A modeling and simulation perspective on the mechanism and function of respiratory complex I

Respiratory complex I is a giant redox-driven proton pump, and central to energy production in mitochondria and bacteria. It catalyses the reduction of quinone to quinol, and converts the free energy released into the endergonic proton translocation across the membrane. The proton pumping sets up the proton electrochemical gradient, which propels the synthesis of ATP. Despite the availability of extensive biochemical, biophysical and structural data on complex I, the mechanism of coupling between the electron and proton transfer reactions remain uncertain. In this work, we discuss current state-of-the-art in the field with particular emphasis on the molecular mechanism of respiratory complex I, as deduced from computational modeling and simulation approaches, but in strong alliance with the experimental data. This leads to novel synthesis of mechanistic ideas on a highly complex enzyme of the electron transport chain that has been associated with a number of mitochondrial and neurodegenerative disorders.


Introduction
Plants, animals and microorganisms fulfill their energetic requirements in the most optimal ways. They utilize highly complicated, yet efficient, metabolic pathways for their continuous growth and development. Depending on the environmental conditions, they are capable of re-engineering their metabolic pathways for efficient energy production. For instance, the nitrogen fixing bacterium Bradyrhizobium japonicum terminates the respiratory chain into two types of cytochrome oxidases; aa 3 -type that is expressed under normal aerobic conditions, whereas the other cbb 3 -type oxidase is employed only under low oxygen tensions due to its very high affinity for dioxygen (K M~7 nM) [1]. This shows that organisms employ multiple strategies to cater their needs for one of the most fundamental molecules in biology, ATP (adenosine triphosphate). In the power house of eukaryotic cells, mitochondrion, ATP is synthesized predominantly through oxidative phosphorylation. In this process, several membranebound respiratory complexes catalyze electron transfer reactions from NADH (E m,7 NAD + /NADH~−320 mV) to the terminal electron acceptor, dioxygen (E m,7 O 2 /H 2 O~+800 mV). These electron transfer reactions are tightly coupled to proton pumping across the membrane leading to the formation of electrochemical proton gradient (or proton motive force, pmf), which is about 150-200 mV depending upon the respiration state of mitochondria. The pmf then drives the ATP synthase to generate ATP from ADP and P i through a rotary mechanism as proposed by Boyer [2], and confirmed by X-ray studies and biophysical measurements [3,4]. The rotary-catalysis mechanism of ATP synthase is now well supported by a wealth of experimental data [5], and also by a number of modeling and simulation studies [6,7], which provided detailed mechanistic insights not easily obtained from the experiments. One major unresolved question in the field is how the proton transfer in the F O part [8,9] is coupled to the ATP synthesis in the F 1 part of the enzyme.
In the classical "linear view" of the electron transport chain (ETC), the respiratory enzymes (ETC complexes and ATP synthase) float in the sea of lipids, and the sequential electron transfer from complexes I/II → III → IV is tightly coupled to the proton pumping across the membrane. In this process of long-range electron transfer, mobile electron careers such as the membrane-bound quinone and water-soluble cytochrome c play key roles. In the inner mitochondrial membrane, ubiquinone (Q 10 , quinone with ten isoprene units) dynamically shuttles between complexes I, II and III transferring electrons coupled with the uptake and release of protons. The ubiquinone structure resembles that of the lipids with a polar head group and a long (ca. 30 carbon) hydrophobic tail. Free energy simulations show that the head group of Q prefers certain locations along the bilayer normal [10,11], and that the Q molecules undergo rapid flip-flop between the two leaflets of the bilayer, which may be important for the optimal turnover of the ETC [11].
A second dominant view in the field is of mitochondrial supercomplexes, in which individual respiratory complexes join together to form a single unit, albeit in different stoichiometric amounts [12]. These so called respirasomes catalyze redox-driven proton pumping in a much tightly packed environment (see also recently solved structures [13,14]), which may be necessary to prevent ROS formation [15,16]. The advantages of having respirasomes as functional entities over individual complexes are not fully understood, and the topic remains highly debated [17]. Interestingly, respiratory supercomplexes (III + IV) have been studied with coarse-grained molecular dynamics (MD) simulations imitating the crowded inner mitochondrial membrane environment [18]. Results reveal an important role of cardiolipin molecules in gluing the respiratory complexes together in a variety of conformational arrangements [18], in agreement with earlier data [19].
In contrast to limited knowledge on supercomplexes, individual respiratory complexes have been studied for several decades through different types of biochemical, biophysical and structural techniques [20][21][22]. They have been characterized in both detergents and liposomal conditions through site-directed mutagenesis studies, spectroscopic and as well as state-of-the-art electrometric techniques [20]. Thanks to these immense efforts, the molecular mechanisms of complexes III and IV are relatively well-understood. Some delicate questions remain to be answered, in which computational approaches are playing the most important role. See refs. [23,24] and refs. [25,26] for recent computational work on complexes IV and III, respectively. In detailed mechanistic discussions below, this existing mechanistic knowledge on cytochrome oxidase (complex IV), cytochrome bc 1 (complex III) and Photosystem II is utilized to delineate complex I working principles.
Among all the ETC enzymes, complex I remains least understood, especially because of its large size and very high complexity. The respiratory complex I comprises 14 to 45 subunits, with mass ranging from 500 kDa to 1 MDa in bacteria to mitochondria, respectively. Earlier, research on complex I progressed slowly due to the unavailability of any suitable biochemical assay to estimate activity and efficiency of the enzyme. However, currently highly active preparations of complex I are being used in various labs for its mechanistic explorations (see refs. [27,28]). The crystal or cryo EM structures of complex I [29][30][31][32][33][34][35] have provided the most spectacular insights into enzyme architecture, in particular, it is now absolutely clear that all three respiratory enzymes (I, III and IV) utilize completely different strategies to generate pmf, even though some microscopic elements and principles may be shared. In the light of these well-understood respiratory and photosynthetic enzymes, we discuss some novel mechanistic ideas on complex I in strong alliance with computational methods and tools.
Computational approaches such as molecular dynamics (MD) simulations are particularly effective when applied to static structural data because they allow spatio-temporal resolutions to be gauged that are currently outside the scope of expensive experimental instruments. Emerging techniques such as super resolution microscopy spans time and length scales of ca. 15 millisecond and 20 nm, respectively. Furthermore, the time-resolved crystallography captures atomic resolution images of processes that occur in femtoseconds to hundreds of picoseconds [36]. However, it is the reactions that occur in hundreds of nanoseconds to hundreds of microseconds for which there are no experimental techniques that would provide high resolution data. Examination of such processes can indeed be performed by modeling and simulation techniques. An additional factor that makes biomolecular computations highly attractive is their ability to simulate catalytic states that are inaccessible to experiments (such as high energy states of low occupancy). In the discussions below on complex I mechanism, such aspects will be explored.

General structure
Respiratory complex I (Fig. 1) is a large membrane protein found in the inner mitochondrial membrane as well as in the plasma membrane of many bacteria. It couples the two-electron exergonic reduction of Fig. 1. The 14 core subunits of complex I from Thermus thermophilus. The membrane-bound subunits and hydrophilic domain are displayed. Ironsulphur clusters are shown in yellow and pink spheres, together with FMN and a quinone molecule. Phospholipid phosphorus atoms (light blue spheres) mark the membrane boundary. The electron transfer and proton pumping routes are shown in purple and green arrows, respectively. Inset shows highly conserved charged and polar amino acid residues that provide connectivity to the hydrophilic axis in the middle of the membrane domain, together with the functionally critical Tyr87 from Nqo4 subunit. Bilayer is also shown with white sticks and pink spheres, in which the tail of a Q 10 molecule extends into. The dynamic loop of Nqo7 is highlighted in a thicker orange representation (see also Fig. 7 below). quinone from NADH to the pumping of protons across the membrane. This creates an electrochemical proton gradient (Δp) across the membrane, which is used to drive the synthesis of ATP, to transport metabolites and to catalyze ATP-ADP exchange. The bacterial complex I consists of about 14 subunits, which are all essential for enzyme catalysis, whereas the mitochondrial version is much larger in size and comprises up to 45 subunits, including 31 auxiliary subunits, functions of which are not known with certainty [37]. Out of the 14 catalytic (core) subunits, seven are membrane-bound and are tightly buried in the lipidic milieu, whereas the rest reside in the mitochondrial matrix or bacterial cytoplasm (Fig. 1). Interestingly, the seven membranebound subunits of mitochondrial complex I are coded exclusively by the mitochondrial genome (mitochondrial DNA), whereas the rest are imported from the cytoplasm. A large number of point mutations in complex I are responsible for several clinically characterized mitochondrial and neurodegenerative diseases [38][39][40][41][42][43][44][45][46]. The assembly of mitochondrial complex I is very complex and is a highly regulated process. It involves a large number of assembly factors, which are responsible for transportation, metal insertion and appropriate folding of the protein [47,48]. With the advent of complexome profiling techniques, there has been a great deal of understanding in this highly dynamics process [49,50].
Due to its large size and complexity, standard X-ray crystallography techniques were only partly successful in the early era of structural characterization of complex I. Instead, cryo electron microscopy (cryo-EM) based structures were the first to reveal the L-shaped structure of the enzyme [47]. The first major breakthrough took place when the structure of the hydrophilic domain of the enzyme was resolved using X-ray crystallography in 2006 at a resolution of 3.3 Å [51]. It revealed an intricate arrangement of FeS clusters and the electron transfer (eT) path from the FNM (flavin mononucleotide) to the terminal electron acceptor, the N2 FeS cluster (see Section 2.2.2 and Fig. 2). Later on, using state-of-the-art protein purification techniques and X-ray crystallography, research groups of Leonid Sazanov and Ulrich Brandt solved the structures of bacterial and eukaryotic complexes, respectively [32,35]. However, another key milestone was achieved in 2013, when the entire structure of complex I from Thermus thermophilus comprising all 14 catalytic subunits was resolved at 3.3 Å resolution [29]. This structure provided the most detailed insights into the electron and proton transfer (eT and pT) pathways, as well as into the possible coupling mechanism that involves a uniquely structured Q binding site and a functionally critical subunit (Nqo8 in Thermus enzyme) with several tilted transmembrane (TM) helices. Continuing this trend, and complemented with the great advancements in cryo-EM technology, 3D-structures of mitochondrial enzymes have been solved recently, at a sufficiently high resolution, thereby providing atomistic insights into the core as well as the overall architecture of 45 subunit mitochondrial complex I [13,31,34]. The structural information has enormously benefited the (re)interpretation of published biochemical and biophysical data on complex I and has provided a strong impetus to the recent modeling and simulation studies.

Electron transfer
Electrons are quantum mechanical particles that exhibit wave-particle duality. The transfer of electrons in proteins is achieved by the assistance of redox-active cofactors buried in protein matrix. Several different types of cofactors are found in nature, which are responsible for optimizing eT rates, which in turn are necessary for efficient enzyme catalysis. Some commonly occurring examples are, hemes (in complex III), Cu centers (in complex IV), FeS clusters (in complex I), etc. The phenomena of intramolecular eT through the buried redox centers in proteins is relatively well-understood [52]. The rate of eT (k eT ) depends on three key factors; the electronic coupling between the initial (before eT) and the final (after eT) state (H AB ), the reorganization energy (λ) and the driving force (ΔG°), as described below in Eq. (1).
There are quite a few theoretical models that describe the electronic coupling (H AB ), such as the pathways model [53] and a model in which the distance between the redox cofactors is the primary determinant of the rate of eT [54]. The reorganization energy is the energy required to change the system from the initial to the final state without actual eT taking place. In the Moser-Dutton model, a generic value of~0.7 eV is suggested [54]. However, for highly solvated proteins such as the peripheral arm of complex I (Fig. 1), this value may deviate to larger values (see ref. [55]). The driving force (ΔG°) is known from equilibrium redox potential measurements, albeit in the "ground state" of the enzyme. Here, time-resolved spectroscopic measurements can sometimes allow estimation of the redox potential during enzymatic turnover [56]. It would be of high mechanistic importance to determine (C) Water-protein-based connection from the bulk to His169 shown with an orange mesh. Water molecules are shown with purple spheres. A highly conserved D408 (Nqo4 subunit) is a part of the connection.
"operational" potentials of various redox couples involved in complex I turnover.

Electron transfer in the peripheral arm; FMN and FeS chain
The eT machinery in complex I is unique in that it resides in the hydrophilic domain of the enzyme in its entirety. This is different from all other respiratory and photosynthetic complexes, which harbour redox cofactors bound to the membrane subunits. Fig. 2 shows the eT path in complex I, which spans 70-80 Å. The X-ray data revealed the possible binding modes of NADH, next to the first electron acceptor; flavin mononucleotide (FMN) [57], stabilized by stacking interactions. The vicinity (~3.4 Å) of donor and acceptor suggests an adiabatic electron transfer with low driving force (E m,7 of NAD + / NADH~−320 mV and FMN/FMNH 2~− 300 mV). Electrons transferred to FMN from NADH via a rapid hydride (H − ) transfer [58] are subsequently passed on to the one-electron carriers, the FeS clusters. There are about 7-9 FeS clusters found in complex I, which are of the type; 2Fee2S or 4Fee4S ( Fig. 2A). It is well-known that the electronic properties of FeS are strongly perturbed by the surrounding protein, as shown in recent ultra-high resolution structural study [59].
Electron transfer theory predicts a fast electron tunneling in the redox chain of complex I [54], which has been verified by the experiments [60,61]. Upon oxidation of FMN (in E. coli enzyme), the two electrons end up on two high-potential FeS clusters, one on N1a cluster, which is segregated from the main eT pathway, and the other on the N2 cluster, which is the direct electron donor to bound ubiquinone or quinone ( Fig. 2) Though, in mammalian enzyme, this electron bifurcation is not observed, owing to apparently different redox potentials of the FeS clusters (N2 and N1a).
It is interesting to note that the first eT (from FMN to N2) occurs in sub-millisecond timescales via pure electron tunneling with no tight coupling to proton pumping, which occurs an order of magnitude slower and also without any significant conformational changes, suggesting that the electron transfer through FeS chain is not directly linked to energy transduction. The long chain of metal centers (Fig. 2) serves what purpose, remains currently unknown, but it may be necessary to make eT to Q kinetically efficient, and maintaining low-levels of ROS production at the FMN site [62]. Moreover, it could be advantageous to segregate the ROS producing FMN center from the N2/Q redox chemistry for reasons not yet fully understood. Computational investigations have provided microscopic insights into the eT through this route, and revealed the possible roles of water molecules in efficient eT, in particular in achieving the rates commensurate with the overall turnover of the enzyme [63,64].

N2 cluster
Until the arrival of structural information, the location of the highpotential FeS cluster (N2) cluster remained controversial [65]. And, this had large repercussions on the proposed molecular mechanisms. Now that the position of N2 cluster is known, and given that the eT from NADH to N2 is fast (sub millisecond), it is generally considered that the proton pumping in complex I is not directly linked to the oxidoreduction of N2, but it is the transfer of electrons to bound Q that drives the pump. In this work, only electron transfer reactions downstream to N2 will be discussed, whereas upstream eT (NADH to N2) will be briefly mentioned, even though the latter may be physiologically important.
N2, a Fe 4 S 4 cluster, has a total charge of −2 or − 3 (including cysteinate ligands), when oxidized or reduced, respectively. However, structural data show that the cluster is surrounded by conserved positively charged residues (Fig. 2B), which would result in a total charge of the region to be neutral when oxidized, and −1 when reduced. Interestingly, a highly conserved residue His169 from Nqo4 subunit has been suggested to act as a redox-Bohr group [66], which would suggest that upon reduction, the charge of the region may stay neutral due to proton uptake from the N-side of the membrane. When fluctuations are induced in the crystal structure by means of MD simulations, water molecules and polar amino acid residues are found to mediate a connectivity between the bulk N-phase and His169 (Fig. 2C). Such a path may be responsible for the reduction (of N2) coupled protonation of His169, supporting its role as a redox-Bohr group [66]. However, the recent data from Hyperfine EPR spectroscopy reveal that His169, though protonatable upon reduction of N2 cluster, is not a strong redox-Bohr group, and the coupling between electron and proton is nonstoichiometric, therefore unable to allow an efficient energy transduction [67]. This is in agreement with the earlier reports [66], which showed that the redox-Bohr group is not required for proton translocation, in turn supporting the current viewpoint that it is the redox reaction of Q that drives proton pumping in respiratory complex I.

Proton pumping
Protons are approximately 1800 times heavier than the electrons. Therefore, they cannot tunnel as long as the electrons do. However, proton transport occurs over longer distances, especially in photosynthetic and respiratory proteins, in which hydrated cavities and ionisable or polar amino acid residues play a critical role in proton diffusion via Grotthus mechanism [68]. The pT can occur either in discrete steps, or in a concerted or semi-concerted fashion on a pre-aligned chain of water molecules, where the latter is known to be exceptionally fast [69]. Recent XFEL [70] and reactive molecular dynamics (MD) [24] data shed light on how hydration-coupled proton transfer drives proton pumping in enzymes.
Proton pumping in complex I has been studied using a range of experimental techniques [71][72][73], and according to the current consensus, the proton pumping stoichiometry is 4H + /2e − . However, it is partly at odds with the number of homologous subunits that seem to participate in proton pumping (see Section 2.3.1), which is three. It is to note that such an asymmetry is not uncommon in respiratory enzymes. For instance, the mammalian F 1 F o ATP synthase has a non-integer H + / ATP ratio (8/3), which probably suggests that the enzyme has evolved to conserve energy in most efficient way possible. On a similar note, it is possible that respiratory complex I is capable of shifting proton pumping stoichiometry depending upon the physiological state.

Proton pumping antiporter-type subunits
Based on a number of site-directed mutagenesis studies, the three subunits Nqo12, Nqo13 and Nqo14 have been suggested to translocate one proton each [74][75][76][77][78], although it is not known for certain. All three subunits are homologous to the Mrp (Multiresistance and pH adaptations) Na + /H + antiporters [79][80][81], and show a conserved stretch of charged residues that lie in the middle of the subunits (Fig. 3). Interestingly, these charged residues follow a fixed pattern, and enable connectivity through the entire membrane arm of complex I [29,82]. Proton transfer pathways have been predicted based on the crystal structure data [29,30,33] (Fig. 3 and Table 1). Even though the path towards the P-side of the membrane, and connection in the middle of the membrane seems conserved in complexes from different organisms, it is the proton uptake route that remains unclear. In T. thermophilus enzyme, the latter is formed by residues from TM helices 5 and 7, whereas in Y. lipolytica complex I it has been suggested to form around TM helices 7 and 8 [33]. This apparent contradiction is yet to be resolved. However, overlay several structures of complex I as well as simulation data provide a consistent picture, as discussed below (see also Fig. 3).
Classical MD simulation studies on the membrane domain of E. coli complex I (minus NuoH subunit) have been pivotal in providing atomistic insights into proton channels [83,84]. Simulation data revealed rapid hydration of the membrane-bound subunits, and formation of water-based networks through which protons can be pumped from the N-side to the P-side of the membrane. An interesting suggestion that emerged from these simulations is the "gating" of protons via hydration-dehydration transitions of the antiporter-type subunits [83].
Recently, large scale atomistic MD simulations have been performed on the entire structure of complex I from Thermus Thermophilus [85] comprising~800,000 atoms. These microsecond simulations of complex I in full membrane-solvent environment are truly state-of-the-art. Data from these simulations reveal hydration patterns that partly overlap with the proton channels predicted from crystal structure analysis [29,30,33] as well as earlier simulation data [83]. The residues that participate in proton transfer pathways are given in Table 1. Interestingly, almost all the residues proposed are conserved in the structurally characterized enzymes, further consolidating the channel prediction from atomistic simulations. Moreover, the data also point out some new candidates for site-directed mutagenesis studies (Table 1).
In another simulation study of complex I from Thermus enzyme [86], the hydration of antiporter type subunits was studied in atomic details. There are significant similarities seen in the two recent simulation studies, primarily being the hydration of the antiporter-type subunits occurring within first hundred ns. The simulations also point to a unifying picture in which proton uptake paths close to the N-side of the membrane are formed somewhat in the middle of the subunits (TM helices 7 and 8), and most likely do not involve the characteristic Lys/ Glu pair present in each antiporter-type subunit. Furthermore, several residues that participate in channel formation are common to both studies, thereby consolidating the proton channel predictions (see Table 1). However, despite these significant similarities, some differences exist, as discussed below.
In ref. [86], authors observed the dissociation of the backbone Hbond between His211 and Leu214 from Nqo13 subunit, and proposed it to be a gate for channel opening. In contrast, the H-bond remains stable in the 500 ns time scales [85]. The residue His211 from Nqo13 is highly conserved, is located in the bulge of TM helix 7, and H-bonds to the Asp557 of the horizontal helix of subunit Nqo12 (Fig. 4). In MD simulations [85], His211 undergoes a conformational change in which it dissociates from Asp557, flips towards the interior of the subunit (Fig. 4), and participate in putative proton uptake route. We envisage that this conformational flexibility of His211 is necessary for function, because mutation of this residue in E. coli enzyme to a basic amino acid such as arginine or lysine [87] has a major effect on the activity, most likely due to the lower rigidity of the region upon ion-pair formation with Asp557 (Asp563 in E. coli, see also ref. [88]). A similar role is envisaged for His193 of Nqo14 subunit, and multi-scale simulation approaches are necessary to understand these aspects to atomic resolution. Simulation data also show the sidechain of Lys216 (Nqo14) to undergo conformational changes; a notion also supported by the structural data on complex I from various organisms (Figs. 3 and 4). The flip of the sidechain seems to provide the required water/protein-based connectivity between the Lys-Glu pair (Figs. 3 and 4), and the rest of the proton channel, necessary for horizontal proton transfer and proton pumping (Section 5).
The three antiporter type subunits may pump one proton each, which would be equivalent to a proton pumping stoichiometry of 3H + / 2 e − , predicted to occur under high membrane potential [89]. However, it has been convincingly shown that complex I pumps 4H + under zero pmf conditions [71][72][73]. Therefore, a fourth unique proton channel must be present, unless one of the antiporter-type subunit pumps more than one proton during the catalytic cycle. In an earlier simulation study on E. coli enzyme [83], the fourth proton channel was predicted to form around residues from NuoJ (Nqo10) and NuoK (Nqo11), which is in agreement with the interpretation of crystal structure data [30,33]. However, in the recent work [86], it was proposed that Nqo8 (NuoH) subunit harbors the fourth proton channel, a notion also supported by the crystal structure data from Thermus enzyme [29]. Therefore, these seemingly contradictory results need further work to be sorted out. However, in contrast to these interpretations, our recent work [85] provides a scenario in which a highly hydrated Nqo8 subunit functions primarily as the coupling element (see below), and the fourth proton channel is in the same location as predicted earlier by crystallographic and simulation approaches, respectively [30,83]. In the scenario currently favored by us, each of the three antiporter-like subunits pump one proton, and the fourth proton is pumped through the Nqo10/11 interface in tight coupling with the redox/protonation reactions in Nqo4/6/8 subunits, which are discussed in much more detail below (Section 5).
The explicit proton transfer was also simulated in the membranebound subunits using hybrid QM/MM (quantum mechanical/molecular mechanical) MD techniques [83]. A rapid transfer of proton from the Nside of the membrane to the conserved and functionally important Glu67 of the Nqo11 subunit (Glu72 of NuoK) was observed, which resides next to another highly conserved and functionally critical Glu32 (Glu36) from the same subunit. Interestingly, Glu32 has been found to undergo conformational change in the recent simulations, a notion also shared by the structural data from bovine enzyme [34], and seems to form the key site in the fourth proton translocation channel. A similar QM/MM analysis of Nqo13 subunit also revealed rapid proton transport to the conserved Glu377 from Lys235 [86]. In agreement with this, the total electric dipole moment of the cluster of water molecules indeed aligns to promote Grotthus-type proton transfer towards Glu377 [85].
Despite the well-known drawbacks associated with conventional simulation approaches such as insufficient sampling or fixed protonation states, these provide detailed insights that are difficult to achieve through experiments, and allows one to delineate proton transfer pathways, which are very difficult to predict from static structural data Fig. 3. Highly conserved residues in proton pumping subunits Nqo12 (A), Nqo13 (B) and Nqo14 (C). Backbone architecture is from T. thermophilus complex I structure, whereas the magenta, cyan and green residues are from Thermus, bovine and Yarrowia lipolytica complex I, respectively. Black arrows indicate the putative proton transfer pathways from N-to the P-side of the membrane. [90]. The synergistic approach combining simulations and experimental data is highly beneficial, in particular for the emerging field of research on complex I.

Inhibitors of complex I
It is necessary to appreciate the significant amount of work done in the field of complex I based inhibitors [91][92][93]. A rather large number of natural or synthetic compounds are known to inhibit complex I reactions with varying degrees, including insecticides and acaricides [94]. It has been suggested that several different types of complex I inhibitors bind in an overlapping fashion in the binding cavity [95], a notion also corroborated by the site-directed mutagenesis data [96,97]. However, their binding sites as well as the mechanism of inhibition are not known with certainty. Given the central role of complex I in mitochondrial bioenergetics, as well as its association with several mitochondrial and neurodegenerative disorders, it would be important to obtain molecular insights into enzyme-inhibitor interactions, which may provide a rationale for structure-based drug design.
Overall the data from inhibition assays suggest the presence of two different inhibitor binding sites in complex I [91], which also led to the viewpoint that there are more than one Q-binding sites involved in biological energy conversion. One of the most well-characterized inhibitors of complex I is Rotenone. It is known to inhibit the downstream eT from N2 (FeS) cluster to Q, suggesting that it may occupy the Qbinding site(s) preventing enzymatic turnover. Crystal and cryo-EM based structural data reveal an unusually long and narrow Q-binding cavity, not seen in other respiratory or photosynthetic complexes [29,33,34]. Given the tight space in the Q-binding site, it is not easy to envisage how bulky molecules such as Rotenone would bind without significant structural (and possibly functional) distortion of complex I. Moreover, it is known that Rotenone has a bent structure [98], and for it to bind in the active site it would require rather large conformational changes, raising the possibility that the site may completely loose its structure and function upon binding. We suggest that Rotenone does not enter the long tunnel-like Q cavity of complex I, instead seals the mouth of the Q-channel by binding in a specific conformation that is congruent with its low K i (~nM range) (Fig. 5A). In agreement with this view, it has been suggested that Rotenone binds to the ND1 subunit [99]. Moreover, addition of phospholipids to the enzyme preparation increases the affinity of Q at the Rotenone sensitive site [100,101], Table 1 Amino acid residues that participate in proton channel formation in three antiporter-type subunits Nqo12-14. The residues in bold are based on MD simulation studies by Haapanen and Sharma [85]. Residue numbers in italics corresponds to the ones also found in ref. [86]. Residues numbering correspond to the following organisms; Tt -Thermus thermophilus, Yl -Yarrowia lipolytica, Oa -Ovis aries, Bt -Bos Taurus, Ec -Escherichia coli.
Piericidin is a strong competitive inhibitor of complex I (K i in nM range) [92], and its binding site close to the N2 center has been suggested based on structural data [29]. This location is also supported by the fact that Val407Met mutation in Rhodobacter capsulatus enzyme is known to decrease Piericidin sensitivity of complex I by about 30 times without affecting the deamino-NADH-ferricyanide oxidoreductase activity [102]; Val407 (Val403 in Thermus enzyme) is located close to the proposed binding site near N2 (Fig. 5B).
Classification of complex I inhibitors led to the suggestions that Rotenone and Piericidin A bind to different locations. This notion also led to the proposal of multiple Q binding sites in complex I [92], such that there are two Q binding sites, one close to the N2 center (Piericidin binding site), and the second where Rotenone binds. Interestingly, neither Rotenone nor Piericidin inhibit complex I from E. coli [103], which could be due to unusually tight binding of a Q molecule. Indeed, it is known that purified E. coli complex I contains one tightly bound quinone molecule [104], the binding location of which remains unknown. Given that both inhibitors fail to inhibit E. coli complex I, it can be proposed that a tightly bound Q resides at the Rotenone binding site or prevents access to both inhibitor binding sites.
Miyoshi and co-workers have synthesized a diverse range of complex I inhibitors by using state-of-the-art chemical techniques [105]. For the bulky inhibitors that bind next to the site near N2 cluster, a rather large conformational change has been proposed to take place [106]. Fenpyroximate, which was earlier thought to bind the distal end of complex I (ND5/Nqo12) subunit [107], was later reassigned to localize at the junction of PSST and 49 kDa subunits, that is next to segments; Ser43-Arg66 (Asn34-Arg57) and Asp160-Arg174 (As-n139-Arg153) of PSST (Nqo6) and 49 kDa (Nqo4) subunits, respectively [108] (Fig. 5B). The quinazoline-based inhibitors are found to fasten next to the residues Asp41-Arg63 of bovine enzyme, which corresponds to the residues Met26-Gly31 and Gly39-Arg42 in Nqo4 from T. thermophilus [109]. The putative binding location is shown in Fig. 5B, next to the Q-binding cavity and close to the region where crucial redox-reactions take place. The two different functional moieties of long molecules such as acetogenins; γ-lactone ring and bis-THF are proposed bind different regions of ND1 subunit (Val144-Glu192) and between 4th TM of ND1 and Arg195 [110,111]. Amilorides, on the other hand [112] attach next to Thr25 -Glu115 segment of Nqo4, and may approach from the side other than the crystallographic Q-binding site suggesting that the cavity is not fully enclosed. The latter point raises an important question on the accessibility of the Q-tunnel from directions different than the one observed in crystal structure.  Interestingly, recent coarse-grained MD simulation data on Photosystem II suggest the presence of three substrate transport channels through which a plastoquinone molecule may access the Q B site [113], and similar scenario can be envisaged for complex I too.

Q-tunnel
The Q-tunnel of complex I is a unique architecture among all known respiratory and photosynthetic complexes that utilize Q or its analogues as a substrate. It is 30-35 Å long, connects the lipidic milieu with the protein interior, and easily accommodates majority of the long-tailed Qs [29]. Even though static structural data points to a single tunnel-like architecture in complex I, it is possible that there may be other cavities that transiently open (and close) during enzyme function complemented by conformational changes in Nqo4/6/8 subunits. The Qtunnel opens right next to the lipid head groups, a region in which Q head group has been found to populate based on free energy simulations [10,11]. It is interesting to note that in enzymes that utilize Q as a substrate, the Q binding sites are predominantly found in this region of the lipid bilayer, thereby maximizing the local occupancy of the substrate [11].
The interior of the Q-tunnel in complex I is uniquely designed such that one side is lined with charged residues, whereas the other half is significantly hydrophobic, which was earlier suggested to be a key factor responsible for Q dynamics in the cavity [32]. This unusual architecture in which a Q molecule would diffuse back-and-forth during enzymatic turnover strongly supports an important mechanistic role of this exceptionally long tunnel, and the diffusion of Q molecule in it. The Q-tunnel takes a turn at a distance of ca. 25 Å from the side chain of Tyr87 (Fig. 1), a key residue in the active-site of complex I. This bending in the Q-tunnel puts a constriction in the cavity, as observed in some of the recent structures [34], and may support a viewpoint that there can be a permanently trapped Q molecule in the enclosed cavity. The observed kink in the Q-tunnel is roughly at the same region where a Q molecule was modeled (20-25 Å from Tyr87 of Nqo4) in a recent simulation study [85], and it is also closer to the functionally critical "Echannel" that connects to the middle of the membrane-bound subunits (see also Fig. 1). This proximity of Q-tunnel to the middle of the proton pumping subunits is likely to be of functional importance, as discussed below (Section 5 and 6).

Quinone binding, reaction and dynamics
The precise binding mode of quinone, which would be biologically relevant for efficient electron transfer from N2, has not been resolved in any of the available structures. Though crystal structures of Thermus and Yarrowia enzymes have revealed certain binding modes; a feat achieved by soaking the crystals with the solution of quinone-based derivatives [29,33]. However, it is to be emphasized that the placement of Q-derivatives in the cavity may yield binding sites or modes that are probably different from the natural mode of Q binding.
Even if the binding arrangement is not known, the binding of a Q molecule next to the N2 cluster is necessary for complex I turnover. Molecular simulations have been particularly useful in identifying possible binding modes of Q at the site near N2 cluster in a dynamic context [114,115]. In these studies, the Q was modeled in its various redox and protonation states (oxidized Q, QH 2 and QH − ), and in one possible arrangement the oxidized Q head group is hydrogen bonded to Tyr87 and His38 of Nqo4 subunit. This arrangement, earlier proposed based on crystallographic and biochemical experiments [29,116,117], was found to be important for the proton-coupled electron transfer (PCET) reaction that takes place upon two electron reduction of Q (from N2). Subsequent, classical MD simulations of hundreds of nanoseconds revealed short-to-long ranged conformational transitions in the Nqo4/ 6/8 subunits that occur after the PCET reaction [85,114]. Though, much longer timescales would be needed to understand binding and unbinding events of Q from this site near N2, simulation of redox/ protonation states that are inaccessible to experiments provide impeccable insights into the putative molecular mechanism of complex I.
Based on the binding of Q near the N2 center, we may describe possible reactions that could occur during stepwise Q reduction from N2 FeS cluster. Two possibilities exist after the first-electron transfer to bound Q at the site near N2, either formation of a neutral or an anionic semiquinone species (SQH ⁎ and SQ −/⁎ , respectively). Earlier calculations showed that the transfer of first electron is not coupled to proton transfer from the nearby residues (protonated His38 and Tyr87 of Nqo4), suggesting that the formed semiquinone species is anionic (SQ −/⁎ ) in nature [114]. A negatively charged semiquinone species is likely to interact strongly with the protonated His38 (and/or neutral Tyr87), which may be necessary to keep SQ bound until the second electron transfer has occurred. If a neutral SQ species forms upon coupled proton transfer either from protonated His38 or Tyr87, this may weaken the binding, leading to its release in the environment, which would be lethal to protein and other cellular components.
The second electron transfer to anionic SQ has been found to be dependent on proton transfer, as shown by the state-of-the-art quantum chemical calculations [115]. This electron transfer again leads to two possible states; QH − and QH 2 formed after one or two proton transfers from protonated His38 and neutral Tyr87 of Nqo4. In fact, earlier calculations as well as experiments suggest both species (QH 2 and QH − ) are capable of catalyzing proton pumping, albeit with different kinetics [114]. In Section 5 below, mechanistic aspects are discussed in terms of the catalytically active QH − species.
Wikström et al. [20] formulated a mechanism in which a Q molecule shuttles between two binding sites within the Q-tunnel. The first Qbinding site is within eT distance from the N2 center, and the second site has been proposed based on inhibitor binding studies, somewhere in the middle of the Q-tunnel. The possible binding of a Q molecule to such a site was recently explored through classical simulations [85], and it was proposed that proton(s) released upon quinol oxidation may be transported to the membrane-bound subunits, thereby, strengthening the coupling between the redox reaction and proton pumping (see Section 5.1). Although mechanistically different from the above, a two-binding site model of proton pumping was also proposed based on the structural data [33] (see also [116]). In this model, it was hypothesized that the two catalytic states (E and P) may correspond to the active (A) and deactive (D) forms of the enzyme, and an inter-conversion between the two is achieved by a coordinated movement of three loops from Nqo7, Nqo8 and Nqo4 subunits (see also Section 7).
Recently, Fedor et al. studied the kinetics of bovine complex I using quinones of various tail lengths [118]. They found that Q 10 has a higher binding affinity, and also faster binding and unbinding rates, and suggested that the dynamics or diffusion of ubiquinone is not rate-limiting (but fast) in the mechanism of complex I [118].

Coupling mechanism
Decades of biochemical and biophysical work on complex I led to a multitude of mechanistic proposals, which have been nicely summarized in an earlier work [119]. From late 1990s to early 2000s, some new ideas were circulated such as based on the Q-cycle mechanism of cytochrome bc 1 [120] and indirect conformational coupling mechanisms [121,122]. However, after the availability of first crystal structure of complex I from T. thermophilus [51] in 2006, many mechanistic proposals were revised [65] to account for the fact that no fixed redox cofactor is present in the membrane domain of complex I. A series of recent reviews highlight the current state-of-the-art in the field [20,62,82,[123][124][125][126][127]. The prevailing viewpoint is that the N2 cluster does not participate in energy conservation, however, its relatively high redox potential makes sure that it is always occupied with an electron during the turnover even though its high redox potential is not critical for the proton pumping activity [66].
Since N2 is the direct electron donor to bound Q, it is possible that during enzyme catalysis a semiquinone (SQ) species may not populate at the latter site. This also means that His169 from Nqo4 subunit, which has been identified as the redox-Bohr group, would remain partly protonated. Since, the head group of bound Q (modeled) and N2 cluster are not far apart (10-12 Å), the former will feel a constant electric field from the N2 region, which may be important in subtle PCET reactions (Q → SQ → QH − /QH 2 ) and quinone dynamics [20].
Early on it became apparent that there may be more than one Q reactive sites in complex I [91,92,128], which may play a role in proton pumping. The proposal was further strengthened by the data from EPR measurements, which showed two semiquinone species (SQ Nf and SQ Ns ) that responded differently to well-known complex I inhibitors and to the membrane potential [129]. A third species has also been characterized (SQ Nvs ) [130]. The fact that certain preparations of complex I contain stoichiometric amounts of tightly bound ubiquinone [131], as well as evidence from the inhibitor binding studies, the idea that complex I functions by involving more than one Q sites cannot be slighted. However, despite this, the binding sites are not known with any certainty.
Based on site-directed mutagenesis and EPR measurements, a mechanistic proposal was delineated by Euro et al. indicating that it is the of the charge of one or two-electron reduced Q species that drives the proton pumping by means of conserved lysines in the middle of the membrane [76]. Sharma et al., using multiscale computational approaches and classical simulations on the entire structure of complex I, found that the two-electron reduction of bound Q at the site near N2 results in abstraction of protons from two nearby conserved residues, and this triggers the conformational change in the active site, which resonates all the way to the middle of the membrane bound subunit Nqo8 [114]. The simulation data, which is supported by structural information, suggests that both electrostatics and conformational coupling are at play in the molecular mechanism of complex I.

Electrostatic and conformational coupling
Wikström's and Verkhovskaya's groups provided a more explicit version of the mechanism [20,82] proposed earlier by Euro et al. [76]. However, one key question remains; is there a permanently trapped Q in the long-tunnel of complex I or if a single Q molecule shuttles in and out of the cavity, thereby coupling to proton pumping. There are arguments for and against each proposal. However, the work under progress in our group provide some insights that may be more in favour of the Q-shuttle mechanism. Our tentative mechanistic proposal or working hypothesis, which is largely based on the proposals by Wikström et al. [20], is displayed in Fig. 6, and can be described as follows: in state I, the oxidized Q (red) binds at the Q-binding site near center N2, whereas another oxidized Q (green) sits at the putative Rotenone binding site at the lipid-protein interface. The protons equilibrate between the N-side of the membrane and the proton binding sites in the middle of membrane-bound subunits Nqo12-14, and in Nqo10/11 (His241, Lys235 and Lys216 of Nqo12, Nqo13 and Nqo14, respectively, and Glu32/Glu67 of Nqo11). The two-electron reduction of Q (red) by N2 is coupled to local proton transfer (blue) from Tyr forming tyrosinate and QH − in state II [114]. The anionic quinol then move towards another location halfway through the tunnel due to the repulsion from tyrosinate, as shown in state III (see also ref. [85]). The quinol at this position gets oxidized to another Q (green) bound at a distance of < 14 Å [20], coupled with the uptake of two protons from the N-side (pink), as shown in state IV (see also ref. [20]). The proton (black) released upon quinol oxidation in the middle of the Q-tunnel, travels ca. 30-40 Å towards the antiporter-type subunits via protonation-deprotonation events of key charged residues in the middle of membrane subunits Nqo7 and Nqo8 ("E-channel") [85]. This initiates protonation/ deprotonation events along the central axis, and combined with electrostatics (red dotted arrow), the protons loaded on Nqo10/11 and Nqo12-14, are pumped towards the P-side of the membrane (state VI). The "pushing" proton (black) is released to the N-phase, whereas the anionic Tyr87 from Nqo4 is reprotonated from the bulk complemented with the back diffusion of Q (red) to the crystallographic site, as shown in state VI. Overall, four protons are pumped to the P-side of the membrane, and two substrate protons (pink) are consumed to form QH 2 .
The protonation/deprotonation events as well as electrostatics are likely to play a significant role in proton transfer through the antiporter-type subunits, and may be described as follows. The proton (black) neutralizes the anionic Glu32/Glu67 pair of Nqo11 subunit, which leads to the transfer of proton to the anionic Glu112 of the conserved Lys/Glu ion-pair in Nqo14. The neutralization causes the dissociation of Lys186 from protonated Glu112, leading to its sidechain displacement towards the protonated Lys216, as also seen crystal structures (Fig. 3). This induces lowering of the pK a of Lys216 and/or Lys345 leading to the ejection of proton towards the P-side of the membrane. After the horizontal interfacial proton transfer from Lys345 to Glu123 of Nqo13, the process repeats, leading to the pumping of one proton each by the three antiporter-type subunits. The fourth proton is pumped after the partial pump cycle is over, that is during relaxation. In this proposed mechanism, the proton movements in direction parallel to the membrane through the central hydrophilic axis drive the pumping perpendicular to the membrane.
The above discussed mechanism of complex I has some interesting analogies to the now rather well-understood proton pumping mechanism of bacterial cytochrome c oxidase [20]. In this enzyme, the protons to be pumped out to the P-side of the membrane are first loaded on to the so-called proton loading site (PLS) in a tight coupling with the electron transfer to the active site. The subsequent transfer of substrate proton to the active site (also driven by eT to the active site), unlatches the electrostatic coupling between the electron at the active site and proton on the PLS, leading to the ejection of the latter to the P-side of the membrane. Similarly, we envisage that the loading of the "proton loading sites" (His241, Lys235 and Lys216 of Nqo12, Nqo13 and Nqo14, respectively) in the middle of the antiporter-type subunits occur due to their high pKa (> pH of the N-side). Finally, these protons on the PLS are pushed out to the P-side of the membrane, driven by the proton released upon oxidation of QH − , that has formed after "local" proton transfer from Tyr87 (see Fig. 6, and above).
It is worth asking what prevents the leak of protons from the P-side to the N-side or during the catalysis, when they are transiently loaded on to the PLS. Such aspects have been very well studied in cytochrome c oxidase, where it was found that the dynamics of acidic residues [132,133], uncharged but polar residues [134] and hydration and dehydration aspects prevent proton leakage from the P-side to the N-side of the membrane [135]. Similar microscopic gates and valves need to operate robustly in complex I, especially under high pmf, when the chances of back leak are high.

Inter-quinone electron transfer (Q2Q eT)
One of the key components of the above proposed model is the inter-Q electron transfer, which is known to occur in Photosystem II, and has been a subject of intense research [136,137]. Building upon the earlier proposals [20], we envisage that the two electron-exchange sites are uniquely designed to achieve inter-Q electron transfer. At the proposed site in the middle of the Q-tunnel, where a QH − (or QH 2 ) will be oxidized, it is necessary to have a proton acceptor group. This condition is partly fulfilled by the site modeled in the middle of the Q tunnel in the recent computational study [85]. At this site, a number of proton acceptor groups exist, namely, Glu223, Glu225 and Glu248 of Nqo8 subunit. In the earlier work [114], it was proposed that proton pump of complex I may function similar to wild-type with a QH − species that is formed at the site near N2, as a result of two-electron reduction of Q, coupled with a single proton uptake from the highly conserved Tyr87.
Such a species was found to be repelled by the negatively charged Tyr87 towards the putative Q-binding site in the middle of the Qtunnel. QH − at the latter may act as a strong reductant when bound next to acidic residues from Nqo8 subunit (Glu223, Glu225 and Glu248), thereby reducing the barrier for eT to oxidized Q bound at another site (Fig. 6). The oxidation-coupled deprotonation of QH − at the site in the middle of the Q-tunnel may provide the required driving force for proton pumping, which may occur through long-range proton transfer towards the membrane-bound subunit (see Fig. 6, also refs. [20,85]). For the second exchangeable Q-site, it may be necessary to have proton-donor groups nearby, so that the coupled electron transfer from trapped Q to exchangeable Q is highly efficient.
Since, a simultaneous two-electron transfer from QH 2 to an oxidized Q is unlikely [138], such a reaction may proceed step-wise through anionic SQ species to prevent their profligation. It would be extremely interesting to apply EPR spectroscopy approaches to identify signals of spin-spin coupled state between two transiently formed anionic SQ species, which would also strengthen Q2Q eT hypothesis.

Active-deactive transition
One unique property of mitochondrial complex I is to undergo active (A) ↔ deactive (D) transitions [139], which may be of physiological relevance [140]. The D form of the enzyme slowly converts to the A form upon turnover, and when no electrons are supplied or consumed, it reverts back to the D form. Although, D/A transitions may be mitochondrial specific, it appears that the complex I from E. coli shows similar transitions, and in which proton pumping activity is uncoupled from the electron transfer in the resting state of the enzyme [28]. This scenario can be compared with the O H /O forms of cytochrome c oxidase, in which the two forms drastically differ in their capability to pump protons across the membrane, likely due to the subtle differences in conformations and/or in operational redox potentials of the active site [141,142].
The focal point of the A/D transition has been the ND3 subunit (NuoA in E. coli and Nqo7 in T. thermophilus), in which a Cys (not conserved in some organisms) undergoes D-state specific chemical modification [140]. The loop carrying Cys (Ser46 in Nqo7 subunit of the Thermus enzyme) is highly solvated, and our simulation data on bacterial enzyme also shows it to be highly dynamic (Fig. 7). It makes contacts with Nqo8, Nqo4 and Nqo6 subunits (Fig. 7), and also partly penetrates the Q-binding site, suggesting its possible regulatory role.
Recently, Blaza et al. [27] using highly active preparations of mitochondrial complex I suggested that in the D state of the enzyme the Q binding cavity is highly disorganized, especially the loop containing His55 and His59 of the 49 kD subunit (His34 and His38 from Nqo4 in Thermus enzyme), and the segment between TM helices 5 and 6 of the   55) shown in crystal structure conformation (orange, PDB id 4HEA) and after 500 ns MD simulation (yellow). The dynamic loop contacts several regions on subunits Nqo4 (lavender), Nqo6 (green) and Nqo8 (pink). ND1 subunit. In two separate structures of complex I, which are also in the D state, large scale movements have been observed in these regions [31,33]. In classical simulations of Thermus enzyme, these segments have been found to rearrange in a redox-dependent manner [85,114]. Now with the availability of structural data on both D and A states, computer simulations would be of great assistance in providing dynamic as well as energetic insights into the D ↔ A transition.

Role of lipids
Lipids are not just partitioning entities, but play an active role in various cellular processes such as signaling and trafficking [143]. They are also important for the stabilization of membrane proteins in complex environments, and some lipids are inherent components of enzyme structure, such that harsh delipidation techniques can render enzyme without function, as is known in the case of tightly-bound cardiolipins in respiratory complexes [144]. Lipids also play a major role in enzyme regulation, for instance, the effect of cholesterol on GPCR regulation is well-known in which it stabilizes a physiologically relevant conformation of the receptor [145].
The bovine mitochondrial complex I has been found to bind CL molecules, which if removed, leads to lower enzymatic activity [146] (see also [147]). The recent structural information on mammalian mitochondrial complexes consolidates this point, and show some putative CL binding sites [13,31]. In addition to CL, other lipids such as PC and PE are also resolved in the structures. A comparison between structural and dynamic data from atomistic simulations reveals conserved lipid binding sites (Fig. 8). The data in Table 2 presents interactions between cardiolipins and charged amino acid residues (side chains) from several core and accessory subunits for the two structures. It is observed that some lipids bind tightly into the subunit-subunit interface, and some are even buried between the horizontal helix and three antiporter-type subunits suggesting a tight lipid-protein packing (Fig. 8). This is not surprising given the elongated shape of complex I membrane arm, which drives proton pumping as a consequence of redox reactions in the Q binding site; a very tight lipid-protein interface would be extremely important for efficient coupling at longer distances. Coarse-grained simulation approaches will be required to understand CL binding and unbinding to mitochondrial complex I, similar to what has been achieved for complexes III and IV of the ETC [148,149].

Translational aspects
Due to the involvement of O 2 , electrons and protons, mitochondria produce significant amount of ROS, which if not controlled can damage the cellular components such as lipids, proteins and DNA. ROS, such as superoxide (O 2 −⁎ ), peroxide (O 2 −2 ) or hydroxyl radical (OH ⁎ ), rapidly react with the cellular components, initiating a chain-like eT reaction. So far, the view point has been that ROS are bad, and are responsible for aging and a number neurodegenerative disorders. But this crammed thinking is rapidly changing with new data suggesting ROS are in fact an important component of signaling, and a low level ROS is critical for enhancement of life [150].
Among the three proton-pumping ETC complexes, complex I and III are the major producers of ROS. The ROS producing sites in complex III have been known for some time [151,152], and based on multi-scale computational approaches, energetics of these processes have been studied [153,154]. It has been proposed that Q o site primarily produces H 2 O 2 , releasing it in the intermembrane space, which is in contrast to the ROS production (primarily O 2 −⁎ ) by FMN site in complex I in to the matrix, in a fairly well established physiologically relevant forward electron transfer mode [152,155,156]. Though the microscopic details of the latter reaction are not known, but one combined experimentalcomputational study identified how ROS production by FMN site may accomplish [157]. The charged residues surrounding the FMN site may modulate the redox potential of FMN/NADH couple so as to fine-tune the ROS production through this site. Besides the FMN site, there are indications of another ROS producing site, which is functional during the reverse electron transfer mode or under conditions of high pmf. It has been proposed that ROS production at the latter site is due to higher semiquinone formation [152], and has been linked to the unique Q-binding site of complex I (Section 4).
The FMN site and the Q-site are two such regions that can be manipulated by means of external factors such as drug molecules. The Qtunnel is long and narrow, and possible structural differences in this site  between the bacterial and human complexes can be utilized for rationalized structure-based drug design. Given that complex I is a focal point of a number of mitochondrial disorders, designing specific inhibitors or allosteric regulators would be extremely important, and will allow bridging of basic sciences to translational aspects. This viewpoint is complemented by the recent developments in anti-tuberculosis drug development, in which respiratory enzymes have emerged as promising drug targets, such as ATP synthase and cytochrome bc 1 [158]. Now, with the availability of the structure of human mitochondrial complex I, computational and experimental approaches are needed more than ever to identify and exploit druggable aspects of the enzyme.

Conclusions
Respiratory complex I is a nanomachine that couples electron transfer from NADH to quinone, to the pumping of protons across the membrane. Despite the availability of its three dimensional structure, the molecular mechanism of redox-coupled proton pumping is unknown. Given the central role of complex I in mitochondrial pathophysiology, it is necessary to delineate its mechanistic and regulatory aspects in full details. Computational approaches have played a major role in providing atomistic insights into mechanisms of other respiratory and photosynthetic complexes. Now, with the availability of high-quality structural data on respiratory complex I, such approaches are unraveling complex I function, in tight association with the experiments.