INTRODUCTION
Influenza virus is the cause of seasonal epidemic and sporadic pandemic flu outbreaks. The hemagglutinin (HA) surface glycoprotein mediates viral recognition of host cells through its interaction with sialic acid receptors (
1,
2). The globular “head” domain of HA is immunodominant, likely due to its accessibility on the surface of viruses, and, consequently, antibodies are rapidly generated against it. However, the HA head undergoes continual antigenic drift, which results in escape from the host immune response through amino acid changes on its surface or by masking neutralizing epitopes with glycans. Antibodies generated against HA are typically strain specific, which necessitates nearly annual vaccine strain reformulations. In contrast, the residues that form the receptor binding site (RBS) are functionally constrained for receptor binding and, thus, have restricted mutational freedom. As such, the RBS is a prime target for virus neutralization by broadly neutralizing antibodies that prevent viral-host interactions (
3). However, the footprint of the sialoglycan receptor on the RBS is much smaller than that of an antibody. Hence, most antibodies that block the RBS also contact the hypervariable regions surrounding it, which leads to strain-specific binding. Nevertheless, a few antibodies that target the RBS display a broader spectrum of reactivity than those that target HA elsewhere on the head (
4–10). S139/1 reaches into the RBS and has heterosubtypic neutralizing activity (
7). C05, which also neutralizes highly divergent viruses, similarly enters the RBS and remarkably accomplishes this interaction using essentially a single antibody loop (
6). CH65 and CH67 are broadly neutralizing H1-specific antibodies and, unlike S139/1 and C05, make use of receptor mimicry (
5,
9); however, CH65 does not neutralize 1918 or 2009 H1 pandemic strains that are now the current seasonal H1 epidemic strains (
5).
We have previously reported the identification and characterization of a human monoclonal antibody, 5J8, that possesses neutralization activity and therapeutic efficacy against H1 viruses spanning decades, including the 1918 and 2009 pandemic viruses (
11). Notably, the pandemic strains contain a basic amino acid insertion at residue 133a (between residues 133 and 134) that has been proposed to sterically clash with other RBS-targeted antibodies (
6,
7). Here, we present the crystal structure of the bacterially expressed HA1 globular head domain from the A/California/07/2009 (H1N1) (Cali07/2009-H1) virus in complex with 5J8 Fab. The complex structure reveals that Lys133a, which is conserved in pandemic H1N1 strains, makes favorable electrostatic interactions with an acidic patch on the antibody. Similar to other RBS-targeted antibodies (
6,
7), avidity through a bivalent IgG extends the antibody's breadth of neutralization and allows it to bind divergent HA strains within the H1 subtype. Most strikingly, 5J8 reaches into the RBS and utilizes receptor mimicry, similar to that of CH65 and CH67 (
5,
9). That these three antibodies all use receptor mimicry and, hence, display a common theme for receptor site recognition, as well as provide complementary coverage of H1 viruses spanning the past four decades, reinforces the RBS as a promising site of vulnerability on the HA.
MATERIALS AND METHODS
Fab and IgG cloning, expression, and purification.
5J8 and CH65 Fab were cloned in a pFastBac dual vector (Invitrogen) with N-terminal gp67 and honeybee melittin secretion signal peptides fused to the heavy and light chains, respectively, and a C-terminal His
6 tag fused to the heavy chain. Recombinant bacmid DNA and baculovirus were generated as previously described (
7). The Fabs were purified by Ni-nitrilotriacetic acid (NTA) (Qiagen) and Mono S (GE Healthcare) chromatography. The purified Fabs were then dialyzed into 20 mM HEPES (pH 7.4) and 150 mM NaCl, flash frozen with liquid nitrogen, and stored at −80°C.
The heavy and light chains of the 5J8 and CH65 IgGs were cloned separately in a phCMV3 vector (Genlantis) with an N-terminal Ig kappa secretion signal peptide and a C-terminal His6 tag fused to the heavy chain. The heavy and light chain plasmids were transiently transfected at a 2:1 ratio into HEK293F suspension cells and incubated for six days. The IgGs were purified by Ni-NTA (Qiagen) and protein A (GE Healthcare) chromatography and then buffer exchanged into 20 mM Tris (pH 8) and 150 mM NaCl.
HA expression and purification.
HA was prepared for binding studies and crystallization as previously described (
7,
12). Briefly, each HA was fused with an N-terminal gp67 signal peptide and a C-terminal BirA biotinylation site, a thrombin cleavage site, a trimerization domain, and a His
6 tag. The HAs were expressed as described above for the Fabs and purified by Ni-NTA. The HAs were either matured by trypsin (New England BioLabs) for crystallization or biotinylated with BirA (
12) for binding studies.
Flagellin-HA1 preparation.
The HA1 protein was derived from a recombinant fusion protein, STF2.HA1, that contains N-terminal
Salmonella enterica serovar Typhimurium flagellin residues 1 to 505 linked to the N terminus of A/California/07/2009 (H1N1) (Cali07/2009-H1) residues 55 to 271 (H3 numbering), which was expressed in batch bioreactor cultures as previously described (
13).
Crystallization.
Apo 5J8 Fab crystals were grown by sitting drop vapor diffusion at 4°C by mixing 0.5 μl of protein (6.3 mg/ml) with 0.5 μl of reservoir solution (0.2 M calcium acetate, 11% [wt/vol] polyethylene glycol 3350 [PEG 3350]). Crystals were cryoprotected in mother liquor supplement with 20% (vol/vol) ethylene glycol, flash cooled, and stored in liquid nitrogen until data collection.
The 5J8 Fab-STF2.HA1 complex was prepared by mixing individually prepared proteins in a 1.1:1 molar ratio and then purified by gel filtration. Fractions corresponding to the complex were pooled and concentrated to 16 mg/ml for crystallization screening. However, STF2.HA1 degraded over time, likely at or around the linker connecting flagellin and HA1, as only HA1 in complex with 5J8 Fab crystallized. Crystals grew at 23°C by sitting drop vapor diffusion by mixing 0.6 μl of protein solution (16 mg/ml) with 0.5 μl of reservoir solution (0.1 M Tris [pH 8.5], 23% [wt/vol] PEG 8000, 0.2 M magnesium chloride). Crystals were cryoprotected in mother liquor supplemented with 15% (vol/vol) glycerol, flash cooled, and stored in liquid nitrogen until data collection.
The trimer-stabilized A/California/04/2009 (H1N1) (Cali04/2009-H1 HA2 E47G) HA crystals were grown by sitting drop vapor diffusion. The HA was concentrated to 17 mg/ml in 20 mM Tris (pH 8.0), 100 mM NaCl, and 0.02% (vol/vol) NaN3 and crystallized in 0.1 M Tris (pH 8.8), 25% (wt/vol) methoxypolyethylene glycol 2000 (MPEG 2000) at 23°C. Crystals were flash cooled in mother liquor supplemented with 12% (vol/vol) ethylene glycol and stored in liquid nitrogen until data collection.
X-ray structure determination and refinement.
X-ray diffraction data for the apo 5J8 Fab were collected to a 1.55-Å resolution at the GM-CA CAT 23ID-D beamline at the Advanced Photon Source (APS). The data were processed in space group P2
12
12
1 using XDS (
14). The structure was determined by molecular replacement with Phaser (
15) by using the variable and constant domains of the anti-HIV-1 V3 Fab 3074 (Protein Data Bank [PDB] accession no. 3MLY, chains H and L) as search models, and two Fab copies were found in the asymmetric unit. The crystal exhibited pseudotranslational symmetry. The model was iteratively built using Coot (
16) and refined in Phenix (
17). Refinement parameters included rigid-body refinement (set for each Ig domain), simulated annealing, and restrained refinement, including translation/libration/screw (TLS) refinement (for each Ig domain).
X-ray diffraction data for the 5J8-Cali07/2009-H1 HA1 complex were collected to 2.25 Å at the Canadian Light Source beamline 08B1-1 (CMGF-BM). The data were processed in space group P3
121 using XDS (
14). The complex was determined by molecular replacement with Phaser (
15) by first using one copy of the HA1 from Cali04/2009-H1 (PDB accession no. 3LZG, chain A) residues 55 to 271 (H3 numbering). Next, one copy of the variable and constant domains of the high-resolution 5J8 Fab structure were used as search models after fixing the position and orientation of the HA head. The model was iteratively built using Coot (
16) and refined in Phenix (
17). Refinement parameters included rigid-body refinement (set for the HA1 and each Ig domain), simulated annealing, and restrained refinement, including TLS refinement (set for the HA1 and each Ig domain).
X-ray diffraction data for the trimer-stabilized HA (Cali04/2009-H1 HA2 E47G) were collected to a 2.20-Å resolution at beamline 12-2 at the Stanford Synchrotron Radiation Lightsource (SSRL). The mutant HA structure was determined by molecular replacement using the program Phaser (
15) by using the native Cali04/2009-H1 HA (PDB accession no. 3LZG) as the starting model. The refinement was performed in Refmac5 (
18) and Phenix (
17), and model building was carried out with Coot (
16).
Sample preparation and imaging by electron microscopy.
Copper grids (400 mesh) were coated in nitrocellulose and a thin layer of carbon. The grids received samples shortly (<20 min) after glow discharging. Negative staining was performed through application of 4 μl of sample (∼0.02 mg/ml of Fab-HA complex diluted in Tris-buffered saline [TBS]) to the grid, with blotting to remove excess sample, followed by two cycles of staining with 4 μl of “Nano W” stain (2% methylamine tungstate [Nanoprobes, Yaphank, NY]), 20 s of incubation, and blotting to remove excess stain.
Micrographs were acquired on an FEI Tecnai Spirit transmission electron microscope (TEM) operating at an accelerating voltage of 120 kV. A Tietz charge-coupled-device camera was used to record 2,048- by 2,048-pixel images at a magnification of ×52,000 and a defocus range of 900 to 1,300 nm. The stage was tilted in five-degree increments, from 0° to 55°, to increase the number of observed orientations. The pixel size was previously calibrated to be 2.65 Å using a two-dimensional catalase crystal. The Leginon software package (
19,
20) was used to automate some steps of data acquisition.
Image processing, volume map determination, and interpretation.
Particles were automatically selected using a difference of Gaussian algorithm (
21) provided in the Appion package (
22), and most subsequent processing steps were facilitated using Appion. Particle boxing was performed using Eman1.9 (
23) and Spider (
24). Xmipp (
25) was used to normalize the boxed images. There was no correction applied for the contrast transfer function. Initial classification was performed with the CL2D program (
26) provided in the Xmipp package (
25). Classification steps were followed by manual analysis in which heterogeneities or low-quality picks were excluded.
Images corresponding to homogeneous complexes were inputted into a projection-matching algorithm implemented in Eman 1.9 (
23). An unliganded HA (PDB accession no. 4FQV) was low-pass filtered to 30 Å and used as an initial model for the reconstruction by first refining against class averages. The resulting map was then refined against raw particles (128- by 128-pixel box size) for 80 to 90 cycles. Volumes were visualized and interpreted using UCSF Chimera (
27). Fourier shell correlation (FSC) curves were calculated using the eotest protocol in Eman 1.9 (
23) and were fit to a tanh-based function to estimate the resolutions at an FSC value of 0.5.
Structural analyses.
Hydrogen bonds and van der Waals contacts were calculated using HBPLUS and CONTACSYM, respectively (
28,
29). Surface area upon Fab binding was calculated using MS (
30). MacPyMOL (DeLano Scientific) was used to render structure figures. Kabat numbering was applied to the coordinate files using the AbNum server (
31). The final coordinates were validated using the JCSG quality control server (version 2.8), which includes MolProbity (
32). Structural alignments to calculate root mean square deviation (RMSD) values were performed by iterative fitting on the alpha carbons using the McLachlan algorithm (
33) as implemented in the program ProFit (A. C. R. Martin and C. T. Porter,
http://www.bioinf.org.uk/software/profit).
Sequence analysis of the antibody epitopes.
The full-length and nonredundant influenza A HA sequences were downloaded from the Influenza Virus Resource at the NCBI database (
34). At the time of download (20 March 2013), the data set includes 3,700 human sequences from the H1 subtype. The sequences were aligned using MUSCLE (
35) and analyzed using GCG (Accelrys) and custom shell scripts (available from the authors upon request).
Kd determination.
Dissociation constant (
Kd) values were determined by biolayer interferometry using an Octet RED instrument (ForteBio, Inc.) as previously described (
12). Briefly, biotinylated HAs at ∼10 to 50 μg/ml in 1× kinetics buffer (1× phosphate-buffered saline [PBS; pH 7.4], 0.01% bovine serum albumin [BSA], and 0.002% Tween 20) were immobilized onto streptavidin-coated biosensors and incubated with various concentrations of Fab or IgG of 5J8 or CH65. All binding data were collected at 30°C. The
kon and
koff values of each Fab or IgG were measured in real time to determine the
Kd values for each HA tested. The sequences of the HA proteins used in the binding studies and the experimental binding curves for each Fab or IgG for fitting
kon and
koff are reported in the supplemental material.
Protein structure accession numbers.
The atomic coordinates and structure factors reported in this paper are deposited in the Protein Data Bank (
www.pdb.org; PDB accession no. 4M5Y, 4M5Z, and 4M4Y). The reconstruction data reported in this paper are deposited in the Electron Microscopy Data Bank (
www.emdatabank.org; EMDB accession no.
EMDB-5731 and
EMDB-5733).
DISCUSSION
Here, we describe the recognition of H1 pandemic viruses by the human monoclonal antibody 5J8. This antibody inserts its HCDR3 into the RBS of HA and thus blocks viral-host interactions. In addition, 5J8 utilizes avidity through bivalency to extend its breadth of recognition and increase its affinity against highly divergent HA strains, as the bivalent IgG is more potent compared to monovalent Fab. Avidity has also been observed in other previously characterized antibodies that target the RBS (
6,
7) as well as in CH65, which has been further characterized in this study (
Table 2). These data indicate that avidity through bivalency is critical for extending the breadth of neutralization and may be a general mode of antibody recognition against the HA RBS. In addition, a bivalent IgG relaxes the specificity of antibody recognition, allowing it to tolerate some of the hypervariable residues in divergent strains that would have otherwise been moderately or weakly bound by monovalent Fab.
The HCDR3 of 5J8 inserts into the HA RBS and closely mimics the natural sialoglycan receptor. The carboxylate of Asp
H100b is oriented nearly identically to that of the sialic acid carboxylate and makes similar hydrogen bonding networks to conserved receptor binding residues. This mode of receptor mimicry has also been observed in related broadly neutralizing H1 antibodies CH65 and CH67 (
5,
9) (n.b. 5J8 uses D3-3*02 and J4*02, while CH65 and CH67 use D1-1 and J6 germ line genes). In addition, the 5J8 Pro
H100a inserts into a universally conserved hydrophobic pocket in the HA RBS that would be occupied by the acetamide group of sialic acid, which has also been similarly targeted by other antibodies (
5–7,
9,
10). The compounding structural information has revealed a number of common binding modes and recognition hot spots. For instance, these binding details may serve as a template for structure-guided drug discovery by combining common recognition elements for the design of small molecules. The antibody-antigen interactions can also be recapitulated through proteins engineered to target the HA RBS, as has been successfully performed against the HA stem (
44,
45). In addition, the structural information can be used to design immunogens that elicit RBS-targeted antibodies, along similar lines as the germ line-targeting engineered outer domain of HIV-1 gp120 (
46).
Although 5J8 and other antibodies mimic certain moieties of the receptor, the region of the binding site occupied by the glycerol moiety of sialic acid is not contacted by these antibodies (
Fig. 9B). As there is only space for a single antibody loop to enter into the binding groove, the level of receptor mimicry therefore has spatial limitations. Sterics also play a role in antibody recognition, as the 133a insertion present in pandemic H1 strains appears to be an important binding determinant for these H1-specific antibodies. For example, binding by 5J8 depends largely on the presence of the 133a insertion, whereas CH65 appears to favor binding to strains without the insertion (
Table 2), although CH67 modestly neutralizes pandemic strains (
5). The 133a insertion may thus dictate the specificity of any subsequent design efforts against the RBS of H1 isolates. Obviously, it is overly simplistic to distinguish the antibodies by a single amino acid, considering they have distinct binding footprints on HA and use different angles of approach (
Fig. 9). However, it is compelling to note that these antibodies complement each other and jointly recognize all H1 human isolates tested since the H1N1 virus reemerged in humans in 1977 (
Table 2). As avidity by bivalent IgG increases the affinity of each antibody to HA, it could be possible to use a bispecific antibody (
47), i.e., with one arm as 5J8 and the other as CH65 or CH67, as a potential therapeutic or diagnostic for existing and emerging H1 viruses.
Our study also highlights the potential for developing an alternative immunogenic and effective vaccination strategy using an
E. coli expression system. Eukaryotic expression systems are widely used for the preparation of recombinant HAs, as these proteins have glycans on their N-linked glycosylation sites. However, it has been previously reported that HA1 can be recombinantly produced and refolded from
E. coli inclusion bodies with similar biophysical properties to HA produced in insect cells (
36). Moreover,
E. coli-expressed HA1 has been shown to elicit a protective immune response in ferrets (
48), suggesting that a protective antibody response can be generated despite the lack of glycan shielding on the surface of HA, at least in the case of the 2009 H1 pandemic strain. In support of this,
E. coli-expressed fusions of flagellin and Cali07/2009-H1 HA1 or A/Solomon Islands/3/1986 HA1 have been shown to elicit protective antibody titers in humans (
49,
50). We show that the bacterially expressed HA1, which was originally expressed as a fusion protein with flagellin, is properly refolded and is recognized by broadly neutralizing human antibody 5J8. As such, these results provide further evidence that this strategy of producing recombinant HA in
E. coli is suitable for vaccination, as also noted from the ferret experiments (
48) and human studies (
49,
50). The structural knowledge of how 5J8 functions, in combination with the other RBS-targeted antibodies, will potentially further aid and inform vaccine and therapeutic design against the influenza A H1 subtype.
ACKNOWLEDGMENTS
We thank H. Tien of the Robotics Core at the Joint Center for Structural Genomics for automated crystal screening, the staff of the APS GM/CA-CAT 23ID-D, CLS 08ID-1, and SSRL BL12-2 for beamline support, X. Dai and M. Elsliger for crystallographic and computational support, W. Yu, Y. Hua, and A. Arnell for materials and expertise, as well as J.-P. Julien, R. L. Stanfield, and L. Kong for helpful discussions. We thank A. Cheng, T. Nieusma, and J. H. Lee for assistance with EM data collection and interpretation.
The work was funded in part by NIH R56 AI099275 (I.A.W.), by the Skaggs Institute for Chemical Biology, by GM080209 from the NIH Molecular Evolution Training Program (P.S.L.), by a Saper Aude fellowship from the Danish Council for Independent Research, Natural Sciences (N.S.L.), and with federal funds from the Office of the Assistant Secretary for Preparedness and Response, BARDA, HHS, contract no. HHSO100201100011C. Isolation and production of the antibody was supported by NIH grant R01 AI106002 and NIH contract HHSN272200900047C (both to J.E.C.). The electron microscopy data presented here were collected at the National Resource for Automated Molecular Microscopy at TSRI, which is supported by the NIH though the P41 program (RR017573) at the National Center for Research Resources. X-ray diffraction data were collected at the Canadian Light Source, which is supported by the Natural Sciences and Engineering Research Council of Canada, the National Research Council Canada, the Canadian Institutes of Health Research, the Province of Saskatchewan, Western Economic Diversification Canada, and the University of Saskatchewan. The GM/CA CAT has been funded in whole or in part with federal funds from National Cancer Institute (Y1-CO-1020) and National Institute of General Medical Sciences (Y1-GM-1104). Use of the Advanced Photon Source was supported by the U.S. Department of Energy, Basic Energy Sciences, Office of Science, under contract no. DE-AC02-06CH11357.
The content is solely the responsibility of the authors and does not necessarily represent the official views of NIGMS or the NIH. This is The Scripps Research Institute manuscript number 23085.
Vanderbilt University submitted a patent covering the diagnostic and therapeutic use of the antibody described in this paper; J.C.K. and J.E.C. are two of the coinventors on that application.