Spatial control of irreversible protein aggregation.

Liquid cellular compartments form in the cyto- or nucleoplasm and can regulate aberrant protein aggregation. Yet, the mechanisms by which these compartments affect protein aggregation remain unknown. Here, we combine kinetic theory of protein aggregation and liquid-liquid phase separation to study the spatial control of irreversible protein aggregation in the presence of liquid compartments. We find that even for weak interactions aggregates strongly partition into the liquid compartment. Aggregate partitioning is caused by a positive feedback mechanism of aggregate nucleation and growth driven by a flux maintaining the phase equilibrium between the compartment and its surrounding. Our model establishes a link between specific aggregating systems and the physical conditions maximizing aggregate partitioning into the compartment. The underlying mechanism of aggregate partitioning could be used to confine cytotoxic protein aggregates inside droplet-like compartments but may also represent a common mechanism to spatially control irreversible chemical reactions in general.

Introduction

Spatial control within living cells is essential to many cellular activities, ranging from the local control of protein activity to the uptake of pathogens or the management of wastes (Alberts, 2017). Understanding the mechanisms underlying regulation of cell activities in space and time is key not only for biological function, but also in view of understanding and eventually controlling cellular dysfunction (Knowles et al., 2011; Knowles et al., 2014; Chiti and Dobson, 2006; Gitler et al., 2017; Michaels et al., 2018). The spatial organization of cellular activities is often associated with membrane-bound organelles that ensure permeation only for certain molecules of specific molecular structure (Neupert and Herrmann, 2007; Wiedemann and Pfanner, 2017; Dukanovic and Rapaport, 2011). Recently, new types of organelles have been discovered that do not possess a membrane. They are referred to as non-membrane-bound compartments and they share most hallmark properties with actual liquid-like droplets (Brangwynne et al., 2009; Brangwynne, 2013; Elbaum-Garfinkle et al., 2015; Zhu and Brangwynne, 2015; Banani et al., 2017). Unlike organelles surrounded by membranes, these non-membrane-bound compartments are formed by liquid-liquid phase separation. In many cases, this phase separation is driven by disfavoring interactions between the constituent molecules of the compartment and the surrounding cyto- or nucleoplasm (Hyman et al., 2014; Brangwynne et al., 2015). The partitioning of other intracellular molecules into such droplet-like compartments is then controlled by their relative interactions with the constituent molecules of the compartment.

These droplet-like compartments are ubiquitous inside living cells (Banani et al., 2017). For instance, they emerge prior to cell division (Brangwynne et al., 2009; Parker and Sheth, 2007), and form as a response to cellular stress (Patel et al., 2015; Malinovska et al., 2013; Molliex et al., 2015). They have been shown to enrich proteins (Hernández-Vega et al., 2017; Woodruff et al., 2017; Mateju et al., 2017) and genetic material (Parker and Sheth, 2007; Saha et al., 2016; Zhang et al., 2015) providing distinct environments for chemical reactions and biological function. The molecules hosted inside these compartments may even be protected against other agents from the cytoplasm (Franzmann et al., 2018) or face conditions facilitating their molecular repair (Mateju et al., 2017; Ganassi et al., 2016; Alberti et al., 2017; Alberti and Carra, 2018; Jain et al., 2016; Specht et al., 2011). In addition to these roles, recent evidence suggests that liquid cellular compartments could play an important role in regulating pathological protein aggregation (Alberti and Hyman, 2016; Shin and Brangwynne, 2017). An example is the irreversible assembly of amyloids into fibrillar aggregates, a process that is linked to a large variety of currently incurable diseases (Dobson, 2003; Knowles et al., 2014; Lashuel et al., 2002; Catalano et al., 2006; Benilova et al., 2012; Campioni et al., 2010), such as Alzheimer’s and Parkinson’s diseases, amyloidosis or type-II diabetes. As another example, a chaperone in yeast uses a prion-like, intrinsically disordered domain to bind and sequester misfolded proteins in protein deposition sites (Grousl et al., 2018; Boczek and Alberti, 2018). Moreover, misfolded and pathological proteins can accumulate inside liquid-like stress granules triggering the aggregation kinetics inside these compartments. The presence of this phase separated compartment can promote the formation of fibrillar aggregates, and prevent aggregation outside the stress granules (Molliex et al., 2015; Mateju et al., 2017). Thus, the corresponding cytotoxic effects of protein aggregates are expected to be strongly localized in space as well. However, whether weak protein interactions are sufficient to significantly change the aggregate concentration in the compartment relative to homogeneous aggregation and how the physical parameters of aggregation and phase separation determine the partitioning of aggregates remains an open question.

Here, we combine the kinetics of irreversible protein aggregation with the theory of liquid-liquid phase separation to develop a model of irreversible assembly of protein fibrils in the presence of droplet-like compartments. We use this model to predict the partitioning of aggregates into the liquid compartment as a function of the fundamental physical parameters underlying aggregation kinetics and phase separation. We find that relatively weak interactions between the protein monomers and the liquid compartment molecules are sufficient to enrich the concentration of aggregates within the liquid compartment by several orders of magnitudes relative to homogeneous aggregation (Figure 1). This strong enrichment of aggregates emerges because the liquid compartment acts as continuous sink of monomers during the aggregation dynamics, thus promoting intra-compartment aggregation but suppressing aggregation outside of the compartment. Moreover, we find that aggregate partitioning is more pronounced for larger (smaller) compartments depending on the relative values of the reaction orders for primary and secondary nucleation. Our results suggest that cellular liquid compartments are ideal to control irreversible protein aggregation in space. In particular, the compartment volume, which is determined by the mean concentration of phase separated protein, represents a relevant control parameter for intra-compartment positioning of aggregate amount and size. The underlying physical mechanism might also be relevant in the context of spatial regulation of other irreversible chemical reactions where liquid compartments act as biomolecular microreactors.

Figure 1.

Partitioning of monomers and aggregates via liquid-like compartments.

Protein aggregation may occur homogeneously inside cells also leading to aggregates inside more sensitive cellular regions (left). A liquid compartment may accumulate monomers and thereby trigger the local formation of aggregates (right). The hardly diffusing aggregates are thus kept away from a more sensible cellular region. Such a spatial segregation of aggregates is ideal for adding functional, drug-like molecules which dominantly dissolve inside the compartment. These molecules may degrade the aggregates or inhibit further growth and nucleation. But most importantly, as these molecules are localized inside the compartment their toxic effects are diminished.

Model for liquid compartments controlling protein aggregation

To capture the interplay between liquid phase separation and protein aggregation kinetics we start with a model of two coexisting phases. One phase could be rich in proteins for example and coexist with a phase rich in another protein component, lipid, or water. Monomers that are prone to aggregate can partition differently into these phases. This partitioning is determined by the relative interactions between the majority components of each phase with the monomers. We consider the case where the partitioning of monomers is close to equilibrium during the kinetics of aggregation. This assumption is well justified since small, weakly interacting molecules such as the aggregating monomers diffuse between seconds and minutes through a cell of size in the order of tens of  (Brangwynne et al., 2009; Griffin et al., 2011), while typical time scales of aggregation in vitro are in the order of hours (see, for example, Cohen et al., 2013). Furthermore, the diffusion of aggregates is highly hindered as long fibrillar aggregates experience a much larger hydrodynamic drag force and can get entangled with cytoskeletal filaments and other assembled fibrils (de Gennes, 1971; Rubinstein, 1987). Finally, at large enough density and size, fibrils may even form solid-like gels (Mateju et al., 2017) further slowing down their mobility. All these effects imply that we may safely neglect diffusion of large aggregates and consider the typical case that monomers diffuse quickly relative to their aggregation kinetics.

We also consider the case where monomers and aggregates are dilute enough to neglect their influence on the composition of the two coexisting protein phases. Typical values of volume fractions for monomers of Amyloid-β, (radius of gyration in the range 1–2 nm [Sajfutdinow et al., 2018]), at physiological concentrations between 100 pM to 1 nM are in the range of 10-9 to 10-8. Time scale separation and dilute monomers together ensure that the compartment can coexist at thermodynamic equilibrium while the partitioning kinetics of monomers may weakly deviate from the partitioning equilibrium. Thus, we first discuss the partitioning of monomers into phase separated compartments at equilibrium and then consider small deviations from this equilibrium to understand its consequences for protein aggregation.

Phase separation and partitioning of monomers at equilibrium

We consider a system of total volume hosting a single liquid compartment (a droplet for example) of a condensed phase I of volume . The compartment itself forms by liquid-liquid phase separation between the two components and . Compartment I is composed of the component and a small fraction of component , while compartment II has a small amount of and a large amount of , as depicted in Figure 1. Each compartment creates a distinct environment for the aggregating monomers.

For simplicity, we discuss the case of an incompressible system where the aggregating monomers ‘m’ and aggregates ‘a’ are dilute, that is and , with and denoting the concentrations of monomers and aggregates and and are the respective molecular volumes. The assumption of dilute monomers and aggregates imply that for an incompressible system, the volume fractions and of the protein components and obey, , where we abbreviate in the following. As a result, the monomers may partition differently into the respective minority and majority phases, but, due to their dilute concentrations, they do not affect the degree of phase separation. Under these circumstances and in the absence of binding processes, the partitioning of monomers in the two phases is governed by the relative interaction strength between the monomers with the and the components, respectively. If is large and positive, monomers favor the presence of the majority component in compartment I. In this case, we expect a more pronounced partitioning of monomers into compartment I. Contrariwise, when is large and negative, monomers favorably partition into compartment II. The degree of monomer partitioning at equilibrium can be calculated using the condition that the chemical potentials of monomers associated with compartment I and II are balanced (see Figure 2(a), and Appendix 1 for the derivation), and allows us to define the monomer partitioningwhere , are the monomer concentrations in phases I and II, respectively, denotes the molecular volume of and molecules, and is the degree of phase separation of the -component. Then the relative partitioning of the total monomer concentration, , is given by the expressions and , where the partition degreecaptures the impact of the relative size of the compartment volume . The volume of the compartment I is in turn controlled by the mean volume fraction of molecules in the system in terms of the relationship , where we neglected the volume contribution of monomers and aggregates due to the considered dilute conditions. For finite sized compartments, the equilibrium volume fractions, and , are slightly increased due to the Laplace pressure. However, for compartments significantly exceeding the size of the molecules the relative increase is weak and is thus neglected in the following (see Appendix 1).

Figure 2.

Monomer partitioning and relative degree of segregation.

(a) The monomer partitioning (Equation 1) exponentially increases with the relative interaction strength (units of ) between the monomers and the and molecules which is defined in the Appendix. Its characteristic increase is set by the degree of phase separation, . Partitioning vanishes at the critical point of phase separation (solid line) and increases with the degree of phase separation (dashed line). Partitioning is largest for (dash-dotted line). Due to the exponential increase, large monomer partitioning can already be reached for weak relative interaction energies of a few . (b) The partition degree (Equation 2) describing the concentration fraction of monomers that resides in the minority phase II of the compartment, decreases with the mean volume fraction of material, , along with increasing compartment volume . Smaller compartments are thus better in enriching the monomer mass concentration.

Model for protein aggregation coupled to non-equilibrium monomer partitioning

Due to the separation of time scales of monomer diffusion and monomer aggregation, the partitioning of monomers into the compartment is close to equilibrium at all times of the aggregation kinetics and thus the relative fraction of monomers is approximately governed by the monomer partitioning , Equation 1. However, as the aggregation kinetics decreases the amount of monomers inside each phase, aggregation couples to the partitioning. This coupling is represented by a diffusive flux of monomer with a rate in each phase, that attempts to maintain the monomer partitioning close to equilibrium. In the limit of a sharp interface separating the liquid compartment from the bulk, there is no aggregation at the interface, . Furthermore, to linear order, the flux between the phases is proportional to the difference of monomer partitioning with respect the equilibrium value (see Appendix 2 for the derivation) and is of the form:where (with as monomer mass) is the monomer mass concentration in compartment , and denotes the rate at which monomer partitioning relaxes back to the equilibrium given by Equation 1. For simplicity, we consider the case where diffusion of monomers is constant and equal in each phase, and not affected by the aggregates.

Very generally, in a homogeneous solution, irreversible protein aggregation results from the combined action of several microscopic events, including (i) primary nucleation, whereby monomers spontaneously interact to form the smallest stable aggregate structures, (ii) fibril elongation, and (iii) secondary (i.e. aggregate-dependent) nucleation processes (Michaels and Knowles, 2014; Michaels et al., 2016; Arosio et al., 2016; Michaels et al., 2018; Törnquist et al., 2018). Secondary nucleation mechanisms (Törnquist et al., 2018) have been found to be active in many aggregating protein systems, ranging from prions to amyloidogenic proteins (Zhu et al., 2003; Kundel et al., 2018; Ruschak and Miranker, 2007; Meisl et al., 2014; Cohen et al., 2013); key examples of such secondary nucleation processes include fibril fragmentation, lateral branching and surface-catalyzed secondary nucleation.

In the presence of a liquid compartment, irreversible protein aggregation of fibrillar structures occurs within each phase as a consequence of both primary and secondary nucleation, and growth of aggregates via their ends, each event occurring with rate constants , , and (Michaels and Knowles, 2014; Michaels et al., 2016; Arosio et al., 2016; Michaels et al., 2018). We have seen that the key term in our model is the difference between the monomer concentration inside and outside of the compartment which leads to the diffusive flux of monomers between the phases (Equation 3a), which connects the effects of phase separation and protein aggregation. The coupled equations describing protein aggregation kinetics in both phases can be written as

Here, Equation (3b) describes the rate of formation of new fibrils in each compartment () through primary nucleation, fragmentation or surface catalyzed secondary nucleation. In the case of primary nucleation, the rate of formation of new aggregates depends solely on the concentration of monomers, where the reaction order describes the concentration dependence of nucleation. For secondary processes, including fragmentation and surface-catalyzed secondary nucleation, the rate of formation of new aggregates is proportional to the aggregate mass concentration; the dependence of the rate on the monomer concentration is described by the reaction order (the case corresponding to fragmentation). Note that both primary and secondary nucleation of aggregates are non-classical, multi-step nucleation processes; hence, the reaction orders and do not necessarily correlate to the physical size of nuclei (Šarić et al., 2016). Equation (3c) captures the build-up of aggregate mass within each compartment due to elongation of existing aggregates, which occurs by monomer addition at their ends. Finally, Equation (3d) models the population balance of monomers in each compartment as a result of two effects: (i) monomer depletion due to aggregate growth (see Equation (3c)) and (ii) the monomer flux between compartments I and II; this flux is given by Equation (3a) and ensures that partitioning is maintained close to the monomer partitioning factor .

While the monomer partitioning factor (Equation 1) governs the constant ratio of the time dependent concentrations in compartment I and II, the partitioning degree (Equation 2) determines how the total monomer concentration, which decays over time as a result of aggregation, is split between the two compartments at any time point during the kinetics of aggregation. As we will see, both parameters will be crucial in controlling the degree of aggregate partitioning into the compartments.

Irreversible aggregation in the presence of phase separated compartments

To understand how protein aggregation kinetics couples to the two phase separated compartments in terms of the physical parameters and , we constructed explicit analytical solutions to the set of non-linear kinetic Equation (3) by exploiting an analogy to classical mechanics (Michaels et al., 2016 and Appendix B for details of the calculations), and compared these with numerical solutions of (Equation 3).

Monomer partitioning affects nucleation and growth of aggregates between the compartments

In the limit of fast monomer diffusion, the aggregation kinetics in each compartment is controlled by a set of effective rate parameters. The relative magnitude of these effective rates between compartment I and II at early times scales with the monomer partitioning as , while at late times, the corresponding ratio of these rates scales with (see Appendix 3, Equation (39) and Equation (40)). Thus, the aggregate growth inside compartment I is faster than in compartment II if there is enrichment of monomers in the condensed phase (). Moreover, the relative magnitudes of growth rate at early times solely depends on the reaction order of primary nucleation, , while at late times, relative growth is determined by the reaction order of secondary nucleation, .

Phase separated compartments mediate a positive feedback for aggregate growth

This difference in growth rates between the phases can be qualitatively explained by the rapid preference of monomers to recover phase equilibrium (Figure 3(a)). The enhanced monomer concentration in compartment I causes aggregates to nucleate first inside compartment I. As a consequence, elongation of aggregates is more pronounced inside compartment I leading to a stronger consumption of monomers. This difference in monomer consumption between the compartments couples to the flux (Equation 3), which forces more monomers to diffuse into compartment I to maintain partitioning equilibrium, even as aggregates grow. This positive feedback mechanism in compartment I is accompanied by negative feedback for compartment II, which continuously loses monomers leading to a slowing down of the aggregation kinetics outside. Thus, the coupling between the aggregation kinetics and phase separation, mediated by diffusion of monomers (Equation 3), is key to determine aggregate enrichment/depletion in each phase.

Figure 3.

Segregation of aggregates into compartment I via positive feedback mediated by phase separation.

(a) Sketch of aggregation kinetics inside the two compartments I and II. Left: Initially, monomers get enriched on a short diffusive time scales due to the partitioning mediated by the phase separated compartments (Equation 1). Center: Monomers slowly aggregate. More aggregates nucleate and grow in compartment I due to the initial partitioning of monomers. This pronounced, initial aggregation causes a continuous monomer flux into compartment I, further promoting aggregation (positive feedback indicated by arrows). Right: Partitioning of monomers together with the positive feedback can cause a very pronounced accumulation of aggregates relative to compartment II. (b) Aggregate concentration as a function of time obtained from solving numerically and analytically Equation 3 actually confirms that aggregates can enrich by several orders of magnitude. (c) The asymptotic concentrations and inside each of the compartment inversely scale for small compartments, while for large compartment I, aggregate enrichment therein vanishes while depletion inside compartment II is dominated by primary nucleation. The asymptotic concentration in the absence of monomer partitioning, , is denoted as . Dashed line are the scalings given in the the main text. Parameters: . (d) Partitioning factor of aggregates inside compartment I as a function of monomer partitioning can reach very large values. The behavior switches from secondary nucleation dominated increase at small compartment I volumes to primary dominated growth at large volumes. Dashed line are the scalings given in Equation (6). (e) The slope of the partitioning factor as a function of mean volume fraction , equivalently speaking, volume of compartment I, changes its sign when partitioning is dominated by primary () or secondary nucleation (). Parameters: (b,e) consistent with weak interactions.

Positive feedback for aggregate growth causes strong aggregate partitioning

To understand this feedback mechanism, we study the time evolution of the aggregate concentration inside each phase, and (Figure 3(b)). The first aggregates are initiated by primary nucleation and solely determined by the monomer concentration. Because monomer concentrations in the compartments are slaved due to the rapid flux that maintains partitioning equilibrium, the time evolution of the aggregate concentrations in the early regime of the aggregation kinetics are slaved as well, following . When aggregates start consuming monomers via elongation, the flux of monomers from compartment II to I causes a saturation of the aggregate concentration outside the compartment II, while the concentration of aggregates in compartment I increases significantly. This rapid increase of growth is facilitated by the continuous influx of monomers (positive feedback). As monomers get depleted in the entire system the growth of aggregates also saturates in compartment I. Most importantly, the resulting asymptotic concentrations at large time scales, and , can differ by several orders of magnitude, even for modest values of corresponding to weak relative interactions.

Enrichment and depletion relative to homogeneous aggregation is determined by the reaction orders

To elucidate the impact of the reaction orders on the aggregation kinetics, we first consider the enrichment of aggregates relative to the case of homogeneous aggregation, that is for . For large values of monomer partitioning, the asymptotic concentrations in compartments I and II at large times relative to the homogeneous aggregate concentration at large times readwhere is a dimensionless numerical prefactor (Appendix 3, Equation 52). We see that for a large monomer partitioning factor , the partitioning of aggregates inside compartment I gets more pronounced, while aggregates in compartment II are more depleted relative to the homogeneous case (Figure 3(c)). Most importantly, the value of the terminal values of aggregate concentrations for given monomer partitioning factor are controlled by the reaction orders for primary and secondary nucleation, and . The role of and results directly from the interplay between aggregate growth and nucleation and their dependence on the monomer concentration.

Aggregate concentration in the compartments is controlled by compartment volume

Having understood the role of the monomer partitioning factor in aggregation kinetics, we now turn to how the asymptotic concentrations of aggregates in each compartment depend on the volume of the compartments. The dependence on compartment volume is given the partition degree . From Equation 2, we see that the partition degree , where the value of one is relevant for small compartments (Figure 2(b)). Following Equations (4) and (5), we see that for a small volume of compartment I, enrichment and depletion exhibit an inverse scaling, i.e. , which is solely dependent on the reaction order for secondary nucleation. Contrariwise, when the volume of compartment I is large, enrichment of aggregates inside I vanishes, while depletion inside compartment II then solely depends on the reaction order for primary nucleation, .

This switch between aggregate partitioning governed by secondary nucleation, to a partitioning solely determined by primary nucleation, arises from primary nucleation events occurring first inside compartment I due to a higher monomer concentration (). Once the first aggregates have formed via primary nucleation inside compartment I, small and large compartments behave fundamentally differently. If compartment I is small, only a few aggregates can form via primary nucleation due to the small compartment size. As aggregates begin to grow earlier in compartment I, the unbalance of monomers causes a flux from II to I. As a consequence of this continuous flux, the secondary nucleation events quickly overwhelm primary nucleation events inside compartment I, while secondary nucleation is suppressed in compartment II. However, if compartment I is large, the aggregation kinetics is similar to that for a homogeneous system because the monomer mass concentration is very close to the total monomer mass in the system and there is only a negligible amount of monomers entering from compartment II. Additionally, in the smaller compartment II where aggregates grow via primary nucleation, the coupling flux continuously removes monomers suppressing primary nucleation. Since compartment I is large, it shows little or no enrichment of aggregates relative to the homogeneous case while inside the small compartment II, aggregates are depleted determined by the lack of primary nucleation events relative to the homogeneous case.

Changes in compartment volume switch the driving mechanism for aggregate partitioning

To quantify the switch in aggregate partitioning as a function of compartment volume, we define the asymptotic aggregate partitioning ratio

As the compartment volume enters the partitioning factor solely via the partition degree , the sign of determines whether larger or smaller compartments lead to a larger partitioning (Figure 3(e)). Indeed, we find that the slope of the partitioning factor scales as . Thus, for , increasing the compartment volume by increasing the amount of -material causes a larger relative partitioning. Conversely, for , larger partitioning can be found for smaller compartment sizes. Consistently, if the nucleation coefficients obey , compartment volume has no impact on the partitioning factor .

This qualitative switch in the mechanism for aggregate partitioning raises the question which systems favor large or small compartment volumes in order to maximize aggregate partitioning . Figure 4 depicts the regimes in terms of the reaction orders characterizing primary and secondary nucleation, and , for which the maximal aggregate partitioning corresponds to smaller and larger compartment volumes. This prediction can be related to specific aggregating systems for which the values of the reaction orders and have been experimentally determined (References see caption of Figure 4). Using these values for the reaction orders, our model predicts that largest partitioning is obtained for large compartments in systems of aggregating tau and yeast prion Ure2p. These two examples belong primarily to the class of systems where the mechanism responsible for the formation of new aggregates in the late stage is fragmentation which has a zero secondary reaction order, (i.e. nucleation is monomer independent). For non-fragmenting systems with , our model predicts different scenarios for aggregating systems: largest aggregate partitioning for large compartment volumes occurs in the case branching systems, such as actin in the presence of the complex Arp2/3, as well as systems proliferating through monomer dependent secondary nucleation with , such as the Islet Amyloid Polypeptide (IAPP). In contrast, largest aggregate partitioning is reached for small compartments in the case of the 40- and 42-residue forms of Amyloid-β peptide (Aβ40 and Aβ42).

Figure 4.

Theoretical predictions of maximal aggregate partitioning for various aggregating systems.

Our predictions are summarized by a phase diagram depicting that aggregating systems characterized by different reaction orders for primary and secondary nucleation, and , show maximal aggregate partitioning for large or small compartments, respectively. The two regions where either large or small compartments lead to a larger partitioning of aggregates is separated by the line determined from Equation 6. For small compartments lead to larger aggregate partitioning, while for , larger compartments are beneficial. To illustrate which scenario might apply to which kind of aggregating system, we indicate the measured values of the primary and secondary reaction orders for a range of systems propagating through fragmentation (blue), lateral branching (green) or monomer-dependent secondary nucleation (red): Tau (Kundel et al., 2018), yeast prion Ure2p (Zhu et al., 2003), IAPP (Ruschak and Miranker, 2007), Amyloid-β40 (for monomer concentrations below 5 μM) (Meisl et al., 2014), Amyloid-β42 (Cohen et al., 2013).

Compartment volume and monomer partitioning control the total amount of aggregates

Our results have demonstrated that aggregates can be effectively partitioned inside liquid-like compartments, raising the question: can compartments also control the total amount of aggregates or their average size? To test this possibility, we compute the difference between the total amount of aggregates formed in the presence of liquid compartments, , compared to the number of aggregates formed in the homogeneous system without compartments, . The homogeneous case can be studied by considering equal compositions of both compartments, that is . This difference between the homogeneous case and the case with compartments be quantified by introducing the relative asymptotic aggregate concentration, , which is positive for an increased pool of aggregates, and negative for a lowered pool of aggregates relative to the homogeneous state. We find that compartments can affect the total number of aggregates relative to the homogeneous system depending on the relative values of the reaction orders for secondary nucleation and aggregate growth, the value of monomer partitioning and the amount of compartment material that in turn regulates compartment size . In particular, for reaction orders , the presence of the liquid droplet always reduces the total amount of aggregates formed relative to the homogeneous system for all values of and and thereby always leading to larger aggregates (Figure 5(a)). However, for , we find a different behaviour. For low partitioning factors , the presence of liquid compartments decreases the total number of aggregates, corresponding to a larger average aggregate size, while for larger values of , more and thereby shorter aggregates form compared to the homogeneous system (Figure 5(b)). This behavior is also affected by compartment volume; the corresponding boundary in the - diagram separates these two regimes corresponding to more but smaller or less but larger aggregates (Figure 5(c)). The role of the reaction order for secondary nucleation on total aggregate concentration and average size can be explained as follows. In a homogeneous system proliferating through secondary nucleation pathways, the average aggregate size in the saturating regime of the aggregation kinetics at long times scales as  (Michaels et al., 2015). Larger values of lead to an increase of monomers in the compartment, favouring both secondary nucleation and aggregate growth by elongation inside the compartment. If , the rate of secondary nucleation is increased by more than elongation, which results in more numerous aggregates and hence shorter aggregates due to the limiting and fixed amount of total monomer mass in the system. The opposite trend is observed when . In summary, a strong partitioning of aggregates inside compartments caused by a strong monomer partitioning (large ) is accompanied by an increase of the total number of aggregates in the system in the presence of secondary nucleation, while in the absence of secondary nucleation, the total amount of aggregates decreases.

Figure 5.

Compartments can change the total aggregate concentration compared to the homogeneous state without compartments.

(a,b) Relative asymptotic aggregate concentration as a function of volume fraction of the compartment material (connected to compartment volume ), where is the concentration of the homogeneous state in the absence of compartments. (a) For secondary reaction order , the total amount of aggregates is decreased compared to the case without compartments for all values of monomer partitioning and compartment material volume fractions and compartment volumes . (b) However, for , the total amount of aggregates is either increased or decreased relative to the homogeneous state. (c) Depending on the value of the monomer partitioning , compartments either lead to more but shorter aggregates (large , larger volume controlled by ) or less but larger aggregates Parameters: (a) , ; (b,c) , .

Discussion

By combining the theories of irreversible protein aggregation kinetics and phase separation, we have shown how liquid compartments can control the position and the total amount of aggregates. The coupling of slow aggregation and rapid phase separation leads to a mechanism whereby even a weak partitioning of monomers is amplified into a relatively large accumulation of aggregates in the compartment. Such partitioning of aggregates is a non-equilibrium effect and thereby not only determined by the phase separation parameters relevant at equilibrium (monomer partitioning and partitioning degree ) but in addition, it depends on kinetic parameters characterizing the aggregation kinetics (e.g. reaction orders and for primary and secondary nucleation). However, several other effects may influence or limit the resulting degree of aggregate partitioning.

Model validity

In our model, we have considered the case that monomers and aggregates do not affect phase separation and phase separation is driven by the competition between the entropic tendency to mix and interactions favoring demixing. Future work could be devoted to extending our model by a coupling between aggregates and the liquid compartment or by entropically driven phase separation, relevant for the assembly of coacervates (Overbeek and Voorn, 1957) or mixtures with depletion interactions. Moreover, our model is restricted to time scales when aggregates hardly diffuse and monomer diffusion is not affected by rheological properties of the aggregates. The observed strong aggregate partitioning may diminish if aggregates significantly diffuse, or if they slow down diffusion of monomers. Furthermore, we have focussed on the case where both phases have the same reaction rates of aggregation. This assumption may be inaccurate for protein-rich phases (Wei et al., 2017) but can be scrutinized using our model (see Appendix 3). We find that due to the power law dependence of the monomer partitioning (Equation 6) differences in aggregation rates must be very large to significantly affect the partitioning of aggregates. A lowered partitioning of aggregates could be caused by the coarsening dynamics of many droplets (Ostwald, 1897; Lifshitz and Slyozov, 1961; Bray, 1994). While coarsening via coalescence would not affect our results at all because aggregates remain confined inside the droplets, dissolving droplets undergoing Ostwald ripening would diminish the degree of aggregate partitioning. However, because the aggregation kinetics varies with compartment size, aggregates in droplets of different size may compete about monomers. This non-equilibrium competition could cause accumulation of more aggregates either in smaller or larger compartments. Overall, for systems where partitioning of monomers is fast relative to the aggregation kinetics, the mechanism underlying the strong partitioning of aggregates proposed in this study could be relevant for several phenomena in living cells. It could have impact on strategies of drug design or serve as a principle to speed up irreversible chemical reactions and can be tested experimentally.

In-vitro realization

Our quantitative predictions of strong aggregate enrichment inside a liquid-like compartment (Figure 3 (b–e) and Figure 4) are experimentally testable using recently developed bulk and microfluidic assays. For example, synthetic liquid biocompartments of tuneable size and composition can be used to locally affect reaction rates and partition proteins (Faltova et al., 2018) and thereby represent attractive platforms to investigate the partitioning and aggregation of different amyloidogenic peptides and proteins, including Amyloid-β. These synthetic compartments are highly flexible and allow to validate the effect of several parameters predicted in this work. For instance, the monomer partitioning factor could be varied in vitro by changing the degree of phase separation (Equation (1) and Figure 2) or by conjugating the proteins with specific sequences capable of tuning recruitment into the liquid compartments (Faltova et al., 2018). Moreover, the compartment volumes can be adjusted by the initial supersaturation via changes in temperatures, which affect the kinetic rate constants only weakly (Cohen et al., 2018). Measuring the concentration or size of aggregates inside and outside of the compartment by epi-fluorescence spectroscopy as a function of time and parameters such as the partitioning factor and compartment volume will allow for tracking aggregate enrichment as a function of compartment volume and test both the scaling predictions and the crossover of the scaling exponent from at small volume to at large volumes. It would be particularly interesting to test this prediction for different amyloid-forming protein systems or for varying the reaction orders and by adjusting the total amount of monomers (Meisl et al., 2014); see Figure 4.

In vivo relevance and implications for drug design

Our model may already provide a framework to explain the phenomena of aggregate partitioning inside living cells. An example of such phenomena could be the partitioning of pericentriolar material into centrosomes (Zwicker et al., 2014) and the spatial organization of aggregates inside stress granules (Molliex et al., 2015; Mateju et al., 2017). The propensity of aggregates to solidify the compartment as reported in Mateju et al. (2017) could be accounted for in our model through a gel-sol transition (Stockmayer, 1943; Harmon et al., 2017). Including the solidification induced by aggregates could lead to additional volume changes of the compartment which in turn may affect the aggregation kinetics. Furthermore, the enrichment of toxic aggregates inside liquid compartments may trigger new directions for drug design against aberrant protein aggregation. Our results suggest to design drugs not only with respect to their ability to interfere with the aggregation kinetics (Arosio et al., 2014) but also with respect to their partitioning properties into the liquid compartments. This strategy is reminiscent of quantifying the potency of low-molecular weighted anesthetics via the Meyer-Overton correlation based on solubility of the anaesthetics in oil (Meyer, 1937; Franks and Lieb, 1978).

General speed-up mechanism for chemical reactions

The reported feedback mechanism of aggregate growth mediated by liquid compartments may represent a general principle to spatially confine and speed up other irreversible chemical processes or to control aggregate amount and average size. Examples may include precipitation of proteins or polymerization kinetics of actin and microtubules (see also Figure 4). Indeed, a speed up of the chemical reactions could be expected due to the increased concentration of educts inside the liquid compartments. Thus, liquid compartments are ideal biomolecular microreactors that enrich the amount of products by dynamically exchanging reactants with their surroundings.

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

Thank you for submitting your article "Spatial control of irreversible protein aggregation" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by Frank Jülicher as Reviewing Editor and Arup Chakraborty as the Senior Editor. The following individual involved in the review of your submission has agreed to reveal his identity: Alexander Grosberg (Reviewer #3).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

The work on "Spatial control of irreversible protein aggregation" presents novel ideas and a theoretical approach to show that protein aggregates could predominantly form in liquid cellular compartments and strongly enrich there. Such compartments could therefore play a role to reduce protein aggregation outside such compartments and to control where irreversible aggregation is permitted to occur. The approach used is simple and elegant. The paper has a number of insights gleaned from both Flory-Huggins and kinetic theories of phase separation and aggregation kinetics, respectively. Presented equations are formulated to provide insight into the rich interplay between monomer partitioning, condensate volume fraction, and aggregation mechanism on the resulting aggregation location and rates under a quantitative framework. The main results are: 1) When monomers that tend to aggregate partition preferentially in a liquid droplet this results in a positive feedback driving aggregation inside droplets, as it causes a flux of monomers from the dilute phase to the droplet. 2) A phase diagram reveals a switch between regions where large or small compartments show maximal aggregate enrichment, respectively. This switch depends on the orders of the primary and secondary aggregate nucleation reactions. This is a strong and interesting paper which presents important theoretical ideas that are relevant for cell biology. The work presented is in principle suitable for eLife. However the authors should carefully consider several specific points raised by the reviewers and in a revised manuscript they should formulate testable predictions more explicitly and improve the links to potential experimental tests.

Essential revisions:

1) The main entrance point to the model is Equation 1 (same as 14). It operates with phase segregation between A-rich and A-poor liquid phases in which monomers and aggregates are also dissolved. This idea is not explained well, because terms (monomer, aggregate, A, B, etc.) are used without definitions. More importantly, what are A's and B's? Presumably A's and B's are also proteins of some sort. It is unclear why it is fair to talk about equilibrium and kinetics of proteins called monomers while assuming other proteins (in fact, perhaps many protein species!) to passively maintain equilibrium volume fractions ϕ and ϕ. One would actually expect that even a small amount of a particular protein called "monomer" could have a significant effect on the balancing of coexisting densities given the multitude of components hidden inside notations A and B.

2) In a regular theory of liquid nuclei (like, e.g., Lifshitz and Slyozov, 1961) an important part of the setup is dependence of equilibrium volume fractions on the nucleus size (the Laplace pressure effect); it is unclear why present theory neglects this altogether. To some extent authors may make an argument that concentration of "monomers" is small and does not affect the droplet size; but as irreversible aggregation continues, aggregates accumulate, and they seem to be bound to start affecting the droplet size eventually.

3) It is also not clear why in the kinetics part it is possible to neglect the fact that one of the phases might be dramatically more viscous than the other. The authors themselves cite works on polymer reputation and even possible solidification, but do not seem to consider the viscosity increase in the subsequent considerations, such as Equation 3. Again, this is possibly justifiable at the initial stage, but perhaps not so easily on a later time. I think that the authors have in mind some self-consistent system of conditions where all these questions may be answered, but they did not present these conditions in a clear way. These conditions and the terminology should be clarified.

4) In the subsection “Mathematical model for liquid compartments controlling protein aggregation”, the authors present an interesting approach to understanding the partition coefficient that is quite useful. However, the authors comment that the partitioning is solely governed by the relative interaction strength Δχ between the monomers with the A and the B components. This statement seems overstated, since other issues such as the relative density of the components and the potential of non-χ type interactions (e. g. direct binding between A and B) may not captured.

5) The text around Equations 3B-3D needs to be elaborated. n1 and n2 need to be defined and the description of the processes being considered requires more detail.

6) It is worth noting that the type of analyses carried out in this paper, including the analogy with Boundary Layer theory, have been considered extensively in the past in the context of gas-liquid reactions that are used extensively in the chemical industry for separation processes. The work of G. Astarita and P.V. Danckwerts are just two examples. Some reference to these studies may be appropriate.

7) It is important to note whether the biological systems placed on the "phase diagram" in Figure 4 are known to correspond to the big or small compartment limits. If not, how might the prediction be tested.

8) More generally the authors should formulate testable predictions more explicitly and improve the links to potential experimental tests.

9) The model assumes that the condensed phase and dilute phase have the same kinetics of growth for aggregation. This is likely inaccurate and thus should be addressed. For example, others have noticed (e.g. Wei et al., 2017), a significant size originated non-ideality in partitioning occurs in many protein-rich phases in vivo and in vitro. This non-ideality would weaken the kinetics of growth for aggregates and potentially limit their size. It is unclear how much such non-ideal contributions would perturb the conclusions thus requiring some discussion.

10) The authors focus on cytoplasm vs. specific biomolecule-dense liquid phases. The authors should not limit themselves to the cytoplasm, since these principles would apply to aggregation in the nucleoplasm as well, and are thus worth at least commenting on.

11) The authors discuss the concept of primary vs. secondary nucleation, and indeed their analysis focuses on key differences in these two modes. But I think they should be introduced, at least in a few sentences, in a qualitative way, prior to diving into the mathematical details.

12) The authors focus on droplet size as a key factor, but the analysis seems to be exclusive with the simplest mapping from volume fraction to droplet size, by taking the condensed phase as all having coalesced into a single droplet. The authors should at least comment on the expected effects of how a distribution of droplet sizes, such as those commonly reported in the intracellular phase separation literature, would manifest in this context.

[Editors' note: further revisions were requested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "Spatial control of irreversible protein aggregation" for further consideration at eLife. Your revised article has been favorably evaluated by Arup Chakraborty (Senior Editor) and a Reviewing Editor.

The manuscript has been improved but there is one remaining issue that needs to be addressed before acceptance, as outlined below:

You have addressed most points raised and have improved the manuscript. However, an important criticism of the reviewers was formulated:

"8) More generally the authors should formulate testable predictions more explicitly and improve the links to potential experimental tests."

The reviewers felt that it was rather unclear how the ideas of the paper could be tested in experiments and whether there are clear predictions from the work for future experiments. In the revision of the manuscript this point was addressed somewhat superficially. This is not easy to do, but we would be grateful if you might try again to address this point more carefully. We anticipate that the paper will be accepted after that.

Essential revisions:

1) The main entrance point to the model is Equation 1 (same as 14). It operates with phase segregation between A-rich and A-poor liquid phases in which monomers and aggregates are also dissolved. This idea is not explained well, because terms (monomer, aggregate, A, B, etc.) are used without definitions. More importantly, what are A's and B's? Presumably A's and B's are also proteins of some sort. It is unclear why it is fair to talk about equilibrium and kinetics of proteins called monomers while assuming other proteins (in fact, perhaps many protein species!) to passively maintain equilibrium volume fractions ϕI and ϕII. One would actually expect that even a small amount of a particular protein called "monomer" could have a significant effect on the balancing of coexisting densities given the multitude of components hidden inside notations A and B.

The reviewers suggest improving the discussion and presentation of the paragraph about the three component phase separation model and why a consideration at phase separation equilibrium of the A-B mixture is valid. We have now revised the corresponding paragraphs and explicitly mention what species A and B could represent (note, in any case, that an exact definition is unnecessary because only physical parameter enter the model). Moreover, we now briefly explain why the binary phase separation between A and B is well equilibrated while the partitioning of monomer deviates weakly from the partitioning equilibrium. Such assumptions are well satisfied because of two reasons: 1) Monomer species are very dilute and thus cannot affect A-B phase equilibrium. 2) Moreover, aggregation kinetic is slow relative to the partitioning of monomers via diffusion. Thus, at each time point during the aggregation kinetics, monomer partitioning is very close to the partitioning equilibrium. For better overview we have now combined all essential assumptions of our model in the first section of the model description.

2) In a regular theory of liquid nuclei (like, e.g., Lifshitz and Slyozov, 1961) an important part of the setup is dependence of equilibrium volume fractions on the nucleus size (the Laplace pressure effect); it is unclear why present theory neglects this altogether. To some extent authors may make an argument that concentration of "monomers" is small and does not affect the droplet size; but as irreversible aggregation continues, aggregates accumulate, and they seem to be bound to start affecting the droplet size eventually.

We agree with the reviewer that we have neglected the effect of the compartment radius R on the equilibrium concentrations. For a binary A-B mixture, this effect can be captured by the Gibbs-Thomson relationship stating that the relative increase on both equilibrium concentrations inside and outside increases, cαlγα/R, where lγα denotes the capillary length inside and outside. However, the capillary length is typically in the order of a few molecular sizes. Thus, as long as the compartment size significantly exceeds the molecular size (which is the typical case inside cells for example), the actual equilibrium concentration values would differ only very weakly relative to the equilibrium concentrations for large R. Of course, this little concentration difference is key for the coarsening kinetics (Ostwald ripening) where the small fraction, lγα/R, competes with another very small value, namely the supersaturation. However, in our work we only consider one compartment and not the coarsening of many compartments. In the presence of monomers the arguments above apply as well because the considered monomer concentrations are so dilute that we can neglect their impact on the equilibrium concentrations and also on the surface tension (in the case they would act as surfactants). We have added a comment to the main text and a paragraph to the Appendix discussing when Laplace pressure effects can be neglected.

Please note that we have estimated typical values for homogeneous volume fraction of aggregating monomers to be 10-9 – 10-8. Hence, even for large partitioning factors, such as e.g. Γ=100, the local concentrations of monomers remain dilute; this allows us to safely neglect the contributions of monomers to droplet volume. Based on the reviewer’s comments, we have now commented on the role of Laplace pressure effects in the appendix (see text after Equation 13) and the impact of monomers on droplet in the main text (see end of paragraph “Model for liquid compartments controlling protein aggregation”).

3) It is also not clear why in the kinetics part it is possible to neglect the fact that one of the phases might be dramatically more viscous than the other. The authors themselves cite works on polymer reputation and even possible solidification, but do not seem to consider the viscosity increase in the subsequent considerations, such as Equation 3. Again, this is possibly justifiable at the initial stage, but perhaps not so easily on a later time. I think that the authors have in mind some self-consistent system of conditions where all these questions may be answered, but they did not present these conditions in a clear way. These conditions and the terminology should be clarified.

We agree that very complex rheological properties could emerge in the later stage of the aggregation kinetics. It is correct that our assumption of a constant monomer diffusion coefficient, while aggregates hardly diffuse, may no more be quantitively correct. One may expect that the monomers could pick up a thickening behaviour due to the interactions with the aggregates on long time scales that would lower their diffusivity as a function of time. We have not included these effects in our model. However, we think that they are worth studying for specific cellular and aggregating systems because rheological properties, in particular, time dependent ones, vary from system to system. Please note that in our manuscript, we present a study of a mechanism which is expected to be qualitatively robust against different rheological properties – of course, quantitative changes of the degree of aggregate partitioning may occur. Based on the reviewer’s comment above we have decided to add a comment to the main text and the outlook.

4) In the subsection “Mathematical model for liquid compartments controlling protein aggregation”, the authors present an interesting approach to understanding the partition coefficient that is quite useful. However, the authors comment that the partitioning is solely governed by the relative interaction strength Δχ between the monomers with the A and the B components. This statement seems overstated, since other issues such as the relative density of the components and the potential of non-χ type interactions (e. g. direct binding between A and B) may not captured.

We thank the reviewers for this comment. Our statement only holds for dilute monomer volume fraction (typical case for the considered aggregating systems discussed in our manuscript) and in the absence of binding processes. We have corrected this statement in our manuscript.

5) The text around Equations 3B-3D needs to be elaborated. n1 and n2 need to be defined and the description of the processes being considered requires more detail.

The reviewers suggest elaborating on the presentation about the primary and secondary reaction orders. Accordingly, we have revised the text around Equations (3B-3D). In particular, we now give a physical interpretation of the reaction orders n1 and n2.

6) It is worth noting that the type of analyses carried out in this paper, including the analogy with Boundary Layer theory, have been considered extensively in the past in the context of gas-liquid reactions that are used extensively in the chemical industry for separation processes. The work of G. Astarita and P.V. Danckwerts are just two examples. Some reference to these studies may be appropriate.

We thank the reviewers for suggestions of references on the boundary layer theory. We have included the proposed references to the discussion of boundary-layer dynamics (see text before Equation 21 in the Appendix).

7) It is important to note whether the biological systems placed on the "phase diagram" in Figure 4 are known to correspond to the big or small compartment limits. If not, how might the prediction be tested.

To the best of our knowledge, no experiments are currently available that would allow us to judge which system corresponds to the small or large compartment limit. However, we appreciate the suggestion to think more of how to test the “phase diagram”. We thus revised the paragraph in the conclusion section, where we have already suggested how our prediction could be tested. Now we also include a short discussion related to Figure 4.

8) More generally the authors should formulate testable predictions more explicitly and improve the links to potential experimental tests.

We have revised the paragraphs in the conclusion section, which now present explicit experimental tests. Moreover, we tried to stress the link to experimental systems by structuring the conclusion section more clearly using subheadings.

9) The model assumes that the condensed phase and dilute phase have the same kinetics of growth for aggregation. This is likely inaccurate and thus should be addressed. For example, others have noticed (e.g. Wei et al., 2017), a significant size originated non-ideality in partitioning occurs in many protein-rich phases in vivo and in vitro. This non-ideality would weaken the kinetics of growth for aggregates and potentially limit their size. It is unclear how much such non-ideal contributions would perturb the conclusions thus requiring some discussion.

It is correct that the assumption of equal kinetic parameters inside and outside will not hold for all phase separating systems combined with aggregation. In our manuscript we wanted to make the point that a strong partitioning of aggregates already occurs for equal kinetic parameters inside and outside. To test how deviations from our ideal assumption affect our conclusions we have derived an equation for the partitioning of aggregates in the case of different growth rates inside and outside. We have now added a corresponding discussion to the main text and show the extended equations in the appendix. We also cite the reference mentioned by the reviewers and discuss Brangwynne et al.’s finding of a semi-dilute mesh of proteins forming droplets that can affect physical properties inside the compartment.

10) The authors focus on cytoplasm vs. specific biomolecule-dense liquid phases. The authors should not limit themselves to the cytoplasm, since these principles would apply to aggregation in the nucleoplasm as well, and are thus worth at least commenting on.

Great point. We now also mention the nucleoplasm (see e.g. Abstract).

11) The authors discuss the concept of primary vs. secondary nucleation, and indeed their analysis focuses on key differences in these two modes. But I think they should be introduced, at least in a few sentences, in a qualitative way, prior to diving into the mathematical details.

Following the reviewers’ suggestion, we have included (before Equation 3) a detailed discussion of primary and secondary nucleation in protein aggregation (see also reply to point 5).

12) The authors focus on droplet size as a key factor, but the analysis seems to be exclusive with the simplest mapping from volume fraction to droplet size, by taking the condensed phase as all having coalesced into a single droplet. The authors should at least comment on the expected effects of how a distribution of droplet sizes, such as those commonly reported in the intracellular phase separation literature, would manifest in this context.

This is a fascinating comment. We would expect that either the smallest or the largest droplets in the system would initially enrich more of the aggregates (depending on the reaction orders of primary vs. secondary nucleation). During the aggregation kinetics there would be a competition about the monomers between the droplets of different size. Droplets of different size thus end up with different aggregate fraction. We now mention these thoughts in the conclusion of our manuscript.

[Editors' note: further revisions were requested prior to acceptance, as described below.]

You have addressed most points raised and have improved the manuscript. However, an important criticism of the reviewers was formulated:

"8) More generally the authors should formulate testable predictions more explicitly and improve the links to potential experimental tests."

The reviewers felt that it was rather unclear how the ideas of the paper could be tested in experiments and whether there are clear predictions from the work for future experiments. In the revision of the manuscript this point was addressed somewhat superficially. This is not easy to do, but we would be grateful if you might try again to address this point more carefully. We anticipate that the paper will be accepted after that.

We were delighted by the positive response of the reviewers to our work. We have included suggestions into our manuscript in the context of experimentally testable predictions. We have also added a new short section on a calculation of the amount of aggregates in the compartments – summarized in a new Figure 5, which we think is a complementary perspective on the rest of the paper.