Orthogonal SARS-CoV-2 Serological Assays Enable Surveillance of Low-Prevalence Communities and Reveal Durable Humoral Immunity

Abstract

We conducted a serological study to define correlates of immunity against SARS-CoV-2. Compared to those with mild coronavirus disease 2019 (COVID-19) cases, individuals with severe disease exhibited elevated virus-neutralizing titers and antibodies against the nucleocapsid (N) and the receptor binding domain (RBD) of the spike protein. Age and sex played lesser roles. All cases, including asymptomatic individuals, seroconverted by 2 weeks after PCR confirmation. Spike RBD and S2 and neutralizing antibodies remained detectable through 5-7 months after onset, whereas α-N titers diminished. Testing 5,882 members of the local community revealed only 1 sample with seroreactivity to both RBD and S2 that lacked neutralizing antibodies. This fidelity could not be achieved with either RBD or S2 alone. Thus, inclusion of multiple independent assays improved the accuracy of antibody tests in low-seroprevalence communities and revealed differences in antibody kinetics depending on the antigen. We conclude that neutralizing antibodies are stably produced for at least 5-7 months after SARS-CoV-2 infection.

Keywords: COVID-19; S2 domain; SARS-CoV-2; antibodies; neutralization; nucleocapsid protein; orthogonal serological tests; receptor binding domain; serological test; serology; spike protein.

Graphical abstract

Figure 1

Assessment of RBD-Based Sensitivity and Specificity in Serological Testing (A) Serum samples (153) from healthy controls and confirmed COVID-19 cases were assessed for RBD reactivity by ELISA and neutralization of live SARS-CoV-2. PRNT₉₀ values were determined as the last dilution by which 90% neutralization occurred. Antibody titers were quantified for RBD by quantifying area under the curve (AUC) across a serial dilution curve. R values were calculated by Pearson correlation test. (B) Pre-2020 negative-control samples (352) and 30 samples from SARS-CoV-2-exposed individuals were screened by ELISA at a single 1:40 dilution against RBD. The blue region indicates overlap of OD values between negative- and positive-control samples. % indicates frequency of negative-control values in this range. Experiments were repeated 3 times. (C) RBD seroreactivity was quantified based on time elapsed from PCR+ confirmation of SARS-CoV-2 infection. (D) Individuals recruited from the community (n = 5,882) were screened for seroreactivity to RBD. (E) PRNT₉₀ analysis from community drawn samples that displayed indeterminate or positive RBD seroreactivity. Samples that neutralized 90% of virions at least at a 1:20 dilution were considered positive. Experiments were repeated at least once. Error bars in (B), (D), and (E) depict mean values of datasets ± standard error of the mean (SEM) and were calculated in GraphPad Prism.

Figure 2

Assessment of S2 and N Antibodies as Secondary Confirmations of Seropositivity (A) Correlations of neutralization and N-specific IgG ELISA titers across 156 serum samples from healthy controls and COVID-19 cases. (B) A sample set of 32 pre-pandemic controls and 30 PCR+ SARS-CoV-2 samples were assayed for seroreactivity to N protein. Blue shaded region indicates overlap between negative and positive controls. Frequency of negative controls in this range is shown. (C) Correlations of neutralization and S2-specific IgG ELISA titers across 151 serum samples from healthy controls and COVID-19 cases. (D) Pre-pandemic negative-control samples (272) were screened for seroreactivity against S2 and compared to 30 PCR-confirmed SARS-CoV-2-exposed sera. (E) Comparison of RBD and S2 seroreactivity across 272 pre-pandemic serum samples. Threshold for RBD positivity as described in Figure 1. Threshold for S2 positivity was set as 5 SDs above the average OD₄₅₀ of the negative-control cohort. (F) ELISA results from indeterminate and putative seropositive samples from community testing. Thresholds for seropositivity were defined as in (E). Red circles indicate samples that have PRNT₉₀ titers of at least 1:20. Experiments were repeated at least once. Error bars in (B) and (D) depict the mean value of datasets ± SEM.

Figure 3

Antibody Responses to SARS-CoV-2 as a Function of Disease Severity and Age (A–C) Antibody titers to RBD (A), S2 (B), and N (C), over time post-onset of SARS-CoV-2 infection symptom grouped by case severity. The negative-control average was determined by calculating the average AUC value of negative-control (n = 25) samples. The p values represent a comparison of fit in a non-linear regression model between displayed groups; p < 0.05 rejects a single non-linear regression model fit for all datasets in the plot and creates two significantly different model fits. (D) PRNT₉₀ values over time post-onset of SARS-CoV-2 infection symptoms. The p values were calculated as in (A)–(C). (E–G) Antibody titers over time post-onset of SARS-CoV-2 infection symptoms from PCR+ confirmed patients or seropositive individuals from community-wide cohort for RBD (E), N (F), and S2 (G), grouped by patient age. (H) PRNT₉₀ values over time post-onset of SARS-CoV-2 infection symptoms grouped by patient age. For (E)–(H), p values were calculated as in (A)–(D).

Figure 4

Antibody Responses to Spike Glycoprotein Are More Stable Than Responses to Nucleocapsid (A–C) Antibody titers for mild infections over time to RBD (A), S2 (B), and N (C) for PCR-confirmed subjects and seropositive samples from community serological testing. Solid lines connect data from individuals sampled serially over time. Blue line depicts smoothing splines curve fit with 4 knots. Dashed line depicts mean values from seronegative controls. (D) Subjects sampled serially were assessed for changes in antibody titers to RBD, S2, and N from the first draw to the last draw collected. Only subjects in which the last draw occurred >6 weeks from onset are shown. The p values were calculated by paired 1-way ANOVA. (E) Neutralizing titers were measured for longitudinal subjects over time post-onset. Solid lines connect data from individuals sampled serially over time. Spline curve (blue line) was generated in Prism using Loess smoothing splines. Pseudo-R² values were calculated as the squared correlation of the predicted outcomes and the actual outcomes from the fitted model.