Correlating the ability of VP24 protein from Ebola and Marburg viruses to bind human karyopherin to their immune suppression mechanism and pathogenicity using computational methods

Introduction

Viruses from the family Filoviridae are negative-stranded RNA viruses having a filamentous shape¹. The first member of this family (Marburg) was discovered in 1967², while the Ebola virus was first discovered in 1976³. Public attention has been drawn to this rare, but deadly disease⁴ ever since the current outbreak in West African countries threatened to rapidly deteriorate into a full-blown epidemic^5,6. Both viruses cause haemorrhagic fever by quickly suppressing innate antiviral immune responses⁷. However, quite surprisingly, the Reston Ebola (REBOV) strain, first identified in monkeys that were imported into Reston in the United States from the Philippines⁸, is non-pathogenic in humans^9,10.

Previously, we have characterized α-helical (AH) structures in Ebola proteins using PAGAL¹¹, and demonstrated that the AHs with characteristically unique feature values are involved in critical interactions with host proteins¹². We showed that the AH from Ebola virus membrane fusion subunit GP2¹³, which is disrupted by a neutralizing antibody derived from a human survivor of the 1995 Kikwit outbreak¹⁴, has a very large hydrophobic moment compared to other AHs in Ebola proteins¹². Similarly, another AH with the highest proportion of negatively charged residues is the binding site of the human karyopherin (KPNA) to the Zaire Ebola (ZEBOV) virus VP24 (ezVP24) protein¹⁵.

In spite of sharing a common ancestry¹⁶, Marburg and Ebola have different antigenicity of the virion glycoprotein¹⁷. Furthermore, the mechanism of immunosuppression is different in these viruses¹⁸. These differences are probably the reason for the reduced mortality observed in Marburg outbreaks. In Ebola, the crucial role of host immune system evasion is accomplished by two proteins: VP35 and VP24¹⁹. Ebola VP24 inhibits interferon (IFN) signaling by hindering the nuclear accumulation of tyrosine-phosphorylated STAT1 by binding KPNA^20,21. In contrast, the Marburg virus abrogates the host immune response by inhibiting IFN-induced tyrosine phosphorylation of STAT1 and STAT2¹⁸ via a moonlighting function matrix protein, VP40²². Specifically, ezVP24 binds KPNA via two AHs (α5 and α6)¹⁵. In Marburg VP24 (mVP24), α5 has distinctively different properties (not easily identified by a sequence or structural alignment), while α6 is just a small turn²³. This explains why mVP24 is not immunosuppressive.

We investigated these AHs in VP24 from the REBOV strain (erVP24). While α5 in erVP24 was similar to that in ezVP24, α6 in erVP24 had different properties caused by the presence of a serine in the place of arginine (S140R). We modeled the apo erVP24 (PDBid:4D9OA) using the ezVP24 in complex with KPNA as a template (PDBid:4U2X) by SWISS-MODEL²⁴, and then docked KPNA to this structure using DOCLASP²⁵. The docked structure helped visualize the ability of Arg140 in ezVP24 to make the correct electrostatic interaction with two glutamic acids, one residing on α5 in VP24, and the other in KPNA. The effect of single mutations in modulating virulence has been well established^26–28. However, our methodology provides a more rational way of finding such critical residues. The possibility of a REBOV mutant gaining immunosuppressive capabilities is particularly disconcerting since the isolation of the REBOV strains from pigs^29–31. We also highlight the possibility of using α5 and α6 from VP24 as epitopes for generating antibodies³² or designing compounds and peptides to inhibit protein-protein interaction³³.

Materials and methods

AHs in proteins were identified using DSSP³⁴. These AHs were then analyzed using PAGAL¹¹. Briefly, the Edmundson wheel is computed by considering a wheel with centre (0,0), radius 5, first residue coordinate (0,5) and advancing each subsequent residue by 100 degrees on the circle, as 3.6 turns of the AH makes one full circle. We compute the hydrophobic moment by connecting the center to the coordinate of the residue and giving it a magnitude obtained from the hydrophobic scale obtained from³⁵. These vectors were then added to calculate the final hydrophobic moment. The color coding for the Edmundson wheel was as follows: all hydrophobic residues were colored red, while hydrophilic residues were colored in blue: dark blue for positively charged residues, medium blue for negatively charged residues and light blue for amides.

The protein structures used in the current work were all identified using the PDBid, and are available at www.rcsb.org. We used the SWISS-MODEL program to model the erVP24 (PDBid:4D9OA) structure using the ezVP24 (PDBid:4U2XA) in complex with KPNA as template. See 4D9OA4U2XA.pdb in Dataset 1 Note the residue numbering is not conserved by SWISS-MODEL. For example, Glu113 in PDBid:4D9OA corresponds to Glu97 in PDBid:4D9OA4U2XA. We used DOCLASP²⁵ to dock KPNA to the modelled structure of erVP24 (See Pymol script ‘dockingKPNAtoRestonVP24.p1m’ for human KPNA and ‘RESTONVP24mouse.p1m’ for mouse KNPA in Dataset 1). ‘4U2XA.4U2XD.maxdist.out.sort’ in Dataset 1 lists the closest atoms of the residues of VP24 (PDBid:4U2XA) that make contact with human karyopherin (PDBid:4U2XD), sorted based on distances.

All protein structures were rendered by PyMol (http://www.pymol.org/). The sequence alignment was done using ClustalW³⁶. The alignment images were generated using SeaView³⁷. Protein structures were superimposed using MUSTANG³⁸.

Results and discussion

Differences in α5 in Ebola and Marburg viruses: explaining why Marburg VP24 is not immunosuppressive

ezVP24 has a 39.6% identity (73.8% similar) with mVP24 (Figure 1a), and there is significant structural homology among VP24 proteins from different strains of Ebola and Marburg (Figure 1b). Yet, the mechanism of immune response suppression is different in these viruses from the Filoviridae family¹⁸. ‘Reasons why Marburg virus VP24 is not immunosuppressive remain elusive’²³. Therefore, we sought to investigate the differences in residues involved in binding KPNA in the ezVP24 and mVP24.

Figure 1. Sequence and structural homology between VP24 proteins from different strains of Ebola and Marburg.

(a) EbZaire: Zaire Ebola, EBSudan: Sudan Ebola, EBReston: Reston Ebola, Mar-Musoke: Marburg Musoke. Multiple sequence alignment was done using ClustalW. Note, that the numbering used by ClustalW is not consistent with the real numbering of the VP24 residues. (b) Structural alignment of PDBid:4M0QA (Ebola Zaire Apo, in red), PDBid:4U2XA (Ebola Zaire complexed, in green), PDBid:4D9OA (Ebola Reston Apo, in blue), PDBid:3VNEA (Ebola Sudan Apo, in yellow) and PDBid:4OR8A (Marburg Musoke Apo, in orange). Structural alignment was done using MUSTANG³⁸. (c) Helices involved in binding human karyopherin (α5 and α6 in magenta). Note, that the α5 is not a helix in Marburg VP24 (PDBid,4OR8A, in orange), but just a small turn.

ezVP24 binds KPNA via two AHs (α5 and α6), residues on loops and a Lys on a β-sheet (Table 1). In mVP24, α5 has different properties (Figure 2a,b and Table 2), while α6 is just a small turn (Figure 1c). These differences in the properties of AHs involved in binding KPNA in eVP24 to those in mVP24 strongly indicates that mVP24 is not immunosuppressive, as is widely accepted¹⁸ (or at least it does not use the same mechanism).

Table 1. Residues in Ebola Zaire VP24 (ezVP24, PDBid:4U2XA) that make contact with human karyopherin (PDBid:4U2XD).

One or more atoms from these residues are within 4 Å of residues from human karyopherin.

Residues in ezVP24 (PDBid:4U2XA)	Secondary structure
GLU/113,GLY/117,LEU/121,ASP/124,TRP/125	α5
THR/129,THR/131, PHE/134,ASN/135,MET/136,ARG/137,THR/138	loops
GLN/139,ARG/140,VAL/141	α6
GLN/184,ASN/185,HIS/186,LEU/201,GLN/202,GLU/203,PRO/204,ASP/205	loops
LYS/218	β9

Table 2. Properties of α5 in VP24 proteins from different strains of Ebola and Marburg.

It can be seen that the Marburg VP24 (mVP24) protein has a distinctly different charge residue composition in the helix. This strongly indicates that mVP24 might not bind human karyopherin, which is the mechanism of immunosuppression by the Ebola VP24 proteins. HM: Hydrophobic moment, RPNR: Ratio of the positive to the negative residues, Len: length of the helix, NCH: number of charged residues.

PDB.Helix	Description	Len	HM	RPNR	NCH
4M0QA.α5	Ebola Zaire Apo	13	2.2	0	1
4U2XA.α5	Ebola Zaire in complex with KPNA	16	4.4	0	2
4D9OA.α5	Ebola Reston Apo	15	3	0	1
3VNEA.α5	Ebola Sudan Apo	14	4.1	0	1
4OR8A.α5	Marburg Apo	16	4.9	0.7	3

Figure 2. Edmundson wheel for α5 of VP24 in ZEBOV strain (eZVP24), Marburg (mVP24) and REBOV (erVP24) viruses.

The color coding for the Edmundson wheel is as follows: all hydrophobic residues are colored red, while hydrophilic residues are colored in blue: dark blue for positively charged residues, medium blue for negatively charged residues and light blue for amides. (a) Apo ezVP24 (PDBid:4M0QA). (b) Apo mVP24 (PDBid:3VNEA). It can be seen that mVP24 has two positively charged residues in the AH, unlike eZVP24. (c) ezVP24 (PDBid:4U2XA) in complex with human karyopherin (PDBid:4U2XD). Note, that Glu113 and Pro114 are now part of the AH, in contrast to the apo AH in (a). (d) Apo erVP24 (PDBid:4D9OA).

S140R substitution in α6 may explain why Ebola Reston strain is non-pathogenic in humans

The REBOV strain ‘does not represent an immediate public health menace on the scale of the African Ebola virus’⁹, possibly due to the generation of antibodies against this strain³⁹. Also, gene expression study of infected cells showed that the ZEBOV and Marburg viruses has fewer activated IFN-inducible genes relative to REBOV⁴⁰. Thus, most likely, the REBOV strain does not have the same immunosuppressive capabilities as the ZEBOV or Sudan strain. While α5 of erVP24 has properties similar to ezVP24 (Figure 2c), α6 in REBOV VP24 (erVP24) is clearly different in hydrophobic moment and residue composition (Figure 3). For example, Arg140 in ezVP24 is replaced with Ser140 in erVP24.

Figure 3. Edmundson wheel for α6 of VP24 in ezVP24, esVP24 and erVP24 viruses.

(a) apo ezVP24 (PDBid:4M0QA). (b) ezVP24 in complex with humans karyopherin (PDBid:4U2X). Note, that the AH is extended by two residues (E143 and Q144) as compared to the apo protein. However, the hydrophobic moment remains the same. (c) α6 of esVP24 (PDBid:3VNEA). (d) α6 of erVP24 (PDBid:3VNEA). It can be seen REBOV VP24 has a different hydrophobic moment than the other, since Ser140 is place of Arg140.

To better visualize and quantify this difference, we docked KPNA to erVP24. First, we modelled the apo erVP24 (PDBid:4D9OA) using the ezVP24 complexed with KPNA (PDBid:4U2X) using SWISS-MODEL²⁴. Subsequently, KPNA was docked to this protein using DOCLASP²⁵.

Figure 4 shows ezVP24 and erVP24 docked to KPNA. In ezVP2, KPNA binding is primarily facilitated by electrostatic attraction between the negatively charged Asp124 in α5 and Lys481 in KPNA (at 3.9 Å)¹², and a hydrogen bond between Arg140 (α6) and Glu475 of KPNA (among other hydrogen bonds, Table 3). Also, the ezVP24 itself is stabilized by an electrostatic bond between the negatively charged Glu113/OE1 (α5) and the positively charged Arg140/NH1 (α6) at 3.4 Å. Note, that this pair is at distance of 12.8 Å in the apo ezVP24 (PDBid:4M0QA). This 8Å conformational change in these AHs emphasizes the role of plasiticity in binding KPNA. In contrast, in the erVP24, the distance between Glu113/OE1 and Ser140/OG changes from 14 Å in the apo enzyme to 6.2 Å in the docked model. Also, the Ser140/OG atom is not positively charged unlike Arg140/NH1. Further, the possibility of Ser140/OG making a hydrogen bond with Glu475 of KPNA is remote, since they are 6.7 Å apart. Thus, we conclude that the mutation R140S is likely to be one of the critical factors for the non-pathogenic nature of REBOV, since this mutation renders erVP24 incapable of binding KPNA. Other factors might include the different susceptibilities of the glycoproteins of ZEBOV and REBOV for furin cleavage⁴¹.

Figure 4. Docking human karyopherin (KPNA) to erVP24.

The erVP24 was modelled using SWISS-MODEL²⁴ using ezVP24 structure complexed with KPNA (PDBid:4U2XA) (See 4D9OA4U2XA.pdb in Dataset 1). The docking was done using DOCLASP²⁵, which superimposes the proteins as well. (a) Superimposition of modelled erVP24 and ezVP24, with bound KPNA. (b) Electrostatic attraction between the negatively charged Glu113/OE1 (α5) and the positively charged Arg140/NH1 (α6) at 3.4 Å, and a hydrogen bond between Arg140 (α6) and Glu475 of KPNA stabilizes the binding. (c) Ser140 replaces Arg140 in erVP24, and fails to make any of the above interactions.

Table 3. Atoms from ZEBOV VP24 (ezVP24) that are closest to the human karyopherin (KPNA) in PDBid:4U2X.

The complete sorted list can be found in ‘4U2XA.4U2XD.maxdist.out.sort’ in Dataset 1. Note, that there is a hydrogen bond between Arg140/NH2 and Glu475/O.

ezVP24 atom	KPNA atom	Distance (Å)
THR/138/OG1	ASP/480/OD2	2.7
ASN/185/ND2	ASP/431/O	2.7
ASN/185/OD1	ARG/398/NH1	2.8
THR/138/N	ASP/480/OD2	2.9
ARG/140/NH2	GLU/475/O	3.0

Docking mouse KPNA to erVP24

We used KPNA from (Mus musculus (mouse) (PDBid:1Y2AC, 50.3% identity and 77.8% similar) to compare the binding of VP24 to KPNA from another related species⁴². Although, REBOV is pathogenic in non-human primates, there are no known structures for KNPA in other primates (See blastkpna.png in Dataset 1). Figure 5 shows the sequence alignment, the superimposed proteins and the mouse KPNA docked to erVP24 using DOCLASP (See RESTONVP24mouse.p1m in Dataset 1). Note, that the interacting residues (Glu475 and Lys481) are conserved. The fact that erVP24 is not immunosuppressive for mouse is further substantiated by a recent study that noted viral replication in all rodents tested, but disease progression occurs only in STAT1 knockouts⁴³. Note, that erVP24 is able to directly bind STAT1 at levels similar to VP24 from other species²¹. However, apparently this binding is not sufficient to inhibit the IFN signalling pathway⁴³. Thus, VP24 and its ability to bind KPNA plays a major role in the ‘Reston-pathogenicity puzzle’⁴⁴. Several putative sites, including a ‘cluster of Reston-specific residues in VP24 is L136, R139 and S140’, have been identified using deuterium exchange mass spectrometry methods⁴⁴. Our computational method, with its associated caveats, identifies the S140 residue as being more critical than the other sites.

Figure 5. Docking mouse karyopherin (KPNA) to erVP24.

The erVP24 was modelled by SWISS-MODEL²⁴ using ezVP24 structure complexed with KPNA (PDBid:1Y2AC) (See RESTONVP24mouse.p1M in Dataset 1). The docking was done using DOCLASP²⁵, which superimposes the proteins as well. (a) Sequence alignment of human and mouse KPNA, showing that the interacting residues are conserved. (b) Superimposition of human (in cyan) and mouse (in wheat) KPNA done using MUSTANG. (c) Docked mouse KPNA (in wheat) to erVP24 (in limegreen). Interacting residues of mouse KNPA residues (Glu475 and Lys481) make similar contact to erVP24.

Conclusions

The ability of a single mutation to significantly alter the immunosuppressive properties of the Ebola proteins is well established^26,27,48. Sequence-based methods (whole genome profiling) are typically used to identify these critical mutations²⁶. Structural studies provide an alternative, and possibly more rational, method to identify such mutations. For example, while double (and not single) mutations are required in VP35 to inhibit protein kinase R activation, it is difficult to rationalize this based on sequence data only²⁸. In the current work, we build on previous work that characterized AH structures in Ebola proteins to rationalize the lack of immunosuppressive properties in the mVP24. ezVP24 binds to KPNA via two AHs (α5 and α5), loops and a residue on a β-sheet. We attribute the lack of immunosuppressive properties of mVP24 to its inability to bind KPNA, which emanates from different characteristics of mVP24 α5 compared to ezVP24 α5. Subsequently, we demonstrate that a single mutation in α6 in the erVP24 might endow it with immunosuppressive properties. We corroborate this conclusion by modelling the apo structure of the erVP24 based on the structure of ezVP24 in complex with KPNA using SWISS-MODEL²⁴, and by docking KPNA to the modelled structure using DOCLASP²⁵. The REBOV strain, first identified in monkeys and imported into the United States from the Philippines⁸, has never caused disease in humans^9,10. However, the isolation of the REBOV strains from pigs in the Philippines^29,30, and recently in China³¹, highlights the significance of finding preventive therapies in the probable scenario that a mutant REBOV for VP24 with immunosuppressive capabilities gets transferred to human handlers. Such a difference does not exist in the VP35 protein, where REBOV VP35 has been used as a model to show how they could silence and sequester double-stranded RNA, which is a key event in immunosuppression⁴⁹. We also reiterate the potential of using these AHs from VP24 as epitopes^50,51 for generating antibodies^32,52,53, or innovating drugs to inhibit protein-protein interaction^33,54–58. The presence of two intrinsically disordered residues proximal to these AHs in the apo structure that gain a AH structure upon binding should encourage antibody search to use both apo and complexed AHs. It is certainly worth investigating whether supplementing ZMapp, a cocktail of three antibodies that has shown reversion of advanced Ebola symptoms in non-human primates⁵⁹, with more antibodies would prove more effective.

Data availability

F1000Research: Dataset 1. Version 2. Data used for SCALPEL search methodology to identify plant alpha helical - antimicrobial peptides in the PDB database.

10.5256/f1000research.5666.d40354⁶⁰