Neutralization of IgM-14 against naturally emerged SARS-CoV-2 variants

Human IgM-14 was previously engineered from its parental IgG-14 by replacing the Fv region of IgG1 with an IgM scaffold in the presence of J-chain co-expression [13]. Unlike IgG-14, which contains two identical Fabs, IgM-14 features ten Fabs (Fig 1A), enhancing its valency and avidity. Accordingly, IgM-14 outperformed IgG-14 in the neutralization of multiple SARS-CoV-2 variants, including B.1.1.7 (Alpha), B.1.351 (Beta), P.1 (Gamma), and even the spike E484A mutant, which is resistant to IgG-14 [12,13].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Antibody neutralization activity and breadth against SARS-CoV-2.

A, Illustration of the architectures of Fab, IgG-14, and IgM-14. B, Construction of spike variants. L, 5’ lead sequence; ORF, open reading frame; E, Envelope; M, Membrane; N, nucleocapsid; UTR: untranslated region; mNG, mNeonGreen. C, Summary of neutralization activities of IgM-14 and IgG-14 against naturally emerged variants. *, data from our previous study. ^#, Fold changes in EC₅₀ relative to the parental USA-WA1/2020 strain. D, Scatter plot comparing EC₅₀ fold changes of IgM-14 and IgG-14 across SARS-CoV-2 variants, normalized to the parental strain. Each point represents a variant. Blue, variants carrying the E484K/Q mutation; gray, variants without the E484K/Q mutation; orange, Omicron strains. E, Scheme of testing IgM-14’s antiviral activity in human airway epithelial (HAE). F-K, antiviral efficacy of IgM-14 against SARS-CoV-2 in HAE cultures. Fluorescent microscopy of HAE infected with mNG BA.1-spike (F) or Delta-spike (I) variants after IgM-14 treatment. Intracellular viral RNA on day 3 after mNG BA.1-spike (G) or Delta-spike (J) infection. Extracellular viral RNA on day 3 after mNG BA.1-spike (H) or Delta-spike (K) infection. Geometric means with 95% confidence intervals (CI) are shown in panels G-H and J-K. One-way ANOVA with Dunn’s multiple comparison corrections was used for statistical analysis. ***, p < 0.001; ****, p < 0.0001.

https://doi.org/10.1371/journal.ppat.1014071.g001

In this study, we further evaluated IgM-14’s neutralization activity against a range of additional SARS-CoV-2 variants, including pre-Omicron variants-D614G, B.1525 (Eta), B.1.526 (Iota), B.1.617.1 (Kappa), B.1.617.2 (Delta), C.37 (Lambda), B.1.618, B.1.621 (Mu), as well as highly mutated Omicron sublineages BA.1, BA,2, BA.3, and a recently emerged JN.1. Pre-Omicron variants, which harbor single or a few spike mutations, were engineered into the ancestral USA-WA1/2020 SARS-CoV-2 strain (WT). Omicron variants, containing over 30 spike amino acid changes, were cloned into a live-attenuated USA-WA1/2020 strain expressing an mNG green fluorescence reporter [15] (Fig 1B and S1 Table). Neutralization activity was assessed using the plaque reduction neutralization test (PRNT) or PRNT-equivalent fluorescence focus reduction neutralization tests (FFRNT) [16].

IgM-14 potently neutralized all pre-Omicron variants, with half-maximal effective concentration (EC₅₀) values ranging from 0.003 to 0.0321 µg/ml, whereas IgG-14 showed significantly lower potency, with EC₅₀ in the range of 0.26-156 µg/ml (Figs 1C and S1). The E484K/Q mutation in the spike of pre-Omicron variants, a well-known antibody resistance marker [13], reduced IgG-14’s neutralization potency by 29.4- to 445.7-fold (Fig 1C and 1D). In contrast, IgM-14 remained less affected by this mutation. It effectively neutralized Eta, Kappa, Mu, and B.1.618-variants carrying E484K/Q, with similar efficiency as against the ancestral strain and variants lacking this mutation, such as Delta, D614G, Alpha, and Lambda. The most significant reduction in IgM-14’s neutralization (10.7-fold) was observed against Beta and Iota variants, which contained E484K and other RBD or NTD mutations (Fig 1C and 1D and S1 Table). Omicron variants become more resistant to both IgG-14 and IgM-14. IgG-14 was almost inactive against all tested Omicron sublineages (EC₅₀ > 100 µg/ml). IgM-14 showed modest neutralization against BA.1 (EC₅₀ of 0.54 µg/ml) and weak activity against BA.2 and BA.3 (EC₅₀s of 22.8-23.9 µg/ml) but was inactive against the Omicron sublineage JN.1 at the highest tested antibody concentration (Fig 1C and 1D).

Consistently, IgM-14 retained modest activity against BA.1 infection in physiologically relevant primary human airway epithelial (HAE) cell cultures. At a concentration of 1.0 µg/ml, IgM-14 markedly reduced the number of mNG-positive cells (Fig 1E and 1F). The estimated EC₅₀ for BA.1 inhibition in HAE ranged from 0.29 to 0.30 µg/ml, based on measurements of extracellular or intracellular viral RNA, respectively (Fig 1G and 1H). As a control, IgM-14 demonstrated substantially greater potency against the Delta variant, effectively reducing mNG-positive cells even at 0.01 µg/ml (Fig 1I). The EC₅₀ for Delta inhibition in HAE was 0.0011 µg/ml for both extracellular and intracellular viral RNA quantifications (Fig 1J and 1K). This represented a 263- to 273-fold decrease in neutralization efficacy against BA.1 compared to Delta.

Collectively, these findings further validate the broader neutralization spectrum of IgM-14 compared to IgG-14 against SARS-CoV-2 variants, and demonstrate the significant immune evasion of Omicron variants from IgM-14-mediated neutralization.

IgM-14 resistance mutations map to the receptor-binding motif within the spike

To map residues critical for IgM-14 neutralization, we conducted serial passaging of live-attenuated mNG USA-WA1/2020 strain in the presence of increasing concentrations of IgM-14 (Fig 2A), following a protocol previously used to select IgG-14-resistant variants [12]. After six passages, three out of four independent selections (SL1–3) yielded variants strongly resistant to IgM-14, with EC₅₀ values exceeding 52 µg/ml, representing a > 17,000-fold increase in potency compared to the unpassaged parental strain.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Characterization of IgM-14-resistant variants.

A, Diagram of the process for selecting IgM-14-resistance in VeroE6 cells. Four independent selections (SL1-4) started at passage 1 (P1) and were passaged 6 times. SL4 was aborted at the second passage. B, Spike mutations occurred in P6. EC₅₀s of IgM-14 against P6 viruses (SL1-3) are shown. *, intact and deletion sequences observed in the selection. C, Construction of mNG SARS-CoV-2 with spike mutations. D, Neutralization curves of IgM-14 and IgG-14 against WT or mutant mNG SARS-CoV-2. E, Summary of the EC₅₀ values derived from panel D. F, Plaque morphologies of WT or mutant mNG SARS-CoV-2 on VeroE6 cells. Growth kinetics of G476D (G) or F486S (H) versus WT mNG SARS-CoV-2 in VeroE6, A549-hACE2, and human airway epithelial (HAE) cultures. Two-way ANOVA with Šidák’s multiple comparison corrections was used for statistical analysis. **, p < 0.01; ***, p < 0.001; ****, p < 0.0001. I, Relative expression of RBD mutants to the WT RBD. Error bars indicate variation across three independent experiments. J, BLI analysis of the binding of WT and mutant RBDs to human ACE2. Association rate (k_on), dissociation rate (k_off), affinity constant (K_D), and R² values of curve fitting are indicated. K, ELISA analysis of IgM-14 and IgG-14 binding to WT (solid circles), G476D (red circles), and F486S (blue circles) RBDs. Error bars indicate variation across three independent experiments.

https://doi.org/10.1371/journal.ppat.1014071.g002

Notably, the development of SARS-CoV-2 resistance to IgM-14 required a longer selection period (22 days versus 10 days for IgG-14) and more passages (6 versus 3 for IgG-14). Furthermore, one of the IgM-14 selections failed to yield resistant mutants, whereas all four IgG-14 selection experiments successfully recovered resistant variants (Fig 2A and our previous data [12]). These data support the notion that IgM-14 imposes a higher barrier to resistance compared to IgG-14.

Sequencing the spike gene from Passage 6 (P6) viruses revealed consensus amino acid changes, including G476D (in SL1 and SL3), F486S (in SL2), and two additional mutations near the furin cleavage site, S686G (in SL1), and Δ675–679 (deletion and the intact sequences co-existing in SL2 and SL3) (Figs 2B, S2A and S2B). G476D and F486S are located within RBM, a region frequently targeted by potent monoclonal antibodies [17,18]. The S686G and Δ675–679 were known Vero cell-adaptive changes that reduce S1 and S2 cleavage [19–21], and are unlikely to contribute to resistance. To validate the role of G476D and F486S in resistance to IgM-14, we introduced each mutation individually into the infectious clone of mNG SARS-CoV-2. Mutant viruses were recovered from VeroE6 cells following electroporation with in vitro-transcribed full-length RNA (Fig 2C). FFRNT analysis showed that IgM-14 failed to neutralize either G476D or F486S at concentrations up to 100 µg/ml (Fig 2D and 2E). Moreover, both mutant viruses exhibited complete resistance to IgG-14, confirming the critical roles of G476 and F486 in mediating neutralization by IgM-14 and IgG-14.

Spike G476D and F486S mutations reduce SARS-CoV-2 fitness

Although both G476D and F486S mutant viruses replicated efficiently on VeroE6 cells, reaching titers exceeding 10⁶ PFU/ml, they formed smaller plaques compared to WT mNG-SARS-CoV-2 (Fig 2F), suggesting attenuated viral replication. To further assess the impact of both mutations on viral fitness, we compared their multicycle growth with WT mNG SARS-CoV-2 in immortalized and primary cell culture models. Both mutants exhibited varying degrees of replication attenuation across VeroE6, A549-hACE2, and HAE cultures (Fig 2G and 2H), demonstrating that these two mutations reduce viral fitness.

To understand how G476D and F486S mediate antibody resistance and impact viral fitness, we expressed, purified, and characterized the recombinant RBDs carrying each mutation. Interestingly, while F486S had only a marginal effect, the G476D mutation drastically reduced RBD expression by approximately 71-fold (Fig 2I). This substantial reduction is possible due to the replacement of a neutral glycine with a negatively charged aspartic acid, which may disrupt RBD folding and compromise protein stability. The binding affinities of G476D and F486S RBDs to human ACE2, as determined by biolayer interferometry (BLI) assay, were 23.1 nM and 8.53 nM, respectively, which were 96- and 36-fold lower than that of the WT RBD (Fig 2J). In enzyme-linked immunosorbent assay (ELISA), neither mutant RBD bound to IgM-14 or IgG-14 (Fig 2K), confirming their role in antibody resistance. These findings demonstrate that IgM-14 targets functionally critical regions of the spike protein involved in receptor binding, and that resistance mutations in this region come at a significant cost to viral fitness.

Structural heterogeneity of the spike/Fab-14 complex

We employed structural approaches to gain deeper insights into the molecular mechanism of IgM-14-mediated neutralization and resistance from naturally occurring and laboratory-acquired variants. Despite multiple attempts, no condition permitting cryo-EM imaging of isolated spike/IgM-14 complexes was found, likely due to the high flexibility and exceptionally strong avidity of IgM-14, which promoted extensive cross-linking and aggregation. Supporting this hypothesis, size‑exclusion chromatography (SEC) of spike: IgM‑14 (3:1) showed a prominent near‑void peak (~8.3 ml; indicative of species exceeding the column’s separation upper limit ~5,000,000 Da) and a broad 8.3-12.6 ml plateau, accompanied by a reduced IgM-14 peak (~12.6 ml), near disappearance of the spike peak (~13.8 ml), and minor late species >15 ml (S3A Fig). These features indicate polydisperse, very high-molecular‑weight assemblies arising from multivalent cross‑linking. As a control, spike: IgG‑14 (2:3) resolved a single, discrete complex at ~12.5 ml and no near‑void material (S3B Fig), consistent with a well‑behaved, non‑aggregated complex. In concert with the SEC, negative-stain electron microscopy revealed spike aggregation in the presence of IgM-14, distinct from the “starfish-like” morphology of IgM or the “mushroom-like” shape of the prefusion trimeric spike alone (S3C-S3E Fig). Notably, elongated, needle-like postfusion spike particles were also observed, plausibly arising from IgM-14-driven S1 dissociation. The appearance of later, smaller SEC species appeared compatible with such S1 release and subunit rearrangement. Together, these results indicated that IgM‑14 promotes spike cross‑linking/aggregation and conformational heterogeneity.

To overcome these technical challenges, we utilized Fab-14, which retains the key antigen-binding region of IgM-14 but lacks the Fc domain. This approach enabled us to successfully determine high-resolution cryo-EM structures of Fab-14 bound to the trimeric spike of the USA-WA1/2020 strain carrying the D614G mutation (hereafter referred to as the D614G spike).

We first established our cryo-EM imaging and reconstruction pipeline by determining the structure of the apo (Fab-free) D614G spike [22,23]. This initial dataset yielded relatively low-resolution cryo-EM maps that were nevertheless sufficient to reveal the overall spike architecture and confirm the correctness of our image reconstruction procedure (S4A-S4C Fig). The maps showed that RBDs undergo a hinge-like motion between up (receptor-accessible) and down (receptor-inaccessible) states, consistent with previous studies [24,25]. All well-defined spike trimers fell into two major conformational classes with roughly equal populations: one with a single RBD in the up state (referred to as “1-up-RBD”), and the other with all three RBDs in the down state (referred to as “3-down-RBD”) (S4C Fig).

Next, we obtained high-resolution structures of Fab-14-bound D614G spike trimers after processing 17,415 micrograph movies followed by 2D classification, 3D reconstruction, and refinement (S5A-S5C Fig). The experiments were conducted at a Fab-14 to spike molar ratio of 5:1. Cryo-EM analysis showed that most spike trimers are bound by one or two Fab-14 molecules, hereafter referred to as “1-Fab” and “2-Fab” complexes. Only 7.2% of spike trimers remain unbound, all adopting the 3-down-RBD conformation (S5C Fig). According to the number of bound Fab-14s and conformational state of the spike, we identified five distinct modes of Fab-14/spike complexes (Figs 3A-3E and S5C).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Five modes of Fab-14 binding to D614G spike trimer.

A, Cryo-EM density of the Fab-14 and spike trimer complex in Mode I. The primary and secondary binding sites are indicated by orange and red circles, respectively. The down RBD involved in the primary binding site is shown in yellow, while the down RBD involved in the secondary binding site is shown in blue. The heavy and light chains of Fab-14 are colored pale cyan and light gray, respectively. B, Cryo-EM density of the Fab-14 and spike trimer complex in Mode II. The two down RBDs are shown in yellow and blue, respectively. The up RBD is shown in green. C, Cryo-EM density of the Fab-14 and spike trimer complex in Mode III. D, Cryo-EM density of the Fab-14 and spike trimer complex in Mode IV. The up RBDs are shown in yellow and green, respectively. The down RBD is shown in blue. E, Cryo-EM density of the Fab-14 and spike trimer complex in Mode V. F, local refined structures of primary and secondary binding sites in the bipartite binding mode. The paperclip motif in the primary binding site is shown in orange. The loop 331-335 in the secondary binding site is shown in dark blue.

https://doi.org/10.1371/journal.ppat.1014071.g003

Mode I (2-down-RBD/1-Fab) accounted for 21.4% of the total particles. In this mode, Fab-14 bridges two adjacent down RBDs in a bipartite manner with the third RBD being poorly resolved due to its local mobility (Figs 3A and S6A). Fab-14 binding appears to disrupt the threefold symmetry of three down RBDs by inducing a modest increase in the up angle of one of the two Fab-bound RBDs, and that of the Fab-free RBD. This reorganization is accompanied by positional shifts, with the centers of mass of the RBDs displaced by 1.2 to 3.0 Å (S6B-S6E Fig), relative to the published apo trimeric spike structures [26].

Mode II (1-up-RBD/1-Fab), which constitutes 26.3% of the total particles, features a single Fab-14 binding to an up RBD (Fig 3B). This mode is structurally heterogeneous and can be further divided into three subgroups with slightly different up angles (110°, 114°, and 117°) of the up RBD (S7A-S7B Fig).

Mode III (1-up-RBD/2-Fab) contains 11.4% of the total particles. In this mode, one Fab-14 molecule binds to an up RBD. At the same time, a second Fab-14 engages the other two down RBDs in a bipartite manner (Fig 3C). The orientation of the Fab-bound up RBD is similar to that found in Mode II, whereas the second Fab-14 binds the two down RBDs in a bipartite manner, similar to Mode I. Notably, to accommodate both Fab molecules, the up angles and position of the two down RBDs in this mode (S8A and S8B Fig) show slight differences compared to those in Mode I (S6B Fig).

Mode IV (2-up-RBD/2-Fab) was observed in 24.9% of the total particles. In this mode, two up RBDs are each bound by a Fab-14 (Fig 3D). Similar to Mode II, this group is heterogeneous. It can be divided into two subgroups (I and II), which differ by the orientation of their Fab-bound RBDs (S9A-S9D Fig) and by the amount of disorder in the down RBD (S5C Fig). In both subgroups, the two Fabs interact with each other via their light chains (S9C Fig).

Mode V (2-up-RBDs/1-Fab) was found in 8.7% of the total particles. In this mode, only one of the two up RBDs is bound by Fab-14 (Fig 3E). The Fab-bound up RBD density is significantly stronger than the free up RBD (S9D Fig). 3D flexible refinement confirmed conformational heterogeneity (S9E-S9F Fig), suggesting that Fab-14 binding may stabilize the RBD in the up conformation.

Two distinct binding sites between Fab-14 and RBD

In all conformations of the spike/Fab-14 complex described above, the Fab-14/RBD unit displayed only two poses. In Modes II, III, IV, and V, Fab-14 binds to a single up RBD, while in Modes I and III, it bridges two adjacent down RBDs. To elucidate these interactions in atomic detail, we next performed local refinement of the Fab-14/RBD region using particle subsets that yielded the highest-quality EM density maps for Mode I and Mode IV subgroup I conformations.

The Mode I cryo-EM map refinement converged to a resolution of 3.4 Å (S5D Fig). In this conformation, Fab-14 engages two down RBDs via two distinct binding sites (Figs 3F, S5F and S5G). The primary site involves the paperclip motif (residues 471–491) within RBM, a region commonly targeted by Class 2 neutralizing antibodies [27,28]. Fab-14 interacts with this site via two heavy-chain complementarity-determining regions (HCDR-2 and HCDR-3) and two light-chain CDRs (LCDR-1 and LCDR-3). The secondary site comprises RBD residues 331–335, located near the hinge region and outside of the RBM, and is engaged with LCDR-2 of Fab-14.

Local refinement of the Mode IV subgroup I conformation revealed that two Fab-14 molecules engage two up RBDs through the paperclip motif as observed in Mode I (S5E, S5H and S10A Figs). Notably, the two Fabs also interact with each other via a light chain loop (residues 13–17) connecting strands A and B (S10B Fig). Structural alignment of the RBD/Fab-14 module at the primary binding site from Mode IV and Mode I showed a high similarity, with an RMSD of 0.21 Å across all Cα atoms (S10C Fig). These observations suggest that the structure of the RBD/Fab-14 module and at the primary binding interface are conserved across conformational modes. Consequently, since the secondary binding site involves a rigid-body rotation of the RBD/Fab-14 module (Fig 3F), it is also likely to be conserved.

The primary site plays a major role in antibody neutralization

The Fab-14 heavy chain dominates the primary site interaction, with HCDR2 and HCDR3 residues forming a tight network of hydrophobic and polar contacts around the RBD paperclip motif (Fig 4A). Specifically, residues S57 and Y60 of the heavy chain coordinated E484. In contrast, Y54, L106, Y110, and Y112 of the heavy chain, along with Y93 of the light chain, form a hydrophobic clamp that captures F486 of the RBD (Fig 4B). F486 is sandwiched within this clamp by π-stacking interactions with R52 and Q104 of the heavy chain, flanking sides of its benzene ring. Notably, R52 is stabilized by ionic bridges with D62 and E102 of the heavy chain, while Q104 forms hydrogen bonds with R56 and Y110 of the heavy chain, as well as with RBD residue Y489. Y110 contributes to additional polar interactions with RBD N487. The light chain provides auxiliary stabilization through LCDR1 and LCDR3, with residues Y32, Y34, and Y93 engaging the loop spanning RBD residues 476–479 (Fig 4A and 4C). Collectively, Fab-14 buries a surface patch formed by RBD residues 456, 475–479, and 484–489, encompassing key determinants of ACE2 recognition. This primary binding site overlaps with the ACE2 interfaces (Fig 4D), suggesting that Fab-14 neutralizes the virus by directly blocking ACE2 engagement.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Structural analysis of the primary binding site.

A, overview of the primary binding site. B, Zoomed-in view of the interaction between HCDR2 and the RBD paperclip motif. Pink dashed lines indicate salt bridges, purple dashed lines show hydrogen bonds. C, Zoomed-in view of the interaction between Fab-14 light chain and RBD residues G476-P479. D, Superposition of Fab-14 and ACE-2 on the same up RBD. The surface of ACE2 and Fab-14 are shown in pink and blue, respectively. E, MM/PBSA analysis of changes in RBD/Fab-14 complex binding free energy changes caused by individual mutations. Data shows the mean ± standard deviations. F, Sequence alignment of the primary binding site between the parental SARS-CoV-2 (USA-WA1/2020) and Omicron variants. The GISAIDs of BA.1, BA.2, BA.3, and JN.1 spikes were EPI_ISL_6640916, EPI_ISL_6795834.2, EPI_ISL_7605591, and EPI_ISL _18237538, respectively.

https://doi.org/10.1371/journal.ppat.1014071.g004

This structural model provides a framework for understanding the resistance mechanisms associated with several critical RBD mutations, G476D, E484K/A/S, and F486S/P, identified in natural variants. These mutations likely perturb this interaction network at the primary binding site, weakening the neutralization activity of IgM-14 and IgG-14. G476D can reduce backbone flexibility, introduce steric hindrance, and alter surface charge (Fig 4C). Substitution of E484 with alanine (A), serine (S), or lysine (K) can disrupt the glutamic acid-mediated polar interaction network, destabilizing HCDR2 binding (Fig 4B). F486S introduces a smaller, hydrophilic side chain, while F486P may alter the backbone conformation of the entire paperclip motif, both significantly impairing hydrophobic interactions and the “sandwich” stacking between Fab-14 and RBD (Fig 4B).

These structural suppositions were, in general, supported by Molecular Mechanics Poisson-Boltzmann Surface Area (MM/PBSA) calculations. Mutations involving interface-buried residues, F486S, F486P, and G476D, showed reduced binding free energy of RBD/Fab-14 complexes, whereas mutations at the peripheral, solvent-exposed E484 resulted in only modest changes (Fig 4E). Omicron variants carrying multiple mutations at the primary binding site, such as BA.1/BA.2/BA.3 (S477N, T478K, and E484A/K), and JN.1 (S477N, T478K, V483_del, E484K, and F486P) (Fig 4F), which likely further weaken Fab-14 binding.

The secondary site contributes to IgM-14 neutralization

The local refinement structural analysis revealed that the second binding site exhibits fewer and weaker interactions, primarily concentrated in the 331–335 loop region of the RBD and Fab-14 LCDR2 (Fig 5A). Notably, the glycan attached to RBD residue N331 forms a stacking-like interaction with the side chain of LCDR2 residue Y62. Additionally, the backbones of RBD residues N331 and I332 interact with the side chain of Y62, while the Oγ atom of the T331 side chain forms hydrogen bonds with the backbone nitrogen atoms of LCDR2 residues G59 and I60. However, the relatively long distances (>3.9 Å) suggest these interactions are weak. Furthermore, the side chain of RBD residue L335 is 3.1 Å away from LCDR2 residue S58, indicating a van der Waals interaction. The glycan at RBD residue N343 interacts with the side chain of Fab-14 heavy chain residue E1 and the backbone of G26. However, the cryo-EM density in these regions indicated a high degree of conformational flexibility, suggesting that these interactions are not particularly stable.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Analysis of the secondary binding site.

A, Structural analysis of the secondary binding site. B, MM/PBSA analysis of changes in RBD/Fab-14 complex binding free energy changes caused by indicated mutations. C-D, Neutralization curves of IgG-14 (C) or IgM-14 (D) against mNG SARS-CoV-2 D614G (reference) and mNG SARS-CoV-2 D614G+N331Q + L335A (N331Q+L335A). E-F, Neutralization curves of IgG-14 (E) or IgM-14 (F) against mNG SARS-CoV-2 D614G+E484K (E484K) and mNG SARS-CoV-2 D614G+E484K + N331Q+L335A (E484K+N331Q + L335A). Means and standard deviation from three independent experiments are shown in panels C-F. G, Summary of EC₅₀ values of IgG-14 or IgM-14 against indicated SARS-CoV-2 mutants. H, A model of IgM-14 complexed with SARS-CoV-2 spike trimer shows different modes of spike-IgM-14 interactions. The coordinates of IgM were taken from a published report⁴¹ by Chen, Q. et al. and adjusted by rigid body movement of the Fab arms while preserving the integrity of the Fab tethers.

https://doi.org/10.1371/journal.ppat.1014071.g005

To assess the biological significance of these secondary-site interactions, we designed three RBD mutants: N331Q, which removed the glycan-mediated interaction; L335A, which disrupted the hydrophobic contact; and the double mutant N331Q/L335A, which abolished both interactions. MM/PBSA analysis suggested that none of the mutations changed the binding free energy for the RBD/Fab-14 complexes (Fig 5B). To experimentally characterize the impact of the secondary site on antibody neutralization, we introduced the N331Q/L335A mutation into the infectious clone of a reference strain mNG SARS-CoV-2 D614G. FFRNT assays showed that the double mutant resulted in a modest 2.2- to 3.25-fold reduction in neutralization potency by IgM-14 and IgG-14 (Fig 5C, 5D and 5G), consistent with weak Fab-14 engagement at the secondary site. Consistent with these results, BLI analysis showed that Fab‑14 binding to RBD fit a 1:1 interaction model across low, medium, and high RBD surface densities (S11 Fig). Increasing the RBD loading resulted in only minor changes in association rates and a slightly slower dissociation, with monophasic off‑rates in all cases. These patterns showed that the contribution of the secondary contact to the overall energetics of Fab-14 binding is negligibly small compared to that of the primary binding site.

To further investigate the role of the secondary binding site, we introduced N331Q/L335A double mutation into the infectious clone of mNG SARS-CoV-2 D614G carrying the E484A mutation, which severely impaired IgG-14 binding at the primary site [12]. As expected, E484A reduced IgG-14’s neutralization potency by 64.1-fold (EC₅₀ of 50.59 µg/ml for E484A versus 0.789 µg/ml for the reference strain), while IgM-14 was only modestly affected (7.25-fold increase in EC₅₀ values; 0.029 µg/ml versus 0.004 µg/ml). Notably, adding N331Q/L335A further reduced IgM-14 potency by 65.8-fold, compared to just a 3.08-fold reduction for IgG-14 (Fig 5E-5G). These data suggest that secondary-site interactions play a more critical role in IgM-14 neutralization than in IgG-14.

In addition to the structural analysis of Fab-14/D614G spike, we determined a cryo-EM structure of the BA.1 spike in the complex with Fab-14 (S12A-S12C Fig). Consistent with the modest neutralization potency of IgM-14, only a small fraction (11.1%) of BA.1 spike particles were bound by a single Fab-14 with all RBDs in the down conformation (S12C Fig). Although the map resolution was insufficient for atomic modeling, the overall architecture resembled Mode I of the Fab-14/G614 spike complex, where Fab-14 engages both primary and secondary binding sites (S13 Fig). This suggests that the conserved secondary binding site may partially compensate for the reduced affinity at the mutated primary site in BA.1. Consistently, these structural analyses highlight the critical role of the secondary site in IgM-14’s neutralization breadth.

Structural model of IgM-14 binding to SARS-CoV-2 spikes via both primary and secondary sites

To understand the structural basis for the role of the secondary binding site in IgM potency, we constructed multiple models of spike/IgM-14 interactions guided by the high-resolution structures of RBD/Fab-14 complexes. Because of its pentameric architecture and inherent structural flexibility, a single IgM-14 molecule can crosslink multiple spike trimers via diverse binding modes. As illustrated in Fig 5H, IgM-14 could engage a spike trimer via the binding Mode IV using a Fab pair from a single monomeric unit. More importantly, IgM-14 can bind a single spike via Modes I and II (one down-RBD and one up-RBD) using Fabs from adjacent monomeric units, with Mode I involving the secondary binding site. The remaining Fab arms remain available to bind additional spike trimers with all potential binding modes, enabling multivalent binding. In contrast, IgG-14, with its two Fab arms tethered to a single Fc hinge, is limited in angular flexibility and spatial reach, restricting it to binding one or two up RBDs or a single down RBD on a spike trimer. Overall, our structural model supports the enhanced avidity and broader spatial accessibility of IgM-14 compared to IgG-14, highlighting the functional importance of the secondary binding site in IgM-mediated neutralization.

Cells and Virus

VeroE6 (ATCC CRL-1586) cells were obtained from the American Type Culture Collection (ATCC, Bethesda, MD). VeroE6 cells expressing TMPRSS2 (JCRB1819) were purchased from SEKISUI XenoTech, LLC. Both cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% fetal bovine serum (FBS; HyClone Laboratories, South Logan, UT) and 1% penicillin/streptomycin (P/S). All cultures were maintained at 37°C with 5% CO₂. A549-ACE2 cells (human alveolar epithelial cells expressing human Angiotensin-converting enzyme 2 [ACE2]) [46] were maintained in high-glucose DMEM supplemented with 10% FBS, 1% P/S, 1% 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES), with 10 µg/ml blasticidin S. The 293F Freestyle cells (ThermoFisher Scientific) were maintained in FreeStyle 293 Expression Medium at 37°C with 8% CO₂. Human airway epithelia (HAE, EpiAirway, AIR-100) were obtained from MatTek Corporation and maintained according to the manufacturer’s instructions. All antibiotics, medium, and supplements were purchased from ThermoFisher Scientific. All cells were tested Mycoplasma negative.

Infectious clone-derived SARS-CoV-2 and mNG SARS-CoV-2 were used in this study [47]. All SARS-CoV-2 handling was conducted at the Biosafety Level 3 facility with approval from the Institutional Biosafety Committee at the University of Texas Medical Branch.

Selection of antibody-resistant viruses

VeroE6 cells were incubated with SARS-CoV-2 mNG (MOI 2) in the presence of IgM-14 at a starting concentration of 0.5 μg/ml. Supernatants were collected when more than 30% of cells showed CPE. 200 µl of supernatants were used for the next infection on fresh VeroE6 cells with increased concentrations of IgM-14. The concentrations of IgM-14 in all 6 passages are the following: Passage 1–2 (P1-2): 0.5 μg/ml; P3-4: 1.2 μg/ml; P5: 4.8 μg/ml; P6: 18 μg/ml. The P6 viruses were used for testing their sensitivity to IgM-14 and for sequencing the spike region.

PRNT

The neutralization of the SARS-CoV-2 US-WA1 strain and the recombinant SARS-CoV-2 variants were performed using the plaque reduction neutralization test (PRNT) as described previously [15]. In brief, antibodies were serially diluted in culture medium and incubated with 100 PFU of viruses at 37°C for 1 h, after which the antibody-virus mixtures were inoculated onto VeroE6 cell monolayers in six-well plates. After 1 h of infection at 37°C, 2 ml of 2% SeaPlaque agar (Lonza) in DMEM containing 2% FBS and 1% P/S was added to the cells. After 2 days of incubation, 2 ml of 2% SeaPlaque agar in DMEM containing 2% FBS, 1% P/S, and 0.01% neutral red (Sigma-Aldrich) was added on top of the first layer. After another 16 h of incubation at 37°C, plaque numbers were counted. The relative infection was obtained by the plaque counts from the antibody-treated groups compared to the untreated groups. The relative infection versus the concentration of the antibody (in log₁₀ scale) was plotted. PRNT₅₀ titers were calculated using a nonlinear regression model in GraphPad software Prism 10.

FFRNT

Fluorescent focus reduction neutralization test (FFRNT) was conducted using mNG SARS-CoV-2 by following a protocol described previously [16]. Briefly, VeroE6 cells were seeded in 96-well microplates (Greiner Bio-One). The cells were incubated overnight. The next day, 3-fold serially diluted IgM-14 and IgG-14 were mixed with each mNG SARS-CoV-2 at 37°C for 1 h. Afterward, the antibody-virus mixtures were loaded onto the pre-seeded VeroE6 cell monolayer. At 1 h post-infection, the inoculum was replaced with 100 μl of overlay medium (supplemented with 0.8% methylcellulose). The plates were incubated at 37°C for 16 h. Raw images of mNG foci were scanned and counted using Cytation 7 (BioTek) imager. The foci in each well were normalized to the non-antibody-treated controls to calculate the relative infectivities. The FFRNTs were performed in duplicate. A four-parameter nonlinear regression model in GraphPad Prism 10 determined the EC₅₀ values.

Generation of SARS-CoV-2 mNG mutants

Recombinant mNG SARS-CoV-2 spike-variant was constructed by engineering the complete spike gene from the indicated variants into an infectious cDNA clone of mNG USA-WA1/2020 as reported previously [47]. Briefly, standard overlap PCRs using specific primers (listed in the S2 Table) were conducted to introduce the spike mutations into the infectious clone of the mNG USA-WA1/2020. The full-length infectious cDNA clones were assembled by in vitro ligation. Subsequently, the genome-length RNAs were synthesized by in vitro transcription. The full-length RNA and N gene RNA transcripts were electroporated into VeroE6-TMPRSS2 cells to rescue the recombinant viruses. After 48-72h of transfection, the supernatants (referred to as P0) were harvested. Then the P0 stocks were inoculated into freshly prepared VeroE6 cells for further amplification. At 48 h post-infection, supernatants (P1) were harvested and clarified with centrifugation at 1000 × g for 10 mins and stored at -80°C.

RNA extraction, RT-PCR, and cDNA sequencing

Cell culture supernatants were mixed with a five-fold excess of TRIzol LS Reagent (ThermoFisher Scientific). Viral RNAs were extracted according to the manufacturer’s instructions. The extracted RNAs were dissolved in 50 μl nuclease-free water. For sequence validation of mutant viruses, 2 µl of RNA samples were used for reverse transcription by using the SuperScript IV one-step RT-PCR kit (ThermoFisher Scientific) with Cov-21115V and Cov-25169R. The resulting DNAs were purified by the QIAquick Gel Purification Kit and sequenced by Sanger sequencing at GENEWIZ (South Plainfield, NJ).

Plaque assay

8 × 10⁵ VeroE6-TMPRSS2 cells per well were seeded into 6-well plates. The next day, 200 μl of 10-fold serially diluted virus was added to pre-seeded cells and incubated at 37°C for 1 h. After that, the inoculum was replaced with 2 ml of overlay medium containing DMEM with 2% FBS, 1% P/S, and 1% Seaplaque agarose (Lonza, Walkersville, MD, USA). After 2 days of incubation at 37°C, 2 ml of overlay medium supplemented with neutral red (Sigma-Aldrich) was added to stain the cells, and plaques were counted on the next day.

Growth kinetics of recombinant SARS-CoV-2 in cell culture

For viral replication kinetics on VeroE6 and A549-hACE cells, 5 × 10⁵ cells per well were seeded in 6-well plates. The next day, cells were infected with corresponding viruses (MOI 0.01 for VeroE6; MOI 0.1 for A549-hACE cells) at 37°C for 1 h. After infection, the inocula were removed, and the cells were washed with PBS three times. Cells were then incubated in DMEM supplemented with 2% FBS at 37°C, 5% CO₂. At each time point, 300 µl of supernatants were collected and 300 µl of fresh culture media were added back to the wells. All samples were stored at -80°C prior to use.

Antiviral testing in HAE culture

For HAE (approximately 5 × 10⁵ cells per well), an equal amount of SARS-CoV-2 mNG and variants (MOI 0.2) was added at the apical side and incubated at 37°C for 2 h. After infection, the inocula were removed, and the culture was washed three times with DPBS at the apical side and followed by incubation at 37°C, 5% CO₂. At each time point, 300 μl of DPBS was added onto the apical side of the airway culture and incubated at 37°C for 30 min to elute progeny viruses. All samples were collected from the apical side and stored at -80°C prior to use.

Antibodies at various concentrations were mixed with mNG SARS-CoV-2s and immediately used to infect HAE cultures (MOI 0.2). At day 3 post-infection, apical DPBS washes were collected as described above. Following DPBS wash collection, cells on the transwell inserts were lysed in 300 µl Trizol reagent (ThermoFisher Scientific). Viral RNAs in both cell lysates and DPBS washes were quantified by standard RT-qPCR. To determine the EC₅₀ values (the concentration of antibody required to inhibit 50% viral infection), relative RNA levels were calculated by normalizing viral RNA in each sample to those of no-antibody-treated control groups. A four-parameter nonlinear regression model in GraphPad Prism 10 was used to determine EC₅₀.

RT-qPCR

To quantify viral yields, 1 volume of supernatant samples was mixed with 5 volumes TRIzol LS regent (ThermoFisher Scientific). SARS-CoV-2 RNA was extracted and eluted in 50 µl nuclease-free water according to the manufacturer’s instructions. The RNA copies of each sample were determined by a standard curve-based quantitative RT-PCR (RT-qPCR) assay using iTaq SYBR Green One-Step Kit (BioRad) on the QuantStudio 7 Flex (ThermoFisher Scientific). An in vitro transcribed 3,839-bp RNA from the SARS-CoV-2 genome was used as a standard, as reported previously [48].

Protein preparation

The mammalian expression plasmid pαH with cDNA encoding SARS-CoV-2 D614G spike HexaPro (S) was kindly provided by Dr. Jason S. McLellan [49]. A codon-optimized cDNA encoding the BA.1 spike sequence was synthesized by Genscript and cloned into the same pαH vector. Both constructs encode spike protein with HexaPro stabilizing mutations and a C-terminal 8 × Histidine tag followed by 2 × Strep-Tag II for purification purposes. D614G and BA.1 spike protein were expressed and purified according by following a protocol reported previously [49]. In brief, FreeStyle 293-F cells were transfected with the plasmids encoding D614G spike or BA.1 spike using polyethyleneimine (Polysciences). After 3 h post-infection, 5 μM kifunensine (Sigma) was added to boost protein expression. On day 4 post-transfection, the supernatants were collected by centrifugation at 7000 × rpm at 4°C for 10 min, followed by passing through a 0.22 μm filter. The spike proteins were purified using Ni-NTA agarose columns (Cytiva) followed by size-exclusion chromatography using a Superose 6 10/300 column in an ÄKTA pure instrument (Cytiva). Spike proteins were finally eluted in a buffer composed of 2 mM Tris, pH 8.0, 200 mM NaCl.

The human ACE2 protein (10108-H08H) was purchased from Sino Biological. The Fc-tagged wild-type RBD protein and the RBD protein containing amino acid mutation F486S or G476D were prepared as described previously [45]. The IgG-14 was prepared as previously described [12]. Briefly, plasmids encoding the IgG-14 were transfected into Expi293F cells. 7 days post-transfection, the supernatants were collected and recombinant IgG-14 was purified using CaptivA HF Protein A Affinity Resin (REPLIGEN). Purified IgG-14 was then digested with papain to generate Fab-14 fragments. The IgM-14 was prepared as previously described [13]. Briefly, the V_H and V_L region of IgG-14 was incorporated into expression vectors encoding the IgM constant region and a human J-chain. IgM-14 was expressed in Expi293F cells and purified by mixed-mode chromatography and anion-exchange chromatography.

ELISA

An Enzyme-Linked Immunosorbent Assay (ELISA) was performed to evaluate the antibody binding to the RBD proteins. The 96-well high-binding ELISA plates were coated with 100 μl per well of 0.2 μg/ml recombinant Fc-tagged RBD proteins overnight at 4°C. The plates were blocked with PBS supplemented with 5% skim milk. After blocking, 100 μl serial dilutions of monoclonal antibodies (diluted in 1% skim milk) were added to the wells and incubated at room temperature for 2 h. The plates were incubated with HRP-conjugated mouse anti-human lambda (Southern Biotech, 9280-05, 1:6,000 diluted in 1% skim milk) for 1 h. Between incubation steps, the plates were washed five times with PBST (0.05% Tween-20). After the final washes, TMB substrate was added at 100 μl per well for color development. The reaction was stopped by adding 50 μl per well 2M H₂SO₄. The OD₄₅₀ nm was read by a SpectraMax microplate reader and analyzed with GraphPad Prism 10.

BLI

Bio-Layer Interferometry (BLI) was used to analyze RBD and ACE2 binding. The Fc-tagged RBD proteins were used for measuring affinity with human ACE2 on the Forte bio Octet RED 96 system (Sartorius, Goettingen, Germany). Briefly, RBD proteins (20 μg/ml) were captured onto protein A biosensors for 300 seconds. The loaded biosensors were then dipped into the kinetics buffer for 10 s for adjustment of baselines. Subsequently, the biosensors were dipped into serially diluted (0.37 ~ 300 nM) human ACE2 protein for 200 seconds to record association kinetics and then dipped into kinetics buffer for 400 s to record dissociation kinetics. A kinetic buffer without ACE2 was used to correct the background. The Octet Data Acquisition 9.0 software was used to collect affinity data. For the fitting of K_D values, Octet Data Analysis software V11.1 was used to fit the curve by a 1:1 binding model and the use of the global fitting method. For RBD and Fab-14 binding, RBD was biotinylated with NHS-PEG4-Biotin (ThermoFisher) and captured on SA sensors (Sartorius) at 5, 20, or 100 µg/ml, titrated with Fab-14 (100-3.7 nM). Binding and dissociation were monitored in PBS, and data were globally fit using a 1:1 binding model with Octet Data Analysis software v13.

Size-exclusion analysis

Purified D614G spike protein was incubated with IgM-14 or IgG-14 at a molar ratio of 3:1 and 2:3, respectively, at a final spike concentration of 0.2 mg/ml in a total volume of 200 µl. The mixture was incubated for 30 min at 4°C prior to analysis. Samples were then subjected to a Superose 6 Increase 10/300 column equilibrated in 2 mM Tris, pH 8.0, 200 mM NaCl.

Cryo-EM sample preparation and imaging

For the D614G spike alone, 4 μL of proteins (0.3 mg/ml) were applied to the QUANTIFOIL R2/2 grid (Quantifoil) that had been plasma-cleaned by Solarus plasma cleaner (Gatan). The grid was plunge-frozen in liquid ethane using the EM GP2 Automatic plunge freezer (Leica) at 22°C and 95% humidity. For spike/Fab-14 complexes, Fab-14 was mixed with the D614G spike or BA.1 spike protein at a molecular ratio of 5:1 to a final concentration of 0.3 mg/ml spike and incubated at 4°C for 30 minutes. Then, 4 μL of samples were applied to the grid as described above.

Grids were loaded into a Titan Krios G3i microscope (ThermoFisher Scientific) equipped with a K3 or Falcon 4 direct electron detector with a GIF Quantum energy filter (20-eV energy slit) (Gatan) and operating at 300 keV. For the D614G spike alone, cryo-EM data were automatically acquired with a Falcon 4 camera using EPU software at a nominal magnification of 105,100 × (corresponding to 0.86 Å per pixel) with a nominal defocus range between -1.0 and -2.5 μm. Forty-frame movie stacks were collected over an exposure time of 1.0 s with a total dose of 40 e⁻ Å⁻¹. For the D614G spike/Fab-14 complex, cryo-EM data were automatically acquired with a K3 camera using SerialEM at a nominal magnification of 105,100 × (corresponds to 0.84 Å per pixel) with a nominal defocus range between -1.0 and -2.5 μm. Forty-four-frame movie stacks were collected over an exposure time of 1.0 s with a total dose of 43.17 e⁻ Å⁻¹. Similar movies were collected from frozen grids of the BA.1 spike/Fab-14 complex using the same microscope under the same conditions, except the stacks were collected over an exposure time of 1.0 s with a total dose of 43.27 e⁻ Å⁻¹. The parameters of data collection were summarized in Supplementary S3-S6 Tables.

For the D614G spike alone, the collected movie fractions were imported into CryoSPARC for image processing [50]. Movie data were motion corrected using Patch Motion Correction. The contrast transfer function (CTF) was estimated using the Patch CTF Estimation method. Micrographs with CTF fit worse than 5 Å were excluded, leaving 10,220 micrographs. A total of 4,045,200 particles were selected using Blob Picker and extracted with 2 × binning (192 pixels). After four rounds of 2D classification, the Rebalance 2D classification was performed, and 317,242 particles of the total 2,095,808 particles were used to create five initial 3D volumes using Ab Initio with C3 symmetry. Three 3D volumes and all the remaining particles were used for five iterative rounds of heterogeneous refinement, resulting in two different conformations. Conformation 1 has one up-RBD with 419,193 particles. Conformation 2 has three down-RBD with 371,184 particles. The 419,193 particles in conformation 1 were re-extracted without binning (384 pixels) and further refined by homogeneous refinement, followed by global and local CTF refinement and a final round of non-uniform refinement with C1 symmetry, yielding a 3.3 Å reconstruction according to GSFSC at 0.143. The 371,184 particles in conformation 2 were re-extracted without binning (384 pixels) and further refined by homogeneous refinement with C3 symmetry, followed by global CTF refinement and a final round of non-uniform refinement with C3 symmetry, yielding a 3.3 Å reconstruction. The detailed information on data processing is shown in S4 Fig and S3 Table.

Cryo-EM data processing

For the D614G spike/Fab-14 complex, collected movie fractions were imported into CryoSPARC and preprocessed similarly as described above. Micrographs with CTF fit worse than 4 Å were excluded, leaving 17,415 micrographs for further processing. A total of 7,261,652 particles were selected using Blob Picker and extracted with 2 × binning (220 pixels). After three rounds of two-dimensional (2D) classification, 695,411 remaining particles were used to create five initial three-dimensional (3D) volumes using Ab Initio. Three 3D volumes were used as input for three iterative rounds of heterogeneous refinement, resulting in various modes. Mode I contained 21.4% of the total particles. The 113,755 particles in Mode I were re-extracted without binning (440 pixels) and first refined using homogeneous refinement, followed by CryoSPARC’s implementation of global CTF refinement and a final round of non-uniform refinement, yielding a 3.3 Å reconstruction. The particles were then refined with Flexible refinement. Local refinement was further performed in cryoSPARC to improve the electron density in the Fab-RBD interaction region. A mask that covers the Fab-14 and the two down RBD was created using UCSF Chimera. The mask was imported into cryoSPARC with dilation and soft padding. Particles were re-centered to the center of the mask, and local refinement was performed, yielding a 3.4 Å reconstruction. Mode II contained 26.3% of the total particles. Heterogeneous refinement was conducted to further classify Mode II into three subgroups with different up angles. The particles were re-extracted and refined as described in Mode I, yielding 3.4 Å reconstruction for each subgroup. Mode III contained 11.4% of the total particles. The particles were re-extracted and refined as described in Mode I, yielding 3.3 Å resolution. Mode IV was observed in 24.9% of the total particles, which were divided into two subgroups using heterogeneous refinement. Local refinement was further performed in cryoSPARC to improve the electron density in the Fab-14/RBD contacts in Mode IV subgroup I. Two masks were created using UCSF Chimera: one covering the two 2-up-RBD and Fab-14, and one covering the remaining region. Masks were imported into cryoSPARC with dilation and soft padding. Particles were subtracted and re-centered, and local refinement was performed, yielding a 4.1 Å reconstruction. Mode V contained 8.7% of the total particles, which were re-extracted and refined as described in Mode I, yielding a 3.4 Å reconstruction. The remaining particles (7.2% of the total particles) are Fab-free, which were re-extracted and refined identically as Mode I, yielding a 3.4 Å reconstruction. Detailed information on data processing is shown in S5 Fig and S3 and S4 Tables.

For Omicron BA.1 spike/ Fab-14 complex, collected movie fractions were imported into CryoSPARC and preprocessed similarly to the D614G spike/Fab-14 complex. Micrographs with CTF fit worse than 4 Å were excluded, leaving 9,282 micrographs. A total of 1,123,585 particles were selected using Blob Picker and extracted with 2 × binning (192 pixels). After three rounds of 2D classification, 626,037 remaining particles were used to create five initial 3D volumes using Ab Initio. Three 3D volumes were used as input for three iterative rounds of heterogeneous refinement, resulting in three different conformations (1–3). The 336,659 particles in conformation 1 were re-extracted without binning (384 pixels) and further refined by homogeneous refinement, followed by CryoSPARC’s implementation of global CTF refinement and a final round of non-uniform refinement, yielding a 3.1 Å reconstruction according to GSFSC at 0.143. The 105,694 particles in conformation 2 were re-extracted and refined as conformation 1, yielding a 3.2 Å reconstruction according to GSFSC at 0.143. The 55,801 particles in conformation 3 were re-extracted and refined identically as conformation 1, yielding a 3.6 Å reconstruction according to GSFSC at 0.143. The detailed information on data processing is shown in Supplementary S11 Fig and S6 Table. Local resolutions of final reconstructions were estimated by CryoSPARC’s implementation of blocres and were colored by a resolution range from 3 Å to 7 Å in UCSF ChimeraX [50,51].

Model building, refinement, and analysis

Cryo-EM structure of the D614G spike (PDB: 7BNM for 3-down-RBD; PDB: 7BNN for 1-up-RBD) and the atomic model of Fab-14 (generated by AlphaFold 3 [52]) were used to build an initial model by rigid body fitting in UCSF Chimera and manual adjustments in Coot. The model was refined iteratively by real-space refinement in Phenix and by manual adjustments and improvements in Coot. Protein-protein interaction was analyzed using UCSF ChimeraX and PyMol (Schrödinger). Figures were prepared using UCSF Chimera and UCSF ChimeraX. To calculate the angles of RBD-apo or RBD-Fab-14 complexes, the axes of RBDs were generated. Then, the angles between the horizontal plane and the axis were calculated with the UCSF Chimera X angle command [53].

Molecular mechanics Poisson-Boltzmann Surface Area calculations

The MM/PBSA-derived binding free energies were calculated using FoldX package [54]. Firstly, the structure of a protein complex was refined by using the FoldX function “RepairPDB”. This procedure “repairs” improbable dihedral angles and van der Waals clashes in a protein. The refined complex with respect to the FoldX energy function was then used as a starting point for a computational mutagenesis procedure with “BuildModel”. A desired mutation was made to the repaired structure and then energy-minimized by sampling rotamers for the mutated residue and its neighboring amino acids. The wild-type protein was then energy-minimized with respect to rotamers of the same set of residues. The mutation-caused energy difference (ΔΔG) between the mutant (ΔG_MT) and wild type (ΔG_WT) was calculated between the two chosen protomers in the complex using the “AnalyseComplex” function. Apply the calculations for different pairs of protomers if the interaction involves more than two protomers. After repeating the calculations 10 times for each mutation, the means and standard deviations were calculated from structures with different local neighbor conformations.

Negative-stain electron microscopy

For negative staining (NS) electron microscopy (EM), 4 μl of IgM-14 at 0.05 mg/ml was applied to a glow-discharged CF200-Cu carbon film grid (Electron Microscopy Sciences). The sample was left on the carbon film for 60 s, followed by negative staining with 2% uranyl formate for 60 s. Micrographs were recorded in a JEOL 2100FS microscope at 60,000 × magnification operated at 200 keV. The images were imported and processed in CryoSPARC. For the D614G spike alone, 4 μl of IgM-14 at 0.1 mg/ml was applied to the grid as described above. Micrographs were recorded and processed as described above, except images were recorded at 80,000 × magnification. For D614G spike and IgM-14 complex, D614G spike and IgM-14 were incubated on ice for 30 minutes at molar ratios of 6:1 or 3:1. The final concentration of D614S spike was kept at 0.1 mg/ml. 4 μl of complex was applied to the grid as described above, and micrographs were recorded and processed as the D614G spike alone.

Statistics

Sample sizes were estimated based on similar research reported in the literature; no statistical method was used to predetermine sample size. No data was excluded from the analysis. All experiments were performed with at least two biological replicates to ensure reproducibility. Investigators were generally not blinded, as experimental conditions required knowledge of sample identities. Statistical tests can be found in the corresponding figure legends. Dots on the bar graphs represent the values from an individual replicate of the experiment. All data were analyzed using the software Prism 10 (GraphPad). Unless otherwise noted, numerical results from repeated independent experiments are presented as mean ± standard deviation (SD). For viral growth kinetics, data were log₁₀-transformed to approximate a normal distribution before statistical analysis.