A Mycobacterium tuberculosis ligand-binding Mn/Fe protein reveals a new cofactor in a remodeled R2-protein scaffold

Abstract

Chlamydia trachomatis R2c is the prototype for a recently discovered group of ribonucleotide reductase R2 proteins that use a heterodinuclear Mn/Fe redox cofactor for radical generation and storage. Here, we show that the Mycobacterium tuberculosis protein Rv0233, an R2 homologue and a potential virulence factor, contains the heterodinuclear manganese/iron-carboxylate cofactor but displays a drastic remodeling of the R2 protein scaffold into a ligand-binding oxidase. The first structural characterization of the heterodinuclear cofactor shows that the site is highly specific for manganese and iron in their respective positions despite a symmetric arrangement of coordinating residues. In this protein scaffold, the Mn/Fe cofactor supports potent 2-electron oxidations as revealed by an unprecedented tyrosine-valine crosslink in the active site. This wolf in sheep's clothing defines a distinct functional group among R2 homologues and may represent a structural and functional counterpart of the evolutionary ancestor of R2s and bacterial multicomponent monooxygenases.

Diiron-carboxylate proteins perform some of the most chemically challenging oxidations observed in nature. The 2 best studied groups are the bacterial multicomponent monooxygenases (BMMs) and the ribonucleotide reductase R2 proteins. BMMs use the diiron cofactor to perform a 2-electron oxidation, the O₂-dependent hydroxylation of hydrocarbons, including the hydroxylation of methane to methanol performed by soluble methane monooxygenase (MMO) via a Fe(IV)-Fe(IV) intermediate (1–3). BMMs have different and usually broad substrate specificities, including alkanes, alkenes, and aromatic compounds. For this reason the proteins and the bacteria that produce them are of great interest for industrial and environmental applications, such as bioremediation of contaminated soil. All diiron hydrocarbon hydroxylases/monooxygenases are multisubunit complexes requiring different protein components for activity, complicating their use in biotechnological applications (4–6).

Ribonucleotide reductases (RNRs) are the only identified enzymes for de novo synthesis of all four deoxyribonucleotides. The R2 subunit of Class-I RNRs is a homodimeric diiron-carboxylate protein that performs a 1-electron oxidation. Standard R2s generate an essential stable tyrosyl radical (Y·) via a Fe(III)-Fe(IV) intermediate (7–10).

Diiron-carboxylate proteins have very similar metal sites, coordinated by 4 carboxylates and 2 histidines, and are believed to share a common evolutionary ancestor (6, 11). Much effort has gone into defining the structural and chemical determinants that direct the systems to perform 1- or 2-electron chemistry (4, 12, 13). A peculiar R2, lacking the radical harboring tyrosine, was identified in the important human pathogen Chlamydia trachomatis (14, 15). It was suggested that the 1-electron oxidizing equivalent was stored at the metal site, as opposed to as a tyrosyl radical, and that this could be an adaptation to produce a radical site that is less sensitive to scavenging by reactive nitrogen and oxygen species produced by the host's immune response. In addition, a number of proteins, previously assigned as standard R2s, were assigned to this group of Chlamydia R2-like proteins, denoted R2c. Recently it was shown that CtR2c possesses a manganese/iron redox cofactor and that the one-electron oxidizing equivalent is stored as a Mn(IV)-Fe(III) species that replaces the Fe(III)-Fe(III)-Y· cofactor in standard R2s (16, 17). Interestingly, it was also shown that not only is the Mn(IV)-Fe(III) form stable to incubations with H₂O₂, but the reduced forms are efficiently activated by H₂O₂ treatment (18).

The Mycobacterium tuberculosis R2c-like protein, Rv0233, is 1 of the 10 most up-regulated proteins, about 7-fold, in the virulent H37Rv M. tuberculosis strain compared with the avirulent bacillus Calmette–Guérin (BCG) vaccine strain and is therefore a possible virulence factor and drug target candidate (19). M. tuberculosis is the causative agent of tuberculosis (TB), one of the worst global killers, with an estimated 1.7 million yearly deaths and a third of the world's population infected. The bacterium boasts one of nature's most elaborate lipid metabolisms that is also key to its virulence, inherent drug resistance and ability to multiply within the macrophage (20). There is a rapid development of multi drug-resistant strains (MDR-TB) and extensively drug resistant TB strains (XDR-TB), resistant also to injectable second-line drugs, are emerging and pose a severe threat to TB control worldwide. Identification of new TB drugs and drug targets is imperative (21).

Results

R2c-Like Proteins Form 2 Subgroups.

When the R2c proteins were discovered, there were only a handful of sequences available for the group (15). Now this number is above 50, allowing detailed sequence analysis. This reveals that R2c-like proteins actually form 2 groups in a phylogenetic tree, 1 group including CtR2c and 1 group including M. tuberculosis Rv0233 (Fig. 1). Alignments show that the Rv0233 group does not contain the conserved C-terminal tyrosine, known to participate in radical transfer from R2 to the catalytic R1 subunit and shown to be essential for activity in CtR2c (9, 22) (Fig. S1). Moreover, a number of organisms with fully sequenced genomes that contain proteins from the Rv0233 group do not contain any other protein component of a Class I RNR system. Together, this suggests that there may be functional differences between the groups, despite an overall sequence similarity and conservation of active site and other key residues.

Fig. 1.

Phylogenetic tree of R2 homologues where the canonical radical harboring tyrosine is replaced by phenylalanine. The locations of C. trachomatis R2c and M. tuberculosis Rv0233 are indicated.

Protein Production and RNR Assay.

Rv0233 was produced as an N-terminal His₆-affinity-tagged protein by overexpression in E. coli. M. tuberculosis H37Rv possesses 1 R1 homologue, nrdE, and 3 genes that encode R2 protein homologues: nrdF1, nrdF2, and Rv0233, presently annotated as nrdB. We were unable to obtain any RNR activity above background from Rv0233 with M. tuberculosis R1 (MtR1) either with or without the addition of 1 equivalent each of Fe(II) and Mn(II) per protein monomer while the MtR2, encoded by nrdF2, was highly active with MtR1 in the established [³H]CDP assay as also shown previously (23). Activities obtained were (activity relative to MtR1+MtR2 in %, average ± 1 SD); MtR1+MtR2: 100 ± 7.1, MtR1+Rv0233: 0.55 ± 0.12, MtR1+Rv0233+Fe(II)+Mn(II): 0.50 ± 0.083, MtR1 only, 0.43 ± 0.083.

Overall Structure.

The protein was crystallized and the structure solved by SAD phasing to 1.9-Å resolution (Table 1). Rv0233 is a homodimer with the same interaction surface and geometry as R2s, producing the well-known heart-shaped dimer (Fig. 2A). The chain could be traced from residue 2 through 290, and the overall structure is virtually identical to that of R2s. The core 8-helix bundle is conserved, and the main differences are observed in helices αD, αE, and the last 2 helices, αG and αH [nomenclature as defined in (24)] (Fig. 2B). The largest backbone structural differences are observed in the N- and C-termini. The position of the C-terminus is interesting in comparison to R2s. The R2 C- terminus is known to interact with R1 and the C-terminal 30–40 residues are disordered in all R2 structures determined to date. The last modeled residues, however, align in space within a radius of only ≈5 Å, most likely relevant for the interaction with R1. In Rv0233, the C-terminal 24 residues are also disordered but the last ordered residue is separated by ≈30 Å compared with R2s. The difference in location of the C-terminus further supports the sequence and biochemical data showing that the protein is not an R2 component of a class I RNR.

Table 1.

Crystallographic data and refinement statistics for Rv0233

	Rv0233 λ = 0.934 Å
Data collection	Mosflm
Beamline	ID14–1
Wavelength, Å	0.934
Space group	P3221
Cell parameters
Å	54.57; 54.57; 176.65
°	90; 90; 120
Resolution, Å	40–1.90 (2.00–1.90)
Completeness, %	99.9 (99.4)
Redundancy	5.2 (4.4)
Rsym, %	6.9 (36.2)
I/σI	17.1 (3.5)
Refinement statistics	Refmac 5.4.0073
Resolution, Å	40–1.9
No. of unique reflections	24918
No. of reflections in test set	1268
Rwork, %	14.9
Rfree, %	17.7
No. of atoms
Protein	2319
Metal ions	2
Ligand	16
Solvent	266
RMSD from ideal values*
Bond length, Å	0.022
Bond angle, °	1.699
Ramachandran outliers, %†	1.5

*Ideal values from (37).

^†Calculated using a strict-boundary Ramachandran plot (38).

Fig. 2.

Structure of Rv0233. (A) Overall dimeric structure of the protein. (B) Superposition of Rv0233 (blue), E. coli R2 (green), and C. trachomatis R2c (red). Helices αD, αE, αG, and αH display the largest differences, the positions of the C-termini are indicated.

Ligand Binding.

A striking difference compared to R2 proteins is the presence of a bound ligand that coordinates directly to the metal site (Fig. 3A). The ligand is modeled as myristic (C₁₄) acid because of the fit to the electron density and the complete lack of H-bond interactions between the ligand tail and the protein. The ligand is bound in a large continuous cavity, similar to the one observed in BMMs with large substrates, for example, toluene monooxygenase (4, 25), extending from the metal site toward the protein surface made up of the loop linking helices αG and αH (Fig. 3B). The cavity is narrow and hydrophobic close to the metal site and widens, once it has passed between helices αB and αE of the metal coordinating 4-helix bundle, to produce a larger cavity with more H-bond possibilities. The narrow part of the cavity is completely occupied by the lipid ligand, while the wider part of the cavity is also solvated by ordered water molecules. In the present structure, the cavity is closed but conformational changes in the loop linking helices αG and αH or rotamer changes of R59, E244, L248, or Y249 would open the cavity to bulk solvent. The cavity is produced without significant movement of the protein backbone compared with R2s. This is achieved by numerous substitutions from larger to smaller sidechains in combination with architectural differences in the second shell of cavity lining residues that allow a number of first shell side chains to position differently, creating the cavity space. Based on the properties of the ligand binding cavity and the bound ligand it seems most likely that the protein is involved in the bacterium's lipid metabolism. Studies to define the in vivo substrate are initiated but complicated by the pathogenicity and very extensive lipid metabolism of M. tuberculosis, including several poorly described mycobacterial-specific pathways.

Fig. 3.

Ligand binding and cavity in Rv0233. (A) Ligand binding and interaction with the metal site shown by an Fo-Fc omit map for the ligand contoured at 0.42 eÅ⁻³. (B) The ligand-binding cavity in Rv0233 shown by the protein molecular surface, the bound ligand is displayed as VdW spheres, ordered water molecules in the cavity are indicated. Residues R59, E244, L248, or Y249, restricting bulk solvent access are shown as sticks.

Structure of the Heterodinuclear Metal Cofactor.

Protein produced in standard rich LB media contains significant amounts of both manganese and iron, approximately 0.7 and 1.2 equivalents, respectively, as determined by ICP-SFMS. The Mn/Fe ratio is also reflected in the relative intensity of the K-level X-ray emission lines from the crystal. Addition of 2 mM MnCl₂ to the expression medium shifts this relation close to unity. Thus, no matter if the protein is produced at Mn:Fe ratios in the expression media of roughly 1:10 (rich media) or 100:1 (rich media + 2 mM MnCl₂), it still contains close to 1 equivalent of both Mn and Fe per polypeptide.

The metal binding properties of the protein suggest to us that this protein, like the related C. trachomatis R2, contains the Mn/Fe-carboxylate redox cofactor. Anomalous diffraction difference maps show that the metal binding is specific with the manganese ion occupying the site closest to the position of the radical harboring tyrosine in standard R2s, thus replacing Fe1 (Fig. 4A). The anomalous data were collected on protein produced in Mn-supplemented media. Still, there is no Mn anomalous signal from the Fe-site above the noise level of the map. Similarly, there is no sign of Fe binding in the Mn site. The metal binding is thus very specific, given that both metal ions are present in sufficient amounts. The manganese ion has fewer carboxylate coordinations, despite the symmetric arrangement of coordinating residues, this difference may affect metal specificity. However, the coordination of the heterodinuclear cofactor is very similar to that observed in diiron BMMs and R2s. In these systems there is also known to be a large flexibility in metal coordination depending on oxidation state and coordinating exogenous ligands (3, 4, 26, 27). The basis for the strikingly strict metal specificity should lie in the metal free and reduced M(II)-M(II) forms because it is at this oxidation state that the metals bind to the protein. It seems very likely that the metal specificity in this system involves outer-sphere effects. The present structure together with existing data on diiron systems should provide a useful tool to study the fundamental processes of metal specificity and redox tuning in proteins. The metal site surroundings and H-bonding distances are depicted in Fig. 4 B and C. The non-coordinating H_xO species that is H-bonded to both the exogenous ligand and Y175 refines to a distance of 3.0 Å from the manganese ion and does not seem to coordinate it directly; the electron density for this ligand is also somewhat weaker than for the most well ordered water molecules, indicative of dynamics or partial occupancy.

Fig. 4.

Rv0233 active site. (A) Anomalous difference maps. Purple: manganese anomalous difference density, contoured at 0.09 eÅ⁻³. Green: iron-specific ddano map, contoured at 0.07 eÅ⁻³. (B and C) Metal-site coordination and hydrogen bonding network of conserved residues in the metal site surrounding, distances in Å.

Tyrosine-Valine Crosslink in the Active Site.

The catalytic potential of the metal site is manifested in the protein by the formation of an unprecedented tyrosine-valine crosslink, likely generated during one of the first redox cycles of the metal site. The phenolic oxygen of Y162 is covalently bound to Cβ of V71, connecting the metal site coordinating helices αB and αE (Fig. 5). The net chemical reaction is a 2-electron oxidation of the V71-Y162 side chain pair with removal of 2 hydrogen atoms, resulting in the crosslink. With the exception of a few reported cases of peptides containing hydroxyvaline, this result is to our knowledge a previously undocumented modification of a valine side chain in a protein. Modification of the very inert aliphatic side chain suggests that the Mn/Fe cofactor is capable of similarly challenging 2-electron oxidations as BMMs. The use of the heterodinuclear site is thus not limited to generating and storing a one-electron oxidizing equivalent as in CtR2c (16). Crosslinking of these amino acids, which are also conserved in the group, possibly prepares the active site for subsequent enzymatic chemistry (see Discussion).

Fig. 5.

Covalent crosslink between V71 and Y162 shown by an Fo-Fc omit map for the residues contoured at 0.42 eÅ⁻³. The π-helical part of αE is illustrated by the n + 5 main chain H-bonds.

We have considered if there could be alternative explanations for the formation of the crosslink than via oxidation by the heterodinuclear site. The only option in this case would be the action of another enzyme. The crosslink is deeply buried and, to be accessible, more than half of the protein would need to be unfolded. Moreover, this would also imply that the expression host, E. coli, possesses a system to form this crosslink in a M. tuberculosis protein that has no close homologues in E. coli. This possibility appears extremely unlikely and cannot be considered a real option for crosslink formation.

A Combination of Conserved Features from both R2s and BMMs.

In the direct vicinity of the metal site, the ligand-binding cavity is established by the substitution of a phenylalanine residue, which is absolutely conserved in R2s, to an alanine (A171) conserved in the Rv0233 group (Fig. S1). This positions the cavity in the same place as the active site in BMMs. Moreover, in certain R2 mutants that are engineered toward 2-electron chemistry, this phenylalanine becomes hydroxylated (12, 28). In the present structure, the bound ligand occupies the same position in space and is thus located at the preferred location for substrate oxidation in diiron carboxylate proteins.

The hydrogen bonding network on the histidine side of the metal site is known to control electron transfer and tuning between 1- and 2-electron chemistry. In this area, the Rv0233 group displays a composite structure of features otherwise unique to R2s or BMMs (Fig. S1 and Fig. 4 B and C). The residues preceding both metal coordinating histidines are absolutely conserved as arginines in BMMs, indicating that they are essential for function or folding (4). The corresponding residues are mainly hydrophobic in R2s and R2cs. The Rv0233 group, like the BMMs has conserved positively charged residues in these positions, although the first is a lysine (K103). This indicates that these residues in BMMs and the Rv0233 group are involved in electron transfer or redox tuning, rather than necessary for folding because the structurally virtually identical R2cs lack them. In addition, the Rv0233 group also possesses the tryptophan, W32 (W48 E. coli R2 numbering), normally a hallmark of R2s and involved as a radical species in cofactor assembly (29). In BMMs, the large side chain of the arginine preceding the first metal coordinating histidine occupies the same position in space as the conserved tryptophan in R2s. In Rv0233, on the other hand, the corresponding K103 amine is involved in a π-cation interaction with the pyrrole ring of W32, most likely tuning its chemical properties and redox potential. The hydrogen bonding network is thus also very similar to the one in the stearoyl-acyl carrier protein desaturases, another group of 2-electron, lipid-oxidizing, diiron carboxylate proteins that display a different dimer interaction geometry than R2s and BMMs (30). Y222 in CtR2c was recently shown to contribute in mediating the 1-electron reduction of the Mn(IV)-Fe(IV) state to produce the active Mn(IV)-Fe(III) state. Mutation of this residue slows down the external 1-electron reduction, thus stabilizing the Mn(IV)-Fe(IV) intermediate (22). The corresponding residue is not conserved as an electron-relay competent residue in the Rv0233 group, something that may stabilize the Mn(IV)-Fe(IV) state and contribute to direct the protein to 2-electron oxidations.

Discussion

Recently, a group of R2 proteins was discovered (15). This group uses a heterodinuclear Mn/Fe redox cofactor that, upon reaction with molecular oxygen, yields a Mn(IV)-Fe(III) oxidation state that is used in place of the diiron-tyrosyl radical system of standard R2s (16). This solution, which actually seems less complex, also appears to be less sensitive to certain radical scavengers. Here, we show that the use of the Mn/Fe-carboxylate cofactor is not limited to the R2c proteins but is also present in a new group of ligand-binding Mn/Fe-carboxylate proteins. On the sequence level this group is easily confused with the Mn/Fe-containing R2c proteins but the present structure allows assignment of available sequences to the different groups. There are a number of sequence particulars that strongly indicate that the proteins in the new group are all ligand-binding oxidases, and we thus propose that they are denoted “R2-like ligand binding oxidases.” The in vivo substrates and products are presently unknown, and may well differ within the group. It seems most likely that the proteins are hydrocarbon oxidases, possibly involving oxygen insertion. The potential for challenging 2-electron oxidations by the heterodinuclear cofactor is demonstrated by the formation of an unprecedented tyrosine-valine crosslink in the active site. This shows that the Mn/Fe cofactor has a richer chemical repertoire than the generation and storage of a one-electron oxidizing equivalent, as in CtR2c, and may be similarly versatile as the diiron-carboxylate cofactor.

We also describe the detailed structure of this cofactor. Even though this structure is for a protein that is not an R2 the extensive structural similarities between the groups strongly suggest that the R2c proteins also have the same arrangement with the Mn ion substituting for Fe1. Since it is known that the manganese assumes an (IV) oxidation state and serves as the radical initiator in the active state of CtR2c its position has great importance for the radical transfer in these systems and should also have implications for the details of radical transfer in the canonical diiron R2 proteins.

Because the residues involved in the tyrosine-valine crosslink are conserved, it likely has relevance for protein function, this role is not obviously apparent. Some interesting features in relation to BMMs can however be noted. Binding of the regulatory protein in BMMs induces structural changes in the active site that increase oxygen reactivity and turnover as well as influence the regiospecificity for hydroxylation (3). CD and MCD studies on MMO show that this is mainly a result of structural changes around Fe2, in particular of E209, corresponding to E167 in Rv0233 (31). In the complex between phenol hydroxylase and its regulatory protein the αE helix adopts a π-helical structure in this region. It was proposed that this feature might mediate the effect of the regulatory protein to the active site (25). A recent study of toluene-4-monooxygenase shows that effector protein binding induces a number of structural changes in the metal site coordinating helices, especially αE, leading to changes in both metal ligation and the active-site channel (32). The Y-V crosslink in Rv0233 puts a strict geometric restraint between helices αB and αE and establishes a π-helix in αE, comprising 2 full turns and including E167, illustrated by the n + 5 main chain H-bonds in Fig. 5. The αE helices in BMMs are more distorted in terms of straightness than αE in Rv0233; still it is noteworthy that the same segment adopts a π-helical structure. A possible consequence is that the crosslink functions as a poor man's regulatory protein, imposing geometric restraints that fix the helix in its π-conformation and thus lock the protein in one of several states otherwise controlled by the regulatory protein in BMMs. It remains to be shown whether this adduct has additional functions or takes part in the chemistry as a cofactor.

Reconstitution of the Mn/Fe cofactor in CtR2c involves a Mn(IV)-Fe(IV) oxidation state (33). This is interesting because the Fe(IV)-Fe(IV) state has never been observed in an R2 protein while methane monooxygenase is known to use this intermediate for substrate oxidation. The phenolic oxygen of the cross-linked Y162 is located 5.1 Å from the iron ion. This is the same distance as the buried radical harboring tyrosine in canonical R2s is to Fe1 (ranging from ≈ 4.6–6.6 Å depending on species and oxidation state). By analogy with these systems we hypothesize that the covalent link may be created via a mechanism involving a Y162 radical produced by the Mn(IV)-Fe(IV) oxidation state (Fig. S2).

Interestingly, the phenolic oxygen of another conserved tyrosine, Y175, present in all proteins in the Rv0233 group, but conserved as Phe in R2cs and standard R2s, is H-bonded to the metal site ligand E202 and via a water molecule to E68 (Fig. 4 B and C). The phenolic oxygen of Y175 lines the substrate-binding cavity and is positioned 4.9 Å from the manganese ion. Similarly to the radical harboring tyrosine in R2s, Y175 is also linked by H-bonds to the metal site. It thus seems reasonable that Y175 can become oxidized to a radical species by the metal site. However, unlike the R2s that bury the tyrosyl radical in a hydrophobic pocket in the protein, this residue is exposed to the substrate. In the present structure Y175 is also H-bonded to the exogenous ligand via a water molecule. This opens the possibility that substrate oxidation and formation of the covalent crosslink proceed via similar mechanisms involving tyrosyl radical-linked high valent metal site intermediates. Based on the positions of the H_xO species, reactive metal-oxygen intermediates produced by dioxygen cleavage are expected to reside on the substrate-binding side of the metal site close to Y175, providing possibilities for oxygen insertion into the substrate.

From the pattern of sequence conservation it seems that the protein is a hybrid between BMMs and R2s. The conserved features of the 2 diiron systems that are merged in Rv0233 must, however, be interpreted in light of that the protein has a high specificity for Mn and Fe and that it is the heterodinuclear cofactor that produces the tyrosine-valine crosslink. The R2-like ligand binding oxidases described here close the circle of a group of proteins that perform both fascinating and important chemistry. They merge structural and functional features from 2 well-studied families, the R2s and BMMs. This group of proteins should provide a key tool to consolidate and test theories about mechanistic differences and similarities.

Evolution of BMMs is believed to have ensued via a gene duplication of a diiron carboxylate protein and subsequent divergence into the catalytic α-subunit and the non-catalytic β-subunit, while retaining the overall fold and dimer interaction geometry. Accessory protein subunits were likely acquired via horizontal gene transfer, which also largely characterizes the spread of BMMs (6, 11). The Mn/Fe-carboxylate proteins could represent structural and functional homologues of an ancient ancestor of R2s and BMMs and Mn/Fe could be considered a possible candidate for the ancestral cofactor. Although the evolutionary relationship is clearly the topic for more detailed analysis, it is apparent that both functions can be housed in very similar homodimeric protein scaffolds with a heterodinuclear Mn/Fe-carboxylate cofactor.

Materials and Methods

Detailed materials and methods are described in SI Materials and Methods.

Bioinformatics.

Sequences encoding R2 homologues but lacking the canonical radical harboring tyrosine were collected by sequence database searching.

Cloning, Protein Expression, Purification, Enzymatic Assays, and Metal Analysis.

The Rv0233 gene was PCR cloned from M. tuberculosis strain H37Rv (20) and overexpressed in E. coli BL21(DE3) grown in LB medium either without metal supplement or with the addition of 2 mM MnCl₂. Protein was purified by affinity and size exclusion chromatography. Ribonucleotide reductase enzymatic activity for M. tuberculosis R1 with Rv0233 was measured using the established [³H]CDP assay. Initial indication that the protein contained more than one metal was obtained by a simple combined luminescence and colorimetric assay (34). Quantitative metal analysis was performed using inductively coupled plasma sector field mass spectrometry. The intensity of the X-ray K-level emission lines were also used to estimate the relative amount of Mn and Fe in the crystal as well as to verify that the crystallization process did not impose any significant enrichment of protein containing a particular metal.

Crystallization, Data Collection, and Structure Determination.

Rv0233 was crystallized using the vapor diffusion method. Diffraction data were collected at the ESRF synchrotron in Grenoble, France. Data collection statistics are shown in Tables 1, S1, and S2. The intrinsic metal cofactor of the protein was used to phase the data by means of single-wavelength anomalous dispersion methods using anomalous data collected at the high energy side of the iron edge. Model statistics are presented in Table 1. To determine metal identity in the different binding sites, anomalous diffraction data were collected at the high-energy side of the Mn-edge (λ = 1.8 Å) and the high-energy side of the Fe-edge (λ = 1.7 Å) (Table S2). Anomalous difference model phased Fourier (DANO) maps were calculated with FFT (35). At λ = 1.8 Å, manganese, but not iron, displays an anomalous signal and these data were used to determine the position of the Mn ion. Since both manganese and iron display anomalous signals at λ = 1.7 Å, an iron-specific “difference DANO” map was calculated using both datasets according to (36). Coordinates and structure factors are deposited in the PDB with id 3EE4.