RNA modulates physiological and neuropathological protein phase transitions.

Abstract

Aggregation of RNA-binding proteins (RBPs) is a pathological hallmark of neurodegenerative disorders like amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD). In these diseases, TDP-43 and FUS RBPs are both depleted from the nuclear compartment, where they are normally localized, and found within cytoplasmic inclusions in degenerating regions of patient postmortem tissue. The mechanisms responsible for the aggregation of these proteins has remained elusive but recent studies suggest that liquid-liquid phase separation (LLPS) might serve as a critical nucleation step in the formation of pathological inclusions. The process of phase separation also underlies the formation and maintenance of several functional membraneless organelles (MLOs) throughout the cell, some of which contain TDP-43, FUS, and other disease-linked RBPs. One common ligand of disease-linked RBPs, RNA, is a major component of MLOs containing RBPs and has been demonstrated to be a strong modulator of RBP phase transitions. While early evidence suggested a largely synergistic effect of RNA on RBP phase separation and MLO assembly, recent work indicates that RNA can also antagonize RBP phase behavior in certain physiological and pathological conditions. In this review, we describe the mechanisms underlying RNA-mediated phase transitions of RBPs and examine the molecular properties of these interactions, such as RNA length, sequence, and secondary structure, that mediate physiological or pathological LLPS.

RNA binding protein aggregation in neurodegenerative disease

The intracellular deposition of misfolded proteins is a well characterized pathological feature of a number of neurodegenerative disorders, such as Alzheimer’s Disease (AD), Parkinson’s Disease (PD), Huntington’s Disease (HD), chronic traumatic encephalopathy (CTE), amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD) (Kumar et al., 2016; McKee et al., 2015; Ross and Poirier, 2004). While typically associated with divergent clinical presentations, pathway perturbations and genetic/environmental factors, the unifying observation of abnormal protein inclusions in postmortem tissue of these patients may suggest some common downstream mechanisms driving neurodegeneration in these diseases.

Though the specific protein components of neuropathological inclusions vary, both across disorders and between subtypes of disease, spanning a wide range of biological functions and subcellular distributions, one class of proteins that is increasingly overrepresented are RNA-binding proteins (Maziuk et al., 2017). For example, aggregation of the DNA/RNA binding protein TDP-43 is estimated to occur in up to ~97% of ALS, ~60% of AD, ~50% of FTD, and ~85% of CTE patients (McKee et al., 2010; Prasad et al., 2019; Tremblay et al., 2011). Aberrant aggregation of different RNA-binding proteins (RBPs) is also documented in neurodegenerative disorders, including: FUS, TAF15, EWSR1, hnRNPA1, hnRNPA2B1, TIA-1, hnRNPA3 and hnRNPD/AUF1 (Table 1). Furthermore, other well-recognized disease-linked proteins like tau and α-synuclein not typically considered “traditional” RNA binding proteins were recently predicted and/or demonstrated to interact with RNA and other nucleic acids (Schaser et al., 2019; Zanzoni et al., 2013; Zhang et al., 2017), suggesting that alterations in protein:RNA interactions or global RNA homeostasis may participate in the pathological deposition of protein inclusions across many different disorders.

Table 1:

Disease-linked RBPs with PrLDs that undergo LLPS.

RNA-binding protein (RBP)	Genetic disease-association	Pathology disease-association	Identified PrLD (1,2)	In vitro LLPS?	Example MLO-association
TDP-43	ALS (3), FTD (4)	ALS (5,6), FTLD (5,6), LATE (7), CTE (8), AD (9,10), CBD (10), PD (11)	a.a. 277–414	Full-length protein (12–14) C-terminal PrLD (15–18)	mRNP transport granules (19,20), Nuclear gems (21,22), Nuclear stress bodies (23), Stress granules (24,25), Paraspeckles (26)
FUS	ALS (27,28), FTD (29,30)	ALS (27,28), FTLD-FUS (31), HD (32), SCA1/2/3 (33)	a.a. 1–237	Full-length protein (34–36) N-terminal PrLD/LCD (34,36,37)	Paraspeckles (38), Nuclear gems (22), Stress granules (39–41), mRNP transport granules (42,43)
EWSR1	ALS (44)	FTLD-FUS (45)	a.a. 1–280	Full-length protein (46,47)	DNA damage response foci (48), Stress granules (49), Paraspeckles (26)
TAF15	ALS (50,51)	FTLD-FUS (45)	a.a. 1–152	Full-length protein (46,47)	DNA damage response foci (48), Stress granules (49,52), Paraspeckles (26)
hnRNPA1	MSP (53), ALS (53)	MSP* (53), IBM* (53), FTLD-FUS (54)	a.a. 186–372	Full-length protein (12,47,55) C-terminal PrLD/LCD (12,55)	Stress granules (12,56), mRNP transport granules (42), Paraspeckles (26)
hnRNPA2B1	MSP (53), ALS (53)	MSP* (53), IBM* (53)	a.a. 197–353	Full-length protein (46,57,58) C-terminal PrLD/LCD (58,59)	Stress granules (53,60), mRNP transport granules (42)
TIA-1	ALS (61,62), WDM* (63,64)	AD (65)	a.a. 292–386	Full-length protein (46,61,66) C-terminal PrLD/IDR (55)	Stress granules (67)
hnRNPA3	Unknown	C9orf72-ALS (68,69)	a.a. 207–378	Full-length protein (46)	mRNP transport granules (70), Stress granules (71)
hnRNPD/AUF1	Unknown	FTLD-FUS (54)	a.a. 262–365	Full-length protein (46)	mRNP transport granules (42)

^*Muscular diseases/pathology

Abbreviations: AD = Alzheimer’s disease; ALS = amyotrophic lateral sclerosis; CBD = corticobasal degeneration; CTE = chronic traumatic encephalopathy; FTD = frontotemporal dementia; FTLD = frontotemporal lobar degeneration; HD = Huntington’s disease; IBM = Inclusion body myopathy; IDR = Intrinsically-disordered region; LATE = Limbic-predominant age-related TDP-43 encephalopathy; LCD = Low-complexity domain; LLPS = Liquid-liquid phase separation; MLO = Membraneless organelle; PD = Parkinson’s disease; MSP = Multisystem proteinopathy; PrLD = Prion-like domain; SCA1/2/3 = Spinocerebellar ataxia types 1, 2 & 3; VCPDM = vocal cord and pharyngeal weakness with distal myopathy; WDM = Welander distal myopathy

While these observations from postmortem patient tissue suggest some participation of RNA-binding protein aggregation in the pathogenesis of neurodegenerative disease, a critical question remains: how do RBP inclusions arise and what upstream mechanisms initiate this event? One major hypothesis that has recently emerged from the fields of polymer chemistry and biophysics revolves around the concept of phase separation. In addition to interactions with nucleic acids, another commonality across many RNA-binding proteins and proteins misfolded in neurodegenerative diseases, is their ability to undergo a particular type of phase transition, called liquid-liquid phase separation (LLPS) and this is proposed to serve as a critical nucleation event for the deposition of pathological protein aggregates (Babinchak and Surewicz, 2020).

Liquid-liquid phase separation in biological systems

In biological terms, phase separation refers to a process by which certain organic molecules dynamically transition from an initially mixed population into separate compartments, or phases. In the same manner that non-biological molecules can transition between different states (i.e. gas, liquid, and solid) based upon variables like temperature, it is becoming increasingly recognized that biological molecules, such as proteins, are found in different biophysical states within a cell (i.e. soluble proteins or solid-like aggregates) based upon changes in the intra- and extracellular environment such as heat (Wallace et al., 2015), nutrient starvation (Narayanaswamy et al., 2009) or proteostatic stress (Sui et al., 2020).

Phase transitions are also utilized to dynamically compartmentalize and organize the cell in the absence of lipid membranes. Complementing more canonical lipid-bound organelles like mitochondria and the endoplasmic reticulum, these “membraneless organelles” (MLOs) can be found all over the cell and include such structures as the nucleolus, RNA transport granules, processing bodies (P-bodies), and paraspeckles (reviewed by Gomes and Shorter, 2019). Following pioneering biophysical investigations into the C. elegans germline P granule (Brangwynne et al., 2009) and the X. laevis oocyte nucleoli (Brangwynne et al., 2011), it is now established that many MLOs arise through the process of liquid-liquid phase separation (LLPS). Commonly presented as analogous to the behavior of oil in water, LLPS involves the selective condensation of molecules (i.e. proteins and nucleic acids) into distinct droplets, or a condensed phase, within a surrounding liquid environment. These condensed droplets exhibit many properties of “classical” liquids, for example: spherical shape, fission/fusion, dripping, surface wetting, and dynamic molecular exchange both within droplets and with the surrounding environment (reviewed by Hyman et al., 2014; Shin and Brangwynne, 2017).

The resulting biophysical characteristics of these condensed assemblies provide a unique advantage over membrane-bound organelles. For example, recent work investigating LLPS at the neuronal synapse suggests that condensed phases function to rapidly concentrate and release specific pre- and post-synaptic components in response to neuronal activity without the need for active transporters typically utilized by membrane-bound compartments (Milovanovic et al., 2018; Zeng et al., 2016). Liquid-like assemblies at the plasma membrane of non-neuronal cells also play a role in T-cell receptor signaling, where downstream signaling proteins condense and organize specific adapter/effector proteins, such as ZAP70 and Nck, while excluding phosphatases that function to dissolve condensates via protein dephosphorylation (Su et al., 2016). These localized condensates promote actin filament assembly at rates that exceeded those resulting from simple membrane targeting of the same molecules (Su et al., 2016). Additional examples of functional biomolecular condensates are being increasingly discovered and associated with novel cellular pathways, thus highlighting that the ability of proteins and other molecules to undergo phase transitions is a fundamental principle of cellular physiology.

Intrinsically-disordered regions (IDRs) as key players in the formation and material properties of phase-separated protein assemblies

As the number and diversity of proteins reported to utilize phase separation across the proteome has grown, so too has our understanding of the molecular determinants governing the formation and biophysical properties of different membraneless assemblies. While the precise domain composition and molecular interactions vary, recent efforts in the purification and proteomic analysis of various MLOs, such as the nucleolus (Andersen et al., 2002, 2005; Scherl et al., 2002), stress granules (Jain et al., 2016; Kuechler et al., 2020) P-bodies (Hubstenberger et al., 2017), and other RNA granules (Han et al., 2012; Kato et al., 2012), suggest that a common feature allowing proteins to undergo phase separation is multivalency. This multivalency can be achieved in a number of ways, but generally involves weak, transient contacts, often through repeated interaction motifs or modules, within and between molecules (reviewed by Posey et al., 2018; Shin and Brangwynne, 2017). Initial investigations utilizing synthetic linear peptides comprised of repeated interaction modules, such as the SH3/PRM (Li et al., 2012) and SUMO/SIM (Banani et al., 2016) domains, in heterotypic reactions have highlighted the importance of valency (i.e. the number of interaction modules) in the ability of proteins to undergo phase separation, but later studies also point to a distinct role for the interspersing of these motifs by flexible linker sequences as a critical determinant of phase behavior in these systems (Harmon et al., 2017). Interestingly, another key feature of proteins found within MLOs by the studies mentioned above, particularly in stress and other RNA granules, was an enrichment of intrinsically disordered domains or intrinsically-disordered regions (IDRs) (Han et al., 2012; Jain et al., 2016; Kato et al., 2012; Kuechler et al., 2020).

IDRs are typically defined as regions of a protein sequence that lack a defined, or ordered, secondary structure and thus are capable of adopting a range of three-dimensional conformations (Dunker et al., 2013). The structural ambiguity of intrinsically-disordered regions are conferred by amino acid compositions of limited sequence complexity, and regions conforming to these characteristics are commonly referred to as low-complexity domains (LCDs) or low-complexity IDRs (van der Lee et al., 2014). IDRs are also typically enriched in charged (i.e. arginine, lysine, glutamic acid), polar (i.e. serine, glutamine, asparagine), and/or aromatic (i.e. phenylalanine, tyrosine, tryptophan) amino acids and contain amino acids like glycine and proline that may convey some structural information (Romero et al., 2001; Vucetic et al., 2003). Based on the specific composition and clustering of certain subsets of these enriched amino acids, IDRs can be further classified into distinct subtypes, such as RG/RGG domains (Elbaum-Garfinkle et al., 2015; Wang et al., 2018b), FG domains (Hülsmann et al., 2012; Schmidt and Görlich, 2015) and prion-like domains (PrLDs, discussed below), that engage in a variety of non-covalent interactions responsible for driving phase transitions. While interaction types can vary both across and within IDRs, they generally involve one or more of the following: i) charge-based interactions (charge-charge or cation-π), ii) dipole-dipole interactions, and iii) π-π stacking of aromatic residues.

IDR amino acid composition dictates biomolecular condensate material properties and function

Recent simulations and mutational studies predict that the heterogeneity of both structural conformation and interaction type (via amino acid composition) in IDRs have profound effects on the formation and biophysical properties of higher-order assemblies (Tüű-Szabó et al., 2018; Wang et al., 2018b). The material properties, in turn, reflect specific functions that are encoded within the sequence of IDRs and other interaction domains within protein components of various membraneless structures. One example of this notion is found within the central channel of the nuclear pore complex, where condensation of intrinsically-disordered FG-nucleoporins form a selectivity barrier of the nuclear pore complex through multivalent interactions between repetitive aromatic patches (i.e. FxFG, GLFG, xxFG) contained within FG domains of these proteins (Li et al., 2016) (Figure 1a). In vitro, the FG domains of several nucleoporins, such as Nup62 (Labokha et al., 2013), Nup98 (Hülsmann et al., 2012), and Nup153 (Milles and Lemke, 2011), undergo hydrogel formation. Despite divergence of exact amino acid sequence, overall amino acid composition and structural disorder is a well-conserved feature of FG domains spanning from yeast to human nucleoporins (Denning and Rexach, 2007). Yeast FG-nucleoporins also form hydrogels when sufficiently saturated in vitro and function as selective permeability barriers in these systems, allowing the rapid translocation of importin-β-bound cargo via successive high-affinity interactions with FG-nucleoporins while blocking the diffusion of non-specific molecules (Celetti et al., 2020; Frey and Görlich, 2007, 2009). Importantly, deletion of FG domains, or mutation of phenylalanine residues to allow liquid-liquid phase separation but prevent hydrogelation impairs the barrier function of nucleoporins like Nup98 (Hülsmann et al., 2012) and Nup100 (Patel et al., 2007), highlighting the functional benefit of hydrogel formation driven by conserved FG-repeat interactions.

Figure 1. Examples of biophysical state dictating structure function in vivo.

(a) Within the central channel of the nuclear pore complex (NPC), hydrophobic interactions between FG-repeat regions of the FG-nucleoporins (i.e. Nup98, Nup62) contribute to the formation of a hydrogel-like assembly. This gel-like state forms a permeability barrier to block the translocation of nonspecific cargo in and out of the nucleus, while allowing for specific interactions between nuclear transport receptors and nucleoporins that drive cargo tranport. (b) The nucleolar protein NPM1 utilizes electrostatic interactions provided by acidic and basic tracts within its central IDR to undergo LLPS as part of the the granular component (GC) of the nucleolus. The liquid-like state of the GC allows for rapid exchange of newly-transcribed and processed rRNA, from the dense fibrillar component (DFC) and the fibrillar core (FC) subcompartments of the nucleolus, and ribosomal subunits, from the surrounding nucleoplasm, that is required for proper ribosomal assembly and export.

Another example is found in the phase behavior of the nucleolar protein, nucleophosmin 1 (NPM1). NPM1 utilizes charge-based interactions driven by acidic and basic tracts within its central IDR to achieve homotypic LLPS in vitro (Mitrea et al., 2018) and heterotypic LLPS with arginine-rich peptides and rRNA. This results in the formation of the liquid-like granular component (GC) of the nucleolus (Feric et al., 2016; Mitrea et al., 2016) (Figure 1b). In addition to compartmentalization of the nucleolar machinery within the surrounding nucleoplasm, the liquid-like state of NPM1 within the granular component is essential for rRNA processing, as solidification of the GC phase by optogenetic oligomerization (Zhu et al., 2019) or the infiltration of toxic arginine-rich dipeptide-repeat proteins (Lee et al., 2016) leads to the accumulation of rRNA precursors and reduction of mature rRNA. The liquid-like material properties of the GC also play an important role in the cellular stress response, facilitating the entry and dynamic storage of misfolded proteins to allow for Hsp70-mediated refolding and prevent irreversible aggregate formation prior to stress recovery (Frottin et al., 2019). Disruption of the liquid-like state of NPM1 through prolonged stress or accumulation of dipeptide-repeat proteins, severely compromises the protein quality control function of the GC phase. Thus, each distinct interaction type (i.e. aromatic or charge-based) encoded within these regions were likely evolutionarily tuned to allow for the formation of assemblies with specific, functionally-beneficial material properties.

Prion-like domains and their relevance to aberrant RNA-binding protein aggregation in neurodegenerative disease

Another subtype of IDR that has been identified to mediate phase separation of various proteins into assemblies with unique biophysical properties are prion-like domains (PrLDs). First identified in the yeast proteins Ure2 and Sup35, PrLDs were shown to drive the assembly of self-propagating, amyloid-like aggregates that were capable of templating their aggregated state to soluble versions of the proteins (Patino et al., 1996; Ter-Avanesyan et al., 1994; Wickner, 1994). Subsequent analysis of these domains revealed that PrLDs were intrinsically-disordered and contained very limited sequence complexity, skewing towards amino acid compositions of glycine and largely uncharged, polar amino acids like glutamine, asparagine, serine, and tyrosine (Verdile et al., 2019).

When the human proteome was later scanned for proteins containing PrLD-like amino acid compositions, a heavy enrichment of proteins identified as containing PrLDs was found in RNA-binding proteins (Alberti et al., 2009; Couthouis et al., 2011; King et al., 2012). Furthermore, a number of these PrLD-containing RBPs overlap with those reported to form pathological inclusions in postmortem patient tissue mentioned above (Table 1), suggesting a role for prion-like domains in the aberrant aggregation of proteins observed in disease (Couthouis et al., 2011). Purified preparations of these disease-linked RBPs also undergo LLPS in vitro and are components of RNA-containing MLOs within the cell, including RNA transport granules, paraspeckles, and nuclear gems (Table 1). The PrLDs of many RBPs are necessary and sufficient for phase separation in these systems (Babinchak et al., 2019; Conicella et al., 2016, 2020; Molliex et al., 2015; Patel et al., 2015; Ryan et al., 2018; Schmidt and Rohatgi, 2016), highlighting the importance of prion-like domains in regulating physiological self-assembly of disease-linked RNA-binding proteins under normal conditions.

While initial assemblies formed by PrLD-containing RBPs in vitro, as well as their PrLDs in isolation, often display liquid-like biophysical properties, extended incubation of certain RBPs (such as FUS, hnRNPA1, and the TDP-43 LCD) promote the maturation or hardening of droplets into more gel-like structures or solid-state aggregates that resemble the pathological state of these proteins observed in patient tissue (Babinchak et al., 2019; Conicella et al., 2016; Lin et al., 2015; Patel et al., 2015; Zhuo et al., 2020). Similar droplet-like states precede amyloid formation of Sup35 in yeast (Patel et al., 2015; Serio et al., 2000), suggesting that PrLDs harbor evolutionarily-conserved elements that allow for the formation of liquid-like compartments in response to physiological stimuli but also confer a propensity for solid-state aggregation under pathological conditions.

Supporting this potential duality, liquid-to-solid state transitions of certain RBPs, also called aberrant phase transitions, are promoted by disease-linked familial mutations within their PrLDs. An ALS-linked mutation in the FUS PrLD (G156E) was recently shown to gradually increase the viscosity of initially liquid-like FUS droplets in vitro and promote emergent fibrillization at incubation times in which wildtype droplets remained dynamic (Patel et al., 2015). Similar results were observed when examining the effect of a multisystem proteinopathy (MSP)-linked PrLD mutation on the in vitro phase behavior of hnRNPA1, which underwent LLPS indistinguishable from wildtype protein at initial timepoints and were readily partitioned into liquid-like droplets with wildtype protein when mixed (Molliex et al., 2015). However, over time, the mutant protein began to fibrillize and emerged from wildtype droplets, eventually seeding wildtype protein into similar fibril-like assemblies. These effects were similarly observed for other PrLD-containing RBPs like TDP-43 and TIA-1 in cellular systems, where disease-linked mutations within these regions corresponded with increased protein droplet viscosity and/or delayed dissolution of membraneless structures containing mutant proteins (Gopal et al., 2017; Mackenzie et al., 2017; Mann et al., 2019; Schmidt and Rohatgi, 2016). Though the maturation of initial protein assemblies has only been directly interrogated for a subset of disease-linked RBPs, it is hypothesized that LLPS may serve as a general initiation event that results in the deposition of these proteins within insoluble inclusions in neurodegenerative diseases.

RNA as a scaffold for promoting physiological phase transitions and MLO formation

In addition to prion-like IDRs, many protein components of MLOs also contain multiple RNA-binding domains, either in the form of canonical RNA-recognition motifs (RRMs) or RGG/ZnF regions (reviewed by Gomes and Shorter, 2019). RNA is major component of a number of MLOs within the cell, including paraspeckles (Bond and Fox, 2009; Clemson et al., 2009), processing bodies (Sheth and Parker, 2003; Teixeira et al., 2005), and neuronal transport granules (Krichevsky and Kosik, 2001). Taken together, this suggests that RBP:RNA interctions may regulate phase separation. Recent in vitro studies uncovered some of the basic mechanisms underlying the effects of RNA on RBP phase behavior. One of the first direct investigations of RNA’s effect on RBP phase separation reported a strong enhancement of LLPS observed of artificial fusion proteins containing RNA-binding elements from the polypyrimidine tract-binding protein (PTB) along with intrinsically disordered regions from various other RBPs (including FUS, hnRNPA1, TIA-1, and others) upon addition of RNA (Lin et al., 2015). The enhancement of PTB assembly by RNA was dependent upon the presence of IDRs, whereas phase separation of PTB alone with RNA only occurred at high protein concentrations (50 μM alone versus 1.25–2.5 μM with IDRs). The addition of RNA did not promote the LLPS of intrinsically disordered regions alone (without PTB RNA-binding regions), suggesting a synergistic effect of PTB:RNA and IDR:IDR interactions in the driving of phase separation. Similar findings were reported for full-length RBPs, including FUS (Burke et al., 2015; Han et al., 2012) and hnRNPA1 (Lin et al., 2015; Molliex et al., 2015), which led to the hypothesis that binding of RNA promotes a high local concentration of IDR-containing RBPs that contribute additional multivalent interactions required to drive phase separation of RNP structures.

The coaction of RNA:RBP and IDR:IDR interactions is particularly important for the assembly of multicomponent MLOs, such as the nucleolus, and was recently conceptually described in terms of client-scaffold relationships (Banani et al., 2016; Ditlev et al., 2018). In brief, the client-scaffold framework describes the formation and composition of phase-separated compartments being dictated by interactions between multivalent “scaffold molecules” (i.e. RNA) and “client molecules” (i.e. RBPs). The enhanced multivalency afforded by interactions between RNA-binding regions within RBPs and specific binding motifs encoded within RNA nucleotide sequence drive heterotypic phase separation (i.e. through lowering the concentration threshold required for a protein to undergo LLPS), and provide some specificity in the recruitment of particular RBPs to ribonucleoprotein (RNP) granules (reviewed by Langdon and Gladfelter, 2018). This concept is well-exemplified in the granular component (GC) subcompartment of the nucleolus by interaction between rRNA and NPM1, serving as scaffold and client respectively. In vitro, rRNA is capable of driving phase separation of NPM1 into co-assemblies that mirror the liquid-like state of the GC phase of the nucleolus in vivo (Feric et al., 2016; Mitrea et al., 2018). Co-assembly is mediated by specific interactions between the C-terminal RNA binding domain of NBM1 and rRNA. Poly-U50 RNA cannot induce NPM1 LLPS and deletion of the RNA binding domain disrupts NPM1 phase separation with rRNA and localization to the GC phase of nucleoli (Feric et al., 2016; Mitrea et al., 2016, 2018). Interestingly, NPM1 localization within the GC requires the N-terminal oligomerization domain (Feric et al., 2016; Mitrea et al., 2016), demonstrating the cooperative effects of homotypic and heterotypic protein:RNA multivalent interactions often underlying the assembly of multicomponent MLOs.

Other factors such as relative scaffold/client concentrations, client affinity/valency, and additional multivalent interactions between different scaffold and client molecules also play a critical role in determining the effect of RNA in promoting MLO assembly and RBP incorporation into these structures (reviewed by Boeynaems et al., 2018). Other work investigating the formation of other naturally-occurring MLOs within intracellular systems support the general framework of a client-scaffold relationship between RBPs and RNA, as the presence of certain RNAs within a cell is required for the assembly of many different cellular RNP compartments including P-bodies (Teixeira et al., 2005), bacterial RNP (BR) bodies (Al-Husini et al., 2018), paraspeckles (Clemson et al., 2009), nuclear amyloid bodies (Audas et al., 2016), and the nucleolus (Berry et al., 2015; Falahati et al., 2016). Furthermore, MLOs, such as nuclear speckles, paraspeckles, Cajal bodies, and nuclear stress bodies, can be formed de novo by the introduction of excess RNA substrates into the cell (Bounedjah et al., 2014; Kaiser et al., 2008; Shevtsov and Dundr, 2011). These observations suggest that RNA is not merely a passive component of many MLOs within the cell but is a necessary and active participant in their assembly and maintenance.

Stress granules as an intersection between physiological and pathological RBP phase transitions

One of the best-characterized multi-component MLOs in the cell are stress granules. First described in chicken embryo fibroblasts (Collier and Schlesinger, 1986; Collier et al., 1988), stress granules are cytoplasmic RNP granules that form in response to cellular stressors or pharmacological/genetic intervention and contain an abundance of RBPs, translation initiation factors, and mRNAs (reviewed by Buchan and Parker, 2009; Protter and Parker, 2016). These granules exhibit liquid-like properties in cells and are thus presumed to arise through the process of LLPS. Stress granules have been proposed to function in translational regulation and mRNA homeostasis, given the presence of stalled translation initiation complexes and the global inhibition of protein synthesis observed in conditions triggering their formation (Kedersha et al., 2000, 2002, 1999; Kimball et al., 2003; Mazroui et al., 2006; Mokas et al., 2009). Stress granules also sequester a number of other non-RNA binding proteins involved in various signaling pathways (Jain et al., 2016) and thus may participate in additional or alternative functions.

Recent in-depth investigations into the molecular mechanisms of stress granule formation show that RNA is a critical factor mediating their initial assembly and maturation in conditions of translational arrest. Specifically, the release of non-translating mRNAs from polysomes upon translational inhibition triggers phase separation of stress granule core proteins like G3BP1/2 upon binding to newly available RNA (Guillén-Boixet et al., 2020; Sanders et al., 2020; Yang et al., 2020). Initial core assemblies of G3BP1/2 and non-translating mRNA then serve as multivalent network nodes that provide a physical platform for the subsequent recruitment of client molecules required for stress granule condensation through crosslinking (Guillén-Boixet et al., 2020; Sanders et al., 2020; Yang et al., 2020). Early studies indicate that recruitment of additional client RBPs is indeed dependent upon their binding to RNA, as deletion or mutation of RNA binding domains of RBPs, such as TDP-43 and FUS, antagonizes their co-localization with stress granules (Bentmann et al., 2012; Chen and Cohen, 2019; Colombrita et al., 2009; Daigle et al., 2013; Guillén-Boixet et al., 2020; Mann et al., 2019), and RNaseI-mediated degradation of single-stranded RNA significantly reduces their enrichment within stress granule fractions upon the induction of cellular stress (Fang et al., 2019).

Stress granules are a major focus in the field of neurodegeneration following the discoveries that disease-linked RBPs like TDP-43, FUS, TIA-1, hnRNPA1 and hnRNPA2/B1 localize to this MLO (Bosco et al., 2010; Colombrita et al., 2009; Dormann et al., 2010; Gal et al., 2011; Guil et al., 2006; Kim et al., 2013; Liu-Yesucevitz et al., 2010; McDonald et al., 2011; Molliex et al., 2015) (see also Table 1). Many of the above-mentioned RBPs exert some regulatory capacity over stress granule formation and disease-linked mutations often enhance recruitment to stress granules (Baron et al., 2013; Bosco et al., 2010; Gilks et al., 2004; Kim et al., 2013; McDonald et al., 2011). Additionally, familial disease-linked mutations in some stress granule components, such as FUS and TIA-1, alter stress granule assembly and/or dynamics (Baron et al., 2013; Mackenzie et al., 2017).

These observations led to the emerging hypothesis linking stress granule dysfunction to the protein depositions in neurodegenerative diseases. Specifically, during the physiological stress response, proteins like TDP-43 and FUS shuttle out to the cytoplasm and are incorporated into stress granules. In normal conditions, stress granules are then disassembled and/or cleared by autophagy during stress recovery, allowing for translational restart of stalled mRNPs and degradation or return to normal function of other RBP components (reviewed by Protter and Parker, 2016). Conversely, during pathological conditions in which stress granule clearance or disassembly is delayed, these liquid-like structures could serve as reaction crucibles to promote the aggregation of proteins like TDP-43 and FUS in a similar mechanism to that observed in vitro (reviewed by Wolozin and Ivanov, 2019). In this sense, the prolonged incubation of aggregate-prone RBPs present at high local concentrations within stress granules could promote a maturation of these assemblies and result in their deposition into insoluble inclusions capable of driving disease. However, while one might expect other stress granule components to co-aggregate with TDP-43 or FUS in patient inclusions as a result of stress granule maturation, postmortem analyses of TDP-43 or FUS proteinopathy patient tissue has yielded largely negative results (Chen and Cohen, 2019; Hirsch-Reinshagen et al., 2017; Mackenzie et al., 2017; Mann et al., 2019). ALS patients harboring TIA-1 mutations shown to delay stress granule disassembly in cellular models, there are TIA-1-positive inclusions nor co-localization of TIA-1 with TDP-43 pathology was observed in postmortem tissue (Hirsch-Reinshagen et al., 2017; Mackenzie et al., 2017). Still, the possibility does exist that initial TDP-43 or FUS aggregate formation may occur within stress granules, which may persist following disassembly of other stress granule components. However, it is still unclear whether seeding of RBP aggregation within stress granules, based on observations in cellular models, recapitulates the pathogenic processes occurring in human disease and remains the focus of continued study.

The role of RNA in the maturation of stress granules also remains an important question in the field. Recent transcriptomic studies have begun to uncover some common elements of SG-enriched RNAs (discussed below) and potential mechanisms driving their partitioning into stress granules (Khong et al., 2017; Matheny et al., 2019; Van Treeck et al., 2018). Additionally, some shifts in the protein and RNA composition of stress granules induced by acute or chronic stress have been reported (Reineke et al., 2018), but future studies aimed at directly modulating the RNA content of these assemblies are needed to determine the contribution of RNA to the maintenance of SG fluidity and prevention of protein aggregation under chronic stress conditions.

RNA antagonizes the phase transitions of RBPs at physiological concentrations

In addition to influencing the formation and composition of membraneless assemblies, studies show that RNA can promote RBP phase separation and incorporation into MLOs both in vitro and in cellular environments. The effects of RNA on RBP LLPS are often recognized as heavily dependent upon RNA:protein molar ratios, with lower RNA concentrations promoting phase separation of excess protein concentrations in most in vitro experiments (Burke et al., 2015; Lin et al., 2015; Molliex et al., 2015; Schwartz et al., 2013; Zhang et al., 2015) (Figure 2a). However, in a cellular environment under physiological conditions, many RBPs reside in the nuclear compartment where these molar ratios are generally reversed (Maharana et al., 2018) (Figure 2b). FUS for example, which has been predicted to be endogenously expressed at ~7 μM within the nucleus, readily undergoes LLPS at the same concentration in vitro (Burke et al., 2015; Monahan et al., 2017), but remains largely soluble within the nuclear compartment in cells (Maharana et al., 2018). While addition of low concentrations of total RNA (~0.4:1 RNA:FUS ratio) enhances in vitro LLPS, multiple studies noted that increasing RNA concentrations to excess, thus mimicking the native intracellular environment, reverses the enhancement of, and prevents, FUS phase separation in vitro (Burke et al., 2015; Maharana et al., 2018; Schwartz et al., 2013). In fact, total RNA was shown to effectively solubilize FUS droplets in vitro at concentrations over 10-fold lower than that predicted to be present within the nucleus under normal conditions (Maharana et al., 2018). A similar biphasic effect of RNA on phase separation in vitro was demonstrated for other RBPs like Whi3 (Zhang et al., 2015), G3BP1 (Yang et al., 2020), and artificial arginine-rich peptides (RP₃/SR₈) (Banerjee et al., 2017), suggesting a complex relationship between RNA and RBP phase behavior at work within the cell (Figure 2b).

Figure 2. RNA concentration is an important determinant of RBP phase behavior in vitro and in vivo.

(a) In vitro LLPS of RBPs can be stimulated by the addition of low concentrations of RNA, wherein excess concentrations of RBPs are allowed to multimerize on single RNA molecules and undergo phase transitions. In contrast, in vitro phase separation of RBPs resulting from molecular crowding or high RBP concentrations can be buffered by the addition of RNA at high/excess concentrations. The increased availability of RNA binding partners can prevent RBP multimerization on single RNAs and thus may prevent/reverse LLPS. (b) The effects of these molar ratios of RBP:RNA can be observed in the cell, where high RNA:RBP ratios in the nucleus maintain RBP solubility and low RNA:RBP ratios in the cytoplasm can promote physiological/pathological phase transitions.

In support of an oppositional role for RNA in physiological contexts, mutation of the RNA-binding domains of RBPs like TDP-43, FUS, Matrin-3, and Whi3 to mitigate RNA binding efficiency promotes LLPS and/or aggregation when localized to their native compartments within the cell (Daigle et al., 2013; Flores et al., 2019; Malik et al., 2018; Mann et al., 2019; Yu et al., 2021; Zhang et al., 2015). Degradation of endogenous RNA substrates exerts a similar effect as artificial RNA-binding domain mutations, resulting in the rapid phase separation of wildtype RBPs, such as FUS, TDP-43, hnRNPA1, EWSR1, and TAF15, upon RNase microinjection in the nucleus (Maharana et al., 2018). Interestingly, phase separation of nuclear RBPs is not observed upon DNase microinjection, suggesting a specific role for RNA in the buffering of these proteins in the cell. In support of this notion, novel ALS/FTD-linked mutations adjacent to the RNA-recognition motifs (RRMs) of TDP-43 were recently identified (Chen et al., 2019). Two of these mutations (K181E/K263E) disrupt RNA binding and promote the intranuclear aggregation of TDP-43, while the D169G mutation neither affected RNA binding or intracellular aggregation of the protein. Additional studies indicate that nuclear aggregation of TDP-43 with compromised RNA-binding capacity are initiated through LLPS, as TDP-43 containing ALS/FTD-linked (P112H/K181E/K263E) or acetylation-mimicking mutations within RNA-recognition motifs was observed to form liquid-like anisosomal structures capable of conversion to more gel- or solid-like structures following Hsp70 inhibition or ATP depletion (Yu et al., 2021).

Recent proteomic analysis of cell and tissue lysates subjected to RNase treatment additionally demonstrated that this phenomenon is not restricted to a small subset of disease-linked RBPs, as over 1300 individual proteins become insoluble following cellular RNA degradation (Aarum et al., 2020). Other disease-associated proteins not typically considered RBPs, such as Tau and Huntingtin, also became aggregated upon RNase digestion, suggesting that RNA:protein interactions contribute to aberrant phase transitions and resulting protein depositions found in a variety of neurodegenerative disorders. Another key observation of this study was the over-representation of proteins associated with the Gene Ontology (GO) terms of “RNA binding” and “translation initiation” that became aggregated upon RNase treatment, which mirrors the major protein composition of stress granules determined by a similar proteomic analysis (Aarum et al., 2020; Markmiller et al., 2018). These results are especially interesting in light of recent work suggesting a protective role for stress granules in the maintenance of TDP-43 and FUS solubility during periods of cellular stress (Gasset-Rosa et al., 2019; Hans et al., 2020; Mann et al., 2019; McGurk et al., 2018; Shelkovnikova et al., 2013). Here, acute RNA-dependent sequestration of proteins like TDP-43, FUS, and others within RNA-rich stress granules may actually prevent their aggregation during environmental conditions that trigger protein misfolding or damage in a similar manner to that proposed for the yeast prion protein Sup35 during pH stress (Franzmann et al., 2018). Osmotic stress conditions that are typically used to initiate stress granule assembly directly trigger TDP-43 insolubility and ubiquitination in the presence of pharmacological inhibitors of stress granule formation (Hans et al., 2020), further supporting a possible stress granule-independent pathway to RBP aggregation during periods of cellular stress. Together, these results suggest that intracellular RNA is an endogenous buffering system evolved to maintain the solubility of RBPs under physiological conditions. RNA likely acts in concert with other molecular chaperones previously shown to prevent aberrant RBP aggregation, including various heat-shock proteins (HSPs) and nuclear import receptors (Carlomagno et al., 2014; Chen et al., 2016; Guo et al., 2018). Breakdown of any of these systems in pathological states, such as the disrupted interaction between ALS-linked mutant FUS and its nuclear import receptor Kapβ2 (Hofweber et al., 2018), may promote the deposition of insoluble inclusions of these proteins observed in disease.

RNA properties differentially regulate RBP phase behavior

The intracellular and in vitro findings described above have shed new light on the complex regulatory role of RNA on the phase behavior of various RBPs. However, considering the variety of RNA species present within the intracellular environment and total RNA preparations used for many in vitro assays, it is challenging to identify RNA species responsible for promoting or inhibiting RBP phase separation in current model systems or the mechanisms by which they may be exerting their effects. Recent investigations into RBP phase behavior with individual RNA species revealed that the properties of RNA molecules exert differential effects on the formation, material properties, and composition of RNP assemblies.

Long RNAs as scaffolds for RBP phase transitions and MLO assembly

Though direct comparisons are limited, RNA length is one variable that mediates phase behavior of RBPs. Longer RNAs exhibit an enhanced ability to drive in vitro protein phase separation, as has been observed with the RBP, PGL-3 (Saha et al., 2016). Recent smFRET studies of LAF-1 and FUS interactions with synthetic oligonucleotide sequences of varying lengths support these studies, where longer oligonucleotides promote multimerization of RBPs with RNA and dynamic protein:RNA interactions that represent the initiation of phase separation (Kim and Myong, 2016; Niaki et al., 2020). Longer RNAs preferentially drive LLPS of the SG-nucleating protein, G3BP1, with a minimum length requirement about 10-fold higher than that required for G3BP1 binding (~250 nt), further demonstrating the importance of protein multimerization upon single RNA molecules to drive LLPS and MLO assembly (Guillén-Boixet et al., 2020; Yang et al., 2020). Interestingly, ultra-course-grained models generated to simulate G3BP1/RNA interactions revealed that longer RNAs promote the formation of an interconnected meshwork of G3BP1 and RNA, resulting in higher-order assembly not observed in simulations involving shorter RNA species (Guillén-Boixet et al., 2020). These phase-separated networks of G3BP1/RNA, achieved via crosslinking mediated by G3BP1 clustering at distinct foci along long RNAs, serve as assembly platforms in the nucleation of stress granules through the recruitment of other protein/RNA components to unoccupied RNA binding sites and additional protein/RNA interconnectivity (Guillén-Boixet et al., 2020; Sanders et al., 2020; Yang et al., 2020). Recent transcriptomic studies highlight the importance of RNA length in the scaffolding of MLO assembly and revealed an enrichment in long RNAs contained within stress granules across various stressors (Khong et al., 2017; Namkoong et al., 2018).

Other anecdotal observations regarding the RNA content of other MLOs, such as paraspeckles and nuclear stress bodies, similarly support a positive relationship between RNA length and scaffolding of MLO assembly. For example, alternative isoforms of the long noncoding RNA NEAT1 have differing effects in nucleating paraspeckle assembly, with replacement of only the longer of the two isoforms (NEAT1_2) able to recover paraspeckle assembly in NEAT1 knockout MEF cells (Fox et al., 2018; Naganuma et al., 2012). The unique ability of the longer NEAT1_2 isoform drives the recruitment and resulting phase separation of PrLD-containing RBPs, like FUS and RBM14, through functionally-redundant elements not present in the shorter NEAT1_1 isoform (Chujo and Hirose, 2017; Yamazaki et al., 2018). A similar mechanism is proposed to underlie the formation of nuclear stress bodies, where heat-induced transcription of highly-repetitive satellite III (HSATIII) long noncoding RNA is linked to the recruitment of various IDR-containing RBPs and nucleation of these droplet-like structures (Biamonti and Vourc’h, 2010; Metz et al., 2004; Ninomiya et al., 2020).

Short RNAs prevent RBP multimerization and higher-order assembly

RNA length might also be a strong determinant to inhibit RBP phase separation. While direct comparisons between RNAs of differing length but similar valence and affinity are lacking, recent investigations of the buffering capacity of different types of RNAs (tRNA, NEAT1 lncRNA, and rRNA) revealed a near-linear relationship between RNA length and the antagonization of FUS LLPS (Maharana et al., 2018) (Figure 3a). Specifically, shorter RNA species (tRNA) were the most efficient solubilizers of FUS droplets, with the longer NEAT1 and rRNAs requiring much higher concentrations in solution for effective inhibition of phase separation (Maharana et al., 2018). Short tRNA species were also shown to outcompete other, presumably longer, RNAs to bind and prevent FUS assembly within stress granules. The above-mentioned smFRET studies of FUS multimerization with polyU RNA of different lengths demonstrates a length-dependence in the buffering of protein phase transitions, with longer polyU species allowing for FUS multimerization and dynamic FRET fluctuations which are representative of condensate formation (Niaki et al., 2020). Conversely, shorter polyU tracts (U10-U30) only allowed binding of single FUS monomers and did not promote FUS multimerization or dynamic RNA interactions in this study.

Figure 3. RNA length contributes to regulation of RBP LLPS and MLO dynamics.

(a) RNA length seemingly inversely correlates with its capacity to buffer phase transitions of RBPs (i.e. FUS). The low binding occupancy of short oligonucleotides and/or tRNA molecules may prevent multiple RBP binding, while longer RNAs (i.e. NEAT1_2 lncRNA, rRNA) may contain many binding sites that enable RNA scaffolding of RBP LLPS. Longer RNA length also may promote RNA:RNA interactions that contribute to LLPS and MLO formation, while shorter RNAs have less capacity to form these multivalent RNA/RBP networks. (b) Examples of different intracellular condensates/MLOs that contain RNAs of different lengths. While short RNA oligonucleotides are capable of effectively buffering RBP phase transitions in the cell (i.e. bait RNA oligonucleotides), longer RNAs tend to promote the formation of MLOs and tune their characteristics. MLOs with longer RNA components may exhibit a less dynamic biophysical state (i.e. myogranules, Titin mRNA) than those containing shorter RNA components (i.e. stress granules, non-translating mRNA) due to an increased propensity for RNA self-interaction and reduced molecular exchange.

Similarly, TDP-43 self-assembly is antagonized by short, TDP-43 targeting RNA oligonucleotides (French et al., 2019; Mann et al., 2019), but conversely can be promoted by long RNAs like NEAT1 lncRNA and titin mRNA that contain multiple binding sites for RBPs on single transcripts (Tollervey et al., 2011; Vogler et al., 2018) (Figure 3b). Thus, the ability of shorter oligonucleotides to inhibit higher-order assembly of these RBPs may be intrinsically linked to the low-occupancy binding capacity of these molecules, in contrast to longer multivalent RNA molecules that act as scaffolds for RBP multimerization and subsequent phase separation (Figure 3a).

RNA secondary structure influences RBP phase transitions

In addition to specific sequence motifs, many RBPs display some structural biases in their interaction with RNA substrates (Dominguez et al., 2018). The RNA secondary structures that are preferred by different RBPs tends to vary, likely as a result of the type and combination of RNA-binding domain(s) utilized for RNA interaction, but structure-specific recognition is often used along with cooperative sequence-based binding to achieve target specificity across RBPs, especially in those utilizing redundant binding motifs (Dominguez et al., 2018). The structural content of RNA molecules positively correlates with the number of individual protein interactors for a given RNA, suggesting that RNA secondary structure may enhance specific RBP binding at distinct transcript loci (Sanchez de Groot et al., 2019). Given these observations, one might expect RNA secondary structure to contribute to the coordinated nucleation of various MLOs. However, recent studies indicate that long, single-stranded, unstructured RNAs are preferred for initiation of stress granule assembly, and were more efficient drivers of G3BP1 LLPS in vitro (Guillén-Boixet et al., 2020; Yang et al., 2020). Transcriptomic analyses of RNAs found within stress granules support a possible lack of structural dependence for RNA scaffolding or incorporation within SGs, revealing weak associations between SG-enriched transcripts and either mRNA folding energy or GC-content (Namkoong et al., 2018). Conversely, RNAs enriched with linear sequence motifs like AU-rich elements (AREs) were more likely to be found within SGs in this study. Similar observations have arisen from investigations of NEAT1_2 lncRNA required for paraspeckle formation (Lin et al., 2018). Structural analysis showed a lack of conservation in secondary structure of the NEAT_2 lncRNA across evolution, despite the maintenance of its essential function in coordinating paraspeckle assembly (Lin et al., 2018). Additional studies identified a set of redundant subdomains within the middle domain of NEAT1_2 that contain multiple binding sites for essential paraspeckle proteins like NONO, SFPQ, FUS, and RBM14. These are proposed to initiate paraspeckle assembly via multimerization and phase separation upon cooperative binding to NEAT1_2 (Yamazaki et al., 2018).

Linear motifs commonly bound by RBPs are predicted to have a lower propensity to form defined secondary structures (Dominguez et al., 2018). This suggests that a higher overall structure of RNAs may diminish their ability to scaffold MLOs assembly by limiting RBP binding accessibility. However, local structural elements within or adjacent to binding motifs are likely important in determining interaction specificity in interactions with certain RBPs and may aid in the nucleation of MLOs (Dominguez et al., 2018). In a similar manner, overall or local structural flexibility of RNA molecules may be a major determinant of their regulation of RBP phase behavior. Multiple investigations show that structural rigidity of nucleic acids, for example due to complementary base-pairing or nucleotide composition, suppress LLPS or favor the assembly of less dynamic assemblies (Boeynaems et al., 2019; Shakya and King, 2018). Modulation of RBP phase behavior by RNA structural flexibility was also observed with endogenous RNA targets (CLN3/BNI1) for the RBP Whi3, found within functionally distinct assemblies in Ashbya gossypii (Langdon et al., 2018; Zhang et al., 2015). Here, Whi3 condensates formed along with CLN3 mRNA, which showed a more-compacted baseline conformation by smFRET, exhibit significantly increased viscosity when compared with the more-extended BNI1 mRNA that exhibited more dynamic interactions with Whi3 and incorporated more Whi3 protein into droplets than CLN3 (Langdon et al., 2018; Zhang et al., 2015). Thus, the structural flexibility of RNA molecules may allow for dynamic conformational remodeling of RNA upon RBP binding, as observed for RBPs like FUS, LAF-1, DDX3X, and Whi3 (Kim and Myong, 2016; Langdon et al., 2018; Loughlin et al., 2019), that promotes RBP multimerization and interactions across RBP/RNA complexes responsible for subsequent phase separation. Conversely, rigid RNA structures, may result in more static or tight monomeric binding that would oppose LLPS.

The influence of secondary structure is highlighted by the capacity of short synthetic oligonucleotides that solubilize protein aggregation induced by RNA digestion in whole cell lysate. The most effective solubilizing molecules containing single-stranded motifs including loops or bulges interspersed by double stranded regions (Aarum et al., 2020). Similar results were demonstrated following RBP precipitation via biotinylated isoxazole (b-isox) treatment in HeLa cell lysate, where mass spectrometry revealed a preferential antagonization of protein aggregation following incubation with highly-structured RNAs, whereas low-structure RNAs produced results similar to background controls (Sanchez de Groot et al., 2019).

Secondary structure drives RNA partitioning into condensates

In addition to directly affecting protein binding and phase behavior, RNA complementarity and secondary structure can also influence the incorporation of RNA into pre-existing membraneless assemblies. For example, secondary structures within nascent pre-rRNA are required for proper sorting in the nucleolus where a stem-loop structure within the 5’ end of the 47S pre-rRNA is necessary and sufficient for incorporation into pre-formed fibrillarin (FBL) droplets in vitro (Yao et al., 2019). Similar results were observed for oligonucleotide incorporation into pre-formed Ddx4 droplets, where ssDNA/ssRNA hairpins were preferentially absorbed over their unstructured ssDNA/ssRNA counterparts (Nott et al., 2016). Both single-stranded oligonucleotides exhibited enhanced partitioning when compared to dsDNA/dsRNA, and those double-stranded molecules that were absorbed were unwound and stabilized in single-stranded conformation within droplets. The favoring of compact, single-stranded nucleic acid structures within Ddx4 droplets was due to a reduced disruption of condensate organization arising from extended and/or rigid conformations that alter the transient interactions required to maintain the dynamic nature of these assemblies (Nott et al., 2016).

Beyond recruitment by RBP components of biomolecular condensates, secondary structure also mediates RNA incorporation into MLOs via RNA:RNA interactions. In the Whi3 studies mentioned above for example, it was shown that in addition to modulating interactions with Whi3 protein alone, RNA structure determined heterotypic RNA:RNA interactions and resulting RNA partitioning into Whi3 droplets formed with other RNA species (Langdon et al., 2018). Exposed regions of RNA complementarity were proposed to drive the differential incorporation of RNAs into Whi3/RNA assemblies, either contained within native structures of mRNAs found co-localized within Ashbya gossypii polarity granules in vivo or artificially induced in normally-segregated mRNAs by thermal unfolding or point mutations. Distinct mRNA partitioning was also observed in the absence of Whi3 protein, highlighting the role for RNA:RNA interactions in the sorting of functionally-related RNAs into specific RNP assemblies.

RNA:RNA interactions and membraneless organelle assembly

RNA self-assembly also contributes to the formation of physiological RNA-containing MLOs like stress granules. It is well known that RNA is capable of self-assembly into higher-order structures (reviewed by Grabow and Jaeger, 2014). However, RNA homopolymers, as well as total yeast RNA, also self-assemble into liquid-like droplets that recruit RBPs like hnRNPA1 (Van Treeck et al., 2018). Sequencing of RNAs enriched within total RNA assemblies created in vitro revealed a striking similarity to the stress granule-enriched transcriptome mentioned above, indicating overlaps in both RNA characteristics (longer RNAs enriched) and specific RNA transcripts contained within these assemblies (Namkoong et al., 2018). In addition to the above observations, alterations in osmotic conditions that favor RNA:RNA interactions (i.e. hyperosmotic stress) (Bounedjah et al., 2012) or the introduction of G-quadruplex-forming RNAs initiate intracellular stress granule assembly (Fay et al., 2017). Together, this raised the hypothesis that intermolecular interactions between RNAs play an active role in physiological RNP assembly (Van Treeck and Parker, 2018).

Homotypic RNA phase separation was recently shown to participate in the formation of pathological RNA foci in repeat expansion disorders like Huntington’s Disease (HD) and C9ORF72-ALS/FTD (Fay et al., 2017; Jain and Vale, 2017). Disease-associated repeat RNA (CAG_n for HD/GGGGCC_n for C9ORF72-ALS/FTD) can phase separate in vitro and form intracellular liquid- or gel-like RNA assemblies in an RNA length-dependent manner (Fay et al., 2017; Jain and Vale, 2017). Phase transitions of repeat RNA arose from the long GC-rich tracts contained within these repeats that allow for multivalent base-pairing and the formation of structural elements, such as hairpins and/or G-quadruplexes, since foci formation was reversed by interrupting complementary base-pairing interactions with blocking antisense oligonucleotides or the nucleic acid intercalator doxorubicin (Jain and Vale, 2017). Another similar repeat-expansion (CGG_n) in the FMR1 5’UTR linked to Fragile-X associated tremor ataxia syndrome (FXTAS) can also drive the formation of nuclear RNA foci (Cid-Samper et al., 2018). In an analogous manner to CAG/GGGGCC repeats described above, these CGG_n repeats recruit RNA-binding proteins in cellular models that are often observed to be sequestered within foci in patient tissue (Cid-Samper et al., 2018; Jain and Vale, 2017). Disruption of distinct hairpin structures formed by these repeats prevented both nuclear foci formation and RBP recruitment, again highlighting the role of RNA structure in mediating homo- and heterotypic RNA/RBP phase separation. Structure-dependent phase transitions of nucleic acids are also observed in non-disease contexts as other G-quadruplex-forming nucleic acids, such as regions of the SHR mRNA and ssDNA from the c-myc promoter region, similarly undergo LLPS that is highly sensitive to structural perturbations and can recruit proteins like linker histone H1 to condensates (Mimura et al., 2020; Zhang et al., 2019).

Taken together, these observations provide insight into the dynamic regulation of RBP phase behavior and MLO formation by distinct RNA species. Similarly, the above studies highlight the potential differences between RNA properties important for initial condensate nucleation and subsequent maintenance/recruitment of additional components to membraneless assemblies. Returning to the conceptual framework of client-scaffold relationships as it relates to MLO dynamics, it is possible that scaffold RNAs exist as a distinct class of components from client RNAs and tend to exhibit divergent characteristics that mirror their function (Ditlev et al., 2018). In this sense, longer, flexible, generally-unstructured RNAs that contain a multitude of different RBP binding motifs likely serve as the building blocks of MLOs through the scaffolding of important multivalent RBP:RNA, RBP:RBP, and RNA:RNA interactions responsible for phase separation. Following initial assembly, functionally-related client RNA molecules may then be partitioned into condensates through specific complementary interactions with scaffold RNAs or interactions with RBP components mediated by specific sequence and/or structural elements, which are then unwound/stabilized upon entry to maintain the biophysical state of the condensate and provide additional multivalency (Ditlev et al., 2018). Conversely, shorter RNAs suppress LLPS and MLOs assembly by limiting RBP multimerization due to the inherently lower binding occupancy of these molecules. High RNA concentrations, such as those observed in the nucleus of cells, result in a similar buffering effect, as the enhanced binding availability of different RNAs likely support monomeric RBP:RNA interactions.

Conclusions and future directions

The emerging field of biomolecular condensates indicates that the assembly and maintenance of MLOs is a crucial component of normal cellular physiology. However, as is the case with many other functional pathways, disruption of this dynamic process can have devastating consequences. The pathological deposition of RBPs in neurodegenerative disease provides an interesting hypothesis that highlights how alterations in the regulation of protein phase transitions drive disease via protein LLPS and MLO dysfunction. One common element of disease-linked RBPs, prion-like domains, may lie at the nexus of the dichotomy between physiological and pathological phase separation. PrLDs allow for functional phase transitions during normal cell physiology but also confer a propensity for aberrant phase transitions in pathological contexts, such as the presence of genetic mutations within these regions (Harrison and Shorter, 2017), diminished proteostasis during aging (Hipp et al., 2019), and/or chronic cellular stress (Baradaran-Heravi et al., 2020). However, recent studies demonstrated that another commonality between disease-linked RBPs, interaction with RNA ligands, has a profound modulatory influence over the assembly and dynamics of membraneless assemblies in vitro and within the cell.

On one hand, RNA exists as a frequent and often critical component of many MLOs in the cell and promotes phase separation of multiple RBPs through a scaffolding mechanism as described above (Banani et al., 2016; Ditlev et al., 2018). In contrast, RNA seems to oppose phase transitions of disease-linked RBPs, both in vitro and within the cell, and prevents toxicity associated with their aberrant assembly (Maharana et al., 2018; Mann et al., 2019). One potential explanation for the contrasting effects of RNA on RBP phase behavior is the relative subcellular concentrations or molar ratios of RBPs and RNAs, such as the nucleus, or in vitro reactions (Burke et al., 2015; Maharana et al., 2018; Zhang et al., 2015), where excess RNA inhibits RBP phase separation (Figure 2a–b). However, recent work indicates that the distinct properties of different RNA species play a critical role in the assembly and composition of membraneless assemblies. Some of the few examples point to a positive correlation between RNA length and the ability to scaffold RBP phase transitions (Niaki et al., 2020; Wang et al., 2020) (Figure 3), but direct evidence for the effects of different RNA secondary structures on MLO assembly and function remains limited to draw precise and generalizable conclusions. Nonetheless, these studies demonstrate a key concept outlined in this review: that RNA is not a uniform entity in its participation in biological phase transitions and should thus be carefully considered in the same detail as diverse protein components of membraneless assemblies. Of course, future studies aimed at exploring the effects of diverse RNA species on the phase behavior of single RBPs and modulating the RNA contents of various MLOs will be needed to truly decipher the intricacies of RNA’s role in MLO form and function. This knowledge, in turn, will hopefully deepen our understanding of biomolecular condensates in cellular physiology and inform future efforts in the development of RNA-based therapeutics to counter pathological RBP phase transitions in disease.

Abbreviations:

AD

Alzheimer’s disease

ALS

amyotrophic lateral sclerosis

CBD

corticobasal degeneration

CTE

chronic traumatic encephalopathy

FTD

frontotemporal dementia

FTLD

frontotemporal lobar degeneration

HD

Huntington’s disease

IBM

Inclusion body myopathy

IDR

Intrinsically-disordered region

LATE

Limbic-predominant age-related TDP-43 encephalopathy

LCD

Low-complexity domain

LLPS

Liquid-liquid phase separation

MLO

Membraneless organelle

PD

Parkinson’s disease

MSP

Multisystem proteinopathy

PrLD

Prion-like domain

SCA1/2/3

Spinocerebellar ataxia types 1, 2 & 3

VCPDM

vocal cord and pharyngeal weakness with distal myopathy

WDM

Welander distal myopathy