Multigenerational Regulation of the Caenorhabditis elegans Chromatin Landscape by Germline Small RNAs

Abstract

In animals, small noncoding RNAs that are expressed in the germline and transmitted to progeny control gene expression to promote fertility. Germline-expressed small RNAs, including endogenous small interfering RNAs (endo-siRNAs) and Piwi-interacting RNAs (piRNAs), drive the repression of deleterious transcripts such as transposons, repetitive elements, and pseudogenes. Recent studies have highlighted an important role for small RNAs in transgenerational epigenetic inheritance via regulation of heritable chromatin marks; therefore, small RNAs are thought to convey an epigenetic memory of genomic self and nonself elements. Small RNA pathways are highly conserved in metazoans and have been best described for the model organism Caenorbabditis elegans. In this review, we describe the biogenesis, regulation, and function of C. elegans endo-siRNAs and piRNAs, along with recent insights into how these distinct pathways are integrated to collectively regulate germline gene expression, transgenerational epigenetic inheritance, and ultimately, animal fertility.

INTRODUCTION

Animal development requires a carefully orchestrated program of gene expression and gene silencing. In the developing germline, the silencing of deleterious elements such as transposons and repetitive elements is vital for transmitting an intact genome to the next generation. Accordingly, there must be biological mechanisms that distinguish deleterious genetic elements from germline-expressed endogenous genes and pass on a memory of that distinction to the next generation. In the model organism Caenorhabditis elegans, collaboration between small RNAs and chromatin modifiers allows for the transgenerational epigenetic inheritance of germline gene silencing and licensing (or activating).

Owing to its rapid development and genetic tractability, C. elegans is an ideal system for the study of small RNA biology and transgenerational epigenetic inheritance. Both transgene silencing and RNA interference (RNAi) are transgenerational processes (3, 9, 18, 61, 65, 90, 112). Histone modification patterns, which are established in parental germlines, are also inherited in progeny (8, 39, 84) and can transmit a memory of parental germline gene expression (100).

How are gene silencing and gene licensing balanced during germline development, and how is that balance successfully achieved in each generation? In the first part of this review, we summarize how small RNAs control germline gene expression via silencing and licensing of distinct populations of germline messenger RNA (mRNA) transcripts. In the second part of this review, we discuss how small RNAs modulate the chromatin landscape to facilitate appropriate gene expression and germline development.

SMALL RNAs

In the years since the discovery of RNAi, the robust silencing of homologous genes caused by double-stranded RNAs (dsRNAs) (36), small noncoding RNAs have emerged as major determinants of gene expression in C. elegans and other animals. Exogenous RNAi (exo-RNAi) has become a widely used tool for research, allowing for targeted-gene knockdown by delivery of exogenous dsRNA and subsequent processing into small interfering RNAs (siRNAs). The proteins that execute the biogenesis and silencing functions of siRNAs also regulate gene expression during organismal development through the generation and effector function of endogenous small noncoding RNAs. Deep sequencing of small RNAs from C. elegans and subsequent studies have uncovered numerous conserved classes of small RNAs regulating diverse processes during development, including microRNAs (miRNAs), endogenous siRNAs (endo-siRNAs), and 21U Piwi-interacting RNAs (piRNAs). Although miRNAs primarily regulate somatic development, endo-siRNAs and piRNAs are expressed in the C. elegans germline and are required for fertility.

In this section, we discuss the biogenesis and regulation of endo-siRNAs and piRNAs (Table 1, Table 2). All small RNAs function in association with an Argonaute protein. The Argonaute protein is the defining component of the RNA-induced silencing complex (RISC). The Argonaute-bound small RNAs guide RISC to target transcripts by base complementarity. The Argonaute proteins all share two characteristic RNA-binding domains: PAZ and PIWI. The PIWI domain is RNase H related, and some Argonautes have retained a catalytic triad, the DDH motif, which enables the cleavage, or slicing, of target RNAs. By convention, small RNAs and their associated RISCs are classified according to their associated Argonaute protein. Following this convention, we discuss two distinct classes of endo-siRNAs: the WAGO class and the CSR-1 class. Both WAGO-class and CSR-1-class endo-siRNAs have a characteristic length of 22 nucleotides (nt) and a strong bias for guanosine at the 5′ position (22G RNAs) (4, 82, 126). The WAGO-class 22G RNAs are the primary effectors of both exo-RNAi and the 26G endo-siRNAs that are expressed in oogenic germline and embryos and bind to the Argonaute ERGO–1 (53, 111, 126), which we do not discuss in detail here (but see Reference 14 for a review).

Table 1.

Summary of key protein factors in small RNA pathways in Caenorhabditis elegans

Factors	WAGO-CLASS 22G RNAs			CSR-1-CLASS 22G RNAs	21U RNAs (piRNAs)
Biogenesis factors	RRF-1, EGO-1, DRH-3, ELK-1			EGO-1, DRH-3, ELK-1	SNPC-4/GEI-11, PRDE-1, TOFU-3, TOFU-4, TOFU-5
	In Mutator focus: MUT-2, MUT-7, MUT-8, MUT-14, MUT-15, MUT-16, SMUT-1, RDE-8
	Outside Mutator focus: RDE-10, RDE-11, RDE-12, RSD-2, RSD-6
Processing factors	ND	ND	ND	CDE-1	HENN-1, PARN-1, TOFU-1, TOFU-2, PID-1
Argonautes	WAGO-1, WAGO-2, WAGO-3, WAGO-4, WAGO-5, WAGO-10, WAGO-11, SAGO-1, SAGO-2, WAGO-7/PPW-1	Germline nuclear: WAGO-9/HRDE-1	Somatic nuclear: WAGO-12/NRDE-3	CSR-1	PRG-1
Other RISC components	ND	NRDE-1, NRDE-2, NRDE-4	NRDE-1, NRDE-2, NRDE-4	ND	ND
Downstream effectors	ND	H3K9me3 HMTs: SET-25, SET-32, MET-2	ND	ND	WAGO 22G RNA pathway
		H3K9me3 maintenance: MORC-1

Abbreviations: endo-siRNA, endogenous small interfering RNA; HMT, histone methyltransferase; ND, not determined; piRNA, Piwi-interacting RNA; RISC, RNA-induced silencing complex.

Table 2.

Genes involved in small RNA pathways in Caenorhabditis elegans

Gene abbreviation	Gene class full name	Relevance
csr-1	Chromosome-segregation and RNAi deficient	Argonaute protein associated with CSR-l-class 22G RNAs
cde-1	Germline co-suppression defective	Catalyzes 3′ uridylation of CSR-l-class 22G RNAs
dcr-1	Dicer related	Ribonuclease involved in processing dsRNA
drh-3	Dicer related helicase	Functions in 22G RNA biogenesis as a component of the RdRP module
ego-1	Enhancer of Glp-one	RdRP functioning in CSR-l-class and WAGO-class 22G RNA biogenesis
elk-1	ELK transcription factor homolog	Functions in 22G RNA biogenesis as a component of the RdRP module
emb-4	Abnormal embryogenesis	Required for nuclear RNAi against intron-containing transcripts
glb-1	Germline helicase	Required for RNAi inheritance
henn-1	HEN1 of nematode	Catalyzes 2′-O-methylation of piRNAs
heri-1	Heritable enhancer of RNAi	Limits duration of RNAi inheritance
hrde-1	Heritable RNAi deficient	Argonaute protein mediating RNAi inheritance
hrde-2, hrde-4	Heritable RNAi deficient	Required for RNAi inheritance
met-1	Histone methyltransferase-like	H3K36 HMT, suppressor of hrde-1 and morc-1
met-2	Histone methyltransferase-like	H3K9 HMT functioning downstream of RNAi
mog-2	Masculinization of germline	Required for RNAi inheritance
morc-1	Mouse microrchidia family CW-type zinc finger protein	Required for RNAi inheritance and chromatin compaction
mut-2, mut-7, mut-8, mut-14, mut-15, mut-16	Mutator	Required for WAGO-class 22G RNA biogenesis
nrde-1, nrde-2, nrde-4	Nuclear RNAi defective	Required for germline and somatic nuclear RNAi
nrde-3	Nuclear RNAi defective	Argonaute protein mediating somatic nuclear RNAi
parn-1	Poly(A)-specific ribonuclease homolog	Required for 3’ trimming of piRNAs
pid-1	piRNA induced silencing defective	Required for accumulation of mature piRNAs
prde-1	piRNA-dependent silencing defective	Required for piRNA biogenesis
prg-1	Piwi related gene	Argonaute protein associated with piRNAs
rde-8, rde-10, rde-11, rde-12	RNAi defective	Required for accumulation of WAGO-class 22G RNAs
rnp-3	RRM RNA binding domain containing	Required for RNAi inheritance
rrf-1	RNA-dependent RNA polymerase family	RdRP functioning in WAGO-class 22G RNA biogenesis
rsd-2, rsd-6	RNAi spreading defective	Required for accumulation of WAGO-class 22G RNAs
set-25, set-32	SET domain containing	H3K9 HMTs functioning downstream of RNAi
smut-1	Synthetic mutator	Required for accumulation of WAGO-class 22G RNAs
snpc-4	Small nuclear RNA activating complex homolog	Required for piRNA biogenesis
tofu-1, tofu-2	Twenty One U-RNA biogenesis Fouled Up	Required for piRNA biogenesis
tofu-3, tofu-4, tofu-5	Twenty One U-RNA biogenesis Fouled Up	Required for accumulation of mature piRNAs
wago-1, wago-2, wago-3, wago-4, wago-5, wago-7, wago-10, wago-11	Worm Argonaute protein	Argonaute proteins associated with WAGO-class 22G RNAs
znfx-1	ZNFX homolog	Required for RNAi inheritance

Abbreviations: dsRNA, double-stranded RNA; HMT, histone methyltransferase; piRNA, Piwi-interacting RNA; RdRP, RNA-dependent RNA polymerase; RNAi, RNA interference.

WAGO-Class 22G RNAs

The WAGO-class 22G RNAs bind a partially redundant set of 12 worm-specific Argonaute proteins (WAGOs) to target protein-coding genes, pseudogenes, transposons, and other repetitive elements (9, 18, 49, 51, 72, 90, 92). The WAGO-class 22G RNAs are enriched in the germline, including mature gametes, and in embryos (27, 53). Biogenesis of the WAGO-class 22G RNAs requires triggering by primary siRNAs (26G RNAs or exo-RNAi) (76, 91, 93, 111) or by piRNAs (63); thus, the WAGO-class 22G RNAs are considered secondary siRNAs. Accordingly, depletion of 26G RNAs or piRNAs, such as by deletions of the genes encoding biogenesis machinery or Argonaute, also causes depletion of the secondary 22G RNAs (11, 29, 63, 111). The binding of the primary siRNAs or piRNAs, in association with RISC, to their mRNA targets triggers 22G RNA biogenesis, utilizing the targeted mRNA as the biogenesis template (11, 29, 63, 76, 91, 93, 111). As such, 22G RNAs facilitate the amplification of 26G RNA- and piRNA-directed gene silencing. Similarly, WAGO-class 22G RNAs are triggered by exo-RNAi to execute the majority of target repression (91, 93, 126).

Amplification of the WAGO-class 22G RNAs.

As with other endo-siRNAs, the WAGO-class 22G RNAs are made by an RNA-dependent RNA polymerase (RdRP) module composed of an RdRP, a DExD/H-box helicase protein, and a Tudor domain protein. In this case, the module includes the partially redundant RdRP RRF-1 or EGO-1, the DExD/H-box helicase DRH-3, and the Tudor domain protein ELK-1 (49). Although exo-RNAi and 26G RNAs are processed by Dicer and are accordingly marked by 5′ monophosphate, the 22G RNAs contain a 5′ triphosphate, suggesting that they are not processed by the single C. elegans Dicer homolog DCR-1 (4, 40, 76, 82). Furthermore, in vitro experiments have shown that 22G RNA accumulation is DCR-1 independent (6).

The recent characterization of the 22G RNA pathway has led to the identification of a novel subcellular structure, the Mutator focus, where many 22G biogenesis factors are concentrated (80, 81, 129). The Mutator foci serve as amplification centers for WAGO-class 22G RNAs downstream of 26G RNAs, piRNAs, and exo-RNAi. Many of the components of these structures were originally identified in genetic screens for mutants with high rates of germline mutations [Mutator (Mut)] (26, 62, 113). The Mutator phenotype arises from increased transposition of DNA transposons such as Tc1 and Tc3 elements (26, 62, 113). Transposon silencing is also piRNA dependent (12, 99); thus, the increased transposition in mutator mutants highlights the essential role of the WAGO-class 22G RNAs in piRNA-mediated transposon silencing (29, 49, 63). Loss of the Mutator focus components is also associated with resistance to RNAi [RNAi defective (Rde)], indicating their involvement in the amplification of secondary 22G RNAs triggered by exo-RNAi (62, 99).

Mutator foci are thought to serve as centers for surveillance of mRNAs as they are exported from the nucleus (81). The components of the Mutator foci include RRF-1, MUT-2, MUT-7, MUT-8, MUT-14, MUT-15, MUT-16, SMUT-1, and RDE-8 (44, 80, 81, 105). Of all the Mutator focus components, only MUT-16 appears to be required for focus formation (80, 81). MUT-16 contains an intrinsically disordered region, which is required for focus formation and for RNAi (108). In diverse biological systems and cellular contexts, there is emerging evidence for the contribution of intrinsically disordered regions to liquid-liquid phase separation of RNA-protein assemblies (reviewed in 69) and Mutator foci likely represent another instance of phase-separated RNA-protein granules (108).

The specific functions of the other Mutator focus components are largely undescribed, but recent evidence suggests that some components might contribute to cleavage and uridylation of mRNA targets to facilitate RdRP recruitment (101, 105). Target uridylation precedes 22G RNA synthesis and depends on the endoribonuclease RDE-8 (105). Biochemical studies of RDE-8 have focused primarily on its role in exo-RNAi, but based on the WAGO-class 22G RNA depletion in rde-8(−/−) mutants, it may function similarly in endo-siRNA biogenesis (105). Taken together, these studies suggest that RDE-8 is recruited to an mRNA target by primary siRNAs (both primary exo-siRNAs and ERGO–1-class 26G RNAs) in association with RISC (105). The interaction between RDE-8 and the mRNA target may be stabilized by other components of Mutator foci, including MUT-15 (105). RDE-8 then cleaves the mRNA target, and the liberated 3′ end is uridylated, possibly by the β-nucleotidyltransferase MUT-2/RDE-3, which then recruits RRF-1 (105).

The accumulation of some WAGO-class 22G RNAs also depends on several factors that are not components of Mutator foci. RDE-10, which has no annotated domains, and RDE-11, a RING-type zinc finger, form a complex and may be involved in stabilizing the mRNA target to facilitate RRF-1 binding and transcription of the secondary22G RNAs (125, 128). During exo-RNAi, RISC recruits RDE-10 to the target mRNA. RDE-11 then contributes to deadenylation and degradation of the RDE-10-bound targets (125, 128). The binding of RDE-10 to target mRNA depends on the DEAD-box helicase RDE-12 (124). RSD-2 and RSD-6, which are associated with RDE-12, are also required for accumulation of some 22G RNAs (83, 103, 124).

Worm-specific Argonautes.

Once the 22G RNAs are synthesized in the Mutator focus, they are loaded onto one of the 12 WAGOs. Worms with mutations in all 12 wagos (so-called mago12 mutants) are temperature-sensitive (ts) sterile and RNAi defective; in single mutants, these phenotypes are absent or not fully penetrant, suggesting that the WAGOs are partially redundant (27, 49, 104, 126). Genetic screens for factors involved specifically in siRNA-mediated silencing in the nucleus have identified two specialized WAGOs, WAGO-12/NRDE-3 (nuclear RNAi defective phenotype) and WAGO-9/HRDE-1 (heritable RNAi defective phenotype), which are discussed below (18, 51).

Multigenerational silencing by RNAi.

RNAi targeting of transcripts expressed in the maternal germline causes multigenerational gene silencing. For example, gfp RNAi of worms expressing a GFP-tagged histone 2B reporter (GFP::H2B) from a transgene controlled by a germline-specific promoter (pie-1) results in highly penetrant silencing of gfp in F₁ and F₂ worms that have never been exposed to exogenous gfp RNAi (9, 18). High-throughput sequencing studies have shown that in the parental P₀ generation grown on gfp RNAi, the siRNAs targeting gfp are 22 nt long with no 5′-nucleotide bias, indicating that they are DCR-1-dependent primary siRNAs (9). After four generations with no exposure to exogenous gfp RNAi, only DCR-1-independent 22G RNAs remain at the gfp::h2b target gene. In addition, the secondary siRNAs have spread upstream and downstream of the original siRNA trigger, indicating that exogenous gfp RNAi in the P₀ generation triggers the amplification of secondary 22G RNAs in subsequent generations (9). Furthermore, the ability of 22G RNAs to both spread from the initial trigger and silence in trans supports additional direct evidence that secondary 22G RNAs, acting downstream of piRNAs, themselves can serve as triggers for additional rounds of amplification and spreading, i.e., the generation of tertiary siRNAs (85). Notably, studies investigating primary and secondary siRNA responses to other exogenous dsRNA triggers have shown that the secondary 22G RNAs do not trigger the amplification of tertiary siRNAs, suggesting that the ability to generate tertiary siRNAs may be context specific (77). Taken together, these findings suggest that at each generation, maternally inherited siRNAs are amplified in the progeny and loaded into the zygote, thus transmitting a multigenerational memory of siRNA-mediated gene silencing.

Multigenerational silencing is a dynamic process that can be passed from the maternal or paternal germline (3). The duration of silencing after RNAi can be increased by exposure to a secondary RNAi trigger, against a different target gene, in inheriting generations (54). Additionally, some RNAi pathway genes that affect the duration of silencing are targeted by endo-siRNAs, suggesting a degree of feedback between siRNA biogenesis and target silencing (54).

The core machinery that transmits this memory is RISC defined by the Argonaute WAGO-9/HRDE-1 (9, 18, 47, 90). The non-Argonaute RISC components NRDE-1, NRDE-2, and NRDE-4 are also required for multigenerational siRNA-mediated silencing (18). The assembled nuclear RISC mediates target silencing in two ways. First, RISC is guided by bound siRNAs to nascent precursor (pre-)mRNA, where it induces the stalling of RNA polymerase (Pol) II (50). Second, RISC recruits histone-modifying enzymes to trimethylate histone 3 at lysine 9 (H3K9) and histone 3 at lysine 27 (H3K27) residues at the corresponding targeted genomic loci; H3K9 and H3K27 methylation is the hallmark of heterochromatin (19, 20, 47, 50, 66). The siRNA-induced H3K9me3 marks can extend far beyond the footprint of the siRNA itself, potentially up to 9 kb from the target sequence (47). HRDE-1 is expressed exclusively in the germline (18) but contributes some degree of inherited silencing in somatic tissue, which is likely established in the maternal germline (68).

The cytoplasmic Argonaute WAGO-4, whose localization partially overlaps with P granules, also contributes to inherited silencing by RNAi (116, 123). WAGO-4 expression is specific to the germline in adult worms and persists specifically in germline precursor cells in the developing embryo (116, 123). One likely function of WAGO-4 is the loading of germline-expressed 22G RNAs into the embryo and the stabilization of those 22G RNAs in germline precursor cells in the embryo to allow for triggering of additional rounds of siRNA amplification in the inheriting generations (123). Although WAGO-4, like HRDE-1, is not expressed in somatic tissue, WAGO-4, but not HRDE-1, is required for the single-generation inheritance of RNAi in somatic tissue (123). One explanation for this finding is that WAGO-4 acts upstream of WAGO-12/NRDE-3 (hereafter called NRDE-3), a specialized WAGO Argonaute that is expressed in somatic nuclei and also required for single-generation inheritance of somatic RNAi (20, 123). Surprisingly, sequencing of WAGO-4-bound 22G RNAs indicates that WAGO-4 may share mRNA targets with the CSR-1 pathway, suggesting that antagonism between these two pathways may control the expression of a shared subset of germline-expressed transcripts (123).

21U piRNAs

The C. elegans piRNAs comprise a large class of germline-enriched, bidirectionally transcribed, 5′-monophosphorylated small RNAs with a characteristic length of 21 nt and a bias for a 5′ uridine (12, 82). More than 15,000 unique piRNAs are transcribed as autonomous units from broad regions on chromosome IV (15, 23, 82). The piRNA promoters are generally AT rich and nucleosome depleted (23, 82). In metazoans, the expression of piRNAs in the germline is a common strategy for transposon repression (7, 12, 17, 29, 42, 45, 46, 126). Transposon repression is critical for germline function; therefore, C. elegans depleted of piRNAs has reduced brood sizes compared with wild type (12, 29, 94, 117). Although transposon repression is a common function of metazoan piRNAs, the mechanism for repression in C. elegans is unique: While piRNAs are required to initiate target silencing, the primary effectors of piRNA-directed silencing are the downstream secondary 22G RNAs (9, 11, 63, 65, 90).

Biogenesis and processing of piRNAs.

Although the piRNA sequences themselves are not conserved in other nematodes, the clusters are syntenic with the 21U RNA-encoding clusters in other Caenorhabditis species (32, 82). Also conserved is the enrichment of an 8-nt core motif: CTGTTTCA, known as the Ruby motif (12, 82). This promoter motif, separated from the transcribed piRNA by an AT-rich spacer of 20–60 nt, is required to drive piRNA expression (15, 23, 82). The conservation of the Ruby motif in other Caenorhabditis species suggests that the regulatory mechanisms that control the piRNA pathway are likely conserved, although the piRNA sequences themselves are not. The precise role of the Ruby motif has not yet been described, but it has been proposed to be a transcription factor binding site (23). The Ruby motif also contributes to sex-specific expression of piRNAs, as the 5′ cytosine of the motif is enriched upstream of piRNAs that are expressed during spermatogenesis, whereas piRNAs expressed during oogenesis have no nucleotide bias at that position (15). Although many piRNAs are expressed sex specifically, enriched in either spermatogenic or oogenic germline, no differences in target selection or effector function between the two groups have been identified.

The piRNAs are transcribed by RNA Pol II as a 23–30-nt precursor species during the fourth larval stage and into adulthood, when germline expansion and maturation are most active (15, 23, 48, 119). piRNA transcription requires the transcription factor SNPC-4/GEI-11 (hereafter called SNPC-4), which binds throughout the piRNA clusters (43, 59). snpc-4(−/−) mutants are depleted for piRNAs and sterile (59). As SNPC-4 binding strength does not correlate with piRNA expression level, SNPC-4 is not thought to act as a canonical transcription factor at these loci (59). SNPC-4 colocalizes with PRDE-1, a casein kinase I (CKI) family member that is expressed exclusively in the germline, where it localizes specifically to chromosome IV (59, 119). PRDE-1 is required for accumulation of piRNA precursors, indicating that it acts upstream of the piRNA Argonaute PRG-1 (119). Also required for piRNA precursor accumulation are three uncharacterized factors identified in a screen for piRNA biogenesis factors: Twenty One U-RNA biogenesis Fouled Up 3, 4, and 5 (TOFU-3, TOFU-4, and TOFU-5) (43). TOFU-4 and TOFU-5 were recently shown to act in a complex with PRDE-1 and SNPC-4 at piRNA promoters (121).

Many recently identified factors are involved in piRNA processing. The piRNAs are 2′-O-methylated by HENN-1, which contributes to their stability and facilitates piRNA maintenance in the developing embryo (13, 58, 70). The exonuclease PARN-1 is required for 3′ trimming of piRNAs (102). In parn-1(−/−) mutants, the untrimmed piRNAs are stably expressed, 2′-O-methylated, and bound by PRG-1, indicating that 3′ trimming is not required for piRNA stability, methylation, or Argonaute loading (102). As parn-1(−/−) mutants have decreased fertility compared with wild type and are defective for piRNA-dependent gene silencing, 3′ trimming is likely required for piRNA effector function, perhaps by recruiting additional components of the silencing machinery (102). The accumulation of mature piRNAs is also dependent on several uncharacterized factors: PID-1, TOFU-1, and TOFU-2 (30, 43).

piRNA target selection.

Unlike siRNAs, piRNA targeting to mRNAs is tolerant of sequence mismatches (63); therefore, target mRNAs cannot be identified on the basis of sequence alone. A recent study used a method of cross-linking, ligation, and sequencing of piRNA-target hybrids to identify piRNA targets in vivo and found that piRNAs target nearly all germline-expressed transcripts (89). This genome-wide study and a separate, targeted analysis using CRISPR-Cas9 demonstrate not only that piRNA binding is tolerant of mismatches at certain positions but also that perfect complementarity in the seed region (nucleotides 2–8) is critical for robust targeting (89, 130).

piRNAs establish multigenerational transgene silencing.

Studies have described an extremely stable, completely penetrant form of transgene silencing termed RNA-induced epigenetic silencing (RNAe) (65, 90). Once initiated, RNAe is thought to be permanently maintained over generations, as spontaneous reversion has never been observed (65, 90). RNAe requires PRG-1 and is thus considered piRNA mediated (65, 90). Interestingly, once a transgene has been silenced by RNAe, prg-1 is dispensable for both gene silencing and expression of piRNA-triggered secondary 22G RNAs (65, 90). Maintenance of silencing requires multiple factors involved in the WAGO-class 22G RNAs, including MUT-7, RDE-3, and HRDE-1, but not the Argonaute that functions in exo-RNAi, RDE-1, suggesting that silencing is mediated by the endogenous WAGO-class 22G RNAs (65, 90). Both pre-mRNA and mRNA of the silenced transcript are downregulated during RNAe, indicating that this repression is both cotranscriptional and posttranscriptional (65, 90). RNAe also has a chromatin-based regulatory component, as RNAe induces hrde-1- and nrde-1-dependent H3K9me3 deposition at the target and requires the H3K9me3 reader HPL-2, one of the C. elegans homologs of mammalian heterochromatin protein 1 (HP1) (65, 90).

When RNAe is inherited through the maternal germline, it can act in trans to suppress active transgenes, indicating that the secondary 22G RNAs are a heritable silencing trigger from the maternal germline that is able to silence a paternally expressed transgene (65). Although deep sequencing of these secondary 22G RNAs shows that they spread upstream of the original piRNA trigger, they are limited to parts of the transgene that encode foreign sequences such as gfp and do not target the transgene-borne endogenous sequences (65, 90). Furthermore, RNAe-mediated cosuppression of other sequences in trans is limited to other foreign sequences and does not occur at endogenous genes, suggesting that endogenous sequences are likely protected from RNAe targeting (90). Targeting by the CSR-1 22G RNA pathway, as discussed below, has been proposed as a mechanism for gene licensing or protection from piRNA-mediated silencing (86).

CSR-1-Class 22G RNAs

The CSR-1 22G RNA pathway is a fascinating worm-specific pathway that regulates germline gene expression by both silencing and activating, or licensing, germline transcripts. How CSR-1 is able to coordinate these apparently contradictory actions is unclear. In this section, we discuss in turn the evidence for licensing and silencing. Although CSR-1 is also a worm-specific Argonaute protein, it functions differently from the other WAGOs and therefore defines its own class of 22G RNAs. One key difference between CSR-1 and WAGO pathways is that the CSR-1-class 22G RNAs are not secondary siRNAs; therefore, their biogenesis is not dependent on the 26G RNAs or piRNAs. Many of the biogenesis factors for CSR-1-class 22G RNAs have been identified, but how they are recruited to a specific set of mRNA templates is unknown. The existing evidence suggests that CSR-1 has several important roles, including modulating germline gene expression through gene licensing and silencing and regulating chromatin organization to promote correct chromosome segregation in the early embryo (10, 22, 25, 41, 86, 110, 118).

Biogenesis of CSR-1-class 22G RNAs.

CSR-1-class 22G RNAs are produced in the germline by an RdRP module composed of the RdRP EGO-1, the Tudor domain-containing protein EKL-1, and the DEAH/D-box helicase DRH-3 (25). These 22G RNAs target germline-expressed genes, a feature conserved in the CSR-1 pathway of the sister species Caenorhabditis briggsae (106).

Although CSR-1 and the WAGO Argonautes bind distinct 22G RNAs, how the 22G RNAs are sorted to the appropriate Argonaute is a significant question in the field. One possibility is that posttranscriptional modifications of the 22G RNAs could direct their loading to a specific RISC. One feature that may help distinguish CSR-1-class 22G RNAs from WAGO-class 22G RNAs is the presence of at least one nontemplated uridine at the 3′ end of many CSR-1-class 22G RNAs (25). Importantly, 60% of CSR-1-bound 22G RNAs are not uridylated; therefore, 3′ uridylation is dispensable for direct small RNA loading onto CSR-1 (25, 110). The β-nucleotidyltransferase CDE-1 catalyzes 3′ uridylation of CSR-1-class 22G RNAs (110). cde-1(−/−) mutants express increased levels of endo-siRNAs compared with wild type, suggesting that 3′ uridylation may function to destabilize 22G RNAs, perhaps to maintain the relatively low abundance of CSR-1-class 22G RNAs compared with the WAGO-class 22G RNAs (25, 49, 110). In fact, endo-siRNAs upregulated in the cde-1(−/−) mutant correspond with downregulation of their target genes (110). Most evidence suggests that CSR-1-class 22G RNAs do not have a substantial role in target silencing (25); therefore, the target downregulation in cde-1(−/−) mutants may indicate that the stabilized CSR-1-class 22G RNAs are misrouted to the WAGO pathway to mediate target silencing (110). Alternatively, it is possible that the increased abundance of 22G RNAs in cde-1(−/−) mutants drives CSR-1-mediated target silencing rather than licensing. Recent evidence suggests that CSR-1 mediates the silencing of genes targeted by abundant 22G RNAs (41); hence, if the increased abundance of the nonuridylated 22G RNAs in cde-1(−/−) mutants is sufficient to drive CSR-1-mediated silencing, the target downregulation could be explained.

Nearly all the major components of the CSR-1 pathway, CSR-1, DRH-3, EGO-1, and CDE-1, localize to perinuclear P granules (25, 41, 110). Although the Tudor domain-containing protein EKL-1 has not been detected in P granules, it is essential for P granule integrity, as are EGO-1, DRH-3, and CSR-1 (25, 109, 115). In mature gametes and embryos, CSR-1 localizes to the nucleus (25). Biochemical studies have shown that CSR-1 binds to nascent pre-mRNAs, chromatin, and RNA Pol II (22, 25, 118) and can be immunopurified from purified sperm and oocyte chromatin (24). However, a study utilizing a single-copy CSR-1 transgene detected its localization at P granules but no association with mitotic chromosomes in embryos (41). Further work is needed to clarify the role of CSR-1 in the nucleus. It is possible that CSR-1 associates with chromatin transiently or at low abundance, making it difficult to detect through imaging.

The csr-1(−/−) mutant is Rde and sterile; the few embryos that are produced by csr-1(−/−) mutants are inviable owing to chromosome segregation defects (126). Mutation or knockdown of the other CSR-1 pathway components (e.g., drh-3, elk-1, and cde-1) leads to similar phenotypes, although in cde-1(−/−) mutants the defects are less severe (25, 110). The details of these chromosome segregation defects are discussed in the section titled CSR-1-Dependent Chromatin.

CSR-1 tunes germline gene expression via target activation and silencing.

Initial studies of the CSR-1 pathway found that its targets were largely unaffected or slightly downregulated in csr-1(−/−) mutants compared with wild type (25). For example, most genes encoding histones are downregulated in the csr-1(−/−) mutant (10). Additionally, global nuclear run-on followed by high-throughput sequencing studies, which measure nascent RNA transcripts, have demonstrated reduced transcription of CSR-1 target genes in both csr-1 partial loss-of-function, or hypomorphic, mutants and drh-3(−/−) mutants (22). As RNA Pol II occupancy at WAGO targets is unaffected in the drh-3(−/−) mutant, this effect is thought to be specific to the CSR-1-class 22G RNAs (22). Taken together, these findings suggest that CSR-1 acts cotranscriptionally to positively regulate its targets.

Further study of the CSR-1 pathway in the germlines of male worms supports the model of CSR-1 positively regulating its targets. By means of immunostaining, csr-1(−/−) mutant males show decreased levels of RNA Pol II and H3K4me2, a hallmark of euchromatin, in the germline (88). Studies of hermaphrodites have shown that csr-1 RNAi causes upregulation of spermatogenesis genes, suggesting that in the hermaphrodite, CSR-1 contributes to gene silencing (21). This could indicate that CSR-1 positively regulates targets in males but negatively regulates them in hermaphrodites. Alternatively, the upregulation of spermatogenesis genes upon RNAi knockdown of csr-1 could be an artifact of disrupted P granules, as P granule disruption is a consequence of the loss of the CSR−1 pathway (21). P granule disruption is also associated with upregulation of spermatogenesis genes independent of the CSR-1 pathway (21).

Gene licensing by CSR-1 may have an important role in protecting germline transcripts from silencing by piRNAs. Some expressed transgenes are capable of mediating small RNA-induced gene activation (RNAa) to desilence sequences in trans (90). For example, in wild-type worms, some transgenes encoding a GFP-tagged endogenous protein are expressed, a gfp(on) allele, whereas others are silenced, a gfp(off) allele. RNAa describes the phenomenon in which these two alleles are crossed into the same worm and the presence of the gfp(on) allele drives the desilencing, or activation, of the gfp(off) allele. Whereas the sequence features that allow this trans-activation are unknown, it is dependent on the CSR-1 22G RNA pathway, as indicated by the identification of 22G RNAs corresponding to the activating allele [gfp(on) per the example] and the finding that RNAa does not occur in the presence of csr-1 RNAi (86). After multiple generations of exposure to the gfp(on) allele, the activation of the gfp(off) allele can be transmitted to subsequent generations (86). Unlike RNAe, the permanent silencing of some transgenes that is established by piRNAs and maintained by the WAGO-class 22G RNAs, RNAa is heritable but not permanent because trans-activated gfp(off) alleles always revert back to the silenced state eventually, even after multiple generations of RNAa-mediated activation (86). As with de novo establishment of RNAe, this reestablishment requires prg-1 (86). These data suggest that CSR-1 acts in opposition to piRNAs to promote expression of its targets.

If RNAa is in fact driven by the CSR-1 pathway, the directed targeting of CSR-1 to a silenced transcript should be sufficient to desilence, or activate, the allele. The recent use of the phage λ-BoxB system to tether CSR-1 to target transcripts has shown that CSR-1 tethering can protect a transcript from silencing by RNAe and that a silenced allele can be activated after multiple generations of CSR-1 tethering (118). Taken together, these studies suggest that licensing by CSR-1 protects transcripts from silencing by piRNAs to promote the expression of germline genes (118). In support of this model, global analysis of piRNA-target interactions using an auxin-inducible degradation system has shown that in CSR-1-depleted worms, the number of unique piRNA-binding sites increases, suggesting that CSR-1 and piRNAs compete for germline targets (89). Both CSR-1 and PRG-1 function primarily at germline P granules. As such, these may be the sites of mRNA surveillance and sorting into endogenous, or self, transcripts that are CSR-1 bound and thus licensed for germline expression. In contrast, foreign, or nonself, transcripts would be targeted by piRNAs and shuttled to adjacent Mutator foci to template 22G RNA amplification and trigger gene silencing. Notably, other gene-licensing mechanisms might exist that act in parallel to CSR-1, as some germline-expressed targets that are intrinsically resistant to targeting by a synthetic piRNA can evade piRNA-mediated silencing even in the presence of csr-1 RNAi knockdown (130).

The existing evidence suggests that CSR-1 can directly repress gene expression by two distinct mechanisms. First, CSR-1 has an intact catalytic triad and is capable of slicing its targets in vitro (6). The recent finding that many 22G RNA target genes are upregulated in worms expressing CSR-1(SIN), a slicer-dead CSR-1 variant, compared with wild-type CSR-1 indicates that (a) CSR-1 slices target mRNAs in vivo and that (b) CSR-1-mediated slicing contributes to target gene silencing (41). The severity of target upregulation in the csr-1(sin) mutant correlates strongly with the density of corresponding 22G RNA reads from CSR-1 immunopurification, suggesting that CSR-1 may preferentially slice highly expressed genes (41). Second, CSR-1 may have an indirect role in gene silencing by regulating local chromatin environments. We discuss this role for CSR-1 in the section titled Chromatin.

On the basis of sequence homology to CSR-1-bound 22G RNAs, more than 4,000 transcripts are targeted by CSR-1-class 22G RNAs (25). Most of these (more than 3,000) are germline-expressed protein-coding genes (25). When a twofold threshold was used for gene upregulation in a csr-1(sin) versus wild-type background, 344 transcripts were significantly upregulated in the csr-1(sin) mutant (41). Of these 344 genes, 133 correspond to CSR-1-class 22G RNAs, suggesting that they are likely direct targets of 22G RNA-dependent CSR-1 slicing activity (41). The remaining upregulated genes, which are not enriched for germline transcripts, may be indirect targets or may be targeted by 22G RNAs that have not yet been identified through CSR-1 immunopurification experiments (41).

P Granules: Centers for Small RNA-Mediated Surveillance of Self and Nonself mRNA Transcripts

P granules are perinuclear phase-separated assemblies of RNAs and proteins in the C. elegans germline and germline precursors (16, 96, 97). Many components of small RNA pathways, including CSR-1, DRH-3, EGO-1, CDE-1, WAGO-1, WAGO-4, and PRG-1, localize to P granules (12, 25, 41, 49, 110, 123). In support of a role for P granules in small RNA pathways, PGL-1, which is required for the formation of P granules, is required for exo-RNAi in a genetically sensitized background (60, 131).

How do these three distinct, small RNA pathways—piRNAs, WAGO-class 22G RNAs, and CSR-1-class 22G RNAs—coexist in the same physical structure? One attractive model is that the CSR-1 and piRNA pathways compete for mRNA binding to promote either activation or silencing of the transcript. In support of this competitive model, for some transgenes, RNAa can be achieved only in the absence of PRG-1, suggesting that piRNAs compete with the licensing pathway (87). Furthermore, depletion of CSR-1 utilizing an auxin-inducible degradation system results in an approximately twofold increase in the number of unique piRNA-binding sites genome wide (89). The emerging model in the field is that after nuclear export mRNAs localize to P granules, where they are targeted for silencing by piRNAs and the downstream WAGO-class 22G RNAs or alternatively for activation by CSR-1-class 22G RNAs.

One important gap in this model is an explanation for how 22G RNAs are sorted to associate either with WAGOs and direct target silencing or with CSR-1 and promote gene expression. Two recent studies suggest that piRNAs help establish a genetic memory between self and nonself that is transmitted to the next generation by WAGO-class and CSR-1-class 22G RNAs. By crossing mutants of two different WAGO-class 22G RNA pathway components, mut-7 and mut-16, de Albuquerque et al. (31) were able assess de novo 22G RNA establishment in the F₁ generation. Although mut-7(−/−) and mut-16(−/−) homozygous mutants are strongly depleted of WAGO-class 22G RNAs, their cross-progeny, which are heterozygous mut-7(+/−); mut-16(+/−), are able to accumulate WAGO-class 22G RNAs (31). Phillips et al. (79) took a similar approach, crossing mut-14(−/−); smut-1(−/−) homozygous mutants with mut-16(−/−) mutants. For clarity, we refer to both approaches interchangeably as 22G RNA resetting. Although 22G RNA resetting in a prg-1(+/+) background results in approximately 30% sterility in the F₁ generation, 22G RNA resetting in a piRNA-depleted mutant background [i.e., prg-1(−/−) or pid-1(−/−)] results in 100% sterility in the F₁ generation (31, 79). This indicates that resetting the WAGO-class 22G RNAs in the absence of piRNAs is detrimental to fertility. If the same experiments are performed while the WAGO pathway is compromised in the F₁ generation, such as by growth on mut-16 RNAi or in a hrde-1(−/−) background, the sterility from resetting in the prg-1(−/−) background can be partially rescued (31, 79), indicating that it is the reintroduction of WAGO-class 22G RNAs in the absence of piRNAs that causes F₁ sterility. Immunopurification of HRDE-1 and WAGO-1 after 22G RNA resetting without piRNAs shows that these WAGOs are enriched for both WAGO-class and CSR-1-class 22G RNAs (31, 79). After resetting in a prg-1(+/+) background, however, immunopurified HRDE-1 is enriched for WAGO-class 22G RNAs only (79). These findings suggest that when 22G RNAs are reset, piRNAs are critical for specifying appropriate Argonaute loading, which in turn is essential for appropriate germline development (31, 79). The aberrant loading of CSR-1-class 22G RNAs onto HRDE-1 RISC may inappropriately silence genes normally required for successful germline development, causing the severe fertility defects that arise when 22G RNA resetting occurs in the absence of piRNAs (31, 79). Taken together, these studies provide compelling evidence that piRNAs direct the loading of 22G RNAs onto the appropriate Argonaute, consequently establishing a self versus nonself distinction for germline transcripts that is enforced in subsequent generations by the 22G RNAs.

How are mRNAs compartmentalized into WAGO-class targets versus CSR-1-class targets within the P granule? One intriguing possibility is that substructures within P granules may define distinct areas for either amplification of and targeting by CSR-1-class 22G RNAs or routing to the Mutator focus for amplification of the WAGO-class 22G RNAs. Consistent with this idea, a recent study found that two factors required for RNAi inheritance, ZNFX-1 and WAGO-4, localize to P granules early in germline development but separate to form a new phase-separated structure, the Z granule, later in germline development (116). The Z granule appears to localize between P granules and Mutator foci and may be involved in shuttling mRNA transcripts between the two (116). The continued development of superresolution microscopy methods and further study of phase-separated liquid condensates will allow for investigation into how P granule structure and physical properties may facilitate the routing of mRNAs to the appropriate 22G pathway.

CHROMATIN

The role of small RNAs in the regulation of chromatin has been of great interest since siRNAs were found to drive H3K9me3 at target genes in Schizosaccharomyces pombe (52, 114). Furthermore, histone modifications established in parental germlines in C. elegans are inherited in progeny (8, 39, 84, 100) and thus could be a mechanism for the multigenerational effects of small RNAs. How small RNAs are able to regulate chromatin is unclear, as are the relative contributions of the small RNAs themselves and the resulting formation of heterochromatin to the multigenerational phenotypes in the relevant mutant backgrounds.

WAGO-Dependent Establishment and Maintenance of Heterochromatin

As described above, the HRDE-1 bound subclass of WAGO-class 22G RNAs drives heterochromatin formation via H3K9me3 and H3K27me3 (18, 19, 47, 66). In hrde-1 mutants, H3K9me3 loss is progressive over generations (18, 120), indicating that HRDE-1-directed heterochromatin is heritable. The mechanisms underlying the HRDE-1-mediated deposition of H3K9me3 have been an active area of study in recent years; these studies have elucidated protein factors involved in both the establishment and the maintenance of HRDE-1-dependent H3K9me3. In this section, we present an overview of the emerging model in this exciting field; however, the many experimental approaches taken to address these questions have led to contradictory results in some cases. One possibility is that distinct subsets of HRDE-1-targets may engage different machinery to establish and maintain heterochromatin. Furthermore, the DNA sequence itself may contribute to heterochromatin formation at certain loci (37).

Most H3K9 methylation in C. elegans is deposited by the histone methyltransferases (HMTs) SET-25, SET-32, and MET-2 (57, 127). All three HMTs contribute to HRDE-1-mediated H3K9 methylation (9, 57, 95, 122). SET-25 and SET-32 appear to be particularly important for H3K9me3 inheritance in the F₁ generation, as set-25(−/−) and set-32(−/−) mutants have strong impairment in inheriting RNAi-mediated silencing in the F₁ generation. A small fraction of set-25(−/−) and set-32(−/−) mutants, however, can maintain silencing in the F₁ generation and transmit silencing to their progeny at wild-type levels, suggesting that SET-25 and SET-32 are dispensable for long-term maintenance of silencing (122).

Given that SET-25 and SET-32 are dispensable for long-term silencing (56, 57), other mechanisms must exist to ensure long-term silencing downstream of siRNAs. The finding that H3K9me3 loss is progressive in hrde-1(−/−) mutants suggests that inherited small RNAs have an important role in the maintenance of H3K9me3 at target sites (18). Other factors that function downstream of siRNAs are also important for the maintenance of heterochromatin at siRNA target loci. For example, the GHKL ATPase MORC-1 has a critical role in the maintenance of both siRNA-mediated silencing and H3K9me3 (95, 120).

The duration of transgenerational silencing may also be controlled at the chromatin level. Some heritable RNAi effects are enhanced in met-2(−/−) mutants (64). HERI-1, which contains a chromodomain, is recruited to chromatin by RNAi, HRDE-1, and SET-32 to limit the duration of silencing (78). The precise role of HERI-1 is unknown, as is the biological importance of suppressing nuclear RNAi-mediated silencing and heterochromatin formation. heri-1(−/−) mutants exhibit spermatogenesis defects, and those defects are dependent on HRDE-1, suggesting that the enhanced activity of the HRDE-1 pathway is detrimental to spermatogenesis (78).

Relevance of Heterochromatin Beyond Gene Silencing

The use of exo-RNAi and transgenes to dissect pathways contributing to both gene silencing and heterochromatin formation has been extremely fruitful in recent years, elucidating many pathway components. What is unclear, however, is how much these findings are applicable with respect to endo-siRNAs and the resulting heterochromatin formation. One important distinction between endogenous and exogenous pathways is that most targets of the endogenous pathway are either transcriptionally silenced or H3K9 trimethylated but not both (72). Indeed, mutations in H3K9 HMTs lead to little transcriptional activation of targets. However, depletion of the same HMTs in a hrde-1(−/−) mutant background leads to greater depression of native 22G RNA targets than in hrde-1(−/−) mutants alone, suggesting that the transcriptional effects of small RNAs and H3K9me3 are additive (57). Even global depletion of H3K9me3 in C. elegans is associated with relatively little transcriptional change, as less than 10% of H3K9-methylated genes are upregulated in met-2(−/−); set-25(−/−) mutants (127). Taken together, these findings suggest that transcriptional repression may not be a primary function for H3K9me3 in C. elegans.

What is the role of endo-siRNA-directed heterochromatin formation if not to repress target transcription? One important function of the germline nuclear RNAi pathway is to preserve the immortality of the germline lineage from one generation to the next. The nuclear RNAi mutants share a ts germline mortal phenotype, a transgenerational progressive sterility defect, indicating that germline nuclear RNAi is essential for transgenerational germline maintenance at elevated temperatures (18, 73). The downstream, chromatin-associated effectors of endo-siRNAs are also required for germline immortality, as set-32(−/−) and morc-1(−/−) mutants share the germline mortal phenotype of hrde-1(−/−), nrde-1(−/−), nrde-2(−/−), and nrde-4(−/−) mutants, suggesting that endo-siRNAs facilitate germline immortality at the chromatin level (95, 120).

The etiology of the germline mortal phenotype of nuclear RNAi mutants is likely related to epigenetic defects rather than to the accumulation of DNA damage over generations, which is a shared feature of the canonical germline mortal mutants that exhibit defects in telomere maintenance and DNA repair (1, 38, 67). Global depletion of H3K9me3 in C. elegans is also associated with the accumulation of DNA damage (127). In the case of nuclear RNAi mutants, the germline mortal phenotype can be reversed by shifting the worms to a lower temperature (73, 83, 94, 95), indicating that DNA damage, which would not be reversible, is not the primary cause of germline mortality. In further support of an epigenetic mechanism driving germline mortality, mutations in met-1, which encodes one of two H3K36 HMTs in C. elegans, suppress the germline mortality phenotype of hrde-1(−/−) and morc-1(−/−) mutants (120). Methylation of H3K36 is a marker of euchromatin or active chromatin. Notably, the mutations in met-1 that rescue morc-1(−/−) germline mortality do not rescue the overexpression of a subset of endo-siRNA target genes in morc-1(−/−) mutants, suggesting that restoring target repression is not essential for germline immortality (120).

We propose that nuclear RNAi has an essential role in the three-dimensional organization of germline chromatin and that loss of that organization in nuclear RNAi mutants is an important driver of germline mortality. We have three primary lines of evidence for this model: (a) In addition to their RNAi inheritance defect and germline mortality phenotype, morc-1(−/−) mutants grown at the nonpermissive temperature have defective compaction of germline chromatin that is also rescued by mutations in met-1. (b) In a wild-type background, sites that are dependent on HRDE-1 for H3K9me3 maintenance tend to occupy sites of high local H3K9me3 and local minima of H3K36me3, and the loss of H3K9me3 enrichment in hrde-1(−/−) and morc-1(−/−) mutants corresponds with increased H3K36me3. (c) In hrde-1(−/−) and morc-1(−/−) mutants, the number of sites that are MET-1 dependent for their H3K36me3 is increased four- to five-fold compared with wild-type backgrounds, suggesting that HRDE-1 and MORC-1 function to limit MET-1 activity, target access, or both (120). Taken together, these findings support a model in which endo-siRNAs and the nuclear RNAi machinery direct germline chromatin compaction and limit the encroachment of euchromatic marks, such as H3K36me3, into heterochromatinized parts of the genome to facilitate transgenerational germline maintenance.

Contribution of DNA Sequence to Multigenerational Silencing

The DNA sequence itself may also modulate expression versus repression of germline transcription. The promoters and introns of germline-expressed genes are enriched for 10-base periodic An/Tn clusters (PATCs) (35). PATCs promote the expression of transgenes that are inserted into regions of heterochromatin (37). PATC density is anticorrelated with 22G RNA density, and the insertion of a PATC-containing intron into a transgene that is otherwise subjected to piRNA-mediated silencing is sufficient to drive stable transgene expression and a corresponding decrease in the number of 22G RNAs targeting the transgene (130). These studies suggest that PATCs may directly oppose endo-siRNA-mediated silencing to promote germline gene expression and that further study is needed to elucidate the mechanism.

CSR-1-Dependent Chromatin

RNA-profiling studies have suggested that CSR-1 silences some genes that are expressed at low levels in wild-type worms. This may occur due to the impact of CSR-1 on local chromatin environments rather than as a direct result of interactions between 22G RNAs and mRNA targets. In a csr-1 hypomorphic mutant, RNA Pol II occupancy is increased at weakly expressed genes (22). This is consistent with a subset of targets that are upregulated in the csr-1(sin) mutant, which expresses a catalytically inactive, or slicer-dead, CSR-1, but are not normally highly expressed or germline enriched; these targets do not correspond to identified CSR-1-bound 22G RNAs (41). Furthermore, genes normally expressed specifically in embryos are precociously expressed in csr-1(−/−) and csr-1(sin) mutant oocytes but are not generally targeted by CSR-1-class 22G RNAs (34). How could this potentially 22G RNA-independent silencing of low-expressing genes be achieved? One explanation is that CSR-1 may be important for the integrity of heterochromatin in the vicinity of its targets and that loss of CSR-1 impairs heterochromatin formation or stability, leading to ectopic expression of normally repressed transcripts. In support of this model, both imaging and biochemical studies have shown that the centromeric histone variant CENP-A is disorganized in csr-1(−/−) or hypomorphic mutants, indicative of a heterochromatin defect (22, 25). In csr-1 hypomorphic mutants, normally silent loci that flank CSR-1 target genes lose their enrichment for the repressive histone mark H3K27me3 and CENP-A and are upregulated (22). Taken together, these findings suggest that CSR−1 regulates the chromatin landscape in the vicinity of its targets to maintain the silenced, heterochromatic state. Loss of this regulation in csr-1 hypomorphic mutants and drh-3(−/−) mutants may explain the global increase in antisense transcription in these mutants (22).

In addition to its contribution to CENP-A and H3K27me3 organization, CSR-1 is required for correct loading of condensins and cohesins onto mitotic chromosomes in early embryos (25). The widespread disorganization of these factors is thought to cause the chromosome segregation defects, such as anaphase bridging and accumulation of abnormal nuclei, in csr-1(−/−) mutants (25). Importantly, CSR-1 slices the transcripts that encode some microtubule assembly proteins, whose misregulation in a csr-1(sin) mutant background also contributes to chromosome segregation defects (41). Although the consequences of this misregulation are most apparent in early embryos, where defects arise at the first mitotic division and continue until embryonic arrest at the 50-cell stage (25), they are not limited to the early embryo. H3K9me2 is more broadly distributed along meiotic chromosomes in csr-1(−/−), drh-3(−/−), and ekl-1(−/−) mutant males than in wild-type males (88). Meiotic chromosomes also show decreased H3K4me2 staining and aberrant overlap of H3K9me2 and H3K4me2 marks, suggesting that the boundaries between heterochromatin and euchromatin are blurred (88). Loss of heterochromatin-euchromatin boundaries may also contribute to the finding that spermatocyte nuclei are more condensed in csr-1(−/−) mutant males than in wild-type males (28), although there are no reports on the distribution of heterochromatic marks in csr-1(−/−) mutant males.

REMAINING QUESTIONS

What Role Do Other Factors Required for RNAi Inheritance Have?

Multiple factors implicated in RNAi inheritance have precise functions that are not yet understood. Among these are several proteins related to splicing, including EMB-4, which binds to HRDE-1 and functions in nuclear RNAi specifically against intron-containing transcripts (2, 107). The splicing factors RNP-3 and MOG-2 are also required to transmit silencing to progeny (71). The possibility of a functional connection between splicing and endo-siRNA function is intriguing, as these two pathways show similar patterns of phylogenetic conservation (98). In the yeast Cryptococcus neoformans, stalled spliceosomes serve as a trigger for endo-siRNA biogenesis (33). In C. elegans, WAGO-class 22G RNA targets tend to be poorly conserved and have weaker splice sites; thus, spliceosomal engagement may serve as a surveillance mechanism that directs specific transcripts for silencing (71).

Four additional factors have been identified in a forward genetic screen for defective RNAi inheritance: GLH-1/VASA, CDE-1, HRDE-2, and HRDE-4 (95). The possibility of a role for CDE-1 in the inheritance of silencing is intriguing because we already know that CDE-1 functions as part of the opposing CSR-1 22G RNA pathway (110). CDE-1 is thought to destabilize the CSR-1-class 22G RNAs by catalyzing the addition of nontemplated uridines on CSR-1-class 22G RNAs (110). It is possible that uridylation may have a yet undescribed role in WAGO-class 22G RNAs or that without CDE-1-mediated uridylation the stabilized CSR-1-class 22G RNAs promote the reactivation of RNAi target genes in progeny in direct opposition to the germline nuclear RNAi pathway.

Does Chromatin Structure Feed Forward into siRNA Biogenesis?

One seeming paradox in the field is how endo-siRNAs, which are transcribed from their mRNA targets, can continue to be expressed when their target loci lie in regions of heterochromatin and thus are presumably transcriptionally inactive. In yeast, siRNA-dependent heterochromatin promotes siRNA amplification (55, 75), and in Drosophila melanogaster, heterochromatin drives piRNA expression (5), suggesting that a feedback loop between small RNA amplification and heterochromatin formation may be a conserved strategy to maintain small RNA expression from heterochromatinized loci. We do not yet have direct evidence that such a mechanism exists in C. elegans, but the finding that PATCs disrupt heterochromatin formation and also seem to inhibit 22G RNA amplification could be a clue that 22G RNA amplification is at least partially dependent on heterochromatin (37, 130).

Alternatively, heterochromatin formation and 22G RNA amplification may be spatially distinct in the C. elegans germline. Single-molecule fluorescence in situ hybridization studies examining the expression of two 22G RNA-targeted long-terminal repeat retrotransposons (Cer3 and Cer8) have shown that in particular germline zones a small fraction of nuclei express long-terminal repeats in wild-type backgrounds and that HRDE-1 and NRDE-2, though essential for silencing in early germline development (through early pachytene), are dispensable for silencing in later germline stages (74). Although the extent to which these findings are generalizable to other targets will require further investigation, it is possible that some degree of developmental relaxation of repression allows sufficient endo-siRNAs to be amplified to trigger silencing for the remainder of germline development (74).

CONCLUDING REMARKS

In summary, germline gene expression and chromatin structure in C. elegans are controlled by a sophisticated network of small RNAs that act both cotranscriptionally and posttranscriptionally and at the chromatin level. Both the small RNAs and the downstream chromatin effects contribute to multigenerational inheritance of gene silencing. In addition to their roles in controlling gene expression, small RNAs contribute to important aspects of germline chromatin architecture that are essential for transgenerational germline maintenance independently of their effects on gene expression. Although some aspects of these pathways are specific to nematodes, for example, the dramatically expanded Argonaute family and the CSR-1 pathway for gene licensing, many essential principles are conserved in other organisms. H3K9 methylation is directed by endo-siRNAs in both Arabidopsis thaliana and S. pombe and by piRNAs in D. melanogaster. The extent to which such regulation might occur in mammalian systems is unclear, but observations of multigenerational effects of environmental exposures and nutritional status in mammals raise the intriguing possibility that similar pathways in humans may exist.