Germline
Graphical Abstract
Highlights
d PIWI CLASH identifies piRNA binding sites transcriptome-wide
d piRNAs engage all germline mRNAs via microRNA-like pairing rules
d piRNA target sites have distinct 22G-RNA patterns on silenced and expressed mRNAs
d Targeting by CSR-1 Argonaute correlates with reduced piRNA binding density
Authors
En-Zhi Shen, Hao Chen,
Ahmet R. Ozturk, ..., Si-Yuan Dai, Zhiping Weng, Craig C. Mello
Correspondence
In Brief
Transcriptome-wide profiling of piRNA targeting rules provides new insights into the interplay between Argonaute
pathways and their physiological roles in C. elegans
Shen et al., 2018, Cell 172, 1–15 February 22, 2018ª 2018 Elsevier Inc.
https://doi.org/10.1016/j.cell.2018.02.002
Identification of piRNA Binding Sites
Reveals the Argonaute Regulatory Landscape of the C. elegans Germline
En-Zhi Shen,1,6Hao Chen,2,3,6Ahmet R. Ozturk,1Shikui Tu,2,4Masaki Shirayama,1,5Wen Tang,1Yue-He Ding,1 Si-Yuan Dai,1Zhiping Weng,2and Craig C. Mello1,5,7,*
1RNA Therapeutics Institute, University of Massachusetts Medical School, Worcester, MA 01605, USA
2Program in Bioinformatics and Integrative Biology, University of Massachusetts Medical School, Worcester, MA 01605, USA
3Bioinformatics Program, Boston University, Boston, MA 02215, USA
4Department of Computer Science and Engineering, and CMaCH center, Shanghai Jiao Tong University, Shanghai, China
5Howard Hughes Medical Institute
6These authors contributed equally
7Lead Contact
*Correspondence:[email protected] https://doi.org/10.1016/j.cell.2018.02.002
SUMMARY
piRNAs (Piwi-interacting small RNAs) engage Piwi Argonautes to silence transposons and promote fertility in animal germlines. Genetic and computa- tional studies have suggested that C. elegans piRNAs tolerate mismatched pairing and in principle could target every transcript. Here we employ in vivo cross-linking to identify transcriptome-wide interac- tions between piRNAs and target RNAs. We show that piRNAs engage all germline mRNAs and that piRNA binding follows microRNA-like pairing rules.
Targeting correlates better with binding energy than with piRNA abundance, suggesting that piRNA concentration does not limit targeting. In mRNAs silenced by piRNAs, secondary small RNAs accumu- late at the center and ends of piRNA binding sites.
In germline-expressed mRNAs, however, targeting by the CSR-1 Argonaute correlates with reduced piRNA binding density and suppression of piRNA- associated secondary small RNAs. Our findings reveal physiologically important and nuanced regula- tion of individual piRNA targets and provide evidence for a comprehensive post-transcriptional regulatory step in germline gene expression.
INTRODUCTION
Argonaute (AGO) proteins and their associated small RNAs are fundamental regulators of transcriptional and post-transcrip- tional gene regulation (Czech and Hannon, 2011; Ghildiyal and Zamore, 2009; Hutvagner and Simard, 2008; Meister, 2013;
Siomi and Siomi, 2009; Thomson and Lin, 2009). Piwi proteins are members of the RNaseH-related Argonaute superfamily and associate with small RNAs (i.e., Piwi-interacting RNAs or piRNAs) to form piRNA-induced silencing complexes (piRISCs)
(Czech and Hannon, 2016; Malone and Hannon, 2009; Weick and Miska, 2014). The genomic origins, sequences, and lengths of animal piRNAs vary, but some of the biological functions of piRISCs appear to be shared. For example, piRISCs are required for fertility and transposon silencing in worms, flies, and mice (Aravin et al., 2001, 2007; Batista et al., 2008; Carmell et al., 2007; Siomi et al., 2011; Thomson and Lin, 2009). A growing number of studies suggest that piRNAs and Piwi Argonautes may regulate many, if not all, germline mRNAs (Fagegaltier et al., 2016; Vourekas et al., 2016; Zhang et al., 2015b).
The C. elegans Piwi protein, PRG-1, binds an abundant class of germline-expressed 21-nucleotide (nt) piRNAs with a 50 uri- dine (21U-RNAs)(Batista et al., 2008; Ruby et al., 2006). Target- ing by the 21U-RNA/PRG-1 piRISC complex recruits an RNA- dependent RNA polymerase (RdRP) that initiates the de novo synthesis of secondary 22-nt small RNAs that are templated directly from the target RNA and exhibit a bias for a 50guanosine residue. These so-called 22G-RNAs engage an expanded group of worm Argonautes (WAGOs) that function downstream of piRNAs to silence transposons and many endogenous genes.
The WAGO pathway is required for long-term maintenance of silencing (Bagijn et al., 2012; Lee et al., 2012).
piRNA targeting in C. elegans permits mismatches, suggesting that thousands of endogenous mRNAs could be targeted by piRNAs (Lee et al., 2012). However, anti-silencing mechanisms are thought to prevent or reduce the sensitivity of endogenous mRNAs to piRNA-mediated silencing. The CSR-1 pathway, for example, is thought to be one arm of a ‘‘self’’ recognition pathway that protects endogenous mRNAs from piRNA surveillance (Seth et al., 2013). CSR-1 engages RdRP-derived small RNAs templated from nearly all germline-expressed mRNAs (Claycomb et al., 2009). However, it is unknown whether CSR-1 blocks PRG-1 targeting directly or if it acts downstream to prevent WAGO recruitment.
The identification of piRNA targets is essential for deciphering the roles of piRNAs in both sequence-directed immunity and more broadly in the regulation of germline gene expression.
Here, we optimize a crosslinking, ligation, and sequencing of hybrids (CLASH) protocol to identify piRNAs and associated
(candidate) target RNA binding sites in C. elegans (Helwak et al., 2013; Van Nostrand et al., 2016; Vourekas and Mourelatos, 2014). We identified 200,000 high-confidence piRNA-target site interactions. The overwhelming majority of interactions were between piRNAs and mRNAs. Bioinformatics analysis of the hybrids revealed that targets are enriched for energetically favorable Watson-Crick pairing with their associated piRNAs.
We show that the seed sequence (i.e., positions 2 to 8) and sup- plemental nucleotides near the 30end (positions 14 to 19) of the piRNA are important determinants of piRNA-target binding and silencing, suggesting that piRNA targeting resembles miRNA targeting.
piRNA target sites defined by CLASH show a non-random pattern of WAGO 22G-RNAs that initiate at both ends and near the center (position 12) of the piRNA target site, consis- tent with local recruitment of RdRP. Analysis of CLASH hybrids obtained from CSR-1-depleted animals suggest that CSR-1 protects its targets from PRG-1 binding and WAGO-depen- dent silencing. Our findings reveal that the entire germline mRNA transcriptome engages piRISC, and suggests how germline Argonaute pathways are coordinated to achieve comprehensive regulation and surveillance of germline gene expression.
RESULTS
PRG-1 CLASH Directly Identifies piRNA-Target Chimeras
We used a modified cross-linking, ligation, and sequencing of hybrids (CLASH) approach to identify RNAs associated with the C. elegans PRG-1-piRISC complex. Briefly, CLASH involves the in vivo cross-linking of RNAs to a protein of interest followed by immunoprecipitation (IP), trimming of RNA ends, ligation to form hybrids between proximal RNAs within the crosslinked complex, cDNA preparation, library construction, and deep sequencing (seeSTAR Methods;Figures 1A–1E). In principle, this procedure should allow the recovery of hybrid-sequence reads formed when piRNAs are ligated to proximal cellular target RNAs within the cross-linked PRG-1 IP complex.
To perform CLASH, we first used CRISPR/Cas9 genome editing to introduce a GFP-TEV-FLAG (GTF) multiplex tag into the endogenous prg-1 locus (Kim et al., 2014). In addition to direct fluorescence detection this tag also permits tandem af- finity purification with a TEV protease-mediated elution after the first affinity step. GTF::PRG-1 exhibited a robust expression and was prominently localized in P-granules (Figure S1A) in a pattern identical to that previously reported in PRG-1 immunolocaliza- tion studies (Batista et al., 2008). Moreover, the GTF::PRG-1 fusion protein was functional, as evidenced by its ability to mediate piRNA-dependent silencing of a gfp::cdk-1 reporter gene (Figure S1B).
We then carefully optimized each step of the CLASH procedure using GTF-PRG-1 (Figures 1A–1E;STAR Methods) (Broughton et al., 2016; Helwak et al., 2013). The tandem affinity purification of GTF::PRG-1 resulted in recovery of a single prom- inent protein of the expected size in silver-stained SDS-PAGE gels (Figure 1B). Although GTF::PRG-1 stably associated with piRNAs under these purification conditions, the recovery of
longer associated RNA required the pretreatment of the worms with ultraviolet light. These crosslinked RNA Protein complexes, RNPs, were then treated with nuclease to trim the long RNAs, fol- lowed by intermolecular ligation to form RNA hybrids between the piRNA and target (Figures 1B–1D). RNA hybrids of approxi- mately 42 nts were recovered by gel purification (Figure 1E) and were used for library construction and deep-sequencing.
In two independent experiments, we found similar distribu- tions of mapped sequence reads (Figures S1C–S1H). Together, these comprised a total of21million reads, including a total of
7-million reads corresponding to 17,192 different piRNAs.
Most of these piRNA-containing reads lacked a hybrid sequence (1,083,172), or the hybrid sequences could not be mapped to the genome because they were too short, or for other reasons (3,946,162). We obtained 2,106,813 hybrid reads composed of a piRNA sequence and a genome-mapping sequence, of which
1.5 million were composed of a single piRNA sequence fused to an mRNA. In addition to mRNA chimeras, we detected piRNAs fused to sequences corresponding to rRNA (137,322 reads), tRNA (11,231 reads), pseudogenes (48,208 reads), lincRNA (2,583 reads), miRNA (1,556 reads), introns (10,529 reads), and transposable elements (19,092 reads).
CLASH Reveals piRNA Target Sites in Germline mRNAs Because mRNA chimeras were by far the most abundant type of hybrid read, we chose to focus on mRNA hybrids in the present study. Altogether a total of 16,385 genes were represented among the piRNA hybrids (Figure 1F). We found that ‘‘soma-spe- cific’’ mRNAs were strongly under-represented in the CLASH data (Figure 1G) (Beanan and Strome, 1992; Li et al., 2014), consistent with the idea that CLASH captures interactions between piRNAs and mRNAs that occur in the germline, and not interactions that occur in lysates. The frequency of recovery of each piRNA by CLASH correlated with its level in the input sample as measured by small RNA sequencing (Figure S1I, r = 0.58, p < 0.005).
The nuclease treatment during the CLASH procedure was optimized to produce chimeras of approximately 40 nucleotides.
Thus, each chimera potentially reflects a piRNA/target mRNA duplex ligated at, or near, one end of the duplex. We noted, how- ever, that not all chimeras contained a full-length piRNA and that the recovered target regions varied in length, indicating some variability in nuclease trimming during the CLASH procedure.
Therefore, prior to searching for base-pairing interactions, we inferred the full-length piRNA and extended the empirically defined target space by adding nucleotides to each end, creating ‘‘ideal’’ piRNA/target RNA pairs (seeSTAR Methods).
We next predicted the most energetically favorable piRNA- mRNA interactions from in silico folding of these ‘‘ideal’’ se- quences and compared it with predicted binding energies in a control dataset with randomly matched pairs (Figure 1H). This analysis showed that stable base-pair interactions were strongly enriched in the recovered piRNA-mRNAs chimeras. In fact, when normalized for mRNA levels, hybrid read counts per target site correlated better with binding energy than with piRNA abundance (seeDiscussion) (Figure 1I). Chimeras in which the piRNA 30end was contiguous with mRNA sequence were roughly 20-fold more frequent than chimeras ligated at piRNA 50 ends
(Figure S1J). These findings are consistent with the idea that piRNA 30ends are more available for ligation to their targets. Taken together, these findings support the idea that CLASH captures proximal mRNAs bound to piRISC via base-pairing interactions.
piRNA Targets Exhibit a Pattern of Discrete Peaks in 22G-RNA Levels
In C. elegans, piRISC recruits RdRP to its targets. Therefore, we wished to examine the pattern of RdRP-dependent 22G-RNA A
UV, 254 nm
5’P 3’OH
RNase digestion First IP
5’P 3’OH
3’P 5’OH
5’P 3’OH
3’P 5’P
Second IP
PNK treatment
5’P 3’P
Intermolecular ligation
5’OH 3’OH
5’OH
Crosslinking
FastAP treatment
P-labeled 3’-linker ligation
32
Elution, SDS-PAGE and transfer to NC, Proteinase K digest- ion, RNA isolation
PNK treatment, 5’ linker ligation, cDNA synthesis, library amplification , Illumina sequencing
150 100 75
50
37
kDa Marker BSA WT gtf::prg-1
+ - +
150
100
75
50 250
gtf::prg-1WT
- + + - + +
gtf::prg-1 - +
gtf::prg-1 - +
150
100
75
50
UV
Silver staining Autoradio- graphy
WB Autoradio-
graphy
WB
10bp ladder
400 300
200
100 bp 500
gtf::prg-1 - + UV
B C D E
32P-labelling
gtf::prg-1WT UV UV
F
Total reads with piRNA (7,136,150)
G
Total genome-mapping hybrid reads
(2,106,813) Total hybrids targeting mRNA
(1,527,120) Total unique piRNA /mRNAs identified
(14,307) / (16,731) both
−1 0 1 2
log10 (CLASH counts / RPKM)
median 4.7e-203 3
−40 −30 −20 −10 0
051015
mRNA, CLASH count >=1 mRNA, CLASH count >= 10 rand mRNA
Frequency (%)
H
< −30
−30 ~
−20
−20 ~
−10
−2
−1 0 1 2
log10 (CLASH counts / target site) n=11,781n=23,094n=82,745
median 1.5e-162 8e-234
n=203,399 n=25,235 n=3,796
2.1e-38 0.29
I
(normalized to target RPKM)
high middle low piRNA abundance
∆G (kcal/mol)
∆G (kcal/mol)
Figure 1. PRG-1 CLASH Identifies piRNA-Target Chimeras in C. elegans
(A) CLASH workflow. PRG-1 (gray oval), piRNA (red line), and mRNA (black line). Linkers indicated as blue (30) and purple (50) rectangles.
(B) Silver stain analysis of GFP IP complexes (red arrow) purified from wild-type (WT) and gtf::prg-1 worms with or without UV irradiation.
(C and D) Autoradiography (left) and western blot (right) of FLAG IP complexes released by TEV cleavage after GFP IP. Panels C and D show independent samples. Red line shows the region excised for library preparation.
(E) Polyacrylamide gel showing library products (red line) isolated for sequencing.
(F) Summary of combined data from two CLASH replicates.
(G) Normalized CLASH counts per gene for soma-specific genes, germline-specific genes, and both. CLASH reads normalized as described inSTAR Methods.
(H) Predicted binding energies (DG, kcal/mol) between piRNAs and target sites identified by CLASH (red) or between randomly matched pairs (blue).
(I) Boxplots of CLASH counts per target site with increasing binding energy or piRNA abundance (low, 0%–33%; medium, 33%–66%; high, 66%–100%). Median, solid black line. Significant differences between groups indicated by p values.
See alsoFigure S1.
production near CLASH-defined piRNA target sites in both WT and prg-1 mutant worms. To do this, we plotted 22G-RNA levels within a 40-nt region centered on the piRNA complementary sequences defined by CLASH. The 50 ends of 22G-RNAs are thought to be formed directly from RdRP initiating at C residues within the target mRNAs. We therefore normalized the 22G-RNA levels initiating at each position to the frequency of C residues within the CLASH-defined targets at each position. Because the CSR-1, and WAGO Argonaute pathway are thought to have opposing functions, resisting and supporting piRNA silencing (Seth et al., 2013; Wedeles et al., 2013), we separately considered predicted piRNA targets within previously defined WAGO and CSR-1 targeted mRNAs (Claycomb et al., 2009; Gu et al., 2009). As a control set, we considered a target region arbi- trarily set 100 nts away (within each mRNA) from of the piRNA binding sites identified by CLASH. In WT animals, 22G-RNA levels were much higher for WAGO targets than for CSR-1 tar- gets, as expected (Figures 2A and 2B,Figures S2A andS2B).
However, piRNA binding sites within both WAGO and CSR-1 targets showed a non-random distribution of 22G-RNA levels across the interval. By contrast, the control regions within the same target mRNAs, but offset from the hybrid sites, exhibited no such patterns (Figures 2G and 2H). WAGO targets exhibited a strong central peak, and clusters of peaks at either end of the piRNA target sites. To describe these patterns, we refer to the mRNA sequences near the target site as follows: t1 through t30 includes the presumptive binding site (t1 to t21) plus 9 nucle- otides 50of the target site (t22 to t30). The mRNA region 30of the
target site consists of nucleotides t–1 through t–11. Strikingly, this analysis revealed a prominent peak in the center of the piRNA complementary region near t12, and smaller peaks centered at t1 and t21 (Figure 2A). CSR-1 targets exhibited a cluster of much smaller peaks near the 30end of the predicted target site, with the largest peak residing in sequences located near t–5 (Figure 2B). The amplitudes of 22G-RNA levels on both the WAGO and CSR-1 targets correlated positively with the predicted free energy of piRNA binding and to a lesser extent with piRNA abundance (Figures 2C–2F andS2C–2F).
The amplitude and position of 22G-RNA peaks differed in prg-1 mutants. For WAGO targets, the central peak at t12 was completely depleted in prg-1 mutants, whereas the terminal peaks were reduced. In CSR-1 targets, the prominent peak located at t–5 disappeared, but new peaks at t1, t6, and t21 became evident (Figure 2F). This analysis suggests that PRG-1 influences both the precise position, and the levels of 22G- RNAs on its targets, and that CSR-1 and WAGO targets differ strikingly in their accumulation of 22G-RNAs in response to piRNA targeting (seeDiscussion).
Patterns of piRNA Targeting
Previous studies have revealed features of Argonaute/small RNA guided targeting, including the importance of ‘‘seed’’ pairing be- tween the target and nucleotides 2 to 8 of the small RNA guide (Bartel, 2009). To explore patterns of piRNA-mediated targeting we used two independent computational strategies. In the first strategy, we considered the in silico predicted folding within a A
B
WAGO targetsCSR-1 targets nomalized reads/C/site/millionnomalized reads/C/site/million
prg-1WT 1.0
0.0 0.5
0.0 0.1
prg-1WT
WT prg-1
C
D
30 21 12 1 -11
1.0
0.0 0.5
piRNA U
30 21 12 1 -11
piRNA U
30 21 12 1 -11
piRNA U
30 21 12 1 -11
piRNA U
WT prg-1
1.0
0.0 0.5
prg-1WT
30 21 12 1 -11
piRNA U
30 21 12 1 -11
piRNA U
E
F
WT
prg-1 All Hybrids All Hybrids with G<-20 All Hybrids with G<-20 &
top 30% abundant piRNA
0.2
0.0 0.1 0.2
0.0 0.1 0.2
0.3 0.3 0.3
0.0 0.2
0.1 0.3
30 21 12 1 -11
piRNA U
30 21 12 1 -11
piRNA U
1.0
0.0 0.5
G
H
prg-1WT
prg-1WT
Figure 2. 22G-RNAs Peak at the Center and Ends of piRNA Binding Sites
(A–H) 22G-RNA 50ends cloned from wild-type (red) or prg-1 (blue) worms mapped at single-nucleotide resolution to an extended 40-nt window around CLASH- defined piRNA target sites. Each plot is centered on a 21-nt piRNA, shown schematically. WAGO (A, C, E) and CSR-1 (B, D, F) targets analyzed separately. All hybrids (A and B), hybrids with DG < –20 kcal/mol (C and D), hybrids from the high abundance piRNAs with DG < –20 kcal/mol (E and F), and control target regions > 100-nt from the defined piRNA target sites (G and H).
See alsoFigure S2.
high-confidence group of ‘‘ideal’’ piRNA/target RNA pairs that were identified by at least 5 sequence reads in our combined datasets. To identify preferred base-pairing patterns within this group of hybrids, we applied the Affinity Propagation clustering algorithm (APcluster) (Frey and Dueck, 2007). This analysis revealed a clearly preferred interaction at the seed region and distinct base-pairing patterns at the 30 supplementary region (Figures 3A and 3B, and Figure S3A). Notably, base-pairing frequencies declined from positions 9 to 13 of the piRNA and increased from positions 14 to 19 (Figure S3B). As expected,
these patterns were not enriched in a set of randomized piRNA target RNA pairs (Figure 3A, andFigure S3A).
In the second approach, we performed base-pairing analysis using a sliding 4-, 5-, or 7-nt window of piRNA sequence to search for Watson-Crick pairing in each RNA target (Figures 3C, andFigure S3C). Consistent with the analysis inFigures 3A and 3B, this approach revealed a pattern of seed and supple- mentary pairing, with a distinct drop in pairing from positions 9 to 13 (Figure 3C andFigure S3C). Taken together, these findings suggest that both seed pairing at positions 2 to 8 and
U piRNA (21 nt)
5’ 3’
Watson-Crick Base-Pairing
CLASH chimeras
A C
% of total read counts
0 20 40 60
1..42..53..64..75..86..97..108..119..1210..1311..1412..1513..1614..1715..1816..1917..2018..21
−0.2
−0.1 0.0 0.1 0.2 0.3
1 11 21
●
●G-C A-T
Ratio of base paring in piRNA:mRNA contact site
%ofallCLASHreads
1 10 21 1 10 21
U piRNA (21 nt)
5’ 3’
Negative control (random targets)
B E
D
shuffled target
sites 0.0 0.2 0.4 0.6 0.8 1.0
A C G U
% of seed matches (nts 2-8)
piRNA nucleotide position
(2-6 nt) (1-6 nt)
(2-7 nt) (3-8 nt) (1-7 nt)
(2-8 nt)1mis match (2-8 nt) (1-8 nt) 0.0
1.0 2 22 42
62 interactions shuffled sequences
5-mer 6-mer 7-mer 8-mer
Complementary region
counting from 5’ nt of piRNA
0.31 0.29
0.21
0.23 0.23
0.42
0.21 0.22
0.14
0.25 0.26
0.23
dinucleotide trinucleotide A sliding 4-mer window
Figure 3. piRNAs Target with miRNA-Like Seed and 30Supplementary Base Pairing
(A) Heatmaps of Watson-Crick base-pairing (black pixels) at each piRNA position for piRNA-mRNA chimeras detected at least 5 times (left), and for negative control (random target) sequences.
(B) Chart showing the percentage of CLASH reads (piRNA-target chimeras; black) with complementarity at the indicated positions. Approximately 70% of piRNA- target interactions possess the tested complementarities. Target sequences with shuffled dinucleotides served as control (gray).
(C) A 4-mer sliding window search for perfect Watson-Crick base pairing between piRNAs and CLASH-defined targets.
(D) Ratios of G:C (red) or A:T (blue) base pairing in piRNA-mRNA duplexes, after deducting ratios from random control.
(E) Percent of seed-matched target sites (left) with the indicated nucleotide at target position 1 (t1; opposite piRNA position 1). Randomized target sequences with shuffled dinucleotides (center) and trinucleotides (right) serve as controls. Data expressed as mean ± 2 s.e.m. from two replicates.
See alsoFigure S3.
supplementary pairing at positions 14 to 19 contribute to piRNA- target RNA binding (Shin et al., 2010; Wee et al., 2012).
To further characterize piRNA-mRNA interactions, we analyzed A:U and G:C base-pair ratios at each position of the piRNA. We found no significant difference between the two base pair ratios within the seed region, but in other regions, we found a bias toward G:C pairing (Figure 3D). Notably, cytosine was strongly over-represented in the target strand immediately 30of the seed complement opposite the 50u, (defined as target strand position 1 cytosine, or ‘‘t1C’’) (Figure 3E). This preference contrasts with t1A preferred by insect Piwis (Wang et al., 2014).
A search for evolutionary conservation using PhyloP scores (Pollard et al., 2010) failed to reveal a preferential conservation for piRNA-mRNA target sites (Figure S3D).
Finally, we compared the features of piRNA target interactions on WAGO and CSR-1 targets. The energetics of piRNA targeting, the patterns of seed and supplementary pairing, and the average C content along the target region were no different be- tween these groups (Figure S3F).
Seed and 30Supplementary Pairing Are Required for Target Silencing
To determine the importance of base pairing interactions along the length of the piRNA/target mRNA hybrid, we used CRISPR genome editing to systematically mutate positions 2 to 21 of an anti-gfp piRNA expressed from the 21ux-1 piRNA locus (Fig- ure 4A,Figures S4A–4F;Seth et al., 2018). We then assayed the ability of each 21ux-1(anti-gfp) mutant piRNA to silence a single- copy cdk-1::gfp transgene over a time course of up to 8 genera- tions (Figures 4B and 4C). Strikingly, we found that individual mismatches in the seed region (i.e., m2 to m8) and 30supple- mental region (i.e., m14 to m21) strongly reduced the ability of 21ux-1(anti-gfp) to silence cdk-1::gfp, but mismatches at the central region (m9 to m13) had a much more mild effect (Fig- ure 4B). By the F2 generation, when fully matched 21ux-1(anti- gfp) piRNA silences cdk-1::gfp by 70% (Figure S4G), mis- matches at positions 2 to 8 or 14 to 21 reduced silencing to less than 10% and 25% (respectively) of animals scored (Fig- ure 4B). By contrast, mismatches at positions 9 to 13 reduced silencing activity only slightly, to approximately 50% at the F2 generation. Mismatches at positions 2 or 3 prevented silencing of cdk-1::gfp, even after 8 generations, demonstrating that pair- ing at positions 2 and 3 is essential for piRNA-mediated silencing. Mutants with mismatches at any of the other 18 posi- tions eventually silenced cdk-1::gfp over the 8 generation time course (Figure 4C,Figure S4H). Consistent with these findings, we also observed by western blotting after generations 4 and 8 that mutants with mismatches at positions 2 to 8 or 14 to 21 pro- duced much higher levels of CDK-1::GFP protein than did mu- tants with mismatches at other positions (Figure 4D).
To further test the importance of pairing in these regions, we selectively mutated positions t3, t15, and t21 of the anti-gfp target site in cdk-1::gfp mRNA to compensate for anti-gfp piRNA muta- tions in guide-strand positions, g3, g15, and g21, each of which strongly diminished silencing. As expected, in the absence of 21ux-1(anti-gfp), these silent mutations did not affect the level of GFP expression (Figures S4I–S4K). Strikingly, target mRNAs with ‘‘re-matching’’ mutations at t3, t15, and t21 were each
rapidly silenced by piRNA strains bearing the corresponding guide mutations (Figures S4K–S4M). Thus the failure of the g3, g15, and g21 point mutant piRNAs to silence wild-type cdk-1::
gfp was caused specifically by the mismatches and not by changes in expression or piRISC loading of the mutant piRNAs.
Lastly, we analyzed 22G-RNA induction for several 21ux- 1(anti-gfp) point mutant strains. As expected, we found that 22G-RNA levels correlated with the degree of GFP silencing observed (Figures 4E and 4F). Overall, these findings confirm the importance of base-pairing within the seed region (nucleo- tides 2 to 8) and within the 30supplemental pairing region (nucle- otides 14 to 21) for efficient piRNA targeting.
Specific piRNA-mRNA Interactions Suppress Endogenous mRNA Targets
To investigate how the base-pairing rules defined by our bioin- formatics and transgene studies affect targeting of an endoge- nous mRNA, we edited the 21ux-1 target site, introducing single mismatches into the predicted 21ux-1/xol-1 target duplex (Fig- ure 5A). XOL-1 is a key regulator of dosage compensation and sex determination in early zygotes, and xol-1 mRNA was recently shown to be regulated by the X chromosome-derived piRNA, 21ux-1 (Tang et al., 2018). Consistent with our findings in the transgene studies, single-nucleotide mismatches within the seed and 30 supplemental pairing regions, but not within the central region, dramatically increased expression of XOL-1 (Fig- ure 5B). The 21ux-1 mutants with mismatches in the seed and 30 supplemental pairing regions were phenotypically similar to 21ux-1 null mutants and enhanced the dosage compensation and sex determination phenotypes (decreased brood size and masculinization of hermaphrodites) caused by silencing the X-signal element sex-1 (Figures 5C and 5D) (Carmi et al., 1998). Thus, mutating a single nucleotide in 21ux-1 dramatically increases both XOL-1 expression and activity.
As with most germline mRNAs we found that xol-1 was tar- geted by multiple piRNAs. We identified a total of 166 CLASH hy- brids containing xol-1 mRNA sequences fused to 40 different piRNAs (Figure 5E). However, given the importance of 21ux-1 in regulating xol-1, and the fact that 21ux-1 is the most abundant piRNA, we were surprised to find that a different piRNA, 21ur-4863, was recovered in xol-1 chimeras at a frequency greater than twice that of 21ux-1 chimeras. Specifically, we identified 8 reads with 21ux-1 fused to its xol-1 target site and 19 reads of 21ur-4863 fused to its xol-1 target site.
We therefore wished to ask if 21ur-4863 is also important for xol-1 regulation. Strikingly, deletion of 21ur-4863 resulted in the upregulation of both xol-1 mRNA and protein levels to a degree similar to that observed in 21ux-1 mutants (Figures 5F and 5G). Similar to the 21ux-1 mutant, the 21ur-4863 deletion mutant enhanced defects in dosage compensation and sex determination caused by sex-1(RNAi): fewer progeny, masculin- ization of hermaphrodites, embryonic lethality, and dumpy (Dpy) body morphology (Figures 5H and 5I,Figure S5A) (Carmi et al., 1998). Thus, 21u-4863 and 21ux-1 are both required for xol-1 silencing—neither is sufficient—suggesting that piRNAs function cooperatively to silence xol-1.
We also examined the consequences of piRNA targeting on fbxb-97 and comt-3, whose mRNAs are also regulated by
5’ pUGUUUCAUAUGAUCUGGGUAU 3’
5’ pUCUUUCAUAUGAUCUGGGUAU 3’
5’ pUGCUUCAUAUGAUCUGGGUAU 3’
5’ pUGUCUCAUAUGAUCUGGGUAU 3’
5’ pUGUUCCAUAUGAUCUGGGUAU 3’
5’ pUGUUUGAUAUGAUCUGGGUAU 3’
5’ pUGUUUCCUAUGAUCUGGGUAU 3’
5’ pUGUUUCACAUGAUCUGGGUAU 3’
5’ pUGUUUCAUCUGAUCUGGGUAU 3’
5’ pUGUUUCAUACGAUCUGGGUAU 3’
5’ pUGUUUCAUAUCAUCUGGGUAU 3’
5’ pUGUUUCAUAUGCUCUGGGUAU 3’
5’ pUGUUUCAUAUGACCUGGGUAU 3’
5’ pUGUUUCAUAUGAUGUGGGUAU 3’
5’ pUGUUUCAUAUGAUCCGGGUAU 3’
5’ pUGUUUCAUAUGAUCUCGGUAU 3’
5’ pUGUUUCAUAUGAUCUGCGUAU 3’
5’ pUGUUUCAUAUGAUCUGGCUAU 3’
5’ pUGUUUCAUAUGAUCUGGGCAU 3’
5’ pUGUUUCAUAUGAUCUGGGUCU 3’
5’ pUGUUUCAUAUGAUCUGGGUAC 3’
3’ .ACAAAGUAUACUAGACCCAUA. 5’
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
mismatches anti-gfp 21U
Target RNA
Single Mismatch
neg pos 2 3 4 5 6 7 8 9 10 11 neg pos 12 13 14 15 16 17 18 19 20 21 Single Mismatch (Position)
anti-gfp 21U
CDK-1::GFP alpha-tubulin
CDK-1::GFP alpha-tubulin
F4F8
D
anti-gfp 21U target site
gfp coding region E
C
A B
anti-gfp 21U target site
gfp coding region
10 reads per million 5 reads per million
0.0 0.5 1.0
F2
F4
F6
F8
CDK-1::GFP (on/total)
anti-gfp 21U sequence, 5’ to 3’
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
U G U U U C A U A C G A U C U G G G U A U
m3
m8
m18 F2
F8
F2
F8
F2
F8 0.0 0.5 1.0
0.0 0.5 1.0
0.0 0.5 1.0
Single Mismatch (Position)
Control Control
F
no piRNA
WT anti-gfp piRNA
m2
m3
m8
m9
m13
m14
22G signal
Figure 4. Seed and 30Supplementary Pair- ing Are Required for Silencing
(A) anti-gfp piRNA (red) and single-nucleotide mismatches (blue) from positions 2 to 21 on the piRNA target site in cdk-1::gfp (black).
(B and C) Graphs of the fraction of GFP-positive worms in the presence of anti-gfp piRNA with single-nucleotide mismatches at the indicated piRNA positions (numbers). Ten worms were randomly picked for fluorescence microscopy at the F2 generation (B) and later generations F4, F6, and F8 (C). Data expressed as mean ± 2 s.e.m. of three experiments.
(D) Western blots of CDK-1::GFP in F4 and F8 worms with single-nucleotide mismatches at each piRNA position (from C). Negative control worms without anti-gfp piRNA. Positive control worms with fully match anti-gfp piRNA.
(E) Schematic of 22G-RNAs targeting gfp in F4 cdk-1::gfp worms with the indicated single- nucleotide mismatches (m2 = position 2 mismatch, etc.). Positions from 50, central, and 30 regions of the piRNA were randomly chosen for analysis. Scale bar, 5 reads per million.
(F) Schematic of 22G-RNAs targeting gfp in cdk- 1::gfp worms with the with the indicated single- nucleotide mismatches (m3, m8, or m18) at the F2 and F8 generations. Scale bar, 5 reads per million.
See alsoFigure S4.
C
GFP::FLAG XOL-1
alpha- tubulin
WT prg-1
21ur -4863
Δ
Relative xol-1 mRNA
F
21ur -4863
Δ
G
25 25 0
0
50
21ur-4863 (ppm: 3.15), chimeric reads:19, ∆G = -31.09 kcal/mol xol-1
50 5'
3' 0
25 A U A G U A G U UG AG UU G A C U U U U
A U U U A UC C G A U C A A C U G A AA CUU CA
G G GA G
reads per million
WT prg-1 0 10 20
chimeric reads
xol-1 no CDS isoform b
isoform c
E
▲
G H
3’ .GCACAAAAAACAAUUUUACGA. 5’
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
21ux-1 target
3’ .GCACAAAAAACAACUUUACGA. 5’
3’ .GUACAAAAAACAACUUUACGA. 5’
3’ .GCACAAAAAAUAACUUUACGA. 5’
xol-1 t2 t11 t14 Amino acid Translation
GLu His Phe Asn Lys Lys His Asp xol-1 silent mutation
▼
▼
▼
5’ pUGUGUUUUUUGUUGAGGUAUC. 3’
21ux-1
I
silent mutation
alpha- tubulin
sex-1 RNAi sex-1 RNAi
WT t2 t11
t14 control RNAi sex-1 RNAi
B A
0 50 100
viability of progeny
prg-1
0.0 0.5 1.0 1.5 2.0
t2 t11 t14
WTprg-1 t2 t11 t14
0 4 8 12
% of pseudo male
WT prg-1 21ur-4863
Δ 0
4 8 12
% of pseudo male
tail morphology
♂
♀♂
♂ normal
normal
pseudo
0 20 40 60
viability of progeny (%)
WT prg-1 21ur-4863
Δ
D
sex-1 RNAi
21ux-1 Δ
21ux-1 (ppm: 858.95), chimeric reads:8, ∆G = -28.1 kcal/mol xol-1
50 5'
3'
0
25 C A U G G A G U UU GU UU U U U G U G U
A G C A U UU C A A C A A A A A A C AC G C CTCGA G
G T TG G GG G
▲
gfp::flag-xol-1 GFP::FLAG
XOL-1
21ux-1 Δ gfp::flag-xol-1
21ux-1 Δ
＋
－ WT
＋ ＋ ＋ ＋
＋
－ WT
＋ ＋
21ux-1 Δ
21ux-1 Δ
Figure 5. 21ur-4863 and 21ux-1 Suppress xol-1 Function in Sex Determination
(A) Diagram of silent mutations at positions t2, t11, and t14 of the 21ux-1 target site in xol-1 (green) to create single-nucleotide mismatches with 21ux-1.
(B) western blot (anti-FLAG) of GFP::FLAG::XOL-1 levels in gfp::flag::xol-1 transgenic worms (+) with intact 21ux-1 (WT), 21ux-1 deletion, or xol-1 silent mutations at position t2, t11, or t14. N2 worms serve as negative control (–).
(C) Bar graphs of percent viable progeny of WT, prg-1 loss of function, 21ux-1 deletion, or xol-1 single-nucleotide mismatch (t2, t11, t14) worms treated with sex- 1(RNAi). n > 150 progeny per experimental group. Data expressed as mean ± 2 s.e.m. of three experiments.
(D) DIC images (upper panel) of typical hermaphrodite, male, and pseudomale worms. Bar graphs (lower panel) show the percent pseudomale in WT, prg-1, 21ux-1 deletion, and xol-1 single-nucleotide mismatch (t2, t11, t14) worms treated with sex-1(RNAi). n > 100 progeny per experimental group. Data expressed as mean ± 2 s.e.m. of three experiments.
(E) Distribution of chimeric xol-1 reads (red) identified by CLASH, and the distribution of xol-1 22G-RNAs (blue) in prg-1 mutant and WT worms. Locations of 21ur-4863 (upper) and 21ux-1 target sites in xol-1 gene indicated by inverted black triangles. Sequences and base pairing (right) of 21ur-4863:xol-1 (upper) and 21ux-1:xol-1 (lower) chimeras. piRNA expression level, number of chimeric reads, and binding energy (DG, kcal/mol) indicated above each chimera. Distribution of 22G-RNAs at single-nucleotide resolution shown below each chimera.
(legend continued on next page)
PRG-1 (Bagijn et al., 2012; Batista et al., 2008; Gu et al., 2009;
Lee et al., 2012). We identified 70 chimeric reads between 21ur-1563 and fbxb-97 (Figure S5B). fbxb-97 mRNA levels were upregulated 1.5-fold in a 21ur-1563 deletion mutant and
8-fold in the prg-1 mutant (Figure S5C). To analyze piRNA regu- lation of comt-3 (Figure S5D), we took the alternative approach of mutating target sequences. We introduced silent mutations into wobble-positions that maintain the comt-3 open reading frame but disrupt 4 piRNA target sites (Figure S5E). comt-3 mRNA levels were markedly increased in the prg-1 mutant and in the comt-3 quadruple-piRNA target site mutant, but were not elevated in a comt-3 single-piRNA target site mutant (Fig- ure S5F). COMT-3::FLAG (introduced by CRISPR) was signifi- cantly elevated (by 1.5 fold) in the quadruple target site mutant (Figure S5G). Taken together, our findings suggest that individual piRNAs exhibit a range of regulatory effects and that multiple piRNAs cooperatively silence individual targets.
CLASH Analysis Reveals Competition between the CSR-1 and PRG-1 Argonaute Pathways
Previous studies suggested that CSR-1 protects its germline mRNA targets from piRNA-mediated silencing (Seth et al., 2013; Shirayama et al., 2012; Wedeles et al., 2013). We sought to test whether CSR-1 protects its targets by preventing PRG-1 from binding. To do this, we used an auxin-inducible degradation (AID) system to conditionally deplete CSR-1 in young adult worms (CSR-1depleted;Figure S6A and S6B) (Zhang et al., 2015a), and then performed CLASH on CSR-1depleted worms in two independent biological replicates. We compared the number of unique piRNA binding sites on CSR-1 targets from CSR-1depleted and wild-type worms. Strikingly, we found that the number of unique piRNA binding sites significantly increased (2 fold) in the CSR-1depleted worms compared to wild-type (Figures 6A and 6B,Figures S6C–S6E). This increase did not result from changes in target mRNA levels, which did not change dramatically during CSR-1 depletion (Figures S6F–
S6H). Increased piRNA targeting is illustrated for dhc-1, whose mRNA levels did not appreciably change (1.5 fold), but whose piRNA targeting was elevated by > 3.4-fold in CSR-1depleted worms (Figure 6C). These results suggest that, when CSR-1 is depleted, mRNAs normally targeted by CSR-1 become bound by additional piRNAs.
To determine whether increased piRNA binding correlates with decreased mRNA levels, we plotted the fold change in mRNA abundance (CSR-1depleted/ WT) versus the fold change in piRNA-binding density (CSR-1depleted/ WT) for 3,820 CSR-1 targets (Figure 6D) (Claycomb et al., 2009). We observed a nega- tive correlation between increased piRNA-binding density and mRNA abundance in CSR-1depletedworms (r = - 0.44). To clearly visualize this relationship, we split the 3,820 CSR-1 targets into
five bins of increasing piRNA binding density and plotted the change in mRNA abundance in CSR-1depletedversus wild-type (Figures 6E). This analysis revealed that, as piRNA binding den- sity increases, mRNA abundance decreases. These findings support the idea that CSR-1 functions, at least in part, upstream of PRG-1 to reduce piRNA targeting.
DISCUSSION
In this study, we took the unbiased approach of directly cross- linking piRNAs to target RNAs in vivo. The resulting transcrip- tome-wide snap-shot of piRNA/target-RNA interactions reveals that all germline mRNAs undergo piRNA surveillance. Our find- ings are consistent with a model for germline gene regulation wherein mRNAs undergo comprehensive post-transcriptional scanning by Argonaute systems. More than 10,000 distinct piRISCs access hundreds of thousands of target sequences on germline mRNAs. Our finding that binding energy was better correlated with hybrid formation than was piRNA abundance, suggests that, for most piRNAs, piRISC concen- tration is not limiting. Thus surveillance by piRISC is both transcriptome -wide and remarkably efficient. Perhaps as yet un- known features of the enigmatic P-granules, where piRISC re- sides and presumably functions, create an environment that facilitates this seemingly daunting task of comprehensive mRNA surveillance (Figure 7; see alsoSeth et al., 2018).
Non-mRNA piRNA Interactions
Although mRNA target sites accounted for greater than 90% of CLASH hybrid reads, we also reproducibly identified CLASH reads mapping to a variety of non-coding RNA species (ncRNAs). For example, over 80,000 CLASH reads and hundreds of different piRNAs were mapped to ncRNA hybrids, including sequences from a single region of the 26S rRNA (Figure S1K).
Interestingly, this rRNA region is also targeted by WAGO 22G- RNAs that were recently reported to downregulate rRNA levels in response to stress (Zhou et al., 2017).
Our studies also identified many interactions between piRNAs and tRNA species. For example, tRNAGlu(CUC) and 21U-8377 formed highly reproducible chimeras that showed thermody- namically stable base-pairing (Figure S1L). Altogether we identi- fied piRNA tRNA hybrids involving 474 different tRNAs and 1225 different piRNAs. The significance of these findings remains to be determined, but it is intriguing that in Drosophila a mutation that leads to accumulation of misprocessed tRNA results in a collapse of Piwi-mediated transposon silencing (Molla-Herman et al., 2015; Yamanaka and Siomi, 2015). We also identified hy- brids between piRNAs and other ncRNAs including microRNAs and annotated long-noncoding, lncRNAs. The identification of these piRNA interactions provides a new lens through which to
(F) Bar graph of xol-1 mRNA levels in WT, prg-1, 21ur-4863 deletion, and 21ux-1 deletion worms measured by RT-qPCR. actin mRNA served as the internal control. Data expressed as mean ± s.d. of three experiments.
(G) Western blot (anti-FLAG) of GFP::FLAG::XOL-1 (top) levels in WT, 21ur-4863 deletion, and 21ux-1 deletion worms. Alpha-tubulin (bottom) was probed as a loading control.
(H and I) Bar graphs of percent viable (H) and pseudomale (I) progeny of WT, prg-1, and 21ur-4863 deletion worms treated with sex-1(RNAi). n > 500 per experimental group. Data expressed as mean ± 2 s.e.m. of three experiments.
See alsoFigures S5.
B A
C
D E
(legend on next page)
explore potential functions and regulation of germline ncRNA species.
Molecular Cross-talk between Germline Argonaute Pathways
Previous genetic studies have revealed interactions between the Piwi pathway and two Argonaute pathways that propagate epigenetic memories of gene expression states: the WAGO pathway, which targets silenced genes, and the CSR-1 pathway, which targets expressed genes. Targeting by WAGO and CSR-1 Argonautes is readily apparent since both engage 22G-RNAs templated directly from the target RNA by RdRP. Therefore, the comprehensive identification of PRG-1/piRNA target sites affords an opportunity to explore how piRNA targeting correlates with 22G-RNA levels across annotated WAGO and CSR-1 tar- geted mRNAs.
A striking and unanticipated pattern of 22G-RNA levels emerged from this analysis. On WAGO-targeted mRNAs, piRNA target sites were correlated with three predominant 22G-RNA peaks, one in the center at t12, and one on each side of the tar- geted site. Interestingly, the central peak at t12 was completely dependent on PRG-1, while the flanking peaks were much less dependent on PRG-1. The flanking peaks that persist in prg-1 mutants may reflect piRNA-initiated 22G-RNAs that function in WAGO-mediated trans-generational silencing. Consistent with this idea, analyses of data from published WAGO IP experiments indicate that 22G-RNAs at these somewhat prg-1-independent flanking sites associate with Argonautes required for propagating piRNA-induced epigenetic silencing (WAGO-1 and WAGO-9) (Figures S2G–S2J). Interestingly, the strongly prg-1-dependent 22G-RNAs generated at t12 associate with WAGO-1 only.
Thus, it will be interesting to learn why WAGO-1 but not WAGO-9 binds these t12-associated species and whether their biogenesis depends on PRG-1-dependent mRNA slicing which is predicted to occur between t10 and t11.
Our findings also shed light on the relationship between PRG-1 and CSR-1 targeting. Depletion of CSR-1 resulted in an increase in both unique and total piRNA hybrid reads on mRNAs targeted by CSR-1. These findings are consistent with genetic findings that CSR-1 protects its targets from PRG-1-induced silencing (Seth et al., 2013; Wedeles et al., 2013). Moreover, piRNA target regions on CSR-1 target mRNAs exhibit a pattern of 22G-RNA accumula- tion that is strikingly different from that observed on WAGO-tar- geted mRNAs. Instead of a central peak and twin flanking peaks, as in WAGO targets, a small but reproducible 22G-RNA peak, positioned just 5 nucleotides 30of the piRNA target site (Figures
S2K and S2L), was evident in CSR-1 target mRNAs. The 22G- RNA distribution around piRNA target sites in CSR-1 mRNAs remained unchanged after the short period of CSR-1 depletion.
This finding suggests that the effect of CSR-1 depletion on pat- terns of WAGO 22G-RNA accumulation, if any, is less rapid and perhaps less direct than its effect on piRNA targeting of these mRNAs. Unfortunately, depletion of CSR-1 leads to adult sterility, precluding a longer-term multi-generational analysis of 22G-RNA patterns. Taken together, our findings suggest that CSR-1 pro- tects its targets from piRNA silencing in two ways; first by reducing the frequency of PRG-1 piRISC binding, and second, perhaps more indirectly, by preventing 22G-RNA accumulation at t12 and flanking regions correlated with WAGO-1 and WAGO-9 targeting.
Rules Governing piRNA Targeting
Our analysis of base-pairing interactions between piRNAs and their targets suggests that animal Piwi- and AGO-clade Argo- nautes have broadly similar patterns of targeting. As previously described for miRNA RISC, we find that piRISC function strongly depends on pairing in the seed region and to a lesser extent on 30 supplemental pairing (Shin et al., 2010). The most significant dif- ference we observe is a shift in 30supplementary pairing from po- sitions 13 to 16 in miRISC to positions 15 to 18 in PRG-1 piRISC (Grimson et al., 2007), perhaps consistent with structural differ- ences between miRISC and piRISC (Matsumoto et al., 2016a).
In addition to base-pairing interactions, both AGO and Piwi Argonautes make direct contact with their target RNAs, including specific amino acid contacts with the t1 nucleotide. Human AGO2 and insect Piwi proteins (i.e., Siwi and Aubergine) exhibit a strong preference for adenosine at t1 (t1A), which differs from our finding that PRG-1 prefers t1C. This preference for C may help ensure that PRG-1 target sites often have optimal positioning of a C residue that can serve as a start site for RdRP-dependent amplification of 22G-RNAs. A comparison of the region in PRG-1 that corresponds to the t1 binding pocket in other Argonautes suggests a possible structural basis for this discrimination for t1C. Whereas the polar hydrophobic amino acid Thr640 in Siwi and Aubergine is thought to bind t1A (Matsumoto et al., 2016a), the corresponding position in PRG-1 is a non-polar hydrophobic leucine (Figure S3E).
Using a sensitive epigenetic silencing assay, we were able to directly validate the importance of pairing at each position of the seed and 30supplemental pairing regions. Silencing was most sen- sitive to the loss of pairing at positions 2 and 3, suggesting that tar- geting is initiated by the first half of the seed region. Remarkably, with the exception of positions 9 to 13, which had very little effect
Figure 6. CSR-1 Prevents piRNA Binding to Its Targets
(A) Box-and-whisker plots of piRNA binding site density in 3,820 CSR-1 targeted mRNAs identified by CLASH in both WT and CSR-1depletedworms. piRNA binding density expressed as the number of normalized CLASH reads per kilobase (target mRNA length) per million mapped reads (RPKM) in WT or CSR-1depleted. Outliers were removed. Mean, median, and p values are indicated. CLASH reads normalized as described inSTAR Methods.
(B) Scatterplot of unique piRNA binding site counts per gene for each CSR-1 target in CSR-1depletedversus WT. For 1466 CSR-1 targets, the number of piRNA binding sites increased > 2-fold after depleting CSR-1. A representative target, dhc-1, shown in panel (C).
(C) Distribution of dhc-1 chimeric reads from WT and CSR-1depletedworms.
(D) Scatterplot of the change in mRNA abundance between WT and CSR-1depletedworms versus the change of piRNA binding density for 3,820 CSR-1 targets.
(E) Boxplots of the change in mRNA expression levels between WT and CSR-1depletedworms for 5 sets of genes with different levels of piRNA binding site changes. Median and mean indicated. Each set is significantly different from the prior one, as indicated by the p values.
See alsoFigures S6.
on silencing, single nucleotide substitutions at any other location from positions 2 to 8 or 14 to 21 dramatically reduced silencing over the first several generations. Mutants with mismatches in the 30supplementary pairing region eventually silenced the target in later generations, but mutants with mismatches in the seed re- gion, especially at g2 and g3, never exhibited full silencing of the target. Thus, seed and 30supplementary pairing are of key impor-
tance to piRNA targeting. Even single-nucleotide changes dramat- ically reduced targeting and extended the number of generations required for penetrant silencing.
The Physiology of piRNA Targeting
In most animals, Piwi mutants are completely sterile, likely due at least in part, to loss of transposon regulation. In worms,
PRG-1
5’
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
piRNA
Transcriptome-wide Scanning and
C U
central seed 3’ supplemental
piRNA Targeting Rule
PRG-1
RdRP-sensitive Spots on WAGO Targets
piRNA-guided Physiological Function
WAGO Targets CSR-1 Targets
1
2
3
5
RdRP
-1 -2 -3 -4 -5 22
23 24 25
26 3’ target
WAGO-1/9-22G WAGO-1-22G (Memory) (Non-memory)
RdRP-sensitive Spots on CSR-1 Targets
4
CSR-1-22G (Memory) 5’
1 12
21
piRNA
C C C C C C U U PRG-1
3’ target
RdRP RdRP RdRP 5’ piRNA
C C C C C C U U PRG-1
3’ target -5
piRNA U
GG G
GG G GG
G
WAGO-1/9-22G (Memory)
1 12
21
RdRP RdRP RdRP RdRP
GG
? ? ? G
piRNA
?
?
destroyed intermediate expressed transcript
Figure 7. Model for a Regulatory Landscape of piRNAs in the C. elegans Germline
most transposons appear to be silenced by epigenetic mecha- nisms—i.e., WAGO 22G-RNAs and heterochromatin path- ways—that maintain transgenerational silencing downstream of PRG-1 (Ashe et al., 2012; Bagijn et al., 2012; Gu et al., 2009; Lee et al., 2012; Shirayama et al., 2012). This additional layer of epigenetic silencing may explain why prg-1 mutants exhibit relatively minor transposon activation and fertility de- fects during early generations, but exhibit declining fertility over multiple generations (i.e., a mortal germline phenotype) (Simon et al., 2014).
PRG-1 is nevertheless constantly required to maintain silencing at some loci. Transgenes exposed to both positive (i.e., CSR-1-dependent) and negative (i.e., piRNA-dependent) signals can achieve a balanced state of regulation, where PRG-1 targeting becomes essential to maintain silencing (Seth et al., 2018). At least a few hundred endogenous mRNAs are significantly upregulated in prg-1 mutants, with a concomitant loss of robust 22G-RNAs levels. One such gene, xol-1, is silenced in the hermaphrodite germline by an X chromosome expressed piRNA, 21ux-1 (Seth et al., 2018; Tang et al., 2018).
Silencing of xol-1 ensures that hermaphrodite offspring respond robustly to signals that initiate dosage compensation and sex determination in the early embryo. Although 21ux-1 is by far the most abundant piRNA species, a piRNA with average abun- dance (21ur-4863) binds xol-1 more efficiently based on the fre- quency of CLASH hybrid identification. 21ur-4863 is predicted to bind xol-1 with higher binding energy than predicted for 21ux-1, highlighting the importance of binding energy rather than abundance in driving piRNA targeting. Surprisingly, both 21ur-4863 and 21ux-1 are required to maintain xol-1 silencing, suggesting that they—and perhaps other—piRNAs coopera- tively silence xol-1. Consistent with this idea, the pattern of 22G-RNA induction along xol-1 extends beyond the regions proximal to these two piRNA target sites, suggesting that addi- tional piRNAs likely contribute to the cooperative regulation of xol-1 mRNA. Indeed, our CLASH experiments identified 40 piRNAs that target different sites in xol-1 mRNA. Similarly, because we tagged the endogenous prg-1 gene with GFP and FLAG to permit tandem-affinity purification, we were able to identify 92 different piRNAs that target sites distributed along the length of gfp (Table S1). Cooperative targeting by these piRNAs could explain why 22G-RNA accumulation occurs broadly along silenced gfp transgenes (Shirayama et al., 2012;
Seth et al., 2018). Remarkably, even though multiple piRNAs regulate xol-1, changing a single nucleotide within the seed or 30supplementary pairing regions of 21ux-1 can disrupt silencing of xol-1 and thus affect the regulation of dosage compensation and sex determination.
In summary, our findings show that piRNAs target the entire germline transcriptome. Together with findings from previous and parallel studies our findings also suggest that piRNAs are remarkably versatile in their control of gene expression. piRNAs can act decisively in one generation to initiate epigenetic silencing that persists for multiple generations without need for further piRNA targeting. piRNAs can act cooperatively to silence germline mRNAs (e.g., xol-1) that would otherwise reac- tivate in each generation. And finally, piRNAs can act gradually, over multiple generations, to progressively silence a germline
mRNA. Understanding how piRNAs achieve these nuanced modes and tempos of regulation may shed light on whole new vistas of post-transcriptional and epigenetic regulation in animal germlines.
STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:
d KEY RESOURCES TABLE
d CONTACT FOR REAGENT AND RESOURCE SHARING
d EXPERIMENTAL MODEL AND SUBJECT DETAILS B C. elegans Strains and Bacterial Strains
d METHOD DETAILS
B CRISPR/Cas9 genome editing
B Small RNA Library Preparation and analysis B mRNA Library Preparation
B Immunoprecipitation and RNA Isolation B RT-qPCR
B Western Blot Analysis B RNAi
B Microscopy
B Auxin-inducible Depletion of CSR-1 B Viability and Pseudomale Development B PRG-1 CLASH Protocol
B Bioinformatic Analysis of CLASH Data
d QUANTIFICATION AND STATISTICAL ANALYSES
d DATA AND SOFTWARE AVAILABILITY SUPPLEMENTAL INFORMATION
Supplemental Information includes six figures and three tables and can be found with this article online athttps://doi.org/10.1016/j.cell.2018.02.002.
ACKNOWLEDGMENTS
We thank P. Zamore and V. Ambros for suggestions; members of Mello and Ambros labs for discussions, D. Conte for comments and edits on the text;
W. Gu for providing the reagents for preparing RNA seq libraries; M. Carmell for proofreading; E. Kittler, D. Wilmot and the Umass Deep Sequencing Core for offering high-throughput sequencing; Caenorhabditis Genetics Center (CGC) for providing strain. This work was supported by an NIH grant (HD078253) to Z. W.; NIH Pathway to Independence Award (GM124460) to W.T.; and NIH grants (GM058800 and HD078253) to C.C.M. C.C.M. is a Ho- ward Hughes Medical Institute Investigator.
AUTHOR CONTRIBUTIONS
Conceptualization, E.S. and C.C.M.; Investigation, E.S., H.C., A.R.O., S.T., M.S., W.T., Y.D., S.D., Z.W. and C.C.M.; Writing-Original draft, E.S. and C.C.M.; Writing-Review & Editing, E.S. and C.C.M.; Supervision, C.C.M and Z.W.
DECLARATION OF INTEREST
The authors declare no competing interests.
Received: January 2, 2018 Revised: January 26, 2018 Accepted: January 31, 2018 Published: February 15, 2018
