The step-by-step assembly mechanism of secreted flavivirus NS1 tetramer and hexamer captured at atomic resolution

Cryo-EM structures of DENV4 NS1 dimer, tetramer, and hexamer

Recombinant DENV1–4 NS1 proteins (residues 1 to 352) were expressed in 293F cells and purified by nickel affinity chromatography and size exclusion chromatography (fig. S1, A to D). The retention volumes of DENV1–4 NS1 on the Superdex 200 Increase column ranged from 10.4 to 10.9 ml, indicating that the purified DENV1–4 sNS1 proteins form oligomers with higher molecular weights than dimer (fig. S1, A to D). The purified DENV4 sNS1 was used for further cryo-EM analysis, and 6561 movies were collected (fig. S2). The two-dimensional (2D) class averages indicated that the sample was a mixture of dimer, tetramer, and hexamer (fig. S2). After ab initio model generation, dimer, tetramer, and hexamer particles were separated through heterogeneous refinement. Final reconstructions yield an NS1 dimer structure (named DENV4_NS1_dimer) at 2.9-Å resolution and an NS1 tetramer structure (named DENV4_NS1_tetramer) at 3.0-Å resolution (Fig. 1, A and B; fig. S2; and Table 1). Two NS1 hexamer structures (named DENV4_NS1_hexamer1 and DENV4_NS1_hexamer2) with slight conformational variation were reconstructed at nominal resolutions of 2.9 and 3.0 Å (Fig. 1, C and D; fig. S2; and Table 1), respectively.

The atomic models of DENV4_NS1_dimer, DENV4_NS1_tetramer, DENV4_NS1_hexamer1, and DENV4_NS1_hexamer2 were built according to the density maps (Fig. 1, E to H). The chains are named Aa, AaBb, and AaBbCc for NS1 dimer, tetramer, and hexamer, respectively. Most of the residues could be traced in the structures, but some residues in the loop region 108–132 (known as the wing tip) are missing because of the density blurring. Residues that could be traced in the wing tips are summarized in fig. S3A. The densities of wing tips in monomer A of structure DENV4_NS1_tetramer and monomers A, a, and b of structure DENV4_NS1_hexamer1 are presented in fig. S3B. The local density map for regions involved in oligomer packing is presented in fig. S4 (A to H). In the structure of DENV4_NS1_hexamer1, some fatty acyl chain–like densities were found sandwiched between dimers Aa, Bb, and Cc. Mass spectrometry (MS) analysis indicated that triacylglycerol could be copurified with the protein (fig. S4I). Therefore, a triacylglycerol molecule was fitted into the density located in the central cavity of the hexamer. As shown in Fig. 1G, the densities of two acyl chains buried in the cavity are continuous. The density of the third acyl chain extending outside the central cavity is discontinuous, suggesting that the third acyl chain of the triacylglycerol molecule undergoes flexibility.

The assembly of DENV4 NS1 tetramer

Full-length DENV4 NS1 contains 352 amino acids, forms 2 α helices and 20 β strands, and is stabilized by six intramolecular disulfide bonds. The NS1 protomer consists of three distinct domains: a β-roll domain (residues 1 to 29, β1-2), a wing domain (residues 38 to 151, β4-7 and α1-2), and a β-ladder domain (residues 181 to 352, β10-20) (Fig. 2, A and B). β3, β8, and β9 form a “connector” domain (residues 30 to 37 and 152 to 180) and connect the wing to the β-roll and β-ladder domains (Fig. 2, A and B). Loop_β8-β9 in the connector domain and loop_β5-β6 in the wing domain are also known as the greasy finger and wing tip, respectively. Assembly of the DENV4_NS1_dimer is mediated by the intertwining of two β-roll domains and the end-to-end stacking of two β-ladder domains from adjacent NS1 protomers (Fig. 2A). An inner surface and an outer surface are defined in the structure of DENV4_NS1_dimer according to previous structural studies on NS1 from other flaviviruses (Fig. 2A) (25). The β-roll domain, loop_β8-β9, and loop_β5-β6 form the “inner” hydrophobic face (Fig. 2A). When superposed on the crystal structure of the NS1 dimer from DENV2 (fig. S5A), the root mean square deviation (RMSD) is 0.84 Å, suggesting that the overall structure of the NS1 dimer from DENV2 and DENV4 is similar to each other.

The structure of the DENV4_NS1_tetramer consists of two NS1 dimers stacked through the inner surfaces and presents a distorted “H” shape (Fig. 2C). The distance between the centroids of monomers A and B is 38.9 Å, while the distance between the centroids of monomers a and b is 50.0 Å (Fig. 2D). When superposed on the structure of DENV4_NS1_dimer, the RMSD values are 0.62 and 0.59 Å for dimers Aa and Bb from DENV4_NS1_tetramer, respectively. Upon the assembly of the tetramer, loop_β1-β2 in the β-roll domain and loop_β8-β9 in the connector domain shift toward and stack with the other dimer (fig. S5B). In addition, loop_β5-β6 in monomer A, which is flexible in the dimer structure, is stabilized in the tetramer structure through interactions with the distal end of the β-ladder domain from monomer B (figs. S3B and S5B).

The buried interface area between dimers Aa and Bb is 1967 Ų. According to the distribution of residues involved in dimer-dimer stacking, a core stacking zone (consisting of the β-roll and the connecter domains) and a peripheral stacking zone (consisting of the distal ends of the wing and β-ladder domains) are defined (Fig. 2C). The core stacking zone could be further divided into three layers, namely, the top layer (top connector domains), the middle layer (β-roll domains), and the bottom layer (bottom connector domains) (Fig. 2C). In the middle layer, the interface between dimers Aa and Bb is mainly stabilized by hydrophobic interactions contributed by the β-roll domains (Fig. 2E and fig. S5C). The β2^a-β1^a-β1^A-β2^A plane packs against the opposite β2^b-β1^b-β1^B-β2^B plane with a dislocation. As shown in Fig. 2E, β1^a, β1^A, and β2^A pack against β2^b, β1^b, and β1^B respectively, while β2^a interacts with the opposite loop_β2-β3 and loop_β8-β9. In the top layer, Trp28^B in loop_β2-β3^B stretches into a small cavity surrounded by β2^a, loop_β2-β3^A, and loop_β8-β9^A, and is sandwiched between F160^A and K14^a (Fig. 2F and fig. S5C). The main chain of Trp28^B and Thr29^B is hydrogen-bonded to the main chain of Leu13^a in β2^a, the side chain of Thr29^B is hydrogen-bonded to the main chain of Cys15^a, and the side chain of Gln31^B is hydrogen-bonded to the side chain of Cys15^a (Fig. 2F and fig. S5C). A hydrogen bond between the main chain of Thr27^B and the main chain of Met162^A contributes to the interactions between loop_β2-β3^B and loop_β8-β9^A (Fig. 2F and fig. S5C). In the bottom layer, interactions between the connector domains from monomers a and b are mainly mediated by hydrophobic stacking between loop_β2-β3^a and loop_β8-β9^b (Fig. 2G and fig. S5C). Compared to the tight stacking in the core stacking zone, loop_β5-β6^A forms a relatively loose stacking with loop_β16-β17^B and loop_β18-β19^B in the peripheral stacking zone (Fig. 2H and fig. S5C). Therefore, the stacking of dimers Aa and Bb in the structure of DENV4_NS1_tetramer1 is mediated mainly by the β-roll and the connector domains.

The assembly of DENV4 NS1 hexamer

The overall structures of DENV4_NS1_hexamer1 are assembled by three NS1 dimers, with the inner surfaces stacked together and the outer surface exposed to the solvent (Fig. 3A). The diameter of the hexamer particle is approximately 100 Å. The centroid-to-centroid distances between monomers A and B, A and C, and B and C are 44.6, 46.9, and 50.8 Å, and those between monomers a and b, a and c, and b and c are 44.9, 44.6, and 50.8 Å, respectively (Fig. 3B). The distance between dimers Bb and Cc is larger than that between dimers Aa and Bb, and Aa and Cc; thus, neither C3 nor D3 symmetry could be fitted into the hexamer. In addition, tilting of NS1 dimers is required for hexamer assembly. The dimer axis is defined by linking the centroids of two monomers in a dimer, and the tile angle of the dimer is evaluated by measuring the angle between two axes. The tilt angles of dimers Bb and Cc relative to dimer Aa are 20.3° and 23.5°, respectively (Fig. 3B). Given the differences in the dimer-to-dimer distances and tile angles, the three NS1 dimers are organized in an irregular pattern around a central axis with variations in the interactions between the dimers. Thus, we believe that the DENV4 NS1 hexamer may be assembled asymmetrically.

The buried interface areas between dimers Aa and Bb, Aa and Cc, and Bb and Cc are 985.6, 913.4, and 386.9 Ų, respectively, indicating that the stacking between dimers Aa and Bb and dimers Aa and Cc is stronger than that between dimers Bb and Cc. In the middle layer of the core stacking zone, the buried interfaces are contributed by the β-roll domains of dimers Aa, Bb, and Cc (Fig. 3A). In this layer, interactions between β-roll domains of dimers Aa and Bb and Aa and Cc are observed but are absent between dimers Bb and Cc (Fig. 3C). The β-roll domains from dimers Aa and Bb interface are held together by two hydrogen bonds between Thr2^b and Gly10^a, Thr2^A and Gly10^B, and hydrophobic interactions between Val6^a and Val6^B (Fig. 3D and fig. S6A). The β-roll domains from dimers Aa and Cc are similarly stacked (Fig. 3D and fig. S6A). The top layer consists of three connector domains from monomers A, B, and C. Residues Met¹⁶² and Phe¹⁶³ in loop_β8-β9 of three monomers stack together to form a central M162F163 hydrophobic cluster (Fig. 3E and fig. S6A). Nearby, two H26W28F160 hydrophobic clusters could be found at the monomer A-B interface and the monomer A-C interface, but not at the monomer B-C interface. The H26W28F160 cluster is formed by hydrophobic interactions between Phe¹⁶⁰ from one monomer and His²⁶ and Trp²⁸ from the neighboring monomer (Fig. 3E and fig. S6A). In the bottom layer, the interaction network between three connector domains is very similar to that in the top layer. A M162F162 hydrophobic cluster stabilizes the stacking in the center, while two H26W28F160 hydrophobic clusters strengthen the Aa-Bb and Aa-Cc stacking (Fig. 3F and fig. S6A).

A small cavity surrounded by the β-roll domains is found in the NS1 hexamer, and the acyl chain–like densities in the cavity allow us to fit a TAG molecule into the cavity (Figs. 1G and 3G). The cavity is sealed by the tightly stacked connector domains at the top and bottom, and by the Aa-Bb and Aa-Cc interfaces at the side. A narrow gate between the β-roll domains of dimers Bb and Cc connects the cavity to the solvent. The two acyl chains of the triacylglycerol molecule are buried in the cavity and form hydrophobic interactions with amino acids in the β-roll domains (Fig. 3G and fig. S6B). The third acyl chain extends out of the cavity through the gate between dimers Bb and Cc. The total buried surface area between the TAG molecule and the NS1 hexamer is 1169.4 Ų, suggesting that the TAG molecule may play a role in stabilizing the hexamer structure.

The structure of DENV4_NS1_hexamer2 is similar to that of DENV4_NS1_hexamer1 but with conformational differences. When dimer Aa from DENV4_NS1_hexamer1 and DENV4_NS1_hexamer2 is superposed, a rotation of 4.5° and 3.9° in dimers Bb and Cc, respectively, is observed (Fig. 3H). The B factors in the distal ends of DENV4_NS1_hexamer2 are higher than those in DENV4_NS1_hexamer1 (fig. S6C), suggesting that these regions are more flexible in hexamer2. This could be further elucidated by the lower local resolution in these regions in the density map of hexamer2 (fig. S2). In addition, the density of TAG in the central cavity of hexamer1 is blurring in hexamer2, suggesting that TAG may bind to hexamer2 with low occupancy.

The assembly of ZIKV and JEV NS1 tetramers

To investigate the oligomeric states of NS1 from other flaviviruses, NS1 proteins from ZIKV and JEV were expressed (ZIKV NS1 was expressed in Sf9 cells; JEV NS1 was expressed in 293F cells), purified, and analyzed by cryo-EM (fig. S7, A and B). 2D averages of ZIKV and JEV sNS1 particles indicated the existence of tetramer and hexamer (fig. S7, A and B). Data collection and processing yielded two ZIKV NS1 tetramers (named ZIKV_NS1_tetramer1 and ZIKV_NS1_tetramer2) in different conformations at resolutions of 2.9 and 3.2 Å, respectively (Fig. 4, A and B; fig. S7C; and Table 1). In addition, a JEV NS1 tetramer structure at 3.2 Å (named JEV_NS1_tetramer) was reconstructed (Fig. 4C, fig. S8A, and Table 1). The atomic models were built for the structure of ZIKV_NS1_tetramer1, ZIKV_NS1_tetramer2, and JEV_NS1_tetramer, and the chains in these structures are named in the order of AaBb. Residues that could be traced in the wingtip are summarized in fig. S8 (B and C).

The overall structures of ZIKV_NS1_tetramer1, ZIKV_NS1_tetramer2, and JEV_NS1_tetramer are assembled by the stacking of the inner surface of NS1 dimers (Fig. 4, A to C). For these three tetramers, the distance between the centroids of monomers A and B is obviously smaller than that between centroids of monomers a and b (Fig. 4, A to C); thus, monomers A and B may stack with each other more tightly. The buried interface area between two dimers of ZIKV_NS1_tetramer1, ZIKV_NS1_tetramer2, and JEV_NS1_tetramer1 is 1161, 1104.4, and 784.4 Ų, respectively. In all three structures, dimer-dimer interactions are mainly mediated by the core stacking zone, while ZIKV_NS1_tetramer1 is further stabilized by interactions in the peripheral stacking zone (Fig. 4, D to F). In the middle layer of the core stacking zone, the relative spatial position of the β2^aβ1^aβ1^Aβ2^A plane and the β2^bβ1^bβ1^Bβ2^B plane is different in three structures (Fig. 4, G to I). In the top layer, the stacking is mainly mediated by the interactions between loop_β2-β3^A and loop_β2-β3^B, while additional interactions contributed by loop_β8-β9 are observed in the structure of ZIKV_NS1_tetramer2 (Fig. 4, J to L).

Proposed models for the step-by-step assembly of NS1 dimer into hexamer

Our atomic models of the DENV4_NS1_hexamer differ from the previously reported model of the DENV2_NS1_hexamer [Protein Data Bank (PDB): 7WUV]. In the DENV4_NS1_hexamer, the assembly is mainly mediated by the β-roll and the connector domains (Fig. 5A). In contrast, in the DENV2_NS1_hexamer, the assembly is mainly mediated by the wing and the β-ladder domains (Fig. 5B, left). To distinguish the two different NS1 hexamers, the DENV4_NS1_hexamer is defined as a side-to-side NS1 hexamer, while the DENV2_NS1_hexamer is defined as a head-to-head NS1 hexamer. Because of the resolution limitations, only the backbones of the wing and the β-ladder domains are modeled for the DENV2_NS1_hexamer (Fig. 5B, left) (24). For the following comparison, a full-length head-to-head hexamer was modeled by fitting three ZIKV NS1 dimers into PDB 7WUV (Fig. 5B, right). The head-to-head hexamer is distinguished from the side-to-side hexamer by the relative positions of the three dimers. In the head-to-head hexamer, the centroid-to-centroid distance between monomers A and B is 36.3 Å, while that between monomers a and b is 75.0 Å (Fig. 5C). The angles of dimers Bb and Cc relative to dimer Aa are 60.2° and 60.9°, respectively (Fig. 5C).

When aligning the four tetramer structures by superposing the Aa dimers, the Bb dimers do not align (fig. S9A). Consequently, we try to define these tetramer structures as distinct conformational or energy states. This classification may allow us to better understand the range of structural possibilities and the energy landscape associated with the formation of different tetramer structures. In the four NS1 tetramer structures, JEV_NS1_tetramer and ZIKV_NS1_tetramer2 are assembled by limited interactions at the core stacking zone. Therefore, the structures of JEV_NS1_tetramer and ZIKV_NS1_tetramer2 are defined as semi-stable tetramer states that may represent the initial contact of two free NS1 dimers. The structures of ZIKV_NS1_tetramer1 and DENV4_NS1_tetramer, in contrast, are further stabilized by the interactions at the peripheral stacking zone. To further define the state of ZIKV_NS1_tetramer1 and DENV4_NS1_tetramer, dimer Aa of these tetramers is superposed on dimer Aa of the side-to-side hexamer (Fig. 5D) and the head-to-head hexamer (Fig. 5E), respectively, and the Cα-RMSD values between dimer Bb of tetramer and hexamer are calculated. Since the Bb dimers are quite similar when superposed, Cα-RMSD of Bb dimers here is used to quantify the average distance without implying structural flexibility of Bb dimers. As shown in Fig. 5D, when superposed on the side-to-side hexamer, the Cα-RMSD value of dimer Bb is 40.79 Å for ZIKV_NS1_tetramer1 and 18.19 Å for DENV4_NS1_tetramer. When superposed on the head-to-head hexamer, the Cα-RMSD value of dimer Bb is 19.34 Å for ZIKV_NS1_tetramer1 and 38.21 Å for DENV4_NS1_tetramer (Fig. 5E). The angles and distances between dimers Aa and Bb also indicate that the conformation of DENV4_NS1_tetramer is closer to the side-to-side hexamer, while the conformation of ZIKV_NS1_tetramer1 is closer to the head-to-head hexamer (Figs. 2D, 3B, 4A, and 5C, and fig. S9B). Therefore, we speculate that DENV4_NS1_tetramer may represent a side-opened transition state from the semi-stable tetramer to the side-to-side hexamer, while ZIKV_NS1_tetramer1 may represent a head-opened transition state from the semi-stable tetramer to the head-to-head hexamer.

Following the structural analysis, the oligomeric states of dimer, semi-stable tetramer, side-opened transition tetramer, side-to-side hexamer, head-opened transition tetramer, and head-to-head hexamer are defined. A three-step process for the assembly of the side-to-side hexamer (steps 1-2-3) and the head-to-head hexamer (steps 1′-2′-3′) is presented in Fig. 5F, respectively. In steps 1 and 1′, dimers Aa and Bb from two NS1 dimers stack together through the core stacking zone to form a semi-stable tetramer. Then, dimer Bb tilts to one side, exposing a side opening and forming the side-opened transition tetramer (step 2). Last, dimer Cc couples to the side-opened transition tetramer to form the side-to-side hexamer (step 3). In the other situation, dimer Bb in the semi-stable tetramer tilts to one head to form the head-opened transition tetramer (step 2′), and then dimer Cc couples to the head opening to form the head-to-head hexamer (step 3′). In addition to the step-by-step assembly pathway, there is also a possibility that three NS1 dimers may directly assemble into hexamers (indicated by step 4 and 4′ in Fig. 5F).

It has been found that there are two DENV4 NS1 hexamers with a slight difference in their conformation. This suggests that each assembly state may have inherent flexibility, making the assembly of NS1 oligomers a dynamic process. To illustrate the potential dynamic and reversibility of the transition between different oligomeric states, we have introduced dotted arrows in Fig. 5F. These arrows indicate that the assembly process can occur in both directions, allowing for the conversion from one conformation to another and vice versa. By incorporating these bidirectional arrows, our aim is to highlight the dynamic nature of the NS1 hexamer assembly and the various conformations it can adopt.

Cell culture

Expi293F suspension cells [Union-Biotech (Shanghai) Co. Ltd.] used for production of recombinant NS1 proteins were cultured in 293 Expression medium [Union-Biotech (Shanghai) Co. Ltd.] and grown in a CO₂ incubator at 37°C with 5% CO₂ maintained on a cell shaker at 110 rpm. Sf9 suspension cells [Union-Biotech (Shanghai) Co. Ltd.] used for production of recombinant NS1 proteins were cultured in Union insect Expression medium [Union-Biotech (Shanghai) Co. Ltd.] and grown in a cell shaker at 110 rpm and maintained at 27°C.

Construction, expression, and purification of DENV and JEV NS1

The coding sequences of DENV 1-4, JEV, and WNV were cloned into the pCAGGS expression vector with the original signal peptide and a C-terminal 6× histidine tag. The mammalian expression plasmids for DENV1–4, JEV, and WNV NS1 were transfected into Expi293F cell lines at a density of ~2.0 million cells/ml with linear polyethylenimine (MAX, Polysciences). Supernatants were harvested 48 to 60 hours after transfection. Proteins were captured by Ni–nitrilotriacetic acid (NTA) affinity resin and washed with 20 mM sodium phosphate (pH 8.0), 150 mM NaCl, and 35 mM imidazole, and eluted with 20 mM sodium phosphate (pH 8.0), 150 mM NaCl, and 250 mM imidazole. The eluted fraction was concentrated using the Amicon Ultra centrifugal filter with a 50-kDa cutoff membrane (Millipore) and further purified in ÄKTA Pure System using a Superdex 200 10/300 GL column (GE Healthcare) equilibrated in 20 mM sodium phosphate (pH 8.0) and 150 mM NaCl.

Construction, expression, and purification of ZIKV NS1

The codon-optimized gene encoding NS1 from ZIKV (GZ02 strain) was synthesized into the baculovirus transfer vector pFastbac1 (GENEWIZ) with an N-terminal GP67 signal peptide and a 6× histidine tag. ZIKV NS1 proteins were produced by infecting suspension cultures of Sf9 cells (Invitrogen) for 48 hours. NS1 proteins were captured by Ni-NTA affinity resin and washed with 20 mM tris-HCl (pH 8.0), 150 mM NaCl, and 35 mM imidazole, and eluted with 20 mM tris-HCl (pH 8.0), 150 mM NaCl, and 250 mM imidazole. The eluted fractions were concentrated using the Amicon Ultra centrifugal filter with a 50-kDa cutoff membrane (Millipore) and further purified in ÄKTA Pure System using a Superdex 200 10/300 GL column (GE Healthcare) equilibrated in 20 mM tris-HCl (pH 8.0) and 150 mM NaCl.

Cryo-EM grid preparation and data collection

For cryo-EM grid preparation, 3.5 μl of purified flavivirus NS1 was applied to the amorphous alloy film (CryoMatrix Au300-R12/13, Zhenjiang Lehua Electronic Technology Co. Ltd.) that was glow-discharged at 15 W for 1 min. The grids were blotted for 3 s at 4°C with 100% humidity using Vitrobot IV (Thermo Fisher Scientific), frozen in liquid ethane, and stored in liquid nitrogen.

The grids containing flavivirus NS1 samples were imaged using a 300-kV Titan Krios Gi3 equipped with a Gatan K3 Summit detector. The movie stacks with 50 frames were automatically collected using SerialEM software at a nominal magnification of ×105,000, corresponding to a pixel size of 0.85 Å. The defocus ranges from −1.2 to −1.8 μm. Each movie stack was exposed in the counted-Nanoprobe mode for 1.96 to 2.48 s, and the total dose rate was approximately 50 to 56 e⁻/Ų.

Cryo-EM data processing of DENV4 NS1

For DENV4 NS1, a total of 6561 movies were collected. The movie stacks were aligned using MotionCorr2 in RELION, and the contrast transfer function (CTF) was estimated using Patch CTF in CryoSPARC. Particles (7,669,350) were auto-picked and extracted in a pixel size of 1.7 Å. 2D classifications were performed to remove junk particles, and 4,608,421 particles were retained. Ab initio model reconstruction generated a dimer, a tetramer, and a hexamer model. Then, particles of dimer, tetramer, and hexamer are separated by heterogeneous refinement.

Particles (386,457 and 837,707) corresponding to the DENV4 NS1 dimer and tetramer were reextracted in RELION in a pixel size of 0.85 Å. After 3D refinement and 3D classification without alignment, 112,941 and 86,204 particles were selected for CTF refinement and polish. The final local refinement in CryoSPARC yielded a 2.9-Å cryo-EM density map for the dimer and a 3.0-Å cryo-EM density map for the tetramer.

For the hexamer, 805,607 particles were reextracted in RELION in a pixel size of 0.85 Å. Then, 3D refinement and 3D classification without alignment were performed in RELION. Two subsets containing 101,485 and 94,411 particles were selected for CTF refinement and polish. The final local refinement in CryoSPARC yielded a 2.9- and 3.0-Å cryo-EM density map for two hexamer structures with different conformations.

Cryo-EM data processing of ZIKV NS1

For ZIKV NS1, a total of 2574 movies were collected and were motion-corrected by MotionCorr2 in RELION. The movie stacks were imported into CryoSPARC for the patch CTF estimation. A total of 3,443,085 particles were auto-picked and extracted in a pixel size of 1.7 Å. 2D classification was performed to remove junk particles, and the remaining 2,460,081 particles were used for ab initio model building and heterogeneous refinement. Two classes of particles corresponding to two tetramer states were separated through heterogeneous refinement. Particles (1,031,375 and 638,113) corresponding to tetramer 1 and 2 were reextracted in RELION in a pixel size of 0.85 Å. After 3D refinement and 3D classification without alignment, 88,058 and 127,526 particles were selected for further CTF refinement and polish. The final local refinement in CyoSPARC resulted in a 2.9-Å cryo-EM density map for tetramer 1 and a 3.2-Å density map for tetramer 2.

Cryo-EM data processing of JEV NS1

For JEV_NS1, a total of 2592 movies were collected. Motion corrections were accomplished by MotionCor2 in RELION. Then, the micrographs were imported into CryoSPARC and the CTF parameters were estimated by Patch CTF. Particles (2,899,763) were auto-picked and extracted in a pixel size of 1.7 Å. Junk particles were removed by 2D classification, resulting in 1,916,160 particles. After ab initio reconstruction and several rounds of heterogeneous refinement, 198,587 particles were reextracted in a pixel size of 0.85 Å and used for the nonuniform refinement, yielding a density map at 3.2-Å resolution.

Cryo-EM data processing of WNV NS1

For WNV_NS1, a total of 2672 movies were collected. Motion corrections were accomplished by MotionCor2 in RELION. Then, the micrographs were imported into CryoSPARC and the CTF parameters were estimated by Patch CTF. Particles (1,175,887) were auto-picked and extracted in a pixel size of 0.85 Å. These particles were used for the subsequent 2D average classification.

Model building and refinement

For structures of DENV4 NS1 oligomers, a model of DENV4 NS1 dimer was initially generated through homogeneous modeling in SWISS-MODEL using the structure of DENV2 NS1 dimer (PDB: 4O6B) as template. One, two, and three copies of the dimer models were fitted into the density maps of DENV4_NS1_dimer, DENV4_NS1_tetramer, and DENV4_NS1_hexamer1 and DENV4_NS1_hexamer2, respectively. The chain IDs were changed, and the chains were merged in Coot to get the oligomer models. After one round of rigid body refinement, one round of morphing, and one round of simulated annealing in Phenix Real_Space_Refinement, residues missing in loop_β5-β6 were manually built in Coot. Then, the models were refined by iterative refinement in Phenix Real_Space_Refinement with default parameters and manual adjustment in Coot. Secondary structure restrains were used for the real-space refinement in Phenix. The geometries of refined models were validated using MolProbity. The figures of the structures were prepared in ChimeraX and PyMOL (Schrödinger LLC).

For the structures of ZIKV NS1 tetramers, two copies of the crystal structure of full-length ZIKV NS1 (PDBV: 5GS6) were fitted into the density maps of ZIKV_NS1_tetramer1 and ZIKV_NS1_tetramer2. Then, the mismatch residues were mutated, the chain IDs were changed, and the chains were merged in Coot to get the tetramer models. The following refinement strategy was similar to that of DENV4 NS1 oligomers.

For the structure of JEV_NS1_tetramer, a model of JEV NS1 dimer was modeled using SWISS-MODEL. Two copies of the dimer models were fitted into the density map of JEV_NS1_tetramer, and the four chains were merged after changing the chain IDs in Coot. The following refinement strategy was similar to that of DENV4 NS1 oligomers.

MS analysis

To extract the lipids from the protein sample, 100 μl of supernatant was added with 300 μl of isopropanol (−20°C precooling) plus 10 μl of SPLASH internal standard solution, vortexed for 1 min, and then kept at −20°C overnight. The sample was subsequently centrifuged, and the supernatant was placed in a vial. For lipid extraction and analysis, a liquid chromatography (LC)–MS system consisting of Waters 2D UPLC (Waters, USA) and Q Exactive mass spectrometer (Thermo Fisher Scientific, USA) was used. The LC was performed on a CSH C18 column (1.7 μm 2.1 × 100 mm, Waters, USA) at 55°C with the flow rate at 0.35 ml/min. Solvent A, a 60% aqueous acetonitrile solution containing 0.1% formic acid and 10 mM ammonium formate, and solvent B, a 10% aqueous acetonitrile solution with 90% isopropanol, 0.1% formic acid, and 10 mM ammonium formate, were used in positive ion mode. In negative ion mode, solvent A, a 60% aqueous acetonitrile solution with 10 mM ammonium formate, and solvent B, a 10% aqueous acetonitrile solution containing 90% isopropanol and 10 mM ammonium formate, were used. Gradient elution was carried out at a flow rate of 0.35 ml/min, with a column temperature of 55°C and a 5-μl sample size.

The MS scan method covered a range of m/z 200 to 2000. The resolution of MS1 was 70,000, with the automatic gain control (AGC) set at 3 × 10⁶, while the maximum injection time was 100 ms. The top three ions were selected for MS2 based on precursor ion intensity. For MS2, the resolution was 17,500, AGC was 1 × 10⁵, and the maximum injection time was 50 ms. The collision energy was applied stepwise, i.e., 15, 30, and 45 eV. For electrospray ionization, a sheath gas flow rate of 40 liters/min and an auxiliary gas flow rate of 10 liters/min were used. The spray voltage was set at 3.8 |KV| for positive ion mode and 3.2 |KV| for negative ion mode. The capillary temperature was maintained at 320°C, and the auxiliary gas heater temperature was set to 350°C.

This PDF file includes: