Cryo-EM structure of Chikungunya virus in complex with the Mxra8 receptor

SUMMARY

Mxra8 is a receptor for multiple arthritogenic alphaviruses that cause debilitating acute and chronic musculoskeletal disease in humans. Herein, we present a 2.2 Å resolution X-ray crystal structure of Mxra8 and 4 to 5 Å resolutioncryo-electron microscopy reconstructions of Mxra8 bound to chikungunya (CHIKV) virus-like particles and infectious virus. The Mxra8 ectodomain contains two strand-swapped Ig-like domains oriented in a unique disulfide-linked head-to-head arrangement. Mxra8 binds by wedging into a cleft created by two adjacent CHIKV E2-E1 heterodimers in one trimeric spike and engaging a neighboring spike. Two binding modes are observed with the fully mature VLP, with one Mxra8 binding with unique contacts. Only the high-affinity binding mode was observed in the complex with infectious CHIKV, as viral maturation and E3 occupancy appear to influence receptor binding site usage. Our studies provide insight into how Mxra8 binds CHIKV and creates a path for developing alphavirus entry inhibitors.

In Brief sentence

Insights into how Chikungunya virus recognizes the entry receptor, Mxra8, paves the way for developing inhibitors to treat alphavirus infection.

Graphical Abstract

Basore et al describe an X-ray crystal structure of Mxra8, an entry receptor for arthritogenic alphaviruses, and cryo-EM structures of Mxra8 bound to CHIKV. Mxra8 has an unusual domain architecture with a head-to-head arrangement, with domain 2 emanating from the loops of domain 1. Mxra8 binds into a complex quaternary cleft formed between two E2-E1 heterodimers within a trimeric spike, while also contacting E1 from a neighboring spike. Mxra8 occupancy is lower when E3 protein is retained on the virion.

INTRODUCTION

Alphaviruses are positive-sense, single-stranded, enveloped RNA viruses and are among the most important arthropod-borne viruses causing disease in humans (Powers et al., 2001). This genus includes chikungunya (CHIKV), Mayaro (MAYV), O’nyong’nyong (ONNV), and Ross River (RRV) viruses, which are emerging beyond their historical boundaries and now cause debilitating acute and chronic polyarthritis affecting millions of people in Africa, Asia, Europe, and the Americas. Despite their epidemic potential, there are no specific therapies or licensed vaccines for any alphavirus infection.

Alphavirus genomes encode four non-structural and five structural proteins. The non-structural proteins are required for virus replication, protein modification, and immune evasion. The structural proteins (capsid (C)-envelope (E)3-E2–6K-E1) are synthesized from a subgenomic promoter and cleaved co- and post-translationally. The E1 envelope glycoprotein participates in cell fusion (Lescar et al., 2001), whereas the E2 envelope glycoprotein binds to entry factors (Smith et al., 1995; Zhang et al., 2005) and initiates clathrin-dependent endocytosis (DeTulleo and Kirchhausen, 1998; Lee et al., 2013; Ooi et al., 2013). The E3 protein is essential for the proper folding of p62 (precursor to E2) and the formation of the p62-E1 heterodimer (Carleton et al., 1997; Mulvey and Brown, 1995) but is cleaved by furin-like proteases during the maturation process in the trans-Golgi network (Heidner et al., 1996). Mature alphaviruses are ~700 Å diameter icosahedral particles that assemble at the plasma membrane and contain a lipid bilayer with 240 embedded E2-E1 heterodimers assembled into 80 trimeric spikes with T=4 icosahedral symmetry (Cheng et al., 1995; Kostyuchenko et al., 2011; Paredes et al., 1993), and a nucleocapsid containing a single copy of genomic RNA.

Crystallographic studies of the precursor p62-E1, the mature E2-E1 glycoprotein complex, and the E1 protein (Lescar et al., 2001; Li et al., 2010; Roussel et al., 2006; Voss et al., 2010) have elucidated the glycoprotein structures. Several alphavirus virions structures also were described by cryo-electron microscopy (cryo-EM) (Chen et al., 2018; Cheng et al., 1995; Hasan et al., 2018; Kostyuchenko et al., 2011; Lee et al., 1998; Li et al., 2010; Pletnev et al., 2001; Zhang et al., 2011; Zhang et al., 2002). The E1 ectodomain consists of three β-barrel domains termed Domain I (DI), DII, and DIII. The fusion peptide is located at the distal end of DII. E1 monomers lie at the base of the surface spikes and form trimers surrounding the icosahedral axes. E2 localizes to a long, thin, leaf-like structure on the top of the spike. The mature E2 protein contains three domains with immunoglobulin (Ig)-like folds: the N-terminal domain A, located at the center; domain B at the lateral tip; and the C-terminal domain C, located close to E1 and the viral membrane.

We recently used a genome-wide CRISPR/Cas9-based screen to identify mouse Mxra8, a two immunoglobulin (Ig)-like domain containing cell adhesion molecule, as a receptor for multiple arthritogenic alphaviruses including CHIKV, RRV, MAYV, and ONNV (Zhang et al., 2018). Importantly, the human ortholog, MXRA8, also bound to CHIKV and other alphaviruses, and its cell surface expression was required for efficient infection of primary human target cells including fibroblasts, skeletal muscle cells, and chondrocytes. Mxra8 bound directly to CHIKV particles and enhanced attachment and internalization into cells, and Mxra8-Fc fusion protein or anti-Mxra8 monoclonal antibodies (mAbs) blocked CHIKV infection of several cell types. Administration of Mxr8a-Fc protein or anti-Mxra8 blocking mAbs to mice reduced CHIKV or ONNV infection and associated joint swelling. Despite defining several biological characteristics of CHIKV interaction with this receptor, structural insight as to how Mxra8 engages the alphavirus spike proteins on the virion is lacking.

To date, there has been limited structural information of the binding of receptors to icosahedral enveloped viruses. Here, we describe the X-ray crystal structure of the Mxra8 ectodomain and cryo-EM reconstructions of CHIKV virus-like particles (VLPs, produced as capsid-E3-E2–6K-E1 but lacking viral RNA) and fully infectious CHIKV in complex with Mxra8. Mxra8 has an unusual architecture, as its two Ig-like domains are oriented in a disulfide-linked head-to-head arrangement, with domain 1 inserted into a loop of domain 2. Mxra8 binds into a cleft formed between two CHIKV E2-E1 heterodimers within a trimeric spike and also extends to engage an adjacent trimeric spike. Two binding modes for Mxra8 were observed with mature CHIKV VLPs, which is consistent with a high and low affinity binding site model supported by surface plasmon resonance measurements. The low affinity binding sites, however, were sterically obscured by the retention of E3 on infectious CHIKV. Overall, our structural analysis defines how CHIKV engages its receptor Mxra8 to facilitate attachment and infection of cells. This information may inform the basis of therapies and improved vaccine designs that mitigate disease of multiple emerging alphaviruses.

RESULTS

Mxra8 protein and CHIKV VLPs

The soluble ectodomain of murineMxra8 (amino acids 23 through 296) was produced as E. coli inclusion bodies that were oxidatively refolded (Fig S1A). Size exclusion chromatography coupled with multi-angle light scattering (SEC-MALS) analysis reveals that refolded Mxra8 is monomeric with a molecular weight of 31 kDa (Fig S1B). Each Mxra8 monomer contains three disulfide linkages as indicated by a monoisotopic molecular weight of 31,092.45 determined by electrospray mass spectrometry. This value corresponds to the predicted weight of the native expression construct, minus the first methionine removed by aminopeptidase and six hydrogens lost during formation of the three disulfide bonds. We also produced a mammalian-expressed soluble form of Mxra8 by engineering a Mxra8-Fc fusion protein (Zhang et al., 2018) with a human rhinovirus 3C (HRV) protease site between the C-terminus of Mxra8 and the Fc domain. After production of Mxra8-Fc fusion protein in HEK-293F cells and purification by protein A affinity chromatography, we cleaved the Fc domain with HRV, and isolated Mxra8 in the flow-through of a second round of protein A agarose affinity chromatography (Fig S1C). SEC-MALS analysis revealed that cleaved mammalian cell-generated Mxra8 is monomeric with a molecular weight of 38 kDa (Fig S1D), and N-terminal sequencing using Edman degradation showed that the mature N-terminus of the protein begins at position 23 (Fig S2). These preparations of Mxra8 were used for the crystallographic and cryo-EM studies described below.

Previous studies have obtained high-resolution structural information about how neutralizing antibodies engage CHIKV using VLPs, which contain the structural proteins but lack infectious viral RNA and can be imaged under lower biosafety conditions (Long et al., 2015; Sun et al., 2013). To begin to define the structural basis for interaction between Mxra8 and arthritogenic alphaviruses (Zhang et al., 2018), we produced CHIKV VLPs (strain 37997) after transient transfection of HEK-293 cells with a plasmid encoding C-E3-E2–6K-E1 (Akahata et al., 2010); equivalent preparation of VLPs were used in phase 1 and 2 human clinical trials for vaccine protection against CHIKV disease ((Chang et al., 2014), NCT02562482 and NCT03483961). Soluble VLPs were collected, buffer exchanged, and monitored for purity by SDS-PAGE and antigenicity by Western blotting with anti-capsid and anti-E2/E1 antibodies (Fig S1E and F). Dynamic light scattering experiments revealed a particle diameter of ~680 Å (Fig S1G), which is close to the expected external diameter of 700 Å for a CHIKV virion (Cheng et al., 1995).

X-ray crystallographic structure of Mxra8

Diffraction quality crystals of bacterially produced murine Mxra8 (ectodomain residues 23 to 296) were obtained in space group I222 with cell dimensions a = 72.59 Å, b = 142.85 Å, and c = 195.59 Å. There are two copies of Mxra8 in the asymmetric unit. The structure was solved by molecular replacement using junctional adhesion molecule-like protein (JAML; PDB code 3MJ6) as a model. There was visible density for all but the first nine and last two residues of chain A, and all but the first seven and last two residues of chain B. The final statistics for the Mxra8 model, refined to a resolution of 2.2 Å, are provided in Table S1. Mxra8 consists of two Ig- V-like domains arranged in a headto-head orientation with the CDR-like loops pointing towards one another. The two FG-loops (CDR3-like) are connected by a disulfide bond (Fig 1A–C). Unexpectedly, the first two β-strands of each domain are swapped such that the full domain 1 is seen as an insertion into the BC-loop (CDR1-like) of domain 2. Domains 1 and 2 are similar in sequence (Fig S2) and nearly identical in fold, with an average root-mean-square deviation (RMSD) of 0.90 Å over 68 Cα positions for chain A and 1.13 Å over 67 positions for chain B in the crystal structure. In comparison, the two domain 1 structures in the asymmetric unit align with a RMSD of 0.4 Å over 110 residues, whereas the two domain 2 structures align with an RMSD of 1.0 Å over 104 residues. The hinge connecting the two domains is flexible, and comparison of the two independent Mxra8 structures indicates that domain 1 can move by at least ~43° relative to D2 (Fig S3)(Hayward and Berendsen, 1998).

Figure 1. Mxra8 crystal structure and topology diagram.

A. Ribbon model of the mouse Mxra8 protein structure. The β-strands are labeled following standard convention. The two Ig-like domains are colored from the N- to C-terminus in a rainbow spectrum of blue to red with the disulfides shown as yellow ball and stick bonds. The positions of the cysteines forming the interdomain disulfide bond are indicated. The N- and C-termini are labeled in lowercase. B. Secondary structure diagram of Mxra8, numbered by sequence positions in the X-ray structure. Colored arrows indicate the β-strands as in panel A. Yellow lines connect the cysteine positions forming disulfide bonds. The AB strand swaps anchor the connections between the D1 and D2 domains. C. Cartoon schematic of Mxra8, with labeled D1 and D2 domains, N- and C-termini, BC linkers, and cysteine residues forming the interdomain disulfide bond. Mxra8 is bowed. A rotation of ~155˚ around the hinge region is required for alignment of the two domains. See Fig S1, S2, S3, and Table S1.

Cryo-EM reconstruction of Mxra8 bound to CHIKV VLPs

Electron micrographs of CHIKV VLPs with or without bound mammalian cell-derived Mxra8 were recorded using a 300 kV Titan Krios cryo-electron microscope with Gatan K2 detector at the Washington University Center for Cellular Imaging (WUCCI) (Fig 2A and Table S2). The images were corrected for beam-induced motion using MotionCor2 (Zheng et al., 2017). Particles were auto-picked from the micrographs and subjected to two-dimensional classification to remove ice contamination and debris (Fig 2B). The remaining particles underwent ab initio three-dimensional reconstruction and refinement in cisTEM (Grant et al., 2018). The reconstructions of CHIKV VLP alone and with Mxra8 were determined at overall resolutions of 4.16 Å and 4.06 Å (Fig 2C and D), respectively, based on Fourier Shell Correlation (FCS) analysis (Fig 2E and F).

Figure 2. Two-dimensional cisTEM analysis of CHIKV particles with or without Mxra8 bound.

A. Raw electron micrographs of CHIKV VLPs. B. Two-dimensional classification scheme for binning of CHIKV VLPs. C-D. Two-dimensional equatorial slices of CHIKV VLP alone (C) or CHIKV VLP in complex with Mxra8 (D). Dimensions: the outer radius of the nucleocapsid shell (~180 Å), lipid bilayer (~240 Å), E1 protein glycoprotein shell (~280 Å), E2 protein spike (~340 Å), and bound Mxra8 (~350 Å) from the viral center. E-F. Fourier shell correlation (FSC) plot versus resolution for CHIKV VLP alone (E) or CHIKV VLP with bound Mxra8 (F). See Fig S1 and Table S2.

The three-dimensional reconstructions showed no large conformational changes in the structure of the CHIKV VLP when Mxra8 is bound (Fig 3A and B). Despite the different chemical environments of the structural glycoproteins within the asymmetric unit of the particle, Mxra8 binds E2-E1 heterodimers in a 1:1 ratio, with 240 Mxra8 molecules bound per virion. The local resolution of the reconstruction of CHIKV VLP-Mxra8 complex indicates that the membrane proximal regions of the E2-E1 heterodimers were the best resolved, with an estimated resolution of 3.9–5.4 Å for E1 ectodomains. In comparison, the radially distant trimeric spikes and Mxra8 were resolved less well, with an estimated resolution of 4.2–8.0 Å for Mxra8 chains (Fig 3C and F). Mxra8 binds to the CHIKV VLP with a complex quaternary epitope, wedging into a cleft created by two E2-E1 heterodimers in one trimeric spike, effectively wrapping around the distal end of one E2-E1 heterodimer (herein referred to as “wrapped”), and contacting an adjacent heterodimer within the same trimeric spike (the “intraspike” heterodimer). Mxra8 also engages another heterodimer on a neighboring spike (the “interspike” heterodimer”) (Fig 3D and E).

Figure 3. Cryo-EM reconstruction of CHIKV particles with or without Mxra8 binding.

A-C. Paired, colored surface representations (top panel) and equatorial cross-sections (bottom panel) of CHIKV VLP (A), CHIKV VLP + Mxra8 (B), and local resolution of CHIKV VLP + Mxra8 (C). The white triangle indicates one icosahedral asymmetric unit. The 5-fold (i5), 3-fold (i3), and 2-fold (i2) icosahedral axes of symmetry are indicated with a pentagon, triangles, and oval, respectively. Trimeric spikes are labeled “i3” if coincident with the i3 axes, and “q3” if on a quasi 3-fold axes. Black arrows: directions of icosahedral symmetry axes (i2, i3, q3, and i5). (A-B) Radial distance color scheme: red, electron dense core and RNA; yellow, capsid; green, membrane lipid; dark blue, E1; cyan, E2 spike; and magenta, Mxra8. D. Zoomed-in view of the q3 spike, highlighting the three E2-E1 heterodimers that interact with Mxra8 at site 3: the wrapped (light grey), the intraspike (medium grey), and the interspike (dark grey) heterodimers. The i5 and i2 axes of symmetry are labeled with a pentagon and oval, respectively. Mxra8 is colored by domain (D2, dark magenta; D1, light magenta). E-F. Side views of Mxra8 at site 3. (E) Mxra8 and the wrapped, intraspike, and interspike heterodimers are colored to match panel D. (F) Mxra8 and the wrapped, intraspike, and interspike heterodimers are colored by local resolution, as reported in panel C: red, 4 Å; yellow, 6 Å; and green, 8 Å; See Table S2 and S3.

Building and refinement of cryo-EM structural models

We iteratively built an atomic model of the CHIKV VLP structural proteins and Mxra8 using a combination of published CHIKV crystal structures and de novo modelling. As a starting point, we used the structures of the CHIKV capsid (PDB: 5H23), CHIKV p62-E1 (PDB: 3N42), modeled transmembrane domains of E1 and E2 from Venezuelan equine encephalitis virus (VEEV) (PDB: 3J0C), and our structure of Mxra8 (Fig 1) to build one assembly into the cryo-EM map of CHIKV VLP bound to Mxra8. Neither chain A nor B of the crystallographic model of Mxra8 initially docked well as one rigid body into the cryo-EM density map (Fig S3); instead, chain A D1 and D2 were docked as two separate rigid bodies and the hinge regions connecting them were re-built manually using COOT (Emsley et al., 2010). These coordinates then were used to build the T=4 asymmetric unit surrounded by the adjacent subunits in neighboring asymmetric units. This process was essential to ensure accurate modeling of all interaction interfaces. This model underwent manual and computational real-space refinement employing COOT and PHENIX (Afonine et al., 2012) (see Methods) (Fig 4A and B, Table S3).

Figure 4. Atomic model of Mxra8 interaction with CHIKV.

A. The refined model of Mxra8 and CHIKV structural proteins (E1, E2, transmembrane (TM) helices, and Capsid) in the electron density map of the CHIKV VLP with Mxra8 reconstruction at threshold σ=1, viewed from the side (left panel) and top (right panel) of the asymmetric unit. B. Cartoon of the asymmetric unit viewed from the side (left panel) and top (right panel) with all domains labeled. “q3” and “i3” refer to the icosahedral and quasi 3-fold axes, respectively. C. Individual E2-E1 subunits at the binding interface, specifically site 1 Mxra8 and its wrapped heterodimer (left panel), site 1 Mxra8 with its intraspike heterodimer (middle panel), and site 3 Mxra8 with its interspike heterodimer (right panel). Mxra8 and structural proteins are colored by domain. Mxra8: dark magenta, D2; light magenta, D1. E1: DI, light grey; DII, medium grey; DIII, dark grey; fusion loop, orange; TM region, black. E2: A domain, light cyan; β-linker, medium cyan; domain B, dark cyan; domain C, medium blue; TM region, dark blue. Capsid, forest green. Domain labels are boxed if the domain is at the binding interface. Disulfide bonds are shown as yellow balls and sticks. D. Ribbon model of the refined site 1 Mxra8 in its electron density map, with the N- to C-terminus in a rainbow spectrum of blue to red and the disulfide bonds shown as yellow balls and sticks. The N- and C-termini are labeled in lowercase, and the β-strands are labeled in uppercase. The density map is viewed at contour level =1.7. E. Zoomed-in view of the protrusion density and atomic structure of the N-linked glycan at residue 118, viewed at contour level =1.4. The value is the standard deviation of density values above the mean in the map. See Fig S2, S3, S6and Table S4.

Our model details the domains and residues of the CHIKV E2-E1 heterodimers at the Mxra8 binding interface. At the primary contacted wrapped heterodimer in all four unique environments, Mxra8 engages the A and B domains of E2 (residues 18, 26–29, 71–72, 74–76 119–121, 123, 178–182, 189, 191, 193, 212–214, 221–223), the fusion loop in E1 (83, 85, 87, 88–91), and domain II of E1 (223, 226–227). On the intraspike heterodimer, Mxra8 interacts with domain A and the -linker of E2 (residues 5–6, 62, 64, 144, 150, 157–160, 263–265, 267). Mxra8 at site 1 also makes unique contacts to domains I and II of E1 (residues 37, 130, 132, 142–143, 145–147, 152, 154, 156–157, 263–265, 267); these interactions were not observed at sites 2, 3, and 4. In contrast, at all binding positions except at site 1, Mxra8 makes additional contacts with domain II of E1 (residues 71–74, 112, 115, 206–207, 209–212) of the interspike heterodimer (Fig 4C, Fig S4, S5, and S6, Table S4). The average overall binding interface at each of the four sites in an asymmetric unit is ~2,100 Ų. The largest contributing interface is between Mxra8 and the wrapped heterodimer (~1,200 Ų), followed by the intraspike heterodimer (~600 Ų), and the interspike heterodimer (~300 Ų).

Mxra8 residues at the binding interface also were identified in our model. Both D1 and the hinge region contribute to the binding of the wrapped heterodimer (residues 56–57, 63–66, 71, 73, 82–86, 89, 91–99, 104, 137, 139, 141, 144, 146, 148, 247, 250–251). To the intraspike heterodimer, E2 is contacted by amino acids in D1 and the hinge region of Mxra8 at all binding sites (residues 60, 62–63, 112, 114–115, 190, 193–196, 198, 200, 227–229), and E1 is contacted by only D2 residues of Mxra8 at binding site 1 (46, 230–232, 236–237, 239–240, 260, 262). To the interspike heterodimer, only D2 amino acids of Mxra8 at binding sites 2, 3, and 4 are at the binding interface (residues 45, 208, 210–213, 217, 219, 235–237, 241, 262–264, 266, 290–291) (Fig 4C, Fig S2, and Table S4).

To evaluate our docking and modeling of Mxra8, we first assessed the general fit of the atomic coordinates and searched for key features in the electron density map. The additional β-strand (termed “H” strand) and its adjacent A strand uniquely fill the “flap” density of D1 (Fig 4D). There is a predicted N-linked glycosylation site in the ectodomain of Mxra8 at residue 118. At this position in our model, a clear density protrudes (Fig 4E). The relative size and shape of this density is similar at each of the four sites in the asymmetric unit. Additional weak density is seen at the C-terminus of Mxra8 at sites 2, 3, and 4; this allowed us to model 10 additional residues at these sites past the G strand of D2 that is beyond the register of the crystal structure.

In support of our model, biolayer interferometry (BLI) competition binding experiments revealed that Mxra8 cannot bind to CHIKV VLPs in the presence of a subset of previously described anti-Mxra8 mAbs that inhibit infection (Zhang et al., 2018) (Fig 5A). Epitope mapping via hydrogen-deuterium exchange mass spectrometry analysis shows that competing mAbs bind to Mxra8 at its interface with CHIKV, whereas non-competing mAbs bind at other sites (Fig 5B and S7A–B). We observed competitive inhibition with mAbs 4E7.D10 and 8F7.E1, which map to an epitope in D1 of Mxra8 in residues adjacent to the E2 A domain of the wrapped heterodimer, and mAb 1G11.E6, whose epitope in D2 residues is adjacent to the E1 fusion loop of the wrapped heterodimer. In contrast, mAbs 1H1.F5 and 3G2.F5, which do not compete in the BLI assay, map to predominantly solvent-exposed peptides in D1 near the interspike heterodimer. If the orientation of the Mxra8 Ig-like domains were inverted, the epitope of mAbs 1H1.F5 and 3G2.F5 would be inaccessible due to wrapped heterodimer contacts with E2 domain B and the E1 fusion loop. The alternate domain-flipped model also is not consistent with the observed ability of mAb 1G11.E6 to compete for virus binding (Fig S7C–D).

Figure 5. Binding competition, epitope mapping, and mutagenesis supporting the Mxra8-CHIKV cryo-EM model.

A. Competition BLI assay traces. CHIKV VLP was captured on the biosensor with an anti-CHIKV mAb and then dipped into wells containing Mxra8 ectodomain alone or with an anti-Mxra8 mAb or isotype control mAb. One experiment of three is shown. B. Surface and ribbon diagrams of Mxra8 at site 3 and the E2-E1 domains at its binding interface. Structural proteins are colored by domain. E1: DI, light grey; DII, medium grey; fusion loop, orange. E2: A domain, light cyan; β-linker, medium cyan; domain B, dark cyan. ‘ denotes domains within the intraspike heterodimer, and “ for the interspike heterodimer; the wrapped heterodimer is labelled without symbols. Surface representation of Mxra8 models are colored with the HDX-mapped epitopes of anti-Mxra8 mAbs, where shades of violet correspond to mAb 1G11.E6, green for mAbs 1H1.F5 and 3G2.F5, and yellow for mAbs 4E7.D10 and 8F7.E1. C. Trans-complementation of 293T cells (which lack endogenous MXRA8) with charge-reversal mutated Mxra8. (Top) CHIKV infection was determined flow cytometric analysis of intracellular E2 protein. (Bottom) Relative surface expression of Mxra8 by flow cytometry using anti-Mxra8 mAbs. Data are pooled from three experiments performed in triplicate, normalized to the WT controls, and the mean values are shown (one-way ANOVA with Dunnett’s post-test: *, P < 0.05; ****, P < 0.0001). D. Ribbon diagram of Mxra8 model at site 1 with side chains of mutated residues shown as balls and sticks. Mutations that result in statistically significant decreases in CHIKV infection are colored. Inset, zoomed-in view of the surface representation of Mxra8 at the interface. Side chains of residues of E2 (cyan ribbon) at this interface are displayed as balls and sticks. E. Surface representation of all CHIKV structural protein domains at the interface of Mxra8 at site 3 (shown as ribbon diagram), colored by E2-E1 heterodimer: wrapped, light grey; intraspike, medium grey; interspike, dark grey. CHIKV protein residues are colored magenta if they lose over 30% solvent surface area upon Mxra8 binding as calculated by PDBePISA (http://www.ebi.ac.uk/pdbe/pisa/). F-G. Zoomed-in view of D1 of Mxra8. (F) CHIKV E2 residues identified as epitopes for neutralizing human anti-CHIKV mAbs (Long et al., 2015; Smith et al., 2015) that also block binding to Mxra8 (Zhang et al., 2018) are labeled and colored blue. (G) Four CHIKV E2 residues at the CHIKV-Mxra8 interface as defined by previous alanine scanning mutagenesis data (Zhang et al., 2018) are labeled and colored green. See Fig S2, S4, S5, S7, and Table S4.

As another test of the model, charge-reversal point mutations were introduced in Mxra8 at wrapped heterodimer interface residues with high percentages of buried surface area as well as other control residues not predicted to be part of the binding interface (Fig S2). WT and mutant Mxra8 cDNAs were transfected into 293T cells, which lack expression of endogenous human MXRA8 (Zhang et al., 2018), and tested for their surface expression and ability to support CHIKV infection. Several mutations (e.g., A49R, V88R, A93R, G117R, and F119R) in Mxra8 affected surface expression and were not analyzed further. For mutations with wild-type levels of Mxra8 surface expression, three substitutions (D89R, G94R, and R97E) resulted in marked reductions in CHIKV infection and two other residues (V98R and Y99R) showed smaller yet reproducible decreases (Fig 5C). All five of these residues are at the interface of domain A of E2 of the wrapped site (Fig 5D). Thus, the binding competition and mapping results support the identified residues at the Mxra8 interface, as well as the positioning of D1 and D2 in our electron density map.

For the interface on CHIKV (Fig 5E), many of the identified E2 contact residues are supported by coincident antibody mapping and mutagenesis data (Fig S5, Table S4). Several neutralizing human mAbs against CHIKV that disrupt Mxra8 binding (Zhang et al. 2018) map to epitopes in the A domain of E2 (Long et al., 2015; Smith et al., 2015) that are shared with the Mxra8 binding site (Fig 5F). Previous alanine scanning mutagenesis mapping data identified four solvent-exposed residues in domain A of E2 as important for optimal Mxra8 binding (Zhang et al., 2018); Mxra8 residues D71, T116, and I121 are at the wrapped interface whereas W64 contacts Mxra8 at the intraspike interface in our model (Fig 5G). In the original study, mutations in E1 were not evaluated for binding to Mxra8 because it was expected that E2 and not E1 contributed to receptor engagement (Strauss et al., 1994; Tucker and Griffin, 1991).

Modes of Mxra8 binding to CHIKV VLPs and infectious virus

We also generated a 4.99 Å cryo-EM reconstruction of CHIKV infectious particles with Mxra8 and built an atomic model (Fig 6A, Tables S2 and S3). In contrast to the CHIKV VLP, strong electron density was seen for the E3 protein in all four chemical environments of the infectious CHIKV (Fig 6B–C), which similarly was observed in other infectious and non-infectious alphavirus particle preparations (Wu et al., 2008; Yap et al., 2017; Zhang et al., 2011). In the CHIKV infectious particles, weaker electron density was seen for Mxra8 compared to the CHIKV VLP reconstruction. Only one of the four potential binding sites in the asymmetric unit showed interpretable Mxra8 density (Fig 6B). This suggested there might be a lower occupancy of Mxra8 binding on the infectious particle compared to on the fully mature VLP, potentially due to steric hindrance by the residual E3 protein in three of four chemical environments. To test this hypothesis, we docked in the difference map of Mxra8 from the mature Mxra8-VLP complex onto our Mxra8-bound CHIKV infectious virus reconstruction and assessed for clashes. E3 appears to sterically obstruct Mxra8 binding at the i3 (site 4) and two of three of the q3 binding sites (sites 2 and 3), allowing for 60 binding sites per virion at site 1 (Fig S8). We note that electron density for Mxra8 on the CHIKV VLP is strongest at site 1, which corresponds to the Mxra8 binding site on the infectious virion (Fig 6C). This finding is consistent with our binding analysis of Mxra8 to CHIKV VLPs by surface plasmon resonance, where the raw traces did not fit well to kinetic and equilibrium 1:1 binding models (Fig 6D). One explanation for the higher occupancy of Mxra8 at one site in an asymmetric unit is a site-specific higher binding affinity, due to the slightly different chemical environments of each of the four E2-E1 heterodimers within the asymmetric unit. We generated a model assuming one high affinity site and three lower but equal affinity sites (see Methods); this model produced a substantially better fit to our SPR binding data (Fig 6E). Mxra8 binds the high affinity site with a kinetically derived K_D of ~84 nM and the lower affinity sites with a K_D of ~270 nM. Similar SPR measurements and two-site binding analysis were performed for human monomeric MXRA8 (as determined by SEC-MALS), which is 78% identical to the murine protein (Fig S2 and Fig S9).

Figure 6. Modes of Mxra8 engagement by CHIKV.

A. Two-dimensional equatorial slice (top panel) and Fourier shell correlation (FSC) plot (bottom panel)of CHIKV infectious particles in complex with Mxra8. B-C. Views of asymmetric units of CHIKV infectious particles (B) or CHIKV VLP (C) with bound Mxra8 electron density at high contour (left panel) and low contour (middle and right panels), colored by structural proteins if within 6 Å of docked model coordinates. Mxra8 coordinates were removed from the model, and E3 was docked in for (B). The value is the standard deviation of density values above the mean. Color scheme: grey, E1; cyan, E2; yellow, E3; capsid, green; Mxra8 and other unexplained density, magenta. D-E. Kinetic sensograms and steady-state analysis of murine Mxra8 binding to CHIKV VLPs fit to a 1:1 binding model (D) or two-site model (one high affinity site and three lower but equal affinity sites) (E). Raw experimental traces are in black, fit traces are in red. Inset, Scatchard plot (4 experiments; mean, standard error of the mean (SEM), and χ² values). See Fig S8 and S9.

DISCUSSION

Cell culture infection experiments with mouse and human cells and in vivo pathogenesis studies in mice defined Mxra8 as a cell surface receptor required for optimal infectivity and induction of musculoskeletal disease by multiple arthritogenic alphaviruses (Zhang et al., 2018). Here, our X-ray crystallography and single particle cryo-EM analyses of Mxra8, VLPs, and infectious virus provide structural insight into how CHIKV engages Mxra8 to facilitate interactions with target cells. Our study adds to the limited structural knowledge of how receptors bind to enveloped icosahedral virions. Only one other structure of an enveloped, icosahedral virus complexed with its receptor exists. In a 25 Å resolution cryo-EM structure, Dengue virus (DENV) was complexed with the carbohydrate recognition domain of DC-SIGN; the only contacts were with protruding N-linked glycans in domain II of the E protein (Pokidysheva et al., 2006). In contrast, we observed a complex network of quaternary protein-protein interactions with Mxra8 engaging two E2-E1 heterodimers within one trimeric spike as well another heterodimer on a neighboring spike. The specific binding determinants we observed are supported by our previous mutagenesis analysis of E2, structure-guided mutations that we introduced into Mxra8, and epitope maps of mAbs against CHIKV and Mxra8 that directly block virus-receptor interactions. Our structures indicate that Mxra8 can bind at four distinct sites in the icosahedral asymmetric unit of the CHIKV VLP but only one site in the infectious virus, which retains E3.

The quaternary interactions formed between Mxra8 and multiple envelope proteins would effectively cross-link CHIKV spikes in a manner analogous to a previously defined broadly neutralizing mAb (CHK-265) that binds domain B on one trimer and domain A on an adjacent spike (Fox et al., 2015). The cross-linking of the viral structural proteins by Mxra8, while facilitating attachment and entry, might create a conundrum for viral fusion in the endosome, which requires domain B on E2 to undergo a substantive conformational shift to expose fully the underlying hydrophobic fusion loop in domain II of E1 (Li et al., 2010; Voss et al., 2010). After clathrin-mediated endocytosis, fusion of CHIKV occurs within the acidic environment of early endosomes (Hoornweg et al., 2016). Although further studies are required, we speculate that some of the Mxra8 binding interactions with CHIKV E1 and E2 may be sensitive to acidic pH, such that upon transiting to the early endosome the cross-linked trimers can separate and allow the structural transitions required for fusion to occur. In preliminary experiments, we observed deceased affinity of binding of Mxra8 to recombinant pE2-E1 under conditions of mildly acidic pH (K. Basore, unpublished results). It is plausible that the strength of Mxra8 interactions with viral proteins may regulate the stage of endocytosis and pH of fusion of some arthritogenic alphaviruses.

The alphavirus structural polyprotein is processed in the endoplasmic reticulum to yield E3-E2 (p62) and E1, which form heterodimers and oligomerize as trimers to generate the immature spike (Uchime et al., 2013). In the Golgi network, furin-like proteases cleave E3 from E2 (yielding E2-E1) to render the spikes optimally fusogenic. However, the cleavage and dissociation event can be variable in a cell-type and virus-type specific manner, as E3 remains covalently or non-covalently associated with the mature virus in some alphaviruses, including Sindbis virus, Semliki Forest virus, and VEEV (Heidner et al., 1996; Zhang et al., 2011; Ziemiecki and Garoff, 1978). The comparison of our cryo-EM structures of Mxra8 bound to CHIKV VLPs and infectious virus reveals a difference in stoichiometry of binding, with 240 Mxra8 proteins bound to CHIKV VLP and only 60 bound to our infectious CHIKV particles. One reason for the decreased occupancy on the infectious CHIKV is the retention of the E3 protein, which appears to occlude Mxra8 binding at three of four chemical environments in the asymmetric unit. Biochemical and structural analysis confirmed that E3 was absent from our CHIKV VLP preparation, possibly because of the mildly alkaline buffers used in the chromatography purification steps. Our binding studies with soluble mouse or human Mxra8 and CHIKV VLPs defined two classes of sites, one of high affinity (~60 nM) and a second of lower affinity (~300 nM). In comparison, infectious CHIKV had only the high-affinity binding site. These data suggest that while Mxra8 can bind mature CHIKV virions in two binding modes with full occupancy, regions of partial maturity that retain E3 will bind in only a single mode. At present, the contribution of the low affinity binding site to infectivity remains unclear, although these same infectious virus preparations still showed a strong Mxra8-dependence (Zhang et al., 2018). It is plausible that E3 retention on some arthritogenic (e.g., SINV) alphaviruses could result in a lack of contribution of Mxra8 to entry and infection. Future studies with fully mature, infectious alphaviruses propagated in cells over-expressing furin (Mukherjee et al., 2016) that lack E3 might address this hypothesis directly.

Mxra8 contains two strand-swapped Ig-like domains oriented in a unique disulfide-linked head-to-head arrangement withdomain 1 essentially being an insertion into the BC loop region of domain 2. Despite this unusual architecture, the order and placement of the β-strands in each Mxra8 domain closely resemble the topology of a typical V-set Ig domain, with the closest match (Holm and Laakso, 2016) being the coxsackievirus and adenovirus receptor (Seiradake et al., 2006). Each Mxra8 domain appears as a two β-sheet sandwich containing a conventional cysteine bridge between strands B and F, and also containing C’ and C” strands. However, some differences exist. In both domains of Mxra8, the A strands occur only in one face, along the G strand (AGFCC’C”). Domain 1 contains an additional β-strand not normally found in Ig-like domains. This H strand allows the A and B strands to complete domain 1 although not in the normal order (i.e., after the G strand). In canonical Ig-like domains, the BC loop passes from one face of the β-sandwich to the other. In Mxra8, the A and B strands are swapped between domains, and the BC loops pass between the two domains to serve as interdomain linkers. The only similar arrangement we could find is an artificially engineered Ig-like monomer that homodimerized by swapping the A and B strands to pack in a head-to-head fashion (Hu et al., 2007).

Our cryo-EM map of Mxra8 bound to CHIKV is corroborated by mutational analysis and coincidence mapping of antibodies that blocked Mxra8 or CHIKV binding. We previously identified solvent-accessible residues W64, D71, T116, and I121 in the A domain of E2 as essential for optimal Mxra8 engagement by using an E2 alanine-scanning mutagenesis library (Fox et al., 2015; Smith et al., 2015; Zhang et al., 2018). Our cryo-EM analysis identified three of these amino acids as direct interface residues (W64, D71, I121) with one additional residue (T116) positioned in close proximity to Mxra8. When charge-reversal point mutations were introduced in Mxra8 residues at the wrapped heterodimer interface (D89, G94, R97, V98, Y99), reductions in CHIKV infection were observed. In further support of the structurally determined Mxra8 binding site on CHIKV, neutralizing human mAbs recognizing epitopes in the A domain inhibited Mxra8 binding, whereas others localizing to distinct sites had less effect (Zhang et al., 2018). The epitopes of the anti-CHIKV mAbs (Smith et al., 2015) that completely blocked Mxra8 binding directly overlap the structural binding sites of Mxra8 and CHIKV E2. Our epitope mapping of a panel of anti-Mxra8 mAbs and experiments assessing their ability to block virus binding provides further support for the cryo-EM atomic model.

Mxra8 is a receptor for multiple arthritogenic but not encephalitic alphaviruses (Zhang et al., 2018). An alignment of amino acid sequences corresponding to CHIKV residues comprising the Mxra8 binding site reveals that determinants are generally conserved among arthritogenic alphaviruses (44% conserved) but are more divergent in encephalitic alphaviruses (14% conserved) (Fig S4 and S5), which likely explains the negligible interaction or binding requirement for infectivity of Mxra8 with VEEV or related encephalitic alphaviruses (Zhang et al., 2018). Cryo-EM structures with other alphaviruses including ONNV, MAYV, and RRV should provide further insight into the critical E2, E1, and Mxra8 contacts that facilitate attachment, entry, and infection. Such analysis could identity targets on either the viral structural proteins or Mxra8 to facilitate the development of agents capable of disrupting these interactions. This approach could form the basis of a therapy that mitigates infection by multiple arthritogenic alphaviruses.

STAR Methods

Contact for reagent and resource sharing

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact author Daved H. Fremont (fremont@wustl.edu).

Experimental models

Viruses

CHIKV strain 181/25 was obtained from the World Reference Center for Emerging Viruses and Arboviruses (S. Weaver and K. Plante, Galveston, TX). The virus was propagated in Vero cells (ATCC) in DMEM supplemented with 10% FBS, concentrated by sucrose gradient ultracentrifugation, and titrated by standard focus-forming assays (Brien et al., 2013; Pal et al., 2013).

VLP production and purification

CHIKV VLPs were produced via transient transfection of C-E3-E2–6K-E1 (CHIKV strain 37997) plasmid DNA (Akahata et al., 2010) into HEK293 cells (obtained from Vaccine Research Center, NIH), and purified via Q Sepharose XL (GE Healthcare, GE17–5072-01) anion chromatography. The peak of interest was diafiltered into a buffer containing 218 mM sucrose, 10 mM potassium phosphate, and 25 mM citrate, pH 7.2. The material was sterile-filtered using a 0.2 μM filter, 500 μL aliquots were made and stored at −80°C. The final material was analyzed by BCA assay for protein concentration, Coomassie-stained SDS-PAGE gel with densitometry, and Western blotting with rabbit polyclonal anti-CHIKV 181/25 (04–0008, IBT Bioservices) and a secondary horseradish peroxidase conjugated goat anti-rabbit antibody (65–6120, ThermoFisher Scientific).

Mxra8 protein generation and purification

To generate recombinant bacterially-derived mouse Mxra8 protein, residues 23–296 (Accession number NM_024263) were codon optimized, synthesized, and inserted into the pET21a expression vector. After sequence verification, Mxra8 was expressed in BL21(DE3) E. coli cells. Cell were grown to an OD₆₀₀ of 0.8, induced with 1 mM IPTG for 5 h at 37°C, resuspended in 50 mM Tris-HCl, 1 mM EDTA, 0.01% NaN₃, and 1 mM DTT (TEND) buffer supplemented with 25% sucrose, and lysed in 50 mM Tris-HCl, 1 mM EDTA, 0.01% NaN₃, 1 mM DTT, 200 mM NaCl, 1% sodium deoxycholate and 1% Triton X-100. Inclusion bodies were isolated and washed extensively in TEND buffer with 100 mM NaCl and 0.5% Triton X-100 and then TEND buffer alone. Inclusion bodies (~200 mg) were denatured in 6M guanidinium hydrochloride, 100 mM Tris-HCl, and 20 mM β-mercaptoethanol. Denatured protein was oxidatively refolded overnight at 4°C in 100 mM Tris-HCl, 400 mM L-arginine, 5 mM reduced glutathione, 0.5 mM oxidized glutathione, 10 mM EDTA and 200 mM phenylmethylsufonyl flouride. Protein was concentrated using a 10 kDa molecular weight cut-off stirred cell concentrator (EMD Millipore) and purified by HiLoad 16/600 Superdex 75 size exclusion chromatography (GE Healthcare). Mxra8 mutants were generated using a QuikChange II XL Site-Directed Mutagenesis Kit (Agilent) and verified by DNA sequencing. Mxra8 mutant proteins were expressed and purified as described above.

To generate mammalian Mxra8 protein, residues 23–336 (Accession number NM_024263) were codon optimized, synthesized, and inserted into the pCDNA3.4 vector upstream of the HRV 3C cleavage sequence (LEVLFQGP) and the mouse IgG2b Fc region. After sequence confirmation, Mxra8-HRV-Fc protein was expressed using the Expi293 mammalian expression system (Thermo Fisher). Briefly, 200 μg of plasmid was diluted in Opti-MEM, incubated with HYPE-5 transfection reagent (OZ Biosciences), and added dropwise to cells (10⁶ cells/ml). Transfected cells were supplemented daily with Expi293 media and 2% (w/v) Cell Boost Supplement (HyClone). Four days post transfection, supernatant was harvested by centrifugation and purified by Protein A Sepharose (Thermo Fisher) chromatography. The mouse IgG2b region was cleaved using the Pierce HRV 3C Protease Solution Kit (Thermo Fisher). Cleaved Fc fragments were depleted using Protein A Sepharose chromatography, and HRV-cleaved Mxra8 was purified by HiLoad 10/600 Superdex 75 size exclusion chromatography (GE Healthcare). Purity was determined by SDS-PAGE analysis.

Multi-angle light scattering analysis

Recombinant Mxra8 proteins were loaded onto a size exclusion chromatography column in sizing buffer (150 mM NaCl, 20 mM HEPES pH 7.4, 0.01% NaN₃) in series with a Dawn HELEOS-II 18-angle light scattering detector (Wyatt), Optilab rEX refractive index monitor (Wyatt), and photodiode array detector 996 (Waters). The light scattering, refractive index change, and UV light absorbance were measured, and the molecular weight of the eluted protein was calculated using the Astra V macromolecular characterization software package (Wyatt).

Mxra8 crystallography and structural analysis

Mouse Mxra8 was crystallized by hanging drop vapor diffusion at 15.0 mg/ml in 100 mM Tris-HCl, pH 8.5, and 8% PEG 8000. Crystals were looped and flash cooled in mother liquor supplemented with 25% ethylene glycol. Data was remotely collected at Advanced Light Source MBC Beamline 4.2.2 and merged and processed with XDS (Kabsch, 2010). Molecular replacement was performed using Phaser in the Phenix software (Adams et al., 2010) and junctional adhesion molecule-like protein (JAML; PDB code 3MJ6) as the search model. Refinement and model building were carried out in Phenix and COOT (Emsley et al., 2010). Data collection and refinement statistics are reported in Table S1. Structures were assessed using Molprobity and figures were generated using Chimera (http://www.cgl.ucsf.edu/chimera) and PyMOL (Bramucci et al., 2012).

SPR-based Mxra8 binding assay

SPR binding experiments were performed on a Biacore T200 system (GE Healthcare) to measure the kinetics and affinity of Mxra8 binding to CHIKV VLPs. Experiments were performed at 30 μl/min and 25°C using HBS-EP (0.01 M HEPES pH 7.4, 0.15 M NaCl, 3 mM EDTA, 0.005% v/v Surfactant P20) as running buffer. CHK-265 mAb (Pal et al., 2013) was immobilized onto a CM5 sensor chip (GE Healthcare) using standard amine coupling chemistry, and CHIKV VLPs were captured. Recombinant Mxra8 was injected over a range of concentrations (2 μM to 20 nM) for 200 sec, followed by a 600 sec dissociation period. As a negative control, murine norovirus was captured with mAb A6.2, as described previously (Nelson et al., 2018). Real-time data was analyzed using BIAevaluation 3.1 (GE Healthcare). For the equal affinity binding sites model, kinetic profiles and steady-state equilibrium concentration curves were fitted using a global 1:1 binding algorithm with drifting baseline. For the model of Mxra8 binding to 1 of 4 sites with higher affinity, the following fits were used:

Equilibrium Analysis:

$$RU=\frac{k_1[A]}{1+k_1[A]}+\frac{3k_2[A]}{1+k_2[A]}$$

where k₁ corresponds to the high affinity site and k₂ denotes the low affinity sites.

Kinetic Analysis:

$$\mathrm{Reaction}\mathrm{equations}:A+B1\underset{k_{d1}}{\overset{k_{a1}}{\rightleftarrows }}AB1A+B2\underset{3k_{d2}}{\overset{3k_{a2}}{\rightleftarrows }}AB2$$

where A=analyte (Mxra8), B1=high affinity sites of ligand (CHIKV VLP), and B2=low affinity sites of ligand (CHIKV VLP)

Differential equations:

$$\frac{d[B1]}{dt}=-\left(k_{a1}[A][B1]-k_{d1}[AB1]\right)$$

$$\frac{d[B2]}{dt}=-3\left(k_{a2}[A][B2]-k_{d2}[AB2]\right)$$

$$\frac{d[AB1]}{dt}=\left(k_{a1}[A][B1]-k_{d1}[AB1]\right)$$

$$\frac{d[AB2]}{dt}=3\left(k_{a2}[A][B2]-k_{d2}[AB2]\right)$$

BLI-based competition binding assay

Binding of Mxra8 and anti-Mxra8 mAbs to captured CHIKV VLP was monitored in real-time at 25°C using an Octet-Red96 device (Pall ForteBio). CHK-265 (100 μg) was mixed with biotin (EZ-Link-NHS-PEG4-Biotin, Thermo Fisher) at a molar ratio of 20:1 biotin:mAb, incubated at room temperature for 30 min, then unreacted biotin was removed by passage through a desalting column (5 mL Zeba Spin 7K MWCO, Thermo Fisher). Biotinylated-CHK-265 was loaded onto streptavidin biosensors (ForteBio) until saturation, typically 10 μg/ml for 2 min, in 10 mM HEPES (pH 7.4), 150 mM NaCl, 3 mM EDTA, and 0.005% P20 surfactant with 1% BSA. CHIKV VLP was then added for 5 min. The biosensors then were dipped into wells containing 1 μM Mxra8 alone or with 1μM of hamster anti-Mxra8 mAbs (1G11.E6, 1H1.F5, 3G2.F5, 4E7.D10, 8F7.E1, or an isotype control mAb) for 5 min, followed by a 5-min dissociation.

Cryo-EM sample preparation, data collection, and single particle reconstruction

CHIKV VLP with and without cleaved Mxra8 and CHIKV infectious virus (181/25 strain) with cleaved Mxra8 in molar excess were flash cooled on holey carbon EM grids in liquid ethane under BSL-2 containment conditions using an FEI Vitrobot (ThermoFisher). Movies of the samples were recorded with the software EPU (Thermo Fisher) using a K2 Summit electron detector (Gatan) mounted on a Bioquantum 968 GIF Energy Filter (Gatan) attached to a Titan Krios microscope operating at 300 keV with an electron dose of 50 e⁻/Ų and a magnification of 18,000x. The movies (30 frames, 300 msec exposure per frame) were corrected for beam-induced motion using MotionCor2 (Zheng et al., 2017). Contrast transfer function parameters of the electron micrographs were estimated using CTFFIND4 (Rohou and Grigorieff, 2015) in cisTEM (Grant et al., 2018). Particles were auto-picked, and underwent reference-free 2D classification, ab initio 3D reconstruction, and 3D refinement in cisTEM. Local resolution was estimated using RELION-3 (Zivanov et al., 2018). Additional information regarding the number of images and particles is listed in Table S2. Structural visualization of the 3D reconstructions were performed using the programs Chimera (http://www.cgl.ucsf.edu/chimera) and PyMOL (Bramucci et al., 2012).

Model building

The initial models of the CHIKV structural proteins (E1, E2, TM regions, and Capsid) with or without Mxra8 were built into the density of a subunit by docking the components from the crystal structures of CHIKV p62-E1 (PDB: 3N42), the CHIKV capsid (PDB: 5J23), and the modeled TM regions of VEEV (PDB: 3J0C) using the fitmap command in Chimera. All components were docked in as one rigid body except for Mxra8, where D1 (residues 70–188) and D2 (32–59; 201–294) were docked as two separate bodies. Amino acid substitutions and Mxra8 hinge regions (residues 60–69; 189–200) were added or built manually in COOT to reflect the strain of CHIKV used for VLP production (CHIKV-37997) and infectious virus (CHIKV-181/25) for cryo-EM studies. The subunit models then underwent real-space refinement using PHENIX with default parameters plus rigid body refinement and secondary-structure and torsion restraints with the initial components as the reference. Rigid bodies for E1 were divided into domains I and II (residues 1–292), domain III (293–381), stem (382–412), and transmembrane region (413–442). E2 was divided into the N-linker region (residues 5–15), domain A (residues 16–134), domain B (residues 173–231), domain C (residues 269–341), β-linker (residues 135–172, 232–268), and the stem, transmembrane region, and cytoplasmic tail (residues 342–423). The capsid protein was divided into two rigid bodies (residues 111–176 and residues 177–261), and Mxra8 was divided into D1 (residues 70–188) and D2 (32–59; 201–294). The refined subunits were used to build the asymmetric unit with adjacent subunits to prevent clashes and optimize interactions. These models underwent further real-space refinement. COOT was used to fix regions manually with poor geometry. After optimization, coordinates of the asymmetric units were checked by MolProbity (Table S3). Contact residues were identified, and buried surface areas were calculated using PDBePISA (www.ebi.ac.uk/pdbe/pisa/).

Mxra8 expression experiments in 293T cells

The C-terminal FLAG tagged mouse Mxra8 corresponding to the transcript (NM_024263) and its corresponding charge reversal mutants were synthesized commercially (Genewiz) in a pCAGGS mammalian expression vector. Plasmids were transfected transiently in 293T cells using Fugene®HD according to the manufacturer’s instructions. One day after transfection cells were inoculated with CHIKV-181/25 (MOI 3) for 12.5 h, and then harvested, fixed with paraformaldehyde (PFA), permeablized, and processed for CHIKV expression using 1 μg/ml of humanized CHK-166 (Pal et al., 2013) and anti-FLAG® M2 antibody (Sigma-Aldrich #1804). After washing, cells were stained with 2 μg/ml of goat anti-human IgG conjugated with Alexa Fluor® 647 (Thermo Scientific) and 2 μg/ml of goat anti-mouse IgG conjugated with Alexa Fluor® 488 (Thermo Scientific) and subjected to flow cytometry analysis on a MACSQuant® Analyzer 10 (Miltenyi Biotec). To assess surface versus intracellular expression of Mxra8, cells were harvested at 36 h post transfection using TrypLE (Thermo Scientific), and incubated with a serum polyclonal hamster anti-mouse Mxra8 antibody (1:400) at 4°C for 25 min. After washing, intact cells were stained with 2 μg/ml of goat anti-Armenian hamster IgG H&L conjugated with Alexa Fluor® 647 (Abcam #173004) for 25 min at 4°C. After two additional washes, cells were fixed with 2% PFA for 10 min at room temperature. After permeablization, cells were stained with 1 μg/ml of anti-FLAG® M2 antibody (Sigma-Aldrich #1804), and then with 2 μg/ml of goat anti-mouse IgG conjugated with Alexa Fluor® 488 (Thermo Scientific). Cells were subjected to flow cytometry analysis and processed using FlowJo software (Tree Star).

Hydrogen-deuterium exchange mass spectrometry

Continuous HDX labeling of Mxra8 with or without the anti-Mxra8 mAbs was performed at 25°C for 0, 10, 30, 60, 120, 900, 3,600 and 14,400 sec as previously described with the following modifications (Yan et al., 2015). Briefly, stock solutions of both Mxra8 with or without the mAbs (50 μM) were prepared in 1X PBS pH 7.4 and incubated at 25°C for at least 1 h. Continuous labeling with deuterium was initiated by diluting 1 μl of the stock samples 25-fold in 24 μl of deuterated 1X PBS buffer (Sigma-Aldrich). HDX control samples (non-deuterated) were prepared in the same way with H₂O. Quenching was performed under reducing condition by adding a solution of 500 mM Tris (2-carboxyethyl)phosphine hydrochloride (TCEP HCl) and 4 M guanidine hydrochloride in 1X PBS buffer pH 7.4 (adjusted using sodium hydroxide to pH 2.5) to the reaction vial at a 1:1 volume ratio. The sample was mixed and incubated for 1 min at 25°C before loading on to a custom-built HDX platform for desalting, on-line pepsin digestion, and reversed-phase separation and directly injected into the mass spectrometer for analysis.

The sample was passed over a custom-packed 2 × 20 mm pepsin column at 200 μL/min; immobilized pepsin was prepared by a published procedure (Busby et al., 2007). The peptides resulting from digestion were captured by a 2.1 × 20 mm Zorbax Eclipse XDB-C8 trap column (Agilent) and desalted at 200 μL/min of H₂O containing 0.1% triflouroacetic acid for 3 min. The peptides were separated by a 2.1 × 50 mm C18 column (2.5 μm XSelect CSH C18; Waters) with a 9.5-min gradient of 5%−100% acetonitrile in 0.1% formic acid at a flow rate of 100 μl/min delivered by a LEAP 3× Ti pump (LEAP technologies, NC). The linear part of the gradient from 0.3 min to 5.5 min raised the acetonitrile content from 15% to 50%, during which time most of the peptides eluted from the C18 column. The entire fluidic system was kept in an ice bath except for the pepsin column to minimize back exchange. Duplicate measurements were carried out for each of the time points.

Acquired spectra were analyzed using HDX workbench software (Pascal et al., 2012) against a peptide set generated as described below. The deuterium level was normalized to the maximum deuterium concentration (96%) contained in the reaction vial. A peptide list used to search the HDX data was identified first by a tandem MS experiment in a data-dependent mode on a LTQ-FT mass spectrometer (Thermo Scientific). The six most abundant ions were submitted to collision-induced dissociation fragmentation. Product-ion spectra were submitted to Byonic™ (Protein Metrics, version 2.11.0 (Bern et al., 2007)) for identification and peptide set build-up, then manually inspected, and the validated peptides were used for the HDX analysis.

HDX data analysis and epitope assignment

The epitopes were identified as regions/sequences of amino acids (not single residues) and determined from peptide level HDX-MS data. The data trends in the overlapping peptides were analyzed using a conservative approach in making assignments. For each peptide, the relative difference in deuterium uptake was calculated by dividing the difference in deuterium uptake between Mxra8 alone and Mxra8 with mAb for each of the time points. Epitope peptides were initially identified as those peptides with relative differences in deuterium uptake of greater than 15% at two or more consecutive time points. If two or more overlapping peptides corresponded to a given region, the larger peptide was chosen. Selected peptides were screened to exclude the regions/sequences of amino acids exhibiting less than 15% relative difference in deuterium uptake between the apo- and the holo- states, as evidenced by another constituent peptide with shorter sequence length.

Data Availability

The authors declare that all data supporting the findings of this study are available within the paper and its Supplementary information. All structures are deposited in the PDB and EMDB databases (PDB 6NK3, 6NK5–6NK7; EMDB 9394–9395). The mapping data of anti-Mxra8 mAbs are deposited in the Immune Epitope Database and Analysis Resource.

Supplementary Material

1

Figure S1. Biochemical characterization of soluble Mxra8 proteins and CHIKV VLPs. Related to Figures 1 and 2. A-D. Mxra8 characterization. A-B. Bacterially-derived Mxra8 ectodomain was expressed in BL21 cells, oxidatively refolded, and purified by size exclusion chromatography. A. Coomassie-stained SDS-PAGE of refolded Mxra8 under non-reducing and reducing conditions. B. SEC-MALS of refolded Mxra8 showing a monomeric form at ~31 kDa. C-D. Mxra8-Fc was produced in Expi-293 cells, purified by protein A affinity chromatography, and digested with HRV protease to release the monomeric form of Mxra8 from the Fc region. C. Coomassie-stained SDS-PAGE under non-reducing and reducing conditions of 5 μg of undigested (left panel) and HRV-digested (right panel) Mxra8. One representative experiment of three is shown. D. SEC-MALS of soluble Mxra8 showing a monomeric form at ~38 kDa. E-G. VLP characterization. After transfection of HEK-293 cells with plasmid DNA encoding CHIKV C-E3-E2–6K-E1 (strain 37997), VLPs were harvested from the supernatant and purified by anion exchange chromatography. E. Coomassie-stained 4–12% SDS-PAGE of 2 μg of purified VLPs. E2 and E1 co-migrate at ~52–53 kDa. Capsid migrates at ~35 kDa. One representative experiment of two is shown. F. Immunoblotting of 0.25 μg of VLPs with a rabbit anti-CHIKV polyclonal antibody. One representative experiment of two is shown. G. Dynamic light scattering analysis of CHIKV VLPs. A single homogeneous peak of 68 nm (680 Å) ± 5.3 is shown. Data are representative of six experiments from two different VLP preparations.

2

Figure S2. Sequence alignment of mouse and human Mxra8 orthologs. Related to Figures 1, 4 and 5. Alignment was performed between mouse (NP_077225) and human (NP_001269514) Mxra8 using PROMALS3D (Papadopoulos and Agarwala, 2007) with mouse Mxra8 numbering. The Figure was prepared using ESPript 3.0 (Robert and Gouet, 2014). Domains 1 (light magenta) and 2 (dark magenta) are indicated above the sequence, along with the labeled secondary structures. Triangles indicate the N- and C- termini of the crystal structure (black) and the cryo-EM model at sites 2, 3, and 4 (red). Red boxes signify conserved residues; white boxes and black letters, non-conserved residues. Numbers under the alignment indicate contacts between Mxra8 and individual E2-E1 heterodimers, given as percentage buried surface area calculated by PDBePISA (http://www.ebi.ac.uk/pdbe/pisa/): 1 represents 10% buried surface area, 2 represents 20% buried surface area, and so on. “Wrapped” denotes contacts to the wrapped E2-E1 heterodimer making the primary contacts to Mxra8. “Intraspike” refers to the intraspike heterodimer, which is adjacent to the wrapped heterodimer but within the same trimeric spike. “Interspike” refers to the interspike heterodimer, which is in the neighboring spike. Red numbers indicate contacts to CHIKV E3.

3

Figure S3. Domain rotation in Mxra8., Related to Figures 1 and 4. (Upper left) Ribbon drawings of the Mxra8 structural models obtained from cryo-EM refinement shown superposed on the basis of the membrane proximal domain 2 (D2). These structures correspond to the four different binding sites (sites 1–4) on the CHIKV VLP. Analysis of the domain movements using DynDom (Hayward and Berendsen, 1998) suggest that all four domains are essentially identical in conformation. (Upper right) In contrast, comparison of the two Mxra8 models from the asymmetric unit of the X-ray crystal structure (chains A and B) show a difference in domain orientation, yielding a rotation of ~ 43˚ around the axis shown in black. Furthermore, the conformations of both chains A and B vary from that seen in the cryo-EM structure of Mxra8 bound to VLP (Lower panels). To facilitate comparison, the position of domain 2 was held constant throughout all four panels. The rotation axes pass through the hinge region near the interdomain disulfide. The domain movements include both bending and twisting motions, as indicated by differences in the relative orientations of the rotation axes. Depiction of helical twists have been removed for clarity.

4

Figure S4. Sequence alignment of E1 proteins of arthritogenic and encephalitic alphaviruses. Related to Figure 5. Amino acidsequence alignment of E1 proteins of arthritogenic (CHIKV, ABX40011; MAYV, AAY45742; RRV, ACV67002; ONNV, AOS52786; SFV, AAM64227; and SINV, AUV65223) and encephalitic alphaviruses (WEEV, AAF60166; EEEV, AAF04796; and VEEV, AAB02517). Structure based sequence alignments were performed between alphaviruses that do (group 1, left margin) or do not (group 2, left margin) require Mxra8 for infection using PROMALS3D with CHIKV numbering. The Figure was prepared using ESPript 3.0. Domains I (light grey), II (medium grey), III (dark grey), the fusion loop (orange), and the E1 stem, TM segment, and cytoplasmic tail (black) are indicated above the sequence, along with the secondary structure features and nomenclature (PDB 3N42; Voss et al, 2010). Red boxes, 100% conserved; White boxes and red letters; homologous residues; White boxes and black letters, non-conserved residues. Numbers under the alignment indicate contacts between Mxra8 and individual E2-E1 heterodimers, given as percentage buried surface area calculated by PDBePISA: 1 represents 10% buried surface area, 2 represents 20% buried surface area, and so on. “Wrapped” denotes contacts to the wrapped E2-E1 heterodimer making the primary contacts to Mxra8. “Intraspike” refers to the intraspike heterodimer, which is adjacent to the wrapped heterodimer but within the same trimeric spike. “Interspike” refers to the interspike heterodimer, which is in the neighboring spike.

5

Figure S5. Sequence alignment of p62 (E3-E2) proteins of arthritogenic and encephalitic alphaviruses. Related to Figure 5. Amino acidsequence alignment of p62 (E3 and E2) proteins of arthritogenic (CHIKV, ABX40011; MAYV, AAY45742; RRV, ACV67002; ONNV, AOS52786; SFV, AAM64227; and SINV, AUV65223) and encephalitic alphaviruses (WEEV, AAF60166; EEEV, AAF04796; and VEEV, AAB02517). Structure based sequence alignments were performed between alphaviruses that do (group 1, left margin) or do not (group 2, left margin) require Mxra8 for infection using PROMALS with CHIKV numbering. The Figure was prepared using ESPript 3.0. E3 (yellow), and E2 domains A (light cyan), B (medium cyan), C (blue), the -linker (medium cyan), and the stem, TM segment, and cytoplasmic tail (dark blue) are indicated above the sequence, along with the secondary structure features and nomenclature (PDB 3N42; Voss et al, 2010). Red boxes, 100% conserved; White boxes and red letters; homologous residues; White boxes and black letters, non-conserved residues. Numbers under the alignment indicate contacts between Mxra8 and individual E2-E1 heterodimers, given as percentage buried surface area calculated by PDBePISA: 1 represents 10% buried surface area, 2 represents 20% buried surface area, and so on. “Wrapped” denotes contacts to the wrapped E2-E1 heterodimer making the primary contacts to Mxra8. “Intraspike” refers to the intraspike heterodimer, which is adjacent to the wrapped heterodimer but within the same trimeric spike. “Interspike” refers to the interspike heterodimer, which is in the neighboring spike.

6

Figure S6. Unique E1 contacts by Mxra8 at site 1. Related to Figure 4. A. Top view of the asymmetric unit of CHIKV VLP with labeled Mxra8 binding sites. Site 1 is the Mxra8 (magenta) site that makes additional contacts to domains I and II of E1 directly beneath it (grey). E2, cyan; capsid, green. B-C. Side views (top panels) and zoomed-in views (bottom panels) of site 1 (B) and site 2 (C), highlighting the unique residues of E1 (orange) that contact site 1 of Mxra8 binding. Contacts were identified from the atomic model using PDBePISA (http://www.ebi.ac.uk/pdbe/pisa/).

7

Figure S7. Epitope mapping of hamster anti-Mxra8 mAbs by hydrogen-deuterium exchange and mass spectrometry. Related to Figure 5. A. Representative HDX kinetic plots of peptides from mAbs of each of the 3 classified groups, both in the presence (black lines) and absence (red lines) of Mxra8. Regions are considered to contain the binding epitopes if they show reduced rates or extents of hydrogen-deuterium exchange. All experiments were performed in duplicate, and data are representative of two experiments. B. Heat map depicting the average difference of deuterium incorporation between the mAb alone and with Mxra8 present across all seven time points (ΔD%). Coloring of the peptides correspond to the mAb classification. Yellow: mAbs 4E7.D10, 8F7.E1; violet: mAbs 1G11.E6, 7F1.D8; green: mAbs 1H1.F5, 3G2.F5, 9G2.D6. Consensus peptides for epitope designation and depiction: 41–51, light violet; 91–107, yellow; 162–173, light green; 185–202, green; 241–256, violet. C-D. Surface and ribbon diagrams of Mxra8 at site 3 and the E2-E1 domains at its binding interface. CHIKV structural proteins are colored by domain. E1: DI, light grey; DII, medium grey; fusion loop, orange. E2: A domain, light cyan; β-linker, medium cyan; domain B, dark cyan. ‘ denotes domains within the intraspike heterodimer, and “ for the interspike heterodimer; the wrapped heterodimer is labelled without symbols. Surface representation of Mxra8 models are colored with the HDX-mapped epitopes of anti-Mxra8 mAbs, where shades of violet correspond to mAb 1G11.E6, green for mAbs 1H1.F5 and 3G2.F5, and yellow for mAbs 4E7.D10 and 8F7.E1. (C) Mxra8 model shown is of the properly docked model, with D2 proximal to the viral membrane. (D) Alternative Mxra8 model where the Mxra8 docking was flipped, positioning D1 proximal to the viral membrane and D2 extending away from the virion. This model underwent an identical building and refinement process as described in the Methods and resulted in a poorer fit into the cryo-EM density map as evident by visual inspection and a lower correlation coefficient.

8

Figure S8. Presence of E3 can sterically hinder Mxra8 binding. Related to Figure 6. A. View of asymmetric unit of CHIKV infectious particles with difference map of Mxra8 docked on. E1, E2, E3, and Mxra8 sites are labeled, with the high affinity site 1 labeled with a square. B-D. Zoomed-in views highlight the amount of steric clashing of Mxra8 at site with E3 from the adjacent spike. We calculated the surface area of occluded Mxra8 density (zoned within 2 Å of E3 coordinates) as follows: site 1 (B), 19.6 Ų; site 2 (C), 218.4 Ų; site 3 (D), 421.4 Ų; and site 4 (E), 400.3 Ų. Color scheme for A-E: grey, E1; cyan, E2; transparent yellow, E3; capsid, green; Mxra8, magenta. Prime labels refer to Mxra8 sites on adjacent spikes. F. Cartoon schematic summary of results: the red triangle and the dashed lines depict two distinct representations of one icosahedral asymmetric unit. The 5-fold (i5), 3-fold (i3), and 2-fold (i2) icosahedral axes of symmetry are indicated with a black pentagon, triangles, and oval, respectively. The white triangles and ovals indicate the quasi 3-fold and 2-fold axes of symmetry, respectively. Trimeric spikes are labeled “i3” if coincident with the i3 axes, and “q3” if on a quasi 3-fold axes. Color scheme: grey, E1; cyan, E2; transparent yellow, E3; site 1 Mxra8, magenta; sites 2, 3, and 4 Mxra8, purple. X denotes sites of large steric clashes.

9

Figure S9. Human Mxra8 binding to CHIKV VLPs. Related to Figure 6. A. SEC-MALS of cleaved human Mxra8 showing a monomeric form at ~34 kDa. B-C. Kinetic sensograms and steady-state analysis of human Mxra8 binding to CHIKV VLPs fit to a 1:1 binding model (B) or a two-site binding model (C) (one high affinity site and three low but equal affinity sites). Raw experimental traces are in black, fit traces are in red. Inset, Scatchard plot (2 experiments; mean, standard error of the mean (SEM), and χ² values).

10

Table S1. Crystallographic data collection and refinement statistics for mouse Mxra8.

Table S2. Summary of Cryo-EM data collection, Related to Figures 2 and 3.

Table S3. Cryo-EM model building and refinement statistics, Related to Figure 3.

Table S4. List of contact residues of CHIKV E2-E1 at Mxra8 binding interface, Related to Figures 4 and 5.

HIGHLIGHTS.

• A 2.3 Å resolution X-ray structure of Mxra8 shows an unusual Ig-like fold topology

• Mxra8 has a complex quaternary binding site that spans CHIKV E2 and E1 proteins

• Mutational and epitope analyses support cryo-EM structures of Mxra8 with CHIKV

• The presence of CHIKV E3 protein on the virion affects the mode of Mxra8 binding