Q reactivity evolved primarily within the NuoD subunit. Key amino acids facing its internal cavity (fig. 1A, right panel) that originally contained the Ni–Fe cofactor of hydrogenases would then change in concert with others lying in the regions interconnected with the NuoB and NuoH subunits, together forming the Q reaction chamber (Baradaran et al. 2013). Detailed sequence analysis of these subunits should then reveal the key amino acid signatures that are linked to the acquisition of Q reactivity from homologous proteins of Ni–Fe hydrogenases, which do not react with membrane quinones. Following this rationale, the investigation has been expanded to answer the questions:

1. Which of these signatures correspond to Q-binding residues in the 3D structure of bacterial and mitochondrial complex I?

2. Are natural substitutions of these residues related to different quinone types that are present in some archaea and bacteria?

3. Do organisms that produce complex I inhibitors show substitutions of (or near to) these residues that could account for natural resistance of their complex I toward the same Q antagonist inhibitors?

The results presented below provide answers to these questions while throwing a new light on the evolution of the structure and function of respiratory complex I.

Results and Discussion

A novel pattern for the evolution of bacterial complex I

Phylogenetic trees based on the highly conserved NuoD/49 kDa/Nad7 subunit of complex I (fig. 1B) produced novel information to interpret the possible evolution of Nuo complexes that is of relevance to the acquisition of Q reactivity. Previous models for the evolution of complex I proposed that membrane-bound hydrogenases such as E. coli hyc (Vignais and Billoud 2007) may either be the progenitors of complex I (Fearnley and Walker 1992; Hedderich 2004) or have a common ancestral origin with Nuo complexes (Moparthi and Hägerhäll 2011). Alternatively, subtypes of group 4 hydrogenases that lack canonical Ni-Fe ligands, such as ehr (energy converting hydrogenase-related complexes— Marreiros et al. 2013) and hyq (hydrogenases quinone reacting in alpha proteobacterial diazotrophs; Sprecher et al. 2012), have been proposed as intermediates in the evolution of complex I. By contrast, the present analysis suggests that the large subunit of group 4 hydrogenases of the ech subtype (energy converting hydrogenases, e.g., CO-induced Ni–Fe hydrogenase of Rhodospirillum rubrum; Coppi 2005; Drennan et al. 2007) may be the precursors of NuoD (figs. 1B and 2, cf. Kashani-Poor et al. 2001).

The most ancestral forms of nuo operons are found in organisms that thrive in oceanic hydrothermal vents: members of the Nautiliales order of epsilon proteobacteria (Campbell et al. 2009) and the deep-branching Desulfurobacterium and Thermovibrio of the phylum Aquificae, now grouped in the order of Desulfurobacteriales (Göker et al. 2011; Giovannelli et al. 2012). These strictly anaerobic organisms share the pathways of H₂-mediated sulfur reduction and reverse tricarboxylic acid cycle (Campbell and Cary 2004; Campbell et al. 2009), which is present also in facultatively anaerobic organisms such as Thermotoga and Magnetococcus (Schübbe et al. 2009) that possess a partial nuo operon matching the mbx complex originally found in Pyrococcus.

Pyrococcus mbx complex functions to support ferredoxin-reducing hydrogenases, re-oxidizing reduced ferredoxin to produce NADPH for biosynthetic purposes when elementary sulfur is available as final electron acceptor (Bridger et al. 2011). The gene sequence of such a complex incorporates the entire operon of a pha/mrp antiporter system (cf. fig. 2A) and uses the Fe–S clusters in the homolog of the NuoI subunit to directly reduce NADP+. In the mbx-like operon of Thermotoga and proteobacteria, instead, the NuoI subunit is substituted by a flavoprotein related to gltD (fig. 2B), the small subunit of NAD(P)-dependent glutamate synthase interacting with NAD(P)H (Suzuki et al. 2005). Hence, this subunit functions as the reduction site for the acceptor substrate NAD(P)+, while the mbx-like operon often terminates with rubrerythrin (fig. 2B), a redox protein probably involved in quenching radicals (Sztukowska et al. 2002). The NuoD homologs of mbx and mbx-like complexes show the vestigial presence of Cys residues at positions corresponding to cysteine ligands of the Ni–Fe cluster of hydrogenases (fig. 3), a feature shared with archean Fpo (F420-phenazine oxidoreductase) complexes (Moparthi and Hägerhäll 2011; Lang et al. 2015). The same vestigial ligands are now evident also in the ancestral Nuo complexes of Desulfurobacteriales and Nautiliales (fig. 2B cf. fig. 3A). In particular, their NuoD proteins maintain the first and third Cys ligands of the Ni–Fe cluster, corresponding to positions 86 and 403 following Thermus numeration (figs. 2B and 3A). The gene sequences of ancestral forms of Nuo complexes are shown in fig. 2 and supplementary fig. S2, Supplementary Material online, which summarize information derived from the alignments of various bacterial subunits and their plastidial homologs (fig. 3A and data not shown).

Fig. 2.—

Gene clusters for Nuo complexes and their ancestors. (A) Graphical illustration of the gene clusters for membrane-bound hydrogenases and a typical complete nuo operon for complex I. The nuo operon of Rhodobacter capsulatus has been simplified by removing open reading frames that intermix with the canonical Nuo subunits (Dupuis et al. 1988). The ech type hydrogenase corresponds to the CO-induced enzyme of R. rubrum (Drennan et al. 2007), whereas the gene sequence at the bottom labeled “mtDNA Jakobides” represents the synthenic groups of complex I subunits in the mitochondrial genome of the jakobide Andalucia (Burger et al. 2013). Genes are color-coded as follows: light gray, P-module; dark gray, Q-module; light blue, N-module. The light blue rectangle inside NuoD subunits indicates the loss of ligands for the Ni–Fe cluster of hydrogenases. (B) Possible evolution of nuo gene sequences. The earliest recognizable sequence of NUO11 operons for complex I is seen in Desulfurobacterium (top part) or, when NuoD is fused with NuoC, in Nautilia (bottom part); these bacteria, therefore, have the most ancestral form of the complex, which is similar to that present in MQ-reacting NUO11 of Chlorobium and the PQ-reacting NDH-1 complex of Cyanobacteria such as Synechocystis (which is split in different synthenic groups and additionally has small subunits involved in ferredoxin binding that are not shown here; Ma and Ogawa 2015). The legend on the top right indicates the different symbols used for cluster N2 ligated to two consecutive cysteines in the NuoB subunit (light blue cube, cf. fig. 1A), the homologous high-potential Fe₄S₄ clusters of hydrogenases and ancestral Nuo operons (yellow cube) (table 2), and NuoD proteins retaining the first (Cys86 with Thermus numeration) and third ligand of the Ni–Fe cluster of hydrogenases (fig. 3A and table 1). Other symbols are as follows: pentagons, flavin cofactors; dark stars with Mo inside, Mo-binding domains; gray stars without Mo, lost Mo-binding domains; dark diamonds, F₂S₂ clusters; dark blue cubes, Fe₄S₄ clusters in subunits NuoF, NuoG, and NuoI; gray dashed cube, cluster N7 present in Thermus and other bacteria (Pohl et al. 2007; Sazanov 2015). The known or likely substrates are indicated on the left of each gene sequence. Additional variations in Nuo gene clusters are shown in supplementary fig. S2, Supplementary Material online.

Fig. 3.—

Alignment of NuoD/Nad7/49-kDa subunit sequences of complex I and acquisition of Q reactivity in complex I. Sequences with the accession number identified on the left were first aligned using the COBALT feature of the BLAST program and then manually refined (Degli Esposti et al. 2015). (A) Alignment blocks of the NuoD/49-kDa subunit showing the four conserved regions that contribute to the architecture of the large Q reaction chamber. The four conserved regions are indicated on top of the alignment and the type of complex is identified on the right. The quinone substrates or the Q antagonist inhibitor typical of each organism are listed on the right of the bottom panel, in which the blue rectangle indicates a cut of the alignment in the region not facing the Q cavity (see fig. 4B for the complete alignment of this part of the protein). Note that Anaeromyxobacter is the only genus of the Myxococcales that does not produce fruiting bodies and antimicrobial antibiotics (Huntley et al. 2011). NuoD sequences of other groups of delta proteobacteria, for instance, Bdellovibrio and Geobacter, are much closer to that of Anaeromyxobacter than to those of antibiotic-producing Myxococcales (their various Nuo gene clusters are reported in supplementary fig. S2, Supplementary Material online). In bold are identical or strictly conserved residues (e.g., I,L,M), whereas residues forming the Q chamber are shown in black and white following the numeration of Thermus subunit equivalent to NuoD (Fig. 1A, Baradaran et al. 2013). Unusual substitutions of phylogenetically conserved residues are reported in red with gray background (e.g., N). D160, residue in mature bovine 49-kDa subunit equivalent to D139 in Thermus that is labeled by an azido acetogenin (Masuya et al. 2014). V in red and yellow highlighted, mutation conferring inhibitor resistance in Rhodobacter (Darrouzet et al. 1988); myxothiazol, naturally occurring complex I inhibitor produced by the organism. (B) Alignment of the N-terminal part of NuoB/PSST subunit. The region comprises the two consecutive Cys ligands of Fe-S cluster N2 and most residues that are in the Q reacting chamber of complex I, from either Thermus or Yarrowia (highlighted in black and white). Unusual amino acid substitutions and conserved cysteines are highlighted in yellow. The organisms colored in red possess rhodoquinone. The sequence of Chlorobium refers to that of Chlorobium limicola, which is representative of NUO11 operons of green sulfur bacteria.

Analysis of Nuo proteins to answer Question 1: Which are the early signatures of Q reactivity?

There is a fundamental functional difference that distinguishes mbx from Fpo complexes. While mbx and related enzyme catalyze electron transport from reduced ferredoxins to NAD(P) + (Bridger et al. 2011), Fpo complexes use as their electron acceptor a quinone analog, methanophenazine (Mph—Abken et al. 1998; Welte and Deppenmeier 2014). This is a membrane quinone chemically related to anti-microbial phenazines produced by Pseudomonas (Turner and Messenger 1986) and also to potent acaricide inhibitors of complex I such as fenazaquin (Hollingworth et al. 1994; Degli Esposti 1998). Hence, Fpo complexes should harbor the first set of molecular signatures enabling reactivity to membrane quinones or their antagonists along the evolution of complex I.

Sequence analysis of the NuoD subunit indicates four protein regions that contain residues forming the Q reacting chamber in Thermus and Yarrowia complex I (cf. fig. 1A) and are partially conserved in the homologous subunit of Fpo complexes (highlighted in black and white, fig. 3A). Two of these regions correspond to the ligand domains of the Ni–Fe cluster of hydrogenases and show a similar pattern of evolution. The other two regions contain the Q-binding residues H38 and D139, respectively, and follow a different pattern of amino acid variation (fig. 3A and table 1). Archean Fpo (Welte and Deppenmeier 2014) does have Y87, which is hydrogen-bonded to a carbonyl of Q but not H38 and D139, which are hydrogen-bonded also to each other in Thermus complex I (Baradaran et al. 2013). Among other residues that form part of the Q reacting chamber (fig. 1A), F146 and D401 appear to be already present in the NuoD homologs of Ech hydrogenases (fig. 3), but never together with Y87, as in Fpo complexes (table 1). Of note, substitution of one Cys ligand of the Ni–Fe cluster in R. rubrum Ech hydrogenase enables its reaction with quinols (Heo et al. 2002), thereby indicating a propensity for Q reactivity.

Table 1.

Evolutionary variation of key residues of the NuoD subunit

Note.—Residues not present in ech hydrogenases and acquired with Mph reactivity in Fpo complexes are in white over brown background, whereas those sporadically present in ech or mbx complexes and then acquired together with MQ(H₂) reactivity in Chlorobi are highlighted in light brown (cf. fig. 5). Of note, Chlorobium residues refer to a consensus sequence of five different species of green sulfur bacteria pivoting on C. limicola NUO11 (cf. fig. 3B).

Although there appears to be no study on the reaction of bacterial or mitochondrial complex I with Mph, this compound is certainly able to accept electrons from the complex, as its midpoint redox potential (−165 mV, Welte and Depenmeier 2014) is generally higher than that of cluster N2, which reduces quinones (Sazanov 2015). Moreover, Mph is related to phenazine methosulfate, which has long been used as a redox mediator in NADH oxidation assays (Löw and Vallin 1964). The concomitant presence of Y87, F146, and D401 in the NuoD sequences of Fpo complexes (table 1) would form the minimal reaction domain for Mph and, consequently, constitute the earliest structural signature for the acquisition of Q reactivity. Indeed, these residues are not present in ancestral forms of Nuo complexes such as mbx, which do not use Q as an electron acceptor (table 1 and fig. 3).

To obtain independent evidence for the initial acquisition of Q reactivity in Nuo complexes, sequence analysis was next applied to the NuoB subunit binding cluster N2, which directly reduces Q (Sazanov 2015). This Fe–S cluster is unusually ligated to two consecutive cysteines in the N-terminal part of the protein (fig. 1A, cf. Sazanov and Hinchcliffe 2006), a feature that has been connected to the protonmotive capacity of complex I (Brandt 2006; Sazanov 2015). Alignment of the NuoB sequences indicates an abrupt appearance of this cysteine doublet in the Fpo complex of Methanosarcina (fig. 3B and table 2). Hence, the peculiar chemistry and protonmotive capacity associated with cluster N2 appeared early in evolution, aside if not instrumental to the acquisition of redox reactivity with the Q analog Mph (table 2). It is also present in the Fpo-like complex of Ca. Methanoplasma termitum (Lang et al. 2015), the NuoD homolog of which has both H38 and D139 (fig. 3 cf. Lang et al. 2015). As it is presently unknown whether Ca. Methanoplasma termitum contains Q analogs as in related archea (Shimada et al. 2001), its Fpo-like complex has been postulated to directly reduce disulfide reductase (Lang et al. 2015). Considering also the widespread transfer of enzymes between bacteria and archea (Nelson-Sathi et al. 2015), it remains difficult to envisage whether the Fpo-like complex of Ca. Methanoplasma termitum represents a truly intermediate step in the structure–function evolution of complex I. Nevertheless, the residues that concomitantly appear in the NuoD subunit of Fpo complexes (table 1) lie away from those appearing in the NuoB subunit (fig. 5).

Table 2.

Evolutionary variation of key residues of the NuoB subunit

Note.—Residues not present in ech hydrogenases and acquired with Mph reactivity in Fpo complexes are in white over brown background, whereas those sporadically present in ech or mbx complexes and then acquired together with MQ(H₂) reactivity in Chlorobi are highlighted in light brown. Chlorobium residues refer to a consensus sequence as in table 1.

Fig. 5.—

Model for progressive acquisition of Q reactivity by complex I. (A) Evolution of the reactivity with quinones. The 5 × 4 matrix represents the set of NuoD residues forming the Q reaction chamber in the crystal structure of complex I (fig. 1A, cf. Baradaran et al. 2013); on the right, it is shown empty and overlapped by residue M51 of the NuoB subunit to emphasize the closeness of the latter subunit in the 3D structure of Yarrowia complex I (Zickermann et al. 2015). The interaction of NuoD with the other Q-reacting subunits, transmembrane NuoH and NuoA, is modeled taking into consideration the membrane topology of the proteins (fig. 1A) and the overall distribution of inhibitor-resistant sites (table 3). Residues colored in dark brown are present in Ech hydrogenases or have been acquired with Mph reactivity in Fpo complexes, whereas those colored in light brown have been acquired with the reactivity with MQ in the NUO11 of Chlorobi and other bacteria (tables 1 and 2). Black hexagons indicate the residues forming the Q reaction chamber in Thermus complex I (Baradaran et al. 2013). Of these, only F147 and S36 of the NuoD subunit are not shown in the scheme; anyway, table 1 reports the evolutionary variation of S36. The yellow hexagons indicate the residues that show about 80% reduction in the NADH-DBQ activity of either Yarrowia (Tocilescu et al. 2007, 2010b) or E. coli complex I (Baradaran et al. 2013; Sinha et al. 2015). (B) Amino acid variation following chemical changes in natural quinone acceptors. The green squares indicate changes in Q-binding residues of complex I subunits that are potentially related to the reactivity toward PQ (table 3). The panel on the right, derived from the 3D images in fig. 1A, shows the principal Q-reacting residues of the NuoD and NuoB subunits as red cylinders; those that are changed in PQ-reacting complex I are indicated to provide structural reference to their position in the scheme on the left. Red diamonds indicate sites showing unusual amino acid changes that are present in the sequences of organisms having the natural Q analog, rhodoquinone (RQ, cf. fig. 4 and table 3).

Of note, analysis of the NuoH subunit revealed the evolutionary change of a single residue (E35, fig. 1A cf. fig. 4A) among those involved in the Q reaction chamber of Thermus complex I (supplementary fig. S1, Supplementary Material online), thereby indicating that this subunit was minimally involved in the acquisition of Q reactivity by complex I.

Fig. 4.—

Alignment of Q-reacting subunits of complex I. (A) Alignment of the NuoH/ND1 subunit highlights amino acid variations related to RQ reactivity. The sequences of bacterial NuoH possessing both ubiquinone and rhodoquinone (RQ) such as R.rubrum do not show unusual amino acid substitutions in the residues that form part of the Q reacting chamber of Thermus complex I (in black and white) or are associated with human pathological mutations (listed in red at the bottom of the alignment, cf. Iommarini et al. 2013). However, invertebrates that have RQ (Van Hellemond et al. 1995) do show unusual substitutions at or around these residues (marked in red with gray background as in fig. 3). Most notable is the substitution of the bulky hydrophobic M42 (in Thermus, F in mammals) with the hydrogen-bonding S, which is shared by nematodes and the anaerobic ciliate Nyctotherus, which also has RQ (Müller et al. 2012). Conversely, the substitutions at position 37 and 52 appear to be restricted to nematodes, the latter being shared with bivalves such as Mytilus, which also have RQ (Van Hellemond et al. 1995). Bivalves additionally show nonconservative substitutions of Q43, a residue forming part of the Q reacting chamber (Baradaran et al. 2013) that is involved in pathological mutations too (Iommarini et al. 2013). See supplementary fig. S1 (Supplementary Material online) for an enlarged alignment with bacterial NuoH subunits and further information. TM1 and AH1: transmembrane helix 1 and amphipatic helix AH1, respectively, in the structure of Thermus complex I (Baradaran et al. 2013). (B) Plant Nad7 and animal 49-kDa subunit show residue changes probably associated with resistance to Q antagonist inhibitors. Alignment blocks overlap and extend those of bacterial NuoD sequences in fig. 3A and show much stronger conservation (in bold) than across bacterial phyla. To date, only the Nad7 sequence of Pongamia or Millettia pinnata (accession: AET62928; Kazakoff et al. 2010) is available from the plants that produce rotenoids. It is aligned with homologous sequences from plants that do not produce rotenoids (Arabidopsis and Brassica) as well as Capsicum, which produces the weak complex I inhibitor capsaicin (Shimomura et al. 1989; Degli Esposti 1998). Of note, the sequence of Capsicum is very similar to that of other members of the Solanoideae family for which natural resistance to rotenoids has been documented, for example, potato. The sequence of the 49-kDa subunit of Amphymedon queenslandica (Srivastava et al. 2011) is aligned together with plant Nad7 and animal homologs to represent the protein variation in demisponges, which produce the complex I inhibitor mycothiazole (Morgan et al. 2010). Unusual or nonconservative substitutions are in red over gray background as in figs. 3, whereas sites of human mutations are in dark red as in A.

Analysis to answer Question 2: Are natural substitutions of key residues related to different quinone types?

Following the analysis described above, the next evolutionary step in the acquisition of Q reactivity is the appearance of H38, which directly binds a Q carbonyl in Thermus complex I (fig. 1A, cf. Sazanov 2015), together with D139 (fig. 5A cf. fig. 1A and table 1). Contrary to the NuoD homologs of Ech hydrogenases and ancestral Nuo complexes, Chlorobium NuoD shows these residues concomitantly with the substitution of the first and third vestigial ligand of the Ni–Fe cluster and the appearance of phylogenetically conserved residues that lie in the vicinity of bound Q in Thermus, in particular T135 (figs. 3A, 4B and table 1). Such a group of structurally interconnected residues appears not only in the NUO11 complexes of Chlorobi but also in that of gram-positive bacteria, for instance, Clostridium pasteurianum (fig. 3A and table 1), organisms that have menaquinone (MQ) as the dominant membrane quinone. The presence of H38 and D139 in plastidial NDH-1 complexes, however, is accompanied by the substitution of T135 with Pro and other nonconservative replacements (table 3 and fig. 5B) that could reflect, at least in part, the adaptation to react with plastoquinone (PQ), the specific quinone of photosynthetic organisms having the NDH-1 complex (Battchikova et al. 2011). Intuitively, PQ reactivity was acquired by changing complex I residues that could accommodate the modification of the double ring of MQ into the single, di-methylated ring of PQ. These residues include strongly conserved C39 in NuoA (Zickermann et al. 2015), which is mutated to Ser in plastidial sequences (fig. 5 and data not shown), and form a concerted set of amino acid variations that is distributed across greater part of the Q reaction chamber, especially at the interface between NuoD and the other subunits of the Q module (fig. 5B). While accommodating the diverse quinone ring of PQ, such variations did not alter the core chemistry of Q interaction, as previously concluded for PQ-reacting cytochrome b₆ (Nelson e al. 2005).

Table 3.

Amino acid variations in complex I subunits associated with Q function and resistance to Q antagonist inhibitors

Note.—Inhibitors are naturally occurring (their natural origin is annotated in the first row of the table), with the exception of quinazolines, for example, DQA. These synthetic inhibitors have been included given the abundance of mutagenesis data (Yamashita et al. 2004; Fendel et al. 2008; Murai et al. 2009; Tocilescu et al. 2010b; Sinha et al. 2015) and their resemblance to methanophenazine (Mph), the natural acceptor substrate for Fpo (Abken et al. 1998). Red squares in the central column indicate residues associated with approximately 80% reduction in DBQ activity after mutation (Tocilescu et al. 2007; Sinha et al. 2015). The question mark under the column of quinazoline inhibitors indicates possible hypersensitivity in Ascaris vs. bovine (Yamashita et al. 2004), whereas that in the column of mycothiazole indicates possible variations deduced from the available sequence of Cnidaria (see text). The asterisk under the columns of piericidin and quinazoline inhibitors indicates the labeling of the equivalent bovine residue with an acetogenin analog that can be displaced by bullatacin and quinazoline inhibitors (Masuya et al. 2014).

What about complex I adaptation to naturally occurring changes in the ring substituents of Q? To provide answers to this additional question, it has been investigated whether specific variation in complex I subunits could be associated with chemically different quinones of organisms having a fully sequenced genome, starting with the archaean Sulfolobus that possesses very peculiar quinones such as caldarellaquinone (Shimada et al. 2001). The sequence of Sulfolobus NuoD does not have H38, while showing Y87 and D139 substituted with residues such as Ile, the side chain of which is unable to establish H bonds with Q carbonyls. These differences are equivalent to those observed in the NuoD sequences of Nautiliales and Desulfurobacteriales (fig. 3), suggesting a common ancestry dating before the acquisition of effective Q reactivity. The function of the Nuo complexes of these organisms presumably consists in electron transfer between ferredoxin and NADP+, as in mbx complexes (cf. Bridger et al. 2011; Lücker et al. 2013).

The introduction of the methoxy ring substituents that differentiate ubiquinone (Q) from MQ and PQ has occurred when oxygen levels in the environment became sufficiently high to sustain the three hydroxylation reactions required in ubiquinone biosynthesis (Aussel et al. 2014). That time corresponded with the separation of aerobic proteobacteria from other prokaryotes (Segata et al. 2013; Degli Esposti et al. 2015). In contrast with recent speculations (Welte and Deppenmeier 2014), complex I subunits of aerobic proteobacteria do not show evident sequence variation that could be correlated with Q rather than MQ reactivity (fig. 3 and results not shown). Indeed, Thermus complex I equally reacts with its physiological MQ substrate and DBQ (Baradaran et al. 2013). However, peculiar changes have been noted in both the NuoD and the NuoH sequences of organisms that possess the rare Q analog, rhodoquinone (2-methoxy,3-amino,5-methyl,6-isoprenyl,2-4p-benzoquinone, RQ; figs. 4 and 5; table 3).

The replacement of one methoxy substituent with an amine group in RQ increases the number of potential H bonds that can be formed with interacting proteins. Hence, additional residues forming H bonds are in principle required for adapting Q-reacting enzymes to RQ, which normally is a poor substrate for respiratory complex I and II (Lenaz et al. 1968). Among bacteria, only R. rubrum and R. photometricum possess high levels of RQ (Hiraishi and Hoshino 1984). Conversely, RQ distribution is relatively dispersed among protists and metazoans adapted to anaerobiosis, such as the nematode Ascaris and the ciliate Nyctotherus, which contain RQ in their anaerobic mitochondria (Müller et al. 2012). Fortunately, there is a valuable model to discern RQ-specific changes, the Q-reducing Qi site in cytochrome b of the cytochrome bc₁ complex, which is functionally equivalent to the Q-reducing site in complex I. Cytochrome b proteins that exclusively react with Q show a notable amino acid variation at the Qi site with respect to cytochrome b proteins that react with either MQ or PQ (supplementary fig. S3, Supplementary Material online): the substitution of the bulky hydrophobic Ile with Ser at position 206 (yeast numbering; Degli Esposti et al. 1993a). S206 is indeed H-bonded to one methoxy group of bound ubiquinone in crystal structures of the bc₁ complex (Berry et al. 1999; Gao et al. 2003; Schütz et al. 2003). Additionally, the cytochrome b of RQ-containing organisms that retain also Q in their membranes, such as Rhodospirillum and Mytilus, shows the substitution of the nearby residue Thr204 with Lys, a change that also increases the capacity of forming H bonds (supplementary fig. S3, Supplementary Material online). In the aligned sequences of the NUOH/ND1 subunit of complex I, a substitution equivalent to that at position 206 of cytochrome b is observed at position 42, where bulky hydrophobic residues (Met in mammals or Phe in Thermus) are substituted by Ser in Nyctotherus and nematodes, but not other invertebrates (fig. 4A). The same organisms, together with those experiencing recurrent cycles of anaerobiosis like Mytilus, also show the substitution of conserved G52 with H bond-forming Ser or Thr (fig. 4A). Moreover, ND sequences from bivalves show nonconservative substitutions of Q43, which is part of the Q reacting chamber in Thermus (fig. 1A cf. fig. 4A). Position 43 and 52 are associated with pathological mutations too (fig. 4A, cf. Iommarini et al. 2013).

All the above changes in the ND1 subunit do not occur in the homologous NuoH subunit of RQ-containing Rhodospirilli, which instead show specific replacement of two Q-interacting residues in NuoD (fig. 3A and supplementary fig. S1, Supplementary Material online) and one in NuoB (table 3). Hence, complex I adaptation to RQ reactivity has followed different evolutionary paths in proteobacteria and mitochondria. Consequently, the sporadic presence of RQ in eukaryotes cannot be considered a bona fide relic from the bacterial ancestors of mitochondria (see Müller et al. 2012 and discussion therein), otherwise complex I adaptation to react with RQ would have involved the same subunits in prokaryotes and eukaryotes.

Answer to Question 3: Complex I changes in organisms producing Q antagonist inhibitors

After obtaining insights on Q reactivity and its adaptation to diverse natural Q substrates, the analysis was extended to the Nuo subunits of organisms that produce Q antagonist inhibitors of complex I (figs. 3A and 4B). Such organisms must have adapted their Q reaction sites so as to minimize the inhibitory action of the Q antagonists they produce, in analogy with earlier studies on natural resistance toward Q antagonist inhibitors of the bc₁ complex (Ghelli et al. 1992; Degli Esposti et al. 1993a; Kraiczy et al. 1996; Ouchane et al. 2002). The most potent inhibitors of complex I derive from plants of the Annonaceae family, for example, rolliniastatin-2 or bullatacin (Degli Esposti 1998); however, no full sequence of complex I subunits from Annonaceae is available to date. Labeling studies in bovine heart mitochondria have indicated that these acetogenins interact with both the ND1 and the 49-kDa subunits of complex I (Masuya et al. 2014). In the latter protein, an acetogenin analog specifically binds to the residue corresponding to D139 in Thermus complex I (fig. 3A); this binding is displaced not only by bullatacin, but also by a quinazoline inhibitor (Masuya et al. 2014).

Piericidin A is next in potency to acetogenins and is produced by organisms of the Streptomyces group of gram-positive bacteria (Takahashi et al. 1965; Chen et al. 2014). The genome of some piericidin-producing Streptomyces strains is currently available and codes for early versions of NUO14 complexes that are expressed in the normal physiology of the bacteria (Surup et al. 2008; Spero et al. 2015). The NuoD and NuoH subunits of these piericidin-producing organisms show remarkable substitutions of amino acids forming the Q reaction chamber in Thermus complex I (table 3), even in comparison with closely related species that produce weak complex I inhibitors such as iromycins (Surup et al. 2008), which have NuoD sequences very close to that of S. coelicolor (Borodina et al. 2005), shown in fig. 3. These substitutions are either unique or very rare and isolated among the numerous sequences from Streptomyces and other Actinomycetes that are now available, thereby defining likely correlations with natural resistance toward piericidin inhibitors. In particular, piericidin-producing strains show the substitution of Y87 with Trp (W, fig. 3A and table 3), an identical change to site-directed mutations that produce severe reduction in DBQ activity and strong resistance to rotenone and other complex I inhibitors (Kashani-Poor et al. 2001; Tocilescu et al. 2010b; Sinha et al. 2015). Another remarkable change occurs at position 401 of NuoD, the same site that was previously found to be responsible for resistance to piericidin A in Rhodobacter mutants (Darrouzet et al. 1998; table 3). These and other sites of experimentally induced or natural resistance toward piericidin A have been mapped in the diagram containing the residues forming the Q reaction chamber in Thermus complex I (fig. 6, cf. figs. 1A and 5). Their distribution appears to define two domains (fig. 6), consistent with earlier evidence that piericidin A binds to two sites in complex I (Gutman et al. 1970; reviewed by Degli Esposti 1998). The first domain is formed by three residues of the NuoH subunit (E35, R36, and R50) and several residues of the NuoD subunit that are either involved in the binding of piericidin A in the crystal structure of Thermus (Baradaran et al. 2013) or lie near the Q reaction chamber, such as M85 and G405 (fig. 1A cf. fig. 6). The second domain includes a few NuoD residues, in particular T135 and R350, which are also linked to natural resistance toward myxococcal antibiotics (fig. 6 and table 3).

Myxococcal antibiotics include potent Q-antagonist inhibitors of bacterial and mitochondrial respiratory chains that are produced by predatory delta proteobacteria of the order Myxococcales, for example, Myxococcus and Stigmatella (Reichenbach et al. 1988; Friedrich et al. 1994; Degli Esposti 1998). Among these Q-antagonists, myxothiazol and stigmatellin inhibit both the bc₁ complex and complex I (Degli Esposti et al. 1993a; Degli Esposti 1998) and are produced in mixtures that vary between species and strains (Reichenbach et al. 1988). Anaeromyxobacter is a clear exception among Myxococcales because it does not produce the fruiting bodies in which the antibiotics are stored (Huntley et al. 2011). Consequently, the sequences of complex I subunits of Anaeromyxobacter can be used as a reference for identifying unusual amino acid changes that may confer resistance to myxococcal Q-antagonist inhibitors. These changes occur at several NuoD sites (H38 to S, D86 to N, T135 to A, R350 and D401 to N) that form part of the Q reacting chamber in Thermus complex I (table 3, figs. 3 and 6, cf. fig. 1A). The same changes are present in the NuoD sequence of Sorangium (fig. 3), a close relative of the Polyangium strains that produce phenoxan—the most potent inhibitor of complex I among myxococcal antibiotics (Kunze et al. 1992; Friedrich et al. 1994)—for which no Nuo gene is currently available. It thus appears that complex I of delta proteobacteria has adapted to counteract all myxococcal Q antagonists by a common set of residue substitutions that are spread across the NuoD and NuoB subunits forming the Q reaction chamber (table 3, figs. 3 and 6). Moreover, the NuoA subunit of Q-antagonist-producing Myxococcales shows the substitution of the highly conserved C39 with Ala (table 3, cf. Zickermann et al. 2015), while recurrently replacing highly conserved Asp residues with Asn in the NuoD subunit (fig. 3 and table 3). Interestingly, natural resistance to myxococcal antibiotics involves sites that are also changed in PQ-reacting proteins (fig. 6B cf. fig. 5B), or are associated with resistance toward rotenone and quinazoline inhibitors (table 3). Hence, complex I evolution could have followed a common structural strategy to reduce the interference of Q antagonist inhibitors and adapt its reactivity to Q substrates different than MQ, which is the standard quinone in delta proteobacteria (Iizuka et al. 2003).

Other naturally occurring Q antagonists include rotenone, the classical inhibitor of complex I (Singer and Ramsay 1994; Degli Esposti 1998). Rotenone is part of the isoflavonoid family of rotenoids produced by Derris and Millettia, leguminosae plants that belong to the Millettiae tribe (Brierley and Smith 1968; Simin et al. 2002; Ngandeu et al. 2008). Among the species producing rotenoids, only the Nad7 sequence of Millettia pinnata is currently available (Kazakoff et al. 2012) and its alignment with homologous sequences from plants that do not produce rotenoids (e.g., Arabidopsis) indicate the substitution of two residues that form part of the Q reaction chamber in complex I: T135 and F146, both changed to Ser (fig. 4B cf. fig. 3). The first substitution is the opposite to a directed mutation introduced in the homologous 49-kDa subunit of Yarrowia, which also has Ser at position 135 and is naturally resistant to rotenone (Fendel et al. 2008; Tocilescu et al. 2010b). Re-introduction of T135 is associated with an increased sensitivity toward rotenone in Yarrowia complex I (Fendel et al. 2008), thereby validating the possibility that its replacement with Ser in some plants confers natural resistance toward rotenoids. This substitution remains unique to rotenone-producing Millettia when combined with the substitution of F146 with Ser (fig. 4B cf. fig. 3). Conversely, Nad7 sequences of Solanoideae plants show the rare substitution of both L143 and M404 (Phe in other eukaryotes) with Ser (fig. 4B). Among these plants is Capsicum, which produces the weak complex I inhibitor capsaicin (Shimomura et al. 1989), as well as potato. Hence, the changes of either L143 or M404 could be associated with natural resistance toward capsaicin and may also be responsible for the strong rotenone resistance documented in potato complex I (Degli Esposti 1998). Of note, a substitution of position 404 with Ser in NuoD has been found to be responsible for in vivo resistance of Helicobacter strains to benzimidazole inhibitors (Mills et al. 2004), compounds that belong to type C inhibitors (Andreani et al. 1995; Degli Esposti 1998). Recently, rotenone has been reported to inhibit also the ferredoxin-mediated NADPH oxidase activity of the NDH-1 complex isolated from the cyanobacterium, Thermosynechococcus elongatus (Hu et al. 2013). The NuoD homolog of this complex has the same variations in the Q reacting chamber as those present in the related Synechocystis shown in fig. 3A. As only two of such variations occur at sites associated with rotenone resistance (table 3), rotenone sensitivity of cyanobacterial NDH-1 complex fits with the structure–function relationships deduced here.

Mycothiazole is a novel complex I inhibitor that has been recently isolated from marine demisponges, for which only the sequence of the 49-kDa subunit of Amphymedon queenslandica is currently available (Srivastava et al. 2011). Its alignment with NuoD homologs shows the substitution of T135 with Ser combined with that of L143 with Asn (figs. 3 and 4B). The combined changes are similar to those found in Millettia but include a different hydrophobic residue forming the Q reaction chamber, namely, L143 instead of F146. However, both residues lie in the vicinity of bound ubiquinone (figs. 1A and 6, cf. Fendel et al. 2008) and therefore could contribute to the binding of either rotenone or mycothiazole, as well as capsaicin (see above). This possibility would imply that mycothiazole may function like rotenone in inhibiting complex I, a possibility that is consistent with some biochemical data (Morgan et al. 2010 cf. Degli Esposti 1998), but needs to be substantiated by detailed studies on complex I of demisponges. In the meantime, complex I sequences available from marine metazoans, such as the cnidarian Nematostella (Putnam et al. 2007), show changes in the Q-interacting residues of the NuoB subunit (reported in table 3 too), which perhaps reflect instances of natural resistance toward mycothiazole and other natural complex I inhibitors released in the ocean, an area or biological research that has yet to be explored.

Rotenone resistance in NUO13 complex I and its evolutionary implications

NUO13 Complex I of E. coli is known to be strongly resistant to rotenone (Friedrich et al. 1994), contrary to the NUO14 complex I of other proteobacteria such as Paracoccus or Rhodobacter capsulatus (Darrouzet et al. 1998; Dupuis et al. 1998). The analysis just presented suggests the first molecular reasons to explain E. coli resistance to rotenone (table 3). Across all Q-reacting subunits of complex I, E. coli sequences show six amino acid replacements that occur at the same sites that are modified in PQ-reacting plastidial subunits, including Cys39 to Ser in NuoA (table 3). In NuoD, the replacement of Q-interacting L138 with Q (fig. 4B) is similar to mutations induced in nearby sites producing resistance to rotenone and DQA in Yarrowia complex I (Tocilescu et al. 2010a, b) and therefore is likely to contribute to rotenone resistance in E. coli (table 3). Four sites that are unusually substituted in E. coli NuoB and NuodD are also changed in Myxococcales (table 3), in agreement with the resistance of E. coli complex I to myxococcal inhibitors too (Friedrich et al. 1994). Conversely, E. coli NuoD shows only the nonconservative substitution of M85 among the sites that could be associated with natural resistance to piericidin (table 3, cf. fig. 1A). This observation is consistent with the moderate sensitivity of E. coli complex I to piericidin A (Friedrich et al. 1994) compared with other proteobacteria (Grivennikova et al. 2003). Interestingly, the changes observed in E. coli subunits are also seen in the NUO13 complex of other species that either have only this type of complex I (e.g., Asaia) or additionally have a NUO14 complex I, as in Rhodopseudomonas palustris (Oda et al. 2008; Spero et al. 2015). It is not clear why some organisms possess two different types of complex I. NUO13 complexes may be predominantly involved under anaerobic conditions (Spero et al. 2015), or in reverse electron transport to supply NADPH for biosynthetic purposes (Sharma et al. 2012; Lücker et al. 2013). However, the NUO14 complex of Rhodobacter is considered to predominantly work in reverse electron transport (Dupuis et al. 1998; Herter et al. 1998).

The present work suggests another explanation for the presence of different Nuo operons in some bacteria, which is based on the spectrum of amino acid changes associated with resistance toward naturally occurring inhibitors (table 3). The NUO13 complex may confer ecological (and consequently also evolutionary) advantages in environments where predatory Myxococcales thrive, for example, soil, because it can provide bioenergy and metabolic activities that remain insensitive to the Q antagonist antibiotics released by competing myxococcal organisms. Indeed, the presence of both NUO13 and NUO14 complex I is common among soil bacteria such as Rhodopseudomonas, which require complex I function in their metabolic adaptation to changes in soil environments (Oda et al. 2008; Spero et al. 2015). Although it has been assumed that myxococcal antibiotics are released to successfully compete with protists and fungi in the soil (Reichenbach et al. 1988), it is plausible that some bacteria might have evolved successful strategies to counteract the same antibiotics.

Functional studies with Q substrates re-define the old concept of two Q reaction sites in complex I

The structural deductions presented so far could be correlated with a wealth of biochemical data that suggested two different Q or inhibitor sites in complex I, an old concept in bioenergetics (Gutman et al. 1970; Singer and Ramsay 1994). Differences in complex I reaction with Q can be discerned by using protonmotive measurements with chemically diverse quinones, as the initial reduction and subsequent protonation of Q substrates contribute differently to the proton pumping capacity of the complex (Degli Esposti et al. 1996; Helfenbaum et al. 1997; Ohshima et al. 1998; Sharma et al. 2015). The bioenergetic capacity of natural Q analogs discussed here and of other quinones is compared in fig. 7 and the results, together with previous reports (Degli Esposti et al. 1996; Helfenbaum et al. 1997), operationally define two Q reaction sites in complex I: the distal, hydrophobic site where the Q substrates enter the complex and are initially reduced—in a reaction intensely coupled to protonmotive activity—and the hydrophilic site where the reduced substrates then move, before being released as products. The latter site produces little membrane potential and is most accessible to hydrophilic quinones such as Q0 (2,3-dimethoxy,5-methyl,p-benzoquinone, cf. Degli Esposti et al. 1996).

Fig. 7.—

Membrane potential generation in bovine complex I with different Q analogs. Generation of membrane potential by complex I was measured by following the absorbance changes of oxonol VI in coupled submitochondrial particles from bovine heart (to a final concentration of 0.15-0.2 mg/ml) activated with NADH and then treated with inhibitors of other respiratory complexes plus nigericin as described earlier (Degli Esposti et al. 1996; Helfelbaum et al. 1997). PQ and RQ refer to the decyl analog of PQ and rhodoquinone, respectively. (A) The reaction was started with 0.1 mM NADH after the quinones had been equilibrated at a final concentration of ca. 0.02 mM for at least 2 min with the particles. (B) The reaction was started with ca. 0.02 mM quinone after the particles had been equilibrated with 0.1 mM NADH for at least 2 min. (C) Reproduction of the absorbance recording at 630 minus 601 nm obtained with a Cary spectrophotometer using the conditions described by Degli Esposti et al. (1996) with only 5 µM NADH, which enabled approximately one-half the membrane potential burst obtained with saturating concentrations of NADH (cf. D). Subsequent addition of a stoichiometric (1) concentration of Q elicited very different levels of membrane potential for Q0 (red solid trace) and DBQ (black solid trace). The dashed lines represent instead the traces obtained with the addition of excess (2, 0.02 mM) of either quinone. (D) The experimental conditions were as in C but the concentration of NADH was 0.1 mM, close to the saturation level for reducing complex I. After the initial addition of excess Q0 (0.02 mM as in C), in-flight addition of 5 µM DBQ produced rapid generation of membrane potential that followed a very similar time-course to that obtained in the absence of Q0 (cf. C). After this was completed, complex I maintained the slowly decaying level of membrane potential generated by Q0 alone, which proceeded until complete exhaustion of excess reductant, as described earlier (Helfenbaum et al. 1997). On the right, such a reaction was interrupted by the addition of the uncoupler CCCP (0.5 µM). The dotted line at the bottom represents the trace obtained on the addition of excess Q0 after the particles had been treated with 0.3 µM rotenone, showing essentially complete inhibition of membrane potential generation. Under the same conditions, a small generation of membrane potential was observed with DBQ (Degli Esposti et al. 1994). The results are representative of three to four similar experiments conducted in part by collaboration with Anna Ghelli and Bruna Benelli (University of Bologna).

The relative efficiency in membrane potential generation decreases in the order of decyl Q, PQ, and RQ, despite the similar hydrophobicity of these analogs; the rates of decyl RQ are even lower than those of Q0 (fig. 7A). All these quinones generate membrane potential when equilibrated with mitochondrial particles before the addition of NADH to initiate the reaction (fig. 7A). However, when they are added after the particles have been equilibrated with NADH, only DBQ and Q-1 produce similar rates of membrane potential generation (fig. 7B). In contrast, PQ, RQ, and Q0 produce hardly any membrane potential under these conditions leading to full reduction of endogenous Q, even if they still function as electron acceptors for the redox reaction of complex I (not shown; cf. Lenaz et al. 1968; Ruzicka and Crane 1970; Degli Esposti et al. 1996; Fato et al. 1996; Ohshima et al. 1998). This remarkable difference in membrane potential generation whether the reaction is started with NADH or Q is simply explained by considering that endogenous QH₂ occupies the hydrophilic site in the complex pre-reduced with NADH and effectively competes with hydrophilic Q0, as well as PQ and RQ, but not with DBQ. Complex I has a much higher affinity for oxidized than reduced Q (Verkhovsky et al. 2012) and this applies also to hydrophobic Q analogs having 9 to 11 carbons in their tail (Degli Esposti et al. 1996); hence, endogenous QH₂ does not compete with these quinones.

The above explanation is consistent with structural data, as hydrophilic quinones and ring-substituted Q analogs such as PQ, once penetrated inside the Q reaction chamber through the common entry, would preferentially react with the amino acids that normally interact with QH₂, which minimally contribute to membrane potential generation in complex I (Degli Esposti et al. 1996; Helfenbaum et al. 1997). Among the quinones shown in fig. 7, only Q-1 interacts with both the distal and the hydrophilic site, especially in the absence of endogenous QH₂; the same would apply to Q-2, which is small enough to elicit double occupancy of the Q reaction chamber, but is more efficient than Q-1 in displacing endogenous QH₂ owing to its increased hydrophobicity (Degli Esposti et al. 1996). Structurally similar to Q-2, piericidin A (cf. fig. 6) would occupy the two sites simultaneously, for it appears to have similar affinity for either site (Gutman et al. 1970).

This rationalization of the differential functional activity of various Q analogs implies that hydrophilic Q0 would not enter the distal site, whereas DBQ would react with the same site even in the presence of excess Q0 (or quinols) at the hydrophilic site. The results presented in fig. 7 sustain this possibility, as they show the additive effect of DBQ (at low concentrations, equivalent to those of NADH in fig. 7C) in generating membrane potential following excess of Q0 (fig. 7D). The transient trace of additional membrane potential has the same time-course and similar intensity as that produced by DBQ in the absence of Q0 (fig. 7C and D). Of note, rapid addition of excess Q0 during DBQ-induced membrane potential had little effect on the recorded absorbance changes. Consequently, DBQ generates membrane potential that is additive to the rotenone-sensitive potential generated by Q0, thereby indicating two Q reaction sites. Indeed, DBQ produces equimolar oxidation of NADH, whereas Q0 induces over-stoichiometric oxidation of NADH until complete exhaustion of the reductant, as also indicated by the trace obtained with excess Q0 in fig 7C (trace 2). The latter effect is owing to the electron sink capacity of hydrophilic quinones including Q-2 that elicit double occupancy and reciprocal dismutation within complex I (Degli Esposti et al. 1996; Helfenbaum et al. 1997).

Conclusions

This work has exploited the abundant genomic information available today to obtain new insights into the evolution of the structure and function of respiratory complex I, an ancient bioenergetic enzyme. The in-depth analysis of catalytic protein subunits helps re-defining the concept that respiratory complex I may contain two sites of interaction with Q substrates or their antagonists, which generated a long-standing debate in bioenergetics after the detailed biochemical studies of T. Singer and co-workers (Gutman et al. 1970; Singer and Ramsay 1994). Subsequent studies with various inhibitors of complex I (Friedrich et al. 1994; Degli Esposti et al. 1994; Degli Esposti 1998) converged with biophysical studies on semiquinones (Magnitsky et al. 2002) and photoaffinity labeling experiments (Schuler et al. 1999; Murai et al. 2009) in sustaining the presence of two Q or inhibitor sites in complex I. In contrast, U. Brandt and co-workers have considered all the above evidence to be compatible with a single large cavity for Q and inhibitors in complex I, predominantly on the basis of binding (Okun et al. 1999) and mutagenesis results (Brandt 2006; Fendel et al. 2008; Tocilescu et al. 2010b). The concept of a single Q or inhibitor site was later sustained by the structural work by A. Sazanov and co-workers, who showed an unusually large Q reacting chamber in Thermus complex I, into which a single molecule of either DBQ or its antagonist piericidin A was bound (Baradaran et al. 2013). However, the structure of Yarrowia complex I (Zickermann et al. 2015) shows that the Q-antagonist DQA is bound to a restricted pocket corresponding only in part with that binding Q in Thermus complex I (fig. 1A, cf. Sazanov 2015). This structural difference has inspired the present work.

In evolutionary stages, structurally different parts of the vast Q reaction chamber (fig. 1A cf. Sazanov 2015) were formed at the time in which complex I acquired reactivity with the quinone analog, Mph, which is rather similar to Q antagonist inhibitors such as quinazolines (tables 1 and 2 and fig. 5). The distal Q site was initially restricted to the NuoD subunit and later incorporated the extrinsic loop following the first transmembrane helix of the NuoH subunit, which frames the Q entry into the complex (figs. 5 and 6, cf. Baradaran et al. 2013). The hydrophilic site would instead correspond to the part of the chamber that accommodates the ubiquinol product, intrinsically more hydrophilic than the oxidized form. Consequently, hydrophilic quinones would preferentially react with this site in competition with endogenous QH₂, whereas hydrophobic quinones would not bind to the same site until fully reduced. Short-chain Q homologs such as Q-2 are likely to bind to both sites simultaneously (Degli Esposti et al. 1996), even if they may have slightly different affinity for one site or the other, as suggested by their enzymatic properties in mutants of Q-binding Y87 (Tocilescu et al. 2010a,b). The chemically similar piericidin A (fig. 6) appears instead to have equivalent binding affinity for both sites in complex I (Gutman et al. 1970; Grivennikova et al. 2003). Of note, the binding data of Okun et al (1999) would be compatible with two piericidin binding sites too, by considering the over-estimation of complex I content in their bovine mitochondrial particles (0.06 nmol per mg of protein, whereas the normal content in the same particles is 0.03–0.04 nmol per mg of protein; Gutman et al. 1970).

The above re-definition of the old concept of two Q reaction sites can be extended to the classification of the large variety of Q antagonists of complex I, previously defined as class I and II (Friedrich et al. 1994) or type A and B (Degli Esposti 1998) inhibitors. Most of such inhibitors can now be re-considered as a single category of chemically different compounds that share the capacity of interacting with both the distal and the hydrophilic Q site. In other words, they effectively have two interaction sites in complex I like Q-2, even if they reach these sites from a common entrance in the complex. Only a few compounds such as rolliniastatins would not compete or overlap with type C inhibitors, a category of compounds introduced by this Author and co-workers (Degli Esposti et al. 1994; Andreani et al. 1995; Degli Esposti 1998) to account for complex I inhibition by myxothiazol and chemicals competing with ubiquinol (Degli Esposti et al. 1993a,b; Degli Esposti et al. 1994).

The above characteristics can now be rationalized in molecular terms by considering the mapping of resistance sites to myxococcal antibiotics and capsaicin, which overlap the second domain of piericidin resistance and many resistance sites for rotenone and quinazoline inhibitors, but not the first domain of piericidin resistance that roughly corresponds to the distal Q site (fig. 6 and table 3). Type C inhibitors, instead, predominantly interact with the hydrophilic Q site in complex I, a concept that introduces a simplified classification of complex I inhibitors following its evolution for adapting to natural Q substrates. This re-classification of inhibitors would not be inconsistent with the evidence that a single molecule of piericidin A is bound to Thermus complex I (Baradaran et al. 2013), because piericidin sensitivity in Thermus membranes is approximately one-order of magnitude lower than in mammalian mitochondria (Meinhardt et al. 1990 cf. Degli Esposti 1998). Consequently, the second interaction site of the inhibitor is likely to be weaker in Thermus than in mammals or other bacteria (Grivennikova et al. 2003), perhaps owing to unusual substitutions in the NuoD subunit such as A to P at position 391 (fig. 4B). The same substitution occurs in the 49-kDa subunit of Yarrowia complex I, which has limited sensitivity to piericidin (Kerscher et al. 1999) compared with closely related fungi such as Cyberlindnera fabiamii (formerly Candida utilis; Coles et al. 1974) having A at position 391 of the same subunit (not shown).

The final conclusion regards the new light that this work is throwing on the evolutionary pattern of complex I following major changes in its function. From the ancestral Nuo complexes retaining the mbx function of transferring electrons from reduced ferredoxin to NADP+, Fpo complexes developed the earliest acquisition of Q reactivity with Mph, which then evolved into MQ- or PQ- oxidoreductase enzymes in photosynthetic Chlorobi and Cyanobacteria. Oxidation of NADH arrived later, once the N-module was incorporated into the initial NUO11 operon (figs. 1B and 2). Overall, complex I has thus changed its function three times before becoming the huge bioenergetic machine we are familiar with. While maintaining the same core structure, local sequence variation in key subunits of the complex has occurred multiple times for adapting to changes in substrates or acquiring inhibitor resistance. These variations have been discerned here to produce insightful structure–function correlations that can now be tested experimentally, either by enlarging the sequence information currently available or undertaking focused functional studies.

Methods

The approaches followed in this work mostly derived from those recently introduced for studying the evolution of mitochondria and bacterial terminal oxidases (Degli Esposti et al. 2014, 2015). In essence, the program Domain Enhanced Lookup time Accelerated BLAST, DELTABLAST (Boratyn et al. 2012), has been used to investigate Nuo proteins in the National Center for Biotechnology Information (NCBI) resources. Alignments of protein sequences have been refined manually from initial drafts obtained with the COBALT resource of the DELTABLAST program (Degli Esposti et al. 2014), complemented with phylogenetic trees obtained with the program PhyML 3.0.

Given the variety in the composition and sequence of nuo operons in prokaryotes (Moparthi and Hägerhäll 2011; Marreiros et al. 2013; Spero et al. 2015), the principal types of gene clusters for bacterial complex I have been identified systematically with the simple annotation of the number of their single-copy subunits. The complete set of these subunits is 14 for the majority of alpha, beta, and gamma proteobacteria (Spero et al. 2015) and therefore, their operon is named NUO14. NUO13 is instead the complete operon for the subset of alpha and gamma proteobacteria, including E. coli, which have NuoD fused with NuoC, whereas NUO12 corresponds to complete operons exhibiting the additional fusion of NuoD with NuoB, as in some deltaproteobacteria (Marreiros et al. 2013; Spero et al. 2015—cf. supplementary fig. S2, Supplementary Material online). Organisms possessing operons lacking the N module of the NuoE-G subunits within their genome are considered to have the ancestral NUO11 operon (or NUO10 when NuoD is fused to NuoC, as in Nautiliales). Intermediate combinations resulting from additional copies of one subunit are labeled NUO11+, that subunit, for instance, NUO11+M for the operon of Desulfovibrio africanus (supplementary fig. S2, Supplementary Material online), whereas combinations that derive from the loss of one or more subunits from a primordial NUO14 cluster are labeled NUO14minus, the missing subunit. When the operon is separated in different sub-clusters, it is defined as split, as, for instance, in the case of Bdellovibrio (supplementary fig. S2, Supplementary Material online). Detailed phylogenetic analysis has indicated that, with the exception of Nautiliales, the various gene clusters present in epsilon proteobacteria (Weerakoon and Olson 2008; Marreiros et al. 2013) fall in this category of derived operons (fig. 1 and data not shown) and therefore cannot be considered as possible intermediates in the evolution of the complete Nuo enzyme as previously considered (Moparthi and Haggerwall 2011). The style reported by T. Yagi and co-workers (Sinha et al. 2015) has been used for the protein subunits of bacterial complex I, for example, NuoD. For proteins that are coded also in mtDNA, the bacterial definition has been combined with the nomenclature of the corresponding mitochondrial gene (e.g., Nad7 for NuoD), as well as the original classification of bovine subunits introduced by J. Walker and colleagues (e.g., 49 kDa for NuoD; Fearnley and Walker 1992).

Three-dimensional images of the crystal structure of the entire respiratory complex I of Thermus (Baradaran et al. 2013) have been obtained with the program CCP4MG, version 2.10.4. Chemical structures of inhibitors and quinones were obtained with the program JChem (ChemAxon.com). Bioinformatic data have been integrated with available biochemical and microbiological information, including unpublished data on complex I activities, which were obtained in coupled and NADH-activated submitochondrial particles from bovine heart following the methods described earlier (Degli Esposti et al. 1994; Degli Esposti et al. 1996).

Supplementary Material

Supplementary figures S1–S3 are available at Genome Biology and Evolution online (http://www.gbe.oxfordjournals.org/).