Senior Scientist, Quantitative Systems Pharmacology Moderna, Inc. Malden, Massachusetts, United States
Disclosure(s):
Kenji Miyazawa: No financial relationships to disclose
Objectives: To compare the structure and sensitivity profiles of two models of humoral immune response: a semi-mechanistic model for Ad26.COV2.S (adenoviral vaccine) and a mechanistic QSP model for LNP-mRNA vaccination. The goal is to identify key drivers of antibody durability and assess how model complexity impacts prediction, interpretation, and application in vaccine development.
Methods: The Dari et al. (2023) semi-mechanistic model uses a theoretical antigen compartment to activate memory B cells, germinal center (GC) B cells, and plasma cells, calibrated to human data over 427 days. The QSP model (Salvaggio et al., 2021) includes LNP uptake, antigen translation, APC activation, and downstream T and B cell dynamics, calibrated to macaque data (Liang et al., 2017). We extended the QSP model to include explicit spike protein-B cell receptor binding. Structural comparison and local sensitivity analysis (± 50% parameter variation) were conducted, evaluating plasma antibody titers. Virtual populations were simulated to examine booster and dose-response effects.
Results: In the Dari model, antibody AUC was strongly modulated by antigen decay and GC B cell dynamics. In contrast, the QSP model demonstrated LNP degradation at the injection site as most impactful – with a 50% increase in LNP half-life led to 4.5-fold longer antibody persistence. This highlights the importance of LNP formulation design on antibody durability. Other sensitive parameters included LNP uptake by dendritic cells, initial DC count, and long-lived plasma cell decay. T cell activation and proliferation showed moderate sensitivity. Virtual population simulations with the QSP model revealed dose-response saturation, with limited benefit in antibody durability beyond a certain dose. Nonlinear immune constraints were also captured by the QSP model, where overstimulation by the primary dose reduces short-interval boosters due to limited antigen-specific T cells and memory cells production. While the Dari model included saturation in B cell proliferation, it lacks upstream T cell dynamics, limiting its scope in simulating antigen competition or exhaustion.
Conclusions: Both models describe the data well but offer different insights. The QSP model supports mechanistic interpretation and formulation design optimization. The semi-mechanistic model, with reduced complexity, enables flexible data-driven parameter estimation. Interestingly, the Dari model effectively captured clinical data, suggesting B cell dynamics alone may be sufficient to describe humoral responses in some contexts. These findings highlight the importance of tailoring model complexity to the scientific questions of interest, balancing mechanistic depth with abstraction to match the modeling goals. While simplified models help fitting and interpretation, QSP models remain powerful for exploring LNP kinetics at the injection site and the T cell-antibody cross-talk on durability.
Citations: [1] Dari A, Jacqmin P, Iwaki Y, et al. Mechanistic modeling projections of antibody persistence after homologous booster regimens of COVID-19 vaccine Ad26.COV2.S in humans. CPT Pharmacometrics Syst Pharmacol. 2023;12(10):1485-1498. doi:10.1002/psp4.13025 [2] Selvaggio G, Leonardelli L, Lofano G, et al. A quantitative systems pharmacology approach to support mRNA vaccine development and optimization [published correction appears in CPT Pharmacometrics Syst Pharmacol. 2022 Apr;11(4):524. doi: 10.1002/psp4.12767.]. CPT Pharmacometrics Syst Pharmacol. 2021;10(12):1448-1451. doi:10.1002/psp4.12721 [3] Liang F, Lindgren G, Lin A, Thompson EA, Ols S, Röhss J, John S, Hassett K, Yuzhakov O, Bahl K, Brito LA, Salter H, Ciaramella G, Loré K. Efficient Targeting and Activation of Antigen-Presenting Cells In Vivo after Modified mRNA Vaccine Administration in Rhesus Macaques. Mol Ther. 2017 Dec 6;25(12):2635-2647. doi: 10.1016/j.ymthe.2017.08.006. Epub 2017 Aug 12. PMID: 28958578; PMCID: PMC5768558.
