Effect of natural mutations of SARS-CoV-2 on spike structure, conformation, and antigenicity

SARS-CoV-2 from alpha to epsilon As battles to contain the COVID-19 pandemic continue, attention is focused on emerging variants of the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) virus that have been deemed variants of concern because they are resistant to antibodies elicited by infection or vaccination or they increase transmissibility or disease severity. Three papers used functional and structural studies to explore how mutations in the viral spike protein affect its ability to infect host cells and to evade host immunity. Gobeil et al. looked at a variant spike protein involved in transmission between minks and humans, as well as the B1.1.7 (alpha), B.1.351 (beta), and P1 (gamma) spike variants; Cai et al. focused on the alpha and beta variants; and McCallum et al. discuss the properties of the spike protein from the B1.1.427/B.1.429 (epsilon) variant. Together, these papers show a balance among mutations that enhance stability, those that increase binding to the human receptor ACE2, and those that confer resistance to neutralizing antibodies. —VV

The emergence of rapidly-spreading variants of SARS-CoV-2, the causative agent for COVID-19, threatens to prolong an already devastating pandemic. Some variants have exhibited resistance in in vitro assays to neutralization by antibodies (Abs) and plasma from convalescent or vaccinated individuals, raising concerns that their resistance may reduce the efficiency of current vaccines (1, 2) (www.cdc.gov/coronavirus/2019-ncov/cases-updates/variantsurveillance/variant-info.html). Additionally, SARS-CoV-2 transmission between humans and animals has been observed in mink farms, leading to culling of large mink populations in Denmark and other countries to prevent establishment of a non-human reservoir of SARS-CoV-2 variants (3). Changes in the spike (S) glycoprotein (4,5) in these variants are under scrutiny due to the S protein's central role in engaging the angiotensin-converting enzyme 2 (ACE2) receptor to mediate cellular entry (6), and its being a dominant target of neutralizing antibodies (nAbs) elicited either by vaccination or natural infection (7,8).
Fall 2020 was marked by the appearance of several fastspreading SARS-CoV-2 variants with S protein variations accumulating in the background of the D614G substitution (24). Some amino acid substitutions recur in variants that originated independently in different geographical locations, suggesting convergent evolution and selective advantages of these changes. Here, we determine structures of, and measure ACE2 and Abs binding to S protein variants. These include a variant that was implicated in SARS-CoV-2 Effect of natural mutations of SARS-CoV-2 on spike structure, conformation, and antigenicity tralize variants that harbor the E484K substitution (2).
We tested several variants in the B.1.351 spike backbone (Fig. 1, figs. S2 and S3, and table S1). We found that the commonly occurring 242-244 deletion, and a rare R246I substitution that is included in some reagent panels and candidate vaccines (39), can each impact not only binding of NTD-directed Abs, but also of RDB-directed Abs DH1041 and DH1043. While binding of NTD-directed nAbs DH1050.1 and DH1050.2 to B.1.1.7 and B.1.351 spikes was dramatically reduced, binding to triple mutant spike and S-GSAS-P.1 (or "P.1-like spike") remained unchanged. This is consistent with neutralization data, where mAbs 5-24 and 4-8 that target the same antigenic supersite as DH1050.1, lost activity against B.1.351 but neutralized P.1 (40).
In summary, our binding data are consistent with biological data obtained in in vitro neutralization assays, thus establishing that our SARS-CoV-2 S ectodomain constructs are an effective mimic of native spikes and supporting their use for studying structural changes due to amino acid substitutions in spike variants.
We next sought to understand the effect of each amino acid substitution on the functional and structural properties of the spike. The ∆FV spike bound ACE2 with ~3.5-fold improved affinity than the D614G spike, resulting from a decreased off-rate, mediated by the Y453F substitution ( Fig.  2D, fig. S9, and table S2). While neither the I692V substitution or ∆H69/V70 affected ACE2 binding affinity, ∆H69/V70 contributed to increased affinity for NTD-directed nAbs DH1050.1 and DH1050.2. The I692V substitution occurs in SD2, where small changes can translate to large movements in the NTD and RBD regions ( Fig. 1) (16,19). In the D614G spike, I692 contacts P600; loss of the methyl due to the I692V substitution increases the distance between P600 and V692 (fig. S10). We observed disorder in the 3D-4 cryo-EM map, accompanied by the largest separation between P600 and V692 of all the ∆FV spike 3-RBD-down structures. This local destabilization around the I692V substitution in 3D-4, together with DDM comparisons and superpositions that showed 3D-4 to be the most asymmetric of the 3-RBD-down structures, and also the most variable in the S2 subunit, suggested a role for the I692V substitution in the 3-RBDdown state disorder.
To define and quantify changes in ∆FV spike domain orientations, and to determine how local changes around the SD2 I692V substitution propagate to adjacent domains, we examined its quaternary structure using a vector representation (19). This was accomplished by assigning a central coordinate to each domain and calculating angles, dihedrals, and distances between different structural elements ( Fig. 2E and supplementary text). Principal components analysis (PCA) of these intra-protomer vector relationships showed that the 3D-4 protomers occupied a distinct cluster (Fig. 2F), consistent with the DDM analysis ( fig. S7, B and C). The two RBD-down protomers in M1(A and C) were similar to 3D-1(A), 3D-2(C), 1U-1(A) and 2U(C) protomers along the first principal component (PC1), with M1(A) separating from M1(C) in PC2 into a 3D-1(A) containing cluster. Both 3D-1(A) and 3D-3(C) occupied extreme positions in the vector set for angles involving the NTD′, subdomains, and the RBD that mimic the 1U-1(A) structure (fig. S11). As constraints on RBD-down protomers are relaxed in spikes with at least one RBD in the "up" position, this may represent a particularly stable protomer position. Together, the vector clustering is consistent with structural observations for the 3D-4 structure and indicates that loss of a single S1 protomer in M1 allowed its two other RBD-down protomers to relax to a configuration resembling RBD-down protomers in 1-RBD-up spikes.
Comparing the ∆FV spike 3-RBD-down structures to our previously published D614G spike structures (PDB ID: 7KE4, 7KE6, 7KE7 and 7KE8) revealed that the 3D-1 and 3D-2 protomers closely matched 7KE4 and 7KE8, respectively, in their intra-protomer φ3 and θ 3 angles (fig. S11, A and B). Two protomers in the 3D-4(B and C) structure resembled two protomers in the 7KE8(A and B) D614G spike structure in their φ 3 dihedrals. Both the 7KE8 and 3D-4 structures displayed marked asymmetry, with the third protomer in each occupying an extreme dihedral angle; in 7KE8(C), the NTD and RBD are rotated toward S2, while 3D-4(A) showed a rotation in the opposite direction ( fig. S11C). Due to contact between SD1 and NTD′, this results in global shifts of S1 elements away from S2. These shifts, together with close contact between S2 and these S1 domains, results in changes in S2 structure leading to the variability observed in our structural analysis ( fig. S7). The large separation of S1 from S2 in the 3D-4(A) protomer (fig. S11C) suggests it could be an intermediate that leads to the S1-dislocated M1(B) state. The 3D-3 structure also lacked a close match (fig. S11, A and B). Alignment of 3D-3 to its most similar D614G down-state trimer structure, 7KE7, indicated similar, albeit less extreme differences in domains, thus suggesting 3D-3 to be yet an- Thus, by combining cryo-EM classifications and vector analysis we track the origin of the observed instability in the ∆FV spike and find evidence of instability in two 3-RBDdown structures (3D-3 and 3D-4) that leads to dislocation of a S1 protomer in M1.
In summary, our data show that interspecies adaptation involved improved receptor binding affinity of the ∆FV spike mediated primarily by the RBD Y453F substitution. The observed increase in RBD-up states may also contribute to higher levels of ACE2 binding by providing more receptor-accessible sites. We found no evidence in the binding data for immune evasion at the dominant neutralization sites; this is consistent with previous findings that neutralization potency of a panel of RBD antibodies was not significantly impacted by Y453F or ∆H69/V70 (38). Structural analysis revealed destabilization of the 3-RBD-down state, and loss of tight regulation of its conformation in the minkassociated ∆FV spike. We can infer from these structures that in the virion associated spike these changes could impact spike stability, possibly leading to premature S1 shedding.
To visualize the impact of the amino acid variations on the spike conformation, we determined cryo-EM structures of the B.1.1.7 spike (Fig. 3, B to D, figs. S13 and S14, and table S3). Multiple populations of the 3-RBD-down and RBD-up states were identified, with higher proportion of RBD-up particles observed for the B.1.1.7 (~1.8:1 RBD-up/RBD-down) compared to the D614G (~0.8:1) (16) and mink-associated ∆FV (~1.3:1) spikes (Fig. 2, A to C). Three populations of 3-RBD-down spike were refined to 3.2-3.6 Å (Fig. 3B, figs. S13 and S14, and table S3), each showing visible asymmetry with weaker density for one of its RBDs (Fig. 3B), suggestive of enhanced mobility. We identified several RBD-up structures, including a typical 1-RBD-up state (Fig. 3C), two 1-RBD-up populations with the "up" RBD and its adjacent NTD disordered (Fig. 3D). We identified states with 2-or 3-RBD up (fig. S13G), that were not detected in the D614G spike (16). Due to their limited particle numbers and preferred orientations of the particles, we were unable to obtain highresolution reconstructions of these populations. Unlike the mink-associated ∆FV spike structures, DDM analysis of the B.1.1.7 structures did not show variability in S2 (fig. S15). The apparent increase in RBD mobility in the B.1.1.7 spike 3-RBD-down structures suggested a reduced barrier for "up" state transition due to weakening of "down" state contacts. RBDs in their "down" state contact an adjacent NTD and another RBD via interprotomer protein-protein and proteinglycan contacts (Fig. 3B, inset) (51,52). Transition from the "down" to "up" state replaces these contacts with differing RBD-to-NTD and RBD-to-RBD contacts (Fig. 3C, inset).
We next sought to understand how variations that are distal from the RBD/NTD region influence the B.1.1.7 spike conformational distribution. These variations spanned multiple domains including SD1 (A570D), SD2 (P681H), HR1 (S982A), CD (D1118H) and the linker between SD2 and fusion peptide (T716I) (Fig. 1A). The P681H substitution located near the furin cleavage site could not be visualized due to disorder in the cryo-EM map in that region. The D1118H substitution, on the other hand, was well-resolved and formed a symmetric histidine triad near the base of the spike ( Fig. 3E and fig. S14, B and C). Although the histidines were positioned too far from each for direct hydrogen bonding, water mediated interactions are feasible at this separation. Moreover, the cryo-EM reconstructions showed evidence for alternate conformations that could place the histidines into closer proximity (fig. S14B). By contrast, the T716I substitution abrogated an intra-protomer hydrogen bond (H-bond) between the Thr716 side chain and Gln1071 main chain carbonyl (Fig. 3F), suggesting a local destabilizing effect.
The A570D and S982A substitutions (Fig. 4, A to E), in the SD1 and HR1 regions, respectively, appeared to be counter posing. The A570D substitution resulted in an interprotomer H-bond with the N856 side chain, reinforcing the stacking of the SD1 loop against the HR1 helix of the adjacent protomer (Fig. 4, A and B). The HR1 S982A substitu- tion, on the other hand, resulted in the loss of an interprotomer H-bond between the S982 and T547 side chains (Fig. 4, C and D). Comparing the "down" (PDB: 7KDK) and "up" (PDB: 7KDL) protomers in the D614G spike (16) showed concerted ~5-6 Å shifts in the A570 and T547 loop positions, with the T547 loop in the "up" protomer shifted farther away from, and no longer within H-bonding distance of S982 (Fig. 4D). Thus, the S982A mutation appears to disable a latch that modulates the RBD up/down equilibrium, thereby increasing RBD "up" propensity ( Fig. 4E). We had previously engineered a construct, named u1S2q, where modulation of a latch involving the A570 loop was implicated in shifting its RBD up/down equilibrium (19).
To gain insight into how the S982-T547 interprotomer latch impacts the spike quaternary structure, we defined a new set of inter-protomer vectors (Fig. 4F). Within each S protomer we defined a "Unit" comprised of the SD1/RBD region, and the NTD/ NTD′ region of the adjacent protomer that it interacts with. These Units are in conformational communication ("Com") through RBD-to-RBD contacts at the apex, as well as through the SD2 subdomain. We examined the relative disposition of the three Units, and of SD2, by using a vector network spanning the trimer. For each structure, the protomer that contained the disordered RBD (termed ProtomerB) showed a marked increase in the intra-protomer angle formed by the NTD′, SD2, and an SD2 anchor (SD2a) point (θ 5′ ) compared to this angle in the other two protomers (Protomer A and Protomer C ). This occurred in conjunction with a shift in the angle between the SD2, SD2a, and SD1 (θ 6′ ; Fig. 4G). These angular changes were accompanied by a rotation of SD1 and SD2a about an axis connecting the NTDʹ and SD2 (φ 8′ ) as well as a compensatory rotation of the SD1 to adjacent NTDʹ (φ 1′ ). This compensatory shift occurs due to differences in the A570D loop positions. With the SD2 orientation relative to S2 largely similar to that of the other protomers, these movements can be ascribed to the S982A and A570D induced movements of SD1. Together, these changes resulted in disengagement of the NTD from the adjacent RBD, explaining the increase in RBD disorder. Thus, the S982A and A570D pairing acts as an allosteric switch through coupled domain movements.
In summary, structural analysis of the B.1.1.7 spike highlights how allosteric effects of variations in distal regions alter RBD disposition. B.1.1.7 balances amino acid substitutions that destabilize the 3-RBD-down or "closed" state to favor RBD-up or "open" states, with those that stabilize the pre-fusion spike conformation. Thus, while the T716I substitution disrupts an intra-protomer H-bond, the D1118H histidine triad appears to play a stabilizing role. Similarly, while the S982A substitution abrogates a H-bond, facilitating RBD "up" movement, the A570D substitution adds a H-bond with N856, stabilizing interactions between HR1 and SD1. The accumulation of stabilizing contacts in the B.1.1.7 spike even as it acquires mutations that enable increased presentation of receptor-accessible RBD-up states, may contribute to stabilizing the pre-fusion spike to prevent premature S1 shedding.

Structural analysis of variants bearing the K417N, E484K and N501Y RBD mutations
Multiple variants that originated independently in different geographical locations show three amino acid substitutions (K417N, E484K and N501Y) in the RBD, suggesting convergent evolution and selective advantage of these substitutions. Of these, the E484K mutation is of particular concern due to its location within nAb epitopes, and it has been shown to reduce or eliminate binding to many potent RBDdirected nAbs (2). The E484K and K417N-E484K-N501Y ("triple mutant RBD") substitutions abolished binding of the potent Class 2 RBD nAbs DH1041 and DH1043 to an RBD construct ( Fig. 5A and fig. S16) (33). We found, however, that high affinity binding of DH1041 and DH1043 to S-GSAS-D614G-E484K (or "E484K spike") and S-GSAS-D614G-K417N-E484K-N501Y (or "triple mutant spike") was retained, albeit at reduced levels (Fig. 5, B and C, and figs. S2, S3, and S9).
To understand why some binding to DH1041 and DH1043 was retained for the E484K variant in the context of a S ectodomain, whereas binding was completely abrogated in the RBD-only construct, we studied the effect of the mutations on RBD conformation using molecular dynamics (MD) simulations to compare the native RBD and the triple mutant RBD (model included residues 327-529 in each). We built Markov state models of transitions between conformational states from large ensembles of short MD simulations for both constructs (figs. S17 to S20 and table S4; ~260 µs total simulation time each). The Markov models were characterized by a hook-like folded RBD tip (the "Hook" state), which resembled the conformation observed in x-ray crystal structures (33,53), and a highly dynamic "Disordered" state in which the RBD tip cycles between a variety of conformations (Fig. 5, D and E, and figs. S18 and S19, C and E). While the native RBD displayed a nearly even proportion of "Hook" vs. "Disordered" states (Fig. 5D), the triple mutant RBD showed a dramatic increase in the "Disordered" state (Fig. 5E). These population differences result from an increased transition rate to the "Disordered" state from the "Hook" state combined with a slower transition rate back to the "Hook" state in the triple mutant RBD compared to the native RBD (figs. S18F and S19F). Monitoring the interactions between residue 484 side chain in each model indicated that the native E484 hydrogen bonding with the F490 backbone in particular acted to stabilize the "Hook" state ( fig. S20). In the "Disordered" state, the K484 side chain forms fewer interactions across the RBD compared to E484 ( fig. S20B). Together, these results are consistent with the loss in binding of Abs DH1041 and DH1043 to the RBD E484K variant and indicate that the E484K substitution destabilizes the native conformation of the RBD tip, hindering binding of Class 2 RBD-directed SARS-CoV-2 neutralizing Abs.
To visualize the impact of RBD tip conformational variability on the spike, we determined cryo-EM structures of the triple mutant spike (Fig. 6A, figs. S21 and S22, and table S3). We identified 3-RBD-down, 1-RBD-up and 2-RBD-up states, as well as intermediate states that showed one RBD in the "up" position and another RBD partially up. 3-RBDdown states accounted for ~12% of the total spike population, and showed considerable disorder in their RBDs, with the disorder being most pronounced for one of the three RBDs and its contacting NTD (Fig. 6A).
We next studied spikes that, in addition to the RBD K417N-E484K-N501Y substitutions, also contained multiple residue changes in the NTD, and a A701V substitution, found in B.1.351 (Fig. 1, Fig. 6, B and C, and figs. S2, S3, S23, and S24). Despite no additional RBD mutations, binding to RBD-directed nAbs was further reduced (Figs. 1D and 6C), showing that amino acid changes outside the RBD have an allosteric effect on the binding of RBD-directed Abs. A cryo-EM dataset of a B.1.351 spike (Fig. 1) revealed ~6:1 ratio of RBD-up to 3-RBD-down structures (Fig. 6B). A "consensus" 3-RBD-down state with 212,753 particles was refined to 3.7 Å, and displayed remarkably weak RBD density in one of the 3 RBDs that also appeared detached from its interprotomer contacting NTD (Fig. 6A, PDB: 7LYM). Taken together, these data implicate the K417N-E484K-N501Y substitutions in the RBD disorder observed in the 3-RBDdown states, and suggest that the E484K-induced conformational disorder in the RBD tip "hook" structure may be the source of the increased RBD-up spike populations due to weakened RBD-to-RBD coupling. In the spike, interprotomer interactions made by the RBD in its up state, and secondary contacts that the bound antibody makes with adjacent RBDs may play a role in stabilizing antibody binding to the E484K mutant (54), explaining the retention of high-affinity binding, albeit at lower levels.
We next asked whether the weakened RBD-RBD and RBD-NTD coupling involving the disordered RBD had an impact on spike quaternary structure. Domain interface mutations are limited to the RBD in the triple mutant and B.1.351 spike variants (Fig. 1A). Asymmetry in the S1 subunit was observed when aligning the SD2 subdomain of each protomer (Fig. 6D). Patterns in the inter-protomer vector network indicated the triple mutant and B.1.351 spikes were similar in their protomer-to-protomer relationships (Fig.  6E). The absolute positions, however, displayed marked dif-ferences (Fig. 6, D and E, and fig. S25), suggesting that the additional mutations in the B.1.351 spike play a role in further modulating spike conformation. Comparing the interprotomer vector networks of these structures with the 3-RBD-down D614G spike structures indicated the B.1.351 structure was most similar to the D614G 7KE8 structure while the triple mutant spike lacked similarity to any of the D614G structures ( fig. S25). This shift toward a more D614Glike state in B.1.351 may indicate the selection of stabilizing mutations to balance the RBD destabilizing mutations. Together, these results show that amino acid variations in the RBD alone can have significant impacts on S1 quaternary structure, and accumulation of additional variations outside the RBD may in turn modulate RBD conformational changes.

Comparing SARS-CoV-2 variant S ectodomain quaternary structure
The structural results presented here indicate that the primary consequence of conformational adjustments in the SARS-CoV-2 variants is increased propensity for RBD exposure. Our data implicate destabilization of the 3-RBD-down state and involvement of a disordered RBD in this conformational difference. In order to compare the different approaches that the variants take toward this destabilization, we examined the inter-protomer network of each variant spike (Fig. 4F), together with a new RBD-to-RBD and RBDto-NTD network (Fig. 7A). It is necessary to define a primary protomer for these comparisons due to the asymmetric nature of the spike. We selected the protomer containing the RBD most distant from its adjacent NTD, often the disordered RBD protomer, for this analysis (this protomer is here designated ProtomerA″; a double prime [″] designation was used for all vector measures and domain/protomer names to signify this change). We also included in our analysis an asymmetric 3-RBD-down reconstruction of our engineered u1S2q S ectodomain (19), and four of our previously published 3-RBD-down D614G spike reconstructions (16). We first examined PCA clustering to identify structurally similar sets (Fig. 7, A and B). The triple mutant and B.1.351 spike structures, as well as ∆FV 3D-1 and 3D-2 clustered with D614G spike structures. The B.1.1.7 and ∆FV 3D-3 structures clustered with u1S2q while ∆FV 3D-4 differed markedly from all others. The separation of the structures into D614Glike and u1S2q-like is consistent with differing RBD destabilization strategies in the variants that harbor the RBD triple mutants relative to the B.1.1.7 and ∆FV spikes. Examination of the primary vectors reporting on the differences observed in these clusters indicated that the typically disordered RBD protomer, ProtomerA″, is the driver of differences between the two clusters. Positioning of SD1 relative to SD2, defined by the angle θ 4″ , in Protomer A″ , and the distances between S2 to SD2 and S2 to NTD′ were each indicators of these differences (Fig. 7, B and C). The interconnected spike domain network suggests that changes in local quaternary arrangements are likely to induce rearrangements in distant domains (Fig. 7D). We therefore examined correlations in quaternary arrangements of SD2, SD2a, SD1, and NTD′ in the full dataset. The variant discriminating SD2 to SD1 angle θ4″ (defined in Fig.  4F) displays a considerable number of correlations with quaternary arrangements throughout the network (Fig. 7, E and F, and figs. S26 to S30). This includes the Protomer B″ and Protomer C″ SD2 to SD1 angles θ 2″ and θ 6″ and the interprotomer dihedral rotation of SD2 positions about axes connecting SD1 and NTD′ between Protomer B″ to Protomer C″ and Protomer C″ to Protomer A″ ; φ 1″ and φ 4″ , respectively (Fig. 7, E and F). These, and correlation with dihedral rotation of Protomer B″ SD2 and Protomer C″ NTD′ about an axis connecting the SD2 anchor and SD1, φ 9″ , are mirrored by the Protomer B″ SD2 to SD1 angle, θ 6″ . The relationships identified show that changes in domain arrangement in one protomer have predictable impacts on the domain arrangements of the other protomers. In the D614G cluster, quaternary arrangements give rise to the marked distance between the disordered RBD and the NTD′ (Fig. 7D). For the triple mutant and B.1.351 spike structures, the RBD tip disorder presumably reduces the stability of its contact with the adjacent RBD, increasing its up-state propensity. Unlike the D614G cluster, in the u1S2q cluster RBDs are all distant from their adjacent NTD (Fig. 7D). Examination of the structures indicated rearrangements occurred in the orientation of SD1 relative to SD2 and S2. The engineered u1S2q contains mutations only in S2 and in the SD1 A570 loop that is adjacent to S2. These together increase the up-state population. It is, therefore, likely that amino acid substitution in SD2 and S2/SD1 in the ∆FV and B.1.1.7 spikes, respectively, are responsible for the increased RBD-up populations in these spikes. Thus, several mechanisms exist by which changes induced in domain interaction strength by spike amino acid substitutions modify RBD positioning.

Discussion
The SARS-CoV-2 spike plays an essential role in virus spread and represents the primary target for neutralizing antibodies. Spike mutations in SARS-CoV-2 variants can impact virus neutralization sensitivity and transmissibility. While many of the currently circulating variants of interest/concern likely arose from some combination of genetic drift, host adaptation, and immune evasion, the virus will increasingly experience pressure from vaccine elicited antibody responses. To prepare for the continued evolution of the virus, it is essential to understand how spike variations impact virus transmissibility and neutralization sensitivity. The increased binding to ACE2, mediated both by affinity enhancing substitutions in the RBD and increased propensity for the receptor-accessible RBD-up states, may contribute to the rapid spread of variants. For the mink-associated variant increased receptor binding may have helped establish infection in a new host. While all human-evolved variants studied here showed reduced binding to antibodies at dominant neutralization epitopes, the mink-associated variant retained similar levels of binding to all antibodies tested, underscoring the role of the human immune response in shaping the course of SARS-CoV-2 evolution. For the minkevolved variant we uncovered evidence for spike instability, that may be the reason why the variant failed to spread widely when transmitted back to humans. For the humanevolved variants, we found that the S protein used different mechanisms for manipulation of its immunodominant regions to converge on a common goal of destabilizing the 3-RBD-down state. While in the B.1.1.7 variant this occurred by modifications in SD1 or SD2 to S2 interaction, for variants harboring the K417N/E484K/N501Y RBD triple substitutions, RBD destabilization was mediated by RBD-RBD contacts. Together, these results show that these variants have modified the S1 subunit domain interaction network to control the functionally critical disposition of the RBD while acquiring antibody resistance and improved transmissibility. We have provided a structurally detailed view of these variants and a framework from which to anticipate further changes to the spike as the pathogen evolves.

Protein purification
On the 6 th day post transfection, spike ectodomains were harvested from the concentrated supernatant. The spike ectodomains were purified using StrepTactin resin (IBA LifeSciences) and size exclusion chromatography (SEC) using a Superose 6 10/300 GL Increase column (Cytiva, MA) equilibrated in 2mM Tris, pH 8.0, 200 mM NaCl, 0.02% NaN3. All steps of the purification were performed at room temperature and in a single day. Protein quality was assessed by SDS-Page using NuPage 4-12% (Invitrogen, CA). The purified proteins were flash frozen and stored at -80°C in single-use aliquots. Each aliquot was thawed by a 20-min incubation at 37°C before use. Antibodies were purified by Protein A affinity and digested to their Fab state using LysC. ACE2 with human Fc tag was purified by Protein A affinity chromatography and SEC (19). RBD constructs were produced and purified as described in Saunders et al. (56).

SPR
Antibody binding to SARS-CoV-2 spike and RBD constructs was assessed using SPR on a Biacore T-200 (Cytiva, MA, formerly GE Healthcare) with HBS buffer supplemented with 3 mM EDTA and 0.05% surfactant P-20 (HBS-EP+, Cytiva, MA). All binding assays were performed at 25°C. Spike variants were captured on a Series S Strepavidin (SA) chip (Cytiva, MA) by flowing over 200 nM of the spike for 60 s at 10 µL/min flowrate. The Fabs were injected at concentrations ranging from 0.625 nM to 800 nM (2-fold serial dilution) using the single cycle kinetics mode with 5 concentration per cycle. For the single injection assay, the Fabs were injected at a concentration of 200nM. A contact time of 60s, dissociation time of 120 s (3600s for DH1047 for the single cycle kinetics) at a flow rate of 50µL/min was used.
The surface was regenerated after each dissociation phase with 3 pulses of a 50mM NaH + 1M NaCl solution for 10 s at 100 µL/min. For the RBDs, the antibodies were captured on a CM5 chip (Cytiva, MA) coated with Human Anti-Fc (using Cytiva Human Antibody Capture Kit and protocol), by flowing over 100nM antibody solution at a flowrate of 5µL/min for 120s. The RBDs were then injected at 100nM for 120 s at a flowrate of 50µL/min with a dissociation time of 30 s. The surface was regenerated by 3 consecutive pulse of 3M MgCl2 for 10s at 100µL/min. Sensorgram data were analyzed using the BiaEvaluation software (Cytiva, MA).

Negative-stain electron microscopy
Samples were diluted to 100 µg/ml in 20 mM HEPES pH 7.4, 150 mM NaCl, 5% glycerol, 7.5 mM glutaraldehyde (Electron Microscopy Sciences, PA) and incubated for 5 min before quenching the glutaraldehyde by the addition of 1 M Tris (to a final concentration of 75 mM) and 5 min incubation. A 5-µl drop of sample was applied to a glow-discharged carbon-coated grid (Electron Microscopy Sciences, PA, CF300-Cu) for 10-15 s, blotted, stained with 2% uranyl formate (Electron Microscopy Sciences, PA), blotted and airdried. Images were obtained using a Philips EM420 electron microscope at 120 kV, 82,000× magnification, and a 4.02 Å pixel size. The RELION (57) software was used for particle picking, and 2D and 3D class averaging.

ELISA assays
Spike ectodomains tested for antibody-or ACE2-binding in ELISA assays as previously described (32). Assays were run in two formats i.e., antibodies/ACE2 coated, or spike coated. For the first format, the assay was performed on 384-well plates coated at 2 µg/ml overnight at 4°C, washed, blocked and followed by two-fold serially diluted spike protein starting at 25 µg/mL. Binding was detected with polyclonal anti-SARS-CoV-2 spike rabbit serum (developed in our lab), followed by goat anti-rabbit-HRP (Abcam, Ab97080) and TMB substrate (Sera Care Life Sciences, MA). Absorbance was read at 450 nm. In the second format, serially diluted spike protein was bound in wells of a 384-well plates, which were previously coated with streptavidin (Thermo Fisher Scientific, MA) at 2 µg/mL and blocked. Proteins were incubated at room temperature for 1 hour, washed, then human mAbs were added at 10 µg/ml. Antibodies were incubated at room temperature for 1 hour, washed and binding detected with goat anti-human-HRP (Jackson ImmunoResearch Laboratories, PA) and TMB substrate. for data processing. Phenix (54,59), Coot (60), Pymol (61), Chimera (62), ChimeraX (63) and Isolde (64) were used for model building and refinement.

Vector based structure analysis
Vector analysis of intra-protomer domain positions was performed as described previously (19)  , and a S2 sheet motif (S2s; residues 717 to 727 and 1047 to 1071). Additional centroids for the NTD (NTDc; residues 116 to 129 and 169 to 172) and RBD (RBD c ; residues 403 to 410) were determined for use as reference points for monitoring the relative NTD and RBD orientations to the NTD′ and SD1, respectively. Vectors were calculated between the following within protomer centroids: NTD to NTD′, NTD′ to SD2, SD2 to SD1, SD2 to CD, SD1 to RBD, CD to S2s, NTDc to NTD, RBD to RBD c . Vector magnitudes, angles, and dihedrals were determined from these vectors and centroids.
Inter-protomer domain vector calculations for the SD2, SD1, and NTD′ used these centroids in addition to anchor residue Cα positions for each domain including SD2 residue 671 (SD2a), SD1 residue 575 (SD1a), and NTD′ residue 276 (NTD′a). These were selected based upon visualization of position variation in all protomers used in this analysis via alignment of all of each domain in PyMol (61). Vectors were calculated for the following: NTD′ to NTD′r, NTD′ to SD2, SD2 to SD2 r , SD2 to SD1, SD1 to SD1 r , and SD1 to NTD′. Angles and dihedrals were determined from these vectors and centroids. Vectors for the RBD to adjacent RBD and RBD to adjacent NTD were calculated using the above RBD, NTD, and RBD c centroids. Vectors were calculated for the following: RBD 2 to RBD 1 , RBD 3 to RBD 2 , and RBD 3 to RBD 1 . Angles and dihedrals were determined from these vectors and centroids. Principal components analysis, K-means clustering, and Pearson correlation (confidence interval 0.95, p<0.05) analysis of vectors sets was performed in R (67). Data were centered and scaled for the PCA analyses. Principal components analysis, K-means clustering, and Pearson correlation (confidence interval 0.95, p < 0.05) analysis of vectors sets was performed in R. Data were centered and scaled for the PCA analyses.

Difference distance matrices (DDM)
DDM were generated using the Bio3D package (68) (73) and Glycam (74) forefields were used throughout. All simulations were performed using the Amber20 pmemd CUDA implementation. The systems were first minimized for 10,000 steps with protein atom restraints followed by minimization of the full system without restraints for an additional 10,000 steps. This was followed by heating of the systems from 0 K to 298 K over a period of 20 ps in the NVT ensemble using a 2 fs timestep using the particle mesh Ewald method for long-range electrostatics and periodic boundary conditions (75). The systems were then equilibrated for 100 ps in the NPT ensemble with the temperature controlled using Langevin dynamics with a frequency of 1.0 ps -1 and 1 atm pressure maintained using isotropic position scaling with a relaxation time of 2 ps (76). A non-bonded cut-off of 8 Å was used throughout and hydrogen atoms were constrained using the SHAKE algorithm (77) with hydrogen mass repartitioning (78) used to allow for a 4 fs timestep. In order to generate an ensemble of RBD tip conformations for initiation of the adaptive sampling routine, we performed one hundred 50 ns simulations in the NVT ensemble with randomized initial velocities for each of the WT and Mut systems. The final frame from each of these simulations was used to initiate the adaptive sampling scheme. Adaptive sampling was performed using the High-Throughput Molecular Dynamics (HTMD v. 1.24.2) package (79). Each iteration consisted of 50-100 independent simulations of 100 ns. Simulations from each iteration were first projected using a dihedral metric with angles split into their sin and cos components for residues 454 to 491. This was followed by a TICA (80) projection using a lag time of 5 ns and retaining five dimensions. Markov state models were   4.0) license, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/. This license does not apply to figures/photos/artwork or other content included in the article that is credited to a third party; obtain authorization from the rights holder before using such material.  Pink outlines identify relationships plotted in (F). Square outline identifies non-significant correlation in the full structure set that was significant in the D614G cluster only correlations. (F) Selected vector relationship plots. Dot color indicates Kmeans cluster assignment from the PCA analysis in (B).