Our understanding of cells has progressed rapidly in recent years, mainly because of technological advances. Modern technology now allows us to observe molecular processes in living cells with high spatial and temporal resolution. At the same time, we are beginning to compile the molecular parts list of cells. However, how all these parts work together to yield complex cellular behavior is still unclear. In addition, the established paradigm of molecular biology, which sees proteins as well-folded enzymes that undergo specific lock-and-key type interactions, is increasingly being challenged. In fact, it is now becoming clear that many proteins do not fold into three-dimensional structures and additionally show highly promiscuous binding behavior. Furthermore, proteins function in collectives and form condensed phases with different material properties, such as liquids, gels, glasses or filaments. Here, I examine emerging evidence that the formation of macromolecular condensates is a fundamental principle in cell biology. I further discuss how different condensed states of living matter regulate cellular functions and decision-making and ensure adaptive behavior and survival in times of cellular crisis.
The paradigm of molecular biology has propelled us forward for many decades; it has generated our current view of biological systems: genes code for proteins that assemble into complex nanomachines. These machines tirelessly perform their functions: adding and removing phosphates, pumping ions against gradients, and churning out metabolites with amazing speed and accuracy. Every one of these machines is an independent agent that dedicatedly performs a unique function for which it has been honed by evolution. Changes in cell function are achieved by altering the activity of these small protein agents, either through changes in gene expression, protein degradation or post-translational modifications. But is this view sufficient to explain the complex behavior of living cells?
The numerous molecular processes in cells yield complex cellular structures, movements and regulatory decisions. We know most of the molecules that underlie these structures, behaviors and decisions, but we often do not know how all these molecules interact and function together. Thus, there is a considerable gap in our mechanistic understanding of how cells work, especially on the mesoscopic level – the level between the nanoscale of molecular machines and the microscale of cells. At the same time, there is an increasing realization that many macromolecules do not function as individuals but in collectives. Could this collective macromolecular behavior be the key for understanding the complex behavior of living cells?
For a chemist, it does not come as a surprise that molecules act in collectives. In fact, a whole subfield of chemistry called colloidal chemistry is dedicated to studying the behavior of many particles. The word colloid is derived from ‘kolla’, the Greek word for glue, because some of the original colloidal solutions were glues. In chemistry, a colloid is a form of matter where one substance of dispersed particles is suspended in another substance. Milk, for example, is a colloidal emulsion of two liquids: lipid-protein drops suspended in water. Cheese is a colloidal gel where a liquid is suspended in a solid-like protein matrix. Not surprisingly, colloids play very important roles in food chemistry. However, the trained biologist of today has almost no exposure to the science of colloids, although, as I will argue, this topic is eminently important for understanding living cells.
In this Review, I introduce the newly emerging biochemistry and biophysics of macromolecular colloids. I examine accumulating evidence on how the collective behavior of macromolecules and their condensation into distinct physical phases is a key principle that affects various aspects of cell biology: intracellular organization, metabolism, signaling, and stress adaptation.
Protein colloids: the collective behavior of protein crowds
Evidence for the collective behavior of proteins has been around for many decades, but its significance for cell biology has only become clear in the past couple of years. Early experiments to probe the colloidal properties of proteins were performed with folded and easy-to-purify model proteins, such as lysozyme or crystallins (Dumetz et al., 2008b; Galkin and Vekilov, 2000; Vekilov, 2004). Researchers found that these proteins can exist in many different physical states; they can be single molecules in dilute solutions, but they can also interact and form several condensed phases: crystals, gels, glasses or dense liquids (Vekilov, 2010) (Figs 1 and 2). Importantly, these condensed matter states have very different physical properties (Fig. 2, Box 1). As I will discuss below, these differences in physical properties are biologically important. Liquids, for example, allow free diffusion of molecular components and thus could be reaction vessels for concentrated biochemistry, whereas in gels or glasses, molecular diffusion is restricted, thus providing a means for suppressing biochemical reactions.
Liquid. In a liquid, there is only order over short distances. Liquids are dynamic and the molecules in liquids rearrange quickly and move around. When forces are applied to deform a liquid, the molecules move away from each other. There is no memory of the initial configuration of molecules.
Solid. In solids, positional order exists over short and long distances. Molecules in a solid are confined to a small cage created by the neighboring molecules. Application of a force leads to the buildup of deformations until the force is balanced by elastic stresses. When the force is removed, the solid relaxes back to its initial state. Thus, the system has a memory of the initial molecule configuration.
Glass. A glass is an amorphous (non-crystalline) solid in which the molecules are jammed together. As a consequence, the molecules diffuse very slowly compared with molecules in a simple liquid. Glasses are non-equilibrium solids that occur as a result of dynamical molecular arrest, whereas a crystal is an equilibrium solid.
Gel. A gel is a percolated system-spanning network of interacting molecules. Because gels are cross-linked networks, they exhibit no flow when in steady state. A hydrogel is a gel composed of a network of hydrophilic polymeric molecules pervaded by water. Hydrogels are highly absorbent and can contain over 90% water.
Crystal. A crystal is a solid material whose molecules are arranged in a highly ordered structure, forming a crystal lattice that extends in all directions. As a consequence, crystals often have clearly visible geometrical shapes. Crystal formation is frequently preceded by LLPS.
Liquid–liquid phase separation (LLPS). In LLPS, an initially homogenous solution of supersaturated molecules spontaneously ‘de-mixes’ into a dilute and a dense liquid phase that then stably coexist. The two phases are separated by a phase boundary that allows exchange of molecules between the two phases. The dense liquid phase is often metastable, and forms crystals over time or becomes trapped in a gel- or glass-like state (Fig. 2).
Supersaturation. Supersaturation is the state of a solution that contains more dissolved molecules than could be dissolved by the solvent under normal circumstances. When the maximum number of molecules is dissolved, adding more molecules causes the dissolved molecules to precipitate. Thus, supersaturated solutions are unstable and will undergo LLPS, crystallization, gel or glass formation (Fig. 2).
Condensation of folded proteins such as lysozyme into liquid, glasses or gels requires that these proteins self-associate. This primarily occurs through transient and relatively weak electrostatic interactions that often require high protein concentrations and changes in salt or pH (Dumetz et al., 2008a; Vekilov, 2010). Self-association can lead to a range of assembled species that exchange with the solution dynamically. These assemblies are typically easily reversed by moving to lower protein concentrations and/or shifting the pH or ionic strength of the solution (Vekilov, 2010).
However, not all condensed phases made up of proteins are easily reversed. In some cases, proteins associate so strongly that it is effectively irreversible on practical time scales and concentration ranges. This means that the resulting assemblies do not dissociate appreciably upon dilution or upon shifts in solution pH or ionic strength. The formation of such assemblies or aggregates typically involves partial unfolding of the constituent proteins. Partially unfolded proteins can generate a range of different aggregated structures (Knowles et al., 2014; Yoshimura et al., 2012). These include highly ordered amyloid fibrils (crystal-like properties) or amorphous aggregates (with material properties of gels or glasses). These assemblies can often only be dissolved through chemical denaturants (for example, urea or ionic detergents such as SDS).
Although early studies with folded model proteins have provided important insights into the phase behavior of proteins, it was unclear whether proteins show colloidal behavior in cells and whether protein colloids affect biological functions and are subject to evolutionary selection. This has changed recently with the discovery that cells contain many membrane-less compartments that are condensed phases of living matter (see below).
The colloidal nature of the intracellular environment
Eukaryotic cells contain two different types of compartments: membrane-bound and membrane-less compartments. Membrane-bound compartments, such as the Golgi complex, use lipid bilayers to separate biomolecules. It is less clear how membrane-less compartments separate molecular components. Examples of membrane-less compartments are ribonucleoprotein (RNP) granules (Hubstenberger et al., 2013), clusters of signaling complexes (Su et al., 2016), nucleoli (Brangwynne et al., 2011) and Balbiani bodies in developing oocytes (Boke et al., 2016) (Fig. 3). Evidence is now accumulating that these structures are different states of condensed living matter, and that similar colloidal mechanisms are at work, as described previously for model proteins such as lysozyme. Consequently, the term ‘biomolecular condensates’ has been proposed to refer to these assemblies (Banani et al., 2017).
Compartment formation through liquid–liquid demixing phase separation has emerged as one of the key principles for forming biomolecular condensates (Banani et al., 2017; Bergeron-Sandoval et al., 2016; Brangwynne et al., 2009; Hyman et al., 2014; Li et al., 2012). Indeed, upon changes in specific physicochemical conditions, many compartment-forming macromolecules such as RNA-binding proteins and RNAs ‘de-mix’ into a dilute and a dense phase. It has been hypothesized that the dense phase forms a liquid-like compartment, with the interface being a compartment boundary that allows selective access of certain components but not others.
Biomolecular condensates are most frequently found in cells exposed to stressful conditions (Rabouille and Alberti, 2017). For example, under conditions of heat stress or energy depletion, so-called stress granules (SGs) form in the cytoplasm of eukaryotic cells (Anderson and Kedersha, 2009; Buchan and Parker, 2009; Kroschwald et al., 2015; Molliex et al., 2015; Patel et al., 2015). These structures contain RNAs and RNA-binding proteins, as well as proteins involved in translation regulation. The material properties of SGs show some variation from organism to organism; they range from liquid-like in mammalian cells to gel- or glass-like in budding yeast (Kroschwald et al., 2015). SGs sequester and store RNAs and proteins to regulate translation in times of cellular crisis.
But how do SGs form? There is increasing evidence that the liquid-like SGs of mammalian cells form through liquid–liquid phase separation (LLPS) (Kroschwald et al., 2015; Molliex et al., 2015; Patel et al., 2015). Although these SGs have liquid-like properties, there also are signs that they contain solid-like ‘cores’ (Jain et al., 2016; Wheeler et al., 2016). It has been postulated that these cores are nuclei that initiate the assembly of SGs (Wheeler et al., 2016). However, it is possible that the cores are liquid-like initially, but then quickly harden with time into a gel- or glass-like state. This mode of compartment formation – LLPS followed by rapid maturation into a solid-like state – also appears to be the way in which the solid-like SGs of budding yeast form (Kroschwald et al., 2015). Our understanding of SGs is still limited, mostly because the interactions that drive SGs formation are still ill-defined. Some interactions appear to be mediated by folded protein domains and RNAs, whereas others involve intrinsically disordered protein regions that undergo promiscuous interactions (Kroschwald et al., 2015).
How do all these different components come together and assemble into one coherent SG? As explained in the paragraph above, it is likely that SG formation involves a nucleating step and that other components are recruited after a nucleus has formed. Recent findings with heat-stressed budding yeast indeed suggest that SG components become insoluble with different kinetics (Wallace et al., 2015). Proteins that respond quickly to heat stress have been termed super-aggregators (Wallace et al., 2015). These super-aggregators could be the nucleators of SGs. Components with more delayed kinetics may be recruited to SGs by interacting with the nucleators. We also have a very limited understanding of the overall organization of all these components within SGs. Do they mix and form one homogeneous phase or do they assemble into a multi-phase system? It will be important to temporally and spatially resolve SG formation in living cells and to reconstruct SGs from purified components.
There are other stress-inducible structures that do not seem to form through LLPS. For example, in budding yeast and Drosophila, there are filaments of assembled proteins (Liu, 2011; Noree et al., 2010). These protein filaments are often composed of one or a few metabolic enzymes, such as CTP synthase, which are repeated in many units along the filament axis. Thus, in a way, protein filaments are solid crystals that grow in one dimension. They assemble in a typical polymerization reaction that may be triggered by changes in metabolite concentrations (Noree et al., 2014) or general physicochemical conditions (e.g. pH) (Petrovska et al., 2014). Indeed, work from my group has shown that changes in protein concentration and alterations in crowding conditions are important, because they strongly impact the equilibrium between the assembled and unassembled states of these proteins (Petrovska et al., 2014). It has been proposed that filament formation either activates or inactivates the activities of these enzymes (Aughey and Liu, 2016; Petrovska et al., 2014). In addition, it is also possible that they serve to protect enzymes from damage or degradation.
As explained above, macromolecular condensates are a frequently observed phenomenon in stressed cells. In agreement, proteome-wide localization studies have discovered many assemblies in the cytoplasm and nucleoplasm of stressed budding yeast (Breker et al., 2013; Narayanaswamy et al., 2009). In fact, it appears that a major part of the cytoplasm of stressed cells can change its material properties in a regulated manner. This was recently shown for bacteria and yeast. In these organisms, the cytoplasm transitions from a fluid- to solid-like state upon energy depletion (Munder et al., 2016; Parry et al., 2014). Importantly, these organisms can quickly fluidize their cytoplasm again when the environmental conditions improve. Remarkably, my group recently showed that in yeast, this fluid-to-solid transition is regulated by internal pH changes and makes the cytoplasm so stiff that the affected cells keep their shape when their cell wall is removed (Munder et al., 2016). This suggests that cells have a far-reaching ability to change the material properties of the cytoplasm, which could be used to regulate the many metabolic processes that take place in this intracellular environment.
The material properties of condensates are subject to regulation in cells
As discussed above for SGs, there is evidence that intracellular condensates can change their material properties. In agreement, recent reconstitution studies performed with purified compartment-forming proteins, such as the RNA-binding proteins FUS or hnRNPA1, show that these proteins initially form liquid droplets, but these droplets become more solid-like with time (Molliex et al., 2015; Patel et al., 2015). This widespread phenomenon has been referred to as ‘molecular aging’. We do not know yet which structural rearrangements underlie the process of molecular aging of FUS and hnRNPA1, but when concentrated in the dense liquid phase, these proteins become increasingly immobile. This is consistent with a transition from liquid to gel or glass, similar to that observed for model proteins such as lysozyme (Fig. 2).
Although FUS ages in the test tube, it forms liquid-like compartments in living cells without any sign of aging (Alberti and Hyman, 2016; Molliex et al., 2015; Patel et al., 2015). This suggests that cells have mechanisms in place to maintain the liquid-like properties of FUS and presumably many other proteins. One possibility is that energy-consuming processes modulate the colloidal properties of these compartments and rearrange their constituents to keep them in a liquid state. Supporting evidence comes from a recent study on RNP granules in Caenorhabditis elegans. Inactivation of an ATP-consuming helicase triggered a transition of RNP granule components from a liquid state to solid crystals (Hubstenberger et al., 2013). This suggests that many membrane-less compartments are active, out-of-equilibrium colloidal materials that require a continuous input of energy.
There are also examples of compartments in living cells that change their physical properties depending on physiological conditions. One example is the centrosome of C. elegans embryos. Recent biochemical reconstitution studies show that centrosome assembly requires a coiled-coil protein called spindle-defective protein 5 (SPD-5) (Woodruff et al., 2015, 2017). Recombinant SPD-5 polymerizes into micrometer-sized porous networks in vitro. Importantly, in the presence of crowding agents, SPD-5 forms dense droplets that are initially dynamic and liquid-like, but then harden and become more solid-like. Centrosomes with liquid-like and solid-like properties have also been observed in different developmental stages of C. elegans embryos (Woodruff et al., 2017), suggesting that the phenomenon of droplet formation and subsequent aging is physiologically relevant.
Another structure with solid-like colloidal properties is the Balbiani body in Xenopus oocytes (Boke et al., 2016). In contrast to the centrosome, the Balbiani body does not appear to go through a liquid intermediate state. Balbiani bodies form through polymerization of the protein Xvelo (also known as Velo1) into a dense, fibrous network that sequesters proteins, RNAs and mitochondria. The Balbiani body is an extremely stable structure. Photobleaching experiments show that there is barely any turnover of assembled Xvelo in Balbiani bodies. It has thus been postulated that the rigid Balbiani body protects the sequestered components from damage and allows storage of important macromolecules and organelles for the next generation (Boke et al., 2016).
Compartments, such as the centrosome and the Balbiani body, are heterogeneous assemblies that contain many different components. Some of these components have a structural role and are required for the integrity of the compartment. These proteins have been termed scaffold proteins (Banani et al., 2016). Others are dispensable for compartment integrity and have been termed client proteins. Clients generally show a higher mobility than scaffolds, because they interact with the scaffolds only transiently. Together, scaffolds and clients form a selective phase that allows access of certain components, but not others.
Thus, we can conclude that the formation of intracellular condensates frequently occurs in cells and can affect a large fraction of the proteome. But what are the molecular features that promote compartment formation by condensation? Fortunately, studies are now beginning to shed light on the molecular principles underlying condensate formation in the intracellular environment, as discussed in the next section.
Molecular features promoting the formation of intracellular condensates
For condensate formation to occur in cells, a network of many interacting macromolecules has to be established. This network can only be stable in space and time if the number of connections is sufficiently high. Not surprisingly then, one of the key features of compartment-forming macromolecules is their valence, and there is increasing experimental evidence that multivalency promotes macroscopic phase separation (Banani et al., 2016; Banjade and Rosen, 2014; Banjade et al., 2015; Li et al., 2012; Zeng et al., 2016). Another important parameter that facilitates intracellular condensate formation is molecular flexibility (Malinovska et al., 2013; Toretsky and Wright, 2014; Uversky et al., 2015).
A class of proteins that combines all these features are intrinsically disordered proteins (IDPs) (Oldfield and Dunker, 2014; Wright and Dyson, 2015). The key hallmark of these proteins is their conformational flexibility. IDPs do not have a lowest energy state that corresponds to the native conformation. Rather, they populate a range of conformations, which differ only slightly in their free energy state. As a consequence, the structure of IDPs is better described as a conformational ensemble, where each conformation is associated with a certain probability of occurrence under a given condition. Another important feature of IDPs is that they have polymer-like properties. The flexibility and polymer-like behavior of IDPs support the formation of large adaptable networks with different dynamics and morphologies.
Compartment-forming IDPs are often characterized by low sequence complexity. For example, prion-like IDPs are enriched for polar amino acids, such as glutamine, asparagine and serine (Alberti et al., 2009). They are called prion-like because of their ability to aggregate and occasionally form structures with infectious prion properties (Malinovska et al., 2013; March et al., 2016). IDPs that contain blocks of charged amino acids are another example; RNA-binding proteins frequently contain repeats enriched for glycine and arginine (RG or RGG repeats) (Nott et al., 2015). Reconstitution experiments over the past few years have provided strong evidence that low-complexity IDPs drive the formation of many physiologically relevant condensates inside cells (Banani et al., 2017).
What are the interactions that drive the condensation of IDPs? The presence of distinct charge patterns suggests that electrostatic interactions play an important role. In agreement, the formation of dense liquid phases by IDPs is strongly sensitive to ionic conditions (Molliex et al., 2015; Nott et al., 2015). In addition, there are short-range interactions, such as dipole–dipole, π–π or cation–π interactions (Brangwynne et al., 2015). Dipole–dipole interactions are frequently observed for prion-like IDPs because they contain many polar amino acids, whereas π–π interactions require the presence of aromatic residues, such as phenylalanine and tyrosine. Cation–π interactions occur between aromatic residues and basic residues such as arginine and appear to be very important for the formation of RNP granules. Generally, interactions between IDPs are weak and their dissociation constant is in the micromolar range. This favors the formation of dynamic assemblies with liquid-like properties.
Although IDPs play a very important role, intracellular condensate formation is not strictly limited to IDPs. Proteins with multiple folded interaction domains have also been shown to condense into compartments (Li et al., 2012). The centrosome protein SPD-5 is another example (Woodruff et al., 2017). Although SPD-5 is partially disordered, it requires the formation of tightly interacting coiled-coils to assemble into dense droplets. In agreement, SPD-5 droplets harden more quickly than IDP droplets and adopt solid-like properties within a few minutes (Woodruff et al., 2017).
At the other end of the disorder-to-order spectrum are metabolic enzymes that assemble into filaments and maintain their globular native folds upon assembly. Filament formation appears to be driven by electrostatic interactions and the steric compatibility of the filament-forming enzymes (Petrovska et al., 2014). Metabolic enzymes are often highly concentrated in the cytoplasm and are characterized by symmetrical geometries, features that appear to favor the assembly of folded enzymes into filaments.
As discussed in this section, the features that drive assembly are diverse. Some assembly reactions require conventional protein folds, whereas others depend on intrinsic disorder and multivalency. But what are the molecular triggers that lead to the formation of intracellular condensates?
Regulating the formation of intracellular condensates
Condensate formation requires the establishment of a network of interactions between a large number of macromolecules. Generally, this can be achieved by changing the binding affinities, the solubility or the effective concentration of the interacting components. Binding affinities can be tuned in several ways. For example, changes in intracellular physicochemical conditions can directly affect protein behavior. Early research with folded model proteins such as lysozyme showed that protein phase behavior is often driven by electrostatic interactions. The reason is that folded proteins carry many amino acids on the surface that are charged and behave as acids or bases. This also means that pH changes strongly modulate protein solubility and phase behavior (Dumetz et al., 2008a; Pace et al., 2009). In addition, protein solubility also depends on the ionic strength of the solution (Collins, 2004; Ruckenstein and Shulgin, 2006). Thus, whether a protein undergoes interactions or not is often determined by a combination of the protein surface charge density, as well as the ionic strength and pH of the solution. It is generally assumed that basic parameters, such as the cytosolic pH or ionic strength, are stable inside cells. However, there is now increasing evidence that these cellular parameters can fluctuate, in particular when cells are stressed. In fact, it has been proposed that cells have learned to interpret such fluctuations as stress signals (Munder et al., 2016; Orij et al., 2011). Thus, the formation of intracellular condensates may be an adaptive strategy to cope with stress and promote organismal survival (Rabouille and Alberti, 2017).
Which molecules could sense such fluctuations in physicochemical conditions in cells? IDPs are exquisitely sensitive to changes in physicochemical parameters, such as pH and ionic strength (Theillet et al., 2014). Changes in pH or salt could shift the conformational ensemble of IDPs and this could directly affect IDP phase behavior. In the future, it will be important to identify the sequence features that act as stress sensors and convert changes in physicochemical conditions into the formation of condensed phases.
Binding affinities can also be regulated through post-translational modifications (PTMs). Phase-separating IDPs often contain residues, such as serines or arginines, which can be modified through phosphorylation or methylation events, respectively. Indeed, evidence for the regulation of colloidal protein behavior by PTMs is accumulating. The starvation-associated Sec bodies in Drosophila that are modified through ADP ribosylation are one example (Aguilera-Gomez et al., 2016; Zacharogianni et al., 2014). The process of ADP ribosylation is mediated by a specific polymerase that is turned on during amino acid starvation and specifically adds ADP-ribose moieties onto components of the Sec body. Another example is provided by SGs in mammalian cells, which are regulated by phosphorylation of certain SG components by kinases (Mc Inerney et al., 2005; Wippich et al., 2013). Moreover, there is evidence that phase separation of the RNA helicase Ddx4 is regulated by methylation of specific arginine residues (Nott et al., 2015).
Another possibility to regulate intracellular condensate formation is through changing the effective concentration of a protein. This can be achieved by either upregulating protein expression, or alternatively, increasing crowding conditions in cells. Indeed, recent studies show that macromolecular crowding strongly favors the formation of intracellular condensates. Addition of crowders, such as polyethylene glycol or Ficoll, to in vitro assembly reactions strongly accelerates the formation of centrosomes or protein filaments (Petrovska et al., 2014; Woodruff et al., 2017). Interestingly, when cells are stressed, for example, during nutrient depletion, they respond by reducing their cell volume, which, in turn, increases the crowding conditions inside the cell (Joyner et al., 2016). Thus, it appears that cells can actively adjust the intracellular crowding conditions in times of crisis, which increases the effective concentration of macromolecules and could thus drive the formation of intracellular condensates.
In summary, there are many ways in which condensate formation can be regulated and induced in cells. Macromolecular condensates also appear to have a variety of useful features, raising questions regarding their functional roles.
The functional repertoire of intracellular condensates
We can envision many different functional roles of intracellular condensates, such as acting as biochemical reaction vessels in which certain enzymatic reactions occur more efficiently than in the bulk solution (Fig. 4A). Some time ago, Paul Srere recognized that cellular metabolism would strongly benefit from compartmentalization. He proposed the so-called metabolon concept, in which metabolic enzymes cluster together and, by doing so, increase the efficiency of enzymatic reactions (Ovádi and Srere, 2000; Srere, 1987). Indeed, there now is increasing evidence for metabolic compartmentalization (Sweetlove and Fernie, 2013; Zecchin et al., 2015). However, it is not known whether this involves condensate formation. DNA damage sites constitute another compartment that presumably enables enzymatic reactions. These assemblies have properties of liquid-like droplets and they recruit diverse enzymes that are dedicated to damage repair in a sequential and highly coordinated manner (Altmeyer et al., 2015; Patel et al., 2015). Concentrating DNA repair enzymes in proximity to a DNA lesion is likely to accelerate the repair process. A third example of a condensate that is thought to enhance biochemical reactions is the processing body (P-body). This liquid-like assembly of ribonucleoproteins is involved in degrading and silencing mRNAs (Kroschwald et al., 2015; Mugler et al., 2016; Sheth and Parker, 2003). It has been suggested that P-bodies concentrate RNA processing enzymes and that this makes mRNA degradation more efficient and specific. Finally, T-cell-receptor signaling clusters form on the cytosolic face of immune cell membranes (Fig. 4B), and these membrane-associated droplets have been proposed to concentrate kinases, while excluding phosphatases, thus amplifying phosphorylation reactions (Su et al., 2016). However, for all of these examples it remains to be rigorously tested whether these droplet compartments enhance specific enzymatic reactions or allow selective access of substrates or products.
To date, there are only few instances for which it has clearly been demonstrated that condensates have biological functions. One example is provided by reconstituted centrosomes. Tubulin enrichment in reconstituted centrosomes by a factor of four or more promotes the formation of microtubule asters, thus providing a foundation for centrosomes as microtubule-organizing centers (Woodruff et al., 2017). The formation of actin filaments at the plasma membrane appears to follow a similar principle. Actin assembly has been shown to be promoted by concentrating actin nucleators such as Arp2/3 (ACTR2/3) at the plasma membrane in phase-separated compartments (Banjade and Rosen, 2014; Li et al., 2012; Su et al., 2016) (Fig. 4C). What these examples have in common is that compartment formation allows the concentration of specific components above a threshold for nucleation. However, to properly perform their functions as compartments, they have to be selective and only allow access of specific components. How compartments achieve their selectivity for certain components, while excluding others, will therefore be an important research area for the future.
In addition to active droplet compartments, there are other condensates that function as compartments for the sequestration and release of specific components. This principle appears to apply to many RNP granules. For example, when under stress, cells form stress granules that sequester RNA and proteins to regulate translation (Jain et al., 2016; Kroschwald et al., 2015) (Fig. 4D). The sequestered RNAs and proteins are released again when cells re-enter into the cell cycle.
Stress and other sudden changes in the environment generate a need to protect, inactivate and store many macromolecules. But what is a good material that facilitates protection, inactivation and storage? Glasses and gels have a high tolerance for heterogeneity, because they have many degrees of freedom, and thus can accommodate many different components; these materials also limit macromolecular diffusion, thus providing protection and potentially inactivation to the stored components. Protein filaments or crystals are also excellent materials for protein storage, but they can only accommodate one or a few components, so they are less tolerant of heterogeneity. One example is the enzyme glutamine synthetase, which forms inactive protein storage filaments when yeast cells enter into quiescence (Petrovska et al., 2014). These filaments dissolve rapidly when cells re-enter the cell cycle, thus providing a ready supply of this growth-limiting enzyme. Stressed cells contain many assemblies that may be adaptive, but whether they promote cellular survival by storing, protecting or inactivating macromolecules during stress remains to be shown.
Finally, it should be mentioned that condensed phases have special capabilities that single macromolecules do not have. This is because condensed phases consist of many different dynamically interacting components. The collective properties of these components can unlock higher-order emergent properties. As a consequence, condensates can exhibit complex properties, such as wetness, adhesiveness, rigidity or elasticity. These features can in principle be harnessed by cells to acquire a new functionality. This largely unexplored area will be a particularly interesting research direction in the future.
Cells are extremely complex and highly organized systems. Scientists have been speculating for many years how they can maintain their elaborate architecture and at the same time perform all their different functional tasks. Molecular cell biology has given us tremendous insight into the molecular parts that underlie the complex behavior of living cells. However, there has been a gaping hole in our understanding of how all these different parts work together to yield complex cellular behavior. Recent research has provided evidence for the colloidal behavior of macromolecules and how protein crowds affect fundamental processes in living cells. Among others, protein colloids have been shown to facilitate intracellular organization, signaling, metabolism, and stress adaptation. However, we have only just started to look at living cells with new eyes and many additional examples are probably yet to be discovered. What are the cellular functions that are regulated by biomolecular condensates? Can we learn the molecular rules underlying condensate formation? How does evolution shape the properties of condensate-forming macromolecular collectives? The answers to these and other related questions will transform our view of cellular and organismal biology. The future has just begun.
I am grateful to Christiane Iserman and Elisabeth Nüske for critical comments on the manuscript.
I acknowledge funding from the Max-Planck Society, the Deutsche Forschungsgemeinschaft and the Volkswagen Foundation.
The author declares no competing or financial interests.