Systematic Investigation of the Liquid-Liquid Phase Separation Propensity of Intrinsically Disordered Proteins by Molecular Simulations.

Liquid-liquid phase separation (LLPS) has been intensively studied since P-granules in Caenorhabditis elegans drew the attention of researchers. Decades of studies have revealed that intrinsically disordered proteins or regions (IDP/Rs) are closely related to phase-separating (PS) proteins, and their multivalency is instrumental to the LLPS propensity. Here, we constructed a comprehensive database of PS and non-PS IDPs and introduced a framework to calculate the exact value of multivalence with molecular dynamics (MD) simulations. Also, we examined the correlation of the calculated multivalence to LLPS. Results show that the simulation-based classification of the LLPS propensity can achieve high consistency with experimental observations despite varying experimental conditions. Besides, inspired by a graph theoretical approach, we determined a recommended cutoff of multivalence for effectively differentiating IDPs with a high LLPS propensity. Last, studies on the role of amino acid types in the fused in sarcoma (FUS) protein and its mutants reveal context-dependent interactions in prion-like and RNA-binding domains. Our work deepens the understanding of LLPS and provides a cost-effective computational framework for studying LLPS for IDPs.