We produced recombinant R. rubrum encapsulin nanocompartments in E. coli by co-expression of the encapsulin (Rru_A0974) and EncFtn (Rru_A0973) proteins, and purified these by sucrose gradient ultra-centrifugation (Figure 1A) (Sutter et al., 2008). TEM imaging of uranyl acetate-stained samples revealed that, when expressed in isolation, the encapsulin protein forms empty compartments with an average diameter of 24 nm (Figure 1B and Figure 1—figure supplement 1A/C), consistent with the appearance and size of the T. maritima encapsulin (Sutter et al., 2008). We were not able to resolve any higher-order structures of EncFtn by TEM. Protein purified from co-expression of the encapsulin and EncFtn resulted in 24 nm compartments with regions in the center that exclude stain, consistent with the presence of the EncFtn within the encapsulin shell (Figure 1C and Figure 1—figure supplement 1B/C).

Figure 1 with 1 supplement see all

Download asset Open asset

We purified recombinant R. rubrum EncFtn as both the full-length sequence (140 amino acids) and a truncated C-terminal hexahistidine-tagged variant (amino acids 1–96 plus the tag; herein EncFtn_sH). In both cases the elution profile from size-exclusion chromatography (SEC) displayed two peaks (Figure 2A). SDS-PAGE analysis of fractions from these peaks showed that the high molecular weight peak was partially resistant to SDS and heat-induced denaturation; in contrast, the low molecular weight peak was consistent with monomeric mass of 13 kDa (Figure 2B). MALDI peptide mass fingerprinting of these bands confirmed the identity of both as EncFtn. Inductively coupled plasma mass spectrometry (ICP-MS) analysis of the SEC fractions showed 100 times more iron in the oligomeric fraction than the monomer (Figure 2A, blue scatter points; Table 1), suggesting that EncFtn oligomerization is associated with iron binding. In order to determine the iron-loading stoichiometry in the EncFtn complex, further ICP-MS experiments were performed using selenomethionine (SeMet)-labelled protein EncFtn (Table 1). In these experiments, we observed sub-stoichiometric metal binding, which is in contrast to the classical ferritins (Le Brun et al., 2010). Size-exclusion chromatography with multi-angle laser light scattering (SEC-MALLS) analysis of samples taken from each peak gave calculated molecular weights consistent with a decamer for the high molecular weight peak and a monomer for the low molecular weight peak (Figure 2C, Table 2).

Figure 2

Download asset Open asset

We purified EncFtn_sH from E. coli grown in MM with or without the addition of 1 mM Fe(NH₄)₂(SO₄)₂. The decamer to monomer ratio in the sample purified from cells grown in iron-supplemented media was 4.5, while that from the iron-free media was 0.11, suggesting that iron induces the oligomerization of EncFtn_sH in vivo (Figure 3A, Table 3). To test the metal-dependent oligomerization of EncFtn_sH in vitro, we incubated the protein with various metal cations and subjected samples to analytical SEC and non-denaturing PAGE. Of the metals tested, only Fe²⁺, Zn²⁺ and Co²⁺ induced the formation of significant amounts of the decamer (Figure 3B, Figure 3—figure supplement 1/2). While Fe²⁺ induces the multimerization of EncFtn_sH, Fe³⁺ in the form of FeCl₃ does not have this effect on the protein, highlighting the apparent preference this protein has for the ferrous form of iron. To determine if the oligomerization of EncFtn_sH was concentration dependent we performed analytical SEC at 90 and 700 µM protein concentration (Figure 3C). At the higher concentration, no increase in the decameric form of EncFtn was observed; however, the shift in the major peak from the position of the monomer species indicated a tendency to dimerize at high concentration.

Figure 3 with 2 supplements see all

Download asset Open asset

We determined the crystal structure of EncFtn_sH by molecular replacement to 2.0 Å resolution (see Table 1 for X-ray data collection and refinement statistics). The crystallographic asymmetric unit contained thirty monomers of EncFtn with visible electron density for residues 7 – 96 in each chain. The protein chains were arranged as three identical annular decamers, each with D5 symmetry. The decamer has a diameter of 7 nm and thickness of 4 nm (Figure 4A). The monomer of EncFtn has an N-terminal 3₁₀-helix that precedes two 4 nm long antiparallel α-helices arranged with their long axes at 25° to each other; these helices are followed by a shorter 1.4 nm helix projecting at 70° from α2 (Figure 4B). The C-terminal region of the crystallized construct extends from the outer circumference of the ring, indicating that the encapsulin localization sequence in the full-length protein is on the exterior of the ring and is thus free to interact with its binding site on the encapsulin shell protein (Sutter et al., 2008).

Figure 4 with 1 supplement see all

Download asset Open asset

The monomer of EncFtn_sH forms two distinct dimer interfaces within the decamer (Figure 4 C/D). The first dimer is formed from two monomers arranged antiparallel to each other, with α1 from each monomer interacting along their lengths and α3 interdigitating with α2 and α3 of the partner chain. This interface buries one third of the surface area from each partner and is stabilized by thirty hydrogen bonds and fourteen salt bridges (Figure 4C). The second dimer interface forms an antiparallel four-helix bundle between helices 1 and 2 from each monomer (Figure 4D). This interface is less extensive than the first and is stabilized by twenty-one hydrogen bonds, six salt bridges, and a number of metal ions.

The arrangement of ten monomers in alternating orientation forms the decamer of EncFtn, which assembles as a pentamer of dimers (Figure 4A). Each monomer lies at 45° relative to the vertical central-axis of the ring, with the N-termini of alternating subunits capping the center of the ring at each end, while the C-termini are arranged around the circumference. The central hole in the ring is 2.5 nm at its widest in the center of the complex, and 1.5 nm at its narrowest point near the outer surface, although it should be noted that a number of residues at the N-terminus are not visible in the crystallographic electron density and these may occupy the central channel. The surface of the decamer has distinct negatively charged patches, both within the central hole and on the outer circumference, which form spokes through the radius of the complex (Figure 4—figure supplement 1).

The electron density maps of the initial EncFtn_sH model displayed significant positive peaks in the mFo-DFc map at the center of the 4-helix bundle dimer (Figure 5—figure supplement 1). Informed by the ICP-MS data indicating the presence of iron in the protein we collected diffraction data at the experimentally determined iron absorption edge (1.74 Å) and calculated an anomalous difference Fourier map using this data. Inspection of this map showed two 10-sigma peaks between residues Glu32, Glu62 and His65 of two adjacent chains, and a statistically smaller 5-sigma peak between residues Glu31 and Glu34 of the two chains. Modeling metal ions into these peaks and refinement of the anomalous scattering parameters allowed us to identify these as two iron ions and a calcium ion respectively (Figure 5A). An additional region of asymmetric electron density near the di-iron binding site in the mFo-DFc map was modeled as glycolic acid, presumably a breakdown product of the PEG 3350 used for crystallization. This di-iron center has an Fe-Fe distance of 3.5 Å, Fe-Glu-O distances between 2.3 and 2.5 Å, and Fe-His-N distances of 2.5 Å (Figure 5B). This coordination geometry is consistent with the di-nuclear ferroxidase center (FOC) found in ferritin (Bertini et al., 2012). It is interesting to note that although we did not add any additional iron to the crystallization trials, the FOC was fully occupied with iron in the final structure, implying that this site has a very high affinity for iron.

Figure 5 with 1 supplement see all

Download asset Open asset

The calcium ion coordinated by Glu31 and Glu34 adopts heptacoordinate geometry, with coordination distances of 2.5 Å between the metal ion and carboxylate oxygens of Glu31 and Glu34 (E31/34-site). A number of ordered solvent molecules are also coordinated to this metal ion at a distance of 2.5 Å. This heptacoordinate geometry is common in crystal structures with calcium ions (Figure 5C) (Katz et al., 1996). While ICP-MS indicated that there were negligible amounts of calcium in the purified protein, the presence of 140 mM calcium acetate in the crystallization mother liquor favors the coordination of calcium at this site. The fact that the protein does not multimerize in solution in the presence of Fe³⁺ may indicate that these metal binding sites have a lower affinity for the ferric form of iron, which is the product of the ferroxidase reaction. A number of additional metal-ions were present at the outer circumference of at least one decamer in the asymmetric unit (Figure 5D). These ions are coordinated by His57, Glu61 and Glu64 from both chains in the FOC dimer and are 4.5 Å apart; Fe-Glu-O distances are between 2.5 and 3.5 Å and the Fe-His-N distances are 4 and 4.5 Å.

Structural alignment of the di-iron binding site of EncFtn_sH to the FOC of Pseudo-nitzschia multiseries ferritin (PmFtn, PDB ID: 4ITW) reveals a striking similarity between the metal binding sites of EncFtn_sH and the classical ferritins (Pfaffen et al., 2013) (Figure 6A). The di-iron site of EncFtn_sH is by necessity symmetrical, as it is formed through a dimer interface, while the FOC of ferritin does not have these constraints and varies in different species at a position equivalent to His65 of the second EncFtn monomer in the FOC interface (His65’) (Figure 6A). Structural superimposition of the FOCs of ferritin and EncFtn brings the four-helix bundle of the ferritin fold into close alignment with the EncFtn dimer, showing that the two families of proteins have essentially the same architecture around the di-iron center (Figure 6B). The linker connecting helices 2 and 3 of ferritin is congruent with the start of the C-terminal helix of one EncFtn monomer and the N-terminal 3₁₀ helix of the second monomer (Figure 6C).

Figure 6 with 1 supplement see all

Download asset Open asset

In order to confirm the assignment of the oligomeric state of EncFtn_sH and investigate further the Fe²⁺-dependent assembly, we used native nano-electrospray ionization (nESI) and ion-mobility mass spectrometry (IM-MS). As described above, by recombinant production of EncFtn_sH in minimal media we were able to limit the bioavailability of iron. Native MS analysis of EncFtn_sH produced in this way displayed a charge state distribution consistent with an EncFtn_sH monomer (blue circles, Figure 7A1) with an average neutral mass of 13,194 Da, in agreement with the predicted mass of the EncFtn_sH protein (13,194.53 Da). Under these conditions, no significant higher order assembly was observed and the protein did not have any coordinated metal ions. Titration with Fe²⁺ directly before native MS analysis resulted in the appearance of a new charge state distribution, consistent with an EncFtn_sH decameric assembly (+22 to +26; 132.65 kDa) (Figure 7A2/3). After instrument optimization, the mass resolving power achieved was sufficient to assign iron-loading in the complex to between 10 and 15 Fe ions per decamer (Figure 7B, inset top right), consistent with the presence of 10 irons in the FOC and the coordination of iron in the Glu31/34-site occupied by calcium in the crystal structure (Δmass observed ~0.67 kDa). MS analysis of EncFtn_sH after addition of further Fe²⁺ did not result in iron loading above this stoichiometry. Therefore, the extent of iron binding seen is limited to the FOC and Glu31/34 secondary metal binding site. These data suggest that the decameric assembly of EncFtn_sH does not accrue iron in the same manner as classical ferritin, which is able to sequester around 4500 iron ions within its nanocage (Mann et al., 1986). Ion mobility analysis of the EncFtn_sH decameric assembly, collected with minimal collisional activation, suggested that it consists of a single conformation with a collision cross section (CCS) of 58.2 nm² (Figure 7B). This observation is in agreement with the calculated CCS of 58.7 nm²derived from our crystal structure of the EncFtn_sH decamer (Marklund, 2015). By contrast, IM-MS measurements of the monomeric EncFtn_sH at pH 8.0 under the same instrumental conditions revealed that the metal-free protein monomer exists in a wide range of charge states (+6 to +16) and adopts many conformations in the gas phase with collision cross sections ranging from 12 nm² to 26 nm² (Figure 7—figure supplement 1). These observations are indicative of an unstructured protein with little secondary or tertiary structure (Beveridge et al., 2014). Thus, IM-MS studies highlight that higher order structure in EncFtn_sH is mediated/stabilized by metal binding, an observation that is in agreement with our solution studies. Taken together, these results suggest that di-iron binding, forming the FOC in EncFtn_sH, is required to stabilize the 4-helix bundle dimer interface, essentially reconstructing the classical ferritin-like fold; once stabilized, these dimers readily associate as pentamers, and the overall assembly adopts the decameric ring arrangement observed in the crystal structure.

Figure 7 with 2 supplements see all

Download asset Open asset

We subsequently performed gas phase disassembly of the decameric EncFtn_sH using collision-induced dissociation (CID) tandem mass spectrometry. Under the correct CID conditions, protein assemblies can dissociate with retention of subunit and ligand interactions, and thus provide structurally-informative evidence as to the topology of the original assembly; this has been termed ‘atypical’ dissociation (Hall et al., 2013). For EncFtn_sH, this atypical dissociation pathway was clearly evident; CID of the EncFtn_sH decamer resulted in the appearance of a dimeric EncFtn_sH subcomplex containing 0, 1, or 2 iron ions (Figure 7—figure supplement 2). In light of the crystal structure, this observation can be rationalized as dissociation of the EncFtn_sH decamer by disruption of the non-FOC interface with at least partial retention of the FOC interface and the FOC-Fe. Thus, this observation supports our crystallographic assignment of the overall topology of the EncFtn_sH assembly as a pentameric assembly of dimers with two iron ions located at the FOC dimer interface. In addition, this analysis provides evidence that the overall architecture of the complex is consistent in the crystal, solution and gas phases.

In light of the identification of an iron-loaded FOC in the crystal structure of EncFtn and our native mass spectrometry data, we performed ferroxidase and peroxidase assays to demonstrate the catalytic activity of this protein. In addition, we also assayed equine apoferritin, an example of a classical ferritin enzyme, as a positive control. Unlike the Dps family of ferritin-like proteins, EncFtn showed no peroxidase activity when assayed with the substrate ortho-phenylenediamine (Pesek et al., 2011). The ferroxidase activity of EncFtn_sH was measured by recording the progress curve of Fe²⁺ oxidation to Fe³⁺ at 315 nm after addition of 20 and 100 µM Fe²⁺ (2 and 10 times molar ratio Fe²⁺/FOC). In both experiments the rate of oxidation was faster than background oxidation of Fe²⁺ by molecular oxygen, and was highest for 100 µM Fe²⁺ (Figure 8A). These data show that recombinant EncFtn_sH acts as an active ferroxidase enzyme. When compared to apoferritin, EncFtn_sH oxidized Fe²⁺ at a slower rate and the reaction did not run to completion over the 1800 s of the experiment. Addition of higher quantities of iron resulted in the formation of a yellow/red precipitate at the end of the reaction. We also performed these assays on purified recombinant encapsulin; which, when assayed alone, did not display ferroxidase activity above background Fe²⁺ oxidation (Figure 8B). In contrast, complexes of the full EncFtn encapsulin nanocompartment (i.e. the EncFtn-Enc protein complex) displayed ferroxidase activity comparable to apoferritin without the formation of precipitates (Figure 8B).

Figure 8 with 1 supplement see all

Download asset Open asset

We attributed the precipitates observed in the EncFtn_sH ferroxidase assay to the production of insoluble Fe³⁺ complexes, which led us to propose that EncFtn does not directly store Fe³⁺ in a mineral form. This observation agrees with native MS results, which indicates a maximum iron loading of 10–15 iron ions per decameric EncFtn; and the structure, which does not possess the enclosed iron-storage cavity characteristic of classical ferritins and Dps family proteins that can directly accrue mineralized Fe³⁺ within their nanocompartment structures.

To analyze the products of these reactions and determine whether the EncFtn and encapsulin were able to store iron in a mineral form, we performed TEM on the reaction mixtures from the ferroxidase assay. The EncFtn_sH reaction mixture showed the formation of large, irregular electron-dense precipitates (Figure 8—figure supplement 1A). A similar distribution of particles was observed after addition of Fe²⁺ to the encapsulin protein (Figure 8—figure supplement 1B). In contrast, addition of Fe²⁺ to the EncFtn-Enc nanocompartment resulted in small, highly regular, electron dense particles of approximately 5 nm in diameter (Figure 8—figure supplement 1C); we interpret these observations as controlled mineralization of iron within the nanocompartment. Addition of Fe²⁺ to apoferritin resulted in a mixture of large particles and small (~2 nm) particles consistent with partial mineralization by the ferritin and some background oxidation of the iron (Figure 8—figure supplement 1D). Negative stain TEM of these samples revealed that upon addition of iron, the EncFtn_sH protein showed significant aggregation (Figure 8—figure supplement 1F); while the encapsulin, EncFtn-Enc system, and apoferritin are present as distinct nanocompartments without significant protein aggregation (Figure 8—figure supplement 1G–I).

The results of the ferroxidase assay and micrographs of the reaction products suggest that the oxidation and mineralization function of the classical ferritins are split between the EncFtn and encapsulin proteins, with the EncFtn acting as a ferroxidase and the encapsulin shell providing an environment and template for iron mineralization and storage. To investigate this further, we added Fe²⁺ at various concentrations to samples of apo-ferritin, EncFtn, isolated encapsulin, and the EncFtn-Enc protein complex, and subjected these samples to a ferrozine assay to quantify the amount of iron associated with the proteins after three hours of incubation. The maximum iron loading capacity of these systems was calculated as the quantity of iron per biological assembly (Figure 8C). In this assay, the EncFtn_sH decamer binds a maximum of around 48 iron ions before excess iron induces protein precipitation. The encapsulin shell protein can sequester about 2200 iron ions before significant protein loss occurs, and the reconstituted EncFtn-Enc nanocompartment sequestered about 4150 iron ions. This latter result is significantly more than the apoferritin used in our assay, which sequesters approximately 570 iron ions in this assay (Figure 8C, Table 5).

Consideration of the functional oligomeric states of these proteins, where EncFtn is a decamer and encapsulin forms an icosahedral cage, and estimation of the iron loading capacity of these complexes gives insight into the role of the two proteins in iron storage and mineralization. EncFtn decamers bind up to 48 iron ions (Figure 8C), which is significantly higher than the stoichiometry of fifteen metal ions visible in the FOC and E31/34-site of the crystal structure of the EncFtn_sH decamer and our MS analysis. The discrepancy between these solution measurements and our MS analysis may indicate that there are additional metal-binding sites on the interior channel and exterior faces of the protein; this is consistent with our identification of a number of weak metal-binding sites at the surface of the protein in the crystal structure (Figure 5D). These observations are consistent with hydrated Fe²⁺ ions being channeled to the active site from the E31/34-site and the subsequent exit of Fe³⁺ products on the outer surface, as is seen in other ferritin family proteins (Pesek et al., 2011; Behera and Theil, 2014). While the isolated encapsulin shell does not display any ferroxidase activity, it binds around 2200 iron ions in our assay (Table 5). This implies that the shell can bind a significant amount of iron on its outer and inner surfaces. While the maximum reported loading capacity of classical ferritins is approximately 4500 iron ions (Mann et al., 1986), in our assay system we were only able to load apoferritin with around 570 iron ions. However, the recombinant EncFtn-Enc nanocompartment was able to bind over 4100 iron ions in the same time period, over seven times the amount seen for the apoferritin. We note we do not reach the experimental maximum iron loading for apoferritin and therefore the total iron-loading capacity of our system may be significantly higher than in this experimental system.

Taken together, our data show that EncFtn can catalytically oxidize Fe²⁺ to Fe³⁺; however, iron binding in EncFtn is limited to the FOC and several surface metal binding sites. In contrast, the encapsulin protein displays no catalytic activity, but has the ability to bind a considerable amount of iron. Finally, the EncFtn-Enc nanocompartment complex retains the catalytic activity of EncFtn, and sequesters iron within the encapsulin shell at a higher level than the isolated components of the system, and at a significantly higher level than the classical ferritins (Andrews, 2010). Furthermore, our recombinant nanocompartments may not have the physiological subunit stoichiometry, and the iron-loading capacity of native nanocompartments is potentially much higher than the level we have observed.

To investigate the structural and biochemical role played by the metal binding residues in the di-iron FOC of EncFtn_sH we produced alanine mutations in each of these residues: Glu32, Glu62, and His65. These EncFtn_sH mutants were produced in E. coli cells grown in MM, both in the absence and presence of additional iron. The E32A and E62A mutants eluted from SEC at a volume consistent with the decameric form of EncFtn_sH, with a small proportion of monomer; the H65A mutant eluted at a volume consistent with the monomeric form of EncFtn_sH (Figure 9). For all of the mutants studied, no change in oligomerization state was apparent upon addition of Fe²⁺ in vitro.

Figure 9

Download asset Open asset

In addition to SEC studies, native mass spectrometry of the apo-EncFtn_sH mutants was performed and compared with the wild-type apo-EncFtn_sH protein (Figure 10). As described above, the apo-EncFtn_sH has a charge state distribution consistent with an unstructured monomer, and decamer formation is only initiated upon addition of ferrous iron. Both the E32A mutant and E62A mutant displayed charge state distributions consistent with decamers, even in the absence of Fe²⁺. This gas-phase observation is consistent with SEC measurements, which indicate both of these variants were also decamers in solution. Thus it seems that these mutations allow the decamer to form in the absence of iron in the FOC. In contrast to the glutamic acid mutants, MS analysis of the H65A mutant is similar to wild-type apo-EncFtn_sH and is present as a monomer; interestingly a minor population of dimeric H65A was also observed.

Figure 10

Download asset Open asset

We propose that the observed differences in the oligomerization state of the E32A and E62A mutants compared to wild-type are due to the changes in the electrostatic environment within the FOC. At neutral pH the glutamic acid residues are negatively charged, while the histidine residues are predominantly in their uncharged state. In the wild-type (WT) EncFtn_sH this leads to electrostatic repulsion between subunits in the absence of iron. Coordination of Fe²⁺ in this site stabilizes the dimer and reconstitutes the active FOC. The geometric arrangement of Glu32 and Glu62 in the FOC explains their behavior in solution and the gas phase, where they both favor the formation of decamers due to the loss of a repulsive negative charge. The FOC in the H65A mutant is destabilized through the loss of this metal coordinating residue and potential positive charge carrier, thus favoring the monomer in solution and the gas phase.

To understand the impact of the mutants on the organization and metal binding of the FOC, we determined the X-ray crystal structures of each of the EncFtn_sH mutants (See Table 4 for data collection and refinement statistics). The crystal packing of all of the mutants in this study is essentially isomorphous to the EncFtn_sH structure. All of the mutants display the same decameric arrangement in the crystals as the EncFtn_sH structure, and the monomers superimpose with an average RMSD_Cα of less than 0.2 Å.

Close inspection of the region of the protein around the FOC in each of the mutants highlights their effect on metal binding (Figure 11 and Figure 11—figure supplement 1–3). In the E32A mutant the position of the side chains of the remaining iron coordinating residues in the FOC is essentially unchanged, but the absence of the axial-metal coordinating ligand provided by the Glu32 side chain abrogates metal binding in this site. The Glu31/34-site also lacks metal, with the side chain of Glu31 rotated by 180° at the C_β in the absence of metal (Figure 11—figure supplement 1). The E62A mutant has a similar effect on the FOC to the E32A mutant, however the entry site still has a calcium ion coordinated between residues Glu31 and Glu34 (Figure 11—figure supplement 2). The H65A mutant diverges significantly from the wild type in the position of the residues Glu32 and Tyr39 in the FOC. E32 appears in either the original orientation as the wild type and coordinates Ca²⁺ in this position, or it is flipped by 180° at the C_β, moving away from the coordinated calcium ion in the FOC. Tyr39 moves closer to Ca²⁺ compared to the wild-type and coordinates the calcium ion (Figure 11—figure supplement 3). A single calcium ion is present in the entry site of this mutant; however, Glu31 of one chain is rotated away from the metal ion and is not involved in coordination.

Figure 11 with 3 supplements see all

Download asset Open asset

Taken together the results of our data show that these changes to the FOC of EncFtn still permit the formation of the decameric form of the protein. While the proteins all appear decameric in crystals, their solution and gas-phase behavior differs considerably and the mutants no longer show metal-dependent oligomerization. These results highlight the importance of metal coordination in the FOC for the stability and assembly of the EncFtn protein.

To address the question of how mutagenesis of the iron coordinating residues affects the enzymatic activity of the EncFtn_sH protein we recorded progress curves for the oxidation of Fe²⁺ to Fe³⁺ by the different mutants as before. Mutagenesis of E32A and H65A reduces the activity of EncFtn_sH by about 40%-55%; the E62A mutant completely abrogates activity, presumably through the loss of the bridging coordination for the formation of the di-nuclear iron center of the FOC (Figure 12). Collectively, the effect of mutating these residues in the FOC confirms the importance of the iron coordinating residues for the ferroxidase activity of the EncFtn_sH protein.

Figure 12 with 1 supplement see all

Download asset Open asset

The quaternary arrangement of classical ferritins into an octahedral nanocage and Dps into a dodecamer is absolutely required for their function as iron storage compartments (Chasteen and Harrison, 1999). The oxidation and mineralization of iron must be spatially separated from the host cytosol to prevent the formation of damaging hydroxyl radicals in the Fenton and Haber-Weiss reactions (Honarmand Ebrahimi et al., 2012). This is achieved in all ferritins by confining the oxidation of iron to the interior of the protein complex, thus achieving sequestration of the Fe³⁺ mineralization product. A structural alignment of the FOC of EncFtn with the classical ferritin PmFtn shows that the central ring of EncFtn corresponds to the external surface of ferritin, while the outer circumference of EncFtn is congruent with the inner mineralization surface of ferritin (Figure 6—figure supplement 1A). This overlay highlights the fact that the ferroxidase center of EncFtn faces in the opposite direction relative to the classical ferritins and is essentially inside out regarding iron storage space (Figure 6—figure supplement 1B, boxed region). Analysis of each of the single mutations (E32A, E62A and H65A) made in the FOC highlights the importance of the iron-coordinating residues in the catalytic activity of EncFtn. Furthermore, the position of the calcium ion coordinated by Glu31 and Glu34 seen in the EncFtn_sH structure suggests an entry site to channel metal ions into the FOC; we propose that this site binds hydrated iron ions in vivo and acts as a selectivity filter and gate for the FOC (Haldar et al., 2011). The constellation of charged residues on the outer circumference of EncFtn (His57, Glu61 and Glu64) could function in the same way as the residues lining the mineralization surface within the classical ferritin nanocage (Le Brun et al., 2010), and given their proximity to the FOC these sites may be the exit portal and mineralization site (Honarmand Ebrahimi et al., 2012).

The absolute requirement for the spatial separation of oxidation and mineralization in ferritins suggests that the EncFtn family proteins are not capable of storing iron minerals due to the absence of an enclosed compartment in their structure (Figure 6—figure supplement 1B). Our biochemical characterization of EncFtn supports this hypothesis, indicating that while this protein is capable of oxidizing iron, it does not accrue mineralized iron in an analogous manner to classical ferritins. While EncFtn does not store iron itself, its association with the encapsulin nanocage suggests that mineralization occurs within the cavity of the encapsulin shell (McHugh et al., 2014). Our ferroxidase assay data on the recombinant EncFtn-Enc nanocompartments, which accrue over 4100 iron ions per complex and form regular nanoparticles, are consistent with the encapsulin protein acting as the store for iron oxidized by the EncFtn enzyme. TEM analysis of the reaction products shows the production of homogeneous iron nanoparticles only in the EncFtn-Enc nanocompartment (Figure 8—figure supplement 1).

Docking the decamer structure of EncFtn_sH into the pentamer of the T. maritima encapsulin Tmari_0786 (PDB ID: 3DKT) (Sutter et al., 2008) shows that the position of the C-terminal extensions of our EncFtn_sH structure are consistent with the localization sequences seen bound to the encapsulin protein (Figure 14A). Thus, it appears that the EncFtn decamer is the physiological state of this protein. This arrangement positions the central ring of EncFtn directly above the pore at the five-fold symmetry axis of the encapsulin shell and highlights a potential route for the entry of iron into the encapsulin and towards the active site of EncFtn. A comparison of the encapsulin nanocompartment and the ferritin nanocage highlights the size differential between the two complexes (Figure 14B) that allows the encapsulin to store significantly more iron. The presence of five FOCs per EncFtn_sH decamer and the fact that the icosahedral encapsulin nanocage can hold up to twelve of decameric EncFtn between each of the internal five-fold vertices means that they can achieve a high rate of iron mineralization across the entire nanocompartment. This arrangement of multiple reaction centers in a single protein assembly is reminiscent of classical ferritins, which has 24 FOCs distributed around the nanocage.

Figure 14

Download asset Open asset

Our structural data, coupled with biochemical and ICP-MS analysis, suggest a model for the activity of the encapsulin iron-megastore (Figure 14C). The crystal structure of the T. maritima encapsulin shell protein has a negatively charged pore positioned to allow the passage of Fe²⁺ into the encapsulin and directs the metal towards the central, negatively charged hole of the EncFtn ring (Figure 4—figure supplement 1). The five metal-binding sites on the interior of the ring (Glu31/34-sites) may select for the Fe²⁺ ion and direct it towards their cognate FOCs. We propose that the oxidation of Fe²⁺ to Fe³⁺ occurs within the FOC according to the model postulated by (Honarmand Ebrahimi et al., 2012) in which the FOC acts as a substrate site through which iron passes and is released on to weakly coordinating sites at the outer circumference of the protein (His57, Glu61 and Glu64), where it is able to form ferrihydrite minerals which can be safely deposited within the lumen of the encapsulin nanocompartment (Figure 14).

Here we describe for the first time the structure and biochemistry of a new class of encapsulin-associated ferritin-like protein and demonstrate that it has an absolute requirement for compartmentalization within an encapsulin nanocage to act as an iron store. Further work on the EncFtn-Enc nanocompartment will establish the structural basis for the movement of iron through the encapsulin shell, the mechanism of iron oxidation by the EncFtn FOC and its subsequent storage in the lumen of the encapsulin nanocompartment.

Genes of interest were amplified by PCR using R. rubrum ATCC 11,170 genomic DNA (DSMZ) as the template and KOD Hot Start DNA Polymerase (Novagen). Primers used in this study are listed in Supplementary file 2. PCR products were visualized in 0.8% agarose gel stained with SYBR Safe (Life Technologies, UK). Fragments of interest were purified by gel extraction (Qiagen, UK) before digestion by endonuclease restriction enzymes (Thermo Fisher Scientific, UK) at 37°C for 1 hr, followed by ligation with similarly digested vector pET-28a(+) or pACYCDuet-1 at room temperature for 1 hr. Ligation product was transformed into chemically competent Escherichia coli Top10 cells and screened against 50 ng/μl kanamycin for pET-28a(+) based constructs or 34 ng/μl chloramphenicol for pACYCDuet-1 based constructs. DNA insertion was confirmed through Sanger sequencing (Edinburgh Genomics, The University of Edinburgh, UK). Sequence verified constructs were transformed into E. coli BL21(DE3) or Tuner(DE3) for protein production. Alternatively, plasmids transformed into E. coli B834(DE3) cells were cultured in selenomethionine medium.

A single colony of E. coli BL21(DE3) or Tuner(DE3) cells, transformed with protein expression plasmid, was transferred into 10 ml LB medium, or M9 minimal medium (MM), supplemented with appropriate antibiotic, and incubated overnight at 37 °C with 200 rpm shaking. The overnight pre-culture was then inoculated into 1 liter of LB medium and incubated at 37 °C with 200 rpm shaking. Recombinant protein production was induced at OD₆₀₀= 0.6 by the addition of 1 mM IPTG and the incubation temperature was reduced to 18°C for overnight incubation. Cells were pelleted by centrifugation at 4000 g for 20 min at 4 °C, and resuspended 10-fold (volume per gram of cell pellet) in PBS to wash cells before a second centrifugation step. Cells were resuspended in 10-times (v/w) of appropriate lysis buffer for the purification method used (see details of buffers below) and lysed by sonication on ice, with ten cycles of 30-second burst of sonication at 10 µm amplitude and 30 s of cooling. Cell lysate was clarified by centrifugation at 20,000 x g, 30 min, 4 °C; followed by filtration using a 0.22 µM syringe filter (Millipore, UK).

Selenomethionine labelled protein was produced by growing a single colony of E. coli B834 (DE3) cells transformed with protein expression plasmids in 100 ml LB medium supplemented with appropriate antibiotic overnight at 37 °C with shaking at 200 rpm. The overnight pre-culture was pelleted by centrifugation 3,000 x g, 4 °C, 15 min and washed twice with M9 minimal medium. The washed cells were transferred to 1 liter of SeMet medium, which contains M9 minimal medium, 40 mg/L of each _L-amino acid (without methionine), 40 mg/L selenomethionine, 2 mM MgSO₄, 0.4% (w/v) glucose and 1 mM Fe(NH₄)₂(SO₄)₂. Cells were incubated at 37 °C with 200 rpm shaking and recombinant protein production was induced at OD₆₀₀= 0.6 by the addition of 1 mM IPTG and the incubation temperature was reduced to 18 °C for overnight incubation. Cells were harvested and lysed as above.

Clarified cell lysate was loaded onto a 5 ml HisTrap column (GE Healthcare, UK) pre-equilibrated with HisA buffer (50 mM Tris-HCl, 500 mM NaCl and 50 mM imidazole, pH 8.0). Unbound proteins were washed from the column with HisA buffer. His-tagged proteins were then eluted by a step gradient of 50% HisA buffer and 50% HisB buffer (50 mM Tris-HCl, 500 mM NaCl and 500 mM imidazole, pH 8.0). Fractions containing the protein of interest, as determined by 15% (w/v) acrylamide SDS-PAGE, were pooled before loading onto a gel-filtration column (HiLoad 16/600 Superdex 200, GE Healthcare) equilibrated with GF buffer (50 mM Tris-HCl, pH 8.0, 150 mM NaCl). Fractions were subjected to 15% SDS-PAGE and those containing the protein of interest were pooled for further analysis.

Co-expressed encapsulin and EncFtn (EncFtn-Enc) and encapsulin protein were both purified according to the protocol used by M. Sutter (Sutter, 2008). Briefly, EncFtn-Enc or encapsulin was expressed based on pACYCDuet-1 vector. The E. coli cells were grown, induced, harvested and sonicated in a similar way as described above. GF buffer used in this purification contains 50 mM Tris-HCl, pH 8.0, and 150 mM NaCl. To remove RNA contamination, the lysate was supplemented with 50 μg/ml RNase A and rotated at 10 rpm and room temperature for 2 hrs, followed by centrifugation at 34,000 x g and 4 °C for 20 min and filtering through 0.22 µM syringe filter. Proteins were pelleted through 38% (w/v) sucrose cushion by ultracentrifugation at 100,000 x g and 4 °C for 21 hrs. 10% - 50% (w/v) sucrose gradient ultracentrifugation was applied to further separate the proteins at 100,000 x g and 4 °C for 17 hrs. Protein was dialyzed against GF buffer to remove sucrose before being used in chemical assays or TEM.

TEM imaging was performed on purified encapsulin, EncFtn, and EncFtn-Enc and apoferritin. Purified protein at 0.1 mg/ml concentration was spotted on glow-discharged 300 mesh carbon-coated copper grids and excess liquid wicked off with filter paper (Whatman, UK). The grids were washed with distilled water and blotted with filter paper three times before staining with 0.2% uranyl acetate, blotting and air-drying. Grids were imaged using a JEM1400 transmission electron microscope and images were collected with a Gatan CCD camera. Images were analyzed using ImageJ (NIH, Bethesda, MD) and size-distribution histograms were plotted using Prism 6 (GraphPad software). To observe iron mineral formation by TEM, protein samples at 8.5 µM concentration including EncFtn_sH, encapsulin, EncFtn-Enc and apoferritin were supplemented with acidic Fe(NH₄)₂(SO₄)₂ at their maximum iron loading ratio in room temperature for 1 hr. The mixtures were subjected to TEM analysis with or without uranyl acetate staining. TEM experiments without Fe loading were repeated three times, a representative set of images are presented here. Proteins loaded with Fe and imaged by TEM were from single preparation.

EncFtn_sH was purified by anion exchange and Superdex 200 size- exclusion chromatography and concentrated to 10 mg/ml (based on extinction coefficient calculation). Crystallization drops were set up using the hanging drop vapor diffusion method at 292 K. Glass coverslips were set up with 1–2 μl protein mixed with 1 μl well solution (0.14 M calcium acetate and 15% (w/v) PEG 3350) and sealed over 1 ml of well solution. Crystals appeared after 5 days and were harvested from the well using a LithoLoop (Molecular Dimensions Limited, UK), transferred briefly to a cryoprotection solution containing well solution supplemented with 1 mM FeSO₄ (in 0.1% (v/v) HCl), 20% (v/v) PEG 200, and subsequently flash cooled in liquid nitrogen. Crystals of the EncFtn_sHsingle mutations were produced in the same manner as for the EncFtn_sH wild-type protein.

All crystallographic datasets were collected on the macromolecular crystallography beamlines at Diamond Light Source (Didcot, UK) at 100 K using Pilatus 6M detectors. Diffraction data were integrated and scaled using XDS (Kabsch, 2010) and symmetry related reflections were merged with Aimless (Evans, 2011). Data collection statistics are shown in Table 4. The resolution cut-off used for structure determination and refinement was determined based on the CC_1/2 criterion proposed by Karplus and Diederichs (2012).

The structure of EncFtn_sH was determined by molecular replacement using PDB ID: 3K6C as the search model, modified to match the sequence of the target protein using Chainsaw (Stein, 2008). A single solution comprising three decamers in the asymmetric unit was found by molecular replacement using Phaser (McCoy et al., 2007). The initial model was rebuilt using Phenix.autobuild (Adams et al., 2010) followed by cycles of refinement with Phenix.refine (Afonine et al., 2012), with manual rebuilding and model inspection in Coot (Emsley et al., 2010). The final model was refined with isotropic B-factors, torsional NCS restraints, and with anomalous group refinement. The model was validated using MolProbity (Chen et al., 2010). Structural superimpositions were calculated using Coot. Crystallographic figures were generated with PyMOL. Multiple sequence alignment of EncFtn and ferritin family proteins was performed using Clustal Omega Sievers and Higgins, 2014 and displayed with Espript 3.0 (Gouet et al., 2003). Model refinement statistics are shown in Table 4. The final models and experimental data are deposited in the PDB and diffraction image files are available at the Edinburgh DataShare repository.

Horse spleen apoferritin purchased from Sigma Aldrich (UK) was dissolved in deaerated MOPS buffer (100 mM MOPS, 100 mM NaCl, 3 g/100 ml Na₂S₂O₄ and 0.5 M EDTA, pH 6.5) (Bauminger et al., 1991). Protein was dialyzed against 1 liter MOPS buffer in room temperature for two days before buffer exchanging to GF buffer (50 mM Tris-HCl, pH 8.0, 150 mM NaCl) in a vivaspin column with 5 kDa cut-off (Sartorius, UK) for several times. Fe content of apoferritin was detected using ferrozine assay (Riemer et al., 2004). Protein concentration was determined using Pierce Microplate BCA Protein Assay Kit. Apoferritin containing less than 0.5 Fe per 24-mer was used in the ferroxidase assay. Apoferritin used in the Fe loading capacity experiment was prepared in the same way with 5–15 Fe per 24-mer.

1 mM and 200 µM Fe(NH₄)₂(SO₄)₂ stock solutions were prepared in 0.1% (v/v) HCl anaerobically. Protein solutions with 20 µM FOC were diluted from ~10 mg/ml frozen stock in GF buffer (50 mM Tris-HCl, pH 8.0 and 150 mM NaCl) anaerobically. Ferroxidase activity was initiated by adding 450 μl protein to 50 μl of acidic Fe(NH₄)₂(SO₄)₂ at the final concentration of 100 µM and 20 µM in the air, respectively. The ferroxidase activity was measured by monitoring the Fe³⁺ formation which gives rise to the change of the absorbance at 315 nm (Bonomi et al., 1996). Absorbance at 315 nm was recorded every second over 1800 s using a quartz cuvette in a JASCO V-730 UV/VIS spectrophotometer (JASCO Inc., Easton, MD). In recombinantly coexpressed nanocompartments the ratio of EncFtn to Enc was assumed as 2 to 1, assuming each of the twelve pentameric vertices of the icosahedral encapsulin were occupied with decameric EncFtn. The data are presented as the mean of three technical replicates with error bars indicating one standard deviation from the mean. Proteins used here were from a single preparation.

In order to determine the maximum iron loading capacity, around 8.5 µM proteins including decameric EncFtn_sH, Encapsulin, EncFtn-Enc and apoferritin were loaded with various amount of acidic Fe(NH₄)₂(SO₄)₂ ranging from 0 to 1700 µM. Protein mixtures were incubated in room temperature for 3 hrs before desalting in Zebra spin desalting columns (7 kDa cut-off, Thermo Fisher Scientific, UK) to remove free iron ions. The protein concentration was determined using PierceMicroplate BCA assay kit (Thermo Fisher Scientific). The protein standard curve was plotted according to the manufacturer. The Fe content in the samples was determined using modified ferrozine assay (Riemer et al., 2004). Briefly speaking, 100 μl protein sample was mixed with 100 μl mixture of equal volume of 1.4 M HCl and 4.5% (w/v) KMnO₄ and incubated at 60 °C for 2 hrs. 20 μl of the iron-detection reagent (6.5 mM ferrozine, 6.5 mM neocuproine, 2.5 M ammonium acetate, and 1 M ascorbic acid dissolved in H₂O) was added to the cooled tubes. 30 min later, 200 μl of the solution was transferred into a well of 96-well plate and the absorbance at 562 nm was measured on the plate reader Spectramax M5 (Molecular Devices, UK). The standard curve was plotted using various concentrations of FeCl₃ (in 10 mM HCl) diluted in the gel-filtration buffer. Three technical repeats were performed for both the ferrozine and microplate BCA assays. Samples analyzed by ICP-MS were prepared in the same way by mixing protein and ferrous ions and desalting. The proteins used in the Fe loading experiment came from a single preparation.

The peroxidase activity of EncFtn_sH was determined by measuring the oxidation of ortho-phenylenediamine (OP) by H₂O₂ Pesek et al. (2011). EncFtn_sH decameric and monomeric fractions purified from MM were both used in the assay. Ortho-phenylenediamine was prepared as a 92.5 mM stock solution in 50 mM Tris-HCl (pH 8.0). 80, 70, 60, 50, 40, 30, 20 and 10 mM of OP were prepared by diluting the stock solution in the 50 mM Tris-HCl (pH 8.0). 100 μl of each diluted OP was added to a 96-well plate in 3 repeats. 1 μl of 32 µM protein was supplemented into each well to a final concentration of 160 nM, followed by the addition of 2 μl of 30% H₂O₂. After 15 min shaking in the dark, the reaction was stopped by adding 100 μl of 0.5 M H₂SO₄. The peroxidase activity was measured by monitoring the absorbance at 490 nm in the SpectraMax M5 Microplate Reader (Molecular Devices) (Pesek et al., 2011).

Protein samples were diluted 50-fold into a solution of 2.5% HNO₃ (Suprapur, Merck, UK) containing 20 µg/L Pt as internal standard. Matrix-matched elemental standards (containing analyte metal concentrations 0 – 1000 µg/L) were prepared by serial dilution from individual metal standard stocks (VWR) with identical solution compositions, including the internal standard. All standards and samples were analyzed by ICP-MS using a Thermo x-series instrument (Thermo Fisher Scientific) operating in collision cell mode (using 3.0 ml min^-1 flow of 8% H₂ in He as the collision gas). Isotopes ⁴⁴Ca, ⁵⁶Fe, ⁶⁶Zn, ⁷⁸Se, and ¹⁹⁵Pt were monitored using the peak-jump method (100 sweeps, 25–30 ms dwell time on 5 channels per isotope, separated by 0.02 atomic mass units) in triplicate. The protein samples used in ICP-MS came from a single protein preparation.

For native MS analysis, all protein samples were buffer exchanged into 100 mM ammonium acetate (pH 8.0; adjusted with dropwise addition of 1% ammonia solution) using Micro Biospin Chromatography Columns (Bio-Rad, UK) prior to analysis and the resulting protein samples were analyzed at a final concentration of ~5 µM (oligomer concentration). In order to obtain Fe-bound EncFtn, 100 µM or 300 µM of freshly prepared FeCl₂ was added to apo-EncFtn_sH (monomer peak) immediately prior to buffer exchange into 100 mM ammonium acetate (pH 8.0). Samples were analyzed on a quadrupole ion-mobility time of flight instrument (Synapt G2, Waters Corp., Manchester, UK), equipped with a nanomate nanoelectrospray infusion robot (Advion Biosciences, Ithaca, NY). Instrument parameters were tuned to preserve non-covalent protein complexes. After optimization, typical parameters were: nanoelectrospray voltage 1.54 kV; sample cone 50 V; extractor cone 0 V; trap collision voltage 4 V; source temperature 80°C; and source backing pressure 5.5 mbar. For improved mass resolution the sample cone was raised to 155 V. Ion mobility mass spectrometry (IM-MS) was performed using the travelling-wave mobility cell in the Synapt G2, employing nitrogen as the drift gas. Typically, the IMS wave velocity was set to 300 m/s; wave height to 15 V; and the IMS pressure was 1.8 mbar. All native MS experiments were performed on samples from two independent protein preparations. For collision cross section determination, IM-MS data was calibrated using denatured equine myoglobin and data was analyzed using Driftscope v2.5 and MassLynx v4.1 (Waters Corp., UK). Theoretical collision cross sections (CCS) were calculated from pdb files using IMPACT software v. 0.9.1 (Marklund, 2015). In order to obtain information on the topology of the EncFtn_sH assembly, gas-phase dissociation of the Fe-associated EncFtn_sH complex was achieved by increasing the sample cone and/or trap collision voltage prior to MS analysis.

Size-exclusion chromatography (ÄKTA-Micro; GE Healthcare) coupled to UV, static light scattering and refractive index detection (Viscotec SEC-MALS 20 and Viscotec RI Detector:VE3580; Malvern Instruments, UK) were used to determine the molecular mass of fractions decamer and monomer of EncFtn_sH in solution individually. Protein concentration was determined by measurement of absorbance at 280 nm and calculated using the extinction coefficient ε0.1%= 1.462 mg⁻¹ ml^-1 cm⁻¹. 100 μl of 1.43 mgml^-1 fractions of EncFtn_sH decamer and 4.03 mg ml ^-1 fractions of EncFtn_sH monomer were run individually on a Superdex 200 10/300 GL size-exclusion column pre-equilibrated in 50 mM Tris-HCl (pH 8.0), 150 mM NaCl at 22°C with a flow rate of 0.5 ml/min. Light scattering, refractive index (RI) and A_280nm were analyzed by a homo-polymer model (OmniSEC software, v 5.1; Malvern Instruments) using the following parameters for fractions of decamer and monomer: the extinction coefficient (dA/dc) at 280 nm was 1.46 AU mg ml⁻¹ and specific refractive index increment (dn/dc) was 0.185 ml g⁻¹. The proteins analyzed by SEC-MALLS came from single protein preparation.

Recombinant EncFtn_sH fractions at 50 µM concentration were incubated with one molar equivalent of metal ions at room temperature for 2 hrs. Half of each sample was mixed with 5 x native loading buffer (65 mM Tris-HCl, pH 8.5, 20% glycerol and 0.01% bromophenol blue) and run on non-denaturing PAGE gels (10% acrylamide) and run in Tris/glycine buffer, 200 V, 4 °C for 50 min. The remaining samples were left for an additional three hours prior to SDS-PAGE (15% acrylamide) analysis. SDS-PAGE gels were run at room temperature at 200 V, room temperature for 50 min. Gels were stained with Coomassie Brilliant Blue R250 and scanned after de-staining in water. The proteins used in this experiment came from single protein preparation.

For analysis of the multimeric state of EncFtn proteins by analytical size-exclusion gel-filtration chromatography (AGF) 25 μl of 90 µM protein was loaded into Superdex 200 PC 3.2/30 column (GE Healthcare) at 15 °C with GF buffer running at 0.05 ml/min and pressure limit 0.45 MPa. In order to use AGF to determine how metal ions influence the assembly of EncFtn_sH, 90 µM EncFtn_sH monomer fractions were mixed with equal molar concentrations of metal ion solutions including FeSO₄ in 0.1% (v/v) HCl, Fe(NH₄)₂(SO₄)₂, FeCl₃, CoCl₂, calcium acetate (CaAc), ZnSO₄ and MnCl₂ at room temperature for 2 hrs prior to AGF analysis. Protein samples without metal titration were also analyzed as a control group. Both monomer and decamer fractions of EncFtn_sH left at room temperature for 2 hrs, or overnight, were also analysed as controls to show the stability of the protein samples in the absence of additional metal ions. The AGF results have been repeated twice using two independent preparations of protein, of which only one representative trace is presented in the paper.

Coordinates and structure factors for the structures presented in this paper have been deposited in the PDB under the following accession codes: EncFtn_sH, 5DA5; EncFtn_sH-E32A, 5L89; EncFtn_sH-E62A, 5L8B; EncFtn_sH-H65A, 5L8G (DOIs for X-ray diffraction image data are shown in Table 4). All MS datasets presented in this paper can be found, in the raw format at http://dx.doi.org/10.7488/ds/1449.

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

This work was supported by a Royal Society Research Grant awarded to JMW (RG130585), a BBSRC New Investigator Grant to JMW and DJC (BB/N005570/1) and a Wellcome Trust Institute Strategic Support Fund grant awarded to set up the XtalPod crystallization facility at the University of Edinburgh. DJC, SH, and KA are funded by the University of Edinburgh. DH is funded by the China Scholarship Council. KJW and ET are funded by the Wellcome Trust and Royal Society through a Sir Henry Dale Fellowship awarded to KJW (098375/Z/12/Z).

We would like to thank the staff on the Macromolecular Crystallography beamlines at Diamond Light Source for their assistance with data collection. We would like to thank Dr Steve Mitchell in the University of Edinburgh and Electron Microscopy Research Services in Newcastle University for the use of Transmission Electron Microscopes. We would like to thank Dr Michael Capeness and Dr. Louise Horsfall for the use of their anaerobic chamber. We would like to thank staff in the Edinburgh Protein Production Facility for their guidance and patience, Dr Janice Brahman, Dr Liz Blackburn, Dr Matthew Nowicki and Dr Martin Wear. We would like to thank Professor Rick Lewis and Dr Arnaud Basle (Newcastle University, UK), and Dr John Berrisford (PDBe, EMBL-EBI, Hinxton, Cambridge) for help identifying the glycolic acid ligand in the ferroxidase center. We would like to thank Prof Dominic Campopiano, Dr Elisabeth Lowe and Laura Tuck for their critical reading of this manuscript and helpful discussions.

© 2016, He et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.