Potent neutralizing antibodies from COVID-19 patients define multiple targets of vulnerability

Phenotyping SARS-CoV-2–specific B cell subsets

We collected a single blood sample from three polymerase chain reaction–confirmed SARS-CoV-2–infected individuals (COSCA1, COSCA2, and COSCA3) ~4 weeks after symptom onset. COSCA1 (a 47-year-old male) and COSCA2 (a 44-year-old female) showed symptoms of an upper respiratory tract infection and mild pneumonia, respectively (Table 1). Both remained in home isolation during the course of COVID-19 symptoms. COSCA3, a 69-year-old male, developed a severe pneumonia and became respiratory insufficient 1.5 weeks after symptom onset, requiring admission to the intensive care unit for mechanical ventilation. To identify S protein–specific antibodies in the sera obtained from all three patients, we generated soluble, prefusion-stabilized S proteins of SARS-CoV-2 using stabilization strategies previously described for S proteins of SARS-CoV-2 and other β-coronaviruses (Fig. 1A) (12, 34). As demonstrated by the size-exclusion chromatography trace, SDS–polyacrylamide gel electrophoresis (PAGE), and blue native PAGE, the resulting trimeric SARS-CoV-2 S proteins were of high purity (fig. S1, A and B). Sera from all patients showed strong binding to the S protein of SARS-CoV-2 in an enzyme-linked immunosorbent assay (ELISA), with end-point titers of 13,637, 6133, and 48,120 for COSCA1, COSCA2, and COSCA3, respectively (Fig. 1B), and showed cross-reactivity to the S protein of SARS-CoV (fig. S1C). COSCA1, COSCA2, and COSCA3 had varying neutralizing potencies against SARS-CoV-2 pseudovirus, with 50% inhibition of virus infection (ID₅₀) values of 383, 626, and 7645, respectively (Fig. 1C), and similar activities against authentic virus (fig. S1D). In addition, all sera showed cross-neutralization of SARS-CoV pseudovirus and authentic SARS-CoV virus, albeit with low potency (fig. S1, E and F). The potent S protein–binding and -neutralizing responses observed for COSCA3 are consistent with earlier findings showing that severe disease is associated with a strong humoral response (35). On the basis of these strong serum binding and neutralization titers, we sorted SARS-CoV-2 S protein–specific B cells for mAb isolation from COSCA1, COSCA2, and COSCA3.

Peripheral blood mononuclear cells were stained dually with fluorescently labeled prefusion SARS-CoV-2 S proteins and analyzed for the frequency and phenotype of specific B cells by flow cytometry (Fig. 2A and fig S2). The analysis revealed a frequency ranging from 0.68 to 1.74% of S protein–specific B cells (S-AF647⁺, S-BV421⁺) among the total pool of B cells (CD19⁺Via-CD3^–CD14^–CD16^–), (Fig. 2B). These SARS-CoV-2 S protein–specific B cells showed a predominant memory (CD20⁺CD27⁺) and plasmablasts/plasma cells (PBs/PCs) (CD20^–CD27⁺CD38⁺) phenotype. We observed a threefold higher percentage of PBs/PCs for SARS-CoV-2 S protein–specific B cells compared with total B cells (P = 0.034), indicating an enrichment of specific B cells in this subpopulation (Fig. 2C). COSCA3, who experienced severe symptoms, showed the highest frequency of PBs/PCs in both total (34%) and specific (60%) B cells (Fig. 2C and fig. S2). As expected, the SARS-CoV-2 S protein–specific B cells were enriched in the immunoglobulin G–positive (IgG⁺) and IgM^–/IgG^– (most likely representing IgA⁺) B cell populations, although a substantial portion of the specific B cells were IgM⁺, particularly for COSCA3 (Fig. 2D).

Genotypic signatures of the SARS-CoV-2–specific antibody response

SARS-CoV-2 S protein–specific B cells were subsequently single-cell sorted for sequencing and mAb isolation. In total, 409 heavy chain (HC) and light chain (LC) pairs were obtained from the sorted B cells of the three patients (137, 165, and 107 from COSCA1, COSCA2, and COSCA3, respectively), of which 323 were unique clonotypes. Clonal expansion occurred in all three patients (Fig. 3A) but was strongest in COSCA3, where it was dominated by HC variable (VH) regions VH3-7 and VH4-39 (34 and 32% of SARS-CoV-2 S protein–specific sequences, respectively). Even though substantial clonal expansion occurred in COSCA3, the median somatic hypermutation (SHM) was 1.4%, with similar SHM in COSCA1 and COSCA2 (2.1 and 1.4%, respectively) (Fig. 3B). These SHM levels are similar to those observed in response to infection with other respiratory viruses (36).

A hallmark of antibody diversity is the heavy chain complementarity-determining region 3 (CDRH3). Because the CDRH3 is composed of V, D, and J gene segments, it is the most variable region of an antibody in terms of both amino acid composition and length. The average length of CDRH3 in the naïve human repertoire is 15 amino acids (37), but for a subset of influenza virus and HIV-1 broadly neutralizing antibodies, long CDRH3 regions of 20 to 35 amino acids are crucial for high-affinity antigen–antibody interactions (38, 39). Even though the mean CDRH3 length of isolated SARS-CoV-2 S protein–specific B cells did not differ substantially from that of a naïve population (37), we observed a significant difference in the distribution of CDRH3 length (two-sample Kolmogorov–Smirnov test, P = 0.006) (Fig. 3C). This difference in CDRH3 distribution can largely be attributed to an enrichment of longer (~20 amino acid) CDRH3s, leading to a bimodal distribution as opposed to the bell-shaped distribution that was observed in the naïve repertoire (Fig. 3C and fig. S3).

Next, to determine SARS-CoV-2–specific signatures in B cell receptor repertoire usage, we compared ImmunoGenetics (IMGT) database–assigned unique germline V regions from the sorted SARS-CoV-2 S protein–specific B cells with the well-defined extensive germline repertoire in the naïve population (Fig. 3D) (37). Multiple VH genes were enriched in COSCA1, COSCA2, and COSCA3 compared with the naïve repertoire, including VH3-33 (P = 0.009) and VH1-24 (P < 0.001) (Fig. 3D). Even though the enrichment of VH1-69 was not significant (P > 0.05), it should be noted that an enrichment of VH1-69 has been shown in response to a number of other viral infections, including influenza virus, hepatitis C virus, and rotavirus (40), and an enrichment of VH3-33 was observed in response to malaria vaccination, whereas the enrichment of VH1-24 appears to be specific for COVID-19 (Fig. 3D) (41). By contrast, VH4-34 (P > 0.05) and VH3-23 (P = 0.018) were substantially underrepresented in SARS-CoV-2–specific sequences compared with the naïve population. Although usage of most VH genes was consistent between COVID-19 patients, VH3-30-3 and VH4-39 in particular showed considerable variability. Thus, upon SARS-CoV-2 infection, the S protein recruits a subset of B cells from the naïve repertoire enriched in specific VH segments and CDRH3 domains.

Identification of unusually potent SARS-CoV-2–neutralizing antibodies

Subsequently, all HC and LC pairs were transiently expressed in human embryonic kidney (HEK) 293T cells and screened for binding to SARS-CoV-2 S protein by ELISA. A total of 84 mAbs that showed high-affinity binding were selected for small-scale expression in HEK 293F cells and purified (table S1). We obtained few S protein–reactive mAbs from COSCA3, possibly because most B cells from this individual were IgM⁺, whereas cloning into an IgG backbone nullified avidity contributions to binding and neutralization present in the serum. To gain insight in the immunodominance of the RBD as well as its ability to cross-react with SARS-CoV, we assessed the binding capacity of these mAbs to the prefusion S proteins and the RBDs of SARS-CoV-2 and SARS-CoV using ELISA. Of the 84 mAbs tested, 32 (38%) bound to the SARS-CoV-2 RBD (Fig. 4, A and B), with seven mAbs (22%) showing cross-binding to SARS-CoV RBD (fig. S4A). We also observed 33 mAbs (39%) that bound strongly to SARS-CoV-2 S but did not bind the RBD, of which 10 mAbs (30%) also bound to the S protein of SARS-CoV (Fig. 4, A and B). Notably, some mAbs that bound very weakly to soluble SARS-CoV-2 S protein in ELISA showed strong binding to membrane-bound S protein, implying that their epitopes are presented poorly on the stabilized soluble S protein or that avidity is important for their binding (table S1). Surface plasmon resonance (SPR) assays confirmed binding of 77 mAbs to S protein and 21 mAbs to the RBD with binding affinities in the nanomolar to picomolar range (table S1).

All 84 mAbs were subsequently tested for their ability to block infection. A total of 19 mAbs (23%) inhibited SARS-CoV-2 pseudovirus infection with varying potencies (Fig. 4C) and, of these, 14 (74%) bound the RBD. Seven of the 19 mAbs could be categorized as potent neutralizers [median inhibitory concentration (IC₅₀) < 0.1 μg/ml], six as moderate neutralizers (IC₅₀ = 0.1 to 1 μg/mL), and six as weak neutralizers (IC₅₀ = 1 to 10 μg/ml). With an IC₅₀ of 0.008 μg/ml, the RBD-targeting antibodies COVA1-18 and COVA2-15 in particular were unusually potent. However, they were quite different in other aspects, such as their HC V gene usage (VH3-66 versus VH3-23), LC usage (VL7-46 versus VK2-30), HC sequence identity (77%), and CDRH3 length (12 versus 22 amino acids). Seventeen of the mAbs also interacted with the SARS-CoV S and RBD proteins and two of these cross-neutralized the SARS-CoV pseudovirus (IC₅₀ = 2.5 μg/ml for COVA1-16 and 0.61 μg/ml for COVA2-02; fig. S4B), with COVA2-02 being more potent against SARS-CoV than against SARS-CoV-2. Next, we assessed the ability of the 19 mAbs to block infection of authentic SARS-CoV-2 virus (Fig. 4C and fig. S4C). Although previous reports suggested a decrease in neutralization sensitivity of primary SARS-CoV-2 compared with pseudovirus (25, 27, 28), we observed very similar potencies for seven of the 19 NAbs, including the most potent NAbs (IC₅₀ = 0.007 and 0.009 μg/ml for COVA1-18 and COVA2-15, respectively; Fig. 4C). NAbs COVA1-18, COVA2-04, COVA2-07, COVA2-15, and COVA2-39 also showed strong competition with ACE2 binding, illustrating that blocking ACE2 binding is their likely mechanism of neutralization (fig. S4D). The RBD-targeting mAb COVA2-17, however, showed incomplete competition with ACE2. This corroborates previous observations that the RBD encompasses multiple distinct antigenic sites, some of which do not involve blocking of ACE2 binding (23, 25, 26). The non-RBD NAbs all bear substantially longer CDRH3s compared with RBD NAbs (fig. S4E), suggesting a convergent, CDRH3-dependent contact between antibody and epitope.

Multiple targets of vulnerability on the SARS-CoV-2 S protein

To identify and characterize the antigenic sites on the S protein and their interrelationships, we performed SPR-based cross-competition assays using S protein, followed by clustering analysis. We note that competition clusters do not necessarily equal epitope clusters but the analysis can provide clues as to the relationship between mAb epitopes. We identified 11 competition clusters, of which nine contained more than one mAb and two contained only one mAb (clusters X and XI; Fig. 5A and fig. S5). All nine multiple-mAb clusters included mAbs from at least two of the three patients, emphasizing that these clusters represent common epitopes targeted by the human humoral immune response during SARS-CoV-2 infection. Three clusters included predominantly RBD-binding mAbs (clusters I, III, and VII), with cluster I forming two subclusters. These three clusters were confirmed by performing cross-competition experiments with soluble RBD instead of complete S protein (fig. S5B). Four clusters (V, VI, XIII, and IX) included predominantly mAbs that did not interact with RBD, and clusters II, IV, X, and XI consisted exclusively of non-RBD mAbs. mAbs with diverse phenotypes (e.g., RBD and non-RBD–binding mAbs) merged together in multiple clusters, suggesting that these mAbs might target epitopes bridging the RBD and non-RBD sites or that they sterically interfere with each other’s binding as opposed to binding to overlapping epitopes. Although clusters II, V, and VIII contained only mAbs incapable of neutralizing SARS-CoV-2, clusters I, III, IV, VI, and VII included both non-NAbs and NAbs. Cluster V was formed mostly by non-RBD–targeting mAbs that also bound to SARS-CoV. However, these mAbs were not able to neutralize either SARS-CoV-2 or SARS-CoV, suggesting that these mAbs target a conserved non-neutralizing epitope on the S protein. Finally, the two non-RBD mAbs COVA1-03 and COVA1-21 formed single-mAb competition clusters (clusters X and XI, respectively) and showed an unusual competition pattern, because binding of either mAb blocked binding by most of the other mAbs, but not vice versa (figs. S5 and S6 and table S2). We hypothesize that these two mAbs allosterically interfere with mAb binding by causing conformational changes in the S protein that shield or impair most other mAb epitopes. COVA1-21 also efficiently blocked virus infection without blocking ACE2, suggesting an alternative mechanism of neutralization than blocking ACE2 engagement (fig. S4C). The SPR-based clustering was corroborated using biolayer interferometry competition assays on a subset of NAbs (fig. S6). Overall, our data are consistent with the previous identification of multiple antigenic RBD sites for SARS-CoV-2 and additional non-RBD sites on the S protein, as described for SARS-CoV and MERS-CoV (32, 42).

To visualize how selected NAbs bound to their respective epitopes, we generated Fab–SARS-CoV-2 S complexes that were imaged by single-particle negative-stain electron microscopy (EM; Fig. 5, B and C, and fig. S7). We obtained low-resolution reconstruction with six Fabs, including five RBD-binding Fabs from three different competition clusters. COVA1-12 overlapped highly with the epitope of COVA2-39, whereas COVA2-04 approached the RBD at a different angle somewhat similar to that of the cross-binding SARS-CoV–specific mAb CR3022 (42). The EM reconstructions confirmed the RBD as the target of these NAbs but revealed a diversity in approach angles (Fig. 5B). Furthermore, whereas four RBD NAbs interacted with a stoichiometry of one Fab per trimer, consistent with one RBD being exposed in the “up state” and two in the less accessible “down state” (13, 43), COVA2-15 bound with a stoichiometry of three per trimer (fig. S7). COVA2-15 was able to bind RBD domains in both the up and down state (Fig. 5D). In either conformation, the COVA2-15 epitope partially overlapped with the ACE2-binding site, and therefore the mAb blocks receptor engagement. The higher stoichiometry of this mAb may explain its unusually strong neutralization potency. None of the epitopes of the five RBD Fabs overlapped with that of CR3022, which is unable to neutralize SARS-CoV-2 (42), although COVA2-04 does approach the RBD from a similar angle as CR3022. The sixth Fab for which we generated a three-dimensional (3D) reconstruction was from the non-RBD mAb COVA1-22 placed in competition cluster IX. EM demonstrated that this mAb bound to the NTD of S1. Such NTD NAbs have also been found for MERS-CoV (44).

Conclusions

Convalescent COVID-19 patients showed strong anti–SARS-CoV-2 S protein–specific B cell responses and developed memory and antibody-producing B cells that may have participated in the control of infection and the establishment of humoral immunity. We isolated 19 NAbs that targeted a diverse range of antigenic sites on the S protein, of which two showed picomolar neutralizing activities (IC₅₀ = 0.007 and 0.009 μg/ml or 47 and 60 pM, respectively) against authentic SARS-CoV-2 virus. This illustrates that SARS-CoV-2 infection elicits high-affinity and cross-reactive mAbs targeting the RBD as well as other sites on the S protein. Several of the potent NAbs had VH segments virtually identical to their germline origin, which holds promise for the induction of similar NAbs by vaccination because extensive affinity maturation does not appear to be a requirement for potent neutralization. The most potent NAbs both targeted the RBD on the S protein and fell within the same competition cluster, but were isolated from two different individuals and bore little resemblance genotypically. Although direct comparisons are difficult, the neutralization potency of these and several other mAbs exceeds the potencies of the most advanced HIV-1 and Ebola mAbs under clinical evaluation, as well as the approved anti-RSV mAb palivizumab (45). Through large-scale SPR-based competition assays, we defined NAbs that targeted multiple sites of vulnerability on the RBD and the additional previously undefined non-RBD epitopes on SARS-CoV-2. This is consistent with the identification of multiple antigenic RBD sites for SARS-CoV-2 and the presence of additional non-RBD sites on the S protein of SARS-CoV and MERS-CoV (32). Subsequent structural characterization of these potent NAbs will guide vaccine design, and simultaneous targeting of multiple non-RBD and RBD epitopes with mAb cocktails paves the way for safe and effective COVID-19 prevention and treatment.