Glossary CPC chromosomal passenger complex C sat saturating concentration FRAP fluorescence recovery after photobleaching IDP intrinsically disordered protein IDR intrinsically disordered regions KD dissociation constant k off dissociation rate constant MVP multivalent protein NoLS nucleolar localization signal NOR nucleolar organizer region NPC nuclear pore complex NTR nuclear transport receptor Nups nucleoporins PRM proline-rich motif PS phase separation RNP ribonucleoprotein SAC spindle assembly checkpoint SSI site-specific interaction Introduction

Many functions of eukaryotic cells are confined within compartments that are not delimited by membranes (and therefore often referred to as membraneless). Examples that have been known for many decades, among a number of others, are the centrosome, the nucleolus, and P-granules (Courchaine et al, 2016; Banani et al, 2017; Shin & Brangwynne, 2017; Ditlev et al, 2018; Lyon et al, 2021). With linear dimensions in the biological mesoscale (between ~100 nm and a few micrometers), compartments are supra-molecular, i.e., larger than their individual macromolecular components. Typically, tens or even hundreds of different macromolecular species, usually in multiple copies, populate each individual compartment and interact in it, conferring different degrees of internal order on compartments (Wu & Fuxreiter, 2016; Goetz & Mahamid, 2020; Peran & Mittag, 2020; Fare et al, 2021; Korkmazhan et al, 2021).

The physicochemical drivers of the biogenesis, maintenance, and disassembly of compartments, and their ability to concentrate macromolecules without an encapsulating membrane, have attracted considerable interest (Banani et al, 2017; Shin & Brangwynne, 2017). Molecular mechanistic biology, with its focus on the identification, characterization, and visualization of discrete and highly specific interactions of macromolecules, has uncovered how the building blocks of compartments assemble and interact reciprocally (see Fig 1A–E for a few examples) (Jones & Thornton, 1996; Sudha et al, 2014). How the building blocks interact to give rise to mesoscale structures devoid of strong internal order, however, is less well understood and harder to model.

Figure 1. Examples of site-specific interactions and their combination in MVPs

(A–E) Cartoon diagrams of various protein domains and proteins discussed in the text (yellow), with their cognate ligands shown in stick. (A) Src homology 3 (SH3) domain with proline-rich motif (PRM); (B) Src homology 2 domain with tyrosine-phosphorylated peptide; (C) Tudor domain with peptide from histone H3 trimethylated on lysine 36; (D) Importin-β with Gly-Leu-Gly-Phe (GLGF) peptide; (E) Enlargement of area boxed in D. The protein data bank (PDB) codes are indicated; (F) Domain organization of the NCK and N-WASP proteins, and sequence of the proline-rich region of N-WASP (Uniprot, human sequence); and (G) Artificial multivalent constructs used by Li et al (2012).

In recent years, a burgeoning new field of research in cell biology, biomolecular phase separation (PS), has promoted a radically different explanation for how membraneless compartments assemble. Under the PS paradigm, classical specific macromolecular interactions are attributed a secondary role in compartment assembly. Compartments are rather proposed to form under the action of one or a few specific PS drivers (almost invariably proteins). Drawing concepts from polymer chemistry (Hyman et al, 2014; Brangwynne et al, 2015), PS drivers are treated as associative polymers that phase-separate thanks to transient, low-affinity, cohesive interactions (Fig 2B and C) (Tanaka, 2011; Hyman et al, 2014; Brangwynne et al, 2015; Uversky et al, 2015; Banani et al, 2017; Shin & Brangwynne, 2017; Boeynaems et al, 2018; Choi et al, 2020).

Figure 2. Examples of IDPs and low-affinity, non-site-specific interactions

(A) Domain organization of three IDPs and specific sequence stretches from each of three low-complexity regions of human FUS; (B) Examples of low-affinity interactions believed to drive homotypic PS of IDPs; (C) Droplets of FUS, DDX4, and LAF-1 (indicated as “mesoscale”) are proposed to arise from multiple nanoscale interactions shown in B; (D) Sticker-and-spacer model for interactions of MVPs (left) and of IDPs (right). Gln and Tyr (Q and Y, respectively) are considered stickers that could interact through dipoles and stacking. The entire figure is an adaptation of Fig 2 from Brangwynne et al (2015).

The PS paradigm proposes that membraneless compartments are (usually) liquid like and that they arise from liquid–liquid (LL) demixing, a process exemplified by the spontaneous, thermodynamically driven separation of oil from water caused by poor solvation (Hyman et al, 2014; Banani et al, 2017). This idea was ignited by provocative observations that P-granules and nucleoli form droplets that fuse, wet neighboring cellular structures, drip, and display rapid internal mobility of its components when assayed by fluorescence recovery after photobleaching (FRAP) (Phair & Misteli, 2000; Brangwynne et al, 2009, 2011). Subsequent studies on two classes of molecules frequently identified in membraneless compartments, multivalent proteins (MVPs; Fig 1F and G) and intrinsically disordered proteins (IDPs), containing low-complexity domains (LCDs, examples of which are shown in Fig 2A), began to provide a molecular and theoretical underpinning to the PS paradigm. For instance, natural or engineered MVPs were shown to form droplets in vitro when highly concentrated (Li et al, 2012; Banani et al, 2016; Harmon et al, 2017), and similar observations were also made with various IDPs [e.g., (Kato et al, 2012; Elbaum-Garfinkle et al, 2015; Lin et al, 2015; Molliex et al, 2015; Nott et al, 2015; Pak et al, 2016)]. These observations supported the progressive identification of MVPs and IDPs as scaffolds driving PS through weak homotypic interactions (i.e., involving the same macromolecule), while other macromolecules attracted to the same compartments through heterotypic interactions (i.e., involving different types of macromolecules) were identified as clients (Hyman & Simons, 2012; Banani et al, 2016, 2017). Collectively, these developments caused the transformation of PS from the abstractions of a useful analogy to a seemingly coherent and unifying paradigm for the biogenesis of membraneless compartments from the nano- to the mesoscale (Hyman & Brangwynne, 2011; Banani et al, 2017).

Macromolecular interactions

Below, we will often refer to three main types of macromolecular interactions. The first (type I) are homotypic or heterotypic interactions established (mainly) by IDPs, i.e., by proteins, or segments thereof, that do not adopt a stable folded conformation. IDPs are usually viewed as arrays of stickers and spacers, with spacers determining overall solubility, and stickers mediating interactions (Tompa & Fuxreiter, 2008; Brangwynne et al, 2015; Martin & Mittag, 2018; Choi et al, 2020; Borcherds et al, 2021; Fare et al, 2021). Stickers are capable of only few, relatively low-affinity and poorly specific types of attractive interactions, including charge–charge, dipole–dipole, cation Π, and Π–Π stacking (Fig 2B).

The second and third types of macromolecular interactions (types II and III) are also homotypic or heterotypic, but are enabled by the specific three-dimensional arrangement (reflecting conformation) and the detailed chemical identity of the binding interfaces. These interactions involve at least one folded domain of a macromolecule and either (type II) a short linear segment of a target macromolecule, which may also fold locally in the process of binding, or (type III) another folded domain. While type I interactions are mediated by a limited set of attractive or repulsive bonds, with the result that their specificity is limited, macromolecular interactions of types II and III usually involve large, conserved surface patches on the interacting macromolecules that exploit the spatial and chemical complementarity of the binding interfaces as well as steric exclusion (Jones & Thornton, 1996; Aloy et al, 2003; Choi et al, 2009; van der Lee et al, 2014).

The PS idea identifies interactions of type I typically as fuzzy, whereas interactions of types II and III are now often defined stereospecific (Hyman & Brangwynne, 2011; Banani et al, 2017). The term stereospecific is used in chemistry to denote formation, recognition, or reaction of enantiomers. If used with reference to biological interactions, the term stereospecific aims to put emphasis on the property that specificity is largely dictated by the complementary geometry and chemistry of binding sites. This use of stereospecific is therefore related to, but deviates considerably from, the traditional definition of the term. For this reason, here I prefer using simply site-specific interaction (abbreviated as SSI) to refer to interactions of types II and III.

SSIs are enormously versatile, as masterly described in a classic review (Mammen et al, 1998). As a rule of thumb, they are titrated between 0.1 and 10 times the dissociation constant (KD) (Jarmoskaite et al, 2020). If the KD is sufficiently small (high affinity), correspondingly low concentrations of binders, normally well below their solubility limit, will suffice. SSIs can be easily and rapidly modulated, e.g., through post-translational modifications, without having to change the concentration of binders. KDs can be extremely low, to the point of making an interaction essentially irreversible, as in the core of many multisubunit macromolecular complexes, or relatively high for regulated interactions, and easily adjustable to non-equilibrium conditions in active processes. Thus, while in principle, both type I and type II/III interactions can reach high affinity (e.g., for type I interactions by harnessing multivalency), the hallmark that distinguishes I from II and III is the low specificity of the first and the high specificity of the latter two.

Condensates

As discussed below, the narrative that compartments are liquid, and driven by low-affinity, low-specificity fuzzy interactions has severe shortcomings. Nonetheless, it has exercised a tremendous influence on the field’s developments. Briefly, after disregarding SSIs as potential drivers of compartment biogenesis on the basis of a supposed incompatibility with the liquid-like appearance of phase-separated compartments (Hyman & Brangwynne, 2011), it licensed PS assays in which putative PS drivers were studied in perfect isolation in vitro, and usually confirmed in this presumed role (prematurely, as explained below). This allowed PS, initially invoked for the assembly of nuclear and cytoplasmic bodies involved in ribonucleoprotein (RNP) assembly and metabolism, such as nucleoli and P-granules, to claim for itself an ever-growing list of cellular territories. PS is now considered a self-evident universal driver of compartment assembly, as implied by the proposal to rename all membraneless compartments biomolecular condensates (Banani et al, 2017; Shin & Brangwynne, 2017; Snead & Gladfelter, 2019) on the basis of their ability to concentrate molecules, that they comprise biological macromolecules, and that they may arise through similar mechanisms [text in italic indicates textual citations (Banani et al, 2017)]. This definition encompasses functionally and compositionally diverse compartments like centrosomes, centromeres, kinetochores, transcriptional super-enhancers, chromatin domains, chromosomes, sites of response to DNA damage and DNA recombination, membrane-associated signaling complexes (which of course are not “membraneless” per se, but nevertheless not surrounded by a membrane), and many more (Hyman & Brangwynne, 2011; Banani et al, 2017; Shin & Brangwynne, 2017; Lyon et al, 2021).

Not unlike the term stereospecific, also the term condensate, as often used in the PS context, carries some ambiguities. Condensation is usually taken to describe the transition of a gas to a liquid, or chemical bond formation between two molecules accompanied by release of water. In PS publications, the term condensation is rather used to imply concentration of biomolecules through a phase change (Banani et al, 2017)—which may be considered a similar analogy as using condensation to describe the compaction of chromosomes upon mitotic entry. It is important to keep in mind, however, that the ability to concentrate molecules, e.g., to accelerate reactions or inhibit them through sequestration (Banani et al, 2017; Shin & Brangwynne, 2017; Lyon et al, 2021), is an attribute that does not support a role of PS more than it supports other mechanisms of biogenesis. SSIs can do that too, and better, as I will discuss. Furthermore, while concentration of specific macromolecules is clearly a feature of membraneless compartment, whether cellular compartments generally reach overall macromolecular concentrations higher than those of the surrounding cellular medium is unclear (Handwerger et al, 2005; Wei et al, 2017). Thus, collectively, condensate should not be taken as a literal descriptor of mesoscale membraneless compartments, at least in relation to our current knowledge of the processes that promote their assembly.

What drives compartment assembly?

Another question raised by the definition of biomolecular condensates is whether it is true that the similar mechanisms from which compartments are proposed to arise imply homotypic fuzzy interactions of PS scaffolds. These interactions are assumed to be the basis of PS and compartment assembly, but as we shall see, this is rather implausible and defies compelling evidence that the only truly general mechanism of compartment biogenesis involves highly specific heterotypic SSIs occurring at concentrations well below the saturating concentration (Csat) of any of the interacting components, i.e., in the one-phase regime. These concerns, to the extent that they call into question the very arguments on which the PS concept has built its success, are highly significant and require thorough motivation. In this essay, I will therefore examine two crucial questions: (i) Are low-specificity, fuzzy interactions of macromolecules as associative polymers literally the primary physicochemical driver of biogenesis of membraneless compartments in cells under physiological conditions (I will refer to this as general PS)? and (ii) Can phase transitions influence a compartment’s solvation and material properties after macromolecules have become concentrated there by more traditional binding mechanisms (I will refer to this as special or restricted PS)? Below I will first discuss arguments indicating that the answer to question 1 is (most) likely no. Later, I will also discuss why the answer to question 2 is instead likely yes, but probably for a relatively small subset of compartments. I will also clarify why demonstrating special PS, let alone general PS (if it exists in our cells), requires standards of proof that studies on PS have almost invariably ignored.

What is general PS?

General PS assumes that low-specificity homotypic interactions of one or very few PS scaffold macromolecules drive, at concentrations above their Csat, a density transition that results in the formation of coexisting dilute and dense phases, each with a fixed concentration of the scaffold (Banani et al, 2017). The dense phase may further promote concentration of other species to complete assembly of a mature compartment (Hyman et al, 2014; Banani et al, 2017). Examples adapted from Brangwynne et al (2015) are shown in Fig 2C.

As noted, MVPs and IDPs were initially identified as likely scaffolds in general PS. Their plausibility as PS drivers reflected (i) their enrichment in various membraneless compartments and (ii) their tendency, especially for the IDP class, to be aggregation prone and implicated in various degenerative pathologies (Alberti & Hyman, 2021). This ignited an extensive program of study of their phase behavior in vitro, with attempts to correlate it with their behavior in living cells and in disease. For instance, studies focused on the associative properties of tight complexes of reciprocally interacting linear MVPs, including engineered ones containing multiple SH3 domains and proline-rich motifs (PRMs), separated by flexible linker spacers (Fig 1F and G) (Li et al, 2012; Banani et al, 2016; Harmon et al, 2017; Case et al, 2019).

While IDPs may be viewed as assemblers, whose distributed linear motifs engage in SSIs with target folded domains (type II interactions) (van der Lee et al, 2014), the PS paradigm considers IDPs usually as associative polymers with alternating stickers and spacers (type I) (Choi et al, 2020). The same sticker-spacer description may be applied to MVPs, where folded domains are stickers-mediating interactions and the linkers between them are spacers (Choi et al, 2020). This description aims to underline a unifying similarity of MVPs and IDPs that might explain why both classes drive PS (Fig 2D). However, the similarity is less obvious than implied by this unifying description, as discussed more thoroughly below, because the stickers in MVPs usually are SSIs that can generate very significant binding affinities and specificity, whereas the stickers of IDPs are relatively low affinity and non-specific, at least in non-aggregative states (Wu & Fuxreiter, 2016; Korkmazhan et al, 2021).

The unintuitive general PS

Fascinating as they may seem, the PS tenets prove on closer scrutiny less intuitive than the simple oil–vinegar analogy suggests. Proteomes may have evolved to maximize specificity and minimize aggregation of proteins (Zhang et al, 2008; Johnson & Hummer, 2011). Neither specificity nor aggregation seem to be of significant concern for proposed general PS mechanisms of condensate biogenesis, which rather focus on low-affinity and low-specificity interactions (Fig 2) at or above the solubility limit of one or more PS drivers (Hyman et al, 2014; Brangwynne et al, 2015). It is hard to imagine how fuzzy interactions of very low specificity would promote selective, thermodynamically driven condensation of any particular macromolecule at typical cellular concentrations, with myriad competing interactions of a similar kind. And how can the same limited set of weak, non-site-specific interactions explain how different compartments maintain their identities and prohibit or enable co-PS? Are active processes involved in this complicated scheme, and if so, how? How is the saturating concentration of every putative PS driver adjusted across different specialized cells with different sizes and compositions of their cytosol, and even more so across different organisms? And how would the formation of toxic aggregates be controlled given the proposed generality of this mechanism for compartment biogenesis? Let us also consider that compartments usually form around defined spatial and temporal cues, delimited by the presence of a primary scaffold (Fig 3), and ideally do not extend beyond them. These cues, be it signaling-active transmembrane receptors or centromeres, attract and concentrate the correct targets almost certainly through SSIs with adequate affinity and specificity, probably also dictating the final concentration of various compartment’s components. If PS drivers were indeed sufficient for PS in the absence of spatial cues, why do they not assemble unrestrained in space and time?

Figure 3. Hierarchy of compartment assembly

Compartments are usually hierarchical. Nephrin, as an element of membrane clusters, is a transmembrane protein whose intracellular domain undergoes regulated phosphorylation. As primary scaffold, it acts as a binding site for the NCK client through its SH2 domain. N-WASP binds NCK through phosphorylation. Regulators of actin polymerization may be further recruited to clusters. During mitosis, the CPC kinase complex is recruited to specific phosphorylated residues of histone H2A and H3 that are enriched in the centromere region. It is therefore a client of the centromere, although it has been suggested to act as scaffold in PS. The kinetochore assembles on the histone H3 variant CENP-A, which is part of a specialized centromeric histone complex. Constitutive centromere-associated network (CCAN) subunits are recruited through interactions with CENP-A. The Knl1-Mis12-Ndc80 (KMN) complex is recruited to CCAN through established SSIs. Not shown are tertiary clients like the spindle checkpoint proteins BUB1 and MAD2 discussed in the text. In Cajal bodies, specific RNAs transcripts, for instance, of histones, but also spliceosome subunits, act as scaffolds for the recruitment of a variety of downstream proteins (Kaiser et al, 2008; Shevtsov & Dundr, 2011). The recruitment hierarchy remains poorly understood. Other nuclear bodies, like the histone body or the nucleolus, also require transcription (see main text for details).

Liquid and site specific

Thus, general PS may seem to counter biological intuition when it proposes an inverted hierarchy of compartment assembly that prioritizes unspecific interactions over more probable site-specific ones (Hyman & Brangwynne, 2011). If in all evidence macromolecules usually become concentrated in their target cellular locales due to highly site-specific interactions, why were these not considered as a mechanism of biogenesis? The primary answer is that they were perceived as incompatible with the liquid-like behavior of supposed phase-separated organelles (Hyman & Brangwynne, 2011). For instance, the kinetics of recovery of many components of compartments in FRAP experiments was deemed too fast for SSIs (Hyman & Brangwynne, 2011). This conclusion appears premature. The kinetics of typical SSIs, as they may be captured by half-time of the bound state at equilibrium, t1/2, are well within boundaries demonstrated by FRAP experiments on liquid-like compartments [for a more thorough and informative discussion on the execution and interpretation of FRAP experiments, including more sophisticated protocols to assess the potential presence of PS, please consult Erdel et al (2020; McSwiggen et al (2019b); Sprague and McNally (2005); Taylor et al (2019) and references therein]. If a macromolecule bound to a receptor in a compartment (rather than getting there through condensation), its FRAP recovery rates would be typically determined by the dissociation rate constant (koff), i.e., by the rate of release of the bleached molecules from their receptor, necessary for replacement with new fluorescent molecules (Sprague & McNally, 2005). The faster the bleached molecules dissociate from their receptor, the faster the recovery.

For example, the spindle assembly checkpoint (SAC) complex BUB1/BUB3 binds its kinetochore receptor with a dissociation constant (KD) of approximately 100 nM and a t1/2 of recovery in FRAP experiments of ~10 s (Overlack et al, 2015, 2017) (Fig 4A and B). Similarly, FRAP recovery rates and fractions for another SAC protein at kinetochores, MAD2, were initially measured in living cells and then reproduced with recombinant proteins in a reconstituted system in vitro, with typical t1/2 of recovery of a few seconds (Howell et al, 2004; Shah et al, 2004; Vink et al, 2006). Other kinetochore components (indicated as “core kinetochore” in Fig 4A) remain connected to the underlying chromatin throughout a cell’s life, without ever exchanging (Jansen et al, 2007). At kinetochores, both rapid and slow/non-existent turnovers reflect SSIs (Fig 4) (Musacchio & Desai, 2017).

Figure 4. Liquid like does not exclude SSIs

(A) KNL1, an IDP at kinetochores, contains multiple sequence-related phosphorylation sites for the recruitment of the BUB1/BUB3 complex, where BUB3 is a phospho–amino acid adaptor. KNL1 docks to CCAN in the core kinetochore. A FRAP experiment on BUB1/BUB3 and on a core kinetochore subunit would lead to fundamentally different conclusions on the nature of the compartment, as the core subunit do not exchange and would not recover, whereas BUB1/BUB3 would exchange in seconds; (B) Hand-drawn curves representing the recovery behavior shown in A; (C) Two imaginary directly interacting MVPs in neighboring phase-separated droplets (1 and 2, where the small differences in color are meant to recognize the origin of molecules in the original droplets) could easily mix if the interaction times of the individual modules allowed relatively rapid exchange.

In FRAP experiments, predicted recovery half-times of macromolecular interactions, just assuming typical KDs for reversible interactions and typical association rate constants (Shammas et al, 2012), will thus extend from sub-seconds to hours, considered to indicate a diffusive liquid-like state or even solid-like state in many claims of PS [typically 1–100 s (McSwiggen et al, 2019b)]. If SSIs turn over rapidly, as they can, they provide a plausible basis for another property of liquid-like membraneless compartment, their fusion. Turning over rapidly, new “mixed” interactions in fusing droplets will progressively replace the identical interactions that existed in the individual droplets, especially if the network is heavily cross-linked through interactions turning over rapidly, as shown in a hypothetical example in Fig 4C [see also discussion in (Erdel & Rippe, 2018)]. In cells, active processes may additionally contribute to promote fusions, as observed for mitotic spindles, large ensembles of microtubules, microtubule cross-linkers, and molecular motors engaged in many reciprocal dynamic SSIs, which can fuse within a few minutes (Gatlin et al, 2009).

Finally, how surface tension/energy of membraneless compartments emerges, and which compartments should or should not show surface tension and sphericity, regardless of the detailed mechanism of biogenesis, are open question on which agnosticism is due [see Erdel and Rippe (2018); McSwiggen et al (2019a); McSwiggen et al (2019b); Peng and Weber, (2019) for further discussions]. It may be speculated that surface energy emerges from the density of the underlying mesh of interactions. Furthermore, we may surmise that macromolecules at the surface of a putative condensate have higher energy than those inside it. In this case, the surface energy may be expected to be high—and the boundary sharp—when the binding energies of the underlying molecular interactions are strong, which usually implies site specificity.

Most compartments may not result from PS

Thus, there are no good reasons to exclude SSIs as drivers of compartment assembly. Moreover, they are significantly more likely to participate in the biogenesis of the great diversity of subcellular compartments than a restricted variety of low-affinity unspecific homotypic interactions. By way of example, the kinetochore is a compartment on chromosomes that concentrate at least 60 distinct polypeptides to control chromosome segregation (Musacchio & Desai, 2017). With the dissection of kinetochores well underway, all interactions characterized so far appear to be site specific, with discrete binding interfaces, whose mutation predictably prevents recruitment of downstream components. This does not reflect an absence of MVPs or IDPs, as several kinetochore components, including CENP-CMif2 and KNL1Spc105, belong to these classes. In addition, we have no evidence that the kinetochore acts as an unspecific “sink” for functionally unrelated macromolecules. Its composition seems to be limited to functional components recruited to it in a site-specific manner, as shown for BUB1/BUB3 and MAD2. This may be the norm, as the compositional identity of compartments is usually well defined. Readers are referred to a discussion on the centrosome as a condensate (Woodruff et al, 2017; Raff, 2019), and contextually to a report on how artificial coacervation promotes filamentation of a bacterial tubulin homolog (Te Brinke et al, 2018).

Seen retrospectively, it may seem puzzling that SSIs were excluded as an underlying possible driver of compartment assembly. Why were they? I suspect the answer is that the field, since early on and progressively more pervasively, accepted that compartments concentrate macromolecules through a mechanism of PS, and continued to support this idea, often in presence of very strong evidence to the contrary, without a rigorous test of the hypothesis, as I will show below. My concerns extend and complement previously formulated objections that many compartments indicated as condensates lack features expected if PS were their driver (Wheeler et al, 2016; McSwiggen et al, 2019b; Erdel et al, 2020). If many compartments do not show hallmarks of PS, and if their biogenesis is likely to reflect mechanisms other than those advocated by its proponents, we should avoid calling them condensates, which should be rather reserved for compartments with evident signs of condensation by PS.

Validation pipeline: part 1

So, there are compartments whose predominant and possibly only drivers of assembly are SSIs. Are there any compartments where PS is an undisputable driver of assembly? How should we investigate this question for other compartments where our knowledge of the assembly mechanisms is limited? As a thought experiment, let us consider a newly discovered membraneless compartment X with several concentrated macromolecules (Fig 5A). Measured FRAP rates and recovery fractions range from essentially no recovery for certain components to full recovery in minutes or seconds for others. These FRAP rates may be taken as an indication that condensate X is partly solid, partly liquid, but also as an indication that SSIs of variable strength are at stake. With hypotheses facing off, what should be done?

Figure 5. Interactions in an imaginary compartment

(A) Compartment X concentrates several components, three of which are shown in blue, brown, and green colors together with their turnover times; (B) An assembly hierarchy might have been identified while investigating the interactions in Compartment X; (C) Upper left: Can a primary client with wild-type sequence be recruited to a scaffold with mutations in a binding site for the primary client? If so, the binding interaction is at least necessary for the recruitment. Upper right: Grafting of binding site of primary client for scaffold allows recruitment of unrelated macromolecule. Minimal binding sites indicate sufficiency. Bottom left: Iterative analysis down the interaction hierarchy. Bottom right: A mutant client that fails to be recruited to X but undergoes PS in vitro like the wild-type counterpart indicates PS in vitro not predictive.

Using a combination of in vitro and in vivo work, SSIs supporters should aim to identify all components of compartment X and their binding regions, mutate them, and show which are crucial for assembly and what hierarchy exists, if any (Fig 5B). What should PS supporters do? Their hypothesis is that solvation exemplified by oil–water is crucial for compartment assembly, irrespective of any existing site-specific interaction with other components. As it first needs to be proven that SSIs are at least insufficient for compartment assembly, one needs to identify, as a premise, all possible SSIs to assess how they contribute to assembly. This is precisely what those supposing SSIs planned to do, and that is how it should be: all possible hypotheses need to be considered, and SSIs should not be excluded from consideration without compelling reasons.

With the characterization of compartment X in progress, SSIs might have or have not been found. If they had not been found, compartment X might qualify for general PS. If they had been found, there would be a prospect for special PS but also for no PS at all. Thus, if SSIs have been found, the next question is whether they are necessary (and ideally sufficient) for compartment assembly in vivo. To assess this, a putative assembly scaffold may be mutated to eliminate the site-specific binding determinant for a client (Fig 5C). In a strategy enabled by separation-of-function mutants, the mutant scaffold will assemble in its correct location if there remain binding sites for the compartment. If the client fails to become recruited, the SSI is at least necessary for recruitment in compartment X. This could be reapplied to the chain of primary, secondary, and higher-order scaffold/clients. Under the PS paradigm, the continued recruitment of the mutant primary scaffold would imply that its solubility has not changed appreciably. If recruitment of the client, on the other hand, is abrogated, the PS model cannot invoke changes in its solubility in the bulk phase, as its sequence is invariant (the mutation is on the scaffold). Conversely, one may also mutate a client to prevent its recruitment into X, and ask if the mutation affects its PS behavior in vitro, i.e., whether PS in vitro is predictive of localization. In addition, as a criterion for sufficiency of SSIs, one may isolate the binding determinants of a client and graft them onto a macromolecule with entirely different features to ask if it becomes recruited in X, which would imply that the compartment does not discriminate for or against other features of the client (such as those that may determine its solubility).

Collectively, the pipeline described above is Part 1 of the test. With compartments being usually very complex, identifying interactions and assessing assembly hierarchies is an extremely demanding task requiring great biochemical prowess and a lot of patience, yielding modest short-term rewards; but nonetheless essential to real progress.

Validation pipeline: part 2

With Part 1 complete, PS may be ready for limelight if SSIs in X were found to be completely absent, or if they were unnecessary or insufficient toward its biogenesis (in order of increasing likelihood). Part 2 of the test should aim to gather hard evidence for PS. The general PS idea implies that a condensate will appear when the concentration of a putative PS driver exceeds Csat (implying there should be a recognizable Csat). To test PS, we cannot manipulate the SSIs we might have identified in Part 1 of the test, as they are crucial thermodynamic parameters that have to remain unaltered when we are trying to assess additional effects of PS. Thus, we can only aim to change Csat, for instance, by increasing the relative concentration of candidate components of X in the dilute bulk phase so that they would not join X, either because they prefer it less or because they prefer the dilute phase more. But how is Csat determined in a cell? We have powerful theories for the behavior of associative polymers in “poor solvent” conditions (Hyman et al