Structures of Respiratory Supercomplex I+III<sub>2</sub> Reveal Functional and Conformational Crosstalk

The mitochondrial electron transport chain complexes are organized into supercomplexes (SCs) of defined stoichiometry, which have been proposed to regulate electron flux via substrate channeling. We demonstrate that CoQ trapping in the isolated SC I+III2 limits complex (C)I turnover, arguing against channeling. The SC structure, resolved at up to 3.8 Å in four distinct states, suggests that CoQ oxidation may be rate limiting because of unequal access of CoQ to the active sites of CIII2. CI shows a transition between “closed” and “open” conformations, accompanied by the striking rotation of a key transmembrane helix. Furthermore, the state of CI affects the conformational flexibility within CIII2, demonstrating crosstalk between the enzymes. CoQ was identified at only three of the four binding sites in CIII2, suggesting that interaction with CI disrupts CIII2 symmetry in a functionally relevant manner. Together, these observations indicate a more nuanced functional role for the SCs.

Aerobic cellular respiration, the process by which cells transfer electrons from sugars, fats, and proteins to molecular oxygen (O2), is central to the energy metabolism of all eukaryotes and many prokaryotes. The final stages of aerobic respiration are carried out in the mitochondria. The mitochondrial electron transport chain (ETC) complexes of the inner mitochondrial membrane (IMM) catalyze the terminal electron transfer reactions. The ETC is composed of four large membrane protein complexes: (1) a H+-pumping NADH-coenzyme Q (CoQ; ubiquinone) oxidoreductase (complex [C]I), (2) a succinate-CoQ oxidoreductase (CII), (3) an obligatorily dimeric H+-pumping CoQH2 (reduced CoQ; ubiquinol)-cytochrome c (cyt c) oxidoreductase (CIII2; cytochrome bc1 complex), and (4) a H+-pumping cyt c oxidase (CIV) responsible for O2 reduction. We now have several atomic models for all of the isolated ETC complexes (Baradaran et al., 2013, Iwata et al., 1995, Iwata et al., 1998, Sun et al., 2005, Tsukihara et al., 1996). Most recently, structures of mammalian mitochondrial CI, the largest and least well characterized ETC complex, have been reported from multiple sources (Agip et al., 2018, Fiedorczuk et al., 2016, Guo et al., 2017, Wu et al., 2016, Zhu et al., 2016). CI is an ∼1 MDa membrane protein complex with 45 protein subunits arranged into two “arms”: a peripheral arm that extends into the mitochondrial matrix and a membrane arm that is embedded in the IMM (Hirst, 2013, Sazanov, 2015).

Although each ETC complex is capable of functioning in isolation (Hatefi et al., 1962), the individual complexes form supercomplexes (SCs) of defined stoichiometry (Schägger and Pfeiffer, 2000). In mammalian heart mitochondria, the majority of CI is found in association with CIII2 and CIV (SC I+III2+IV; respirasome) or in association with CIII2 alone (SC I+III2). Recent structural work has defined the arrangement of the individual complexes within the mammalian mitochondrial respirasome (Gu et al., 2016, Guo et al., 2017, Letts et al., 2016b, Sousa et al., 2016, Wu et al., 2016). Recent electron cryo-tomography on mitochondria from mammals, plants, and yeast revealed that it is not the respirasome that is structurally most conserved across kingdoms but SC I+III2 (Davies et al., 2018). However, the subunits shown to be responsible for the interactions between CI and CIII2 in mammalian mitochondria (Letts et al., 2016b) are either completely absent or largely truncated in plants and yeast (Letts and Sazanov, 2015, Subrahmanian et al., 2016), suggesting convergent evolution and, in turn, an important but still undefined physiological role for SC I+III2 in energy metabolism (Davies et al., 2018).

The possible physiological functions of the SCs remain controversial (Letts and Sazanov, 2017, Milenkovic et al., 2017), with proposals including roles in the stability of the individual complexes (Acín-Pérez et al., 2004, Calvaruso et al., 2012, Schägger et al., 2004), CoQ substrate channeling between CI and CIII2 (Bianchi et al., 2004, Lapuente-Brun et al., 2013, Lenaz et al., 2016), the reduction of reactive oxygen species (ROS) production (Lopez-Fabuel et al., 2016), and the prevention of non-specific protein aggregation in the IMM (Blaza et al., 2014). Evidence has been mounting against the hypothesis that substrate channeling is a significant function of the respirasome, especially against the notion that there are two pools of CoQ in the IMM, one associated with SCs and one freely diffusing (Enríquez, 2016, Lapuente-Brun et al., 2013). Recent respirasome structures (Gu et al., 2016, Guo et al., 2017, Letts et al., 2016b, Sousa et al., 2016, Wu et al., 2016) do not show any protein subunits blocking the free exchange of CoQ from CI or CIII2 within the membrane pool. Moreover, kinetic analyses indicate that only a single pool of CoQ exists in the membrane (Blaza et al., 2014, Fedor and Hirst, 2018, Gupte et al., 1984, Kröger and Klingenberg, 1973a, Kröger and Klingenberg, 1973b). To settle this debate, new experimental frameworks are needed to further explore the physiological roles of SCs.

Here, we report the isolation of functional SC I+III2 from ovine heart mitochondria. The preparation was highly active when isolated in amphipols, providing a soluble functional respiratory unit eminently suitable for detailed functional studies. Our results indicated that limiting amounts of CoQ-10 were co-purified with the SC and that CoQ trapping by SC particles in fact decreased rates of electron transport. Structural characterization of the SC particle by cryo-electron microscopy (cryoEM) resulted in reconstructions at 4.2–4.6 Å overall resolution of multiple structural three-dimensional (3D) classes and up to 3.8 Å resolution in focused refinements. In contrast to the isolated ovine CI (Fiedorczuk et al., 2016), subunit NDUFA11 (B14.7), the traverse helix from ND5 subunit, and a few other peripheral areas of the complex were well ordered within the SC, confirming the stabilizing role of CIII2 on CI (Letts and Sazanov, 2017). The structures here allowed us to improve the completeness and accuracy of the ovine CI model and to obtain an atomic model of ovine CIII2, with clear density for endogenous CoQ bound to three of the four possible sites. Comparison of local map resolution between SC structures revealed a CI state-dependent conformational flexibility in CIII2’s cytochrome b (MT-CYB) subunit, indicating crosstalk between the two complexes. Using a focus-revert-classify strategy to separate distinct states of CI, we resolved six open and one major closed conformation. These additional states expand on the structural states previously observed in the bovine, ovine, and mouse CI (Agip et al., 2018, Blaza et al., 2018, Fiedorczuk et al., 2016, Zhu et al., 2016) and add an intriguing possibility that a key transmembrane (TM) helix rotates during the catalytic cycle.

Our approach presents a new experimental framework for structure-function analysis of the physiological roles of mitochondrial SCs. Our biochemically defined system supports a stabilizing rather than a substrate-channeling function for SCs and suggests a more subtle functionally relevant interaction between CI and CIII2.

When purifying CI, we observed that the yield from the first chromatographic step (anion exchange Q-column) was lower when the membranes were extracted with the branched chain detergent lauryl maltose neopentyl glycol (LMNG) than when using the single-chain detergent dodecyl-maltoside (DDM) (Letts et al., 2016a). In LMNG, a significant proportion of CI did not elute from the column until the high salt wash (Figure S1A). By modifying the gradient, two clear peaks of CI NADH:FeCy activity could be identified (Figures S1B and 1A). The second peak contained both CI NADH:FeCy activity and high absorption at 420 nm, indicating the presence of heme groups (Figures S1B and 1A). When this peak was concentrated and run over either another round of anion exchange (Mono Q) or a size exclusion chromatography (SEC) column, it separated into two distinct peaks, one with NADH:FeCy activity and the other with the majority of the A420 signal (Figures S1C–S1E). Mass spectrometry demonstrated that these peaks corresponded to isolated CI and CIII2 (Figures S1F–S1H; Table S1). By adding the amphipathic polymer (amphipol) A8-35 directly following elution of the CI + CIII2 peak on the first anion exchange step, a peak containing both CI NADH:FeCy activity and A420 signal was isolated by SEC (Figures 1B and 1C). Blue Native (BN)-PAGE indicated that this peak was SC I+III2 (Figure 1D). This was confirmed by mass spectrometry, which identified the presence of all expected CI and CIII2 subunits (Figures 1E and 1H; Table S1).Figure 1Purification of SC I+III2 from Ovine Mitochondria(A) Q-Sepharose anion-exchange column chromatogram of LNMG-extracted washed mitochondrial membranes. Chromatograms show A280 (blue line), A420 (red line), percentage buffer B (containing 1M NaCl) (dashed black line), and CI NADH:FeCy activity (lime green triangles) throughout.

Complex (C)I elutes in two distinct peaks: the first peak is isolated CI, and the second peak is SC I+III2 and excess CIII2.(B) Superose 6 size exclusion column chromatogram of amphipol-(A8-35)-exchanged SC from the second peak fraction of the anion-exchange step shown in (A) in the absence of detergent.(C) Superose 6 size exclusion column chromatogram of amphipol-stabilized SC I+III2 from peak fraction containing SC I+III2 in (B).(D) BN-PAGE gel of the purified SC I+III2 with isolated CI and CIII2 shown for comparison.(E) SDS-PAGE of purified SC I+III2 with labels for some subunits identified by mass spectrometry (MS). Some labels were excluded for clarity.See also Figure S1 and Table S1.

In the absence of amphipol, the complexes dissociated over time and separated during subsequent purification steps (Figures S1C–S1E). Given that lipids are important to stabilize SCs (Mileykovskaya and Dowhan, 2014), de-lipidation is a possible cause for the separation of the SC I+III2 during purification in the absence of amphipols. We determined that the isolated SCs were lipid-protein particles containing on average 122 ± 8 lipid molecules per SC (Table S2). Given the ∼4:1 phospholipid/cardiolipin ratio of the IMM (Horvath and Daum, 2013), this suggests that each SC particle contains ∼24 cardiolipin molecules and ∼98 standard phospholipids, mainly phosphatidylcholine and phosphatidylethanolamine with minor amounts of phosphatidylinositol and phosphatidylserine.

The amphipol-stabilized SC I+III2 displayed all expected enzymatic activities (see Figure 2A for a schematic of SC I+III2 substrate binding sites), that is, CI NADH:CoQ oxidoreductase activity and CIII2 CoQH2:cyt c oxidoreductase activity via the Q-cycle mechanism (Cramer et al., 2011). Independent of whether NADH oxidation or cyt c reduction was monitored, clear hyperbolic concentration-activity curves for both NADH and cyt c were seen in the absence of any added CoQ analogs (Figures 2B and 2C). This indicated that endogenous CoQ-10 co-purified with the SC particles. This co-purification is expected even in the absence of CoQ-10 trapping in the membrane, because of CoQ-10’s high hydrophobicity (Persson et al., 2005) and the fact that the SCs, which contain several CoQ-10 binding sites, are being extracted into an aqueous environment.Figure 2NADH:cyt c Oxidoreductase Activity of Isolated SC I+III2(A) Schematic shows positions of CI (blue) and CIII2 (green) within the SC and the different catalytic sites. The FMN, FeS clusters, and heme groups are shown colored by atom: carbon in gray, nitrogen in blue, oxygen in red, sulfur in yellow, and iron in orange. The two major conformations of the Rieske FeS domain of the CIII subunit UQCRFS1 are indicated with Qp-proximal in orange and c1-proximal in red.

The gray area indicates the approximate extent of the amphipol-lipid belt.(B) [NADH]-activity curves, NADH oxidation (top), and cyt c reduction (bottom) throughout, in standard buffer (SB) plus 100 μM cyt c.(C) [cyt c]-activity curves in SB plus 100 μM NADH.(D) [DQ]-activity curve in SB plus 100 μM NADH and 100 μM cyt c.(E) [NADH]-activity curves in SB plus 10 μM DQ and 100 μM cyt c added.(F) [cyt c]-activity curves in SB plus 10 μM DQ and 100 μM NADH.(G) [DQ]-activity curves in lipid-detergent (LD) buffer plus 100 μM NADH and 100 μM cyt c.(H) [NADH]-activity curves in LD buffer plus 100 μM DQ and 100 μM cyt c.(I) [cyt c]-activity curves in LD buffer plus 100 μM DQ and 100 μM NADH.Data are mean ± SEM.See also Figure S2 and Table S3.

As NADH is a two-electron donor and cyt c is a single-electron acceptor, two molecules of cyt c are reduced per NADH oxidized. The observed “rate-coupling” ratio of Kcats (cyt c reduction/NADH oxidation) was 1.80 ± 0.05 (Table S3), near the expected ratio of 2.0 for perfect rate coupling, indicating that electrons from NADH are directly transferred to cyt c without a significant buildup of reduced CoQH2. Time courses of NADH oxidation and cyt c reduction showed an expected ∼2-fold greater final concentration of reduced cyt c relative to NADH added (Figures 3A–3F). A possible side reaction may be the reduction of cyt c by superoxide generated from the reduced flavin of CI (Pryde and Hirst, 2011), bypassing CoQ-10 and CIII2. To prevent this, all reaction buffers contained 50 U/mL superoxide dismutase (SOD). Together, the high rate coupling and expected final concentration of reduced cyt c indicated that very few electrons were escaping to ROS. Nevertheless, the activity of the isolated SC I+III2 was low relative to what was measured in ovine mitochondrial membranes or isolated mammalian mitochondrial CI using various CoQ analogs (Hatefi and Stiggall, 1978, Letts et al., 2016a, Sharpley et al., 2006, Shinzawa-Itoh et al., 2010), suggesting that the activity of the isolated SC I+III2 may be limited by insufficient levels of co-purified CoQ-10.Figure 3BSA Uncouples CI and CIII2 Activity in SC I+III2, but [DQ] Alone Has Little Effect(A–F) SC I+III2 activity time courses after addition of 5 μM NADH to SB plus 100 μM cyt c (A); 5 μM NADH to SB plus 100 μM cyt c and 10 μM DQ (B); 10 μM NADH to SB plus 100 μM cyt c and 10 μM DQ (C); 10 μM NADH to SB plus 100 μM cyt c and 100 μM DQ(D); 10 μM NADH to SB plus 100 μM cyt c, 100 μM DQ, and 10 μM BSA (E); and 10 μM NADH to LD buffer plus 100 μM cyt c and 100 μM DQ (F).

In each panel, the [NADH] (blue circles) and [reduced cyt c] (red squares) are shown at each time point (n = 8, data are represented as mean ± SEM). Initial rates and final concentrations are denoted by blue lines for [NADH] and red lines for [cyt c] and indicted in the panels (mean ± SEM). Green triangles represent calculated values of DQH2 at each time point assuming [DQH2](t) = {[NADH](initial)-[NADH](t)}-{[Red. cyt c](t)/2} and plotted as mean ± SEM.(G) [DQ]-activity curves, NADH oxidation (top), and cyt c reduction (bottom) throughout, in SB plus 10 μM BSA, 100 μM NADH, and 100 μM cyt c.(H) [NADH]-activity curves in SB plus 10 μM BSA, 100 μM DQ, and 100 μM cyt c.(I) [Cyt c]-activity curves in SB plus 10 μM BSA, 100 μM DQ, and 100 μM NADH.(J) [NADH]-activity curves in SB plus 100 μM DQ and 100 μM cyt c.(K) [Cyt c]-activity curves in SB plus 100 μM DQ and 100 μM NADH.(L) NADH oxidation activity in the absence of cyt c at the indicated DQ concentrations in SB: rotenone, 2.5 μM rotenone; BSA, 10 μM BSA; LD buf., lipid-detergent buffer.Data are mean ± SEM, n = 8–12. Not significant (n.s.) indicates p > 0.01.See also Table S3.

To address this, we tested different concentrations of the CoQ analog decyl-ubiquinone (DQ) in the reaction buffer (Figure 2D). The addition of small amounts of DQ (∼10 μM) resulted in a large increase in both NADH oxidation and cyt c reduction activity; however, further addition of DQ resulted in a reduction of activity relative to the peak at 10 μM DQ (Figure 2D). The non-hyperbolic nature of the [DQ]-activity curve has been attributed to its hydrophobic “detergent-like” properties (Estornell et al., 1993, Letts et al., 2016a). If DQ solubility dictated the shape of the [DQ]-activity curve, factors such as ionic strength should affect the profile. To test this, we exchanged the buffer from HEPES to potassium phosphate, maintaining the same pH, and re-measured the concentration activity curve (Figure S2A). This buffer change resulted in a significantly different profile at low DQ concentrations, with a broader high-activity peak (5–20 μM DQ) in phosphate compared with HEPES (tightly centered around 10 μM; Figures S2B and S2C). This is consistent with the shapes of the curves resulting from solubility effects and the low activity seen at high [DQ] being due to poor DQ solubility.

Given that maximal NADH:cyt c oxidoreductase activity was seen with 10 μM DQ, this concentration was used to evaluate NADH and cyt c activity (Figures 2E and 2F). In the case of the [NADH]-activity curve, a hyperbolic profile was seen, with a higher Vmax (∼5-fold increase) and higher apparent Km than when no DQ was added (Figures 2B, 2C, 2E, and 2F). However, although the activity was increased at all [cyt c] when 10 μM DQ was added, the [cyt c]-activity curves are no longer hyperbolic, instead showing a pronounced decrease in activity at high [cyt c] (Figure 2F). Given that this is not seen at higher [DQ] (see below), this reduction in activity may stem from a non-specific interaction between cyt c and DQ. The rate-coupling ratio from the [NADH] activity curves was 1.90 ± 0.06 (Table S3), again reflecting efficient electron transfer from NADH to cyt c (Figure 3B).

Although addition of DQ increased the activity of the amphipol-stabilized SC I+III2 (Figure 2D), the NADH oxidation activity of CI was still lower than that seen in mitochondrial membranes or solubilized in detergent (Letts et al., 2016a), indicating that some aspect of SC I+III2 was limiting CI turnover. The CIII2 activity observed in SC I+III2 (102 ± 2 s−1; Table S3) is consistent with that measured for isolated eukaryotic CIII2 (S. cerevisiae, 60–270 s−1 [Covian and Trumpower, 2008, Gutierrez-Cirlos and Trumpower, 2002, Nett et al., 2000], and Bos taurus, ∼600 s−1 [Brandt and Okun, 1997]) and for mitochondria of living cultured cells (∼250 s−1 [Kim et al., 2012, Ripple et al., 2013]). However, the rate of ∼100 cyt c reduced per CIII2 per second is on the low end of the reported range, suggesting that the activity of CIII2 may also be impeded in the amphipol-stabilized SC.

To determine whether CI and CIII2 activities are limited by confinement in the SC or whether isolation by this protocol results in lower activity in general, the activities of the amphipol-stabilized SC were assayed in lipid-detergent (LD) buffer containing 0.25 mg/mL dioleoylphosphatidylcholine (DOPC)/cardiolipin (CL) 4:1, 0.1% CHAPS (w/v), and 0.1% LMNG (w/v). This LD mixture allows maximal activity for isolated ovine CI (Letts et al., 2016a). The addition of lipid and detergent to the reaction mixture resulted in a significant shift in the [DQ] activity curve with maximal activity at higher [DQ] (∼100 μM) and marked increases in both NADH oxidation and cyt c reduction activity (Figure 2G). The [NADH] activity curves in these conditions were hyperbolic and displayed increased Vmax and higher apparent Km values than when measured in the absence of lipid and detergent (Figures 2H and 2E). Addition of lipid and detergent uncoupled NADH oxidation from cyt c reduction and resulted in robust NADH oxidation in the absence of any added cyt c (Figure 2I). However, the addition of sufficient cyt c (≥50 μM) did increase the NADH oxidation activity of CI, indicating that CIII2 activity was still able to influence CI and that their association may not be wholly disrupted (Figure 2I). When cyt c reduction was monitored, the [cyt c] activity curve was hyperbolic, resulting in a 2-fold increase in Vmax relative to the maximum activity seen in the absence of lipid and detergent (8.22 ± 0.09 U/mg versus 4.11 ± 0.09 U/mg). This rate of ∼230 cyt c reduced per second per CIII2 agrees more closely with fluxes measured in the mitochondria of living cells (Kim et al., 2012, Ripple et al., 2013). Given that upon addition of the LD buffer, the CI activity increased by a larger factor than CIII2 activity, the rate-coupling ratio decreased to 1.21 ± 0.02 (Table S3).

When a time course of the reaction was examined, a mismatch between the amount of NADH oxidized versus cyt c reduced at each time point was observed (Figure 3F). Moreover, the reduction of cyt c became bi-phasic with a fast phase of reduction that coincided with the oxidation of NADH by CI and a slow phase of cyt c reduction after all the NADH was oxidized (Figure 3F). The two phases of cyt c reduction activity again suggest that the amphipol-stabilized SC I+III2 may not be completely disrupted in the LD buffer. The initial fast phase of cyt c reduction likely corresponds to DQH2 diffusing quickly within the SC between CI and CIII2 and the slow phase to DQH2 that escapes the SC into the LD buffer and then must re-encounter CIII2 to be oxidized (Figure 3F). Overall, there is a buildup of reduced DQH2 during the experiment as CI activity outpaces that of CIII2. This is consistent with there being a kinetic advantage to keeping the CI and CIII2 active sites close together, as would be expected for a diffusion-coupled process (Hackenbrock et al., 1986). Nonetheless, diffusion of CoQ has been shown not to be rate limiting during respiration (Chazotte and Hackenbrock, 1988), and hence decreasing the diffusion distance between CI and CIII2 is not likely to be the main physiological role of SCs. Comparing the LD buffer data with those in the absence of detergent indicates that CI activity is limited in the amphipol-stabilized SC I+III2 particle.

We also note that the NADH oxidation and cyt c reduction activities of the amphipol-stabilized SC I+III2 were sensitive to known CI and CIII2 inhibitors (Figure S3). We tested the CI inhibitors rotenone and piericidin A, as well as Na+/H+-antiporter amiloride inhibitors. Also, the QP-site CIII2 inhibitor myxothiazol and the QN-site inhibitor antimycin A both showed strong inhibition of SC activities, with Hill coefficients close to 2, suggesting that binding to either of the QP or QN sites in the dimer may be sufficient to inhibit CIII2 activity (Figures S3C and S3D).

To better understand what was limiting the activity of the SC, we measured SC activity at different DQ concentrations in the presence or absence of the non-specific hydrophobic carrier protein BSA (Figures 3G–3I). The addition of 10 μM BSA to the reaction mixture shifted the [DQ]-activity curve to higher [DQ], and the curves became more “hyperbolic-like” (Figure 3G). This indicates that higher concentrations of DQ are needed to make the substrate available for electron transport in the SC I+III2 particles in the presence of BSA and hence that BSA reduces the partitioning of DQ into the SC lipid-protein particles at low [DQ]. When we measured the NADH oxidation activity in the presence of BSA, we saw a slight but significant (p < 0.01) increase in Vmax, whereas there was a significant decrease in the Vmax of cyt c reduction (Figures 2E, 2F, and 3H). This results in a reduction of the rate-coupling ratio similar to that seen in LD buffer (Table S3), indicating that BSA provides an alternate path for DQH2 out of the SC I+III2 particle. To test whether the higher concentration of DQ was partially responsible for the differences in LD buffer and in the presence of BSA, we measured [NADH] activity and [cyt c] activity curves with 100 μM DQ (Figures 3J and JK). As expected from the [DQ] activity curves (Figure 2D), the Vmax measured with 100 μM DQ was lower than that using 10 μM DQ (Figures 3J and 3K). However, the decreases in NADH oxidation and cyt c reduction activities were proportional, thus maintaining a high rate-coupling ratio between CI and CIII2 (Table S3).

Whether the ability of SC-associated DQ to exchange with the bulk DQ pool is dependent on [DQ] can also be estimated by measuring NADH oxidation in the absence of any added cyt c (Figure 3L). In these conditions, reduced DQH2 cannot be readily oxidized by CIII2, and thus the only source of oxidized CoQ to maintain CI activity is via exchange of DQ from the bulk pool into the SC I+III2 particles. In the absence of DQ and cyt c, NADH oxidation activity was undetectable (Figures 2C and 3L). This shows that the small amount of co-purified CoQ-10 (Figures 2B and 2C) is rapidly reduced and that there is no readily available mechanism for the regeneration of oxidized CoQ-10. When 10 μM DQ was added, some NADH oxidation was maintained even in the absence of added cyt c (Figures 2F and 3L), indicating that the more soluble CoQ analog DQ is able to slowly exchange with the bulk pool. Surprisingly, the rate of NADH oxidation in the absence of added cyt c was independent of the concentration of DQ, with no significant difference in the rate of NADH oxidation over 10–200 μM DQ (Figure 3L). This low rate of NADH oxidation was nearly completely inhibited by rotenone (Figure 3L), indicating that NADH was not oxidized via ROS production at the flavin site of CI (Pryde and Hirst, 2011) but was dependent on DQ reduction at the Q-tunnel of CI.

The fact that NADH oxidation is independent of [DQ] indicates that not all DQ is accessible to CI and that at 10 μM DQ, the accessible pool of DQ is already saturated. The accessible DQ pool is likely the DQ that has partitioned into the SC I+III2 particles because of its hydrophobic nature; hence, the low level of activity seen in the absence of cyt c would represent the exchange of SC-associated DQH2 with bulk solvent DQ. In our conditions, the exchange of SC-associated DQH2 with bulk solvent DQ occurred at a rate of ∼7.0 ± 0.1 s−1 (Figure 3L). If the rate-limiting step of NADH oxidation in the absence of cyt c were DQ exchange, then we would expect the addition of BSA or LD buffer to significantly increase the rate. As shown in Figures 2I, 3I, and 3L, this is the case. Overall, these data indicate that in the absence of cyt c, CI activity is limited by exchange of “local” SC-associated DQ with the bulk solvent pool. Thus, when cyt c is present, the increase in CI activity is directly attributable to the ability of CIII2 to re-oxidize local CoQH2.

We used the amphipol-stabilized SC I+III2 to solve the atomic structure of the SC to resolutions up to 3.8 Å (in focused refinements) by cryoEM (Tables 1 and S4), using our improved protocol for model refinement (STAR Methods). CI appeared more stable within the SC than previous preparations of isolated ovine CI (Fiedorczuk et al., 2016), allowing us to improve the model in some peripheral areas that were previously disordered. Initial 3D classification identified four classes of SC I+III2 particles (Figures 4A–4D; Figure S4C). These differed mainly in the relative angles between the CI peripheral arm and membrane arm and in the angle between CI and CIII2 (Figures 4E and 4F). We concluded that the most distinct class with respect to the position of the CI peripheral arm belonged to the closed state (or active form) of CI (Agip et al., 2018, Fiedorczuk et al., 2016). The assignment of the closed state was based on clear density adjacent to the CI Q-tunnel for the NDUFS2 β1-β2 loop (residues Gly52–Gly60) and for the ND3 TM helix 1 (TMH1)-TMH2 loop (residues Pro25–Lys54; Figure 4G). In all other classes of SC I+III2, these loops were disordered (Figure 4G), suggesting that they are distinct forms of the open state (or deactive form) of CI (Blaza et al., 2018, Fiedorczuk et al., 2016). Previously, only a single open state had been identified for CI (Agip et al., 2018, Blaza et al., 2018). We identified striking changes in TMH3 of ND6 between the open- and closed-state structures. Whereas in the closed state, ND6 TMH3 was a clear α-helix (residues Leu52–Met74), in each of the open states, the helix was interrupted by a π-bulge midway across the membrane (residues Tyr60–Met65; Figure 4H).

The result of this π-to-α transition between the open and closed states of CI is a rotation of ∼100° for the C-terminal half of ND6-TM3 (residues Met64–Met74; Figures 4H and 4I).Table 1CryoEM Map and Model Refinement and Validation StatisticsReconstructionSC I+III2 ClosedSC I+III2 Open 1SC I+III2 Open 2SC I+III2 Open 3CI Peripheral ArmCI Membrane ArmCIII2CI IsolatedCI ClosedCI Open 1CI Open 2CI Open 3Number of particles39,86335,64030,83614,230178,121174,334102,31457,16022,10726,97825,40421,913Accuracy of rotations (°)0.4540.3980.4300.4891.370.6401.570.5230.5210.4600.4740.551Accuracy of translations (pixels)0.3100.3000.3000.3760.5670.4180.6110.3890.3810.3110.3370.390Box size (pixels)512512512512400512364512512512512512Final resolution (Å)4.24.24.24.63.83.93.94.14.34.14.24.4MAP SHARPENING B FACTOR (Å2)−80−75−80−55−80−90−90−80−75−70−80−95PDB ID6QBX6QC36QC26QC46Q9D6Q9B6Q9E6QA96QC56QC66QC86QC7EMDB ID449344954494449644804479448144824497449845004499RefinementSoftwarePhenix 1.14 real-space-refineInitial model (PDB code)5LNK and 1PPJ5LNK5LNK1PPJ5LNK5LNK5LNK5LNK5LNKMap/model correlationModel resolution (Å)4.24.24.24.63.94.04.04.14.34.14.24.4d99 (Å)4.34.34.34.74.04.14.14.34.44.24.34.4FSC model 0.5 (Å)4.44.34.34.73.84.04.04.24.44.24.34.4Map CC (masked)0.710.750.760.750.820.780.800.780.770.770.760.75Model compositionNon-hydrogen atoms97,04996,89796,93896,70527,66238,01331,99765,40365,69165,34465,39365,353Protein residues12,09212,05912,06312,0713,4634,6273,9888,0918,1368,0858,0928,086Number of chains656565651829204545454545Number of ligands and cofactors21212121121111313131313Number of lipids3790012963552Atomic displacement parameters (ADP)Protein average (Å2)56.1266.1161.49103.4078.173.982.1123.8880.8373.9453.3765.28Ligand average (Å2)67.1064.6682.11144.0860.972.795.3140.18104.9483.4880.5980.71RmsdsBond lengths (Å)0.0060.0100.0090.0060.0060.0070.0070.0070.0080.0060.0080.007Bond an

gles (°)1.131.291.221.081.081.241.021.191.261.161.241.18Ramachandran plotFavored (%)88.6288.1088.4589.0489.0388.6291.3487.9687.5189.6488.0387.62Allowed (%)11.2611.7711.4710.8210.7711.318.6311.9812.3310.2911.9112.31Disallowed (%)0.130.130.080.130.200.070.030.060.160.080.060.08ValidationMolProbity score1.982.042.001.981.931.921.782.022.051.902.012.00Clash score7.108.137.567.296.376.175.177.667.996.227.597.14Rotamer outliers (%)0.400.750.570.380.240.270.420.440.720.550.570.47EMRinger score1.211.231.260.332.231.642.311.161.101.421.230.85Figure 4SC I+III2 Structures Reveal State-Dependent Conformational Changes in the CI Membrane Arm(A–D) CryoEM densities for the (A) closed class, (B) open class 1, (C) open class 2, and (D) open class 3.(E) Overlay of the models for the different SC classes aligned by the CI membrane arm shown as cartoons. Models colored as in (A)–(D) and viewed from CI side (left) and the mitochondrial matrix (right). Differences in the relative positions of the CI peripheral arm and CIII2 are indicated by arrows.(F) Slice through the CI peripheral arm and CIII2 at the position indicated by the red dashed line in (E). Models are shown as ribbons and colored as in (E). Approximate boundaries between subunits are indicated by black lines.(G) CryoEM density for the closed state CI (left, blue density) and open state 1 (right, green density) for the NDUFS2 β1-β2 loop in the Q-tunnel (top) and the ND3 TMH1-TMH2 loop (bottom). Models are shown as cartoons and colored by subunit: green, NDUFS2; red, NDUFS7; light green, ND1; white, ND3; and pink, ND6. The black and white hexagon indicates approximate binding site for CoQ.(H) ND6 TM3 from the closed state (left, blue) and open state 1 (right, green) viewed from the same side. The π-bulge in the open state 1 is indicated.

Models shown as cartoons with side chains as sticks colored by atom, with nitrogen blue, oxygen red, sulfur yellow, and carbon colored as the cartoon helix.(I) View from the mitochondrial matrix looking at ND6 TMH3 in the closed state (top) and open state 1 (bottom). Models shown as cartoons and colored by subunit: ND4L in green, ND6 in light blue, ND3 in blue-green, ND1 in gray, and NDUFS2 in green. Positions of the ND6 TMH3 side chains Tyr70 and Phe68 are shown.See also Figure S4 and Table S4 for initial processing.

The three major changes between the closed and open states—the ordering of the NDUFS2 β1-β2 loop in the Q-tunnel, the ordering of the ND3 TMH1-TMH2 loop, and the π-to-α transition of ND6-TM3—occur at the interface of the CI peripheral and membrane arms and were accompanied by a significant rotation between the two arms of the complex, bringing NDUFA5 and NDUFA10 into close contact (Figure 4F). As shown in Figure 4I, the ordered ND3 TMH1-TMH2 loop in the closed state crosses directly above ND6 TMH3. This loop, however, was disordered in the open states. Additional large conformational changes were seen in this area: the short three-stranded β sheet of NDUFS2 that harbors the β1-β2 loop was adjacent to ND6 TMH3 in the closed state but rotated away from TMH3 by as much as 10 Å in the open states (Figure 4I). This movement was accompanied by the striking rotation of the side chains of Phe68 and Tyr70 because of the π-bulge-to-α-helix rearrangement. The analogous structural differences were also observed between the active and deactive forms of CI for mouse CI (Agip et al., 2018), indicating that such large conformational transitions are conserved.

Because of the hinge-like motions between the different regions of the SC, we carried out focused refinements on distinct sub-regions to generate higher quality density maps. These focused refinements were performed on the peripheral arm of CI, on the membrane arm of CI and on CIII2, resulting in reconstructions at 3.8–3.9 Å resolution and improved atomic models for each region of the SC (Figure S5; Tables 1 and S5).

Although we were able to separate three distinct CI open states using the standard 3D classification, we reasoned that because of the systematic worsening map quality along the peripheral arm of CI (Figure 7; Figure S7), averaging of particles with diverse angles between the membrane and peripheral arms was still occurring. Hence, we developed a strategy for the separation of different CI states on the basis of the angle between the peripheral and membrane arms. Termed “focus-revert-classify,” an initial focused refinement around the peripheral arm of CI is performed, maximizing the differences in the relative positions between the membrane arms of each CI, followed by “reverting” to a mask around the membrane arm for classification (Figures 5A and S6A).Figure 5Focus-Revert-Classify Strategy Results in at Least Six Different CI Open Structures(A) Schematic of focus-revert-classify strategy for separating CI particles on the basis of the angle between the peripheral and membrane arms.(B) Overlay of CI models generated from the focus-revert-classify strategy aligned by the CI membrane arm. Models colored with the closed state blue and the different open states from dark green to yellow-green according to overall resolution. Viewed from CI side (left) and the mitochondrial matrix (right). Differences in the relative positions of the CI peripheral arm indicated by arrows.(C) Slice through the CI peripheral arm at the position indicated by red dashed line in (B). Models shown as ribbons and colored as in (B). Approximate boundaries between subunits are indicated by black lines, subunits labeled.See also Figure S6.

The focus-revert-classify strategy resulted in the separation of nine distinct structural classes of CI ranging in resolution from 4.1 to 7.5 Å (Figures 5B and S6A). Of these, seven correspond to distinct open states in which the Q-site loops are disordered, one to a closed state in which the loops around the Q site are well ordered and the final class to a closed-like state at 7.0 Å, precluding the assignment of any of the active site loops. All six CI open state structures at ≤6.5 Å each had distinct angles between the peripheral and membrane arms (Figures 5B and 5C). The closed state differs from the open states as described above for the closed state of the SC (Figure 5C). When comparing the particles between the classes of the original 3D classifications of the SC, isolated CI (Figure S4C) and focus-revert-classify (Figure S6A), it is clear that the closed state particles are distinct from all other classes (Figure S6B). The majority of particles that we identified as having CI in the closed state in the original classifications were also found in the closed state class after application of the focus-revert-classify strategy. This was not the case for the three original open state classes, which were more distributed between the six focus-revert-classify open classes. These data indicate that the closed state of CI is a distinct conformation and the open state is an ensemble of nearly continuous conformations differing in the relative positions of CI’s membrane and peripheral arms.