Inhibition of Ebola virus glycoprotein-mediated cytotoxicity by targeting its transmembrane domain and cholesterol

The high pathogenicity of the Ebola virus reflects multiple concurrent processes on infection. Among other important determinants, Ebola fusogenic glycoprotein (GP) has been associated with the detachment of infected cells and eventually leads to vascular leakage and haemorrhagic fever. Here we report that the membrane-anchored GP is sufficient to induce the detachment of adherent cells. The results show that the detachment induced through either full-length GP1,2 or the subunit GP2 depends on cholesterol and the structure of the transmembrane domain. These data reveal a novel molecular mechanism in which GP regulates Ebola virus assembly and suggest that cholesterol-reducing agents could be useful as therapeutics to counteract GP-mediated cell detachment. The GP protein of the Ebola virus is involved in the detachment of infected cells, which eventually leads to vascular leakage and contributes to haemorrhagic fever. Here Hacke et al.show that the membrane-anchored subunit of GP is sufficient to induce cell detachment, and that cholesterol contributes to this process.

T he current Ebola virus outbreak is the largest and most severe in the history of this virus. Ebola virus poses a threat to thousands of people in West Africa, with outbreaks reported in Guinea, Liberia, Sierra Leone and Nigeria 1,2 . Despite the devastating consequences of Ebola virus infection, the treatment options remain limited and experimental. Recently, the use of cholesterol-reducing agents as an adjuvant therapy to reduce the sepsis-like side effects of Ebola virus disease has been suggested 3,4 . The dysregulation of the immune response and a heavy inflammatory reaction, described as a 'cytokine storm', are induced on EBOV infection 5 . Whether the detachment of infected endothelial cells and the concomitant leakage of the endothelial barrier are important factors in the pathogenesis of Ebola virus disease remains controversial [6][7][8][9][10][11] . Among other factors, the cytotoxicity of EBOV has been attributed to the envelope glycoprotein (GP) 6 ; however, the molecular mechanism underlying the pathogenicity of this virus is unclear. GP is an essential virulence factor that plays a key role in viral entry, membrane fusion, epitope shielding and EBOV assembly [12][13][14][15][16][17] . Virulence has been attributed not only to the membraneanchored GP but also to the soluble form of this protein, sGP, which results from RNA editing 18 . sGP increases endothelial permeability by acting as an activator of neutrophils and macrophages, which regulates the host immune response [19][20][21][22] .
Recently, GP shedding was shown to impact endothelial permeabilization 23 . In addition, the interaction between b1-integrin and GP was shown to contribute to the loss of adhesion of GP-expressing cells 7,24 . Furthermore, a role for the GP 2 subunit in counteracting interferon-induced antiviral response via the protein tetherin (BST-2) has been proposed 25 . GP is a type I membrane protein that exhibits high N-and O-glycosylation, and undergoes proteolytical cleavage through furin in the trans-Golgi network, which yields the large amino-terminal GP 1 subunit and the small carboxy-terminal GP 2 subunit 26 . The GP 2 subunit contains a putative fusion peptide and a transmembrane domain (TMD) and is palmitoylated twice on the short cytoplasmic tail 27,28 . Owing to an intramolecular disulfide bond, GP 1 and GP 2 do not dissociate on furin cleavage 29 . The mature GP forms a trimer that comprises of disulfide-bridged GP 1,2 subunits 30,31 . The ectopic expression of GP is sufficient to induce the detachment of adherent cells 32 , and a mucin-like domain within subunit GP 1 has been suggested to strongly impact this process 6 . Cell detachment accounts for the loss of the endothelial barrier of infected blood vessels, which putatively leads to vascular leakage and contributes to the high pathogenicity of the virus 6,9 . Ebola virions are filamentous in shape, and the matrix protein VP40 and GP are sufficient to produce protrusions at the plasma membrane of adherent cells 33,34 . The protrusions resulting from VP40 are virus like, whereas the structures resulting from GP alone appear pleiomorphic 35,36 . The combination of VP40 and GP is sufficient to produce filamentous virus-like particles that exhibit the morphology of Ebola virions 37,38 .
Here we report that GP-induced cell detachment might be attributed to the GP 2 subunit and that cytotoxicity is strongly impacted through cholesterol and a GXXXA motif in the TMD. Furthermore, we show that the GP 2 subunit induces filamentous protrusions at the plasma membrane that also depend on cholesterol and the GXXXA motif, which implies a role for GP in the assembly of the filamentous Ebola virus.

Results
A mucin-like domain within GP has been attributed to the induction of cell detachment 6 . The isolated GP 2 subunit has not been analysed in this context, and notably, the contribution of GP 2 to cell detachment or the production of virus-like filaments was not observed using yellow fluorescent protein (YFP)-GP 2 fusions 25 . These results left a possibility that the mucin-like domain, rather than being involved in mediating detachment, acts as a spacer to maintain an appropriate distance between GP 1 and GP 2 and thereby allow the interaction of GP 2 subunits (Fig. 1a). In this case, the deletion of the mucin-like domain then would reduce the relative distance of GP 1 to the membrane, causing GP 1 to interfere with the oligomerization of GP 2 . Accordingly, YFP-GP 2 might not induce cell detachment, as the bulky YFP moiety may have taken the role of GP 1 in the GP 1,2 -Dmucin construct, potentially disturbing GP 2 oligomerization and masking the contribution of this protein to cell detachment.
FLAG-GP2 induces protrusions at the plasma membrane. We observed the presence of a seam-like arrangement of Gly and Ala residues with a central G/A/S-XXX-G/A/S motif in the TMD of GP 2 . These motifs are well-established hubs for TMD-TMD interactions 39 ; thus, these motifs likely contribute to the oligomerization of GP. To challenge the hypothesis that TMD-TMD interactions might be disrupted due to the steric hindrance resulting from the presence of a bulky protein (unduly close GP 1 in the mucin-like domain deletions or YFP in YFP-GP 2 ), we designed constructs based on the GP from Ebola Zaire (ZEBOV) by using a small FLAG tag directly fused to the GP 2 domain (as depicted in Fig. 1a; Supplementary Table 2). This tag, which contained eight amino acids (with a mass of B2 kDa, including the linker peptide), was not expected to exert strong steric hindrance.
Surprisingly, immunofluorescence and scanning electron microscopy revealed that the transient transfection of HeLa and various other cell lines (Supplementary Fig. 1) with FLAG-GP 2 constructs resulted in the drastic formation of filaments at the plasma membrane (Fig. 1b,c and Fig. 2). The transfection of FLAG-GP 1,2 led to the formation of dot-like structures (Fig. 2a,b, left panel) that likely resemble the previously described GP 1,2 -induced pleiomorphic structures 36 . The FLAG-GP 2induced filaments were, on average, 14.7 ± 0.44 mm in length, and B50 nm in diameter ( Supplementary Fig. 3a). The observed filament length is consistent with the dimensions observed for the virus-like particles produced through VP40 and GP 1,2 (ref. 36). The observed reduction in diameter (80 nm for virions) might reflect the lack of other viral proteins. The filaments exhibited a high extent of peripheral branching, with fine and widespread extensions (Fig. 1b,c). Strikingly, a strong reduction in the volume of the cells was observed concomitant with the production of these filaments and frequently resulted in rod-like cell remnants ( Fig. 1b; Supplementary Movies 1

-3).
Cytotoxicity of GP is encoded in its GP 2 subunit. Next, we analysed the potential induction of HeLa cell detachment through FLAG-GP 2 . Although HEK cells are frequently used to study GP-induced detachment 6,32 , we performed quantitative measurements in HeLa cells, which adhere more strongly and are less prone to detachment than HEK cells. HeLa cells were transfected with either FLAG-GP 2 or FLAG-GP 1,2 constructs for various times, and the ratio of detached to adherent cells was determined through flow cytometry (Fig. 1d). As a control, we employed plasma membrane-resident ( Supplementary Fig. 3f) asialoglycoprotein receptor 1 (ASGR1), which forms trimers similar to Ebola GP. At 36 h post transfection, B60% of the FLAG-GP 2 -transfected cells were detached, whereas ASGR1transfected cells did not exceed 10% detachment, although the protein levels were higher (Fig. 1d). Importantly, FLAG-GP 1,2transfected cells achieved a similar degree of detachment (B70%, ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8688 after 36 h), which indicated that the propensity to induce detachment is attributed to the GP 2 subunit and not the mucin-like domain within GP 1 as previously reported 6 .
A GXXXA motif in the TMD of GP impacts cytotoxicity. Single G/A/S-XXX-G/A/S motifs are able to mediate TMD-TMD interactions 39 . We therefore speculated that the presence of multiple repeats of these motifs, aligned on one side of the TMD of GP ( Supplementary Fig. 5b), would enhance GP oligomerization and potentially influence the extent of filament formation and detachment. To this end, we changed the central GXXXA motif in FLAG-GP 2 to LXXXL (as highlighted in Fig. 1a and Supplementary Fig. 5b). This mutant was analysed through immunofluorescence and scanning electron microscopy. Notably, filament formation was strongly inhibited, with the average lengths reduced to 4.7 ± 0.33 mm (Supplementary Fig. 3a). When the propensity of the mutant to confer detachment was analysed, a significant B50% reduction was observed compared with wild-type FLAG-GP 2 (Fig. 3b), which suggested that the GXXXA motif contributes to the extent of filament formation and GP 2 -mediated cell detachment. Strikingly, a single point mutation in full-length FLAG-GP 1,2 , yielding GP 1,2 -GXXXL, exhibited a drastic reduction of 450% in detachment compared with the wild-type construct (Fig. 3a).
On the basis of the immunofluorescence analysis of cells expressing FLAG-GP 2 and co-stained for actin, we observed microfilaments at the basis of the FLAG-GP 2 -induced filaments ( Fig. 1b and Fig. 2a,b). We therefore incubated the cells with cytochalasin B (CB), an inhibitor of actin filament polymerization, and observed that GP 2 -transfected cells no longer exhibited GPinduced filaments, although FLAG-GP 2 was significantly localized to the plasma membrane ( Supplementary Fig. 3d), which suggested a role for actin in the observed phenotype, that is, potentially acting as a scaffold for GP-induced filaments.
Cholesterol impacts GP-induced cytotoxicity. Notably, the endomembrane system in FLAG-GP 1,2 -and FLAG-GP 2 - transfected cells appears unaffected ( Supplementary Fig. 2a,b). The partitioning of GP into (cholesterol-dependent) detergentresistant membranes 40 suggested that in addition to the GXXXA motif, the oligomerization of GP might be affected through the increased level of cholesterol at the plasma membrane. A cholesterol gradient exists across the secretory pathway 41 (low in the endoplasmic reticulum (ER) and high in the plasma membrane), and would hence allow the formation of filaments at the plasma membrane, but not in the endomembrane system. To analyse the specific interactions of GP with cholesterol, we utilized a well-established photoaffinity-labelling assay [42][43][44][45] . Radioactively labelled (tritiated) photoactivatable cholesterol was fed to HeLa cells expressing either FLAG-GP 1,2 , FLAG-GP 2 or their respective variants. At 24 h post transfection, the cells were irradiated with ultraviolet, thereby crosslinking the photolabile cholesterol analogue to the proteins and lipids in the immediate proximity (r3 Å). As a control, the equally trimeric and plasma membrane-resident protein ASGR1 was employed. Although ASGR1 exhibited weak labelling despite high expression levels ( Supplementary Fig. 3e), both FLAG-GP 1,2 and FLAG-GP 2 were highly and efficiently labelled (Fig. 2e). SDS-resistant oligomers of FLAG-GP 2 corresponding to dimers and trimers were observed. Strikingly, a strong reduction in photocholesterol labelling was observed for both FLAG-GP 1,2 -GXXXL and FLAG-GP 2 -LXXXL (Fig. 2e). The reduction in cholesterol labelling of the GXXXA motif point mutants was correlated with a strong decrease in cell detachment of B50%, respectively ( Fig. 3a,b). Therefore, we next analysed the effect of lowering cholesterol in the plasma membrane of FLAG-GP 2 and FLAG-GP 1,2 -transfected cells with regard to filament production and cell detachment. Methyl-b-cyclodextrin (MBCD) is a wellestablished tool for lowering the cholesterol content in the plasma membranes of mammalian cells 46 . We exposed cells transfected with full-length FLAG-GP 1,2 , FLAG-GP 2 or variants mutated in the central GXXXA motif to 0.5 mM MBCD at 6 h post transfection, which significantly lowered the levels of cholesterol at the time of analysis 24 h post transfection (425%; Supplementary Fig. 4a). Subsequently, we measured the extent of detachment induced through the respective proteins in the presence and absence of MBCD using flow cytometry and immunofluorescence. Strikingly, MBCD significantly reduced (450%) the level of detachment induced through FLAG-GP 1,2 and FLAG-GP 2 constructs (Fig. 3a,b), whereas for FLAG-GP 1,2 -GXXXL and FLAG-GP 2 -LXXXL, detachment was not affected after treatment with MBCD ( Supplementary Fig. 3b). Treatment with MBCD alone did not significantly induce the detachment of cells, changes in morphology or interfered with expression of GP ( Supplementary Figs 3c, 4c and 7). Accordingly, when we incubated the cells with the statin simvastatin before transfection, a significant reduction in the cholesterol levels was observed (430%; Supplementary Fig. 4b), and GP-induced detachment was significantly reduced B40% (Fig. 3d).
Interestingly, the addition of cholesterol to cells using cholesterol-loaded MBCD strongly increased GP-induced cell detachment (Fig. 3e). Furthermore, the inhibitory effect of MBCD on GP-induced detachment was restored when the cells were incubated with MBCD-complexed cholesterol (Fig. 3c).
Next we expanded the analysis of filoviral GP 2 subunits to Marburg virus (MARV) and Lloviu cuevavirus (LLOV), the only two remaining genera of Filoviridae 47,48 . Comparisons of the TMDs of the respective GP 2 subunits of these viruses ( Supplementary Fig. 5a) revealed that EBOV and LLOV share a high extent of sequence identity, including the G/A/S-XXX-G/A/S motif, but that MARV lacks a comparable sequence ( Supplementary Fig. 5b). Analogous to EBOV GP 2 , we examined FLAG-LLOV-and FLAG-MARV-GP 2 for the induction of cell detachment and production of plasma membrane protrusions. MARV FLAG-GP 2 induced neither virus-like-filament formation at the plasma membrane, nor significant cell detachment, while the GXXXA motif containing LLOV FLAG-GP 2 triggered cell protrusions and cell detachment comparable to EBOV FLAG-GP 2 , and this effect was susceptible to treatment with MBCD ( Supplementary Fig. 5c,d). These data highlight the correlation between the presence of G/A/S-XXX-G/A/S motifs in the TMD of filoviral GP, cell detachment and cholesterol.
Next we employed the cholesterol-specific probe filipin to examine the enrichment of cholesterol in GP 2 -induced cell protrusions 49 . Although GP 2 -expressing cells are efficiently labelled at the plasma membrane and its filamentous protrusions, staining of the plasma membrane is strongly reduced in MBCD-treated cells (concomitant with the absence of protrusions), and FLAG-GP 2 -LXXXL-expressing cells exhibit low filipin co-staining (Fig. 3g). We speculate that the co-aggregation of cholesterol into Ebola virions and the concomitant depletion of plasma membrane cholesterol might also contribute to pathogenicity, as the sequestration of cholesterol would also likely affect the function of cholesterolinteracting receptors and signalling proteins in the plasma membrane, such as the T-cell receptor or MHC class II molecules 50,51 .
Altogether, these data suggest that the GXXXA motif mediates interactions among GP trimers, generating higher, cholesterolcompetent, GP oligomers. Thus, the mutation of this motif correlates with a loss of MBCD sensitivity and a marked reduction in detachment. The titration of MBCD revealed a sigmoidal dose response for both GP 1,2 and GP 2 constructs, with half-maximal inhibition of detachment at similar concentrations (FLAG-GP 1,2 : 197.5 mM ± 25.4 s.e., and FLAG-GP 2 : 295.1 mM ± 98.5 s.e.; Fig. 3f). Thus, we concluded that the cytotoxicity conferred through GP depends on cholesterol and is additionally influenced by the presence of a GXXXA motif in the TMD. Although these data strongly implicated cholesterol as the major lipid involved in cytotoxicity, we cannot exclude that other lipids were indirectly affected in these experiments. A reduction in the cholesterol levels at the plasma membrane might reflect differences in the composition of specific lipid microdomains targeted through GP, for example, polyunsaturated phospholipids resulted in deformations of the plasma membrane 52 , and phosphatidylserine levels in the cytoplasmic leaflet might also be affected.
Nonetheless, these data suggest that cholesterol-reducing treatments are potential candidates for therapeutics to counteract the pathogenicity induced through Ebola GP; these treatments could also be employed after infection with Ebola virus as an addition to the established post exposure regimens [53][54][55][56][57][58][59][60] . Cyclodextrins have been patented as antiviral agents for the treatment of infections resulting from various enveloped viruses 61 . Treatment with MBCD is even more promising because the IC 50 reported here is B20-fold lower than those determined for the herpes, vaccinia, Epstein-Barr and hepatitis C viruses 61 .
FLAG-GP 2 -induced filaments form by retraction. We next monitored the production of the FLAG-GP 2 -induced filaments through live-cell imaging using a TIRF-DIC (total internal reflection fluorescence-differential interference contrast) microscope set-up. As depicted in Fig. 4c and in Supplementary Movie 1, after 370 min and within the course of B70 min, the cells contract with filamentous structures left behind while the cell body retracts, which suggests that the filaments are formed through retraction rather than outgrowth. Moreover, bubbles were observed underneath the cell, and the cells rounded up before eventual detachment, which suggests a correlation between the production of plasma membrane protrusions and cell detachment.
A subset of host proteins exhibits a GP-like TMD arrangement. Subsequently, we analysed the amino-acid sequences of host single-spanning plasma membrane-resident proteins to characterize the seam-like arrangement of Gly and Ala residues present in the TMD of GP. Strikingly, only 26 proteins exhibited more than five Gly, Ala or Ser residues within a seam-like arrangement in the TMD. A total of 14 of these candidates were implicated in cell adhesion or cytoskeleton organization (Fig. 4d,  Supplementary Table 1). Among these TMDs are known interactors of GP, such as b1-integrin 24 . Nine proteins involved in the immune response (for example, the a-chain of HLA class II) were also identified. These findings suggest that GP 1,2 might bind to and interfere with these candidates via the GXXXA motif.

Discussion
As depicted in Fig. 4a, we propose the following model for the formation of GP-induced filaments-at the onset of assembly, GP-TMDs trimerize in a cholesterol-independent manner. During Ebola virus assembly, trimeric GP subunits interact within cholesterol-enriched membrane domains via GXXXAmotifs in the TMDs, generating a lattice-like structure that predominantly contains GP and excludes host proteins (Fig. 4a,  centre). When a critical lattice size is formed, this membrane domain collapses around an actin filament that serves as a scaffold for plasma membrane filament formation (Fig. 4b).
Matrix protein VP40 participates in this process through oligomerization at the cytoplasmic face of the plasma membrane 34,62 (Supplementary Fig. 6), further stabilizing the filamentous structures in the viral context and putatively additionally exerting an outward-directed force on the plasma membrane.
In summary, based on the results obtained in the present study, we propose a novel mechanism for Ebola virus assembly at the plasma membrane, based on cholesterol-dependent interactions among GPs, leading to the collapse of large GP aggregates around actin filaments. Rather than through outgrowth, the formation of these filaments reflects the growth of GP aggregates within the plasma membrane and the subsequent wrapping of these aggregates around actin scaffolds (Fig. 4a,b). This mechanism might be closely associated with cell detachment, and suggests two novel drug targets to counter GP-mediated cell detachmentcholesterol, which could be targeted using cholesterol-reducing agents, such as MBCD and statins, and the central GXXXA-motif situated within a seam-like assembly of Gly and Ala residues in the TMD of GP.
Plasmids and constructs. FLAG-tagged cDNA for Ebola GP 2 and GP were purchased as synthetic genes from GeneArt (Life Technologies) based on the reverse-translated amino-acid sequences of Uniprot Q05320 (VGP_EBOZM) using the most likely codon usage for H. sapiens (Supplementary Table 2). Using EcoRI and NotI sites, the synthetic genes were cloned into pEGFP-N1 vectors, from which EGFP had been removed via EcoRI/NotI digestion. Site-directed mutagenesis was performed using the QuikChange Site-Directed Mutagenesis Kit (Stratagene). FLAG-tagged ASGR1-cDNA was obtained from Origene (clone # RC205686).
For transfection, the culture medium was removed from a monolayer of 90% confluent cells. The cells were washed with PBS (1 Â ; PBS, Sigma-Aldrich) and subsequently trypsinized (0.05% Trypsin-EDTA; Gibco by Life Technologies). A total of 1.5 Â 10 6 cells were resuspended in a final volume of 400 ml of electroporation buffer (RPMI-1640-medium supplemented with 10% fetal bovine serum) and placed in a 0.4 cm electrode gap cuvette (PEQLAB). 15 mg FLAG-GP 2 was mixed with the cell suspension and incubated at 4°C for 10 min. Cells were electroporated using an electric pulse generator (BioRad Gene Pulser Xcell) under the following conditions-square wave electric pulse, 250 V and 40 ms pulse length. Subsequently, the cells were resuspended in the respective growth media and plated for further assays. After 4 h, the culture medium was replaced with fresh complete growth medium. The images were acquired after 24 h through confocal microscopy using a Zeiss LSM 510 unit mounted on an Axiovert 200 inverted microscope with a 1.4 Plan-APOCHROMAT Â 63 oil objective.
Confocal microscopy. The cells were plated on 35 mm glass-bottomed dishes (MatTek Corp.) and transfected using FuGENE HD. After 24 h, the cells were washed three times with PBS (pH 7.4) and fixed with 3.6% paraformaldehyde (Sigma-Aldrich) for 20 min on ice. For endomembrane staining, the cells were permeabilized using 0.5% (w/v) Triton-X-100 for 5 min at RT. After washing three times, for blocking, cells were incubated with 1% bovine serum albumin (BSA; Carl Roth) in PBS for 1 h. Immunofluorescence was performed with a dilution of primary ARTICLE antibodies at 1:200 and secondary antibodies at 1:1,000 in 0.5% BSA in PBS. The primary antibody dilutions were added to the cells for 1 h, followed by washing three times with PBS for 5 min. Secondary antibody dilutions were subsequently added for 30 min. Alexa647-conjugated WGA was used at final concentration of 10 mg ml À 1 for 20 min, and 10 nmol of Atto565-conjugated phalloidin (Sigma) was added to 1 ml of methanol and incubated with the cells at a dilution of 1:100 in PBS for 30 min. The samples were examined through confocal microscopy using a Zeiss LSM 510 unit mounted on an Axiovert 200 inverted microscope with a 1.3 NA Plan-NEOFLUAR Â 40 oil objective, equipped with an argon laser. Image processing was performed using Zen 2011 and ImageJ software.
Quantification of cell detachment using flow cytometry. The cells were plated on 6-well dishes (Corning Inc.) and transfected with the indicated contructs using FuGENE HD. At the indicated timepoints, the dishes were placed on ice. The culture medium containing detached cells was transferred to 1.5 ml tubes and centrifuged for 6 min (10,000 r.p.m.). The supernatant was discarded, and the pellet was resuspended in 200 ml serum-free DMEM. The adherent cells were washed once with PBS, and subsequently incubated with 300 ml of EDTA-containing cell dissociation buffer (Gibco) at 37°C for 5 min. Subsequently, 300 ml of serum-free DMEM was added and the resuspended cells were transfered to 1.5 ml tubes. The ratio of detached versus total cells was quantified using a FACSCalibur flow cytometer (Beckton Dickinson).
Fluorescence stain of membrane cholesterol and quantification. A filipin stock solution was prepared in dimethylsulphoxide at a concentration of 50 mg ml À 1 . For membrane staining, unpermeabilized paraformaldehyde-fixed cells were incubated with a 1:100 dilution of filipin solution in PBS containing 1% BSA for 30 min before incubation with the antibodies (primary antibody: 1:200; secondary antibody: 1:1,000) and then subjected to confocal microscopy. To quantify the membrane cholesterol population, six-well plates with adherent cells were treated accordingly, and filipin fluorescence was measured using a MolecularDevices SpectraMax M5 microplate reader at an excitation wavelength of 355 nm and emission wavelength of 460 nm (ref. 49).

Treatment of cells with MBCD and CB.
A stock solution of CB was prepared at 5 mg ml À 1 in dimethylsulfoxide. To deplete cholesterol and inhibit actin polymerization, MBCD and CB were added to the medium at the concentrations indicated 6 h post transfection and incubated overnight.
Bioinformatics. All proteins annotated as membrane proteins were downloaded from Uniprot 63 . These proteins were homology reduced to 30% using CD-HIT 64 . To map the TMDs of these proteins, we used DG-pred 65 with the constraint that a TMD shows a D-G score o0. To identify putative TMD-TMD interaction interfaces that were enriched in the Gly, Ala and Ser residues, we generated an algorithm that analyses the TMDs from the z axis. On the basis of the assumption of the regular spacing of the residues in these alpha-helical segments, the total number of small side chain-containing residues (Gly, Ala and Ser) and the relative distribution of these residues around the TMD were obtained. Three data sets corresponding to the different subcellular localizations of the membrane proteins (ER, Golgi and plasma membrane) were generated. To evaluate the data, the plasma membrane protein data set was analysed for proteins containing Z5 residues of Gly, Ala or Ser positioned at an angle of 180°-as observed in the TMD of Ebola GP (Fig. 4d). Falsely annotated entries and those specific to early developmental stages and proteins exclusively present in certain tissues were omitted for clarity.
Depletion and replenishment of cholesterol in the cells. MBCD-complexed cholesterol from Sigma was added to the cells at 6 h after transfection for 30 min at a concentration of 100 mM in normal growth medium. The cells were washed with PBS and incubated overnight in growth medium. For the rescue experiments, the cells were incubated in 5 mM MBCD for 60 min before transfection and washed three times with PBS. At 6 h after transfection, water-soluble cholesterol was added as indicated.
Treatment of cells with a statin. The cells were cultured in simvastatin at 2 mM for 4 days before experiments in normal growth medium. The cells then were subsequently transfected and treated as described above.
Scanning electron microscopy. The cells were plated onto 35 mm glass-bottomed dishes (MatTek), and transfected for 24 h with different GP variants. The cells were fixed with 2% glutaraldehyde in 0.1 M sodium phosphate buffer (Sørensen), pH 7.4 for 2 h, rinsed with buffer and dehydrated with an ascending series of ethanol. The samples were critical point dried using carbon dioxide and the slides were coated with gold (5 nm). The samples were analysed using a ZEISS ULTRA 55-field emission scanning electron microscope, using the Everhart-Thornley secondary electron detector.
Live-cell imaging. HeLa cells were plated onto 35-mm MatTek glass-bottomed dishes and transfected for 24 h with 3.3 mg FLAG-GP 2 and 10 ml FuGENE HD transfection reagent in complete DMEM medium. After 5 h, the cells were washed with PBS and a HEPES-based imaging medium (Invitrogen). Microscopy was performed on a Nikon TIRF-2 system with Eclipse Ti inverted microscope, using a Nikon Apo-TIRF 1.5 NA Â 60 oil objective in differential interference contrast mode with a Hamamatsu 1394 ORCA-ERA camera and an incubation chamber adjusted to 37°C and 5% CO 2 . Imaging was commenced at 5 h post transfection, with an acquisition rate of one image per 10 min, for a total duration of 16 h using NIS-Elements Microscope Imaging Software.
In vivo photoaffinity labelling with cholesterol. FLAG-tagged GP 2 and GP 1,2 constructs and variants thereof and ASGR1, as a negative control, were probed for protein-cholesterol interaction in vivo. For this purpose, photoactivatable cholesterol was employed for photolabelling experiments as previously described 66 . In brief, HeLa cells were grown in six-well dishes and transfected using FuGENE HD (1 mg DNA þ 3 ml transfection reagent in 100 ml Opti-MEM (Invitrogen) at B60% confluency. After 16 h, the cells were labelled with 75 mCi of ( 3 H)-photocholesterol in 3 ml of DMEM supplemented with 10% delipidated FCS for 8 h. The cells were ultraviolet irradiated for 10 min on ice and lysed for 1 h in 100 ml lysis buffer (50 mM HEPES-NaOH, pH 7.4, 100 mM NaCl, 5 mM EDTA, 1% Triton-X-100 (v/ v), 0.5% deoxycholate (w/v), and protease inhibitor cocktail (cOmplete mini, Roche)). Post-nuclear supernatants were subjected to immunoprecipitation using M2-(anti-FLAG)-affinity gel (Sigma-Aldrich) according to the manufacturer's instructions. The proteins were subsequently eluted in 4 Â SDS-polyacrylamide gel electrophoresis sample buffer containing 16% SDS and subjected to SDSpolyacrylamide gel electrophoresis (10-20% Tris/Tricine gradient gels, Invitrogen), western blotting and the quantitative immunodetection of FLAG-tagged proteins using a LI-COR infrared imager. Radioactively labelled proteins were detected by digital autoradiography (Beta Imager 2000, Biospace Lab), using acquisition times of 16 h.