Structural and functional ramifications of antigenic drift in recent SARS-CoV-2 variants

The COVID-19 pandemic has already lasted for more than a year, but new infections are still escalating throughout the world. Although several different COVID-19 vaccines have been deployed globally, a major concern is the emergence of antigenically distinct severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) variants of concern (VOCs). In particular, the B.1.1.7 lineage that arose in the UK (1) and quickly became dominant, the B.1.351 (also known as 501Y.V2) lineage in South Africa (2), the B.1.1.28 lineage (and its descendant B.1.1.28.1, also known as P.1/501Y.V3) in Brazil (3), and B.1.232/B.1.427/B.1.429 (also called CAL.20C and CAL.20A) in the United States (4) have raised serious questions about the nature, extent, and consequences of antigenic drift in SARS-CoV-2. In the receptor binding site (RBS) of the spike (S) protein receptor-binding domain (RBD), the B.1.1.7 lineage has acquired an Asn⁵⁰¹ → Tyr (N501Y) mutation, the B.1.351 and P.1 lineages share this mutation along with Lys⁴¹⁷ → Asn/Thr (K417N/T) and Glu⁴⁸⁴ → Lys (E484K), whereas the California variants have a Leu⁴⁵² → Arg mutation that is also present in the B.1.617 variant, which was first isolated in India, with Glu⁴⁸⁴ → Gln (5). E484K has also been detected in a few B.1.1.7 genomes (1) (Fig. 1A). We therefore investigated the structural and functional consequences of such mutations on neutralizing antibodies (nAbs) isolated from COVID-19 convalescent patients, as well as their effect on angiotensin-converting enzyme 2 (ACE2) receptor binding.

N501Y was previously reported to enhance binding to the human ACE2 receptor (6, 7). Here, we quantified binding of K417N, E484K, N501Y, and double and triple combinations in the RBD to ACE2 by biolayer interferometry (Fig. 1A and fig. S1). N501Y indeed increased RBD binding to ACE2 relative to wild-type RBD (binding affinity K_D = 3.3 nM versus 7.0 nM), whereas K417N substantially reduced ACE2 binding (41.6 nM). E484K slightly reduced binding (11.3 nM). N501Y could rescue binding of K417N (9.0 nM), and the triple mutant K417N/E484K/N501Y (as in B.1.351) had similar binding (6.5 nM) to the wild type (Fig. 1A and fig. S1). Consistently, K417N/T mutations are associated with N501Y in naturally circulating SARS-CoV-2. Among 585,054 SARS-CoV-2 genome sequences in the GISAID database (5 March 2021) (8), about 95% of K417N/T mutations occur with N501Y, despite N501Y being present in only 21% of all analyzed sequences. In contrast, only 36% of E484K mutations occur with N501Y.

We and others have shown that most SARS-CoV-2 nAbs that target the RBD and their epitopes can be classified into different sites and subsites (9–12). Certain IGHV genes are highly enriched in the antibody response to SARS-CoV-2 infection, with IGHV3-53 (11, 13–16) and IGHV3-66, which differ by only one conservative substitution (V12I), and IGHV1-2 (11, 17, 18) being the most enriched IGHV genes used among 1593 RBD-targeting antibodies from 32 studies (11, 14–44) (Fig. 1B). We investigated the effects of the prevalent SARS-CoV-2 mutations on neutralization by these major classes of antibodies found in multiple donors, and the consequences for current vaccines and therapeutics.

K417N and E484K in VOCs B.1.351 and P.1 have been reported to decrease the neutralizing activity of sera as well as neutralizing monoclonal antibodies (mAbs) isolated from COVID-19 convalescent plasma and vaccinated individuals (45–62). Relative to wild-type SARS-CoV-2, B.1.351 is more resistant to neutralization by convalescent plasma and mRNA vaccine sera by a factor of ~8 to 14, and P.1 is more resistant by a factor of ~2.6 to 5 (63–68). Some variants are able to escape neutralization by some nAbs (e.g., LY-CoV555, 910-30, COVOX-384, S2H58, C671, etc.), whereas others retain activity (e.g., 1-57, 2-7, mAb-222, S309, S2E12, COV2-2196, C669, etc.) (50, 57, 63, 66, 69, 70). Here, we selected a representative panel of 17 human nAbs isolated from COVID-19 patients or humanized mice to study the escape mechanism. These nAbs cover all known neutralizing sites on the RBD and include those encoded by V genes that are the most frequently used and also significantly enriched (Fig. 1B). We tested the activity of a panel of nAbs against wild-type (Wuhan strain) SARS-CoV-2 pseudovirus and single mutants K417N and E484K (Fig. 1C). Binding and neutralization of four and five antibodies out of the 17 tested were abolished by K417N and E484K, respectively. Strikingly, binding and neutralization by all six highly potent IGHV3-53 antibodies (71) that we tested were abrogated by either K417N (RBS-A/class 1) or E484K (RBS-B/class 2) (Fig. 1, C and D, and fig. S2). In addition, binding and neutralization of IGHV1-2 antibodies was severely reduced for the E484K mutation (Fig. 1, C and D, and fig. S2).

We next examined 54 SARS-CoV-2 RBD-targeting human antibodies with available structures. The antibody epitopes on the RBD can be classified into six sites—four RBS subsites, RBS-A, B, C, and D; CR3022 site; and S309 site (fig. S3) (72)—that are generally related to the four classes assigned in (10) (Fig. 1C). Of 23 IGHV3-53/3-66 antibodies, 21 target RBS-A (Fig. 2A). All IGHV1-2 antibodies with known structures bind to the RBS-B epitope. A large fraction of antibodies in these two main families make contact with Lys⁴¹⁷, Glu⁴⁸⁴, or Asn⁵⁰¹ in their epitopes (Fig. 2A) (73). Almost all RBS-A antibodies interact extensively with Lys⁴¹⁷ and Asn⁵⁰¹, whereas most RBS-B and RBS-C antibodies contact Glu⁴⁸⁴, and most RBS-C antibodies interact with Leu⁴⁵². We also examined the buried surface area (BSA) of Lys⁴¹⁷, Glu⁴⁸⁴, and Asn⁵⁰¹ upon interaction with these RBD-targeting antibodies (fig. S3C). The extensive BSA confirmed why mutations at positions 417 and 484 affect binding and neutralization. Antibodies targeting RBS-D, or the cross-neutralizing S309 and CR3022 sites, are minimally or not involved in interactions with these four RBD mutations (Fig. 2A and fig. S3C).

IGHV3-53/3-66 RBD antibodies can adopt two different binding modes (9, 10), which we refer here to as binding modes 1 and 2 (74), with distinct epitopes and approach angles (Fig. 2B and fig. S4). All IGHV3-53/3-66 RBD antibodies to date with binding mode 1 have a short, heavy-chain complementarity-determining region 3 (CDR H3) of <15 amino acids (Kabat numbering) and bind RBS-A (13, 16, 32), whereas those with binding mode 2 have a longer CDR H3 (≥15 amino acids) and target RBS-B (9, 10, 75). These dual binding modes enhance recognition of this antibody family for the SARS-CoV-2 RBD, although most IGHV3-53/3-66 RBD antibodies adopt binding mode 1 (Fig. 2 and fig. S4). Lys⁴¹⁷ is a key epitope residue for antibodies with IGHV3-53/3-66 binding mode 1 (Fig. 2B and fig. S4). IGHV3-53 germline residues V_H Tyr³³ and Tyr⁵² make hydrophobic interactions with the aliphatic moiety of Lys⁴¹⁷, and its ε-amino group interacts with CDR H3 through a salt bridge (Asp⁹⁷ or Glu⁹⁷), hydrogen bond (H-bond), or cation-π interaction (Phe⁹⁹) (Fig. 2B). K417N/T would diminish such interactions and therefore affect antibody binding and neutralization, providing a structural explanation for K417N escape in IGHV3-53/3-66 antibodies with binding mode 1 (Fig. 1, C and D, Fig. 2B, and fig. S2). In contrast, IGHV3-53 antibodies with binding mode 2 do not interact with RBD-Lys⁴¹⁷ (fig. S4) but with Glu⁴⁸⁴ through H-bonds with CDR H2 (Fig. 2C). Consistently, binding and neutralization of IGHV3-53 antibodies with binding mode 2 (fig. S4) are abolished by E484K but not K417N (Fig. 1, C and D, and fig. S2). Interestingly, unlike most IGHV3-53 antibodies that are sensitive to K417N/T or E484K, a recently discovered IGHV3-53-encoded mAb-222, which binds RBS, retains activity against P.1 and B.1.351. The mAb-222 light chain could largely restore the neutralization potency of other IGHV3-53 antibodies, which suggests that light-chain interactions can compensate for loss of binding of K417N/T by the heavy chain (63). However, this antibody may represent only a small portion of IGHV3-55/3-66 antibodies that can neutralize VOCs.

Among the IGHV genes used in RBD antibodies, IGHV1-2 is also highly enriched over the baseline frequency in the antibody repertoire of healthy individuals (76) and is second only to IGHV3-53/3-66 (Fig. 1B). We compared three structures of IGHV1-2 antibodies, namely 2-4 (27), S2M11 (30), and C121 (10), that target RBS-B. Despite being encoded by different IGK(L)V genes, 2-4 (IGLV2-8), S2M11 (IGKV3-20), and C121 (IGLV2-23) share a nearly identical binding mode and epitope (Fig. 3A). Structural analysis reveals that the V_H Gly²⁶-Tyr-Thr-Phe-Thr-Gly-(Tyr)-Tyr³³, Trp⁵⁰-(Ile)-Asn/Ser-(Pro)-X-Ser-X-Gly-Thr⁵⁷, and Thr⁷³-Ser-(Ile)-Ser/Thr⁷⁶ motifs are important for RBD binding (fig. S5, A to D). Although only a small part of the epitope interacts with the light chains of 2-4, S2M11, and C121, V_L residues 32 and 91 (and also residue 30 in some antibodies) play an important role in forming a hydrophobic pocket together with V_H residues for binding RBD-Phe⁴⁸⁶, which is another key binding residue in such classes of antibodies (9) (fig. S5, E to I). Three other IGHV1-2 antibodies, 2-43, 2-15, and H4, also bind in a similar mode (77), further highlighting the structural convergence of IGHV1-2 antibodies in targeting the same RBD epitope. All IGHV1-2 antibodies to date form extensive interactions with Glu⁴⁸⁴ (Fig. 3A and fig. S3C). In particular, germline-encoded V_H Tyr³³, Asn⁵² (somatically mutated to Ser⁵² in C121), and Ser⁵⁴ are involved in polar interactions with the RBD-Glu⁴⁸⁴ side chain that would be altered by substitution with Lys (Fig. 3A) and thereby diminish binding and neutralization of IGHV1-2 antibodies against E484K (Fig. 1, C and D, and fig. S2).

We previously isolated another potent IGHV1-2 antibody, CV05-163, targeting the SARS-CoV-2 RBD (fig. S6) from a COVID-19 patient (17). CV05-163 likely represents a shared antibody response for IGHV1-2 RBD antibodies across patients (fig. S7). Negative-stain electron microscopy (nsEM) of CV05-163 in complex with the SARS-CoV-2 S trimer illustrates that it can bind in various stoichiometries, including molar ratios of 1:1, 2:1, and 3:1 (Fab to S protein trimer), and can accommodate RBDs in both up and down conformations (fig. S8). We also determined a crystal structure of Fab CV05-163 with SARS-CoV-2 RBD and Fab CR3022 to 2.25 Å resolution (Fig. 3B, figs. S9 to S11, and tables S1 and S2) and found that it does indeed bind RBS-B (Fig. 3) and makes extensive interactions with Glu⁴⁸⁴ through H-bonds (V_H Trp⁵⁰ and V_H Asn⁵⁸) and a salt bridge (V_L Arg⁹¹) (Fig. 3B); these interactions explain why CV05-163 binding and neutralization were diminished with E484K (Fig. 1, C and D, and fig. S2). However, CV05-163 is rotated 90° (Fig. 3B) relative to IGHV1-2 antibodies 2-4, S2M11, and C121 (Fig. 3A). Thus, IGHV1-2 antibodies, akin to IGHV3-53/66 (75), can engage the RBD in two different binding modes, both of which are susceptible to escape by E484K but not by K417N (Fig. 1, C and D).

A further group of antibodies target the back side of the RBS ridge (RBS-C) (9). To date, five nAbs isolated from COVID-19 patients are known to bind RBS-C: CV07-270 (17), BD-368-2 (38), P2B-2F6 (18), C104 (10), and P17 (78). These RBS-C nAbs also interact with Glu⁴⁸⁴ (Fig. 4A), mainly through an arginine in CDR H3, which suggests that E484K may have an adverse impact on RBS-C antibodies. Indeed, binding and neutralization by CV07-270 was abrogated by E484K (Fig. 1C and fig. S2). Intriguingly, these five RBS-C antibodies are encoded by five different IGHV genes (79), but they all target a similar epitope with similar angles of approach. In addition, neutralization by REGN10933, a potent antibody used for therapeutic treatment, was reduced to a lesser extent by K417N and E484K (Fig. 1C) (28). REGN10933 binds at a slightly different angle from RBS-A antibodies and other RBS-B antibodies. Lys⁴¹⁷ then interacts with CDRs H1 and H3 of REGN10933, whereas Glu⁴⁸⁴ contacts CDR H2 (fig. S12). Overall, our results demonstrate that RBS mutations K417N and E484K can either abolish or extensively reduce the binding and neutralization of several major classes of SARS-CoV-2 RBD antibodies.

Two other non-RBS sites that are distant from Lys⁴¹⁷ and Glu⁴⁸⁴ have been repeatedly shown to be neutralizing sites on the SARS-CoV-2 RBD, namely the CR3022 cryptic site and the S309 proteoglycan site (9) (Fig. 1C and Fig. 4B). Antibodies from COVID-19 patients can neutralize SARS-CoV-2 by targeting the CR3022 site, including COVA1-16 (80), S304, S2A4 (31), and DH1047 (41). Recently, we isolated antibody CV38-142 that targets the S309 site (81). Antibodies targeting these two epitopes are often cross-reactive with other sarbecoviruses, as these sites are more evolutionarily conserved than the RBS. To test the effect of the K417N and E484K mutations on nAbs that target the S309 and CR3022 sites, we assessed binding and neutralization by CV38-142 and COVA1-16 to SARS-CoV-2. Both mutations have minimal effect on these antibodies (Fig. 1C and Fig. 4C).

The most potent neutralizing antibodies to SARS-CoV-2 generally tend to target the RBS (table S3), as they directly compete with receptor binding. Such RBS antibodies often interact with Lys⁴¹⁷, Glu⁴⁸⁴, or Asn⁵⁰¹ and are therefore sensitive to RBS mutations at these positions in the VOCs. On the other hand, antibodies targeting the CR3022 and S309 sites are often less potent but are less affected by the VOCs, as their epitopes do not contain mutated residues. In fact, recent studies have shown that sera from convalescent or vaccinated individuals can retain neutralization activity, albeit reduced, against the mutated variants (48, 50, 82), which is possibly due to antibodies targeting other epitopes including the CR3022 and S309 sites. Thus, the CR3022 and S309 sites are promising targets to avoid interference by SARS-CoV-2 mutations observed to date.

As SARS-CoV-2 continues to circulate in humans and increasing numbers of COVID-19 vaccines are administered, herd immunity to SARS-CoV-2 should be approached locally and globally. However, as with other RNA viruses, such as influenza and HIV (83), further antigenic drift is anticipated in SARS-CoV-2. Within-host antigenic drift has also been observed in an immunosuppressed COVID-19 patient who had low titers of neutralizing antibodies that allowed the emergence of N501Y and E484K mutations (84). Whereas antibody responses elicited by the wild-type lineage that initiated the COVID-19 pandemic have been well characterized in natural infection (11, 14–18, 20, 23, 26, 27, 29) and vaccination (50, 85–87), data on the immune response to VOCs are now only starting to emerge (88) and will clarify the similarity and differences in the antibodies elicited. Because SARS-CoV-2 is likely to become endemic (89), the findings here and in other recent studies can be used to fast-track the development of more broadly effective vaccines and therapeutics.