Evidence for antibody as a protective correlate for COVID-19 vaccines

Abstract

A correlate of protection (CoP) is urgently needed to expedite development of additional COVID-19 vaccines to meet unprecedented global demand. To assess whether antibody titers may reasonably predict efficacy and serve as the basis of a CoP, we evaluated the relationship between efficacy and in vitro neutralizing and binding antibodies of 7 vaccines for which sufficient data have been generated. Once calibrated to titers of human convalescent sera reported in each study, a robust correlation was seen between neutralizing titer and efficacy (ρ = 0.79) and binding antibody titer and efficacy (ρ = 0.93), despite geographically diverse study populations subject to different forces of infection and circulating variants, and use of different endpoints, assays, convalescent sera panels and manufacturing platforms. Together with evidence from natural history studies and animal models, these results support the use of post-immunization antibody titers as the basis for establishing a correlate of protection for COVID-19 vaccines.

Keywords: COVID-19; Correlate of protection; SARS-CoV-2; Vaccine.

Fig. 1

Correlation between antibody responses and efficacy rate for 7 COVID-19 vaccines. Panels A and B display correlations of antibody responses for neutralization and ELISA assay ratios, respectively, without HCS calibration. Panels C and D display the same vaccine-induced responses, but with HCS calibration. Data included in correlation analyses are described in Tables S1 and S2. Dot size corresponds to the number of cases reported for Phase III efficacy analyses. The y-axis is estimated log risk ratio reported on the vaccine efficacy scale. The x-axis is ratio of the peak geometric mean neutralization titer or ELISA titer at 7–28 days post vaccination, relative to HCS. Error bars indicate 95% confidence Intervals (except for Oxford/AZ antibody responses, which represent ratios of median titers with interquartile ranges) with dashed line showing non-parametric LOESS fit. A rank correlation value was calculated with R² in a linear model utilized for variance explanation.

Fig. 2

Impact of post-hoc analyses to assess efficacy against ancestral strain on correlation. To assess the impact that variation in circulating strains may have on correlation between neutralizing antibody titer (Panel A) or binding antibody titer (Panel B) and efficacy, post-hoc analyses that calculate efficacy against the dominant ancestral strain D614G (Novavax) or calculate efficacy at sites without circulating VOCs (Janssen/J&J) were substituted for primary endpoint efficacy estimates (blue dots). Binding antibody ratio for J&J (Panel B) was calculated from US-specific Phase III data, calibrated to HCS titers published with Phase I/II immunogenicity data, as noted in Table S2.

Fig. 3

Impact of exploratory analysis of Oxford/AstraZeneca immunogenicity and efficacy by interval between doses on correlation. Publication of a pooled analysis of Phase 3 trial data for Oxford/AstraZeneca suggests that interval between doses in Phase 3 studies varied from 4 to 12+ weeks, and that both vaccine efficacy and immunogenicity data varied by dose interval. The immunogenicity data from Phase 1/2, which corresponded to a 4-week dose interval, may therefore not be representative of immunogenicity generated in the Phase 3 study, or correspond to pooled vaccine efficacy estimates. To assess the impact on correlation, exploratory analyses of dose intervals < 6 week and ≥ 12 weeks, and corresponding immunogenicity and efficacy, were substituted for the pooled vaccine efficacy estimate used in Fig. 1. Exploratory analyses are denoted by blue dots. Ratios for neutralizing antibody titer (Panel A) and binding antibody titer (Panel B) were generated for Oxford/AstraZeneca using the HCS median titer from Phase 1/2 publication, as noted in Tables S1 and S2.

Fig. 4

Prediction of vaccine efficacy using a cross-validation leave-one-out analysis. A non-parametric Bayesian approach was used to evaluate antibody titers as a potential correlate. Neutralizing antibody titer (A) and binding antibody titer (B) were each evaluated using leave-one-out cross-validation: the immune response and VE from six of the trials was used to build a model to predict the VE in the seventh trial based on the observed immune response distribution. Predicted VE is shown for each trial (red dots and 95% confidence interval). Observed VE and immune response data are also plotted for reference (black dots).