Main

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has undergone considerable evolution since its initial discovery, leading to the emergence of several variants of concern (VOCs) including Alpha2,3,4,5,6, Beta5,6,7,8,9,10, Gamma11 and Delta12,13. These variants that have multiple mutations on their S protein show enhanced transmissibility and resistance to antibody neutralization13. Recently, a new variant named Omicron (B.1.1.529), which was first reported in South Africa in November 2021, was classified as the fifth VOC by the World Health Organization (WHO)14. Omicron bears 37 mutations in its S protein relative to the original SARS-CoV-2 strain15,16. As a consequence, Omicron has been observed to extensively escape neutralization by previously developed neutralizing monoclonal antibodies (mAbs) or sera from vaccinated or convalescent individuals15,17,18,19,20,21,22. Among all of the Omicron S mutations, 15 are present in the receptor-binding domain (RBD) that mediates binding of the virus to its host cell receptor, angiotensin-converting enzyme 2 (ACE2), which is also a major target for neutralizing antibodies23,24,25,26,27. However, Omicron still uses ACE2 as its entry receptor22. Moreover, the Omicron S appears to have an increased binding affinity to ACE2 relative to the wild-type (WT) S15,16,28.

The high transmissibility and greatly enhanced resistance to antibody neutralization observed for Omicron makes this VOC particularly threatening. Therefore, further understanding of the nature of Omicron is of substantial importance and may help in developing countermeasures against this VOC. To address this from a structural aspect of how Omicron binds to the ACE2 receptor and how it recognizes or evades neutralizing antibodies raised to the original virus, here we performed cryo-electron microscopy (cryo-EM) analyses on the Omicron S trimer and its complexes with ACE2 or with neutralizing mAbs.

Closed and open states of Omicron S protein

We prepared a prefusion-stabilized trimeric S protein of the SARS-CoV-2 Omicron variant (Extended Data Fig. 1) and determined its cryo-EM structures. Three cryo-EM maps, including an all RBD down conformation (termed Omicron S-close), an RBD-1 up open conformation (termed Omicron S-open), and an RBD-1 up while RBD-3 disordered open conformation (termed Omicron S-open-2), were determined at 3.08, 3.40 and 3.41 Å resolution, respectively (Fig. 1a, b, Extended Data Fig. 2a–e, Extended Data Table 1). For the Omicron S-close state, the three protomers are well resolved and they display similar conformation with their RBDs in the down position (Fig. 1c, Extended Data Fig. 2d). The Omicron S-close appears more twisted and compact in the RBDs than the G614 S-close structure1 (Fig. 1d). In addition, in the Omicron S-open state, the RBDs are slightly more twisted and compact than that of the G614 S-open state1 (Fig. 1e). There is no linoleic acid (termed LA) in the Omicron S-close, S-open and S-open-2 maps, and LA has been suggested to lead to more compact RBDs29. Thus, the Omicron S trimer is more compact than that of G614, and this is not caused by binding of LA. Moreover, in the Omicron S-open-2 map, the RBD-3 density appears mostly disordered, indicating an extremely dynamic nature of RBD-3 (Fig. 1a). Further 3D variability analysis30 on the Omicron S trimer dataset revealed an intrinsic rising up motion of RBD-1, which could alter the original RBD-1–RBD-3 contact and destabilize RBD-3, making it extremely dynamic and may transiently rise up (Fig. 1f, Supplementary Video 1).

Fig. 1: Cryo-EM structures of the SARS-CoV-2 Omicron S trimer.
figure 1

a, Cryo-EM maps of the Omicron S-close, S-open and S-open-2 states. Protomers 1, 2 and 3 are shown in light green, royal blue and gold, respectively; this colour scheme is followed throughout. In S-open-2, the mostly disordered RBD-3 is indicated by a red dotted ellipsoid. b, Atomic model of the Omicron S-open, with mutations labelled and indicated by a red sphere. c, Side view of the overlaid protomers of the Omicron S-close. d, Top view of the overlaid RBDs of the Omicron S-close (violet red) and the G614 S-close (PDB: 7KRQ; dark grey), indicating a twist of the Omicron S-close relative to that of G614. e, Top view of the overlaid RBDs of the Omicron S-open (light green) and the G614 S-open (PDB: 7KRR; purple), indicating a twist of the Omicron S-open relative to that of G614. f, One representative 3D variability analysis motion of the Omicron S. The two left maps illustrate the top view of two extremes in the motion, with the RBD-3 indicated by a black dotted ellipsoid; the top view of the overlaid two extreme maps is shown on the right. g, Newly formed hydrogen bonds (black dashed line) in the interfaces of protomer 1–protomer 3 (blue dashed line box) and protomer 1–protomer 2 (red dashed line box) of the S-close state (see also Extended Data Fig. 2f).

Full size image

The population distribution of the Omicron closed and open states (S-open and S-open-2) is about 60.8% and 39.2%, respectively (Extended Data Fig. 2a), displaying a considerable population shift to the closed state than that of the Beta and Kappa S protein (both around 50.0:50.0% open-transition ratio) or that of the Delta S protein (75.3:24.7% open-transition ratio) from our recent studiues10,31, with these S structures obtained in the same sample preparation and data processing schemes. Thus, the Omicron S trimer appears more prone to the closed state and potentially stabilized relative to the counterparts of the Delta, Kappa, Beta or G614 variants. The population distribution of the closed and open states of these S trimers varies among different studies, which could potentially be due to subtle difference in the chemical condition used by different research groups32,33,34,35. Inspection of the protomer interaction interface of Omicron S-close revealed new hydrogen bond interactions induced by the unique Omicron substitutions in the SD1 and S2 regions (Fig. 1g, Extended Data Fig. 2f, Extended Data Table 2). Specifically, T547K from the SD1 subdomain of protomer 1 forms a new hydrogen bond with N978 from the S2 subunit of protomer 3, potentially enhancing the S1–S2 interaction between the two protomers; N856K or N764K from protomer 1 can form hydrogen bonds with T572 or Q314 from protomer 2, respectively. We also observed multiple new such interactions between N317 or R319 of protomer 1 and D737 of protomer 3 (Fig. 1g, Extended Data Fig. 2f). The SD1 and SD2 of S1 is the hinge for RBD upward rotation31,35,36,37,38, and disturbing the SD1–S2 interface could destabilize the interface and increase S1 mobility36. Collectively, these extra hydrogen bonds mainly induced by Omicron substitutions contribute greatly to strengthen the inter-protomer and S1–S2 interactions, markedly stabilizing the Omicron S trimer and inhibiting its transformation towards the fusion-prone open state and subsequent shielding of S1.

Structure of enhanced S–ACE2 interaction

Compared with the WT strain, Omicron bears 15 mutations in the RBD region, nine of which are located in the receptor-binding motif (RBM)15. We assessed whether these mutations affect the human ACE2 receptor-binding ability of the Omicron S trimer by performing biolayer interferometry assay. We found that the ACE2-binding affinity of the Omicron S (dissociation constant (Kd) = 80 nM) is comparable to that of the Delta S (Kd = 88 nM) but is about threefold higher than that of the G614 S (Kd = 237 nM; Fig. 2a), consistent with other recent reports15,16,28.

Fig. 2: Structural basis of the enhanced Omicron S–ACE2 interaction.
figure 2

a, Measurement of the binding affinity between the ACE2 monomer and the S trimer of the G614 (left), Delta (middle) and Omicron (right) variants using biolayer interferometry. Biotinylated S trimers were loaded onto streptavidin sensors and were then allowed to interact with different concentrations of ACE2 (shown on the right). Raw sensorgrams and fitting curves are shown in colour and grey, respectively. Association and dissociation phases are divided by the red dashed lines. Kdis, dissociation rate; Kon, ‘on-rate’. b, Cryo-EM maps of the Omicron S–ACE2 complex in three distinct conformational states. In the S–ACE2-C2 and S–ACE2-C3 maps, the density of RBD-2-associated or RBD-3-associated ACE2 appears weaker than that of the stably associated ACE2 on RBD-1 (see also Extended Data Fig. 3c). ACE2 is shown in violet red. This colour scheme is followed throughout. c, Density map of the focus-refined Omicron RBD-1–ACE2 and the zoomed-in view of the RBD–ACE2 interaction interface, showing the side chain densities of Q493R, G496S, Q498R, N501Y and Y505H on the RBM. d, The substituted residues R493, S496, R498 and H505 of the Omicron RBM form new interactions with E35, D38, Q42 and K353 of ACE2 (the spring represents the salt bridge, and the black dashed line represents the hydrogen bond) relative to that in the WT RBD–ACE2 (PDB: 6M0J; dark grey). A newly formed hydrogen bond without substitution is shown by a red dashed line. e, Interaction interface areas between ACE2 and the RBD of Omicron, Delta (PDB: 7W9I) and WT (PDB: 6M0J), analysed using PISA. f, The electrostatic surface properties of Omicron, Delta (PDB: 7W9I) and WT (PDB: 6M0J) RBDs, with the mutated residues indicated. The black outlines depict the footprint of ACE2 on the RBD. g, The electrostatic surface property of ACE2, with residues in proximity to RBD-1 (less than 4 Å) indicated (related to Extended Data Table 3).

Source data

Full size image

Next, we carried out cryo-EM analysis on the Omicron S in complex with the human ACE2 peptidase domain (Extended Data Fig. 3a). We obtained three cryo-EM maps of the S trimer engaged with ACE2, including a one RBD-up state (termed Omicron S–ACE2-C1), a two RBD-up state (termed Omicron S–ACE2-C2) and an all three RBD-up state (termed Omicron S–ACE2-C3) at 3.69, 3.70 and 4.04 Å resolution, respectively (Fig. 2b, Extended Data Fig. 4, Extended Data Table 1). The population distribution among Omicron S–ACE2-C1, S–ACE2-C2 and S–ACE2-C3 is about 43.9%, 41.2% and 14.9%, respectively (Extended Data Fig. 3b), displaying an obvious higher one RBD-up C1 population (43.9%) than that of the Beta, Kappa or Delta variants (C1 population ranges from 8.3% to 14.1%) observed in our recent studies10,31, with these S–ACE2 structures obtained in the same sample preparation and processing schemes. Together, the Omicron S trimer exhibits less ability to transform to the more RBD-up C2 or C3 states in the presence of ACE2 than that of the Beta, Kappa and Delta variants.

We further focus-refined the stably associated Omicron RBD-1–ACE2 region and obtained a 3.67 Å resolution structure (Fig. 2c, Extended Data Figs. 3, 4), which revealed many new interactions between the RBM substitutions and ACE2 compared with that of the WT RBD–ACE2 (ref. 26). Specifically, the RBM Q493R and Q498R result in three new salt bridges with the ACE2 E35 and D38, respectively; the RBM G496S and Y505H with ACE2 K353, the RBM Q498R with ACE2 Q42, and the RBM S477N with ACE2 Q19 also form new hydrogen bonds, respectively (Fig. 2d, Extended Data Table 3), generally in line with recent studies16,28,39,40,41,42. Moreover, we observed an extra hydrogen bond between RBM T500 and ACE2 D355 (Fig. 2d). Our previous study showed that Y505A obviously decreased the binding affinity of ACE2 (ref. 35), thus the Omicron Y505H substitution may maintain or even enhance ACE2 binding. Meanwhile, the K417N substitution, which occurred in Omicron as well as in Beta and Delta, is known to markedly reduce the binding of ACE2 through abolishing multiple salt bridges and/or hydrogen bonds with ACE2 D30 (refs. 26,43,44). Together, these newly formed RBM–ACE2 interactions may compensate for the loss of some of the original RBM–ACE2 interactions due to the residue changes introduced into the Omicron RBM.

Further inspection of the surface property showed that the substitutions in RBM, especially Q493R, G496S, Q498R and Y505H, render the substituted site within the ACE2 interaction footprint more positively charged, which could strengthen the interaction of RBM with the overall negatively charged ACE2 in the interaction interface (Fig. 2f, g). Corroborating this, the Omicron RBD–ACE2 interaction area (920.2 Å2) was enlarged compared to that of the WT (843.3 Å2), whereas it was comparable to that of the Delta RBD–ACE2 (928.4 Å2)31 (Fig. 2e). This is also in agreement with our biolayer interferometry data showing that the ACE2-binding affinity of the Omicron S is similar to that of the Delta S but is higher than that of the G614 S (Fig. 2a).

Omicron sensitivity to neutralizing mAbs

We compared five previously generated mAbs—2H2, 3C1, 8D3, S5D2 and S3H3 (refs. 45,46)—for neutralization of the WT (Wuhan-Hu-1 strain), Delta or Omicron pseudoviruses (Fig. 3a, b). The half-maximal inhibitory concentration (IC50) values of the mAbs 3C1, 2H2, 8D3 and S3H3 against Delta were comparable (less than 2.5-fold variation) to the corresponding ones against WT, whereas S5D2 was still neutralizing to Delta (IC50 = 734.6 ng ml−1) but was about 90-fold less potent. In Omicron neutralization tests, 3C1, 8D3 and S5D2 lost neutralization activity (IC50 > 10 μg ml−1). However, 2H2 and S3H3 remained highly effective against Omicron with IC50s being 30.4 and 53.3 ng ml−1, respectively, despite observing a 3.3-fold increase (relative to the WT) in the IC50 value for 2H2. Thus, 2H2 and S3H3 are two potent neutralizing mAbs against Omicron.

Fig. 3: Neutralization and binding activities of the mAbs against Omicron and Delta variants.
figure 3

The mAbs were raised to WT RBD or S trimer proteins. a, Neutralization IC50 values and fold changes in neutralization potency for the Delta and Omicron variant pseudoviruses (PVs) compared to the WT pseudovirus. A minus sign (−) denotes decrease. The yellow highlighting indicates a more than tenfold decrease; the red highlighting denotes a more than 1,000-fold decrease. b, Neutralization of the mAbs towards WT, Delta and Omicron pseudoviruses. All mAbs were fourfold serially diluted. Data are expressed as mean ± s.e.m. of four replicate wells. The horizontal black dotted lines indicate 0% neutralization. c, Binding activities of the mAbs to recombinant S trimers of the WT, Delta and Omicron strains were tested by ELISA. Serially diluted S trimer proteins were coated onto the ELISA wells. Data are mean ± s.d. of triplicate wells. Neutralization and ELISA data are representative of two independent experiments with similar results.

Source data

Full size image

We then compared the binding ability of the five mAbs to the WT, Delta and Omicron S proteins by ELISA (Fig. 3c). For mAb S5D2, its binding to the Delta S and to the Omicron S was nearly abolished; for mAbs 3C1 and 8D3, their reactivity profile with the Delta S closely resembled that towards the WT S but their binding to the Omicron S was substantially reduced; for mAb 2H2, its binding curve to the Omicron S was similar to those towards the WT S and the Delta S, despite the binding efficiency to the Omicron S being slightly lower; meanwhile,mAb S3H3 produced nearly identical binding curves to the three S proteins. Overall, the antigen-binding ability of the mAbs was in good agreement with their neutralization potency towards specific variant pseudovirus (Fig. 3). Collectively, the above data showed that Omicron remains sensitive to binding and neutralization by the mAbs 2H2 and S3H3, whereas it displays resistance to 3C1, 8D3 and S5D2.

Structure of the Omicron S–S3H3 complex

The mAb S3H3 is a unique neutralizing antibody that binds to the SD1 region of the WT S46. Our cryo-EM study revealed two states of the Omicron S in complex with S3H3 Fab. Both structures showed two engaged Fab densities on the SD1 of protomer 2 and protomer 3, but with the RBD-1 in the up (termed Omicron S-open–S3H3) or down (termed Omicron S-close–S3H3) conformations (Fig. 4a, b). The two maps were resolved to the resolution of 3.48 Å and 3.64 Å, respectively (Extended Data Fig. 5, Extended Data Table 1). Compared with the Omicron S-open, the S trimer in S-open–S3H3 exhibits a slight twist and the RBD-1 displays a 9.1º downward rotation (Fig. 4c), making the S trimer seemingly less ‘open’ as a whole. SD1 shows a slight downward rotation (Fig. 4c).

Fig. 4: Cryo-EM analyses on the Omicron S–S3H3 Fab complex.
figure 4

a, b, Side and top views of the cryo-EM map of the Omicron S-open–S3H3 (a) and S-close–S3H3 complex (b), with the heavy chain (HC) and light chain (LC) of S3H3 Fab in medium blue and violet red, respectively. This colour scheme is followed throughout this figure. c, Conformational comparison between Omicron S-open–S3H3 (light green) and Omicron S-open (orange), indicating a slight twist of the RBDs of S-open–S3H3 and the downward rotations of RBD-1 (up to 9.1º) and SD1. d, Model map fitting of the focus-refined Omicron SD1–S3H3 structure, and the zoomed-in view of the Omicron SD1–S3H3 interaction interface. The side chain densities at the interface were well resolved. e, The S3H3 binding on SD1 of protomer 2. f, The structural elements involved in the interaction between S3H3 Fab and SD1 are labelled. The SD1 T547K substitution is also shown in red. The residues of SD1 in proximity to S3H3 (less than 4 Å) are indicated and coloured in transparent orange (related to Extended Data Table 4). g, The SD1–S3H3 interaction interface analysed using PISA, with the major involved structural elements labelled (the spring represents the salt bridge, and the black dashed line represents the hydrogen bond).

Full size image

We further focus-refined the SD1–S3H3 Fab region and obtained a map at 3.61 Å resolution, with most of the side chain densities well resolved (Fig. 4d). The heavy chain of S3H3 Fab contributes more to the interactions with SD1 than the light chain does, that is, all three heavy-chain complementarity determining regions (CDRs) of S3H3 and its CDRL1 and CDRL3 interact with T323–E324 and the three loops (loop532–537, loop554–556 and loop581–584) of SD1 (Fig. 4e, f, Extended Data Table 4). Specifically, S32 of CDRL1 forms hydrogen bonds with S555 and I584 of SD1, respectively, D102 of CDRH3 forms a hydrogen bond with T581 from loop581–584, and D55 of CDRH2 forms a salt bridge with K537 from loop532–537 (Fig. 4g, Extended Data Table 4), thus constituting an intense interaction network between S3H3 Fab and SD1. A single mutation, T547K, is present in the SD1 of Omicron; however, this mutation was located outside the footprint of S3H3 (Fig. 4f), and thus will not affect the Omicron S–S3H3 interaction. Collectively, S3H3 binds to the extremely conserved SD1 region, therefore retains binding and neutralizing activity towards major VOCs including Omicron.

Discussion

In this study, we performed cryo-EM and biochemical analysis on the Omicron S trimer and its complex with the ACE2 receptor. We captured both the closed and the open states of the Omicron S trimer (Fig. 1a). In contrast to the S trimer of the Delta, Beta and Kappa variants10,31, the Omicron S-close and S-open structures appear more twisted and compact than their counterpart of the G614 strain1 (Fig. 1d, e), which may hinder its spike transformation towards the fusion-prone open state and shielding of S1. This could be related to enhanced inter-protomer and S1–S2 interactions induced by unique Omicron substitutions (T547K, N856K and N764K in SD1 and S2) (Fig. 1g).

Corroborating to the enhanced inter-protomer and S1–S2 interactions of Omicron, our cryo-EM analysis revealed that for the Omicron S trimer, the dominantly populated conformation is the closed state with all the RBDs buried, possibly leading to ‘conformational masking’ that may prevent antibody binding and neutralization at sites of receptor binding, similar to that described for the HIV-1 envelope47,48. Such an Omicron conformational masking mechanism could potentially affect antibodies that bind to the up RBDs (such as classes 1, 2 and 4 RBD antibodies49), contributing to the observed extensive neutralization escape by Omicron. However, for the Delta S trimer, our recent work showed that the dominant population is in the open state, indicating that the conformational masking mechanism may be less effective for the Delta variant31,35. This could contribute to the striking immune evasion of the Omicron variant15,17,18,19,20,21,22.

We also captured three states for the Omicron S–ACE2 complex (Fig. 2b). Unlike the Delta S, which tends to bind to three ACE2 in majority31, Omicron mainly binds to one or two ACE2 (Extended Data Fig. 3b). Further focus-refining of the RBD-1–ACE2 structure demonstrated that the substitutions on the RBM of Omicron (especially Q493R, G496S, Q498R, S477N and Y505H) result in the formation of new salt bridges and hydrogen bonds, and more complementary electrostatic surface properties (Fig. 2d, f, g), which together may compensate the abolished original RBM–ACE2 interactions26,43,44, leading to enhanced interactions with ACE2 and potentially enhanced transmissibility of the Omicron variant.

The present study showed that 2H2 and S3H3 retain potent neutralization towards Omicron and Delta (Fig. 3). Further structural study revealed a unique binding epitope of S3H3 within the SD1 region (Fig. 4). The binding of S3H3 to S trimer may function as a ‘lock’ to block the releasing of S1 from S2, resulting in inhibition of virus entry. The SD1 region targeted by S3H3 is extremely conserved among SARS-CoV-2 variants (Fig. 4f), thus explaining the cross-neutralization ability of S3H3 towards Omicron, Delta and other variants46. These findings also suggest a possibility to design SD1-based broad-spectrum SARS-CoV-2 vaccines. It is somewhat surprising that 2H2, whose epitope largely overlaps with RBM45, remained highly neutralizing against Omicron, despite the loss of the neutralization potency of this antibody to the Kappa and Beta variants due to their E484Q or E484K substitution10. Docking of our previous RBD-bound 2H2 Fab structure45 onto our current Omicron RBD-1 structure from the S-open state revealed that the 2H2 Fab could be reasonably well accommodated without major clashes with RBD (Extended Data Fig. 6a). In particular, the E484A substitution in Omicron does not appear to create steric hindrance with 2H2 due to the smaller size of the Ala side chain. In addition, the Omicron RBM substitutions, such as Q493R, Q498R and Y505H, do not introduce extra clashes between RBM and 2H2, and also make the epitope surface more positively charged than the WT RBD-1 (ref. 45) (Extended Data Fig. 6b), potentially allowing better interaction with the 2H2 paratope, which tends to be more negatively charged (Extended Data Fig. 6c).

We found that Omicron could escape three RBD antibodies, including 8D3, S5D2 and 3C1 (refs. 45,46). Both 8D3 and S5D2 are class 1 RBD antibodies and they share similar epitopes centred around loop477–489 (refs. 31,46). Docking of the RBD-bound 8D3 or S5D2 Fab structures31,46 onto the Omicron RBD-1 structure revealed that several Omicron residues, especially S477N, may potentially clash with the 8D3 Fab (Extended Data Fig. 6d); the S477N and T478K substitutions in Omicron may break the hydrogen bond network between RBD and S5D2 Fabs (Extended Data Fig. 6e). 3C1, whose epitope mainly involves the relatively stable RBD core45, belongs to the class 3 RBD antibody49. Analysis of the docked 3C1 structure on the Omicron RBD-1 showed that the S375F substitution may contribute to the escape of Omicron from 3C1 binding and neutralization through altering the interaction interface (Extended Data Fig. 6f).

In summary, the present study reveals that the Omicron spike acquires an increased RBM–ACE2 interaction network contributed by new hydrogen bonds and salt bridges and more favourable surface properties, and therefore maintains a strong affinity to ACE2, providing a possible explanation to the high infectivity and transmissibility of Omicron. In addition, our study suggests that, besides individual residue substitutions in RBD antibody epitopes, a conformational masking mechanism may also contribute to the extensive antibody evasion by Omicron. Moreover, our work demonstrates that Omicron remains sensitive to S3H3, an antibody that targets the extremely conserved SD1 region. Our findings provide structural insights into how Omicron maintains high transmissibility while greatly evading immunity, and may also inform design of broadly effective vaccines against emerging variants.

Methods

Protein expression and purification

To express the SARS-CoV-2 Omicron variant S glycoprotein ectodomain, the mammalian codon-optimized gene-coding SARS-CoV-2 (hCoV-19 Botswana R42B90_BHP_000842207 2021, GISAID ID: EPI_ISL_6752027) S glycoprotein ectodomain with double proline substitutions and ‘GSAS’ substitution at the furin cleavage site was cloned into the pcDNA 3.1+ vector35. A C-terminal T4 fibritin trimerization motif, a TEV protease cleavage site, a FLAG tag and a His tag were cloned downstream of the S glycoprotein ectodomain (Extended Data Fig. 1a). The constructs of prefusion-stabilized S proteins of SARS-CoV-2 G614 and Delta (B.1.617.2) variants were prepared as previously reported31. A gene encoding the human ACE2 peptidase domain (Q18–D615) with an N-terminal IL-10 signal peptide and a C-terminal His tag was cloned into the pcDNA 3.4 vector35. The recombinant proteins were prepared as previously reported in a published protocol35. In brief, the constructs were transiently transfected into HEK293F cells (Thermo Fisher) using polyethylenimine (PEI). Note that the HEK293F cell line has not recently been tested for mycoplasma contamination. Three days after transfection, the supernatants were harvested by centrifugation, and then passed through 0.45-μm filter membrane. The clarified supernatants were added with 20 mM Tris-HCl pH 7.5, 200 mM NaCl, 20 mM imidazole and 4 mM MgCl2, and incubated with Ni-NTA resin at 4 °C for 1 h. The Ni-NTA resin was recovered and washed with 20 mM Tris-HCl pH 7.5, 200 mM NaCl and 20 mM imidazole. The protein was eluted by 20 mM Tris-HCl pH 7.5, 200 mM NaCl and 250 mM imidazole.

Biolayer interferometry assay

Before the biolayer interferometry (BLI) assay, Ni-NTA-purified recombinant S trimer proteins of the G614, Delta and Omicron SARS-CoV-2 variants were further purified by gel-filtration chromatography using a Superose 6 increase 10/300 GL column (GE Healthcare) pre-equilibrated with PBS. Then, 70 μg of purified S trimer proteins of the G614, Delta and Omicron variants were separately biotinylated using the EZ-Link Sulfo-NHS-LC-LC-Biotin kit (Thermo Fisher) and then purified by Zeba spin desalting columns (Thermo Fisher).

Binding affinities of S trimers to ACE2 were determined by BLI analysis on an Octet Red96 instrument (Pall FortéBio). In brief, biotinylated S trimer proteins (approximately 40 μg ml−1) were immobilized onto streptavidin (SA) biosensors (Pall FortéBio). After washing with kinetic buffer (0.01 M PBS with 0.02% Tween 20 and 0.1% bovine serum albumin), these sensors were incubated with threefold serial dilutions of ACE2 monomer protein for 500 s. Subsequently, the biosensors were allowed to dissociate in kinetic buffer for 500 s. The data were analysed using the Octet Data Analysis 11.0 software to calculate affinity constants.

Neutralization

Luciferase (Luc)-expressing pseudoviruses bearing SARS-CoV-2 S proteins were constructed based on the HIV-1 backbone. In brief, HEK 293T cells (American Type Culture Collection) in a 10-cm dish were co-transfected using PEI (polysciences) with 10 μg pCMV-dR8.91 packaging plasmid, 10 μg recombinant pLVX-IRES-ZsGreen1 plasmid containing the luciferase reporter gene, and 2 μg recombinant pVAX1 plasmids encoding the SARS-CoV-2 S proteins. The cells were incubated with the transfection mixture for 6 h, and then 5 ml of fresh DMEM medium with 10% FBS was added to each dish. After incubation overnight, the media in the dishes were replaced with fresh DMEM medium (10% FBS). At 48 h post-transfection, the culture supernatant was harvested and frozen at −80 °C before use.

All mAbs were fourfold serially diluted and tested by a pseudovirus neutralization assay with human ACE2-overexpressing HEK 293T cells (293T-hACE2), which were generated in our previous study and verified by western blot, following our previous protocol45. Note that the 293T-hACE2 cell line has not recently been tested for mycoplasma contamination. Two days after pseudovirus infection, luciferase activity was measured. Data were analysed by non-linear regression using GraphPad Prism 8 to calculate the IC50.

ELISA

To test the binding activities of recombinant Omicron S protein with our previously developed anti-SARS-CoV-2 mAbs45,46, recombinant S trimer proteins from WT45, Delta or Omicron SARS-CoV-2 strains were twofold serially diluted and coated onto ELISA plates at 37 °C for 2 h. The plates were blocked with 5% milk in PBS-Tween 20 (PBST) at 37 °C for 1 h. After washing with PBST, the plates were incubated with 50 ng per well of each of the anti-SARS-CoV-2 mAbs45,46 at 37 °C for 2 h, followed by horseradish peroxidase (HRP)-conjugated anti-mouse IgG (1:5,000 dilution; Sigma) at 37 °C for 1 h. After washing and colour development, absorbance was measured at 450 nm. ELISA data were analysed by non-linear regression using GraphPad Prism 8. Note that for neutralization and ELISA assays, no statistical method was used to predetermine sample size, and no blinding or randomization protocols were used.

Omicron S–S3H3 Fab complex formation

The Omicron S–S3H3 Fab complex was prepared following our previously reported protocol46. In brief, purified S3H3 IgG was incubated with papain (300:1 w/w) in PBS buffer (in the presence of 20 mM l-cysteine and 1 mM EDTA) for 3 h at 37 °C. The reaction was quenched by 20 mM iodoacetamide. Fab was purified by running over a HiTrap DEAE FF column (GE Healthcare) pre-equilibrated with PBS. Omicron S protein was incubated with S3H3 Fab in a 1:6 molar ratio on ice for 1 h. The Omicron S–S3H3 Fab complex was purified by size-exclusion chromatography using a Superose 6 increase 10/300 GL column (GE Healthcare) in 20 mM Tris-HCl pH 7.5, 200 mM NaCl and 4% glycerol. The complex peak fractions were concentrated and assessed by SDS–PAGE.

Cryo-EM

Sample preparation

To prepare the cryo-EM sample of the Omicron S trimer, a 2.2 μl aliquot of the sample was applied on a plasma-cleaned holey carbon grid (R 1.2/1.3, Cu, 200 mesh; Quantifoil). The grid was blotted with Vitrobot Mark IV (Thermo Fisher Scientific) at 100% humidity and 8 °C, and then plunged into liquid ethane cooled by liquid nitrogen. To prepare the cryo-EM sample of the Omicron S–ACE2 complex, the purified Omicron S trimer was incubated with ACE2 in a 1:4 molar ratio on ice for 20 min and then vitrified using the same condition. The purified Omicron S–S3H3 complex was vitrified using the same procedure as for the Omicron S sample.

Data collection

Cryo-EM movies of the samples were collected on a Titan Krios electron microscope (Thermo Fisher Scientific) operated at an accelerating voltage of 300 kV. For the three datasets, the movies were collected at a magnification of ×64,000 and recorded on a K3 direct electron detector (Gatan) operated in the counting mode (yielding a pixel size of 1.093 Å), and under a low-dose condition in an automatic manner using EPU 2.11 software (Thermo Fisher Scientific). Each frame was exposed for 0.1 s, and the total exposure time was 3 s, leading to a total accumulated dose of 50.2 e–/Å2 on the specimen.

3D reconstruction

For each dataset, the motion correction of image stack was performed using the embedded module Motioncor2 in Relion 3.1 (refs. 35,50,51), and CTF parameters were determined using CTFFIND4.1.8 (ref. 52) before further data processing. Unless otherwise described, the data processing was performed in Relion 3.1.

For the Omicron S dataset (Extended Data Fig. 2), 600,845 particles remained after reference-free 2D classification in cryoSPARC v3.3.1 (ref. 30). After two rounds of 3D classification and further focused 3D classification on the RBD-1 region, we obtained an Omicron S-close map from 108,509 particles and an RBD-1-up open conformation from 69,873 particles. We then preformed focused 3D classification on the RBD-3 region of the open-state dataset and obtained an Omicron S-open map from 30,967 particles and a S-open-2 map from 38,906 particles. After Bayesian polishing and CTF refinement, the Omicron S-close, S-open and S-open-2 datasets were independently loaded into cryoSPARC v3.3.1 (ref. 30) and refined to the resolution of 3.08 Å, 3.40 Å and 3.41 Å, respectively, using non-uniform refinement. The overall resolutions for all of the cryo-EM maps in this study were determined based on the gold-standard criterion using a Fourier shell correlation (FSC) of 0.143. Moreover, we performed 3D variability analysis on the Omicron S trimer dataset containing 178,382 particles in cryoSPARC v3.3.1 to capture its continuous conformational dynamics30.

For the Omicron S–ACE2 dataset (Extended Data Fig. 3), 1,268,072 particles remained after reference-free 2D classification. After two rounds of 3D classification and further focused 3D classification on the RBD-1–ACE2 region, we obtained an Omicron S–ACE2 map from 141,538 particles. After Bayesian polishing and CTF refinement, the map was reconstructed to 3.53 Å resolution. We then focused on the RBD-2 region for further classification and obtained two conformations, with RBD-2 in the ‘down’ position (43.9%), termed S–ACE2-C1, or in the ‘up’ position. The RBD-2 up dataset was further focused 3D classified on the RBD-3 region. We then obtained two conformations with RBD-3 in the down or up position, termed S–ACE2-C2 and S–ACE2-C3, respectively. The three datasets were independently loaded into cryoSPARC v3.3.1 and refined using non-uniform refinement to 3.69 Å, 3.70 Å and 4.04 Å resolution, respectively. Here, after obtaining the 3.53 Å resolution map of Omicron S–ACE2, we performed further local refinement on the RBD-1–ACE2 region (indicated by a black dotted ellipsoid in Extended Data Fig. 3a) in cryoSPARC v3.3.1 to acquire a 3.67 Å resolution map of this region.

For the Omicron S–S3H3 dataset (Extended Data Fig. 5), a similar data processing procedure was adapted as described for the Omicron S dataset to obtain a 3.56 Å resolution S–S3H3 map from 238,121 particles. We then carried out focused 3D classification on the RBD-1 region, followed by non-uniform refinement in cryoSPARC v3.3.1, and obtained a 3.48 Å resolution S-open–S3H3 map from 162,221 particles and a 3.64 Å resolution S-close–S3H3 map from 75,900 particles. In addition, after obtaining the 3.56 Å resolution map, we performed focused 3D classification on the S3H3–SD1 region of protomer 2 (highlighted by an orange dotted ellipsoid in Extended Data Fig. 5), leading to a dataset of 101,192 particles, which was further local refined on the S3H3–SD1 region in cryoSPARC v3.3.1, deducing a 3.61 Å resolution map of this region. All of the obtained maps were post-processed through deepEMhancer53.

Atomic model building

To build an atomic model for the Omicron S-open structure, we used the atomic model of Delta S-open (PDB: 7W92) from our previous study as the initial model31. We first fit the model into our Omicron S-open map in Chimera by rigid body fitting, then manually substituted the mutations of the Omicron variant in COOT54. Subsequently, we flexibly refined the model against our Omicron S-open map using ROSETTA55. Finally, we used the phenix.real_space_refine module in Phenix 1.19.2-4158 for the S trimer model refinement against the map56. For the S-close model, we utilized the down protomer from our recent Delta S-transition (PDB: 7W94)31 structure as the initial template, and followed a similar procedure described above for model refinement. For the Omicron S–ACE2 and the local refined RBD-1–ACE2 structures, we used the Delta S–ACE2 model (PDB: 7W98 and 7W9I)31 as the initial template, and followed a similar procedure described above for model refinement. For better fitting in the dynamic ACE2 region of the Omicron S–ACE2-C1, C2, and C3 maps, we merged the better resolved Omicron RBD-1–ACE2 model with the other portion of the original model to make a complete model, and refined it against the corresponding unsharpened map. For the Omicron S–S3H3 and the local refined RBD-1–S3H3 structures, we utilized our recent Beta S–S3H3 model (PDB: 7WDF and 7WD8)46 as the template, and followed a similar procedure described above for model refinement. The atomic models were validated using the Phenix.molprobity command in Phenix. Analyses of the interaction interface were conducted through the PISA server57.

UCSF Chimera and ChimeraX were applied for figure generation, rotation measurement and Coulombic potential surface analysis58,59.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this paper.