Germ Granules Coordinate RNA-based Epigenetic Inheritance Pathways

Summary:

Germ granules are biomolecular condensates that promote germ cell totipotency in animals. In C. elegans, MEG-3 and MEG-4 function redundantly to assemble germ granules in germline blastomeres. Here, we show that meg-3/4 mutant animals exhibit defects in RNA interference (RNAi) that are transgenerationally disconnected from the meg-3/4 genotype. Similar non-Mendelian inheritance is associated with other mutations disrupting germ granule formation, indicating that loss of germ granules is the likely cause of the observed disconnects between genotype and phenotype. meg-3/4 animals produce aberrant siRNAs that are propagated for ≅10 generations in wild-type descendants of meg-3/4 ancestors. Aberrant siRNAs inappropriately and heritably silence germline-expressed genes including the RNAi gene sid-1, suggesting that transgenerational silencing of sid-1 underlies inherited defects in RNAi. We conclude that one function of germ granules is to organize RNA-based epigenetic inheritance pathways and that germ granule loss has consequences that persist for many generations.

Graphical Abstract

eTOC blurb

Parentally deposited small non-coding RNAs direct heritable gene regulation in the C. elegans germline. Dodson and Kennedy provide evidence that biomolecular condensates known as germ granules spatially organize these small RNA-based epigenetic inheritance pathways. Disrupting germ granules triggers changes in small-RNA-based gene regulation that can be inherited across generations.

Introduction:

Cells contain many non-membrane-bound organelles (termed liquid droplet organelles or biomolecular condensates) that consist of proteins and RNAs that self-assemble via liquid-liquid phase separations (Shin & Brangwynne 2017). Examples of biomolecular condensates include nucleoli, processing (P) bodies, Cajal bodies, stress granules, neuronal granules, and germ granules (Shin & Brangwynne 2017). Germ granules are biomolecular condensates that form in the germ cells of many metazoans to help maintain totipotency of the germline (Voronina et al. 2011). The mechanism(s) by which germ granules promote germ cell health are largely unknown (Voronina et al. 2011; Voronina 2013). According to current models, one major function of biomolecular condensates such as germ granules may be to bring specific proteins and nucleic acids together in space and time to help organize the complex RNA processing pathways underlying gene regulation (Toretsky & Wright 2014; Weber & Brangwynne 2012).

Germ granules in C. elegans (referred to as P granules) are present in germ cells during all stages of development (Strome & Wood 1983). During early embryonic cell divisions, P granules assemble asymmetrically in the germline-destined portion of the zygotic cytoplasm (Brangwynne et al. 2009). MEG-3 and MEG-4 are intrinsically disordered proteins that are expressed during early embryogenesis and that function redundantly to nucleate P granule formation in early embryos (Wang et al. 2014; Smith et al. 2016). Another protein, DEPS-1, contributes to P granule formation during most stages of germline development (Spike et al. 2008). For much of development, P granules localize to the outer nuclear membrane directly adjacent to nuclear pores. Multiple lines of evidence suggest that P granules help surveil and/or process mRNAs as they transit through nuclear pores and enter the cytoplasm (Pitt et al. 2000; Sheth et al. 2010). For example, newly synthesized mRNAs have been observed transiting P granules (Sheth et al. 2010), and specific mRNAs concentrate in P granules during specific stages of germline development (Seydoux & Fire 1994). Additionally, a number of RNA quality control proteins, including small RNA pathway components, localize to P granules (see below). Finally, in the absence of P granules, somatic genes become improperly expressed in the germline (Updike et al. 2014; Knutson et al. 2017). Thus, P granules likely function to store and surveil mRNAs in germ cells.

Non-coding RNAs, such as PIWI-interacting RNAs (piRNAs) and endogenous small interfering RNAs (endo-siRNAs), contribute to RNA quality control in many eukaryotes and, in some cases, also transmit epigenetic information from parent to progeny (Bošković & Rando 2018). In C. elegans, many instances of transgenerational epigenetic inheritance (TEI) are directed by endo-siRNAs (Rechavi & Lev 2017). Current models posit that, during TEI, endo-siRNAs are deposited into the embryo via the egg or sperm. Parentally deposited endo-siRNAs then act as guide molecules to identify cognate mRNAs, recruit RNA-dependent RNA Polymerases (RdRPs), and amplify endo-siRNA populations. Repetition of this process each new generation allows endo-siRNA-based gene regulatory information to pass across multiple generations. Heritably maintained endo-siRNA populations bind Argonaute proteins such as HRDE-1 to regulate gene expression in germ cells (Buckley et al. 2012; Shirayama et al. 2012). Many of the genomic loci targeted for heritable silencing by endo-siRNAs in C. elegans are pseudogenes and cryptic loci, suggesting that the endo-siRNA system may have evolved to silence unwanted germline RNAs (Gu et al. 2009). This heritable gene regulatory pathway is likely important for germ cell function, as mutations that disrupt this process also cause a germline mortal (Mrt) phenotype in which germ cell function deteriorates over generations (Buckley et al. 2012; Spracklin et al. 2017; Wan et al. 2018). In summary, C. elegans possess an RNA-based mode of epigenetic inheritance driven by generationally repeated amplification of endo-siRNAs by RdRPs, followed by silencing of unwanted RNAs by germline-expressed Argonautes such as HRDE-1.

Because endo-siRNA-based gene regulation is both complex and heritable, it is likely that cells regulate and organize endo-siRNA biogenesis in ways that prevent runaway heritable silencing of the wrong mRNAs. P granules may provide this organization, as many endo-siRNA pathway proteins are known to localize to P granules, including the RNase III enzyme Dicer, the Dicer-related factor DRH-3, the endo-siRNA-binding Argonautes WAGO-1 and CSR-1, the piRNA-binding Argonaute PRG-1, and the RdRP EGO-1 (Claycomb et al. 2009; Batista et al. 2008; Beshore et al. 2011; Gu et al. 2009). Other endo-siRNA pathway factors localize to other germline biomolecular condensates (WAGO-4 and ZNFX-1 in Z granules and RRF-1 in Mutator foci) that form ordered multi-condensate structures with P granules in adult C. elegans germ cells (Wan et al. 2018; Phillips et al. 2012). Together, the data hint that various C. elegans germ granules may act as organizational hubs that help connect endo-siRNA pathway proteins with the correct mRNAs to help ensure fidelity and accuracy of endo-siRNA-based gene regulation.

Here, we show that mutations that disrupt germ granule formation trigger the production of aberrant endo-siRNAs that inappropriately silence germline-expressed genes. Due to the heritable nature of endo-siRNAs, aberrations in gene expression are inherited across multiple generations, even after germ granules have been restored. We conclude that one function of germ granules is to organize and coordinate RNA-based epigenetic inheritance pathways and that the loss of this organizational function has consequences that can persist for many generations.

Results:

meg-3/4 animals exhibit defects in experimental RNAi.

RNA interference (RNAi) can be triggered experimentally in C. elegans by feeding animals bacteria that express dsRNAs targeting specific C. elegans mRNA sequences (referred to hereinafter as experimental RNAi) (Timmons et al. 2001). Mutations in deps-1 disrupt P granule formation in adult C. elegans germ cells and also lead to defects in experimental RNAi, suggesting that P granules may contribute in some way to small RNA-based gene regulation in germ cells (Spike et al. 2008). To further investigate a potential link between P granules and germline small RNA pathways, we asked if other mutations that affect P granule formation also cause defects in experimental RNAi. MEG-3 and MEG-4 (together, MEG-3/4) are two intrinsically disordered proteins that are redundantly required for P granule assembly during early embryogenesis, but not during larval development or in the adult germline (Wang et al. 2014; Smith et al. 2016). tm4259 and ax2026 are putative null alleles of meg-3 and meg-4, respectively (Wang et al. 2014; Mitani 2009). RNAi targeting either pos-1 or egg-4/5, which are essential germline genes, causes sterility in C. elegans (Fig. 1A and Fig. S1A) (Parry et al. 2009). meg-3(tm4259) meg-4(ax2026) animals remained fertile after either pos-1 RNAi or egg-5 RNAi, suggesting that MEG-3/4 are required for experimental RNAi in germ cells (Fig. 1A). meg-3(tm4259) meg-4(ax2026) animals also failed to silence a germline-expressed gfp reporter gene after exposure to gfp RNAi, confirming that MEG-3/4 contribute to RNAi-based gene silencing in germ cells (Fig. S1B). meg-3(tm4259) meg-4(ax2026) animals responded normally to RNAi targeting genes expressed in the soma, suggesting that the role of MEG-3/4 in promoting experimental RNAi may be restricted to the germline (Fig. S1C) (Wan et al. 2018). ax3055 and ax3052 are independently isolated deletion alleles of meg-3 and meg-4, respectively (Smith et al. 2016). Similar to meg-3(tm4259) meg-4(ax2026) animals, meg-3(ax3055) meg-4(ax3052) animals responded normally to RNAi targeting somatically expressed genes, but failed to respond to RNAi targeting the germline-expressed pos-1 and egg-4/5 genes (Fig. 1A and Fig. S1A,C). The data show that MEG-3/4 contribute to dsRNA-based gene silencing in the germline and thereby support the idea that P granule formation is somehow important for dsRNA-based gene silencing.

Figure 1. Transgenerational disconnect between meg-3/4 genotype and phenotype.

(a) L2 larvae of the indicated genotypes were fed either bacteria expressing dsRNAs derived from the pos-1 gene, which is required for embryonic viability, or bacteria containing the control vector, L4440. When the animals became adults, they were allowed to lay broods, and % hatching embryos was scored. Black dots represent individual broods (n = 18). Error bars represent +/− standard deviations of the mean (gray bars). (b) Top panel: schematic of genetic crosses. We first marked meg-3(tm4259) meg-4(ax2026) with dpy-3(e27) (dpy-3 is 0.1 cM from meg-3 and 0.8 cM from meg-4). Then, meg-3(tm4259) meg-4(ax2026) dpy-3(e27) animals were crossed to wild-type males, and heterozygous progeny were identified by PCR-based genotyping of both meg-3 and meg-4. As a mutant control cross, meg-3(tm4259) meg-4(ax2026) dpy-3(e27) animals were crossed to meg-3(tm4259) meg-4(ax2026) males. Lines that were either homozygous (control cross) or heterozygous for meg-3/4 mutations were maintained for 22 generations (data presented later in the paper will clarify why this was necessary for this experiment). All lines remained heterozygous for dpy-3(e27). The progeny of homozygous or heterozygous meg-3/4 lines were isolated, and meg-3 and meg-4 were genotyped by PCR. These progeny were used to establish 6 lineages that were maintained for 14 generations under normal growth conditions. Bottom panel: at the indicated generations, animals from each lineage were exposed to pos-1 RNAi and % hatching embryos was scored. Black dots represent individual lineages, colored bars represent the median value of % viable progeny. (c) A meg-3(tm4259) meg-4(ax2026) dpy-3(e27) chromosome was maintained in a heterozygous state for ≅45 generations in animals that were homozygous for two fluorescent protein markers: pgl-1::rfp (Wan et al. 2018), which marks P granules (magenta), and gfp::h2b (Ashe et al. 2012), which marks chromatin (green). PGL-1::RFP and GFP::H2B were visualized in first-generation meg-3/4 animals (indicated by Dpy phenotype) and their meg-3/4(+) and ++/meg-3/4 siblings (indicated by non-Dpy phenotype). Fluorescent micrographs of three embryos in the uterus of one adult are shown. Arrows indicate P granules. The percentage of F₁ adults containing embryos with normal PGL-1::RFP expression and the number of adults scored are indicated. Scale bar, 10 microns. See also Figure S1.

Transgenerational disconnect between meg-3/4 genotype and phenotype.

While conducting genetic crosses and RNAi experiments with meg-3(tm4259) meg-4(ax2026) animals (henceforth, meg-3/4), we noticed that the RNAi-defective (Rde) phenotype associated with meg-3/4 was transgenerationally disconnected from the meg-3/4 genotype. For instance, the meg-3/4 offspring of heterozygous (++/meg-3/4) parents surprisingly responded to pos-1 RNAi (Fig. 1B). Amazingly, descendants of these newly generated meg-3/4 mutants did not become Rde until MEG-3/4 function had been absent for 5–9 generations (Fig. 1B). [Note: for reasons that will become apparent below, the ++/meg-3/4 parents of the animals described above were maintained as heterozygotes for >20 generations prior to isolation of meg-3/4 homozygous progeny in these experiments.] PGL-1 is a commonly used marker of P granules, and PGL-1 fails to concentrate into the germline blastomeres of MEG-3/4(−) embryos (Wang et al. 2014; Smith et al. 2016). To ensure that MEG-3/4 function was lost in the meg-3/4 animals described above, we monitored the subcellular localization of a PGL-1::RFP fluorescent marker protein in the various offspring of ++/meg-3/4 parents. We found that PGL-1::RFP concentrated properly in wild-type and ++/meg-3/4 embryos, but not in meg-3/4 embryos (Fig. 1C). Importantly, the meg-3/4 progeny in this experiment responded normally to RNAi, even though these animals did not possess detectable embryonic P granules (Fig. S1D). Thus, MEG-3/4 function is transgenerationally disconnected from Rde phenotypes. The fact that meg-3/4 animals do not become Rde for many generations after the loss of MEG-3/4 indicates that MEG-3/4 do not play a direct role in dsRNA-based gene silencing. Rather, loss of MEG-3/4 triggers a process that indirectly impairs RNAi over many generations. Henceforth, we use the term “phenotypic lag” to refer to situations where a phenotype does not appear for many generations after genotype is established.

Ancestral loss of P granules is associated with phenotypic hangovers.

Outcrossing meg-3/4 animals to wild type revealed a second type of transgenerational disconnect between meg-3/4 genotype and phenotype. When meg-3/4 animals were crossed to wild-type males, both the wild-type and meg-3/4 F₂ progeny of the cross were Rde (Fig. 2A,B). Amazingly, lineages established from the wild-type progeny of meg-3/4 ancestors remained defective for pos-1 RNAi for 9–11 generations before eventually reverting to a wild-type phenotype (Fig. 2B). Control crosses between wild type and wild type did not produce progeny exhibiting RNAi defects (Fig. 2A,B). Wild-type lineages derived from the meg-3/4 outcross were also defective for RNAi targeting egg-4/5 for >10 generations (Fig. S2A). Interestingly, putting wild-type descendants under strong selection for inheriting Rde (by maintaining lines on egg-5 RNAi) did not prolong the generational perdurance of the Rde phenotype (Fig. S2A). When a different set of meg-3/4 alleles, meg-3(ax3055) and meg-4(ax3052), were outcrossed, two out of three crosses also produced wild-type descendants that were RNAi-defective for multiple generations (Fig. S2B). Offspring from the third cross did not show an Rde phenotype (Fig. S2B), which may be related to the observation that a small percentage (approximately 5–10%) of meg-3(ax3055) meg-4(ax3052) individuals do not exhibit an Rde phenotype (see Fig. 1A and Fig. S1A). The fact that wild-type descendants of meg-3/4 animals typically retain an Rde phenotype for many generations indicates that mechanisms unrelated to meg-3/4 genotype exist to propagate the Rde phenotype across generations. Henceforth, we use the term “phenotypic hangover” to refer to the transgenerational inheritance of a phenotype in the absence of the mutant genotype that originally triggered the phenotype. The remainder of this paper investigates the mechanism underlying meg-3/4-associated phenotypic hangovers.

Figure 2. Ancestral loss of P granules is associated with phenotypic hangovers.

(a) Schematic of genetic crosses used to generate animals scored for RNAi responsiveness in (b). Note: for this experiment, meg-3(tm4259) meg-4(ax2026) animals had been maintained in a homozygous state for dozens of generations prior to outcross. (b) 15 lineages were established from F₂ progeny and were maintained for 24 generations under normal growth conditions. At the indicated generations, animals from each lineage were exposed to pos-1 RNAi and % hatching embryos was scored. Black dots represent individual lineages, colored bars represent the median value of % viable progeny. (c) deps-1(bn124) animals were crossed to wild-type males and descendants of the crosses were scored for pos-1 RNAi sensitivity as described in (b). (d) rde-4(ne301) animals were crossed to wild-type males, and wild-type and rde-4 progeny were scored for pos-1 RNAi sensitivity. See also Figure S2.

Phenotypic hangovers are likely initiated by failure to assemble P granules.

MEG-3/4 nucleate P granule assembly in germline blastomeres (Wang et al. 2014; Smith et al. 2016). We wondered if the Rde hangovers associated with loss of MEG-3/4 might be related to the role of MEG-3/4 in P granule assembly. DEPS-1 is a C. elegans protein that 1) localizes to P granules, 2) is partially required for P granule assembly (primarily in adult germ cells), and 3) contributes to experimental RNAi in the germline (Spike et al. 2008). To test our model that P granule disruption causes Rde hangovers, we asked whether mutations in deps-1 could, like meg-3/4, trigger Rde hangovers. We crossed deps-1(bn124) animals to wild-type males, isolated homozygous wild-type or homozygous deps-1 progeny, and measured RNAi responsiveness in lineages established from these animals. Both wild-type and mutant progeny of deps-1(bn124) ancestors were Rde, and lineages established from the wild-type progeny of deps-1(bn124) ancestors remained Rde for 2–5 generations after they had become genetically wild-type for deps-1 (Fig. 2C). RDE-4 is a dsRNA-binding protein that is thought to play a direct role in dsRNA-mediated gene silencing by working with Dicer to process dsRNAs into siRNAs (Tabara et al. 2002). RDE-4 has no known role in P granule assembly. Unlike meg-3/4 or deps-1 outcrosses, rde-4 outcrosses showed Mendelian inheritance of the Rde phenotype, indicating that (as expected) not all RNAi-related factors are associated with Rde hangovers (Fig. 2D). Altogether, our data suggest that Rde hangovers are likely triggered by an ancestral loss of P granules.

The inheritance phase of Rde hangovers is not associated with obvious defects in germ granule morphology or localization.

We wondered if Rde hangovers might be caused by inherited defects in P granule assembly. To test this idea, we first generated animals that were homozygous mutant for meg-3/4 and that expressed PGL-1::RFP, a fluorescent marker of P granules (Wan et al. 2018). We next outcrossed pgl-1::rfp; meg-3/4 animals (after animals had been meg-3/4 mutants for ≅10 generations) to pgl-1::rfp; meg-3/4(+) males and monitored PGL-1::RFP subcellular localization in embryos in the uteri of F₂ progeny from this cross. Whereas embryonic P granules failed to concentrate properly in meg-3/4 homozygous progeny, as expected, embryonic P granules formed properly in animals that had just become wild-type for meg-3/4 (Fig. 3A). P granules also appeared normal in the adult germline in wild-type F₂ progeny (Fig. S3A). We conclude that, although Rde hangovers are likely initiated by the loss of P granules, the maintenance of Rde hangovers is not associated with the inheritance of obvious defects in P granule morphology or localization.

Figure 3. MEG-3/4 help organize endo-siRNA pathways.

(a) gfp::h2b; pgl-1::rfp; meg-3/4 dpy-3 animals that had been maintained in the homozygous state for dozens of generations were crossed to gfp::h2b; pgl-1::rfp males (WT). In the F₂ generation, meg-3/4(+) and ++/meg-3/4 adults (indicated by non-Dpy phenotype) or meg-3/4 homozygous adults (indicated by Dpy phenotype) were imaged. Fluorescent micrographs of three embryos in the uterus of one adult are shown. Arrows indicate P granules. The percentage of F₂ adults containing embryos with normal PGL-1::RFP expression and the number of adults scored are indicated. Scale bar, 10 microns. (b) Loss of HRDE-1 or NRDE-2 suppresses the RNAi defect associated with meg-3/4. Individual animals of the indicated genotypes were scored for pos-1 RNAi sensitivity as described in Figure 1A (n = 18). Error bars represent +/− standard deviations of the mean (gray bars). WT, wild type. (c) Volcano plot showing log2 fold change in the # of endo-siRNAs targeting each C. elegans gene in meg-3(tm4259) meg-4(ax2026) animals relative to wild type on the x-axis and the −log10 adjusted p-value on the y-axis. Dots shown in red indicate endo-siRNA pools that were significantly different between genotypes. rde-11 and sid-1 endo-siRNA pools are labeled for reasons that will become clear in Figure 5. (d) The number of endo-siRNAs targeting all MEG-3/4-regulated genes normalized to the total number of small RNAs sequenced from each sample are shown for each replicate of wild type and meg-3/4. k=1000. (e) Overlap between the list of MEG-3/4-regulated genes and published lists of genes targeted by CSR-1-bound endo-siRNAs (Claycomb et al. 2009), genes in the WAGO class (Gu et al. 2009), and genes targeted by HRDE-1-bound endo-siRNAs (Buckley et al. 2012). Numbers of genes overlapping and non-overlapping between lists are indicated. See also Figure S3 and Table S1.

hrde-1 suppresses the Rde phenotype associated with meg-3/4 animals.

Small RNAs are major vectors for transgenerational epigenetic inheritance (TEI) in plants and animals. Current models posit that, during endo-siRNA-directed TEI in C. elegans, a bolus of endo-siRNAs is deposited into the embryo via the egg or sperm. Parentally deposited endo-siRNAs then act as guide molecules to identify cognate mRNAs and recruit RdRP enzymes, which then amplify endo-siRNA populations. Finally, repetition of this process each generation allows gene regulatory information to pass across multiple generations. The germline-expressed Argonaute HRDE-1 is, for unknown reasons, required for endo-siRNA-directed TEI in C. elegans (Buckley et al. 2012; Ashe et al. 2012; Shirayama et al. 2012; Luteijn et al. 2012). To ask if endo-siRNA-based TEI might somehow underlie Rde hangovers, we tested whether the Rde phenotype associated with meg-3/4 animals depended on HRDE-1. Indeed, tm1200, a deletion allele of hrde-1, suppressed the RNAi defect, but not the P granule defect, associated with meg-3/4 animals (Fig. 3B and Fig. S3B). HRDE-1 acts with Nuclear RNAi Defective-2 (NRDE-2) to drive endo-siRNA-based TEI (Guang et al. 2010; Buckley et al. 2012). NRDE-2 was also required for meg-3/4 animals to exhibit an Rde phenotype (Fig. 3B). These data support the idea that endo-siRNA-directed TEI contributes in some way to the RNAi defect associated with meg-3/4 animals.

MEG-3/4 help organize endo-siRNA pathways.

How might the endo-siRNA system contribute to RNAi defects and Rde hangovers? A number of endo-siRNA pathway factors localize to P granules (see Introduction). Thus, P granules may bring specific Argonautes, RdRPs, and mRNAs together in space and time in ways that help produce the correct numbers and types of endo-siRNAs each generation. This idea led us to the following model that might explain the mechanistic underpinnings of Rde hangovers. First, the disruption of P granules, via mutations like meg-3/4 or deps-1, disorganizes endo-siRNA pathways, leading to aberrant production of endo-siRNAs targeting one or more genes required for RNAi (termed RNAi gene-x genes). Second, disorganized endo-siRNAs propagate across generations to transgenerationally silence RNAi gene-x (via HRDE-1), resulting in an Rde hangover. Note: for this model to work, RNAi gene-x would need to contribute to experimental RNAi, but not endo-siRNA-based TEI. To test our model, we sequenced small RNAs (≅15–30 nucleotides) using a 5’-phosphate-independent cloning method capable of sequencing C. elegans endo-siRNAs (see Method Details). We sequenced small RNAs from replicates of wild-type and meg-3/4 animals, as well as genetically wild-type animals whose ancestors had been meg-3/4 3–25 generations prior to sequencing. Reads were mapped to the C. elegans genome, and the number of small RNAs mapping antisense to each C. elegans gene was quantified. We searched for genes that were differentially targeted by small RNAs in wild-type and meg-3/4 animals (adjusted p-value < 0.05 and log2 fold change > 1 or < −1). The analysis identified 94 and 396 genes that were targeted by more or fewer small RNAs, respectively, in meg-3/4 animals than in wild-type animals (termed MEG-3/4-regulated genes) (Fig. 3C and Table S1). Small RNAs targeting MEG-3/4-regulated genes were mostly 22 nucleotides in length, and the majority of 22-nucleotide RNAs initiated with guanosine (Fig. S3C). Given that such features are hallmarks of endo-siRNAs, we henceforth refer to these small RNAs as MEG-3/4-regulated endo-siRNAs (Billi et al. 2014). Although the number of MEG-3/4-regulated endo-siRNAs went up or down in meg-3/4 animals on a gene-by-gene basis, the total number of endo-siRNAs targeting all MEG-3/4-regulated genes remained similar in wild-type and meg-3/4 animals (Fig. 3D). Thus, loss of MEG-3/4 alters the degree to which some genes are targeted by endo-siRNAs, but does not have a major effect on the overall number of endo-siRNAs produced. Previous studies have sub-categorized C. elegans endo-siRNAs into two major groups: endo-siRNAs associated with the Worm-specific Argonautes (WAGOs), and endo-siRNAs associated with the Argonaute CSR-1 (Claycomb et al. 2009; Gu et al. 2009). MEG-3/4-regulated endo-siRNAs were largely WAGO-class endo-siRNAs (p < 0.0001) (Fig. 3E). WAGO-class endo-siRNAs engage a number of WAGO-class Argonautes, including HRDE-1, to regulate gene expression (Billi et al. 2014). MEG-3/4-regulated endo-siRNAs were enriched for HRDE-1-interacting endo-siRNAs (p < 0.0001) (Fig. 3E) (Buckley et al. 2012). Therefore, loss of MEG-3/4 and embryonic P granules is associated with aberrant levels of WAGO-class, HRDE-1-associated endo-siRNAs.

Wild-type descendants of meg-3/4 animals inherit aberrant endo-siRNA populations.

Our model predicts that disruptions to endo-siRNA populations should persist in genetically wild-type descendants of meg-3/4 animals for the duration of Rde hangovers (approximately ten generations). We therefore analyzed levels of MEG-3/4-regulated endo-siRNAs across generations in wild-type descendants of meg-3/4 animals. Hierarchical clustering using the complete linkage method revealed three discernable patterns of inheritance (Fig. 4A). Most endo-siRNAs that were abundant in meg-3/4 animals relative to the wild-type control remained high in the initial generations of the Rde hangover and then diminished to wild-type control levels over the course of 25 generations (Class 1 in Fig. 4A). Endo-siRNAs that were reduced in meg-3/4 animals relative to the wild-type control fell into two major categories: a) endo-siRNAs that remained low in the initial generations of the hangover and increased slowly, many of which never fully recovered to wild-type control levels by the F₂₅ generation (Class 2 in Fig. 4A); and b) endo-siRNAs that remained low in the initial generations of the hangover but recovered relatively quickly (Class 3 in Fig. 4A). Hierarchical clustering using divisive analysis also identified discrete patterns of endo-siRNA inheritance that resemble the categories described above (Fig. S4A). Both clustering analyses showed that the majority of MEG-3/4-regulated endo-siRNA pools progressed from mutant-like levels to wild-type levels over the course of 25 generations. A small proportion of MEG-3/4-regulated endo-siRNAs deviated from this general pattern of inheritance. For example, 6 pools of heritable endo-siRNAs remained at mutant-like levels for 25 generations and showed little evidence of reversion to wild-type levels (Fig. S4B). hrde-1 affected the levels of 4 of these 6 endo-siRNA pools, suggesting that this remarkably stable inheritance was epigenetic (and not genetic) in nature (see Discussion). Interestingly, many aberrant endo-siRNA pools were unaffected by hrde-1 in a meg-3/4 mutant background, yet were still inherited in wild-type descendants of meg-3/4 animals (Fig. S4C,D). Thus, HRDE-1-independent mechanisms of inheritance may exist in C. elegans.

Figure 4. Wild-type descendants of meg-3/4 inherit aberrant endo-siRNA populations.

(a) Z scores of the levels of endo-siRNAs targeting each MEG-3/4-regulated gene. Biological replicates were plotted side-by-side for wild type (WT), meg-3(tm4259) meg-4(ax2026) (meg-3/4), and wild-type animals descending from meg-3/4 mutant animals (P₀) for the indicated number of generations (F₃-F₂₅). Genes were sorted into groups via hierarchical clustering (complete linkage method). The three major clusters of endo-siRNAs are shown and are indicated as Class 1–3. (b) Biplot of the top two principal components (PC1 and PC2) determined by principal component analysis. Points represent individual small RNA sequencing libraries, which are color-coded by genotype and by generation. Two biological replicates were analyzed for each genotype and each generation. Each PC represents a source of variation, and the distribution of points along a given PC axis indicates the degree of correlation between libraries for that PC. Note: PC1 explains most of the variation in the sequencing data. Prior to PCA, read counts were subjected to a regularized log transformation using the rlog function in DESeq2 (Love et al. 2014). See also Figure S4.

The analyses described above indicate that aberrant endo-siRNA populations are inherited over many generations in animals whose ancestors lacked P granules. To further understand this remarkable pattern of inheritance, we subjected our sequencing data to principal component analysis (PCA). Briefly, PCA identifies and ranks sources of variation (termed principal components, or PCs) between datasets and then shows how the datasets relate to one another with regard to each PC. In the case of our small RNA sequencing data, for example, samples with similar small RNA profiles should cluster together in a plot of the highest-ranking PCs. PCA of our small RNA sequencing data revealed that the largest source of variation, PC1, corresponds to genotype, with meg-3/4 animals and wild type exhibiting the least similarity (Fig. 4B). Interestingly, wild-type descendants of meg-3/4 animals dispersed along the PC1 axis by generation: earlier generations more closely resembled meg-3/4 animals, whereas later generations more closely resembled wild type, providing further evidence that wild-type descendants of meg-3/4 animals evolve over the course of many generations from a mutant-like state to a wild-type state (Fig. 4B). PCA also revealed that, in each generation, biological replicates cluster together (Fig. 4B). Therefore, even though the recovery of endo-siRNA pools may take many generations, this process is reproducible and, therefore, largely deterministic.

Ancestral loss of P granules is associated with heritable silencing of RNAi genes.

Our model also predicts that aberrant endo-siRNA populations should inappropriately silence genes needed for experimental RNAi. Indeed, two of the top 12 most significantly upregulated endo-siRNA pools mapped to genes with known roles in experimental RNAi in C. elegans (sid-1 and rde-11) (Fig. 3C and Table S1). SID-1 is a putative transmembrane protein that is expressed in the soma and germline, where it is thought to act as a channel for transporting dsRNA into cells (Feinberg & Hunter 2003; Winston et al. 2002; Shih & Hunter 2011). RDE-11 is a zinc finger protein that is thought to function in the amplification step of RNAi (Yang et al. 2012; Zhang et al. 2012). sid-1 and rde-11 endo-siRNAs were elevated 9-fold and 13-fold in meg-3/4 mutant animals, respectively, and remained elevated in wild-type descendants of meg-3/4 animals for approximately 10 generations (Fig. 5A,B and Fig. S5). Thus, the generational kinetics of Rde hangovers and aberrant sid-1/rde-11 endo-siRNA inheritance are quite similar. To determine the effects of these endo-siRNAs on gene expression, we used quantitative RT-PCR to measure the levels of sid-1 and rde-11 mRNA. sid-1 and rde-11 mRNA levels were down 5-fold and 6-fold in meg-3/4 animals, respectively, and remained low in wild-type descendants of meg-3/4 animals for approximately 10 generations (Fig. 5B and Fig. S5B). Heritable downregulation of sid-1/rde-11 expression therefore also correlates with Rde hangovers. The data hint that runaway silencing of sid-1 or rde-11 (or both) may be responsible for meg-3/4-associated Rde hangovers. Since the RNAi defect associated with meg-3/4 depends on HRDE-1, we asked whether the misregulation of sid-1/rde-11 in meg-3/4 animals also depends on HRDE-1. Indeed, hrde-1 suppressed both the increased levels of sid-1/rde-11 endo-siRNAs and the decreased levels of sid-1/rde-11 mRNAs observed in meg-3/4 animals (Fig. 5A,B and Fig. S5). Previous studies have shown that HRDE-1 interacts with endo-siRNAs targeting sid-1 and rde-11, suggesting that the role of HRDE-1 in sid-1/rde-11 silencing is likely to be direct (Buckley et al. 2012). Thus, HRDE-1 promotes both sid-1/rde-11 silencing and germline Rde in the absence of MEG-3/4, and likely propagates these defects in wild-type animals whose ancestors lacked MEG-3/4. We conclude that ancestral loss of P granules is associated with heritable defects in gene expression, and that defects specific to sid-1 and/or rde-11 may be the cause of Rde hangovers.

Figure 5. Ancestral loss of P granules is associated with heritable changes in sid-1 expression.

(a) endo-siRNAs sequenced from animals of the indicated genotypes (and generations after outcross) that map to the sid-1 locus. A schematic of the sid-1 locus is shown below. Counts were normalized to total number of reads. (b) Top panel: endo-siRNA reads mapping to the sid-1 locus were quantified in replicates and across generations during Rde hangovers. Counts were normalized using the median of ratios method (DESeq2). Replicates are shown in red and blue. k=1000. Bottom panel: qRT-PCR was used to quantify sid-1 mRNA in the samples shown in the top panel. sid-1 mRNA values are shown relative to the mRNA values of nos-3, a germline-expressed gene. (c) Adults of the indicated genotype were injected with 200 ng/ul of pos-1 dsRNA in either one or both arms of the germline (n ≥ 9 animals per condition). After 18–22 hours of recovery, animals were allowed to lay eggs and % hatching embryos was scored. Some of the variability observed in meg-3/4 and sid-1 animals may be due to the fact that responses of sid-1 mutants to dsRNA injection change with time post-injection, and the timescale of this change varies from animal to animal(Wang & Hunter 2017). Error bars represent +/− standard deviations of the mean (gray bars). **, p-value < 0.01; ***, p-value < 0.001 (Student’s t-test). WT, wild type. (d) Model for initiation of Rde hangovers: P granules coordinate and organize endogenous small interfering RNA (endo-siRNA) biogenesis by concentrating endo-siRNA pathway factors (blue ovals) such as RNA-dependent RNA polymerases (RdRPs) and Argonautes (AGOs) together with the appropriate mRNAs. In the absence of MEG-3/4 and therefore P granules, endo-siRNA pathway factors engage the “wrong” pool of mRNAs (long red lines) and thereby initiate the production of aberrant endo-siRNAs (short red lines). (e) Model to explain transgenerational disconnects between meg-3/4 genotype and phenotype. Upon the loss of MEG-3/4, germ cells begin to produce slightly higher-than-normal levels of endo-siRNAs that target one or more genes required for experimental RNAi (RNAi gene-x). Due to the heritable nature of the endo-siRNA pathway, endo-siRNAs targeting RNAi gene-x accumulate slowly over generations in the absence of MEG-3/4 (lag). After 5–9 generations of aberrant endo-siRNA production/inheritance, these endo-siRNAs reach a level that causes silencing of RNAi gene-x; hence the defect in RNAi. After re-introduction of MEG-3/4, aberrant endo-siRNA pools continue to propagate across generations and continue to silence RNAi gene-x in genetically wild-type animals (hangover). Over the course of ten generations, genetic systems reassert their control over endo-siRNA biogenesis, returning endo-siRNA pools to normal. See also Figure S5.

rde-11 and sid-1 mutants respond idiosyncratically to dsRNAs injected directly into the germline (Zhang et al. 2012; Winston et al. 2002; Wang & Hunter 2017). For instance, rde-11 mutants show a partial RNAi defect in response to low doses of injected dsRNA (0.5 ng/ul), but respond normally to high doses of injected dsRNA (20 ng/ul or 200 ng/ul) (Zhang et al. 2012). This differential sensitivity may be due to the presumed role of RDE-11 in amplifying gene silencing signals (Zhang et al. 2012; Yang et al. 2012). sid-1 mutants, on the other hand, are sensitive to the number of gonad arms injected with dsRNA (the C. elegans germline is comprised of two distinct gonad arms) (Winston et al. 2002; Wang & Hunter 2017). Because C. elegans RNAi is systemic (capable of spreading between tissues), dsRNA injection into a single gonad arm (or the soma) can trigger gene silencing in the germ cells of both gonad arms in wild-type animals (Fig. 5C) (Fire et al. 1998). However, in sid-1 mutants, which are defective for systemic RNAi, dsRNA injection only triggers gene silencing in the injected gonad and this silencing fails to spread to the other gonad arm (Winston et al. 2002; Wang & Hunter 2017). We took advantage of these observations to ask if aberrant silencing of rde-11 or sid-1 might be the cause of Rde hangovers: We injected meg-3/4 with a high dose of pos-1 dsRNA (200 ng/ul) and asked if meg-3/4 animals either 1) behaved like rde-11 mutants (responded like wild-type to a high dose of dsRNAs), or 2) behaved like sid-1 mutants (responded more robustly to injection of both gonad arms vs. a single arm). meg-3/4 animals exhibited an Rde phenotype following injection of 200 ng/ul of pos-1 dsRNA into either one or two gonad arms, suggesting that aberrant rde-11 silencing is not solely responsible for the Rde phenotype of meg-3/4 animals (Fig. 5C). Injecting both gonad arms of meg-3/4 animals triggered more robust gene silencing than injection of a single gonad arm (Fig. 5C). sid-1 mutant animals behaved similarly to meg-3/4 animals in this assay (Fig. 5C). By contrast, rde-1 mutant animals, which are fully defective for autonomous RNAi (Tabara et al. 1999), failed to respond to dsRNA injection regardless of the number of gonad arms injected (Fig. 5C). The data suggest that the RNAi defect associated with meg-3/4 is largely systemic in nature. Thus, the data are consistent with a model in which the Rde phenotype associated with meg-3/4 is caused by aberrant silencing of the sid-1 gene. Note: for this model to be correct, aberrant silencing of the sid-1 gene would need to be limited to germ cells, as sid-1 mutant animals, but not meg-3/4 mutant animals, exhibit RNAi defects in the soma (Fig. S1C) (Winston et al. 2002).

Discussion:

Here, we show that mutations disrupting P granules trigger phenotypes that are transgenerationally disconnected from genotype. Disconnects correlate with alterations in endogenous small RNA levels and gene expression patterns, which persist across generations in the wild-type descendants of animals that lacked P granules. The data suggest that one function of germ granules is to organize and coordinate RNA-based modes of epigenetic inheritance. Upon loss of this germ granule-based organization, epigenetic defects are propagated on a generational timescale.

We find that MEG-3/4 regulate the degree to which particular genes are targeted by endo-siRNAs. How might MEG-3/4 (and therefore embryonic P granules) regulate endo-siRNA pathways? A subset of maternally deposited mRNAs and endo-siRNA pathway proteins (e.g. EGO-1, PRG-1, CSR-1, and WAGO-1) localize to P granules (Seydoux & Fire 1994; Vought et al. 2005; Claycomb et al. 2009; Batista et al. 2008; Gu et al. 2009). Thus, one function of P granules may be to spatially organize endo-siRNA biogenesis by ensuring that endo-siRNA pathway factors interact only with the proper mRNA targets (Fig. 5D). Such organization may be necessary to prevent the endo-siRNA system, which is inherently dangerous due to its feed-forward and heritable nature, from targeting functionally important germline genes for runaway heritable silencing. We propose that, in the absence of the organizing capabilities of P granules, one or more small RNA pathway factors fails to connect with its correct mRNA targets, resulting in over-targeting of some mRNAs and under-targeting of other mRNAs (Fig. 5D). In this issue of Developmental Cell, Ouyang et al., 2019 report that P granules limit piRNA-based gene silencing. The piRNA-binding Argonaute PRG-1, which directs the production of endo-siRNAs bound by HRDE-1 (Ashe et al. 2012; de Albuquerque et al. 2015; Phillips et al. 2015), is required for the RNAi defect and aberrant gene silencing that occurs in meg-3/4 animals (Ouyang et al., 2019, this issue of Developmental Cell). In support of an organizational role for P granules, MEG-3/4 regulate the localization of both PRG-1 and specific mRNAs that become mis-targeted for silencing in meg-3/4 animals (Ouyang et al., 2019, this issue of Developmental Cell).

Because endo-siRNAs can be inherited, aberrant endo-siRNAs produced in the absence of P granules propagate for many generations, even after P granules have been restored. We find that wild-type descendants of meg-3/4 ancestors can inherit aberrant endo-siRNA and mRNA patterns for over 25 generations. The fact that most endo-siRNA pools eventually return to wild-type levels (and do so reproducibly in biological replicates) indicates that, although recovery may take many generations, it is still largely a deterministic process. Thus, the information that ultimately dictates which mRNAs should and should not be channeled into the endo-siRNA pathway is likely hardwired in the genome. This hardwiring likely involves piRNAs, which are genomically encoded small RNAs with the ability to initiate endo-siRNA biogenesis in C. elegans (Ruby et al. 2006; Batista et al. 2008; Das et al. 2008; Gu et al. 2009). Interestingly, we also find that some changes in endo-siRNA pools triggered by the loss of embryonic P granules were stably inherited for at least 25 generations. Epigenetic inheritance at these loci may represent situations in which endo-siRNA-based inheritance is so efficient that it has become largely divorced from genetic/piRNA control. Finally, we also identified a number of MEG-3/4-regulated endo-siRNA pools that were unaffected by hrde-1, hinting that HRDE-1-independent mechanisms for small RNA-based TEI may exist in C. elegans.

Our data argue against a direct role for embryonic P granules in experimental RNAi, as we have documented multiple examples of a) animals that lack embryonic P granules but respond normally to germline RNAi, and b) animals that possess P granules but fail to respond to germline RNAi. How, then, might meg-3/4 mutations impair RNAi? A likely explanation is that the loss of P granules in meg-3/4 triggers the production of endo-siRNAs that inappropriately and heritably silence one or more genes required for experimental RNAi in the germline (Fig. 5E). Multiple lines of evidence support this idea. First, meg-3/4 animals produce abnormally high levels of endo-siRNAs that inappropriately silence sid-1 and rde-11, two genes required for experimental RNAi. Second, aberrant levels of sid-1/rde-11 endo-siRNAs and aberrant silencing of sid-1/rde-11 mRNA are inherited for approximately 10 generations and, therefore, occur concomitantly with Rde hangovers. Third, aberrant sid-1/rde-11 silencing and the meg-3/4 RNAi defect both depend on the same factor: HRDE-1. Lastly, our dsRNA injection data point specifically to sid-1 as the gene whose aberrant silencing triggers Rde hangovers, as meg-3/4 mutant animals show a systemic defect in germline RNAi that mimics that of sid-1 mutant animals. For the above reasons, we speculate that inappropriate and heritable silencing of the sid-1 locus in germ cells is the cause of Rde hangovers. Proving this model will require additional work, which could include experimentally restoring sid-1 expression to wild-type levels in animals undergoing Rde hangovers. A related model might explain the phenotypic lags we observe in meg-3/4 animals. Specifically, loss of P granules (via introduction of meg-3/4 mutations) might cause a small increase in sid-1 endo-siRNAs that, initially, is not sufficient to effectively silence sid-1. In subsequent generations, however, sid-1 endo-siRNAs would originate from two sources: 1) continued mild overproduction of endo-siRNAs (due to the absence of P granules), and 2) parental deposition of endo-siRNAs that were produced in previous generations. According to this model, over four to five generations, sid-1 endo-siRNAs would accumulate to levels sufficient to silence the sid-1 locus (Fig. 5E). Sequencing endo-siRNAs during a meg-3/4 phenotypic lag would be a strong test of this idea.

In addition to sid-1 and rde-11, hundreds of other genes are also mistargeted by aberrant endo-siRNAs in the wild-type descendants of meg-3/4 animals (Fig. 4). It is possible that inappropriate silencing of these genes could trigger additional types of phenotypic lags or hangovers that have not yet been documented. Furthermore, mutations disrupting germ granule assembly or organization at other developmental time points might trigger a distinct set of genotype/phenotype disconnects, as the types of mRNAs that would be available to inappropriately enter the endo-siRNA pathway would likely change throughout development. Supporting this idea, some of the genes misregulated in deps-1 animals (which exhibit defects in P granule assembly in both embryos and adult germ cells) were also identified by us in meg-3/4 animals (which lack embryonic P granules), whereas other regulated genes were unique to each genotype (Table S2) (Spike et al. 2008). Notably, deps-1 mutations decrease the expression of rde-4, a gene required for RNAi; therefore, reduced levels of RDE-4 (and not SID-1) may underlie the germline RNAi defect of deps-1 mutants (Spike et al. 2008).

C. elegans possess a robust mode of TEI thought to transmit an RNA-based memory of germline gene expression programs from parent to progeny. We find that perturbing this mode of TEI (by disrupting P granule formation) leads to heritable defects in germline gene expression programs. Given that systems exist to transmit epigenetic information across generations, it stands to reason that genetic or environmental perturbations that alter the quality or quantity of this information would have heritable effects, even if the initiating perturbation were short-lived. Although heritable alterations to the epigenome could conceivably be adaptive, it is more likely that such changes would have negative impacts on organismal fitness, which could persist for generations. Such considerations may also apply to other modes of TEI and other animals, as mutations affecting chromatin or gene regulatory factors in both C. elegans and mammals have also been linked to phenotypic hangovers (Greer et al. 2011; Nelson et al. 2012; Siklenka et al. 2015). Given that many genes, pathways, and environmental signals are likely to impinge upon the germline epigenome, phenotypic lags and phenotypic hangovers, such as those documented here, may turn out to be a fairly common phenomenon. Exploring how much phenotypic variation is contingent upon ancestral genotype, assessing the generational perdurance of this type of variation in different animals, and asking if phenotypic hangovers ever contribute to disease inheritance in humans will be important questions for future studies to address.

STAR Methods:

Lead Contact and Materials Availability

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Scott Kennedy (kennedy@genetics.med.harvard.edu). Strains are available upon request from the authors.

Experimental Model and Subject Details

C. elegans were grown at 20°C for all experiments. Unless otherwise indicated, animals were maintained on Nematode Growth Medium (NGM) plates seeded with E. coli OP50. Genetic crosses described in this study were between hermaphrodites and males. Phenotypic analyses were performed on hermaphrodites. A list of strains used in this study can be found in the Key Resources Table.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Bacterial and Virus Strains
Escherichia coli OP50	Caenorhabditis Genetics Center (CGC)	RRID:WB-STRAIN:OP50
Escherichia coli HT115(DE3)	CGC	RRID:WB-STRAIN:HT115(DE3)
Ahringer RNAi Libraries in E. coli HT115(DE3)	Kamath et al., 2003	RRID:SCR_017064
gfp RNAi clone	Author’s laboratory	N/A
Chemicals, Peptides, and Recombinant Proteins
TRIzol Reagent	Invitrogen	Cat#15596018
T4 RNA Ligase 2, truncated	New England BioLabs	Cat#M0242L
Antarctic Phosphatase	New England BioLabs	Cat#M0289L
T4 Polynucleotide Kinase	New England BioLabs	Cat#M0201L
T4 RNA Ligase 1 (ssRNA Ligase)	New England BioLabs	Cat#M0204L
Critical Commercial Assays
15% Criterion TBE-Urea Polyacrylamide Gel	Bio-Rad	Cat#3450091
Superscript III First-Strand Synthesis System	Invitrogen	Cat#18080051
iTaq Universal SYBR Green Supermix	Bio-Rad	Cat#1725120
MEGAscript T7 Transcription Kit	Invitrogen	Cat#AM1333
Deposited Data
Small RNA sequencing	This study	GEO: GSE134683
Experimental Models: Organisms/Strains
C. elegans: N2 wild type	CGC	N2 Bristol
C. elegans: meg-3(tm4259) meg-4(ax2026) X	(Wang et al., 2014)	JH3225
C. elegans: meg-3(ax3055) meg-4(ax3052) X	(Smith et al., 2016)	JH3475
C. elegans: rde-4(ne301) III	(Tabara et al., 2002)	WM49
C. elegans: mjIs31[pie-1p::gfp::h2b] II	(Ashe et al., 2012)	SX461
C. elegans: mjIs31[pie-1p::gfp::h2b] II; meg-3(tm4259) meg-4(ax2026) X	This study	YY1501
C. elegans: mjIs31[pie-1p::gfp::h2b] II; rde-1(ne219) V	Author’s laboratory	YY1568
C. elegans: deps-1(bn124) I	(Spike et al., 2008)	DG3226
C. elegans: hrde-1(tm1200) III	National BioResource Project	YY538
C. elegans: nrde-2(gg091) II	(Guang et al., 2010)	YY502
C. elegans: hrde-1(tm1200) III; meg-3(tm4259) meg-4(ax2026) X	This study	YY1512
C. elegans: nrde-2(gg091) II; meg-3(tm4259) meg-4(ax2026) X	This study	YY1513
C. elegans: dpy-3(e27) X	CGC	CB27
C. elegans: meg-3(tm4259) meg-4(ax2026) dpy-3(e27) X	This study	YY1317
C. elegans: mjIs31[pie-1p::gfp::h2b] II; pgl-1(gg547[pgl-1::3xflag::tagrfp]) IV	This study	YY1514
C. elegans: mjIs31[pie-1p::gfp::h2b] II; pgl-1(gg547[pgl-1::3xflag::tagrfp]) IV; meg-3(tm4259) meg-4(ax2026) X	This study	YY1523
C. elegans: mjIs31[pie-1p::gfp::h2b] II; hrde-1(tm1200) III; pgl-1(gg547[pgl-1::3xflag::tagrfp]) IV; meg-3(tm4259) meg-4(ax2026) X	This study	YY1548
C. elegans: mjIs31[pie-1p::gfp::h2b] II; pgl-1(gg547[pgl-1::3xflag::tagrfp]) IV; dpy-3(e27) X	This study	YY1549
C. elegans: mjIs31[pie-1p::gfp::h2b] II; hrde-1(tm1200) III; pgl-1(gg547[pgl-1::3xflag::tagrfp]) IV	This study	YY1552
C. elegans: mjIs31[pie-1p::gfp::h2b] II; pgl-1(gg547[pgl-1::3xflag::tagrfp]) IV; meg-3(tm4259) meg-4(ax2026) dpy-3(e27) X	This study	YY1553
C. elegans: sid-1(qt9) V	(Winston et al., 2002)	HC196
C. elegans: rde-1(ne219) V	(Tabara et al., 1999)	WM27
Oligonucleotides
nos-3 mRNA Forward: 5’ GGAGGCTATCGGCAGTATCA 3’	This study	N/A
nos-3 mRNA Reverse: 5’ GTGGCCCTGCTTGAGGATTA 3’	This study	N/A
rde-11 mRNA Forward: 5’ GATTTCGGACTCCCTATGTGGAC 3’	This study	N/A
rde-11 mRNA Reverse: 5’ GTAGAGATACAGTCCGTCCAGC 3’	This study	N/A
sid-1 mRNA Forward: 5’ CGGCGAATGAATCCATCTAT 3’	This study	N/A
sid-1 mRNA Reverse: 5’ CGGGAGCTATGAAGACGAAG 3’	This study	N/A
Software and Algorithms
cutadapt 1.14	(Martin, 2011)	RRID:SCR_011841
Bowtie 1.2.2	(Langmead et al., 2009)	RRID:SCR_005476
featureCounts	(Liao et al., 2014)	RRID:SCR_012919
RStudio	(RStudio Team, 2016)	RRID:SCR_000432
DESeq2 1.22.2	(Love et al., 2014)	RRID:SCR_015687
pheatmap 1.0.12	(Kolde, 2015)	RRID:SCR_016418
DEGreport 1.18.1	(Pantano, 2019)	N/A
deepTools 3.0.2	(Ramírez et al., 2016)	RRID:SCR_016366
Sushi 1.20.0	(Phanstiel et al., 2014)	N/A
ZEN Digital Imaging for Light Microscopy	Zeiss	RRID:SCR_013672
Fiji	(Schindelin et al., 2012)	RRID:SCR_002285
Other
Axio Observer.Z1 microscope	Zeiss	N/A
ORCA-Flash 4.0 CMOS camera	Hamamatsu	N/A
Plan-Apochromat 63×/1.4 Oil DIC M27 objective	Zeiss	N/A
Plan-Apochromat 20×/0.8 M27 objective	Zeiss	N/A
EC Plan-Neofluar 10×/0.3 Ph1 M27 objective	Zeiss	N/A

Method Details

Bacteria-mediated RNA interference:

Animals were fed E. coli HT115 expressing dsRNAs targeting the indicated genes. E. coli HT115 containing the L4440 vector was used as a no-RNAi control in Figure 1A, Figure 3B, and Figure S1. RNAi clones, with the exception of gfp RNAi, were obtained from the Ahringer C. elegans RNAi library (Kamath et al. 2003). For tests using pos-1 RNAi or egg-5 RNAi, RNAi feeding began at the L2/L3 larval stages. gfp RNAi feeding began at the L3 larval stage, and gfp silencing was scored in the same generation at the adult stage. Somatic RNAi assays were performed by plating embryos onto bacteria expressing the indicated dsRNAs, and silencing was scored in the same generation at the adult stage. Details on scoring and sample size are described in figure legends.

Phenotypic lag experiments:

The dpy-3 locus is approximately 0.1 cM from meg-3 and 0.8 cM from meg-4 (meg-3 and meg-4 are approximately 0.7 cM apart). We first marked a meg-3(tm4259) meg-4(ax2026) chromosome with dpy-3(e27). Then, we crossed meg-3(tm4259) meg-4(ax2026) dpy-3(e27) animals to wild-type males and maintained a meg-3(tm4259) meg-4(ax2026) dpy-3(e27) chromosome in a heterozygous state for at least 22 generations. Multiple independent lines were established in this manner. Every generation, non-Dpy progeny were singled from non-Dpy parents that gave rise to both Dpy and non-Dpy progeny. Prior to performing lag experiments, lines were genotyped for meg-3 and meg-4 by PCR to check whether meg-3(tm4259) and meg-4(ax2026) were still linked to dpy-3(e27). The progeny of animals that had been heterozygous for 22 generations were singled, tested for sensitivity to pos-1 RNAi, and then genotyped for meg-3 and meg-4 by PCR (n = 72). In parallel, siblings of the progeny tested for RNAi were singled under normal growth conditions, genotyped for meg-3 and meg-4 by PCR, and wild-type and meg-3/4 animals were used to establish lines (6 wild-type lines and 10 meg-3/4 lines). Lines were maintained under normal growth conditions. Every subsequent generation, a small pool of animals (3–5) from each line was tested for sensitivity to pos-1 RNAi. Lines were re-genotyped for meg-3 and meg-4 approximately every 5 generations. As a control, meg-3(tm4259) meg-4(ax2026) dpy-3(e27) animals were crossed to meg-3(tm4259) meg-4(ax2026) males, and the dpy-3(e27) allele was maintained in a heterozygous state for at least 22 generations, whereas meg-3(tm4259) and meg-4(ax2026) were maintained in a homozygous state. Fifteen meg-3(tm4259) meg-4(ax2026) dpy-3(e27) lines were then established and tested in parallel with the lines described in the first cross. To examine the P granule phenotype of newly generated meg-3/4 mutants (Fig. 1C), similar crosses were performed as described above, except animals also contained pgl-1(gg547[pgl-1::3xflag::tagrfp]) to mark P granules and mjIs31[pie-1p::gfp::h2b] to mark chromatin. For this experiment, meg-3/4 genotype was inferred from the Dpy phenotype.

Phenotypic hangover experiments:

In general, hangover experiments began by crossing animals homozygous for the mutant allele(s) of interest [e.g. meg-3(tm4259) meg-4(ax2026)] to wild-type males (P₀ generation). Multiple independent crosses were performed for each experiment. In the F₂ generation, L2/L3 larvae were singled, tested for sensitivity to pos-1 (or egg-5) RNAi, and then genotyped by PCR. Siblings of those animals were singled under normal growth conditions, genotyped by PCR, and were then used to establish either homozygous wild-type or homozygous mutant lines. Unless otherwise indicated, lines were maintained under normal growth conditions for the duration of the experiment. Each subsequent generation, a small pool of animals (3–5) from each line was tested for sensitivity to pos-1 (or egg-5) RNAi. Lines were re-genotyped by PCR every 5–10 generations. As a control, wild-type animals descending from a cross between wild-type hermaphrodites and wild-type males were tested for sensitivity to pos-1 (or egg-5) RNAi in parallel with the wild-type descendants of mutant animals. To examine the P granule phenotype in wild-type descendants of meg-3(tm4259) meg-4(ax2026) animals (Fig. 3A and Fig. S3A), similar crosses were performed as described above, except animals also contained pgl-1(gg547[pgl-1::3xflag::tagrfp]) to mark P granules and mjIs31[pie-1p::gfp::h2b] to mark chromatin. In addition, the meg-3(tm4259) meg-4(ax2026) chromosome was marked with dpy-3(e27), and meg-3/4 genotype was inferred from the Dpy phenotype.

Sample collection for small RNA-seq and qRT-PCR:

The following crosses were performed in parallel in preparation for RNA isolation: 1) meg-3(tm4259) meg-4(ax2026) animals were crossed to wild-type males (P₀ generation), and wild-type descendants of this cross were collected in generations F₃ through F₇, F₁₀, F₁₅, and F₂₅; and 2) as a control, wild-type animals were crossed to wild-type males (P₀ generation), and the descendants of this cross, which were all wild type, were collected in generation F₃ (this sample is referred to as the wild-type control). For each type of cross, biological replicates were derived from different parents (P₀s). To amass enough animals for RNA isolation, 30–50 wild-type lines were established in the F₂ generation (for each biological replicate), and lines were pooled starting in the F₃ generation. meg-3(tm4259) meg-4(ax2026) animals that had been homozygous mutant for dozens of generations were collected as a mutant control. Approximately 10% of meg-3(tm4259) meg-4(ax2026) animals do not develop a full germline, and 27% of meg-3(tm4259) meg-4(ax2026) animals are sterile (have empty uteri) (Wang et al. 2014). To help control for the proportion of germ cells in each sample, adults with empty uteri were removed with a standard worm pick prior to sample collection. Adult worms were washed two times with M9 Buffer, resuspended and vortexed for 30 seconds in TRIzol, then flash-frozen in liquid nitrogen and stored at −80°C. Total RNA was isolated by TRIzol extraction.

Small RNA library preparation and sequencing:

RNAs ranging from approximately 15 to 30 nucleotides were gel-purified from total RNA (20 ug) on a 15% polyacrylamide/urea gel and then ligated to a 3’ adapter using T4 RNA ligase 2, truncated (New England BioLabs). To enable the cloning of 5’-triphosphorylated RNAs, samples were treated with Antarctic Phosphatase (New England BioLabs) followed by treatment with T4 polynucleotide kinase (New England BioLabs) as described previously (Gent et al. 2009). Prior to ligation of the 5’ adapter, 3’-ligated small RNAs and any excess 3’ adapter were hybridized to the oligo that would eventually be used as a primer for reverse transcription. This step was taken to help minimize adapter-dimer formation (McReynolds & Munafo 2014). To help avoid cross-contamination, the 5’ adapter was modified to contain the Illumina genomic sequencing primer annealing site followed by an additional 4 nucleotides at the 3’ end. Two different 5’ adapters (ending in either AGCG or CGUC) were mixed in a 1:1 ratio, and the mix was ligated to each sample using T4 RNA ligase I (New England BioLabs). Every sample was treated with the same mix of 5’ adapters. Libraries were amplified and multiplexed with a 6-nucleotide 3’ barcode, then pooled for next-generation sequencing on a NextSeq500 (Biopolymers Facility, HMS).

Computational analysis of small RNA-seq:

First, custom scripts were used to select reads starting with the last 4 nucleotides of the 5’ adapters (either AGCG or CGTC). Cutadapt 1.14 was used to trim the 3’ adapter (cutadapt -a CTGTAGGCACCATCAATAGATCGGAAGAGCAC -m 14 -- discard-untrimmed) and the in-line portion of the 5’ adapter (cutadapt -u 4) (Martin 2011). Trimmed reads were then mapped to the C. elegans genome (WormBase release WS260) using Bowtie 1.2.2 (Langmead et al. 2009). No mismatches were allowed. The number of reads mapping antisense to each gene was determined using featureCounts (featureCounts -s 2) (Liao et al. 2014). Raw counts were then normalized by the median of ratios method using DESeq2 1.22.2 in R (Love et al. 2014; RStudio Team 2016). Differential analyses were performed using DESeq2 1.22.2 (Love et al. 2014). To identify MEG-3/4-regulated endo-siRNA pools that did not recover to wild-type levels by the F₂₅ generation (Fig. S4B), DESeq2 was used to find genes that were differentially targeted by small RNAs in wild-type F₂₅ descendants of meg-3/4 animals (the last generation of the hangover that was tested) and wild-type control animals (adjusted p-value < 0.05 and log2 fold change > 1 or < −1), and the resulting genes were tested for overlap with the list of MEG-3/4-regulated genes. To identify MEG-3/4-regulated endo-siRNA pools that were unaffected by hrde-1 (Fig. S4C,D), DESeq2 was used to find genes that were differentially targeted by small RNAs in meg-3/4 and hrde-1; meg-3/4 animals (adjusted p-value < 0.05 and log2 fold change > 1 or < −1), and the resulting genes were tested for overlap with the list of MEG-3/4-regulated genes. Heatmaps (Fig. 4A and Fig. S4C) were clustered by row (genes) and scaled by row using pheatmap 1.0.12 in R (Kolde 2015). The gene clusters shown in Figure S4A were generated using DEGreport 1.18.1 in R (Pantano 2019). Principal component analysis of rlog-transformed counts (Fig. 4B) was performed using the rlog and plotPCA functions in DESeq2 1.22.2 (Love et al. 2014). Custom scripts were used to extract the features of small RNAs that mapped antisense to MEG-3/4-regulated genes. Coverage plots (Fig. 5A and Fig. S5A) were generated as follows: first, bedGraph files normalized by counts per million were produced for the forward and reverse strands using deepTools 3.0.2 (bamCoverage -bs 5 --normalizeUsing CPM --samFlagExclude 16) and (bamCoverage -bs 5 --normalizeUsing CPM --samFlagInclude 16) (Ramírez et al. 2016); then, bedGraph files were plotted in R using Sushi 1.20.0 (Phanstiel et al. 2014).

Quantitative RT-PCR:

Using the total RNA prepared as described above, mRNA was reverse-transcribed into cDNA using the SuperScript III First-Strand Synthesis System (Invitrogen). Quantitative RT-PCR was performed with the iTaq Universal SYBR Green Supermix (Bio-Rad) and the primers listed in the Key Resources Table. Cycle threshold values were calibrated to a standard curve generated using a 4-point, 1:2 dilution series of wild-type control cDNA. PCR reactions were performed in technical triplicate for each biological replicate.

Microscopy:

Animals were immobilized in M9 Buffer containing 0.05% sodium azide and mounted on glass slides. Images were taken with a wide-field Zeiss Axio Observer.Z1 microscope equipped with an ORCA-Flash 4.0 CMOS camera (Hamamatsu) and the following Zeiss objectives: Plan-Apochromat 63×/1.4 Oil DIC M27, Plan-Apochromat 20×/0.8 M27, and EC Plan-Neofluar 10×/0.3 Ph1 M27. Embryos were imaged in utero. Images were acquired with ZEN software (Zeiss) and compiled in Fiji (Schindelin et al. 2012).

Synthesis and microinjection of pos-1 dsRNA:

pos-1 dsRNA was synthesized in vitro using a MEGAscript T7 Transcription Kit (Invitrogen). The transcription template was PCR-amplified from the Ahringer pos-1 RNAi clone and contained the pos-1 insert sequence as well as the flanking T7 promoters. pos-1 dsRNA was injected into one or both gonad arms of young adults at a concentration of 200 ng/ul. At least 9 animals were injected per condition. 18–22 hours post-injection, animals were singled and allowed to lay embryos for approximately 20 hours. Progeny and unhatched eggs were counted 24 hours later. A small fraction of meg-3/4 animals did not lay eggs and were therefore excluded from the analysis. Mutant alleles were the following: meg-3(tm4259) meg-4(ax2026), sid-1(qt9), and rde-1(ne219).

Quantification and Statistical Analysis

Differential analyses for small RNA sequencing were performed with the Wald test using DESeq2 1.22.2, and genes with both an adjusted p-value < 0.05 and a log2 fold change >1 or < −1 were deemed significant (Love et al. 2014). Significance values reported in the description of Figure 3E were calculated with the one-sided Fisher’s exact test using the fisher.test function in R (RStudio Team 2016). Significance values shown in Figure 5C were calculated with the Student’s t-test (two-tailed, unequal variances) using Excel. All error bars represent standard deviation. Sample sizes are indicated in the figure legends and Method Details.

Data and Code Availability

The small RNA sequence data generated during this study are available from Gene Expression Omnibus (GEO) under the accession number GEO: GSE134683.

Supplementary Material

3

Table S1. List of MEG-3/4-regulated endo-siRNA targets, related to Figure 3.

Highlights.

• Disrupting germ granules leads to aberrant levels of endogenous small RNAs.

• Aberrantly expressed small RNAs misregulate germline gene expression.

• Gene expression defects caused by germ granule loss are inherited across generations.