Abstract
Lujo virus (LUJV) has emerged as a novel and highly fatal human pathogen. Despite its membership among the Arenaviridae, LUJV does not classify with the known Old and New World groups of that viral family. Likewise, LUJV was recently found to use neuropilin-2 (NRP2) as a cellular receptor instead of the canonical α-dystroglycan (α-DG) or transferrin receptor 1 (TfR1) utilized by Old World (OW) and New World (NW) arenaviruses, respectively. The emergence of a deadly new pathogen into human populations using an unprecedented entry route raises many questions regarding the mechanism of cell recognition and the risk that Arenaviruses are further diversifying their infection strategies. To provide the basis for combating LUJV in particular, and to increase our general understanding of the molecular changes that accompany an evolutionary switch to a new receptor for Arenaviruses, we used X-ray crystallography to reveal how the GP1 receptor-binding domain of LUJV (LUJVGP1) recognizes NRP2. Our structural data imply that LUJV is evolutionary closer to OW than to NW arenaviruses. Structural analysis supported by experimental validation further suggests that NRP2 recognition is metal ion dependent and that the complete NRP2 binding is formed in the context of the trimeric spike. Taken together, our data provide the mechanism for the cell attachment step of LUJV, the evolutionary relationship between the GP1 domain of this novel pathogen and other arenaviruses, and indispensable information for combating LUJV.
Introduction
The Arenaviridae family of viruses includes viruses that reside in rodent hosts (mammarenaviruses), producing chronic, mostly asymptomatic infections1. Through the consumption of contaminated foods or by inhaling aerosolized feces, several mammarenaviruses can infect humans and cause acute diseases, including hemorrhagic fevers with high morbidity and fatality rates1,2. These enveloped, bi-segmented RNA viruses are genetically and geographically divided to the OW mammarenaviruses, endemic to West Africa, and the NW mammarenaviruses, endemic to South and North America3. OW mammarenaviruses, including the notorious Lassa virus (LASV), utilize α-DG as a cell entry receptor4,5. Most of the NW mammarenaviruses, including the pathogenic Junín, Machupo, Guanarito, and Sabiá viruses, utilize TfR1 as an entry receptor6,7. Receptor recognition is achieved by the GP1 receptor-binding domains, which are part of the class-I trimeric spike complexes on the viral surfaces8. Following viral internalization into an endocytic compartment of the cell, the spike complexes mediate pH-dependent fusion of the viral and intracellular membranes to deliver the viral genomes into the cell cytoplasm9. The GP1 domains of both OW and NW mammarenaviruses adopt the same fold10-14, but isolated GP1 domains from OW mammarenaviruses show substantial conformational differences compared with their trimer-associated counterparts11,12,15. During cell entry, LASV dissociates from its α-DG receptor and binds lysosomal-associated membrane protein 1 (LAMP1)16. We have found that the binding to LAMP1 triggers membrane fusion by the spike complex of LASV17, and that utilizing LAMP1 is a unique mechanism for LASV12. In the case of LASV, the altered conformation that GP1 adopts is needed for LAMP1 binding11 and may have other uncharacterized functions.
Recently, LUJV was identified as a novel pathogenic mammarenavirus during a limited but highly fatal outbreak in Southern Africa18. Although LUJV has emerged in Africa, which serves as the ecological niche for most of the OW mammarenaviruses, phylogenetic analysis indicated that LUJV is distinct from the OW and NW mammarenaviruses18. A genetic screen revealed that LUJV utilizes NRP2 as a cell entry receptor and that the tetraspanin CD63 must also be present for productive cell entry19. This study also showed that the binding of LUJV to NRP2 is reduced in acidic conditions19, which is reminiscent of the LAMP1 switching mechanism in LASV. To uncover the structure of the receptor-binding module of LUJV, to reveal how LUJV recognizes NRP2, and to elucidate the evolutionary relationship between LUJV and the OW and NW mammarenaviruses, we studied the structure of LUJVGP1 in complex with NRP2 using X-ray crystallography. Analysis of the structure provided the molecular mechanism that allows LUJVGP1 to bind NRP2 and uncovered the involvement of a metal ion in mediating this recognition. Structural analysis further suggests the formation of a quaternary binding site for NRP2 in the context of the trimeric spike. The structure of LUJVGP1 implies that LUJV is evolutionarily closer to OW than to NW mammarenaviruses, but unlike the OW viruses LUJVGP1 seems to maintain its trimer-associated conformation as an isolated domain, an insight that will be important for devising strategies to combat LUJV.
Results and discussion
Determining the structure of LUJVGP1 in complex with NRP2
To reveal how LUJV recognizes NRP2, we co-expressed 6xHis-tagged versions of LUJVGP1 (residues 74-199) and the first CUB domain of NRP2 (residues 27-144) as secreted proteins using insect cells. The LUJVGP1/NRP2 complex was purified by immobilized metal affinity chromatography followed by size exclusion chromatography. The LUJVGP1/NRP2 complex formed thin needle-like crystals from which we were able to collect X-ray diffraction data to 2.44 Å resolution (Supplementary Table 1). Using the available structure of NRP220 (PDB: 2QQK) as a search model for molecular replacement (MR) in Phaser21, we were able to find 2 copies of the first CUB domain in the asymmetric unit. In contrast, none of the available GP1 structures from mammarenaviruses or homology models based on their structures were able to provide a MR solution for LUJVGP1. We hence used phases from the partial solution of NRP2 to trace LUJVGP1 into electron density maps. Initial tracing was done with Phenix.AutoBuild22, followed by manual building of the model into density-modified maps using Coot23. The final model consists of residues 87-196 for LUJVGP1 and residues 24-143 for NRP2, with two copies of the complete complex that relate to each other with translational non-crystallographic symmetry (Fig. 1a). Electron density for residues 74-86 of LUJVGP1 and residues 144-145 of NRP2 was missing and hence these residues were not modeled.
LUJVGP1 adopts a structure that resembles the trimer-associated native conformation of ‘Old World’ mammarenaviruses
Based on its glycoprotein spike complex (GPC) sequence, LUJV does not cluster with OW or NW mammarenaviruses19. LUJV further uses a distinct receptor compared to other viruses in this family19. It is thus unclear how different is the GP1 receptor-binding domain of LUJV compared to the other viruses from this family. The LUJVGP1/NRP2 crystal structure reveals that LUJVGP1 has a central β-sheet flanked by loops on one side and helices and loops on the other (Fig. 1b), in a configuration that resembles the structures of GP1s from OW11,12,15,24 and NW10,13,14 mammarenaviruses. Comparing LUJVGP1 to GP1 domains from LASV and Machupo virus (MACV) as representatives of OW and NW viruses reveals interesting structural differences (Fig. 1c). Whereas the central β-sheet is somewhat conserved, with the exceptions of a longer β-strand 6 in LUJVGP1 and the absence of an N’ terminal β-strand preceding β1 of LUJVGP1 (Fig. 1b and 1c), the relative orientations of the helices with respect to the central β-sheet are not maintained (Supplementary Fig. 2). In addition, the conformations of most of the loops that connect the secondary structure elements greatly differ. A prominent β-hairpin11, which contributes to the LAMP1 binding site on LASVGP112 but is shared by other OW mammarenaviruses that do not utilize LAMP1 during cell entry12, is missing in LUJVGP1. Instead, this hairpin is reduced in LUJVGP1 to a short loop that connects α-helix 3 with β-strand 6 similarly to GP1 domains from NW mammarenaviruses (Fig. 1b and Supplementary Fig. 1). However, LUJVGP1 is missing a disulfide bond (D3) in this region that is found in both OW and NW mammarenaviruses (Fig. 1b and 1c). Despite these differences and the low overall sequence similarity, LUJVGP1 clearly belongs to the same fold family as GP1 domains from other arenaviruses.
Interestingly, GP1 domains from OW mammarenaviruses like LASV and Morogoro virus adopt a characteristic altered conformation11,12 compared to the trimer-associated native conformation15. In the case of LASV, this altered conformation makes it compatible for binding LAMP111,17, and it may further serve for immunological evasion. These conformational changes of LASVGP1 involve the rearrangement of termini that disrupt the first β-strand as well as significant rearrangements of the helices (Supplementary Fig. 3). The termini of LASV as well as of all the GP1 structures from other mammarenaviruses solved to date are linked by a characteristic disulfide bond (D1) (Fig. 1b, 1c and Supplementary Fig. 1). In LUJVGP1 however, the termini are linked differently by a disulfide between Cys195 at the C’ terminus and Cys89 on β1. Cys89 of LUJV is not located on equivalent region as in GP1 from MACV or LASV (Fig. 1c and Supplementary Fig. 1). In the structure of LASVGP1 that was determined in the context of the native trimeric spike complex15, the cysteine that corresponds to Cys89 of LUJV is located on the extra β-strand that precedes β-strand 1 of LUJVGP1 (Fig. 1c). Likewise, the equivalent disulfide bond in MACV also links the C’ terminus of the protein to an extra β-strand at the N’ terminus. In the current crystal structure, the first 13 residues of LUJVGP1, which could potentially have formed such extra β-strand, are disordered. Although the absence of a preceding β-strand may resemble one attribute of the altered LASVGP1 conformation when isolated, the overall structure of LUJVGP1 is more similar to the trimer-associated conformation of LASVGP1 than to isolated LASVGP1 or to MACVGP1. This similarity is visually evident and further reflected by the lower RMSD (Fig. 1c and Supplementary Fig. 2). This observation implies that LUJVGP1 as an isolated domain maintains its trimer-associated native structure.
LUJVGP1 recognizes NRP2 using an intricate network of polar interactions
Two types of interfaces between LUJVGP1 and NRP2 are found in the crystal structure (Supplementary Fig. 4). One involves a flat surface on NRP2 and has a combined buried surface area (BSA) of 624 Å2(309 and 315 Å2on LUJVGP1 and NRP2, respectively). The second interface has a combined BSA of 1158 Å2 (613 and 545 Å2on LUJVGP1 and NRP2, respectively) and involves loops from both proteins that together with their side-chains form geometrically compatible interaction surfaces (Fig. 2a). We thus consider the first interface as a crystal contact and the second as the biologically relevant binding interface. On LUJVGP1, the short β-strand 3 and its flanking residues, as well as the loop that connects α-helix 2 with β-strand 4 (α2β4 loop), form most of the binding site (Fig. 2a). A major NRP2 determinant recognized by LUJVGP1 is a calcium-binding site formed by loops on the receptor surface (Fig. 2b). Though calcium binding sites were already observed in other domains of NRP220, this site on the first CUB domain was not occupied in the previously determined structure (supplementary Fig. 5). The binding of a calcium ion stabilizes the conformations of Asp127 and Glu79 from NRP2 (supplementary Fig. 5) pre-organizing them for interaction with Lys110 of LUJVGP1 (Fig. 2b). By coordinating a main-chain carbonyl of Arg130 (Fig. 2b), the calcium ion stabilizes the loop between residues 128-130 of NRP2, which is otherwise mobile (supplementary Fig. 5). The conformation of this loop is important for LUJVGP1 binding, as it positions Arg130 of NRP2 to form hydrogen bonds with two main-chain carbonyls from the LUJVGP1 α2β4 loop (Fig. 2b). This particular rotamer of Arg130 that LUJVGP1 recognizes is stabilized by a hydrogen bond between the Arg130 guanidino group and the calcium-positioned carboxylic acid of Glu79 (Fig. 2b). In the absence of a calcium ion, the local architecture of NRP2 would be disrupted (supplementary Fig. 5), which will likely prevent binding of LUJVGP1. Indeed, the first NRP2-CUB domain fused to an Fc portion of an antibody can stain HEK293 cells that express the LUJV spike complex as long as EGTA is not added to the staining solution (Fig. 2c).
The loop that stretches from Glu79 to Asp86 of NRP2 bears several charged residues that are also utilized by LUJVGP1 for binding. Both Lys80 and Asp82 of NRP2 form salt-bridges with the counter-charged Glu157 and Lys105 of LUJVGP1, respectively (Fig. 2d). The main-chain carbonyl group of Lys80 from NRP2 forms a polar interaction (likely a hydrogen bond) with Asp142 of LUJVGP1. In addition, Tyr85 and Tyr128 of NRP2 form hydrogen bonds with the main-chain amide group of Lys110 and the main-chain carbonyl group of Cys111 from LUJVGP1, respectively. Beside these polar interactions, Val139 and Thr140 from the α2β4 loop of LUJVGP1 participate in Van der Waals interactions inside a hydrophobic pocket that is formed by the aliphatic portions of some NRP2 residues like Glu79 and Glu77 (Fig. 2e). Additional hydrophobic interactions are formed by His81 of NRP2 with non-polar atoms on LUJVGP1 (Fig. 2d). An adjacent Asn107 on the surface of NRP2 makes yet another hydrogen bond with Glu161 from LUJVGP1. Altogether, the interface between LUJVGP1 and NRP2 is mostly polar.
The NRP2 binding site on LUJVGP1 spans a region that mediates α-dystroglycan recognition by OW mammarenaviruses
Previous studies provided structural and biochemical data for the recognition of α-DG and TfR1 receptors by OW and NW mammarenaviruses. Structural studies by Abraham et al.25 illustrated that the TfR1 binding site on GP1 of NW mammarenaviruses is formed on the face of the central β-sheet together with the flanking loops (Fig. 3a). By superimposing LUJVGP1 with MACVGP1 crystallized in complex with TfR1, it is evident that the NRP2 recognition site comprises a different surface of GP1 (Fig. 3a). In contrast, the NRP2-binding surface of LUJVGP1 appears to be close to residues that contribute to α-DG binding in OW mammarenaviruses (Fig. 3b). Previous studies with the α-DG-tropic OW lymphocytic choriomeningitis virus (LCMV) pointed to a few GP1 residues important for binding to its α-DG receptor26,27. These residues were later mapped on the structure of the trimeric spike complex of LASV and found to be at its apex near the trimer interface15. Interestingly, Tyr150, Asn148, and Ile254 of LASV that contribute for α-DG binding appear at the NRP2 binding site when LASVGP1 is superimposed on LUJVGP1 (Fig. 3b). Although LUJV and LASV do not share any similar sequences or local loop conformations in this particular region, LUJV has evolved to utilize the same overall region of GP1 for NRP2 recognition as OW mammarenaviruses use for binding α-DG. This observation implies that LUJV is evolutionary closer to the OW than the NW mammarenaviruses, a notion that is further supported by the higher structural similarity of LUJVGP1 to LASVGP1-NATIVE compared with MACVGP1 (Fig. 1c). In the absence of structural information for α-DG recognition by OW mammarenaviruses, the LUJVGP1/NRP2 structure may further hint about additional GP1 regions that have a potential to contribute for binding to α-DG.
NRP2 recognition in the context of the trimeric spike hints for a combined quaternary binding site
Since LUJVGP1 superimposes reasonably well with LASVGP1 in its trimeric, associated, native conformation (Fig. 1c), we utilized the trimeric structure of the LASV spike complex that was determined by Hastie et al.15, to gain insights about how NRP2 might be recognized in the context of the trimeric spike. By superimposing the LUJVGP1/NRP2 complex on the three copies of LASVGP1-NATIVE, we found that NRP2 will bind the apex of the spike complex (Fig. 4a).
Considering only the first CUB domain of NRP2, its binding angle is such that in principle up to three copies could be bound simultaneously without steric clashes (Fig. 4a). In this model, each NRP2 makes the contacts described above to one LUJV GP1 domains, but it also close to a neighboring LUJVGP1, potentially making additional contacts that are trimer dependent (Fig. 4b), as postulated for the α-DG binding of OW mammarenaviruses15 (Fig. 3b). In particular, NRP2 is in position to contact α-helix 1 and a loop connecting β-strands 4 and 5 (β4β5 loop) of the neighboring LUJVGP1. These contacts are near the glycan that is attached to Asn112 of LUJVGP1 (Fig. 1b and Fig. 4b). This glycan may restrict the access of antibodies to α-helix 1 and the β4β5 loop region as well as the β-strand 3 and α2β4 loop, which are nearby in the trimeric configuration. Although in this model the N-acetylglucosamine partially clashes with the NRP2 (Fig. 4b), a slight movement of the glycan toward the axis of the trimer will prevent such clashing. The positioning of LUJVGP1 in a trimeric configuration based on the spike complex of LASV is not accurate enough to elucidate the exact putative interactions that α-helix 1 and β4β5 loop may form with NRP2. Nevertheless, examining the contact electrostatic potential of NRP2 in the region predicted to form such interactions reveals a negatively charged patch (Fig. 4b). The surface that forms by α-helix 1 and β4β5 loop of LUJVGP1 is positively charged (Fig. 4b) and hence complementary. This observation supports the possibility that the complete NRP2 binding site forms by two neighboring LUJVGP1 molecules, helping to explain the fairly modest BSA upon complex formation that we observed in the crystal structure (Fig. 2a).
Dissociation of LUJVGP1 from NRP2 in acidic conditions is likely due to changes in NRP2
The demonstration by Raaben et al. that LUJVGP1 dissociates from NRP2 at acidic pH19 raises the question whether pH-dependent binding could be explained at the structural level. Dissociation from NRP2 is important for efficient cell entry as it presumably allows LUJV to switch to CD6319 that may act as a triggering factor for membrane fusion in a similar way to LAMP1 in the case of LASV17. Since histidine has a side-chain pKa of about 6, it is an obvious candidate for controlling pH-dependent protein-protein interactions. Of all the histidine residues in the LUJVGP1/NRP2 complex, only His81 on the surface of NRP2 is located at the interface of the complex (Fig. 5a). Interestingly, His81 is not involved in any obvious polar interaction but rather makes Van Der Waals interactions with an apolar pocket on LUJVGP1 that is flanked by positively charged residues (Fig. 2d and Fig. 5b). Potentially, a protonated and positively charged His81 could thus repel LUJVGP1, which may promote the dissociation of the proteins. However, based on our structural analysis, we propose an additional mechanism that may act in concert with His81 to break the interaction between LUJVGP1 and NRP2 during cell entry. Since the LUJVGP1 interaction with NRP2 depends on a calcium ion bound to the first CUB domain of NRP2 (Fig. 2b, 2c and Supplementary Fig. 5), losing this calcium ion during cell entry will cause LUJVGP1 to dissociate from NRP2. The fact that no calcium ion was observed bound in the previously determined structure of the full-length NRP220 indicates a limited affinity to calcium. In the early endosomes the calcium concentration drops to the low micromolar range, which is ∼1000 fold less compared to the extracellular space28 where LUJVGP1 first attaches to NRP2. Together with the acidification that can help to protonate the negatively charge groups that coordinate the calcium ion (Fig. 2b), NRP2 may thus lose its calcium upon entering early endosomes and thus further promote the dissociation of LUJV.
Overall, the structural data that we provided elucidate that LUJV utilizes a binding site formed by β-strand 3 and α2β4 loop on one monomer and may further span α-helix 1 and β4β5 loop on an adjacent LUJVGP1 in the context of the trimer. We postulate that a monoclonal antibody that will target any of these sites on LUJVGP1 in the context of the trimeric spike would block binding to NRP2 and thus may neutralize the virus. Hence, we propose to focus efforts to elicit such antibodies as a way to provide potential therapeutics against LUJV.
Methods
Expression and purification of recombinant proteins
GP1LUJV coding DNA (residues 74-199) was amplified from chemically synthesized LUJV GPC (Genscript). The gene was subcloned into the pACgp67b expression vector (BD Biosynthesis) to include an N-terminal 6x-His tag. Neuropilin 2 (NRP2) binding site fragment (residues 27-146) was chemically synthesized with C-terminal 6x-His tag, and was subcloned into the pACgp67b vector. Both GP1LUJV and NRP2 were co-expressed as a secreted proteins using the baculovirus system as we previously described11. Cell media was collected, clarified using centrifugation and buffer exchanged to TBS (20 mM Tris-HCl pH 8.0, 150 mM sodium chloride) using a tangential flow filtration system (Millipore). LUJVGP1/NRP2 complex was captured using a HiTrap IMAC fast-flow Ni+2 (GE Healthcare) affinity column, and further purified using size exclusion chromatography with Superdex 75 10/300 column (GE Healthcare) in TBS buffer. Fc-fused NRP2 was expressed in HEK293 cells adapted to suspension cells (Expression Systems). Transfections were done using linear 25 kDa polyethylenimine (PEI) (Polysciences) at 1 mg of plasmid DNA per 1 L of culture at cell density of 1 M/ml. Media were collected after 5 days of incubation and supplemented with 0.02% (wt/vol) sodium azide and PMSF. Fusion proteins were isolated using protein-A affinity chromatography (GE Healthcare).
Crystallization
We used vapor diffusion in sitting drops method for crystallization screens. For crystallization experiments we used a Mosquito® crystallization robot (TTP labs) to set 60, 120, and 180 nL drops of protein with 120 nL reservoir of commercially available crystallization screens. Initial crystallization hits for LUJVGP1/NRP2 were identified using PEGRx HT™ (Hampton) screen. Crystallization conditions were manually optimized. We further identified using additive screen HT™ (Hampton) the tri-peptide glycyl-glycyl-glycine as an additive that improves crystals’ morphology. Crystals for diffraction experiments were obtained by mixing the LUJVGP1/NRP2 complex solution at 12 mg/mL with 0.03M glycyl-glycyl-glycine, 25.9% PEG 6000, 0.09M Bis-Tris Propane pH 9.5 at 20° C. Crystals were briefly soaked in mother liquor solution supplemented with 25 % ethylene glycol for cryo preservation before flash cooling in liquid nitrogen.
Data collection, structure solution and refinement
X-ray diffraction data were collected at the European Synchrotron Radiation Facility (ESRF) at beamline ID23-2 using a Pilatus 3 2M detector. Diffraction data were collected to a resolution of 2.4 Å. Images were indexed, integrated, and scaled using Xia229 pipeline that made use of aimless30, CCP431, Dials32, and Pointless33. A molecular replacement solution using the first CUB domain of NRP220 (PDB: 2QQK) was found using Phaser21. We used Phenix.AutoBuild22 to start tracing LUJVGP1 and subsequently extend the model to completion using manual building in Coot23.
Cell stain
HEK293T cells were seeded on poly-L-Lysine pre-coated cover slips in 24-well plates and transfected with a plasmid (pcDNA3) encoding LUJV GPC using PEI-MAX (polysciences) reagent. At 48 h post transfection cells were fixed with pre-warmed 3.7% formaldehyde (PFA) solution in PBS and blocked with 3% BSA in PBS. Cover slips were incubated with NRP2-Fc diluted in 1% BSA-PBS at a concentration of 30 μg/ml, with or without addition of 5 mM EGTA (Sigma), followed by staining with Cy3-conjugated anti-Human Fc (Jackson) and mounting with mounting media supplemented with DAPI (GBI Labs). Cells were imaged at ×10 magnification and images were processed using ImageJ34.
Accession code
Atomic model for the LUJVGP1/NRP2 complex as well as structure factors were deposited to the protein data bank under accession code 6GH8.
Author contributions
H.C.D. together with I.K. produced, purified and crystallized the LUJVGP1/NRP2 complex. H.C.D. and R.D. collected diffraction data. R.D. solved and analyzed the structure and wrote the manuscript with the help of H.C.D and I.K.
Acknowledgments
Diffraction experiments were performed in beamline ID23-2 at the European Synchrotron Radiation Facility (ESRF), Grenoble, France. We are grateful to Chloe Zubieta at the ESRF for providing assistance in using beamline ID23-2. We thank professor Randy Read from University of Cambridge for his invaluable advises and contribution in analyzing our crystallographic data. We thank professor Deborah Fass for providing critical comments and suggestions. Ron Diskin is incumbent of the Tauro career development chair in biomedical research. Research in the Diskin lab is supported by a research grant from the Enoch Foundation, a research grant from the Abramson Family Center for Young Scientists, a research grant from Ms. Rudolfine Steindling, by the Minerva Foundation with funding from the Federal German Ministry for Education and Research, and by a grant from the Israel Science Foundation (grant No. 682/16).