A membrane-associated condensate drives paternal epigenetic inheritance in C. elegans

Transgenerational epigenetic inheritance (TEI) describes the transmission of gene-regulatory information across generations without altering DNA sequences, and allows priming of offspring towards transposable elements (TEs) and changing environmental conditions. One important mechanism that acts in TEI is based on small non-coding RNAs. Whereas factors for maternal inheritance of small RNAs have been identified, paternal inheritance is poorly understood, as much of the cellular content is extruded during spermatogenesis. We identify a phase separation-based mechanism, driven by the protein PEI-1, which is characterized by a BTB-BACK domain and an intrinsically disordered region (IDR). PEI-1 specifically secures the Argonaute protein WAGO-3 within maturing sperm in C. elegans. Localization of PEI granules in mature sperm is coupled, via S-palmitoylation, to myosin-driven transport of membranous organelles. pei-1-like genes are also found in human and often expressed in testis, suggesting that the here identified mechanism may be broadly conserved.


INTRODUCTION
Transgenerational epigenetic inheritance (TEI) is a non-genetic way of inheriting gene regulatory information across generations, while leaving genome sequences unchanged. It has been found to 35 mediate gene regulation in relation to environmental conditions, neuronal activity and more (Bošković and Rando, 2018;Perez and Lehner, 2019). Germ cell-resident mechanisms that drive TEI have been found in both plants and animals, and include both nuclear (chromatin-based) as well as cytoplasmic mechanisms (Bošković and Rando, 2018;Castel and Martienssen, 2013;Perez and Lehner, 2019). A class of molecules that has been clearly linked to TEI is that of small RNAs, most notably short-40 interfering RNAs (siRNAs) and Piwi-interacting RNAs (piRNAs) (Bošković and Rando, 2018). These act as sequence-specific co-factors for Argonaute proteins, which in turn can impose gene-regulatory effects upon base-pairing of the small RNA to a target transcript. Such effects can range from target RNA destabilization to chromatin modification (Castel and Martienssen, 2013;Hutvagner and Simard, 2008;Peters and Meister, 2007). 45 In C. elegans, small RNAs with an established role in TEI are the 22G RNAs (de Albuquerque et al., 2015;Buckley et al., 2012;Mao et al., 2015;Phillips et al., 2015;Wan et al., 2018;Xu et al., 2018). These small RNAs are generated by RNA-dependent RNA Polymerases (RdRPs), in a process that also requires Mutator proteins, such as MUT-16, and are bound by a highly diversified, worm-specific Argonaute (WAGO) family Phillips et al., 2012;Yigit et al., 2006;Zhang et al., 2011). 22G RNAs 50 represent secondary siRNAs, as they are generated in response to target recognition by so-called primary Argonaute proteins, such as the C. elegans Piwi protein PRG-1, or the Argonaute protein that mediates exogenous RNAi: RDE-1 (Das et al., 2008;Yigit et al., 2006). However, following establishment, 22G RNA responses can be inherited across multiple generations, in some cases close to indefinitely, in absence of the original trigger (Luteijn and Ketting, 2013). Interestingly, both 55 maternal as well as paternal inheritance of such responses has been described (Alcazar et al., 2008;Grishok et al., 2000;Lev et al., 2019;Wan et al., 2018). Finally, inheritance of 22G RNAs plays an important role in maintaining an endogenous 22G RNA pool that does not target essential genes. In absence of maternal and paternal 22G RNA inheritance severe sterility can develop, caused by 22G RNAs that erroneously silence genes required for germline development (de Albuquerque et al., 2015;60 Phillips et al., 2015). This phenotype is named Mutator Induced Sterility (Mis).
Molecular mechanisms linked to small RNA pathways are often organized in non-membranous organelles (Voronina et al., 2011), that often form through phase separation, driven by proteins with intrinsically disordered regions (IDRs), and/or proteins that are multivalent with regard to interactions with other proteins and RNA (Banani et al., 2017). In fact, the IDRs and multivalent character of these 65 molecules are key characteristics for enhancing their phase separating behavior. In C. elegans, three such phase-separated condensates have been described to accommodate and affect small RNA-driven pathways: P granules, Z granules and Mutator foci (Phillips et al., 2012;Updike and Strome, 2010;Wan et al., 2018). Interestingly, none of these condensates has been found in mature sperm. In fact, it has been demonstrated that P granules dissociate during spermatogenesis (Updike and Strome, 2010). In 70 contrast, P granules and Z granules have been closely connected to maternal TEI in C. elegans (Wan et al., 2018). In mammals, a dedicated paternal condensate has been described: the chromatoid body (CB) (Kotaja and Sassone-corsi, 2007). However, its function relates to RNA decay and translational repression during spermatogenesis, not to inheritance of specific factors. Finally, biomolecular condensates are well-described throughout germ cell development, but no such structures have been 75 identified in mature sperm of any organism (Voronina et al., 2011). Possibly, this is linked to the fact that spermatogenesis is accompanied by a massive reduction of the intracellular content (Ellis and Stanfield, 2014). This also happens in C. elegans: during meiosis II a residual body is formed into which much of the spermatid cytoplasm is discarded, including, for instance, the spermatogenesis-expressed Argonaute protein ALG-3 (Conine et al., 2010). 80 Given the fact that much of the cytoplasmic content of spermatids is lost before spermatozoa are formed, one may question if paternal TEI can be mediated through the cytoplasm, or whether such inheritance via the father is restricted to chromatin-based mechanisms. Here we show that this is not the case. We identify the cytoplasmic Argonaute protein WAGO-3 as a factor required for paternal TEI and demonstrate that it is maintained within the cytoplasm of mature sperm. We identify the protein 85 PEI-1, containing a BTB-BACK domain, as well as an IDR, as a factor that drives the formation of spermspecific condensates (PEI granules), to which WAGO-3, but no other tested WAGO protein, is attracted. PEI granules are juxtaposed to membranous organelles such as mitochondria, and like mitochondria, depend on myosin-dependent transport to be maintained with the maturing sperm cell. This process critically depends on S-palmitoylation, suggesting that PEI granules are coupled to membranes by a 90 lipid-anchor. We find that pei-1-like genes are also found in human, where they are expressed in, and have functions related to spermatogenesis, suggesting that the mechanism we here identify is broadly conserved.

WAGO-3 is required for an immortal germline and associates with paternal small RNAs
Many RNAi factors of C. elegans have been shown to be needed for proper germ cell development across generations, or germline immortality (Perez and Lehner, 2019). We found that WAGO-3 is also required for germline immortality, as out-crossed wago-3 mutant animals became sterile in 100 subsequent generations ( Figure 1A), implicating WAGO-3 in a transgenerational mechanism that maintains germ cell health. We also analyzed how WAGO-3 affects the expression of a germlinespecific mCherry::H2B transgene whose silencing can be initiated by the C. elegans Piwi protein PRG-1 (Bagijn et al., 2012;Lee et al., 2012;Ozata et al., 2019). Once established, this silencing can be maintained in absence of PRG-1 for many generations and is referred to as RNAe (RNAi-induced 105 epigenetic silencing) (Ozata et al., 2019). We crossed an RNAe-silenced version of this transgene into wago-3 mutants and analyzed its activity by microscopy. We found that WAGO-3 is not acutely required for RNAe, but that it is required for the stable transgenerational maintenance of the RNAedriven silencing of this transgene ( Figure 1B), indicating that WAGO-3 has a role in TEI.
Next, we aimed to determine the endogenous targets of WAGO-3, by sequencing WAGO-3-bound 110 small RNAs. To achieve this, we inserted a gfp::3xflag encoding sequence directly downstream of the endogenous start codon of wago-3, to enable immunoprecipitation (IP) experiments. Following IP from adult animals, we sequenced small RNAs from both input and anti-FLAG IP samples. WAGO-3 is predominantly associated with small RNAs that have characteristics of 22G RNAs (Gu et al., 2009) (Figures 1C and S1A). Based on sequence complementarity we identified 2166 WAGO-3 targets. These 115 targets cover the vast majority of known Mutator targets, while WAGO-3 and CSR-1 targets were almost mutually exclusive Phillips et al., 2012) ( Figure 1D). Notably, nearly 70 % of all annotated transposable elements were represented in WAGO-3-bound 22G RNAs (Figures 1E and  1F), consistent with its role in TE silencing (Robert et al., 2005;Vastenhouw et al., 2003). The vast majority of WAGO-3 associated 22G RNAs (83%) target exonic sequences over the whole gene body of 120 both germline-expressed and soma-expressed protein-coding genes (Figures 1E and S1B-S1D). Among germline-expressed genes, WAGO-3 displayed a small, but significant preference for genes that are spermatogenic over genes that are oogenic ( Figure S1B). Finally, we found a significant overlap of protein-coding target genes between WAGO-3 and sperm-derived 22G RNAs (Stoeckius et al., 2014) ( Figure 1G and 1H). We note that these numbers are most likely underestimates since the sperm used 125 to determine small RNA target genes was contaminated with 30 % spermatocytes (Stoeckius et al., 2014), and thus also included WAGO-1, CSR-1 and possibly other Argonaute associated 22G RNAs ( Figures S2G-S2I). We conclude that WAGO-3 is involved in TEI and binds to 22G RNAs known to be paternally transmitted to the subsequent generation.

WAGO-3 is expressed throughout germline development 130
We performed confocal microscopy to determine the expression pattern of GFP::3xFLAG::WAGO-3 (from here on referred to as WAGO-3). We detected specific and global expression throughout germline development, with localization to P granules in mitotic, meiotic and primordial germ cells ( Figures S2A and S2B). Notably, we also found strong WAGO-3 signals within the spermatheca, where spermatozoa are stored, indicating that WAGO-3 is also expressed during spermatogenesis and that it 135 is maintained in mature sperm.

WAGO-3 is guided into sperm by PEI-1
Immunoprecipitation of WAGO-3 from late-L4 stage hermaphrodites, a stage during which spermatogenesis is ongoing, followed by label-free quantitative mass spectrometry (IP-MS/MS) identified a number of WAGO-3 co-enriched proteins. Among them, known P granule components like 140 DEPS-1, PRG-1 and WAGO-1 were identified ( Figure 2A). We focused, however, on the protein F27C8.5, which we named PEI-1 (Paternal Epigenetic Inheritance defective-1) (Figures 2A and 2B). We generated endogenously tagged pei-1 alleles and found that PEI-1::mTagRFP-T (from here on referred to as PEI-1) was exclusively expressed during later stages of spermatogenesis, both in males ( Figure S2C) as well as in L4 hermaphrodites ( Figure S2D). 145 We analyzed spermatogenic gonads of late-L4 stage hermaphrodites to look at PEI-1 and WAGO-3 expression more closely. In naïve germ cells, WAGO-3 localized to peri-nuclear P granules, marked by PGL-1 ( Figure S2A). However, subcellular alterations occurred from the primary spermatocyte stage onwards. First, WAGO-3 was found to accumulate in distinct, non-peri-nuclear, cytoplasmic foci . Second, as previously described (Updike and Strome, 2010), P granules began 150 to disappear (Figures 2E and S2E). Third, PEI-1 started to be expressed and always co-localized with WAGO-3 in cytoplasmic foci ( Figures 2F and S2D). Using a publicly available deletion allele, we assessed the requirement of PEI-1 for the subcellular localization of WAGO-3 during sperm cell maturation. In pei-1(ok1050) mutants, WAGO-3 was absent from spermatozoa and instead was found in the residual body (Figures 2D; 2G; S2F), which contains discarded material that is not needed in mature sperm (Ellis 155 and Stanfield, 2014). Thereby WAGO-3 followed the same fate as other Argonaute proteins like WAGO-1, ALG-3 and CSR-1 (Figures 2H and S2G-S2I). Conversely, loss of WAGO-3 did not affect PEI-1 localization ( Figure S3). We conclude that WAGO-3 is special amongst C. elegans Argonaute proteins, as it is maintained in maturing sperm by re-localizing to cytoplasmic foci before P granules disappear. Furthermore, PEI-1 is a novel protein that marks these cytoplasmic foci and is required for proper 160 WAGO-3 localization, and segregation into mature sperm.

WAGO-3 and PEI-1 are crucial for paternal epigenetic inheritance of silencing information
We implemented a genetic system to test the roles of WAGO-3 and PEI-1 in parental TEI in an endogenous setting. We and others have previously shown that C. elegans individuals need parental 22G RNAs to specify the correct targets, particularly when also piRNAs are absent: Absence of parental 165 22G RNAs in animals that lack piRNAs but can make 22G RNAs results in sterility due to inappropriate gene silencing ( Figure 3A). This phenotype was named Mutator Induced Sterility (Mis) (de Albuquerque et al., 2015;Phillips et al., 2015). We used this Mis phenotype to assess the ability of sperm or oocyte to provide silencing information that could lead to a rescue of the sterility. First, we tested whether either maternal or paternal 22G RNAs would be sufficient to prevent the Mis phenotype. As expected, 170 this was indeed what we found ( Figure 3B-C, top two bars). We next tested whether WAGO-3 or PEI-1 were required for either paternal and/or maternal rescue of the Mis phenotype. The offspring of both wago-3, as well as pei-1 mutant males showed high degrees of sterility, similar to that observed from mut-16 mutant males ( Figure 3B), indicating that WAGO-3 and PEI-1 are required for paternal rescue of the Mis phenotype. In contrast, neither PEI-1 nor WAGO-3 was required for maternal rescue of this 175 phenotype, whereas loss of mut-7 did trigger sterility ( Figure 3C). We conclude that PEI-1 and WAGO-3 are specifically required in the male, to provide embryos with paternal silencing memory that guides 22G RNA production in the offspring.

Granule formation and WAGO-3 interaction are mediated via different PEI-1 domains
PEI-1 has a bimodal composition in terms of intrinsically ordered and intrinsically disordered regions. 180 The N-terminal region is predicted to predominantly fold into alpha helical structures and likely adopts a BTB fold followed by a BACK domain ( Figure 4A). While the BTB domain is described to mediate protein-protein interactions (Collins et al., 2001), the molecular function of the BACK domain remains speculative (Stogios and Privé, 2004). The C-terminal part of PEI-1 is predicted to be intrinsically disordered ( Figure 4A). Following these predictions, we further edited the endogenous 185 pei-1::mTagRfp-t locus by introducing defined deletions to generate six different, fluorescently tagged PEI-1 variants ( Figure 4B). We then performed co-localization studies with WAGO-3 to determine effects on its subcellular distribution ( Figure 4C). Deletion of the BTB domain resulted in normally appearing foci, which were accompanied by a diffuse cytoplasmic signal of both WAGO-3 and PEI-1 in spermatozoa. Loss of the BACK domain led to a similar phenotype, but in addition affected the number 190 and homogeneity of cytoplasmic foci throughout spermatogenesis. PEI-1 lacking both the BTB and BACK domain caused a more severe phenotype. Although some foci were still present in maturing sperm, no foci, and only diffuse signals were detectable in spermatozoa. We note that none of these three deletion mutants resulted in PEI-1 or WAGO-3 accumulation in the residual body, and that both proteins always co-localized. 195 Upon deletion of the IDR of PEI-1, WAGO-3 displayed diffuse distribution in spermatocytes, localized to residual bodies and was absent from spermatozoa. We note that the enrichment in the residual body was possibly based on active sorting, as a transcriptional GFP::3xFLAG reporter from the endogenous wago-3 locus was found to be evenly distributed over the residual body and the budding spermatids, resulting in GFP positive spermatozoa ( Figures S4A-S4C). Intriguingly, the remaining, 200 structurally ordered part of PEI-1 did not accumulate in the residual body, but localized to faint foci in maturing sperm cells, indicating that the remaining PEI-1 protein can form larger assemblies with correct segregation. Additional removal of the BACK domain did not further affect the subcellular localization of WAGO-3: it remained enriched in the residual body. However, the remaining BTB::mTagRFP-T fusion-protein no longer formed foci, and was evenly distributed between residual 205 body and budding spermatids, reminiscent of the symmetric segregation of free GFP::3xFLAG. To our surprise, the largest deletion of PEI-1, which only left a few N-and C-terminal amino acids, was found to localize to unknown structures that were segregated into spermatids . Interestingly, co-staining of mitochondria revealed a mutually exclusive localization pattern ( Figures  S4F and S4G). We conclude that the IDR of PEI-1 is essential to bind WAGO-3 and contributes to foci 210 formation, while the BTB and BACK domains play an important role in forming and stabilizing PEI-1 foci during the entire process of spermatogenesis, and continue to do so in spermatozoa. Additionally, both the BACK domain and the IDR are required for the asymmetric sorting of PEI-1 into the spermatids.

PEI granules are independent condensates that retain WAGO-3 via hydrophobic interactions
Membrane-less compartments, like P granules and Mutator foci, are phase-separated condensates 215 known to play crucial roles in germ cell biology (Phillips et al., 2012;Updike and Strome, 2010). We tested whether PEI-1 foci depend on these condensates by removing MUT-16, disrupting Mutator foci, or DEPS-1, disrupting P granule assembly (Phillips et al., 2012;Spike et al., 2008). Neither MUT-16 nor DEPS-1 was required for proper localization of PEI-1 during spermatogenesis, indicating that the formation of PEI-1 foci does not depend on these known, phase-separated condensates ( Figure S3). 220 Hence, PEI-1 foci represent an independent, novel entity, which we will from here on refer to as PEI granules.
Phase separation can involve various types of interaction including pi/pi, cation/pi, electrostatic and hydrophobic interactions (Hyman et al., 2014). Aliphatic compounds, such as 1,6-hexanediol, were shown to disrupt weak hydrophobic interactions, and hence affect phase separation driven by such 225 interactions (Kroschwald et al., 2017). Thus, we monitored WAGO-3 and PEI-1 in isolated male-derived spermatocytes and budding spermatids in the presence of different concentrations of 1,6-hexanediol ( Figure 5A). WAGO-3 was found to be highly sensitive, as no PEI granule localization was detected anymore in the presence of merely 1.25 % 1,6-hexanediol. This indicates a major role for hydrophobic interactions in WAGO-3 localization to PEI granules. Of note, WAGO-3 relocated preferentially to the 230 residual body upon 1,6-hexanediol treatment, revealing an ongoing drive of these cells to relocate proteins into the residual body. In contrast, PEI-1 foci were found to be more resistant, especially in budding spermatids where even a 5 % 1,6-hexanediol treatment did not cause disassembly of PEI granules. These results suggest that the interaction between WAGO-3 and PEI granules is mostly hydrophobic, while PEI granules themselves significantly depend on other, or additional types of 235 interactions, possibly mediated by the BTB and BACK domains.
To better understand the physical properties of PEI granules, we asked to which extent they exchange material with the cytoplasm. Therefore, we measured the fluorescence recovery after photobleaching (FRAP) of WAGO-3 in both P granules and PEI granules ( Figures 5B and 5C). As previously reported, proteins localizing to the liquid-phase of P granules exhibit high recovery rates (Brangwynne et al., 240 2009;Putnam et al., 2019). Similarly, we found that WAGO-3 also showed very rapid fluorescence recovery in P granules (t1/2 = 3.2 +/-1.5 s). Strikingly, WAGO-3 exhibited much slower exchange dynamics when localizing to PEI granules in budding spermatids (t1/2 = 47.3 +/-17.6 s). Moreover, we found that the mobile fraction of WAGO-3 within the condensates is slightly reduced in PEI granules compared to P granules. This suggests that PEI granules are more gel-like than P-granules. The 245 prevalence of certain amino acids has been shown to modulate the material property of phase-separated condensates (Wang et al., 2018). In particular, glycine residues were shown to maintain liquidity, while serine and glutamine residues promote hardening of condensates. Thus, we analyzed the amino acid composition of the intrinsically disordered regions that were predicted for PEI-1, and compared that to IDR compositions of PGL-1 and PGL-3, both known to localize to the liquid 250 phase of P granules (Hanazawa et al., 2011;Kawasaki et al., 1998Kawasaki et al., , 2004, and of MEG-3 and MEG-4, both reported to form gel-like assemblies (Putnam et al., 2019) ( Figure S5). While glutamine was not found to be strongly enriched in any of the IDRs, serine was significantly enriched in all of them, and most highly in PEI-1, MEG-3 and MEG-4. In addition, PGL-1 and PGL-3 IDRs were strongly enriched for glycine, whereas such enrichment was absent from PEI-1, MEG-3 and MEG-4 IDRs. This similar amino 255 acid profile between PEI-1 and MEG-3/4 IDRs is consistent with the idea that PEI granules are more gel-like than P-granules.
Finally, we live-imaged PEI granules by monitoring WAGO-3, and found that they were rather static. We did not observe any major movements in a period of one hour ( Figure 5D and Movie S1), suggesting an attachment to bigger structures, which prevents any fusion or fission events between individual 260 foci. Everything considered, we conclude that PEI granules are stable, independent condensates that retain WAGO-3 via hydrophobic interactions.

Segregation of PEI granules depends on membranous organelle transport
We next addressed the question of how PEI granules are transported into budding spermatids. A substantial amount of cellular material, including free ribosomes, the endoplasmic reticulum and the 265 Golgi apparatus, is asymmetrically segregated into residual bodies during the second meiotic division. Only a few organelles, like the nucleus, mitochondria and the sperm-specific fibrous body-membranous organelles (FB-MOs) are exclusively sorted into budding spermatids (Ellis and Stanfield, 2014). Given the very low mobility of PEI granules, we reasoned that they might be associated with one of these membranous organelles. A previous study showed that the myosin VI 270 motor protein SPE-15 is required for proper segregation of mitochondria and FB-MOs into budding spermatids (Kelleher et al., 2000). Thus, we asked whether the asymmetric sorting of PEI granules is subject to the same principle. In absence of SPE-15, PEI granules were detected in both budding spermatids and residual bodies ( Figure 6A), revealing that their subcellular distribution indeed depends on SPE-15, and is thus likely driven by the segregation of mitochondria and/or FB-MOs. We also 275 performed confocal microscopy to determine the subcellular localization of mitochondria in maturing sperm cells in relation to PEI granules. This revealed PEI-1 localization close to mitochondria ( Figure S6A; Movies S2 and S3), without showing any overlaps. To probe FB-MO association, we looked at WAGO-3 and SPE-45::mCherry localization in spermatids that budded off the residual body ( Figure S6B). SPE-45 is a protein known to localize to the membranous organelles of FB-MOs, which 280 are found close to the cell membrane in spermatids. While SPE-45::mCherry showed this peripheral distribution, WAGO-3 foci were detected throughout the cytoplasm, suggesting that WAGO-3 is not stably associated with FB-MOs.
Additional evidence for association of PEI granules with mitochondria comes from an IP-MS/MS experiment on late-L4 stage animals expressing PEI-1::3xMYC from the endogenous pei-1 locus 285 ( Figures 6B and 6C). This experiment identified 172 enriched proteins. The vast majority (82 %), including PEI-1 and WAGO-3, were also detected in a proteomic analysis of male-derived spermatids . Strikingly, 110 proteins (64 %) were previously identified in a proteomic study on purified mitochondria (Jing et al., 2009). We also identified 43 ribosomal proteins (25 %). Both sets are fully consistent with interaction of PEI granules with mitochondria, as previous proteomic and 290 microscopic analyses identified and visualized cytosolic ribosomes on the surface of mitochondria (Gold et al., 2017;Jing et al., 2009). We also identified major sperm protein (MSP), which is the main component of the fibrous body (FB) (Ellis and Stanfield, 2014). We note that MSP was also identified on isolated mitochondria (Jing et al., 2009), suggesting a common contamination or possible interaction of both organelles. Taken together, we propose that PEI granules interact with 295 mitochondria, and possibly with FB-MOs, which transport PEI granules into spermatids in a myosin VIdependent manner.

Correct segregation of PEI granules requires S-palmitoylation
S-palmitoylation is a covalent attachment of palmitic acid to a protein and can serve as membrane anchor, guiding proteins to membranous structures. S-palmitoylation typically occurs on the Golgi 300 complex or on Golgi-derived membranes (Tabaczar et al., 2017). SPE-10, a sperm-specific palmitoyltransferase, has been reported to localize to FB-MOs, and to be required for the stable interaction between the FB and the Golgi-derived membranous part (Gleason et al., 2006). In absence of SPE-10, defects of FB-MOs become apparent during the second meiotic division: the FBs dissociate prematurely from the membranous parts and end up in the residual body, where they frequently bud 305 off in so-called cytoplasts. We found that PEI granules were severely defective in spe-10 mutants. Already in spermatocytes, a stage well before the second meiotic division, PEI-1 localized to large and irregularly shaped patches, which were found arranged along the cell membrane ( Figure 6A). These patches, like the much smaller wild-type PEI granules, were very static and did not show signs of fusion or fission (Movie S4). At later spermatogenic stages, large PEI-1 aggregates were detected in the 310 residual body, leaving no detectable PEI-1 signal in the spermatids ( Figure 6A). No PEI-1 signal was detected in cytoplasts ( Figure 6A), suggesting that the PEI-1 aggregates in the residual body of spe-10 mutants were not FB-associated. Our results show that the molecular function of SPE-10, i.e. Spalmitoylation, is not restricted to FB-MO stabilization, but also affects both shape and segregation of PEI granules. WAGO-3 and PEI-1 still co-localized in absence of SPE-10, indicating that S-palmitoylation 315 does not affect the interaction between both proteins. Rather, we hypothesize that it mediates the association of PEI granules with mitochondria.

DISCUSSION
Our work identifies WAGO-3 as a cytoplasmic Argonaute protein that is inherited via the sperm, and 320 how WAGO-3 is secured within the maturing sperm cells by joining a sperm-specific condensate made by PEI-1. Our findings are summarized in Figure 7, but a number of additional aspects will be discussed below.

PEI granule formation and WAGO specificity
We identified PEI-1, a conserved protein in terms of domain composition that is required for WAGO-3 325 localization in mature sperm. PEI-1 exclusively marks a novel, sperm-specific compartment, the PEI granule, which ensures proper subcellular segregation of WAGO-3. Both the IDR and the structurally ordered region of PEI-1 are able to form assemblies, and likely contribute to granule formation. However, we also found differences. The BTB and BACK domains of PEI-1 have a strong effect on the stability of PEI granules, whereas the IDR is essential for WAGO-3 recruitment. We note that BTB 330 domains can mediate both homo-and heteromeric oligomerization (Collins et al., 2001), which provides multivalency, a property known to drive phase separation (Banani et al., 2017). We propose that the BTB and BACK domains stabilize the PEI-1 IDR-IDR interactions, allowing the formation of stable granules, whereas the IDR also specifies interactions with other proteins, such as WAGO-3.
Although a number of Argonaute proteins have been reported in spermatocytes (Batista et al., 2008;335 Buckley et al., 2012;Conine et al., 2010;Wan et al., 2018), we did not find any of these in spermatozoa within the spermatheca. WAGO-1 and CSR-1 have specifically been proposed to be maintained during sperm cell maturation (Conine et al., 2010(Conine et al., , 2013, however, using tagged proteins expressed from endogenous loci, we could not confirm their presence in mature sperm. Especially WAGO-1 has a very similar protein sequence compared to WAGO-3, making the difference between WAGO-3 and WAGO-340 1 localization during sperm development remarkable. Further experiments will be required to identify the origin of this differential behavior.

Transport of PEI granules
PEI granules are segregated to sperm via myosin VI dependent transport that is also known to be required for proper localization of mitochondria and FB-MOs (Kelleher et al., 2000). Our data suggest 345 that mitochondria are an important vector of transport for the PEI granules, even though we cannot exclude a function for FB-MOs in this process. Interestingly, association of RNAi-related pathways with mitochondria has been described in various animals and include spermatogenic structures like pibodies, piP-bodies, chromatoid bodies and mitochondria-associated ER membranes (Wang et al., 2020). It appears that interaction with mitochondria may be used for various purposes in the context 350 of small RNA pathways, ranging from small RNA biogenesis (Ge et al., 2019;Munafò et al., 2019) to the here identified role in transport and TEI.
An important question to further resolve is how PEI granules can interact with membranous structures. We show that SPE-10 mediated S-palmitoylation plays an important role in PEI granule localization, suggesting that anchoring of PEI granules may proceed via a lipid modification. It remains unknown 355 whether PEI-1 itself is targeted for S-palmitoylation, but the striking subcellular localization of a large PEI-1 deletion that only maintains a few amino acids at both N-and C-termini provides an intriguing clue. The subcellular structures identified by this variant are not mitochondria. By principle of exclusion, they might well be FB-MOs. The observed structures indeed resemble FB-MO morphology, as revealed by electron microscopy (Fabig et al., 2019). This would imply that the most terminal amino 360 12 acids of PEI contain information for FB-MO localization, where PEI-1, or an associated protein, could be palmitoylated by SPE-10 (Gleason et al., 2006). In this scenario, the large PEI-1 deletion visualizes a normally transient association of PEI-1 with FB-MOs, in order to interact with the SPE-10 enzyme. Interestingly, both the BACK domain and the IDR, which we show are required for proper spermatid localization of PEI-1, contain predicted palmitoylation sites ( Figure S6C), suggesting that PEI-1 itself 365 may be modified by SPE-10. More experiments will be needed to test these hypotheses.
Why are PEI granules needed and how do they release their cargo in the embryo?
Why are PEI granules required for paternal TEI and how can they release their contents upon fertilization? The answer to the first question could be related to the fact that in most nematodes the nuclear membrane breaks down during late stages of spermatogenesis (Yushin and Malakhov, 2014). 370 In mature sperm, the highly condensed genome is encapsulated by an RNA-protein halo (Ward et al., 1981). Since P granules are closely associated with the nuclear envelope, they may not be able to act as carrier of epigenetic information in sperm. Indeed, except WAGO-3, all described Argonaute proteins that were reported in P granules are excluded from mature sperm. Hence, a different, non nuclear envelope-associated condensate may be required to stabilize and keep specific proteins in 375 maturing sperm. A different condensate may allow add specificity towards Argonaute inheritance. Whereas P granules are known to house many Argonaute proteins, only one, or a selected set may be appropriate to load into mature sperm. The reduced exchange dynamics of PEI granules compared to P granules might provide another important aspect for why PEI granules are required: PEI granules may be more viscous than P granules, allowing more robust transport. We do note that WAGO-3 380 recovery in PEI granules is still faster than that typically found in gel-like assemblies (Putnam et al., 2019). This may be related to the fact that in order to function in TEI, WAGO-3, together with its 22G RNA cofactors, has to be released into the oocyte, as paternal mitochondria are degraded during early development (Sato and Sato, 2017). We can envisage a number of possibilities of how this release may work. First, upon fertilization paternally inherited factors are massively diluted, which may cause PEI 385 granule disassembly as concentrations drops below critical thresholds for phase separation. Thus, WAGO-3 proteins might efficiently diffuse into the zygote. Second, maternally deposited factors may stimulate WAGO-3 release by mediating post-translational protein modifications, which have been shown to regulate phase separation and affect physical properties of biomolecular condensates (Hofweber and Dormann, 2019). Third, S-palmitoylation is reversible (Tabaczar et al., 2017), raising the 390 possibility that PEI granules as a whole are released after fertilization. It will be interesting to experimentally test these, and other possibilities for how paternal WAGO-3 can be used to prime embryonic 22G RNA biogenesis.

PEI-1 related proteins in other species
How conserved is the mechanism we uncovered? Based on primary sequence, PEI-1 is nematode 395 specific ( Figure S7A). However, when domain organization is considered, PEI-1-related proteins can be easily identified, for example in human ( Figure S7B). Particularly the human protein BTBD7, which is expressed in various tissues including testis, has a domain organization that closely resembles that of PEI-1. Interestingly BTBD7 carries a predicted myristoylation site close to its N-terminus, suggesting it may be membrane bound. The other human proteins shown in Figure S7B are also known to be 400 expressed in testis, and for two of these, BTBD18 and GMCL1 (which is the human homolog of Drosophila gcl), functions during spermatogenesis have been described (Kleiman et al., 2003;Zhou et al., 2017). Interestingly, BTBD18 forms nuclear foci (Zhou et al., 2017), and GMCL1 has been described to interact with IDRs found in primate-specific GAGE proteins, and to affect their localization (Gjerstorff et al., 2012), similar to what we find for PEI-1 and WAGO-3. Hence, it seems likely that the mechanism 405 we here reveal is broadly conserved in germ cell biology, and possibly also beyond.

ACKNOWLEDGEMENTS
We thank all of the current and former members of the Ketting laboratory for helpful discussions and feedback on the manuscript. We are grateful to Helge Grosshans for critical reading of the manuscript. 410 A special thanks to Miroslav Dörr and Svenja Hellmann for excellent technical and experimental support. Clara Werner of the Institute for Molecular Biology Genomics Core Facility is thanked for small RNA library preparation. We would like to thank the Institute for Molecular Biology Media Laboratory, Microscopy, Proteomic and Genomic Core Facilities for consumables, equipment and experimental support. Some strains were provided by the Caenorhabditis Genetics Center (

DECLARATION OF INTEREST 425
The authors declare no competing interests.

Figure. 1. WAGO-3 is required for an immortal germline and associates with paternal small RNAs. 430
A.
Mortal germline assay representing loss of fertility of out-crosses strains with indicated genotype at 25°C. B.
Fluorescence microscopy showing re-activation of a piRNA-targeted transgene in absence of both PRG-1 and WAGO-3. Expression of the transgene was determined after seven generations of homozygosity. No activity was detected in the first double mutant generation. Percentage 435 in lower right corner shows fraction of animals expressing the transgene (n ≥ 50). Scale bars: 30 µm. C.
Mean abundance of defined small RNA populations in the indicated small RNA libraries. D.
Venn diagram showing overlap of protein-coding target genes between WAGO-3, CSR-1 and the Mutator complex. 440 E. Distribution of WAGO-3 target genes. F.
Genomic coverage and distribution of WAGO-3 targeted transposons. G.
Bar chart showing how many of the top sperm-derived 22G RNA target genes (protein-coding), or of all annotated protein-coding genes, were also identified as WAGO-3 target genes. Statistical significance was tested with a Chi-square test (***: p ≤ 0.001).

445
H. Bar chart showing how many of the top WAGO-3 target genes (protein-coding), or of all annotated protein-coding genes, were also identified as sperm-derived 22G RNA target genes. Statistical significance was tested with a Chi-square test (***: p ≤ 0.001, **: p ≤ 0.01, ns: p > 0.05). See also Figure S1. 450 versus the genome-edited strain (GFP::3xFLAG::WAGO-3). The Y-axis represents −log10(p-value) of observed enrichments. Dashed lines show thresholds at P = 0.05 and twofold enrichment. Blue and green data points represent above and below threshold, respectively. WAGO-3 and PEI-1 are highlighted with red data points. B.

C.
Schematic summarizing spermatogenesis in C. elegans. D.
Confocal maximum intensity projections of male-derived budding spermatids expressing GFP::3xFLAG::WAGO-3 in presence and absence of PEI-  22G RNAs, whereas PRG-1 is required for piRNA biogenesis and function. B-C.
A. PEI-1 protein composition in terms of secondary and fixed tertiary structure. Prediction of naturally disordered regions was performed using PONDR VSL2 and PONDR VL3 algorithms. Secondary structure elements were predicted using Jpred4 (Drozdetskiy et al., 2015). Helices are marked as green tubes, and sheets as magenta arrows. 485 B.
Confocal maximum intensity projections of isolated spermatocytes (male-derived), budding spermatids (male-derived) and spermatozoa within the spermatheca (hermaphrodite) expressing indicated proteins. White arrow heads mark budding spermatids, red arrows 490 indicate residual bodies. Hoechst33342 was used to stain DNA. Asterisks indicate optical sections. Scale bars: 4 µm. See also Figures S4.

Figure 5. PEI granules are independent condensates that retain WAGO-3 via hydrophobic 495 interactions.
A. Confocal micrographs of isolated male-derived spermatocytes and budding spermatids after treatment with 1,6-hexanediol. White arrow heads mark budding spermatids, red arrows indicate residual bodies. Hoechst33342 was used to stain DNA. Scale bars: 4 µm. B.
Time sequence of GFP::3xFLAG::WAGO-3, taken from movie S1. Images are confocal maximum intensity projections of an isolated male-derived spermatocyte, which is marked by a dashed 505 circle. Scale bar: 4 μm. See also Figure S3 and S5 and Movie S1.
Volcano plot representing label-free proteomic quantification of PEI-1::3xMYC 515 immunoprecipitation experiments from late-L4 stage hermaphrodite extracts. The X-axis indicates the mean fold enrichment of individual proteins in the control (wild-type PEI-1) versus the genome-edited strain (PEI-1::3xMYC). The Y-axis represents −log10(p-value) of observed enrichments. Dashed lines show thresholds at P = 0.05 and twofold enrichment. Blue and green data points represent above and below threshold, respectively. PEI-1 is highlighted with a red 520 data point. C.
Venn diagram and bar chart showing comparison of enriched proteins from (B) with published spermatid proteome, mitochondria proteome and ribosomal proteins. See also Figure S6 and Movies S2-S4.

C. elegans culture and strains 530
Unless otherwise stated, all worm strains were cultured according to standard laboratory conditions at 20°C on Nematode Growth Medium (NGM) plates seeded with Escherichia coli OP50 (Brenner, 1974). Animals for IP-MS/MS experiments were grown on egg plates (90 mm diameter) (Schweinsberg and Grant, 2013) for one generation, synchronized by bleaching, and then grown on standard NGM plates (90 mm diameter) for one generation before harvest. Egg plates were generated by thoroughly 535 mixing egg yolk with 50 ml LB media/egg. Following incubation at 65°C for 2-3 hours, the mixture was allowed to cool to room temperature before adding 10 ml OP50 culture/egg. About 10 ml was put on top of standard NGM plates (90 mm diameter) and incubated at room temperature. Next day, excess liquid was decanted and egg plates were incubated at room temperature for another two days.

Mortal germline assay
All mutant strains were confirmed and out-crossed four times before starting the experiment. For each strain, 90 L2 or L3 animals were distributed to 15 NGM plates (90 mm diameter), resulting in six larvae per plate. Animals were grown at 25°C. Worms were picked onto fresh plates every second generation. 545 The experiment was stopped after 17 generations.

Mutator-induced sterility crosses
All strains were confirmed and out-crossed two times before setting up crosses. We note that outcrossing ensured comparable results as an enhanced Mis phenotype was observed when using nonout-crossed animals. The transgenic allele otIs45 [unc-119p::gfp] V. was always present in paternal 550 strains and served as mating control to avoid picking self-fertilized offspring. Only L2 stage F1 animals were picked onto individual plates to avoid any biased selection. After three days, male or dead F1 animals were excluded from the analysis. Fertility of F1 animals was determined by the presence of F2 animals after another two to four days.

Immunoprecipitation experiments 625
Unless otherwise stated, synchronized animals were cultured at 20°C until late-L4 stage, harvested with M9 buffer and frozen on dry ice in sterile water and 200 µl aliquots. Aliquots were thawed on ice, mixed with same volume of 2x lysis buffer (50 mM Tris HCl pH 7.5, 300 mM NaCl, 3 mM MgCl2, 2 mM DTT, 0.2 % Triton™ X-100, cOmplete™ Mini EDTA-free Protease Inhibitor Cocktail (Art. No. 11836170001, Roche)) and sonicated using a Bioruptor® Plus device ( according to the manufacturer's instructions. Extracts were separated from non-conjugated Dynabeads™, combined with antibody-conjugated Dynabeads™ and incubated with rotation for 2 h at 4°C. Following three washes with 500 µl 1x wash buffer, antibody-conjugated Dynabeads™ were resuspended in 25 µl 1.2x Novex™ NuPAGE™ LDS sample buffer supplemented with 120 mM DTT. For RIP experiments, immunoprecipitations were performed as described above with the following 650 modifications: i) adult animals were harvested, ii) soluble worm extract was diluted to 650 µl and a total protein concentration of 7 µg/µl, of which 150 µl served as input sample for later RNA extraction, iii) antibody-conjugated Dynabeads™ were resuspended in 50 µl nuclease-free water.
Immunoprecipitation experiments associated with mass spectrometry and small RNA sequencing were performed in quadruplicates and triplicates, respectively. 655
22G reads mapping to features in the custom annotation were counted using htseq-count v0.9.0 (Anders et al., 2015) (-s no -m intersection-nonempty). Differential targeting analysis was carried out in R® using DESeq2 (Love et al., 2014) with a strict cut-off for the adjusted p-value of 0.01. A cut-off for fold-change (IP versus input) was determined by fitting a Gaussian to the fold-change-distribution of 740 reads mapping to miRNA, which are known not to bind WAGOs, and choosing the value that corresponds to a false discovery rate (FDR) of 5% for this RNA species; here log2-fold-change > 1.3.
Protein-coding target genes of WAGO-3 were compared to: i) protein-coding target genes of CSR-1 , ii) protein-coding target genes of siRNAs downregulated in mut-16 mutant animals (Phillips et al., 2014), and iii) protein-coding target genes of sperm-derived 22G RNAs 745 (Stoeckius et al., 2014). To determine germline expression, protein-coding target genes of WAGO-3 were compared to lists of genes expressed in the C. elegans germline of either fem-3 or fog-2 mutant animals (Ortiz et al., 2014).
Reads mapping to intronic, exonic, or untranslated regions were counted using a custom Python script. Reads mapping at exon-intron junctions were counted as 0.5 intronic and 0.5 exonic regardless of the 760 spanned region.
Time series of spermatocytes expressing GFP::3xFLAG::WAGO-3 were acquired with a fluorescence 780 spinning disk confocal microscope (SDCM) from Visitron Systems (VisiSope 5Elements) based on a Nikon Ti-2E stand and a spinning disk from Yokogawa (CSU-W, 50 µm pinhole) controlled by the VisiView® software. The microscope was equipped with a 60x plan apochromatic water immersion objective (CFI Plan Apo VC, NA 1.2), a twofold magnification lens in front of the sCMOS camera (BSI, Photometrics), and a stage-top incubation chamber for live imaging (20°C, ambient CO2). The sample 785 was excited by an argon laser at λex = 488 nm (200 mW, power set to 20 %) and the emission was detected in a range of λem = 500 -550 nm (ET525/50m, Chroma).

FRAP
FRAP measurements were performed on a TCS SP5 Leica confocal microscope equipped with a FRAPbooster and a HCX PL APO 63x water objective (NA 1.2). An entire granule was bleached in a fixed 790 region of interest (ROI) (0.9 µm diameter), while two additional control ROIs of same size were used to detect fluorescence emission of an unbleached granule and background signal, respectively. Five pre-bleach frames were recorded (5x 0.374 s/frame), followed by two bleach frames (2x 0.374 s/frame), and 3 sets of post-bleach frames (10x 0.5 s/frame, 10x 5 s/frame, 15x 10 s/frame). Data analysis including full scale normalization and curve fitting using a double term exponential equation 795 was performed using EasyFRAP-web (Koulouras et al., 2018) B C protein-coding (1794) transposon (124) pseudogene (118) ncRNA (51) snoRNA (44) other (