Ubiquitin (Ub) is a small, highly conserved protein that can be covalently attached to substrate proteins and other biomolecules, in a process called ubiquitination. The modification of Ub itself with further Ubs at various residues gives rise to a vast combinatorial space of linear and branched polyUb chains. Ubiquitination can lead to a wide variety of outcomes depending on the linkage type, which is often referred to as the “ubiquitin code”. Structural modeling of polyUb bound to the “writers” (i.e., ligases), “readers” (i.e., Ub-binding domains), and “erasers” (i.e., deubiquitinases) of the Ub code can illuminate the function and characteristics of these proteins, and their role in the broader Ub ecosystem. Recently, AlphaFold (AF) has emerged as a powerful tool for protein complex structure prediction. However, without the possibility to include interchain linkages, its applicability to polyUb chains has been limited.
We explore two strategies to induce isopeptide-bond mimetic linkages in AF predictions of polyUb complexes. First, we use correlated cysteine mutations to induce linkage-specific proximity of Ubs, which is applicable in both AF2 and 3. Second, we utilize AF3’s ability to covalently link small compounds to biomolecular residues to introduce explicit linkages. We show that in comparison with naïve predictions (using unmodified Ubs), both strategies substantially improve the placement of diUb in interactions with known Ub binders. Further, the well-defined linkages via covalent bonds in AF3 allow the reliable modeling of even complex, branched polyUb chains, making the vast combinatorial space of Ub linkages accessible to AF modelling.