The deletion at the S1/S2 boundary of spike protein propels the virus to enter cells through the endosomal pathway

We observed the same phenomena that others have reported, an instability of the SARS-CoV-2 S1/S2 boundary^9,10,11. Using the patient-isolated SARS-CoV-2 SH01 strain, we performed three rounds of plaque purification in Vero E6 cells in the presence of trypsin and observed no mutations in any of the structural genes (Sfull virus). However, after two additional rounds of passage without trypsin, a 21-nucleotide deletion at the S1/S2 cleavage site was acquired, disrupting the RRAR motif (Fig. 1a and Supplementary Fig. 1a). Unexpectedly, this presumed cell culture adaptation could be prevented by adding trypsin to the media or by ectopically expressing the serine protease TMPRSS2 in Vero E6 cells during a continuous passaging experiment (Supplementary Fig. 1b, c). We designated the plaque-purified deletion clone as Sdel virus and detected no additional mutations in the full-length genome when compared to the Sfull virus (Supplementary Fig. 1d). Compared to Sfull, the deletion-bearing Sdel virus exhibited a dramatic increase in infectivity as measured by the greater percentage of nucleocapsid (N) antigen-positive cells in wild-type Vero E6 (hereafter Vero cells), Vero plus trypsin, Vero expressing TMPRSS2, and A549 cells expressing the receptor ACE2 (Fig. 1b and Supplementary Fig. 2a). Conversely, in human Calu-3 lung epithelial cells, the Sdel virus replicated slower than the Sfull clone (Fig. 1b), similar to previous reports using a pseudovirus or fully infectious, mutant virus^7,9. Moreover, we found that pseudovirus bearing the S protein from Sfull, Sdel, or a RRAR mutant variant (R682S, R685S)¹⁶ had a phenotype similar to infectious viruses used in these cell types (Fig. 1c). Of note, infection using either the Sdel S- or mutant variant S (R682S, R685S)-bearing pseudovirus was decreased by approximately ten-fold in Calu-3 cells, highlighting the critical role of these basic residues at the S1/S2 boundary in infectivity.

a Sequence alignment of spike protein encompassing the cleavage site between S1 and S2 subunits. The spike proteins of SARS-CoV-2 without (Sfull strain) and with (Sdel strain) deletion were used to compare with that of SARS-CoV. The insertion of multi-basic amino acids in spike protein of SARS-CoV-2 was shown in red. b Comparison of the replication property between Sfull and Sdel strains in different cell lines. The percentage of nucleocapsid (N) protein-positive cells was analyzed by imaging-based analysis following virus infection (two or more experiments; n = 6 except for Calu-3 in which n = 8; one-way ANOVA with Dunnett’s test; mean ± s.d.). c Evaluation of entry efficiency in different cell lines infected with pseudoviruses bearing spike protein Sfull, Sdel, or S mutant (R682S, R685S). Data are normalized to the Sfull of individual experiments (two experiments; n = 6; one-way ANOVA with Dunnett’s test; mean ± s.d.). d Effect of TMPRSS2 serine protease inhibitor Camostat and cysteine protease inhibitor E-64d on Sfull or Sdel infection in different cell lines (two experiments; n = 4 or 6; one-way ANOVA with Dunnett’s test; mean ± s.d.). Data shown were normalized to the untreated group of individual experiments. ****P < 0.0001; n.s. not significant.

To assess the impact of the S1/S2 junction deletion on viral entry pathways, cells were treated with camostat mesilate, a TMPRSS2 inhibitor that blocks viral fusion at the plasma membrane, and/or E-64d (aloxistatin), an inhibitor that blocks the protease activity of cathepsins B and L, which are required for the endosomal membrane fusion (Fig. 1d and Supplementary Fig. 2b). We observed apparent S1/S2 cleavage for Sfull virus but not for Sdel in multiple cell types (Supplementary Fig. 3a). Sfull virus infection, as measured by N antigen-positive cells, was sensitive to inhibition by E-64d but not camostat in Vero cells (Fig. 1d), suggesting that Sfull virus enters the TMPRSS2-negative Vero cells through the endosomal pathway. When TMPRSS2 was expressed, both camostat and E-64d inhibited the infectivity of Sfull, indicating that expression of TMPRSS2 could promote the membrane fusion entry pathway. Remarkably, E-64d and camostat had no effect on Sfull virus in A549-ACE2 cells, suggesting that in this cell Sfull may use other TMPRSS2 homologs or trypsin-like proteases to activate fusion at the plasma membrane since TMPRSS2 expression is absent in A549 cells^17,18. We observed a similar phenotype even when cells were treated with a high concentration of inhibitors (Supplementary Fig. 2c). In Calu-3 cells that express the TMPRSS2, camostat completely blocked the Sfull infection, but E-64d had minimal effects, suggesting that Sfull preferentially enters Calu-3 cells via the plasma membrane fusion pathway.

For the Sdel virus, E-64d significantly inhibited infection in Vero, Vero-TMPRSS2, and A549-ACE2 cells, whereas camostat did not reduce the infection, even in Vero-TMPRSS2 cells (Fig. 1d). It is noteworthy that Sdel was sensitive to both inhibitors in Calu-3 cells unlike the Sfull virus, and these two compounds exerted a synergetic effect on Sdel infection. This suggests Sdel utilizes both plasma membrane and endosomal fusion pathways in Calu-3 cells. Notably, the spike protein of SARS-CoV does not have the insertion of multiple basic residues at the S1/S2 cleavage site (Fig. 1a). We found that, E-64, but not camostat, efficiently inhibited SARS-CoV pseudovirus infection in multiple cell types (Supplementary Fig. 2d). In all, these results demonstrate that: (a) the Sfull virus with intact spike protein are endocytosed into the TMPRSS-negative cells (e.g., Vero), (b) the deletion at the S1/S2 junction site propels the virus to enter cells through the endosomal pathway (e.g., A549 and Calu-3), (c) the more efficient endosomal entry is acquired by the deletion at the S1/S2 site (e.g., Vero and A549), and (d) the entry pathways of SARS-CoV may resemble the Sdel virus. So far, we fully characterized the ensodomal entry properties of Sdel virus, which could be served as a model to dissect the entry process in specific cell types.

Genome-wide CRISPR/Cas9 screen identifies entry factors using Sdel virus as model

Genome-wide CRISPR/Cas9 screens have enabled the identification of host factors required for efficient virus infection^{19,20,21,22,23}. A lack of suitable human physiologically relevant cell lines and the spike protein-induced syncytia formation in cells have made such a screen to identify the host factors that regulate the entry of SARS-CoV-2 very challenging. We found that Sdel virus preferentially enters A549-ACE2 cells via the endosomal pathway, replicates robustly, does not cause syncytia, and efficiently results in cell death. Because of these properties, the Sdel virus represented an ideal virus model to investigate the endosomal entry process. To this end, we performed a genome-wide, cell survival-based screen with the Sdel virus in A549-ACE2 cells transduced with a library of single-guide RNAs (sgRNAs) targeting 19,114 human genes (Fig. 2a)²⁴. The vast majority of transduced cells inoculated with Sdel virus died within 7 days of infection. Surviving cells were harvested and expanded for a second round of challenge with Sdel. The remaining surviving cells were expanded and subjected to genomic DNA extraction, sgRNA sequencing, and data analysis (Supplementary Data 1).

The top candidates from the CRISPR screen were determined according to their MAGeCK score (Fig. 2b). The top hit was ACE2, the cellular receptor that confers susceptibility to SARS-CoV-2, which confirmed the validity of the screen. Additionally, the gene encoding cathepsin L (CTSL), a target of our earlier assay using E-64d that is known to be important for activating SARS-CoV virion membrane fusion with the endosome²⁵, also was identified, again confirming the utility of the screening strategy.

We chose the top 32 genes with a cutoff of false discovery rate (FDR) <0.15 to validate. For each specific gene target, A549-ACE2 cells were transduced with two independent sgRNAs and then infected with Sdel virus. The percentage of N protein-positive cells was determined by image-based analysis. Remarkably, editing of all 32 genes resulted in a statistically significant reduction of Sdel infection compared to cells receiving the control sgRNA (Fig. 2c). Most of these genes were associated with the endolysosome, including components of the retromer complex, the COMMD/CCDC22/CCDC93 (CCC) complex, Wiskott–Aldrich syndrome protein and SCAR homologue (WASH) complex, and actin-related protein 2/3 (Arp2/3) complex, which have significant roles in endosomal cargo sorting^26,27. We also identified genes encoding the WD Repeat Domain 81 (WDR81)–WDR91 complex, which was detected in a previous genetic screen for regulators of endocytosis and the fusion of endolysosomal compartments²⁸. Similarly, we identified the gene encoding Transcription Factor Binding To IGHM Enhancer 3 (TFE3), which may regulate lysosomal positioning in response to starvation or cholesterol-induced lysosomal stress²⁹. We also validated NPC Intracellular Cholesterol Transporter 1 (NPC1) and NPC2, which regulate intracellular cholesterol trafficking, as important for Sdel infection^30,31. In addition, the gene for Activating Signal Cointegrator complex 3 (ASCC3), which functions as a negative regulator of the host defense response, was identified in our screen³². From these hits, we selected representative genes to validate for cell-type specificity in HeLa-ACE2 cells, finding that all the genes tested greatly reduced infection with Sdel virus (Supplementary Fig. 4a).

To define the stage of viral infection that each of the 32 validated genes acted, one representative sgRNA per gene was selected for study in A549-ACE2 cells. Due to its known antiviral activity, ASCC3 was not targeted. We confirmed that editing of these genes did not affect cell viability (Supplementary Fig. 4b). The gene-edited cells were infected with pseudovirus bearing the spike protein of Sdel virus or, as a control, the glycoprotein of vesicular stomatitis virus (VSV-G) (Fig. 3a, b). Consistent with data from the fully infectious Sdel virus, editing any of the selected genes markedly inhibited Sdel pseudovirus infection whereas only editing of some retromer-associated genes and the Arp2/3 complex significantly reduced the VSV-G pseudovirus infection (Fig. 3a, b). These results suggest that most of the genes identified act at the virus entry step and are specific to SARS-CoV-2. Notably, pseudovirus bearing the spike protein of SARS-CoV, which lacks the multiple basic residues at the S1/S2 junction as Sdel, exhibited a phenotype similar to Sdel pseudovirus and Sdel live virus (Fig. 3c). Editing of these genes, including those encoding CTSL, cholesterol transporters NPC1/2, WDR81/91, and TFE3, markedly reduced infection, suggesting that Sdel and SARS-CoV may utilize similar entry machinery in this cell type (Fig. 3c). Intriguingly, these edited genes also significantly reduced the infection by pseudovirus bearing the spike protein of Middle East Respiratory Syndrome coronavirus (MERS-CoV) in A549-ACE2-DPP4 cells (Fig. 3d). Although the furin cleavage site is present at the S1/S2 boundary of MERS-CoV³³, it preferentially enters the A549 cell via endosomal pathway as indicated by its sensitivity to E-64d inhibitor (Supplementary Fig. 2e). This is possibly due to the lack of proper protease to activate the plasma membrane fusion pathway in A549 cells for MERS-CoV. In sum, these results indicate that the host genes identified from CRISPR screen are required for the entry in A549 cells by pseudotyped SARS-CoV and MERS-CoV, and SARS-CoV-2 with deletion at the S1/S2 boundary.

To determine whether these genes identified impact Sfull virus infection, one representative sgRNA per gene was tested (Fig. 3e). Given the low infection efficiency with pseudotyped virions bearing the spike protein of Sfull virus, we only used the Sfull live virus to validate these genes. The editing efficiency of some of these genes by sgRNAs was confirmed by western blotting (Supplementary Fig. 3b). As expected, editing of CTSL did not reduce infection (Fig. 3e), as the Sfull virus enters A549-ACE2 cells via an endosomal-independent pathway (demonstrated in Fig. 1d). Also, editing of NPC1, NPC2, or TFE3 that functions in endolysosomes had a negligible impact on Sfull virus infection (Fig. 3e). In general, editing of genes encoding complexes that regulate the retrieval and recycling of cargo significantly reduced infection, albeit to a lesser extent than observed with the Sdel live virus (Fig. 3e). U18666A, a cationic sterol, binds to the NPC1 protein to inhibit cholesterol export from the lysosome, resulting in impaired endosome trafficking, late endosome/lysosome membrane fusion^34,35,36. U18666A has been shown to inhibit the S protein-driven entry of SARS-CoV, MERS-CoV, and the human coronaviruses NL63 and 229E, with the most efficient inhibition observed with SARS-CoV³⁷. The antiviral effect of U18666A on type I feline coronavirus has also been characterized in vitro and in vivo^38,39. We found that, pretreating A549-ACE2 cells 2 h prior to or post infection had no inhibitory effect on Sfull virus (Fig. 3f). In contrast, Sdel virus was more sensitive to U18666A, even when used for treatment 2 h post infection, presumably due to Sdel preferential usage of the endosomal entry pathway in A549 cells, which is consistent with the result of its sensitivity to E-64d as demonstrated in Fig. 1d.

Host genes that regulate the surface expression of receptor ACE2 are identified

The Sdel-validated genes that also affected Sfull infectivity were largely multi-protein complexes (Figs. 2c and 3e). These complexes are important for maintaining plasma membrane and lysosomal homeostasis by maintaining expression of key integral proteins, including signaling receptors and transporters^26,40. We hypothesized that disruption of these complexes might affect the binding or transit of virions. To this end, we performed binding and internalization assays using Sfull virus in A549-ACE2 cells. The genes COMMD3, VPS29, and CCDC53, which encode proteins that comprise CCC, retromer, and WASH complexes, respectively, were each edited; effects on expression were confirmed by western blotting except for the VPS29 (Supplementary Fig. 3b). Notably, binding and internalization of Sfull virions to these gene-edited bulk cells was significantly decreased when compared to the cells edited with control sgRNA (Fig. 4a).

a The effect on virion binding and internalization in gene-edited cells. A549-ACE2 cells were incubated with SARS-CoV-2 Sfull infectious virus on ice for binding or then switched to 37 °C for internalization. Viral RNA was extracted for RT-qPCR analysis (two experiments; n = 4; one-way ANOVA with Dunnett’s test; mean ± s.d.). b, c Surface expression of receptor ACE2 was decreased in gene-edited cells as measured by flow cytometry using S1-Fc recombinant protein or anti-ACE2 antibody (2 experiments; n = 7 (b) or 6 (c); one-way ANOVA with Dunnett’s test; mean ± s.d.). d, e Surface and total expression of receptor ACE2 were decreased in gene-edited cells. The plasma membrane proteins were biotin-labeled and immunoprecipitated by streptavidin beads for western blotting. One representative blot was shown (d) and data are pooled from four independent experiments, quantified, and normalized to the controls of individual experiments (e) (four experiments; n = 4; one-way ANOVA with Dunnett’s test; mean ± s.d.). f, g The impact on viral production in CCDC53 gene-edited Calu-3 cells. The mixed cell population was infected with Sfull (f) or Sdel (g) to assess the virus yield (two experiments; n = 6; two-way ANOVA with Sidak’s test). *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001; n.s. not significant.

The entry receptor ACE2 is critical for SARS-CoV-2 infection. To determine whether cell surface expression of ACE2 is regulated by these complexes, gene-edited cells (COMMD3, VPS29, VPS35, CCDC53, CCDC22, and NPC1) were incubated with S1-Fc recombinant protein or an anti-ACE2 antibody, and binding was measured by flow cytometry (Fig. 4b, c). It showed that editing of these genes apparently perturbed the surface expression of ACE2, with the exception of the cholesterol transporter gene NPC1. To further confirm these findings, we biotinylated the surface proteins of these gene-edited cells, immunoprecipitated with streptavidin, and performed western blotting and quantification (Fig. 4d, e). A significant reduction of surface ACE2 was observed across the different gene-edited cells except for NPC1-edited cells, consistent with the data above from flow cytometry-based surface binding assays. To correlate the significance of this finding for virus infection, we edited CCDC53, which had the greatest reduction in virion internalization as shown in Fig. 4a, in Calu-3 lung cells. Viral yield was approximately ten-fold lower in the CCDC53-edited bulk cells compared to control sgRNA-edited cells at 24 h for Sfull and 48 h for Sdel (Fig. 4f, g). These results suggest retrieval and recycling complexes identified in our screen regulate expression of the ACE2 receptor, which is required for optimal SARS-CoV-2 infection.

SARS-CoV-2 entry is elegantly regulated by endosomal cargo sorting complexes

To distinguish the complexes important for virus infection, we edited additional genes. The retriever complex is another retromer-like complex that mediates cargo recycling and consists of the genes DSCR3, C16orf62, and VPS29⁴¹. VPS29 and C16orf16 that were identified in our screen, also are shared functionally by the retromer and CCC complexes^42,43. Sorting Nexin 17 (SNX17) acts as a cargo adaptor associated with retriever and the adaptor SNX31 (ref. ⁴¹). SNX27 and SNX3 are two additional cargo adaptors associated with the retromer complex⁴⁴. To test these genes, which were not identified in our screen, we introduced three sgRNAs per gene in A549-ACE2 cells and infected with Sdel virus. The editing efficiency of SNX17 and SNX27 was confirmed by western blotting (Supplementary Fig. 5). Among the genes chosen for examination, only the retromer-associated adaptor SNX27 was required (Supplementary Fig. 5), highlighting the importance of the retromer complex over the retriever one for virus infection.

The COMMD proteins of CCC complex are a 10-member family (COMMD1-10)⁴⁵ that act as cargo-binding adaptors^46,47. Of these 10 proteins, we identified the genes encoding all of them in our screen except for COMMD1, 6, and 9 (Fig. 2c). Knockout of the COMMD1, 6, and 9 increases the low-density lipoprotein cholesterol levels in the plasma membrane, thereby maintaining lipid raft composition⁴⁸. In our experiments, editing each of these three genes as well as cholesterol uptake-related genes (LDLR, SRB1, CD36, LRP1) did not impact Sdel infection in A549-ACE2 or HeLa-ACE2 cells (Supplementary Fig. 6a, b), suggesting that these members of the COMMD protein family function differently. Notably, knockout of COMMD1 did not affect expression of COMMD3 or CCDC22 in our study as opposed to previous work (Supplementary Fig. 6c)^46,48, which needs to be further characterized. Overall, our experiments demonstrate that SARS-CoV-2 entry is regulated by endosomal cargo sorting complexes. Understanding how these complexes regulate the sorting of incoming virions might enable development of host-directed antiviral agents to control COVID-19.

The switch of virion entry pathways modulates the infection and transmission in hamsters

In the culture of A549 lung epithelial cells, we demonstrated that the deletion at the S1/S2 boundary of spike protein of SARS-CoV-2 resulted in a switch from the plasma membrane to endosomal fusion pathway for entry. Using this model, we uncovered a suite of host genes that regulate the virion endosomal entry and surface expression of receptor ACE2. In Calu-3 lung cells, which model more physiologically relevant airway epithelial cells, this switch led to a less efficient entry process. Since virus entry is the first step in establishing infection, we hypothesized that deletion at the S1/S2 boundary propelling the viral entry to endosomal pathway might reduce virus infectivity and transmissibility in vivo.

Indeed, using the golden Syrian hamster model, a previous study showed that a SARS-CoV-2 variant with a 30-nucleotide deletion at the S1/S2 junction caused milder disease and less viral infection in the trachea and lungs compared to a virus lacking the deletion⁹. Here, we extended the study and systemically evaluated the tissue tropism and transmissibility. Following intranasal inoculation of golden Syrian hamsters, nasal turbinates, trachea, lungs, heart, kidney, spleen, duodenum, brain, serum, and feces were collected. Sfull virus replicated robustly and reached peak titer at day 1 post infection, with a mean titer 31-, 126-, and 1259-fold higher than Sdel in the turbinates, trachea, and lungs, respectively (Fig. 5a). While Sdel virus replication was delayed, no significant differences were observed by day 4 in these three tissues. At days 2 and 4, five pieces of fresh feces were collected from each hamster. Although no infectious virus was detected by focus-forming assay, viral RNA levels were higher in fecal samples for Sfull (20- and 40-fold) than Sdel at days 2 and 4, respectively (Fig. 5b). Likely related to this, no infectious virus was detected in the duodenum, and Sfull RNA was 6.3-fold higher than Sdel at day 4 (Supplementary Fig. 7a). In serum, we detected no difference in viremia at day 1, but Sfull RNA was 63- and 32-fold higher than Sdel at days 2 and 4, respectively (Supplementary Fig. 7b). In other extrapulmonary organs, infectious virus was not consistently detected. In general, brain tissue had the highest viral RNA copy number, and all organs showed higher levels of Sfull RNA at day 2 or 4 compared to Sdel except for the liver and kidneys (Supplementary Fig. 7c–g). Body weight of six hamsters challenged with Sfull or Sdel virus were monitored daily, and the weight loss was only observed in hamsters inoculated with Sfull and decreased as much as ~18% at days 5 and 6 (Supplementary Fig. 7h).

a Viral load in the tissues of nasal turbinate, trachea, and lung. Tissues were harvested at days 1, 2, and 4 post challenge of Sfull or Sdel virus (n = 6 per day). b Viral RNA in fecal samples. Fresh fecal samples were collected at days 2 and 4 post infection of Sfull or Sdel strain (n = 6 per day) for qRT-PCR. c Transmission of Sfull or Sdel strain in hamsters by direct contact exposure. Naïve hamsters (n = 6) were each co-housed with one inoculated donor at day 1 for 3 days. Hamsters were sacrificed and the indicated tissues were harvested for titration. The dashed lines represent the limit of detection by focus-forming assay. Median viral titers (a–c): two-tailed Mann–Whitney test; *P < 0.05; **P < 0.01; n.s. not significant. d H&E staining of lung sections of contact hamsters. Representative images are shown from n = 6 hamsters. Scale bar, 100 μm. e, f RNA ISH of lung and nasal turbinate sections of contact hamsters. Representative images are shown from n = 6 hamsters. Scale bar, 100 μm. g Model of the role of S1/S2 boundary and host factors in regulating cell entry, pathogenicity, and transmissibility of SARS-CoV-2. SARS-CoV-2 enters cells via two pathways. The virus (Sfull) with intact spike protein preferentially enters cells at the plasma membrane (early entry pathway) in airway epithelial cells (Calu-3) or respiratory tract tissues expressing the proteases (e.g., TMPRSS2) to activate the membrane fusion. The deletion at S1/S2 junction site in spike (Sdel), however, propels the virus to enter cells through the endosomal pathway (late entry pathway), which is less efficient than the fusion pathway at the plasma membrane. Host factors such as CTSL, NCP1/2, WDR81/91, and TFE3 are critical for the endosomal entry. Both entry pathways are initiated with virion binding to receptor ACE2 at the cell surface that is regulated by host factors including retromer, CCC, and WASH complexes, etc. The more efficient early entry pathway in respiratory tract with intact spike protein than the late pathway promotes virus production, pathogenesis, and transmission in a hamster model. The SARS-CoV with spike lacking the insertion of multi-basic amino acids may resemble the Sdel virus and enter cell less efficiently than SARS-CoV-2 resulting in relatively low transmissibility.

To determine the impact of deletion at the S1/S2 junction on transmissibility by direct contact exposure, six hamsters were inoculated intranasally with Sfull or Sdel virus. At 24 h post inoculation, each donor hamster was transferred to a new cage and co-housed with one naïve hamster for 3 days. For donors (day 4 post inoculation), tissue samples were processed (Fig. 5a, b and Supplementary Fig. 7). For contact hamsters (day 3 post-exposure), nasal turbinate, trachea, and lungs were collected for infectious virus titration and histopathological examination. The average titers in turbinate, trachea, and lungs from Sfull-exposed hamsters reached 6.6, 6.2, and 6.1 logs, respectively (Fig. 5c). Unexpectedly, no infectious virus was detected in these three tissues from Sdel-exposed hamsters (Fig. 5c). In lung sections from hamsters that were exposed to Sfull-infected animals, we observed mononuclear cell infiltrate, protein-rich fluid exudate, hyaline membrane formation, and hemorrhage (Fig. 5d). In contrast, no or minimal histopathological change was observed in the lung sections from hamsters that were exposed to Sdel-infected animals (Fig. 5d). To examine viral spread in the lungs, we performed RNA in situ hybridization (ISH). Viral RNA was clearly detected in bronchiolar epithelial cells in hamsters exposed to Sfull-infected animals (Fig. 5e) whereas it was rarely detected in hamsters exposed to Sdel-infected animals. Similarly, abundant RNA was observed in the nasal turbinate epithelium (Fig. 5f). These results indicated that transmission of Sfull from infected hamsters to co-housed naïve hamsters was efficient whereas the deletion at the S1/S2 boundary in the S protein of Sdel markedly reduced transmission.