Superoxide activates uncoupling proteins by generating carbon-centered radicals and initiating lipid peroxidation: studies using a mitochondria-targeted spin trap derived from alpha-phenyl-N-tert-butylnitrone.

Although the physiological role of uncoupling proteins (UCPs) 2 and 3 is uncertain, their activation by superoxide and by lipid peroxidation products suggest that UCPs are central to the mitochondrial response to reactive oxygen species. We examined whether superoxide and lipid peroxidation products such as 4-hydroxy-2-trans-nonenal act independently to activate UCPs, or if they share a common pathway, perhaps by superoxide exposure leading to the formation of lipid peroxidation products. This possibility can be tested by blocking the putative reactive oxygen species cascade with selective antioxidants and then reactivating UCPs with distal cascade components. We synthesized a mitochondria-targeted derivative of the spin trap alpha-phenyl-N-tert-butylnitrone, which reacts rapidly with carbon-centered radicals but is unreactive with superoxide and lipid peroxidation products. [4-[4-[[(1,1-Dimethylethyl)-oxidoimino]methyl]phenoxy]butyl]triphenylphosphonium bromide (MitoPBN) prevented the activation of UCPs by superoxide but did not block activation by hydroxynonenal. This was not due to MitoPBN reacting with superoxide or the hydroxyl radical or by acting as a chain-breaking antioxidant. MitoPBN did react with carbon-centered radicals and also prevented lipid peroxidation by the carbon-centered radical generator 2,2'-azobis(2-methyl propionamidine) dihydrochloride (AAPH). Furthermore, AAPH activated UCPs, and this was blocked by MitoPBN. These data suggest that superoxide and lipid peroxidation products share a common pathway for the activation of UCPs. Superoxide releases iron from iron-sulfur center proteins, which then generates carbon-centered radicals that initiate lipid peroxidation, yielding breakdown products that activate UCPs.

The mitochondrial respiratory chain is a major source of superoxide and derived reactive oxygen species (ROS) 1 in vivo (1,2). Pathological oxidative damage to mitochondria in diseases and aging is a consequence of this ROS production, although there are many protective and adaptive responses to prevent and repair oxidative damage (2,3).
Recently, it has been shown that the GDP-sensitive proton conductance catalyzed by uncoupling proteins (UCPs) increases on exposure to superoxide (4) or to lipid peroxidation breakdown products such as 4-hydroxy-2-trans-nonenal (HNE) (5). Three UCPs have been characterized in mammals (6). UCP1 increases the proton conductance of the mitochondrial inner membrane in brown adipose tissue (BAT) for thermogenesis; in contrast, the physiological role of its homologs UCP2 and UCP3 is uncertain. One possibility (7) is that UCPs limit the magnitude of the protonmotive force, ⌬p, and thus decrease mitochondrial ROS production, which increases dramatically at high ⌬p (8). The complete dependence of proton conductance through UCP2 and UCP3 on superoxide or HNE (4,5) suggests an appealing feedback mechanism in which ROS or oxidative damage activate uncoupling by UCPs, lower ⌬p, and decrease ROS production (7). That UCP2 and UCP3 have a role in preventing mitochondrial oxidative damage is supported by the increased expression of UCPs in response to elevated mitochondrial oxidative damage (6,9) and by the observation that oxidative damage is increased in mitochondria from UCP3 Ϫ/Ϫ mice (10,11).
To investigate the putative antioxidant role of UCPs, it is important to know whether superoxide and lipid peroxidation products interact with UCPs by distinct mechanisms (e.g. alternative allosteric sites on UCPs) or at different points on the same pathway (e.g. by superoxide exposure leading to the formation of lipid peroxidation products). To distinguish between these possibilities, antioxidants that block different components of the putative oxidative damage cascade leading from superoxide to peroxidation products such as HNE should be informative. Previously we found that the superoxide stimulation of UCPs was blocked by mitochondria-targeted antioxidants (12). These compounds are selectively directed to and accumulate within mitochondria due to their conjugation to the lipophilic triphenylphosphonium cation (13,14). Targeting derivatives of ubiquinol (MitoQ) or ␣-tocopherol (MitoVit E) to mitochondria showed that it was superoxide within the matrix that led to UCP activation (12). However, as both these compounds react with a range of ROS and lipid peroxidation intermediates (14,15), it was not possible to distinguish between superoxide itself or a downstream product of the ROS cascade as the UCP activator (12).
We have now synthesized a more selective mitochondriatargeted antioxidant, MitoPBN, which is derived from ␣-phenyl-N-tert-butylnitrone (PBN). The spin trap PBN was chosen because of its high reactivity with carbon-centered radicals (16,17), whereas its reactivity with superoxide and peroxyl radicals is low, and it does not prevent lipid peroxidation by a chain-breaking antioxidant mechanism (18). MitoPBN should be accumulated within mitochondria and there react preferentially with carbon-centered radicals but not with superoxide. Here we show that MitoPBN prevents the activation of proton conductance through UCPs by superoxide but not by HNE. Furthermore, a carbon-centered radical generator stimulates UCPs, and this activation is also blocked by MitoPBN. These data suggest that superoxide and lipid peroxidation products are both components of a single oxidative damage pathway that activates UCPs. This pathway starts with superoxide as the initiator of a cascade of phospholipid peroxidation reactions. These reactions form lipid peroxidation breakdown products (such as HNE) that then activate the UCPs.

EXPERIMENTAL PROCEDURES
Synthetic Chemistry-1 H (300 MHz) and 31 P (121 MHz) NMR spectra were recorded on a Varian Unity 300 spectrometer in chlorform-d 1 standardized against the solvent peak or external 85% (v/v) phosphoric acid, respectively. UV spectra were obtained on a Varian Cary 500 Scan spectrophotometer in 95% (v/v) ethanol solutions. Elemental analyses were performed in the Campbell Microanalytical Laboratory, University of Otago. High resolution mass spectra (ESMS) were obtained by Associate Professor G. Willett, School of Chemistry, University of New South Wales, Sydney, Australia.
[4-(4-Formylphenoxy)butyl]triphenylphosphonium Iodide (1 ϭ Mitobenzaldehyde, Scheme 1)-Sodium hydride (241 mg, 6.03 mmol, 60% (w/v) suspension in oil) was added to a dry Schlenk tube containing a magnetic stirrer and held under an argon atmosphere. The sodium hydride was washed 3 times with pentane and then dried in vacuo (0.1 mm Hg). Dimethylformamide (5 ml) was then added, and the suspension was stirred for 10 min at room temperature. A solution of phydroxybenzaldehyde (633 mg, 5.43 mmol) in dimethylformamide (5 ml) was added dropwise to the reaction vessel causing bubbling and the appearance of a yellow/orange precipitate. After 2.5 h stirring at room temperature, the reaction mixture had become an orange suspension. A solution of (4-iodobutyl)triphenylphosphonium iodide (3.11 g, 5.43 mmol) (19) in dimethylformamide (9 ml) was then added dropwise to the ice-cooled reaction mixture, which was subsequently allowed to warm to room temperature overnight to give a clear yellow solution. Distilled water (50 ml) was then added, and the mixture was partitioned with dichloromethane (3 times, 30 ml). The combined organic layers were dried (MgSO 4 ) and evaporated to dryness in vacuo. The residual oil was dissolved in minimal dichloromethane and precipitated with excess ether, and the solvent layer was decanted. The precipitate was then redissolved in minimal dichloromethane and precipitated with excess ether, and the solvent layer was decanted. This precipitation process was repeated 9 times. The residue was dried under reduced pressure for 3 h yielding the monohydrate of 1 as a pale yellow solid (2.65 g, 4.53 mmol, 84%). UV spectroscopy gave a max of 275 nm (⑀ 19,725 M Ϫ1 ⅐cm Ϫ1 . The NMR data were as follows: 1 H NMR ␦ 9.86 (1H, s, CHO), 7.6 -7.9 (17H, m, ArH), 6 (13 ml) and activated zinc (21) (2.75 g, 42.1 mmol) were added in single aliquots to an ice-cooled solution of 1 (1.98 g, 3.51 mmol) in dry absolute alcohol (13 ml) and 4-Å molecular sieves (4 g). This mixture was stirred under argon for 10 min. A mixture of glacial acetic acid (5.05 g, 84.2 mmol) in dry absolute alcohol (15 ml) was then added dropwise over 100 min. The mixture was then stirred for 2 h in an ice bath, after which time it was stored at 4°C for 9 days, with daily agitation. The white precipitate was filtered off, and the green solution was evaporated to dryness in vacuo. The residue was dissolved in dichloromethane (20 ml) and washed twice with aqueous potassium bromide (15 ml, 20%, pH 7.0). The organic layer was dried (MgSO 4 ) and evaporated in vacuo to a volume of 2 ml. This solution was then added dropwise to ether (100 ml) with brisk stirring. A pale yellow precipitate formed, and once it had settled the solvent layer was decanted. The resulting precipitate was dissolved in minimal dichloromethane and added dropwise to excess ether with brisk stirring. Once the precipitate had settled, the solvent layer was decanted. This precipitation process was repeated 3 times. The residue was dried in vacuo for 2 h to give 2 as a pale yellow solid (1.00 g, 1.69 mmol, 48%). UV spectroscopy gave a max of 305 nm (⑀ 18,960 m Ϫ1 ⅐cm Ϫ1 ). The NMR data were as follows: 1 Table I).
Measurement of Uptake of MitoPBN by Mitochondria-An ion-selective electrode for MitoPBN was constructed and inserted into the stirred chamber of an oxygen electrode thermostatted at 30°C (22)(23)(24). Calibration of the electrode response by additions of MitoPBN from 10 -30 M gave a response that was a linear function of log 10 [MitoPBN] with a slope of ϳ60 mV, as predicted by the Nernst equation.
Mitochondrial Preparations and Incubations-Rat liver mitochondria were prepared by homogenization and differential centrifugation (25). Oxygen electrode experiments with liver mitochondria were in KCl medium (120 mM KCl, 10 mM HEPES, 1 mM EGTA, pH 7.2) in the 3-ml stirred and thermostatted chamber of a Clark-type oxygen electrode (Rank Brothers, Bottisham, Cambridge, UK). Rat kidney, skeletal muscle, and brown adipose tissue mitochondria were prepared as described (12) in isolation medium (250 mM sucrose, 5 mM Tris-HCl, 2 mM EGTA, pH 7.4). For BAT mitochondria the medium was supplemented with 1% (w/v) defatted BSA, and the mitochondrial pellet was then washed twice in isolation medium without BSA. Mitochondrial pellets were suspended in isolation medium, and the protein concentration was determined by the biuret method using BSA as a standard (26).
Yeast cultures (Saccharomyces cerevisiae strain DBY746, MAT␣ leu2-3,112 his3⌬1 trp1-289 ura3- 52) were grown aerobically at 28°C to mid-log phase in lactate-containing medium (2% DL-lactic acid, 0.3% yeast extract, 0.05% glucose, 0.05% CaCl 2 ⅐2H 2 O, 0.05% NaCl, 0.06% MgCl 2 ⅐6H 2 O, 0.1% KH 2 PO 4 , 0.1% NH 4 Cl (all w/v), pH 5.5) (27) , and the resulting supernatant then spun for 10 min at 12,000 ϫ g, 4°C. The mitochondrial pellet was resuspended in 0.6 M sorbitol buffer, homogenized as before, and centrifuged (5 min at 1,500 ϫ g). The supernatant obtained was centrifuged for 10 min at 12,000 ϫ g, 4°C. The final mitochondrial pellet was resuspended in 0.6 M sorbitol buffer, and the protein concentration was measured by the bicinchoninic acid assay using BSA as a standard (28). Aliquots of the mitochondrial preparation were mixed with 10 mg/ml BSA as a cryoprotectant, snapfrozen on dry ice, stored at Ϫ80°C, and thawed prior to use. Under these conditions the mitochondria retained a membrane potential that was indistinguishable from that of freshly isolated yeast mitochondria as confirmed by the uncoupler-sensitive uptake of [ 3 H]TPMP (data not shown).
Assays-The TBARS assay was used to quantitate lipid peroxidation (14). Rat liver mitochondria (4 mg of protein) were suspended in 2 ml of buffer (100 mM KCl, 10 mM Tris-HCl, pH 7.6, 10 mM succinate, 8 g of rotenone/ml) supplemented with ethanol carrier or test compound. After 5 min of preincubation, oxidative stress was induced by addition of 100 M FeSO 4 and 300 M ascorbic acid, and 40 min later the incubation was divided into two 800-l aliquots and 400 l of thiobarbituric acid (TBA; 0.05% w/v in 10 ml of H 2 O, 10 ml of perchloric acid) was added to each aliquot. Samples were heated at 100°C for 15 min, cooled on ice, and then transferred to a glass tube, and 3 ml of water and then 3 ml of butanol were added. After vortexing, the organic layer was isolated by centrifugation, and 200-l aliquots were analyzed in a fluorometric plate reader ( Ex 515 nm, Em 553 nm) and compared with a standard curve of 0 -30 nmol of 1,1,3,3-tetraethoxypropane.
To assess reactivity of different molecules with the hydroxyl radical, we used ferrous iron to generate the hydroxyl radical and then measured the hydroxylation of benzoic acid (29). This was done in 30 mM NaP i , pH 7.4, 40 mM NaCl containing 690 M sodium benzoate, 30 M EDTA, and 200 M FeCl 2 . Compounds were incubated at 37°C for 60 min, then cooled on ice, and diluted in 3 ml of buffer, and the fluorescence was measured in a stirred 3-ml system ( Ex 305 nm, Em 407 nm).
To study the interaction of different compounds with superoxide, xanthine oxidase (XO; 0.01 units/ml) in 50 mM KP i , pH 7.5, 1 mM EDTA, 100 M DTPA supplemented with 500 M hypoxanthine was used to generate superoxide. Superoxide production was measured as the rate of reduction of 50 M acetylated cytochrome c (Sigma) at 550 nm in a 1-ml cuvette at 30°C.
To measure the oxidation of ferrous iron, 110 M FeCl 2 was incubated in Chelex-treated 50 mM NaCl, 5 mM Tris-HCl, pH 7, under argon. At various times 100-l samples were removed, added to 900 l of 1 mM FerroZine (30) in the same buffer, and the FerroZine-Fe(II) complex was assayed spectrophotometrically (⑀ 564 ϭ 27.9 ϫ 10 3 M Ϫ1 ⅐cm Ϫ1 ) (30). Fe(II) oxidation over 30 min and the absorbance of the tested compounds at 564 nm were both negligible. To confirm that oxidation of Fe(II) produced Fe(III) we used desferrioxamine, which chelates Fe(III) (desferrioxamine-Fe(III), ⑀ 428 ϭ 2.8 ϫ 10 3 M Ϫ1 ⅐cm Ϫ1 ). This was done under argon as desferrioxamine in the presence of oxygen rapidly oxidizes Fe(II) to Fe(III). For this 1 mM desferrioxamine was stirred in an air-tight 3-ml cuvette under argon with 100 M FeCl 2 and after an injection of TEMPO there was rapid (Ͻ1 s) oxidation of Fe(II) to Fe(III).
Aconitase activity was measured by a coupled enzyme assay linking isocitrate production by aconitase to NADP reduction by isocitrate dehydrogenase (⑀ 340 NADPH ϭ 6.22 ϫ 10 3 M Ϫ1 ⅐cm Ϫ1 ) (31,32). The background rate of NADPH formation was determined in the presence of fluorocitrate (100 M), a competitive inhibitor of aconitase, and was always less than 10% of the initial rate. Aliquots of frozen yeast mitochondria were thawed rapidly, washed in mannitol buffer (0.6 M mannitol, 10 mM Tris maleate, pH 6.8, 5 mM KP i , 0.5 mM EDTA), and resuspended in this buffer at 0.2-0.3 mg of protein/ml. MitoPBN or TPMP (from stocks in dimethyl sulfoxide) and substrate were added, and the mitochondria were incubated in a shaking water bath at 30°C. Samples were removed at various time points, snap-frozen on dry ice, and thawed prior to assaying. The aconitase assay was adapted for a 96-well plate format with a 10-l sample added to 190 l of assay buffer (50 mM Tris-HCl, pH 7.4, 0.6 mM MnCl 2 , 5 mM sodium citrate, 0.2 mM NADP ϩ , 0.1% v/v Triton X-100, and 0.4 units/ml isocitrate dehydrogenase) at 30°C and assayed with A 340 readings at 15-s intervals over 7 min. The resulting slopes of multiple samples (typically 6) were averaged.
The oxidation of cis-parinaric acid (cPA) was monitored fluorometrically ( Ex ϭ 324 nm; Em ϭ 413 nm) in a 3-ml fluorometer cuvette at SCHEME 1. Synthesis of MitoPBN (2). Ph, phenyl.   (33). Beef heart mitochondrial membranes, prepared as described (34), were incubated in 50 mM KP i buffer, pH 8.0, and after 40 s, cis-parinaric acid (0.5 M) was added, and its oxidation was monitored. The absorption spectrum of MitoPBN overlaps with the excitation spectrum of cPA; therefore, for comparisons all experiments were adjusted to the same maximum 100% fluorescence immediately following addition of cPA. Proton Leak Measurements-Mitochondria (0.35 mg of protein/ml) from kidney or skeletal muscle were incubated in 120 mM KCl, 5 mM KP i , 3 mM HEPES, and 1 mM EGTA, pH 7.2, at 37°C, with 5 M rotenone, 80 ng of nigericin/ml, and 1 g of oligomycin/ml. BAT mitochondria were incubated in 50 mM KCl, 1 mM EGTA, 4 mM KP i , 5 mM HEPES, pH 7.2, and 1% w/v defatted BSA at 37°C with 5 M rotenone, 80 ng of nigericin/ml, and 1 g of oligomycin/ml. Respiration rate and membrane potential were measured simultaneously using electrodes sensitive to oxygen and TPMP (4). The TPMP electrode was calibrated with five sequential 0.5 M additions of TPMP and then substrate was added, 4 mM succinate for kidney or skeletal muscle mitochondria or 10 mM ␣-glycerophosphate for BAT mitochondria. Membrane potential was varied by adding malonate (up to 1 mM) for kidney and skeletal muscle or KCN (up to ϳ100 M) for BAT mitochondria. After each run, 0.2 M FCCP was added to release TPMP for base-line correction. When MitoPBN was used, the TPMP electrode was calibrated with five sequential additions of 9:1 TPMP:MitoPBN to final concentrations of 2.25 and 0.25 M respectively. For simplicity, the TPMP binding correction was assumed to be 0.4/(l per mg protein) (24) for mitochondria from all tissues; this will have caused small systematic errors in membrane potential in muscle mitochondria or when Mi- Proton leak kinetics for BAT mitochondria were measured as described in A and B, except that the membrane potential was generated using 10 mM ␣-glycerophosphate and was gradually decreased by sequential addition of KCN (up to ϳ100 M). Data are means Ϯ range of two independent experiments each performed in duplicate. E and F, MitoPBN prevents activation of UCP3 in rat skeletal muscle mitochondria by exogenous superoxide. Proton leak kinetics were measured as in A and B. Data are means Ϯ S.E. of three independent experiments each performed in duplicate. G and H, no effect of MitoPBN on HNE activation of UCP2. Proton leak kinetics were measured in rat kidney mitochondria as in A and B in the presence of 0.3% (w/v) defatted BSA, with or without 35 M HNE and 250 nM MitoPBN as indicated. Data are means Ϯ range of two independent experiments each performed in duplicate. The stimulation of leak by HNE is GDP-sensitive (5). *, respiration rate at the highest common potential in the relevant panel is significantly different (p Ͻ 0.05, Student's paired t test) from control (left-hand panels) or MitoPBN (right-hand panels).  's t test. F, oxidative damage to the respiratory chain. Rat liver mitochondria (1 mg of protein/ml) were incubated with 20 mM paraquat or 5 mM tert-butylhydroperoxide (tBHP) for 10 min at 30°C in sucrose-based isolation medium toPBN was present. Exogenous superoxide was generated using xanthine (50 M) and xanthine oxidase (0.01 units/3.5 ml assay) (4). Xanthine and xanthine oxidase were added before the TPMP (or TPMP:MitoPBN) calibration and incubated with mitochondria for 10 -15 min before addition of substrate. To assess the statistical significance of the shifts in leak curves caused by superoxide, we generally compared respiration rates at the highest common membrane potential for pairs of curves from 3 independent experiments using Student's t test for paired data.

RESULTS AND DISCUSSION
Synthesis of MitoPBN-Acyclic nitrones such as PBN are usually formed by reductive condensation of a benzaldehyde with a nitroalkane (16,20). Nitrone formation is a slow equilibrium process that can be shifted in favor of the nitrone by dehydrating agents. The phosphonium unit could be attached to the basic PBN core at a number of positions via carbon or heteroatom, and the length of the alkyl chain between PBN and the phosphonium unit is variable. Anticipating adverse sensitivity of the nitrone to chemical manipulation, the phosphonium functionality was introduced first using a phenoxide/ alkylation methodology (35) to link (4-iodobutyl)triphenylphosphonium iodide to para-hydroxybenzaldehyde before incorporating the nitrogen function in the final step (Scheme 1). Reaction of (4-iodobutyl)triphenylphosphonium iodide with the anion derived from para-hydroxybenzaldehyde gave Mito-Benzaldehyde (1) in 84% yield (Scheme 1). Reductive condensation of 1 with 2-methyl-2-nitropropane under conditions optimized using para-methoxybenzaldehyde and based on reported methodology (20) gave low yields of MitoPBN (2). The reaction was modified and carried out in dry absolute alcohol in the presence of molecular sieves (16) and over 9 days gave pure 2 in 48% yield, as indicated by 1 H and 31 P NMR analysis.
Uptake of MitoPBN by Mitochondria-To determine whether MitoPBN was accumulated by energized mitochondria, an ionselective electrode for MitoPBN was used. In Fig. 1 the electrode response below 5 M was calibrated by sequential MitoPBN additions, and then a membrane potential was generated by addition of succinate, leading to accumulation of MitoPBN by mitochondria. Dissipation of the membrane potential with the uncoupler FCCP led to the immediate release of MitoPBN from the matrix (Fig. 1). The external concentration of MitoPBN following uptake was 0.91 Ϯ 0.08 M indicating that mitochondria had accumulated 2.0 Ϯ 0.2 nmol of MitoPBN/mg of protein (n ϭ 3). The mitochondrial volume under these conditions (0.5-0.9 l/mg) (36 -39) gives an intramitochondrial MitoPBN concentration of 2.2-4 mM. The Nernst equation implies that this accumulation ratio of 2,400 -4,400-fold corresponds to a membrane potential greater than the expected value of ϳ180 mV (24). This overestimation is due to MitoPBN binding reversibly to the matrix surface of the inner membrane (24,39,40); correcting for the expected potential suggests that 65-75% of accumulated MitoPBN is bound, compared with ϳ60% for the TPMP (36) and is consistent with their relative hydrophobicities (Table I).
To see if MitoPBN disrupted mitochondrial function, we compared its effects on respiration of rat liver mitochondria with those of TPMP, PBN, and with Mitobenzaldehyde (1), a possible breakdown product of MitoPBN (41). None of these compounds affected resting, phosphorylating, or uncoupling respiration at 10 M or lower, but there were minor inhibitory and uncoupling effects of the triphenylphosphonium containing compounds at 25 M and above (data not shown). Therefore, MitoPBN concentrations of 25 nM to 10 M were used for most of the experiments reported here, and controls with TPMP were done to check that there were no nonspecific effects on mitochondria.
MitoPBN Blocks Activation of UCP1, UCP2, and UCP3 by Superoxide but Not by HNE- Fig. 2A shows that kidney mitochondria exposed to exogenous superoxide from xanthine and xanthine oxidase showed an increase in proton conductance that was fully prevented by the specific UCP inhibitor, GDP, as reported previously (4). The superoxide activation of UCP2 was completely abolished by 250 nM MitoPBN (Fig. 2B). However, the same concentration of PBN did not affect the superoxideinduced proton conductance ( Fig. 2A). 25 nM MitoPBN also blocked the superoxide-stimulated leak, and even 2.5 nM Mi-toPBN attenuated the effect, whereas PBN concentrations of at least 10 M were required to block superoxide activation (data not shown). This more than 400-fold increased potency of Mi-toPBN over PBN can be explained by the accumulation of MitoPBN within the mitochondrial matrix, as demonstrated in Fig. 1. N-tert-Butylhydroxylamine, a hydrolysis product of PBN that accumulates in PBN stock solutions, accounts for some of the protective effects of PBN in cell culture (41); however, 100 M N-tert-butylhydroxylamine did not affect superoxide-induced proton leak (data not shown). Therefore, this breakdown product does not contribute to the effects of MitoPBN on superoxide-induced leak.
Kidney mitochondria contain only UCP2 (4,42), so the experiments in Fig. 2, A and B the superoxide stimulation of proton leak by other UCPs, we investigated mitochondria from BAT (Fig. 2, C and D) and skeletal muscle (Fig. 2, E and F). Superoxide activation of proton conductance in BAT mitochondria operates primarily through UCP1 (4). As reported previously (4), the basal activity of UCP1 in BAT mitochondria was inhibited by GDP because UCP1, unlike UCP2 and UCP3, is activated in the absence of superoxide by endogenous factors including fatty acids. As before (4), this basal proton conductance of UCP1 was further stimulated by superoxide in a GDP-sensitive manner (Fig. 2C). MitoPBN prevented this activation (Fig. 2D), so it prevents superoxide activation of UCP1. Superoxide activation of proton conductance in skeletal muscle mitochondria operates through UCP3, because the effect is absent in mitochondria from UCP3 Ϫ/Ϫ mice (4). Superoxide activated proton conductance in skeletal muscle mitochondria (Fig. 2E) as reported previously (4). MitoPBN prevented this activation (Fig. 2F), so it also blocks superoxide activation of UCP3. Thus, the abolition of superoxide activation of UCPs by MitoPBN is general to all three mammalian UCPs.
Lipid peroxidation products such as HNE activate UCPs by a GDP-sensitive mechanism (5). In contrast to its inhibitory effect on superoxide-activated proton conductance, MitoPBN had no effect on the activation of UCP2 by HNE in kidney mitochondria (Fig. 2, G and H).
The mitochondria-targeted spin trap MitoPBN blocks the superoxide-induced increase in UCP proton conductance with more than 400-fold greater potency than the untargeted PBN because it is accumulated within the mitochondria. This finding indicates that MitoPBN blocks the UCP activation pathway within the matrix. This conclusion is consistent with our earlier findings that the mitochondria-targeted antioxidants Mi-toQ and MitoVit E act within mitochondria to prevent superoxide activation of UCPs (12). In contrast, MitoPBN did not affect the stimulation of proton conductance by HNE. This difference suggests two scenarios: either superoxide activates UCPs by generating lipid peroxidation products such as HNE, and MitoPBN prevents superoxide activation by blocking the production (but not the activity) of such lipid peroxidation products; or the stimulation of UCP proton conductance by superoxide and HNE occur by distinct processes. To distinguish between these possibilities, we next investigated how MitoPBN interacts with the ROS generated within mitochondria exposed to superoxide.
Interactions of MitoPBN with Reactive Oxygen Species-When superoxide was generated from xanthine and xanthine oxidase in the absence of mitochondria, MitoPBN concentrations up to 500 M did not decrease the superoxide dismutasesensitive reduction of cytochrome c (data not shown); therefore, MitoPBN does not react significantly with superoxide. To see if MitoPBN affected the reactivity of superoxide within the mitochondrial matrix, we measured the rate of inactivation of the matrix enzyme aconitase, which is particularly sensitive to damage by superoxide (32). Neither the spontaneous inactivation of aconitase by endogenous superoxide nor the high rate of inactivation induced by the redox cycler paraquat were prevented in yeast mitochondria by MitoPBN (Fig. 3A).
During the proton leak measurements shown in Fig. 2, deenergized mitochondria were exposed to exogenous superoxide which moves into the matrix, presumably by passage of its conjugate acid (pK 4.8) (43) through the phospholipid bilayer , and cPA was added after 1.5 min, and its fluorescence decay was measured. A typical experiment is shown; two independent repeats gave similar results, as did replacing desferrioxamine with an alternative iron chelator DTPA (2 mM). (12). MitoPBN did not affect the inactivation of aconitase in de-energized rat liver mitochondria exposed to exogenous superoxide (Fig. 3B); therefore, MitoPBN does not inhibit superoxide movement through the mitochondrial membrane. Thus the effects of MitoPBN on the superoxide-induced leak are not due to it reacting with superoxide, preventing its uptake into mitochondria or affecting its reactivity within the matrix.
Superoxide dismutates to hydrogen peroxide, which generates the very reactive hydroxyl radical in the presence of ferrous ions. To assess the reactivity of MitoPBN with the hydroxyl radical, we measured its ability in vitro to prevent hydroxylation of benzoic acid by hydroxyl radicals generated by the Fenton reaction (29) (Fig. 3C). MitoPBN, PBN, and TPMP trapped the hydroxyl radical with IC 50 values of ϳ77, ϳ143, and ϳ419 M, respectively, whereas an equimolar mixture of PBN and TPMP gave an IC 50 of ϳ100 M (Fig. 3C). Therefore, the reactivity of MitoPBN with the hydroxyl radical is marginally greater than that of PBN, probably due to the bulky triphenylphosphonium group. The rate constant for the reaction of PBN with the hydroxyl radical is 6.1-8.5 ϫ 10 9 M Ϫ1 ⅐s Ϫ1 (44,45), suggesting that MitoPBN reacts with the hydroxyl radical at close to the diffusion limit, in common with most organic compounds (46). The IC 50 for TPMP was only ϳ5.4-fold greater than that of MitoPBN, but TPMP concentrations 200-fold greater than those of MitoPBN had no effect on the activation of UCPs by superoxide. Therefore, although MitoPBN does react rapidly with the hydroxyl radical, this is not how it blocks the activation of proton leak by superoxide.
Mitochondria exposed to superoxide release ferrous iron from aconitase and other FeS proteins. This ferrous iron can then catalyze the initiation of lipid peroxidation (47,48). Therefore, the possibility that MitoPBN could prevent UCP activation by intercepting ferrous iron was addressed. There was no reaction between MitoPBN and ferrous iron in vitro under anaerobic conditions (Fig. 3D). Free radicals react with nitrones such as MitoPBN to generate short lived nitroxides, and the stable nitroxide TEMPO rapidly oxidized ferrous to ferric iron (Fig. 3D). However, when ferrous iron was incubated aerobically to generate ROS and transient nitroxides, MitoPBN did not stimulate its oxidation, even at 500 M (Fig. 3E). Only when PBN was added at very high concentrations (100 mM) did it affect iron oxidation. Hence MitoPBN does not prevent UCP activation by interacting with the ferrous iron released on exposure of mitochondria to superoxide.
To see if MitoPBN protected the respiratory chain from oxidative damage, we exposed rat liver mitochondria to superoxide by using the redox cycler Paraquat, or we oxidized the glutathione pool with the glutathione peroxidase substrate tertbutylhydroperoxide. Both treatments substantially decreased the rate of uncoupled respiration due to generalized oxidative damage to the respiratory chain, but MitoPBN gave no protection against either form of oxidative damage (Fig. 3F). In summary, MitoPBN does not prevent superoxide from activating UCPs by reacting with superoxide, ferrous iron, or the hydroxyl radical, or through general antioxidant protection.
Trapping of Carbon-centered Radicals by MitoPBN-PBN reacts rapidly with carbon-centered radicals, so we next determined whether MitoPBN also underwent this reaction. UV photolysis of H 2 O 2 in ethanol generated hydroxyl radicals that reacted rapidly with ethanol to yield the carbon-centered ␣-hydroxyethyl radical (17). PBN gave the well characterized ␣-hydroxyethyl radical adduct shown in Fig. 4 (A N ϭ 15 (Fig. 4), and when a mixture of MitoPBN and PBN was exposed to ␣-hydroxyethyl radicals, the spectra were overlapping and additive (data not shown). However, whereas the ␣-hydroxyethyl radical adduct of PBN was long lived with negligible loss over 80 min, that of MitoPBN decayed more rapidly, with no signal detectable from 1 mM MitoPBN ϳ40 min after UV irradiation (data not shown). The faster decay was not due to an intermolecular interaction between the cation and the ␣-hydroxyethyl radical adduct of PBN, as UV irradiation of an equimolar mixture of TPMP and PBN generated a long lived radical adduct of PBN (data not shown). However, radical adducts of para-methoxy-PBN decay more rapidly than those of PBN due to the electron-donating methoxyl group (50). Thus the faster decay of MitoPBN radical adducts is probably a consequence of the electron-donating ether linkage between the PBN moiety and the triphenylphosphonium. No oxygen-centered radical adducts were detected when 500 M MitoPBN was exposed to hydroxyl radicals generated by the Fenton reaction or to superoxide generated by xanthine oxidase/hypoxanthine (data not shown), consistent with the short lifetime of the adducts formed between PBN and oxygen-centered radicals and the low rate of reaction between PBN and superoxide (16). Therefore, the rapid reaction of Mi-toPBN with carbon-centered radicals may be how MitoPBN blocks the activation of UCPs by superoxide. As carbon-centered radicals occur during the initiation of lipid peroxidation, we next investigated the effects of MitoPBN on lipid peroxidation.
Effects of MitoPBN on Lipid Peroxidation-Lipid peroxidation is initiated by abstraction of a hydrogen atom to generate a carbon-centered radical on phospholipid fatty acyl chains. These carbon-centered radicals react with oxygen to form lipid hydroperoxides, which drive a self-propagating chain reaction. Once initiated, the lipid peroxidation chain reaction can be prevented by chain-breaking antioxidants such as vitamin E or ubiquinol. PBN, however, is a poor chain-breaking antioxidant (18). MitoPBN was also ineffective at preventing lipid peroxidation in liver mitochondria exposed to ferrous iron/ascorbate (Fig. 5A), in contrast to MitoVit E and MitoQ (14,15). This observation suggests that MitoPBN cannot block lipid peroxidation once it is initiated by excess pro-oxidant. Although PBN is ineffective as a chain-breaking antioxidant, it can prevent initiation of lipid peroxidation by reaction with carbon-centered radicals (18). As MitoPBN reacts rapidly with carbon-centered radicals (Fig. 4), we next set out to test whether it too could block the initiation of lipid peroxidation by reacting with carbon-centered radicals.
We initiated lipid peroxidation by using the water-soluble carbon-radical generator 2,2Ј-azobis(2-methylpropionamidine) dihydrochloride (AAPH), which decomposes spontaneously at 37°C to generate two carbon-centered radicals (Scheme 2). Lipid peroxidation in beef heart mitochondrial membranes was measured as the loss of fluorescence of cPA, a polyunsaturated fatty acid that partitions into phospholipid bilayers and loses its fluorophore on oxidation, so it can be used as a convenient marker of phospholipid peroxidation in membranes (33). AAPH led to a loss of cPA fluorescence that was prevented by Mi-toPBN (Fig. 5B), consistent with MitoPBN reacting with carbon-centered radicals and blocking initiation of lipid peroxida- SCHEME 2 tion. Exposing mitochondrial membranes to superoxide also decreased cPA fluorescence, and this reaction too was blocked by MitoPBN (Fig. 5C). Superoxide-induced lipid peroxidation in mitochondrial membranes was prevented by the iron chelator desferrioxamine (Fig. 5D), suggesting that superoxide alone was unable to initiate lipid peroxidation. In these experiments the iron probably became associated with the membranes during preparation and storage, as well as being released from FeS centers within respiratory complexes on exposure to superoxide. The superoxide anion is insufficiently reactive to initiate lipid peroxidation by abstraction of a hydrogen atom from a fatty acid (51). Its conjugate acid, the hydroperoxyl radical, can initiate lipid peroxidation; however, the rate of hydrogen atom abstraction from isolated fatty acids by this radical (k ϳ0.3-1.2 ϫ 10 3 M Ϫ1 ⅐s Ϫ1 ) (51) is far lower than that of the hydroxyl radical (k ϳ10 9 M Ϫ1 ⅐s Ϫ1 ), and the pK of superoxide (4.8) (43) means that the amount of the hydroperoxyl radical present is small. Consequently the initiation of lipid peroxidation by superoxide is largely iron-dependent in these systems.
These data suggest that MitoPBN reacts with carbon-centered radicals and stops the initiation of lipid peroxidation. This conclusion is consistent with the known rapid reaction between PBN and carbon-centered radicals (k ϳ10 6 -10 7 M Ϫ1 ⅐s Ϫ1 (17,45)). In contrast, MitoPBN will not stop lipid peroxidation once initiated by other sources. This is because the rate of reaction of PBN with the peroxyl radicals essential for propagating lipid peroxidation is low (k ϳ40 -200 M Ϫ1 ⅐s Ϫ1 ) (52), and the radical adducts thus formed are very unstable (18). . ) generated outside mitochondria by XO, or within mitochondria by the respiratory chain, inactivates iron-sulfur center-containing enzymes such as aconitase, expelling ferrous iron. The superoxide is also dismutated (mostly by the matrix Mn-superoxide dismutase) to hydrogen peroxide (H 2 O 2 ), which reacts with ferrous iron by the Fenton reaction to generate the hydroxyl radical ( ⅐ OH). The hydroxyl radical extracts a hydrogen atom (H ⅐ ) from a bis-allylic position on an unsaturated fatty acyl chain of a phospholipid, leaving a carbon-centered fatty acyl radical that can be quenched by spin traps such as MitoPBN. The carbon-centered radical reacts with oxygen to initiate formation of a peroxyl radical, which then propagates a chain reaction of lipid peroxidation leading to a complex mixture of lipid peroxidation breakdown products. Two of these, 4-hydroxynonenal (HNE) and 4-hydroxyhexenal, are shown. HNE and other lipid peroxidation breakdown products activate UCPs, thereby increasing the proton conductance of the mitochondrial inner membrane. This activation of UCPs decreases the protonmotive force, and we have proposed elsewhere (7) that is part of a regulatory mechanism to decrease superoxide production by the respiratory chain when ROS production is too high.
UCP Activation by Superoxide Exposure Leading to Lipid Peroxidation Products-MitoPBN blocks superoxide activation of UCPs, but this is not due to its reaction with superoxide, iron, or the hydroxyl radical. As MitoPBN reacts rapidly with carbon-centered radicals, one possibility is that superoxide acts within mitochondria to generate carbon-centered radicals on phospholipid acyl chains. How this might occur is outlined in Fig. 6. Mitochondria exposed to endogenous or exogenous superoxide undergo inactivation of iron-sulfur center proteins, such as aconitase or respiratory chain complexes, expelling ferrous iron (48). This ferrous iron reacts with hydrogen peroxide, produced by dismutation of superoxide catalyzed mostly by mitochondrial Mn-superoxide dismutase, to generate hydroxyl radicals by Fenton chemistry. These hydroxyl radicals attack the fatty acyl chains of mitochondrial phospholipids to initiate formation of carbon-centered radicals. The carbon-centered radicals then react with oxygen to form peroxyl radicals, which in turn propagate a cascade of lipid peroxidation. Depending on the fatty acyl chain that is attacked, and on the particular breakdown pathway that ensues, lipid peroxidation leads to the formation of large amounts of reactive small lipid fragments such as HNE (from n-6 fatty acyl groups such as 20:4(n-6), arachidonyl), hydroxyhexenal (from n-3 fatty acyl groups such as 22:6(n-3), docosahexaenoyl), and malondialdehyde (53), most or all of which can activate the proton conductance of UCPs.
This scenario explains how MitoPBN can act at the beginning of this cascade to intercept carbon-centered radicals and prevent the initiation of lipid peroxidation. It also explains why MitoPBN did not prevent the activation of UCPs by HNE, which occurs downstream. Most importantly, this model indicates how the apparently unrelated activation of UCPs by superoxide and HNE can occur by a common pathway. MitoQ and MitoVit E also block superoxide-induced leak, but as they can act as chain-breaking antioxidants, and also react with superoxide, their effect on superoxide activation of UCPs is far less informative than that of MitoPBN.
As well as being consistent with the data presented here, the model in Fig. 6 has testable predictions. One prediction is that the release of iron within mitochondria is necessary for superoxide to initiate lipid peroxidation and UCP activation; another is that the generation of carbon-centered radicals in the phospholipid bilayer would lead to UCP activation and that this should be prevented by MitoPBN.

Requirement for Mitochondrial Iron for Superoxide Activation of UCPs-
The initiation of lipid peroxidation when mitochondria are exposed to superoxide probably arises from the release of ferrous iron from FeS centers such as aconitase (48), as the direct initiation of lipid peroxidation by superoxide is slow, even when it is protonated (51). Therefore, we investigated whether addition of iron chelators affected the activation of UCPs by superoxide. Most of the iron chelators we investigated could not be used during measurements of proton conductance of isolated mitochondria due to membrane impermeance, excessive uncoupling, or limited solubility. Furthermore, many chelators have complicated interactions with ROS (54). However, we found that bipyridyl was usable up to 5 mM, although it did cause considerable nonspecific uncoupling, limiting the concentrations that could be tested. Bipyridyl attenuated the superoxide activation of proton conductance through UCP2 in kidney mitochondria (data not shown), consistent with superoxide acting to increase leak via changes in intramitochondrial iron.
Activation of UCPs by Carbon-centered Radicals-If superoxide activates UCPs by generating carbon-centered radicals within the phospholipid bilayer, then generating such radicals directly should activate UCPs. To test this we added the carbon-centered radical generator AAPH to kidney mitochondria. 2 mM AAPH strongly increased the proton conductance, and this activation was fully prevented by addition of GDP, indicating that AAPH activated UCP2 and did not uncouple by causing nonspecific damage to the mitochondria (Fig. 7A). Carboxyatractylate, a specific inhibitor of the adenine nucleotide translocase, also prevented activation by AAPH (Fig. 7A). These results faithfully echo the effects on UCPs and the adenine nucleotide translocase seen previously with HNE (5). Mi-toPBN completely blocked this activation of UCP2 by AAPH (Fig. 7B), suggesting that MitoPBN was able to prevent initiation of lipid peroxidation and hence the activation of UCP2 by AAPH. Together these data are consistent with the general model (Fig. 6) that superoxide activates UCPs through the generation of carbon-centered radicals within mitochondria and that MitoPBN blocks this by preventing the initiation of lipid peroxidation through reaction with carbon-centered radicals.
Conclusions-In the work reported here, MitoPBN was synthesized by generating a phenoxide from p-hydroxybenzaldehyde, which was then coupled to the butyltriphenylphosphonium FIG. 7. Activation of UCP proton conductance by carbon-centered radicals and prevention by MitoPBN. A, activation of proton conductance of UCP2 by carbon-centered radicals. Carbon-centered radicals were generated by incubating kidney mitochondria (0.35 mg of protein/ml) with AAPH (2 mM) for about 10 min. Proton leak kinetics were measured as described in Fig. 2. Membrane potential was generated using 4 mM succinate and gradually decreased by sequential additions of malonate up to 1 mM. GDP (0.5 mM) or carboxyatractylate (CAT; 2.5 M) were present where indicated. AAPH was added before TPMP. Data are means Ϯ S.E. of three independent experiments. B, prevention of carbon radical activation of leak by MitoPBN. Kidney mitochondria were incubated with 2.25 M TPMP and 250 nM MitoPBN in the presence or absence of AAPH, and the proton leak kinetics were measured as described in A. AAPH was added before TPMP and MitoPBN. Data are means Ϯ S.E. of three independent experiments. *, respiration rate at the highest common potential in the relevant panel is significantly different (p Ͻ 0.05, Student's paired t test) from control (A) or MitoPBN (B). moiety of (4-iodobutyl)triphenylphosphonium through an ether linkage by nucleophilic displacement of iodide (Scheme 1). The benzaldehyde (1) was then converted to a nitroxide to give Mi-toPBN (2). Introducing the active moiety after conjugation to the lipophilic cation contrasts with the conventional approach of creating the triphenylphosphonium cation in the last step in the synthesis by reaction of triphenylphosphine with a halogenated precursor (14,15,40). The success of this procedure expands the number of possible synthetic routes to mitochondria-targeted molecules.
MitoPBN has proven to be a powerful tool to probe pathways of ROS-induced changes in isolated mitochondria. It blocks the activation of proton conductance by superoxide from the matrix side of the mitochondrial inner membrane, but it does not affect the stimulation of conductance by the lipid peroxidation product HNE. The prevention of UCP activation by superoxide is not due to the reaction of MitoPBN with superoxide, iron, or the hydroxyl radical, or by MitoPBN acting as a chain-breaking antioxidant. However, MitoPBN does react strongly with carbon-centered radicals. This reactivity prevents the initiation of lipid peroxidation by superoxide, and the activation of UCPs by a carbon-centered radical generator. Therefore, superoxide probably stimulates UCPs from within mitochondria by attacking and inactivating iron-sulfur center enzymes, leading to the release of ferrous iron. In the presence of hydrogen peroxide, this leads to the formation of hydroxyl radicals, which generate carbon-centered radicals on phospholipids. These carbon-centered radicals undergo further reactions and degrade to form lipid peroxidation breakdown products such as HNE, which go on to activate UCPs.
This model shows how the activation of UCPs by superoxide and through lipid peroxidation breakdown products such as HNE lie on the same pathway (Fig. 6). However, the way in which such products activate UCPs is still unclear. The physiological role of activation of mild uncoupling by UCP2 and UCP3 through the superoxide-initiated lipid peroxidation pathway may be to lower the protonmotive force and decrease endogenous superoxide production in the mitochondrial matrix. This mechanism provides a simple negative feedback loop, with HNE and other lipid peroxidation products as the mediators, to ensure that superoxide production in the mitochondrial matrix is minimized, at the expense of the efficiency of energy conservation (7).
PBN has a number of protective pharmacological properties in vivo, although the mechanism of these effects is often unclear, as they are not always associated with general antioxidant efficacy (reviewed in Refs. 55 and 56). Our finding of a dramatic biological effect of a PBN derivative without significant general antioxidant efficacy suggests that some of the pharmacological effects of PBN may occur by preventing the initiation of lipid peroxidation by carbon-centered radicals and by interfering with signaling cascades that rely on such generation. The selectivity of MitoPBN for mitochondrial carboncentered radicals suggests that it may be a promising candidate to alter mitochondrial oxidative damage and UCP function in vivo.
In summary, we have synthesized a novel mitochondriatargeted spin trap, MitoPBN, and used it to show that superoxide activates UCPs through the formation of carbon-centered radicals and to infer that the activation of UCPs by superoxide and lipid peroxidation breakdown products such as HNE occurs on the same pathway. These findings further support the hypothesis that a major function of UCPs is to lower the mitochondrial membrane potential and thereby decrease ROS production in response to increased mitochondrial ROS or oxidative damage.