Liquid-Liquid Phase Separation

Why LLPS matters

• Reorganizes textbook cell biology. Before 2009, cells were drawn as a network of membrane-bound organelles. LLPS adds a second class — membraneless condensates that compartmentalize biochemistry through demixing rather than lipid bilayers. Roughly 30 to 40 distinct condensates have been catalogued in mammalian cells, including the nucleolus, Cajal bodies, nuclear speckles, P granules, P-bodies, stress granules, and signaling clusters at receptor microclusters.
• Concentrates reagents 10x to 100x. Inside a condensate, the dense phase contains the multivalent species at concentrations 10 to 100 times higher than the surrounding cytoplasm. For an enzyme cascade to work efficiently, this is a powerful catalytic enhancement. The nucleolus uses LLPS to bring together rDNA, RNA Pol I, processing factors, and ribosomal proteins to assemble ~7500 ribosomes per minute in a HeLa cell.
• Buffers protein concentration. Above the critical concentration, additional protein partitions into the dense phase rather than raising the dilute-phase concentration. The cytoplasmic free pool stays buffered at C_sat (saturation concentration) regardless of total expression. This makes condensates natural concentration regulators — a feature exploited at signaling hubs to set thresholds.
• Enables fast, reversible reorganization. Stress granules form within 5 to 10 minutes of arsenite or heat shock and dissolve within 1 to 2 hours of stress relief. No protein synthesis is required for either step — the dynamics are purely thermodynamic. Dissolution returns translation factors and mRNAs to the cytoplasm rapidly.
• Drives transcriptional control at super-enhancers. Master transcription factors and Mediator coactivator condense at super-enhancer loci, concentrating RNA Pol II and drug-targetable BRD4 into ~300 nm transcriptional condensates. The Young lab (Sabari et al., Science 2018) showed that BRD4 inhibition by JQ1 dissolves these condensates and coordinately downregulates super-enhancer-driven oncogenes — a possible mechanism for selective cancer dependency.
• Pathological aging causes ALS/FTD. FUS, TDP-43, hnRNPA1, EWSR1, TAF15 — many ALS/FTD-causing proteins phase separate normally and have prion-like low-complexity domains. Disease mutations accelerate the liquid-to-gel-to-fibril transition. ALS aggregates in postmortem motor neurons reflect persistent stress granules that have hardened over years.
• New drug-discovery target class. Drugs that selectively partition into condensates can deliver therapy at 10 to 100x higher local concentration than in the surrounding cytoplasm. Cisplatin, doxorubicin, and tamoxifen have all been shown to partition into specific condensates, contributing to their efficacy and selectivity. Companies (Dewpoint Therapeutics, Faze Medicines) have raised hundreds of millions to develop condensate-modulating small molecules.

Common misconceptions

• Phase separation is just protein aggregation. No — aggregates are solid, irreversible, and molecularly amorphous. LLPS produces liquid droplets that are reversible, exchange components with the surrounding cytoplasm in seconds (FRAP recovery), and fuse on contact like oil drops. Aggregation is what happens when a condensate ages pathologically.
• One protein is enough to drive phase separation. Phase separation requires multivalent interactions — typically multiple weak binding sites per molecule. Single-domain proteins that bind a single partner cannot drive LLPS no matter how strong the interaction. The hallmark architecture is a long IDR with repeated motifs (FUS prion-like domain has dozens of Tyr-Gly-Gly repeats), or a folded multivalent scaffold like SH3 connected by linkers to multiple PRMs.
• 1,6-hexanediol is a specific test for LLPS. Hexanediol disrupts hydrophobic interactions broadly and dissolves many condensates within 30 to 60 seconds — but it also disrupts the cytoskeleton, nuclear pore selectivity, and membrane traffic. A hexanediol-sensitive structure is consistent with LLPS, not diagnostic. Combine with FRAP, fusion, and in vitro reconstitution.
• RNA is just cargo in condensates. RNA actively participates as a phase-separation driver. RNA-binding proteins like FUS use both protein-protein and protein-RNA contacts; the RNA itself contributes valence through its multiple binding sites. RNase treatment can either dissolve a condensate (if RNA is a scaffold) or condense it (if RNA was buffering by competing for binding sites).
• All membraneless compartments are phase-separated. Some are; some aren't. Nucleoli and stress granules pass the criteria. Heterochromatin domains, polycomb bodies, and the centrosome have liquid character but also include scaffolded protein-protein interactions; the field distinguishes pure phase separation from polymer-polymer phase separation, percolation, and scaffold-mediated assembly. The term "biomolecular condensate" is now preferred over "phase-separated body" to remain agnostic about mechanism.
• Phase separation explains everything. A 2022 backlash (Musacchio's commentary in EMBO Journal and others) emphasized that some claims of LLPS in cells are based on weak evidence — fluorescence enrichment alone, hexanediol sensitivity alone — and that the field has overreached. Robust evidence requires reconstitution, FRAP, fusion behavior, and concentration-dependent demixing.

How phase separation works

The thermodynamics is borrowed from polymer physics. Two well-mixed components with weak attractive interactions will demix into two phases if the attraction is strong enough to overcome the entropy of mixing. The "stickers and spacers" framework, formalized by Rohit Pappu, Cliff Brangwynne, and others around 2017, treats multivalent IDR proteins as polymers with stickers (the binding motifs — aromatic Tyr/Phe/Trp residues for π-π stacking, Arg for cation-π, charged blocks for electrostatic) connected by spacers (the rest of the IDR sequence, which tunes the polymer's flexibility and excluded volume). Above a saturation concentration C_sat — typically 1 to 10 µM in vitro for FUS, hnRNPA1, TDP-43, and similar proteins — the network of weak sticker-sticker contacts crosses a percolation threshold and the system demixes. Below C_sat, the protein remains uniformly soluble.

The interaction types are diverse but all weak. π-π stacking between aromatic Tyr/Phe/Trp residues produces ~1 to 2 kT per contact. Cation-π between Arg and aromatic residues is a special case favored in many phase-separating IDRs. Charge-charge interactions between blocky polyelectrolytes drive complex coacervation (the wet biology version of the rubber-cement chemistry textbooks discuss). Hydrogen bonds between Gln/Asn-rich tracts (the basis of prion-like domains in yeast Sup35 and mammalian FUS) provide additional valence. None of these is a strong, specific binding event — the strength comes from many weak contacts in parallel, the way Velcro hooks bind one at a time but the patch holds firmly.

In cells, regulation of phase separation happens by tuning the multivalence. Phosphorylation of FUS's prion-like domain, of TDP-43, of nuclear-pore FG repeats, or of SUMO-SIM scaffolds in PML bodies all reduce the multivalent attraction and dissolve condensates. Methylation of Arg residues (e.g., FUS arginines) reduces cation-π and dissolves. RNA binding adds valence (concentrating partners to a 1D scaffold) and can promote condensation; competing RNA can compete out scaffolding and dissolve. Stress responses use these levers — eIF2α phosphorylation arrests translation, ribosomes detach from mRNA, exposed mRNA + RNA-binding proteins seed stress granules within minutes. When stress relieves, ATP-dependent chaperones (Hsp70, Hsp40) and DEAD-box RNA helicases dissolve the granules within an hour.

P-bodies vs stress granules vs nucleoli vs Cajal bodies

Property	P-bodies	Stress granules	Nucleoli	Cajal bodies
Location	Cytoplasm	Cytoplasm	Nucleus (largest body)	Nucleus (1–5 per cell)
Size	200–500 nm	200–800 nm	1–5 µm (~10% of nuclear vol)	~0.5 µm
Constitutive vs induced	Constitutive	Induced (heat, arsenite, virus)	Constitutive (G1, S, G2)	Constitutive in active cells
Marker proteins	Dcp1/Dcp2, Xrn1, Lsm1-7	G3BP1, TIA-1, eIF3	Nucleophosmin, fibrillarin	Coilin, SMN
Cargo	Decapped/decay mRNAs	Stalled 48S complexes, mRNAs	rDNA, pre-rRNA, ribosomal proteins	snRNPs, scaRNAs
Function	mRNA decay and storage	Translation arrest under stress	Ribosome biogenesis (~7500/min)	snRNP/snoRNP maturation
Disassembly	Slow turnover	1–2 h after stress relief	Mitotic disassembly	Persists; varies with activity
Disease relevance	Reduced in cancer	Persistent in ALS/FTD (TDP-43)	Hypertrophic in cancer	SMN reduced in spinal muscular atrophy

Liquid condensate vs gel vs fibrillar aggregate

Property	Liquid condensate	Hydrogel/maturation	Solid fibrillar aggregate
Internal mobility	FRAP recovery in seconds	FRAP recovery in minutes (partial)	No FRAP recovery
Shape	Spherical (surface tension)	Mostly round, some asymmetry	Often asymmetric, fibrous
Fusion behavior	Fuses to spherical droplet	Fuses slowly or partially	Does not fuse
Reversibility	Reversible (solute-conc dependent)	Slowly reversible	Largely irreversible
Hexanediol sensitivity	Dissolves in 30–60 s	Partial	Resistant
Underlying interactions	Many weak, dynamic	More cross-linked, partial structure	β-sheet stacking (amyloid)
Cellular role	Functional compartment	Possibly storage; toxic precursor	Pathological (ALS, FTD, AD, PD)
Diagnostic	Liquid droplet behavior	Slowed dynamics	Thioflavin S/T positive

Famous experiments

• Brangwynne & Hyman P granules (2009). Cliff Brangwynne, working with Tony Hyman at MPI-CBG Dresden, showed that P granules in C. elegans 1-cell embryos are spherical liquid droplets that fuse on contact, drip under gravity in centrifuged embryos, and exchange protein with the cytoplasm in seconds. Published in Science 324: 1729–1732. Citations now exceed 4500. Reframed cell biology around phase separation.
• Rosen lab engineered SH3-PRM system (2012). Mike Rosen and colleagues at UT Southwestern showed that purified linear repeats of SH3 domains and proline-rich motifs (PRMs) phase separate in vitro above a critical concentration, with droplet formation requiring at least 3 valences. The paper (Nature 483: 336–340) established the in vitro biophysics and the multivalence requirement.
• FUS prion-like domain phase separation (Kato et al. 2012; Patel et al. 2015; Murakami et al. 2015). Showed that FUS's low-complexity, Gly-Tyr-Ser-Arg-rich N-terminal domain phase separates in vitro at micromolar concentrations and ages from liquid to gel to fibril. ALS-causing FUS mutations accelerate aging. Connected LLPS to neurodegeneration.
• Sabari, Young transcriptional condensates (2018). Benjamin Sabari and Richard Young at the Whitehead showed that master transcription factors (OCT4, MED1, BRD4) form ~300-nm transcriptional condensates at super-enhancers. JQ1 inhibition dissolves BRD4 condensates and selectively downregulates super-enhancer-driven oncogenes — possibly explaining the long-puzzling selectivity of BET-inhibitor cancer therapy. Science 361: eaar3958.
• Optogenetic Corelet and OptoDroplet (Bracha, Brangwynne 2018; Shin, Brangwynne 2017). Engineered Cry2 light-activated condensation systems let experimenters trigger phase separation in living cells on a 1 to 10 second timescale and dissolve them in similar times. Provides causal manipulation rather than correlation, used to test the function of specific condensates and to titrate concentrations across the phase boundary in real time.