CryoEM studies of E53 antibody Fab fragment in complex with WNV and DENV

CryoEM experiments failed to form complexes of E53 Fab fragments with mature WNV, although it is possible that a low occupancy of potential E53 binding sites would not become evident in the icosahedral averaging procedure required for the image reconstruction. In contrast, when E53 Fab was incubated with immature WNV or DENV particles, it formed stable icosahedral complexes. The cryoEM reconstructions of immature WNV and DENV complexed with E53 Fab were determined to 15 and 23 Šresolution, respectively. These showed the anticipated trimeric E protein spikes (molecules A, B and C in Figure 1A) as observed in studies of the uncomplexed immature viruses (Zhang et al, 2003, 2007) (Figure 3; Supplementary Figure 1). The separation of the lipid membrane from the nucleocapsid core, a measure of the quality of the cryoEM reconstruction, was seen only in the DENV complex (Supplementary Figure 1). Fitting of the atomic structures of E and pr into the electron density of the complex required only minor movement of the individual E and pr molecules, compared with the known icosahedral arrangement in immature virions (Supplementary Tables III and IV). In both reconstructions, density representing the bound E53 Fab molecules was associated with the fusion loops on molecules A and B, but not with molecule C (Figure 3B and C). Indeed, in immature flavivirus particles, the fusion loops on molecules A and B (Figure 1A) are marginally accessible to water (accessible surface areas (Hubbard and Thornton, 1993) are 678 and 596 Ų, respectively), but the fusion loop on molecule C is completely covered by the pr peptide and neighbouring E proteins.

Open in a separate window

Figure 3

CryoEM reconstruction of immature WNV in complex with Fab fragments of the anti-fusion-loop antibody E53. (A) Stereoview showing the immature WNV–E53 Fab complex at 15 Å resolution. The density is coloured according to radius (red: 190 Å; yellow: 220 Å; green: 235 Å; light blue: 260 Å; blue: 305 Å). The black triangle shows an icosahedral asymmetric unit. The dashed circles mark the positions where the E53 Fab is bound to the fusion loop of molecules A and B, respectively. The scale bar represents 100 Å. Similar results were obtained for immature DENV in complex with E53 Fab (Supplementary Figure 1). (B) Stereoview showing the difference density between the immature WNV–E53 Fab complex and immature WNV (blue) superpositioned onto the surface of immature WNV (gold) at 24 Å resolution. The black triangle marks an icosahedral asymmetric unit. The scale bar represents 100 Å. (C) The difference density (blue mesh) is shown relative to the immature E protein structure coloured as in Figure 1.

Crystallographic structure determination

Data were collected at the Advanced Light Source beamline 4.2.2 (Lawrence Berkeley Laboratories, Berkeley, CA) by the oscillation method at a wavelength of 1.24 Å at 100 K and recorded with a CCD detector. Data were processed, scaled and merged with d^*trek (Pflugrath, 1999). The crystals belong to space group P2₁2₁2 with unit cell dimensions of a=146.41 Å, b=160.14 Å and c=43.88 Å, containing one E53 Fab–E complex per asymmetric unit. Initial phases were obtained by molecular replacement (Collaborative Computational Project, 1994) using the coordinates of WNV E (PDB accession code 2HG0) and an IgG1 Fab (PDB accession code 2IGF). An atomic model was iteratively built in O (Jones et al, 1991) and refined with CNS (Brunger et al, 1998) and Phenix (Adams et al, 2002) and contains residues 7–144, 163–297 of E, 1–208 of the E53 light chain, 1–127 and 136–211 of the E53 heavy chain, and no solvent molecules. The electron density for the Fab variable domains and DII of the E protein including the fusion and AB-loop epitope were of good quality, while significant regions of the Fab constant domains and E protein DI and DII regions distal from the combining site were of very poor quality. Electron density corresponding to DIII was highly ambiguous and no coordinates for this region have been incorporated. The final 3.0-Å resolution model was refined using six translation-libration-screw domains (two per chain) with the program Phenix (Adams et al, 2002) to an R_work=24.7% and R_free=34.4% for all F>0 (Supplementary Table I). Procheck analysis of the coordinates indicates that 98.9% of residues are located in favoured and additionally allowed regions of a Ramachandran plot.

WNV and DENV propagation and purification

Mature and immature WNV was produced and purified from Vero cells as described earlier (Kaufmann et al, 2006; Zhang et al, 2007). For immature DENV, C6/36 Aedes albopictus mosquito cells were grown in MEM supplemented with 10% fetal bovine serum (FBS) following standard cell culture procedures. Confluent cells were infected with DENV-2 at a multiplicity of infection of 0.2 in the presence of 5% FBS. Medium was replaced 24 h after infection with MEM containing 20 mM NH₄Cl. Cell culture supernatant was harvested 3 days after infection and virus was purified as described for WNV.

Complex formation, cryoEM and 3D image reconstructions

Purified WNV particles were incubated with E53 Fab in the presence of 100 mM NaCl at 4°C overnight, using a ratio of about five Fab fragments per E protein. Purified immature DENV particles were incubated with Fab at 37°C for 30 min and then at 4°C for 2 h, using a ratio of about two Fab fragments per E protein. Micrographs of the frozen-hydrated sample were recorded on Kodak (Rochester, NY) SO-163 films with a CM300 FEG transmission electron microscope (Philips, Eindhoven, The Netherlands). Images were taken at a nominal magnification of × 47 000 and a total electron dose of 12–15 e⁻/Ų. The cryoEM micrographs were digitized on a Nikon 9000 scanner (Tokyo, Japan) with a 6.35-μm step size, and subsequently sets of four pixels were averaged to sample the specimen at 2.69 Å intervals. The program RobEM (Baker, 2004) was used to select a total of 4143 particles from 84 micrographs for the immature WNV–E53 Fab complex and a total of 2741 particles from 23 micrographs for the complex of immature DENV with E53 Fab. The defocus level was determined by fitting the theoretical microscope contrast transfer functions (CTFs) to the incoherent sum of the Fourier transforms of all particle images from each micrograph. The 3D reconstruction was computed using CTF phase-corrected images. The reconstruction was initiated by using a cryoEM density map of immature WNV as a model. The particle orientations were determined with SPIDER (Frank et al, 1996), and the 3D electron density map was calculated with a modified version of XMIPP (Sorzano et al, 2004) assuming icosahedral symmetry. Only 3927 and 2741 particles of the WNV and DENV complex, respectively, were selected to calculate the final 3D electron density maps. Selection was based on correlation with the model projections and stability of the particle centre position used. The resolution of the resultant map was estimated by comparing structure factors for the virus shell computed from two independent half-data sets. The estimated resolution was based on determining the spacing frequency at which the correlation between the two independent data sets became less than 0.5.

One measure of the map quality is the resolution of the lipid leaflets. The above procedure did not give a good representation of the lipid bilayer in the immature WNV–E53 Fab reconstruction (Supplementary Figure 1). Thus, as an alternative reconstruction technique, the Polar Fourier Transform (PFT) (Baker and Cheng, 1996) reciprocal space procedure was used for both WNV and DENV. This gave considerably better representations of the membrane region of these viruses, but the quality of the density representing the glycoprotein was reduced. This might suggest that the PFT method is the better procedure indicating that the interpretation of the Fab density in the XMIPP reconstruction could be inaccurate. However, the excellent agreement of the cryoEM density of the E53 Fab–virus complex with the crystal structure of the soluble E ectodomain in complex with E53 Fab showed that the reconstruction based on the modified XMIPP procedure was accurate. Thus, the lack of a clear separation of the lipid bilayers in the membrane of the reconstructions is not the result of disorder in the lipid due to the E53 binding, but a limitation of the reconstruction techniques.

Fitting of X-ray coordinates into the cryoEM electron density

The fitting of the E and E53 Fab atomic coordinates to the immature virusFab complexes followed earlier procedures (Kaufmann et al, 2006) using the program EMfit (Rossmann et al, 2001). First, the E glycoprotein (PDB codes 3C6D (Li et al, 2008) and 2OF6 (Zhang et al, 2007) of DENV and WNV, respectively) and pr (PDB code 3C5X (Li et al, 2008) for DENV and a model for the pr of WNV generated by SWISS-MODEL (Arnold et al, 2006)) were fitted independently into the cryoEM density (Supplementary Table III). Each E molecule was divided into two rigid bodies, DI+DIII (residues 1–51+133–196+281–400 for WNV and residues 1–49+132–192+280–395 for DENV) and DII+pr. The position and orientation of each rigid body was determined one at a time by making a complete 3D angular search, followed by rotational and translational refinement. The density corresponding to the best fit was set to zero before fitting the next rigid body. Domain DI+DIII was fitted by using a maximum distance restraint of 10 Å between the termini of strands connecting DI and DII. The results showed that the general arrangement of the E ectodomain and pr in the virus–E53 Fab complexes did not significantly change in comparison to the structure of the unliganded immature virus.

Similarly, the atomic coordinates of E53 Fab, extracted from the crystal structure of the E53 Fab–WNV E protein complex, were fitted into the cryoEM density of the complex. For independent fitting, the Fab was divided into two rigid bodies, the variable domain (residues 1–121 chain H and 1–107 chain L) and the constant domain (residues 122–221 chain H and 108–213 chain L). Although the variable domain fitted well into the density associated with molecule A (Figure 1A), the constant domain required some adjustment, accounting for a small change of the elbow angle from 144° to 153° compared with the crystal structure (Figure 4). Furthermore, a good fit of the E53 variable domain into the cryoEM density was obtained when DII and the Fab variable domain of the crystal structure were fitted as one rigid body, verifying the conservation of the mode of binding between the E protein and the variable domain of E53.

However, fitting of the E53 Fab variable domain into the difference density adjacent to the B molecule fusion loop showed that the E53 variable domain had rotated by about 35° relative to the E protein in comparison with the crystal structure. When the crystal structure of the E53 Fab–E complex was superpositioned onto the B molecule in the EM map there were clashes between the three-fold symmetry-related Fab molecules. Thus, the immature WNV structure, the complex of E53 Fab with the B molecule, is slightly different than that observed in the crystal structure and with molecule A. Moreover, to avoid steric hindrance, the elbow angle of the Fab associated with the B molecule changed from 144° to 171°.

A rough measure of the Fab occupancies was obtained by comparing the average height of the densities (‘sumf' (Rossmann et al, 2001)) between the DII+pr and the Fab variable domain. The height of density associated with the Fab molecules was lower compared with the density of DII of E (73 and 57% of the variable domain density for molecules A and B, respectively). This is most likely due to incomplete occupancy of the binding sites by the Fab molecules. The Fab bound to the B molecule has a lower occupancy compared with molecule A, probably as a result of steric hindrance. The height of the density for the more distant Fab constant domains is even further diminished (61 and 83% of the variable domain density in the A and B molecules, respectively). This suggests that either the elbow angle in the Fab molecule is somewhat flexible.

The elbow angle between Fab variable and constant domains was calculated using the program Elbow Angle Calculation (http://proteinmodel.org/AS2TS/RBOW/index.html).

Data deposition

Atomic coordinates and structure factors for the reported crystal structure have been deposited with the Protein Data Bank under accession code 3I50 (rcsb053971). The cryoEM density maps were deposited in the EM databank under accession number EMD-5103 (immature WNV–E53 Fab) and EMD-5102 (immature DENV–E53 Fab). The fitted complex structures were deposited in the Protein Data Bank under accession codes 3IXX (for WNV–E53 Fab) and 3IXY (for DENV–E53 Fab).

We are grateful to Sheryl Kelly and Carol J Greski for help in the preparation of the paper. The work was supported by the NIH grants RO1-AI76331 (MGR), R01-AI073755 (MSD), U01-AI061373 (MSD), and PDVI grants TR-17 (RJK and MGR) and DR-5 (DHF). We are grateful to the Keck Foundation grant for the purchase of the CM300 field emission gun electron microscope used in this study. MVC processed and analysed cryoEM data, fitted X-ray structures in cryoEM maps and wrote the paper. BK prepared Fab complexes of mature and immature WNV. SL prepared Fab complex with immature DENV. GEN, BRC, CAN, JTW and DHF produced the WNV E–Fab complex crystals and performed the crystallographic analysis. VAK provided computational support for cryoEM reconstruction. HAH and PRC collected cryoEM data. RJK was involved in study design. MSD produced E53 Fab for cryoEM. MGR was involved in study design, data analysis and paper writing. MVC, BK, SL, RJK, MSD, MGR and DHF discussed and analysed the results and edited the paper.