SARS-CoV-2 Omicron-B.1.1.529 leads to widespread escape from neutralizing antibody responses

Summary On 24th November 2021, the sequence of a new SARS-CoV-2 viral isolate Omicron-B.1.1.529 was announced, containing far more mutations in Spike (S) than previously reported variants. Neutralization titers of Omicron by sera from vaccinees and convalescent subjects infected with early pandemic Alpha, Beta, Gamma, or Delta are substantially reduced, or the sera failed to neutralize. Titers against Omicron are boosted by third vaccine doses and are high in both vaccinated individuals and those infected by Delta. Mutations in Omicron knock out or substantially reduce neutralization by most of the large panel of potent monoclonal antibodies and antibodies under commercial development. Omicron S has structural changes from earlier viruses and uses mutations that confer tight binding to ACE2 to unleash evolution driven by immune escape. This leads to a large number of mutations in the ACE2 binding site and rebalances receptor affinity to that of earlier pandemic viruses.

In brief A comprehensive analysis of sera from vaccinees, convalescent patients previously infected by multiple variants, and potent monoclonal antibodies from early in the COVID-19 pandemic reveals a substantial overall reduction in the ability to neutralize the SARS-CoV-2 Omicron variant, which seems to ameliorate with a third vaccine dose. Structural analyses of the Omicron RBD suggest that selective pressure balances key changes that increase affinity for ACE2 with other changes in the receptor-binding motif that disfavor ACE2 binding but facilitate immune escape.

INTRODUCTION
Since the end of 2020, a series of viral variants have been emerging in different regions, and some have caused large out-breaks. Alpha  and, more recently, Delta , have had the greatest global reach, whilst Beta , Gamma , and Lambda (Colmenares-Mejía et al., 2021), despite causing large outbreaks in Southern Africa and South America, did not become dominant in other parts of the world. Indeed, Beta was later displaced by Delta in South Africa.
The rapid emergence of Omicron (https://www.who.int/news/ item/26-11-2021-classification-of-omicron-(b.1.1.529)-sars-cov-2-variant-of-concern) in the background of high Beta immunity implies that the virus may have evolved to escape neutralization by Beta-specific serum . Within Spike (S), Omicron has 30 substitutions plus a deletion of 6 and an insertion of 3 residues, whereas in all the other proteins there are a total of 16 substitutions and 7 residue deletions. Particular hotspots for the mutations are the angiotensin converting enzyme 2 (ACE2) receptor binding domain (RBD) (15 amino acid substitutions) and the N-terminal domain (NTD) (3 deletions totaling 6 residues, 1 insertion, and 4 substitutions).
S mediates cellular interactions. It is a dynamic, trimeric structure (Walls et al., 2017Wrapp et al., 2020), which can be lipid bound (Toelzer et al., 2020) and tightly associated in a ''closed'' form or unfurled to expose one or more RBDs, allowing both receptor binding and increased access to neutralizing antibodies. Once bound to a cell, S undergoes cleavage and a drastic elongation, converting it to the post-fusion form.
Most potent neutralizing antibodies target the ACE2 footprint Lan et al., 2020;Liu et al., 2021b), occupying $880 Å 2 at the outermost tip of the RBD (the neck and shoulders, referring to the torso analogy ) and preventing cell attachment. A proportion of antibodies are able to cross-neutralize different variants , and a few of these bind to a motif surrounding the Nlinked glycan at residue 343 Liu et al., 2021b). These latter antibodies, exemplified by S309 (Pinto et al., 2020), can cross-react with SARS-CoV-1 but do not block ACE2 interaction, and destabilizing the S-trimer may be their mechanism of action. Neutralizing anti-NTD mAbs do not block ACE2 interaction and bind to a so-called supersite on the NTD (Cerutti et al., 2021;Chi et al., 2020); however, they generally fail to provide a broad protection as the supersite is disrupted by a variety of NTD mutations present in the variants of concern (VOC). Moreover, some NTD-binding antibodies were shown to have an infectivity-enhancing effect by inducing the open form of S (Liu et al., 2021c).
In this report, we study the neutralization of Omicron by a large panel of sera collected from convalescents of early pandemic, Alpha-, Beta-, Gamma-, and Delta-infected individuals, together with vaccinees who had received three doses of the Oxford/As-traZeneca (AZD1222) or the Pfizer BioNtech (BNT16b2) vaccines. There is a widespread reduction in the neutralization activity of sera from multiple sources, and we use these data to plot an antigenic map, where Omicron is seen to occupy the most distant position from early pandemic viruses, which form the basis for current vaccines.
We show that Omicron escapes neutralization by the majority of potent monoclonal antibodies (mAbs) arising after both early pandemic and infection with Beta variant. Utilizing a large bank of structures (n = 29) from panels of potent mAbs, which includes most of the mAbs developed for prophylactic or therapeutic use, we describe the mechanism of escape caused by the numerous mutations present in Omicron RBD (Baum et al., 2020).
Analysis of the binding of ACE2 to RBD and structural analysis of the Omicron RBD indicate that changes at residues 498 and 501 of the RBD have locked ACE2 binding to the RBD in that region sufficiently strongly to enable the generation of a plethora of less favorable changes elsewhere, providing extensive immune escape and, in the process, resulting in a final net affinity for ACE2 similar to the early pandemic virus.

Phylogeny of Omicron
Omicron has changes throughout its proteome, but S changes dominate, with 30 amino acid substitutions plus 6 residues deleted and 3 inserted (Figures 1 and 2). Ten of these were found previously in at least two lineages (D614G was mutated early on and maintained throughout). Of those ten, six have the same amino acid substitution in >75% of the sequences, and only one (E484A) has a unique substitution in Omicron (in Beta and Gamma it is a Lys). Figure S1A shows the number of mutant sequences per residue at positions undergoing mutations in independent lineages. This can be interpreted in two ways: one is that the later mutations are epistatic to one another and thus are more difficult to reach, or that they do not contribute to virus fitness.
The Omicron RBD has 15 changes in total, as described in the next section. The NTD also has numerous changes, including 4 amino acid substitutions, 6 amino acids deleted, and 3 amino acids inserted, also described in the next section. Several mutations found in Omicron occur in residues conserved in SARS-CoV-1 and in many other Sarbecoviruses (Figure 1). These observations agree with the Pango classification (Rambaut et al., 2020), which places Omicron at a substantial distance from all other variants.
Mapping of Omicron RBD mutations compared with Alpha, Beta, Gamma, and Delta The Alpha variant has a single change in the RBD at N501Y (Figure 2D;Supasa et al., 2021), which occupies the right shoulder and contributes to the ACE2 binding footprint. Beta has two further mutations in the RBD: K417N and E484K, at the back of the neck and left shoulder, respectively ( Figure 2E), which are also part of the ACE2 footprint ( Figure 2C; Zhou et al., 2021). Gamma mutations are similar: K417T, E484K, and N501Y . Delta mutations: L452R in front of the neck, and T478K on the far side of the left shoulder, fall just peripheral to the ACE2 binding footprint ( Figure 2F; Liu et al., 2021a). All of these variants have at least one RBD mutation in common with Omicron. Of the 15 Omicron changes in the RBD, nine map to the ACE2 binding footprint: K417N, G446S, S477N, E484A, Q493R, G496S, Q498R, N501Y, Y505H, with N440K and T478K just peripheral ( Figures 2B and 2C). Additionally, mutations occur on the right flank: G339D, S371L, S373P, and S375F ( Figure 2B), the last three of which are adjacent to a lipid-binding pocket ( Figure S1B) (Toelzer et al., 2020;Carrique et al., 2020). This pocket has been seen occupied by a lipid similar to linoleic acid in an unusually rigid state of S, where all RBDs are found in a locked-down configuration stabilized by lipid-bridged quaternary interactions between adjacent RBDs. However, this lipid-bound form has been seen rarely; instead, the pocket is usually empty and collapsed, with the RBD alternating between looser down and up conformations.
We presume that this is because the pocket readily empties of lipid during protein purification. Indeed, rapidly prepared virus particles tend to have RBDs closer to locked-down state (Ke et al., 2020). Loss of lipid promotes RBD presentation to the target cell.
Until now, the antigenic properties of variant viruses have been well described by assuming that each mutation produces only a local change in structure, and we used this assumption to rationalize the serological impact of the mutations in Omicron. We present structural data later to qualify this assumption.

Omicron NTD mutations
The mutations seen in the NTD lie on exposed flexible loops, which differ from those in SARS-CoV-1 and are likely antigenic ( Figure 1A). The pattern of deletions and insertions seen in Omicron consistently changes those loops that are most different from SARS-CoV-1 to be more SARS-CoV-1-like, at least in length. Of the N1, N3, and N5 loops, which comprise the antibody supersite, Omicron has a substitution at G142D and deletion of residues 143-145 in N3, which would mitigate against binding by a number of potent neutralizing antibodies, e.g., 4A8 and mAb 159 (Chi et al., 2020;Dejnirattisai et al., 2021b). The deletion of residues 69 and 70 in N2 has also occurred in the Alpha variant, whereas the deletion at residue 211, the substitution at 212, and the insertion at 214 are unique to Omicron. All these changes are on the outer surface and are likely antigenic.

Neutralization of Omicron by convalescent serum
We isolated Omicron virus from the throat swab of an infected case in the UK. Following culture in VeroE6 cells transfected with TMPRSS2, the S gene sequence was confirmed to be the Omicron consensus with the additional mutation A701V, which is present in a small number of Omicron sequences.
We have collected convalescent serum/plasma with the indicated median day of sampling, from individuals infected early in the pandemic (n = 32, median day 42) before the emergence of the VOC, along with cases infected with Alpha (n = 18, median day 18), Beta (n = 14, median day 61), Gamma (n = 16, median day 63), and Delta (n = 42, median day 38). Neutralization assays were performed against Omicron and compared with neutralization titers for Victoria (an early pandemic strain), Alpha, Beta, Gamma, and Delta.
In summary, Omicron causes widespread escape from neutralization by serum obtained following infection by a range of SARS-CoV-2 variants, meaning that previously infected individuals will have little protection from infection with Omicron, although it is hoped that they will still maintain protection from severe disease.

Vaccination and infection in combination increases Omicron neutralization titers
We have collected sera from Delta-infected cases and because Delta spread in the UK during the vaccination campaign, we Shown with Alpha, Beta, Delta, and Omicron variants (Omicron repeated on the lower line for clarity). Binding sites for the early pandemic potent antibodies  and the potent Beta antibodies  are depicted using iron heat colors (black < straw < yellow < white) to indicate relative levels of antibody contact and commercial antibody contacts are depicted with the pairs of antibodies in red and blue (purple denotes common interactions). Totally conserved residues are boxed on a red background on the upper rows, while on the final row Omicron mutations are boxed in red. Secondary elements are denoted above the alignment. Figure produced in part using ESPript (Robert and Gouet, 2014). obtained sera from three different groups-Delta infection only (n = 19) ( Figure 3E), Delta infection following vaccination (n = 9), and vaccination following Delta infection (n = 8) ( Figure 3F). Neutralization assays against early pandemic, Alpha, Beta, Gamma, Delta, and Omicron viruses were performed. Compared with Delta-only infected individuals, sera from cases who had received the vaccine and been infected by Delta showed substantially higher neutralization to all viruses tested-early pandemic, with Delta-infected and vaccinated sera showing a 7.9-fold (p < 0.0001) increase in the neutralization of Omicron compared with Delta infection alone. To confirm the boosting effect of vaccination, we collected a paired blood sample from 6 Delta cases before and after vaccination, which clearly demonstrated the boosting effect of infection and vaccination ( Figure S2).  (Pettersen et al., 2021). Related to Figure S1.

Increased neutralization of Omicron by third-dose booster vaccination
In a number of countries, booster programs have been launched to counter waning immunity and the increasing frequency of breakthrough infections with Delta. To examine the effect of booster vaccination, we tested neutralization of Victoria, Delta, and Omicron viruses using sera from individuals receiving 3 doses of ADZ1222 (n = 41) or BNT162b2 (n = 20). For ADZ1222, the serum was obtained 28 days following the second and third doses ( Figure 3G). For BNT162b2, the serum was obtained 28 days, 6 months, immediately prior to the third dose, and 28 days following the third dose ( Figure 3H).
At 28 days following the third dose, for ADZ1222, the neutralization titer to Omicron was reduced 12.7-fold (p < 0.0001) compared with Victoria and 3.6-fold (p < 0.0001) compared with Delta; for BNT162b2, the neutralization titer to Omicron was reduced 14.2-fold (p < 0.0001) compared with Victoria and 3.6-fold (p < 0.0001) compared with Delta. The neutralization titers for Omicron were boosted 2.7-fold (p < 0.0001) and 34.2fold (p < 0.0001) following the third dose of ADZ1222 and BNT162b2, respectively, compared with 28 days following the second dose. Of concern, and as has been noted previously, neutralization titers fell substantially between 28 days and 6 months following the second dose of the BNT162b2 vaccine, although we did not measure titers at 6 months following the second dose of AZD1222.
In summary, neutralization titers against Omicron are boosted following a third vaccine dose, meaning that the campaign to deploy booster vaccines should add considerable protection against Omicron infection. Article Effect of Omicron mutations on antibodies elicited by early pandemic virus We have previously reported a panel of 20 potent neutralizing antibodies (50% focus reduction in neutralization test [FRNT50] < 100 ng/mL) isolated from cases infected with early pandemic viruses (Wuhan) . Neutralization assays against Omicron were performed and compared with neutralization assays of early pandemic, Alpha, Beta, Gamma, and Delta , (F) Delta before vaccination or Delta after vaccination (n = 17), (G) before and after the third dose of AZD1222 (n = 41), and (H) 4 weeks, 6 months after the second dose, before the third, and after the third dose of BNT162b2 (n = 20). In (A-E) comparison is made with neutralization titers to Victoria, Alpha, Beta, Gamma, and Delta previously reported in Dejnirattisai et al. (2021aDejnirattisai et al. ( , 2021b, Supasa et al. (2021), Zhou et al. (2021), and Liu et al. (2021b), in (G) the data points for Victoria and Delta titers on BNT162b2 are taken from Flaxman et al. (2021). Geometric mean titers are shown above each column. The Wilcoxon matched-pairs signed rank test was used for the analysis, and two-tailed p values were calculated.
The binding sites of these antibodies were mapped together with other published structures to 5 epitopes (based on the position of the center of gravity of each antibody) either by direct structural studies or competition analyses . According to the torso analogy , these were designated as follows: neck, left shoulder, right shoulder, right flank, and left flank ( Figure 2B). In Figures  5A-5D, we show the mapping of the density of centroids to the surface of the RBD with the Omicron mutations shown as spikes (the information is also mapped to the primary structure in Fig Dejnirattisai et al., 2021a;Huang et al., 2021); therefore, this cryptic epitope might be an important target for therapeutic antibody applications and cross-protective vaccine antigen (Pinto et al., 2020). We demonstrate the continued binding of this class of antibodies later on.
Nineteen of the twenty most potent (FRNT50 < 100 ng/mL) neutralizing mAbs are mapped to the ACE2 binding site across the neck and shoulder epitopes of the RBD, and 5 of these are classified as public IGVH3-53 (immunoglobulin heavy chain variable gene family) antibodies Yuan et al., 2020). Mapping these onto the RBD surface ( Figure 5B) shows that the centroids are highly concentrated in the neck region. IGVH3-53 mAbs were especially common in early pandemic responses, and although their centroid is at the neck, they are orientated in such a way that their light chain CDRs interact with the right shoulder ( Figure S3). Most IGVH3-53 mAbs are sensitive to the N501Y mutation, although some, such as mAb 222 or Beta-27, can still neutralize 501Y-containing viruses Liu et al., 2021b). Omicron mutation Y505H has a direct interaction with the L1 and L3 CDRs of mAb 222  and, together with Q493R, is likely responsible for the 12.6-fold reduction in the neutralization titer of mAb 222 ( Figure 4A; Table S1).
The neutralizing activity of mAbs 88, 316, and 384 is knocked out for Omicron ( Figure 4A; Table S1); all interact with E484 (mAb 316 via H1 and H2) within the left shoulder epitope, and the E484A mutation is unfavorable. For mAb 316, Q493R will also likely be deleterious due to contacts with H1 and H3. Broadly neutralizing mAb 58 binds at the front of the RBD, reaching toward the right flank in an area that is relatively clear of mutations and thus is unaffected ( Figure S3). MAb 278 binds more of the right shoulder, with L3 in contact with G446, and the G446S mutation in Omicron knocks out activity ( Figure S3).
MAb 170 will be affected by Q493R and Q498R, which directly interact with L1 and H3, respectively ( Figure S3). Q498R is between G496S and G446S ( Figure 2B), and G446 is in proximity to H1; together, these mutations knock out the activity of mAb 170 ( Figure 4A; Table S1). The binding sites of selected potent antibodies are shown in Figure 5E. All of these, with the exception of mAb58, are affected by the mutations in Omicron. To understand the resilience of mAb58, we determined the structure of a ternary complex of an early pandemic RBD with Fabs for mAbs 58 and 158 (Table S2), confirming that its epitope includes no residues mutated in Omicron ( Figure S3).

Effect of Omicron mutations on antibodies elicited by the Beta variant
We derived a panel of 27 potent Beta antibodies (FRNT < 100 ng/ mL) , and this revealed a surprisingly skewed response with 18/27 potent antibodies targeting the Beta mutations: E484K, K417N, and N501Y. This is seen in Figure 5C, where the focus on residues in the shoulders has spread the centroid patch out toward several Omicron mutation sites. This information is mapped to the primary structure in Figure 1A, and a schematic of the binding of the four potent cross-reactive antibodies is shown in Figure 5F. While K417N and N501Y are conserved in Omicron, E484 is mutated to an alanine, which seems a likely escape mutation from either 484E (early pandemic/Alpha) or 484K (Beta).
A large number of Beta mAbs target the 501Y mutation, including a public antibody response mediated through IGVH4-39 (n = 6) and the related IGVH4-30 (n = 1) . Many are likely to be sensitive to the numerous mutations in this region: N440K, G446S, Q493R, G496S, Q498R, and Y505H. In total, 11 antibodies contact 501Y;  and -29 bind epitopes dependent on 417N/T together with 501Y (antibodies in italic are VH4-39 or VH4-30 and the neutralization of Omicron for those in bold is completely knocked out). Beta mAbs targeting the back of the neck epitope (Beta-22, -29, and -30) will be affected, for example, in the case of Beta-29, H1 makes extensive interactions with residues Q493, G496, and Y505 ( Figure S4).

Figure 4. mAb neutralization curves
(A-C) FRNT curves for mAb from (A) early pandemic, (B) Beta infected cases or (C) commercial sources. Omicron neutralization is compared with curves for Victoria, Alpha, Beta, Gamma, and Delta, which have been previously reported Supasa et al., 2021;Zhou et al., 2021;Liu et al., 2021b). Neutralization titers are reported in Table S1. Related to Figure S2. ll OPEN ACCESS Figure 5. Relative antibody contact (A-D) RBD surface produced in PyMOL and rendered in mabscape using iron heat colors (gray < blue < glowing red < yellow < white) to indicate relative levels of antibody contact. Antibody contact is calculated for each surface vertex as the number of antibodies within a 10 Å radius by their known or predicted positions from earlier mapping studies Liu et al., 2021b). Outward facing cones are placed at the nearest vertex to each mutated residue on the RBD surface. Drawn back and front views for (A) all RBD-reactive antibodies isolated from early pandemic, (B)strongly neutralizing antibodies (<100 ng/mL) from Article Beta-44 binding to the left shoulder epitope has already been shown to be sensitive to T478K, while the combination of S477N and T478K in Omicron is likely to be more deleterious. Interestingly, several of the antibodies (Beta-40, -54, -55, -56, and Beta-22 and -29 [501Y 417N/T]) retain some activity, and this is explained later on with reference also to the structure of the Omicron RBD/Fab 55 complex.
Four Beta mAbs potently cross-neutralize all Alpha, Beta, Gamma, and Delta variants   Figure 5F). Of these, Beta-27 is a VH3-53 antibody that contacts Q493 and Y505 in a similar way to mAb222 and shows reduced neutralization of Omicron ( Figure 4B; Table S1). Beta-47, a VH1-58 antibody, has contacts with S477 and Q493, likely leading to the observed reduction in neutralization of Omicron.
Beta-49 and -50, which belong to the IGVH1-69 gene family, bind similarly to the right flank and are knocked out by Omicron ( Figure 4B; Table S1). They lie directly on RBD G339 and would clash with G339D. Beta-53 also binds to the right flank, with H1 contacting residue 339 and likely clashing with G339D. L1 likely contacts G446S, leading to the observed two-log reduction in Omicron neutralization compared with Beta ( Figure S4).

Effect of Omicron mutations on current antibody therapeutics
Various individual antibodies or cocktails of antibodies (usually recognizing different epitopes to reduce the risk of escape [Sun et al., 2021]) have been licensed for use, and the aggregate of their binding is shown in Figure 5G. This illustrates the strong correlation of binding with sites of mutation (this is mapped to the primary structure in Figure 1A) and neutralization of Omicron is markedly reduced in most ( Figure 4C; Table S1). Specifically, they are as follows: Regeneron 10933 and 10987: Regeneron 10933 (Weinreich et al., 2021) binds to the back of the left shoulder and 10987 to the right shoulder ( Figures 5G and S5); activity of both is knocked out on Omicron ( Figure 4C). REGN10933 is unable to effectively neutralize Beta, being sensitive to E484K , and H2 contacts Q493, so that neutralization of Omicron is almost completely lost. REGN10987 contacts N440 and G446 causing complete loss of neutralization ( Figure S5).
AZD8895 and AZD1061: AZD8895, a VH1-58 antibody, binds to the back of the left shoulder and activity on Omicron is reduced 230-fold compared with Victoria due to contacts with S477 (H3) and Q493 (H2). AZD1061, binding the front of the right shoulder is reduced 268-fold (Figures 5G and S5) due to L2 and H3 contacts with the G446 loop. AZD7442 (a combination of AZD8895 and AZD1061) maintains neutralizing activity against Omicron, although reduced 30.3-fold compared with Victoria.
LY-CoV016 and 555: The activity of both antibodies on Omicron is knocked out. LY-CoV016 is a VH3-53 antibody and extensive interactions with N501 and Y505 via L1 and L3 make it vulnerable to mutations at these residues ( Figures 5G and  S5). LY-CoV555 (Sun and Ho, 2020) is sensitive to the E484K mutation in Delta  and also contacts Q493.
ADG 10, 20, 30: All Adagio antibodies suffer considerable loss of activity against Omicron ( Figure 4C). The activity of ADG10 and ADG30 were completely lost, while ADG20 activity was reduced 276-fold.
Effect on RBD/ACE2 interaction Fitness of a virus can stem from higher infectivity or evasion of the immune system. One way to identify mutations that increase binding affinity is by selection, using a randomly mutated RBD displayed on the yeast surface for ACE2 binding to obtain the highest affinity clone RBD-62. Mutations fixed for higher affinity binding included N501Y, E484K, S477N, and most prominently, Q498R ( Figure 6A; Zahradník et al., 2021b). Interestingly, Q498R was selected only at later stages. This is explained by the 2-fold reduction in affinity as a single mutation ( Figure 6A). However, in combination with the N501Y mutation, the affinity is increased 26-fold, more than any other mutation analyzed. Adding to this, the S477N mutation, one obtains a 37-fold increase in binding ( Figure 6B). These three mutations, selected through in vitro evolution, were found together for the first time in the Omicron variant.
We measured the affinity of Omicron RBD for ACE2 using SPR and yeast display titration. Perhaps surprisingly, the affinity was on par with that of the early virus, 8 and 7 nM, respectively, using SPR ( Figures 6B and S6A) and 2.9 and 1.9 nM using yeast display titration (SPR and yeast display titration data strongly correlate but with a constant shift in absolute values [Zahradník et al., 2021b]). This implies that the increased affinity imparted by S477N, Q498R, and N501Y is being offset by other mutations in the ACE2 footprint. We measured the affinities of the other single mutations in the ACE2 binding footprint of Omicron (using yeast display titration), as shown in Figures 6B and 6C, and they provide a rationale for this. T478K in the presence of N501Y decreased the positive effect of the latter by 2-fold. Y505H reduces the binding of Q498R, N501Y double mutant by 50%. G496S and the triple-mutation S371L, S373P, and S375F reduce binding by 2-and 2.2-fold, respectively. The effect of changing the triple-mutant (S371L, S373P, and S375F) back to the wildtype sequence was even more pronounced in the background of Omicron, in which the affinity increased from 2.9 to 0.4 nM (7fold). Moreover, this back-to-wild-type triple-mutant increased the expression on the surface of yeast 10-fold relative to Omicron. This indicates a functional role in increasing the fitness of the virus for this triple-mutant, which requires the binding enhancement provided by the Q498R, N501Y double mutant. E484A (instead of the Lys found in other variants, Figure 6A) was neutral. While K417N (found in the Beta variant) on its own , blue (75%-50%), and magenta (50%-25%). Information about the distribution and frequency of S-protein mutations and the spatiotemporal characterization of SARS-CoV-2 lineages were retrieved from www.outbreak.info (Mullen et al., 2020) and GISAID database (Elbe and Buckland-Merrett, 2017).* Same evolutionary origin, a number of evolutionary non-related lineages with given or similar mutation (Zahradnik et al., 2021c), b log(10) number of the observed Omicron mutation at the given position as determined on 14 th November 2021, c same as b but total log(10) number of changes at the given position. d Fold-change in binding as determined by yeast-surface display. Fold-change is the ratio between original RBD KD and the mutant RBD KD for binding human ACE2. Article decreases binding substantially, the effect on binding when combined with other mutations is smaller ( Figure 6B). Two single mutations found specifically in Omicron, Q493R, and N440K did increase binding, probably due to increasing the electrostatic complementarity between ACE2 (negatively charged) and the RBD (positively charged) ( Figure 6D).
Comparing the structure of the complex of the pM affinity RBD-62 with ACE2 (Zahradnik et al., 2021b;PDB:7BH9) to that of Omicron, RBD bound to Beta-55 antibody (described later on, see Table S2; Figure 6C) shows high similarity with an RMSD of 0.55 Å over 139 residues. Importantly, the locations of the binding-enhancing mutations 477N, 498R, and 501Y are conserved between the two, despite the RBD-62 being bound to ACE2, while Omicron RBD is not. This shows that these residues are pre-arranged for tight binding, implying low entropic penalty of binding.

Antigenic cartography
We used the matrix of neutralization data generated in Figure 3 to place Omicron on an antigenic map, with a method similar to that developed for analysis of the Delta variant , where we model individual viruses independently and allow for serum-specific scaling of the responses (STAR Methods). This model works well; the measured and modeled responses are shown in Figures 7A and 7B (with 1,600 observations and 215 parameters, the residual error is 9.1%). The results are well described in three dimensions (see Video S1) and are shown projected into two dimensions in Figures 7C and 7D. It will be seen that the previous variants are placed in a planetary band around a central point, with Delta opposed to Beta and Gamma; however, Omicron is displaced a large distance out of this plane, almost on a line drawn from the central point perpendicular to the planetary band, illustrating vividly how Omicron dramatically expands our view of the antigenic landscape of SARS-CoV-2.
The structural impact of the numerous mutations in S We first used Alphafold2 (Jumper et al., 2021) to predict the Omicron RBD structure ( Figure S1B). The top-ranked structure was very similar to the early pandemic RBD (RMSD for Cas 0.71 Å , residues 334-528), with a significant difference in the region of the triple serine mutations 371-375, on the right flank (Figure S1B). We then determined the high-resolution crystal structure of the Omicron RBD domain in complex with two Fabs: Beta-55 and EY6A ( Figure S6B and Table S2) Liu et al., 2021b;Zhou et al., 2020). The RBD structure is indeed close to that of early pandemic viruses (RMSD 0.9 Å for 187 Ca) with the only significant change at the 371/373/375 triple serine mutations ( Figure 7E). The rearrangement in this region is essentially an amplified version of that predicted by Alphafold2, suggesting that such algorithms have some value in predicting the effect of dense mutations as seen in Omicron RBD. The mutations S371L, S373P, and S375F are all changes from small, flexible polar serine residues to bulkier, less flexible hydrophobic residues. Interestingly, all the Omicron S mutations involve single codon changes apart from S371L, which requires two changes from TCC to CTC, indicative of underlying strong selection pressure and functional change. Although the rearrangement in Omicron is quite modest, it is exactly this region of the structure that undergoes a larger conformation change when lipid is bound into the pocket ( Figure S1B). Changes in the serine-rich loop allow the attached helix to swing out, opening the pocket for lipid binding. It is possible that the increased rigidity and the entropic penalty of exposing hydrophobic residues may disfavor lipid binding to Omicron, which would alter the properties of the virus, explaining the selection of these changes.
The binding of EY6A to the left flank of the RBD is essentially unchanged from that observed previously  (dissociation constant [K D ] 7.8 and 6.8 nM for early pandemic and Omicron RBDs, respectively, by SPR) (Figures 7F and  S6A). This cryptic epitope, which is highly conserved for functional reasons, is a good target for broadly neutralizing therapeutic antibodies.
Beta-55, as predicted earlier , binds to the right shoulder, around residue 501. Interestingly, the epitope includes several residues mutated in Omicron from the early pandemic virus (including Q498R, N501Y, and Y505H) ( Figures  7E, S6B, and S6C). It is remarkable that despite these significant changes, neutralization is relatively little affected. The neutralization result was confirmed by measurements of the binding affinity, 177 pM and 204 pM for the early pandemic and Omicron RBDs, respectively ( Figure S6A). To confirm the structural basis, we also determined the crystal structure of an analogous ternary complex formed with early pandemic RBD (Table S2). As expected, the details of the interaction are essentially identical. If we extend the analysis of the 501Y targeting antibodies by comparing the structures of Beta-6, -24, -40, and -54, we find subtle explanations; thus, Beta-24 and some others are knocked out due to a clash with CDR-L1 created by the Q493R mutation showing sera as columns against challenge variants as rows. Sera are grouped into blocks according to the eliciting variant. The reference neutralization titer for each block is calculated as the average of all titers when challenged with the variant that elicited the serum. In the case of vaccine sera this was taken as the average of all best neutralization titers. Therefore, colors within a single block express the relative neutralization titer with respect to this reference. (B) Shows an example of the equivalent model generated from one run of antigenic map refinement using the same reference offsets as calculated for (A). (C) Shows a view of the three-dimensional antigenic map for variants of concern. The distance between two points corresponds to the drop-off in neutralization titer used in (B). (D) Same antigenic space as (C) but rotated 90 , to look downward form Omicron. (E) Overlay of the X-ray structure of Omicron (red) on the early pandemic (Wuhan) RBD (gray) and the predicted model of the Omicron RBD in black, drawn as cartoons. The structural change effected by the S371L, S373P, and S375F mutations is shown enlarged in the inset. (F) X-ray structure of ternary complex of Omicron RBD with Beta-55 and EY6A Fabs. The Omicron RBD is shown as a gray semi-transparent surface with mutated residues in magenta. Fabs are drawn as cartoons, heavy chain in magenta and light chain in blue. Related to Figures S1 and S6 and Video S1.
( Figure S6D), whereas for antibodies Beta-40, -54, and -55, this mutation can be accommodated. In addition, the Q498R mutation may create a hydrogen bond in Beta-40 or a salt bridge in Beta-54 to CDR-H3, which may compensate for the loss of binding affinity due to changes around residue 501 ( Figure S6E). Thus, the surprising resilience of several of the 501Y targeting antibodies may be because the mutated residues in this region are not ''hotspots'' of interaction, and mutations can sometimes be accommodated without significant impact on affinity. This may suggest that a major driver for evolution was the less 501focused responses to early viruses.

DISCUSSION
The first 4 Omicron sequences were deposited on 24 th November 2021. Within days, distant international spread was seen, and has caused great concern due to its high transmissibility and ability to infect previously exposed or vaccinated individuals. Only 3 weeks after the virus was first detected in the UK, Omicron cases outnumbered Delta in London, with the number of daily new cases in the UK larger than that recorded during any other previous time in the pandemic. Over the next weeks, disease severity will become clearer.
The density of mutational changes (including deletions and insertions) found in Omicron S is extraordinary, being more than three times that observed in previous variants. Within S, as observed for other variants, the NTD, RBD, and the furin cleavage site region are hotspots for mutation (Zahradník et al., 2021b), and within the RBD, mutations are concentrated on the ACE2 interacting surface and the right flank.
Most potent neutralizing antibodies bind on or in proximity to the ACE2 footprint (neck and shoulder epitopes) and block interaction of S with ACE2, thereby preventing viral attachment to the host cell. There are two other classes of potent neutralizing mAbs, first antibodies binding in close proximity to the N343 glycan (right flank epitope) exemplified by Vir S309 (Pinto et al., 2020), which includes the Beta-49, -50, and -53 antibodies  used in our analysis. These mAbs bind distant from the ACE2 binding site, do not block ACE2 interaction, and destabilize the S-trimer, which may be their mechanism of action. Finally, antibodies binding to the supersite on the NTD can also be potently neutralizing, although the mechanism of action of NTD antibodies remains obscure (Cerutti et al., 2021;Chi et al., 2020;Dejnirattisai et al., 2021a). Multiple mutations at all three of these sites-the receptor-binding site, proximal to N343 glycan, and NTD-are found in Omicron and lead to substantial reduction in neutralization titers for naturally immune or vaccine sera, with many showing complete failure of neutralization. This, together with the widespread failure of potent mAb to neutralize Omicron, points to a driver of immune evasion for their evolution.
The left flank epitope, which is not mutated in Omicron, is used by antibodies that do not block ACE2 binding but are protective (Barnes et al., 2020;Dejnirattisai et al., 2021a;Hastie et al., 2021;Huang et al., 2021;Zhou et al., 2020). Here, we demonstrate structurally and by affinity measurements that this epitope is conserved and unchanged in Omicron.
Following repeated rounds of selection by yeast display for high ACE2 affinity, RBD-62 (I358F, V445K, N460K, I468T, T470M, S477N, E484K, Q498R, and N501Y) emerged as the highest affinity clone with a 1,000-fold increase in affinity for ACE2 from 17 nM for Wuhan RBD to 16 pM for RBD-62. It is striking that the key contributors for the high affinity of RBD-62 are present in Omicron. Interestingly, the combination of mutations K417M, E484K, Q493R, Q498R, and N501Y also emerged after 30 passages in mouse lungs (Roy Wong et al., 2021). This mouse-adapted virus was highly virulent and caused more severe disease. The appearance of E484K, Q493H/R, Q498R, and N501Y in yeast display and mouse adaptation experiments are strong indications that the tighter binding to ACE2 also facilitates more efficient transmission.
However, in Omicron, overall affinity for ACE2 is not increased, suggesting a different strategy. Since mutations S477N, Q498R, and N501Y are likely to increase ACE2 affinity by 37-fold, we hypothesize that these changes, also found in RBD-62, serve to anchor the RBD to ACE2, leaving the rest of the receptor-binding motif more freedom to develop further mutations, including those that reduce ACE2 affinity, in a quest to evade the neutralizing antibody response. Indeed, K417N, T478K, G496S, Y505H, and the triple S371L, S373P, S375F reduce affinity to ACE2 while driving immune evasion. All this is achieved with very minimal structural changes in the isolated Omicron RBD ( Figure 7E).
These observations provide a valuable lesson on the plasticity of protein-protein binding sites, maintaining nM binding affinity (Cohen-Khait and Schreiber, 2016). Thus, the extreme concentration of potent neutralizing antibodies around the 25 amino acid receptor footprint of ACE2 suggests that this would be an Achilles heel for SARS-CoV-2, with ACE2 placing constraints on its variability (this is why receptor-binding sites are often hidden [Rossmann et al., 1985]). However, in practice, the extraordinary plasticity of this site, allowing it to absorb mutational change while retaining affinity for ACE2, is a potent weapon to evade the antibody response. Such camouflage of receptorbinding sites has been observed before (see, for example, Acharya et al., 1989), but it seems that by acquiring a lock on the ACE2 receptor at one point, through 498 and associated mutations, many other less energetically favorable changes can be tolerated, fueling antigenic escape. Thus, by mutating the receptor-binding site, the virus can modulate ACE2 affinity and potentially transmissibility while evading the antibody response.
How Omicron evolved is under debate. The results presented here suggest that immune evasion is a primary driver in its evolution, sacrificing affinity-enhancing mutations to optimize immune-evading mutations. This could, for instance, occur in a single immunocompromised individual, with further evolution in rural, unmonitored populations (Clark et al., 2021). Virus evolution has been previously observed in chronically infected HIV+ individuals and other immunocompromised cases, leading to the expression of the N501Y, E484K, and K417T mutations (Cele et al., 2021;Karim et al., 2021;Kemp et al., 2021). What seems beyond doubt from the ratio of nonsynonymous to synonymous mutations (only one synonymous mutation in all of S) is that the evolution has been driven by strong selective pressure on S. It has been predicted that increasing immunity by natural infection or vaccination will increase the selective pressure to find a susceptible host, either by increased transmissibility or antibody evasion. It appears that Omicron has achieved both ll OPEN ACCESS of these goals, although our data only speak directly to antibody evasion.
In addition to changes in the ACE2 footprint, Omicron RBD possesses a triplet of mutations from serines to more bulky, hydrophobic residues, a motif not found in any other Sarbecoviruses. This introduces structural changes and may lead to loss of the ability to form the lipid-binding pocket, which might normally aid release of the virus from infected cells. One of these mutations requires a double change in the codon, reinforcing its significance, and it is conceivable that there is synergy with the change at residue 498, perhaps explaining why this mutation has not established itself earlier.
For most mAbs, the changes in interaction are so severe that the activity is completely lost or greatly impaired. This also extends to the set of mAbs developed for clinical use-the activity of most is lost, AZD8895 and ADG20 activity is substantially reduced, whereas the activity of Vir S309 is more modestly reduced.
Omicron has now got a foothold in many countries. In the UK, it has an estimated doubling time of 2.5 days and 2 doses of vaccine appear to give low protection from infection, whereas 3 doses give better protection. There is considerable concern that Omicron will rapidly replace Delta and cause a large and sharp peak of infection in early 2022. It is likely that substantial increases in transmissibility and immune evasion are contributing to the explosive rise in Omicron infections. At present, the only option to control the spread of Omicron, barring social distancing and mask wearing, is to pursue vaccination with Wuhan-containing antigen to boost the response to sufficiently high titers to provide some protection. However, the antigenic distance of Omicron may mandate the development of vaccines against this strain. There will then be a question of how to use these vaccines; vaccination with Omicron will likely give good protection against Omicron but will not give good protection against other strains. Therefore, it seems possible that Omicron may cause a shift from the current monovalent vaccines containing Wuhan S to multivalent vaccines containing an antigen, such as Wuhan or Alpha, at the center of the antigenic map and Omicron or other S genes at the extreme peripheries of the map, similar to the polyvalent strategies used in influenza vaccines.
In summary, we have presented data showing that the huge number of mutational changes present in Omicron lead to a substantial knockdown of neutralizing capacity of immune serum and failure of mAb. This appears to lead to a fall in vaccine effectiveness, but it is unlikely that vaccines will completely fail and it is hoped that although vaccine breakthroughs will occur, protection from severe disease will be maintained, perhaps by T cells. It is likely that the vaccine-induced T cell response to SARS-CoV-2 will be less affected than the antibody response. Third-dose vaccine boosters substantially raise neutralization titers to Omicron and are the mainstay of the response to Omicron in countries, such as the UK. Widespread vaccine breakthroughs may mandate the production of a vaccine tailored to Omicron, and failure of mAbs may likewise lead to the generation of secondgeneration mAbs targeting Omicron.
A question asked after the appearance of each new variant is whether SARS-CoV-2 has reached its limit for evolution. Analyzing the mutations in Omicron shows that, except for S371L, all other mutations require only single-nucleotide changes. Two-nucleotide mutations and epistatic mutations are more difficult to reach, but they open up vast untapped potential for future variants. Global control measures are critical to avoid this.

Limitations
The neutralization assays presented in this paper are performed in vitro and do not fully quantify the antibody response in vivo, where complement and antibody-dependent cell-mediated cytotoxicity may contribute to virus control. Evasion of the antibody response may allow reinfection with Omicron, but the role of the T cell response, which is not measured here, is likely to contribute to the control of infection and disease severity.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

DECLARATION OF INTERESTS
G.R.S. sits on the GSK Vaccines Scientific Advisory Board and is a founder member of RQ Biotechnology. J.Z. and G.S. declare the Israel patent application no. 23/09 /2020-277,546 andUnited States patent application no. 16/12/ 2020-63/125,984, entitled methods and compositions for treating coronaviral infections. Oxford University holds intellectual property related to the Oxford-Astra Zeneca vaccine. A.J.P. is Chair of UK Dept. Health and Social Care's (DHSC) Joint Committee on Vaccination & Immunisation (JCVI) but does not participate in the JCVI COVID-19 committee, and is a member of the WHO's SAGE. The views expressed in this article do not necessarily represent the views of DHSC, JCVI, or WHO. The University of Oxford has entered into a partnership with AstraZeneca on coronavirus vaccine development. The University of Oxford has protected intellectual property disclosed in this publication. S.C.G. is co-founder of Vaccitech (collaborators in the early development of this vaccine candidate) and is named as an inventor on a patent covering use of ChAdOx1-vectored vaccines and a patent application covering this SARS-CoV-2 vaccine (PCT/GB2012/000467). T.L. is named as an inventor on a patent application covering this SARS-CoV-2 vaccine and was a consultant to Vaccitech for an unrelated project during the conduct of the study. S.J.D. is a Scientific Advisor to the Scottish Parliament on COVID-19.

Lead contact
Resources, reagents and further information requirement should be forwarded to and will be responded by the lead contact, David I Stuart (dave@strubi.ox.ac.uk).

Materials availability
Reagents generated in this study are available from the lead contact with a completed Materials Transfer Agreement.

Data and code availability
The coordinates and structure factors of the crystallographic complexes are available from the PDB with accession codes given in Table S2. Mabscape is available from https://github.com/helenginn/mabscape, https://snapcraft.io/mabscape. The data that support the findings of this study are available from the corresponding authors on request.
Virus-containing supernatants were harvested at 80% CPE and spun at 3000 rpm at 4 C before storage at -80 C. Viral titres were determined by a focus-forming assay on Vero cells. Victoria passage 5, Alpha passage 2 and Beta passage 4 stocks Gamma passage 1, Delta passage 3 and Omicron passage 1 were sequenced to verify that they contained the expected spike protein sequence and no changes to the furin cleavage sites.

Cloning of Spike and RBD
Expression plasmids of wild-type and Omicron spike and RBD were constructed encoding for human codon-optimized sequences from wild-type SARS-CoV-2 (MN908947) and Omicron (EPI_ISL_6640917). Wild-type Spike and RBD plasmids were constructed as described before . Spike and RBD fragments of Omicron were custom synthesized by GeneArt (Thermo Fisher Scientific GENEART) and cloned into pHLsec and pNEO vectors, respectively, as previously described Supasa et al., 2021;Zhou et al., 2021). Both constructs were verified by Sanger sequencing after plasmid isolation using QIAGEN Miniprep kit (QIAGEN).

Protein production
Protein expression and purification were conducted as described previously Zhou et al., 2020). Briefly, plasmids encoding proteins were transiently expressed in HEK293T (ATCC CRL-11268) cells. The conditioned medium was concentrated using a QuixStand benchtop system. His-tagged Omicron RBD were purified with a 5 mL HisTrap nickel column (GE Healthcare) and further polished using a Superdex 75 HiLoad 16/60 gel filtration column (GE Healthcare). Twin-strep tagged Omicron spike was purified with Strep-Tactin XT resin (IBA lifesciences).
$4mg of ACE2 was mixed with homemade His-tagged 3C protease and DTT (final concentration 1mM). After incubation at 4 C for one day, the sample was flowed through a 5 mL HisTrap nickel column (GE Healthcare). His-tagged proteins were removed by the nickel column and purified ACE2 was harvested and concentrated.

IgG mAbs and Fab purification
To purify full length IgG mAbs, supernatants of mAb expression were collected and filtered by a vacuum filter system and loaded on protein A/G beads over night at 4 C. Beads were washed with PBS three times and 0.1 M glycine pH 2.7 was used to elute IgG. The eluate was neutralized with Tris-HCl pH 8 buffer to make the final pH=7. The IgG concentration was determined by spectrophotometry and buffer exchanged into PBS.
To express and purify Fabs 158 and EY6A, heavy chain and light chain expression plasmids of Fab were co-transfected into HEK293T cells by PEI. After culturing cells for 5 days at 37 C with 5% CO2, culture supernatant was harvested and filtered using a 0.22 mm polyethersulfone filter. Fab 158 was purified using Strep-Tactin XT resin (IBA lifesciences) and Fab EY6A was purified with Ni-NTA column (GE HealthCare) and a Superdex 75 HiLoad 16/60 gel filtration column (GE Healthcare).
AstraZeneca and Regeneron antibodies were provided by AstraZeneca, Vir, Lilly and Adagio antibodies were provided by Adagio. For the antibodies heavy and light chains of the indicated antibodies were transiently transfected into 293Y cells and antibody purified from supernatant on protein A. Fab fragments of 58 and beta-55 were digested from purified IgGs with papain using a Pierce Fab Preparation Kit (Thermo Fisher), following the manufacturer's protocol.

Surface Plasmon Resonance
The surface plasmon resonance experiments were performed using a Biacore T200 (GE Healthcare). All assays were performed with a running buffer of HBS-EP (Cytiva) at 25 C.
To determine the binding kinetics between the SARS-CoV-2 RBDs and ACE2 / monoclonal antibody (mAb), a Protein A sensor chip (Cytiva) was used. ACE2-Fc or mAb was immobilised onto the sample flow cell of the sensor chip. The reference flow cell was left blank. RBD was injected over the two flow cells at a range of five concentrations prepared by serial twofold dilutions, at a flow rate of 30ml min À1 using a single-cycle kinetics programme. Running buffer was also injected using the same programme for background subtraction. All data were fitted to a 1:1 binding model using Biacore T200 Evaluation Software 3.1.
X-ray data collection, structure determination, and refinement Crystals were mounted in loops and dipped in solution containing 25% glycerol and 75% mother liquor for a second before being frozen in liquid nitrogen. Diffraction data were collected at 100 K at beamline I03 of Diamond Light Source, UK. All data were collected as part of an automated queue system allowing unattended automated data collection (https://www.diamond.ac.uk/Instruments/ Mx/I03/I03-Manual/Unattended-Data-Collections.html). Diffraction images of 0.1 rotation were recorded on an Eiger2 XE 16M