Biochemical Journal

Research article

Production of a human neutralizing monoclonal antibody and its crystal structure in complex with ectodomain 3 of the interleukin-13 receptor α1

Nicholas T. Redpath, Yibin Xu, Nicholas J. Wilson, Louis J. Fabri, Manuel Baca, Arna E. Andrews, Hal Braley, Ping Lu, Cheryl Ireland, Robin E. Ernst, Andrea Woods, Gail Forrest, Zhiqiang An, Dennis M. Zaller, William R. Strohl, Cindy S. Luo, Peter E. Czabotar, Thomas P. J. Garrett, Douglas J. Hilton, Andrew D. Nash, Jian-Guo Zhang, Nicos A. Nicola


Gene deletion studies in mice have revealed critical roles for IL (interleukin)-4 and -13 in asthma development, with the latter controlling lung airways resistance and mucus secretion. We have now developed human neutralizing monoclonal antibodies against human IL-13Rα1 (IL-13 receptor α1) subunit that prevent activation of the receptor complex by both IL-4 and IL-13. We describe the crystal structures of the Fab fragment of antibody 10G5H6 alone and in complex with D3 (ectodomain 3) of IL-13Rα1. Although the structure showed significant domain swapping within a D3 dimer, we showed that Arg230, Phe233, Tyr250, Gln252 and Leu293 in each D3 monomer and Ser32, Asn102 and Trp103 in 10G5H6 Fab are the key interacting residues at the interface of the 10G5H6 Fab–D3 complex. One of the most striking contacts is the insertion of the ligand-contacting residue Leu293 of D3 into a deep pocket on the surface of 10G5H6 Fab, and this appears to be a central determinant of the high binding affinity and neutralizing activity of the antibody.

  • asthma
  • interleukin-13
  • interleukin-4
  • monoclonal antibody
  • receptor
  • TH2 immune response


Asthma is a heterogeneous and chronic inflammatory disease of the airways that is characterized by airway inflammation, reversible airway obstruction, mucus hypersecretion, airway hyper-responsiveness and airway wall remodelling [1]. In a significant proportion of patients, the disease manifests as an allergic hypersensitivity dominated by an immune response polarized to the TH2 phenotype [2]. This activation results in the secretion of TH2 cytokines [IL (interleukin)-4, IL-13 and IL-5] that either directly or indirectly trigger many of the pathologies associated with this disease including IgE production, eosinophilia, mucus secretion by lung epithelial cells and airway smooth muscle hyper-responsiveness. IL-4 is thought to be important in skewing the T-cell response to the TH2 type and in promoting antibody class switching to the IgE type. Whereas IL-13 has similar effects to IL-4, it also has direct roles in stimulating mucus secretion by lung epithelial cells. IL-5 is the major cytokine affecting eosinophil production and activation [3].

The similarity of the biological effects of IL-4 and IL-13 is partly explained by the use of common cell-surface receptor complexes: they each can bind to and activate a receptor complex made up of IL-13Rα1 (IL-13 receptor α1) and IL-4Rα (IL-4 receptor α). On the other hand, only IL-4 can bind to an alternative signalling receptor complex made up of IL-4Rα and the common γ chain receptor (γc) [4,5].

Although IL-5-gene-deleted mice have reduced eosinophil numbers, they nonetheless display airway hyper-reactivity, independent of IL-5, in an ovalbumin-induced model of asthma [6]. In human trials, anti-IL-5 antibodies did not completely eliminate eosinophils from the airways and did not inhibit antigen-induced late airway responses [7,8]. Similarly, despite early promise, clinical trials targeting IL-4 inhibition have been disappointing [3].

Although clinical trials with anti-IL-13 reagents are still at an early stage [9], several lines of evidence suggest that IL-13 and IL-13Rα1 are critically involved in asthmatic responses. Genetic studies in mice have shown that IL-13 plays critical roles in the regulation of IgE synthesis, mucus hypersecretion, sub-epithelial fibrosis and eosinophil infiltration into the lung, and it has been associated with the regulation of chemokine receptors including CCR5 (CC chemokine receptor 5) [10]. In addition, genetic polymorphisms in IL-13, its receptor subunits or its signalling molecules have been associated with the severity of human asthmatic disease [1113].

We originally cloned the mouse IL-13Rα1 subunit and showed that the high-affinity IL-13 receptor consisted of this subunit as well as the IL-4Rα subunit. We also showed that this receptor complex was an alternative receptor for IL-4 [4]. The human equivalent was subsequently cloned and shown to have identical functional properties [14,15]. Structural studies have revealed the molecular details of IL-13 and IL-4 recognition by the identical receptor complex of IL-13Rα1 and IL-4Rα [5]. Although these structures look very similar, the molecular assembly order is different and can lead to differential biological responses in the same cell in response to the two cytokines. Clearly there would be potential benefit in a therapeutic approach that targets both IL-13 and IL-4 that are strongly associated with asthma.

In the present paper, we describe the production of fully human neutralizing monoclonal antibodies against h (human) IL-13Rα1, show that they inhibit the binding and biological activities of IL-13 and IL-4 on human cells and define the binding epitope and structural details of the binding interaction.


Animal experimentation

All animal experiments were approved by the animal ethics committee of the Walter and Eliza Hall Institute in accordance with the NHMRC (National Health and Medical Research Council) Australian code of practice for the care and use of animals for scientific purposes.

Immunization of Medarex™ mice with hIL-13Rα1 and selection of hybridomas

Mice transgenic for human immunoglobulin genes (Medarex) [16] were immunized intraperitoneally with 50 μg of the hIL-13Rα1 ECD (extracellular domain) [17] in complete Freund's adjuvant and then boosted repeatedly with hIL-13Rα1 ECD formulated in either incomplete Freund's adjuvant or PBS. Before fusion, mice were injected intravenously with 10 μg of hIL-13Rα1 ECD in PBS and, 4 days later, spleens were harvested and fusions performed as previously described [18,19].

Hybridoma supernatants were screened for hIL-13Rα1 binding activity using a standard ELISA format with plate-bound ECD, and for IL-13 antagonist activity in the TARC (thymus- and activation-related chemokine) release assay described below. Selected hybridomas were cloned by limit dilution.

Affinity maturation of 10G5 by phage display

Affinity maturation of monoclonal antibody 10G5 was achieved by randomly mutating amino acids within the variable H (heavy-chain) and L (light-chain) CDR (complementarity-determining region) 3 followed by expression and selection using phage display (Supplementary Table S1 at Libraries were constructed as described previously [20] using ‘stop template’ versions of pFab10G5 for the Kunkel mutagenesis method [21] with mutagenic oligonucleotides designed to simultaneously repair the stop codons and introduce mutations at the designed sites. The mutagenesis reactions were then electroporated into Escherichia coli SS320 then phage production was initiated with addition of M13-KO7 helper phage (Invitrogen) before incubation at 30°C for 18 h. Sequencing analysis of approximately 100 clones from each of the small-scale libraries showed 100% diversity of mutant clones at the designed amino acid locations.

10G5 HCDR3 and LCDR3 libraries were panned against IL-13Rα1. Briefly, four rounds of selection were performed by incubating phage with decreasing concentrations of immobilized biotinylated hIL-13Rα1 (10, 1, 0.1 and 0.01 μg) in 2% (w/v) dried skimmed milk in PBST (PBS containing 0.1% Tween 20) for 20 min at room temperature (22°C). Downstream panning steps were carried out on the basis of standard protocols [22]. Individual clones were picked after the second and third round of selection and the Fab cassette and the light chain were PCR-amplified and sequenced essentially as described in [23]. Phage ELISA was used as the primary assay to determine the ability of the phage-bound recombinant Fab fragments to recognize biotinylated hIL-13Rα1 immobilized on streptavidin plates.

Site-directed mutagenesis and phage display of hIL-13Rα1 and hIL-13Rα1 fragments

The ECD of hIL-13Rα1 (fibronectin-like domains D1, D2 and D3) (GenBank® accession number NP_001551) [14] as well as sequences encoding D2–D3 or individual domains (D1, amino acids 26–105; D2, amino acids 106–202; D3, amino acids 203–319) were directionally cloned into a phagemid vector (phg3-1) and used subsequently as the templates for all expression constructs.

IL-13Rα1 constructs were transfected into E. coli and phage production was initiated by addition of M13-KO7 helper phage (Invitrogen) before incubation at 30°C for 18 h. Phage was harvested by precipitation with 4% PEG [poly(ethylene glycol)] 600/0.5 M NaCl and resuspended in 100 μl of PBS. IL-13Rα1-expressing phage supernatants were tested for antibody binding by ELISA plates coated with 10G5 (or control antibodies) with detection by 100 μl HRP (horseradish peroxidase)-conjugated anti-M13 antibody diluted 1:2500 in 1% (w/v) dried skimmed milk in PBS (GE Healthcare). After washing the plates three times with PBST, the signal was developed for 2–3 min using 100 μl/well of TMB/E (3,3′,5,5′-tetramethybenzidine) substrate (Chemicon International), the reaction was stopped upon addition of 2 M phosphoric acid (50 μl/well), and absorbance was measured at 450 nm.

Determination of antibody-binding affinity by surface plasmon resonance

Antibody-binding affinity was performed using a Biacore™ 4000 biosensor (GE Healthcare). In one set of experiments, goat anti-(human IgG) (Invitrogen H10500) was immobilized on a CM5 sensorchip following the manufacturer's instructions to approximately 7000 RU (relative units). Antibodies 10G5 or 10G5-6 were captured on spots 1 and 5 of each flow cell for 2 min at 0.3μg/ml to approximately 200 RU. Soluble hIL-13Rα1 was injected over each flow cell for 5 min and dissociation was monitored for a further 30 min. The analysis was performed with receptor prepared in a 2-fold dilution series between 40 and 1.25 nM, with each concentration analysed twice. The analysis was performed at a flow rate of 30 μl/min in 10 mM Hepes, 150 mM NaCl, 3 mM EDTA and 0.005% Tween 20 (pH 7.4) at 37°C. Responses from spots 2 and 4 of each flow cell (in which 10G5 or 10G5-6 were not captured, but otherwise treated identically), were subtracted from those of spots 1 and 5 respectively. Responses from blank injections were then subtracted from those of all other samples to produce double-referenced data suitable for kinetic analysis. Double-referenced sensorgrams were fitted using non-linear regression to a model describing 1:1 kinetics, including a term for mass transport limitation. The Rmax value was fitted locally with association rate (ka), dissociation rate (kd) and equilibrium dissociation constant (Kd) fitted globally. In another set of experiments, to analyse the effect of mutation of Leu293 in hIL-13Rα1, 10G5H6 Fab was coupled directly to a CM5 sensorchip to a level of approximately 1000 RU and background subtraction was achieved using a flow cell on which an unrelated Fab fragment had been immobilized.

Determination of antibody-binding affinity by KinExA

All KinExA experiments were carried out at room temperature. The antigen hIL-13Rα1 was serially 2-fold diluted into a 1× assay buffer (1 mg/ml BSA in PBS, pH 7.4) containing a fixed amount of IgG or Fab. Samples were allowed to reach equilibrium by pre-incubation for 12–24 h and free antibody was measured using 1 μg/ml Cy5 (indodicarbocyanine)-labelled mouse anti-(human Fab) or goat anti-(human Fcγ) for IgG. Samples were run in duplicate for each experiment. At least two curves were run using different antibody concentrations, one close to the predicted Kd and one at least 3-fold higher. The equilibrium titration data were analysed using a 1:1 binding model provided by KinExA software. N-curve analysis for final Kd measurement was performed using KinExA software.

Protein production

hIL-13Rα1 D3–His6 was expressed in plasmid pET26b, which contains a pelB periplasmic localization signal sequence in frame with the expressed protein, in E. coli BL21(DE3)pLysS cells. Cells were grown to a D600 of 0.8–1, induced with 0.1 mM IPTG (isopropyl β-d-thiogalactopyranoside) then grown overnight at 19°C. Since the majority of D3–His6 was secreted into the medium, the concentrated cell supernatant was purified on an Ni-NTA (Ni2+-nitrilotriacetate) column (Qiagen) followed by Q-Sepharose ion-exchange and Superdex 75 gel-filtration chromatography. In order to remove the C-terminal His6 tag, the protein was treated with 30 units of carboxypeptidase A/mg of D3 (Sigma) for 6 h at 37°C before separation on a Superdex 75 column.

hIL-13Rα1 D1–D3 was expressed as a secreted N-terminal FLAG–His6-tagged protein in Sf21 insect cells using established protocols [24] and purified from cell medium as above.

For expression of anti-hIL-13Rα1 Fab fragments, the Fab light and heavy chain fragments were co-expressed in the same plasmid (pET26b or pBR322) with a pelB periplasmic localization signal sequence in frame with the light chain and an OmpA signal sequence in-frame with the heavy chain fragment. There was also a His6 tag at the C-terminus of the heavy chain fragment. The Fab was purified on Ni-NTA, followed by an hIL-13Rα1 D3 affinity column.

Crystallization and structural determination

The final crystals of 10G5H6 Fab were obtained through a few rounds of microseeding and crystallized in 1.2 M (NH4)2SO4, 0.5% PEG8000 and 0.1 M Hepes (pH 7.2). For the 10G5H6 Fab–hIL-13Rα1 D3 complex, the protein was concentrated to an A280 of 17 and the crystals grown from 1.6 to 2.0 M Li2SO4, 0.1 M Ches [2-(N-cyclohexylamino)ethanesulfonic acid] buffer (pH 9.5) by the hanging-drop vapour-diffusion method.

Data from 10G5H6 Fab crystals were collected on a home source at 100 K and processed with HKL2000 [25]. The structure was solved by molecular replacement using PHASER [26], using one Fab of PDB code 1T04 as a search model. The model was checked for stereochemical correctness using RAMPAGE of the CCP4 suite [27]. The Ramachandran plot indicated that 97.1% of residues fall in the favoured regions and 2.9% fall in the allowed regions.

Data from 10G5H6 Fab–hIL-13Rα1 D3 crystals were collected at 100K on the MX2 beamline at the Australian Synchrotron and processed with the XDS package [28]. Structures were solved by molecular replacement using uncomplexed 10G5H6 Fab as a search model with PHASER [26], whereas D3 was gradually built into the model starting from two strands. Structure refinement was carried out with PHENIX [29] and AutoBuster (BUSTER version 2.10.0; Global Phasing) and manual rebuilding was undertaken using COOT [30]. The Ramachandran plot indicated that 94.7% of residues fall in the favoured regions, 4.9% fall in the allowed regions and 0.4% (Ala241 and Ala304) fall in outlier regions.

Cell-based bioassays

STAT6 luciferase reporter assay

A HEK (human embryonic kidney)-293 cell line stably transfected with a luciferase reporter gene under the control of a STAT6 (signal transducer and activator of transcription 6) promoter was stably transfected with hIL-13Rα1 and hIL-4Rα (293-STAT6-Luc-IL-13R). These cells were plated at 5×104 cells/well in 96-well flat-bottomed plates in RPMI 1640 medium, 10% FBS (fetal bovine serum) and GlutaMAX™ (Invitrogen). Recombinant hIL-13 or hIL-4 (R&D Systems) was added and cells were incubated at 37°C for 18 h under 5% CO2. Luciferase activity was determined by Britelite Plus Kit (PerkinElmer) and these cells responded to both hIL-13 and hIL-4 equivalently with an EC50 of ~1 ng/ml.

Eotaxin [CCL (CC chemokine ligand) 11] release

NHDF (neonatal human dermal fibroblast) cells (Lonza CC-2509) were maintained in FGM™-2 medium (Lonza CC-3132) at 37°C under 5% CO2. Cells were plated in flat-bottomed 96-well plates at 105 cells per well and PMA (20 ng/ml) was added for 1 h. Recombinant hIL-13 or hIL-4 was added and cells incubated at 37°C for 18 h under 5% CO2 before assaying eotaxin in the supernatant by ELISA (R&D Systems).

TARC (CCL17) release

PBMCs (peripheral blood mononuclear cells) were isolated from normal donors with informed consent (Australian Red Cross) and plated at 5×105 cells/well in 96-well flat-bottomed plates in RPMI 1640 medium, 10% FBS and GlutaMAX™. Recombinant hIL-13 or hIL-4 was added and cells were incubated at 37°C for 48 h under 5% CO2 before assaying TARC in the supernatant by ELISA (R&D Systems).

CD23 up-regulation

PBMCs were isolated and plated as above except that cells were incubated for 18 h. Cells were then incubated with FITC-conjugated anti-CD14 and PE (phycoerythrin)-conjugated anti-CD23 antibodies (BD Biosciences) at 4°C for 30 min.


Generation of human neutralizing antibodies against hIL-13Rα

By immunizing Medarex™ HuMab mice with soluble hIL-13Rα1, we generated a panel of hybridoma clones that express fully human antibodies against recombinant hIL-13Rα1. Characterization of the antibodies identified a candidate, designated 10G5, with strong to moderate neutralizing activity of both IL-13 and IL-4 signalling in normal human fibroblasts expressing IL-13Rα1 and IL-4Rα (Figure 1) and with high affinity for hIL-13Rα1 as determined by surface plasmon resonance (Kd 3.65±0.04 nM, n=4).

Figure 1 10G5 antibody inhibits IL-13 binding and activity

(a) Inhibition of IL-13 binding to hIL-13Rα1. Purified hIL-13Rα1 (8 μg/ml) was incubated without (A) or with (B) 10G5 monoclonal antibody (50 μg/ml) before passing over a hIL-13 biosensor chip. (b) Inhibition of IL-13- and IL-4-induced eotaxin (CCL11) production from NHDFs. NHDFs were cultured with PMA and incubated with increasing concentrations of 10G5 before stimulation with IL-13 (left) at various concentrations (∆, 4 ng/ml; □, 20 ng/ml; ○, 200 ng/ml) or IL-4 (right) at various concentrations (∆, 0.2 ng/ml; □, 1 ng/ml; ○, 10 ng/ml) for 18 h. Supernatants were collected and assayed for eotaxin production. These experiments were repeated four times with representative graphs shown.

The 10G5 antibody was subjected to affinity maturation (as outlined in Supplementary Table S1). Initially three phage display libraries covering the variable HCDR3 (FPNWGSFDY) and another three libraries covering the variable LCDR3 (QQYET) were generated. These libraries all contained >108 functional diversity and, between them, covered all combinations of amino acids at every position randomized in each set. In addition to the randomization of amino acid residues in HCDR3 and LCDR3, a deletion of one amino acid residue for HCDR3 and addition of one or two amino acid residues for LCDR3 were also made. In the first round of optimization, the most significant leads were a F100M mutation in HCDR3 (MPNWGSFDY; designated 10G5H6), and a separate, less potent mutant possessing a pair of changes in the C-terminal residues of HCDR3 (FPNWGSLDH). Notably, only the first, seventh and ninth positions of HCDR3 appeared to be conducive to randomization without loss of activity. Shortening of the HCDR3 by one amino acid residue abolished the binding activity of 10G5 to IL-13Rα1.

No significant potency improvement was obtained from the original LCDR3 libraries (keeping the HCDR3 as wild-type), with the exception that it became obvious that insertion of additional residues in the very short LCDR3 of 10G5 abolished binding activity. A second-round LCDR3 library was generated using the 10G5H6 (MPNWGSFDY) mutant as the VH (variable heavy) chain to search for synergy with the modified LCDR3. A set of LCDR3 mutants was obtained containing small neutral residues in the last two positions of LCDR3, including QQYAT, QQYAS, QQYGS and QQYST (Supplementary Table S1), the best of which was QQYAS, as determined in an in vitro dissociation-based binding assay (results not shown). A second-round library of HCDR3 was constructed in which three amino acid positions were randomized (XPNWGSXDX; randomized positions indicated by X), as dictated by the first-round results. The best results from this library are shown in Supplementary Table S1. A series of specific constructs was then made to test various combinations; the best three of which all contained the same sequence in HCDR3 (MPNWGSLDH) and LCDR3 (QQYAS) and possessed Kd values for hIL-13Rα1 in solution of 20–30 pM. Two of these also contained inadvertent framework mutations. 10G5-6 was the lead VH/VL (variable light) combination chosen for development due to the least number of framework changes. The amino acid sequences of HCDR3 and LCDR3 for parental 10G5, 10G5H6 and 10G5-6 along with their corresponding IL-13Rα1-binding affinities and cell-based potencies are shown in Table 1. Notably, a single amino acid change from 10G5 to 10G5H6 (F100M in HCDR3) resulted in increased binding affinity of nearly 10-fold and an increase in potency of ~3-fold. The additional changes from 10G5H6 to 10G5-6 were two amino acid changes in the light chain (E93A and T94S) and two additional changes in the heavy chain (F106L and Y108H) that further increased binding affinity 4-fold and potency ~1.5-fold.

View this table:
Table 1 Binding affinity and inhibitory capacity of 10G5 and affinity-matured neutralizing antibodies

Equilibrium dissociation constants (Kd) (mean±S.D. from three replicates) for binding hIL-13Rα1 were determined using kinetic exclusion on a KinExA™ instrument and 50% inhibitory doses (IC50) were determined by titration in the NHDF assay stimulated with IL-13 at 20 ng/ml.

Inhibitory activities of antibodies against hIL-13Rα1

Both 10G5 and 10G5-6 were shown to completely block binding of IL-13 to recombinant hIL-13Rα1 in a surface plasmon resonance assay and to efficiently inhibit IL-13- and IL-4-stimulated production and release of the T-lymphocyte chemokine TARC (CCL17) from A549 human lung epithelial cells as well as expression of a STAT6–luciferase reporter in HEK-293 cells engineered to express hIL-4Rα and IL-13Rα1 (Figure 2). 10G5-6 also completely inhibited IL-13-induced up-regulation of CD23 and CCL17 release from human PBMCs (Figure 2). Both antibodies more efficiently inhibited IL-13 action compared with IL-4 action (~30-fold difference in EC50) (Figure 2).

Figure 2 Inhibition of IL-13- and IL-4-induced chemokine production and activation of human cells and cell lines

HEK-293 IL-4R/IL-13R luciferase cells (a) and A549 human lung epithelial cells (c) were incubated with increasing concentrations of 10G5 (□) or 10G5-6 (∆) then stimulated with IL-13 for 18 h. HEK-293 IL-4R/IL-13R luciferase cells (b) and A549 human lung epithelial cells (d) were incubated with increasing concentrations of 10G5-6 then stimulated with IL-4 (■) or IL-13 (○) for 18 h. Cells were assayed for luciferase activity (a and b) or supernatants were collected and assayed for CCL17 production (c and d). For (e) and (f), PBMCs were prepared from normal human blood and incubated with increasing concentrations of 10G5-6 then stimulated with IL-13 for 18 h. Supernatants were collected and assayed for CCL17 production by ELISA (e) or cells were stained with anti-CD14 and anti-CD23 antibodies and the percentage of CD14+, CD23+ cells were determined by flow cytometry (f). All experiments were repeated at least four times with representative graphs shown. Results in (a)–(e) are means±S.E.M. from duplicate wells. mAb, monoclonal antibody.

The IL-13Rα1 ECD consists of three fibronectin type III-like modules, so we prepared constructs encoding the full-length ECD and truncated fragments that contained one or two of these modules. The protein constructs were displayed on M13 phage as gene III fusions, which allowed for binding assays by phage ELISA. Only phage display constructs containing the third (membrane-proximal) fibronectin type III module (D3) showed significant binding to 10G5 (Figure 3a), showing that the 10G5-binding epitope was localized to this region of IL-13Rα1. Since 10G5 bound to human, but not mouse, IL-13Rα1, we generated a series of human–mouse chimaeric molecules (again by phage display) to more precisely define the binding epitope in the D3 region of hIL-13Rα1. The series of chimaeras (HM1–HM11 shown in Supplementary Figure S1 at allowed us to identify two linear segments of the IL-13Rα1 sequence that contained residues critical for antibody binding (Figure 3b). Finer mapping of the binding epitope was then carried out by individual alanine mutations of each human/mouse difference in these regions. This series of IL-13Rα1 ECD mutants, displayed on phage, identified Phe233, Phe249, Tyr250 and Gln252 as having the strongest effects on antibody binding (Figure 3c).

Figure 3 Epitope mapping of 10G5 by phage ELISA

Phage-displayed hIL-13R constructs were tested for binding to various monoclonal antibodies. 10G5 or anti-IL-13R control antibodies [8B4, a D2-specific monoclonal (control 1), and 2B6, a D1/2-specific monoclonal (control 2)] were coated on to plates and phage-displayed ECDs of the hIL-13R (a), phage-displayed human–mouse (HM) mutant receptors (b) or phage-displayed hIL-13R-containing point mutations (c) were incubated with the coated antibodies. Bound receptor was detected with a HRP-conjugated anti-phage antibody. (a) Results are A450 values. (b and c) Results for 10G5 binding to mutant receptors is expressed as log binding index as determined from the following equation: log[100×(EC50 10G5 receptor mutant/EC50 control 1 receptor mutant)/(ΣEC50 10G5 receptor mutants/ΣEC50 control 1 receptor mutants)]. The experiments were repeated twice. Representative histograms are shown.

Structures of 10G5H6 and 10G5H6 Fab–D3 complex

To refine further the epitope of 10G5 and to understand the structural basis for the improved affinity attributed to the M100F substitution in HCDR3, we crystallized the Fab fragment of 10G5H6 antibody and the complex of the 10G5H6 Fab antibody with the D3 subdomain of hIL-13Rα1. We solved the structure of 10G5H6 Fab at 2.43 Å (1 Å=0.1 nm) resolution and the structure of the complex at 2.61 Å.

The crystal structure of 10G5H6 Fab was refined to 2.43 Å with R/Rfree factors of 17.58%/22.61% (Table 2). The final model contained one 10G5 Fab molecule in the asymmetric unit with six disordered residues (133–138) from the heavy chain. The structure of 10G5H6 showed that HCDRs and LCDRs form a deep pocket able to bind to D3 of IL-13Rα1.

View this table:
Table 2 Data collection and refinement statistics of 10G5H6 and 10G5H6 Fab–D3

Values in parentheses represent highest resolution shell.

For the 10G5H6 Fab–D3 complex, the initial model containing only 10G5H6 Fab was refined using AutoBuster. A good difference FoFc electron density map was built manually with two polyalanine strands. Through iterative rounds of refinement and model building, the whole D3 domain was placed into the final model. The structure of the 10G5H6 Fab–D3 complex was determined at 2.61 Å resolution with R/Rfree factors of 19.67%/22.42% (Table 2). The asymmetric unit contained one copy of the D3 domain bound to one Fab fragment of 10G5H6 (Figure 4a) (solvent content approximately 75%). The structure has a number of interesting features, especially in the D3 domain (Figures 4a and 4b). The most striking are the swapping of the N-terminal strands A and B and C-terminal strand G between two D3 monomers (Figure 4b and see below). There are also five Ches buffer molecules found in the complex; two of the Ches molecules occupied the hydrophobic pocket in D3 that was created by the swapping of B–C and F–G loops.

Figure 4 Structure of the 10G5H6 Fab–D3 complex

(a) Ribbon diagram of 10G5H6 Fab–D3 complex (D3, green; light chain, yellow; heavy chain, cyan). (b) Structure of the D3 domain-swapped dimer in the complex. (c) Ribbon diagram of domain-swapped dimeric D3 with two 10G5 Fabs. (d) Comparison of the CDRs of unbound 10G5H6 (light chain, marine; heavy chain, pink) and bound 10G5H6 (light chain, yellow; heavy chain, cyan) with large conformational change residues indicated.

The crystal structures of 10G5H6 Fab in both complexed and uncomplexed states aligned well with an RMSD (root mean square deviation) of 0.68 Å for 423 Cαs. Upon D3 binding, only minor conformational changes (<1 Å) were seen in most of the CDRs, with the exception of HCDR3. The most important residues (Asn102 and Trp103) from HCDR3 are involved in six hydrogen bonds and extensive van der Waals contacts in the interface (see below) and undergo an approximate 2 Å shift upon D3 binding. The side chain of Arg60 from the heavy chain undergoes a shift to stabilize the 310 helix at the C-terminus of D3. Another important residue, Tyr92, in LCDR3 undergoes a dramatic side-chain shift to contact Leu293 of D3 (Figure 4d).

Interface of the 10G5H6 Fab–D3 complex

The antibody–antigen interface buries a total surface area of 2199 Å2 (1137 Å2 of D3 accessible surface area and 1062 Å2 of 10G5H6). Analysis of the complex structure indicated that 22 residues of D3 participate in the antibody–antigen interaction. These residues belong to four discontinuous fragments: Pro202 and Asp203 at the N-terminus of D3 form two hydrogen bonds with Asp56 and Ser57, both from the heavy chain (Figures 5a and 5b); residues 226–235 from the B–C loop and strand C; residues 248–253 from strand C′; and residues 289–296 from the C-terminus of strand F followed by the F–G loop which contains a 310 helix (residues 292–294) (Figure 5a). The total interaction between 10G5H6 Fab and D3 consists of 16 hydrogen bonds (<3.5 Å), three salt bridges and 197 van der Waals contacts (<4 Å). All 16 hydrogen bonds involve HCDRs 1–3 and LCDRs 1–3 of 10G5H6. As for D3, strands C, C′ and the F-strand-loop participate in forming four, five and five hydrogen bonds respectively.

Figure 5 The interface of 10G5H6 Fab–D3

(a) The four fragments containing the epitope of D3 (green) lie on the 10G5 surface. Important residues on D3 are shown as sticks except for Leu293 which is shown as balls. The key residues on 10G5H6 are also labelled. (bd) The hydrogen-bonding interaction network between 10G5H6 and D3. Oxygen, nitrogen and sulfur atoms are coloured red, blue and yellow/orange respectively. In each case, D3 elements are in green, 10G5H6 heavy chain is in cyan and 10G5H6 light chain is in yellow.

Arg230 from the B–C loop forms three salt bridges with Asp56 HCDR2 and a hydrogen bond with Trp34 of HCDR1 (Figure 5b). Phe233 from strand C of D3 is a key residue and its phenyl group makes extensive van der Waals contact with the side chain of Trp103 (HCDR3) of 10G5H6 Fab heavy chain (Figure 5b). Another important residue from strand C is Leu232, which forms a hydrogen bond with Asn102 (H3) of 10G5H6 Fab (Figure 5b). The phenyl group of Tyr250 of strand C′ of D3 stacks with the side chain of Trp103 (H3) as well. Gln252 of strand C′ forms four hydrogen bonds with Ser32 (H1) and Asn102 (H3), both from the heavy chain of 10G5H6 Fab (Figure 5c). In addition, the main-chain oxygen atom of Val248 of D3 hydrogen-bonds with the OG atoms of Ser54 (L2). One of the most striking contacts is the insertion of Leu293 of D3 into the deep pocket on the surface of 10G5H6 Fab (Figure 5d). Leu293, in the middle of a 310 helix, was positioned approximately 4 Å from five residues that line the pocket of 10G5H6 (Trp34, Trp48, Val51 and Met100 in the heavy chain fragment and Tyr92 in the light chain) (Figure 5d and Supplementary Figure S2 at The backbone oxygen atom of Leu293 of D3 hydrogen-bonds with NE of Arg60 in the heavy chain fragment, which also forms another hydrogen bond with the main-chain oxygen atom of Tyr295. Asn291, another important residue in D3, forms two hydrogen bonds with Tyr92 (LCDR3) and Asn102 (HCDR3). Additionally, Thr290 of D3 makes one hydrogen bond with Tyr33 (LCDR1) (Figure 5d).

Comparison of D3 with that in intact receptor

The D3 domain of the 10G5H6 Fab–D3 complex displays some unusual features. The global alignment between D3 within the 10G5H6 Fab–D3 complex and D3 within the IL-13 receptor complex (PDB code 3BPO) was broken because of the extensive N- and C-terminal strand swapping in the 10G5H6 Fab–D3 complex. The B–C loop is stretched out and residues 226–229 form a helix. This extended B–C loop separates strands A and B from strands C, C′, E and F so that the strands A and B belong to one monomer and strands C, C′, E and F belong to another monomer (Figure 4b). In addition, the extended F–G loop allows swapping of the G strand. The extensive domain swapping favours a more stable dimeric D3.

Although the overall alignment between D3 of the 10G5H6 Fab–D3 complex and D3 of the IL-13 receptor complex was broken, strands C, C′ and F align well with an RMSD of 0.87Å for 23 Cαs.

The key epitope residues Arg230, Phe233, Tyr250, Gln252 and Leu293 are located close to or on these three strands (Figure 6a). Phe233 fitted well (RMSD of 1.6Å for Cα) between the two D3 complexes and the Cα of Tyr250 and Gln252 shifted 2.3 and 3.5 Å respectively. The Cα of Arg230 shifted 4.5 Å, but the biggest shift (5.6 Å for Cα) seen was for Leu293 (Figure 6b).

Figure 6 Comparison of the D3 domain structures seen in IL-13/IL-13Rα1 (PDB code 3BPO) or 10G5H6 Fab–D3 complexes

(a) Overall superimposition of D3 dimer from our complex (green and blue) with D3 from the holoreceptor complex 3BPO (magenta). (b) The relative orientations of Leu293 and other residues from our structure (green) and 3BPO (magenta) are superimposed.

The B–C loop in the D3 domain of the full-length receptor is stabilized by interactions with both IL-13 (Arg230 of the D3 domain forms hydrogen bonds with Phe107 and Arg108 of IL-13 and the disulfide bond Cys231–Cys294 of D3 is in van der Waals contact with Phe107 of IL-13) and the D2 domain (Asn226 of D3 hydrogen-bonds with the main-chain oxygen atom of Asn112 in D2, whereas Phe227 in D3 interacts with Leu113 in D2) [5]. It is possible that in the absence of these stabilizing interactions, the B–C loop may be quite mobile.

Since Leu293 in D3 is an important contact residue for binding IL-13 and IL-4 [5] and appears to make a significant shift in our structure relative to that in the full-length ECD [5], we tested the binding of 10G5H6 Fab to wild-type compared with L293A mutant full-length hIL-13Rα1 ECD by surface plasmon resonance. The mutant receptor bound 10G5H6 with nearly 100-fold lower affinity than did wild-type (Kd 52 nM compared with 0.57 nM) (Figure 7), most of the change in affinity being due to a greater than 30-fold increase in the kd of the L293A mutant compared with wild-type. Similar results were seen with L293A D3 domain compared with wild-type D3 domain (Supplementary Figure S3 at

Figure 7 Effect of Leu293 mutation in D3 on binding affinity to 10G5H6

Binding of full-length ECDs of wild-type or L293A hIL13Rα1 to immobilized 10G5H6 Fab measured using surface plasmon resonance. The concentrations (nM) of receptor used are indicated. Representative sensorgrams are shown and the mean±S.D. Kd values are shown (n=3).


In the present paper, we have described the generation and neutralizing activities of humanized antibodies that recognize the D3 domain of hIL-13Rα1. The 10G5 antibody and the affinity-matured derivatives (10G5H6 and 10G5-6) inhibited the biological actions of both IL-13 and IL-4, including chemokine production by human cell lines, activation and chemokine production by human PBMCs and eotaxin release from normal human dermal fibroblasts at doses below 10 μg/ml. The differential inhibitory doses of antibody required to inhibit IL-13 compared with IL-4 could be a result of the different order of assembly of the ternary complexes with antibody inhibiting the first step of IL-13 receptor assembly, but only the second step of IL-4 receptor assembly [5].

Crystal structures were determined for the Fab fragment of 10G5H6 alone and in complex with the D3 domain of hIL-13Rα1. Surprisingly, the latter structure did not consist of the expected D3 monomer in complex with Fab 10G5H6, but rather a dimer in complex with two Fab fragments (Figure 4c). Relative to the published structure of D3 in hIL-13Rα1, our D3 fragment displayed N- and C-terminal domain swapping that was intrinsic to dimer formation. Since the swapped A, B and G strands showed extensive crystal lattice packing contacts, it is likely that this domain swapping favoured crystal formation in our experiments. It was observed during the purification of hIL-13Rα1 D3 that it existed as both monomeric and dimeric forms that could be separated by gel filtration. It seems likely, given the results of the structural determination, that the dimeric D3 observed during purification resulted from domain swapping. Although monomeric D3 was isolated for use in 10G5H6 complex formation, the possibility cannot be excluded that there was dimeric D3 in the preparation or that dimer formation occurred subsequent to isolation of the monomer, either before or during crystallization.

Three-dimensional domain swaps of this type were first observed for diphtheria toxin and ribonuclease A oligomerization and have now been observed for a large number of different proteins [31,32].

The major deviations (other than the domain swap) between the D3s in the two structures involved a 3.7 Å movement of Cys231 and a 4.5 Å movement of Arg230 with a consequent 5.6 Å movement of the important IL-13-binding residue Leu293. Comparison of 10G5H6 Fab alone or in complex with D3 indicated some similarly dramatic movements especially of light chain Tyr92 which forms part of the binding pocket for Leu293 of D3. Leu293 (Leu319 in the nomenclature of LaPorte et al. [5]) is a key interacting residue of IL-13Rα1 forming Site IIa contacts with both IL-4 and IL-13 [5]. Since 10G5H6 Fab bound with high affinity to both D3 and the full-length ectodomain of hIL-13Rα1 and, in both cases, the affinity of this interaction was decreased approximately 100-fold by the L293A mutation, it is likely that the B–C loop structure in the receptor, in which Leu293 is located, displays significant mobility. Indeed the B–C loop conformation is among the most variable in fibronectin III domain structures of other receptors and in hIL-13Rα1, it makes direct interactions with bound ligands. It is possible therefore that the B–C loop in the unliganded receptor has a conformation quite different from that seen in the liganded state.

Superimposition of our complex structure on the D3 domain of the holoreceptor complex [5] indicates that the Fab fragment occupies much of the space normally occupied by the D2 domain of the IL-4Rα chain (Supplementary Figure S4 at It is therefore likely that 10G5 inhibits IL-13 and IL-4 binding and biological activity both by binding ligand-contact residues in D3 (Leu293), thus inhibiting ligand binding, as well as occupying the site required for IL-13Rα1 D3 to interact with ligand and with IL-4Rα D2.

The first affinity-matured version of 10G5 (10G5H6) contained a single amino acid change of Phe100 to methionine. Computational replacement of Met100 in the 10G5H6 complex structure with phenylalanine resulted in steric overlap with the important contact residue Leu293 of hIL-13Rα1, suggesting a plausible explanation for the increased affinity of 10G5H6 (Supplementary Figure S5 at

Most other changes in the HCDR3 region resulted in loss of binding because they included the contact residues Asn102 or Trp103 or residues adjacent to them. However, mutations F106L and Y108H in HCDR3 increased affinity, presumably by subtly altering the distance or orientation of these contact residues. Similarly mutated residues in the LCDR3 region that increased binding affinity (E93A and T94S) are not ligand-contact residues, but are directly adjacent to the contact residues Gln91 and Tyr92 (Figure 5).

In summary, we have reported the generation of human neutralizing antibodies against hIL-13Rα1 that inhibit IL-13 and IL-4 actions and thus may find clinical use in treating asthma. The structure of the antigen–antibody complex revealed the key antibody-contact residues in hIL-13Rα1 required to inhibit IL-13 binding, especially Leu293. Furthermore, mutation analysis showed how the binding affinity of the antibody could be improved.


Arna Andrews, Louis Fabri, Hal Braley, Andrea Woods, Cindy Luo, Thomas Garrett, Ping Lu, Cheryl Ireland, Robin Ernst and Gail Forrest designed and performed experiments. Peter Czabotar interpreted experiments. Nicholas Redpath, Yibin Xu, Nicholas Wilson, Manuel Baca, Douglas Hilton, Andrew Nash, Jian-Guo Zhang, Zhiqiang An, Dennis Zaller, William Strohl and Nicos Nicola designed experiments, interpreted and summarized data. Nicos Nicola wrote the paper with input from all authors.


The work was supported by a Merck, Sharp & Dohme Research Grant (2009-2010), the Australian National Health and Medical Research Council [grant numbers 257500 and 461219], the Australian Cancer Research Foundation, the Victoria Government Operational Infrastructure Scheme, and the Australian Government Independent Research Institutes Infrastructure Support Scheme.


We thank P.M. Colman and M.C. Lawrence for structural insights and continual support. We also thank the IL-13R antibody team at Merck Research Laboratories for their collective contribution to the project. Data were collected on the MX1 and MX2 beamlines at the Australian Synchrotron, Melbourne, VIC, Australia. Crystallization trials were performed at the Bio21-C3 (Collaborative Crystallisation Centre). N.J.W., A.E.A., L.J.F., H.B., M.B. and A.N.D. are or were employees of CSL Ltd. P.L., C.I., R.E.E., A.W., G.F., Z.A., D.M.Z. and W.R.S. are or were employees of Merck, Sharp and Dohme.


  • Co-ordinates for 10G5H6 Fab and 10G5H6 Fab–ectodomain 3 complex have been deposited in the PDB under codes 4HWE and 4HWB respectively.

Abbreviations: CCL, CC chemokine ligand; CDR, complementarity-determining region 3; Ches, 2-(N-cyclohexylamino)ethanesulfonic acid; ECD, extracellular domain; FBS, fetal bovine serum; HCDR, heavy-chain CDR; HEK, human embryonic kidney; HRP, horseradish peroxidase; IL, interleukin; IL-4Rα, IL-4 receptor α; IL-13Rα1, IL-13 receptor α1; h, human; LCDR, light-chain CDR; NHDF, neonatal human dermal fibroblast; Ni-NTA, Ni2+-nitrilotriacetate; PBMC, peripheral blood mononuclear cell; PBST, PBS containing Tween 20; PEG, poly(ethylene glycol); RMSD, root mean square deviation; RU, relative units; STAT6, signal transducer and activator of transcription 6; TARC, thymus- and activation-related chemokine; VH, variable heavy; VL, variable light


View Abstract