Broad and potent activity against SARS-like viruses by an engineered human monoclonal antibody

1The COVID-19 pandemic, caused by the β-coronavirus severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), presents an urgent global health crisis. Although two vaccines and two monoclonal antibody (mAb) therapies have been authorized for emergency use by the U.S. Food and Drug Administration, it is unknown whether these vaccines and treatments will provide broad protection against newly emerging SARS-CoV-2 strains that originate in humans or animal reservoirs. Furthermore, the recurrent zoonotic spillover of CoVs into the human population, along with the broad diversity of SARS-like CoVs circulating in animal reservoirs, suggests that new pathogenic CoVs are likely to emerge in the future and underscores the need for broadly active countermeasures.

2As in other CoVs, the SARS-CoV-2 spike (S) protein mediates viral entry and is the only known target for neutralizing antibodies (nAbs). Although SARS-CoV and SARS-CoV-2 share 76% amino acid identity in their S proteins, only a limited number of cross-neutralizing antibodies have been described to date. These rare broadly neutralizing antibodies (bnAbs) represent an attractive opportunity for therapeutic drug stockpiling to prevent or mitigate future outbreaks of SARS-related CoVs, but their limited neutralization potency may translate into suboptimal protective efficacy or impractical dosing regimens. In this study, we show that such bnAbs can be engineered for improved neutralization potency while retaining neutralization breadth, and we demonstrate that these bnAbs can provide broad protection in vivo.

3We isolated several antibodies from the memory B cells of a 2003 SARS survivor that cross-neutralize multiple SARS-related viruses with relatively modest potency. Although breadth and potency are often opposing characteristics, we sought to engineer these bnAbs for improved neutralization potency against SARS-CoV-2 while maintaining or improving neutralization breadth and potency against other SARS-like viruses. Because binding affinity and neutralization potency are generally well correlated, we employed yeast surface display technology to improve the binding affinities of three of the bnAbs (ADI-55688, ADI-55689, and ADI-56046) for a prefusion-stabilized SARS-CoV-2 S protein.

4Yeast display libraries were generated by introducing diversity into the heavy- and light-chain-variable genes of ADI-55688, ADI-55689, and ADI-56046 through oligonucleotide-based mutagenesis and transformation into Saccharomyces cerevisiae by homologous recombination. After four rounds of selection with a recombinant SARS-CoV-2 S1 protein, improved binding populations were sorted, and 20 to 50 unique clones from each lineage were screened for binding to SARS-CoV-2 S (Fig. 1, A and B, and fig. S1). The highest-affinity binders from each of the three lineages bound to the S protein with monovalent equilibrium dissociation constants (KDs) in the picomolar range, representing 25- to 630-fold improvements in binding relative to their respective parental clones and surpassing the affinities of several clinical-stage nAbs (S309, REGN10987, REGN10933, and CB6/LY-CoV16) (Fig. 1B and fig. S2A). To determine whether the improvements in SARS-CoV-2 S binding affinity translated to enhanced neutralization potency, we selected 9 to 14 affinity-matured progeny from each lineage and evaluated them for SARS-CoV-2 neutralizing activity in a murine leukemia virus (MLV) pseudovirus assay. All of the affinity-matured antibodies showed improved neutralizing activity relative to that of their parental clones, and the most-potent neutralizers from each lineage (ADG-1, ADG-2, and ADG-3) displayed neutralization half-maximal inhibitory concentrations (IC50s) that were comparable to or lower than those observed for the clinical SARS-CoV-2 nAb controls (Fig. 1B). Notably, however, we observed no correlation between the binding affinities and neutralization potencies of ADG-1, -2, and -3 and the clinical-stage antibodies, suggesting that neutralization potency is more tightly linked to fine epitope specificity than binding affinity to prefusion S (fig. S2B).

5Because in vitro engineering can lead to polyreactivity with potential risks of off-target binding and accelerated clearance in vivo, we evaluated ADG-1, ADG-2, and ADG-3 in a polyreactivity assay that is predictive of serum half-life in humans. All three antibodies lacked polyreactivity in this assay, indicating a low risk for poor pharmacokinetic behavior (fig. S3). The three antibodies also showed low hydrophobicity, a low propensity for self-interaction, and thermal stabilities within the range observed for clinically approved antibodies (fig. S3). Thus, the process of in vitro engineering appeared not to negatively affect biophysical properties that are often linked to characteristics such as serum half-life, ease of manufacturing, ability to formulate to high concentrations, and long-term stability.

6To determine whether the process of affinity engineering affected neutralization breadth, we evaluated ADG-1, ADG-2, and ADG-3, as well as their respective parental antibodies, for neutralizing activity against a panel of representative clade 1 sarbecoviruses (SARS-CoV, SHC014-nLuc, SARS-CoV-2-nLuc, and WIV-1-nLuc). SHC014 and WIV-1 were selected because these two bat SARS-like viruses readily replicate in primary human airway cells, suggesting the potential for direct transmission to humans. Consistent with the MLV–SARS-CoV-2 assay results, ADG-2 displayed highly potent neutralizing activity against authentic SARS-CoV-2-nLuc, with an IC50 comparable to or lower than those observed for clinical-stage SARS-CoV-2 nAbs (Fig. 1C and fig. S4). Furthermore, in contrast to the benchmark nAbs, ADG-2 displayed high neutralization potency against authentic SARS-CoV and the two bat SARS-related viruses, with IC50s between 4 and 8 ng/ml (Fig. 1C and fig. S4). ADG-3 and the clinical nAb S309 also cross-neutralized all four sarbecoviruses but with markedly lower potency than that of ADG-2.

7On the basis of its potent cross-neutralization and favorable biophysical properties, we selected ADG-2 as a lead therapeutic candidate and confirmed its potent neutralizing activity in an alternative authentic SARS-CoV-2 neutralization assay (IC50 ~ 1 ng/ml) (Fig. 1, C and D, and fig. S4). Because SARS-CoV-2 D614→G (D614G) has emerged as the dominant pandemic strain, we also performed neutralization studies with MLV–SARS-CoV-2 D614G and confirmed that ADG-2 retains potent neutralizing activity against this strain (fig. S5).

8We further assessed the breadth of sarbecovirus recognition by ADG-2 by measuring its apparent binding affinity (KDApp) to a panel of 17 representative sarbecovirus receptor binding domains (RBDs) expressed on the surface of yeast. Thirteen viruses were selected from clade 1—representing the closest known relatives of SARS-CoV-2 (GD-Pangolin and RaTG13) to the most divergent (SHC014 and Rs4231)—as well as four viruses from the distantly related clades 2 and 3, which do not use ACE2 as a host receptor (Fig. 2A). Recombinant hACE2-Fc and the benchmark SARS-CoV-2 nAbs described above were included as controls. In agreement with prior reports, hACE2 recognized only clade 1 RBDs and bound with higher affinity to SARS-CoV-2 than to SARS-CoV (Fig. 2B).

9Consistent with their broadly neutralizing activities, S309, ADG-2, and ADG-3 displayed broad binding reactivity to clade 1 sarbecovirus RBDs, with ADG-2 and ADG-3 strongly binding 12 of 13 viruses and S309 binding all 13 (Fig. 2B). Notably, ADG-2 bound with high affinity (KDApp = 0.24 to 1.12 nM) to every clade 1 sarbecovirus RBD that exhibited detectable hACE2 binding, supporting the high degree of ADG-2 epitope conservation among sarbecoviruses that can use hACE2 as a receptor. By contrast, ADG-1 bound to only 9 of 13 viruses, and CB6/LY-CoV16, LY-CoV555, REGN10987, and REGN10933 bound to only the closest evolutionary neighbor(s) of SARS-CoV-2, consistent with their narrow neutralization profiles (Figs. 2B and 1C).

10Prior studies have shown that RBD mutants that are resistant to commonly elicited SARS-CoV-2 nAbs are circulating in the human population. We therefore sought to assess the breadth of ADG-2 binding to naturally circulating SARS-CoV-2 variants that contain single amino acid substitutions in the RBD. ADG-1, ADG-3, and the benchmark SARS-CoV-2 nAbs were included as comparators. Using the yeast surface display platform, we expressed the 30 most frequently observed SARS-CoV-2 RBD variants reported in the GISAID database as well as 6 naturally circulating SARS-CoV-2 variants reported to be resistant to certain SARS-CoV-2 nAbs. One or more of the 36 SARS-CoV-2 variants exhibited loss of binding to ADG-1, CB6/LY-CoV16, LY-CoV555, REGN10987, and REGN10933, as defined by >75% loss relative to the wild-type (WT) construct (Fig. 2C). Notably, the loss-of-binding variants identified for REGN10987 and REGN10933 partially overlapped with those identified in previous in vitro neutralization escape studies, validating the use of RBD display for the prediction of antibody escape mutations. By contrast, ADG-2, ADG-3, and S309 bound to all 36 variants at levels ≥50% of those for WT SARS-CoV-2 (Fig. 2C). This result, combined with the substantial neutralization breadth observed for these three mAbs (Figs. 1C and 2, B and D), indicates a potential link between epitope conservation and resistance to viral escape. Finally, all of the antibodies retained high-affinity binding to the N501Y variant, which is the only RBD mutation present in the rapidly spreading B.1.1.7 lineage, suggesting that this newly emerging SARS-CoV-2 strain is likely susceptible to neutralization by these antibodies.

11To gain further insight into the antigenic surface recognized by ADG-2, we generated a mutagenized yeast surface display RBD library and performed rounds of selection to identify RBD variants that exhibited loss of binding to ADG-2 relative to the WT construct (Fig. 3A and fig. S6, A and B). To exclude mutations that globally disrupt the conformation of the RBD, a final round of positive selection was performed using a mixture of recombinant hACE2-Fc and two RBD-directed mAbs (S309 and CR3022) that target nonoverlapping epitopes distinct from the ADG-2 binding site (figs. S6B and S7). Selected RBD mutants were individually tested for binding to ADG-2, recombinant hACE2-Fc, CR3022, and S309 to confirm site-specific knockdown mutations (fig. S6C). Substitutions at only four RBD positions—D405E, G502E/R/V, G504A/D/R/S/V and Y505C/N/S—specifically abrogated ADG-2 binding (Fig. 3B). These four residues are highly conserved among the clade 1 sarbecovirus subgenus and invariant among SARS-CoV-1, SARS-CoV-2, SHC014, and WIV1 viruses (Fig. 3C), providing a molecular explanation for the breadth of binding and neutralization exhibited by ADG-2. Consistent with the conservation of these residues among clade 1 sarbecoviruses, none of the substitutions that affected ADG-2 binding were present in full-length SARS-CoV-2 sequences deposited in the GISAID database at a frequency >0.001% as of 9 December 2020. In addition, three of the four identified mutations that abrogate ADG-2 binding lie within the hACE2 binding site, and at least one mutation at each position (G502E/R/V, G504V, and Y505C/N/S) also abrogated hACE2 binding (Fig. 3B), likely accounting for their absence among circulating SARS-CoV-2 isolates. These results suggest that the evolutionary conservation of the ADG-2 epitope is likely directly linked to ACE2 binding.

12To support the results of this experiment, we performed low-resolution cryo–electron microscopy (cryo-EM) of the complex of ADG-2 bound to prefusion-stabilized SARS-CoV-2 S. This yielded a ~6-Å resolution three-dimensional reconstruction that clearly had at least one ADG-2 Fab bound to an RBD in the up conformation and allowed us to dock in previously determined high-resolution models of the SARS-CoV-2 spike and a homologous Fab (Fig. 3D; fig. S8, A to D; and table S1). Consistent with our fine epitope mapping experiments (Fig. 3B and fig. S7C), the epitope recognized by ADG-2 overlaps the hACE2-binding site and each position identified by epitope mapping clustered to the cleft between the heavy and light chains of ADG-2 (Fig. 3D). This epitope also partially overlaps with those recognized by frequently observed “class 1” SARS-CoV-2 nAbs (Fig. 3E). However, in contrast to previously reported nAbs in this class, ADG-2 binds with a divergent angle of approach and displays broadly neutralizing activity (Figs. 3E and 1C and fig. S8E). Thus, ADG-2 binds to a highly conserved motif through a distinct angle of approach.

13Because Fc-mediated effector functions can contribute to protection independently of viral neutralization, we assessed the ability of ADG-2 to induce antibody-dependent natural killer (NK) cell activation and degranulation, antibody-dependent cellular phagocytosis mediated by monocytes and neutrophils, and antibody-mediated complement deposition using in vitro effector function assays. Benchmark SARS-CoV-2 nAbs S309 and REGN10987 were included as comparators. Notably, though ADG-2, S309, and REGN10987 showed comparable recruitment of phagocytosis (Fig. 4B), these antibodies differed with respect to complement deposition and NK cell activation (Fig. 4, A and C): S309 showed reduced complement deposition compared with ADG-2 and REGN10987, and ADG-2 showed superior NK cell activation to both S309 and REGN10987 (Fig. 4). In summary, ADG-2 robustly triggers diverse Fc-mediated effector activities with potencies comparable or superior to those of current SARS-CoV-2 clinical antibodies.

14Finally, we tested the ability of ADG-2 to provide broad in vivo protection in immunocompetent mouse models of SARS and COVID-19 using mouse-adapted SARS-CoV (MA15) and SARS-CoV-2 (MA10), respectively. Balb/c mice were prophylactically treated with either 200 μg of ADG-2 or phosphate-buffered saline (PBS) via intraperitoneal injection 12 hours before intranasal challenge with an MA15 or MA10 dose of 103 plaque-forming units (PFU). All mice were monitored daily for weight loss and changes in respiratory function, and groups of mice were euthanized at day 2 or 4 postinfection to allow for measurement of virus replication in the lung and analysis of lung histopathology. We observed substantial, progressive weight loss in sham-treated mice infected with both viruses along with increases in enhanced pause (Penh), a calculated measure of airway resistance. By contrast, mice treated prophylactically with ADG-2 demonstrated minimal weight loss, no change in Penh, and no signs of gross pathology at the time of harvest (Fig. 5, A and B). Furthermore, prophylactic antibody treatment prevented viral replication in the lungs at both 2 and 4 days postinfection (dpi). We next investigated the ability of ADG-2 to act antivirally against SARS-CoV-2 MA10 in a therapeutic setting. Mice were treated with 200 μg of ADG-2 or PBS 12 hours after intranasal challenge with a 103-PFU dose of MA10. Mice given therapeutic ADG-2 had intermediate levels of weight loss, moderate respiratory function changes, and some gross lung pathology—significantly more than prophylactically treated mice but significantly less than sham-treated mice (Fig. 5C). Therapeutic antibody treatment also resulted in a significant reduction in lung viral loads at 4 dpi, but not at 2 dpi, relative to sham-treated mice. We conclude that ADG-2 treatment can reduce disease burden in mice infected with both SARS-CoV MA15 and SARS-CoV-2 MA10.

15Since the beginning of the COVID-19 pandemic, a plethora of potently neutralizing SARS-CoV-2 antibodies have been isolated, and some have rapidly advanced to clinical trials. However, the epitopes recognized by most of these nAbs are highly variable among other clade 1 sarbecoviruses, hence limiting their neutralization breadth and increasing their susceptibility to antibody escape mutations. Here, we describe an engineered antibody that neutralizes SARS-CoV-2 with a potency that rivals current lead SARS-CoV-2 clinical nAbs but also broadly neutralizes other clade 1 sarbecoviruses, potently triggers Fc-mediated effector functions, and provides significant protection against SARS and COVID-19 in mouse models. Thus, ADG-2 represents a promising candidate for the prevention and treatment of not only COVID-19, but also future respiratory diseases caused by pre-emergent SARS-related CoVs. Furthermore, our fine epitope mapping and structural studies demonstrate that ADG-2 employs a distinct angle of approach to recognize a highly conserved epitope that overlaps the receptor binding site. This epitope represents an Achilles’ heel for clade 1 sarbecoviruses and hence is an attractive target for the rational design of “pan-SARS” vaccines that aim to elicit similar broadly protective antibodies.