Yeast Frataxin Sequentially Chaperones and Stores Iron by Coupling Protein Assembly with Iron Oxidation*

We have investigated the mechanism of frataxin, a conserved mitochondrial protein involved in iron metabolism and neurodegenerative disease. Previous studies revealed that the yeast frataxin homologue (mYfh1p) is activated by Fe(II) in the presence of O 2 and assembles stepwise into a 48-subunit multimer ( (cid:1) 48 ) that sequesters > 2000 atoms of iron in 2–4-nm cores structurally similar to ferritin iron cores. Here we show that mYfh1p assembly is driven by two sequential iron oxidation reactions: A ferroxidase reaction catalyzed by mYfh1p induces the first assembly step ( (cid:1) 3 (cid:1) 3 ), followed by a slower autoxidation reac- tion that promotes the assembly of higher order oligomers yielding (cid:1) 48 . Depending on the ionic envi- ronment, stepwise assembly is associated with accumulation of 50–75 Fe(II)/subunit. Initially, this Fe(II) is loosely bound to mYfh1p and can be readily mobilized by chelators or made available to the mitochondrial enzyme ferrochelatase to synthesize heme. Transfer of mYfh1p-bound 20 min at 30 °C. The ferro- chelatase reaction was stopped by adding 1 M NaOH and pyridine (176 (cid:3) l each), and iron-deuteroporphyrin was measured by the pyridine hemochromogen method (29) with a (cid:8) (cid:2) for the (reduced (cid:6) oxidized) difference spectrum (30). Competition assays were designed to test if transfer of Fe(II) from mYfh1p to ferrochelatase can occur in the presence of a Fe(II) chelator, as was done by others to study the transfer of copper from a metallochaperone to its target protein (31). Unlike in the transfer equilibrium between a copper chap- erone and its target protein, we measured the end product of the transfer reaction, i.e. heme. Thus, upstream and downstream steps that are also susceptible to iron chelation had to be considered in choosing an appropriate chelator. We have shown that mYfh1p assembles stepwise into an 840-kDa molecule sequestering up to 50–75 Fe(II) atoms per subunit; this iron is accessible to direct chelation until it is oxidized and incorporated into a stable ferrihydrite mineral (Refs. 22 and 24 and this study). Yeast ferrochelatase is a homodimer of (cid:1) 80 kDa containing one Fe(II) substrate binding site and one protoporphyrin binding cleft per subunit; heme synthesis requires the insertion of the Fe(II) atom into the porphyrin ring (32). A second site in each ferrochelatase subunit is thought to be involved in the initial Fe(II) binding or enzyme regulation (32). Citrate is a relatively weak Fe(II) chelator (Fe(II)-binding constant (cid:5) 10 4 M (cid:6) 1 (33) believed to one of the most abundant ligands to the “free” iron pool in (Refs. and and pH 8.0, 150 m M KCl, or ( C ) 10 m M HEPES-KOH, pH 7.3, 10 m M MgCl 2 . In A and B , in the presence of mYfh1p there is some initial overshoot of the electrode of (cid:1) 3 (cid:3) M O 2 , indicating that the electrode response time (half-life (cid:5) 6 s) is not adequate to monitor the fast reaction under these conditions.

Mitochondria require micromolar concentrations of iron to support the heme and the iron-sulfur cluster biosynthetic pathways (1,2). Making this iron bioavailable while limiting its participation in free radical reactions is an essential function accomplished by mechanisms that remain largely uncharacterized (2)(3)(4). The importance of these mechanisms is exemplified by Friedreich ataxia (FRDA), a severe neuro-and cardio-degenerative disease (5) in which mitochondria lack the ability to handle iron properly (reviewed in Ref. 6). FRDA is caused by defects in frataxin, a conserved nucleus-encoded mitochondrial protein of as yet unknown function (6,7). Studies in Saccharomyces cerevisiae have shown that the loss of frataxin results in accumulation of iron in mitochondria, widespread oxidative damage to mitochondrial and nuclear DNA via Fenton chemistry, and impaired respiration (8 -11). This phenotype can be explained by new findings that yeast frataxin is required for the biosyntheses of iron-sulfur clusters (12)(13)(14)(15)(16) and heme (17), two processes critical for maintenance of mitochondrial iron homeostasis (18,19).
An open question is how frataxin influences two different iron-dependent pathways and also provides protection from iron toxicity. We have proposed that such diverse roles could be reconciled if the basic function of frataxin were to bind and store iron in a bioavailable and nontoxic form (20). Our studies with recombinant yeast frataxin have shown that the protein is activated by Fe(II) in the presence of O 2 and forms an oligomeric species (␣ 3 ) that catalyzes Fe(II) oxidation (21). When the Fe(II) concentration exceeds the iron-loading capacity of ␣ 3 , stepwise assembly of ␣ 3 oligomers yields a 48-subunit multimer (␣ 48 ) that sequesters ϳ2,400 atoms of ferric iron. The multimer is a regular spherical particle with a hydrodynamic radius of ϳ11 nm and contains small iron cores of 2-4 nm (22) with Fe-O and Fe-Fe interactions similar to those found in ferritin iron cores (23). Similarly, recombinant human frataxin assembles during expression in Escherichia coli yielding regular spherical particles of ϳ1 MDa and ordered polymers of these particles that sequester up to 10 atoms of iron per subunit in small cores structurally identical to the yeast frataxin iron cores (23). High molecular weight forms of frataxin can be detected by gel filtration and Western blotting in yeast cells or mouse cardiac tissue, and the native protein binds stoichiometric amounts of 55 Fe in metabolically labeled yeast cells (24,25). These previous findings support the idea that frataxin, like ferritin, has an iron storage role. Here, we have tested if frataxin might also serve as a reservoir of bioavailable iron. We describe the coupled stepwise-assembly/iron-oxidation reaction of yeast frataxin and show that this mechanism is compatible with both iron chaperone and storage functions. serum albumin were from Sigma, and bovine brain calmodulin from Calbiochem. All buffers and solutions were made with milli-Q-deionized water (18 M⍀). Stock solutions of ferrous ammonium sulfate (2-10 mM) were freshly prepared in water previously deaerated by purging with argon gas (Ͻ0.2 ppm O 2 ). Calmodulin and albumin were desalted into the appropriate buffer using NAP-25 columns (Amersham Biosciences). The mature forms of yeast frataxin (mYfh1p and mYfh1p[C98A]) and yeast ferrochelatase were expressed in E. coli (24,26) and purified as previously described (22,27). The construct for expression of mYfh1p[C98A] was created via PCR-mediated site-directed mutagenesis as previously described (24). Human H-and L-chain apoferritin homopolymers (28) (designated H-and L-apoferritin) were a generous gift of P. Arosio (Brescia University, Brescia, Italy) and S. Levi (Ospedale San Raffaele, Milano, Italy). Protein concentration was determined from the absorbance and extinction coefficient (⑀ 280 nm ϭ 20,000/44,200/ 27,900/34,000 M Ϫ1 cm Ϫ1 for mYfh1p, bovine serum albumin, H-and L-chain apoferritin, respectively, and ⑀ 276 nm ϭ 3,000 M Ϫ1 cm Ϫ1 for calmodulin). Iron concentration was either directly measured by inductively coupled plasma mass spectrometry (ICP-MS) at the Mayo Metals Laboratory or deduced from the concentration of Fe[BIPY] 3 2ϩ (⑀ 520 nm ϭ 9,000 M Ϫ1 cm Ϫ1 ) (21).
Electrode Oximetry, Ultrafiltration, Gel Filtration, and Fluorescence Measurements-Measurements of dissolved O 2 concentration were performed with a MI-730 micro-O 2 electrode (Microelectrodes, Inc.) (21). The drift of the electrode was ϳ2 M/60 min at 30 or 20°C. Iron binding by mYfh1p and other proteins were measured by ultrafiltration with a molecular mass cutoff of 5 kDa (21). To analyze stepwise assembly of ␣ 48 , independent samples containing identical concentrations of mYfh1p and Fe(II) were incubated at 30°C for different periods of time. Each sample was rapidly cooled down to 4°C to stop assembly, and analyzed by Superdex 200 or Sephacryl 300 gel filtration (22). Tryptophan fluorescence intensity was measured in a Quanta Master fluorimeter (Photon Technology International, Ontario, Canada) with a monochromator bandwidth of 2-4 nm and a pathlength of 4 mm. Excitation was at 294 nm, and tryptophan emission was quantitated from the area under the emission band integrated from 300 to 400 nm.
Fe[BIPY] 3 2ϩ and Ferrochelatase Assays-Fe(II) was added to purified mYfh1p monomer, H-or L-apoferritin, or calmodulin in 10 mM HEPES-KOH, pH 7.3, and each sample (8 ml) was incubated at 30°C. Two aliquots were withdrawn at the indicated time points. BIPY was added to the first aliquot (500 l) at a final concentration of 2 mM, and after 5 min of incubation at room temperature, the concentration of Fe[BIPY] 3 2ϩ was determined (21). Ferrochelatase and deuteroporphyrin IX were added to the second aliquot (300 l) at final concentrations of 2 and 118 M, 2 and 200 M, or 4 and 400 M, respectively, and incubation was continued for an additional 20 min at 30°C. The ferrochelatase reaction was stopped by adding 1 M NaOH and pyridine (176 l each), and iron-deuteroporphyrin was measured by the pyridine hemochromogen method (29) with a ⌬⑀ (545-530) nm ϭ 15.3 mM Ϫ1 cm Ϫ1 for the (reduced Ϫ oxidized) difference spectrum (30). Competition assays were designed to test if transfer of Fe(II) from mYfh1p to ferrochelatase can occur in the presence of a Fe(II) chelator, as was done by others to study the transfer of copper from a metallochaperone to its target protein (31). Unlike in the transfer equilibrium between a copper chaperone and its target protein, we measured the end product of the transfer reaction, i.e. heme. Thus, upstream and downstream steps that are also susceptible to iron chelation had to be considered in choosing an appropriate chelator. We have shown that mYfh1p assembles stepwise into an 840-kDa molecule sequestering up to 50 -75 Fe(II) atoms per subunit; this iron is accessible to direct chelation until it is oxidized and incorporated into a stable ferrihydrite mineral (Refs. 22 and 24 and this study). Yeast ferrochelatase is a homodimer of ϳ80 kDa containing one Fe(II) substrate binding site and one protoporphyrin binding cleft per subunit; heme synthesis requires the insertion of the Fe(II) atom into the porphyrin ring (32). A second site in each ferrochelatase subunit is thought to be involved in the initial Fe(II) binding or enzyme regulation (32). Citrate is a relatively weak Fe(II) chelator (Fe(II)-binding constant ϭ 10 4 M Ϫ1 ) (33) believed to represent one of the most abundant ligands to the "free" iron pool in vivo (Refs. 19 and 34 and Refs. therein). In ferrochelatase assays performed under strictly anaerobic conditions, Fe(II) can be provided as a ferrous citrate salt (35). Thus, citrate should not be able to remove the Fe(II) ion from ferrochelatase after the transfer or to destabilize heme, as has been shown to occur with thiol reagents (36). Moreover, at the neutral pH and under the aerobic conditions used in our assays, citrate will promote rapid autoxidation of Fe(II) (37) such that any mYfh1p-bound Fe(II) mobilized by citrate will be rapidly oxidized and excluded from the reaction. Both citrate and a stronger chelator, EDTA (EDTA Fe(II)-binding constant ϭ 10 14 M Ϫ1 ) (38), were used in competition assays. We used citrate/total iron ratios ranging from 0.06:1 to 166:1, and citrate/ferrochelatase ratios ranging from 1:1 to 2500:1, which encompass and exceed the ratios used in anaerobic ferrochelatase assays (citrate/Fe ϭ 1:1; citrate/ferrochelatase ϭ 28:1) (35). EDTA/total iron ratios ranged from 0.33:1 to 7:1 and EDTA/ferrochelatase ratios from 5:1 to 100:1.

RESULTS
Stepwise Assembly of mYfh1p Is Coupled with Two Sequential Iron Oxidation Reactions-At Fe(II)/mYfh1p ratios Յ0.5, mYfh1p catalyzes Fe(II) oxidation with a stoichiometry of ϳ2 Fe(II)/O 2 and production of H 2 O 2 (ferroxidase reaction) (21). A ϳ50-kDa oligomer (␣ 3 ) is responsible for this activity suggesting that three mYfh1p subunits may form one binuclear ferroxidation site (21). Here, we have analyzed the iron oxidation reaction of mYfh1p at concentrations of iron (40 -75 Fe(II)/ mYfh1p) that encompass the iron loading capacity of mYfh1p (50 -75 Fe(III)/mYfh1p depending on the ionic environment) and result in stepwise assembly of ␣ 3 to yield iron-loaded ␣ 48 (22,24). Fig. 1A shows representative O 2 consumption curves recorded when 100 M Fe(II) was incubated in 10 mM HEPES-KOH, pH 7.3, in the absence or presence of 2 M mYfh1p (Fe(II)/mYfh1p ϭ 50/1). In buffer without protein there was an initial lag phase due to the time required to generate sufficient hydrolyzed Fe(III) to initiate autoxidation of Fe(II) (39) (Fig.  1A, black plot). The final Fe(II)/O 2 stoichiometry was 3.7 Ϯ 0.3 (n ϭ 3) as expected for autoxidation (39). In the presence of mYfh1p, the initial rate of O 2 consumption was faster compared with buffer, consistent with ferroxidase activity, but became slower after the first 4 min (Fig. 1A, red plot). The final Fe(II)/O 2 stoichiometry was 3.6 Ϯ 0.3 (n ϭ 3), indicating that ferroxidation was rapidly overcome by autoxidation. We were unable to detect any H 2 O 2 released into the solution, which should be expected given the high Fe(II)/mYfh1p ratio used in these experiments (21,40).
In Fig. 1B, gel filtration was used to determine the speciation of mYfh1p during the iron oxidation reaction described above. Experimental conditions were similar to those employed for electrode oximetry in Fig. 1A except that both the protein and the iron concentrations were increased 2-fold to enable detection of mYfh1p by absorbance measurements. Such an increase is not expected to change the rate of ␣ 48 assembly to a significant degree (22,24), and therefore the gel filtration data (Fig. 1B) can be correlated with the two phases in the O 2 consumption curve of mYfh1p (Fig. 1A). The initial faster phase (Fig. 1A, 0-4 min, red plot) was associated with the assembly of an oligomer of ϳ50 kDa (Fig. 1B, 3 min), while the subsequent slower phase (Fig. 1A, 4-50 min, red plot) was associated with stepwise assembly of higher order oligomers (Fig. 1B, 6, 10, and 30 min), in agreement with the previously described progression, ␣ 3 ␣ 3 3 ␣ 6 3 ␣ 12 3 ␣ 24 3 ␣ 48 (22,24). The A 280 of assembled mYfh1p species increased in a time-dependent manner (Fig. 1B). At the end of the iron oxidation reaction, the A 280 of ␣ 48 was much higher than the A 280 of monomer analyzed in the absence of any added Fe(II) (Fig. 1B, peak ␣). These results are consistent with progressive oxidation of Fe(II) to Fe(III) and formation of a ferrihydrite-like mineral (which, unlike Fe(II), absorbs at 280 nm) within the assembled protein (23).
We showed previously that the mature form of Yfh1p is generated by cleavage of an N-terminal mitochondrial targeting signal between residues 51-52 (24,41). This cleavage eliminates one cysteine residue at position 32. The mature form of the protein (amino acids 52-174), which is the form used in our experiments, contains one cysteine residue at position 98. Thus, an alternative explanation for the results in Fig. 1B could be that chelation of iron by cysteine residues from different mYfh1p subunits leads to formation of metal-thiolate ag-gregates (42). However, others have reported that when yeast frataxin is treated with iodoacetamide to block any exposed cysteine residues and subsequently incubated with 20 equivalents of Fe(II), iron-dependent oligomerization is not affected (43). We obtained similar results using a mYfh1p variant in which cysteine 98 was replaced by an alanine residue (data not shown). We therefore conclude that mYfh1p assembly is driven by iron oxidation: A ferroxidase reaction catalyzed by mYfh1p is associated with the first assembly step (␣ 3 ␣ 3 ), followed by a slower autoxidation reaction associated with assembly of higher order oligomers to ultimately yield ␣ 48 .
The Ferrous Iron Sequestered by mYfh1p Is Bioavailable-The time required to complete the iron oxidation reaction of mYfh1p is in the order of hundreds of seconds (Fig. 1A), much longer than the iron oxidation reaction of ferritin, which is in the order of tens of seconds (Ref. 40 and data not shown). At the beginning of its reaction, however, mYfh1p rapidly sequesters up to 50 -75 Fe(II)/subunit, which are then progressively oxidized within the assembled protein (Ref. 21,22, and 24 and data presented below). Given that mobilization of iron from ferritin is inefficient in the absence of reducing agents (44), we hypothesized that the coupling of a slow iron autoxidation reaction with stepwise assembly might enable mYfh1p to serve as a temporary reservoir and a chaperone for Fe(II). We therefore measured iron mobilization during mYfh1p assembly by use of ␣,␣Ј-bipyridine (BIPY), a chelator that preferentially binds Fe(II) (45), or purified yeast ferrochelatase, a mitochondrial enzyme that catalyzes the insertion of Fe(II) into protoporphyrin IX to yield heme (reviewed in Ref. 46). Iron mobilization from human H-or L-apoferritin (28) was analyzed in parallel. These two proteins are pre-assembled 24-subunit shells with a negatively charged inner surface that promotes iron autoxidation and mineralization; in addition, H-apoferritin has 24 dinuclear ferroxidation sites (28,47). As a negative control we used calmodulin, a calcium-binding protein with a molecular mass and an isoelectric point similar to those of mYfh1p (17 versus 14 kDa, and 4.09 versus 4.34). Reactions were started by addition of a fixed concentration of Fe(II) (30 M) to buffer in the absence or presence of protein (0.4 M; Fe(II)/subunit ϭ 75:1 for all proteins tested). At successive time points, an aliquot was withdrawn and divided in two parts that were immediately incubated with either BIPY (2 mM) or ferrochelatase (2 M) and deuteroporphyrin IX (118 M). The halflife of BIPY-accessible iron estimated from a single exponential fitting was 21.5 min in the presence of mYfh1p compared with 1.0, 4.0, 6.7, and 8.8 min in the presence of H-apoferritin, L-apoferritin, buffer only, and calmodulin, respectively ( Fig.  2A). Similarly, ferrous iron was more accessible to ferrochelatase in the presence of mYfh1p relative to buffer or calmodulin (Fig. 2B). BIPY can bind Fe(III) and/or reduce Fe(III) to Fe(II) although with lower affinity compared with Fe(II) (48 -50), suggesting that the BIPY accessible iron mobilized from mYfh1p could represent a mixture of both ferrous and ferric iron. However, the concentrations of BIPY-accessible iron at successive time points in the presence of mYfh1p ( Fig. 2A) were in the same order as the concentrations of ferrochelatase-accessible iron measured under similar conditions (Fig. 2B). We therefore conclude that the iron mobilized by direct chelation (i.e. BIPY accessible iron) during mYfh1p assembly is largely in ferrous form, becoming progressively less accessible as it is oxidized to Fe(III). The Fe(II) that can be mobilized by direct chelation can also be donated to ferrochelatase. This should involve a direct mYfh1p-ferrochelatase interaction given that there was no deuteroheme synthesis in samples containing Fe(II), mYfh1p, and deuteroporphyrin IX but not ferrochelatase (data not shown).  Fig. 2, A and B. Three different sets of additions followed. In set 1, 2 M ferrochelatase was added to mYfh1p-Fe(II) after 15 min of incubation, and the incubation continued for another 5 min to allow putative protein-protein interactions to take place. Then, citrate and deuteroporphyrin IX (120 M) were added in rapid sequence and the incubation continued for an additional 20 min, after which deuteroheme levels were measured. In set 2, Fe(II) was incubated with mYfh1p for 15 min as above, after which citrate was added and the incubation continued for 5 min to allow the chelator to access mYfh1p-Fe(II) prior to the putative docking of ferrochelatase onto mYfh1p. Then, ferrochelatase and deuteroporphyrin IX were added in rapid sequence. In set 3, Fe(II) was incubated with mYfh1p for 20 min after which citrate, ferrochelatase, and deuteroporphyrin IX were added in rapid sequence. The time courses in Fig. 2, A and B indicate that 20 min after addition of 30 M Fe(II) to 10 mM HEPES-KOH, pH 7.3, at 30°C, little residual Fe(II) is present in buffer without mYfh1p (Ͻ3 M; Fig. 2A, black plot) while ϳ16 M Fe(II) is still available in the presence of mYfh1p ( Fig. 2A, red plot). Therefore, in all the three sets described above, heme synthesis will largely depend on ϳ16 M mYfh1p-Fe(II). In all cases, the small size of citrate (Ͻ9 Å) may allow this compound to penetrate the protein and directly access mYfh1p-Fe(II), similar to mobilization of ferritin-iron by direct-chelation (44). However, at neutral pH and in the presence of atmospheric O 2 , which are the conditions used in these assays, citrate will promote rapid autoxidation of Fe(II) (37). Therefore, any mYfh1p-Fe(II) mobilized by citrate will be rapidly oxidized and excluded from the reaction. If ferrochelatase has a high affinity for binding to mYfh1p, as would be expected for a specific intermolecular interaction, the yield of the transfer reaction should be the same between set 1 and set 3. If docking of ferrochelatase onto mYfh1p hampers the ability of citrate to penetrate mYfh1p and directly chelate Fe(II), the yield of the transfer reactions in set 2 should be lower compared with set 1 and set 3. In set 1, the addition of 2-500 M citrate (Fe(II)-binding constant ϭ 10 4 M Ϫ1 ) resulted in a ϳ6 -25% drop in heme levels; however, increasing the citrate concentration to 2 and 5 mM (corresponding to a 66 -166-fold molar excess over the total iron concentration and a 1000 -2500-fold molar excess over the ferrochelatase concentration) did not cause any significant additional decrease (Fig. 2C, gray plot). Compared with set 1, heme synthesis was further decreased in set 2 (Fig. 2C, pink plot) but remained unchanged in set 3 (Fig. 2C, blue plot). These results can be explained by two parallel reactions: (i) transfer of Fe(II) from mYfh1p to ferrochelatase yielding heme; (ii) direct chelation of mYfh1p-Fe(II) by citrate. It appears that mobilization of mYfh1p-Fe(II) by citrate increased with increasing chelator concentrations up to a maximum limited value that was augmented if the chelator was added to mYfh1p-Fe(II) 5 min before the addition of ferrochelatase (Fig. 2C, pink plot). We will show below that a fraction of mYfh1p-Fe(II) can be released into the solution during ultrafiltration. The fraction of mYfh1p-Fe(II) accessible to citrate most likely corresponds to this labile 2ϩ and deuteroheme were determined as described under "Experimental Procedures." The bars represent the mean Ϯ S.D. of 3 (each protein) or 5 (buffer) independent measurements. The traces show single exponential fittings to the data. C, heme synthesis assays were performed under the conditions used in B. Three sets of assays were analyzed: mYfh1p-Fe(II) ϩ FC ϩ [Cit ϩ PPIX], mYfh1p, and Fe(II) were incubated for 15 min, ferrochelatase (FC) was added for 5 min, followed by citrate (Cit) and protoporphyrin IX (PPIX), and the incubation continued for 20 min (n ϭ 2); mYfh1p-Fe(II) ϩ Cit ϩ [FC ϩ PPIX], same as above except that citrate was added to mYfh1p-Fe(II) for 5 min, followed by ferrochelatase and protoporphyrin IX (n ϭ 1); mYfh1p-Fe(II) ϩ Cit ϩ FC ϩ PPIX, mYfh1p and Fe(II) were incubated for 20 min, then citrate, ferrochelatase, and protoporphyrin IX were added in this order (n ϭ 1). Heme levels measured at the end of each assay are plotted versus the citrate concentration.

Transfer of Fe(II) from mYfh1p to Ferrochelatase Occurs in the Presence of Citrate-Copper
mYfh1p-Fe(II) pool (see Table II, 10 min). Thus, in the three sets of assays shown in Fig. 2C, the Fe(II) consumed by citrate affected the yield of the transfer reaction. However, once chelation of Fe(II) by citrate reached its maximum, heme levels did not change significantly even in the presence of a large excess of citrate. This indicates that the transfer reaction is mostly independent of the presence of a physiologic ferrous iron chelator, suggesting that the transfer occurs via a specific intermolecular interaction. Two alternative ways by which Fe(II) transfer might occur include (i) release of Fe(II) from mYfh1p  D, red plots). In other duplicate samples, 16 mM dithionite was added instead of EDTA, and the incubation continued for 10 min at 30°C (E-F, orange plots). Each sample was then treated and analyzed by gel filtration as described above. An equal amount of monomer without any added iron was similarly analyzed (green plot). The peaks denoted by asterisks probably represent dithionite and small levels of ferric oxides. into the solution, and (ii) release of Fe(II) from mYfh1p via general pores on the protein surface. In these scenarios, ferrochelatase and citrate would directly compete for the same Fe(II) pool or the same docking sites. Under either condition, heme synthesis would be expected to decrease proportionally to an increase in the citrate concentration which is not observed in Fig. 2C. Therefore, the results in Fig. 2C better fit a model where ferrochelatase and citrate access mYfh1p-Fe(II) via different paths. Furthermore, the higher levels of heme detected in set 1 and set 3 compared with set 2 suggest that docking of ferrochelatase onto mYfh1p may hamper the ability of citrate to penetrate mYfh1p. Addition of 10 -200 M EDTA (Fe(II)binding constant ϭ 10 14 M Ϫ1 ) (38) resulted in a marked inhibition of heme synthesis (ϳ30% heme synthesized relative to assays without EDTA). However, this strong chelator is expected to mobilize most mYfh1p-bound Fe(II) as is the case for BIPY ( Fig. 2A; see also Fig. 3). EDTA could also chelate Fe(II) from ferrochelatase after the transfer and/or destabilize heme, as has been reported for thiol reagents (36).
The Ferrous Iron Sequestered by mYfh1p Is Loosely Associated with the Protein-To analyze the interaction between mYfh1p and iron during stepwise assembly, Fe(II) (30 M) was incubated in the absence or presence of mYfh1p or L-apoferritin (Fe(II)/subunit ϭ 75:1) under conditions similar to those used in the BIPY and ferrochelatase assays described above. After 10 or 60 min of incubation, each sample was subjected to ultrafiltration with a molecular mass cutoff of 5 kDa (21). In the absence of protein, extensive precipitation of insoluble ferric oxyhydroxides was observed at both time points as expected (Table I). In the presence of L-apoferritin, most iron was recovered in a protein-bound form after either 10 or 60 min of incubation and only very little free iron was observed (Table I), reflecting rapid oxidation of Fe(II) within the protein shell as also seen in Fig. 2A. After 10 min of incubation in the presence of mYfh1p, similar levels of iron (ϳ12 M) were detected in protein-bound and free form (Table I). After 60 min of incubation, the mYfh1p-bound iron increased to 19 M, corresponding to a Fe/mYfh1p stoichiometry of ϳ50/1 (22,24), while free iron decreased to ϳ3 M (Table I). Under the conditions used in this experiment, the iron sequestered by mYfh1p consists of mostly Fe(II) after 10 min of incubation and Fe(III) after 60 min (see Fig. 2, A and B). Therefore, a significant proportion of the Fe(II) sequestered by mYfh1p was released into the solution during ultrafiltration (Table I, 10 min), indicating that this Fe(II) is loosely bound to the protein. However, if Fe(II) was allowed to oxidize inside mYfh1p prior to ultrafiltration, iron release was greatly reduced (Table I, 60 min). We conclude that Fe(II) is loosely bound to mYfh1p but it is not released into the solution unless the binding equilibrium is perturbed as it occurs during  ultrafiltration. Once Fe(II) is oxidized to Fe(III), the iron is more tightly bound to the protein.
Iron Oxidation Stabilizes mYfh1p Assemblies-To analyze iron mobilization from ␣ 48 , samples containing 40 M protein and 1.6 mM Fe(II) were incubated at 30°C. These concentrations changed the reaction kinetics as compared with Fig. 1B, such that ␣ 48 was formed within 2 min of incubation. At different time points, one sample was rapidly cooled down to 4°C to stop assembly (22) and immediately analyzed by gel filtration. A duplicate sample was first treated with EDTA (20 mM; EDTA/Fe(II) ϭ 12.5), a chelator that binds both Fe(II) and Fe(III) with high affinities (10 14 and 10 25 M Ϫ1 , respectively) (38), incubated for an additional 60 min at 30°C, and finally analyzed by gel filtration. In samples that had not been treated with EDTA, mYfh1p monomer (␣) assembled into ␣ 48 and there was a progressive increase in the A 280 of this species at successive time points (Fig. 3, A-F,  black plots). Addition of EDTA after 2 min of incubation resulted in disassembly of ␣ 48 back to smaller assembly intermediates, with a concomitant decrease in the A 280 due to mobilization of mYfh1p-bound iron by direct chelation (Fig.  3A, red plot). A similar result was obtained upon addition of EDTA after 10 min of incubation, although the shift from ␣ 48 to smaller assembly intermediates was less pronounced and the levels of EDTA-accessible iron were significantly decreased compared with the 2-min sample (Fig. 3B, red plot).
Upon addition of EDTA after 1 or 16 h of incubation, protein disassembly was no longer observed and there was a timedependent decrease in the levels of EDTA-accessible iron (Fig. 3, C-D, red plots). These results are consistent with a model in which iron oxidation and mineralization are an integral part of mYfh1p assembly (23). The Fe(II) sequestered by ␣ 3 is progressively oxidized and incorporated into a ferrihydrite crystallite. As the crystallite increases in size, stepwise assembly of trimers is promoted by the alignment and binding of one crystallite to another. As mineralization proceeds, the proportion of iron that can be mobilized by direct chelation decreases, while the stability of mYfh1p assemblies increases. In agreement with this model, EDTA caused time-dependent disassembly of ␣ 48 into smaller oligomers, whereas the reducing agent, dithionite, caused quantitative disassembly of ␣ 48 back to monomer in a time-independent manner (Fig. 3, E and F, orange plots). Others and we have found that the cysteine residue present in the mYfhp sequence is not required for stepwise assembly (Ref. 43 and data not shown). These observations exclude the possibility that the disassembly induced by treatment with dithionite was due to reduction of metal-thiolate aggregates.
mYfh1p Catalyzes Oxidation of Fe(II) in Different Ionic Environments-To investigate a recent report that iron binding by frataxin takes place only at very low ionic strength (43), we analyzed whether mYfh1p exhibits ferroxidase activity in the presence of 150 mM KCl, close to the concentration believed to exist in mitochondria (43,51). Fig. 4 shows representative O 2 consumption curves recorded when 48 M Fe(II) was incubated in 10 mM HEPES-KOH, pH 7.0, 150 mM KCl, in the absence or presence of 96 M mYfh1p (Fe(II)/mYfh1p ϭ 0.5). Except for the presence of 150 mM KCl, these are standard conditions to detect the ferroxidase activity of mYfh1p (21). O 2 consumption was facilitated in the presence of mYfh1p compared with buffer without added protein (Fig. 4). This was not the case for samples containing 96 M albumin instead of mYfh1p (Fig. 4, inset). A stoichiometric Fe(II)/O 2 ratio of 2.1 Ϯ 0.3 (n ϭ 3) was determined for the completed reaction of mYfh1p ( Fig. 4 and data not shown), consistent with the presence of ferroxidase activity (47,52). Stoichiometric Fe(II)/O 2 ratios of 3.8 Ϯ 0.3 (n ϭ 2) and 4.6 Ϯ 0.2 (n ϭ 2), consistent with autoxidation, were otherwise measured for the completed reactions of albumin and buffer without added protein, respectively ( Fig. 4 and data not  shown). These results demonstrate that mYfh1p binds and catalyzes oxidation of Fe(II) at physiologic concentrations of salt.
In aqueous solution, the rate of spontaneous Fe(II) oxidation is influenced by the ionic strength as well as the interaction of Fe(II) with different anions (53)(54)(55). Upon addition of 100 M Fe(II) to 10 mM HEPES-KOH, pH 7.3, O 2 consumption was significantly slowed down in the presence of 150 mM KCl compared with buffer without added salt (compare black plots in Figs. 1A and 5A) as expected (55). Fig. 5A shows a representative O 2 consumption curve for 2 M mYfh1p upon addition of 100 M Fe(II) (Fe(II)/mYfh1p ϭ 50) in 10 mM HEPES-KOH, pH 7.3, 150 mM KCl. There is an initial phase (0 -2 min) during which Fe(II) oxidation is much faster compared with buffer without protein, consistent with ferroxidase activity, followed by a prolonged phase during which Fe(II) oxidation proceeds at a similar slow rate as in buffer without protein (Fig. 5A). The  (Fig. 5C), conditions that were previously reported to prevent iron binding by frataxin (43). These results confirm that two sequential iron oxidation reactions take place at high Fe(II)/mYfh1p ratios in different ionic environments: A faster reaction catalyzed by mYfh1p is followed by a slower autoxidation reaction. In addition, a comparison of the O 2 consumption curves for mYfh1p in Fig. 1A and 5, A-C indicates that both reactions are influenced by the ionic environment, which may depend on salt effects on the protein fold (see below) and the reactivity of Fe(II) toward O 2 (53)(54)(55).
Salt Affects the Rate of Stepwise Assembly but Not the Iron Binding Capacity of mYfh1p-Evidence reported above (Figs.  1, A and B and 3) and elsewhere (23) is consistent with a model in which iron oxidation and biomineralization are integral parts of mYfh1p assembly. Thus, the report that iron-dependent self-assembly of frataxin is inhibited at physiologic concentrations of salt (43) might be explained by salt effects on the kinetics of iron oxidation, and thus on the kinetics of mYfh1p assembly. We analyzed a time course of mYfh1p assembly under conditions similar to those used in Fig. 1B except for the presence of 150 mM KCl during assembly. At the end of a 10-min incubation we could only detect low levels of ␣ 3 and larger oligomers, with ␣ 48 becoming detectable at 20 min (Fig.  5D). The rate of ferrihydrite mineral accumulation (estimated from the increase in the A 280 of ␣ 48 at successive time points) was slower in the presence of 150 mM KCl compared with buffer without added salt (compare Figs. 1B and 5D), consistent with the respective O 2 consumption curves (Figs. 1A and 5A, red plots). In addition, the levels of residual monomer (peak ␣) decreased only minimally at successive time points in the presence of 150 mM KCl (Fig. 5D). One possible interpretation of these results is that the presence of salt inhibits iron binding by mYfh1p, hence iron-dependent protein aggregation is also inhibited (43). On the other hand, iron binding by mYfh1p is expected to occur efficiently in the assembly reactions analyzed in Fig. 5D because the protein exhibits ferroxidase activity under similar experimental conditions (Figs. 4 and 5A). Thus, an alternative explanation is that the salt slows down stepwise assembly of ␣ 3 to ␣ 48 due to the inhibitory effect of KCl on the rate of Fe(II) autoxidation as discussed above. This results in the release of Fe(II) during gel filtration leading to protein disassembly, similar to what we observed in Table I and Fig. 3. If this explanation is valid, the iron loading capacity of ␣ 48 assembled in the presence of 150 mM KCl should not be impaired. We therefore analyzed assembly of ␣ 48 under conditions similar to those employed in Fig. 5D except that the protein and the iron concentration were increased 10-and 8-fold, respectively, to enable determination of the Fe/mYfh1p stoichiometry. Upon gel filtration, fractions corresponding to ␣ 48 were analyzed for protein concentration by SDS/PAGE and for iron concentration by ICP-MS, and a stoichiometric ratio of ϳ70 -75 Fe/mYfh1p was determined (n ϭ 2), which is higher than the ϳ50/1 ratio determined for ␣ 48 samples assembled in the absence of added salts (Refs. 22 and 24 and Table I,   Fractions from Superdex 200 gel filtration (fractionation range 10 -600 kDa) corresponding to ␣ 48 were further analyzed by Sephacryl 300 gel filtration (fractionation range 10 kDa to 1.5 MDa) and the same macromolecular species of ϳ1 MDa was observed for samples assembled in the absence or presence of 150 mM KCl (data not shown). We conclude that physiologic KCl concentrations (43) slow down the rate of Fe(II) autoxidation and thereby influence the rate at which ␣ 3 assembles into ␣ 48 , but do not impair and may even improve the iron loading capacity of mYfh1p. An analogous effect has been described for ferritins, where the more slowly incorporating L-rich variants have larger iron cores than H-rich ferritins (56).
The Iron Chaperone Properties of mYfh1p Are Enhanced by Its Stepwise Assembly-A possible advantage of being able to control Fe(II) autoxidation and stepwise assembly via changes in the ionic environment might be to prolong the availability of the Fe(II) bound to mYfh1p. To test this, salt effects on iron mobilization from mYfh1p were analyzed by incubating 100 M Fe(II) in the absence or presence of 2 M mYfh1p in 10 mM HEPES-KOH, pH 7.3, at increasing KCl concentrations. Fig.  6A shows that more iron is accessible to BIPY and for a longer time in the presence of mYfh1p compared with buffer. Moreover, this effect is enhanced at increasing salt concentrations. This is not the case for buffer without protein where added salt has little or no effect on iron accessibility to BIPY after the first 10 min of incubation (Fig. 6A). There was a progressive increase in tryptophan fluorescence intensity when mYfh1p monomer was incubated in the presence of increasing KCl concentrations, with a maximum change between 100 and 150 mM KCl (Fig. 6B), suggesting that salt concentrations close to physiologic conditions may influence the fold of the protein and improve its function. Table II further shows that mYfh1p keeps more iron available to BIPY or ferrochelatase and for a longer time relative to buffer even though the concentration of Fe(II) decreases at a similar rate in samples with or without mYfh1p (Table II). These results demonstrate that mYfh1p does not simply provide a passive storage compartment for Fe(II), but also enhances the availability of the stored Fe(II), possibly by keeping it at a high concentration within the protein. At any given time point in the presence of mYfh1p, the concentration of BIPY-accessible iron reflects the estimated concentration of residual Fe(II) ( Table II). In contrast, the concentration of ferrochelatase-accessible iron is lower at 10 min but becomes comparable to the estimated Fe(II) concentration at later time points (Table II,  , which should exclude the possibility that these two reagents were limiting relative to the available Fe(II). A similar effect is seen in Fig. 2, A and B, where the concentration of deuteroheme is significantly lower than that of Fe[BIPY] 3 2ϩ at the two earliest time points but not at later time points. From these data it would appear that the transfer of Fe(II) from mYfh1p to ferrochelatase becomes more efficient as mYfh1p assembles into progressively larger oligomers. DISCUSSION The need to maintain a supply of bioavailable iron while avoiding iron toxicity is a central problem in biology (57). In aqueous solutions under aerobic conditions, iron toxicity depends on its tendency to catalyze production of free radicals as illustrated by the iron-catalyzed Haber-Weiss reaction (57) shown in Equations 1 to 3.
Our work indicates that the mature form of the yeast frataxin homologue (mYfh1p) has a mechanism to control iron oxidation and availability in vitro. Initial studies showed that in the presence of Fe(II) and O 2 , mYfh1p monomer assembles stepwise yielding ␣ 3 3 ␣ 6 3 ␣ 12 3 ␣ 24 3 ␣ 48 . The assembly intermediates in this progression were found to accumulate increasing Fe(III) levels, reaching an apparent maximum loading capacity of 50 Fe(III) per subunit in ␣ 48 (22,24). We have recently shown that two different iron oxidation reactions take place during the initial assembly step (␣ 3 ␣ 3 ). A ferroxidase reaction with a stoichiometry of ϳ2 Fe(II)/O 2 is evident at Fe(II)/mYfh1p ratios Յ 0.5, while an autoxidation reaction with a stoichiometry of ϳ4 Fe(II)/O 2 becomes predominant at ratios between 0.5 and 1.5 (21). At Fe(II)/mYfh1p ratios Յ 0.5, only a small fraction of the H 2 O 2 expected from the ferroxidase reaction (47,52) is detected in solution, and concomitant oxidative degradation of mYfh1p suggests that most H 2 O 2 reacts with the protein itself (21). Here, we have analyzed the reaction of mYfh1p at saturating concentrations of iron (40 -75 Fe(II)/ mYfh1p). These ratios were chosen because they encompass the iron loading capacity of mYfh1p and result in stepwise assembly of ␣ 3 to yield iron-loaded ␣ 48 (22,24). Under these conditions, a faster initial phase is rapidly overcome by a slower phase, with a final stoichiometric Fe(II)/O 2 ratio of ϳ4 and no detectable H 2 O 2 released in the solution, consistent with ferroxidase activity being rapidly overcome by autoxida-  tion (Figs. 1A, 5, A-C, and data not shown). The initial phase correlates with assembly of ␣ 3 , while autoxidation is associated with assembly of higher order oligomers yielding ␣ 48 (Figs. 1B and 5D). Oligomerization enables mYfh1p to rapidly sequester up to 50 -75 Fe(II)/subunit depending on the ionic environment. Initially, Fe(II) is loosely bound to the protein and can be readily mobilized, whereas ferric iron is more tightly bound (Tables I and II; Figs. 2, 3, and 6A). We therefore postulate that the mYfh1p reaction is as follows (at Fe(II)/mYfh1p ratios Ն 0.5) in Equation 4, Postulated mechanism of action of mYfh1p. A, assembly at low Fe(II)/mYfh1p ratios that do not exceed the iron loading capacity of ␣ 3 . B, assembly at saturating Fe(II)/mYfh1p ratios. Red and black diamonds symbolize Fe(II) and Fe(III), respectively. Yeast ferrochelatase is shown as a soluble dimer with one molecule of protoporphyrin IX (IX) and one Fe(II)-binding site per subunit (32); in vivo, ferrochelatase is bound to the inner mitochondrial membrane with the iron binding site exposed on the matrix side (67). where (mYfh1p) n represents any of the species formed during stepwise assembly.
This reaction would appear to provide two main advantages compared with spontaneous Fe(II) oxidation in solution (see Equation 1): (i) H 2 O is expected to represent the predominant product of O 2 reduction; (ii) the iron bound to mYfh1p is in a readily accessible form until it is converted to a water-soluble ferrihydrite mineral (23), which is stored within the assembled protein.
One limitation is that our analyses were carried out under controlled conditions of pH (7, 7.3, or 8), temperature (20 or 30°C), and ionic strength (0.01-0.15 N), which are different from the much more complex mitochondrial matrix environment in living cells. The mitochondrial matrix has an alkaline pH of ϳ8 (58) and contains significant concentrations of certain salts (2 M CaCl 2 , 0.8 mM MgCl 2 , and 100 mM KCl) and ironchelating molecules (51,59). These factors are known to influence the rate of Fe(II) oxidation (33,60) and could interfere with the iron oxidation reaction of mYfh1p. However, the ferroxidase activity of mYfh1p (Figs. 4 and 5, A-C), as well as its iron loading capacity and ability to enhance iron availability (Fig. 6A) were conserved at salt concentrations and pH values close to those believed to exist in the mitochondrial matrix. This suggests that what the protein can do in vitro most likely reflects its function in vivo.
Current reports strongly suggest that yeast frataxin controls the iron required for the in vivo biosyntheses of iron-sulfur clusters (13)(14)(15)(16) and heme (17). A recent study further shows that human frataxin functions as an iron donor for assembly of [2Fe-2S] clusters in ISU-type proteins in vitro (61). Here, we performed deuteroheme synthesis assays to assess the availability of the Fe(II) bound to mYfh1p using a physiologic Fe(II) chelator, i.e. the mitochondrial enzyme ferrochelatase. Given that there was no deuteroheme synthesis in samples lacking ferrochelatase (data not shown), we postulate that Fe(II) is near the outer surface of mYfh1p whereby it is transferred to ferrochelatase. We have shown that this transfer can occur in the presence of an excess of a physiologic ferrous iron chelator (Fig. 2C) and at physiologic ionic strength (Table II), and becomes more efficient as mYfh1p assembles into progressively larger oligomers (Table II). These data strongly suggest that transfer of Fe(II) from mYfh1p to ferrochelatase involves an intermolecular interaction. This conclusion is in accord with a recent report showing that: (i) zinc-protoporphyrin, not heme, is synthesized in yeast cells lacking Yfh1p, consistent with a specific role of frataxin in making iron available to ferrochelatase, and (ii) Yfh1p and ferrochelatase physically interact with each other in Biacore experiments (17).
The ability of yeast frataxin to provide iron to such diverse proteins as ISU-type proteins and ferrochelatase indicates that frataxin is different from copper chaperones, which deliver copper ions to specific partners (62). Copper chaperones contain the motif, M(T/H)CXXC, also present in their target proteins, that bind metal ions via the two cysteine residues (62). Metal transfer requires the docking of the chaperone and target protein with their metal binding domains close to each other, followed by metal exchange via formation of intermediates that involve the cysteine residues from both proteins (62). It would appear that yeast frataxin is a different type of metallochaperone, acting as a general reservoir of Fe(II) atoms and making them available to different users perhaps via hydrophobic interactions mediated by the conserved neutral surface found on the protein (43). Importantly, if the Fe(II) is not transferred to iron users, it is oxidized and stored in a water-soluble mineral within the assembled protein (23).
Our previous findings (21-24) together with the results re-ported here support the following mechanism for yeast frataxin (Fig. 7). Monomer is activated by Fe(II) in the presence of O 2 and forms an oligomer, ␣ 3 , with a negatively charged inner surface (63,64) that sequesters Fe(II) from the solution (Fig.  7A). The oligomer catalyzes oxidation of Fe(II) to Fe(III) and further promotes nucleation of a ferrihydrite crystallite at the negatively charged surface. If the Fe(II) concentration exceeds that of the ferroxidase sites on the protein (Ͼ0.5 Fe(II)/subunit), ferroxidation is rapidly overcome by a slower autoxidation reaction at the surface of the growing crystallite. At high Fe(II)/mYfh1p ratios, the initial ferrihydrite crystallite grows into a larger particle (Fig. 7B). Nichol et al. (23) have proposed that the iron core of frataxin forms via a process similar to bacterial biomineralization (65). According to this model, alignment and binding of one ferrihydrite particle to another leads to the interaction of trimers, which further facilitates biomineralization (Fig. 7B). During this process, frataxin acts as a chaperone, donating residual Fe(II) to ferrochelatase or ISUtype proteins to support heme and iron-sulfur cluster biosynthesis, respectively (Fig. 7B). Interestingly, the Fe(II) chaperone function appears to have been separated from the Fe(III) storage function in human frataxin. Ongoing studies in our laboratory show that the human frataxin monomer acts as a Fe(II) donor to ferrochelatase, consistent with a recent report that human frataxin monomer acts as an iron donor to ISUtype proteins in vitro (61). On the other hand, the assembled form of human frataxin (25) has ferroxidase activity 2 and is able to store Fe(III) in iron cores structurally identical to the yeast frataxin iron cores (Ref. 23).
Observations we have made previously in vivo support this model. When native frataxin is analyzed by gel filtration, the protein is detected as a distribution of species over a broad molecular mass range (from ϳ13 to Ͼ600 kDa) corresponding to monomer and progressively larger molecules (24,25). This suggests that stepwise assembly occurs in mitochondria and that monomer may be in equilibrium with higher order oligomers. Further support comes from structural studies. Scanning transmission electron microscopy data (22) and extended x-ray absorption fine structure analysis (23) indicate that yeast and human frataxin iron cores are composed of small ferrihydrite crystallites. The three-dimensional structures of human and bacterial frataxin show a highly conserved negatively charged surface similar to the anionic surface involved in the iron storage mechanism of ferritin (63,64,66). This surface could be involved in iron oxidation and nucleation, and facilitate biomineralization by keeping the growing ferric iron crystallites in a soluble form. Point mutations of carboxylate residues in this surface compromise stepwise assembly of bacterial frataxin (43) as well as the ferroxidase activity and stepwise assembly of yeast frataxin. 2 A second highly conserved uncharged surface predicted to be involved in protein-protein interactions (63,64) could mediate interactions between frataxin and iron users. Indeed, yeast frataxin and ferrochelatase were found to interact with each other by Biacore studies (17). This evidence and our work support the hypothesis that frataxin could work both as a chaperone for Fe(II) when mitochondrial iron is limiting, and as a storage compartment for Fe(III) when iron is in excess.