A quantitative inventory of yeast P body proteins reveals principles of compositional specificity

P bodies are archetypal biomolecular condensates that concentrate proteins and RNA without a surrounding membrane. While dozens of P body proteins are known, the concentrations of components in the compartment have not been measured. We used live cell imaging to generate a quantitative inventory of the major proteins in yeast P bodies. Only 7 proteins are highly concentrated in P bodies (5.1-15 μM); the 24 others examined are appreciably lower (most ≤ 2.4 μM). P body concentration correlates inversely with cytoplasmic exchange rate. Sequence elements driving Dcp2 concentration into P bodies are distributed across the protein and act cooperatively. Strong correlation between interaction valency of components and P body concentration suggests how the composition of other condensates may be estimated from known interaction data. Our data indicate that P bodies, and probably other condensates, are biochemically simpler than suggested by proteomic analyses, with implications for specificity, reconstitution and evolution.


INTRODUCTION
Eukaryotic cells contain numerous compartments that concentrate specific sets of molecules without a surrounding membrane (Banani et al., 2017;Shin and Brangwynne, 2017). These structures, referred to as biomolecular condensates, are related to a variety of biological processes. Examples include cytoplasmic processing bodies (P bodies) associated with RNA metabolism (Decker and Parker, 2012); promyelocytic leukemia nuclear bodies (PML NBs) involved in transcription, DNA damage repair, and anti-viral responses (Lallemand-Breitenbach and de Thé, 2010); signaling clusters in T cell activation (Su, 2016); and HP1 clusters in heterochromatin organization (Larson et al., 2017;Strom et al., 2017). While recent data indicate that condensates can have a variety of material properties and sometimes contain substructures (Banani et al., 2017;Shin and Brangwynne, 2017), significant portions of many condensates behave as dynamic liquids, showing fusion and fission, surface tension that makes them generally round, and rapid exchange of many components with the surrounding cytoplasm or nucleoplasm (Brangwynne et al., 2009(Brangwynne et al., , 2011Dundr et al., 2004;Wang et al., 2014;Weidtkamp-Peters et al., 2008). These behaviors, and the sharp response of some condensates to concentration or temperature changes, suggest that liquid-liquid phase separation plays an important role in generating these compartments in cells (Brangwynne et al., 2009;Falahati and Wieschaus, 2017;Li et al., 2012;Riback et al., 2017;Saha et al., 2016;Smith et al., 2016). The activities of biomolecular condensates are thought to derive from the assembly of specific collections of functionally related molecules into a unique physical environment. Thus, understanding condensates as chemical entities requires knowledge and understanding of their compositions.
Large scale proteomics studies have been performed to determine comprehensive inventories of molecules that localize to different condensates, including stress granules Markmiller et al., 2018;Youn et al., 2018), P bodies (Hubstenberger et al., 2017;Youn et al., 2018), and nucleoli (Anderson, 2002). There are also extensive studies on the localization of individual molecules to different condensates. Cumulatively, these studies generate partially overlapping lists of tens to hundreds of proteins and, in some recent work, RNA molecules in different condensates. Although some transcriptomic studies and a few analyses of individual proteins have been quantitative Klingauf et al., 2006;Leung et al., 2006;Wheeler et al., 2017), proteomic studies to date have been only qualitative. So both the relative and absolute concentrations of proteins in condensates are largely unknown, as are ways of predicting these features from existing data. Additionally, the relationships between concentration and dynamics of molecules in condensates have not been systematically explored.
We recently proposed a model to explain certain features of the compositions of biomolecular condensates (Banani et al., 2017;Ditlev et al., 2018). The model is based on distinctions between two classes of molecules: scaffolds and clients. Scaffolds are defined as condensate components that contribute to the assembly of the compartment.
Thus, deletion of scaffolds decreases the size and/or numbers of condensates. In contrast, clients are not necessary for integrity of the structure, but are recruited by interacting with scaffolds. Scaffolds often have high valency of interaction elements and exhibit slow molecular dynamics, while clients often have lower valency and faster dynamics. Where studied, there are generally few scaffolds and many clients in natural condensates (Ditlev et al., 2018). Implicit in the model is that scaffolds should have higher concentration within the condensate than clients, since the former recruit the latter. The model was generated based on the behaviors of engineered multivalent macromolecules, where the distinction between scaffold and client was stark by design. In more complicated natural condensates, however, the distinction between scaffolds and clients may be more ambiguous, and could depend on expression levels and cell conditions (Ditlev et al., 2018). Nevertheless, initial features of the model were validated by the behaviors of several molecules in two natural condensates, PML NBs and yeast P bodies.
To examine the composition of a complex cellular condensate in a quantitative manner, we have performed a systematic analysis of yeast P bodies, an archetypical biomolecular condensate. P bodies are protein-and mRNA-rich cytoplasmic condensates conserved from yeast to mammals (Ingelfinger et al., 2002;Sheth and Parker, 2003;van Dijk et al, 2002). They are thought to participate in RNA metabolism, modulating mRNA decay and/or acting as sites of RNA storage during cellular stress (Aizer et al., 2014;Bhattacharyya et al., 2006;Cougot et al., 2004;Sheth and Parker, 2003). Taking advantage of an available yeast GFP library, we used quantitative fluorescence microscopy to measure the absolute concentrations of 31 P body resident proteins within the condensates and in the surrounding cytoplasm. We also measured their dynamic properties using fluorescence recovery after photobleaching (FRAP). We find that P body proteins segregate into two groups based on their concentrations in P bodies and dynamics. Members of the first group (Dcp2, Edc3, Pat1, Xrn1, Lsm1, Dhh1, Upf1) are highly concentrated in P bodies (5.1-15 µM) and exchange slowly with the cytoplasm. All genetically-identified protein scaffolds of yeast P bodies are in this group. In contrast, members of the second group are substantially less concentrated in P bodies (all except Sbp1, 0.7-2.4 µM) and exhibit faster dynamics. These data are consistent with the scaffold-client model of condensate composition, and suggest that while a condensate may contain many components, only a small number are highly concentrated there. In a molecular dissection of Dcp2, we find that the N-terminal domain, multivalent C-terminal domain, and central high affinity Edc3 binding site all contribute to partitioning of the protein into P bodies and to its dynamic exchange with the cytoplasm. Moreover, the Nterminal and C-terminal domains can act cooperatively to promote recruitment of Dcp2 into P bodies. This highly distributive organization suggests a ready means of quantitatively modulating condensate composition during evolution. Surprisingly, a simplistic analysis of the known interactions of the major P body proteins reveals a strong correlation between the valency of partner binding and concentration in the compartment, suggesting that the composition of other biomolecular condensates may be predictable from known interaction data. We suggest that similar quantitative analyses will be important in understanding the assembly, compositions, and chemical and biological functions of other biomolecular condensates.

Strategies for quantification of P body proteins
In order to understand the chemical nature of a condensate it is important to both know its members, and also their concentrations. While recent experiments have generated multiple lists of the components of RNP granules, very little information is available on the relative and absolute concentrations of these components in the compartments. Two criteria must be met to accurately measure protein concentrations in a condensate based on fluorescence intensities alone: 1) the size of the condensate must be larger than the point spread function (PSF, i.e. the diffraction limit) of the microscope used in the analysis since the fluorescence intensities are diluted for small objects (Fink et al., 1998); 2) the composition of the condensate cannot change over time, allowing reliable comparisons among multiple condensates in multiple cells. P bodies in wild type S. cerevisiae are typically diffraction-limited in size under normal conditions. But they become larger when mRNA decay is decreased, e.g. when mRNA decay proteins such as Dcp1 are deleted or when cells respond to stresses such as glucose starvation (Teixeira et al., 2005.
Initially we analyzed GFP-tagged proteins expressed in a dcp1Δ strain, because the effects of stress are time-dependent and thus more difficult to systematically measure across a series of strains. We used confocal microscopy, limiting our analyses to P bodies that were larger than the x-y PSF of our microscope, and correcting for the diluting effect of the larger z-PSF based on an assumption that the structures were spherical.
Using this approach, we individually expressed 31 reported P body resident proteins tagged with GFP at their C-termini in their chromosomal location under their endogenous promoters in a dcp1Δ strain (Figure S1A, and Table S2). For each protein, we measured the absolute concentration in the P body and in the surrounding cytoplasm, based on cellular fluorescence intensity and independent calibration of fluorescence versus GFP concentration on our microscope. We calculated the partition coefficient (PC) as the ratio of these two values.
We also used fluorescence recovery after photobleaching (FRAP) to measure the exchange between P bodies and the cytoplasm. Of the 31 proteins, 19 showed sufficiently punctate distributions in mid-log phase to permit analysis (PC > ~2). We presume these puncta are P bodies because of previous reports that the proteins localize to P bodies and because in all cases tested we observed co-localization of the proteins with an mCherry-tagged P body marker, Edc3 ( Figure S1B). We refer to these 19 proteins as regular P body proteins hereafter. The remaining 12 proteins were distributed relatively uniformly in the cytoplasm and their P body concentrations could not be analyzed; these may be stress or strain specific proteins that do not concentrate in P bodies under our experimental conditions (Table S2).
Two observations suggest that the GFP tag probably does not strongly affect protein behaviors. First, the PC values and dynamics of Edc3 and Dhh1 tagged with GFP and mCherry were similar ( Figure S2A and S2B). Second, the PC values and dynamics of Dcp2 are nearly identical with an N-terminal or C-terminal GFP tag ( Figure S2C). In a screen of this size it is not practical to validate each protein with multiple tags at multiple locations, and it remains possible that some other proteins are affected by the tag.

Seven proteins/assemblies are highly concentrated in P bodies
We obtained PC for all 19 regular P body proteins, which revealed subclasses of these molecules. Average PC values of the 19 regular P body proteins had a wide range, with a maximum of 133 (Dcp2) steadily decreasing to a minimum of ~5 (Eap1, Ssd1) ( Figure   1A). Only a few proteins had high partition coefficients, including Dcp2 (133 ± 13), Edc3 (133 ± 8) and Pat1 (107 ± 12); the majority had PC values <20. PC values of a given protein varied between 25 -45% of the average, both among P bodies in a given cell and between cells, although the intracellular variation tended to be slightly smaller. With only 1 component labeled in each cell, it was not possible to assess correlations between PC values of different components (see discussion).
We also obtained absolute concentrations of each protein in P bodies, which ranged from ~0.7 µM to 15 µM and revealed two notable features of P bodies. First, seven proteins, Dcp2, Edc3, Pat1, Xrn1, Lsm1 (likely representing the entire, constitutive Lsm1-7 assembly (Sharif and Conti, 2013), Dhh1 and Upf1 -had average P body concentrations of 8.9 -12 uM, with a slightly higher concentration (15 uM) for Dcp2 and slightly lower concentration (5 µM) for Upf1. We refer to these components as core P body proteins. A second striking observation was that there was a clear, sharp distinction between highly concentrated and more diffuse proteins ( Figure 1B). After the core components most other proteins have P body concentrations < 2.5 uM, with most ~1 µM. The only exceptions is Sbp1 (4.6 uM), which also has a very high cytoplasmic concentration, affording it a high P body concentration even with a small PC value ( Figure 1, Table S1).
Sbp1 also is not highly enriched in glucose-starved wild type P bodies (see below); for these reasons we do not include it in the core group, even though it has modestly high concentration in dcp1Δ P bodies. Thus, there is a large drop in P body concentration between the seven core proteins, and most other non-core proteins.
Two observations suggest that our data are consistent with previously reported measurements. First, the stoichiometry of protein complexes including Pat1/Lsm1 (1:1) and CCR4/Pop2(1:1) was the same as in previous biochemical characterizations (Basquin et al., 2012;Sharif and Conti, 2013). Upf1 concentrates in P bodies more than Upf2/Upf3 (which show 1:1 stoichiometry), consistent with previous findings that Ufp1 functions upstream, and independently of Upf2 and Upf3, which act as a constitutive complex (Sheth and Parker, 2006). Second, the total cellular concentrations that we measured are similar to a unified quantitative protein abundance database of S.
By knowing the concentrations of 31 proteins in P bodies analyzed here, and their molecular weights, we were able to calculate the fractional contribution of the core components to the total protein mass of P bodies ( (#$ * &'($) where C i and MWT i are the P body concentration and molecular weight of the ith species, respectively). This estimate assumes conservatively that the 12 reported P body residents that lacked puncta in our analysis (Table S2) had the same concentration in P bodies as Psp2, the least concentrated component we could observe. It also neglects the (probably small) contributions of other proteins that might be in P bodies but were not analyzed here.
By these calculations, the seven core proteins, including the whole Lsm1-7 complex, constitute approximately 83% of the total protein mass of P bodies (see Methods).
Together, these data show that while yeast P bodies contain many different proteins, the majority of these are only weakly concentrated in the compartment, and a relatively small number make up the bulk of the structure.

Core proteins tend to have slow dynamics
We also examined the dynamics of the 19 regular P body proteins using FRAP. For each GFP-fusion expressing strain, entire P bodies (0.4 -0.8 um) were photobleached and the fluorescence recovery curves were fit to single exponentials. The exponential rate constant of exchange (k) and extent of recovery were used to assess their dynamics. This analysis revealed that the regular P body proteins exchange with the cytoplasm with very different rates and fractional recoveries ( Figure S3 and Table S1). At one end of the distribution, Dcp2 and Edc3 showed no measurable recovery on a 150 second timescale.
Since we observed heterogeneity of both partitioning and dynamics, we asked whether these parameters are related. For each protein, we plotted the exchange rate (k), and extent of recovery against P body concentration (Figure 2A and 2B). Although exchange rate and recovery for both core proteins and non-core proteins spanned broad ranges, the core proteins (Dcp2, Edc3, Pat1, Xrn1, Lsm1, Upf1, and Dhh1) tended to have slower exchange rates and less fractional recovery than the non-core proteins (Figure 2A and 2B). On average, for the core group, the exchange rate was five-fold smaller than the non-core group, and recovery fraction was two-fold lower (Figure 2A and 2B).
Three observations argue that the slower recovery dynamics of the core proteins are not simply due to smaller fluorescent pools in cytoplasm, but rather reflect different interactions in the P body. First, neither exchange rate nor fractional recovery correlates with protein concentrations in cytoplasm ( Figure S4A and S4B). Second, less than 50% of the total pool of each protein is localized to the P body (< 8% for most), suggesting that there are still substantial fluorescent pools in the cytoplasm ( Figure 3). Third, we also performed inverse FRAP (iFRAP), which is insensitive to the size of fluorescent pools, for several proteins. In iFRAP, the entire cytoplasm except one P body was bleached and the loss of fluorescence in the P body was followed over time. In each case, the exchange rate and fractional recovery measured by iFRAP was similar to those measured by FRAP ( Figure S4C and S4D). Collectively, our data suggest the seven core P body proteins tend to exchange slowly with the surrounding cytoplasm, while non-core proteins tend to exchange more rapidly.

Partitioning and dynamics of proteins are not strongly affected by Dcp1 deletion
Since the above analysis was all performed in a dcp1∆ strain, we wanted to measure PC and P body dynamics under a different set of conditions to evaluate if the Dcp1 deletion made a substantial impact on P body composition or dynamics. For this experiment, we analyzed wild type strains after 30-60 min of glucose starvation, a stress that is known to promote P body assembly (Teixeira et al., 2005). By several criteria, we observed similar results to our analysis in dcp1∆ strains.
First, similar to the dcp1∆ strains, during glucose deprivation, Dcp2, Edc3, Pat1, Lsm1, Xrn1, Upf1, and Dhh1, remain the most concentrated proteins in wild type P bodies, although Pat1 and Lsm1 dropped somewhat compared to the dcp1∆ strains ( Figure S5A and S5B). Two additional proteins also partition strongly into wild type P bodies under glucose starvation, Dcp1 and Pby1. Dcp1 binds with high affinity to Dcp2, and Pby1, in turn, binds strongly to Dcp1, explaining its absence in dcp1Δ P bodies (Krogan et al., 2006). In contrast to the core group, in these experiments only two members of the noncore group, Hek2 and Sbp1, partition sufficiently strongly into P bodies to permit analysis suggesting that the partitioning of low occupancy clients is lower under glucose starvation (see discussion).
We also observed that the dynamics of all proteins were qualitatively similar in glucose deprivation as compared to the dcp1Δ strains. The core proteins have slower dynamics, although there is no clear distinction between core and non-core concentrated groups because of limited number of the latter ( Figure S5C and S5D). Nevertheless, these data suggest that partitioning and dynamics of proteins are qualitatively similar in the wild type strains under glucose starvation and the Dcp1 deletion strains, especially for the core proteins. We thus performed all subsequent experiments in dcp1Δ strains, unless stated otherwise.

P bodies do not strongly sequester their resident proteins
The biological importance of concentrating certain proteins into P bodies is unclear. One proposed function is that the P body could sequester molecules, inhibiting their activities in the cytoplasm. Similarly, P bodies have been suggested to store mRNAs or proteins under cellular stress, which could then be returned to the cytoplasm when the stress is resolved (Aizer et al., 2014;Bhattacharyya et al., 2006). To examine the efficiency of protein sequestration, we quantified the fractions of each P body protein in the compartments (F P , the ratio of integrated fluorescence intensity within P bodies to the total fluorescence intensity in the entire cytoplasm), with the caveat that very small P bodies would likely be too dim to be observed (but see below). We first measured F P in mid-log phase without stress. Under these conditions, the most concentrated proteins, Dcp2 and Edc3, are on average ~30% sequestered in observable P bodies ( Figure 3).
For the other highly concentrated proteins (Pat1, Xrn1, Lsm1, Upf1, and Dhh1), about 10% is sequestered in visible P bodies. The fraction in P bodies is even smaller for less concentrated proteins such that only about 5% of each is in visible P bodies.
We further asked if the degree of P body sequestration in dcp1Δ strains increases under glucose starvation . Thus, we measured F P after 4h of glucose starvation. Although F P increased for most of the proteins, visible P bodies still sequester less than 40% of each of the concentrated proteins ( Figure S6A). We next grew cells to stationary phase to test if P bodies convert to sequestration sites at different growth stages. Again, less than 50% of each protein except Dcp2 and Edc3 is in P bodies ( Figure   S6B).
One possible explanation for the low degree of sequestration measured here is that many P bodies are too small to be observed by standard confocal microscopy. This would give artifactually low F P values, as a significant fraction of P body-associated protein would not be accounted for in the integrated P body fluorescence intensity. Nevertheless, we can estimate an upper limit of F P for most proteins by assuming that all P bodies in a given cell have identical compositions independent of size, and that the F P value for the most sequestered protein (Edc3), is, in fact 1 when small P bodies are properly accounted for.
With this assumption, the estimated maximum F P (F P,max ) for each protein would be F P,max = F P * 1/ F P(Edc3) . Even with this conservative estimate, most proteins are only sequestered to < 20% in P bodies under all conditions.
Together, our quantitative measurements indicate that P bodies do not strongly sequester their resident proteins under the conditions examined here. This is because the volume of P bodies is very small compared to the total cytoplasmic volume (e.g. ~ 0.5% volume fraction in the dcp1Δ strains), so that even for a protein with a high PC value, most of the molecules are located in the surrounding cytoplasm rather than in the condensate. Our data are similar to other observations wherein only a small fraction of Argonaute protein accumulates in P bodies (Leung et al., 2006) and only ~18% of G3bp1 accumulates in mammalian stress granules . Our data of course do not rule out the possibility that sequestration could be higher under different conditions, nor do they speak to sequestration/storage of RNA, which could have even higher PC values than proteins, perhaps due to non-equilibrium processes (Ditlev et al., 2018;Hubstenberger et al., 2017).

Elements controlling Dcp2 partitioning and dynamics are distributed across the protein
We have shown that proteins concentrate into P bodies to different degrees and with distinct dynamic behaviors. We next sought to understand what molecular features could control these properties, using Dcp2 as an example, as one of the most highly concentrated and least dynamic P body components. We divided Dcp2 into three distinct regions ( Figure 4A). The N-terminal domain (NTD) of the protein possesses decapping activity and binds to Dcp1 and mRNAs (Deshmukh et al., 2008). The multivalent Cterminal domain (CTD) contains multiple short helical leucine rich motifs (HLMs) that bind to Edc3. The analogous CTD of S. pombe Dcp2 phase separates with Edc3 through these multivalent interactions (Fromm et al., 2014). Finally, near the center of Dcp2, the first HLM (HLM1) appears to bind Edc3 with appreciably higher affinity than all other HLMs , and mutations to this motif impair Dcp2 partitioning into P bodies (Harigaya et al., 2010). We expressed different Dcp2 mutants in dcp1Δdcp2Δ strains, in which P bodies are still formed, as indicated by punctate Edc3 localization ( Fig S7A), but they are smaller than in yeast expressing full length Dcp2, which is consistent with earlier results showing Dcp2 contributes to P body assembly . The analysis of these variants identified three molecular elements of Dcp2 that affect its partitioning into P bodies.
First, we found that N-terminally truncated Dcp2 (Dcp2 DN), which removes the RNA binding domain of the protein, partitions into P bodies less efficiently than the full-length protein (Dcp2 FL), with PC values of 72 ± 8 and 124 ± 9, respectively. FRAP analysis revealed that Dcp2 DN had an appreciably higher recovery fraction, 0.35 over 150 seconds, compared to Dcp2 FL, which does not recover at all in this timeframe ( Figure   4B). This demonstrates that the N-terminal domain of Dcp2 promotes P body accumulation.
Second, mutating HLM1 (Dcp2 DH1) to alanine in the full length protein strongly decreased the PC to ~26, and increased the recovery after photobleaching to 0.65 over 150 seconds. Thus, both the NTD and HLM1 contribute to concentrating Dcp2 into P bodies and decreasing its exchange with the cytoplasm.
Third, truncation of the CTD (Dcp2 300) had no significant effect on the PC and dynamics of Dcp2 ( Figure 4B). However, additional data suggest that the other HLMs in Dcp2's Cterminal extension can contribute to P body targeting when HLM1 is missing. Specifically, removing the C-terminal domain from Dcp2 DH1 (Dcp2 300 DH1) abolished its recruitment to P bodies ( Figure 4C). Presumably, the C-terminal domain can contribute to Dcp2 partitioning in this context through the other HLMs interacting, albeit less effectively, with Edc3 (Fromm et al., 2014).
Since the N-terminal domain, HLM1, and C-terminal domain are all required for efficient partitioning and maintaining the characteristic slow dynamics of Dcp2, we conclude that elements controlling protein partitioning and dynamics are distributed across the protein.

RNA binding and turnover affect Dcp2 partitioning and dynamics
We further dissected the N-terminal region of Dcp2 and asked how interactions with RNA, an important scaffold of P bodies, affect partitioning and dynamics in dcp1Δdcp2Δ strains.
Starting with the Dcp2 300 fragment, we either mutated previously reported RNA binding residues, R170, K212, K216, and R229, to alanine, or introduced repulsive interactions by mutating K212 and K216 to aspartate, generating Dcp2 300 AAAA, and Dcp2 300 ADDA, respectively ( Figure 5A) (Deshmukh et al., 2008). PC values of both mutants decreased to ~85, from ~132 for wild type Dcp2 300 ( Figure 5B), which is similar to the PC for the Dcp2∆N construct which is missing the entire N-terminal domain ( Figure 4B).
Mutating the four RNA binding residues increased dynamics slightly, as both mutants recovered to 0.3 over 150 seconds ( Figure 5C). These data suggest that binding to RNA enhances Dcp2 partitioning and maintains its slow dynamics of exchange.
To investigate the effect of catalytic activity on Dcp2 partitioning and dynamics, we made mutations in the Dcp2 DH1 construct, which affords a better dynamic range than the wild type protein ( Figure 4B). We mutated W50 and D54, which were previously implicated in RNA cap recognition and hydrolysis, to alanine, giving Dcp2 DH1 WD ( Figure   5A) Floor et al., 2010). In a dcp1Ddcp2D strain, this protein had PC values and dynamics similar to those of the parent Dcp2 DH1 ( Figure S7B). One explanation for this behavior is that catalytic activity of Dcp2 is already low in the absence of Dcp1, so that subtle differences caused by mutations would not be observed. We thus analyzed Dcp2 DH1 and Dcp2 DH1 WD in a dcp2D strain, where the former protein should reconstitute significant mRNA decapping activity while the latter should not. Formation of P bodies was promoted by expressing Dcp2 DH1 WD ( Figure 5D), which is consistent with mutations that block decapping catalysis leading to increased P bodies (Sheth and Parker, 2003;Teixeira and Parker, 2007). Dcp2 DH1 WD formed six P bodies per cell on average, compared to two P bodies per cell for Dcp2 DH1 ( Figure 5E), despite being expressed at similar level ( Figure S7C). More importantly, changing the catalytic rate of decapping changed the accumulation of Dcp2 in P bodies and altered its dynamics.
Specifically, ~10% of Dcp2 DH1 WD was concentrated in visible P bodies compared with ~4% for Dcp2 DH1 ( Figure 5E). The exchange rate also decreased from 0.041 s -1 to 0.017s -1 and fractional recovery decreased from 0.85 to 0.65 in the WD mutant ( Figure   5F). These data indicate that defective hydrolysis of the RNA cap promotes accumulation of Dcp2 in P bodies, and decreases Dcp2 dynamics. However, it is unclear whether the changes are caused by increasing amounts of cellular mRNA, which occurs when decapping is blocked, or by disrupting catalytic activity of Dcp2 per se.

Recruitment elements of Dcp2 act cooperatively
Different regions of Dcp2 contribute to partitioning into P bodies and modulate the dynamics of the protein. We next asked whether these regions can act cooperatively to promote P body recruitment. We analyzed three Dcp2 mutant fragments in dcp1Δdcp2Δ strains: 1) Dcp2 300 DH1, which lacks HLM1 and can only bind to RNA; 2) a C-terminal Dcp2 fragment where five out of ten HLMs have been inactivated by mutation, Dcp2C D5H, which binds Edc3 but not RNA; and 3) fusion of Dcp2 300 DH1 and Dcp2C D5H, Dcp2 DH1 D5H, which binds to both RNA and Edc3 ( Figure 6A and 6D). The three mutants expressed at similar level in dcp1Δdcp2Δ strains ( Figure S7C). We analyzed recruitment of the three proteins into P bodies using Edc3 as a marker for the compartments. Dcp2 300 DH1 and Dcp2C D5H are measurably recruited into microscopic P bodies in only ~20% and ~30% of cells, respectively. However, the fusion of the two fragments, Dcp2 DH1 D5H, is recruited into P bodies in ~95 % of cells ( Figure 6B and Figure 6C). Moreover, partition coefficients of Dcp2 300 DH1 and Dcp2C D5H 2.5 and 3.5 respectively, while the partition coefficient of Dcp2 DH1 D5H is much higher, 31 ( Figure   6D).
Thus, while Dcp2 300 DH1 and Dcp2C D5H are each recruited only weakly to P bodies, their fusion, Dcp2 DH1 D5H, is recruited strongly. This indicates that recruitment elements can act cooperatively when fused in cis, even when they recognize distinct ligands (in this case, Edc3 and RNA) within a condensate ( Figure 6E).

Partitioning into P bodies correlates with interaction valency
Since Dcp2 fragments partition into P bodies stronger when binding to both Edc3 and RNA, we wondered whether proteins, in general, partition into condensates more efficiently when they can interact with multiple molecules within the compartment. We summarized the number of identified protein-protein interacting elements and expected binding sites per mRNA molecule, expanding a previously published table (Rao and Parker, 2017) to include the 19 regular P body proteins in this study (Table S4). We only counted one valence if a protein interacts with multiple molecules through the same site.
The total valency thus is an estimation of how many molecules one protein can interact with simultaneously. Surprisingly, given that our analysis did not include any information on binding affinity or avidity, the partition coefficients, P body concentrations, and fractional recovery values correlate strongly with total valency (Pearson r = 0.8) ( Figure   7A, 7B and 7D). The highly concentrated proteins generally show multivalent interactions with each other and with less concentrated proteins, whereas the less concentrated proteins typically only interact with one, usually RNA, or two molecules. Protein dynamics, i.e. exchange rates, correlate with total valency to a lesser extent ( Figure 7C).
Although there are some assumptions made in the correlative analysis (see comments for each protein, Table S4), our data suggest that multivalent interactions of highly concentrated proteins probably contribute to their higher partitioning and slower dynamics.
This result suggests a way to predict the relative concentration of proteins, and the likelihood of a protein functioning as a scaffold or client, in other condensates based on known interaction data.

DISCUSSION
This work presents the first quantitative description of an RNP granule, which is representative of the broader class of non-membrane bound organelles referred to as biomolecular condensates. This analysis has revealed features of yeast P bodies that are generalizable to other condensates.

Distinctions between scaffolds and clients can explain compositions of natural biomolecular condensates.
Our work has revealed two clear classes of proteins within yeast P bodies, which exhibit the properties of scaffolds and clients as defined in our previous work (see Introduction) (Banani et al., 2016). One class of proteins, which we name the core P body proteins, are a set of highly enriched proteins with PC ≥ 30 and P body concentrations > 5 µM, including Dcp2, Edc3, Pat1, the Lsm1-7 complex, Xrn1, Dhh1, and Upf1. On average, these proteins also show slower dynamics of exchange from P bodies and a small fraction of exchangeable molecules. Remarkably, the three most stable proteins within P bodies, Dcp2, Pat1, and Edc3, are the three major proteins shown to affect P body assembly (Buchan et al., 2008;Decker et al., 2007;Teixeira and Parker, 2007). Other core P body members also show multivalent interactions with each other (Rao and Parker, 2017;Swisher and Parker, 2011), and at least for Dhh1, Upf1, and the Lsm1-7 complex can contribute to P body assembly (Decker et al., 2007;Mugler et al., 2016;Sheth and Parker, 2006 -Peters et al., 2008). Analogously, the scaffold of the pericentriolar matrix in C.elegans, SPD-5, is stable and less dynamic, whereas the clients PLK-1, TPXL-1, SPD-2, and ZYG-9, are more mobile (Woodruff et al., 2017). Together, these data suggest that condensates may generally be organized around a relatively small number of highly concentrated, less dynamic scaffolds and large number of less concentrated, more dynamic clients. This construction would provide a relatively simple route for a condensate to appear during evolution, in that only a small number of proteins would need to develop the ability to assemble dynamically (either with themselves or with each other). Once this structure was established, other proteins could evolve the ability to interact with the scaffolds and consequently be recruited into the compartment (see also below). A composition of this nature also indicates that condensates are biochemically less complicated than suggested by proteomics studies, where tens to hundreds of proteins have been annotated as residents of particular structures. Our quantitation indicates that most components are present in only small amounts, and the majority of the protein mass derives from only a few molecules. Such an understanding greatly simplifies efforts to reconstitute condensates in vitro, and can frame models of their biochemical functions.

Compositional specificity
A major question regarding biomolecular condensates is how the composition of the Our data on Dcp2 recruitment to P bodies suggests one means of solving this problem.
We found that two fragments of Dcp2, which bind RNA and Edc3, respectively, partition only weakly into P bodies individually. Yet when fused together, they are recruited strongly. This behavior is analogous to previous observations that cooperativity between IDRs and folded domains can promote recruitment into both phase separated droplets in vitro and P bodies in cells (Lin et al., 2015;Protter et al., 2018). This suggests that in general, protein elements that bind scaffolds weakly, either specifically or non-specifically, will not partition strongly into a condensate. But when two such elements are fused together, even if they bind different scaffolds, they can be recruited strongly. This is analogous to, and in some cases may be mechanistically similar to, avidity effects in canonical molecular interactions, where high affinity can be achieved through multivalent binding. Such effects will greatly narrow the specificity of recruitment, even for scaffolds that individually bind ligands promiscuously. This mechanism also provides ready routes for evolution of new clients through genetic rearrangements that fuse together multiple low-affinity interaction elements. Distributing recruitment across a large number of interaction elements would render composition less susceptible to mutations, which could lead to evolutionary selection. Therefore, partition into condensates through multiple binding elements may be evolutionarily favored to allow compositional specificity of biomolecular condensates.

Estimating composition and dynamics of biomolecular condensates through interaction data
Our analyses provide means to predict compositional and dynamic properties of other condensates that have not yet been analyzed quantitatively. We observed strong correlations of the concentration, PC and fractional recovery after photobleaching of P body components with the valency of interaction between components. In addition, we observed that all P body scaffolds are within the core group of components, while the non-core group contains exclusively clients. If these relationships hold generally, the relative valency of molecules within a condensate should allow prediction of the relative concentrations and dynamics of those molecules, and estimation of scaffolds and clients for the structure. The valency of direct interactions, in turn, can be determined from careful curation of low-(biochemical, yeast two-hybrid, etc.) and high-throughput (proteomic) interaction data for a condensate.
We emphasize that despite yielding strong correlations, our analysis here was extremely simplistic. Many factors beyond simply the number of non-exclusive binding partners may affect the partitioning and dynamics of components of biomolecular condensates. These include binding affinities, avidity effects, unknown interactions and active cellular processes. For example, clients with higher binding affinities for a scaffold should partition into condensates more strongly (Banani et al., 2016), and stronger binding between scaffolds should increase their partitioning (Wei et al., 2017). Enzymatic activities and mechanical forces exerted by ATP-consuming machines may alter protein dynamics and partitioning. Thus, simple correlative analyses of valency should be viewed as a starting point to further explore the chemical and physical natures of biomolecular condensates.

Quantitative considerations of condensate function
Quantitative analyses as we have performed play an important role in assessing the functions of condensates. For example, several condensates, including P bodies, have been proposed as sites for sequestration/storage of biomolecules, in part because they appear as qualitatively bright puncta by microscopy, and in part because their disruption can activate certain processes (Arimoto et al., 2008;Decker and Parker, 2012;Li et al., 2000). However, as described above, when total condensate volume (typically < 1-2 % of cytoplasm/nucleoplasm even when accounting for sub-diffraction puncta, (Figure 3, Leung et al., 2006;Rao and Parker, 2017) and partition coefficients (2~200) are quantified, it is evident that only a small fraction of most molecular species are sequestered within condensates. While in some processes small changes in the amount of available species in the bulk cytoplasm/nucleoplasm could have functional consequences, in others, different mechanisms must be considered. For example, rather than sequestration, an inhibitory catalytic function within a condensate could also explain the activation of a process upon condensate disruption. Such considerations will further advance the already significant impact of biochemical reconstitutions of phase separation on understanding the functions of biomolecular condensates (Fromm et al., 2014;Schütz et al., 2017;Woodruff et al., 2017).

Conclusion
Our quantitative analyses have revealed that P bodies are compositionally simpler than previously envisioned, with only a small number of highly concentrated components.
Among these are all of the genetically defined P body scaffolds. Rates of exchange between P bodies and the surrounding cytoplasm loosely correlate inversely with See also Table S1, Figure S1, Figure S2 and Figure S5.

Figure 2. Core proteins tend to have slow dynamics
Exchange rate (A) or fractional recovery (B) was plotted as a function of absolute concentrations in P bodies, respectively (mean ± SEM). Core and non-core proteins are indicated by red and black symbols, respectively. Graphs on the right show averaged exchange rate or fractional recovery (red lines) in core and non-core groups. Analysis of significance was calculated by Wilcoxon rank-sum test. *** and **** indicate p values less than 0.001 and 0.0005, respectively.
See also Table S1, Figure S2, Figure S3, Figure S4, and Figure S5.    See also Figure S7. Significance was calculated by the Wilcoxon rank-sum test. **** indicates p values less than 0.0005.
(E) Specific recruitment to P bodies can be achieved by distributing elements across Dcp2.
See also Figure S7.  Comparison of protein abundance (molecules per cell) measured in this study to previous unified database (Ho et al., 2018).  were plotted as a function of absolute concentrations in cytoplasm (mean ± SEM), respectively. P body core and non-core proteins are indicated by red and black symbols, respectively.
(C/D) Exchange rate (C) or fractional decay (D) measured by iFRAP are qualitatively similar to values measured using FRAP. The fractional decay of more rapidly exchanging proteins measured by iFRAP is systematically smaller than measured using FRAP due to the longer time to bleach the whole cytoplasm in the former experiments.
P body core and non-core proteins are indicated by red and black symbols, respectively. Left, proteins found in both wild type and dcp1Δ strains ordered by their PC in dcp1Δ strains. Right, proteins only found in P bodies in wild type strains.
(C/D) Exchange rate (C) or fractional recovery (D) was plotted as a function of absolute concentration in P bodies (mean ± SEM). P body core and non-core proteins are indicated by red and black symbols, respectively.

Yeast strains
Yeast strains used in this study are listed in Table S2. Yeast strains carrying plasmids were constructed using lithium acetate-based transformation (Gietz and Schiestl, 2007).

Plasmid construction
Plasmids used in this study are listed in Table S3. All Dcp2 mutants were expressed under the DCP2 promoter on a low-copy centromeric plasmid (pRP1902) as previously reported (Harigaya et al., 2010). Dcp2 point mutations were made by site-directed mutagenesis using KOD Xtrem Hot start DNA Polymerase followed by Dpn1 digestion.

Yeast growth conditions
dcp1Ddcp2D strains and dcp2D strains expressing GFP-Dcp2 mutants were grown in synthetic medium lacking uracil but containing 2% glucose. dcp1Ddcp2D strains expressing both GFP-Dcp2 mutants and Edc3-mCherry were grown in the same media also lacking lysine. Glucose starvation in Figures S5 and S6 was performed by exchanging with the corresponding synthetic medium lacking 2% glucose for durations indicated in the text. Stationary stage in Figure S6 was achieved by growing cells for 5 days and OD 600 > 6. For imaging, cells were grown at 30 °C until OD 600 = 0.4 ~ 0.6, and then immobilized on concanavalin-A (Sigma-Aldrich) coated glass bottom dishes (MatTek).

Image acquisition and analysis
All images were acquired using a Leica SP8 Laser Scanning Confocal Microscope using a 100 X 1.4 NA oil immersion objective and photomultiplier tube (PMT) detector. Images were analyzed using Fiji.

Identification of P bodies
For strains expressing only GFP tagged protein, P bodies were identified by thresholding the GFP fluorescence intensity using the MaxEntropy algorithm in Fiji. For stains expressing both GFP tagged proteins and Edc3-mCherry in Figure 6, Edc3-mCherry signals were thresholded (MaxEntropy) to identify P bodies, and created masks.
Partitioning of the Dcp2 mutants was assessed by the ratio of average GFP intensity within in the mask (P body identified by Edc3-mCherry signal) to the intensity outside the mask (cytoplasm). Positive GFP puncta were determined when the ratio is greater than 1.5. Because the formation of and partitioning into biomolecular condensates are sensitive to protein expression levels, we eliminated cells with low expression (bottom 10% of the populations) and high expression (top 10% in the populations) of GFP and mCherry.

Measurements of absolute concentrations in P bodies and cytoplasm, and partition coefficients
To quantify the absolute concentrations of GFP-tagged proteins in P bodies, we assume that P bodies are spherical (based on the similar diameters in x and y when measured), and correct their measured intensities based on the point spread function (PSF) of our microscope (Fink et al., 1998). We determined the PSF using 200 nm fluorescent microspheres (Invitrogen) imaged with the same optics, filters, zoom settings and pinhole settings used throughout our study. We then modeled (Matlab) the intensity-diluting effect of the PSF when imaging spheres of different sizes through convoluting the PSF with the sphere. This yielded a correction curve relating sphere diameter to the fraction of true maximum intensity actually measured in the image (Fink et al., 1998), assuming all fluorescence intensity derived from the sphere and none derived from the surroundings.
We limited our cellular analyses to P bodies with measured diameter > 330 nm (most are 400-800 nm in dcp1D strains), which is 1.1 times larger than the x-y PSF, and thus can be accurately measured as the full width at half maximum intensity (FWHM) of the object.
From the measured diameter in the x-y dimension, an assumption of spherical shape, and the correction curve, we determined the calibration factor (CF) for the P body intensity. To determine P body intensity (I Pbody,measured ), we first measured the maximum intensity of P body and then drew a one pixel circle around it to find the surrounding pixels.
9 pixels including the maximal one were averaged to get the I Pbody,measured . Since the correction was based on the assumption that all fluorescence intensity derived from the object, we applied it only to the incremental intensity of the P body over the cytoplasm (I Pbody,measured -I cyto ); cytoplasmic intensity also contributes to I Pbody, measured , but is homogenous across the cell and should not be corrected for the PSF effect. The real maximum intensity of P body(I Pbody ) was thus calculated as [(I Pbody,measured -I cyto )/CF + I cyto ].
Cytoplasm intensity was calculated by averaging the mean intensities of three ROIs the same size as P bodies in the cytoplasm.
We used standard curves of the fluorescence intensities of GFP solutions imaged with identical parameters as yeast to convert I Pbody and I cyto to absolute concentrations in P bodies (C Pbody ) and the cytoplasm (C cyto ). Because the intensities of P bodies marked by different proteins have a large dynamic range, to avoid saturation of our camera, we imaged them using different laser powers and gain settings, and generated different GFP standard curves accordingly. Partition coefficients were calculated as C Pbody /C cyto .

Fraction of protein mass in P bodies attributable to the core proteins
The core P body proteins are Dcp2, Edc3, Pat1, Xrn1, the Lsm1-7 complex, Upf1, and Dhh1. We assumed that the 12 reported P body proteins that lacked clear punctate localization in our studies here (Table S2) have the same P body concentrations as the least concentrated regular protein, Psp2 (Table S1). The mass of the ith protein in P bodies was calculated as the absolute P body concentration (C i ) multiplied by the molecular weight (MWT i ). The fraction of protein mass of the core proteins was  Table S1.

Total fractions in P bodies and number of molecules per cell
We collected z-stacks of yeast cells with a 0.22 µm step size. To calculate the number of molecules in cytoplasm (N cyto ), we first measured the cell volume. Diameters in x, y and z directions were measured manually with Otsu thresholding to determine the cell boundaries in Fiji. Cell volume was calculated as V cell = 4/3*π*(x/2)*(y/x)*(z/2), assuming that yeast cells are ellipsoidal. Previous studies have shown that the cytoplasmic volume of a yeast cell is about 67% of the total cell volume (Uchida et al., 2011;Yamaguchi et al., 2011). To calculate the number of molecules in P bodies (N Pbody ), we measured the x-y diameter of each P body in the cell and calculated its volume by assuming a spherical shape. The concentrations in the cytoplasm (C cyto ) and P body (C Pbody ) were determined as described above. N cyto and N Pbody were calculated as N cyto = V cell x 0.67 x C cyto x N A (6.02 x 10 23 ). N Pbody = V Pbody x C Pbody x N A (6.02 x 10 23 ). The fraction of molecules in P bodies is thus N Pbody /(N cyto + N Pbody ).

Fluorescence recovery after photobleaching (FRAP)
Selected P bodies were bleached using an 0.5W 488 nm laser at 60% laser power.
Images were collected from a single plane using a 2.5 airy unit pinhole at 5 s intervals for (I background ) were first subtracted. Because yeast cells are small, the cytoplasm may be bleached slightly while bleaching the P bodies. We thus measured the average fluorescence intensities of cytoplasm (excluding the bleached P body) before bleaching (I cytobefore ) and in the first frame after bleaching (I cytoafter ) to account for this effect. An unbleached P body was used to correct for the photo-bleaching during image acquisition in the recovery phase (I unbleached ). The corrected recovered intensities (I recovery ) were normalized to the intensities pre-bleaching (I pre-bleaching ). where I 0 , I ¥ and k were fit as intensity immediately after bleach, intensity at long times and the rate constant for recovery, respectively. The fractional recovery was calculated as: For iFRAP, the whole cytoplasm except one P body was bleached 3 times for a total of 1.5 s using a 0.5W 488 nm laser at 100% laser power. Fluorescence intensities were analyzed as in the FRAP experiments above. Because intensities of the unbleached P body were likely affected during bleaching, we normalized the intensity to the intensities of P body in the first frame after photo-bleaching (I pre ).  where I 0 , I ¥ and k were fitted parameters. The fractional decay was calculated as:

Quantifications and Statistical Analysis
Detailed statistics including number of cells analyzed, mean value, standard deviation and standard error of the mean are indicated in each figure legend. The Wilcoxon ranksum test was performed using GraphPad Prism (GraphPad software). Significance was determined as: ***, p < 0.001; ****, p < 0.0005.

Fractions in P bodies
Dcp2 ΔH1 WD