From the ‘black box' to ‘domino effect' mechanism: what have we learned from the structures of respiratory complex I

In the respiratory chain, oxidative phosphorylation (OXPHOS) enzymes are responsible for energy production (in the form of ATP) in eukaryotic cells and in many bacteria [1]. In mitochondria the chain consists of four large inner membrane-embedded protein complexes — complex I (CI) or NADH-ubiquinone oxidoreductase, complex II (CII) or succinate dehydrogenase, complex III₂ (CIII₂) or cytochrome bc₁ oxidoreductase and complex IV (CIV) or cytochrome c oxidase [2]. The proton motive force (pmf) generated by the concerted action of these enzymes drives ATP synthase [3]. Over the years we have worked on most respiratory enzymes, studying them either on their own or as part of supercomplexes, in which they are often organised in situ [4]. Our aim is to understand the basic mechanisms of these huge molecular machines, especially complex I. In the course of the 1990s, as the structures of the respiratory enzymes were first determined, it became clear that they all operate on very different principles of coupling between redox reactions and proton translocation – Q-cycle for CIII₂ [5], direct coupling for CIV [6] and rotary for ATPase [7]. However, complex I, being the largest and most elaborate (comprising up to 45 subunits of ∼1 MDa MW in total in mammals) assembly of the chain, remained the ‘block box' for much longer. All that was known about complex I structure was that it has an L-shape formed by the peripheral and membrane arms [8].

We first studied bovine complex I by 2D electron crystallography [9], which was one of the main methods to study membrane proteins earlier but is rarely used nowadays, surpassed by the new single particle EM approaches. Using a simpler bacterial complex I (14 conserved subunits) as a model, we could show that antiporter-like subunits ND5/NuoL and ND4/NuoM (bovine/E. coli nomenclature) form a distal part of the membrane arm, with limited information otherwise due to low resolution [10,11]. At that time we decided to switch to complex I from Thermus thermophilus, as likely more stable enzyme, even though it was never purified before. This has lead to a first major breakthrough in 2006 — structure of the peripheral arm determined by X-ray crystallography [12]. Our extensive efforts on the crystallisation of E. coli enzyme eventually lead to the X-ray structure of the membrane arm [13]. Crystals of the entire E. coli enzyme were obtained but never diffracted to high resolution, however that was fortunately not the case for T. thermophilus. Although T. thermophilus crystals were always twinned, severely complicating structure solution, in 2013 we finally were able to solve the first structure of the entire complex I [14]. Shortly after that the cryo-EM ‘resolution evolution' happened, prompting us to apply these new approaches to mammalian complex I, resulting in the first nearly complete (88% atomic) structure of the mammalian (ovine) complex I [15]. It should be noted that simultaneously published structure of the bovine enzyme [16] was only ∼53% atomic due to lower resolution (i.e. 47% of residues had no side chains).

Thus, by 2016 we knew the structures of bacterial and mammalian complex I, allowing for a range of hypotheses on the coupling mechanism to be developed in the subsequent years. These were mostly based on the long-range conformational changes due to a large distance (up to 200 Å) between electron transfer reactions (happening in the peripheral arm) and proton translocation across the membrane arm [17–24]. The proposals varied in many details in the absence of hard experimental evidence (high-resolution structures with resolved waters) on what happens with enzyme during turnover. Recently we succeeded in resolving such structures first for mammalian [25] and then for bacterial [26] enzyme, allowing us to develop a detailed, robust and universal coupling mechanism of complex I. It turned out to be radically different from previous models, leading to intense discussions but also some emerging consensus, as presented below.

The general architecture of complex I is conserved, comprising fourteen core subunits (in bacteria) equally divided between the peripheral arm (PA) and the membrane arm (MA) (Figure 1A). Eukaryotes contain additional supernumerary (or accessory) subunits, which form a shell around the core (Figure 1B). These (mammalian enzyme has 31 such subunits) are likely not involved in the catalytic reaction but assist with the assembly, stability and regulation of the complex I [15,27].

The seven conserved core subunits of the PA can be assigned to the N- (NADH-binding) and Q- (quinone-binding) modules (Table 1). Here, electrons from NADH are accepted (as a hydride) by flavin-mononucleotide (FMN) and then passed by electron tunnelling along a chain of seven iron-sulfur clusters (ending with cluster N2) to the final acceptor quinone (Figure 1A) [26,28]. Additional cluster N1a lies off-path ‘upstream' of FMN, and its role could be to temporarily store an electron to prevent flavosemiquinone formation and reduce ROS generation [29,30]. The ninth FeS cluster N7, far off the main path, exists only in some bacteria and is probably an evolutionary vestige [31]. Since the largest drop in redox potential along the chain occurs between cluster N2 and the quinone/quinol pair, the crucial energy-releasing step in the reaction is therefore likely to be quinone reduction/protonation [17] or perhaps even its release out of the binding cavity as the potential of the Q/QH₂ pair bound near N2 is likely similar to the N2 potential [19]. This has been confirmed by real-time measurements of electron transfer reactions in the peripheral arm, which also ruled out the involvement of a long-lived semiquinone radical in the mechanism [32]. For each NADH oxidised and quinol produced, four protons are pumped from the matrix to the inter-membrane space (IMS) [33].

In the MA, ND1/NuoH subunit forms part of the PA–MA interface (Figure 1A,B). It is followed by three small subunits, ND3/NuoA, ND6/NuoJ and ND4L/NuoK, which form, together with part of ND1, the E-channel containing key glutamates [14]. The rest of MA is formed by three homologous antiporter-like subunits (ALS), ND2/NuoN, ND4/NuoM and ND5/NuoL, containing 14 conserved transmembrane helices (TM) each. A central axis of charged residues connects these subunits, suggesting a way of long-distance communication along the MA (Figure 1C) [14,15,26].

ALS share a cation/proton antiporter MRP fold [34] of two inverted symmetric five TM repeats forming one half of the putative proton translocating channel each, exposed either to the matrix/cytosol (N-terminal repeat) or to the IMS/periplasm (C-terminal repeat). Both halves contain a key conserved lysine LysTM7/12 (or GluTM12 in ND4), connected in each ALS by a central LysTM8 (or HisTM8 in ND5) into apparently a full channel across the membrane. In each ALS, LysTM7 forms an ion pair with a conserved GluTM5. All of these key residues sit on breaks in TM helices (loops in TM7/TM12, π-bulge in TM8), which may render the central hydrophilic axis flexible. This has led to various mechanistic proposals involving long-range conformational changes. We suggested [14,21] that quinone reactions could ‘shift' the central MA axis so that half-channels are exposed either to the matrix or to IMS, resulting in pumping of one proton per ALS and an additional one via E-channel, matching the experimental value [33]. Alternative ‘electrostatic wave' mechanism involved co-ordinated forward and backward waves of conformational changes and charge interactions from quinone site to the tip of ND5 subunit [20,35]. An earlier version of a similar model was based on mutagenesis studies of the conserved residues of the central axis [18]. A ‘two-stroke' model suggested that two pump modules are driven in two conformational strokes generated by stabilisation of the anionic forms of semiquinone and ubiquinol [36]. While some elements of the earlier models, such as the critical role of key lysines, have stood the test of time, others turned out to be incompatible with the recent cryo-EM results [25,26,37].

Already the first mammalian complex I structures revealed that it can exist in two major structural states [15,16], which we called ‘closed' and ‘open'. Initially we referred to them as such [15] due to apparently larger angle between the two arms in the open state, although it rather reflects a sideways tilting (with some twisting) movement of the PA (Figure 2B). One of the main differences between the states is a tight, fully enclosed (except for narrow exit into the lipid bilayer) quinone (Q) cavity in the closed state and a wider, open also to the matrix, cavity in the open state. A similar open-to-closed state transition in the E. coli enzyme (and also in Chaetomium [38] and Drosophila[39]) reflects instead of tilting mostly a twisting motion of PA (open-to-closed anticlockwise when observed from the PA tip, Figure 2B). Therefore, instead of referring to the PA–MA angle, the closed/open state terminology is still applicable to all species by referring to the state of Q cavity.

Q binding site is a long (∼25 Å) channel at the PA/MA interface, with a very narrow entry point from the lipid bilayer at one end and a headgroup-binding site at another, deep end, near cluster N2 [25]. The cavity is lined by flexible loops: 49 kDa/NuoCD β1–β2 loop, ND1/NuoH TM5–6 loop and two PSST/NuoB loops (Figure 2C). ND3/NuoA TM1–2 loop seals the cavity from the matrix and embraces PA/MA interface. Most of these loops are disordered in the open state, leading to a wider channel, which facilitates quinone movement into the cavity at the start of the catalytic cycle and quinol movement out at the end. Opening to the matrix facilitates these movements further by allowing waters to come in and out as necessary [26]. On the other hand, in the closed state all these loops re-order, so that the Q channel becomes narrow, tightly engulfing bound quinone. Matrix opening gets closed, so that two substrate protons needed for protonation of reduced Q can come only from protein (nearby 49 kDa/NuoCD His/Asp pair) and not from bulk water.

Further changes happen in the E-channel. Highly conserved ND6 TM3 in the open state has a π-bulge in the middle of helix, where key Y60 (ovine residue numbering in this section as some features are specific to mammals) interacts with other polar residues of the central axis. In this conformation, conserved large hydrophobic TM3 residues block the path for proton transfer along the axis, which is otherwise uninterrupted all the way from Q cavity to the tip of ND5 (Figure 1C). Remarkably, in the closed state half of TM3 rotates almost 180°, so that helix straightens, π-bulge disappears and its hydrophobic residues move out of the path (this feature was thus recently referred to as a π-gate [38]). This fills in the gap with waters so that continuous proton pathway is established along the entire central axis (Figure 1C). This is complemented by a dramatic flip of the conserved _ND1Y142 (prefix indicates subunit) from its outward-facing conformation (open state) into the central proton path (closed state). This switch of tyrosine is conserved in all known open/closed structures from different species and is an easily recognisable indicator of the enzyme state.

Overall, these open-to-closed changes are highly conserved between the species, e.g. ovine and E. coli (Figure 2C) with only slight variations. For example, 49 kDa β1–β2 loop in closed states is always retracted from the cavity to allow Q to bind in the deep site (Q_d) near N2. However, in the open state, this loop can be either extended into the cavity (E. coli turnover, ovine NADH reduced) or can be disordered (ovine turnover). For open-like states in other species, in Yarrowia [37] it appears to be retracted, but is extended in NDH complex from cyanobacteria [40,41]. Therefore, in the closed state β1–β2 loop appears to be always retracted, while in the open state it is more variable, with tendency to be either extended [26] or disordered (but likely still occupying the deep end of the cavity as Q_d is not observed [25]) under turnover conditions. The ND1 TM5–6 loop is always ordered in a very similar conformation in all known closed state structures, while its degree of disorder in the open state can vary (Figure 2C). The first PSST loop (residues 48–51) becomes a β-strand in the open mammalian enzyme [25] (and also in Chaetomium[38]) but remains a loop in bacteria [26] (Figure 2C). The second PSST loop (residues 75–80) flips in the open mammalian enzyme, so that conserved R77 completely changes orientation, which does not happen in bacteria. The rotation of _ND6TM3 is accompanied by large movements of TM3–4 loop, which differ in detail between bacteria and mammals (Figure 2C) but the loop still communicates with key ND3 TM1–2 loop in all cases.

There are two absolutely conserved features in _ND6TM3 and ND3 loop, which clearly define either open or closed state in all studied species. Closed state can be defined as the one without π-bulge in _ND6TM3 (thus a connected central axis) and with ordered ND3 TM1–2 loop (thus enclosed Q cavity). Open state can be defined as the one with π-bulge in _ND6TM3 (thus disconnected central axis) and with disordered central part of ND3 TM1–2 loop (thus opened to matrix Q cavity). Both of these features are central to the coupling mechanism as discussed below. Additionally, in the closed state all the key loops are ordered forming tight Q channel, while in the open state cavity-lining loops may be disordered to a different degree in different species, leading to enlarged on average Q cavity. It should be noted that key loops as built in the deposited PDB models for a range of species are not always well defined by cryo-EM density, so one has to consult the deposited maps to make conclusions on the conformation of these loops.

The ordering of key loops is ensured by many conserved salt bridges and polar interactions formed in the closed state, which are all broken in the open state (Table 2). Most of these interactions are conserved from bacteria to mammals, consistent with their key mechanistic role. One interesting difference is that in mammals the PSST loop flips, bringing conserved R77 to interact either with shallow site (Qs) -bound quinone in the closed state or with key _49 kDaD160 (from the Q-protonating pair) in the open state [25]. In bacteria the homologous R87 does not flip, but it also interacts with similarly positioned Q in the closed state, suggesting conserved interactions in this intermediate Q site. Native quinone with its long tail can bind there only temporarily on the way to the deep site, while short-tail quinones used in cryo-EM experiments can occupy both sites simultaneously [25,26].

Until recently mammalian complex I was unique by showing closed and open states in apo conditions (meaning here as purified, no substrates added) [15,42]. It is also unique by showing a transition into the so-called deactive state with high-energy barrier, requiring prolonged incubation at elevated temperature (30–37°C) without substrates [43]. Deactive state resembles open state as all the key loops are disordered and there is a π-bulge in _ND6TM3 [42]. Due to this similarity it was suggested that deactive and open state are one and the same [44]. However, ‘as purified' ovine enzyme shows mostly open state in cryo-EM analyses and yet it does not show any time delay in developing activity after addition of substrates [25]. This delay is a defining characteristic of the deactive state, as enzyme needs several cycles of slow turnover to recover from deactive conformation into active state. Thus, fully active mammalian enzyme can have various proportions of open/closed states (from mostly open to mostly closed) depending on species and purification procedure [45].

We have shown that true deactive state differs from open state by the complete relocation of ND6 TM4 [25]. It should be noted that our deposited maps for deactive states (differing by PA–MA angle) were post-processed with optimal B-factor for the bulk of molecule. Less well-ordered peripheral parts are better visualised by applying blurring B-factor of about 100 (e.g. in Coot) or by filtering density to ∼4 Å. Then the relocated ND6 TM4 density becomes very clear (e.g. PDB IDs 6ZKU, 6ZKV). The C-terminus of the lateral ND5 helix and subunit B14.7, in the vicinity of ND6 TM4, are mostly disordered in the deactive enzyme. This is however also the case for some of the more open classes of the active enzyme, including apo conditions. Together with the fact that deactive enzyme can be re-activated this suggests rather local disorder in this area in the most open (i.e. with the largest PA-MA angle) classes and not a loss of subunits/degradation. However, the relocation of ND6 TM4 is unique to deactive enzyme and is not observed in any other open classes. It puts ND6 TM3–4 loop in between PA and MA like a spanner in the works, explaining enzyme deactivation and a high barrier for active/deactive transition.

Another still often-used characteristic of the deactive state is the sensitivity of enzyme to NEM exposure, which modifies _ND3C39 (Figure 2C). However, this cysteine is exposed (thus accessible to NEM) both in open and in deactive states (while it is tucked away into Q cavity in closed state), so NEM assay is not specific for deactive enzyme but instead indicates the proportion of open plus deactive (if any) states. The only specific biochemical assay for deactive state is the delay in developing activity, as noted above.

Structures of (apo) complex I from other species have shown so far mostly open-like states (judging by _ND6TM3 and density for key loops) — e.g. T. thermophilus (PDB 6Y11), Arabidopsis (PDB 7AR8, 7ARB), Polytomella (PDB 7ARD) and NDH complex (PDB 6NBY, 6HUM). This alone would argue against considering open state as equivalent to deactive, since these species do not show any active/deactive transition. To understand this distribution of conformations, we can consider that closed–open–deactive states of mammalian enzyme are part of the same conformational trajectory with large energy barriers in between (due to a high barrier for deactive transition) (Figure 2A). Other species usually would have lower closed-to-open state barrier (as there is no deactive state) and so would ‘slip' into lower energy open state in the absence of turnover (i.e. for ‘as purified’ apo enzyme). Therefore mammalian apo enzyme can be seen in both closed and open states, while for most other species only open state can be observed. That is however, not always the case, as Chaetomium [38] and Drosophila [39] apo enzymes (both lacking deactive transition) were recently observed in both states, probably due to higher energy barrier between closed and open states in these species. These observations demonstrated that open/closed states are not unique for mammalian enzyme. For species showing only open apo state it would be expected that under turnover a closed state would appear, since if open/closed transition is a part of catalytic cycle then closed state should be populated to some degree during turnover (Figure 2A). This is exactly what we observed with E. coli enzyme [26], finally settling the debate. It is now clear that open/closed transitions are part of catalytic cycle in all species, while the deactive state is an ‘extreme' stabilised form of open state, specific for mammalian enzyme, where it may be important in preventing ischemia/reperfusion injury [46].

One distinct example requiring further clarification is Yarrowia enzyme. It is the only other species, apart from E. coli and ovine, studied under turnover so far. However, both apo and turnover states were observed to be open-like, judging by _ND6TM3 and ND3 loop state [37]. Yarrowia enzyme ‘as purified’ (apo) exists in a ‘low barrier' deactive state, from which it can be quickly reactivated. It is distinct from mammalian deactive state as key Q loops are mostly ordered and there is no relocation of _ND6TM4 [37]. Under turnover ND1 loop reforms from apo conformation, accommodating Q binding, but ND3 loop is disordered and π-bulge in _ND6TM3 remains, indicating that the single dominant observed under turnover Yarrowia state is open. The ratio of closed-to-open states under turnover can be variable, from 10 [25] to 54% [26] closed for ovine or 4 to 24% closed for E. coli [26]. This reflects relative energy levels of each state under turnover, depending on conditions (pH) and species. Since minor 3D classes (such as 5–10% closed) are challenging to classify out with standard cryo-EM techniques, it is likely that further processing, perhaps with our focus-reverse-classify approach, developed specifically for this purpose [47], will reveal closed state in Yarrowia as well.

Since we established that complex I cycles between open and closed conformations during turnover, and we had high-resolution structures (with waters resolved) of these states from two evolutionary distant species (ovine [25] and E. coli [26]), the stage for developing a first experiment-based complex I mechanism was set. The analysis of structures brought many surprises, contradicting previous assumptions. First, despite large conformational changes around Q cavity and E-channel, discussed above, there was virtually no difference in the conformations of ALS subunits between open and closed states, including, crucially, during turnover, in both species. This rules out previous mechanisms where conformation/solvent exposure of proton channels in ALS was proposed to change. Instead, data strongly supports a purely electrostatic mechanism for the ALS, as also indicated by apparent changes in the protonation state of some key residues. Several glutamates in the E-channel consistently become charged (side chain cryo-EM density disappears) in closed state [25,26].

Another surprise from the recent structures is the unique hydration pattern of the ALS, with the ND5 being much more hydrated at the IMS/periplasm side than ND2/ND4, first observed in mammalian enzyme [25] and later in yeast [37] and E. coli [26]. This unique hydration profile is consistent with all available complex I and NDH structures, as well as with molecular dynamics simulations [37], although in one case conclusions differed [48]. This suggests that ND2 and ND4, lacking any polar residues or waters at the IMS side, do not possess viable hydrated proton pathways leading towards IMS and only ND5 has a full proton input and output pathway (Figure 1C). _ND5HisTM8 may rotate on the TM8 π-bulge to establish a connection either towards the rest of the central axis or towards _ND5GluTM12, thus re-distributing incoming protons and preventing proton back-leak. ND4 appears also to have proton input pathway from matrix/cytosol with a similar switch/gate at _ND4LysTM8 [26]. ND2 probably does not have such an input (Figure 1C), but this may depend on species, and whether ND2 has matrix input pathway or not is not important for our mechanism, which will work in both cases. The E-channel is dry on both sides of the membrane, ruling out its direct involvement in proton pumping. This suggested an unexpected possibility that all the protons are ejected through ND5. This model is consistent with a very distinct sequence conservation pattern of ND5, conserved all along input, central axis connection and output pathways, while ND2/ND4 are conserved mostly along the central axis.

On the basis of this new structural knowledge we have developed a novel coupling mechanism of complex I, first proposed for mammalian enzyme [25], and then further elaborated and refined using E. coli data [26]. We set out to explain, on the basis of minimal assumptions, all the unusual structural features of complex I, including open-to-closed transition and ND5-only proton exit. After some permutations, we eventually arrived at a very robust and straightforward, in our opinion, ‘domino effect' mechanism, which naturally accounts for (and is based on) these unexpected new findings.

First, we consider that in open state Q cavity widens and opens to matrix (site W in Figure 3), although the entry site from lipids (site Q in Figure 3) remains narrow. Quinone is observed bound in shallow/median sites near the entrance but not at the deep site near N2, prevented by the extended/disordered 49 kDa/NuoCD β1–β2 loop. Therefore we propose that in open state quinone comes in or quinol comes out of cavity, and opening to the matrix is necessary to allow accompanying waters to come out/in to fill the cavity, as quinone tail would block the narrow lipid entry point. In closed state the cavity is fully enclosed and tightly engulfs quinone, which is now bound in the deep site. In this state quinone cannot move but can accept electrons from N2. The two protons needed to finish the reaction come from nearby 49 kDa His59-Asp160 (NuoCD His228–Asp329) pair, as only a few remaining in cavity waters cannot provide them. The cavity is enclosed on all sides except a connection through a funnel of charged residues to the E-channel, and then only in closed state it is connected further to the rest of central MA axis. This provides a natural way for redox reaction to initiate proton translocation events, because the two protons needed to re-protonate 49 kDa/NuoCD His/Asp pair have to come from the central axis (as evidenced by de-protonation of several E-channel glutamates).

Second, the ND5-only proton exit, along with the absence of any conformational changes in ALS, lead us to realise that this actually ensures tight ALS communication to each other in a series of proton transfer events, which must happen, unexpectedly, mostly along the central MA axis.

The proposed [26] cycle (Figure 3 and Supplementary Movie S1) starts with Step 1, where quinone temporarily binds in shallow/median site. Site W is open for waters to flow out of the cavity and give space to the incoming quinone. In this state the ALS are maximally protonated at the key TM8 sites (Lys/HisTM8) by protons coming into NuoL/NuoM (for simplicity in the description of the cycle we use only E. coli nomenclature/residue numbering as shown in Figure 1C) from the cytoplasm and redistributed along the central axis until the _NuoJTM3 break. For the ease of following proton transfers we indicate any protonated residues just by a ‘+' sign, leaving unprotonated ones empty. The break prevents futile proton leak into the Q cavity and back to the cytosol. For the closely interacting TM7/TM5 ion pair sites (LysTM7/GluTM5), the proton is proposed to reside in TM5 site, as suggested by cryo-EM observations [25,26]. The residues in TM7/TM5 ion pairs are thus uncharged and interact weakly. The TM12 (Lys/GluTM12) sites have a lower pKa and tuned to remain unprotonated in this state due to electrostatic interactions with protonated TM8 and TM5 sites. The exception is NuoL TM12 site, which is protonated as it is distal and so does not have a TM5 partner. (Step 2) Bound quinone traverses into the deep site, triggering the open-to-closed transition, so that the W site is closed off and the Q cavity tightly engulfs quinone. _NuoJTM3 rotates, establishing the uninterrupted proton path from the Q cavity all the way to the MA tip. Quinone accepts two electrons from cluster N2, and the unstable charged quinone intermediate is immediately protonated by the coordinating His/Asp pair. Since the Q cavity is sealed, the protons for the re-protonation of His/Asp pair come from the central axis (_HD79 and _NGluTM5). (Step 3) In a ‘minimal' interpretation (Occam's razor) of the subsequent events, de-protonation of these residues first triggers proton transfer from TM8 to TM7 site in NuoN, due to the removal of large positive charge around TM5 area. In a series of ‘domino effect' events, the removal of _NTM8 charge allows _MTM5 proton to hop on _NTM12 site, repeated in NuoM/L by _MTM8 to _MTM7, _LTM5 to _MTM12 and _LTM8 to _LTM7 hops (indicated by black arrows in Figure 3 Step 3). The exact sequence of each proton hop is given in Supplementary Movie S1. De-wetting of the TM8 area due to de-protonation, as observed in MD simulations [49], would prevent the back-flow of protons (along with possible role of _LHisTM8/_MLysTM8 switches). Effectively, due to the ‘forcefully’ protonated TM12 sites and a shift of proton from TM5 to TM7 sites (so that the residues in TM7/TM5 ion pairs are now charged and interact strongly) the enzyme will now be in a highly energised state, akin to stacked dominos ready to fall. (Step 4) The presence of the freshly produced quinol in the deep site along with the re-protonated state of coordinating residues triggers the transition from the closed to the open state, so that the Q cavity widens and the W site opens, allowing waters to come in and help quinol on its way out. The TM8 sites (and _KE72/E36) can be fully protonated from the cytosol, blocked off from the Q cavity by _JTM3. In total at this stage six protons (four to be pumped and two substrate) will enter the central axis. Five of them can enter via NuoL/M, re-distributed along the central axis to TM8 sites in NuoL, M and N, as well as to _KE72 and _KE36, while the sixth proton for _HD79 can come via open Q cavity. In this scenario the re-protonation of these sites is rather ‘passive', the key to coupling being that TM12 and TM7/TM5 sites state is fully controlled by quinone reactions. (Step 5) Electrostatic interactions with the protonated TM8 and TM7 sites (red dashes in Step 4) lead to a large decrease in pKa's of TM12 residues, forcing them to lose their protons. In NuoL the TM12 proton would be ejected directly into the periplasm/IMS. In the reverse wave of the ‘domino effect' this will initiate a sequence of proton hops from _LTM8 to _LTM12, _LTM7 to _LTM8 and _MTM12 to _LTM5, repeating twice more in NuoM/N and ending with _KE72 donating proton to _NTM5. The simple natural basis for this transfer of protons along the central axis is the appearance of a ‘vacancy' on the ‘left' of the chain and the electrostatic ‘pressure' of the incoming proton from the ‘right' (or reverse in Step 3). Effectively, after the cycle is repeated three times, each time ending closer to NuoL, in the end the _MTM12, _NTM12, _KE72 and _KE36 sites would transfer their protons along the central axis towards NuoL, adding up as the four protons ejected into the periplasm (Supplementary Movie S1). This brings the system back to Step 1, with TM8 protonated and a proton in TM7/TM5 ‘switch' sitting again on TM5, thus the cycle re-starts. Crucially, for the mechanism to work, a TM12 proton from NuoN/M must be transferred to the neighbouring NuoM/L TM5 and not directly to the periplasm, as otherwise the process will not be initiated in the next subunit (i.e. a domino will fall without tripping the next one). Similarly, in Step 3 it is essential for protons to hop across subunit interfaces from TM5 to TM12 sites. Therefore, our mechanism naturally explains initially counter-intuitive NuoL-only exit. The pump works with protons moving along the entire central axis either towards Q cavity (Steps 2–3) or in the reverse wave (Step 5), thus the periplasm/IMS side must be shielded from the solvent everywhere except the NuoL exit.

The mechanism also explains the reverse electron transport in complex I: under these conditions high pmf and a reduced quinone/quinol pool would promote reverse reaction by driving charge transitions in ALS in reverse to those in Figure 3. Translocation of protons into the matrix (the reversal of Steps 5 and 4, driven by high pmf and accompanied by quinol entering the cavity) would be coupled to the transfer of protons from the Q coordinating residues into the central axis (the reversal of Steps 3 and 2), creating a double negative charge near Q deep site. It would promote quinol binding as well as lower the N2 redox potential, enabling reverse electron transfer from N2 to FMN and NAD⁺. The presence of ready proton acceptors in the form of de-protonated Q coordinating residues will allow the quinol oxidation reaction to complete.

The ‘domino effect' mechanism is straightforward, robust and naturally explains the NuoL-only proton exit, why the Q entry site is so narrow, why the W site exists and why ND3/NuoJ TM3 rotates. The arrangement of key TM12, TM8 and TM7/TM5 sites appears to be a minimum necessary to allow for the mechanism.

Despite NuoL-only exit, all three ALS and the E-channel are essential, being responsible for the eventual transfer of one pumped proton each. Therefore, the varying number of ALS is related to the number of protons pumped per cycle in each of evolutionary-related complexes, such as MRP [34] and membrane-bound hydrogenases, according to the available redox energy [50]. The mechanism appears to be conserved in these proteins: the Q-like cavity encloses different substrates, such as sodium ions, plastoquinone, hydrogen or polysulfide, but the principle of the redox charge action via the lateral proton transfer along the central axis remains fully applicable [34].

Recent cryo-EM structures provided a strong experimental basis for our detailed complex I mechanism (Figure 3), which was published only a few months ago [26] and so it remains to be seen whether alternative proposals will continue to proliferate as intensively as they were until now or whether a wider consensus will start to emerge in the field.

Our initial proposal on ND5-only proton exit [25] was supported by experimental work and MD simulations on Yarrowia enzyme [37], as well as on a related MRP antiporter [51]. On the other hand, others did not agree [48], even though MD data seem to indicate very little, if any, hydration on the IMS side in ND4, ND2 and E-channel, in stark contrast with ND5 (Figure 2A in [48]).

The details of the coupling mechanism itself are also very different in recent proposals. In Yarrowia work [37] the mechanism involves charged quinone intermediates, which shuttle twice per cycle between deep and shallow sites. However, such long-lived intermediates are energetically unfavourable and have not been observed experimentally [52]. The interactions between chemical and pumped protons at the putative proton-loading site (PLS) near the Q shallow site would require fine tuning and ‘orchestration' of events to ‘push' protons into E-channel. Also, it is not clear how those two ‘pushed' protons would drag along another two pumped protons. Overall, the mechanism seems to be unnecessary complicated and potentially ineffective (moving quinone twice per cycle instead of once would lose energy). In another proposal, stemming from MD simulations, a single proton is also ‘pushed' from the shallow Q site into E-channel but it drives the entire cycle [49,53]. Here, in the ‘forward' wave, ion pairs in ALS ‘open', shifting protons from key TM8 to TM12 residues. The ‘reverse' wave is then imitated by proton release from ND5/NuoL to IMS, so that TM8 residues are re-protonated from the matrix, ion pairs close and each ALS (and E-channel) pumps out one proton each. Here, again, using just one proton to initiate the whole cycle seems ineffective, putting a lot of strain on a single event. There are some similarities with our mechanism as the forward wave energises the MA and the reverse wave releases protons into IMS, however, the direction and sequence of proton transfer events are completely different. And of course the mechanism is not consistent with growing consensus on ND5-only output.

It should be noted that a putative mechanism for proton translocation in complex I, also termed ‘domino effect', was suggested earlier [54]. However, despite some similarity involving consecutive electrostatic interactions between the ALS, the basic principles there were very different from our mechanism, as proton input/output was proposed to happen individually across each ALS for each pumped proton and proton transfers between ALS and a single NuoL output, a key to our mechanism, were not envisaged.

The debate on whether the open state and deactive state of mammalian complex I are one and the same is now settled, in our opinion, with, among others, recent publications on E. coli [26], Chaetominum [38] and Drosophila [39]. It is clear that both closed and open states are part of the catalytic cycle by active enzyme, while deactive (mammalian) and resting (bacterial, with disrupted Q cavity) [26] states are special off-pathway states. In a recent review, in an attempt to explain experimental observations, it was suggested that different species may enter a deactive-like state either rarely (WT mouse) or often (ND6 P25L mouse mutant) during catalysis, depending on how fast relaxing they are into deactive state [44]. However, it was still concluded that only closed complex I is active in the catalytic cycle and any open/deactive states are off-pathway. This is hard to imagine mechanistically, as a long quinone molecule would have to squeeze in and out of very narrow Q passage ∼200 times per second. The ‘only closed complex is active' notion also contradicts a growing number of examples of species showing closed/open states but no active/deactive transition. It also contradicts the striking fact that E. coli (with more examples from other species likely to follow) shows closed state only during turnover, otherwise being in a mixture of open/resting states [26].

What experiments can be done to clarify any remaining disputes/questions? Clearly, more cryo-EM data from enzyme undergoing turnover needs to be collected from more species. Only such experiments can reveal a full repertoire of 3D conformations of the complex. It is important to process the data carefully and extensively, as minor 3D states can be difficult to identify in the routine classification. Here, either our focus-reverse-classify approach [47] or the latest 3D classification methods in Cryosparc, focused on the E-channel/Q cavity area [39], will be useful. Most of species adopt open state in apo, but some show mostly closed state, e.g. Tetrahymena [55], mouse [42], and possibly Paracoccus denitrificans [44]. For these species it is likely that open state will appear under turnover, which would be an important result. It is possible that pH of the buffer, in which turnover is performed before flash-freeze for EM grids, may need to be adapted. We have observed that, consistent with de-protonation of key E-channel residues in the closed state, the proportion of this state is increased with pH [26]. Therefore, if the proportion of the minor open state needs to be increased (either in apo or during turnover), the pH should be decreased to ∼pH 5, and vice versa (pH 8–9 can be used to promote closed state).

The results of extensive mutagenesis experiments performed with E. coli and other species [26] are strongly consistent with our mechanism, but do not fully differentiate it from others, as e.g. central MA axis residues are essential in any mechanism. Perhaps introducing periplasm-side proton exits into ND4/ND2 by mutations into polar residues by homology with ND5 may help – it should not interfere with ‘one-proton-per-ALS' mechanism, but would disrupt our mechanism. The importance of sealing the Q cavity in the closed state may be verified by mutations of the central part of ND3 loop with an aim to introduce leaks, which should abolish proton pumping. Some residues around NuoJ/ND6 TM3 p-bulge were mutated, with loss of activity and proton pumping in E. coli V65G mutant [56]. Another important residue to mutate here is V63, which blocks the proton path in the open state. Its mutations to Gly or Ala may re-connect the path in the open state, leading to proton leaks into the Q cavity and disrupting proton pumping. One of the key residues not yet mutated is NuoL H254 from TM8, which links proton input path with the central axis (Figure 1C).

More direct ways to observe any movements in complex I (such as open/closed transitions during turnover) are experimentally challenging but are becoming feasible. One possibility is to use high-speed atomic force microscopy (AFM) to image movies of single complex I molecules, absorbed on mica sheets, during turnover. The methodology currently approaches required special resolution (∼5 Å) to resolve potential movements of the PA tip, although temporal resolution (∼100 ms) [57] will need to be improved to match complex I turnover rate (∼5 ms). Alternatively, the sample may be imaged cooled down (as we did for mammalian enzyme [25]) or with very limiting substrate concentrations to slow down turnover. Another approach could be to label appropriate residues on the surface of PA and MA so that single-molecule FRET signals [58] can be recorded and compared in the absence and presence of turnover. Finally, time-resolved cryo-EM methods are now approaching ∼5 ms resolution [59,60] and could be applied to visualise first steps in complex I reaction after addition of substrates. However, achieving necessary temporal synchronisation of the reaction after mixing with substrates may be challenging. It could be better pursued using e.g. caged NADH and laser light to initiate the reaction, which can also potentially achieve microsecond time resolution [61].

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

Leonid Sazanov: Conceptualization, Funding acquisition, Writing — original draft, Writing — review and editing.

ALS

antiporter-like subunits

FMN

flavin-mononucleotide

IMS

inter-membrane space

MA

membrane arm

PA

peripheral arm

TM

transmembrane