The pathophysiology of neurodegenerative disease: Disturbing the balance between phase separation and irreversible aggregation

Abstract

Liquid-liquid phase separation (LLPS) brings together functionally related proteins through the intrinsic biophysics of proteins in a process that is driven by reducing free energy and maximizing entropy. The process of LLPS allows proteins to form structures, termed membrane-less organelles. These diverse, dynamic organelles are active in a wide range of processes in the nucleus, cytoplasm, mitochondria and synapse, and ranging from bacteria to plants to eukaryotes. RNA and DNA present long chained charged polymers that promote LLPS. Consequently, many RNA binding proteins (RBPs) and DNA binding proteins form membrane-less organelles. However, the highly concentrated phase separated state creates conditions that also promote formation of irreversible protein aggregates. Mutations in RNA and DNA binding proteins that increase the stability of irreversible aggregates also increase the accumulation of irreversible aggregates directly and from membrane-less organelles. Many of the RBPs that exhibit disease-linked mutations carry out cytoplasmic actions through stress granules, which are a pleiotropic type of RNA granule that regulates the translational response to stress. Phosphorylation and oligomerization of tau facilitates its interactions with RBPs and ribosomal proteins, affecting RNA translation; we propose that this is a major reason that tau becomes phosphorylated with stress. Persistent stress leads to the accumulation of irreversible aggregates composed of RBPs or tau, which then cause toxicity and form many of the hallmark pathologies of major neurodegenerative diseases. This pathophysiology ultimately leads to multiple forms of neurodegenerative diseases, the specific type of which reflects the temporal and spatial accumulation of different aggregating proteins.

1. Introduction

This chapter will explore the thesis that the origins of many neurodegenerative diseases derive from the double edge sword that is the biology of protein based liquid-liquid phase separation (LLPS). LLPS has been known to exist as a chemical phenomenon for decades, however the essential role of LLPS in the biology of proteins only became apparent in a series of seminal papers published between 2012 and 2015.^1–6 These papers demonstrated that many proteins can coalesce to form into phase separated structures that are referred to as “membrane-less organelles”.⁷ Since then, we have come to understand that LLPS underlies the biology of most interactions between proteins and RNA or DNA, and likely extends to many more proteins throughout the cell. LLPS coalesces proteins into droplets that contain relatively high concentrations of aggregation prone proteins. These protein droplets have the benefit of bringing functionally relevant proteins together into close proximity, however, the benefit of bringing these proteins together into close proximity creates an inherent problem by increasing the rate of formation of irreversible, insoluble aggregates.

The biology of LLPS presents a fundamental problem for organisms as they age. In young organisms these aggregates pose little problem because the catabolic autolysosomal system is sufficiently active to remove the aggregates as they form, which prevents their accumulation. As we age, though, autolysosomal flux diminishes, reducing the ability to remove irreversible protein aggregates.^8–10 The progressive accumulation of these aggregates represents the most consistent pathological feature of late onset neurodegenerative disease, and is thought to underlie subsequent neurodegeneration. Correspondingly, many genetic risk factors for neurodegenerative disease constitute genetic changes that increase the rate of formation or interfere with removal of protein aggregates; these genetic changes accelerate the accumulation of irreversible proteins aggregates which promotes the development of neurodegenerative disease.

2. The biophysics of LLPS

The basic concepts underlying LLPS are remarkably simple.^11,12 When we think of phase separation, we typically think of one immiscible liquid forming drops within another liquid, such as oil in water. The droplet formation is driven by thermodynamics as the two liquids separate because separating into homogenous states is a lower energy state. Hydrogen bonding between water molecules is disrupted by having lipids dispersed in the aqueous phase, which propels lipids to coalesce into droplets. The process reduces the overall “free energy” of the mixture. Each phase, though, remains highly dynamic because all of the interactions are relatively weak.

The same ideas apply to LLPS of proteins. Many proteins contain regions that exhibit a low complexity of amino acids. These regions contain many small segments, termed “stickers” that weakly associate with each other.^11,13 These weak interactions cause the proteins to stay close to each other because small regions of the proteins are always weakly sticking to each other. Herein lies the origin of the phase-separated protein droplet; the proteins associate and coalesce together due to weak interactions at multiple different sites but because the interactions are weak the protein mixture remains highly mobile and dynamic, which fulfills the criteria of a liquid droplet.

2.1. Phase separation is driven by a decrease in free energy

From the water vapor generated during boiling of the water to ice melting in cold coffee, phase separation of matters occurs every day in life. Cells exhibit similar phase-based phenomena, exhibiting abundant evidence for a strong role of phase separation in cell biology, the structure of nucleoli and Cajal bodies in the nucleus, and the role of stress granules (SGs) in regulating cytoplasmic mRNA translation.^14–17 The definition of a membrane-less organelle is a nonstoichiometric condensation of protein and/or nucleic acid molecules through a spontaneous or energy driven process by lowering the free energy of the system. These organelles exhibit a liquid-like behavior, high diffusion coefficient of content molecules, high fluidity and dynamic organization, which together lead to a generalized state termed Liquid-Liquid Phase Separation (LLPS).

As a spontaneous or energy driven process, LLPS results from a change in free energy of the system.^11,18,19 Change in free energy is a physical definition which quantifies the difference in energy state of two different states. To discuss the state of randomness or the degree of freedom, entropy is measurement for how “free” the molecule can be. The fundamental law is that the end state of a system with no interactions is always in its maximum entropy state, in other words, matters have a tendency to mix or be “chaotic” in the system. In statistical mechanics, entropy is the ratio of system entropy given by a discrete orientation of particles in a given space. Consider a system with a confined space of two different molecules A and B separated by a membrane or boundary. When the membrane or boundary is released, the molecules tend to mix randomly in this confined space. This tendency is driven by a change in entropy, as defined by a change in number of states the molecules could orientate itself in this given space. This entropy is also called mixing entropy. As entropy is a quantity that describe the change of the system, we have to consider the number of possible arrangements for molecules initially (boundary on), W_on, and number of possible arrangements for molecules afterwards (boundary off), W_off. Initially, as two molecules are confined in each individual space, there is only one possible arrangement in this state. Therefore, W_on is 1. After the boundary is removed, with number of lattices being N and number of A molecules being n_a, the number of possible arrangements for this state is $W_{off}=\frac{N!}{n_a!\left(N-n_a\right)!}$. In Boltzmann’s theory, the entropy is the change of possible arrangements between two states: $S=k_B\text{ ln }\frac{W_{off}}{W_{on}}$. For this example, the entropy of the system is $\frac{S}{k_B}=-n_a\text{ ln }{n}_a+N\text{ ln }N-\left(N-n_a\right)\text{ ln }\left(N-n_a\right)$. In practice, the number of molecules is hard to measure but the concentration is measurable. Switching the entropy into an volumetric view (by dividing the whole equation by volume of the system), the expression for entropy per volume space is $\frac{S}{V{k}_B}=-{\Phi }_a\text{ ln }{\Phi }_a-\left(1-{\Phi }_a\right)\text{ ln }\left(1-{\Phi }_a\right)$, where Φ_a is the volume fraction of a molecules.²⁰ As the volume fraction of molecule type “A” rises, the entropy per volume increases to a maximal point then decreases to 0, modeling a bell shape behavior (Fig. 1A). Entropy of the system would reach a maximum at the end state of the system, therefore for this example of molecule types “A” and “B”, the molecules will unmix randomly till a point that there is an equal volume fraction of each molecule. From this model, the entropy drives the system to an unmixed state. The system reaches a final state with the highest entropy: with 0.5 volume fraction of molecule A and 0.5 volume fraction of molecule B in a mixing state. With the entropy change of volume fraction equal to 0, the system reaches a locally stable point.

Fig. 1.

Single particle-phase separation thermodynamic model. (A) Mathematical model of entropy of mixing for the two-particle system. The x-axis is the volume fraction of the particles (red or green). The y-axis is the entropy of the system. The maximum entropy of mixing is achieved when the volume fraction of the particular particle reaches 50% (i.e., 0.5 on the X axis). (B) The change in free energy of mixing of the two-particle system. The x-axis is the volume fraction of particle a and the y-axis is the free energy of the system. As the exchange parameter, X_ab, increases (becomes more positive), the system becomes more energetically unfavorable towards mixing and phase separation is favored. (C) Diagrams showing the change in free energy of mixing of the two-particle system at different temperatures. On the left, when the temperature is significantly lower than miscible temperature zone (T_c), the corresponding volume fractions of the particles (red or green) at two local minimums of change in free energy occur at positions A′ and B′. This correlates to the boundary condition of phase separation. In the middle, when the temperature is lower or close to miscible temperature zone (T_c), the corresponding volume fractions of a at two local minimums of change in free energy occur at A″ and B″. On the right, when the temperature is significantly larger than to miscible temperature zone (T_c), the change in free energy of the system is negative which indicates the system favors mixing without phase separation. (D) A phase diagram constructed based on a schematic showing the change in free energy of mixing of the two-particle system at different temperatures. A′, A″, B′, and B″ show the boundary of the two phases conditions and identify the temperature of the corresponding volume fraction. At volume fraction above the T_c results in a system that is mixed with no phase separation.

Since the second law of thermodynamic indicates that the only favorable system is the mixing of molecules in a given space, then how would phase separation occur? Besides considering all molecules as simple objects in a lattice model, molecules actually have interactions between others. To model this interaction, we consider the interaction of molecules with neighboring molecules to form bonds. Bond formations allow electron sharing hence lowering the energy of the system. In this example, three different bonds would occur, w_aa, w_bb, w_ab which account for the interaction energy between A and A molecules, B and B molecules, and A and B molecules, respectively. By modeling the overall energy interaction and applying Bragg-Williams Mean-field approximation for the number of A and B interactions, the total interaction energy is:

$$U=\left(\frac{z{w}_{aa}}{2}\right)N_a+\left(\frac{z{w}_{bb}}{2}\right)N_b+kT{\chi }_{ab}\frac{N_a{N}_b}{N}$$

$${\chi }_{ab}=\frac{z}{kT}\left(w_{ab}-\frac{w_{aa}+w_{bb}}{2}\right)$$

For the equation as proved in Dill and Bromberg’s book, Molecular Driving Forces,²⁰ z is the number of contact bonds that form with neighboring molecules. N is the number of molecules in the system. N_a and N_b are the number of A and B molecules in the system. χ_ab is the exchange parameter that describe the energy difference between mixing and unmixing. K is the Boltzman Constant and T is the temperature of the system; together kT is the thermal energy of a particular state. The overall energy of the system is a general expression of balance between A-A homogenous interaction, B-B homogenous interactions, and A-B heterogeneous interactions. The exchange parameters account for the balancing of these different interactions. If the heterogeneous interactions is more favorable (more energy drop) than homogenous interactions, then more As would neighbor Bs instead of a cluster of As and another cluster of Bs (Fig. 1B).

Combining both the overall interaction energy with the entropy of mixing, the equation shown below is the free energy of mixing which accounts for the general free energy change of the system.

$$\frac{\Delta F}{NkT}={\Phi }_a\text{ ln }{\Phi }_a+\left(1-{\Phi }_a\right)\text{ ln }\left(1-{\Phi }_a\right)+{\chi }_{ab}{\Phi }_a\left(1-{\Phi }_a\right)$$

The key factor that drives phase separation is the exchange parameter. χ_ab is a temperature dependent factor that accounts for the balancing of mixing and unmixing of the system through interactions.²⁰ For miscible liquids, when χ_ab ≫ 0 the system favors heterogeneous interactions over homogeneous interactions, $w_{ab}\gg \frac{w_{aa}+w_{bb}}{2})$. Despite this system not favoring phase separation, the enthalpy of interactions and entropy of mixing are both driving the system to mix by lowering the free energy of the system. On the other hand, for immiscible liquid, χ_ab ≪ 0, where the homogeneous interactions are more favorable than the heterogeneous interactions, the system would have exhibit behaviors that are dependent on temperature. Rearranging the previous equation, the expression for free energy is obtained as below:

$$\Delta F=RT\left[{\Phi }_a\text{ ln }{\Phi }_a+\left(1-{\Phi }_a\right)\text{ ln }\left(1-{\Phi }_a\right)\right]+RT{\chi }_{ab}{\Phi }_a\left(1-{\Phi }_a\right)$$

Since χ_ab is inversely dependent on the temperature term, $\frac{z}{kT}$. The expression could be rearranged again shown below:

$$\Delta F=RT\left[{\Phi }_a\text{ ln }{\Phi }_a+\left(1-{\Phi }_a\right)\text{ ln }\left(1-{\Phi }_a\right)\right]+{\text{X}}_{ab}{\Phi }_a\left(1-{\Phi }_a\right)$$

X_ab is the molar term for χ_ab that ${\text{X}}_{ab}={\chi }_{ab}RT=z{N}_a\left(w_{ab}-\frac{w_{aa}+w_{bb}}{2}\right)$. This term is not temperature dependent. Therefore, free energy is determined by the entropy of mixing which is dependent of temperature and the enthalpy of interaction which is independent of temperature. In the case of immiscible liquids, the enthalpy is held constant (ideally), while the contribution of entropy changes with temperature to allow mixing or unmixing to occur.

As the temperature changes, the contribution of the entropy of mixing becomes more dominant than the enthalpy of interactions for dictating a free energy decrease (RT[−Φ_alnΦ_a−(1−Φ_a)ln(1−Φ_a)]≫ X_abΦ_a(1−Φ_a)), and vice versa. When the entropy of mixing is dominant, the system favors a state of mixing rather than phase separation. But when the temperature is too low, the system does not reach any state because the change in free energy would not be negative (Fig. 1C.) Moreover, when the system reaches a temperature threshold, the system favors phase separation stably. This temperature threshold is called the miscible temperature zone (T_C, where “c” refers to a critical point for phase transition). Consequently, the system is separated into three possible compositions by temperature (Fig. 1C).

When the temperature of the system is far below T_C, the system does not proceed to make any changes to the initial state, which is the state that both molecules are separated. The interpretation of this composition is such, that the contribution of entropy of mixing (the tendency of mixing) to free energy change is far lower than the contribution of homogeneous interactions to free energy change; therefore, the system stays interacting homogeneously. At composition A′ and B′, the system has two local minima, the change in free energy at this point is zero indicating a stable system. Any composition other than A and B would eventually reach A and B state with no fluctuation of free energy. When the temperature of the system is around T_C, the system reaches the lowest temperature to phase separated stably. The phase separation occurs at composition A and B, which are the local free energy minima. Any composition other than A′ and B′ would finally reach A′ and B′ state with no fluctuation of free energy. Finally, when the temperature of the system if far above T_C, the system undergoes mixing with one minima of free energy at point A″. All of the compositions other than A″ would lower the free energy to mix down to composition A″. This is the same for miscible fluid. As (RT[−Φ_alnΦ_a−(1−Φ_a)ln(1−Φ_a)]≫X_abΦ_a(1−Φ_a), the free energy of the system can be simply approximated to: ΔF≈RT[−Φ_alnΦ_a−(1−Φ_a)ln(1−Φ_a)].

By drawing relationship between the composition points of minima free energy with distinctive temperatures, the phase diagram of phase separation of two molecules is obtained. (Fig. 1D) In practice, this regular solution model by Hallience applies to the majority of two-system single molecule compositions mixing in solution (ranging from small molecule atoms or ions to simple chemical components). By modeling the entropy of the system and measure the effective enthalpy of interaction of molecules, a theoretical phase diagram can be obtained. Locating the composition of molecules and the temperature of the system, the phase diagram can determine the state in which that system exists.

2.2. Polymeric phase separation

Previously we have discussed the physical basis for phase separation of simple molecules in a regular solution, which only applies to single or small molecules in a confined system. However, phase separation of proteins represents polymeric structures that are composed of long stretches of amino acids. The larger size of polymeric compounds means the space occupied by the protein during phase separation is large and the entropy of that molecule would change with a change in molecule size. In the 1940s, Flory and Huggins proposed the non-idealized polymer unmixing model in single molecule solvent.^21,22 The model treats polymer as a solvent-sized chain segment in the same lattice model done on single molecule phase separation.

The Flory-Huggins model of polymer unmixing reflects the concept of mixing. As polymers have a randomness of linear combination with chains that connect each solvent-sized segment in space, they recapitulate the entropy of mixing in statistical thermodynamics. They count the possible configurations by using a “chain growth” method, where the polymer chain is a growing solvent in a linear orientation (in 2D) segment in space. This concept modifies the entropy of mixing to the following equation.^21,22

$$\frac{S}{M{k}_B}=-{\varphi }_s\text{ ln }{\varphi }_s-\left(\frac{{\varphi }_p}{N}\right)\text{ ln }\left({\varphi }_p\right)$$

In this equation, M is the total number of lattice states available in the system. N is the number of polymeric chains in the system. ϕ_s and ϕ_p denote as the volume fractions of singe molecule solvent and polymeric chain, respectively. k_B is the Boltzmann constant. As the number of molecules on the polymer increases, the free energy of mixing is skewed right and shifts down with the maximum entropy occurring at a lower degree of freedom and a higher polymer volume fraction (Fig. 2A). For example, a protein with 300 amino acids chains will exhibit a higher volume fraction to reach the maximum entropy in the final state than a protein with 10 amino acid chains (Fig. 2B).

Fig. 2.

Polymer phase separation thermodynamic model. (A) A mathematical model of entropy of mixing for the polymeric system, where the x-axis is the volume fraction of the polymer and the y-axis is the entropy of the system. As the number (N) of the monomer in the polymer chain increases, the entropy of the system decreases and skewers to the right. (B) A mathematical model of entropy of mixing for 30 amino acids and 500 amino acids protein system. The x-axis is the volume fraction of the polymer and the y-axis is the entropy of the system. The 500 amino acid protein exhibits a lower entropy of mixing than the 30 amino acid protein, which indicates that longer proteins require larger volume fractions for the system to reach the maximum entropies (the thermodynamic endpoint of mixing). (C) A mathematical model of the change in free energy of a 500 amino acid protein with various exchange parameters. The x-axis is the volume fraction of the polymer and the y-axis is the change in free energy of the system. The change in free energy of the polymer is skewed towards the right (influenced by the change in entropy) compared to the free energy of the single-particle system. Moreover, with a decrease in entropy, the system exhibits a lower exchange parameter for phase separation with the transition occurring at ~X_ab=2.

The enthalpy of interaction remains the same with simplification of ignoring the two covalent bonds already attached on the molecule in the polymer. Consequently, the change in free energy of the system is:

$$\frac{\Delta F}{MkT}={\Phi }_s\text{ ln }{\Phi }_s+\left(\frac{{\varphi }_p}{N}\right)\text{ ln }\left({\varphi }_p\right)+{\chi }_{sp}{\Phi }_s{\Phi }_p$$

$${\chi }_{sp}=\frac{z}{kT}\left(w_{sp}-\frac{w_{ss}+w_{pp}}{2}\right)$$

For example, a 500 amino acid protein sequence with different exchange parameters (from 0, 1 to 10) exhibits different phase separation profiles with distinctive changes to the free energy diagram. For an exchange parameter lower than 2, this 500 amino acid protein would undergo mixing. When χ_sp =2, the system has two minima which means that the system would undergo phase separation to the composition of those two minima. When χ_sp>2, the system has a positive free energy indicating that the system would undergo phase separation spontaneously (Fig. 2C).

2.3. Polymeric biomolecule interactions

As previously mentioned, the important variable for determining the separation point of the miscibility gap and miscible temperature zone is the exchange parameter. The exchange parameter contains three variables that are dependent on the polymeric compound itself and the solvent/molecule it is surrounded in: w_sp, w_ss, and w_pp. These represent the contact energy of the solvent-polymer, solvent-solvent and polymer-polymer, respectively. The contact energy is a combination of all energy possible for the two-compounds described as such. To understand the phase separation in greater detailed, models for the energy of interactions are required.

2.4. Bioinformatics modeling with force field

A popular approach to model the interactions and conformations of large molecules is to model the energy as a combination of columbic interaction between charged molecules, spring forces that are required for bond deformation, the periodic potential for rotations around bonds, and the Jones (van der Waals) potential for non-bonded interactions.²⁰ Multiple novel interactions that drives biomolecule phase separation were also proposed with this goal. Based on this interaction model, a machine learning based algorithm, termed CAMELOT, was developed to predict protein phase separation by analyzing the force field of proteins in a distinctive solvent phase.¹⁹

A more common way of analyzing protein interactions is to employ the idea of sticker and spacer; where sticker is the domain that has stronger interactions while spacer is a non-interactive domain. The space does have covalent bonds attached on the sequence introducing string energy and configuration constrains.¹¹ The machine learning based algorithm, LASSI, analyzes sticker and spacer behavior to predict protein phase separation behaviors and conditions in a given solvent.¹¹

Most recently, databases have been built to identify protein phase separation based on protein structures, domain interaction strengths and force fields affecting proteins. The PhaSePro database combines previous LLPS data on proteins through manual curation.²³ The LLPSDB database is similar to LLPSDB, and contains 1182 entries representing 273 independent proteins and 2394 specific conditions related to specific examples of protein LLPS analysis.²⁴

2.5. LLPS depends on particular interactions of specific polymeric forms

Recent studies have begun to dissect the “grammar” guiding the effects of amino acids and nucleotides on protein phase separation and the driving force of prion-like RBPs.^13,25 This molecular grammar proposes that polar-pi interactions (which lead to electron stabilization) allow members of the FET proteins (Fused in sarcoma (FUS), Ewing’s sarcoma breakpoint region 1 (EWSR1) and TATA-box binding protein associated factor 15 (TAF15)) to undergo homogenous spontaneous LLPS in physiological conditions. In this model, the number of polar-pi interactions directly contribute to the driving force (the exchange parameter) that control homogenous LLPS.¹³

A related approach applied bioinformatics to investigate the x-ray crystallography of folded proteins based on RCSB Protein Data Bank (PDB).²⁶ This study indicated that that the hydrogen bonding of protein frequency often increases with sp2 sidechains that undergo pi-contacts, thereby identifying cooperative binding resulting from electrostatic and geometric stabilization of bonds by sp2 sidechains with the surrounding solvent. This study also demonstrated long-range planar pi-contacts are more generally shared in the field of protein-protein interactions than intra-protein interactions. Based on pi-pi contact frequency by the PDB, they developed PScore prediction method to predict protein sequences with the highest probability of undergoing phase separation.²⁶

2.6. Multiphase LLPS

The reality of biological systems, though, is that LLPS results from multiple phases that co-separate (ternary, quaternary or multi-species phase separation of dynamic proteins, lipids, or nucleotide chains). Bohidar proposed a statistical thermodynamic model for the condition of ternary phase separation to occur²⁷:

$$\frac{{\left({\sigma }_{23}\right)}^2}{\sqrt{I}}\ge \frac{8}{3\alpha }\left(2{\chi }_{231}-1\right)$$

The behavior of just such a complex system was later investigated by Brangwynne’s group, with a series of studies of the biophysical behavior of the molecules that form the core of the nucleolus: NPM1, FIB1 and RNA.^3,28 These studies added the role of surface tension into the biology phenomenon demonstrating empirical proportionality to the Flory parameter.^3,28

2.7. Chemical factors regulating LLPS

The biophysical concepts described above translate into strong empirical studies in living systems. The process of LLPS is controlled by the primary sequence of the target protein, and also to the environmental conditions. The first principle that we will consider is the role of the primary sequence in controlling phase separation. The most important protein domains controlling LLPS are termed low complexity domains (LCDs) or intrinsically disordered protein regions (IDPRs).^29,30 This was shown by proteomics demonstrating that the majority of natural membrane-less organelles are enriched in in intrinsically disordered proteins.^29,31 IDPRs are associated with the assembly of membrane-less organelles, which are implicated in numerous diseases such as cancer, cardiovascular disease, and neurodegenerative disease.³² The interactions observed in membraneless organelles are highly dynamic and exhibit supramolecular “fuzziness”.³³ These features of membrane-less organelles can be in part attributed to the LCDs and LDPRs in many RBPs and proteins involved in phase separation.

The composition of these LCDs consists of a high fraction of alanine, glycine, glutamine, arginine, tyrosine, serine and proline.^34,35 These uncharged amino acids impart an intrinsic disorder to the regions, allowing the structure to fluctuate in a dynamic manner.¹³ The weak structure, though, masks an internal language controlling the tendency to phase separate. The structure of LCDs can be considered to be groups of spacers and stickers.^11,13 The chains are composed of sequences of simple amino acids (e.g., alanine and glycine) that do not impart structure. The chains are interspersed with short regions (e.g., 6 amino acids) that contain glutamine, arginine and tyrosine, which are amino acids that are capable of forming weak bonds termed Pi-bonds. These Pi-bonds enable the weak interactions that drive LLPS). The presence of multiple stickers in each LCD allows for sufficient interactions to drive LLPS. The sticker: sticker interaction, though, is weak and therefore transient, which means that the LCD interactions are highly dynamic, creating the conditions for the complex to exist in a liquid state.

The strength of the sticker interactions can be regulated by post-translational modifications (PTMs) of the amino acids. The effects of PTMs varies for each particular protein, but in the case of FUS, increasing charges promotes interactions, which leads to larger liquid droplets. Thus, phosphorylation of serine or tyrosine promotes droplet formation,¹³ while methylation of arginine inhibits droplet formation.³⁶ The length of the spacer also impacts on phase separation, with longer spacers promoting independent action of the stickers and less phase separation.^11,13

Other chemicals in the cell strongly impact on phase separation of proteins. RNA and DNA play a critical role in LLPS because they act as a poly-ionic scaffold that promotes phase separation.³⁷ Thus, many of the complexes of proteins in the nucleus function as phase separated droplets; these components are discussed in more detail in subsequent parts of this review. The nucleolus is the most visible such droplet, but other complexes are also phase separated. Transcription complexes act as phase separated proteins, as gem bodies, PML bodies, Cajal bodies and nuclear speckles.^2,15,38,39 Long non-coding RNAs act in the nucleus to control the phase separation of proteins that regulate chromatin structure, and even the proteins around telomeric complexes are phase separated.^40,41 In the cytoplasm, mRNA acts as an essential component that is recognized by RNA binding proteins and promotes phase separation of the very same proteins, forming structures such as P-bodies and stress granules. Salt, pH and temperature regulate phase separation by controlling the charge interactions and kinetic energy.^1–6 Each of these factors has an optimal concentration that favors phase separation. Frequently, the optimal concentration coincides with the normal state of the cell. Thus, the chemical nature of the cell promotes LLPS of proteins.

3. Types of RNA granules

The types of RNA granules that form are determined by the protein components. Each type of RNA granule has characteristic components. Although the biology governing formation of RNA granules is poorly understood, the protein components of each type of granule suggest that specific groups of protein tend to coalesce together in membrane-less organelles.⁴² It seems likely that the structural interactions of co-phase separating proteins integrates with binding of related mRNA transcripts to promote formation of each type of RNA granule.^43,44 Emerging evidence suggests that long non-coding RNAs (lncRNAs) also contribute to formation of different types of RNA granules.⁴⁰ The section below will outline the regulation of some of the major types of RNA granules that contain disease-linked RNA binding proteins.

Multiple different techniques are used to determine the physical state of a membrane-less organelle, and the type of organelle.¹¹ Live imaging techniques are used to determine the mobility of organelle components. These methods include FRAP to determine the rate of recovery after bleaching, evidence of droplet fusion or fission, and for in vitro studies, fluorescence polarization to determine rate of spinning and single particle movement. The shape of the organelle can also provide insight into the state, because liquid droplets tend to be round, while gels are irregular in shape. The type of membrane-less organelle is generally determined by imaging hallmark proteins, which are described below.

3.1. Stress granules

Two of the most prominent types of RNA granules present in the cytoplasm are stress granules (SGs) or processing bodies (P-bodies) (Fig. 3). These organelles have distinct functions in the cells. In SGs the mRNA and associated proteins coalesce as a mechanism to protect cells from stress. Meanwhile, P-bodies function to store or degrade translationally repressed mRNA. The corresponding mechanisms of collection of mRNA into these physiological granules also differ mechanistically. It is also important to note that SGs are likely pleiotropic. Although SGs are commonly referred to as one homogeneous species, the reality is much more complicated. Proteomic studies show that SGs differ among cell types and stresses.⁴² Study of disease linked proteins show that the stresses leading to granules that contain Tar DNA binding protein-43 (TDP-43), FUS or T-cell intracellular antigen 1 (TIA1) differ.^45–48

Fig. 3.

RBP and protein aggregation in the cell during stress. RNA-binding proteins (RBPs) such as, TIA1, TDP-43, and hnRNPs are predominately nuclear however, during stress (A) they translocate to the cytoplasm through the nuclear pore complex (NPC) and interact with cytoplasmic RBPs (G3BP, PABP, FMRP, GW182). (B) Nucleation occurs when the RBPs and proteins (Dcp1/2, Edc3/4, Xrn1, Lsm 1–7) phase separate with translationally stalled mRNPs with pre-initiation complexes (PIC-mRNPs) to form the “cores” of the stress granules (SGs) and P-bodies (PBs). RBPs and proteins also interact with mRNPs to form transport granules. (C) Subsequently, additional RBPs and proteins phase separate to form the “shells” of the SGs and PBs. (D) After acute stress has resolved, the transient RNA granules disperse through interaction with disaggregases (such as, VCP and Transportin) or are degraded through the autolysosome. The autolysosome is formed first by the surrounding of the aggregate by the LC3+ “limiting membrane”. Once completely surrounded the now autophagosome fuses with the lysosome. The lysosome is capable of enzymatically digesting the aggregate. (E) Following granule dispersal, the predominately nuclear RBPs return to the nucleus. (F) Alternatively, chronic stress can prevent SG dispersal leading to SGs becoming gel-like and aggregating. RBPs receive post-translational modification such as, phosphorylation that affect their mobility. (G) These phosphorylated RBPs can aggregate to form oligomers. (I) and (J) RBPs and other proteins (such as, phosphorylated tau) can phase separate and seed their own aggregation independent of the SG pathway. (K) With continued stress these protein oligomers can aggregate further to form fibrils or amorphous aggregates leading to cellular injury. Alternatively, aggregated proteins can be degraded in the autolysosomal cycle.

SGs are formed from mRNA with stalled pre-initiation complexes (PICs) that consist of small 40S ribosomal subunits and translation initiation factors.^42,49 During basal conditions, the PIC is formed by the eIF4F complex. The eIF4F complex is comprised of eIF4E which recognizes the 5′ end of the mRNA, eIF4A which recognizes the 3′-poly(A) tail-bound PABP, and eIF4G which is the backbone.⁵⁰ eIF4F is joined by both the eIF2 ternary complex and the eIF3–40S ribosome to form the 48S PIC. This complex then joins the 60S ribosome to initiate mRNA translation.⁵⁰ The formation of the initial eIF4F is partially regulated by the serine/threonine kinase, mTOR.⁵¹ mTOR phosphorylates the eIF4E-binding protein (4E-BP) which prevents binding of 4E-BP to eIF4E. However, during stress mTOR is inactivated and the hypo-phosphorylated 4E-BP binds to eIF4E, preventing functional eIF4F formation.^51,52 Inactivation of mTOR can be achieved through stressors such as, oxidative stress, UV radiation, and aggregated proteins. Preventing the formation of the eIF4F complex results in the recruitment of RBPs to the 40S-mRNA complex into SGs.

The eIF2 ternary complex is also an important checkpoint in the formation of SGs. eIF2alpha brings the initiator tRNA $\left({\text{tRNA}}_{\text{i}}^{\text{Met}}\right)$ to the 40S ribosomal subunit.⁵⁰ This complex recognizes the start AUG codon on mRNA and begins translation through the exchange of GTP. However, during stress eIF2alpha is phosphorylated on serine 51 and the GTP exchange mechanism is inhibited. There are four different kinases that can phosphorylate eIF2alpha depending on the type of stress.^53–55 This phosphorylation leads to stalled 40S-mRNA complexes that associate in the cytoplasm with RBPs to form SGs.

SGs assembly begins with the translocation of RNA binding proteins (RBPs) from the nucleus to the cytoplasm through the nuclear pore (Fig. 3A). Once in the cytoplasm, RBPs such as TIA1, TIAR, TTR and GTPase activating protein binding protein 1 (G3BP1) interact with stalled 40S-mRNA complexes to form SG nuclear cores (Fig. 3B).^56–58 It is important to note, though, that recent evidence highlights a potentially more important roles for G3BP1 in regulating turnover of RNA and stalled ribosomes.^59,60 Other RBPs can join the SG to form the shell. The SG classically persists only transiently, and disperses once the stress is removed with the help of disaggregases, such as valocin containing protein 1 (VCP1).⁶¹ After dispersal, the nuclear RBPs return to the nucleus and normal mRNA processing is restored.

3.2. Disease associated mutations in RNA binding proteins are linked to SG biology

Many of the genes that exhibit mutations associated with amyotrophic lateral sclerosis (ALS) or frontotemporal dementia (FTD) are genes whose protein products are involved in SG biology, and many of these are RNA binding proteins. Mutations in the RNA binding proteins TDP-43, FUS, ATXN2, EWSR1, TAF15, heterogeneous nuclear ribonucleoprotein (HNRNPA A1 and A2B1), MATR3, ANG and TIA1 are all associated with ALS or FTD.^62–68 The mutations generally act to increase the accumulation of aggregated forms of these proteins. Studies increasingly suggest that the biggest effect of the mutations is to reduce the rate at which aggregates disperse, once formed. The aggregation promoting mutations tend to occur in the LCD domains, however they can also occur in other parts of the protein that impact on aggregation propensity. FUS presents a notable exception to this generalization because most of the disease-linked mutations in FUS act to disrupt the nuclear localization signal that is located at the C-terminus. These mutations act to increase the amount of FUS in the cytoplasm, which results in more FUS granules, more persistent FUS granules and thus more time to form irreversible aggregates.

The most common mutation associated with familial ALS is a hexanucleotide repeat expansion in front of the gene C9orf72.^69,70 This mutation produces multiple types of pathology, RNA foci and dipeptide repeats (DPRs); we discuss these in more detail below.^71,72 The DPRs are sticky and interfere with many functions in neurons, including increasing the accumulation of persistent SGs.^73–75

Mutations associated with ALS also occur in proteins that regulate removal of persistent pathological SGs and irreversible aggregates (Fig. 3D and K). In the sequestrome (SQSTM1/p62) system, VCP and UBQLN2 both play important roles in removing such aggregates, and mutations in both proteins are associated with ALS, FTD and muscular dystrophies.^76–79 These mutations reduce the ability of these proteins to remove aggregates of other proteins. The disease mechanism for SQSTM1 is not entirely clear though because SQSTM1 also utilizes LLPS in its biology, which means that mutations might also act by increasing the accumulation of SQSTM1 aggregates.⁸⁰ The biology of SQSTM1 is discussed in detail below.

3.3. P-bodies

P-bodies form via LLPS due to the interaction of mRNA decay machinery and translationally halted mRNA.⁸¹ The mRNA decay proteins that make up P-bodies include the 5′ decapping complex, Dcp1/Dcp2, a 5′ to 3′ exonuclease, Xrn1, and enhancers of mRNA decapping, Edc3, Edc4, Pat1, LSm1–7, and DDX6. The process of LLPS with these P-body proteins is similar to that for SG proteins, the only difference being the actual protein and mRNA components of the P-bodies. Like SGs, P-bodies have a distinct core-shell arrangement with core nucleating proteins occurring toward the center and secondary nucleating components occurring in a more peripheral position.⁸²

The process of phase separation of complex RNA granules is through formation of multi-component complexes containing more than one phase separated species. P-bodies contain the mRNA decay machinery that phase separate around a core Dcp1:Dcp2:Edc3 complex. The decapping enzyme, Dcp2, has multiple short linear motifs in its disordered region which facilitate its phase separation.⁸³ The Edc3 protein has three domains: a Lsm N-terminal domain, an IDPR, and C-terminal YjeF-N domain.⁸⁴ The N-terminal of Edc3 is necessary for protein dimerization which stabilizes its interaction with Dcp2 while the IDPR region interacts with mRNA. The exonuclease is recruited to the P-body by Dcp1 which activates Dcp2 and binds Xrn1.^85,86

These proteins influence the size and number of the P-bodies. Increasing the protein Edc4 increases P-body number and size while silencing this protein eliminates P-bodies. In contrast, silencing Dcp2 increases P-body number and size. A microprotein, non-annotated P-body dissociating polypeptide (NoBody), binds to Edc4 and is involved in P-body formation.⁸⁷ Overexpression of NoBody eliminates P-bodies, while silencing the gene increases the number of P-bodies in the cell. In cells, P-bodies are constitutively present but can also form de novo with increased levels of translationally stalled mRNPs in the cytoplasm. Despite the presence of mRNA decay machinery, the role of P-bodies is unclear. P-bodies may be involved in translational repression, RNA storage, and/or RNA decay.^81,88,89

3.4. Nuclear pore components

LLPS also plays a critical role in nucleocytoplasmic transport through nuclear pore complexes (NPCs). NPCs are macromolecular complexes involved in the active transport of proteins, ribosomes, mRNA, and rRNA across the membrane and the diffusion of small molecules across the nuclear envelope (NE). The proteins of the nuclear pore exist in a phase separated state (LLPS).⁹⁰ Permeability of the NPC barrier is selective so that large molecules, ≥30 kDa in mass or ≥5 nm in size, are not able to freely diffuse without assistance.^91,92 This assistance is gained through nuclear transport receptors (NTRs) and a coupled energy system mediated by the protein RanGTPase.⁹³ There are many types of NTRs with selectivity for different cargo, these NTRs can include importins/exportins, NTF2, and Hikeski.⁹³ The importin/exportin family is particularly interesting because recent evidence indicates that they also act as disaggregases.^36,94,95 This disaggregase function appears to act on proteins that accumulate in neurodegenerative diseases, such as FUS, HNRNPs and TDP-43, which suggests a potential role in the pathophysiology of neurodegenerative diseases related to these proteins.^36,94,95 The NPC also contains nucleoporins (Nups) which have three important functions in the NPC¹: they form a scaffold by binding to the NE,² they form the permeability barrier and³ they bind NTRs to facilitate transport. The characteristic feature of Nups that allow them to accomplish these functions is their dipeptide motifs of phenylalanine-glycine (FG) found on long uncharged IDPRs. FG-Nup monomers phase separate due to the interactions between their IDPRs, but there are different degrees of cohesiveness due to heterogeneity in length and sequencei. FG-Nups, specifically Nup98 and Nup49, phase separate in vitro into liquid droplets prior to forming a hydrogel. The permeability barrier formed by FG-Nups in the NPC is present in both phases in vitro. Nups also appear to play an important role in neurodegenerative diseases because they exhibit a strong tendency to form cytoplasmic aggregates as other pathological protein aggregates accumulate.^48,96–99

3.5. Nuclear bodies

The nucleus is filled with complexes of proteins that interact with DNA or RNA, which are termed nuclear bodies (NBs). Virtually all of these complexes exist as phase separated membrane-less organelles. One of the largest NBs is the nucleolus. The nucleolus is an energy dependent site of ribosome biogenesis. Nucleoli are comprised of a fibrillar center (FC), a dense fibrillar component (DFC), and a granular component (GC).³ The FC forms the core and contains the transcription machinery including, RNA polymerase (Pol I). The DFC is enriched with the protein fibrillarin (FIB1), a small nucleolar ribonucleoprotein (snoRNP).^100,101 A striking demonstration of the liquid nature of nucleolar components comes from imaging studies showing that DFCs from different nucleoli are able to fuse together and form a core with a GC surrounding it.³ The GC has an important role in nucleolus formation due to the protein nucleophosmin (NPM1/B23). This protein participates in phase separation through binding of arginine-rich short linear motifs (R-motifs) to its N-terminal oligomerization domain (OD) which contains a short-disordered region.¹⁰² Phase separation of NPM1 is enhanced by the presence of RNA through interactions with the N-terminal domain, although the C-terminal domain also binds nucleic acids.¹⁰² Finally, although LLPS is largely a passive phenomenon, ATP is required for transport of nucleolar components from the nucleus to the cytoplasm.^103,104

Super-enhancers are another type of NB that acts via LLPS. Super-enhancers are transcriptional complexes that contain RNA polymerase II (Pol II), transcription factors, RBPs, and co-activators. These super-enhancers phase separate into membrane-less organelles in order to facilitate rapid and specific gene expression.¹⁰⁵ Three of the RBPs in these complexes are notable because mutations in each protein causes ALS: FUS, EWSR1, and TAF15, these proteins phase separate by binding the C-terminal of Pol II.^106,107

Cajal bodies (CBs) are NBs that contain the splicing machinery and telomerase activity. CBs are the site of small nuclear ribonucleoprotein (snRNP) biogenesis and facilitates formation of the spliceosome. The spliceosome contains uridine(U)-rich snRNAs U1, U2, U4, U5, and U6 each of which is bound to Sm proteins. Cajal bodies provide a good example of a membrane-less organelle that exhibits clear core and shell structure because the core of CBs are resistant to high salt, heparin, and urea.¹⁰⁸ The snRNPs undergo preprocessing in CBs before moving to paraspeckles and nucleoli, respectively. Despite being membraneless organelles, interactions in CBs among the low-complexity protein p-80 coilin create clearly defined coiled structure that is evident from electron microscopy images.¹⁰⁹ Coilin is required for tethering of snRNPs while Nopp140 is necessary to tether small CB specific RNPs (scaRNPs) in CBs.¹¹⁰ Coilin also presents an excellent example of how post-translational modifications can regulate phase separation and RNA granule function. Coilin is a phosphoprotein and undergoes many post-translation modifications that alter its interactions.^111,112 Increasing coilin phosphorylation decreases its self-association and mobility.¹¹¹ Thus, CBs present an example where the role of phase separation and the regulation of phase separation are readily apparent.

Promyelocytic leukemia (PML) bodies are stress-responsive NBs involved in DNA damage, telomere functioning,⁴¹ and viral monitoring.^2,113 The PML protein has an important N-terminal RING finger domain that is required for PML body formation. PMLs present the complexity of how post-translational regulate membrane-less organelles/RNA granules. Oxidation causes the PML protein to form a spherical shell that binds the E2 small ubiquitin-like modifier (SUMO)-conjugating enzyme (UBC9).¹¹⁴ The UBC9 protein is recruited through the SUMO-interacting motif (SIM), which ultimately facilitates SUMOlyation.¹¹⁴ The addition of SUMO to the structure allows recruitment of new components that segregate to the core.¹¹⁴ Thus, oxidation ultimately leads to a complex phase separated membrane-less organelle with a core and shell structure.

Paraspeckles are NBs with a role in transcription, RNA and protein sequestration, and miRNA processing. Paraspeckles are arranged into distinct core, shell, and bridge zones like many other NBs. FUS and RNA binding motif 14 (RBM14) are found in the core and are required paraspeckles to form in vivo.^115,116 Paraspeckles present a clear example of how RNA can selectively regulate a membrane-less organelle. The 23,000 nucleotide noncoding RNA (lncRNA) NEAT1_2 acts as a scaffold enabling paraspeckle formation and stability.^40,117 Increasing the concentration of NEAT1_2 results in an increased numbers and size of paraspeckles.¹¹⁸ The structure, function and regulation of each of the NBs provides distinct insights into the mechanisms that control the process of LLPS and the structures that we variously term RNA granules or membrane-less organelles.

3.6. The tension between LLPS and irreversible aggregation

LLPS has profound biological benefits because it provides a powerful yet low energy mechanism for enabling spatial organization of proteins. However, the nature of LLPS also presents an inherent problem because the elevated concentration of proteins in phase separated droplets increases the rate at which they can transition to insoluble aggregates. The interaction of stickers among LCDs that allows for phase separation brings proteins to a lower energy state. Some types of proteins, though, can achieve an even lower energy state by forming amyloids, often achieved through β-sheet interactions.^119,120 These transitions are accelerated by the higher concentration of each protein present in phase separated protein droplets. Once these insoluble conformations form, the proteins begin to aggregate and they serve as a substrate for seeding aggregation of other similar proteins. With time the accumulation of aggregates interferes with cell functioning.

The tension between the benefit of LLPS and the harm of protein aggregation is something that evolution has addressed. The cell has an entire system designed to minimize protein aggregation and produce a state of proteostasis. Our cells produce numerous heat shock proteins, chaperones and disaggregases that can normalize the conformation or detect the aggregates and send them for degradation through the autolysosomal machinery.^121,122 The system of heat shock proteins (HSPs) and chaperones is transcriptionally regulated by heat shock factor-1 (HSF1), which itself is regulated by HSP90 proteins. Protein aggregates that are not re-solubilized become labeled by ubiquitin and/or identified by SQSTM1, whereupon they are shunted through the autolysosomal system (Fig. 3D and K).^123,124 Dividing cells have an additional protective mechanism in the armory because during cell division protein aggregates are caged by vimentin and then differentially shunted to one of the two daughter cells, allowing one cell to continue functioning well while sacrificing the other daughter cell as a “garbage dump”.¹²⁵ While we are young, the autolysosomal system efficiently removes all of the deleterious protein aggregates, allowing the cell to function optimally. As we age, though, the autolysosomal system begins to work less efficiently, in part because of decreased fusion of the autophagosome with the lysosome, and because of reduced lysosomal function.^9,10,126 The reduced flux through the autolysosomal system enables the accumulation of insoluble proteins. These insoluble proteins interfere with cellular function and ultimately lead to age related degeneration. Hence, the very nature of LLPS ultimately sets the stage for age-related cell injury and degeneration.

Our genomes all differ, though, and many people harbor polymorphisms or mutations in genes that are necessary for cell proteostasis. Mutations in proteins such as VCP, SQSTM1, charged multivesicular body protein 2B (CHMP2B) and UBQLN2 interfere with proteostasis, and cause early onset neurodegenerative diseases or muscular dystrophies. Dysfunction of the proteins made by these genes impairs proteostasis enabling the accumulation of protein aggregates as we age. The resulting diseases generally appear in non-dividing tissues, such as neurons and muscle cells. Meanwhile, dividing cells appear to be less sensitive to these mutations, perhaps because these dividing cells retain the ability to maintain the health of the population by shunting protein aggregates into selected daughter cells. Thus, mutations or polymorphisms in our proteostasis machinery set the stage for age related degenerative diseases that largely appear in non-dividing tissues.

3.7. Aggregate removal: SQSTM1/p62 and ubiquitin

Phase separation even plays a role removal of aggregated proteins. SQSTM1/p62 has been frequently mentioned in this review because it functions to identify aggregated proteins and target them to the autolysosomal or proteasomal system for degradation. LLPS plays an important role in the biology of SQSTM1 because it phase separates into liquid droplets in the presence of poly-ubiquitinated proteins.¹²⁷ The characteristics of ubiquitination affect the speed and likelihood of phase separation. Longer ubiquitin chains facilitate faster phase separation. Additionally, the type of ubiquitin chain affects the rate of phase separation. Ubiquitin forms multiple types of linkages, with K48 and K63 being the most common. SQSTM1 identifies K48 polyubiquitination, and targets proteins to the proteasome, while K63 linkage tends to target along proteasome independent pathways, including the autolysosomal pathway. The K63 poly-ubiquitinated chains found on proteins destined to undergo autophagy stimulate SQSTM1 phase separation.¹²⁷ Conversely, free K63- and K48-poly-ubiquitinated chains inhibit phase separation.¹²⁸ Phase separation of SQSTM1 is also regulated by binding to other proteins. For instance, binding of LC3B to SQSTM1 leads to a slowing of phase separation of the ubiquitinated protein.¹²⁹ Note the absence of RNA or DNA from the discussion of SQSTM1 phase separation; this is a clear example of a process of LLPS that does not require polynucleotides, which suggests that the role of LLPS in cell biology extends well beyond that or RNA or DNA binding proteins.

4. Pathological hallmarks of neurodegenerative diseases

The confluence of aging with genetic mutations or polymorphisms ultimately reduces functioning of the proteostasis machinery to a point where aggregated proteins begin to accumulate. Although the vast majority of the >800 RNA binding proteins likely exhibit LLPS, only a very small fraction of these proteins actually accumulates as protein aggregates in disease.¹²¹ The RNA binding proteins that accumulate in disease are those possessing LCDs that are extremely hydrophobic or convert to β-sheets readily. These proteins are prone to form insoluble protein aggregates, which become the hallmark protein aggregates in that characterize neurodegenerative diseases and muscular dystrophies (Fig. 3G–K).

4.1. Amyotrophic lateral sclerosis and FTD-TDP

The relationship between RNA binding proteins and disease first became apparent with the discovery of TDP-43 as the major protein that accumulates in the spinal cords of patients with Amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD).^130–132 TDP-43 is a nuclear protein that normally functions as a component of the spliceosome, but translocates to the cytoplasm during times of stress, associating with SGs.¹³³ TDP-43 can phase separate due to interactions in its LCD.³⁷ TDP-43 fibrils are also capable of inducing further TDP-43 aggregation in cells (Fig. 3I and J).^48,134,135 Disease associated mutations in TDP-43 tend to occur in the LCD, and increase the propensity of the protein to aggregate, which shifts the balance between LLPS and irreversible aggregation such that aggregated TDP-43 accumulates in the neurons over time. The aggregates typically accumulate in the cytoplasm, occurring with a corresponding loss of nuclear TDP-43.^133,136 In some cases of FTD (Type IV), visible accumulations of TDP-43 aggregates are present in the nucleus, but studies suggest that stress can cause TDP-43 can to form insoluble, non-functional nuclear aggregates even when no visible aggregates are present in the nucleus.^136–138 The loss of nuclear TDP-43 function, perhaps combined with the loss of TDP-43 function at other sites outside of the nucleus (e.g., at synapses) is thought to precipitate the neurodegeneration that we associate with ALS.

Mutations in other RNA binding proteins also lead to ALS and FTD. As mentioned, most of these RNA binding proteins are characterized by protein sequences that are particularly hydrophobic, which means that small changes in aggregation propensity caused by mutations are sufficient to tip the LLPS/aggregation equilibrium for these proteins such that they accumulate as pathological aggregates with age. After TDP-43, FUS/TLS is perhaps the best characterized of the disease-associated RNA binding protein. The most important function of FUS/TLS appears to be in the regulation of responses to DNA damage, where it quickly localizes to DNA and then later translocates to the cytoplasm where it becomes associated with a type of RNA granule resembling stress granules.^45,46 Mutations in FUS/TLS occur in two regions. Most mutations occur at the C-terminus in the nuclear localization signal; the mutations reduce nuclear transport of FUS/TLS, which over time leads to the accumulation of cytoplasmic aggregates, neuropathology and neurodegeneration. A much smaller set of mutations occur in the LCD domain, where they act in a manner similar to mutations in other RNA binding proteins by increasing the aggregation propensity.^84,139

The most common ALS/FTD mutation is a hexanucleotide (G₄C₂) repeat expansion in an intronic region of C9ORF72.^69,70 These expansions account for 30% of familial ALS cases and about 3% of sporadic ALS cases. Mutations in C9ORF72 are observed in a diverse array of neuropsychiatric diseases, ranging from FTD to schizophrenia. The repeat expansions can range from dozens of repeats to >1500 repeats. The repeat expansions cause formation of RNA lariats, RNA foci and Dipeptide repeats (DPRs). Accumulating evidence suggests that the DPRs are the toxic species that causes neurodegeneration.¹⁴⁰ DPRs form because the repeat sequences interfere with normal translation, leading to RAN translation, which results in inappropriate translation of the hexanucleotide repeat.^71,72,141 The resulting translation products include 5 DPRs, that differ depending on the coding frame that is translated: glycine–alanine (GA), glycine–arginine (GR), proline–alanine (PA), proline–arginine (PR) and glycine–proline (GP). The mechanism of toxicity associated with the repeats appears to be highly pleiotropic, but includes interference in nuclear-cytoplasmic transport, dysfunction of SG formation and protein synthesis, induction of TDP-43 aggregation and interactions with NPSM1 in nucleoli to decrease mobility and binding affinity to rRNA.^{74,75,142–145}

4.2. Alzheimer’s disease and other tauopathies

RNA binding proteins also contribute to the pathophysiology of tauopathies, including Alzheimer’s disease (AD) and all types of FTD.^{130,136,146,147} Alzheimer’s disease (AD) is characterized by extra-cellular β-amyloid plaques and intracellular hyperphosphorylated neurofibrillary tangles (NFTs). The pathophysiology of AD is thought to stem from the amyloid cascade hypothesis, but the genetics of late onset AD (LOAD) also suggests an important role for inflammation.^148,149 Our current understanding of AD posits that high levels oligomeric Aβ (oAβ) cause stress for neurons.^150–152 Familial cases of AD produce oAβ at sufficient levels to induce neurotoxicity directly. For cases of LOAD, the oAβ accumulates slowly, but the toxicity is augmented by the increased inflammation that is associated with aging. Together, oAβ and inflammation act together to injure the neuron. The mechanism of toxicity of oAβ is likely pleiotropic, but the most important mechanism of action appears to be NMDA receptors which act in concert with PrP to cause excitotoxicity.^148,153 As the neurons become injured signals are sent to microglia which elicit an inflammatory response.

Stress signals arising from oAβ and inflammation elicit a stress response that is transmitted throughout the neuron. The microtubule associated protein tau (MAPT) plays a key role in transmitting this signal. This discussion will begin with a brief review of the biology of MAPT. MAPT is intrinsically disordered and lysine-rich, which results in inhomogeneous charge distribution¹⁵⁴). Specific tau isoforms are more likely to undergo phase separation than others.^155–157 Full length tau contains two N-terminal inserts, two proline-rich regions, and four 32-residue long imperfect repeat sequences.¹⁵⁸ The 3R and 4R tau are more likely to be found in NFTs in AD patient brains.¹⁵⁹ MAPT normally functions in the axon where it promotes microtubule stability, which is important to maintaining the immensely long axonal arbors.¹⁵⁸

Stress dramatically changes MAPT biology. Stress activates serine/threonine proline directed kinases, such as GSK3β, CDK5 and MARK, which phosphorylates tau in a process that is commonly referred to as hyperphosphorylation.^158,160 Hyperphosphorylated tau behaves as a different species, exhibiting three critical differences: (A) No binding to microtubules, (B) Cytoplasmic accumulation and (C) Rapid oligomerization. These three characteristics combine together allowing oligomeric tau (oTau) to play very different roles. In the cytoplasm phosphorylated oTau regulates the stress response of the protein synthesis/RNA translation pathway, while at the synapse phosphorylated oTau regulates synaptic transmission.^47,161–165

The relationship between mRNA and RNA binding proteins is particularly interesting. Tau has been known interact with RNA since 1996, when it was demonstrated that RNA promotes tau fibrillization.¹⁶⁶ The significance of this interaction has only become apparent recently. Biophysical studies demonstrate that tau phase separates in a manner that is increased by the presence of RNA.^155,156 This sensitivity to RNA belies a more profound biological role for tau in RNA metabolism. Hyperphosphorylated tau (pTau) rapidly generates oligomeric tau (oTau), and these species selectively interact with many components of RNA metabolism. pTau (and presumably oTau) regulates the translational stress response; it both promotes and is required for a normal translational stress response, which is associated with stress granule formation.⁴⁷ pTau also binds to the ribosomal protein RPS6.^162,163 Tau also appears to play a role in ribosomal genesis, because the tau isoform, 2N4R, localizes to the nucleolus under basal conditions, but moves away from the nucleolus to other parts of the nucleus upon phosphorylation during stress.¹⁶⁷ Many proteins change function upon phosphorylation or oligomerization. For example, the binding profiles of p53 differs markedly between monomeric and oligomeric forms.¹⁶⁸ Oligomerization is also known to initiate signaling of many receptors, and regulates the function of many intracellular signaling cascades.¹⁶⁹ The biological logic of tau phosphorylation and oligomerization might lie in the exact same processes, where phosphorylation and oligomerization of tau switches its binding partners and functions from that of regulating microtubules in the axon to that of regulating the translational stress response in the cytoplasm.

5. Conclusion

Accumulating evidence indicates that phase separation plays a profoundly important function in regulating cell biology. LLPS brings together functionally related proteins through the intrinsic biophysics of proteins, allowing them to form membrane-less organelles that carry out specific functions. RNA and DNA present long chained charged polymers that promote LLPS, and consequently most RNA and DNA binding proteins carry out their functions through a process of LLPS. However, the very nature of bringing together proteins with low complexity domains renders them prone to aggregation. This creates an inherent tension between LLPS and irreversible protein aggregation, which the cell handles through an integrated set of proteins and organelles that maintain proteostasis. Age and/or mutations impair the equilibrium between LLPS and irreversible protein aggregation. As this equilibrium tilts towards the accumulation of irreversible protein aggregates, non-dividing cells such as neurons become dysfunctional and begin to degenerate. This pathophysiology ultimately leads to multiple forms of neurodegenerative diseases, the specific type of which reflects the temporal and spatial accumulation of different aggregating proteins.