Complex III Releases Superoxide to Both Sides of the Inner Mitochondrial Membrane*

2 O 2 and superoxide. Incubation was stopped, and aconitase activity measurements were begun by the addition of 1 volume (100 (cid:1) l) of 50 m M Tris, 0.6 m M MnCl 2 , 60 m M citrate, 0.2% Triton X-100, 100 (cid:1) M NADP (cid:4) , and 1 unit of isocitrate dehydrogenase (Sigma). Fluorometric measurements were then started immediately (Fluoroskan-FL Ascent type 374 microplate reader). The “blank” to measure aconitase-indepen-dent NADP (cid:4) reduction consisted of the same buffer except with isoci- trate dehydrogenase omitted. The slope of the increase in NADPH fluorescence was taken as the amount of aconitase activity. Following previous work (29), the results were expressed as a percentage relative to control, which, in our case, consisted of untreated skeletal muscle mitochondria. mitochondria ( (cid:1) 0.4 mg/ml) were incubated in either Amplex red reaction buffer (with SOD) or with 5 (cid:1) M MCLA in 125 m M KCl, 5 m M HEPES, 2 m M K 2 HPO 4 . For comparative purposes, rates of superoxide release with MCLA and H 2 O 2 release with Amplex red were expressed relative to state 1 (mitochondria without added substrate), which was defined as ‘100.‘ As reported many times before, antimycin A (cid:4) succinate as well as rotenone (cid:4) glutamate/malate yielded a drastic increase in H 2 O 2 release ( white bars ). However, only antimycin A (cid:4) succinate (hence superoxide from Complex III) caused an increase in the MCLA signal (superoxide release, gray bars ). The MCLA superoxide release with rotenone (cid:4) glutamate/malate was not different from glutamate/malate alone. The MCLA chemiluminescent signal was insensitive to added catalase but was completely eliminated by the exogenous addition of 200 units/ml SOD ( black bars ). Each experiment was repeated at least three times with S.E. shown. Essentially the same result (no increase in superoxide release by addition of rotenone to glutamate/malate-respiring mitochondria) was obtained using DHE fluorescence to measure superoxide release (data not shown). We thus confirmed the conclusions reached from Fig. 1: 1) while Complex I produces large amounts of superoxide, none escapes from intact mitochondria, and 2) contrary to Complex I, Complex III readily releases superoxide from intact mitochondria. succ , succinate; G / M , glutamate/malate. Taken together, these data indicate that superoxide derived from Complex III was able to reach the mitochondrial matrix and that this superoxide from the Q o site semiquinone.

The mitochondrial electron transport chain is the main source of ATP in the mammalian cell and thus is essential for life (1). However, during energy transduction, a small number of electrons "leak" to oxygen prematurely (2)(3)(4), forming the oxygen free radical superoxide (O 2 . in its anionic form and HO 2 ⅐ in its protonated form), which has been implicated in the pathophysiology of a variety of diseases including Parkinson's, Hun-tington's, and Alzheimer's diseases as well as the aging process itself (5,6). The detrimental significance of mitochondrial electron transport chain-derived superoxide is well illustrated by the lethal phenotype of mice lacking the mitochondrial matrix superoxide dismutase (Sod2) gene (7). While an increasing number of investigators have focused their attention on the potential pathological effects of mitochondrial superoxide (and its derivatives), there is a dearth of information on the mechanisms of deleterious superoxide formation by the electron transport chain. Indeed the state of knowledge has not changed much since the mid-1970s (4). As such, diminishing the rate of mitochondrial free radical production remains an elusive therapeutic strategy in the treatment of disease in which superoxide is thought to be involved. In this work, we seek to expand this area of investigation. The basic facts of superoxide production can be summarized as follows. At the ultrastructural level, Complexes I and III are the main sites of mitochondrial superoxide production (2,8,9). In Complex I, the most likely sites of electron leakage are the iron-sulfur clusters (Refs. 8 and 10, although some evidence also points to the flavin (10)), while in Complex III, it is the Q o semiquinone (2,3,(11)(12)(13). However, at the atomic level, the chemical details of either reaction remain completely unknown (for example, how does oxygen actually reach its reduction site?).
An important area of controversy, directly relevant to understanding the mechanism of superoxide formation, is to which side of the inner mitochondrial membrane either Complex I or Complex III releases superoxide (either to the mitochondrial matrix side or the cytoplasmic side (2)). Anionic superoxide (O 2 . (aq) ) is highly membrane-impermeable (14 -17) such that biologically it is highly compartmentalized, i.e. there is no flux between the pools of matrix and cytoplasmic superoxide (16 -19). The first studies on electron transport chain-derived superoxide production concluded that most superoxide must be extruded to the matrix side since superoxide was readily released from antimycin A-or rotenone-treated submitochondrial particles (SMPs, 1 in which the matrix side faces the medium) but not from intact mitochondria (Refs. 20 -23; for reviews, see Refs. 24 and 25). However, concomitant measurements of H 2 O 2 production also revealed that only about half of the total electron leak in the presence of antimycin A (thus Complex IIIderived (21,24)) could be explained by net (outward) O 2 . release from SMPs. The view that most superoxide production was directed toward the matrix was unchallenged until very recently when the x-ray structure of Complex III was solved (Ref. 26; for a review, see Ref. 27). F. L. M. pointed out that the x-ray structure of Complex III unambiguously shows the ubiquinol oxidation site (Q o site), the locus of superoxide production in Complex III (11), to be located immediately adjacent to the intermembrane space (IMS) and quite distant from the matrix (2). Based on this structural evidence, it was argued that some fraction of superoxide derived from the Q o site must be released to the IMS and that, if the data supporting superoxide release to the matrix were correct, it was the ability of Q o site-derived superoxide to reach the matrix that begged an explanation. Taking the latter data to be correct (21,24,28,29), we formulated a simple model that could explain the release of Q o site superoxide to both sides of the inner mitochondrial membrane (2). Subsequently Han et al. (30) (35)). Alternatively, the technique used to reach the estimates of superoxide release in Refs. 31 and 33, and thereby the conclusion that all superoxide from Complex III is released into the IMS (31,33,34), is known to overestimate the flux of superoxide (36). In the present report (preliminary results were presented at the 2003 Society for Free Radical Biology and Medicine conference (37)), we address previous experimental shortcomings and we challenge the newly popular opinion that Complex III only releases superoxide to the cytoplasmic side of the inner membrane. Using a refined version of the assay of St. Pierre et al. (31), we show that while superoxide derived from the Q o site of Complex III was indeed released from intact skeletal muscle mitochondria, this net extramitochondrial superoxide release accounts for no more than ϳ50% of H 2 O 2 production (by extension ϳ50% of total superoxide production) from Complex III even in mitochondria lacking CuZn-SOD (from Sod1Ϫ/Ϫ mice (38)). To demonstrate that the remaining ϳ50% of superoxide is released into the matrix, we avoided the use of SMPs (noting the potentially confounding factors), instead using the aconitase inhibition assay (39) in intact mitochondria. Superoxide release by Complex III into the matrix was corroborated by the observation of a profound inhibition of aconitase activity, a sensitive target of superoxide exclusively located in that compartment. This inhibition of aconitase occurred even in the presence of exogenously added SOD and catalase, verifying that neither H 2 O 2 nor re-entry of released superoxide could account for this effect. Our data are thus fully in agreement with initial conclusions from studies in SMPs (24), confirming that Complex III-derived superoxide can indeed reach the ma-trix. We discuss two models of superoxide production that explain this observation.

EXPERIMENTAL PROCEDURES
Chemicals-Unless stated otherwise, all chemicals used in this study were obtained from Sigma.
Animals-All mice used in this study were in the C57B6/J background and housed in the vivarium of the Audie L. Murphy Veterans Affairs Hospital. CuZn-SOD knock-out (Sod1Ϫ/Ϫ) mice were generated by the laboratory of C. Epstein (38). Mice used in this study were between 3 and 8 months of age. Animals were anesthetized and sacrificed by cervical dislocation. All procedures were approved by the subcommittee for animal studies at the Audie L. Murphy Veterans Affairs Hospital.
Mitochondrial Purification-Mitochondria were purified from whole hind limb skeletal muscle (predominantly gastrocnemius and soleus) according to Chappell and Perry (40,41). Hind limb skeletal muscle was excised, weighed, bathed in 150 mM KCl, and placed in Chappell-Perry buffer with nagarse. The minced skeletal muscle was homogenized with an all glass homogenizer. The homogenate was centrifuged for 10 min at 600 ϫ g, and the supernatant was passed through two cheesecloth layers and centrifuged at 14,000 ϫ g for 10 min. The resultant pellet was washed once in modified Chappell-Perry buffer with 0.5% bovine serum albumin and once in modified Chappell-Perry buffer without bovine serum albumin. Mitochondria were used immediately. Respiration was measured with a Clark electrode as described by Estabrook (42 (31)(32)(33). A weakness of this approach is that superoxide production is overestimated because superoxide can both reduce and oxidize HRP (36,44). To address this problem of the previous procedure (31-33), we added 20 M acetylated cytochrome c (cyt c) to the Amplex red reaction buffer to act as a superoxide "sink" (instead of HRP). As such, in the absence of SOD, extramitochondrially released superoxide reduces acetylated cyt c (a reaction that does not yield H 2 O 2 as a product). When SOD is added, extramitochondrial superoxide reacts with it, forming H 2 O 2 instead. Thus, the increase in H 2 O 2 formation upon SOD addition can be used to estimate net extramitochondrial superoxide release. Superoxide Production-Superoxide production was measured by three direct methods: 1) MCLA (modified Cypridina luciferin analog) chemiluminescence, 2) dihydroethidine (DHE) fluorescence, and 3) inhibition of aconitase activity.
The photomultiplier was set to default with an integration time of 1000 ms. The MCLA signal was quantified as an integral of 20 s of continuous measurement and expressed as relative luminescence units/mg of mitochondrial protein. The reaction was conducted in 100 l of reaction buffer containing ϳ0.5 mg/ml mitochondrial protein. The reaction buffer contained 125 mM KCl, 10 mM HEPES, 5 mM MgCl 2 , 2 mM K 2 HPO 4 , and 5 M MCLA. As a positive control, we used the xanthine/ xanthine oxidase system (49). 1 mM xanthine ϩ 0.1 unit/ml xanthine oxidase caused a ϳ100-fold increase in chemiluminescent signal as compared with xanthine or xanthine oxidase alone, which did not differ significantly from reaction buffer alone. Addition of SOD (CuZn-SOD from erythrocytes) decreased the MCLA signal by over 98% (data not shown).
As a confirmation of the MCLA assay, we used dihydroethidine (Molecular Probes). This compound is oxidized by superoxide to a fluorescent product with an excitation maximum at 498 nm and an emission maximum at 598 nm (50). DHE (50 M in the same buffer as described above) was tested with xanthine ϩ xanthine oxidase. Confirming previous work (50), a steady increase in fluorescence could be detected for which the slope (the rate of superoxide formation) was reduced to zero by addition of 100 units/ml SOD (data not shown).
The third method of superoxide detection used in this work is the aconitase inhibition method developed by Gardner and Fridovich (29,39,51). Aconitase catalyzes the reversible isomerization of citrate to isocitrate (cis-aconitate being the intermediate). A low redox potential 4Fe-4S iron-sulfur cluster is required for this activity, and superoxide reacts with the latter at around ϳ10 6 M Ϫ1 s Ϫ1 (51, 52), inactivating the enzyme. While it is known that other reactive species, such as H 2 O 2 , can also inactivate aconitase, the reaction of aconitase with superoxide is several orders of magnitude faster than that with H 2 O 2 . Thus, aconitase activity is a sensitive index of superoxide levels both in vivo and in vitro (18,28,29,53,54). Aconitase is usually present in both the mitochondrial matrix and the cytoplasm (but encoded by two different genes resulting in different molecular weight proteins). In skeletal muscle, however, only mitochondrial matrix aconitase is present (aconitase 2), making this tissue ideal for probing the question of superoxide release directed specifically toward the matrix.
Aconitase activity is assayed (in detergent-dispersed samples) by measuring NADP ϩ reduction by citrate in the presence of isocitrate dehydrogenase (the conversion of citrate to isocitrate by aconitase being the rate-limiting step). We used a fluorometric method (excitation at 355 nm and emission at 460 nm) to quantify the reduction of NADP ϩ . Mitochondria (ϳ0.4 mg of protein/ml) were aliquoted in 96-well plates (100 l of pH 7.44, 125 mM KCl, 10 mM HEPES, 5 mM MgCl 2 , 2 mM K 2 HPO 4 ) and incubated at 30°C up to 40 min as indicated. Substrates (9 mM succinate and 5 mM glutamate/malate) and inhibitors (10 M antimycin A, 10 M rotenone, 0.4 M stigmatellin, and 0.4 M myxothiazol) were added as indicated. For each experiment, 200 units/ml SOD and 900 units/ml catalase were also added to eliminate extramitochondrial H 2 O 2 and superoxide. Incubation was stopped, and aconitase activity measurements were begun by the addition of 1 volume (100 l) of 50 mM Tris, 0.6 mM MnCl 2 , 60 mM citrate, 0.2% Triton X-100, 100 M NADP ϩ , and 1 unit of isocitrate dehydrogenase (Sigma). Fluorometric measurements were then started immediately (Fluoroskan-FL Ascent type 374 microplate reader). The "blank" to measure aconitase-independent NADP ϩ reduction consisted of the same buffer except with isocitrate dehydrogenase omitted. The slope of the increase in NADPH fluorescence was taken as the amount of aconitase activity. Following previous work (29), the results were expressed as a percentage relative to control, which, in our case, consisted of untreated skeletal muscle mitochondria.

Complex I-derived Superoxide Is Vectorially Released into the Matrix and Does Not Escape from Intact Mitochondria-When
mitochondria respire with an NADH-linked substrate (such as glutamate/malate), rotenone inhibition causes a profound increase in superoxide production (8,55). It was reported that exogenous SOD addition does not further increase H 2 O 2 release from rotenone-inhibited mitochondria (31). This observation was taken as evidence that no Complex I-derived superoxide was released from mitochondria and that Complex I releases superoxide exclusively into the matrix, although St. Pierre et al. (31) cautioned that this was by no means conclusive since dismutation of superoxide by CuZn-SOD (whether in the intermembrane space or as a cytoplasmic contaminant) could not be ruled out.
Here we expand on this observation using a refined technique and using mitochondria isolated from mice lacking CuZn-SOD. Addition of 200 units/ml exogenous SOD did not increase the rate of mitochondrial H 2 O 2 release (as measured with Amplex red with 20 M acetylated cyt c) in glutamate/ malate ϩ rotenone-treated mitochondria even when the mito- chondria were derived from CuZn-SOD null (Sod1Ϫ/Ϫ) mice (Fig. 1A).
Since this indirect method is not suited to detecting low levels of superoxide, we wished to follow up on this observation by measuring superoxide release directly. Superoxide release was measured chemiluminescently with 5 M MCLA on the same skeletal muscle mitochondria preparations that were used above for H 2 O 2 assays. For comparative purposes, the data in Fig. 2 were normalized with respect to oxidant production in state 1 (i.e. mitochondria without substrate (45)). Addition of glutamate/malate increased the MCLA signal by around ϳ2-fold compared with mitochondria alone (state 1), but addition of rotenone with glutamate/malate did not result in any further increase in MCLA chemiluminescence as compared with glutamate/malate alone (Fig. 2). Yet rotenone ϩ glutamate/malate drastically increased H 2 O 2 release (measured with Amplex red), indicating that while superoxide was not being released from, it was still being produced in mitochondria, most likely being released into the matrix. In corroboration of this idea, a strong MCLA signal was observed in inverted SMPs (in which the matrix side faces the outside medium) in response to treatment with 100 M NADH ϩ rotenone (data not shown). This result was reported several times previously using different methods (22,23) and indicates that superoxide produced by Complex I under rotenone inhibition is released into the matrix. To prove that this is also true in intact mitochondria, we measured inhibition of mitochondrial matrix aconitase activity, a sensitive target of superoxide. Addition of 10 M rotenone to mitochondria respiring on 5 mM glutamate/ malate resulted in a profound inhibition of aconitase activity (Fig. 3) Fig. 2, data show that, in contrast to the effects of rotenone, antimycin A readily induced a large MCLA chemiluminescent signal in intact mitochondria that was completely eliminated by addition of exogenous SOD. These data thus indicate that Complex III can release superoxide to the cytoplasmic side of the inner mitochondrial membrane (i.e. into the IMS). We sought to quantify what fraction of Complex III-derived superoxide could be accounted for by this extramitochondrial superoxide release. . or H 2 O 2 from playing any part in the observed aconitase inhibition. Aconitase activity was measured 17 min after the incubation of mitochondria with substrates (and inhibitors) was begun. Following previous work (29), results were expressed as a percentage of initial (before incubation) aconitase activity. The experiment was repeated three times with S.E. shown. For comparative purposes, rates of superoxide release with MCLA and H 2 O 2 release with Amplex red were expressed relative to state 1 (mitochondria without added substrate), which was defined as '100.' As reported many times before, antimycin A ϩ succinate as well as rotenone ϩ glutamate/malate yielded a drastic increase in H 2 O 2 release (white bars). However, only antimycin A ϩ succinate (hence superoxide from Complex III) caused an increase in the MCLA signal (superoxide release, gray bars). The MCLA superoxide release with rotenone ϩ glutamate/malate was not different from glutamate/malate alone. The MCLA chemiluminescent signal was insensitive to added catalase but was completely eliminated by the exogenous addition of 200 units/ml SOD (black bars). Each experiment was repeated at least three times with S.E. shown. Essentially the same result (no increase in superoxide release by addition of rotenone to glutamate/malate-respiring mitochondria) was obtained using DHE fluorescence to measure superoxide release (data not shown). We thus confirmed the conclusions reached from Fig. 1: 1) while Complex I produces large amounts of superoxide, none escapes from intact mitochondria, and 2) contrary to Complex I, Complex III readily releases superoxide from intact mitochondria. succ, succinate; G/M, glutamate/malate. In Fig. 1, we present data addressing this question. We improved upon the methodology of previous investigators (Refs. 31 and 33 and see "Experimental Procedures") and performed the assays in mitochondria from mice lacking CuZn-SOD (38). In mitochondria treated with rotenone and antimycin A and supplemented with succinate, addition of SOD increased H 2 O 2 production ϳ2-fold, indicating that ϳ50% of total H 2 O 2 generated by Complex III can be explained by direct extramitochondrial release of superoxide. This leaves ϳ50% of total electron leak unaccounted for, yet our experiments conducted with Sod1Ϫ/Ϫ mitochondria ruled out dismutation of superoxide in the IMS. We hypothesize that the remaining 50% of H 2 O 2 release resulted from superoxide released into the matrix (and subsequently dismutated by Mn-SOD).
Complex III-derived Superoxide Reaches the Mitochondrial Matrix-Pioneering data using SMPs suggested that Complex III-derived superoxide can reach the mitochondrial matrix (20,21,24). In the present study we revisited this question in intact mitochondria using aconitase inactivation to measure superoxide release into the mitochondrial matrix.
In Fig. 4, the effect of antimycin A on aconitase activity in succinate-supplemented mitochondria is shown. After a 40-min incubation, aconitase activity was decreased by about 95% as compared with succinate only-treated mitochondria. To eliminate the possibility of Complex I-derived superoxide production due to reverse electron flow, we performed the same experiments with added rotenone. These experiments were conducted with 900 units/ml catalase and 200 units/ml superoxide dismutase present to rule out that H 2 O 2 or re-entry of superoxide was responsible for aconitase inactivation (Figs. 4 and 5). Next we show that the antimycin A-dependent inhibition of aconitase was largely prevented by addition of 0.4 M stigmatellin (Fig. 5), which binds to the distal Q o site and prevents superoxide formation at Complex III (12,27,56,57). Although we cannot rule out completely that superoxide derived from other sources may play a role in the inhibition of aconitase brought on by antimycin A, the fact that this can be largely prevented by stigmatellin indicates that the majority of aconitase inhibi-tion was solely due to superoxide derived from the Q o site. Finally we repeated the experiments in Fig. 5 at pH 8.75 and 6.75 and in the presence of the uncoupler carbonyl cyanide p-trifluoromethoxyphenylhydrazone. These conditions did not change the results obtained in Fig. 5 (data not shown). We conclude from these data that superoxide derived from the Q o site does, in fact, reach the matrix by a pathway that is independent of bulk pH and proton gradient.  (16,44). The protonated form of superoxide, HO 2 ⅐ (pK a , 4.9), is readily membrane-permeable but is only present in very low amounts at physiological pH (less than 0.2% at pH 7.44 (44)). Thus, in the cell each membranebound compartment has its own SOD isozyme, and overexpression of one will not compensate for the loss of another SOD isozyme in a different compartment (18,19). Given the fact that superoxide is very selective in its targets (58) and that once released into a membrane-bound compartment it cannot escape, understanding the sidedness of superoxide release by its sites of generation is essential to clarify the role of mitochondrial superoxide production in pathophysiology.

Vectorial Release of Complex I-derived Superoxide into the
In rotenone-inhibited Complex I (in the presence of NADHlinked substrates), the oxygen reduction site (or locus of superoxide generation) is somewhere between the flavin (FMN) and the rotenone binding sites (8,23,59). Since the x-ray structure of Complex I has not yet been elucidated, the exact position of the rotenone binding site remains unknown. However, the rough outline of Complex I is known from cryoelectron microscopy (60,61); the approximate loci of the rotenone and the flavin binding sites are known from biochemical and mutagenic studies (62). Finally it is also known that iron-sulfur clustermodifying agents (chloromercury benzoate and ethoxyformic FIG. 4. Antimycin A treatment drastically inhibits mitochondrial matrix aconitase activity even in the presence of extramitochondrial SOD and catalase. ϳ0.4 mg/ml mitochondrial protein was incubated at 30°C in the presence of 200 units/ml SOD, 900 units/ml catalase, 9 mM succinate, 10 M antimycin A, and 10 M rotenone as indicated. Rotenone was added to prevent reverse electron transfer through Complex I. The results were expressed as a percentage of initial aconitase activity (untreated mitochondria). The fact that antimycin A was able to inhibit aconitase activity despite the presence of exogenous SOD and catalase can only be explained by superoxide from Complex III released directly into the matrix. The reaction was stopped with the addition of aconitase reaction buffer (see "Experimental Procedures") at 0 (white bars), 10 (light gray bars), 20 (dark gray bars), and 40 min (black bars). The experiment was repeated at least four times with S.E. shown. anhydride) profoundly decrease reactive oxygen species production elicited by rotenone with NADH-linked substrates (8). Thus, the site of superoxide formation must be somewhere on the hydrophilic fraction, which consists of the matrix-protruding arm of Complex I (Fig. 6). One would then hypothesize that superoxide produced from this locus should be exclusively released into the mitochondrial matrix. The data presented in this study support this hypothesis.
Using direct methods of superoxide detection (MCLA and DHE), we failed to observe a significant increase in superoxide release from intact skeletal muscle mitochondria (Figs. 1 and 2) in response to the addition of rotenone with glutamate/malate (an NADH-linked electron source). Yet the same treatment, as was reported many times before (8,55), drastically increased H 2 O 2 production (indicating that superoxide was being produced in, but not released out from, intact mitochondria). Rotenone ϩ glutamate/malate led to a profound inhibition of aconitase activity in intact mitochondria, indicating that superoxide was being released into the matrix (Fig. 3). The question why MCLA (and DHE) would not react with matrix superoxide inevitably arises, since these probes are relatively large hydrophobic molecules that readily permeate cells, thus readily entering into the mitochondrial matrix. The simple answer is that both probes probably reach the matrix but are vastly out-competed (for reacting with superoxide) by the ϳ10 M Mn-SOD in that compartment (Table I). Based on the kinetic data in Table I 6. The exclusive release of Complex I-derived superoxide into the mitochondrial matrix is in agreement with the hypothesis that the iron-sulfur clusters are the locus of oxygen reduction (8). Cryoelectron microscopy studies have delineated a "shoelike" general shape of Complex I (61). EPR as well as sitedirected mutagenesis studies place the iron-sulfur clusters in the long hydrophilic, matrix-protruding arm (62). The rotenone binding site is located at the hydrophobic/hydrophilic interface (62). Our present data indicate that, in the presence of rotenone, Complex I releases superoxide exclusively into the mitochondrial matrix. This result is exactly what one would expect from the current notion that it is the hydrophilic arm iron-sulfur clusters that are the source of superoxide (8). UQH 2 , ubiquinol.

FIG. 5. Stigmatellin, which blocks superoxide production at the Q o site (13, 37, 57), prevents matrix aconitase inactivation.
Mitochondria were treated with 10 M rotenone (to prevent electron backup through Complex I) and 9 mM succinate (control) as well as 10 M antimycin A and 0.4 M stigmatellin as indicated. The inhibitors were preincubated with mitochondria for 5 min before succinate addition (to allow equilibration). The reactions (with added SOD and catalase) were then incubated at 30°C for 20 min at which time aconitase activity was measured. Each bar represents the average of three experiments with S.E. shown. Antimycin A caused a substantial inhibition of aconitase activity, indicating superoxide release into the matrix. Concomitant addition of stigmatellin with antimycin A largely prevented this inhibition (very similar results were obtained with myxothiazol, data not shown). Stigmatellin is a Q o site inhibitor known to block antimycin A-induced superoxide formation from Complex III (37,56,57). Stigmatellin prevents superoxide formation by sterically blocking ubiquinol binding and formation of the Q o semiquinone (12,27). Taken together, these data indicate that superoxide derived from Complex III was able to reach the mitochondrial matrix and that this superoxide originated from the Q o site semiquinone. SOD) also failed to show any increase in extramitochondrial O 2 . release in response to rotenone ϩ glutamate/malate, confirming the aforementioned results. Since more or less identical results were obtained with mitochondria from mice lacking CuZn-SOD (which is present in the intermembrane space and the cytoplasm (63)), we may conclude that, despite high levels of Complex I superoxide generation toward the matrix, no detectable levels escape from intact mitochondria. This finding strongly supports the hypothesis that the locus of superoxide production in Complex I is somewhere on the hydrophilic, mitochondrial matrix-protruding arm (Fig. 6). We suggest that the simplest mechanism for superoxide production at this site is a single electron transfer to O 2 in the aqueous phase. An additional implication from our results is that the extramitochondrial superoxide release in the absence of respiratory inhibitors (Fig. 2, "succinate" and "Glutamate/malate") cannot be due to Complex I, and we posit that it must have originated from Complex III.
Complex III-derived Superoxide Is Released to Both the Matrix and the Intermembrane Space-Much more is known about the structure (several independent x-ray structures now published (26,64,65)) and mechanism of proton pumping (the modified Q cycle (27)) of Complex III than of Complex I. Yet despite 30 years of research, the mechanism of superoxide production at Complex III remains incompletely understood. In brief summary, Complex III superoxide production was first observed as H 2 O 2 released by mitochondria in response to antimycin A (21, 66, 67) that was subsequently (in experiments with SMPs) demonstrated to have originated as the stoichiometric dismutation product of O 2 . (21,68). The experiments with SMPs also led to the long held conclusion that Complex III-derived superoxide must be released exclusively to the mitochondrial matrix (for a review, see Ref. 24) because SMPs are inverted (meaning the matrix side faces the medium). Subsequently Turrens et al. (11) determined that the Q o semiquinone, rather than cytochrome b L (for a detailed review of the structure and function of Complex III, see Ref. 27), was the source of superoxide under antimycin A treatment. More recently, the resolution of the x-ray structure demonstrated unambiguously that the Q o site is immediately adjacent to the intermembrane space (Fig. 7). The close proximity of the Q o site to the IMS previously led us to conclude that a fraction of Complex III-derived superoxide must invariably be released to the cytoplasmic side of the inner membrane (2). This was experimentally demonstrated shortly afterward (30). The position of the Q o site in the x-ray structure of Complex III also made the possibility of superoxide release into the matrix much more problematic, considering the high membrane impermeability of anionic superoxide (14 -16); indeed most investigators have taken the position that Complex III only releases superoxide to the cytoplasmic side of the inner mitochondrial membrane (31,33), in disagreement with the initial work in SMPs (21,68). Despite some in vivo evidence (29), data from SMPs could be explained away as an artifact of preparation (for example, incompletely inverted SMPs, sheets, or other membrane fragments (35)). We believe, however, the SMP data to be consists of a dimer of 11 subunits each; for simplicity only a monomer of the three catalytically active subunits is shown. Cytochrome b is in red, cytochrome c 1 is in blue, and the Rieske iron-sulfur protein is in yellow. To the right is a schematic representation of the structure with the major cofactors and ubiquinone binding sites indicated. The ubiquinol oxidation site (Q o site), the locus of superoxide formation (11), is formed mostly by cytochrome b and covered by the Rieske iron-sulfur protein.
In this x-ray structure, it is occupied by stigmatellin (26). Structures in which this site is occupied by myxothiazol and other Q o site inhibitors have also been published (27). For a detailed review on the structure and mechanism of Complex III, see Ref. 27. valid and that an objective reading of the literature circumstantially supports that Complex III releases superoxide to both sides of the inner mitochondrial membrane (2). In this work, we provide direct supporting evidence for this in intact mitochondria. St. Pierre et al. (31,32) and Han et al. (33) reported that up to 70% of the total H 2 O 2 release can be accounted for by net extramitochondrial superoxide release (31)(32)(33). However, the methodology used to estimate the flux of superoxide in previous work relied on HRP as the superoxide sink (which is known to cause overestimation of the superoxide flux (36), see "Experimental Procedures"). We modified the methodology taking the above into consideration. First, we used lower levels of HRP (1 unit of HRP/ml for the Amplex red assay used here versus Ͼ10 units/ml in Refs. 31 and 33). Second, we added 20 M acetylated cyt c as a superoxide sink. As shown in Fig. 1, ϳ35% of the total H 2 O 2 production from Complex III (succinate ϩ antimycin A) was due to net extramitochondrial superoxide release in wildtype mouse skeletal muscle mitochondria. As such, our data are about equal to the lower range of values of previous work (31)(32)(33) (69). We thus conclude that superoxide derived from the Q o site can indeed reach the matrix, independent of bulk pH, in agreement with previous work in SMPs and in vivo (29).
Two Models That Can Explain the Release of Complex IIIderived Superoxide to Both Sides of the Inner Membrane-As we pointed out in the Introduction, the ability of membrane-impermeable superoxide to reach the matrix requires an explanation. FIG. 8. Two models to explain how Complex III releases superoxide to both sides of the inner mitochondrial membrane. We propose two models to explain how superoxide derived from the Q o site can reach both sides of the inner mitochondrial membrane. In the first mechanism (shown in gray arrows), a neutral ubisemiquinone (QH ⅐ ) could diffuse out of the Q o site along the tunnel that (the x-ray structure shows) connects the Q o site to the bulk hydrophobic lipid phase. At the lipid/aqueous phase interface, the ubisemiquinone would deprotonate (the pK a is considerably lower in the aqueous phase as compared with aprotic solvents (27)) and react with oxygen to form aqueously solvated superoxide We have previously proposed two alternative hypotheses, outlined in Fig. 8 . would have to move against the membrane potential) even if hypothetical anion channels capable of channeling superoxide did exist. Liu (70) suggested that aqueous O 2 . in the IMS could be protonated to HO 2 ⅐ , which could then diffuse through the inner membrane. This model cannot explain our present findings since neither raising the pH to 8.75 nor adding an uncoupler (both treatments drastically decrease the concentration of aqueous protons in the IMS) attenuated the inactivation of matrix aconitase (data not shown).
But we argue that two previously proposed hypotheses can accommodate these new data. Our first hypothesis (13) posits that a neutral (uncharged) semiquinone escapes from the Q o site through the hydrophobic tunnel that connects the Q o site to the membrane lipid phase and opens up roughly in the middle of the inner mitochondrial membrane (see Fig. 8; for a detailed discussion on the structure of Complex III, see Refs. 2 and 71). From there, the neutral semiquinone would be able to diffuse to both sides of the inner membrane with more or less equal likelihood. At the lipid/aqueous phase interface, the semiquinone would release its proton and react with O 2 , forming aqueous, anionic O 2 . . Equal propensity of the semiquinone to diffuse to either side of the membrane would explain why superoxide release to both sides of the inner mitochondrial membrane is observed. How well does such a scenario fit with the current understanding of the catalytic mechanism of Complex III? The ability of the semiquinone to escape from the Q o site would be a natural consequence of the Q cycle in its current form (27,72,73) since the Q cycle proposes a very thermodynamically unstable, hence very loosely bound, Q o semiquinone (72). Our second hypothesis (2) states that oxygen reacts with a neutral semiquinone in the hydrophobic part of the Q o site to form HO 2 ⅐ . At first sight, this hypothesis would seem very similar to the proposal of Liu (70) thus suffering the same drawbacks. However, the critical difference between these models is that our hypothesis has HO 2 ⅐ forming in the hydrophobic Q o site by protons supplied from the oxidation of ubiquinol, independent of bulk pH. HO 2 ⅐ , contrary to O 2 . , can be formed in non-aqueous environments (74,75) and is readily membrane-permeable (14 -17). The proton for the formation of HO 2 ⅐ would come from either a direct proton-coupled electron transfer (76) from neutral, singly protonated semiquinone or via shuttling of the proton from the acidic residues in the Q o site (77). Our experiments with varied pH led us to conclude that the former most likely occurs, which would also be consistent with the pH independence of the energy of activation of Complex III (78), indicating that during catalysis the Q o site does not equilibrate with bulk aqueous phase protons (27). Once formed in the Q o site, HO 2 ⅐ would then diffuse along the hydrophobic tunnel into the lipid phase, which opens up roughly in the middle of the inner mitochondrial membrane. From there, HO 2 ⅐ could diffuse to both sides of the inner mitochondrial membrane with more or less equal probability. We have previously discussed in detail the feasibility, in terms of thermodynamics and structural requirements, of the HO 2 ⅐ hypothesis (2).
Arguing every detail of this hypothesis is beyond the scope of the present work, but an immediate objection must be addressed to make the hypothesis credible. It could be argued that HO 2 ⅐ is too reactive to cross membranes since it would react with polyunsaturated fatty acids at diffusion-controlled rates. To counter this point, empirical evidence already attests that HO 2 ⅐ is readily able to cross membranes containing polyunsaturated fatty acids (14,16), and the rate constant for HO 2 ⅐ reacting with linoleic, linolenic, and arachidonic acids is only ϳ10 3 M Ϫ1 s Ϫ1 (44,79,80), 6 orders of magnitude lower than the diffusion limit.
Future experiments will be needed to verify or falsify either of our proposed hypotheses. Nevertheless having demonstrated that the Q o site of Complex III releases superoxide toward both the mitochondrial matrix and the IMS, we argue that at present the two hypotheses we put forward best explain this observation.