N-(4-Hydroxyphenyl)retinamide suppresses SARS-CoV-2 spike protein-mediated cell-cell fusion and viral infection in vitro

The coronavirus disease (COVID-19) pandemic, caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), persists worldwide with limited therapeutic options. Since membrane fusion between SARS-CoV-2 and host cells is essential for the early step of the infection, the membrane compositions, including sphingolipids, in host cells are considered to affect the viral infection. However, the role of sphingolipids in the life cycle of SARS-CoV-2 remains unclear. Here, we assessed several inhibitors of sphingolipid metabolism enzymes against SARS-CoV-2 spike protein-mediated cell-cell fusion and viral infection in vitro. Among the compounds tested, only N-(4-hydroxyphenyl)retinamide (4-HPR, also known as fenretinide), an inhibitor of dihydroceramide Δ4-desaturase 1 (DES1) and well known for having antitumour activity, suppressed cell-cell fusion (50% effective concentration [EC 50 ] = 4.1 µM) and viral infection ([EC 50 ] = 4.4 µM), wherein the EC 50 values are below its plasma concentration in previous clinical trials on tumours. DES1 catalyses the introduction of a double bond in dihydroceramide, and the inhibition eciencies observed were consistent with an increased ratio of saturated sphinganine-based lipids to total sphingolipids and the decreased cellular membrane uidity. These ndings, together with the accumulated clinical data regarding the safety of 4-HPR, make it a likely candidate drug to treat COVID-19. DSP1-7 and DSP8-11 reassociate eciently, resulting in reconstitution of a functional RL and GFP to generate luminescent and uorescent signals, respectively. (b) Cell viability was examined using the WST-8 assay. Results are normalised to the rate of cell viability in vehicle/ dimethyl sulfoxide (DMSO)-treated cells. (c) The susceptibility of cell-cell fusion was examined using the DSP-based cell-cell fusion assay. Results are normalised to the rate of cell-cell fusion in vehicle/DMSO-treated cells. Values represent the mean ± S.D. from three independent experiments. Statistical signicance was determined by one-way ANOVA followed by Dunnett test for multiple comparisons; * p < 0.05, ** p < 0.01 when compared with vehicle/DMSO-treated cells. CC50 and EC50 values were determined using GraphPad Prism software. CC50, 50% cytotoxic concentration; 50% effective

Sphingolipids are critical to all stages of the viral life cycle, including the binding of human rhinovirus 5 , entry of in uenza virus 6 , replication of hepatitis C virus 7 , and new particle budding of adenovirus 8 . We found that sphingomyelin synthase 2, involved in the synthesis of sphingomyelin (SM), can promote HIV-1 envelope-mediated membrane fusion 9 . Recently, Vitner et al. 10  Sphingolipid metabolism involves a dynamic network of molecules, including important bioactive signalling molecules ( Fig. 1) 11,12 . De novo biosynthesis of sphingolipids occurs in the endoplasmic reticulum and begins with the condensation of L-serine with palmitoyl-CoA to produce 3ketodihydrosphingosine in a reaction catalysed by serine palmitoyltransferase (SPT). 3-Ketodihydrosphingosine is reduced to dihydrosphingosine, which is N-acylated to dihydroceramide (DHCer) by ceramide synthase (CerS). The N-acyl chain length of DHCer is determined by the speci city of different CerSs; C16:0, C18:0, C24:0, and C24:1 are the major fatty acids incorporated into DHCer in mammalian cells 13 . The formation of ceramide (Cer), the core structure of sphingolipids, involves insertion of a single double bond into DHCer by dihydroceramide Δ4-desaturase 1 (DES1). Cer, and to a lesser extent DHCer, is further metabolised to form complex sphingolipids, such as sphingomyelins (SM and DHSM, respectively) and glucosylceramides (GlcCer and DHGlcCer, respectively) in the Golgi apparatus. Thus, DHCer, DHSM, and DHGlcCer are composed of saturated sphingoid base backbones (sphinganine), while Cer, SM, and GlcCer are composed of unsaturated sphingoid base backbones (sphingosine: Sph). Most of the genes encoding sphingolipid-metabolizing enzymes, as well as several inhibitors of sphingolipid-metabolizing enzymes, have been identi ed [14][15][16][17][18][19][20][21][22][23][24] .
SARS-Cov-2 needs to penetrate the cell membrane in order to infect cells. Therefore, in the present study, we tested a compound, N-(4-hydroxyphenyl)retinamide (4-HPR, also known as fenretinide), which would change the membrane environment in such a way that makes virus penetration di cult, thereby preventing infection. We found that 4-HPR could prevent viral entry into host cells, due to decreased membrane uidity. Furthermore, to understand the function of sphingolipids in SARS-CoV-2 infection, we analysed the quantitative sphingolipid metabolome of inhibitor-treated cells using liquid chromatography-tandem mass spectrometry (LC-MS/MS). We found that the quality of the sphingoid base backbone (saturated or unsaturated), and not the total quantity of sphingolipids, is involved in SARS-CoV-2 infection.

4-HPR inhibits SARS-CoV-2 spike (S) protein-mediated cell-cell fusion
Since membrane fusion between SARS-CoV-2 and host cells is essential for the early step of the infection, the membrane compositions, including sphingolipids, in host cells are considered to affect the viral infection. To investigate the roles of sphingolipids in SARS-CoV-2 S protein-meditated membrane fusion, we employed the dual split protein (DSP)-based cell-cell fusion assay using SARS-CoV-2 mimicking cells and sphingolipid-manipulated target cells treated with an inhibitor of sphingolipid metabolism enzyme.
We next assessed the effects of the 13 compounds against the SARS-CoV-2 S protein-mediated cell-cell fusion at nontoxic concentrations of each compound (Fig. 2c). The analysis of cell-cell fusion revealed that 4-HPR, threo-PPMP, and erythro-PPMP inhibited DSP activity by more than 50% compared with that of the solvent-only control containing dimethyl sulfoxide (DMSO). Importantly, luciferase activity derived from cells carrying the preformed DSP1-7/DSP8-11 reporter complex was not affected by 4-HPR, threo-PPMP, and erythro-PPMP ( Supplementary Fig. 1a), indicating that the suppression of luciferase activity re ects the inhibition of cell-cell fusion by these compounds. However, little or no anti-membrane fusion activity was found for myriocin, fumonicin B1, GT11, CBE, HPA12, GW4869, amitriptyline, ceranib-2, SKI-II, and FTY720.
To investigate changes in sphingolipid levels in inhibitor-treated cells, we next quanti ed sphingolipids using LC-MS/MS. Importantly, the levels of all species with a distinct acyl chain in DHCer, DHGlcCer, and DHSM (panels g-l in Supplementary Figs. 2, 3, and 4) and total levels of DHCer, DHGlcCer, and DHSM (lower panels in Fig. 3a-c) increased in 5 µM 4-HPR-treated cells compared with that in 10 µM GT11treated cells. This observation was despite the fact that both compounds inhibit the same enzyme, DES1. The ratio of saturated sphinganine-based lipids to total sphingolipids was close to 20% in 5 µM 4-HPRtreated cells, but 10% for 2.5 µM 4-HPR or 10 µM GT11 treatments (Fig. 3e). We found that the increased ratio was consistent with the inhibition e ciencies of cell-cell fusion ( Fig. 2c and Fig. 3e).
As expected, treatment with myriocin and fumonicin B1, inhibitors of de novo sphingolipid biosynthesis, signi cantly decreased the levels of most species with a distinct acyl chain in Cer, GlcCer, and SM (panels a-f in Supplementary Figs. 2, 3, and 4), and total levels of Cer, GlcCer, and SM (upper panels in Fig. 3a-c) compared with that for DMSO treatment. As a result, cells treated with myriocin and fumonicin B1 showed a reduction of approximately 65% and 79% in total sphingolipid levels, respectively (Fig. 3d). These changes had little effect on SARS-CoV-2 S protein-mediated cell-cell fusion, indicating that the total quantity of sphingolipid was not involved.
Contrary to the expectations, cells treated with 10 µM erythro-PPMP, an inactive stereoisomer of GCS inhibitor threo-PPMP, exhibited a decreased level of GlcCer compared with that in DMSO-treated cells, although to a lesser extent than that in cells treated with threo-PPMP (upper panel in Fig. 3b). Given that myriocin and fumonicin B1 reduced the levels of all GlcCer species as much as, or more than that by threo-PPMP (panels a-f in Supplementary Fig. 3), no correlation was observed between the GlcCer levels and inhibition e ciency against cell-cell fusion. Thus, the inhibition of cell-cell fusion caused by 10 µM threo-PPMP and 10 µM erythro-PPMP might be potential off-target effects.

4-HPR inhibits SARS-CoV-2 infection in vitro
We further examined antiviral activity of three compounds 4-HPR, threo-PPMP, and erythro-PPMP, which inhibited SARS-CoV-2 S protein-mediated cell-cell fusion by more than 50%. In addition to these compounds, myriocin and GT11 were also used to investigate the function of sphingolipids in viral infection. When VeroE6 TMPRSS2 cells were exposed to various concentrations of the compounds for 3 days, myriocin, at a concentration above 20 µM, exhibited no effect on cell viability (Fig. 4a). In contrast, threo-PPMP and erythro-PPMP exhibited cytotoxicity at concentrations above 20 µM, while 4-HPR and GT11 showed cytotoxicity above 10 µM.
The analysis of sphingolipid levels with LC-MS/MS revealed that the ratio of saturated sphinganinebased lipids to total sphingolipids in 4-HPR-treated cells increased in a concentration-dependent manner, exhibiting 25% inhibition for 5 µM 4-HPR, 17% for 2.5 µM 4-HPR, and 12% for 1.3 µM 4-HPR treatment ( Fig. 4c). Furthermore, no signi cant difference in the ratio of saturated sphinganine-based lipids to total sphingolipids was observed between 1.3 µM 4-HPR and 5 µM GT11 treatments (Fig. 4c). These results were consistent with the inhibition e ciencies of viral infection (Fig. 4b). Thus, the quality of the sphingoid base backbone, saturated or unsaturated, was found to be involved in SARS-CoV-2 infection.
The cells treated with myriocin showed approximately 50% reduction in total sphingolipid levels ( Fig. 4d), and those with threo-PPMP showed approximately 60% reduction in total GlcCer levels (upper panel in Supplementary Fig. 5b) compared with that by treatment with DMSO. These results indicated that the total quantity of sphingolipids, including GlcCer, was not involved in SARS-CoV-2 infection.

4-HPR-treated cells display decreases in membrane uidity
Finally, we examined the cellular effect of 4-HPR to obtain mechanistic insight into the inhibitory effect of 4-HPR on SARS-CoV-2 S protein-mediated cell-cell fusion and viral infection. Because cell-cell fusion and viral infection were considered to be strictly dependent on the cell-surface levels of the SARS-CoV-2 receptor ACE2, we examined the effect of 4-HPR on the expression of ACE2 in 293FT/ACE2/TMPRSS2/DSP1-7 cells and VeroE6 TMPRSS2 cells. Importantly, no signi cant differences were observed in the ACE2 expression levels in the 293FT/ACE2/TMPRSS2/DSP1-7 cells treated with 5 µM 4-HPR compared with that in the cells treated with DMSO or 10 µM GT11 (Fig. 5a). Similar results were obtained with VeroE6 TMPRSS2 cells ( Supplementary Fig. 9a). These results indicated that the inhibition of cell-cell fusion and viral infection by 4-HPR was not attributable to the ACE2 expression levels on the cell surface.
Lipid rafts are de ned as detergent-resistant sphingolipid-rich microdomains of cellular membranes 26 . Several viral receptors are known to be localised in the lipid rafts and are involved in e cient viral infection 27,28 . We next examined whether 4-HPR treatment affected the localization of ACE2 on the membrane microdomain. When we treated 293FT/ACE2/TMPRSS2/DSP1-7 cells with the non-ionic detergent TritonX-100, ACE2 was found in the detergent-solubilised membrane fraction in which the nonraft marker transferrin receptor protein was localised, but not otillin, the raft marker protein (Fig. 5b).
Furthermore, ACE2 did not redistribute to the lipid raft domains even in the presence of SARS-CoV-2 S protein stimulation (Fig. 5b). Importantly, we observed no dramatic difference in ACE2 localization in the membrane microdomain between the DMSO and 4-HPR treatments (Fig. 5b), indicating that the inhibition of cell fusion by 4-HPR was not attributable to ACE2 localization on the membrane microdomain.
Lipid saturation is known to be important for membrane uidity 29 . In addition, it has been demonstrated that an increase in DHSM is associated with the rigidi cation of the cell membrane, which can inhibit HIV-1 infection 30 . We therefore examined the effect of 4-HPR on cellular membrane uidity using lipophilic pyrene probes, which undergo excimer formation with increasing membrane uidity 31 . Cells treated with 5 µM and 2.5 µM 4-HPR exhibited decreased membrane uidity by about 50% and 20%, respectively, compared with that in cells treated with DMSO (Fig. 5c). In addition, membrane uidity was not signi cantly different between the cells treated with 2.5 µM 4-HPR and 10 µM GT11. Importantly, these results were consistent with the inhibition e ciencies of SARS-CoV-2 S protein-mediated cell-cell fusion (Fig. 2c) and the increased ratio of saturated sphinganine-based lipids to total sphingolipids (Fig. 3e).
Similar results were obtained with VeroE6 TMPRSS2 cells ( Supplementary Fig. 9b). These ndings suggested that a decrease in cellular membrane uidity by 4-HPR might be a major cause of the inhibition of SARS-CoV-2 S protein-mediated cell-cell fusion and viral infection.

Discussion
In this study, we clearly showed that 4-HPR could prevent SARS-CoV-2 entry into cells due to a decrease in membrane uidity, and not the location of the ACE2 receptor within the membrane compartments. Furthermore, we found that the quality of the sphingoid base backbone, and not the total quantity of sphingolipids, was involved in SARS-CoV-2 infection.
Interestingly, no signi cant activity against cell-cell fusion or viral infection was found for myriocin, although the total sphingolipid levels signi cantly decreased in the myriocin-treated cells compared with that in the DMSO-treated cells. Lipid rafts, sphingolipid-enriched microdomains in cellular membranes, were not a major localization site for ACE2 in 293FT/ACE2/TMPRSS2/DSP1-7 cells. This might explain why the quantity of sphingolipids did not appear to be involved in SARS-CoV-2 infection. We observed no clear localization of ACE2 in VeroE6 TMPRSS2 cells, as determined biochemically using TritonX-100 solubilization, since the expression level of ACE2 in these cells was lower than in 293FT/ACE2/TMPRSS2/DSP1-7 cells (data not shown). It has been reported that ACE2 is localised within non-raft domains in Chinese hamster ovary cells 32 and VeroE6 cells 33 , whereas other groups have reported ACE2 localization within raft domains in VeroE6 cells 34,35 . Although the cause of discrepancies between these ndings is unknown, importantly, in this study ACE2 did not redistribute to the lipid raft domains even in the presence of SARS-CoV-2 S protein stimulation. Given that, we propose that ACE2 localization on the membrane microdomain is not essential for SARS-CoV-2 entry into host cells at least in our infection study.
We found that the increased ratio of saturated sphinganine-based lipids (DHCer, DHGlcCer, and DHSM) to total sphingolipids was correlated with the inhibition e ciencies of cell-cell fusion and viral infection. The antiviral activity of 4-HPR has been proved in different viruses [36][37][38][39] ; however, the mechanism of inhibition remains unclear. Finnegan et al. 36 reported that 4-HPR increases cellular Cer, and not DHCer, levels, which is thought to be a cause of inhibition of HIV-1 infection. More speci cally, at that time it was thought that 4-HPR elevates Cer levels, but not DHCer, by activation of SPT and CerS 40,41 . However, it was later revealed that the increased lipid level is due to DHCer rather than Cer; DHCer accumulation is caused by 4-HPR inhibition of DES1 42 . Therefore, here, we rst clari ed the relationship between the accumulation of saturated sphinganine-based lipids by 4-HPR and suppression of viral infection. Among the saturated sphinganine-based lipids in 293FT/ACE2/TMPRSS2/DSP1-7 cells treated with 4-HPR, DHSM constituted 65%, DHCer 34.5%, and DHGlcCer 0.5% (data not shown). Further research is needed to elucidate the role of each lipid involved in infection, which will aid in establishing the functions of saturated sphinganinebased lipids against SARS-CoV-2 infection.
Treatment with 4-HPR increased the ratio of saturated sphinganine-based lipids to total sphingolipids more than that by treatment with GT11 in 293FT/ACE2/TMPRSS2/DSP1-7 cells and VeroE6 TMPRSS2 cells, although GT11 competitively inhibits DES1 activity in rat liver microsomes higher than 4-HPR (IC 50 of GT11 = 0.69 µM; IC 50 of 4-HPR= 2.32 µM) 43 . It is also known that 4-HPR increases the levels of reactive oxygen species intracellularly 44 , and DES1 activity is inhibited by oxidative stress 45 . Casasampere et al. 12 suggested that 4-HPR inhibition of DES1 may occur indirectly through increased oxidative species. Therefore, both direct and indirect inhibitory effects of 4-HPR on DES1 might have effectively increased the ratio of saturated sphinganine-based lipids to total sphingolipids more than that by GT11.
Our ndings clearly indicated that 4-HPR potently inhibited SARS-CoV-2 S protein-mediated membrane fusion in a cell-cell fusion assay system (EC 50

= 4.1 µM) and viral infection in vitro (EC 50 = 4.4 µM). 4-HPR
is a synthetic derivative of all-trans-retinoic acid, which has been widely investigated for the treatment of cancer [46][47][48][49] . In a phase I study of 4-HPR oral powder administered to patients with relapsed/refractory neuroblastoma, the day 6 mean peak plasma concentration has been noted as 21 µM 4-HPR at a dose of 1500 mg/m 2 per day 46 . Given that the EC 50 value for 4-HPR against SARS-CoV-2 infection was below its plasma concentration (4.4 µM vs. of 21 µM) in this study, 4-HPR might serve as a potential therapeutic agent for COVID-19, as predicted by another group 50 . Furthermore, a low-toxicity pro le of 4-HPR has been observed in many clinical trials, as well as in long-term treatments [46][47][48][49] . Therefore, the ndings of this study, together with the accumulated clinical data regarding the safety of 4-HPR, make it a likely candidate drug to treat COVID-19. Furthermore, 4-HPR could be a potential preventative therapy against SARS-CoV-2 infection, which unlike the vaccines could be administered via an inhaler or a spray to the airways, perhaps by self-administration, thereby relieving the pressure on the healthcare system. Nonetheless, further studies using animal models, such as ferrets, are needed to clarify the role of 4-HPR in SARS-CoV-2 infection in vivo.

DSP-based cell-cell fusion assay
The cell-cell fusion assay was conducted as previously described by Yamamoto et al. 25 , with slight modi cations. Target cells (293FT/ACE2/TMPRSS2/DSP1-7) and effector cells (293FT/SARS-CoV-2 S/DSP8-11) were seeded in 24-well plates at 4 × 10 4 cells/well. On the following day, the medium of the target cells was exchanged with D-MEM containing 10% FCS and various concentrations of each of the compounds to be tested, and the cells were incubated for an additional 2 days. Cells were washed with phosphate-buffered saline (PBS) and were detached using Cellstripper containing EDTA but no trypsin (Corning, Christiansburg, VA, USA). Cells were resuspended in serum-free D-MEM and centrifuged at 500 × g for 2 min. After aspirating the suspension, cells were resuspended in serum-free D-MEM containing 1% Nutridoma SP (Roche, Basel, Switzerland) and 6 mM EnduRen (Promega, Madison, WI, USA), a substrate for RL. The effector and target cells were mixed in the wells of a 96-well plate, and after incubating at 37°C for 4 h the RL activity was measured using a SpectraMax i3x Microplate Reader (Molecular Devices, San Jose, CA, USA).

Antiviral assay
VeroE6 TMPRSS2 cells were seeded in 96-well plates (5 × 10 3 cells/well). On the following day, the cells were cultured with each of the tested compounds for 3 days before adding SARS-CoV-2 05-2N . The cells were inoculated at a multiplicity of infection of 0.01. After culturing the cells with the speci c compounds and SARS-CoV-2 05-2N for 3 days, the level of cytopathic effect observed in SARS-CoV-2-exposed cells was determined using the WST-8 assay.

Lipid extraction and quanti cation of sphingolipids by LC-MS/MS
Lipid extraction and quanti cation of sphingolipids by LC-MS/MS were performed as described previously [52][53][54] . 293FT/ACE2/TMPRSS2/DSP1-7 and VeroE6 TMPRSS2 cells were seeded in 6-well plates at 2 × 10 5 cells/well and 7 × 10 4 cells/well, respectively. On the following day, the cells were cultured with each of the tested compounds for 2 days (293FT/ACE2/TMPRSS2/DSP1-7 cells) or 3 days (VeroE6 TMPRSS2 cells), after which the cells were washed once with cold PBS, collected in 100 mL of cold PBS, and then homogenised by sonication. Part of the sample (5 mL) was used in the bicinchoninic acid protein assay to determine the amount of protein. Total lipids were extracted by adding 375 mL of chloroform:methanol (1:2, v/v) containing 40 pmol of each component, speci cally D31-Cer, D31-SM, and D3-GlcCer. Deuterated D31-Cer, D31-SM, and D3-GlcCer were used as internal standards for Cer and DHCer, SM and DHSM, and GlcCer and DHGlcCer, respectively. After sonicating the single-phase mixture, 100 mL of chloroform:methanol:5N NaOH (1:2:0.8, v/v/v) was added and the solution was incubated for 1 h at 37 °C, followed by neutralisation with acetic acid. Subsequently, 158 mL of chloroform and 158 mL of water were added and the mixture was then vigorously shaken using a vortex, before centrifuging for 1 min at 13,000 × g at 4 °C. The lower phase was withdrawn and dried and then resuspended in acetonitrile:methanol (1:1, v/v), sonicated for 10 s, centrifuged at 14,000 × g for 5 min, and the supernatant was transferred to vials. The concentrations of Cer, DHCer, GlcCer, DHGlcCer, SM, and DHSM were analysed using the QTRAP4500 instrument (SCIEX, Framingham, MA, USA). Sphingolipids containing C16:0, C18:0, C20:0, C22:0, C24:0, and C24:1 fatty acids were detected using a multiple reaction monitoring method, as described in Supplementary Table 1.

Analysis of cell-surface expression of ACE2
293FT/ACE2/TMPRSS2/DSP1-7 cells in 24-well plates were treated with the indicated concentrations of the compounds for 2 days. Cells were harvested by Cellstripper (Corning) and xed with 4% formaldehyde before incubation with either anti-ACE2 antibody (Proteintech) or mouse IgG1 isotype control (BioLegend) antibody at 4 °C. Cells were then stained with PE-conjugated secondary antibody (Thermo Fisher Scienti c) and analysed using the FACSCanto II instrument (BD Biosciences, San Jose, CA, USA). Data were analysed with FlowJo software (Tree Star Inc., San Carlos, CA, USA).
S protein stimulation of 293FT/SARS-CoV-2 S/DSP8-11 cells and isolation of lipid raft domains Target cells (293FT/ACE2/TMPRSS2/DSP1-7), S protein expressing cells (293FT/SARS-CoV-2 S/DSP8-11), and cells not expressing S protein (293FT/DSP8-11) were seeded at 7 × 10 5 cells per 10-cm-diameter dish. On the following day, the medium of the target cells was exchanged with D-MEM containing 10% FCS and 5 mM 4-HPR, and the cells were incubated for an additional 2 days. Cells were then washed with PBS and detached by Cellstripper (Corning) before resuspending in serum free D-MEM and centrifuging at 500 × g for 2 min. After aspirating the suspension, the cells were resuspended in serum-free D-MEM containing 1% Nutridoma SP (Roche). The target cells were mixed with 293FT/SARS-CoV-2 S/DSP8-11 (presence of SARS-CoV-2 S protein stimulation) or 293FT/DSP8-11 (absence of SARS-CoV-2 S protein stimulation) cells, incubated for 30 min, and then washed with cold PBS. Subsequently, sucrose gradient analysis for ACE2 was performed as described previously 55 with slight modi cations. The pellets were homogenised in 2 mL of TNE buffer (25 mM Tris-HCl, pH 7.5, 150 mM NaCl, and 1 mM EDTA) with 0.1% TritonX-100. The sucrose content of the homogenate was then adjusted to 40% by adding 80% sucrose. A linear sucrose gradient (5-30%) in 7 mL of TNE buffer without TritonX-100 was layered over the lysate. The gradients were centrifuged for 17 h at 200,000 × g at 4 °C using a P40ST rotor (Hitachi Koki Co., Ltd, Tokyo, Japan). Eleven fractions were collected from the top of the gradient, followed by immunoblot analysis using anti-ACE2, anti-transferrin receptor (non-raft marker), and anti-otillin 1 (raft marker) antibodies (all from Abcam).

Analysis of cellular membrane uidity
Analysis of membrane uidity was performed as described previously 56 with slight modi cations. Brie y, 293FT/ACE2/TMPRSS2/DSP1-7 cells in 24-well plates were treated with the indicated concentrations of the compounds for 2 days. Cells were detached by Cellstripper (Corning), resuspended in PBS, and centrifuged at 500 × g for 2 min. The suspension was aspirated and the cells were labelled with a solution of 15 mM pyrenedecanoic acid and 0.08% F-127 in perfusion buffer, in the dark for 20 min with rocking at 25 °C. Excess probe was removed and the cells were washed with PBS. The cells were resuspended in 450 mL PBS; 200 mL was aliquoted into a 96-well plate and uorescence was measured using an excitation wavelength of 350 nm and two emission wavelengths: 400 nm for the monomer and 470 nm for the excimer, using a SpectraMax i3x Microplate reader (Molecular Devices). The ratio of excimer to monomer uorescence is related to cellular membrane uidity, with a higher ratio re ecting greater uidity.

Statistical analysis
All statistical analyses were performed with GraphPad PRISM 6 (GraphPad Software, San Diego, CA, USA).