Structure of the human frataxin-bound iron-sulfur cluster assembly complex provides insight into its activation mechanism

The core machinery for de novo biosynthesis of iron-sulfur clusters (ISC), located in the mitochondria matrix, is a five-protein complex containing the cysteine desulfurase NFS1 that is activated by frataxin (FXN), scaffold protein ISCU, accessory protein ISD11, and acyl-carrier protein ACP. Deficiency in FXN leads to the loss-of-function neurodegenerative disorder Friedreich’s ataxia (FRDA). Here the 3.2 Å resolution cryo-electron microscopy structure of the FXN-bound active human complex, containing two copies of the NFS1-ISD11-ACP-ISCU-FXN hetero-pentamer, delineates the interactions of FXN with other component proteins of the complex. FXN binds at the interface of two NFS1 and one ISCU subunits, modifying the local environment of a bound zinc ion that would otherwise inhibit NFS1 activity in complexes without FXN. Our structure reveals how FXN facilitates ISC production through stabilizing key loop conformations of NFS1 and ISCU at the protein–protein interfaces, and suggests how FRDA clinical mutations affect complex formation and FXN activation.


Introduction
Iron-sulfur clusters (ISC) are inorganic cofactors essential in all life forms with common roles in electron transfer, radical generation, and structural support. 1 In eukaryotes, the de novo ISC assembly machinery is located in the mitochondrial matrix and requires a core complex comprising the proteins NFS1, ISD11, ACP, and ISCU (SDAU). 1,2 The NFS1 cysteine desulfurase facilitates a pyridoxal 5' phosphate (PLP) cofactor to generate the sulfane sulfur from L-cysteine, and deliver it to the ISCU scaffold protein. 3,4 The accessory protein ISD11/LYRM4 is unique in eukaryotes, and was shown to stabilize NFS1 and interact directly with the acyl carrier protein ACP/NDUFAB1. 5,6 ISCU utilizes three of its conserved cysteine residues (Cys69, Cys95, Cys138) to combine the sulfane sulfur from NFS1 with an iron source, resulting in ISC formation. ISCU then exploits the highly conserved 'LLPVK' motif for interaction with the chaperones, such as GRP75/HSCB, 7 for the downstream delivery to apo-targets. Whereas electrons required for ISC assembly most likely involves mitochondrial ferredoxin/ferredoxin reductase combination, 8 the iron source remains unclear.
An intronic GAA repeat of FXN gene, resulting in deficiency of the frataxin (FXN) protein, causes autosomal recessive Friedreich's ataxia (FRDA). 9 The in vivo loss of FXN results in oxidative stress due to iron accumulation in the mitochondria, rendering FRDA a fatal and debilitating condition with chelation therapy as the only mainstay treatment option. FXN is a key allosteric regulator of ISC assembly, and stimulates NFS1 activity by binding the SDAU complex to form the five-way active SDAUF complex. [10][11][12] Zn 2+ ion has been found to completely inhibit the SDAU complex in vitro, although its activity is restored by addition of FXN. 13 Recently, crystal structures of SDA/SDAU/SDAU-Zn 2+ complexes without the key component FXN have been published, 5,14 which attributed the zinc inhibition to the sequestration of key NFS1 catalytic residue Cys381, but could not serve as template to understand the molecular roles of FXN activator. To this end, we pursued structure determination of the SDAUF complex, coupled with FXN binding studies, to decipher the FXN-mediated activation mechanism.

Results and Discussion
Recombinant production and cryo-electron microscopy of the SDAUF complex. FXN binding to the SDAU complex is dynamic, yielding low-µM dissociation constants (Kd) by bio-layer interferometry (BLI) (Supplementary Fig. 1a-c), hence presenting challenges to isolate the SDAUF complex intact with all 5 components in proper stoichiometry. Our several attempts to generate the SDAUF complex by reconstitution of individually expressed components ( Supplementary Fig. 2a,b) did not fully incorporate FXN. To remedy this, we co-expressed in E. coli a plasmid containing His6-ISD11-NFS1-ISCU, with a plasmid containing His6-FXN ( Supplementary Fig. 2c). This produced excess FXN, shifting equilibrium towards formation of a stable and active SDAUF complex comprising human SDUF co-purified with E. coli ACP (ACPec). We attempted to make the 5-way all human complex, without e coli ACP, by inserting human ACP (NDUFAB1) into the second site of the vector containing His6-FXN. Upon co-expression with the plasmid containing His6-ISD11-NFS1-ISCU, we observed a heterogeneous complex containing an approximately equimolar mixture of the desired human and contaminating E.coli ACP ( Supplementary Fig. 2d). Based on previous reports, 5 and the functional conservation of human and E. coli ACP, we continued our experiments with a homogenous complex containing E. coli ACP with human SDUF (hereafter "SDAUF"). The as isolated complex could still be inhibited by Zn 2+ due to the dissociation equilibrium of FXN ( Supplementary Fig. 3, 'SDAUF'), and the addition of more purified ISCU further exacerbated the Zn 2+ inhibition ( Supplementary Fig. 3, 'SDAUF+U'). Zn 2+ inhibition was fully reversed upon further FXN supplementation ( Supplementary  Fig. 3, 'SDAUF+F' and 'SDAUF+U+F'), explaining the need for excess FXN to maintain its bound state within the five-way complex.
Overall architecture of the SDAUF complex. Our human SDAUF-Zn 2+ structure, the first FXN-bound complex from any organism, is a symmetric heterodecamer comprising 2 copies each of the five proteins i.e. (NFS1)2(ISD11)2(ACPec)2(ISCU-Zn 2+ )2(FXN)2. Structurally it constitutes a (NFS1-ISD11-ACPec)2 homodimeric core, with one ISCU appended to each long end of the core, and one FXN fitted into the cavity next to each ISCU (Fig. 1a,b). This architecture agrees with small-angle x-ray scattering (SAXS) analysis (Fig. 1c,d and Supplementary Fig. 7a,b), whereby a theoretical SAXS profile back-calculated from our SDAUF-Zn 2+ cryo-EM structure shows a good fit to experimental scattering data ( 2 =1.98). While our five-way complex superimposes well with four-way SDAU/SDAU-Zn 2+ structures from Boniecki et al 5 within the (NFS1-ISD11-ACPec)2 core (rmsd 0.6 Å), there is significant displacement of ISCU, up to 2.0 Å away from the core, in our FXN-bound complex ( Supplementary Fig. 7c). interaction is observed, contrasting previous predictions with oligomeric FXN. 16,17 A key feature of FXN binding is its simultaneous interactions with both NFS1 protomers of the complex (Fig. 2a), which definitively supports previous predictions from crosslinking, SAXS and NMR studies. 5,18,19 Importantly, this requires a homodimeric arrangement of NFS1 within the complex, consistent with the SDAU conformation observed by Boniecki et al 5 Supplementary Fig. 1d). Therefore, NFS1, via two extensive interfaces and C-terminus, anchors FXN to interact with ISCU. This explains why FXN alone cannot bind ISCU without NFS1. 10,12,20 FXN binds to two key regions on ISCU. One ISCU-FXN interface is through the conserved ISCU Ala-loop (Ala66-Asp71), contributing the conserved Cys69 that is required for ISC biosynthesis and interacts with FXN Asn151 as well as Zn 2+ coordinating ligand Asp71. This interaction, which may account for the weaker binding caused by the FXN(N151A) variant (Fig. 2b,d), is mediated by significant changes of the ISCU Ala-loop conformation in SDAUF as compared with SDAU-Zn 2+ (rmsd ~6Å), and zinc-free SDAU (rmsd ~2Å) structures (Fig. 3a). The other, more predominant ISCU-FXN interface is through the conserved ISCU L131PPVKLHCSM140 sequence motif (Fig. 3b).
This region (Fig. 2c,d), connecting ISCU helices α3 and α4, contains: the 'L131PPVK135' sequences recognized by the GRP75/HSCB chaperones for downstream ISC delivery, 7 Cys138 the proposed sulfur acceptor for the NFS1 sulfane, 21 and Met140, a residue reportedly determining if ISC biosynthesis is FXN-dependent (as in eukaryotes) or FXN-independent (prokaryotes). 22 Previous structures of zinc-bound ISCU reveal a helical conformation for the L131PPVK135 region. 5 In our structure, the displacement of ISCU caused by FXN ( Supplementary Fig. 7d) is associated with the ISCU L131PPVK135 helix becoming loosened and more flexible, allowing His137 to pack against invariant FXN Trp155 (Fig. 2c,d). ISCU Pro133 and Val134 also pack against FXN Thr142  Supplementary Fig. 1d).
Substitution of Met140 in yeast Isu, to the amino acids (Ile, Leu, or Val) observed at the equivalent prokaryotic position (Fig. 3b), obviated the need for Yfh1 (FXN equivalent) and reversed ΔYfh1 phenotype. 23 Our structure shows that ISCU Met140 packs against FXN Pro163  Fig. 2b and Supplementary Fig. 1d).
FXN modifies the Zn 2+ environment and its influence within the complex. FXN binding to SDAU also influences the ISCU Zn 2+ environment and NFS1 Cys-loop. In the reported SDAU-Zn 2+ structure, ISCU Asp71, Cys95 and His137, and NFS1 Cys381 (from Cys-loop) form the Zn 2+ ligation. 5 This structure explained the Zn 2+ -dependent inhibition of NFS1 activity, due to sequestration of the catalytic Cys381 away from turning over the substrate cysteine. In our SDAUF-Zn 2+ structure, zinc is ligated by ISCU Asp71, Cys95 and Cys138 (Fig. 2d). This rearranged metal coordination frees up ISCU His137 (now 3.7 Å away from Zn 2+ ) to interact with FXN Trp155, Leu156 carbonyl backbone, Pro163, and ISCU Lys135 carbonyl backbone. Importantly, NFS1 Cys381 is also freed from Zn 2+ ligation, now available for sulfur transfer (Fig. 3a). Our structure captured a novel conformation of NFS1 Cys-loop that positioned Cys381 approximately halfway between the NFS1 active site and conserved ISCU Cys residues, as part of a loop trajectory of 27 Å that could take place during ISC assembly (Fig. 4). Supported by the activity assay ( Supplementary Fig. 3), our data therefore reveals how FXN unlocks the zinc inhibition of SDAU complex, to activate NFS1 into a conformation that is now poised for its incoming substrate cysteine. 13

Conclusion
To conclude, our structure elucidates how FXN binding to SDAU complex causes significant conformational changes to ISCU, which unlocks the zinc inhibition and primes its key regions (Ala-loop and LPPVK region) to facilitate mobility of NFS1 Cys-loop for sulfide formation and transfer during ISC assembly (Fig. 4). This work provides the framework for future mechanistic studies on the dynamics of SDAUF complex during next steps in ISC biosynthesis. For example, the 'LPPVK' region of ISCU, proposed to be important for downstream chaperone binding, is buried at the FXN interface in our structure, raising the questions whether FXN would be released from the complex to make way for binding ferredoxin/chaperones, 24 and whether an ISC-loaded ISCU would be replaced by 'apo-ISCU' during the ISC biosynthesis cycle.
This work also represents one of very few reported cryo-EM structures of <200 kDa and >3.5 Å resolution for both membrane and soluble proteins. Our ability to visualize ligands and cofactors in such depth shows the advancement of modern cryo-EM for structure determination. We expect more examples to follow as the technique is applied to a broad range of clinicallyrelevant targets that remain intractable for x-ray crystallography.

Cloning, expression and purification of human ISCU, FXN, and NFS1-ISD11
For bi-cistronic co-expression of NFS1-ISD11 and tri-cistronic co-expression of NFS1-ISD11-ISCU, a DNA fragment encoding His-tagged ISD11, non-tagged NFS1 (Δ1-55), and for tri-cistronic additional non-tagged ISCU (Δ1-34) separated by an in-frame ribosomal binding site, was sub- FXN variants were constructed using the Q5 ® Site-Directed Mutagenesis Kit (NEB) and confirmed by sequencing of plasmid DNA and intact mass spectrometry of purified proteins. Cells transformed with the above plasmids were grown in 100 mL auto-induction Terrific Broth at 37 °C for 5 hours and then 20 °C for 2 days. Cell pellets were resuspended in Lysis Buffer (50 mM HEPES, pH 7.5, 500 mM NaCl, 5 % Glycerol, 2 mM TCEP, 1 µg/mL Benzonase, 1:1000 EDTA-free protease inhibitor (Merck), and 5 mg/mL lysozyme), left at room temperature for 30 min and then added 1 mL/g cells 10 % Triton X-100 and frozen for >1hour. Purification was similar to above but gel filtration was replaced with a PD-10 desalting column. All variants and wild-type were checked for stability using differential scanning fluorimetry and values recorded in Fig. 2e.

Differential scanning fluorimetry
DSF was performed in a 96-well plate using an Mx3005p RT-PCR machine (Stratagene). Each well (20 µl) consisted of protein (2 µM in buffer containing 50 mM HEPES, pH 7.5, 250 mM NaCl, 5% glycerol, and 2 mM TCEP), SYPRO-Orange (Invitrogen, diluted 1000-fold of the manufacturer's stock). Fluorescence intensities were measured from 25 to 96 °C with a ramp rate of 1 °C/min. Tm was determined by plotting the intensity as a function of temperature and fitting the curve to a Boltzmann equation. Final graphs were generated using GraphPad Prism. Assays were carried out in triplicate.

Methylene blue activity assay
Sulfide production, due to cysteine desulfurase enzyme activity, was measured using the methylene blue colorimetric assay as described previously. 25,26 The standard assay was Complexes were at 0.1 mg/mL and loaded to the streptavin coated sensors. The concentration for FXN used ranged from 500 mM to 1 nM. Measurements were performed using a 90 second association step followed by a 60 second dissociation step on a black 384-well plate with tilted bottom (ForteBio). The baseline was stabilized for 30 sec prior to association and signal from the reference sensors was subtracted. A plot of response vs.
[FXN] was used for Kd determination using one site-specific binding fit in GraphPad Prism (GraphPad Software).

Small angle X-ray scattering (SAXS)
SAXS experiments for the SDAU and SDAUF complex were performed at 0.99 Å wavelength Diamond Light Source at beamline B21 coupled to the Shodex KW403-4F size exclusion column (Harwell, UK) and equipped with Pilatus 2M two-dimensional detector at 4.014 m distance from the sample, 0.005 < q < 0.4 Å -1 (q = 4π sin θ/λ, 2θ is the scattering angle). The samples were in a buffer containing 300 mM NaCl, 25 mM Hepes 7.5, 1 mM TCEP 2 % Glycerol, 1% Sucrose and the measurements were performed at 20 C. The data were processed and analyzed with Scatter and the ATSAS program package. 27 Scatter was used to calculate the radius of gyration Rg and forward scattering I(0) via Guinier approximation and to derive the maximum particle dimension Dmax and P(r) function. The ab initio model was derived using DAMMIF 28 . 20 individual models were created, then overlaid and averaged using DAMAVER. 29 FoxS 30,31 server was used for comparison of theoretical and experimental data.

Grid preparation and data acquisition
3.5 µL of 1.5 mg/ml purified SDAUF complex was applied to the glow-discharged Quantifoil Au R1.2/1.3 grid (Structure Probe), and subsequently vitrified using a Vitrobot Mark IV (FEI Company). In order to overcome an orientation bias, n-octyl-β -d-glucopyranoside (BOG, Anatrace) was added to the sample prior freezing. Cryo grids were loaded into a Titan Krios transmission electron microscope (ThermoFisher Scientific) operating at 300 keV with a Gatan K2 Summit direct electron detector. Images were recorded with SerialEM in super-resolution mode with a super resolution pixel size of 0.543 Å and a defocus range of 1.2 to 2.5 μm. Data were collected with a dose rate of 5 electrons per physical pixel per second, and images were recorded with a 10s exposure and 250 ms subframes (40 total frames) corresponding to a total dose of 42 electrons per Å 2 . All details corresponding to individual datasets are summarized in Supplementary Table S1.

Electron microscopy data processing
A total of 4,260 dose-fractioned movies were gain-corrected, 2 x binned (resulting in a pixel size of 1.086 Å), and beam-induced motion correction using MotionCor2 32 with the dose-weighting option. The SDAUF particles were automatically picked from the dose-weighted, motion corrected average images using Gautomatch. CTF parameters were determined by Gctf. 33 A total of 1,316,416 particles were then extracted using Relion 2.0 34 with a box size of 200 pixels.
The 2D, 3D classification and refinement were performed with Relion 2.0. Two rounds of 2D classification and one round of 3D classification were performed to select the homogenous particles. After selecting particle coordinates, per-particle CTF estimation was refined using the program Gctf. 33 One set of 267,153 particles was then submitted to 3D auto-refinement with C2 symmetry imposed and resulted in a 3.2 Å map ( Supplementary Fig. S4, S5). All 3D classifications and 3D refinements were started from a 60 Å low-pass filtered version of an ab initio map generated with VIPER. 35 To evaluate the contribution of imposed symmetry in the result, 3D refinement was repeated using the same set of 267,153 particles without imposing symmetry and produced a 3.4 Å map (Supplementary Fig. S4). Since the overall structures with/without imposing symmetry are nearly identical, the C2 symmetry density map was used for model building. All resolutions were estimated by applying a soft mask around the protein complex density and based on the gold-standard (two halves of data refined independently) FSC = 0.143 criterion. Prior to visualization, all density maps were sharpened by applying different negative temperature factors using automated procedures, 36 along with the half maps, were used for model building. Local resolution was determined using ResMap 37 (Supplementary Fig. S4).

Model building and refinement
The initial template of the SDAUF complex was derived from a homology-based model calculated by SWISS-MODEL. 38 Each subunit was docked into the C2 symmetry EM density map using Chimera 39 and followed by manually adjustment using COOT. 40 The model was independently subjected to global refinement and minimization in real space using the module phenix.real_space_refine in PHENIX 41 against separate EM half-maps with default parameters.
The model was refined into a working half-map, and improvement of the model was monitored using the free half map. The geometry parameters of the final models were validated in Coot and using MolProbity and EMRinger. 42 These refinements were performed iteratively until no further improvements were observed. The final refinement statistics were provided in Supplementary Table S1. Model overfitting was evaluated through its refinement against one cryo-EM half map. FSC curves were calculated between the resulting model and the working half map as well as between the resulting model and the free half and full maps for cross-validation ( Supplementary Fig. S1). Figures were produced using PyMOL 43 and Chimera. 39 Fig. 1 SDAUF-Zn 2+ structure and FXN-NFS1 interactions. a. Cryo-EM density of SDAUF-Zn 2+ structure (NFS1/NFS1' slate, ISD11/ISD11' magenta, ACP/ACP' light green, ISCU/ISCU', Cyan, FXN/FXN' orange). b. Cartoon representation of SDAUF-Zn 2+ complex. 4'-phosphopantetheine acyl chain (8Q1, yellow), Zn 2+ ion (ZN, red) and pyridoxal 5'-phosphate (PLP, wheat) are shown as sticks/spheres. c. Scattering data from a sample of SDAUF used in cryo-EM was collected (black points) and fit to the theoretical SAXS profile back-calculated from the SDAUF-Zn 2+ cryo-EM structure (red line) with  2 =1.98. d. The ab initio envelope calculated from SAXS data for the SDAUF sample was superimposed with the cryo-EM structure e. Each FXN (orange cartoon) binds to a cavity formed by interface (surface colored by electrostatic potential) of NFS1 homodimer and one ISCU. Yellow dotted lines denote protein boundaries.  Supplementary Fig. S1b) and melting temperature (Tm) of FXN variants. c. Interface of FXN β-sheet (orange) with ISCU LPPVK-region (cyan) and NFS1 Cys-loop (slate). d. Interface of FXN with ISCU Ala-loop, LPPVK-region, and Zn 2+ ion (sphere). Inset, viewpoints of panels c and d within SDAUF-Zn 2+ complex.