Article The fitness cost and benefit of phase-separated protein deposits Natalia Sanchez de Groot1,2,3,*,† , Marc Torrent Burgas1,4,† , Charles NJ Ravarani1 , Ala Trusina5, Salvador Ventura6 & M Madan Babu1,** Abstract macromolecular assemblies (Veis, 2011). The formation of protein deposits can affect biological processes as a result of a loss and/or Phase separation of soluble proteins into insoluble deposits is gain of function and has been implicated in disorders such as associated with numerous diseases. However, protein deposits can Alzheimer’s or Parkinson’s disease (Chiti & Dobson, 2006; Jahn & also function as membrane-less compartments for many cellular Radford, 2008; Babu et al, 2011; Gsponer & Babu, 2012; Sanchez de processes. What are the fitness costs and benefits of forming such Groot et al, 2012; Ciryam et al, 2013; Lin et al, 2015; Woerner et al, deposits in different conditions? Using a model protein that phase- 2016). However, protein sequences predisposed to form deposits are separates into deposits, we distinguish and quantify the fitness found in all kingdoms of life suggesting a neutral or advantageous contribution due to the loss or gain of protein function and deposit effect on cell fitness (Li et al, 2012; Newby & Lindquist, 2013; formation in yeast. The environmental condition and the cellular Berchowitz et al, 2015; Khan et al, 2015; Chavali et al, 2017a). In demand for the protein function emerge as key determinants of fact, it has been shown that phase separation-promoting sequences fitness. Protein deposit formation can influence cell-to-cell varia- are essential to build membrane-less structures and higher-order tion in free protein abundance between individuals of a cell popu- assemblies with several biological functions (Maji et al, 2009; lation (i.e., gene expression noise). This results in variable Bershtein et al, 2012; de Groot et al, 2012; Gsponer & Babu, 2012; manifestation of protein function and a continuous range of Ciryam et al, 2013; Toretsky & Wright, 2014; Lin et al, 2015; Miller phenotypes in a cell population, favoring survival of some individu- et al, 2015; Nott et al, 2015; Suresh et al, 2015; Wallace et al, 2015; als in certain environments. Thus, protein deposit formation by Xiang et al, 2015; Zhu & Brangwynne, 2015; Chavali et al, 2017a; phase separation might be a mechanism to sense protein concen- Holehouse & Pappu, 2018). tration in cells and to generate phenotypic variability. The select- Protein deposit formation due to phase separation inside the cell able phenotypic variability, previously described for prions, could is a complex process that depends on a number of features such as be a general property of proteins that can form phase-separated the physicochemical properties of the polypeptide sequence (e.g., assemblies and may influence cell fitness. hydrophobicity, net charge; de Groot et al, 2006), local protein concentration (e.g., high vs. low concentration; Ciryam et al, 2013; Keywords cell fitness; natural selection; phase separation; phenotypic diver- Levy et al, 2014; Stepanenko et al, 2016), and its interaction with sity; protein deposit cellular components (e.g., chaperones, RNA; De Baets et al, 2011; Subject Categories Protein Biosynthesis & Quality Control; Quantitative Gsponer & Babu, 2012; Sanchez de Groot et al, 2012; Miller et al, Biology & Dynamical Systems 2015; Zhang et al, 2015; Jain et al, 2016; Pak et al, 2016; Maharana DOI 10.15252/msb.20178075 | Received 26 October 2017 | Revised 14 March et al, 2018). The different combinations of these features result in 2019 | Accepted 15 March 2019 the formation of deposits with different physicochemical and Mol Syst Biol. (2018) 15: e8075 dynamic properties (Fig 1A). According to their viscoelastic charac- teristics and their ability to exchange components with the cyto- plasm, protein assemblies can adopt a wide range of states that go Introduction from highly dynamic states with liquid droplet properties to almost static states with solid-like properties (Fig 1A; Franzmann et al, The exposure of certain polypeptide segments in a protein to the 2018). In this manner, phase-separated assemblies can be classified solvent can trigger a process in which proteins phase-separate into as liquid–liquid (e.g., nucleolus) or liquid–solid (e.g., amyloid 1 Medical Research Council Laboratory of Molecular Biology, Cambridge, UK 2 Bioinformatics and Genomics Programme, Centre for Genomic Regulation (CRG), Barcelona, Spain 3 Universitat Pompeu Fabra (UPF), Barcelona, Spain 4 Systems Biology of Infection Lab, Department of Biochemistry and Molecular Biology, Universitat Autònoma de Barcelona, Barcelona, Spain 5 Niels Bohr Institute, University of Copenhagen, Copenhagen, Denmark 6 Institut de Biotecnologia i Biomedicina and Departament de Bioquímica i Biologia Molecular, Universitat Autònoma de Barcelona, Barcelona, Spain *Corresponding author. Tel: +34 933160380; E-mail: natalia.sanchez@crg.eu **Corresponding author. Tel: +44 1223 267066; E-mail: madanm@mrc-lmb.cam.ac.uk † These authors contributed equally to this work ª 2019 MRC Laboratory of Molecular Biology. Published under the terms of the CC BY 4.0 license Molecular Systems Biology 15: e8075 | 2019 1 of 20 Molecular Systems Biology Natalia Sanchez de Groot et al aggregate) according to their viscoelastic and dynamic properties pituitary hormones (Maji et al, 2009); the phase-separated state, (e.g., deformation and coalescence; Morley et al, 2002; Jahn & such as the amyloid-like form of Rim4 (Berchowitz et al, 2015); or Radford, 2008; Kaganovich et al, 2008; Escusa-Toret et al, 2013; in both states (e.g., monomeric Orb2 represses, whereas oligomeric Toretsky & Wright, 2014; Lin et al, 2015; Miller et al, 2015; Zhang Orb2 activates translation; Khan et al, 2015). Therefore, the forma- et al, 2015; Zhu & Brangwynne, 2015; Jain et al, 2016; Boeynaems tion of phase-separated structures is not only associated with et al, 2018; Franzmann et al, 2018; Alberti et al, 2019). diseases but can positively influence cell fitness under different This variety of assemblies with distinct dynamic behavior conditions. For instance, the recruitment of certain metabolic (Fig 1A) is employed by cells to regulate key biological processes enzymes into cytoplasmic reservoirs/deposits enhances yeast such as gametogenesis (Berchowitz et al, 2015), programmed necro- survival during periods of starvation and stress (O’Connell et al, sis (Li et al, 2012), hormone storage (Maji et al, 2009), or memory 2014; Petrovska et al, 2014; Suresh et al, 2015; Wallace et al, 2015; maintenance (Khan et al, 2015), among other functions (Toretsky & Riback et al, 2017; Franzmann et al, 2018). Thus, beyond prion Wright, 2014; Miller et al, 2015; Chakrabortee et al, 2016; Mitrea & domain-containing proteins, the extensive number of reported self- Kriwacki, 2016). Proteins that are able to phase-separate can assembly events suggests that this is likely to be an intrinsic prop- perform their biological functions in the soluble state such as the erty of the polypeptide chain (Dobson, 1999; Monsellier et al, A Properties of the intracellular B System to measure the effect of protein assemblies protein phase separation on cell fitness Soluble Liquid Solid (Diffuse) (Droplet) (Compact) Functional Reporter Aggregation region region prone region I II III (URA3) (GFP) (Aβ) Dynamic Static D Aβ C Ura3p Ura3p GFP GFP Ura3p Reaction Ura3p Activity + Uracil Non-essential - Uracil Essential + 5FOA* Toxic URA3sol URA3agg Ura3p catalyses two reactions: Orotidine Uridine Supports 5’-monophosphate monophosphate cell division (OMP) (UMP) Impairs 5-Fluoroorotic acid 5-Fluorouracil cell division (5-FU) (5-FOA) 2μm *Uracil is present in the media to ensure cell growth and survival. Figure 1. Model proteins and design of experimental system. A Protein phase separation can lead to the formation of protein assemblies with varying dynamics. Soluble proteins may phase-separate into liquid-like droplets or insoluble deposits that, according to their viscoelastic properties and their ability to exchange components with the cytoplasm, vary from being dynamic to static. B Design of the modular system to measure the effect of protein phase separation on cell fitness. C Growth media composition and the essential/non-essential/toxic roles of Ura3p. D The model proteins consist of a fusion between Ura3p and GFP (URA3sol, left) and the amyloid-b-peptide of 42 residues (Ab) (URA3agg, right). After 18 h of expression, Ura3psol remains homogeneously distributed whereas Ura3pagg is accumulated into intracellular foci. 2 of 20 Molecular Systems Biology 15: e8075 | 2019 ª 2019 MRC Laboratory of Molecular Biology Natalia Sanchez de Groot et al Molecular Systems Biology 2008). The ability for proteins to phase-separate tends to be encoded is a polypeptide segment whose biochemical activity can be essen- in specific regions or domains (Appendix Fig S1; Moore et al, 2008; tial, non-essential, or toxic for the cell. We chose the endogenous Marsh & Teichmann, 2010; Li et al, 2012; Berchowitz et al, 2015; yeast enzyme orotidine-50 -phosphate decarboxylase (Ura3p) Khan et al, 2015; Hervas et al, 2016). Such a modular organization involved in the production of pyrimidine nucleotides and widely permits independent evolution of protein regions and facilitates the employed for positive and negative selection (Seiple et al, 2006; emergence of proteins with new properties through recombination Fig 1C). Ura3p activity is essential in yeast cells that are grown in (Moore et al, 2008; Marsh & Teichmann, 2010). the absence of uracil. However, Ura3p activity is non-essential when The ultimate fitness consequences of phase-separated structures grown in the presence of uracil. Furthermore, Ura3p activity is toxic will depend on the function of the protein, the components seques- in the presence of an alternative substrate named 5-fluoroorotic acid tered in the assembly, dynamic nature, and the molecular structure (5FOA), as it leads to the production of a toxic compound (5-fluor- of the phase-separated assembly, among other factors. Due to the ouracil, 5FU) leading to cell division arrest and cell death (Seiple complexity and the diverse effects associated with phase separation, et al, 2006; Fig 1C). The second component is a reporter (green flu- there is a need to develop rational approaches that permit the quan- orescent protein; GFP) that allows monitoring the integrity and the tification of the effects of forming such deposits on cell fitness in distribution/location of the protein in the cell (Fig 1B and D). The multiple environments and determine how proteins with phase third component is a phase separation-promoting segment that leads separation-prone segments are selected for, or against, in a cell to the formation of intracellular, insoluble protein deposits, which population. To begin unraveling the effect of these factors, we have in our system is the 42-amino acid amyloid-b-peptide (Ab) (de Groot designed a model protein that phase-separates to primarily form & Ventura, 2006; Villar-Pique & Ventura, 2013; Sanchez de Groot insoluble deposits and describe a population genetics approach. et al, 2015; Fig 1D). This allows us to disentangle and quantify the effect of the following We built two different constructs: one encoding Ura3p fused to factors associated with phase separation in different environments: GFP (to obtain a protein with no or low phase separation potential, the fitness change associated with (i) deposit formation, (ii) the loss Ura3psol) and another that includes the Ab peptide to promote the of an essential biochemical activity of the protein that forms phase separation process (Ura3pagg) (Fig 1D; Materials and Meth- the deposit, and (iii) the gain of a beneficial effect due to deposit ods). Instead of using Ura3psol as a control, we initially considered formation. fusing Ura3p-GFP to a soluble variant of the Ab42 (Villar-Pique & Ventura, 2013; Sanchez de Groot et al, 2015; e.g., a non-foci- forming variant). However, the different soluble variants are not Results always completely soluble in yeast, since after fractionation they are still found in the insoluble part and they form deposits in some of Model for protein phase separation: one protein, three roles the stress environments investigated in the current work (Villar-Pique & Ventura, 2013). Hence, we decided that the addition Various studies have reported diverse and often conflicting effects in of a soluble variant of Ab42 will not be a suitable control for gener- terms of the beneficial and detrimental effects of protein deposit ating the soluble version of Ura3p. We also decided against fusing formation on cell fitness (Maji et al, 2009; Geiler-Samerotte et al, another soluble protein with a length similar to Ab42 but with a dif- 2011; Sanchez de Groot et al, 2012; Escusa-Toret et al, 2013; ferent sequence because URA3-GFP fusion is already a large protein Tomala et al, 2014). These differences are understandable if one and the inclusion of a “random” short soluble peptide sequence considers the complexity of deposit formation, the differences in the might not affect the cost significantly. dynamic nature of the deposit, biochemical function of the protein We integrated the two chimeric genes (Ura3psol and Ura3pagg) forming the deposit, and the diversity in the experimental conditions into a stable genomic region (TRP1 locus) to ensure steady expres- in which the studies have been carried out. Furthermore, prior stud- sion in multiple generations. We included an inducible promoter ies in the literature typically do not explicitly discriminate the dif- (GAL1) to control transcription and guarantee expression under dif- ferent phase separation processes or the different fitness effects ferent environments (Materials and Methods; Appendix Fig S2). associated with proteins (e.g., loss vs. gain of function due to Their integration in the S. cerevisiae genome resulted in two strains deposit formation). Here, we measure and disentangle the effects of (URA3sol and URA3agg) with the same genomic background and phase separation of a model protein. To trigger the phase separation with similar mRNA expression levels (Fig EV2). This ensures that process, instead of mutating an endogenous protein to destabilize it the fitness cost of expressing the constructs as gratuitous proteins as has been done before (Geiler-Samerotte et al, 2011; Tomala et al, will remain similar in the two strains (Dekel & Alon, 2005; Pena 2014), we designed a modular system that allows us to disentangle et al, 2010; Plata et al, 2010; Geiler-Samerotte et al, 2011; Kafri and quantify the cost/benefit of protein phase separation while et al, 2016) and hence minimally influence the measurements of fit- tuning different roles influencing this process (e.g., the essentiality/ ness effects (next section). We monitored competitive growth and non-essentiality/toxicity of a protein). The model protein phase- formation of deposits in different environments by changing the separates from a mainly soluble, functionally active state into a osmotic pressure (1 M sorbitol, 0.5 M NaCl), oxidation level primarily insoluble, functionally less active state (Materials and (0.5 mM H2O2, 1 mM DTT), and temperature (37, 30 and 25°C) and Methods, Figs 5 and EV1, and Appendix Fig S2). in the presence of a chemical chaperone (0.5 M proline). After 18 h Mimicking the domain organization seen in nature (Derkatch of induction at standard growth conditions (30°C) in mid-log phase, et al, 1996; Moore et al, 2008; Marsh & Teichmann, 2010; Ura3psol remains distributed through the cytoplasm, whereas Appendix Fig S1), our model protein is made up of three modular Ura3pagg forms stable, non-dynamic protein deposits similar to components (Fig 1B; Materials and Methods). The first component IPODs (insoluble protein deposits; Kaganovich et al, 2008), as ª © 2019 MRC Laboratory of Molecular Biology Molecular Systems Biology 15: e8075 | 2019 3 of 20 Molecular Systems Biology Natalia Sanchez de Groot et al measured by FRAP experiments (Figs 1D and EV3). Although some et al, 2015). This growth phase will ensure a constant doubling of these assemblies may contain other proteins (Rothe et al, 2018), time, a young population, and a low number of aged protein assem- prior findings suggest that the vast majority of them are likely to be blies (Hill et al, 2016). The selection coefficient (S) is related to the Ura3pagg (Morell et al, 2011; Sanchez de Groot et al, 2015). difference in growth rate between the two strains (Fig 2B) and quantifies how much the cell fitness increases (positive values) or Quantifying selection for/against phase separation in decreases (negative values) due to the formation of deposits in different environments URA3agg in comparison with URA3sol. The results obtained at standard growth conditions (30°C) show We experimentally determined how the formation of intracellular that the formation of Ura3pagg deposits can be neutral (in the pres- protein deposits is selected for, or against, in a population when the ence of uracil; non-essential), deleterious (in the absence of uracil; two strains (URA3sol and URA3agg) are grown in competition essential), or even advantageous (in the presence of 5FOA; toxic) (Fig 2A). We grew mixed cell cultures (1:1 initial proportion) for for yeast, depending on the composition of the growth medium three days in exponential phase and measured the selection coeffi- (Fig 2C; average S-values from two biological replicates). Therefore, cient (S) by PCR (Fig 2A; Appendix Table S1; Materials and Meth- without changing the protein sequence or the genotype, we can ods; Chevin, 2011; Geiler-Samerotte et al, 2011; Sanchez de Groot quantify different overall effects of deposit formation on cell fitness A COMPETITION EXPERIMENT D Outcome of protein Medium Ura3p Activity phase separation URA3sol + Uracil Non-essential Neutral vs. URA3agg - Uracil Essential Detrimental + 5FOA Toxic ENVIRONMENT osmotic pressure temperature E Overal fitness effect of phase separation oxidative levels Measured Selection Coefficient ( S ) * Selection (S) ∞ -0.08 -0.04 0.00 0.04 0.08 0.12 PHASE TRANSITION loss of function deposit formation 37ºC gain of function Major effects 30ºC detrimental neutral H2O2 B DTT SELECTION COEFFICIENT ; = 1/ Sorbitol C Overal fitness effect of phase separation Minor effects 25ºC Measured Selection Coefficient ( S ) Medium S+Uracil S+Uracil S-Uracil S+5FOA NaCl Environment S-Uracil 30ºC 0.002 -0.045 0.055 Proline S+5FOA *in all cases SE<0.05 Figure 2. Protein phase separation and selection coefficient. A Experimental design to measure the selection coefficient (S) upon growing URA3agg in competition with URA3sol. B The selection coefficient, S, is proportional to the difference in growth rates between URA3agg and URA3sol (Materials and Methods). x is the growth rate, which corresponds to the inverse of doubling time, s. C, D The S-values measured at standard conditions (30°C) indicate that depending on the media composition (+Ura, Ura, +5FOA), the formation of protein deposits can be neutral, deleterious, or even advantageous for the cell as the function becomes non-essential, essential or toxic, respectively. E S-values measured in different environments. In all cases, the standard error is below 5%. See also Appendix Table S1. Color scale: purple in the absence of uracil (-Ura), white in the presence of uracil (+Ura), and orange in the presence of 5FOA (+5FOA). At 37°C, the heat stress together with the inability to fold Ura3p has such a strong fitness effect that the strains barely grew in the absence of uracil, impeding the measurement of S. A bar of purple-white diagonal lines indicates a presumed value of S at 37°C without uracil. 4 of 20 Molecular Systems Biology 15: e8075 | 2019 ª 2019 MRC Laboratory of Molecular Biology Natalia Sanchez de Groot et al Molecular Systems Biology (Fig 2D). In this manner, the system we developed allows for the (FFOCIagg/FTOTALagg) and a proportionality constant/coefficient (a) quantification of the overall effects (both negative and positive fit- that depends on the environment (Fig 3A, equations 2 and 3). This ness effects) of deposit formation upon protein phase separation in constant/coefficient is indicative of the magnitude of the effect of different environments (Fig 2E and Appendix Table S1). We find a forming the protein deposit on the fitness in a particular environ- wide range of selection coefficients suggesting that URA3agg is dif- ment and is considered to remain the same for an environment irre- ferentially selected for or against when the role of the protein is dif- spective of whether Ura3p is non-essential, essential or toxic (i.e., ferent (essential/non-essential/toxic; in different media) and in when grown in +Uracil, Uracil or +5FOA; see Materials and Meth- different environments (Fig 2E). Changing the oxidation levels ods). The effect of reduced Ura3p activity due to deposit formation (0.5 mM H2O2, 1 mM DTT) or increasing the temperature (37°C) is defined as the product of the apparent free, active protein has a major effect on fitness compared to conditions in which we (FCYTOagg/FTOTALsol, ratio of free protein in the URA3agg strain to the changed the osmotic pressure (1 M sorbitol, 0.5 M NaCl), or maximum that can be available in a cell) and a proportionality decreased the temperature (25°C) or when we added a chemical constant/coefficient (b for cost/c for benefit) that is indicative of the chaperone (0.5 M proline). This observation raises the question as magnitude of the effect of the loss/gain of biochemical activity on to how and why the phase separation of the same protein leads to fitness in a particular environment (Fig 3A, equations 4 and 5, such differences in selection coefficients under different environ- Materials and Methods). Although both strains (URA3agg and ments. We investigate this question in the next sections (Fig EV4). URA3sol) present similar transcript levels (Fig EV2), due to deposit formation, a fraction of translated Ura3pagg can be rapidly removed Disentangling the different effects of protein phase separation through autophagy or other mechanisms (Villar-Pique & Ventura, 2013; Sanchez de Groot et al, 2015; Fig EV1). This may hence lead The overall fitness effect (cost/benefit) of protein phase separation to a reduction in FTOTALagg when compared to FTOTALsol (Villar-Pique is not only determined by the loss or gain of the biochemical activity & Ventura, 2013; Sanchez de Groot et al, 2015; Fig EV1; of the protein, but also due to the cost of deposit formation (e.g., Appendix Fig S2; Appendix Table S1; Materials and Methods). amino acid sequestration in deposits, sequestration of ATP-depen- dent chaperone activity and other proteins, toxicity of the assembly; Environment modulates the magnitude of the effects of phase Maji et al, 2009; Olzscha et al, 2011;; Gsponer & Babu, 2012; separation on cell fitness Sanchez de Groot et al, 2012; Suraweera et al, 2012; Tomala et al, 2014; Patel et al, 2017); Fig 3A, equation 1). To infer the impact of Environment can modulate the selection pressure on variants in a these effects, we considered that the measured selection coefficient population; genotypes and characteristics that are beneficial in (S) is determined by a combination of three factors: (i) the cost of certain environments could become detrimental in others (Pena deposit formation, (ii) the cost of losing the essential Ura3p et al, 2010; Bershtein et al, 2012). We examined how the fraction of biochemical activity, and (iii) the benefit of gaining a protective deposited vs. free Ura3p, the magnitude of effect, the fitness cost/ function against the toxic activity by sequestering Ura3p into benefit of deposit formation, and the selection coefficient (S) vary in deposits (Fig 3A, equation 2). In conditions where Ura3p is not different environments. essential (with uracil), any change in cell fitness primarily depends When Ura3p activity is not essential for a cell, although the frac- on the fitness cost of deposit formation (Fig 3A, equation3). In the tion of the deposited protein varies considerably in the different absence of uracil, the overall effect on fitness includes not only the environments, the measured selection coefficient remains closer to cost of deposit formation but also the cost of reducing Ura3p activity 0 (0 means no difference in cell fitness; Fig 3B, first plot; gray dots). due to deposit formation (Fig 3, equation 4). Finally, in the pres- However, when Ura3p activity is essential or toxic, the fraction of ence of 5FOA, the effect on fitness includes the cost of deposit the deposited protein in the different environments and the selection formation, and the fitness benefit of reducing the toxic activity of coefficient tend to vary considerably (e.g., S = 0.103, without Ura3p by sequestering the protein into the deposit (Fig 3, equa- uracil at 37°C and S = 0.055, with 5FOA at 30°C) (Fig 3B, first plot; tion 5). Although some Ura3p activity can be present in deposits purple and orange dots). Consistently, in terms of the fraction of (O’Connell et al, 2014; Wallace et al, 2015), this is likely to be free Ura3p, although the amount of free protein varies in the dif- significantly reduced (e.g., due to protein conformational changes, ferent environments, the measured selection coefficient does not dif- and restricted access to substrate) compared to the free, diffusible fer much when Ura3p is not essential (Fig 3B, second plot; gray well-folded protein as in the URA3sol strain (Suresh et al, 2015). dots). When Ura3p is essential or toxic, the fraction of free protein To relate the measured selection coefficient with the different in the different environments and the selection coefficient tend to effects of deposit formation, we quantified the total fluorescence vary considerably (Fig 3B, second plot; purple and orange dots). In (FTOTALagg) and the amount of fluorescence in deposit/foci addition, we find that the magnitude of effect for the cost of deposit (FFOCIagg), and the cytosol (FCYTOagg) for individual cells of the formation (a) is considerably lower than that of the effects associ- URA3agg population using confocal microscopy (Fig 3A). The ated with protein function, both with loss of essential enzymatic measured fluorescence is proportional to the amount of protein activity (b) and gain of protection against the toxic activity (c) in present in the cell (Soboleski et al, 2005). For the URA3sol strain, the different environments (Fig 3B, third plot). These observations the total fluorescence (FTOTALsol) is the same as cytosolic fluores- quantify and reflect the essentiality and lethality associated with the cence (FCYTOsol) and is indicative of the maximum amount of free, biochemical reaction catalyzed by Ura3p. Moreover, whereas the active Ura3p that can be available in our designed system (Materials cost of deposit formation remains relatively stable in the different and Methods). The cost of forming a deposit is defined as the environments (Fig 3B, fourth plot; gray dots), the fitness cost/bene- product of the fraction of the protein in deposits (phase-separated) fit associated with reduced Ura3p activity varies significantly in ª © 2019 MRC Laboratory of Molecular Biology Molecular Systems Biology 15: e8075 | 2019 5 of 20 Molecular Systems Biology Natalia Sanchez de Groot et al A B FFOCIagg FCCYTOagg FCCYTOagg FOCI CIagg gg YTOagg YT YTOagg YT S FTOTALagg FTTOTALsol FTTOTALsol TOTALagg T OTALsol T OTALsol T Fraction of deposited protein 1.2 -Uracil / Essential 25ºC 1.05 +Uracil / Non-essential +5FOA / Toxic Overall 0.9 Sorb. Fraction of deposited fitness of forming a loss of function due gain of function due Eq.1 DTT protein deposit to deposit formation to deposit formation 0.75 effect 25ºC 0.6 H2O2 Sorb. H2O2 37ºC 30ºC 0.45 Sorb. 30ºC cost of forming cost of losing the Factors Ura3p deposits essential Ura3p activity from Ura3p toxicity 37ºC NaCl NaCl 0.3 Pro. Pro. 30ºC 25ºC Pro. · · · S fraction of fraction of free fraction of free Eq.2 0.15 DTT NaCl H2O2 DTT deposited protein Ura3p protein Ura3p protein 0 -0.12 -0.08 -0.04 0 0.04 FFOCIagg S+Uracil - - Eq.3 “Non-essential” FTOTALagg FFFOCIagg F FCYTOagg S OCI CIagg gg FCYTOCYTOagg FCYTOagg FTTOTALagg FTOTALsol FTOTALsol FFOCIagg FCYTOagg OTALagg T TOTALsol T TOTALsol T S-Uracil - Eq.4 “Essential” FTOTALagg FTOTALsol Fraction of free protein FFOCIagg FCYTOagg S+5FOA - Eq.5 0.5 “Toxic” FTOTALagg FTOTALsol -Uracil / Essential +Uracil / Non-essential Pro. 0.4 +5FOA / Toxic Fraction of free protein C |S| average Average 0.3 25ºC free protein Pro. 25ºC Standard condition Sorb. 30ºC 30ºC 0.050 0.179 30ºC 0.2 DTT NaCl Sorb. 30ºC 30ºC H2O2 Pro. NaCl H2O2 DTT NaCl 25ºC Proline 0.005 0.360 Proline 0.1 Sorb. 37ºC H2O2 DTT 25ºC 0.006 Lower More free 0.225 25ºC selection protein 37ºC 0.0 pressure Sorbitol 0.008 0.194 Sorbitol -0.12 -0.08 -0.04 0.00 0.04 NaCl 0.005 0.162 NaCl FFFOCIagg OCI CIagg gg FCCYTOagg YTOagg YT FCCYTOagg YTOagg YT DTT 0.014 0.177 DTT S FTTOTALagg FTTOTALsol FTTOTALsol OTALagg T OTALsol T OTALsol T Higher Less free H2O2 0.019 selection protein 0.154 H2O2 pressure 37ºC 0.103 0.073 37ºC 0.4 Loss of essential activity (β) 30ºC 0.3 Deposit formation (α) α, β and γ) Average of S Loss of toxic activity (γ) (absolute values) Average of FCYTOagg/FTOTALsol 0.2 H2O2 DTT Bigger Smaller More Less 0.1 Sorb. 0.10 0.00 0.30 0.00 Sorb. 25ºC 25ºC Pro. 0.0 NaCl |S| average = (|S-URA|+|S+FOA|)/2 Pro. -0.1 D H2O2 NaCl DTT -0.2 -0.3 Environment 30ºC 37ºC -0.4 -0.5 Protein Cell state -0.12 -0.08 -0.04 0.00 0.04 Components of the model Ura3p GFP Aβ Changes in cellular FFOCIagg FOCI CIagg gg FCYTOagg CYT YTOagg FCYTOagg CYT YTOagg S FTOTALagg FTOTALsol FTOTALsol demand TOTALagg T TOTALsol T TOTALsol T Actions Function Phase for Ura3p produced or activity Uracil/ 5FU separation activity 0.08 Resulting FCYTOagg FFOCIagg Cost of loss of essential actvivity β/γ· FTOTALsol α FTOTALagg Supply Demand Cost of deposit formation 30ºC 0.04 Big cost / benefit Not a big cost H2O2 DTT Pro. 25ºC Sorb. 25ºC Pro. 0.00 NaCl Sorb. Overall Fitness H2O2 DTT NaCl -0.04 30ºC Selection detrimental neutral -0.08 37ºC -0.12 -0.12 -0.08 -0.04 0.00 0.04 Figure 3. 6 of 20 Molecular Systems Biology 15: e8075 | 2019 ª 2019 MRC Laboratory of Molecular Biology Natalia Sanchez de Groot et al Molecular Systems Biology ◀ Figure 3. Fitness cost and benefit analysis. A We measured the intensity of fluorescence emitted by the fluorescent reporter (GFP) to estimate (i) the amount, (ii) the location, and (iii) the deposited/diffused state of the model proteins. Accordingly, for each cell analyzed, we measured the fluorescence intensity of its cytoplasm (FTOTALsol for URA3sol and FCYTOagg for URA3agg) and of its foci (FFOCIagg). We used these measurements to estimate the different effects of protein phase separation on cell fitness. The overall fitness effect of forming a protein deposit depends on three factors: foci formation, loss of function, and gain of function (Equation 1). In our system, we can distinguish between the cost of forming Ura3p deposits, cost of losing the essential Ura3p activity, and the benefit from protection against Ura3p toxicity. Each cost and benefit effect depends on the ratio of deposited or free protein and the specific environment. Accordingly, we can split these cost and benefit effects into two components: one associated with the amount of deposited or free protein (FFOCIagg/FTOTALagg or FCYTOagg/FTOTALsol) and a magnitude of effect defined by the environmental conditions (a, b, or c) (Equation 2). For the non-essential, essential, and toxic roles (+Ura, Ura, and +5FOA) of Ura3p, the number and type of effects applicable are different (equations 3–5). B Plots showing the distribution and relationship between the fractions of deposited or free Ura3p, its magnitude of effect on fitness (a/b/c), and the selection coefficient (S) in different environments. C Average of the selection coefficient (left) and average of the cytosolic fluorescence (equivalent to the fractional abundance of the free protein in URA3agg, right). D Schematic representation highlighting how phase separation of the same protein can result in different fitness effects depending on the environment. different environments. Thus, in certain environments the amount Ura3p demand. Hence, the amount of free protein (ratio of FCYTOagg/ of free Ura3p is more important for cell fitness than in others (e.g., FTOTALsol, in other words the Ura3p supply) as well as the quality of 37°C and 25°C, without uracil; Fig 3B, second plot). In line with this the enzyme determines the overall fitness effect (Fig 3C and D). observation, the environments with higher average absolute selec- Given the extensive variability in fitness of a strain expressing a tion coefficient (Fig 3C, darker blue) tend to have lower average protein that forms deposits in different environments (Figs 2 and 3), apparent free protein in the URA3agg strain (Fig 3C, lighter green; we computationally investigated how the environment could influ- Pearson correlation coefficient = 0.64). ence the population frequency of variants carrying this protein at Due to the modular nature of the designed protein, a change in the population level. the environment can (i) affect the phase separation process (e.g., higher temperature can accelerate deposit formation; de Groot & History of the environments determines population frequency of Ventura, 2006) and/or (ii) affect the activity of Ura3p, for instance, strains that can form phase-separated deposits by enhancing protein folding (e.g., presence of the chemical chaper- one proline; De Los Rios & Goloubinoff, 2012) or misfolding (e.g., Our experiments were performed with a 1:1 population (URA3agg: higher levels of oxidative stress). Interestingly, environments that URA3sol), grown in an exponential phase and with no resource limi- affect protein folding and hence Ura3p activity can directly influence tation. However, in nature, variants that can form phase-separated the supply of nucleotides in a cell (Fig 3D). Moreover, the environ- structures arise sporadically in an already growing population, ment can also affect the cellular state and the growth rate and cause possibly in more severe conditions (Newby & Lindquist, 2013). To changes in the cellular demand for Ura3p activity (e.g., low temper- investigate how such a variant changes in frequency with time in a ature reduces speed of cell division, hence DNA replication and, population, and whether such a variant can take over the thus, rate of nucleotide consumption; Fig 3D, right panel). In this population in a defined timescale, we developed a computational context, it has been demonstrated that the growth rate affects cell model and simulated a scenario where one individual in a fitness and the sensitivity to environmental stresses; for instance, population of a million cells acquires a mutation that results in the rapidly growing cells tend to be more sensitive to antibiotics than formation of protein deposits (Fig 4A; Materials and Methods). slow-growing or stationary cells (Berney et al, 2006; Lu et al, 2009). Using the experimentally measured selection coefficient values (S) Thus, specific environments can influence protein folding, and (Appendix Table S1), we followed how an individual that carries a hence the total amount of free/deposited protein, and modulate the protein prone to form deposits is selected for, or against, in a cell cell state by influencing the supply/demand for a particular population in diverse scenarios. From the simulations, we infer that biochemical activity (Fig EV4). In this way, different environments in a stable environment and when deposit formation is beneficial, can modulate the magnitude of the fitness costs and benefits of the variant can become a dominant member of the population (i.e., sequestering Ura3pagg in deposits. more than 50%) within ~50–5,000 generations (Fig 4B and C, blue Overall, these results help explain why the phase separation of a color range; Appendix Table S2). As expected, if protein phase sepa- single protein can lead to different selection coefficients in the dif- ration is detrimental to cell fitness in an environment, the variants ferent environments. Specifically, in our model protein, cell fitness containing the protein that form deposits never achieve population primarily relies on the balance between the supply/demand of frequencies higher than the starting condition and get “diluted” with Ura3p activity and to a much lesser extent on the cost of deposit time (Fig 4C, orange color spectrum). formation. This means that when uracil is present in the media, the In nature, however, the growth environment can fluctuate more cell does not require Ura3p (no demand). Hence, a change in the or less rapidly over time (e.g., day/night cycles affect growth amount (free/deposited) or quality of this enzyme (active, folded/ temperatures daily; Newby & Lindquist, 2013). This means that inactive, misfolded) in a specific environment does not affect cell fit- an individual with a protein variant that can phase-separate can be ness significantly. In agreement with this possibility, in these envi- selected to different extents as determined by the sequence/ ronments, there is no correlation between the fraction of the history of environments (Sh) that the population experiences deposited protein (or free protein) and the selection coefficient (Kussell & Leibler, 2005; Leibler & Kussell, 2010; Fig 4D and (Fig 3B, first and second panels, gray dots). However, when the Appendix Table S3). In rapidly fluctuating and alternating environ- protein activity is essential or toxic (i.e., in the absence of uracil or ments, as long as the average selection coefficient over time (or presence of 5FOA), each environmental condition has an associated selection coefficient of the history of environments, Sh) is positive, ª © 2019 MRC Laboratory of Molecular Biology Molecular Systems Biology 15: e8075 | 2019 7 of 20 Molecular Systems Biology Natalia Sanchez de Groot et al A One mutant in a D t1 t2 t3 t4 wild-type population of 106 cells Sh mutant Time Time E1 E2 E1 E2 ... ? How does this S E1 . t1 + S E2 . t2 + S E1 . t3 + S E2 . t4 mutant evolve in Sh = different scenarios? 4 B E 100 100 +5FOA/NaCl +5FOA/Proline % of mutants in the population +5FOA/Sorbitol %of mutants in the population 80 80 +5FOA/DTT tn tn+1 +5FOA/H2O2 60 +5FOA/30ºC 60 +URA/NaCl +URA/H2O2 40 +5FOA/25ºC 40 +URA/30ºC +URA/37ºC 20 20 0 0 0 1000 2000 3000 4000 5000 6000 7000 0 50 100 150 200 250 300 Time (generations of WT) Time (generations of WT) C F 12000 10 2 Time to reach 50% (generations of WT) +5FOA/30ºC +5FOA/H2O2 1 +5FOA/DTT +5FOA/Sorbitol 10000 +5FOA/Proline Fraction of mutant cells 10 -2 +5FOA/NaCl +URA/NaCl +URA/30ºC 8000 10 -4 +5FOA/25ºC +URA/H2O2 +URA/37ºC 6000 10 -6 +URA/Proline +URA/DTT +URA/Sorbitol 10 -8 -URA/Sorbitol 4000 -URA/NaCl -URA/Proline 10 -10 +URA/25ºC -URA/DTT 2000 -URA/25ºC 10 -12 -URA/H2O2 -URA/30ºC -URA/37ºC 0 10 -14 1 10 10 2 10 3 10 4 10 5 10 6 0.0 .005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 Time (generations of WT) Selection coefficient of the history of environments (Sh) Figure 4. Computational simulation and estimation of population frequencies. A Schematic of the question addressed for the computational simulation. B Population frequency of the phase separation-prone variant over time (in number of generations with respect to wild-type strain at 30°C, 2.7 h) in different non- oscillating conditions with a positive measured selection coefficient (S) for protein phase separation. A dashed line indicates the point where the mutant is present in 50% of the population. The corresponding value in the x-axis for each profile denotes the time required to reach this point. C Population frequency of the phase separation-prone variant over time in different environmental conditions with positive and negative S. The dashed line indicates the initial fraction of mutant cells in the population, and the colored lines indicate the fraction of mutant cells in the population at different time points (generations of WT is used as the unit of reference). D Diagram showing how the frequency of the mutant cells can change in fluctuating environments (environment, E, and time in environment, t). The sequence of environments (history) experienced by a mutant determines the overall selection coefficient (Sh). E Population frequency of the variant over time in two different alternating environments. Black line, profile of population frequency of the variant, where, at the moment in which the first mutant cell appears, the environment positively selects it; however, in the subsequent environment the selection is negative. Gray line, profile of population frequency of the variant, where, at the moment in which the first mutant cell appears, the environment negatively selects it; however, in the subsequent environment the selection is positive. The area of the plot is divided into gray (E1) and white (E2) sections to indicate the fluctuating environment with positive and negative S. F History of environments resulting from any possible combination of two environments (e.g., “environment 1” 25°C and “environment 2″ 0.5 M proline), from those experimentally tested in this work (actual data available in Appendix Table S3). 8 of 20 Molecular Systems Biology 15: e8075 | 2019 ª 2019 MRC Laboratory of Molecular Biology Natalia Sanchez de Groot et al Molecular Systems Biology the frequency of the phase separation-prone variant in the popula- population, the history of environments (Sh) can determine the tion becomes higher over time (Fig 4E). It is worth noting that the persistence and frequency of a phase separation-prone variant frequency of the variant becomes higher irrespective of whether the (Fig 4F). immediate environment when the mutation is acquired is beneficial (Fig 4E; dark line) or detrimental (Fig 4E; light line) to fitness, as Protein phase separation: a source of phenotypic variability long as the environment fluctuates and Sh is positive. Nevertheless, at a given time, the frequency of the phase separation-prone variant Microscopy image analysis reveals the existence of stochastic, cell- can be higher if the mutation is acquired in a favorable environment to-cell variation in total protein abundance (i.e., expression noise compared to a detrimental environment (Fig 4E; dark line vs. lighter Soboleski et al, 2005; Acar et al, 2008; Jothi et al, 2009; Burga et al, line at tn and tn+1). These simulations highlight that for the same 2011; Chevin, 2011; Ravarani et al, 2016) as measured by GFP Variability in protein phase separation Variability in free protein abundance A C Fraction of deposited protein Fraction of free protein D 1.0 1.0 URA3agg / +URA / 30ºC / Non-essential Less phase separated More free protein 0.8 0.8 FCYTOagg/FTOTALagg FFOCIagg/FTOTALagg 0.6 0.6 15 μm More phase separated Less free protein 0.4 0.4 URA3sol / +URA / 30ºC / Non-essential 0.2 0.2 Average percentage of deposited and free protein for the URA3agg population 0.0 0.0 15 μm -URA +URA +5FOA -URA +URA +5FOA 0.25 0.32 0.39 FFOCI 0.75 0.68 0.61 B FTOTAL C.V. URA3agg p-value C.V. URA3sol C.V. FFOCI p-value C.V. FCYTO -URA +URA +5FOA -URA / Essential 62.44 1.2x10-159 53.12 -URA / Essential 45.85 1.7x10-66 15.67 +URA / Non-essential 69.48 9.9x10-168 62.56 +URA / Non-essential 49.74 7.9x10-106 24.27 Selection against Selection for FCYTO large foci size large foci size +5FOA / Toxic 62.51 2.1x10-91 49.47 +5FOA / Toxic 49.00 1.9x10-48 30.73 Variability in free protein abundance Variability in extent of function Diversity in phenotype Differential selection E G H URA3agg/+5FOA/30ºC I URA3agg/+5FOA/30ºC J URA3agg/+URA/30ºC URA3agg / +5FOA / 30ºC / Toxic High activity H p = 2.7x10-5 Toxic Non-essential 8e 05 8e 05 Stronger phenotype 80% of free protein 8 p= 4.5x10-4 p= 0.36 400 Density (no. of cells) Density (no. of cells) 2 20% of phase 1N FCYTOagg/area (a.u.) sseparated protein 300 15 μm 4e 05 4e 05 Moderate activity 1N 2N 200 F URA3 GFP Aβ 60% of free protein 40% of phase separated protein Weaker phenotype 100 0 0 5FU Uracil Low activity More protein Less protein 0 5x104 10x104 15x104 0 5x104 10x104 15x104 production production 30% of free protein deposited deposited DNA fluorescence (a.u.) DNA fluorescence (a.u.) 70% of phase separated protein Impairs Supports cell division cell division FFOCI≥70%FTOTAL FFOCI≤70%FTOTAL FFOCI≥70%FTOTAL FFOCI≤70%FTOTAL FFOCI≥70%FTOTAL FFOCI≤70%FTOTAL Figure 5. Impact on cell-to-cell variability and phenotypic diversity. Panels (A–D) show that the variability in foci formation contributes to variability in free protein abundance. A Confocal microscopy images of URA3agg and URA3sol incubated for 18 h at 30°C. B Coefficient of variation and P-values (Wilcoxon test) of the total fluorescence measured from populations of URA3agg and URA3sol. C Boxplot showing the distribution of foci fluorescence (left) or cytoplasmic fluorescence (right), from the foci containing subset of a population of URA3agg grown at 30°C. Coefficient of variation and P-values of each boxplot are shown below. D Schematic showing how the formation of protein deposit can affect the abundance of free protein in a cell. Pie charts showing the average fraction of deposited and free protein of URA3agg in different media compositions and at 30°C. E Confocal microscopy image of URA3agg incubated in the presence of 5FOA after 18 h at 30°C. F Schematic showing that in our system, the phase separation-promoting region can modulate Ura3p activity and thus its phenotypic effect in a cell. G Schematic highlighting that higher/lower level of free protein inside the cell results in higher/lower Ura3p activity leading to the manifestation of a stronger/weaker phenotype. H Boxplots of the cytoplasmic fluorescence normalized by the cell area (FCYTOagg/cell area) for individual cells of the URA3agg population in 5FOA. Light green box, cells with foci containing at least 70% of the total fluorescence (big foci). Dark green box, cells with foci containing less than 70% of the total fluorescence (small foci). I, J Cell cycle arrest phenotype. Density plots showing the fluorescence distribution of individual cells from the URA3agg population stained with propidium iodide to measure their DNA content. The cells were incubated for 18 h at 30°C without (I) and with 5FOA (J). Data information: (C, H) For the boxplots: the central line represents the median, the notches represent 95% confidence interval of the median, the box shows the interquartile range (IQR), the whiskers add 1.5 times the IQR to the 75 percentile (box upper limit) and subtract 1.5 times the IQR from the 25 percentile (box lower limit). We used the Wilcoxon test to compare samples. ª © 2019 MRC Laboratory of Molecular Biology Molecular Systems Biology 15: e8075 | 2019 9 of 20 Molecular Systems Biology Natalia Sanchez de Groot et al fluorescence; Fig 5A, Appendix Fig S3). Under standard growth during DNA synthesis (Fig 5F; Seiple et al, 2006). Since the forma- conditions (in which both strains present similar transcript levels; tion of phase-separated structures reduces the amount of free/active see Fig EV2), we find that the cell-to-cell variation in protein abun- protein (Fig 5D), individuals with larger amount of protein in dance is higher for URA3agg compared to URA3sol irrespective of deposits (FFOCI > 70% FTOTAL) will have little free protein and whether Ura3p activity is essential, non-essential, or toxic (as display less Ura3p activity (Fig 5G), and hence should less often measured by the coefficient of variation; CV = 100r/l; ratio of the exhibit cell cycle arrest. Similarly, individuals with smaller amount standard deviation to the mean abundance; Fig 5B). However, if we of protein in deposits will display higher Ura3p activity (Fig 5G), focus on just one genotype, regardless of URA3sol or URA3agg, we and hence should more often exhibit cell cycle arrest. In line with obtain higher noise when uracil is added. This is consistent with the this expectation, we observe that within the same population, indi- current view that the maximum population variance is achieved vidual cells with more protein recruited into phase-separated under conditions with no selection (Thompson & Cubillos, 2017). deposits tend to have less free protein after normalizing for cell area Interestingly, the cell-to-cell variation in protein abundance is higher (Fig 5H). Furthermore, DNA quantification reveals that cells with in the deposits compared to the cytosol for URA3agg irrespective of more protein in phase-separated deposits tend to display cell cycle whether Ura3p is essential non-essential or toxic (Fig 5C). These arrest less often compared to those with less protein in deposits data suggest that the extent of protein deposit formation by phase (Fig 5I). Indeed, in conditions where Ura3p activity is not essential, separation can modulate the abundance of free/active protein in a no cell cycle arrest is observed (Fig 5J). cell (Fig 5D). We find that when Ura3p activity is essential or toxic, Taken together, our results suggest that the cell-to-cell variability on average, cells with a higher fraction of phase-separated protein in the extent of protein phase separation can lead to variability in are selected against (i.e., higher free protein is selected for) or for the abundance of free, functional protein, which in turn can gener- (i.e., less free protein is selected for), respectively (Fig 5D). When ate a phenotypic continuum in a genetically identical cell population Ura3p is not essential, the average foci size is in between what is (Fig 6A). It should be noted that depending on the molecular struc- observed for the other two growth media, suggesting no obvious ture and the properties of the phase-separated assembly (e.g., insol- selection for foci size (or free protein) (Fig 5C). These results indi- uble deposits or liquid droplets), one could find an opposite cate that even within an isogenic population, the extent of deposit behavior where enzyme activity may be higher within certain types formation between individuals can be selected for or against under of phase-separated structures (Shin & Brangwynne, 2017). Never- different selective pressures. This effect is similar to what one theless, such phenotypic diversity, induced by variable protein observes in prion strain selection when yeast cells are grown under phase separation, can provide a selective advantage under certain different stress conditions (Holmes et al, 2013; Chakrabortee et al, conditions and could enhance the likelihood of survival of some 2016). individuals from a population after an environment alteration. Simi- In conditions where Ura3p activity is toxic, we observe that cells lar benefits have been attributed to the prion conformational switch. are elongated, tend to be larger, and display cell division problems An important difference, however, is that in the latter case, instead (Fig 5E). This is consistent with the fact that in the presence of of generating a phenotypic continuum, the emergence of different 5FOA, free/active Ura3p produces 5FU and leads to cell cycle arrest prion strains provides new heritable phenotypes with distinct and A Variability in Variability in Diversity in Stochastic nature of free protein foci size phenotype phase separation abundance B Environmental change Time Protein phase separation: 2-5 min Adaptation processes mRNA Synthesis: 20 min Protein Synthesis: 40 min Protein Degradation: 50 min Changes in DNA: hours-years Figure 6. Phenotype modulators. A The stochasticity of deposit formation due to phase separation will define the foci size, which in turn will influence the amount of free protein in the cytosol. The abundance of the free protein (or the phase-separated structure) will ultimately determine the cellular phenotype. The behavior and phenotypic outcome may also be influenced by the molecular structure and properties of the phase-separated structure (e.g., complex coacervates, liquid droplets). B Timeline for the different processes that are important for adaptation/response to an environmental change. Time estimates were obtained from published datasets (e.g., Wang et al, 2002; Belle et al, 2006; Escusa-Toret et al, 2013). 10 of 20 Molecular Systems Biology 15: e8075 | 2019 ª 2019 MRC Laboratory of Molecular Biology Natalia Sanchez de Groot et al Molecular Systems Biology largely non-overlapping phenotypes/functional states (Derdowski indicate that the interplay between the protein state (e.g., soluble or et al, 2010; Halfmann et al, 2010; Holmes et al, 2013; Newby & deposited) and the functional demand placed by the cell defines the Lindquist, 2013; Farkash-Amar et al, 2014; Chakrabortee et al, overall effect of protein phase separation on cell fitness. It also high- 2016; Newby et al, 2017; Fig EV5). Our observations suggest that lights that the specific growth environment is a key determinant of beyond prions, polypeptide chains with certain phase separation/ the effect of phase separation on cell fitness. Using computational aggregation/deposit-forming propensity can contribute to selectable simulations, we further showed that the frequency and persistence phenotypic heterogeneity to an isogenic cell population. In other of a phase separation-prone variant within a population are deter- words, the selectable phenotypic variability described for prions is mined by the history of environments experienced by this popula- likely to be a general property of any polypeptide segment that can tion (Fig 4G). In fact, it has been observed in nature that the history form phase-separated protein deposits. of environments (memory) can be “encoded” by aggregation-prone Q/N-rich segments (Wiltzius et al, 2009; Caudron & Barral, 2013; O’Connell et al, 2014; Zhang et al, 2015). Discussion Our observations on individual cells in an isogenic population indicate that the extent and onset of protein phase separation may Our understanding of the process of protein phase separation has vary between cells (Fig 5). While this may be influenced by dif- changed rapidly in the last decade. Currently, there is consensus ferent cellular processes (Chalancon et al, 2012), protein phase that intracellular protein condensates are associated with a wide separation might further contribute to changes in free protein level range of effects, ranging from pathological to beneficial (Aguzzi & and leads to cell-to-cell heterogeneity in free protein abundance. Altmeyer, 2016). Clearly, not all deposits have the same molecular The existence of cytoplasmic chaperones that can dissolve protein structure (Brangwynne et al, 2015; Molliex et al, 2015; Zhang et al, macromolecular assemblies such as Hsp104p (Duennwald et al, 2015; Feric et al, 2016; Lee et al, 2016). In the cellular milieu, the 2012), and variability in their abundances between cells, could different types of deposits differ in their composition, dynamics, and further contribute to the ability of the phase separation-prone cellular location. This in combination with the biochemical function protein to be released or refold from deposits to become active. Such of the molecules that participate in phase separation determines the mechanisms that generate cell-to-cell variability in abundance can overall effect on fitness (Morley et al, 2002; Gsponer & Babu, 2012; act in synergy and still be distinct from the ability of chaperones to Lin et al, 2015; Miller et al, 2015; Molliex et al, 2015; Zhang et al, help fold (i.e., buffer) deleterious mutants during or after protein 2015; Zhu & Brangwynne, 2015; Banani et al, 2016; Feric et al, synthesis (Burga et al, 2011), leading to stochastic changes in 2016; Mitrea & Kriwacki, 2016; Pak et al, 2016; Chavali et al, protein abundance between individuals. 2017b; Riback et al, 2017). To investigate how phase-separated A population should generate enough phenotypic variability to protein deposits can affect fitness and to disentangle the effects of ensure that some cells will be able to survive an unexpected envi- different factors on cell fitness, we studied a model protein with a ronmental change (Kussell & Leibler, 2005; Koonin, 2007; Acar modular architecture (Fig 1B) that is prevalent in nature (Bornberg- et al, 2008; Leibler & Kussell, 2010). In this sense, like in the case of Bauer & Alba, 2013; Lees et al, 2016; Chavali et al, 2017a). Such a a prion conformational switch, phase separation can be a general protein architecture may be widespread, possibly because it allows mechanism of adaptation by resulting in rapid and transient mani- the properties of the polypeptide segment that promotes phase sepa- festation of loss/gain of function compared to mechanisms that ration (e.g., concentration at which the protein forms deposits) to involve changes in the mRNA, or protein levels (Wang et al, 2002; evolve independently of the biochemical function that is mediated Kussell & Leibler, 2005; Belle et al, 2006; Leibler & Kussell, 2010; by the rest of the protein. Proteins without this modular architecture Escusa-Toret et al, 2013; Fig 6B). In this context, variation in the may also tune their aggregation propensity; however, since function formation of deposits, and variable stoichiometry of the deposits and aggregation propensity are linked in such proteins, evolution of between individuals of a population can further generate phenotypic their sequences will likely be more constrained. diversity (Fig 6A). In other words, since deposit formation is depen- Our work is consistent with previous studies (Geiler-Samerotte dent on protein concentration (Veis, 2011; Ciryam et al, 2013; et al, 2011; Escusa-Toret et al, 2013; Tomala et al, 2014) and Brangwynne et al, 2015), different individuals can form phase-sepa- provides quantitative insights as well as a framework to decouple rated assemblies at different abundances by “sensing” their concen- the fitness cost and benefit of protein phase separation, and the tration in cells (Gsponer & Babu, 2012). This may lead to variable associated loss/gain of protein function in different environments. and rapid manifestation of phenotypes by transiently sequestering Overall, the data suggest that in our system, yeast cells can tolerate proteins into deposits (or restricting their localization in a cell) in and are adapted to intracellular protein phase separation of the different environments. Interestingly, in most cases this phenom- model protein that we have designed. In future, this modular archi- enon is triggered because proteins have cellular concentrations tecture can be adapted to assess the specific contribution of different above their saturation level (Ciryam et al, 2013). For these supersat- types of phase separation-promoting sequences on cell fitness. In urated proteins, formation of such deposits can be protective when our system, the effect on fitness is primarily driven by either the loss toxic concentrations are reached in a cell (Bershtein et al, 2012). or gain of protein function as a consequence of sequestering the The modular nature of the phase transition-promoting segment free/active/functional protein into insoluble deposits (Fig 3C). also helps with phenotypic diversity and the manifestation of new Furthermore, the environment can modulate the consequences of phenotypes during evolution, since such segments can evolve inde- protein phase separation in strains with the same genomic back- pendently of the biochemical function mediated by the rest of the ground, enabling situations where protein deposit formation is polypeptide (Bornberg-Bauer & Alba, 2013; Boke et al, 2016; Lees either beneficial or detrimental to fitness (Fig 2C). These results et al, 2016). This may possibly explain why many regulators such ª © 2019 MRC Laboratory of Molecular Biology Molecular Systems Biology 15: e8075 | 2019 11 of 20 Molecular Systems Biology Natalia Sanchez de Groot et al as transcription factors (e.g., Snf5) contain phase separation- A multicloning site flanked by a 65-bp region that is homologous promoting segments (e.g., Q/N-rich segments) independently of to TRP1 (TRP start vector) was designed and introduced in a pMA structured domains that perform specific biochemical function such vector (GeneArt Life Technologies) (Appendix, Vectors and as DNA binding (Chavali et al, 2017a; Chong et al, 2018; Sabari Primers). The TRP1 locus was selected as the insertion position for et al, 2018). It is known that cell-to-cell variation in protein abun- the proteins (Ura3psol and Ura3pagg), because of its close proximity dance of such regulators can lead to differential activation of down- to the centromere and the high expression levels of genes in this stream regulatory networks and drive the phenotypic differences in region. These properties minimize the frequency of gene silencing an isogenic population (Jothi et al, 2009; Halfmann et al, 2012; and ensure a steady state expression level through several genera- Holmes et al, 2013; Gemayel et al, 2015). Thus, variability in tions. A yeast-optimized GFP codon and the SpHis5 selection deposit formation between individuals can lead to cell-to-cell dif- marker were extracted from a pKT128 vector (Sheff & Thorn, 2004) ferences in phenotypes by affecting the abundance of the free/active and inserted between the restriction sites PacI and SacI. The pGAL1 protein, as observed with foci size variability, and cell cycle arrest promoter was obtained from a pESC-URA vector (Agilent Technolo- phenotype in the URA3agg population (Fig 5). gies) and inserted between HindIII and SalI. The same vector was The findings presented here suggest that in addition to variability used to extract the locus encoding the URA3 gene plus an additional in the process of gene expression, and protein folding between indi- 12-residue linker (Linker 1: GGTACCGCTAGTGGTTCTGCTGGTTCT viduals of a population (Kussell & Leibler, 2005; Acar et al, 2008; GCGATTAAC). URA3 was inserted between pGAL1 and GFP using Jothi et al, 2009; Leibler & Kussell, 2010; Burga et al, 2011; Chevin, the In-Fusion HD Cloning Kit (Clontech) (TRP-URA vector). 2011; Holmes et al, 2013; Ravarani et al, 2016), phase separation- The sequence of Ab42 was optimized for yeast using the codon induced cell-to-cell variability in free protein abundance is likely to usage reported in the Codon Usage Database (http://www.kazusa. be an important phenomenon that can facilitate adaptability in dif- or.jp/codon/) and was built using four DNA oligos. Ab42 was ferent conditions. Thus, studies on the effects of phase-separated inserted 30 prime of the GFP (TRP-URA-AB vector) using the In- structures such as intracellular protein deposition should not only Fusion HD Cloning Kit (Clontech) and included a 12-residue linker involve a global investigation of the cost and benefit but also in between (Linker 2: GGTGGAAGTGCTAATGGTACTTCTGGTGC consider cell-to-cell variability of phenotypes in the population TAGTGGT). (Farkash-Amar et al, 2014). Both linkers were designed to be yeast codon optimized and Inside the cell, a myriad of proteins phase-separate by different enriched in small amino acids to provide flexibility and avoid steric mechanisms. However, the cellular complexity does not end here hindrances. Additionally, these sequences have a low net charge since these assemblies can interact with each other, and with other and no aggregation-prone regions to prevent interference with the molecules and are subjected to a constant transformation due to the phase separation process triggered by Ab42. This second property quality control machinery (Farkash-Amar et al, 2014; Brangwynne was measured using the AGGRESCAN algorithm (Conchillo-Sole et al, 2015; Lin et al, 2015; Nott et al, 2015; Zhang et al, 2015; Zhu et al, 2007; de Groot et al, 2012). The cellular behavior of Ab42 has & Brangwynne, 2015; Banani et al, 2016; Bolognesi et al, 2016; Feric been extensively studied in bacteria and yeast, both by others and et al, 2016; Pak et al, 2016; Wu & Fuxreiter, 2016). Overall, in an by us (de Groot & Ventura, 2006; de Groot et al, 2006; Morell et al, in vivo system, phase separation can affect the cellular processes at 2011; Villar-Pique et al, 2012; Sanchez de Groot et al, 2015). In multiple levels and result in different fitness costs and benefits. In yeast, the formation of deposits/granules appears to protect cells our system, the protein model phase-separates from a largely against the toxic species (e.g., soluble oligomers; de Groot & soluble active state to a primarily insoluble inactive state. Obvi- Ventura, 2006; Villar-Pique & Ventura, 2013; Sanchez de Groot et al, ously, in more complex systems (e.g., dynamic assemblies, active 2015). compartments, complex coacervates) there are more variables to be studied than those analyzed here. However, the quantitative Rationale and design of model proteins insights disentangling the various effects, as well as the conceptual and methodological framework presented in this study, may be In this work, we designed model proteins to study the effects of adapted (e.g., by testing phase separation-promoting sequences with protein phase separation on cell fitness. With this aim, we designed different molecular structures and properties, adding new elements, a polypeptide with a modular arrangement, similar to those recently and in different genetic backgrounds) to understand the effect on found in nature (McGlinchey et al, 2011; Kato et al, 2012; Kedersha cell fitness of phase separation of diverse proteins and how this et al, 2013; Sleeman & Trinkle-Mulcahy, 2014; Fig 1C). This setup determines their selection within a population. allows us to tune different properties while studying the cost/benefit of the phase separation event (see main text). To build our model proteins, we chose the yeast endogenous Materials and Methods Ura3p enzyme as the functional domain. This provides a versatile framework for assessing the different roles of the enzyme, as the Strains and vectors biochemical activity can be essential, non-essential, or toxic depend- ing on the composition of the growth medium. All the strains employed in this work are based on Y03157 (BY4741; To report the expression, distribution, and location of the fusion Mat a; his3D1; leu2D0; met15D0; ura3D0; YBR020w::kanMX4) protein, we added the enhanced GFP (F64L/S65T) as the reporter. obtained from Euroscarf deletion collection. In this strain, the GAL1 This GFP variant includes a mutation at the position 64 (F>L) that gene is missing, thereby hindering cells to consume galactose (the allows its correct production at 37°C (Cormack et al, 1996; Day & inducer) as a carbon source (Geiler-Samerotte et al, 2011). Davidson, 2009). This extra stability is crucial for the present work 12 of 20 Molecular Systems Biology 15: e8075 | 2019 ª 2019 MRC Laboratory of Molecular Biology Natalia Sanchez de Groot et al Molecular Systems Biology to ensure the integrity of the reporter under stress conditions. strain. The microscopy images demonstrate that there is still cytoso- Importantly, as reported in earlier studies in E. coli, even under lic fluorescence in the Ura3pagg strain (Figs 5 and EV1). These data strong overexpression, the folding of this enhanced GFP fused with suggest that the fitness effects associated with deposit formation of phase separation-promoting peptide is faster than the phase separa- Ura3p are smaller than those obtained with a true auxotroph tion process itself. This property results in the formation of fluores- (Mulleder et al, 2012). cent foci at a temperature range between 25 and 37°C (de Groot & For the mathematical modeling of this system (see Selection coef- Ventura, 2006; de Groot et al, 2006). Furthermore, to distinguish ficient calculation), we have considered some generalizations for between the effect of the environment and foci formation on the the analysis of the different parameters influencing the phase sepa- steady state abundance of the model protein (GFP fluorescence), ration event and its effect on cell fitness. We have subtracted the our system includes a soluble control (URA3sol) to indicate the protein exchange between the phase-separated assembly and the maximum amount of active GFP that can be made available in a cell surrounding milieu from our analyses based on two experimental at a certain environment. Hence, in the unlikely event in which a observations. First, FRAP analyses show that Ura3pagg forms a GFP molecule will misfold due to reasons not associated with the stable, non-dynamic deposit similar to IPODs (Fig EV3). Second, phase separation process, the normalization against the control our analyses are done keeping the culture at exponential phase and strain will account for such a situation. after long expression times (from 18 h to 3 days). This setup allows Finally, Ab42 was added as a tag to promote phase separation Ab42-GFP deposits to reach an equilibrium in which most cells and study the effect of deposit formation. This peptide is the main contain just one big focus (see Fig 5) and the amount of protein component of the plaques found in patients with Alzheimer’s synthesized/degraded appears to be constant (Morell et al, 2011). disease (Chiti & Dobson, 2006; Olzscha et al, 2011; Ciryam et al, Thus, in these conditions we consider that the cells reach a “popula- 2013; Pak et al, 2016; Woerner et al, 2016). In vitro, Ab42 assem- tion equilibrium” in which some properties of the culture are bles into amyloid fibrils driven by its hydrophobic residues. When constant: (i) The number of newborn cells is constant (exponential overexpressed in unicellular models, this peptide accumulates into phase); (ii) the cells are adapted to the environment (after 18–72 h), inclusion bodies (de Groot & Ventura, 2006; de Groot et al, 2006; and therefore, the environmental pressure/selection is constant; and Plata et al, 2010; Morell et al, 2011; Villar-Pique et al, 2012; Escusa- (iii) considering assumptions (i) and (ii), the noise distribution of Toret et al, 2013; Villar-Pique & Ventura, 2013; Sanchez de Groot Ura3p expression and the number of foci-forming cells in the popu- et al, 2015; Pak et al, 2016). Ab42 has no intrinsic function in lation remain stable (Morell et al, 2011; Sanchez de Groot et al, S. cerevisiae and, according to our previous studies, its expression 2015). This premise also considers that the cell-to-cell heterogeneity has almost no effect on yeast growth, at least under optimal grow in foci formation is a final outcome of the inherent variability of conditions (Villar-Pique & Ventura, 2013; Sanchez de Groot et al, cellular processes such as protein expression and quality control 2015). Consistent with this, it is worth emphasizing that we activity (e.g., chaperones; Chalancon et al, 2012; Farkash-Amar measured no significant difference in growth that was associated et al, 2014; Ravarani et al, 2016). with the addition of Ab42 in conditions where Ura3p is non-essen- Under the microscope, Ura3psol seems homogeneously distrib- tial and no stress is introduced (Fig 2). An alternative strategy to uted throughout the cytosol whereas part of Ura3pagg is accumu- induce the phase separation of Ura3p could be the alteration of its lated into foci. It has been reported that not all the protein in the sequence with the aim to destabilize the protein and to induce foci soluble fraction is active (e.g., soluble oligomers) and that not all formation. However, this mutant variant can also affect Ura3p struc- the aggregate protein is inactive (i.e., fluorescence into foci; Villar- ture and enzymatic activity, increasing the complexity of the system Pique & Ventura, 2013). However, at least in our system, due to the and making it more difficult to determine the origin of the cell fit- static nature of the analyzed foci, the Ura3p recruited in them will ness alterations. We also considered the addition of a less aggrega- have limited access to its substrate (irrespective of its conformation tion-prone Ab42 peptide; however, all point mutations studied have and functional state). In addition, theoretically, the free Ura3p a similar growth rate (Villar-Pique & Ventura, 2013; Sanchez de should have less constraints to acquire a correct fold than the Groot et al, 2015), and to obtain a fully soluble variant, multiple protein located in the foci (Morell et al, 2011). Overall, based on mutations are necessary (Wurth et al, 2002). Therefore, to gener- these premises, we assume that the protein located in the cytosolic ate the soluble version of the model protein, we used the version fraction is more active and accessible to substrate than the protein without a tag. In short, we designed the model protein in a modu- located in the assembly, so we consider that the residual activity lar manner with the addition of a phase separation-promoting tag located in the foci is negligible. It should be noted that depending (i.e., separate functional and aggregation-prone region in the same on the nature of the phase-separated structure (e.g., liquid droplets), polypeptide). one could find an opposite behavior where enzyme activity may be With the above-mentioned elements, we built two different higher within certain types of phase-separated structures (Shin & constructs: (i) One was designed to control for the intrinsic fitness Brangwynne, 2017). cost of overexpressing a soluble protein (Plata et al, 2010; Geiler- Samerotte et al, 2011; Kafri et al, 2016) (Ura3p fused to GFP, Media composition and environmental conditions Ura3psol), and (ii) the other one was designed to investigate the effect of protein deposit formation (Ura3p fused to GFP and Ab42 In all the assays, the strains were grown in SD -HIS media contain- peptide, Ura3pagg, Fig 1D). By comparing the overall cell fitness ing a mixture of sugars and amino acids. Since protein concentra- upon expressing these two constructs, we obtain the fitness effects tion is a critical factor for foci formation, to guarantee high levels of of cells expressing the phase-separated Ura3pagg. It is worth expression we measured the effect of galactose concentration (the mentioning that the obtained phenotype is not that of a full null inducer) on URA3sol expression by monitoring the GFP fluorescence ª © 2019 MRC Laboratory of Molecular Biology Molecular Systems Biology 15: e8075 | 2019 13 of 20 Molecular Systems Biology Natalia Sanchez de Groot et al using flow cytometry (Appendix Fig S2D). Maximum levels of foci was subtracted from the total fluorescence to obtain the amount expression were obtained between 0.5 and 2%. Below 0.5% of of soluble protein in each cell. The cell size analysis shows a positive galactose, the fluorescence declines abruptly, so we chose 1% as a correlation with the foci size (Appendix Fig S3), probably due to reliable concentration to keep high levels of induction. To obtain a aging and yeast asymmetric division, which retains protein deposits fresh culture before inducing Ura3p expression, single colonies were in mother cells (which tend to bigger) producing clean daughter cells picked and grown overnight in 2% glucose. This culture was (which tend to be smaller) where there protein assembly has to start employed to inoculate SD -HIS in 2% raffinose and grown for 6 h. de novo (Coelho et al, 2014). For the propidium iodide assay, we Then, it was inoculated in fresh media with 2% raffinose and 1% adapted the macro to additionally measure the fluorescence in the galactose. SD -His -Ura was employed to test the essentiality of propidium iodide channel for each element detected. Ura3p. SD -His with uracil and 5FOA (Zymo Research) was used to analyze the phase separation effects of a toxic Ura3p activity. Immunochemistry In all cases, the media were adjusted to pH 4.5 to ensure (when required) the 5FOA permeability. To test different environments, Yeast cells were grown for 18 h in media with 2% raffinose and 1% the different compounds were added before the pH adjustment. The galactose. 20 ml of culture was divided in two and centrifuged. For final concentrations tested are 0.5 mM H2O2, 1.5 mM DTT, 0.5 M total fraction, a pellet was resuspended in 75 ml of lysis buffer NaCl, 1 M sorbitol and 0.5 M proline. The cells were grown at (10 mM Tris–HCl, pH 1/4 8, 150 mM NaCl, 0.05% Tween-20, 10% 220 rpm and 30°C before starting the experiments to test the effect glycerol, 5 mM EDTA, 1 mM DTT, 2 mM PMSF) and 25 ml of a specific environment. Subsequently, the cultures were incu- NuPAGE LDS sample buffer with 2.5% 2-mercaptoethanol (w/w) bated at 30° with the corresponding concentration of the above- and incubated at 100°C for 5 min. For fraction separation, a pellet mentioned chemical or without the chemical but shifting the was resuspended in 75 ml Y-PER yeast protein extraction reagent temperature to 25° or 37°C. (Thermo Scientific) supplemented with 0.1 mM PMSF and incu- bated at room temperature with agitation for 20 min. Then, the Cell cycle arrest sample was centrifuged to separate the two fractions: soluble (su- pernatant) and insoluble (pellet resuspended in 75 ml of PBS). Cells grown until exponential phase were fixed by incubating them 25 ml of NuPAGE LDS sample buffer with 2.5% 2-mercaptoethanol for 1 h with ethanol 70% at 4°C. Then, the cells were treated with (w/w) was added to both fractions and then incubated at 100°C. To 1 mg/ml RNase A (Sigma) for 2 h at 37°C. The samples were separate the proteins, 5 ml of the total fraction and 10 ml of soluble stained with propidium iodide (50 lg/ml) for 1 h before analyzing or insoluble fractions were eluted into a precast NOVEX NuPAGE 4– them under a confocal microscope. The fluorescence of propidium 12% gels in denaturing conditions. The Invitrogen iBlot system was iodide was employed as a measure of DNA content and an indica- used to transfer proteins to PVDF membranes. After blocking, tion of the cell cycle arrest. membranes were incubated overnight at 4°C with anti-GFP rabbit antibody (Santa Cruz sc-8334) or anti-PGK1 mouse antibody (Novex Confocal microscopy 459250) diluted 1:1,000 and 1:10,000, respectively. Secondary incu- bation with anti-Protein G HRP conjugate (Millipore 18-161) at Cells were grown at exponential phase with 1% galactose and the 1:10,000 was performed at RT for 1 h. Images were taken with an corresponding environmental condition for 18–20 h before acquir- Amersham Imager 600. ImageJ 2.0.0-rc-48 software was used to ing the images. The images were acquired with a Zeiss 710 (Carl quantify protein bands. Zeiss) with an objective of 63×, an excitation laser of 488 nm, and emission window between 581 and 750 nm. The cells expressing In vivo half-life measurement soluble Ura3p (Ura3psol) displayed fluorescence that is homoge- neously distributed through the cytoplasm, whereas the cells that Yeast cells were grown for 18 h in media with 2% raffinose and additionally express the Ab (Ura3pagg) showed fluorescence in foci. 1% galactose. Protein production was then stopped by cleaning At least 100 cells were captured for each strain and environment. the cells and changing the media to SD -HIS with 2% glucose. All For the propidium iodide assay, the fluorescence was excited with samples were adjusted to the same concentration before monitor- 561 nm and the emission was collected between 566 and 719 nm. ing the turbidity (absorbance, 600 nm) and fluorescence (excita- tion 480 nm, emission 510 nm) with a TECAN Infinite 200. Image processing Samples were agitated for 10 s before each measurement and during 450 s between time points. The fluorescence was measured All the images were processed with Fiji (ImageJ). For each cell, the as the ratio between fluorescence and turbidity at every time point. program measured the cell area, the cytosolic fluorescence, and the To calculate the ratio of fluorescence loss, the data were fitted to a fluorescence contained in the aggregates. A macro was created to Boltzmann’s sigmoid with GraphPad PRISM 5 software (GraphPad perform these measures automatically. The parameters of this macro Software). for a whole cell analyze the presence of elements with (i) a fluores- cent intensity between 20 and 255 a.u., (ii) an area (size) between 8 Competition experiments and 50 lm2, and (iii) a circularity between 0.25 and 1. To detect foci, a mask was generated including the elements with a fluorescent Cells pre-induced with 1% galactose were grown in SD -His over- intensity between 40 and 255 a.u. and an area between 0.2 and night and then inoculated into fresh media for 4 h to reach exponen- 10 lm2. For the cells expressing Ura3pagg, the fluorescence from the tial phase. Then, the OD600 nm of the cultures was measured, and the 14 of 20 Molecular Systems Biology 15: e8075 | 2019 ª 2019 MRC Laboratory of Molecular Biology Natalia Sanchez de Groot et al Molecular Systems Biology two strains were mixed in equal concentration (1:1) and this mixture where P is the proportion of a certain strain in the population, Nagg was employed to inoculate the different media. At this point, we took is the number of cells of URA3agg, and Nsol is the number of cells a sample for time 0. At least two different duplicates of each culture of URA3sol at time t. were analyzed. The cells were kept at exponential phase to guaran- The proportion of each strain at a certain time point is: tee a stable duplication time and to minimize the number of old and 0 dead cells. For each culture, samples were collected at six different Nagg Nagg 2t-agg Nagg 0 ¼ 0 t- ¼ 0 2tð-agg -sol Þ (2) time points (0, 17, 34, 51, 58, 75, and 82 h). At each time point, the Nsol Nsol 2 sol Nsol cells were centrifuged and the genomic DNA was extracted. After taking the samples, a dilution was performed to maintain the where N0 is the initial concentration (concentration at time 0) and cultures at OD600 nm between 0.005 and 0.6. x is the growth rate. The measurement could be simplified using To quantify the proportion of each strain, a pair of oligos (solF/ the base 2 logarithm and adding the difference between the growth solR and aggF/aggR) was designed to obtain a specific product close ratio of the two strains, which can be obtained from the competi- to 100 bp (Appendix, Vectors and Primers). The oligos to quantify tion experiments: URA3sol anneal between URA3 and GFP. For URA3agg, the primers 0 anneal specifically to the Ab sequence. The specificity and efficiency Nagg Nagg log2 ¼ log2 0 þ tð-agg  -sol Þ ! of each pair of oligos were tested against the plasmids containing the Nsol N inserted cassettes (Fig EV2). A control PCR with an empty plasmid   sol (3) d Nagg (TRP start) produced no amplification product. For each qPCR assay, log2 ¼ -agg  -sol dt Nsol two standard slopes of plasmids encoding for Ura3psol and Ura3pagg     (pESC-URA and TRP-URA-AB vectors) were included in the same d 1 Nagg 1 d Nagg S¼ log2 ! S¼ log2 (4) plate to correct for PCR efficiency and measure the number of gene dt log2 e Nsol log2 e dt Nsol copies (Fig EV2). The DDCT method was employed to analyze each qPCR (Teste et al, 2009). Log2 of the ratio of the strains (URA3agg/ By combining (3) and (4), the selection coefficient could be URA3sol) was plotted. The resultant slope corresponds to the dif- calculated as: ference between the growth rates of the two strains. Before the optimization of the above-described method of growth 1  S¼ -agg  -sol (5) measurement, we also tried to implement other approaches previ- log2 e ously reported as highly efficient to measure the effects of protein misfolding on cell fitness (Geiler-Samerotte et al, 2011). However, Population frequency simulation several experimental problems appeared after altering the environ- mental conditions. For example, we were not able to follow by flow The doubling times (s) employed in the simulations were cytometry the competition between yeast strains expressing a interpolated from the selection coefficient values (S) and the experi- soluble and insoluble protein variant tagged with different fluores- mentally measured doubling times of URA3sol (the wild-type) cent markers (i.e., YFP and GFP) at 37°C or 0.5 mM H2O2. The (Appendix Table S1; Appendix Table S2). extreme growth conditions employed influenced the marker’s fold- To simulate the time evolution of the mutant fraction in a ing and fluorescence spectrum, impeding the correct identification mixed population, we started with a mixed population consisting of the populations. of one mutant cell (URA3agg) and 106 wild-type cells (URA3sol). In the simulation, we used the logarithmic scale to handle large Doubling time numbers, but the final numbers were converted to linear scale whenever needed. We used an Euler integration scheme (with URA3sol and URA3agg were grown separately in 96-well plates. The Dt = 60 min) to numerically calculate number of cells, N (t), at turbidity (OD600 nm) and fluorescence (450 nm excitation and any given time, t: 510 nm emission) were recorded at 72 h with a Tecan Infinite M200 Dt Pro. Before starting the assay, the cells were pre-induced with ln ðNðtÞÞ ¼ ln ðNðt  DtÞÞ þ ln 2 (6) s SD -His 2% raffinose–1% galactose overnight and then incubated in fresh media for 4 h to reach exponential phase. The time-course where s is the doubling time (s = 1/ϖ). To simulate alternating assay was started with an OD600 nm of 0.02 (Cary 50 Bio). A spline environments, we used the same integration scheme as above, but interpolation approach was applied with the RStudio program to the doubling times were re-assigned to match a specific environ- measure the slope of each growth curve, resulting in the doubling ment, i.e., every 12 h. All simulations were performed in time s (Appendix Table S2). MATLAB. Selection coefficient calculation Measurement of the effect of protein phase separation The selection coefficient is defined as in Chevin (Chevin, 2011): In our experiments, the effects of Ura3p phase separation on cell fitness are determined by three factors: (i) the cost of deposit   d P d Nagg formation, (ii) the cost of losing the essential Ura3p biochemical Selection ¼ S ¼ ln ¼ ln (1) dt 1P dt Nsol activity, and (iii) the benefit of gaining a protective function by ª © 2019 MRC Laboratory of Molecular Biology Molecular Systems Biology 15: e8075 | 2019 15 of 20 Molecular Systems Biology Natalia Sanchez de Groot et al FFOCIagg FCYTOagg sequestering Ura3p in the deposit. The effect of foci formation is S ¼ a þ c ! FTOTALagg FTOTALsol comprised of the cost due to the sequestration of material and   (9) FFOCIagg FTOTALsol consumption of energy (e.g., amino acid sequestration, ATP- c ¼ S  a  FTOTALagg FCYTOagg dependent chaperone activity). The other two factors are associ- ated with protein function that existed before foci formation (loss of function) and the generation of new beneficial effect after foci The three effects can be integrated in one general equation that formation (gain of protection against the toxic function). In our could be applied for any of the environments and Ura3p roles experiments, the loss of function leads to a decrease in pyrimidi- (essential, non-essential, toxic) analyzed (Fig 3): nes (in the absence of uracil), whereas the gain of protective func- FFOCIagg FCYTOagg FCYTOagg tion is due to the prevention of toxic 5-fluorouracil production (in S ¼ a þ b þ c (10) FTOTALagg FTOTALsol FTOTALsol the presence of 5FOA). We consider that each effect is determined by the amount of phase-separated protein (FFOCIagg/FTOTALagg), the amount of soluble protein (FCYTOagg/FTOTALsol), and a coefficient of mRNA expression levels proportionality that captures the effect of the specific environment (a, b, c) (Fig 3, see main text). These coefficients indicate the 5 ml of fresh cells grown in 1% galactose was centrifuged and magnitude of the effect on fitness of forming the deposit and loss suspended in 1 ml of 0.2 M lithium acetate–1% SDS solution. After and gain of protein function due to phase separation in a particu- 5 min at 70°C, 3 ml of TRIzol was added. Next, the RNA was lar environment. extracted with TRIzol Reagent following the instructions provided For a particular environment, we consider a to remain the same by Life Technologies, and the concentration of purified mRNA was irrespective of whether Ura3p is non-essential, essential, or toxic measured. Retrotranscription was performed using the RevertAid H (i.e., when grown in +Uracil, Uracil, or +5FOA). In conditions Minus First Strand cDNA Synthesis Kit (Thermo scientific) and the where Ura3p activity is neither essential nor toxic, we consider that random hexamer primers included with the kit. The concentration there is no loss or gain of function and that the differences between of the cDNA generated was adjusted before running the qPCR exper- URA3agg and URA3sol are a consequence of the phase separation iment. The reactions were done using the SYBR Green PCR Master process. Thus, the cost of protein deposit formation due to phase Mix (Life technologies) and a ViiA7 (Applied Biosystems). The pair transition could be expressed as: of primers (FmRNA/RmRNA) was designed using Primer BLAST to anneal to the GFP region and not to the yeast genome (Appendix, Vectors and Primers). The mRNA of a reference gene (ALG9) was FFOCIagg SFTOTALagg S ¼ a ! a¼ (7) measured to normalize the data. The primers to amplify this gene FTOTALagg FFOCIagg were obtained from Teste et al (2009). Transcript quantification was done using the DCt method where the variation between where FFOCIagg is the fluorescence contained in the deposits and URA3sol and URA3agg was calculated as DDCt. The measured fold FCYTOagg is the florescence from the cytoplasm. The ratio quantifies change between URA3sol and URA3agg is 1.07. the extent of phase separation. a is the magnitude of the effect of forming the deposit in specific environment. By considering that a remains the same for a specific environ- Data availability ment when grown in the different growth media, we can estimate the magnitude of effect for the loss (b) and gain (c) of function We provide all relevant datasets in the Appendix. when grown in the absence of uracil and in the presence of 5FOA, respectively. Therefore, when Ura3 activity is essential, the magni- Expanded View for this article is available online. tude of effect of the loss of protein function due to phase separation can be expressed as: Acknowledgements This work was supported by the Medical Research Council (MC_U105185859; FFOCIagg SFCYTOagg M.M.B., M.T., C.R., and N.S.G.), the Center for Models of Life through Danish S ¼ a þ b ! FTOTALagg FTOTALsol National Research Foundation (A.T.), Marie Curie Actions (FP7-PEOPLE-2012-   (8) FFOCIagg FTOTALsol IEF-330352, to M.T.; and FP7-PEOPLE-2011-IEF299105, to N.S.G.), FEBS Long- b ¼ S  a  FTOTALagg FCYTOagg Term Fellowships (N.S.G.), Beatriu de Pinos fellowships (M.T.), and the Ministe- rio de Economía y Competitividad (SAF2017-82158-R, SAF2015-72518-EXP, and where FTOTALsol is the fluorescence of the cells expressing Ura3psol RYC-2012-09999; M.T.). N.S.G. is a recipient of the MRC Centenary Award. measured in the corresponding environment. FTOTALsol represents M.M.B. is a Lister Institute Research Prize Fellow. We thank Pavithra Chavali, the 100% of active enzyme that is available to a cell in a particular Xiaohan Li, Greet De Baets, Natasha Latysheva, Sreenivas Chavali, Guilhem environment for a non-aggregation-prone Ura3psol. FCYTOagg/ Chalancon, Louis Maddox, and Anthony Fitzpatrick for critically reading this FTOTALsol measures the amount of activity that is lost in the aggre- paper. We thank Gian Gaetano for supporting N.S.G. and Sean Munro for the gating strain. gift of strain Y03157. In conditions where Ura3 activity is toxic (i.e., uracil and 5FOA are added in the medium), the magnitude of effect of the Author contributions gain of protection against the toxic activity (c) could be calculated MMB, NSG, and SV conceived the project. MMB and NSG designed the project as below: and wrote the manuscript. MTB and CNJR helped with experimental design. 16 of 20 Molecular Systems Biology 15: e8075 | 2019 ª 2019 MRC Laboratory of Molecular Biology Natalia Sanchez de Groot et al Molecular Systems Biology NSG, MTB performed the experiments, analyzed the data, and performed the Chakrabortee S, Byers JS, Jones S, Garcia DM, Bhullar B, Chang A, She R, Lee L, computational and statistical analyses. AT performed the computational simu- Fremin B, Lindquist S, Jarosz DF (2016) Intrinsically disordered proteins lations. NSG led the project and MMB supervised the project. All authors were drive emergence and inheritance of biological traits. Cell 167: involved in discussion and preparation of the final manuscript. 369 – 381.e312 Chalancon G, Ravarani CN, Balaji S, Martinez-Arias A, Aravind L, Jothi R, Babu Conflict of interest MM (2012) Interplay between gene expression noise and regulatory The authors declare that they have no conflict of interest. network architecture. Trends Genet 28: 221 – 232 Chavali S, Chavali PL, Chalancon G, de Groot NS, Gemayel R, Latysheva NS, Ing-Simmons E, Verstrepen KJ, Balaji S, Babu MM (2017a) Constraints and References consequences of the emergence of amino acid repeats in eukaryotic proteins. Nat Struct Mol Biol 24: 765 – 777 Acar M, Mettetal JT, van Oudenaarden A (2008) Stochastic switching as a Chavali S, Gunnarsson A, Babu MM (2017b) Intrinsically disordered proteins survival strategy in fluctuating environments. Nat Genet 40: adaptively reorganize cellular matter during stress. Trends Biochem Sci 42: 471 – 475 410 – 412 Aguzzi A, Altmeyer M (2016) Phase separation: linking cellular Chevin LM (2011) On measuring selection in experimental evolution. Biol Lett compartmentalization to disease. Trends Cell Biol 26: 547 – 558 7: 210 – 213 Alberti S, Gladfelter A, Mittag T (2019) Considerations and challenges in Chiti F, Dobson CM (2006) Protein misfolding, functional amyloid, and studying liquid-liquid phase separation and biomolecular condensates. human disease. Annu Rev Biochem 75: 333 – 366 Cell 176: 419 – 434 Chong S, Dugast-Darzacq C, Liu Z, Dong P, Dailey GM, Cattoglio C, Heckert A, Babu MM, van der Lee R, de Groot NS, Gsponer J (2011) Intrinsically Banala S, Lavis L, Darzacq X, Tjian R (2018) Imaging dynamic and selective disordered proteins: regulation and disease. Curr Opin Struct Biol 21: low-complexity domain interactions that control gene transcription. 432 – 440 Science 361: eaar2555 Banani SF, Rice AM, Peeples WB, Lin Y, Jain S, Parker R, Rosen MK (2016) Ciryam P, Tartaglia GG, Morimoto RI, Dobson CM, Vendruscolo M (2013) Compositional control of phase-separated cellular bodies. Cell 166: Widespread aggregation and neurodegenerative diseases are associated 651 – 663 with supersaturated proteins. Cell Rep 5: 781 – 790 Belle A, Tanay A, Bitincka L, Shamir R, O’Shea EK (2006) Quantification of Coelho M, Lade SJ, Alberti S, Gross T, Tolic IM (2014) Fusion of protein protein half-lives in the budding yeast proteome. Proc Natl Acad Sci USA aggregates facilitates asymmetric damage segregation. PLoS Biol 12: 103: 13004 – 13009 e1001886 Berchowitz LE, Kabachinski G, Walker MR, Carlile TM, Gilbert WV, Schwartz Conchillo-Sole O, de Groot NS, Aviles FX, Vendrell J, Daura X, Ventura S TU, Amon A (2015) Regulated formation of an amyloid-like translational (2007) AGGRESCAN: a server for the prediction and evaluation of “hot repressor governs gametogenesis. Cell 163: 406 – 418 spots” of aggregation in polypeptides. BMC Bioinformatics 8: Berney M, Weilenmann HU, Ihssen J, Bassin C, Egli T (2006) Specific growth 65 rate determines the sensitivity of Escherichia coli to thermal, UVA, and Cormack BP, Valdivia RH, Falkow S (1996) FACS-optimized mutants of the solar disinfection. Appl Environ Microbiol 72: 2586 – 2593 green fluorescent protein (GFP). Gene 173: 33 – 38 Bershtein S, Mu W, Shakhnovich EI (2012) Soluble oligomerization provides a Day RN, Davidson MW (2009) The fluorescent protein palette: tools for beneficial fitness effect on destabilizing mutations. Proc Natl Acad Sci USA cellular imaging. Chem Soc Rev 38: 2887 – 2921 109: 4857 – 4862 De Baets G, Reumers J, Delgado Blanco J, Dopazo J, Schymkowitz J, Rousseau Boeynaems S, Alberti S, Fawzi NL, Mittag T, Polymenidou M, Rousseau F, F (2011) An evolutionary trade-off between protein turnover rate and Schymkowitz J, Shorter J, Wolozin B, Van Den Bosch L, Tompa P, Fuxreiter protein aggregation favors a higher aggregation propensity in fast M (2018) Protein phase separation: a new phase in cell biology. Trends degrading proteins. PLoS Comput Biol 7: e1002090 Cell Biol 28: 420 – 435 De Los Rios P, Goloubinoff P (2012) Protein folding: chaperoning protein Boke E, Ruer M, Wuhr M, Coughlin M, Lemaitre R, Gygi SP, Alberti S, Drechsel evolution. Nat Chem Biol 8: 226 – 228 D, Hyman AA, Mitchison TJ (2016) Amyloid-like self-assembly of a cellular Dekel E, Alon U (2005) Optimality and evolutionary tuning of the expression compartment. Cell 166: 637 – 650 level of a protein. Nature 436: 588 – 592 Bolognesi B, Lorenzo Gotor N, Dhar R, Cirillo D, Baldrighi M, Tartaglia GG, Derdowski A, Sindi SS, Klaips CL, DiSalvo S, Serio TR (2010) A size threshold Lehner B (2016) A concentration-dependent liquid phase separation limits prion transmission and establishes phenotypic diversity. Science 330: can cause toxicity upon increased protein expression. Cell Rep 16: 680 – 683 222 – 231 Derkatch IL, Chernoff YO, Kushnirov VV, Inge-Vechtomov SG, Liebman SW Bornberg-Bauer E, Alba MM (2013) Dynamics and adaptive benefits of (1996) Genesis and variability of [PSI] prion factors in Saccharomyces modular protein evolution. Curr Opin Struct Biol 23: 459 – 466 cerevisiae. Genetics 144: 1375 – 1386 Brangwynne CP, Tompa P, Pappu RV (2015) Polymer physics of intracellular Dobson CM (1999) Protein misfolding, evolution and disease. Trends Biochem phase transitions. Nat Phys 11: 899 – 904 Sci 24: 329 – 332 Burga A, Casanueva MO, Lehner B (2011) Predicting mutation outcome from Duennwald ML, Echeverria A, Shorter J (2012) Small heat shock proteins early stochastic variation in genetic interaction partners. Nature 480: potentiate amyloid dissolution by protein disaggregases from yeast and 250 – 253 humans. PLoS Biol 10: e1001346 Caudron F, Barral Y (2013) A super-assembly of Whi3 encodes memory of Escusa-Toret S, Vonk WI, Frydman J (2013) Spatial sequestration of misfolded deceptive encounters by single cells during yeast courtship. Cell 155: proteins by a dynamic chaperone pathway enhances cellular fitness 1244 – 1257 during stress. Nat Cell Biol 15: 1231 – 1243 ª © 2019 MRC Laboratory of Molecular Biology Molecular Systems Biology 15: e8075 | 2019 17 of 20 Molecular Systems Biology Natalia Sanchez de Groot et al Farkash-Amar S, Zimmer A, Eden E, Cohen A, Geva-Zatorsky N, Cohen L, Milo Kafri M, Metzl-Raz E, Jona G, Barkai N (2016) The cost of protein production. R, Sigal A, Danon T, Alon U (2014) Noise genetics: inferring protein Cell Rep 14: 22 – 31 function by correlating phenotype with protein levels and localization in Kaganovich D, Kopito R, Frydman J (2008) Misfolded proteins partition individual human cells. PLoS Genet 10: e1004176 between two distinct quality control compartments. Nature 454: Feric M, Vaidya N, Harmon TS, Mitrea DM, Zhu L, Richardson TM, Kriwacki 1088 – 1095 RW, Pappu RV, Brangwynne CP (2016) Coexisting liquid phases underlie Kato M, Han TW, Xie S, Shi K, Du X, Wu LC, Mirzaei H, Goldsmith EJ, nucleolar subcompartments. Cell 165: 1686 – 1697 Longgood J, Pei J, Grishin NV, Frantz DE, Schneider JW, Chen S, Li L, Franzmann TM, Jahnel M, Pozniakovsky A, Mahamid J, Holehouse AS, Nuske Sawaya MR, Eisenberg D, Tycko R, McKnight SL (2012) Cell-free formation E, Richter D, Baumeister W, Grill SW, Pappu RV, Hyman AA, Alberti S of RNA granules: low complexity sequence domains form dynamic fibers (2018) Phase separation of a yeast prion protein promotes cellular fitness. within hydrogels. Cell 149: 753 – 767 Science 359: eaao5654 Kedersha N, Ivanov P, Anderson P (2013) Stress granules and cell signaling: Geiler-Samerotte KA, Dion MF, Budnik BA, Wang SM, Hartl DL, Drummond DA more than just a passing phase? Trends Biochem Sci 38: 494 – 506 (2011) Misfolded proteins impose a dosage-dependent fitness cost and Khan MR, Li L, Perez-Sanchez C, Saraf A, Florens L, Slaughter BD, Unruh JR, Si trigger a cytosolic unfolded protein response in yeast. Proc Natl Acad Sci K (2015) Amyloidogenic oligomerization transforms Drosophila Orb2 from a USA 108: 680 – 685 translation repressor to an activator. Cell 163: 1468 – 1483 Gemayel R, Chavali S, Pougach K, Legendre M, Zhu B, Boeynaems S, van der Koonin EV (2007) Chance and necessity in cellular response to challenge. Mol Zande E, Gevaert K, Rousseau F, Schymkowitz J, Babu MM, Verstrepen KJ Syst Biol 3: 107 (2015) Variable glutamine-rich repeats modulate transcription factor Kussell E, Leibler S (2005) Phenotypic diversity, population growth, and activity. Mol Cell 59: 615 – 627 information in fluctuating environments. Science 309: 2075 – 2078 de Groot NS, Aviles FX, Vendrell J, Ventura S (2006) Mutagenesis of the Lee KH, Zhang P, Kim HJ, Mitrea DM, Sarkar M, Freibaum BD, Cika J, central hydrophobic cluster in Abeta42 Alzheimer’s peptide. Side-chain Coughlin M, Messing J, Molliex A, Maxwell BA, Kim NC, Temirov J, properties correlate with aggregation propensities. FEBS J 273: Moore J, Kolaitis RM, Shaw TI, Bai B, Peng J, Kriwacki RW, Taylor JP 658 – 668 (2016) C9orf72 dipeptide repeats impair the assembly, dynamics, and de Groot NS, Castillo V, Grana-Montes R, Ventura S (2012) AGGRESCAN: function of membrane-less organelles. Cell 167: 774 – method, application, and perspectives for drug design. Methods Mol Biol 788.e717 819: 199 – 220 Lees JG, Dawson NL, Sillitoe I, Orengo CA (2016) Functional innovation from de Groot NS, Ventura S (2006) Effect of temperature on protein quality in changes in protein domains and their combinations. Curr Opin Struct Biol bacterial inclusion bodies. FEBS Lett 580: 6471 – 6476 38: 44 – 52 Gsponer J, Babu MM (2012) Cellular strategies for regulating functional and Leibler S, Kussell E (2010) Individual histories and selection in heterogeneous nonfunctional protein aggregation. Cell Rep 2: 1425 – 1437 populations. Proc Natl Acad Sci USA 107: 13183 – 13188 Halfmann R, Alberti S, Lindquist S (2010) Prions, protein homeostasis, and Levy ED, Kowarzyk J, Michnick SW (2014) High-resolution mapping of protein phenotypic diversity. Trends Cell Biol 20: 125 – 133 concentration reveals principles of proteome architecture and adaptation. Halfmann R, Jarosz DF, Jones SK, Chang A, Lancaster AK, Lindquist S (2012) Cell Rep 7: 1333 – 1340 Prions are a common mechanism for phenotypic inheritance in wild Li J, McQuade T, Siemer AB, Napetschnig J, Moriwaki K, Hsiao YS, Damko E, yeasts. Nature 482: 363 – 368 Moquin D, Walz T, McDermott A, Chan FK, Wu H (2012) The RIP1/RIP3 Hervas R, Li L, Majumdar A, Fernandez-Ramirez Mdel C, Unruh JR, Slaughter necrosome forms a functional amyloid signaling complex required for BD, Galera-Prat A, Santana E, Suzuki M, Nagai Y, Bruix M, Casas-Tinto S, programmed necrosis. Cell 150: 339 – 350 Menendez M, Laurents DV, Si K, Carrion-Vazquez M (2016) Molecular Lin Y, Protter DS, Rosen MK, Parker R (2015) Formation and maturation of basis of Orb2 amyloidogenesis and blockade of memory consolidation. phase-separated liquid droplets by RNA-binding proteins. Mol Cell 60: PLoS Biol 14: e1002361 208 – 219 Hill SM, Hao X, Gronvall J, Spikings-Nordby S, Widlund PO, Amen T, Jorhov A, Lu C, Brauer MJ, Botstein D (2009) Slow growth induces heat-shock Josefson R, Kaganovich D, Liu B, Nystrom T (2016) Asymmetric inheritance resistance in normal and respiratory-deficient yeast. Mol Biol Cell 20: of aggregated proteins and age reset in yeast are regulated by Vac17- 891 – 903 dependent vacuolar functions. Cell Rep 16: 826 – 838 Maharana S, Wang J, Papadopoulos DK, Richter D, Pozniakovsky A, Poser I, Holehouse AS, Pappu RV (2018) Functional implications of intracellular phase Bickle M, Rizk S, Guillen-Boixet J, Franzmann TM, Jahnel M, Marrone L, transitions. Biochemistry 57: 2415 – 2423 Chang YT, Sterneckert J, Tomancak P, Hyman AA, Alberti S (2018) RNA Holmes DL, Lancaster AK, Lindquist S, Halfmann R (2013) Heritable buffers the phase separation behavior of prion-like RNA binding proteins. remodeling of yeast multicellularity by an environmentally responsive Science 360: 918 – 921 prion. Cell 153: 153 – 165 Maji SK, Perrin MH, Sawaya MR, Jessberger S, Vadodaria K, Rissman RA, Jahn TR, Radford SE (2008) Folding versus aggregation: polypeptide confor- Singru PS, Nilsson KP, Simon R, Schubert D, Eisenberg D, Rivier J, mations on competing pathways. Arch Biochem Biophys 469: 100 – 117 Sawchenko P, Vale W, Riek R (2009) Functional amyloids as natural Jain S, Wheeler JR, Walters RW, Agrawal A, Barsic A, Parker R (2016) ATPase- storage of peptide hormones in pituitary secretory granules. Science 325: modulated stress granules contain a diverse proteome and substructure. 328 – 332 Cell 164: 487 – 498 Marsh JA, Teichmann SA (2010) How do proteins gain new domains? Genome Jothi R, Balaji S, Wuster A, Grochow JA, Gsponer J, Przytycka TM, Aravind L, Biol 11: 126 Babu MM (2009) Genomic analysis reveals a tight link between McGlinchey RP, Yap TL, Lee JC (2011) The yin and yang of amyloid: insights transcription factor dynamics and regulatory network architecture. Mol from alpha-synuclein and repeat domain of Pmel17. Phys Chem Chem Syst Biol 5: 294 Phys 13: 20066 – 20075 18 of 20 Molecular Systems Biology 15: e8075 | 2019 ª 2019 MRC Laboratory of Molecular Biology Natalia Sanchez de Groot et al Molecular Systems Biology Miller SB, Mogk A, Bukau B (2015) Spatially organized aggregation of Riback JA, Katanski CD, Kear-Scott JL, Pilipenko EV, Rojek AE, Sosnick TR, misfolded proteins as cellular stress defense strategy. J Mol Biol 427: Drummond DA (2017) Stress-triggered phase separation is an adaptive. 1564 – 1574 Evolutionarily tuned response. Cell 168: 1028 – 1040.e1019 Mitrea DM, Kriwacki RW (2016) Phase separation in biology: functional Rothe S, Prakash P, Tyedmers J (2018) The insoluble protein deposit (IPOD) in organization of a higher order. Cell Commun Signal 14: 1 yeast. Front Mol Neurosci 11: 237 Molliex A, Temirov J, Lee J, Coughlin M, Kanagaraj AP, Kim HJ, Mittag T, Sabari BR, Dall’Agnese A, Boija A, Klein IA, Coffey EL, Shrinivas K, Abraham BJ, Taylor JP (2015) Phase separation by low complexity domains promotes Hannett NM, Zamudio AV, Manteiga JC, Li CH, Guo YE, Day DS, Schuijers J, stress granule assembly and drives pathological fibrillization. Cell 163: Vasile E, Malik S, Hnisz D, Lee TI, Cisse II, Roeder RG et al (2018) 123 – 133 Coactivator condensation at super-enhancers links phase separation and Monsellier E, Ramazzotti M, Taddei N, Chiti F (2008) Aggregation propensity gene control. Science 361: eaar3958 of the human proteome. PLoS Comput Biol 4: e1000199 Sanchez de Groot N, Torrent M, Villar-Pique A, Lang B, Ventura S, Gsponer J, Moore AD, Bjorklund AK, Ekman D, Bornberg-Bauer E, Elofsson A (2008) Babu MM (2012) Evolutionary selection for protein aggregation. Biochem Arrangements in the modular evolution of proteins. Trends Biochem Sci 33: Soc Trans 40: 1032 – 1037 444 – 451 Sanchez de Groot N, Gomes RA, Villar-Pique A, Babu MM, Coelho AV, Ventura Morell M, de Groot NS, Vendrell J, Aviles FX, Ventura S (2011) Linking amyloid S (2015) Proteome response at the edge of protein aggregation. Open Biol protein aggregation and yeast survival. Mol BioSyst 7: 1121 – 1128 5: 140221 Morley JF, Brignull HR, Weyers JJ, Morimoto RI (2002) The threshold for Seiple L, Jaruga P, Dizdaroglu M, Stivers JT (2006) Linking uracil base polyglutamine-expansion protein aggregation and cellular toxicity is excision repair and 5-fluorouracil toxicity in yeast. Nucleic Acids Res 34: dynamic and influenced by aging in Caenorhabditis elegans. Proc Natl Acad 140 – 151 Sci USA 99: 10417 – 10422 Sheff MA, Thorn KS (2004) Optimized cassettes for fluorescent protein Mulleder M, Capuano F, Pir P, Christen S, Sauer U, Oliver SG, Ralser M (2012) tagging in Saccharomyces cerevisiae. Yeast 21: 661 – 670 A prototrophic deletion mutant collection for yeast metabolomics and Shin Y, Brangwynne CP (2017) Liquid phase condensation in cell physiology systems biology. Nat Biotechnol 30: 1176 – 1178 and disease. Science 357: eaaf4382 Newby GA, Lindquist S (2013) Blessings in disguise: biological benefits of Sleeman JE, Trinkle-Mulcahy L (2014) Nuclear bodies: new insights prion-like mechanisms. Trends Cell Biol 23: 251 – 259 into assembly/dynamics and disease relevance. Curr Opin Cell Biol 28: Newby GA, Kiriakov S, Hallacli E, Kayatekin C, Tsvetkov P, Mancuso CP, 76 – 83 Bonner JM, Hesse WR, Chakrabortee S, Manogaran AL, Liebman SW, Soboleski MR, Oaks J, Halford WP (2005) Green fluorescent protein is a Lindquist S, Khalil AS (2017) A genetic tool to track protein aggregates and quantitative reporter of gene expression in individual eukaryotic cells. control prion inheritance. Cell 171: 966 – 979.e918 FASEB J 19: 440 – 442 Nott TJ, Petsalaki E, Farber P, Jervis D, Fussner E, Plochowietz A, Craggs TD, Stepanenko OV, Povarova OI, Sulatskaya AI, Ferreira LA, Zaslavsky BY, Bazett-Jones DP, Pawson T, Forman-Kay JD, Baldwin AJ (2015) Phase Kuznetsova IM, Turoverov KK, Uversky VN (2016) Protein unfolding in transition of a disordered nuage protein generates environmentally crowded milieu: what crowding can do to a protein undergoing responsive membraneless organelles. Mol Cell 57: 936 – 947 unfolding? J Biomol Struct Dyn 34: 2155 – 2170 O’Connell JD, Tsechansky M, Royall A, Boutz DR, Ellington AD, Marcotte EM Suraweera A, Munch C, Hanssum A, Bertolotti A (2012) Failure of amino acid (2014) A proteomic survey of widespread protein aggregation in yeast. Mol homeostasis causes cell death following proteasome inhibition. Mol Cell BioSyst 10: 851 – 861 48: 242 – 253 Olzscha H, Schermann SM, Woerner AC, Pinkert S, Hecht MH, Tartaglia GG, Suresh HG, da Silveira Dos Santos AX, Kukulski W, Tyedmers J, Riezman H, Vendruscolo M, Hayer-Hartl M, Hartl FU, Vabulas RM (2011) Amyloid-like Bukau B, Mogk A (2015) Prolonged starvation drives reversible aggregates sequester numerous metastable proteins with essential cellular sequestration of lipid biosynthetic enzymes and functions. Cell 144: 67 – 78 organelle reorganization in Saccharomyces cerevisiae. Mol Biol Cell 26: Pak CW, Kosno M, Holehouse AS, Padrick SB, Mittal A, Ali R, Yunus AA, Liu DR, 1601 – 1615 Pappu RV, Rosen MK (2016) Sequence determinants of intracellular phase Teste MA, Duquenne M, Francois JM, Parrou JL (2009) Validation of reference separation by complex coacervation of a disordered protein. Mol Cell 63: genes for quantitative expression analysis by real-time RT-PCR in 72 – 85 Saccharomyces cerevisiae. BMC Mol Biol 10: 99 Patel A, Malinovska L, Saha S, Wang J, Alberti S, Krishnan Y, Hyman AA (2017) Thompson DA, Cubillos FA (2017) Natural gene expression variation studies in ATP as a biological hydrotrope. Science 356: 753 – 756 yeast. Yeast 34: 3 – 17 Pena MI, Davlieva M, Bennett MR, Olson JS, Shamoo Y (2010) Evolutionary Tomala K, Pogoda E, Jakubowska A, Korona R (2014) Fitness costs of minimal fates within a microbial population highlight an essential role for protein sequence alterations causing protein instability and toxicity. Mol Biol Evol folding during natural selection. Mol Syst Biol 6: 387 31: 703 – 707 Petrovska I, Nuske E, Munder MC, Kulasegaran G, Malinovska L, Kroschwald Toretsky JA, Wright PE (2014) Assemblages: functional units formed by S, Richter D, Fahmy K, Gibson K, Verbavatz JM, Alberti S (2014) Filament cellular phase separation. J Cell Biol 206: 579 – 588 formation by metabolic enzymes is a specific adaptation to an advanced Veis A (2011) A review of the early development of the thermodynamics of state of cellular starvation. Elife 3: e02409 the complex coacervation phase separation. Adv Colloid Interface Sci 167: Plata G, Gottesman ME, Vitkup D (2010) The rate of the molecular clock and 2 – 11 the cost of gratuitous protein synthesis. Genome Biol 11: R98 Villar-Pique A, de Groot NS, Sabate R, Acebron SP, Celaya G, Fernandez- Ravarani CN, Chalancon G, Breker M, de Groot NS, Babu MM (2016) Affinity Busquets X, Muga A, Ventura S (2012) The effect of amyloidogenic and competition for TBP are molecular determinants of gene expression peptides on bacterial aging correlates with their intrinsic aggregation noise. Nat Commun 7: 10417 propensity. J Mol Biol 421: 270 – 281 ª © 2019 MRC Laboratory of Molecular Biology Molecular Systems Biology 15: e8075 | 2019 19 of 20 Molecular Systems Biology Natalia Sanchez de Groot et al Villar-Pique A, Ventura S (2013) Protein aggregation propensity is a crucial Wu H, Fuxreiter M (2016) The structure and dynamics of higher-order determinant of intracellular inclusion formation and quality control assemblies: amyloids, signalosomes, and granules. Cell 165: 1055 – 1066 degradation. Biochim Biophys Acta 1833: 2714 – 2724 Wurth C, Guimard NK, Hecht MH (2002) Mutations that reduce aggregation Wallace EW, Kear-Scott JL, Pilipenko EV, Schwartz MH, Laskowski PR, Rojek of the Alzheimer’s Abeta42 peptide: an unbiased search for the sequence AE, Katanski CD, Riback JA, Dion MF, Franks AM, Airoldi EM, Pan T, determinants of Abeta amyloidogenesis. J Mol Biol 319: 1279 – 1290 Budnik BA, Drummond DA (2015) Reversible, specific, active aggregates Xiang S, Kato M, Wu LC, Lin Y, Ding M, Zhang Y, Yu Y, McKnight SL (2015) of endogenous proteins assemble upon heat stress. Cell 162: 1286 – The LC domain of hnRNPA2 adopts similar conformations in hydrogel 1298 polymers, liquid-like droplets, and nuclei. Cell 163: 829 – 839 Wang Y, Liu CL, Storey JD, Tibshirani RJ, Herschlag D, Brown PO (2002) Zhang H, Elbaum-Garfinkle S, Langdon EM, Taylor N, Occhipinti P, Bridges AA, Precision and functional specificity in mRNA decay. Proc Natl Acad Sci USA Brangwynne CP, Gladfelter AS (2015) RNA controls PolyQ protein phase 99: 5860 – 5865 transitions. Mol Cell 60: 220 – 230 Wiltzius JJ, Landau M, Nelson R, Sawaya MR, Apostol MI, Goldschmidt L, Zhu L, Brangwynne CP (2015) Nuclear bodies: the emerging biophysics of Soriaga AB, Cascio D, Rajashankar K, Eisenberg D (2009) Molecular nucleoplasmic phases. Curr Opin Cell Biol 34: 23 – 30 mechanisms for protein-encoded inheritance. Nat Struct Mol Biol 16: 973 – 978 License: This is an open access article under the Woerner AC, Frottin F, Hornburg D, Feng LR, Meissner F, Patra M, Tatzelt J, terms of the Creative Commons Attribution 4.0 Mann M, Winklhofer KF, Hartl FU, Hipp MS (2016) Cytoplasmic protein License, which permits use, distribution and reproduc- aggregates interfere with nucleocytoplasmic transport of protein and RNA. tion in any medium, provided the original work is Science 351: 173 – 176 properly cited. 20 of 20 Molecular Systems Biology 15: e8075 | 2019 ª 2019 MRC Laboratory of Molecular Biology