Main

We expressed and purified C-terminally truncated Zaire ebolavirus nucleoprotein (NP), containing residues 1−450 (NP 1−450) in a human cell line (Fig. 1a), in which it sequestered cellular RNA and assembled into a rigid helix indistinguishable from the viral nucleocapsid core3,4,5,7. The sample was imaged without fixation using cryo-electron microscopy (cryo-EM), and its structure was determined by single-particle analysis (SPA) (Fig. 1b, Extended Data Fig. 1a–e, Extended Data Table 1). The complex forms a left-handed helical tube with outer and inner diameters of 295 and 175 Å, respectively, and features a characteristic zipper-like arrangement of parallel C-terminal α-helices (Fig. 1b–d). Its diameter and helical symmetry correspond closely to those of the recently reported Ebola virion reconstructed by cryo-electron tomography (cryo-ET)4,5. Our cryo-EM map enabled generation of an atomic model, including clearly resolved RNA nucleotides (Fig. 1e, Extended Data Figs. 1d, 1f, 2).

Fig. 1: Cryo-EM reconstruction and atomic model of Ebola NP 1−450 helix.
figure 1

a, Structural layout of the Ebola NP sequence. The full-length NP contains 739 amino acids divided into an N-terminal arm domain, an NP core consisting of N- and C-terminal lobes, a disordered linker and a C-terminal tail. b, The iso-electron potential surface of NP–RNA complex reconstruction (contoured at 3σ above average) was calculated from 117,552 single-particle images. The RNA strand and a single NP subunit are highlighted in red and orange, respectively. c, d, The isolated NP subunit (2.5σ) with our atomic model (RCSB Protein Data Bank code (PDB): 5Z9W) in ribbon representation. e, Isolated electron potential map (5.0σ) of RNA superimposed with the atomic model of RNA. RNA bases were modelled as uracil. There are 6 nucleotides per NP. Scale bars, 20 Å.

Full size image

The transition from the RNA-free to the RNA-bound state requires conformational changes in both RNA and NP. Encapsidation of RNA by NP has been addressed by two leading models. One model has NP encapsidating the RNA in a closed conformation with a hinge-like motion around the junction between N- and C-terminal lobes6,8,9,10,11,12,13,14. The other has RNA and a phosphoprotein competing directly for the RNA-binding cleft15,16 (Extended Data Fig. 3). To discriminate between these possibilities, we compared our atomic model of RNA-bound NP with structures of RNA-free monomeric NP chaperoned by phosphoprotein (VP35)6,8. We found that the relative positions around the junction—specifically, α-helices α9−α12 in the N-terminal lobe and α13−α15 in the C-terminal lobe—do not change significantly, which indicates that the transition is not a simple hinge motion. Binding regions of RNA and the N-terminal peptide of VP35 do not overlap, weakening the case for direct competition. These observations imply that the transition to the closed conformation is a consequence of multiple elements. In Ebola virus NP, the β-hairpin (β3−β4, which is unique to the family Filoviridae) is rotated approximately 90° between RNA-free and RNA-bound states (Fig. 2a, Extended Data Fig. 3, Supplementary Video 1). Simultaneously, the adjacent 310 helix η2 and α-helices α16−α18 are shifted to form a ‘closed’ state (Fig. 2a, Extended Data Fig. 3a). The C-terminal helix α18 clamps the RNA strand and serves as a fastener to cover the RNA in its cleft, as previously reported5. A short 310 helix η5 (residues 329−333), which does not exist in the RNA-free state, shortens and directs the adjacent loop 334–338 to secure the RNA tightly (Fig. 2a, b, Extended Data Fig. 3a).

Fig. 2: NP–RNA interactions.
figure 2

a, Comparison of an RNA-free, monomeric NP model (PDB: 4YPI, grey) and RNA-bound NP (PDB: 5Z9W, coloured, from this study). The conformational transition to the encapsidated state is orchestrated by a 90° rotation of the β-hairpin, formation of a new secondary structural element (η5) and shift of the C-terminal α-helices. Arrows indicate transitions from RNA-free (open) state to a bound (closed) state. The green arrow indicates formation of a 310 helix η5 and shifting of the polypeptide chain. See also Supplementary Video 1. b–d, Close-up view of the NP–RNA binding region. Blue dashed lines depict predicted hydrogen bonds. b, 310 helix η5, loop 334–338 and helix α18 coordinate RNA bases and flexibly adopt the RNA path. c, Interactions between positively charged NP residues and RNA-backbone phosphates. d, Possible hydrogen bond of E309 with the adjacent RNA pentose hydroxyl group. e, Schematic of RNA nucleotides within the nucleoprotein. Two RNA base triplets with opposing orientations are coordinated by each NP subunit. The RNA bases are numbered sequentially in the 3′ to 5′ direction, representing negative-sense genomic RNA, facing inward (green) towards the helical tube or outward (red). A set of specific electrostatic interactions on both the NP core and α18 determine nucleotide orientation. Secondary structure labels are based on our atomic model (see Extended Data Fig. 2).

Full size image

The viral RNA is firmly secured in the NP–RNA complex within a positively charged ellipsoid cleft with a cross-section approximately 5 × 11 Å (Extended Data Fig. 4). The conformation of the RNA strand follows an undulating path along the helical NP strand. We found 6 RNA nucleotides clamped in the cleft (Fig. 1c, d), confirming predictions4,5,6,8,9,17. These nucleotides adopt a ‘3-bases-inward, 3-outward’ configuration (Figs. 1e, 2a, b), similar to measles virus and parainfluenza virus 513,18. Coordination is achieved by hydrogen bonding and electrostatic interactions between NP and negatively charged phosphate groups of RNA.

Polar residues on helix η5 and loop 334–338 are within hydrogen-bonding distance of the RNA backbone, adjacent to helix α17, and clamping helix α18 (Fig. 2b). Therefore, this region not only aligns the RNA but also serves as an interface for NP–RNA interactions. Positively charged residues K160, K171, R174, R298 and R401—some of which are important for RNA binding in filovirus NP8,19—were found in close proximity to RNA nucleotides, with their side chains pointing towards RNA-backbone phosphates (Fig. 2c). K248, which is non-essential for RNA binding8,19, faces the RNA bases. These common residues suggest a shared structural motif for RNA interactions within the family Filoviridae. Unlike in measles virus, we found no indication of π–π stacking as a possible cause of the switch between ‘3-in’ and ‘3-out’ nucleotide directions13. The location of R298 on the inner side of the cleft enables it to interact with the RNA backbone. Combined with the restricted space in the cavity and R401 on helix α18 on the outside, these structural constraints may provide an alternative mechanism for the switch. In addition, E309 is within hydrogen-bonding distance of the closest RNA nucleotide on the next NP subunit (Fig. 2d). These specific interactions with the RNA backbone are consistent with the sequence-independent NP–RNA interaction of negative-sense RNA viruses12,13,15,20,21,22,23,24,25 (Fig. 2e).

The N-terminal arm domain of NP is essential for NP oligomerization6,8. Our map clearly shows that the N-terminal arms cross-link subunits on the inside of the helical tube (Fig. 3a–c, Extended Data Fig. 5a, d), confirming previous observations5. Furthermore, N-terminal residues have been predicted to associate with the same binding site as the N terminus of VP355,6. VP35 binds to a hydrophobic pocket in the NP C-terminal lobe and acts as a chaperone to keep NP in its monomeric RNA-free state6. To clarify the mechanism of NP oligomerization, we examined the interface between the N-terminal arm and the pocket of adjacent NPs (Fig. 3b, c). The N-terminal arm binds to the same location as VP35 within this pocket via I24, L25 and A27 on its short N-terminal α-helix (Fig. 3c, Extended Data Fig. 5a–d). Whereas VP35 straddles the β-hairpin in the RNA-free monomeric form6,8 (Extended Data Fig. 5a, c), adjacent NP subunits extend their N-terminal arm to interact directly with the rotated β-hairpin and the pocket in the NP–RNA complex (Fig. 3c). Furthermore, the N-terminal arm is tucked behind K257 on the rotated β-hairpin and the pocket (Fig. 3c, d, Extended Data Fig. 5d). Compared with monomeric NP, we discovered that loop 220–227 moves to create a groove between itself and helix α4, which makes space for the N-terminal arm of the adjacent NP (Fig. 3e, Extended Data Fig. 5e). This conformational change exposes two amino acids (D226 and D229) that serve as electrostatic anchors for partners (R37 and R39) on the N-terminal arm. This observation is consistent with a previous report of increased solvent exchange in this area upon oligomerization6.

Fig. 3: Intrastrand and interstrand NP–NP interactions.
figure 3

We refer to the NP–RNA band generically as a ‘strand’, ‘intrastrand interactions’ in the lateral direction and ‘interstrand interactions’ in the axial direction. a, An atomic model of the NP–RNA complex. Larger and smaller dashed boxes outline close-ups of two adjacent NP molecules (b) and interstrand NP–NP interface (g), respectively. b, An NP pair viewed from inside the helix. Dashed boxes delineate respective close-up figures. c, Intrastrand NP–NP interactions between the N-terminal arm and the hydrophobic pocket. The β-hairpin is shown in dark grey. d, Schematics of the hydrophobic pocket illustrating the binding mode of the N-terminal arm in the RNA-bound state (top) and VP35 in the RNA-free state (bottom). e, Contacts between the N-terminal arm and the N-terminal lobe. f, The contact region between α17 of NPn and α18 of NPn + 1 subunits. The blue dashed line indicates a hydrogen bond. g, Interstrand NP–NP interactions (with subunits NPn ± 24 and NPn ± 25) formed by 3 pairs of amino acid side chains, K110–E349, E92–Q371 and R105–Q381.

Full size image

Lateral assembly of NP monomers relies on NP–NP and NP–RNA interactions. Beside the N-terminal arm, multiple contact regions are essential for RNA encapsidation26. Whereas the N-terminal arm of vesicular stomatitis virus is co-located with the NP–NP interface, contact regions in NP 1–450 are more spatially separated. Stoichiometric excess of VP35 does not dissociate the NP oligomer after the NP–RNA complex has been formed4,6,27. Our data and previous reports show that the NP–RNA complex retains lateral binding of NP subunits to form a flat ribbon-like strand, even where the NP–RNA complex is uncoiled4,6,7,8 (Extended Data Fig. 1a, b). We identified a major NP–NP interface between the parallel C-terminal extended α-helices α17 and α18 (Fig. 3a, b, f), which contains surface patches formed by highly conserved hydrophobic residues (Fig. 3f, Extended Data Fig. 5b). These patches are buried in the NP oligomer. In non-physiological packing within the crystallographic unit cells of filovirus NP, α17 and α18 of adjacent NPs also make contacts in almost identical configurations8,19 (Extended Data Fig. 5f), which emphasizes the high specificity of these helix–helix interactions. The function of the hydrophobic interface may be to align the C-terminal α-helices to form the characteristic zipper-like arrangement throughout the NP strand.

In our structure of NP 1–450, the interstrand interface is uniquely determined by NP–NP interactions. In the axial direction, any arbitrary NPn subunit shares contact regions with two pairs of neighbours located above and below (Fig. 3a). Three pairs of amino acids, K110–E349, E92–Q371 and R105–Q381, are located near enough for electrostatic interaction to occur between the zipper-like C-terminal α-helices and the β-sheet of the N-terminal lobe (Fig. 3g). Of note, most other amino acids at the interstrand interface within 2−6 Å of each other are also charged or polar (Extended Data Fig. 6), suggesting that the nature of interstrand NP–NP interaction is mainly electrostatic. This broad distribution of charged residues may allow distance variation without rupture upon helix bending, and would explain the phenotypic flexibility of filoviruses. The hydrophilic character of the interstrand interface further explains how the NP–RNA complex can easily uncoil, yet remain stable in solution.

Oligomeric NP structures determined at intermediate resolution by cryo-ET5 and SPA28 have previously been reported. The helical parameter of our NP–RNA complex is consistent with the full virion in the cryo-ET study5. However, our structure differs from the bacterially expressed, in vitro-assembled NP (residues 25–457)28, which formed helical tubes of heterodimeric NP subunits with nearly twice the diameter of native Ebola virus. That structure has fundamentally different helical symmetry and contains no RNA density (Extended Data Fig. 7). Nevertheless, this N-terminal truncation mutant can still assemble into a paired back-to-back helical configuration with a partially disordered C-terminal helix located away from the RNA cleft. This further emphasizes the importance of the first 24 N-terminal residues for cooperative control in alignment and oligomerization of NP subunits, the clamping helix α18 and the RNA itself during assembly of the NP–RNA complex under physiological conditions.

Here we report the structure of the Ebola virus NP 1–450 RNA complex at near-atomic resolution. Insights from our study provide opportunities for rational design of novel antivirals—for example, creation of peptidomimetics competing with the binding site for the N-terminal arm helix or the C-terminal hydrophobic patches, or by cross-linking the interstrand interface. On the basis of its 6 nucleotides per NP subunit and genome size of the Zaire ebolavirus (18,960 bases), the length of the nucleocapsid necessary to accommodate a single copy of genomic RNA is 950 μm. This dimension is consistent with the average length of the filamentous Ebola virion29,30. Furthermore, a histogram of virion length shows a peak distribution with integer multiples of this distance30. Because the outer surface of mature Ebola nucleocapsid is in contact with matrix protein VP40, the NP–RNA complex may serve as a yardstick to determine virion length, causing pinch-off at its ends. Our structure supports conclusions by previous studies5,6—chiefly, the important role of the N-terminal arm in subunit linking and in competing with the VP35 N-terminal peptide. Encapsidation of the viral genome takes place concomitantly with genome replication, and lateral NP–NP interactions are required for assembly of the nucleocapsid. Therefore, NP oligomerization in open conformation by self-alignment via the hydrophobic interface must precede RNA clamping or occur simultaneously. Because NP oligomerization and NP–RNA binding are cooperative, the end of the RNA strand would terminate assembly.

Methods

Purification of the Ebola NP–RNA complex

Human embryonic kidney 293T cells (tested mycoplasma-free, authenticated by DNA fingerprinting and obtained from Kawaoka Lab at Institute of Medical Science, University of Tokyo) were transfected with a plasmid encoding truncated nucleoprotein (NP residues 1–450) of Zaire ebolavirus (strain Mayinga-76, GenBank code AF499101), the type species of the genus Ebolavirus31. Three days after transfection, cells were rinsed with Tris-HCl buffer (10 mM Tris-HCl (pH 7.8), 150 mM NaCl, 1 mM EDTA), and lysed with 0.1% Nonidet P-40 in buffer supplemented with a protease inhibitor (Roche) and an RNase inhibitor (Promega). The lysate was centrifuged at 20,000g at 4 °C for 10 min to remove insoluble compounds. Supernatant was loaded onto a discontinuous 25–40% (wt/wt) CsCl gradient and centrifuged at 250,000g at 4 °C for 1 h. Absorbance of fractions at 280 nm was measured, and peak fractions were subjected to centrifugation at 200,000g at 4 °C for 30 min. The pellet was suspended in the buffer and stored at 4 °C.

Cryo-EM specimen preparation

Two microlitres purified NP–RNA complex solution at protein concentration of 0.3 mg ml−1 was applied to C-Flat CF-1.2/1.3-4C grids (Protochips), previously treated with H2/O2 plasma (Gatan Solarus). Grids were blotted on a Vitrobot Mark IV (Thermo Fisher Scientific) before plunge freezing in liquid ethane.

Cryo-electron microscopy

Images of the liquid nitrogen-cooled sample were acquired on a Titan Krios cryo-TEM (Thermo Fisher Scientific) equipped with an energy filter (Gatan Quantum 968) at 300 kV in EFTEM nanoprobe mode. Data were acquired automatically as movies with the Leginon System32 on a K2 Summit detector (Gatan). For details, see Extended Data Table 1.

Image processing

Super-resolution movie frames were 2× Fourier-cropped, gain-normalized, aligned, dose-weighted and summed using MOTIONCOR233. Helices were manually traced with E2HELIXBOXER32. Subsequent helical image processing steps were performed with SPRING34 and RELION 2.135,36. Defocus values were estimated with CTFFIND437. Classes with visual details were selected after two rounds of 2D classification in RELION 2.1. Helical symmetry parameters (helical pitch and subunits per turn) were exhaustively explored by multiple real-space correlation searches with the program SEGCLASSRECONSTRUCT of the software package SPRING34. The search provided many correlation peaks corresponding to various combinations of parameters. Then, 3D classification in RELION 2.1 was performed with symmetry search around the respective peaks (search range for pitch ± 10%, rotation ± 0.6 subunits per turn). Eighty-eight per cent of the segments fell into three of four classes with nearly identical helical symmetry and yielded reasonable reconstructions at high resolution. These three classes were combined and submitted to further 3D refinement with an additional symmetry search (resulting in no significant change in helical symmetry). The helical symmetry parameter converged to pitch = 73.56 Å, 24.44 subunits per turn (rise = 3.01 Å, rotation = 14.73°). A SPIDER38-generated cylinder-shaped electron density map was used as an initial alignment reference for 3D classification and refinement in RELION 2.1. The resolution estimate of the final reconstruction was based on Fourier shell correlation (FSC) between independently refined half sets using the high-resolution noise substitution method39 and a soft-edged mask. Local resolution was calculated with RELION 2.1. The locally-sharpened map was normalized with MAPMAN40. Detailed image processing conditions are listed in Extended Data Table 1.

Atomic model building and refinement

Atomic modelling of the NP 1−450 helix used template-based and de novo structure modelling. First, a previously reported crystal structure of Ebola NP residues 38−385 (PDB: 4YPI)8 served as the initial model and was docked as a rigid body into our 3.6 Å-resolution cryo-EM map. Next, N- and C-terminal residues of NP, and RNA nucleotides, were added, and the model was adjusted manually with COOT41. The RNA was built de novo with COOT. During manual model building, stereochemistry of peptide bonds was monitored using a Ramachandran plot42. N- and C-terminal residues 1−18 and 407−450 were not modelled owing to a lack of interpretable map features, although the highly flexible disordered C-terminal region was recognizable in 2D class-average images (Extended Data Fig. 1c). The model was refined including data to 3.6 Å resolution with phenix.real_space_refine43 in the presence of 8 surrounding subunits including an extended RNA strand. Secondary structure elements for protein and base-pair stacking for RNA were estimated by phenix.secondary_structure_restraints43, and used as restraints for the refinement. The model was validated with MolProbity44. The unambiguous assignment of nucleic acid density in our map represents an independent element of map validation. Detailed conditions of model refinement and validation are given in Extended Data Table 1. The electron potential map and atomic model were visualized with UCSF Chimera and ChimeraX45,46.

Statistics and reproducibility

Data collection and atomic model statistics are listed in Extended Data Table 1. The method at the core of this work is cryo-EM and single particle analysis. The reconstruction was assembled from a total number of 117,552 single particle images. These images represent projections of a common 3D object (the NP–RNA complex) and were averaged in three dimensions to result in the present reconstruction. The process used two independent sets, each containing half the number of images. These half-set reconstructions were compared to each other using a FSC, which measures the degree of correspondence as a function of resolution. Using the established criterion for resolution measurement (FSC = 0.143 for independent datasets), the two half-set reconstructions agreed up to a spatial resolution of 3.6 Å.

Reporting Summary

Further information on research design is available in the Nature Research Reporting Summary linked to this paper.