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

Abstract

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, and the most predictive genes pointed to inflammatory and tissue repair programs such as IL6/JAK/STAT3 signalling. When transferred to mouse blood, the model reliably separated high- from low-reactogenicity formulations, although the intermediate class was less well resolved, suggesting shared transcriptional programs between compartments and coordinated innate responses. When applied to human blood, the classifier identified Fluad (MF59-adjuvanted) as the most reactogenic licensed vaccine, in agreement with elevated C-reactive protein and ReactoScore values, whereas smaller differences among the remaining vaccines were not resolved. These results align with clinical safety data and suggest that early transcriptomic signatures in mice can identify highly reactogenic formulations in humans. 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 screening of vaccine tolerability.