Micellar theory

Starch granules of corn

The micellar theory of Carl Nägeli was developed from his detailed study of starch granules in 1858.^[2] Amorphous substances such as starch and cellulose were proposed to consist of building blocks, packed in a loosely crystalline array to form what he later termed “micelles”. Water could penetrate between the micelles, and new micelles could form in the interstices between old micelles. The swelling of starch grains and their growth was described by a molecular-aggregate model, which he also applied to the cellulose of the plant cell wall. The modern usage of 'micelle' refers strictly to lipids, but its original usage clearly extended to other types of biomolecule, and this legacy is reflected to this day in the description of milk as being composed of 'casein micelles'.

Colloidal phase separation theory

Glycogen granules in Spermiogenesis in Pleurogenidae (Digenea)

The concept of intracellular colloids as an organizing principle for the compartmentalization of living cells dates back to the end of the 19th century, beginning with William Bate Hardy and Edmund Beecher Wilson who described the cytoplasm (then called 'protoplasm') as a colloid.^[3][4] Around the same time, Thomas Harrison Montgomery Jr. described the morphology of the nucleolus, an organelle within the nucleus, which has subsequently been shown to form through intracellular phase separation.^[5] WB Hardy linked formation of biological colloids with phase separation in his study of globulins, stating that: "The globulin is dispersed in the solvent as particles which are the colloid particles and which are so large as to form an internal phase",^[6] and further contributed to the basic physical description of oil-water phase separation.^[7]

Colloidal phase separation as a driving force in cellular organisation appealed strongly to Stephane Leduc, who wrote in his influential 1911 book The Mechanism of Life: "Hence the study of life may be best begun by the study of those physico-chemical phenomena which result from the contact of two different liquids. Biology is thus but a branch of the physico-chemistry of liquids; it includes the study of electrolytic and colloidal solutions, and of the molecular forces brought into play by solution, osmosis, diffusion, cohesion, and crystallization."^[8]

The primordial soup theory of the origin of life, proposed by Alexander Oparin in Russian in 1924 (published in English in 1936)^[9] and by J.B.S. Haldane in 1929,^[10] suggested that life was preceded by the formation of what Haldane called a "hot dilute soup" of "colloidal organic substances", and which Oparin referred to as 'coacervates' (after de Jong^[11]) – particles composed of two or more colloids which might be protein, lipid or nucleic acid. These ideas strongly influenced the subsequent work of Sidney W. Fox on proteinoid microspheres.

Support from other disciplines

Micelle caseine

When cell biologists largely abandoned colloidal phase separation, it was left to relative outsiders – agricultural scientists and physicists – to make further progress in the study of phase separating biomolecules in cells.

Beginning in the early 1970s, Harold M Farrell Jr. at the US Department of Agriculture developed a colloidal phase separation model for milk casein micelles that form within mammary gland cells before secretion as milk.^[12]

Also in the 1970s, physicists Tanaka & Benedek at MIT identified phase-separation behaviour of gamma-crystallin proteins from lens epithelial cells and cataracts in solution,^{[13][14][15][16][17]} which Benedek referred to as 'protein condensation'.^[18]

Lens epithelium containing crystallin. Hand-book of physiology (1892)

In the 1980s and 1990s, Athene Donald's polymer physics lab in Cambridge extensively characterised phase transitions / phase separation of starch granules from the cytoplasm of plant cells, which behave as liquid crystals.^{[19][20][21][22][23][24][25][26]}

In 1991, Pierre-Gilles de Gennes received the Nobel Prize in Physics for developing a generalized theory of phase transitions with particular applications to describing ordering and phase transitions in polymers.^[27] Unfortunately, de Gennes wrote in Nature that polymers should be distinguished from other types of colloids, even though they can display similar clustering and phase separation behaviour,^[28] a stance that has been reflected in the reduced usage of the term colloid to describe the higher-order association behaviour of biopolymers in modern cell biology and molecular self-assembly.

Phase separation revisited

Advances in confocal microscopy at the end of the 20th century identified proteins, RNA or carbohydrates localising to many non-membrane bound cellular compartments within the cytoplasm or nucleus which were variously referred to as 'puncta/dots',^{[29][30][31][32]} 'signalosomes',^[33][34] 'granules',^[35] 'bodies', 'assemblies',^[32] 'paraspeckles', 'purinosomes',^[36] 'inclusions', 'aggregates' or 'factories'. During this time period (1995-2008) the concept of phase separation was re-borrowed from colloidal chemistry & polymer physics and proposed to underlie both cytoplasmic and nuclear compartmentalization.^{[37][38][39][40][41][42][43][44][45][46]}

Since 2009, further evidence for biomacromolecules undergoing intracellular phase transitions (phase separation) has been observed in many different contexts, both within cells and in reconstituted in vitro experiments.^{[47][48][49][50][51][52][53]}

The newly coined term "biomolecular condensate"^[54] refers to biological polymers (as opposed to synthetic polymers) that undergo self assembly via clustering to increase the local concentration of the assembling components, and is analogous to the physical definition of condensation.^[55][54]

In physics, condensation typically refers to a gas–liquid phase transition.

In biology the term 'condensation' is used much more broadly and can also refer to liquid–liquid phase separation to form colloidal emulsions or liquid crystals within cells, and liquid–solid phase separation to form gels,^[1] sols, or suspensions within cells as well as liquid-to-solid phase transitions such as DNA condensation during prophase of the cell cycle or protein condensation of crystallins in cataracts.^[18] With this in mind, the term 'biomolecular condensates' was deliberately introduced to reflect this breadth (see below). Since biomolecular condensation generally involves oligomeric or polymeric interactions between an indefinite number of components, it is generally considered distinct from formation of smaller stoichiometric protein complexes with defined numbers of subunits, such as viral capsids or the proteasome – although both are examples of spontaneous molecular self-assembly or self-organisation.

Mechanistically, it appears that the conformational landscape^[56] (in particular, whether it is enriched in extended disordered states) and multivalent interactions between intrinsically disordered proteins (including cross-beta polymerisation),^[57] and/or protein domains that induce head-to-tail oligomeric or polymeric clustering,^[58] might play a role in phase separation of proteins.

Cytoplasmic condensates

• Lewy bodies
• Stress granule
• P-body
• Germline P-granules – oskar
• Starch granules
• Glycogen granules^[59]
• Frodosomes (Dact1)^[60]
• Corneal lens formation and cataracts^{[13][14][15][16][17][18]}
• Other cytoplasmic inclusions such as pigment granules or cytoplasmic crystals
• Purinosomes^[36]
• Misfolded protein aggregation such as amyloid fibrils or mutant Haemoglobin S (HbS) fibres in sickle cell disease
• Signalosomes, such as the supramolecular assemblies in the Wnt signaling pathway.^[61][62]
• It can also be argued that cytoskeletal filaments form by a polymerisation process similar to phase separation, except ordered into filamentous networks instead of amorphous droplets or granules.
• Bacteria Ribonucleoprotein Bodies (BR-bodies)- In recent studies it has been shown that bacteria RNA degradosomes can assemble into phase‐separated structures, termed bacterial ribonucleoprotein bodies (BR‐bodies), with many analogous properties to eukaryotic processing bodies and stress granules.^[63]
• FLOE1 granules: FLOE1 is a prion-like seed-specific protein that controls plant seed germination via phase separation into biomolecular condensates.^[64]

Nuclear condensates

Formation and examples of nuclear bodies

• Nucleolus^[51]
• Cajal body
• Paraspeckle
• Synaptonemal complex

Other nuclear structures including heterochromatin form by mechanisms similar to phase separation, so can also be classified as biomolecular condensates.

Plasma membrane associated condensates

• Membrane protein, or membrane-associated protein, clustering at neurological synapses, cell-cell tight junctions, or other membrane domains.^[65]

Secreted extracellular condensates

• Secreted thyroglobulin colloid and colloid nodules of the thyroid gland
• Secreted casein ‘micelles’ of the mammary gland
• Serum albumin and globulins
• Secreted lysozyme^[66][42]

Lipid-enclosed organelles and lipoproteins are not considered condensates

Typical organelles or endosomes enclosed by a lipid bilayer are not considered biomolecular condensates. In addition, lipid droplets are surrounded by a lipid monolayer in the cytoplasm, or in milk, or in tears,^[67] so appear to fall under the 'membrane bound' category. Finally, secreted LDL and HDL lipoprotein particles are also enclosed by a lipid monolayer. The formation of these structures involves phase separation to from colloidal micelles or liquid crystal bilayers, but they are not classified as biomolecular condensates, as this term is reserved for non-membrane bound organelles.

Liquid biomolecular condensates

Liquid–liquid phase separation (LLPS) generates a subtype of colloid known as an emulsion that can coalesce to from large droplets within a liquid. Ordering of molecules during liquid–liquid phase separation can generate liquid crystals rather than emulsions. In cells, LLPS produces a liquid subclass of biomolecular condensate that can behave as either an emulsion or liquid crystal.

The term biomolecular condensates was introduced in the context of intracellular assemblies as a convenient and non-exclusionary term to describe non-stoichiometric assemblies of biomolecules.^[54] The choice of language here is specific and important. It has been proposed that many biomolecular condensates form through liquid–liquid phase separation (LLPS) to form colloidal emulsions or liquid crystals in living organisms, as opposed to liquid–solid phase separation to form crystals/aggregates in gels,^[1] sols or suspensions within cells or extracellular secretions.^[68] However, unequivocally demonstrating that a cellular body forms through liquid–liquid phase separation is challenging,^{[69][47][70][71]} because different material states (liquid vs. gel vs. solid) are not always easy to distinguish in living cells.^[72][73] The term "biomolecular condensate" directly addresses this challenge by making no assumption regarding either the physical mechanism through which assembly is achieved, nor the material state of the resulting assembly. Consequently, cellular bodies that form through liquid–liquid phase separation are a subset of biomolecular condensates, as are those where the physical origins of assembly are unknown. Historically, many cellular non-membrane bound compartments identified microscopically fall under the broad umbrella of biomolecular condensates.

In physics, phase separation can be classified into the following types of colloid, of which biomolecular condensates are one example:

In biology, the most relevant forms of phase separation are either liquid–liquid or liquid–solid, although there have been reports of gas vesicles surrounded by a phase separated protein coat in the cytoplasm of some microorganisms.^[76]

Wnt signalling

One of the first discovered examples of a highly dynamic intracellular liquid biomolecular condensate with a clear physiological function were the supramolecular complexes (Wnt signalosomes) formed by components of the Wnt signaling pathway.^[44][61][62] The Dishevelled (Dsh or Dvl) protein undergoes clustering in the cytoplasm via its DIX domain, which mediates protein clustering (polymerisation) and phase separation, and is important for signal transduction.^{[29][30][31][32][34][44]} The Dsh protein functions both in planar polarity and Wnt signalling, where it recruits another supramolecular complex (the Axin complex) to Wnt receptors at the plasma membrane. The formation of these Dishevelled and Axin containing droplets is conserved across metazoans, including in Drosophila, Xenopus, and human cells.

P granules

Another example of liquid droplets in cells are the germline P granules in Caenorhabditis elegans.^[68][47] These granules separate out from the cytoplasm and form droplets, as oil does from water. Both the granules and the surrounding cytoplasm are liquid in the sense that they flow in response to forces, and two of the granules can coalesce when they come in contact. When (some of) the molecules in the granules are studied (via fluorescence recovery after photobleaching), they are found to rapidly turnover in the droplets, meaning that molecules diffuse into and out of the granules, just as expected in a liquid droplet. The droplets can also grow to be many molecules across (micrometres)^[47] Studies of droplets of the Caenorhabditis elegans protein LAF-1 in vitro^[77] also show liquid-like behaviour, with an apparent viscosity ${\displaystyle \eta \sim 10}$Pa s. This is about a ten thousand times that of water at room temperature, but it is small enough to enable the LAF-1 droplets to flow like a liquid. Generally, interaction strength (affinity)^[78] and valence (number of binding sites)^[53] of the phase separating biomolecules influence their condensates viscosity, as well as their overall tendency to phase separate.

Design and control

Despite the dynamic nature and lack of binding specificity that govern the formation of biomolecular condensates, synthetic condensates can still be engineered to exhibit different behaviors. One popular way to conceptualize condensate interactions and aid in design is through the "sticker-spacer" framework.^[85] Multivalent interaction sites, or "stickers", are separated by "spacers", which provide the conformational flexibility and physically separate individual interaction modules from one another. Proteins regions identified as ‘stickers’ usually consist of Intrinsically Disordered Regions (IDRs) that act as "sticky" biopolymers via short patches of interacting residues patterned along their unstructured chain, which collectively promote LLPS.^[86] By modifying the sticker-spacer framework, i.e. the polypeptide and RNA sequences as well as their mixture compositions, the material properties (viscous and elastic regimes) of condensates can be tuned to design novel condensates.^[87]

Other tools outside of tuning the sticker-spacer framework can be used to give new functionality and to allow for high temporal and spatial control over synthetic condensates. One way to gain temporal control over the formation and dissolution of biomolecular condensates is by using optogenetic tools. Several different systems have been developed which allow for control of condensate formation and dissolution which rely on chimeric protein expression, and light or small molecule activation.^[88] In one system,^[89] proteins are expressed in a cell which contain light-activated oligomerization domains fused to IDRs. Upon irradiation with a specific wavelength of light, the oligomerization domains bind each other and form a ‘core’, which also brings multiple IDRs close together because they are fused to the oligomerization domains. The recruitment of multiple IDRs effectively creates a new biopolymer with increased valency. This increased valency allows for the IDRs to form multivalent interactions and trigger LLPS. When the activation light is stopped, the oligomerization domains disassemble, causing the dissolution of the condensate. A similar system^[90] achieves the same temporal control of condensate formation by using light-sensitive ‘caged’ dimerizers. In this case, light-activation removes the dimerizer cage, allowing it to recruit IDRs to multivalent cores, which then triggers phase separation. Light-activation of a different wavelength results in the dimerizer being cleaved, which then releases the IDRs from the core and consequentially dissolves the condensate. This dimerizer system requires significantly reduced amounts of laser light to operate, which is advantageous because high intensity light can be toxic to cells.

Optogenetic systems can also be modified to gain spatial control over the formation of condensates. Multiple approaches have been developed to do so. In one approach,^[91] which localizes condensates to specific genomic regions, core proteins are fused to proteins such as TRF1 or catalytically dead Cas9, which bind specific genomic loci. When oligomerization is trigger by light activation, phase separation is preferentially induced on the specific genomic region which is recognized by fusion protein. Because condensates of the same composition can interact and fuse with each other, if they are tethered to specific regions of the genome, condensates can be used to alter the spatial organization of the genome, which can have effects on gene expression.^[91]

As biochemical reactors

Synthetic condensates offer a way to probe cellular function and organization with high spatial and temporal control, but can also be used to modify or add functionality to the cell. One way this is accomplished is by modifying the condensate networks to include binding sites for other proteins of interest, thus allowing the condensate to serve as a scaffold for protein release or recruitment.^[92] These binding sites can be modified to be sensitive to light activation or small molecule addition, thus giving temporal control over the recruitment of a specific protein of interest. By recruiting specific proteins to condensates, reactants can be concentrated to increase reaction rates or sequestered to inhibit reactivity.^[93] In addition to protein recruitment, condensates can also be designed which release proteins in response to certain stimuli. In this case, a protein of interest can be fused to a scaffold protein via a photocleavable linker. Upon irradiation, the linker is broken, and the protein is released from the condensate. Using these design principles, proteins can either be released to, or sequestered from, their native environment, allowing condensates to serve as a tool to alter the biochemical activity of specific proteins with a high level of control.^[92]

Coarse-grained molecular models

Molecular dynamics and Monte Carlo simulations have been extensively used to gain insights into the formation and the material properties of biomolecular condensates.^[96] Although molecular models of different resolution have been employed,^[97][98][99] modelling efforts have mainly focused on coarse-grained models of intrinsically disordered proteins, wherein amino acid residues are represented by single interaction sites. Compared to more detailed molecular descriptions, residue-level models provide high computational efficiency, which enables simulations to cover the long length and time scales required to study phase separation. Moreover, the resolution of these models is sufficiently detailed to capture the dependence on amino acid sequence of the properties of the system.^[96]

Several residue-level models of intrinsically disordered proteins have been developed in recent years. Their common features are (i) the absence of an explicit representation of solvent molecules and salt ions, (ii) a mean-field description of the electrostatic interactions between charged residues (see Debye–Hückel theory), and (iii) a set of "stickiness" parameters which quantify the strength of the attraction between pairs of amino acids. In the development of most residue-level models, the stickiness parameters have been derived from hydrophobicity scales^[100] or from a bioinformatic analysis of crystal structures of folded proteins.^[101][102] Further refinement of the parameters has been achieved through iterative procedures which maximize the agreement between model predictions and a set of experiments,^{[103][104][105][106][107][108]} or by leveraging data obtained from all-atom molecular dynamics simulations.^[102]

Residue-level models of intrinsically disordered proteins have been validated by direct comparison with experimental data, and their predictions have been shown to be accurate across diverse amino acid sequences.^{[103][104][105][102][107][109][108]} Examples of experimental data used to validate the models are radii of gyration of isolated chains and saturation concentrations, which are threshold protein concentrations above which phase separation is observed.^[110]

Although intrinsically disordered proteins often play important roles in condensate formation,^[111] many biomolecular condensates contain multi-domain proteins constituted by folded domains connected by intrinsically disordered regions.^[112] Current residue-level models are only applicable to the study of condensates of intrinsically disordered proteins and nucleic acids.^{[113][102][114][115][116][108]} Including an accurate description of the folded domains in these models will considerably widen their applicability.^[117][96]