Redox-dependent Changes in Molecular Properties of Mitochondrial Apoptosis-inducing Factor*

Mitochondrial apoptosis-inducing factor (AIF) is a central player in the caspase-independent cell death pathway whose normal physiological function remains unclear. Our study showed that naturally folded mouse AIF very slowly reacts with NAD(P)H (kcat of 0.2-0.01 s-1) forming tight, dimeric, and air-stable FADH2-NAD(P) charge-transfer complexes ineffective in electron transfer. FAD reduction is accompanied by a conformational change involving AIF-specific N-terminal and regulatory 509–559 peptides and the active site His453, and it affects susceptibility of AIF to calpain and AIF-DNA interaction, the two events critical for initiating caspase-independent apoptosis. Based on our results, we propose that formation of long lived complexes with NAD(P)H and redox reorganization may be functionally important and enable AIF to act as a redox-signaling molecule linking NAD(P)H-dependent metabolic pathways to apoptosis.

Mitochondrial apoptosis-inducing factor (AIF) is a central player in the caspase-independent cell death pathway whose normal physiological function remains unclear. Our study showed that naturally folded mouse AIF very slowly reacts with NAD(P)H (k cat of 0.2-0.01 s ؊1 ) forming tight, dimeric, and airstable FADH 2 -NAD(P) charge-transfer complexes ineffective in electron transfer. FAD reduction is accompanied by a conformational change involving AIF-specific N-terminal and regulatory 509 -559 peptides and the active site His 453 , and it affects susceptibility of AIF to calpain and AIF-DNA interaction, the two events critical for initiating caspase-independent apoptosis.

Based on our results, we propose that formation of long lived complexes with NAD(P)H and redox reorganization may be functionally important and enable AIF to act as a redox-signaling molecule linking NAD(P)H-dependent metabolic pathways to apoptosis.
Apoptosis-inducing factor (AIF) 2 is a highly conserved, phylogenetically old mitochondrial flavoprotein implicated in embryonic development, cardiac cell survival, carcinogenesis, and neurodegenerative disorders. In the late 1990s, AIF was identified as a key player in the caspase-independent cell death pathway (1,2). These early studies suggested that mature AIF consists of residues 102-612 (⌬1-101), resides in the intermembrane space (IMS) of normal mitochondria and, upon an apoptotic stimulus, translocates to the nucleus and triggers chromatin condensation and large scale DNA fragmentation by cooperating with other proteins such as endonuclease G and cyclophilin A (CypA) (1,3,4). Recently, it has been proven that mature AIF starts at residue 54 and is anchored to the inner mitochondrial membrane (5), whereas the apoptogenic ⌬1-101/102 form is produced upon proteolytic processing involving calpain I and cathepsins B, L, and S (6,7).
Although the apoptotic function of AIF has been extensively studied (reviewed in Ref. 8), the role of the protein in normal mitochondria remains unclear. To date, functional and structural studies have been carried out mainly on the ⌬1-120 and shorter forms of recombinant AIF, which were purified from bacteria as apoproteins and refolded to incorporate FAD (1, 9 -11). Based on the redox properties and structural similarity to the disulfide reductase family of enzymes, AIF was proposed to function as a superoxide-generating NADH oxidase (9) or electron transferase (10). Other studies indicated that AIF might act in mitochondria as an antioxidant (12)(13)(14) or decrease reactive oxygen species levels indirectly by assisting biogenesis/maintenance of respiratory complexes I and III (15)(16)(17) and regulating glutathione levels (18). Finally, a recent observation that mitochondria in AIF-deficient neurons were fragmented, with aberrant cristae, led to a suggestion that AIF participates in maintaining mitochondrial structure (19). Despite the fact that there is a strong indication that AIF is linked to free radical homeostasis and oxidative stress, the mechanism and precise role of AIF in mitochondria remain elusive.
One problem in elucidating the physiological function of AIF is that only truncated and refolded forms of the recombinant protein have been investigated (9 -11, 20, 21). In this study, molecular and redox properties of naturally folded mature and apoptogenic AIF have been characterized and compared. Both flavoproteins were found to slowly react with NAD(P)H forming dimeric, air-stable FADH 2 -NAD(P) charge-transfer complexes (CTCs) with notable differences in the kinetics of NAD(P)H dissociation and CTC decay imposed by dissimilarities in the length of the N terminus. The redox-linked monomer-dimer transition, similarly occurring in the mitochondrial protein, was accompanied by a conformational change that involved a unique 509 -559 insertion, a potential protein-protein interaction site, and affected susceptibility of AIF to proteases and AIF-DNA interaction, critical for initiation of the caspase-independent apoptotic pathway. Our results challenge previous concepts and suggest that AIF is a redox-signaling molecule whose normal and apoptogenic functions are controlled by NAD(P)H. rials. Due to problems with expression of AIF⌬1-53 H453L, the mutation was introduced into a shorter ⌬1-77 construct, and the resulting protein was expressed in the JM109 strain of E. coli. Wild-type (WT) AIF⌬1-77 was used for comparative analysis.
Spectral Measurements-Kinetics of AIF reduction with NAD(P)H was monitored under anaerobic conditions at 452 nm and 25°C using a SX.18MV stopped-flow spectrophotometer (Applied Photophysics, UK). Anaerobic samples were prepared as described previously (22). Redox potentials of AIF were determined spectrophotometrically in xanthine saturated 100 mM phosphate, pH 7.0 -9.0, containing 2 M benzyl viologen (Ϫ359 mV) and catalytic amounts of xanthine oxidase (23). The dissociation rate constant for the AIF⅐NADH complex and redox activities of AIF were measured as described in the supplemental materials. To compare stability of the NAD(P)Hbound forms, AIF was sealed in an anaerobic cuvette, reduced with subequimolar NAD(P)H, and purged with pure oxygen to initiate flavin oxidation that was monitored at 452 nm and 25°C. CD spectra of 2.5-5 M AIF in the absence and presence of a 20-fold excess of NAD(P)(H) were recorded in 12 mM phosphate, pH 7.0, using a 0.1-cm path length cuvette and a Jasco J-720 spectropolarimeter.
Determination of the Molecular Size of AIF-The molecular size of recombinant AIF was determined by gel filtration on Superdex-200 (Amersham Biosciences) in 50 mM phosphate, pH 7.0. 5-10 nmol of AIF in the presence or absence of a 20-fold excess of NAD(P)(H) was used per injection. The molecular size of mitochondrial AIF was estimated using a combination of two-dimensional blue native (BN) and SDS-PAGE (24) and immunoblotting. Mice liver mitochondria (10 mg/ml) untreated or treated with 5 mM NAD(H) for 15 min at 25°C, were incubated with digitonin (3 mg/mg of protein) for 30 min at 4°C and centrifuged at 53,000 rpm for 10 min. The supernatant fraction was subjected to 3-13% BN-PAGE followed by two-dimensional 5-15% SDS-PAGE. Separated proteins were transferred to the polyvinylidene difluoride membrane and probed with an anti-AIF antibody raised against the 517-531 peptide (1:1000, Chemicon) and a secondary antibody conjugated with alkaline phosphatase (1:3000, Bio-Rad).
Proteolytic Digests-10 M AIF in 20 mM Tris-HCl, pH 7.5, 2 mM CaCl 2 was incubated with trypsin (0.06 mg/ml, Sigma) or human calpain I (5 units/ml, Calbiochem) at 37°C in the absence and presence of 2 mM NAD(P)(H). Proteolytic sites were identified by the N-terminal amino acid sequencing of the digestion products.
AIF Release from Mitochondria-Mice liver mitochondria were isolated by standard differential centrifugation. Protein concentrations were measured by the Pierce BCA method. AIF release was induced as described previously (6,7). Briefly, 400 g of mitochondria was incubated for 15 min at room temperature in the respiration medium (125 mM KCl, 2 mM K 2 HPO 4 , 20 mM HEPES, pH 7.0, 4 mM MgCl 2 , 3 mM ATP, 1 mM ADP, 1 M ruthenium red) containing either 5 mM malate/glutamate or 5 mM succinate and 2 M rotenone with or without 0.5-5 mM NAD(P)(H) and 100 M calpeptin (Calbiochem), then placed for 60 min at 30°C untreated or treated with a mixture of calpain (5 units/ml), 200 M CaCl 2 , and either 200 nM calpain-activated Bid (cBid, Calbiochem) or 5 mM atractyloside (Atr, Calbiochem), then centrifuged for 30 min at 13,400 ϫ g to collect supernatants. Released proteins were separated on 4 -15% SDS gels, and AIF was detected by Western blotting as described above.
AIF-DNA Interaction-Interaction of AIF⌬1-101 with linearized double-stranded DNA was examined using gel-retardation assays (25). 1-100 g of AIF was incubated with a 100-bp DNA ladder (180 -250 ng, New England Biolabs or Invitrogen) for 15 min at room temperature in 20 mM Hepes, pH 7.0, in the absence and presence of 2 mM NAD(P)(H) and equimolar amounts of recombinant human CypA (Calbiochem) or heat shock protein 70 (Hsp70) and electrophoresed in 50 mM Tris acetate on 1.2-2% agarose gel containing 0.6 g/ml ethidium bromide. The ability of AIF⌬1-101 to degrade DNA was examined as described by Cande et al. (4). 20 nM AIF was incubated with 200 ng of a pNEB plasmid (2.7 kb, New England Biolabs) with or without 20 nM CypA for 1-4 h at 37°C. The resulting mixtures were resolved on 1.2% agarose gels and visualized with ethidium bromide.

RESULTS
The N-terminal Affinity Tag Affects AIF Folding-Folding of flavoenzymes is critical for their function, because it defines protein-flavin interactions and redox properties of the cofactor. Previously, AIF was cloned with a thioredoxin-or poly-histidine affinity tag attached to the N terminus, close to the FAD binding site, and was expressed and purified as an apoprotein that had to be refolded to incorporate FAD (1, 9 -11, 20). From our studies on putidaredoxin reductase (22), a structural homologue of AIF, we learned that the affinity tag must be placed at the C terminus and removed at the end of purification to allow in vivo FAD incorporation and to preserve redox properties of the flavoprotein. When cloned this way, both mature (⌬1-53) and apoptotic (⌬1-101) forms of mouse AIF were expressed in E. coli as soluble, FAD-bound proteins (supplemental Fig. 1S), redox properties of which differed significantly from those of refolded AIF⌬1-120 (9).
AIF Produces Air-stable FADH 2 -NAD(P) Charge-transfer Complexes upon Reduction with NAD(P)H-The first difference between the proteins was that the midpoint redox potential of our AIF was 50 -56 mV lower (Table 1) and poised 16 -22 mV below that of the NAD/NADH couple (E m,7 ϭ Ϫ320 mV). The manner of interaction with reduced pyridine nucleotides (PNs) was another striking dissimilarity. Refolded AIF⌬1-120 was reported to react with equimolar amounts of NAD(P)H without formation of an FADH 2 -NAD(P) CTC (9). In flavoproteins, CTCs are produced as a result ofcharge-transfer interactions between the parallel-stacked isoalloxazine ring of the reduced flavin (donor) and the nicotinamide ring of NAD(P) (acceptor) and absorb in the long wavelength region, where position of the absorption maximum reflects affinity of the complex: the greater the interaction between the reacting molecules, the longer the wavelength of the CTC absorption band (26). During reduction of refolded AIF, the 600 -800 nm band was observed only when NAD(P)H was present in excess, and its absorption maximum depended on the AIF:PN ratio (9). In contrast, high affinity CTC ( max ϭ 770 nm) was stoichio-metrically formed upon addition of NAD(P)H to our flavoprotein (Fig. 1, A and B). 1 mol of NADH per mol of AIF was needed to completely reduce the flavin, indicative of very tight cofactor binding. Reaction with NADPH proceeded considerably slower within several hours (versus tens of minutes for NADH), and a 2-fold excess of the reductant was required to reach equilibrium and to fully reduce FAD. Similar to the refolded protein, our AIF did not stabilize FAD semiquinone intermediates. No changes in the AIF absorbance spectrum were detected in the presence of NAD or NADP.
Kinetic measurements showed that AIF had a 100-fold higher preference for NADH over NADPH (Table 1) but reacted with either cofactor orders of magnitude slower than other PN-dependent flavoenzymes. Having determined the second order rate constant for the AIF⅐NADH complex formation (k on , 1.7 ϫ 10 2 M Ϫ1 s Ϫ1 ) and the rate constant for NADH dissociation (k off , 2.8 Ϯ 0.5 ϫ 10 Ϫ5 s Ϫ1 , supplemental Fig. 2S) we estimated that K d for NADH was very low, 150 Ϯ 10 nM.
Another remarkable feature of AIF was resistance of its NAD(P)Hbound forms to oxidation. Normally, it takes 2 days for air oxygen to fully oxidize FAD in CTC at ambient temperatures producing spectral changes opposite to those observed during the reduction step (not shown). To speed up the oxidation process, we measured kinetics of CTC decay under pure oxygen atmosphere. This reaction was multiphasic and accelerated with time ( Fig. 1, C and D) most likely due to formation of multiple redox intermediates. Importantly, complexes with NADPH had a shorter lifetime than those with NADH, wherein AIF⌬1-53 was oxidized 2-fold slower than the corresponding redox forms of AIF⌬1-101. This was the first indication that the extended N terminus assists CTC formation/stabilization. AIF Possesses Low Electron Transferring Ability-Kinetic parameters, such as low NADH binding rates, high K m and low K d for the cofactor, as well as the low redox potential of the flavin and high stability of the NADH-bound form, suggest that AIF cannot effectively transfer reducing equivalents from NADH to other molecules. Indeed, AIF catalyzed reduction of one-and two-electron acceptors with turnover numbers several orders of magnitude lower than those of traditional electron transferring flavoenzymes ( Table 2). Notably, the 2,6-dichlorophenolindophenol (DCIP)and cytochrome c reduction rates did not depend on whether catalase or superoxide dismutase was present in the reaction mixture, meaning that electrons to the acceptors came directly from AIF. The rate of nitro blue tetrazolium (NBT) reduction, on the other hand, was decreased by ϳ30% in the presence of superoxide dismutase and thus was partially mediated by superoxide anion, which is quite typical for oxidoreductases. Owing to high stability of CTC with NADH, the NADH oxidase activity of AIF⌬1-53 was virtually undetectable, and that of AIF⌬1-101 was very weak (Ͻ0.2 min Ϫ1 ). This is in contrast to refolded

FIGURE 1. Recombinant AIF forms dimeric, air-stable charge-transfer complexes upon NAD(P)H binding.
A and B, anaerobic titration of 20 M AIF⌬1-53 with NADH and NADPH, respectively. Equimolar amounts of NADH and ϳ2.0 mol of NADPH per mol of AIF were required to fully reduce FAD (insets A and B, respectively). K d for NADPH calculated from the hyperbolic fit to the plot in inset B was 4.2 Ϯ 0.8 M. C and D, kinetics of FAD oxidation in CTC formed between AIF⌬1-101 (traces a and b) or ⌬1-53 (c and d) and NADPH (a and c) or NADH (b and d). Solutions of NAD(P)H-bound AIF were purged with pure oxygen to speed up CTC decay that was monitored at 452 nm and 25°C. A dotted line in panel C indicates the absorbance of fully oxidized AIF. "Fraction remaining" was calculated as (⌬A max Ϫ ⌬A t )/⌬A max to determine t1 ⁄2 (shown by arrows).
a Values in parentheses are for WT AIF⌬1-77. b Dependence of k obs for the reaction of AIF⌬1-77 H453L reduction with NADH on AIF concentration is hyperbolic with k lim of 0.6 s Ϫ1 . AIF⌬1-120 that was shown to oxidize NAD(P)H considerably faster and catalyze NAD(P)H-dependent reduction of cytochrome c and NBT exclusively via superoxide anion radicals (9). Finally, neither form of our flavoprotein could transfer electrons from NADH to superoxide anion, hydrogen peroxide, ascorbate free radical, or dehydroascorbate, which suggests that AIF is not an antioxidant. AIF Undergoes Dimerization and a Conformational Change upon NAD(P)H Binding-Redox properties of AIF suggest that either the flavoprotein has evolved as a low turnover electron transferase or plays another role in mitochondria where formation of stable CTC is functionally important. To test the latter possibility, we analyzed how PN binding affects molecular properties of AIF. Gel-filtration experiments showed that oxidized AIF⌬1-101 and ⌬1-53 existed in solution as monomers ( Fig. 2A). NAD(P) had no effect on the elution profile (not shown), but reduction with NAD(P)H caused a 2-fold increase in the molecular size of both proteins (Fig. 2, B and C). Thus, AIF dimerization may be one of the factors stabilizing the FADH 2 ⅐NAD(P) complex. A small monomeric peak seen in the elution profiles of the NADPH-reduced proteins (Fig. 2C) reflects partial dissociation of the lower affinity ligand, wherein the larger amount of    FEBRUARY 29, 2008 • VOLUME 283 • NUMBER 9 monomers produced by the NADPH-bound apoptotic form (10 versus 5% in mature AIF) was another sign that the N-terminal peptide is recruited in CTC formation/stabilization. We analyzed also how PN cofactors affect the secondary structure of AIF. Far UV CD spectra of ligand-free AIF⌬1-53 and ⌬1-101 were highly similar but not identical (Fig. 2, D and  E), suggesting that the N terminus of mature AIF may adopt an ordered conformation in solution. Only subtle changes in the CD spectra of both proteins were detected in the presence of oxidized cofactors. In contrast, complex formation with NAD(P)H caused significant perturbations in the 190 -250 nm region. Computer analysis of the spectroscopic data identified a common trend in the NAD(P)H-induced structural changes in AIF, such as an increase in the ␤-strand and turn content and a decrease in the relative percentage of random coil (supplemental Fig. 3S).

Interaction of AIF with NAD(P)H
Redox-linked Conformational Changes in the N-terminal and 509 -559 Peptides-Given that proteolytic processing of the AIF N terminus is a critical event leading to caspase-independent apoptosis, we examined how PN cofactors affect susceptibility of the mature recombinant flavoprotein to specific and nonspecific proteases, calpain I and trypsin, respectively. According to the N-terminal amino sequencing, calpain attacks oxidized AIF⌬1-53 at two sites, Gly 102 -Leu 103 and Glu 118 -Gly 119 , producing 55 and 53-kDa fragments regardless of whether or not NAD(P) was present, and only at Gly 102 -Leu 103 when AIF was reduced with NAD(P)H (Fig. 2F). Inability of calpain to access the Glu 118 -Gly 119 site in the AIF⅐NAD(P)H complex was the third indication that the N-terminal peptide, previously designated as a linker connecting the membrane-spanning and catalytic fragments of AIF with no functional role (1), is actively involved in AIF⅐NAD(P)H association. Unlike calpain, trypsin proteolyzed only one Arg 126 -Ala 127 bond in oxidized AIF, and two bonds, Arg 126 -Ala 127 and Lys 517 -Ser 518 , in the NAD(P)H-reduced protein (Fig. 2G). The redox-sensitive trypsinolysis site is situated in the AIF-specific 509 -559 insertion, part of which forms two short ␣-helices and blocks the active site in the ligand-free protein (supplemental Fig. 4S) (10,20). Evidently, the Lys 517 -containing helical stretch must unfold upon reduction to become accessible to trypsin. Because the content of random coil decreases and more ␤-strands are formed in the reduced protein, the 509 -559 peptide likely adopts a different conformation rather than becoming fully disordered when the AIF⅐NAD(P)H complex is formed.
Native AIF Undergoes Redox-dependent Monomer-Dimer Transition, and Its Release from Mitochondria Is Inhibited by NAD(P)(H)-To confirm that the recombinant flavoprotein is structurally and functionally similar to the native one, we investigated properties of mitochondrial AIF. As two-dimensional BN-SDS-PAGE analysis showed, native AIF exists as an equimolar mixture of monomers and dimers in the digitoninsolubilized fraction of rodent liver mitochondria that were untreated or treated with NAD(P) prior to addition of the detergent, and predominantly dimeric when the incubation media contained NAD(P)H (Fig. 3, A and B). When NADH and NBT, an electron acceptor that turns dark-blue upon reduction, were used as substrates for in-gel reductase activity assay, only high molecular weight respiratory supracomplexes but not bands corresponding to AIF were detected (Fig. 3D). Thus, similar to recombinant AIF, the native flavoprotein forms long lived dimeric complexes with reduced cofactors and possesses a low NADH oxidase activity.
Another common feature between the proteins was susceptibility to calpain. AIF released from mitochondria treated with cBid or Atr (inducers of morphological changes and the inner membrane remodeling (27)) migrated as the 55-kDa fragment of recombinant AIF (Fig. 4A) and, according to Yuste et al. (7), was also processed at Gly 102 -Leu 103 . The smaller digestion product corresponding to the 53-kDa fragment of recombinant AIF could be detected only when permeabilized mitochondria were incubated with calpain for extended periods of time. Based on these results, one can conclude that AIF in healthy or freshly permeabilized, respiring mitochondria predominantly exists as a PN-bound dimer and converts into a ligand-free monomer when reduced cofactors are consumed.
Next, we tested how addition of various PN into the respiration media affects AIF proteolysis. Surprisingly, not only reduced but also oxidized cofactors were found to inhibit AIF proteolysis/release (Fig. 4, B and C), with NADP/NADPH hav-   A and B, reduction-induced dimerization of mitochondrial AIF. Mice liver mitochondria were solubilized with digitonin in the absence or presence of 2 mM NAD(P)(H) and separated by two-dimensional BN-SDS-PAGE followed by immunodetection with anti-AIF antibody raised against the 517-531 peptide. C, to prove that an increase in the molecular size of native AIF was due to redox-dependent monomerdimer transition and not because of complex formation with other proteins or other artifacts, recombinant AIF⌬1-53 was subjected to a similar procedure. D, in-gel NBT reductase activity assay. Digitonin-solubilized mitochondria were separated by BN-PAGE, and the excised gel strip was soaked for 3 h in 50 mM Tris-HCl, pH 7.5, 2 mM NBT, and 2 mM NADH. Lacking accumulation of a dark-blue NBT reduction product in the area corresponding to AIF monomers and dimers is indicative of a low NADH oxidase activity of the flavoprotein.
ing a stronger effect than NAD/ NADH. The protective action of all PN was more pronounced in mitochondria respiring on malate/glutamate than in the succinate/rotenone media. This suggests that the fully functional respiratory system is required for the maximal AIF protection, which in turn could be due to specific AIF-complex I interactions or because of an optimal lipid and protein environment created during turnover of the complete electron transfer chain. The PN effect was also concentrationdependent, reaching maximum at 2-5 mM. At these levels, AIF release was diminished to the same extent as in the presence of calpeptin, a specific inhibitor of calpain. Because most of mitochondrial AIF appears to be already PN-bound during the course of our experiments, one can speculate that the externally added cofactors act indirectly by modulating mitochondrial permeability, morphology, and protein-protein interactions. Elucidation of the exact mechanism underlying our observations will require special investigations. NAD(P)H Binding Affects AIF-DNA Interaction-It has been established that, upon proteolysis and the outer mitochondrial membrane permeabilization, detached AIF translocates to the nucleus of a dying cell to trigger chromatin condensation and large scale DNA fragmentation (1). Although it is still unknown how AIF executes its apoptogenic function, there are reports demonstrating that refolded AIF⌬1-120 can directly bind DNA and RNA in a sequence-independent manner and degrade DNA in cooperation with CypA (4,20,25). CypA is also required for translocation of AIF from the cytosol to the nucleus (28). Heat shock protein 70 (Hsp70), on the other side, inhibits AIF-mediated apoptosis by retaining the flavoprotein in the cytoplasm and prevents AIF-induced chromatin condensation in a cellfree system (29 -31). We evaluated effects of PN, CypA, and Hsp70 on the DNA-binding ability of our   (lanes 11-14). AIF⌬1-101. It was found that electrophoretic mobility of linearized double-stranded DNA was perturbed at high AIF:DNA ratios (Ͼ40 -50:1 (w/w)), was not affected by magnesium ions and NAD(P), and was greatly diminished in the presence of NAD(P)H (Fig. 5A). This is in contrast to refolded AIF, 10-fold lower amounts of which were required to observe DNA retardation that was somewhat assisted by Mg 2ϩ and NAD(P)H and promoted by NADP (25). Some of these discrepancies could in part be due to a 6-histidine tag attached to the protein studied previously. As we found, tag-fused AIF has a 10-fold higher affinity and forms high molecular weight complexes with DNA at high protein concentrations (supplemental Fig. 4S).
Addition of CypA to the AIF-DNA mixture did not change the DNA migration pattern, whereas, in the presence of Hsp70 the inhibiting effect of NADH on the AIF-DNA interaction was eliminated (Fig. 5B). Because under these conditions the DNAbinding ability of Hsp70 itself is considerably lower (supple-mental Fig. 4S) and the protein does not perturb AIF⅐NADH interaction and stability of CTC, we conclude that notable changes in DNA retardation were caused by formation of a specific complex between Hsp70 and reduced AIF. Another important observation was that our AIF⌬1-101 displayed negligible DNA nicking and linearizing abilities and, unlike the refolded flavoprotein (4), could not degrade supercoiled plasmid DNA in the presence of CypA (Fig. 5C). This suggests that in the nucleus of a dying cell AIF may recruit proteins other than CypA to effectively initiate chromatin degradation or, as Yuste et al. found (32), may not be involved in DNA fragmentation at all.
The Active Site His 453 Is Crucial for NADH Binding-To get an insight into how CTC is stabilized and how redox-linked changes in the active site are transmitted to the surface, we compared crystal structures of AIF and other oxidoreductases of the glutathione reductase family. As seen from a structural overlay (Fig. 6A), the active sites are highly conserved except one distinct residue in AIF, His 453 , substituted by Leu or Ile in other homologues. The H453L replacement was found to profoundly affect redox properties of AIF. Compared with WT, the mutant had a 56-mV higher redox potential (Table 1), two orders of magnitude lower affinity for NADH (K d of 40 M), more effectively catalyzed electron transfer reactions (Table 2) and reacted with the cofactor without CTC formation (Fig. 6B), producing a structurally different and easily oxidizable intermediate (Fig. 6C-F). Hence, His 453 is a key residue that not only modulates the FAD redox potential but also defines the manner of AIF⅐NADH interaction. Being situated in the vicinity of the isoalloxazine ring of FAD, the imidazole ring of His 453 could directly interact and optimally orient the nicotinamide group for CTC formation, wherein its positional shift, shown by an arrow in Fig. 6A, could transmit the redox signal to the surface via adjacent peptides. Given that, similar to the H453L mutant, refolded AIF does not form CTC with equimolar NAD(P)H (9), it is plausible to suggest that the His 453 -containing peptide is misplaced during refolding. Evidently, determination of three-dimensional structures of correctly folded oxidized and NADH-bound AIF is necessary to precisely elucidate the redox mechanism and identify alterations caused by refolding.  (45)). AIF residues are labeled. A hypothetical movement of the side chain of His 453 to establish a direct hydrogen bond with the nicotinamide and assist in orienting and tighter binding of the cofactor is shown by an arrow. B, anaerobic titration of AIF⌬1-77 H453L with NADH. K d for NADH calculated from the plot shown in the inset was 300-fold higher than that determined for WT (40 M versus 150 nM, respectively). C and D, CD signals produced by WT (solid line) and H453L AIF⌬1-77 (dotted line) indicate that the oxidized proteins have similar folds (C) but the NADH-bound intermediates are structurally different (D). E and F, proteolytic digests of AIF H453L with calpain and trypsin, respectively. In contrast to WT, binding of NADH to the mutant did not protect the Glu 118 -Gly 119 site from a calpain attack (E) and led to complete digestion by trypsin (F).

DISCUSSION
There are two important conclusions that follow from this work. First, our finding that the manner of FAD incorporation and the length of the N terminus define redox properties of AIF implies that previous studies involving truncated, refolded, and tag-fused fragments of the protein need to be carefully reexamined, whereas future investigations must ensure that correctly folded AIF is employed. Secondly and most importantly, our results provide novel insights into the redox mechanism of AIF suggesting that, in addition to catalyzing electron transfer reactions, the flavoprotein may play a role in redox signaling.
Protein monomer-oligomer transition is a key process in cellular signal transduction. The AIF⅐NAD(P)H complex formation triggers not only dimerization but also a conformational change in the 509 -559 peptide and other, yet to be identified parts of the molecule. We propose that these redox-linked structural changes are functionally important. The 509 -559 stretch of amino acids contains two regulatory elements (supplemental Fig. 5S): the proline-rich motif (PPXXPXXPXXP), a putative recognition/interaction site for Src homology 3 domains found in numerous cytoskeletal and signaling proteins (33) and capable of mediating assembly of large multiprotein complexes (34), and the proline/glutamate/serine/threoninerich (PEST) sequence, a potential proteolytic signal and phos-phorylation site. Some of the PEST sequences, present in key metabolic enzymes that exhibit rapid changes in concentration and must be short-lived (35), are conditional proteolytic signals masked in one form of the protein and activated in another via conformational reorganization or serine/threonine phosphorylation. Given that the conformation and solvent exposure of the 509 -559 peptide are defined by the redox state of FAD, it is plausible to hypothesize that the AIF-specific insertion represents a regulatory element through which redox changes in the active site are transmitted to the surface and further to the neighboring/interacting proteins. This way, NAD(P)H binding could control AIF-mediated signaling pathways as well as the lifetime, nuclear translocation of the flavoprotein, and, as demonstrated here, its interaction with DNA.
Mutagenesis studies on human AIF⌬1-120 help to understand how ligand association and a conformational switch in the regulatory peptide could change affinity for DNA. Ye and coworkers (20) showed that replacement of Lys 510 and Lys 518 (Lys 509 and Lys 517 in mouse AIF, respectively) with Ala prevented DNA binding and induction of cell death. Being spatially oriented in oxidized AIF (supplemental Fig. 5S), these lysines could contribute to a positive electrostatic surface potential and promote interaction with the negatively charged DNA. Elimination of positive charges or, as this study predicts, NAD(P)H- The AIF-mediated protein-protein interactions protect the membrane linker from proteolysis and assist cristae formation and proper assembly/functioning of the respiratory chain. When NAD(P)H are depleted, AIF becomes a ligand-free monomer and looses contacts with the neighboring molecules. Alternatively, multiprotein complexes involving AIF could be disrupted due to aberrations in the inner membrane. In either case, changes in mitochondrial morphology are reversible and result in transient redox uncoupling and increased reactive oxygen species levels. Finally, activation of calpain and cathepsin proteases coupled with an increase in permeability of the outer mitochondrial membrane result in AIF proteolysis and release from the IMS (point of no return). Ligand-free AIF monomers translocate to the nucleus and initiate caspase-independent apoptosis, wherein the NAD(P)H-bound dimers may be retained in the cytosol. Even if PN-bound AIF reaches or newly forms in the nucleus, it would have a lower ability to trigger apoptotic changes because of the lower affinity to DNA. At any stage, redox-linked structural reorganization may affect post-translational modification of AIF and, as a consequence, its degradation, retention in the cytosol and interaction with known and yet to be discovered partners. This way, NAD(P)H could control both vital and lethal functions of AIF.
induced AIF dimerization and a conformational switch in the 509 -559 peptide could change the surface charge distribution and lower affinity for DNA.
Our results suggest also that the redox-induced structural reorganization may affect interaction of AIF with Hsp70 (Fig.  5B), a cryoprotectant that inhibits apoptosis through association and retaining AIF in the cytosol (29 -31). When cellular levels of NAD(P)H are sufficiently high, a preferable sequestering of reduced AIF by Hsp70 would be beneficial, because it could slow down or prevent initiation of an AIF-mediated cascade of apoptotic events and give the damaged cell a chance to recover. And vice versa, depletion of reduced PN leading to AIF oxidation and monomerization would manifest serious aberrations in the cellular homeostasis and retaining of the flavoprotein in the cytoplasm would be of little help. Although we did not obtain any evidence for cooperation between AIF and CypA in DNA binding and degradation, a possibility of their direct interaction exists and needs to be further explored. Nucleocytoplasmic transport can be regulated by post-translational modification (phosphorylation, methylation, and ubiquitination) of target proteins (36). Given that the 509 -559 loop is a putative phosphorylation site and AIF was shown to undergo serine and threonine phosphorylation (9) and ubiquitination (37), it is possible that interaction of the flavoprotein with CypA as well as other partners, such as T-cell ubiquitin ligand and X-linked inhibitor of apoptosis (37,38), is controlled by the redox state-dependent modification of the flavoprotein.
In contrast to our results, several in vivo and in vitro studies led to a conclusion that the apoptogenic function of AIF does not depend on its oxidoreductase activity (1,2,9,39). As a control in apoptotic assays, these studies employed refolded, polyhistidine-tagged recombinant or intracellularly expressed AIF fused to the green fluorescent protein. Because truncation, misfolding, and attachment of charged or bulky markers could perturb not only the FAD redox potential, NAD(P)H binding, and dimerization of AIF but also its interaction with DNA and other proteins, there is a possibility that the PN-binding ability of the control proteins was perturbed and, as a consequence, assessment of the redox dependence of the apoptogenic action of AIF was inaccurate. Another observation used as an argument is that AIFsh, a naturally occurring isoform consisting of the C-terminal residues 353-613, provokes the same apoptotic effect as AIF⌬1-120 (21). The structure of the 353-613 fragment derived from the x-ray model of AIF⌬1-120 (supplemental Fig. 6S) suggests that AIFsh must aggregate or adopt a different conformation in solution to stabilize long surface loops and bury hydrophobic patches. If the structures of AIFsh and AIF⌬1-120 differ, their apoptotic mechanisms may differ as well.
Finally, for the first time, we showed that processing/release of mitochondrial AIF can be regulated by PN cofactors. Little is known about PN homeostasis in the IMS, and elucidation of the mechanism underlying our findings requires special investigations. Nevertheless, based on the results presented here and reported previously (12)(13)(14)(15)(16)(17)19) as well as the facts that massive depletion of cellular NAD(P)H is an early event in apoptosis (40), mitochondria are the major store of NAD that can leak into or be replenished from the cytosol (41,42), and the redox state of the NAD(P)(H) pool is an important modulator of mitochondrial function (43), we hypothesize that mitochondrial AIF may act as a redox sensor whose normal and apoptogenic functions are controlled by NAD(P)H according to the following mechanism (Fig. 7). Under normal conditions, when reduced cofactors are abundant in mitochondria, AIF predominantly exists as a NAD(P)H-bound dimer and participates in redox cycling, cristae formation, and proper assembly/functioning of the respiratory chain directly or indirectly via protein-protein interactions. PN depletion and aberrations in the inner membrane can transiently disrupt multiprotein complexes and increase reactive oxygen species production (reversible, preapoptotic state). Prolonged PN deficiency and morphological changes coupled with protease activation and increased permeability of the outer membrane lead to AIF proteolysis and release from the IMS (point of no return). Ligand-free monomers translocate to the nucleus and initiate caspase-independent apoptosis, wherein PN-bound dimers may be retained in the cytosol. If NAD(P)H-bound AIF does enter the nucleus, its apoptogenic ability would be considerably lower due to the lower affinity to DNA. Elucidation of whether AIF can function via the outlined mechanism and link PN-dependent metabolic pathways to apoptosis is vital for better understanding of cellular communications and developing new therapeutic strategies and should be the focus of future studies.