Second Coordination Sphere Effects on the Mechanistic Pathways for Dioxygen Activation by a Ferritin: Involvement of a Tyr Radical and the Identification of a Cation Binding Site

Abstract Ferritins are ubiquitous diiron enzymes involved in iron(II) detoxification and oxidative stress responses and can act as metabolic iron stores. The overall reaction mechanisms of ferritin enzymes are still unclear, particularly concerning the role of the conserved, near catalytic center Tyr residue. Thus, we carried out a computational study of a ferritin using a large cluster model of well over 300 atoms including its first‐ and second‐coordination sphere. The calculations reveal important insight into the structure and reactivity of ferritins. Specifically, the active site Tyr residue delivers a proton and electron in the catalytic cycle prior to iron(II) oxidation. In addition, the calculations highlight a likely cation binding site at Asp65, which through long‐range electrostatic interactions, influences the electronic configuration and charge distributions of the metal center. The results are consistent with experimental observations but reveal novel detail of early mechanistic steps that lead to an unusual mixed‐valent iron(III)‐iron(II) center.


Introduction
O 2 activating metalloenzymes are common catalysts in biology and found in all forms of life. Many of these enzymes utilize mononuclear iron in their active site and react as either monooxygenases or dioxygenases. [1,2] In addition to these mononuclear iron enzymes, there are a number of nonheme diiron enzymes that activate O 2 in their catalytic cycles. [3] Among these oxygen activating nonheme diiron enzymes are the soluble methane monooxygenases involved in the hydroxylation of methane in soil, [4] as well as the aldehyde deformylating dioxygenases, [5] and the toluene monooxygenases. [6] Another O 2 activating nonheme diiron enzyme is ferritin, which has various biological roles and has been implicated in iron(II) detoxification and protection against oxidative stress, [7] in addition to providing an accessible store of iron under conditions of limitation. Ferritins are widely distributed across all kingdoms of life due to their importance in iron homeostasis. [8] Animal ferritins are composed of a mixture of Hand L-chain subunits that co-assemble to produce a heteropolymeric protein cage in which the H-chains harbor a diiron site, the ferroxidase center, and the L-chains do not. [7a,9] Ferritins isolated from other organisms are polymers containing only Hchain like subunits. [10] Structural models derived from high resolution X-ray diffraction data are now available for many ferritins. These reveal an identical 4 α-helical fold in all examples and similar coordination of the metal ions at the ferroxidase center in the H-chain proteins, illustrated in Figure 1a. The two ions are bridged by the carboxylate group of a glutamate residue (Glu 66 ). Studies in which ferrous iron exposure times were varied identified one high-affinity (Fe1) and one low-affinity iron site (Fe2). [11] The coordination of the high affinity Fe1 site is completed by the sidechain of a second carboxylate (Glu 33 ), a histidine (His 69 ) and a water molecule. The lower affinity Fe2 site is coordinated only by a second carboxylate (Glu 110 ) and a water molecule. A strictly conserved tyrosine residue (Tyr 40 ) is located close to, but not within bonding distance of, Fe2 and has been identified as important in the mechanism of several ferritins. [12] The similarity of protein structure together with the identification of a common reaction intermediate and product [13] has been invoked to argue in favor of a 'universal mechanism' of ferritin activity involving 2 reaction pathways. [14] In pathway 1, dioxygen binds to a diiron(II) center and is converted into H 2 O 2 , concomitant with oxidation of the diiron center to Fe 3 + (Figure 1b). An alternative dioxygen activation channel (through pathway 2, Figure 1b) requires binding and oxidation of 3 equivalents of Fe 2 + together with oxidation of the active site Tyr residue (Tyr 40 ) coupled to the reduction of dioxygen to water. However, this proposal remains controversial for several reasons. Prokaryotic ferritins contain a third iron binding site, site C, close to the catalytic center that is clearly defined in crystal structures of proteins exposed to iron. [15] Disruption of this binding site by mutagenesis has been shown to affect catalysis in prokaryotic ferritins [16] and the binding of a third equivalent of iron is invoked in the universal mechanism under high iron loadings. In the H-chain like proteins of prokaryotes three conserved glutamate residues act as ligands to site C. One of these acts as a bridging ligand between site C and the low affinity site of the ferroxidase center. In contrast there is considerable variation in the equivalent residue of the true H-chains, with a non-coordinating side chain at this position in many instances ( Figure S1, Supporting Information). It has been shown that the site C of prokaryotic ferritins can be disrupted by using site directed mutagenesis to introduce a non-coordinating residue in place of Glu 52 calling into question the existence of such a site in the animal H-chains. [16,17] In the universal mechanism both site C and the conserved tyrosine residue are proposed to play identical roles in all ferritins. [12c] However, the consequences of substitution of Tyr for the non-oxidizable Phe or of a site C ligand for a noncoordinating residue vary between proteins. It is also known that the stoichiometry of ferritin catalyzed iron-O 2 reaction varies between the extremes of 2 : 1 and 4 : 1 for different proteins. [10] Consequently the view persists that a degree of mechanistic diversity exists within the ferritin family, perhaps reflecting variation in primary role between iron storage and oxidative stress response but also the environmental niche occupied by the organism in which the protein was identified. [7b] SynFtn is a ferritin isolated from the prokaryotic marine cyanobacterium Synechoccocus sp. CC9311 but bears some resemblance to the H-chain subunits of the animal proteins. Specifically the three glutamate residues that comprise site C are absent from the peptide chain with a serine residue, Ser 146 , in place of the glutamate that bridges site C and the low affinity site of the ferroxidase center in other prokaryotic ferritins. Structural models derived from X-ray diffraction data from iron enriched crystals revealed only two areas of electron density associated with metal ion binding at SynFtn ferroxidase centers and that Ser 146 does not act as a ligand to the low affinity site. [12a] Therefore, the first coordination sphere of iron bound to the ferroxidase center of SynFtn is identical to that reported for animal H-chains. Indeed, one of the PDB entries for iron enriched SynFtn crystals, 6GKA, [12a,18] was used to produce the illustration in Figure 1. The first detectable intermediate in SynFtn reactivity is a previously unreported mixed valent iron(II)/iron(III) ferroxidase center together with a protein based radical assigned to the side chain of Tyr 40 . The diiron(III)-peroxo species thought to be common to all other ferritins and related diiron enzymes [19] was not reported as a reaction intermediate for this protein. SynFtn reactivity therefore appears to contradict one of the central arguments in favor of a universal mechanism of ferritin function -similarity in structure inevitably results in similarity of reaction mechanism.
Several computational studies on ferritin and analogous non-heme O 2 -activating diiron enzymes have been reported. [20] However, most of these studies used small cluster models or QM/MM with a modest QM region. To be specific, none of the reported computational studies included the Tyr 40 residue in the models and hence no pathway was ever studied that includes the involvement of this residue. To gain insight into the mechanism of ferritin activity we conducted a computational study into the various possible reaction mechanisms, including the use of large active site models to probe second coordination sphere effects. The latter is relevant because in recent work we showed the importance of the secondcoordination sphere in quantum chemical calculations, whereby the long-range electrostatic, dipole and electric field perturba- The universal reaction mechanism of dioxygen reduction to hydrogen peroxide or water at the diiron site of ferritins through pathway 1 (independent of site C occupation) and pathway 2 (with Fe 2 + occupation at site C). tions influence optimized geometries and energetics. [21] Our calculations indicate that Tyr 40 is rapidly oxidized and deprotonated upon O 2 binding and enables facile reduction of O 2 to H 2 O 2 on the diiron center. Our studies also imply an important role of the second-coordination sphere Asp 65 residue that can hold a cation or may be protonated during the catalytic cycle.

Model set-up and validation
A large active-site cluster model was created based on the structure of SynFtn as deposited in the 6GKA PDB file (see Figure 2). [12a,18] Dioxygen was added to the Fe2 atom and hydrogen atoms and solvent (water) added in Chimera. [22] The model includes the 4 α-helical chains around the diiron center with all residues pointing outward as truncated to Gly residues. In addition, a number of polar and charged residues that potentially can induce a local dipole moment or act as part of the hydrogen bonding network have been taken into consideration. The full reaction mechanism was calculated for model B and BP, which differ in the protonation state of Asp 65 : namely, it is protonated in model BP while deprotonated in model B.
Thus, a pK a analysis in PropKa predicts Glu 62 and Asp 65 to have deprotonated amino acid chains. However, based on a visual inspection of the structure, we reasoned that since these anionic groups point toward each other that they may be part of a cation binding site although no cation is found in the crystal structure. Therefore, both models B and BP were tested here. A further expanded model C of 439 atoms was tested with Asp 65 deprotonated, but gave the same structure and electronic configuration as model B.

Geometry optimization of large cluster models
Following the manual addition of a dioxygen group bound to one of the Fe 2 + ions (Fe2) of the ferroxidase center (see model setup in Methods section), geometry optimization leads to essentially identical end-on superoxo structures for all residues conserved in each of the large cluster models (B, BP and C). This is due to the presence of a strong hydrogen bonding network included in the models that restrains the systems and maintains the site of iron and oxygen binding in a rigid conformation. An overlay of the optimized geometry of 11 Re B with the crystal structure coordinates deposited under 6GKA PDB file puts most protein chains in virtually identical positions and retains the basic features of the biochemical system ( Figure S14, Supporting Information). The superoxo group is hydrogen bonded to both the side chain of Gln 143 and the alcohol group of Ser 146 , while the phenol group of Tyr 40 forms a hydrogen bond with the carboxylate of Glu 110 . Overall, the extra-large model C resulted in the same fold and electronic configuration and structural features as model B. We therefore continued the work with model B and BP only.

Calculations on unprotonated Asp model B
As previous computational studies on diiron complexes showed that they contain many close-lying spin and electronic configurations in the various complexes, [20,23] we tested the reaction mechanism on multiple spin state surfaces in various possible configurations for each complex. For all spin states for model B, optimization resulted in a small redistribution of electrons, giving a diiron(II)-superoxo complex with an unpaired electron on the superoxo group, as well as an unpaired electron shared between the carboxylate groups of Glu 62 and Asp 65 in the orbital labeled as π* Asp . Note, that in none of the reactant complexes, Re B or Re C , was spin density found on the Tyr 40 residue. Attempts were made to swap molecular orbitals and find an iron(III)-superoxo with closed-shell deprotonated Asp 65 instead; however, during the SCF convergence the system returned to an Asp 65 radical instead. Consequently, in our model without cation bound to Glu 62 and Asp 65 and both residues in their deprotonated form, an electron transfer happens from these carboxylates to the iron center that creates an electronic configuration with an Asp radical as the lowest energy structure for model Re B . This is unusual in bioinorganic chemistry as no experimental evidence for Asp radicals has ever been found in ferritin proteins. As such, model B may be a spurious result due to a missing charge and may not be a realistic model of ferritin enzymes. We conclude, therefore, that we expect that in reality Asp 65 is probably protonated or binds a cation.
To find out if system B with deprotonated Glu 62 and Asp 65 would be catalytically active, we nevertheless, as a test of principle, investigated the potential energy landscape of dioxygen conversion into H 2 O 2 using ferritin model B on the lowest energy singlet, triplet, quintet, septet, nonet and undecaplet spin states. In all spin states there was radical character on Asp 65 (Table S10, Supporting Information) in most configurations along the mechanism with 7 Re B as the lowest in energy. The reaction for model B starts from a diiron(II)superoxo structure by proton abstraction from the bridging water molecule to form an iron(II)-iron(III)-hydroperoxo species with barriers below 11 kcal mol À 1 . As the pK a of an ionized Tyr residue is low (pK a = À 2), [24] a proton coupled electron transfer from Tyr 40 transfers a proton to Glu 110 while an electron is moved from Tyr 40 to the diiron system to form a diiron(II)hydroperoxo species with a barrier of 15.2 kcal mol À 1 above reactants ( Figure S16, Supporting Information). A final proton transfer from Glu 110 shuttles the proton to the hydroperoxo group to form H 2 O 2 in a slightly overall endothermic reaction by 5.1 kcal mol À 1 . As the highest barrier along the calculated mechanism is only 19.9 kcal mol À 1 this implies that model B could be reactive and convert dioxygen to H 2 O 2 on the diiron(II) center of ferritin despite showing an unusual radical center on Asp 65 . However, the structural orientation of the side chains of Asp 65 and Glu 110 imply space for a cation to bind there. As such, we think it more likely that Asp 65 is either protonated or bears a cation during the reaction mechanism. Nevertheless full details of the predicted transition states and intermediates for the unprotonated Asp model B and a discussion on the calculations are given in the Supporting Information.

Calculations on Asp protonated model BP
Upon visual inspection of the model B structures, we decided to repeat some of the calculations with a proton added to the Asp 65 residue, i. e., Model BP. In particular, the Asp 65 side-chain of each subunit may either bind a cation such as Na + or K + or even Fe 2 + . Alternatively the side chains of Glu 62 and Asp 65 may share a proton during the catalytic cycle of ferritin. While the overall spin state of the reactant complex is the same as that of model B, there is an increased charge of + 1. The optimized geometries of structures 11,9,7,1 Re BP have no spin density located on the two carboxylate residues. These optimized geometries are shown in Figure 3 and an overlay of the optimized geometries of 7 Re B and 11 Re BP (right-hand-side of Figure 3) shows them to be almost identical. Therefore, as expected, the protein fold and second-coordination sphere are not affected by protonation of Asp 65 and the characteristic features of the enzyme structure stay the same. The major difference between the 7 Re B and the Re BP structures relates to Tyr 40 , which is deprotonated in Re BP and bears an unpaired electron, while it is closed-shell and protonated in Re B .
During the geometry optimization of Re BP the proton from the phenol group of Tyr 40 spontaneously moved to Glu 110 . Consequently, Glu 110 in the ferritin environment is much more basic than Tyr 40 and accepts its proton rapidly. Attempts to swap molecular orbitals and create an electronic configuration without the Tyr 40 radical in Re BP failed and did not converge. Hence, the lowest energy state has a radical on Tyr 40 . The molecular orbitals of the reactant complexes are shown in Figure S13, Supporting Information. The relevant molecular orbitals mainly originate from the metal 3d sets of orbitals on Fe1 and Fe2 and their interactions with first-coordination sphere ligands. Overall, all reactant complexes of model BP have an electronic configuration with two iron(II) centers and a radical on superoxo and Tyr 40 , i. e. π* xy,Fe1 2 π* xz,Fe1 1 π* yz,Fe1 1 σ* z2,Fe1 1 σ* x2-y2,Fe1 1 π* xy,Fe2 2 π* xz,Fe2 1 π* yz,Fe2 1 σ* z2,Fe2 1 σ* x2-y2,Fe2 1 π* OO 1 π* Tyr 1 . The eight unpaired electrons on Fe1/Fe2 are ferromagnetically coupled to the unpaired electrons on the superoxo and Tyr 40 units in 11 Re BP , while they are antiferromagnetically coupled in 7 Re BP . In 1 Re BP the electrons on Fe1 and superoxo are up-spin, whereas those on Fe2 and Tyr 40 are down-spin. Due to the same electron distribution in 1 Re BP , 7 Re BP , 9 Re BP and 11 Re BP these structures are close in energy and, indeed, we find them all within 1.5 kcal mol À 1 , with 1 Re BP lowest in energy. As a result equilibration between the four reactant complexes will be  11,9,7 Re BP with bond lengths in angstroms. Relative energies (ΔE + ZPE values) are in kcal mol À 1 . The right-hand-side shows an overlay of the 11 Re BP (light blue) and 7 Re B (amber) optimized geometries. possible and all will have a finite lifetime. Moreover, reactivity patterns originating from all reactant complexes can be expected through multistate reactivity. [19] In all structures, the OÀ O distance is short at 1.420 Å in 11 Re BP , 9 Re BP and 1 Re BP while it is 1.394 Å in 7 Re BP .
The mechanism of dioxygen conversion into H 2 O 2 on the diiron center of SynFtn for model BP is illustrated in Scheme 1 and Figure 4. First, proton transfer from the water molecule bridging the two iron ions to the terminal oxygen atom of the iron(II)-superoxo group occurs, along with electron transfer from the iron(II) to the superoxo group via a transition state TS1 BP . This forms intermediate IM1 BP , which features an iron(II)iron(III)-hydroperoxo species. Barriers for the proton coupled electron transfer are ΔE + ZPE = 10.1 (10.1) kcal mol À 1 on the undecaplet (nonet) spin state surfaces, while the open-shell singlet spin state is at 9.5 kcal mol À 1 . Energetically these barriers are slightly lower than those for 11,9 TS1 B reported in the Supporting Information Figure S16. The transition states are product-like with long HÀ OH distances on the bridging water of 1.316 (1.318) [1.534] Å for 11 TS1 BP ( 9 TS1 BP ) [ 1 TS1 BP ], while the distance to the terminal oxygen of the superoxo is much shorter: 1.168 Å for both 11,9 TS1 BP and 1.056 Å for 1 TS1 BP . The nonet and undecaplet transition states also have relatively large imaginary frequencies, i1068 and i1072 cm À 1 , respectively, for the OÀ HÀ O stretch vibration between the peroxo group and the μ-water ligand, while it is much smaller for the singlet spin structure.
After the transition states, the resulting iron(II)-iron(III)hydroperoxo complex (IM1 BP ) retains the radical on Tyr 40 and a protonated Glu 110 group. The latter releases its proton to the distal oxygen of the iron(III)-hydroperoxo species via barriers of 21.2 kcal mol À 1 on the nonet and undecaplet spin states, while the septet and singlet are much higher lying. The 9,11 TS2 BP structures are shown on the right-hand-side of Figure 4. They have imaginary frequencies of i854( i856) cm À 1 for the undecaplet (nonet) spin states. Both TS2 BP structures are relatively central, i. e. neither early nor late, with the transferring hydrogen atom almost midway between the donor and acceptor oxygen atom, namely the distances for the donating OÀ H interaction is 1.244 (1.257) Å and the distances for the accepting OÀ H interaction is 1.215 (1.207) Å for 11 TS2 BP ( 9 TS2 BP ).
The FeÀ O distances toward the H 2 O 2 group have elongated to values above 2.2 Å. Interestingly, the 1 TS2 BP barrier is relatively high in energy and much higher than those on the undecaplet and nonet spin states. The same is true for the singlet spin product complex. This is surprising as the electronic configuration of the singlet, nonet and undecaplet products all have an iron(II)iron(III) complex coupled to a radical on Tyr 40 . However, in the nonet and undecaplet the two iron centers have all metal electrons in up-spin, while they are antiferromagnetically coupled in the singlet spin state. Moreover, in the singlet spin state the iron(II) center ferromagnetically couples with the tyrosinyl radical, while it is antiferromagnetic with the iron(III) center. Since most metal-type orbitals are smeared out

ChemBioChem
Research Article doi.org /10.1002/cbic.202200257 over the full diiron system this results in a higher singlet spin product complex than the higher-spin orientations. Several attempts were made to swap orbitals to find a lower energy singlet spin product state, but all converged to the same state shown in Figure 4.
Overall, the potential energy landscapes for dioxygen activation by Re B and Re BP show similar patterns with a small initial proton coupled electron transfer event with barriers of 9-11 kcal mol À 1 , followed by a larger second barrier of around 20 kcal mol À 1 for H 2 O 2 formation. The details of the electronic changes during the reaction mechanism of oxygen activation by the protonated Asp 65 model Re BP are given in Scheme 1. As discussed above, the singlet spin state reactant is the ground state with two antiferromagnetically coupled iron(II) centers coupled to a superoxo and a tyrosyl radical. Upon proton transfer from the μ-H 2 O group to superoxo a high-spin state is formed ( 11 IM1 BP ) that is built up from an iron(III)-hydroperoxo group in sextet spin coupled to a quintet spin iron(II) ion and a tyrosyl radical. A final proton transfer from Glu 110 to iron(III)hydroperoxo generates the H 2 O 2 product and leaves a mixedvalent Fe(III)-Fe(II) center with a nearby tyrosyl radical.
The mechanism described for model Re BP closely matches experimental data for SynFtn. Thus, Bradley et al. [12a] detected both a mixed valent iron(III)-iron(II) ferroxidase center and a tyrosyl radical during the catalytic cycle, consistent with our findings for model Re BP and its reaction mechanism. In conclusion, our work points to the mechanism initiating from Re BP as the one that best links to experimental work. Even though the mechanism and barriers for the unprotonated model Re B are similar to those for model Re BP , a radical on Tyr 40 is formed only at the final step leading to a mixed-valent intermediate and product structure. As such, we believe the Asp 65 group is either protonated or binds a cation and thereby guides the reaction to mechanism Re BP .
Interestingly, a recent study of a D65A variant of SynFtn revealed a significant, unexplained effect on the mechanism of iron(II) oxidation, [24] consistent with the results reported here suggesting an important functional role for this residue in determining the mechanistic path of iron oxidation/O 2 activation. A survey of ferritin sequences reveals that neither Glu 62 nor Ser 146 are conserved within H-chain peptides ( Figure S1, Supporting Information). There is considerable variation in the residues at the position equivalent to 146 whilst that equivalent to position 62 is typically Gln, His or occasionally Ala. The residue equivalent to position 65 is invariably a Glu in the other proteins surveyed. Comparison between coordinate sets deposited in the PDB shows Asp 65 of SynFtn to adopt a conformation distinct from the Glu of other ferritins ( Figure S1, Supporting Information). The results of our calculations therefore provide a rationale for the unique reactivity reported for SynFtn with other ferritins utilizing an alternative pathway to O 2 activation due to the absence of a hydroxyl on residue 146 and a different electronic potential in the region of residues 62 and 65 of the SynFtn peptide chain.
We also explored possible reaction mechanisms that proceed via a μ-1,2-peroxo intermediate. In particular, for the deprotonated Asp 65 model we located its optimized geometry and found it higher in energy than the lowest reactant complex by ΔE + ZPE = 8.5 kcal mol À 1 . Moreover, its formation barrier from a diiron(II)-peroxo reactant complex was calculated to be well over 25 kcal mol À 1 . This barrier is far higher than the alternative proton coupled electron transfer to form the mixed valent iron(II)-iron(III)-hydroperoxo species. Therefore, the μ-1,2peroxo diiron complex will not take part in the catalytic reaction . Potential energy profile (ΔE BS2 + ZPE BS1 data in kcal mol À 1 ) for dioxygen activation by ferritin models 11,9,7,1 Re BP as calculated in Gaussian. Optimized geometries of key transition states in the mechanism with bond lengths in angstroms and the imaginary frequency in cm À 1 . mechanism of SynFtn and consequently is a side-product from an alternative higher energy pathway. Our studies match previous computational studies on benzoyl coenzyme A epoxidase, [23b] that also assigned the μ-1,2-peroxo diiron(III) complex as a side reaction channel of a higher energy pathway.

Conclusion
A series of density functional theory calculations on the catalytic cycle of ferritin based on a model of SynFtn are reported using large cluster models of well over 300 atoms that contain the first-and second-coordination sphere around the diiron center. The calculations suggest that formation of the first experimentally observed reaction intermediate is preceded by several rapid electron and proton transfer events offering new insight into the mechanism of O 2 activation by ferritin enzymes specifically, and diiron proteins more broadly. Specifically our calculations show that rather than oxidation of Tyr 40 by an iron(II)-iron(III)-superoxo species, as was originally proposed, [10] SynFtn first forms a diiron(II)-superoxo species and a tyrosyl radical. The iron(II)-iron(III)-hydroperoxo form of the diiron site is then generated via subsequent facile electron and proton transfer processes.
The end point of our calculations here is an iron(II)-iron(III)hydrogen peroxide complex and a Tyr radical. Experimental evidence indicates that the hydrogen peroxide readily dissociates from the ferroxidase center, and that the Tyr radical is rapidly quenched. According to the scheme of Bradley et al., the radical centered on Tyr 40 is translocated to a remote diiron(II) catalytic center, which has not encountered O 2 , by a series of electron transfer steps, resulting in an iron(III)-iron(II) ferroxidase center with no hydroperoxo group bound. In this way, the overall reaction of 24 diiron(II) ferroxidase centers with 12 molecules of O 2 results in 12 molecules of H 2 O 2 and 24 iron(III)-iron(II) ferroxidase centers. [12a] The calculations shown in this work highlight a strong hydrogen-bonding network in the active site that positions the diiron center and dioxygen and enables facile proton transfer. In particular, geometry optimizations indicate that the alcohol group of Ser 146 stabilizes the superoxo species bound to Fe2 via a hydrogen bonding interaction. We also propose that protonation of, or cation binding between, the carboxylates of Asp 65 and Glu 62 is key to determining SynFtn reactivity. With a proton/ cation located on the Asp 65 group, a mechanism of O 2 activation is found in which the first proton is delivered by a bridging water molecule, while the second one is shuttled in from Tyr 40 via Glu 110 . The mechanism produces a mixed-valent iron(III)iron(II) product complex and relies on the nearby Tyr 40 that forms a tyrosyl radical, in excellent agreement with experimental observation. Overall our studies highlight the importance of second-coordination sphere effects in the catalytic mechanism of nonheme diiron enzymes, where hydrogen bonding interactions position polar residues and assist with proton-relay mechanisms. In addition, long-range electrostatic interactions influence electronic configurations and spin-state stabilities that are essential for chemical catalysis.

Experimental Section
Model set-up: We initially ran a 500 ns molecular dynamics (MD) simulation on the ferritin structure starting from the crystal structure coordinates (6GKA PDB file) and with hydrogen atoms added under pH 7 conditions as well as a water layer. The MD simulation (Supporting Information) shows that the structure is highly rigid throughout with a very stable fold and secondcoordination environment. Based on these results, models of the active site of ferritin were created and validated as described previously and followed standard procedures for the set-up of cluster models. [26] The cluster models were created from the 6GKA PDB file as retrieved from the protein databank. [12a,18] The PDB file describes a monomer with both iron atoms included but without dioxygen. Hydrogen atoms were added in Chimera using pH 7 conditions, [22] which assumed all Arg and Lys side-chains were protonated and all Asp and Glu side-chains taken to be in their deprotonated forms. The protonation states of all acidic and basic residues were further confirmed with PropKa that also assigned the Glu 62 and Asp 65 residues to be deprotonated. However, upon visual inspection of the protein structure we decided to investigate models with Asp 65 in its deprotonated form (models A, B, C) and in its protonated form (model BP). His 29 was assumed to be doubly protonated, while all other histidine residues were singly protonated. Large active site cluster models were generated that incorporate the first-, second-and third-coordination sphere of the catalytic center and further studied with density functional theory methods. These models often describe short-and long-range electrostatic interactions well and are the method of choice for enzymatic reaction mechanisms. [26,27] To gain experience with the chemical system, we first explored a small cluster model of 121 atoms (model A), see Supporting Information for details. Subsequently, we investigated more realistic models with a large number of second-and third-coordination sphere residues included: Model B with 340 atoms and Model BP with 341 atoms. Model B included the two iron atoms bridged by a water molecule and the carboxylate of Glu 66 which is part of the peptide chain Glu 62 -Ala 63 -Lys 64 -Asp 65 -Glu 66 -Gln 67 -Ser 68 -His 69 , whereby the Ala 63 , Lys 64 , Gln 67 and Ser 68 residues were truncated to Gly. Another peptide chain incorporated in the model covered the sequence from His 29 to Tyr 40 . Again, the amino acid side chains pointing away from the active site were truncated to Gly, i. e. Ile 30 , Ser 31 , Ile 32 , Arg 34 , Tyr 35 , Val 38 and Thr 39 . The Glu 110 ligand of one of the iron atoms was included as part of the short peptide chain Glu 110 -Ala 111 -Asp 112 -Thr 113 with the Ala and Asp residues shortened to Gly. A final peptide chain Gln 143 -Ile 144 -Gln 145 -Ser 146 , with the two middle residues without side chains donating hydrogen bonding interactions into the active site. Lastly, the model contains the side chain of Phe 150 as toluene and eleven water molecules. A dioxygen group was manually added to one of the iron(II) atoms (Fe2) in an end-on conformation. Our model is overall charge neutral and was investigated with odd spin multiplicity (M = 1, 3, 5, 7, 9, 11). To test the reproducibility of the model, we expanded model B further by adding Ala 70 -Val 71 -His 72 -Phe 73 , Phe 107 -Gln 108 -Met 109 and Thr 114 with the side chains of Ala 70 , Val 71 , Phe 107 and, Gln 108 truncated to Gly to get model C of 439 atoms and minimized its geometry 11 Re C . The structure and electronic configuration of 11 Re C showed little difference to that of 11 Re B ; hence the work was continued with model B and BP only.
Despite the fact that the Glu 62 and Asp 65 side chains are considered to have low pK a values, actually their local environment on visual inspection does not show a nearby positive counter charge. Therefore, we investigated a protonated model whereby a proton was added between the two carboxylate side chains of Glu 62 and Asp 65 : Re BP . The full mechanism of dioxygen activation on several spin states surfaces was also investigated for model Re BP . All models were run without geometric constraints and a comparison of optimized geometries with crystal structure coordinates only showed minor deviations.
Procedures: Density functional theory (DFT) calculations were performed on the ferritin models using the Gaussian-09 software package, [28] and utilized the unrestricted B3LYP hybrid density functional method. [29] Geometry optimizations, frequencies, intrinsic reaction coordinate scans and geometry scans were performed with an LANL2DZ basis set on iron with core potential and 6-31G on the rest of the atoms: basis set BS1. [30] To correct the energies, single point calculations on the optimized geometries were performed with the LACV3P + basis set on iron with core potential and 6-311 + G* on the rest of the atoms: basis set BS2. All calculations were run with the continuum polarized conductor model (CPCM) included with a dielectric constant mimicking chlorobenzene. [31] For a number of structures, dispersion corrected DFT was tested, but resulted in energies very similar to those obtained without dispersion, in agreement with previous work. [32] The methods used for this work have been extensively tested for analogous systems and reproduce spin-state orderings, product distributions and free energies of activation well. [33] Transition states were obtained through full geometry optimizations and were characterized with a single imaginary mode for the correct transition. For a selection of transition states we ran intrinsic reaction coordinate scans (IRCs) that confirmed the connection of the TS with the two local minima, see Supporting Information. Free energies were obtained from unscaled vibrational frequencies and entropies at 298 K. All energies reported in this work were obtained using UB3LYP/BS2 with solvent and zero-point corrections included.