Since emerging in late 2019 severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2) is estimated to have led to 772 million infections and 7 million deaths (https://covid19.who.int/). Effective vaccines and natural infection have led to increasing herd immunity, greatly reducing mortality but not preventing infection. SARS-CoV-2 therefore, still causes significant mortality in high-risk groups such as the immunosuppressed or elderly who show reduced or absent responses to vaccination.
All SARS-CoV-2 vaccines are designed to induce an antibody response to the spike protein (S), which is delivered in a variety of formats; RNA, viral vector, protein, or inactivated virus1–3. Analysis of panels of S-specific mAbs by several laboratories has led to considerable understanding of the antigenic landscape of S and sites of mAb binding4–7. To date two domains of the S1 region of spike, N-terminal (NTD)8 and receptor binding (RBD)7,9 have been described as binding sites for potent neutralizing mAbs. Most potent anti-RBD mAbs bind on or near the receptor binding motif5, a small 25 amino acid patch at the tip of the RBD. mAbs binding here neutralize by blocking binding of S to the SARS-CoV-2 cellular receptor angiotensin converting enzyme 2 (ACE-2)10. A second group of anti-RBD antibodies, exemplified by S309, bind close to the N-linked glycan at residue N453, do not block interaction with ACE2 and may act to destabilize the S-trimer11. Potent anti-NTD mAbs bind to a so-called supersite on the NTD, do not block ACE2 interaction and their mechanism of action is not well understood8. All commercially developed mAbs to date target the RBD12–14, whilst anti-NTD mAbs tend to be variant specific, due to extensive mutation of the supersite between variants.
Both the RBD and NTD are hot spots of mutation in SARS-CoV-2. RBD mutations can impart selective advantages to the virus, firstly some, for example the N501Y mutation found in the Alpha variant, increase the affinity to ACE2 and are believed to drive increased transmissibility15. Secondly, mutations in the RBD and NTD may lead to escape from neutralizing antibody responses16. Increasing herd immunity generates intense selective pressure on the virus to break through pre-existing immunity17. Evolution of S has therefore been rapid with many mutations mapping closely to the sites of interaction of potent mAbs in the RBD and NTD16. SARS-CoV-2, in a period of 4 years since its emergence, has evolved variants that escape all mAbs developed for clinical use.
In this study we report the generation of a panel of mAbs from vaccinated volunteers who suffered breakthrough BA.4/5 infections. We characterise these structurally and functionally and document a steady attrition of activity to a succession of variants culminating in the loss of activity of the whole panel to some currently circulating strains.
Generation of mAbs from BA.4/5 infection samples
Blood was taken from 11 triple vaccinated volunteers > 23 days (median 38) after PCR test confirmed SARS-CoV-2 BA.4 infection (n = 3) or BA.5 infection (n = 8). Focus reduction neutralisation tests (FRNT) were performed on Victoria (an early pandemic strain), together with BA.2, BA.4 and BA.5 live virus, to select the highest titre samples for antibody production (Fig. 1A). Seven samples with the highest titres against BA.4 or BA.5 were used for mAb production.
PBMCs were stained with BA.4/5 S trimer (the S sequence is the same for BA.4 and BA.5) and single IgG positive memory B cells were sorted (Fig. 1B). Meanwhile, 4 samples were stained with an S trimer which we term BA.4 + all, constructed to harbour additional mutations seen in recent sub-lineages (G339H, R346T, L368I, K444R, V445P, G446S, N450D, L452M, N460K, V483A, E484R, F486S, F490V and S494P) in a BA.4/5 background. From the selected cells, a degenerate PCR reaction was used to amplify heavy and light chains, which were assembled into an expression vector using Gibson assembly and the products expressed by transient transfection. Supernatants were tested for binding to full length BA.4/5 or BA.4 + all S trimer, BA.4/5 or BA.4 + all RBD and BA.4/5 NTD. From 861 sorted cells 414 antibodies were recovered leading to the selection of 28 potent RBD or NTD binding mAbs (BA.4/5–34, 35 and 36 were derived from the BA.4 + all sort) showing 50% focus reduction neutralization titres (FRNT) of BA.5 virus < 100 ng/ml. Heavy chain (HC) gene usage corresponded to: IGHV1-69 (5/28), IGHV3-9 (4/28), and public gene family IGHV3-53 (4/28) and IGHV3-66 (5/28) (Fig. 1C, Table S1). Somatic mutation was comparable to a previous set of antibodies we developed following BA.1 infection, significantly greater than seen in early pandemic mAbs (Fig. 1D). Six mAbs showed little or no ACE2 blocking ability (BA.4/5 − 3, 4, 12, 15, 20 and 33) (Fig. 1E).
Cross-reactivity of anti-BA.4/5 antibodies
Pseudo-typed virus neutralisation assays18 were used to test the antibodies against 26 variants seen throughout the pandemic with particular emphasis on Omicron sub-lineages (Fig. 2). All potent BA.4/5 mAbs, except BA.4/5–33 and − 36, cross-neutralize early pandemic pseudovirus Victoria (IC50 < 100 ng/ml) and may have been selected from B cell clones originally generated following vaccination (Figure S1). BA.4/5–33 and − 36 were the only two anti-NTD antibodies we isolated and were specific to BA.2 derived variants.
Most BA.4/5 mAbs showed > 5-fold reduction of neutralization titre on at least one Omicron sub-lineage, compared to BA.4/5 (Fig. 2A). Against circulating variants isolated after the emergence of BA.4/5 the activities of these mAbs were completely knocked out for 12/28, 14/28, 18/28, and 16/28 on the variants BA.2.75.2, BQ.1.1, BN.1, and CH.1.1 respectively. Against the latest dominating variants (XBB and its sub lineages), the activities of the mAbs were also severely impaired, with 18/28 knocked out by XBB, XBB.1 and XBB.1.5. The most striking knockout was observed on XBB.1.5.10 and XBB.1.5.70, which have F456L and L455F + F456L (‘FLip’) mutations respectively on the XBB.1.5 background (Fig. 2B). XBB.1.5.70 containing the FLip mutation led knock out of activity of the few remaining antibodies retaining activity against XBB.1.5 including the broadly neutralizing mAbs BA.4/5 − 1 and BA.4/5 − 2 together with mAbs 22, 28, 31, 34 and 35, leading to knock out of activity of all 28/28 mAbs from the BA.4/5 panel.
BA.2.86 is a newly emerging variant first recognised in August 2023, compared to its closest ancestor, BA.2, it has 38 amino acid changes in S with 14 changes in RBD including a deletion (ΔV483). The large number of changes in S has led to the concern that it may show a greater level of immune escape. Here we tested BA.2.86 neutralisation by the BA.4/5 panel of mAbs, where activity was similar to XBB.1.5, likely because BA.2.86 lacks mutations at residues 455 and 456. However, mutation L455S has been observed in 856 reported BA.2.86 sequences and has been named JN.1. Neutralization assays show that JN.1 knocks out mAbs BA.4/5–8, -10, -28, and − 35, which show neutralizing activity against BA.2.86.
Structures of anti-BA.4/5 mAbs
To elucidate the binding mode and detailed interactions of the most broadly neutralizing anti-BA.4/5 mAbs, we determined crystal structures of complexes of Delta-RBD19 with BA.4/5 − 1 and EY6A20, Delta-RBD with BA.4/5 − 2 and Beta-4921, Delta-RBD with BA.4/5–9 and Delta-RBD with BA.4/5–35 (Figs. 3, 4, S2 and Table S2).
BA.4/5 − 1 (IGHV4-39) is an RBD binding antibody that, according to the visual analogy of the RBD to a human torso7, attaches at the back of the RBD, with the HC binding at the top of the neck and the LC at the back of the left shoulder, making a large footprint of 1310 Å2 (HC 670 Å2 and LC 640 Å2) (Fig. 3A, E, F, 4C and Figure S2A). This antibody heavily overlaps the ACE2 binding site; of the 35 RBD residues in the BA.4/5 − 1 footprint, 20 overlap with the ACE2 footprint. In line with this a large number of mutation sites in the Omicron variants have either direct contact with, or are on the footprint of BA.4/5 − 1, including 405, 408, 417, 476, 477, 484, 486, 490, 493–494, 498, 501 and 505, but interestingly most of these mutations have little or no effect on the neutralization potency of BA.4/5 − 1, which maintains quite broad cross-reactivity (Fig. 2A). RBD residues L455, F456, Y489 and Q493 make extensive hydrophobic interactions (≤ 4 Å) with CDR-H3, while F486, G476 and S477 contact CDR-L3 of BA.4/5 − 1. F486 also makes ring stacking contacts with Y35 and Y60 from the framework regions of the HC (Fig. 3E,F). Mutation of residue F486 has been observed in recently identified SARS CoV-2 variants: F486V in BA.4/5 and BQ.1, F486S in BA.2.75.2 and XBB and F486P in BA.2.10.4 and XBB.1.5 (Fig. 2B). It appears that despite close interaction with residue 486, BA.4/5 − 1 can tolerate mutations of F486 showing only modest reduction in titres to some of the newly described variants compared to BA.4/5 (Fig. 2A). Neutralization titres to BS.1 (a rarely described variant, which has failed to achieve a major break-through) were reduced 32-fold compared to BA.4/5 or BA.2.12.1. BS.1 contains the unique mutation G476S (Fig. 2B) compared to the other variants tested in Fig. 2A, G476 has close contact with Y92 of CDR-L3 and the change to Ser likely disrupts this interaction (Fig. 3F).
BA.4/5 − 2 (IGHV3-30), a broadly neutralizing mAb, binds the RBD with the HC at the back of the left shoulder and light chain at the back of the neck. The HC makes most contacts with the RBD, with a footprint of 650 Å2 compared to 305 Å2 for the LC (Fig. 3B, 4D and S2B). Of the 27 residues within this footprint, 12 overlap with the footprint of ACE2. CDR-H3 makes extensive interactions with RBD residues 416–417, 420–421, 455–460, 473 and 489. In contrast, CDR-L3 makes only weak contact to T415 of the RBD. CDR-H1 contacts residues 475–477 and 486–487 (Fig. 3G,H). Although residues 405, 408, 417, 460, 476–477 and 505, which are mutated in some recent variants (Fig. 2B), have direct contacts with BA.4/5 − 2, BA.4/5 − 2 retains the ability to broadly neutralize all the variants containing mutations at these residues except those also containing F456L or L455F and F456L mutations (Fig. 2A).
BA.4/5–9 (IGVH1-46) is bound similarly to BA.4/5 − 1 with its HC sitting on top of the neck of the RBD, but rotated by about 35º such that the LC locates at the back rather than at the back of the left shoulder as in BA.4/5 − 1, making a footprint of 1210 Å2 (770 Å2 by HC, 440 Å2 by LC) (Figs. 3C,I,J,4E and S2C). CDR-H3 contacts L455 and F456 through G101 and N103, while contacts to F456 from CDR-L3 are extensive, mainly from the main chain atoms of residues 92–94 and the Cβ of W94 (Fig. 3J). Interestingly BA.4/5–9 is not sensitive to the F456L mutation of XBB.1.5.10, but is sensitive to L455S in JN.1 whilst the L455F and F456L double mutation knocks out the activity of the mAb. CDR-H2 is located directly on top of Q493 making a single contact through P53 to the latter, and the Q493R mutation found in BA.1, BA.2 and some subvariants thereof diminishes the neutralization ability of BA.4/5–9. CDR-L1 interacts extensively with residues 415–416 and 420–421 none of which are in the ACE2 footprint and have not yet shown significant mutations.
BA.4/5–35 is highly unusual in that it belongs to the so-called public family of antibodies IGHV3-66, closely related to IGHV3-53, and binds a well characterised epitope (Fig. 3D, K-M, 4F and S2D). Such antibodies were very common in early responses5 and many escape mutations map to the shared epitope, leading to the general abolition of neutralisation. The structure reveals how BA.4/5–35 dodges these mutations to remain highly cross-reactive. Of published Fabs in the PDB the BA.4/5–35 HC is most similar to Omi-322 (83% sequence identity for Vh). A substitution (D100G) in the HC creates space for the side chain of residue W94 of the light chain to stack against the peptide of HC residue 100. This leads to a significant refolding of the LC CDR3 and positions the whole LC further away from the RBD. The overall effect is that HC contacts are maintained but the LC is lifted away from the escape mutations, conferring broad cross-reactivity (Fig. 3K-M). A recently described, but low frequency mutation in BQ.1.1, A475V which maps to the BA.4/5–35 binding site, leads to a significant fall in activity.
The L455F and F456L double “FLip” mutations knock out all BA.4/5 mAbs
Our structures show that the four most broadly neutralising anti-BA.4/5 mAbs have substantial (12 to 21 residues) overlap with the ACE2 footprint. Mutations of many of these footprint residues (417, 445–446, 460, 477, 484, 486, 498, 501 and 505) have no effect on these mAbs, whilst other mutations show moderate effects (3- to 7-fold for G476S in BS.1 variant on BA.4/5 − 1, BA.4/5 − 2 and BA.4/5–35) or target just one of the mAbs, (A475V impacts BA.4/5–35 and Q493R knocks out BA.4/5–9). However, all four Fabs have close contacts with residues 455 and 456 (Fig. 4) and the L455F and F456L double “FLip” mutations, present in some of the recent fastest spreading variants, knock out all of them. Whilst the F/L switch only requires a minimal third base switch in the codon (Phenyalanine: UUU or UUC, Leucine: UUA or UUG), the effect on the protein is a substantial change in side-chain volume, sufficient to disrupt surface complimentarity between tightly interacting surfaces whilst maintaining hydrophobic properties. Hydrophobic surface patches are typically hallmarks of protein-protein interfaces and this region is central to the ACE2 footprint, indeed these mutations have also been reported to increase ACE2 affinity23. We note that IGHV3-53/66 mAbs have become common among the potent antibodies induced from post-Omicron variant infections, e.g. 8 out of 26 BA.4/5 mAbs and 5 out of 10 XBB.1.5 mAbs24. These all have close contacts with 455 and 456. It is clear that the L455F and F456L mutations have been generated under the pressure of a group of antibodies bound at the back of the RBD and that these mutations act in synergy23.