AI-guided epitope engineering of a SARS-CoV-2 spike antigen for broad sarbecovirus neutralization

Abstract

The rapid evolution of SARS-CoV-2 and the ongoing risk of zoonotic spillover highlight the need for vaccines that provide broad protection beyond strain-specific immunity. Here, we present a structure-guided, AI-enabled strategy for rational antigen design that enhances cross-reactive B-cell epitope recognition across betacoronaviruses. By integrating comparative sequence analysis with conformational epitope prediction, we identified conserved epitope hotspots within the Spike receptor-binding domain and introduced targeted motif-level substitutions into a SARS-CoV-2 BA.2 backbone. Engineered Spike immunogens, delivered as mRNA-lipid nanoparticle vaccines, elicited robust antibody responses in mice and demonstrated broad neutralizing activity against antigenically diverse SARS-CoV-2 variants, including BA.5 and XBB.1.5, as well as zoonotic sarbecoviruses. Selected designs reached neutralisation breadth comparable to an Omicron-adapted clinical benchmark for shared SARS-CoV-2 antigens, while extending neutralising activity to divergent zoonotic sarbecoviruses. These findings show that minimal, structure-guided epitope remodeling can enhance the presentation of conserved epitopes and improve antibody breadth, providing a generalizable framework for designing vaccines resilient to viral evolution and future pandemic threats.