A translational transcriptomic signature of vaccine reactogenicity for the evaluation of novel formulations

Accurately predicting vaccine reactogenicity at the preclinical stage remains a major challenge in vaccine development, as conventional animal studies and in vitro assays capture general inflammation but fail to quantify local or systemic reactogenicity relevant to humans. Using transcriptomic data from the BioVacSafe consortium encompassing seven vaccines and immunostimulants in mice and five licensed vaccines in humans, we developed a cross-compartment and cross-species predictive model of vaccine reactogenicity. Reactogenicity classes were defined in mouse muscle based on the magnitude of transcriptomic responses and literature evidence. A penalized ordinal regression model was trained to predict both discrete classes and continuous scores of reactogenicity. Transcriptomic profiles from mouse muscle were highly predictive of reactogenicity, with key genes enriched in inflammatory and tissue repair pathways such as IL6/JAK/STAT3 signalling. The model retained strong performance when transferred to mouse blood and revealed shared transcriptional programs between compartments, suggesting coordinated innate responses. When applied to human blood, the classifier correctly ranked licensed vaccines by reactogenicity, identifying Fluad (MF59-adjuvanted) as the most reactogenic, in agreement with elevated C-reactive protein and ReactoScore values, while Engerix-B, Varilrix, and Stamaril were classified as low-reactogenicity formulations. These results align with clinical safety data and demonstrate that early transcriptomic signatures in mice can predict human reactogenicity profiles. Our study presents a pan-vaccine, cross-species transcriptomic signature that bridges preclinical and clinical data, offering a foundation for translational biomarkers and mechanism-informed assessment of vaccine tolerability.