Introduction

Dengue virus (DENV) is a member of the flavivirus genus and is the cause of significant healthcare problems and economic burden worldwide, partially due to the lack of effective therapeutics and the limited efficacy of licensed vaccines, Dengvaxia® and QDENGA®^{1, 2}. While the majority of DENV infections are mild or asymptomatic, severe dengue infections, marked by vascular leakage, can be life-threatening and even fatal. The severity of dengue infections has been demonstrated to be correlated to high sera levels of sNS1 in clinical studies^{3, 4}. The viral nonstructural protein 1 (NS1) is a highly conserved and multifunctional protein that exists as intracellular membrane-associated NS1, a presumed GPI-anchored outer plasma membrane-associated NS1, and secreted NS1 (sNS1) during viral infection⁵. NS1 could interact with a myriad of proteins associated with its immune evasion and virotoxin roles⁶. The interactions of NS1 with viral E protein⁷ and the NS4A-2K-NS4B precursor protein⁸ have been shown to be important in virus production and replication respectively. Intracellular NS1 found in the ER lumen is an essential part of the membranous viral RNA replication compartment⁷. Upon release from the cells, sNS1 enters the blood circulation where in vitro and in vivo mouse evidence have demonstrated its ability to induce vascular permeability, a hallmark of severe dengue, either independently⁹ or through inducing pro-inflammatory responses¹⁰.

The high-resolution crystal structures of NS1 dimer have been reported for DENV¹¹, WNV¹¹, and ZIKV^{12, 13} which provide a conserved molecular view of flaviviral NS1. Mature NS1 is 352 amino acids long with an apparent molecular mass between 40 – 50 kDa depending on its glycosylation state. NS1 has a three-domain architecture, a hydrophobic β-roll (residues 1-29), an α/β wing (38-151), and a β-ladder (181-352). The connector segments between the wing and β-ladder domains, residues 30-37 and 152-180, form a 3-stranded β-sheet. The dimer has a distinct cross shape with the wings extending from the central β-ladder which has an extended β-sheet that faces the hydrophobic β-roll and a ‘‘spaghetti loop” on the opposite hydrophilic outer face that lacks structured elements^11–13. sNS1 was reported to be a barrel-shaped hexamer with lipid cargo held together by hydrophobic interactions based on biophysical and low-resolution EM analysis^14–16. CryoEM structures of recombinant sNS1 (rsNS1) were recently reported¹⁷. The rsNS1 is mainly tetrameric with only 3% of the rsNS1 population found to be in the hexameric form¹⁷. Other studies have implied that sNS1 interacts with high-density lipoprotein (HDL) and its major protein components, such as ApoA1 and ApoB^{14, 16}. Additionally, sNS1 appears to utilize the scavenger receptor B1, a known receptor for HDL, as the cognate receptor in cultured cells¹⁷. Since the interactions of sNS1 with the target cells implicated in pathogenesis require the hydrophobic β-roll side to be exposed¹¹, the mechanism by which the oligomeric sNS1 with its lipid load in the central channel dissociates and associates have been a topic of intense research interest^18–20. However, the native structure of the extracellular NS1 during viral infection remains unclear.

To closely mimic of sNS1 circulating in dengue patients’ sera, we obtained infection-derived sNS1 (isNS1) from the culture supernatant of Vero cells infected with either the DENV2 WT (isNS1wt) or T164S mutant (isNS1ts). The T164S mutation in the greasy finger loop between the wing and β-ladder interdomain of NS1 was identified from a DENV2 epidemic in Cuba in 1997, where a correlation with enhanced clinical disease severity was observed²¹. We demonstrated that this single mutation in NS1 could directly cause lethality in mice and increased sNS1 secretion²², which served as an epidemiologically relevant tool to study the biochemical and structural characteristics of sNS1 in this study.

We determined that the isNS1 is a complex of the NS1 dimer embedded on a single HDL particle composed of ApoA1. By applying integrative structural tools, we report the cryoEM structural model of the purified isNS1:HDL complex at ∼7 Å resolution with protein-protein interactions between NS1 and ApoA1 defined by crosslinking mass spectrometry (XL-MS). Furthermore, the Ab56.2 binding site model in the cryoEM model of isNS1:ApoA1 complex was elucidated using hydrogen deuterium exchange mass spectrometry (HDX-MS) as well as NS1 peptide competition ELISA. Interrogation of DENV-infected mouse sera and a human patient suggests that the sNS1:HDL complex exists in vivo as a similar complex to isNS1.

Results

Native isNS1 is a complex of NS1 with ApoA1

We purified sNS1 using a monoclonal antibody 56.2 from the supernatant of the infected Vero cell cultures (Fig. 1a and Supplementary Fig. 1a-b). Initial negative stain screening of the isNS1wt (Supplementary Fig. 1c) showed some 2D classes of NS1 dimers presumably docked onto a spherical density, mirroring the in vitro reconstitution of rsNS1 with HDL complex negative stain microscopy results by Benfrid et al. (2022)²⁰. Both immunoaffinity purified isNS1wt (Fig. 1b and Supplementary Fig. 2a) and isNS1ts (Supplementary Fig. 2a and 3b) retained a molecular size of ∼250 kDa as detected with Coomassie Blue (Fig. 1b and Supplementary Fig. 2a) and on a Western blot using Ab56.2 following separation on a Native-PAGE (Fig. 1b), as previously reported²². The isNS1wt migrated slower than the rsNS1 protein (a gift from Shu et al. 2022²³), indicating possible conformational or oligomeric differences between the two NS1 species (Fig. 1b). On a reducing SDS-PAGE, we identified two major bands of approximately 50 kDa and 25 kDa in isNS1wt, the former corresponding to the single monomeric NS1 band seen in rsNS1 (Fig. 1c). Preliminary mass ID identified the 25 kDa protein as bovine apolipoprotein A1 (ApoA1), a major component of HDL in bovine serum-containing culture media, which has a reported molecular weight of 28 kDa²⁴. Hence, the two bands were further validated on a Western blot, which confirmed the 50 kDa band as NS1 and the 25 kDa band as ApoA1 respectively (Fig. 1c). The same observations were seen in isNS1ts (Supplementary Fig 3).

Fig. 1.

Next, we elucidated the composition of the prominent 250 kDa band observed in the Native-PAGE by label-free quantification (LFQ) analysis of proteins using liquid chromatography–mass spectrometry (LC-MS). We obtained LC-MS data for the ∼250 kDa protein excised from the Native-PAGE separation of the immunoaffinity purified isNS1wt (represented in Fig. 1b, purple box as Gel Band 1); an aliquot of the elute fraction from immunoaffinity purification (depicted in Fig. 1a blue box and referred to as “Elute”) and the 50 kDa and 25 kDa bands excised after separation of the elute fraction on a reducing SDS-PAGE (Fig. 1c, represented by red and green boxed labelled as Gel Band 2 and 3 respectively) as positive controls. The raw data for the ion intensities of the samples tested are as detailed in Supplementary File 1. We found that the combined LFQ intensity of NS1 and ApoA1 accounted for more than 77.6% and 88.6% of the total ion intensity count (TIC) detected from the elute fraction (Fig. 1a, blue box) and the 250 kDa band (Fig. 1b, the purple box labelled ‘1’) respectively, indicating that NS1 and ApoA1 were the major proteins in the elute fraction and the ∼250 kDa band from immunoaffinity purified isNS1wt (Fig. 1d). Accordingly, as depicted in Fig. 1d, NS1 accounted for 96.85% of the TIC for the 50 kDa band (Fig. 1c, red box ‘2’), while the intensity of ApoA1 accounted for 96.23% of the TIC for the 25 kDa band (Fig. 1c, green box ‘3’). Similar LFQ analysis of LC-MS data for the purified isNS1ts mutant showed that NS1 and ApoA1 were the major components (Supplementary Fig. 2) of the respective Elute and ∼250 kDa gel band on Native-PAGE (Supplementary Fig. 2a; 3a, purple box). The corresponding 50 kDa and 25 kDa gel bands on th reducing SDS-PAGE (Supplementary Fig. 2b, 3b) identified as NS1 and ApoA1 by immunoblotting with the respective antibodies (Supplementary Fig. 3b), were further confirmed by LC-MS as NS1 and ApoA1 (Supplementary Fig. 2c, 3c, Supplementary file 1). Taken together, the results from the Native PAGE and LFQ by LC-MS strongly indicated that the isNS1 purified from the supernatant of infected cells in vitro is an approximately 250 kDa complex of NS1 and ApoA1.

CryoEM structure of isNS1

To gain insights into the molecular organization of this native isNS1, we determined the cryoEM structure of isNS1 (Supplementary Table 1 and Supplementary Fig. 4-7). The 2D classes of isNS1ts are observed to be heterogenous spheres with some internal features, some of which resembling the cross-shaped NS1 protruding from the sphere (Supplementary Fig. 4b). However, the 3D reconstruction of isNS1ts alone was observed to have no discernible features beyond a spherical-like density with varying dimensions averaging around 106 Å by 77 Å (Supplementary Fig. 4). To obtain higher resolution structural information on the complex, we collected data for the ternary complexes of the isNS1wt:Ab56.2, isNS1wt:Fab56.2 and the isNS1wt:Fab56.2 (Fig. 2a and Supplementary Fig. 5-6). The use of antibodies as fiduciary markers is a well-proven approach for solving the structures of small proteins^{25, 26}. Overall, both samples resulted in similarly distinguishable 2D class averages (Fig 2b-c) which show asymmetrical binding of one unit of Fab56.2 to the dimeric cross-shaped sNS1 associated with a lower-density sphere, presumably HDL.

Fig. 2.

In the isNS1wt:Fab56.2 ternary complex dataset, we further observed two apparent but rare sub-classes of free sNS1wt dimer bound to two units of Fab56.2 representing 3.5% of the total population of picked particles (Fig. 2c, red dashed boxes). While the cryoEM map reconstructions remain limited at low resolutions (Supplementary Fig. 5-6), they could be fitted with the NS1 dimer and Fab56.2 models (Fig. 2d) predicted separately using AlphaFold^{27, 28}, with a correlation value of 0.75 to the fitted regions using a simulated 5 Å map from the NS1 dimer and Fab models. The AlphaFold predicted model of the NS1 dimer was used despite the availability of the crystallographic dimer model (PBS ID: 6WER) because the latter had a few unmodelled residues (aa 118-128)^{7, 11}. The predicted model upon superimposition on the crystallographic model, nonetheless, has a RMSD of 0.67 Å indicating that these two models were highly similar²⁹. The map-model fitting showed that the epitope sites were at the tip of the C-terminal β-ladder region. The approximate dimensions of the spherical HDL could also be measured at 82 Å by 65 Å (Supplementary Fig 5-6) which aligned with existing reports of spherical HDL being measured at 70-120 Å in diameter with predominantly homodimeric ApoA1 arranged as an anti-parallel double-belt stabilized by intermolecular salt bridges^{30, 31}. We predicted a model of an antiparallel dimer of ApoA1 using AlphaFold2^{27, 28} without the first 58 residues of ApoA1 as the N-terminal domain is highly flexible³⁰. The model has a cross-sectional distance of approximately 78 Å and fits into the spherical density in the cryoEM map. The NS1 dimer appears semi-embedded on the HDL particle through its hydrophobic surface, with a conformation that exposed only one end of the β-ladder domain to which Ab56.2 and Fab56.2 bind. Overall, the cryoEM model indicated that the purified ∼250 kDa isNS1 is a complex of a NS1 dimer and HDL, consistent with the gel and LFQ analysis of LC-MS data presented above.

Similarly, we determined the structures of the ternary complex of isNS1ts mutant with Fab56.2 (Fig. 3a). Interestingly, 2D class averages of isNS1ts:Fab56.2 dataset showed a significant sub-population of a free NS1ts dimer bound to two units of Fab56.2 (53.7%), in addition to the HDL spheres (23.6%) and isNS1ts dimer:HDL:Fab56.2 (22.7%) classes (Fig. 3b and Supplementary Fig. 7). The predicted structure of the isNS1ts dimer bound to Fab56.2 could be fitted with an overall correlation value of 0.5 using map simulated from atoms at 5 Å (Fig. 3c). Achieving atomic resolution to map the Fab binding interface with the NS1 β-ladder domain remains a challenge due to the preferred orientation (Supplementary Fig. 7). A comparison of the cryoEM density maps of free isNS1ts dimer:Fab56.2 (Fig. 3D, grey) and isNS1ts dimer:HDL:Fab56.2 ternary complex (Fig. 3d, yellow) fitted with a correlation value of 0.71 based on the NS1 map region, revealed a pose difference of the Fab56.2 between the free isNS1ts dimer and the HDL-bound isNS1ts dimer (Fig. 3d, inset). Further comparison of the density maps of the isNS1 dimer:HDL population in the Fab56.2 ternary complexes of isNS1wt (Fig. 3e, purple) and isNS1ts (Fig. 3e, yellow) revealed conformational similarities of the isNS1 dimer:HDL complex.

Fig. 3.

XL-MS maps the inter- and intra-molecular architecture of the isNS1

Next to refine our understanding of the NS1 dimer:HDL complex of the isNS1 indicated by the LFQ of LC-MS (Fig. 1d and Supplementary Fig. 3d) and cryoEM results (Fig 2d and 3d), we probed the interactions between NS1 and ApoA1 - the major protein moiety of HDL - using crosslinking mass spectrometry (XL-MS)³² on the immunoaffinity purification eluate fraction (Fig. 4a, Supplementary Fig. 8a-b). We identified 28 NS1:NS1 and 29 ApoA1:ApoA1 intramolecular crosslinks, as well as 25 NS1:ApoA1 intermolecular crosslinks for isNS1wt (Fig. 4b; Supplementary File 2a-b). As depicted in Fig. 4b-c, the NS1 residues that are involved in both intra- and inter-molecular crosslinks were located mainly on the β-roll and wing domains, with ∼34% of the NS1:ApoA1 crosslinks being located on the β-roll domain. The NS1:NS1 intramolecular crosslink pairs were further validated on the predicted NS1 dimer model by implementing a 30 Å cut-off distance for the Cα atoms of crosslinked residues, a criterion that was met by 85% of the NS1:NS1 intramolecular crosslinks (24 out of 28; Fig. 4d). The NS1:ApoA1 intermolecular crosslinks for the immunoaffinity purified isNS1wt were found to be located between the β-roll and wing domains of NS1 and the helices 3, 4, 8 and 10 of ApoA1 (Fig. 4b, d). Applying the same cut-off value of <30 Å distance between the Cα atoms of the NS1:ApoA1intermolecular crosslinked residues and validating it on the cryoEM model of isNS1wt in this study (Fig. 4d-e) satisfied 76% of the crosslinks (19 out of 25 intermolecular crosslinks between NS1 dimer and ApoA1). In the case of ApoA1:ApoA1 intramolecular crosslinks, the <30 Å distance between their Cα atoms cut-off satisfied 62% of the crosslinks (18 out of 29 ApoA1:ApoA1 intramolecular crosslinks). The higher number of violated crosslinks in ApoA1:ApoA1 compared to those in NS1:NS1 could be due to the flexible and dynamic nature of ApoA1 dimers which may prevent it from assuming a fixed relative position among the ApoA1 dimer population in the HDL. Overall, the XL-MS data and structural validation agreed with the cryo-EM model of isNS1 being a complex of the NS1 dimer embedded on a HDL particle composed of an ApoA1 dimer (Fig 4d-e).

Fig. 4.

Additionally, the XL-MS mapping of isNS1ts mutant (Supplementary Fig. 8c-d) followed by rigid-body fitting into the corresponding cryoEM map (Supplementary Fig. 8e) revealed a similar crosslink pattern to isNS1wt, except that there were fewer crosslink sites (Supplementary Fig. 8f, Supplementary File. 2a). This indicated that isNS1ts dimer may have a weaker affinity to HDL in comparison to the isNS1wt dimer. This finding is consistent with the cryoEM data for isNS1ts where we found that free isNS1ts dimers bound to Fab56.2 formed a significant sub- population (34.5%) of the total picked particles (Fig. 3b, Supplementary Fig. 7b).

Given the reported structures of hexameric and tetrameric rsNS1 from mammalian expression systems, insect cells^{11, 15, 16, 29} and Expi293 HEK cells³³, we further examined whether rsNS1 oligomerizes in the presence of exogenously added human HDL, using XL-MS. The crosslinked rsNS1 alone appeared as higher oligomer bands of dimers, tetramers, and hexamers on the SDS- PAGE analysis (Fig. 4f, lane 3) which differed from the crosslinked isNS1 (Fig 4f, lane 3). Interestingly, the addition of human HDL and crosslinker resulted in a smearing band ranging from 30-150 kDa with the loss of the well-defined monomeric rsNS1 band (Fig 4f, lane 4). In the crosslinked rsNS1 in the absence of HDL, we identified 17 NS1:NS1 intramolecular crosslinks at the β-roll and wing domains (Fig. 4f-g, Supplementary File 3a-b). In the presence of human HDL, we identified 30 ApoA1:ApoA1, 4 NS1:NS1, and 6 NS1:ApoA1 crosslinks (Supplementary File 3a-b). Most crosslinks identified are ApoA1:ApoA1 intramolecular crosslinks in the rsNS1:ApoA1 samples (Fig. 4g). Only 1 out of 6 rsNS1:ApoA1 intermolecular crosslinks was within the 30 Å cutoff distance when mapped to the NS1:ApoA1 cryoEM model (Fig. 4e). This was in stark contrast with the isNS1 samples where the number of crosslinks between ApoA1:ApoA1, sNS1:sNS1, and sNS1:ApoA1 was similar (Fig 4b). Overall, this suggests that rsNS1 may form oligomers of dimers but not a complex with HDL under the tested conditions.

Ab56.2 binds at the C-terminal β-ladder region of NS1

The cryoEM map for isNS1wt:Fab56.2 reliably placed the epitope site of Ab/Fab56.2 at the C- terminal b-ladder region, although the precise identification of the epitope site was hindered by the limited resolution of the cryoEM maps (Fig. 2-3, Supplementary Fig. 4-7). Attempts to improve the resolution were challenging so we resorted to using a truncated form of recombinant secreted NS1 consisting of only the C-terminal region (rsNS1c; residues 172-352) to determine Ab56.2 epitope. The rsNS1c fragment of residues 172-352 was previously used for structural studies of dengue^{34, 35} and other flaviviruses, West Nile virus³⁴, Zika virus^{35 36}, and Japanese encephalitis virus³⁷. The epitope region of Ab56.2 was first confirmed by overlapping 15-mer NS1 peptide competition ELISA for Ab56.2 binding (Supplementary Fig. 9a, Supplementary Table 2). The greatest competition was observed at the NS1 peptide spanning residues 316-330 out of 301-340 which showed a decrease in OD450 absorbance (Supplementary Fig. 9b). We further applied the hydrogen-deuterium exchange mass spectrometry (HDX-MS) to monitor the conformational dynamics between Fab56.2 and NS1c to obtain higher peptide-level resolution for the binding interface. Comparative HDX results revealed global scale protection against deuterium exchange of NS1c in the presence of Fab56.2 (Fig. 5a, Supplementary Fig. 10a), indicative of reduced conformational dynamics. Large-scale differences in deuterium exchange were primarily observed across peptides spanning the dimer interface (residues 174-186) and at the tip of the C-terminal β-ladder (residues 286-347) of NS1c (Fig. 5b). Multiple overlapping peptides, spanning residues 286-347 (Fig. 5b), showed protection against deuterium exchange even at the longer labeling times (100 min) in NS1c:Fab56.2 complex (Supplementary Fig. 10a, Supplementary File 4). These results suggest that Fab56.2 bound stably with NS1c. Coupled with the observation that the most significant change in the differential plot was at residues 300-310 (Fig. 5a), these results collectively indicate that NS1c binds stably to Fab56.2 at the primary epitope site of residues 300-310.

Fig. 5.

Comparative HDX results of NS1c-bound and free Fab56.2 revealed protection against deuterium exchange of all three complementarity-determining regions (CDR) of the heavy chain (Fig. 5c, Supplementary Fig. 10b) and CDRL2 of the light chain (Fig. 5d, Supplementary Fig. 10c). Upon binding to NS1c, peptides spanning CDRH3 (residues 89-116) showed the largest magnitude decrease in deuterium exchange at all labeling time points, while CDRH2 showed minor significant differences. In addition, only the CDRL2 loop (residues 49-56) of the light chain of Fab56.2 showed significantly decreased deuterium exchange. Mapping of the deuterium exchange profiles of heavy and light chains onto a model of Fab56.2 (Fig. 5e) revealed that CDRH1-3 and CDRL2 loops were spatially co-localized to form the paratope site to bind NS1c, with the heavy chain being the primary anchor. Collectively, these results indicated that residue 300-310 of NS1 act as the primary epitope site to bind Fab56.2 and Fab56.2 binds stably with NS1c.

Circulating NS1 is in the form of NS1:HDL complex

Lastly, we sought to provide in vivo evidence of NS1 and ApoA1 interaction using sera samples from DENV-infected mice and a human patient. Since we previously showed that DENV2 T164S mutant virus led to more severe disease in the AG129 mouse model²², we used the same infection model to obtain pooled mice serum for immunoaffinity purification of sNS1ts as described in Supplementary Fig. 1. The mouse sera immunoaffinity purified sNS1ts was detected at ∼250 kDa by both anti-ApoA1 and anti-NS1 antibodies in a Western blot analysis following separation on a Native PAGE (Fig. 6a), supporting our mass spectrometry finding that ApoA1 is part of the ∼250 kDa isNS1 protein complex purified from infected Vero cells (Fig. 1d). We next examined if the isNS1:HDL complex can be detected in dengue patients using serum samples obtained from the CELADEN clinical trial³⁸. A primary infection sample with high levels of sNS1 (>15 µg/ml) detected by commercial Platelia™ NS1 ELISA was selected for this purpose to avoid the confounding factor of the presence of anti-NS1 Abs in the secondary infected patients and to pull-down sufficient sNS1 from the patient serum. The patient serum was first subjected to a pre-clearing step using Protein AG resin to reduce the abundance of human IgGs that may interfere with the downstream analysis. We then subjected the pre-cleared patient serum to immunoprecipitation using commercial polyclonal anti-ApoA1 antibodies to pull down and detect ApoA1 and possibly sNS1. Increasing the anti-ApoA1 antibody from 10 to 50 µg coupled onto Protein AG resin for pull-down resulted in a corresponding increase in the sNS1 amount detected by ELISA (Fig. 6b), suggesting a specific association between sNS1 and ApoA1. Collectively, despite the limited amount of sNS1 in sera samples from DENV-infected mice or humans, our data points to the association of ApoA1 with sNS1 in vivo.

Fig. 6.

Discussion

In this study, by integrating biochemical and biophysical approaches including immunoaffinity purification, cryoEM, XL-MS and HDX-MS, we have successfully purified and characterised a native population of isNS1. We determined the cryo-EM structure of a natively secreted NS1 from DENV2-infected cells as a ternary complex of NS1 dimer embedded on a single HDL particle. The structure is validated by in-solution cross-linking mass spectrometry (XL-MS) analysis of the cryoEM samples. The NS1:ApoA1 complex is mediated through defined interactions between the β-roll and wing domains of NS1 and helices 3, 5, 8, and 10 of (bovine) ApoA1 – the main protein moiety of HDL. Additionally, we were able to preliminarily observe NS1:HDL complexes in purified sNS1 from the sera of DENV2-infected mice and further detect and quantify NS1 that co-immunoprecipitated with ApoA1 in the serum of a DENV1-infected patient, further supporting the in vivo and clinical relevance of the NS1:HDL complexes. We also captured a population of free sNS1 dimers that are bound to two copies of Fab56.2 from the same sample. These sNS1 dimers may be separated from HDL complex or the oligomeric form upon Fab56.2 binding. There may be a dynamic equilibrium between the HDL-bound NS1 and oligomeric NS1, regulated by an interplay between NS1 secretion and circulating HDL.

Secreted NS1 from DENV1-infected cells was shown to be a detergent-labile complex of ∼300 kDa that contains lipid species similar to that of HDL¹⁴. Benfrid et al. (2022) recently reported that dimers of rsNS1 bind to HDL when healthy plasma was spiked with purified rsNS1 and subsequently immunoaffinity purified²⁰. Indeed, for the ∼250 kDa PAGE band that was thought to be hexameric^14–16, it could be a complex of dimeric NS1 (∼100 kDa) with HDL (∼164 kDa reported by Benfrid et al., 2022²⁰) composed of ApoA1 homodimer (∼50 kDa) arranged as an antiparallel dimer along with the lipids. Similar to our isNS1 findings, NS1 dimers from the rsNS1 were found to be anchored on the surface of the spheroid human HDL particles, albeit in varying ratios from one to three dimers docked on a single HDL particle ¹⁵. These differences in ratios could be attributed to variations in the concentration of sNS1 used for spiking and immunoaffinity purification. The concentration of the rsNS1 (400 μg/mL) used for spiking in the mentioned study was more than 10-fold higher than that in the concentrated crude supernatant we used for immunoaffinity purification (∼30 μg/mL). Given that circulating spheroid HDL can consist of two or more copies of ApoA1 in a double-belt model or multifoil model³⁹ respectively within the same HDL population, it is theoretically possible that under high concentrations of rsNS1, the NS1 dimers could associate with multiple ApoA1 molecules on the HDL particle. Nonetheless, the structure of a NS1 dimer docked onto a single HDL particle is likely to be more physiologically relevant, considering that the concentration used in our study is closer to the maximum clinically reported concentration of sNS1 in the sera of DENV patients (50 μg/mL)^{3, 40, 41}. Furthermore, the observation that NS1 interacts with HDL regardless of the source of mammalian ApoA1 is unsurprising, given the high sequence similarity of bovine and mouse HDL to human HDL of approximately 79% and 65% respectively (Supplementary Fig. 11) and that the structure of ApoA1 is highly conserved due to the functional significance of the amphipathic α-helices in HDL assembly and lipid interactions.

While we observed a significant number of NS1:ApoA1 crosslinks (29 crosslinks, Fig 4b) via XL-MS in isNS1wt, only 6 NS1:ApoA1 crosslinks (Fig 4g) were detected between rsNS1 and HDL, suggesting weaker interaction between the two species compared to that in the isNS1 complex. However, the reconstitution experiment in the aforementioned study demonstrated that rsNS1 could disassemble into its dimeric building blocks in the presence of human serum, allowing for anchorage onto HDL particles²⁰. These findings collectively suggest that once the higher-order oligomers of NS1 dimers are formed, they become highly stable and may require assistance from other serum proteins to dissociate and bind to HDL. HDL is known to recruit and interact with at least 96 different serum proteins for its functions^{42, 43}. Alternatively, it could be due to the difference between insect cell (Drosophila S2)-expressing rsNS1 used in the reconstitution experiment²⁰ and the commercial HEK293-expressing rsNS1^{10, 44} used in our study for crosslinking with purified human HDL. Hexamers were reported in insect cell-expressing rsNS1^{15, 16} while tetramers were the predominant form from Expi293F-expressing rsNS1³³. Hence, such structural differences may affect the binding pattern between the two different sources of rsNS1. Overall, the intermolecular interaction residues identified using in-solution XL-MS (Fig. 4b) do support the fitted 3D model of isNS1 as a complex of NS1:HDL (Fig. 4d-e; Supplementary Fig. 8e).

High molecular weight recombinant NS1 was also interpreted to be a hexameric NS1 oligomer based on results obtained from SEC, SAXS, chemical crosslinking, negative staining EM methods, and crystallographic packing analysis^{11, 14, 15} depicted in Supplementary Fig. 12a. Shu et al. (2022) recently reported the first high-resolution cryoEM structure of rsNS1 that is predominantly tetrameric with two NS1 dimers interacting with each other via the hydrophobic interactions³³, with only ∼3.1% out of all rsNS1 particles in the hexameric form³³ (Supplementary Fig. 12b). However, these hexamer models highly disagree with our XL-MS data (Fig. 4b, Supplementary Fig. 8c) since the β-roll domain in these models is presumably forming the lipid core with little space (∼17 – 20 Å) for interaction with another protein. Hence, it is improbable to accommodate the ApoA1 at the core of the hexamer which we illustrated in Supplementary Figure 12a-c using a sphere with a radius of 30 Å to represent the maximum range between the Cα atoms that can be cross-linked to the β-roll (residue Ser2). These multimeric NS1 species may still co-exist with NS1:HDL during viral infection given that rsNS1 form oligomers such as tetramers and hexamers in the absence of ApoA1 and its associated lipid content from HDL. Furthermore, the presence of a significant population of free isNS1ts dimers in complex with Fab56.2 and fewer crosslinks identified between isNS1ts and ApoA1 point towards the possibility that isNS1ts is more dynamic in its association with host proteins than isNS1wt. The dynamic picture of the different forms of sNS1 that is further enriched in the presence of anti-NS1 antibodies has been recently reviewed ^{5, 45, 46}.

Finally, the detection and quantification of sNS1 from an ApoA1 pull-down using commercial antibodies in DENV1-infected patient serum³⁸ provided preliminary evidence that HDL-bound sNS1 complexes are clinically relevant. This is irrespective of the source of mammalian ApoA1 which are highly conserved in sequence and length, for which human and bovine ApoA1 shares ∼78% sequence identity (Supplementary Fig. 11). Although the sample size of mice and dengue patient is small and the NS1 antibodies could bind to serum proteins non-specifically, our observation of sNS1 association with ApoA1 from in vivo natural infection is supported by Benfrid et al. (2022) study which detected sNS1 in complex with HDL in the sera of hospitalized DENV patients²⁰.

Overall, our work provided a better understanding of the native structure of sNS1 in DENV pathogenesis. The functional role of HDL-bound NS1 protein in disease is now a topic of contention. Contrary to the known protective roles of HDL such as against endotoxic shock caused by lipopolysaccharides^47–51, Benfrid et al. (2022) showed that sNS1:HDL complex triggers proinflammatory signals when tested with primary human macrophages²⁰. The presence of sNS1:HDL complexes as a predominant population of sNS1 in our in vitro infection model corroborates with a previous report of an interaction between insect-derived recombinant NS1 and the classical HDL/ApoA1-binding receptor known as SR-BI¹⁷, which is expressed on hepatocytes⁴². This interaction in hepatocytes have been demonstrated to trigger endocytosis of sNS1 and subsequent downstream effects including vascular permeability, which is a characteristic feature of severe dengue^{17, 52, 53}. Internalisation of sNS1 is critical for increased virus production in hepatocytes⁵² and NS1-mediated permeability in endothelial cells⁵³. Furthermore, NS1:HDL complexes have been shown to induce the production of pro-inflammatory cytokines in macrophages and their presence has been detected in the sera of hospitalised DENV patients²⁰. Yet, this hypothesis is confounded by other reports. While rsNS1 has been shown to trigger a TLR4-mediated release of proinflammatory cytokines to induce vascular leakage^{54, 55}, Coelho et al. (2021) further showed that lipid-free recombinant ApoA1 neutralizes the proinflammatory effects caused by insect-derived recombinant intracellular NS1⁵⁵. Furthermore, exogenous administration of purified HEK-293 derived NS1 failed to worsen in vivo vascular leakage in sublethally infected mice⁴³. This underscores the need for physiologically relevant studies to reconcile these discrepancies and gain a comprehensive understanding of the role of NS1-HDL complex in DENV pathogenesis.

In conclusion, despite the study limitations that our isNS1wt and isNS1ts were immunoaffinity purified with one specific monoclonal antibody (Ab56.2), our structural model, obtained by CryoEM and verified by XL-MS, of infection-derived sNS1 shows a NS1 dimer embedded on a single HDL particle composed of ApoA1 dimer. Future investigation of this work using infected mouse sera from other DENV strains and different flaviviruses would inform the structural and functional relevance of the HDL-bound NS1 protein complex amongst flaviviral NS1 as well as provide a basis for the development of NS1-directed therapeutics.

Supporting information

Supplementary File

Acknowledgements

We thank the scientific facility support from NTU Institute of Structural Biology and Protein Product Platform. The authors acknowledge the Cryo-Electron Microscopy Facility at Center for Bioimaging Science, Department of Biological Science, National University of Singapore the scientific and technical assistance. Recombinant DENV2 NS1 protein is a generous gift from Dr Shu Bo and Professor Lok Shee Mei. We thank members of the DL and SV labs for their support.

Funding

This research is supported by the Singapore Ministry of Education under its Singapore Ministry of Education Academic Research Fund Tier 2 (T2EP30220-0020) to DL and National Medical Research Council of Singapore (MOH-OFIRG18may-0006; MOH-OFIRG20nov-0002) to SV, A*STAR core funding and Singapore National Research Foundation under its NRF-SIS “SingMass” scheme to R.M.S, Career Development Award 2021 (A*STAR BMRC) to W.W.P., Career Development Fund 2021 (A*STAR BMRC) to S.Z., and LKCMedicine Dean’s Postdoctoral Fellowship to C.B.L.A.

Author contributions

Conceptualization: SV and DL; Investigation: BLAC, AQN, WWP, KWKC, ZS, SSL, MJGW, MMC, JGL, EEO, SV and DL; Writing: BLAC, AQN, SV, and DL with input from all authors.

Competing interests

Authors declare that they have no competing interests.

Data availability

The data that support this study are available from the corresponding authors upon reasonable request. The cryoEM density maps have been deposited in the Electron Microscopy Data Bank (EMDB) under accession codes EMD-36483 (isNS1ts:Fab56.2) and EMD-36480 (Fab56.2:isNS1ts:HDL). The crosslinking MS raw files and the search results can be downloaded from https://repository.jpostdb.org/preview/14869768463bf85b347ac2 with the access code: 3827. The HDX-MS data is deposited to the ProteomeXchange consortium via PRIDE partner repository⁵⁶ with the dataset identifier PXD042235.