The endodomain of the SARS-CoV S protein binds to the F1 lobe of the FERM domain of ezrin in yeast
We hypothesized that cellular proteins may interact with the carboxy-terminal endodomain of the S envelope protein and either help or restrict processes involving S. To identify such cellular factors, we performed a genomic yeast two-hybrid screen using the SARS-CoV S endodomain as bait (). A random-primed cDNA library from human placenta was screened and 233 positive clones were identified. The most prominent result of the screen was the identification of the interaction between S endodomain and the cellular protein ezrin () with a total of 82 positive clones (35%) corresponding to this protein (Accession number: NM_003379; ). The S - ezrin interaction was classified with high confidence score (Predicted Biological Score
A; ) and was not found in parallel screens performed with other viral baits 
and data not shown). These data indicate that ezrin specifically interacts with SARS-CoV S endodomain in yeast.
The F1 lobe of the ezrin FERM domain binds to the SARS-CoV S endodomain.
To determine the binding region of ezrin to S endodomain, the 82 ezrin cDNA sequences were aligned using ClustalW multiple sequence alignment program. All ezrin cDNA shared a common sequence of 264 base pairs corresponding to amino acids 6 to 93 (). Interestingly, this region corresponds almost completely to the F1 lobe of the N-terminal FERM domain (amino acids 2 to 82 
, ). This suggests that interaction of ezrin with S endodomain requires a complete F1 lobe. This region was subsequently named ezrin F1/S Binding Domain or F1/SBD ().
S endodomain is sufficient to pull-down ezrin from epithelial cell lysates
To confirm the interaction of S endodomain with human ezrin, GST-pull down assays were performed. Briefly, we used GST-fused S endodomain (GST-Sendo) protein bound to glutathione-coupled sepharose beads to pull down endogenous ezrin protein from cell lysates of HeLa human epithelial cells (). Glutathione-sepharose beads alone and GST protein fused to the ezrin FERM domain and bound to beads (GST-FERM) were used as negative and positive controls, respectively ( lanes 2–3). GST-FERM was able to pull down ezrin from the HeLa lysate. This result was expected as the FERM domain has the property to specifically interact with the C-ERMAD of ezrin. Furthermore, GST-Sendo could coprecipitate ezrin in a dose-dependent manner ( lanes 4, 5, 6). This result confirms the interaction of the SARS-CoV S endodomain with the human ezrin protein.
In vitro confirmation of the interaction between SARS-CoV S endodomain and ezrin.
We then decided to analyze the interaction of the S endodomain with ezrin from the SARS-CoV-permissive Vero E6 cell line. Similarly, the GST-Sendo could precipitate the Vero E6 endogenous ezrin protein ( lanes 2–3). Incubation of GST-Sendo with Vero E6 cell lysates in the presence of rabbit serum against the endodomain of S significantly diminished the interaction ( lanes 4–5). Moreover, GST protein alone was not able to pull down ezrin ( lane 6), further confirming that the 39 amino-acids of SARS-CoV S endodomain are responsible for the interaction. These results collectively show that the S endodomain is capable of interacting specifically with ezrin in vitro.
The last 8 carboxy-terminal residues and the membrane proximal cysteine cluster of SARS-CoV S endodomain are involved in ezrin binding in vitro
To map the amino-acids on S that are important for the interaction with ezrin, we decided to construct a series of S endodomain mutants and analyze their capacity to bind ezrin in GST pull down assays (). Analysis of the SARS-CoV S endodomain sequence revealed three regions with distinctive characteristics: a carboxy-terminal basic region (amino-acids 1248–1254), a central acidic stretch (amino-acids 1239–1244) and a series of clusters of cysteine residues (amino-acids 1217–1236). We reasoned that these regions could be involved in the interaction with ezrin. First, we constructed truncation mutants by deleting the carboxy-terminal regions either containing the basic cluster or both the basic cluster and the acidic stretch. These truncation mutants consisted in deletion of eight and nineteen amino-acids, respectively (Sendo Δ8 and Sendo Δ19, ). As expected, glutathione-conjugated beads and GST-coupled beads could not pull down ezrin whereas beads coupled to GST-Sendo wt could precipitate ezrin efficiently from a Vero E6 cell lysate ( lanes 2–6). Both Δ8 and Δ19 truncations were able to significantly diminish interaction with ezrin at similar levels, but not abrogate it ( lanes 7, 8, 9, 10). These data show that the eight carboxy-terminal amino-acid stretch, which is rich in basic residues, is important for ezrin binding and that another motif in the endodomain may contribute to the interaction. Conversely, the acidic stretch does not seem involved in the interaction.
Characterization of interactions determinants of S endodomain binding to ezrin.
We then analyzed the implication of the nine cysteines in the interaction with ezrin. These residues are found in the membrane proximal region of the S endodomain. To simplify the analysis, we decided to pool cysteines into four clusters, namely C1 to C4, of two to three residues (). While mutations of C2, C3, and C4 clusters had little impact on the interaction with ezrin ( compare lanes 9–14 with lanes 5–6), mutation of the cluster C1 was able to almost completely abrogate the interaction ( lanes 7–8). Consistently, mutation of all clusters of cysteines also greatly reduced binding to ezrin ( lanes 15–16). This result indicates that the C1 membrane proximal cysteine cluster is important for binding to ezrin in vitro.
Since endodomain truncations and cysteine to alanine mutations of the SARS-CoV S endodomain were found to diminish interaction with ezrin individually, both types of modifications were next tested simultaneously (Sendo Δ8
C1-4 and Sendo Δ19 C1-4). Additionally, a truncation mutant was tested where the cysteine cluster-containing portion (residues 1217 to 1236) of S endodomain was deleted (Sendo ΔC). Also, the positively-charged residue lysine 1227 and the polar residue threonine 1220 were point-mutated into alanines to investigate their possible involvement in the S-ezrin interaction (Sendo K1227A and Sendo T1220A; ). As expected, the ΔC truncation almost completely abrogated the interaction with ezrin, although a faint band was still observed further indicating that the C-terminal half of S endodomain contains some interaction determinants ( lane 6). The point mutation on the lysine K1227 did not significantly affect binding with ezrin. The point mutation on the threonine T1220 had no effect on binding with ezrin. Confirming our previous results, S endodomain mutants containing both truncations (Δ8 or Δ19), and the cysteine mutations (C1-4) were not able to bind ezrin ( lanes 4–5). Together, these data show that the membrane proximal cysteine cluster C1 and the last eight C-terminal residues are important determinants of interaction of SARS-CoV S endodomain with ezrin in vitro.
Ezrin is present at the site of entry of Spike-pseudotyped lentiviral particle
Considering the major role of S in virus entry and membrane fusion, we first assessed whether S endodomains could be accessible to ezrin at early stages of infection. To mimic SARS-CoV entry, we used Spike-pseudotyped lentiviral particles (SARSpp), which were previously shown to enter cells in an ACE-2, cathepsin-L, and low pH dependent pathway, faithfully recapitulating the entry process of native SARS-CoV virions 
. Vero E6 epithelial cells stably expressing RFP-ezrin fusion protein were infected with GFP-Vpr SARSpp. At 30 minutes post-infection, cells were analyzed using a total internal reflection fluorescence microscope (TIRFM) to visualize events occurring at proximity of the plasma membrane and minimize background from the other layers of the cells. Green dots corresponding to GFP-Vpr SARSpp were readily observed and found associated with RFP-ezrin enriched domains (, white arrowheads, Supporting movie S1). This result indicates that ezrin is present at sites of SARSpp entry, possibly coating endosomes following internalization of particles. Similar data were obtained using a spinning disc confocal microscope on live cells expressing RFP-ezrin (data not shown). This would suggest that after the formation of the fusion pore, S endodomains facing the host cell cytosol have the possibility to interact with ezrin molecules.
Ezrin accumulates at sites of entry of SARS-CoV S pseudotyped particles.
Knock down of ezrin by siRNA increases SARS-CoV S-mediated entry
To investigate the potential role of ezrin in S-dependent entry, we decided to silence the expression of ezrin by specific siRNA. HeLa-F5 cells, stably expressing the SARS-CoV receptor ACE2, were transfected with siRNAs that target ezrin mRNA or non-targeting control siRNAs. The ezrin siRNA transfection decreased expression significantly, with a quantified 80% knock down at the protein level (). Ezrin siRNA was shown to slightly increase transduction levels by Δenvpp and VSVGpp (around 3- and 5-fold increases, respectively) (). Interestingly, the highest fold-change observed was for SARSpp where S-mediated entry was enhanced more than 12-fold. These data indicate that ezrin expression silencing has a slight enhancing effect on entry of viral particles uptake in general, as well as a more specific highly increasing effect on entry of SARS-CoV S pseudotyped particles. This suggests a negative regulatory role for ezrin in SARS-CoV S-mediated entry.
Knock down of ezrin by siRNA increases entry of SARS-CoV S pseudotyped particles.
Mutation of the ezrin-binding domain on S endodomain favors SARSpp entry into Vero E6 cell
To confirm that ezrin binding to S endodomain negatively regulates SARSpp entry, we have generated mutated SARS-CoV S proteins that contain modified endodomains and pseudotyped them. The mutated SARS S pseudotyped particles were designed by taking into account the biochemical pull-down data we have established (). To completely abolish ezrin interaction, we constructed a S mutant with cysteine to alanine mutations of the first cysteine cluster and with a deletion of the last 8 amino acids (SΔ8 C1). Considering that the C1 mutation could have severe consequences on fusion 
, we also generated a second S mutant with only a deletion of the 8 last amino acids of the C-terminus (SΔ8), a modification which should partially alter interaction with ezrin. SΔ8 and SΔ8 C1 S proteins could be incorporated into lentiviral particles (, upper panel) and similar amounts of particles were found in concentrated supernatants, as indicated by detection of the HIV p24 protein (, lower panel). Levels of incorporation of wt S, SΔ8 and SΔ8 C1 were quantified by densitometry and normalized to p24 levels. The ratio of S incorporation was 1
3.9 for wt S, SΔ8 and SΔ8 C1, respectively. Entry of mutated S pseudotyped particles along with wt S SARSpp was analyzed in SARS-CoV-susceptible Vero E6 cells (). Both mutated SARSpp were found to transduce Vero E6 cells more efficiently than wt S SARSpp. Although mutation of C1 is expected to induce a defect in fusion 
, a 2-fold increase in transduction was measured for SΔ8 C1 SARSpp. This increase can be explained by both the efficient incorporation of SΔ8 C1 in pseudotyped particles and the lack of interaction with ezrin. Interestingly, significantly higher levels of transduction were observed for the SΔ8 SARSpp, with an increase of transduction of approximately 11-fold compared to wt S SARSpp. This result is in agreement with previously published data where murine leukemia virus (MLV)-based SΔ8 SARSpp were found to transduce Vero cells more efficiently than wt S pseudotyped particles 
. In our study, better incorporation of SΔ8 could be in part responsible for the increase of transduction. However, the difference in fold-change that was observed for SΔ8 incorporation into pseudoparticles and cellular transduction (2-fold compared to 11-fold increase, respectively) suggests that partial disruption of interaction with ezrin contributes to this enhancement of transduction. Taken together, these data are in agreement with the hypothesis that ezrin binding to S endodomain decreases entry of viral particles.
S C-terminal mutations that decrease ezrin interaction enhance transduction by pseudotyped particles in Vero E6 cells.
Expression of the ezrin FERM domain increases susceptibility of Vero E6 cells to SARS-CoV infection
To further study the role of ezrin during SARS-CoV infection, we produced clonal Vero E6 stable cell lines that either express wild-type ezrin (ezrinwt
) or the N-terminal FERM domain of ezrin (ezrinFERM
) fused to the green fluorescent protein (GFP). The latter form of ezrin is known to have a dominant negative effect on endogenous ezrin 
. The levels of expression and subcellular localization of GFP-ezrinwt
in selected clones were monitored by flow cytometry and fluorescence microscopy, respectively (). Flow cytometry analysis showed that clones GFP-ezrinwt
had means of fluorescence intensities of 2×103
, respectively (). As expected, GFP-ezrinwt
distributed in the cell cytosol with occasional enrichments at the cell cortex, whereas GFP-ezrinFERM
was almost exclusively found at the cell cortex, in membrane ruffles and lamellipodia (). We verified that Vero E6, and clones GFP-ezrinwt
expressed similar levels of ACE-2 receptor at cell surface (data not shown).
Expression of the FERM domain of ezrin increases Vero E6 cell susceptibility to SARS-CoV infection.
The stable cell lines and control Vero E6 cells were first infected with SARS-CoV. Infection levels were then monitored by quantitative real-time PCR (qRT-PCR) on the viral N gene at 3, 6 and 24 hours post infection () and by immunofluorescence assay at 24 hours post-infection (). During the whole time course of infection, levels of N RNAs measured in infected GFP-ezrinwt cells were comparable to those counted for control Vero E6 cells (). Normalized N levels at early time points ranged between 3.2×104 and 3.6×104 and reached approximately 2.6×106 at 24 hours post-infection for these two cell lines (~2 log increase between 6 and 24 hours). Interestingly, although the replication rate at 24 hours was lower in clone GFP-ezrinFERM than control Vero E6 cells (normalized N levels of 1×106), a slight enhancement was observed at early time points (normalized N RNAs ranged between 5.2×104 and 6×104). Similar profiles were observed for qRT-PCR on viral ORF1b (data not shown). This result shows that expression of the dominant negative FERM domain of ezrin slightly enhances early, but not late stages of infection.
In parallel to qRT-PCR analysis, cells were analyzed by immunolabeling and fluorescence microscopy at 24 hours post-infection (). S was labeled to mark infected cells. All cells were stained for either actin or nuclei. For each cell line, the percentage of infected cells was calculated (). Infection rates in Vero E6 and Vero E6 GFP-ezrinwt cells were similar with 2.5% of S-positive cells. Interestingly, a 5.8-fold increase in the percentage of infected cells (15% of cells) was observed for the GFP-ezrinFERM cell line, when compared with control cells. This higher susceptibility to infection is consistent with higher levels of N RNAs detected at early time points post-infection for the clone GFP-ezrinFERM. These data indicate that partial disruption of ezrin function by expression of its dominant negative form increases host cell susceptibility to infection.
Expression of the ezrin FERM domain increases S-mediated entry
To verify that the enhanced permissiveness to SARS-CoV infection observed in cells expressing the ezrin FERM domain is due to a higher efficiency of entry of virions, we investigated levels of transduction of control, GFP-ezrinwt and GFP-ezrinFERM clonal Vero E6 cells by SARSpp (). Interestingly, whereas the transduction level of GFP-ezrinwt was not significantly different than in control Vero E6 cells, GFP-ezrinFERM cells showed a ~15-fold increase in luciferase activity values. This result indicates that clone GFP-ezrinFERM is more permissive to SARS-CoV S-mediated entry and is consistent with the higher rates of infection observed with SARS-CoV in this cell line ().
Expression of the FERM domain of ezrin enhances entry of SARS-CoV S pp.
To exclude a clonal effect, a new batch of polyclonal Vero E6 cells expressing GFP-ezrinFERM was generated. As control, we used Vero E6 cells transduced to express a non-relevant HcRed protein. Expression level of GFP-ezrinFERM was monitored by flow cytometry (). These cells were used to carry out infections using SARSpp (). Consistently with previous data on clonal cell lines, polyclonal cells expressing GFP-ezrinFERM showed a 9-fold increase in luciferase activity, indicating higher susceptibility to SARSpp infection. Of note, SARSpp infection was not enhanced in cells selected to express high amounts of GFP-ezrinFERM (data not shown). Altogether, our data indicate that partial disruption of ezrin activity by expression of low levels of GFP-ezrinFERM enhances cell susceptibility to S-mediated entry and support a restraining role of ezrin at this stage.
Expression of the FERM domain of ezrin by target cells enhances S-mediated cell-cell fusion
Our previous functional experiments have demonstrated a negative regulatory role for ezrin during S-mediated entry. To characterize further this phenomenon, we questioned whether ezrin could be involved in the S-mediated fusion process, during which S endodomains become accessible to the cytosol. To that end, we chose to study the effect of expression of the FERM domain of ezrin on S-dependent fusogenicity in an in vitro cell-cell fusion assay. In this experiment, HeLa cells stably expressing both S and a HcRed fluorescent marker (HeLa HcRed Spike) were co-incubated with GFP, GFP-ezrinwt or GFP-ezrinFERM Vero E6 stable cell lines. As expected, control HeLa HcRed cells were not able to fuse with any of the three Vero E6 cell lines, as no syncytia were found ( panels a to c). Similarly, no syncytia were observed for HeLa HcRed Spike cells incubated with Vero E6 stable cell lines but not activated by trypsin treatment ( panels d, f and h). Conversely, about 4.5% of nuclei were found in syncytia when HeLa HcRed Spike cells were incubated in presence of either GFP or GFP-ezrinwt Vero E6 cells after trypsin activation ( panels e and g and ). Interestingly, 8% of nuclei were found to be in syncytia in the condition where HeLa HcRed Spike cells were co-cultured with GFP-ezrinFERM cells after trypsin activation ( panel i and ). This 2-fold increase of fusion, linked to expression of GFP-ezrinFERM in target cells, was consistently found in four independent experiments. This result shows that target cells expressing the FERM domain of ezrin are more susceptible to S-dependent cell-cell fusion and further indicates that ezrin plays a restrictive role during SARS-CoV S-dependent fusion process.
Effect of wt or FERM ezrin expression on S-mediated cell-cell fusion.