Endogenous RNAi pathways in C. elegans^*

The phenomenon of RNA interference (RNAi), wherein introduction of dsRNA results in silencing of homologous genes, was first observed and reported in C. elegans. In addition, when the first small RNA, the lin-4 microRNA, was identified, it was thought to be a curiosity of C. elegans biology until the discovery of the conserved let-7 family of microRNAs. Since then, thousands of unique endogenous small RNAs have been found in C. elegans falling into three distinct classes: microRNAs, endogenous small interfering RNAs (endo-siRNAs), and Piwi-interacting RNAs (piRNAs). All three types of small RNAs bind to Argonaute effector proteins, recognize target transcripts exhibiting partial or perfect complementarity, and direct target regulation that is primarily inhibitory in nature. Yet microRNAs, endo-siRNAs, and piRNAs vary greatly in biogenesis mechanisms, specific protein cofactors, and even effector function. The details of C. elegans microRNA biology have been previously described (Fischer, 2010) and are addressed in this review only insofar as they intersect with endo-siRNA and piRNA biology. Similarly, a thorough examination of exogenously-derived siRNAs and exogenous RNAi (exo-RNAi), and its function as an anti-viral silencing response, is beyond the scope of this review; however, the gene silencing mechanisms initiated by exo-RNAi are discussed, as C. elegans exo-RNAi engages a downstream endo-RNAi amplification pathway also shared by primary endo-siRNAs and piRNAs.

The gene silencing capacity of antisense RNA was first described in C. elegans over two decades ago (Fire et al., 1991). The effective interfering agent was subsequently determined to be double-stranded RNA (dsRNA), and its incredible potency suggested the existence of a catalytic or amplification mechanism engaged by exogenous dsRNA (Fire et al., 1998). This dsRNA is processed into primary exo-siRNAs by the conserved RNase III enzyme Dicer (Zamore et al., 2000; Bernstein et al., 2001), increasing the ratio of trigger to target, but still to a degree insufficient to explain the potency of exo-RNAi. Further studies of the small RNA effector populations during exo-RNAi in C. elegans revealed that primary exo-siRNAs are not the ultimate effectors of interference; rather, they trigger vigorous production of secondary siRNAs by RNA-dependent RNA polymerases (RdRPs), amplifying the signal and focusing the interference response on expressed sequences (Sijen et al., 2001). Before the discovery of endogenous silencing pathways in C. elegans, isolation of mutations causing deficiencies in both exo-RNAi and endogenous silencing phenomena such as transposon silencing suggested a shared mechanism (Tabara et al., 1999; Ketting et al., 1999; Ketting and Plasterk, 2000). Mutations in core molecular machinery of RNAi also were found to result in phenotypes indicating essential endogenous roles. Loss of DCR-1, the sole C. elegans ortholog of Dicer, results in profound cell fate specification defects and germline abnormalities leading to lethality and sterility (Grishok et al., 2001; Ketting et al., 2001; Knight and Bass, 2001); the RdRP EGO-1 is required for germline development (Smardon et al., 2000); and mutants defective in transposon silencing and RNAi show compromised fertility (Ketting et al., 1999; Tijsterman et al., 2002a; Vastenhouw et al., 2003; Chen et al., 2005).

The discovery of the first microRNA, lin-4, in C. elegans suggested that endogenous small RNAs also initiate transcript regulation through antisense mechanisms (Lee et al., 1993; Wightman et al., 1993). After the identification of a second microRNA, let-7, in 2000 (Reinhart et al., 2000), cloning approaches were successfully applied to expand dramatically the repertoire of microRNAs in C. elegans (Lau et al., 2001; Lee and Ambrose 2001). It was not until a decade after the discovery of the first microRNA that other endogenous small RNAs were identified in C. elegans (Ambros et al., 2003; Lim et al., 2003), and only in 2006 did small RNA deep sequencing first reveal the incredible diversity of C. elegans small RNAs (Ruby et al., 2006). Among the species identified were a large pool of 5’ guanosine antisense small RNAs identified as endo-siRNAs that appeared to represent distinct 26- and 22-nucleotide (nt) subpopulations, later determined to correspond to primary and secondary endo-siRNAs, respectively (Ruby et al., 2006; Pak and Fire, 2007; Sijen et al., 2007; Han et al., 2009; Conine et al., 2010; Vasale et al., 2010). Subsequent dissection of these 26G and 22G RNA populations identified unique subgroups with largely overlapping biogenesis requirements but that engage different effector pathways distinguished by the particular Argonaute protein(s) interacting with the small RNAs. C. elegans encodes 27 Argonaute proteins. Roughly half of these proteins belong to a worm-specific Argonaute (WAGO) clade and bind 22G RNAs of particular subclasses, the WAGO 22G RNAs and CSR-1 22G RNAs; the former represent secondary siRNAs that effect target silencing, whereas the latter are a class of siRNAs complementary to germline-expressed transcripts that do not silence target genes but rather promote proper chromosome organization during mitosis (Claycomb et al., 2009; Gu et al., 2009). Also identified in the initial deep sequencing dataset were the 21U RNAs, 5’ uridine 21-nt small RNAs later determined to represent the piRNAs of C. elegans (Batista et al., 2008; Das et al., 2008). This review summarizes the literature contributing to our current understanding of the C. elegans 26G RNAs, WAGO 22G RNAs, CSR-1 22G RNAs, and 21U RNAs, discussing mechanisms of triggering, biogenesis, and effector function, where known. Whereas microRNAs are required for diverse developmental and physiological processes in the soma, endo-siRNAs and 21U RNAs serve as the guardians of the immortal germline, constituting a complex, interconnected, and tremendously robust system for surveillance of the C. elegans genome.

Early deep sequencing of C. elegans small RNAs revealed a distinct population of 26 nt species that are largely anti-sense to annotated genes and therefore classified as endo-siRNAs (Ruby et al., 2006). Like endo-siRNAs sequenced earlier in C. elegans (Ambros et al., 2003), these 26 nt species show a 5’ guanosine bias and are thus termed 26G RNAs (Ruby et al., 2006; Asikainen et al., 2008; Han et al., 2009; Stoeckius et al., 2009; Gent et al., 2010) (Table 1). 26G RNAs are 5’ monophosphorylated (Ruby et al., 2006; Han et al., 2009; Stoeckius et al., 2009; Gent et al., 2010). 26G RNAs are quite enriched in male and female germlines (Han et al., 2009) where they comprise two distinct subpopulations that are temporally isolated and bound by unique, germline-specific effector complexes. 26G RNAs in the spermatogenic gonad are bound by redundant Argonautes ALG-3 and ALG-4, whereas 26G RNAs in oogenic gonad are bound by the ERGO-1 Argonaute (Han et al., 2009; Conine et al., 2010; Vasale et al., 2010). ERGO-1 class 26G RNAs are also highly abundant in embryos (Han et al., 2009; Stoeckius et al., 2009) and perdure through early larval development (Gent et al., 2010).

Little is known about how transcripts are selected for targeting by 26G RNAs. 26G RNAs map primarily to protein-coding genes with a strong antisense bias, although some target unannotated loci (Ruby et al., 2006; Han et al., 2009; Stoeckius et al., 2009). They are transcribed from spliced mRNA templates by the RNA-dependent RNA polymerase RRF-3 (Gent et al., 2009; Han et al., 2009; Conine et al., 2010; Gent et al., 2010), as indicated by the sequencing of rare species spanning exon-exon junctions (Ruby et al., 2006; Han et al., 2009; Gent et al., 2010) and by the loss of complementary 26G RNAs upon nonsense-mediated decay of template mRNA (Han et al., 2009).

Targets of ALG-3/4 class 26G RNAs are highly enriched for transcripts classified as spermatogenesis-enriched (Reinke et al., 2004; Gent et al., 2009; Han et al., 2009; Conine et al., 2010), explaining earlier observations connecting endo-siRNAs primarily with regulation of sperm function genes (Ruby et al., 2006; Asikainen et al., 2007). ALG-3/4 class 26G RNAs map across the lengths of their target sequences, but preferentially target transcript 5’ and 3’ termini; transcripts exhibiting higher 5’ UTR targeting appear to be more efficiently silenced (Conine et al., 2010).

Although biogenesis of ERGO-1 class 26G RNAs initiates in oogenesis, their targets are depleted of germline-intrinsic transcripts (Reinke et al., 2004; Han et al., 2009); accordingly, ERGO-1 class 26G RNAs primarily regulate zygotic targets throughout development (Han et al., 2009; Gent et al., 2010). The majority of abundant ERGO-1 class 26G RNAs map to within 5 Mb of chromosome ends at more gene-poor regions (Vasale et al., 2010; Fischer et al., 2011). Unlike ALG-3/4 class 26G RNAs, ERGO-1 class 26G RNAs are excluded from the first ~100 nt of target transcripts (Vasale et al., 2010). Roughly half of ERGO-1 class 26G RNAs map to coding loci, lincRNAs, or pseudogenes and half to loci that likely correspond to unannotated transcription units (Vasale et al., 2010). Many of their targets represent recently duplicated, non-conserved genes, suggesting ERGO-1 class 26G RNAs may buffer expression of newly acquired genes (Vasale et al., 2010; Fischer et al., 2011). A large-scale proteomics dataset (Schrimpf et al., 2009) shows peptides corresponding to less than 20% of ERGO-1 class 26G RNA targets, but 54% of all annotated coding genes, indicating that very few of these targets represent functional, coding loci (Fischer et al., 2011).

The ERI complex mediates 26G RNA biogenesis and is named for the enhanced (exogenous) RNAi (Eri) phenotype that is associated with compromise of ERGO-1 class 26G RNA function (Table 2 and Figure 1). Loss of ERI complex members also results in temperature-sensitive (ts) sterility at 25°C due to defective spermatogenesis as well as a high incidence of male progeny (Him) phenotype, indicative of increased X chromosome nondisjunction, due to compromised ALG-3/4 class 26G RNA function. The Eri and ts sterile phenotypes are explained in greater detail below. At the heart of the ERI complex is a core RdRP module consisting of RRF-3, DRH-3, and ERI-5. Associated with this module are DCR-1, accessory factors ERI-1b, ERI-3, RDE-4, and possibly the RNA phosphatase PIR-1. Analyses of DCR-1, ERI-1b, and ERI-5 complexes immuno-purified from gravid adult and embryo have revealed interactions with each other member of the ERI complex (Duchaine et al., 2006; Thivierge et al., 2012), and the ~850 kD estimated mass of the ERI complex is similar to the summed mass of these proteins, ~810 kD (Thivierge et al., 2012). The ERI complex shares only DCR-1 and RDE-4 with the RDE (exogenous RNAi defective) complex, which mediates processing of dsRNA trigger to initiate the exo-RNAi pathway (Tabara et al., 2002). DCR-1 and ERI-1 also both interact with ERI-9, another factor implicated in ERGO-1 class 26G RNA biogenesis (Thivierge et al., 2012), suggesting that the ERI complex includes other transient interactors.

ERI complex factors show some interdependence and hierarchy of assembly. Most notably, ERI-5 is proposed to tether the RdRP module to DCR-1 to potentiate 26G RNA biogenesis. While the other factors are required for biogenesis of 26G RNAs, loss of ERI-5 merely attenuates their expression due to partial compensation by paralog EKL-1 (Thivierge et al., 2012). EKL-1 interacts with DRH-3 (Gu et al., 2009) and substitutes for ERI-5 in the RdRP module, but it does not interact with DCR-1 (Thivierge et al., 2012). Untethered to DCR-1, this EKL-1 RdRP module still produces 26G RNAs of normal genomic distribution and in a DCR-1-dependent manner, but 26G RNA accumulation is impaired (Thivierge et al., 2012).

The ERGO-1 and ALG-3/4 Argonautes are required for accumulation of their respective 26G RNA populations (Han et al., 2009; Conine et al., 2010; Gent et al., 2010; Vasale et al., 2010), but they are not found in the ERI complex and are thought to act downstream of biogenesis. While many of the Argonautes encoded by C. elegans do not show conservation of the three critical catalytic residues mediating slicer activity, all of the Argonautes that bind primary small RNAs (ERGO-1, ALG-3/4, PRG-1/2, RDE-1, ALG-1/2, and, putatively, CSR-1) show intact catalytic triads in their RNase H-related PIWI domains (Yigit et al., 2006). Although the slicer activity of an Argonaute refers generally to the ability to catalyze target cleavage, it appears also to play a critical role in effector complex maturation through passenger strand removal. The catalytic activity of RDE-1, the Argonaute that binds primary exo-siRNAs, is required only for efficient removal of the passenger strain; thus, the catalytic mutant of RDE-1 shows only a partial Rde phenotype attributable to impaired target mRNA interaction (Steiner et al., 2009). A putative ERGO-1 catalytic mutant exhibits an Eri phenotype (Fischer et al., 2011). This indicates that catalytic activity is required for ERGO-1-mediated triggering of WAGO 22G RNAs, but 26G RNA and passenger strand levels have not been assessed. The ergo-1 mutant shows vastly depleted ERGO-1 class 26G RNA levels but two-fold increased passenger strand levels (Fischer et al., 2011), supporting a role for ERGO-1 slicer activity in liberating the passenger strand from a 26G RNA dsRNA intermediate. In comparison, both mature ERGO-1 class 26G RNAs and passenger strands are depleted by loss of ERI-1 or ERI-6/7, suggesting function upstream of 26G RNA duplex formation (Fischer et al., 2011). Similar analyses of catalytically inactive ALG-3 or ALG-4 have not been reported.

Loss of factors involved in ERGO-1 and/or ALG-3/4 class 26G RNA biogenesis or accumulation results in desilencing of the relevant population of complementary target mRNAs (Duchaine et al., 2006; Lee et al., 2006; Gent et al., 2009; Han et al., 2009; Pavelec et al., 2009; Conine et al., 2010; Gent et al., 2010; Vasale et al., 2010) (Table 3). Factors that play lesser roles in accumulation such as ERI-5 or Mutator foci MUT proteins show less pronounced silencing defects (Duchaine et al., 2006; Zhang et al., 2011). While several factors have been identified that appear to contribute exclusively to ERGO-1 class 26G RNA accumulation (Pavelec et al., 2009; Fischer et al., 2011; Billi et al., 2012; Kamminga et al., 2012; Montgomery et al., 2012; Zhuang and Hunter, 2011), ALG-3 and ALG-4 remain the only factors known to be specifically required for accumulation of ALG-3/4 class 26G RNAs (Han et al., 2009; Conine et al., 2010). 26G RNAs repress target mRNA expression in their cognate cell types; loss of 26G RNAs does not result in ectopic expression within other cell types of the native male or female germline or inappropriate target expression in the opposite germline (Han et al., 2009).

During 26G RNA biogenesis, DCR-1 likely catalyzes the cleavage of the mRNA template in generating the dsRNA intermediate (Welker et al., 2011), thereby reducing mRNA levels. Subsequently, 26G RNA Argonaute recognition may also trigger target cleavage. However, neither is likely to represent a significant means of target silencing, as WAGO 22G RNAs triggered by 26G RNA targeting are the major effectors of the 26G RNA endo-RNAi pathway. These secondary siRNAs, discussed in depth below, are strictly required for 26G RNA target silencing (Conine et al., 2010; Gent et al., 2010; Vasale et al., 2010), as well as silencing of other small RNA pathway targets (Figure 2). It is not known precisely how 26G RNA RISC association triggers 22G formation, but target transcripts show 26G RNA-dependent accumulation of WAGO 22G RNAs (Conine et al., 2010; Gent et al., 2010; Vasale et al., 2010), and many 22G and 26G RNAs originate from the same 5’ nt (Gent et al., 2010).

22G RNAs amplify the silencing signal of 26G RNAs and increase the perdurance of the repression. ERGO-1 class 26G RNA levels peak in embryos and decline significantly throughout larval development (Han et al., 2009; Gent et al., 2010; Vasale et al., 2010); however, ERGO-1 class 26G RNAs primarily target transcripts that are not germline-intrinsic (Reinke et al., 2004; Han et al., 2009), and triggering the production of WAGO 22G RNAs enables their silencing influence to persist through larval development (Han et al., 2009; Gent et al., 2010; Vasale et al., 2010). Similarly, 26G RNAs and ALG-3/4 are depleted in mature sperm, but WAGO 22G RNA Argonaute WAGO-1 is abundant in mature sperm (Conine et al., 2010). This precludes significant inheritance of paternal 26G RNAs, but evidently not dependent 22G RNAs: an ALG-3/4 class 26G RNA-dependent 22G RNA can be detected even at the L1 larval stage (Han et al., 2009).

Analysis of 26G RNA-dependent 22G RNAs further suggests that 26G RNAs target transcripts in trans as well as in cis. The X-cluster (Ambros et al., 2003) describes a region of the X chromosome from which abundant 22Gs, but none or very few 26Gs, are generated (Vasale et al., 2010). Accumulation of X-cluster 22G RNAs requires ERGO-1 (Yigit et al., 2006) and a particular ERGO-1 class 26G RNA derived from the K02E2.11 transcript, which shares no other significant homology with the X-cluster (Montgomery et al., 2012). This particular 26G RNA shows multiple putative target sites throughout the X-cluster. None is perfectly complementary, and the best sites contain a minimum of three adjacent, central mismatches and two wobble pairs (Montgomery et al., 2012). This indicates that 26G RNA targeting of mRNA targets to initiate 22G RNA production in trans is mismatch-tolerant, suggesting immense targeting capacity.

Early capture of C. elegans small RNAs using a protocol insensitive to 5’ structure indicated that a majority of antisense small RNAs are 22 nt in length (Ambros et al., 2003). Protocols selective for 5’ monophosphorylated species failed to recover these sequences effectively (Ruby et al., 2006; Lim et al., 2003), revealing a 5’ structure different from that of the 26G RNAs. Subsets of these small RNAs were subsequently shown to exhibit a 5’ triphosphate (Pak and Fire, 2007) and a 3’ hydroxyl (Pak and Fire, 2007; Sijen et al., 2007; Gu et al., 2009). Finally, 22 nt siRNAs in C. elegans show a prominent 5’ guanosine bias (Ambros et al., 2003; Ruby et al., 2006; Gu et al., 2009), distinguishing these as 22G RNAs.

As a whole, 22G RNAs target about ~50% of the annotated coding genome, with most 22G RNAs targeting unique genome sequences (Gu et al., 2009). 22G RNAs mapping antisense to mRNAs show enrichment primarily at transcript 3’ ends, consistent with RdRP engagement at the mRNA 3’ terminus, but some also show enrichment at 5’ ends (Gu et al., 2009). Most 22G RNAs are germline expressed and deposited into embryo, potentially coinherited with their mRNA targets (Gu et al., 2009). The 22G RNAs comprise two distinct classes of small RNAs that are synthesized by similar RdRP modules but engage unique pathways mediated by specific Argonautes and cofactor proteins to effect entirely distinct outcomes. The WAGO 22G RNAs bind semi-redundant Argonautes of the worm-specific WAGO clade to mediate silencing of certain protein-coding genes, transposons, pseudogenes, and cryptic loci through both transcriptional and post-transcriptional mechanisms (Guang et al., 2008; Gu et al., 2009; Burkhart et al., 2011; Burton et al., 2011; Ashe et al., 2012; Buckley et al., 2012; Gu, S.G., et al. 2012; Luteijn et al., 2012; Shirayama et al., 2012). The CSR-1 22G RNAs bind CSR-1, another WAGO Argonaute, and target germline-expressed genes to fulfill an essential role in promoting chromosome segregation (Claycomb et al., 2009). Thus, 22G RNAs target both silent and expressed loci genome-wide to maintain the germline.

RdRPs RRF-1 and EGO-1 both contribute to biogenesis of 22G RNAs (Figure 2). The ego-1; rrf-1 double mutant lacks 22G RNAs synthesized de novo in the germline (Gu et al., 2009). EGO-1 and RRF-1 proteins show greater than 50% sequence identity (Smardon et al., 2000) and collaborate to generate WAGO 22G RNAs, but EGO-1 alone is required for CSR-1 22G RNAs (Claycomb et al., 2009; Gu et al., 2009). Accordingly, the ego-1 transcript is classified as germline-intrinsic (Reinke et al., 2004), and ego-1 mRNA and protein are enriched in young adult, adult, and embryo but nearly absent during larval development (Smardon et al., 2000; Vought et al., 2005; Claycomb et al., 2009). EGO-1 protein is also enriched in the nuclear fraction of cell lysates (Maine et al., 2005). In contrast, a transgene expressing GFP::RRF-1 that rescues the rrf-1 mutant phenotype shows robust somatic expression, with prominent cytoplasmic and weak nuclear localization (Aoki et al., 2007). Despite their different expression patterns, the rrf-1 gene is encoded directly downstream of ego-1 in an operon (Mangone et al., 2010).

Because exo-RNAi produces WAGO 22G RNAs to mediate transcript knockdown (Sijen et al., 2001; Pak and Fire, 2007; Sijen et al., 2007), the RNAi sensitivity phenotypes associated with loss of RRF-1 versus EGO-1 reveal their respective contributions to the WAGO pathway. RRF-1 is required in somatic tissues for 22G RNA accumulation and therefore exo-RNAi (Sijen et al., 2001; Gent et al., 2010; Vasale et al., 2010), whereas no role for EGO-1 is detected during RNAi in the soma (Smardon et al., 2000; Sijen et al., 2001). Accordingly, 22G RNAs mapping to ERGO-1 class 26G RNA targets are largely RRF-1-dependent (Vasale et al., 2010). In the germline, however, EGO-1 and RRF-1 are partially redundant for WAGO 22G RNA biogenesis. The rrf-1 mutant germline is sensitive to exo-RNAi of germline transcripts (Sijen et al., 2001), revealing compensation by EGO-1. This explains why loss of RRF-1 does not recapitulate the ts sterility and other germline phenotypes associated with total compromise of WAGO 22G RNA target silencing. Loss of EGO-1 results in a partial germline Rde phenotype (Smardon et al., 2000). Interestingly, injection of ego-1 dsRNA enhances the germline defects of an ego-1 null mutant (Smardon et al., 2000), suggesting the possibility that RRF-1 may be recruited, although insufficiently, to CSR-1 22G RNA biogenesis in the absence of EGO-1. EGO-1 and phenotypes associated with its loss are discussed further with CSR-1 22G RNAs, below (Section 22).

WAGO 22G RNAs serve as the critical amplification pathway upon which most primary small RNAs in C. elegans converge. Accordingly, WAGO 22G RNAs are generated from and target transcripts targeted by 26G RNAs, primary exo-siRNAs, RDE-1-scavenged small RNAs, and 21U RNAs. The requirements for triggering of WAGO 22G RNAs by each of these primary small RNA types are addressed below (Sections 9.1, 9.2, 9.3, and 9.4).

Secondary siRNAs in C. elegans were first identified during exo-RNAi as a population of small RNAs that are antisense to a target mRNA but not derived from the initial dsRNA trigger (Sijen et al., 2001). However, the existence of some means of amplification had been hypothesized several years prior. It had been noted that a few molecules of dsRNA are sufficient to deplete a cellular pool of target mRNAs, arguing against simple stoichiometric interaction and indicating the involvement of a catalytic and/or amplification mechanism (Fire et al., 1998; Montgomery and Fire, 1998). Although trigger dsRNA cleavage generates some measure of amplification, this was recognized as still insufficient to explain the potent silencing ability of exogenous dsRNA (Sijen et al., 2001). The essential contribution of RdRP activity to this amplification was subsequently established by demonstrating complete insensitivity of the rrf-1 mutant soma to injection of prepared siRNA duplexes (Sijen et al., 2001).

WAGO 22G RNAs are generated from spliced transcripts targeted by the triggering primary small RNAs (Pak and Fire, 2007). RdRP initiation appears to be non-random however as 22G RNAs show some phasing across a target transcript, and similar sets of 22G RNAs are produced in different transgenic lines (Sijen et al., 2007). 22G RNA biogenesis shows a limited degree of spreading from the target site of the primary small RNA (Sijen et al., 2001). During exo-RNAi, spreading occurs primarily 5’ of the primary RNA target site with respect to the sense of the target transcript (Sijen et al., 2001; Pak and Fire, 2007; Sijen et al., 2007). ERGO-1 class 26G RNAs rarely target the first ~100 nt of a transcript, but 22G RNAs show robust coverage of this interval at these same targets (Vasale et al., 2010). Similarly, 21U RNAs trigger 22G RNA production primarily 5’ of the target site (Lee et al., 2012). This may have implications for transgenic construct design, as transgenes encoding foreign sequence 3’ to endogenous sequence are more likely to evade silencing by 21U RNAs (Shirayama et al., 2012); presumably, 5’ spread of 22G RNA biogenesis from foreign to endogenous sequence could be detrimental, resulting in selection for evaders. The range of the spreading is relatively modest and may vary by primary small RNA abundance. Exo-RNAi appears to generate secondary siRNAs with a range of 100-180 nt 5’ to the region targeted by dsRNA (Alder et al., 2003). 21U RNAs trigger production of 22G RNAs within the surrounding 40-100 nt (Bagijn et al., 2012; Lee et al., 2012).

The observation that some natural isolates of C. elegans are mutators (strains that have a high spontaneous germ line mutation rate) not only led to the use of these strains in genetic screens but also to research into the causal “mutagen”, DNA transposons of the Tc1/mariner family, including Tc1 and Tc3. To understand how activity of these transposons is regulated, genetic screens were performed to identify mutants that display increased Tc1 or Tc3 transposition in the germline (Collins et al., 1987; Ketting et al., 1999; Vastenhouw et al., 2003). The description of the first mutator (Mut) gene identified, mut-7 (Ketting et al., 1999), and of the first genetic screen for exo-RNAi-defective (Rde) mutants (Tabara et al., 1999), provided the first evidence of a role for RNAi in transposon silencing in the germline: many Rde mutants displayed Mut phenotypes and many Mut mutants displayed Rde phenotypes. These and subsequent screens for related phenomena (Ketting et al., 1999; Vastenhouw et al., 2003; Grishok et al., 2005; Kim et al., 2005; Robert et al., 2005) recurrently identified members of a group of proteins that have now been implicated in formation of Mutator foci. These perinuclear processing compartments interact with RRF-1 and are required for WAGO 22G RNA amplification and target silencing in the germline (Phillips et al., 2012). The six proteins that assemble to form Mutator foci are MUT-16, MUT-7, MUT-8/RDE-2, MUT-2/RDE-3, MUT-15, and MUT-14. Consistent with their critical, non-redundant role in the WAGO 22G RNA pathway that silences transposons, loss of any of these proteins results in mutator (Mut), ts sterile, Him, and Rde phenotypes (Collins et al., 1987 Ketting et al., 1999; Tijsterman et al., 2002a; Vastenhouw et al., 2003; Chen et al., 2005; Tops et al., 2005; Gu et al., 2009; Phillips et al., 2012), although a mut-14 mutant carrying a missense mutation shows only a germline Rde phenotype (Tijsterman et al., 2002a; Phillips et al., 2012).

MUT-16 is a worm-specific protein containing proline-rich and glutamine/asparagine-rich regions (Vastenhouw et al., 2003). A translational MUT-16::GFP fusion protein is expressed broadly in cytoplasm and nuclei (Vastenhouw et al., 2003). MUT-16 is also required for accumulation of ERGO-1 class 26G RNAs, whereas other Mutator foci components are not strictly required (Zhang et al., 2011). MUT-7 is a 3’ to 5’ exonuclease (Ketting et al., 1999) that is conserved in animals (Phillips et al., 2012). A mut-7::gfp transcriptional fusion shows expression in many different cell types in larva and adult (Hunt-Newbury et al., 2007). MUT-8/RDE-2 contains no known domains and may only exist to recruit MUT-7 (Tops et al., 2005; Phillips et al., 2012). MUT-2/RDE-3 is a β nucleotidyltransferase; mutations of conserved residues at the active site produce Rde phenotypes, suggesting polymerase activity is required for function (Chen et al., 2005). MUT-2/RDE-3 expressed in frog oocytes shows no polymerase activity, unlike homolog CDE-1 (Kwak and Wickens, 2007). MUT-15 contains no known domains. MUT-14 is a DEAD box RNA helicase (Tijsterman et al., 2002a).

The RDE-10/RDE-11 protein complex amplifies 22G RNAs and potentiates degradation of the targets that serve as their templates (Yang et al., 2012; Zhang et al., 2012). RDE-10 and RDE-11 are worm-specific genes (Yang et al., 2012). WormBase reports RDE-10 as containing a Maelstrom domain, which adopts an RNase H fold that may confer nuclease or RNA-binding functionality. RDE-11 contains a RING-type zinc finger domain (Yang et al., 2012). Mutation of a key zinc-coordinating residue in RDE-11 abrogates its interaction with RDE-10 in vitro and results in failure to rescue rde-11 mutant phenotypes (Yang et al., 2012). Thus, zinc binding facilitates the formation of the RDE-10/RDE-11 complex, possibly by promoting folding of RDE-11.

Early analysis of C. elegans Argonautes revealed significant expansion of the protein family (Grishok et al., 2001). A large subset of these cluster phylogenetically into a WAGO clade that is roughly equally distant from the conserved PIWI and AGO clades (Yigit et al., 2006). Included in these are CSR-1, the Argonaute that mediates the CSR-1 22G RNA pathway (Claycomb et al., 2009), and related protein C04F12.1. Also variably included is the divergent RDE-1, the first C. elegans Argonaute to be linked to RNAi (Tabara et al., 1999). The other 12 Argonaute proteins in this clade constitute the WAGO Argonautes of the WAGO 22G RNA pathway. These 12 Argonautes bind and stabilize WAGO 22G RNAs semiredundantly to mediate target silencing. Simultaneous loss of all 12 Argonautes in the MAGO12 Argonaute mutant results in penetrant ts sterile, Him, and Rde phenotypes and massive depletion of 22G RNAs that are not dependent on CSR-1 (Claycomb et al., 2009; Gu et al., 2009). The ts sterility of the MAGO12 mutant can only be partially rescued by wild type sperm, likely because loss of 21U RNAs, one of the triggers of 22G RNA biogenesis, results in ts fertility defects in female germline as well as male (Wang and Reinke, 2008). These WAGO-dependent 22G RNAs are also globally depleted in mut-7 and mut-2/rde-3 mutants (Gu et al., 2009), reflecting the contribution of Mutator foci components to WAGO 22G RNA accumulation (Phillips et al., 2012).

Although the phylogenetic relationships between the Argonautes can vary depending on whether full-length protein or specific domains were used for phylogenetic analysis, the 12 WAGO Argonautes of the cognate 22G RNA pathway (Grishok et al., 2001; Yigit et al., 2006; Montgomery et al., 2012) are usually clustered into three subclades referred to by phylogenetic branch: Branch 1 comprises WAGO-1 (R06C7.1), WAGO-2 (F55A12.1), WAGO-3 (PPW-2/Y110A7A.18), WAGO-4 (F58G1.1), WAGO-5 (ZK1248.7); Branch 2 comprises WAGO-6/8 (SAGO-2/F56A6.1), WAGO-7 (PPW-1/C18E3.7), WAGO-8/6 (SAGO-1/K12B6.1); and Branch 3 comprises WAGO-9 (HRDE-1/C16C10.3), WAGO-10 (T22H9.3), WAGO-11 (Y49F6A.1), WAGO-12 (NRDE-3/R04A9.2) (Gu et al., 2009). wago-1, wago-2, wago-3/ppw-2, and wago-5 transcripts are classified as germline-intrinsic (Reinke et al., 2004). wago-4 and wago-9/hrde-1 transcripts are classified as oogenesis-enriched, whereas wago-10 is spermatogenesis-enriched (Reinke et al., 2004). The wago-12/nrde-3 transcript is classified as mixed oogenesis-somatic (Reinke et al., 2004). The others are not assigned germline enrichment classifications (Reinke et al., 2004). WAGO-12/NRDE-3 and WAGO-9/HRDE-1 will be discussed in greater detail below in the context of the nuclear RNAi pathway.

Loss of single WAGO Argonautes can cause mild defects in exo-RNAi sensitivity and germline transposon silencing, such as the germline Rde phenotype observed in a natural C. elegans isolate (CB4856) that lacks functional WAGO-3/PPW-2 (Tijsterman et al., 2002b; Vastenhouw et al., 2003; Yigit et al., 2006). Simultaneous loss of multiple WAGO Argonautes within the same phylogenetic branch produces more significant impairment. The MAGO strain lacks Branch 2 and WAGO-4 Argonautes and shows a somatic and germline Rde phenotype and a ts fertility defect (Yigit et al., 2006). The Quadruple mutant lacks all of the Branch 1 Argonautes but WAGO-1 and shows an Rde phenotype (Gu et al., 2009). Similarly, WAGO Argonautes bind and stabilize WAGO 22G RNAs semiredundantly. In exo-RNAi of the unc-22 transcript, Branch 2 Argonautes bind secondary siRNAs redundantly, and overexpression of any Branch 2 Argonaute rescues the RNAi defect of the MAGO strain, which otherwise fails to accumulate unc-22 secondary siRNAs (Yigit et al., 2006). Still, particular 22G RNAs in germline and soma show different dependencies upon different WAGO Argonautes. The Quadruple mutant, while showing a germline Rde phenotype, still generates wild-type levels of a germline F37D6.3 22G RNA, whereas loss of the remaining Branch 1 Argonaute, WAGO-1, severely depletes this species (Gu et al., 2009). Loss of all five Branch 1 Argonautes in the Quintuple mutant results in dramatic reduction of germline 22G RNAs, whereas a MAGO+2 mutant lacking Branch 2, WAGO-3, and WAGO-4 Argonautes (Gu et al., 2009) shows near wild type levels of germline 22G RNAs, but loss of a somatic Y47H10A.5 22G RNA triggered by RDE-1 (Gu et al., 2009).

Early evidence suggested that exo-RNAi occurred primarily through a post-transcriptional silencing mechanism. Injection of dsRNA segments corresponding to intron and promoter sequences does not elicit interference (Fire et al., 1998), and injection of dsRNA targeting a transcript encoded in an operon generally does not decrease levels of transcripts encoded downstream of the target within the operon appreciably (Montgomery et al., 1998). Little progress has been made in understanding cytoplasmic, post-transcriptional target silencing by WAGO 22G RNAs. It is unlikely to involve Argonaute-mediated target cleavage, as WAGO Argonautes lack the catalytic triad residues required for slicer activity (Yigit et al., 2006); furthermore, target cleavage induced by introduction of 5’ triphosphorylated siRNAs into cell lysate is predominantly mediated by CSR-1 (Aoki et al., 2007), which is not involved in the WAGO 22G RNA pathway (Claycomb et al., 2009; Gu et al., 2009). This suggests that other nucleases may be recruited to the target mRNA to mediate mRNA degradation in the WAGO pathway. It is possible that RDE-10/RDE-11-mediated target deadenylation and degradation represents the main post-transcriptional mechanism for target silencing (Yang et al., 2012). In the germline, post-transcriptional target silencing may occur at P granules, as GFP::WAGO-1 is expressed in the germline and localizes to perinuclear P granules (Gu et al., 2009), like Argonautes ALG-3, PRG-1, and CSR-1 (Batista et al., 2008, Wang and Reinke, 2008; Claycomb et al., 2009; Conine et al., 2010). Earlier data connecting the nonsense-mediated decay (NMD) pathway with exo-RNAi likely resulted from the presence of the mut-16(mg461) mutation in the mutant strains of smg-2, smg-5, and smg-6 genes, which encode NMD pathway proteins (Zhang et al., 2011). This explanation is consistent with an initial report that failed to demonstrate a role for the SMG proteins in exo-RNAi (Montgomery et al., 1998), although a transgenic system for studying transcriptional gene silencing showed impairment of repression upon depletion of smg-5 by RNAi (Grishok et al., 2005).

Like other WAGO Argonautes, NRDE-3 binds WAGO 22G RNAs. Immuno-purified NRDE-3 complexes contain primarily 5’ triphosphorylated ~22 nt endo-siRNAs (Guang et al., 2008). Consistent with production by a cytoplasmic RdRP acting on spliced transcripts, NRDE-3-associated siRNAs map to exonic sequences, even when RNAi is triggered by dsRNA containing introns (Guang et al., 2008). NRDE-3 primarily binds somatic 22G RNAs triggered by ERGO-1 class 26G RNAs, such as E01G4.5 22G RNAs, but also binds secondary siRNAs during exo-RNAi targeting a somatic transcript (Guang et al., 2008; Gent et al., 2010; Guang et al., 2010). Due to redundancy among the WAGO Argonautes, loss of NRDE-3 does not significantly impair stability of these 22G RNAs (Guang et al., 2008).

Upon binding a cytoplasmic WAGO 22G RNA, NRDE-3 translocates to the nucleus to execute target silencing (Guang et al., 2008). Nuclear localization is required for NRDE-3 function: an NLS-defective NRDE-3 mutant protein that binds siRNAs at wild type levels but does not redistribute to the nucleus fails to rescue nrde-3 mutant phenotypes (Guang et al., 2008). siRNA binding is required for translocation to the nucleus, as mutating the NRDE-3 PAZ domain to abrogate siRNA binding also results in failure to localize to the nucleus (Guang et al., 2008). Similarly, loss of siRNA cofactors abolishes nuclear localization. As NRDE-3 binds primarily somatic 22G RNAs triggered by ERGO-1 class 26G RNAs, GFP::NRDE-3 is cytoplasmic in eri-1, ergo-1, mut-7, rde-4, mut-2, eri-9, eri-6, and dcr-1(mg375) helicase mutants (Guang et al., 2008; Pavelec et al., 2009; Fischer et al., 2011). The nuclear localization of GFP::NRDE-3 is rescued upon introduction of dsRNA in an eri-1 mutant, but not an rde-4 mutant, which cannot initiate exo-RNAi (Guang et al., 2008); this suggests that siRNA binding is necessary and sufficient for nuclear redistribution of nuclear Argonautes.

After translocation to the nucleus, the loaded WAGO 22G RNA directs NRDE-3 to associate with unspliced target pre-mRNA (Guang et al., 2008). NRDE-3 targeting recognition appears to require very high sequence complementarity (Montgomery et al., 2012): A reporter transgene with a single target site for an X-cluster 22G RNA is efficiently silenced in a wild type background and shows primarily NRDE-3-dependent silencing. Silencing is abrogated by mismatches at positions 1-3, 4-5, 12-14, or deletion of a nucleotide at position 4. Silencing is impaired by mismatches at positions 9-11 or deletion or insertion of a nucleotide at position 13. Mismatches at position 13 or 20-22 are tolerated.

The proteins NRDE-2, NRDE-1, and NRDE-4 mediate the transcription silencing effects of nuclear Argonautes. After NRDE-3 binds a WAGO 22G RNA, translocates to the nucleus, and recognizes a nascent pre-mRNA target, NRDE-3 associates with NRDE-2 to recruit NRDE-1 to the transcript and direct its deposition on chromatin through a mechanism that requires NRDE-4 (Guang et al., 2008; Guang et al., 2010; Burkhart et al., 2011). NRDE-2, NRDE-1, and NRDE-4 are also engaged by HRDE-1 to mediate nuclear RNAi in the germline (Ashe et al., 2012; Buckley et al., 2012; Luteijn et al., 2012). Accordingly, loss of any of these factors results in the same Mrt phenotype observed with loss of HRDE-1 (Buckley et al., 2012), as well as loss of sensitivity to nuclear exo-RNAi (Guang et al., 2010; Burkhart et al., 2011).

Nuclear RNAi terminates transcription and promotes establishment of a repressed chromatin state at genomic loci targeted by exo-RNAi as well as multiple loci targeted by endo-siRNAs and 21U RNAs (Burkhart et al., 2011; Buckley et al., 2012; Gu, S.G., et al., 2012; Luteijn et al., 2012; Shirayama et al., 2012). Nuclear RNAi requires interaction with target pre-mRNA, as targeting 5’ to the 5’ UTR does not result in silencing (Gu, S.G., et al., 2012). NRDE-3, NRDE-2, and NRDE-1 bind to pre-mRNA at the target site but are not detected in association with pre-mRNA 5’ to the target site, presumably due to co-transcriptional splicing of nascent transcripts (Moore and Proudfoot, 2009; Guang et al., 2010; Burkhart et al., 2011). Nrde targeting inhibits transcription elongation, as Pol II occupancy increases near the target site but not at initiation sites (Guang et al., 2010). Nuclear run-on assays show Pol II inhibition occurring ~2 kb downstream of the target site, correlating with decreased Pol II occupancy by chromatin immunoprecipitation 3’ to the target site (Guang et al., 2010; Burkhart et al., 2011). Because transcription is inhibited, NRDE-3, NRDE-2, and NRDE-1 do not bind pre-mRNA sequences encoded 3’ to the target site; however, Nrde factors appear to remain associated with pre-mRNA fragments after silencing (Guang et al., 2010; Burkhart et al., 2011).

Nrde pathway targeting results in enrichment of H3K9 trimethylation at genomic target sites that correlates with the transcription inhibition (Burkhart et al., 2011), consistent with the established repressive role of H3K9 methylation. Whereas NRDE-3 and NRDE-2 recruit NRDE-1 to the RNA target near or 5’ to the target site, NRDE-1 requires NRDE-4 for chromatin interaction and associates with chromatin predominantly 3’ to the site of RNAi (Burkhart et al., 2011). Accordingly, H3K9 trimethylation occurs across the gene but peaks 3’ to the target site (Burkhart et al., 2011). The chromatin response to nuclear RNAi occurs at the subgenic level, as shifting the RNAi trigger site shifts the interval of H3K9me3 deposition (Gu, S.G., et al., 2012). During exo-RNAi-triggered nuclear RNAi, secondary siRNAs accumulate mainly at exonic regions proximal to the dsRNA trigger site, but both exons and introns show high H3K9me3 levels after targeting (Gu, S.G., et al., 2012).

Although many of the studies of NRDE-3 function in nuclear RNAi were conducted under exo-RNAi, NRDE-3-mediated nuclear RNAi is also important for transcriptional silencing triggered by a subset of endogenous small RNAs in the soma (Burkhart et al., 2011). As mentioned above (see Section 15), NRDE-3 binds somatic WAGO 22G RNAs, many of which are triggered by ERGO-1 class 26G RNAs (Guang et al., 2008; Gent et al., 2010; Guang et al., 2010). 22G RNAs targeting the ERGO-1 target E01G4.5 are abundantly bound by NRDE-3 (Guang et al., 2008) and direct chromatin modification at this locus through the somatic nuclear RNAi pathway (Burkhart et al., 2011). Loss of members of the ERI complex, NRDE-3, or downstream Nrde pathway factors results in a subtle but detectable increase in E01G4.5 transcription inhibition, increased transcript levels, and depletion of H3K9me3 marks at the E01G4.5 locus (Burkhart et al., 2011). The strong influence of NRDE-3 on H3K9me3 deposition at the E01G4.5 locus is uncommon, and levels of many ERGO-1 targets show little or no effect of loss of the somatic Nrde pathway (Burkhart et al., 2011), consistent with significant overlap between cytoplasmic and nuclear WAGO Argonautes in silencing of somatic targets (Gu et al., 2009). Nonetheless, in the absence of ERGO-1 class 26G RNAs, the somatic Nrde pathway can be heavily engaged by exo-RNAi, as indicated by the emergence of RNAi phenotypes for polycistronic transcripts in an eri-1 mutant (Bosher et al., 1999; Kennedy et al., 2004). In fact, exo-RNAi of some monocistronic transcripts also engages NRDE-3 in the absence of ERGO-1 class 26G RNAs, as exo-RNAi targeting unc-73 and dpy-13 results in less severe phenotypes in an eri-1; nrde-3 double mutant than in an eri-1 mutant (Guang et al., 2008). While loss of ERGO-1 class 26G RNAs does sensitize the somatic nuclear RNAi system for phenotypic analysis, an eri-1 or equivalent mutant background is not necessary for the transcription- and chromatin-level effects of nuclear RNAi (Burkhart et al., 2011).

Although NRDE-3 acts primarily in somatic cells, the silencing effects of somatic nuclear RNAi are transmissible for one generation (Burton et al., 2011). Exo-RNAi in the parental generation generates secondary siRNAs abundantly bound by NRDE-3; inheriting progeny also show high levels of NRDE-3-bound secondary siRNAs from embryo through adulthood, although these are not inherited into F2 embryos (Burton et al., 2011). In the absence of NRDE-1 or NRDE-4, these secondary siRNAs are still inherited and associate with NRDE-3, but levels of NRDE-3-bound secondary siRNAs are not maintained throughout development (Burton et al., 2011). This indicates that somatic nuclear RNAi is not required for inheritance of secondary siRNAs but is required for continued expression of these siRNAs in progeny.

H3K9me3 marks established in the parental generation at RNAi target loci are lost in progeny during the embryo stage, possibly due to the absence of NRDE-3 in early embryogenesis (Guang et al., 2008; Burton et al., 2011). Silencing is then reestablished through NRDE-3 and downstream Nrde factors during development and strengthened to exceed the H3K9me3 levels in the parental generation (Burton et al., 2011). The mechanism is likely the same in progeny as in the parental generation as NRDE-3 siRNA binding and nuclear translocation are still required for inherited silencing (Burton et al., 2011). However, the increased H3K9me3 deposition in progeny suggests that the inheritance of a silencing signal may further promote the ability of nuclear RNAi to repress chromatin, possibly by enhancing coupling of siRNA generation and H3K9me3 deposition in progeny. Importantly, this reestablishment of silencing requires the activity of NRDE-3 in progeny: homozygous nrde-3 mutant progeny of heterozygous nrde-3 hermaphrodites subjected to exo-RNAi do not show inheritance of silencing (Burton et al., 2011). The converse has not been tested; it is not known whether somatic nuclear RNAi is required in the parental generation for RNAi inheritance and chromatin repression in progeny. Furthermore, it is as yet unclear whether somatic nuclear RNAi is required for establishment or maintenance of transcriptional silencing during RNAi inheritance in the soma.

Somatic nuclear RNAi in the parental generation directs siRNA expression and H3K9 trimethylation in inheriting progeny through one of two possible mechanisms. NRDE-3-bound secondary siRNAs may be directly deposited into germ cells to direct embryonic silencing in association with other WAGO Argonautes, then re-bind NRDE-3 to promote maintenance of siRNA expression through nuclear RNAi during development. Alternatively, a different silencing signal such as primary small RNAs may be deposited in germ cells to promote expression of siRNAs in progeny that bind NRDE-3 and re-engage nuclear RNAi (Burton et al., 2011). In support of the latter mechanism, NRDE-3-bound WAGO 22G RNAs do not appear to initiate their own amplification at target sites (Montgomery et al., 2012), suggesting that primary small RNA trigger may be required for secondary siRNA re-amplification in progeny. One likely possibility is that primary exo-siRNAs are directly inherited from hermaphrodites exposed to dsRNA. Indeed, rde-1 transcript is abundant in hermaphrodite germline and detected at high levels in embryos of all stages (NEXTDB clone 26b3), suggesting that primary exo-siRNAs may be inherited bound to RDE-1 to trigger NRDE-3-mediated nuclear RNAi in inheriting progeny. Such a mechanism would explain why somatic nuclear RNAi is heritable for only a single generation.

Like NRDE-3, HRDE-1 engages the Nrde downstream machinery to direct transcriptional silencing and H3K9me3 deposition at loci targeted by exo-RNAi and endogenous small RNAs; however, HRDE-1 binds germline WAGO 22G RNAs and engages the downstream Nrde pathway to direct transgenerational silencing of germline transcripts (Ashe et al., 2012; Bagijn et al., 2012; Buckley et al., 2012; Gu, S.G., et al., 2012; Luteijn et al., 2012; Shirayama et al., 2012). Loss of HRDE-1, NRDE-2, NRDE-1, or NRDE-4 results in the Mrt phenotype described above (Buckley et al., 2012), likely due to desilencing of germline targets such as transposons that pose a threat to genome integrity when active. Further testing of mutants for various upstream and downstream 21U RNA, 26G RNA, and WAGO 22G RNA pathway factors is necessary to determine the respective contributions of these pathways to promoting germline immortality. Curiously, the Mrt phenotype of hrde-1 and nrde-2 mutants manifests only at 25°C, whereas nrde-1 and nrde-4 mutants are Mrt at 25°C and 20°C (Buckley et al., 2012), suggesting the chromatin machinery of the Nrde pathway may have additional roles in germline maintenance.

Multigenerational inheritance of transcriptional silencing requires germline expression of the targeted transcript (Grishok et al., 2005). Most likely, an mRNA template must be present in the germline for RdRP to generate siRNAs capable of propagating silencing across each generation; this may account for the limited perdurance of NRDE-3-mediated somatic nuclear RNAi (Burton et al., 2011). In contrast, HRDE-1-mediated nuclear RNAi in the germline results in continued siRNA production and silencing over multiple generations that maintain H3K9 trimethylation (Ashe et al., 2012; Buckley et al., 2012; Gu, S.G., et al., 2012; Luteijn et al., 2012; Shirayama et al., 2012). Accordingly, factors such as MUT-7 and RDE-3 that are required for 22G RNA accumulation are also required for maintenance of transgenerational silencing (Luteijn et al., 2012; Shirayama et al., 2012). Screens using transgenerationally silenced exo-RNAi and 21U RNA sensor transgenes have identified a set of chromatin factors necessary for maintenance of transgenerational RNAi: the chromo domain protein HPL-2 (semiredundantly with HPL-1), Polycomb complex protein MES-3, Trithorax complex protein MES-4, SET domain SET-32, and putative H3K9 methyltransferase SET-25 (Ashe et al., 2012; Shirayama et al., 2012). These proteins may be recruited by NRDE-1 to mediate H3K9 methylation or otherwise promote a chromatin state favorable for continued siRNA biogenesis to maintain nuclear RNAi.

In the parental generation exposed to dsRNA, HRDE-1 is dispensable for germline silencing; however, HRDE-1 is required in the F1 generation for inheritance of parental germline silencing, in the F2 generation for inheritance from the F1, and so on, as indicated by a failure of homozygous hrde-1 mutant offspring to inherit germline transgene silencing from a hrde-1 heterozygous hermaphrodite (Buckley et al., 2012). This result differs somewhat from another study, which shows that loss of HRDE-1 or MUT-7 reactivates a silenced locus only in the second generation of hrde-1 homozygosity (Luteijn et al., 2012), suggesting that epigenetic silencing may be established in the parental germline.

Surprisingly, the germline nuclear RNAi pathway appears to require cytoplasmic WAGO Argonautes for initiation: during exo-RNAi targeting the oogenesis-enriched smg-1 transcript (Reinke et al., 2004), the target site fails to become enriched for H3K9me3 marks in the MAGO mutant strain (Gu, S.G., et al., 2012). This definitively indicates that the nuclear and cytoplasmic WAGO Argonautes do not compete for siRNA cofactors. Rather, cytoplasmic WAGO Argonautes promote nuclear RNAi through an unknown mechanism. It is not yet established whether WAGO Argonautes bind 22G RNAs irreversibly or reversibly; if the latter, perhaps HRDE-1 appropriates WAGO 22G RNAs from the cytoplasmic Argonautes for nuclear silencing. NRDE-3 does not appear to be required for establishment or maintenance of transgenerational silencing (Buckley et al., 2012; Luteijn et al., 2012), nor is a sensitizing mutation in endo-RNAi factors such as ERI-1 or RRF-3 required (Gu, S.G., et al., 2012). Other WAGO Argonautes do, however, contribute variably to HRDE-1 target silencing, as some 21U RNA sensor transgenes show dependence on additional Argonautes for complete silencing (Bagijn et al., 2012; Shirayama et al., 2012).

While establishment of transgenerational silencing requires factors involved in the formation of the initial trigger, such as RDE-1 in the case of primary exo-siRNAs (Buckley et al., 2012; Gu, S.G., et al., 2012) or PRG-1 in the case of 21U RNAs (Ashe et al., 2012; Bagijn et al., 2012; Lee et al., 2012; Luteijn et al., 2012; Shirayama et al., 2012), maintenance of transgenerational silencing does not (Ashe et al., 2012; Buckley et al., 2012; Lee et al., 2012; Luteijn et al., 2012; Shirayama et al., 2012). Once established, HRDE-1-mediated nuclear silencing is stable for multiple generations and can be inherited from either the male or female germline as a dominant signal, although silencing may show a generational lag when inherited from the male (Ashe et al., 2012; Luteijn et al., 2012; Shirayama et al., 2012). This silencing in trans is mediated by a diffusible agent, namely, secondary 22G RNAs, capable of repressing homologous DNA regardless of whether the primary small RNA target site is present (Luteijn et al., 2012). These 22G RNAs map further upstream of the initiating 21U RNA target site than observed during early, PRG-1-dependent silencing (Bagijn et al., 2012), suggesting that transgenerational RNAi may entail additional spreading during maintenance (Luteijn et al., 2012). 22G RNAs whose expression is stimulated at transgenerationally silenced 21U RNA targets to maintain silencing may not correspond precisely to the species triggered by 21U RNA targeting, as the latter appear to be PRG-1-dependent and are lost when transgenerationally silenced loci are introduced into a prg-1 mutant background (Ashe et al., 2012).

After exposure to dsRNA, secondary siRNA accumulation and association with HRDE-1 decrease progressively across generations (Buckley et al., 2012). Transgenerational silencing initiated by endogenous small RNAs shows a similar effect, as endogenous targets of HRDE-1 silencing show progressive loss of H3K9me3 enrichment and increased expression over successive generations in a hrde-1 mutant background (Buckley et al., 2012). This suggests that endogenous primary small RNA triggers may be required for periodic reinforcement of established transgenerational silencing. It is intriguing to speculate what function the nuclear RNAi pathways serve. While nuclear RNAi may simply provide a heritable signal to direct silencing of aberrant and potentially dangerous transcripts, it has also been suggested that targeting of these endogenous loci by nuclear RNAi may also fulfill a larger function in regulating chromatin dynamics.

The 22G RNAs that are not bound by WAGO Argonautes are the CSR-1 22G RNAs (Figure 4). Named for their sole Argonaute effector, CSR-1, these 22G RNAs do not induce target silencing but rather associate with chromatin to promote proper organization or assembly of targets within the holocentric chromosomes of C. elegans (Claycomb et al., 2009). CSR-1 22G RNAs are produced in the germline by the same DRH-3- and EKL-1-containing RdRP module as WAGO 22G RNAs, but require EGO-1 and not RRF-1 for accumulation (Claycomb et al., 2009; Gu et al., 2009). Although CSR-1 22G RNAs share the features and biogenesis machinery of the secondary siRNAs engaged by WAGO Argonautes, there is no evidence that CSR-1 22G RNAs are themselves secondary siRNAs.

CSR-1 22G RNAs largely target germline-expressed genes. Immuno-purified CSR-1 complexes are enriched for a set of 22G RNAs that are antisense to over 4,000 protein-coding genes, whereas microRNAs, 21U RNAs, and 22G RNAs targeting repetitive sequences, pseudogenes, and intergenic or unannotated loci are depleted (Claycomb et al., 2009). Similarly, comparison of deep sequencing libraries generated from csr-1 and ego-1 mutants (Claycomb et al., 2009) with drh-3 and ekl-1 mutants (Gu et al., 2009) reveals that CSR-1 and EGO-1 are required for accumulation of 22G RNAs antisense to protein-coding genes, many of which also require DRH-3 and EKL-1 (Claycomb et al., 2009). Most CSR-1 22G RNAs are depleted in the absence of germline (Gu et al., 2009), consistent with the germline-restricted expression pattern of EGO-1 (Smardon et al., 2000; Maine et al., 2005; Vought et al., 2005; Claycomb et al., 2009). Less than 1% of 22G RNAs recovered in immuno-purified CSR-1 complexes target repetitive elements and pseudogenes (Claycomb et al., 2009).

The 22G RNA RdRP module is discussed in detail above; the following sections focus on the features and functions of EGO-1, DRH-3, and EKL-1 only as they pertain to the CSR-1 22G RNA pathway. In situ hybridization indicates that concentrations of ego-1, drh-3, ekl-1, and csr-1 transcripts are highest in gonad and early embryo (She et al., 2009), in line with other transcript and protein localization data (Smardon et al., 2000; Reinke et al., 2004; Maine et al., 2005; Vought et al., 2005; Nakamura et al., 2007; Claycomb et al., 2009). DRH-3 and EKL-1 are not germline-restricted, as they also function in WAGO 22G RNA biogenesis in germline and soma (Claycomb et al., 2009; Gu et al., 2009).

EGO-1, DRH-3, EKL-1, and CSR-1 represent the main components of the CSR-1 22G RNA pathway (Claycomb et al., 2009). Earlier phenotypic analyses and focused or genome-wide RNAi screens have identified an overlapping set of phenotypes associated with depletion of these proteins that include embryonic lethality, larval arrest, and sterility (Qiao et al., 1995; Fraser et al., 2000; Smardon et al., 2000; Maeda et al., 2001; Zipperlen et al., 2001; Kamath et al., 2003; Simmer et al., 2003; Rual et al., 2004; Fernandez et al., 2005; Sonnichsen et al, 2005; Duchaine et al., 2006; Lehner et al., 2006; Yigit et al., 2006). These four factors were first implicated in a common pathway by their shared enhancement of ksr-1 lethality (Ekl) phenotype, which describes a failure to specify excretory duct cell fate in late embryogenesis (Rocheleau et al., 2008). Transcripts whose loss produces an Ekl phenotype are expressed in maternal germline to regulate embryonic gene expression (Rocheleau et al., 2008).

Although CSR-1 shows cleavage activity in cell lysate (Aoki et al., 2007), CSR-1 does not appear to silence CSR-1 22G RNA targets: these targets are expressed in germline, oocyte, and embryo and do not show changes in expression upon loss of CSR-1 or DRH-3 (Claycomb et al., 2009; Gu et al., 2009). Rather, EGO-1, DRH-3, EKL-1, and CSR-1 localize to chromosomes to promote proper organization and alignment of metaphase chromosomes and proper orientation of kinetochores to opposing spindle poles (Claycomb et al., 2009). Loss of these factors disrupts chromosome condensation and cohesion: depletion of any by RNAi results in highly disorganized condensin and cohesin protein loading onto mitotic chromosomes in embryos (Claycomb et al., 2009). Loss of these factors also disrupts kinetochore assembly. Whereas inner centromeric proteins HCP-3/CENP-A and HCP-4/CENP-C should be poleward-localized on both sides of metaphase plate, embryos depleted of CSR-1 22G RNA pathway factors show dramatically disorganized loading of these proteins on metaphase chromosomes; outer kinetochore protein KLP-7/MCAK and spindle checkpoint protein BUB-1 are also misloaded onto mitotic chromosomes (Claycomb et al., 2009). Thus, the CSR-1 22G RNA pathway directly influences chromosome organization to promote proper segregation. The X chromosome is depleted of genes expressed in the germline (Reinke et al., 2004) and therefore of CSR-1 22G RNA targets. Nevertheless, the X chromosome usually segregates correctly in mitosis, suggesting that other mechanisms may govern X chromosome segregation (Claycomb et al., 2009).

The CSR-1 22G RNA pathway also regulates H3K9me2 distribution. H3K9me2 becomes enriched on unpaired chromosomes and high-copy extrachromosomal arrays during meiosis (Kelly et al., 2002; Bean et al., 2004). Loss of EGO-1, DRH-3, EKL-1, or CSR-1 results in reduced H3K9me2 accumulation on unpaired chromosomes such as the male X chromosome and ectopic H3K9me2 accumulation on paired and synapsed chromosomes (Maine et al., 2005; She et al., 2009). Whereas wild-type male pachytene nuclei show a single focus of H3K9me2 labeling, ego-1, drh-3, ekl-1, or csr-1 mutant nuclei with normal pachytene morphology lack a single strong focus of H3K9me2 labeling, and some show multiple bright H3K9me2 foci with higher overall levels of H3K9me2 (Maine et al., 2005; She et al., 2009). Mutant nuclei, however, may be large, morphologically abnormal, and polyploid, with diffuse chromosome morphology and multiple H3K9me2 foci and high overall H3K9me2 (She et al., 2009). This demonstrates a role for CSR-1 22G RNAs in specifying chromatin modification. However, as CSR-1 22G RNAs target protein-coding genes expressed in germline (Claycomb et al., 2009; Gu et al., 2009), it is likely that CSR-1 and associated 22G RNAs are excluding repressive H3K9me2 marks from target genomic loci rather than depositing them.

Another function of CSR-1 22G RNAs may be to define areas of germline transcription. This is consistent with a proposed role for CSR-1 22G RNAs in acting as a self-recognition system in opposition to 21U RNA-directed silencing of DNA perceived as non-self; targeting of transcripts by CSR-1 may therefore represent a licensing process that helps to maintain expression of bona fide germline transcripts (Ashe et al., 2012; Shirayama et al., 2012). As CSR-1 22G RNAs are generated in germline through the activity of RdRP EGO-1, CSR-1 22G RNA targets are robustly expressed in maternal germline and may therefore engage CSR-1 as nascent transcripts during gametogenesis; this may define CSR-1 chromatin domains that could perdure through embryogenesis independent of transcription (Claycomb et al., 2009). Such a mechanism would account for the prominent localization of CSR-1 to chromatin in mature oocytes (Claycomb et al., 2009). Presumably, CSR-1 establishes euchromatic genomic regions and defines the boundaries of adjacent kinetochores, potentially by mediating deposition of activating chromatin marks and HCP-3/CENP-A incorporation, respectively. In embryos, CENP-A is loaded at low density onto about half of the genome, inverse to regions transcribed in germline and early embryo, and defines centromeric regions (Gassmann et al., 2012). Pre-existing CENP-A nucleosomes are not necessary to guide recruitment of new CENP-A nucleosomes, as CENP-A is not transmitted by sperm during fertilization and is unloaded and reloaded during oogenic meiotic prophase (Gassmann et al., 2012). Germline transcription of a genomic region may exclude CENP-A incorporation in progeny, with reinforcement by zygotic transcription in early embryogenesis. Domains targeted by the CSR-1 22G RNA pathway and regions enriched for the centromeric histone variant CENP-A appear to be mutually exclusive (Claycomb et al., 2009); therefore, the CSR-1 22G RNA pathway may serve to transmit patterns of germline transcription to early embryo. Alternatively, H3K36 methylation may fulfill this role, as its pattern also correlates inversely with CENP-A occupancy (Rechtsteiner et al., 2010; Gassmann et al., 2012).

Finally, the CSR pathway also appears to play a critical role in maturation of replication-dependent core histone mRNAs. Processing of replication-dependent core histone pre-mRNAs requires endonucleolytic cleavage in the 3’UTR just downstream of the stem-loop that in many organisms is directed by the U7 snRNA (Marzluff et al., 2008). Nematodes, however, lack the U7 snRNA (Dávila López and Samuelsson, 2008), and CSR-1 22G RNAs may fulfill this role in its stead (Avgousti et al., 2012): a majority of histone genes (66%) are depleted by loss of CSR-1. Immuno-purified CSR-1 complexes are enriched for 22G RNAs targeting all core histone genes, and EGO-1-dependent 22G RNAs map to histone pre-mRNAs at the 3’ end sites of mature histone mRNAs (Avgousti et al., 2012). CSR-1 binds these 22G RNAs to target histone pre-mRNAs and either directly cleaves them or facilitates their cleavage. Loss of CSR-1 22G RNA pathway factors results in the accumulation of unprocessed histone mRNAs, with a corresponding decrease in histone production that could explain the phenotypes of these mutants (Avgousti et al., 2012). For example, histone biogenesis defects in cultured cells result in S-phase extension and cell cycle delay (Wagner et al., 2005; Ideue et al., 2009) reminiscent of the extended transition zone observed in ego-1 mutant germlines (Smardon et al., 2000). Additionally, RNAi-mediated knockdown of a single core histone gene causes chromosome segregation defects and sterility similar to the phenotypes of CSR-1 22G RNA pathway mutants (Kodama et al., 2002; Sonnichsen et al, 2005). Consistent with this explanation, overexpression of transgenic histone pre-mRNAs that do not require cleavage rescues lethality associated with csr-1 or ego-1 depletion by RNAi (Avgousti et al., 2012). Phenotypes similar to those shown by CSR-1 22G RNA pathway mutants also result upon depletion of the stem-loop-binding protein (SLBP), CDL-1. SLBP is required for histone mRNA processing, stabilization, and efficient translation (Marzluff et al., 2008). In the germline, CDL-1::GFP shows nuclear enrichment in developing oocytes, much like CSR-1 (Claycomb et al., 2009; Avgousti et al., 2012). RNAi-mediated knockdown of cdl-1 results in accumulation of unprocessed histone mRNA, severely depletes histone levels, and causes adult sterility and chromosome segregation defects in early embryo (Kodama et al., 2002; Pettitt et al., 2002; Avgousti et al., 2012). In contrast, RNAi-mediated knockdown of condensin component smc-4 does not result in histone processing and accumulation defects, showing that this is not a general feature associated with compromised chromosome segregation (Avgousti et al., 2012). These results suggest that CSR-1 and CDL-1 may collaborate to promote histone maturation in oocytes, ensuring that stores of histones will be sufficient to sustain embryonic division until zygotic transcription begins. It is interesting to note that the sterility associated with compromise of the CSR-1 22G RNA pathway may be independent of the P granule defects, as loss of cdl-1 does not disrupt P granules (Avgousti et al., 2012). Finally, intestinal cell endoreduplication also requires high levels of histone proteins (see WormBook chapter The C. elegans intestine ). Possibly related to somatic expression observed for the longer csr-1 isoform (Claycomb et al., 2009), CSR-1 protein is detected in adult intestinal nuclei (Avgousti et al., 2012). These cells show depletion of histone H2B upon RNAi knockdown of csr-1, suggesting that CSR-1 function in histone processing may not be restricted to the germline (Avgousti et al., 2012).

The CSR-1 22G RNA pathway requires the conserved β nucleotidyltransferase CDE-1 for uridylation of 22G RNAs to restrict their accumulation (van Wolfswinkel et al., 2009). The cde-1 transcript (Reinke et al., 2004) is detected by in situ hybridization throughout the entire gonad but not in other tissues (van Wolfswinkel et al., 2009). In the male or hermaphrodite spermatogenic germline, CDE-1 localizes to bright granules in close proximity to the condensing DNA of mature sperm. In the oogenic germline, CDE-1 localization is cytoplasmic with perinuclear enrichment. Within the embryonic P lineage, CDE-1 localizes to perinuclear granules, colocalizing mainly with PGL-1; in mitotic cells, CDE-1 shows EGO-1-dependent localization to the outer edges of condensing chromosomes at prometaphase and the poleward sides of chromosomes at metaphase, with expression detectable through anaphase. CDE-1 immuno-purifies with EGO-1 in embryo extract, but CSR-1 is not required for EGO-1 accumulation. Loss of EGO-1 results in dispersed localization of CDE-1 over the whole metaphase plate and a CDE-1 halo surrounding the DNA in condensing sperm. Depletion of CSR-1 results in very faint but correct CDE-1 localization to the metaphase plate, but no association of CDE-1 granules with DNA in sperm. Loss of EGO-1 or CSR-1 does not affect CDE-1 localization to P-granules, and EGO-1 and CSR-1 show normal localization in the absence of CDE-1. CDE-1 protein is not detected in intestine (van Wolfswinkel et al., 2009).

CDE-1 is a catalytically active nucleotidyltransferase. Immuno-purified CDE-1 complexes preferentially catalyze the addition of uridines to small RNA 3’ ends in vitro (van Wolfswinkel et al., 2009), consistent with earlier findings expressing CDE-1 in Xenopus oocytes (Kwak and Wickens, 2007). Uridine tails added in vitro are short, suggesting low processivity, and can be blocked by 2’-O-methylation of the substrate small RNA 3’ end (van Wolfswinkel et al., 2009). In vivo, CDE-1 uridylates a portion of CSR-1 22G RNAs. Endo-siRNAs, but not microRNAs, show addition of one or more untemplated 3’ uridines that are lost in the absence of CDE-1 in adults; however, some endo-siRNAs in young adult still show untemplated uridines in a cde-1 mutant, suggesting redundant terminal uridylation activity at earlier stages (van Wolfswinkel et al., 2009). It is not clear whether CDE-1 uridylates 22G RNAs already bound by CSR-1; however, no association between the two proteins is detected in embryo extract (van Wolfswinkel et al., 2009). Uridylation by CDE-1 appears to destabilize CSR-1 22G RNAs, as those frequently found uridylated in wild type are increased upon loss of CDE-1, whereas levels of microRNAs, 21U RNAs, and other antisense small RNAs are not (van Wolfswinkel et al., 2009). Accordingly, immuno-purification of CSR-1 complexes from a cde-1 mutant recovers many more siRNAs, although the distribution of targets does not change (van Wolfswinkel et al., 2009). This suggests that CDE-1 acts to reduce the half-life of CSR-1 22G RNAs, ensuring that CSR-1 is only partially loaded, and, indeed, CSR-1 22G RNAs are rarer by far than WAGO 22G RNAs (Claycomb et al., 2009; Gu et al., 2009). Their modest accumulation may also explain why CSR-1 22G RNA targets are not silenced, as only the more abundantly WAGO 22G RNA-targeted loci are appreciably silenced (Claycomb et al., 2009; Gu et al., 2009). CDE-1-mediated uridylation and partial loading of CSR-1 may serve to restrict CSR-1 22G RNAs to the CSR-1 22G RNA pathway, as loss of CDE-1 results in erroneous gene silencing likely due to misloading of CSR-1 22G RNAs onto WAGO 22G RNA Argonautes (van Wolfswinkel et al., 2009). This may explain the phenotypes of cde-1 mutants, which show defects in cosuppression, increased Tc1 transposon activity, and partial insensitivity to RNAi in the germline (Robert et al., 2005; van Wolfswinkel et al., 2009). In the absence of CDE-1-mediated uridylation, excess CSR-1 22G RNAs compete with WAGO 22G RNAs for Argonatute binding, compromising exo-RNAi in the germline. Similarly, Tc1-targeting siRNAs are decreased, possibly due to decreased stabilization by available WAGO Argonautes, resulting in Tc1 activation and increased dsDNA breaks and apoptosis (van Wolfswinkel et al., 2009).

Loss of CDE-1 leads to meiotic and embryonic mitotic segregation defects with a resulting Him phenotype, presumably due to erroneous targeting of CSR-1 22G RNA targets by WAGO Argonautes (van Wolfswinkel et al., 2009). Germline nuclei do not show mitotic missegregation, but diakinetic oocytes often show univalents, suggesting impaired pairing of homologous chromosomes during meiosis (van Wolfswinkel et al., 2009). cde-1 mutant embryos show metaphase plate disorganization similar to that observed in csr-1 and ego-1 mutant embryos, with abnormal CENP-A deposition, and occasional severe spindle defects and polar body retention (Smardon et al., 2000; Yigit et al., 2006; Claycomb et al., 2009; van Wolfswinkel et al., 2009). In analysis of cde-1 mutant hermaphrodite x wild-type male hybrid embryos, both parental chromosomes are affected by major segregation defects, indicating that missegregation is not limited to the parental germline and persists into embryonic development (van Wolfswinkel et al., 2009). Consistent with the collaborative but antagonistic functions of CDE-1 and CSR-1 22G RNA pathway factors, the embryonic lethality of cde-1 mutants is less severe than that of csr-1 and ego-1 mutants (van Wolfswinkel et al., 2009). CDE-1 also regulates processing of pre-let-7 through uridylation (Heo et al., 2008) and may act in additional roles in other small RNA pathways.

The following pertains to canonical or “Type 1” 21U RNAs. “Type 2” 21U RNAs are associated with transcription start sites (TSSs) of protein-coding mRNAs and other Pol II transcripts and are discussed below.

Early deep sequencing of C. elegans small RNAs identified a highly diverse population of 21 nt small RNAs with a 5’ uridine bias and no other common sequence features (Ruby et al., 2006). These 21U RNAs are 5’ monophosphorylated (Ruby et al., 2006) and terminally methylated by HENN-1 methyltransferase (Ruby et al., 2006; Billi et al., 2012; Kamminga et al., 2012; Montgomery et al., 2012). Unlike endo-siRNAs, 21U RNAs are not generated from mRNA templates by RdRPs but rather transcribed directly from genomic loci primarily clustered in two gene-depleted regions of chromosome IV (Ruby et al., 2006; Gu, W., et al., 2012). 21U RNA loci are vastly depleted for overlap with exons. Those that do overlap show no sense or antisense bias (Ruby et al., 2006). 21U RNA loci do not show a prominent strand bias and overlap less than expected by chance (Ruby et al., 2006), consistent with their independent transcription from short precursors, (Gu, W., et al., 2012; Billi et al, 2013). Although the larger genomic clusters of 21U RNA loci show synteny across rhabditids, the 21U RNAs themselves show no sequence conservation (Ruby et al., 2006; de Wit et al., 2009). This lack of conservation suggests that evolutionary pressure has maximized the diversity of 21U RNA sequences, in line with their purported role as a genetic immune system capable of selectively targeting non-self transcripts (Ashe et al., 2012; Lee et al., 2012; Shirayama et al., 2012). Consistent with their free evolution, the sequence content of a 21U RNA appears to be irrelevant for expression, with the exception of the 5’ uridine (Billi et al, 2013).

21U RNAs are robustly expressed in the germline (Batista et al., 2008; Das et al., 2008) and detected in early embryo, but their levels decline across development (Stoeckius et al., 2009). A majority of 21U RNAs show differential expression in male and female germlines (Billi et al, 2013; Shi et al., 2013). These male and female germline-enriched subpopulations of 21U RNAs direct the production of nonoverlapping 22G RNAs that target distinct transcripts. Targets of male germline-enriched 21U RNAs are depleted for transcripts enriched in spermatogenesis but not oogenesis (Reinke et al., 2004), suggesting selection against evolution of male-expressed 21U RNAs that target male germline genes (Billi et al, 2013; Shi et al., 2013). Female germline-enriched 21U RNAs do not show depletion of spermatogenesis-enriched transcripts and are paradoxically enriched for oogenesis-enriched transcripts. While the significance of this is uncertain, the distinct enrichment patterns suggest that male and female germline-enriched 21U RNAs are subject to different evolutionary pressures (Billi et al, 2013; Shi et al., 2013). Female germline-enriched 21U RNAs are preferentially detected in embryos (Billi et al, 2013; Shi et al., 2013), similar to the pattern observed for ERGO-1 class 26G RNAs (Han et al., 2009).

21U RNAs target transcripts by directing Argonaute PRG-1 to imperfectly complementary sites with up to three, or possibly even four, mismatches (Bagijn et al., 2012; Lee et al., 2012). 21U RNA targets are depleted of protein-coding transcripts but not pseudogene or transposon transcripts (Bagijn et al., 2012). As 21U RNA targeting triggers local production of WAGO 22G RNAs (Das et al., 2008; Lee et al., 2012), 21U RNA targets are enriched for WAGO 22G RNA targets and depleted for CSR-1 22G RNA targets (Lee et al., 2012). As 21U RNAs are capable of triggering transgenerational silencing that persists for tens of generations in the absence of PRG-1 and additional 21U RNA trigger (Ashe et al., 2012; Luteijn et al., 2012), identification of target transcripts is challenging. Targets can be approximated as those transcripts that show robust 21U RNA-dependent 22G RNA accumulation or those that show upregulation upon loss of PRG-1, although the latter are more likely to report a specific subset of targets whose silencing is more heavily weighted toward cytoplasmic than nuclear RNAi. Possibly, the rapid evolution of 21U RNA sequences and the profoundly stable silencing exerted by 21U RNAs renders definition of a specific target set irrelevant, as many previously silenced targets have likely been eroded through genetic drift. So may be the eventual fate of sequences not protected from 21U RNA targeting by a self-recognition system such as CSR-1 22G RNAs are proposed to represent (Ashe et al., 2012; Shirayama et al., 2012).

21U RNAs are not transcribed by RdRP complexes, nor does their biogenesis appear to require DCR-1 or any of the factors directly involved in endo-siRNA biogenesis (Batista et al., 2008; Das et al., 2008). Rather, 21U RNAs are encoded as independent Pol II transcriptional units (Gu, W., et al., 2012; Billi et al, 2013) (Figure 5). In C. elegans and other rhabditids, 21U RNA loci show a common upstream sequence arrangement of two motifs, the large and small motifs, that together drive the regulated expression of 21U RNAs in the germline (Ruby et al., 2006; Cecere et al., 2012; Gu, W., et al., 2012; Billi et al, 2013). In C. elegans and C. briggsae, 21U RNA loci distribution shows a pattern of enrichment at the start and end of full-length DNA transposon loci in sense and antisense orientations, respectively, suggesting that 21U RNA upstream motifs may serve as traps for triggering 21U RNA biogenesis against recent integrations (Bagijn et al., 2012).

In association with an upstream large motif or at a TSS, the YR pair of the YRNT small motif appears sufficient to direct csRNA expression, whereas the thymidine is required for 21U RNA accumulation. The accumulation of csRNAs, but not mature 21U RNAs, associated with YNRV small motifs suggests that the 5’ uridine of 21U RNAs is dispensable for transcription but plays a role in mature 21U RNA stabilization or processing (Gu, W., et al., 2012). In support of this, mutation of the thymidine to G or A abrogates accumulation of a cognate 21G or 21A species (Billi et al, 2013). Furthermore, comparison of other C. elegans isolates with YRNT motifs where the N2 isolate shows YRNV reveals that substitution for a genomic thymidine results in accumulation the associated 21U RNA (Gu, W., et al., 2012; Billi et al, 2013). The presence of a uridine at position 3 of the csRNA may be required for removal of the m⁷GpppAmNm cap-2 structure to generate the mature 21U RNA 5’ end; however, recombinant PRG-1 shows preferential target cleavage in association with 5’ uridine ssRNA (Bagijn et al., 2012), suggesting PRG-1 may purify candidate small RNAs processed from csRNAs by selectively binding and stabilizing those with 5’ uridines. Identification of decapping machinery responsible for csRNA 5’ end processing will address this question definitively. Trimming of the csRNA 3’ end to achieve the mature 21 nt length may occur after PRG-1 binding. While the vast majority of small RNA deep sequencing reads corresponding to 21U RNA loci are 21 nt in length, a small portion show extensions, with a significant bias toward 3’ extensions (Kato et al., 2009). These 3’ extended reads may represent biogenesis intermediates and support the hypothesis that 3’ trimming occurs after 5’ processing. In silkworm, PIWI Argonaute-bound small RNAs are trimmed after Argonaute loading by a 3’ to 5’ exonuclease until the length is sufficiently short for anchoring of the 3’ end in the Argonaute (Kawaoka et al., 2011). While the 26 nt length of 26G RNAs is determined by DCR-1 processing, both 22G RNAs and 21U RNAs are DCR-1-independent (Aoki et al., 2007; Batista et al., 2008; Das et al., 2008; Pavelec et al., 2009). Perhaps a common exonuclease trims 22G RNAs and 21U RNAs.

Mapping of csRNAs reveals that a majority are not associated with upstream core CTGTTTCA sequences; rather, many of these map bidirectionally to TSSs of protein-coding and other Pol II transcripts (Gu, W., et al., 2012). Like chromosome IV csRNAs associated with large motifs, these TSS-associated csRNAs are also processed into mature 21U RNAs when encoded by YRNT motifs (Gu, W., et al., 2012). The abundance of csRNAs at a particular promoter correlates with the abundance of longer capped reads originating from that promoter, suggesting that these csRNAs are products of Pol II initiation at active promoters and may reflect promoter-proximal pausing (Gu, W., et al., 2012). The significance of these TSS-associated 21U RNAs is as yet unknown. Much as Pol II initiation at canonical TSSs produces both csRNAs and longer pre-mRNA transcripts, a small fraction of 21U RNA loci show both csRNAs and corresponding longer reads (Cecere et al., 2012; Gu, W., et al., 2012). This suggests that promoter elements and 21U RNA upstream motifs may both promote Pol II association and transcription initiation, but 21U RNA loci generally lack sequence elements or chromatin configuration necessary to promote further transcript elongation. csRNA distribution does reveal differences in 21U RNA and canonical promoters, as bidirectional transcription is not reported to occur at 21U RNA loci. 21U RNAs are required for full fertility (Yigit et al., 2006; Batista et al., 2008; Das et al., 2008; Wang and Reinke, 2008), so it seems possible that csRNAs mapping to canonical promoters are incidental byproducts of selection for Pol II activity that produces 21U RNA precursors. However, promoter-associated csRNAs are observed in other metazoan genomes (Rasmussen and Lis, 1995; Nechaev et al., 2010), suggesting rather that 21U RNA production may originally have arisen incidental to this more ancient process of unknown significance.

As a model organism, C. elegans rose to fame for its utility in the study of cell differentiation and organ development, yet this nematode has proven one of the richest models for study of small RNAs. The earlier challenges—developing methods to capture small RNAs of unknown properties, distinguishing the unique subclasses, identifying the factors involved in their biogenesis and effector mechanisms—have been met due to the ingenuity of pioneers in the field. Advances in sequencing technology have allowed us to catalogue the small RNAs of C. elegans to greater depth than ever anticipated, and new approaches continue to reveal additional complexity, such as the recently identified TSS-associated 21U RNAs found through the novel CapSeq method (Gu, W., et al., 2012). Genetic screens have also been invaluable in defining the protein factors that act in these pathways. Early screens identified suites of candidate small RNA pathway factors using complex RNAi-related phenotypes as readouts for incompletely understood underlying mechanisms. The roles of these candidates are still being elucidated, for example, chromatin factors previously implicated in transgene or transposon silencing (Kelly and Fire, 1998; Vastenhouw et al., 2003; Grishok et al., 2005; Kim et al., 2005) have only just been positioned downstream of the nuclear RNAi pathway and transgenerational silencing (Ashe et al., 2012; Shirayama et al., 2012).

In this young field of small RNAs, large, fundamental questions at every step remain unanswered. In biogenesis: How are transcripts selected for targeting by 26G RNAs? Currently, there is only speculation that secondary structure may somehow contribute to recruitment of biogenesis factors. In amplification: How does primary small RNA recognition trigger local biogenesis of secondary siRNAs? The key may lie in Mutator foci or other such physical cellular substructures that serve as amplification factories. In silencing: What are the chief post-transcriptional mechanisms by which siRNAs repress target mRNAs? The cleavage activity of many C. elegans Argonautes appears dispensable or non-conserved, contradicting the RNAi canon of Argonaute-mediated target cleavage. And particularly in broader function: What are the molecular mechanisms by which CSR-2 22G RNAs promote chromosome segregation? Why does loss of male 26G RNAs lead to defects in spermatogenesis? How does loss of the nuclear RNAi pathway compromise germline immortality? The final questions that bear upon us are those of conservation. Mammalian piRNAs show entirely unique biogenesis mechanisms, although downstream aspects of piRNA biology are shared with the 21U RNAs. This does not exclude the possibility that mammals express small RNAs from tiny transcriptional units like 21U RNA loci; discovery of small RNA classes and mechanisms in mammals is not yet at saturation. Although many of the pathways found in C. elegans are unlikely to be conserved in higher animals lacking the greatly expanded Argonaute family and its worm-specific clade, analogous or partially overlapping pathways may function in the germlines of other organisms. microRNAs, once regarded as a peculiarity of C. elegans genetics, have emerged as master regulators of most developmental processes. Endo-siRNAs and piRNAs appear to occupy a similar position of prominence in the C. elegans germline and embryo. It remains to be seen how this role translates in higher organisms.