Multi-scale simulations of protein phase separation in gene regulation
Loading...
Date issued
Authors
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Reuse License
Abstract
The formation of distinct phase-separated condensates of bio-macromolecules underpins specific regulation in cells. In this thesis, I elucidate how phase-separated condensates regulate biochemical processes by providing distinct chemical environments through particle-based multi-scale simulations. The disordered C-terminal domain of RNA polymerase II (CTD) forms condensates in both unphosphorylated and phosphorylated (pCTD) states. Understanding the molecular driving forces behind CTD phase separation offers insights into how these condensates regulate transcription. To uncover the molecular determinants of CTD phase separation, I integrate nuclear magnetic resonance (NMR) data with atomistic molecular dynamics simulations using the refined hierarchical chain growth (RHCG) method. This reveals local structural preferences and highlights the key role of tyrosine-proline contacts in driving condensation. These insights aid in validating the coarse-grained simulations, which capture how temperature and post-translational modifications (PTMs) influence condensate dynamics. I further explore whether unphosphorylated and phosphorylated CTD form mixed or distinct condensates by calculating interfacial properties from coarse-grained models. To understand how phosphorylation modulates CTD condensate behaviour, an algorithm for simulating stochastic phosphorylation dynamics is developed. This approach provides a mechanistic understanding of how phosphorylation alters the phase behaviour of CTD condensates. By dynamically tuning protein interactions, phosphorylation modulates condensate stability, facilitating transitions between different functional states. This coupling between phosphorylation dynamics and phase separation highlights the regulatory potential of post-translational modifications in transcriptional condensates. Finally, I introduce an active learning framework combining Bayesian optimization and coarse-grained simulations to study sequence determinants of phase behavior. This approach identifies sequence-property relationships for disordered proteins, including their self-interactions and interactions in phase-separated condensates. Using second virial coefficients, I predict peptide self-interactions and identify sequences that bind to CTD condensates. This approach allows the rational design of multiphasic condensates with tunable morphologies. These findings highlight the importance of phase separation and post-translational modifications in regulating transcription. The active learning methods offer key insights for designing functional condensates and controlling cellular processes.