Control of mitochondrial superoxide production by reverse electron transport at complex I

The generation of mitochondrial superoxide (O2˙̄) by reverse electron transport (RET) at complex I causes oxidative damage in pathologies such as ischemia reperfusion injury, but also provides the precursor to H2O2 production in physiological mitochondrial redox signaling. Here, we quantified the factors that determine mitochondrial O2˙̄ production by RET in isolated heart mitochondria. Measuring mitochondrial H2O2 production at a range of proton-motive force (Δp) values and for several coenzyme Q (CoQ) and NADH pool redox states obtained with the uncoupler p-trifluoromethoxyphenylhydrazone, we show that O2˙̄ production by RET responds to changes in O2 concentration, the magnitude of Δp, and the redox states of the CoQ and NADH pools. Moreover, we determined how expressing the alternative oxidase from the tunicate Ciona intestinalis to oxidize the CoQ pool affected RET-mediated O2˙̄ production at complex I, underscoring the importance of the CoQ pool for mitochondrial O2˙̄ production by RET. An analysis of O2˙̄ production at complex I as a function of the thermodynamic forces driving RET at complex I revealed that many molecules that affect mitochondrial reactive oxygen species production do so by altering the overall thermodynamic driving forces of RET, rather than by directly acting on complex I. These findings clarify the factors controlling RET-mediated mitochondrial O2˙̄ production in both pathological and physiological conditions. We conclude that O2˙̄ production by RET is highly responsive to small changes in Δp and the CoQ redox state, indicating that complex I RET represents a major mode of mitochondrial redox signaling.


heart mitochondria. Measuring mitochondrial H 2 O 2 production at a range of proton-motive force (⌬p) values and for several coenzyme Q (CoQ) and NADH pool redox states obtained with the uncoupler p-trifluoromethoxyphenylhydrazone, we show that O 2
. production by RET responds to changes in O 2 concentration, the magnitude of ⌬p, and the redox states of the CoQ and NADH pools. Moreover, we determined how expressing the alternative oxidase from the tunicate Ciona intestinalis to oxidize the CoQ pool affected RET-mediated O 2 . production at complex I, underscoring the importance of the CoQ pool for mitochondrial O 2 . production by RET. An analysis of O 2 . production at complex I as a function of the thermodynamic forces driving RET at complex I revealed that many molecules that affect mitochondrial reactive oxygen species production do so by altering the overall thermodynamic driving forces of RET, rather than by directly acting on complex I. These findings clarify the factors controlling RET-mediated mitochondrial O 2 . production in both pathological and physiological conditions. We conclude that O 2 . production by RET is highly responsive to small changes in ⌬p and the CoQ redox state, indicating that complex I RET represents a major mode of mitochondrial redox signaling.
Superoxide (O 2 . ) 3 is the proximal reactive oxygen species (ROS) formed within mitochondria, with most O 2 . being very rapidly converted to H 2 O 2 by manganese superoxide dismutase (MnSOD) within the matrix (1,2). As well as contributing to oxidative damage, H 2 O 2 acts as a redox signal, both within the mitochondria and in the cytosol (3)(4)(5)(6). This mode of signal transduction arises via the reversible oxidation of protein thiols that pass on the modification to effector proteins as a redox relay (3)(4)(5)(6). Whereas there are a number of potential mitochondrial sources of O 2 . (2,7,8), respiratory chain complex I is considered to be a major contributor (1). The production of O 2 . at complex I can be driven by reverse electron transport (RET) by a highly reduced coenzyme Q (CoQ) pool and a large protonmotive force (⌬p), which together drive electrons backward through complex I and lead to a dramatic increase in O 2 . production ( Fig. 1) (9). This process has been known since the 1960s (10) but was tacitly assumed to be an in vitro curiosity of no physiological relevance (1,11,12). However, there is now considerable evidence that RET at complex I is a physiological process that underlies mitochondrial redox signaling in a range of situations (11,13) while also leading to pathological oxidative damage during ischemia-reperfusion injury (1,14,15). A particularly intriguing aspect of RET is that it does not require damage to, or inhibition of, the respiratory chain (1,9). As RET responds sensitively to physiological variables, O 2 . production by complex I can be modulated under physiological conditions (1,9). Hence, there is considerable interest in understanding the mechanisms by which mitochondria regulate O 2 .
production at complex I by RET as a physiological signaling process. Here, we quantified the factors that determine RET in isolated heart mitochondria. To do this, we measured H 2 O 2 generation as a function of the membrane potential (⌬) and of the reduction state of the CoQ and NADH pools. This was done by altering ⌬ with the uncoupler carbonyl cyanide p-trifluoromethoxyphenylhydrazone (FCCP) and also oxidizing the CoQ pool by ectopic expression within heart mitochondria of the alternative oxidase (AOX) from Ciona intestinalis (16 production on these three variables was measured in parallel (Fig. 2). ⌬p was measured as ⌬ from the distribution of the lipophilic methyltriphenylphosphonium (TPMP) cation (21) (Fig. 2B). Whereas this does not measure the pH gradient (⌬pH) across the inner membrane, ⌬pH is a smaller (ϳ10 -20%) component of ⌬p than ⌬ and here is assumed to be invariant. The redox state of the CoQ pool was assessed by measuring its percentage reduction by reverse-phase HPLC (Fig. 2C). The redox state . production by RET. This occurs when the ⌬p (a combination of the ⌬ and the ⌬pH) is high and the CoQ pool is reduced. Q, ubiquinone; QH 2 , ubiquinol.

Complex I reverse electron transport
of the NADH pool was inferred from NAD(P)H fluorescence. Although this cannot distinguish between NADH and NADPH, the mitochondrial NADPH pool is thought to contribute less to changes in this variable (22); hence, this method gives a reasonable assessment of the NADH pool redox state (Fig. 2D).
The addition of succinate to heart mitochondria led to extensive H 2 O 2 production ( Fig. 2A), a large ⌬ (Fig. 2B), and highly reduced CoQ (Fig. 2C) and NAD(P)H (Fig. 2D) pools. The addition of the complex I inhibitor rotenone decreased H 2 O 2 production without affecting ⌬ (Fig. 2B) or the CoQ pool (Fig.  2C), but there was oxidation of the NAD(P)H pool (Fig. 2D). These findings are consistent with H 2 O 2 production originating from complex I by RET (1). To explore the dependence of RET on ⌬ and the redox state of the CoQ and NAD(P)H pools, we measured these variables with increasing amounts of the uncoupler FCCP. The gradual decrease in H 2 O 2 production that resulted ( Fig. 2A) was associated with a decrease in ⌬ (Fig.  2B) and oxidation of the CoQ (Fig. 2C) and NAD(P)H (Fig. 2D) pools.
The matrix pH has been suggested to alter RET directly at complex I, independently of its role as the ⌬pH component of ⌬p (23,24). Hence, we next used the K ϩ /H ϩ exchanger nigericin to decrease the matrix pH from ϳ7.7 to that of the incubation medium (pH 7.4), thereby abolishing the ⌬pH component of ⌬p. Importantly, the magnitude of ⌬p will not change, due to a compensatory increase of ϳ20 mV in ⌬ (Fig. 2B). Nigericin resulted in an oxidation of the CoQ (Fig. 2C) and NAD(P)H pools (Fig. 2D) and a decrease in H 2 O 2 efflux ( Fig. 2A). Together, these data show a strong dependence of RET at complex I on the magnitude of ⌬p and on the CoQ and NAD(P)H redox states.

Effect of the AOX on O 2 . production by RET
To analyze the effects of CoQ pool redox state on O 2 . production by RET, independently of effects on ⌬p, we utilized mice expressing the AOX from C. intestinalis (16). AOX transfers electrons from CoQH 2 directly to O 2 , bypassing complex IV, and thus acts as a safety valve to prevent the excessive reduction of the CoQ pool (25,26). The AOX protein was present in heart mitochondria from AOX ϩ/Ϫ knock-in mice (Fig. 3A) and was catalytically active, as respiration in mitochondria from WT mice was inhibited by cyanide, whereas mitochondria from AOX mice continued to respire, but this residual respiration was sensitive to the AOX inhibitor N-propyl gallate (Fig. 3B). Mitochondrial H 2 O 2 efflux during succinate oxidation was decreased by AOX expression as shown previously (16) (Fig.  3C). This was not due to a decrease in ⌬ (Fig. 3D); however, the CoQ pool was more oxidized in mitochondria containing AOX (27) (Fig. 3E)

Effects of therapeutic compounds on complex I RET
A number of potentially therapeutic compounds are thought to act, at least in part, by decreasing mitochondrial ROS production. Therefore, we set out to assess some of these compounds to determine whether they altered RET at complex I. The compounds tested were as follows: MitoQ, a mitochondria-targeted antioxidant based on ubiquinone (29); decylTPP, which contains the mitochondria-targeting triphenylphosphonium (TPP) cation and which is frequently used as a control compound to correct for nonspecific effects of MitoQ (21); SS31, a peptide composed of D-Arg-Dmt-Lys-Phe-NH 2 (30) (where Dmt represents 2,6-dimethyltyrosine), whose therapeu-tic effects are thought to be due to its interactions with mitochondria; CN-POBS, an inhibitor of mitochondrial O 2 . production by RET at complex I (12); and the antidiabetic biguanides

Figure 3. Effect of AOX expression on mitochondrial H 2 O 2 production. A,
AOX expression in mouse heart mitochondria. Heart mitochondria from AOX ϩ/Ϫ mice or WT littermate controls were analyzed by Western blotting for AOX using the mitochondrial outer membrane protein voltage-dependent anion channel as a loading control. Mitochondria from three separate AOX ϩ/Ϫ and WT mice were assessed. B, O 2 consumption by mitochondria from AOX ϩ/Ϫ and WT mice. Mitochondria (200 g of protein/ml) were incubated at 37°C in an oxygen electrode and respiration was initiated by the addition of succinate (10 mM) followed by KCN (1 mM) and propyl gallate (50 M). Traces are typical of experiments repeated with at least three independent mitochondrial preparations for each condition. C, H 2 O 2 production by heart mitochondria from AOX ϩ/Ϫ and WT mice. Mitochondria (200 g of protein/ml) were incubated with succinate (10 mM), and H 2 O 2 production was assessed. n ϭ 3 (WT) or 6 (AOX). D, ⌬ of heart mitochondria from AOX and WT mice. Mitochondria (500 g of protein/ml) were incubated at 37°C for 5 min with succinate (10 mM), and ⌬ was assessed. n ϭ 4. E, CoQ redox state of heart mitochondria from AOX and WT mice. Mitochondria (1 mg of protein/ml) were incubated at 37°C for 2 min with succinate (10 mM), and the CoQ redox state was assessed. n ϭ 4. *, p Ͻ 0.05; Error bars, S.E.  . production by RET to some extent ( Fig. 5A). As RET at complex I is very sensitive to the magnitude of ⌬ and redox state of the CoQ pool, these compounds may affect RET indirectly by altering these bioenergetic variables rather than by directly interacting with complex I to block RETdependent O 2 . production. Consistent with this interpretation, all of the compounds decreased ⌬ (Fig. 5B), and some affected the redox states of the CoQ and NAD(P)H pools (Fig. 5, C and  D). Together, these findings suggest that the effects of these compounds on RET at complex I may be indirect due to effects on ⌬ and perhaps on the redox states of the NAD(P)H and CoQ pools. production by RET at complex I to these two physiological variables.

The thermodynamic driving force for RET at complex I
Whereas Fig. 6 shows clearly that decreasing ⌬ and oxidizing the CoQ pool lowers RET, it does not provide a quantitative basis to allow us to infer whether the effects of nigericin shown in Fig. 2A and those of the various compounds shown in Fig. 5A are due to direct interactions with complex I itself or are indirect effects due to altering the driving forces of RET. Therefore, we next determined how O 2 . production during RET depends on the overall thermodynamic driving force across complex I. The direction of electron flow at complex I is determined by the balance of thermodynamic driving forces across the complex (Fig. 1). During forward electron transfer at complex I, two electrons are passed from NADH to CoQ.

Reaction 1
The driving force for the transfer of two electrons from NADH to CoQ is ⌬E h .

Complex I reverse electron transport
This driving force, 2⌬E h , is used to pump four protons across the mitochondrial inner membrane against the ⌬p. Hence, for forward electron movement to occur, the thermodynamic requirement is as follows.
However, when the ⌬p is high and/or the ⌬E h is decreased, RET can occur provided the following is true.
We can thus calculate the thermodynamic driving force (⌬G) for RET, where F is the Faraday constant (33).
Rearranging to express the thermodynamic driving force for RET as a positive number in V gives the following.
From this, we can use the data from the FCCP titration in Fig.  2 to calculate Ϫ⌬G/F, the driving force for RET (see "Experimental procedures"). This analysis yields a plot of H 2 O 2 production by RET as a function of the thermodynamic driving force across complex I (Fig. 7A). This shows that when Ϫ⌬G/F Ͻ 0, there is a residual background level of ROS production, but as soon as the driving force for RET passes a threshold and Ϫ⌬G/F Ͼ 0, there is a dramatic and steep increase in O 2 . production by RET. This analysis confirms that the O 2 . production by RET requires a sufficient thermodynamic force to reverse electron transport at complex I and further shows the steep dependence of O 2 . production on this driving force. A benefit of describing H 2 O 2 production by RET at complex I as a function of its overall thermodynamic driving force is that it enables us to quantify whether compounds that affect O 2 .
production by RET do so by altering the drivers of this process or by acting directly on complex I. A compound that only affects O 2 . production by RET indirectly through altering ⌬p and ⌬E h would lie on the curve shown in Fig. 7A. In contrast, a compound that directly affected complex I independently of the thermodynamic drivers of RET, would lie below this curve. We first applied this analysis to nigericin, which decreases H 2 O 2 production by RET ( Fig. 2A) but which also led to a more oxidized CoQ pool (Fig. 2C) without affecting ⌬p. Carrying out this analysis, including accounting for changes in matrix pH on E h of the NADH and CoQ pools (see "Calculations") indicated that the decrease in H 2 O 2 production by nigericin was due to its effects on the CoQ pool redox state (Fig. 7A). In contrast, the decrease of H 2 O 2 production by the complex I inhibitor rotenone ( Fig. 2A) was not due to changes in the thermodynamic driving forces for RET, as these data lay below the trend line in Fig. 7A. If the compounds that affect mitochondrial ROS production assessed in Fig. 5 decrease mitochondrial ROS production independently of the drivers of RET, then they should lie below the trend line shown in Fig. 7A, as was the case for rotenone. When the data from Fig. 5 were analyzed to show the effect of these compounds on the thermodynamic driving forces for RET, the results lay above the trend line, showing the dependence of H 2 O 2 production on the thermodynamic driving forces across complex I (Fig. 7B). Hence, these data suggest that the effects of the compounds analyzed in Fig. 5 on ROS production by RET are more likely to be accounted for by their effects on ⌬p and/or ⌬E h rather than due to specific inhibitory effects on complex I.

Discussion
We investigated the dependence of O 2 . production by RET at complex I within isolated mitochondria on E h of the NAD(P)H and CoQ pools, ⌬p, matrix pH, and [O 2 ]. This approach confirmed that O 2 . production by RET at complex I is favored by a high ⌬p and a reduced CoQ pool, with exquisite sensitivity to small changes in these two drivers (Fig. 6). This analysis was extended to show that O 2 . production by RET at complex I was also highly responsive to small changes in the overall thermodynamic driving force for RET across the complex (Fig. 7A). Two sites have been proposed for O 2 . production by complex I during RET: the FMN of the complex I NADH-binding site (9,34,35) or the CoQ-binding site (36,37 . production (40).
A further point to note is that the term RET is often interpreted as requiring NAD ϩ reduction at the FMN site of complex I. This is not the case, as O 2 . production by RET at complex I occurs when the NAD ϩ /NADH pool is highly reduced and there is no net electron flow from complex I into this pool (e.g. Fig. 2D). Thus, we favor a model in which O 2 reacts with a FMNH Ϫ to generate O 2 . , whereas the FMN still exchanges electrons with the matrix NAD ϩ /NADH pool, although there is no net electron transfer (Fig. 8).
The generation of O 2 . by RET at complex I could be described as a function of the thermodynamic driving forces, ⌬p and ⌬E h , across the complex. As well as illustrating the factors that drive RET, this analysis allowed us to integrate the effects of the forces driving RET. Applying this to the decrease in RET when the matrix pH decreased suggested that the lower rate of RET could be accounted for by the change in the ⌬E h between the CoQ and NADH pools and may not be due to a direct effect of pH on complex I itself (24). Furthermore, this approach suggests that compounds such as MitoQ and metformin that affect mitochondrial ROS metabolism may do so indirectly, rather than by specific interactions with complex I. However, it is important to note that the calculation of Ϫ⌬G/F requires a number of assumptions and combines several technically challenging experimental measurements; hence, systematic errors will affect accuracy and precision. In addition, we have assumed that the CoQ pool interacts to the same extent with all complexes, and any effects of supercomplex formation on this were not considered (42). Nevertheless, our work does indicate novel approaches for determining how compounds impact mitochondrial ROS production and has implications for the interpretation of experiments using these compounds. For example, MitoQ is widely used in vitro and in vivo, where it acts as a chain-breaking antioxidant decreasing oxidative damage (29,43). As the excessive accumulation of hydrophobic TPP compounds disrupts mitochondrial function (44,45), controls with compounds with matched physicochemical properties are essential to correct for nonspecific effects. Thus, some "antiox-

Complex I reverse electron transport
idant" effects of lipophilic cations may be due to mild disruption of ⌬p that lowers mitochondrial O 2 . production by RET.
Similarly, some of the beneficial effects of less hydrophobic cations, such as metformin in vivo, may be associated with limiting O 2 . production by RET, in addition to stimulating AMP-activated protein kinase by inhibiting complex I (46,47). Mitochondrial O 2 . production by RET was initially investigated in vitro, and at that time its relevance to in vivo physiology was not considered (10). Recently, mitochondrial ROS production by RET has been demonstrated in multiple situations in vivo. For example, during ischemia, high levels of succinate accumulate, and it is oxidized upon reperfusion to drive a burst of ROS via RET (14,48). More generally, RET at complex I occurs as a redox-signaling pathway in inflammation (13), contributes to lifespan in flies (49), and is part of the oxygen-sensing mechanism of the carotid body (49). The linear dependence of ROS production by RET on O 2 (Fig. 4), which has been shown previously by others (50 -53), further adds to its appeal as a potential component of the carotid body O 2 sensor (49). One possibility is that O 2 . production by RET at complex I accounts for most of the redox signaling from the mitochondrion to the rest of the cell (3)(4)(5)(6). The appeal of RET as a mitochondrial redox signal is illustrated in Figs. 6 and 7, which show the tremendous sensitivity of RET to ⌬p and the redox status of the CoQ pool. The magnitude of ⌬p is directly linked to ATP demand, whereas the redox state of the CoQ pool reflects electron supply to the respiratory chain. Thus, RET provides a sensitive mechanism for the real-time feedback of the most critical aspects of mitochondrial status to the rest of the organelle and to the cell (Fig. 8).
In summary, we have provided a thermodynamic underpinning to O 2 . production by RET at complex I. This analysis highlights the potential for this mechanism to play a key role in mitochondrial redox signaling and demonstrates how to investigate the effects of RET on various physiological and pathological processes.

Mitochondria isolation
All procedures were performed in accordance with the UK Guide for the Care and Use of Laboratory Animals (PPL: 70/7538). Rat hearts were collected from 10 -12-week-old female Wistar rats (Charles River). C57Bl/6 mice carrying a single copy of the C. intestinalis AOX gene in the Rosa26 locus were generated as described (16). AOX mice and their WT littermate controls of both sexes were used at 8 -12 weeks of age to prepare heart mitochondria. Rats were killed by stunning followed by cervical dislocation. Mice were killed by cervical dislocation only. Hearts were removed into ice-cold STE buffer (250 mM sucrose, 5 mM Tris-HCl (pH 7.4, KOH), and 1 mM K-EGTA, supplemented with 0.1% (w/v) fatty acid-free BSA. Heart mitochondria were isolated by homogenization and differential centrifugation (700 ϫ g for 3 min; 3 ϫ 5,500 ϫ g for 10 min) at 4°C. Mitochondrial protein content was determined using the bicinchoninic acid assay with BSA as a standard.

Hydrogen peroxide efflux
H 2 O 2 efflux from mitochondria was assayed using a plate reader fluorometer (SpectraMax GeminiXS, Molecular Devices; used at medium sensitivity). Resorufin (the product of Amplex Red oxidation) fluorescence was detected using ex ϭ 570 nm and em ϭ 585 nm. Mitochondria (2 mg protein/ml) were incubated with 2.5 M Amplex Red (Invitrogen), 5 units/ml HRP in STE with 10 mM potassium succinate at 37°C. H 2 O 2 production rates were linear over 10 min but were measured from 0 to 2 min to facilitate comparison with other measurements. The H 2 O 2 response was calibrated using freshly prepared H 2 O 2 standards (⑀ 240 ϭ 43.5 M Ϫ1 cm Ϫ1 ) that were added sequentially to mitochondrial incubations lacking only succinate to generate a linear calibration curve. Mitochondriatargeted test compounds were added 30 s before measurement.

Mitochondrial ⌬⑀
Mitochondrial ⌬⑀ was measured by the uptake of radiolabeled [ 3 H]TPMP as described (21). Mitochondria (2 mg protein/ml) were incubated at 37°C with 10 mM succinate, 500 nM TPMP supplemented with [ 3 H]TPMP (50 nCi/ml) with the test compounds for 2 min in 250 l of medium in Eppendorf tubes. Mitochondria were pelleted by centrifugation (10,000 ϫ g for 30 s). Supernatant (200 l) was removed, and the pellets were dried with a rolled-up tissue and then solubilized in 40 l of 20% (v/v) Triton X-100. Both the supernatant and pellets were then added to scintillant (Ultima-Gold liquid scintillant, Perkin-Elmer Life Sciences) and incubated for 1 h at room temperature and then vortexed, and [ 3 H]TPMP content was assessed using a TriCarb LCS counter (PerkinElmer Life Sciences) counter with appropriate quench controls. To calculate the ⌬⑀, first the accumulation ratio was calculated assuming a mitochondrial matrix volume of 0.6 l/mg protein, and the mitochondrial ⌬⑀ was then calculated from the Nernst equation, assuming 40% binding of TPMP and that this was independent of mitochondrial ⌬⑀ and consistent across all conditions (21).

CoQ extraction and detection
Mitochondria (2 mg of protein/ml) were incubated in 500 -700 l of STE with 10 mM potassium succinate at 37°C on a shaking heat block for 2 min. At the end of the incubation, mitochondria were rapidly pelleted by centrifugation (10,000 ϫ g for 30 s), the supernatant was removed, and pellets were snapfrozen in a dry ice/ethanol bath and stored at Ϫ80°C until analysis. Immediately before HPLC analysis, the pellets were homogenized in 0.5 ml of ice-cold, nitrogen-purged 1-propanol in an ice-cold glass-on-glass homogenizer, 100 l of ice-cold H 2 O was added, and the samples were centrifuged (16,000 ϫ g at 4°C for 5 min). Supernatants (200 l) were immediately analyzed by HPLC on a 150 ϫ 4.6-mm, 3 Hypersil ODS column (Thermo). Solvent A was MeOH, 50 mM NaClO 4 ; solvent B was EtOH, 50 mM NaClO 4 . The gradient was 60% to 50% A over 15 min at a flow rate of 0.8 ml/min at 45°C. Identity was established by retention time compared with authentic standards at the absorbance maxima for CoQ and CoQH 2 (260 and 290 nm, respectively). CoQ 9 redox state was determined from the peak areas of ubiquinone and ubiquinol at 292.5 nm, the isosbestic point for oxidized and reduced CoQ 9 . The percentage reduc-Complex I reverse electron transport tion of the CoQ 9 pool was calculated as area of the reduced peak divided by the sum of both peak areas. Control incubations under conditions designed to maximally oxidize and reduce the CoQ pool demonstrated that this approach accurately reported the CoQ redox state, as the percentage reduction increased to ϳ80% in the presence of cyanide or anoxia, whereas inhibition of the respiratory chain with malonate decreased the percentage reduction to ϳ20%.

Redox state of mitochondrial NAD(P)H/NAD(P) pools
The redox state of the NAD(P)H/NAD(P) ϩ pool was determined by monitoring NAD(P)H fluorescence using a plate reader fluorimeter (SpectraMax GeminiXS; Molecular Devices) using ex ϭ 365 nm and em ϭ 450 nm. Mitochondria were incubated as described for measuring H 2 O 2 efflux. The signal was calibrated by subtraction of background fluorescence (mitochondria with no additions), and maximal reduction of the pool was set by incubating mitochondria with 5 mM malate and 5 M rotenone for 5 min.

Combined respirometry and H 2 O 2 production measurements
Combined respiration and H 2 O 2 production by mitochondria was assessed using an Oxygraph2K (O2K) respirometer (Oroboros, Innsbruck, Austria) with a fluorescence LED module attachment. To assess the effect of O 2 concentration on H 2 O 2 production, mitochondria (200 -250 g of protein/ml) were suspended in 2 ml of KCl buffer (120 mM KCl, 10 mM Hepes, 1 mM EGTA, pH 7.2) supplemented with 50 units/ml SOD, 4 units/ml HRP, 0.2 mg/ml fatty acid-free BSA, 25 M Amplex Red with stirring at 37°C. Respiration was initiated by the addition of either 5 mM glutamate and 5 mM malate; 10 mM succinate; or 10 mM succinate with 5 M rotenone. Where indicated, incubations were supplemented at the start with 5 M rotenone, 500 nM FCCP, 1 M antimycin A, or 1 M nigericin. The concentration of O 2 was adjusted by bubbling the buffer with N 2 . Amplex Red fluorescence was measured via the O2K fluorometer, and the corresponding voltage changes were calibrated via titration of known amounts of H 2 O 2 (500 nM to 5 M) in the presence of mitochondria, SOD, fatty acid-free BSA, HRP, and Amplex Red.

Western blotting
Mitochondrial pellets (ϳ250 g of protein) were solubilized on ice in 50 l of lysis buffer (100 mM Tris, 300 mM NaCl, 0.05% Nonidet P-40, pH 7.4, supplemented with protease and phosphatase inhibitors (Roche Applied Science). Protein was then quantified by the BCA assay and diluted in 4ϫ loading buffer (Invitrogen), and 10 g of protein was separated by SDS-PAGE on a 10% gel and transferred to polyvinylidene difluoride membrane. Membranes were incubated with a 1:20,000 dilution of rabbit serum raised against two AOX peptides (FKIETNDST-DEPNIEVENFPC and CVNHDLGSRKPDEQNPYPPGQ (49)) and a mouse monoclonal antibody against the voltage-dependent anion channel (1:1,000; Abcam ab14734)) and visualized using a LI-COR Odyssey flatbed scanner with anti-mouse and anti-rabbit secondary antibodies conjugated to IRDye 680RD and IRDye 800CW, respectively.

Calculations
This section describes calculations required to determine the thermodynamic driving force for RET from the data in Figs. 2 and 5 to generate the graphs shown in Fig. 7. The thermodynamic driving force for RET is derived from Equation 5. In mV, it is as follows.
⌬p ϭ ⌬ ϩ 18.1 (Eq. 7) As the matrix pH is 7.7, E h for the NAD ϩ /NADH couple in mV is as follows.

Statistical analysis and experimental design
Data were expressed as mean Ϯ S.E., with p values calculated using a two-tailed Student's t test for pairwise comparisons whereas one-way analysis of variance (ANOVA) followed by Tukey's post hoc test was used for multiple comparisons. Statistical analyses were performed using GraphPad Prism version 7 software.