Complete mapping of mutations to the SARS-CoV-2 spike receptor-binding domain that escape antibody recognition

Abstract

Antibodies targeting the SARS-CoV-2 spike receptor-binding domain (RBD) are being developed as therapeutics and make a major contribution to the neutralizing antibody response elicited by infection. Here, we describe a deep mutational scanning method to map how all amino-acid mutations in the RBD affect antibody binding, and apply this method to 10 human monoclonal antibodies. The escape mutations cluster on several surfaces of the RBD that broadly correspond to structurally defined antibody epitopes. However, even antibodies targeting the same RBD surface often have distinct escape mutations. The complete escape maps predict which mutations are selected during viral growth in the presence of single antibodies, and enable us to design escape-resistant antibody cocktails-including cocktails of antibodies that compete for binding to the same surface of the RBD but have different escape mutations. Therefore, complete escape-mutation maps enable rational design of antibody therapeutics and assessment of the antigenic consequences of viral evolution.

Figure 1.. A yeast-display system to completely map SARS-CoV-2 RBD antibody escape mutations.

(A) Yeast display RBD on their surface. The RBD contains a c-myc tag, enabling dual-fluorescent labeling to quantify both RBD expression and antibody binding of RBD by flow cytometry. (B) Per-cell RBD expression and antibody binding as measured by flow cytometry for yeast expressing unmutated RBD and one of the RBD mutant libraries. (C) Experimental workflow. Yeast expressing RBD mutant libraries are sorted to purge RBD mutations that abolish ACE2 binding or RBD folding. These mutant libraries are then labeled with antibody, and cells expressing RBD mutants with decreased antibody binding are enriched using FACS (the “antibody-escape” bin; see Figure S1 for gating details). Both the initial and antibody-escape populations are deep sequenced to identify mutations enriched in the antibody-escape population. The deep-sequencing counts are used to compute the “escape fraction” for each mutation, which represents the fraction of yeast cells with a given RBD mutation that falls into the antibody-escape sort bin. The escape fractions are represented in logo plots, with tall letters indicating mutations that strongly escape antibody binding.

Figure 2.. Complete maps of escape mutations from 10 human monoclonal antibodies.

(A) Properties of the antibodies as reported by Zost et al. (2020a). SARS-CoV-2 neutralization potency is represented as a gradient from black (most potent) to white (non-neutralizing). Antibodies that bind SARS-CoV-1 spike or compete with RBD binding to ACE2 or rCR3022 are indicated in black. (B) Structure of the SARS-CoV-2 RBD (PDB: 6M0J, (Lan et al., 2020)) with residues colored by whether they are in the core RBD distal from ACE2 (orange), in the receptor-binding motif (RBM, light blue), or directly contact ACE2 (dark blue). ACE2 is in gray. RBD sites where any antibody in the panel selects escape mutations are indicated with spheres at their alpha carbons. (C) Maps of escape mutations from each antibody. The line plots show the total escape at each RBD site (sum of escape fractions of all mutations at that site). Sites with strong escape mutations (indicated by purple at bottom of the line plots) are shown in the logo plots. Sites in the logo plots are colored by RBD region as in (B), with the height of each letter representing the escape fraction for that mutation. Note that different sites are shown for the rCR3022-competing antibodies (top four) and all other antibodies (bottom six). (D) Multidimensional scaling projection of the escape-mutant maps, with antibodies having similar escape mutations drawn close together. Each antibody is shown with a pie chart that uses the color scale in (B) to indicate the RBD regions where it selects escape mutations.

Figure 3.. Neutralization assays validate antibody escape maps.

For each of the four indicated antibodies, we chose two mutations that our maps indicated should escape antibody binding, and one or two mutations that should not escape binding. Logo plots show the escape maps for the sites of interest, with the tested mutations that should escape antibody binding in red. Dot plots show the fold change in neutralization (inhibitory concentration 50%, IC50) relative to the unmutated (wildtype) spike measured using spike-pseudotyped lentiviral particles. Fold changes greater than one (dashed gray line) mean a mutation escapes antibody neutralization. Points in red correspond to the mutations expected to mediate escape, and those in blue correspond to mutations not expected to escape (blue letters are not visible in the logo plots as they do not have substantial effects in the mapping). The dotted pink line at the top of some plots indicates the upper limit to the dynamic range; points on the line indicate a fold change greater than or equal to this value. See Figure S3A for the raw neutralization curves, and Figures S3B,C for similar validation for the non-neutralizing antibody rCR3022.

Figure 4.. Structural mapping of antibody binding and escape.

(A-D) For each antibody, the structure shows the RBD surface (PDB 6M0J) colored by the largest-effect escape mutation at each site, with white indicating no escape and red indicating the strongest escape mutation for that antibody. Antibodies are arranged so that those with similar structural epitopes are in the same panel, namely by whether their epitopes are in (A) the core of the RBD, (B) the ACE2-binding ridge, (C) the opposite edge of the RBM, or (D) the saddle of the RBM surface. (E) Crystal structure of the rCR3022-bound RBD (PDB 6W41), with Fab in purple and RBD colored according to sites of escape as in (A). (F) For 5 monoclonal antibodies, Fab bound to SARS-CoV-2 spike ectodomain trimer was visualized by negative-stain electron microscopy (EM). The RBD is modeled as a surface representation, colored according to sites of escape as in (A). Fab chains are modeled in gold. Detailed EM collection statistics are in Table S1. Antibody names are colored according to Figure 2B: core-binding, orange; RBM-binding, cyan; ACE2 contact site-binding, dark blue. See https://jbloomlab.github.io/SARS-CoV-2-RBD_MAP_Crowe_antibodies/ for interactive versions of the escape-colored structures in (A-D).

Figure 5.. Functional and evolutionary constraint on antibody escape mutations.

(A) Variation at sites of antibody escape among currently circulating SARS-CoV-2 viruses. For each site of escape from at least one antibody, we counted the sequences in GISAID with an amino-acid change (there were 93,858 sequences at the time of the analysis). Sites with at least 5 GISAID variants are shown ordered by count; Black cells indicate antibodies with escape mutations at that site. Sites are in orange for the core RBD, light blue for the RBM, and dark blue for ACE2 contact residues. Antibodies are colored according to where the majority of their sites of escape fall. Figure S4 shows similar data broken down by amino-acid change and without count thresholding. (B) Escape maps (as in Figure 2C), with letters colored according to how deleterious mutations are for ACE2 binding or RBD expression effects (Starr et al., 2020). Only sites of escape for each antibody are depicted. Similar logo plots for all antibodies are shown in Figure S5. (C) Mutational constraint on sites of escape. For each antibody, the mean effects of all 19 possible amino acid mutations at sites of escape on ACE2 binding and RBD expression are shown. N_eff) in the sarbecovirus RBD alignment at sites of escape for (D) Top: effective number of amino acids (N_eff is a measure of the variability of a site (the exponentiated Shannon entropy), and each antibody. ranges from 1 for a position that is conserved across all sequences to an upper limit of 20 for a site where ll amino acids are present at equal frequency. Bottom: escape fraction for each sarbecovirus RBD homolog from the yeast display selections.

Figure 6.. Viral escape-mutant selections with individual antibodies and antibody cocktails.

(A) Results of viral selections with five individual monoclonal antibodies. The number of replicates where escape variants were selected are indicated, color coded according to whether escape was selected frequently (red) or rarely (white). The mutations present in the RBD of the selected escape variants are indicated. (B) Each point represents a different amino-acid mutation to the RBD, with the x-axis indicating how strongly the mutation ablates antibody binding in our escape maps (larger values indicate more escape from binding) and the y-axis indicating how the mutation affects ACE2 binding (negative values indicate impaired ACE2 binding). The point shapes indicate whether or not mutations are accessible by single-nucleotide changes, and whether they were selected in viral escape experiments. All selected mutations were accessible by single-nucleotide changes. Note that the only accessible escape mutation from COV2–2165 that is not deleterious to ACE2 binding is D420Y, but this mutation is highly deleterious to expression of properly folded RBD (Figure 5B and S5). (C) Results of viral selections with antibody cocktails, with the last three columns showing the number of replicates with escape out of the total tested. The data for the single antibodies are repeated from (A). In all panels, antibody names are colored according to where in the RBD the majority of their sites of escape fall: orange for the core RBD, light blue for the RBM, and dark blue for ACE2 contact residues. See Figure S6 for additional data relevant to this figure.