S-palmitoylation and sterol interactions mediate antiviral specificity of IFITMs

Abstract

Interferon-induced transmembrane proteins (IFITM1, 2 and 3) are important antiviral proteins that are active against many viruses, including influenza A virus (IAV), dengue virus (DENV), Ebola virus (EBOV), Zika virus (ZIKV) and severe acute respiratory syndrome coronavirus (SARS-CoV). IFITM proteins exhibit specificity in activity, but their distinct mechanisms of action and regulation are unclear. Since S-palmitoylation and cholesterol homeostasis are crucial for viral infections, we investigated IFITM interactions with cholesterol by photoaffinity crosslinking in mammalian cells along with molecular dynamic simulations and nuclear magnetic resonance analysis in vitro. These studies suggest that cholesterol can directly interact with S-palmitoylated IFITMs in cells and alter the conformation of IFITMs in membrane bilayers. Notably, we discovered that the S-palmitoylation levels regulate differential IFITM protein interactions with cholesterol in mammalian cells and specificity of antiviral activity towards IAV, SARS-CoV-2 and EBOV. Our studies suggest that modulation of IFITM S-palmitoylation levels and cholesterol interaction influence host susceptibility to different viruses.

Graphical Abstract

INTRODUCTION

Interferons (IFNs) provide an important host response to infection and induce the expression of many genes that protect against diverse pathogens¹. Amongst the hundreds of IFN-stimulated genes, interferon-induced transmembrane proteins (IFITMs) have emerged as crucial IFN effectors that prevent viral infection of host cells². In particular, IFITM1, 2 and 3 have been shown to restrict infection by many enveloped viruses including IAV, DENV, EBOV, ZIKV, SARS-CoV, hepatitis C virus (HCV), human immunodeficiency virus (HIV) and others in mammalian cell lines^3–5 and mouse models^6,7. Notably, single nucleotide polymorphisms (SNPs) in humans, such as rs12252-C, which encodes a truncated form of IFITM3, have been associated with severe H1N1 IAV infection⁷. Also, IFITM3 SNP rs34481144, which results in lower mRNA expression, has been correlated with reduced CD8+ T cells in patient airways and increased IAV infection severity⁸. Moreover, IFITM3 SNP rs12252-C has been identified as a risk factor for severity of SARS-CoV-2 infection amongst COVID-19 hospitalized patients in two independent studies^9,10. IFITMs are also expressed in embryonic stem cells independent of IFN stimulation and are important for their resistance to viral infections¹¹. IFITMs reduce susceptibility of placental trophoblasts to viral infections and inhibit trophoblast cell fusion, an essential process for fetal development mediated by syncytin, a retroviral envelope-like protein^12,13. These significant studies in cellular and animal models as well as preliminary clinical observations highlight an important function for IFITM3 in host immunity against viral infections. Beyond antiviral immunity, IFITM3 has been shown to be critical in phosphoinositide 3-kinase (PI3K) signal amplification for rapid expansion of B cells with high affinity to antigen¹⁴. IFITM3 is also implicated in neuroinflammatory conditions as it regulates γ-secretase activity and amyloid-β production in patients with late-onset Alzheimer’s disease¹⁵. These studies highlight broader roles of IFITMs in host physiology and disease, which requires a more detailed understanding of their mechanisms of action and regulation.

IFITM1, 2 and 3 exhibit unique cellular properties and are highly regulated by posttranslational modifications^2,16. All three IFITM proteins show antiviral activity, but IFITM3 is the most active against many viruses^2,3,4. In mammalian cells, IFITM2 and IFITM3 are distributed through early and late endosomes and lysosomes whereas IFITM1, which has a truncated N terminus localizes to the plasma membrane^2,4. The cellular localization of IFITMs is an important determinant of their antiviral mechanism specificity, since enveloped viruses enter cells at different sites (for example, plasma membrane versus endocytic pathway) to deliver their genetic contents for replication in host cells. IFITMs are small type IV single-pass transmembrane proteins with amphipathic helices that interact with the inner membrane leaflet^17,18 and are post-translationally modified by S-palmitoylation, phosphorylation, and ubiquitination. IFITM3 is S-palmitoylated at membrane-juxtaposed Cys residues (Cys71, 72 and 105)¹⁹, but Cys72 is the major site of fatty acid-modification^20,21 and the most important for antiviral activity^20,22. Recent molecular dynamic simulation and NMR studies show that fatty-acylation at Cys72 stabilizes IFITM3 amphipathic helix-membrane interaction²³. Notably, Cys72 is highly conserved across IFITM orthologs in mice, bats, and humans^20,23,24. Live-cell imaging studies also show that Cys72 is essential for the trafficking of IFITM3-positive vesicles with incoming virus particles during infection²². In addition to fatty-acylation, Tyr20 phosphorylation regulates IFITM3 plasma membrane localization and endocytosis, whereas Lys ubiquitination at residues 24, 83, 88, and 104, and especially at Lys24, is important for IFITM3 trafficking and turnover in cells²⁵.

The antiviral activity of IFITMs is also regulated by other protein and lipid interactions. For example, proteomic analysis of the S-palmitoylated IFITM3 interactome has identified many membrane-associated protein interaction partners^26–28, including the p97/VCP ATPase that contributes to IFITM3 lysosomal turnover and antiviral activity^26,29,30. Further, yeast two-hybrid screening revealed IFITM3 interaction with vesicle-membrane-protein-associated protein A (VAPA), which has been implicated in cholesterol accumulation in endolysosomal compartments and IFITM3 antiviral activity^30,31. However, other studies indicate that high cholesterol accumulation in these compartments by U18666A pretreatment or down-regulation of Niemann-Pick C1 (NPC1) does not inhibit IAV hemagglutinin-mediated viral membrane fusion³². Nonetheless, membrane cholesterol composition plays a critical role in viral infection by altering membrane mechanical properties and modulating membrane curvature during hemifusion or lipid mixing³³. Furthermore, specific interactions of viral fusion proteins with cholesterol and other lipids are crucial for successful viral entry into cells^33,34. Recent live-cell imaging studies show that IFITM3 colocalize with virus particles and traffics to lysosomes but does not inhibit membrane lipid mixing or hemifusion^22,35. Despite its lack of effect on viral membrane hemifusion with endosomal membranes^22,32, IFITM3 could potentially block fusion pore formation and release of viral genetic material in the cytosol. It has been suggested that the IFITM3 amphiphatic helices could modulate membrane curvature and stiffness to block fusion pore formation, a process that can be promoted by the presence of cholesterol in model membranes^17,36. However, whether cholesterol directly binds IFITMs and affects their antiviral activity is unclear.

To investigate IFITM-cholesterol interactions, we employed chemical biology tools together with in silico and structural approaches. To characterize cholesterol interaction with IFITMs in mammalian cells, we synthesized a cholesterol photoaffinity reporter (x-alk-chol) and performed proteomic as well as targeted protein photocrosslinking studies. These studies demonstrated that x-alk-chol can photocrosslink IFITM3 expressed in IFN-α-stimulated cells. Subsequent mutagenesis studies showed that S-palmitoylation and a cholesterol binding motif in the transmembrane domain of IFITM3 can influence x-alk-chol photoaffinity labeling and antiviral activity. Our molecular dynamics simulation suggests that cholesterol alters the conformation of IFITM3 in membrane bilayers, which was supported by NMR studies of truncated IFITM3 protein constructs reconstituted in membrane bicelles. Surprisingly, IFITM2 was photocrosslinked by x-alk-chol much less efficiently compared to IFITM3 even though both contain the identified cholesterol binding motif and are more than 80 percent identical at the protein level. Our subsequent analysis showed that the IFITM S-palmitoylation levels are consistent with their x-alk-chol photocrosslinking and specificity of antiviral activity towards IAV, SARS-CoV-2 and EBOV. Collectively, our results suggest that cholesterol can directly interact with IFITMs in mammalian cells, modulate S-palmitoylated IFITM conformations in membrane bilayers, and are consistent with specificity in antiviral activity.

RESULTS AND DISCUSSION

Photoaffinity labeling of proteins in HeLa cells with a sterol reporter.

To investigate cholesterol binding of IFITMs in mammalian cells, we synthesized a bifunctional cholesterol analog (x-alk-chol) that has a diazirine in position 6 for crosslinking to interacting proteins on UV light exposure (Supplementary Figure 1). This cholesterol photoaffinity probe (x-alk-chol) also has a terminal alkyne handle on the alkyl side chain for Cu^I-catalyzed azide-alkyne cycloaddition (CuAAC) to azide-fluorophore for fluorescence detection or azide-biotin for affinity enrichment of interacting proteins (Supplementary Figure 1). Such photoaffinity cholesterol probes have been used to study sterol-protein interactions in cells previously^37,38. Maestro modeling of x-alk-chol showed similar molecular topology as cholesterol (Supplementary Figure 1). To probe cholesterol-binding proteins in cells, HeLa cells were treated with x-alk-chol and irradiated with UV light for x-alk-chol crosslinking with interacting proteins. The cell lysate was then subjected to CuAAC with azide-rhodamine for in-gel fluorescence profiling of x-alk-chol crosslinked proteins (Figure 1A). Many proteins in HeLa cells were crosslinked with x-alk-chol in a UV-dependent and dose-dependent manner (Figure 1B, C). Cells delivered excess cholesterol following x-alk-chol treatment showed lower labeling suggesting potential cholesterol competition with x-alk-chol (Figure 1D).

Figure 1. In-gel profiling of x-alk-chol labeled proteins in HeLa cells.

A) Experimental scheme for labeling of proteins with x-alk-chol in live cells (created with Biorender.com). B) In-gel fluorescence profiling of x-alk-chol crosslinked HeLa cell proteins shows UV dependent x-alk-chol labeling of many proteins in HeLa cell proteome. Cells were incubated with x-alk-chol for 30 min and irradiated with UV light (365 nm) for 5 min following which cell lysates were reacted with azide-rhodamine for fluorescence gel scanning. C) In-gel fluorescence profiling of x-alk-chol concentration dependent labeling of HeLa cell proteins. Cells were incubated with different concentrations of x-alk-chol for 30 min and irradiated with UV light (365 nm) for 5 min following which cell lysates were reacted with azide-rhodamine for fluorescence gel scanning. In-gel fluorescence profiling shows x-alk-chol concentration dependent labeling of proteins. α-Tubulin western blotting shows comparable protein loading. D) In-gel fluorescence profiling of cholesterol competition with x-alk-chol labeling of proteins. Cells were incubated with x-alk-chol for 30 min following which cells were treated with 10 μM cholesterol in mβCD and then irradiated with UV light (365 nm) for 5 min. Then cell lysates were reacted with azide-rhodamine for fluorescence gel scanning. In-gel fluorescence profiling shows cholesterol competition with x-alk-chol labeling of proteins. α-Actin western blotting shows comparable protein loading.

Photoaffinity labeling identifies IFITM3 as a cholesterol binding immune associated protein.

To enrich and identify x-alk-chol labeled proteins in HeLa cells, we subjected the cell lysate to CuAAC with azide-biotin, affinity purified with streptavidin beads and subjected to on-bead protease digestion for mass spectrometric analysis (Figure 2A). A summary of the proteomics data revealed enrichment of 330 proteins in UV light irradiated samples including known cholesterol binding proteins like caveolin-1 (CAV1) and NPC1, among others (Figure 2A)^39,40. Bioinformatic analysis of the enriched proteins suggest x-alk-chol labeling of endoplasmic reticulum (ER) and plasma membrane resident membrane proteins primarily (Figure 2B). Many of the high confidence x-alk-chol labeled proteins were previously identified in a proteome-labeling profile of trans-sterol probe (Supplementary Figure 2)³⁷. Gene ontology analysis of the enriched proteins enriched identified many membrane-associated proteins involved in lipid biosynthesis, and metabolism (Supplementary Figure 3), To characterize cholesterol binding to IFITMs and other immune-associated proteins, we profiled x-alk-chol crosslinked proteins in IFN-α stimulated HeLa cells (Figure 2C). In addition to the proteins identified in non-stimulated cells, we recovered IFITM3 and many other immune-associated proteins from the proteomic data (Figure 2C). Gene ontology analysis of the proteins enriched in IFN-α stimulated cells identified many other membrane-associated proteins involved in immunity (Figure 2D, Supplementary Figure 3), which may warrant further investigation as many of these hits are important host factors involved in virus infection including SARS-CoV-2^41,42. These results show that x-alk-chol can photocrosslink many known cholesterol-binding proteins as well as capture many immune-associated proteins including IFITM3.

Figure 2. Proteomic analysis of x-alk-chol labeled proteins in HeLa cells.

A) Volcano plot showing x-alk-chol binding proteins in HeLa cells. Cells were treated with x-alk-chol (10 μM) for 30 min and UV (365 nm) irradiated for 5 min. The cell lysates were further reacted with azide-biotin for enrichment of x-alk-chol labeled proteins with streptavidin beads and identification by mass spectrometry. Volcano plot shows enrichment of many reported cholesterol binding proteins (in blue) for UV treated sample. The x-axis is the difference of means between the UV treated and untreated samples and the y-axis is the log of the probability of that difference determined by the t-test. The minimum values for a valid protein are p<0.05 and a difference of means of 2 (FDR=0.0001, S0=1, n=3 replicates). B) Gene Ontology (GO) analysis shows subcellular localization of proteins enriched in UV treated samples. C) Volcano plot showing x-alk-chol crosslinked proteins in interferon (IFN-α) stimulated HeLa cells. Volcano plot shows enrichment of many reported cholesterol binding proteins (in blue) and new cholesterol binding immunity associated proteins (in red) for UV treated sample. The x-axis is the difference of means between the UV treated and untreated samples and the y-axis is the log of the probability of that difference determined by the t-test. The minimum values for a valid protein are p<0.05 and a difference of means of 2 (FDR=0.0001, S0=1, n=3 replicates). D) Gene Ontology (GO) analysis shows subcellular localization of proteins enriched in Interferon and UV treated samples.

S-Palmitoylated IFITM3 interacts with membrane cholesterol.

S-palmitoylated membrane proteins have been suggested to partition into cholesterol-rich liquid-ordered membrane microdomains^43,44. IFITM3 is S-palmitoylated at Cys71, 72 and 105 (Figure 3A). Western blot analysis of IFN-α stimulated and x-alk-chol treated HeLa cells showed a mobility shift for IFITM3 in UV-irradiated samples (Figure 3B) resembling that seen with CAV1, a known cholesterol binding protein (Figure 3B). To confirm x-alk-chol photocrosslinking to IFITM3, the cell lysate was subjected to CuAAC with azide-biotin and affinity purified with streptavidin beads. Western blot analysis showed significant enrichment of IFITM3 and CAV1 in the UV-irradiated sample (Figure 3B). Further treatment of cells with x-alk-chol in the presence of cholesterol decreases x-alk-chol labeling of IFITM3, suggesting cholesterol competition with reporter (Supplementary Figure 4). To validate IFITM3-cholesterol interaction, we performed x-alk-chol crosslinking studies with overexpressed HA tagged IFITM3 WT in HEK293T cells. In-gel fluorescence detection of anti-HA immunoprecipitated sample shows UV-dependent x-alk-chol crosslinking of HA-IFITM3 (Figure 3C). To identify the role of S-palmitoylation in IFITM3 cholesterol binding, we tested x-alk-chol labeling of IFITM3 PalmΔ construct, a Cys71, 72 and 105 to Ala triple mutant (Figure 3C, D, Supplementary Figure 5A, B). IFITM3 PalmΔ shows significantly less x-alk-chol labeling. Bioinformatic analysis of IFITM3 orthologues from great apes and rodentia shows that IFITM3 has a putative cholesterol binding domain. IFITM3 in great apes have a cholesterol consensus domain CARC ¹⁰⁴KCLNIWALIL¹¹³ (in pink) which lies N-terminus to the transmembrane domain (Supplementary Figure 6). Many membrane proteins have CARC motifs or its mirror code CRAC motif to mediate interactions with cholesterol^45,46. To investigate role of this motif in IFITM3-cholesterol interaction, we made a CARCΔ construct by replacing Lys104 with Ala and Trp109 with Ile. We observe that x-alk-chol photocrosslinking efficiency of CARCΔ is significantly decreased (Figure 3C, D, Supplementary Figure 5A, B), suggesting role of the motif in IFITM3-cholesterol interaction. Mouse IFITM3, which has the three Cys residues but not the CARC motif conserved, shows significantly lower x-alk-chol labeling in HEK293T cells (Supplementary Figure 7A, B). IFITM3 Lys104 to Ala and Trp109 to Ile single mutants also show reduction in x-alk-chol photocrosslinking (Supplementary Figure 8A, B). IFITM3 single Cys71 and Cys72 to Ala mutants show similar x-alk-chol photocrosslinking as IFITM3 WT but Cys105 to Ala mutant shows some increase in x-alk-chol photocrosslinking. This may be due to less steric interactions at the CARC domain on mutation of Cys105 to Ala (Supplementary Figure 9A, B). S-palmitoylation levels of IFITM3 CARCΔ were analyzed by metabolic labeling with alk-16 labeling¹⁹. IFITM3 CARCΔ shows similar S-palmitoylation levels as IFITM3 WT (Supplementary Figure 10). We also analyzed the subcellular localization of IFITM3 CARCΔ by co-expressing myc-tagged IFITM3 WT with HA-tagged IFITM3 CARCΔ. IFITM3 CARCΔ shows similar endolysosomal localization as IFITM3 WT (Supplementary Figure 11A, B). These results suggest that CARC domain of IFITM3 is important for its interaction with cholesterol but does not impact S-palmitoylation or subcellular localization.

Figure 3. IFITM3 S-Palmitoylation and cholesterol binding regulates antiviral activity.

A) S-palmitoylated IFITM3 in a membrane bilayer (created with Biorender.com). Key domains and post translational modifications are highlighted. Each secondary structure is colored differently, amphipathic helix 1 (AH1 from residue 62 to 67) in purple, amphipathic helix 2 (AH2 from residue 76 to 85) in green and transmembrane domain (TM from residue 109 to 131) in orange. Cholesterol binding motif in pink. B) Western blot analysis of x-alk-chol binding proteins CAV1 and IFITM3. Interferon (IFN-α) treated HeLa cells were incubated with x-alk-chol (10 µM) for 30 min and UV-irradiated for 5 min and after cell lysis, proteins were separated by SDS-PAGE. Western blot analysis shows slower migration rate for CAV1 and IFITM3 for UV treated sample, suggesting x-alk-chol crosslinking. Western blot analysis of streptavidin-biotin pull down of x-alk-chol crosslinked proteins. Cell lysates from input was reacted with azide-biotin for streptavidin pull down of x-alk-chol crosslinked proteins. Further western blot analysis shows UV dependent pull down of endogenous CAV1 and IFITM3. C) x-alk-chol labeling of overexpressed HA-tagged IFITM3 wildtype and cholesterol binding mutants in HEK293T cells. In-gel fluorescence profiling shows UV-dependent fluorescence signal for IFITM3 WT. IFITM3 PalmΔ and CARCΔ shows significantly lower fluorescence signal, suggesting lower level of x-alk-chol labeling. Anti-HA blot shows IFITM3 expression for each construct. D) Quantitative analysis of x-alk-chol labeling of IFITM3 constructs in C. The fluorescence signal normalized to protein level was plotted for each construct. Data represents the mean and s.e.m. of three independent experiments. P values were determined by one-way anova with a Dunnett’s multiple comparisons test. **P<.01, ***P<.001. E) Influenza A virus (IAV) infection of A549 IFITM1/2/3 KO cells stably expressing IFITM3 constructs. Virus nucleoprotein (NP) levels were examined by flow cytometry for percentage of infection analysis. Data represents the mean and s.e.m. for triplicates. F) rVSV-SARS-CoV-2 S infection of A549 IFITM1/2/3 KO cells stably expressing IFITM3 constructs. Viral Infectivity was examined by counting the eGFP-positive cells using fluorescent microscopy. Data represents the mean and s.e.m. for five independent experiments.

To analyze the significance of IFITM3 interaction with cholesterol, we evaluated antiviral activity of IFITM3 cholesterol binding mutants against IAV and SARS-CoV-2. A549 IFITM1/2/3 KO – ACE2 cells stably expressing IFITM3 constructs were used for the infection studies (Supplementary Figure 12A, B). Cells expressing IFITM3 WT shows antiviral activity against IAV whereas loss of function construct IFITM3 PalmΔ has little or no activity¹⁹. Cells expressing IFITM3 CARCΔ construct shows significant loss of resistance to IAV infection (Figure 3E, Supplementary Figure 13). HEK293T cells expressing these IFITM3 constructs show similar trend in antiviral activity against IAV (Supplementary Figure 14A, B). Next, we tested antiviral activity of these cell lines against a recombinant vesicular stomatitis virus bearing SARS-CoV-2 spike (rVSV-SARS-CoV-2 S)⁴⁷, because recent studies show that SARS-CoV-2 spike protein interaction with cholesterol is important for virus entry and pathological syncytia formation³⁴. Though IFITM3 is shown to be protective against SARS-CoV-2 infections in mice⁴⁸, early cell based studies show conflicting results for both inhibition of SARS-CoV-2 pseudotyped virus infection and SARS-CoV-2 spike protein mediated cell-cell fusion^34,49,50,51. Furthermore, although overexpression of IFITMs restrict SARS-CoV-2 infection, endogenous levels of IFITMs was suggested to have a proviral effect^52,53. Our A549 IFITM1/2/3 KO cells provides an excellent system to evaluate the activity of IFITMs in SARS-CoV-2 entry (Supplementary Figure 12A, B). Cells expressing IFITM3 WT shows antiviral activity against rVSV-SARS-CoV-2 S whereas IFITM3 PalmΔ or CARCΔ mutants have minimal or no antiviral activity (Figure 3F). Thus, IFITM3 cholesterol interaction might play an important role in blocking virus fusion and release of genetic material in host cytosol during IAV and SARS-CoV-2 infection entry.

Cholesterol alters IFITM3 conformations in membrane bilayers from molecular dynamics simulations.

To explore the effect of cholesterol on IFITMs in membrane bilayers, we performed molecular dynamics (MD) simulations of apo-IFITM3 and C72, C105 S-palmitoylated (palm) IFITM3 in a 1,2-dimyristoyl-sn-glycero-3-phosphocholine (DMPC) membrane bilayer with or without 20% cholesterol (Figure 4A, Supplementary Figure 15A and Supplementary table 1)(Pogozheva et al. 2022; Garst et al. 2021). We measured tilt angles of each helical domain with respect to the membrane normal (i.e., Z-axis), Z positions of the center of mass of each helix with respect to the bilayer center (i.e., Z=0), as well as the interaction patterns of each residue to quantify behaviors of each helix namely, AH1 (residues 62–67), AH2 (residues 76–85) and TM (residues 109–131). Each helical domain shows different behaviors on addition of cholesterol in DMPC membrane systems. However, we observed similar change in trends for the helices in both apo-IFITM3 and S-palm IFITM3. C72, 105 S-palm IFITM3 shows more interactions of AH1 and Loop2 (residues 86–108) with DMPC membrane than apo-IFITM3 (Figure 4B, Supplementary Figure 15B, Supplementary table 1). This is consistent with our previous studies, which showed that S-palmitoylation at Cys72 increased IFITM3 AH1 interaction with DMPC membranes⁵⁴. In the presence of cholesterol in the membrane, IFITM3 interacts with cholesterol via the residues around AH1, Loop1 and TM domain. MD simulation of IFITM3 CARCΔ shows subtle changes in cholesterol binding (Supplementary Figure 16). In the presence of cholesterol, both apo- and S-palm IFITM3 AH1 shows further increase in membrane interactions as its tilt angles are closer to 90°, a completely horizontal orientation. AH1 has a tilt angle of 86.85° for apo-IFITM3 and 77.7° for S-palm IFITM3 in cholesterol containing systems versus 101.53° for apo-IFITM3 and 108.4° for S-palm IFITM3 in DMPC-only systems (Supplementary table 2). Additionally, AH1 domain was more deeply embedded into the membrane in the presence of cholesterol, thus making more frequent interactions with lipid tails (Figure 4B, C, Supplementary Figure 15B). AH2, on the other hand, responses differently to the presence of cholesterol in comparison to AH1. AH2 tilt angle shows values larger than 90°, meaning that it becomes more vertical in a membrane system with cholesterol (Supplementary table 2). In addition, presence of cholesterol does not affect the position of AH2 along the membrane normal in contrast to AH1 (Figure 4C). For TM domain, the tilt angle tends to decrease in the cholesterol membrane system for both apo-IFITM3 and S-palm IFITM3 (Figure 4D). Such tilt angle changes are known to occur to maximize a hydrophobic match between the lipid bilayer and the transmembrane domain⁵⁵. Cholesterol is known to increase the thickness and order of a lipid bilayer⁵⁶, as we observed for all cholesterol containing membrane systems (Supplementary Figure 17). Interestingly, we observed significant changes in the interaction pattern for Loop2 in membrane with and without cholesterol for both apo-IFITM3 and S-palm IFITM3. Loop2 is a part of the highly conserved CD225 domain which contains a basic patch consisting of R85, R87, and K88, as well as a ⁹¹GxxxG⁹⁵ motif implicated in IFITM3 oligomerization and antiviral activity^24,57. We see increased hydrophobic interactions in Loop2 in the DMPC system, whereas the cholesterol containing membrane system has more hydrophilic (protein-water and protein-lipid headgroup) interactions (Supplementary Figure 18–21). Since DMPC is a fully saturated lipid and is well-mixed with rigid cholesterol, it leads to a more ordered membrane phase. This makes it difficult for Loop 2 to insert into the membrane hydrophobic core and have hydrophobic interactions. Thus, cholesterol-induced changes in IFITM3-membrane interactions might in turn influence IFITM3 interactions with other lipids and proteins as well as its activity.

Figure 4. Molecular dynamics simulation of S-palmitoylated IFITM3 in cholesterol-containing membrane bilayer.

A) Simulation snapshots of C72, 105 S-palm IFITM3 in DMPC (top) and DMPC + Chol membrane bilayers (bottom). B) Interaction frequency of each residue of C72, 105 S-palm IFITM3 in DMPC (top) and DMPC + Chol membrane bilayers (bottom) interacting with surrounding environment, DMPC lipid headgroups (red), DMPC lipid tails (gray), water (blue), cholesterol headgroup (magenta), and cholesterol tails (pink). Each graph shows the interaction frequency within 4 Å from each residue. C) The distance of the center of mass for each helical domain (AH1, AH2 and TM) to the membrane center (Z=0) for apo-IFITM3 and C72, 105 S-palm IFITM3 in DMPC or DMPC + Chol. Data represents the mean and SE for triplicates. D) The averaged tilt angle of each helical domain for apo-IFITM3 and C72, 105 S-palm IFITM3 in DMPC or DMPC + Chol. Averaged tilt angle was measured by calculating the angle between the principal axis of each helix and the membrane normal (Z-axis). Data represents the mean and SE for triplicates.

To structurally characterize IFITM3 in a cholesterol-containing bilayer, we employed a bicelle membrane system. However, in the presence of full length IFITM3, the short chain lipid, 1,2-dihexanoyl-sn-glycero-3-phosphatidylcholine (DHPC) could not solubilize DMPC liposomes to form membrane bicelles. We, therefore, designed several truncated constructs for bicelle reconstitution. IFITM3⁸⁹⁻¹³³, which excluded the amphipathic helices, expressed well after overnight induction, and were purified using His-affinity purification (Supplementary Figure 22A, B). We characterized ¹⁵N-labeled IFITM3⁸⁹⁻¹³³ in bicelles with a protein to lipid ratio of 1:50 and final q value between 0.5 and 0.6 using 2D 1H-15N TROSY spectra (Supplementary Figure 23). We observed subtle changes in some specific residues of the TM domain though full assignment of the TM residues could not be performed due to low signal.

IFITMs show different S-palmitoylation levels, x-alk-chol labeling and antiviral activity.

We next analyzed cholesterol binding of human IFITM1, IFITM2 and IFITM3. All three IFITM proteins have conserved Cys71, 72, 105 and cholesterol binding motif (Figure 5A). We used in-gel fluorescence profiling to evaluate photocrosslinking of IFITMs with x-alk-chol. In-gel fluorescence detection and quantification normalized to protein levels showed x-alk-chol photocrosslinking to IFITM1 and IFITM3 but not with IFITM2 (Figure 5B, C, Supplementary Figure 24A, B). Differential photocrosslinking of IFITM2 and IFITM3 with x-alk-chol is surprising since the protein sequence alignment shows 83% sequence identity. They also showed similar subcellular localization by immunofluorescence confocal microscopy (Supplementary Figure 25). However, the S-palmitoylation levels of IFITM2 and IFITM3 by alk-16 labeling correlated with x-alk-chol photocrosslinking (Figure 5D, E, Supplementary Figure 24C, D). These results suggest that IFITM2 and IFITM3 interactions with cholesterol are consistent with their S-palmitoylation levels.

Figure 5. Cholesterol binding, S-palmitoylation and antiviral activity of IFITM proteins.

A) Alignment, and topology of IFITM proteins. The distinct residues in IFITM2 and IFITM3 are in cyan. IFITM3 has an additional Phe residue in red. CARC Lys, Trp and Leu residues are highlighted in pink. B) x-alk-chol labeling of overexpressed HA-tagged IFITM proteins in HEK293T cells. In-gel fluorescence profiling shows UV-dependent fluorescence signal for IFITMs. Anti-HA blot shows IFITM expression for each construct. The fluorescence band marked with asterisk (*) is used for further quantification. C) Quantitative analysis of x-alk-chol labeling of IFITM constructs in b, respectively. The fluorescence signal normalized to protein level was plotted for each construct. Data represents the mean and s.e.m. of three independent experiments. P values were determined by one-way anova with a Dunnett’s multiple comparisons test. 0.5<ns, ****P<.0001. D) S-palmitoylation analysis of overexpressed HA-tagged IFITM constructs in HEK293T cells. In-gel fluorescence profiling shows UV-dependent fluorescence signal for IFITMs. Anti-HA blot shows IFITM3 expression for each construct. E) Quantitative analysis of alk-16 labeling of IFITM3 constructs in d. The fluorescence signal normalized to protein level was plotted for each construct. Data represents the mean and s.e.m. of three independent experiments. P values were determined by one-way anova with a Dunnett’s multiple comparisons test. 0.5<ns, **P<.01, ***P<.001. F) Influenza A virus (IAV) infection of HEK293T and A549 IFITM1/2/3 KO cells expressing IFITM constructs. Cells were infected with IAV (MOI 10) for 6 h. Virus nucleoprotein (NP) levels were examined by flow cytometry using anti-NP staining for percentage of infection analysis. Data represents the mean and s.e.m. for three independent experiments. P values were determined by one-way anova with a Tukey’s multiple comparisons test. ***P<.001. G) SARS-CoV-2 entry in A549 IFITM1/2/3 KO cells stably expressing IFITMs. Cells were infected with rVSV-eGFP SARS-CoV-2 for 7 h. Viral Infectivity was examined by counting the eGFP-positive cells using fluorescent microscopy. Data represents the mean and s.e.m. for five independent experiments. P values were determined by one-way anova with a Tukey’s multiple comparisons test. 0.5<ns, **P<.01, ****P<.0001. H) EBOV entry in A549 IFITM1/2/3 KO cells stably expressing IFITMs. Cells were infected with rVSV-eGFP EBOV GP for 7 h. Viral Infectivity was examined by counting the eGFP-positive cells using fluorescent microscopy. Data represents the mean and s.e.m. for five independent experiments. P values were determined by one-way anova with a Tukey’s multiple comparisons test. 0.5<ns, **P<.01, ****P<.0001.

We then evaluated the antiviral activity of IFITM proteins with IAV, SARS-CoV-2 and EBOV entry. IAV and SARS-CoV-2 use sialylated EGFR⁵⁸ and ACE2⁵⁹ as entry factor respectively, whereas EBOV, another virus entering from late endosomes, has been shown to use NPC1 as an entry factor^60–62. For these experiments, we stably expressed the IFITM proteins in A549 IFITM1/2/3 KO – ACE2 cells (Supplementary Figure 26A, B). All IFITM proteins show comparable expression levels in A549 IFITM1/2/3 KO – ACE2 cells (Supplementary Figure 26C). Consistent with previous reports^63,64, expression of plasma membrane resident IFITM1 did not inhibit infection by these viruses that enter through low pH endosomal compartments (Figure 5F, G, H). In contrast, we found that IFITM3 exhibits greater antiviral activity against IAV and rVSV-SARS2 S, whereas IFITM2 is more active against an rVSV bearing the EBOV spike glycoprotein, GP (rVSV-EBOV GP)⁶⁵ (Figure 5F, G, H). These results suggest that differential interactions with cholesterol and S-palmitoylation impact the antiviral activity of IFITM proteins towards different viruses.

Conclusion

IFITMs have emerged as important IFN-induced effectors in host immunity against diverse viruses but require further mechanistic insight. Here, we explored IFITM-cholesterol interactions using chemical biology approaches and in silico, structural methods. We employed photoaffinity crosslinking to investigate direct cholesterol-protein interactions in cells. Our proteomic analysis of cholesterol photoaffinity crosslinking revealed many candidate sterol-interacting proteins, including endogenously expressed IFITM3 in interferon (IFN-α)-stimulated cells. Further targeted crosslinking studies identified a cholesterol binding motif adjacent to the transmembrane domain of IFITM3. Infection studies with IFITM3 cholesterol binding mutants suggest that IFITM3-cholesterol interaction can play an important role in blocking virus entry into host cells mediated by the IAV and SARS-CoV-2 spike glycoproteins. This is in agreement with recent in vitro work, which shows that cone-shaped lipids like cholesterol facilitate IFITM3-induced membrane curvature to inhibit viral fusion with host membranes³⁶. Moreover, we show that S-palmitoylation of IFITM3 is crucial for cholesterol photoaffinity crosslinking in cells. Molecular dynamics simulation revealed potential effects of cholesterol on IFITM conformations and membrane interactions in lipid bilayers, which was supported in part by our NMR studies of IFITM3 transmembrane domain reconstituted in membrane bicelles. Our in silico studies also suggested that cholesterol may impact IFITM3 AH1 and Loop2 interactions with the membrane. This is particularly interesting since IFITM3 AH1 is important in inhibiting viral infection in cells^17,36 and IFITM3 Loop2 contains motifs implicated in IFITM3 oligomerization and antiviral activity^24,57. Additional studies with different IFITM proteins led to the surprising discovery that IFITM2 exhibits significantly less x-alk-chol crosslinking than IFITM3, even though both contain the conserved Cys and cholesterol binding domain.

Further analysis by metabolic labeling revealed that S-palmitoylation level of IFITM2 is significantly less than IFITM3, which may explain the differential x-alk-chol crosslinking of the IFITMs. Interestingly, infection studies in A549 cell lines shows that IFITM3 has greater antiviral activity against IAV and SARS-CoV-2 infection, whereas IFITM2 is more active against EBOV, that enter host cells through interactions with NPC1 in late endosomes^60–62,66. This is in agreement with previous observations that IFITM3 restricts viruses like IAV or flaviviruses more efficiently than cathepsin-dependent viruses like EBOV^60–62,66. Interestingly, amphotericin B, which can bind cholesterol, has been reported to inhibit IFITM3 restriction of IAV but not EBOV⁶³, suggesting that differential engagement of cholesterol-rich membranes or endosomes by IFITMs can impact selective inhibition of IAV versus EBOV infection. Our results now provide direct evidence for this possibility, as we show that higher levels of S-palmitoylation enhance IFITM3 interaction with cholesterol and inhibits viruses like IAV and SARS-CoV-2, whereas IFITM2 exhibits lower S-palmitoylation levels and cholesterol interactions but more effectively inhibits EBOV, which may engage cellular membranes with less cholesterol that contain NPC1 transporter. Therefore, while both IFITM2 and IFITM3 are present in the endolysosomal pathway, their differential S-palmitoylation levels and cholesterol interactions may provide specificity against viruses that enter through different endosomal trafficking routes and membrane compartments. Therefore, it will be interesting to investigate the cellular and biochemical mechanisms that regulate differential palmitoylation and depalmitoylation of IFITMs.^67,68 Our study highlights how direct cholesterol photoaffinity labeling in cells can reveal unpredicted covalent and non-covalent lipid regulation of IFN effectors that maybe harnessed to modulate their activity and/or specificity towards different viruses in the future.

Methods

Labeling of HeLa cell proteome with x-alk-chol.

HeLa cells were incubated with different concentrations (5, 10, 15 and 20 μM) of x-alk-chol in mβCD for 30 min. For cholesterol competition, cells were first treated with 10 μM of x-alk-chol for 30 min and were further treated with 10 μM cholesterol for 30 min. Cells were washed with PBS and UV treated for 5 min. Cells were lysed in 1% Brij 97 in 50 mM TEA, 150 mM NaCl pH 7.4, 1X Roche protease inhibitor, and 1,500 units/mL benzonase (EMD). Protein concentrations were determined by the BCA assay.

45 μl of the samples (50 μg of protein) were added to 5 μl of CuAAC reactant solution (0.5 μl of 10 mM azido-rhodamine (final concentration 100 μM), 1 μl of 50 mM freshly prepared CuSO₄·5H₂O in H₂O (final concentration 1 mM), 1 μl of 50 mM freshly prepared TCEP (final concentration 1 mM) and 2.5 μl of 10 mM Tris[(1-benzyl-1H-1,2,3-triazol-4-yl)methyl]amine (TBTA) (final concentration 500 μM)). The samples were rocked at room temperature for 1 h and protein was precipitated with methanol-chloroform-water mixture (4:1.5:3). After air-drying, the protein pellets were dissolved in 1X Laemmli sample buffer (20 μl) and were heated for 10 min at 95 °C and separated by gel electrophoresis.

Proteomic analysis.

HeLa cells were grown to confluency in 10 cm dishes. Cells were stimulated with IFN-α (100 µg/mL) overnight. Next cells were treated with x-alk-chol (10 μM) for 30 min and UV treated for 5 min. Cells were lysed with 4% SDS in 50 mM HEPES buffer with 150 mM NaCl. 2 mg of cell lysate was reacted with azido-biotin for 4 h (1 mM CuSO₄, 1 mM TCEP, 200 μM TBTA, 200 μM azido-biotin) at 1 mg/mL. Protein was precipitated using methanol (8 mL), chloroform (3 mL), and water (6 mL) overnight at −20 °C. Protein pellets were dried for 1 h and solubilized in 200 μL of 4% SDS buffer. The sample was diluted to 1 mg/mL with 50 mM HEPES buffer. 25 μL of streptavidin beads were added to the samples and were nutated for 1 h. The beads were washed with 1% SDS, 5 M Urea, and PBS. Samples were split and half the beads were boiled in 20 μL of 4% SDS loading buffer and separated by SDS-PAGE for western blot analysis. The other half was suspended in 200 μL of 25 mM ammonium bicarbonate, reduced with 1 mM DTT for 30 min, and then alkylated with 50 mM iodoacetamide in the dark for 30 min. Then, the beads were washed with 200 μL of 25 mM ammonium bicarbonate and suspended in 50 μL of 25 mM ammonium bicarbonate. 0.1 μg of trypsin was added, and the samples were digested at 37 °C overnight. The supernatant was collected and dried on a speedvac and solubilized in 5% acetonitrile/1% formic acid for LC-MS/MS analysis.

Extracted tryptic peptides were desalted on a trap column following separation on a 12 cm/75 μm reversed phase C₁₈ column (Nikkyo Technos Co., Ltd. Japan). A 180-minute gradient increasing from 10% B to% 45% B in 133 minutes (A: 0.1% Formic Acid, B: Acetonitrile/0.1% Formic Acid) were delivered at 200 nL/min. The liquid chromatography setup (Dionex, Boston, MA, USA) was connected to an Orbitrap XL (Thermo, San Jose, CA, USA) operated in top-8-CID-mode with MS spectra measured at a resolution of 60,000@m/z 400. Acquired tandem MS spectra were extracted using Maxquant queried against the Uniprot complete human database and processed using Perseus. Absent values were imputed based on the normal distribution of values. To determine if a protein was a valid hit, the false discovery rate was lower than 0.01% and the mean difference of the control and labeled samples had be greater than 2 with p<0.05 determined by the t-test. Database for Annotation, Visualization and Integrated Discovery (DAVID) was used for Gene Ontology (GO) enrichment analysis.