Physiological, pathological, and targetable membraneless organelles in neurons

Abstract

Neurons require unique subcellular compartmentalization to function efficiently. Formed from proteins and RNAs through liquid-liquid phase separation, membraneless organelles have emerged as one way in which cells form distinct, specialized compartments in the absence of lipid membranes. We first discuss membraneless organelles that are common to many cell types as well as those that are specific to neurons. Interestingly, many proteins associated with neurodegenerative disease are found in membraneless organelles, particularly stress and transport granules. Next, we review possible links between neurodegeneration and membraneless organelles and the hypothesis that the protein and RNA inclusions formed in disease are related to the functional complexes occurring inside these membraneless organelles. Finally, we discuss the hypothesis that protein posttranslational modifications, which can alter phase separation, can modulate membraneless organelle formation and provide potential new therapeutic strategies for currently untreatable neurodegenerative diseases.

Liquid-liquid phase separation and membraneless organelles

Cells, especially neurons, have many internal compartments. While many compartments (i.e. organelles) are membrane bound, recent work has highlighted the prevalence of specific compartments not surrounded by a phospholipid bilayer, called membraneless organelles (MLOs, see glossary). Liquid-liquid phase separation (LLPS), a phenomenon where mixtures of two or more components self-segregate into distinct liquid phases (for example demixing of oil and water), appears to underlie the formation of many MLOs. In cells, proteins, nucleic acids, and other biomolecules demix from the surrounding solution to form a liquid phase densely packed with specific biomolecules, still containing water, and distinct in composition from the surrounding cytoplasm or nucleoplasm. Many distinct MLOs are found in the nucleus and cytoplasm of eukaryotes and in prokaryotes [1]. Nuclear MLOs include the nucleolus [2, 3] and paraspeckles [4, 5], whereas cytoplasmic MLOs include stress granules [6–9], RNA transport granules, and P-bodies [10]. Many of these structures have long been identified as granules or puncta, whereas other structures have been discovered recently. The diversity of MLOs implies a diversity of biological functions, as also suggested by the distinct composition of protein and nucleic acid components that make up MLOs [8]. Despite this complexity, the proposed biophysical functions of MLOs seem to fall into a few broad categories (reviewed in [11]), including: sequestering molecules (e.g. preventing RNA translation, as in transport and stress granules), regulating reaction kinetics by concentrating components, and altering reaction specificity by sequestering specific components to increase the local concentration (as in P-bodies) [11].

In this review, we briefly discuss the biophysics of LLPS before describing membraneless organelles specific to the nervous system. We then examine the role of LLPS and MLOs in disorders of the nervous system, particularly neurodegenerative diseases. Finally, we discuss emerging evidence that posttranslational modifications can modulate LLPS, and hence may offer potential therapeutic targets in the treatment of MLO-associated neurodegenerative diseases.

What drives LLPS?

Identification of the protein components and the mechanisms behind the formation and dissociation of MLOs has been an area of intense research. Hundreds of protein [12] and RNA [13] components of RNA granules have been reported. The proteome of stress granules, transient MLOs that form and sequester RNAs to prevent translation when the cell is exposed to exogenous stress, have been extensively characterized [8]. Many of the identified granule-associated proteins contain RNA-binding domains and/or intrinsically disordered regions (IDRs) [8, 12] (Box 1). IDRs of granule-associated proteins are often enriched in certain amino acids or sequence motifs that make them distinct from other IDRs. Some granule-associated IDRs contain arginine, often arranged in RG or RGG motifs that have unique roles in protein and nucleic acid interactions (reviewed in [14]). Others are classified as prion-like domains (PLDs), defined by amino acid sequence compositional similarity to yeast prion proteins (i.e. rich in asparagine and glutamine), which are prone to functional self-assembly [15] (Box 1). In addition, many granule proteins with PLDs harbor neurodegenerative disease-associated mutations and are thought to drive aberrant cytoplasmic aggregation (reviewed in [16]).

Text box.

Disordered proteins, prion-like domains, and in vitro LLPS

Many granule proteins, particularly those capable of LLPS in vitro, are intrinsically disordered proteins (IDPs) or have intrinsically disordered regions (IDRs). These proteins or domains have no persistent secondary structure but are structurally dynamic and sample many different conformations. IDPs and IDRs can be involved in protein-protein or protein-nucleic acid interactions, and self IDR-IDR interactions contribute to the ability of a protein to undergo LLPS. Multivalency, the ability to interact with multiple partners, is an important characteristic of proteins capable of LLPS. Some IDPs or IDRs are also low complexity (LC) in that their amino acid sequence is biased towards a small number of amino acids that are overrepresented when compared to the average proteome. Some of these LC sequences have short repeats of a few amino acids, forming repeat motifs (including Gly-Tyr-Gly or Arg-Gly-Gly) that may be important in the formation of MLO and can also serve as PTM sites. Some LC domains are also prion-like. Prion-like domains (PLDs) are similar in amino acid composition to yeast prion proteins because they contain a high proportion of polar amino acids with few hydrophobic or charged residues. It is possible that misfolded or aggregated PLDs can propagate via cell-to-cell transfer although this has not been demonstrated for most PLDs involved in neurodegenerative diseases.

While in vivo experiments are necessary to understand the physiological relevance of LLPS, in vitro experiments are critical to understanding the forces and interactions holding these phase-separated states together. Often, these in vitro experiments are performed on recombinant protein, although it can be difficult to purify and work on proteins with disordered and prion-like or aggregation prone domains. The experimenter will induce LLPS by changing salt conditions, removing denaturants, changing protein concentration, changing the temperature, or adding multivalent binding partners. Observing the presence or absence of LLPS at many conditions can lead to a better understanding of the physical properties of the protein and the forces that drive it to phase separate (i.e. primarily hydrophobic, electrostatic, etc.). Understanding the forces driving LLPS may help lead to drugs targeting LLPS or protein aggregation as we better understand how to disrupt or promote LLPS and aggregation. Additionally, experimenters can test individual types of PTM or even individual PTM to determine their impact on LLPS, which both aids our understanding of the forces behind LLPS and may lead to more targeted therapies.

Given the diversity of biomolecular components in MLOs, identification of binding/interaction motifs that drive phase separation is ongoing (Figure 1). Interaction between IDRs has been suggested to drive phase separation in some MLOs [17–19]. Structural studies have shown that the monomeric and in vitro phase separated states formed by several proteins, including the disease-associated proteins FUS and hnRNPA2, as well as DDX4, are predominantly structurally disordered [17, 18, 20]. A mutational and sequence-comparison approach found that interactions between tyrosine and arginine are critical to phase separation of FUS [21]. Furthermore, a bioinformatic study found that planar π-π interactions are enriched in many proteins known to undergo LLPS [22]. Hence, IDR repeats act as multivalent motifs, forming multiple simultaneous weak interactions between disordered binding partners (including themselves), contributing to IDR-driven LLPS [23]. Distinctly, protein secondary structure can play a role in LLPS. Phase separation of the prion-like C-terminal domain of TDP-43 is stimulated by self-interaction of a transiently structured α-helix formed by a conserved aliphatic-rich segment, and can be tuned by mutations altering α-helicity [19, 24]. Furthermore, interactions between several multivalent structured domains connected by flexible linkers, structured domains and IDRs, and structured domains and other predominantly disordered polymers (i.e. multidomain RNA-binding proteins with mRNAs) can also mediate phase separation [25–27].

Figure 1: Interactions driving MLO formation.

Many different types of interactions have been shown to drive or contribute to LLPS and MLO formation. These include RNA-RNA, protein-RNA (through RNA-RNA binding domain (RBD) interactions or RNA-IDR/IDP interactions), and different kinds of protein-protein interactions, like self-interactions between IDRs or interactions between an IDP and structured protein. Granule components exchange with the surrounding environment, from the dilute phase to the condensed phase. Within the granule, RNA and protein molecules both form multivalent interactions, meaning that one molecule can interact with multiple others simultaneously.

Translating in vitro biophysical studies to the interactions mediating LLPS in cells remains a challenge. Although cellular MLOs contain multiple proteins and RNAs, most in vitro and biophysical studies to date have focused on single or a small subset of proteins known to be in the MLO. Additionally, many studies have used crowding agents such as polyethylene glycol (PEG) to mimic the crowded intracellular environment in order to observe LLPS at physiological protein concentrations (<1 μM). Although crowding agents are generally assumed to be inert and to induce LLPS primarily by excluding volume, they do not replicate the diverse molecules present in the cellular environment [28] and may result in LLPS “false positives,” leading some to use purified proteins or cell lysate as crowders. Regardless, crowding agents may confound experiments to elucidate the biophysical interactions underlying LLPS. As such, careful controls with several different crowders (including their absence [29]) are important. In addition, experiments designed to test the interactions driving LLPS and how those interactions are altered by the presence of other granule components are needed for further elucidating the molecular interactions leading to LLPS.

Specialized membraneless organelles in the nervous system

Whereas many MLOs are found across different cell types, some neuron-specific MLOs have been discovered (Figure 2). These neuron-specific MLOs are notable in their functional diversity and unique constituent proteins and interactions. A prominent example of a neuronal cellular compartment where MLOs are found is the post-synaptic density (PSD). PSDs are protein-rich compartments adjacent to the post-synaptic membrane. PSDs concentrate neurotransmitter receptors, exchange components with the surrounding cytoplasm, and are altered as a result of synaptic plasticity. Like other MLOs, the primary components of PSDs, PSD-95 and SynGAP, are together capable of undergoing LLPS in vitro and in vivo [30]. PSD LLPS appears to be stabilized by intermolecular interactions between the folded PDZ domain of PSD-95 and a protein-binding motif of SynGAP [30]. Interestingly, mutations to SynGAP disrupting both its trimeric structure (and thus multivalent architecture) and LLPS lead to greater dispersion of SynGAP from the PSD and stronger synapses in response to long-term potentiation stimuli [30]. Therefore, structural organization of SynGAP may be important for dynamic localization to the phase-separated PSD and regulation of synaptic strength [30].

Figure 2: Membraneless organelles in the nervous system.

The nervous system relies heavily on MLOs and has a few specialized membraneless compartments, including the postsynaptic density (PSD), the synapsin-synaptic vesicle phase, and the active zone. Neurons use transport granules to move RNAs to sites of local translation. The myelin sheath is dependent on MLOs to transport myelin basic protein mRNA and hold lipid membranes in close proximity with myelin basic protein.

Like the post-synapse, recent evidence suggests that pre-synapse architecture is also maintained by multiple MLOs. One of these neuron-specific liquid phases is formed by synapsin, a protein that associates with the membrane of synaptic vesicles [31]. Unlike SynGAP and PSD-95, synapsin contains an IDR that is necessary and sufficient for LLPS [31]. In vitro phase separated synapsin can recruit proteins containing SH3 (Src homology 3) domains, which bind proline-rich IDRs, to partition into the liquid phase [31]. Unique to MLOs described thus far, this synapsin phase is also capable of sequestering lipid vesicles in vitro and loss of synapsins in mouse neurons leads to decreased and more dispersed vesicles at the synaptic terminal [31]. Additionally, phosphorylation of synapsin by CamKII dissociates droplets of synapsin with and without liposomes [31]. In addition to organization of synaptic vesicles by synapsin, the organization of the pre-synaptic active zone has also been suggested to resemble a liquid MLO [32]. RIM, a protein that is critical for the formation of the active zone, is capable of LLPS in vitro. Further, the SH3 domains of its binding partner RIM-BP can stimulate formation of in vitro LLPS droplets of RIM and incorporate into them [32]. Additionally, the cytoplasmic tail of the N-type voltage gated calcium channel α1 subunit can incorporate into RIM and RIM-BP phase-separated droplets, clustering the cytoplasmic tail of the calcium channels on a lipid bilayer as a potential model for pre-synaptic organization [32]. Though it remains unclear if (1) these pre-synaptic assemblies of synapsin and RIM truly function physiologically as MLOs, (2) they remain separate or co-assemble, or (3) other MLOs organize additional ion channels and components of neurotransmitter release and response, these studies do highlight the elegant way neurons may use MLOs to cluster materials necessary for signal transmission at synapses, while continuing to be primed for neurotransmitter release.

In neuronal processes, transport granules, or neuronal granules, play a critical role in the formation of long-term memory (reviewed in [33]). Long-term memory requires new gene expression and protein synthesis to become stable [34]; one hypothesis is that transport granules localize transcripts to synapses and repress translation until synaptic activity induces translation required for long-term memory formation. Many RNA binding proteins have been implicated in memory-associated mRNA transport, including FMRP, CPEB1 (cytoplasmic polyadenylation element binding protein), hnRNPA2, and staufen (reviewed in [33, 35]). These RNA binding proteins have different mRNA targets and may make distinct transport granules, although there is some evidence that several different mRNAs can be found in the same granule [36]. For example, hnRNPA2 [36] and FMRP [37] both associate with CamKII mRNA. Additionally, Ataxin-2, in particular its IDRs, is required for granule formation and subsequent long-term memory in Drosophila [37, 38]. Another hypothesis that has been raised as a mechanism for long-term memory formation is the prion hypothesis (reviewed in [35]), which states that synaptic activity induces a conformational change in proteins like CPEB that leads to the formation of a structured, stable oligomer complex that functions distinctly from the monomeric protein. While some transport granule proteins contain PLDs, there is no definitive evidence for a physiological structured prion form of other transport granule proteins. Rather, it is possible that the high proportion of prion-like domains in RNA binding proteins encodes their ability to undergo LLPS. Indeed, even the archetypal yeast prion protein, Sup35, which forms protein fibrils capable of protein-based phenotypic inheritance (i.e. information transfer), has been suggested to have evolved its prion domain to undergo in cell LLPS [39]. Though some transport granule proteins have been shown to undergo LLPS, transport granules have received relatively little attention compared to the better-studied stress granules. To deepen our understanding of transport granules, future studies should target their longevity, liquidity, protein and RNA components, and role in disease pathogenesis.

MLOs in the nervous system are not limited to neurons. In oligodendrocytes, the formation of the myelin sheath may involve MLOs. The mRNA encoding myelin basic protein (MBP), one of the major protein components of myelin, binds hnRNPA2 in transport granules for transport to sites of local translation in oligodendroglial processes (reviewed in [40]). It is thought that after MBP mRNA is locally translated, MBP oligomerizes to bring the two sides of the membrane close together and prevent the diffusion of soluble and membrane proteins into the myelin sheath [41]. Purified recombinant MBP can undergo LLPS in vitro when exposed to high pH, which neutralizes the highly positive net charge of MBP leading to MBP self-interaction [41]. A similar net charge neutralization leading to assembly may occur when MBP encounters the negatively charged membranes present in the cell [41]. However, it remains unclear if LLPS observed in vitro is present in vivo or how the phospholipid membrane and other myelin components affect MBP phase separation.

In summary, the diversity and crucial function of MLOs in neurons and oligodendrocytes is striking and there may be as yet unidentified MLOs in the nervous system. Furthermore, other cell types appear to use phase separation for the formation of system-specific MLOs. Indeed, a recent report found granules containing TDP-43 and RNA in muscle [42]. However, the heavy reliance of the nervous system on MLOs that contain aggregation-prone domains may explain the observation of inclusions of these proteins and the high number of LLPS prone proteins mutated in neurodegenerative diseases.

LLPS and neurodegenerative disease

As more genetic mutations are associated with inherited neurodegenerative disease, more granule components and proteins that undergo LLPS emerge as players in disease (reviewed in [43]). Although it remains unclear why ubiquitously expressed RNP granule proteins aggregate in neurodegenerative disease, a few hypotheses have emerged (Figure 3). One hypothesis is that increased protein concentration inside MLOs stochastically leads to the formation of persistent protein interactions (with greater probability in the presence of a disease mutation) that then leads to stable aggregate formation and neurodegeneration [44]. Another hypothesis is that disease-associated mutations (or the presence of aberrant RNAs or peptides derived from repeat expansion transcription or translation [45, 46]) reduce the liquidity of MLOs, disrupting their function and preventing their dissociation [46]. A third hypothesis is that aggregates of proteins not normally found in MLOs accumulate within MLOs, particularly stress granules. For instance, one study found aggregated forms of mutant SOD1, a protein associated with familial amyotrophic lateral sclerosis (ALS), accumulated in stress granules [7]. In contrast, granule associated RNA-binding proteins with aggregation prone prion-like domains may form inclusions away from MLOs or granules due to the absence of high concentrations of solubilizing RNA [47, 48]. Finally, phase separation could increase the formation of oligomers on pathway to toxic aggregates [42, 49]. While many neurodegenerative diseases have been associated with MLOs, primarily stress granules, the relationship is much clearer for some diseases than others. Below we will discuss the current data that supports the link between MLOs and several neurodegenerative diseases (Figure 4 and Table 1).

Figure 3: Hypotheses relating MLOs and aggregate formation in neurodegenerative disease.

A number of hypotheses exist to explain the connection between MLOs and protein inclusions/aggregates found in disease. Increased toxic oligomer/aggregate formation: The high protein concentration in a granule may increase the likelihood that mutant proteins will aggregate and not be dissociated with the rest of the granule. Solidification: What was once a liquid phase may persist and solidify. Accumulation of misfolded proteins: Misfolded or aggregated proteins may accumulate in MLOs. The high concentration inside MLOs may increase the formation of toxic oligomers stochastically, so they are not dissolved when the fluid parts of the granule dissociate.

Figure 4: Granules and neurodegenerative diseases/associated proteins.

(A) Neurodegenerative disease-associated proteins grouped by the type of granule where they have been identified in cells (stress or transport granules) or whether they are capable of undergoing LLPS in vitro. (B) Neurodegenerative diseases grouped by the type of granule(s) that has been implicated in that disease. Fragile X has been added to the schematic given the roles of FMRP, its associated protein, as an mRNA transport granule component and the recently raised possibility that it undergoes LLPS (see main text).

Table 1:

Summary of neurodegenerative diseases and LLPS/MLOs.

Protein	Diseases associated with	Evidence for disease association	Granule types	Evidence for In vitro LLPS	Evidence for In vivo LLPS or granule
FUS	ALS/FTD	Point mutations (NLS and others)	Stress	[18, 44, 89]	[44, 89]
TDP-43	ALS/FTD, AD	Point mutations, truncations	Stress, transport	[19, 90]	[47, 48, 90, 105]
hnRNPA1	ALS/FTD/MSP, MS	Point mutations in ALS/FTD/MSP, mislocalized in MS [64]	Stress, transport	[6]	[6]
hnRNPA2	ALS/FTD/MSP	Point mutation in ALS/FTD/MSP	Stress, transport	[17]	[106]
C9ORF72	ALS/FTD, SCA	G4C2 expansion in ALS/FTD, SCA [72]	Stress	[45, 46, 52]	[45, 46, 51, 52]
UBQLN2	ALS/FTD	Point mutations in ALS/FTD	Stress	[54, 107]	[54]
TIA1	ALS/FTD, AD, MS, SMA	Point mutations in ALS/FTD, mislocalized in AD [61] MS [64] and SMA [77]	Stress	[55, 56]	[55, 56]
Profilin	ALS/FTD, HD	Point mutations in ALS/FTD, modifies HD aggregation in vitro [66]	Stress		[57]
Ataxin-2	ALS/FTD, SCA	polyQ expansion, 32 or more in SCA, 29–32 is risk factor for ALS [73]	Stress, transport	[38]	[38]
Tau	AD	Mutated in AD	Stress	[29, 60, 108]	[29]
DJ-1	PD	Mutated in PD	Stress (P bodies)		[67]
Huntingtin	HD	PolyQ expansion in HD	Stress	[65, 66]	[65]
Staufen-1	SCA	Modifies SCA - recruited to aggregates and increased expression [74]	Transport		[74]
SMN	SMA	Mutated/reduced levels in SMA	Stress, transport		[77, 78]
FMRP	Fragile X, FXTAS	5’ UTR repeat expansion in Fragile X, FXTAS	Stress, transport	[22, 80]	[8, 13]

Amyotrophic lateral sclerosis/frontotemporal dementia (ALS/FTD):

ALS and FTD exist on a spectrum, with overlapping genetic causes and some patients displaying symptoms of both diseases. ALS is defined by loss of upper and/or lower motor neurons leading to paralysis, while FTD is characterized by degeneration of the frontal and temporal lobes and corresponding behavioral changes. After the identification of aggregates of mutated SOD1 in 1993, the number of identified genes mutated in ALS patients started increasing rapidly [50]. Interestingly, many of the new genes identified are RNA-binding proteins or proteins with functions related to RNA processing and many contain PLDs. While several ALS/FTD-associated RNA binding proteins are found in RNA granules [12]. FUS was the first to be shown to be capable of undergoing LLPS in vitro and in vivo [44]. Correspondingly, ALS-associated mutations in FUS can drive a liquid-to-solid transition [44]. Reports of similar phenomenon for other ALS/FTD-associated proteins followed, including TDP-43 [19], hnRNPA1 [6], hnRNPA2 [17], dipeptide repeats encoded by C90RF72 [45, 46, 51,52] (which can also alter nucleolar LLPS [53]), UBQLN2 [54]. TIA1 [55, 56], profilin [57], and ataxin-2 [38]. Other ALS/FTD-associated proteins may be related to MLO regulation, including destruction through autophagy (e.g. VCP, SQSTM1/p62) or posttranslational modification (PTM) and regulation of granule proteins (e.g. TBK1) [58].

Alzheimer’s disease (AD):

AD is the most common form of dementia. While amyloid plaques and tau tangles are the most recognizable aggregates in (most) AD patients, inclusions of the RNA-binding, granule-associated protein TDP-43 is often present and correlated with cognitive impairment [59]. Hence, RNA-granule disruption may play a role in AD. LLPS of phosphorylated tau may be associated with tau function [29] and phosphorylation increases LLPS of the tau microtubule-binding region [60]. AD-associated recombinant mutant tau forms solid aggregates within liquid phases in vitro [29], co-localizes with stress granules, associates with the ALS-associated protein TIA1, and increases the number and size of stress granules in cultured neurons [61]. Additionally, internalized extracellular phosphorylated tau associates with stress granules and alters their fluidity [62]. Therefore, alteration in tau LLPS may play a role in AD. Due perhaps to its extracellular localization, there have not yet been reports of amyloid-β, the other main driver of AD pathology, directly associating with membraneless organelles, though soluble amyloid-P(1–42) can induce stress granule formation in microglia [63].

Multiple sclerosis (MS):

MS is an autoimmune disease where the immune system attacks and destroys the myelin sheath. While MS is characterized by inflammation and neurodegeneration, the molecular mechanism is unclear. Exposing SK-N-SH cells (neuronal mimics) to the proinflammatory cytokine IFNγ leads to the formation and persistence of hnRNPA1-containing stress granules even after removing the stressor [64]. Additionally, patient brains showed hnRNPA1 and TIA1 redistribution from the nucleus to the cytoplasm where they accumulate in aggregates or persistent, large stress granules [64]. Hence, distinct MLOs may play functional and pathological roles in a single disease, and efforts to therapeutically modulate LLPS may need to be highly targeted to specific MLOs and their components.

Huntington’s Disease (HD):

HD is an inherited neurodegenerative disease caused by a poly-glutamine (polyQ) expansion in the huntingtin (HTT) protein. While the genetic cause of HD has been known since the early 1990s, it remains unclear why neurons degenerate, although both loss-of-function and gain-of-toxic-function of HTT have been implicated. Recently, multiple reports have shown that N-terminal fragments of HTT containing polyQ expansion can initially undergo LLPS before converting to aggregates [65, 66]. Additionally, some liquid phases of polyQ-containing HTT partially co-localize with stress granule markers [65]. Interestingly, profilin, product of the PFN1 ALS-associated gene, preferentially binds to the monomeric phase of poly-Q HTT and reduces the extent of LLPS and fibrillar aggregate formation in vitro [66], introducing profilin as a possible cross-disease therapeutic target.

Parkinson’s Disease (PD):

PD is a neurodegenerative disorder characterized by movement impairments, loss of dopamine neurons and other symptoms. While mutations in a few genes have been associated with PD, PARK7, which encodes DJ-1, localizes to both stress granules and P-bodies after oxidative stress and immunoprecipitates with known stress granule proteins [67]. Additionally, overexpression of parkin is able to rescue motor defects of a FUS-ALS model without altering amount or localization of FUS [68]. A recent report showed that alpha-synuclein is able to undergo LLPS and a liquid to solid transition in vitro and in cells [69]. Interestingly, parkin (PRKN) and PTEN-induced kinase 1 (PINK1) genetically interact with VCP (Transitional endoplasmic reticulum ATPase, an ALS/FTD gene) to degrade mitochondria [70] and VCP is also required for stress granule autophagy [71], tenuously linking quality control of these compartments through PD genes.

Spinocerebellar ataxia (SCA):

SCA is a group of neurodegenerative movement disorders. Some types of SCA are associated with the ataxin protein family, some of which are also found in stress granules [8, 12]. Repeat expansions in C90RF72, which result in aberrant RNA and protein products and cause ALS [45, 46, 51, 52], are also associated with SCA [72]. SCA type 2 is caused by polyQ expansion in ataxin-2 with 32 or more repeats while 29–32 repeats is a risk factor for ALS [73]. One group found an RNA-dependent interaction between polyQ-expanded ataxin-2 and staufenl (STAU1), an RNA-binding protein found in both transport granules and stress granules [74]. Interestingly, polyQ-expanded ataxin-2 increased protein levels of STAU1 and co-localized with STAU1 in stress granules and aggregates [74].

Spinal muscular atrophy (SMA):

SMA is a neuromuscular disorder, often with onset in early infancy or childhood, causing progressive muscle weakness associated with loss-of-function of the survival of motor neuron (SMN) protein. SMN, together with Gemin proteins, forms large nuclear particles called Gems, and helps form small nuclear ribonucleoproteins (snRNPs). SMN is transported in cytoplasmic granules to dendrites and axons, including the growth cone and synaptic terminals [75, 76], though the function, protein components (beyond Gemin2 [75] and hnRNPR [76]), and RNA components remain unclear. SMN was also found in a stress granule proteome [8] and co-localizes with TIA-1 in stress granules when overexpressed, potentially facilitating stress granule formation [77]. Knockdown of SMN reduces stress granule number and decreases cell viability after exposure to exogenous stress [78]. Interestingly, SMN is reported to interact both physically and functionally with LLPS-prone proteins including FUS [79]. Therefore, it is possible that SMN partitions into liquid phases and that Gems themselves are an MLO.

Fragile X syndrome:

Whereas most of our coverage of diseases linked to MLOs focused on neurodegenerative disorders, mechanisms related to some of those discussed earlier could be of relevance to fragile X syndrome, and we discuss these briefly next. Fragile X syndrome is a neurodevelopmental disorder associated with intellectual disability and is primarily caused by a repeat expansion in the 5’ untranslated region of FMR1, which encodes the FMRP protein. Somewhat similarly to expansions in ATXN2, which can cause ALS or SCA depending on repeat number, intermediate numbers of repeats in FMR1 can cause fragile-X associated tremor/ataxia (FXTAS). FMRP is a well-studied mRNA transport granule component [33] and has been found in stress granules [8, 12]. Recently, the C-terminal disordered region of FMRP was shown to undergo LLPS in vitro [22, 80]. As the repeat expansion in FRM1 leads to depletion of FMRP, FMRP may not aggregate from the phase-separated state, unlike many of the other proteins discussed earlier. Instead, fragile X syndrome is a loss-of-function disorder, one element of which may be improper localization and translation of the mRNAs carried by FMRP-containing transport granules.

Disease summary:

While some relationships are stronger than others, the number of neurodegenerative diseases that are linked to MLOs, and in particular stress granules, is notable, especially given the diversity of proteins/genes involved and the distinct symptoms patients display. Given that knock out and knock down of some stress granule proteins modify cellular toxicity across several ALS models [81], altering stress granule protein levels may be a potential therapeutic avenue to explore.

Modulating LLPS: physiological modification and therapeutic opportunity

Physiological mechanisms controlling the location and timing of MLO formation/dissolution are beginning to emerge [10, 82]. One mechanism that cells may use is interaction with binding partner proteins. For example, PY-NLS binding protein karoypherin beta 2 disrupts FUS aggregation [83–85]. Other biopolymers such as DNA and RNA can also modulate LLPS and regulation of poly(ADP-ribose) has shown promise as a drug-targetable LLPS modification [86]. An additional means of regulation appears to be post-translational modification of MLO components, including phosphorylation, methylation, acetylation, and others (see below). As so many neurodegenerative disease-associated proteins are found in MLOs and/or are capable of LLPS in vitro, modulating phase transition behavior via PTMs may be an attractive cross-disease putative therapeutic target to reduce or prevent protein aggregation and inclusion formation by altering the biophysical interactions of proteins. While many PTMs are physiologically found in IDRs and RNA-binding regions of granule proteins, only some have been studied for their effect on LLPS or granule formation (Table 2). Below we discuss the current evidence for PTMs altering LLPS or aggregation, and how these PTM may be targeted by therapeutics.

Table 2:

Summary of post-translational modifications, their structures, the effect they have on proteins, and the evidence for PTMs altering LLPS.

Phosphorylation:

Protein phosphorylation refers to the covalent attachment of a phophoryl group, most commonly to a serine, threonine, or tyrosine hydroxyl group. Each phosphate group adds on average nearly two negative charges to the protein at physiological pH [87, 88]. This is a large amount of localized charge, particularly for LLPS-prone proteins, some of which have specific charge patterning or depletion of charged residues in LLPS-prone regions. This increased negative charge may be exploited to alter phase separation of granule proteins. Indeed, serine/threonine phosphorylation of FUS reduces phase separation and prevents aggregation [25, 89]. Although FUS is often studied as the “poster child” for phase separation, its lack of charge in the LLPS-prone LC domain and high positive charge in the RGG domains make it unusual among RNA-binding proteins, many of which have less segregated amino acid patterning. However, serine/threonine phosphorylation is also important for granule formation or LLPS for other proteins, including tau [29, 60] and the C. elegans P-granule MEG proteins [82]. Interestingly, although the C-terminus of TDP-43 is hyperphosphorylated in patient inclusions, a single substitution mimicking serine phosphorylation in the structured N-terminal domain disrupts linear homopolymerization, dramatically decreasing LLPS in vitro, fluidizing TDP-43 reporter droplets in cells, and disrupting TDP-43 splicing function [90]. While most work on phosphorylation and LLPS has focused on serine/threonine phosphorylation, tyrosine phosphorylation has been shown to release mRNA for translation from hnRNPA2-containing transport granules [91]. These results demonstrating modulation of LLPS and aggregation via phosphorylation are intriguing and promising steps toward therapeutics, although more work is needed to identify the active kinases, determine when and in what pattern proteins are phosphorylated, and their effects on LLPS, granule formation/function, and conversion to toxic species.

Methylation:

Methyl (CH₃) groups can be added to both arginine and lysine. Unlike phosphorylation, methylation does not change the charge of the residue, but increases the amino acid size, hydrophobicity, and alters charge distribution, hence altering biomolecular interactions. Multiple methyl groups can be added to a single residue: up to three methyl groups can be added to lysine, while arginine can be monomethylated, asymmetrically or symmetrically dimethylated (see Table 2). Asymmetric arginine dimethylation reduces LLPS of granule proteins including Ddx4, hnRNPA2, and FUS [17, 20, 85, 92]. Interestingly, increased expression of protein arginine methyl transferase 1 (PRMT1, the main human enzyme catalyzing asymmetric dimethylation of RGG motifs) and asymmetric arginine demethylation was found in ALS patients, and the ratio of asymmetrically methylated arginine to unmethylated arginine is a predictor of disease prognosis [93]. G3BP1 arginine methylation has been shown to reduce stress granule formation and exposure to stress decreases its methylation [94]. Notably, lysine methylation has not yet been linked to LLPS or granule formation, although few lysine methylation sites have been identified in LLPS-associated proteins thus far [95], perhaps due to difficulty in proteomic detection of these proteins [89]. The effect of arginine methylation on LLPS and its correlation with ALS disease severity indicates methylation may be a promising therapeutic target.

Acetylation:

Acetyl groups can be added at lysine residues and protein N-termini. Lysine hyperacetylation of tau reduces phase separation in vitro [96]. Interestingly, lysine acetylation of TDP-43 in its RNA-binding domain increases inclusion formation in cells [97]. TDP-43 acetylation was found in a patient with ALS but not in a patient with FTD [97] where increased aggregation in cells and patient tissue implies that membraneless organelle formation/dissociation or dynamics may also be altered. A recent report found that lysine acetylation of DDX3X impairs LLPS in vitro and deacetylation is required for stress granule assembly in cells [98]. Although most proteins are N-terminally acetylated [99], the specific effect of this modification on LLPS or granule formation has not yet been studied.

Poly(ADP-ribose):

Poly(ADP-ribose) (PAR) is a PTM where linear or branched chains of ADP-ribose molecules are added to proteins. Initial work on the importance of PAR in LLPS was performed on FUS; both PAR and FUS are found at sites of DNA damage [100] and PAR seeds LLPS of FUS through interaction with FUS RGG domains though FUS itself is not known to be directly PARylated [44, 101]. Recently, PAR polymerases (PARPs) were identified as suppressors of TDP-43-associated neurodegeneration in a genomic screen in Drosophila [102]. Interestingly, PAR directly binds TDP-43 (which is not known to be PARylated), increases phase separation of TDP-43 [102], and reduces its aggregation in vitro [103]. PAR levels are increased in the nuclei of ALS patients, but a small molecule inhibitor of PARPs showed promise in reducing the toxicity of TDP-43 in cultured neurons [86]. It is important to note that the identity of the relevant PARylated protein remains unclear in these studies. One study found that hnRNPA1 is PARylated at K298, which alters LLPS of hnRNPA1 and its co-phase separation with TDP-43 [104]. Identifying PARylated protein(s) in MLOs is an important next step towards fully understanding the effect of PAR on MLOs and the potential for PAR as an MLO therapeutic.

The large effect PTMs and biopolymer modulators can have on phase separation and aggregation of these disease-associated proteins makes them attractive potential therapeutic targets. However, directly targeting the enzymes that generate these PTMs may be problematic because they have many other physiological targets. Simple knockdown or overexpression of these enzymes may have adverse effects on many other pathways and proteins. More targeted approaches, including designing enzymes that have few off-target effects, or carefully modulating the interplay between addition and removal of specific PTMs, may be more successful.

Concluding remarks

Although the field is young, MLOs appear to be particularly important for both normal function of neurons and disease pathogenesis (see Outstanding Questions). Additional links between MLOs and diseases of the nervous system may continue to be found. Furthermore, neurons have specialized MLOs and more may be identified as advances in microscopy are applied to neurons and glia. Finally, MLOs in the nervous system that are important in the development or progression of neurodegenerative diseases may be therapeutic targets through PTM of MLO proteins.

Outstanding questions box.

Distinct MLOs can have overlapping protein and RNA components. How does the cell distinguish between different types of granules? For example, are there different protein and/or RNA components in transport granules to distinguish the composition, function, and localization of the granule?

Previous work on the biophysical interactions that stabilize granules has focused on one or a few components. As granules seem to have many components, what are the biophysical interactions holding granules together, and can they be targeted by therapeutics? How do the interactions specify the liquidity/solidity of the MLO? How does granule dissociation or longevity affect the liquidity/solidity of a granule?

Many neurodegenerative-disease associated proteins are found in MLOs or able to undergo LLPS. Are there additional ones, that have not been identified so far? Are membraneless organelles a common link between neurodegenerative diseases? Could it be the case that some MLO proteins are so essential for development, and their function intolerant to mutation, that no human disease variants are expected to be found due to lack of organismal viability?

Neurodegenerative diseases are often characterized by selective loss or degeneration of one or a few types of neurons. Are some types of neurons more prone to stress granule formation and therefore to stress granule-related disease pathogenesis? Is there a relationship between axon length and role of transport granules in specific disease pathogenesis? What is the role of non-neuronal CNS cells (glia, oligodendrocytes) and their MLOs in disease?

Emerging evidence suggests that PTMs alter LLPS of MLO proteins, their localization, or their granule recruitment. When are these PTMs present in vivo and are different kinds of PTMs present at the same time? In what patterns are PTMs present (i.e. which sites are simultaneously modified)? For MLO proteins that shuttle between the nucleus and cytoplasm, is there a difference in PTM state between these locations as some proteins are only recruited to MLO in one location?

Highlights.

Neurons have a number of unique membraneless compartments formed by phase separation, including the postsynaptic density and elements of the pre-synapse.

Many neurodegenerative-disease associated proteins are found in membraneless organelles and/or are capable of phase separation; a number of hypotheses link membraneless compartments to toxic aggregates observed in disease.

Posttranslational modifications, including phosphorylation, methylation, and PARylation, are potential therapeutic targets for modulating phase separation and aggregation. Further research is needed, however, to fully understand the effects of posttranslational modifications on membraneless organelles and to develop targeted therapeutics.

Glossary

Liquid-liquid phase separation (LLPS)

a physical phenomenon where, in the case of cells, proteins, RNAs, and other biomolecules de-mix from the surrounding mixture and form a distinct liquid phase within the liquid of the nucleus/cytoplasm (or surrounding solution); evidence exists for LLPS both biochemically and in cells

Membraneless organelle (MLO):

distinct puncta observed in cells that are not surrounded by a phospholipid membrane; their formation is thought to be mediated by phase separation from the surrounding liquid

π-π interaction

refers to the non-covalent interaction between π (i.e. sp² hybridized) molecular orbitals of chemical groups, including those found in proteins. This interaction may include the stacking of two aromatic amino acids and contacts formed by the peptide bond region of amino acids. This interaction contributes to phase separation of many disordered proteins, especially those enriched in π-orbital containing residues

Posttranslational modification (PTM)

a chemical signal covalently added to a protein by specific enzymes that catalyze the reaction; distinct enzymes exist to remove the chemical modification

RNA/RNP granule

a punctate structure observed in cells. While their existence has been known for decades, more recently they have been recognized to be membraneless organelles. Most RNA/RNP granules contain both RNA and protein (RNP=ribonucleoprotein)