Structural Basis for Differential Neutralization of Ebolaviruses

There are five antigenically distinct ebolaviruses that cause hemorrhagic fever in humans or non-human primates (Ebola virus, Sudan virus, Reston virus, Taï Forest virus, and Bundibugyo virus). The small handful of antibodies known to neutralize the ebolaviruses bind to the surface glycoprotein termed GP1,2. Curiously, some antibodies against them are known to neutralize in vitro but not protect in vivo, whereas other antibodies are known to protect animal models in vivo, but not neutralize in vitro. A detailed understanding of what constitutes a neutralizing and/or protective antibody response is critical for development of novel therapeutic strategies. Here, we show that paradoxically, a lower affinity antibody with restricted access to its epitope confers better neutralization than a higher affinity antibody against a similar epitope, suggesting that either subtle differences in epitope, or different characteristics of the GP1,2 molecules themselves, confer differential neutralization susceptibility. Here, we also report the crystal structure of trimeric, prefusion GP1,2 from the original 1976 Boniface variant of Sudan virus complexed with 16F6, the first antibody known to neutralize Sudan virus, and compare the structure to that of Sudan virus, variant Gulu. We discuss new structural details of the GP1-GP2 clamp, thermal motion of various regions in GP1,2 across the two viruses visualized, details of differential interaction of the crystallized neutralizing antibodies, and their relevance for virus neutralization.


Introduction
Ebolaviruses are filamentous viruses with a negative-sense RNA genome that cause severe hemorrhagic fever with high lethality. There are five different ebolaviruses, each named after the location of the outbreak in which it was first discovered: Ebola virus (EBOV; formerly known as Zaire ebolavirus), Sudan virus (SUDV), Reston virus (RESTV), Taï Forest virus (TAFV), and Bundibugyo virus (BDBV) [1,2]. These virions differ by 40-50% in primary amino acid sequence and are antigenically distinct. Sudan virus was the first ebolavirus to be discovered when it caused an outbreak of hemorrhagic fever among cotton factory workers in Nzara, Sudan, in 1976 [3]. Travel of these patients and their caretakers spread the disease to other villages including Maridi, where in the hospital at which several of the patients were treated, the doctor-in-charge and 61 of the 154 nursing staff became infected [3]. The variant of Sudan virus linked to this outbreak is termed Boniface (or SUDV-Bon) after the index case. This emergence of SUDV-Bon ultimately led to 284 cases of disease and 151 deaths.
A separate variant, termed Gulu (or SUDV-Gul), is a modern variant linked to a disease outbreak in the Gulu municipality of Uganda in 2000 [4]. The Gulu outbreak was the largest outbreak of Ebola virus disease yet recorded, with 425 cases and 224 deaths and a large number of cases among children and adolescents. The glycoprotein (GP 1,2 ) of SUDV-Bon and SUDV-Gul differ in sequence by ~5% and an ideal diagnostic or immunotherapeutic for Sudan virus would be able to recognize and neutralize both the Boniface and Gulu variants. An even better diagnostic or therapeutic would be able to recognize all ebolaviruses, taking advantage of structural information of the distinct GP 1,2 glycoproteins of each virus.
The trimeric, membrane-attached glycoprotein GP 1,2 is the only virally encoded protein present on the virus surface and is critical for virus adhesion, entry and internalization into target cells. In infected cells, GP 1,2 is cleaved by furin to yield two subunits, termed GP 1 and GP 2 , that remain linked by a disulfide bond. GP 1 is responsible for recognition and engagement of new target cells, while GP 2 drives fusion of the viral membrane with the endosomal membrane of the target cell. Antibodies against EBOV GP 1,2 have been tested for protection against EBOV challenge in various models with mixed outcomes [5][6][7][8][9][10]; although recent studies are quite promising [11].
A crystal structure of trimeric GP 1,2 from Ebola virus variant Mayinga (EBOV-May) in its prefusion form (PDB code 3CSY) [12] revealed that the three GP 1 subunits are gathered and tethered into a bowl/chalice shape in the viral surface trimer by the three GP 2 subunits that wrap around the GP 1 s. In crystal structures of the post-fusion six-helix bundle conformation of GP 2 , heptad repeat 1 (HR1) adopts a single long helix [13][14][15]. However, in the prefusion structure, the hydrophobic fusion peptides of GP 2 (residues 529-535) wrap around the outside of the trimer, packing against the GP 1 subunits of adjacent monomers, and only the first of two heptad repeats in GP 2 is visible. This single helix is broken into four separate sections in the prefusion structure: HR1 A-D . The crystal structure also illustrated that within GP 1 , putative receptor-binding regions are somewhat sequestered inside the bowl of the chalice at the top of GP 1,2 . A subsequent crystal structure of SUDV-Gul GP 1,2 (PDB code 3S88) [16] illustrates that its GP 1,2 has a similar fold but curiously, different electrostatic characteristics than EBOV-May GP 1,2 . EBOV-May GP 1,2 is neutral/basic while SUDV-Gul GP 1,2 is acidic at the base where the GP 1 and GP 2 subunits meet. In addition, the ectodomain of SUDV GP 1,2 is more susceptible to proteolysis than EBOV GP 1,2 . Treatment of SUDV GP 1,2 with either thermolysin or cathepsin L/B results in significant degradation of GP 1,2 [17]. It is possible that the different electrostatics may contribute to the differing behavior of the GP 1,2 in the endosome [16]. Recent studies also show that the GP 1,2 of filoviruses exhibit distinct protease preferences. The GP 1,2 of EBOV is strongly dependent on the endosomal protease cathepsin B while the GP 1,2 of SUDV is not dependent on the protease [18,19].
Both EBOV-May GP 1,2 and SUDV-Gul GP 1,2 structures were determined in complex with antibodies. EBOV-May GP 1,2 was crystallized in complex with an EBOV-specific human antibody termed KZ52, which was derived from a human survivor of a natural infection in Kikwit, Zaire (EBOV-May), in 1995 [5]. SUDV-Gul GP 1,2 was crystallized in complex with a novel Fab termed 16F6 that was raised by immunization of mice with irradiated SUDV-Bon virions [16]. Unexpectedly, the epitopes of KZ52 and 16F6 overlap, at the base of the GP 1,2 spike where GP 2 meets GP 1 [16], suggesting that the two antibodies may function in the same way and that the shared site may be a hot spot for neutralization of ebolaviruses.
Here we report new molecular details of neutralization of the different GP 1,2 s by respective antibodies. Surface Plasmon Resonance and neutralization assays reveal that paradoxically, the antibody with lower affinity for recombinant GP 1,2 , with hindered access to GP 1,2 epitope, has better neutralizing capacity. Here, we also report a new crystal structure, to 3.35 Å resolution, of SUDV-Bon GP 1,2 from the historical 1976 Boniface variant, in complex with mAb 16F6. This SUDV-Bon GP 1,2 structure allows better visualization of key contacts between GP 1 and GP 2 and the "chain reversal regions" of GP 2 involved in conformational changes during fusion (a step likely blocked by both 16F6 and KZ52) [20]. Comparison of the now three available structures of ebolavirus GP 1,2 s (SUDV-Bon, SUDV-Gul and EBOV-May) provides new insights into particular components of the two neutralizing epitopes and the puzzle of what constitutes an effective neutralizing antibody response against ebolaviruses.

Crystal Structure of Sudan Virus Boniface GP 1,2
The crystal structure of SUDV-Bon GP 1,2 bound to antibody 16F6 was determined to a resolution of 3.35 Å using molecular replacement. The crystallographic asymmetric unit contains one GP 1,2 monomer bound to one 16F6 Fab fragment. The final model contains residues 32-191, 213-285, 300-311 and 510-614 of GP 1,2 and residues 1-220 and 1-212 of the heavy and light chains of 16F6, respectively. A disulfide bond between C53 and C609 covalently links the GP 1 and GP 2 subunits together, and is only visible in structures of SUDV, but not EBOV GP 1,2 . In all three ebolavirus GP 1,2 structures, GP 1 and GP 2 bury a surface area of ~2400 Å 2 on each other and are held together by numerous hydrogen bonds and non-bonded contacts. Although GP 1,2 was treated with peptide-Nglycosidase F (PNGase F), clear electron density is still observed for the first two monosaccharides of glycans attached to N257 (NAG 850) in GP 1 and N563 (NAG 901) in GP 2 indicating that these sites (in both SUDV and EBOV GP 1,2 ) are resistant to PNGaseF digestion ( Figure 1). The trimeric form of the GP 1,2 is obtained by application of 3-fold crystal symmetry, and has an approximate dimension of 95 Å  95 Å  90 Å. Four lysines (95, 114, 115, and 140) in the putative receptor-binding head of GP 1 have been identified as critical for attachment of EBOV to target cells [21]. Three of these (114, 115, and 140) are strictly conserved among all known ebolaviruses, while position 95 is occupied by glutamine in some variants of SUDV (but is lysine in the SUDV-Bon and SUDV-Gul sequences crystallized). In all three ebolavirus GP 1,2 structures, K95 is buried deep in the chalice bowl of GP 1 and projects into GP 2 rather than out toward the target cell and receptor ( Figure 1). Inside GP 1,2 , K95 forms a hydrogen bond with the main-chain carbonyl of T576 and forms van der Waals interactions with L573, which is located in the loop between HR1 B and HR1 C of GP 2 . Hence the crystal structures support biochemical predictions [21] that the observed functional importance of K95 may not be in direct receptor engagement, but rather in proper maintenance or springing of the prefusion GP 1,2 assembly. Alternatively, K95 may become better exposed for interaction with host factor(s) in the endosome if any conformational changes occur as a result of receptor binding. K115 also points down toward the viral membrane, but is solvent exposed and could potentially reorient upon receptor engagement. K114 and K140 project into solvent towards the target cell and may serve as direct contacts for receptor.
In the prefusion conformation of ebolavirus GP 1,2 , the first heptad repeat of GP 2 is broken into four structural segments (HR1 A-D ) that wrap around GP 1 . We observe in all three structures (SUDV-Gul, SUDV-Bon, and EBOV-May) that residues R580-T581 of HR1 C form a short  strand that assembles into a continuous  sheet with the K95-S99  strand of the GP 1 receptor-binding head. These residues are completely conserved among all ebolaviruses and marburgviruses.
Other structural details not previously reported include the noncovalent assembly by which GP 1 is anchored to GP 2 . Five  strands of the GP 1 base and two antiparallel  strands of the GP 2 internal fusion loop combine to form a continuous seven-stranded twisted  sheet. The GP 1 component of that sheet and a second small  sheet of the GP 1 base together assemble a semicircular GP 1 horseshoe surface that clamps the first heptad repeat of GP 2 in its metastable, prefusion structure (Figure 2 A,B). In order for GP 2 to rearrange into its post-fusion six-helix bundle structure [13,15] both the continuous GP 1 -GP 2  sheet and the horseshoe clamp must be broken. The C-terminal heptad repeat 2 is disordered in all three SUDV-Bon, SUDV-Gul and EBOV-May structures, which may result from the functional mobility of this region as well as the lack of the transmembrane regions that tether GP 1,2 on the viral surface.

Comparison of SUDV-Bon and SUDV-Gul Glycoproteins
The Boniface and Gulu variants of Sudan GP 1,2 differ in sequence by ~ 5%. The clones of SUDV-Bon and SUDV-Gulu

Implications for fusion
A CX 6 CC motif in GP 2 that is conserved among all ebolaviruses is visible in both structures of SUDV GP 1,2 . This motif contains the disulfide bond that anchors GP 1 to GP 2 (C53-C609) and a second disulfide bond within GP 2 (C601-C608). The glycoproteins of some retroviruses have a CXXC motif within the receptor-binding subunit that isomerizes to release the disulfide bond between the receptorbinding and fusion subunits [20]. Filoviruses, however, lack such a motif in GP 1 , and it is currently unknown if the GP 1 -GP 2 disulfide link, which is encoded by a different motif, is retained or released during fusion. It has been speculated that endosomal thiolreductases could reduce the C53-C609 GP 1 -GP 2 bond [22], and mild reduction of the enzymatically processed 19 kDa GP 1,2 does confer binding to liposomes (an ability presumably attained via conformational change, as unreduced 19 kDa GP 1,2 does not bind liposomes) [23]. The neighboring, C601-C608 intra-GP 2 disulfide bond however, is observed in pre-fusion as well as post-fusion crystal structures and thus, probably remains intact during fusion. Given the close proximity of both the GP 1 -GP 2 and GP 2 -GP 2 disulfide bonds, reduction is either very specific for the GP 1 -GP 2 disulfide bond or plays no role during fusion. Separate possibilities are that GP 1 simply rotates out of the way, or that continual enzymatic processing of GP 1 by cathepsins digests enough of GP 1 to remove steric hindrance to GP 2 conformational rearrangement [22].

Figure 2.
Secondary structure interactions between SUDV-Bon GP 1 (colored blue and purple) and GP 2 (colored light grey). (A) Close up of the continuous seven-stranded twisted  sheet formed by five  strands of the GP 1 base (2, 1, 13, 3, and 6 in blue) and two antiparallel  strands of the GP 2 internal fusion loop (19 and 20 in grey). (B) The first heptad repeat of GP 2 (which contains the grey coils labeled HR1A and HR1B and the loop afterwards) is clamped by a horseshoe surface formed by five  strands of GP 1 base (, a second small  sheet of the GP 1 base (andand a glycan (pink ball-and-stick). To orient the reader, key residues are labeled.
A recent study by Delos et al. has revealed residues in the "chain reversal region" of GP 2 critical for fusion of Ebola virions [20]. This region comprises a short stretch of hydrophobic residues, a glycine-glycine pair, a CX 6 CC motif, and a bulky hydrophobic residue after the motif. Mutational analysis in EBOV GP 1,2 has shown that residues R587, F592, W597, G598, H602, and I610 are critical for viral infectivity. All the above residues except H602 (which is R in SUDV and RESTV) are conserved in SUDV. The better ordered structures of SUDV provide us with newer insights into the role of the conserved residues in the prefusion conformation, and are reported here. Interactions made by the side chains of these residues in the prefusion-form in SUDV and the post-fusion form in analogous EBOV (strain Mayinga, PDB code 1EBO) are shown in Table 1. In the pre-fusion state, the conserved residues make interactions with GP 1 (hydrogen bonding to T60 and stacking interaction against L57 and I185 ( Figure 3A); the  denotes residues from a 3-fold related monomer). W597 is involved in a stacking interaction with other W597 residues from two 3-fold related monomers, suggesting its role in stabilizing the trimeric form in the heptad repeat region. Definitive density was not seen for the side chains of R602 and I610 in either SUDV-Bon or SUDV-Gul GP 1,2 and we could not assign any interactions of these residues; positions 602 and 610 are thus modeled as alanine. However, in the post-fusion state, the conserved residues make interactions solely with residues in GP 2 ( Figure 3B). In addition, the conserved residues make interactions with different residues in the pre-fusion and post-fusion forms suggesting a greater role of these residues during the fusion process. Prefusion SUDV-Bon is used here as it is better ordered than prefusion EBOV-May. Interacting residues are shown in ball and stick. In (A), different monomers of GP 1 are colored blue and purple and the three copies of GP 2 are all colored light grey. In (B), the three copies of GP 2 are colored different shades of grey. Equivalent residues in the 3-fold related protomers are labeled with ´ and ´´ respectively. Hydrogen bonds are shown as red dashed lines. The residue R609* in the postfusion form is an engineered mutation to replace the cysteine residue (C609) in the native protein that is involved in a disulfide bond with C53 of GP 1 .

Interactions between 16F6 and SUDV GP
The complementarity determining regions (CDRs) H1 and H3 of 16F6 form a network of hydrogen bonds, van der Waals interactions and one salt bridge to the GP 1 base. CDR L2 also hydrogen bonds to the GP 1 base and forms additional hydrophobic interactions to the stem region of the internal fusion loop of GP 2 (Figure 4). Specific interactions between 16F6 and the glycoprotein have not been previously reported and are shown in Table 2. The heavy chain and light chain of 16F6 bury a surface area of ~1630 Å 2 between them. The antibody 16F6 interacts with GP 1,2 primarily using its heavy chain, burying an area of ~350 Å 2 with GP 1 and ~200 Å 2 with GP 2 . The interface between GP 1,2 and 16F6 is predominantly hydrophobic with the exception of four hydrogen bonds. Residues at the interface of SUDV-Bon GP 1,2 and 16F6 (cutoff distance of 3.5 Å). GP 1 is colored purple, GP 2 is colored white, the 16F6 heavy chain is colored orange and the light chain is colored pale yellow. Hydrogen bonds are shown as red dashed lines.

Thermal Motion in GP
Comparison of B-factor values (an atomic displacement parameter arising from thermal vibration of atoms and static disorder of atoms in different unit cells of the protein crystal) of key portions of GP 1 and GP 2 in SUDV GP 1,2 reveals that motion predominates in the glycan cap regions, the C-terminal half of the fusion loop, and the visible C-terminal regions of GP 2 ( Figure 5). How does SUDV compare to EBOV GP 1,2 in this regard? Deuterium Exchange Mass Spectrometry (DXMS) reveals that although GP 1 of SUDV and EBOV exhibit nearly identical rates of exchange of amide hydrogens with solvent deuterium, all regions of GP 2 of SUDV, including the fusion loop, heptad repeats, disulfidecontaining linker and C-terminal regions, are fundamentally more mobile than those of EBOV GP 1,2 [17] (Figure 6). Interestingly, the disulfide-containing linker regions of GP 2 are only visible in crystals of SUDV GP 1,2 , not EBOV GP 1,2 . The unique crystal packing environment of the SUDV I23 unit cell and the acute angle made by the bound 16F6 antibody may have constrained this region. Another difference in mobility between SUDV and EBOV GP 1,2 relevant for antibody binding is the N terminus of GP 2 , which is released from GP 1 by furin cleavage. In EBOV GP 1,2 , this peptide is hydrogen bonded to the GP 1,2 core, is bound between heavy and light chains in the central region of the KZ52 paratope and forms a significant portion of its epitope. By contrast, in the SUDV structure, the same peptide is disordered and is not bound by 16F6. Instead, 16F6 binds the underlying GP 1,2 core beneath the peptide (Figure 7).
One explanation could be that the GP 2 N terminus is simply tacked down in the EBOV GP 1,2 crystal structure by KZ52 binding. A more likely explanation, supported by DXMS, is that the mobility of this region fundamentally differs between SUDV and EBOV, even in the absence of antibody binding. Key differences in sequence between SUDV and EBOV at this site may explain this observation. The anchor point of the GP 2 N terminus to the GP 1,2 core is residue G509. At position 509, EBOV contains a proline, which may restrict mobility, while SUDV contains a glycine, which may enhance mobility. Further, residues N506 and Q508 of EBOV use both their terminal oxygen and nitrogen atoms to form a network of hydrogen bonds to the EBOV GP 1,2 core; specifically, to amino acids as well as the attached glycan of heptad repeat 1. By contrast, position 506 is K in SUDV-Gul and Q in SUDV-Bon, and position 508 is T in both SUDV GP 1,2 (Gulu and Boniface). None of those hydrogen bonds made by EBOV are observed for SUDV, and the specific hydrogen bonds made by the side-chain oxygens of N506 and Q508 in EBOV are not possible for the corresponding residues in SUDV. The key differences in sequence and inherent mobility of N-term region of SUDV GP 1,2 reflects in a marginally varied epitope for binding of 16F6. Figure 6. Plots of number of deuterons vs. time. These compare the deuteration levels of key peptides of EBOV GP 1,2 (shown in blue) to their counterparts in SUDV-Gul GP 1,2 (shown in magenta), measured over a time scale of 10-1000 sec using deuterium exchange mass spectrometry [17].

Comparison of Antibody-GP complexes
The crystal structures reveal that the 16F6 and KZ52 antibodies bind at overlapping epitopes that contain residues from both GP 1 and GP 2 . However, the number of interactions made by each antibody with its respective antigen differs (Table 3). 16F6 interacts with SUDV GP 1,2 via 7 hydrogen bonds and 19 non-bonded contacts. KZ52 interacts with EBOV GP 1,2 via 13 hydrogen bonds and 29 nonbonded contacts. Hence, the binding of KZ52 to EBOV GP 1,2 is expected to be stronger than the binding of SUDV GP 1,2 to 16F6. Accordingly, Surface Plasmon Resonance studies reveal that the K d of KZ52 binding to recombinant, soluble EBOV GP 1,2 ectodomain is 6.31 ± 3.70 nM while the binding of 16F6 to recombinant, soluble SUDV GP 1,2 ectodomain is 1.94 ± 1.40 µM to SUDV-Bon, and 1.59 ± 1.0 µM to SUDV-Gul. Further, the angle the antibodies make when binding against the core of the GP 1,2 differs dramatically. KZ52 binds GP 1,2 at an angle more parallel to the viral membrane, while 16F6 is angled 50° downward toward the membrane, such that the C-terminal, constant regions of 16F6 are 50 Å closer to the viral membrane than those of KZ52 (Figure 8). The hinge regions attached to the membrane-proximal end of the 16F6 Fab must therefore adopt acute angles so that the other Fab and Fc of the IgG do not penetrate the viral membrane. It is a bit surprising that the 16F6 IgG can reach down to access the epitope from that angle.  As a result of the higher affinity for the GP 1,2 ectodomain, greater number of contacts, and presumably easier access to its epitope, one might expect KZ52 to neutralize better than 16F6. Surprisingly, however, 16F6 neutralizes SUDV GP 1,2 -bearing VSIV better than KZ52 neutralizes EBOV GP 1,2 -bearing vesicular stomatitis Indiana virus (VSIV) [16] (data replotted here as Figure 9). At all IgG concentrations, a greater portion of virus remains un-neutralized by KZ52 than by 16F6. Even small differences in neutralization capacity may confer significant functional variation in vivo for filoviruses because a single p.f.u. can be a lethal dose for primates.
Our data suggests that binding affinity and access to the epitope alone do not account completely for antibody neutralization efficacy; the composition of the particular epitopes and the susceptibility of the different GP 1,2 , themselves, to be neutralized are also important. For example, although the epitopes overlap, KZ52 is shifted more towards GP 2 , burying mostly GP 2 while 16F6 buries more of GP 1 on SUDV GP 1,2 [16]. Affinity for transmembrane-anchored GP 1,2 could differ somewhat from affinity for the GP 1,2 ectodomain as well, although the epitopes of these antibodies are quite some distance (>50 Å) from the transmembrane anchor.
In addition, the 16F6 and KZ52 Fabs are rotated relative to each other when bound to GP 1,2 . KZ52 binds with its heavy chain "down" toward the membrane, while 16F6 binds with its heavy chain "up" toward GP 1 . In addition, the crystal structures reveal that the heavy chain of 16F6 interacts directly with both heptad repeat 1 of GP 2 (specifically HR1 B ) and the  2 strand of GP 1 . By contrast, KZ52 interacts primarily with the N-terminal peptide of GP 2 and the turn region prior to the HR1 helix. KZ52 is unable to make the same contacts as 16F6 because (1) its heavy and light chain are switched in space relative to those of 16F6 and (2) the N-terminal peptide of GP 2 is nestled into the center of the paratope in EBOV (Figure 7). The N-terminal peptide of GP 2 is probably less important for GP 1,2 function than are the HR1 helix and GP 1 base strand. Hence, an antibody that anchors a less-important terminal peptide might be expected to be less effective than one that anchors a heptad repeat critical for conformational change to its prefusion GP 1 -sheet clamp. Hence, 16F6 binds more weakly, but may bind more effectively. These antibodies are thought to neutralize by blocking conformational changes during fusion. Perhaps better bridging of GP 1 and GP 2 by 16F6 more effectively anchors the pair in a pre-fusion conformation.
Alternatively, key differences between EBOV and SUDV GP 1,2 may render them differently susceptible to neutralization. GP 2 of SUDV is less stably ordered than GP 2 of EBOV. Perhaps 16F6 has recognized a greater portion of GP 1 because its GP 2 is a "moving target". Certainly, the increased mobility of the GP 2 N-terminal peptide in SUDV means that more of underlying GP 1,2 structure is available for antibody interaction, antibody anchoring of the prefusion conformation and greater neutralization. Another difference is that EBOV GP 1,2 has a greater requirement for cathepsin cleavage than does SUDV GP 1,2 [18], and crystal structures reveal that the two GP 1,2 have different electrostatics at their trimer interface. Electrostatics at the trimer interface may influence how the low pH environment of the endosomes determines susceptibility of GP 1,2 to triggering [16].
In summary, we find by structural and functional analysis, that the antibody of higher affinity and presumably easier access to a shared site is not necessarily the most effective at viral neutralization. Other characteristics of the epitope bound, including which amino acids are contained in the individual epitopes and the functional importance of these amino acids, may confer neutralization susceptibility. A conundrum for filoviruses has been that antibody efficacy in vitro and in vivo do not always correlate. Here, we demonstrate that certain in vitro characteristics of antibody binding strength, epitope access, and in vitro neutralization also do not necessarily correlate. In addition to binding tightly, an effective antibody may need to bind correctly.
Development of effective immunotherapeutics against filoviruses will thus require detailed and coupled structural and functional characterization of candidate antibodies. It will also require production and analysis of larger panels of antibodies against the filoviruses than have been currently described so that we may compare several different antibodies within the same competition group, in order to determine what the most effective antibodies are and why this is so. This knowledge base will also clarify what would constitute an ideal antibody response elicited by vaccination, and will aid in development of vaccines designed to elicit these types of antibodies. Side chain disordered Residue disordered No interactions Note that  and  denote residues from the other two monomers related by the three-fold axis  *Interactions were calculated with the PDBsum server and confirmed by visual inspection with COOT. The cutoff distance for non-bonded interactions listed above is 3.5 Å. Table 3. Comparison of the 16F6 and KZ52 epitopes on SUDV and EBOV GP 1,2 , respectively.
Residues at the shared intersection of the two epitopes are colored orange. Residues listed are conserved between SUDV and EBOV, unless specified.

Residues of SUDV GP 1,2 bound by 16F6
Residues of EBOV GP 1,2 bound by KZ52 S32 P34 T39 (note: residue 39 is H in EBOV)  . The ability of 16F6 and KZ52 IgGs to neutralize SUDV GP 1,2 -bearing VSIV or EBOV GP 1,2 -bearing VSIV, plotted in a log (A) and linear (B) scales. VSIV particles were pre-incubated with the indicated concentrations of antibody for one hour at room temperature, and then exposed to Vero cells at 37° C. Viral infectivity was scored at 16 hours post-infection. The starting titers were 910 7 IU/ml for EBOV and 310 7 IU/ml for SUDV on Vero cells [16].

Molecular Biology
Sudan virus Boniface (SUDV-Bon) glycoprotein (GP 1,2 ) was cloned into the pDISPLAY vector (Invitrogen) for expression in mammalian cells with an Ig K-chain leader sequence in place of its natural signal sequence, and a Factor Xa-cleavable N-terminal HA tag for purification. The construct includes residues 33-313 and 473-637, with deletion of the mucin-like domain (residues 314-472) and C-terminal transmembrane domain (residues 638-676). The GP 1,2 gene was amplified by PCR using oligonucleotides containing BglII restriction sites at their 5 ends and SalI restriction sites at the 3 end. The PCR fragment was then cloned into pDISPLAY using BglII and SalI DNA restriction and T4 ligase DNA ligation to produce the IgGK secretion leader-HA-FactorXa-ARS-GPdelta mucin deltaTM plasmid. GP 1,2 is composed of more than 50% oligosaccharides by weight, primarily because of a heavily glycosylated mucin-like domain that is non-essential for cellular attachment or entry [24].

Protein Expression
Recombinant SUDV GP 1,2 was transiently expressed in HEK293T cells by calcium phosphate precipitation in ten-layer CellStacks (Corning). The DNA-calcium phosphate mixture was added to 70% confluent cells grown in DMEM plus 1x penicillin-streptomycin and 5% (v/v) fetal bovine serum. The supernatant was harvested four days post-transfection, concentrated using a Centramate tangential flow system and purified by anti-HA immunoaffinity chromatography. GP 1,2 was natively deglycosylated using Peptide:N-glycosidase F (PNGaseF) in PBS, overnight at room temperature. The extent of the digestion was followed by SDS-PAGE.

Structural Determination of SUDV GP-Fab 16F6 Complex.
For crystallization, deglycosylated SUDV GP 1,2 was mixed with excess Fab 16F6 and incubated for 1 hour at 4° C and subequently purified on a Superdex-200GL 10/300 column equilibrated with 10 mM Tris-HCl, pH 7.5 and 150 mM NaCl. Fractions containing trimeric SUDV GP 1,2 -Fab16F6 complex were concentrated to 10 mg/ml and initial crystallization conditions were screened using a Topaz Fluidigm system at 20° C. Initial crystal hits were obtained in the presence of 20% PEG 3350 and 0.2 M lithium citrate and were translated and optimized by hanging-drop vapor diffusion. Initial crystals grew over a period of one month at room temperature up to a maximum size of 0.2  0.2  0.2 mm 3 and have a cubic morphology. Initial crystals diffracted up to 4.5 Å at the Beamline 8.2.2 of the Advanced Light Source (ALS, Berkeley). Improved crystals were obtained by adding a mixture of 3 L of the SUDV GP 1,2 -Fab16F6 complex at 10 mg/mL in 10 mM Tris HCl pH 7.5 and 0.5 L of 20% benzamidine hydrochloride to 3 L of mother liquor crystallization solution 15% (w/v) PEG 3350, and 0.2 M lithium citrate. These crystals grew in 3-4 weeks at room temperature to a maximum size of 0.2  0.3  0.3 mm 3 and have a star-shaped morphology. The crystals were cryoprotected with 17.5% (v/v) glycerol plus mother liquor before flash cooling in liquid nitrogen. The crystals diffracted beyond 3.2 Å resolution, but due to radiation damage, a complete dataset was collected with good statistics up to 3.35 Å, at ALS beamline 8.3.1. The data were indexed, integrated and scaled using HKL2000 [25]. The SUDV GP 1,2 -Fab16F6 crystals belong to space group I23, with unit cell dimensions of a=b=c=194 Å. The solvent content is approximately 65% [26] with one monomer of the GP 1,2 -Fab complex in the asymmetric unit. Data collection statistics are shown in Table 4.
The structure of SUDV GP 1,2 was determined by molecular replacement using an edited polyalanine model based on the sequence alignment and on the conservation of structure elements between SUDV GP 1,2 and the previously determined EBOV-May GP 1,2 structure [12]. The edited poly-alanine search model comprised residues 62-113, 129-186, 215-260 from GP 1 and 503-595 from GP 2 . An initial molecular replacement solution was found using the first 4.8 Å data set using PHASER [26]. A partial solution for the glycoprotein model was obtained and revealed unambiguous extra electron density for key parts that were absent from the search model. This partial solution was fixed and the Fab position was determined using poly-alanine models of separate variable and constant domains of a different Fab (PDB 1JLP) [27], with CDRs deleted. Interpretable electron density maps with clear secondary structural elements and solvent boundaries were obtained, with the biologically relevant trimer generated by the crystallographic three-fold axis that extends through the diagonal vertices of the cubic spacegroup I23. This initial solution was assembled into a complete search model and used with the subsequent 3.35 Å data set in PHASER (LL-gain=604, TFZ=33.8, RFZ=10.4). After an initial step of rigid body refinement, the structure was refined with CNS [28] using torsion-angle simulatedannealing refinement with a maximum-likelihood amplitude target. The initial model was iteratively improved by visual inspection and model building using COOT [29] and refinement in CNS [28]. In the final rounds of refinement, riding hydrogens, grouped atomic displacement and TLS parameters were implemented using PHENIX [30]. The structure was refined to final R work and R free values of 21.98% and 28.44% and contains SUDV-Bon GP 1,2 residues 32-191, 213-285, 300-311 and 510-614, and Fab 16F6 residues 1-220 (heavy chain) and 1-212 (light chain). Electron density was missing for GP 1,2 residues 192-212, 286-299, 312-313, 473-509 and 615-637. Weak or discontinuous electron density is observed in the outer regions of the glycan cap (residues 261-285 and 300-307) and these regions were tentatively assigned as poly-alanine fragments. Despite the treatment of the GP 1,2 with PNGaseF, two glycan chains resistant to deglycosylation at N257 on GP 1 and N563 on GP 2 were preserved. These glycans show a high B-factor (B average = 242.5 Å 2 ), but there is unambiguous electron density for the chitobiose core and up to the first mannose residues of these glycans. Final refinement statistics and analysis are shown in Table 4. Atomic coordinates and structure factors are deposited in the RCSB Protein Data Bank under accession number 3VE0. The biologically relevant trimer was generated by the crystallographic three-fold axis that extends through the diagonal vertices of the cubic spacegroup I23.

Antibody binding studies
The binding affinity of KZ52 to EBOV-May GP 1,2 and 16F6 to SUDV-Gul and -Bon GP 1,2 was measured using BIAcore-2000 Surface Plasmon Resonance spectrophotomer. All the samples were buffer exchanged to HBS-EP buffer (10 mM Hepes, pH 7.4, 150 mM Sodium Chloride, 3 mM EDTA, and 0.005% Polysorbate 20) prior to complexation. KZ52 and 16F6 were immobilized at 25 °C on various channels of carboxymethylated dextran CM5 sensor chips to ~200 response units using standard amine coupling protocol. EBOV GP 1,2 , SUDV-Bon GP 1,2 and SUDV-Gul GP 1,2 were injected over channels coated with KZ52 and 16F6 respectively. The glycoproteins were injected sequentially at increasing concentrations of 1, 2.5, 5, 10 and 20 µM respectively at a flow rate of 10 µL/min for 250 sec and dissociation of the complexes were measured over a period of ~600 sec. The chromatogram data were analyzed using a 1:1 Langmuir binding model to determine the k a (on), k d (off) and K d (dissociation) values using the BIAevaluation software.

Antibody Neutralization Studies
VSIV pseudotypes bearing EBOV or SUDV GP and expressing eGFP were generated and concentrated by ultracentrifugation as described previously [31,32]. Virus-antibody incubations were carried out as follows: Concentrated virus (1-3 µL, ~110 5 IU) was mixed with PBS containing antibody at the indicated concentration (0-40 µg/mL final; total reaction volume, 20 µL). Reactions were incubated at room temperature for 1 hour, and viral infectivity was determined by titration on Vero cell monolayers. Briefly, Vero cells were exposed to samples for 1 hour at 37° C and infectivity was scored at 16 hours post-infection by manually enumerating GFP-positive cells, as described previously [32].

Deuterium Exchange Mass Spectrometry Experiments
A detailed description of DXMS methodology was described previously [17]. Briefly, quench conditions that produced an optimal pepsin fragmentation pattern were established for both EBOV and SUDV GP1,2. Then, the non-deuterated samples, functional deuterated samples and the equilibriumdeuterated back-exchange control samples were prepared by mixing the GP 1,2 proteins with nondeuterated (8.3 mM Tris, 150 mM NaCl, in H 2 O, pH 7.15), deuterated (8.3 mM Tris, 150 mM NaCl, in D 2 O, pD READ 7.2) and equilibrium-deuterated (1% formic acid in 99.9% D 2 O) buffers, separately. Later, optimized quench buffer was added to stop the reactions. Finally, the samples were passed over pepsin column, and the resulting peptides were collected on a C18 trap and separated using a C18 reversed phase column. Data were acquired in both data-dependent MS/MS mode and MS1 profile mode by a LCQ mass spectrometer, and the data analyzed by SEQUEST (Thermo Finnigan Inc.) and DXMS explorer (Sierra Analytics Inc., Modesto, CA).

Conclusions
We report the crystal structure of SUDV-Bon GP 1,2 bound to 16F6 and describe new structural details of the complex including thermal motion of GP 2 , approach angle and access to the antibody epitope and molecular interactions between antibody and GP 1,2 . Surface Plasmon Resonance studies, also presented here, reveal that 16F6 binds SUDV-Gul/Bon GP 1,2 with a lower affinity than KZ52 binding to EBOV-May GP 1,2 . Paradoxically, however, 16F6 confers better neutralization than KZ52. Together, these studies suggest that the key to effective neutralization might lie in binding the epitope correctly rather than in affinity alone.