April 7
@
4:00 PM
–
5:00 PM
Kayla Sprenger
University of Colorado Boulder
Boulder, CO
Towards Modeling Antibody–Virus Coevolution and Escape: Integrating Evolutionary Modeling, Molecular Dynamics Simulations, and Interpretable Machine Learning
In the Rationally Designed Immunotherapeutics and Interfaces (RDI) Lab, we integrate computational modeling, immuno-engineering, molecular biophysics, and machine learning to understand—and ultimately control—how immune systems respond to rapidly evolving viral pathogens. A key challenge in designing vaccines against such pathogens is engineering immunogens that elicit broadly neutralizing antibodies (bnAbs), which target conserved regions of viral surface proteins and thereby bind diverse viral variants. Yet, even the most potent bnAbs isolated from infected individuals to date have proven susceptible to escape by viral mutations that weaken or abolish bnAb binding. Notably, neutralization datasets frequently reveal escape mutations at sites distal to the antibody-bound epitope, suggesting that allosteric and epistatic effects play a key role in modulating binding. In most cases, the mechanistic basis by which these distal mutations confer escape remains poorly understood, limiting our ability to design vaccine immunogens or antibody-based therapeutics that are resistant to escape mutations.
To address this gap, this talk will highlight our use of evolutionary frameworks to model B cell affinity maturation against static sequences of HIV-1–derived immunogens administered via time-varying immunization protocols. This approach enables us to understand how different vaccine strategies shape antibody lineages and guide them toward broadly neutralizing responses. In parallel, coupling these models of immune evolution with dynamic viral fitness landscapes enables identification of escape pathways that may be exploited in vivo, thus informing the iterative design of immunogens and immunization strategies capable of eliciting fully escape-resistant bnAbs. To further resolve the mechanistic basis of escape, this talk will also describe our use of atomistic molecular dynamics simulations and interpretable machine learning to characterize how distal mutations propagate dynamical changes through the structure of HIV-1’s Envelope (Env) spike protein to abrogate antibody binding and neutralization. Complementing these structural insights, we have developed an interpretable protein language model framework trained on HIV-1 sequence data and bnAb neutralization profiles. This framework identifies context-dependent mutational effects that rewire long-range residue-level communication networks governing antibody sensitivity. Together, our work provides a mechanistic foundation for designing next-generation immunogens against highly mutable pathogens like HIV-1, as well as antibody-based therapeutics that are more robust to rapid viral evolution.