Introduction

The family of Paramyxoviridae contains many clinically important viruses, such as human parainfluenza virus 1-3, measles virus, and Nipah virus (NiV). Paramyxoviruses code for two glycoproteins on the viral membranes for virus-cell membrane fusion. The receptor-binding proteins (RBP, HN/H/G) engage the host receptors and activate the refolding of the fusion protein (F) that merges the virus and cell membranes¹. Paramyxovirus-F is a class I fusion protein, along with human immunodeficiency virus (HIV) envelope, influenza virus hemagglutinin, and severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2) spike¹. Paramyxovirus-F is a trimeric transmembrane protein with a “tree-like” structure at its prefusion conformation. Upon triggering by the RBP/receptor complex, paramyxovirus-F mediates fusion by inserting its N-terminal fusion peptides into the target membrane and undergoes a conformational change from pre- to post-fusion conformations, similarly to other class I fusion proteins. The refolding of the paramyxovirus-F overcomes the energy barrier for membrane fusion and leads to virus-cell and cell-cell membrane fusion.

NiV is a well-studied paramyxovirus due to its capability of human-human and human-animal transmissions, high mortality and morbidity rates, and the unavailability of vaccines or therapeutics for human use^2,3. NiV is closely related to Hendra virus, Cedar virus, and the newly identified Langya virus⁴. NiV-F is expressed on the cell surface as an fusion-inactive form, activated in the endosome by cathepsin B and L cleavage, and exocytosed to the cell surface as a fusion-active form^5,6. NiV-G binds to ephrinB2 and/or -B3 receptors in host cells and triggers the refolding of F that leads to virus-cell membrane fusion⁷. Three complementary models of the NiV fusion activation mechanism have been proposed in the past. In the first model, the NiV-G and F form a complex before receptor engagement. The F/G interaction maintains F at a prefusion conformation. The receptor binding to G induces conformational changes in G, leading to the release and refolding of F. This model is supported by co-immunoprecipitation and conformational antibody binding assays^8,9. In the second model, NiV-F exists as hexamer-of-trimers, and the fusion peptides are sequestered at the hexameric interface. The activation of one single trimer by the NiV-G/ephrinB2 complex can be transmitted to all six trimers via the interfaces of the hexamer-of-trimers, leading to efficient F-triggering and fusion pore formation. This model is supported by the crystal structure of a GCN4-decorated NiV-F ectodomain and mutagenesis analysis of key amino acid residues at the hexameric interfaces¹⁰. In a third model, NiV-F and G do not form a complex before receptor binding. The NiV-G/ephrinB2 engagement clusters ephrinB2 and triggers F. F triggering leads to reduced mobility of a portion of F molecules that potentially contributes to fusion pore formation. This model is supported by the nanoscale distribution of NiV-F and G on cell membranes resolved by super-resolution microscopy and a single-particle tracking assay of NiV fusion protein at the interface between supported lipid bilayer and live cell membranes^11,12. Although these models explain the NiV membrane fusion mechanisms from different perspectives, a link between the molecular structure and the fusion machinery on biological membranes remains unclear.

We attempted to address this link by analyzing the nano-organization of NiV-F at the prefusion state on biological membranes using single-molecule localization microscopy (SMLM) and mutagenesis analyses. We have shown that NiV-F forms clusters on biological membranes that are isolated from NiV-G¹¹. Here, we show that NiV-F forms regular-sized nanoclusters on cell membranes regardless of the expression level or endosomal cleavage. The estimated size of NiV-F clusters on cell membrane is similar to that of the hexamer-of-trimer assemblies formed by the GCN4-decorated soluble NiV-F. The NiV-F nano-organization is altered by mutations at the hexameric interface and the putative oligomerization motifs at the transmembrane domain. NiV-F molecules enriched in nanoclusters favor membrane fusion activation. The NiV-F nanoclusters are stabilized by the interactions between NiV-F, the endocytosis adaptor complex AP-2, and the clathrin coat at the cell membranes. In summary, our study reveals the NiV-F nano-organization on the biological membranes and provides novel insights into the NiV fusion machinery in situ.

NiV-F forms regular-sized clusters that are not affected by the surface expression level

We recently published that NiV-F formed distinctive nanoclusters on the plasma membrane regardless of the presence of NiV-G or ephrinB2¹¹. To investigate the determinants of the NiV-F nano-organization, we used SMLM to probe the NiV-F at various conditions. SMLM generates super-resolution images by localizing spatially far-apart fluorophores with nanometer precision¹³. PK13 cells have a flat morphology that is suitable for SMLM imaging. PK13 cells express little endogenous ephrinB2 and -B3 receptors for NiV and thus can eliminate any potential receptor-dependent clustering. Cell surface NiV-F was detected using a FLAG tag in its extracellular domain¹¹. The FLAG tags were employed to detect NiV-F because of the availability of high affinity and specificity anti-FLAG antibodies for a high labeling density, which is vital for the reconstruction accuracy of SMLM images¹⁴. To rule out the effect of the FLAG tag on NiV-F clustering, we compared NiV-F clusters formed by NiV-F-FLAG to that of a NiV-F-HA construct. Both HA and FLAG tags were inserted after amino acid residue 104, right before the fusion peptide (Fig. S1A). Visual examination of the SMLM images reveals that NiV-F-FLAG and NiV-F-HA form similar clusters (Fig. S1B). The clustering tendency, indicated by Hopkins’ index, is comparable between NiV-F-HA and NiV-F-FLAG (Fig. S1C), suggesting that NiV-F clustering is not affected by specific epitope tags. To test whether the FLAG tag affects the function of NiV-F, we pseudotyped NiV-G-HA and NiV-F-FLAG or un-tagged NiV-F on the recombinant vesicular stomatitis virus (VSV) with glycoprotein (G) replaced by a Renilla luciferase gene (VSV-ΔG-rluc) and determined the viral entry levels (VSV/NiV pseudoviruses). The virus entry induced by NiV-F-FLAG is comparable to that of untagged NiV-F (Fig. S1D), verifying that the FLAG tag has minimal effect on the membrane fusion activity of NiV-F.

Next, we investigated whether the nano-organization of NiV-F depends on the cell surface expression levels. The total cell surface expression level (CSE) of NiV-F in a selected cell was measured before the cell was subjected to SMLM imaging. The SMLM images show that NiV-F forms similar nanoclusters in both high-expression and low-expression cells (Fig. 1A and B).

Fig. 1.

Our data show that the clustering tendency of NiV-F is independent of its CSE in individual cells (Fig. 1C). To gain a quantitative understanding of NiV-F nanoscale organization, we used a cluster identification method DBSCAN (Density-Based Spatial Clustering of Applications with Noise) that links the closely situated localizations in a propagative manner. We created cluster and density maps using Clus-DoC for representative high-expression and low-expression cells (Fig. 1D and E)¹⁵. Our data show that the estimated diameter of the NiV-F clusters ranges from 20-34 nm, with a peak at 24 nm (Fig. 1F). Nonetheless, the precise values of the cluster diameter are indicative rather than definite in SMLM as it could be affected by the labeling complex and the blinking property of the fluorophore. The distribution of the cluster diameter shows that ∼60% of NiV-F clusters have similar sizes, indicating that NiV-F clusters maintain a quite uniform appearance. Interestingly, we noticed that the size of the clusters (Fig. 1G) and localization density (Fig. 1H) in clusters were similar between high- and low-expression cells, suggesting that the NiV-F organization in nanoclusters is not affected by the expression levels. Moreover, our data show that the high-expression cells have more clusters than the low-expression cells (Fig. 1I). These results suggest that changes in CSE may affect the number of NiV-F clusters on the plasma membrane (Fig. 1I), but not the clustering tendency of NiV-F (Fig. 1C), the size of the clusters (Fig. 1G), or the localization density within clusters (Fig. 1H). These data imply that the distribution and nano-organization of NiV-F are tightly regulated and may be important for membrane fusion activation. We also analyzed NiV-F distribution in different cell lines. HeLa cells express ephrinB2 and -B3, and thus are NiV permissive cells. The overall NiV-F nano-organization seems quite similar on PK13 and HeLa cells, agreeing with our previous results (Fig. S2A)¹¹. The Hopkins’ indices are also similar for NiV-F on PK13 and HeLa cells (Fig. S2B). The DBSCAN analysis shows that the percentage localizations in clusters (Fig. S2C), the relative density defined as the ratio of the localization density in clusters to the total localization density in the selected region (Fig. S2D), and cluster diameters (Fig. S2E) are comparable for NiV-F on PK13 and HeLa cells. These results suggest that NiV-F clusters are minimally affected by the cell lines of expression or the presence of the ephrinB2 and/or -B3.

The NiV-F nanoclusters are not affected by NiV-F cleavage

NiV-F is synthesized in host cells as inactive precursor F₀ and cleaved by cellular proteases cathepsin L and cathepsin B in the endosomes at an acidic pH to generate the fusion-active, disulfide-linked F₁-F₂ construct. The F₁-F₂ is subsequently recycled to the cell surface to induce cell-cell fusion and incorporated into virus particles. We hypothesized that the F₀ precursor and the F₁-F₂ active forms co-existed in the same clusters on the cell surface. To test this hypothesis, we used a pan-cysteine cathepsin inhibitor E64d to inhibit the NiV-F cleavage. E64d inhibited the cell-cell fusion induced by NiV-F and NiV-G in HeLa cells (Fig. S3), agreeing with the previous fusion inhibitory effects in Vero and MDCK cells^5,6. Our SMLM images show that E64d treatment does not result in any significant change in the clustering of NiV-F in HeLa cells (Fig. 2A). The clustering extent of NiV-F was not altered by the E64d treatment, as shown by comparable Hopkin’s indices (Fig. 2B). Additionally, quantitative parameters of the nano-organization such as percentage of localizations in clusters (Fig. 2C), relative density (Fig. 2D), cluster size (Fig. 2E), and the total density of the regions (Fig. 2F) were comparable. Our data show that clusters formed by non-cleaved NiV-F in cells treated by E64d are similar to those formed by a mixture of cleaved and non-cleaved NiV-F in the control cells, suggesting that F cleavage does not alter F clustering on the cell membrane. These data support our hypothesis that the non-cleaved precursor F₀ and cleaved F₁-F₂ coexist in the same cluster on the cell membrane.

Fig 2.

Mutations destabilizing the NiV-F hexamer-of-trimer assembly alter its nano-organization on the plasma membrane

We noticed that the uniformity in the NiV-F cluster morphology resembled the hexameric assembly of soluble NiV-F decorated by GCN4¹⁰. Next, we investigated whether NiV-F clusters were stabilized by the hexameric interface. As reported, L53D and V108D mutants destabilize the hexameric interface and demonstrate decreased membrane fusion ability; Q393L stabilizes the hexameric interface and demonstrates increased fusion ability¹⁰. We inserted the FLAG tag into the ectodomains of these mutants at the same position as that of the NiV-F. We verified that the relative fusion ability to their F-WT counterpart was comparable between the FLAG-tagged and untagged mutants (Fig. S4A and S4B)¹⁰. The FLAG-tagged mutants showed similar CSE levels to NiV-F (Fig. S4C). All mutants showed some levels of processing, although the cleavage of L53D was the least efficient compared to F-WT and other mutants (Fig. S4D).

Additionally, the FLAG-tagged NiV-F mutants showed a similar trend in virus entry as that of cell-cell fusion when pseudotyped to a recombinant VSV virus (Fig. S4E and S4F), agreeing with the previous results obtained using the NiV-F constructs with a C-terminal AU1 tag¹⁰. We also generated VLPs expressing NiV-M-β-lactamase (NiV-M-Bla), NiV-G-HA, and NiV-F-FLAG or mutants for an entry kinetics assay. The entry kinetics was measured in real-time by conversion of green to blue fluorescence resulting from CCF2-AM dye cleavage by the β-lactamase upon the entry of VLPs^16,17. Our results show that the hyperfusogenic mutant Q393L shows more efficient VLP entry than that of WT, while VLPs bearing the hypofusogenic mutant L53D and V108D are less efficient in entry than that of the WT (Fig. S4G and S4H). It is noteworthy that the incorporation of NiV-G seems to be impacted by the expression of L53D in both VSV/NiV pseudovirions and VLPs (Fig. S4F and S4H), and the deficient virus entry induced by L53D may be partly due to the decreased attachment to the target cells (Fig. S4E and S4G). Nonetheless, the level of L53D on the pseudovirions and VLPs seems comparable to that of the WT (Fig. S4F and S4H). These data support that mutations at the hexameric interface significantly impact the fusion activity of NiV-F on both cell and virus surface.

NiV-F mutants L53D, V108D, and Q393L were expressed on PK13 cells and subjected to SMLM imaging. The images show that clusters formed by L53D and V108D are smaller and more dispersed than those of the F-WT and Q393L (Fig. 3A). The Hopkin’s index for Q393L is significantly higher than that of F-WT, while the Hopkin’s indices for L53D and V108D are slightly lower than that of F-WT (Fig. 3B). Consistently, a lower percentage of L53D and V108D localizations are segregated into clusters than that of F-WT, while a similar percentage of Q393L and F-WT localizations form clusters (Fig. 3C). In addition, clusters of L53D and V108D are less packed than that of F-WT, with 48-, 49-, and 55-fold denser than the total localization density in the region, respectively (Fig. 3D). The clusters of Q393L are not significantly denser than that of F-WT (Fig. 3D). Our analysis also shows that L53D and V108D form smaller clusters compared to the F-WT (Fig. 3E). The Q393L clusters are of a similar size to that of F-WT (Fig. 3E). The total density of localizations and the total CSE levels for all constructs are comparable (Fig. 3F and Fig. S4C), suggesting that differences in clustering are not due to variable numbers of localizations or levels of expression. Collectively, these data show that the mutations destabilizing the hexameric interface (L53D and V108D) make NiV-F localizations more dispersed and form smaller and less packed clusters. Q393L that stabilizes the hexameric interface promotes the clustering of NiV-F localizations (Fig. 3B) but does not significantly alter the organization and morphology of individual clusters (Fig. 3C-F). These data suggest that the NiV-F clusters are susceptible to mutations at the hexameric interface. Considering that the uniformed cluster size is similar to the assemblies formed by the soluble GCN4-NiV-F, we propose that NiV-F may organize into hexamer-of-trimer on the plasma membrane.

Fig. 3.

Mutations at the NiV-F hexameric interface affect its distribution on the VLP membrane

To further investigate the organization of F mutants in viral membranes, we imaged the FLAG-tagged NiV-F constructs on VLPs produced by 293T cells expressing NiV-M-GFP, NiV-G, and NiV-F-WT, L53D, V108D, or Q393L mutants. The incorporation of NiV-F is comparable among the NiV-F constructs (Fig. 4A). The L53D and V108D on VLPs are less cleaved compared to that of the F-WT and Q393L, as suggested by the weaker NiV-F₂ bands (Fig. 4A). Notably, the incorporation of NiV-G is largely abrogated in the VLPs that expressing L53D (Fig. 4A).

Fig. 4.

To probe the organization of NiV-F constructs on VLPs, the F-constructs on VLP membranes were stained using Alexa Fluor 647 and subjected to SMLM imaging. The GFP fluorescence on NiV-M was used to locate the VLPs. Fig. 4B shows that F-WT and F-Q393L form distinctive clusters, while localizations of L53D and V108D are more dispersed on a z = 100 nm projection. This agrees with their distributions on the plasma membrane (Fig. 3A). To gain a quantitative insight into the distribution of the F constructs on the three-dimensional (3D) VLPs, we used an algorithm named OPTICS (Ordering Points To Identify Clustering Structure)¹⁸. To identify clusters, the OPTICS algorithm sorts out the localizations by calculating the reachability distance between two adjacent localizations in a propagative manner and generates a sequence of the ordered localizations¹⁸. In the plot of reachability distance vs. the sequence of ordered localizations, a sudden increase in the reachability distance marks the cutoff of a cluster (Fig. S5A). The shade between neighboring peaks represents localizations that are classified in one cluster (Fig. S5A). OPTICS can identify sub-clusters on a 3D VLP that may not be recognized by DBSCAN that uses one fixed parameter for the entire dataset. Fig. S5B shows the OPTICS plots of the localizations of F-WT and mutants on representative VLPs (Fig. 4B).

The noticeable kinks suggest that F-WT and Q393L form distinctive clusters on VLPs, while the irregular, small kinks indicate that L53D and V108D are more dispersed and form smaller clusters on VLPs (Fig. 4B and Fig. S5B). To gain a population insight on the nano-organization of the F constructs on VLPs, we developed a one-dimensional convolutional neural network (1D CNN) to classify the ordered sequence of reachability distances of the NiV-F localizations obtained by the OPTICS algorithm¹⁹. The distribution patterns of the localizations of the F-WT and Q393L partition into one category, and L53D and V108D in another (Fig. 4C). These results indicate that L53D and V108D form smaller and more dispersed clusters than F-WT and Q393L on the VLP membranes, agreeing with that of the plasma membrane (Fig. 3). Our results indicate that the trimer-trimer interaction at the hexameric interface is key in stabilizing the nano-organization of NiV-F on both cell and viral membranes, and NiV-F do not seem to rearrange during the incorporation into VLPs. Nonetheless, our data do not rule out the possibility that the association of nucleocapsids with the plasma membrane during assembly may reorganize NiV-F on the authentic virus membranes. Since the incorporation levels of NiV-M and -G varies among mutants, the nano-organization of NiV-F is likely associated with the virion incorporation or budding.

The NiV-F clusters are dispersed upon the disruption of the interactions between transmembrane domains (TMD)

Evidence shows that TMD of class I fusion protein can self-associate in the absence of the rest of the protein and is important for membrane fusion²⁰. Previous studies show that mutations in the Leucine-Isoleucine Zipper (LI zipper) motif lead to the dissociation of an HeV-F TMD-derived peptide, decreased stability, and the fusion ability of the whole HeV-F protein²¹. As the TMD domains of the NiV- and HeV-F demonstrate a 94% similarity, we mutated leucine (488 and 509) and isoleucine (495 and 502) residues of the LI zipper in NiV-F TMD to alanine (LI4A) (Fig. S6A). Similar to their counterparts of HeV-F, the NiV-F-LI4A showed a decreased F processing (Fig. S6B), cell-cell fusion activity (Fig. S6C and S6D), and CSE levels (Fig. S6E)²¹. The NiV-F-LI4A failed to incorporate into VSV/NiV pseudovirions (Fig. S6F), and thus did not induce any virus entry (Fig. S6G). Visual inspection of the images suggests that LI4A forms bigger clusters than F-WT (Fig. 5A). However, the LI4A seems less likely to cluster than F-WT, indicated by a lower Hopkin’s index (Fig. 5B). Similarly, the portion of LI4A localizations partitioned into clusters is slightly lower than that of WT (Fig. 5C). Interestingly, we noticed that clusters formed by LI4A are significantly bigger (Fig. 5D) than that of F-WT and with a similar localization density within the clusters (Fig. 5E). The data suggest that interactions between TMDs of NiV-F monomers play a role in stabilizing its nano-organization. Dissociation of TMDs by mutations may result in increased space between F monomers and thus leads to bigger clusters on the cell membrane.

Fig. 5.

The NiV-F nanoclusters are stabilized by endocytosis components

NiV-F can be endocytosed for endosomal cleavage. Next, we investigated whether NiV-F clusters are stabilized during endocytosis. NiV-F contains an endocytosis sorting signal YSRL and an additional YY motif at its cytoplasmic tail (Fig. S7A). Mutation of these tyrosine residues to alanine almost diminished NiV-F cleavage⁶. A FLAG-tagged NiV-F-YA construct containing the aforementioned mutations resulted in significantly less F₂ band than F-WT (Fig. S7B) and completely abrogated cell-cell fusion in 293T cells (Fig. S7C and S7D), although the CSE levels were comparable to that of F-WT (Fig. S7E). The YA mutant did not incorporate into the VSV/NiV pseudovirions (Fig. S7F) and thus did not mediate virus entry (Fig. S7G). By visually inspecting the SMLM images, we noticed that F-YA was less clustered than that of F-WT (Fig. 6A). Indeed, the Hopkin’s analysis suggests that F-YA is less clustered than F-WT (Fig. 6B). The DBSCAN analysis shows that lower percentage of F-YA localizations segregate into clusters (Fig. 6C) than that of the F-WT, and clusters formed by F-YA are smaller (Fig. 6D) and less dense (Fig. 6E) than that of the F-WT. Notably, there is no significant difference in the total density for F-YA and F-WT SMLM images (Fig. 6F), indicating that the differences in cluster organizations do not result from overall protein expression or the stochastic blinking properties of the fluorophore. These results suggest that the endocytosis sorting signal of NiV-F may facilitate the enrichment of NiV-F into clusters on the cell surface, preparing them for endosomal cleavage and/or incorporation into viruses.

Fig. 6.

It is established that the YxxФ sorting signal on the endocytosis cargo is bound and enriched by the heterotetrameric adaptor protein (AP) complexes, which further recruit the assembly of the clathrin coat^22,23. We hypothesize that the clustering of NiV-F is promoted by the assembly of the clathrin coat via the NiV-F-AP-2 interactions. As suggested by a yeast two-hybrid analysis, NiV-F interacts with μ1, μ2, μ3 and μ4 subunits of the AP complexes AP-1, AP-2, AP-3, and AP-4, and mutations of Y525, Y542, and Y543 almost completely abolished the interaction between AP-2 and NiV-F²⁴. Indeed, the co-immunoprecipitation assay shows that less AP2μ2 is pulled down by F-YA than that of F-WT, suggesting that F-YA has a reduced interaction with AP-2 (Fig. S7H). Next, we examined the NiV-F clusters in HeLa cells treated by the endocytosis inhibitor pitstop2. Pitstop2 blocks clathrin-mediated endocytosis by obstructing the binding of the accessory protein (e.g. AP-2) and the clathrin terminal domain, and potentially preventing the assembly of the clathrin coat²⁵. The SMLM images (Fig. 6G) show that the NiV-F localizations are more dispersed upon pitstop2 treatment, agreeing with the lower Hopkin’s index (Fig. 6H) and a smaller percentage of localizations in clusters (Fig. 6I) in pitstop2 treated group compared to that of the control group. Interestingly, we noticed that the NiV-F clusters in pitstop2 treated cells were larger than the control cells (Fig. 6J), although with similar densities (Fig. 6K). The total density is consistent between the pitstop2 treated and control groups (Fig. 6L). The overall dispersed NiV-F distribution upon pitstop2 treatment may potentially be caused by the disrupted assembly of clathrin coat. In combination, these results suggest that the enrichment of NiV-F by AP-2 and the subsequent assembly of the clathrin-coated endosomes are important in stabilizing the F clusters on cell membranes. We envision that the endocytosis components may also facilitate the NiV-F clustering in virus particles as previous proteomic studies show the presence of endocytosis related proteins, such as clathrin, AP-2, and dynamin in NiV VLPs^26–28.

Discussion

Here we used SMLM to resolve the arrangement and organization of NiV-F on the biological membranes at a precision of 10 nm. Our data supports the following conclusions: 1) NiV-F is organized into nanoclusters on the biological membranes and this organization is independent of the protein expression level or endosomal cleavage; 2) the NiV-F nano-organization is susceptible to mutations at the trimer interface on the NiV-F ectodomain and the putative oligomerization motif on the transmembrane domain; 3) NiV-F sequestered in nanoclusters favors membrane fusion activation; 4) the interactions among NiV-F, the AP-2 complex, and the clathrin coat assembly stabilize the NiV-F clusters. We propose that the NiV-F nanoclusters are the fundamental unit of the NiV fusion machinery, and this organization facilitates membrane fusion triggering by a mixed population of NiV-F molecules with varied degrees of cleavage and opportunities of interacting with NiV-G/receptor complex.

The nano-organization of NiV-F informs the coordinated membrane fusion triggering by a mixed population of NiV-F. This notion is supported by the following observations from this and previous studies: 1) NiV-F molecules form distinctive, regular-sized nanoclusters on the cell and virus membranes (Fig. 1A-E); 2) the cleaved, active F co-exist with the noncleaved, inactive F in a nanocluster, as suggested by that the NiV-F nanoclusters are resistant to the cathepsin inhibitor, E64d (Fig. 2); 3) the NiV-F and G are segregated into different clusters¹¹; and 4) the full-length NiV-F and -G proteins do not show stable interactions either before or after ephrinB2 activation¹². These observations indicate that the NiV-F triggering may be a result of transient interactions between the F molecules at the edge of the clusters with the adjacent ephrinB2/G complex. Moreover, we observed that the NiV-F sequestering into clusters is favorable for membrane fusion activation (Fig. 3, 4, and 5), highlighting the importance of the spatial arrangement of NiV-F in its fusion activity. Therefore, it is likely that the triggered, fusion-active NiV-F at the cluster peripheral can facilitate other F molecules in the same cluster to fulfill the conformational changes, and thus maximize the energy to merge the opposing membranes. In contrast, the correlation between the clustering pattern of NiV-F and its incorporation into virions is not obvious. Although the YA and LI4A mutants that showed aberrant clusters cannot incorporate into VSV/NiV pseudovirions (Fig. S6F and S7F), the clustering-deficient hexameric interface mutants L53D and V108D were incorporated into VLPs (Fig. 4A) and VSV/NiV pseudovirisons (Fig. S4F) comparable to that of NiV-F-WT. Therefore, the significance of NiV-F clustering for assembly and budding of the virions may be context-dependent.

A previous study revealed a hexamer-of-trimer assembly of soluble, GCN4-decorated, prefusion NiV-F. Using SMLM, we estimated that the NiV-F clusters on the cell plasma membrane had a diameter between 24-26 nm (Fig. 1F). This estimation agrees with the size of the hexameric soluble NiV-F-GCN4 assemblies¹⁰. Notably, the density of clusters (Fig. 1I), but not the size and localization density of individual clusters (Fig. 1G and H), is affected by the surface expression levels of NiV-F, implying that the NiV-F clusters are highly regulated. A collaborative effort of multiple copies of viral envelope proteins is favorable for receptor binding and overcoming the energy barrier for membrane fusion. Nanoclusters formed by viral glycoproteins have been observed for HIV-1^29,30, Herpes virus³¹, and Influenza virus³². The viral restriction factor, serine incorporator protein 5 (SERINC5), inhibits HIV-1 fusion by disrupting the HIV env nanoclusters, highlighting the importance of the nanoclusters in the function of the viral glycoproteins³³. More interestingly, the NiV-F distribution becomes more dispersed, characterized by smaller and less dense clusters, upon the mutation of two key residues, L53 and V108, at the hexameric interface, suggesting that the NiV-F clusters are likely to be the hexamer-of-trimer assemblies on the cell and viral membranes (Fig. 3 and 4).

The hexamer-of-trimer NiV-F is observed on the VLP surface by electron tomography¹⁰. The NiV-F hexamer-of-trimers are arranged into a soccer ball-like structure, with one trimer being part of multiple hexamer-of-trimers. Nonetheless, the hexagonal lattice has been reported for heamagglutinin-esterase-fusion (HEF) proteins on the surface of the influenza C virus³⁴ and Env glycoprotein on the Foamy virus³⁵. The sharing of one trimer among hexamer-of-trimers and conformational arrangements in the ectodomain of viral glycoproteins were shown to stabilize the contacts for lattice formation^10,34,35. In our study, the NiV-F clusters observed on the cell plasma membrane are more likely to be individual hexamer-of-trimers estimated by size (Fig. 1). We speculate that the formation of the hexagonal lattice may be a result of increased concentration of NiV-F on VLP membrane. A study on Measles virus shows only limited high order of MeV-F assemblies at some virus assembly sites in recMeV-(H-118 ∇41×) cells³⁶, however, we have not observed increased NiV-F localizations at NiV-M positive membrane domains on the cell body of NiV-M, F, and G co-expression PK13 cells⁸. Therefore, the formation of NiV-F lattice on the VLP membrane may occur either at a late stage of virus assembly or after budding. Additionally, cellular factors may also play a role in maintaining individual NiV-F hexamer-of-trimers on the plasma membrane, as a proteomic analysis reveals that NiV-F is an interacting partner of many vesicular trafficking and actin cytoskeleton factors^26,28.

We also identified that the endosomal components as key host factors in maintaining the NiV-F nano-organization on the cell membrane. Our data suggest that both the disruption of the F/AP-2 interaction and the inhibition of clathrin assembly by pitstop2 result in dispersed NiV-F distribution. AP-2 is the key adaptor of clathrin-mediated endocytosis. It binds cargo and PtdIns(4,5)P₂-containing membrane via multiple interface, and undergoes conformational changes to append to the clathrin lattice^25,37. Our data indicate that AP-2-NiV-F interaction can enrich and strengthen the F nanoclusters (Fig. 6A-F and Fig. S7H), and potentially facilitate the uptake of NiV-F by endocytosis for endosomal cleavage and activation. It is reported that the clustering of CD44, modulated by N-glycosylations, facilitates the uptake of CD44 by endocytosis. Furthermore, the assembly of the clathrin coat at the plasma membrane may also strengthen the F clusters, because pitstop2 that inhibits the clathrin-coat assembly by blocking the clathrin terminal domain²⁵ prevents the cluster formation of NiV-F. It is plausible that both clathrin and AP-2 facilitate NiV-F clustering on cell and virus membranes because of the similar organization of F on virus and cell membranes and the presence of clathrin and AP-2 in VLPs^26,28.

In conclusion, our observations provide direct evidence on the nano-organization and distribution of NiV-F on the cell and viral membranes and shed lights to the fusion activation mechanisms. Therefore, it would be critical to elucidate host factors that maintain NiV-F clusters and the interacting motif in NiV-F, which may represent a new therapeutic strategy for NiV infections. It would also be interesting to analyze the reorganization of the NiV-F clusters upon triggering by the NiV-G/receptor complex on live cell membranes by combining single-molecule imaging and the supported lipid bilayer technologies¹². Additionally, a precise stoichiometry of individual NiV-F clusters on the cell and virus membranes could facilitate the vaccine design.

Materials and methods

Cell lines and plasmids

PK13, HeLa, Vero and HEK293T cells were cultured at 37°C and 5% CO₂ in DMEM (Sigma-Aldrich, D6429) complemented with 10% fetal bovine serum (Invitrogen, 12483-020). Cells were passaged using phosphate-buffered saline (PBS, Invitrogen, 10010-049) and 0.25% Trypsin-EDTA solution (Invitrogen, 25002-072). Cells were monitored routinely for mycoplasma contamination using a mycoplasma detection PCR kit (ABM, G238). A FLAG (NiV-F-FLAG) or HA (NiV-F-HA) tag was inserted after residue 104 of codon-optimized NiV-F in pcDNA3.1 vector. An AU1(NiV-F-AU1) tag was inserted at the C-terminal of codon-optimized NiV-F in pCAGGS vector. The untagged NiV-F does not contain any epitope tag. NiV-F-YA, L53D, V108D, Q393L, and LI4A were produced by site-directed mutagenesis using the NiV-F-FLAG in pcDNA3.1 vector as a template^6,10,21. AP2μ2-mcherry plasmid is a gift from Christien Merrifield (Addgene # 27672). The HA-tagged NiV-G, GFP-tagged NiV-M, and β-lactamase-NiV-M were constructed previously^7,16,38.

Immunofluorescence for SMLM

For SMLM imaging on cells, 1×10⁵ PK13 or HeLa cells were seeded on coverslips (Marienfeld #1.5H, 18 mm) coated with 2.5 μg fibronectin (Sigma-Aldrich, F4759-2 mg) in a 12-well plate, and transfected with 1 μg NiV-F variants using lipofectamine 3000 (Invitrogen, L3000015) on the following day. E64d was dissolved in methanol and pitstop2 in DMSO as recommended by the manufacturer. E64d was added to cells at 3 hrs post-transfection for a total treatment time of 15 hrs. Pitstop2 was added to cells at 10 hrs post-transfection for a total treatment time of 8 hrs. At 18 hrs post-transfection, cells were fixed with PBS containing 4% paraformaldehyde (PFA; Electron Microscopy Sciences; 50980487) and 0.2% glutaraldehyde (Sigma-Aldrich, G5882-50 ml) for 90 min at room temperature. Cells were treated with signal enhancer image-IT-Fx (Life Technologies, I36933) for 30 min at room temperature, and then blocked using BlockAid (Life Technologies, B10710) for 1 hr at room temperature. The FLAG-tagged NiV-F and mutants were detected by the anti-FLAG mouse monoclonal antibody (Sigma-Aldrich, F1804) and an Alexa Fluor® 647 conjugated donkey anti-mouse secondary antibody (Invitrogen, A31571). Cells were incubated with primary antibody overnight at 4°C, and then with the secondary antibody for 1 hr at room temperature. Each antibody incubation is followed by five PBS washes, 5 min each time. Cells were then fixed in PBS containing 4% PFA for 10 min at room temperature.

SMLM setup, imaging, and data analysis

SMLM setup. Imaging was performed on a custom-built SMLM. Briefly, the microscope was built upon an apochromatic TIRF oil-immersion objective lens (Nikon, 60x; numerical aperture 1.49). Four lasers were used for excitation: a 639 nm laser (MRL-FN-639, 500 mW) for exciting Alexa Fluor^® 647, a 532 nm laser (MGL-III-532-300 mW) for exciting cy3B, a 488 nm laser (MBL-F-473-300 mW) for exciting GFP, and a 405 nm laser (MDL-III-405-100 mW) for reactivating Alexa Fluor 647 and cy3B. For SMLM imaging, the exciting and reactivating lasers were combined into a single path using dichroic mirrors and then expanded and collimated using a beam expander. The incident light beam was focused on the back focal plane of the objective lens. A translation stage allowed the beam to be shifted between the center and edge of the objective lens so that the incident light emerging from the objective lens reaches the sample at various angles. For SMLM imaging of NiV-F at the cell and virus membranes at the cell-coverslip contact, the incident angle of the illumination light was adjusted to near or greater than the critical angle of the water-glass interface to allow total internal reflection and ensure low background fluorescence for single fluorophore detection³⁹. The emission fluorescence was separated using appropriate dichroic mirrors and filters (Semrock) and detected by electron-multiplying charge-coupled devices (EMCCD; Ixon, Andor). The full width at half maxima of the image of the point spread function was magnified in size to ∼3 camera pixels (106 nm per pixel), ensuring the optimal fluorophore localization³⁹. A feedback loop, composed of a tracking camera (Andor Zyla SCOMS camera) and a computer-controlled piezo stage (Thorlabs 3-Axis NanoMax 300), was employed to control the sample drift to <1 nm laterally and 2.5 nm axially⁴⁰.

Before SMLM imaging, fluorescence beads (Life technologies, F8799) were added to samples as fiducial markers for drift control. Samples were immersed in imaging buffer [TN buffer (50 mM Tris (pH 8.0) and 10 mM NaCl), 0.5 mg/ml glucose oxidase, 40 μg/ml catalase, 10% glucose, and 50 mM mercaptoethylamine (MEA)]¹⁴. The expression level of the protein of interest in individual cells was determined by measuring the average emission fluorescence intensity of an area of 27×27 μm². To ensure sample stability during SMLM, a fiducial marker was continuously tracked by the tracking camera. A brief procedure for SMLM imaging is described as follows: 1) the sample was exposed to the imaging light beam [639 nm laser (1 kW/cm²)] at a 90° incident angle to cause a subset of the fluorophores to switch to the fluorescent state; 2) the incident light angle was adjusted close to or at the critical angle at the water-glass interface to achieve total internal reflection; 3) a low density of activated fluorophores were recorded to avoid fluorophore overlap; 4) another subset of fluorophores was stochastically activated by exposure to the activation light (405 nm). This process was repeated until 40,000 images were acquired. Custom-written software in MATLAB (Mathworks) was used to identify activated fluorophores in each image, determine their precise locations by Gaussian fitting, and reconstruct SMLM images³⁹. Briefly, each image was first convolved with a Gaussian kernel to remove noise and background, and thresholded to search for local maxima. Next, two fitting steps were applied to determine the peak center position, the amplitude, and the width.

Then, filtering and trail generation were combined to remove erroneous peaks and group multiple peaks originating from one activated fluorophore. Lastly, a list of fluorophore positions was reconstructed into a SMLM image. Clusters of NiV-F localizations on cell surface resolved by SMLM were identified and characterized using ClusDoC¹⁵. The min points and ε were set at 4 and 20, respectively.

VLP production and immunofluorescence

To produce NiV VLPs, HEK293T cells were transfected by NiV-F-FLAG or mutants, NiV-G-HA, and NiV-M-GFP at a 9:1:5 ratio by using polyethylenimine (PEI) at 1 mg/ml (Polysciences, 23966-100). At 48hrs posttransfection, the supernatant of the cell culture was collected and subjected to ultracentrifugation on a 20% sucrose cushion at 125, 392 rcf for 90 min. The VLP-containing pellets were resuspended in 5% sucrose-NTE buffer. VLPs were bound to 2.5 μg fibronectin-coated 18 mm coverslips at 4°C overnight, followed by fixation using PBS containing 4% PFA and 0.2% glutaraldehyde. The protocols of immunofluorescence, SMLM setup, and SMLM imaging for cells were followed to stain and image NiV-F-FLAG and mutants on VLPs using SMLM. A widefield image of GFP was acquired and superimposed on the SMLM image of NiV-F. VLPs were identified as GFP-positive particles. The distribution of NiV-F localizations on VLPs was analyzed by using a custom-written OPTICS algorithm based on C++ and Point Cloud Library. Briefly, OPTICS segregates closely-situated data points into clusters from a 3D SMLM dataset¹⁸. OPTICS algorithm depends on two parameters: 1) ε describes the radius for searching neighbors, and the number of neighbors of point p is called Nε(p); 2) minPts is a user-defined parameter; Point p is a core point if it has more than p neighbors around it (including point p itself). The core-distance and reachability-distance are calculated for cluster segregation. The core-distance is the Euclidean distance of the minptsth closest points to point p (Equation (1)).

The reachability-distance of another Point p to Point p is either the Euclidean distance between point p and point p or the core-distance of p, whichever is bigger (Equation (2)).

The process begins by randomly selecting a point (point A). If the number of neighbors of point A is larger than the user-defined minPts (set at 10), the algorithm outputs point A without a reachability-distance value. Next, OPTICS obtains point A’s nearest neighbor (point B) and outputs the B-A reachability-distance value. Point B is the new starting point. OPTICS recalculates the reachability-distance of all unprocessed neighbors of point B and selects the next point to process, as described above. This process repeats until all points in the dataset have been visited. The final output of OPTICS is a sequence of reachability-distances for each point.

The classification of the ordered sequence of reachability-distances was performed by using custom-written one-dimensional convolutional neural network (1D CNN) based on Python and Tensorflow. Briefly, CNN extracts features from OPTICS output dataset by using convolutional layers for subsequent classification of the space information hidden in the OPTICS datasets. The convolutional layers use a 1D sliding window to slide along the sequence data⁴¹. Each sliding window captures local information and passes it to the next layer. Multiple CNN layers work sequentially to build a high-dimensional representation of the dataset. A fully connected layer that transforms high-dimensional features into classification results is achieved by CNN convolution. We use t-SNE (t-distributed stochastic neighbor embedding) to reduce dimensionality on a trained neural network layer for visualization in Fig 4C⁴².

SDS-PAGE, and western blot analysis and co-immunoprecipitation

For western blot analysis to confirm the expression of NiV-F-FLAG and mutants, HEK293T cells were seeded in 6-well plate and transfected with 2.5 μg pcDNA3.1 empty vector, NiV-F-FLAG, or mutants. At 28 or 48 hrs posttransfection, cells were lysed in RIPA buffer (Millipore-Sigma, 20-188) supplemented with protease inhibitor (Sigma-Aldrich, 11836170001) on ice for 30 min. The cell lysates were collected after centrifuge at 16,000 ×g for 20 min at 4°C. The cell lysates were supplemented with 1x SDS loading dye [60 mM Tris-HCl (pH=6.8); 2% SDS; 10% glycerol, 0.025% Brophenol blue] and 15 mM DTT (Thermo Scientific, R0861), and heated at 95°C for denature. The denatured cell lysates were loaded into 10% polyacrylamide gels for SDS-PAGE. Proteins were transferred to PVDF membrane, pore size 0.45 μm (Cytiva, GE10600021).

Membrane was blocked with 1% Bovine serum albumin (Sigma-Aldrich, A9647-50G) in PBS and incubated with anti-FLAG mouse monoclonal antibody (Sigma-Aldrich, F1804). An HRP-conjugated goat anti-mouse secondary antibody and the clarity Western ECL substrate (BioRad, 1705060) were used for protein detection. Images were acquired using Chemic Doc MP Imaging System (BioRad).

For co-immunoprecipitation experiments, HEK293T cells were transfected with the following combinations 1) 2.3 μg empty pcDNA 3.1 vector and 0.2 μg AP2μ2-mcherry, 2) 0.2 μg AP2μ2-mcherry and 2.3 μg NiV-F-FLAG, and 3) 0.2 μg AP2μ2-mcherry and 2.3 μg NiV-F-YA. At 48 hrs post-transfection, cells from 1 well of a 6-well plate were washed with PBS and lysed in 200 μl lysis buffer provided with the μMACS DYKDDDDK isolation kit, and supplemented with protease inhibitors. Cells were isolated on ice for 30 min. Cell debris was removed by centrifuge at 16,000 x g for 20 min at 4°C. 60 μl of cell lysate was set aside for immunoblot analysis, and the rest was used for immunoprecipitation, as recommended by the manufacturer. 6 μl anti-DYKDDDDK microbeads (Miltenyi Biotec, 130-101-591) were added to 140 μl cell lysates and incubated for 30min on ice. μ columns (Miltenyi Biotec, 130-101-591) were prepared according to the manufacturer’s instructions. Lysate ran over the columns, and microbeads were washed according to the manufacturer’s instruction. 20 μl preheated elution buffer (95°C) was added to the column before eluting the bound immunoprecipitated protein in 50 μl elution buffer. Elute was separated by 10% SDS-PAGE, and proteins were immunoblotted by mouse anti-FLAG and rabbit anti-mcherry primary antibodies. The HRP-conjugated goat anti-mouse and goat anti-rabbit secondary antibodies were used for protein detection.

Flow cytometry

HEK293T cells were seeded in 6-well plate and transfected with 2.5 μg pcDNA3.1 empty vector, NiV-F-FLAG, or mutants. Cells were collected in 1 mL phosphate-buffered saline (PBS) containing 10 mM EDTA after 28hrs post-transfection. The collected cells were incubated on ice for 1 h with primary antibody mouse anti-flag diluted in 1:200. Samples were washed twice in fluorescence-activated cell sorting (FACS) buffer (0.1% FBS with PBS). After washing, samples were incubated with fluorescent anti-mouse Alexa Fluor 647 antibody 1:400 on ice for 45 min. After being washed twice again, samples were read on a flow cytometer (Attune® NxT™ Acoustic Focusing Cytometer, Thermo Fisher) The results were analyzed through FlowJo. The mean fluorescent intensities (MFI) were normalized to the MFIs of the NiV F-WT.

Cell-cell fusion

HEK293T cells were seeded in the 12-well plate and transfected with 1μg total DNA of NiV-G and wt-F or mutant-F with a 3:1 ratio using Lipofectamine 3000. Cells were fixed with 4% PFA after 18 hrs post-transfection. Four or more nuclei within a common cytoplasm were considered syncytium. Syncytia were quantified by counting the number of nuclei in the syncytium per 10 × filed (5 fields are counted per group) under microscope TE2000U.

Viral entry kinetics assay

HEK293T cells were loaded with CCF2-AM in the loading buffer by using a CCF2-AM loading kit (Invitrogen) and incubated at room temperature for 90 min. Loaded cells were washed three times and resuspended in 60 μl of cold fusion buffer (25mM HEPES, 2mMGlutamine, 1mM CaCl2, 2.5mM probenecid in HBSS). 110 μl fusion buffer, 10 μl of VLPs, and 60 μl of loaded cells were added into a black, clear-bottom 96-well plate on ice. The VLPs were allowed to attach to target cells at 4°C for 1 hr. Viral entry was measured by a Tecan fluorometer preheated to 37°C. Fluorescence was recorded every 3 min for 4 h. A blue-to-green (B/G) fluorescence ratio was determined for each well at each time point. The background control B/G ratio of VLPs expressing only MG and MF was then subtracted from each sample ratio as indicated, and the B/G ratio was plotted against time.

VSV/NiV pseudovirus production, quantitative PCR (qPCR), and viral entry quantification

HEK293T cells in a 10 cm dish were transfected with NiV-F-Flag and NiV-G-HA. pCDNA3.1 vector was transfected as the negative control. At 18-20 hrs post-transfection, cells were infected by a recombinant VSV-ΔG-renilla luciferase virus in infection buffer (0.1% fetal bovine serum in PBS). After 48 hrs, the supernatant was collected and concentrated by ultra-centrifugation through a 20% sucrose cushion in NTE buffer at 125,000 × g at 4℃ for 90 min. The supernatant was removed and the pellet was re-suspended in the appropriate amount of 5% sucrose in NTE buffer. The expression levels of NiV-F and -G proteins in pseudoviruses were analyzed by Western blot, and the VSV genome copy number was quantified by qPCR. Briefly, the viral RNA of NiV/VSV pseudoviruses was extracted by using a Qiagen viral RNA extraction kit (QIAGEN, 52904), reverse transcribed to cDNA by the SuperScript III first-strand synthesis system (Invitrogen, 18080051). The VSV viral cDNA was quantified by using a qPCR Taqman probes (FAM-CGG TAT TTT TCC ATA ATT CAA GTA ATC TGC T-TAMRA) and the VSV-P plasmid was used as a standard to establish the standard curve. To quantify virus entry, Vero cells in a 96-well plate were infected using 10-fold serial virus dilutions. Infections were done for 2 h in infection buffer. After 24 hrs, the luciferase activity was measured using a Renilla luciferase assay kit (Promega E2810). The relative light units (RLUs) were plotted against the viral genome copy number per milliliter in GraphPad Prism.

Acknowledgements

We thank the multiscale imaging facility for instrument use. We thank Dr. Youssef Chebli at McGill University and Dr. David Scriven at the University of British Columbia for the support on laser scanning confocal microscopy and single-molecule imaging, respectively. Funding. This work is supported by funding from the Canadian Institutes of Health Research to Q.L. (183861) and the Coronavirus Variants Rapid Response Network to Q.L. (175622). Author contribution: Conceptualization: Q.W. and Q.L.; Image acquisition: Q.W.; Data analysis and code writing: Q.W., J.L., and Q.L.; Functional analysis: Q.W., Y.L., V.K., G.L.M., J.W.; Resources: Q.L., Supervision: Q.L.; Funding Acquisition: Q.L.. Writing: original draft: Q.W., J.L., and Q.L.; revision: all authors. Competing Interests: The authors declare that they have no competing interests. Code, reagents, and data availability: Custom-written codes are deposited in GitHub or available upon request. Reagents can be provided by the corresponding author on a completed material transfer agreement. Requests for the reagents and data should be submitted to qian.liu3@mcgill.ca. Declare of interests. The authors declare no competing interests.