Beyond aggregation: pathological phase transitions in neurodegenerative disease

Abstract

Over the past decade, phase transitions have emerged as a fundamental mechanism of cellular organization. In parallel, a wealth of evidence has accrued indicating that aberrations in phase transitions are early events in the pathogenesis of several neurodegenerative diseases. Here we review the key evidence of defects at multiple levels, from phase transition of individual proteins to the dynamic behavior of complex, multicomponent condensates in neurodegeneration. We also highlight two concepts, dynamical arrest and heterotypic buffering, that are key to understanding the relationship between pathological phase transitions and pleiotropic defects in cellular functions and the accrual of proteinaceous deposits at end stage disease. These insights not only inform upon disease etiology, but are also likely to guide the development of therapeutic interventions to restore homeostasis.

Proteinaceous deposits in neuronal tissues have long been recognized as a hallmark of late-onset neurodegenerative diseases. Over the last 20 years, the dominant paradigm to relate protein deposits to cellular demise has centered on the concept of aggregation. Mechanistically, protein deposits are thought to impose neomorphic, toxic gains of function intrinsic to the deposited proteins, and much research in the last two decades has been centered on the physicochemical nature and aggregation state of the presumed toxic species (e.g., monomers vs. oligomers vs. fibrils). A wealth of evidence has accrued in recent years leading to an evolution in the aggregation paradigm into a deeper and more direct understanding of how these protein deposits arise and relate to cellular dysfunction and death; namely, via pathological phase transitions. Among this additional evidence is a deeper understanding of how cells are organized by phase transitions and how they govern vital biological processes. In particular, recognition that disease proteins participate in these phase transitions, and genetic insights demonstrating that disease-causing mutations promote this process, has invoked the concept of pathological phase transitions. These advances have implications not only for understanding gain and/or loss of function of proteins associated with disease, but also for the biological function of the condensates (Box 1) to which they belong. Here we review these lines of evidence and synthesize these correlations into a framework that seeks to draw a direct line from genetics and biophysics to molecular mechanisms of disease and pathology. This framework is important not only in defining disease etiology, but is also likely to guide the development of therapeutic interventions to restore homeostasis.

Box 1: Glossary of terms.

Condensate: A generic term used to refer to membraneless cellular structures that concentrate biomolecules. These structures can form through reversible phase transitions.

Phase transition: Refers to abrupt, highly cooperative changes to order parameters caused by the breaking of symmetries that in turn lead to changes in the state of matter.

Symmetry: Symmetry refers to the invariance of a physical system to operations such as translations of molecules along, or rotations about, defined axes. Disorder is characterized by a state of high symmetry and disorder-to-order transitions occur by the breaking of specific symmetries.

Phase separation: A type of phase transition in which a system separates into one or more coexisting phases. In a binary mixture, the coexisting phases are dense and dilute phases. If the dense and dilute phases that result from phase separation are liquids, then the transition is referred to as liquid-liquid phase separation (LLPS).

Percolation: Multivalent macromolecules behave like associative polymers that form reversible, non-covalent crosslinks. When the number of crosslinks crosses a threshold known as the percolation threshold, a majority of the molecules are incorporated into a single system-spanning (percolated) network.

Material properties: With respect to biomolecular condensates, this refers to features such as viscosity, elasticity, and surface tension of the dense phase. These features are manifestations of elastic and dissipative moduli that are governed by the extent of physical crosslinking and the timescales for making and breaking crosslinks within condensates. These material properties influence the spatial organization and diffusion of macromolecules within the dense phase, as well as selective permeability to molecules entering the condensate. These material properties are tightly regulated and directly linked to condensate function.

Dynamical arrest: This refers to the situation in which the efficiency for making and breaking of crosslinks within a condensate is reduced to an extent that dynamism is lost. A network of physical crosslinks can trap macromolecules in arrested phases characterized by irregular (aspherical) morphologies and immobile molecules. When crosslinks are made and broken efficiently, the resulting condensate can have liquid properties. Excessive crosslinking or reduced efficiency in the breaking of crosslinks, such as accompanies many disease mutations, results in arrested dynamism, altered material properties, and impaired function.

Heterotypic buffering: This term describes the ability of heterotypic interactions to buffer against the deleterious effects of homotypic interactions that can drive pathological liquid-to-solid phase transitions. It also refers to the positive effects of heterotypic interactions that suppress dynamical arrests.

Disease Proteins Undergo Phase Transition via Homotypic Interactions

A phase transition (Box 1) is a sharp change to one or more physical properties of a physico-chemical system. In macromolecular solutions, the relevant physical properties are termed symmetries (Box 1). A disordered system has high symmetry in that measurable properties of the system are invariant to molecular translations, rotations, vibrations, density fluctuations, and changes in conformation. Thus, in general terms, phase transitions represent the breaking of one or more symmetries. Phase transitions abound in nature and are particularly important in cell biology.

A system comprising macromolecules plus solvent can undergo a particular type of phase transition known as phase separation (Box 1). Beyond a system-specific threshold concentration, the macromolecular solution can separate into two phases: a dense phase enriched in macromolecules that coexists with a dilute phase that is deficient in macromolecules. Of particular relevance in biological systems are liquid-liquid phase separation (LLPS) and liquid-to-solid phase transitions. In LLPS, macromolecules separate from solution to form a dense liquid phase. Pathological fibrils that are found in late-stage neurodegeneration form via liquid-to-solid phase transitions, where the precursor liquid phase can either be the dilute solution of macromolecules or the dense phase formed via LLPS.

Many proteins associated with neurodegeneration, including tau, α-synuclein, TDP-43, hnRNPA1, TIA1, and FUS, are capable of fibril formation in vitro (1–7). Although this process is often referred to as aggregation, phase transitions and aggregation are distinct phenomena: whereas fibril formation is mediated by homotypic interactions and governed by the principles of phase transitions, aggregation refers to the sticking of molecules to one another, unconstrained by concentration thresholds or accompanied by symmetry breaking. This distinction between pathological phase transition and aggregation is more than academic because it is informative with respect to how such fibrils may arise in a pathological context, the cellular processes that may be disturbed by pathological phase transitions, and how such pathology may be reversed.

Beginning in 2015, it was appreciated that many neurodegenerative disease-related proteins not only assemble into fibrillar solids, but also undergo LLPS to form liquid droplets. First illustrated for hnRNPA1 (8, 9), TDP-43 (8), and FUS (10, 11), it was determined that many disease proteins exhibit distinct concentration thresholds that correspond to the onset of two types of phase transitions: one threshold for LLPS, and a higher threshold for liquid-to-solid phase transition (8, 10). It was also shown that the liquid-to-solid phase transition can be enhanced within the liquid phase (8, 10). Similar observations have been made for tau (12), α-synuclein (13), huntingtin (14), and TIA1 (7). These observations have highlighted two distinct routes to fibril formation (Figure 1). Fibril formation may be initiated by a combination of primary and secondary nucleation (15) in dilute solution. Alternatively, fibril formation may occur via liquid-to-solid phase transition within the dense liquid phase. In the latter case, the condensed liquid state that arises from LLPS facilitates fibril formation by concentrating proteins and enabling the crossing of the threshold concentration for liquid-to-solid phase transition. These routes are not mutually exclusive and might differ for different proteins and different contexts (e.g., in vitro vs. in living cells).

Fig. 1. Two non-exclusive routes lead to fibril formation in neurodegenerative diseases.

Fibril formation may be initiated by a primary and secondary nucleation in dilute solution with subsequent growth through templating additional units. Alternatively, fibril formation may occur via liquid-to-solid phase transition within the dense liquid phase. In the condensed liquid state, fibril formation is facilitated by concentrating proteins and bringing them closer to the threshold for liquid-to-solid phase transition. These routes are not mutually exclusive and may be influenced by context.

Neurodegenerative disease-causing mutations in the low complexity domains (LCDs) of hnRNPA1 (6), hnRNPA2B1 (6), TDP-43 (16), FUS (10, 11), and TIA1 (7) are known to reduce the concentration threshold for LLPS. These mutations can also reduce the threshold for liquid-to-solid phase transitions within dense liquid phases, giving rise to pathological phase transitions. Likewise, it was recently shown that disease-causing mutations in α-synuclein also alter the threshold for liquid-to-solid phase transition (13). Importantly, the biophysical properties of proteins can be regulated by local changes in the cellular milieu (e.g., pH) or chemical modifications. Indeed, a number of post-translational modifications associated with disease in these proteins also reduce the threshold for phase transitions driven by homotypic interactions that promote fibril formation (12, 13, 17).

Neurodegenerative Disease Proteins Are Constituents of Complex Condensates

Whereas it is evident that a common feature of purified disease proteins is their ability to undergo phase transitions driven by homotypic interactions, and that such phase transitions are promoted by disease mutations, the situation is far more complex in living cells. Moreover, degeneration of neurons cannot be explained solely by fibril formation by a disease protein. Rather, understanding the pathogenesis of neurodegeneration requires consideration of underlying cellular processes that are corrupted over time. Notably, many proteins associated with neurodegeneration reside primarily within cellular bodies known as biomolecular condensates that assemble via phase separation and encompass hundreds of distinct protein and nucleic acid components (13, 18–22) (Figure 2). Biomolecular condensates are distinct from simple liquid droplets or solids formed via phase transitions mediated by homotypic interactions of specific proteins. Instead, they form and dissolve via reversible phase transitions of multicomponent systems that are controlled by dynamic networks of homotypic and heterotypic interactions (23).

Fig. 2. Condensates arise through a network of heterotypic and homotypic interactions.

Condensates form through phase separation of multiple types of macromolecules. In RNP granules, for example, multivalent proteins and RNA molecules participate in a network of homotypic and heterotypic interactions that collectively determine the concentration threshold for LLPS and the material properties of the resulting condensate. The material properties of biomolecular condensates, such as viscosity, elasticity, and surface tension of the dense phase are governed by the extent of physical crosslinking and the timescales for making and breaking crosslinks within condensates. These material properties influence the spatial organization and diffusion of macromolecules within the dense phase, as well as selective permeability to molecules entering the condensate. These material properties are tightly regulated and directly linked to condensate function.

In cells, biomolecular condensation provides spatial and temporal control over cellular components and biochemical reactions (24). A plethora of condensates are found in cells spanning a vast range of sizes and compositions. For example, the central channels of nuclear pores are small condensates composed of a few different biomolecules, whereas ribonucleoprotein (RNP) granules are large, complex condensates containing hundreds of distinct biomolecules that function as discrete membraneless organelles. Indeed, a cell may be viewed as a complex, dynamic network of condensates that are in constant communication through exchange of materials. Biomolecular condensates provide advantages over membrane-mediated compartments in that they concentrate macromolecules in space while enabling rapid exchange of constituents with the surrounding intracellular milieu (25). Moreover, many condensates can assemble or disassemble rapidly in response to changes in cellular state. The functional consequences of this dynamic organization include positive and negative regulation of biochemical processes. For example, condensation of multiple enzymes in a common pathway can promote “substrate channeling” wherein the intermediary metabolic product of one enzyme is passed directly to another enzyme without its release into solution, thereby increasing overall efficiency of the pathway (26). Such a mechanism may underlie the regulation of multiple glycolytic enzymes at neuronal synapses in an activity-dependent manner (27). Condensate formation via phase separation may also have the opposite effect, wherein sequestering one or more constituents in the dense phase may negatively regulate biological activities in the dilute phase (28). Condensates can also orchestrate the assembly of complex higher-order structures, such as ribosome subunit assembly in the nucleolus (29).

Dynamism in Complex Biomolecular Condensates

The dynamic behavior of condensates reflects the nature of the interactions that underlie their assembly – namely, weak, transient interactions among multivalent biomolecules that form non-covalent crosslinks of varying strengths and durations. Above a system-specific threshold, the system becomes populated with sufficient crosslinks to form a system-spanning network that holds the condensate together, a phenomenon known as percolation (Box 1). Thus, unlike phase transitions that are driven purely by homotypic interactions, condensate formation requires the crossing of a collective threshold defined by condensate-specific networks of homotypic and heterotypic interactions (23). The macromolecular partners engaging in physical crosslinks will evolve dynamically, and if such crosslinks are made and broken efficiently, the condensate can be highly dynamic and exhibit liquid properties (30). The extent of networking and the timescales for making and breaking crosslinks contribute directly to the material properties (Box 1) of a condensate. These include properties such as viscosity, elasticity, and surface tension. These material properties are tightly regulated because they influence the spatial organization and diffusion of macromolecules within the dense phase, selective permeability to molecules entering the dense phase, exchange of constituents with the light phase, and ultimately condensate function.

Many proteins associated with neurodegenerative diseases reside within distinct biomolecular condensates. For example, disease-related RNA-binding proteins such as TDP-43, FUS, hnRNPA1, hnRNPA2B1, and TIA1 are constituents of multiple types of RNP condensates that control the fate of RNA molecules as they transit through processes such as splicing, nuclear export, trafficking in the cytoplasm, translation, and degradation (31). Importantly, disease mutations alter the balance of homotypic and heterotypic interactions in these RNP assemblies, thereby changing their material properties, even in the absence of pathological liquid-to-solid phase transitions (32). Indeed, a recurrent observation is that disease-causing mutations lead to dynamical arrest (Box 1) of RNP granules that can impair functions with adverse consequences for RNA metabolism (6, 7, 10–13, 33). At the same time, concentration of proteins in the dense liquid phase increases the probability of a liquid-to-solid phase transition – particularly in proteins harboring prion-like low complexity domains. However, the rarity of this pathological event points to the presence of mechanisms that hold such pathological transitions in check. The primary factor suppressing potentially deleterious, excess homotypic interactions is the collective effects of functional networks of heterotypic interactions within condensates, an effect we term heterotypic buffering (Box 1). Additional checks, most notably the activities of chaperones, are in place to reverse any excess homotypic interactions that escape heterotypic buffering. The concept of heterotypic buffering is particularly useful as a framework to understand sporadic neurodegenerative disease, which culminates in the same pathology as disease arising from rare genetic mutations. According to this view, a variety of insults may intersect at a common point, collectively altering the dynamic network of condensate interactions in such a way as to impair heterotypic buffering, leading to pathological liquid-to-solid transitions and / or dynamical arrest (Figure 3).

Fig. 3. Disruption of the physiologically relevant interplay between homotypic and heterotypic interaction can lead to pathological phase transitions.

Within multicomponent condensates, homotypic interactions are typically buffered by an abundance of heterotypic interactions. This buffering guards against pathological homotypic interactions that may give rise to liquid-to-solid phase transition and fibril formation. In this simplified example, two types of macromolecules are illustrated, each depicted with two interacting domains (rectangles) connected by a spacer region (black line). Purple macromolecules can form homotypic interactions with other purple macromolecules or heterotypic interactions with blue macromolecules. Blue macromolecules can form heterotypic interactions with purple macromolecules. Under normal conditions (left), the concentration of blue macromolecules is such that heterotypic interactions buffer any homotypic interactions. Four scenarios are shown in which this buffering is disrupted: (1–2) imbalances in the relative concentrations of condensate constituents, (3) competition for heterotypic interactions by the presence of an additional molecule, and (4) pathological mutations that favor homotypic interactions. In each case, excessive homotypic interactions can lead to the deposition of solid-like structures within condensates if they escape protein quality control mechanisms.

Defining the Relationship Between Pathological Phase Transitions and Disease

Understanding the process of neurodegeneration requires consideration not only of how end-stage proteinaceous deposits arise, but how specific cellular processes are corrupted over time to give rise to neuronal dysfunction and death. Such disturbances can be considered from two perspectives: first, a mode of toxicity in which pathological consequences arise directly from unchecked homotypic phase transition leading to fibril formation; and second, a mode of toxicity in which pathological phase transitions have broader functional consequences by altering key properties of biomolecular condensates, thereby impairing the ability of condensates to regulate biological activities.

With regard to the first mode of toxicity, the accumulation of disease proteins in fibrillar deposits may directly impair cellular function. One possible mechanism is that pathological protein deposits create a sink that depletes cells of that particular protein. For example, accumulation of fibrillar TDP-43 or FUS pathology in the cytoplasm is accompanied by gradual depletion from nuclei (34) and this nuclear depletion may lead to a partial loss of function (35, 36). Pathological proteinaceous deposits may also sequester additional factors leading to their functional depletion. For example, it has been suggested that recruitment of chaperones to such pathology may deplete the capacity of protein quality control mechanisms, with widespread implications (37).

Alternatively, pathological phase transitions may exert broad cellular toxicity by influencing the network of interactions that define the nature and function of biomolecular condensates. A prominent example is ALS-FTD, which arises from pathological variants in at least seven different RNA-binding proteins, including TDP-43 (38), FUS (39, 40), hnRNPA1 (6), hnRNPA2 (6), TIA1 (7, 41), matrin 3 (42), ataxin 2 (43), and annexin A11 (44). All of these proteins reside within biomolecular condensates that are distributed throughout the nucleus and cytoplasm of cells and govern many aspects of RNA metabolism. Disease-associated mutations in these RNA-binding proteins alter the material properties of their native condensates, and it is therefore unsurprising that ALS-FTD is associated with widespread disturbance of RNA metabolism (45). It remains to be determined whether disease-causing mutations in other condensate-resident proteins (e.g., tau and α-synuclein) also impair functions of specific condensates. Beyond impairing function, perturbation of condensate material properties can simultaneously enhance the driving forces for additional symmetry-breaking operations, notably liquid-to-solid phase transition. As a result, dynamically arrested condensates can also become crucibles driving the formation of fibrillar deposits arising from homotypic interactions, as described above (Figure 1). Thus, disturbance of the material properties of condensates can drive pathology through two consequences that are inextricably linked: impairment of native condensate function and production of proteinaceous pathology.

Remarkably, disease-causing mutations in a variety of proteins that are not typically thought of as constituents of condensates can also cause dynamical arrest. These include proteins involved in the maintenance and clearance of condensates, such as VCP (46–48), UBQLN2 (49, 50), and OPTN (51, 52). Furthermore, pathological poly-dipeptides arising from expanded C9ORF72 produce widespread disturbances in biomolecular condensate function. Specifically, arginine-containing poly-dipeptides (polyGR and polyPR) become concentrated within biomolecular condensates and alter their material properties through extensive interactions with LCDs (53). For example, C9ORF72-related poly-dipeptides impair the dynamics and material properties of nucleoli resulting in reduced ribosome biogenesis (53–55) and impaired ability of the nucleolus to buffer against nuclear protein misfolding (56), impair the central channel of the nuclear pore (57), and disturb the dynamics of stress granules and RNA transport granules (53, 58).

The mechanisms described above focus on pathological phase transitions of proteins. However, we note that RNA molecules are also well-suited to driving phase transitions. Indeed, pathological expansion of RNA repeats, such as those observed in myotonic dystrophy types 1 and 2 and C9ORF72-related ALS-FTD, is marked by pathological RNA phase transition resulting in RNA foci (59). These RNA-driven pathological phase transitions can also initiate toxicity via mechanisms that are remarkably similar to those observed with protein deposits, including sequestration of RNAs and RNA-binding proteins (60).

Perspectives

A decade’s worth of research has driven an evolution of the commonly held perspective that neurodegeneration is caused by aggregation of misfolded, toxic proteins towards the view that a common underlying principle in neurodegeneration is pathological phase transitions. Appreciating this insight is important because it focuses our attention squarely on the dynamic cellular condensates that are assembled from these proteins. Pathological phase transitions of disease proteins, irrespective of which route they take to fibril formation, are inextricably linked to the functions of the condensates in which they reside. According to this view, the primary manifestations of cellular dysfunction in the context of disease are two-fold: (1) altered material properties due to dynamical arrest of condensates and (2) pathological liquid-to-solid transitions. Accordingly, reversing these defects should be the objective of therapeutic intervention.

As described above, the percolation threshold and material properties of condensates are defined by the dynamic network of homotypic and heterotypic interactions that define them. Indeed, it is now evident that manipulation of individual constituents is sufficient to alter the percolation threshold and material properties of specific condensates. Thus, one promising approach to restoring material properties to condensates is to target individual constituents that are deemed druggable by conventional criteria, with the desired effect of shifting the percolation threshold and material properties of a specific condensate in the desired direction. For example, in the setting of ALS there is evidence of altered material properties and functions of a variety of RNP granules, such as stress granules and RNA transport granules. It has been proposed that depletion of ataxin-2 ameliorates neurodegeneration in a mouse model of ALS through precisely this mechanism, resulting in decreased TDP-43 pathology (61). Another attractive target to restore normal material properties in this way may be G3BP, which serves as a more central node in the assembly of these condensates. Indeed, knockdown or inhibition of G3BP enhances local translation in neuronal processes, protects against axonal injury, and promotes axonal regeneration (62). Beyond targeting the network of condensate constituents to influence material properties, an alternative approach might be to target key pathological phase transitions themselves, such as assembly of pathological TDP-43 or FUS fibrils. In this strategy, one might exploit the chaperones that specifically target these assemblies. Specifically, it has recently been shown that nuclear import receptors (karyopherins) function as chaperones for and can reverse pathological phase transitions of their clients. Indeed, the activity of KapB2 reverses pathological phase transitions by proteins harboring the cognate PY-NLS (i.e., FUS, hnRNPA1, hnRNPA2), whereas the activity of KapB1 together with importin-α reverses pathological phase transitions of TDP-43 in vitro and in vivo (63). Thus, strategies to augment the activity of these chaperones may be beneficial in a disease setting. Related to this, VCP serves as an important segregase in the dismantling of stress granules and perhaps other condensates. Indeed, it was recently shown that a small molecule agonist of ULK1/2 kinases, which phosphorylate and activate VCP, accelerates stress granule disassembly (64). It is likely that our improved understanding of the causes of dynamical arrest, pathological liquid-to-solid transitions, and the loss of heterotypic buffering will pave the way for targeting functional restoration and regulation of condensates as potent therapeutic strategies.