key: cord-0820272-wrqtjc0t
authors: Mesquita, Francisco S.; Abrami, Laurence; Sergeeva, Oksana; Turelli, Priscilla; Kunz, Béatrice; Raclot, Charlène; Montoya, Jonathan Paz; Abriata, Luciano A.; Peraro, Matteo Dal; Trono, Didier; D’Angelo, Giovanni; van der Goot, F. Gisou
title: S-acylation controls SARS-Cov-2 membrane lipid organization and enhances infectivity
date: 2021-03-15
journal: bioRxiv
DOI: 10.1101/2021.03.14.435299
sha: 021cd9318b74a2b61788a04f7c6f3da88adde21f
doc_id: 820272
cord_uid: wrqtjc0t

SARS-CoV-2 virions are surrounded by a lipid bilayer which contains membrane proteins such as Spike, responsible for target-cell binding and virus fusion, the envelope protein E and the accessory protein Orf3a. Here, we show that during SARS-CoV-2 infection, all three proteins become lipid modified, through action of the S-acyltransferase ZDHHC20. Particularly striking is the rapid acylation of Spike on 10 cytosolic cysteines within the ER and Golgi. Using a combination of computational, lipidomics and biochemical approaches, we show that this massive lipidation controls Spike biogenesis and degradation, and drives the formation of localized ordered cholesterol and sphingolipid rich lipid nanodomains, in the early Golgi where viral budding occurs. ZDHHC20-mediated acylation allows the formation of viruses with enhanced fusion capacity and overall infectivity. Our study points towards S-acylating enzymes and lipid biosynthesis enzymes as novel therapeutic anti-viral targets.

β-Coronaviruses (CoVs) belong to a family of large, positive-stranded RNA, enveloped viruses capable of infecting a wide range of hosts and with great zoonotic potential (V'kovski et al., 2020) . Human-tropic species can cause from mild sub-clinical upper respiratory symptoms (i.e. common colds) to a life-threatening form of atypical pneumonia termed severe acute respiratory syndrome (SARS) (V'kovski et al., 2020; Wang et al., 2020) , such as COVID-19 caused by SARS-CoV-2 (SARS coronavirus type 2) (Bojkova et al., 2020; Gordon et al., 2020) . Despite the impressive advances in understanding SARS-CoV-2 over the last year, a precise molecular understanding of the biogenesis of the virus and of the factors that govern infectivity is still lacking.

CoVs are surrounded by a membrane in which at least 3 structural proteins are embedded: Spike (S), Membrane (M), and Envelope (E) (Boson et al., 2021; Siu et al., 2008; V'kovski et al., 2020) as well as accessory factors such as Orf3a (Ito et al., 2005; Tan et al., 2004) . Viral assembly requires E and M (Boson et al., 2021; Siu et al., 2008) but not Spike (Siu et al., 2008) . Spike is however essential for infectivity as it mediates attachment to the host cell receptors, angiotensin-converting enzyme 2 (ACE2) for SARS-CoV-2 (Hoffmann et al., 2020b; Letko et al., 2020) , and triggers the fusion between viral and target cell membranes, to release the viral genome into the cytoplasm. It is therefore the target of most vaccine strategies (Dai and Gao, 2020; Hoffmann et al., 2020a) .

Spike, E and M are synthesized by ribosomes associated to the ER, where folding and quaternary assembly occurs, such as trimerization of Spike. They may also undergo post-translation modifications such as N-glycosylation on their luminal domains (Watanabe et al., 2020) and S-acylation (commonly referred to as S-palmitoylation) on their cytosolic domains. Both S and E have been reported to undergo S-acylation in the case of SARS-CoV-1, mouse hepatitis virus (MHV) and other CoVs (Boscarino et al., 2008; Lopez et al., 2008; McBride and Machamer, 2010; Petit et al., 2007; Thorp et al., 2006; Tseng et al., 2014) . S-acylation, which consists in the covalent attachment of medium chain fatty acids (frequently palmitate, C16) to the sulphur atom of cytosolic cysteines, is very frequent in mammalian cells and considered a hallmark of viral envelope proteins Schmidt, 1982) . S-acylation of Spike has been proposed to mediate its association to lipid microdomains, promote syncytia formation upon transfection into cells and reduce infectivity of MHV and of pseudotyped particles (McBride and Machamer, 2010; Nguyen et al., 2020; Sanders et al., 2020; Thorp et al., 2006) .

Here, we present a comprehensive study of the role of S-acylation in the SARS-CoV-2 infection cycle, including the detailed characterisation of Spike acylation, the identification of additional acylated viral proteins and host enzymes involved, and the importance of acylation in viral biogenesis and infection. We identify ZDHHC20 as the main acyltransferase responsible for the lipidation of SARS-CoV-2 Spike, E and Orf3a.

We demonstrated that Spike acylation plays several roles, promoting biogenesis by protecting from premature targeting to ER degradation pathway, increasing the Spike half-life, and controlling the lipid composition of its immediate membrane environment.

Lipidomic analyses reveal the unique lipid composition of SARS-CoV-2 virions, enriched in specific early Golgi sphingolipids, quite distinct from that of viruses budding from the plasma membrane. Finally, we show that S-acylation and sphingolipid metabolism promote SARS-CoV-2 infection, pointing towards S-acylation and lipid biogenesis related enzymes as novel potential anti-viral targets.

identify the acyltransferases responsible for the modification of Spike, we individually depleted the 23 ZDHHC enzymes by RNA interference in Hela cells (ZDHHC1 to 24, ZDHHC10 does not exist). Downregulation of three family members, ZDHHC8, 9 and 20, led to a >50% drop in 3 H-palmitate incorporation into ectopically expressed Spike ( Figure 1E , S1C). A subset screen, carried in near-haploid HAP-1 cells knocked out for individual ZDHHCs, confirmed these findings and indicated a principal role for ZDHHC20 ( Figure 1F , S1D). Spike acylation upon depletion of the individual ZDHHC8, 9 or 20 could be restore by expression of the correspondent siRNA-resistant myctagged ZDHHC enzyme, but not a random acyltransferase ( Figure S1E ). In line with our findings in HAP1 cells, ectopic expression of ZDHHC20 could also fully rescue Spike acylation upon depletion of ZDHHC8 or ZDHHC9, further supporting a prominent role for ZDHHC20 ( Figure S1E ). We next tested the involvement of the identified ZDHHC enzymes during infection in Vero E6 cells, using monkey specific siRNAs. Expression of ZDHHCs varies significantly between cell types (Figure S1G), with ZDHHC20 being more abundant in Vero E6 cells when compared to the human cell lines used in this study ( Figure S1D ).

We first targeted ZDHHC8, 9, and 20 using an siRNA mix (MixT). Acylated Spike could no longer be retrieved using the Acylrac capture assay ( Figure S1E ). We then silenced the enzymes individually, and included additional ZDHHCs for comparison ( Figure 7G and S7G). Unfortunately, efficient depletion (below 50%) of monkey ZDHHC8 also led to a decrease in monkey ZDHHC20 mRNA in Vero cells, an effect we did not observed in human cell lines ( Figure S1H ). We therefore omitted ZDHHC8 from further analysis in Vero cells, but maintained the use of MixT. Knocking down ZDHHC9 and 20 led to a strong decrease in S-acylation of Spike during infection ( Figure 1G ). Since Spike is not the only acylated SARS-CoV-2 protein, we tested M, E and Orf3a, which all have cytosolic cysteines ( Figure S1I ) (Boscarino et al., 2008; Issa et al., 2020, p. 3) . Acylrac analysis indicated that all three proteins are acylated during infection. Interestingly, E and Orf3A, but not M, could no longer be retrieved when cells were pre-treated with siMixT or siZDHHC20 ( Figure 1H ). Altogether, these observations show that Spike is a substrate for ZDHHC9 and ZDHHC20 (monkey and human), and ZDHHC8 (human), and that ZDHHC20 also modifies E and Orf3a.

Coronaviridae Spike proteins contain a conserved cysteine-rich stretch of amino acids in their cytosolic tail ( Figure 2A ) (Gelhaus et al., 2014; Thorp et al., 2006) . SARS-CoV-2 Spike has 10 cysteines within its first 20 cytosolic amino acids ( Figure 2A ) and was recently highlighted as one of the most cysteine-rich proteins encoded by animal viruses (Sanders et al., 2020) . To study the relative importance of these residues, we generated four mutants changing groups of 2-3 adjacent cysteines to alanine ( Figure   2A ). In transfected cells, mutation of the two most membrane-proximal cysteines (group I) led to an 80% drop in 3 H-palmitate incorporation into the Spike protein ( Figure   2BC ), as previously observed for SARS-CoV-1 Spike (Petit et al., 2007) . Mutation of the next three cysteines (group II) also had a major, albeit, lower effect with a ≈40% drop in 3 H-palmitate incorporation ( Figure 2BC ). Mutating cysteines from both groups I and II had an impact almost equivalent to that of changing all 10 to alanine ( Figure   2BC ). Yet, cysteines from groups III and IV also undergo S-acylation, since mutating both groups alone or in conjunction with group II still significantly reduced 3 H-palmitate incorporation ( Figure 2BC ). Altogether this analysis indicates that all 10 cysteines are targets of S-acylation and lipidation of those closest to the membrane is required for the others to be modified, possibly because it brings the more distal cysteines closer to the membrane, and therefore more accessible to the ZDHHC active site.

To gain more insight into the dynamics of this process, we measured 3 H-palmitate incorporation kinetics in transfected HeLa cells. Incorporation of 3 H-palmitate into WT Spike was rapid, reaching a plateau within about 3 h, in great contrast with group Iand group II-mutated variants, where it was very slow and limited ( Figure 2D and S2A).

Mutation of groups III and IV cysteines had no major effect on S-palmitoylation kinetics but resulted in lower final levels of incorporation, consistent with the removal of 3, or 2 acceptor amino acids, respectively ( Figure 2D ). Thus, membrane proximal cysteines play a more influential role in Spike S-acylation than their more distal counterparts.

We also evaluated 3 H-palmitate incorporation into Spike during a single-round of infection in Vero E6 cells. We used a higher MOI for infection (≈0.5 to 1) and added radioactive 3 H-palmitate 4 h post-viral inoculation, when Spike started being detected by flow cytometry and significant viral RNA replication was observed by QPCR ( Figure   S2A -C). Remarkably, the kinetics of Spike 3 H-palmitate incorporation were essentially identical for infected and transfected cells ( Figure 2EF and S2D).

We next investigated where during viral biogenesis Spike acylation occurs. After synthesis, folding and assembly, Spike trimers are transported to the ER-Golgi intermediate compartment (ERGIC) where viral assembly and budding take place (de Haan and Rottier, 2005; Klein et al., 2020; Stertz et al., 2007) . Assembled virions are subsequently transported to the plasma membrane to be released through exocytic mechanisms involving the Golgi apparatus and/or lysosomal compartments (Ghosh et al., 2020) . To determine if Spike acylation occurs during protein biogenesis, and/or after transport from the ER to the Golgi, we monitored 3 H-palmitate incorporation under the effects of the protein synthesis inhibitor cycloheximide (CHX), and the Golgi-disrupting drug Brefeldin A. Incorporation into WT Spike was decreased by ≈50% when cells were treated with either CHX or Brefeldin A ( Figure 2G ), suggesting that S-palmitoylation is initiated shortly after protein synthesis and continues once the protein has exited the ER. This is consistent with previous reports on SARS-CoV-1 and MHV Spike proteins indicating that their palmitoylated versions were present in both pre-and post-Golgi compartments (McBride and Machamer, 2010; van Berlo et al., 1987) . We next determined the effect of the drugs on the S-palmitoylation of two additional mutants, bearing respectively only the first 2 (groups II, III and IV mutated) or 5 (groups III and IV mutated) most membrane-proximal cysteines ( Figure 2G ). With the former mutant Spike, sensitivity to brefeldin A was lost, suggesting that acylation of the 2 juxta-membrane cysteines occurs in the ER, whereas the latter had a profile similar to that of WT indicating that more distal cysteines are modified in downstream compartments.

We also monitored the kinetics of 3 H-palmitate incorporation when silencing the Spikemodifying ZDHHC enzymes. ZDHHC20 has a broad distribution throughout the entire biosynthetic pathway from the ER to the plasma membrane ( Figure S2E ). ZDHHC9 expression appears to be more restricted to the ER, while ZDHHC8 is more concentrated in the perinuclear-Golgi region ( Figure S2E ). Silencing of any of the three enzymes had a drastic effect on 3 H-palmitate incorporation, but silencing ZDHHC20 caused the earliest and most dramatic effect ( Figure 2H ), phenocopying the group I cysteine mutant ( Figure 2D ).

Altogether, these results indicate that following synthesis, Spike can undergo acylation on its two most juxtamembranous cysteines in the ER by ZDHHC20, a step required for the modification of the more distal cysteines by ZDHHC20, 8 or 9 in the ER and Golgi.

The above experiments, whether by Acylrac or 3 H-palmitate incorporation, demonstrate that acylation occurs but do not indicate whether only a small percentage or all Spike molecules undergo the modification. A variant of Acylrac has therefore been developed where after removal of the acyl chains using hydroxylamine, the newly freed up cysteines are labelled with PEG (5 kDa), leading to molecular weight shifts (Howie et al., 2014; Percher et al., 2017) . For small proteins, each PEGylation reaction leads to an approximate 5 kDa shift in the SDS-PAGE migration pattern. When proteins are large, and the number of acylation sites is important, molecular weight shifts are less predictable. When Spike was transfected into Vero E 6 cells, we observed a weak high molecular weight smear, but the full length or S2 Spike bands remained essentially unaltered ( Figure 3A , left panel). In contrast, PEGylation of proteins from infected cells led to the disappearance of the S2 band ( Figure 3A , right panel and 3B). It is unclear where exactly the S2 band shifted to. It could be the band migrating slightly higher than the full-length form. Extensive PEGylation may also hinder the full-length Spike from entering into the SDS gels. The disappearance of the S2 band nevertheless indicates that all cleaved Spike proteins were highly acylated during infection. A kinetic analysis of PEGylation indicated that Spike S-acylation occurred as early as 2 h after viral inoculation ( Figure S3A ), consistent with the rapid dynamics observed in our previous 3 H-palmitate incorporation experiments ( Figure 2) .

We also performed PEGylation on the virion containing supernatants of infected Vero E6 cells and on isolated virions, observing similar shifts ( Figure 3CD ). The PEG induced band shifts in S2 were abolished when ZDHHC enzymes were silenced with MixT or ZDHHC20 siRNA ( Figure 3B -D and S3B). Correspondingly capturing of acylated Spike from supernatants of infected cells was also abrogated when cells were silenced with MixT ( Figure S3C ). The PEGylation analysis, combined with the 3 H-palmitate incorporation experiments, indicate that the entire Spike population produced by infected cells undergoes massive and rapid acylation, apparently on all 10 cysteines.

S-acylation can be reversed by the action of deacylating enzymes, the acyl protein thioesterases (APTs) (Zaballa and Goot, 2018) . Using a metabolic 3 H-palmitate pulseand-chase approach in transfected HeLa cells, we found that WT Spike lost its palmitate with a half-live of about 3 h, as did group III and IV mutants ( Figure 3E and S3D), and that this process could be blocked using the general APT inhibitor Palmostatin B ( Figure S3E ). These data prompted us to test if S-acylation of Spike is also dynamic in infected cells.

We performed a similar 3 H-palmitate decay analysis during SARS-CoV-2 infection of Vero E6 cells. 3 H-Palmitate was added 14 h post-inoculation for 2 h, and cells were further incubated in label-free medium. While depalmitoylation of cellular proteins (detected by autoradiography of total cell extracts) was readily observed ( Figure S3F and 3G), there was no significant loss of palmitate from Spike ( Figure 3FG ). This observation suggests that Spike deacylation barely occurs during infection of Vero E6 cells. The difference with depalmitoylation of Spike during transfection could be due to rapid segregation of Spike from the acyl thioesterases during formation of virions in the early Golgi, while the Spike cysteines remain cytosolically exposed to acyl thioesterases in transfected cells (model in Fig. 3H ).

S-acylation has been reported to modify the turnover rate of proteins (Zaballa and Goot, 2018) . We therefore tested whether this was the case for Spike. We used two different methods, 35 S Cys/Met metabolic pulse-chase experiments, which monitor the evolution of a protein starting from its synthesis, and cycloheximide chase, which monitors the degradation of fully folded mature proteins. The 35 S Cys/Met pulse period was chosen based on the plateau of palmitate incorporation, i.e. 4 h ( Figure 2D ). The decay in the 35 S signal observed for WT Spike indicated an apparent half-life of ≈4 h, while that observed for the 10C-A cysteine-less mutant was ≈1 h ( Figure 3I and S3G).

The accelerated early decay for the 10C-A mutant suggests that this derivative is rapidly targeted to degradation in the ER, whereas acylation protects the WT protein from this premature process. The targeting to degradation is indicative of inefficient folding, recognized by the ER quality control machinery. A previous study did not reveal an effect of SARS-Cov-1 Spike S-acylation on protein degradation (McBride and Machamer, 2010) . In these experiments, the 35 S Cys/Met metabolic labelling time was short, 20 min, but we know now this is insufficient to significantly populate the acylated Spike species ( Figure 2D ) (Abrami et al., 2019) . Acceleration of Spike degradation following its synthesis was also observed when silencing the ZDHHC enzymes ( Figure 3J ). Silencing ZDHHC9 or 20 was comparable to siMixT or mutating all cysteines, while silencing of ZDHHC8 had a less pronounced effect ( Figure 3K ).

Comparison of the total levels 35 S Cys/Met-labelled Spike at 20 h of chase showed that acylation, primary by ZDHHC20, leads to a 3-fold increase in total Spike ( Figure   3K ). Thus, acylation is crucial for efficient Spike biogenesis. In addition, through the cyloheximide chase approach, we found that S-acylation also slows down the turnover of mature, fully folded, Spike ( Figure 3L and S3H). Therefore, by increasing the flux of newly synthesized Spike through the ER quality control gateway and by slowing down the turnover of the fully folded protein, Sacylation dually increases the levels of Spike available for virus assembly.

Following synthesis, Spike assembles into trimers (Delmas and Laude, 1990) , which Given this very significant contribution, we analyzed the possible impact of Spike acylation on the surrounding lipid organization, using coarse-grained (CG) molecular dynamics (MD) simulations. We compared the behavior of palmitoylated and nonpalmitoylated Spike in a model membrane. We set up a membrane composed of 50% dipalmitoylphosphatidylcholine (DPPC), 30% dilinoleylphosphatidylcholine (DLiPC) and 20% cholesterol (CHOL), and parameterized it using the MARTINI CG force field, which is known to capture lipid nanodomain formation (Lorent et al., 2017; Risselada and Marrink, 2008) . We first reproduced the results of Lorent et al., 2017 of microdomain formation (a liquid-disordered DLiPC-rich phase and a liquid-ordered DPPC+CHOL-rich phase) within ≈1 µs when starting from randomly distributed lipids.

We subsequently prepared two systems, again with lipids distributed randomly, and introduced the transmembrane domain (TMD) of Spike followed by the cytosolic tail residues (residues 1201 to the end, Uniprot P0DTC2), either fully S-palmitoylated or not at all. We ran each system 10 times for ≈3.7-3.8 µs in CG MD simulations, starting from different random seeds ( Figure 4A ). Analysis of the MD simulations led to two predictions. While the cytoplasmic tails of the protomers were initially introduced in an extended form, acylation of all cysteine residues led to the collapse of the cytosolic tails, bringing the palmitate moieties in close proximity of the TMD and, leading to a ≈9-fold local increase in the acyl chain concentration ( Figure 4AB ). Such compaction is not an artefact of the coarse-grained description of the protein, as it was also observed in atomistic MD simulations using the CHARMM36m force field ( Figure 4B ).

Spike preferentially associated with the disordered phase when unmodified, but rather partitioned into the ordered phase when S-palmitoylated. This preference was dictated by the cytosolic half of the TMD ( Figure 4AB and S4A). In contrast, the luminal half of the TMD interacted preferentially with the disordered phase, irrespective of Spike acylation ( Figure 4B and S4A). These simulations raised the interesting possibility, not tested here, that acylated Spike could drive trans-bilayer lipid asymmetry.

Given the abundance of Spike trimers in viruses, and the predicted compaction of the saturated acyl chains around the TMD, the local lipid organization of Spike is expected to be ordered. Consistently, it was previously observed that transfected Spike associates with detergent resistant membranes (DRMs), as we could confirm ( Figure   S4BC ). DRMs, of which caveolin-1 is a marker, have been considered a useful biochemical readout for the association of proteins with cholesterol-rich lipid domains, even though not equivalent to the ordered lipid domains in cells .

Such domains are predominantly found in the plasma membrane, the endosomal system and the late Golgi (van Meer et al., 2008) . The ER and the early Golgi are presumed to be devoid of such domains since their cholesterol and sphingolipid contents are very low. Therefore the ERGIC, the intermediate compartment between the ER and the Golgi, into which CoVs bud (Klein et al., 2020) , is not expected to contain cholesterol-rich lipid domains.

To probe the lipid environment of Spike, we first tested whether transfected intracellular Spike could be found in DRMs. For this, we biotinylated the surface proteins of transfected cells, prepared DRMs and performed streptavidin pull downs on each fraction to separate plasma membrane (bound) from intracellular (unbound) proteins. Remarkably, intracellular WT Spike was exclusively found in the DRM fractions, while cell surface WT Spike was distributed throughout the gradient ( Figure   4CD ). The intracellular 10C-A mutant in contrast was detergent-soluble ( Figure S4DE ) and also less expressed at the cell surface ( Figure S4FG ). We next analyzed the distribution of cholesterol in Spike-expressing cells using the cholesterol-binding fungal metabolite, filipin. Filipin staining was more than 2-fold higher in WT than in acylation-deficient 10C-A Spike-expressing cells ( Figure 4EF ). In addition, staining was atypically concentrated in the perinuclear area ( Figure 4F ). Note that as previously observed, expression of WT, but not 10C-A Spike, promoted cellto-cell fusion leading to the formation of large multi-nucleated syncytia (McBride and Machamer, 2010) ( Figure 4F and S4I).

We next performed the DRM analysis on SARS-CoV-2 infected cells. Spike was then exclusively found in DRMs ( Figure 4GH ), without needing to separate intracellular from surface expressed proteins. Spike-DRM association appeared mostly intracellular as indicated by its resistance to cholesterol extraction from the plasma membrane using extracellular addition of ß-methylcyclodextrin (ßMCD). Partitioning of the viral M-protein also remain unaltered, while ßMCD treatment did solubilize caveolin-1 which is predominantly found at the plasma membrane ( Figure 4G -I). The DRM association of Spike was lost when cells were treated with the MixT siRNA pool or ZDHHC20 siRNA ( Figure 4J -L and S4I). In contrast, silencing of ZDHHC9 had no effect ( Figure 4L and S4I). Thus ZDHHC20-mediated acylation of Spike in SARS-CoV-2 infected cells leads to it association with intracellular DRMs, presumably derived from the ERGIC, as well as from the newly formed virions on their way to the extracellular space ( Figure 3H ).

As mentioned, the ERGIC is not expected to have cholesterol and sphingolipid rich ordered domains. The presence of acylated Spike in DRMs in infected cells therefore raised the possibility that viral proteins drive the formation of lipid domains in the ERGIC. To test this hypothesis, we monitored the behavior of ERGIC53, a transmembrane protein that accumulates in the ERGIC. In uninfected cells, ERGIC53 was detergent soluble ( Figure S4J ) as expected. In contrast, ERGIC53 was mostly detected in DRMs when cells were infected with SARS-CoV-2 ( Figure 4JM ). This was due to protein acylation since ERGIC53 was detergent soluble in infected cells treated with the siRNA pool MixT, or siZDHHC20 but not siZDHHC9 ( Figure 4JM and S4I).

Altogether this analysis suggests that ZDHHC20-mediated acylation of Spike in infected cells drives the formation of ordered domains in the ERGIC.

We next investigated whether S-acylation of Spike influences its lipid environment in viral particles. We generated and purified Spike-pseudotyped viral-like particles (VLPs) from HEK293T cells, using an HIV-derived lentivector-based system. First, we verified that Spike present in VLPs is acylated ( Figure S5A ). Depletion of individual ZDHHC enzymes (8, 9 and 20) or all three combined (siMixT) did not significantly affect VLP production (measured by ELISA against the HIV antigen p24), incorporation of Spike into them, nor the ratio between cleaved S2 fragment and fulllength Spike ( Figure S5B -E). ZDHHC silencing however drasticall affected the ability of Spike to co-fractionate with DRMs ( Figure 5A -C). More specifically, DRM association of Spike was lost when silencing ZDHHC20, reduced when silencing ZDHHC8, and unaltered when silencing ZDHHC9 ( Figure 5C and Figure S5F ).

We next generated VLPs containing either WT or 10C-A Spike, or made non-typed (bald) viruses ( Figure S5I ). Both forms of Spike could be efficiently detected in HEK293T cells, with a slight increase for the 10C-A mutant ( Figure 5D and S5G), and displayed an equivalent ratio of cleaved vs. full length protein ( Figure S5H ). In contrast, Spike incorporation ( Figure 5EF ) and the ratio of cleaved vs. full length ( Figure 5EG ) was reduced by more than 30% in 10C-A containing VLPs. Such a decrease in Spike incorporation was not observed when ZDHHC expression was modified using siRNAs, likely because the silencing is incomplete and as observed acylation is highly efficient.

As in MixT-treated cells, 10C-A Spike did not partition into DRMs within VLPs ( Figure   5GH ). Interestingly, the HIV p17 matrix protein also associated with DRMs when acylated Spike was present ( Figure 5GI ). Thus, as observed with ERGIC53, acylation of Spike influences its lipid environment in VLPs.

To test whether this is a local effect around Spike or a broader effect on the lipid composition of the entire VLP membrane, we performed liquid chromatography-mass spectrometry (LC-MS)-based shotgun lipidomics of purified VLPs and their producing HEK293T cells. Comparison of the relative lipid levels indicated an enrichment of sphingolipids in the VLPs, irrespective of the presence or S-acylation of Spike ( Figure   5J ). Thus, S-acylation of Spike induced localized formation of ordered lipid microdomains restricted to the local environment of Spike in VLPs ( Figure 5K ).

We next extended our biochemical analyses to SARS-CoV-2 virions. Virions were purified from infected Vero E6 cells (48 h post inoculation) silenced for acyltransferase expression using the MixT pool. Similar levels of Spike and M in virions and cells were observed irrespective of acylation, with a small, yet, significant decrease of the Spike to M ratio in siMixT-derived virions ( Figure 6AB ). The relative proportion of the S2 fragment compared to the total amount of Spike was also largely unaltered by siMixT ( Figure 6C ). Well defined viruses were observed by cryo-electron microscopy under both conditions, with clearly visible Spikes complexes in the characteristic prefusion conformation ( Figure 6D ). The virions obtained from siMixT-treated cells, however, appeared less homogenous in size and slightly but significantly larger on average ( Figure 6DE ). We modelled the tails of the distribution of virion diameters using a Generalized Pareto Distribution, which enables the analysis of the exceedances of a particular dataset. The shape factor (ξ) for virions derived from control cells was negative, of -0.20 ±0.19, indicating a low propensity to generate extreme diameters ( Figure 6E and S2A). In contrast, for virions derived from siMixT-treated cells, ξ was 0.64 ±0.34, clearly showing an increased frequency of extreme diameters ( Figure 6E and S2B). Thus, ZDHHC9 and 20 contribute to the tight control of viral particle dimensions.

We next analyzed the DRM association of Spike within SARS-CoV-2 envelopes.

Virions were submitted to detergent solubilization in the cold. Note that for both VLPs and virions, the detergent to membrane ratio is higher than when solubilizing cells, due to the low amount of material, which generates a high risk of over-solubilization.

Nevertheless, Spike was co-fractionated in DRMs in a MixT dependent manner, as observed for VLPs ( Figure 6FG ). We next determined the lipid composition of SARS-CoV-2 virions and compared it to that of parental cells. We found that virions were deprived of non-membrane lipids (triglycerides >80% less represented), poor in glycerolipids (25% less represented), and enriched in sphingolipids (40% more represented) ( Figure 6H and S6C-F). Virions were also enriched in phosphatidic acid which displays a small head-group and a bulky acyl tail that contributes to negative membrane curvature and may facilitate intraluminal vesicle budding (Egea-Jimenez and Zimmermann, 2018; Zhukovsky et al., 2019). More strikingly, hexosylceramide (HexCer) species (most likely glucosylceramide-GlcCer) were found to be on average 330% more represented in virions than in parental cells ( Figure 6H and S6C). GlcCer, the precursor of glycosphingolipids, is produced on the cytosolic side of early Golgi cisternae where GlcCer synthase resides (Halter et al., 2007) . Once produced, GlcCer is translocated to the lumenal leaflet of either Golgi cisternae or, following non-vesicular transport, of the TGN (D' Angelo et al., 2013 Angelo et al., , 2007 . In the luminal leaflet of the Golgi/TGN membrane, GlcCer is readily converted to lactosylceramide and complex glycosphingolipids. Thus, GlcCer primarily populates the early segment of the biosynthetic pathway. It is widely accepted that virions retain/select a lipid composition for their envelope that is characteristic of their budding site, and this has been used as a mean to determine the lipid composition of the plasma membrane of polarized cells for example (Heaton and Randall, 2011; Ketter and Randall, 2019) . Our analysis therefore provides the first detailed insight into the lipid composition of the SARS-CoV-2 budding site and shows biochemically that these viruses bud in the early secretory pathway, consistent with morphological studies (Klein et al., 2020; Stertz et al., 2007) .

Although the preparations of VLPs and SARS-CoV-2 virions were generated from different cells, HEK293 and Vero E6 respectively, we did attempt comparing their lipid composition. While this comparison is to be taken cum grano salis (with a grain of salt), it indicates that SARS-CoV-2 virions are enriched both in sphingolipids and in GlcCer when compared to VLPs. Consistent with findings on VLPs, we found that the overall lipid composition of the virions is not markedly affected by protein acylation ( Figure 6JK and S6C-F). We also monitored the relative cholesterol amounts in our viral preparations normalized to the M protein. Our analysis revealed that virions derived from siMixT-treated cells contained approximately 30% less cholesterol than control virions ( Figure 6L ), fully consistent with the prediction that Spike S-acylation promotes the formation of ordered lipid nanodomains in the ERGIC ( Figure 4 ).

Thus, S-acylation does not drastically affect the overall lipid composition of SARS-CoV-2 viruses but instead modifies the lipid environment in the vicinity of Spike promoting the formation of cholesterol-rich lipid domains (model in Figure 3H ). This local effect is consistent with the fact that Spike trimers are rather sparsely distributed within the viral membrane as shown by cryo-tomography studies (Ke et al., 2020; Klein et al., 2020; Yao et al., 2020) .

Having characterized the biochemical effects of S-acylation on the lipid composition and organization in VLPs and SARS-CoV-2 virions, we set up to address how Sacylation affects infection. We used three complimentary approaches. First, to focus specifically on the role of Spike, we used the Spike-pseudotyped VLP system and monitored the receptor binding and fusion/entry steps. Secondly, we studied the impact of S-acylation on full SARS-CoV-2 virus by: on one hand, monitoring how wildtype viruses can infect cells depleted of specific ZDHHC enzymes; on the other, how viruses derived from siZDHHC-silenced cells can infect naïve control cells, the latter addressing the infectivity of virus produced with altered protein acylation.

To analyze binding to the ACE-2 receptor, we compared the competitive binding between Spike pseudotyped VLPs (WT or 10C-A) and purified Spike receptor binding domain (RBD), using a modified human ACE2 ELISA kits. Both WT and 10-CA pseudotyped VLPs displayed equivalent levels of competition with purified RBD, whereas bald VLPs did not display significant binding ( Figure 7A ). Next, we performed synchronized fusion assays by incubating ACE-2-expressing HeLa cells with VLP supernatants at 4°C, washing unbound particles and allowing subsequent fusion at 37°C for 30 min. Prior to cell harvesting, surface-attached particles were removed using a stringent acid-wash and fused and/or internalized particles were monitored by western blot. While it was possible to detect ACE2-dependent fusion of WT Spike-S2, no Spike 10C-A was detected, indicating that 10C-A-pseudotyped viral particles were unable to fuse with target cells ( Figure 7B and S7A). We also monitor the infectivity of these VLPs by transducing them into ACE2-expressing HEK293 cells and monitoring luciferase expression after 3 days. The infectivity of WT Spike-pseudotyped VLPs was more than 4 times higher than that of the same amount of 10C-A containing VLPs, or bald, non-typed particles ( Figure 7C ). In line with this, Spike-pseudotyped VLPs, but not VSVG-pseudotyped VLPs ( Figure S7B ), produced from cells silenced for ZDHHC20 also displayed lower infectivity when compared to VLPs obtained from control cells ( Figure 7D ). Silencing ZDHHC8 or ZDHHC9 did not significantly compromise the infectivity of VLPs. Thus, ZDHHC20-mediated acylation of Spike drives its association with lipid domains, promotes its incorporation into VLPs, and increases fusion and infectivity of the VLPs. We next analyzed the general importance of S-acylation during infection of the SARS-CoV-2 virus itself. We first chose a 24 h infection at low MOI (≈0.1). We monitored the levels of two viral RNAs, E and RNA-dependent RNA polymerase-RdRp transcripts.

Both dropped by about 75% in siMixT treated infected cells, as well as in the supernatants of these cells, which contain released virions ( Figure 7E) . A similar reduction in viral RNA was observed when ZDHHC20 was silenced, both in the cells and in supernatants ( Figure 7F ). Silencing ZDHHC9 also reduced viral RNA but to a lesser extent ( Figure 7F ).

These experiments indicate that ZDHHC20 and 9 strongly affect the infection cycle of SARS-CoV-2. The time point chosen (24 h) allows for 2 to 3 rounds of infection, i.e. multiple rounds of entry, replication, viral biogenesis and release. To estimate the importance of acylation for the replication and/or infectivity, we analyzed infection at a shorter time, 10 h, but this required the use of a higher MOI (≈1 to 1.5) for detection.

In these conditions, no significant change in viral RNA in either cells or supernatant was detected when silencing ZDHHC20, a moderate decrease was observed in cells depleted for ZDHHC9 ( Figure 7G ). Consistently, the levels of Spike and M proteins in infected lysates and supernatants were also comparable between conditions ( Figure   S7C ). Within the limitations of high MOIs and incomplete loss of ZDHHC expression following siRNA, the observation suggest that ZDHHC9 influences the first round of infection, while ZDHHC20 might not.

Next, we utilized the viral supernatants produced from the 10 h infections in cells silenced for either ZDHHC20 or ZDHHC9, and used them to infect naïve control cells.

Supernatants were first adjusted to comparable amounts of E-RNA copies per ml ( Figure S7D ). Approximately 50 E copies were then added per target cell for infection.

The infectivity was monitored 6 h after viral inoculation by measuring the total RNA of E and RdRP, as well as sub-genomic E RNA (E-sub_gen), which is a more accurate proxy for viral replication. The infectivity of virions produced from either ZDHHC20 and ZDHHC9 silenced cells was 25% to 30% lower than that of control virions, based on total and sub-genomic E RNA, respectively ( Figure 7H) . A similar trend was observed for RdRp RNA. These data indicate that ZDHHC20-mediated S-acylation contributes to SARS-CoV-2 infection by promoting the production of fully infectious virions.

Finally, given the here identified importance of cholesterol and sphingolipid-rich domains, we tested whether pharmacological inhibition of sphingolipid biogenesis, could affect SARS-CoV-2 infection. We used the fungal metabolite Myriocin, which targets serine palmitoyltransferase-1 at early steps of sphingosine biosynthesis and fully depletes sphingolipids in cells in 5 days. We also tested FTY720, or fingolimod, a drug approved for treatment of multiple sclerosis, which while derived from Myriocin potently affects downstream enzymes, ceramide synthases. FTY720 was found to efficiently modulate the intracellular levels of sphingolipids with shorter treatment times (Berdyshev et al., 2009; Lahiri et al., 2009) . Vero E6 cells were either treated with Myriocin for 5 days and subsequently infected with SARS-CoV-2 for 24 h or first infected with SARS-CoV-2 and 1 h after inoculation exposed to FTY720 for the remaining 24 h. Both treatments, Myriocin and FTY720, phenocopied the silencing of the acyltransferases, leading to a drastic reduction of viral RNA both in the cells and in supernatants after 24 h of infection ( Figure 7I ). Thus, disruption of sphingolipid cascades can also be used as target to disrupt cellular infection of SARS-CoV-2.

Membrane proteins of envelope viruses were amongst the first shown to be Sacylated, yet in-dept molecular understanding of this process and its precise consequences has lagged behind. Here, based on a combination of functional genomics, biochemical, kinetic and computational studies, we show that during SARS-CoV-2 biogenesis in infected cells, Spike protein is rapidly and efficiently S-acylated following its synthesis in the ER, primarily through the action of the ZDHHC20 enzyme.

This prevents premature degradation and strongly promotes the biogenesis of Spike, which is subsequently transported to the ERGIC where it arrives with 30 acyl chains decorating each trimer. The presence of these saturated lipids drives the formation of ordered membrane domains around Spike, enriched in GlcCer, sphingomyelin and cholesterol. Using VLPs pseudotyped with WT or acylation deficient Spike, as well as SARS-CoV-2 virions produced from cells silenced for ZDHHC8/9/20 expression, we could show that acylation greatly enhances viral fusion and infectivity. Our findings are fully consistent and complementary to those by Brangwynne and co-workers who recently identified, through a screening approach, that cholesterol present in the membrane of VLPs is critical for Spike-mediated fusion (Sanders et al., 2020) . Our study shows that S-acylation, and the ZDHHC20 enzyme in particular, constitutes a promising drug target for coronavirus infection. Interestingly, ZDHHC20 is also involved in the acylation of the hemagglutinin of Influenza virus and the recent elucidation of its structure indicates that this acyltransferase should be specifically druggable (Rana et al., 2018) . This study sets the ground for a re-emergent interest in the study of protein S-acylation and also lipid biosynthetic pathways as important regulatory mechanisms of infection by coronaviruses and enveloped viruses in general. 

The authors declare no competing interests. Results are mean ± SD, n = 3 independent experiments. p values were obtained by C One-way or G Two-way ANOVA with Dunnet's or Tukey's multiple comparison tests respectively. Figure S4J) and ERGIC53 (M) in each fraction as in C. Results are mean ± SD, n = 3 independent experiments. All p values were obtained by two-way ANOVA with Sidak's multiple comparison test except for E where one-way ANOVA with tukey's multiple comparison test was used. Quantitative inhibition of RBD binding was compared to culture medium and normalized to the mean of WT-pseudotyped VLP values. B. VLP preparations were also used for synchronized fusion assays (see experimental procedures) using HeLa or ACE-2-HeLa cells. Surface attached particles were removed by acid wash and internalized Spike S2 signal quantified by western blot (see Figure S7A ) and expressed as fraction of WT-pseudotyped VLP signal in HeLapACE-2 cells. AB.

Results are mean ± SEM, and each dot represents an independent preparation, n=3. Figure S5C ). Virion supernatants from Control and ZDHHC9 or 20 silenced cells with comparable viral protein content and equivalent number of E copies per ml (see Figure S5D ) were used to re-infect standard Vero E6 cells using 50 viral E copies per cell. Viral infectivity was assessed by quantifying total viral RNA (E, RdRp) and sub genomic viral RNA (E sub gen) 6 h after inoculation.

Results are mean ± SEM and each dot represent the average RNA levels of an individual independent experiment (n≥3) quantified for at least three biological replicates for each sample. 

All resource/reagent requests will be available upon contact to Francisco Mesquita (Francisco.mesquita@epfl.ch).

Cell culture HAP1 cells were obtained from Horizon Discovery, Vero E6 cells were obtained from the American Type Culture Collection (ATCC) and maintained in DMEM medium supplemented with 10% FCS, penicillin and streptomycin; HELA cells were maintained in MEM medium supplemented with 10% FCS and complemented with 2 mM glutamine and 1% of non-essential amino-acids. For lentiviral vector production, 293T cells were cultured in DMEM supplemented with 10% FBS for maintenance and transfection, and in EpiSerf for production. All cells were kept at 37°C in a humidified atmosphere of 5% CO2 and were tested negative for mycoplasma.

The 293T-ACE2 stable cell line was generated by transduction with an ACE2-irespuro encoding lentivector (produced as described below and at a MOI of 0.5) followed by puromycin selection and maintained in DMEM 10% FCS with 1µg/ml puromycin.

All viral stocks were produced and isolated from supernatants of Vero E6 cells, Plasmids expressing human ZDHHC proteins were cloned in pcDNA3.1 with MYC-tag (provided by the Fukata lab).

The ACE2 sequence (from pHAGE2-EF1aInt-ACE2-WT a kind gift from J.D. Bloom) was swapped with the LacZ cassette of the pRRL-PGK-lacZ-ires-puro lentivector to generate the pRRL-ACE2-ires-puro vector.

All oligonucleotides, namely QPCR primers and siRNA oligos used in this study are described in detail in Supplementary table 1 and 2.

Unless otherwise indicated, plasmids were transfected into 293T, Hela or Vero cells for 24 or 48 hours (3 µg/ 9.6cm 2 plate) using Transit-X2RTM (Mirius) transfection reagent. For control transfections, we used an empty pcDNA6.2 plasmid. siRNA for Hela cells or for Vero cells were purchased from Qiagen (see supplementary   table) . For gene silencing, cells were transfected for 72 hours with (15 pmol/ 9.6cm 2 plate) using Transit-X2RTM (Mirius) transfection reagent. Control siRNA and species specific siRNA oligos targeting human or monkey ZDHHCs are described in Supplementary table 1.

All infections for experimental analysis were done using passage 3 SARS-CoV-2 stocks. Vero E6 cells seeded to a confluency of 90 to 100%, were, washed twice in warm serum-free medium and inoculated with the indicated MOI of SARS-CoV-2, diluted in serum free medium (5 ml for T75; 2 ml for T25; 1 ml for 6-well plates). 1 hour after inoculation cells were washed with complete medium and infection allowed to proceed for the indicated time points in DMEM supplemented with 2.5% FCS, penicillin and streptomycin (unless otherwise indicated: 10 ml for T75; 4 ml for T25; 2 ml for 6well plates). Cells cultured in T75/T25 flasks for biochemical analysis, or 6-well plates for qPCR and Flow cytometry experiments, were infected with a MOI of ≈0.1, except for time-course experiments measuring: radioactive 3 H-Palmitic acid -incorporation;decay; and Acyl-pegylation dynamics, in which cells were infected with a MOI of ≈0.5 to 1. For infections involving 3 H-Palmitic acid, medium without serum IM (see below)

was used for all washes, viral inoculation and throughout infection.

For immunoprecipitations, cells were lysed 30 min at 4°C in IP buffer (0.5%NP40, 500

mM Tris-HCl pH 7.4, 20 mM EDTA, 10 mM NaF, 2 mM benzamidine, and a cocktail of protease inhibitors), centrifuged 3 min at 2000 g and supernatants were precleared with protein G-agarose conjugated beads and supernatants were incubated 16 h at 4°C with specified antibodies and beads.

To follow S-acylation, transfected or infected cells were incubated 1 hour in medium without serum IM (Glasgow minimal essential medium buffered with 10 mM Hepes, pH 7.4), followed by 2 hours or indicated hours at 37°C in IM with 200 µCI /ml 3 H palmitic acid, washed with IM and incubated different times at 37°C with complete medium prior immunoprecipitation overnight. Beads were incubated 5 min at 90°C in reducing sample buffer prior to SDS-PAGE. Immunoprecipitates were split into two, run on 4-20% gels and analyzed either by autoradiography ( 3 H-palmitate) after fixation (25% isopropanol, 65% H2O, 10% acetic acid), gels were incubated 30 min in enhancer Amplify NAMP100, and dried; or Western blotting. Autoradiograms and western blotting were quantified using the Typhoon Imager (Image QuantTool, GE healthcare).

For metabolic labeling, transfected cells were washed with methionine /cysteine free medium, incubated 4hours pulse at 37°C with 50 µCi/ml 35 S-methionine/cysteine, washed and further incubated for different times at 37°C in complete medium with a 10-fold excess of non-radioactive methionine and cysteine. Proteins were immunoprecipitated and analyzed by SDS-PAGE. Autoradiography and western blotting were quantified using the Typhoon Imager (Image QuantTool, GE healthcare).

Protein synthesis was blocked by 1 hour or different indicated time treatment with 10 µg/ml cycloheximide at 37°C in serum free medium. To block protein transport from ER to Golgi, we used Brefeldin A at 5 µg/ml for 1 hour at 37°C in serum free medium.

To sequester cholesterol, beta-metyl-cyclodextrin was used at 10 mM for 30 min in serum free medium. As inhibitor of depalmitoylation, Palmostatin B was used at 10 µM for 4 hours in serum free medium. To block serine palmitoyltransferase, denovo sphingolipid biosynthesis, cells were cultured either in Myriocin at 2.5 µM, 5 days prior infection and during the course of infection, or with a synthetic myriocin analogue,

Fingolimod-FTY20 at 2 µM or 4 µM during infection.

Protein S-palmitoylation was assessed by the Acyl-RAC assay as previously described (Werno and Chamberlain, 2015) 

This PEG switch assay is used to PEGylate previously palmitoylated cysteines following removal of hydroxylamine as previously described (Plain et al., 2020) . To block free cysteine, cells or supernatant or purified virions were lysed and incubated in 400 μl buffer (2.5% SDS, 100 mM HEPES, 1 mM EDTA, 100mM Maleimide pH 7.5, and protease inhibitor cocktail) for 4 h at 40°C. To remove excess unreacted maleimide, proteins were acetone precipitated and re-suspended in buffer (100 mM HEPES, 1 mM EDTA, 1% SDS, pH 7.5). Previously palmitoylated cysteines were revealed by treatment with 250 mM hydroxylamine (NH2OH) for 1 hour at 37°C. Cell lysates were desalted using Zebra spin columns and incubated 1 hour at 37°C with 2mM 5kDa PEG: methoxypolyethylene glycol maleimide. Reaction was stopped by incubation in SDS sample buffer with ß-mercaptoethanol for 5 min at 95°C. Samples were separated by SDS-PAGE and analyzed by immunoblotting. A fraction of the cell lysate was saved before the addition of 5kDa PEG as the input.

Approximately 1 × 10 7 cells were re-suspended in 0.5 ml cold TNE buffer ( 

Surface biotinylation was performed on transfected cells. Cells were allowed to cool down shaking at 4°C for 15 min to arrest endocytosis. Cells were then washed three times with cold PBS and treated with EZ-Link Sulfo-NHS-SS-Biotin No weight for 30 min shaking at 4°C. Cells were then washed 3 times for 5 min with 100mM NH4Cl and lysed in 1% Tx-100 to do DRMs or in IP Buffer for 1h at 4°C. Lysate were then centrifuged for 5 min at 5000rpm and the supernatant incubated with streptavidin agarose beads overnight on a wheel at 4°C. Beads were washed with IP buffer 5 times and the proteins were eluted from the beads by incubation in SDS sample buffer with ß-mercaptoethanol for 5 min at 95°C. Samples were separated by SDS-PAGE and analyzed by immunoblotting. Western blots were developed using the ECL protocol and imaged on a Fusion Solo from Vilber Lourmat. Densitometric analysis was performed using the software Bio-1D from the manufacturer.

Plasmids expressing SARS-CoV-2 Spike WT-HA or SARS-CoV-2 Spike 10C-A-HA proteins, pHAGE2-CMV-Luc-ZSgreen, Hgpm2, REV1b, Tat1b (a kind gift from J.D.

Bloom) were co-transfected into 293T cells for 24 hours with the following ratio 3/9/2/2/2 (18 µg/ 56.7cm 2 plate) using Fugene transfection reagent. The following day, cells were transferred in EpiSerf medium, and cell supernatants were collected after 8 hours and 16 hours. Harvested supernatants were pooled, clarified by low-speed centrifugation, filtered to remove cell debris (5 ml were kept at this stage for infectivity assays) and concentrated by centrifugation at 47000g through a 20%(w/v) sucrose cushion for 90min. Concentrated virus like particles (VLPs) were immediately resuspended in 300 μl of cold TNE buffer (25mM Tris-HCl, pH 7.5, 150 mM NaCl, 5 mM EDTA, and 1% Triton X-100) for biochemical experiments.

Binding between Spike-pseudotyped VLPs and ACE-2 was determining using an 

Non-concentrated VLPs suspensions were adjusted for p24 content (as monitored by ELISA) and used to transduce 293T-ACE2 cells (technical replicates n=12).

Transduced cells were analyzed 3 days later with the Bright-Glo Luciferase assay system.

For time-course optimization ( Fig S2) and 24 h replication experiments (Fig 7) Infected

Vero-E6 cells and correspondent supernatants (150 μl), cultured in 6-well plates (2 ml culture medium per well -3 independent wells per condition) were harvested at indicated time points and lysed in 500 μl of mastermix QVL lysis buffer from E.Z.N.A.

Viral RNA kit used for Viral RNA extraction according to manufacturer's instructions.

RNA concentration was measured and 500 ng or 1000 ng of total RNA was used for cDNA synthesis using iScript. Samples were then used to perform QPCR analysis (using E specific primers) as previously described. A standard curve was generated by plotting the Ct values against the number of E copies per ml indicated by the serial dilution of the Estandards (stock at 1e5 copies / µl) and used to extrapolate the number of E copies per ml in samples.

Vero-E6 cell cultured in T75 culture flasks (≈80-90% confluency after 72 h siRNA transfection) were infected with MOI=1 as described and further incubated in 5 ml culture medium per flask using 3 independent flasks per condition. 10 h after inoculation cells were washed scrapped and harvested with correspondent supernatants. Cells (1/2 of T75) and correspondent supernatants (150 μl) were processed for QPCR analysis or lysed in IP buffer and processed for Western blot as described. Remaining supernatants were aliquoted and stored at -80°C for titration of viral E copies. After titration and adjustment to viral E RNA copies non-concentrated SARS-CoV-2 supernatants were used to infect confluent Vero E6 cell monolayers cultured in 12 well plates using an approximate ration or 50 E copies per cultured host cell. At least 3 wells were infected per supernatant per condition. Infection was done as described until 6 h post inoculation when cells were washed lysed and processed for QPCR analysis as described 

Approximately 10 ml (per T75 flask) of pre-cleared culture supernatants from 48 h infections were aliquoted into 1.5 ml microtubes and concentrated by centrifugation at 47000g through a 20%(w/v) sucrose cushion (200 μl) for 90min. Medium and sucrose supernatants were discard, pellets allowed to dry and re-suspended and pooled in:

400 μl of cold 1%-Triton X TNE buffer for DRM analysis; or 500 µl 1:1 mixture of methanol:chloroform for LPDX analysis.

Total lipid extracts were prepared using a standard MTBE protocol followed by a methylamine treatment for total lipid analysis by mass spectrometry. Briefly, cell pellets or 

Lipid extracts (2 μL injection volume in ChCl3:MeOH 2:1) were separated over an 8 min gradient at a flow rate of 200 μL/min on a HILIC Kinetex Column (2.6lm, 2.1 × 50 mm2) on a Shimadzu Prominence UFPLC xr system (Tokyo, Japan). Mobile phase A was acetonitrile:methanol 10:1 (v/v) containing 10 mM ammonium formate and 0.5% formic acid while mobile phase B was deionized water containing 10 mM ammonium formate and 0.5% formic acid. The elution of the gradient began with 5% B at a 200 μL/min flow and increased linearly to 50% B over 7 min, then the elution continued at 50% B for 1.5 min and finally, the column was re-equilibrated for 2.5 min. MS data were acquired in fullscan mode at high resolution on a hybrid Orbitrap Elite (Thermo Fisher Scientific, Bremen, Germany). The system was operated at 240,000 resolution (m/z 400) with an AGC set at 1.0E6 and one microscan set at 10-ms maximum injection time. The heated electrospray source HESI II was operated in positive mode at a temperature of 90 C and a source voltage at 4.0KV. Sheath gas and auxiliary gas were set at 20 and 5 arbitrary units, respectively, while the transfer capillary temperature was set to 275 °C.

Mass spectrometry data were acquired with LTQ Tuneplus2.7SP2 and treated with To analyze and quantify free cholesterol (filipin) staining, we used the MetaXpress Custom Module editor software from Molecular Devices, as in previous publications (Larios et al., 2020; Moreau et al., 2019) . We first segmented each individual nucleus using the PI channel, we then used the HA channel to mask the expressing cells. The final masks were applied to all original fluorescent images and measurements per cell (nucleus) and average per well were extracted. The same analysis pipeline was applied to all images.

The model of trimeric Spike's TM helix + C-terminal region (residues 1201-end in Uniprot P0DTC2) was built by extending laterally the C-terminus of model 1_1_2 built from PDB 6VSB by (Woo et al., 2020) . Membrane insertion was carried out with CHARMM-GUI's web tool in atomistic or coarse-grained (CG) lipid bilayers as required (Hsu et al., 2017; Jo et al., 2008; Qi et al., 2015) . For CG molecular dynamics (MD) simulations, we parametrized, also using CHARMM-GUI, the protein with the MARTINI 2.2p CG force field (Marrink et al., 2007) , the DPPC, DLiPC and cholesterol with MARTINI CG force field for lipids (Marrink et al., 2004) (notice that DLiPC corresponds to MARTINI's DIPC lipid), and S-palmitoylated cysteines with the parameters provided by (Atsmon-Raz and Tieleman, 2017). Membranes for CG MD simulations contained 50% DPPC, 30% DLiPC and 20% cholesterol, which we verified forms rafts within 1 μs of CG MD simulation without proteins inserted, as shown by (Lorent et al., 2017) . For atomistic MD simulations, we parameterized the protein, POPC membrane and palmitoylated cysteine with CHARMM36m as implemented in CHARMM-GUI. All MD simulations were run in Gromacs (Abraham et al., 2015) using standard settings from CHARMM-GUI for CG and atomistic membrane simulations, at 303 K. In atomistic MD we used a 12 Å cutoff for nonbonded interactions, PME, NVT equilibration followed by NPT production with semi-isotropic pressure coupling to 1 atm and 2 fs timestep; in CG MD we used standard MARTINI equilibration for membranes followed by NPT with semi-isotropic pressure coupling to 1 atm and 20 fs timestep. Simulations were visualized in VMD (Humphrey et al., 1996) and quantitative analysis were carried out through custom VMD Tcl and bash scripts. Membrane perturbation maps were built with MEMBPLUGIN for VMD (Guixà-González et al., 2014) .

are mean ± SEM, and each dot represents an independent VLP preparation F. Western blot analysis of fractionation profiles from VLPs generated as in B and used for quantification in Figure 5C . G, H.

HEK293Tcells corresponding to the representative blots showed in Figure 5D . p values were obtained by students t test I. Concentration of non-typed, bald, or HA-Spike (WT or 10C-A mutant) pseudotyped VLP suspensions generated as in Figure 5D -J, measured by ELISA for the lentiviral protein p24.

Results are mean ± SEM, and each dot represents an independent preparation. 

GROMACS: High performance molecular simulations through multi-level parallelism from laptops to supercomputers

Dynamic Radiolabeling of S-Palmitoylated Proteins

Parameterization of Palmitoylated Cysteine, Farnesylated Cysteine, Geranylgeranylated Cysteine, and Myristoylated Glycine for the Martini Force Field

SARS-CoV-2 (COVID-19) by the numbers

FTY720 inhibits ceramide synthases and up-regulates dihydrosphingosine 1-phosphate formation in human lung endothelial cells

Proteomics of SARS-CoV-2-infected host cells reveals therapy targets

Envelope Protein Palmitoylations Are Crucial for Murine Coronavirus Assembly

The SARS-CoV-2 envelope and membrane proteins modulate maturation and retention of the spike protein, allowing assembly of virus-like particles

Viral targets for vaccines against COVID-19

Glycosphingolipid synthesis requires FAPP2 transfer of glucosylceramide

Vesicular and non-vesicular transport feed distinct glycosylation pathways in the Golgi

Molecular Interactions in the Assembly of Coronaviruses

Assembly of coronavirus spike protein into trimers and its role in epitope expression

Phospholipase D and phosphatidic acid in the biogenesis and cargo loading of extracellular vesicles

Hemagglutinin of Influenza A, but not of Influenza B and C viruses is acylated by ZDHHC2, 8, 15 and 20

Toward the identification of ZDHHC enzymes required for palmitoylation of viral protein as potential drug targets

Palmitoylation of the Alphacoronavirus TGEV spike protein S is essential for incorporation into viruslike particles but dispensable for S-M interaction

Altan-Bonnet, N., 2020. β-Coronaviruses Use Lysosomes for Egress Instead of the Biosynthetic Secretory Pathway

MEMBPLUGIN: studying membrane complexity in VMD

Pre-and post-Golgi translocation of glucosylceramide in glycosphingolipid synthesis

Multifaceted roles for lipids in viral infection

A Multibasic Cleavage Site in the Spike Protein of SARS-CoV-2 Is Essential for Infection of Human Lung Cells

SARS-CoV-2 Cell Entry Depends on ACE2 and TMPRSS2 and Is Blocked by a Clinically Proven Protease Inhibitor

Substrate recognition by the cell surface palmitoyl transferase DHHC5

CHARMM-GUI Martini Maker for modeling and simulation of complex bacterial membranes with lipopolysaccharides

VMD: Visual molecular dynamics

SARS-CoV-2 and ORF3a: Nonsynonymous Mutations, Functional Domains, and Viral Pathogenesis. mSystems 5

Severe Acute Respiratory Syndrome Coronavirus 3a Protein Is a Viral Structural Protein

CHARMM-GUI: A web-based graphical user interface for CHARMM

Structures and distributions of SARS-CoV-2 spike proteins on intact virions

Virus Impact on Lipids and Membranes

SARS-CoV-2 structure and replication characterized by in situ cryo-electron tomography

Ceramide synthesis is modulated by the sphingosine analog FTY720 via a mixture of uncompetitive and noncompetitive inhibition in an Acyl-CoA chain length-dependent manner

ALIX-and ESCRT-III-dependent sorting of tetraspanins to exosomes

Functional assessment of cell entry and receptor usage for SARS-CoV-2 and other lineage B betacoronaviruses

Lipid Rafts: Controversies Resolved, Mysteries Remain

Importance of Conserved Cysteine Residues in the Coronavirus Envelope Protein

Structural determinants and functional consequences of protein affinity for membrane rafts

Coarse Grained Model for Semiquantitative Lipid Simulations

The MARTINI Force Field: Coarse Grained Model for Biomolecular Simulations

Palmitoylation of SARS-CoV S protein is necessary for partitioning into detergent-resistant membranes and cell-cell fusion but not interaction with M protein

Drug-induced increase in lysobisphosphatidic acid reduces the cholesterol overload in Niemann-Pick type C cells and mice

Spike glycoprotein and host cell determinants of SARS-CoV-2 entry and cytopathic effects

Mass-tag labeling for the determination of protein Sfatty acylation

Palmitoylation of the cysteine-rich endodomain of the SARS-coronavirus spike glycoprotein is important for spike-mediated cell fusion

CHARMM-GUI Martini Maker for Coarse-Grained Simulations with the Martini Force Field

Fatty acyl recognition and transfer by an integral membrane S-acyltransferase

The molecular face of lipid rafts in model membranes

SARS-CoV-2 Requires Cholesterol for Viral Entry and Pathological Syncytia Formation (preprint)

Acylation of virol. spike glycoproteins: A feature of enveloped RNA viruses

The M, E, and N Structural Proteins of the Severe Acute Respiratory Syndrome Coronavirus Are Required for Efficient Assembly, Trafficking, and Release of Virus-Like Particles

The intracellular sites of early replication and budding of SARScoronavirus

A Novel Severe Acute Respiratory Syndrome Coronavirus Protein, U274, Is Transported to the Cell Surface and Undergoes Endocytosis

Palmitoylations on Murine Coronavirus Spike Proteins Are Essential for Virion Assembly and Infectivity

SARS-CoV envelope protein palmitoylation or nucleocapid association is not required for promoting virus-like particle production

Fatty acid acylation of viral proteins in murine hepatitis virus-infected cells

Membrane lipids: where they are and how they behave

Coronavirus biology and replication: implications for SARS-CoV-2

A novel coronavirus outbreak of global health concern

Site-specific glycan analysis of the SARS-CoV-2 spike

S-acylation of the Insulin-Responsive Aminopeptidase (IRAP): Quantitative analysis and Identification of Modified Cysteines

Developing a Fully Glycosylated Full-Length SARS-CoV-2 Spike Protein Model in a Viral Membrane

Molecular Architecture of the SARS-CoV-2 Virus

The molecular era of protein S-acylation: spotlight on structure, mechanisms, and dynamics

Uninfected cells were used as control. Results are mean±SEM B-D. Western blot and autoradiographic analysis of Spike IP fractions from: B, Hela cells siRNA-depleted for the indicated ZDHHCs or transfected with control siRNAs (Ctrl), and expressing Spike as described and quantified in Figure 1E

KO cells expressing Spike described and quantified in Figure 1F; or D from HeLa cells processed as in B but co-transfected with the siRNA resistant plasmids expressing myc-tagged ZDHHC8, 9, 20 or 6. In (-) cells were transfected with empty plasmid. Screening experiments were carried at 48 to 72 h of Spike expression, with Spike cleavage being more pronounced at later times. Yet, following a 2 h 3H-palmitate pulse, the majority of labeled Spike was full-length. E. ZDHHC mRNA levels of in different cell lines. Values are TAG per Million (TPM) corresponding to 10*6 X ((reads mapped to transcript/transcript length)/ SUM (reads mapped to transcript/transcript length))

Acyl-RAC capture assay in Vero E6 cells transfected (72 h) with an siRNA pool to silence ZDHHC8, ZDHHC9 and ZDHHC20 (siMixT) or siRNA control (Ctrl) and subsequently

Cell lysate fractions (input), S-palmitoylated proteins (detected after hydroxylamine treatment, +NH2OH) and control fractions (-NH2OH) were analyzed by western blot against Spike. G. Western blot, against Spike and GAPDH, and autoradiographic analysis of total cell extracts (TCE) or

Spike IP-fractions from Vero E6 cells transfected with siRNAs targeting individual ZDHHCs or control siRNA (siCtrl) and infected 24 h with SARS-CoV-2, MOI=0.1. Cells were metabolically labeled throughout the 24 h of infection with 3 H-palmitic acid and analyzed for Spike S-palmitoylation quantified in Figure 1G. H. Q-PCR analysis of mRNA levels in Vero E6

Control (siCtrl) or the indicated siRNAs. Results are mean ± SEM, and each dot represents and independent experiment n ≥3. I. Alignment of the juxtamembranenous regions of the SARS-CoV-1 or

Transmembrane domains are shown in blue and cytosolic cysteine residues highlighted in yellow

CoV-1 (P59637)

CoV-1 (P59596)

FIGURE S2 -Complement to Figure 2

Vero E6 were left uninfected (red line) or infected with SARS-CoV-2 MOI≈0.5/1 for the indicated time points. Cells were either harvested and processed for flow cytometry with Spike antibodies (A-B) or processed for quantification of total E and RdRp Viral RNA by Q-PCR

Spike signal per cell (GeoMean a.u.) (B)

quantified for 3 independent biological replicates. Results are mean ±SD. D. Representative autoradiography ( 3 H-Palm) images and western blot analysis of 3 H-Palmitic acid incorporation by Spike WT transfected in HeLa cells as described in Figure 2F.E. Confocal images of Vero E6 cells ectopically expressing the indicated myc-ZDHHC enzymes (24 h) fixed, and immunolabelled with myc antibodies

Acy-PEG exchange assay in lysates from Vero E6 cells infected with SARS-CoV-2, (A) at high

S-acylated proteins were labelled with PEG-5KDa (+mPEG) following hydroxylamine treatment (NH2OH) and analyzed with control non-labelled (-mPEG), and input cell lysate fractions (input) by western blot against Spike or (A) Calnexin as positive control. In (A) a parallel analysis (right panel) was performed using double of the protein content. Yellow asterisks indicate shifted Spike bands at 2 and 4 h p.i. C. Acylrac capture assay from pre-cleared, filtered supernatants of Vero E6 siRNAdepleted and infected as in B. Hela cells transfected for 24 to 48 h with plasmids expressing untagged Spike were left untreated or pretreated with Palmostatin B for 4 h in serum free medium. Cells were metabolically labeled for a 2 h pulse with 3 H-palmitic acid, and chased in complete medium for the indicated times

Similar analysis was used to quantify the data depicted in Figure 3E. F. Autoradiography ( 3 H-Palm) images and western blot analysis of total cell lysate fractions described and quantified in Figure 3G G. Representative Autoradiography images and GH western blot analysis of: (G) 35 S-Met/Cys pulse chase; and (H) Cyclohexamide (CHX) pulse experiments described and quantified in Figure 3H and 3K

FIGURE S4 -Complement of Figure 4

Center and Right) correspond to Left and Center of Figure 4A in the main text, but seen from the external side. B. Fractionation and DRM isolation (see experimental procedures) of Vero E6 cells expressing HA-tagged Spike (WT or 10C-A mutant -24 h). Fractions were analyzed by western blot with anti HA (Spike) and Caveolin1 antibodies. C. Quantification of Spike-HA (all forms) in each fraction divided by the total signal in all the 6 fractions

Quantification of HA-Spike signal in each fraction as in D. FG -Surface biotinylation assay as in C and G quantification of total surface proteins pulleddown by streptavidin-beads and analyzed, with correspondent total cell lysate fractions (TCE), by western blot against HA (HA Spike), LRP6 (surface control) and GAPDH (negative control). H. Values for Spike surface expression is shown divided by the amount of

Results are mean± SEM and each dot represent an independent experiment, n=3. I. Confocal images of Vero E6 cells transfected as in B immunolabelled for HA-Spike, and stained with Hoechst (nuclei)

IJ Western blot analysis of fractions from: I Vero E6 cells transfected with siZDHHC20 or siZDHHC9 (72 h) and infected with SARS-CoV-2, 24 h MOI ≈0.1 or J uninfected cells Vero E6 cells, processed as in C

Concentration of VLP suspensions generated in Figure 5A-C, measured by ELISA for the lentiviral protein p24. Results are mean ± SEM, and each dot represents an independent preparation. C, D, E. Western blot analysis, and quantification of VLPs generated in Figure 5A-C. Blots were analyzed for HA Spike, and p17 lentiviral matrix protein. D. HA-Spike levels were normalized against p17 levels and set to 1 for WT. E Spike cleavage was quantified as the ratio between S2 and S2+S full length. Results as fraction of WT-pseudotyped VLP signal in HeLapACE-2 cells (see Figure 7B). The increased exposure of western blot membranes to retrieve signal from the internalize S2 fragment originated the appearance of an unspecific band indicated in the gel. B Control infectivity assays as done as in Figure 7D using VSVG-pseudotyped VLPs suspensions harvested from siCtrl or siZDHHC20-silenced cells adjusted to the lentiviral p24 content. C Western Blot analysis, probing for viral proteins Spike and M and host cellular GAPDH, from infected total cell extracts (Vero E6 cells) and correspondent supernatants retrieve after 10 h of infection with SARS-CoV-2 at high MOI (1-2) as described in Figure 7G and used to re