Modeling brings additional insights into the kinetics of SARS-CoV-2 neutralizing antibody

Mélanie Prague, Quentin Clairon, Jérémie Guedj, Rodolphe ThiébautmedRxiv 2021 · DOI

Abstract

Following the paper by Seow et al. published in Nature Microbiology, we reanalyzed the publicly available data using dynamical models of humoral response. The main conclusion is that, from available data, we can demonstrate that the decline of neutralizing antibodies (as measured with $ID_{50}$) is biphasic, which is compatible with two types of antibody secreting cells (short lived and long lived). We found that lower bound of half life of the long-lived antibody secreting cells is 450 days. Moreover, our model predicts that the neutralizing antibody response could be more durable than suggested (up to 129 days for individuals with no requirement of supplemental oxygen and up to 175 days for others). A result which adds insight on the longevity of immune response.

Introduction

Early in the COVID-19 pandemic, a major concern was the "waning immunity" observed in recovered patients. Descriptive studies, such as the influential paper by Seow et al. (2020), showed antibody titers dropping significantly within the first few months, leading to fears that protection might be fleeting. However, raw data often masks underlying biological mechanisms. Simple descriptive statistics do not account for the different types of cells producing antibodies. This study re-evaluates the Seow et al. dataset—comprising repeated measures of neutralizing antibody titers ($ID_{50}$) from 83 individuals—using mathematical modeling to determine if the observed decline reflects a simple linear "fade-out" or a more complex immune structure involving both short-term and long-term cellular responses.

Methods

The researchers applied a dynamical model of humoral response, a framework previously used to study Ebola virus kinetics. The model assumes two distinct populations of Antibody Secreting Cells (ASCs):

  1. Short-lived ASCs ($S$): Responsible for the initial burst of antibodies.
  2. Long-lived ASCs ($L$): Responsible for sustained, low-level production over time.

Statistical analysis utilized Non-linear Mixed-Effects Modeling with the Stochastic Approximation Expectation-Maximization (SAEM) algorithm via Monolix software. This approach allows for robust parameter estimation by leveraging the entire dataset, even for individuals with sparse data. The biphasic model was statistically compared against a simplified model that ignored long-lived cells.

Results

The analysis confirmed that the decline of neutralizing antibodies is biphasic rather than linear. An initial rapid drop, driven by the decay of short-lived cells (half-life $\approx$ 4.8 days), is followed by a much slower decay phase supported by long-lived cells. The biphasic model significantly outperformed the simpler model ($P=0.001$), providing strong evidence for the existence of long-lived plasma cells in the SARS-CoV-2 response.

Biphasic decay and probability of detectable antibodies over time.

Interestingly, disease severity (defined by the requirement for supplemental oxygen) influenced the magnitude of the response but not the rate of decay. Patients with severe symptoms showed a 14-fold increase in initial short-lived cell activity, yet the "biological clock" of the immune cells ticked at a similar rate across groups.

The model predicts that the time to reach undetectable antibody levels ($ID_{50} < 50$) is longer than previously estimated. For severe cases, the mean duration is 175 days, while for non-severe cases, it is 129 days. Crucially, the model allows for an estimated lower bound for the half-life of long-lived antibody-producing cells of at least 450 days.

Discussion

While these results offer a more optimistic view of immune longevity, several limitations remain. The 450-day half-life is a "lower bound" estimate due to the relatively short three-month follow-up period of the original dataset; longer-term studies are necessary for confirmation. Additionally, while the model identifies these cell populations mathematically, biological characterization in a laboratory setting is required to physically identify these long-lived plasma cells. Finally, high individual variability was observed, particularly in non-severe cases, where some individuals were predicted to lose detectable antibodies rapidly. This underscores the potential need for personalized booster strategies despite the overall durability of the response.

Summary updated

Abstract

Abstract Following the paper by Seow et al. published in Nature Microbiology, we reanalyzed the publicly available data using dynamical models of humoral response. The main conclusion is that, from available data, we can demonstrate that the decline of neutralizing antibodies (as measured with ID50) is biphasic, which is compatible with two types of antibody secreting cells (short lived and long lived). We found that lower bound of half life of the long-lived antibody secreting cells is 450 days. Moreover, our model predicts that the neutralizing antibody response could be more durable than suggested (up to 129 days for individuals with no requirement of supplemental oxygen and up to 175 days for others). A result which adds insight on the longevity of immune response.
BibTeX
@misc{prague2021modeling,
  author = {Mélanie Prague and Quentin Clairon and Jérémie Guedj and Rodolphe Thiébaut},
  title = {{Modeling brings additional insights into the kinetics of SARS-CoV-2 neutralizing antibody}},
  year = {2021},
  doi = {10.1101/2021.10.13.21264693},
  howpublished = {medRxiv},
}
Figure