DESIGNING AND SIMULATING SEQUENCE-ENCODED PHASE SEPARATION IN DISORDERED AND MULTIDOMAIN PROTEINS

Abstract

Biomolecular condensates (BMCs) are membraneless organelles that can form through liq- uid–liquid phase separation. They help organize some of a cell’s biochemistry. It has been observed that improper formation of these condensates can be associated with neurodegenerative diseases and cancer. Much of the behavior associated with condensate formation and dynamics can be studied by examining the intrinsically disordered proteins involved in their formation. The phase behavior of intrinsically disordered proteins is strongly influenced by their amino acid sequence, which affects whether they form condensates under specific conditions such as temperature, protein concentration, and other physical or solution conditions. Two things matter most, which residues are present and how they are arranged along the chain. Coarse-grained simulations can be used to study the formation and dynamics of biomolecular condensates. They can help predict how a protein sequence may behave under different conditions before performing wet-lab experiments. However, several practical gaps remain. Existing patterning measures, including sequence charge decoration and sequence hydropathy decoration, are more suitable for comparing sequences with similar lengths and overall compositions. This makes them hard to use as a general design tool. It is also unclear whether descriptors that predict the phase behavior of an individual protein can predict its behavior when it forms a condensate with a second protein. A related gap is how the size of a large client and its interaction strength together decide whether it gets pulled in or excluded. It is also unclear whether folded domains can remain intact in coarse-grained simulations, which is necessary for realistically studying most natural proteins. This dissertation works through these gaps using coarse grained molecular dynamics simulation. Most of the work is based on the HPS-Urry model, together with sequence-design tools and statistical measures used to quantify mixing. We introduced a shuffle-based normalization method. It compares patterning across sequences of different length and composition. We used it in a Monte Carlo algorithm to create new sequences with specific phase behavior. For proteins that combine folded and disordered regions, we added a network of internal restraints. This holds the folded part together while the rest of the chain moves freely. Using this approach, we found that whether a sequence mixes with or separates from a partner protein is mainly determined by a single factor. It depends mainly on whether the sequence can undergo phase separation on its own. It depends less on how different its patterning is from the partner’s. We also built an entropy based measure of mixing. It remains effective for systems with unequal compositions, where the standard demixing index may not work well. In our simple particle simulations, the interaction strength required to recruit a large client into a condensate scaled with its radius rather than its volume. A similar pattern was seen when comparing short and long chains of the same polymer, the longer chain had a greater number of stickers. It reached a higher concentration in the dense phase and moved more slowly within the condensate. We applied our folded domain model to the SARS-CoV-2 nucleocapsid protein. We compared it against experimental scattering and single molecule data. It showed that phosphorylation of the protein’s disordered linker creates new contacts across the dimer. This explains why phosphorylation weakens RNA binding and softens the resulting condensate. This work provides a framework for studying and designing of intrinsically disordered proteins. It introduces a sequence-patterning design tool for quickly designing sequences with specific properties. It provides a simple predictor of whether two proteins will share a condensate across different compositions. It shows how size and interaction strength jointly decide what a condensate can recruit. And it shows that folded domains can be added to sequence based simulations without losing their structure, which allows more realistic and disease-relevant proteins to be studied computationally. Finally, it links a molecular mechanism to experimentally measured condensate properties.