Crystal structure of two anti-porphyrin antibodies with peroxidase activity.

We report the crystal structures at 2.05 and 2.45 Å resolution of two antibodies, 13G10 and 14H7, directed against an iron(III)-αααβ-carboxyphenylporphyrin, which display some peroxidase activity. Although these two antibodies differ by only one amino acid in their variable λ-light chain and display 86% sequence identity in their variable heavy chain, their complementary determining regions (CDR) CDRH1 and CDRH3 adopt very different conformations. The presence of Met or Leu residues at positions preceding residue H101 in CDRH3 in 13G10 and 14H7, respectively, yields to shallow combining sites pockets with different shapes that are mainly hydrophobic. The hapten and other carboxyphenyl-derivatized iron(III)-porphyrins have been modeled in the active sites of both antibodies using protein ligand docking with the program GOLD. The hapten is maintained in the antibody pockets of 13G10 and 14H7 by a strong network of hydrogen bonds with two or three carboxylates of the carboxyphenyl substituents of the porphyrin, respectively, as well as numerous stacking and van der Waals interactions with the very hydrophobic CDRH3. However, no amino acid residue was found to chelate the iron. Modeling also allows us to rationalize the recognition of alternative porphyrinic cofactors by the 13G10 and 14H7 antibodies and the effect of imidazole binding on the peroxidase activity of the 13G10/porphyrin complexes.

Introduction

Hemoproteins contain iron-protoporphyrin IX or heme as the prosthetic group, whose divalent iron atom can reversibly bind molecules such as molecular oxygen, leading to a wide range of biological functions [1]. Chemical or biotechnological models of hemoproteins have thus long been developed in order to create selective catalysts for industrial and fine chemistry and to predict the oxidative metabolism of new drugs [2], [3], [4], [5]. Examples include the de novo design of heme proteins, including that of membrane-soluble proteins [6], [7]. Peroxidases appear to be the easiest hemoproteins to be mimicked. Indeed, their active site consists of the iron(III)-porphyrin moiety encapsulated in the apoprotein. On one side, the heme iron is bound to an axial histidine residue (proximal ligand) and on the other side to the peroxide substrate to lead to an iron-oxo complex. The radical cation on the iron (IV)-oxo porphyrin ring can be delocalized onto proximal protein side chains [8]. The reducing cosubstrate does not bind to a well-defined site on the inside of the protein, as peroxidases restrict access of substrates to the heme-oxo complex, so that the electron transfer occurs to the meso edge of the heme [9]. Heterolytic cleavage of the O-O bond is assisted by general acid base catalysis through the concerted action of the distal histidine and arginine residues [10]. A major problem in homogeneous metalloporphyrin systems mimicking hemoproteins is that the catalyst is often destroyed by oxidation during the course of the reaction and it is difficult to combine reactivity and selectivity in these models. The use of a protein such as xylanase A [11] or an antibody mimicking the protein matrix of heme enzymes not only prevents aggregation and intermolecular self-oxidation of the catalyst, but can also influence the selectivity of the reaction [12]. As the antibody has the role of a host molecule that enhances the function of porphyrin, the porphyrin itself can be used as the hapten to induce the antibodies.

In order to generate antibodies with peroxidase activity, mice have been immunized against iron(III)-α,α,α,β-meso-tetrakis-orthocarboxyphenyl-porphyrin (Fe(ToCPP)) (Figure 1) [13], [14]. Two antibodies, 13G10 and 14H7, were found to bind the porphyrin hapten with nanomolar affinities and enhance its peroxidase activity. The 13G10-Fe(ToCPP) and 14H7-Fe(ToCPP) complexes catalyzed the oxidation of 2,2′-azinobis (3-ethylbenzothiazoline-6-sulfonic acid) (ABTS) 5 to 8 fold more effectively than the cofactor alone, with kcat reaching 540 min−1 and kcat/Km (ABTS) 6.2 104 M−1 min−1 at pH 4.6 for the 13G10 complex [15]. The antibodies protected the cofactor from oxidative degradation, allowing more than one thousand turnovers before destruction. In addition, it was shown that the 13G10-Fe(ToCPP) complex possessed a remarkably thermostable peroxidase activity and that it was able to use not only H2O2 as the oxidant but also a wide range of hydroperoxides [15], [16]. Compared with the peroxidase enzymes, the two antibodies possess a low affinity toward H2O2 (K = 9–16 mM) but a higher affinity than the cofactor alone (K = 42 mM) [13]. Because no amino acid was found to coordinate the iron atom in the 13G10 and 14H7/Fe(ToCPP) complexes [14], it was thought that an imidazole ligand could mimic the proximal histidine of peroxidases and enhance the catalytic activity. Fe(ToCPP) and less hindered iron(III)-tetraarylporphyrins, bearing only one or two carboxyphenyl substituents, were found to bind only one imidazole ligand when complexed to antibody 13G10, suggesting that the complexes are good mimics for peroxidases [16]. Whereas 50 mM imidazole inhibited the peroxidase activity of the 13G10-Fe(ToCPP) complex, it increased that of 13G10 in complex with porphyrins bearing two meso-[ortho-carboxyphenyl] substituents by a factor 15.

Figure 1

General structures of the cofactors used in this work.

R1α = R2α = R3α = R4β = COOH, R2β = H: iron(III)- α,α,α,β-mesotetrakis-orthocarboxyphenyl-porphyrin Fe(ToCPP) hapten used to induce the antibodies 13G10 and 14H7; R1α = R2α = COOH, R2β = R3α = R4β = H: αα-Fe(DoCPP); R1α = R2β = COOH, R2α = R3α = R4β = H: αβ-Fe(DoCPP); R1α = COOH, R2α = R2β = R3α = R4β = H: Fe(MoCPP).

To get insight into the structural basis for the different binding/activities of the variously substituted porphyrins, we have determined the crystal structure of the fragment antigen binding (Fab) of antibodies 13G10 and 14H7 and modeled their interaction with the different iron-porphyrins using molecular docking.

Materials and Methods

Purification and Characterization of Fabs 13G10 and 14H7

The antibodies were produced in ascitic fluid as described [13]. After ammonium sulfate precipitation, the IgGs (IgG1, λ) were loaded on a protein A column in 3 M NaCl, 1.5 M glycine pH 8.9 and eluted by 0.1 M citrate pH 5. The Fab 13G10 (resp. 14H7) was generated by papain digestion of the antibody at 37°C under standard conditions (30 mM Tris pH 7.4, 138 mM NaCl, 1.25 mM EDTA, 1.5 mM 2-mercaptoethanol) using a 3% papain to antibody ratio (w/w) and a 10 h (resp. 8 h) digestion time. Undigested IgG and Fc fragment were removed by DEAE anion exchange chromatography followed by gel filtration on a Sephacryl S100 HR column. The Fabs were further purified by ion exchange chromatography on a mono Q FPLC column by a NaCl gradient in 20 mM bistrispropane buffer at pH 7.2 for Fab 14H7 or in 20 mM diethanolamine buffer at pH 8.8 for Fab 13G10.

Sequence Determination

N-terminal sequencing of the H chain of 13G10 showed that the first three amino acid residues of the Fab were missing. mRNA from 13G10 hybridoma cells (4 106 cells) were isolated using a mRNA purification kit from Dynal. cDNAs were prepared in 50µl using the Omniscript Reverse Transcriptase kit from Qiagen and stored at −20°C until use. The heavy chain variable region fragment (about 420 bp) of each cDNA was amplified using a set of 12 forward primers MHV1 to MHV12 previously described [17] and reverse primer IgG1∶ 5′GGATCCCGGGCCAGTGGATAGACAGATG complementary to the beginning of the constant region. The λ-light chain fragment (about 690 bp) of each cDNA was amplified using a mixture of 2 forward primers, NL1∶ 5′ATGGCCTGGATTTCACTTATAC and NL2∶ 5′ATGGCCTGGACTCCTCTCTTC, corresponding to mouse λ-light chain leader sequence, and a mixture of 4 reverse primers, CL1 5′GCAGGAGACAGACTCTTCTCCAC, CL2 5′GCACGAGACAGACTCTTCTCCAC, CL3 5′GCAGGGGACAAACTCTTCTCCAC, and CL4 5′GCACGGGACAAACTCTTCTCCAC, complementary to mouse CLterminal sequence.

Each cDNA mixture (3 µl) was amplified with the Polymerase Chain Reaction [18] using TFL polymerase (Promega) on a MJ Research mini cycler with the following programm. A 10 min denaturation step at 94°C was followed by 30 cycles of 30 sec denaturation at 94°C, 45 sec hybridization at 59°C and 1 min 30 sec elongation at 72°C, a final step of 10 min at 72°C was performed to ensure completion of the amplification. Amplified products were purified on 0.8% low melting agarose (Sigma) and ligated into pGEM®Teasy vector (Promega). Electrocompetent E. coli TG1 cells {D(lac pro) supE thi hsdD5 F' traD35 proAB LacIq LacZDM15} were transformed with the ligation mixture by electroporation using a Cell porator electroporation system equipped with a Voltage Booster (Life technologies) according to manufacturer's recommendations. Plasmid DNAs were extracted from transformed cells and submitted to dideoxy sequencing [19]. For each antibody, clones were originated from at least two independant Polymerase Chain Reactions.

Crystallization

Crystals of 13G10 were grown in 26.5% PEG 2000, 0.2 M MgCl2, 0.1 M Tris pH 8.5, 10% glycerol. Crystals were flash cooled in a nitrogen stream at 100 K in the same solution containing 10% glycerol. Crystals of 14H7 were grown in 25% PEG 4000, 0.1 M ammonium acetate, 0.1 M sodium cacodylate pH 6.5. A single capillary-mounted crystal kept at 4°C was used for data collection. This explains the low redundancy for the 14H7 data.

X-Ray data collection and structure determination

Diffraction data for Fab 13G10 and Fab 14H7 were recorded at the ID14-1 station of ESRF and the LURE DW32 station, respectively. Data were processed with DENZO and SCALEPACK [20] (Table 1). The structures were solved by molecular replacement with the program AMoRe [21]; the models used were the Fv domain (PDB code 1mfa) and the CL-CH1 dimer (PDB code 1mfe) of the murine anticarbohydrate antibody Se155-4, which belongs to the same IgG1, λ class [22]. The atomic model of 13G10 was refined alternating cycles of model reconstruction with O [23] and refinement with CNS [24], whereas the atomic model of 14H7 was refined alternating cycles of model reconstruction with COOT and refinement with PHENIX [25] using the twin option. The final refinement statistics are given in Table 1. Several residues at the N-terminus of the heavy chain and CDRH1 of four heavy chains that were disordered were not included in the model. Figures were drawn with PYMOL. ThrL51 in CDRL2, which lies in the disallowed region of the Ramachandran plot, belongs to a γ turn, as commonly observed in all antibody structures.

Table 1

Data collection and refinement statistics for Fab 13G10 and 14H7.

Data collection	Fab13G10	Fab14H7
Space group	C2	P21
Number of molecules in the asymmetric unit	1	8
unit cell	a = 55.8 Å, b = 62.7 Å, c = 113.5 Å	a = 56.2 Å, b = 228.2 Å, c = 146.6 Å
	α = 90, β = 91.7°, γ = 90°	α = 90, β = 90.1°, γ = 90°
resolution	25–2.05 Å	20–2.55 Å
(outer resolution shell)	2.09–2.05 Å	2.64–2.55 Å
unique reflections	23580	111576
completeness*	93.7% (91.4%)	93.7% (87.8%)
mean I/sigma*	23.7 (4.6)	7.4 (2.1)
Rsym *	0.037 (0.19)	0.109 (0.51)
redundancy*	2.5 (2.1)	1.2 (1.2)
Refinement statistics
resolution	20.0–2.05	20.0–2.45
protein residues	425	3356
water molecules	188	359
glycerol molecules	3	–
Mg2+ molecule	1	–
Rcryst	0.224	0.279
Rfree ‡	0.275	0.328
Deviations from ideal geometry (rms)
bond length deviation (Å)	0.008	0.003
bond angle deviation (°)	1.52	0.75
B values
Average B value of protein atoms (Å2)	49.0	46.5
Average B value of water molecules (Å2)	51.1	25.5
Average B value of glycerol (Å2)	76.3	–
B value of Mg2+ (Å2)	45.0	–
Ramachandran plot
Most favored (%)	86.1	79.6
Additionally allowed (%)	11.4	17.5
Generously allowed (%)	1.1	1.4
Disallowed (%)	1.4	1.5

Values for highest-resolution shell are given in parentheses.

4.5% and 4.7% of the data were set aside for the Rfree calculation during the entire refinement.

Molecular Modeling

Quantum mechanical calculations were carried out to model the structures of the tetra-, bi- and mono substituted porphyrins. These structures were fully optimized with the density functional B3LYP [26], [27], as implemented in Gaussian09 [28]. The double-z basis set LANL2DZ [29] and its associated pseudo potential were used for the iron and the split valence 6–31 g** [30], [31] for the other atoms (C, N, H and O). Calculations were carried out for the high spin ferric species with the iron coordinated by the four porphyrin nitrogen atoms. The structures of the 13G10 and 14H7 antibodies were processed with the UCSF Chimera package [32] prior to docking in order to remove the water, ions and glycerol molecules, calculate the protonation states of the amino acids and add hydrogen atoms. Protein-ligand dockings were undertaken for both antibody structures following a recently published protocol that accounts for the presence of the metal in the ligand [33]. Calculations were carried out using the program GOLD (version 5.1) [34] and the Chemscore [35] scoring function. The metal ion was modeled thanks to an in house parameterization of an additional atom type in GOLD, which mimics the chelating capacity of the metal by interacting with amino acids with potential Lewis basis capacities. For the docking of imidazole, the parameters of the Chemscore scoring function used for the iron atom were that of cytochrome P450, as implemented in GOLD5.0. Because the Mg2+ ion lies at the interface between the hypervariable loops in the structure of 13G10 and probably mimics the iron of the porphyrin hapten, the docking calculations were performed using a 20 Å sphere around the Nδ atom of the neighboring residue AsnH33. Based on initial rigid dockings, full flexibility was allowed for TyrL34 using the Dunbrack rotamer library [36]. The bonds between the carboxyphenyl groups and the porphyrin moiety were frozen to maintain the correct α/β configuration. For each docking, an in house approach, based on a statistical analysis of a large PDB set of metal-bound proteins, was used in order to identify those residues that could directly bind to the metal. Based on the distances between the metal and the Cα carbons of all neighboring residues, no direct interaction was identified between the metal and the protein. Therefore, no further QM/MM refinements of the complexes were carried out [33].

Results

Amino Acid Sequence of Fab 13G10 and Fab 14H7

13G10 and 14H7 are murine monoclonal antibodies that were found to belong to the IgG1, λ-class. The sequences of their variable chains (Figure 2) differ by one residue in the light chain, outside of the complementary determining regions (CDRs) [37], and 17 residues in the heavy chain. In CDRH1, the sequences differ at positions H23 (Lys/Thr), H28 (Thr/Ser), H30 (Thr/Ser) and H34 (Met/Ile) for 13G10 and 14H7, respectively. CDRH2 has the same sequence in both antibodies. Their CDRH3, which is a main determinant of the combining site, has the same length but contains, in addition to Ser/Ala mutations at positions 93 and 97, two different amino acids with the same hydrophobic nature at position H100b (Ile/Leu) and H100c (Met/Leu).

Figure 2

Amino acid sequences of the VL and VH domains of 13G10 and 14H7.

Dots denote sequence identity to antibody 13G10. H denotes the heavy chain and L the light chain. Numbering of the antibody residues follows the Kabat nomenclature [40] and the definition of the hypervariable regions, indicated in bold, is from North et al. [37]. The sequence in italics has also been determined by Edman degradation of the aminoterminal part of the protein. The nucleotide sequences of 13G10 and 14H7 have been deposited in the Genbank database, except for the constant heavy chains whose sequences have not been determined: Genbank accession numbers 13G10H, ; 13G10L, for the mRNA and , for the corresponding protein sequence; 14H7H, ; 14H7L, for the mRNA and , for the corresponding protein sequence. The previously published amino acid sequences [14] were not those of antibodies 13G10 and 14H7. The germline sequences are indicated below the sequence of the antibodies.

Mature antibody genes are first formed by the assembly of the variable (V) and junction (J) gene segments for the light chain variable (VL) domain and by the assembly of the variable (V), junction (J) and diversity (D) gene segments for the heavy chain variable (VH) domain; affinity maturation then introduces into antibodies somatic mutations that increase binding affinity to the hapten or antigen [38]. For mice (Mus Musculus) antibodies, only three germline Vλsegments (compared with 180 Vκ segments), and four Jλ segments have been uncovered [39], which explains the poor diversity of the λ-light chain variable sequences compared with the κ-class. Compared with the germline Vλ1/Jλ1 sequence [40], there is one difference in the variable light chain of 13G10 and 14H7, Leu at position L96 (instead of Trp) in CDRL3 that arises from diversity at the junction between the V and J gene segments (Figure 2). In addition, 14H7 displays a Thr instead of a Lys at position L39, which is located outside the recombinant site. The nucleotide sequence of the 13G10 and 14H7 VH domains shows that they are derived from a gene in the J558 family, the largest VH family. The 13G10 VH sequence has highest homology to germline gene sequence J558.42 (98% identity at the nucleotide level) [41]. This translates into 94 (resp. 87) identical amino acid residues out 98 in VH for 13G10 and 14H7, respectively (Figure 2). The nucleotide sequence indicates that the VH region of antibodies 13G10 and 14H7 is encoded by the combination of J558.42 with a D gene segment that could not be identified because it is too short and JH4.

Comparison of the 13G10 and 14H7 VL and VH sequences to that of the germline genes indicates that 13G10 and 14H7 did not undergo much affinity maturation in the V segments. Among the rare somatic mutations, the ones that belong to the combining sites are LeuL96 (CDRL3), ValH50 (CDRH2) and SerH93 (CDRH3), for 13G10 (Table 2, Figure 2). On the other hand, the exceedingly hydrophobic nature of CDRH3 of 13G10 and 14H7 is due to extensive affinity maturation in the D segment so that the germline D segment could not be identified from the nucleotide sequence. Alanine at position H101 comes from a somatic mutation.

Table 2

Comparison of the aminoacids that contact small haptens in the structurally characterized λ-light chain antibodies.

PDB code		13G10/14H7*	Se155-41mfa [22]	CHA2551ind [79]	NC101etz [83]	88C6/121yuh [80]	2D12.51nc2 [85]	RS2-G192ntf [82]	10G61jnh [84]	N1G91ngp [81]
L32	CDRL1	Tyr	His	Tyr	Tyr	Tyr	Tyr	Tyr	Tyr	Tyr
L34	–	Asn	Asn	Asn	Ile	Asn	Asn	Asn	Asn	Asn
L91	CDRL3	Trp	Trp	Trp	Trp	Trp	Trp	Trp	Trp	Trp
L93	–	Ser	Asn	Ser	Ser	Ser	Ser	Ser	Ser	Ser
L94	–	Asn	Asn	Asn	Asn	Asn	Asn	Asn	Asn	Asn
L96	–	Leu	Trp	Trp	Trp	Trp	Trp	Trp	Phe	Trp
H33	CDRH1	Asn	Trp	Thr	Gly	Leu	Gly	Trp	Trp	Trp
H35	–	His	His	Ser	Gly	His	His	His	Gln	His
H47	FR2	Trp	Trp	Trp	Leu	Trp	Trp	Trp	Trp	Trp
H50	CDRH2	Val	Ala	Thr	Asp	Arg	Val	Thr	Ala	Arg
H52	–	Tyr	Tyr	Leu	Trp	Asp	Trp	Tyr	Tyr	Asp
H52A	–	Pro	Pro	Ser		Pro		Pro	Pro	Pro
H53	–	Gly	Asn	Gly	Asn	Asn	Ser	Gly	Gly	Asn
H56	–	Asp	Ala	Phe	Lys	Val	Gly	Asn	Asp	Gly
H58	–	Ser	Phe	Phe	Tyr	Lys	Ala	Tyr	Arg	Lys
H95	CDRH3	Gly	Gly	His	Arg	Tyr	Arg	Gly	Gly	Tyr
H96	–	Gly	Gly	Arg	Thr	Ala	Gly	Ser	Arg	Asp
H97	–	Ala/Ser	His		Phe	Tyr	Ser	Leu	Ser	Tyr
H98	–				Ser	Cys	Tyr	Tyr	Leu	Tyr
H99	–				Tyr	Arg	Pro	Tyr	Tyr	Gly
H100	–				Tyr		Tyr	Asn		Ser
H100	–				Tyr
H100	–				Gly
H100	–				Ser
H100	–				Ser			Asn
H100	–		Gly		Phe			Tyr
H100	–	Gly	Tyr		Tyr		Asn	Gly	Tyr	Ser
H100	–	Ile	Tyr		Tyr	Pro	Tyr	Trp	Thr	Tyr
H100		Met/Leu	Gly		Phe	Met	Phe	Phe	Met	Phe
H101		Ala	Asp	Val	Asp	Asp	Asp	Gly	Asp	Asp

Underligned letters are residues that form direct or water-mediated hydrogen-bonds or salt-bridge to the hapten. A bold letter indicates that the residue is in van der Waals contacts with the hapten.

For 13G10 and 14H7, the hapten has been modeled by docking (see Table 3).

Table 3

Statistical analysis of the lowest energy structures obtained in the docking calculations of the different cofactors in the 13G10 and 14H7 antibody structures.

Antibody	Ligand	Buried surface area (Å2)1	Score (kJ/mol)2	ΔG (kJ/mol)2	Shbond (kJ/mol)2	Hydrogen bondingInteractions(distances in Å)	Slipo(kJ/mol)2	Hydrophobic contacts	Sclash(kJ/mol)2	Clashe (distancesin Å)	Sinternal(kJ/mol)2	Iron distance(Å)3
13G10	Fe(ToCPP)	119.3	41.49	−49.13	3.38	β-COO−/SerH58 OH (2.95)α1- COO−/AsnH33 ND (3.08)α1- COO−/HisH35 NE (2.68)α1- COO−/GlyH96 NH (3.05)α1- COO−/GlyH100a NH (2.82)	276.65	TyrL32, AsnL34, TrpL91,AlaL93, AsnL94, LeuL96,AsnH33, HisH35, ValH50IleH51, TyrH52, AspH56,TyrH57, SerH58, GlyH95,GlyH96, AlaH97, GlyH100a,IleH100b, MetH100c	3.02	C38– SerH58HB (1.89)C64– TrpL91H (2.3)C64– TrpL91C (3.1)	4.62	1.62
	αα–Fe(DoCPP)	133.4	42.01	−45.26	3.19	α1- COO−/AsnH33 ND (2.97)α1- COO−/HisH35 NE (2.88)α1- COO−/GlyH96 N (2.87)α1- COO−/GlyH100a N (2.9)α2- COO−/AlaH97 N (3.7)	248.9	TyrL32, TrpL91, TyrH52,AsnH33, HisH35, ValH50,IleH51, AspH56, TyrH57,SerH58, GlyH95, GlyH96,AlaH97, GlyH100a, IleH100b	0.87	C70– GlyH100aHA(1.97)	2.38	0.83
	αβ–Fe(DoCPP)	125.87	40.51	−43.79	2.67	α- COO−/AsnH33 ND (3.06)α- COO−/HisH35 NE (2.85)α- COO−/GlyH96 N (2.85)α -COO−/GlyH100a N (2.6)	251.09	TyrL32, AsnL34, TrpL91,TyrH52, AsnH33, HisH35,ValH50, IleH51, AspH56,TyrH57, SerH58, GlyH95,GlyH96, AlaH97, GlyH100a,IleH100b	0.92	C58(H) – SerH58HB2 (2.15)	2.36	0.74
	Fe(MoCPP)	126.91	42.1	−43.84	2.64	COO−/AsnH33 ND (3.04)COO−/HisH35 NE (2.67)COO−/GlyH100a N (2.73)	252.46	TyrL32, AsnL34, TrpL91,TyrH52, AsnH33, HisH35,ValH50, AspH56, TyrH57,SerH58, GlyH95, GlyH96,AlaH97, GlyH100a, IleH100b	0.29	–	1.46	1.75
14H7	Fe(ToCPP)	101.53	31.54	−35.62	3.26	α1-COO−/TYRH52 OH (2.63)α2-COO−/TYRH32 OH (2.96)α2-COO−/SERH97 OH (2.96)β- COO−/ASNL94 ND (3.05)	164.57	TyrL32, TrpL91, AsnL94,TyrH32, AsnH33, HisH35,TyrH52, AsnH54, AspH56,ThrH57, SerH58, GlyH100a,LeuH100b	0.22	–	3.87	TyrH32 OH (4.5)
	αα–Fe(DoCPP)	116.34	29.84	−32.08	1.98	α1-COO−/SERH97 OG (2.58)α2-COO−/TYRH32 OH (2.76)	170.96	TyrL32, TrpL91, TyrH32,TyrH52, AspH56, GlyH100a,LeuH100b	0.12	–	2.12	TyrH32 (4.01)
	αβ–Fe(DoCPP)	102.66	29.43	−31.64	1.99	α–COO−/TYRH32 OH (2.73)α–COO−/SERH97 OG (2.80)	166.85	TyrL32, TrpL91, AsnL94,TyrH32, TyrH52, AsnH54,AspH56, ThrH57, SerH58,GlyH100a, LeuH100b	0.19	C70– SERH58 OH (2.1)	2.03	TyrH32 (4.13)
	Fe(MoCPP)	109.55	30.63	−32.17	1.88	COO−/TYRH52 OH (2.6)	174.54	SerL30, TyrL32, TrpL91,TrpL93, TyrH52, SerH97,GlyH100a.	0.15	–	1.4	TyrL32 OH (2.30)

The buried surface area was obtained by subtracting the molecular surfaces (calculated using the UCSF Chimera environment) of the nonbonded cofactor and of the antibody alone, from that of the complex and by dividing the result by 2.

The ChemScore scoring is defined as: Score = −(ΔG+Sclash+Sinternal) where the total free energy change that occurs upon ligand binding ΔG = −5.4800 −3.3400*Shbond −6.0300*Smetal −0.1170*Slipo +2.5600*H(rot). Sinternal is the energy term for the internal rotations of the cofactor, Smetal that for the metal interactions and Hrot that for the frozen rotable bonds.

For 13G10, distance between the iron atom in the modeled complex and Mg2+ in the crystal structure. For 14H7, distance to iron.

Determination of the Structures of Fab 13G10 and Fab 14H7

The crystal structures of Fab 13G10 and Fab 14H7 were determined at 2.05 Å and 2.55 Å, respectively (Table 1). Fab 13G10 crystallized with one molecule in the asymmetric unit, whereas the crystals of Fab 14H7 contained eight Fab molecules in the asymmetric unit and were pseudomerohedrally twinned [42], [43], [44]. Twinning was detected using the xtriage program included in PHENIX [25]. The cumulative local intensity deviation distribution statistics [45] did not indicate twinning: <|L|> = 0.488 (untwinned: 0.500; perfect twin: 0.375);  = 0.319 (untwinned: 0.333; perfect twin: 0.200). However, the results of the H-test [46] on acentric data gave a <|H|> value of 0.046 (0.50: untwinned; 0.0∶50% twinned) and a