Dipeptides are minimalistic but sufficient for liquid-liquid phase separation

Liquid-liquid phase separation (LLPS) of proteins mediates the assembly of biomolecular condensates involved in physiological and pathological processes. Identifying the minimalistic building blocks and the sequence determinant of protein phase separation is of urgent importance but remains challenging due to the enormous sequence space and difficulties of existing methodologies in characterizing the phase behavior of ultrashort peptides. Here we demonstrate computational tools to efficiently quantify the microscopic fluidity and density of liquid-condensates/solid-aggregates and the temperature-dependent phase diagram of peptides. Utilizing our approaches, we comprehensively predict the LLPS abilities of all 400 dipeptide combinations of coded amino acids based on 492 micro-second molecular dynamics simulations, and observe the occurrences of spontaneous LLPS. We identify 54 dipeptides that form solid-like aggregates and three categories of dipeptides with high LLPS propensity. Our predictions are validated by turbidity assays and differential interference contrast (DIC) microscopy on four representative dipeptides (WW, QW, GF, and VI). Phase coexistence diagrams are constructed to explore the temperature dependence of LLPS. Our results reveal that aromatic moieties are crucial for a dipeptide to undergo LLPS, and hydrophobic and polar components are indispensable. We demonstrate for the first time that dipeptides, minimal but complete, possess multivalent interactions sufficient for LLPS, suggesting that LLPS is a general property of peptides/proteins, independent of their sequence length. This study provides a computational and experimental approach to the prediction and characterization of the phase behavior of minimalistic peptides, and will be helpful for understanding the sequence-dependence and molecular mechanism of protein phase separation.SignificanceProtein liquid-liquid phase separation (LLPS) is associated with human health and diseases. Identifying the minimalistic building blocks and sequence determinants of LLPS is of urgent importance but remains computationally challenging partially due to the lack of methodologies characterizing the liquid condensates. Herein we provide approaches to evaluate LLPS ability of dipeptides, and screen all 400 dipeptides by MD simulations combined with multi-bead-per-residue models which capture key interactions driving LLPS that are missing in one-bead-per-residue models. Three categories of LLPS dipeptides are identified and the experimentally-verified QW dipeptide is by far the smallest LLPS system. Our results suggest that dipeptides, minimal but complete, possess multivalent interactions sufficient for LLPS, and LLPS is a general property of peptides/proteins, independent of their length.