To identify RNAs that associate with P granules in vivo, we performed individual-nucleotide resolution UV crosslinking and immunoprecipitation (iCLIP) experiments (Huppertz et al., 2014) on MEG-3 and PGL-1 proteins tagged at each locus with GFP. We chose these two proteins because MEG-3 and PGL-1 are essential (with their respective paralogs MEG-4 and PGL-3) to assemble the gel (MEG) and liquid (PGL) phases of embryonic P granules (Updike et al., 2011; Hanazawa et al., 2011; Putnam et al., 2019). Early embryos (1 to 100 cell stage) were exposed to ultraviolet light, lysed, and the cross-linked protein/RNA complexes were immunoprecipitated using an anti-GFP antibody (Figure 1—figure supplement 1A). As a control, we also used embryos expressing GFP alone. The GFP immunoprecipitates were washed stringently, lightly treated with nuclease to trim the bound RNAs, extracted for RNA and deep-sequenced. Sequencing reads were mapped back to the C. elegans genome (ws235) and used to determine read counts per locus. Read counts obtained in the control GFP iCLIP were used to define a background threshold. We identified 657 transcripts that were reproducibly recovered above the GFP background threshold in two independent MEG-3::GFP iCLIP experiments (‘MEG-3-bound transcripts’; Figure 1B, Figure 1—figure supplement 1B and Supplementary file 1). In contrast, we identified only 18 transcripts above background in the two PGL-1::GFP iCLIPs, despite abundant PGL-1::GFP protein in the immunoprecipitates (Figure 1B, Figure 1—figure supplement 1A and Supplementary file 1). 15/18 of the PGL-1-bound transcripts were also in the MEG-3-bound list (Supplementary file 1). We compared the average normalized read count (RPKM) across the two iCLIPs to transcript abundance in P blastomeres (Figure 1C and Figure 1—figure supplement 1C) or in whole embryos (Figure 1—figure supplement 1D), and detected no strong correlation, suggesting that MEG-3 binds a specific subset of mRNAs. Low abundance mRNAs, however, were underrepresented in the iCLIPs and therefore could have been missed in our analysis (Figure 1—figure supplement 1C–D).

Among the MEG-3-bound transcripts were nos-2, pos-1, mex-1 and gld-1, the four transcripts previously reported to localize to P granules (Figure 1C). To determine whether other MEG-3-bound transcripts localize to P granules, we used single molecule fluorescent in situ hybridization (smFISH) (Raj et al., 2008) to examine the distribution of various transcripts in embryos. We initially analyzed nine transcripts: nos-2, six other transcripts in the MEG-3-bound list, and two transcripts recovered in the MEG-3 iCLIPs that did not meet the GFP background cut off. For each transcript, we determined the average granule size in the P₂ blastomere and compared that to the average raw read count in the MEG-3::GFP iCLIPs and observed a strong correlation (R = 0.92) (Figure 1D). In this analysis, nos-2 clustered with two other genes also in the MEG-3-bound list (F35G2.1 and F35C11.5). Extrapolating from this correlation, we predicted that transcripts that ranked higher than the nos-2 cluster in the MEG-3-bound list should all localize to P granules. In total, we examined 18 transcripts among this 492-gene set and found that, as predicted, all localized robustly to P granules (Figure 1E, Figure 1—figure supplement 1E and Figure 1—figure supplement 2). We also examined seven transcripts that ranked below the nos-2 cluster and found none that localized to P granules in all P blastomeres, although we observed occasional, weak localization to P granules (Figure 1—figure supplement 2; csr-1, R04D3.3, sip-1, ZC155.4, gly-20, T20F5.7 and fib-1). Finally, we examined six genes that were not recovered reproducibly in the iCLIPs and had above average expression in P blastomeres (RPKM = 7). We found none that localized to P granules (Figure 1E (T26A5.2), Figure 1—figure supplement 2). We conclude that ranking at or above the nos-2 cluster in the MEG-3-bound list is a stringent metric for predicting transcripts with robust P granule localization. We designate this 492-gene set as ‘P granule transcripts’ (Supplementary file 1). This gene list corresponds to the top 75% of the 657 MEG-3-bound transcripts, and ~3% of all transcripts expressed in early embryos (15,345 transcripts detected by RNAseq) (Lee et al., 2017).

For all 18 P granule transcripts analyzed by smFISH, localization to P granules was inefficient with many molecules also found diffusely in the cytoplasm (Figure 1E, Figure 1—figure supplement 1E, Figure 1—figure supplement 2). We calculated the number of molecules in P granules and cytoplasm for two transcripts among the top five ranked in the MEG-3-bound list. We found that only 21 ± 3% puf-5 and 34 ± 3% Y51F10.2 molecules localized to P granules in the P₂ blastomere. Because P granules occupy only a small portion of the P₂ cell volume (5.9 ± 2%), this enrichment corresponds to a ~ 6 fold increase in concentration over the cytoplasm.

The iCLIP results suggest that mRNAs are recruited to P granules as part of the MEG phase rather than the PGL phase. Consistent with this hypothesis, nos-2 mRNA still localized to granules in embryos lacking pgl-1 and pgl-3, but not in embryos lacking meg-3 and meg-4 (Figure 1F). We obtained similar results in in situ hybridization experiments against poly-A RNA and against Y51F10.2, one of the transcripts identified in both the MEG-3-bound and PGL-1-bound lists (Figure 1—figure supplement 1F, Supplementary file 1). In Drosophila, mRNAs have been proposed to be recruited to germ granules via interaction with piRNAs complexed with the PIWI-class Argonaute Aubergine (Vourekas et al., 2016). We found that nos-2 and Y51F10.2 still localized to P granules in embryos lacking the PIWI-class Argonaute PRG-1 (Figure 1F and Figure 1—figure supplement 1F). We conclude that mRNA recruitment to embryonic P granules depends on MEG-3 and MEG-4 and does not require PGL-1 and PGL-3 or the Argonaute PRG-1.

The majority of reads in the MEG-3::GFP iCLIPs mapped to protein-coding mRNA transcripts (Figure 2—figure supplement 1). The iCLIP protocol yields short (~10–30 bp) reads that correspond to sequences cross-linked to MEG-3::GFP (‘toeprints’) (Huppertz et al., 2014). Metagene analysis of the MEG-3::GFP toeprints revealed that MEG-3 binds transcripts throughout the coding and UTR regions, with a preference for 3’UTR sequences (Figure 2A, Figure 2—figure supplement 1A). We analyzed the MEG-3 toeprints for possible motifs and found no evidence for any sequence preference (Materials and methods). MEG-3-bound transcripts tended to be longer on average than other embryonic transcripts (Figure 2B) and were enriched for transcripts known to be targeted by translational repressors expressed in oocytes, including OMA-1, GLD-1 and LIN-41 and CGH-1 (Boag et al., 2008; Scheckel et al., 2012; Tsukamoto et al., 2017) (Figure 2—figure supplement 1B). Ribosome profiling experiments confirmed that MEG-3-bound transcripts are on average less protected by ribosomes than other embryonic transcripts (Figure 2C, Figure 2—figure supplement 2A for P granule transcripts). To determine whether low ribosome coverage can be used to predict P granule localization, we ranked embryonic mRNAs based on ribosome occupancy. We focused this analysis on a set of mRNAs previously defined as enriched in P blastomeres (Lee et al., 2017) to avoid mRNAs transcribed in somatic cells which could complicate the analysis. We identified 19 mRNAs that ranked in the lowest ribosome occupancy class (ribosome occupancy <0.1; Figure 2—figure supplement 2B). 65% of these mRNAs were among the ‘P granule transcripts’ set defined above, including cbd-1 which strongly localizes to P granules (Figure 1E, Figure 1—figure supplement 2). Among the remainder, we selected three transcripts for analysis by in situ hybridization and found that all three localized weakly to P granules. These three transcripts also exhibited a low ranking in the MEG-3::GFP iCLIPs (gly-20, ZC155.4 and R04D3.3, Figure 1—figure supplement 2). We conclude that low ribosome occupancy is one criterion for enrichment in P granules, although efficiency of enrichment in P granules may depend on other factors.

The profiling data suggest that MEG-3 does not bind to any specific RNA sequence and simply favors ‘free mRNAs’ not covered by ribosomes. If so, under conditions where translation is globally inhibited, previously cytoplasmic, well translated mRNAs might be expected to re-localize to P granules. We found that a brief incubation at 30°C (15 min heat shock) was sufficient to disassemble polysomes in embryos (Figure 2—figure supplement 2C) (McCormick and Penman, 1969; Shalgi et al., 2013). We analyzed five non-P granule transcripts, chosen for their high ribosome occupancy under non-heat shock conditions, and remarkably found that all five accumulated in P granules after heat shock (Figure 2D–E and Figure 2—figure supplement 2D). Accumulation in P granules was observed in wild-type embryos, but not in embryos depleted of MEG proteins (Figure 2—figure supplement 2D). We conclude that global translational inhibition is sufficient to recruit new transcripts to P granules.

We noticed that MEG-3::GFP condensates became larger upon heat shock. In zygotes, MEG-3 molecules exist in two states: a fast-diffusing, dilute pool in the cytoplasm and slow-diffusing condensed pool in P granules (Wu et al., 2019). Growth of MEG-3 granules under heat shock suggested that additional MEG-3 molecules condense in P granules under conditions when translation is repressed globally. To examine this further, we measured the size of MEG-3::GFP granules in embryos treated with drugs that block translation. For these experiments, we used mex-5 mex-6 mutant embryos, which lack embryonic polarity and assemble MEG-3 granules throughout the cytoplasm (Smith et al., 2016). We found that treatment with puromycin, which causes ribosomes to dissociate from transcripts, increased the size of MEG-3::GFP granules by ~4 fold, as also observed following heat-shock (Figure 2F and G). In contrast, treatment with cycloheximide, which stalls ribosomes on transcripts, did not change the size of MEG-3::GFP granules (Figure 2F and G). These findings parallel the divergent effect of puromycin and cycloheximide on the assembly of stress granules (Kedersha et al., 2000; Aulas et al., 2017). These results confirm that recruitment to P granules is driven more by translational status than by specific mRNA sequences.

Among the 18 P granule transcripts analyzed by in situ hybridization, we noticed two (nos-2 and Y51F10.2) that transitioned to a more diffuse cytoplasmic localization in the last P blastomere, the germline founder cell P₄ (Figure 3A,B). All other transcripts, in contrast, remained in P granules and their levels diminished starting in the P₄ stage as is typical for maternal mRNAs (Figure 3—figure supplement 1A) (Seydoux and Fire, 1994). As mentioned above, nos-2 codes for a Nanos homolog that becomes translated in P₄ (Subramaniam and Seydoux, 1999). Y51F10.2 is a new P granule transcript that had not been characterized before. We tagged the Y51F10.2 open reading frame with a small epitope by genome editing, and found that like nos-2, Y51F10.2 is translated specifically in P₄ (Figure 3B). nos-2 translation is regulated by proteins that compete for binding to the nos-2 3’ UTR (Jadhav et al., 2008). In particular, translation is repressed prior to the P₄ stage by the RNA-binding protein MEX-3 and activated in P₄ by the RNA-binding protein POS-1 (Jadhav et al., 2008) (Figure 3A). We found that the same is true for Y51F10.2 (Figure 3B). In mex-3(RNAi) embryos, we detected NOS-2 and Y51F10.2 proteins precociously as early as the 4 cell stage. In contrast, in pos-1(RNAi) embryos, NOS-2 and Y51F10.2 proteins were not expressed (Figure 3A,B). Remarkably, we found that these mex-3 and pos-1 RNAi treatments had opposite effects on RNA localization. In mex-3(RNAi) embryos, nos-2 and Y51F10.2 mRNAs did not localize to P granules. In contrast, in pos-1(RNAi) embryos, nos-2 and Y51F10.2 mRNAs remained in P granules through P₄ (Figure 3A,B). These observations confirm a link between P granule localization and translational repression.

Translational repression could be a cause and/or an effect of localization to P granules. To explore this possibility, we examine the translational status of nos-2 and Y51F10.2 in meg-3 meg-4 mutants which lack P granules. Surprisingly, we found that the translational timing of nos-2 and Y51F10.2 was unaffected in meg-3 meg-4 mutants. nos-2 and Y51F10.2 were translationally silent prior to the P₄ stage and began translation in P₄ in meg-3 meg-4 mutants as in wild-type (Figure 3A,B), although protein levels appeared reduced (see below). To examine the translational status of other mRNAs in meg-3 meg-4 mutants, we repeated the ribosome profiling experiments in meg-3 meg-4 embryos. We found that MEG-3-bound transcripts as a class maintained low ribosome occupancy in meg-3 meg-4 mutants as in wild-type (Figure 3—figure supplement 1B). Only 25 embryonic mRNAs showed differential ribosomal occupancy in meg-3 meg-4 embryos versus wild-type, and strikingly most showed lower ribosome occupancy (Figure 3—figure supplement 1C). These results confirm that neither localization to P granules nor binding to MEG-3 is a requirement for translational silencing. Translation silencing, however, appears to be a requirement for localization to P granules, with translational activation correlating with P granule exit.

We noticed that nos-2 and Y51F10.2 transcript and protein levels were lower in the germline founder cell P₄ in meg-3 meg-4 mutants compared to wild-type (Figure 3A,B). Germline blastomeres are generated by a series of asymmetric divisions that give rise to somatic blastomeres and successive germline blastomeres (Figure 1A) (Strome and Wood, 1982). In wild-type embryos, P granule-associated mRNAs segregate preferentially to germline blastomeres with the P granules during each asymmetric division (Figure 3A–B). In contrast, in meg-3 meg-4 mutants, P granule-associated mRNAs segregate symmetrically at each division, resulting in lower levels in germline blastomeres compared to wild-type (Figure 3A–B). Maternal mRNAs are degraded more rapidly in somatic blastomeres than in germline blastomeres (Seydoux and Fire, 1994), and this post-division asymmetry is maintained in meg-3 meg-4 embryos (Figure 3A–B). As expected for equal segregation to somatic blastomeres followed by degradation, we found that P granule transcripts were present on average at lower levels overall in meg-3 meg-4 embryos compared to wild-type as determined by RNAseq (Figure 3C). We conclude that recruitment into P granules serves to enrich maternal mRNAs in germline blastomeres, where mRNAs are stabilized by a P granule-independent mechanism.

30% of meg-3 meg-4 mutants develop into sterile adults (Figure 3D) (Wang et al., 2014), raising the possibility that failure to enrich maternal mRNAs in the germline founder cell P₄ compromises germline development. This hypothesis predicts that meg-3 meg-4 mutants should be hyper-sensitive to loss of P granule-associated mRNAs coding for germ cell fate determinants. To explore this prediction, we examined the effect of combining deletions in nos-2 and Y51F10.2 with deletions in meg-3 and meg-4. Embryos derived from mothers carrying deletions in nos-2 or Y51F10.2 were close to 100% fertile (Figure 3D). In contrast, Y51F10.2; nos-2 double mutant mothers laid 46 ± 15% sterile progeny that lacked a germline (maternal-effect sterility), suggesting that Y51F10.2 functions redundantly with nos-2 in germ cell fate specification (Figure 3D). Remarkably, the majority of nos-2; meg-3 meg-4 and Y51F10.2; meg-3 meg-4 triple mutants grew into sterile animals lacking a germline (Figure 3D). This synthetic effect is consistent with a role for meg-3 and meg-4 in concentrating transcripts coding for germ cell fate regulators, including nos-2 and Y51F10.2, beyond a threshold required for robust germline development. The higher penetrance maternal-effect sterility of nos-2; meg-3 meg-4 mutants compared to Y51F10.2; nos-2 mutants suggests that other P granule transcripts also contribute to germ cell fate (Figure 3D). We conclude that the primary function of the MEG-3 phase is to preferentially segregate maternal mRNAs to the germline founder cell P₄ to ensure robust germ cell specification. Importantly, the MEG-3 phase is NOT essential for translational silencing or for preferential stabilization of maternal mRNAs in germline blastomeres compared to somatic blastomeres.

The iCLIP toeprints suggest that MEG-3 binds mRNAs with no sequence specificity. We showed previously that recombinant MEG-3 readily condenses with total RNA extracted from C. elegans embryos (Putnam et al., 2019). To investigate whether MEG-3 shows any bias when presented with specific sequences, we synthesized nine fluorescently labeled RNAs (800-1300nt size range) corresponding to embryonic transcripts with strong, minimal, or no localization to P granules in vivo under normal culture conditions. Each transcript (20 ng/µL) was combined with recombinant His-tagged MEG-3 (500 nM; Figure 4—figure supplement 1A) in condensation buffer containing 150 mM salt. In the absence of RNA, MEG-3 formed irregular assemblies with a broad size range (Figure 4A). Addition of RNA led to the formation of more uniformly sized assemblies with radii of less than 400 nm in size (Figure 4A and Figure 4—figure supplement 2A,B), consistent with the size of MEG-3 condensates in vivo (Putnam et al., 2019). We used this size difference to distinguish between ‘aggregates’ that form independent of RNA, and ‘condensates’ that form with RNA (Figure 4B; Materials and methods). We found that all nine transcripts stimulated the formation of MEG-3 condensates and became enriched in the condensates with similar efficiencies (Figure 4B–C, Figure 4—figure supplement 2C). We observed no RNA condensates in the absence of MEG-3 even at high RNA concentrations (80 ng/µL), confirming that our conditions do not induce RNA-only aggregation (Figure 4—figure supplement 1B).

MEG-3 has a predicted pI of 9.3 (ExPASy) and thus could potentially interact with the sugar-phosphate backbone of RNA through electrostatic interactions. To explore this hypothesis, we examined MEG-3 condensation behavior in the presence of varying concentrations of salt and RNA (nos-2 transcript). Again, we distinguished aggregates from condensates based on size. In the absence of RNA, MEG-3 formed aggregates under all salt concentrations tested (Figure 4D, Figure 4—figure supplement 2D). In high salt (500 mM) conditions, MEG-3 continued to form aggregates even in the presence of nos-2 RNA and these aggregates did not recruit nos-2 RNA (Figure 4D, Figure 4—figure supplement 2D,E). Decreasing salt concentrations and increasing RNA concentrations shifted the balance from MEG-3 aggregates to MEG-3/RNA condensates. Remarkably, at the lowest concentrations of NaCl, high concentrations of RNA caused MEG-3 to solubilize with no visible aggregates or condensates (Figure 4D, Figure 4—figure supplement 2D,F,G). These observations are consistent with MEG-3 interacting with RNA in a salt-sensitive manner and suggest that electrostatic interactions with RNA compete with the MEG-3/MEG-3 interactions that lead to aggregation.

To determine whether RNA length affects MEG-3 condensation behavior, we tested RNAs of varying sizes in the condensation assay. We found that short RNAs (30 and 100 nt) were not as efficient as longer RNAs at stimulating MEG-3 condensation (Figure 5A,B, Figure 5—figure supplement 1A). We previously showed that MEG-3 protein becomes immobilized in MEG-3/RNA condensates, with no MEG-3 exchange detected within one minute of condensation (Putnam et al., 2019). To determine whether RNAs also become trapped in MEG-3 condensates, we performed FRAP experiments to measure the rate of RNA exchange between dilute and condensed phases comparing RNAs of different lengths. We found that short RNAs (30 and 100 nt) were mobile in MEG-3 condensates. In contrast, longer RNAs (200 nt and higher), including four full-length transcripts, exhibited no detectable exchange (Figure 5C,D, Figure 5—figure supplement 1B). We conclude that MEG-3 interacts most efficiently with mRNA-sized RNAs, which become trapped in the MEG-3 condensates.

To examine whether mRNAs also associate stably with the MEG phase of P granules in vivo, we labeled permeabilized live embryos with the RNA dye SYTO 14. As expected, we observed intense SYTO 14 fluorescence in MEG-3-positive granules (Figure 5E, Figure 5—figure supplement 1C). We verified in vitro that SYTO 14 fluorescence is sensitive to RNA and does not interact significantly with MEG-3 or PGL-3 in the absence of RNA (Figure 5—figure supplement 1D). When released from embryos by laser puncture of the eggshell, MEG-3 granules remain stable in aqueous buffer, whereas PGL-1 and PGL-3 dissolve immediately (Putnam et al., 2019). We found that MEG-3 granules remained positive for SYTO 14 ex vivo for over 1.5 min (the maximum time tested; Figure 5E,F). These observations indicate that mRNAs associate stably with the MEG phase ex vivo as observed in vitro, and confirm that the majority of RNAs in embryonic P granules do not reside in the PGL phase.

P granules were the first RNA granules reported to behave like ‘liquid droplets’ based on observations of the PGL phase (Brangwynne et al., 2009). In this study, we demonstrate that the PGL phase of P granules is neither necessary not sufficient for RNA recruitment to embryonic P granules, which occurs through an independent gel-like phase organized by MEG-3 (and its paralog MEG-4). Several lines of evidence indicate that mRNAs are recruited to P granules by direct binding to MEG proteins. First, iCLIP experiments demonstrate a strong correlation between MEG-3-binding and P granule localization (Figure 1D). Second, localization to granules requires meg-3 and meg-4 and does not require pgl-1 and pgl-3 (Figure 1F and Figure 1—figure supplement 1F). Third, mRNAs in P granules are in a phase that is resistant to dilution and high temperature like the MEG phase and unlike the PGL phase. Finally, despite lacking a canonical RNA binding domain, MEG-3 binds RNA robustly in vitro (Smith et al., 2016) and in vivo (Figure 1—figure supplement 1A), and properties of MEG-3/RNA condensates assembled in vitro match properties of the MEG phase in vivo (Figure 4 and Figure 5).

We previously showed that the MEG phase is gel-like (Putnam et al., 2019) and we demonstrate here that mRNAs become kinetically trapped in the MEG phase. Low RNA dynamics have also been reported in the germ granules of other organisms (Jamieson-Lucy and Mullins, 2019; Trcek and Lehmann, 2019). Specification of the germline using germ granules and other asymmetrically-segregated maternal factors (germ plasm) is a derived trait that arose independently several times during metazoan evolution (Kulkarni and Extavour, 2017). Proteins that scaffold germ granule assembly in different organisms DrosophilaOskar, vertebrate Xvelo/Bucky Ball, and C. elegans MEGs) are non-homologous, but all contain predicted intrinsically-disordered domains and form non-dynamic condensates in vitro. Oskar and Bucky Ball bind Nanos RNA in vitro (Yang et al., 2015; Krishnakumar et al., 2018), as we show here for MEG-3. Condensation of disordered protein domains with RNA, therefore, may be a common driver of germ granule assembly in a wide range of animals. Trapping RNAs in a solid, gel-like phase could be beneficial for long term storage of RNAs during oogenesis and for maintaining a pool of maternal mRNAs in germline precursors before the onset of zygotic transcription (P₄ stage; Seydoux and Dunn, 1997).

What defines which mRNAs are recruited into germ granules? We report here that recruitment of nos-2 RNA to P granules requires MEX-3, a translational repressor that recognizes specific sequences in the nos-2 3’ UTR (Jadhav et al., 2008). Similarly, in vertebrate embryos, mRNA recruitment to the Balbiani body depends on 3’ UTR sequences recognized by sequence-specific mRNA-binding proteins (Jamieson-Lucy and Mullins, 2019). In Drosophila, recruitment to polar granules also depends on 3’UTR sequences implicated in translational repression, as well as potential RNA:piRNA and RNA:RNA homotopic interactions (Trcek and Lehmann, 2019). We suggest that sequence-specific protein:RNA and/or RNA:RNA interactions involving 3’ UTR sequences define a pool of low translation mRNAs. By virtue of its low ribosome occupancy, this pool is preferentially captured by non-sequence specific, intrinsically-disordered proteins in germ plasm for condensation and segregation to the embryonic germline.

Our findings highlight several similarities between embryonic P granules and stress granules. Stress granules are RNA granules that form in the cytoplasm of stressed cells to store mRNAs that have exited the translational pool (Buchan and Parker, 2009; Ivanov et al., 2019). As we report here for P granules, the stress granule transcriptome favors long mRNAs that are ribosome-depleted (Kedersha et al., 2000; Khong et al., 2017; Aulas et al., 2017). Stress granule assembly is enhanced by treatments that release mRNAs from polysomes (heat-shock and puromycin), as we show here for P granules. Recruitment of mRNAs to stress granules is an inefficient process, with only a minority of molecules for most mRNA species localizing in the granules (Khong et al., 2017). Recruitment into P granules is also inefficient with only ~30% of molecules localizing to granules for two of the most robust P granule transcripts described here. Translational repression is required for localization to P granules, but P granules are not required for translational repression or for mRNA stability, as is also true for stress granules (Kedersha et al., 2016). Finally, P granules contain proteins also found in stress granules (poly-A binding protein, TIA-1 and the DDX3/LAF-1 RNA helicase) and P granules interact with P bodies, as do stress granules (Gallo et al., 2008; Shih et al., 2012; Elbaum-Garfinkle et al., 2015). Our findings suggest that, like stress granules, embryonic P granules function downstream of the translational regulation machinery to temporarily hold mRNAs not engaged with ribosomes until their degradation or translation in the germline blastomere P₄. A recent pre-print (Parker et al., 2020) also reports that translational repression is a prerequisite for RNA localization to P granules. Interestingly, P bodies also have been reported to favor translationally repressed mRNAs and to recruit additional mRNAs under translational stress (Hubstenberger et al., 2017; Matheny et al., 2019). Low translation, therefore, may be a common requirement for entry into cytoplasmic granules. Our analyses indicate, however, that not all translationally-repressed mRNAs are recruited equally strongly to P granules. A key question for the future will be to understand what factors beyond low translation contribute to enrichment in P granules.

What is the function of P granules? A unique characteristic of P granules is their polarized assembly (Strome and Wood, 1982). P granules assemble preferentially in cytoplasm destined for germline blastomeres (Smith et al., 2016) and consequently mRNAs in P granules are preferentially segregated to germline blastomeres during the first embryonic cleavages. Because mRNA localization to P granules is inefficient, this enrichment is relatively weak but, over 4 cell divisions, boosts mRNAs levels in the germline founder cell P₄ over what would have been achieved by equal segregation. meg-3 meg-4 mutants, which lack P granules, segregate P granule mRNAs equally to germline and somatic blastomeres, leading to reduced levels in P₄ compared to wild-type. meg-3 meg-4 mutants are 30% sterile and exquisitely sensitive to loss of germ cell fate regulators. These observations are consistent with previous studies, which showed that embryonic P granules, while non-essential, are required for robust germ cell fate specification (Gallo et al., 2010). Embryonic P granules are also required to maintain small RNA homeostasis across generations, likely via the PGL phase which contains Argonaute proteins and other epigenetic factors (Ouyang et al., 2019; Dodson and Kennedy, 2019). We propose that embryonic P granules are two-phase, dual-cargo condensates whose main function is to maximize transmission of maternal mRNAs (MEG phase) and epigenetic factors (PGL phase) to the germline founder cell P₄. Importantly, our findings demonstrate that P granules are not essential for translational repression or preferential RNA stability in germline blastomeres.

What P granule mRNAs code for germ cell fate regulators? Our analyses identified 492 predicted ‘P granule transcripts’. This list is unlikely to be exhaustive, especially since low abundance mRNAs were not recovered efficiently in the MEG-3 iCLIP experiments. We surveyed 18 P granule transcripts by in situ hybridization and identified two that exit P granules and become translated in the germline founder cell P₄: the known germ cell fate regulator nos-2 and a previously uncharacterized transcript Y51F10.2. Y51F10.2 codes for the C. elegans orthologue of human TRIM32, a member of the broadly conserved TRIM-NHL protein family implicated in protein ubiquitination and mRNA regulation (Tocchini and Ciosk, 2015). Our genetic findings indicate that 1) Y51F10.2 functions in germ cell fate specification redundantly with nos-2 and 2) additional germ cell fate regulators likely exist among the other P granule mRNAs. It appears unlikely, however, that all P granule-enriched transcripts function in germ cell specification. Unlike nos-2 and Y51F10.2, the other P granule mRNAs we surveyed by in situ hybridization turn over before ever releasing from P granules (Figure 3—figure supplement 1), and at least some are known to function in unrelated processes (e.g. cbd-1; a chitin-binding protein required for egg shell biogenesis Johnston et al., 2010). The MEG phase, therefore, may serve as a relatively non-specific repository for untranslated or low-translation mRNAs, many of which do not function in germ cell fate specification. It is interesting to note that the first step for mRNA selection in the Drosophila polar granules also has been proposed to rely on a relatively non-specific mechanism, in that case involving mRNA:piRNA interactions (Vourekas et al., 2016). Similarly, recruitment of mRNAs to stress granules is mostly non-specific with over 80% of mRNA species recruited (Khong et al., 2017). Sequence non-specific RNA condensation may therefore be a common strategy to spatially segregate mRNAs in cells and embryos.

C. elegans were cultured according to standard methods (Brenner, 1974). RNAi knockdown experiments were performed by feeding on HT115 bacteria (Timmons and Fire, 1998). Feeding constructs were obtained from Ahringer or OpenBiosystem libraries. The empty pL4440 vector was used as negative control. Bacteria were grown at 37°C in LB + ampicillin (100 µg/mL) media for 5–6 hr, induced with 5 mM IPTG for 30 min, plated on NNGM (nematode nutritional growth media) + ampicillin (100 µg/mL) + IPTG (1 mM) plates, and grown overnight at room temperature. Embryos isolated by bleaching gravid hermaphrodites were put onto RNAi plates directly. To culture larger number of worms for iCLIP and ribosome profiling experiments, worm cultures were started from synchronized L1s (hatched from embryos incubated in M9 overnight) onto NA22 or RNAi bacteria containing plates and grown to gravid adults. Early embryos were harvested from gravid adults.

To verify the efficiency of RNAi treatments for knocking down meg genes, we scored animals exposed to the same RNAi feeding conditions for maternal-effect sterility. For meg-1(vr10) strain on meg-2 RNAi, sterility was 95 ± 0.6% at 20°C; meg-1 meg-3 meg-2(RNAi) meg-4(RNAi) maternal effect sterility was 100 ± 0%. To verify the RNAi efficiency of targeting mex-3 and pos-1, embryonic lethality was assayed. Cohorts of 10–20 mothers were allowed to lay eggs for periods ranging from 2 to 4 hr. Embryos were then counted, and adults were scored four days later. The embryonic lethality for both mex-3 and pos-1 were 100 ± 0% and 98 ± 0.2% respectively.

CRISPR generated lines were created as in Paix et al. (2017) as indicated in the strain in the key resources table. Guides and repair temples used for CRISPR are listed in the key resources table.

Synchronized L1 worms were seeded on plates containing HT115 bacteria transformed with pL4440 vector and cultured at 25°C for ∼48 hr. Additional bacteria were added to ensure enough food to support development. Early embryos were collected by bleaching gravid hermaphrodites. Small aliquots of embryos were collected from each experiment and staged by DAPI-staining. 70 ± 7% of embryos were before or at ~100 cell stage. ∼400 μL packed embryos were then resuspended in 2 mL footprint lysis buffer [20 mM Tris-Cl (pH8.0), 140 mM KCl, 1.5 mM MgCl2, 1% Triton X-100, 0.1 mg/mL CHX] and lysed in a Spex 6870 freezer mill. After clarification lysate by sequential centrifugation at 3000 rpm followed by 17,000xg, 100 µL (5% of lysate) of lysates were saved for mRNAseq. For ribosome profiling, lysates containing 300 μg of total RNA were treated with 100 units of RNaseI (Ambion) for 30 min at 25°C. 40 units of SUPERaseIN (ThermoFisher) were added to prevent further digestion from RNaseI. Monosomes were isolated by sucrose gradients (10–50%) and centrifuged at 40,000 rpm for 3 hr in a SW41 rotor (Beckman Coulter). The extracted RNA was size-selected (15–34 nt) after running on 15% denaturing PAGE gels. An oligonucleotide adapter was ligated to the 3’ end of isolated fragments. After ribosomal RNA depletion using RiboZero (Illumina), the following steps were performed: reverse transcription using SuperScript III reverse transcriptase (Thermo Fisher Scientific), circularization using CirLigase I (Lugicen) and PCR amplification (Schuller et al., 2017). Libraries were sequenced on a HiSeq2500 machine at facilities at the Johns Hopkins Institute of Genetic Medicine.

Read counts obtained in the control GFP iCLIP were used to define background threshold (read count = 60.) We defined MEG-3 bound transcripts by excluding transcripts with MEG-3::GFP iCLIP read counts < 60 as shown in Figure 1B (stippled horizontal line at log_e60 = 4.1). To define P granule transcripts, we used the rank order of F35G2.1 (Rank 388), the left most gene in the nos-2 cluster as shown in Figure 1D, as the cut off. Genes with rank order better than 388 in either one of MEG-3::GFP iCLIP were defined as P granule transcripts (Supplementary file 1).

smFISH probes were designed using Biosearch Technologies’s Stellaris Probe Designer. The fluorophores used in this study were Quasar570 and Quasar670. For sample preparation, embryos were extruded from adults on poly-lysine slides (0.01%) and subjected to freeze-crack followed by cold methanol fixation at −20° C. Samples were washed five times in PBS+0.1%Tween20 and fixed in 4% PFA (Electron Microscopy Science, No.15714) in PBS for one hour at room temperature. Samples were again washed four times in PBS+0.1%Tween20, twice in 2x SCC, and once in wash buffer (10% formamide, 2x SCC) before blocking in hybridization buffer (10% formamide, 2x SCC, 200 ug/mL BSA, 2 mM Ribonucleoside Vanadyl Complex, 0.2 mg/mL yeast total RNA, 10% dextran sulfate) for 30 min at 37° C. Hybridization was then conducted by incubating samples with 50–100 nM probe solutions diluted in hybridization buffer overnight at 37° C. Following hybridization, samples were washed twice in wash buffer at 37° C, twice in 2x SCC, once in PBS-Tween20 (0.1%), and twice in PBS. Lastly, samples were mounted using VECTASHIELD Antifade Mounting Media with DAPI or Prolong Diamond Antifade Mountant.

Fluorescence confocal microscopy was performed using a Zeiss Axio Imager with a Yokogawa spinning-disc confocal scanner. Embryo images were taken using Slidebook v6.0 software (Intelligent Imaging Innovations) using a 63x objective. Embryos were staged by DAPI-stained nuclei in optical Z-sections and multiple Z-sections were taken to include germ cells. For In vitro condensation reactions, images are single planes taken using a 40x objective unless otherwise indicated. For fluorescence super-resolution microscopy, images were acquired using ZEISS LSM 880-AiryScan (Carl Zeiss) equipped with a 63X objective. Images were processed using ZEN imaging software (Carl Zeiss). Equally normalized images were exported via either Slidebook v6.0 or ZEN, and contrasts of images were equally adjusted between control and experimental sets. For in vitro fluorescence recovery after photobleaching experiments, images were acquired using Zeiss LSM 800 GaAsp. Images are single confocal planes imaged using a 63x objective every 3 s during a recovery phase of 300 s. All image analyses were conducted using the Fiji image-processing package (http://fiji.sc/Fiji).

As in sample preparation for smFISH experiments, embryos were extruded from adults and subjected to freeze-crack on poly-lysine slides followed by cold methanol fixation for 15 min and then cold acetone for 10 min. Slides were blocked twice in PBS-Tween20 (0.1%)-BSA (0.1%) for 30 min at room temperature, and incubated with 90 μl primary antibody overnight at 4°C in a humidity chamber. Antibody dilutions (in PBST/BSA): mouse K76 1:10 (DSHB), Rat α-OLLAS 1:80 (Gift from Dr. Jeremy Nathans), mouse α-FLAG M2 1:500 (Sigma F3165). Secondary antibodies (Molecular Probes/Thermo Fisher Sci.) were applied for 1 ~ 2 hr at room temperature.

RNA was extracted from embryos or cleared embryo lysates using TRIZOL. The aqueous phase was transferred to Zymo-SpinTM IC Column (Zymo research R1013) for concentration and DNase I treatment as described in manual. RNA quality was assayed by Agilent Bioanalyzer using Agilent RNA 6000 Pico Chip. All RNAs used for library preparation had RIN (RNA integrity number)>9. For mRNA-seq library construction, 0.5 µg of total RNA was treated with Ribo-Zero Gold Epidemiology rRNA Removal Kit. Libraries were then prepared following the TruSeq RNA Library Prep Kit v2 instruction. All sequencing was performed using the Illumina HiSeq2500 at the Johns Hopkins University School of Medicine Genetic Resources Core Facility.

mRNAs were transcribed using T7 or SP6 mMessageMachine (Thermofisher) using manufacturer’s recommendation. 1 µL of ChromaTide Alexa Fluor 488–5-UTP or 546–14-UTP (Thermofisher) were added to transcription reactions to fluorescently trace label mRNAs. Template DNA for transcription reactions was obtained by PCR amplification from plasmids. nos-2 fragments were generated by PCR amplification from the 5’ end of the full length nos-2 template DNA. Free NTPs and protein were removed by lithium chloride precipitation. RNAs were resuspended in water and stored at −20°C. The integrity of RNA products was verified by agarose gel electrophoresis.

30 nt oligo polyU RNAs were ordered from IDT either unlabeled or with a 3’ FAM modification. Oligos were resuspended in water aliquoted and stored at −80°C. Labeled and unlabeled oligo RNAs were mixed together and used at final concentrations of 20 ng/uL including 25 nM fluorescently labeled oligo.

Protein condensation was induced by diluting proteins out of storage buffer into condensation buffer containing 25 mM HEPES (pH 7.4), salt adjusted to a final concentration of 150 mM (37.5 mM KCl, 112.5 mM NaCl), and RNA. Unless otherwise indicated, for all co-assembly experiments, we used 500 nM MEG-3 and 20 ng/µL RNA. MEG-3 solutions contained 25 nM fluorescent trace labels with either 488, 555, or 647 (indicated in figure legends). Condensate reactions with the RNA dye contained a final concentration of 100 nM SYTO 14. Condensation reactions were incubated at room temperature for 30 min or as indicated, before spotting onto thin chambered glass slides (ERIE SCIENTIFIC COMPANY 30-2066A) with a coverslip. Images used for quantification are single planes acquired using a 40x oil objective over an area spanning 171 × 171 μm.

To quantify the ratio of protein or RNA in condensates, a mask was created by thresholding images, filtering out objects of less than four pixels to minimize noise, applying a watershed filter to improve separation of objects close in proximity, and converting to a binary image by the Otsu method using the nucleus counter cookbook plugin. Minimum thresholds were set to the mean intensity of the background signal of the image plus 1–2 standard deviations. The maximum threshold was calculated by adding 3–4 times the standard deviation of the background. Using generated masks, the integrated intensity within each object was calculated. To remove non-specific background signal the mean intensity of an image field in the absence of the labeled component was subtracted from each pixel yielding the total intensity of each object.

Histograms of MEG-3 intensity were generated by taking the log(10) of total intensity for each MEG-3 object, Log(I). Objects in 2–3 experimental replicates of 4 images were identified and quantified as described above the number of objects for each Log(I) value was binned (bin size = 0.2 Log(I) units), and normalized to the total number of objects. The percent of objects in each bin was averaged for each experimental replicate (Figure 4B, Figure 4—figure supplement 2E, Figure 5—figure supplement 1A).

The percent of MEG-3 in aggregates or condensates was determined by comparing histograms of reactions in which all objects are aggregates (500 mM NaCl or no RNA) or condensates (150 mM NaCl and 20–80 ng/µL nos-2 RNA) as illustrated in Figure 4B, Figure 4—figure supplement 2E, Figure 5—figure supplement 1A. The minimum at the intersection of the two conditions was calculated. The percentage of MEG-3 objects with an intensity above or equal to the intersection were classified as aggregates and the fraction of objects with an intensity below the intersection were classified as condensates.

To calculate the fraction of RNA in MEG-3 condensates/aggregates, the background corrected sum of RNA fluorescent intensity in each MEG-3 object was divided by the total intensity of RNA fluorescence in the imaged area (Figure 4C).

Radii of MEG-3 condensates were estimated by imaging condensation reactions of 500 nM MEG-3 and 40 ng/µL RNA (Figure 4—figure supplement 2A). Images used for quantification were single planes acquired using a 100x oil objective over an area spanning 68 × 68 μm. four experimental replicates of 16 images were identified and quantified (>1500 objects/replicate) as described above, and radii were calculated from the area of each object. Calculated radii are an overestimate and represent upper limits for actual condensate size. The number of objects for each radii was binned (bin size = 0.06 µm), and normalized to the total number of objects. The percent of objects in each bin was averaged for each experimental replicate (Figure 4—figure supplement 2B).

20 µL condensation reactions (prepared as described above) were added to a chambered coverglass (Grace BioLabs) and imaged using a Zeiss LSM 800 GaAsp. Bleaching was performed using 100% laser power in the 488, 546, or 647 channels. Regions slightly larger than the condensates (radius ≈ 3 µM) were photobleached. A single confocal plane was imaged using a 63x objective every 3 s during a recovery phase of 300 s.

FRAP analysis was performed as described in Putnam et al. (2019). Briefly, fluorescence recovery was corrected for background and normalized to the initial granule intensity using the equation: nI = (I-I^bkg)/(Iⁱ-I^bkgi), where nI is the background corrected and normalized fluorescence intensity, I is the intensity of the FRAPed granule, I^bkg is the fluorescence intensity outside of the condensate, Iⁱ is the initial intensity before bleaching, and I^bkgi is the initial background intensity. Recovery rates were determined by fitting individual traces to a first order equation nI = (A^rec·(1-e^-kt), where A^rec is the fluorescence recovery amplitude and k is the rate of fluorescence recovery. For RNAs where fluorescence recovery was in the linear range for the entire time course, initial recovery rates were calculated by fitting to a linear equation nI = kt, where k is the initial rate of fluorescence recovery (Figure 5D, Figure 5—figure supplement 1B).

1 mM SYTO 14 RNA Dye (ThermoFisher) and 1 mM JF₆₄₆ (JF₆₄₆, Janelia Grimm et al., 2015) dissolved in DMSO were diluted 500 fold into osmolarity calibrated Egg buffer [118 mM NaCl, 48 mM KCl, 2 mM CaCl₂, 2 mM MgCl₂25 mM HEPES pH 7.3, 340 ± 5 mOsm] to reach 2 µM stock solutions. perm-1(RNAi) adult gravid hermaphrodites expressing MEG-3::HALO (Wu et al., 2019) were dissected and egg shell permeabilized embryos were released into 10 µL egg buffer containing 1 µM SYTO 14 and JF₆₄₆ for 10 min in a humid chamber to prevent evaporation. After drug treatment, embryos were washed 3X with egg buffer without drug. Approximately 200–20 µm polysterene beads (Bangs Laboratories) suspended in egg buffer were added to prevent embryo compression, and placed on slides for imaging. Embryo contents were extruded by puncturing the eggshell near the anterior region of the germline blastomere using a 3i Ablate! laser system at 532 nm pulse setting with a power level of 155 (Putnam et al., 2019). All embryo images are Z stack maximum projections using a Z step size of 1 μm, spanning the depth of the embryo. Images were acquired in the 488 and 647 channel every 10 s using a 63x objective (Figure 5E, Figure 5—figure supplement 1C).

To quantify SYTO 14 and MEG-3::Halo persistence in granules, photomicrographs acquired as described above were analyzed using FIJI. Not all SYTO 14 granules were MEG-3::Halo positive, and are potentially P bodies (Gallo et al., 2008). Only SYTO 14 granules also positive for MEG-3::Halo were quantified. MEG-3::Halo granules were identified using the nucleus counter plugin as described above. The total intensity of objects was quantified for both 488 (SYTO 14) and 647 (Halo) channels. Total fluorescence intensity was calculated before (I_B) and after (I_A) extrusion and used to calculate a fluorescence ratio (I_A/I_B). Photobleaching was minimal for MEG-3::Halo; however it was significant for SYTO 14. To correct for photobleaching, total fluorescence intensity was corrected for photobleaching using the average photobleaching rate calculated from the cytoplasm of intact embryos in the imaging area. For some embryos, granules left the field of view and could not be counted. The I_A/I_B, therefore, is a minimal estimate of the fraction MEG-3 and SYTO 14 that remained in the granule phase after extrusion (Figure 5F).

All data were plotted and statistical analysis was conducted using Graphpad Prism seven software. Fitting of recovery curves in FRAP experiments was conducted using Kaleidagraph (Synergy) software.

Sequencing datasets and processed results generated in this paper are available at GEO accession GSE139881 for iCLIP (GSE139878), embryonic RNAseq (GSE139879) and ribosome profiling results (GSE139880).