LIFE SCIENCES

Most, if not all, proteins are subjected to distinct post-translational modifications, i.e. the covalent, typically enzymatic modification of a protein after its biosynthesis. Ubiquitylation refers to the post translational modification of proteins by attachment of Ubiquitin (Ub) or Ub oligomers. Ub is a 76 amino acid protein that is covalently linked to a substrate protein via an isopeptide bond formation between its C terminus and a lysine residue of the substrate. Ub itself can be ubiquitylated at eight different positions which leads to the formation of different types of Ub chains. In the cell, Ub chains of different linkage topology and length appear to signal respectively modified proteins for different fates or functions, i.e. ubiquitylation appears to act as a code to store and transmit information inside the eukaryotic cellular system.

Even though a significant amount of experimental Ub chain structures are available specific recognition mechanisms are still not entirely understood. Here, combined experimental and computational approaches are needed that shed light on the diverse, dynamic conformational equilibria of Ub chains. Within the present project, the computational resources on the JUWELS system were used to generate extensive simulation datasets of Ub oligomers (dimers, trimers and tetramers) of all eight linkage types. Molecular dynamics (MD) simulations were performed at two levels of resolution: coarse grained (CG) simulations were essential to sample the highly diverse conformational equilibria of the chains, including the formation of a multitude of protein-protein interfaces between the different Ub units. These CG simulations were complemented by atomistic simulations that provide the necessary chemical details and the necessary link to experimental data. In order to fully characterize Ub-Ub interactions and the relevant protein interfaces also extensive simulations of systems with multiple Ub monomers were generated. Beyond the specific Ub context the so generated data are an optimal basis to investigate the impact of protein concentration on protein aggregation in dense solutions.

The HPC resources have been chiefly used for the production of trajectory data. Post processing, analysis and visualization are performed on local machines. To analyze the vast amount of high-dimensional data, an efficient, highly scalable analysis framework was urgently required. To this end, a neural network-based dimensionality reduction technique was used to obtain a two-dimensional representation of the conformational space. With the density-based clustering algorithm HDBSCAN characteristic conformational states were detected. Last but not least a metric was identified that allows to assess the (dis)similarity between conformational spaces (from simulations of different Ub chains or from simulations at different resolution levels). In the context of the dual-resolution framework, the two-dimensional conformational free-energy landscapes were found to be elemental for the systematic connection between the resolution levels and, thus, to optimally utilize the CG simulations to obtain equilibrated atomistic ensembles.

This suite of methods was found to be capable of characterizing Ub chain configurations with different linkage topologies and chain lengths and the linkage dependent impact on protein-protein interaction patterns. First it was found that ubiquitin dimers exhibit characteristic linkage type-dependent properties in solution, such as interface stability and the character of contacts between the subunits, which can be directly correlated with experimentally observed linkage-specific properties.

This analysis was extended to trimers and tetramers. There new features were found since the different linkages exhibit different propensities for compact and extended conformations. Moreover, the extent to which chain elongation affects the protein-protein interface between adjacent Ub moieties was found to be highly linkage dependent. This analysis is still ongoing and it is to be expected that the simulation data will help to explain the experimentally observed chain length specific effects.

In summary, the simulation data obtained in this project have contributed to a better understanding of Ub signaling and to the development and improvement of analysis techniques for the characterization of protein-protein interactions.

The project has significant relevance beyond the Ub system in the sense that the methods can be generally used for the simulation and characterization of multidomain proteins and the characterization of protein-protein interactions. Importantly, the low-dimensional maps that are generated to identify conformational states are also central to an extension of the dual-scale framework that is currently being employed. In backmapping-based sampling, the CG model explores the conformational phase space and subsequently an atomistic ensemble is produced that not only re-samples the (potentially flawed) free energy landscape of the CG model but corresponds to the atomistic Hamiltonian.

Backmapping-based sampling relies on the here developed metric to monitor the deviation of CG and the evolving atomistic free energy landscapes and to systematically improve the ensemble generated from the CG model. The so-obtained atomistic information is of great importance for the interpretation of experimental (e.g. NMR) data.