Liquid–Liquid Phase Separation

In cell biology and soft-matter biophysics, biomolecular condensates are micron- and sub-micron-scale intracellular compartments that concentrate specific proteins and nucleic acids (RNA and DNA) by 10- to 100-fold relative to the surrounding cytoplasm or nucleoplasm without an enclosing lipid bilayer membrane. First demonstrated as a physical phase transition in living cells by Clifford Brangwynne, Anthony Hyman, Frank Jülicher, and colleagues in 2009—who showed that germline P-granules in Caenorhabditis elegans embryos behave as liquid droplets that wet surfaces, fuse into spheres under surface tension, drip under shear stress, and dissolve/condense across a sharp saturation concentration—liquid–liquid phase separation (LLPS) has unified the study of dozens of classic membrane-less organelles.

Across the eukaryotic cell, biomolecular condensates organize essential biochemical processes: in the nucleus, they form the tripartite nucleolus (ribosome biogenesis), Cajal bodies (snRNP assembly), nuclear speckles (splicing factor storage), paraspeckles (NEAT1 lncRNA), PML bodies, transcriptional super-enhancer condensates (MED1/BRD4/RNA Pol II), and constitutive heterochromatin (HP1α); in the cytosol, they form stress granules (G3BP1/FUS/TDP-43), processing bodies (P-bodies), germline P-granules, the centrosomal pericentriolar material (SPD-5), and the postsynaptic density (PSD).

Why does a homogeneous solution of macromolecules spontaneously demix into two coexisting liquid phases—a dilute phase of volume fraction φ_dil (equals the saturation concentration c_sat) and a dense condensed droplet phase of volume fraction φ_dense—even though demixing lowers translational configurational entropy (ΔS_mix < 0)?

Formulated independently by Paul Flory and Maurice Huggins in 1941–1942 on a three-dimensional lattice of site volume v₀, the Flory–Huggins free energy of mixing per unit volume, f_FH(φ) = ΔG_mix / V, for a polymer of chain length N (in solvent monomer units, where N ≈ 100–1,000 for a protein or RNA) at volume fraction φ in an aqueous solvent of volume fraction (1 − φ) is given by the balance between translational mixing entropy and the Flory interaction parameter χ:

f_FH(φ) / (k_BT / v₀) = (φ / N) ln(φ) + (1 − φ) ln(1 − φ) + χ · φ (1 − φ)

where k_B is the Boltzmann constant, T is absolute temperature, and χ is the dimensionless Flory interaction parameter comparing the average free energy of polymer–solvent (P–S) contacts against polymer–polymer (P–P) and solvent–solvent (S–S) contacts across lattice coordination number z:

χ = ( z / 2 k_BT ) [ 2 ε_PS − ( ε_PP + ε_SS ) ]

1. Why Macromolecules (N ≫ 1) Phase Separate Easily: Notice that the first term—the translational entropy of the macromolecule, (φ / N) ln(φ)—is divided by the degree of polymerization N. When N = 1 (small molecules like glucose or NaCl), translational entropy is large and opposes demixing. When a polypeptide or RNA has chain length N = 300 to 2,000, the macromolecular mixing entropy per unit volume shrinks by 300- to 2,000-fold, meaning that even weak, transient attractive contacts between polymer chains (ε_PP < 0, making χ > χ_critical = ½ (1 + N^−1/2)² ≈ 0.5) are sufficient to make ∂²f_FH/∂φ² negative and drive spontaneous phase separation!
2. The Phase Diagram: Binodal (Coexistence) vs. Spinodal Curves:
   • Binodal Curve (Common-Tangent Construction, μ_P^dil = μ_P^dense): Defines the thermodynamic equilibrium phase boundary (saturation concentration c_sat = φ_dil). Between the binodal and spinodal curves (∂²f / ∂φ² > 0), the solution is metastable, and droplets form via nucleation and growth (requiring an initial free-energy barrier ΔG* = 16πγ³ / 3(Δg)² set by interfacial surface tension γ).
   • Spinodal Curve (∂²f_FH / ∂φ² = 0): Inside the spinodal boundary (∂²f / ∂φ² < 0), the homogeneous solution is thermodynamically unstable to infinitesimally small concentration fluctuations, undergoing barrier-free spinodal decomposition into bicontinuous phase networks.

Because real proteins and RNAs are heteropolymers rather than uniform Flory–Huggins homopolymers, Rohit Pappu, Michael Rosen, Tanja Mittag, Clifford Brangwynne, and colleagues developed the "stickers-and-spacers" percolation model (adapted from Michael Rubinstein and Albert Semenov's associative polymer theory) coupled with the "Scaffold-and-Client" model (Banani et al., 2016):

1. Multivalent Modular Domain Arrays (Folded Scaffolds): First demonstrated in vitro by Li, Rosen, and colleagues (2012), mixing a protein composed of m tandem SH3 domains with a ligand composed of n tandem PRM (Proline-Rich Motif) Short Linear Motifs triggers liquid droplet condensation specifically when the valency reaches m, n ≥ 3 to 5. Below valency 3, the chains form only small soluble dimers or trimers; at valency ≥ 4, branched cross-linking drives a sol-to-gel percolation transition coupled to phase separation, with the saturation concentration c_sat dropping sharply as valency increases. Similarly, in the stress granule hub G3BP1, homodimerization of the N-terminal NTF2L domain combines with the C-terminal RRM and RGG RNA-binding domains to cross-link long single-stranded mRNA polymers (valency > 10 binding sites per mRNA) into a percolated network.
2. Intrinsically Disordered Low-Complexity Domains (LCDs / Prion-Like Domains) — The Cation–π and π–π Grammar: Many condensate scaffolds—such as FUS, EWSR1, TAF15 (the FET family), TDP-43 (TARDBP), hnRNPA1, and DDX4/LAF-1—drive phase separation via long intrinsically disordered regions (IDRs) enriched in uncharged polar residues (Gly, Ser, Gln, Asn — the spacers that control chain solvation volume and prevent irreversible aggregation) punctuated at regular intervals by aromatic and charged stickers:
   • Cation–π Stickers (Tyrosine + Arginine): Quantitative mutagenesis of the FET family by Jie Wang, Hyman, Pappu, and colleagues (2018) demonstrated that the saturation concentration c_sat of FUS-family proteins is governed quantitatively by the product of the number of tyrosine (Tyr) residues in the prion-like domain and the number of arginine (Arg) residues in the RNA-binding RGG domain: c_sat ∝ ( n_Tyr · n_Arg )⁻¹. Planar guanidinium cations of Arg form favourable cation–π quadrupole interactions (~8 to 15 kJ/mol in water) with the electron-rich aromatic π-clouds of Tyr (and Phe), acting as reversible cross-links that break and reform on the microsecond-to-millisecond timescale. Notably, mutating Arg → Lys increases c_sat by an order of magnitude because the spherical ammonium group of Lys lacks the planar delocalized π/hydrogen-bonding geometry of Arg's guanidinium group.
   • Aromatic π–π Stickers (Tyr–Tyr and Phe–Phe): Drive homotypic condensation in hnRNPA1 low-complexity domains and in the cohesive GLFG hydrogel of Nup98 inside the Nuclear Pore Complex.
3. Multiphase Immiscibility (Core–Shell Architecture of the Nucleolus): What happens when two different condensates have distinct interfacial tensions γ₁ and γ₂ with the nucleoplasm and with each other (γ₁₂)? Just as oil and water form concentric droplets when their surface tensions satisfy Antonow's / Neumann's wetting inequalities, Marina Feric, Brangwynne, and colleagues (2016) proved that the three concentric sub-compartments of the nucleolus—the innermost Fibrillar Center (FC, RNA Pol I), surrounded by the Dense Fibrillar Component (DFC, Fibrillarin FBL), encased within the outer Granular Component (GC, Nucleophosmin NPM1)—are three immiscible coexisting liquid phases ordered radially by their surface tensions (γ_DFC > γ_GC)! Newly transcribed 47S pre-rRNA emerges at the FC/DFC interface, is methylated/pseudouridylated as it diffuses through the hydrophobic FBL phase (DFC), and is assembled with ribosomal proteins in the outer acidic/basic-tract NPM1 phase (GC).

Because the protein concentration inside a liquid biomolecular condensate reaches 200 to 400 mg/mL (~5 to 15 mM)—roughly 100-fold higher than in the surrounding bulk cytoplasm—the liquid droplet is actually a metastable state relative to the global thermodynamic free-energy minimum of a cross-β amyloid fibril (in accordance with Wilhelm Ostwald's 1897 Step Rule, whereby a supersaturated solution nucleates the kinetically accessible liquid phase first before slowly converting to the thermodynamically stable crystal/solid phase).

In healthy cells, ATP (which acts as a biological hydrotrope at its physiological concentration of 5–10 mM, solubilizing hydrophobic disordered domains even independently of enzymatic hydrolysis!), Heat Shock Proteins (Hsp70/Hsp40/Hsp104), Karyopherin-β2 (Transportin-1 / TNPO1, which binds the PY-NLS of FUS and hnRNPA1 to dissolve droplets), and post-translational phosphorylation (CK2, DNA-PK) or arginine methylation (PRMT1) continuously fluidize stress granules and dissolve them once cellular stress subsides.

In Amyotrophic Lateral Sclerosis (ALS) and Frontotemporal Dementia (FTD), however, patient-derived disease mutations—such as FUS (P525L, G156E), TDP-43 (A315T, M337V, G298S), hnRNPA1 (D262V), or C9orf72 hexanucleotide (GGGGCC) repeat expansion producing arginine-rich poly-GR and poly-PR dipeptide repeats—accelerate the irreversible liquid-to-solid ("aging" or "maturation") phase transition (Patel et al., 2015):

• Fluorescence Recovery After Photobleaching (FRAP) of freshly formed wild-type FUS or TDP-43 droplets shows rapid internal molecular rearrangement (recovery half-time t_1/2 ≈ 2 to 15 seconds).
• In mutant FUS G156E or TDP-43 M337V droplets (or droplets poisoned by C9orf72 poly-PR peptides that outcompete natural Arg–Tyr stickers), the interior of the liquid droplet nucleates steric-zipper cross-β amyloid fibrils within hours, arresting FRAP recovery (t_1/2 → ∞), converting spherical droplets into rigid "sea-urchin" or fibrillar hydrogel aggregates, and sequestering essential splicing factors and nucleoporins in motor neurons.

How do scientists prove in a living cell that a punctate structure is a liquid condensate (LLPS) rather than a solid aggregate or a stoichiometric complex?

Because many diffraction-limited puncta (< 250 nm) appear round under standard fluorescence microscopy regardless of their physical state, biophysicists apply four quantitative experimental criteria to diagnose LLPS:

1. Spherical Coalescence (Fusion Relaxation Time τ_fusion = (η / γ) · L): When two droplets touch, surface tension γ drives them to fuse and relax into a single larger sphere with a characteristic relaxation time τ_fusion proportional to droplet radius L (where the slope η / γ measures the inverse capillary velocity, typically ~0.1 to 10 s/μm for cellular condensates).
2. Concentration-Dependent Binodal Threshold (c_sat) and Optogenetic (OptoDroplet / Corelet) Reversibility: Quantitative titration (or blue-light-induced oligomerization using Cry2- or iLID-tagged disordered domains, developed by Brangwynne and colleagues) demonstrates a sharp saturation concentration c_sat: below c_sat no droplets form; above c_sat the dilute-phase concentration remains buffered at c_sat while the volume fraction of droplets grows linearly, and turning off the light (or diluting the cell) dissolves the droplets reversibly.
3. Internal Mobility via Half-Droplet FRAP: Photobleaching half of a single droplet confirms rapid internal mixing (Seconds) distinct from simple boundary exchange.
4. Sensitivity to Post-Translational Valency Tuning: Phosphorylation, acetylation, or RNase digestion predictably shifts the binodal boundary. (Note: aliphatic alcohols like 1,6-hexanediol melt weak hydrophobic condensates in vitro, but must be interpreted cautiously in live cells due to membrane and kinase toxicity).

Why does physiological ATP (5–10 mM) dissolve protein condensates even without ATPase enzymes?

An individual cell requires only micromolar (10–100 μM) concentrations of ATP to saturate the active sites (K_m) of virtually all cellular kinases, myosin/kinesin motors, and chaperones—raising the longstanding puzzle of why all living cells maintain cytoplasmic ATP at 5 to 10 millimolar (5,000–10,000 μM). In 2017, Avinash Patel, Anthony Hyman, and colleagues discovered that at 5–10 mM, the amphiphilic ATP molecule—consisting of a hydrophobic aromatic adenine ring linked to a hydrophilic, negatively charged triphosphate tail—functions as a chemical hydrotrope (even when replaced by non-hydrolyzable AMP-PNP or ATP-γ-S!). By transiently binding via π–π and cation–π contacts to exposed aromatic/hydrophobic stickers on intrinsically disordered proteins (FUS, TAF15, hnRNPA3, TDP-43) while presenting its hydrated triphosphate charge to water, millimolar ATP keeps the crowded 200–300 mg/mL cytoplasm soluble and prevents age-related pathological protein aggregation when cellular ATP levels decline.

1. Brangwynne, C. P., Eckmann, C. R., Courson, D. S., Rybarska, A., Hoege, C., Gharakhani, J., Jülicher, F., & Hyman, A. A. (2009). "Germline P granules are liquid droplets that localize by controlled dissolution/condensation." Science, 324(5935), 1729–1732. https://doi.org/10.1126/science.1172046
2. Li, P., Banjade, S., Cheng, H. C., Kim, S., Chen, B., Guo, L., ... & Rosen, M. K. (2012). "Phase transitions in the assembly of multivalent signalling proteins." Nature, 483(7389), 336–340. https://doi.org/10.1038/nature10879
3. Patel, A., Lee, H. O., Jawerth, L., Maharana, S., Jahnel, M., Hein, M. Y., ... & Hyman, A. A. (2015). "A liquid-to-solid phase transition of the ALS protein FUS accelerated by disease mutation." Cell, 162(5), 1066–1077. https://doi.org/10.1016/j.cell.2015.07.047
4. Wang, J., Choi, J. M., Holehouse, A. S., Lee, H. O., Zhang, X., Jahnel, M., ... & Hyman, A. A. (2018). "A molecular grammar governing the driving forces for phase separation of prion-like RNA binding proteins." Cell, 174(3), 688–699. https://doi.org/10.1016/j.cell.2018.06.006