Skip to main content
Journal of Virology logoLink to Journal of Virology
. 2026 Oct 14;100(11):e01001-26. doi: 10.1128/jvi.01001-26

Variant escape site mapping and characterization of SARS-CoV-2 nucleocapsid rapid antigen test antibodies

Alasdair D Keith 1, Meredith M Keen 1,2, Filipp Frank 1,3, Wilbur A Lam 2,3,4,5, Morgan Greenleaf 2, Eric A Ortlund 1,2,✉
Editor: Rebecca Ellis Dutch6
PMCID: PMC13776708  PMID: 42872825

ABSTRACT

Rapid antigen tests (RATs) were a critical tool for diagnosing severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection during the height of the pandemic, but their performance depends on antibodies that remain robust to viral evolution. Since most commercial RATs target the viral nucleocapsid protein, mutations in this domain can compromise binding and test performance. To systematically define this risk, we used deep mutational scanning (DMS) to map escape mutations for 82 antibodies across 43 commercially available tests, generating the most comprehensive escape database of SARS-CoV-2 viral mutations to date. Most antibodies target conformational epitopes, with 47 (57%) mapping to the RNA-binding domain and 23 (28%) to the dimerization domain. Principal component analysis and clustering revealed escape phenotypes, including shared features outside of predicted epitopes. As the pandemic proceeded, we noted a change in industry preference from IgG to IgM-type antibodies, which was associated with changes to epitope sub-class prevalence. Analysis of paired capture-detection antibodies revealed a preference for targeting the same structural domain (20/33 or 61%), indicating a slight preference for test antibodies to hit orthogonal yet proximal sites. Together, the results establish a comprehensive resource for understanding how SARS-CoV-2 nucleocapsid protein mutations affect antibody binding and the phenotypes that underlie RAT antibody selection. This resource and approach can help prioritize antibodies for follow-up performance testing and support the design of more mutation-resilient diagnostics.

IMPORTANCE

It is critical during a pandemic for diagnostic tests to be accurate and for the false negative rate to be kept low. The rapidly changing nature of SARS-CoV-2 exposed a weakness with using rapid antigen tests to diagnose infection, as mutations to the viral nucleocapsid protein can cause the virus to evade detection. In this work, we systematically tested the robustness of 82 antibodies from 43 separate commercially available tests by deliberately mutating the nucleocapsid protein and investigating which changes caused weakened binding to the test antibodies. In addition to determining which specific tests were at risk from certain mutations, we collated our datasets into a meta-analysis, allowing us to identify some best practices for designing robust tests. We, therefore, expect that this resource will support the design of more mutation-resilient diagnostics, whether for future variants of SARS-CoV-2 or in applying these principles to other respiratory viruses, such as influenza virus.

KEYWORDS: nucleocapsid, rapid antigen test diagnostics, SARS-CoV-2, deep mutational scanning

INTRODUCTION

Rapid antigen tests (RATs) played a central role in decentralized diagnosis during the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) pandemic, but their reliability depends on antibodies that remain robust to viral evolution (1). This issue is particularly acute if these viral changes include mutations to the nucleocapsid protein (NP) since the diagnostic efficacy of RATs typically depends on antibodies that bind to this protein. Despite these risks, there has been limited ability to systematically evaluate how viral mutations affect the performance of diagnostic antibodies, with traditional approaches limited by time and cost constraints. As such, there is a critical need for scalable methods that can prospectively define the susceptibility of diagnostic antibodies to antigenic variation. This is a pressing public health concern since, during the pandemic, significant decreases in test sensitivity were observed, such as a reduction from near 90 to 0–26% sensitivity upon the emergence of the Omicron variant for all 10 of the most commonly used RATs at the time (2, 3). False negatives are especially dangerous from a public health perspective, as they can result in protective measures being mistakenly lifted, facilitating viral transmission (4), as well as discouraging vulnerable individuals from seeking urgent medical treatment.

Anticipating these issues, a U.S. Rapid Acceleration of Diagnostics (RADx) Variant Task Force was set up in January 2021 (5). This task force studied the sensitivities of RATs against multiple arising variants (6) and has published recommendations for best practices for variant testing panels (7). Further groups from other countries have conducted similar sensitivity studies (8), such as for the later BA.2.86 variant (9). These studies are informative but do not provide direct information on antibody-antigen interactions. A method of evaluating RAT antibody robustness against emerging viral strains can be of significant utility. Such tests provide information on epitope-paratope interactions, thus highlighting which antibodies may be prone to escape when the SARS-CoV-2 nucleocapsid mutates at certain regions.

We previously developed a protocol combining mammalian cell surface display with deep mutational scanning (DMS) to measure the effect of SARS-CoV-2 nucleocapsid point mutations on escape from binding to antibodies used in commercially available RATs (Fig. 1) (10). DMS is a powerful, high-throughput technique that utilizes deep sequencing to determine how mutations respond to an external pressure (11–13) or how they impact protein stability and/or binding (14, 15). We can determine antibody-antigen variant binding affinities, allowing for the inference of epitope-paratope interactions. In our assay, we express nucleocapsid variants on the surface of mammalian cells and incubate with the RAT antibody. By selecting a population of low-affinity antibody-antigen binders and comparing against a reference population, following sequencing, we can determine which nucleocapsid mutations cause escape from binding to the antibody. This approach effectively covers all possible single non-synonymous mutations across the full length of the nucleocapsid protein. We typically recognize regions of high escape as the binding site or an important secondary site exerting allosteric effects on binding. These data allow us to infer epitopes with high confidence (13, 16) and, therefore, predict vulnerability to antigenic variation to prioritize antibodies for follow-up performance testing.

Fig 1.

Diagram of a DMS workflow for nucleocapsid variants using a lentiviral construct, surface display with FACS sorting, PacBio and Illumina sequencing, and a heatmap showing escape scores by N site.

Deep mutational scanning (DMS) strategy. (A) Schematic of lentiviral construct used to generate a stable mammalian cell line. (B) Schematic for surface expression. First, nucleocapsid is expressed as a dimer on the surface of HEK293 cells. Cells are then incubated with the RAT antibody, followed by incubation with an appropriate secondary antibody and an anti-Myc antibody to control for surface expression levels. (C) DMS workflow. Starting from a barcoded library of nucleocapsid, in step 1, PacBio sequencing was applied to generate a lookup table associating nucleocapsid variants with barcodes. Therefore, all following assays could proceed using short-read Illumina sequencing, as described in step 2. First, the barcoded library was transduced into HEK293 cells, then exposed to the anti-N, anti-Myc, and secondary antibodies. FACS was then used to select for weakest binders, which comprises our escape population. RNA was extracted from this population as well as from an unsorted reference population. Both sets of RNA were reverse-transcribed and amplified, then the barcodes were sequenced. By referring to the lookup table, nucleocapsid variants were inferred from the barcodes. The data were then filtered and, by comparing the escape and reference population statistics, an escape score per nucleocapsid variant was generated. These scores can be visualized using heatmaps. CMV, cytomegalovirus; FACS, fluorescence-activated cell sorting; GFP, green fluorescent protein; IRES, internal ribosomal entry site; LTR, long terminal repeat; nt, nucleotide; SP, signal peptide; TM, transmembrane.

We previously published a study at the height of the pandemic in August 2022, in which we tested 17 antibodies from 11 different antigen tests (10). These results have been used to guide test design and inform risk assessment for federal governing bodies, such as the National Institutes of Health (NIH) and U.S. Food and Drug Administration (FDA). We have now substantially expanded our analysis to 82 antibodies from 43 separate tests provided by external companies, representing the largest collection of single-site escape scores for SARS-CoV-2 RAT mAbs to date. This expanded data set allows us not only to map escape mutations at scale, but also to ask whether commercial RAT antibodies converge on recurrent escape phenotypes, whether antibody choices shifted over time as viral lineages changed, and whether common design principles govern paired capture and detection antibodies. By integrating this large number of escape map data sets with comparative and temporal analyses, we have established a resource for understanding how NP variation affects diagnostic antibody binding. This asset should, therefore, inform researchers interested in understanding the shared features of mAbs used in commercial tests.

RESULTS

Escape phenotype mapping

We leveraged our existing DMS library based on the Wuhan viral strain covering 99% of non-synonymous point mutations. Briefly, we transduced HEK293 with a lentiviral construct to generate a stable cell line expressing a single nucleocapsid variant per cell. We then applied fluorescence-activated cell sorting to select the lowest 15% of binders (see Fig. S1 for the gating scheme) and sequenced DNA from this population, alongside an unsorted reference sample, to calculate the relative abundance of each variant. Variants enhanced in the population of lowest binders constitute ‘escape’ mutations, i.e., mutations that compromise binding between nucleocapsid and the antibody being tested.

Overall, we investigated 82 antibodies, with epitopes inferred for each from escape heatmap plots (Data S1) and collated (Table 1; Table S1). KD values for binding between each antibody and wild-type nucleocapsid protein (WT NP) are also provided in Table 1, with full binding curves available in Data S1.

TABLE 1.

Anti-nucleocapsid antibody propertiesa

Company Test Antibody KD (nM) IgG/IgM Epitope class PCA escape phenotype
Abbott BinaxNOW ADK-214 1.0 IgG DD DD-LNK (1)
NanoDitech Nano-Check ND1 1.3 IgG DD DD-LNK (1)
Sekisui Flu SARS-CoV-2 Combo Home Test S32213 0.59 IgG DD DD-LNK (1)
Sekisui Flu SARS-CoV-2 Combo Home Test S32217 0.064 IgG DD DD-LNK (1)
Acon EUA210494 Flowflex mAb-N 0.12 IgG DD DD-INTERFACE (2)
ANP Technologies NIDS COVID-19 Antigen Rapid Test N-Ab-3 0.18 IgG DD DD-INTERFACE (2)
AvantGen AccuRate COVID-19 SelfCheck C7 4.7 IgG DD DD-INTERFACE (2)
AvantGen AccuRate COVID-19 SelfCheck F5mod 0.48 IgG DD DD-INTERFACE (2)
CorDx COVID-19 Ag Test Ab339 0.7 IgG DD DD-INTERFACE (2)
GenBody COVID-19 Ag 2F4 6.6 IgG DD DD-INTERFACE (2)
GenBody COVID-19 Ag 3C3 1.3 IgG DD DD-INTERFACE (2)
Luminostics (ClipHealth) Clip Flu + COVID HomeTest mAb1 0.27 IgG DD DD-INTERFACE (2)
Mologic COVI-Go Self-Test 5B1 1.2 IgG DD DD-INTERFACE (2)
Princeton BioMeditech Corporation (PBMC) Status COVID-19 Antigen Test Ab2 1.5 IgG DD DD-INTERFACE (2)
Quidel QuickVue At-Home OTC COVID-19 Test 1C1 1.9 IgG DD DD-INTERFACE (2)
Quidel Sofia SARS Antigen FIA Ab #166 3.6 IgG DD DD-INTERFACE (2)
CorDx COVID-19 Ag Test Ab063 12 IgM DD DD-FLANK (3)
OraSure Technology InteliSwab COVID-19 Rapid Test mAb-OTI-1 1.3 IgG DD DD-FLANK (3)
Roche/SD/Bionote Biosensor Pilot COVID-19 At-Home Test MAD4904-P-2209 (B) 2.0 IgG DD DD-FLANK (3)
Roche/SD/Bionote Biosensor Pilot COVID-19 At-Home Test mAb22-048 (C) 1.0 IgM DD DD-FLANK (3)
Siemens Atellica IM or ADVIA Centaur SARS-CoV-2 Antigen Test rSN-327 0.65 IgG DD DD-FLANK (3)
Acon EUA210494 Flowflex mAb-L 1.6 IgM RBD RBD-GENERAL (4)
AzureBiotech Fastep COVID-19 Antigen Home Test mAb5 1.1 IgG RBD RBD-GENERAL (4)
Ellume COVID-19 Home Test MM05 3.3 IgG RBD RBD-GENERAL (4)
Healgen Heal-Check Rapid COVID-19 + Influenza A/B Antigen Test Mab2 3.9 IgM RBD RBD-GENERAL (4)
MaximBio ClearDetect COVID-19 Antigen Home Test 01127RC17602 7.6 IgG RBD RBD-GENERAL (4)
OraSure Technology InteliSwab COVID-19 Rapid Test mAb-OTI-2 0.38 IgG RBD RBD-GENERAL (4)
Phase Scientific SARS-CoV-2 Antigen Tests Labeling Antibody #200619 0.22 IgG RBD RBD-GENERAL (4)
Quanterix Simoa SARS-CoV-2 N Protein Antigen Test MM05 3.3 IgG RBD RBD-GENERAL (4)
Watmind SARS-CoV-2 Ag Self-Test Kit COVID-19-PS-MAB115 2.7 IgG RBD RBD-GENERAL (4)
Watmind SARS-CoV-2 Ag Self-Test Kit COVID-19-PS-MAB117 2.6 IgG RBD RBD-GENERAL (4)
Healgen Heal-Check Rapid COVID-19 + Influenza A/B Antigen Test Mab117 8.2 IgG RBD LIN-CTD (5)
Mologic COVI-Go Self-Test 5E2 1.7 IgG Linear LIN-CTD (5)
NanoDitech Nano-Check ND2 6.7 IgG RBD LIN-CTD (5)
Phase Scientific SARS-CoV-2 Antigen Tests Capture Antibody #200707 5.0 IgG RBD LIN-CTD (5)
Quidel Sofia SARS Antigen FIA R040 0.78 IgG Linear LIN-CTD (5)
University of California, San Diego (UCSD) Not applicable #9 7.4 IgG DD LIN-CTD (5)
Luminostics (ClipHealth) Clip COVID Rapid Antigen Test mAb1 0.35 IgG RBD RBD-CBODY (6)
Luminostics (ClipHealth) Clip COVID Rapid Antigen Test mAb2 0.037 IgG RBD RBD-CBODY (6)
MaximBio ClearDetect COVID-19 Antigen Home Test 01127RC17604 5.4 IgG RBD RBD-CBODY (6)
Osang OHC COVID-19 Antigen Self Test FPZ 0.062 IgG RBD RBD-CBODY (6)
Quanterix Simoa SARS-CoV-2 N Protein Antigen Test R004 12 IgG RBD RBD-CBODY (6)
Quidel QuickVue At-Home OTC COVID-19 Test 1A7 0.11 IgG RBD RBD-CBODY (6)
BD Veritor At-Home COVID-19 Test C518 3.0 IgG RBD RBD-SBODY (7)
BD Veritor At-Home COVID-19 Test C524 0.3 IgG RBD RBD-SBODY (7)
BD Veritor At-Home COVID-19 Test C706 4.1 IgG RBD RBD-SBODY (7)
Ellume COVID-19 Home Test MM08 1.6 IgG RBD RBD-SBODY (7)
LumiQuick Boson Rapid SARS-CoV-2 Antigen Test Card mAb-2 1.2 IgG None identified RBD-SBODY (7)
Osang OHC COVID-19 Antigen Self Test C706 1.1 IgG RBD RBD-SBODY (7)
Qorvo Biotechnologies Omnia SARS-CoV-2 Antigen Test C706 4.1 IgG RBD RBD-SBODY (7)
TwistBiosciences Rapid Antigen SARS-CoV-2 Test MM08 1.6 IgG RBD RBD-SBODY (7)
TwistBiosciences Rapid Antigen SARS-CoV-2 Test C706 4.1 IgG RBD RBD-SBODY (7)
AzureBiotech Fastep COVID-19 Antigen Home Test mAb4 5.7 IgG None identified LIN-LNK (8)
Siemens Atellica IM or ADVIA Centaur SARS-CoV-2 Antigen Test rSN-69 0.31 IgG Linear LIN-LNK (8)
Siemens Atellica IM or ADVIA Centaur SARS-CoV-2 Antigen Test SNS-25 1.9 IgG Linear LIN-LNK (8)
Boston Molecules Not known 3A7 54 IgG Linear LIN-NTD (9)
Boston Molecules Not known 11B12 2.8 IgG Linear LIN-NTD (9)
LumiQuick Boson Rapid SARS-CoV-2 Antigen Test Card mAb-1 0.4 IgG DD LIN-NTD (9)
Princeton BioMeditech Corporation (PBMC) Status COVID-19 Antigen Test Ab1 1.2 IgG None identified LIN-NTD (9)
LumiraDx SARS-CoV-2 Ag Test Fab4 (biotin) 2.3 IgG RBD RBD-CTD (10)
LumiraDx SARS-CoV-2 Ag Test Fab5 0.65 IgG RBD RBD-CTD (10)
Siemens Atellica IM or ADVIA Centaur SARS-CoV-2 Antigen Test rSN-25 1.8 IgG RBD RBD-CTD (10)
Siemens Atellica IM or ADVIA Centaur SARS-CoV-2 Antigen Test rSN-122 0.27 IgG RBD RBD-CTD (10)
WebbDx Porta-Kleer Test Capture Antibody 0.60 IgG RBD RBD-CTD (10)
WebbDx Porta-Kleer Test Detector Antibody 3.41 IgG RBD RBD-CTD (10)
Xtrava Health SPERA COVID-19 Ag Test Clone 07 3.4 IgG RBD RBD-CTD (10)
Xtrava Health SPERA COVID-19 Ag Test Clone 09 4.3 IgG RBD RBD-CTD (10)
Luminostics (ClipHealth) Clip Flu + COVID HomeTest mAb2 1.8 IgG Linear MIXED (11)
Osang Quickfinder COVID-19/Flu Antigen Self Test BMRnp311 0.26 IgG RBD MIXED (11)
Osang Quickfinder COVID-19/Flu Antigen Self Test BMRnp314 2.3 IgG RBD MIXED (11)
Phase Scientific INDICAID HA811-36MB 4.1 IgG None identified MIXED (11)
Phase Scientific INDICAID HA811-37R 0.58 IgG RBD MIXED (11)
Quidel-Ortho Not known R001 41 IgG Linear MIXED (11)
Abbott BinaxNOW N18 0.83 IgG RBD RBD-SIGNATURES (12)
AccessBio CareSuperb COVID-19 Home Test R0112 0.52 IgM RBD RBD-SIGNATURES (12)
AccessBio CareSuperb COVID-19 Home Test R0113 0.22 IgM RBD RBD-SIGNATURES (12)
BD Veritor System for Rapid Detection of SARS-CoV-2 FPZ0639 0.071 IgM RBD RBD-SIGNATURES (12)
BD Veritor System for Rapid Detection of SARS-CoV-2 FPZ0643 0.048 IgM RBD RBD-SIGNATURES (12)
iHealth COVID-19/Flu A&B Rapid Test & Flu A&B/COVID-19/RSV Rapid Test Capture Antibody 1.2 IgG RBD RBD-SIGNATURES (12)
iHealth COVID-19/Flu A&B Rapid Test & Flu A&B/COVID-19/RSV Rapid Test Labeling Antibody 0.051 IgG RBD RBD-SIGNATURES (12)
Wondfo WELLlife COVID-19/Influenza A&B Test Mab15 0.06 IgM RBD RBD-SIGNATURES (12)
Wondfo WELLlife COVID-19/Influenza A&B Test Mab19 2.4 IgM RBD RBD-SIGNATURES (12)
ANP Technologies NIDS COVID-19 Antigen Rapid Test MM05 3.5 IgG RBD –
Qorvo Biotechnologies Omnia SARS-CoV-2 Antigen Test C524 0.5 IgG RBD –
Qorvo Biotechnologies Not known C518 3.0 IgG RBD –
a

Antibodies are organized by PCA escape phenotype. Epitope class column was determined by inspection of heatmaps (available from Data S1) and the PCA escape phenotype column by principal component analysis (PCA). Validation of our PCA suggested the optimal number of clusters, k, was 12. We assign each of these clusters a number and a label: 1, DD-LNK; 2, DD-INTERFACE; 3, DD-FLANK; 4, RBD-GENERAL; 5, LIN-CTD; 6, RBD-CBODY; 7, RBD-SBODY; 8, LIN-LNK; 9, LIN-NTD; 10, RBD-CTD; 11, MIXED; 12, RBD-SIGNATURES. Cases where the inspected epitope and PCA-derived escape phenotype classes are inconsistent are highlighted in bold font. “–” indicates PCA phenotype classifications for three of the 85 antibodies are missing, as the data from these submissions were no longer available for use, leaving 82 in total. To assist the temporal analysis presented later in this paper, Table S1 presents the same data organized by date of receipt from the clients. DD, dimerization domain; RBD, RNA-binding domain.

Some companies market RATs containing antibodies sourced from contract research organizations (CROs) and other parties, as it is common practice for test developers to license mAbs already proven as effective for use in their own test formats. As such, it is possible for formats to differ while using the same mAbs. For example, at the beginning of the pandemic, multiple companies used the MM05 and MM08 antibodies developed by SinoBiological and the C518, C524, and C706 antibodies available from Advanced Immunochemical, Inc. We also wished to determine whether any companies were using the same antibodies but submitted to us under different names. We, therefore, created a correlogram, which shows escape phenotype correlations between all the antibodies we studied (Fig. 2). The MM05 antibodies used by Ellume and Quanterix were submitted to us as separate samples but we confirmed they were from the same source, so we ran a single DMS assay, hence the same KD values and epitopes described in Table 1 for these antibody/company pairings. The same was true for the MM08 antibody used by Ellume and TwistBiosciences and the C706 antibody used by BD, Osang, and TwistBiosciences. This translates to a Pearson value of 1.000 for these pairings within our correlogram (Fig. 2). All other DMS assays were run independently, even when companies were using the same name for their antibodies. Eighty-two of our 85 antibodies were analyzed, with the submissions from Qorvo Biotechnologies (antibodies C524 and C518) and ANP Technologies (MM05) remaining confidential.

Fig 2.

Triangular correlation matrix of 82 anti-nucleocapsid antibodies showing Pearson coefficients from 1 to −0.142. Clustered dark blue blocks indicate highly correlated escape patterns among antibody pairs.

Anti-nucleocapsid antibody correlogram. Eighty-two antibodies are plotted. Correlations between antibodies are calculated for the maximum number of variants common to the two DMS data sets being compared minus variants at sites 2–14, 18–30, and 32–41, which are associated with Myc anti-Myc binding, and 51–53, 71–73, 84–87, 110–116, 130, and 132, which are located in the core of the RBD and so are associated with disruption of domain stability instead of with epitopes. Boxes are color-coded according to the Pearson coefficient, ρ, scaling from +1 (dark blue) to –1 (dark red). Antibody correlates exceeding ρ = 0.65 are provided and ranked, alongside the antibody pair with the lowest correlation.

The strongest correlation for antibodies tested in separate assays was for BD Biosciences FPZ0639 and the biotinylated version of this IgM, FPZ0643 (ρ = 0.883). Throughout our analyses, we had to consider that the DMS technique is subject to sampling error (17, 18), which can be exacerbated by inconsistencies in the drawing of FACS gates between assays (19). Therefore, the fact that these antibodies gave such similar escape phenotypes gave us confidence that antibody pairs with high correlations were likely to be the same, but perhaps with different formulations or chemically modified to enable different test configurations.

Similarly, a high correlation (ρ = 0.807) between the Labeling Antibody #200619 from Phase Scientific and mAb-L from Acon Laboratories suggested they were derived from the same source. Both have effectively linear escape phenotypes contained in the RBD (Data S1). Phase Scientific had listed their antibody as an IgG1, whereas Acon listed theirs as an IgM, so we were surprised by this high level of correlation. Our hypothesis is that one of these antibodies was misclassified, although it is possible that one of these companies reformatted the IgG to an IgM or vice versa.

The next-strongest correlation (ρ = 0.789) was between the R004 antibody from Quanterix and the mAb2 antibody from the Clip COVID Rapid Antigen Test marketed by Luminostics (ClipHealth). This pair was identified as being the same antibody in our previous publication (10) through simple inspection of the respective heatmaps. Therefore, despite being submitted to us and marketed under different names, our DMS method could identify that they were likely purchased from the same source (in this case, from SinoBiological). Due to a high correlation between the R004 antibody and 1A7 from Quidel (ρ = 0.645), we conclude that 1A7 is likely to be from this same source.

Of the 82 antibodies investigated, our inspection of heatmaps suggested that 47 (57%) had conformational epitopes in the RNA-binding domain (RBD); 23 (28%) had conformational epitopes in the dimerization domain (DD); and eight (10%) epitopes were linear. We could not confidently identify an epitope for four antibodies: PBMC Status COVID-19 Antigen Test Ab1, LumiQuick Boson Rapid SARS-CoV-2 Antigen Test Card mAb-2, AzureBiotech Fastep COVID-19 Antigen Home Test mAb4, and Phase Scientific INDICAID HA811-36MB. Of the linear epitopes, five were in the C-terminal unstructured region; two (both antibodies submitted by Boston Molecules) were located in the N-terminal unstructured region; and one (Siemens Atellica IM or ADVIA Centaur SARS-CoV-2 Antigen Test SNS-25) was in the unstructured linker region.

Principal component analysis: PCs

We performed a principal component analysis (PCA) to identify more precisely the functional phenotypes within the nucleocapsid driving escape from the RAT mAbs (Fig. 3). We extracted the top 250 loadings for each of the PC1 positive, PC1 negative, PC2 positive, and PC2 negative sets and determined whether any sites were appearing multiple times in each (Fig. S2A through D), allowing us to infer which sites were influential in separating the components. We found that, along the PC1 axis, highly positive sites all occurred in the DD (Fig. S2A), whereas highly negative ones were all contained in the RBD (Fig. S2B). Along the PC2 axis, all the highly positive sites also occurred in the DD; indeed, many of the sites that have large positive values along PC1 are the same as those that have large positive values along PC2 (compare Fig. S2A and C). For negative values, this was not the case: the highly negative sites along PC2 did not typically match those from PC1 (compare Fig. S2B and D). Most sites in the PC2 negative case are located in the RBD, indicating that this is an important feature separating antibodies that is orthogonal to the other DD/RBD distinction found along the PC1 axis. Antibodies with high escape scores in the DD are typically found in the top-right quadrant, and those with high escape scores in the RBD are in the top-left quadrant. Antibodies with high escape outside these two domains are located near the origin, with a further unique cluster of antibodies with high escape in the RBD located at the bottom of the plot (Fig. 3).

Fig 3.

PCA scatter plot showing 12 epitope clusters of anti-nucleocapsid antibodies. RBD clusters occupy negative PC1; DD clusters positive PC1; and linear phenotypes are central. Cluster 12 is a notable outlier.

Anti-nucleocapsid antibody escape score principal component analysis (PCA) and clustering. The gap method for validation of our PCA suggested that the optimal number of clusters, k, was 12. Of the 12 clusters, three were for linear escape phenotypes, three were located in the DD, five in the RBD, and one was difficult to classify. Meta clusters are hand-drawn to identify antibodies with strong escape scores with contiguous sites or in the RBD and DD. Antibodies with RBD-located phenotypes are typically mapped in the top-left quadrant of the PCA diagram (negative PC1 and positive PC2), those with DD-located phenotypes in the top-right quadrant (positive PC1 and positive PC2), and the linear phenotypes in a central position. Cluster 12 (meta cluster RBDx), which ostensibly contains antibodies with an RBD-located phenotype, is an outlier: this cluster occupies a neutral PC1, highly negative PC2 position. We labeled the clusters according to escape score properties, as explained in the text: 1 = DD-LNK; 2 = DD-INTERFACE; 3 = DD-FLANK; 4 = RBD-GENERAL; 5 = LIN-CTD; 6 = RBD-CBODY; 7 = RBD-SBODY; 8 = LIN-LNK; 9 = LIN-NTD; 10 = RBD-CTD; 11 = MIXED; 12 = RBD-SIGNATURES.

By inspecting loadings (Table S2; Fig. S2), we determined the variants contributing most to the primary and secondary components in both the positive and negative directions; these variants, therefore, constitute those which differentiate mAbs to the greatest extent. G335Y had the largest positive loading value for PC1 and second-highest value for PC2; this is a mutation located within the DD, which involves a highly disruptive change from a small to a large amino acid. F66C, in the RBD, had the largest negative loading value for PC1. R276L has the largest negative loading values for PC2. Alternative mutations at this site do not have large negative loading values (see Table S2), suggesting that conversion to leucine is important. R276 is located in the DD and forms an interaction with T366 on an unstructured loop at the CTD (Fig. S2F); conversion to leucine disrupts the complementary polarity of this interaction.

Principal component analysis: clusters

Following standard validation methods, we selected 12 as our cluster number, k (see Materials and Methods and Fig. S3 to S5 for details). Of these clusters, three described conformational escape phenotypes located primarily in the DD; three described linear escape phenotypes; five described conformational escape phenotypes primarily in the RBD; and one contained a mix of types.

To visualize the general escape phenotype for each cluster, we plotted heatmaps, calculated the average per-site escape scores, and mapped the highest scoring sites onto three-dimensional (3D) structures (Fig. 4 to 6). For details on these calculations, see Materials and Methods.

Fig 4.

Dot heatmaps show nucleocapsid escape scores across sites 2 to 419 for DD-LNK, DD-INTERFACE, DD-FLANK, and RBD-SIGNATURES antibody clusters, with matching 3D structures highlighting epitope locations.

Three-dimensional (3D) heatmap and structural depiction of escape phenotypes by cluster (Set 1: DD features and RBD-SIGNATURES). Heatmaps show the average scores for antibodies within a given cluster. The x-axis depicts the nucleocapsid site (2–419), and the y-axis gives the amino acid. Black dots throughout the heatmap represent the wild-type sequence. All other dots are color-coded according to the escape score, with a deeper purple color indicating higher escape. The masked blocks represent variants that are either associated with Myc anti-Myc binding or are located in the core of the RBD and so can disrupt nucleocapsid stability when mutated; variants within these blocks, therefore, do not contribute to the epitope, even if they have high escape scores. Bar charts above the heatmaps give the average escape scores per site. A pink / purple bar occurs when the score exceeds 1 / 2 standard deviation(s) above the mean, and a red star is given when at least eight individual mutations at that site are at least 1 standard deviation above the mean. The number of antibodies in each cluster is provided in circled text. For each cluster, a full depiction of the nucleocapsid dimer is given (top) in addition to magnified regions (bottom). Color-coding of the 3D structures is by domain type (NTD = gray; RBD = cyan; LNK = orange; DD = pink; CTD = green) with a lighter shade used to depict chain A and a darker one for chain B. Sites with strong and moderate escape scores (as defined in Materials and Methods) are highlighted in purple and pink. (A) DD-LNK cluster. As well as an epitope in the dimerization domain, members of this class have a strong feature in the linker region. (B) DD-INTERFACE cluster. Members have epitopes in the DD that run across the interface between the two chains. (C) DD-FLANK cluster. Members have epitopes in the DD distal from the inter-chain interface. (D) RBD-SIGNATURES cluster. Members have epitopes in the RBD plus have a range of signature strong escape mutations situated throughout the protein.

Fig 6.

Dot heatmaps show nucleocapsid escape scores acros sites 2-419 for LIN-NTD, LIN-CTD, LIN-LNK and MIXED antibody clusters, with matching 3D structures highlighting epitope locations.

Three-dimensional heatmap and structural depiction of escape phenotypes by cluster (Set 3: linear features and MIXED). Heatmap and structure formats are as described in Fig. 4. (A) LIN-NTD cluster. Members have linear epitopes in the N-terminal domain. (B) LIN-CTD cluster. Members have linear epitopes in the C-terminal domain. (C) LIN-LNK cluster. Members have linear epitopes in the linker domain. (D) MIXED cluster. Members within this cluster have a range of epitope types.

The first three clusters in Fig. 4 all describe antibodies with escape phenotypes in the DD. The cluster described in Fig. 4A is separated from the two others, as it has a feature in the LNK domain (particularly at sites 214, 216, and 225) in addition to the DD; therefore, we labeled this cluster as DD-LNK. These sites on the linker are proximal to the DD sites, suggesting that antibodies in this class bind to the nucleocapsid in such a way that they ‘nestle’ between the DD and the LNK, with the LNK adjusting to ‘wrap across’ the antibody. Antibodies in Fig. 4B and C have distinct escape phenotypes in the DD. While phenotypes in Fig. 4B are situated at the interchain interface, those in Fig. 4C are located at the flanks, facing away from their opposite chains; we, therefore, label these clusters as DD-INTERFACE and DD-FLANK, respectively (details on this distinction are in Fig. S6). These two clusters can be distinguished by a region of strong escape at sites 299–306 found in DD-FLANK but not in DD-INTERFACE, whereas DD-INTERFACE contains more high-escape sites around site 340.

Figure 4D shows a cluster containing members with escape phenotypes in the RBD, but this cluster is situated far from the other RBD-based clusters on the PCA plot. We infer that this separation is due to a set of high-escape mutations unique to this cluster, and so we named it RBD-SIGNATURES. Compared to those in other clusters, antibodies within this cluster have a higher rate of being escaped when sites are mutated to cysteine residues. This effect is particularly pronounced in the RBD, with the A156C and L159C carrying especially high escape scores. The R276L mutation (which has the highest negative PC2) also typically carries a very high escape score for members of this cluster. A selection of mutations to bulky tryptophan or tyrosine residues (D128W, D128Y, G170W, T271F, and T271Y) also cause high escape. The heatmaps in Fig. 5 all display clusters located in the top-left quadrant of the PCA plot and contain antibodies with escape phenotypes in the RBD. The cluster described in Fig. 5A contains a mixture of RBD-based phenotypes, with some being located at the β-turn and others across the main body, and so is denoted as RBD-GENERAL (see Fig. S6 for further details). The clusters shown by Fig. 5B and C both contain members with phenotypes located across the main body of the RBD (generally corresponding to sites within the range 133–160). Phenotypes in Fig. 5C are generally larger and more spread out than those in Fig. 5B, which tend to be concentrated at precise surfaces. We, therefore, denote the Fig. 5B cluster as RBD-CBODY (where ‘c’ stands for ‘concentrated’) and the cluster in Fig. 5C as RBD-SBODY (where ‘s’ stands for ‘spread’). The cluster in Fig. 5D is distinguished from the others by having an additional strong escape feature in the CTD, and so is denoted as RBD-CTD. We hypothesize that the flexible CTD loop wraps around the antibodies within this cluster to anchor it in place as it binds to the RBD. Multiple members of the RBD-GENERAL and RBD-CBODY clusters appear to have common features, as suggested by the proximity of these two clusters in the PCA plot.

Fig 5.

Dot heatmaps show nucleocapsid escape scores across sites 2-419 for RBD-GENERAL, RBD-CBODY, RBD-SBODY and RBD-CTD antibody clusters, with matching 3D structures highlighting epitope locations.

Three-dimensional heatmap and structural depiction of escape phenotypes by cluster (Set 2: RBD features). Heatmap and structure formats are as described in Fig. 4. (A) RBD-GENERAL cluster. Members have a variety of epitopes in the RBD, including some with epitopes incorporating the RBD β-turn. (B) RBD-CBODY cluster. Members have epitopes in the main body of the RBD, which are relatively (compared to RBD-SBODY) continuous in sequence space. (C) RBD-SBODY cluster. Members have epitopes in the main body of the RBD, which are relatively (compared to RBD-CBODY) spread out in sequence space. (D) RBD-CTD cluster. Members have epitopes in the RBD with an additional strong feature in the CTD.

The clusters described in Fig. 6A through C are all situated near the center of the PCA plot, with their members having approximately neutral PC1 and PC2 values. All contain members with linear escape phenotypes and are differentiated according to where these phenotypes are located. The cluster depicted in Fig. 6A consists of antibodies with phenotypes in the N-terminal domain, so is named LIN-NTD, while that in Fig. 6B contains antibodies with escape phenotypes in the C-terminal domain, so is named LIN-CTD, and that in Fig. 6C has antibodies with phenotypes in the linker region, so is named LIN-LNK.

The cluster depicted in Fig. 6D is also located in a central horizontal position of the PCA plot, but with more negative PC2 values. Multiple different antibody types appear in this cluster, so we labeled it MIXED. Two members (R001 from Quidel-Ortho and mAb2 from the Clip Flu + COVID HomeTest marketed by Luminostics [ClipHealth]) have strong linear escape phenotypes in the CTD (see heatmaps in Data S1). The latter is situated closer to members of the LIN-CTD cluster, away from the other members of the MIXED cluster, suggesting it has been misclassified. This mAb2 antibody is situated close to 5E2 from Mologic on the PCA plot; Fig. 2 shows that these two antibodies are highly correlated, supporting this misclassification hypothesis. Of the four other members in this cluster, three have linear escape phenotypes within the RBD, and one (HA811-36MB from Phase Scientific) has no obvious phenotype. This cluster, therefore, largely consists of antibodies with linear phenotypes.

Binding affinities (KD values) are provided for each antibody (Table 1) and average values per cluster shown, stratified according to whether the antibody was an IgG (Fig. S7) or an IgM (Fig. S8). On average, IgGs in the RBD-CBODY cluster had the lowest KD values, suggesting that a continuous strip of residues within the body of the RBD creates a particularly strong escape phenotype. Generally, IgGs with linear phenotypes outside either of the structured domains bind with less strength, especially when situated in the CTD. Meanwhile, IgMs within the RBD-SIGNATURES cluster have significantly lower KD values than other IgMs.

Tracking clusters over the pandemic

We tested RAT antibodies from April 2021 through April 2024. Over this period, many companies were updating and developing the antibodies used in their RATs, and it is unclear whether this was to respond to emerging SARS-CoV-2 nucleocapsid mutations or for other reasons, such as test compatibility and overall performance. We, therefore, sought to determine whether such changes corresponded to changes in cluster type over time. In other words, did the development of new antibodies signal any changes regarding the targeting of particular regions of the nucleocapsid?

We show that there was indeed a change in cluster type over time (Fig. 7A). At the beginning of the pandemic, the escape phenotypes described in clusters RBD-SBODY, LIN-NTD, RBD-CBODY, DD-INTERFACE, RBD-GENERAL, and LIN-CTD were those mostly being targeted; furthermore, whereas the first three of these clusters ceased being identified within the first year of our analyses, the latter three persisted through to April 2023. Between months 12 and 23 (March 2022 to February 2023), antibodies in clusters LIN-LNK, DD-FLANK, and RBD-CTD all emerged then disappeared. Finally, at months 25 (April 2023) and 28 (July 2023), respectively, antibodies in clusters DD-LNK and RBD-SIGNATURES emerged. For ease of reference, we refer to these groupings as the first, second, and third waves.

Fig 7.

Heatmap, area chart, and box plots show SARS-CoV-2 antibody cluster prevalence across three waves, the prevalence of particular nucleocapsid mutations between lineages, nucleocapsid escape scores by cluster, and RAT antibody combination frequencies.

Antibody cluster and SARS-CoV-2 lineage changes over time. (A) Cluster identification timecourse heatmap and dendrogram. Over the 37-month period, we calculate the percentage of antibodies received, separated by the clusters the antibodies belong to. We see a change in cluster prevalence over time, approximately corresponding to three ‘waves.’ (B) Plot of SARS-CoV-2 lineages over time (all countries, 01/2020–01/2026). Data are from GISAID (20) and the graph generated with Nextstrain (21). Frequencies are color-coded by clade, with different color hues within each clade denoting lineages. (C) SARS-CoV-2 nucleocapsid mutation prevalences across prominent lineages. Data are from GISAID (20) and the report generated from the outbreak.info resource (22). Lineages were selected according to their frequencies from panel B and availability in outbreak.info. All mutations occurring in nucleocapsid between the Wuhan sequence and these lineages are listed on the horizontal axis. Bracketed numbers denote the number of sequences available and analyzed during report generation. (D) Box plots of the average escape scores (z-normalized) for each nucleocapsid point mutation appearing across lineages, as identified in panel C. Within each plot, boxes are separated by cluster type. The key associates boxes with cluster types. (E) Heatmap of RAT antibody combinations sorted according to their PCA assignments. The heatmap is color-coded by the number of times the cluster combination is identified. For example, a count of three (yellow) indicates three separate tests where the antibodies used belonged to the two clusters being considered.

We compared these data against the most prominent SARS-CoV-2 clades over this period (Fig. 7B). Assuming an approximately 3–4-month delay as companies adapted to new strains of the virus (23), we see that the emergence of the second wave of clusters corresponds to the BA (Omicron) strain, which began appearing in late 2021 (24, 25). These new strains typically included mutations in the nucleocapsid (P13L, DEL31-33, R203K, G204R, and S413R) (20) not seen in previous clades (Fig. 7C). These results do not necessarily imply that companies were deliberately changing their antibodies in response to SARS-CoV-2 nucleocapsid mutations, as other market forces would have been at play; instead, we are simply noting that an association exists.

We collected available data on nucleocapsid mutations across different strains (Fig. 7C) from outbreak.info (22). We then focused on the escape scores for each of these mutations (omitting DEL31-33, as we did not have DMS data for this alteration since this was not a point mutation). A higher score denotes a stronger ability to escape binding (the reverse of what is desired for a successful RAT), and results are stratified by antibody cluster type (Fig. 7D).

We regard the escape scores for D3L and P13L as contaminated by our assay set-up, where mutations at these sites can interfere with Myc signaling. Certain residues in the NTD are potentially blocked from binding to the anti-nucleocapsid antibody due to steric clash with the anti-Myc antibody. Nevertheless, despite this important caveat, we note that, on average, the antibodies in the DD-LNK and RBD-SIGNATURES clusters have particularly high escape from nucleocapsid bearing the D3L mutation. This mutation arose in the Alpha strain but did not persist into future strains, ameliorating potential issues with selection of the mAbs for RATs as Alpha quickly drifts with the emergence of new strains. Regarding the P13L mutation, it appears that escape against antibodies in the LIN-NTD and RBD-SIGNATURES clusters is also high. The P13L variant is present in very small quantities of the Alpha and Delta sequences that were analyzed, but became almost ubiquitous when Omicron emerged and has continued to be prominent in subsequent strains. Therefore, according to these data (which do not have the benefit of clinical comparison due to the limitations of our study), there is a risk that antibodies belonging to these two clusters will struggle to detect the presence of virus effectively.

Three members of the DD-INTERFACE cluster (F5mod and C7 from AvantGen and 2F4 from GenBody) and one from the DD-FLANK cluster (mAb-OTI-1 from OraSure) all had large escape scores from the D63G mutation. This surprised us because all of these antibodies have escape phenotypes in the DD, whereas site 63 is in the RBD. Interestingly, for each of these high-scoring antibodies, it appears that mutation must be to glycine, with alternative D63 mutations not eliciting a strong escape response. This D63G mutation appears in the Delta strain but thereafter disappears. Structural studies have shown that the D63G mutation increases RNA binding affinity, which is achieved by the loss of a salt-bridge between the G63 residue and R89, therefore modulating the flexibility of the latter (26).

One member of MIXED (HA811-36MB from Phase Scientific, which has no obvious epitope) has very high escape from the D377Y mutation, which arose with Delta and has persisted at low levels since then (Fig. 7C). Also, one member of DD-FLANK [MAD4904-P-2209(B) from Roche/SD/Bionote] has very high escape from the S413R mutation, a defining convergent mutation for Omicron. This site is far removed from the DD epitope for this antibody, so the site and epitope, therefore, appear to be unrelated.

In the final 10 months of the study period (August 2023–April 2024), over half of the antibodies belonged to the RBD-SIGNATURES cluster, which was the most recent to appear (July 2023). This change coincided with an increased use of IgMs within the RATs. Antibodies in RBD-SIGNATURES may have the highest susceptibility to escape from the late 2023 Q229K variant (Fig. 7D), which is concerning given the preponderance of antibodies belonging to this cluster currently used in RATs. As discussed earlier, it appears that escape against antibodies belonging to the RBD-SIGNATURES cluster is high for the P13L mutation, although those particular results should be considered more cautiously. Interestingly, the third wave clusters (DD-LNK and RBD-SIGNATURES) also perform poorest against the S235F mutation, although that mutation only appeared in the Alpha strain before disappearing.

Introduction of IgMs

As the pandemic progressed, research into IgMs as possible diagnostic antibodies in RATs expanded (27). Despite typically being more complicated to produce, the higher valency of IgMs increases avidity, conceivably allowing for detection of antigen at lower concentrations, thus improving RAT performance. Starting in April 2022, we tested 10 IgMs and investigated whether IgMs would produce distinct escape patterns compared to IgGs. We found that these IgMs do not form a unique cluster but instead cluster with IgGs. Six of the IgMs were grouped into the RBD-SIGNATURES cluster, which contains nine members in total. The three non-IgMs were the Capture and Labeling antibodies from iHealth and N18 from Abbott. Interestingly, these members of the cluster are closest to the center of the plot (i.e., they have the least negative PC2 values), as shown in Fig. 3. The IgMs in this cluster, by contrast, all have very negative PC2 values; highly negative PC2 values, therefore, seem to be associated with the presence of IgMs rather than IgGs within this cluster. Of the further four IgMs, two were identified as belonging to the DD-FLANK cluster and two to the RBD-GENERAL cluster.

Combinations of antibodies in RATs

Most RATs rely on two antibodies recognizing orthogonal epitopes: a mobile labeling antibody that binds to the antigen when the sample is applied and an immobilized capture antibody at the test line that binds the antigen-label complex as it flows past.

Of the 33 tests for which we had defined pairing, 20 contained capture and labeling antibodies recognizing non-overlapping epitopes in the same domain (Data S1), 15 of which were in the same cluster, and five of which were grouped into different clusters but still hit the same structured domain (four RBD-RBD and one DD-DD combination) (Fig. 7E; Table S3). Therefore, with 61% of antibody combinations hitting the same domain across all RATs, we concluded it must be advantageous for antibodies to have epitopes close to one another despite the necessity to avoid epitope overlap. We recognize that it is beneficial for both antibodies in a RAT to target more conserved and stable domains, plus when the same domain is targeted, this ensures that both antibodies interact with a consistent conformation.

These results clearly show that clusters do not precisely define epitopes but instead give a range of epitopes with common features, such as overlapping or proximal escape sites, and conformational or linear structure.

DISCUSSION

A major public health goal during a pandemic is to efficiently and reliably identify individuals infected with the pathogen. Rapid antigen tests contribute significantly to these efforts; however, test sensitivity can be compromised when new variants emerge. During the SARS-CoV-2 pandemic, the nucleocapsid protein was the main target of diagnostic RATs, and so when this antigen mutated as new strains of the virus emerged, it became important to determine whether the diagnostic antibodies being used were at risk of being escaped.

Using deep mutational scanning, we can determine in a high-throughput manner which nucleocapsid mutations result in escape from antibody detection. Typically, variants of high escape are grouped into distinct patterns. These regions of high escape can either be due to nucleocapsid destabilization, direct antigen-antibody interaction (i.e., epitopes), or allosteric effects. Through understanding these epitopic and allosteric features, we can predict which diagnostic antibodies should be flagged for follow-up testing when new nucleocapsid mutants emerge.

Over the course of the pandemic, we analyzed 82 diagnostic antibodies, generating a large variant escape resource that can be used to infer epitopes, inform test design, improve surveillance, and identify potential risk. This provided a wealth of data with which to derive general principles governing antibody-antigen binding and determine whether RATs were sufficiently diverse to prevent a single nucleocapsid mutation from compromising most tests on the market. Using an exploratory threshold of ρ ≥ 0.65, we identified 24 antibodies as potentially identical or closely related across RATs. Our DMS technique, therefore, provides a powerful method for identifying related or shared antibody sources used in RATs without requiring mass spectrometry.

The first principal component (PC1) of the antibody escape scores from a PCA differentiated antibodies with strong escape in the dimerization domain (positive) and RNA-binding domain (negative). Antibodies with strong escape across linear sites or with weak escape typically had neutral PC1 values. Meanwhile, antibodies with strongly negative PC2 values were characterized by strong escape scores at a selection of seemingly random signature sites across the nucleocapsid.

The antibodies clustered into 12 separate groups based on escape score patterns. Though such groups do not directly give epitopes, they suggest antibodies with similar binding and allosteric features. Three of these groups were characterized by strong escape across linear sites outside the two structured domains; three groups had strong escape in the DD; and five groups exhibited strong escape in the RBD. We note that favored cluster types shifted over the course of the pandemic, with a particularly prominent switch following the emergence of Omicron.

A limitation of this study was our lack of access to RAT sensitivity and specificity data, thus preventing us from correlating our escape score classifications with test performance. Nevertheless, we foresee our results being useful to clinical researchers with access to such data and who wish to identify factors contributing to good or poor test performance. If such researchers know the antibodies used in the RAT they studied, and these coincide with those we have investigated, they can identify variants prone to high escape, which may become relevant as SARS-CoV-2 continues to evolve.

A further limitation is the omission of multi-point mutations from our analysis, which is a significant consideration, as nucleocapsid mutations often accumulate between virus strains. Our DMS libraries did contain a limited volume of multi-point mutations, with double-point mutations comprising ~1% of the total. Higher-order mutations were present at an even lower frequency. Creating libraries with higher coverage of multi-point mutations for investigation with DMS presents numerous practical challenges, such as integration with an appropriate display platform, efficiency of selection, and cost of sequencing. It may be possible to use machine learning to combine single-site DMS data sets with co-evolutionary information to make accurate multi-site escape predictions, although this technology is still developing.

Our study does not explicitly evaluate post-translational modifications (PTMs) despite SARS-CoV-2 nucleocapsid possessing phosphorylation and glycosylation sites (28, 29). We aimed to mitigate this limitation by using mammalian cells to ensure that any PTMs were close to physiological conditions.

Pandemic preparedness is now a public health imperative. We believe our DMS procedure could be adapted to investigate alternative viruses of concern, and we have already begun preparing H3N2 Influenza A libraries for a future study. If our DMS results can be quickly integrated with diagnostic test performance data, we believe antibodies could be selected more efficiently for use in RATs during future pandemics. Antibody development could be guided to limit susceptibility to escape by single point mutations by ensuring the availability of antibodies targeting diverse regions of a given antigen.

MATERIALS AND METHODS

Cells and viruses

We used “Viral Production Cells” (a HEK293 subline) from Thermo Fisher’s LV-MAX Lentiviral Production System for all surface-display experiments and cultured cells in LV-MAX Production medium.

Lentiviral construct and library design

Our lentiviral system and method of library design were previously introduced in Frank et al. (10). Briefly, we inserted a codon-optimized Wuhan SARS-CoV-2 nucleocapsid protein sequence (Uniprot ID: P0DTC9) into a lentiviral construct (Fig. 1A). With this design, we ordered a site-saturation library (TwistBioscience) containing almost all possible single point mutations at sites 2–419 of the nucleocapsid and added 15-nucleotide barcodes by PCR (five amplification cycles). Barcodes were linked to variants using PacBio library sequencing. Gibson assembly and packaging vectors (psPAX2 and pMD2G) were used to clone and package these libraries into lentivirus. Stable cell lines were generated using a GFP reporter; by infecting our Viral Production (HEK293) Cells at a multiplicity of infection (MOI) of 0.1, we ensured that >90% of our cells were infected with just one viral particle. GFP+/Myc+ cells were then sorted and expanded for screening against the diagnostic antibodies.

Flow cytometry to characterize antibody binding

As detailed in Frank et al. (10), we first measured antibody binding to wild-type nucleocapsid to derive an optimal concentration at which to conduct our fluorescence-activated cell sorting assays with the nucleocapsid library. We measured binding between our given antibody and a cell line stably expressing wild-type Wuhan nucleocapsid on the cell surface over a fivefold dilution series. Following washing and incubation, cells were stained with labeled anti-Myc and PE-labeled secondary antibodies, and data were collected with either a FACSCanto (BD) or Northern Lights (Cytek) instrument. We then obtained binding curves and KD values.

Fluorescence-activated cell sorting to determine escape mutations

For each assay, we incubated cells from our GFP+/Myc+ stable cell library against our antibody of interest. Following washing, we then stained these cells with suitable host-specific anti-IgG/anti-IgM and anti-Myc antibodies. Fluorescence-activated cell sorting was then performed using either a FACS Aria II (BD) at the Emory Flow Cytometry Core or a FACS Melody (BD). Diagonal gates on antibody vs anti-Myc signal plots were drawn to select for cells with the lowest 15% of antibody signal while accounting for cell surface expression (Fig. 1C). Approximately 1 × 106 cells were selected in this way per assay, plus a reference population of 5 × 106 was submitted each time escape populations were to be processed for sequencing.

Next-generation sequencing of cell populations

A detailed procedure is given in Frank et al. (10). Briefly, post-sort and reference cell samples were each washed, and then RNA was extracted. The extracted RNA was reverse-transcribed to cDNA using a specific primer designed to anneal immediately downstream of the barcode. Barcodes were then amplified using PCR (eight rounds), and the products were purified. Following preparation with the Illumina NexteraXT DNA Library Preparation Kit and validation, we then sequenced our samples on an Illumina NovaSeq 6000 at PE100 or PE26 × 91. Generally, we read to a depth of 10 million reads for sorted samples and 50 million reads for each reference sample.

DMS data analysis and escape score calculation

We wrote custom scripts and adapted pipelines from others to analyze our sequences (14, 15). We started with simple custom Python and R scripts, but now have a robust pipeline available from our GitHub (https://github.com/Ortlund-Laboratory/SARS-CoV-2_Epitope_Mapping/tree/main/Workflow_to_Determine_Escape_Scores). Briefly, we make use of a Snakemake pipeline consisting primarily of Jupyter notebooks, which we largely adapted from Greaney et al. (15). We parse barcodes from our sequences, and then map these to nucleocapsid variants using the lookup table we generated with PacBio sequencing. We summed barcode counts associated with the same variant to give an overall count for each variant. For each sample, we calculated abundances of the variant, ni, against the entire population, N, where N is the sum of all counts (N = ∑ini). We calculated escape scores per variant, Ei, by taking the ratio of these abundances between the sorted and reference populations:

Ei=(ni,sort∑ini,sort)(ni,ref∑ini,ref) 

A series of filtration steps are performed to control for outliers and conditions (such as when abundance values in the reference population are unusually low, giving artificially high escape scores), as detailed in our notebooks available at https://github.com/Ortlund-Laboratory/SARS-CoV-2_Epitope_Mapping/tree/main. Per-site scores are calculated by taking the average of Ei values at a given site, irrespective of the amino acid mutated to. For certain analyses (see below), z-normalization of the suite of Ei scores was necessary to allow for comparison between antibodies.

Correlation of escape data

We created the correlogram (Fig. 2) by collecting the escape data for the antibodies we had studied into a single dataframe. Eighty-two of the 85 antibodies we had studied were included, with MM05 from ANP Technologies and C518 and C524 from Qorvo Biotechnologies removed because we could no longer access those data sets. Sites 51–53, 71–73, 84–87, 110–116, 130, and 132 were removed from our analysis because these are located in the central portion of the nucleocapsid RBD and have a significant effect on protein surface-expressed folding fitness when mutated, rather than influencing direct binding between the nucleocapsid and the antibody. Likewise, variants in the NTD associated with Myc anti-Myc binding ‘stickiness,’ including all variants at sites 36–41, the V, L, I, M, F, Y, and W variants for sites 2–14, 18–30, and 32–35, plus variants C, S, and T at site 5, and variants D and E at sites 32–35, were removed. Pearson coefficients were then calculated between all antibodies. See correlogram_and_PCA.R in our GitHub (https://github.com/Ortlund-Laboratory/SARS-CoV-2_Epitope_Mapping/tree/main/Correlogram_and_PCA) for the full workflow.

Principal component analysis (PCA)

Prior to PCA, it was necessary to z-normalize all data sets. As with the correlation analysis, we removed selected variants to adjust for Myc anti-Myc binding ‘stickiness’ and nucleocapsid folding fitness. To conduct the PCA, we used the prcomp function in R. To determine the optimal cluster number, k, we tried the Elbow (30) and Silhouette (31) methods. These methods gave very different results for k (7 and 19), as shown in Fig. S3 and S4, respectively. We, therefore, included the Gap (32) method (Fig. S5) to potentially act as a tie-breaker. This method gave k = 12, which we decided to proceed with for two reasons: it was approximately mid-way between the other two methods; and the Gap method is seen as more statistically rigorous than the other methods for k determination. See correlogram_and_PCA.R in our GitHub (https://github.com/Ortlund-Laboratory/SARS-CoV-2_Epitope_Mapping/tree/main/Correlogram_and_PCA) for the full workflow.

Determination of top escape sites per cluster

The following results were also obtained using our correlogram_and_PCA.R script: as in other analyses, we removed selected variants by adjusting for Myc anti-Myc binding ‘stickiness’ and nucleocapsid folding fitness. Per-site escape scores for each antibody were determined by calculating the average across all per-variant escapes scores at each site, irrespective of the amino acid being converted to. Per-cluster per-site escape scores were then calculated by averaging these per-site escape scores by all the antibodies present in the cluster. In this way, sites of particular consequence for the various clusters emerge due to their high escape values. For the structural representations in Fig. 4 to 6, sites of ‘strong escape’ (color-coded purple) are identified as those which have a score at least 2 standard deviations above the mean across the cluster considered as a whole plus are at least 0.5 standard deviations above the mean for all individual antibodies within the cluster; sites of ‘moderate escape’ (color-coded pink) only meet the criterion that the site has a score at least 2 standard deviations above the mean across the cluster as a whole.

At the time of publication, there was no published structure of the full SARS-CoV-2 nucleocapsid structure whether in monomeric or dimeric form; therefore, to create a 3D structure depicting the nucleocapsid dimer, as shown in Fig. 4 to 6, we used AlphaFold 3 (33) and the Wuhan sequence (Uniprot ID: P0DTC9).

Tracking clusters and lineages over time

Of the 82 antibodies we had organized into clusters, we knew the date of receipt for 73. For the nine other antibodies, this information was missing, as we no longer had access to the uptake forms, which we used to determine the date of receipt. These nine missing antibodies were C706 and FPZ from Osang, COVID-19-PS-MAB115 and COVID-19-PS-MAB117 from Watmind, and rSN-25, rSN-69, rSN-122, rSN-327, and SNS-25 from Siemens. With the remaining 73 antibodies, we calculated the percentage of antibodies received per cluster for the 37-month period (04/2021–04/2024) in which we conducted the analysis (Fig. 7A). Using pheatmap (34) in R, we created a dendrogram to visualize hierarchical clustering.

To determine the SARS-CoV-2 lineages that were circulating over this period, we used the online Nextstrain resource (21), which in turn uses data from GISAID (20). Using the date range 01/2020–01/2026 and considering all countries, we generated a Frequencies by Clade graph at https://nextstrain.org/ncov/open/global/all-time?tl=none and adapted the result for Fig. 7B.

The lineages labeled in Fig. 7B are those that circulated prominently over the course of the pandemic. To determine whether any of these lineages carry mutations in the nucleocapsid, we generated a report using the outbreak.info online resource (22), which also uses data gathered from GISAID (20). Not all of these relevant lineages were included in outbreak.info; we used as many as were available. We used the output from this report to generate Fig. 7C.

We then cross-referenced the point mutations to nucleocapsid incurred between lineages against the escape scores we generated for these exact mutations. We grouped these scores by the clusters the antibodies belonged to and displayed the results as box plots (Fig. 7D).

ACKNOWLEDGMENTS

We gratefully acknowledge all data contributors (i.e., the authors and their originating laboratories responsible for obtaining the specimens) and their submitting laboratories for generating the genetic sequence and metadata and sharing via the GISAID Initiative, on which part of this research is based. Some FACS experiments were performed by the Flow Cytometry Core Facility of the Emory University School of Medicine. We thank Julie Sullivan from the Atlanta Center for Microsystems-Engineered Point-of-Care Technologies (ACME POCT), Emily Kennedy from OOMVELT LLC, and Eric Lai from PharmaDx LLC for coordination and implementation of RADx-based research.

Research was supported by the National Institute of Biomedical Imaging and Bioengineering of the National Institutes of Health (under award numbers 75N92019P00328, U54EB015408, and U54EB027690) as part of the Rapid Acceleration of Diagnostics (RADx) initiative. Next-generation sequencing services were provided by the Emory NPRC Genomics Core in part supported by NIH P51 OD011132. Sequencing data were acquired on an Illumina NovaSeq 6000 funded by NIH S10 OD026799.

Contributor Information

Eric A. Ortlund, Email: eortlun@emory.edu.

Rebecca Ellis Dutch, University of Kentucky College of Medicine, Lexington, Kentucky, USA.

DATA AVAILABILITY

Our DMS escape score data sets, code, workflows, and filtration steps are provided in our openly accessible group GitHub repository (https://github.com/Ortlund-Laboratory/SARS-CoV-2_Epitope_Mapping) and are stored permanently on Zenodo at https://doi.org/10.5281/zenodo.20801728. Sequencing fastq files have been uploaded to the National Center for Biotechnology Information (NCBI) and are available within BioProject PRJNA1526356.

SUPPLEMENTAL MATERIAL

The following material is available online at https://doi.org/10.1128/jvi.01001-26.

Supplemental material. jvi.01001-26-s0001.docx.

Fig. S1 to S8; Tables S1 to S3.

jvi.01001-26-s0001.docx (2.9MB, docx)
DOI: 10.1128/jvi.01001-26.SuF1
Data. jvi.01001-26-s0002.pdf.

Individual antibody epitopes.

jvi.01001-26-s0002.pdf (10.2MB, pdf)
DOI: 10.1128/jvi.01001-26.SuF2
Data. jvi.01001-26-s0003.pdf.

Individual antibody epitopes.

jvi.01001-26-s0003.pdf (10.5MB, pdf)
DOI: 10.1128/jvi.01001-26.SuF3

ASM does not own the copyrights to Supplemental Material that may be linked to, or accessed through, an article. The authors have granted ASM a non-exclusive, world-wide license to publish the Supplemental Material files. Please contact the corresponding author directly for reuse.

REFERENCES

  • 1. Smith-Jeffcoat SE, Mellis AM, Grijalva CG, Talbot HK, Schmitz J, Lutrick K, Ellingson KD, Stockwell MS, McLaren SH, Nguyen HQ, Rao S, Asturias EJ, Davis-Gardner ME, Suthar MS, Kirking HL, RVTN-Sentinel Study Group . 2024. SARS-CoV-2 viral shedding and rapid antigen test performance — respiratory virus transmission network, November 2022–May 2023. MMWR Morb Mortal Wkly Rep 73:365–371. doi: 10.15585/mmwr.mm7316a2 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2. Krenn F, Dächert C, Badell I, Lupoli G, Öztan GN, Feng T, Schneider N, Huber M, Both H, Späth PM, Muenchhoff M, Graf A, Krebs S, Blum H, Durner J, Czibere L, Kaderali L, Keppler OT, Baldauf H-M, Osterman A. 2023. Ten rapid antigen tests for SARS-CoV-2 widely differ in their ability to detect Omicron-BA.4 and -BA.5. Med Microbiol Immunol 212:323–337. doi: 10.1007/s00430-023-00775-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3. Høeg TB, Prasad V. 2023. Rapid antigen testing for COVID-19: decreasing diagnostic reliability, potential detrimental effects and a lack of evidence to support continued public funding of community-based testing. Public Health Pract (Oxf) 6:100451. doi: 10.1016/j.puhip.2023.100451 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4. Woloshin S, Patel N, Kesselheim AS. 2020. False negative tests for SARS-CoV-2 infection - challenges and implications. N Engl J Med 383:e38. doi: 10.1056/NEJMp2015897 [DOI] [PubMed] [Google Scholar]
  • 5. Creager R, Blackwood J, Pribyl T, Bassit L, Rao A, Greenleaf M, Frank F, Lam W, Ortlund E, Schinazi R, Greninger A, Cirrincione M, Gort D, Kennedy E, Samuta A, Shaw M, Walsh B, Lai E. 2021. RADx variant task force program for assessing the impact of variants on SARS-CoV-2 molecular and antigen tests. IEEE Open J Eng Med Biol 2:286–290. doi: 10.1109/OJEMB.2021.3116490 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6. Rao A, Westbrook A, Bassit L, Parsons R, Fitts E, Greenleaf M, McLendon K, Sullivan JA, O’Sick W, Baugh T, Bowers HB, Frank F, Wang E, Le M, Frediani J, Roychoudhury P, Greninger AL, Jerris R, Pollock NR, Ortlund EA, Roback JD, Lam WA, Piantadosi A. 2023. Sensitivity of rapid antigen tests against SARS-CoV-2 Omicron and Delta variants. J Clin Microbiol 61:e0013823. doi: 10.1128/jcm.00138-23 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7. Bassit L, Bowers HB, Greenleaf M, Sabino C, Lai E, Yu G, Piantadosi A, Wang E, O’Sick W, McLendon K, Sullivan JA, Schinazi RF, Damhorst GL, Lam W, Rao A. 2024. Protocol for the creation and characterization of SARS-CoV-2 variant testing panels using remnant clinical samples for diagnostic assay testing. STAR Protoc 5:103146. doi: 10.1016/j.xpro.2024.103146 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8. Byrne RL, Owen RS, Aljayyoussi G, Greenland-Bews C, Kontogianni K, Somasundaran A, Wooding D, Williams CT, LSTM Diagnostics Group, Falcon Steering Group, de Vos M, Body R, Adams ER, Escadafal C, Edwards T, Cubas-Atienzar AI. 2025. Comparison of the analytical and clinical sensitivities of 34 rapid antigen tests with prevalent SARS-CoV-2 variants of concern during the COVID-19 pandemic in the UK. Microbiol Spectr 13:e0074925. doi: 10.1128/spectrum.00749-25 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9. Bown A, Sweed A, Catton M, Nelthorpe-Cowne J, Conti-Frith H, Elderfield R, Terrey J, Abib H, Lui C, Fisher E, Bewley KR, Coombes NS, Robinson D, Agrawal S, Hallis B, Blandford E, Fowler T, Williamson DA, Vipond R. 2024. Detection of SARS-CoV-2 BA.2.86 by lateral flow devices. Int J Infect Dis 139:168–170. doi: 10.1016/j.ijid.2023.12.010 [DOI] [PubMed] [Google Scholar]
  • 10. Frank F, Keen MM, Rao A, Bassit L, Liu X, Bowers HB, Patel AB, Cato ML, Sullivan JA, Greenleaf M, Piantadosi A, Lam WA, Hudson WH, Ortlund EA. 2022. Deep mutational scanning identifies SARS-CoV-2 nucleocapsid escape mutations of currently available rapid antigen tests. Cell 185:3603–3616. doi: 10.1016/j.cell.2022.08.010 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11. Araya CL, Fowler DM. 2011. Deep mutational scanning: assessing protein function on a massive scale. Trends Biotechnol 29:435–442. doi: 10.1016/j.tibtech.2011.04.003 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12. Hietpas RT, Jensen JD, Bolon DNA. 2011. Experimental illumination of a fitness landscape. Proc Natl Acad Sci USA 108:7896–7901. doi: 10.1073/pnas.1016024108 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13. Keen MM, Keith AD, Ortlund EA. 2025. Epitope mapping via in vitro deep mutational scanning methods and its applications. J Biol Chem 301:108072. doi: 10.1016/j.jbc.2024.108072 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14. Starr TN, Greaney AJ, Hilton SK, Ellis D, Crawford KHD, Dingens AS, Navarro MJ, Bowen JE, Tortorici MA, Walls AC, King NP, Veesler D, Bloom JD. 2020. Deep mutational scanning of SARS-CoV-2 receptor binding domain reveals constraints on folding and ACE2 binding. Cell 182:1295–1310. doi: 10.1016/j.cell.2020.08.012 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15. Greaney AJ, Starr TN, Gilchuk P, Zost SJ, Binshtein E, Loes AN, Hilton SK, Huddleston J, Eguia R, Crawford KHD, Dingens AS, Nargi RS, Sutton RE, Suryadevara N, Rothlauf PW, Liu Z, Whelan SPJ, Carnahan RH, Crowe JE Jr, Bloom JD. 2021. Complete mapping of mutations to the SARS-CoV-2 spike receptor-binding domain that escape antibody recognition. Cell Host Microbe 29:44–57. doi: 10.1016/j.chom.2020.11.007 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16. Hanning KR, Minot M, Warrender AK, Kelton W, Reddy ST. 2022. Deep mutational scanning for therapeutic antibody engineering. Trends Pharmacol Sci 43:123–135. doi: 10.1016/j.tips.2021.11.010 [DOI] [PubMed] [Google Scholar]
  • 17. Rubin AF, Gelman H, Lucas N, Bajjalieh SM, Papenfuss AT, Speed TP, Fowler DM. 2017. A statistical framework for analyzing deep mutational scanning data. Genome Biol 18:150. doi: 10.1186/s13059-017-1272-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18. Rubin AF, Gelman H, Lucas N, Bajjalieh SM, Papenfuss AT, Speed TP, Fowler DM. 2018. Correction to: a statistical framework for analyzing deep mutational scanning data. Genome Biol 19:17. doi: 10.1186/s13059-018-1391-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19. Freudenberg J, Rao J, Howard MK, Macdonald C, Greenwald NF, Coyote-Maestas W, Pimentel H. 2026. Accurate variant effect estimation in FACS-based deep mutational scanning data with Lilace. Genome Biol 27:48. doi: 10.1186/s13059-026-03934-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20. Khare S, Gurry C, Freitas L, Schultz MB, Bach G, Diallo A, Akite N, Ho J, Lee RT, Yeo W, Curation Team GC, Maurer-Stroh S. 2021. GISAID’s role in pandemic response. China CDC Wkly 3:1049–1051. doi: 10.46234/ccdcw2021.255 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21. Hadfield J, Megill C, Bell SM, Huddleston J, Potter B, Callender C, Sagulenko P, Bedford T, Neher RA. 2018. Nextstrain: real-time tracking of pathogen evolution. Bioinformatics 34:4121–4123. doi: 10.1093/bioinformatics/bty407 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22. Gangavarapu K, Latif AA, Mullen JL, Alkuzweny M, Hufbauer E, Tsueng G, Haag E, Zeller M, Aceves CM, Zaiets K, Cano M, Zhou X, Qian Z, Sattler R, Matteson NL, Levy JI, Lee RTC, Freitas L, Maurer-Stroh S, GISAID Core and Curation Team, Suchard MA, Wu C, Su AI, Andersen KG, Hughes LD. 2023. Outbreak.info genomic reports: scalable and dynamic surveillance of SARS-CoV-2 variants and mutations. Nat Methods 20:512–522. doi: 10.1038/s41592-023-01769-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23. Nguyen NNT, McCarthy C, Lantigua D, Camci-Unal G. 2020. Development of diagnostic tests for detection of SARS-CoV-2. Diagnostics (Basel) 10:905. doi: 10.3390/diagnostics10110905 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24. Callaway E. 2021. Heavily mutated Omicron variant puts scientists on alert. Nature 600:21–21. doi: 10.1038/d41586-021-03552-w [DOI] [PubMed] [Google Scholar]
  • 25. Viana R, Moyo S, Amoako DG, Tegally H, Scheepers C, Althaus CL, Anyaneji UJ, Bester PA, Boni MF, Chand M, et al. 2022. Rapid epidemic expansion of the SARS-CoV-2 Omicron variant in southern Africa. Nature 603:679–686. doi: 10.1038/s41586-022-04411-y [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26. Dhamotharan K, Korn SM, Wacker A, Becker MA, Günther S, Schwalbe H, Schlundt A. 2024. A core network in the SARS-CoV-2 nucleocapsid NTD mediates structural integrity and selective RNA-binding. Nat Commun 15:10656. doi: 10.1038/s41467-024-55024-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27. Huang Y, Pu Y, Weng Y, Wu Y, He Q, Litchev S, Zhao L, Ding H, Lai Y, Li J, Zheng X, Chen J, Xiong X, Lu S, Gao F, Gao M, Kong Q. 2024. Development and diagnosis performance of IgM-based rapid antigen test for early detection of SARS-CoV-2 infection in a large cohort of suspected COVID-19 cases — USA, Poland, and Sweden, 2021–2022. China CDC Wkly 6:953–961. doi: 10.46234/ccdcw2024.199 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28. Bouhaddou M, Memon D, Meyer B, White KM, Rezelj VV, Correa Marrero M, Polacco BJ, Melnyk JE, Ulferts S, Kaake RM, et al. 2020. The global phosphorylation landscape of SARS-CoV-2 infection. Cell 182:685–712. doi: 10.1016/j.cell.2020.06.034 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29. Supekar NT, Shajahan A, Gleinich AS, Rouhani DS, Heiss C, Chapla DG, Moremen KW, Azadi P. 2021. Variable posttranslational modifications of severe acute respiratory syndrome coronavirus 2 nucleocapsid protein. Glycobiology 31:1080–1092. doi: 10.1093/glycob/cwab044 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30. Thorndike RL. 1953. Who belongs in the family? Psychometrika 18:267–276. doi: 10.1007/BF02289263 [DOI] [Google Scholar]
  • 31. Rousseeuw PJ. 1987. Silhouettes: A graphical aid to the interpretation and validation of cluster analysis. J Comput Appl Math 20:53–65. doi: 10.1016/0377-0427(87)90125-7 [DOI] [Google Scholar]
  • 32. Tibshirani R, Walther G, Hastie T. 2001. Estimating the number of clusters in a data set via the gap statistic. J R Stat Soc Series B Stat Methodol 63:411–423. doi: 10.1111/1467-9868.00293 [DOI] [Google Scholar]
  • 33. Abramson J, Adler J, Dunger J, Evans R, Green T, Pritzel A, Ronneberger O, Willmore L, Ballard AJ, Bambrick J, et al. 2024. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature 630:493–500. doi: 10.1038/s41586-024-07487-w [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34. Kolde R. 2025. pheatmap: pretty heatmaps. R package version 1.0.13. https://github.com/raivokolde/pheatmap.

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplemental material. jvi.01001-26-s0001.docx.

Fig. S1 to S8; Tables S1 to S3.

jvi.01001-26-s0001.docx (2.9MB, docx)
DOI: 10.1128/jvi.01001-26.SuF1
Data. jvi.01001-26-s0002.pdf.

Individual antibody epitopes.

jvi.01001-26-s0002.pdf (10.2MB, pdf)
DOI: 10.1128/jvi.01001-26.SuF2
Data. jvi.01001-26-s0003.pdf.

Individual antibody epitopes.

jvi.01001-26-s0003.pdf (10.5MB, pdf)
DOI: 10.1128/jvi.01001-26.SuF3

Data Availability Statement

Our DMS escape score data sets, code, workflows, and filtration steps are provided in our openly accessible group GitHub repository (https://github.com/Ortlund-Laboratory/SARS-CoV-2_Epitope_Mapping) and are stored permanently on Zenodo at https://doi.org/10.5281/zenodo.20801728. Sequencing fastq files have been uploaded to the National Center for Biotechnology Information (NCBI) and are available within BioProject PRJNA1526356.


Articles from Journal of Virology are provided here courtesy of American Society for Microbiology (ASM)

RESOURCES