SARS-CoV-2 spike protein: structure, viral entry and variants

SARS-CoV-2 entry.

SARS-CoV-2 infection of a host cell initiates with fusion of its membrane with the host cell membrane. Although membrane fusion is energetically favorable, repulsive electrostatic and hydration forces cause high kinetic barriers when the two membranes are in close apposition^2,3. Free energy to overcome these kinetic barriers is generated when the virus-encoded S protein refolds from a high-energy, metastable prefusion conformation to a low-energy, stable postfusion state^4–6. The S protein is a heavily glycosylated type I membrane protein that is anchored in the viral membrane by a transmembrane segment. It also contains another hydrophobic region, known as fusion peptide (FP), capable of interacting with the host membrane. The S protein belongs to the class I viral fusion proteins, a category that includes all other coronavirus S proteins, HIV-1 envelope glycoprotein, influenza A virus hemagglutinin and Ebola virus glycoprotein, reflecting their structure and function similarities. First produced as a trimeric single-chain precursor, the S protein is cleaved at a polybasic amino acid sequence into two fragments: S1 for receptor binding and S2 for membrane fusion^7–9. The cleavage is catalyzed by the host protease furin in the trans-Golgi network of the virus-producing cells. To infect a new host cell, the S protein binds to the entry receptor angiotensin converting enzyme 2 (ACE2) on the host cell surface for attachment, and is further processed at a second cleavage site in S2, named S2’ site, by another host protease – either TMPRSS2 (transmembrane serine protease 2) for cell surface entry or cathepsin L for endosomal entry^1,10,11. Once activated, the prefusion S protein trimer undergoes large conformational changes to insert the FP into the target cell membrane and subsequently refolds into the stable postfusion structure, which leads to the fusion of the viral and target cell membranes and the release of the viral RNA genome into the cytosol of the host cell¹.

Structure of SARS-CoV-2 spike protein.

The protein sequence of the original Wuhan-Hu-1 strain of SARS-CoV-2 contains 1273 amino acid residues and 22 N-linked glycosylation sites^12–15 (Fig. 1a). After the furin-site cleavage, three copies of each S1 and S2 fragment comprise the mature S protein. This S protein is anchored to the viral membrane by its transmembrane segment, presenting a large ectodomain on the virion surface. S1 includes the N-terminal domain (NTD), a receptor-binding domain (RBD) and two C-terminal domains (CTD1 and CTD2), whereas S2 contains FP, heptad repeat 1 (HR1), central helix (CH), connector domain (CD), heptad repeat 2 (HR2), transmembrane (TM) segment and the cytoplasmic tail (CT).

Cryogenic electron tomography (cryo-ET) studies with chemically inactivated SARS-CoV-2 preparations in the early days of the pandemic show that each virion has on average of ~25 S proteins, randomly distributed on the surface, with approximately 97% in the prefusion conformation and 3% in the postfusion conformation^16,17 (Fig. 1b). The prefusion S protein trimers adopt either a closed and symmetrical structure with three RBDs in the down conformation that represents a receptor-inaccessible state, or an open and asymmetrical structure with one RBD in the up conformation that represents a receptor-accessible state^18,19 (Figs. 1c and 1d). These prefusion trimers can also break the three-fold symmetry at their membrane-proximal stem region and bend towards the viral membrane at various angles^16,17,20 (Fig. 1b). The postfusion S proteins form long rods that are very rigid with their long axis perpendicular to the viral membrane plane^17,21. Single-particle cryo-EM with recombinant spike proteins has revealed additional features at the atomic level^{18,19,22–24} (Figs. 1c and 1d). In the prefusion structures of both the stabilized soluble and full-length, detergent-solubilized trimers, S1 shows a ‘V’ shape with the NTD forming one arm and the RBD, CTD1 and CTD2 forming the other. These four domains wrap around the central helical bundle made up by the prefusion S2 trimer. The three RBDs of the S protein trimer constitute its apex and sample either the ‘up’ or ‘down’ conformation. The three NTDs are located at the periphery of the S protein trimer, with each contacting the RBD from its neighboring protomer. The CTD1 and CTD2 pack underneath the RBD against S2 and between the two adjacent NTDs.

There are four stacked β-sheets and several connecting flexible loops in the NTD, which is also covered by some N-linked glycans^18,19,22. This domain is recognized by potent neutralizing antibodies^25–27, which suggests that it is functionally critical or at least close to other functional sites. The RBD consists of a core structure that has a five-stranded anti-parallel β-sheet with short connecting α-helices on both sides, and an extended loop (that is, the receptor-binding motif (RBM)), that packs against one side of the core structure to create the surface for receptor binding^28,29 (Fig. 1e). The RBD in its down conformation covers the central helical bundle of S2 together with two other RBDs in the S protein trimer, while leaning on the CTD1 from the same polypeptide chain and the NTD from a neighboring protomer. This configuration partially occludes the RBM, preventing it from binding ACE2. When the RBD flips up, the adjacent CTD1 and NTD move aside to accommodate RBD movement, fully exposing the RBM. The two CTDs are formed mainly by β-structures (Figs. 1c and 1d). If the RBD is considered an insertion between two anti-parallel β-strands in the CTD1, the CTD1 can also be viewed as an insertion between two β-strands in the CTD2. The CTD1 packs underneath the RBD and rotates outwards when the RBD adopts the up conformation. The CTD2 has two stacked, four-strand β-sheets. An interstrand loop of one β-sheet contains the S1/S2 cleavage site and one strand is the N-terminal segment of S2. Therefore, S1 dissociation from S2 requires destabilization of the CTD2 to free the N-terminal end of S2.

Structural details of the prefusion conformation.

Structural studies of the full-length S protein trimers in the prefusion conformation have identified the 630 loop (residues 620 to 640) and fusion peptide proximal region (FPPR; residues 828 to 853) as control elements that may modulate RBD movement and control the kinetics of the S protein structural rearrangements^22,23 (Fig. 1c). The 630 loop is largely disordered in the Wuhan-Hu-1 S protein trimer, from which S1 dissociates spontaneously even in absence of ACE2, but it becomes structured in the trimer bearing a D614G substitution (G614 trimer) that remains fixed in S protein since its emergence early in the pandemic^22,23. The structured 630 loop packs along a hydrophobic surface of the CTD2, which is largely solvent-exposed in the Wuhan-Hu-1 S protein trimer. The ordered 630 loop that stabilizes the CTD2 by closing off the exposed hydrophobic surface seems to limit S1 shedding, enhancing the stability of a cleaved S protein trimer. However, the 630 loop can also become disordered when the RBD flips up even in the G614 trimer²³. Likewise, the FPPR is structured when the RBD is in its down conformation but disordered when the RBD flips up. The structured FPPR, abutting the opposite side of CTD1 from the RBD, would clash with the CTD1 if the RBD moves up and thus it helps clamp the S protein trimer in the closed, RBD-down conformation²². The CTD1 is probably a structural relay between RBD and FPPR that detects displacement on either side. Overall, the RBD up movement apparently pushes both the 630 loop and FPPR out of their original positions in the closed conformation, making them invisible in cryo-EM maps^22,23. Interestingly, two distinct conformations of the three-RBD-down state, referred to as being ‘locked’ and ‘closed’, respectively, have been reported, possibly reflecting additional conformational dynamics of the FPPR and 630-loop³⁰.

In the prefusion S protein trimer, three S2 fragments fold around a long three-stranded coiled-coil formed by the CH²² (Fig. 1c; ref). HR1 is fragmented into four separate α-helices. Two of them, including the longest, assemble into a nine helix-bundle with the CH coiled–coil comprising the most rigid part of the entire S protein trimer. The CD region forms a β-hairpin and expands into a three-stranded β-sheet with another β-strand (residues 718-729; β^718-729), which wraps around the C-terminal end of the CH coiled-coil. The FP is largely buried by the region surrounding the S2’ cleavage site and the N-terminal portion of S2 from a neighboring protomer. The remaining HR2, TM and CT segments are not visible in all high-resolution structures of the prefusion S protein trimers, but are present in low-resolution cryo-ET reconstructions of the virus-bound S protein trimers that tilt away with an angle from the 3-fold axis¹⁶.

Structural details of the postfusion conformation.

In the postfusion state, S1 dissociates as a monomer and the S2 trimer adopts a rigid, rod-like shape^17,21. The membrane-interacting segments of S2 are functionally important, but missing in many postfusion structures of coronavirus spikes. A recent cryo-EM structure of the intact postfusion SARS-CoV-2 S2 reconstituted in lipid nanodiscs has revealed structural details of its membrane-interacting regions²⁴. The ectodomain of the postfusion S2 in the lipid bilayer adopts a structure nearly identical to that of the detergent-solubilized protein²². HR1 and CH form a long central coiled–coil as the core structure. (Fig. 1f). The connector β-sheet together with the CH coiled–coil do not rearrange between the prefusion and postfusion conformations. The segment of residues 737-769 in the S1/S2-S2’ fragment forms three consecutive α-helices (3H), which packs tightly against the groove of the CH coiled–coil. The S2’ site is not cleaved in this protein preparation but invisible in the structure. In the C-terminal half of S2, the β-strand of residues 1127-1135 (β^1127-1135) upstream of HR2 from the adjacent protomer becomes part of the connector β sheet to expand it into four strands, projecting HR2 towards the TM region. In addition, the two-turn helix of residues 1148-1155 (α^1148-1155) wedges between two neighboring 3Hs and a longer helix of HR2 to form the six-helix bundle with the HR1 coiled–coil, stabilizing the rigid postfusion structure.

The structure has identified the segment of residues 8660-910 as the real FP among nine membrane-spanning helices (three for each protomer) found in the lipid bilayer. The helix formed by the FP extends the central coiled–coil to ~218Å and inserts well into the lipid bilayer (Figs. 1f and 1g), contributing substantially to the rigidity of the entire postfusion structure. This first FP helix is followed by a sharp U-turn within the lipid bilayer, and another membrane-spanning helix, thereby sending the adjacent FPPR back to the ectodomain side of the membrane. Six transmembrane helices of the FP of three protomers pack tightly together to make up a structure with a blunted cone shape (Fig. 1f). The other three membrane-spanning helices are formed by the TM segment and they gently wrap around the FP cone. The visible portion of the CT is embedded horizontally in the cytosolic headgroup region of the membrane to cap the tip of the FP cone. There is an intimate FP-TM interaction in the SARS-CoV-2 postfusion spike, as predicted for class I viral fusion proteins. The disulfide-locked FPPR has its C-terminal half embedded in the membrane, and its N-terminal half, directly connected to the disordered s816-834, projecting away from the membrane (Figs. 1f and 1g). The s816-834 segment, which is immediately downstream of the S2’ cleavage site and the previously assigned FP, is not inserted into the membrane.

Transition from the prefusion conformation to the postfusion state is accompanied by formation of the long postfusion central helix that translocates the FP by 185 Å. The FP reconfigures into a hairpin-like wedge and inserts in the target membrane (Fig. 1g), and it is also a site for the TM helix to dock in the final stages of membrane fusion. In addition, the HR2 must fold back to pack along the groove of the HR1–CH coiled–coil to form the stable postfusion bundle.

Structural evidence for the fusion-intermediate states.

For class I viral fusion proteins, the transition from prefusion to postfusion conformations is believed to go through prehairpin intermediate states that can be targeted by HR2-derived peptide fusion inhibitors^31,32. Two independent cryo-ET studies have indeed imaged distinct prehairpin intermediates that bridge the two fusing membranes at different angles^33,34. These data confirm the existence of the structural intermediates with life spans long enough for imaging, consistent with their susceptibility to antibodies or fusion inhibitors. Nevertheless, these reconstructions are still of low resolution and required molecular modeling to interpret, which suggests the structural heterogeneity of these samples still poses technical challenges, limiting understanding of these intermediate states at the atomic level. A recent study has revealed an ACE2-induced early fusion intermediate conformation in which S2 has partially refolded to translocate the FP while S1 remains associated despite the furin cleavage³⁵.

Interactions between the spike protein and ACE2, and other alternative receptors.

ACE2, previously identified as a receptor for other coronaviruses^36,37, is the main entry receptor for SARS-CoV-2 and binds the RBD to promote viral entry^10,13. It is a type I membrane glycoprotein that contains an ectodomain with metallopeptidase activity, which functions as a key negative regulator of the renin-angiotensin system (RAS) for modulating vascular function^38,39. The peptidase active site does not overlap with the RBD-binding site and this enzyme activity is not required for its receptor function. Cryo-EM structures of the soluble complexes of ACE2 and S protein trimer show that the trimer can bind one, two or three ACE2s in the RBD-up conformations^40,41. ACE2 binding pushes the NTD outwards slightly, but seemingly leaves S2 largely unchanged. The crystal structures of the RBD–ACE2 complex reveal that the extended RBM forms a slightly concave outer surface to bind the N-terminal helix of the catalytic domain of ACE2 (Refs. ^28,29 ) (Fig. 1e), There are 20 residues of ACE2 and 17 residues from the RBD that form networks of hydrophilic, side chain interactions. Hydrogen bonds and salt bridges between a series of polar residues dominate the RBD–ACE2 interactions. Additional hydrophobic interactions also contribute to receptor engagement.

ACE2 is the best-characterized receptor for SARS-CoV-2, but its low expression levels in respiratory and olfactory epithelial cells are not fully consistent with the pathology of the virus^42–44. Several alternative receptors have thus been suggested, including CD147, C-type lectins, phosphatidylserine receptors, heparan sulfate, neuropilin-1 (NRP1) and more recently TMEM106B^45,46. Lectins, phosphatidylserine receptors and heparan sulfate are the so-called ‘attachment factors’, rather than receptors, because they are not virus-specific and do not support efficient SARS-CoV-2 infection in the absence of ACE2 (Refs. ^47–50). CD147 is a transmembrane glycoprotein expressed ubiquitously in epithelial and immune cells, and viral entry was increased with higher levels of CD147, but direct binding of CD147 to the S protein has not been demonstrated^51–54. NRP1 is abundantly expressed in the respiratory and olfactory epithelia, and seems to bind the C-terminal end of S1 after the S1/S2 cleavage to enhance TMPRSS2-mediated entry in the cells with low levels of ACE2 (Refs. ^55,56). It may represent an entry factor on the epithelia that promotes efficient cell, vascular and tissue penetration of SARS-CoV-2 (Refs. ^57,58). NRP1 has also been identified as a receptor for reovirus and shown to enhance hepatitis B virus infection^59,60. TMEM106B is a lysosomal transmembrane protein, and its luminal domain is found to bind the RBM of the SARS-CoV-2 S protein. It enhances S protein-mediated syncytia formation in an ACE2-independent manner. Moreover, selected RBD-specific monoclonal neutralizing antibodies can function as an entry receptor⁶¹, providing a plausible explanation for how the virus infects those apparently nonpermissive cells, albeit inefficient virus replication^62,63. Finally, customized entry receptors can be built by grafting computationally designed virus-binding domains on transmembrane scaffolds to support sarbecovirus entry into cells⁶⁴, further demonstrating that the specific ACE2–S protein interaction is not required for SARS-CoV-2 entry under special circumstances.

Functional roles of host proteases.

Unlike other sarbecoviruses, SARS-CoV-2 has acquired a four-residue insertion (Pro-Arg-Arg-Ala) at the S1/S2 boundary, creating a furin-recognition site to cleave the S protein in virus-producing cells. Furin is a calcium-dependent serine endoprotease in the subtilisin-like proprotein convertase family^65–67, and preferentially cleaves a polybasic sequence (RXR/KR|X). The furin site in the SARS-CoV-2 spike seems to promote infection of lung cells and also contributes to viral pathogenesis in animal models^8,68–70.

After the furin cleavage at the S1/S2 junction, a downstream S2’ site is also cleaved to activate the S protein for membrane fusion. This S2’ cleavage is catalyzed by either TMPRSS2 at the cell surfaces or cathepsin L within endosomes. TMPRSS2 is a type II transmembrane serine protease that cleaves at an arginine or lysine residue, and belongs to the TMPRSS family of proteases. Although its role in healthy physiology is unknown, its contribution to respiratory virus infection is well documented⁷¹. TMPRSS2 is present in many epithelial tissues, but only nasal, bronchial and gastrointestinal epithelia express both ACE2 and TMPRSS2 (Refs. ^42,44,72). Proteolytic activation of spike by TMPRSS2 and downstream membrane fusion seems to require both ACE2 engagement and previous cleavage by furin at the S1/S2 boundary¹⁰. Some Omicron subvariants (for example, BA.1) seem to use TMPRSS2 less and depend more on the endocytic entry than the pre-Omicron variants^73–75, possibly leading to a change in cellular tropism away from TMPRSS2-expressing cells, and explaining the altered tissue tropism and pathogenesis^73,74,76,77. Cathepsins are proteases with endo- or exo-peptidase activities that are responsible for protein degradation at low pH (4.5–5.0) in the late endosomes and lysosomes. Cathepsin L is a matrix-degrading cysteine protease that is expressed in all tissues and cell types, and associated with chronic inflammation^78–80. Blocking the cathepsin L-dependent endosomal pathway only is insufficient to prevent SARS-CoV-2 infection, at least for pre-Omicron variants, consistent with the limited antiviral efficacy of endosomal acidification inhibitors such as hydroxychloroquine⁸¹.

Membrane fusion.

High-resolution structures of the SARS-CoV-2 S protein have revealed its critical components, including the RBD, 630 loop, CTD2, FPPR, FP, HR1, HR2 and TM. In the native S protein trimer before binding to ACE2, the FPPR and 630 loop help retain the RBDs in the down conformation^22,23. Intrinsic protein dynamics allow the RBD to flip up and sample its up conformation occasionally (Fig. 2). ACE2 binding captures the RBD-up conformation, and move both the 630 loop and FPPR out of their original positions in the closed S protein trimer, which leads to the early fusion intermediate conformation that exposes the otherwise buried S2’ site for proteolytic cleavage by TMPRSS2 or cathepsin L³⁵. The movement of the 630 loop destabilizes the CTD2 and frees the N-terminal segment of S2 from S1, which is likely to result in S1 dissociation, a consequence of the earlier furin-catalyzed precleavage of S1–S2 boundary. Departure of S1 then initiates a cascade of refolding events in the metastable prefusion S2, transitioning through the extended prehairpin intermediate states and then collapsing into a stable postfusion structure. Accompanying these transitions, the zipping-up of HR1 into the continuous three-stranded coiled–coil drives translocation of the FP, which inserts into the target-cell membrane²⁴. Folding back of HR2 onto the HR1 to form the postfusion structure effectively places the FP and TM at the same end of the molecule, dragging the membranes with which they interact close to each other, and ultimately leading to membrane fusion.

Emergence of new variants.

Impact of spike mutations in new variants.