Single-Virus Fusion Measurements Reveal Multiple Mechanistically Equivalent Pathways for SARS-CoV-2 Entry

The severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) betacoronavirus has spread globally since late 2019, causing over 600 million confirmed infections and over 6.5 million deaths as of November 2022 (1). As with many coronaviruses, it binds to a cell surface receptor and is activated for membrane fusion and cell entry by proteolytic cleavage of its spike protein by a host protease. The primary receptor for SARS-CoV-2 binding, as with SARS-CoV binding, is ACE2 (2–6). This is a common, but not universal, binding receptor among betacoronaviruses infecting humans (7); Middle East respiratory syndrome coronavirus (MERS-CoV) utilizes DPP4 as a means of host-cell attachment (8). However, the biochemical steps following viral attachment and their roles in cell entry and physiological infection vary more substantially. Canonical descriptions of viral glycoprotein-mediated fusion typically use the terms “priming” for proteolysis that leaves the glycoprotein in a fusion-competent form and “triggering” for receptor-binding or protonation events that trigger a fusogenic conformational change. As we will discuss, SARS-CoV-2 has at least two proteolysis events: one cleaving the spike (S) protein to form S1 and S2 and one that cleaves S2 to form a fusion-active S2′ fragment. The sequencing of the final proteolysis and receptor binding may be flexible, as our data suggest. We therefore will follow an alternate convention in the literature to use “activation” for the second proteolysis step (9).

The host proteases responsible for activation and the subsequent subcellular location of viral membrane fusion vary substantially among betacoronaviruses (10). SARS-CoV was canonically thought to utilize cathepsins present in late endosomes for activation, entering via the endocytic pathway (11–13), although it can also enter at the cell surface (14, 15) in a TMPRSS2-dependent fashion, likely including in respiratory epithelial tissues (16). MERS-CoV can be activated by TMPRSS2 (17, 18), permitting cell surface entry, although infectivity requires two proteolytic steps, the first of which may be furin mediated (19–21). Initial data on SARS-CoV-2 have indicated the potential for each of these activation and entry mechanisms (6, 22–27), with some ensuing debate about the biochemical and physiological relevance of cell surface versus endosomal entry and whether this depends on cell type. In particular, variations in the efficiency with which SARS-CoV-2 variants of concern infect lung parenchyma, bronchi, and nasal epithelial tissues have been correlated with differential protease expression, subsequent transmissibility, and severity of disease (28–30).

The site of host proteolytic cleavage limits how early a virus can undergo fusion and entry, but it does not by itself establish the site of entry. A second trigger may be required, as in the case of Ebola (11, 31, 32) the subcellular compartment where proteolysis occurs may not be permissive for entry. Targeted proteolytic inhibitors and inhibitors of intracellular trafficking provide information on the sites of entry (6, 33), but they also potentially perturb membrane composition. An alternative is to biochemically isolate subcellular compartments and exogenously trigger fusion (34). This permits more biochemically precise and controlled testing of cellular requirements for viral entry.

Studying viral entry in reconstituted systems has yielded substantial insight into the biochemical requirements for entry, including the effects of membrane composition and receptor chemical structure (34–42). The viral membrane itself does appear important to the mechanism, whether via distinct composition or distinct spike protein organization, as both coronavirus and influenza virus cell-cell fusion have well-described differences from virus-cell fusion (43–45). Single-virus assays have enabled an added level of mechanistic sophistication, as they permit analysis of heterogeneity among viral particles and straightforward estimates of stoichiometry from single-event waiting times (36, 38, 40, 46–48). As a complement to biochemical reconstitution, single-virus fusion studies on isolated cellular membranes held under exogenous control (34, 36, 49) permit both the control of reconstituted systems and the physiological membrane environment of cells. This is the approach reported here.

Viral membrane fusion is a mechanistically complex process, involving multiple steps of activation and rearrangement of both the viral fusion proteins and the interacting membranes (50–52). Under normal conditions, the activation and rearrangement of the fusion proteins are believed to be rate limiting for viruses where this has been studied in detail (35, 40, 53, 54). One key mechanistic parameter is the apparent stoichiometry, or the number of viral fusion proteins required to achieve fusion. This is important because if we assume that there exists some free energy barrier to viral membrane fusion and all participating copies of the fusion protein are equivalent, then each protein contributes the same free energy to overcoming the barrier and achieving fusion (55). The required protein stoichiometry thus reports on both the underlying barrier to fusion and the contribution each fusion protein makes: if either of these changes, the protein stoichiometry will change accordingly. This is still true if the above assumptions are not strictly correct: should fusion proteins differ in their energetic contribution, the apparent stoichiometry will still reflect any changes to the fusion protein contributions. Stoichiometry can be estimated from single-virus fusion kinetics using a number of techniques (38, 46, 56) and thus provides a key parameter for assessing viral entry mechanisms. Because factors such as virus-to-virus heterogeneity can complicate these estimates (57), a more stringent alternate comparison is to assess the statistical agreement between cumulative distribution functions for fusion, plots where the fraction of particles fused is assessed versus time. Both techniques are used in this study.

Here, we report single-virus fusion studies of SARS-CoV-2 pseudovirions with isolated plasma membrane and controlled exogenous protease treatment. We show that SARS-CoV-2 spike protein can be activated for fusion by a diverse range of host proteases present in the extracellular environment, on the cell surface, and within endosomes. We also show that the plasma membrane is permissive for viral entry through the point of membrane fusion and that additional endosomal factors are not required for membrane fusion. This supports an “opportunistic” model of SARS-CoV-2 entry, where protease activation can occur at several different stages of viral transport and can lead to fusion with whichever cellular membrane is present at the time.

To measure single-virus fusion mediated by SARS-CoV-2 spike proteins, we used an approach previously developed for influenza and Zika viruses (48, 58) and highly similar to related assays for HIV and coronavirus fusion (36, 49). Briefly (Fig. 1), target membranes—in this case host cell plasma membrane vesicles—are immobilized in a microfluidic flow cell. Virus is labeled with a lipid-phase dye at a self-quenching concentration, protease activated, added to the flow cell, and allowed to bind and fuse. Fusion is assessed via lipid mixing between the virus and the host cell membranes, detected as fluorescence dequenching in optical microscopy.

In this case, we used three different pseudovirus systems. Pseudoviruses were expressed on HIV, vesicular stomatitis virus (VSV), and murine leukemia virus (MLV) backgrounds. All of these pseudoviruses infected Vero cells successfully (see Fig. S1 in the supplemental material) but did not carry SARS-CoV-2 genomes and thus could not replicate. Host cell plasma membranes were isolated by blebbing giant plasma membrane vesicles from cultured cells (59), biotinylating these membranes, and then immobilizing them in a flow cell using poly-l-lysine (PLL)-polyethylene glycol (PEG)-biotin and streptavidin as previously reported (55). Because we observed proteolytic activation of SARS-CoV-2 spike protein to be relatively slow (Fig. S2 and S3), fusion assays were performed by preincubating viral particles with the protease of interest and then allowing the activated particles to bind. Cleavage states of the spike protein were assessed via anti-S2 immunoblotting (Fig. S3). In our immunoblots of MLV and HIV as well as previously published immunoblots of the VSV pseudoviruses provided for use in this study (60), a small amount of the spike protein was initially in the S0 form, but the majority was in the cleaved S1/S2 form. Activation of spike protein by further cleavage of S2 to form an S2′ or similar fragment can be observed in the immunoblots. Additional cleavage products likely result from secondary inactivating cleavage events (3). Secondary cleavage events make it challenging to quantitate the fraction of S2′ versus S2, but a robust fusion is observed with a substantial fraction of S2 spike remaining, suggesting that a relatively small fraction of S2′ cleaved spikes on a (pseudo)virion may be sufficient to support fusion. We also observed activation and fusion without preincubation when membrane-bound proteases are present on the target membranes and proteolysis occurs subsequent to receptor binding (Fig. 2). Prior studies have shown greater rates of S2′-forming proteolysis after ACE2 binding, likely due to conformational exposure of the cleavage state (61). Our results are entirely compatible with these findings and simply indicate that with sufficient protease exposure, S2′ formation and fusion activation can occur prior to ACE2 binding. This has been subsequently confirmed by surrogate receptor assays that replace ACE2 entirely and still observe protease-dependent fusion, although the rates of fusion are accelerated by addition of soluble ACE2 in trans (62).

For testing of exogenous proteases, Vero cell plasma membrane vesicles were used due to their low levels of TMPRSS2 expression yet good ACE2 receptor expression. As a control, plasma membrane vesicles from HEK293T cells that express low levels of ACE2 were immobilized in microfluidic flow cells and HIV pseudovirus allowed to bind in a fashion identical to Vero or Calu-3 plasma membrane vesicles. Twenty-one-fold fewer pseudoviruses remained after washing, showing a strong dependence on ACE2 expression (Fig. S4). Furthermore, minimal fusion (<0.5% of particles dequenching) was observed to Vero cell membranes in the absence of exogenous protease (Fig. S5), while more fusion was observed to Calu-3 cell membranes with accompanying higher levels of TMPRSS2 expression (Fig. 2). Prior work has shown that ACE2 mRNA levels are ~7-fold higher in Calu-3 cells than in Vero cells and >10-fold higher than in Vero E6 cells (63).

Analysis of fusion kinetics by proteolytically activated pseudovirions is made challenging by the fact that proteolytic activation and binding are substantially slower than subsequent fusion: for HIV pseudovirions, the time scale for activation is 4 to 5 min (Fig. S2), while the time scale for fusion is 71 s. We corrected for this by taking virions preincubated with protease that then show distinct binding and fusion events on fluorescence microscopy (Fig. 1). These selection criteria may bias the kinetic analysis by not including viruses where binding and fusion occur within the same frame (<1 s), but this provides a lower limit for fusion. Additional data on total fluorescence change from lipid mixing integrated across the 133- by 133-μm microscope field of view are given in the supplemental material, and representative fields of view are shown in Fig. S6 and movies available at https://doi:10.5281/zenodo.5718787. Cumulative distribution functions were calculated for trypsin-activated fusion by SARS-CoV-2 spike protein on each of the three pseudovirus genetic backgrounds. These results (Fig. 3) show the fastest fusion on a VSV background and the slowest on an MLV background, with HIV-based pseudovirions intermediate. Single-virus waiting times were statistically different between VSV and either MLV or HIV (P ≤ 0.001 via 2-tailed Kolmogorov-Smirnov test with Bonferroni correction) but not between HIV and MLV backgrounds (P = 0.69; 2-tailed Kolmogorov-Smirnov test). Furthermore, fusion of HIV pseudovirions to Vero plasma membrane vesicles after preincubation with trypsin shows indistinguishable kinetics from fusion of these pseudovirions to Calu-3 plasma membrane vesicles without protease preactivation (P > 0.5; 2-tailed Kolmogorov-Smirnov test). However, pseudovirions did not fuse to Vero plasma membrane vesicles in the absence of protease treatment. These data suggest that proteolytic activation by TMPRSS2 on Calu-3 cell membranes is not rate limiting in this case.

Plasma membrane vesicles are somewhat heterogeneous in size. We took advantage of this to determine the relationship between target membrane curvature and fusion kinetics. Prior work (64) has established that the total fluorescence intensity of a membrane-incorporated dye is proportional to vesicle surface area, so the square root of this intensity is proportional to vesicle radius. Labeling of Vero cell plasma membranes with DiO prior to vesicle production and measurement of lipid-mixing kinetics yielded an assessment of this relationship on a single-vesicle, single-virus level as we have done previously for influenza virus fusion (65). These measurements showed no correlation (Spearman rank correlation of 0.27) between plasma membrane vesicle curvature and time to lipid mixing (Fig. 4).

Additionally, to verify that endosomal maturation is not required for productive infection, we incubated Vero cells with (i) SARS-CoV-2 virus-like particles (VLPs) or (ii) trypsin-activated SARS-CoV-2 VLPs in the presence of 5 nM bafilomycin A, a concentration previously shown to inhibit endosomal entry of SARS-CoV-2 pseudovirus into Vero cells with minimal cytotoxicity (66). As assessed by bulk luciferase assays (Fig. 5; Fig. S7), trypsin-treated particles underwent some inactivation, but the activated ones became independent of bafilomycin A inhibition in TMPRSS2^low Vero E6 cells. The majority of trypsin activation occurred prior to ACE2 binding, as demonstrated when the serine protease inhibitor aprotinin was added to trypsinized pseudovirus prior to incubation with cells. Aprotinin was added at a final concentration of 30 μM; tests using equal amounts of trypsin and a fluorogenic substrate showed that the 50% effective concentration (EC₅₀) for aprotinin in this situation is below 0.1 μM. This demonstrates that productive virus-like particle infection does not necessarily require endosomal acidification or the activity of late-endosomal proteases. Our data are highly concordant with the results previously reported using infectious SARS-CoV-2, where pretreatment of virus with trypsin overcame the inhibitory effects of blocking endosomal acidification (67). Differing results have been reported on this in the literature, including a demonstration using single-virus tracking that inhibition of endocytosis could arrest infection by VSV pseudoviruses (68). The subtle differences between experimental conditions used in these and the single-virus tracking experiments may yield future insight into SARS-CoV-2 biology: some of these may include exposure to protease prior to ACE2 receptor binding, potential alterations in plasma membrane composition and organization induced by dynasore treatment (69) to inhibit endocytosis in the single-virus tracking study, or differences among the viral and pseudoviral constructs used.

Because pseudoviruses have a different budding and assembly process from infectious SARS-CoV-2, we also compared single-virus lipid-mixing kinetics of VLPs formed using S, M, N, and E proteins from SARS-CoV-2 (70). Such particles undergo an assembly process similar to infectious virus and bud from the endoplasmic reticulum (71) Single-virus lipid-mixing kinetics by SARS-CoV-2 VLPs (Fig. S8) were slightly slower but not statistically different from those of MLV and HIV pseudovirus fusion; VSV pseudoviruses underwent lipid mixing significantly faster than all three other types of particles (P < 0.001; Kolmogorov-Smirnov test with Bonferroni correction).

In most viral fusion systems studied thus far, lipid mixing has been best described as a stochastic process requiring N independent kinetic steps, where each independent step represents the activation of one fusion protein (38, 40, 55, 72, 73). The minimum number of fusion proteins in the most-likely pathway for fusion can then be calculated using the inverse of the normalized variance for the single-event fusion time: N_min = <T>²/var(T) for lipid-mixing times T (57, 74). The apparent fusion protein stoichiometry for SARS-CoV-2 spike fusion is between 1 and 3: N_min values were calculated at 0.89 (95% confidence interval [CI], 0.61 to 1.6) for HIV pseudoviruses, 1.8 (95% CI, 1.2 to 3.4) for MLV pseudoviruses, and 0.29 (95% CI, 0.12 to 0.84) for VSV pseudoviruses. Interestingly, both HIV and VSV pseudoviruses show N_min values of <1. This has previously been discussed in the context of dynamic disorder or fluctuations in single-molecule reaction rates (46, 57, 75) or as a consequence of nonlinear reaction pathways (74, 76). In this case, we hypothesize that the observations are due to static disorder or heterogeneity: multiple populations of pseudovirus that have different fusion rates due to different spike protein densities or morphologies. Immunostaining of pseudovirions with antispike antibodies can explain the difference between MLV and VSV fusion rates (Fig. 6) and qualitatively the difference between MLV and HIV. Figure 6 clearly shows a higher median number of spikes per particle for VSV than MLV; the remaining unexplained factor is what makes HIV fusion kinetics (both N_min and kinetic curves in Fig. 3) lie roughly intermediate between VSV and MLV rather than more closely resembling VSV. We hypothesize that the remaining difference is due to differences in pseudovirion morphology and spike distribution on the pseudovirion surface (cryo-electron microscopy [cryo-EM] images in Fig. S9). VSV pseudoviral particles are well characterized as having a more VSV-like morphology than retroviral (HIV or MLV) pseudovirus scaffolds. In addition, the spike densities on VSV particles do appear somewhat higher on micrographs, but we do not regard our cryomicrographs as sufficiently high contrast to be definitive.

At a coarser level, we hypothesize that the maximum likelihood stoichiometry is identical between viruses but that the rates of fusion vary across pseudovirus backgrounds, likely due to changes in spike density. We specify maximum likelihood stoichiometry to allow for stochastic variation in the number of fusion proteins used in individual fusion events but with the hypothesis that the most utilized stoichiometry is the same across pseudoviruses. We therefore fit cumulative distribution functions to a gamma distribution model previously used to parameterize fusion waiting times (38), where the fraction of particles undergoing lipid mixing, f, is given bywhere Γ(N) is a gamma function. We performed two sets of fits, either allowing all parameters to vary independently (Fig. S10) or constraining N to be identical across pseudovirus backgrounds but allowing the parameter τ to vary (Fig. 3b). The results show high-quality fits with a constraint of common stoichiometry (root mean square errors [RMSEs] = 0.08, 0.03, and 0.06 for VSV, HIV, and MLV, respectively, and Akaike information criterion (77) values of 2,470 versus 2,530 for unconstrained stoichiometry provide a statistical measure supporting the simpler model of a common stoichiometry), thus suggesting a single most likely fusion protein stoichiometry. In combination with the single-turnover variance analysis (N_min) discussed above, we conclude that there is likely static disorder present in the sample (i.e., different pseudoviruses have different spike protein arrangements on their surfaces) but the spike protein stoichiometry required for fusion likely does not vary between pseudovirus backgrounds. Instead, the static disorder manifests as different apparent fusion rates. The relative spike protein content of individual virions varies substantially (Fig. 6), supporting this notion.

Fusion mediated by each SARS-CoV-2 spike-bearing pseudovirus was measured at multiple trypsin concentrations to assess differences in fusogenicity of these different constructs. Results show differences in fusion efficiency (Fig. S11) with overall monotonic increases as a function of protease concentration; when efficiency plateaus, we attribute this to saturation. Cumulative distribution functions calculated for HIV pseudovirions treated with 50 μg/mL and 100 μg/mL trypsin show similar fusion kinetics (Fig. 7a): the data are compatible with a model in which the stoichiometry of fusion is identical and the fusion rates are within error of each other. The single-particle waiting time distributions were not significantly different from each other (P > 0.97; 2-tailed Kolmogorov-Smirnov test). These data are again compatible with a similar fusion protein stoichiometry across pseudovirus backgrounds, subject to heterogeneous populations as discussed above; this may, however, vary between virus-cell fusion and cell-cell fusion as has been suggested for influenza (44). Interestingly, the amount of S2 protein cleaved to form S2′ or a similar activated fragment may differ between proteases (variable band intensity in Fig. S3) but did not appear to alter the kinetics of fusion. Furthermore, the precise molecular weight of these activated fragments varied somewhat between proteases (Fig. S3) but did not alter the measured kinetics.

Finally, we tested the ability of multiple trypsin-like proteases to activate SARS-CoV-2 pseudovirions for fusion. We tested each of the following proteases. Human airway trypsin-like protease (HAT), also known as TMPRSS11D, is a serine protease expressed in both lower and upper airway tissues, found in sputum, and also expressed in other tissues (78–80). HAT has previously been suggested as a candidate for SARS-CoV-2 proteolytic activation (26). Cathepsins B and L are late-endosomal proteases implicated in activation of other viruses, including SARS-CoV (12, 13, 22, 81). No increase in fusion efficiency was observed at pH 5 compared to pH 7.4 despite prior structural evidence suggesting pH-dependent conformational difference in SARS-CoV-2 spike (82); it is possible that these changes may affect binding but not fusion. Finally, TMPRSS2 is a cell surface protease expressed in type 2 alveolar cells, among others, that has been implicated in MERS proteolytic activation and is a primary candidate for SARS-CoV-2 proteolytic activation (6, 23). Each of these was indeed capable of cleaving (Fig. S2 and S3) and activating SARS-CoV-2 spike protein for fusion (Fig. 7b; Fig. S11 and S12). None of the single-particle waiting time distributions were significantly different from each other (P > 0.25; 2-tailed Kolmogorov-Smirnov test), although TMPRSS2 shows slightly but not significantly faster fusion.

These data support an opportunistic model of SARS-CoV-2 activation and entry, where the virus can robustly utilize multiple parallel pathways for infection with mechanistic indifference. Proteolytic activation by cleavage at the S2′ site can occur in the airway extracellular milieu, at the cell surface, or within late endosomes. This is consistent with prior reports of multiple proteases being capable of activating SARS-CoV-2 (67, 83–86). Furthermore, proteolytic activation can either precede or follow receptor binding; these events need not occur in sequence, and the resulting fusions are mechanistically indistinguishable. We also demonstrate that the plasma membrane is capable of supporting SARS-CoV-2 fusion in addition to the previously reported endosomal fusion site (86, 87). Strikingly, these parallel pathways not only coexist but involve the same number of kinetic steps and thus likely the same mechanism, as probed by single-virus kinetics. We therefore propose a model (Fig. 8) in which the site of SARS-CoV-2 entry is determined based on the proteases present in a given tissue and the rates of spike protein cleavage relative to the rates of viral attachment and endocytosis. Where extracellular or cell surface proteolysis is rapid, entry will tend to occur at the cell surface, whereas if extracellular and cell surface proteolysis is slow, entry will tend to occur within endosomes. This model is also supported by other cell biology results (67, 83, 88) that have been published recently (many since initial posting of the manuscript for this article), which we here demonstrate biochemically.

The key proteolytic triggering event is believed to be formation of the S2′ fragment of the SARS-CoV-2 S protein, which releases the fusion peptide (6, 89) and likely potentiates conformational rearrangement to form the postfusion form of the protein (90). This proteolytic event may be facilitated by conformational changes elsewhere in the spike protein (91, 92). Trypsin, TMPRSS2, and TMPRSS11D are all serine proteases with compatible cleavage sites just upstream of the fusion peptide, forming the canonical S2′ fragment. Cathepsin L, however, has previously been described to cleave SARS-CoV-1 somewhat farther N terminal to the fusion peptide yet still leads to functional activation (12, 19, 93). Cathepsins have also been described as functionally activating SARS-CoV-2 (33, 84), and our results show that either cathepsin produces fusion proteins that act with the same kinetics and stoichiometry, suggesting that the additional residues do not impact activation and fusion mechanism. Our results on fusion kinetics thus show that cleavage proximal to the fusion peptide or farther N terminal are functionally and, we believe, mechanistically equivalent in activating SARS-CoV-2 S for entry.

Our single-virus fusion measurements utilize biosafety level 2 (BSL2) pseudoviral systems rather than infectious SARS-CoV-2. However, the robustness of the results across multiple pseudovirus backgrounds suggests that the observations are fundamental features of SARS-CoV-2 spike-mediated fusion regardless of the viral core. Furthermore, since observed fusion rates (although not the mechanisms) do vary with apparent spike protein density, SARS-CoV-2 viral variants that express more functional spike protein are expected to be more infectious. This may be a mechanism for the infectiousness of the D614G mutant (94). We also note that measurements of lipid mixing do not probe viral core exposure, and it is unlikely yet possible that differences in protease activation could affect fusion pore opening and viral uncoating. Such a finding would diverge from the body of work on other viral families (40, 95), where activation kinetics primarily affect hemifusion and lipid mixing rather than downstream fusion pore opening.

The ability of SARS-CoV-2 to utilize parallel entry pathways may contribute to effects where inhibitors targeting a single host protease, such as camostat mesylate, directed at TMPRSS2, are effective in cell and tissue models but show less robust efficacy in clinical trials (6, 96–98). Our data suggest that clinically effective inhibition of proteolytic activation may benefit from combination therapy to target multiple host proteases. Potent monotherapies may sufficiently reduce overall infection to have a clinical effect, but we suggest that multitargeted therapy may be more efficacious overall. We also show in particular that extracellular proteases such as TMPRSS11D can productively activate SARS-CoV-2 spike for viral membrane fusion, likely in advance of ACE2 receptor binding. Effective protease-targeted therapies thus need to consider the airway extracellular milieu in addition to the cell and tissue types typically used to assess betacoronavirus entry.

The opportunistic activation of SARS-CoV-2 also has important implications for viral evolution and host tissue infection. SARS-CoV-1 was canonically thought of as utilizing the endosomal entry pathway (2, 11–13, 43, 99), although substantial evidence suggests that it too can be activated by TMPRSS2 and other proteases and can enter at the cell surface (14, 15, 100, 101). Most recently, while this work was in review, the B.1.1.529, or Omicron, variant of SARS-CoV-2 was shown to have reduced TMPRSS2 sensitivity and concomitant increased utilization of endosomal entry pathways (29, 87). This change is correlated with differences in relative susceptibility of different airway tissues to infection and changes to the resulting pathology. It is thus likely that the ability of SARS-like betacoronaviruses to utilize multiple proteases for entry widens the potential range of tissues that can be infected and facilitates ready adaptation to new host and tissue environments.

We thank Jesse Bloom and Judith White for the gift of reagents and M. Cervantes, G. Morbioli, A. Villamil Giraldo, R. Rawle, and J. White for helpful discussions. Electron cryomicroscopy was performed by K. Dryden at the Molecular Electron Microscopy Core at the University of Virginia.

This work was supported by grants from the Commonwealth Health Research Board (207-01-18), UVA Global Infectious Diseases Institute, and Knut and Alice Wallenberg Foundation (KAW2015.0198 and KAW2020.0209) to P.M.K.

Conceptualization, A.S. and P.M.K.; Methodology, A.S., M.C., and P.M.K.; Investigation, A.S., M.C., S.T.B., and T.H.; Data Analysis, A.S., M.C., T.H., and P.M.K.; Writing – Original Draft, A.S. and P.M.K.; Writing – Review & Editing, A.S., M.C., S.T.B., T.H., and P.M.K.

We declare no conflict of interest.