Deep learning-enabled scaffolding of spatial arrays of PfCSP epitopes

Abstract

Malaria is a leading cause of disease in developing countries. The licensed malaria vaccine RTS,S/AS01 confers partial protection in part due to the elicitation of circumsporozoite protein (CSP) antibodies, of which those to the CSP repeat and junctional regions offer the most potent protection. Anti-repeat region antibodies, including the protective antibody L9, frequently develop mutations that promote inter-Fab contacts when bound to CSP in "spiral" quaternary structures. As a first step toward the design of immunogens that elicit L9-like antibodies, we utilized generative deep learning models to design epitope-scaffolds that incorporated up to three junctional repeat epitopes with structural conformations and relative spatial orientations matching those of the multivalent complex of CSP bound to three copies of L9. Affinity and structural studies demonstrated accurate scaffolding of two epitopes with the intended relative orientation, and less accurate positioning of the third epitope. This study demonstrates proof of principle for design of multi-epitope scaffolds with pre-determined relative epitope spatial positioning. The study also represents an initial step toward development of multi-epitope immunogens to elicit antibodies that utilize homotypic interactions to bind pathogen in multivalent clusters.