Axial Ligation and Redox Changes at the Cobalt Ion in Cobalamin Bound to Corrinoid Iron-Sulfur Protein (CoFeSP) or in Solution Characterized by XAS and DFT

A cobalamin (Cbl) cofactor in corrinoid iron-sulfur protein (CoFeSP) is the primary methyl group donor and acceptor in biological carbon oxide conversion along the reductive acetyl-CoA pathway. Changes of the axial coordination of the cobalt ion within the corrin macrocycle upon redox transitions in aqua-, methyl-, and cyano-Cbl bound to CoFeSP or in solution were studied using X-ray absorption spectroscopy (XAS) at the Co K-edge in combination with density functional theory (DFT) calculations, supported by metal content and cobalt redox level quantification with further spectroscopic methods. Calculation of the highly variable pre-edge X-ray absorption features due to core-to-valence (ctv) electronic transitions, XANES shape analysis, and cobalt-ligand bond lengths determination from EXAFS has yielded models for the molecular and electronic structures of the cobalt sites. This suggested the absence of a ligand at cobalt in CoFeSP in α-position where the dimethylbenzimidazole (dmb) base of the cofactor is bound in Cbl in solution. As main species, (dmb)CoIII(OH2), (dmb)CoII(OH2), and (dmb)CoIII(CH3) sites for solution Cbl and CoIII(OH2), CoII(OH2), and CoIII(CH3) sites in CoFeSP-Cbl were identified. Our data support binding of a serine residue from the reductive-activator protein (RACo) of CoFeSP to the cobalt ion in the CoFeSP-RACo protein complex that stabilizes Co(II). The absence of an α-ligand at cobalt not only tunes the redox potential of the cobalamin cofactor into the physiological range, but is also important for CoFeSP reactivation.


Introduction
The cobalamin cofactor (Cbl, also denoted vitamin B 12 ) since its discovery in 1925 has attracted much research interest [1][2][3][4]. Cbl is essential for all mammals [5] and in bacteria it is involved in carbon oxide (CO x ) conversion pathways related to potential renewable energy applications [6,7]. Anaerobic CO 2 reduction along the bacterial Wood-Ljungdahl pathway includes several unique enzymes [8,9]. The corrinoid iron-sulfur protein (CoFeSP) carries a Cbl cofactor [10,11] and shuttles a methyl group from methyl-transferase bound methyl-tetrahydrofolate to acetyl-CoA synthase. The latter enzyme, after receiving a CO group derived from CO 2 reduction by carbon monoxide dehydrogenase, synthesizes acetyl-CoA for many metabolic reactions [12]. CoFeSP alternates in the methyl transfer cycle between Co(III)-CH 3 and Co(I) states [13]. The Co(I) state is prone to oxidative inactivation generating Co(II), which can be reductively reactivated in an ATP-dependent reaction catalyzed by the reductiveactivator protein (RACo) [14][15][16]. Redox and ligation changes at cobalt in Cbl in the CoFe-SP-RACo system thus are essential in the CO x conversion pathway.
Cobalamin is among the most complex non-polymeric compounds in nature and consists of a unique corrin hetero-macrocycle binding a central cobalt ion by four equatorial nitrogen ligands [17]. Two axial cobalt ligands (α and ß) may be bound in addition. The α-ligand in Cbl in solution or in prototypic Cbl-proteins in the so-called base-on configuration is the nitrogen atom of a dimethylbenzimidazole (dmb) group connected to the corrin ring (Fig 1). Replacement of the dmb ligand (base-off) by a water species or by other amino acids occurs in many proteins [1][2][3]. Crystal structures of isolated CoFeSP and of the protein in complex with RACo or methyl transferase have been reported [12,14,16,18,19]. In all CoFeSP structures, the dmb group is folded away from the corrin so that the α-site apparently is vacant (Fig 1). However, it could also be occupied by a crystallographically less visible (disordered) water species or even by the hydroxyl group of a nearby threonine residue modeled at about 3.5 Å to cobalt in the structures. The ß-site in CoFeSP-Cbl can be occupied by a water species (AqCbl), a methyl group (MeCbl) [20], or may be vacant (Fig 1). In the CoFeSP-RACo protein complex, binding of the hydroxyl group of a serine (Ser 398) of RACo to cobalt at the ß-position has been shown [14][15][16].
Binding of the axial ligands is closely related to the cobalt oxidation state [4]. In Cbl, both in solution and bound to proteins, the formal Co(I), Co(II), and Co(III) states are associated with low-spin (3d 8 , 3d 7 , 3d 6 ) valence electron configurations [22][23][24]. Only Co(II) thus is EPR active. A decrease of the oxidation state may be accompanied by a decreasing number of axial ligands, meaning that (L = ligand) (L α)Co III (Lß), Co II (Lß) or (L α)Co II , and Co I species may prevail [24,25], but in protein environments deviations from such configurations may occur. Control of the axial cobalt ligation in group-transferring Cbl-enzymes such as CoFeSP is important in the reactions. However, relating the redox state to the axial ligation of cobalt can be difficult both by crystallography and spectroscopy. For example, in solution Cbl mixtures of base-on/off states may occur, in crystal structures of Cbl-proteins axial ligands may be unresolved, or certain spectroscopic methods do not provide structural and electronic parameters or only for selected cobalt redox states. Further insight in the cobalt site structures in the CoFeSP-Cbl-RACo system is required to understand the interplay of protein-protein interactions, redox transitions, axial ligand exchange, and methyl group transfer.
Here, we employed X-ray absorption spectroscopy (XAS) at the Co K-edge in combination with density functional theory (DFT) to study redox and coordination changes at cobalt in CoFeSP-Cbl in comparison to Cbl in solution. XAS in principle facilitates oxidation state, metal-ligand bond lengths, and site symmetry determination for solution and protein systems and can be applied to all spin and oxidation states of metal sites [26][27][28][29]. In particular the XAS features due to resonant 1s electron excitation into unoccupied valence levels (for example with Co(3d) character) in the so-called pre-edge absorption spectral region (core-to-valence transitions, ctv), which can be calculated by DFT [30][31][32][33], are sensitive to the molecular and electronic structure of the Cbl cofactor [34][35][36][37][38][39]. Pronounced alterations of the ctv spectra upon changes at cobalt were observed, which were reproduced by the computational approach. Combination of experimental and theoretical analyses has established relations between the Xray spectroscopic features and the redox state and axial ligation at the cobalt centers, thereby providing structural models for the AqCbl and MeCbl cofactors in CoFeSP and in the CoFe-SP-RACo protein complex.

Sample preparation
CoFeSP and RACo proteins from Carboxydothermus hydrogenoformans were heterologously overexpressed in Escherichia coli following previously established protocols [15,16,19] and protein purification and biochemical treatments were performed under anoxic conditions in 95% N 2 and 5% H 2 atmosphere at room temperature in a glove box. Synthetic cobalamin (denoted AqCbl ox , CNCbl ox , and MeCbl ox ) containing Co(III) was purchased from Sigma-Aldrich, all chemicals were at least analysis grade. Purified CoFeSP (25 μM) was reconstituted in 20 mM TRIS-HCl buffer (pH 8.0) with synthetic AqCbl ox or MeCbl ox (40 μM) by overnight incubation at 25°C, subsequently unbound cofactor was removed and the protein concentrated using Vivaspin 500 concentrators (10 kDa cut-off). CoFeSP-AqCbl ox and CoFeSP-MeCbl ox samples for XAS contained 1.0±0.1 mM protein as determined by the Bradford method [40]. A glass-forming agent was not present in the XAS samples. Titanium(III)-citrate (2 mM) was added to CoFeSP-Cbl ox samples for cofactor reduction (red). The CoFeSP-AqCbl-RACo protein complex was prepared according to the previously reported protocol [14][15][16]. Solution samples of synthetic cobalamins (7 mM) were prepared by solvation of AqCbl ox , CNCbl ox , and MeCbl ox powders in 20 mM TRIS-HCl buffer (pH 8.0) and chemical reduction was achieved by addition of sodium dithionite (100 mM) to obtain CNCbl red or titanium-citrate (40 mM) to  [12]) showing a base-off configuration (dmb ligand not bound to cobalt in α-position). Ligand X at the ß-position (light green) at cobalt can be absent or can be a water species, a methyl group, or an oxygen from the side chain of RACo-Ser398 in the CoFeSP-RACo complex [14]; red balls show resolved water molecules. (b) Structure of Cbl in base-on configuration [21]; X can be a water, cyanide, or methyl species. Color code: magenta, Co; blue, N; red, O; grey, C; dark green, P; protons were omitted for clarity.
obtain AqCbl red samples. Aliquots of samples (50 μL) were loaded into Kapton-covered acrylic glass holders for XAS and frozen in liquid nitrogen. Optical absorption spectra of samples using 2 μl aliquots were recorded on a Specord 50 Plus instrument (Analytik Jena, Germany).
Total reflection X-ray fluorescence analysis TXRF [41] was performed for metal content determination in protein samples using a PicoFox instrument (Bruker, Berlin, Germany). Protein samples were mixed (v/v 2:1) with a gallium concentration standard (Sigma, 50 mg/L) and three measurements were carried out per sample using 5 μl aliquots.

X-ray absorption spectroscopy
XAS at the Co K-edge was performed at the bending-magnet beamline KMC-1 at BESSY (Helmholtz-Center for Materials and Energy Berlin) with the storage ring operated in top-up mode (250 mA). The excitation energy was tuned by a Si[111] double-crystal monochromator. Kα-fluorescence-detected XAS spectra were collected using an energy-resolving 13-element germanium detector (Canberra) on samples held in a liquid-helium cryostat (Oxford) at 20 K. The detector was shielded against scattered X-rays by a 10 μm iron foil. The K-edge inflection point at 7709 eV of a simultaneously measured cobalt metal foil was used for calibration of the energy axis. Detector deadtime corrected XAS spectra (scan duration~30 min) were averaged (up to 9 scans, 2 scans per sample spot) for signal-to-noise ratio improvement. No radiation induced spectral changes (i.e. in the XANES) were observed for increasing XAS scan numbers on single sample spots. XAS data processing was carried out as previously described [27] to yield normalized XANES and EXAFS spectra. Simulation of k 3 -weighted EXAFS spectra in kspace was carried out using the in-house software SimX and phase functions calculated with FEFF7.0 (S 0 2 = 0.85) [42,43]. In the fits the number of C-atoms was set to the values corresponding to the corrin ring. Fourier-transforms of EXAFS spectra were calculated for k = 1.8-12.2 Å -1 using cosine windows extending over 10% of both k-range ends. The XANES pre-edge features were extracted by polynomial spline subtraction with the program XANDA [44]. Multiple-scattering theory simulations of K-edge spectra were performed with the FEFF9.0 code [45] using model structures based on the cobalamin crystal structure in CoFeSP (PDB entry 2H9A, 1.9 Å resolution [12]); for details see the Supporting Information (Fig C in S1 File).

Density functional theory calculations
Starting geometries for DFT were derived from cobalamin crystal structures, in which the axial cobalt ligands were modified and structures were truncated to minimize calculation times ( Fig  1; the nucleotide loop and amide side chains of Cbl were removed, in base-off models the dmb was removed, in base-on models a benzimidazole group mimicked the dmb ligand, as further ligands, OH -, H 2 O, CH 3 -, or CNgroups were added). The total charge and spin multiplicity of the models was set to the desired low-spin cobalt oxidation state [21,46]. The model structures were geometry-optimized using the Gaussian09 package [47], the B3LYP functional [48], and a triple-zeta-valence-plus-polarization basis set (TZVP) [49] on the Soroban computer cluster of the Freie Universität Berlin. The theoretical approach was selected because it provided spectra which near-quantitatively agree both in absolute and relative shapes with the experimental data, besides of showing good agreement between experimental and calculated site geometries. Natural population analysis (NPA) charges [50] were calculated with the NBO-5 program [51]. The pre-edge features (ctv) in the XANES were calculated by DFT using the ORCA program [52,53] on the basis of the geometry-optimized model structures as previously described [30][31][32][33]. The calculated ten ctv transitions (sticks) at lowest energies were broadened by Gaussian functions (FWHM 2.5 eV), 158.7 eV shifted on the energy axis, and their amplitudes were scaled (x900) for comparison with experimental ctv spectra.

Cofactor content and oxidation state
Cobalamin (Cbl) species were investigated when bound to the CoFeSP enzyme, in the CoFe-SP-RACo protein complex, and in solution samples serving as reference materials. The expected axial cobalt ligations in the samples included dimethylbenzimidazole (dmb) or water (Aq) species at the α position (occupied by the dmb ligand in crystalline Cbl) or Aq, cyanide (CN), or methyl (Me) group species at the ß position (opposite to the dmb ligand). CoFeSP-Cbl and solution Cbl samples in the oxidized state (ox) and after chemical reduction (red) were compared. We denote the various redox and axial ligation species of cobalt as (α-ligand)Co x (ßligand) (x = valence state) in the following. The cobalt concentrations in the Cbl-reconstituted protein samples were determined by TXRF, which on average yielded 0.8±0.1 Co ions per CoFeSP containing AqCbl or MeCbl (Table 1). This suggested close to stoichiometric reconstitution of CoFeSP with the cofactors. The mean amount of 3.5±0.5 Fe ions per CoFeSP protein was in reasonable agreement with the near-quantitative presence of the [4Fe4S] cluster in CoFeSP. The mean Fe to Co ratio was 4.6 ±0.2, which for 4 Fe ions in the [4Fe4S] cluster per protein, suggested~0.85 Co ions per CoFeSP, in agreement with the protein to cobalt ratios. The increased Fe to Co ratio of 6.5 in the CoFeSP-AqCbl-RACo sample was in good agreement with two additional Fe ions in the sample compared to CoFeSP-AqCbl, due to the presence of close to one RACo protein containing a [2Fe2S] cluster per CoFeSP.
Optical absorption spectra (Fig A in S1 File) of the solution Cbl samples confirmed the expected quantitative presence of Co(III) in AqCbl ox , CNCbl ox , and MeCbl ox , and showed mostly Co(II) in AqCbl red and a Co(II) species in CNCbl red . For the protein samples, the absorption spectra (Fig A in S1 File) indicated the expected Co(III) in the cofactor in CoFeSP--MeCbl ox , suggested dominance of Co(III) in oxidized CoFeSP-AqCbl and of Co(II) in CoFe-SP-AqCbl-RACo, and showed preferentially Co(II) in CoFeSP-AqCbl red with minor (~30%) Co(I) amounts only in this sample. Electron paramagnetic resonance spectroscopy (EPR) detecting only the Co(II)-containing cofactor was used to quantify the relative Co(II) contents in the protein samples (Fig B in S1 File). This showed that CoFeSP-AqCbl ox contained~30% Co(II) and, considering also the optical spectra,~70% Co(III), CoFeSP-AqCbl red contained 70% Co(II), and CoFeSP-AqCbl-RACo near-quantitative amounts of Co(II) (~85%) ( Table 1). The altered EPR signal shape of CoFeSP-AqCbl-RACo (Fig B in S1 File) also suggested near-stoichiometric RACo binding to CoFeSP [15,19], in agreement with the TXRF data.

EXAFS on the Cbl systems
Simulation of EXAFS spectra facilitates determination of interatomic distances such as the cobalt-ligand bond lengths with~0.02 Å precision in favorable cases. Visual inspection of the EXAFS spectra of the Cbl and CoFeSP-Cbl samples revealed a dominant Fourier-transform (FT) peak due to Co-C/N/O bonds from the corrin ring and the axial ligands and smaller features at larger distances mostly due to second-sphere Co-C corrin interactions (Fig 2). The fit analysis ( Table 2) revealed typical bond lengths (~1.87 Å) of the equatorial Co-N corrin ligands in the AqCbl ox , CNCbl ox , and MeCbl ox solution samples, which were only~0.02 Å elongated in AqCbl red and CNCbl red . The second-sphere EXAFS features were well described by a mean Co-C corrin distance of~2.9 Å and a multiple-scattering contribution with an apparent N-C distance in the corrin ring of~1.4 Å. These Co-N/C corrin distances are in agreement with Cbl crystal structures [54][55][56][57] and earlier XAS data [34,36]. Axial cobalt ligands also were discernable in the EXAFS. For AqCbl ox , the dmb (α) and water (ß) ligands showed relatively similar (1.96 ±0.08 Å) bond lengths at cobalt, attributed to a slightly longer Co-N and shorter Co-O bond [56]. Both bonds were~0.3 Å elongated in AqCbl red ( Table 2). For CNCbl ox , the Co-C bond was~0.04 Å shorter than the Co-N corrin bonds as in crystalline CNCbl ox [55] and the Co-N dmb  bond was similar to AqCbl ox . Lower coordination numbers and elongated axial bonds (~2.13 Å) in CNCbl red suggested that one ligand possibly was detached. For MeCbl ox , the longer and shorter axial bonds likely were attributed to the dmb (~2.20 Å) and CH 3 (~1.95 Å) ligands [54]. Overall, the fit results showed that axial cobalt ligation changes dominated the EXAFS spectral variations.
The CoFeSP-Cbl samples showed similar Co-N/C corrin distances as found for solution Cbl in the EXAFS fits (Table 2), revealing the integrity of the base-off cofactor in the reconstituted protein [18,19]. For CoFeSP-AqCbl ox , lower coordination numbers of the axial ligands compared to solution AqCbl ox suggested only one axial ligand. Two detectable Co-O bond lengths were attributed to a larger contribution (~2.0 Å) from 5-coordinated Co(III) and a smaller contribution (~2.3 Å) from 5-coordinated Co(II). CoFeSP-AqCbl red showed significantly (~0.02 Å) shorter Co-N corrin bonds compared to CoFeSP-AqCbl ox , presumably due to the minor Co , R F represents the mean root square deviation in % between the experimental Fourier-isolated k-space EXAFS spectrum in the given reduced-distance range of the fit and the fit curve) *parameters that were fixed at given physically reasonable values in the fits # 2σ 2 was coupled to yield the same values for the~2.9 Å Co-C shell (N Co-C was set to the crystallographic distances in the~2.9-3.3 Å range, the Debye-Waller factor reflects this distance distribution with more emphasis on the 8 shorter Co-C distances).
A further Co-N-C multiple-scattering shell with the same N and 2σ 2 values as for the Co-C shell was included in the fits (apparent N-C distances given in parenthesis). The 2σ 2 values for the Co-N and Co-C/N/O shells were chosen to provide best fit results. Two lines for a given coordination shell mean that both distances were included in the respective fit. We note that splitting of the axial ligation shells in the fit procedure is tentative due to the~0.1 Å distance discrimination limit of our k = 13 Å -1 EXAFS data [58]. We note that the small N-values of the second Co-C/N/O shell with relatively long distances for CoFeSP-MeCbl (fit 19) and CoFeSP-RACo (fit 21) may not be significant and suggest dominance of 5-coordinated cobalt sites (see Fig F in  (I) contribution, and predominance of one axial~2.3 Å bond, attributed to a water species at Co(II). CoFeSP-AqCbl-RACo revealed only one significant short axial ligand bond (~2.1 Å); a longer interaction (~2.5 Å) showed a small and possibly insignificant coordination number (Fig F in S1 File). The short bond may reflect the Co-O Ser interaction in the CoFeSP-RACo complex. CoFeSP-MeCbl ox revealed only one axial ligand (~2.0 Å) due to the Co(III)-CH 3 interaction, which was slightly longer than in solution MeCbl ox (Table 2).

XANES spectral analysis
The XANES spectrum is sensitive to the spin and oxidation state of the metal, as well as to the chemical nature and symmetry of its ligands. The XANES of the solution Cbl and CoFeSP-Cbl samples revealed overall similar shapes (Fig 3), as explained by the spectral dominance of the equatorial N corrin ligands at cobalt. The significantly different K-edge energies (Fig 4) thus likely were related to the cobalt redox and axial ligation changes. Reference K-edge energies for the cobalt redox states species were derived from synthetic complexes (Fig C in S1 File) and were determined as~7715.5 eV for Co(I),~7718.3 eV for Co(II), and~7721.1 eV for Co(III) species, revealing a~2.8 eV edge energy increase per single-electron cobalt oxidation (Fig 4). All Cbl and CoFeSP-Cbl samples showed K-edge energies in the Co(II) to Co(III) region, well above the Co(I) level (Fig 4). The K-edge energies for MeCbl ox , AqCbl ox , and CNCbl ox were centered around Co(III), with AqCbl ox located at the mean Co(III) energy and a~1 eV difference between CNCbl ox and MeCbl ox . The~2.8 eV lower K-edge energies for AqCbl red and CNCbl red sugegsted near-quantitative Co(II) contents. For the protein samples, the K-edge energy of CoFeSP-MeCbl ox was closest to the Co(III) level, but the edge shape differed strongly from solution MeCbl ox (Figs 3 and 4). The edge energy for CoFeSP-AqCbl ox was lower than the mean Co(III) level due to Co(II) admixture and a~1.5 eV lower edge energy for CoFe-SP-AqCbl red reflected the increased Co(II) content. The K-edge energy for CoFeSP-AqCbl-RACo was close to the Co(II) level, but the different edge shape compared to CoFeSP-AqCbl red suggested a coordination change at cobalt. Qualitative multiple-scattering K-edge simulations on structural models for the cobalt sites (Fig D in S1 File) fairly reproduced the experimental K-edge shape and energy differences between AqCbl ox , CNCbl ox , and MeCbl ox (base-on octahedral cobalt sites) (Fig E in S1 File). Simulations for the base-off sites in CoFeSP-Cbl showed that replacement of the dmb by a water ligand in 6-coordinated sites results in lower edge energies compared to the base-on structures for AqCbl ox and MeCbl ox similar to the experimental data, α-ligand removal decreased the edge energy compared to the 6-coordinated sites by~1 eV, elongation of the axial bond as in CoFeSP-AqCbl red resulted in a small (~0.5 eV) edge energy decrease, and the edge energy of a square-planar cobalt site was close to the Co(II) level. These results suggested that the lowered edge energies in the CoFsSP-Cbl ox compared to the solution Cbl ox samples at least in part were explained by loss of one axial ligand and the K-edges of CoFeSP-AqCbl red and the CoFeSP-Cbl-RACo complex reflected a different axial ligand.

The XANES pre-edge feature
The pre-edge absorption in the K-edge reflects resonant 1s electron excitation into unoccupied valence levels with (partial) Co(3d) character (core-to-valence transitions, ctv). Pronounced differences in the ctv spectra were observed between the Cbl systems (Fig 3, insets). The small ctv feature in AqCbl ox was further decreased in AqCbl red . A larger ctv amplitude for CNCbl ox compared to AqCbl ox was further increased in CNCbl red . MeCbl ox showed the largest ctv feature among the solution samples. An almost negligible ctv feature was observed for CoFe-SP-AqCbl ox . CoFeSP-AqCbl red showed a much larger amplitude at higher energies and broader envelope of the ctv feature. Also CoFeSP-AqCbl-RACo showed a ctv amplitude increase, but a shift to lower energies. CoFeSP-MeCbl ox exhibited by far the largest ctv feature, exceeding that of solution MeCbl ox .Density functional theory (DFT) was employed to generate geometry-optimized model structures of the cobalt sites and to calculate ctv features on their basis (Fig 5). The (dmb)Co III (OH 2 ) site from DFT showed metal-ligand bond lengths in agreement with crystal structures and our EXAFS data (Table 3). This structure also reproduced the small ctv feature of AqCbl ox , whereas an OHligand yielded a too large ctv amplitude (Fig 5). The diminished ctv amplitude in AqCbl red was best reproduced using a (dmb)Co II (OH 2 ) site. Co(II) or Co(I) sites in which the water ligand, the dmb ligand, or both ligands were absent yielded larger ctv amplitudes and/or lower or higher peak energies disagreeing with the experimental data (Fig 5). A (dmb)Co III (CN) site reproduced the ctv feature of CNCbl ox well. For CNCbl red , however, the increased ctv amplitude was only calculated for a Co II (CN) site (baseoff), whereas (dmb)Co II (CN) or (dmb)Co II sites yielded too small and shifted ctv features. The large ctv feature for MeCbl ox was reproduced by the expected (dmb)Co III (CH 3 ) geometry. This shows that the ctv feature is a specific indicator of cobalt redox and ligation changes.
The small ctv feature of CoFeSP-AqCbl ox was seemingly described by a 6-coordinated (OH 2 )Co III (OH 2 ) site ( Fig 5). However, the experimental ctv feature likely was increased by a Co(II) admixture so that a 5-coordinated Co III (OH 2 ) site with a weak ctv feature at lower energies accounted equally well for the CoFeSP-AqCbl ox spectrum. The broader and larger ctv feature of CoFeSP-AqCbl red was best explained by dominance of the large ctv feature of a 5-coordinated Co II (OH 2 ) site and minor contributions of weak ctv features from a Co(I) site without axial ligands (Fig 5). The large ctv peak at lower energies of CoFeSP-AqCbl-RACo was well reproduced assuming a Co II (O Ser ) site, i.e. binding of the hydroxyl group of the serine of the RACo protein to cobalt at ß-position in the absence of an α-ligand, in agreement with the CoFeSP-RACo crystal structure [16]. Amplitude and energy of the largest ctv feature of CoFeSP-MeCbl ox were reasonably reproduced only by a 5-coordinated Co III (CH 3 ) site, whereas a (OH 2 )Co III (CH 3 ) site showed a much too small ctv feature (Fig 5).

Molecular structures of the cobalt sites
The analysis of the EXAFS, XANES, and ctv spectra using DFT (and multiple-scattering) calculations, as well as the TXRF, optical absorption, and EPR data, converged towards consistent cobalt site assignments (Fig 6). Solution Aq/CN/MeCbl samples showed the expected octahedral base-on (dmb)Co III (ß-ligand) configurations. AqCbl red likely contained a (dmb) Co II (OH 2 ) species with a weak water ligand (~2.5 Å) whereas CNCbl red seemingly preferred a base-off Co II (CN) configuration with an elongated (~2.1 Å) Co-CN bond under our conditions. Co(III) species thus were generally 6-coordinated and Co(II) species preferred 6-or 5-coordinated geometries in solution Cbl.
The cobalt ion in CoFeSP-Cbl protein showed a tendency towards lower coordination numbers compared to the same metal oxidation state in solution Cbl. CoFeSP-AqCbl containing Co(III) presumably contained a 5-coordinated Co III (OH 2 ) site as main species. Contributions from octahedral (OH 2 )Co III (OH 2 ) sites, however, were not excluded. Single-electron reduction likely resulted in a Co II (OH 2 ) site (Fig 6). CoFeSP-MeCbl ox showed a clearly 5-coordinated Co III (CH 3 ) site, meaning that water species at the α-position were undetectable. The cobalt spectral changes in the CoFeSP-RACo protein complex supported binding of the serine side chain of RACo to Co(II) at the ß-position (Fig 6).

Electronic structure considerations
Calculated ctv spectra for the relevant low-spin cobalt site species were analyzed in terms of the electric dipole and quadrupole contributions to the underlying electronic transitions and of the metal/ligand characters of the target MOs. The ctv spectra were dominated (>75%) by formally selection-rule forbidden electric dipole transitions in 6-coordinated base-on and 5-coordinated base-off Co(III) and Co(II) sites with water ligands (Table 4). Increased contributions from allowed quadrupole transitions in the base-off sites lead to increased ctv intensities. Increased quadrupole contributions (up to~50%) for Co(III) and Co(II) sites with two water ligands account for non-negligible ctv intensities in these symmetric structures. The ctv spectra of CH 3 -, CN-, or O Ser -ligand containing Co(III) sites showed almost exclusive dipole transitions, their more intense ctv features resulted from increased ligand characters of target MOs (Table 4). Dominating corrin character of target MOs for the corresponding Co(II) sites explained their more intense ctv features. The small contributions (<15%) of water ligands to target MOs generally exceeded those of dmb, but influenced the ctv intensities only moderately. The LUMO, corresponding to the lowest-energy ctv transition, and the target MO for the maximal-intensity ctv transition were compared for the main cobalt site species (Fig 7). For (dmb)Co III (OH 2 ) the LUMO was delocalized on the corrin ring and the highest-intensity target MO showed predominant Co-3d(z 2 ) character oriented along the axial ligands. These MO locations were reversed when dmb was replaced by water. Enhanced delocalization of both orbitals over the corrin ring occurred in the absence of the α-ligand. Loss of the α-ligand further caused a 1-2 eV decrease of the HOMO and LUMO energies and a~50% decrease of the LUMO-HOMO energy gap (ΔE) from~3 eV in (dmb)Co III (OH 2 ) to~2 eV in Co III (OH 2 ), mostly due to a larger relative E(LUMO) drop (Table 5). These energy changes are expected to facilitate reduction of Co III (OH 2 ) at more positive potentials than (dmb)Co III (OH 2 ). Exchange or loss of the Co(III) α-ligand further caused a cobalt charge increase by a factor up to~1.5. For Co(II) species, less pronounced changes and LUMO delocalization onto the corrin rather independent of the α-ligand and more delocalized MOs with Co(d) character were found. However, a~50% decreased ΔE compared to (dmb)Co II (OH 2 ) was observed only for (OH 2 ) Co II (OH 2 ), due to a larger relative E(HOMO) drop, whereas Co II (OH 2 ) showed an even slightly increased ΔE (Table 5). Compared to (dmb)Co II (OH 2 ), (OH 2 )Co II (OH 2 ) may thus be  [12,19] e [14] f [18] DFT data refer to geometry-optimized model structures with the indicated cobalt oxidation states and axial ligations; bond lengths from EXAFS (Table 2) were placed in the table to match the other data best and facilitate species comparison.
doi:10.1371/journal.pone.0158681.t003 harder to reduce, but Co II (OH 2 ) may be reduced at most positive potentials. The charge on cobalt for most Co(II) species was even slightly more positive compared to the Co(III) sites and the surplus negative charge was thus mostly located on the corrin ring.
For MeCbl, loss of the dmb ligand left the LUMO delocalization almost unchanged, but increased the valence level delocalization onto the corrin (Fig 7). Loss of the α-ligand rather increased ΔE due to a smaller relative E(LUMO) drop in Co III (CH 3 ) compared to (dmb) Co III (CH 3 ) (Table 5), making the Co III (CH 3 ) species easier to reduce. In addition, the charges

Discussion
Molecular and electronic structures of cobalamin species bound to CoFeSP or in solution were characterized using XAS in combination with DFT calculations. The observed K-edge energies are affected both by axial coordination and formal cobalt oxidation state changes, in agreement with earlier studies [38,39,59]. Transition from octahedral cobalt sites in solution Cbl and reference compounds to square-pyramidal sites in CoFeSP-Cbl leads to relatively lower edge energies and significant shape changes, although the Co-N corrin bond length shows only minor changes due to redox and geometry changes at cobalt. The EXAFS spectra were dominated by the Co-N corrin bonds, but facilitated estimation of the axial ligand bond lengths, which were elongated for Co(II) species as in crystal and DFT structures. The pre-edge absorption due to core-to-valence electronic excitations (ctv) revealed pronounced spectral variations in response to cobalt redox and site geometry changes. Interpretation of the ctv spectra in terms of resonant electronic excitation of a 1s core electron into unoccupied valence levels with variable metal/ligand characters was achieved using DFT. Good agreement between experimental and calculated ctv spectra was obtained for the solution Cbl and CoFeSP-Cbl systems, as previously found for other metal complexes (see, e.g., [30][31][32][33][60][61][62][63]). The ctv intensity variations were consistently explained by changes in the cobalt/ligand character ratio of the target MOs and, to a lesser extent, by electric dipole/quadrupole contribution variations of the underlying electronic transitions due to axial ligation changes. This showed for example that the intense ctv features of cobalt sites with a methyl ligand are related to significant CH 3 character of the target MOs, thus unambiguously establishing a Co III (CH 3 ) site in CoFeSP-MeCbl. The ctv-XAS/DFT combination appears to be viable for redox state and ligation geometry assignment of cobalt sites in cobalamin.
The discriminated cofactor species revealed a trend for fewer ligands at cobalt in CoFeSP-Cbl compared to solution Cbl for the same oxidation state. Octahedral (dmb) Co III (OH 2 ) and (dmb)Co II (OH 2 ) sites were dominant in oxidized and reduced solution AqCbl. XAS and DFT showed a tendency for detachment of the water species from the Co(II) ion. However, the transition from (dmb)Co III (CN) to Co II (CN) species suggested preference for detachment of the weaker dmb ligand upon cobalt reduction for CNCbl in solution. MeCbl in solution showed the anticipated (dmb)Co III (CH 3 ) structure. The spectroscopic and theoretical data converged to the same cobalt site structures in solution Cbl, corroborating the adequacy of the applied theory level (B3LYP/TZVP) for cobalamin structure description.
Crystal structures of CoFeSP-Cbl have shown the dmb group in base-off configuration in the protein [12,14,18,19]. The crystal data furthermore were interpreted as showing the absence also of water species at the α-position. The oxygen of a threonine side chain (Thr374) at the α side was modeled at 3.2-4.6 Å to cobalt in different crystals (Fig 1), suggesting the absence of a Co-O Thr bond. However, the ß-ligand bond length at cobalt also varied considerably or a ß-ligand was not assigned [12,14,18,19]. These results could be related to site heterogeneity in the crystals, which may render detection of axial cobalt ligands difficult.
Our results suggest that oxidized CoFeSP-AqCbl contains mostly Co III (OH 2 ) sites. Contributions from (OH 2 )Co III (OH 2 ) sites, however, were not completely ruled out by our data. The multiple scattering calculations of cobalamin XANES spectra (Fig D), K-edge energies from XANES simulations (Fig E), correlation of EXAFS fit parameters (Fig F), supporting references. (PDF)