Structural Insights into the Neutralization Properties of the Fully Human, Anti-interferon Monoclonal Antibody Sifalimumab*

Background: We investigated the molecular basis of human IFN-α2A recognition by sifalimumab. Results: We determined the structure of the complex between the Fab of sifalimumab and IFN-α2A. Conclusion: The interferon-neutralizing properties of sifalimumab result from direct competition for IFN-α2A binding to IFN receptor subunit 1 and not IFN receptor subunit 2. Significance: These data provide the basis for the mechanism of action of sifalimumab.

Interferons (IFNs) belong to the "4-helical cytokines" superfamily (1) and can be grouped into types I, II, and III. IFN-␥ and IFN-are the only known members of the type II and III IFNs, respectively (2,3), whereas type I IFNs constitute a family of cytokines expressed from more than 15 genes. Most notably, these include IFN-␣, IFN-␤, IFN-, IFN-, IFN-⑀, IFN-␦ and IFN-. The critical role of IFNs in modulating the host mammalian responses to infections has been well documented (4 -6). More recently, IFNs have also been shown to be key immunoregulatory cytokines. As such, they play a central role in the onset of various autoimmune diseases (7,8). Direct evidence includes the observation that autoimmune-predisposed mice deficient in the IFN-␣/-␤ receptor exhibit significantly reduced disease manifestations such as the presence of antierythrocyte autoantibodies, hemolytic anemia, anti-DNA autoantibodies, and kidney disease (9). In particular, systemic lupus erythematosus, type I diabetes, and Sjögren syndrome, as well as thyroid diseases, have now been linked to the action of IFN-␣ (10). The existence of at least 13 subtypes within the IFN-␣ family (11) further complicates a thorough understanding of these pathways. To contribute to the treatment of autoimmune diseases, AstraZeneca/MedImmune has developed sifalimumab, a fully human monoclonal antibody that binds to, and inhibits the actions of multiple IFN-␣ subtypes.
We sought to understand the molecular basis of human IFN-␣2A recognition by sifalimumab. For this purpose, we solved the x-ray crystal structure of the complex between the Fab fragment of this antibody and IFN-␣2A. The structures of several type I human IFNs (e.g. IFN-␣2A, IFN-␣2B, and IFN-␤), unbound or bound to a single chain Fv (scFv), have already been determined using either x-ray crystallography or nuclear magnetic resonance (NMR) (12)(13)(14)(15)(16). However, this study describes the first three-dimensional structure of a human type I IFN bound to a therapeutic antibody currently in human. Our data permitted us to describe in detail the corresponding interface and provide a molecular understanding the interferon-neutralizing properties of sifalimumab.
Protein Expression, Purification, Crystallization, and X-ray Data Collection-Detailed purification, crystallization, and data collection procedures have been previously described (18). In short, crystals of the sifalimumab Fab⅐IFN-␣2A complex diffracting to 3.0 Å were obtained using vapor diffusion. The orthorhombic crystals belonged to the I222 space group with unit cell parameters a ϭ 134.82, b ϭ 153.26, c ϭ 163.49 Å. The crystals exhibited a relatively loose packing with a solvent content and Matthew's coefficient of 59.3% and 3.02 Å 3 Da Ϫ1 , respectively. Two sifalimumab Fab⅐IFN-␣2A complexes were in the asymmetric part of the unit cell.
Structure Determination and Refinement-Diffraction images were integrated and scaled using HKL 2000 (19). Molecular replacement, refinement, and electron density calculation were completed via the CCP4 (Collaborative Computational Project Number 4) program suite (20). The crystal structure of the sifalimumab Fab⅐IFN-␣2A complex was solved using molecular replacement and refined at 3.0-Å resolution. The search model for sifalimumab Fab consisted of the Fab portion of another antibody from AstraZeneca/MedImmune whose structure was determined at 2.17-Å resolution (21). The sequence identities between the Fab portions of sifalimumab and the search model were 95.3 and 78.6% for the light and heavy chains, respectively. The non-identical amino acids were first modeled as alanine during the molecular replacement procedure and initial refinement/model building rounds. A very clear solution was obtained for the 2 sifalimumab Fab molecules in the asymmetric unit using both PHASER (22) and MolRep (23). For the IFN-␣2A portion, 3 human type I IFN structures were available in the Protein Data Bank (PDB) (24) at the time of the study (2008). These corresponded to PDB codes 1ITF (human IFN-␣2A exhibiting 100% sequence identity with IFN-␣2A of this study; NMR-solved), 1RH2 (human IFN-␣2B exhibiting 99% sequence identity with IFN-␣2A of this study, x-ray-solved at 2.9 Å resolution), and 1AU1 (human IFN-␤ exhibiting 39% sequence identity with IFN-␣2A of this study, x-ray solved at 2.2-Å resolution). None of these 3 potential models yielded a clear molecular replacement solution with PHASER or Mol-Rep. However, the phases obtained through the solution of both sifalimumab Fab molecules yielded very clear electron density for the proximal region of IFN-␣2A. Two rounds of Fab-only refinement and model adjustment using the "O" software (25) further improved the electron density quality of the antigen and made it possible to build 3 of 5 helices manually. The resulting partial model was then superimposed on the structure of human IFN-␣2B (PDB ID 1RH2), which differed from IFN-␣2A by only one amino acid (R23K). The tight non-crystallographic symmetry restraints were used throughout the refinement of the model with Refmac5 (26). The substituted alanine residues were changed to their respective bona fide counterparts when permitted by the corresponding electron densities. The first 14 amino acids of both IFN-␣2A molecules in the asymmetric unit were built last, because of the larger conformational differences. The latter may provide a reasonable explanation for not obtaining a clear solution during the molecular replacement procedure. Upon completion, the model was analyzed using the TLS Motion Determination (TLSMD) program running on its Web server (27,28). Further refinement was carried out in TLS and restrained refinement mode using Refmac5. For this purpose, each of 3 different polypeptides were divided into 4 parts in accordance with results from the TLSMD server. More precisely, sifalimumab light chain was divided into sections corresponding to residues 1-40, 41-106, 107-177, and 178 -215. Likewise, the sifalimumab heavy chain was divided into sections corresponding to residues 1-62, 63-129, 130 -185, and 186 -219. Finally, IFN-␣2A was divided into sections corresponding to residues 1-20, 21-51, 52-113, and 114 -157. The same portions were used for NCS restraints application. As we previously described (18), the IFN-␣2A antigen used in this study included one extra threonine and one extra serine residue on its N-terminal end (numbered Ϫ1 and 0, respectively). Amino acids Ϫ1, 0, and 1, along with N⑀2 of His-7, were found to coordinate a total of 2 Ni 2ϩ ions (first and third). Another two Ni 2ϩ ions (second and fourth) were coordinated by His-190 in the sifalimumab light chain. The fifth and sixth Ni 2ϩ ions did not have histidine residues in the coordination sphere. Ni 2ϩ ions were the only divalent metal ions present in the crystallization mixture, and could be identified by their fit to the corresponding electron density.
Analysis of Sifalimumab Binding to IFN-␣2A-The interaction of immobilized IFN-␣2A with sifalimumab was monitored using a KinExA 3000 instrument (Sapidyne Instruments, Boise, ID). IFN-␣2A was first coated onto UltraLink Biosupport beads (Pierce, Rockford, IL) at concentrations of 5 and 10 g/ml in 0.05 M NaHCO 3 , pH 9.0, overnight at 4°C according to the manufacturer's instructions. Coated beads were then separated from unreacted IFN-␣2A using a gentle pulse spin and blocked for ϳ15 min at 22°C with 1 M Tris, pH 8.0, bovine serum albumin, 10 mg/ml. The slurry was then spun and the blocking solution removed. The blocking step was repeated for 2 h at 22°C. Beads were then resuspended in 27 ml of run buffer (phosphate-buffered saline (PBS), pH 7.4, 0.02% NaN 3 ) and packed into a column. Typically, sifalimumab was prepared at concentrations of 40 and 200 pM. IFN-␣2A was then titrated across these IgG solutions at concentrations of 313 fM to 16 nM and 980 fM to 50 nM, respectively, and incubated for 1-4 days at room temperature. The amount of free IgG in the samples was derived from the fluorescence signal obtained after the passing of Cy5-labeled goat anti-human IgG F(abЈ) 2 (typically 1 g/ml; Jackson ImmunoResearch Laboratories, West Grove, PA) through the column. The dissociation constant (K D ) was determined by fitting the individual equilibrium titration data to a 1:1 binding model using the KinExA Pro 1.0.3. software.

Structure of a Therapeutic mAb Bound to IFN-␣2A
Receptor-ligand Competition by Sifalimumab-The ability of sifalimumab to inhibit the human IFNAR1/IFN-␣2A and IFNAR2/IFN-␣2A interactions was monitored using a ProteOn XPR36 instrument (Bio-Rad). The extracellular domain of IFNAR1 (MedImmune) was immobilized to the EDAC/Sulfo-NHS-activated surface of on a GLC biosensor chip (Bio-Rad) using standard amine coupling (200 nM in 10 mM sodium acetate buffer, pH 5.0) at a density of ϳ4,800 resonance units according to the manufacturer's instructions. The extracellular domain of IFNAR2 (MedImmune) was also immobilized using standard amine coupling (50 nM in 10 mM sodium acetate buffer, pH 4.0) at a density of ϳ2,000 resonance units. IFN-␣2A and sifalimumab were prepared in PBS, pH 7.4, containing 0.005% Tween 20. Sifalimumab competition was assessed by 2 consecutive injections of IFN-␣2A and a mixture of IFN-␣2A and sifalimumab over the IFNAR1 or IFNAR2 surfaces. For IFNAR1, IFN-␣2A was first injected at 200 g/ml (30 l/min for 120 s), which was followed by a second injection (30 l/min for 120 s) of the IFN-␣2A (200 g/ml)/sifalimumab (100 g/ml) mixture. For IFNAR2, IFN-␣2A was first injected at 10 g/ml (30 l/min for 120 s), which was followed by a second injection (30 l/min for 120 s) of the IFN-␣2A (10 g/ml)/ sifalimumab (100 g/ml) mixture. The extent of competition was derived from the additional binding detected from the second injection. All sensorgram data were processed by ProteOn Manager 3.1 software (Bio-Rad), and the binding graphs were prepared with Prism (GraphPad).

Results and Discussion
Sifalimumab Fab/IFN-␣2A Three-dimensional Structure-We successfully determined the x-ray crystal structure of the complex between the Fab of an anti-human IFN-␣ therapeutic antibody (sifalimumab) and IFN-␣2A. The corresponding refinement statistics are given in Table 1. Two sifalimumab Fab fragments in the asymmetric unit superimposed with an r.m.s. deviation of 0.31 Å (maximum displacement of 1.3 Å was for C␣/136 in the heavy chain and 1.2 Å for C␣/204 in the light chain). This value is well within the estimated overall coordinate error value of 0.34 Å. In addition, the elbow angles were calculated for both molecules as described (29) and separately estimated at 170.9°and 172.5°. These values are again well within the significance limit of 2-3° (29). Thus, we concluded that the 2 sifalimumab Fab molecules in the asymmetric unit were essentially identical. Both antigen molecules (IFN-␣2A) exhibited a r.m.s. deviation of 0.6 Å when superimposed. However, the greatest differences occurred away from the Fab/IFN-␣2A interface, and close to the IFN-␣2A C-terminal region

Structure of a Therapeutic mAb Bound to IFN-␣2A
(maximum displacement of 3 Å was for C␣/156). As with sifalimumab Fab, both antigen molecules in the asymmetric unit could be considered essentially identical. Therefore, all subsequent descriptions of the antibody/antigen interface were made using 1 of the 2 complexes (namely chains A, B, and C in our PDB ID 4YPG). The interface contributed by the sifalimumab Fab portion could best be described as a canyon, whereas IFN-␣2A exhibited a remarkable shape complementarity to this groove as indicated in Fig. 1A. A shape complementarity of ϳ0.674 between sifalimumab Fab and IFN-␣2A was calculated using the "sc" program from the CCP4 suite (Collaborative Computational Project Number 4) (20). This represents a very high degree of complementarity between both partners. For comparison purposes, the shape complementarity between sifalimumab Fab heavy and light chains was estimated at 0.669. The charge com-plementarity between sifalimumab Fab and IFN-␣2A is provided in Fig. 1, B and C. The basic surfaces of both variable regions of the antibody exhibited very good complementarity to the acidic surface of the antigen. Finally, we also noted that the side of the antigen of the sifalimumab Fab/IFN-␣2A contact interface was formed mainly by parts of B (residues 49 -69), C (residues 78 -101), and D (residues 112-133) helices as illustrated in Fig. 1D.
The buried surface area upon formation of the complex between sifalimumab Fab and IFN-␣2A was estimated at more than 1,300 Å 2 (as calculated from the PISA server). The contact interface included ϳ20 amino acids from each of the antibody polypeptides and ϳ40 amino acids from the antigen. In particular, the heavy and light chains of sifalimumab Fab covered ϳ600 and 700 Å 2 of solvent-accessible area upon complex formation, respectively. The list of all hydrogen bonds between both chains of sifalimumab Fab and IFN-␣2A can be found in Table 2. It is worth noting that the heavy and light chain CDRs of sifalimumab Fab contributed unequally to the interaction with IFN-␣2A. Indeed, 4 intermolecular hydrogen bonds involved the heavy chain first CDR (CDRH1, Fig. 2A and Table  2), whereas 9 intermolecular hydrogen bonds involved the light chain first CDR (CDRL1; Fig. 2B and Table 2). Asn-55/N␦2 in the second CDR of the sifalimumab Fab heavy chain (CDRH2) made a lone hydrogen bond with IFN-␣2A Asp-2/O␦2 (Fig. 2C CDR3 (E). In all panels, the antigen is shown in red. Sifalimumab residues were numbered consecutively. Dotted lines represent hydrogen bonds.

TABLE 3 Affinity measurement for the binding of sifalimumab to IFN-␣2A
The dissociation constant (K D ) was determined using a KinExa instrument as described under "Experimental Procedures."

K D pM
Sifalimumab 44 27-65 a The 95% confidence interval indicated the range over which the measured K D is thought to vary due to the reproducibility of the instrument. The residual error between the fitted and theoretical curves was 2.7%.
and Table 2). The contribution of the second CDR of the sifalimumab Fab light chain (CDRL2) for binding to IFN-␣2A appeared somewhat less specific, in that all corresponding intermolecular hydrogen bonds only involved the main chain CDRL2 atoms of Gly-51 and Arg-55 ( Fig. 2D and Table 2). Finally, the spatially close CDRH3 and CDRL3 of the sifalimumab Fab created 2 intermolecular hydrogen bonds each with the antigen (Fig. 2E and Table 2). In CDRL3, Tyr-92/O and Arg-97/N2 created one hydrogen bond each with IFN-␣2A Gln-61/N⑀2 and Glu-96/O⑀1, respectively. CDRH3 was only involved in contacts with IFN-␣2A Glu-96/O⑀1,2 through its main chain Ile-101/N atom. Interestingly, 2 intermolecular hydrogen bonds were mediated by one residue in the second framework of the sifalimumab light chain (Tyr-50). In summary, sifalimumab CDRL1 and CDRH1 made the largest contribution to formation of the high affinity complex between IFN-␣2A and sifalimumab (K D shown in Table 3). Importantly, none of the side chains in sifalimumab CDRL2 and CDRH3 were involved in hydrogen bonds with IFN-␣2A. Radhakrishnan et al. (12) observed that human IFN-␣2B (99% identical to human IFN-␣2A due to one amino acid difference, namely K23R) can form Zn 2ϩ ion-mediated homodimers. Our model did not indicate such a dimerization mechanism because no ions were found to be shared between 2 contacting IFN-␣2A molecules (despite the presence of Ni 2ϩ in the crystallization mixture). This comforts the notion that the active form of IFN-␣ molecules is monomeric, as determined by Klaus et al. (13).
Implications for Sifalimumab Mechanism of Action-The type I IFN receptor (shared by all human type I IFNs) comprises 2 major transmembrane subunits, namely IFNAR1 and IFNAR2 (30,31). IFN-␣ binds to IFNAR2 with a much faster k on and slower k off than those measured for IFNAR1 (32,33). Therefore, a two-step assembling mechanism was proposed for formation of the tertiary IFN signal complex, in which IFN-␣ first binds IFNAR2 and then recruits IFNAR1 (32,34). The present study provides important clues related to sifalimumab mechanism of action.
A number of studies have revealed critical residues on both IFNAR2 and IFN-␣2 through mutagenesis, NMR, and x-ray crystallography (35)(36)(37)(38)(39)(40). As can be observed when superimposing the x-ray structure of the human IFN-␣2⅐IFNARI⅐IFNAR2 ternary complex (PDB ID 3SE3) (40) with the sifalimumab⅐IFN-␣2A complex (Fig. 3), both sifalimumab and human IFNAR2 bind to opposite sides of the IFN molecule. In addition, various human IFN-␣2 residues identified by Piehler et al. (37) as critical for binding to IFNAR2 (namely Leu-30, Arg-33, Arg-144, Ala-145, Met-148, and Arg-149) were also found to be on the opposite side of the sifalimumab binding site (Fig. 3). These data agree with the observation that sifalimumab can bind to IFN-␣2A⅐IFNAR2 complexes (Fig. 4A), and, thus, rule out a mechanism of action in which the antibody interferes with the corresponding receptor/ligand interaction.
In summary, we conclude that sifalimumab acts as a direct competitive inhibitor of the IFN-␣2/IFNAR1 interaction. A coherent model of the mechanism of action for sifalimumab thus emerges, and indicates the antibody does not interfere with the first step in the response to IFN-␣2, namely the ligand⅐IFNAR2 complex formation. Rather, we suggest that sifalimumab sterically interferes with the recruitment of IFNAR1 and prevents formation of the IFN-␣2⅐IFNAR1⅐IFNAR2 ternary complex.