Predictive and comparative analysis of Ebolavirus proteins

Abstract

Ebolavirus is the pathogen for Ebola Hemorrhagic Fever (EHF). This disease exhibits a high fatality rate and has recently reached a historically epidemic proportion in West Africa. Out of the 5 known Ebolavirus species, only Reston ebolavirus has lost human pathogenicity, while retaining the ability to cause EHF in long-tailed macaque. Significant efforts have been spent to determine the three-dimensional (3D) structures of Ebolavirus proteins, to study their interaction with host proteins, and to identify the functional motifs in these viral proteins. Here, in light of these experimental results, we apply computational analysis to predict the 3D structures and functional sites for Ebolavirus protein domains with unknown structure, including a zinc-finger domain of VP30, the RNA-dependent RNA polymerase catalytic domain and a methyltransferase domain of protein L. In addition, we compare sequences of proteins that interact with Ebolavirus proteins from RESTV-resistant primates with those from RESTV-susceptible monkeys. The host proteins that interact with GP and VP35 show an elevated level of sequence divergence between the RESTV-resistant and RESTV-susceptible species, suggesting that they may be responsible for host specificity. Meanwhile, we detect variable positions in protein sequences that are likely associated with the loss of human pathogenicity in RESTV, map them onto the 3D structures and compare their positions to known functional sites. VP35 and VP30 are significantly enriched in these potential pathogenicity determinants and the clustering of such positions on the surfaces of VP35 and GP suggests possible uncharacterized interaction sites with host proteins that contribute to the virulence of Ebolavirus.

Introduction

Zaire Ebolavirus, the pathogen for Ebola Hemorrhagic Fever (EHF) with a 25–90% fatality rate,¹ continues to threaten people's lives. The current (2013 – Jun. 2015) West African outbreak of EHF has infected more than 27,000 people and caused 11,000 deaths.² The genus Ebolavirus contains 5 known species: Bundibugyo (BDBV), Reston (RESTV), Sudan (SUDV), Taï Forest (TAFV) and Zaire ebolavirus (ZEBOV).³ The current outbreak is associated with ZEBOV.⁴ Four Ebolavirus species cause EHF in human, with the sole exception being RESTV.⁵ RESTV can cause EHF to long-tailed macaque (Macaca fascicularis). People who had contact with RESTV-infected monkeys tested positive for RESTV antibodies but did not develop symptoms associated with EHF.⁵

Ebolavirus belongs to the order Mononegavirales and the family Filoviridae.³ Its RNA genome encodes the following 7 protein products: Envelope glycoprotein (GP), Nucleoprotein (NP), RNA-dependent RNA polymerase L (L), Membrane-associated protein VP24 (VP24), Minor nucleoprotein VP30 (VP30), Polymerase cofactor VP35 (VP35), and Matrix protein VP40 (VP40). The GP transcript can be edited,⁶ and the gene product can be processed by host protease, giving rise to 4 alternative forms of gene products: GP1,2; GP1,2delta; sGP and ssGP. Host furin can cleave the longest product translated from edited GP mRNA and generate GP1,2, which consists of 2 peptide chains connected by a disulfide bond,^7,8 GP1 and GP2. GP1,2 is assembled on the membrane of Ebolavirus and mediates cell entry. GP1,2delta is the processed product after removal of the C-terminal transmembrane region of GP1,2 by host ADAM17.⁹ Other products of the GP gene, sGP and ssGP are translated from the unedited mRNA and alternatively edited mRNA, respectively.^10,11 These products share the N-terminal 295 residues with GP1,2, but differ in their short tails (69 and 3 residues, respectively). GP1,2delta, sGP and ssGP may prevent the neutralizing antibodies from binding GP1,2 on the virus surface, contributing to the immune evasion of the virus.¹²

In addition to serving as structural components, the Ebolavirus proteins play multiple roles in the virus life cycle. GP mediates cell entry^13,14 and membrane fusion^15,16 between the virus and the host cell. NP encapsidates the genome and protects it from nucleases.^17,18 VP30 is a transcription anti-terminator^19,20 and regulates the switch between transcription and replication.^21,22 VP35 acts as a cofactor of the polymerase,^23,24 and VP40 may also play a role in genome replication and transcription.²⁵ VP24 and VP35 participate in viral nucleocapsid assembly,¹⁸ and VP40 is essential for virus budding and assembly.^26-28 In addition, GP, VP24, VP30, VP35 and VP40 interact with multiple host proteins to complete the viral life cycle and to suppress the host immune response.

Three-dimensional (3D) structures are available for a number of Ebolavirus proteins. Interpreting available experimental data and sequence variation among Ebolavirus species in the context of the 3D structures not only allows researchers to understand detailed mechanisms for cell entry, virus assembly and immune suppression, but also provides promising leads for structure-based drug design. In the current study, we predict the 3D structure and functional sites for Ebolavirus protein domains that are not yet characterized. In addition, we compare sequences of Ebolavirus proteins' interacting partners from RESTV-resistant primates with those from RESTV-susceptible monkeys. Elevated sequence divergence for GP and VP35's interaction partners suggests that these 2 viral proteins may be responsible for host specificity in RESTV. Finally, we compare the protein sequences from different Ebolavirus species to detect positions that are conserved among human pathogenic species but different in non-pathogenic RESTV (RESTV-specific mutations). Mapping of these RESTV-specific mutations and known functional sites to the 3D structures reveals clusters of RESTV-specific mutations on the surfaces of GP, VP35 and VP24. These clusters do not overlap with the known functional sites and may suggest novel interaction sites with host proteins.

Materials and Methods

Sequence analysis of Ebolavirus proteins

The protein sequences of ZEBOV were downloaded from the UniProt database²⁹ and submitted to the MESSA web server³⁰ to predict the secondary structure,^31,32 disordered regions,^33-36 transmembrane helices,^37-41 signal peptides,^38,39,42 coiled coils⁴³ and detect structure templates.^44,45 The 3D structures are mostly known, except for protein L, the N-terminal zinc-finger domain of VP30 and the coiled-coil region of VP35. For proteins and domains without known structure, we considered putative structural templates detected by HHpred,⁴⁵ iTASSER^46,47 and known structures for proteins of similar function from other families of RNA virus in PDB and ECOD databases.⁴⁸ Once a candidate structural template was detected, we further validated its relationship to the Ebolavirus protein by similarity in function, compatibility between the predicted secondary structure⁴⁹ of the Ebolavirus protein and the 3D structure of the template, conservation of residues in the Ebolavirus protein that were aligned to the active sites of the template, and the consistency among multiple structural templates. Sequences of the structural templates and the ZEBOV protein were aligned by Promals3D^50,51 and the alignments were manually adjusted to ensure that the corresponding secondary structure elements in different templates were aligned together. Based on these alignments and knowledge about functional sites in the template structures from literature, the active sites of the uncharacterized Ebolavirus domains were predicted.

Identification of positions associated with human pathogenicity

We downloaded protein sequences of 124 Ebolavirus samples from 5 Ebolavirus species⁴ at www.sciencemag.org/content/345/6202/1369/suppl/DC1, aligned them using MAFFT,⁵² and evaluated the similarity between amino acids at a certain position using BLOSUM62 scores.⁵³ We considered a position in the sequence alignment to be associated with the loss of human pathogenicity if it satisfies the following 2 criteria. First, the similarity in amino acids at this position from pathogenic species is always higher than the similarity between RESTV and a pathogenic species. Second, the average similarity in amino acids at this position from 4 pathogenic species (BDBV, TAFV, SUDV and ZEBOV) is significantly (p-value < 0.05) higher than that between RESTV and pathogenic species. In order to calculate the p-value for each position, we obtained an estimate of the background distribution for the positional difference between the average sequence similarity within a group of any 4 Ebolavirus species (all possible combinations except the one with all 4 pathogenic species) and the average sequence similarity between a fifth species and those in the group. This distribution suggests that a difference larger than 2 is associated with p-value less than 0.05. Enrichment of these pathogenicity-associated positions in each protein was measured by a binomial test (p = total number of pathogenicity-associated positions/total length of all proteins, m = number of selected positions in this protein, N = length of this protein). These pathogenicity-associated positions and the functional sites reported in literature were further mapped to the known 3D structures of Ebolavirus proteins.

Results and Discussion

3D structure and functional sites prediction for Ebolavirus proteins

Domain diagrams of all the Ebolavirus proteins are shown in Figure 1. Variable positions among the different Ebolavirus species are marked as black lines above the domain diagram. The average protein sequence identity between different Ebolavirus species ranges from 60% to 80% (Table S1). Ebolavirus proteins contain a significant fraction (20%) of structurally disordered regions, and the fraction of variable positions in these regions is significantly higher (p < 0.01) than in the structurally ordered regions. The 3D structures of globular regions are mostly known (Table S2)^54-71 except for the N-terminal zinc-finger domain of VP30, the coiled-coil domain of VP35, and protein L. Identification and analysis of structurally characterized homologs allowed us to predict the structure of the zinc-finger domain in VP30, the overall topology of NP, and the structure and catalytic sites for the catalytic domains of protein L.

Figure 1.

Domain diagrams for Ebolavirus proteins and coverage of the proteins by experimentally determined and predicted structures. The domains of each protein are represented by boxes on a thread and the positions that are variable among different Ebolavirus species are marked by black sticks above the domain diagrams. The band below is aligned to the domain diagram and the color of this band indicates the prediction status of the corresponding region. The color codes are: green, regions that are structurally characterized and adopt globular structure; red, regions that are experimentally determined but are intrinsically disordered; blue, regions with predicted 3D structure; yellow, coiled coil; cyan, transmembrane helix; purple, signal peptide; orange, predicted intrinsically disordered regions; gray, predicted regions that have a propensity to adopt global structure but 3D structure cannot be predicted;. Abbreviations: SP, signal peptide; FP, fusion peptide; TMH, transmembrane helix.

The zinc-finger domain of VP30

The zinc-finger domain of VP30 binds zinc⁷² and contains a conserved C-x8-C-x4-C-x3-H motif. A search using the VP30 zinc finger motif (residues 70–95) as a query against the SUPERFAMILY⁷³ database with HHpred web server (MSA generation method: HHblits, Maximal MSA Generation iterations: 3, Score secondary structure: yes, Alignment mode: local) reveals similarity (Probability: 52.4; Identity: 35%; E-value: 2.2) to the CCCH zinc finger superfamily (seed: SCOP domain d1m9oa_). Although this is not the best hit according to HHpred probability, it has the highest coverage and is the only one (probability cutoff: 20) that contains all the zinc-binding residues. In addition, a scan of PDB sequences with the conserved pattern C-x(8)-C-x(4)-C-x(3)-H using ScanProsite⁷⁴ reveals exactly the same motif in CCCH zinc fingers (PDB id: 2d9n). All the CCCH zinc fingers belongs to one homologous group in the ECOD database,⁴⁸ and this family contains the N-terminal domain of the transcription antiterminator M2-1 from another Mononegavirales, Pneumovirus (4C3B⁷⁵ and 4CS7,⁷⁶ alignment shown in Fig. S1). In addition to their common function, the C-terminal domain of M2-1 and Ebolavirus VP30 share the same topology (Fig. 2A, B). M2-1 uses a C-x7-C-x5-C-x3-H motif to bind zinc, which is connected to an α-helix at its C-terminus. The VP30 zinc-finger domain very likely adopts a similar structure (Fig. 2C, D), as supported by the presence of a similar C-x8-C-x4-C-x3-H motif and a predicted α-helix following the motif.

Figure 2.

Structure prediction for N-terminal domain of VP30. (A) 3D structure (PDB id: 2I8B) for VP30 C-terminal domain; (B) 3D structure (PDB ID: 4C3B) for Pneumovirus M2-1 C-terminal domain; (C) 3D structure (PDB ID: 4C3B) for Pneumovirus M2-1 N-terminal domain, which was used as template to predict the structure for the VP30 N-terminal domain; (D) structure model for the Ebolavirus VP30 N-terminal domain.

The N-terminal domain of NP

NP has 2 structural domains that are connected by a long disordered linker of about 240 amino acids. The C-terminal domain (PDB id: 4QAZ) is shared among Filoviridae and is involved in protein-protein interaction.⁵⁴ The N-terminal domain is likely shared among Mononegavirales. Known 3D structures of NP from several virus families^77-81 in this order possess the same topology (Fig. 3A–D). Structures of NP from Rhabdoviridae and Paramyxoviridae families are determined in complex with ssRNA (Fig. 3A, C), and they both clamp around the RNA using positively charged grooves (Fig. 3G, H) between the 2 subdomains after a remarkable conformational change compared to the RNA-free form (Fig. 3C, D). The RNA-bound NPs oligomerize to form a ring (Fig. 3I, J), but the oligmerization interfaces vary: Rhabdoviridae pack the ssRNA inside the ring formed by NPs while ssRNA binds on the outside of the NP oligomer in Paramyxoviridae.

Figure 3.

Structures of Mononegavirales Nucleoproteins. The virus family is labeled below. (A–D) Monomeric structures (PDB IDs: 2GIC, 1N93, 2WJ8, and 4CO6) of Nucleoproteins from Mononegavirales. The structures are colored in rainbow; (E, F) The electrostatic potential mapped onto the surface of Nucleoprotein structures (PDB ids: 2GIC and 2WJ8). Blue area corresponds to positively charged surface and the red area corresponds to negatively charged surface; (G) Structure model for the N-terminal domain of Ebolavirus NP; (H) Experimental structure of the N-terminal domain of Ebolavirus NP (PDB id: 4YPI); (I) The electrostatic potential mapped on to the surface of the experimental structure of the N-terminal domain of Ebolavirus NP; (J, K) Structure complex of RNA and Nucleoproteins from Rhabdoviridae and Paramyxoviridae (PDB ids: 2GIC and 2WJ8).

We predicted that the N-terminal domain of Ebolavirus NP adopts the same conserved topology as the other viral NPs and suggested that its structure is similar to the NP from Nipah virus (PDB id: 4CO6).⁸¹ The 3D structure of this domain was released while our manuscript was under review and supported our prediction (Fig. 3E, F). The available 3D structures for Ebolavirus NP^71,82 were all determined in the absence of RNA. But its similarity to the NPs of other Mononegavirales and the presence of a positively charged groove between the 2 subdomains suggest a similar RNA binding mode.

The RNA-dependent RNA polymerase catalytic domain of protein L

Sequence analysis suggests that the N-terminal half of protein L functions as a RNA-dependent RNA polymerase (RdRP), and is responsible for both DNA replication and transcription. HHpred⁴⁵ detects the Bunyavirus RdRP (PDB id: 5AMR⁸³) as a structural template (Probability: 84%). The alignment between Ebolavirus RdRP and Bunyavirus RdRP includes both the catalytic domain and a helical bundle connected to its C-terminus. These two domains are conserved among known structures of RdRPs from RNA viruses^84-88 (Fig. 4A–C). Known RdRPs from RNA viruses share the same topology except for Birnavirus RdRP, which has a circular permutation in the catalytic domain. This structural conservation of RdRPs across different groups of RNA virus suggests that the RdRP of Ebolavirus also adopts the same topology. Secondary structure prediction for the Ebolavirus RdRP is consistent with the topology adopted by most RNA viruses, but not with the circular permuted structure from Birnavirus (Fig. 4E).

Figure 4.

Structures of the catalytic domains of RNA-dependent RNA polymerases (RdRP) from RNA viruses and the structure model for Ebolavirus RdRP. The virus family is labeled below. The structures are colored in rainbow, with equivalent secondary structure elements from different structures colored similarly, except for the Birnaviridae RdRP, which has a circularly permutated topology. The functional sites used to coordinate Mg²⁺ are shown as sticks and colored in magenta. (A–C) Overall structure of the core domains of RdRPs from RNA viruses (PDB IDs: 2R7O, 1GX5, and 5AMR); (D) close up view of the classic arrangement of functional sites for the core domains of RdRPs from RNA viruses; (E–F) overall structure and close up view of the functional sites for the RdRP from Birnaviridae (dsRNA virus); PDB id: 2PGG; (G–H) structure model for the core domains of ZEBOV RdRP and close up view of the predicted active sites.

Multiple sequence alignment and 3D structures suggest a conserved catalytic mechanism of RdRP from RNA viruses. Two conserved Asp residues that are used to coordinate Magnesium ions in the catalytic site are in the same position in the 3D structures⁸⁹ (Fig. 4). A sequence alignment of these RdRPs (Fig. S2) allows us to predict the catalytic sites for Ebolavirus RdRPs: D632D and D742. These two positions are conserved among close homologs of Ebolavirus RdRP detected by PSI-BLAST.⁹⁰ The second conserved Asp residue immediately follows a conserved Gly residue, forming a GD motif. Another Asp residue after the GD motif also participates in coordinating Mg²⁺ in most of the templates (Fig. 4D). However, this residue is not conserved in Ebolavirus and Birnavirus RdRPs. Alternatively, Birnavirus RdRP has a Glu residue after the first conserved Asp (Fig. 4F), which is in the correct position to bind Mg²⁺. Similarly, a conserved Glu residue (634E) in the same position in the Ebolavirus RdRP may participate in Mg²⁺ binding, and the arrangement of these active site residues likely resembles that in Birnavirus RdRP.

The methyltransferase domain of protein L for mRNA capping

Addition of a 7-methylguanosine cap to the 5′ end of mRNA is essential for its subsequent translation and stability in eukaryotic cells.⁹¹ The C-terminal half of protein L is responsible for mRNA capping, and it contains an S-adenosyl-L-methionine-dependent methyltransferase domain that likely works in this process. HHpred detects several structural templates (Fig. 5) for this domain with probabilities above 95%. A sequence alignment between the Ebolavirus methyltransferase domain and the detected templates (Fig. S3) reveals that 3 residues, K1816, D1927, and K1962, are aligned to the conserved catalytic residues in the templates.⁹² In addition, the “GEGAGA” motif at positions 1836–1841 of Ebolavirus protein L is aligned to the conserved S-adenosyl-L-methionine-binding motif in the templates. This motif is also conserved in sequences from Filoviridae, suggesting a similar function in co-factor binding.

Figure 5.

Structural model and templates for the mRNA capping methyltransferase domain in Ebolavirus protein L. The structures are colored in rainbow. Equivalent secondary structure elements from different structures are colored in the same color. The co-factor, S-adenosyl-L-methionine, is shown as sticks. (A) structure model for the mRNA capping methyltransferase domain of ZEBOV; (B–E) experimental structures of other methyltransferase domains.

Interaction between Ebolavirus proteins and host proteins

RESTV causes EHF symptoms to Asian cynomolgus monkeys (Macaca fascicularis), but not human and African green monkeys (Chlorocebus aethiops).⁵ This difference in susceptibility between closely related hosts is likely due to the sequence divergence in the host proteins that interact with virus proteins. Therefore, comparing the interacting partners of virus proteins from different hosts may provide insight into how host specificity is determined and further suggest the mechanism for RESTV's loss of human pathogenicity. The known interacting partners in the host for each Ebolavirus protein are summarized in Table 1.

Table 1.

Host proteins that functionally interact with Ebolavirus proteins, and their divergence level between RESTV-susceptible and RESTV-resistant species

Name	Host protein	Functional implication	Chlorocebus vs Macaca	Homo vs Macaca
GP	NPC113,14	Receptor for the virus	6 (99.5%)	28 (97.8%)
GP	TIM-199		11 (96.5%)***	56 (80.2%)***
GP	CD209 100,101	Facilitate cell entry in specific cell types	15 (95.9%)***	31 (92.1%)***
GP	CLEC4M 100,101		Not available in Macaca and Chlorocebus
GP	CLEC10A93		5 (98.4%)	42 (86.7%)
GP	FOLR1102		3 (98.8%)	8 (96.9%)
GP	FURIN7,8	Process GP to GP1,2	1 (99.9%)	9 (98.9%)
GP	CTSB103	Process GP1,2 and initiate membrane fusion	2 (99.4%)	10 (97.0%)
GP	CTSL103		3 (99.1%)	14 (95.8%)
GP	ADAM179	Process GP1,2 to GP1,2delta	1 (99.9%)	3 (99.6%)
GP	Dynamin (multiple)104	Activates endothelial cells, reduces their barrier function	0∼1 (99.9˜100%)	2∼6 (99.3˜99.8%)
GP	ITGAV104		2 (99.8%)	9 (99.1%)
VP24	STAT161	Inhibit JAK-STAT pathway for interferon sensing	0 (100%)	5 (99.3%)
VP24	KPNA556		0 (100%)	2 (99.6%)
VP24	MAPK14 (p38)105	Prevent phosphorylation and inhibit interferon sensing	0 (100%)	1 (99.6%)
VP30	PPP1C21	Dephosphorylate VP30, control replication-transcription switch	0 (100%)	0 (100%)
VP30	PPP2C21		0 (100%)	0 (100%)
VP30	Dicer106	Antagonize RNAi machinery that could target viral RNA	5 (99.7%)	10 (99.5%)
VP30	TRBP106		2 (99.5%)	3 (99.2%)
VP35	Dicer106	Antagonize RNAi machinery that could target viral RNA	5 (99.7%)	10 (99.5%)
VP35	TRBP106		2 (99.5%)	3 (99.2%)
VP35	ILF3 (DRBP76)107	Inhibit the effect of interferon	0 (100%)	3 (99.7%)
VP35	IKBKϵ108	Block phosphorylation of IRF-3 by TBK-1 and IKBKϵ, inhibiting interferon production	4 (99.4%)	15 (97.9%)
VP35	TBK-1108		2 (99.7%)	8 (98.9%)
VP35	IRF-3108		2 (99.5%)	17 (96.0%)
VP35	PACT109	Inhibit its role as RIG-I activator	0 (100%)	0 (100%)
VP35	PKR110	Inhibit the effect of interferon	42 (92.4%)***	110 (80%)***
VP35	UBE2I111	Use SUMO E2 enzyme (UBE21) and E3 ligase (PIAS1) to modify IRF7 and inhibit its function	0 (100%)	0 (100%)
VP35	PIAS1111		0 (100%)	0 (100%)
VP35	IRF-7111		9 (98.2%)	35 (92.9%)
VP35	DLC8112	May regulate viral life cycle	1 (98.9%)	0 (100%)
VP40	Sec24C113	Virus utilize COPII vesicular transport system for life cycle	7 (99.4%)	24 (97.8%)
VP40	TSG10128	Virus uses multi-vesicular body biogenesis pathway for budding	0 (100%)	0 (100%)
VP40	ABL1114	ABL1 controls budding/release by phosphorylating VP40	5 (99.6%)	13 (98.8%)
VP40	NEDD4115	NEDD4 facilitates budding by adding ubiquitin to VP40	5 (99.5%)	28 (97.8%)
VP40	Tubulin (multiple)116	Virus utilize host cytoskeleton in its life cycle	0 (100%)	0 (100%)
VP40	Actin (1 and 2)117		0 (100%)	0 (100%)
VP40	IQGAP1118		2 (99.9%)	9 (99.4%)

^***significantly (p<0.05) elevated divergence level.

The known host proteins that interact with VP24, VP30, and VP40 are highly similar between the RESTV-resistant (Chlorocebus and human) and RESTV-susceptible species (Macaca), suggesting that they may not be responsible for RESTV's loss of human pathogenicity. In contrast, 7 most divergent host partners interact either with GP or VP35. Three of them (marked in Table 1) show significantly (P < 0.05) elevated divergence between the susceptible and resistant species, including Hepatitis A virus cellular receptor 1 (TIM-1) and pathogen-recognition receptor CD209 that interact with GP and facilitates cell entry, as well as the interferon-induced, dsRNA-activated kinase PKR that is inhibited by VP35.

The elevated divergence level for interacting partners of GP and VP35 in the host suggests that VP35 and GP may play important roles in determining host specificity. This is consistent with some indirect experimental data. RESTV GP pseudotyped viruses show significantly lower ability to infect human cells and to damage human endothelial cells than that of ZEBOV GP pseudotyped viruses.⁹³ In addition, RESTV GP shows lower ability to deplete T cells and to down-regulate interferon-stimulated gene expression compared to ZEBOV GP.^94,95 Meanwhile, ZEBOV VP35 shows stronger Interferon inhibition than RESTV VP35 in human cells.⁶⁸ However, direct studies of all RESTV proteins' effect in cells of both RESTV-susceptible and RESTV-resistant species are needed to prove our hypothesis.

Interpreting residues associated with RESTV's loss of human pathogenicity in the context of 3D structure and known functional sites

We consider positions that are associated with the loss of pathogenicity in RESTV as those that are always and significantly more similar among pathogenic species (BDBV, TAFV, SUDV and ZEBOV) than between RESTV and the pathogenic species. We referred to them as “RESTV-specific mutations.” We identified 215 such positions (Table 2), and VP30 and VP35 are significantly enriched in such mutations.

Table 2.

Positions in ZEBOV that are likely associated with the loss of human pathogenicity by RESTV

Name	UniProt ID	Length	P-value	Mutations associated with the loss of human pathogenicity
GP	Q05320	676	0.457	F31I, Q44K, V45A, E156N, S196A, L199A, S210T, Y261R, T269S, T283P, S307H, T335P, E337T, H339N, E345T, H354L, E359T, A361E, A427M, G488K, R498K, R500K, N514D, D607S, K622E, I627K, Q638H, D642L, W644L, T659I
L	Q05318	2212	0.690	V66T, E93T, Q109H, N120A, V128T, E130I, F132T, L146V, L179F, N201T, T202I, A221S, Q223L, H227Q, V229L, P262V, V263D, S274L, L283V, Y312F, A326S, T330D, S343Y, E350D, T361S, L365F, I402N, Q447H, P450S, D465N, R654H, E689S, S847A, S868A, F896Y, L925F, A954S, S995T, T1024N, R1073K, A1119S, Q1149P, S1154L, P1163A, K1171D, D1189S, A1214S, R1217K, D1237E, Q1253N, Y1322L, R1354K, T1366A, I1408M, S1436N, K1461Q, S1473C, L1488Y, S1506A, A1538S, V1562L, E1564S, T1571K, Q1608I, H1619L, L1624Y, C1628S, D1744G, E1752P, S1769G, Q1782L, R1792H, W1822L, V1850T, R1916N, K1938Q, E1941R, V1955Y, Q2024G, P2038V, S2077T, K2078G, R2079L, E2098D, Q2105L, Q2108E, Y2131F, L2157V, R2168H, R2175K, L2177F, M2186L, L2203F
NP	P18272	739	0.587	R4G, T15G, S30T, R39K, I52M, R105K, M137L, F212Y, K274R, S279A, K373R, K374R, A411L, K416N, Y421Q, D426E, D435N, Q442L, D443E, T453I, V458A, D492E, Q507S, S511I, N551R, T563S, E633L, S647K, A705R, T714Y, D716N
VP24	Q05322	251	0.932	T131S, N132T, M136L, Q139R, T226A, S248L
VP30	Q05323	288	0.010	G20P, V25S, Y39R, T52N, V53L, T63I, E93D, T96N, R98H, K107R, S111I, L116S, N117Q, A120S, Q135S, T150I, Q157R, R196H, E205D, R262A, S268Q
VP35	Q05127	340	0.019	T5L, L25T, S26T, E48D, D76E, C79Y, N80V, E85K, S92M, V97T, Q98S, S106A, A154S, T159V, E160D, G167K, S174A, I258T, E269D, A290V, A291P, V314A, Q329K
VP40	Q05128	326	0.786	M14N, T46V, P85T, A128I, G201N, F209L, Q245P, H269Q, T277Q, V323H, E325D

P-value: binomial test for enrichment of residues that may be associated with RESTV's loss of human pathogenicity in each protein.

43 of the RESTV-specific mutations can be mapped to known 3D structures of Ebolavirus proteins. None of them overlap with functional sites that are proved to be crucial by mutagenesis and 6 of them overlap with interaction surfaces (summarized in Table 3) on these structures. They may affect the binding affinities but would not likely abolish the interactions. One loop (129–141) of VP24 at the boundary of the interacting surface between VP24 and KPNA5⁵⁰ contains 4 RESTV-specific mutations (T131S, N132T, M136L, Q139R, within the red circle in Fig. 6C). These mutations may affect the binding affinity between RESTV and KPNA5 in RESTV, weakening the immune suppression by RESTV. One mutation to GP (N514D) is at the boundary of its interacting surface with neutralizing antibodies from a human survivor⁵⁵ and this mutation may affect the efficiency of the ZEBOV antibodies to antagonize RESTV.

Table 3.

Experimentally characterized functional sites in Ebolavirus proteins

Name	Residues	Function	Experimental evidence
GP	40	Glycosylated by host	N40D loss ability to infect119
GP	41–43, 503–511, 513, 514	Interact with antibody	On the interacting surface with neutralizing antibody 55
GP	51, 68, 86, 99, 109,111, 113, 122, 139, 154, 159, 161, 162, 171, 176, 183–185	Maintain the hydrophobic core structure	W86A, Y99A, Y109A, H139A, H154A, F159A, L161A, Y162A, Y171A, F176A, F183A reduce expression, reduce viral incorporation and abolish infectivity; L111A, I113A, L122A reduce viral incorporation and abolish infectivity; L51A, L68A, L184A, I185A abolish infectivity120
GP	53, 108, 121, 135, 147, 511, 556, 601, 608, 609	Disulfide bond	C53G, C108A, C121G, C135S, C147S, C511G, C556S, C601S, C608G, C609G reduce expression and abolish infectivity119
GP	55, 85, 103, 117, 178	Hydrophilic to maintain the structure	E85A, E103A, E178A reduce expression; E85A, E103A, D117A, E178A reduce viral incorporation; D55A, E103A, D117A, E178A loss ability to infect120
GP	529, 531–533, 535–537	Fusion peptide	I529A, W531R, W531A, I532R, P533R, F535R, G536R, G536A, P537R loss ability to infect15
GP	57, 63, 64, 88, 95, 170	Cell entry	L63K, L63A reduce expression; L57A, L57F, L57I, L57K, L63K, L63A, L63F, R64E, R64A, F88E, F88A, K95E, K95A, I170A, I170E loss ability to infect120
VP24	96–98, 106–121	Interact with STAT1	Show reduced hydrogen exchange rate upon binding61
VP24	113, 115, 117, 121, 124, 125, 128–131, 134–141, 184–186, 201–205, 218	Interact with KPNA5	On the interacting surface in crystal structure with KPNA5 (PDB id: 4U2X)56
VP24	50, 71, 147, 187	Adapt to new host	T50I mouse adaptation98; M71I, L147P, and T187I guinea pig adaptation97
VP30	179, 180, 183, 197	Activate transcription	Mutation to Ala reduces interaction with nucleocapsid; K180A, K183A, E197A block transcription activation59
VP30	143, 146	Phosphorylation	T143A, T143D, T146A, T146D inhibit transcription21
VP35	239, 240, 309, 312, 319, 322, 339	Bind dsRNA	K309A, K319A reduce dsRNA binding; F239A, H240A, R312A, R322A, K339A abolishes dsRNA binding70
VP35	239, 240, 309, 312, 319, 322, 339	IRF-3 inhibition	K309A, K319A reduce IRF-3 inhibition; F239A, H240A, R312A, R322A, K339A greatly reduce IRF-3 inhibition70
VP35	235, 240	Polymerase cofactor	F235A, H240A impair replication of mini-genome24
VP35	312, 322, 339	Bind DRPB76	Mutation to alanine reduce ability to bind DRPB76107
VP35	309, 312	Inhibit RNAi	K309A and R312A lost the inhibition effect121
VP35	305, 309, 312	Inhibit PKR	Mutant any 2 to alanine abolish the inhibition122
VP40	303–308	Interact with Sec24C	303–306A and 305–308A cannot interact with Sec24C, and reduce virus-like particles113
VP40	51–54, 96–101, 212–214, 286–291, 303–308, 314–316	Release of virus-like particles	51–52A, 53–54A, deletion of 96–101, K212A, L213A, R214A, 286–288A, 289–291A, 303–306A, 305–308A reduce the release of virus-like particles113,123,124
VP40	127, 129, 130, 283, 286, 293, 295, 298, 309–317	Membrane localization	K127A, T129A, N130A, P283L, P286L, I293A, L295A, V298A and deletion of 309–317 reduce membrane localization27, 125
VP40	226–255	Interaction with microtubules	Deletion of 226–240 or 241–255 abolish ability to protect microtubules from depolymerization116
VP40	213, 293, 295, 298	Penetrate membrane	Mutation to alanine reduces membrane localization126

Figure 6.

Mapping of RESTV-specific residues, functional sites and interaction surfaces to known 3D structures of Ebolavirus proteins. The structure is shown in ribbon; the functional sites are shown as sticks; and positions with RESTV-specific mutations and alternate host (rodent) adaptation residues are shown as spheres. Carbon atoms of the functional sites and sites with RESTV-specific mutations are colored to show the property of that residue: RESTV-specific surface residues are in magenta; RESTV-specific buried residues are in white; RESTV-specific residues that belong to interaction surfaces are in cyan; known functional sites are in yellow; disulfide bonded and alternate host (rodent) adaptation residues are in orange; predicted functional residues are in blue. Other atoms are colored as follows: oxygen (red); nitrogen (blue) and sulfur (orange). Circles highlight clusters of RESTV-specific residues that are discussed in the text. (A, B) GP (PDB id: 3CSY); (C, D) VP24 (PDB id: 4U2X); (E) VP30 (PDB id: 2I8B); (F) VP35 (PDB id: 3L26); (G) VP40 (PDB id: 1ES6).

Mapping of the RESTV-specific mutations to the 3D structures revealed a couple of mutation clusters in GP and VP35, which may be related to RESTV's difference in pathogenicity (Fig. 6). A first cluster is in the C-terminal subdomain of GP. The cluster consists of 3 mutations on the surface: Y261R, T269S, and S307H (inside the blue circle in Fig. 6A). The functional role of this subdomain is not clear, and the cell entry of ZEBOV is mostly mediated by the interaction between N-terminal 150 residues of GP and cell receptors like NPC1 and TIM-1. One possibility is that it may interact with other host proteins, such as lectins that facilitate the infection of Ebolavirus. In contrast, another cluster of mutations (Q44K, and V45A, inside the magenta circle in Fig. 6B) may affect the interaction between GP and the cell receptors. Even more, the mutation E156N is close to functional sites that are shown by mutagenesis to be important for maintaining the infectivity of ZEBOV. Therefore, these RESTV-specific mutations of GP may cause a significantly lower infectivity in RESTV and contribute to the loss of human pathogenicity.

RESTV-specific mutations (A290V, A291P, V314A, and Q329K) in VP35 form a cluster (inside the pink circle in Fig. 6F) on the opposite side of the dsRNA-binding surface of VP35. Host immune suppression by VP35 is mainly related to its interaction with dsRNA, but the loss of dsRNA-binding ability does not completely abolish VP35-mediated immune suppression.⁹⁶ This observation indicates the existence of other mechanisms for immune suppression by VP35, where the surface enriched in RESTV-specific mutations may play a role. One RESTV-specific mutation (T226A) is adjacent to the position in VP24 that is mutated (T50I) during adaptation to mice^97,98 (orange circles in Fig. 6D). This adaptation site is not close to any known functional sites, but the clustering of the adaptation site and RESTV-specific mutation suggests the possibility that they are at the interface of some uncharacterized interaction with other host proteins.

Funding

This work was supported in part by the National Institutes of Health (GM094575 to NVG) and the Welch Foundation (I-1505 to NVG).

Disclosure of Potential Conflicts of Interest

No potential conflicts of interest were disclosed.

Acknowledgments

We thank Drs. R. Dustin Schaeffer and Lisa N. Kinch for critical reading of the manuscript. Qian Cong is a Howard Hughes Medical Institute International Student Research fellow.

Supplemental Material

Supplemental data for this article can be accessed on the publisher's website.