poly(UG)-tailed RNAs in genome protection and epigenetic inheritance

a, Illumina MiSeq was performed (n = 1 biological experiment) on oma-1 pUG PCR products derived from WT and rde-3(-) worms, with or without oma-1 dsRNA. The number of sequenced pUG RNAs (y-axis) mapping to each pUGylation site (x-axis) is shown. Inset, total number of sequenced oma-1 pUG RNAs from indicated samples and total number of these sequenced pUG RNAs in which the oma-1 sequence was spliced. b, MiSeq-sequenced oma-1 pUG RNAs were sorted into four groups on the basis of the nucleotide at the last templated position (−1) of the oma-1 mRNA. The percentage of oma-1 pUG RNAs (MiSeq reads) with each nucleotide in the −1 position is shown. Logo analysis was then performed on each of the four groups to determine the probability of finding each nucleotide at the first position of the pUG tail (+1), as well as at the second-to-last templated nucleotide of oma-1 (−2). This analysis showed that if the last templated nucleotide of the oma-1 mRNA fragment was an A or a C, then RDE-3 was equally likely to add a U or a G as the first nucleotide of an elongating pUG tail. If, however, the last templated nucleotide was a U or G, then RDE-3 preferentially added a G or U, respectively, as the first nucleotide in an elongating pUG tail. Note, to perform the analyses in this figure, we assumed that if a U or G could have been genomically encoded, then it was. If, instead, RDE-3 added the U or G shown in the −1 position as the first nucleotide of the pUG tail, then these data show that the second nucleotide that RDE-3 prefers to add is a G after a U or a U after a G. CCA-adding rNT enzymes modify the 3′ termini of transfer RNAs (tRNAs) with non-templated CCA nucleotides. The mechanism by which these enzymes add non-templated nonhomopolymeric stretches of nucleotides is thought to involve allosteric regulation of the nucleotide-binding pocket by the 3′ nucleotide of a substrate tRNA⁶². A similar mechanism may explain how RDE-3 can add pUG tails to its mRNA substrates. For instance, when the 3′ nucleotide of an RDE-3 substrate is a U, the rNTP binding pocket of RDE-3 might adopt a structure that preferentially binds G and vice versa when the 3′ nucleotide of an RDE-3 substrate is a G. Such a model could explain how a single rNT enzyme adds perfectly alternating U and G nucleotides to RNA substrates. There are also alternative models for how RDE-3 might add pUG tails to an RNA. These include: (1) the existence of a poly(AC) nucleic acid template used by RDE-3 during pUG tail synthesis, (2) the existence of one or more rNTs that cooperate with RDE-3 to produce pUG tails, or (3) the possibility that RDE-3 binds and incorporates UG or GU dinucleotides. We disfavour the first two possibilities, as these models are difficult to reconcile with the observation that RDE-3 adds UG repeats to tethered RNAs in yeast or in Xenopus oocytes⁵. The third proposed model may be true, but because our sequencing shows that pUG tails can initiate with either a U or a G (this figure, Supplementary Table 1), then RDE-3 would need to be able to bind both UG and GU dinucleotides. Determining the mechanism by which RDE-3 adds pUG tails will probably involve structural studies and/or in vitro pUGylation assays using recombinant RDE-3 protein.

a, gfp::h2b, rde-3(-); gfp::h2b and WT (no gfp::h2b) worms were fed E. coli expressing either empty vector control or gfp dsRNA. b, WT and rde-3(-) worms were fed E. coli expressing empty vector control and either oma-1 or dpy-11 dsRNA. In a, b, gfp, dpy-11 and oma-1 pUG RNAs were detected using the assay outlined in Fig. 1a. Data are representative of three biologically independent experiments. c, rde-1(ne219); oma-1(zu405ts) worms were injected with either an oma-1 (n = 6) or a gfp (n = 10) pUG RNA. n = 3 for no injection. The percentage of embryos hatched was scored for the progeny of injected worms. Panels below the x-axis show RNAs run on 2% agarose gel to assess RNA integrity. Data are mean ± s.d. d, rde-1(ne219); gfp::h2b worms were injected with either an oma-1 or a gfp pUG RNA (n = 10 for both, 3 for no injection). Data are mean ± s.d. of percentage of progeny with gfp::h2b silenced. In c, d, all pUG tails were 36 nt in length.

a, Worms of the indicated genotypes (all harbouring the oma-1(zu405ts) mutation) were treated with or without oma-1 dsRNA. For each experiment, the percentage of embryos hatched was scored at 20 °C and averaged for six individual worms per treatment for each genotype. rde-1(ne219) mutants, which cannot respond to dsRNA³, serve as a control for this experiment. Data are mean ± s.d of three biologically independent experiments. b, Control or rde-3(ne298) worms (all rde-1(ne219); oma-1(zu405ts) background) were injected with oma-1 pUG RNAs and the percentage of embryos hatched was scored at 20 °C. n = 10 noninjected and 16 injected worms for control. n = 8 noninjected and 14 injected worms for rde-3(ne298). Data are mean ± s.d.

a, b, rde-1(ne219); oma-1(zu405ts) worms were injected with: an oma-1 pUG RNA consisting of the sense or antisense strand of the same 541-nt-long oma-1 mRNA fragment (beginning at the aug) with a 36-nt 3′ pUG tail (a; n = 9 for both; n = 3 for no injection); oma-1 pUG RNAs consisting of oma-1 mRNA fragments of varying lengths (with position 1 starting at the aug of the oma-1 mRNA sequence) all appended to a 36-nt pUG tail (b; n = 6 (no injection), 10 (1–50), 17 (1–100), 8 (1–270), 9 (271–540) and 15 (1–540)). In a, b, percentage of embryonic arrest was scored at 20 °C. Data are mean ± s.d.

a, mRNAs upregulated in rde-3(-) mutants (Supplementary Table 2) were compared to published lists of: (1) RNAs targeted by CSR-1-bound endo-siRNAs⁴⁹, (2) piRNA-targeted mRNAs (based on predictive and experimental approaches)⁵⁰, and (3) WAGO-class mRNAs²⁶. P values were generated using a one-sided Fisher’s exact test. This analysis showed statistically significant overlap between the mRNAs upregulated in rde-3(-) mutants and both piRNA targets and WAGO-class mRNAs. b–d, Total RNA was extracted from WT or rde-3(-) worms. The assay outlined in Fig. 1a was used to detect pUG RNAs for two DNA transposons (Tc4v and Tc5) and a retrotransposon (Cer3) that were significantly upregulated in rde-3(-) worms (b); predicted protein-coding mRNAs that were significantly upregulated in rde-3(-) worms (c); and two randomly selected mRNAs whose expression does not change in rde-3(-) mutants (d). Results are representative of three biologically independent experiments. The same reverse-transcribed samples were used for c and d and, therefore, the gsa-1 loading control is the same for both panels.

a, dpy-11 and oma-1 pUG PCR (Fig. 1a) were performed on total RNA from glp-1(q224/ts) worms grown at 15 °C (germ cells present) or 25 °C (approximately 99% of germ cells absent), with or without oma-1 and dpy-11 dsRNA. Data are representative of two biologically independent experiments. The samples in a are the same as those used in Fig. 3e and, therefore, the gsa-1 loading control is the same. b, oma-1 pUG PCR was performed on total RNA extracted from wild-type, rde-3(-) and mut-16(pk710) worms, with or without oma-1 dsRNA. Data are representative of four biologically independent experiments. c, RT–qPCR was used to quantify levels of oma-1 pUG RNAs in wild-type, rde-3(-) and mut-16(pk710) worms, with or without oma-1 dsRNA. Data are represented as fold change in the levels of oma-1 pUG RNAs with or without oma-1 dsRNA (y-axis) for each strain (x-axis). n = 3 biologically independent samples per treatment for each strain. Data are mean ± s.d. d, RT–qPCR was used to quantify levels of Tc1 pUG RNAs in wild-type, rde-3(-) and mut-16(pk710) worms. The RNA samples used for d are the same as those used in c, except that the data for the samples with and without oma-1 dsRNA were pooled for each strain. n = 6 biologically independent samples for each strain. Data are mean ± s.d. The analyses in c and d showed that mut-16 mutants produced more oma-1, but fewer Tc1, pUG RNAs than wild-type worms. The increased levels of oma-1 pUG RNAs in mut-16(pk710) worms was also suggested by the results in b. Together, these data suggest that Mutator foci probably have an important role in coordinating pUG RNA biogenesis in germ cells, as pUG RNA levels become misregulated in mut-16(pk710) mutants.

a, A biological replicate of the experiment shown in Fig. 4d was performed. oma-1(SNP) pUG or pGC RNAs were injected into rde-1(ne219); oma-1(zu405ts) germlines. SNP ___location is indicated with the dotted line. Injected worms were collected 1–4 h after injection, total RNA was isolated and small RNAs (20–30 nt) were sequenced. The distribution of 22G siRNAs mapping antisense to oma-1 is shown, with 22G siRNA reads normalized to reads per million total reads. oma-1 pUG (but not pGC) RNA injection triggered 22G siRNA production near the site of the pUG tail (pUG-specific 22G siRNAs). For unknown reasons, both pUG and pGC RNA injections triggered production of small RNAs around 400 bp 5′ of either tail. b, The length distribution of small-RNA reads mapping antisense to oma-1 is shown for small RNAs sequenced after oma-1(SNP) pUG RNA injections (Fig. 4d and a). c, The proportion of 22-nt small RNAs mapping antisense to oma-1 containing 5′ A, U, G or C is shown.

a, oma-1(zu405ts) worms were fed bacteria expressing empty vector control or oma-1 dsRNA and the percentage of embryos hatched at 20 °C was scored for six generations. Data are mean ± s.d. of three biologically independent experiments. For each experiment, the percentage of embryos hatched at 20 °C was averaged for six individual worms per treatment for each genotype. b, rde-1(ne219); oma-1(zu405ts) worms were injected with co-injection marker alone (n = 12) or co-injection marker + oma-1 pUG RNA (n = 19) and the percentage of embryos hatched at 20 °C was scored for four generations in lineages of worms established from injected parents (see Methods for details of experimental set-up). Data are mean ± s.d. P values were generated using two-tailed unpaired Student’s t-test. c, c38d9.2 and Tc1 pUG RNA expression quantified by RT–qPCR in embryos collected from wild-type, rde-3(-) or MAGO12 worms. Fold change is normalized to rde-3(-). Each point (n) represents a biologically independent replicate, n = 3 independent replicates per strain. Data are mean ± s.d. d, Same experiment as Fig. 5d. rde-1(ne219); oma-1(zu405ts) worms were injected with an oma-1(SNP) pUG RNA or with co-injection marker only. Co-injection marker-expressing F₁ progeny were picked and allowed to lay their F₂ broods. oma-1 pUG PCR was performed on total RNA from F₂ progeny. Shown are data from three biological replicates. e, Two biological replicates of small RNAs sequenced from the progeny of rde-1(ne219); oma-1(zu405ts) worms injected with oma-1(SNP) pUG or pGC RNAs are shown. Dotted line indicates the ___location of the SNP incorporated into oma-1. The distribution of 22G siRNAs mapping antisense to oma-1 is shown, with 22G siRNA reads normalized to reads per million total reads. In Fig. 4d and Extended Data Fig. 7a, small RNAs were sequenced 1–4 h after injection and 100% of 22G siRNAs antisense to the region of the engineered SNP in oma-1 were found to encode the complement of the SNP. Shown here, less than 1% of 22G siRNAs from the progeny of injected worms encoded the SNP complement. siRNAs mapping near the pUG tail were observed only after oma-1(SNP) pUG RNA injection (pUG-specific siRNAs). For unknown reasons, both oma-1(SNP) pUG and pGC RNAs triggered production of small RNAs 5′ of the pUG-specific siRNAs. It is possible that these siRNAs were triggered by systems that respond to foreign RNAs, such as the piRNA system. Further work will be needed to determine the aetiology of these siRNAs. f, Same experiment as Fig. 5e. oma-1(zu405ts) hermaphrodites were fed oma-1 dsRNA and crossed to rde-3(ne298); oma-1(zu405ts) males. F₂ progeny from this cross were genotyped for rde-3(ne298). WT and rde-3(ne298) homozygous F₃ progeny were phenotyped for the percentage of embryos hatched at 20 °C. Three biologically independent crosses (P₀ 1–3) were performed. Data are mean ± s.d. P values were generated using two-tailed unpaired Student’s t-tests.

Initiation: exogenous and constitutive (that is, genomically encoded such as dsRNA or piRNAs) triggers direct RDE-3 to pUGylate RNAs previously fragmented by factors in the RNAi pathway. Maintenance: pUG RNAs are templates for secondary siRNA synthesis by RdRPs. Argonaute proteins (termed WAGOs) bind secondary siRNAs and: (1) target homologous RNAs for transcriptional and translational silencing^29,34,63,64, as well as (2) direct the cleavage and de novo pUGylation of additional mRNAs. In this way, cycles of pUG RNA-based siRNA production and siRNA-directed mRNA pUGylation maintain silencing over time and across generations. This model shows germline perinuclear condensates termed Mutator foci as the likely sites of pUG RNA biogenesis in germ cells for several reasons. RDE-3 localizes to Mutator foci¹⁷ and we show, in Fig. 3d, that endogenous pUG RNAs localize to Mutator foci. The fact that enzyme and enzyme product both localize to Mutator foci suggests that Mutator foci may be sites of RNA pUGylation. In addition, although pUG RNAs are still made in mut-16 mutants (Extended Data Fig. 6b–d), which lack Mutator foci, the levels of both dsRNA-triggered and endogenous pUG RNAs are misregulated. Thus, while RDE-3 still has enzymatic activity in the absence of Mutator foci, these perinuclear condensates are probably coordinating target recognition and pUGylation in wild-type worms. Indeed, both the endonuclease RDE-8, which cleaves mRNAs targeted by dsRNA¹², and the RdRP RRF-1¹⁷ also localize to Mutator foci, further suggesting that pUG RNA–siRNA cycling occurs in Mutator foci. Previous studies have shown that animals lacking RDE-3 still produce some 22G endo-siRNAs, including 22G siRNAs that associate with the Argonaute CSR-1 and whose biogenesis depends upon the RdRP EGO-1^26,65. Thus, EGO-1 may also produce some 22G siRNAs via a pUG RNA-independent mechanism. A previous study showed that, in rrf-1 mutants that lack germlines, sel-1 RNAi causes a small fraction of sel-1 mRNA fragments to be uridylated in a largely RDE-3-dependent manner in the soma¹². These results suggest that, in somatic tissues, RDE-3 may add non-templated Us to the 3′ termini of mRNA fragments generated during RNAi. It was proposed that this uridylation may be important for turnover or decay of RNAi targets¹². Our work, combined with this earlier data about RDE-3-dependent uridylation¹², suggests two models. First, RDE-3 may possess two distinct catalytic activities: uridylation and pUGylation. According to this model, RDE-3 might add Us or UGs depending on context (for example, cell or tissue type, or developmental timing). Alternatively, the mRNA uridylation observed in the soma could depend on RDE-3 and the pUGylation system, but may be mediated by another, currently unknown, poly(U) polymerase.

a, The gel shown is the same as in Fig. 5a, except that oma-1 pUG RNAs from the P₀ generation are included for WT and rde-3(-) worms. Data are representative of three biologically independent experiments. b, oma-1 pUG RNA reads from MiSeq (n = 1 biological experiment) were mapped to oma-1 and the length of the oma-1 mRNA portion of each pUG RNA read was determined (y-axis). In the box plot, the box represents the interquartile range (IQR) and the centre line shows the median of lengths at the indicated generations after dsRNA treatment. The y-axis starts at the aug of the oma-1 mRNA. The whiskers extend to values below and above 1.5 × IQR from the first and third quartiles, respectively. Data beyond the end of the whiskers are outliers and plotted as points. These data support the results in a, showing that the oma-1 mRNA fragments appended to get pUG tails, and thus oma-1 pUG RNAs, get shorter in each generation during RNAi-triggered TEI. c, A ratchet model to explain pUG RNA shortening. pUG RNA shortening may be due to the 3′→5′ directionality of RdRPs, which, during the maintenance phase of pUG–siRNA cycling (see model in Extended Data Fig. 9), causes each turn of the pUG–siRNA cycle to trigger cleavage and pUGylation of target mRNAs at sites more 5′ than in the previous cycle. Eventually, pUG RNAs are too short to act as RdRP templates, cycling cannot be maintained and silencing ends. Additional support for the ratchet model comes from Fig. 5c, which shows that RNAi-triggered pUG RNAs are longer in MAGO12 mutant worms than in wild-type worms. Note: the P₀ worms in Fig. 5c were exposed to dsRNA continuously from embryos to adulthood, when they were collected. These longer pUG RNAs are probably due to continued initiation of pUGylation triggered by the exogenously provided dsRNA without downstream pUG–siRNA cycling. In the absence of this cycling, pUG RNA shortening does not occur. Finally, a number of recent studies in C. elegans have reported transgenerational inheritance of acquired traits, which lasts three to four generations^{66,67,68,69,70,71}. As shown in a, the expression of oma-1 RNAi–directed pUG RNAs also endures for three to four generations. These shared generational timescales of inheritance hint that the inheritance of acquired traits in C. elegans may be mediated by pUG RNAs whose generational ‘half-life’ is limited to three to four generations owing to the built-in brake on TEI provided by pUG RNA shortening.