Heparan sulfate promotes ACE2 super-cluster assembly to enhance SARS-CoV-2-associated syncytium formation

The mechanism of syncytium formation, caused by spike-induced cell-cell fusion in severe COVID-19, is largely unclear. Here we combine chemical genetics with 4D confocal imaging to establish the cell surface heparan sulfate (HS) as a critical host factor exploited by SARS-CoV-2 to enhance spike’s fusogenic activity. HS binds spike to facilitate ACE2 clustering, generating synapse-like cell-cell contacts to promote fusion pore formation. ACE2 clustering, and thus, syncytium formation is significantly mitigated by chemical or genetic elimination of cell surface HS, while in a cell-free system consisting of purified HS, spike, and lipid-anchored ACE2, HS directly induces ACE2 clustering. Importantly, the interaction of HS with spike allosterically enables a conserved ACE2 linker in receptor clustering, which concentrates spike at the fusion site to overcome fusion-associated activity loss. This fusion-boosting mechanism can be effectively targeted by an investigational HS-binding drug, which reduces syncytium formation in vitro and viral infection in mice.


Introduction
SARS-CoV-2, a single-stranded RNA virus coated with a membrane derived from host cells, injects its genetic materials into the host cytoplasm at the cell surface or following receptor-mediated endocytosis 1,2 . Both entry routes require the fusion of viral membranes with host membranes, which depends on the cell surface receptor ACE2 and the viral glycoprotein spike 3 , a single-spanning homotrimeric membrane protein 4 . During viral biogenesis, the spike undergoes a furin-mediated cleavage at the S1 site, generating two covalently linked fragments, S1 and S2 5 . The S1 fragment contains the receptor binding domain (RBD) that exists in at least two conformations: In the up conformation, the spike binds ACE2 with high a nity 6 , inducing TMPRRS2-mediated cleavage of the S2 fragment to expose a fusogenic peptide that drives the fusion of viral membranes with the plasma membrane 5,7,8 .
ACE2-binding also induces receptor-mediated endocytosis, transferring viral particles to late endosomes/lysosomes where a lysosomal protease activates the spike in a similar way to promote membrane fusion 2,7,9-12 . Intriguingly, in addition to ACE2, recent studies have established the cell surface heparan sulfate (HS) as a critical co-receptor that assists ACE2 in viral entry 9,[13][14][15][16] . HS refers to a class of negative chargeenriched polysaccharides attached to the speci c membrane and secretory proteins collectively termed heparan sulfate proteoglycans (HSPG) 17 . Many viruses attach to the cell surface rst by binding to HS 18 . Recent studies showed that HS could bind spike directly, forming an ACE2-containing ternary complex to promote SARS-CoV-2 endocytosis 9,13,15,19,20 . The role of HS in SARS-CoV-2 infection is critical in cells with low levels of ACE2 expression 14,21,22 . Accordingly, endocytosis-mediated SARS-CoV-2 cell entry can be inhibited by HS-binding drugs or HS mimetic compounds 9,13,23 .
While ACE2-mediated viral endocytosis has been extensively studied, the mechanism underlying spikeinduced membrane fusion needs to be better characterized. The spike-ACE2 interactions induce the the medium as an indicator of cytotoxicity detected no signi cant cell death in PIXN-treated samples except in prolonged treatment with the highest dose (20 µM) ( Figure 1G).
While numerous studies have established HS as an assisting factor for SARS-CoV-2 entry in vitro 19 , it is unclear whether HS affects SARS-CoV-2 infection in vivo. We therefore used a mouse model to further evaluate the anti-SARS-CoV-2 activity of PIXN ( Figure 1H). To this end, we injected PIXN intravenously at two doses (50 µg/kg and 100 µg/kg) into K18-hACE2 transgenic mice expressing human ACE2 33 . We then infected the mice with live USA-WA1/2020. Seventy-two hours post-infection, we collected lung tissues and measured the viral load by qRT-PCR (Figure 1 H). We found that for animals treated with PIXN at 100 µg/kg, the viral load was reduced by ~60% compared to the control group, while at a lower dose, a minor reduction was observed ( Figure 1I). These results show that PIXN, as an HS-binding drug, can mitigate SARS-CoV-2 infection in vivo, although the activity is weaker than that seen in cell-based assays. The low in vivo anti-SARS-CoV-2 activity may be due to PIXN binding to HS in non-targeting tissues, which reduces its concentration in the lung.

PIXN and MTAN interact with HS via speci c sulfate groups
We next used a synthetic HS hexasaccharide (the six sugar moieties are designated as A to F, respectively) with de ned sulfate groups (named 6-mer-NS2S3S6S) to characterize how PIXN and MTAN interact with HS ( Figure 2A) because the HS 6-mer binds these drugs similarly as longer heterogenous HS chains (see below). We rst used NMR to measure chemical shift perturbations of HS 6-mer NS2S3S6S caused by PIXN. MTAN was omitted from this experiment because of its self-aggregation property at the concentration required for the NMR study. The result showed that most spectral changes happened on the sugar moieties B-D, centering around the C-3 position in N-sulfo-D-glucosamine B and the anomeric position in glucuronic acid C ( Figure 2B, C). Reciprocal titration of HS 6-mer NS2S3S6S to PIXN showed that the two -NH groups on the arms were most signi cantly affected ( Figure 2D). These results suggest that HS might use speci c sulfate groups (e.g., 2S, NS, and 6S) in the sugar moieties B-D to interact with the -NH groups of PIXN. This model is consistent with the observation that the -NH groups are conserved in MTAN.
To validate the NMR results, we used a binding assay based on the fact that both MTAN and PIXN absorb light at 650 nm 9 and that the interaction of these drugs with HS or HS 6-mer NS2S3S6S resulted in a dose-dependent reduction in Ab 650 ( Figure 2E, Figure S2A). Plotting the change in Ab 650 over HS concentrations suggested that PIXN bound to 6-mer-NS2S3S6S with a similar a nity as MTAN. However, the maximum binding for PIXN was about two-fold higher than that of MTAN ( Figure 2F). As expected, an HS 6-mer analog bearing no sulfate group (6-mer-0S) did not interact with either PIXN or MTAN ( Figure 2G and Figure S2B), demonstrating a sulfate-dependent interaction with these drugs.
To identify the sulfate group(s) involved in drug binding, we performed a binding study using synthetic HS 6-mers containing only given types of sulfate groups ( Figure S2C). The results showed that the drug binding depends on both sulfate number and position ( Figure 2H, I); Low binding was generally detected for HS 6-mer carrying three NS groups. For MTAN, the addition of three 6S groups (NS6S) increased the a nity dramatically, while having additional 2S and 3S groups (NS6S2S or NS2S3S6S) did not further improve MTAN binding ( Figure 2H). By contrast, 6-mer NS6S only had a modestly increased a nity for PIXN compared to 6-mer NS, while adding 2S and 3S maximized the a nity to PIXN ( Figure 2I). Thus, while PIXN and MTAN both bind HS via sulfate groups, they have different preferences for sulfate position.

Pharmacological inhibition of HS mitigates SARS-CoV-2-induced cell-cell fusion
Having established PIXN and MTAN as HS-binding drugs that block endocytosis-mediated entry of SARS- CoV-2, we tested whether these drugs inhibited SARS-CoV-2 entry via fusion at the plasma membrane. We chose the Delta variant (B.1.617.2) because recent studies suggested that the spike of this variant has the most robust membrane fusion-stimulating activity compared to spikes of other variants 34 . Consistent with this view, when Vero TA6 cells infected with the Delta variant were stained with an antibody against the SARS-CoV-2 nucelocapsid protein (NP), we detected viral particles as small puncta throughout the cytoplasm ( Figure 3A). Cells treated with either PIXN or MTAN during infection still contained NP-positive puncta. However, the intensity was reduced by ~50% ( Figure 3A, B). Thus, both PIXN and MTAN can inhibit Delta entry via the plasma membrane.
As expected from the robust fusogenic activity of the Delta spike, we observed many multinuclear syncytia in cells infected with the Delta variant ( Figure 3C, D). Presumably, the fusion of the viral envelope with the plasma membrane redistributes the spike to the cell surface, causing infected cells to fuse with nearby cells. Consistent with this notion, infection with the endosome-bound USA-WA1/2020 generated few syncytia under the same conditions ( Figure S3A). In Delta-infected samples, some syncytia are giant, containing up to 80 nuclei. Their formation requires at least seven rounds of fusion. Given the low MOI used, it is surprising that the limited spike molecules transferred from the viral particles can maintain such high fusogenic activity, particularly as each round of fusion increases the cell surface area, which further reduces the spike concentration. Interestingly, when cells infected with the Delta variant were treated with PIXN or MTAN, syncytium size was much reduced, with most syncytia containing only 4-6 nuclei. These ndings suggest a fusion-boosting mechanism that helps overcome fusion-associated fusogen dilution, which is sensitive to HS-binding drugs.

HS promotes spike-induced syncytium formation
To elucidate the role of HS in spike-induced cell-cell fusion, we optimized a co-culture-based fusion assay ( Figure 3E). We co-transfected 293T cells with the Delta spike-and mCherry-expressing plasmids (Effector) and then added these cells in suspension to a monolayer of 293T cells stably expressing ACE2-GFP (Acceptor). Spike-expressing cells were round and mostly mCherry-positive, but after fusing with ACE2 cells, they attached to the surface and attened into irregular shapes. After incubation, unfused cells were removed by washing. The fusion e ciency could be monitored by counting the number of nuclei per syncytium under uorescence microscopy. In this assay, syncytium formation strictly depends on the spike in effector cells and ACE2 in acceptor cells 35 . Additionally, both PIXN and MTAN inhibited syncytium formation in this assay ( Figure S3B, C) 35 , suggesting that the assay recapitulates virusassociated syncytium formation.
We next created HS-de cient ACE2-GFP cells by CRISPR-mediated knockout (KO) of SLC35B2, a Golgilocalized sulfate transporter essential for the sulfation of HS chains. Staining cells with a super-charged cyan uorescent protein (CFP+) ( Figure 3H), a positive charge-bearing uorescence protein that binds HS with high a nity, con rmed the HS de ciency in the KO cells 9 . When SLC35B2 KO or control ACE2-GFP cells were co-cultured with spike/mCherry cells, syncytia formed by SLC35B2 KO cells were signi cantly smaller than in wild-type control despite similar levels of ACE2-GFP expression ( Figure 3F, G). Notably, the cell fusion defect of SLC35B2 de cient cells was rescued entirely when heparin, a heavily sulfated HS, was added to the medium, attributing the phenotype to HS de ciency. Heparin treatment also increased the size of syncytia in wild-type (WT) cells. These results demonstrate a critical role for the cell surface HS in spike-induced cell-cell fusion, which can be substituted by unanchored heparin.
To further validate the role of HS in membrane fusion, we treated ACE2-GFP cells with heparinase I/III to remove the cell surface HS ( Figure 3H). Like SCL35B2 KO cells, heparinase-treated ACE2-GFP cells, when co-cultured with spike/mCherry cells, produced syncytia of smaller sizes ( Figure S3D, E). Likewise, when heparinase-treated Vero TA6 cells were infected with the Delta strain, we found that heparinase treatment signi cantly reduced the syncytium size ( Figure 3I, J). Because removing HS from acceptor cells alone is su cient to mitigate cell-cell fusion and because high concentrations of heparin are required to restore the fusion activity in SLC35B2 KO cells, it appears that HS acts most effectively when positioned in cis to ACE2-containing membranes (see discussion).

HS enables synapse-like cell-cell contacts in spike-mediated syncytium formation
To fuse with acceptor cells, an effector cell needs to go through the following conceptual steps: 1) forming cell-cell contact, 2) generating a small fusion pore, 3) the pore widening into a signi cant gap, and 4) complete fusion of the two cells ( Figure 4A). We used the co-culture assay to narrow down the steps affected by PIXN. Speci cally, in addition to measuring the syncytium size ( Figure 4B), we counted the number of mononuclear cells that have both mCherry and ACE2-GFP ( Figure 4C). These cells are in a semi-fusion state (F3 in Figure 4A) in which a small fusion pore has formed, allowing mCherry to enter the ACE2-GFP cell, but the membrane boundary between the two cells remains largely intact. In vehicletreated reactions, ~4% of ACE2-GFP cells were detected in this semi-fusion state after 30 min incubation with spike/mCherry cells. In contrast, in PIXN-treated conditions, the number of semi-fusion cells increased to ~9% ( Figure 4C, Figure S4A). After further incubation, semi-fusion cells were almost undetectable in the control reaction because most cells had completed at least one round of fusion.
However, in the presence of PIXN, ~8% of ACE2 cells remained in the semi-fusion state. These results suggest that PIXN delays the opening of the fusion pore to inhibit cell-cell fusion.
To elucidate the precise step that involves HS, we used 4D live cell confocal microscopy to monitor the effect of PIXN on cell-cell fusion in real-time. Analyses of randomly selected fusion events con rmed that PIXN treatment delayed both the entry of mCherry into ACE2-GFP cells and the appearance of a visible pore. However, once a visible pore was formed, the pore widening was unaffected ( Figure 4C, D). These results suggest that HS facilitates fusion pore formation.
How does HS inhibition delay the formation of fusion pores? Since biochemical studies showed that neither PIXN nor MTAN inhibited HS binding to the spike ( Figure S4B) and that PIXN did not affect the interaction of the spike with ACE2 in vitro regardless of whether HS was present ( Figure S4C), HS is likely dispensable for the initiation of cell-cell contact. Consistent with this notion, when spike cells contact ACE2 cells, the spike is processed by the cell surface TMPRSS2, generating an S2' fragment and a fusion peptide. This process was also unaffected by PIXN or knockout of SLC35B2 in acceptor cells ( Figure S4D, E).
Transmission electron microscopy showed that 12 min after mixing spike-and ACE2-cells, these cells form synapse-like cell-cell contacts characterized by long juxtaposed plasma membranes interrupted by bubble-like structures. The paralleled membranes, unseen when ACE2 cells were incubated with cells without spikes ( Figure S4F), are only separated by a gap of less than 20 nm (Figures 4E, right panels, 4F). Importantly, we frequently observed obscure membrane boundaries at these membrane contact sites, probably caused by localized fusion events. By contrast, in PIXN-treated samples, cell-cell contacts were seen, but membranes were only held together at a few spots, leaving signi cant gaps in between, and fusion pores were rarely seen (Figures 4E, left panels, 4F). Thus, HS facilitates the formation of tight membrane junctions between the effector and acceptor cells in spike-induced cell-cell fusion.
HS enhances ACE2 clustering at the fusion sites 4D live cell confocal imaging revealed that within minutes following the contact of spike cells with ACE2-GFP cells, ACE2-GFP was rapidly concentrated at the cell-cell contact sites ( Figure 4C, video 1), consistent with a recent report 26 . The ACE2-GFP clusters expanded over time, forming synapse-like structures with a diameter of 10-20 µm. Fluorescent recovery after photobleaching (FRAP) suggested that ACE2-GFP was almost entirely immobile in these clusters ( Figures 5A, B), suggesting that ACE2 clusters are likely formed by ordered intermolecular interactions as opposed to liquid-liquid phase separation. As expected, immunostaining with spike antibodies also detected spike accumulation at the cell-cell contact sites ( Figure 5C). Since recombinant spike fragments containing the receptor binding domain (RBD) could completely block ACE2 cluster formation ( Figure 5D), ACE2 clustering requires spike-ACE2 interactions. Interestingly, 3D confocal imaging revealed enrichment of actin laments around the ACE2 clusters, resembling those in the immune synapse ( Figure S5). Altogether, these results suggest that the interaction of the spike with ACE2 drives ACE2 into clustered super-assemblies together with the spike at the membrane contact sites. Because ACE2 clustering always precedes the entry of mCherry into the ACE2-GFP cell, ACE2 clustering must play a role in spike-induced cell-cell fusion.
To determine whether HS regulates ACE2 cluster formation, we compared the size of ACE2 clusters in WT cells to that in the SLC35B2 KO cells because the latter lack HS. When SCL35B2 KO ACE2-GFP cells were incubated with spike/mCherry cells, the size of ACE2 clusters was much smaller than those in WT ACE2-GFP cells ( Figure 5E, G). Likewise, PIXN treatment of ACE2-GFP cells signi cantly reduced the size of ACE2 clusters ( Figures 5F, H). Kinetic analysis of cluster formation by 4D confocal imaging further con rmed the ACE2 clustering defect when the cell surface HS is inhibited by PIXN ( Figure 5I). These results suggest that HS is required for spike-induced ACE2 super-cluster formation.
HS acts via a conserved ACE2 linker to promote receptor clustering and cell-cell fusion As HS does not change the a nity of the spike for ACE2, we postulated that HS might induce a conformational change in ACE2, exposing a motif that drives receptor clustering. ACE2 is a singlespanning membrane protein with a large extracellular domain (ECD) for spike and ligand binding. The ECD is connected via an unstructured linker to the transmembrane domain, which mediates receptor dimerization. Intriguingly, albeit with no designated function, the unstructured linker is evolutionarily conserved ( Figure 6A).
To test whether the linker segment (LS) might be critical for spike-induced cell-cell fusion, we created a 293T cell line expressing a mutant ACE2-GFP with the LS replaced by a synthetic linker bearing a glycineserine repeat (ACE2-GS) of the same length. Immunoblotting and uorescence imaging showed that ACE2-GS-GFP was expressed at a similar level as WT ACE2-GFP and localized to the cell surface and endocytic vesicles similarly to WT ACE2-GFP ( Figures S6A, B). Furthermore, binding studies showed that ACE2-GS-GFP bound to the spike with a similar a nity as WT ACE2-GFP ( Figure S6C). Nevertheless, when co-cultured with spike-expressing cells, the ACE2-GS-GFP cells had dramatically reduced fusion activity, resulting in syncytia smaller than those formed by WT ACE2-GFP cells ( Figure 6B, C). Notably, while heparin stimulated the fusion of ACE2-GFP cells with spike cells, it failed to do so when ACE2-GS-GFP cells were incubated with spike cells ( Figure 6D), suggesting that HS acts via the ACE2 LS to promote syncytium formation. Moreover, when co-cultured with spike cells, ACE2-GS-GFP cells only formed ACE2 clusters with reduced size ( Figure 6E, Figure S6D); PIXN did not further reduce the size of ACE2 clusters in ACE2-GS-GFP cells. These results suggest that the conserved LS in ACE2 is critical for HS's function in ACE2 clustering and spike-mediated cell-cell fusion.
HS enhances spike-mediated ACE2 clustering in a cell-free system We developed a cell-free receptor clustering assay to test whether HS directly promotes ACE2 clustering ( Figure 7A). To this end, puri ed biotinylated ACE2 1-740 bearing a C-terminal 6xHis tag was labeled with Alexa 565 and anchored to a cover glass coated with a lipid bilayer that consisted of 1-palmitoyl-2-oleoylsn-glycero-3-phosphocholine (POPC), 1,2-dioleoyl-sn-glycero-3-phosphoethanolamine-N-[methoxy(polyethylene glycol)-5000] (PEG-5000 PE), and 1, 2-dioleoyl-sn-glycero-3-[(N-(5-amino-1carboxypentyl)iminodiacetic acid)succinyl] (DOGS-NTA). This protein contains just the ACE2 ECD and the conserved LS. In our method, the protein is anchored on the lipid bilayer in a highly mobile state, and importantly, with a topology mimicking membrane-anchored full-length ACE2. When membrane-anchored ACE2 1-740 was incubated with recombinant spike trimer, round-shaped bright speckles were formed due to ACE2 clustering, while incubating with buffer or HS alone did not induce ACE2 clustering. Consistent with our model, ACE2 clustering with spike alone was ine cient, but when HS and spike were added together, ACE2 cluster formation was signi cantly accelerated ( Figure 7B, C). Like ACE2 clusters in cells, ACE2 in in vitro-assembled clusters was also immobile, as demonstrated by FRAP ( Figure 7D). These results suggest that HS directly promotes ACE2 clustering in the presence of the spike.
Surprisingly, adding PIXN did not abolish spike/HS-induced ACE2 clustering. However, under this condition, ACE2 clustering generated mostly elongated laments ( Figure S7), suggesting that PIXN alters the mode of ACE2 self-assembly. The differential effect of PIXN on ACE2 clustering in cells vs. in vitro suggests that additional factors modulate ACE2 self-interaction at the cell surface, preventing lamentlike ACE2 clusters in the presence of PIXN.

Discussion
The cause of clinical heterogeneity among SARS-CoV-2-infected patients is unclear, but it is suggested that spike-induced cell-cell fusion may be a contributor to in ammation, thrombosis, or lymphopenia observed in severe COVID-19 patients because multi-nuclei-containing syncytia were often seen in damaged lung tissues from posthumous COVID-19 organs [25][26][27] , and because spike-induced syncytia can rapidly internalize lymphocytes, resulting in a unique cell-in-cell structure seen in COVID-19 tissues but not in other types of pneumonia 25 . When expressed in neurons, spike can even drive the fusion of neurons with neurons or glia, which might contribute to the neurologic symptoms associated with long COVID-19 36 . Furthermore, spike-induced cell-cell fusion may contribute to viral transmission. Since viruses do not need to exit the infected cells, this transmission route allows the virus to escape immune surveillance, particularly antibody-mediated neutralization 28,29 . Intriguingly, spikes from SARS-CoV-2 variants of concern have different fusogenic activities with the Delta spike being the most potent fusogen. Coincidentally, studies have associated the emergence of the Delta variant with an increased risk of COVID-19-related hospitalization 37,38 .
In vitro, spike's fusion-inducing activity can sustain over multiple rounds of fusion reactions despite the fact that the spike concentration at the cell surface is exponentially reduced after each round of fusion. How can the SARS-CoV-2 spike be so e cient at inducing membrane fusion? Given the fast fusion reaction, new protein synthesis is likely unable to compensate for the rapid reduction of the surface spike.
Our data support a model in which HS facilitates an allosteric conformational change in spike-bound ACE2, allowing a conserved linker to promote ACE2 clustering. When clustered ACE2 engages spike on the opposite membrane, these multivalent protein interactions also concentrate the spike ( Figure 5C) while bringing the effector and acceptor membranes together to form the synapse-like structure in preparation for membrane fusion. This model would explain the potent fusogenic activity of the spike despite the constant reduction of its concentration at the cell surface due to cell-cell fusion and other turnover mechanisms ( Figure 7E).
Intriguingly, our results suggest that HS functions predominantly on the ACE2-containing acceptor membrane because HS on spike-containing membranes fails to compensate for the loss of HS on the ACE2 cells. Thus, HS might interact with the spike via a speci c con guration not achievable when positioned in cis to the spike. Furthermore, free HS can promote ACE2 clustering and membrane fusion when added to the medium at high concentrations, suggesting that membrane association may increase local HS concentration and thus enhance its a nity to the spike. Consistent with this idea, HSPG antibody staining did detect HS in small puncta on the cell surface 9 .
Our study establishes PIXN and MTAN as HS-binding drugs targeting speci c sulfate groups in HS.
Although both drugs were initially reported as DNA topoisomerase inhibitors 39  in THF (20 mL) under argon atmosphere was added ethane-1,2-diamine (0.6 mL, 8.97 mmol, 10.0 eq.). The resultant solution was stirred at 50 ℃ for 24 h. The mixture was diluted with methanol (10 mL) and the solvents were removed under vacuum. The crude product was puri ed by column chromatography (30% MeOH: NH 4 OH (9:1) in CH 2 Cl 2 ) to give compound LC-1541 as a blue gum (95 mg, 56%). 1  The following compounds were prepared in a fashion similar to the one described for LC-1541.
LC1539. 1  Chemoenzymatic synthesis of HS 6-mers A total of seven 6-mers were synthesized in this study using the chemoenzymatic synthetic approach 44 . These 6-mers are differed in the number of sulfo groups as well the presence or absences of 2-O-sulfated iduronic acid (IdoA2S) residue. All synthesis was initiated from glucuronide para-nitrophenyl (GlcA-pNP), which is commercially available (Carbosyn). To synthesize the 6-mers without an IdoA2S residue, the synthesis involved the use of heparosan synthase 2 from Pasteurella multocida (pmHS2) and UDP-GlcNAc (or UDP-GlcNTFA, NTFA represents N-tri uoroacetylated glucosamine was incubated with pmHS2 (30 mg) and UDP-GlcNAc (3 mM) in a 100 mL of the reaction buffer containing 25 mM Tris-HCl (pH 7.2), 5 mM MnCl 2 . The reaction was incubated at 37 °C overnight, and the 2-mer product was puri ed by a C-18 reverse phase column. The 2-mer product was further elongated to 3-mer in the 100 mL reaction buffer (pH 7.2) containing UDP-GlcA and puri ed on a C-18 column. The elongation and puri cation steps were repeated until the desired 6-mer was achieved. The nal compound was con rmed for structural identity with Mass Spec and purity was checked with analytical HPLC. The pmHS2 enzyme, UDP-GlcA and UDP-GlcNTFA were made according to the protocol described in a prior publication 45,46 . Additional modi cation steps, including N-sulfation, 6-O-sulfation were completed using N-sulfotransferase and 6-Osulfotransferase isoform 3, respectively 46 . To install an IdoA2S residue, 2-O-sulfotransferase and C5pimerase were employed. The 3-O-sulfation to prepare NS2S6S3S 6-mer, 3-O-sulfotransferanse was used. The purity of the products was con rmed by high resolution anion exchange HPLC, and the molecular weight (MW) was determined by electrospray ionization mass spectrometry. As shown below, the purity of the 6-mers was in the range of 92% to 99%, and the measured MW was very close to the calculated MW (Calc MW). laboratory equipped with advanced access control devices and by trained personnel equipped with powered air-purifying respirators.
Pseudoviral particle entry assay HEK293T-ACE2-GFP cells were seeded in white, transparent bottom 96-well microplates at 20,000 cells per well in 100 µL growth medium and incubated at 37 °C with 5% CO2 overnight (~16 h). The growth medium was carefully removed, and 50 µL PP or PP-containing compounds were added to each well. The plates were then spinoculated by centrifugation at 1500 rpm (453× g) for 45 min and incubated for 24 h (48 h for Calu-3 cells) at 37 °C, 5% CO2 to allow cell entry of PP and the expression of luciferase. After incubation, the supernatant was carefully removed. Then 50 µL/well of Bright-Glo luciferase detection reagent (Promega) was added to assay plates and incubated for 5 min at room temperature. The luminescence signal was measured by a Victor 1420 plate reader (PerkinElmer). For ACE2-GFP cells, the GFP signal was also determined by the plate reader. Data were normalized with wells containing PP but no compound as 100%, wells mock-treated with phosphate buffer saline (PBS) as 0%, and the ratio of luciferase to the corresponding GFP intensity was calculated.

SARS-CoV-2 infection in a 3D EpiAirway model
Human bronchial epithelial cells (HBEC's 3D-EpiAirway™) were seeded into culture inserts for 6-well plates one day before viral infection. Before adding drugs or virus, accumulated mucus from the tissue surface was removed by gently rinsing the apical surface twice with 400 μL TEER buffer. All uids from the tissue surface were carefully removed to leave the apical surface exposed to the air. MTAN was diluted into the assay medium and placed at room temperature before co-treatment with a virus (MOI of 0.1) onto the apical and basal layers for one h. Following one h treatment, the virus was removed from the apical layer.  Assay, Promega). Samples (5 μL) were further diluted with the LDH Buffer (95 μL) and incubated with an equal volume of LDH Detection Reagent. Luminescence was recorded after 60 min incubation at room temperature. As a negative control, we included a no-cell sample in determining the culture medium background. We used tissues treated with the apoptosis-inducing drug bleomycin as a positive control.
ATP-based cytotoxicity assay with Echo/Antiecho-TPPI gradient selection using decoupling during acquisition and multiplicity editing during the selection step.
Thirty-two dummy scans and 20 scans with decoupling during the acquisition period with 1.5 seconds relaxation delay were accumulated, using a matrix size of 2048x256 datapoints.
Bidimensional HSQC-TOCSY data were acquired by using 20 scans per increment using a 2048x256 datapoints matrix with zero-lling in F1 to 2048x2048 points. 1H-13C HSQC-TOCSY were performed using 1.5 seconds relaxation delay and 100 msec mixing time.
For NOE experiments, the samples were prepared by dissolving hexasaccharide (1 mg) and Pixantrone (0.1 mg) in D2O, reaching a molar ratio of hexasaccharide/Pixantrone 2:1. NOESY experiments were performed at 313 K. A total of 24 scans was collected for each free induction decay (matrix 2048x256 points), the data were zero-lled to 2048x2048 points before Fourier transformation, and mixing time values of 120 ms was used.
Drug and protein binding studies HS or HS 6-mers of different concentrations were added to MTAN or PIXN (50uM). After a brief incubation, absorbance was measured from 500 nm to 700 nm by Nanodrop 2000 (Thermo Fisher Scienti c). The change in OD650 was determined and tted into a binding curve using GraphPrism 9.0. Next, samples were rinsed and washed two times for ten minutes each in water and incubated with 1% uranyl acetate overnight at 4 °C. The following day samples were rinsed and washed in water for ten minutes and gradually dehydrated through a graded ethanol series followed by propylene oxide. Samples were then in ltrated in a gradient mix of propylene oxide and resin (Embed 812 resin) before being in ltrated with three changes of pure resin and embedded in 100% resin and baked at 60ºC for 48 hours. Ultrathin sections (65 nm) were cut on an ultramicrotome (Leica EM UT7), and digital micrographs were acquired on JOEL JEM 1200 EXII operating at 80 Kv and equipped with an AMT XR-60 digital camera.
In vitro ACE2 clustering assay To prepare small unilateral vesicles (SUV), we used glass syringes to prepare a DOGS-NTA lipid mixture in a glass vial as follows: rinse a glass vial with chloroform, and then add ~1 mL of chloroform plus individual lipids (POPC, 2.98mg; DOGS-NTA, 0.085mg, PEG-5000 PE, 0.023 mg). The lipid mixture was dried with a stable ow of nitrogen and then in a vacuum desiccator for 2 h. Resuspend the dried lipids in 1.5 mL of PBS and vortex. Transfer the resuspension into two 1.5 mL conical microcentrifuge tubes. Freeze the lipid resuspension in liquid nitrogen and thaw immediately in a water bath at room temperature. Repeat the freeze-thaw for 30 cycles. The cloudy solution will become clear over the freezethaw cycles. The lipid resuspensions were centrifuged at 22,000 × g for 45 min at 4 °C. The SUVcontaining supernatant was collected in a clean tube.
Ibidi u-Slide 8 well Glass Bottom chambers were soaked in 5% Hellmanex III for 24 h (pre-heated to 50 °C) overnight, rinsed extensively with ultrapure water, and air dried. The glass surface was then treated with Imaging processing and statistical data analysis Fluorescence confocal images were acquired by a Nikon CSU-W1 SoRa microscope equipped with a temperature control enclosure and a CO 2 control. 3D or 4D image reconstructions and analyses were done by Imaris software (Licensed to NIH). Fluorescence intensity was analyzed by open-source Fiji software. To this end, images were converted to individual channels, and regions of interest were drawn for measurement. Statistical analyses were performed using either Excel or GraphPad Prism 9.0. P values were calculated by Student's t-test using Excel or one-way ANOVA by GraphPad Prism 9.0. Linear curve tting, nonlinear curve tting, and IC50 calculation were done with GraphPad Prism 9.0. For nonlinear tting, the inhibitor vs. response -variable slope model or the exponential decay model was used. Images were prepared by Photoshop and Illustrator (Adobe). Data processing and reporting are adherent to the community standards.     HS facilitates spike-induced ACE2 clustering in a cell free system (A) A schematic illustration of the in vitro ACE2-clustering assay.