AlphaFold modeling of polyubiquitin complexes and covalently linked proteins
1.The paper presents two strategies for improving AlphaFold’s ability to model covalently linked ubiquitin chains and their complexes: one based on cysteine mutations and another using covalent linkers that mimic isopeptide bonds.
2.The most impactful finding is that introducing a propane-based isopeptide-bond mimetic in AlphaFold 3 allows for accurate prediction of both di- and polyubiquitin structures with predefined linkages, including branched chains and complex architectures.
3.Standard AlphaFold struggles to predict polyubiquitin structures due to the high sequence conservation of ubiquitin and the combinatorial explosion of linkage types, which suppresses coevolution signals and prevents the formation of valid inter-Ub linkages.
4.To address this, the authors first applied correlated cysteine mutations to encourage proximity between linkage sites. This approach improved structural predictions in AlphaFold 2.3 and AlphaFold 3 by favoring disulfide bond formation between modified ubiquitins.
5.However, the most robust improvements came from the second approach: using AlphaFold 3’s new bonded ligand functionality to introduce a propane molecule as a covalent linker, mimicking the natural isopeptide bond between ubiquitins.
6.This covalent linker strategy successfully placed ubiquitins in positions matching experimental structures, reduced RMSD values, and generalized well to triUb and polyUb systems. It even captured elusive conformations that are challenging to resolve experimentally.
7.For particularly challenging complexes (e.g., synthetic antibody-ubiquitin complexes), combining covalent linkers with cross-linking restraints further improved model accuracy, highlighting the approach’s flexibility.
8.The linker-based method enables modeling of arbitrarily linked polyubiquitin chains, including mixed and branched topologies, and can be extended to other covalent PTMs, hybrid Ub-like chains, and small-molecule-constrained complexes such as PROTACs.
9.The authors provide open-source scripts to automate polyUb generation and demonstrate AlphaFold 3’s potential in decoding the "ubiquitin code" by structurally resolving linkage-specific conformations and binding modes.
💻Code:
github.com/bio-phys/polyUb-A…
📜Paper:
biorxiv.org/content/10.1101/…
#AlphaFold3 #Ubiquitin #StructuralBiology #DeepLearning #ComputationalBiology #ProteinModeling #PostTranslationalModification #PolyUb