Hemagglutinin and neuraminidase induce membrane zippering in influenza A virus infected cells
To investigate the structural context of vRNP clustering sites, we performed in situ cryo-electron tomography (cryo-ET) of cryo-focused ion beam (cryo-FIB) milled A549 human lung epithelial cells infected with influenza A/Puerto Rico/8/1934 H1N1 virus (hereafter, PR8) or with a pandemic H3N2 strain A/Hong Kong/1/68 (hereafter, HK68). At 16 hours post-infection (hpi), we observed extensive membrane remodelling in all infected cells (fig. 1 a–d). Strikingly, we were able to detect two different types of membrane pairing, both resembling a zipper, forming membrane sheets and double membrane vesicles of variable sizes. Since zippers could be distinguished by structurally distinct luminal protein arrays obvious in the tomograms, we classified them as zipper type I (fig. 1 a, b; supplementary video 1) and type II (fig. 1 c, d; supplementary video 2). While type I was formed by a single layer with C4 symmetry and membrane-to-membrane distance of 34 nm (fig. 1 e), type II consisted of two parallel arrays with membrane-to-membrane distance of 38 nm (fig. 1 h). Type I zippers were found only in HK68-infected cells, whereas type II zippers were found in all cells infected with PR8 and HK68.
Subtomogram averaging revealed that type I zippers were formed by neuraminidase (NA) forming a crystalline array with lateral spacing of 10 nm (fig. 1 e–g) and head-to-head angular displacement of 29° (fig. 1 k, l; supplementary video 3). Furthermore, we could demonstrate that type II zippers were formed by two opposing hemagglutinins (HA) in pre-fusion conformation (fig. 1 m, n). While the NA zipper was formed by NA-NA lateral interaction with alternating up and down NA headgroups, the HA zipper was formed by HA-HA ectodomain head-to-head interactions. Importantly, we could show that NA zippers can be connected to the plasma membrane (fig. 1 b), thus demonstrating that NA delivery to the plasma membrane in HK68 infected cells is driven by NA unzippering. Since NA zippers were not observed in PR8-infected cells, we sought to investigate possible structural differences between NA-PR8 (subtype 1) and NA-HK68 (subtype 2) in the NA-NA interacting regions. We used AlphaFold234 to predict HK68 and PR8 NA structures and fitted them into the subtomogram average (fig. 1 k, l). The prediction of the NA tetrameric headgroup corresponded well to experimentally determined NA but the stem region was predicted unstructured. The structure predictions revealed differences in surface charge distribution between PR8 and HK68 NA which could explain the lack of NA-NA zippering in PR8 infection (extended data fig. 1). Interestingly, a previous study showed that the expression of NA subtype 2 forms filamentous virus-like particles27 indicating that NA subtype 2 can form organised arrays.
In contrast to the symmetric NA arrays, HA arrays showed HA-HA lateral spacing varying between 6 and 9 nm (fig. 1 i, j) which is consistent with HA-HA spacing on isolated virions35,36. HA-remodelled membranes were highly variable in size, with approximate diameters between 60 and 1220 nm (extended data fig. 3 a). Based on the extent of HA-zippering, we classified the HA-remodelled membranes into (i) single membranes coated with HA, (ii) double membranes with fully zippered HA and (iii) partial zippers displaying both phenotypes in the same membrane. 46% of observed HA-membranes were single membrane vesicles, whereas 28% were partially and 26% fully zippered double membranes that formed flat cisternae, autophagy-like compartments or double-membrane vesicles (extended data fig. 3 b). Since not all HA-containing organelles formed an HA-zipper (fig. 2 a, b; extended data fig. 2 a, b), we termed all HA-containing organelles HA-remodelled organelles and only those which show HA-HA zippering are termed as HA-zippers. Further, we also tested if HA-remodelling is dependent on other viral factors such as M2 which was shown to modulate autophagy in IAV-infected cells37. To this end, we performed in situ cryo-ET on HEK-293T cells transfected with HA, which showed that HA alone induces membrane remodelling indistinguishable from that observed in infected A549wt cells (extended data fig. 2 c). Overall, our data suggest that HA and NA form arrays inducing membrane remodelling. We propose that the HA-induced membrane remodelling is a general hallmark of IAV infection and is solely induced by the HA glycoprotein.
vRNPs interact with HA-remodelled membranes
On the cytoplasmic sides of HA-remodelled membranes, we regularly observed cylindrical densities with helical striation reminiscent of vRNPs (fig. 1 h; supplementary video 2). Subtomogram averaging of the densities confirmed the double-helical structure with major and minor grooves of 85 Å and 65 Å (fig. 1 o) which is consistent with the previously published structure of vRNPs from isolated virions (fig. 1 p)3. vRNPs formed variably dense accumulations associated with 74% (n=109) of HA-remodelled membranes (extended data fig. 3 c) and only a single event was observed where vRNPs interacted with a vesicle not containing HA (extended data fig. 2 g). Interestingly, while vRNP accumulations were also observed at 8 hpi on 64% (n=36) of HA-remodelled membranes, we observed 8 events of vRNPs interacting with non-HA containing membranes showing budding of coated vesicles that likely represent ER exit-sites (extended data fig. 2 d). This indicates that vRNPs accumulate on HA-remodelled membranes during later stages of infection. In the case of HA-remodelled double membranes, vRNPs were found on both sides of the double membrane (fig. 1 c, h). A subset of vRNPs was directly in contact with HA-remodelled membranes, while vRNPs that were not directly interacting were found close to membrane-interacting vRNPs. Occasionally, vRNPs associated with HA-remodelled membranes formed a zone devoid of any cellular material such as ribosomes (extended data fig. 2 e, f), resembling a biomolecular condensate (supplementary videos 4, 5). Thus, our data agree with previously published studies23,38–40 showing NP signal in the vicinity of ER-exit sites. NA-zippers found in HK68 infected cells (n=5) did not carry vRNPs besides one single example where vRNPs were found on a NA-zipper in a region devoid of NA (extended data fig. 2 h). Hence, our data demonstrate that HA-remodelled membranes play a major, yet unexplored role in cytoplasmic vRNP transport and clustering.
vRNP density is increased upon interaction with HA-remodelled membranes in a Rab11a-dependent manner
The small GTPase Rab11a has been extensively demonstrated to be essential for all stages of cytoplasmic vRNP transport14,15,41. This suggests that Rab11a may also be important for vRNP delivery to HA-remodelled membranes, or that HA-remodelled membranes themselves are the trafficking organelle delivering vRNPs to the plasma membrane. To test this hypothesis, we performed in situ cryo-ET on PR8-infected A549 cells constitutively overexpressing Rab11a wild-type (A549-Rab11wt) or Rab11a dominant-negative S25N mutant (hereafter, A549-Rab11dn) fused to green fluorescent protein (GFP)42. Both cell lines displayed the same HA-driven membrane remodelling as A549wt cells (fig. 2 a–c), with no significant differences in sizes of HA-remodelled membranes (extended data fig. 3 a). All aforementioned phenotypes of HA-zippering observed in A549wt cells were also found in A549-Rab11wt and A549-Rab11dn cells (extended data fig. 3 b). Notably, we observed a reduced association of vRNPs with HA-remodelled membranes in A549-Rab11dn cells. In contrast to A549wt and A549-Rab11wt cells where 77% and 90% of analysed HA-remodelled membranes carried vRNPs, respectively, we observed only 40% of HA-remodelled membranes with vRNPs in A549-Rab11dn cells (extended data fig. 3 c). In addition, the number and density of vRNPs apparent in the tomograms were lower in A549-Rab11dn cells. Overall, our data show that vRNP accumulation on HA-remodelled membranes, but not the membrane remodelling itself, depends on the activity of Rab11a.
To gain insights into vRNP-vRNP interactions and their spatial distribution on HA-remodelled membranes, individual vRNPs were segmented and subjected to nearest-neighbour analysis (fig. 2 d–i). This allowed close inspection of vRNP-vRNP interactions in the context of membrane binding and quantitative assessment of the degree of clustering. The analysis confirms a significant increase in vRNP-vRNP distances in A549-Rab11dn cells compared to A549wt (p = 4.9e-07) and A549-Rab11wt (p = 9.8e-08) cells, even though the vRNP-membrane distances are only significantly different between A549-Rab11dn and -Rab11wt cells (p = 1.7e-3) (extended data fig. 3 d, e). These differences prompted us to investigate the clustering of vRNPs in more detail and validate clustering against a model with randomly distributed vRNPs. To this end, vRNP clusters were identified based on vicinity using a threshold of 4 nm. This revealed vRNP clusters of up to 15 vRNPs on HA-remodelled membranes in A549 and A549-Rab11wt cells (fig. 2 m, n) in contrast to only 4 vRNPs in A549-Rab11dn cells (fig. 2 o). Furthermore, increasing cluster size correlated with a shorter distance to HA-remodelled membranes (fig. 2 p). Clustering is reduced in a random vRNP model, independent of the vicinity threshold (extended data fig. 3 f, g). Overall, this data showed that HA-remodelled membranes serve as a platform that increases vRNP density at the HA-membrane surface.
We next aimed to substantiate our hypothesis that HA-remodelled membranes themselves are the Rab11a-positive trafficking organelle that delivers vRNPs to the plasma membrane. To this end, we tested the localization of Rab11a by immunofluorescence confocal microscopy on PR8 infected A549-Rab11wt cells immunostained against HA and NP, the latter as a proxy for vRNPs. In agreement with numerous previous studies15,16,41, Rab11a and NP colocalized into large puncta distributed throughout the cytoplasm (extended data fig. 4 a–d). Surprisingly, we observed a low colocalization of HA with Rab11a (extended data fig. 4 e). Segmentation analysis revealed that 18.5% and 16.7% of HA puncta were interacting with Rab11a and NP puncta, respectively (extended data fig. 4 f). We therefore sought to validate the localization of Rab11a by performing cryo-correlative light and electron microscopy (CLEM) to localize GFP-Rab11a directly on cryo-FIB milled lamellae. This allowed us to direct cryo-ET data acquisition towards sites positive for GFP-Rab11a. Of 21 tomograms acquired at GFP-Rab11a positive areas, 86% contained HA-remodelled membranes (extended data fig. 4 g–j). Our data indicate that Rab11a binds to HA-remodelled membranes to deliver vRNPs.
M1 forms cylindrical assemblies in the nucleus which disassemble in the cytoplasm
In the current model of the IAV replication cycle, vRNPs are assembled in the nucleus of infected cells and exported to the cytoplasm in a complex with nuclear export protein (NEP) and M143. In agreement with this model, we resolved a cluster of vRNPs in the nucleus of an A549 cell infected with PR8 at 8 hpi (fig. 3 a, b, g, h; supplementary video 6), confirming that vRNPs are fully assembled in the nucleus. In addition to vRNPs, our data on the nuclei of both PR8 and HK68 infected cells revealed hollow cylinders (fig. 3 c, d, i–l) with lengths between 65 and 285 nm (fig. 3 o) and inner diameters of 21 nm. Their outer diameter was highly variable, between 47 and 61 nm. Radial averaging of cylinder top views revealed 2 to 4 layers with a spacing of approx. 7 nm (fig. 3 p). These assemblies structurally correspond to previously reported M1 cylindrical structures assembled using a purified M1 protein at high salt concentration28. Similar cylindrical structures were also found inside virions upon acidification as an M1 layer disassembly product44 and in H1N1 pdm2009 infected cells45. Collectively, this shows that M1 cylinders are present in cells infected by different IAV strains. Based on the measured nuclear pore complex (NPC) diameter in infected cells (110–155 nm), the M1 cylindrical assemblies can pass through the NPC. This is further supported by our data showing M1 cylindrical assemblies in the cytoplasm in the vicinity of HA-remodelled membranes (fig. 3 e, f, m, n; supplementary video 7) and close to viral budding sites (extended data fig. 5). Interestingly, cytoplasmic M1 assemblies were significantly shorter (fig. 3 o) and more often loosely coiled (nucleus: 8%; cytoplasm: 31%). This indicates that M1 forms cylindrical assemblies in the nucleus that upon export to the cytoplasm become unstable and disassemble. Multilayered M1 cylinders are stabilized by hydrophobic interactions between M1 N-terminal domains28. These are likely disrupted by M1 recruitment to the plasma membrane by M246, where interactions with the glycoproteins47 and negatively charged phospholipids48,49 stabilize a single-layer M1 matrix in the budding virions29. We propose that the transition from M1 cylinder to M1 layer is the inverse process to matrix layer disruption during virus entry at endosomal pH and delivers a large amount of building blocks to the plasma membrane to aid virus budding.
M1 layer formation drives the incorporation of vRNPs into the budding virions
To further shed light on the vRNP bundling process and the role of M1, we investigated IAV budding profiles by cryo-ET performed on cryo-FIB milled lamellae. Multiple stages of virion budding were observed (fig. 4 a, b): initial membrane bending induced by a small patch of a spherically curved M1 layer (4%, fig. 4 c, i); budding virions with incorporated vRNPs with open M1 layer (42%, fig. 4 d, e, j, k); closed or partially closed M1 layer (28%, fig. 4 f, l), fully assembled with budding neck (26%, fig. 4 g, m) and released viruses (not quantified, fig. 4 h, n). Importantly, cryo-ET allowed us to advance the mechanistic understanding of the budding process. In the virions with a partially closed M1 layer, the M1 layer was not fully attached to the membrane in the virions and had free curved ends interacting with vRNPs (fig. 4 e, k). This suggests that the M1 layer assembly precedes its attachment to the plasma membrane and M1 thereby drives the incorporation of vRNPs into the budding virions. This is in agreement with previous studies showing that budding of virus-like particles containing M1 layer is independent of vRNPs27. It is noteworthy that vRNP bundles reminiscent of a 7+1 parallel configuration5,6 were found neither on HA-remodelled membranes nor on the plasma membrane, but only in fully assembled virions. Hence, we propose that vRNPs only assemble in a parallel 7+1 bundle upon contact with a patch of M1 layer at the plasma membrane, presumably as a result of energetically most favourable packaging geometry within the M1 helically organized scaffold.
In summary, here presented data allowed us to uncover the interplay of several viral proteins during IAV late replication stages and propose a model (fig. 5). We identified HA as a mediator of membrane remodelling which provides a platform to probe vRNP-vRNP interactions in a Rab11a dependent manner. This provides a large membrane surface that could allow for vRNP sorting and clustering as previously determined by FISH and super-resolution microscopy20,21. This work reconciles previous contradicting models showing that vRNPs interact with membranes24 but engage in vRNP-vRNP interactions to form membrane-associated biomolecular condensates23. Further studies will unravel the role of HA-induced membrane remodelling in co-infection and reassortment which is critical in leading to viral progeny with new antigenic footprint. In addition, pinpointing the HA-vRNP interaction and Rab11a recruitment to HA on the molecular level or abrogation of HA-induced membrane remodelling could lead to effective inhibition of vRNP clustering and virus replication. Interestingly, we show that HA compartments form closed double-membrane vesicles occasionally entrapping vRNPs (fig. 2; extended data fig. 2 a, b). This indicates that a fraction of vRNPs is released from the cell upon fusion of the double-membrane vesicle with the plasma membrane. We propose that this mechanism could lead to the formation of HA vesicles containing unspecifically incorporated vRNPs (fig. 5), which could serve as an immunological decoy and function similarly to defective interfering particles50. Finally, we report that M1 assembles into large cylindrical structures in the nucleus that disassemble in the cytoplasm, leading to the formation of the M1 layer, thus unravelling a potential pan-flu antiviral avenue by targeting M1 cylinder disassembly. Overall, this data highlights the multifunctional role of the M1 protein and uncovers the existence of the M1 cylindrical assemblies in infected cells, which has so far been identified only in vitro or inside virions at low pH.