Architecture of the Human Mitochondrial Iron-Sulfur Cluster Assembly Machinery*

Fe-S clusters, essential cofactors needed for the activity of many different enzymes, are assembled by conserved protein machineries inside bacteria and mitochondria. As the architecture of the human machinery remains undefined, we co-expressed in Escherichia coli the following four proteins involved in the initial step of Fe-S cluster synthesis: FXN42–210 (iron donor); [NFS1]·[ISD11] (sulfur donor); and ISCU (scaffold upon which new clusters are assembled). We purified a stable, active complex consisting of all four proteins with 1:1:1:1 stoichiometry. Using negative staining transmission EM and single particle analysis, we obtained a three-dimensional model of the complex with ∼14 Å resolution. Molecular dynamics flexible fitting of protein structures docked into the EM map of the model revealed a [FXN42–210]24·[NFS1]24·[ISD11]24·[ISCU]24 complex, consistent with the measured 1:1:1:1 stoichiometry of its four components. The complex structure fulfills distance constraints obtained from chemical cross-linking of the complex at multiple recurring interfaces, involving hydrogen bonds, salt bridges, or hydrophobic interactions between conserved residues. The complex consists of a central roughly cubic [FXN42–210]24·[ISCU]24 sub-complex with one symmetric ISCU trimer bound on top of one symmetric FXN42–210 trimer at each of its eight vertices. Binding of 12 [NFS1]2·[ISD11]2 sub-complexes to the surface results in a globular macromolecule with a diameter of ∼15 nm and creates 24 Fe-S cluster assembly centers. The organization of each center recapitulates a previously proposed conserved mechanism for sulfur donation from NFS1 to ISCU and reveals, for the first time, a path for iron donation from FXN42–210 to ISCU.

ety of functions. In yeast and animal cells, mitochondria are a main site of Fe-S cluster assembly (1), which involves highly conserved protein systems derived from the main bacterial Fe-S cluster assembly system (2,3). All of these systems utilize IscUtype proteins as scaffolds upon which new Fe-S clusters are assembled and additional components, including the following: (i) pyridoxal phosphate-dependent cysteine desulfurases (bacterial IscS/yeast Nfs1/human NFS1), which in eukaryotes require a stabilizing binding partner (yeast Isd11/human ISD11) that provides elemental sulfur (4); (ii) members of the frataxin family (bacterial CyaY/yeast Yfh1/human FXN) that provide elemental iron (5)(6)(7)(8) and also regulate the cysteine desulfurase activity (9,10); and (iii) the electron donor chain formed by ferredoxin reductase and ferredoxin (11).
Biochemical studies have shown that Fe-S cluster assembly requires direct interactions among the proteins described above (6,7,(11)(12)(13). In the specific case of the human system, two different isoforms of frataxin, FXN  and FXN 81-210 , participate in Fe-S cluster assembly using different modes of interaction with [NFS1]⅐[ISD11] and ISCU as well as different iron binding capacities (14). These isoforms are generated through sequential cleavage of the FXN protein precursor (FXN  ) by the mitochondrial processing peptidase upon import of FXN  to the mitochondrial matrix (15)(16)(17). As determined by size-exclusion chromatography of normal human cell extracts, the native state of FXN  is a monomer-oligomer equilibrium, whereas the native state of FXN 81-210 is monomeric (7). Similarly in Escherichia coli, synthesis of recombinant FXN  yields monomeric and oligomeric species (7), whereas synthesis of recombinant FXN 81-210 yields monomer only (7,12). In vitro, FXN 81-210 binds to the pre-formed [NFS1]⅐[ISD11]⅐[ISCU] complex (7,12,18), whereas FXN  oligomer forms stable contacts with [NFS1]⅐[ISD11] both in the absence or presence of ISCU (7). All of these interactions are iron-independent (7,12,18). Complexes containing native oligomeric FXN  , NFS1, and ISCU could be isolated upon fractionation of normal human cell extracts by size-exclusion chromatography followed by coimmunoprecipitation (7,19), underscoring the stability of these complexes. In contrast, native FXN 81-210 was only recovered in fractions that contained no detectable NFS1 and only traces of ISCU (7,19 complex and FXN 81-210 monomer. However, in both instances a large excess of FXN 81-210 monomer was required for complex formation (13,18), which may explain why this complex could not be detected under native conditions in human cells (7,19).
In the context of the complexes described above, both FXN  and FXN 81-210 stimulate the cysteine desulfurase activity of [NFS1]⅐ [ISD11] in the presence of ISCU and L-cysteine in vitro (12,19). Likewise, both isoforms support Fe-S cluster synthesis in the presence of [NFS1]⅐[ISD11], ISCU, L-cysteine, and ferrous iron under anaerobic conditions that prevent iron oxidation (7). However, under aerobic conditions that promote oxidation of Fe 2ϩ to Fe 3ϩ , FXN 81-210 monomer binds Յ1 iron equivalents (7,13) and is unable to support Fe-S cluster synthesis (7). In contrast, under the same aerobic conditions, oligomeric FXN  chelates up to 10 molar eq of iron, and this ferric iron-loaded oligomer is able to support Fe-S cluster synthesis (7), similar to the oligomeric forms of E. coli and yeast frataxin (5,6). The differences described above led us to propose that, in physiological conditions, FXN 81-210 may bind low levels of iron and support basal levels of Fe-S cluster assembly via transient interactions with the [NFS1] 2 ⅐[ISD11] 4 ⅐ [ISCU] 2 complex, and that stable [FXN  ] n ⅐[NFS1] n ⅐ [ISD11] n ⅐[ISCU] n complexes may provide a mechanism to increase the rate of Fe-S cluster assembly when the demand exceeds the low iron-binding capacity of FXN   (14). In agreement with the view that the two FXN isoforms have complementary roles, different disease states result from mutations that alter the FXN 81-210 /FXN 42-210 molar ratio. FXN  is significantly more depleted than FXN 81-210 in patients with Friedreich ataxia (7), who have a progressive clinical course with death at a young age. Conversely, certain point mutations in MPP that cause FXN 81-210 depletion and FXN  accumulation are associated with a non-progressive form of cerebellar ataxia (17).
Based on structural analyses of complexes from bacteria (20)(21)(22) and yeast (8) 2 subcomplexes with one ISCU subunit bound at each end, and one FXN 81-210 subunit bound in a pocket between each NFS1 and ISCU1 pair, which would result in the presence of one Fe-S cluster assembly site at each end of the complex (13). The mechanism of sulfur transfer to each of these sites involves a flexible loop of NFS1 containing the invariant cysteine residue, Cys-381, which is persulfurated through the cysteine desulfurase reaction (10,12,20,21). Binding of FXN 81-210 to [NFS1] 2 ⅐[ISD11] 4 ⅐[ISCU] 2 accelerates persulfide formation on NFS1 and is also thought to induce a conformational change in ISCU that enhances the transfer of sulfur from NFS1 to Cys-138, the primary or main persulfide acceptor of ISCU (23)(24)(25) (note that Cys-138 corresponds to Cys-104 if the ISCU amino acid sequence is numbered starting from the mature protein N terminus instead of the precursor protein N terminus). Studies of the yeast system further suggest that sulfur-donation to ISCU may require conformational changes in NFS1 to (i) expose the NFS1 substrate-binding site, (ii) bring the flexible loop of NFS1 close to the substrate-binding site to enable formation of persulfurated Cys-381, and (iii) bring the NFS1 flexible loop close to the Fe-S cluster assembly site to enable sulfur transfer from persulfurated Cys-381 to ISCU (10,26).
While the studies summarized above suggest a conserved path for sulfur donation from NFS1 to ISCU, the path for iron-donation remains undefined. In the context of the [FXN 81-210 ] 2 ⅐[NFS1] 2 ⅐[ISD11] 4 ⅐[ISCU] 2 complex it was proposed that one monomer of FXN 81-210 could bind in a pocket between the desulfurase and the scaffold through the frataxin iron-binding surface (13). However, binding of FXN 81-210 to [NFS1] 2 ⅐[ISD11] 4 ⅐[ISCU] 2 blocks the ability of FXN 81-210 to bind iron, which has been attributed to the fact that the frataxin iron-binding surface forms direct contacts with a positively charged patch on NFS1 (13). Whereas binding of FXN 81-210 to [NFS1] 2 ⅐[ISD11] 4 ⅐[ISCU] 2 also blocks the entry of iron into this complex, the addition of L-cysteine and the activation of the cysteine desulfurase activity restore iron entry even in the presence of bound FXN 81-210 (13). The mechanisms underlying these effects remain unexplained and the path for iron entry into the [NFS1] 2 ⅐[ISD11] 4 ⅐[ISCU] 2 complex is still unknown.
FXN  , bacterial CyaY and yeast Yfh1 share a strong propensity to oligomerize in vitro (5,7,27,28) and in vivo (7,29,30). We have shown that unlike CyaY or Yfh1, FXN  oligomerizes in an iron-independent manner, which involves the N-terminal region of the protein upstream of Ser-81 (7). Although Yfh1 oligomerization is iron-dependent (27), certain point mutations enable Yfh1 to oligomerize in an iron-independent manner just like FXN   (29,31). Using one of these Yfh1 variants, Yfh1 Y73A , we were able to uncouple oligomerization of Yfh1 from iron binding and to study the complex formed by the apo-forms of Yfh1 and Isu1 (32). The structural model of the complex recapitulates the known mechanism for sulfurdonation from Nfs1 to Isu1 and also reveals a potential path for iron-donation from Yfh1 to Isu1 (32). The functional similarities between oligomeric Yfh1 Y73A and FXN  , and in particular their common ability to store iron (29,33) and to stimulate Fe-S cluster assembly (6,7), suggest that oligomeric FXN  may be suitable to gain insight into the path for iron-donation in the human system. We therefore utilized a bacterial four-protein co-expression system to mimic the mitochondrial environment, and purified a stable and active macromolecular complex formed by FXN  with ISCU, NFS1 and ISD11. The architecture of the complex suggests a coordinated mechanism for the transfer of iron and sulfur to ISCU for the synthesis of [2Fe-2S] clusters.   , NFS1, and ISCU in normal human cell extracts. In addition, in yeast cells expressing human frataxin precursor instead of yeast frataxin precursor, we detected complexes containing heterologous FXN  bound to endogenous Nfs1 and Isu1, which underscored the structural and functional conservation between the yeast and the human Fe-S cluster assem-bly systems (7). These previous data together with the high degree of conservation between prokaryotic and mitochondrial Fe-S cluster assembly systems (2,3,34) led us to hypothesize that simultaneous co-expression of FXN  , NFS1, ISD11 , and ISCU in E. coli might mimic the mitochondrial environment and yield similar complexes in sufficient quantities to enable structural studies. We therefore co-expressed the four proteins in E. coli as described under "Experimental Procedures." At the end of the protein expression phase, size-exclusion chromatography and SDS-PAGE analysis of the total bacterial cell extract revealed that similar levels of FXN  , ISCU, and ISD11 were co-eluted with NFS1, which was present in large excess relative to the other three proteins (data not shown). Accordingly, nickel affinity chromatography of the total bacterial cell extract, utilizing a His 6 tag fused with the N terminus of NFS1, yielded one soluble pool containing all four proteins and a second pool containing NFS1 alone that was prone to aggregation likely due to the absence of ISD11 (data not shown) (35). Size-exclusion chromatography of the soluble nickel affinity chromatography pool, followed by affinity chromatography utilizing a Strep⅐Tag II fused with the N terminus of FXN  , yielded a soluble protein pool that once again contained all four proteins (Fig. 1A). As analyzed by sizeexclusion chromatography and SDS-PAGE, the vast majority of the proteins in this pool were eluted together in one peak (Fig.  1B). To verify that the four proteins in the peak were bound to each other, the purified complex was incubated with DTSSP cross-linker, which can be cleaved by use of reducing agents. This sample was separated by size-exclusion chromatography, and the center of the peak containing the complex (fractions 56 -60) was analyzed by SDS-PAGE in the presence or absence of ␤-mercaptoethanol. In the presence of the reducing agent, we detected the four individual proteins (Fig. 1C). In the absence of the reducing agent, we detected a single band that migrated in the high molecular mass region of the gel, with only traces of the individual proteins in the 45-to 11-kDa region of the gel, consistent with the presence of an almost completely cross-linked [FXN   (Fig. 1D).

Isolation and Biochemical Characterization of Human
Purified complex was also incubated with a non-hydrolysable cross-linker, bis[sulfosuccinimidyl]suberate (BS 3 ), 3 and analyzed by size-exclusion chromatography. The chromatogram showed one peak with the same elution volume of the uncrosslinked or DTSSP cross-linked complex, and SDS-PAGE in the presence of ␤-mercaptoethanol revealed a single band that migrated in the high molecular mass region of the gel, as observed with DTSSP cross-linker (Fig. 1, E (6,7). To verify this further, purified complex was incubated for 1 h in the presence of 5 mM ␤-mercaptoethanol and subsequently purified by size-exclusion chromatography in TN150 buffer containing 5 mM ␤-mercaptoethanol. As analyzed by SDS-PAGE, reducing conditions during size-exclusion chromatography resolved high molecular weight species present in fractions 38 -50; however, the elution profile of the complex was indistinguishable regardless of the presence of the reducing agent (Fig. 1, B versus F). After another size-exclusion chromatograph of the complex in the absence of reducing agent, we analyzed fractions 56 -60 by reducing and non-reducing SDS-PAGE. In non-reducing conditions, we detected the four protein components of the complex and only faint bands in the region of the gel Ͼ94 kDa (Fig. 1G), where the whole complex migrates (see Fig. 1, D and E). In addition, there were three bands that were not present under reducing conditions (Fig. 1G). Mass spectrometry identification of proteins in each of these three bands (supplemental Table S1) indicated that band* and band** most likely represent faster migrating forms of ISCU and NFS1, respectively, resulting from intramolecular disulfide bonds that make each protein more compact. The data further indicated that band*** most likely represents a slower migrating form of FXN  , probably a dimer given the apparent molecular mass of ϳ46 kDa (Fig. 1G), which may result from an inter-molecular disulfide bond between FXN 42-210 subunits. It is unlikely that inter-molecular disulfide bonds are required to stabilize oligomeric FXN 42-210 as the protein contains only one Cys residue, Cys-50, which is outside of the N-terminal region necessary and sufficient for oligomerization (residues 56 -78) (36). Therefore, based on all of these results together, we concluded that disulfide bonds do not play a significant role in the formation or stability of the [FXN Protein staining and densitometry of SDS-polyacrylamide gels suggested the complex might consist of approximately equimolar amounts of FXN  , ISCU, and ISD11 and an apparently 2-3-fold excess of NFS1 (Fig. 1, A and B). The fact that NFS1 appeared to be present in excess of ISD11 was in contrast with previous observations that the yeast orthologues, Nfs1 and Isd11, form a complex that contains stoichiometric amounts of each protein in a 1:1 ratio (37,38). Given that NFS1 is significantly larger than ISD11 (ϳ45 kDa versus ϳ11 kDa), we considered the possibility that equimolar levels of NFS1 might correspond to ϳ4 times higher local protein concentrations inside the gel, resulting in enhanced stainability of NFS1 compared with ISD11 (39). To avoid this issue, we determined the stoichiometry of the four proteins in the complex via amino acid analysis of fractions 58 and 59. We compared different predicted amino acid compositions of the complex to the experimentally measured composition, and we then calculated the variation for each amino acid and the total % variation ( Table 1). Different predicted amino acid compositions of the complex were based on the assumption that the complex might be made up of FXN  , ISCU, NFS1, and ISD11 in a 1:1:1:1 or 1:1:2:1 or 1:1:1:2 or 2:1:1:1 or 1:2:1:1 molar ratio. Independent amino acid analysis of fractions 58 and 59 yielded almost identical experimental compositions, the average of which was used in the calculations. The total % variation for the 1:1:1:1, 1:1:2:1, 3 The abbreviations used are: BS 3  and ISCU in an iron-independent manner (7).
Finally, to assess the suitability of using chemical cross-linking to probe the complex structure, we assessed the ability of the uncross-linked and BS 3 cross-linked complex to catalyze Fe-S cluster formation in the presence of L-cysteine and Fe 2ϩ after both the streptavidin affinity chromatography and the FIGURE 1. Co-expression of human FXN 42-210 , NFS1, ISD11, and ISCU proteins in E. coli yields a stable and active four-protein complex. A, histidinetagged NFS1 was co-expressed in E. coli with ISD11, ISCU, and streptavidin-tagged FXN  , and a complex containing all four proteins was purified as described under "Experimental Procedures." B, Sephacryl S-300 size-exclusion chromatography and SDS-PAGE analysis of purified complex. C-E, complex eluted from the StrepTrap TM HP affinity column was treated with DTSSP (C and D) or BS 3 (E) cross-linker and purified by size-exclusion chromatography, after which fractions 56 -60 were analyzed by SDS-PAGE in the presence (C and E) or absence (D) of the reducing agent ␤-mercaptoethanol (␤-ME). F, complex eluted from the StrepTrap TM HP affinity column was incubated for 1 h in the presence of 5 mM ␤-mercaptoethanol, purified by size-exclusion chromatography in TN150 buffer containing 5 mM ␤-mercaptoethanol, and analyzed by SDS-PAGE. G, after size-exclusion chromatography in the absence of reducing agent (B), fractions 56 -60 were analyzed by SDS-PAGE in the absence or presence of ␤-mercaptoethanol. Asterisks denote protein bands that are only observed under non-reducing conditions (see supplemental Table S1 for details). H, indicated complex preparations were tested for their ability to catalyze Fe-S cluster assembly; Complex 1 and Complex 2 denote complex after the third and the last purification steps. Assays were performed anaerobically in the presence of 5 M complex, 1 mM L-cysteine, 50 M Fe 2ϩ as described under "Experimental Procedures." Each of the plots shows the mean Ϯ S.D. of three independent measurements with two different complex preparations. I, dynamic light scattering measurements were performed on BS 3 -cross-linked complex freshly eluted from the size-exclusion chromatography column, and the hydrodynamic radius (R h ) was obtained as described under "Experimental Procedures." J, elution volumes of the human complex (Complex) and the indicated protein complexes analyzed by Sephacryl S-300 size-exclusion chromatography. Asterisk denotes ferritin dimer. Although the predicted molecular mass of the human complex (ϳ2 MDa) is greater than that of ferritin (700 -800 kDa due to iron content), these two complexes are eluted according to their similar dimensions (40). The fact that the human complex elutes slightly later than ferritin may also be due to retardation through weak adsorption to the gel. size-exclusion chromatography purification steps, and we detected very similar activity in both samples (Fig. 1H). Complex cross-linked with BS 3 was also able to catalyze Fe-S cluster formation with slower kinetics but a similar final yield as compared with the uncross-linked complex (Fig. 1H). Chemical cross-linking data later indicated that the slower kinetics of the cross-linked complex probably reflected cross-linking of flexible structural elements involved in Fe-S cluster assembly (see Fig. 14F).
Dynamic light scattering of pooled fractions 56 -60 revealed a hydrodynamic radius (R h ) of 7.4 Ϯ 0.3 nm with unimodal distribution (Fig. 1I). We measured and compared the R h and elution profiles of the [FXN    (32), and the horse spleen ferritin 24-mer (R h ϭ 7.5 Ϯ 0.2). The four complexes were eluted with similar retention volumes between ϳ66 and 74 ml (measured between the ferritin and the human complex peaks in Fig. 1J), a result consistent with their similar R h (40) and with the size-exclusion pore dimension of the Sephacryl S-300 gel matrix (ϳ13 nm) (41).
EM  Fig. 2A and not shown), consistent with the R h measured by dynamic light scattering and the behavior of the complex in size-exclusion chromatography. These data together suggested we were dealing with an overall homogeneous population of roughly globular particles. We took EM images of the uranyl acetate-stained particles at a magnification of 115,000 with an Å/pixel ratio of 1.034. We collected 4124 particles and used the ϳ786 best particles to generate reference free class averages using the EMAN2 software package ( Fig. 2A) (42). Although all class averages exhibited similar globular shape and diameter of ϳ15 nm, each class exhibited unique structural details ( Fig. 2A), suggesting the complex adsorbed to the carbon film of the EM grids with different orientations. Given the high degree of conservation between the yeast and the human systems (7,15), and based on the measured 1:1:1:1 stoichiometry of the complex, we hypothesized that the architecture of the human four-protein complex might be similar to the recently reported architecture of the [Yfh1] 24 24 complex, which has 432 symmetry (PDB code 5T0V) (32). Therefore, initial 3D models were generated from the class averages without or with 432 symmetry applied ( Fig. 2A). Refinement of these models using the larger set of ϳ4124 particles yielded models of similar globular shape (Fig. 2, B-D and E-G). Then, structure factors were calculated from the EM density map of the refined 3D model without symmetry applied, and a self-rotation function was obtained (43). Similar calculations were completed using the EM density map of the refined 3D model with 432 symmetry applied. These data revealed the presence of 432 symmetry in both models (Fig. 2, H versus I). Using the 0.143 cutoff (44), the FSC plots indicated resolutions of 15.0 and 13.5 Å, respectively, for the model without and with symmetry applied (Fig. 2, J and K). Resolutions

TABLE 1
Amino acid analysis of human four-protein complex The experimental amino acid composition of purified complex, as determined via amino acid analysis, is compared with three expected compositions based on the three potential FXN  /ISCU/NFS1/ISD11 molar ratios shown. For each molar ratio, the variation between the experimental and expected number of each amino acid and the total % variation is shown. Sixteen amino acids were quantified; tryptophan and cysteine are degraded by the hydrolysis process, and asparagine and glutamine are detected as aspartic and glutamic acid, respectively (75).

Architecture of the Human Fe-S Cluster Assembly Machinery
obtained with the 3-criterion (13.7 and 15.5 Å) and the 1/2 bit criterion (16.0 and 14.9 Å) indicated that at these resolutions we had collected information significantly above the noise level and sufficient for interpretation (45). We segmented the refined 3D models without and with 432 symmetry applied using Chimera (46), and we identified volumes with similar shapes around the 3-and 4-fold symmetry axes (Fig. 2, C, D (32). Based on this, FXN  and ISCU trimers were modeled as described under "Experimental Procedures." Part of the extended N-terminal regions of FXN  and ISCU (residues 42-93 and 36 -50) were predicted to have a high degree of flexibility, and therefore N-terminally truncated trimers were created. We docked them into the segmented EM density map of the refined 3D model with 432 symmetry applied both separately (cross-correlation functions ϭ 0.58 and 0.65, respectively) or together as a [FXN  ] 3 ⅐[ISCU] 3 sub-complex (cross-correlation function ϭ 0.62), and we selected the latter mode of docking. Subsequently, nearby unoccupied densities were filled by modeling the N termini of FXN  and ISCU using information obtained from chemical cross-linking (described in detail later). Nearby remaining segmented density of the 3D model revealed unoccupied volumes with similar shape to the homology model of the NFS1 monomer, which was docked into 24 similar volumes, with a cross-correlation coefficient of 0.57. After molecular dynamics simulations were completed (see below), we attempted docking of the simulated structures of the [FXN  ] 3 ⅐[ISCU] 3 sub-complex and NFS1 monomer into the segmented EM density map of the refined 3D model without symmetry applied. We were able to fit eight [FXN  ] 3 ⅐[ISCU] 3 sub-complexes (cross-correlation functions ranged from 0.37 to 0.54) and 24 NFS1 monomers (crosscorrelation functions ranged from 0.41 to 0.56) (supplemental Fig. S1).
After fitting of FXN  , ISCU, and NFS1 into the EM density map, cross-linking data and the presence of some unoccupied volumes just above the NFS1 subunits lead us to conclude that the ISD11 protein might be located on the surface of the complex. However, the unoccupied volumes at the surface of the refined 3D model were not sufficiently delineated from the background to enable docking of ISD11 monomers into the segmented EM map. Therefore, based on the measured NFS1: ISD11 stoichiometry of 1:1 and the cross-linking data, we modeled 24 ISD11 monomers on the surface of the 3D model with 432 symmetry applied but did not include ISD11 in the simulations described below.
Molecular Dynamics Flexible Fitting for Docked Structures-Attempts to simulate the entire docked structure of the complex (hereafter designated [FXN   24 ) were unsuccessful due to the very large number of FIGURE 2. Transmission EM and single particle analysis of human fourprotein complex. A, electron micrographs of purified uranyl acetate-stained complex particles were obtained, and images were processed with the EMAN2 software package. Shown is a gallery of class averages, with one representative particle from each class and the corresponding projection of the initial 3D reconstructions without symmetry applied (Projection 1 columns) and with 432 symmetry applied (Projection 2 columns). Particles, class averages, and projections representing the 2-, 3-, and 4-fold axis of the complex and intermediate orientations are shown sequentially from the left to the right starting with the top row. The particle diameter was 15.2 Ϯ 0.8 nm (average of 103 particles with 2-fold orientations), 15.1 Ϯ 0.7 nm (average of 123 particles with 3-fold orientations), and 15.4 Ϯ 0.7 nm (average of 91 particles with 4-fold orientations). B-G, refined 3D models were generated without imposed symmetry (B) or with 432 symmetry applied (E). Both models were segmented using Chimera, and the 4-fold (C and F) and 3-fold (D and G) axes were identified and colored in green and blue, respectively. H and I, stereographic projection plots of the kappa ϭ 90°(4-fold), kappa ϭ 120°(3-fold), and kappa ϭ 180°(2-fold) sections of the self-rotation function of the EM density map of the refined 3D model of the complex without symmetry applied (H) and with 432 symmetry applied (I), obtained using POLARRFN. J and K, PDBe Fourier shell correlation (FSC) server was used to calculate and plot the FSC curve for the refined 3D model without (J) and with 432 symmetry applied (K). The dashed red line shows where the FSC curve crosses the correlation value of 0.143. atoms involved (Ͼ100,000). However, we were able to subject one-half of the structure, consisting of six NFS1 dimers docked on top of four [FXN  ] 3 ⅐[ISCU] 3 sub-complexes, to mild simulation using Molecular Dynamics Flexible Fitting (47,48) as described under "Experimental Procedures." Prior to the simulation, residues 42-98 in the N-terminal region of FXN  were repositioned guided by cross-linking data. In addition, the N-terminal flexible portions of ISCU (residues [35][36][37][38][39][40][41][42][43][44][45][46][47][48][49] and NFS1 (residues 55-66) were removed to avoid steric clashes during the simulation. The simulation improved the fitting of the structure into the EM density map, as the cross-correlation coefficient improved from 0.46 to 0.62. After the simulation, the fold of the FXN 42-210 monomer (Fig. 3A) and the overall arrangement of monomers in the FXN 42-210 trimer ( Fig. 3B) remained very similar to those in the docked structure, except for the N-terminal region (residues 42-93) that had been manually modeled into a different configuration based on cross-linking data (Fig. 3, A and B). The fold of the ISCU monomer (Fig. 3C) and the overall arrangement of monomers in the ISCU trimer ( Fig. 3D) also matched the initial docked structure closely except for the absence of N-terminal residues 35-49 and for a small movement of loops L2, L3, and helix ␣5 (Fig. 3, C and D). The fold of the NFS1 monomer ( Fig.  3E) and the overall arrangement of the NFS1 dimer ( Fig. 3F) remained very similar to those in the docked structure except for the following: (i) the absence of N-terminal residues 55-66; (ii) the C-terminal region that had been manually modeled into a different configuration based on cross-linking data; and (iii) a small movement of loop L2 and helices ␣1, ␣2, ␣8, and ␣13 (Fig.  3, E and F). As mentioned above, the modeled 24 ISD11 monomers ( Fig. 3G) were not included in the simulation.
Overall Architecture of the Human Fe-S Cluster Assembly Complex-For the purpose of visualizing the entire complex structure, the simulated one-half was aligned with itself into the EM density map of the refined 3D model, followed by structure improvement iterations with the program Coot to resolve small steric clashes between NFS1 subunits at the interface between the two halves of the structure ( Cross-linking Analysis-We used chemical cross-linking and limited proteolysis of the complex, followed by MS/MS identification of cross-linked peptides as an independent means to  3 sub-complexes and 24 NFS1 monomers into the EM density map of the refined 3D model with 432 symmetry applied, one-half of the docked structure was subjected to molecular dynamics simulations and energy minimizations as described under "Experimental Procedures." FXN 42-210 monomer (A) and trimer (B) before (magenta ribbon; N, N terminus) and after (salmon ribbon; NЈ, N terminus); ISCU monomer (C) and trimer (D) before (orange ribbon; N, N terminus) and after (yellow ribbon; NЈ, N terminus); and NFS1 monomer (E) and dimer (F) before (blue ribbon; N, N terminus) and after (light blue ribbon; NЈ, N terminus) molecular dynamics simulations and energy minimizations. G, model of ISD11 monomer (purple ribbon; N, N terminus) obtained using the I-TASSER web resource, which was not included in the simulated structure as described under "Results." obtain information about protein-protein interactions in the [FXN   (49). Because the cross-linked complex was enzymatically active (Fig. 1H), it was suitable to obtain structural information. A summary of the results is presented in Table 2 and a detailed description in supplemental Table S2. Of the 124 cross-linked peptides identified, 43% had false discovery rate (FDR) Յ5% and 90% had FDR Յ25% ( Table 2). The identified cross-links involved 13/13 Lys residues of FXN   (Fig. 6A), 19/20 Lys residues of ISCU (Fig. 6E), 21/21 Lys residues of NFS1 (Fig. 7A), and 6/6 Lys residues of ISD11 (Fig. 7F). In addition, the identified cross-links involved the N-terminal amino groups and several Ser, Thr, and Tyr residues of FXN  , ISCU, NFS1, or ISD11. Based on the general guidelines for this type of analysis (49), the large number of identified cross-links indicated a tight-binding four-protein complex.
Next, we assessed the degree of agreement between the structure of the complex and the cross-linking data by measuring, in the simulated half-structure of the complex, the distances between all possible pairs of cross-linked residues within any given cross-linked peptide (supplemental Table S2), by identifying mean distances (Ϯ S.D.) equal to or lower than the distance constraints or the maximum allowable distance constraints, or greater than the maximum allowable distance constraints (highlighted in light gray, dark gray, and yellow, respectively, in supplemental Table S2). Agreement between the simulated structure and any given cross-linked peptide was established if at least one of the mean distances measured in the simulated half-structure was within the distance constraints or the maximum allowable distance constraints (Table 2 and supplemental Table S2). When using the distance constraints, our simulated structure is in agreement with 80/124 or ϳ65% of the cross-linked peptides identified (Table 2). When using the maximum allowable distance constraints, the agreement increases to ϳ90% (Table 2). There is apparent disagreement between the simulated half-structure of the complex and 13/124 or ϳ10% of the cross-linked peptides identified ( Table  2). These 13 peptides (denoted by asterisks in supplemental Table S2) involve flexible regions of the four proteins. Distances measured in the structure between cross-linked residues from 10 of the 13 peptides are only ϳ3-9 Å above the maximum allowable distance constraints, a difference that could be explained by the flexibility of the cross-linked regions.
At the 3-fold axis of the complex, which is formed by one [FXN    (Fig. 8B), the crosslinks support the same protein-protein interaction surfaces described above as well as protein-protein interactions, not present in the 2-fold axis, between each of the three peripheral  3 subcomplexes, the cross-links once again support the same protein-protein interaction surfaces described above (Figs. 8C  and 9, A-E and G).

JOURNAL OF BIOLOGICAL CHEMISTRY 21307
unique interfaces with buried surface area (BSA) of Ͼ300 Å 2 (Figs. 10 -12), and four additional minor interfaces with BSA of Յ150 Å 2 (data not shown). FXN 42-210 -ISCU contacts include three main interfaces with BSA of ϳ1646 Å 2 (average of six recurring interfaces), 682 Å 2 (average of six recurring interfaces), and 331 Å 2 (unique interface). About 30 residues of ISCU and 30 residues of FXN  are present in the largest interface of ϳ1646 Å 2 within the [FXN  ]⅐[ISCU] heterodimer, 23 of which are invariant in eukaryotes and 20 are invariant in mammals ( Fig. 10A; supplemental Table S3). The FXN 42-210 N terminus and several residues from the N terminus as well as the ␤-sheet of ISCU together contribute to this interface. Residues in the C terminus of FXN  and two flexible ␣-helices, ␣2 and ␣3, of ISCU also contribute to this interface (Fig. 10A). All residues shown in Fig.  10A are predicted to form hydrogen bonds between the two proteins. In addition, several conserved residues are involved in salt bridges between ISCU and FXN 42-210 (supplemental Table  S3). Charge analysis of this interface in the context of the entire [FXN  ] 3 ⅐[ISCU] 3 sub-complex shows that the ISCU trimer surface is mostly positively charged, whereas the FXN 42-210 trimer surface is mostly negatively charged, suggesting that electrostatic interactions may contribute to formation and stability of the sub-complex (Fig. 10, B-D and C-E). Another interface is present between one FXN 42-210 subunit and one adjacent ISCU subunit within the same [FXN   Fig. 11A; supplemental Table S3). The FXN 42-210 N-terminal flexible loop L3 and several residues from C-terminal helices ␣4 and ␣5 of ISCU contribute to this interface. Several conserved residues are predicted to form salt bridges between ISCU and FXN    Table S3). In addition, residues in and around the PVK motif (shown as a red ribbon) of one ISCU subunit are involved in hydrophobic interactions with one FXN 42-210 subunit from the adjacent [FXN  ] 3 ⅐[ISCU] 3 sub-complex (Fig. 11G).
There are two ISCU-ISCU interfaces with BSA of ϳ412 Å 2 (average of six recurring interfaces) and 298 Å 2 (unique interface), respectively. The first is formed by three ISCU subunits of the same trimer around the 3-fold axis, involving hydrogen bonds between the highly conserved Glu-81 and Gly-83 residues, and salt bridges between residues in the C terminus of each subunit, which are invariant in mammals (Fig. 11, C and H; supplemental Table S3). The second ISCU-ISCU interface is formed between two ISCU subunits of adjacent ISCU trimers via interactions between residues Cys-69 -Cys-130, where Cys-69 is part of the highly conserved Fe-S cluster coordination site of ISCU (Fig. 11D). Additionally, this interface involves hydrophobic residues in and around the PVK motifs of two ISCU subunits from adjacent [FXN  ] 3 ⅐[ISCU] 3 sub-complexes (Fig. 11I).
as well as between loops L7 of one subunit and loops L7 and L9 of another subunit ( Fig. 11E; supplemental Table S3). Interestingly, this interface includes a salt bridge between Glu-100 and Arg-165 (Fig. 11E), which is involved in a pathological human mutation, R165C, found in Friedreich ataxia patients (52). The second FXN 42-210 -FXN  interface is formed by two adjacent trimers of FXN  , between loop L3 and helix ␣3 of one FXN 42-210 subunit in one trimer, and loop L6 of another FXN  subunit in the adjacent trimer ( Fig. 11F; supplemental Table S3). The pair Glu-101-Lys-135, which is invariant in eukaryotes, is predicted to be involved in a salt bridge formation between two adjacent trimers of FXN   (Fig. 11F).
The NFS1-NFS1 dimer interface with BSA of ϳ1600 Å 2 (unique interface, sum of the two symmetrical interfaces shown in Fig. 11J) is formed by residues conserved among eukaryotes.  Four residues in this interface are predicted to form hydrogen bonds between the C-terminal part of one NFS1 subunit, and loops L18 and L19 of the other NFS1 subunit (Fig. 11J; supplemental Table S3). Interestingly, other conserved residues involved in this interface are hydrophobic or partially hydrophobic (supplemental Table S3). Hydrophobic interactions between the two NFS1 subunits (Fig. 11, J and K) create a hydrophobic pocket around the two Fe-S cluster assembly sites of ISCU (Figs. 5, B and D, and 11J).
Each of the two NFS1 subunits in the [FXN  ] 3 ⅐ [ISCU] 3 ⅐[NFS1] 2 sub-complex interacts with two ISCU subunits from the two adjacent ISCU trimers via two distinct interfaces with BSA of ϳ1285 and 315 Å 2 (average of two recurring interfaces each). The first interface involves one NFS1 subunit and one ISCU subunit in the ISCU trimer immediately underneath (Fig. 12A). The flexible N-terminal region as well as helix ␣9, sheet ␤10, and helix ␣13 of NFS1 interact with strand ␤3, and helix ␣1 of ISCU ( Fig. 12A; supplemental Table S3). This interface is predicted to involve salt bridges between NFS1 and ISCU (Fig. 12, A and C; supplemental Table S3). However, some of the residues involved in this interface are hydrophobic. Residues from strands ␤1-␤3 of ISCU and residues from helices ␣8 and ␣13 and loops L15 and L23 of NFS1 may be involved in hydrophobic interactions between the two proteins ( Fig. 12B; supplemental Table S3). Interestingly, residues Leu-63, Val-72, and Phe-94 of yeast Isu1 (corresponding to the invariant Leu-63, Val-72, and Phe-93 residues of ISCU) were previously shown to be important for Isu1-Nfs1 interactions in yeast (53). The second interface is formed between the active site of one NFS1 subunit, including Cys-381, created by loops L18 and L19, and the active site of one ISCU subunit from the neighboring ISCU trimer, including Cys-69 and His-137, created by loops L2, L4, and L7 (Fig. 12D). This interface is predicted to involve mainly hydrogen bonding and one salt bridge between the invariant Glu-387 and Lys-91 of NFS1 and ISCU (Fig. 12E).
There are two FXN 42-210 -NFS1 interfaces with BSA of ϳ1895 and 926 Å 2 (average of two recurring interfaces each), involving the same NFS1 subunit and two different FXN  subunits within the same trimer immediately underneath. The first interface involves a flexible N-terminal region as well as loops L13, L14, L18, and L23 of NFS1, and loop L3, helix ␣3, loop L4, and helix ␣4 of FXN   (Fig. 12F; supplemental Table S3). Interestingly, Cys-381 from the active center of NFS1 is located close to negatively charged residues from helix ␣3 of FXN   (Fig. 12G), which are predicted to be involved in iron binding (see below). Most of the residues from this interface are invariant in eukaryotes and are predicted to form hydrogen bonds and several salt bridges ( Fig. 12G; supplemental Table  S3). The second FXN 42-210 -NFS1 interface involves invariant residues from NFS1 N-terminal region, loops L12 and L13, and helices ␣8 and ␣13, and non-conserved N-terminal residues of FXN   (Fig. 12H), some of which form hydrogen bonds and salt bridges (Fig. 12I; supplemental Table S3).  SEPTEMBER 30, 2016 • VOLUME 291 • NUMBER 40

JOURNAL OF BIOLOGICAL CHEMISTRY 21311
Surface and charge analysis shows that two adjacent [FXN] 3 ⅐[ISCU] 3 sub-complexes form a groove with one negatively charged patch and one positively charged patch on each side (Fig. 12K, yellow and green asterisks). The corresponding interface formed by the two NFS1 subunits has two patches of opposite charges (Fig. 12L, yellow and green asterisks), suggesting that the electrostatic interactions may contribute to complex formation (Fig. 12, J-M 3 sub-complexes, consisting of two adjacent ISCU trimers bound on top of two adjacent FXN 42-210 trimers, are present at each of the 2-fold axes of the complex (Figs. 5 and 13). Two ISCU subunits, one from each of the two ISCU trimers (designated trimers 1 and 2), face each other with their respective [2Fe-2S] coordinating sites formed by the invariant Cys-69, Cys-95, His-137, and Cys-138 residues (Figs. 5A and 13, B and  C). These residues are involved in Fe-S cluster coordination in prokaryotic IscU scaffolds (21,54,55). In each site, these residues are within ϳ2-4 Å from each other, suitable to coordinate one [2Fe-2S] cluster (21). The two [2Fe-2S] cluster coordinating sites are above the interface between the two FXN 42-210 trimers, which is formed by two adjacent subunits from each of the two trimers (designated trimers 1 and 2). Alignment with two available structures of iron-bound Yfh1 Y73A trimer (PDB codes 2FQL and 4EC2) (31,56) identified potential iron-binding sites that may be implicated in iron transfer to each of the two [2Fe-2S] cluster coordinating sites (Fig. 13, B and C). The first iron-binding site is in the channel at the 3-fold axis of each FXN 42-210 trimer, formed by three Asp-167 residues (Asp-143 in Yfh1), one from each subunit of the trimer (e.g. trimer 1 in Fig. 13C). From this site, iron could move to a second site formed by residues Thr-142 and Ser-157 (Thr-118 and Ala-133 in Yfh1) from one subunit of the opposite FXN 42-210 trimer (trimer 2 in Fig. 13C). In addition, alignment with the [Yfh1] 24 ⅐[Isu1] 24 complex (32) suggests the presence of a third iron-binding site formed by residues Glu-96, Glu-100, Glu-101, and Asp-104 (His-74, Asp-78, Asp-79, and Asp-82 in Yfh1) of trimer 1, together with residues Asp-124, Ser-126, and Thr-133 (Asp-101, Glu-103, and Thr-110 in Yfh1) of trimer 2 (Fig. 13C). This site is reminiscent of the ferroxidation site of E. coli ferritin (Fig. 13D) (57); and interestingly, in Yfh1, residues His-74, Asp-78, Asp-79, Asp-82, and His-83 compose the ferroxidation site, and residues Asp-101 and Glu-103 are part of the mineralization site of the protein (28,58). Alignment of cobalt-bound CyaY monomer (PDB code 2EFF) identified a fourth potential iron-binding site formed by residue Asp-104 (Asp-82 in Yfh1) from trimer 1 and Asp-124 (Asp-101 in Yfh1) from trimer 2. Four of these sites are present at the 4-fold axis of the complex (Fig. 13E) and may constitute the mineralization site of the protein, as described for the Yfh1 24-mer (28,58).
The PVK motif of Isu1 is important for Yfh1-Isu1 interactions in yeast (50). The PVK motif of ISCU is close to the Fe-S cluster assembly site and may interact with both FXN  subunits immediately underneath (i.e. in the same [FXN    G and I, and 13B). Moreover, PISA analysis predicts hydrogen bonding between the two opposite ISCU subunits involving Cys-69 -Cys-130 (Fig. 11D), which may contribute to the stability of the cluster assembly site.
Docking of 24 NFS1 monomers into the EM density map of the refined 3D model results in the presence of one symmetric NFS1 dimer resting on top of two [FXN  ] 3 ⅐[ISCU] 3 subcomplexes at each of the 2-fold axes of the complex (Fig. 5, B  and D). The invariant catalytic Cys-381 of NFS1 is on a flexible loop at 6.4 Ϯ 1.8 Å from Cys-69 and 10.3 Ϯ 1.4 Å from Cys-138 of ISCU (average of eight distances measured between the backbone ␣-carbons of Cys-381 and Cys-69 or C-381 and Cys-138 in the simulated half-structure of the complex) (Figs. 5D  and 13, A-C). Cys-69 and Cys-138 are proposed acceptors of the persulfide formed on Cys-381 of NFS1 (21,23,25). The fold of the NFS1 and ISCU monomers is very similar to the fold of Archaeoglobus fulgidus and E. coli IscS and IscU monomers (Fig. 14, A-C) and IscS overlap each other, whereas ISCU is rotated ϳ90°relative to IscU on the same plane, which brings the Fe-S cluster coordination site of ISCU closer to the NFS1 flexible loop (Fig.  14, D and E). This configuration is supported by cross-links between NFS1 and ISCU as well as NFS1 and FXN   (Fig.  14F).

Discussion
Although FXN 81-210 and [NFS1]⅐[ISD11] have been shown to work together to promote the assembly of new Fe-S clusters on the scaffold ISCU (12,13), the architecture of the complex formed by these proteins is not yet defined, and an integrated mechanism for the concerted delivery of iron and sulfur to ISCU is not yet known. To gain new information, we focused on FXN 42-210 , the longest known isoform of human frataxin, which forms oligomers that can participate in Fe-S cluster assembly using a more stable mode of interaction with [NFS1]⅐[ISD11] and ISCU as well as a higher iron binding capacity as compared with monomeric FXN 81-210 (14). We reconstituted in E. coli and purified a stable and functional macromolecular complex consisting of stoichiometric amounts of FXN 42-210 , NFS1, ISD11, and ISCU. Then, via negative staining EM single-particle analysis, we obtained a 3D model of the complex at a resolution of ϳ14 Å. Segmentation of the EM density map of the 3D model, followed by sequential docking of [FXN    24 complex. This structure fits into the EM density map of the 3D model without steric clashes and in a manner consistent with the measured 1:1:1:1 stoichiometry of the four protein components of the complex. The structure also fulfills distance constraints obtained from chemical cross-linking analysis at all main protein-protein interfaces present at the 2-, 3-, and 4-fold axes of the complex. These interfaces are compatible with the formation of hydrogen bonds, salt bridges, and hydrophobic interactions, often involving conserved residues. Finally, the structure of the complex is supported by the presence of 24 Fe-S cluster assembly centers, where the arrangement of cluster coordinating residues of ISCU relative to the catalytic flexible loop of NFS1 and the iron-binding residues of FXN 42-210 not only recapitulate the known mechanism for sulfur donation from NFS1 to ISCU but also provide a path for iron donation from FXN  to ISCU.
At the center of the structure, eight symmetric FXN 42-210 trimers form a cubic 24-mer (Fig. 4D). The frataxin structure can be adapted to oligomerize as evidenced by the ability of bacterial CyaY to form tetramers and of yeast Yfh1 to form trimers at low iron to protein ratios (Յ2:1) (28,59), and by the ability of both CyaY and Yfh1 to form higher order oligomers at increasing iron to protein ratios (5,27,56). Unlike CyaY or Yfh1, FXN  oligomerizes in an iron-independent manner in human, yeast, and bacterial cells (7), which is also an intrinsic property of the frataxin structure because certain point mutations result in iron-independent oligomerization of Yfh1 both in yeast and bacterial cells (29).
One symmetric ISCU trimer binds on top of each of the eight trimers that form the FXN 42-210 24-mer. IscU-type scaffolds have been shown to exist in monomeric or oligomeric states in the absence of protein partners (6,54,55,60,61). However, binding of a distribution of Isu1 monomer, dimer, and trimer to Yfh1 Y73A 24-mer resulted in the assembly of a complex in which one symmetric Isu1 trimer was bound on top of each of the eight trimers present in the Yfh1 Y73A 24-mer, in a manner that created 24 [2Fe-2S] cluster assembly sites (32). Similarly, conformational changes leading to stabilization of bacterial IscU were observed upon binding of IscU to the co-chaperone HscB (62). These previous data and our present results together indicate that binding of ISCU to its protein partners may result in new conformations required for ISCU activity in Fe-S cluster assembly.
Two adjacent [FXN  ] 3 ⅐[ISCU] 3 sub-complexes are present at each of the 2-fold axes of the complex, such that two ISCU subunits, one from each sub-complex, face each other with their respective [2Fe-2S] coordinating sites (Figs. 5, A and D, and 13A). Each of these sites is close to iron-binding sites formed by the two adjacent FXN 42-210 trimers (trimer 1 and trimer 2) underneath the two ISCU trimers (Fig. 13, B and C). Biochemical and structural studies of bacterial, yeast, and mammalian Fe-S cluster assembly complexes have led to a conserved mechanism for sulfur donation from the cysteine desulfurase to the scaffold (3,13,22,23,26). This mechanism predicts that the flexible loop of NFS1 must first move close to the L-cysteine substrate-binding site to enable formation of persulfurated Cys-381 (10,20,21,23), and then move close to the Fe-S cluster assembly site to enable sulfur transfer from persulfurated Cys-381 to ISCU (10,26). It was also shown that binding of the shorter frataxin isoform, FXN     Table  S2f, p. 1), ISCU-NFS1 cross-links K147-S385, K135-S389, and K135-Y390 (supplemental Table S2g, pp. 1 and 3), and NFS1-NFS1 cross-links K180 -S377 and K180 -S379 (supplemental Table S2c, p. 1). Cross-links are shown as dotted lines; cross-linked residues are highlighted in red.
[NFS1] 2 sub-complex in our model is consistent with a similar although simpler mechanism. Cys-381 of NFS1 is at ϳ11 Å from the NFS1 substrate-binding site and at ϳ9 Å from Cys-138 of ISCU. This suggests that the flexible loop of NFS1 would have to move toward the substrate-binding site to allow formation of the persulfurated Cys-381; however, only small conformational changes in ISCU would be required to enable sulfur transfer from Cys-381 to Cys-138 (Figs. 5D and 14, D and E).
Our structure may also help to explain how human frataxin may serve as both the activator of NFS1 and the iron donor, which remains unclear (13,23 (8,13). However, it has been suggested that this mode of frataxin binding could interfere with sulfur transfer (26) as well as iron entry (13). In contrast, the positioning of ISCU relative to FXN  and NFS1 in our structure allows for FXN  and NFS1 to simultaneously bind to ISCU and to stimulate Fe-S cluster synthesis (Fig. 1, A-H   changes the orientation of ISCU and thus the position of the cluster coordinating site relative to the catalytic loop of NFS1 (Fig. 14, D and E)   may also induce conformational changes that make the substrate-binding site of NFS1 more accessible to L-cysteine, thereby activating the cysteine desulfurase activity (10). Modeling of previously identified iron-or cobalt-binding sites of Yfh1 or CyaY into our structure further suggests a mechanism for iron transfer from FXN  to ISCU, which is fully contained within one [[FXN    (Fig. 13C). FXN 42-210 trimer 1 residues, Tyr-175 and Glu-96, may be involved in the transfer of iron from the 3-fold axis of trimer 1 to the second site on trimer 2, and Tyr-143 of trimer 2 may help the transfer of iron from this site to His-137 of ISCU trimer 1. The structure also suggests that in the absence of ISCU, FXN 42-210 -bound iron could take a different path, from the 3-fold axis of trimer 1 to the putative ferroxidation site formed by trimer 1 together with trimer 2 (via Tyr-95 and Glu-96 of trimer 1), followed by transfer to the mineralization site at the 4-fold axis of the complex (Fig. 13,  C-E). This alternate path may serve to detoxify FXN 42-210 -bound iron that is not utilized (Fig. 13, B-E), as proposed previously (63).
The model described above is consistent with several previous observations. The PVK motif of Isu1 was shown to be important for Isu1-Yfh1 interactions in yeast (50). In our structure, the PVK motif of ISCU is close to the Fe-S cluster assembly site and, based on both cross-linking data and PISA interface analysis, it may interact with both the flexible loop of NFS1 at a distance of ϳ7-8 Å, which contains the catalytic Cys-381, as well as with loops L4 and L6 of FXN 42-210 at a distance of ϳ5 Å, which contain the iron-binding site formed by Thr-142 (loop L4) and Ser-157 (loop L6) (Fig. 13B). These interactions may contribute to the stability of the cluster assembly site and facilitate sulfur and iron delivery to ISCU. In addition, it was previously shown that replacement of Leu-63, Val-72, or Phe-94 of yeast Isu1 (corresponding to Leu-63, Val-72, and Phe-93 of human ISCU) with alanine residues affects Isu1 interactions with [Nfs1]⅐[Isd11] (50,53). It was suggested that residues Leu-63, Val-72, and Phe-94 of Isu1 are critical for interaction with hydrophobic residues of Nfs1, Met-482, Pro-478, and Leu-479 (Met-442, Pro-438, and Leu-439 in NFS1) (53). Indeed, Leu-63, Val-72, and Phe-93 of ISCU in our structure are close to hydrophobic residues of NFS1, including Met-436, Pro-438, and Leu-439 but also Trp-440, Met-442, and Val-443, which are located on the C-terminal portion of NFS1 (Fig. 12B). In addition, a G50E mutation in ISCU was shown to affect ISCU interactions with NFS1 (64). In our structure, Gly-50 is located on the flexible portion of the ISCU N terminus, close to helix ␣13 of NFS1 (Fig. 12B). A Gly to Glu substitution may reduce the flexibility of the N-terminal portion of ISCU and may affect formation of hydrophobic interactions with NFS1. Finally, certain frataxin mutations identified in Friedreich ataxia patients are located in the ␤-sheet region of the protein and affect interactions of frataxin with other protein partners (52,65   to interact with NFS1 in vivo (19). In vitro, however, the W155R mutation affected the ability of both FXN  and FXN 81-210 to stimulate the cysteine desulfurase activity of NFS1 and to promote Fe-S cluster synthesis (19). We proposed that in the context of oligomeric FXN 42-210 , Trp-155 might play a role independent of (or not limited to) providing a binding site for the [NFS1]⅐[ISD11] and/or ISCU. Based on the structure presented here, we further propose that the W155R and other pathogenic mutations in the ␤-sheet of FXN  (N146K, I154F, and R165C) may interfere with sulfur transfer from NFS1 to ISCU. In our model, residues in the ␤-sheet of FXN 42-210 are within 5 Å from the PVK motif of ISCU (Fig.  13B). Any disturbance to the PVK motif may affect both Cys-381 in the NFS1 flexible loop, at ϳ7-8 Å from the PVK motif, as well as His-137 and Cys-138 of ISCU, which are in the same flexible loop as Pro-133, Val-134, and Lys-135.
Thus, the structural model of the complex presented in this paper provides a path for concerted iron and sulfur donation from FXN  and NFS1 to ISCU, and it also helps to explain the effects of mutations in conserved amino acids of the three proteins (50, 53, 64), including pathogenic mutations in frataxin involved in Friedreich ataxia (52,65).

Experimental Procedures
DNA Constructs Used for Four-protein Co-expression in E. coli-Two DNA fragments encoding residues 35-168 of ISCU, which comprise the predicted mature form of this protein (66), and residues 42-210 of FXN, which comprise the longest native isoform of FXN (7), were PCR-amplified and cloned into the pET-52b(ϩ) 3C/LIC vector using the LIC Duet Adaptor and LIC Cloning kit (Novagen) according to the manufacturer's protocol. This cloning enables the simultaneous expression of ISCU and of FXN  with an N-terminal Strep⅐Tag II. Two DNA fragments corresponding to residues 56 -458 of NFS1 and 6 -92 of ISD11, comprising the predicted mature forms of these proteins (6,38), were respectively cloned into the BglII and HindIII sites and the NcoI and EcoRI sites of the pCDFDuet-1 vector (Novagen). This cloning enables the simultaneous expression of ISD11 and of NFS1 with an N-terminal His 6 tag. The two plasmids encoding FXN  and ISCU and also NFS1 and ISD11 were co-transformed in the E. coli strain BL21 (DE3) (Novagen) for four-protein co-expression.
Purification of Four-protein Complex-E. coli cells were grown at 25°C in 2 liters of Luria broth containing 100 g/ml ampicillin and 25 g/ml streptomycin. Protein expression was induced at A 600 ϭ 0.6 with 0.5 mM isopropyl ␤-D-1-thiogalactopyranoside (Thermo Fisher Scientific) and growth continued at 15°C for ϳ18 h. Cells were harvested and disrupted by sonication in 50 mM NaH 2 PO 4 , pH 8.0, 300 mM NaCl, 10 mM imidazole, containing 1 tablet/50 ml of complete protease inhibitor mixture EDTA-free (Roche Diagnostics). The four-protein complex was isolated from total bacterial cell extract using four consecutive chromatography steps, which included nickel affinity chromatography (5 ml-HiTrap nickel column), preparative size-exclusion chromatography (16mm ϫ 60-cm column packed with Sephacryl S-300 support), streptavidin-Sepharose TM affinity chromatography (1 ml of StrepTrap TM HP column), and preparative size-exclusion chromatography as above (GE Healthcare). The complex was eluted from the HiTrap nickel column using a 50 -150 mM imidazole gradient in buffer containing 50 mM NaH 2 PO 4 , pH 8.0, 300 mM NaCl. Fractions containing all four proteins were analyzed by Sephacryl S-300 size-exclusion chromatography in buffer containing 50 mM Tris-HCl, pH 8.0, 150 mM NaCl (TN150). Fractions containing all four proteins were loaded on a StrepTrap TM HP column, and the complex was eluted with buffer containing 20 mM NaH 2 PO 4 , pH 7.4, 150 mM NaCl, 6 mM KCl, 2.5 mM d-desthiobiotin. Fractions containing the complex were finally analyzed by Sephacryl S-300 size-exclusion chromatography in TN150 buffer. The four-protein complex was stable and soluble in this buffer. The yield of purified complex was ϳ2 mg/liter bacterial culture, which could be increased to ϳ6 mg/liter bacterial culture when we used two StrepTrap TM HP columns in tandem for the third purification step. Protein concentrations were routinely determined using a BCA kit (Thermo Fisher Scientific) and are expressed as total protein.

SDS-PAGE and Mass
Spectrometry-For reducing or nonreducing SDS-PAGE, protein samples were mixed v/v with 2ϫ Laemmli buffer in the absence or presence of ␤-mercaptoethanol (350 mM final concentration), which was added freshly to the buffer from a 14.3 M stock solution. Samples were heated at 90°C for 10 min and analyzed on 18% Criterion TM precast gels (Bio-Rad). For mass spectrometric analysis of protein bands, samples were analyzed under non-reducing conditions as described above, and the gel was stained with Bio-Safe Coomassie G-250 stain (Bio-Rad). Bands of interest were excised and proteins digested in situ with trypsin, and the extracted peptides were identified by nano-flow liquid chromatography electrospray tandem mass spectrometry as described previously (32).
Amino Acid Analysis-This analysis was performed at the AAA Service Laboratory Inc., Damascus, OR. Aliquots of fractions containing complex eluted from the Sephacryl S-300 sizeexclusion chromatography were hydrolyzed in 6 N HCl, 2% (v/v) phenol for 22 h at 110°C. After hydrolysis, samples were dried, resuspended in 100 l of sampling buffer (sodium citrate, pH 2.20), and injected onto a Hitachi L8900 amino acid analyzer, using post-column ninhydrin derivatization.
Fe-S Cluster Assembly Assays-These assays were performed anaerobically as described previously (7).
Iron Measurements-Iron was measured at the Metals Laboratory, Mayo Clinic, using an inductively coupled plasma mass spectrometer in collision cell mode as described previously (32).
Transmission Electron Microscopy and Single Particle Analysis-Fractions freshly eluted from the Sephacryl S-300 column were diluted to 0.1-0.2 mg of protein/ml in TN150 buffer. Carbon-coated copper grids (400 mesh, EMS) were glow-discharged for 30 s using a DV-502A vacuum evaporator (Denton Vacuum Inc.). After optimization, protein concentrations between 0.2 and 0.3 mg/ml were found to give the best distribution of particles on the grids. Each grid was first preincubated for 1 min on a 20-l TN150 buffer drop on Parafilm M (Bemis Co., Inc.). Excess buffer was blotted, and a 10-l drop of protein sample was placed on the grid for ϳ2 min. Excess protein sample was blotted, and the grid was then washed two times by placing a 7-l drop of sterile water on the grid for 1 min. Excess water was blotted, and the grid was stained with 5 l of 1% (w/v) uranyl acetate for 5 and 30 s by successively placing two separate drops of uranyl acetate on the grid, with excess stain drawn off after each application. The grid was then left to dry on forceps for at least 30 min and stored at room temperature. Electron micrographs were collected at the University of Minnesota Characterization Facility using an FEI Tecnai G 2 F30 field emission gun cryo-transmission electron microscope with acceleration voltage at 300 kV equipped with a 4k ϫ 4k ultrascan CCD camera (4096 ϫ 4096 pixels) with an Electron Energy-Loss Gatan imaging filter. Images were collected at a defocus range from Ϫ0.2 to Ϫ3.0 m and at a magnification of 115,000-fold (1.034 Å/pixel). The EMAN2 software package (42) was used to perform contrast transfer function correction and to further process the images to generate initial and refined 3D models of the complex. To assess the presence of symmetry, structure factors were calculated from the EM density map of a refined 3D model without symmetry applied, and a self-rotation function was calculated by POLAR-RFN of the CCP4 package (43). Similar calculations were completed using the EM density map of a refined 3D model with 432 symmetry applied. We used the PDBe FSC server to calculate and plot the FSC curve for the refined 3D models described above. For each model, the program calculated the FSC curve from two reconstructions generated in EMAN2 using the even or odd half of the particle data set (42). The program also provided the resolution of the reconstruction measured where the FSC curve crosses correlation values of 0.5, 0.333, or 0.143, as well as the resolution based on the 3criterion and the 1/2 bit criterion (45).
Docking of Structures into the EM Density Maps of the 3D Models-A homology model of the FXN 42-210 monomer was generated based on the x-ray crystal structure of one subunit of the Yfh1 Y73A trimer (PDB code 3OEQ) (56) using the I-TASSER web resource (67). A model of the FXN 42-210 trimer was then generated by aligning three FXN 42-210 monomers with the N-terminal region (residues 42-93) removed, with the x-ray crystal structure of the Yfh1 Y73A trimer. Homology models of ISCU and NFS1 monomers were generated based on available crystal and NMR structures of bacterial orthologues using the I-TASSER web resource. Then, a model of ISCU trimer was generated by aligning three ISCU monomers with the N-terminal region (residues 35-50) removed with the Isu1 trimer structure from the 3D model of the yeast [Yfh1] 24 ⅐ [Isu1] 24 complex (32). As no structures were available for ISD11, a model of the ISD11 monomer was generated based solely on the ISD11 primary amino acid sequence using the I-TASSER web resource. The pyridoxal 5Ј-phosphate co-factor was modeled into NFS1 based on the A. fulgidus [IscS] 2 ⅐[IscU] 2 complex structure (PDB code 4EB5). The quality of each model was assessed from the model's Ramachandran plot and ProQ-Protein Quality Predictor measures. Quality parameters for each of the models are summarized in supplemental Table S4. The program Chimera (46) was used to segment the EM density map of the refined 3D model obtained by negative staining transmission EM and single particle analysis and to visualize and dock structures into the segmented map. FXN  and ISCU trimers were docked both separately or together as a [FXN  ] 3 ⅐[ISCU] 3 sub-complex, which yielded similar cross-correlation functions, and the latter mode of docking was selected. NFS1 was docked as individual monomeric units. Volumes of the EM map predicted to be occupied by ISD11 were at the surface of the refined 3D model and were not sufficiently delineated from the background to enable docking of ISD11 structures into the segmented map. Therefore, based on the ISD11/NFS1 stoichiometry of ϳ1:1 measured in the complex, 24 ISD11 monomers were modeled on the surface of the complex structure guided by the cross-linking data; however, ISD11 monomers were not included in the simulations described below.
Molecular Dynamics Flexible Fitting for Docked Structures-Aided by cross-linking data, the FXN  and ISCU N termini (residues 42-93 and 36 -50) were manually added to the docked structures and modeled into nearby unoccupied volumes of the EM map of the refined 3D model with 432 symmetry applied. The EM map was then converted to Situs format to enable rigid-body docking of the entire complex structure into the EM map prior to performing molecular dynamics simulations and energy minimizations. We were unable to simulate the entire structure containing Ͼ100,000 atoms due to the computational limitations of our setup. To overcome this problem, only one-half of the structure, consisting of six NFS1 dimers docked on top of four [FXN  ] 3 ⅐ [ISCU] 3 sub-complexes, was subjected to molecular dynamic simulations and energy minimizations using NAnoscale Molecular Dynamics, NAMD 2.10 (68), followed by structure improvement iterations with the program Coot (69), as described previously (32). This process yielded a structure that closely matched the [FXN  ] 3 ⅐[ISCU] 3 and NFS1 homology models used for the initial docking, with a Ramachandran plot showing 91% of the residues in the favorable region, a MolProbity score of 2.81 with only 1.2% ␤-carbon deviations Ͼ0.25 Å (70), and a Z score of Յ11, which together indicated that the simulated structure had reasonable geometry (69). For the purpose of visualizing the entire complex structure, the simulated half was aligned with itself into the EM density map, followed by structure improvement iterations with the program Coot to resolve small steric clashes between NFS1 subunits at the interface between the two halves of the structure.
Chemical Cross-linking, Limited Proteolysis, and Tandem Mass Spectrometry-For the initial characterization of the complex, we used the water-soluble cross-linker, DTSSP (Thermo Fisher Scientific), which consists of an eight-atom spacer arm with an amine-reactive NHS ester at each end and a central disulfide bond that can be cleaved with reducing agents. Four-protein complex freshly eluted from the Sephacryl S-300 column (1-2 mg of total protein in 1 ml of TN150 buffer) was incubated with DTSSP at a protein/ DTSSP molar ratio of 1:60 for 30 min at room temperature, and the reaction was quenched by the addition of 1 M Tris-HCl buffer, pH 7.5, to a final concentration of 20 mM. This sample was fractionated again on the Sephacryl S-300 column, and fractions were analyzed by SDS-PAGE on an 18% Criterion Tris-HCl precast gel (Bio-Rad), in the presence or absence of the reducing agent ␤-mercaptoethanol. To perform EM studies and to identify protein-protein interfaces in the complex, we used the water-soluble unbreakable crosslinker, BS 3 (Thermo Fisher Scientific), a monobifunctional N-hydroxysuccinimide ester that reacts primarily with the ⑀-amino group of the side chain of Lys residues and the ␣-amino group of the N terminus of polypeptides and more weakly with the OH group of the side chains of Tyr, Thr, or Ser residues (71,72). Complex freshly eluted from the Strep-Trap TM HP affinity column (1 ml containing 1-2 mg of total protein) was incubated with BS 3 at a protein/BS 3 molar ratio package was used to visualize protein surfaces in the PyMOL program.
Author Contributions-O. G. and W. R. contributed equally to the execution of all experiments, wrote the initial draft of the paper, and prepared the figures. E. C. A. helped with the development of the EM protocol. D. Y. S. provided technical assistance with complex isolation and biochemical characterization. S. A. K. and J. R. T. contributed to the structural analysis of the complex and revised the paper. G. I. coordinated the study, revised the paper, and contributed to the preparation of the figures. All authors analyzed the results and approved the final version of the manuscript.