Prion-like domains as epigenetic regulators, scaffolds for subcellular organization, and drivers of neurodegenerative disease

1Alternative protein folding underpins prion-based phenomena. Prions are infectious proteins that can adopt many functionally distinct conformations, at least one of which is self-replicating. Whereas the mammalian prion protein (PrP) forms prions that cause devastating neurodegenerative diseases, prions in yeast are often benign and can even confer selective advantages. Amyloid fibril formation is a unifying feature of both mammalian and fungal prion phenomena, and indeed underpins several age-related neurodegenerative diseases including Alzheimer’s disease and Parkinson’s disease. Thus, understanding key determinants of this basic biophysical process as it relates to yeast prions will help illuminate new therapeutic strategies for human disease.

2The yeast prion states, [PSI+] and [URE3], are embodied by self-replicating conformers of their protein determinants Sup35 and Ure2, respectively. Sup35 is a translation termination factor in yeast; the yeast prion state [PSI+] is associated with a nonsense suppression phenotype. Prion-switching behavior of Sup35 can be readily assessed in yeast harboring a premature stop codon in their ADE1 gene: [psi~] cells (which lack Sup35 prions) are red on rich media and require adenine whereas [PSI+] cells appear white on rich media and can grow on media lacking adenine.

3Yeast prion formation is typically mediated by modular and transferable prion domains (PrDs), which are low complexity sequences enriched in polar, uncharged amino acids such as glutamine, asparagine, tyrosine, and serine as well as glycine. An exception is found in Mod5, which forms beneficial [MOD+] prions without a canonical PrD, and instead harbors an amyloid core-forming region enriched in hydrophobic residues. Importantly, deletion of the PrD from Sup35 and Ure2 does not affect normal protein function. Conversely, PrDs can be appended to model proteins to generate novel engineered prions. Furthermore, the sequence of the PrD can be scrambled and still encode prion behavior. Thus, it is the amino acid composition and not the exact primary sequence of a PrD, which is critical for prionogenesis.

4The modular, transferable, and randomizable nature of PrDs has proven enormously instructive in the quest to discover novel prions. A hidden Markov model, based on the amino acid composition of PrDs of Sup35, Ure2, and Rnq1, was developed to query the S. cerevisiae genome for novel prions, yielding ~200 candidates (cPrDs). To characterize these, chimeras consisting of cPrDs fused to the C-terminal domain of Sup35 were expressed in yeast and cPrDs were assessed based on their ability to mimic the [PSI+] phenotype (e.g. red-white colony color switching). Further refinement of cPrDs by biochemical and yeast phenotypic analysis led to the discovery of Mot3 prions. Importantly, not all cPrDs conferred prion behavior, and subsequent algorithms have been used as filters to improve prediction accuracy and even to design PrDs capable of bona fide prionogenesis.

5[MOT3+] is a prion state formed by self-replicating con-formers of its protein determinant, Mot3. Mot3 is a transcription factor that modulates a variety of processes including mating, carbon metabolism, and stress response by repressing anaerobic genes such as DAN1 during aerobic growth. Perhaps most importantly, [MOT3+] governs the acquisition of multicellular growth phenotypes in yeast through transcriptional regulation of FLO11. [MOT3+] enables acquisition of facultative multicellular growth phenotypes including invasive growth on poor nitrogen sources, complex colony morphology as a starvation response, and flocculation in liquid media. Furthermore, similar results were found in wild strains harboring [MOT3+]. These findings illustrate the potential of prions to facilitate rapid adaptation to environmental cues.

6Yeast prions can confer selective advantages in various circumstances, but can also be neutral or detrimental in other settings. The beneficial phenotypes conferred by yeast prions are often observed under stress conditions, which has led to the suggestion that yeast prions constitute bet-hedging devices, which can reveal potentially adaptive genetic diversity in fluctuating environments. This process is facilitated by the conformational range of PrDs, which can access multiple, distinct cross-β structures or strains. Moreover, protein folding is exquisitely sensitive to environment, allowing even subtle changes to favor one conformation, and therefore one function, over another. In the case of Sup35 prions, [PSI+]-mediated stop codon read-through allows expression of cryptic genetic variation that accumulates in 3′ untranslated regions at many genetic loci. Indeed, Sup35 prions act as evolutionary capacitors that release cryptic genetic variation under stress to facilitate the rapid evolution of adaptive traits.

7The adaptive significance of yeast prions, particularly Sup35 prions, has been contested. There has been a lack of evidence that Sup35 and Ure2 prions arise in wild yeast (although Rnq1 prions were readily found), leading to speculation that prions were merely ‘diseases’ or artifacts of laboratory cultivation. However, a survey of 690 wild yeast strains from diverse biological niches found 10 strains harboring Sup35 prions, 43 harboring Rnq1 prions, and 6 harboring Mot3 prions. Prions conferred a range of phenotypes that increased fitness of these yeast strains under a wide variety of stresses, and the prion phenotypes could become genetically fixed, thus fulfilling key predictions for bet-hedging prions. However, several of the advantageous [PSI+]-dependent phenotypes of wild yeast strains were not replicated in another study. Nonetheless, the preponderance of evidence suggests that yeast prion proteins undergo environmentally-sensitive, alternative folding to effect epigenetic changes that increase fitness in response to fluctuating environments.

8Do human proteins contain domains similar in amino acid composition to yeast PrDs? We applied an updated PrD detection algorithm, PLAAC (for Prion-Like Amino Acid Composition, with default core length of 60 and alpha set to 0.5), to the human genome (Ensembl GRCh38.p5, release 83) and uncovered 240 genes out of ~20,000 protein-coding genes (~1.2%) harboring a domain compositionally similar to annotated yeast PrDs, termed a prion-like domain (PrLD). The complete list of human proteins with a PrLD, including the location of each PrLD, is presented in Table S1. Remarkably, 72/240 (30%) of these proteins are annotated with the gene ontology (GO) molecular function “RNA binding” and 79/240 (~33%) with the GO molecular function “DNA binding” (Table 1). These are among the nine terms from the generic GO slim (an abridged versions of the full gene ontology) that are significantly enriched for human PrLD-containing proteins (Fisher’s exact test, with Holm’s adjusted p<0.05); the others are molecular functions “transcription factor activity, protein binding,” “nucleic acid binding transcription factor activity,” and “transcription factor binding”; biological processes “chromosome organization”, “mRNA processing,” and cellular components “nucleoplasm” and ‘nucleolus’ (Table 1); 174/240 (~73%) of human PrLD-containing proteins were annotated in at least one of these categories (Fig. 1). There are also three GO Slim categories in which PrLD-containing proteins are significantly underrepresented: molecular functions “signal transducer activity” and cellular components “plasma membrane” and “mitochondrion” (Table 1). Our findings suggest that RNA-binding proteins (RBPs) and DNA-binding proteins that reside primarily in the nucleus are significantly overrepresented among the collection of PrLD-containing human proteins. Thus, PrLDs feature prominently at the critical functional interfaces between nucleic acid and protein.

9Several human RBPs with PrLDs including ataxin 1, ataxin 2, TDP-43, FUS, TAF15, EWSR1, hnRNPA1, and hnRNPA2 (Table S1) feature prominently in the pathology and genetics of a number of fatal neurodegenerative diseases, including amyotrophic lateral sclerosis (ALS), frontotemporal dementia (FTD), and spinocereballar ataxias. For example, TDP-43 mislocalizes from the nucleus to cytoplasmic inclusions in degenerating neurons in ALS and FTD, and is an intrinsically aggregation prone protein. The PrLD of TDP-43 confers this intrinsically aggregation-prone property. Almost all ALS- and FTD-linked TDP-43 mutations lie in the PrLD, and several of these mutations can promote deleterious TDP-43 misfolding and enhance proteotoxicity in diverse model systems. Likewise, multisystem proteinopathy (MSP) can be caused by missense mutations in the PrLD of hnRNPA1 or hnRNPA2. These mutations alter a gatekeeper aspartate residue and introduce a potent steric zipper motif into the PrLD, which accelerates formation of self-templating hnRNPA1 and hnRNPA2 fibrils. Furthermore, polyglutamine expansions in the PrLD of ataxin 1 cause spinocereballar ataxia 1, and promote ataxin 1 aggregation. These three striking examples and many others suggest that human proteins with PrLDs are prone to deleterious misfolding events that underpin neurodegenerative disease. Thus, special attention to these proteins is urgently warranted. We suggest that human proteins bearing a PrLD (Table 1) should be scrutinized as potential etiological agents of various degenerative diseases, which might be revealed via gene sequencing and histopathological examination of protein localization.

10Do human proteins with PrLDs form bona fide prions? Currently, there is no evidence that any human protein with a PrLD can form a prion like PrP, which can naturally transmit devastating neurodegenerative disease between individuals. Nonetheless, PrLDs do enable proteins to spontaneously form self-templating fibrils in isolation. Intriguingly, ataxin 1 bearing a polyQ expansion within the PrLD can form oligomeric structures that induce local spread of ataxin 1 pathology in transgenic mice. Moreover, TDP-43 and TDP-43 fragments containing the PrLD (193–414) can form fibrils that elicit TDP-43 aggregation in cell culture. Furthermore, detergent-insoluble fractions from ALS brains contain TDP-43 fibrils and induce TDP-43 aggregation in cell culture. Thus, TDP-43 may access a prion-like conformation, which may even be transmitted across axon terminals. Indeed, phosphorylated TDP-43 pathology in ALS has been interpreted to spread in a sequential manner with highly discernible stages that might indicate involvement of axonal pathways. Prion-like conformers have been proposed to underlie this spreading phenomenon in ALS and other disorders. Although intriguing, compelling proof of formation of prions will require their de novo construction from purely synthetic protein and an ability to infect wild-type mice with a neurodegenerative disease, as has been achieved with PrP and α–synuclein. Even then, only PrP has been shown to form prions that can spread disease naturally between individuals in a population. Evidence is currently lacking that α–synuclein conformers can be infectious in this way.

11These various connections with neurodegenerative disease have led to a negative view of PrLDs in human proteins. However, PrLDs are found in 240 human proteins, and so presumably may serve some beneficial or essential function. Indeed, many genes that encode proteins containing PrLDs are essential in mammals. Furthermore, unlike the PrDs of Sup35 and Ure2, the PrLDs of several human proteins, including TDP-43, hnRNPA1, and hnRNPA2 play a critical role in protein function. For example, the PrLD of TDP-43 is not required for RNA- or DNA-binding activity, but is critical for alternative splicing of some mRNAs and for protein-protein interactions with other hnRNPs, including hnRNPA1, hnRNPA2, and FUS, as well components of the Dicer and Drosha complexes. Likewise, the PrLDs of hnRNPA1 and hnRNPA2 make important contributions to the splicing activity of these proteins. Recent reports suggest that PrLDs may have an important role in critical phase transition events that provide organizational scaffolds for various membraneless organelles, including RNP granules and nuclear subcompartments.

12In addition to classical membrane-delimited organelles, the eukaryotic cell is also organized by membraneless organelles. These include nucleoli, Cajal bodies, gems, paraspeckles, and PML bodies in the nucleus and processing (P) bodies, stress granules, and P granules in the cytoplasm. Yet, many questions remain about the physical basis by which these compartments form and function. One hypothesis that has gained significant attention is that these membraneless organelles form through phase separation. However, precisely what phase architecture these membraneless organelles adopt has remained the subject of intense scrutiny.

13Early evidence that membraneless organelles may be liquid-like came from the study of P granules. In the Caenorhabditis elegans embryo, polarization along the anterior-posterior axis leads to accumulation of P granules in the embryo posterior to mark germ cells. Three-dimensional tracking of fluorescently labeled P granule components revealed that upon symmetry breaking, P granules form spontaneously in the embryo posterior in the vicinity of polarity proteins. Thus, P granule formation is driven by local decreases in the saturating concentration of P granule components in the embryo posterior and an increased flux of P granule components into the embryo posterior. Furthermore, P granules exhibit classic liquid properties such as fusion, dripping and wetting. Together, these data strongly argued for liquid-liquid phase separation as a mechanism for subcellular cytoplasmic architecture.

14A small molecule screen for compounds to promote neuronal progenitor cell differentiation into mature neurons led to the serendipitous discovery that biotinylated 5-aryl-isoxazole-3-carboxyamide (b-isox) selectively precipitates known protein components of ribonucleoprotein (RNP) granules, including TDP-43 and FUS. In vitro, recombinant full-length FUS phase transitions to a hydrogel-like state, but very high protein concentrations are required. Furthermore, it was observed that hydrogels formed by the PrLD of FUS were capable of retaining soluble FUS PrLD, suggesting that hydro-gel assembly is mediated by homotypic interactions in the FUS PrLD. Electron microscopy of FUS PrLD hydrogels revealed a composition of amyloid-like fibrils. X-ray diffraction analysis showed prominent reflections at 4.6–4.7 Å and 10 Å typical of cross-β structure common to amyloid fibrils, again strongly suggesting an amyloid-like structural basis to hydrogel architecture. These hydrogels and fibrils were readily dissolved upon exposure to SDS or mild (37 °C) heating, suggesting that they were distinct and more dynamic than the SDS-resistant amyloid fibrils formed by yeast prions. This finding was consistent with previous work demonstrating that purified FUS spontaneously assembles into SDS-soluble fibrils. However, whether the hydrogels formed in vitro were reflective of RNP granules found in cells remained uncertain.

15Multiple reports have suggested that RNA granules exist as predominantly liquid compartments and not gel-like compartments in cells. For example, in cells and in vitro, FUS forms dynamic assemblies that are rapidly recruited to sites of DNA damage, and display physical characteristics of liquid droplets as predicted by classical physics of polymeric phase transitions, including fast internal dynamics, spherical morphology, and a propensity for two droplets to readily fuse when in close contact with one another. Importantly, the PrLD mediates the phase transition as deletion of the PrLD abrogates droplet assembly. Detailed study of RBPs hnRNPA1, Lsm4, Tia1, and Pub1 revealed similar behavior. Droplet formation was influenced by the presence of molecular crowding agents, ionic strength, and presence of RNA or other polyanions, such as poly-ADP ribose.

16The identity of the bound RNA tunes the biophysical properties of the RNP granule. In the filamentous fungus Ashbya gossypii, the RBP Whi3 possesses a PrLD that enables assembly into liquid droplets to organize cyclin transcripts (CLN3) at sites of nuclear division and formin transcripts (BNI1) at polarity sites where new branch sites are located. Microrheology studies revealed that Whi3 droplets bound to BNI1 are less viscous than CLN3-bound droplets. Additionally, BNI1 droplets fuse with one another faster than CLN3 droplets. This suggests that client RNA identity is critical to tuning the biophysical properties of RNP granules, and that different physical properties of RNP granules may be optimized for specific cellular functions. This latter point raises a provocative parallel with yeast prions, where conformational diversity gives rise to distinct prion strains with unique phenotypes. It is possible that bound RNAs may define ‘strains’ of RNP granules by tuning their biophysical properties.

17The case that biophysical properties of RNP granules reflect their functional role is perhaps most strikingly made in S. cerevisiae. Specifically, whereas P bodies are constitutively active sites involved in mRNA processing and degradation and exist as liquid droplets in yeast, stress granules are inactive storage sites for proteins and RNA that form rapidly upon onset of stress, and are solid, gel-like aggregates. Yeast rely on the protein disaggregase and hexameric AAA+ ATPase, Hsp104, to maintain the fluidity of P bodies. Metazoa lack an Hsp104 homolog, and curiously in mammalian cells P bodies and stress granules exist in more liquid-like states. Thus, the powerful disaggregase activity of Hsp104 may enable yeast cells to readily exploit solid, gel, and liquid states in RNP granules.

18What structure(s) do the PrLDs of RNA-binding proteins adopt in liquid droplets? Solution NMR study of the FUS PrLD in monodisperse solution and condensed into droplets revealed that the PrLD retains disordered character in droplets. This finding suggests a model in which these liquid droplets maintain rapid internal dynamics while being held together by transient intermolecular contacts between adjacent PrLDs.

19In an attempt to further address this question, a mass spectrometry-based chemical footprinting method has been employed in which N-acetylimidazole (NAI) is used to acetylate serine, tyrosine, lysine, threonine, arginine, and asparagine side chains in proteins. Using two model proteins, recombinant glutathione-S transferase (GST) and poly-ADP-ribose polymerase (PARP) isolated from HEK 293T cell nuclei, it was shown that solvent-accessible side chains (as assessed by available crystallographic data for these two proteins) are more readily acetylated. Thus, the acetylation pattern or “footprint” for a given protein can be used as a conformational proxy.

20This technique was then deployed to show that the PrLD of hnRNPA2 in droplets or polymerized into hydrogels in vitro, or hnRNPA2 isolated from nuclei adopt similar conformations, as assessed by their NAI footprints. This congruence might suggest that cross-β fibrillization underpins both phase transitions to liquid droplets and hydrogel formation. Curiously, however, hnRNPA2 PrLD fused to maltose binding protein (MBP) to maintain the hnRNPA2 PrLD in a soluble, monomeric state also displayed a footprint qualitatively similar to that obtained for hydrogels, liquid droplets, and fibrils. Therefore, it is difficult to interpret precisely how the hnRNPA2 chemical footprints relate to hnRNPA2 structure. One possibility is that the observed NAI footprint in the monomeric state may be due to contamination by small amounts of fibrils. This possibility is supported by the fact that the intensity of the footprint progressively increased with time after MBP cleavage, and suggests that the NAI footprinting method detects hnRNPA2 fibril abundance. Alternatively, a fraction of hnRNPA2 PrLDs might exhibit cross-β structure even in the context of monomeric, soluble hnRNPA2. Although the hnRNPA2 PrLD is predicted to be intrinsically disordered in the soluble, monomeric state, circular dichroism studies suggest that this domain may adopt β-sheet-rich structures in solution. Another possibility is that similar regions within the hnRNPA2 PrLD may be invariably solvent accessible (or inaccessible) in distinct structures for monomeric forms in solution, in liquid phases, and in cross-β fibrils, and consequently the NAI footprint does not resolve between them. We suggest that further structural studies using complementary techniques in addition to NAI foot-printing are needed to further resolve the structure of the hnRNPA2 PrLD in various soluble, liquid, and fibrillar states.

21Parker and colleagues observed that stress granules contain stable subcompartments that can be isolated in cell lysates, thus suggesting that stress granules are not simply liquids. Instead, they propose that stress granules contain stable, gel-like cores surrounded by a dynamic liquid shell. They demonstrate that the cores of stress granules isolated from yeast are larger than their counterparts from mammalian cells, thus reconciling their data with that of Alberti and colleagues. The existence of a stable, gel-like core within stress granules may even provide an explanation for the observation by McKnight and colleagues that nuclear hnRNPA2 has a similar chemical footprint to recombinant hnRNPA2 polymerized into hydrogels. However, it remains unclear whether cross-β polymerization is at the root of the stable stress granule core. Thus, the structure of PrLDs within RNP granules is likely to remain the subject of intense focus in future studies. Moreover, other membraneless organelles, such as the nucleolus, contain distinct subcompartments. It will be important to determine whether these are due to separated gel and liquid phases, or immiscible liquid phases with different viscosities.

22What is the connection between the functional role of PrLDs in beneficial phase transitions in the formation of membraneless organelles and the deleterious misfolding events that these domains undergo in neurodegenerative disease? Recent work in vitro demonstrates that liquid droplets composed of FUS and hnRNPA1 harboring disease-linked mutations in their PrLDs (FUSG156E and hnRNPA1D262V) mature to a solid, hydrogel-like state more rapidly than droplets formed by wild-type protein. Moreover, hydrogel-like forms of mutant FUS have been specifically associated with neurodegenerative phenotypes in a C. elegans model of FUS proteinopathy. Thus, a direct link emerges between the biophysical propensity of these proteins to adopt more solid-like structures and neurodegeneration.

23Interestingly, the majority of ALS-causing mutations in FUS are found not in the PrLD but in the C-terminal region, where they disrupt a proline-tyrosine (PY) nuclear localization signal. Mutations in the FUS PY-NLS lead to persistent cytoplasmic FUS mislocalization, which correlates with ALS severity, but curiously does not directly increase the biophysical propensity of FUS to aggregate. However, the biophysics of membraneless organelle assembly shed new light on this observation. Membraneless organelles such as nucleoli and P granules in Caenorhabditis elegans have the property that local concentration of granular components drive granule droplet condensation. Thus, a model emerges where impaired nuclear import of FUS leads to persistent cytoplasmic FUS droplets that mature to more intractable solid aggregates, in accordance with findings in vitro. Therefore, strategies to boost FUS nuclear import or maintain FUS droplet fluidity should represent a significant therapeutic opportunity for ALS and FTD.

24In S. cerevisiae, the protein-remodeling factor Hsp104 regulates the formation, elimination, and propagation of beneficial yeast prions (Sweeny and ). Hsp104 severs yeast prions to ensure their dissemination to daughter cells upon division, and is required to clear solid stress granules in yeast and to maintain the fluid integrity of yeast P bodies. Indeed, it is hypothesized that this robust disaggregase machinery coevolved with solid stress granules as a way for yeast to cope with their vulnerability to environmental fluctuation. However, metazoa lack a clear Hsp104 homolog, and fibrils formed from human RBPs with PrLDs represent an intractable substrate for wild-type Hsp104. Recently, however, engineered forms of Hsp104 have been generated that potently suppress the aggregation and toxicity of various disease-linked RBPs with PrLDs, including TDP-43, FUS, and TAF15. Moreover, these potentiated Hsp104 variants dissolve preformed TDP-43, FUS, and TAF15 fibrils in vitro. Thus, Hsp104 could represent a disruptive technology to enhance metazoan proteostasis to counter RBP misfolding that causes neurodegenerative disease such as ALS and FTD.

25In metazoan cells and yeast, the Hsp110, Hsp70, and Hsp40 protein-disaggregase machinery, contributes to the clearance of stress granules. Hsp70 and Hsp40 chaperones often get sequestered and inactivated by misfolded protein aggregates. Thus, enhancement or engineering of this disaggregase machinery might also open potential therapeutic avenues for ALS, FTD, and a variety of other neurodegenerative disorders.

26Deletion of several autophagy-related genes and Cdc48 adaptor proteins gives rise to constitutive stress granules in S. cerevisiae. Cdc48, or valosin-containing protein (VCP) in humans, is another hexameric AAAþ ATPase that can promote autophagy, and mutations in VCP cause familial forms of ALS and MSP. These diseases are characterized by the formation of cytoplasmic protein aggregates containing RBP components of stress granules. Deletion of the autophagy gene ATG7 or siRNA-mediated knockdown of VCP leads to impaired ability to clear stress granules in mammalian cells, and disease-causing mutations in VCP cause accumulation of constitutive stress granules that contain TDP-43. However, it remains unclear whether the sole role of Cdc48/VCP is to target stress granule components for autophagic degradation or whether Cdc48/VCP may have also have an active role in disaggregation of stress granules and reactivation of their components. Indeed, an exciting possibility is that Cdc48/VCP may represent a triage center for stress granule components, effecting the reactivation of salvageable components and the degradation of others.

27Here, we have reviewed recent advances in our understanding of prion-like phenomena and architecture across biology, from epigenetic regulation in the simple model organism S. cerevisiae to complex mechanisms of eukaryotic subcellular organization. We suggest that PrDs and PrLDs may have been biologically conserved for their wide-ranging biological utility. The ability of prions to rapidly switch between distinct conformational and functional states confers selective advantages for yeast in the face of environmental stress. In fungi and metazoa, PrLDs now have a clear functional role in mediating the reversible coalescence of RNP granules. However, in humans the persistence and maturation of these RNP granules via complex mechanisms leads to pathological protein accumulation and neurodegenerative disease (Fig. 2). We suggest that PrLDs may be general scaffolds for membraneless subcellular organization. However, this activity places PrLDs at risk to accessing deleterious misfolding trajectories that cause neurodegenerative disease. PrLDs have a very distinctive amino acid composition, but this role in subcellular compartmentalization via phase transitions and simultaneous risk of protein misfolding may extend to other intrinsically unfolded, low complexity domains with different amino acid composition. Regardless, we suggest that deleterious misfolding events can be reversed by select protein disaggregases, which could have important therapeutic applications.