Structural basis of Zika virus NS1 multimerization and human antibody recognition

59 The Zika virus (ZIKV) is a mosquito-transmitted flavivirus related to the dengue virus that has 60 emerged in recent global outbreaks, with previously unreported neurological or birth 61 abnormalities. There remains a large, unmet need for efficacious vaccines and antiviral agents 62 against ZIKV. Flavivirus non-structural protein 1 (NS1) is the only secreted viral protein that has 63 proven useful in diagnostics. However, its physiologically relevant structural forms and 64 pathogenesis mechanisms remain unclear. While many antibodies targeting NS1 with mapped 65 immunodomain epitopes have been reported, antibody-NS1 structures are starting to emerge, 66 which may guide vaccine and therapeutic design. Here, we present high-resolution cryoEM 67 structures of ZIKV recombinant secreted NS1 (rsNS1) and its complexes with three human 68 monoclonal antibodies, as well as evidence of ZIKV infection-derived secreted NS1 (isNS1) 69 binding to HDL. We showed that ZIKV rsNS1 forms tetramers and filamentous repeats of 70 tetramers. The studied antibodies (AA12, EB9, GB5) did not disrupt the ZIKV NS1 tetramers as 71 they bound to the wing and connector subdomain to the β-ladder. Our study reveals new insights 72 into NS1 multimerization, highlights the need to distinguish the polymorphic nature of rsNS1 and 73 isNS1, and expands the mechanistic basis of the protection conferred by antibodies targeting 74 NS1.

• rsNS1 forms tetramers, which agrees with the loose tetramers reported at 8.3 Å by Shu     et al. (2022).The rsNS1 dimer-dimer orientation is rotated by ~ 90 °, and the interface is not rigid, with a larger degree of movement observed in the antibody-bound complexes.
• Additionally, ZIKV rsNS1 at lower concentrations showed a filamentous form driven largely by complementary surface charge interactions of the wing ladder, and their lowresolution structure was resolved at 8 Å, corroborating the ZIKV rsNS1 crystal contact form solved by Xu  • Antigen-binding sites for the antibodies studied in this work were mapped at the connector subdomain between the wing and the β-ladder of the NS1 protein, which represents a new subset of antibody-NS1 complex structures.
• Overall, NS1 has the propensity to form various oligomeric structures, and the physiologically more relevant isNS1 of flaviviruses appears to be dimers docked onto HDL.
The usefulness of rsNS1 as an immunogen for vaccine development hinges on its wellcharacterized rsNS1 oligomeric form.

Introduction
Zika virus (ZIKV) belongs to the Flavivirus genus, which includes other viruses that can cause severe human diseases, such as dengue virus (DENV).ZIKV is transmitted by the Aedes mosquito and has re-emerged in recent ZIKV epidemics, with previously unreported cases of congenital microcephaly and Guillain-Barré syndrome in adults.ZIKV remains a public health threat 1,2 with no effective antivirals or vaccines available.
One therapeutic or vaccine target is the multifunctional non-structural protein 1 (NS1).NS1 can be found intracellularly, where it is essential for virus replication 3,4,5 , docked on lipid rafts of the cell surface membrane 6 , and abundantly secreted into the bloodstream of infected individuals during the febrile phase 7 .Interestingly, studies have shown that secreted NS1 (sNS1) has a pathophysiological influence on tissue tropism because of its different surface charge profiles 8,9 .
Earlier efforts by our group and others have shown that NS1-targeted monoclonal antibodies (mAbs) are protective in vivo either through Fcγ-dependent 10,11 and/or Fcγ-independent pathway 12,13,14,15 , without the known antibody-dependent enhancement effect of envelope protein-targeted mAbs.We also showed that the NS1-based ZIKV vaccine can inhibit pathogenicity in vivo 16 .However, autoimmunity concerns regarding NS1-targeted mAbs hinder vaccine development as reviewed 17,18,19 , underscoring the need for complete structural details that underpin antibody-dependent protection.
Unraveling how antibodies recognize and bind to their epitopes is crucial for delineating the mechanisms by which antibodies protect against disease.While a large body of work has focused on studying neutralizing antibodies against the flavivirus envelope (E), there is still a gap in knowledge concerning the role and mechanisms by which NS1 antibodies provide protection.NS1 is comprised of three distinct domains, an N-terminal hydrophobic β-roll dimerization domain (residues 1-29) formed by the intertwining of two protomers, an α/β-wing domain (residues 38-151) that extends out from the central β-ladder domain (residues 181-352) giving the dimer a distinct crossed shape 8,20,21 .The connector segments between the three domains, residues 30-37 and 152-180, form a 3-stranded β-sheet.The β-ladder domain has an inner face of β-strand rungs and a flexible loop (residues 108-129) that forms the membrane-associated surface together with the β-roll and the "greasy finger" (residues 159-163) of the wing domain.
The unstructured and hydrophilic outer face of NS1 is characterized by the "spaghetti loop" (residues 219-272) of the β-ladder domain where most immunodominant and highly targeted regions have been mapped 11,22,23 .Notably, the first high-resolution cryoEM structures of DENV2 recombinant secreted NS1 (rsNS1) and in complex with 5E3 by Shu et al. (2022) showed that rsNS1 predominantly forms tetramers 24 .This challenges the longstanding view that secreted NS1 (sNS1) is a barrel-shaped hexamer with lipid-cargos 25,26,27 .Furthermore, our preprint presented cryoEM structures of DENV2 infection-derived secreted NS1 (isNS1), which showed NS1 dimers complexed with HDL instead of hexamers 28 .This supports earlier reports showing rsNS1 association with HDL, which can trigger pro-inflammatory responses 29,30 and uses scavenger receptor B1 (SRB1) as a cell receptor in cultured cells 31 .The observation that flavivirus rsNS1 is primarily a tetramer when recombinantly derived, while its physiological form, isNS1, is docked on HDL needs to be explored further.
To date, there have been five published reports on antibody-NS1 complex structures.Fab 22NS1 in complex with West Nile virus (WNV) C-terminal NS1 (NS1c; residues 172-352) which was found bound to the loop face of the C-terminal β-ladder domain 32 .While Fab/ScFv 2B7 with DENV1/2 NS1 13 , Fab 1G5.3 with DENV2 NS1c (residues 174-352) and ZIKV NS1c (residues 272-339) 12 , and Fab 5E3 with DENV2 NS1 24 similarly bind to the distal end of the C-terminal β-ladder domain at different orientations.The resolved structures of antibodies 2B7 13 , 1G5.3 12 , and 5E3 24 bound to the NS1 β-ladder domain suggested that the protective effect through an Fcγ-independent pathway was due to steric hindrance of the sNS1 interaction with lipid membranes.
To further characterize how antibodies recognize and interact with ZIKV NS1, we determined the cryoEM maps of ZIKV rsNS1 alone and in complex with human monoclonal antibodies (AA12, EB9, and GB5), at 2.9 Å, 3.1 Å, 2.8 Å, and 3.8 Å respectively.Similar to the "loose" tetramers reported at 8.3 Å by Shu et al 24 , we only observed tetramers of ZIKV rsNS1 (dimer of dimer).In addition, ZIKV rsNS1 at a lower concentration forms a filamentous chain driven largely by the complementary surface charge interactions of the wing ladder.The three antibodies used in our study did not disrupt ZIKV NS1 tetramers into dimers, unlike what was reported for 5E3 with DENV2 rsNS1 dimers 24 and 56.2 with DENV2 isNS1 dimers 33 .Compared to the available structures, we present a new subset of antibody-NS1 complex structures at antigen-binding sites between the wing and β-ladder of the NS1 protein.Lastly, we showed that ZIKV isNS1 is found in complex with apoA1; they are dimers docked onto HDL, similar to earlier reports for DENV 30,33 .
Overall, our study highlights the polymorphic nature of rsNS1 as compared to the more physiologically relevant isNS1 and unravels a plausible mechanistic basis for human mAb (AA12, EB9, and GB5)-mediated protection 10,14 .The potential contribution of rsNS1 to biotherapeutic and vaccine development hinges on a structurally defined sample.

Results
Recombinant secreted ZIKV NS1 forms asymmetric tetramers.We expressed and purified fulllength recombinant ZIKV NS1 (Uganda strain MR766; Zv sNS1MR766wt) with C-terminus polyhistidine (HIS)-tag from the supernatant of Expi293 cells.Unexpectedly, we observed that ZIKV rsNS1 forms filaments that are flexible and composed of asymmetric tetramers (dimer pairs) stacked along their exposed wing-ladder domain surfaces (Supplementary Fig. 1).Helical reconstruction of ZIKV rsNS1 filaments resulted in a low-resolution map of 8 Å, which could be fitted with ZIKV rsNS1 dimers, and the resulting electrostatic surface was shown to be complementary to the filamentous form (Supplementary Fig. 1).To achieve a higher resolution cryoEM structure of ZIKV rsNS1, we obtained a suitable dataset for single-particle analysis by increasing the sample concentration on graphene grids 34 (Supplementary Fig. 2).Similar to the ZIKV rsNS1 filaments, we only observed asymmetric tetramers, and the overall structure resolved to 2.9 Å (Figure 1a-b, Supplementary Fig. 2).Based on model fitting, the rsNS1 dimer-dimer interacts with the β-roll and is oriented almost perpendicularly to each other (Figure 1b-c), with an additional interaction between Phe163 in the greasy finger of the wing domain and Asp274 of the β-ladder within 3.5 Å at the opposite protomer (Figure 1d).Given that the CryoEM map for one dimer was of poor quality owing to the observed smearing, the hydrophobic interface was not rigid (Figure 1d).This is apparent upon local focused refinement and 3D flexible refinement, which is indicative of its motion hinging around the β-roll interface with up to 15.02 °rotation (Figure 1a-b, Supplementary Fig. 2, Movie File 1).
Comparison of recombinant secreted ZIKV NS1 asymmetric tetramers with prior CryoEM and crystal structures.Next, we superimposed our ZIKV rsNS1 tetramer map with the DENV2 rsNS1 "loose" tetramer map (EMDB-32842; Figure 2a), the corresponding models (PDB:7WUU), and an existing crystal structure of ZIKV rsNS1 BeH819015 strain (PDB:5GS6), which could be tetrameric based on crystal packing (Figure 2b) 8,24 .The structures were very similar overall (RMSD:1.93),with our model capturing a different pose within the trajectory of the dimer movement (Figure 2b).The representative 2D classes of ZIKV rsNS1 (Figure 2c, top; Supplementary Figure 2 Characterization of human mAbs AA12, EB9, and GB5 against ZIKV NS1.The cryoEM maps of ZIKV rsNS1 in complex with human monoclonal antibodies (AA12, EB9, and GB5) were obtained at 3.1 Å, 2.8 Å, and 3.8 Å respectively (Supplementary Figures 3-5).ZIKV rsNS1 was incubated with Fab AA12, EB9, or GB5.In addition, anti-Fab nanobody (AfNb), which binds to and increases the rigidity of the elbow linker between the variable and constant domains of the Fab LC, was added to aid in structure determination 35,36 .Both the heavy and light chains of Fabs AA12, EB9, and GB5 bind similarly to the surface of the ZIKV rsNS1 wing and β-ladder connector subdomain region with an observable weak density for AfNb, which helped in the initial model docking (Figure 3a).ZIKV rsNS1 remained tetrameric in all three cases, as seen in the 2D and 3D class averages, but the map density of one dimer was much weaker (Supplementary Figures 3-5).This is indicative of a greater motion range of the dimer than ZIKV rsNS1 alone, as shown for the Fab EB9 data, with up to 28.2 °rotation observed (Supplementary Figures 3).Distinct classes with varied numbers of Fab molecules bound to the theoretical maximum of four were observed for Fab AA12 and EB9 (Supplementary Figures 3-4).Fab GB5 dataset only had up to two fab molecules bound to the exposed faces of the tetramer across all the class averages and with fewer good particles picked (Supplementary Figures 5).While the lower avidity observed for GB5 may be due to poorer sample quality or differences in data collection (Supplementary Table 1), it also correlates to being the weakest binder among the three 10 .To increase the map quality and determine the binding region, focused local refinements on one of the Fab variable domains and the bound ZIKV rsNS1 dimer, as well as subtraction of the poorly resolved dimer region, were performed (Supplementary Figures 3-5).C2 symmetry was also imposed on the Fab GB5 dataset during non-uniform refinement 37 to obtain the final map.This improved the Fab GB5 map resolution from 4.25 to 3.76 Å based on the gold standard FSC curve cut-off at 0.143 and a corresponding improvement in the EMRinger 38 score moving from 0.82 to 1.08 performed (Supplementary Figure 5).The EMRinger Score assesses the goodness of fit between the model and the cryo-EM map, a value of 1.0 is usual for initial models refined for 3.2-3.5Åresolution maps while values higher than 2 are expected for models with higher resolution.The binding epitopes of Fab AA12, EB9, and GB5 were mapped to the wing (residues R99-R103, E146) and connector subdomains (residues D174-L177), which are generally poorly conserved sites (Figure 3b-c, Supplementary Table 2, Supplementary Figure 6).Compared with the other antibody-NS1 complex structures obtained by superimposing the NS1 β-ladder domains, our structures represent a new subset of epitope regions compared to 22NS1 32 , 1G5.2 12 , 2B7 13 , 5E3 24 , and 56.2 33 (Figure 3e).

ZIKV infection-derived NS1 from Vero cells is predominantly a dimer docked on high-density
lipoproteins.Finally, we sought to understand whether earlier reports that DENV2 sNS1 are predominantly dimers docked onto HDL 30,33 are generalizable to other flaviviruses.Using the same infection model and antibody pull-down of isNS1 33 with mAb EB9, we detected apoA1, a major component of HDL co-eluted with isNS1 (Figure 4a-c).We also observed distinctive wingshaped protrusions of ZIKV NS1 dimers docked onto a spherical density that represents HDL in the 2D class averages of the eluted sample under negative-stain electron microscopy (Figure 4d).These representative 2D class averages were similar to those found for DENV2 sNS1 30,33 , providing more support for the observation of infection-derived flavivirus sNS1 as dimers docked onto HDL.2022) 24 showing that recombinantly derived secreted DENV2 NS1 (rsNS1) are mostly tetramers and new evidence on how similar rsNS1 could be bound to HDL 30 , the authors maintained the view that rsNS1 is hexameric without any critical discussion.

Discussion
Here, we provide further evidence that the recombinantly derived secreted ZIKV NS1 also forms tetramers with a distinct ability to oligomerize into filamentous structures (Figure 1-2, Supplementary Figure 1-2).The direct relationship between the crystal contacts (PDB:5GS6) 8 and the native multimeric ZIKV rsNS1 under cryoEM (Figure 2d, left) indicates the possibility that rsNS1 forms a microtubule-like structure observed in the crystal contacts of the other available ZIKV rsNS1 structures (PDB:5K6K; Figure 2d, right) 21 .A bundle of two such ZIKV rsNS1 microtubles (~220 Å overall diameter) may explain the long and thin tubular protrusions extending out of liposomes observed by Ci et al. (2020) when ZIKV rsNS1 was added 40 .While the potential pathophysiological implications of rsNS1 filaments or microtubules are unknown and could simply be induced by the in vitro environment, such structures and earlier work by us 33 and others 24,30 highlight discrepancies regarding the true soluble form of oligomeric NS1.There are clear structural differences in sNS1 among flaviviruses, wherein recombinantly derived NS1 forms tetramers either exclusively or with some hexamers, and infection-derived sNS1 forms dimers bound to HDL.
Our antibody-NS1 structures rationalize the prior observation of the in vivo and Fcγ-dependent protection conferred by human monoclonal antibodies AA12, EB9, and GB5 10 .We found that unlike other antibodies such as 1G5.3 12 , 2B7 13 , 5E3 24 and 56.2 33 that bind to the membrane interaction domains of NS1, the antibodies AA12, EB9, and GB5 10 do not interfere with the membrane interaction of NS1.Conservation analysis of these B cell epitopes to the exposed charged surface of the wing and connector subdomain β-ladder helps to explain why such antibodies generally do not show cross-reactivity 10,11,23 and are useful for flavivirus-specific rapid antigen assays 23 .Among the NS1 immunodominant regions 22 , it remains a keen interest to solve the structures of antibodies bound to the β-roll and connector subdomain to the wing region such as the mAbs 3G2 and 4B8 14 .The membrane-associated face of the NS1 may be exposed and unprotected when bound to antibodies such as 1G5.3 12 , 2B7 13 , 5E3 24 , or 56.2 33 to allow antibodies to target the β-roll region.
In conclusion, we emphasize the need to distinguish the source of sNS1 whether it is derived recombinantly (rsNS1) or from infection models (isNS1) in the literature.The key findings of our study provide important structural insights into NS1 alone and when bound to antibodies that will guide flavivirus therapeutic and vaccine development.
) and DENV2 rsNS1 (Figure 2c, bottom) reproduced from Shu et al. (2022) 24 illustrates that only the "loose" tetramer form is observed in our study.Our ZIKV rsNS1 filament structure corroborates the ZIKV rsNS1 crystal contact form (PDB:5GS6) solved by Xu et al. (2016) 8 , as shown in Figure 2d.Further examination of the crystal contact of the only other available ZIKV rsNS1 structure by Brown et al. (2016; PDB:5K6K) 21 of the same MR766 strain as used in our study, showed that ZIKV rsNS1 could also form a microtubule-like structure with a central hydrophobic cavity of ~16.5 Å.
Since the seminal work performed by Gutsche et al. (2011) 26 and Muller et al. (2012) 27 , sNS1 is thought to be a hexameric protein that can dissociate to bind to cellular membranes and contribute to cytopathic effects and disease 17 .While a recent review 39 acknowledged the new cryo-EM structures by Shu et al. (

Figure 1 .
Figure 1.ZIKV MR766 rsNS1 cryoEM structure.(a) cryoEM density map contoured at 0.3, coloured near atoms of (b) fitted model with its three domains, being the β-roll (orange, residues 1-30), wing (blue, residues 31-179), and β-ladder (cyan, residues 180-352).The bottom dimer of NS1 in (b) is recolored in grey.The side and top views are as shown.Bi-directional arrows indicate the rotational angles measured of the dimer movement observed.Green and red dashed boxes in (b) outlined the tetramer interface with close-up views of (c) the β-roll region and (d) between the wing and β-ladder of the dimers in the asymmetric tetramer.The key interacting residue backbone and side chains are labelled and shown as sticks colored by heteroatom and the NS1 is recolored by lipophilicity potential according to the key, higher attribute value means more hydrophobic.The secondary structures are labelled as shown in grey.Measured distances are shown for (d).(e) Schematic representation of the NS1 dimer-dimer interactions observed at βroll interface.Hydrogen bonds are shown as dashed lines.Hydrophobic residues and the distances between them are coloured in dark goldenrod as per colour key in (c).

Figure. 2 .
Figure. 2. ZIKV MR766 rsNS1 cryoEM structure in comparison with Zv rsNS1 crystal and DENV rsNS1 CryoEM structures.(a) cryoEM map fitting of rsNS1 tetramer between ZIKV (2.9 Å, this study, in cyan) and DENV (8.3 Å, EMDB-32842, in grey), contoured at 0.3 and 4.66 respectively, with a calculated correlation of 0.33.(b) The corresponding rsNS1 tetramer model superposition from this study (in cyan), DENV (PDB: 7WUU, in grey), and based on ZIKV crystal packing (PDB: 5GS6, in pink).The tetramer interface at the β-roll is highlighted by the dashed grey box and the close-up view is shown on the right.RMSD values reported are across all matched pairs as calculated using the matchmaker to a single chain for the overall model (left) and to the β-roll (residues 1-30, right).(c) Representative 2D class averages of the ZIKV MR766 rsNS1 cryoEM dataset are shown in the cyan box, grouped by the side and top views.Black scale bar, 100 Å, as indicated.2D class averages of the DENV rsNS1 cryoEM study from Figure 1d of Bo et al. (2022) is reproduced herein (bottom) and boxed in grey for comparison.(d) Two reported crystal packing forms of the ZIKV rsNS1 from strains BeH819015 (PDB: 5GS6, left) and MR766 (PDB: 5K6K, right) and its symmetry contacts shown at 30 Å. Representative 2D class averages of ZIKV rsNS1 filaments observed in our study (bottom left, white scale bar, 100 Å).NS1 dimers opposite

Figure 3 .
Figure 3. ZIKV MR766 rsNS1 in complex with human antibody fragments (Fab) EB9/AA12/GB5 and the antifab nanobody (AfNb).(a) Cartoon model superposition of the complexes by a matchmaker with respect to EB9 complex with RMSD value across all 352 paired atoms as indicated in the figure to be 0.89 and 1.25 for AA12 and GB5 respectively.AfNb is colored in pink, the Fab light and heavy chains are colored in dark and dim grey respectively, and the individual chains of the NS1 dimer is colored in cornflower blue and orange.Antibody-antigen binding interface is highlighted in an orange dashed box with details of the key interacting residue side chains shown respectively for (b) EB9, (c) AA12, and (d) GB5 as sticks colored by heteroatom.Backbone is rendered as tubes, and only the NS1 is recolored by residue conservation across flaviviruses according to the key.The alignment is shown in Supp.Fig. 6.Hydrogen bonds are shown as dashed lines and the complementarity-determining regions (CDRs) are labelled in (b).

Figure 4 .
Figure 4. ZIKV MR766 sNS1 from the supernatant of infected Vero cells.(a) The silver-stained gel of 100 ng of Crude and Elute immunoaffinity fractions separated on a 10% SDS-PAGE gel (reducing/denaturing conditions).The maximum well volume (30uL; 0.06% of total volume) of the Wash immunoaffinity fraction was loaded, as the total protein concentration is below the Bradford assay detection limit.(b) Western Blot validation of 50 kDa (boxed in red) and 25 kDa (boxed in orange) bands identified from (a) using an anti-ZIKV NS1 antibody (left; a gift from Yap Thai Leong, Experimental Drug Development Centre, A*STAR) and ApoA1 antibody respectively (right; Biorbyt, orb240478), when 125 µg and 500 ng total protein of Crude and Elute immunoaffinity fractions respectively, were separated on a 4-20% SDS-PAGE (reducing/denaturing conditions).The amount of NS1 in 500 ng of total protein in the Crude supernatant is below the detection limit on the western blot.(c) Western Blot detection of ZIKV NS1 (left) and ApoA1 (right) when 125 µg and 500 ng of total protein of Crude and Elute immunoaffinity fractions respectively, were separated on a 10% Native-PAGE, using the same set of antibodies as described in (b).(d) The three most representative 2D class averages from negative electron micrographs taken of the elute fraction.Red arrows highlight the distinctive