Both VP40(31–212) and VP40(31–194) form ring-like oligomeric structures in solution and were crystallized in space groups P6₂22 and P422, respectively. The P6₂22 crystals contained the full-length construct VP40(31–212), while the P422 crystals were only obtained upon proteolysis at residue 54 in the crystallization drop, as confirmed by SDS-PAGE analysis and N-terminal sequencing of dissolved crystals. The structure of the P422 crystal form, thus containing VP40(55–194), was solved by a combination of single isomorphous replacement (SIR) and multiwavelength anomalous diffraction (MAD) methods. The final model contains residues 69 to 192 and a short stretch of very well-defined single-stranded RNA, containing the sequence 5′-U-G-A-3′ (phasing and refinement statistics are presented in Table 1).Table 1

Crystallographic Statistics

The P6₂22 crystal form [VP40(31–212)], which contains two molecules per asymmetric unit, was subsequently solved by Patterson search techniques. The chain traces in both crystal forms are well superimposable, with only slight rms deviations for the Cα atoms of both molecules (0.37 Å for molecule A and 0.42 Å for molecule B). Accordingly, only the structure as present in the tetragonal crystals will be discussed. The P6₂22 model contains residues 69 to 190 (or 191 in molecule B) and two single-stranded triribonucleotides as found in the P422 structure. Residues 31 to 68 and 189(190) to 212 were disordered in the crystal and could not be assigned (for refinement statistics, see Table 1).

Both NTD protein variants analyzed here form toroidal octameric rings (assembled by crystallographic symmetry) with a wide central pore (Figure 1) . The ring structures are composed of four dimeric antiparallel β sandwich domains (Figure 1A). Each domain contains six antiparallel β strands forming two sheets and three α helices that pack laterally against the β structure. The ring structure has an outer diameter of ∼84 Å and an inner diameter of ∼17 Å (Figure 1B); the width of the particle is ∼42 Å (Figures 1C). In each monomer, the C-terminal ends (residues 184 to 194) reach over to the neighboring monomer, which contacts them in an extended conformation. This arrangement leads to the positioning of the C-terminal ends at the outer side of the ring, pointing away from it, almost at the same horizontal level and only 15 Å apart (Figure 1C). This suggests that the C-terminal domains, which were not present in the construct, could be positioned at the side, above or below the ring. The structure lacks ∼70 residues at the N terminus and its N-terminal end is positioned at the outside of the ring, close to the C-terminal end (Pro 185) (Figure 1B). Each monomer binds a short stretch of RNA at the dimer-dimer interface, characterized by excellent electron density for a triribonucleotide (Figure 1D).Figure 1

The N-Terminal Domain of Ebola Virus VP40 Forms Octameric Ring-like Structures Mediated by Specific ssRNA Binding

(A) Ribbon drawing of the antiparallel dimer formed by the N-terminal domain of VP40. The two monomers are shown in different colors and the secondary structure elements are labeled [9]. Note that the C-terminal end (residues 189 to 194) interacts in an extended conformation with the outer β sheet of its neighbor.

(B and C) Ribbon drawing of the ring structure of VP40. (B) View from the top, and (C) view from the side. The RNA molecule, bound at the dimer-dimer interface, is shown as an all-atom model. Each dimer (such as the one shown in A) is drawn in a different color and the dimensions as well as the positions of the N and C termini are indicated (the RNA has been omitted in C).

(D) Stereo view of a σ_A-weighted electron density omit map (mF_obs − DF_calc) contoured at 2σ showing the superimposed final model for the triribonucleotide with the sequence U-G-A as an all-atom model and with a Cα tracing of the adjacent NTD molecule.

The interface of the antiparallel dimer occupies 1250 Ų. The dimer is stabilized by salt bridges from Glu160A (molecule A) to Arg148B (molecule B) and Arg151B, respectively; further polar interactions are found between Trp95A and Gln184B (Figure 2A) . The effect of the interactions is 2-fold due to the antiparallel nature of the dimer. The stability of the dimer is further enhanced by hydrophobic core interactions (residues Trp95, Pro97, Phe161, Ile74, and Ile82 from molecules A and B) (Figure 2B) as well as residues 189 to 194 (A and B), which interact in an extended conformation on the outside of the ring with a neighboring molecule (Figure 1A). All polar residues except Glu160, which is substituted by Asn, and all of the hydrophobic residues involved in the stabilization of the interface are conserved between VP40 sequences from Ebola and Marburg viruses [9]. The interface of the antiparallel dimer coincides with the interface occupied by the N- and C-terminal domains in the closed monomeric conformation, and residues covered by both interfaces are approximately the same [9]. This indicates that the C-terminal domain has to move away to allow the formation of the NTD dimer as suggested [11].Figure 2

Close-Up View of the Molecular Interactions at the Interface of the Antiparallel Dimer

The view is along the two-fold axis similar to Figure 1A (rotated by 180°). Panel (A) shows hydrophilic interactions and panel (B) focuses on the hydrophobic core in the center of the two-fold axis. Secondary structure elements are labeled accordingly. Note that most of the residues involved in the specific contacts are conserved between Ebola and Marburg virus VP40 (see Figure 3 in [9]).

The dimer-dimer interactions bury a surface of 990 Ų (involving molecules A and C). This is dominated by hydrophobic interactions complemented with polar main chain contacts, including hydrogen bonds between the amide of Gly141A and the carbonyl of Tyr171C, and the oxygen of Thr173A and the amide of Gly139C. In addition, residues from the loop structure connecting β strand 4 and α helix 3 are sandwiched between two loop structures from a neighboring molecule, namely one connecting β strands 1 and 2 and the other one bridging α helix 4 with β strand 6. This creates a weak hydrophobic core on both corresponding ends of the β sandwich structures. The formation of the dimer-dimer interface also generates the binding pocket for the specific recognition of a single-stranded ribonucleotide segment, which dominates the interface (Figure 3A) .Figure 3

RNA Conformation and RNA Protein Interactions

(A) Two RNA molecules bind at the molecular interface generated by the dimer-dimer formation. The two protomers are shown in yellow and light yellow and the triribonucleotide is drawn as an all-atom model.

(B) Stereo view of two complete symmetry-related triribonucleotides containing the sequence 5′-U-G-A-3′, bound at the dimer-dimer interface. The two molecules are in white and in light gray with labels containing one asterisk. In addition, the right panel includes also a uridine (label underlined) from a neighboring RNA molecule, interacting specifically with Uri1* (for clarity, only Uri1 of the triribonucleotide is shown). The hydrogen bonding distances within the triribonucleotides are indicated as dashed lines. Note the parallel base pair stacking by the central guanine (Gua2 and Gua2*) as well as the parallel orientation of two symmetry-related uracil bases (Uri1* and Uri1). Both riboses from the uridine and the guanine are in the C₂′-endo conformation while the guanine ribose is in C₃′-endo [29].

(C) Close-up stereo view of protein-RNA interactions, shown for a single RNA molecule; van der Waals contacts are not shown with the exception of the parallel base pair stacking of F125. Note that most of the interactions are contributed by the guanosine phoshate. Hydrogen bonding distances are indicated by dashed lines and contributions of symmetry-related residues are marked with an asterisk (N154*).

The RNA oligonucleotide is composed of a 5′-phosphate-depleted uridine residue (Uri1R, bound to molecule A), followed by a guanosine phosphate (Gua2R) and an adenosine phosphate (Ade3R) at the 3′ end. Both termini are pointing into the interior of the pore, being accessible for bulk solvent and, putatively, for 3′- and/or 5′-elongated RNA molecules Figure 1, Figure 3. The high quality of the final electron density map, including the observed interactions, permitted unambiguous identification of this Uri-Gua-Ade nucleotide (Figure 1D). Uri1R displays its sugar puckering as C₂′-endo [29] (Table 2) , which is stabilized by a number of polar interactions between symmetry-related Uri1R residues. The base moiety is positioned in syn conformation, fixed by an internal hydrogen bond (Uri1R O₅′-O₂, 3.3 Å). This positions the aromatic base ring perfectly parallel to that of a symmetry-equivalent Uri1R* residue (3.9 Å between planes), resulting in a very strong π interaction (Figure 3B).Table 2

Conformational Parameters of the Bound Triribonucleotide

Nucleoside Uri1R is connected with Gua2R by a phosphate-linker in extended conformation, and no interactions are observed between these two residues. Gua2R is also puckered as purely C₂′-endo, with its O₂′ atom stabilized by an interaction with the protein moiety (see below). The base, equally in syn conformation, is accommodated in a pocket in the inner pore surface and at the interface between protein protomers (see below). This pocket is formed by protein residues and by the perfectly parallel base plane of a symmetry-related Gua2R* residue.

The sugar-phosphate backbone folds back after Gua2R to reach Ade3R, so that both the guanine and adenine bases come to interact with each other, being approximately perpendicular (Gua2R N₂-Ade3R N₁, 2.9 Å). A further interaction of the adenine is observed with the preceding phosphate group (Ade3R N₆-Gua2R O2P, 3.3 Å). The sugar puckering of this 3′-terminal residue is C₃′-endo, enabling another interaction with the sugar of the preceding residue (Gua2R O₂′-Ade3R O₄′, 3.4 Å), and the base is positioned in anti (Figure 3B and Table 2).

Inside of the central cavity of the ring, the dimer-dimer interface is stabilized by the interaction with the ssRNA segment (Figure 3A). The main specificity of the observed trinucleotide binding resides in the central guanosine phosphate. The guanine base is placed in a deep pocket located at the interface between two protein protomers related by crystallographic symmetry (residues from the symmetry-related polypeptide are characterized by an asterisk) and interacts mainly with Arg134 (Gua2R O₆-Arg134 Nϵ, 2.9 Å; Gua2R N₇-Arg134 Nη2, 2.9 Å) and Phe125 (parallel ring stacking, 3.4 Å between aromatic planes) of one protein chain. The backbone phosphate group of Gua2R is linked via a double-headed salt bridge with Arg134* (Gua2R O₂P-Arg134 Nη1, 2.9 Å; Gua2R O₅′-Arg134 Nη2, 3.0 Å), while the sugar moiety contacts the protein main chain (Gua2R O₂′-Gly126 N, 3.1 Å) (Figure 3C)

Fewer contacts are observed between the flanking residues of the oligonucelotide and the protein. Nucleoside Uri1R is not involved in any polar interactions with the protein octamer, and only hydrophobic forces are established by the aromatic base to the side chains of Leu132* and Leu158*. The 3′-terminal nucleotide Ade3R is placed in a shallow cavity lined out by the protein main chain (His123-Gly126) and the side chains of Tyr171, Asn154*, and Ile152*. It interacts directly via its base with the protein through one hydrogen bond (Ade3R N₁-Asn154* Nδ2, 3.0 Å) and via its sugar moiety through a second one (Ade3R O₂′-Gly126 O, 3.0 Å) (Figure 3C).

The observed structural features and the fact that the protein oligomer was capable of sequestering a triribonucleotide of a particular sequence from the expression host E. coli strongly suggest that the octameric VP40 N-terminal domain displays a marked selectivity for the specified sequence, in particular for a central guanosine phosphate. Furthermore, the implication of the sugar O₂′ atoms in a number of key interactions (both intra- and intermolecular) further permit to discard the potential binding of a trideoxyribonucleotide with an equivalent sequence. Therefore, the observed binding must be considered both RNA- and sequence-specific. In addition, all polar residues involved in the RNA interaction are strictly conserved between Ebola and Marburg VP40 sequences and some van der Waals contacts are substituted by conservative changes.

The NTD of the monomeric closed conformation [9] and the protomer within the oligomeric structure can be overlaid with an rms deviation (matching the Cα atoms of residues 72 to 189) of 2.8 Å. Most of the differences observed lie in the loop regions and the positions of α helix 2 and β strand 6. In particular, α helix 2 moves 3.3 Å (at Cα position 117) and β strand 6 4.3 Å (at Cα position 173), resulting in major differences within this region (Figure 4A) .Figure 4

Comparison of the NTDs Derived from the Closed Monomeric Conformation and from the Ring Structure Unveils Major Conformational Changes

(A) Superposition of Cα atoms 71 to 191 results in an rms deviation of 2.8 Å. Sites of major conformational movements are indicated with arrows. The NTD from the closed monomeric conformation is shown in red and the one from the octamer structure in yellow.

(B) Schematic overview of the two major conformational changes in VP40. An N-terminal loop (gray) and the C-terminal domain (gray) from the closed VP40 conformation must change their conformation to achieve octamerization. This is indicated by the ribbon drawing of the three regions involved; the potential movement of the two domains with respect to the N-terminal domain is highlighted by arrows.

Part of the NTD construct VP40(31–212), which was used for crystallization, reveals SDS resistance when separated under nonboiling conditions on SDS-PAGE. A high molecular weight form is detected that migrates close to the 150 kDa marker protein (Figure 5A , lane 2). The same oligomeric form is observed when crystals belonging to space group P6₂22 are dissolved and separated under nonboiling conditions (Figure 5A, lane 3). This indicates that the SDS resistance is an intrinsic feature of octameric ring-like VP40 in complex with RNA. In addition, monomeric VP40 adopts the same oligomeric SDS-resistant structure when incubated with 2 M urea in the presence of E. coli nucleic acids. The formation of ring-like structures was also confirmed by electron microscopy (data not shown). The oligomeric form of SDS-resistant VP40 in complex with nucleic acids thus migrates above the 220 kDa marker protein (Figure 5B; lane 1), while monomers and dimers are detected under boiling conditions (Figure 5, lane 2).Figure 5

SDS Resistance of Oligomeric Ring-like VP40 in Complex with RNA

(A) SDS resistance of VP40(31–212). Lane 1, boiled sample; lane 2, nonboiled sample; lane 3, nonboiled sample of crystals of VP40(31–212) in space group P6₂22.

(B) SDS resistance of full-length VP40. Lane 1, full-length recombinant VP40 destabilized in the presence of E. coli nucleic acids and separated on SDS-PAGE; lane 1, nonboiled; lane 2, boiled. Lane 3, detection of full-length VP40 in Ebola virus infected cells under nonboiling conditions and boiled (lane 4). Note that the high molecular weight form in lane 3 corresponds to the size of the oligomeric full-length VP40 generated in vitro (lane 1). The monomeric form and the proposed octameric form are both indicated by arrows and a band corresponding to a dimer (lane 2) is indicated by an asterisk.

(C) SDS resistance analysis of VP40 present in purified Ebola virus particles. Samples were separated under boiled (lane 1) and nonboiled conditions (lane 2).

Samples were separated on gradient SDS-PAGE and positions of marker proteins are indicated. Bands in (A) were detected by Coomassie blue staining and in (B) and (C) by Western blot.

VP40(31–212) was expressed and purified as described [10], and a C-terminally truncated form of VP40, VP40(31–194), was cloned, expressed, and purified as described for VP40(31–212). For crystallization, 1 μl of VP40(31–212) (10 mg/ml) and 1 μl of well solution (100 mM HEPES [pH 7.5], 1.5 M NH₄H₂PO₄, 15% MPD), and 1 μl of VP40(31–194) (13 mg/ml) and 1 μl of well solution (100 mM Na-acetate [pH 4.6], 35% MPD, 4% glycerol) were mixed and equilibrated against the well buffer. VP40(31–194) was nonspecifically proteolyzed in the crystallization drop, and the N terminus of a crystal was determined to start at position 55 by chemical sequencing [VP40(55–194)]. For cryo-protection, VP40(31–212) crystals were briefly transferred to well solution containing 5% glycerol, and VP40(31–194) crystals were transferred into well solution containing 15% glycerol before flash cooling in a stream of liquid nitrogen.

VP40(55–194) crystallizes in space group P422 with unit cell dimensions of a = b = 80.51 Å and c = 46.77 Å, with one molecule per asymmetric unit and a Matthews-parameter V_M = 2.5 ų/Da (50% solvent; [40]). One native data set was collected at beamline ID14-EH1 (ESRF, Grenoble). A heavy-ion derivative was obtained by soaking a crystal during 6 hr in the cryo-protecting buffer containing 200 μM potassium tetra-chloroplatinate(II), and one data set was collected on a rotating anode source. A multiple-anomalous diffraction (MAD) experiment at the platinum L_III absorption edge was performed at beamline BM-14 (ESRF, Grenoble). Therefore, two data sets, at λ = 1.069 Å (absorption peak) and λ = 0.979 Å (hard remote), were collected from the same crystal, with data extending to 1.63 Å (Table 1). All diffraction data were indexed and integrated with DENZO [41], scaled, merged, and reduced with SCALA within the CCP4 package [42].

A single platinum site was located in an anomalous difference-Patterson map (bound to Met89 Sδ). Phases up to 1.63 Å resolution were calculated from this single refined site with the program SHARP [43], from the phases provided by the two-wavelength MAD (synchrotron data) and the SIRAS (rotating-anode data) experiments (Table 1). A posterior density modification step (solvent flipping) with SOLOMON [42] rendered an electron density map of excellent quality that permitted density improvement and automated chain tracing using wARP/ARP [44]. Subsequent final corrections of the model, the introduction of solvent molecules where appropriate if present in σ_A-weighted 2mF _obs − DF _calc and mF _obs − DF _calc maps, and identification and building of the triribonucleotide were performed with the programs O [45] and TURBO-FRODO [46]. Positional and temperature-factor crystallographic refinement was performed with the program CNS v. 1.0 [47]. As a last step, REFMAC5 was used with TLS refinement [48] to better account for the anisotropic thermal displacements. The final refined model consists of protein residues 69 to 192 (molecule A), ribonucleotides 1 to 3 (chain R), and 158 solvent molecules. It has been refined employing native data to 1.6 Å resolution, with a final crystallographic R value of 16.4% and an R_free value of 18.2%. The model has good stereochemistry (Table 1) and all residues lie within the allowed regions of a Ramachandran plot, as checked with PROCHECK [49].

VP40(31–212) crystallizes in space group P6₂22 with unit cell dimensions a = b = 79.61 Å and c = 239.18 Å, and two monomers per asymmetric unit (V_M = 2.7 ų/Da; 55% solvent contents). A native data set to 2.6 Å resolution was collected at beamline ID14-EH4 at the European Synchrotron Radiation Facility (ESRF, Grenoble) and processed as described (see Table 1). The structure was solved by Patterson-search techniques using the superimposed coordinates of the VP40(55–194) structure and those of the NTD as present in the full-length molecule [9]. The rotation function could only be solved employing the program BEAST [50]. Translation functions were then computed with AMoRe [51], resulting in a correlation coefficient (CC) of 47.0% and an R value of 47.9%. This calculation corroborated P6₂22 as the correct space group. An electron density map calculated with the obtained model phases revealed extra density in the central cavity of the octameric protein ring, confirming the presence of the triribonucleotide (not used during the searches nor for phasing) and thus the correctness of the solution. The structure was improved in several cycles of manual model building and crystallographic refinement as mentioned above. The final model includes protein residues 69 to 190 (molecule A), residues 69 to 191 (molecule B), two triribonucleotides of sequence 5′-U-G-A-3′ (chains R and S), and 113 water molecules. Two chloride anions at the interface between octamers were tentatively assigned based on very strong electron density peaks that could not be explained by solvent molecules. Residues 31 to 68 and 191/192 to 212 in each protomer are disordered (33% of the total). Therefore, the structure could not be refined beyond a crystallographic R value of 30.6% (R_free value of 32.9%; see Table 1). The model has good stereochemistry (Table 1) and all residues lie in allowed regions of the Ramachandran plot, with just one exception. The final model displays high temperature factors for the ribonucleotide moieties. One of the two triribonucleotides clearly displays electron density for a phosphate group on the 5′ terminus, indicating that ssRNA extends in a flexible way in that direction.