Structural basis of the broadly neutralizing anti-interferon-α antibody rontalizumab

Abstract

Interferons-alpha (IFN-α) are the expressed gene products comprising thirteen type I interferons with protein pairwise sequence similarities in the 77–96% range. Three other widely expressed human type I interferons, IFN-β, IFN-κ and IFN-ω have sequences 29–33%, 29–32% and 56–60% similar to the IFN-αs, respectively. Type I interferons act on immune cells by producing subtly different immune-modulatory effects upon binding to the extracellular domains of a heterodimeric cell-surface receptor composed of IFNAR1 and IFNAR2, most notably anti-viral effects. IFN-α has been used to treat infection by hepatitis-virus type C (HCV) and a correlation between hyperactivity of IFN-α-induced signaling and systemic lupus erythematosis (SLE), or lupus, has been noted. Anti-IFN-α antibodies including rontalizumab have been under clinical study for the treatment of lupus. To better understand the rontalizumab mechanism of action and specificity, we determined the X-ray crystal structure of the Fab fragment of rontalizumab bound to human IFN-α2 at 3Å resolution and find substantial overlap of the antibody and IFNA2 epitopes on IFN-α2.

Introduction

Systemic lupus erythematosis (lupus) is a chronic disease of dysregulated immune function characterized by the presence of self-reactive antibodies and multi-organ impacts. The causes of lupus are poorly understood and there is currently no cure.^1,2 Chronic use of broadly immune-suppressive treatments effective for the wide range of clinical manifestations, for example, corticosteroids, can produce very serious side effects, even contributing to mortality.³ During the past two decades, medical research into autoimmunity has revealed several molecular targets in the immune system worthy of consideration for safer chronic treatment of lupus, including members of the tumor necrosis factor superfamily (e.g., BLyS),⁴ regulatory members of immune activation systems⁵ (CTLA4⁶, B7RP-1⁷), cytokines secreted from activated immune cells (e.g., interleukin-6⁸), B-cell antigens⁹ (CD20, CD22) and interferons-alpha (IFN-α).^10,11 In 2011, the anti-BLyS antibody belimumab was approved by the United States FDA for treatment of lupus, and the other targets remain subjects of clinical research².

A potential causative role for interferons (IFNs) in lupus arose when use of interferon-α2 (IFN-α2) for treatment of blood cancer¹² or hepatitis-virus type C (HCV)¹³ was shown to correlate with induction of lupus-like symptoms. IFN-α2 is one of several expressed human gene products with protein sequence homologies between 77 and 96%, referred to as subtypes of IFN-α. These and IFN-β and IFN-ω are widely expressed and classified as Type I interferons based on their recognition by the cognate heterodimeric cell-surface receptor IFNAR1/IFNAR2. More restricted expression is seen for two other Type I interferons, IFN-κ and IFN-ε. The array of interferons classified as Type I grows more complicated when considering other species.

Because of the association between IFN-α and lupus, the mouse antibody 9F3 was raised using a mixture of interferons (leukocyte interferon-α) as immunogen followed by antibody screening for broad reactivity with individual purified IFN-α subtypes. The humanized version, rontalizumab (formerly 9F3v13) is broadly neutralizing of IFN-αs but not other Type I interferons.¹⁴ During study of rontalizumab, we grew interested in the specific residues on IFN-α contacted by rontalizumab to learn the molecular basis for reactivity with human IFN-α subtypes, orthologues and other Type I interferons. A detailed picture of interactions between IFN-α2 and its receptor has been reported by Thomas et al.,¹⁵ allowing us the understand the molecular mechanism by which rontalizumab prevents binding of IFN-α to the receptor. We report here the X-ray crystal structure of the Fab fragment of rontalizumab (ronta-Fab) bound to human IFN-α2 at 3Å resolution, in a monoclinic lattice with 8 complexes in the crystallographic asymmetric unit.

Results

The arrangement of the 8 IFN-α2/ronta-Fab complex molecules within the asymmetric unit was assessed using inter-complex contact regions for each complex which occluded the greatest area from the bulk solvent region. We found local 2-fold axes and local translational symmetry (Fig. 1). Complexes “1” and “3” are related by a local 2-fold axis (about 5° from parallel to the ac plane), as are complexes “6” and “4.” These two local rotation axes are almost coincident, and the two pairs are separated by a 47Å translation. The interface buries about 950Ų of surface on each complex, mostly between the C-terminal ends of the Vκ domains, and between CH1 and the neighboring IFN-α2 [Fig. 1(A)]. Complexes “5” and “8” share a close crystal packing contact of about 780Ų that is not a rotation axis, mediated by Vκ and IFN-α2 from complex “5” and the complex “8” heavy chain in a region around the link between VH and CH1 domains [Fig. 1(B)]. This interface also characterizes the close crystal packing contact between complexes “2” and “7,” except that complex “7” includes no atoms for either CL or CH1 domains due to poor electron density. The complex pairs 5/8 and 2/7 are related by a 49Å translation. The 7 Fabs for which all domains are part of the final model share an elbow angle close to 134° (Table I), and there is a volume that could accommodate the constant region from complex “7” with another instance of this elbow angle. Two additional constant regions are characterized by relatively weak electron density and relatively high thermal factors. Non-crystallographic symmetry (NCS) restraints provided a beneficial effect on Rfree, but elbow angles were not part of these restraints. Refinement experiments guided application of NCS restraints. The average of all pairwise root-mean-square deviations of Cα atoms after superpositioning of variable regions and IFN-α2 is 0.44Å with a standard deviation of 0.16Å. Since best results were obtained with relatively tight restraints, it is an appropriate and useful simplification to describe only a single complex molecule in detail (“4” = chains D, J, and K for IFN-α2, light chain and heavy chain, respectively).

Figure 1.

Interactions between members of the crystallographic asymmetric unit of IFN-α2/ronta-Fab. (A) Complexes “1” and “3” viewed along a local 2-fold axis (light chains magenta, heavy chains yelllow, IFN-α2 blue or green). The complexes “4” and “6” bear a similar relationship. (B) Complexes “5” and “8” show a contact between complex “5” light chain (magenta) and complex “8” heavy chain (orange). The complexes “2” and “7” bear a similar relationship.

Table I.

Summary Metrics for the Eight Complexes in the Asymmetric Unit

	Average B-factor by domain (Å2)					Elbow angle (°)
Complex	IFN-α2	Vκ	CL	VH	CH1
1	65	51	58	56	57	132
2	89	78	144	83	135	136
3	78	54	55	64	54	134
4	77	54	58	58	55	137
5	85	72	113	81	102	128
6	76	57	62	62	57	136
7	84	82	–	83	–	–
8	78	84	94	79	105	131

IFN-α2 is a five α-helical bundle protein¹⁶ with one long cross-over extended segment between helices A and B (Fig. 2). The long axes of the helices are close to parallel and anti-parallel. The overall structure of IFN-α2 in the ronta-Fab complex is very similar to those prior instances where an IFN-α has been characterized in complex with one or both members of its receptor¹⁵ (PDB entries 3S9D and 3SE3, respectively) or with a different antibody fragment¹⁷ (PDB 3UX9), superposition of Cα atoms yielding root-mean-square deviations of about 0.5Å after discounting discordant termini and residues in poorly ordered loop regions. The ronta-Fab paratope (antigen combining surface) is a broad groove lying between CDR-H2 and the relatively long CDR-L1 (Fig. 2). The six ronta-Fab complementarity determining region (CDR) loops (L1, L2, L3, H1, H2, and H3) can be classified¹⁸ as L1-15-1, L2-8-1, L3-9-cis7-2, H1-13-1, H2-10-1 and H3 anchor-1, respectively. The class we find for CDR-L3 is relatively rare and is the only Proline-containing L3 class of this length that does not usually start with a Gln-Gln dipeptide sequence¹⁸; ronta-Fab starts Gln-His. Otherwise the loop conformations are the most common for their length and the structural correspondence is high except for CDR-L1. As discussed below, CDR-L1 is a key contact for IFN-α2 which may be associated with moderate excursions from the established conformational norm.

Figure 2.

The ronta-Fab antigen-combining region is a broad groove formed by protrusions of the relatively long CDR-L1 on one side and CDR-H2 on the other. The contacted parts of IFN-α2 are α-helices and an extended segment running approximately parallel to this broad groove. Ronta-Fab residues are colored as for Figure 3.

IFN-α2 is found lying in the broad groove described above, with the longer helical axes running parallel to it, along the interface between the light chain and heavy chain. Most contacts in IFN-α2 are on the long A-B cross-over segment and helices D and E (Fig. 3). The IFN-α2 surface in contact with IFNAR2 in the receptor complex structure¹⁵ includes all these same elements, and we conclude rontalizumab exerts its biochemical and biologic effects by blocking receptor binding. Total IFN-α2 surface area buried by ronta-Fab¹⁹ is 1030Å^2. The antibody light chain and heavy chain contributions to the contact surface are unequal: the light chain contributes 610Ų and the heavy chain 420Ų. This imbalance is in the opposite direction than is typically observed for antibody-protein interfaces,²⁰ and arises from a large contribution from CDR-Ll and relative small contributions from CDR-H3 and CDR-H1. The three side chains providing the most surface are from CDR-L1 (Tyr30, 129 Ų), CDR-L3 (Trp92, 79 Ų) and CDR-H2 (Tyr54, 85Ų) (Fig. 4). Overall, the amino acid sidechains from IFN-α2 and ronta-fab involved in binding are variable in their nature (polar, non-polar and charged), so that H-bonding, van der Waals, and salt-bridge interactions are all present. The shape complementarity statistic,²¹ Sc, of the IFN-α2/ronta-Fab interface is 0.64, a typical value for antigen/antibody complexes. The binding affinities of rontalizumab for the IFN-αs shown in Table II span three orders of magnitude from 18 pM to 35 nM. Even the weakest affinity, for IFN-α10, is consistent with the size and nature of the interaction which the structure with IFN-α2 suggests should be highly relevant for all of them.

Figure 3.

Ronta-Fab paratope. Residues are colored by solvent-accessible surface area lost when bound to IFN-α2. Red = more than 60Ų, orange = 40–60 Ų and yellow 20–40 Ų. Elements of IFN-α2 nearest the Fab are shown in semi-transparent green.

Figure 4.

Close-up view of interactions between ronta-Fab (light chain light blue, heavy chain dark blue) and IFN-α2 (green) for distances up to 4Å. Ronta-Fab residue labels are underlined. Dotted grey lines are potential H-bonds with distances between depicted atoms up to 3.3Å. Ronta-Fab side chains are colored according to surface area buried by proximity to IFN-α2 (see Figure 3).

Table II.

Binding Analysis of Rontalizumab for Interferons

interferon	ka (Ms−1)	kd (s−1)	KD (nM)
IFN-α2	2.02 × 106	3.71 × 10−5	0.018
IFN-α8	8.11 × 105	1.62 × 10−5	0.020
IFN-α5	1.07 × 106	3.37 × 10−5	0.031
IFN-α21	1.11 × 106	5.12 × 10−5	0.046
IFN-α17	9.91 × 105	3.24 × 10−4	0.33
IFN-α14	6.31 × 105	3.26 × 10−4	0.52
IFN-α16	1.31 × 106	9.42 × 10−4	0.72
IFN-α1	4.66 × 105	3.89 × 10−4	0.83
IFN-α7	1.88 × 105	2.43 × 10−4	1.3
IFN-α4	8.51 × 105	5.52 × 10−3	6.5
IFN-α6	8.94 × 105	9.76 × 10−3	10.9
IFN-α10	6.4 × 105	2.27 × 10−2	35.4
IFN-β	no binding

Discussion

Biological effects of rontalizumab arise from preventing the interaction between IFN-αs and their heterodimeric cell-surface receptor IFNAR1/IFNAR2. The X-ray structure of the ternary complex between IFN-α2 and this receptor has been determined.¹⁵ With the IFN-α2/ronta-Fab structure we report here, and assuming the other IFN-αs are bound in very similar ways, the molecular mechanism of IFN-α blockade by rontalizumab is made clear (Fig. 5). There is significant overlap in the regions of IFN-α2 bound by IFNAR2 and ronta-Fab (Fig. 6); it is simply impossible for an IFN-α to bind IFNAR2 when it is already bound by rontalizumab. This mode of action for an anti-IFN-α antibody has been observed previously, in the complex between a single-chain Fv and IFN-α1¹⁷ (Fig. 6).

Figure 5.

Ronta-Fab interferes with IFN-α2 receptor binding. After superposition of IFN-α2 (green), the extracellular domains of receptor component IFNAR2 (magenta) occupies much of the same volume occupied by ronta-Fab (blue surfaces). IFNAR2 is taken from PDB entry 3SE3, which also includes three domains of IFNAR1 (not shown) which do not overlap with ronta-Fab. Signaling by IFN-α2 requires the intact heterodimeric IFNAR1/IFNAR2 receptor, thus binding of rontalizumab is inconsistent with IFN-α2 signaling.

Figure 6.

Structural epitopes on IFN-α from interaction with ronta-Fab, IFNAR2 and and an anti-IFN-α1b single-chain Fv. Minor differences in the shape of IFN-α surfaces arise from details of residues that are represented in electron density or are flexible and some sequence differences. Both ronta-Fab and the single-chain Fv have contact epitopes that show strong overlap with the IFNAR2 binding surface.

The intuitive logic of mating a helical bundle and antibody via the helical long-axes and a broad groove between light and heavy variable domains, as found for the IFN-α2/ronta-Fab complex, is poorly consistent with the body of experimental findings from among other helical-cytokine/antibody structures. Among 12 such structures in the Protein Data Bank (PDB), Fab complexes of interleukin-34²² (IL-34) (PDB entries 4DKE and 4DKF) are also arranged in approximately the same orientation. On the other hand, Fab complexes of IL-13^23,24 (PDB entries 3G6D and 4I77) have helical axes approximately perpendicular to the V_L/V_H interface. The most common general arrangement has the helical “ends” and associated loops in proximity to the Fab paratope (e.g., complexes of IL-23²⁵ (PDB entry 3D85), thrombopoietin²⁶ (PDB entries 1V7N and 1V7M), IL-10²⁷ (PDB entry 1LK3) as well as some additional IL-13 complexes²⁸ [PDB entries 3L5W, 3L5Y, 3L5X)]. The scFv complex of IFN-α1¹⁷ (PDB entry 3UX9) shows the helical axes and V_L/V_H interface at about a 45° angle. There is a preponderance among these examples of antibodies selected for their recognition of epitopes with a targeted biological role, but there should be no doubt that highly specific and high affinity antigen recognition by antibodies can involve essentially any combination of primary, secondary and tertiary structural features.²⁰

Among Type I interferons, ronta-Fab is broadly specific for the 13 human IFN-α subtypes.¹⁴ Structural reasons for poor reactivity with the other Type I interferons are easy to find. Using the known structure of IFN-ω¹⁵ (PDB entry 3SE4), three changes (Ala → His at residue 19, Ala → Met at residue 145 and Ser → Phe at residue 152 (IFN-α2 numbering)) each may create a deleterious steric overlap with CDR-L1, and their combination probably makes a conformational accommodation by CDR-L1 even more energetically disfavored [Fig. 7(A)]. For human IFN-β²⁹ (PDB entry 1AU1), the Alanine → Tryptophan difference at residue 19 and the Ser → Tyr difference at residue 152 are intolerant of CDR-L1 of ronta-Fab [Fig. 7(B)]. Additionally, the Glu→Thr difference at residue 141 eliminates a probable H-bond to light chain Tyr30. There is not a deposited molecular structure of IFN-κ, but amino acid sequence analysis suggests some top candidates for relatively poor reactivity with ronta-Fab; Met → Arg at residue 148 is close to CDR-L1, and the Ser → Tyr at residue 152 is the same difference argued above for IFN-β. Similar reasoning allows us to predict that ronta-Fab would react poorly to with mouse IFN-αs, based on the presence in almost all subtypes of a Met → Trp difference at residue 148, as well as relatively low overall sequence homology to human IFN-α2 (∼60%).

Figure 7.

(A) Structural rationale for lack of binding of IFN-ω by ronta-Fab. IFN-ω from PDB entry 3SE4 (beige) is shown after superposing it on IFN-α2 (green), with concentration on the region near ronta-Fab CDR-L1 (light blue). Only three side chains of IFN-ω are shown (yellow labels), where they differ from those of IFN-α2 and are consistently much larger. These differences, which are incompatible with the conformation of CDR-L1, are very likely to be among the causes for a lack of reactivity between rontalizumab and IFN-ω. (B) Structural rationale for lack of binding of IFN-β by ronta-Fab. IFN-β from PDB entry 1AU1 (orange) has steric conflicts similar to those shown in part A.

Despite the very significant overlap between their epitopes on IFN-α2 (Fig. 6), the important functional distinction between ronta-Fab and IFNAR2 is manifest in their contrasting abilities to recognize Type I interferons IFN-β, IFN-κ and IFN-ω. IFNAR2 recognizes all type I interferons as part of their cognate receptor (IFNAR1/IFNAR2), but ronta-Fab recognizes only IFN-αs. As described above, a Histidine side chain at residue 19 of IFN-ω would not be tolerated by ronta-Fab, but is part of the epitope for IFNAR2 as seen in the complex determined by X-ray crystallography¹⁵ (PDB entry 3SE4). Interestingly, the Tryptophan side chain at residue 19 of IFN-β (PDB entry 1AU1) would clash with IFNAR2 after docking according to the IFN-ω/IFNAR2 complex, but use of the different low energy side chain χ1 torsion angle (∼180°), as seen for His19 in the IFN-ω/IFNAR2 complex, is very likely to alleviate this potential steric problem.

If one inspects sequences of the human IFN-αs to rationalize the rontalizumab affinity rankings in Table II, it is sensible to restrict attention to (1) residues that contact ronta-Fab and to (2) only the most significant differences in affinity (Fig. 8). There is a single residue most nearly consistent with the lowest four affinity values, from IFN-αs 7, 4, 6 & 10, which differ from IFN-α2 by between 70 and 2000-fold. Residue 37, Gly in IFN-α2, is changed to Arg for IFN-αs 7, 6 & 10 (but not 4). This position is part of the AB-connecting segment (Fig. 4), is near CDR-H2, and has main chain torsion angles that are low energy only for Glycine. However, some of the reasonably well-recognized homologues also do not have Gly (Glu for IFN-α8 and 14) so there are clearly other factors involved, which may include combination effects from other changed side chains and other structural differences among IFN-αs. A relatively small affinity difference, about 10-fold, between the four highest affinity IFN-αs and the fifth one (IFN-α17) is well correlated with the Leu at position 153 changing to Phe. Most other lower affinity subtypes also bear Phe at position 153.

Figure 8.

Sequence alignment for human IFN-αs ranked according to increasing K_D values in Table I. For sequences other than that of IFN-α2, only differences from IFN-α2 are shown. The symbol above a position denotes a residue with significant loss of exposed surface area due to ronta-Fab. The symbol ⋆ denotes residues important for binding receptor. Residue position Gly37 is circled in red (see text).

Materials and Methods

Production of interferons for binding assay

IFN-α1, -α10 and -α7 were expressed in insect cells using baculovirus and purified using a C-terminal poly-Histidine (α1 and α10) or Fc affinity tag (α7). IFN-α21 and IFN-β were transiently expressed in 293 cells and purified using a C-terminal flag affinity tag. IFN-α16 was purchased from R&D Systems (Minneapolis). The other IFNs were expressed with poly-Histidine tags in bacteria and refolded. E. coli pastes (6–10 g pellets) were suspended in 10 volumes (w/v) of 20 mM Tris (pH 8) containing 7M guanidine HCl. Solid sodium sulfite and sodium tetrathionate were added to make final concentrations of 0.1 and 0.02M, respectively, and the solution stirred overnight at 4°C. The solution was clarified by centrifugation and loaded onto a Qiagen Ni-NTA metal chelate column equilibrated in 20 mM Tris (pH 8.6) containing 6M guanidine HCl. The column was washed with 38 mM imidazole, protein eluted with 250 mM imidazole, fractions with the desired protein (SDS-PAGE) were pooled and diluted to 50 μg/mL with buffer containing 20 mM Tris (pH 8.6), 2.5M urea, 0.3M NaCl, 20 mM glycine, 1 mM EDTA, and 5 mM cysteine. The refolding mixture was incubated overnight at 2–8°C and then adjusted to pH 3.0 with trifluoroacetic acid (TFA). The refolding mixture was loaded onto a RP-HPLC Vydac C₄ column (1.0 × 25 cm) equilibrated with 0.1%TFA in water and eluted with a linear gradient of acetonitrile (from 25-65%) in 0.1% TFA at 3 ml/min for 35 minutes. Fractions were pooled according to SDS-PAGE, and the acetonitrile content evaporated in a stream of N₂. The RP-HPLC pool was loaded onto a 25 mL HiTrap Desalting column (GE Healthcare) equilibrated with 10 mM HEPES (pH 6.8), and 0.15M NaCl. The eluate was sterile filtered through a 0.22μm filter unit (MILLEX-GV). Protein purity was assessed by SDS-PAGE (>95% purity) and protein concentration determined by Pierce BCA protein assay.

Biacore assay

Binding experiments were carried out using a capture methodology and analyzed with a Biacore 3000. Briefly, 5500-7000 RU of goat anti-human Fc (Jackson #109-006-098) were coupled to a CM5 chip at 20 µg/mL in 10 mM acetate, pH 5.0 using amine chemistry. Then 400-500 RU of antibody was captured at the surface, and then IFN was flowed over the capture surface. Binding was observed over a period of 4.5-5 min, and dissociation was observed for 20-30 min. The surface was regenerated with 10 mM glycine pH 1.5 following each cycle. All binding was carried out in 10 mM Hepes pH7.4, 0.15M NaCl, 3 mM EDTA, 0.005% Surfactant P20. Binding data were fit to the 1:1 Langmuir Binding Model, with Mass Transfer limitations. Dissociation constants (K_D) were determined as the quotient of association rate divided by dissociation rate (k_a/k_d). There was less than 1% standard error in the binding constants for all but three IFN-α subtypes (IFN α-8, −10 and −7 showed <2.5% error).

Production of IFN-α2 for crystallography

A codon-optimized pET15b vector carrying an N-terminal hexa-histidine affinity tag and human IFN-α2 residues 24-188 was expressed in BL21 E. coli. Three liters of fermentation broth were centrifuged to produce a 4.5g cell pellet which was lysed by suspension at 4°C in 300 mL CAPS pH 9.7, 400 mM NaCl, 5 mM imidazole with protease inhibitor cocktail (Roche) and passed three times through a microfluidizer. The clear supernatant from centrifugation of the lysis suspension was batch-loaded onto Ni-NTA affinity resin at 4°C for 30 min, the resin gravity loaded into two columns, washed with 50 mM imidazole (50 mL each) and then eluted with 250 mM imidazole (15 mL each). Size exclusion chromatography was performed using a Superdex-75 column (S75) and protein pooled according to SDS-PAGE, yielding 265 mg in 150 mL. To the pooled protein was added 10,000 units of human thrombin (Calbiochem) and 2 mM CaCl₂, and this mixture was dialyzed against 20 mM TRIS pH 8, 400 mM NaCl, 2 mM CaCl₂ at room temperature for 48 h. A second dialysis step was performed with 20 mM NaOAc pH 5, 400 mM NaCl at room temperature for 18 h, and the protein solution then concentrated to 25 mL for a polishing run over the S75 column before forming the complex with ronta-Fab.

Production of ronta-Fab

The Fab fragment of humanized 9F3 version 13 was produced in E. coli. Two hundred grams of cell pellet were lysed as described above except the lysis buffer was phosphate-buffered saline (PBS) with 25 mM EDTA and 1 mM PMSF. The affinity step was performed on a Protein-G affinity column and elution was performed using 0.58% aqueous acetic acid. Ion exchange chromatography followed on an SP Sepharose column charged with 1M NaCl and equilibrated with 20 mM MES pH 5.5 (buffer A). Before loading, the protein solution was diluted 5× with 0.58% aqueous acetic acid. The column was washed to baseline with buffer A and then eluted with 0–100% 20 mM MES pH 5.5, 500 mM NaCl over 20 column volumes. Size exclusion was performed on an S75 column equilibrated with 20 mM sodium acetate pH 5, 150 mM NaCl, and elution fractions pooled according to SDS-PAGE.

Formation and crystallization of the IFN-α2/ronta-Fab complex

Purified IFN-α2 and ronta-Fab were combined in a 3:1 molar ratio, incubated overnight at 4°C, purified by size exclusion on a Superdex 200 column equilibrated with 20 mM sodium acetate pH 5, 150 mM NaCl and concentrated to 20 mg/mL. Crystallization trials were performed using sitting drops and commercially available screening solutions. The crystals used for data collection were grown in hanging drops made from 2 µL protein and 2 µL reservoir containing 10 mM Tris pH 8.5, 200 mM lithium sulfate, 25% (v/v) PEG3350 and 1% dioxane. When mature, crystals were harvested and passed through solutions of reservoir augmented with 5%, 15%, and 20% glycerol before preserving for data collection by sudden immersion in liquid nitrogen.

Structure determination by X-ray crystallography

Diffraction data extending to 2.9Å were collected at ALS beamline 5.0.2 on a single-axis goniostat in a primitive monoclinic lattice with a long b axis (330 Å), after testing several crystals for acceptable relationships between the data collection rotation axis and the long unit cell dimension. Data reduction was performed using HKL2000³¹ and elements of the CCP4 suite³². Based on test calculations of the Matthews coefficient V_M, we estimated between six and eight complexes in the crystallographic asymmetric unit. The structure was solved by the method of molecular replacement,³³ using the X-ray structure of ovine IFN-τ³⁴ (PDB accession code = 1B5L) and an X-ray structure of a homologous Fab (PDB 1FVD, stripped of CDR-loops) as search probes in a previously collected 3.5Å isomorphous data set (99% complete, Rsymm = 0.080). A Fab probe with optimal elbow angle was identified using a series of rotation function calculations³⁵ with Fab models for which the elbow angle had been adjusted in a series of 5° steps. Using this Fab model, PHASER³³ successfully placed six Fabs, which were refined as 24 rigid bodies.³⁶ With this partial model used as known, PHASER placed four copies of IFN-τ. Two more copies of IFN-τ were placed manually by analogy to the others. Electron density map averaging and solvent flattening were applied to aid corrections and additions to the model. The application of the new 2.9Å data set to refinement and map calculation of the six-complex model revealed two more complexes which were placed using PHASER. The constant region of one ronta-Fab is not part of the final model, it being too poorly represented by electron density. Display of electron density and manipulation of models used XtalView³⁷ and Coot.³⁸ Refinements were performed using Refmac5³⁶ and Phenix.refine.³⁹ Due to overlaps, data between 3.0Å and 2.9Å were only 65% complete and were not used for the final refinements, which employed TLS treatment of thermal factors and NCS restraints on coordinates. Figure 9 shows electron density from the beginning and end of refinement. See Table III for data reduction and refinement statistics.

Figure 9.

Representative electron density (2mFo-dFc contoured at 1X rmsd) of IFN-α2/ronta-Fab. (A) After rigid-body refinement of a partial solution with six complexes and with R = 44%. (B) At the end of refinement. The depicted region is a part of that shown in Fig. 4. For clarity, the maps are contoured only around the depicted amino acid residues. Green atoms are from IFN-α2, light blue from the ronta-Fab light chain and darker blue from the ronta-Fab heavy chain.

Table III.

X-Ray Data Collection and Refinement IFN-α2/Ronta-Fab (PDB 4Z5R)

Data collection
X-ray source	ALS 5.0.2
Wavelength (Å)	1.0000
Space group	P21
Unit cell edges a, b, c (Å)	90.83, 331.86, 98.14
Unit cell angles α, β, γ (°)	90, 111.28, 90
VM (Å3/Dalton)	2.6
Resolution (Å)a	50–2.90 (3.12–3.00) (3.00–2.90)
Rsym	0.095 (0.286) (0.346)
Number of observations	395734
Unique reflections	109926
Completeness (%)	92.1 (78.9) (65.2)
I/σI	12.2 (3.3) (2.3)
Wilson B (Å2)	59
Refinement
Resolution (Å)	50–3.00
Reflctns [F > 0σ(F)]	99974
RWORK, RFREE	0.223, 0.251
Complexes/asymmetric unit	8
Protein residues	4354
Water molecules	0
Sulfate ions	8
Atoms	34056
Mean B-factor (Å2)	75
Rmsd bonds (Å)	0.010
Rmsd angles (˚)	1.4
Number of TLS groups	38
Ramachandran (%)	94/4/2

^aValues in parentheses refer to the resolution shells defined here.

ACCESSION NUMBER: The final coordinates and structure factors have been deposited at the Protein Data Bank⁴⁰ under accession code 4Z5R.

Conclusion

Lupus patients lack safe and effective treatments. Prominent among clinical investigations of potential medicines are candidates designed to eliminate or reduce the effects of INF-α signaling.³⁰ We have shown how a Fab fragment from the experimental antibody medicine rontalizumab blocks interaction of IFN-αs with an obligatory component of their heterodimeric receptor (IFNAR2) and rationalized the specificity of rontalizumab for IFN-αs over other Type I interferons. Based on the concept that signaling by IFN-αs requires both IFNAR2 and IFNAR1, rontalizumab can be expected to prevent IFN-α-induced signals.

Glossary

CDR

complementarity determining region

HCV

hepatitis-virus type C

IFN-α

Interferons-alpha

NCS

Non-crystallographic symmetry

SLE

systemic lupus erythematosis

TFA

trifluoroacetic acid