Formation, Function, and Pathology of RNP Granules

Abstract

Ribonucleoprotein (RNP) granules are diverse membrane-less organelles that form through multivalent RNA-RNA, RNA-protein and protein-protein interactions between RNPs. RNP granules are implicated in many aspects of RNA physiology, but in most cases their functions are poorly understood. RNP granules can be described through four key principles. First, RNP granules often arise because of the large size, high localized concentrations, and multivalent interactions of RNPs. Second, cells regulate RNP granule formation by multiple mechanisms including post-translational modifications, protein chaperones, and RNA chaperones. Third, RNP granules impact cell physiology in multiple manners. Finally, dysregulation of RNP granules contributes to human diseases. Outstanding issues in the field remain determining the scale and molecular mechanisms of RNP granule function and how granule dysfunction contributes to human disease.

etoc

Ribonucleoprotein (RNP) granules are diverse membrane-less organelles visible in the light microscope. While their diversity suggests functional differences, this Perspective focusses on their four key principles and discusses outstanding issues to determine the scale and molecular mechanisms of RNP granule function and how granule dysfunction contributes to human disease.

Introduction

RNP granules are diverse membrane-less intracellular RNA-protein assemblies (Figure 1), operationally defined as granules when visible in the light microscope. Some RNP granules are constitutively present in cells including cytosolic processing bodies (P-bodies), TIS granules, cytoplasmic U snRNP bodies (U-bodies), nuclear Cajal bodies, the nucleolus, nuclear speckles, and paraspeckles ^1,2. Some RNP granules, such as stress granules or P-bodies form, or enlarge, during translation repression and ribosome run-off ^1,2. RNP granules can also form from double stranded RNA (dsRNA) that play a role in the innate immune response ³. Other RNP granules can be unique to distinct cell types. For example, neurons contain RNP transport granules ⁴, while many embryos contain RNP granules called germ granules that localize maternal RNAs for spatially and temporally regulated translation during development, including P-granules, nuage, Balbiani bodies, chromatoid bodies, and sponge bodies⁵. RNP granules can also be found in bacteria, such as the bacterial RNP-bodies (BR-bodies) ⁶.

Figure 1: RNP granule examples.

A diverse set of ubiquitous, cell type-specific or stress induced RNP granules. Objects are not to scale and RNP granule numbers are not representative of cellular conditions.

The wide diversity of RNP granules suggests that there will be differences in the form and function of different granules, although in this review we focus on the similarities between RNP granules. We review fundamental principles of RNP granules, illustrated with specific examples, and provide some thoughts, perspectives, and future issues in this field.

Different RNP granules have been proposed to have diverse impacts on RNA processing, RNP assembly, translation and/or storage of messenger RNAs (mRNAs), localization of RNAs, and RNA degradation. As discussed below, RNP granules can affect biochemical reactions by creating a high local concentration of specific molecules, sequestering molecules away from other reactants to alter specificity, localizing RNAs to certain regions of the cell for local function or segregation during cell division, or by creating biochemical environments that have unique properties. Understanding RNP granule composition, formation, and regulation is critical to fully decipher RNP granule function and pathology.

Distinct RNP granules contain different sets of RNAs and proteins. This conclusion is based on examining specific proteins or mRNAs by immunofluorescence, or single molecule RNA fluorescence in situ hybridization (smFISH) analysis, as well as unbiased proteomic and transcriptomic analysis of purified RNP granules or using BioID/Apex approaches, in which reactive molecules are localized to specific compartments/assemblies to label other co-localized components ^7-17. One limitation of BioID/Apex approaches is that they will also identify many interactions independent of RNP granules since most of the molecules of a given granule component are typically present in the bulk cytosol or nucleoplasm. This limitation can be avoided by analyzing purified RNP granules and/or extensive validation of granule components identified by proximity labeling.

Most nuclear RNP granules are dominated by a specific set of RNAs. For example, Cajal bodies are enriched in small nuclear RNAs (snRNAs), small nucleolar RNAs (snoRNAs), and small Cajal body-specific RNAs (scaRNAs) ¹⁸, while paraspeckles are defined by the nuclear paraspeckle assembly transcript (NEAT1) RNA ¹⁹. In contrast, cytosolic stress granules, P-bodies, and P granules typically contain a heterogenous mix of mRNAs ^7,8,12. Although each RNP granule has its own characteristic proteome, many RNA binding proteins (RBPs) can be observed in multiple RNP granules. For example, stress granules can contain RBPs^10,20 also present in the nucleolus ¹⁴ or nuclear paraspeckles ¹¹. Similarly, U-bodies contain snRNPs with Sm and Lsm proteins that are also found in Cajal and histone locus bodies ²¹.

RNP Granule Assembly

RNP granules form through multivalent protein-protein, RNA-protein, or RNA-RNA interactions between individual RNPs in a process referred to as biomolecular condensation. This umbrella term has grown to encompass any membrane-less organelle, body, granule, speckle, aggregate, or puncta, due to the common multivalent assembly mechanisms for concentrating biological molecules, including liquid–liquid phase separation (LLPS) ^22,23, liquid-solid-phase separation (LSPS), or phase separated assisted percolation (PSAP) ²⁴. These assembly mechanisms are characterized by material properties including liquid-like droplets with a physically defined boundary generated by surface tension, or gel or solid-like structures. However, alternative condensation/assembly mechanisms can also exist. For example, dynamic hubs were proposed to form due to increased local concentrations of diffusive biomolecules around a less mobile scaffold such as a transcription site or a membrane ^25-28. We refer to RNP granules as biomolecular condensates independent of the underlying mechanism by which condensates form, or any specific material properties ²⁴. Multiple reviews provide a more detailed explanation on the biophysical and biochemical parameters of these assembly mechanisms, explain the difficulties of distinguishing the underlying mechanisms in cells and provide suggestions for their characterization ^24,28-30.

RNPs: The building blocks of RNP granules

Since RNP granules are assemblies of RNPs, it is important to consider RNP organization. Some RNPs of small functional RNAs are well described. For example, the U1 small nuclear RNP (snRNP), as revealed by cryo-EM or X-ray structural analysis consists of a single 165 bases RNA³¹ (MW = 165 bases x 330 g/mole/base = 54.5 kDa) with multiple associated proteins such that the RNP is ~20% RNA/80% protein by weight (assuming a total protein mass of 240 kDa ³¹). In this assembly, only a few regions of the snRNA are exposed and available for potential intermolecular RNA-RNA interactions. This implies that the assembly of snRNPs into Cajal bodies might be more dependent on protein interactions. This fits with the critical role for the multivalent coilin protein in Cajal body assembly, where the N-terminal domain, important for multimerization and protein binding, is sufficient for Cajal body assembly ³².

In contrast, several observations imply that messenger RNPs (mRNPs) will have significant amounts of RNA bases that are not bound by protein and could be available for intermolecular RNA-RNA interactions, although this is an understudied area. First, density gradient analyses demonstrate that the nuclear Balbiani ring premessenger RNP particles, which are very large mRNPs expressed in the salivary gland of Chironomus tentans, are ~50-60% protein by mass/40-50% RNA ³³. Assuming a similar ratio occurs in other mRNPs (we use 50/50 for simplicity), this implies that a 3 kb mRNA (MW = 3000 bases x 330 g/mole/base = 990 kDa) would be bound by a total protein mass of ~990 kDa. This corresponds to 18 ~55 kDa proteins. Interestingly, this estimate is similar to the number of RBPs bound to mRNAs calculated from nuclease protection analyses ^34,35. Thus, if the average RBP is bound to 10 nucleotides of a 3 kb mRNA, this argues that only 6% of the RNA bases are bound by RBPs. Moreover, mRNAs show increased chemical reactivity to dimethyl sulfate (DMS) in vivo as compared to in vitro ³⁶, which implies a significant fraction of mRNA sequences within cells are unprotected from chemical modification, and thereby could be available for intermolecular RNA-RNA interactions. Lastly, by comparing the number of molecules and their nuclear or cytosolic location of the 110 most abundant RBPs in U-2 OS cells ³⁷, it is estimated that the nucleus contains ~16x more RBPs/nucleotide of mRNA than the cytosol. This is consistent with cytosolic mRNAs typically being engaged with ribosomes, implying that cytosolic untranslating mRNAs will have substantial amounts of RNA regions not bound by RBPs ³⁸.

The RNA molecules found in RNP granules are often much larger than RNP granule proteins (Figure 2A). For example, the average mRNA in a stress granule is 7.1 kb or 2.34 million Daltons ⁸, which is 34 times larger than the stress granule assembly factor, G3BP1 (~52 kDa) (Figure 2A). Assuming a mass ratio of 50% protein/50% mRNA, an 7.1 kb mRNA would contain ~42 proteins of ~55 kD (Figure 2A). This creates an mRNP with numerous sites for potential interactions with other mRNPs though protein-protein or intermolecular RNA-RNA interaction, with over 50% of the surface area estimated to be exposed RNA (Figure 2A). These interaction sites can be considered analogous to stickers, separated from each other by spacers, discussed in some models of condensate formation ²⁴.

Figure 2: Prevalent Role of RNA.

A) Comparison of the average size of a SG enriched mRNA, an average mRNA and an average 55 kDa protein drawn to scale with size estimates from Khong and Parker, RNA, 2020. RNA bound RBPs are shown as colored circles. Estimates for the RBP number and RNA and protein surface area are indicated. B) Possible types of intermolecular RNA interactions.

Larger RNA or RNP size allows for multiple interactions between individual RNAs or RNPs and a corresponding increase in avidity. Because of this principle, larger RNAs are more prone to self-assembly in vitro ³⁹. Further, the composition of RNAs in stress granules, P-bodies, P granules, and BR bodies is all biased towards longer RNAs ^6-8,12. Parsimoniously, longer mRNAs are more enriched in RNP granules due to more sites for intermolecular RNA-RNA interactions, and more potential binding sites for RBPs and protein-protein interactions between RNPs ⁴⁰.

The importance of protein-protein interactions

Genetic and biochemical evidence has demonstrated a role for protein-protein interactions in promoting RNP granule formation ^30,41. For example, mutations in, or deletion of, the dimerization domains of G3BP1 block the assembly of stress granules ^42,43. Similarly, a C-terminal domain in MEG-3 interacts with PGL proteins to promote P granule assembly in worms ⁴⁴. P-body assembly in yeast also depends on interactions between several proteins, including Pat1, the Lsm1-7p complex, Edc3, Dcp2, and Dhh1, that use folded domains and short linear motifs (SLiMs) to interact ⁴⁵.

Protein-protein interactions between intrinsically disordered regions (IDRs) of RBPs are not as essential for RNP granule formation as frequently suggested. IDRs of RBPs were initially suggested to be important in RNP granule formation since some purified RNP granule proteins self-assemble in vitro in an IDR dependent manner, and those in vitro assemblies show dynamic behavior similar to RNP granules in cells ^46-48. However, mutational analyses argue that RNP granule formation is mainly driven by sequence-specific interactions between well-folded domains or between well-folded domains and SLiMs found in IDRs ^30,45,49. For example, mutations in the PGL-1 dimerization domains block P granule assembly ⁵⁰. Similarly, mutations in the coilin N-terminal domain that impair coilin multimerization or Nopp140 binding prevent Cajal body assembly ³². Since IDRs can weakly interact with many proteins, they may contribute to RNP granule formation in cells when coupled to proteins undergoing specific interactions by an increase in avidity to enhance the overall complex assembly ⁵¹. In addition, several IDRs are RNA binding domains ⁵², and therefore might contribute to RNP granules by enhancing RNA-protein interactions.

The protein-protein interactions that contribute to RNP granule assembly can vary in their specificity and strength. Proteins within some RNP granules such as the Balbani body in humans, and the Rim4 RNP granule in yeast, form very stable, yet specific and reversible, cross-beta structures similar to amyloids ^53,54. Alternatively, specific protein-protein interactions of nano- and micromolar affinities can form between well-folded protein domains as well as between SLiMs and well folded domains ^30,45. On the other extreme, interactions such as charge-charge, pi-pi, cation-pi ⁵⁵, and short cross-beta interactions referred to as low-complexity aromatic-rich kinked segments (LARKS), can be both of lower specificity and affinity ⁵⁶.

Overall, while protein-protein interactions are diverse and vary in their specificity and strength, sequence-specific interactions emerge as the main driver of RNP granule formation, while IDRs contribute to the overall complex assembly and their properties.

The importance of RNA

RNA is thought to be a critical component of many RNP granules since most RNP granules including stress granules, nuclear speckles, Cajal bodies, and the nucleolus disassemble or rearrange when RNA is limited by either transcriptional repression or nucleolytic degradation ^57-61. This is consistent with RNP granule formation being directly connected to the concentration of RNA components. It should be noted that some RNP granules appear more dependent on protein-based assembly mechanisms. For example, germ granules in Drosophila formed from the Oskar protein are resistant to RNAse treatment once formed arguing for a protein dominated assembly mechanism ⁶². Similarly, P-bodies and super-enhancer condensates containing eRNAs, persist after RNA degradation ⁵⁷.

RNA affects RNP granule formation by serving as a platform for the recruitment of RBPs. Mutational analyses show that RBP domains are often required for RNP granule assembly. For example, the RNA binding domains of G3BP1 are essential for stress granule assembly in vivo ^63,64. Similarly, mutations in the RNA-binding domain of Edc3 block P-body assembly in yeast ⁶⁵. The combination of RNA binding and protein-protein interaction domains in multiple RNP granule components allows such proteins to bind to RNAs, and then through protein-protein interactions contribute to the formation of a higher order RNP granule assembly.

RNA also affects RNP granule assembly by forming intermolecular RNA-RNA interactions. This was first demonstrated by genetic evidence that the oskar and bicoid mRNAs form intermolecular duplexes targeting them to distinct RNP granules for transport to opposite poles of the Drosophila oocyte ^66,67. These are genetically selected interactions that allow specific biological function. However, RNAs can also form random/promiscuous interactions between molecules. Because of this property, essentially any RNA is capable of self-assembly in vitro ^39,68. This appears biological relevant since RNA self-assembly in vitro can largely recapitulate the yeast stress granule transcriptome ³⁹. In addition, the formation of RNA assemblies in cell lysates is enhanced by RNAs that are more prone to stable intermolecular interactions ^69,70.

RNA is prone to self-assembly for multiple reasons. First, cellular RNAs are often large and thereby have a large surface area available for interactions (Figure 2A). Second, RNA folding leads to multiple features, including hairpin loops and single stranded bulges, that can interact in trans with other RNAs. Such hairpin loops, referred to a kissing loops, are often utilized in genetically programmed interactions, such as the intermolecular interactions of the oskar mRNA ⁶⁶ and multiple other contexts ⁷¹. Kissing-loop interactions generally only involve a few bases in the loop. Thus, long RNAs with many stem-loops, are expected to form random interactions between stem-loops, and single stranded bulges. Such intermolecular interactions will contribute to intermolecular RNA—RNA interactions, as has been seen with the condensation of riboswitches in vitro ⁷². When complementary, hairpin regions have a high propensity to re-arrange into duplexes, with duplexes being favored at increasing RNA concentrations and metal ions ^73-75, which could further stabilize RNA self-assemblies. Finally, RNA is also prone to self-assembly due to the diverse mechanisms of possible RNA-RNA interactions. Specifically, RNAs can interact in trans by Watson-Crick and non-Watson Crick base-pairs, triple helix interactions, hydrogen-bonding and electrostatic interactions involving bases and/or the ribose or phosphate backbone, co-axial stacking, and metal ions ^{26,73,74,76-78} (Figure 2B). Such non-Watson Crick interactions are what allows RNA condensation of homopolymers ^79,80.

While there is limited direct evidence for intermolecular RNA-RNA interactions within cellular RNP granules, intermolecular crosslinks within RNA condensates in vitro have been demonstrated ⁸¹. Moreover, RNA crosslinking studies in unstressed human or mouse cells have identified over 1000 mRNA hybrids ^82-85. Intramolecular interactions are mostly restricted to the untranslated regions (UTRs), whereas intermolecular RNA-RNA interactions are spread across the entire transcript ⁸⁴. Therefore, we hypothesize that such intermolecular interactions would be more abundant upon ribosome run off and increased RNA concentration in a granule. However, it remains formally possible that intermolecular RNA interactions don’t occur within RNP granules, though this would require cellular mechanisms to prevent intermolecular RNA interactions given the high RNA concentration and its propensity for self-assembly within RNP granules.

We consider the role of promiscuous RNA-RNA interactions in promoting RNP granule formation analogous to protein aggregation ⁸⁶. Protein aggregation is a non-native, nonfunctional stage, where unfolded or misfolded proteins expose hydrophobic regions prone to forming promiscuous intermolecular interactions ⁸⁷. We suggest that an analogous situation exists for RNA, where exposed RNA regions due to loss of ribosomes or RBPs allows for promiscuous RNA-RNA interactions resulting in RNA aggregation ⁸⁶. Such promiscuous interactions between mRNAs could occur frequently simply by chance. For example, two 3 kb mRNAs have over 30,000 potential 4 base-pair interactions. Even if 99% of these interactions are limited by intramolecular RNA structures or RBPs, two random RNAs could potentially interact at multiple sites. For mRNAs, such interactions could be normally prevented by translating ribosomes ⁸⁶. Upon stress, in analogy to protein aggregates, RNA aggregates could either be disaggregated by RNA chaperones or cleared from cells by autophagosomes, as observed for stress granules^88,89. If irreversible, we speculate that persistent RNA aggregates may contribute to disease, which might be most prevalent under conditions that compromise RNA aggregation disassembly machineries (see section on Regulation of RNP granules by the RNA chaperone network for a more detailed explanation).

RNP granule properties

Regardless of the specific assembly mechanism and composition, RNP granules share four general features.

Impacts of thermodynamics and kinetics

While biomolecular condensation is a concentration driven assembly of macromolecules, several observations argue that RNP granules in cells are not at strict thermodynamic equilibrium and are therefore affected by the energy dependent rates of components entering and exiting the granule (kinetic control). For example, the persistence of nuclear Cajal bodies, nuclear speckles, and the nucleolus is dependent on the continued energy dependent synthesis of nascent RNAs ^90-92. However, these RNP granules are stable once isolated from the cell ^93,94, implying the presence of energy dependent disassembly mechanisms, which could involve RNA helicases, protein chaperones, RNA processing reactions, and protein modifications. Similarly, the formation of stress granules is limited by energy dependent disassembly mechanisms ⁸¹.

It is anticipated that granule composition can be impacted by the kinetics of assembly. For example, since the mRNAs in stress granules are essentially static⁹⁵, stress granule mRNA composition is likely to be controlled by the kinetics of mRNP assembly into stress granules, and not their thermodynamic stability. In contrast, when the interactions of individual RNPs in a granule are sufficiently weak, or the time scale sufficiently long, to allow for continual exchange and spatial rearrangement, one expects RNPs to enter granules in a thermodynamically controlled manner and form the most stable interactions over time. This principle has not been demonstrated for any granule to date. One possible example of a thermodynamically controlled RNP granule might be the granules at the posterior pole of Drosophila oocytes, where translationally repressed mRNPs are delivered to the germ plasm during a prolonged period of oogenesis, and those mRNPs then self-assort over time into smaller homotypic RNP assemblies, referred to as clusters ⁹⁶.

Reaching a critical concentration for assembly

The formation of RNP granules is driven by binding interactions between molecules that are concentration dependent. Moreover, because the interactions are highly cooperative and involve multiple molecules, the condensation to an assembled state occurs with a very steep concentration dependence.

Biomolecular condensation of dispersed components

RNP granules can form by condensation of dispersed components in the cytosol or nucleus due to an increased concentration or a change in affinity between components. For example, stress granules form when the concentration of ribosome free mRNPs in the cytosol increases ⁹⁷. Condensation occurs because individual RNPs can interact with each other through protein-protein, protein-RNA or RNA-RNA interactions, and when the concentration of RNPs reaches a critical point, there are sufficient RNPs available for a highly cooperative transition to an assembled state. In model systems, this concentration is referred to as a saturation concentration Csat, but for the multicomponent RNP granule assembly in cells, individual proteins when overexpressed do not show definable Csat ⁹⁸. One limitation of understanding RNP granules is that it has not been tested if RNP granule assembly shows a Csat when examining the endogenous RNA components of granules, which are more likely to be the limiting factors for assembly.

Biomolecular condensation through increased local concentration

Condensation can also occur when RNPs are produced or anchored at specific sites, leading to high localized subcellular concentrations. For example, during Drosophila oogenesis, translationally repressed mRNPs are transported to, and anchored at, the posterior pole of the oocyte ⁹⁹. Subsequently transported mRNPs interact with the initial anchored population, a variety of other mRNAs and synthesized germ plasm proteins to form germ granules ¹⁰⁰.

The recruitment of RNPs to membranes can also create high local concentrations that promote RNP granule assembly ¹⁰¹. For example, Whi3 granules in the filamentous fungus Ashbya, stress granules and P-bodies can be stably associated with the ER ^102,103 and P granules with nuclear pores ^59,104. Moreover, RNP granules can also associate with endosomes, lysosomes, or microtubules ¹⁰⁵.

The formation of several nuclear RNP granules is likely to occur because of the production of a high local concentration of specific RNAs at transcription sites and the subsequent recruitment of RBPs and RNAs. For example, the nucleolus, histone locus body, and Cajal bodies form at the sites of transcription of their resident rRNAs, histone mRNAs and snRNAs, respectively ^18,106-108. In each of these cases, the arrayed nature of these genes creates a very high local concentration of the specific RNAs leading to a unique RNP granule formation^18,106-108. Nuclear speckles have a more heterogeneous composition but are similarly surrounded by active genes that deliver nascent mRNA transcripts into nuclear speckles ¹⁰⁶. In contrast, paraspeckles form at the NEAT1 gene even as a single locus because the large size of the NEAT1 RNA and it’s high transcription rate can facilitate paraspeckle formation ¹⁰⁹. Additional evidence that the high local concentration of specific RNAs nucleates the assembly of these RNP granules is that tethering NEAT1, histone, or satellite RNAs to specific chromatin regions is sufficient to drive assembly of a paraspeckle, a histone locus body, or a satellite RNA assembly at those sites ¹¹⁰.

Thus, RNP granule assembly can occur when RNPs capable of assembly reach the critical concentration for assembly. Since RNPs can have multiple sites for both protein and RNA interactions, the critical concentration is not expected to be of a single protein, but instead of the RNP, although this has not been directly tested. Developing approaches to directly measure the concentrations of specific cellular RNPs could address this issue. Ideally such approaches could also allow the assessment of assemblies below the size easily observed in the light microscope.

Formation of smaller assemblies or clusters below the critical concentration

Below a certain concentration, it is expected that smaller RNP assemblies (Figure 3A), sometimes referred to as clusters, will form. RNP granule components are particularly prone to forming such clusters since they form through a complex network of interactions, and loss of any one protein component generally does not prevent the other interactions from occurring. For example, in yeast cells, mutations that abolish visible P-body formation still lead to the formation of smaller P-bodies when assessed by nanoparticle tracking ¹¹¹. Similarly, biochemical fractionation, imaging approaches, and proximity-based detection methods suggest clusters of mRNPs related to stress granules can form even in the absence of stress ^10,13,20. Such clusters are also predicted from the equilibrium binding of RNP components, and have been observed in in vitro studies of some RBPs ¹¹², model LLPS systems ^113,114, and with misfolded proteins in cells ¹¹⁵.

Figure 3: Common RNP granule properties.

A) Smaller RNP clusters exist below a critical concentration. Formation of larger granules is a cooperative process, triggered by an increase in protein-protein, protein-RNA and RNA-RNA interactions. B) Cooperative assembly could be driven by two energetic contributions: 1) Compared to smaller RNP clusters, larger clusters contain more binding sites leading to overall stronger RNP interactions through increased avidity. 2) RNPs inside granules display stronger intermolecular interactions compared to RNPs interacting with the solvent. C) After initial RNP granule formation, new additional interactions lead to the stabilization of the RNP granule. D) Interactions between homotypic (the same granule) or heterotypic (two different) RNP granules lead to fusion or docking, respectively. Weak or stable homotypic RNP granule interactions lead to faster or slower fusion. E) Homotypic or heterotypic RNP interactions can also lead to subassemblies within a granule or lead to core shell architectures.

The initial formation of clusters would then be expected to create larger building blocks for the RNP granule (Figure 3A), which by virtue of their larger size, are expected to have more interaction sites, leading to higher avidity between each cluster (Figure 3B). Strikingly, lattice light-sheet microscopy (LLSM) video analysis of stress granule formation has revealed the initial formation of small G3BP1 assemblies even 5 min after translation repression that then coalesce into larger easily observed stress granules ¹¹⁶. The possibility of smaller clusters of RNP granule components that gives rise to large RNP granules raises a key issue of whether biological function occurs at the RNP, cluster, or at the larger RNP granule scale.

These observations suggest that an important step in the assembly of RNP granules is a transition from cluster formation to a highly cooperative assembly into a larger RNP granule (Figure 3A). Although this process is not understood in detail, it is likely driven by two energetic contributions (Figure 3B). First, cooperative assembly can be enhanced when the average cluster is large enough to have more favorable binding interactions with other clusters (Figure 3B, left), leading to assembly of the larger clusters into the condensate, with smaller clusters excluded (which has been seen for model protein assemblies¹¹³). Second, when the RNP granule becomes large enough that some RNPs are completely inside the assembly their inclusion into the granule will have additional energetic benefits since they no longer have the energetic cost of interacting with the surrounding solvent (Figure 3B, right) ¹¹⁵. These two factors imply that the formation of assemblies with a discreet inside portion surrounded by RNPs interacting with the bulk cytosol or nucleoplasm, could create unique biochemical environments allowing specific functions.

Stabilization driven by additional interactions after initial RNP granule formation

The initial formation of an RNP granule creates a high local concentration of RNAs and proteins, which is expected to lead to additional molecular interactions, thereby further stabilizing the RNP granule (Figure 3C). This principle is predicted from theoretical analyses ¹¹⁷ and has been observed within RNA-based assemblies in vitro, where recruitment of mRNAs to the surface of an RNA assembly leads to new intermolecular interactions between mRNAs ¹¹⁸. Moreover, creating high local concentrations of RBPs with IDRs that are prone to forming aberrant protein fibers can increase the rate of fiber formation either in vitro ^46-48 or in cells ^119,120. In cells, such stabilized RNA-mediated assemblies are important for proper mRNA localization during Drosophila oogenesis ¹²¹. Since RNP granules contain these inherent positive feedback loops, cells must utilize energy consuming mechanisms to limit RNP granule formation (see section on Regulation of RNP granule assembly).

Such reinforcing molecular interactions can explain why a wide range of RNP granules are stable in lysates including P-bodies ⁶⁰, stress granules ¹⁰, P granules ¹²², Cajal bodies, the nucleolus, and nuclear speckles ^93,94. Similarly, since RNA is much larger than proteins and can form a larger number of RNA-based interactions, it explains why RNAs are generally very static in RNP granules, and in some cases rigidly positioned ^95,123. This contrasts with many RNP granule protein components that are highly dynamic ⁹⁷, and is consistent with RNA being an important scaffold of many RNP granules ¹²⁴.

Homotypic and heterotypic RNP interactions

While homotypic or heterotypic interactions commonly describe interactions involving two identical or two different domains or proteins, there can also be homotypic or heterotypic interactions between related RNPs, clusters of RNPs, or RNP granules. The nature of these homotypic or heterotypic interactions gives rise to RNP granule docking and fusion, as well as subdomains within RNP granules.

RNP granule fusion or docking

Many RNP granules can interact and fuse, or dock, at their surfaces (Figure 3D). This can be described by differences in homotypic and heterotypic interactions of RNP granules. When homotypic RNP granules (e.g. two stress granules) dock, they typically fuse into a single granule (Figure 3D, left), which is observed with multiple RNP granules including P-bodies, stress granules, and the nucleolus ^125,126. The fusion of two homotypic RNP granules can be understood as the rearrangement of the molecular interactions driving assembly to allow for formation of new interactions between components of each granule. The rate of RNP granule fusion is a function of the overall strength/dynamics of the underlying molecular interactions with weak dynamic interactions rapidly rearranging into a sphere to minimize surface area. When interactions are quite stable, the fusion is slow, and the RNP granule can have a more mesh-like appearance ^118,127. RNP granules can become, or appear round, due to surface tension, diffusive molecules around a scaffold or a smaller, diffraction limited size ^25,125. In cells, the rapid dynamics of molecular interactions in RNP granules that allow fusion are maintained by ATP dependent processes ^125,128.

Heterotypic RNP granules frequently dock, but do not merge (Figure 3D, right). This is observed for cytoplasmic stress granules and P-bodies ¹²⁶, germ granules ¹²⁹ and various nuclear granules ^130,131. The docking of different RNP granules can be explained by each RNP granule having unique homotypic interactions (e.g. stress granule RNPs preferentially interacting with themselves) driving their assembly but having RNA or protein molecules on their surfaces that can form heterotypic granule interactions (e.g. some stress granule RNPs can interact with P-body RNPs) ¹³². Docking of heterotypic granules has been proposed to facilitate RNPs movement between stress granules and P-bodies ¹³³. However, such movements must be quite rare since despite a large number of single molecule observations, only one individual mRNA molecule has been observed to transition between a P-body and stress granule ¹³⁴. In contrast, numerous mRNPs have been observed to transition between the surfaces of P-bodies and stress granules by cytoplasmic diffusion ^95,134. Similarly, in vitro, bulk movement of RNA from one condensate to another through the diffuse phase can be observed following DEAD-box protein ATP hydrolysis and release of the mRNA ¹³⁵. Thus, docking sites may not be preferred sites of movement of RNAs between granules, which instead occurs through diffusion in the cytoplasm/nucleoplasm.

Subdomains and surface properties

Portions of individual RNP granules can have different composition and properties (Figure 3E), which could allow for different functional biochemical environments. For example, stress granules, paraspeckles, the nucleolus, and P granules contain regions of different protein composition, often described as subassemblies or core-shell architecture ^{10,44,136-138}. Distinct subdomains (Figure 3E, left) within a granule can allow for compartmentalization and ordering of biochemical reactions. For example, in the nucleolus, the presence of specific compartments is suggested to allow for vectorial transport of maturing ribosomes ⁹⁸. The simplest model is that distinct subdomains arise because individual proteins or RNAs have stronger interactions with a subset of the granule components.

Differential composition can also occur at the surface compared to the interior (Figure 3E, right), which can influence whether RNP granules dock and/or fuse, as well as regulating the exchange of RNP granule components. For example, the MEG3/4 proteins can bind the surface of P granules, alter their surface tension, which then leads to slower rates of fusion, and surface exchange of molecules ¹³⁹.

An important goal for future experiments will be to determine the specific homo- and heterotypic interactions that allow for the formation of RNP granule subdomains and docking. Understanding such interactions may also lead to additional insights into the regulation and functions of RNP granules.

Regulation of RNP granule assembly

RNP granule assembly is regulated in multiple manners including post-translational modifications of RNP granule proteins, protein chaperones, RBPs, and DEAD-box RNA helicases.

Post-translational modifications

Modifications known to affect RNP granule assembly include methylation, phosphorylation, acetylation, N-linked glycosylation, PARylation, and ubiquitination ¹⁴⁰. Modifications of RNP granule proteins can be understood as changing the affinity of critical protein-protein and protein-RNA interactions to alter the assembly parameters of the RNP granule. For example, asymmetric versus symmetric arginine demethylation of ligands of the SMN (survival of motor neuron) protein determine whether nuclear gems and Cajal bodies formed segregated granules or dock ¹³⁰. Alternatively, post-translational modifications can trigger active disassembly mechanisms. For example, during heat shock recovery in mammalian cells ubiquitination of G3BP1 leads to its extraction from stress granules by the VCP ATPase, triggering stress granule disassembly ^141,142.

Regulation of RNP granules by protein chaperones

Protein chaperones and their adaptor proteins can regulate the formation as well as disassembly of RNP granules. RNP granule clearance after stress is promoted by chaperones presumably by facilitating mRNA reentry into translation and/or by reducing protein interactions that stabilize granules ^141,143-145. Moreover, the CCT complex (chaperonin containing tailless complex polypeptide 1 complex), which is known to assist in protein folding, limits the formation of stress granules and P-bodies, at least in yeast ^10,146.

Regulation of RNP granules by the RNA chaperone network

Cells regulate RNP granule formation through a network of proteins that modulate RNA-RNA interactions, which we refer to as the RNA chaperone network ¹¹⁶. This network exists because RNA, like protein, can form promiscuous intermolecular interactions or aggregates.

One member of the RNA chaperone network is eIF4A, a DEAD box helicase. While eIF4A is an essential regulator of translation initiation, it’s high expression levels suggest additional roles in RNA metabolism ¹¹⁸. eIF4A binds RNA in an ATP-dependent manner, thereby limiting the self-assembly of RNA in vitro and reducing stress granule formation in cells ¹¹⁸. A role for eIF4A, and other RNA helicases, in limiting RNA condensation is analogous to protein chaperones, such as HSP70, limiting the aggregation of misfolded proteins. eIF4A binds to, and destabilizes, short less stable duplexes ^147,148, leading to duplex disassembly, ATP hydrolysis and eIF4A release. As eIF4A does not directly use the energy of ATP hydrolysis to unwind RNA duplexes ¹⁴⁹, eIF4A is not efficient at resolving long stable duplexes ¹⁴⁸. However, the ability to bind and destabilize weak random RNA duplexes is an ideal property for an RNA chaperone to limit random RNA condensation. This suggests that other DEAD-box, and related DEAH-box proteins, may play similar roles in limiting inappropriate RNA-RNA interactions.

Another member of the RNA chaperone network is the abundant RBP YB-1, which reduces in vitro RNA assembly and, when overexpressed in cells, reduces stress granule formation ¹⁵⁰. Such abundant monomeric RBPs can be considered analogous to small heat shock proteins (sHsps), a class of protein chaperones limiting irreversible protein aggregation ^87,151. We anticipate that other components of the RNA chaperone network will include nucleases that limit the intracellular concentration of RNA ^6,152,153, or RNA modification enzymes that destabilize RNA-RNA interactions.

DEAD-box RNA helicases as ATP dependent switches in regulating RNP granules

DEAD-box proteins can positively and negatively regulate RNP granules in multiple manners ¹⁵⁴. Since DEAD-box proteins can displace proteins from RNAs ¹⁵⁵, they can play a role in removing proteins that might promote or inhibit RNP granule formation. Moreover, several DEAD-box proteins found in RNP granules contain accessory domains that can promote RNP granule formation, either through oligomerization and/or binding other RNP granule components to increase multivalency, as shown for DDX3 in stress granules ¹⁵⁶ and for DDX6 in P-bodies ¹⁵⁷.

Since the RNA-binding of DEAD-box proteins is released by ATP hydrolysis, they become ideal proteins to regulate both the formation and dissolution of RNP granules ^135,158. For example, while Dhh1, the yeast homolog of DDX6, promotes the formation of P-bodies, mutations inhibiting Dhh1's ATPase activity increases P-bodies ¹⁵⁸. Thus, it is not surprising to find DEAD-box proteins in almost all cellular RNP granules ¹⁵⁹, and the regulation through their ATPase activities, play important roles in the assembly and resolution of RNP granules.

Functions of RNP Granules

Historically, RNP granules have been assumed to have a function because they exist. However, two observations raise the possibility that the large RNP granules may simply form due to the biophysical properties of cells and RNPs. First, increased molecular crowding will shift the equilibrium towards granule formation ^160,161. Second, RNPs are particularly prone to forming granules given their large size and multivalent properties (Figure 2). Thus, some perturbations might trigger RNP granule assembly without a specific biological function.

Consistent with RNP granules forming due to biophysical properties of RNPs, where examined, only a small fraction of the components localizes to the granule. For example, stress granules only contain ~15% of the untranslating mRNPs and ~18% of the G3BP1 protein ^162,163. Similarly, yeast P-bodies contain a small fraction of P-body components ¹⁶⁴. Three possible mechanisms could explain why only a minority of a component is localized to the RNP granule. First, only specific modified forms of a component may partition into the granules, as seen with the active caspase-3 showing increased partitioning into stress granules as compared to inactive caspase-3 ¹⁶⁵. Second, localization could be reduced due to a limiting RNP granule component. For example, P-body partitioning of the decapping enzymes is increased when the pool of capped untranslated mRNPs increases ^166,167. Finally, and more commonly suggested, the components concentration within the granule could be defined by their saturation concentration and has been suggested as a mechanism for intracellular concentration buffering ⁵⁵. While this was shown to be the case for endogenously tagged nucleophosmin (NPM1) and nucleoli during interface when comparing to mitosis where nucleoli are disassembled ¹⁶⁸, this could not be recapitulated for nucleoli, stress granules, Cajal bodies and P-bodies upon overexpression and their associated key scaffolding proteins NPM1, G3BP1, coilin, and DCP1A⁹⁸, presumably because these RNP granules are dependent on limiting RNA scaffolds.

One possibility is that the function of an RNP “granule” may occur at scales below the observed larger RNP granule. For example, the formation of an mRNA decapping complex on mRNAs may be sufficient for mRNA decapping, and P-body formation is simply a consequence of the multivalent nature of that assembly ^65,169,170. Similarly, although RIG-I can form condensates when induced by interferon, the activation of RIG-I signaling occurs before, and in the absence of, condensate formation ¹⁷¹. Indeed, one hypothesis is that most large RNP granules simply reflect the formation of smaller functional multivalent assemblies prone to “incidentally” forming higher order RNP granules ².

Taken together, this suggests that the biophysical properties of proteins and RNAs can lead to the formation of RNP granules with small perturbations from homeostasis. However, analogous to how the biophysical properties of amphipathic lipids have allowed the formation of biological membranes, the biophysical properties driving RNP granule assembly have been used by evolution to create function. Whether all RNP granules have a "function" per se remains to be evaluated. However, because RNP granules alter the concentrations and interactions of intracellular molecules, we anticipate that all RNP granules will impact cellular physiology in some way. For example, stress granules, which form partially by an RNA aggregation process limited by ATP driven mechanisms¹⁷², can alter cell cycle control and cell survival ¹⁶⁵ even though their molecular functions during active stress is not yet fully understood. Fundamentally, the key question that remains to be addressed is whether function, and/or impact on cell physiology, occurs at the individual RNP, at smaller clusters of RNPs, or at the larger granule.

Limitations of assigning RNP granule function

Historically, RNP granules have been suggested to have a diversity of biological functions based on two general types of observations. First, RNP granules are often proposed to have functions because molecules involved in a particular pathway are concentrated in the assembly. For example, P-bodies were proposed to be involved in RNA degradation by concentrating components of the RNA degradation machinery ¹⁶⁷, although this interpretation was challenged by the small fraction of decapping machinery present in P-bodies ¹⁶⁴ and mutations preventing P-body formation do not robustly affect mRNA decay ⁶⁵.

Similarly, many cytosolic RNP granules such as stress granules, germ and neuronal mRNP granules, were proposed to be involved in translation repression since they increase during translation repression and are primarily composed of untranslating mRNPs ⁵. However, it is now clear that while untranslating mRNPs are more efficiently recruited into such RNP granules, the RNP granule is generally not required for translation repression per se. For example, miRNA-based translation repression of mRNAs occurs in cells whether or not P-bodies can assemble ¹⁷³, and cells efficiently repress translation during stress whether or not stress granules can be assembled ⁶³. Similarly, loss of the MEG-3/4 proteins in C. elegans reduces P granule formation, but does not limit translation repression of developmentally regulated mRNAs ¹².

RNP granules may regulate translation in two manners that remain to be fully examined. First, it is unknown whether the recovery of translation for mRNAs packaged into mRNP granules will require additional specific RNP granule disassembly steps for their re-entry into translation. Second, it is possible that some RNP granules might have signaling properties that lead to the repression of translation. For example, mutations in DDX3X that drive medulloblastoma trigger constitutive stress granule assembly and repress translation globally in a manner dependent on stress granule formation ¹⁷⁴.

A second line of evidence used to infer RNP granule function is that when loss of a protein disrupts an RNP granule, and alters a cellular function, the granule is inferred to perform that function. However, since most granule assembly factors have other biological roles, this argument is compromised by possible granule-independent functions of mutated proteins.

Proposed functions for RNP granules

Despite these limitations, experiments have suggested diverse functions for RNP granules with differing degrees of experimental support. We discuss a subset of these functions to illustrate four potential, and distinct, biophysical mechanisms by which RNP granules affect cellular function (Figure 4). In many cases, a critical issue that needs to be resolved is whether the "function" only requires smaller subassemblies, or the larger visible granule.

Figure 4: Biophysical principles by which RNP granules could lead to function.

A) Increase in high local concentration could lead to increase in reaction or assembly rates. B) Sequestration or sponging biomolecules into RNP granules could inhibit their activity in the cytoplasm or nucleoplasm. C) Transport of RNP granules for localized translation. D) Unique environments e.g. RNP granules could prevent irreversible aggregation of biomolecules.

First, RNP granules can promote the rate of biochemical interactions and/or reactions by creating a high local concentration of molecules (Figure 4A). For example, the high local concentration of NLRP6 in dsRNP granules is suggested to increase the rate of inflammasome signaling ¹⁷⁵. Similarly, the formation of Cajal bodies is thought to increase the rate of snRNP assembly by concentrating the protein and RNA components ¹⁷⁶. The concentration of molecules in RNP granules, and on their surfaces, may also create environments that promote translation. This is suggested by the observations that during mouse spermiogenesis, the FXR1 RBP promotes RNP granule formation that recruits the translation machinery to promote translation of mRNAs targeted to that RNP granule ¹⁷⁷. It is anticipated that concentrating enzymes and/or substrates involved in RNA processing reactions within RNP granules such as the histone locus body, the nucleolus, or nuclear speckles might increase RNA processing reactions.

Second, sequestration of components into RNP granules has been suggested to reduce their concentration and function in the bulk cytosol or nucleoplasm, with granules essentially acting like a sponge (Figure 4B). For example, stress granules have been shown to limit apoptotic signaling by the sequestration of proteins that normally promote apoptosis ^165,178. Similarly, in several human diseases repeat expansion RNAs form RNP granules, which can sequester RBPs leading to loss of function phenotypes ¹⁷⁹, although whether the formation of the RNP granule per se, or just the increase in competing binding sites in the repeat expansion RNA, is causing the loss of function phenotype is unclear. Sequestration of mRNAs may also serve in some cases to repress their translation, and/or access to degradative enzymes. This was demonstrated to occur in reconstituted systems where assembly of CAPRIN, FMRP and RNA can create assemblies that limit the access of the RNA to exogenous deadenylation or translation machinery ¹⁸⁰. Sequestration and storage of mRNAs into RNP granules may provide another level of regulation limiting their translation. Consistent with this possibility, it is important during development that mRNAs accumulate in P-bodies ¹⁸¹, and Balbani bodies ⁵³, which are then released for translation at specific times.

Even when an RNP granule appears to sequester only a fraction of a protein or RNA, this could still impact cell physiology for three reasons. First, there could be smaller similar assemblies sequestering additional molecules. Second, many regulatory events in cells show ultrasensitivity. Finally, it can be that a larger fraction of a specific modified/functional form of a protein or RNA is preferentially sequestered.

A third general function of RNP granules is to affect the localization of mRNAs, e.g. in neuronal or germ granules (Figure 4C). For example, P granules in C. elegans are required for proper concentration of mRNAs in the germline blastomere ¹². Similarly, the interaction of mRNP granules with the surface of lysosomes can allow their transport to dendrites in neurons for localized translation ¹⁰⁵.

Finally, RNP granules could also create unique chemical environments, particularly once they have reached the condensate scale and create an internal structure that could have different properties than bulk cytosol or nucleoplasm (Figure 4D). For example, stress granules have been shown to have a lower pH than the surrounding cytosol ¹⁸². Different chemical conditions within a condensate has also been illustrated by in vitro experiments where condensates of P-body components alter Dcp2 conformation and decapping rates ¹⁸³. Similarly, RNP granules might provide a unique folding environment for macromolecules, which is suggested by RNAs serving as potent protein chaperones ¹⁸⁴, which might occur in the nucleolus ¹⁸⁵. Moreover, RNP granules may alter RNA folding either by preventing RNA aggregation by proteins limiting the formation of RNA-RNA interactions ^86,186, or by increasing intermolecular RNA-RNA interactions due to the high local concentration of RNAs ¹⁸⁷. RNP granules might also affect the chemical properties of the surrounding cytosol or nucleoplasm, which is suggested by the preliminary evidence that RNA condensation can affect the bulk diffusion of cytosolic macromolecules ¹⁸⁸.

RNP Granules in Disease

RNP granules affect human diseases in two distinct manners. First, the normal physiological formation of an RNP granule can alter cell physiology in a manner that alters the course of a disease, such as stress granules contributing to tumor progression. Alternatively, mutations that lead to the loss of specific RNP granules or formation of an aberrant RNP granule can alter cell physiology in abnormal manners leading to disease by a diversity of mechanisms including sequestering RBPs, or by enhancing the rates of protein aggregate formation.

Cancer

Stress granules show multiple connections to tumor progression ¹⁸⁹. First, stress granules are observed in multiple tumors ^174,190,191. Second, signaling pathways that enhance tumor progression can also promote stress granule formation ¹⁹². Third, increased expression of multiple stress granule assembly factors correlates with poor patient prognosis ^190-192. Finally, stress granule formation is caused by many chemotherapies ^191,192. One prevalent interpretation is that stress granules help cancer cells survive perturbations in the tumor microenvironment or chemotherapeutic stress.

Several models have been proposed for how stress granules allow cell survival including the hypothesis that concentration of proapoptotic proteins such as RACK1, TRAF2, RhoA/ROCK1 and active caspases in stress granules reduces apoptosis ^165,178,192. Other studies argue stress granules promote cell survival by reducing reactive oxygen species ¹⁹³.

Increased stress granules caused by genetic mutations may also play a role in tumor progression. Mutations in DDX3X, a DEAD-box helicase involved in translation initiation and stress granule formation, frequently occur in medulloblastoma ¹⁷⁴. These cancer-associated mutations in DDX3X both reduce global translation and lead to the formation of stress granules. Strikingly, reducing aberrant stress granule formation by either DDX3X N-terminal deletion, or G3BP1/2 knockout, restores translation, suggesting that stress granule formation per se is altering translation and thereby contributing to tumor progression in some manner ¹⁷⁴.

Antiviral response

RNP granule formation also plays roles in the antiviral responses. This was suggested by the observations that many viruses, including noroviruses, polio, Dengue virus, and the MERS and SARS2 coronaviruses, subvert the formation of stress granules either by cleavage, or inhibition, of G3BP1/2 ¹⁹⁴. A popular model is that G3BP1/2 promotes the formation of a type of antiviral stress granule, which is proposed to enhance the activation of the innate immune system ¹⁹⁴. Similarly, some viruses were shown to disrupt P-bodies¹⁹⁵. However, whether and how P-bodies play a role in antiviral response is unclear.

RNP granules forming from dsRNA are suggested to play a role in the activation of the innate immune system. Specifically, RNP granules containing dsRNA and NLRP6 have been suggested to activate NLRP6, while distinct dsRNA granules containing multiple dsRBPs referred to as dsRNA-induced foci (dRIFs), modulate the activation of PKR ³.

Neurodevelopmental Diseases

Several neurological or neurodegenerative diseases are affected by the formation of aberrant RNP granules. In some cases, the aberrant RNP granule is triggered by alterations in DEAD-box proteins that modulate RNP granules. For example, specific mutations in DHX30 or DDX3X lead to the formation of constitutive stress granule-like assemblies and cause a defect in neurodevelopment leading to intellectual disabilities ^196,197. Interestingly, rare mutations in DDX6, a DEAD-box helicase that both promotes P-body formation and can affect stress granule composition ¹⁹⁸, also leads to intellectual disabilities ¹⁹⁹, although whether this is directly connected to alterations in P-bodies or stress granules remains to be determined.

Neuro- and muscular degenerative diseases

Aberrant RNP granules also affect a variety of neuromuscular diseases including amyotrophic lateral sclerosis (ALS). In this case, a reoccurring theme is that a variety of mutations and alterations lead to the formation and/or persistence of aberrant RNP granules that can contain proteins misfolded into highly stable beta-amyloid like structures. For example, mutations in several stress granule-enriched proteins such as TDP-43, FUS, Ataxin-2, hnRNPA1, hnRNPA2B1, EWSR1 lead to protein aggregation associated with ALS, inclusion body myopathy (IBM), and other forms of muscle and neurodegenerative disorders ²⁰⁰. Moreover, the concentration of these proteins either in phase separations in vitro, or in optically triggered RNP granules in cells, leads to an increased rate of amyloid formation ^46-48,119. Taken together, this suggests excess or persistent RNP granules may contribute to the formation of misfolded proteins, with pathological downstream consequences ^201,202.

Similarly, defects in protein chaperones such as DNAJB6, BAG6, or the VCP complex, that disassemble RNP granules, as well as defects in the clearance of RNP granules by autophagy ²⁰², can be causative in these diseases. This accumulation of aberrant RNP granules appears to be relevant to disease progression since decreasing RNP granules by depletion of the stress granule assembly factor, ataxin2, can rescue neurotoxicity in model systems ^203,204.

The mechanisms by which aberrant RNP granules lead to pathology remain incompletely understood. In some cases, the sequestration of RBPs in aberrant granules can lead to loss of function in the nucleus and alterations in RNA processing. For example, the accumulation of TDP-43 in cytoplasmic inclusions in motor neurons in ALS leads to detrimental alterations in RNA processing ²⁰⁵. Alternatively, although it has not been clearly demonstrated, aberrant RNP granules might activate signaling pathways that eventually lead to cell death.

RNP granules have also been suggested to enhance the formation of tau fibers or aggregates, which occurs in diverse neurodegenerative diseases referred to as tauopathies ²⁰⁶. The central observation is that fibrillar tau aggregates in cell line and mouse models of disease preferentially form in conjunction with nuclear speckles, and with a related cytosolic RNP granule, referred to as a mitotic interchromatin granule or cytosolic speckle ^207,208. Consistent with these observations, the SRRM2 protein, which is found in nuclear and cytosolic speckles, is observed in tau aggregates in postmortem tissues from tauopathy patients ^207,209.

Diseases of Repeat Expansions

Aberrant RNP granules also occur, and may contribute to the pathology, in some diseases caused by repeat expansions such as the expansion of a G4C2 repeat in the C9orf72 intron causing ALS, or a CUG repeat in the 3' UTR of the DMPK gene causing myotonic muscular dystrophy. In both these cases, RNP granules are thought to form containing transcripts with the repeat expansion ²¹⁰. Whether these granules solely form at sites of transcription, or form in part by intermolecular RNA-RNA interactions ⁶⁸ remains to be determined. In either case, such repeat containing RNAs sequester RBPs leading to alterations in RNA regulation and downstream pathology ²¹⁰.

Conclusion and Future Directions

RNP granule impacts on cell physiology

Major issues to address going forward are determining the impacts of RNP granules on cell physiology, the scale of assembly that creates the relevant functional biochemical environment, and the molecular mechanisms of those impacts. Five main experimental approaches will be important to define and then understand how RNP granules impact cells.

First, the correlation of RNP granule formation and function is strengthened if multiple different perturbations that limit the formation of a specific RNP granule in independent manners lead to the same phenotype. For example, multiple different mutations in Lsm14A, or overexpression of the NBDY micropeptides, all of which limit P-body formation, lead to increased MARF1 endonuclease activity, strengthening the conclusion that P-bodies per se regulate MARF1 activity ²¹¹.

A second approach to link function to RNP granules is a "granule rescue” experiment, which restores RNP granule formation independent of the initial perturbation and thereby demonstrates that RNP granule formation per se is the causative event. For example, stress granule formation can be restored in G3BP1/2 KO cells by hyperosmotic stress. Such "granule rescue" experiments have demonstrated roles for stress granules per se in the regulation of nuclear-cytoplasmic transport during stress ²¹². Such "granule rescue" experiments can be performed by restoring granule function with artificial proteins designed to solely restore RNP granule formation, and lack any other potential complicating function ^43,213.

Third, imaging approaches to examine biological processes at the single molecule level both inside and outside of RNP granules can be powerful, particularly if coupled with concurrent imaging approaches to measure the size of the assembly being imaged. For example, imaging decapping rates by different fluorescent tags on the cap and body of the mRNA has provided clear evidence that decapping rates can be different in different RNP condensates, at least in vitro ¹⁸³. Similarly, the use of Suntag reporters that allow the imaging of translation on individual mRNAs, has shown that while most mRNAs in stress granules are not translated, translating mRNAs can associate with stress granules, at least transiently ⁹⁵ and in some cases can be localized within stress granules ²¹⁴.

Fourth, since the fractions of molecules in RNP granules are often small, to examine potential functions resulting from sequestration, it is necessary to carefully determine the fraction of the sequestered component in RNP granules, its exchange rate, and how that change in concentration might affect a biological process. For example, stress granules are proposed to reduce apoptosis by sequestering executioner caspases. Strikingly, while only 10% of total caspase-3 is localized in stress granules, more than 40% of the active caspase-3 is sequestered in stress granules, providing a rationale for their effect on apoptosis ^165,178. However, this is a major limitation in many RNP granule studies where the quantitative assessment of the fractions of components in granules, and their specific modification status, are lacking.

Finally, understanding the molecular functions of RNP granules will require in vitro systems that can be subject to robust experimental manipulation.

Defining the composition and dynamics of RNPs

Another critical issue to understand is the composition and dynamics of entire individual mRNPs, which is poorly understood. E.g. How many RBPs are really bound to an mRNA at a given time? How heterogenous and dynamic are RBPs on individual mRNAs from the same transcriptional unit? How much RNA is exposed in an mRNP? This is extremely important not only for understanding how RNPs assemble into RNP granules, but will also provide new insights into the mechanisms and regulation of RNA processing, localization, translation, and RNA decay. This is currently a difficult area to study for mRNAs due to their heterogeneous nature of mRNAs, but might be approachable in model systems with high levels of specific mRNPs, such as viral infections, and by using mutations and/or inhibitors to trap RNPs at specific stages of their lifecycle.

Understanding granules in disease and therapeutic intervention

A final important area is to understand the molecular mechanisms by which both normal and aberrant RNP granules contribute to human disease, and to determine if therapeutic intervention can be beneficial. For example, pharmacological targeting of stress granule formation has been proposed to be useful for limiting tumor progression ²¹⁵, and for limiting the aggregation of RBPs that contribute to neuromuscular degenerative diseases ²⁰⁰. Similarly, understanding the molecular mechanisms of how RNP granules affect disease will allow more targeted interventions that might be more beneficial. For example, cytoplasmic tau aggregation was observed to nucleate from mitotic interchromatin granules and cytoplasmic speckles (CSs) ²⁰⁸. If the molecular features of those assemblies can be identified that promote tau fiber propagation, compounds targeting those features could be novel therapeutic agents.

An interesting potential area is a fuller understanding of the mechanisms that limit RNP granule formation, and whether their failure contributes to human disease. Here it will be of interest to determine if some diseases mediated through protein aggregation of RBPs such as FUS, TDP-43, Annexin-11, Ataxin-2, are also affected by RNA aggregation. A role for combined protein-RNA aggregation in disease could explain the rare mutations in different RNA helicases in patients with similar neurodevelopmental diseases and persistent RNP granules ^196,197.