Glycans at the Core: Computational-Experimental Investigations of Complex Carbohydrates

Abstract

Next to nucleic acids and proteins, glycans represent a third class of biological sequence, composed of monosaccharides assembled into complex and often branched structures. Glycans modify various biological molecules, most commonly proteins and lipids, and engage in a diverse range of functions, primarily through interactions with specific glycan-binding proteins known as lectins. Throughout this thesis, multiple aspects of glycans, lectins, and glycosylation mechanisms are explored.

The remarkable diversity of both glycans and lectins imposes experimental challenges for characterizing the binding specificity of newly discovered lectins. To address this, we introduce LectinOracle, a deep learning model that combines transformer-based protein representations with graph convolutional neural networks for glycans, enabling accurate prediction of lectin-glycan interactions.

In parallel, we employ an extensive array of experimental techniques to thoroughly characterize a newly identified plant lectin from Cucumis melo, investigating its glycan-binding specificity, binding kinetics, and solving the structure of the N-terminal domain in complex with glycan ligands.

Finally, we challenge a long-standing paradigm in the field of glycobiology: that O-GalNAc glycosylation is restricted to proteins destined for secretion. In this study, we conclusively demonstrate that nuclear proteins can be modified with extended O-GalNAc-type glycans through a mechanism that depends on Golgi-resident biosynthetic enzymes. Our findings suggest the existence of a novel pathway in which nuclear proteins are actively shuttled to and from the secretory pathway.

Altogether, the work presented in this thesis contributes significantly to the advancement of key areas in glycobiology, spanning computational modeling, structural biology, and fundamental insights into glycosylation mechanisms.