Solution structure and antibody binding studies of the envelope protein domain III from the New York strain of West Nile virus.

The solution structure of domain III from the New York West Nile virus strain 385-99 (WN-rED3) has been determined by NMR methods. The West Nile domain III structure is a beta-barrel structure formed from seven anti-parallel beta-strands in two beta-sheets. One anti-parallel beta-sheet consists of beta-strands beta1 (Phe(299)-Asp(307)), beta2 (Val(313)-Tyr(319)), beta4 (Arg(354)-Leu(355)), and beta5 (Lys(370)-Glu(376)) arranged so that beta2 is flanked on either side by beta1 and beta5. The short beta4 flanks the end of the remaining side of beta5. The remaining anti-parallel beta-sheet is formed from strands beta3 (Ile(340)-Val(343)), beta6 (Gly(380)-Arg(388)), and beta7 (Gln(391)-Lys(399)) arranged with beta6 at the center. Residues implicated in antigenic differences between different West Nile virus strains (and other flaviviruses) and neutralization are located on the outer surface of the protein. Characterization of the binding of monoclonal antibodies to WN-rED3 mutants, which were identified through neutralization escape experiments, indicate that antibody neutralization directly correlates with binding affinities. These studies provide an insight into theoretical virus-receptor interaction points, structure of immunogenic determinants, and potential targets for antiviral agents against West Nile virus and highlight differences between West Nile virus and other flavivirus structures that may represent critical determinants of virulence.

In 2002, the mosquito-borne West Nile virus (WNV) 1 (family Flaviviridae, genus Flavivirus) was responsible for the largest outbreak of arthropod-borne encephalitis recorded in the Western hemisphere. In that year, 4156 human infections and 284 deaths were reported in the United States (1). In 2003, the WNV epidemic continued with large numbers of human and animal disease across North America, and it was detected for the first time in Mexico and Central America (e.g. Refs. [2][3][4]. The objective of this study was to solve the solution structure for the putative envelope (E) protein, the receptor-binding domain of WNV with the long term goal of using these findings for the development of structure-based vaccines or antiviral agents. Currently, there are no approved vaccines for WNV or therapeutic treatments for West Nile encephalitis.
The flaviviruses are small, enveloped, positive-sense RNA viruses that are transmitted primarily by either mosquitoes or ticks. The translation of the single open reading frame within the viral genome, followed by co-and posttranslational cleavage, results in ten viral proteins, three structural proteins (core, membrane, and E) and seven non-structural proteins. The non-structural proteins are involved in viral replication and pathogenesis, whereas the three structural proteins are assembled into the mature virus particle.
The E protein is the major surface protein of the flavivirus virions. The E protein is also the primary immunogen and plays a central role in virus attachment and entry to cells via membrane fusion. The x-ray crystallographic structures of the E protein ectodomains of both the tick-borne central European encephalitis virus (5) and the mosquito-borne dengue-2 virus (6) have been solved. Both proteins contain three distinct structural domains (domains I, II, and III) that correspond to previously characterized antigenic domains (7). Domain III (D3) of the E protein was initially proposed as the likely receptorbinding domain of the flaviviruses because of its structural characteristics. These include an IgC-like fold and a four-amino-acid loop that contains an RGD integrin-binding motif in several of the mosquito-borne flaviviruses (5). More recent studies have shown that D3 is directly associated with binding of dengue-2 virus (DEN2V) (8), WNV, and the tick-borne Langat virus 2 to cells. Domain 3 of WNV (WN-rED3) has also been shown to contain epitopes recognized by virus-neutralizing monoclonal antibodies (9). X-ray crystallography (5, 6) and cryo-electron microscopy studies (6,10,11) of the mosquitoborne WNV, DEN2V, and yellow fever virus have found that the E protein is arranged in dimeric form on the virus surface. Located at the 5-and 3-fold axes of symmetry, D3 projects slightly above the virion surface, allowing access to potential receptor molecules and antibodies. The 5-fold axis is arranged such that there is a "pore" between the D3 molecules. When the flaviviruses bind to a target cell, they are internalized into an endosome that is acidified. The drop in pH causes a dramatic conformational shift in the E protein that exposes the viral fusion peptide (12), allows the endosomal and viral membranes to fuse, and provides for the release of the viral RNA into the cell cytoplasm.
The role of the flavivirus E-D3 in virus-receptor binding makes it an attractive target for the development of vaccines, antiviral agents, and/or diagnostic antigens. The solution structure of the WN-rED3 described here, along with that of the closely related Japanese encephalitis virus (JEV) (13), provides fundamental knowledge for the development of structurebased vaccines and antiviral agents to protect against and treat diseases caused by WNV and related flaviviruses.

EXPERIMENTAL PROCEDURES
Protein Expression and Purification-Uniformly 13 C, 15 N-labeled recombinant WN-rED3 from neuroinvasive lineage 1 West Nile virus strain 385-99 was overexpressed as a maltose-binding protein fusion using the pMAL-c2X vector as described previously (9), and the 115residue WN-rED3 was purified using size-exclusion chromatography. Residues Leu 6 -Ser 112 of the 115-residue protein align with residues Leu 297 -Ser 403 of the 385-99 envelope protein.  were included in this construct to increase protein solubility. In addition, the QuikChange XL site-directed mutagenesis kit (Stratagene) was used to derive E-D3 fusion proteins encoding mutations at K307R or T330I. These mutations had been shown previously to affect neutralization of WNV by monoclonal antibodies (9).
Monoclonal Antibody Neutralization and Binding Assays-Neutralization assays with four WNV-specific neutralizing antibodies, designated 7H2, 5H10, 5C5, and 3A3 (Bioreliance), against WNV NY99 and variants monoclonal antibody (MAb) R -5H10 and -5C5 were performed with MAbs at 10 nM concentration as described elsewhere (9). The binding characteristics of the four MAbs to recombinant NY99 E-D3 and the K307R and T330I mutant E-D3 MBP fusion proteins were investigated using a titration enzyme-linked immunosorbent assay similar to that described by Lin and Wu (15). Briefly, wells of 96-well enzyme-linked immunosorbent assay plates were coated overnight at 4°C with 60 ng/well purified fusion protein (equivalent to ϳ11 ng of E-D3). After blocking in phosphate-buffered saline containing 0.5% Tween 20 (phosphate-buffered saline-Tween) and 3% bovine serum albumin, serial dilutions of MAbs from 100 to 0.003 nM were dispensed into replicate wells and incubated at room temperature (25°C) for 60 min. After washing three times with phosphate-buffered saline-Tween, horseradish peroxidase-labeled anti-mouse Ig antiserum was added to each well, and the plates were incubated again, prior to washing and addition of TMB substrate. Color development was for 15 min, after which the reactions were stopped with 3 M HCl, and absorbance values were read at 450 nm. Binding curves and K d values were determined using Sigmaplot (SPSS Inc., Chicago, IL).
NMR Spectroscopy and Generation of Restraints-The NMR sample contained 0.7 mM protein in 50 mM K 2 HPO 4 (pH 6.8), 100 mM NaCl, 10 mM NaN 3 , and 0.1 mM EDTA in 90% H 2 O, 10% D 2 O. All NMR experiments were acquired on Varian UnityPlus 600 or 750 MHz spectrometers at 25°C. Sequence-specific chemical shifts for the backbone atoms were obtained from three-dimensional HNCA, HNCACB (16), and HNCO (17) experiments. The backbone assignments were verified through sequential NH-NH and NH-H␣ nuclear Overhauser effects (NOEs) in a 150-ms mixing time 15 N-edited three-dimensional nuclear Overhauser enhancement spectroscopy-heteronuclear single quantum correlation spectrum (18). Side chain chemical shifts were obtained from three-dimensional H(CCO)NH-total correlation spectroscopy (TOCSY) (19), CC(CO)NH-TOCSY, TOCSY-HSQC{ 1 H, 15 N}, and HCCH-TOCSY (20) experiments. The CC(CO)NH-TOCSY experiment also aided in the assignments of the C-␣ and C-␤ carbons in cases of degeneracy in the HNCACB experiment and in the assignments of the Asn and Gln side chain amide groups. Aromatic chemical shifts were assigned based on a CT-1 H, 13 C-HSQC spectrum, NOE data, and a HCCH-TOCSY spectrum with the carbon frequency centered on the aromatic carbons. Spectra were processed using VNMR v6.1b (Varian, Inc.) or Felix98 (MSI, Inc.) software.
The program SANE (21) was used to facilitate the assignment of some NOE cross-peaks and for the generation of restraints. Within the SANE program, chemical shift, distance cutoff, and secondary structure filters were used. The NOE restraints were separated into three bins based on their volumes, and the upper distance limits of the restraints were set to 2.8, 3.8, and 5.5 Å. The 1239 NOE-based restraints (see Table I) consist of 299 intra-residue, 433 sequential, 104 medium range, and 403 long range NOE restraints.
The program TALOS (22) was used to derive / angle restraints. After two rounds of preliminary molecular dynamics calculations, those dihedral angles inconsistent with the NOE data were removed or changed to the / angle exhibited by fewer of the proteins in the TALOS data bank. Additionally, dihedral angle and chiral angle restraints were also used.
Restrained Molecular Dynamics Calculations-One hundred random structures were generated by annealing the protein at 1500 K, obtaining the coordinates every 5 ps, and minimizing the structures obtained. During the annealing process, chiral restraints were applied to maintain the correct chirality for all amino acids. Each of these structures was then subjected to 15 ps of restrained molecular dynamics with the following protocol. During the first 5 ps, the restraints were increased linearly from 0 to 20 kcal/mol Å 2 , the non-bonded interactions (electrostatic, van der Waals, and hydrogen bonding terms) were set to 10% of their normal values, and the temperature was increased from 100 to 700 K. During the next 5 ps the non-bonded interactions were increased linearly from 10 to 100%, while the temperature remained set to 700 K. During the final 5 ps, the temperature was decreased linearly from 700 to 100 K while the full force constants of the restraints and the nonbonded interactions were applied. Each of the structures obtained was subsequently subjected to 2000 steps of energy minimization. Fifteen structures with low restraint penalties were then chosen for the structural ensemble. All calculations were done with the SANDER module within AMBER6 (23). Structures were visualized with MIDAS (23) or MOLMOL (24), which was used to generate Figs. 1 and 3. Atomic coordinates for fifteen structures have been deposited with the Protein Data Bank (PDB ID 1S6N) and the chemical shifts have been deposited at the BioMagResBank under accession code 6046 (25).  (  Table I). The dihedral angle restraints used were very restrictive in angular space; the tightest restraint was Ϯ3°, and 75% of the dihedral angles were restricted to less than Ϯ20°from the target angle. Therefore, a violation cutoff of 20°is reasonable for the restraints used. The r.m.s.d. restraint error was 0.0156 Ϯ 0.0017 Å, and the r.m.s.d. angle restraint error was 1.60 Ϯ 0.06°. The structural ensemble has an average pairwise atomic r.m.s.d. of 0.65 Ϯ 0.12 Å for the backbone atoms and 0.98 Ϯ 0.17 Å for all heavy atoms for residues 13-108 of the protein.

Quality of the Final
The program PROCHECK (14) was used to analyze the quality of the ensemble of structures. Ramachandran analysis of the non-glycine, non-proline residues indicated 98% of the residues are in the two most favored regions of the Ramachandran plot. Specifically, 76.8% are in the most favored region, 21.2% are in the additionally allowed region, 0.7% are in the generously allowed region, and 1.2% are in the disallowed region. Ser 400 is in the disallowed region in two of the fifteen structures, whereas Lys 307 , which is near the disulfide bond between Cys 305 and Cys 336 , is in the disallowed region in every structure.
Interaction of WNV-specific Neutralizing Monoclonal Antibodies with E-D3-Previously we have described the preliminary determination of in vitro neutralization characteristics of four WNV-specific MAbs that bound to WNV E-D3 (9). To better characterize the interactions of these MAbs with E-D3 and the relationship between MAb binding and virus neutralization we titrated the specific binding of each MAb with the WN-rED3 MPB fusion protein in enzyme-linked immunosorbent assays. Based on these titrations ( Fig. 2A), the calculated K d values for each MAb were comparable, ranging between 0.1 and 0.2 nM. Interestingly, the maximal binding of MAb 7H2 was 20% higher than those of the other MAbs (Fig. 2), which was consistent with the greater neutralizing activity of this Mab against WNV NY99 (Fig. 2, Table II).
Neutralization escape variants selected with MAbs 5C5 (designated Mab R -5C5) or 5H10 (MAb R -5H10) encoded mutations at K307R or T330I, respectively (9). When mapped onto the DEN2 E protein dimer structure, the WNV E-D3 NMR structure clearly shows that these residues are surface-exposed and located spatially near to each other on the upper outside face of E-D3 (Fig. 3B). To better quantify the effects of these mutations on the epitopes recognized by the anti-WNV MAbs, the ability of all four MAbs to bind to recombinant E-D3 proteins that encoded the K307R or T330I mutations was assessed. Despite the close proximity of residues 307 and 330 on the E-D3 surface (Fig. 1), significantly different mutation effects were observed in the binding of the four MAbs (Fig. 2B). The maximal binding of MAb 7H2 (at 100 nM concentration) to either mutant protein was ϳ2-fold lower than its binding to the wild-type WNV NY99 ED3. In contrast, either mutation resulted in ϳ5-10-fold reductions in the binding of either the 5H10 or 3A3 MAbs to levels that were only slightly higher than the background. The binding of MAb 5C5 was affected differently by the two mutations. At concentrations up to and including 100 nM, binding to the K307R protein was not detected, whereas only an ϳ2-fold reduction in peak binding occurred with the T330I mutant. In general, the binding data were consistent with neutralization data obtained for the four MAbs against the MAb R -5C5 and -5H10 variant strains (Table II). The data also confirmed our previous hypothesis that distinctly different binding interactions can occur at this surface despite the relatively small E-D3 surface area available for antibody binding.
NMR studies of the binding of a JEV-specific neutralizing MAb binding to JEV E-D3 have indicated that multiple interactions occur with residues across the entire upper surface of E-D3 (13). However, escape from neutralization by that JEVspecific MAb was primarily associated with mutations at the outside edge of E-D3 in the ␤2-␤3 loop (15). This region is homologous to the location of the antigenically significant mutations that we have reported in WNV (9). In JEV-rED3, the positively charged K307 of WN-rED3 is replaced by a negatively charged glutamic acid, and residues T330 and T332 are replaced by serines. This suggests that the upper edge of domain III formed by the loop linking ␤2 and ␤3 represents an important antigenic determinant for both WNV and JEV. The fact that none of these mutations have been associated with viability or virulence changes in either JEV (15) or WNV (9) suggests that other surface-exposed residues in E-D3 may represent critical receptor binding elements and determinants of tropism and virulence for these viruses. DISCUSSION The solution structure of domain III from the envelope protein of the West Nile virus represents the first high-resolution structure from the immunogenically important E protein of this medically important virus. The structure is similar to the domain III structures of both the DEN2V and the JEV but has significant differences that most likely contribute to the antigenic and tropism differences between these mosquito-borne flaviviruses. A number of E-D3 residues in WNV lineage 1 strains that are different from those in some WNV lineage 2 strains, JEV, and/or DEN2V strains are located near the top of the ␤-barrel (Fig. 1B). Among them are Lys 307 and Thr 330 , which affect antibody binding and escape from neutralization (see below). In addition, residues Thr 332 , Tyr 329 , Leu 312 , and Ala 369 are also solvent-exposed, may play a role in the determination of antigenic specificities, and serve as additional starting points for mutation studies.
Comparison with Domains III of the Dengue-2 and Japanese Encephalitis Virus E Proteins-Although the WN-rED3 structure is similar to domain III from both DEN2V (6) and the JEV (JEV-rED3) (13) E proteins, there are significant differences. When aligning the backbone atoms of the six major ␤-strands present in domain III of all three proteins, WN-rED3 has an atomic r.m.s.d. of 2.5 and 2.8 Å versus the domain III of DEN2V and JEV, respectively. Small differences do exist between the 1.2 a A 20°-limit for restraint errors was used due to the small error bars used on the TALOS derived distance restraints, which were as small as Ϯ3°and were less than Ϯ20°for 75% of the dihedral restraints.

FIG. 2.
A, binding of WNV-specific neutralizing monoclonal antibodies (7H2, 5H10, 5C5, and 3A3) to WN-rED3. B, binding of monoclonal antibodies at 100 nM concentration to either WN-rED3 or its mutants K307R and T330I. sheet structures of the three proteins (Table III). The first two ␤-strands of WN-rED3, spanning Phe 309 -Asp 317 and Val 323 -Tyr 329 , are identical in length to strands ␤1 and ␤2 in the both the JE-rED3 and the DEN2 domain III structures. Strand ␤3 of WN-rED3, spanning Ile 340 -Val 343 , is one residue shorter on the C-terminal end compared with the analogous strands ␤3 (Ile 339 -Ale 343 ) in JE-rED3 and ␤4 (Phe 337 -Met 340 ) in DEN2V. The tworesidue strand ␤4 of WN-rED3 (Arg 354 -Leu 355 ) coincides with the last two residues of the DEN2 twisted ␤-strand ␤5 (Val 347 -Leu 351 ). The ␤5 strand of DEN2V appears to be two sequential ␤-strands that are three and two residues long, twisted ϳ90°r elative to each other. An analogous ␤-strand is not present in the JE-rED3 structure. Strand ␤5 (Lys 370 -Glu 376 ) of WN-rED3 is one residue longer than ␤4 of JE-rED3 (Lys 369 -Met 374 ) on the C-terminal end, and it is one residue longer than ␤7 (Val 365 -Glu 370 ) of DEN2 on the N-terminal end. ␤-Strands ␤6 (Gly 380 -Arg 388 ) and ␤7 (Gln 391 -Lys 399 ) of WN-rED3 are each two residues longer than the analogous strands, ␤8 (Gly 374 -Ile 380 ) and ␤9 (Leu 387 -Lys 393 ), in DEN2V. The analogous strands ␤5 (Ser 381 -Gly 386 ) and ␤6 (Gln 391 -Trp 396 ) in JE-rED3 are each one residue shorter than those of the DEN2V and three residues shorter than those of WN-rED3. Two short interacting ␤-strands that are present in DEN2V domain III, ␤3 (Cys 333 -Lys 334 ) and ␤6 (Ile 357 -Val 358 ), are not observed in either the WN-rED3 or JE-rED3 solution structures. The two ␤-strands not present in the JEV and WNV domain III solution structures are near the top of the DEN2V ␤-barrel. As previously suggested (13), the absence of these two strands may be because of the lack of dimeric head-to-tail interactions between domain III and domains I and II, which are not part of the present protein studied or the JEV domain III solution structure.
Although the secondary structures and global folds of the West Nile, Japanese encephalitis, and dengue-2 virus domain III proteins are similar, the protein surfaces presented as antigens or targets for potential therapeutic drugs are significantly different (Fig. 4). Domain III forms a pentameric pore in the virion, and the residues near this interface are shown in Fig. 4, at the top. For WN-rED3, residues Thr 332 , Ala 365 , Thr 366 , and Asn 368 form a small pocket at this site. The width of the pocket between residues Thr 332 and Thr 366 is 6.85 Å. A similar but wider pocket is present in the JEV-rED3 structure. In the JEV-rED3 structure, the pocket is formed by residues Ser 330 , Asp 331 , Pro 333 , Ser 363 , Ser 364 , and Asn 366 . The width across the pocket between residues Ser 330 and Ser 364 , which are analogous to Thr 332 and Thr 366 in WN-rED3, is 8.95 Å. Thus, in the West Nile virus, this pocket is about 2 Å narrower. In addition, the pocket in the JEV-rED3 structure involves more residues, including the negatively charged Asp 331 . In the West Nile structure, the side chain of the analogous residue Asp 333 is situated away from this pocket.
The electrostatic charges are also significantly different near these pockets and on the face of the proteins. The electrostatic surface formed in WN-rED3 by residue Lys 307 sandwiched between two negatively charged residues, Asp 333 and Glu 390 (Fig. 4A, top right), is very different from that formed by JEV-rED3 (Fig. 4B, top right). In the JEV-rED3 structure, the negatively charged residue Asp 388 is situated between two positive residues, Lys 306 and Lys 389 .
On the central face of the structures shown in Fig. 4, Asp 326 and Lys 370 of WN-rED3 align well with residues Glu 324 and Lys 368 of JE-rED3. However, in the WN-rED3 structure an additional positively charged residue, Lys 310 , is just above and to the right of these residues, whereas in the JEV-rED3 struc-ture an additional positively charged residue, Lys 311 , is directly under the aforementioned residues. These subtle structural changes are responsible for protein surfaces that interact differently with ligands such as antibodies, potential drugs, and possibly cell surface proteins. As an example, the rings of positive lysines in each of the proteins, such as Lys 307 , Lys 310 , Lys 370 , and Lys 337 in WN-rED3, serve as attractive targets for nucleic acid-based aptamer strategies. Thus, the WN-rED3 structure provides direction for future drug development and mutation effect studies and provides insight in the analysis of current mutation studies.
Antibody neutralization experiments indicate mutations K307R and T330I have differential effects with respect to the antibody used in the experiment. At 100 nM concentration, which is 50 -100-fold greater than the K d for binding of each MAb to wild-type NY99 rE-D3, an ϳ2-fold reduction in binding was observed for MAb 7H2 to either the K307R or T330I mutants, relative to the NY99 protein. MAbs 5H10 and 3A3 exhibited a 5-10-fold reduction in binding to either of the mutants. However, MAb 5C5 binding was decreased Ͼ10-fold to the K307R mutant but was decreased only 2-fold to the T330I mutant. These differences in binding correlated with differences in the ability of these MAbs to neutralize WNV strains encoding those mutations and confirmed that, despite the relatively small size of E-D3, distinctly different binding interactions can occur at its upper surface. In addition, the clustering of antigenically significant mutations at the upper outside edge of E-D3 in both WNV and JEV suggest that other surface-exposed residues may represent critical receptor binding determinants. Accordingly, these solvent-exposed residues represent potentially important targets for rational design of vaccines or therapeutic agents to control WNV disease.
This high-resolution structure of the WN-rED3 provides significant insight toward understanding the immunological differences between a mosquito-borne flavivirus, in general, and members of the JEV serocomplex, more specifically. These subtle structural differences, particularly in areas that constitute virus-specific neutralizing epitopes, may provide the basis for the lack of significant immunological cross-protection between viruses such as WNV, JEV, and DEN2V, despite the antigenic cross-reactivity that exists between them. The structural differences and variability of surface exposed residues of the putative receptor-binding domain III may also contribute to differences in host cell tropism and to subsequent disease pathogenesis of a particular flavivirus (e.g. neurotropic WNV and JEV versus DEN). The identification of conserved and variable structures and surface residues provides potential targets for the development of structure-based antiviral agents that are targeted against WNV infection, or potentially, infection by all flaviviruses.